rdelaage
/
postgresql
mirror of https://git.postgresql.org/git/postgresql.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity
postgresql/src/backend/executor/execExprInterp.c

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

4682 lines
127 KiB
C
Raw Normal View History

Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
/*-------------------------------------------------------------------------
*
* execExprInterp.c
* Interpreted evaluation of an expression step list.
*
* This file provides either a "direct threaded" (for gcc, clang and
* compatible) or a "switch threaded" (for all compilers) implementation of
* expression evaluation. The former is amongst the fastest known methods
* of interpreting programs without resorting to assembly level work, or
* just-in-time compilation, but it requires support for computed gotos.
* The latter is amongst the fastest approaches doable in standard C.
*
* In either case we use ExprEvalStep->opcode to dispatch to the code block
* within ExecInterpExpr() that implements the specific opcode type.
*
* Switch-threading uses a plain switch() statement to perform the
* dispatch. This has the advantages of being plain C and allowing the
* compiler to warn if implementation of a specific opcode has been forgotten.
* The disadvantage is that dispatches will, as commonly implemented by
* compilers, happen from a single location, requiring more jumps and causing
* bad branch prediction.
*
* In direct threading, we use gcc's label-as-values extension - also adopted
* by some other compilers - to replace ExprEvalStep->opcode with the address
* of the block implementing the instruction. Dispatch to the next instruction
* is done by a "computed goto". This allows for better branch prediction
* (as the jumps are happening from different locations) and fewer jumps
* (as no preparatory jump to a common dispatch location is needed).
*
* When using direct threading, ExecReadyInterpretedExpr will replace
* each step's opcode field with the address of the relevant code block and
* ExprState->flags will contain EEO_FLAG_DIRECT_THREADED to remember that
* that's been done.
*
* For very simple instructions the overhead of the full interpreter
* "startup", as minimal as it is, is noticeable. Therefore
Support arrays over domains. Allowing arrays with a domain type as their element type was left un-done in the original domain patch, but not for any very good reason. This omission leads to such surprising results as array_agg() not working on a domain column, because the parser can't identify a suitable output type for the polymorphic aggregate. In order to fix this, first clean up the APIs of coerce_to_domain() and some internal functions in parse_coerce.c so that we consistently pass around a CoercionContext along with CoercionForm. Previously, we sometimes passed an "isExplicit" boolean flag instead, which is strictly less information; and coerce_to_domain() didn't even get that, but instead had to reverse-engineer isExplicit from CoercionForm. That's contrary to the documentation in primnodes.h that says that CoercionForm only affects display and not semantics. I don't think this change fixes any live bugs, but it makes things more consistent. The main reason for doing it though is that now build_coercion_expression() receives ccontext, which it needs in order to be able to recursively invoke coerce_to_target_type(). Next, reimplement ArrayCoerceExpr so that the node does not directly know any details of what has to be done to the individual array elements while performing the array coercion. Instead, the per-element processing is represented by a sub-expression whose input is a source array element and whose output is a target array element. This simplifies life in parse_coerce.c, because it can build that sub-expression by a recursive invocation of coerce_to_target_type(). The executor now handles the per-element processing as a compiled expression instead of hard-wired code. The main advantage of this is that we can use a single ArrayCoerceExpr to handle as many as three successive steps per element: base type conversion, typmod coercion, and domain constraint checking. The old code used two stacked ArrayCoerceExprs to handle type + typmod coercion, which was pretty inefficient, and adding yet another array deconstruction to do domain constraint checking seemed very unappetizing. In the case where we just need a single, very simple coercion function, doing this straightforwardly leads to a noticeable increase in the per-array-element runtime cost. Hence, add an additional shortcut evalfunc in execExprInterp.c that skips unnecessary overhead for that specific form of expression. The runtime speed of simple cases is within 1% or so of where it was before, while cases that previously required two levels of array processing are significantly faster. Finally, create an implicit array type for every domain type, as we do for base types, enums, etc. Everything except the array-coercion case seems to just work without further effort. Tom Lane, reviewed by Andrew Dunstan Discussion: https://postgr.es/m/9852.1499791473@sss.pgh.pa.us
2017-09-30 19:40:56 +02:00
* ExecReadyInterpretedExpr will choose to implement certain simple
* opcode patterns using special fast-path routines (ExecJust*).
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
*
* Complex or uncommon instructions are not implemented in-line in
* ExecInterpExpr(), rather we call out to a helper function appearing later
* in this file. For one reason, there'd not be a noticeable performance
* benefit, but more importantly those complex routines are intended to be
* shared between different expression evaluation approaches. For instance
* a JIT compiler would generate calls to them. (This is why they are
* exported rather than being "static" in this file.)
*
*
Update copyright for 2023 Backpatch-through: 11
2023-01-02 21:00:37 +01:00
* Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
* Portions Copyright (c) 1994, Regents of the University of California
*
* IDENTIFICATION
* src/backend/executor/execExprInterp.c
*
*-------------------------------------------------------------------------
*/
#include "postgres.h"
Split tuptoaster.c into three separate files. detoast.c/h contain functions required to detoast a datum, partially or completely, plus a few other utility functions for examining the size of toasted datums. toast_internals.c/h contain functions that are used internally to the TOAST subsystem but which (mostly) do not need to be accessed from outside. heaptoast.c/h contains code that is intrinsically specific to the heap AM, either because it operates on HeapTuples or is based on the layout of a heap page. detoast.c and toast_internals.c are placed in src/backend/access/common rather than src/backend/access/heap. At present, both files still have dependencies on the heap, but that will be improved in a future commit. Patch by me, reviewed and tested by Prabhat Sabu, Thomas Munro, Andres Freund, and Álvaro Herrera. Discussion: http://postgr.es/m/CA+TgmoZv-=2iWM4jcw5ZhJeL18HF96+W1yJeYrnGMYdkFFnEpQ@mail.gmail.com
2019-07-08 17:58:05 +02:00
#include "access/heaptoast.h"
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
#include "catalog/pg_type.h"
Identity columns This is the SQL standard-conforming variant of PostgreSQL's serial columns. It fixes a few usability issues that serial columns have: - CREATE TABLE / LIKE copies default but refers to same sequence - cannot add/drop serialness with ALTER TABLE - dropping default does not drop sequence - need to grant separate privileges to sequence - other slight weirdnesses because serial is some kind of special macro Reviewed-by: Vitaly Burovoy <vitaly.burovoy@gmail.com>
2017-04-06 14:33:16 +02:00
#include "commands/sequence.h"
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
#include "executor/execExpr.h"
#include "executor/nodeSubplan.h"
#include "funcapi.h"
#include "miscadmin.h"
#include "nodes/nodeFuncs.h"
#include "parser/parsetree.h"
#include "pgstat.h"
Make the order of the header file includes consistent in backend modules. Similar to commits 7e735035f2 and dddf4cdc33, this commit makes the order of header file inclusion consistent for backend modules. In the passing, removed a couple of duplicate inclusions. Author: Vignesh C Reviewed-by: Kuntal Ghosh and Amit Kapila Discussion: https://postgr.es/m/CALDaNm2Sznv8RR6Ex-iJO6xAdsxgWhCoETkaYX=+9DW3q0QCfA@mail.gmail.com
2019-11-12 04:00:16 +01:00
#include "utils/array.h"
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
#include "utils/builtins.h"
#include "utils/date.h"
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
#include "utils/datum.h"
Make plpgsql use its DTYPE_REC code paths for composite-type variables. Formerly, DTYPE_REC was used only for variables declared as "record"; variables of named composite types used DTYPE_ROW, which is faster for some purposes but much less flexible. In particular, the ROW code paths are entirely incapable of dealing with DDL-caused changes to the number or data types of the columns of a row variable, once a particular plpgsql function has been parsed for the first time in a session. And, since the stored representation of a ROW isn't a tuple, there wasn't any easy way to deal with variables of domain-over-composite types, since the domain constraint checking code would expect the value to be checked to be a tuple. A lesser, but still real, annoyance is that ROW format cannot represent a true NULL composite value, only a row of per-field NULL values, which is not exactly the same thing. Hence, switch to using DTYPE_REC for all composite-typed variables, whether "record", named composite type, or domain over named composite type. DTYPE_ROW remains but is used only for its native purpose, to represent a fixed-at-compile-time list of variables, for instance the targets of an INTO clause. To accomplish this without taking significant performance losses, introduce infrastructure that allows storing composite-type variables as "expanded objects", similar to the "expanded array" infrastructure introduced in commit 1dc5ebc90. A composite variable's value is thereby kept (most of the time) in the form of separate Datums, so that field accesses and updates are not much more expensive than they were in the ROW format. This holds the line, more or less, on performance of variables of named composite types in field-access-intensive microbenchmarks, and makes variables declared "record" perform much better than before in similar tests. In addition, the logic involved with enforcing composite-domain constraints against updates of individual fields is in the expanded record infrastructure not plpgsql proper, so that it might be reusable for other purposes. In further support of this, introduce a typcache feature for assigning a unique-within-process identifier to each distinct tuple descriptor of interest; in particular, DDL alterations on composite types result in a new identifier for that type. This allows very cheap detection of the need to refresh tupdesc-dependent data. This improves on the "tupDescSeqNo" idea I had in commit 687f096ea: that assigned identifying sequence numbers to successive versions of individual composite types, but the numbers were not unique across different types, nor was there support for assigning numbers to registered record types. In passing, allow plpgsql functions to accept as well as return type "record". There was no good reason for the old restriction, and it was out of step with most of the other PLs. Tom Lane, reviewed by Pavel Stehule Discussion: https://postgr.es/m/8962.1514399547@sss.pgh.pa.us
2018-02-14 00:52:21 +01:00
#include "utils/expandedrecord.h"
SQL/JSON: add standard JSON constructor functions This commit introduces the SQL/JSON standard-conforming constructors for JSON types: JSON_ARRAY() JSON_ARRAYAGG() JSON_OBJECT() JSON_OBJECTAGG() Most of the functionality was already present in PostgreSQL-specific functions, but these include some new functionality such as the ability to skip or include NULL values, and to allow duplicate keys or throw error when they are found, as well as the standard specified syntax to specify output type and format. Author: Nikita Glukhov <n.gluhov@postgrespro.ru> Author: Teodor Sigaev <teodor@sigaev.ru> Author: Oleg Bartunov <obartunov@gmail.com> Author: Alexander Korotkov <aekorotkov@gmail.com> Author: Amit Langote <amitlangote09@gmail.com> Reviewers have included (in no particular order) Andres Freund, Alexander Korotkov, Pavel Stehule, Andrew Alsup, Erik Rijkers, Zihong Yu, Himanshu Upadhyaya, Daniel Gustafsson, Justin Pryzby. Discussion: https://postgr.es/m/CAF4Au4w2x-5LTnN_bxky-mq4=WOqsGsxSpENCzHRAzSnEd8+WQ@mail.gmail.com Discussion: https://postgr.es/m/cd0bb935-0158-78a7-08b5-904886deac4b@postgrespro.ru Discussion: https://postgr.es/m/20220616233130.rparivafipt6doj3@alap3.anarazel.de Discussion: https://postgr.es/m/abd9b83b-aa66-f230-3d6d-734817f0995d%40postgresql.org
2023-03-29 12:11:36 +02:00
#include "utils/json.h"
#include "utils/jsonb.h"
SQL/JSON: support the IS JSON predicate This patch introduces the SQL standard IS JSON predicate. It operates on text and bytea values representing JSON, as well as on the json and jsonb types. Each test has IS and IS NOT variants and supports a WITH UNIQUE KEYS flag. The tests are: IS JSON [VALUE] IS JSON ARRAY IS JSON OBJECT IS JSON SCALAR These should be self-explanatory. The WITH UNIQUE KEYS flag makes these return false when duplicate keys exist in any object within the value, not necessarily directly contained in the outermost object. Author: Nikita Glukhov <n.gluhov@postgrespro.ru> Author: Teodor Sigaev <teodor@sigaev.ru> Author: Oleg Bartunov <obartunov@gmail.com> Author: Alexander Korotkov <aekorotkov@gmail.com> Author: Amit Langote <amitlangote09@gmail.com> Author: Andrew Dunstan <andrew@dunslane.net> Reviewers have included (in no particular order) Andres Freund, Alexander Korotkov, Pavel Stehule, Andrew Alsup, Erik Rijkers, Zihong Yu, Himanshu Upadhyaya, Daniel Gustafsson, Justin Pryzby. Discussion: https://postgr.es/m/CAF4Au4w2x-5LTnN_bxky-mq4=WOqsGsxSpENCzHRAzSnEd8+WQ@mail.gmail.com Discussion: https://postgr.es/m/cd0bb935-0158-78a7-08b5-904886deac4b@postgrespro.ru Discussion: https://postgr.es/m/20220616233130.rparivafipt6doj3@alap3.anarazel.de Discussion: https://postgr.es/m/abd9b83b-aa66-f230-3d6d-734817f0995d%40postgresql.org
2023-03-31 22:34:04 +02:00
#include "utils/jsonfuncs.h"
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
#include "utils/lsyscache.h"
Make the order of the header file includes consistent in backend modules. Similar to commits 7e735035f2 and dddf4cdc33, this commit makes the order of header file inclusion consistent for backend modules. In the passing, removed a couple of duplicate inclusions. Author: Vignesh C Reviewed-by: Kuntal Ghosh and Amit Kapila Discussion: https://postgr.es/m/CALDaNm2Sznv8RR6Ex-iJO6xAdsxgWhCoETkaYX=+9DW3q0QCfA@mail.gmail.com
2019-11-12 04:00:16 +01:00
#include "utils/memutils.h"
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
#include "utils/timestamp.h"
#include "utils/typcache.h"
#include "utils/xml.h"
/*
* Use computed-goto-based opcode dispatch when computed gotos are available.
* But use a separate symbol so that it's easy to adjust locally in this file
* for development and testing.
*/
#ifdef HAVE_COMPUTED_GOTO
#define EEO_USE_COMPUTED_GOTO
#endif /* HAVE_COMPUTED_GOTO */
/*
* Macros for opcode dispatch.
*
* EEO_SWITCH - just hides the switch if not in use.
* EEO_CASE - labels the implementation of named expression step type.
* EEO_DISPATCH - jump to the implementation of the step type for 'op'.
* EEO_OPCODE - compute opcode required by used expression evaluation method.
* EEO_NEXT - increment 'op' and jump to correct next step type.
* EEO_JUMP - jump to the specified step number within the current expression.
*/
#if defined(EEO_USE_COMPUTED_GOTO)
Perform slot validity checks in a separate pass over expression. This reduces code duplication a bit, but the primary benefit that it makes JITing expression evaluation easier. When doing so we can't, as previously done in the interpreted case, really change opcode without recompiling. Nor dow we just carry around unnecessary branches to avoid re-checking over and over. As a minor side-effect this makes ExecEvalStepOp() O(log(N)) rather than O(N). Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2017-12-29 21:38:15 +01:00
/* struct for jump target -> opcode lookup table */
typedef struct ExprEvalOpLookup
{
const void *opcode;
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
ExprEvalOp op;
Perform slot validity checks in a separate pass over expression. This reduces code duplication a bit, but the primary benefit that it makes JITing expression evaluation easier. When doing so we can't, as previously done in the interpreted case, really change opcode without recompiling. Nor dow we just carry around unnecessary branches to avoid re-checking over and over. As a minor side-effect this makes ExecEvalStepOp() O(log(N)) rather than O(N). Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2017-12-29 21:38:15 +01:00
} ExprEvalOpLookup;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
/* to make dispatch_table accessible outside ExecInterpExpr() */
static const void **dispatch_table = NULL;
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
Perform slot validity checks in a separate pass over expression. This reduces code duplication a bit, but the primary benefit that it makes JITing expression evaluation easier. When doing so we can't, as previously done in the interpreted case, really change opcode without recompiling. Nor dow we just carry around unnecessary branches to avoid re-checking over and over. As a minor side-effect this makes ExecEvalStepOp() O(log(N)) rather than O(N). Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2017-12-29 21:38:15 +01:00
/* jump target -> opcode lookup table */
static ExprEvalOpLookup reverse_dispatch_table[EEOP_LAST];
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
#define EEO_SWITCH()
#define EEO_CASE(name) CASE_##name:
#define EEO_DISPATCH() goto *((void *) op->opcode)
#define EEO_OPCODE(opcode) ((intptr_t) dispatch_table[opcode])
#else /* !EEO_USE_COMPUTED_GOTO */
#define EEO_SWITCH() starteval: switch ((ExprEvalOp) op->opcode)
#define EEO_CASE(name) case name:
#define EEO_DISPATCH() goto starteval
#define EEO_OPCODE(opcode) (opcode)
#endif /* EEO_USE_COMPUTED_GOTO */
#define EEO_NEXT() \
do { \
op++; \
EEO_DISPATCH(); \
} while (0)
#define EEO_JUMP(stepno) \
do { \
op = &state->steps[stepno]; \
EEO_DISPATCH(); \
} while (0)
static Datum ExecInterpExpr(ExprState *state, ExprContext *econtext, bool *isnull);
static void ExecInitInterpreter(void);
/* support functions */
static void CheckVarSlotCompatibility(TupleTableSlot *slot, int attnum, Oid vartype);
Verify that expected slot types match returned slot types. This is important so JIT compilation knows what kind of tuple slot the deforming routine can expect. There's also optimization potential for expression initialization without JIT compilation. It e.g. seems plausible to elide EEOP_*_FETCHSOME ops entirely when dealing with virtual slots. Author: Andres Freund Discussion: https://postgr.es/m/20181105210039.hh4vvi4vwoq5ba2q@alap3.anarazel.de
2018-11-16 07:00:30 +01:00
static void CheckOpSlotCompatibility(ExprEvalStep *op, TupleTableSlot *slot);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
static TupleDesc get_cached_rowtype(Oid type_id, int32 typmod,
Redesign the caching done by get_cached_rowtype(). Previously, get_cached_rowtype() cached a pointer to a reference-counted tuple descriptor from the typcache, relying on the ExprContextCallback mechanism to release the tupdesc refcount when the expression tree using the tupdesc was destroyed. This worked fine when it was designed, but the introduction of within-DO-block COMMITs broke it. The refcount is logged in a transaction-lifespan resource owner, but plpgsql won't destroy simple expressions made within the DO block (before its first commit) until the DO block is exited. That results in a warning about a leaked tupdesc refcount when the COMMIT destroys the original resource owner, and then an error about the active resource owner not holding a matching refcount when the expression is destroyed. To fix, get rid of the need to have a shutdown callback at all, by instead caching a pointer to the relevant typcache entry. Those survive for the life of the backend, so we needn't worry about the pointer becoming stale. (For registered RECORD types, we can still cache a pointer to the tupdesc, knowing that it won't change for the life of the backend.) This mechanism has been in use in plpgsql and expandedrecord.c since commit 4b93f5799, and seems to work well. This change requires modifying the ExprEvalStep structs used by the relevant expression step types, which is slightly worrisome for back-patching. However, there seems no good reason for extensions to be familiar with the details of these particular sub-structs. Per report from Rohit Bhogate. Back-patch to v11 where within-DO-block COMMITs became a thing. Discussion: https://postgr.es/m/CAAV6ZkQRCVBh8qAY+SZiHnz+U+FqAGBBDaDTjF2yiKa2nJSLKg@mail.gmail.com
2021-04-13 19:37:07 +02:00
ExprEvalRowtypeCache *rowcache,
bool *changed);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
static void ExecEvalRowNullInt(ExprState *state, ExprEvalStep *op,
ExprContext *econtext, bool checkisnull);
/* fast-path evaluation functions */
static Datum ExecJustInnerVar(ExprState *state, ExprContext *econtext, bool *isnull);
static Datum ExecJustOuterVar(ExprState *state, ExprContext *econtext, bool *isnull);
static Datum ExecJustScanVar(ExprState *state, ExprContext *econtext, bool *isnull);
static Datum ExecJustAssignInnerVar(ExprState *state, ExprContext *econtext, bool *isnull);
static Datum ExecJustAssignOuterVar(ExprState *state, ExprContext *econtext, bool *isnull);
static Datum ExecJustAssignScanVar(ExprState *state, ExprContext *econtext, bool *isnull);
Support arrays over domains. Allowing arrays with a domain type as their element type was left un-done in the original domain patch, but not for any very good reason. This omission leads to such surprising results as array_agg() not working on a domain column, because the parser can't identify a suitable output type for the polymorphic aggregate. In order to fix this, first clean up the APIs of coerce_to_domain() and some internal functions in parse_coerce.c so that we consistently pass around a CoercionContext along with CoercionForm. Previously, we sometimes passed an "isExplicit" boolean flag instead, which is strictly less information; and coerce_to_domain() didn't even get that, but instead had to reverse-engineer isExplicit from CoercionForm. That's contrary to the documentation in primnodes.h that says that CoercionForm only affects display and not semantics. I don't think this change fixes any live bugs, but it makes things more consistent. The main reason for doing it though is that now build_coercion_expression() receives ccontext, which it needs in order to be able to recursively invoke coerce_to_target_type(). Next, reimplement ArrayCoerceExpr so that the node does not directly know any details of what has to be done to the individual array elements while performing the array coercion. Instead, the per-element processing is represented by a sub-expression whose input is a source array element and whose output is a target array element. This simplifies life in parse_coerce.c, because it can build that sub-expression by a recursive invocation of coerce_to_target_type(). The executor now handles the per-element processing as a compiled expression instead of hard-wired code. The main advantage of this is that we can use a single ArrayCoerceExpr to handle as many as three successive steps per element: base type conversion, typmod coercion, and domain constraint checking. The old code used two stacked ArrayCoerceExprs to handle type + typmod coercion, which was pretty inefficient, and adding yet another array deconstruction to do domain constraint checking seemed very unappetizing. In the case where we just need a single, very simple coercion function, doing this straightforwardly leads to a noticeable increase in the per-array-element runtime cost. Hence, add an additional shortcut evalfunc in execExprInterp.c that skips unnecessary overhead for that specific form of expression. The runtime speed of simple cases is within 1% or so of where it was before, while cases that previously required two levels of array processing are significantly faster. Finally, create an implicit array type for every domain type, as we do for base types, enums, etc. Everything except the array-coercion case seems to just work without further effort. Tom Lane, reviewed by Andrew Dunstan Discussion: https://postgr.es/m/9852.1499791473@sss.pgh.pa.us
2017-09-30 19:40:56 +02:00
static Datum ExecJustApplyFuncToCase(ExprState *state, ExprContext *econtext, bool *isnull);
Reduce code duplication for ExecJust*Var operations. This is mainly in preparation for adding further fastpath evaluation routines. Also reorder ExecJust*Var functions to be consistent with the order in which they're used. Author: Andres Freund Discussion: https://postgr.es/m/CAE-ML+9OKSN71+mHtfMD-L24oDp8dGTfaVjDU6U+j+FNAW5kRQ@mail.gmail.com
2019-10-01 00:00:21 +02:00
static Datum ExecJustConst(ExprState *state, ExprContext *econtext, bool *isnull);
Don't generate EEOP_*_FETCHSOME operations for slots know to be virtual. That avoids unnecessary work during both interpreted execution, and JIT compiled expression evaluation. Both benefit from fewer expression steps needing be processed, and for interpreted execution there now is a fastpath dedicated to just fetching a value from a virtual slot. That's e.g. beneficial for hashjoins over nodes that perform projections, as the hashed columns are currently fetched individually. Author: Soumyadeep Chakraborty, Andres Freund Discussion: https://postgr.es/m/CAE-ML+9OKSN71+mHtfMD-L24oDp8dGTfaVjDU6U+j+FNAW5kRQ@mail.gmail.com
2019-10-01 01:06:16 +02:00
static Datum ExecJustInnerVarVirt(ExprState *state, ExprContext *econtext, bool *isnull);
static Datum ExecJustOuterVarVirt(ExprState *state, ExprContext *econtext, bool *isnull);
static Datum ExecJustScanVarVirt(ExprState *state, ExprContext *econtext, bool *isnull);
static Datum ExecJustAssignInnerVarVirt(ExprState *state, ExprContext *econtext, bool *isnull);
static Datum ExecJustAssignOuterVarVirt(ExprState *state, ExprContext *econtext, bool *isnull);
static Datum ExecJustAssignScanVarVirt(ExprState *state, ExprContext *econtext, bool *isnull);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
expression eval: Reduce number of steps for agg transition invocations. Do so by combining the various steps that are part of aggregate transition function invocation into one larger step. As some of the current steps are only necessary for some aggregates, have one variant of the aggregate transition step for each possible combination. To avoid further manual copies of code in the different transition step implementations, move most of the code into helper functions marked as "always inline". The benefit of this change is an increase in performance when aggregating lots of rows. This comes in part due to the reduced number of indirect jumps due to the reduced number of steps, and in part by reducing redundant setup code across steps. This mainly benefits interpreted execution, but the code generated by JIT is also improved a bit. As a nice side-effect it also ends up making the code a bit simpler. A small additional optimization is removing the need to set aggstate->curaggcontext before calling ExecAggInitGroup, choosing to instead passign curaggcontext as an argument. It was, in contrast to other aggregate related functions, only needed to fetch a memory context to copy the transition value into. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de https://postgr.es/m/5c371df7cee903e8cd4c685f90c6c72086d3a2dc.camel@j-davis.com
2020-02-24 23:39:22 +01:00
/* execution helper functions */
static pg_attribute_always_inline void ExecAggPlainTransByVal(AggState *aggstate,
AggStatePerTrans pertrans,
AggStatePerGroup pergroup,
ExprContext *aggcontext,
int setno);
static pg_attribute_always_inline void ExecAggPlainTransByRef(AggState *aggstate,
AggStatePerTrans pertrans,
AggStatePerGroup pergroup,
ExprContext *aggcontext,
int setno);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
Speedup ScalarArrayOpExpr evaluation ScalarArrayOpExprs with "useOr=true" and a set of Consts on the righthand side have traditionally been evaluated by using a linear search over the array. When these arrays contain large numbers of elements then this linear search could become a significant part of execution time. Here we add a new method of evaluating ScalarArrayOpExpr expressions to allow them to be evaluated by first building a hash table containing each element, then on subsequent evaluations, we just probe that hash table to determine if there is a match. The planner is in charge of determining when this optimization is possible and it enables it by setting hashfuncid in the ScalarArrayOpExpr. The executor will only perform the hash table evaluation when the hashfuncid is set. This means that not all cases are optimized. For example CHECK constraints containing an IN clause won't go through the planner, so won't get the hashfuncid set. We could maybe do something about that at some later date. The reason we're not doing it now is from fear that we may slow down cases where the expression is evaluated only once. Those cases can be common, for example, a single row INSERT to a table with a CHECK constraint containing an IN clause. In the planner, we enable this when there are suitable hash functions for the ScalarArrayOpExpr's operator and only when there is at least MIN_ARRAY_SIZE_FOR_HASHED_SAOP elements in the array. The threshold is currently set to 9. Author: James Coleman, David Rowley Reviewed-by: David Rowley, Tomas Vondra, Heikki Linnakangas Discussion: https://postgr.es/m/CAAaqYe8x62+=wn0zvNKCj55tPpg-JBHzhZFFc6ANovdqFw7-dA@mail.gmail.com
2021-04-08 13:51:22 +02:00
/*
* ScalarArrayOpExprHashEntry
* Hash table entry type used during EEOP_HASHED_SCALARARRAYOP
*/
typedef struct ScalarArrayOpExprHashEntry
{
Datum key;
uint32 status; /* hash status */
uint32 hash; /* hash value (cached) */
} ScalarArrayOpExprHashEntry;
#define SH_PREFIX saophash
#define SH_ELEMENT_TYPE ScalarArrayOpExprHashEntry
#define SH_KEY_TYPE Datum
#define SH_SCOPE static inline
#define SH_DECLARE
#include "lib/simplehash.h"
static bool saop_hash_element_match(struct saophash_hash *tb, Datum key1,
Datum key2);
static uint32 saop_element_hash(struct saophash_hash *tb, Datum key);
/*
* ScalarArrayOpExprHashTable
* Hash table for EEOP_HASHED_SCALARARRAYOP
*/
typedef struct ScalarArrayOpExprHashTable
{
saophash_hash *hashtab; /* underlying hash table */
struct ExprEvalStep *op;
Remove size increase in ExprEvalStep caused by hashed saops 50e17ad28 increased the size of ExprEvalStep from 64 bytes up to 88 bytes. Lots of effort was spent during the development of the current expression evaluation code to make an instance of this struct as small as possible. Making this struct larger than needed reduces CPU cache efficiency during expression evaluation which causes noticeable slowdowns during query execution. In order to reduce the size of the struct, here we remove the fn_addr field. The values from this field can be obtained via fcinfo, just with some extra pointer dereferencing. The extra indirection does not seem to cause any noticeable slowdowns. Various other fields have been moved into the ScalarArrayOpExprHashTable struct. These fields are only used when the ScalarArrayOpExprHashTable pointer has already been dereferenced, so no additional pointer dereferences occur for these. Here we also make hash_fcinfo_data the last field in ScalarArrayOpExprHashTable so that we can avoid a further pointer dereference to get the FunctionCallInfoBaseData. This also saves a call to palloc(). 50e17ad28 was added in 14, but it's too late to adjust the size of the ExprEvalStep in that version, so here we just backpatch to 15, which is currently in beta. Author: Andres Freund, David Rowley Discussion: https://postgr.es/m/20220616233130.rparivafipt6doj3@alap3.anarazel.de Backpatch-through: 15
2022-07-06 09:40:32 +02:00
FmgrInfo hash_finfo; /* function's lookup data */
FunctionCallInfoBaseData hash_fcinfo_data; /* arguments etc */
Speedup ScalarArrayOpExpr evaluation ScalarArrayOpExprs with "useOr=true" and a set of Consts on the righthand side have traditionally been evaluated by using a linear search over the array. When these arrays contain large numbers of elements then this linear search could become a significant part of execution time. Here we add a new method of evaluating ScalarArrayOpExpr expressions to allow them to be evaluated by first building a hash table containing each element, then on subsequent evaluations, we just probe that hash table to determine if there is a match. The planner is in charge of determining when this optimization is possible and it enables it by setting hashfuncid in the ScalarArrayOpExpr. The executor will only perform the hash table evaluation when the hashfuncid is set. This means that not all cases are optimized. For example CHECK constraints containing an IN clause won't go through the planner, so won't get the hashfuncid set. We could maybe do something about that at some later date. The reason we're not doing it now is from fear that we may slow down cases where the expression is evaluated only once. Those cases can be common, for example, a single row INSERT to a table with a CHECK constraint containing an IN clause. In the planner, we enable this when there are suitable hash functions for the ScalarArrayOpExpr's operator and only when there is at least MIN_ARRAY_SIZE_FOR_HASHED_SAOP elements in the array. The threshold is currently set to 9. Author: James Coleman, David Rowley Reviewed-by: David Rowley, Tomas Vondra, Heikki Linnakangas Discussion: https://postgr.es/m/CAAaqYe8x62+=wn0zvNKCj55tPpg-JBHzhZFFc6ANovdqFw7-dA@mail.gmail.com
2021-04-08 13:51:22 +02:00
} ScalarArrayOpExprHashTable;
/* Define parameters for ScalarArrayOpExpr hash table code generation. */
#define SH_PREFIX saophash
#define SH_ELEMENT_TYPE ScalarArrayOpExprHashEntry
#define SH_KEY_TYPE Datum
#define SH_KEY key
#define SH_HASH_KEY(tb, key) saop_element_hash(tb, key)
#define SH_EQUAL(tb, a, b) saop_hash_element_match(tb, a, b)
#define SH_SCOPE static inline
#define SH_STORE_HASH
#define SH_GET_HASH(tb, a) a->hash
#define SH_DEFINE
#include "lib/simplehash.h"
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
/*
* Prepare ExprState for interpreted execution.
*/
void
ExecReadyInterpretedExpr(ExprState *state)
{
/* Ensure one-time interpreter setup has been done */
ExecInitInterpreter();
/* Simple validity checks on expression */
Assert(state->steps_len >= 1);
Assert(state->steps[state->steps_len - 1].opcode == EEOP_DONE);
/*
* Don't perform redundant initialization. This is unreachable in current
* cases, but might be hit if there's additional expression evaluation
* methods that rely on interpreted execution to work.
*/
if (state->flags & EEO_FLAG_INTERPRETER_INITIALIZED)
return;
Perform slot validity checks in a separate pass over expression. This reduces code duplication a bit, but the primary benefit that it makes JITing expression evaluation easier. When doing so we can't, as previously done in the interpreted case, really change opcode without recompiling. Nor dow we just carry around unnecessary branches to avoid re-checking over and over. As a minor side-effect this makes ExecEvalStepOp() O(log(N)) rather than O(N). Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2017-12-29 21:38:15 +01:00
/*
* First time through, check whether attribute matches Var. Might not be
* ok anymore, due to schema changes. We do that by setting up a callback
* that does checking on the first call, which then sets the evalfunc
* callback to the actual method of execution.
*/
state->evalfunc = ExecInterpExprStillValid;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
/* DIRECT_THREADED should not already be set */
Assert((state->flags & EEO_FLAG_DIRECT_THREADED) == 0);
/*
* There shouldn't be any errors before the expression is fully
* initialized, and even if so, it'd lead to the expression being
* abandoned. So we can set the flag now and save some code.
*/
state->flags |= EEO_FLAG_INTERPRETER_INITIALIZED;
/*
* Select fast-path evalfuncs for very simple expressions. "Starting up"
Support arrays over domains. Allowing arrays with a domain type as their element type was left un-done in the original domain patch, but not for any very good reason. This omission leads to such surprising results as array_agg() not working on a domain column, because the parser can't identify a suitable output type for the polymorphic aggregate. In order to fix this, first clean up the APIs of coerce_to_domain() and some internal functions in parse_coerce.c so that we consistently pass around a CoercionContext along with CoercionForm. Previously, we sometimes passed an "isExplicit" boolean flag instead, which is strictly less information; and coerce_to_domain() didn't even get that, but instead had to reverse-engineer isExplicit from CoercionForm. That's contrary to the documentation in primnodes.h that says that CoercionForm only affects display and not semantics. I don't think this change fixes any live bugs, but it makes things more consistent. The main reason for doing it though is that now build_coercion_expression() receives ccontext, which it needs in order to be able to recursively invoke coerce_to_target_type(). Next, reimplement ArrayCoerceExpr so that the node does not directly know any details of what has to be done to the individual array elements while performing the array coercion. Instead, the per-element processing is represented by a sub-expression whose input is a source array element and whose output is a target array element. This simplifies life in parse_coerce.c, because it can build that sub-expression by a recursive invocation of coerce_to_target_type(). The executor now handles the per-element processing as a compiled expression instead of hard-wired code. The main advantage of this is that we can use a single ArrayCoerceExpr to handle as many as three successive steps per element: base type conversion, typmod coercion, and domain constraint checking. The old code used two stacked ArrayCoerceExprs to handle type + typmod coercion, which was pretty inefficient, and adding yet another array deconstruction to do domain constraint checking seemed very unappetizing. In the case where we just need a single, very simple coercion function, doing this straightforwardly leads to a noticeable increase in the per-array-element runtime cost. Hence, add an additional shortcut evalfunc in execExprInterp.c that skips unnecessary overhead for that specific form of expression. The runtime speed of simple cases is within 1% or so of where it was before, while cases that previously required two levels of array processing are significantly faster. Finally, create an implicit array type for every domain type, as we do for base types, enums, etc. Everything except the array-coercion case seems to just work without further effort. Tom Lane, reviewed by Andrew Dunstan Discussion: https://postgr.es/m/9852.1499791473@sss.pgh.pa.us
2017-09-30 19:40:56 +02:00
* the full interpreter is a measurable overhead for these, and these
* patterns occur often enough to be worth optimizing.
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
*/
if (state->steps_len == 3)
{
ExprEvalOp step0 = state->steps[0].opcode;
ExprEvalOp step1 = state->steps[1].opcode;
if (step0 == EEOP_INNER_FETCHSOME &&
Perform slot validity checks in a separate pass over expression. This reduces code duplication a bit, but the primary benefit that it makes JITing expression evaluation easier. When doing so we can't, as previously done in the interpreted case, really change opcode without recompiling. Nor dow we just carry around unnecessary branches to avoid re-checking over and over. As a minor side-effect this makes ExecEvalStepOp() O(log(N)) rather than O(N). Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2017-12-29 21:38:15 +01:00
step1 == EEOP_INNER_VAR)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
Perform slot validity checks in a separate pass over expression. This reduces code duplication a bit, but the primary benefit that it makes JITing expression evaluation easier. When doing so we can't, as previously done in the interpreted case, really change opcode without recompiling. Nor dow we just carry around unnecessary branches to avoid re-checking over and over. As a minor side-effect this makes ExecEvalStepOp() O(log(N)) rather than O(N). Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2017-12-29 21:38:15 +01:00
state->evalfunc_private = (void *) ExecJustInnerVar;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
return;
}
else if (step0 == EEOP_OUTER_FETCHSOME &&
Perform slot validity checks in a separate pass over expression. This reduces code duplication a bit, but the primary benefit that it makes JITing expression evaluation easier. When doing so we can't, as previously done in the interpreted case, really change opcode without recompiling. Nor dow we just carry around unnecessary branches to avoid re-checking over and over. As a minor side-effect this makes ExecEvalStepOp() O(log(N)) rather than O(N). Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2017-12-29 21:38:15 +01:00
step1 == EEOP_OUTER_VAR)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
Perform slot validity checks in a separate pass over expression. This reduces code duplication a bit, but the primary benefit that it makes JITing expression evaluation easier. When doing so we can't, as previously done in the interpreted case, really change opcode without recompiling. Nor dow we just carry around unnecessary branches to avoid re-checking over and over. As a minor side-effect this makes ExecEvalStepOp() O(log(N)) rather than O(N). Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2017-12-29 21:38:15 +01:00
state->evalfunc_private = (void *) ExecJustOuterVar;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
return;
}
else if (step0 == EEOP_SCAN_FETCHSOME &&
Perform slot validity checks in a separate pass over expression. This reduces code duplication a bit, but the primary benefit that it makes JITing expression evaluation easier. When doing so we can't, as previously done in the interpreted case, really change opcode without recompiling. Nor dow we just carry around unnecessary branches to avoid re-checking over and over. As a minor side-effect this makes ExecEvalStepOp() O(log(N)) rather than O(N). Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2017-12-29 21:38:15 +01:00
step1 == EEOP_SCAN_VAR)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
Perform slot validity checks in a separate pass over expression. This reduces code duplication a bit, but the primary benefit that it makes JITing expression evaluation easier. When doing so we can't, as previously done in the interpreted case, really change opcode without recompiling. Nor dow we just carry around unnecessary branches to avoid re-checking over and over. As a minor side-effect this makes ExecEvalStepOp() O(log(N)) rather than O(N). Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2017-12-29 21:38:15 +01:00
state->evalfunc_private = (void *) ExecJustScanVar;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
return;
}
else if (step0 == EEOP_INNER_FETCHSOME &&
step1 == EEOP_ASSIGN_INNER_VAR)
{
Perform slot validity checks in a separate pass over expression. This reduces code duplication a bit, but the primary benefit that it makes JITing expression evaluation easier. When doing so we can't, as previously done in the interpreted case, really change opcode without recompiling. Nor dow we just carry around unnecessary branches to avoid re-checking over and over. As a minor side-effect this makes ExecEvalStepOp() O(log(N)) rather than O(N). Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2017-12-29 21:38:15 +01:00
state->evalfunc_private = (void *) ExecJustAssignInnerVar;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
return;
}
else if (step0 == EEOP_OUTER_FETCHSOME &&
step1 == EEOP_ASSIGN_OUTER_VAR)
{
Perform slot validity checks in a separate pass over expression. This reduces code duplication a bit, but the primary benefit that it makes JITing expression evaluation easier. When doing so we can't, as previously done in the interpreted case, really change opcode without recompiling. Nor dow we just carry around unnecessary branches to avoid re-checking over and over. As a minor side-effect this makes ExecEvalStepOp() O(log(N)) rather than O(N). Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2017-12-29 21:38:15 +01:00
state->evalfunc_private = (void *) ExecJustAssignOuterVar;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
return;
}
else if (step0 == EEOP_SCAN_FETCHSOME &&
step1 == EEOP_ASSIGN_SCAN_VAR)
{
Perform slot validity checks in a separate pass over expression. This reduces code duplication a bit, but the primary benefit that it makes JITing expression evaluation easier. When doing so we can't, as previously done in the interpreted case, really change opcode without recompiling. Nor dow we just carry around unnecessary branches to avoid re-checking over and over. As a minor side-effect this makes ExecEvalStepOp() O(log(N)) rather than O(N). Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2017-12-29 21:38:15 +01:00
state->evalfunc_private = (void *) ExecJustAssignScanVar;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
return;
}
Support arrays over domains. Allowing arrays with a domain type as their element type was left un-done in the original domain patch, but not for any very good reason. This omission leads to such surprising results as array_agg() not working on a domain column, because the parser can't identify a suitable output type for the polymorphic aggregate. In order to fix this, first clean up the APIs of coerce_to_domain() and some internal functions in parse_coerce.c so that we consistently pass around a CoercionContext along with CoercionForm. Previously, we sometimes passed an "isExplicit" boolean flag instead, which is strictly less information; and coerce_to_domain() didn't even get that, but instead had to reverse-engineer isExplicit from CoercionForm. That's contrary to the documentation in primnodes.h that says that CoercionForm only affects display and not semantics. I don't think this change fixes any live bugs, but it makes things more consistent. The main reason for doing it though is that now build_coercion_expression() receives ccontext, which it needs in order to be able to recursively invoke coerce_to_target_type(). Next, reimplement ArrayCoerceExpr so that the node does not directly know any details of what has to be done to the individual array elements while performing the array coercion. Instead, the per-element processing is represented by a sub-expression whose input is a source array element and whose output is a target array element. This simplifies life in parse_coerce.c, because it can build that sub-expression by a recursive invocation of coerce_to_target_type(). The executor now handles the per-element processing as a compiled expression instead of hard-wired code. The main advantage of this is that we can use a single ArrayCoerceExpr to handle as many as three successive steps per element: base type conversion, typmod coercion, and domain constraint checking. The old code used two stacked ArrayCoerceExprs to handle type + typmod coercion, which was pretty inefficient, and adding yet another array deconstruction to do domain constraint checking seemed very unappetizing. In the case where we just need a single, very simple coercion function, doing this straightforwardly leads to a noticeable increase in the per-array-element runtime cost. Hence, add an additional shortcut evalfunc in execExprInterp.c that skips unnecessary overhead for that specific form of expression. The runtime speed of simple cases is within 1% or so of where it was before, while cases that previously required two levels of array processing are significantly faster. Finally, create an implicit array type for every domain type, as we do for base types, enums, etc. Everything except the array-coercion case seems to just work without further effort. Tom Lane, reviewed by Andrew Dunstan Discussion: https://postgr.es/m/9852.1499791473@sss.pgh.pa.us
2017-09-30 19:40:56 +02:00
else if (step0 == EEOP_CASE_TESTVAL &&
step1 == EEOP_FUNCEXPR_STRICT &&
state->steps[0].d.casetest.value)
{
Perform slot validity checks in a separate pass over expression. This reduces code duplication a bit, but the primary benefit that it makes JITing expression evaluation easier. When doing so we can't, as previously done in the interpreted case, really change opcode without recompiling. Nor dow we just carry around unnecessary branches to avoid re-checking over and over. As a minor side-effect this makes ExecEvalStepOp() O(log(N)) rather than O(N). Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2017-12-29 21:38:15 +01:00
state->evalfunc_private = (void *) ExecJustApplyFuncToCase;
Support arrays over domains. Allowing arrays with a domain type as their element type was left un-done in the original domain patch, but not for any very good reason. This omission leads to such surprising results as array_agg() not working on a domain column, because the parser can't identify a suitable output type for the polymorphic aggregate. In order to fix this, first clean up the APIs of coerce_to_domain() and some internal functions in parse_coerce.c so that we consistently pass around a CoercionContext along with CoercionForm. Previously, we sometimes passed an "isExplicit" boolean flag instead, which is strictly less information; and coerce_to_domain() didn't even get that, but instead had to reverse-engineer isExplicit from CoercionForm. That's contrary to the documentation in primnodes.h that says that CoercionForm only affects display and not semantics. I don't think this change fixes any live bugs, but it makes things more consistent. The main reason for doing it though is that now build_coercion_expression() receives ccontext, which it needs in order to be able to recursively invoke coerce_to_target_type(). Next, reimplement ArrayCoerceExpr so that the node does not directly know any details of what has to be done to the individual array elements while performing the array coercion. Instead, the per-element processing is represented by a sub-expression whose input is a source array element and whose output is a target array element. This simplifies life in parse_coerce.c, because it can build that sub-expression by a recursive invocation of coerce_to_target_type(). The executor now handles the per-element processing as a compiled expression instead of hard-wired code. The main advantage of this is that we can use a single ArrayCoerceExpr to handle as many as three successive steps per element: base type conversion, typmod coercion, and domain constraint checking. The old code used two stacked ArrayCoerceExprs to handle type + typmod coercion, which was pretty inefficient, and adding yet another array deconstruction to do domain constraint checking seemed very unappetizing. In the case where we just need a single, very simple coercion function, doing this straightforwardly leads to a noticeable increase in the per-array-element runtime cost. Hence, add an additional shortcut evalfunc in execExprInterp.c that skips unnecessary overhead for that specific form of expression. The runtime speed of simple cases is within 1% or so of where it was before, while cases that previously required two levels of array processing are significantly faster. Finally, create an implicit array type for every domain type, as we do for base types, enums, etc. Everything except the array-coercion case seems to just work without further effort. Tom Lane, reviewed by Andrew Dunstan Discussion: https://postgr.es/m/9852.1499791473@sss.pgh.pa.us
2017-09-30 19:40:56 +02:00
return;
}
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
}
Don't generate EEOP_*_FETCHSOME operations for slots know to be virtual. That avoids unnecessary work during both interpreted execution, and JIT compiled expression evaluation. Both benefit from fewer expression steps needing be processed, and for interpreted execution there now is a fastpath dedicated to just fetching a value from a virtual slot. That's e.g. beneficial for hashjoins over nodes that perform projections, as the hashed columns are currently fetched individually. Author: Soumyadeep Chakraborty, Andres Freund Discussion: https://postgr.es/m/CAE-ML+9OKSN71+mHtfMD-L24oDp8dGTfaVjDU6U+j+FNAW5kRQ@mail.gmail.com
2019-10-01 01:06:16 +02:00
else if (state->steps_len == 2)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
Don't generate EEOP_*_FETCHSOME operations for slots know to be virtual. That avoids unnecessary work during both interpreted execution, and JIT compiled expression evaluation. Both benefit from fewer expression steps needing be processed, and for interpreted execution there now is a fastpath dedicated to just fetching a value from a virtual slot. That's e.g. beneficial for hashjoins over nodes that perform projections, as the hashed columns are currently fetched individually. Author: Soumyadeep Chakraborty, Andres Freund Discussion: https://postgr.es/m/CAE-ML+9OKSN71+mHtfMD-L24oDp8dGTfaVjDU6U+j+FNAW5kRQ@mail.gmail.com
2019-10-01 01:06:16 +02:00
ExprEvalOp step0 = state->steps[0].opcode;
if (step0 == EEOP_CONST)
{
state->evalfunc_private = (void *) ExecJustConst;
return;
}
else if (step0 == EEOP_INNER_VAR)
{
state->evalfunc_private = (void *) ExecJustInnerVarVirt;
return;
}
else if (step0 == EEOP_OUTER_VAR)
{
state->evalfunc_private = (void *) ExecJustOuterVarVirt;
return;
}
else if (step0 == EEOP_SCAN_VAR)
{
state->evalfunc_private = (void *) ExecJustScanVarVirt;
return;
}
else if (step0 == EEOP_ASSIGN_INNER_VAR)
{
state->evalfunc_private = (void *) ExecJustAssignInnerVarVirt;
return;
}
else if (step0 == EEOP_ASSIGN_OUTER_VAR)
{
state->evalfunc_private = (void *) ExecJustAssignOuterVarVirt;
return;
}
else if (step0 == EEOP_ASSIGN_SCAN_VAR)
{
state->evalfunc_private = (void *) ExecJustAssignScanVarVirt;
return;
}
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
}
#if defined(EEO_USE_COMPUTED_GOTO)
/*
* In the direct-threaded implementation, replace each opcode with the
* address to jump to. (Use ExecEvalStepOp() to get back the opcode.)
*/
expression eval, jit: Minor code cleanups. This mostly consists of using C99 style for loops, moving variables into narrower scopes, and a smattering of other minor improvements. Done separately to make it easier to review patches with actual functional changes. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:04:50 +01:00
for (int off = 0; off < state->steps_len; off++)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
expression eval, jit: Minor code cleanups. This mostly consists of using C99 style for loops, moving variables into narrower scopes, and a smattering of other minor improvements. Done separately to make it easier to review patches with actual functional changes. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:04:50 +01:00
ExprEvalStep *op = &state->steps[off];
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
expression eval, jit: Minor code cleanups. This mostly consists of using C99 style for loops, moving variables into narrower scopes, and a smattering of other minor improvements. Done separately to make it easier to review patches with actual functional changes. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:04:50 +01:00
op->opcode = EEO_OPCODE(op->opcode);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
}
expression eval, jit: Minor code cleanups. This mostly consists of using C99 style for loops, moving variables into narrower scopes, and a smattering of other minor improvements. Done separately to make it easier to review patches with actual functional changes. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:04:50 +01:00
state->flags |= EEO_FLAG_DIRECT_THREADED;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
#endif /* EEO_USE_COMPUTED_GOTO */
Perform slot validity checks in a separate pass over expression. This reduces code duplication a bit, but the primary benefit that it makes JITing expression evaluation easier. When doing so we can't, as previously done in the interpreted case, really change opcode without recompiling. Nor dow we just carry around unnecessary branches to avoid re-checking over and over. As a minor side-effect this makes ExecEvalStepOp() O(log(N)) rather than O(N). Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2017-12-29 21:38:15 +01:00
state->evalfunc_private = (void *) ExecInterpExpr;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
}
/*
* Evaluate expression identified by "state" in the execution context
* given by "econtext". *isnull is set to the is-null flag for the result,
* and the Datum value is the function result.
*
* As a special case, return the dispatch table's address if state is NULL.
* This is used by ExecInitInterpreter to set up the dispatch_table global.
* (Only applies when EEO_USE_COMPUTED_GOTO is defined.)
*/
static Datum
ExecInterpExpr(ExprState *state, ExprContext *econtext, bool *isnull)
{
ExprEvalStep *op;
TupleTableSlot *resultslot;
TupleTableSlot *innerslot;
TupleTableSlot *outerslot;
TupleTableSlot *scanslot;
/*
* This array has to be in the same order as enum ExprEvalOp.
*/
#if defined(EEO_USE_COMPUTED_GOTO)
static const void *const dispatch_table[] = {
&&CASE_EEOP_DONE,
&&CASE_EEOP_INNER_FETCHSOME,
&&CASE_EEOP_OUTER_FETCHSOME,
&&CASE_EEOP_SCAN_FETCHSOME,
&&CASE_EEOP_INNER_VAR,
&&CASE_EEOP_OUTER_VAR,
&&CASE_EEOP_SCAN_VAR,
&&CASE_EEOP_INNER_SYSVAR,
&&CASE_EEOP_OUTER_SYSVAR,
&&CASE_EEOP_SCAN_SYSVAR,
&&CASE_EEOP_WHOLEROW,
&&CASE_EEOP_ASSIGN_INNER_VAR,
&&CASE_EEOP_ASSIGN_OUTER_VAR,
&&CASE_EEOP_ASSIGN_SCAN_VAR,
&&CASE_EEOP_ASSIGN_TMP,
&&CASE_EEOP_ASSIGN_TMP_MAKE_RO,
&&CASE_EEOP_CONST,
&&CASE_EEOP_FUNCEXPR,
&&CASE_EEOP_FUNCEXPR_STRICT,
&&CASE_EEOP_FUNCEXPR_FUSAGE,
&&CASE_EEOP_FUNCEXPR_STRICT_FUSAGE,
&&CASE_EEOP_BOOL_AND_STEP_FIRST,
&&CASE_EEOP_BOOL_AND_STEP,
&&CASE_EEOP_BOOL_AND_STEP_LAST,
&&CASE_EEOP_BOOL_OR_STEP_FIRST,
&&CASE_EEOP_BOOL_OR_STEP,
&&CASE_EEOP_BOOL_OR_STEP_LAST,
&&CASE_EEOP_BOOL_NOT_STEP,
&&CASE_EEOP_QUAL,
&&CASE_EEOP_JUMP,
&&CASE_EEOP_JUMP_IF_NULL,
&&CASE_EEOP_JUMP_IF_NOT_NULL,
&&CASE_EEOP_JUMP_IF_NOT_TRUE,
&&CASE_EEOP_NULLTEST_ISNULL,
&&CASE_EEOP_NULLTEST_ISNOTNULL,
&&CASE_EEOP_NULLTEST_ROWISNULL,
&&CASE_EEOP_NULLTEST_ROWISNOTNULL,
&&CASE_EEOP_BOOLTEST_IS_TRUE,
&&CASE_EEOP_BOOLTEST_IS_NOT_TRUE,
&&CASE_EEOP_BOOLTEST_IS_FALSE,
&&CASE_EEOP_BOOLTEST_IS_NOT_FALSE,
&&CASE_EEOP_PARAM_EXEC,
&&CASE_EEOP_PARAM_EXTERN,
Rearrange execution of PARAM_EXTERN Params for plpgsql's benefit. This patch does three interrelated things: * Create a new expression execution step type EEOP_PARAM_CALLBACK and add the infrastructure needed for add-on modules to generate that. As discussed, the best control mechanism for that seems to be to add another hook function to ParamListInfo, which will be called by ExecInitExpr if it's supplied and a PARAM_EXTERN Param is found. For stand-alone expressions, we add a new entry point to allow the ParamListInfo to be specified directly, since it can't be retrieved from the parent plan node's EState. * Redesign the API for the ParamListInfo paramFetch hook so that the ParamExternData array can be entirely virtual. This also lets us get rid of ParamListInfo.paramMask, instead leaving it to the paramFetch hook to decide which param IDs should be accessible or not. plpgsql_param_fetch was already doing the identical masking check, so having callers do it too seemed redundant. While I was at it, I added a "speculative" flag to paramFetch that the planner can specify as TRUE to avoid unwanted failures. This solves an ancient problem for plpgsql that it couldn't provide values of non-DTYPE_VAR variables to the planner for fear of triggering premature "record not assigned yet" or "field not found" errors during planning. * Rework plpgsql to get rid of the need for "unshared" parameter lists, by dint of turning the single ParamListInfo per estate into a nearly read-only data structure that doesn't instantiate any per-variable data. Instead, the paramFetch hook controls access to per-variable data and can make the right decisions on the fly, replacing the cases that we used to need multiple ParamListInfos for. This might perhaps have been a performance loss on its own, but by using a paramCompile hook we can bypass plpgsql_param_fetch entirely during normal query execution. (It's now only called when, eg, we copy the ParamListInfo into a cursor portal. copyParamList() or SerializeParamList() effectively instantiate the virtual parameter array as a simple physical array without a paramFetch hook, which is what we want in those cases.) This allows reverting most of commit 6c82d8d1f, though I kept the cosmetic code-consolidation aspects of that (eg the assign_simple_var function). Performance testing shows this to be at worst a break-even change, and it can provide wins ranging up to 20% in test cases involving accesses to fields of "record" variables. The fact that values of such variables can now be exposed to the planner might produce wins in some situations, too, but I've not pursued that angle. In passing, remove the "parent" pointer from the arguments to ExecInitExprRec and related functions, instead storing that pointer in a transient field in ExprState. The ParamListInfo pointer for a stand-alone expression is handled the same way; we'd otherwise have had to add yet another recursively-passed-down argument in expression compilation. Discussion: https://postgr.es/m/32589.1513706441@sss.pgh.pa.us
2017-12-21 18:57:41 +01:00
&&CASE_EEOP_PARAM_CALLBACK,
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
&&CASE_EEOP_CASE_TESTVAL,
&&CASE_EEOP_MAKE_READONLY,
&&CASE_EEOP_IOCOERCE,
&&CASE_EEOP_DISTINCT,
Do execGrouping.c via expression eval machinery, take two. This has a performance benefit on own, although not hugely so. The primary benefit is that it will allow for to JIT tuple deforming and comparator invocations. Large parts of this were previously committed (773aec7aa), but the commit contained an omission around cross-type comparisons and was thus reverted. Author: Andres Freund Discussion: https://postgr.es/m/20171129080934.amqqkke2zjtekd4t@alap3.anarazel.de
2018-02-16 06:55:31 +01:00
&&CASE_EEOP_NOT_DISTINCT,
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
&&CASE_EEOP_NULLIF,
Add back SQLValueFunction for SQL keywords This is equivalent to a revert of f193883 and fb32748, with the addition that the declaration of the SQLValueFunction node needs to gain a couple of node_attr for query jumbling. The performance impact of removing the function call inlining is proving to be too huge for some workloads where these are used. A worst-case test case of involving only simple SELECT queries with a SQL keyword is proving to lead to a reduction of 10% in TPS via pgbench and prepared queries on a high-end machine. None of the tests I ran back for this set of changes saw such a huge gap, but Alexander Lakhin and Andres Freund have found that this can be noticeable. Keeping the older performance would mean to do more inlining in the executor when using COERCE_SQL_SYNTAX for a function expression, similarly to what SQLValueFunction does. This requires more redesign work and there is little time until 16beta1 is released, so for now reverting the change is the best way forward, bringing back the previous performance. Bump catalog version. Reported-by: Alexander Lakhin Discussion: https://postgr.es/m/b32bed1b-0746-9b20-1472-4bdc9ca66d52@gmail.com
2023-05-17 03:19:17 +02:00
&&CASE_EEOP_SQLVALUEFUNCTION,
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
&&CASE_EEOP_CURRENTOFEXPR,
Identity columns This is the SQL standard-conforming variant of PostgreSQL's serial columns. It fixes a few usability issues that serial columns have: - CREATE TABLE / LIKE copies default but refers to same sequence - cannot add/drop serialness with ALTER TABLE - dropping default does not drop sequence - need to grant separate privileges to sequence - other slight weirdnesses because serial is some kind of special macro Reviewed-by: Vitaly Burovoy <vitaly.burovoy@gmail.com>
2017-04-06 14:33:16 +02:00
&&CASE_EEOP_NEXTVALUEEXPR,
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
&&CASE_EEOP_ARRAYEXPR,
&&CASE_EEOP_ARRAYCOERCE,
&&CASE_EEOP_ROW,
&&CASE_EEOP_ROWCOMPARE_STEP,
&&CASE_EEOP_ROWCOMPARE_FINAL,
&&CASE_EEOP_MINMAX,
&&CASE_EEOP_FIELDSELECT,
&&CASE_EEOP_FIELDSTORE_DEFORM,
&&CASE_EEOP_FIELDSTORE_FORM,
Support subscripting of arbitrary types, not only arrays. This patch generalizes the subscripting infrastructure so that any data type can be subscripted, if it provides a handler function to define what that means. Traditional variable-length (varlena) arrays all use array_subscript_handler(), while the existing fixed-length types that support subscripting use raw_array_subscript_handler(). It's expected that other types that want to use subscripting notation will define their own handlers. (This patch provides no such new features, though; it only lays the foundation for them.) To do this, move the parser's semantic processing of subscripts (including coercion to whatever data type is required) into a method callback supplied by the handler. On the execution side, replace the ExecEvalSubscriptingRef* layer of functions with direct calls to callback-supplied execution routines. (Thus, essentially no new run-time overhead should be caused by this patch. Indeed, there is room to remove some overhead by supplying specialized execution routines. This patch does a little bit in that line, but more could be done.) Additional work is required here and there to remove formerly hard-wired assumptions about the result type, collation, etc of a SubscriptingRef expression node; and to remove assumptions that the subscript values must be integers. One useful side-effect of this is that we now have a less squishy mechanism for identifying whether a data type is a "true" array: instead of wiring in weird rules about typlen, we can look to see if pg_type.typsubscript == F_ARRAY_SUBSCRIPT_HANDLER. For this to be bulletproof, we have to forbid user-defined types from using that handler directly; but there seems no good reason for them to do so. This patch also removes assumptions that the number of subscripts is limited to MAXDIM (6), or indeed has any hard-wired limit. That limit still applies to types handled by array_subscript_handler or raw_array_subscript_handler, but to discourage other dependencies on this constant, I've moved it from c.h to utils/array.h. Dmitry Dolgov, reviewed at various times by Tom Lane, Arthur Zakirov, Peter Eisentraut, Pavel Stehule Discussion: https://postgr.es/m/CA+q6zcVDuGBv=M0FqBYX8DPebS3F_0KQ6OVFobGJPM507_SZ_w@mail.gmail.com Discussion: https://postgr.es/m/CA+q6zcVovR+XY4mfk-7oNk-rF91gH0PebnNfuUjuuDsyHjOcVA@mail.gmail.com
2020-12-09 18:40:37 +01:00
&&CASE_EEOP_SBSREF_SUBSCRIPTS,
Renaming for new subscripting mechanism Over at patch https://commitfest.postgresql.org/21/1062/ Dmitry wants to introduce a more generic subscription mechanism, which allows subscripting not only arrays but also other object types such as JSONB. That functionality is introduced in a largish invasive patch, out of which this internal renaming patch was extracted. Author: Dmitry Dolgov Reviewed-by: Tom Lane, Arthur Zakirov Discussion: https://postgr.es/m/CA+q6zcUK4EqPAu7XRRO5CCjMwhz5zvg+rfWuLzVoxp_5sKS6=w@mail.gmail.com
2019-02-01 16:50:32 +01:00
&&CASE_EEOP_SBSREF_OLD,
&&CASE_EEOP_SBSREF_ASSIGN,
&&CASE_EEOP_SBSREF_FETCH,
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
&&CASE_EEOP_DOMAIN_TESTVAL,
&&CASE_EEOP_DOMAIN_NOTNULL,
&&CASE_EEOP_DOMAIN_CHECK,
&&CASE_EEOP_CONVERT_ROWTYPE,
&&CASE_EEOP_SCALARARRAYOP,
Speedup ScalarArrayOpExpr evaluation ScalarArrayOpExprs with "useOr=true" and a set of Consts on the righthand side have traditionally been evaluated by using a linear search over the array. When these arrays contain large numbers of elements then this linear search could become a significant part of execution time. Here we add a new method of evaluating ScalarArrayOpExpr expressions to allow them to be evaluated by first building a hash table containing each element, then on subsequent evaluations, we just probe that hash table to determine if there is a match. The planner is in charge of determining when this optimization is possible and it enables it by setting hashfuncid in the ScalarArrayOpExpr. The executor will only perform the hash table evaluation when the hashfuncid is set. This means that not all cases are optimized. For example CHECK constraints containing an IN clause won't go through the planner, so won't get the hashfuncid set. We could maybe do something about that at some later date. The reason we're not doing it now is from fear that we may slow down cases where the expression is evaluated only once. Those cases can be common, for example, a single row INSERT to a table with a CHECK constraint containing an IN clause. In the planner, we enable this when there are suitable hash functions for the ScalarArrayOpExpr's operator and only when there is at least MIN_ARRAY_SIZE_FOR_HASHED_SAOP elements in the array. The threshold is currently set to 9. Author: James Coleman, David Rowley Reviewed-by: David Rowley, Tomas Vondra, Heikki Linnakangas Discussion: https://postgr.es/m/CAAaqYe8x62+=wn0zvNKCj55tPpg-JBHzhZFFc6ANovdqFw7-dA@mail.gmail.com
2021-04-08 13:51:22 +02:00
&&CASE_EEOP_HASHED_SCALARARRAYOP,
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
&&CASE_EEOP_XMLEXPR,
SQL/JSON: add standard JSON constructor functions This commit introduces the SQL/JSON standard-conforming constructors for JSON types: JSON_ARRAY() JSON_ARRAYAGG() JSON_OBJECT() JSON_OBJECTAGG() Most of the functionality was already present in PostgreSQL-specific functions, but these include some new functionality such as the ability to skip or include NULL values, and to allow duplicate keys or throw error when they are found, as well as the standard specified syntax to specify output type and format. Author: Nikita Glukhov <n.gluhov@postgrespro.ru> Author: Teodor Sigaev <teodor@sigaev.ru> Author: Oleg Bartunov <obartunov@gmail.com> Author: Alexander Korotkov <aekorotkov@gmail.com> Author: Amit Langote <amitlangote09@gmail.com> Reviewers have included (in no particular order) Andres Freund, Alexander Korotkov, Pavel Stehule, Andrew Alsup, Erik Rijkers, Zihong Yu, Himanshu Upadhyaya, Daniel Gustafsson, Justin Pryzby. Discussion: https://postgr.es/m/CAF4Au4w2x-5LTnN_bxky-mq4=WOqsGsxSpENCzHRAzSnEd8+WQ@mail.gmail.com Discussion: https://postgr.es/m/cd0bb935-0158-78a7-08b5-904886deac4b@postgrespro.ru Discussion: https://postgr.es/m/20220616233130.rparivafipt6doj3@alap3.anarazel.de Discussion: https://postgr.es/m/abd9b83b-aa66-f230-3d6d-734817f0995d%40postgresql.org
2023-03-29 12:11:36 +02:00
&&CASE_EEOP_JSON_CONSTRUCTOR,
SQL/JSON: support the IS JSON predicate This patch introduces the SQL standard IS JSON predicate. It operates on text and bytea values representing JSON, as well as on the json and jsonb types. Each test has IS and IS NOT variants and supports a WITH UNIQUE KEYS flag. The tests are: IS JSON [VALUE] IS JSON ARRAY IS JSON OBJECT IS JSON SCALAR These should be self-explanatory. The WITH UNIQUE KEYS flag makes these return false when duplicate keys exist in any object within the value, not necessarily directly contained in the outermost object. Author: Nikita Glukhov <n.gluhov@postgrespro.ru> Author: Teodor Sigaev <teodor@sigaev.ru> Author: Oleg Bartunov <obartunov@gmail.com> Author: Alexander Korotkov <aekorotkov@gmail.com> Author: Amit Langote <amitlangote09@gmail.com> Author: Andrew Dunstan <andrew@dunslane.net> Reviewers have included (in no particular order) Andres Freund, Alexander Korotkov, Pavel Stehule, Andrew Alsup, Erik Rijkers, Zihong Yu, Himanshu Upadhyaya, Daniel Gustafsson, Justin Pryzby. Discussion: https://postgr.es/m/CAF4Au4w2x-5LTnN_bxky-mq4=WOqsGsxSpENCzHRAzSnEd8+WQ@mail.gmail.com Discussion: https://postgr.es/m/cd0bb935-0158-78a7-08b5-904886deac4b@postgrespro.ru Discussion: https://postgr.es/m/20220616233130.rparivafipt6doj3@alap3.anarazel.de Discussion: https://postgr.es/m/abd9b83b-aa66-f230-3d6d-734817f0995d%40postgresql.org
2023-03-31 22:34:04 +02:00
&&CASE_EEOP_IS_JSON,
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
&&CASE_EEOP_AGGREF,
&&CASE_EEOP_GROUPING_FUNC,
&&CASE_EEOP_WINDOW_FUNC,
&&CASE_EEOP_SUBPLAN,
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
&&CASE_EEOP_AGG_STRICT_DESERIALIZE,
&&CASE_EEOP_AGG_DESERIALIZE,
Change function call information to be variable length. Before this change FunctionCallInfoData, the struct arguments etc for V1 function calls are stored in, always had space for FUNC_MAX_ARGS/100 arguments, storing datums and their nullness in two arrays. For nearly every function call 100 arguments is far more than needed, therefore wasting memory. Arg and argnull being two separate arrays also guarantees that to access a single argument, two cachelines have to be touched. Change the layout so there's a single variable-length array with pairs of value / isnull. That drastically reduces memory consumption for most function calls (on x86-64 a two argument function now uses 64bytes, previously 936 bytes), and makes it very likely that argument value and its nullness are on the same cacheline. Arguments are stored in a new NullableDatum struct, which, due to padding, needs more memory per argument than before. But as usually far fewer arguments are stored, and individual arguments are cheaper to access, that's still a clear win. It's likely that there's other places where conversion to NullableDatum arrays would make sense, e.g. TupleTableSlots, but that's for another commit. Because the function call information is now variable-length allocations have to take the number of arguments into account. For heap allocations that can be done with SizeForFunctionCallInfoData(), for on-stack allocations there's a new LOCAL_FCINFO(name, nargs) macro that helps to allocate an appropriately sized and aligned variable. Some places with stack allocation function call information don't know the number of arguments at compile time, and currently variably sized stack allocations aren't allowed in postgres. Therefore allow for FUNC_MAX_ARGS space in these cases. They're not that common, so for now that seems acceptable. Because of the need to allocate FunctionCallInfo of the appropriate size, older extensions may need to update their code. To avoid subtle breakages, the FunctionCallInfoData struct has been renamed to FunctionCallInfoBaseData. Most code only references FunctionCallInfo, so that shouldn't cause much collateral damage. This change is also a prerequisite for more efficient expression JIT compilation (by allocating the function call information on the stack, allowing LLVM to optimize it away); previously the size of the call information caused problems inside LLVM's optimizer. Author: Andres Freund Reviewed-By: Tom Lane Discussion: https://postgr.es/m/20180605172952.x34m5uz6ju6enaem@alap3.anarazel.de
2019-01-26 23:17:52 +01:00
&&CASE_EEOP_AGG_STRICT_INPUT_CHECK_ARGS,
&&CASE_EEOP_AGG_STRICT_INPUT_CHECK_NULLS,
Extend ExecBuildAggTrans() to support a NULL pointer check. Optionally push a step to check for a NULL pointer to the pergroup state. This will be important for disk-based hash aggregation in combination with grouping sets. When memory limits are reached, a given tuple may find its per-group state for some grouping sets but not others. For the former, it advances the per-group state as normal; for the latter, it skips evaluation and the calling code will have to spill the tuple and reprocess it in a later batch. Add the NULL check as a separate expression step because in some common cases it's not needed. Discussion: https://postgr.es/m/20200221202212.ssb2qpmdgrnx52sj%40alap3.anarazel.de
2020-03-05 02:20:20 +01:00
&&CASE_EEOP_AGG_PLAIN_PERGROUP_NULLCHECK,
expression eval: Reduce number of steps for agg transition invocations. Do so by combining the various steps that are part of aggregate transition function invocation into one larger step. As some of the current steps are only necessary for some aggregates, have one variant of the aggregate transition step for each possible combination. To avoid further manual copies of code in the different transition step implementations, move most of the code into helper functions marked as "always inline". The benefit of this change is an increase in performance when aggregating lots of rows. This comes in part due to the reduced number of indirect jumps due to the reduced number of steps, and in part by reducing redundant setup code across steps. This mainly benefits interpreted execution, but the code generated by JIT is also improved a bit. As a nice side-effect it also ends up making the code a bit simpler. A small additional optimization is removing the need to set aggstate->curaggcontext before calling ExecAggInitGroup, choosing to instead passign curaggcontext as an argument. It was, in contrast to other aggregate related functions, only needed to fetch a memory context to copy the transition value into. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de https://postgr.es/m/5c371df7cee903e8cd4c685f90c6c72086d3a2dc.camel@j-davis.com
2020-02-24 23:39:22 +01:00
&&CASE_EEOP_AGG_PLAIN_TRANS_INIT_STRICT_BYVAL,
&&CASE_EEOP_AGG_PLAIN_TRANS_STRICT_BYVAL,
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
&&CASE_EEOP_AGG_PLAIN_TRANS_BYVAL,
expression eval: Reduce number of steps for agg transition invocations. Do so by combining the various steps that are part of aggregate transition function invocation into one larger step. As some of the current steps are only necessary for some aggregates, have one variant of the aggregate transition step for each possible combination. To avoid further manual copies of code in the different transition step implementations, move most of the code into helper functions marked as "always inline". The benefit of this change is an increase in performance when aggregating lots of rows. This comes in part due to the reduced number of indirect jumps due to the reduced number of steps, and in part by reducing redundant setup code across steps. This mainly benefits interpreted execution, but the code generated by JIT is also improved a bit. As a nice side-effect it also ends up making the code a bit simpler. A small additional optimization is removing the need to set aggstate->curaggcontext before calling ExecAggInitGroup, choosing to instead passign curaggcontext as an argument. It was, in contrast to other aggregate related functions, only needed to fetch a memory context to copy the transition value into. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de https://postgr.es/m/5c371df7cee903e8cd4c685f90c6c72086d3a2dc.camel@j-davis.com
2020-02-24 23:39:22 +01:00
&&CASE_EEOP_AGG_PLAIN_TRANS_INIT_STRICT_BYREF,
&&CASE_EEOP_AGG_PLAIN_TRANS_STRICT_BYREF,
&&CASE_EEOP_AGG_PLAIN_TRANS_BYREF,
Improve performance of ORDER BY / DISTINCT aggregates ORDER BY / DISTINCT aggreagtes have, since implemented in Postgres, been executed by always performing a sort in nodeAgg.c to sort the tuples in the current group into the correct order before calling the transition function on the sorted tuples. This was not great as often there might be an index that could have provided pre-sorted input and allowed the transition functions to be called as the rows come in, rather than having to store them in a tuplestore in order to sort them once all the tuples for the group have arrived. Here we change the planner so it requests a path with a sort order which supports the most amount of ORDER BY / DISTINCT aggregate functions and add new code to the executor to allow it to support the processing of ORDER BY / DISTINCT aggregates where the tuples are already sorted in the correct order. Since there can be many ORDER BY / DISTINCT aggregates in any given query level, it's very possible that we can't find an order that suits all of these aggregates. The sort order that the planner chooses is simply the one that suits the most aggregate functions. We take the most strictly sorted variation of each order and see how many aggregate functions can use that, then we try again with the order of the remaining aggregates to see if another order would suit more aggregate functions. For example: SELECT agg(a ORDER BY a),agg2(a ORDER BY a,b) ... would request the sort order to be {a, b} because {a} is a subset of the sort order of {a,b}, but; SELECT agg(a ORDER BY a),agg2(a ORDER BY c) ... would just pick a plan ordered by {a} (we give precedence to aggregates which are earlier in the targetlist). SELECT agg(a ORDER BY a),agg2(a ORDER BY b),agg3(a ORDER BY b) ... would choose to order by {b} since two aggregates suit that vs just one that requires input ordered by {a}. Author: David Rowley Reviewed-by: Ronan Dunklau, James Coleman, Ranier Vilela, Richard Guo, Tom Lane Discussion: https://postgr.es/m/CAApHDvpHzfo92%3DR4W0%2BxVua3BUYCKMckWAmo-2t_KiXN-wYH%3Dw%40mail.gmail.com
2022-08-02 13:11:45 +02:00
&&CASE_EEOP_AGG_PRESORTED_DISTINCT_SINGLE,
&&CASE_EEOP_AGG_PRESORTED_DISTINCT_MULTI,
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
&&CASE_EEOP_AGG_ORDERED_TRANS_DATUM,
&&CASE_EEOP_AGG_ORDERED_TRANS_TUPLE,
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
&&CASE_EEOP_LAST
};
Static assertions cleanup Because we added StaticAssertStmt() first before StaticAssertDecl(), some uses as well as the instructions in c.h are now a bit backwards from the "native" way static assertions are meant to be used in C. This updates the guidance and moves some static assertions to better places. Specifically, since the addition of StaticAssertDecl(), we can put static assertions at the file level. This moves a number of static assertions out of function bodies, where they might have been stuck out of necessity, to perhaps better places at the file level or in header files. Also, when the static assertion appears in a position where a declaration is allowed, then using StaticAssertDecl() is more native than StaticAssertStmt(). Reviewed-by: John Naylor <john.naylor@enterprisedb.com> Discussion: https://www.postgresql.org/message-id/flat/941a04e7-dd6f-c0e4-8cdf-a33b3338cbda%40enterprisedb.com
2022-12-08 14:30:01 +01:00
StaticAssertDecl(lengthof(dispatch_table) == EEOP_LAST + 1,
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
"dispatch_table out of whack with ExprEvalOp");
if (unlikely(state == NULL))
return PointerGetDatum(dispatch_table);
#else
Assert(state != NULL);
#endif /* EEO_USE_COMPUTED_GOTO */
/* setup state */
op = state->steps;
resultslot = state->resultslot;
innerslot = econtext->ecxt_innertuple;
outerslot = econtext->ecxt_outertuple;
scanslot = econtext->ecxt_scantuple;
#if defined(EEO_USE_COMPUTED_GOTO)
EEO_DISPATCH();
#endif
EEO_SWITCH()
{
EEO_CASE(EEOP_DONE)
{
goto out;
}
EEO_CASE(EEOP_INNER_FETCHSOME)
{
Verify that expected slot types match returned slot types. This is important so JIT compilation knows what kind of tuple slot the deforming routine can expect. There's also optimization potential for expression initialization without JIT compilation. It e.g. seems plausible to elide EEOP_*_FETCHSOME ops entirely when dealing with virtual slots. Author: Andres Freund Discussion: https://postgr.es/m/20181105210039.hh4vvi4vwoq5ba2q@alap3.anarazel.de
2018-11-16 07:00:30 +01:00
CheckOpSlotCompatibility(op, innerslot);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
slot_getsomeattrs(innerslot, op->d.fetch.last_var);
EEO_NEXT();
}
EEO_CASE(EEOP_OUTER_FETCHSOME)
{
Verify that expected slot types match returned slot types. This is important so JIT compilation knows what kind of tuple slot the deforming routine can expect. There's also optimization potential for expression initialization without JIT compilation. It e.g. seems plausible to elide EEOP_*_FETCHSOME ops entirely when dealing with virtual slots. Author: Andres Freund Discussion: https://postgr.es/m/20181105210039.hh4vvi4vwoq5ba2q@alap3.anarazel.de
2018-11-16 07:00:30 +01:00
CheckOpSlotCompatibility(op, outerslot);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
slot_getsomeattrs(outerslot, op->d.fetch.last_var);
EEO_NEXT();
}
EEO_CASE(EEOP_SCAN_FETCHSOME)
{
Verify that expected slot types match returned slot types. This is important so JIT compilation knows what kind of tuple slot the deforming routine can expect. There's also optimization potential for expression initialization without JIT compilation. It e.g. seems plausible to elide EEOP_*_FETCHSOME ops entirely when dealing with virtual slots. Author: Andres Freund Discussion: https://postgr.es/m/20181105210039.hh4vvi4vwoq5ba2q@alap3.anarazel.de
2018-11-16 07:00:30 +01:00
CheckOpSlotCompatibility(op, scanslot);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
slot_getsomeattrs(scanslot, op->d.fetch.last_var);
EEO_NEXT();
}
EEO_CASE(EEOP_INNER_VAR)
{
int attnum = op->d.var.attnum;
/*
* Since we already extracted all referenced columns from the
* tuple with a FETCHSOME step, we can just grab the value
* directly out of the slot's decomposed-data arrays. But let's
* have an Assert to check that that did happen.
*/
Assert(attnum >= 0 && attnum < innerslot->tts_nvalid);
*op->resvalue = innerslot->tts_values[attnum];
*op->resnull = innerslot->tts_isnull[attnum];
EEO_NEXT();
}
EEO_CASE(EEOP_OUTER_VAR)
{
int attnum = op->d.var.attnum;
/* See EEOP_INNER_VAR comments */
Assert(attnum >= 0 && attnum < outerslot->tts_nvalid);
*op->resvalue = outerslot->tts_values[attnum];
*op->resnull = outerslot->tts_isnull[attnum];
EEO_NEXT();
}
EEO_CASE(EEOP_SCAN_VAR)
{
int attnum = op->d.var.attnum;
/* See EEOP_INNER_VAR comments */
Assert(attnum >= 0 && attnum < scanslot->tts_nvalid);
*op->resvalue = scanslot->tts_values[attnum];
*op->resnull = scanslot->tts_isnull[attnum];
EEO_NEXT();
}
EEO_CASE(EEOP_INNER_SYSVAR)
{
Move EEOP_*_SYSVAR evaluation out of line. This mainly de-duplicates code. As evaluating a system variable isn't the hottest path and the current inline implementation ends up calling out to an external function anyway, this is OK from a performance POV. The main motivation for de-duplicating is the upcoming slot abstraction work, after which there's not guaranteed to be a HeapTuple backing the slot. Author: Andres Freund, Amit Khandekar Discussion: https://postgr.es/m/20181105210039.hh4vvi4vwoq5ba2q@alap3.anarazel.de
2018-11-07 20:08:45 +01:00
ExecEvalSysVar(state, op, econtext, innerslot);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
EEO_NEXT();
}
EEO_CASE(EEOP_OUTER_SYSVAR)
{
Move EEOP_*_SYSVAR evaluation out of line. This mainly de-duplicates code. As evaluating a system variable isn't the hottest path and the current inline implementation ends up calling out to an external function anyway, this is OK from a performance POV. The main motivation for de-duplicating is the upcoming slot abstraction work, after which there's not guaranteed to be a HeapTuple backing the slot. Author: Andres Freund, Amit Khandekar Discussion: https://postgr.es/m/20181105210039.hh4vvi4vwoq5ba2q@alap3.anarazel.de
2018-11-07 20:08:45 +01:00
ExecEvalSysVar(state, op, econtext, outerslot);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
EEO_NEXT();
}
EEO_CASE(EEOP_SCAN_SYSVAR)
{
Move EEOP_*_SYSVAR evaluation out of line. This mainly de-duplicates code. As evaluating a system variable isn't the hottest path and the current inline implementation ends up calling out to an external function anyway, this is OK from a performance POV. The main motivation for de-duplicating is the upcoming slot abstraction work, after which there's not guaranteed to be a HeapTuple backing the slot. Author: Andres Freund, Amit Khandekar Discussion: https://postgr.es/m/20181105210039.hh4vvi4vwoq5ba2q@alap3.anarazel.de
2018-11-07 20:08:45 +01:00
ExecEvalSysVar(state, op, econtext, scanslot);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
EEO_NEXT();
}
EEO_CASE(EEOP_WHOLEROW)
{
/* too complex for an inline implementation */
ExecEvalWholeRowVar(state, op, econtext);
EEO_NEXT();
}
EEO_CASE(EEOP_ASSIGN_INNER_VAR)
{
int resultnum = op->d.assign_var.resultnum;
int attnum = op->d.assign_var.attnum;
/*
* We do not need CheckVarSlotCompatibility here; that was taken
* care of at compilation time. But see EEOP_INNER_VAR comments.
*/
Assert(attnum >= 0 && attnum < innerslot->tts_nvalid);
Fix mishandling of resjunk columns in ON CONFLICT ... UPDATE tlists. It's unusual to have any resjunk columns in an ON CONFLICT ... UPDATE list, but it can happen when MULTIEXPR_SUBLINK SubPlans are present. If it happens, the ON CONFLICT UPDATE code path would end up storing tuples that include the values of the extra resjunk columns. That's fairly harmless in the short run, but if new columns are added to the table then the values would become accessible, possibly leading to malfunctions if they don't match the datatypes of the new columns. This had escaped notice through a confluence of missing sanity checks, including * There's no cross-check that a tuple presented to heap_insert or heap_update matches the table rowtype. While it's difficult to check that fully at reasonable cost, we can easily add assertions that there aren't too many columns. * The output-column-assignment cases in execExprInterp.c lacked any sanity checks on the output column numbers, which seems like an oversight considering there are plenty of assertion checks on input column numbers. Add assertions there too. * We failed to apply nodeModifyTable's ExecCheckPlanOutput() to the ON CONFLICT UPDATE tlist. That wouldn't have caught this specific error, since that function is chartered to ignore resjunk columns; but it sure seems like a bad omission now that we've seen this bug. In HEAD, the right way to fix this is to make the processing of ON CONFLICT UPDATE tlists work the same as regular UPDATE tlists now do, that is don't add "SET x = x" entries, and use ExecBuildUpdateProjection to evaluate the tlist and combine it with old values of the not-set columns. This adds a little complication to ExecBuildUpdateProjection, but allows removal of a comparable amount of now-dead code from the planner. In the back branches, the most expedient solution seems to be to (a) use an output slot for the ON CONFLICT UPDATE projection that actually matches the target table, and then (b) invent a variant of ExecBuildProjectionInfo that can be told to not store values resulting from resjunk columns, so it doesn't try to store into nonexistent columns of the output slot. (We can't simply ignore the resjunk columns altogether; they have to be evaluated for MULTIEXPR_SUBLINK to work.) This works back to v10. In 9.6, projections work much differently and we can't cheaply give them such an option. The 9.6 version of this patch works by inserting a JunkFilter when it's necessary to get rid of resjunk columns. In addition, v11 and up have the reverse problem when trying to perform ON CONFLICT UPDATE on a partitioned table. Through a further oversight, adjust_partition_tlist() discarded resjunk columns when re-ordering the ON CONFLICT UPDATE tlist to match a partition. This accidentally prevented the storing-bogus-tuples problem, but at the cost that MULTIEXPR_SUBLINK cases didn't work, typically crashing if more than one row has to be updated. Fix by preserving resjunk columns in that routine. (I failed to resist the temptation to add more assertions there too, and to do some minor code beautification.) Per report from Andres Freund. Back-patch to all supported branches. Security: CVE-2021-32028
2021-05-10 17:02:29 +02:00
Assert(resultnum >= 0 && resultnum < resultslot->tts_tupleDescriptor->natts);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
resultslot->tts_values[resultnum] = innerslot->tts_values[attnum];
resultslot->tts_isnull[resultnum] = innerslot->tts_isnull[attnum];
EEO_NEXT();
}
EEO_CASE(EEOP_ASSIGN_OUTER_VAR)
{
int resultnum = op->d.assign_var.resultnum;
int attnum = op->d.assign_var.attnum;
/*
* We do not need CheckVarSlotCompatibility here; that was taken
* care of at compilation time. But see EEOP_INNER_VAR comments.
*/
Assert(attnum >= 0 && attnum < outerslot->tts_nvalid);
Fix mishandling of resjunk columns in ON CONFLICT ... UPDATE tlists. It's unusual to have any resjunk columns in an ON CONFLICT ... UPDATE list, but it can happen when MULTIEXPR_SUBLINK SubPlans are present. If it happens, the ON CONFLICT UPDATE code path would end up storing tuples that include the values of the extra resjunk columns. That's fairly harmless in the short run, but if new columns are added to the table then the values would become accessible, possibly leading to malfunctions if they don't match the datatypes of the new columns. This had escaped notice through a confluence of missing sanity checks, including * There's no cross-check that a tuple presented to heap_insert or heap_update matches the table rowtype. While it's difficult to check that fully at reasonable cost, we can easily add assertions that there aren't too many columns. * The output-column-assignment cases in execExprInterp.c lacked any sanity checks on the output column numbers, which seems like an oversight considering there are plenty of assertion checks on input column numbers. Add assertions there too. * We failed to apply nodeModifyTable's ExecCheckPlanOutput() to the ON CONFLICT UPDATE tlist. That wouldn't have caught this specific error, since that function is chartered to ignore resjunk columns; but it sure seems like a bad omission now that we've seen this bug. In HEAD, the right way to fix this is to make the processing of ON CONFLICT UPDATE tlists work the same as regular UPDATE tlists now do, that is don't add "SET x = x" entries, and use ExecBuildUpdateProjection to evaluate the tlist and combine it with old values of the not-set columns. This adds a little complication to ExecBuildUpdateProjection, but allows removal of a comparable amount of now-dead code from the planner. In the back branches, the most expedient solution seems to be to (a) use an output slot for the ON CONFLICT UPDATE projection that actually matches the target table, and then (b) invent a variant of ExecBuildProjectionInfo that can be told to not store values resulting from resjunk columns, so it doesn't try to store into nonexistent columns of the output slot. (We can't simply ignore the resjunk columns altogether; they have to be evaluated for MULTIEXPR_SUBLINK to work.) This works back to v10. In 9.6, projections work much differently and we can't cheaply give them such an option. The 9.6 version of this patch works by inserting a JunkFilter when it's necessary to get rid of resjunk columns. In addition, v11 and up have the reverse problem when trying to perform ON CONFLICT UPDATE on a partitioned table. Through a further oversight, adjust_partition_tlist() discarded resjunk columns when re-ordering the ON CONFLICT UPDATE tlist to match a partition. This accidentally prevented the storing-bogus-tuples problem, but at the cost that MULTIEXPR_SUBLINK cases didn't work, typically crashing if more than one row has to be updated. Fix by preserving resjunk columns in that routine. (I failed to resist the temptation to add more assertions there too, and to do some minor code beautification.) Per report from Andres Freund. Back-patch to all supported branches. Security: CVE-2021-32028
2021-05-10 17:02:29 +02:00
Assert(resultnum >= 0 && resultnum < resultslot->tts_tupleDescriptor->natts);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
resultslot->tts_values[resultnum] = outerslot->tts_values[attnum];
resultslot->tts_isnull[resultnum] = outerslot->tts_isnull[attnum];
EEO_NEXT();
}
EEO_CASE(EEOP_ASSIGN_SCAN_VAR)
{
int resultnum = op->d.assign_var.resultnum;
int attnum = op->d.assign_var.attnum;
/*
* We do not need CheckVarSlotCompatibility here; that was taken
* care of at compilation time. But see EEOP_INNER_VAR comments.
*/
Assert(attnum >= 0 && attnum < scanslot->tts_nvalid);
Fix mishandling of resjunk columns in ON CONFLICT ... UPDATE tlists. It's unusual to have any resjunk columns in an ON CONFLICT ... UPDATE list, but it can happen when MULTIEXPR_SUBLINK SubPlans are present. If it happens, the ON CONFLICT UPDATE code path would end up storing tuples that include the values of the extra resjunk columns. That's fairly harmless in the short run, but if new columns are added to the table then the values would become accessible, possibly leading to malfunctions if they don't match the datatypes of the new columns. This had escaped notice through a confluence of missing sanity checks, including * There's no cross-check that a tuple presented to heap_insert or heap_update matches the table rowtype. While it's difficult to check that fully at reasonable cost, we can easily add assertions that there aren't too many columns. * The output-column-assignment cases in execExprInterp.c lacked any sanity checks on the output column numbers, which seems like an oversight considering there are plenty of assertion checks on input column numbers. Add assertions there too. * We failed to apply nodeModifyTable's ExecCheckPlanOutput() to the ON CONFLICT UPDATE tlist. That wouldn't have caught this specific error, since that function is chartered to ignore resjunk columns; but it sure seems like a bad omission now that we've seen this bug. In HEAD, the right way to fix this is to make the processing of ON CONFLICT UPDATE tlists work the same as regular UPDATE tlists now do, that is don't add "SET x = x" entries, and use ExecBuildUpdateProjection to evaluate the tlist and combine it with old values of the not-set columns. This adds a little complication to ExecBuildUpdateProjection, but allows removal of a comparable amount of now-dead code from the planner. In the back branches, the most expedient solution seems to be to (a) use an output slot for the ON CONFLICT UPDATE projection that actually matches the target table, and then (b) invent a variant of ExecBuildProjectionInfo that can be told to not store values resulting from resjunk columns, so it doesn't try to store into nonexistent columns of the output slot. (We can't simply ignore the resjunk columns altogether; they have to be evaluated for MULTIEXPR_SUBLINK to work.) This works back to v10. In 9.6, projections work much differently and we can't cheaply give them such an option. The 9.6 version of this patch works by inserting a JunkFilter when it's necessary to get rid of resjunk columns. In addition, v11 and up have the reverse problem when trying to perform ON CONFLICT UPDATE on a partitioned table. Through a further oversight, adjust_partition_tlist() discarded resjunk columns when re-ordering the ON CONFLICT UPDATE tlist to match a partition. This accidentally prevented the storing-bogus-tuples problem, but at the cost that MULTIEXPR_SUBLINK cases didn't work, typically crashing if more than one row has to be updated. Fix by preserving resjunk columns in that routine. (I failed to resist the temptation to add more assertions there too, and to do some minor code beautification.) Per report from Andres Freund. Back-patch to all supported branches. Security: CVE-2021-32028
2021-05-10 17:02:29 +02:00
Assert(resultnum >= 0 && resultnum < resultslot->tts_tupleDescriptor->natts);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
resultslot->tts_values[resultnum] = scanslot->tts_values[attnum];
resultslot->tts_isnull[resultnum] = scanslot->tts_isnull[attnum];
EEO_NEXT();
}
EEO_CASE(EEOP_ASSIGN_TMP)
{
int resultnum = op->d.assign_tmp.resultnum;
Fix mishandling of resjunk columns in ON CONFLICT ... UPDATE tlists. It's unusual to have any resjunk columns in an ON CONFLICT ... UPDATE list, but it can happen when MULTIEXPR_SUBLINK SubPlans are present. If it happens, the ON CONFLICT UPDATE code path would end up storing tuples that include the values of the extra resjunk columns. That's fairly harmless in the short run, but if new columns are added to the table then the values would become accessible, possibly leading to malfunctions if they don't match the datatypes of the new columns. This had escaped notice through a confluence of missing sanity checks, including * There's no cross-check that a tuple presented to heap_insert or heap_update matches the table rowtype. While it's difficult to check that fully at reasonable cost, we can easily add assertions that there aren't too many columns. * The output-column-assignment cases in execExprInterp.c lacked any sanity checks on the output column numbers, which seems like an oversight considering there are plenty of assertion checks on input column numbers. Add assertions there too. * We failed to apply nodeModifyTable's ExecCheckPlanOutput() to the ON CONFLICT UPDATE tlist. That wouldn't have caught this specific error, since that function is chartered to ignore resjunk columns; but it sure seems like a bad omission now that we've seen this bug. In HEAD, the right way to fix this is to make the processing of ON CONFLICT UPDATE tlists work the same as regular UPDATE tlists now do, that is don't add "SET x = x" entries, and use ExecBuildUpdateProjection to evaluate the tlist and combine it with old values of the not-set columns. This adds a little complication to ExecBuildUpdateProjection, but allows removal of a comparable amount of now-dead code from the planner. In the back branches, the most expedient solution seems to be to (a) use an output slot for the ON CONFLICT UPDATE projection that actually matches the target table, and then (b) invent a variant of ExecBuildProjectionInfo that can be told to not store values resulting from resjunk columns, so it doesn't try to store into nonexistent columns of the output slot. (We can't simply ignore the resjunk columns altogether; they have to be evaluated for MULTIEXPR_SUBLINK to work.) This works back to v10. In 9.6, projections work much differently and we can't cheaply give them such an option. The 9.6 version of this patch works by inserting a JunkFilter when it's necessary to get rid of resjunk columns. In addition, v11 and up have the reverse problem when trying to perform ON CONFLICT UPDATE on a partitioned table. Through a further oversight, adjust_partition_tlist() discarded resjunk columns when re-ordering the ON CONFLICT UPDATE tlist to match a partition. This accidentally prevented the storing-bogus-tuples problem, but at the cost that MULTIEXPR_SUBLINK cases didn't work, typically crashing if more than one row has to be updated. Fix by preserving resjunk columns in that routine. (I failed to resist the temptation to add more assertions there too, and to do some minor code beautification.) Per report from Andres Freund. Back-patch to all supported branches. Security: CVE-2021-32028
2021-05-10 17:02:29 +02:00
Assert(resultnum >= 0 && resultnum < resultslot->tts_tupleDescriptor->natts);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
resultslot->tts_values[resultnum] = state->resvalue;
resultslot->tts_isnull[resultnum] = state->resnull;
EEO_NEXT();
}
EEO_CASE(EEOP_ASSIGN_TMP_MAKE_RO)
{
int resultnum = op->d.assign_tmp.resultnum;
Fix mishandling of resjunk columns in ON CONFLICT ... UPDATE tlists. It's unusual to have any resjunk columns in an ON CONFLICT ... UPDATE list, but it can happen when MULTIEXPR_SUBLINK SubPlans are present. If it happens, the ON CONFLICT UPDATE code path would end up storing tuples that include the values of the extra resjunk columns. That's fairly harmless in the short run, but if new columns are added to the table then the values would become accessible, possibly leading to malfunctions if they don't match the datatypes of the new columns. This had escaped notice through a confluence of missing sanity checks, including * There's no cross-check that a tuple presented to heap_insert or heap_update matches the table rowtype. While it's difficult to check that fully at reasonable cost, we can easily add assertions that there aren't too many columns. * The output-column-assignment cases in execExprInterp.c lacked any sanity checks on the output column numbers, which seems like an oversight considering there are plenty of assertion checks on input column numbers. Add assertions there too. * We failed to apply nodeModifyTable's ExecCheckPlanOutput() to the ON CONFLICT UPDATE tlist. That wouldn't have caught this specific error, since that function is chartered to ignore resjunk columns; but it sure seems like a bad omission now that we've seen this bug. In HEAD, the right way to fix this is to make the processing of ON CONFLICT UPDATE tlists work the same as regular UPDATE tlists now do, that is don't add "SET x = x" entries, and use ExecBuildUpdateProjection to evaluate the tlist and combine it with old values of the not-set columns. This adds a little complication to ExecBuildUpdateProjection, but allows removal of a comparable amount of now-dead code from the planner. In the back branches, the most expedient solution seems to be to (a) use an output slot for the ON CONFLICT UPDATE projection that actually matches the target table, and then (b) invent a variant of ExecBuildProjectionInfo that can be told to not store values resulting from resjunk columns, so it doesn't try to store into nonexistent columns of the output slot. (We can't simply ignore the resjunk columns altogether; they have to be evaluated for MULTIEXPR_SUBLINK to work.) This works back to v10. In 9.6, projections work much differently and we can't cheaply give them such an option. The 9.6 version of this patch works by inserting a JunkFilter when it's necessary to get rid of resjunk columns. In addition, v11 and up have the reverse problem when trying to perform ON CONFLICT UPDATE on a partitioned table. Through a further oversight, adjust_partition_tlist() discarded resjunk columns when re-ordering the ON CONFLICT UPDATE tlist to match a partition. This accidentally prevented the storing-bogus-tuples problem, but at the cost that MULTIEXPR_SUBLINK cases didn't work, typically crashing if more than one row has to be updated. Fix by preserving resjunk columns in that routine. (I failed to resist the temptation to add more assertions there too, and to do some minor code beautification.) Per report from Andres Freund. Back-patch to all supported branches. Security: CVE-2021-32028
2021-05-10 17:02:29 +02:00
Assert(resultnum >= 0 && resultnum < resultslot->tts_tupleDescriptor->natts);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
resultslot->tts_isnull[resultnum] = state->resnull;
if (!resultslot->tts_isnull[resultnum])
resultslot->tts_values[resultnum] =
MakeExpandedObjectReadOnlyInternal(state->resvalue);
else
resultslot->tts_values[resultnum] = state->resvalue;
EEO_NEXT();
}
EEO_CASE(EEOP_CONST)
{
*op->resnull = op->d.constval.isnull;
*op->resvalue = op->d.constval.value;
EEO_NEXT();
}
/*
* Function-call implementations. Arguments have previously been
* evaluated directly into fcinfo->args.
*
* As both STRICT checks and function-usage are noticeable performance
* wise, and function calls are a very hot-path (they also back
* operators!), it's worth having so many separate opcodes.
Marginal improvement for generated code in execExprInterp.c. Avoid the coding pattern "*op->resvalue = f();", as some compilers think that requires them to evaluate "op->resvalue" before the function call. Unless there are lots of free registers, this can lead to a useless register spill and reload across the call. I changed all the cases like this in ExecInterpExpr(), but didn't bother in the out-of-line opcode eval subroutines, since those are presumably not as performance-critical. Discussion: https://postgr.es/m/2508.1506630094@sss.pgh.pa.us
2017-09-29 17:32:05 +02:00
*
* Note: the reason for using a temporary variable "d", here and in
* other places, is that some compilers think "*op->resvalue = f();"
* requires them to evaluate op->resvalue into a register before
* calling f(), just in case f() is able to modify op->resvalue
* somehow. The extra line of code can save a useless register spill
* and reload across the function call.
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
*/
EEO_CASE(EEOP_FUNCEXPR)
{
FunctionCallInfo fcinfo = op->d.func.fcinfo_data;
Marginal improvement for generated code in execExprInterp.c. Avoid the coding pattern "*op->resvalue = f();", as some compilers think that requires them to evaluate "op->resvalue" before the function call. Unless there are lots of free registers, this can lead to a useless register spill and reload across the call. I changed all the cases like this in ExecInterpExpr(), but didn't bother in the out-of-line opcode eval subroutines, since those are presumably not as performance-critical. Discussion: https://postgr.es/m/2508.1506630094@sss.pgh.pa.us
2017-09-29 17:32:05 +02:00
Datum d;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
fcinfo->isnull = false;
Marginal improvement for generated code in execExprInterp.c. Avoid the coding pattern "*op->resvalue = f();", as some compilers think that requires them to evaluate "op->resvalue" before the function call. Unless there are lots of free registers, this can lead to a useless register spill and reload across the call. I changed all the cases like this in ExecInterpExpr(), but didn't bother in the out-of-line opcode eval subroutines, since those are presumably not as performance-critical. Discussion: https://postgr.es/m/2508.1506630094@sss.pgh.pa.us
2017-09-29 17:32:05 +02:00
d = op->d.func.fn_addr(fcinfo);
*op->resvalue = d;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
*op->resnull = fcinfo->isnull;
EEO_NEXT();
}
EEO_CASE(EEOP_FUNCEXPR_STRICT)
{
FunctionCallInfo fcinfo = op->d.func.fcinfo_data;
Change function call information to be variable length. Before this change FunctionCallInfoData, the struct arguments etc for V1 function calls are stored in, always had space for FUNC_MAX_ARGS/100 arguments, storing datums and their nullness in two arrays. For nearly every function call 100 arguments is far more than needed, therefore wasting memory. Arg and argnull being two separate arrays also guarantees that to access a single argument, two cachelines have to be touched. Change the layout so there's a single variable-length array with pairs of value / isnull. That drastically reduces memory consumption for most function calls (on x86-64 a two argument function now uses 64bytes, previously 936 bytes), and makes it very likely that argument value and its nullness are on the same cacheline. Arguments are stored in a new NullableDatum struct, which, due to padding, needs more memory per argument than before. But as usually far fewer arguments are stored, and individual arguments are cheaper to access, that's still a clear win. It's likely that there's other places where conversion to NullableDatum arrays would make sense, e.g. TupleTableSlots, but that's for another commit. Because the function call information is now variable-length allocations have to take the number of arguments into account. For heap allocations that can be done with SizeForFunctionCallInfoData(), for on-stack allocations there's a new LOCAL_FCINFO(name, nargs) macro that helps to allocate an appropriately sized and aligned variable. Some places with stack allocation function call information don't know the number of arguments at compile time, and currently variably sized stack allocations aren't allowed in postgres. Therefore allow for FUNC_MAX_ARGS space in these cases. They're not that common, so for now that seems acceptable. Because of the need to allocate FunctionCallInfo of the appropriate size, older extensions may need to update their code. To avoid subtle breakages, the FunctionCallInfoData struct has been renamed to FunctionCallInfoBaseData. Most code only references FunctionCallInfo, so that shouldn't cause much collateral damage. This change is also a prerequisite for more efficient expression JIT compilation (by allocating the function call information on the stack, allowing LLVM to optimize it away); previously the size of the call information caused problems inside LLVM's optimizer. Author: Andres Freund Reviewed-By: Tom Lane Discussion: https://postgr.es/m/20180605172952.x34m5uz6ju6enaem@alap3.anarazel.de
2019-01-26 23:17:52 +01:00
NullableDatum *args = fcinfo->args;
expression eval, jit: Minor code cleanups. This mostly consists of using C99 style for loops, moving variables into narrower scopes, and a smattering of other minor improvements. Done separately to make it easier to review patches with actual functional changes. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:04:50 +01:00
int nargs = op->d.func.nargs;
Marginal improvement for generated code in execExprInterp.c. Avoid the coding pattern "*op->resvalue = f();", as some compilers think that requires them to evaluate "op->resvalue" before the function call. Unless there are lots of free registers, this can lead to a useless register spill and reload across the call. I changed all the cases like this in ExecInterpExpr(), but didn't bother in the out-of-line opcode eval subroutines, since those are presumably not as performance-critical. Discussion: https://postgr.es/m/2508.1506630094@sss.pgh.pa.us
2017-09-29 17:32:05 +02:00
Datum d;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
/* strict function, so check for NULL args */
expression eval, jit: Minor code cleanups. This mostly consists of using C99 style for loops, moving variables into narrower scopes, and a smattering of other minor improvements. Done separately to make it easier to review patches with actual functional changes. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:04:50 +01:00
for (int argno = 0; argno < nargs; argno++)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
Change function call information to be variable length. Before this change FunctionCallInfoData, the struct arguments etc for V1 function calls are stored in, always had space for FUNC_MAX_ARGS/100 arguments, storing datums and their nullness in two arrays. For nearly every function call 100 arguments is far more than needed, therefore wasting memory. Arg and argnull being two separate arrays also guarantees that to access a single argument, two cachelines have to be touched. Change the layout so there's a single variable-length array with pairs of value / isnull. That drastically reduces memory consumption for most function calls (on x86-64 a two argument function now uses 64bytes, previously 936 bytes), and makes it very likely that argument value and its nullness are on the same cacheline. Arguments are stored in a new NullableDatum struct, which, due to padding, needs more memory per argument than before. But as usually far fewer arguments are stored, and individual arguments are cheaper to access, that's still a clear win. It's likely that there's other places where conversion to NullableDatum arrays would make sense, e.g. TupleTableSlots, but that's for another commit. Because the function call information is now variable-length allocations have to take the number of arguments into account. For heap allocations that can be done with SizeForFunctionCallInfoData(), for on-stack allocations there's a new LOCAL_FCINFO(name, nargs) macro that helps to allocate an appropriately sized and aligned variable. Some places with stack allocation function call information don't know the number of arguments at compile time, and currently variably sized stack allocations aren't allowed in postgres. Therefore allow for FUNC_MAX_ARGS space in these cases. They're not that common, so for now that seems acceptable. Because of the need to allocate FunctionCallInfo of the appropriate size, older extensions may need to update their code. To avoid subtle breakages, the FunctionCallInfoData struct has been renamed to FunctionCallInfoBaseData. Most code only references FunctionCallInfo, so that shouldn't cause much collateral damage. This change is also a prerequisite for more efficient expression JIT compilation (by allocating the function call information on the stack, allowing LLVM to optimize it away); previously the size of the call information caused problems inside LLVM's optimizer. Author: Andres Freund Reviewed-By: Tom Lane Discussion: https://postgr.es/m/20180605172952.x34m5uz6ju6enaem@alap3.anarazel.de
2019-01-26 23:17:52 +01:00
if (args[argno].isnull)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
*op->resnull = true;
goto strictfail;
}
}
fcinfo->isnull = false;
Marginal improvement for generated code in execExprInterp.c. Avoid the coding pattern "*op->resvalue = f();", as some compilers think that requires them to evaluate "op->resvalue" before the function call. Unless there are lots of free registers, this can lead to a useless register spill and reload across the call. I changed all the cases like this in ExecInterpExpr(), but didn't bother in the out-of-line opcode eval subroutines, since those are presumably not as performance-critical. Discussion: https://postgr.es/m/2508.1506630094@sss.pgh.pa.us
2017-09-29 17:32:05 +02:00
d = op->d.func.fn_addr(fcinfo);
*op->resvalue = d;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
*op->resnull = fcinfo->isnull;
strictfail:
EEO_NEXT();
}
EEO_CASE(EEOP_FUNCEXPR_FUSAGE)
{
Handle EEOP_FUNCEXPR_[STRICT_]FUSAGE out of line. This isn't a very common op, and it doesn't seem worth duplicating for JIT. Author: Andres Freund
2018-03-21 01:32:21 +01:00
/* not common enough to inline */
ExecEvalFuncExprFusage(state, op, econtext);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
EEO_NEXT();
}
EEO_CASE(EEOP_FUNCEXPR_STRICT_FUSAGE)
{
Handle EEOP_FUNCEXPR_[STRICT_]FUSAGE out of line. This isn't a very common op, and it doesn't seem worth duplicating for JIT. Author: Andres Freund
2018-03-21 01:32:21 +01:00
/* not common enough to inline */
ExecEvalFuncExprStrictFusage(state, op, econtext);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
EEO_NEXT();
}
/*
* If any of its clauses is FALSE, an AND's result is FALSE regardless
* of the states of the rest of the clauses, so we can stop evaluating
* and return FALSE immediately. If none are FALSE and one or more is
* NULL, we return NULL; otherwise we return TRUE. This makes sense
* when you interpret NULL as "don't know": perhaps one of the "don't
* knows" would have been FALSE if we'd known its value. Only when
* all the inputs are known to be TRUE can we state confidently that
* the AND's result is TRUE.
*/
EEO_CASE(EEOP_BOOL_AND_STEP_FIRST)
{
*op->d.boolexpr.anynull = false;
/*
* EEOP_BOOL_AND_STEP_FIRST resets anynull, otherwise it's the
* same as EEOP_BOOL_AND_STEP - so fall through to that.
*/
/* FALL THROUGH */
}
EEO_CASE(EEOP_BOOL_AND_STEP)
{
if (*op->resnull)
{
*op->d.boolexpr.anynull = true;
}
else if (!DatumGetBool(*op->resvalue))
{
/* result is already set to FALSE, need not change it */
/* bail out early */
EEO_JUMP(op->d.boolexpr.jumpdone);
}
EEO_NEXT();
}
EEO_CASE(EEOP_BOOL_AND_STEP_LAST)
{
if (*op->resnull)
{
/* result is already set to NULL, need not change it */
}
else if (!DatumGetBool(*op->resvalue))
{
/* result is already set to FALSE, need not change it */
/*
* No point jumping early to jumpdone - would be same target
* (as this is the last argument to the AND expression),
* except more expensive.
*/
}
else if (*op->d.boolexpr.anynull)
{
*op->resvalue = (Datum) 0;
*op->resnull = true;
}
else
{
/* result is already set to TRUE, need not change it */
}
EEO_NEXT();
}
/*
* If any of its clauses is TRUE, an OR's result is TRUE regardless of
* the states of the rest of the clauses, so we can stop evaluating
* and return TRUE immediately. If none are TRUE and one or more is
* NULL, we return NULL; otherwise we return FALSE. This makes sense
* when you interpret NULL as "don't know": perhaps one of the "don't
* knows" would have been TRUE if we'd known its value. Only when all
* the inputs are known to be FALSE can we state confidently that the
* OR's result is FALSE.
*/
EEO_CASE(EEOP_BOOL_OR_STEP_FIRST)
{
*op->d.boolexpr.anynull = false;
/*
* EEOP_BOOL_OR_STEP_FIRST resets anynull, otherwise it's the same
* as EEOP_BOOL_OR_STEP - so fall through to that.
*/
/* FALL THROUGH */
}
EEO_CASE(EEOP_BOOL_OR_STEP)
{
if (*op->resnull)
{
*op->d.boolexpr.anynull = true;
}
else if (DatumGetBool(*op->resvalue))
{
/* result is already set to TRUE, need not change it */
/* bail out early */
EEO_JUMP(op->d.boolexpr.jumpdone);
}
EEO_NEXT();
}
EEO_CASE(EEOP_BOOL_OR_STEP_LAST)
{
if (*op->resnull)
{
/* result is already set to NULL, need not change it */
}
else if (DatumGetBool(*op->resvalue))
{
/* result is already set to TRUE, need not change it */
/*
* No point jumping to jumpdone - would be same target (as
* this is the last argument to the AND expression), except
* more expensive.
*/
}
else if (*op->d.boolexpr.anynull)
{
*op->resvalue = (Datum) 0;
*op->resnull = true;
}
else
{
/* result is already set to FALSE, need not change it */
}
EEO_NEXT();
}
EEO_CASE(EEOP_BOOL_NOT_STEP)
{
/*
* Evaluation of 'not' is simple... if expr is false, then return
* 'true' and vice versa. It's safe to do this even on a
* nominally null value, so we ignore resnull; that means that
* NULL in produces NULL out, which is what we want.
*/
*op->resvalue = BoolGetDatum(!DatumGetBool(*op->resvalue));
EEO_NEXT();
}
EEO_CASE(EEOP_QUAL)
{
/* simplified version of BOOL_AND_STEP for use by ExecQual() */
/* If argument (also result) is false or null ... */
if (*op->resnull ||
!DatumGetBool(*op->resvalue))
{
/* ... bail out early, returning FALSE */
*op->resnull = false;
*op->resvalue = BoolGetDatum(false);
EEO_JUMP(op->d.qualexpr.jumpdone);
}
/*
* Otherwise, leave the TRUE value in place, in case this is the
* last qual. Then, TRUE is the correct answer.
*/
EEO_NEXT();
}
EEO_CASE(EEOP_JUMP)
{
/* Unconditionally jump to target step */
EEO_JUMP(op->d.jump.jumpdone);
}
EEO_CASE(EEOP_JUMP_IF_NULL)
{
/* Transfer control if current result is null */
if (*op->resnull)
EEO_JUMP(op->d.jump.jumpdone);
EEO_NEXT();
}
EEO_CASE(EEOP_JUMP_IF_NOT_NULL)
{
/* Transfer control if current result is non-null */
if (!*op->resnull)
EEO_JUMP(op->d.jump.jumpdone);
EEO_NEXT();
}
EEO_CASE(EEOP_JUMP_IF_NOT_TRUE)
{
/* Transfer control if current result is null or false */
if (*op->resnull || !DatumGetBool(*op->resvalue))
EEO_JUMP(op->d.jump.jumpdone);
EEO_NEXT();
}
EEO_CASE(EEOP_NULLTEST_ISNULL)
{
*op->resvalue = BoolGetDatum(*op->resnull);
*op->resnull = false;
EEO_NEXT();
}
EEO_CASE(EEOP_NULLTEST_ISNOTNULL)
{
*op->resvalue = BoolGetDatum(!*op->resnull);
*op->resnull = false;
EEO_NEXT();
}
EEO_CASE(EEOP_NULLTEST_ROWISNULL)
{
/* out of line implementation: too large */
ExecEvalRowNull(state, op, econtext);
EEO_NEXT();
}
EEO_CASE(EEOP_NULLTEST_ROWISNOTNULL)
{
/* out of line implementation: too large */
ExecEvalRowNotNull(state, op, econtext);
EEO_NEXT();
}
/* BooleanTest implementations for all booltesttypes */
EEO_CASE(EEOP_BOOLTEST_IS_TRUE)
{
if (*op->resnull)
Improve implementation of EEOP_BOOLTEST_* opcodes. Both Andres and I were happy with "*op->resvalue = *op->resvalue;", but Coverity isn't; and it has a point, because some compilers might not be smart enough to elide that. So remove it. In passing, also avoid doing unnecessary assignments to *op->resnull when it's already known to have the right value.
2017-03-26 21:57:02 +02:00
{
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
*op->resvalue = BoolGetDatum(false);
Improve implementation of EEOP_BOOLTEST_* opcodes. Both Andres and I were happy with "*op->resvalue = *op->resvalue;", but Coverity isn't; and it has a point, because some compilers might not be smart enough to elide that. So remove it. In passing, also avoid doing unnecessary assignments to *op->resnull when it's already known to have the right value.
2017-03-26 21:57:02 +02:00
*op->resnull = false;
}
/* else, input value is the correct output as well */
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
EEO_NEXT();
}
EEO_CASE(EEOP_BOOLTEST_IS_NOT_TRUE)
{
if (*op->resnull)
Improve implementation of EEOP_BOOLTEST_* opcodes. Both Andres and I were happy with "*op->resvalue = *op->resvalue;", but Coverity isn't; and it has a point, because some compilers might not be smart enough to elide that. So remove it. In passing, also avoid doing unnecessary assignments to *op->resnull when it's already known to have the right value.
2017-03-26 21:57:02 +02:00
{
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
*op->resvalue = BoolGetDatum(true);
Improve implementation of EEOP_BOOLTEST_* opcodes. Both Andres and I were happy with "*op->resvalue = *op->resvalue;", but Coverity isn't; and it has a point, because some compilers might not be smart enough to elide that. So remove it. In passing, also avoid doing unnecessary assignments to *op->resnull when it's already known to have the right value.
2017-03-26 21:57:02 +02:00
*op->resnull = false;
}
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
else
*op->resvalue = BoolGetDatum(!DatumGetBool(*op->resvalue));
EEO_NEXT();
}
EEO_CASE(EEOP_BOOLTEST_IS_FALSE)
{
if (*op->resnull)
Improve implementation of EEOP_BOOLTEST_* opcodes. Both Andres and I were happy with "*op->resvalue = *op->resvalue;", but Coverity isn't; and it has a point, because some compilers might not be smart enough to elide that. So remove it. In passing, also avoid doing unnecessary assignments to *op->resnull when it's already known to have the right value.
2017-03-26 21:57:02 +02:00
{
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
*op->resvalue = BoolGetDatum(false);
Improve implementation of EEOP_BOOLTEST_* opcodes. Both Andres and I were happy with "*op->resvalue = *op->resvalue;", but Coverity isn't; and it has a point, because some compilers might not be smart enough to elide that. So remove it. In passing, also avoid doing unnecessary assignments to *op->resnull when it's already known to have the right value.
2017-03-26 21:57:02 +02:00
*op->resnull = false;
}
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
else
*op->resvalue = BoolGetDatum(!DatumGetBool(*op->resvalue));
EEO_NEXT();
}
EEO_CASE(EEOP_BOOLTEST_IS_NOT_FALSE)
{
if (*op->resnull)
Improve implementation of EEOP_BOOLTEST_* opcodes. Both Andres and I were happy with "*op->resvalue = *op->resvalue;", but Coverity isn't; and it has a point, because some compilers might not be smart enough to elide that. So remove it. In passing, also avoid doing unnecessary assignments to *op->resnull when it's already known to have the right value.
2017-03-26 21:57:02 +02:00
{
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
*op->resvalue = BoolGetDatum(true);
Improve implementation of EEOP_BOOLTEST_* opcodes. Both Andres and I were happy with "*op->resvalue = *op->resvalue;", but Coverity isn't; and it has a point, because some compilers might not be smart enough to elide that. So remove it. In passing, also avoid doing unnecessary assignments to *op->resnull when it's already known to have the right value.
2017-03-26 21:57:02 +02:00
*op->resnull = false;
}
/* else, input value is the correct output as well */
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
EEO_NEXT();
}
EEO_CASE(EEOP_PARAM_EXEC)
{
/* out of line implementation: too large */
ExecEvalParamExec(state, op, econtext);
EEO_NEXT();
}
EEO_CASE(EEOP_PARAM_EXTERN)
{
/* out of line implementation: too large */
ExecEvalParamExtern(state, op, econtext);
EEO_NEXT();
}
Rearrange execution of PARAM_EXTERN Params for plpgsql's benefit. This patch does three interrelated things: * Create a new expression execution step type EEOP_PARAM_CALLBACK and add the infrastructure needed for add-on modules to generate that. As discussed, the best control mechanism for that seems to be to add another hook function to ParamListInfo, which will be called by ExecInitExpr if it's supplied and a PARAM_EXTERN Param is found. For stand-alone expressions, we add a new entry point to allow the ParamListInfo to be specified directly, since it can't be retrieved from the parent plan node's EState. * Redesign the API for the ParamListInfo paramFetch hook so that the ParamExternData array can be entirely virtual. This also lets us get rid of ParamListInfo.paramMask, instead leaving it to the paramFetch hook to decide which param IDs should be accessible or not. plpgsql_param_fetch was already doing the identical masking check, so having callers do it too seemed redundant. While I was at it, I added a "speculative" flag to paramFetch that the planner can specify as TRUE to avoid unwanted failures. This solves an ancient problem for plpgsql that it couldn't provide values of non-DTYPE_VAR variables to the planner for fear of triggering premature "record not assigned yet" or "field not found" errors during planning. * Rework plpgsql to get rid of the need for "unshared" parameter lists, by dint of turning the single ParamListInfo per estate into a nearly read-only data structure that doesn't instantiate any per-variable data. Instead, the paramFetch hook controls access to per-variable data and can make the right decisions on the fly, replacing the cases that we used to need multiple ParamListInfos for. This might perhaps have been a performance loss on its own, but by using a paramCompile hook we can bypass plpgsql_param_fetch entirely during normal query execution. (It's now only called when, eg, we copy the ParamListInfo into a cursor portal. copyParamList() or SerializeParamList() effectively instantiate the virtual parameter array as a simple physical array without a paramFetch hook, which is what we want in those cases.) This allows reverting most of commit 6c82d8d1f, though I kept the cosmetic code-consolidation aspects of that (eg the assign_simple_var function). Performance testing shows this to be at worst a break-even change, and it can provide wins ranging up to 20% in test cases involving accesses to fields of "record" variables. The fact that values of such variables can now be exposed to the planner might produce wins in some situations, too, but I've not pursued that angle. In passing, remove the "parent" pointer from the arguments to ExecInitExprRec and related functions, instead storing that pointer in a transient field in ExprState. The ParamListInfo pointer for a stand-alone expression is handled the same way; we'd otherwise have had to add yet another recursively-passed-down argument in expression compilation. Discussion: https://postgr.es/m/32589.1513706441@sss.pgh.pa.us
2017-12-21 18:57:41 +01:00
EEO_CASE(EEOP_PARAM_CALLBACK)
{
/* allow an extension module to supply a PARAM_EXTERN value */
op->d.cparam.paramfunc(state, op, econtext);
EEO_NEXT();
}
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
EEO_CASE(EEOP_CASE_TESTVAL)
{
/*
* Normally upper parts of the expression tree have setup the
* values to be returned here, but some parts of the system
* currently misuse {caseValue,domainValue}_{datum,isNull} to set
* run-time data. So if no values have been set-up, use
* ExprContext's. This isn't pretty, but also not *that* ugly,
* and this is unlikely to be performance sensitive enough to
* worry about an extra branch.
*/
if (op->d.casetest.value)
{
*op->resvalue = *op->d.casetest.value;
*op->resnull = *op->d.casetest.isnull;
}
else
{
*op->resvalue = econtext->caseValue_datum;
*op->resnull = econtext->caseValue_isNull;
}
EEO_NEXT();
}
EEO_CASE(EEOP_DOMAIN_TESTVAL)
{
/*
* See EEOP_CASE_TESTVAL comment.
*/
if (op->d.casetest.value)
{
*op->resvalue = *op->d.casetest.value;
*op->resnull = *op->d.casetest.isnull;
}
else
{
*op->resvalue = econtext->domainValue_datum;
*op->resnull = econtext->domainValue_isNull;
}
EEO_NEXT();
}
EEO_CASE(EEOP_MAKE_READONLY)
{
/*
* Force a varlena value that might be read multiple times to R/O
*/
if (!*op->d.make_readonly.isnull)
*op->resvalue =
MakeExpandedObjectReadOnlyInternal(*op->d.make_readonly.value);
*op->resnull = *op->d.make_readonly.isnull;
EEO_NEXT();
}
EEO_CASE(EEOP_IOCOERCE)
{
/*
* Evaluate a CoerceViaIO node. This can be quite a hot path, so
* inline as much work as possible. The source value is in our
* result variable.
*/
char *str;
/* call output function (similar to OutputFunctionCall) */
if (*op->resnull)
{
/* output functions are not called on nulls */
str = NULL;
}
else
{
FunctionCallInfo fcinfo_out;
fcinfo_out = op->d.iocoerce.fcinfo_data_out;
Change function call information to be variable length. Before this change FunctionCallInfoData, the struct arguments etc for V1 function calls are stored in, always had space for FUNC_MAX_ARGS/100 arguments, storing datums and their nullness in two arrays. For nearly every function call 100 arguments is far more than needed, therefore wasting memory. Arg and argnull being two separate arrays also guarantees that to access a single argument, two cachelines have to be touched. Change the layout so there's a single variable-length array with pairs of value / isnull. That drastically reduces memory consumption for most function calls (on x86-64 a two argument function now uses 64bytes, previously 936 bytes), and makes it very likely that argument value and its nullness are on the same cacheline. Arguments are stored in a new NullableDatum struct, which, due to padding, needs more memory per argument than before. But as usually far fewer arguments are stored, and individual arguments are cheaper to access, that's still a clear win. It's likely that there's other places where conversion to NullableDatum arrays would make sense, e.g. TupleTableSlots, but that's for another commit. Because the function call information is now variable-length allocations have to take the number of arguments into account. For heap allocations that can be done with SizeForFunctionCallInfoData(), for on-stack allocations there's a new LOCAL_FCINFO(name, nargs) macro that helps to allocate an appropriately sized and aligned variable. Some places with stack allocation function call information don't know the number of arguments at compile time, and currently variably sized stack allocations aren't allowed in postgres. Therefore allow for FUNC_MAX_ARGS space in these cases. They're not that common, so for now that seems acceptable. Because of the need to allocate FunctionCallInfo of the appropriate size, older extensions may need to update their code. To avoid subtle breakages, the FunctionCallInfoData struct has been renamed to FunctionCallInfoBaseData. Most code only references FunctionCallInfo, so that shouldn't cause much collateral damage. This change is also a prerequisite for more efficient expression JIT compilation (by allocating the function call information on the stack, allowing LLVM to optimize it away); previously the size of the call information caused problems inside LLVM's optimizer. Author: Andres Freund Reviewed-By: Tom Lane Discussion: https://postgr.es/m/20180605172952.x34m5uz6ju6enaem@alap3.anarazel.de
2019-01-26 23:17:52 +01:00
fcinfo_out->args[0].value = *op->resvalue;
fcinfo_out->args[0].isnull = false;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
fcinfo_out->isnull = false;
str = DatumGetCString(FunctionCallInvoke(fcinfo_out));
/* OutputFunctionCall assumes result isn't null */
Assert(!fcinfo_out->isnull);
}
/* call input function (similar to InputFunctionCall) */
if (!op->d.iocoerce.finfo_in->fn_strict || str != NULL)
{
FunctionCallInfo fcinfo_in;
Marginal improvement for generated code in execExprInterp.c. Avoid the coding pattern "*op->resvalue = f();", as some compilers think that requires them to evaluate "op->resvalue" before the function call. Unless there are lots of free registers, this can lead to a useless register spill and reload across the call. I changed all the cases like this in ExecInterpExpr(), but didn't bother in the out-of-line opcode eval subroutines, since those are presumably not as performance-critical. Discussion: https://postgr.es/m/2508.1506630094@sss.pgh.pa.us
2017-09-29 17:32:05 +02:00
Datum d;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
fcinfo_in = op->d.iocoerce.fcinfo_data_in;
Change function call information to be variable length. Before this change FunctionCallInfoData, the struct arguments etc for V1 function calls are stored in, always had space for FUNC_MAX_ARGS/100 arguments, storing datums and their nullness in two arrays. For nearly every function call 100 arguments is far more than needed, therefore wasting memory. Arg and argnull being two separate arrays also guarantees that to access a single argument, two cachelines have to be touched. Change the layout so there's a single variable-length array with pairs of value / isnull. That drastically reduces memory consumption for most function calls (on x86-64 a two argument function now uses 64bytes, previously 936 bytes), and makes it very likely that argument value and its nullness are on the same cacheline. Arguments are stored in a new NullableDatum struct, which, due to padding, needs more memory per argument than before. But as usually far fewer arguments are stored, and individual arguments are cheaper to access, that's still a clear win. It's likely that there's other places where conversion to NullableDatum arrays would make sense, e.g. TupleTableSlots, but that's for another commit. Because the function call information is now variable-length allocations have to take the number of arguments into account. For heap allocations that can be done with SizeForFunctionCallInfoData(), for on-stack allocations there's a new LOCAL_FCINFO(name, nargs) macro that helps to allocate an appropriately sized and aligned variable. Some places with stack allocation function call information don't know the number of arguments at compile time, and currently variably sized stack allocations aren't allowed in postgres. Therefore allow for FUNC_MAX_ARGS space in these cases. They're not that common, so for now that seems acceptable. Because of the need to allocate FunctionCallInfo of the appropriate size, older extensions may need to update their code. To avoid subtle breakages, the FunctionCallInfoData struct has been renamed to FunctionCallInfoBaseData. Most code only references FunctionCallInfo, so that shouldn't cause much collateral damage. This change is also a prerequisite for more efficient expression JIT compilation (by allocating the function call information on the stack, allowing LLVM to optimize it away); previously the size of the call information caused problems inside LLVM's optimizer. Author: Andres Freund Reviewed-By: Tom Lane Discussion: https://postgr.es/m/20180605172952.x34m5uz6ju6enaem@alap3.anarazel.de
2019-01-26 23:17:52 +01:00
fcinfo_in->args[0].value = PointerGetDatum(str);
fcinfo_in->args[0].isnull = *op->resnull;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
/* second and third arguments are already set up */
fcinfo_in->isnull = false;
Marginal improvement for generated code in execExprInterp.c. Avoid the coding pattern "*op->resvalue = f();", as some compilers think that requires them to evaluate "op->resvalue" before the function call. Unless there are lots of free registers, this can lead to a useless register spill and reload across the call. I changed all the cases like this in ExecInterpExpr(), but didn't bother in the out-of-line opcode eval subroutines, since those are presumably not as performance-critical. Discussion: https://postgr.es/m/2508.1506630094@sss.pgh.pa.us
2017-09-29 17:32:05 +02:00
d = FunctionCallInvoke(fcinfo_in);
*op->resvalue = d;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
/* Should get null result if and only if str is NULL */
if (str == NULL)
{
Assert(*op->resnull);
Assert(fcinfo_in->isnull);
}
else
{
Assert(!*op->resnull);
Assert(!fcinfo_in->isnull);
}
}
EEO_NEXT();
}
EEO_CASE(EEOP_DISTINCT)
{
/*
* IS DISTINCT FROM must evaluate arguments (already done into
Change function call information to be variable length. Before this change FunctionCallInfoData, the struct arguments etc for V1 function calls are stored in, always had space for FUNC_MAX_ARGS/100 arguments, storing datums and their nullness in two arrays. For nearly every function call 100 arguments is far more than needed, therefore wasting memory. Arg and argnull being two separate arrays also guarantees that to access a single argument, two cachelines have to be touched. Change the layout so there's a single variable-length array with pairs of value / isnull. That drastically reduces memory consumption for most function calls (on x86-64 a two argument function now uses 64bytes, previously 936 bytes), and makes it very likely that argument value and its nullness are on the same cacheline. Arguments are stored in a new NullableDatum struct, which, due to padding, needs more memory per argument than before. But as usually far fewer arguments are stored, and individual arguments are cheaper to access, that's still a clear win. It's likely that there's other places where conversion to NullableDatum arrays would make sense, e.g. TupleTableSlots, but that's for another commit. Because the function call information is now variable-length allocations have to take the number of arguments into account. For heap allocations that can be done with SizeForFunctionCallInfoData(), for on-stack allocations there's a new LOCAL_FCINFO(name, nargs) macro that helps to allocate an appropriately sized and aligned variable. Some places with stack allocation function call information don't know the number of arguments at compile time, and currently variably sized stack allocations aren't allowed in postgres. Therefore allow for FUNC_MAX_ARGS space in these cases. They're not that common, so for now that seems acceptable. Because of the need to allocate FunctionCallInfo of the appropriate size, older extensions may need to update their code. To avoid subtle breakages, the FunctionCallInfoData struct has been renamed to FunctionCallInfoBaseData. Most code only references FunctionCallInfo, so that shouldn't cause much collateral damage. This change is also a prerequisite for more efficient expression JIT compilation (by allocating the function call information on the stack, allowing LLVM to optimize it away); previously the size of the call information caused problems inside LLVM's optimizer. Author: Andres Freund Reviewed-By: Tom Lane Discussion: https://postgr.es/m/20180605172952.x34m5uz6ju6enaem@alap3.anarazel.de
2019-01-26 23:17:52 +01:00
* fcinfo->args) to determine whether they are NULL; if either is
* NULL then the result is determined. If neither is NULL, then
* proceed to evaluate the comparison function, which is just the
* type's standard equality operator. We need not care whether
* that function is strict. Because the handling of nulls is
* different, we can't just reuse EEOP_FUNCEXPR.
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
*/
FunctionCallInfo fcinfo = op->d.func.fcinfo_data;
/* check function arguments for NULLness */
Change function call information to be variable length. Before this change FunctionCallInfoData, the struct arguments etc for V1 function calls are stored in, always had space for FUNC_MAX_ARGS/100 arguments, storing datums and their nullness in two arrays. For nearly every function call 100 arguments is far more than needed, therefore wasting memory. Arg and argnull being two separate arrays also guarantees that to access a single argument, two cachelines have to be touched. Change the layout so there's a single variable-length array with pairs of value / isnull. That drastically reduces memory consumption for most function calls (on x86-64 a two argument function now uses 64bytes, previously 936 bytes), and makes it very likely that argument value and its nullness are on the same cacheline. Arguments are stored in a new NullableDatum struct, which, due to padding, needs more memory per argument than before. But as usually far fewer arguments are stored, and individual arguments are cheaper to access, that's still a clear win. It's likely that there's other places where conversion to NullableDatum arrays would make sense, e.g. TupleTableSlots, but that's for another commit. Because the function call information is now variable-length allocations have to take the number of arguments into account. For heap allocations that can be done with SizeForFunctionCallInfoData(), for on-stack allocations there's a new LOCAL_FCINFO(name, nargs) macro that helps to allocate an appropriately sized and aligned variable. Some places with stack allocation function call information don't know the number of arguments at compile time, and currently variably sized stack allocations aren't allowed in postgres. Therefore allow for FUNC_MAX_ARGS space in these cases. They're not that common, so for now that seems acceptable. Because of the need to allocate FunctionCallInfo of the appropriate size, older extensions may need to update their code. To avoid subtle breakages, the FunctionCallInfoData struct has been renamed to FunctionCallInfoBaseData. Most code only references FunctionCallInfo, so that shouldn't cause much collateral damage. This change is also a prerequisite for more efficient expression JIT compilation (by allocating the function call information on the stack, allowing LLVM to optimize it away); previously the size of the call information caused problems inside LLVM's optimizer. Author: Andres Freund Reviewed-By: Tom Lane Discussion: https://postgr.es/m/20180605172952.x34m5uz6ju6enaem@alap3.anarazel.de
2019-01-26 23:17:52 +01:00
if (fcinfo->args[0].isnull && fcinfo->args[1].isnull)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
/* Both NULL? Then is not distinct... */
*op->resvalue = BoolGetDatum(false);
*op->resnull = false;
}
Change function call information to be variable length. Before this change FunctionCallInfoData, the struct arguments etc for V1 function calls are stored in, always had space for FUNC_MAX_ARGS/100 arguments, storing datums and their nullness in two arrays. For nearly every function call 100 arguments is far more than needed, therefore wasting memory. Arg and argnull being two separate arrays also guarantees that to access a single argument, two cachelines have to be touched. Change the layout so there's a single variable-length array with pairs of value / isnull. That drastically reduces memory consumption for most function calls (on x86-64 a two argument function now uses 64bytes, previously 936 bytes), and makes it very likely that argument value and its nullness are on the same cacheline. Arguments are stored in a new NullableDatum struct, which, due to padding, needs more memory per argument than before. But as usually far fewer arguments are stored, and individual arguments are cheaper to access, that's still a clear win. It's likely that there's other places where conversion to NullableDatum arrays would make sense, e.g. TupleTableSlots, but that's for another commit. Because the function call information is now variable-length allocations have to take the number of arguments into account. For heap allocations that can be done with SizeForFunctionCallInfoData(), for on-stack allocations there's a new LOCAL_FCINFO(name, nargs) macro that helps to allocate an appropriately sized and aligned variable. Some places with stack allocation function call information don't know the number of arguments at compile time, and currently variably sized stack allocations aren't allowed in postgres. Therefore allow for FUNC_MAX_ARGS space in these cases. They're not that common, so for now that seems acceptable. Because of the need to allocate FunctionCallInfo of the appropriate size, older extensions may need to update their code. To avoid subtle breakages, the FunctionCallInfoData struct has been renamed to FunctionCallInfoBaseData. Most code only references FunctionCallInfo, so that shouldn't cause much collateral damage. This change is also a prerequisite for more efficient expression JIT compilation (by allocating the function call information on the stack, allowing LLVM to optimize it away); previously the size of the call information caused problems inside LLVM's optimizer. Author: Andres Freund Reviewed-By: Tom Lane Discussion: https://postgr.es/m/20180605172952.x34m5uz6ju6enaem@alap3.anarazel.de
2019-01-26 23:17:52 +01:00
else if (fcinfo->args[0].isnull || fcinfo->args[1].isnull)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
/* Only one is NULL? Then is distinct... */
*op->resvalue = BoolGetDatum(true);
*op->resnull = false;
}
else
{
/* Neither null, so apply the equality function */
Datum eqresult;
fcinfo->isnull = false;
Reduce excessive dereferencing of function pointers It is equivalent in ANSI C to write (*funcptr) () and funcptr(). These two styles have been applied inconsistently. After discussion, we'll use the more verbose style for plain function pointer variables, to make it clear that it's a variable, and the shorter style when the function pointer is in a struct (s.func() or s->func()), because then it's clear that it's not a plain function name, and otherwise the excessive punctuation makes some of those invocations hard to read. Discussion: https://www.postgresql.org/message-id/f52c16db-14ed-757d-4b48-7ef360b1631d@2ndquadrant.com
2017-09-07 18:06:23 +02:00
eqresult = op->d.func.fn_addr(fcinfo);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
/* Must invert result of "="; safe to do even if null */
*op->resvalue = BoolGetDatum(!DatumGetBool(eqresult));
*op->resnull = fcinfo->isnull;
}
EEO_NEXT();
}
Do execGrouping.c via expression eval machinery, take two. This has a performance benefit on own, although not hugely so. The primary benefit is that it will allow for to JIT tuple deforming and comparator invocations. Large parts of this were previously committed (773aec7aa), but the commit contained an omission around cross-type comparisons and was thus reverted. Author: Andres Freund Discussion: https://postgr.es/m/20171129080934.amqqkke2zjtekd4t@alap3.anarazel.de
2018-02-16 06:55:31 +01:00
/* see EEOP_DISTINCT for comments, this is just inverted */
EEO_CASE(EEOP_NOT_DISTINCT)
{
FunctionCallInfo fcinfo = op->d.func.fcinfo_data;
Change function call information to be variable length. Before this change FunctionCallInfoData, the struct arguments etc for V1 function calls are stored in, always had space for FUNC_MAX_ARGS/100 arguments, storing datums and their nullness in two arrays. For nearly every function call 100 arguments is far more than needed, therefore wasting memory. Arg and argnull being two separate arrays also guarantees that to access a single argument, two cachelines have to be touched. Change the layout so there's a single variable-length array with pairs of value / isnull. That drastically reduces memory consumption for most function calls (on x86-64 a two argument function now uses 64bytes, previously 936 bytes), and makes it very likely that argument value and its nullness are on the same cacheline. Arguments are stored in a new NullableDatum struct, which, due to padding, needs more memory per argument than before. But as usually far fewer arguments are stored, and individual arguments are cheaper to access, that's still a clear win. It's likely that there's other places where conversion to NullableDatum arrays would make sense, e.g. TupleTableSlots, but that's for another commit. Because the function call information is now variable-length allocations have to take the number of arguments into account. For heap allocations that can be done with SizeForFunctionCallInfoData(), for on-stack allocations there's a new LOCAL_FCINFO(name, nargs) macro that helps to allocate an appropriately sized and aligned variable. Some places with stack allocation function call information don't know the number of arguments at compile time, and currently variably sized stack allocations aren't allowed in postgres. Therefore allow for FUNC_MAX_ARGS space in these cases. They're not that common, so for now that seems acceptable. Because of the need to allocate FunctionCallInfo of the appropriate size, older extensions may need to update their code. To avoid subtle breakages, the FunctionCallInfoData struct has been renamed to FunctionCallInfoBaseData. Most code only references FunctionCallInfo, so that shouldn't cause much collateral damage. This change is also a prerequisite for more efficient expression JIT compilation (by allocating the function call information on the stack, allowing LLVM to optimize it away); previously the size of the call information caused problems inside LLVM's optimizer. Author: Andres Freund Reviewed-By: Tom Lane Discussion: https://postgr.es/m/20180605172952.x34m5uz6ju6enaem@alap3.anarazel.de
2019-01-26 23:17:52 +01:00
if (fcinfo->args[0].isnull && fcinfo->args[1].isnull)
Do execGrouping.c via expression eval machinery, take two. This has a performance benefit on own, although not hugely so. The primary benefit is that it will allow for to JIT tuple deforming and comparator invocations. Large parts of this were previously committed (773aec7aa), but the commit contained an omission around cross-type comparisons and was thus reverted. Author: Andres Freund Discussion: https://postgr.es/m/20171129080934.amqqkke2zjtekd4t@alap3.anarazel.de
2018-02-16 06:55:31 +01:00
{
*op->resvalue = BoolGetDatum(true);
*op->resnull = false;
}
Change function call information to be variable length. Before this change FunctionCallInfoData, the struct arguments etc for V1 function calls are stored in, always had space for FUNC_MAX_ARGS/100 arguments, storing datums and their nullness in two arrays. For nearly every function call 100 arguments is far more than needed, therefore wasting memory. Arg and argnull being two separate arrays also guarantees that to access a single argument, two cachelines have to be touched. Change the layout so there's a single variable-length array with pairs of value / isnull. That drastically reduces memory consumption for most function calls (on x86-64 a two argument function now uses 64bytes, previously 936 bytes), and makes it very likely that argument value and its nullness are on the same cacheline. Arguments are stored in a new NullableDatum struct, which, due to padding, needs more memory per argument than before. But as usually far fewer arguments are stored, and individual arguments are cheaper to access, that's still a clear win. It's likely that there's other places where conversion to NullableDatum arrays would make sense, e.g. TupleTableSlots, but that's for another commit. Because the function call information is now variable-length allocations have to take the number of arguments into account. For heap allocations that can be done with SizeForFunctionCallInfoData(), for on-stack allocations there's a new LOCAL_FCINFO(name, nargs) macro that helps to allocate an appropriately sized and aligned variable. Some places with stack allocation function call information don't know the number of arguments at compile time, and currently variably sized stack allocations aren't allowed in postgres. Therefore allow for FUNC_MAX_ARGS space in these cases. They're not that common, so for now that seems acceptable. Because of the need to allocate FunctionCallInfo of the appropriate size, older extensions may need to update their code. To avoid subtle breakages, the FunctionCallInfoData struct has been renamed to FunctionCallInfoBaseData. Most code only references FunctionCallInfo, so that shouldn't cause much collateral damage. This change is also a prerequisite for more efficient expression JIT compilation (by allocating the function call information on the stack, allowing LLVM to optimize it away); previously the size of the call information caused problems inside LLVM's optimizer. Author: Andres Freund Reviewed-By: Tom Lane Discussion: https://postgr.es/m/20180605172952.x34m5uz6ju6enaem@alap3.anarazel.de
2019-01-26 23:17:52 +01:00
else if (fcinfo->args[0].isnull || fcinfo->args[1].isnull)
Do execGrouping.c via expression eval machinery, take two. This has a performance benefit on own, although not hugely so. The primary benefit is that it will allow for to JIT tuple deforming and comparator invocations. Large parts of this were previously committed (773aec7aa), but the commit contained an omission around cross-type comparisons and was thus reverted. Author: Andres Freund Discussion: https://postgr.es/m/20171129080934.amqqkke2zjtekd4t@alap3.anarazel.de
2018-02-16 06:55:31 +01:00
{
*op->resvalue = BoolGetDatum(false);
*op->resnull = false;
}
else
{
Datum eqresult;
fcinfo->isnull = false;
eqresult = op->d.func.fn_addr(fcinfo);
*op->resvalue = eqresult;
*op->resnull = fcinfo->isnull;
}
EEO_NEXT();
}
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
EEO_CASE(EEOP_NULLIF)
{
/*
Change function call information to be variable length. Before this change FunctionCallInfoData, the struct arguments etc for V1 function calls are stored in, always had space for FUNC_MAX_ARGS/100 arguments, storing datums and their nullness in two arrays. For nearly every function call 100 arguments is far more than needed, therefore wasting memory. Arg and argnull being two separate arrays also guarantees that to access a single argument, two cachelines have to be touched. Change the layout so there's a single variable-length array with pairs of value / isnull. That drastically reduces memory consumption for most function calls (on x86-64 a two argument function now uses 64bytes, previously 936 bytes), and makes it very likely that argument value and its nullness are on the same cacheline. Arguments are stored in a new NullableDatum struct, which, due to padding, needs more memory per argument than before. But as usually far fewer arguments are stored, and individual arguments are cheaper to access, that's still a clear win. It's likely that there's other places where conversion to NullableDatum arrays would make sense, e.g. TupleTableSlots, but that's for another commit. Because the function call information is now variable-length allocations have to take the number of arguments into account. For heap allocations that can be done with SizeForFunctionCallInfoData(), for on-stack allocations there's a new LOCAL_FCINFO(name, nargs) macro that helps to allocate an appropriately sized and aligned variable. Some places with stack allocation function call information don't know the number of arguments at compile time, and currently variably sized stack allocations aren't allowed in postgres. Therefore allow for FUNC_MAX_ARGS space in these cases. They're not that common, so for now that seems acceptable. Because of the need to allocate FunctionCallInfo of the appropriate size, older extensions may need to update their code. To avoid subtle breakages, the FunctionCallInfoData struct has been renamed to FunctionCallInfoBaseData. Most code only references FunctionCallInfo, so that shouldn't cause much collateral damage. This change is also a prerequisite for more efficient expression JIT compilation (by allocating the function call information on the stack, allowing LLVM to optimize it away); previously the size of the call information caused problems inside LLVM's optimizer. Author: Andres Freund Reviewed-By: Tom Lane Discussion: https://postgr.es/m/20180605172952.x34m5uz6ju6enaem@alap3.anarazel.de
2019-01-26 23:17:52 +01:00
* The arguments are already evaluated into fcinfo->args.
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
*/
FunctionCallInfo fcinfo = op->d.func.fcinfo_data;
/* if either argument is NULL they can't be equal */
Change function call information to be variable length. Before this change FunctionCallInfoData, the struct arguments etc for V1 function calls are stored in, always had space for FUNC_MAX_ARGS/100 arguments, storing datums and their nullness in two arrays. For nearly every function call 100 arguments is far more than needed, therefore wasting memory. Arg and argnull being two separate arrays also guarantees that to access a single argument, two cachelines have to be touched. Change the layout so there's a single variable-length array with pairs of value / isnull. That drastically reduces memory consumption for most function calls (on x86-64 a two argument function now uses 64bytes, previously 936 bytes), and makes it very likely that argument value and its nullness are on the same cacheline. Arguments are stored in a new NullableDatum struct, which, due to padding, needs more memory per argument than before. But as usually far fewer arguments are stored, and individual arguments are cheaper to access, that's still a clear win. It's likely that there's other places where conversion to NullableDatum arrays would make sense, e.g. TupleTableSlots, but that's for another commit. Because the function call information is now variable-length allocations have to take the number of arguments into account. For heap allocations that can be done with SizeForFunctionCallInfoData(), for on-stack allocations there's a new LOCAL_FCINFO(name, nargs) macro that helps to allocate an appropriately sized and aligned variable. Some places with stack allocation function call information don't know the number of arguments at compile time, and currently variably sized stack allocations aren't allowed in postgres. Therefore allow for FUNC_MAX_ARGS space in these cases. They're not that common, so for now that seems acceptable. Because of the need to allocate FunctionCallInfo of the appropriate size, older extensions may need to update their code. To avoid subtle breakages, the FunctionCallInfoData struct has been renamed to FunctionCallInfoBaseData. Most code only references FunctionCallInfo, so that shouldn't cause much collateral damage. This change is also a prerequisite for more efficient expression JIT compilation (by allocating the function call information on the stack, allowing LLVM to optimize it away); previously the size of the call information caused problems inside LLVM's optimizer. Author: Andres Freund Reviewed-By: Tom Lane Discussion: https://postgr.es/m/20180605172952.x34m5uz6ju6enaem@alap3.anarazel.de
2019-01-26 23:17:52 +01:00
if (!fcinfo->args[0].isnull && !fcinfo->args[1].isnull)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
Datum result;
fcinfo->isnull = false;
Reduce excessive dereferencing of function pointers It is equivalent in ANSI C to write (*funcptr) () and funcptr(). These two styles have been applied inconsistently. After discussion, we'll use the more verbose style for plain function pointer variables, to make it clear that it's a variable, and the shorter style when the function pointer is in a struct (s.func() or s->func()), because then it's clear that it's not a plain function name, and otherwise the excessive punctuation makes some of those invocations hard to read. Discussion: https://www.postgresql.org/message-id/f52c16db-14ed-757d-4b48-7ef360b1631d@2ndquadrant.com
2017-09-07 18:06:23 +02:00
result = op->d.func.fn_addr(fcinfo);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
/* if the arguments are equal return null */
if (!fcinfo->isnull && DatumGetBool(result))
{
*op->resvalue = (Datum) 0;
*op->resnull = true;
EEO_NEXT();
}
}
/* Arguments aren't equal, so return the first one */
Change function call information to be variable length. Before this change FunctionCallInfoData, the struct arguments etc for V1 function calls are stored in, always had space for FUNC_MAX_ARGS/100 arguments, storing datums and their nullness in two arrays. For nearly every function call 100 arguments is far more than needed, therefore wasting memory. Arg and argnull being two separate arrays also guarantees that to access a single argument, two cachelines have to be touched. Change the layout so there's a single variable-length array with pairs of value / isnull. That drastically reduces memory consumption for most function calls (on x86-64 a two argument function now uses 64bytes, previously 936 bytes), and makes it very likely that argument value and its nullness are on the same cacheline. Arguments are stored in a new NullableDatum struct, which, due to padding, needs more memory per argument than before. But as usually far fewer arguments are stored, and individual arguments are cheaper to access, that's still a clear win. It's likely that there's other places where conversion to NullableDatum arrays would make sense, e.g. TupleTableSlots, but that's for another commit. Because the function call information is now variable-length allocations have to take the number of arguments into account. For heap allocations that can be done with SizeForFunctionCallInfoData(), for on-stack allocations there's a new LOCAL_FCINFO(name, nargs) macro that helps to allocate an appropriately sized and aligned variable. Some places with stack allocation function call information don't know the number of arguments at compile time, and currently variably sized stack allocations aren't allowed in postgres. Therefore allow for FUNC_MAX_ARGS space in these cases. They're not that common, so for now that seems acceptable. Because of the need to allocate FunctionCallInfo of the appropriate size, older extensions may need to update their code. To avoid subtle breakages, the FunctionCallInfoData struct has been renamed to FunctionCallInfoBaseData. Most code only references FunctionCallInfo, so that shouldn't cause much collateral damage. This change is also a prerequisite for more efficient expression JIT compilation (by allocating the function call information on the stack, allowing LLVM to optimize it away); previously the size of the call information caused problems inside LLVM's optimizer. Author: Andres Freund Reviewed-By: Tom Lane Discussion: https://postgr.es/m/20180605172952.x34m5uz6ju6enaem@alap3.anarazel.de
2019-01-26 23:17:52 +01:00
*op->resvalue = fcinfo->args[0].value;
*op->resnull = fcinfo->args[0].isnull;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
EEO_NEXT();
}
Add back SQLValueFunction for SQL keywords This is equivalent to a revert of f193883 and fb32748, with the addition that the declaration of the SQLValueFunction node needs to gain a couple of node_attr for query jumbling. The performance impact of removing the function call inlining is proving to be too huge for some workloads where these are used. A worst-case test case of involving only simple SELECT queries with a SQL keyword is proving to lead to a reduction of 10% in TPS via pgbench and prepared queries on a high-end machine. None of the tests I ran back for this set of changes saw such a huge gap, but Alexander Lakhin and Andres Freund have found that this can be noticeable. Keeping the older performance would mean to do more inlining in the executor when using COERCE_SQL_SYNTAX for a function expression, similarly to what SQLValueFunction does. This requires more redesign work and there is little time until 16beta1 is released, so for now reverting the change is the best way forward, bringing back the previous performance. Bump catalog version. Reported-by: Alexander Lakhin Discussion: https://postgr.es/m/b32bed1b-0746-9b20-1472-4bdc9ca66d52@gmail.com
2023-05-17 03:19:17 +02:00
EEO_CASE(EEOP_SQLVALUEFUNCTION)
{
/*
* Doesn't seem worthwhile to have an inline implementation
* efficiency-wise.
*/
ExecEvalSQLValueFunction(state, op);
EEO_NEXT();
}
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
EEO_CASE(EEOP_CURRENTOFEXPR)
{
/* error invocation uses space, and shouldn't ever occur */
ExecEvalCurrentOfExpr(state, op);
EEO_NEXT();
}
Identity columns This is the SQL standard-conforming variant of PostgreSQL's serial columns. It fixes a few usability issues that serial columns have: - CREATE TABLE / LIKE copies default but refers to same sequence - cannot add/drop serialness with ALTER TABLE - dropping default does not drop sequence - need to grant separate privileges to sequence - other slight weirdnesses because serial is some kind of special macro Reviewed-by: Vitaly Burovoy <vitaly.burovoy@gmail.com>
2017-04-06 14:33:16 +02:00
EEO_CASE(EEOP_NEXTVALUEEXPR)
{
Code review for NextValueExpr expression node type. Add missing infrastructure for this node type, notably in ruleutils.c where its lack could demonstrably cause EXPLAIN to fail. Add outfuncs/readfuncs support. (outfuncs support is useful today for debugging purposes. The readfuncs support may never be needed, since at present it would only matter for parallel query and NextValueExpr should never appear in a parallelizable query; but it seems like a bad idea to have a primnode type that isn't fully supported here.) Teach planner infrastructure that NextValueExpr is a volatile, parallel-unsafe, non-leaky expression node with cost cpu_operator_cost. Given its limited scope of usage, there *might* be no live bug today from the lack of that knowledge, but it's certainly going to bite us on the rear someday. Teach pg_stat_statements about the new node type, too. While at it, also teach cost_qual_eval() that MinMaxExpr, SQLValueFunction, XmlExpr, and CoerceToDomain should be charged as cpu_operator_cost. Failing to do this for SQLValueFunction was an oversight in my commit 0bb51aa96. The others are longer-standing oversights, but no time like the present to fix them. (In principle, CoerceToDomain could have cost much higher than this, but it doesn't presently seem worth trying to examine the domain's constraints here.) Modify execExprInterp.c to execute NextValueExpr as an out-of-line function; it seems quite unlikely to me that it's worth insisting that it be inlined in all expression eval methods. Besides, providing the out-of-line function doesn't stop anyone from inlining if they want to. Adjust some places where NextValueExpr support had been inserted with the aid of a dartboard rather than keeping it in the same order as elsewhere. Discussion: https://postgr.es/m/23862.1499981661@sss.pgh.pa.us
2017-07-14 21:25:43 +02:00
/*
* Doesn't seem worthwhile to have an inline implementation
* efficiency-wise.
*/
ExecEvalNextValueExpr(state, op);
Identity columns This is the SQL standard-conforming variant of PostgreSQL's serial columns. It fixes a few usability issues that serial columns have: - CREATE TABLE / LIKE copies default but refers to same sequence - cannot add/drop serialness with ALTER TABLE - dropping default does not drop sequence - need to grant separate privileges to sequence - other slight weirdnesses because serial is some kind of special macro Reviewed-by: Vitaly Burovoy <vitaly.burovoy@gmail.com>
2017-04-06 14:33:16 +02:00
EEO_NEXT();
}
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
EEO_CASE(EEOP_ARRAYEXPR)
{
/* too complex for an inline implementation */
ExecEvalArrayExpr(state, op);
EEO_NEXT();
}
EEO_CASE(EEOP_ARRAYCOERCE)
{
/* too complex for an inline implementation */
Support arrays over domains. Allowing arrays with a domain type as their element type was left un-done in the original domain patch, but not for any very good reason. This omission leads to such surprising results as array_agg() not working on a domain column, because the parser can't identify a suitable output type for the polymorphic aggregate. In order to fix this, first clean up the APIs of coerce_to_domain() and some internal functions in parse_coerce.c so that we consistently pass around a CoercionContext along with CoercionForm. Previously, we sometimes passed an "isExplicit" boolean flag instead, which is strictly less information; and coerce_to_domain() didn't even get that, but instead had to reverse-engineer isExplicit from CoercionForm. That's contrary to the documentation in primnodes.h that says that CoercionForm only affects display and not semantics. I don't think this change fixes any live bugs, but it makes things more consistent. The main reason for doing it though is that now build_coercion_expression() receives ccontext, which it needs in order to be able to recursively invoke coerce_to_target_type(). Next, reimplement ArrayCoerceExpr so that the node does not directly know any details of what has to be done to the individual array elements while performing the array coercion. Instead, the per-element processing is represented by a sub-expression whose input is a source array element and whose output is a target array element. This simplifies life in parse_coerce.c, because it can build that sub-expression by a recursive invocation of coerce_to_target_type(). The executor now handles the per-element processing as a compiled expression instead of hard-wired code. The main advantage of this is that we can use a single ArrayCoerceExpr to handle as many as three successive steps per element: base type conversion, typmod coercion, and domain constraint checking. The old code used two stacked ArrayCoerceExprs to handle type + typmod coercion, which was pretty inefficient, and adding yet another array deconstruction to do domain constraint checking seemed very unappetizing. In the case where we just need a single, very simple coercion function, doing this straightforwardly leads to a noticeable increase in the per-array-element runtime cost. Hence, add an additional shortcut evalfunc in execExprInterp.c that skips unnecessary overhead for that specific form of expression. The runtime speed of simple cases is within 1% or so of where it was before, while cases that previously required two levels of array processing are significantly faster. Finally, create an implicit array type for every domain type, as we do for base types, enums, etc. Everything except the array-coercion case seems to just work without further effort. Tom Lane, reviewed by Andrew Dunstan Discussion: https://postgr.es/m/9852.1499791473@sss.pgh.pa.us
2017-09-30 19:40:56 +02:00
ExecEvalArrayCoerce(state, op, econtext);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
EEO_NEXT();
}
EEO_CASE(EEOP_ROW)
{
/* too complex for an inline implementation */
ExecEvalRow(state, op);
EEO_NEXT();
}
EEO_CASE(EEOP_ROWCOMPARE_STEP)
{
FunctionCallInfo fcinfo = op->d.rowcompare_step.fcinfo_data;
Marginal improvement for generated code in execExprInterp.c. Avoid the coding pattern "*op->resvalue = f();", as some compilers think that requires them to evaluate "op->resvalue" before the function call. Unless there are lots of free registers, this can lead to a useless register spill and reload across the call. I changed all the cases like this in ExecInterpExpr(), but didn't bother in the out-of-line opcode eval subroutines, since those are presumably not as performance-critical. Discussion: https://postgr.es/m/2508.1506630094@sss.pgh.pa.us
2017-09-29 17:32:05 +02:00
Datum d;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
/* force NULL result if strict fn and NULL input */
if (op->d.rowcompare_step.finfo->fn_strict &&
Change function call information to be variable length. Before this change FunctionCallInfoData, the struct arguments etc for V1 function calls are stored in, always had space for FUNC_MAX_ARGS/100 arguments, storing datums and their nullness in two arrays. For nearly every function call 100 arguments is far more than needed, therefore wasting memory. Arg and argnull being two separate arrays also guarantees that to access a single argument, two cachelines have to be touched. Change the layout so there's a single variable-length array with pairs of value / isnull. That drastically reduces memory consumption for most function calls (on x86-64 a two argument function now uses 64bytes, previously 936 bytes), and makes it very likely that argument value and its nullness are on the same cacheline. Arguments are stored in a new NullableDatum struct, which, due to padding, needs more memory per argument than before. But as usually far fewer arguments are stored, and individual arguments are cheaper to access, that's still a clear win. It's likely that there's other places where conversion to NullableDatum arrays would make sense, e.g. TupleTableSlots, but that's for another commit. Because the function call information is now variable-length allocations have to take the number of arguments into account. For heap allocations that can be done with SizeForFunctionCallInfoData(), for on-stack allocations there's a new LOCAL_FCINFO(name, nargs) macro that helps to allocate an appropriately sized and aligned variable. Some places with stack allocation function call information don't know the number of arguments at compile time, and currently variably sized stack allocations aren't allowed in postgres. Therefore allow for FUNC_MAX_ARGS space in these cases. They're not that common, so for now that seems acceptable. Because of the need to allocate FunctionCallInfo of the appropriate size, older extensions may need to update their code. To avoid subtle breakages, the FunctionCallInfoData struct has been renamed to FunctionCallInfoBaseData. Most code only references FunctionCallInfo, so that shouldn't cause much collateral damage. This change is also a prerequisite for more efficient expression JIT compilation (by allocating the function call information on the stack, allowing LLVM to optimize it away); previously the size of the call information caused problems inside LLVM's optimizer. Author: Andres Freund Reviewed-By: Tom Lane Discussion: https://postgr.es/m/20180605172952.x34m5uz6ju6enaem@alap3.anarazel.de
2019-01-26 23:17:52 +01:00
(fcinfo->args[0].isnull || fcinfo->args[1].isnull))
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
*op->resnull = true;
EEO_JUMP(op->d.rowcompare_step.jumpnull);
}
/* Apply comparison function */
fcinfo->isnull = false;
Marginal improvement for generated code in execExprInterp.c. Avoid the coding pattern "*op->resvalue = f();", as some compilers think that requires them to evaluate "op->resvalue" before the function call. Unless there are lots of free registers, this can lead to a useless register spill and reload across the call. I changed all the cases like this in ExecInterpExpr(), but didn't bother in the out-of-line opcode eval subroutines, since those are presumably not as performance-critical. Discussion: https://postgr.es/m/2508.1506630094@sss.pgh.pa.us
2017-09-29 17:32:05 +02:00
d = op->d.rowcompare_step.fn_addr(fcinfo);
*op->resvalue = d;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
/* force NULL result if NULL function result */
if (fcinfo->isnull)
{
*op->resnull = true;
EEO_JUMP(op->d.rowcompare_step.jumpnull);
}
*op->resnull = false;
/* If unequal, no need to compare remaining columns */
if (DatumGetInt32(*op->resvalue) != 0)
{
EEO_JUMP(op->d.rowcompare_step.jumpdone);
}
EEO_NEXT();
}
EEO_CASE(EEOP_ROWCOMPARE_FINAL)
{
int32 cmpresult = DatumGetInt32(*op->resvalue);
RowCompareType rctype = op->d.rowcompare_final.rctype;
*op->resnull = false;
switch (rctype)
{
/* EQ and NE cases aren't allowed here */
case ROWCOMPARE_LT:
*op->resvalue = BoolGetDatum(cmpresult < 0);
break;
case ROWCOMPARE_LE:
*op->resvalue = BoolGetDatum(cmpresult <= 0);
break;
case ROWCOMPARE_GE:
*op->resvalue = BoolGetDatum(cmpresult >= 0);
break;
case ROWCOMPARE_GT:
*op->resvalue = BoolGetDatum(cmpresult > 0);
break;
default:
Assert(false);
break;
}
EEO_NEXT();
}
EEO_CASE(EEOP_MINMAX)
{
/* too complex for an inline implementation */
ExecEvalMinMax(state, op);
EEO_NEXT();
}
EEO_CASE(EEOP_FIELDSELECT)
{
/* too complex for an inline implementation */
ExecEvalFieldSelect(state, op, econtext);
EEO_NEXT();
}
EEO_CASE(EEOP_FIELDSTORE_DEFORM)
{
/* too complex for an inline implementation */
ExecEvalFieldStoreDeForm(state, op, econtext);
EEO_NEXT();
}
EEO_CASE(EEOP_FIELDSTORE_FORM)
{
/* too complex for an inline implementation */
ExecEvalFieldStoreForm(state, op, econtext);
EEO_NEXT();
}
Support subscripting of arbitrary types, not only arrays. This patch generalizes the subscripting infrastructure so that any data type can be subscripted, if it provides a handler function to define what that means. Traditional variable-length (varlena) arrays all use array_subscript_handler(), while the existing fixed-length types that support subscripting use raw_array_subscript_handler(). It's expected that other types that want to use subscripting notation will define their own handlers. (This patch provides no such new features, though; it only lays the foundation for them.) To do this, move the parser's semantic processing of subscripts (including coercion to whatever data type is required) into a method callback supplied by the handler. On the execution side, replace the ExecEvalSubscriptingRef* layer of functions with direct calls to callback-supplied execution routines. (Thus, essentially no new run-time overhead should be caused by this patch. Indeed, there is room to remove some overhead by supplying specialized execution routines. This patch does a little bit in that line, but more could be done.) Additional work is required here and there to remove formerly hard-wired assumptions about the result type, collation, etc of a SubscriptingRef expression node; and to remove assumptions that the subscript values must be integers. One useful side-effect of this is that we now have a less squishy mechanism for identifying whether a data type is a "true" array: instead of wiring in weird rules about typlen, we can look to see if pg_type.typsubscript == F_ARRAY_SUBSCRIPT_HANDLER. For this to be bulletproof, we have to forbid user-defined types from using that handler directly; but there seems no good reason for them to do so. This patch also removes assumptions that the number of subscripts is limited to MAXDIM (6), or indeed has any hard-wired limit. That limit still applies to types handled by array_subscript_handler or raw_array_subscript_handler, but to discourage other dependencies on this constant, I've moved it from c.h to utils/array.h. Dmitry Dolgov, reviewed at various times by Tom Lane, Arthur Zakirov, Peter Eisentraut, Pavel Stehule Discussion: https://postgr.es/m/CA+q6zcVDuGBv=M0FqBYX8DPebS3F_0KQ6OVFobGJPM507_SZ_w@mail.gmail.com Discussion: https://postgr.es/m/CA+q6zcVovR+XY4mfk-7oNk-rF91gH0PebnNfuUjuuDsyHjOcVA@mail.gmail.com
2020-12-09 18:40:37 +01:00
EEO_CASE(EEOP_SBSREF_SUBSCRIPTS)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
Support subscripting of arbitrary types, not only arrays. This patch generalizes the subscripting infrastructure so that any data type can be subscripted, if it provides a handler function to define what that means. Traditional variable-length (varlena) arrays all use array_subscript_handler(), while the existing fixed-length types that support subscripting use raw_array_subscript_handler(). It's expected that other types that want to use subscripting notation will define their own handlers. (This patch provides no such new features, though; it only lays the foundation for them.) To do this, move the parser's semantic processing of subscripts (including coercion to whatever data type is required) into a method callback supplied by the handler. On the execution side, replace the ExecEvalSubscriptingRef* layer of functions with direct calls to callback-supplied execution routines. (Thus, essentially no new run-time overhead should be caused by this patch. Indeed, there is room to remove some overhead by supplying specialized execution routines. This patch does a little bit in that line, but more could be done.) Additional work is required here and there to remove formerly hard-wired assumptions about the result type, collation, etc of a SubscriptingRef expression node; and to remove assumptions that the subscript values must be integers. One useful side-effect of this is that we now have a less squishy mechanism for identifying whether a data type is a "true" array: instead of wiring in weird rules about typlen, we can look to see if pg_type.typsubscript == F_ARRAY_SUBSCRIPT_HANDLER. For this to be bulletproof, we have to forbid user-defined types from using that handler directly; but there seems no good reason for them to do so. This patch also removes assumptions that the number of subscripts is limited to MAXDIM (6), or indeed has any hard-wired limit. That limit still applies to types handled by array_subscript_handler or raw_array_subscript_handler, but to discourage other dependencies on this constant, I've moved it from c.h to utils/array.h. Dmitry Dolgov, reviewed at various times by Tom Lane, Arthur Zakirov, Peter Eisentraut, Pavel Stehule Discussion: https://postgr.es/m/CA+q6zcVDuGBv=M0FqBYX8DPebS3F_0KQ6OVFobGJPM507_SZ_w@mail.gmail.com Discussion: https://postgr.es/m/CA+q6zcVovR+XY4mfk-7oNk-rF91gH0PebnNfuUjuuDsyHjOcVA@mail.gmail.com
2020-12-09 18:40:37 +01:00
/* Precheck SubscriptingRef subscript(s) */
if (op->d.sbsref_subscript.subscriptfunc(state, op, econtext))
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
EEO_NEXT();
}
else
{
Renaming for new subscripting mechanism Over at patch https://commitfest.postgresql.org/21/1062/ Dmitry wants to introduce a more generic subscription mechanism, which allows subscripting not only arrays but also other object types such as JSONB. That functionality is introduced in a largish invasive patch, out of which this internal renaming patch was extracted. Author: Dmitry Dolgov Reviewed-by: Tom Lane, Arthur Zakirov Discussion: https://postgr.es/m/CA+q6zcUK4EqPAu7XRRO5CCjMwhz5zvg+rfWuLzVoxp_5sKS6=w@mail.gmail.com
2019-02-01 16:50:32 +01:00
/* Subscript is null, short-circuit SubscriptingRef to NULL */
EEO_JUMP(op->d.sbsref_subscript.jumpdone);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
}
}
Renaming for new subscripting mechanism Over at patch https://commitfest.postgresql.org/21/1062/ Dmitry wants to introduce a more generic subscription mechanism, which allows subscripting not only arrays but also other object types such as JSONB. That functionality is introduced in a largish invasive patch, out of which this internal renaming patch was extracted. Author: Dmitry Dolgov Reviewed-by: Tom Lane, Arthur Zakirov Discussion: https://postgr.es/m/CA+q6zcUK4EqPAu7XRRO5CCjMwhz5zvg+rfWuLzVoxp_5sKS6=w@mail.gmail.com
2019-02-01 16:50:32 +01:00
EEO_CASE(EEOP_SBSREF_OLD)
Support subscripting of arbitrary types, not only arrays. This patch generalizes the subscripting infrastructure so that any data type can be subscripted, if it provides a handler function to define what that means. Traditional variable-length (varlena) arrays all use array_subscript_handler(), while the existing fixed-length types that support subscripting use raw_array_subscript_handler(). It's expected that other types that want to use subscripting notation will define their own handlers. (This patch provides no such new features, though; it only lays the foundation for them.) To do this, move the parser's semantic processing of subscripts (including coercion to whatever data type is required) into a method callback supplied by the handler. On the execution side, replace the ExecEvalSubscriptingRef* layer of functions with direct calls to callback-supplied execution routines. (Thus, essentially no new run-time overhead should be caused by this patch. Indeed, there is room to remove some overhead by supplying specialized execution routines. This patch does a little bit in that line, but more could be done.) Additional work is required here and there to remove formerly hard-wired assumptions about the result type, collation, etc of a SubscriptingRef expression node; and to remove assumptions that the subscript values must be integers. One useful side-effect of this is that we now have a less squishy mechanism for identifying whether a data type is a "true" array: instead of wiring in weird rules about typlen, we can look to see if pg_type.typsubscript == F_ARRAY_SUBSCRIPT_HANDLER. For this to be bulletproof, we have to forbid user-defined types from using that handler directly; but there seems no good reason for them to do so. This patch also removes assumptions that the number of subscripts is limited to MAXDIM (6), or indeed has any hard-wired limit. That limit still applies to types handled by array_subscript_handler or raw_array_subscript_handler, but to discourage other dependencies on this constant, I've moved it from c.h to utils/array.h. Dmitry Dolgov, reviewed at various times by Tom Lane, Arthur Zakirov, Peter Eisentraut, Pavel Stehule Discussion: https://postgr.es/m/CA+q6zcVDuGBv=M0FqBYX8DPebS3F_0KQ6OVFobGJPM507_SZ_w@mail.gmail.com Discussion: https://postgr.es/m/CA+q6zcVovR+XY4mfk-7oNk-rF91gH0PebnNfuUjuuDsyHjOcVA@mail.gmail.com
2020-12-09 18:40:37 +01:00
EEO_CASE(EEOP_SBSREF_ASSIGN)
EEO_CASE(EEOP_SBSREF_FETCH)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
Support subscripting of arbitrary types, not only arrays. This patch generalizes the subscripting infrastructure so that any data type can be subscripted, if it provides a handler function to define what that means. Traditional variable-length (varlena) arrays all use array_subscript_handler(), while the existing fixed-length types that support subscripting use raw_array_subscript_handler(). It's expected that other types that want to use subscripting notation will define their own handlers. (This patch provides no such new features, though; it only lays the foundation for them.) To do this, move the parser's semantic processing of subscripts (including coercion to whatever data type is required) into a method callback supplied by the handler. On the execution side, replace the ExecEvalSubscriptingRef* layer of functions with direct calls to callback-supplied execution routines. (Thus, essentially no new run-time overhead should be caused by this patch. Indeed, there is room to remove some overhead by supplying specialized execution routines. This patch does a little bit in that line, but more could be done.) Additional work is required here and there to remove formerly hard-wired assumptions about the result type, collation, etc of a SubscriptingRef expression node; and to remove assumptions that the subscript values must be integers. One useful side-effect of this is that we now have a less squishy mechanism for identifying whether a data type is a "true" array: instead of wiring in weird rules about typlen, we can look to see if pg_type.typsubscript == F_ARRAY_SUBSCRIPT_HANDLER. For this to be bulletproof, we have to forbid user-defined types from using that handler directly; but there seems no good reason for them to do so. This patch also removes assumptions that the number of subscripts is limited to MAXDIM (6), or indeed has any hard-wired limit. That limit still applies to types handled by array_subscript_handler or raw_array_subscript_handler, but to discourage other dependencies on this constant, I've moved it from c.h to utils/array.h. Dmitry Dolgov, reviewed at various times by Tom Lane, Arthur Zakirov, Peter Eisentraut, Pavel Stehule Discussion: https://postgr.es/m/CA+q6zcVDuGBv=M0FqBYX8DPebS3F_0KQ6OVFobGJPM507_SZ_w@mail.gmail.com Discussion: https://postgr.es/m/CA+q6zcVovR+XY4mfk-7oNk-rF91gH0PebnNfuUjuuDsyHjOcVA@mail.gmail.com
2020-12-09 18:40:37 +01:00
/* Perform a SubscriptingRef fetch or assignment */
op->d.sbsref.subscriptfunc(state, op, econtext);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
EEO_NEXT();
}
EEO_CASE(EEOP_CONVERT_ROWTYPE)
{
/* too complex for an inline implementation */
ExecEvalConvertRowtype(state, op, econtext);
EEO_NEXT();
}
EEO_CASE(EEOP_SCALARARRAYOP)
{
/* too complex for an inline implementation */
ExecEvalScalarArrayOp(state, op);
EEO_NEXT();
}
Speedup ScalarArrayOpExpr evaluation ScalarArrayOpExprs with "useOr=true" and a set of Consts on the righthand side have traditionally been evaluated by using a linear search over the array. When these arrays contain large numbers of elements then this linear search could become a significant part of execution time. Here we add a new method of evaluating ScalarArrayOpExpr expressions to allow them to be evaluated by first building a hash table containing each element, then on subsequent evaluations, we just probe that hash table to determine if there is a match. The planner is in charge of determining when this optimization is possible and it enables it by setting hashfuncid in the ScalarArrayOpExpr. The executor will only perform the hash table evaluation when the hashfuncid is set. This means that not all cases are optimized. For example CHECK constraints containing an IN clause won't go through the planner, so won't get the hashfuncid set. We could maybe do something about that at some later date. The reason we're not doing it now is from fear that we may slow down cases where the expression is evaluated only once. Those cases can be common, for example, a single row INSERT to a table with a CHECK constraint containing an IN clause. In the planner, we enable this when there are suitable hash functions for the ScalarArrayOpExpr's operator and only when there is at least MIN_ARRAY_SIZE_FOR_HASHED_SAOP elements in the array. The threshold is currently set to 9. Author: James Coleman, David Rowley Reviewed-by: David Rowley, Tomas Vondra, Heikki Linnakangas Discussion: https://postgr.es/m/CAAaqYe8x62+=wn0zvNKCj55tPpg-JBHzhZFFc6ANovdqFw7-dA@mail.gmail.com
2021-04-08 13:51:22 +02:00
EEO_CASE(EEOP_HASHED_SCALARARRAYOP)
{
/* too complex for an inline implementation */
ExecEvalHashedScalarArrayOp(state, op, econtext);
EEO_NEXT();
}
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
EEO_CASE(EEOP_DOMAIN_NOTNULL)
{
/* too complex for an inline implementation */
ExecEvalConstraintNotNull(state, op);
EEO_NEXT();
}
EEO_CASE(EEOP_DOMAIN_CHECK)
{
/* too complex for an inline implementation */
ExecEvalConstraintCheck(state, op);
EEO_NEXT();
}
EEO_CASE(EEOP_XMLEXPR)
{
/* too complex for an inline implementation */
ExecEvalXmlExpr(state, op);
EEO_NEXT();
}
SQL/JSON: add standard JSON constructor functions This commit introduces the SQL/JSON standard-conforming constructors for JSON types: JSON_ARRAY() JSON_ARRAYAGG() JSON_OBJECT() JSON_OBJECTAGG() Most of the functionality was already present in PostgreSQL-specific functions, but these include some new functionality such as the ability to skip or include NULL values, and to allow duplicate keys or throw error when they are found, as well as the standard specified syntax to specify output type and format. Author: Nikita Glukhov <n.gluhov@postgrespro.ru> Author: Teodor Sigaev <teodor@sigaev.ru> Author: Oleg Bartunov <obartunov@gmail.com> Author: Alexander Korotkov <aekorotkov@gmail.com> Author: Amit Langote <amitlangote09@gmail.com> Reviewers have included (in no particular order) Andres Freund, Alexander Korotkov, Pavel Stehule, Andrew Alsup, Erik Rijkers, Zihong Yu, Himanshu Upadhyaya, Daniel Gustafsson, Justin Pryzby. Discussion: https://postgr.es/m/CAF4Au4w2x-5LTnN_bxky-mq4=WOqsGsxSpENCzHRAzSnEd8+WQ@mail.gmail.com Discussion: https://postgr.es/m/cd0bb935-0158-78a7-08b5-904886deac4b@postgrespro.ru Discussion: https://postgr.es/m/20220616233130.rparivafipt6doj3@alap3.anarazel.de Discussion: https://postgr.es/m/abd9b83b-aa66-f230-3d6d-734817f0995d%40postgresql.org
2023-03-29 12:11:36 +02:00
EEO_CASE(EEOP_JSON_CONSTRUCTOR)
{
/* too complex for an inline implementation */
ExecEvalJsonConstructor(state, op, econtext);
EEO_NEXT();
}
SQL/JSON: support the IS JSON predicate This patch introduces the SQL standard IS JSON predicate. It operates on text and bytea values representing JSON, as well as on the json and jsonb types. Each test has IS and IS NOT variants and supports a WITH UNIQUE KEYS flag. The tests are: IS JSON [VALUE] IS JSON ARRAY IS JSON OBJECT IS JSON SCALAR These should be self-explanatory. The WITH UNIQUE KEYS flag makes these return false when duplicate keys exist in any object within the value, not necessarily directly contained in the outermost object. Author: Nikita Glukhov <n.gluhov@postgrespro.ru> Author: Teodor Sigaev <teodor@sigaev.ru> Author: Oleg Bartunov <obartunov@gmail.com> Author: Alexander Korotkov <aekorotkov@gmail.com> Author: Amit Langote <amitlangote09@gmail.com> Author: Andrew Dunstan <andrew@dunslane.net> Reviewers have included (in no particular order) Andres Freund, Alexander Korotkov, Pavel Stehule, Andrew Alsup, Erik Rijkers, Zihong Yu, Himanshu Upadhyaya, Daniel Gustafsson, Justin Pryzby. Discussion: https://postgr.es/m/CAF4Au4w2x-5LTnN_bxky-mq4=WOqsGsxSpENCzHRAzSnEd8+WQ@mail.gmail.com Discussion: https://postgr.es/m/cd0bb935-0158-78a7-08b5-904886deac4b@postgrespro.ru Discussion: https://postgr.es/m/20220616233130.rparivafipt6doj3@alap3.anarazel.de Discussion: https://postgr.es/m/abd9b83b-aa66-f230-3d6d-734817f0995d%40postgresql.org
2023-03-31 22:34:04 +02:00
EEO_CASE(EEOP_IS_JSON)
{
/* too complex for an inline implementation */
ExecEvalJsonIsPredicate(state, op);
EEO_NEXT();
}
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
EEO_CASE(EEOP_AGGREF)
{
/*
* Returns a Datum whose value is the precomputed aggregate value
* found in the given expression context.
*/
Move per-agg and per-trans duplicate finding to the planner. This has the advantage that the cost estimates for aggregates can count the number of calls to transition and final functions correctly. Bump catalog version, because views can contain Aggrefs. Reviewed-by: Andres Freund Discussion: https://www.postgresql.org/message-id/b2e3536b-1dbc-8303-c97e-89cb0b4a9a48%40iki.fi
2020-11-24 09:45:00 +01:00
int aggno = op->d.aggref.aggno;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
Assert(econtext->ecxt_aggvalues != NULL);
Move per-agg and per-trans duplicate finding to the planner. This has the advantage that the cost estimates for aggregates can count the number of calls to transition and final functions correctly. Bump catalog version, because views can contain Aggrefs. Reviewed-by: Andres Freund Discussion: https://www.postgresql.org/message-id/b2e3536b-1dbc-8303-c97e-89cb0b4a9a48%40iki.fi
2020-11-24 09:45:00 +01:00
*op->resvalue = econtext->ecxt_aggvalues[aggno];
*op->resnull = econtext->ecxt_aggnulls[aggno];
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
EEO_NEXT();
}
EEO_CASE(EEOP_GROUPING_FUNC)
{
/* too complex/uncommon for an inline implementation */
ExecEvalGroupingFunc(state, op);
EEO_NEXT();
}
EEO_CASE(EEOP_WINDOW_FUNC)
{
/*
* Like Aggref, just return a precomputed value from the econtext.
*/
WindowFuncExprState *wfunc = op->d.window_func.wfstate;
Assert(econtext->ecxt_aggvalues != NULL);
*op->resvalue = econtext->ecxt_aggvalues[wfunc->wfuncno];
*op->resnull = econtext->ecxt_aggnulls[wfunc->wfuncno];
EEO_NEXT();
}
EEO_CASE(EEOP_SUBPLAN)
{
/* too complex for an inline implementation */
ExecEvalSubPlan(state, op, econtext);
EEO_NEXT();
}
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
/* evaluate a strict aggregate deserialization function */
EEO_CASE(EEOP_AGG_STRICT_DESERIALIZE)
{
/* Don't call a strict deserialization function with NULL input */
Change function call information to be variable length. Before this change FunctionCallInfoData, the struct arguments etc for V1 function calls are stored in, always had space for FUNC_MAX_ARGS/100 arguments, storing datums and their nullness in two arrays. For nearly every function call 100 arguments is far more than needed, therefore wasting memory. Arg and argnull being two separate arrays also guarantees that to access a single argument, two cachelines have to be touched. Change the layout so there's a single variable-length array with pairs of value / isnull. That drastically reduces memory consumption for most function calls (on x86-64 a two argument function now uses 64bytes, previously 936 bytes), and makes it very likely that argument value and its nullness are on the same cacheline. Arguments are stored in a new NullableDatum struct, which, due to padding, needs more memory per argument than before. But as usually far fewer arguments are stored, and individual arguments are cheaper to access, that's still a clear win. It's likely that there's other places where conversion to NullableDatum arrays would make sense, e.g. TupleTableSlots, but that's for another commit. Because the function call information is now variable-length allocations have to take the number of arguments into account. For heap allocations that can be done with SizeForFunctionCallInfoData(), for on-stack allocations there's a new LOCAL_FCINFO(name, nargs) macro that helps to allocate an appropriately sized and aligned variable. Some places with stack allocation function call information don't know the number of arguments at compile time, and currently variably sized stack allocations aren't allowed in postgres. Therefore allow for FUNC_MAX_ARGS space in these cases. They're not that common, so for now that seems acceptable. Because of the need to allocate FunctionCallInfo of the appropriate size, older extensions may need to update their code. To avoid subtle breakages, the FunctionCallInfoData struct has been renamed to FunctionCallInfoBaseData. Most code only references FunctionCallInfo, so that shouldn't cause much collateral damage. This change is also a prerequisite for more efficient expression JIT compilation (by allocating the function call information on the stack, allowing LLVM to optimize it away); previously the size of the call information caused problems inside LLVM's optimizer. Author: Andres Freund Reviewed-By: Tom Lane Discussion: https://postgr.es/m/20180605172952.x34m5uz6ju6enaem@alap3.anarazel.de
2019-01-26 23:17:52 +01:00
if (op->d.agg_deserialize.fcinfo_data->args[0].isnull)
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
EEO_JUMP(op->d.agg_deserialize.jumpnull);
/* fallthrough */
}
/* evaluate aggregate deserialization function (non-strict portion) */
EEO_CASE(EEOP_AGG_DESERIALIZE)
{
FunctionCallInfo fcinfo = op->d.agg_deserialize.fcinfo_data;
expression eval: Don't redundantly keep track of AggState. It's already tracked via ExprState->parent, so we don't need to also include it in ExprEvalStep. When that code originally was written ExprState->parent didn't exist, but it since has been introduced in 6719b238e8f. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:06:16 +01:00
AggState *aggstate = castNode(AggState, state->parent);
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
MemoryContext oldContext;
/*
* We run the deserialization functions in per-input-tuple memory
* context.
*/
oldContext = MemoryContextSwitchTo(aggstate->tmpcontext->ecxt_per_tuple_memory);
fcinfo->isnull = false;
*op->resvalue = FunctionCallInvoke(fcinfo);
*op->resnull = fcinfo->isnull;
MemoryContextSwitchTo(oldContext);
EEO_NEXT();
}
/*
* Check that a strict aggregate transition / combination function's
* input is not NULL.
*/
expression eval, jit: Minor code cleanups. This mostly consists of using C99 style for loops, moving variables into narrower scopes, and a smattering of other minor improvements. Done separately to make it easier to review patches with actual functional changes. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:04:50 +01:00
EEO_CASE(EEOP_AGG_STRICT_INPUT_CHECK_ARGS)
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
{
expression eval, jit: Minor code cleanups. This mostly consists of using C99 style for loops, moving variables into narrower scopes, and a smattering of other minor improvements. Done separately to make it easier to review patches with actual functional changes. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:04:50 +01:00
NullableDatum *args = op->d.agg_strict_input_check.args;
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
int nargs = op->d.agg_strict_input_check.nargs;
expression eval, jit: Minor code cleanups. This mostly consists of using C99 style for loops, moving variables into narrower scopes, and a smattering of other minor improvements. Done separately to make it easier to review patches with actual functional changes. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:04:50 +01:00
for (int argno = 0; argno < nargs; argno++)
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
{
expression eval, jit: Minor code cleanups. This mostly consists of using C99 style for loops, moving variables into narrower scopes, and a smattering of other minor improvements. Done separately to make it easier to review patches with actual functional changes. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:04:50 +01:00
if (args[argno].isnull)
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
EEO_JUMP(op->d.agg_strict_input_check.jumpnull);
}
EEO_NEXT();
}
expression eval, jit: Minor code cleanups. This mostly consists of using C99 style for loops, moving variables into narrower scopes, and a smattering of other minor improvements. Done separately to make it easier to review patches with actual functional changes. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:04:50 +01:00
EEO_CASE(EEOP_AGG_STRICT_INPUT_CHECK_NULLS)
Change function call information to be variable length. Before this change FunctionCallInfoData, the struct arguments etc for V1 function calls are stored in, always had space for FUNC_MAX_ARGS/100 arguments, storing datums and their nullness in two arrays. For nearly every function call 100 arguments is far more than needed, therefore wasting memory. Arg and argnull being two separate arrays also guarantees that to access a single argument, two cachelines have to be touched. Change the layout so there's a single variable-length array with pairs of value / isnull. That drastically reduces memory consumption for most function calls (on x86-64 a two argument function now uses 64bytes, previously 936 bytes), and makes it very likely that argument value and its nullness are on the same cacheline. Arguments are stored in a new NullableDatum struct, which, due to padding, needs more memory per argument than before. But as usually far fewer arguments are stored, and individual arguments are cheaper to access, that's still a clear win. It's likely that there's other places where conversion to NullableDatum arrays would make sense, e.g. TupleTableSlots, but that's for another commit. Because the function call information is now variable-length allocations have to take the number of arguments into account. For heap allocations that can be done with SizeForFunctionCallInfoData(), for on-stack allocations there's a new LOCAL_FCINFO(name, nargs) macro that helps to allocate an appropriately sized and aligned variable. Some places with stack allocation function call information don't know the number of arguments at compile time, and currently variably sized stack allocations aren't allowed in postgres. Therefore allow for FUNC_MAX_ARGS space in these cases. They're not that common, so for now that seems acceptable. Because of the need to allocate FunctionCallInfo of the appropriate size, older extensions may need to update their code. To avoid subtle breakages, the FunctionCallInfoData struct has been renamed to FunctionCallInfoBaseData. Most code only references FunctionCallInfo, so that shouldn't cause much collateral damage. This change is also a prerequisite for more efficient expression JIT compilation (by allocating the function call information on the stack, allowing LLVM to optimize it away); previously the size of the call information caused problems inside LLVM's optimizer. Author: Andres Freund Reviewed-By: Tom Lane Discussion: https://postgr.es/m/20180605172952.x34m5uz6ju6enaem@alap3.anarazel.de
2019-01-26 23:17:52 +01:00
{
expression eval, jit: Minor code cleanups. This mostly consists of using C99 style for loops, moving variables into narrower scopes, and a smattering of other minor improvements. Done separately to make it easier to review patches with actual functional changes. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:04:50 +01:00
bool *nulls = op->d.agg_strict_input_check.nulls;
Change function call information to be variable length. Before this change FunctionCallInfoData, the struct arguments etc for V1 function calls are stored in, always had space for FUNC_MAX_ARGS/100 arguments, storing datums and their nullness in two arrays. For nearly every function call 100 arguments is far more than needed, therefore wasting memory. Arg and argnull being two separate arrays also guarantees that to access a single argument, two cachelines have to be touched. Change the layout so there's a single variable-length array with pairs of value / isnull. That drastically reduces memory consumption for most function calls (on x86-64 a two argument function now uses 64bytes, previously 936 bytes), and makes it very likely that argument value and its nullness are on the same cacheline. Arguments are stored in a new NullableDatum struct, which, due to padding, needs more memory per argument than before. But as usually far fewer arguments are stored, and individual arguments are cheaper to access, that's still a clear win. It's likely that there's other places where conversion to NullableDatum arrays would make sense, e.g. TupleTableSlots, but that's for another commit. Because the function call information is now variable-length allocations have to take the number of arguments into account. For heap allocations that can be done with SizeForFunctionCallInfoData(), for on-stack allocations there's a new LOCAL_FCINFO(name, nargs) macro that helps to allocate an appropriately sized and aligned variable. Some places with stack allocation function call information don't know the number of arguments at compile time, and currently variably sized stack allocations aren't allowed in postgres. Therefore allow for FUNC_MAX_ARGS space in these cases. They're not that common, so for now that seems acceptable. Because of the need to allocate FunctionCallInfo of the appropriate size, older extensions may need to update their code. To avoid subtle breakages, the FunctionCallInfoData struct has been renamed to FunctionCallInfoBaseData. Most code only references FunctionCallInfo, so that shouldn't cause much collateral damage. This change is also a prerequisite for more efficient expression JIT compilation (by allocating the function call information on the stack, allowing LLVM to optimize it away); previously the size of the call information caused problems inside LLVM's optimizer. Author: Andres Freund Reviewed-By: Tom Lane Discussion: https://postgr.es/m/20180605172952.x34m5uz6ju6enaem@alap3.anarazel.de
2019-01-26 23:17:52 +01:00
int nargs = op->d.agg_strict_input_check.nargs;
expression eval, jit: Minor code cleanups. This mostly consists of using C99 style for loops, moving variables into narrower scopes, and a smattering of other minor improvements. Done separately to make it easier to review patches with actual functional changes. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:04:50 +01:00
for (int argno = 0; argno < nargs; argno++)
Change function call information to be variable length. Before this change FunctionCallInfoData, the struct arguments etc for V1 function calls are stored in, always had space for FUNC_MAX_ARGS/100 arguments, storing datums and their nullness in two arrays. For nearly every function call 100 arguments is far more than needed, therefore wasting memory. Arg and argnull being two separate arrays also guarantees that to access a single argument, two cachelines have to be touched. Change the layout so there's a single variable-length array with pairs of value / isnull. That drastically reduces memory consumption for most function calls (on x86-64 a two argument function now uses 64bytes, previously 936 bytes), and makes it very likely that argument value and its nullness are on the same cacheline. Arguments are stored in a new NullableDatum struct, which, due to padding, needs more memory per argument than before. But as usually far fewer arguments are stored, and individual arguments are cheaper to access, that's still a clear win. It's likely that there's other places where conversion to NullableDatum arrays would make sense, e.g. TupleTableSlots, but that's for another commit. Because the function call information is now variable-length allocations have to take the number of arguments into account. For heap allocations that can be done with SizeForFunctionCallInfoData(), for on-stack allocations there's a new LOCAL_FCINFO(name, nargs) macro that helps to allocate an appropriately sized and aligned variable. Some places with stack allocation function call information don't know the number of arguments at compile time, and currently variably sized stack allocations aren't allowed in postgres. Therefore allow for FUNC_MAX_ARGS space in these cases. They're not that common, so for now that seems acceptable. Because of the need to allocate FunctionCallInfo of the appropriate size, older extensions may need to update their code. To avoid subtle breakages, the FunctionCallInfoData struct has been renamed to FunctionCallInfoBaseData. Most code only references FunctionCallInfo, so that shouldn't cause much collateral damage. This change is also a prerequisite for more efficient expression JIT compilation (by allocating the function call information on the stack, allowing LLVM to optimize it away); previously the size of the call information caused problems inside LLVM's optimizer. Author: Andres Freund Reviewed-By: Tom Lane Discussion: https://postgr.es/m/20180605172952.x34m5uz6ju6enaem@alap3.anarazel.de
2019-01-26 23:17:52 +01:00
{
expression eval, jit: Minor code cleanups. This mostly consists of using C99 style for loops, moving variables into narrower scopes, and a smattering of other minor improvements. Done separately to make it easier to review patches with actual functional changes. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:04:50 +01:00
if (nulls[argno])
Change function call information to be variable length. Before this change FunctionCallInfoData, the struct arguments etc for V1 function calls are stored in, always had space for FUNC_MAX_ARGS/100 arguments, storing datums and their nullness in two arrays. For nearly every function call 100 arguments is far more than needed, therefore wasting memory. Arg and argnull being two separate arrays also guarantees that to access a single argument, two cachelines have to be touched. Change the layout so there's a single variable-length array with pairs of value / isnull. That drastically reduces memory consumption for most function calls (on x86-64 a two argument function now uses 64bytes, previously 936 bytes), and makes it very likely that argument value and its nullness are on the same cacheline. Arguments are stored in a new NullableDatum struct, which, due to padding, needs more memory per argument than before. But as usually far fewer arguments are stored, and individual arguments are cheaper to access, that's still a clear win. It's likely that there's other places where conversion to NullableDatum arrays would make sense, e.g. TupleTableSlots, but that's for another commit. Because the function call information is now variable-length allocations have to take the number of arguments into account. For heap allocations that can be done with SizeForFunctionCallInfoData(), for on-stack allocations there's a new LOCAL_FCINFO(name, nargs) macro that helps to allocate an appropriately sized and aligned variable. Some places with stack allocation function call information don't know the number of arguments at compile time, and currently variably sized stack allocations aren't allowed in postgres. Therefore allow for FUNC_MAX_ARGS space in these cases. They're not that common, so for now that seems acceptable. Because of the need to allocate FunctionCallInfo of the appropriate size, older extensions may need to update their code. To avoid subtle breakages, the FunctionCallInfoData struct has been renamed to FunctionCallInfoBaseData. Most code only references FunctionCallInfo, so that shouldn't cause much collateral damage. This change is also a prerequisite for more efficient expression JIT compilation (by allocating the function call information on the stack, allowing LLVM to optimize it away); previously the size of the call information caused problems inside LLVM's optimizer. Author: Andres Freund Reviewed-By: Tom Lane Discussion: https://postgr.es/m/20180605172952.x34m5uz6ju6enaem@alap3.anarazel.de
2019-01-26 23:17:52 +01:00
EEO_JUMP(op->d.agg_strict_input_check.jumpnull);
}
EEO_NEXT();
}
Extend ExecBuildAggTrans() to support a NULL pointer check. Optionally push a step to check for a NULL pointer to the pergroup state. This will be important for disk-based hash aggregation in combination with grouping sets. When memory limits are reached, a given tuple may find its per-group state for some grouping sets but not others. For the former, it advances the per-group state as normal; for the latter, it skips evaluation and the calling code will have to spill the tuple and reprocess it in a later batch. Add the NULL check as a separate expression step because in some common cases it's not needed. Discussion: https://postgr.es/m/20200221202212.ssb2qpmdgrnx52sj%40alap3.anarazel.de
2020-03-05 02:20:20 +01:00
/*
* Check for a NULL pointer to the per-group states.
*/
EEO_CASE(EEOP_AGG_PLAIN_PERGROUP_NULLCHECK)
{
AggState *aggstate = castNode(AggState, state->parent);
AggStatePerGroup pergroup_allaggs =
Pre-beta mechanical code beautification. Run pgindent, pgperltidy, and reformat-dat-files. This set of diffs is a bit larger than typical. We've updated to pg_bsd_indent 2.1.2, which properly indents variable declarations that have multi-line initialization expressions (the continuation lines are now indented one tab stop). We've also updated to perltidy version 20230309 and changed some of its settings, which reduces its desire to add whitespace to lines to make assignments etc. line up. Going forward, that should make for fewer random-seeming changes to existing code. Discussion: https://postgr.es/m/20230428092545.qfb3y5wcu4cm75ur@alvherre.pgsql
2023-05-19 23:24:48 +02:00
aggstate->all_pergroups[op->d.agg_plain_pergroup_nullcheck.setoff];
Extend ExecBuildAggTrans() to support a NULL pointer check. Optionally push a step to check for a NULL pointer to the pergroup state. This will be important for disk-based hash aggregation in combination with grouping sets. When memory limits are reached, a given tuple may find its per-group state for some grouping sets but not others. For the former, it advances the per-group state as normal; for the latter, it skips evaluation and the calling code will have to spill the tuple and reprocess it in a later batch. Add the NULL check as a separate expression step because in some common cases it's not needed. Discussion: https://postgr.es/m/20200221202212.ssb2qpmdgrnx52sj%40alap3.anarazel.de
2020-03-05 02:20:20 +01:00
if (pergroup_allaggs == NULL)
EEO_JUMP(op->d.agg_plain_pergroup_nullcheck.jumpnull);
EEO_NEXT();
}
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
/*
expression eval: Reduce number of steps for agg transition invocations. Do so by combining the various steps that are part of aggregate transition function invocation into one larger step. As some of the current steps are only necessary for some aggregates, have one variant of the aggregate transition step for each possible combination. To avoid further manual copies of code in the different transition step implementations, move most of the code into helper functions marked as "always inline". The benefit of this change is an increase in performance when aggregating lots of rows. This comes in part due to the reduced number of indirect jumps due to the reduced number of steps, and in part by reducing redundant setup code across steps. This mainly benefits interpreted execution, but the code generated by JIT is also improved a bit. As a nice side-effect it also ends up making the code a bit simpler. A small additional optimization is removing the need to set aggstate->curaggcontext before calling ExecAggInitGroup, choosing to instead passign curaggcontext as an argument. It was, in contrast to other aggregate related functions, only needed to fetch a memory context to copy the transition value into. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de https://postgr.es/m/5c371df7cee903e8cd4c685f90c6c72086d3a2dc.camel@j-davis.com
2020-02-24 23:39:22 +01:00
* Different types of aggregate transition functions are implemented
* as different types of steps, to avoid incurring unnecessary
* overhead. There's a step type for each valid combination of having
* a by value / by reference transition type, [not] needing to the
* initialize the transition value for the first row in a group from
* input, and [not] strict transition function.
*
* Could optimize further by splitting off by-reference for
* fixed-length types, but currently that doesn't seem worth it.
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
*/
expression eval: Reduce number of steps for agg transition invocations. Do so by combining the various steps that are part of aggregate transition function invocation into one larger step. As some of the current steps are only necessary for some aggregates, have one variant of the aggregate transition step for each possible combination. To avoid further manual copies of code in the different transition step implementations, move most of the code into helper functions marked as "always inline". The benefit of this change is an increase in performance when aggregating lots of rows. This comes in part due to the reduced number of indirect jumps due to the reduced number of steps, and in part by reducing redundant setup code across steps. This mainly benefits interpreted execution, but the code generated by JIT is also improved a bit. As a nice side-effect it also ends up making the code a bit simpler. A small additional optimization is removing the need to set aggstate->curaggcontext before calling ExecAggInitGroup, choosing to instead passign curaggcontext as an argument. It was, in contrast to other aggregate related functions, only needed to fetch a memory context to copy the transition value into. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de https://postgr.es/m/5c371df7cee903e8cd4c685f90c6c72086d3a2dc.camel@j-davis.com
2020-02-24 23:39:22 +01:00
EEO_CASE(EEOP_AGG_PLAIN_TRANS_INIT_STRICT_BYVAL)
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
{
expression eval: Don't redundantly keep track of AggState. It's already tracked via ExprState->parent, so we don't need to also include it in ExprEvalStep. When that code originally was written ExprState->parent didn't exist, but it since has been introduced in 6719b238e8f. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:06:16 +01:00
AggState *aggstate = castNode(AggState, state->parent);
expression eval: Reduce number of steps for agg transition invocations. Do so by combining the various steps that are part of aggregate transition function invocation into one larger step. As some of the current steps are only necessary for some aggregates, have one variant of the aggregate transition step for each possible combination. To avoid further manual copies of code in the different transition step implementations, move most of the code into helper functions marked as "always inline". The benefit of this change is an increase in performance when aggregating lots of rows. This comes in part due to the reduced number of indirect jumps due to the reduced number of steps, and in part by reducing redundant setup code across steps. This mainly benefits interpreted execution, but the code generated by JIT is also improved a bit. As a nice side-effect it also ends up making the code a bit simpler. A small additional optimization is removing the need to set aggstate->curaggcontext before calling ExecAggInitGroup, choosing to instead passign curaggcontext as an argument. It was, in contrast to other aggregate related functions, only needed to fetch a memory context to copy the transition value into. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de https://postgr.es/m/5c371df7cee903e8cd4c685f90c6c72086d3a2dc.camel@j-davis.com
2020-02-24 23:39:22 +01:00
AggStatePerTrans pertrans = op->d.agg_trans.pertrans;
AggStatePerGroup pergroup =
Pre-beta mechanical code beautification. Run pgindent, pgperltidy, and reformat-dat-files. This set of diffs is a bit larger than typical. We've updated to pg_bsd_indent 2.1.2, which properly indents variable declarations that have multi-line initialization expressions (the continuation lines are now indented one tab stop). We've also updated to perltidy version 20230309 and changed some of its settings, which reduces its desire to add whitespace to lines to make assignments etc. line up. Going forward, that should make for fewer random-seeming changes to existing code. Discussion: https://postgr.es/m/20230428092545.qfb3y5wcu4cm75ur@alvherre.pgsql
2023-05-19 23:24:48 +02:00
&aggstate->all_pergroups[op->d.agg_trans.setoff][op->d.agg_trans.transno];
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
expression eval: Reduce number of steps for agg transition invocations. Do so by combining the various steps that are part of aggregate transition function invocation into one larger step. As some of the current steps are only necessary for some aggregates, have one variant of the aggregate transition step for each possible combination. To avoid further manual copies of code in the different transition step implementations, move most of the code into helper functions marked as "always inline". The benefit of this change is an increase in performance when aggregating lots of rows. This comes in part due to the reduced number of indirect jumps due to the reduced number of steps, and in part by reducing redundant setup code across steps. This mainly benefits interpreted execution, but the code generated by JIT is also improved a bit. As a nice side-effect it also ends up making the code a bit simpler. A small additional optimization is removing the need to set aggstate->curaggcontext before calling ExecAggInitGroup, choosing to instead passign curaggcontext as an argument. It was, in contrast to other aggregate related functions, only needed to fetch a memory context to copy the transition value into. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de https://postgr.es/m/5c371df7cee903e8cd4c685f90c6c72086d3a2dc.camel@j-davis.com
2020-02-24 23:39:22 +01:00
Assert(pertrans->transtypeByVal);
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
if (pergroup->noTransValue)
{
expression eval: Reduce number of steps for agg transition invocations. Do so by combining the various steps that are part of aggregate transition function invocation into one larger step. As some of the current steps are only necessary for some aggregates, have one variant of the aggregate transition step for each possible combination. To avoid further manual copies of code in the different transition step implementations, move most of the code into helper functions marked as "always inline". The benefit of this change is an increase in performance when aggregating lots of rows. This comes in part due to the reduced number of indirect jumps due to the reduced number of steps, and in part by reducing redundant setup code across steps. This mainly benefits interpreted execution, but the code generated by JIT is also improved a bit. As a nice side-effect it also ends up making the code a bit simpler. A small additional optimization is removing the need to set aggstate->curaggcontext before calling ExecAggInitGroup, choosing to instead passign curaggcontext as an argument. It was, in contrast to other aggregate related functions, only needed to fetch a memory context to copy the transition value into. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de https://postgr.es/m/5c371df7cee903e8cd4c685f90c6c72086d3a2dc.camel@j-davis.com
2020-02-24 23:39:22 +01:00
/* If transValue has not yet been initialized, do so now. */
ExecAggInitGroup(aggstate, pertrans, pergroup,
op->d.agg_trans.aggcontext);
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
/* copied trans value from input, done this round */
expression eval: Reduce number of steps for agg transition invocations. Do so by combining the various steps that are part of aggregate transition function invocation into one larger step. As some of the current steps are only necessary for some aggregates, have one variant of the aggregate transition step for each possible combination. To avoid further manual copies of code in the different transition step implementations, move most of the code into helper functions marked as "always inline". The benefit of this change is an increase in performance when aggregating lots of rows. This comes in part due to the reduced number of indirect jumps due to the reduced number of steps, and in part by reducing redundant setup code across steps. This mainly benefits interpreted execution, but the code generated by JIT is also improved a bit. As a nice side-effect it also ends up making the code a bit simpler. A small additional optimization is removing the need to set aggstate->curaggcontext before calling ExecAggInitGroup, choosing to instead passign curaggcontext as an argument. It was, in contrast to other aggregate related functions, only needed to fetch a memory context to copy the transition value into. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de https://postgr.es/m/5c371df7cee903e8cd4c685f90c6c72086d3a2dc.camel@j-davis.com
2020-02-24 23:39:22 +01:00
}
else if (likely(!pergroup->transValueIsNull))
{
/* invoke transition function, unless prevented by strictness */
ExecAggPlainTransByVal(aggstate, pertrans, pergroup,
op->d.agg_trans.aggcontext,
op->d.agg_trans.setno);
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
}
EEO_NEXT();
}
expression eval: Reduce number of steps for agg transition invocations. Do so by combining the various steps that are part of aggregate transition function invocation into one larger step. As some of the current steps are only necessary for some aggregates, have one variant of the aggregate transition step for each possible combination. To avoid further manual copies of code in the different transition step implementations, move most of the code into helper functions marked as "always inline". The benefit of this change is an increase in performance when aggregating lots of rows. This comes in part due to the reduced number of indirect jumps due to the reduced number of steps, and in part by reducing redundant setup code across steps. This mainly benefits interpreted execution, but the code generated by JIT is also improved a bit. As a nice side-effect it also ends up making the code a bit simpler. A small additional optimization is removing the need to set aggstate->curaggcontext before calling ExecAggInitGroup, choosing to instead passign curaggcontext as an argument. It was, in contrast to other aggregate related functions, only needed to fetch a memory context to copy the transition value into. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de https://postgr.es/m/5c371df7cee903e8cd4c685f90c6c72086d3a2dc.camel@j-davis.com
2020-02-24 23:39:22 +01:00
/* see comments above EEOP_AGG_PLAIN_TRANS_INIT_STRICT_BYVAL */
EEO_CASE(EEOP_AGG_PLAIN_TRANS_STRICT_BYVAL)
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
{
expression eval: Don't redundantly keep track of AggState. It's already tracked via ExprState->parent, so we don't need to also include it in ExprEvalStep. When that code originally was written ExprState->parent didn't exist, but it since has been introduced in 6719b238e8f. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:06:16 +01:00
AggState *aggstate = castNode(AggState, state->parent);
expression eval: Reduce number of steps for agg transition invocations. Do so by combining the various steps that are part of aggregate transition function invocation into one larger step. As some of the current steps are only necessary for some aggregates, have one variant of the aggregate transition step for each possible combination. To avoid further manual copies of code in the different transition step implementations, move most of the code into helper functions marked as "always inline". The benefit of this change is an increase in performance when aggregating lots of rows. This comes in part due to the reduced number of indirect jumps due to the reduced number of steps, and in part by reducing redundant setup code across steps. This mainly benefits interpreted execution, but the code generated by JIT is also improved a bit. As a nice side-effect it also ends up making the code a bit simpler. A small additional optimization is removing the need to set aggstate->curaggcontext before calling ExecAggInitGroup, choosing to instead passign curaggcontext as an argument. It was, in contrast to other aggregate related functions, only needed to fetch a memory context to copy the transition value into. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de https://postgr.es/m/5c371df7cee903e8cd4c685f90c6c72086d3a2dc.camel@j-davis.com
2020-02-24 23:39:22 +01:00
AggStatePerTrans pertrans = op->d.agg_trans.pertrans;
AggStatePerGroup pergroup =
Pre-beta mechanical code beautification. Run pgindent, pgperltidy, and reformat-dat-files. This set of diffs is a bit larger than typical. We've updated to pg_bsd_indent 2.1.2, which properly indents variable declarations that have multi-line initialization expressions (the continuation lines are now indented one tab stop). We've also updated to perltidy version 20230309 and changed some of its settings, which reduces its desire to add whitespace to lines to make assignments etc. line up. Going forward, that should make for fewer random-seeming changes to existing code. Discussion: https://postgr.es/m/20230428092545.qfb3y5wcu4cm75ur@alvherre.pgsql
2023-05-19 23:24:48 +02:00
&aggstate->all_pergroups[op->d.agg_trans.setoff][op->d.agg_trans.transno];
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
expression eval: Reduce number of steps for agg transition invocations. Do so by combining the various steps that are part of aggregate transition function invocation into one larger step. As some of the current steps are only necessary for some aggregates, have one variant of the aggregate transition step for each possible combination. To avoid further manual copies of code in the different transition step implementations, move most of the code into helper functions marked as "always inline". The benefit of this change is an increase in performance when aggregating lots of rows. This comes in part due to the reduced number of indirect jumps due to the reduced number of steps, and in part by reducing redundant setup code across steps. This mainly benefits interpreted execution, but the code generated by JIT is also improved a bit. As a nice side-effect it also ends up making the code a bit simpler. A small additional optimization is removing the need to set aggstate->curaggcontext before calling ExecAggInitGroup, choosing to instead passign curaggcontext as an argument. It was, in contrast to other aggregate related functions, only needed to fetch a memory context to copy the transition value into. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de https://postgr.es/m/5c371df7cee903e8cd4c685f90c6c72086d3a2dc.camel@j-davis.com
2020-02-24 23:39:22 +01:00
Assert(pertrans->transtypeByVal);
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
expression eval: Reduce number of steps for agg transition invocations. Do so by combining the various steps that are part of aggregate transition function invocation into one larger step. As some of the current steps are only necessary for some aggregates, have one variant of the aggregate transition step for each possible combination. To avoid further manual copies of code in the different transition step implementations, move most of the code into helper functions marked as "always inline". The benefit of this change is an increase in performance when aggregating lots of rows. This comes in part due to the reduced number of indirect jumps due to the reduced number of steps, and in part by reducing redundant setup code across steps. This mainly benefits interpreted execution, but the code generated by JIT is also improved a bit. As a nice side-effect it also ends up making the code a bit simpler. A small additional optimization is removing the need to set aggstate->curaggcontext before calling ExecAggInitGroup, choosing to instead passign curaggcontext as an argument. It was, in contrast to other aggregate related functions, only needed to fetch a memory context to copy the transition value into. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de https://postgr.es/m/5c371df7cee903e8cd4c685f90c6c72086d3a2dc.camel@j-davis.com
2020-02-24 23:39:22 +01:00
if (likely(!pergroup->transValueIsNull))
ExecAggPlainTransByVal(aggstate, pertrans, pergroup,
op->d.agg_trans.aggcontext,
op->d.agg_trans.setno);
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
EEO_NEXT();
}
expression eval: Reduce number of steps for agg transition invocations. Do so by combining the various steps that are part of aggregate transition function invocation into one larger step. As some of the current steps are only necessary for some aggregates, have one variant of the aggregate transition step for each possible combination. To avoid further manual copies of code in the different transition step implementations, move most of the code into helper functions marked as "always inline". The benefit of this change is an increase in performance when aggregating lots of rows. This comes in part due to the reduced number of indirect jumps due to the reduced number of steps, and in part by reducing redundant setup code across steps. This mainly benefits interpreted execution, but the code generated by JIT is also improved a bit. As a nice side-effect it also ends up making the code a bit simpler. A small additional optimization is removing the need to set aggstate->curaggcontext before calling ExecAggInitGroup, choosing to instead passign curaggcontext as an argument. It was, in contrast to other aggregate related functions, only needed to fetch a memory context to copy the transition value into. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de https://postgr.es/m/5c371df7cee903e8cd4c685f90c6c72086d3a2dc.camel@j-davis.com
2020-02-24 23:39:22 +01:00
/* see comments above EEOP_AGG_PLAIN_TRANS_INIT_STRICT_BYVAL */
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
EEO_CASE(EEOP_AGG_PLAIN_TRANS_BYVAL)
{
expression eval: Don't redundantly keep track of AggState. It's already tracked via ExprState->parent, so we don't need to also include it in ExprEvalStep. When that code originally was written ExprState->parent didn't exist, but it since has been introduced in 6719b238e8f. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:06:16 +01:00
AggState *aggstate = castNode(AggState, state->parent);
expression eval: Reduce number of steps for agg transition invocations. Do so by combining the various steps that are part of aggregate transition function invocation into one larger step. As some of the current steps are only necessary for some aggregates, have one variant of the aggregate transition step for each possible combination. To avoid further manual copies of code in the different transition step implementations, move most of the code into helper functions marked as "always inline". The benefit of this change is an increase in performance when aggregating lots of rows. This comes in part due to the reduced number of indirect jumps due to the reduced number of steps, and in part by reducing redundant setup code across steps. This mainly benefits interpreted execution, but the code generated by JIT is also improved a bit. As a nice side-effect it also ends up making the code a bit simpler. A small additional optimization is removing the need to set aggstate->curaggcontext before calling ExecAggInitGroup, choosing to instead passign curaggcontext as an argument. It was, in contrast to other aggregate related functions, only needed to fetch a memory context to copy the transition value into. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de https://postgr.es/m/5c371df7cee903e8cd4c685f90c6c72086d3a2dc.camel@j-davis.com
2020-02-24 23:39:22 +01:00
AggStatePerTrans pertrans = op->d.agg_trans.pertrans;
AggStatePerGroup pergroup =
Pre-beta mechanical code beautification. Run pgindent, pgperltidy, and reformat-dat-files. This set of diffs is a bit larger than typical. We've updated to pg_bsd_indent 2.1.2, which properly indents variable declarations that have multi-line initialization expressions (the continuation lines are now indented one tab stop). We've also updated to perltidy version 20230309 and changed some of its settings, which reduces its desire to add whitespace to lines to make assignments etc. line up. Going forward, that should make for fewer random-seeming changes to existing code. Discussion: https://postgr.es/m/20230428092545.qfb3y5wcu4cm75ur@alvherre.pgsql
2023-05-19 23:24:48 +02:00
&aggstate->all_pergroups[op->d.agg_trans.setoff][op->d.agg_trans.transno];
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
Assert(pertrans->transtypeByVal);
expression eval: Reduce number of steps for agg transition invocations. Do so by combining the various steps that are part of aggregate transition function invocation into one larger step. As some of the current steps are only necessary for some aggregates, have one variant of the aggregate transition step for each possible combination. To avoid further manual copies of code in the different transition step implementations, move most of the code into helper functions marked as "always inline". The benefit of this change is an increase in performance when aggregating lots of rows. This comes in part due to the reduced number of indirect jumps due to the reduced number of steps, and in part by reducing redundant setup code across steps. This mainly benefits interpreted execution, but the code generated by JIT is also improved a bit. As a nice side-effect it also ends up making the code a bit simpler. A small additional optimization is removing the need to set aggstate->curaggcontext before calling ExecAggInitGroup, choosing to instead passign curaggcontext as an argument. It was, in contrast to other aggregate related functions, only needed to fetch a memory context to copy the transition value into. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de https://postgr.es/m/5c371df7cee903e8cd4c685f90c6c72086d3a2dc.camel@j-davis.com
2020-02-24 23:39:22 +01:00
ExecAggPlainTransByVal(aggstate, pertrans, pergroup,
op->d.agg_trans.aggcontext,
op->d.agg_trans.setno);
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
EEO_NEXT();
}
expression eval: Reduce number of steps for agg transition invocations. Do so by combining the various steps that are part of aggregate transition function invocation into one larger step. As some of the current steps are only necessary for some aggregates, have one variant of the aggregate transition step for each possible combination. To avoid further manual copies of code in the different transition step implementations, move most of the code into helper functions marked as "always inline". The benefit of this change is an increase in performance when aggregating lots of rows. This comes in part due to the reduced number of indirect jumps due to the reduced number of steps, and in part by reducing redundant setup code across steps. This mainly benefits interpreted execution, but the code generated by JIT is also improved a bit. As a nice side-effect it also ends up making the code a bit simpler. A small additional optimization is removing the need to set aggstate->curaggcontext before calling ExecAggInitGroup, choosing to instead passign curaggcontext as an argument. It was, in contrast to other aggregate related functions, only needed to fetch a memory context to copy the transition value into. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de https://postgr.es/m/5c371df7cee903e8cd4c685f90c6c72086d3a2dc.camel@j-davis.com
2020-02-24 23:39:22 +01:00
/* see comments above EEOP_AGG_PLAIN_TRANS_INIT_STRICT_BYVAL */
EEO_CASE(EEOP_AGG_PLAIN_TRANS_INIT_STRICT_BYREF)
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
{
expression eval: Don't redundantly keep track of AggState. It's already tracked via ExprState->parent, so we don't need to also include it in ExprEvalStep. When that code originally was written ExprState->parent didn't exist, but it since has been introduced in 6719b238e8f. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:06:16 +01:00
AggState *aggstate = castNode(AggState, state->parent);
expression eval: Reduce number of steps for agg transition invocations. Do so by combining the various steps that are part of aggregate transition function invocation into one larger step. As some of the current steps are only necessary for some aggregates, have one variant of the aggregate transition step for each possible combination. To avoid further manual copies of code in the different transition step implementations, move most of the code into helper functions marked as "always inline". The benefit of this change is an increase in performance when aggregating lots of rows. This comes in part due to the reduced number of indirect jumps due to the reduced number of steps, and in part by reducing redundant setup code across steps. This mainly benefits interpreted execution, but the code generated by JIT is also improved a bit. As a nice side-effect it also ends up making the code a bit simpler. A small additional optimization is removing the need to set aggstate->curaggcontext before calling ExecAggInitGroup, choosing to instead passign curaggcontext as an argument. It was, in contrast to other aggregate related functions, only needed to fetch a memory context to copy the transition value into. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de https://postgr.es/m/5c371df7cee903e8cd4c685f90c6c72086d3a2dc.camel@j-davis.com
2020-02-24 23:39:22 +01:00
AggStatePerTrans pertrans = op->d.agg_trans.pertrans;
AggStatePerGroup pergroup =
Pre-beta mechanical code beautification. Run pgindent, pgperltidy, and reformat-dat-files. This set of diffs is a bit larger than typical. We've updated to pg_bsd_indent 2.1.2, which properly indents variable declarations that have multi-line initialization expressions (the continuation lines are now indented one tab stop). We've also updated to perltidy version 20230309 and changed some of its settings, which reduces its desire to add whitespace to lines to make assignments etc. line up. Going forward, that should make for fewer random-seeming changes to existing code. Discussion: https://postgr.es/m/20230428092545.qfb3y5wcu4cm75ur@alvherre.pgsql
2023-05-19 23:24:48 +02:00
&aggstate->all_pergroups[op->d.agg_trans.setoff][op->d.agg_trans.transno];
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
Assert(!pertrans->transtypeByVal);
expression eval: Reduce number of steps for agg transition invocations. Do so by combining the various steps that are part of aggregate transition function invocation into one larger step. As some of the current steps are only necessary for some aggregates, have one variant of the aggregate transition step for each possible combination. To avoid further manual copies of code in the different transition step implementations, move most of the code into helper functions marked as "always inline". The benefit of this change is an increase in performance when aggregating lots of rows. This comes in part due to the reduced number of indirect jumps due to the reduced number of steps, and in part by reducing redundant setup code across steps. This mainly benefits interpreted execution, but the code generated by JIT is also improved a bit. As a nice side-effect it also ends up making the code a bit simpler. A small additional optimization is removing the need to set aggstate->curaggcontext before calling ExecAggInitGroup, choosing to instead passign curaggcontext as an argument. It was, in contrast to other aggregate related functions, only needed to fetch a memory context to copy the transition value into. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de https://postgr.es/m/5c371df7cee903e8cd4c685f90c6c72086d3a2dc.camel@j-davis.com
2020-02-24 23:39:22 +01:00
if (pergroup->noTransValue)
ExecAggInitGroup(aggstate, pertrans, pergroup,
op->d.agg_trans.aggcontext);
else if (likely(!pergroup->transValueIsNull))
ExecAggPlainTransByRef(aggstate, pertrans, pergroup,
op->d.agg_trans.aggcontext,
op->d.agg_trans.setno);
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
expression eval: Reduce number of steps for agg transition invocations. Do so by combining the various steps that are part of aggregate transition function invocation into one larger step. As some of the current steps are only necessary for some aggregates, have one variant of the aggregate transition step for each possible combination. To avoid further manual copies of code in the different transition step implementations, move most of the code into helper functions marked as "always inline". The benefit of this change is an increase in performance when aggregating lots of rows. This comes in part due to the reduced number of indirect jumps due to the reduced number of steps, and in part by reducing redundant setup code across steps. This mainly benefits interpreted execution, but the code generated by JIT is also improved a bit. As a nice side-effect it also ends up making the code a bit simpler. A small additional optimization is removing the need to set aggstate->curaggcontext before calling ExecAggInitGroup, choosing to instead passign curaggcontext as an argument. It was, in contrast to other aggregate related functions, only needed to fetch a memory context to copy the transition value into. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de https://postgr.es/m/5c371df7cee903e8cd4c685f90c6c72086d3a2dc.camel@j-davis.com
2020-02-24 23:39:22 +01:00
EEO_NEXT();
}
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
expression eval: Reduce number of steps for agg transition invocations. Do so by combining the various steps that are part of aggregate transition function invocation into one larger step. As some of the current steps are only necessary for some aggregates, have one variant of the aggregate transition step for each possible combination. To avoid further manual copies of code in the different transition step implementations, move most of the code into helper functions marked as "always inline". The benefit of this change is an increase in performance when aggregating lots of rows. This comes in part due to the reduced number of indirect jumps due to the reduced number of steps, and in part by reducing redundant setup code across steps. This mainly benefits interpreted execution, but the code generated by JIT is also improved a bit. As a nice side-effect it also ends up making the code a bit simpler. A small additional optimization is removing the need to set aggstate->curaggcontext before calling ExecAggInitGroup, choosing to instead passign curaggcontext as an argument. It was, in contrast to other aggregate related functions, only needed to fetch a memory context to copy the transition value into. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de https://postgr.es/m/5c371df7cee903e8cd4c685f90c6c72086d3a2dc.camel@j-davis.com
2020-02-24 23:39:22 +01:00
/* see comments above EEOP_AGG_PLAIN_TRANS_INIT_STRICT_BYVAL */
EEO_CASE(EEOP_AGG_PLAIN_TRANS_STRICT_BYREF)
{
AggState *aggstate = castNode(AggState, state->parent);
AggStatePerTrans pertrans = op->d.agg_trans.pertrans;
AggStatePerGroup pergroup =
Pre-beta mechanical code beautification. Run pgindent, pgperltidy, and reformat-dat-files. This set of diffs is a bit larger than typical. We've updated to pg_bsd_indent 2.1.2, which properly indents variable declarations that have multi-line initialization expressions (the continuation lines are now indented one tab stop). We've also updated to perltidy version 20230309 and changed some of its settings, which reduces its desire to add whitespace to lines to make assignments etc. line up. Going forward, that should make for fewer random-seeming changes to existing code. Discussion: https://postgr.es/m/20230428092545.qfb3y5wcu4cm75ur@alvherre.pgsql
2023-05-19 23:24:48 +02:00
&aggstate->all_pergroups[op->d.agg_trans.setoff][op->d.agg_trans.transno];
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
expression eval: Reduce number of steps for agg transition invocations. Do so by combining the various steps that are part of aggregate transition function invocation into one larger step. As some of the current steps are only necessary for some aggregates, have one variant of the aggregate transition step for each possible combination. To avoid further manual copies of code in the different transition step implementations, move most of the code into helper functions marked as "always inline". The benefit of this change is an increase in performance when aggregating lots of rows. This comes in part due to the reduced number of indirect jumps due to the reduced number of steps, and in part by reducing redundant setup code across steps. This mainly benefits interpreted execution, but the code generated by JIT is also improved a bit. As a nice side-effect it also ends up making the code a bit simpler. A small additional optimization is removing the need to set aggstate->curaggcontext before calling ExecAggInitGroup, choosing to instead passign curaggcontext as an argument. It was, in contrast to other aggregate related functions, only needed to fetch a memory context to copy the transition value into. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de https://postgr.es/m/5c371df7cee903e8cd4c685f90c6c72086d3a2dc.camel@j-davis.com
2020-02-24 23:39:22 +01:00
Assert(!pertrans->transtypeByVal);
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
expression eval: Reduce number of steps for agg transition invocations. Do so by combining the various steps that are part of aggregate transition function invocation into one larger step. As some of the current steps are only necessary for some aggregates, have one variant of the aggregate transition step for each possible combination. To avoid further manual copies of code in the different transition step implementations, move most of the code into helper functions marked as "always inline". The benefit of this change is an increase in performance when aggregating lots of rows. This comes in part due to the reduced number of indirect jumps due to the reduced number of steps, and in part by reducing redundant setup code across steps. This mainly benefits interpreted execution, but the code generated by JIT is also improved a bit. As a nice side-effect it also ends up making the code a bit simpler. A small additional optimization is removing the need to set aggstate->curaggcontext before calling ExecAggInitGroup, choosing to instead passign curaggcontext as an argument. It was, in contrast to other aggregate related functions, only needed to fetch a memory context to copy the transition value into. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de https://postgr.es/m/5c371df7cee903e8cd4c685f90c6c72086d3a2dc.camel@j-davis.com
2020-02-24 23:39:22 +01:00
if (likely(!pergroup->transValueIsNull))
ExecAggPlainTransByRef(aggstate, pertrans, pergroup,
op->d.agg_trans.aggcontext,
op->d.agg_trans.setno);
EEO_NEXT();
}
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
expression eval: Reduce number of steps for agg transition invocations. Do so by combining the various steps that are part of aggregate transition function invocation into one larger step. As some of the current steps are only necessary for some aggregates, have one variant of the aggregate transition step for each possible combination. To avoid further manual copies of code in the different transition step implementations, move most of the code into helper functions marked as "always inline". The benefit of this change is an increase in performance when aggregating lots of rows. This comes in part due to the reduced number of indirect jumps due to the reduced number of steps, and in part by reducing redundant setup code across steps. This mainly benefits interpreted execution, but the code generated by JIT is also improved a bit. As a nice side-effect it also ends up making the code a bit simpler. A small additional optimization is removing the need to set aggstate->curaggcontext before calling ExecAggInitGroup, choosing to instead passign curaggcontext as an argument. It was, in contrast to other aggregate related functions, only needed to fetch a memory context to copy the transition value into. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de https://postgr.es/m/5c371df7cee903e8cd4c685f90c6c72086d3a2dc.camel@j-davis.com
2020-02-24 23:39:22 +01:00
/* see comments above EEOP_AGG_PLAIN_TRANS_INIT_STRICT_BYVAL */
EEO_CASE(EEOP_AGG_PLAIN_TRANS_BYREF)
{
AggState *aggstate = castNode(AggState, state->parent);
AggStatePerTrans pertrans = op->d.agg_trans.pertrans;
AggStatePerGroup pergroup =
Pre-beta mechanical code beautification. Run pgindent, pgperltidy, and reformat-dat-files. This set of diffs is a bit larger than typical. We've updated to pg_bsd_indent 2.1.2, which properly indents variable declarations that have multi-line initialization expressions (the continuation lines are now indented one tab stop). We've also updated to perltidy version 20230309 and changed some of its settings, which reduces its desire to add whitespace to lines to make assignments etc. line up. Going forward, that should make for fewer random-seeming changes to existing code. Discussion: https://postgr.es/m/20230428092545.qfb3y5wcu4cm75ur@alvherre.pgsql
2023-05-19 23:24:48 +02:00
&aggstate->all_pergroups[op->d.agg_trans.setoff][op->d.agg_trans.transno];
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
expression eval: Reduce number of steps for agg transition invocations. Do so by combining the various steps that are part of aggregate transition function invocation into one larger step. As some of the current steps are only necessary for some aggregates, have one variant of the aggregate transition step for each possible combination. To avoid further manual copies of code in the different transition step implementations, move most of the code into helper functions marked as "always inline". The benefit of this change is an increase in performance when aggregating lots of rows. This comes in part due to the reduced number of indirect jumps due to the reduced number of steps, and in part by reducing redundant setup code across steps. This mainly benefits interpreted execution, but the code generated by JIT is also improved a bit. As a nice side-effect it also ends up making the code a bit simpler. A small additional optimization is removing the need to set aggstate->curaggcontext before calling ExecAggInitGroup, choosing to instead passign curaggcontext as an argument. It was, in contrast to other aggregate related functions, only needed to fetch a memory context to copy the transition value into. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de https://postgr.es/m/5c371df7cee903e8cd4c685f90c6c72086d3a2dc.camel@j-davis.com
2020-02-24 23:39:22 +01:00
Assert(!pertrans->transtypeByVal);
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
expression eval: Reduce number of steps for agg transition invocations. Do so by combining the various steps that are part of aggregate transition function invocation into one larger step. As some of the current steps are only necessary for some aggregates, have one variant of the aggregate transition step for each possible combination. To avoid further manual copies of code in the different transition step implementations, move most of the code into helper functions marked as "always inline". The benefit of this change is an increase in performance when aggregating lots of rows. This comes in part due to the reduced number of indirect jumps due to the reduced number of steps, and in part by reducing redundant setup code across steps. This mainly benefits interpreted execution, but the code generated by JIT is also improved a bit. As a nice side-effect it also ends up making the code a bit simpler. A small additional optimization is removing the need to set aggstate->curaggcontext before calling ExecAggInitGroup, choosing to instead passign curaggcontext as an argument. It was, in contrast to other aggregate related functions, only needed to fetch a memory context to copy the transition value into. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de https://postgr.es/m/5c371df7cee903e8cd4c685f90c6c72086d3a2dc.camel@j-davis.com
2020-02-24 23:39:22 +01:00
ExecAggPlainTransByRef(aggstate, pertrans, pergroup,
op->d.agg_trans.aggcontext,
op->d.agg_trans.setno);
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
EEO_NEXT();
}
Improve performance of ORDER BY / DISTINCT aggregates ORDER BY / DISTINCT aggreagtes have, since implemented in Postgres, been executed by always performing a sort in nodeAgg.c to sort the tuples in the current group into the correct order before calling the transition function on the sorted tuples. This was not great as often there might be an index that could have provided pre-sorted input and allowed the transition functions to be called as the rows come in, rather than having to store them in a tuplestore in order to sort them once all the tuples for the group have arrived. Here we change the planner so it requests a path with a sort order which supports the most amount of ORDER BY / DISTINCT aggregate functions and add new code to the executor to allow it to support the processing of ORDER BY / DISTINCT aggregates where the tuples are already sorted in the correct order. Since there can be many ORDER BY / DISTINCT aggregates in any given query level, it's very possible that we can't find an order that suits all of these aggregates. The sort order that the planner chooses is simply the one that suits the most aggregate functions. We take the most strictly sorted variation of each order and see how many aggregate functions can use that, then we try again with the order of the remaining aggregates to see if another order would suit more aggregate functions. For example: SELECT agg(a ORDER BY a),agg2(a ORDER BY a,b) ... would request the sort order to be {a, b} because {a} is a subset of the sort order of {a,b}, but; SELECT agg(a ORDER BY a),agg2(a ORDER BY c) ... would just pick a plan ordered by {a} (we give precedence to aggregates which are earlier in the targetlist). SELECT agg(a ORDER BY a),agg2(a ORDER BY b),agg3(a ORDER BY b) ... would choose to order by {b} since two aggregates suit that vs just one that requires input ordered by {a}. Author: David Rowley Reviewed-by: Ronan Dunklau, James Coleman, Ranier Vilela, Richard Guo, Tom Lane Discussion: https://postgr.es/m/CAApHDvpHzfo92%3DR4W0%2BxVua3BUYCKMckWAmo-2t_KiXN-wYH%3Dw%40mail.gmail.com
2022-08-02 13:11:45 +02:00
EEO_CASE(EEOP_AGG_PRESORTED_DISTINCT_SINGLE)
{
AggStatePerTrans pertrans = op->d.agg_presorted_distinctcheck.pertrans;
AggState *aggstate = castNode(AggState, state->parent);
if (ExecEvalPreOrderedDistinctSingle(aggstate, pertrans))
EEO_NEXT();
else
EEO_JUMP(op->d.agg_presorted_distinctcheck.jumpdistinct);
}
EEO_CASE(EEOP_AGG_PRESORTED_DISTINCT_MULTI)
{
AggState *aggstate = castNode(AggState, state->parent);
AggStatePerTrans pertrans = op->d.agg_presorted_distinctcheck.pertrans;
if (ExecEvalPreOrderedDistinctMulti(aggstate, pertrans))
EEO_NEXT();
else
EEO_JUMP(op->d.agg_presorted_distinctcheck.jumpdistinct);
}
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
/* process single-column ordered aggregate datum */
EEO_CASE(EEOP_AGG_ORDERED_TRANS_DATUM)
{
/* too complex for an inline implementation */
ExecEvalAggOrderedTransDatum(state, op, econtext);
EEO_NEXT();
}
/* process multi-column ordered aggregate tuple */
EEO_CASE(EEOP_AGG_ORDERED_TRANS_TUPLE)
{
/* too complex for an inline implementation */
ExecEvalAggOrderedTransTuple(state, op, econtext);
IS JSON predicate This patch intrdocuces the SQL standard IS JSON predicate. It operates on text and bytea values representing JSON as well as on the json and jsonb types. Each test has an IS and IS NOT variant. The tests are: IS JSON [VALUE] IS JSON ARRAY IS JSON OBJECT IS JSON SCALAR IS JSON WITH | WITHOUT UNIQUE KEYS These are mostly self-explanatory, but note that IS JSON WITHOUT UNIQUE KEYS is true whenever IS JSON is true, and IS JSON WITH UNIQUE KEYS is true whenever IS JSON is true except it IS JSON OBJECT is true and there are duplicate keys (which is never the case when applied to jsonb values). Nikita Glukhov Reviewers have included (in no particular order) Andres Freund, Alexander Korotkov, Pavel Stehule, Andrew Alsup, Erik Rijkers, Zihong Yu, Himanshu Upadhyaya, Daniel Gustafsson, Justin Pryzby. Discussion: https://postgr.es/m/cd0bb935-0158-78a7-08b5-904886deac4b@postgrespro.ru
2022-03-03 19:02:53 +01:00
EEO_NEXT();
}
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
EEO_CASE(EEOP_LAST)
{
/* unreachable */
Assert(false);
goto out;
}
}
out:
*isnull = state->resnull;
return state->resvalue;
}
Perform slot validity checks in a separate pass over expression. This reduces code duplication a bit, but the primary benefit that it makes JITing expression evaluation easier. When doing so we can't, as previously done in the interpreted case, really change opcode without recompiling. Nor dow we just carry around unnecessary branches to avoid re-checking over and over. As a minor side-effect this makes ExecEvalStepOp() O(log(N)) rather than O(N). Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2017-12-29 21:38:15 +01:00
/*
* Expression evaluation callback that performs extra checks before executing
* the expression. Declared extern so other methods of execution can use it
* too.
*/
Datum
ExecInterpExprStillValid(ExprState *state, ExprContext *econtext, bool *isNull)
{
/*
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
* First time through, check whether attribute matches Var. Might not be
* ok anymore, due to schema changes.
Perform slot validity checks in a separate pass over expression. This reduces code duplication a bit, but the primary benefit that it makes JITing expression evaluation easier. When doing so we can't, as previously done in the interpreted case, really change opcode without recompiling. Nor dow we just carry around unnecessary branches to avoid re-checking over and over. As a minor side-effect this makes ExecEvalStepOp() O(log(N)) rather than O(N). Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2017-12-29 21:38:15 +01:00
*/
CheckExprStillValid(state, econtext);
/* skip the check during further executions */
state->evalfunc = (ExprStateEvalFunc) state->evalfunc_private;
/* and actually execute */
return state->evalfunc(state, econtext, isNull);
}
/*
* Check that an expression is still valid in the face of potential schema
* changes since the plan has been created.
*/
void
CheckExprStillValid(ExprState *state, ExprContext *econtext)
{
TupleTableSlot *innerslot;
TupleTableSlot *outerslot;
TupleTableSlot *scanslot;
innerslot = econtext->ecxt_innertuple;
outerslot = econtext->ecxt_outertuple;
scanslot = econtext->ecxt_scantuple;
expression eval, jit: Minor code cleanups. This mostly consists of using C99 style for loops, moving variables into narrower scopes, and a smattering of other minor improvements. Done separately to make it easier to review patches with actual functional changes. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:04:50 +01:00
for (int i = 0; i < state->steps_len; i++)
Perform slot validity checks in a separate pass over expression. This reduces code duplication a bit, but the primary benefit that it makes JITing expression evaluation easier. When doing so we can't, as previously done in the interpreted case, really change opcode without recompiling. Nor dow we just carry around unnecessary branches to avoid re-checking over and over. As a minor side-effect this makes ExecEvalStepOp() O(log(N)) rather than O(N). Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2017-12-29 21:38:15 +01:00
{
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
ExprEvalStep *op = &state->steps[i];
Perform slot validity checks in a separate pass over expression. This reduces code duplication a bit, but the primary benefit that it makes JITing expression evaluation easier. When doing so we can't, as previously done in the interpreted case, really change opcode without recompiling. Nor dow we just carry around unnecessary branches to avoid re-checking over and over. As a minor side-effect this makes ExecEvalStepOp() O(log(N)) rather than O(N). Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2017-12-29 21:38:15 +01:00
switch (ExecEvalStepOp(state, op))
{
case EEOP_INNER_VAR:
{
int attnum = op->d.var.attnum;
CheckVarSlotCompatibility(innerslot, attnum + 1, op->d.var.vartype);
break;
}
case EEOP_OUTER_VAR:
{
int attnum = op->d.var.attnum;
CheckVarSlotCompatibility(outerslot, attnum + 1, op->d.var.vartype);
break;
}
case EEOP_SCAN_VAR:
{
int attnum = op->d.var.attnum;
CheckVarSlotCompatibility(scanslot, attnum + 1, op->d.var.vartype);
break;
}
default:
break;
}
}
}
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
/*
* Check whether a user attribute in a slot can be referenced by a Var
* expression. This should succeed unless there have been schema changes
* since the expression tree has been created.
*/
static void
CheckVarSlotCompatibility(TupleTableSlot *slot, int attnum, Oid vartype)
{
/*
* What we have to check for here is the possibility of an attribute
Make new expression eval code reject references to dropped columns. Formerly, a Var referencing an already-dropped column was allowed and would always produce a NULL value. However, that behavior was implemented in slot_getattr which the new expression code doesn't use; thus there is now a risk of returning theoretically-deleted data. We had regression test cases that purported to exercise this, but they failed to expose any problem, apparently because plpgsql filters the dropped column and produces an output tuple that has a NULL there already. Ideally the DROP or ALTER attempt in these test cases would get rejected due to dependency checks; but until that happens, let's modify the behavior so that we fail the query during executor start. This was already true for the related case of a column having changed type underneath us, and there's no obvious reason why we need to be laxer for dropped columns. In passing, adjust the error messages in CheckVarSlotCompatibility to include the composite type name. In the cases shown in the regression tests this is always just "record", but it should be more useful in actual stale-plan cases, where the slot tupdesc would be a table's tupdesc directly. Discussion: https://postgr.es/m/16803.1490723570@sss.pgh.pa.us
2017-03-29 00:05:03 +02:00
* having been dropped or changed in type since the plan tree was created.
* Ideally the plan will get invalidated and not re-used, but just in
* case, we keep these defenses. Fortunately it's sufficient to check
* once on the first time through.
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
*
* Note: ideally we'd check typmod as well as typid, but that seems
* impractical at the moment: in many cases the tupdesc will have been
* generated by ExecTypeFromTL(), and that can't guarantee to generate an
* accurate typmod in all cases, because some expression node types don't
Make new expression eval code reject references to dropped columns. Formerly, a Var referencing an already-dropped column was allowed and would always produce a NULL value. However, that behavior was implemented in slot_getattr which the new expression code doesn't use; thus there is now a risk of returning theoretically-deleted data. We had regression test cases that purported to exercise this, but they failed to expose any problem, apparently because plpgsql filters the dropped column and produces an output tuple that has a NULL there already. Ideally the DROP or ALTER attempt in these test cases would get rejected due to dependency checks; but until that happens, let's modify the behavior so that we fail the query during executor start. This was already true for the related case of a column having changed type underneath us, and there's no obvious reason why we need to be laxer for dropped columns. In passing, adjust the error messages in CheckVarSlotCompatibility to include the composite type name. In the cases shown in the regression tests this is always just "record", but it should be more useful in actual stale-plan cases, where the slot tupdesc would be a table's tupdesc directly. Discussion: https://postgr.es/m/16803.1490723570@sss.pgh.pa.us
2017-03-29 00:05:03 +02:00
* carry typmod. Fortunately, for precisely that reason, there should be
* no places with a critical dependency on the typmod of a value.
*
* System attributes don't require checking since their types never
* change.
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
*/
if (attnum > 0)
{
TupleDesc slot_tupdesc = slot->tts_tupleDescriptor;
Form_pg_attribute attr;
if (attnum > slot_tupdesc->natts) /* should never happen */
elog(ERROR, "attribute number %d exceeds number of columns %d",
attnum, slot_tupdesc->natts);
Change tupledesc->attrs[n] to TupleDescAttr(tupledesc, n). This is a mechanical change in preparation for a later commit that will change the layout of TupleDesc. Introducing a macro to abstract the details of where attributes are stored will allow us to change that in separate step and revise it in future. Author: Thomas Munro, editorialized by Andres Freund Reviewed-By: Andres Freund Discussion: https://postgr.es/m/CAEepm=0ZtQ-SpsgCyzzYpsXS6e=kZWqk3g5Ygn3MDV7A8dabUA@mail.gmail.com
2017-08-20 20:19:07 +02:00
attr = TupleDescAttr(slot_tupdesc, attnum - 1);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
Make new expression eval code reject references to dropped columns. Formerly, a Var referencing an already-dropped column was allowed and would always produce a NULL value. However, that behavior was implemented in slot_getattr which the new expression code doesn't use; thus there is now a risk of returning theoretically-deleted data. We had regression test cases that purported to exercise this, but they failed to expose any problem, apparently because plpgsql filters the dropped column and produces an output tuple that has a NULL there already. Ideally the DROP or ALTER attempt in these test cases would get rejected due to dependency checks; but until that happens, let's modify the behavior so that we fail the query during executor start. This was already true for the related case of a column having changed type underneath us, and there's no obvious reason why we need to be laxer for dropped columns. In passing, adjust the error messages in CheckVarSlotCompatibility to include the composite type name. In the cases shown in the regression tests this is always just "record", but it should be more useful in actual stale-plan cases, where the slot tupdesc would be a table's tupdesc directly. Discussion: https://postgr.es/m/16803.1490723570@sss.pgh.pa.us
2017-03-29 00:05:03 +02:00
if (attr->attisdropped)
ereport(ERROR,
(errcode(ERRCODE_UNDEFINED_COLUMN),
errmsg("attribute %d of type %s has been dropped",
attnum, format_type_be(slot_tupdesc->tdtypeid))));
if (vartype != attr->atttypid)
ereport(ERROR,
(errcode(ERRCODE_DATATYPE_MISMATCH),
errmsg("attribute %d of type %s has wrong type",
attnum, format_type_be(slot_tupdesc->tdtypeid)),
errdetail("Table has type %s, but query expects %s.",
format_type_be(attr->atttypid),
format_type_be(vartype))));
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
}
}
Verify that expected slot types match returned slot types. This is important so JIT compilation knows what kind of tuple slot the deforming routine can expect. There's also optimization potential for expression initialization without JIT compilation. It e.g. seems plausible to elide EEOP_*_FETCHSOME ops entirely when dealing with virtual slots. Author: Andres Freund Discussion: https://postgr.es/m/20181105210039.hh4vvi4vwoq5ba2q@alap3.anarazel.de
2018-11-16 07:00:30 +01:00
/*
* Verify that the slot is compatible with a EEOP_*_FETCHSOME operation.
*/
static void
CheckOpSlotCompatibility(ExprEvalStep *op, TupleTableSlot *slot)
{
#ifdef USE_ASSERT_CHECKING
/* there's nothing to check */
if (!op->d.fetch.fixed)
return;
/*
* Should probably fixed at some point, but for now it's easier to allow
Fix misc typos in comments. Spotted mostly by Fabien Coelho. Discussion: https://www.postgresql.org/message-id/alpine.DEB.2.21.1901230947050.16643@lancre
2019-01-23 12:39:00 +01:00
* buffer and heap tuples to be used interchangeably.
Verify that expected slot types match returned slot types. This is important so JIT compilation knows what kind of tuple slot the deforming routine can expect. There's also optimization potential for expression initialization without JIT compilation. It e.g. seems plausible to elide EEOP_*_FETCHSOME ops entirely when dealing with virtual slots. Author: Andres Freund Discussion: https://postgr.es/m/20181105210039.hh4vvi4vwoq5ba2q@alap3.anarazel.de
2018-11-16 07:00:30 +01:00
*/
Make TupleTableSlots extensible, finish split of existing slot type. This commit completes the work prepared in 1a0586de36, splitting the old TupleTableSlot implementation (which could store buffer, heap, minimal and virtual slots) into four different slot types. As described in the aforementioned commit, this is done with the goal of making tuple table slots extensible, to allow for pluggable table access methods. To achieve runtime extensibility for TupleTableSlots, operations on slots that can differ between types of slots are performed using the TupleTableSlotOps struct provided at slot creation time. That includes information from the size of TupleTableSlot struct to be allocated, initialization, deforming etc. See the struct's definition for more detailed information about callbacks TupleTableSlotOps. I decided to rename TTSOpsBufferTuple to TTSOpsBufferHeapTuple and ExecCopySlotTuple to ExecCopySlotHeapTuple, as that seems more consistent with other naming introduced in recent patches. There's plenty optimization potential in the slot implementation, but according to benchmarking the state after this commit has similar performance characteristics to before this set of changes, which seems sufficient. There's a few changes in execReplication.c that currently need to poke through the slot abstraction, that'll be repaired once the pluggable storage patchset provides the necessary infrastructure. Author: Andres Freund and Ashutosh Bapat, with changes by Amit Khandekar Discussion: https://postgr.es/m/20181105210039.hh4vvi4vwoq5ba2q@alap3.anarazel.de
2018-11-17 01:35:11 +01:00
if (slot->tts_ops == &TTSOpsBufferHeapTuple &&
Verify that expected slot types match returned slot types. This is important so JIT compilation knows what kind of tuple slot the deforming routine can expect. There's also optimization potential for expression initialization without JIT compilation. It e.g. seems plausible to elide EEOP_*_FETCHSOME ops entirely when dealing with virtual slots. Author: Andres Freund Discussion: https://postgr.es/m/20181105210039.hh4vvi4vwoq5ba2q@alap3.anarazel.de
2018-11-16 07:00:30 +01:00
op->d.fetch.kind == &TTSOpsHeapTuple)
return;
if (slot->tts_ops == &TTSOpsHeapTuple &&
Make TupleTableSlots extensible, finish split of existing slot type. This commit completes the work prepared in 1a0586de36, splitting the old TupleTableSlot implementation (which could store buffer, heap, minimal and virtual slots) into four different slot types. As described in the aforementioned commit, this is done with the goal of making tuple table slots extensible, to allow for pluggable table access methods. To achieve runtime extensibility for TupleTableSlots, operations on slots that can differ between types of slots are performed using the TupleTableSlotOps struct provided at slot creation time. That includes information from the size of TupleTableSlot struct to be allocated, initialization, deforming etc. See the struct's definition for more detailed information about callbacks TupleTableSlotOps. I decided to rename TTSOpsBufferTuple to TTSOpsBufferHeapTuple and ExecCopySlotTuple to ExecCopySlotHeapTuple, as that seems more consistent with other naming introduced in recent patches. There's plenty optimization potential in the slot implementation, but according to benchmarking the state after this commit has similar performance characteristics to before this set of changes, which seems sufficient. There's a few changes in execReplication.c that currently need to poke through the slot abstraction, that'll be repaired once the pluggable storage patchset provides the necessary infrastructure. Author: Andres Freund and Ashutosh Bapat, with changes by Amit Khandekar Discussion: https://postgr.es/m/20181105210039.hh4vvi4vwoq5ba2q@alap3.anarazel.de
2018-11-17 01:35:11 +01:00
op->d.fetch.kind == &TTSOpsBufferHeapTuple)
Verify that expected slot types match returned slot types. This is important so JIT compilation knows what kind of tuple slot the deforming routine can expect. There's also optimization potential for expression initialization without JIT compilation. It e.g. seems plausible to elide EEOP_*_FETCHSOME ops entirely when dealing with virtual slots. Author: Andres Freund Discussion: https://postgr.es/m/20181105210039.hh4vvi4vwoq5ba2q@alap3.anarazel.de
2018-11-16 07:00:30 +01:00
return;
/*
* At the moment we consider it OK if a virtual slot is used instead of a
* specific type of slot, as a virtual slot never needs to be deformed.
*/
if (slot->tts_ops == &TTSOpsVirtual)
return;
Assert(op->d.fetch.kind == slot->tts_ops);
#endif
}
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
/*
* get_cached_rowtype: utility function to lookup a rowtype tupdesc
*
* type_id, typmod: identity of the rowtype
Redesign the caching done by get_cached_rowtype(). Previously, get_cached_rowtype() cached a pointer to a reference-counted tuple descriptor from the typcache, relying on the ExprContextCallback mechanism to release the tupdesc refcount when the expression tree using the tupdesc was destroyed. This worked fine when it was designed, but the introduction of within-DO-block COMMITs broke it. The refcount is logged in a transaction-lifespan resource owner, but plpgsql won't destroy simple expressions made within the DO block (before its first commit) until the DO block is exited. That results in a warning about a leaked tupdesc refcount when the COMMIT destroys the original resource owner, and then an error about the active resource owner not holding a matching refcount when the expression is destroyed. To fix, get rid of the need to have a shutdown callback at all, by instead caching a pointer to the relevant typcache entry. Those survive for the life of the backend, so we needn't worry about the pointer becoming stale. (For registered RECORD types, we can still cache a pointer to the tupdesc, knowing that it won't change for the life of the backend.) This mechanism has been in use in plpgsql and expandedrecord.c since commit 4b93f5799, and seems to work well. This change requires modifying the ExprEvalStep structs used by the relevant expression step types, which is slightly worrisome for back-patching. However, there seems no good reason for extensions to be familiar with the details of these particular sub-structs. Per report from Rohit Bhogate. Back-patch to v11 where within-DO-block COMMITs became a thing. Discussion: https://postgr.es/m/CAAV6ZkQRCVBh8qAY+SZiHnz+U+FqAGBBDaDTjF2yiKa2nJSLKg@mail.gmail.com
2021-04-13 19:37:07 +02:00
* rowcache: space for caching identity info
* (rowcache->cacheptr must be initialized to NULL)
* changed: if not NULL, *changed is set to true on any update
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
*
Redesign the caching done by get_cached_rowtype(). Previously, get_cached_rowtype() cached a pointer to a reference-counted tuple descriptor from the typcache, relying on the ExprContextCallback mechanism to release the tupdesc refcount when the expression tree using the tupdesc was destroyed. This worked fine when it was designed, but the introduction of within-DO-block COMMITs broke it. The refcount is logged in a transaction-lifespan resource owner, but plpgsql won't destroy simple expressions made within the DO block (before its first commit) until the DO block is exited. That results in a warning about a leaked tupdesc refcount when the COMMIT destroys the original resource owner, and then an error about the active resource owner not holding a matching refcount when the expression is destroyed. To fix, get rid of the need to have a shutdown callback at all, by instead caching a pointer to the relevant typcache entry. Those survive for the life of the backend, so we needn't worry about the pointer becoming stale. (For registered RECORD types, we can still cache a pointer to the tupdesc, knowing that it won't change for the life of the backend.) This mechanism has been in use in plpgsql and expandedrecord.c since commit 4b93f5799, and seems to work well. This change requires modifying the ExprEvalStep structs used by the relevant expression step types, which is slightly worrisome for back-patching. However, there seems no good reason for extensions to be familiar with the details of these particular sub-structs. Per report from Rohit Bhogate. Back-patch to v11 where within-DO-block COMMITs became a thing. Discussion: https://postgr.es/m/CAAV6ZkQRCVBh8qAY+SZiHnz+U+FqAGBBDaDTjF2yiKa2nJSLKg@mail.gmail.com
2021-04-13 19:37:07 +02:00
* The returned TupleDesc is not guaranteed pinned; caller must pin it
* to use it across any operation that might incur cache invalidation.
* (The TupleDesc is always refcounted, so just use IncrTupleDescRefCount.)
*
* NOTE: because composite types can change contents, we must be prepared
* to re-do this during any node execution; cannot call just once during
* expression initialization.
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
*/
static TupleDesc
get_cached_rowtype(Oid type_id, int32 typmod,
Redesign the caching done by get_cached_rowtype(). Previously, get_cached_rowtype() cached a pointer to a reference-counted tuple descriptor from the typcache, relying on the ExprContextCallback mechanism to release the tupdesc refcount when the expression tree using the tupdesc was destroyed. This worked fine when it was designed, but the introduction of within-DO-block COMMITs broke it. The refcount is logged in a transaction-lifespan resource owner, but plpgsql won't destroy simple expressions made within the DO block (before its first commit) until the DO block is exited. That results in a warning about a leaked tupdesc refcount when the COMMIT destroys the original resource owner, and then an error about the active resource owner not holding a matching refcount when the expression is destroyed. To fix, get rid of the need to have a shutdown callback at all, by instead caching a pointer to the relevant typcache entry. Those survive for the life of the backend, so we needn't worry about the pointer becoming stale. (For registered RECORD types, we can still cache a pointer to the tupdesc, knowing that it won't change for the life of the backend.) This mechanism has been in use in plpgsql and expandedrecord.c since commit 4b93f5799, and seems to work well. This change requires modifying the ExprEvalStep structs used by the relevant expression step types, which is slightly worrisome for back-patching. However, there seems no good reason for extensions to be familiar with the details of these particular sub-structs. Per report from Rohit Bhogate. Back-patch to v11 where within-DO-block COMMITs became a thing. Discussion: https://postgr.es/m/CAAV6ZkQRCVBh8qAY+SZiHnz+U+FqAGBBDaDTjF2yiKa2nJSLKg@mail.gmail.com
2021-04-13 19:37:07 +02:00
ExprEvalRowtypeCache *rowcache,
bool *changed)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
Redesign the caching done by get_cached_rowtype(). Previously, get_cached_rowtype() cached a pointer to a reference-counted tuple descriptor from the typcache, relying on the ExprContextCallback mechanism to release the tupdesc refcount when the expression tree using the tupdesc was destroyed. This worked fine when it was designed, but the introduction of within-DO-block COMMITs broke it. The refcount is logged in a transaction-lifespan resource owner, but plpgsql won't destroy simple expressions made within the DO block (before its first commit) until the DO block is exited. That results in a warning about a leaked tupdesc refcount when the COMMIT destroys the original resource owner, and then an error about the active resource owner not holding a matching refcount when the expression is destroyed. To fix, get rid of the need to have a shutdown callback at all, by instead caching a pointer to the relevant typcache entry. Those survive for the life of the backend, so we needn't worry about the pointer becoming stale. (For registered RECORD types, we can still cache a pointer to the tupdesc, knowing that it won't change for the life of the backend.) This mechanism has been in use in plpgsql and expandedrecord.c since commit 4b93f5799, and seems to work well. This change requires modifying the ExprEvalStep structs used by the relevant expression step types, which is slightly worrisome for back-patching. However, there seems no good reason for extensions to be familiar with the details of these particular sub-structs. Per report from Rohit Bhogate. Back-patch to v11 where within-DO-block COMMITs became a thing. Discussion: https://postgr.es/m/CAAV6ZkQRCVBh8qAY+SZiHnz+U+FqAGBBDaDTjF2yiKa2nJSLKg@mail.gmail.com
2021-04-13 19:37:07 +02:00
if (type_id != RECORDOID)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
Redesign the caching done by get_cached_rowtype(). Previously, get_cached_rowtype() cached a pointer to a reference-counted tuple descriptor from the typcache, relying on the ExprContextCallback mechanism to release the tupdesc refcount when the expression tree using the tupdesc was destroyed. This worked fine when it was designed, but the introduction of within-DO-block COMMITs broke it. The refcount is logged in a transaction-lifespan resource owner, but plpgsql won't destroy simple expressions made within the DO block (before its first commit) until the DO block is exited. That results in a warning about a leaked tupdesc refcount when the COMMIT destroys the original resource owner, and then an error about the active resource owner not holding a matching refcount when the expression is destroyed. To fix, get rid of the need to have a shutdown callback at all, by instead caching a pointer to the relevant typcache entry. Those survive for the life of the backend, so we needn't worry about the pointer becoming stale. (For registered RECORD types, we can still cache a pointer to the tupdesc, knowing that it won't change for the life of the backend.) This mechanism has been in use in plpgsql and expandedrecord.c since commit 4b93f5799, and seems to work well. This change requires modifying the ExprEvalStep structs used by the relevant expression step types, which is slightly worrisome for back-patching. However, there seems no good reason for extensions to be familiar with the details of these particular sub-structs. Per report from Rohit Bhogate. Back-patch to v11 where within-DO-block COMMITs became a thing. Discussion: https://postgr.es/m/CAAV6ZkQRCVBh8qAY+SZiHnz+U+FqAGBBDaDTjF2yiKa2nJSLKg@mail.gmail.com
2021-04-13 19:37:07 +02:00
/*
* It's a named composite type, so use the regular typcache. Do a
* lookup first time through, or if the composite type changed. Note:
* "tupdesc_id == 0" may look redundant, but it protects against the
* admittedly-theoretical possibility that type_id was RECORDOID the
* last time through, so that the cacheptr isn't TypeCacheEntry *.
*/
TypeCacheEntry *typentry = (TypeCacheEntry *) rowcache->cacheptr;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
Redesign the caching done by get_cached_rowtype(). Previously, get_cached_rowtype() cached a pointer to a reference-counted tuple descriptor from the typcache, relying on the ExprContextCallback mechanism to release the tupdesc refcount when the expression tree using the tupdesc was destroyed. This worked fine when it was designed, but the introduction of within-DO-block COMMITs broke it. The refcount is logged in a transaction-lifespan resource owner, but plpgsql won't destroy simple expressions made within the DO block (before its first commit) until the DO block is exited. That results in a warning about a leaked tupdesc refcount when the COMMIT destroys the original resource owner, and then an error about the active resource owner not holding a matching refcount when the expression is destroyed. To fix, get rid of the need to have a shutdown callback at all, by instead caching a pointer to the relevant typcache entry. Those survive for the life of the backend, so we needn't worry about the pointer becoming stale. (For registered RECORD types, we can still cache a pointer to the tupdesc, knowing that it won't change for the life of the backend.) This mechanism has been in use in plpgsql and expandedrecord.c since commit 4b93f5799, and seems to work well. This change requires modifying the ExprEvalStep structs used by the relevant expression step types, which is slightly worrisome for back-patching. However, there seems no good reason for extensions to be familiar with the details of these particular sub-structs. Per report from Rohit Bhogate. Back-patch to v11 where within-DO-block COMMITs became a thing. Discussion: https://postgr.es/m/CAAV6ZkQRCVBh8qAY+SZiHnz+U+FqAGBBDaDTjF2yiKa2nJSLKg@mail.gmail.com
2021-04-13 19:37:07 +02:00
if (unlikely(typentry == NULL ||
rowcache->tupdesc_id == 0 ||
typentry->tupDesc_identifier != rowcache->tupdesc_id))
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
Redesign the caching done by get_cached_rowtype(). Previously, get_cached_rowtype() cached a pointer to a reference-counted tuple descriptor from the typcache, relying on the ExprContextCallback mechanism to release the tupdesc refcount when the expression tree using the tupdesc was destroyed. This worked fine when it was designed, but the introduction of within-DO-block COMMITs broke it. The refcount is logged in a transaction-lifespan resource owner, but plpgsql won't destroy simple expressions made within the DO block (before its first commit) until the DO block is exited. That results in a warning about a leaked tupdesc refcount when the COMMIT destroys the original resource owner, and then an error about the active resource owner not holding a matching refcount when the expression is destroyed. To fix, get rid of the need to have a shutdown callback at all, by instead caching a pointer to the relevant typcache entry. Those survive for the life of the backend, so we needn't worry about the pointer becoming stale. (For registered RECORD types, we can still cache a pointer to the tupdesc, knowing that it won't change for the life of the backend.) This mechanism has been in use in plpgsql and expandedrecord.c since commit 4b93f5799, and seems to work well. This change requires modifying the ExprEvalStep structs used by the relevant expression step types, which is slightly worrisome for back-patching. However, there seems no good reason for extensions to be familiar with the details of these particular sub-structs. Per report from Rohit Bhogate. Back-patch to v11 where within-DO-block COMMITs became a thing. Discussion: https://postgr.es/m/CAAV6ZkQRCVBh8qAY+SZiHnz+U+FqAGBBDaDTjF2yiKa2nJSLKg@mail.gmail.com
2021-04-13 19:37:07 +02:00
typentry = lookup_type_cache(type_id, TYPECACHE_TUPDESC);
if (typentry->tupDesc == NULL)
ereport(ERROR,
(errcode(ERRCODE_WRONG_OBJECT_TYPE),
errmsg("type %s is not composite",
format_type_be(type_id))));
rowcache->cacheptr = (void *) typentry;
rowcache->tupdesc_id = typentry->tupDesc_identifier;
if (changed)
*changed = true;
}
return typentry->tupDesc;
}
else
{
/*
* A RECORD type, once registered, doesn't change for the life of the
* backend. So we don't need a typcache entry as such, which is good
* because there isn't one. It's possible that the caller is asking
* about a different type than before, though.
*/
TupleDesc tupDesc = (TupleDesc) rowcache->cacheptr;
if (unlikely(tupDesc == NULL ||
rowcache->tupdesc_id != 0 ||
type_id != tupDesc->tdtypeid ||
typmod != tupDesc->tdtypmod))
{
tupDesc = lookup_rowtype_tupdesc(type_id, typmod);
/* Drop pin acquired by lookup_rowtype_tupdesc */
ReleaseTupleDesc(tupDesc);
rowcache->cacheptr = (void *) tupDesc;
rowcache->tupdesc_id = 0; /* not a valid value for non-RECORD */
if (changed)
*changed = true;
}
return tupDesc;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
}
}
/*
* Fast-path functions, for very simple expressions
*/
Reduce code duplication for ExecJust*Var operations. This is mainly in preparation for adding further fastpath evaluation routines. Also reorder ExecJust*Var functions to be consistent with the order in which they're used. Author: Andres Freund Discussion: https://postgr.es/m/CAE-ML+9OKSN71+mHtfMD-L24oDp8dGTfaVjDU6U+j+FNAW5kRQ@mail.gmail.com
2019-10-01 00:00:21 +02:00
/* implementation of ExecJust(Inner|Outer|Scan)Var */
static pg_attribute_always_inline Datum
ExecJustVarImpl(ExprState *state, TupleTableSlot *slot, bool *isnull)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
ExprEvalStep *op = &state->steps[1];
int attnum = op->d.var.attnum + 1;
Verify that expected slot types match returned slot types. This is important so JIT compilation knows what kind of tuple slot the deforming routine can expect. There's also optimization potential for expression initialization without JIT compilation. It e.g. seems plausible to elide EEOP_*_FETCHSOME ops entirely when dealing with virtual slots. Author: Andres Freund Discussion: https://postgr.es/m/20181105210039.hh4vvi4vwoq5ba2q@alap3.anarazel.de
2018-11-16 07:00:30 +01:00
CheckOpSlotCompatibility(&state->steps[0], slot);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
/*
* Since we use slot_getattr(), we don't need to implement the FETCHSOME
* step explicitly, and we also needn't Assert that the attnum is in range
* --- slot_getattr() will take care of any problems.
*/
return slot_getattr(slot, attnum, isnull);
}
Reduce code duplication for ExecJust*Var operations. This is mainly in preparation for adding further fastpath evaluation routines. Also reorder ExecJust*Var functions to be consistent with the order in which they're used. Author: Andres Freund Discussion: https://postgr.es/m/CAE-ML+9OKSN71+mHtfMD-L24oDp8dGTfaVjDU6U+j+FNAW5kRQ@mail.gmail.com
2019-10-01 00:00:21 +02:00
/* Simple reference to inner Var */
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
static Datum
Reduce code duplication for ExecJust*Var operations. This is mainly in preparation for adding further fastpath evaluation routines. Also reorder ExecJust*Var functions to be consistent with the order in which they're used. Author: Andres Freund Discussion: https://postgr.es/m/CAE-ML+9OKSN71+mHtfMD-L24oDp8dGTfaVjDU6U+j+FNAW5kRQ@mail.gmail.com
2019-10-01 00:00:21 +02:00
ExecJustInnerVar(ExprState *state, ExprContext *econtext, bool *isnull)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
Reduce code duplication for ExecJust*Var operations. This is mainly in preparation for adding further fastpath evaluation routines. Also reorder ExecJust*Var functions to be consistent with the order in which they're used. Author: Andres Freund Discussion: https://postgr.es/m/CAE-ML+9OKSN71+mHtfMD-L24oDp8dGTfaVjDU6U+j+FNAW5kRQ@mail.gmail.com
2019-10-01 00:00:21 +02:00
return ExecJustVarImpl(state, econtext->ecxt_innertuple, isnull);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
}
Reduce code duplication for ExecJust*Var operations. This is mainly in preparation for adding further fastpath evaluation routines. Also reorder ExecJust*Var functions to be consistent with the order in which they're used. Author: Andres Freund Discussion: https://postgr.es/m/CAE-ML+9OKSN71+mHtfMD-L24oDp8dGTfaVjDU6U+j+FNAW5kRQ@mail.gmail.com
2019-10-01 00:00:21 +02:00
/* Simple reference to outer Var */
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
static Datum
Reduce code duplication for ExecJust*Var operations. This is mainly in preparation for adding further fastpath evaluation routines. Also reorder ExecJust*Var functions to be consistent with the order in which they're used. Author: Andres Freund Discussion: https://postgr.es/m/CAE-ML+9OKSN71+mHtfMD-L24oDp8dGTfaVjDU6U+j+FNAW5kRQ@mail.gmail.com
2019-10-01 00:00:21 +02:00
ExecJustOuterVar(ExprState *state, ExprContext *econtext, bool *isnull)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
Reduce code duplication for ExecJust*Var operations. This is mainly in preparation for adding further fastpath evaluation routines. Also reorder ExecJust*Var functions to be consistent with the order in which they're used. Author: Andres Freund Discussion: https://postgr.es/m/CAE-ML+9OKSN71+mHtfMD-L24oDp8dGTfaVjDU6U+j+FNAW5kRQ@mail.gmail.com
2019-10-01 00:00:21 +02:00
return ExecJustVarImpl(state, econtext->ecxt_outertuple, isnull);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
}
Reduce code duplication for ExecJust*Var operations. This is mainly in preparation for adding further fastpath evaluation routines. Also reorder ExecJust*Var functions to be consistent with the order in which they're used. Author: Andres Freund Discussion: https://postgr.es/m/CAE-ML+9OKSN71+mHtfMD-L24oDp8dGTfaVjDU6U+j+FNAW5kRQ@mail.gmail.com
2019-10-01 00:00:21 +02:00
/* Simple reference to scan Var */
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
static Datum
Reduce code duplication for ExecJust*Var operations. This is mainly in preparation for adding further fastpath evaluation routines. Also reorder ExecJust*Var functions to be consistent with the order in which they're used. Author: Andres Freund Discussion: https://postgr.es/m/CAE-ML+9OKSN71+mHtfMD-L24oDp8dGTfaVjDU6U+j+FNAW5kRQ@mail.gmail.com
2019-10-01 00:00:21 +02:00
ExecJustScanVar(ExprState *state, ExprContext *econtext, bool *isnull)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
Reduce code duplication for ExecJust*Var operations. This is mainly in preparation for adding further fastpath evaluation routines. Also reorder ExecJust*Var functions to be consistent with the order in which they're used. Author: Andres Freund Discussion: https://postgr.es/m/CAE-ML+9OKSN71+mHtfMD-L24oDp8dGTfaVjDU6U+j+FNAW5kRQ@mail.gmail.com
2019-10-01 00:00:21 +02:00
return ExecJustVarImpl(state, econtext->ecxt_scantuple, isnull);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
}
Reduce code duplication for ExecJust*Var operations. This is mainly in preparation for adding further fastpath evaluation routines. Also reorder ExecJust*Var functions to be consistent with the order in which they're used. Author: Andres Freund Discussion: https://postgr.es/m/CAE-ML+9OKSN71+mHtfMD-L24oDp8dGTfaVjDU6U+j+FNAW5kRQ@mail.gmail.com
2019-10-01 00:00:21 +02:00
/* implementation of ExecJustAssign(Inner|Outer|Scan)Var */
static pg_attribute_always_inline Datum
ExecJustAssignVarImpl(ExprState *state, TupleTableSlot *inslot, bool *isnull)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
ExprEvalStep *op = &state->steps[1];
int attnum = op->d.assign_var.attnum + 1;
int resultnum = op->d.assign_var.resultnum;
TupleTableSlot *outslot = state->resultslot;
Verify that expected slot types match returned slot types. This is important so JIT compilation knows what kind of tuple slot the deforming routine can expect. There's also optimization potential for expression initialization without JIT compilation. It e.g. seems plausible to elide EEOP_*_FETCHSOME ops entirely when dealing with virtual slots. Author: Andres Freund Discussion: https://postgr.es/m/20181105210039.hh4vvi4vwoq5ba2q@alap3.anarazel.de
2018-11-16 07:00:30 +01:00
CheckOpSlotCompatibility(&state->steps[0], inslot);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
/*
* We do not need CheckVarSlotCompatibility here; that was taken care of
* at compilation time.
*
* Since we use slot_getattr(), we don't need to implement the FETCHSOME
* step explicitly, and we also needn't Assert that the attnum is in range
Fix mishandling of resjunk columns in ON CONFLICT ... UPDATE tlists. It's unusual to have any resjunk columns in an ON CONFLICT ... UPDATE list, but it can happen when MULTIEXPR_SUBLINK SubPlans are present. If it happens, the ON CONFLICT UPDATE code path would end up storing tuples that include the values of the extra resjunk columns. That's fairly harmless in the short run, but if new columns are added to the table then the values would become accessible, possibly leading to malfunctions if they don't match the datatypes of the new columns. This had escaped notice through a confluence of missing sanity checks, including * There's no cross-check that a tuple presented to heap_insert or heap_update matches the table rowtype. While it's difficult to check that fully at reasonable cost, we can easily add assertions that there aren't too many columns. * The output-column-assignment cases in execExprInterp.c lacked any sanity checks on the output column numbers, which seems like an oversight considering there are plenty of assertion checks on input column numbers. Add assertions there too. * We failed to apply nodeModifyTable's ExecCheckPlanOutput() to the ON CONFLICT UPDATE tlist. That wouldn't have caught this specific error, since that function is chartered to ignore resjunk columns; but it sure seems like a bad omission now that we've seen this bug. In HEAD, the right way to fix this is to make the processing of ON CONFLICT UPDATE tlists work the same as regular UPDATE tlists now do, that is don't add "SET x = x" entries, and use ExecBuildUpdateProjection to evaluate the tlist and combine it with old values of the not-set columns. This adds a little complication to ExecBuildUpdateProjection, but allows removal of a comparable amount of now-dead code from the planner. In the back branches, the most expedient solution seems to be to (a) use an output slot for the ON CONFLICT UPDATE projection that actually matches the target table, and then (b) invent a variant of ExecBuildProjectionInfo that can be told to not store values resulting from resjunk columns, so it doesn't try to store into nonexistent columns of the output slot. (We can't simply ignore the resjunk columns altogether; they have to be evaluated for MULTIEXPR_SUBLINK to work.) This works back to v10. In 9.6, projections work much differently and we can't cheaply give them such an option. The 9.6 version of this patch works by inserting a JunkFilter when it's necessary to get rid of resjunk columns. In addition, v11 and up have the reverse problem when trying to perform ON CONFLICT UPDATE on a partitioned table. Through a further oversight, adjust_partition_tlist() discarded resjunk columns when re-ordering the ON CONFLICT UPDATE tlist to match a partition. This accidentally prevented the storing-bogus-tuples problem, but at the cost that MULTIEXPR_SUBLINK cases didn't work, typically crashing if more than one row has to be updated. Fix by preserving resjunk columns in that routine. (I failed to resist the temptation to add more assertions there too, and to do some minor code beautification.) Per report from Andres Freund. Back-patch to all supported branches. Security: CVE-2021-32028
2021-05-10 17:02:29 +02:00
* --- slot_getattr() will take care of any problems. Nonetheless, check
* that resultnum is in range.
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
*/
Fix mishandling of resjunk columns in ON CONFLICT ... UPDATE tlists. It's unusual to have any resjunk columns in an ON CONFLICT ... UPDATE list, but it can happen when MULTIEXPR_SUBLINK SubPlans are present. If it happens, the ON CONFLICT UPDATE code path would end up storing tuples that include the values of the extra resjunk columns. That's fairly harmless in the short run, but if new columns are added to the table then the values would become accessible, possibly leading to malfunctions if they don't match the datatypes of the new columns. This had escaped notice through a confluence of missing sanity checks, including * There's no cross-check that a tuple presented to heap_insert or heap_update matches the table rowtype. While it's difficult to check that fully at reasonable cost, we can easily add assertions that there aren't too many columns. * The output-column-assignment cases in execExprInterp.c lacked any sanity checks on the output column numbers, which seems like an oversight considering there are plenty of assertion checks on input column numbers. Add assertions there too. * We failed to apply nodeModifyTable's ExecCheckPlanOutput() to the ON CONFLICT UPDATE tlist. That wouldn't have caught this specific error, since that function is chartered to ignore resjunk columns; but it sure seems like a bad omission now that we've seen this bug. In HEAD, the right way to fix this is to make the processing of ON CONFLICT UPDATE tlists work the same as regular UPDATE tlists now do, that is don't add "SET x = x" entries, and use ExecBuildUpdateProjection to evaluate the tlist and combine it with old values of the not-set columns. This adds a little complication to ExecBuildUpdateProjection, but allows removal of a comparable amount of now-dead code from the planner. In the back branches, the most expedient solution seems to be to (a) use an output slot for the ON CONFLICT UPDATE projection that actually matches the target table, and then (b) invent a variant of ExecBuildProjectionInfo that can be told to not store values resulting from resjunk columns, so it doesn't try to store into nonexistent columns of the output slot. (We can't simply ignore the resjunk columns altogether; they have to be evaluated for MULTIEXPR_SUBLINK to work.) This works back to v10. In 9.6, projections work much differently and we can't cheaply give them such an option. The 9.6 version of this patch works by inserting a JunkFilter when it's necessary to get rid of resjunk columns. In addition, v11 and up have the reverse problem when trying to perform ON CONFLICT UPDATE on a partitioned table. Through a further oversight, adjust_partition_tlist() discarded resjunk columns when re-ordering the ON CONFLICT UPDATE tlist to match a partition. This accidentally prevented the storing-bogus-tuples problem, but at the cost that MULTIEXPR_SUBLINK cases didn't work, typically crashing if more than one row has to be updated. Fix by preserving resjunk columns in that routine. (I failed to resist the temptation to add more assertions there too, and to do some minor code beautification.) Per report from Andres Freund. Back-patch to all supported branches. Security: CVE-2021-32028
2021-05-10 17:02:29 +02:00
Assert(resultnum >= 0 && resultnum < outslot->tts_tupleDescriptor->natts);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
outslot->tts_values[resultnum] =
slot_getattr(inslot, attnum, &outslot->tts_isnull[resultnum]);
return 0;
}
Reduce code duplication for ExecJust*Var operations. This is mainly in preparation for adding further fastpath evaluation routines. Also reorder ExecJust*Var functions to be consistent with the order in which they're used. Author: Andres Freund Discussion: https://postgr.es/m/CAE-ML+9OKSN71+mHtfMD-L24oDp8dGTfaVjDU6U+j+FNAW5kRQ@mail.gmail.com
2019-10-01 00:00:21 +02:00
/* Evaluate inner Var and assign to appropriate column of result tuple */
static Datum
ExecJustAssignInnerVar(ExprState *state, ExprContext *econtext, bool *isnull)
{
return ExecJustAssignVarImpl(state, econtext->ecxt_innertuple, isnull);
}
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
/* Evaluate outer Var and assign to appropriate column of result tuple */
static Datum
ExecJustAssignOuterVar(ExprState *state, ExprContext *econtext, bool *isnull)
{
Reduce code duplication for ExecJust*Var operations. This is mainly in preparation for adding further fastpath evaluation routines. Also reorder ExecJust*Var functions to be consistent with the order in which they're used. Author: Andres Freund Discussion: https://postgr.es/m/CAE-ML+9OKSN71+mHtfMD-L24oDp8dGTfaVjDU6U+j+FNAW5kRQ@mail.gmail.com
2019-10-01 00:00:21 +02:00
return ExecJustAssignVarImpl(state, econtext->ecxt_outertuple, isnull);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
}
/* Evaluate scan Var and assign to appropriate column of result tuple */
static Datum
ExecJustAssignScanVar(ExprState *state, ExprContext *econtext, bool *isnull)
{
Reduce code duplication for ExecJust*Var operations. This is mainly in preparation for adding further fastpath evaluation routines. Also reorder ExecJust*Var functions to be consistent with the order in which they're used. Author: Andres Freund Discussion: https://postgr.es/m/CAE-ML+9OKSN71+mHtfMD-L24oDp8dGTfaVjDU6U+j+FNAW5kRQ@mail.gmail.com
2019-10-01 00:00:21 +02:00
return ExecJustAssignVarImpl(state, econtext->ecxt_scantuple, isnull);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
}
Support arrays over domains. Allowing arrays with a domain type as their element type was left un-done in the original domain patch, but not for any very good reason. This omission leads to such surprising results as array_agg() not working on a domain column, because the parser can't identify a suitable output type for the polymorphic aggregate. In order to fix this, first clean up the APIs of coerce_to_domain() and some internal functions in parse_coerce.c so that we consistently pass around a CoercionContext along with CoercionForm. Previously, we sometimes passed an "isExplicit" boolean flag instead, which is strictly less information; and coerce_to_domain() didn't even get that, but instead had to reverse-engineer isExplicit from CoercionForm. That's contrary to the documentation in primnodes.h that says that CoercionForm only affects display and not semantics. I don't think this change fixes any live bugs, but it makes things more consistent. The main reason for doing it though is that now build_coercion_expression() receives ccontext, which it needs in order to be able to recursively invoke coerce_to_target_type(). Next, reimplement ArrayCoerceExpr so that the node does not directly know any details of what has to be done to the individual array elements while performing the array coercion. Instead, the per-element processing is represented by a sub-expression whose input is a source array element and whose output is a target array element. This simplifies life in parse_coerce.c, because it can build that sub-expression by a recursive invocation of coerce_to_target_type(). The executor now handles the per-element processing as a compiled expression instead of hard-wired code. The main advantage of this is that we can use a single ArrayCoerceExpr to handle as many as three successive steps per element: base type conversion, typmod coercion, and domain constraint checking. The old code used two stacked ArrayCoerceExprs to handle type + typmod coercion, which was pretty inefficient, and adding yet another array deconstruction to do domain constraint checking seemed very unappetizing. In the case where we just need a single, very simple coercion function, doing this straightforwardly leads to a noticeable increase in the per-array-element runtime cost. Hence, add an additional shortcut evalfunc in execExprInterp.c that skips unnecessary overhead for that specific form of expression. The runtime speed of simple cases is within 1% or so of where it was before, while cases that previously required two levels of array processing are significantly faster. Finally, create an implicit array type for every domain type, as we do for base types, enums, etc. Everything except the array-coercion case seems to just work without further effort. Tom Lane, reviewed by Andrew Dunstan Discussion: https://postgr.es/m/9852.1499791473@sss.pgh.pa.us
2017-09-30 19:40:56 +02:00
/* Evaluate CASE_TESTVAL and apply a strict function to it */
static Datum
ExecJustApplyFuncToCase(ExprState *state, ExprContext *econtext, bool *isnull)
{
ExprEvalStep *op = &state->steps[0];
FunctionCallInfo fcinfo;
Change function call information to be variable length. Before this change FunctionCallInfoData, the struct arguments etc for V1 function calls are stored in, always had space for FUNC_MAX_ARGS/100 arguments, storing datums and their nullness in two arrays. For nearly every function call 100 arguments is far more than needed, therefore wasting memory. Arg and argnull being two separate arrays also guarantees that to access a single argument, two cachelines have to be touched. Change the layout so there's a single variable-length array with pairs of value / isnull. That drastically reduces memory consumption for most function calls (on x86-64 a two argument function now uses 64bytes, previously 936 bytes), and makes it very likely that argument value and its nullness are on the same cacheline. Arguments are stored in a new NullableDatum struct, which, due to padding, needs more memory per argument than before. But as usually far fewer arguments are stored, and individual arguments are cheaper to access, that's still a clear win. It's likely that there's other places where conversion to NullableDatum arrays would make sense, e.g. TupleTableSlots, but that's for another commit. Because the function call information is now variable-length allocations have to take the number of arguments into account. For heap allocations that can be done with SizeForFunctionCallInfoData(), for on-stack allocations there's a new LOCAL_FCINFO(name, nargs) macro that helps to allocate an appropriately sized and aligned variable. Some places with stack allocation function call information don't know the number of arguments at compile time, and currently variably sized stack allocations aren't allowed in postgres. Therefore allow for FUNC_MAX_ARGS space in these cases. They're not that common, so for now that seems acceptable. Because of the need to allocate FunctionCallInfo of the appropriate size, older extensions may need to update their code. To avoid subtle breakages, the FunctionCallInfoData struct has been renamed to FunctionCallInfoBaseData. Most code only references FunctionCallInfo, so that shouldn't cause much collateral damage. This change is also a prerequisite for more efficient expression JIT compilation (by allocating the function call information on the stack, allowing LLVM to optimize it away); previously the size of the call information caused problems inside LLVM's optimizer. Author: Andres Freund Reviewed-By: Tom Lane Discussion: https://postgr.es/m/20180605172952.x34m5uz6ju6enaem@alap3.anarazel.de
2019-01-26 23:17:52 +01:00
NullableDatum *args;
expression eval, jit: Minor code cleanups. This mostly consists of using C99 style for loops, moving variables into narrower scopes, and a smattering of other minor improvements. Done separately to make it easier to review patches with actual functional changes. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:04:50 +01:00
int nargs;
Support arrays over domains. Allowing arrays with a domain type as their element type was left un-done in the original domain patch, but not for any very good reason. This omission leads to such surprising results as array_agg() not working on a domain column, because the parser can't identify a suitable output type for the polymorphic aggregate. In order to fix this, first clean up the APIs of coerce_to_domain() and some internal functions in parse_coerce.c so that we consistently pass around a CoercionContext along with CoercionForm. Previously, we sometimes passed an "isExplicit" boolean flag instead, which is strictly less information; and coerce_to_domain() didn't even get that, but instead had to reverse-engineer isExplicit from CoercionForm. That's contrary to the documentation in primnodes.h that says that CoercionForm only affects display and not semantics. I don't think this change fixes any live bugs, but it makes things more consistent. The main reason for doing it though is that now build_coercion_expression() receives ccontext, which it needs in order to be able to recursively invoke coerce_to_target_type(). Next, reimplement ArrayCoerceExpr so that the node does not directly know any details of what has to be done to the individual array elements while performing the array coercion. Instead, the per-element processing is represented by a sub-expression whose input is a source array element and whose output is a target array element. This simplifies life in parse_coerce.c, because it can build that sub-expression by a recursive invocation of coerce_to_target_type(). The executor now handles the per-element processing as a compiled expression instead of hard-wired code. The main advantage of this is that we can use a single ArrayCoerceExpr to handle as many as three successive steps per element: base type conversion, typmod coercion, and domain constraint checking. The old code used two stacked ArrayCoerceExprs to handle type + typmod coercion, which was pretty inefficient, and adding yet another array deconstruction to do domain constraint checking seemed very unappetizing. In the case where we just need a single, very simple coercion function, doing this straightforwardly leads to a noticeable increase in the per-array-element runtime cost. Hence, add an additional shortcut evalfunc in execExprInterp.c that skips unnecessary overhead for that specific form of expression. The runtime speed of simple cases is within 1% or so of where it was before, while cases that previously required two levels of array processing are significantly faster. Finally, create an implicit array type for every domain type, as we do for base types, enums, etc. Everything except the array-coercion case seems to just work without further effort. Tom Lane, reviewed by Andrew Dunstan Discussion: https://postgr.es/m/9852.1499791473@sss.pgh.pa.us
2017-09-30 19:40:56 +02:00
Datum d;
/*
* XXX with some redesign of the CaseTestExpr mechanism, maybe we could
* get rid of this data shuffling?
*/
*op->resvalue = *op->d.casetest.value;
*op->resnull = *op->d.casetest.isnull;
op++;
expression eval, jit: Minor code cleanups. This mostly consists of using C99 style for loops, moving variables into narrower scopes, and a smattering of other minor improvements. Done separately to make it easier to review patches with actual functional changes. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:04:50 +01:00
nargs = op->d.func.nargs;
Support arrays over domains. Allowing arrays with a domain type as their element type was left un-done in the original domain patch, but not for any very good reason. This omission leads to such surprising results as array_agg() not working on a domain column, because the parser can't identify a suitable output type for the polymorphic aggregate. In order to fix this, first clean up the APIs of coerce_to_domain() and some internal functions in parse_coerce.c so that we consistently pass around a CoercionContext along with CoercionForm. Previously, we sometimes passed an "isExplicit" boolean flag instead, which is strictly less information; and coerce_to_domain() didn't even get that, but instead had to reverse-engineer isExplicit from CoercionForm. That's contrary to the documentation in primnodes.h that says that CoercionForm only affects display and not semantics. I don't think this change fixes any live bugs, but it makes things more consistent. The main reason for doing it though is that now build_coercion_expression() receives ccontext, which it needs in order to be able to recursively invoke coerce_to_target_type(). Next, reimplement ArrayCoerceExpr so that the node does not directly know any details of what has to be done to the individual array elements while performing the array coercion. Instead, the per-element processing is represented by a sub-expression whose input is a source array element and whose output is a target array element. This simplifies life in parse_coerce.c, because it can build that sub-expression by a recursive invocation of coerce_to_target_type(). The executor now handles the per-element processing as a compiled expression instead of hard-wired code. The main advantage of this is that we can use a single ArrayCoerceExpr to handle as many as three successive steps per element: base type conversion, typmod coercion, and domain constraint checking. The old code used two stacked ArrayCoerceExprs to handle type + typmod coercion, which was pretty inefficient, and adding yet another array deconstruction to do domain constraint checking seemed very unappetizing. In the case where we just need a single, very simple coercion function, doing this straightforwardly leads to a noticeable increase in the per-array-element runtime cost. Hence, add an additional shortcut evalfunc in execExprInterp.c that skips unnecessary overhead for that specific form of expression. The runtime speed of simple cases is within 1% or so of where it was before, while cases that previously required two levels of array processing are significantly faster. Finally, create an implicit array type for every domain type, as we do for base types, enums, etc. Everything except the array-coercion case seems to just work without further effort. Tom Lane, reviewed by Andrew Dunstan Discussion: https://postgr.es/m/9852.1499791473@sss.pgh.pa.us
2017-09-30 19:40:56 +02:00
fcinfo = op->d.func.fcinfo_data;
Change function call information to be variable length. Before this change FunctionCallInfoData, the struct arguments etc for V1 function calls are stored in, always had space for FUNC_MAX_ARGS/100 arguments, storing datums and their nullness in two arrays. For nearly every function call 100 arguments is far more than needed, therefore wasting memory. Arg and argnull being two separate arrays also guarantees that to access a single argument, two cachelines have to be touched. Change the layout so there's a single variable-length array with pairs of value / isnull. That drastically reduces memory consumption for most function calls (on x86-64 a two argument function now uses 64bytes, previously 936 bytes), and makes it very likely that argument value and its nullness are on the same cacheline. Arguments are stored in a new NullableDatum struct, which, due to padding, needs more memory per argument than before. But as usually far fewer arguments are stored, and individual arguments are cheaper to access, that's still a clear win. It's likely that there's other places where conversion to NullableDatum arrays would make sense, e.g. TupleTableSlots, but that's for another commit. Because the function call information is now variable-length allocations have to take the number of arguments into account. For heap allocations that can be done with SizeForFunctionCallInfoData(), for on-stack allocations there's a new LOCAL_FCINFO(name, nargs) macro that helps to allocate an appropriately sized and aligned variable. Some places with stack allocation function call information don't know the number of arguments at compile time, and currently variably sized stack allocations aren't allowed in postgres. Therefore allow for FUNC_MAX_ARGS space in these cases. They're not that common, so for now that seems acceptable. Because of the need to allocate FunctionCallInfo of the appropriate size, older extensions may need to update their code. To avoid subtle breakages, the FunctionCallInfoData struct has been renamed to FunctionCallInfoBaseData. Most code only references FunctionCallInfo, so that shouldn't cause much collateral damage. This change is also a prerequisite for more efficient expression JIT compilation (by allocating the function call information on the stack, allowing LLVM to optimize it away); previously the size of the call information caused problems inside LLVM's optimizer. Author: Andres Freund Reviewed-By: Tom Lane Discussion: https://postgr.es/m/20180605172952.x34m5uz6ju6enaem@alap3.anarazel.de
2019-01-26 23:17:52 +01:00
args = fcinfo->args;
Support arrays over domains. Allowing arrays with a domain type as their element type was left un-done in the original domain patch, but not for any very good reason. This omission leads to such surprising results as array_agg() not working on a domain column, because the parser can't identify a suitable output type for the polymorphic aggregate. In order to fix this, first clean up the APIs of coerce_to_domain() and some internal functions in parse_coerce.c so that we consistently pass around a CoercionContext along with CoercionForm. Previously, we sometimes passed an "isExplicit" boolean flag instead, which is strictly less information; and coerce_to_domain() didn't even get that, but instead had to reverse-engineer isExplicit from CoercionForm. That's contrary to the documentation in primnodes.h that says that CoercionForm only affects display and not semantics. I don't think this change fixes any live bugs, but it makes things more consistent. The main reason for doing it though is that now build_coercion_expression() receives ccontext, which it needs in order to be able to recursively invoke coerce_to_target_type(). Next, reimplement ArrayCoerceExpr so that the node does not directly know any details of what has to be done to the individual array elements while performing the array coercion. Instead, the per-element processing is represented by a sub-expression whose input is a source array element and whose output is a target array element. This simplifies life in parse_coerce.c, because it can build that sub-expression by a recursive invocation of coerce_to_target_type(). The executor now handles the per-element processing as a compiled expression instead of hard-wired code. The main advantage of this is that we can use a single ArrayCoerceExpr to handle as many as three successive steps per element: base type conversion, typmod coercion, and domain constraint checking. The old code used two stacked ArrayCoerceExprs to handle type + typmod coercion, which was pretty inefficient, and adding yet another array deconstruction to do domain constraint checking seemed very unappetizing. In the case where we just need a single, very simple coercion function, doing this straightforwardly leads to a noticeable increase in the per-array-element runtime cost. Hence, add an additional shortcut evalfunc in execExprInterp.c that skips unnecessary overhead for that specific form of expression. The runtime speed of simple cases is within 1% or so of where it was before, while cases that previously required two levels of array processing are significantly faster. Finally, create an implicit array type for every domain type, as we do for base types, enums, etc. Everything except the array-coercion case seems to just work without further effort. Tom Lane, reviewed by Andrew Dunstan Discussion: https://postgr.es/m/9852.1499791473@sss.pgh.pa.us
2017-09-30 19:40:56 +02:00
/* strict function, so check for NULL args */
expression eval, jit: Minor code cleanups. This mostly consists of using C99 style for loops, moving variables into narrower scopes, and a smattering of other minor improvements. Done separately to make it easier to review patches with actual functional changes. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:04:50 +01:00
for (int argno = 0; argno < nargs; argno++)
Support arrays over domains. Allowing arrays with a domain type as their element type was left un-done in the original domain patch, but not for any very good reason. This omission leads to such surprising results as array_agg() not working on a domain column, because the parser can't identify a suitable output type for the polymorphic aggregate. In order to fix this, first clean up the APIs of coerce_to_domain() and some internal functions in parse_coerce.c so that we consistently pass around a CoercionContext along with CoercionForm. Previously, we sometimes passed an "isExplicit" boolean flag instead, which is strictly less information; and coerce_to_domain() didn't even get that, but instead had to reverse-engineer isExplicit from CoercionForm. That's contrary to the documentation in primnodes.h that says that CoercionForm only affects display and not semantics. I don't think this change fixes any live bugs, but it makes things more consistent. The main reason for doing it though is that now build_coercion_expression() receives ccontext, which it needs in order to be able to recursively invoke coerce_to_target_type(). Next, reimplement ArrayCoerceExpr so that the node does not directly know any details of what has to be done to the individual array elements while performing the array coercion. Instead, the per-element processing is represented by a sub-expression whose input is a source array element and whose output is a target array element. This simplifies life in parse_coerce.c, because it can build that sub-expression by a recursive invocation of coerce_to_target_type(). The executor now handles the per-element processing as a compiled expression instead of hard-wired code. The main advantage of this is that we can use a single ArrayCoerceExpr to handle as many as three successive steps per element: base type conversion, typmod coercion, and domain constraint checking. The old code used two stacked ArrayCoerceExprs to handle type + typmod coercion, which was pretty inefficient, and adding yet another array deconstruction to do domain constraint checking seemed very unappetizing. In the case where we just need a single, very simple coercion function, doing this straightforwardly leads to a noticeable increase in the per-array-element runtime cost. Hence, add an additional shortcut evalfunc in execExprInterp.c that skips unnecessary overhead for that specific form of expression. The runtime speed of simple cases is within 1% or so of where it was before, while cases that previously required two levels of array processing are significantly faster. Finally, create an implicit array type for every domain type, as we do for base types, enums, etc. Everything except the array-coercion case seems to just work without further effort. Tom Lane, reviewed by Andrew Dunstan Discussion: https://postgr.es/m/9852.1499791473@sss.pgh.pa.us
2017-09-30 19:40:56 +02:00
{
Change function call information to be variable length. Before this change FunctionCallInfoData, the struct arguments etc for V1 function calls are stored in, always had space for FUNC_MAX_ARGS/100 arguments, storing datums and their nullness in two arrays. For nearly every function call 100 arguments is far more than needed, therefore wasting memory. Arg and argnull being two separate arrays also guarantees that to access a single argument, two cachelines have to be touched. Change the layout so there's a single variable-length array with pairs of value / isnull. That drastically reduces memory consumption for most function calls (on x86-64 a two argument function now uses 64bytes, previously 936 bytes), and makes it very likely that argument value and its nullness are on the same cacheline. Arguments are stored in a new NullableDatum struct, which, due to padding, needs more memory per argument than before. But as usually far fewer arguments are stored, and individual arguments are cheaper to access, that's still a clear win. It's likely that there's other places where conversion to NullableDatum arrays would make sense, e.g. TupleTableSlots, but that's for another commit. Because the function call information is now variable-length allocations have to take the number of arguments into account. For heap allocations that can be done with SizeForFunctionCallInfoData(), for on-stack allocations there's a new LOCAL_FCINFO(name, nargs) macro that helps to allocate an appropriately sized and aligned variable. Some places with stack allocation function call information don't know the number of arguments at compile time, and currently variably sized stack allocations aren't allowed in postgres. Therefore allow for FUNC_MAX_ARGS space in these cases. They're not that common, so for now that seems acceptable. Because of the need to allocate FunctionCallInfo of the appropriate size, older extensions may need to update their code. To avoid subtle breakages, the FunctionCallInfoData struct has been renamed to FunctionCallInfoBaseData. Most code only references FunctionCallInfo, so that shouldn't cause much collateral damage. This change is also a prerequisite for more efficient expression JIT compilation (by allocating the function call information on the stack, allowing LLVM to optimize it away); previously the size of the call information caused problems inside LLVM's optimizer. Author: Andres Freund Reviewed-By: Tom Lane Discussion: https://postgr.es/m/20180605172952.x34m5uz6ju6enaem@alap3.anarazel.de
2019-01-26 23:17:52 +01:00
if (args[argno].isnull)
Support arrays over domains. Allowing arrays with a domain type as their element type was left un-done in the original domain patch, but not for any very good reason. This omission leads to such surprising results as array_agg() not working on a domain column, because the parser can't identify a suitable output type for the polymorphic aggregate. In order to fix this, first clean up the APIs of coerce_to_domain() and some internal functions in parse_coerce.c so that we consistently pass around a CoercionContext along with CoercionForm. Previously, we sometimes passed an "isExplicit" boolean flag instead, which is strictly less information; and coerce_to_domain() didn't even get that, but instead had to reverse-engineer isExplicit from CoercionForm. That's contrary to the documentation in primnodes.h that says that CoercionForm only affects display and not semantics. I don't think this change fixes any live bugs, but it makes things more consistent. The main reason for doing it though is that now build_coercion_expression() receives ccontext, which it needs in order to be able to recursively invoke coerce_to_target_type(). Next, reimplement ArrayCoerceExpr so that the node does not directly know any details of what has to be done to the individual array elements while performing the array coercion. Instead, the per-element processing is represented by a sub-expression whose input is a source array element and whose output is a target array element. This simplifies life in parse_coerce.c, because it can build that sub-expression by a recursive invocation of coerce_to_target_type(). The executor now handles the per-element processing as a compiled expression instead of hard-wired code. The main advantage of this is that we can use a single ArrayCoerceExpr to handle as many as three successive steps per element: base type conversion, typmod coercion, and domain constraint checking. The old code used two stacked ArrayCoerceExprs to handle type + typmod coercion, which was pretty inefficient, and adding yet another array deconstruction to do domain constraint checking seemed very unappetizing. In the case where we just need a single, very simple coercion function, doing this straightforwardly leads to a noticeable increase in the per-array-element runtime cost. Hence, add an additional shortcut evalfunc in execExprInterp.c that skips unnecessary overhead for that specific form of expression. The runtime speed of simple cases is within 1% or so of where it was before, while cases that previously required two levels of array processing are significantly faster. Finally, create an implicit array type for every domain type, as we do for base types, enums, etc. Everything except the array-coercion case seems to just work without further effort. Tom Lane, reviewed by Andrew Dunstan Discussion: https://postgr.es/m/9852.1499791473@sss.pgh.pa.us
2017-09-30 19:40:56 +02:00
{
*isnull = true;
return (Datum) 0;
}
}
fcinfo->isnull = false;
d = op->d.func.fn_addr(fcinfo);
*isnull = fcinfo->isnull;
return d;
}
Reduce code duplication for ExecJust*Var operations. This is mainly in preparation for adding further fastpath evaluation routines. Also reorder ExecJust*Var functions to be consistent with the order in which they're used. Author: Andres Freund Discussion: https://postgr.es/m/CAE-ML+9OKSN71+mHtfMD-L24oDp8dGTfaVjDU6U+j+FNAW5kRQ@mail.gmail.com
2019-10-01 00:00:21 +02:00
/* Simple Const expression */
static Datum
ExecJustConst(ExprState *state, ExprContext *econtext, bool *isnull)
{
ExprEvalStep *op = &state->steps[0];
*isnull = op->d.constval.isnull;
return op->d.constval.value;
}
Don't generate EEOP_*_FETCHSOME operations for slots know to be virtual. That avoids unnecessary work during both interpreted execution, and JIT compiled expression evaluation. Both benefit from fewer expression steps needing be processed, and for interpreted execution there now is a fastpath dedicated to just fetching a value from a virtual slot. That's e.g. beneficial for hashjoins over nodes that perform projections, as the hashed columns are currently fetched individually. Author: Soumyadeep Chakraborty, Andres Freund Discussion: https://postgr.es/m/CAE-ML+9OKSN71+mHtfMD-L24oDp8dGTfaVjDU6U+j+FNAW5kRQ@mail.gmail.com
2019-10-01 01:06:16 +02:00
/* implementation of ExecJust(Inner|Outer|Scan)VarVirt */
static pg_attribute_always_inline Datum
ExecJustVarVirtImpl(ExprState *state, TupleTableSlot *slot, bool *isnull)
{
ExprEvalStep *op = &state->steps[0];
int attnum = op->d.var.attnum;
/*
* As it is guaranteed that a virtual slot is used, there never is a need
* to perform tuple deforming (nor would it be possible). Therefore
* execExpr.c has not emitted an EEOP_*_FETCHSOME step. Verify, as much as
* possible, that that determination was accurate.
*/
Assert(TTS_IS_VIRTUAL(slot));
Assert(TTS_FIXED(slot));
Assert(attnum >= 0 && attnum < slot->tts_nvalid);
*isnull = slot->tts_isnull[attnum];
return slot->tts_values[attnum];
}
/* Like ExecJustInnerVar, optimized for virtual slots */
static Datum
ExecJustInnerVarVirt(ExprState *state, ExprContext *econtext, bool *isnull)
{
return ExecJustVarVirtImpl(state, econtext->ecxt_innertuple, isnull);
}
/* Like ExecJustOuterVar, optimized for virtual slots */
static Datum
ExecJustOuterVarVirt(ExprState *state, ExprContext *econtext, bool *isnull)
{
return ExecJustVarVirtImpl(state, econtext->ecxt_outertuple, isnull);
}
/* Like ExecJustScanVar, optimized for virtual slots */
static Datum
ExecJustScanVarVirt(ExprState *state, ExprContext *econtext, bool *isnull)
{
return ExecJustVarVirtImpl(state, econtext->ecxt_scantuple, isnull);
}
/* implementation of ExecJustAssign(Inner|Outer|Scan)VarVirt */
static pg_attribute_always_inline Datum
ExecJustAssignVarVirtImpl(ExprState *state, TupleTableSlot *inslot, bool *isnull)
{
ExprEvalStep *op = &state->steps[0];
int attnum = op->d.assign_var.attnum;
int resultnum = op->d.assign_var.resultnum;
TupleTableSlot *outslot = state->resultslot;
/* see ExecJustVarVirtImpl for comments */
Assert(TTS_IS_VIRTUAL(inslot));
Assert(TTS_FIXED(inslot));
Assert(attnum >= 0 && attnum < inslot->tts_nvalid);
Fix mishandling of resjunk columns in ON CONFLICT ... UPDATE tlists. It's unusual to have any resjunk columns in an ON CONFLICT ... UPDATE list, but it can happen when MULTIEXPR_SUBLINK SubPlans are present. If it happens, the ON CONFLICT UPDATE code path would end up storing tuples that include the values of the extra resjunk columns. That's fairly harmless in the short run, but if new columns are added to the table then the values would become accessible, possibly leading to malfunctions if they don't match the datatypes of the new columns. This had escaped notice through a confluence of missing sanity checks, including * There's no cross-check that a tuple presented to heap_insert or heap_update matches the table rowtype. While it's difficult to check that fully at reasonable cost, we can easily add assertions that there aren't too many columns. * The output-column-assignment cases in execExprInterp.c lacked any sanity checks on the output column numbers, which seems like an oversight considering there are plenty of assertion checks on input column numbers. Add assertions there too. * We failed to apply nodeModifyTable's ExecCheckPlanOutput() to the ON CONFLICT UPDATE tlist. That wouldn't have caught this specific error, since that function is chartered to ignore resjunk columns; but it sure seems like a bad omission now that we've seen this bug. In HEAD, the right way to fix this is to make the processing of ON CONFLICT UPDATE tlists work the same as regular UPDATE tlists now do, that is don't add "SET x = x" entries, and use ExecBuildUpdateProjection to evaluate the tlist and combine it with old values of the not-set columns. This adds a little complication to ExecBuildUpdateProjection, but allows removal of a comparable amount of now-dead code from the planner. In the back branches, the most expedient solution seems to be to (a) use an output slot for the ON CONFLICT UPDATE projection that actually matches the target table, and then (b) invent a variant of ExecBuildProjectionInfo that can be told to not store values resulting from resjunk columns, so it doesn't try to store into nonexistent columns of the output slot. (We can't simply ignore the resjunk columns altogether; they have to be evaluated for MULTIEXPR_SUBLINK to work.) This works back to v10. In 9.6, projections work much differently and we can't cheaply give them such an option. The 9.6 version of this patch works by inserting a JunkFilter when it's necessary to get rid of resjunk columns. In addition, v11 and up have the reverse problem when trying to perform ON CONFLICT UPDATE on a partitioned table. Through a further oversight, adjust_partition_tlist() discarded resjunk columns when re-ordering the ON CONFLICT UPDATE tlist to match a partition. This accidentally prevented the storing-bogus-tuples problem, but at the cost that MULTIEXPR_SUBLINK cases didn't work, typically crashing if more than one row has to be updated. Fix by preserving resjunk columns in that routine. (I failed to resist the temptation to add more assertions there too, and to do some minor code beautification.) Per report from Andres Freund. Back-patch to all supported branches. Security: CVE-2021-32028
2021-05-10 17:02:29 +02:00
Assert(resultnum >= 0 && resultnum < outslot->tts_tupleDescriptor->natts);
Don't generate EEOP_*_FETCHSOME operations for slots know to be virtual. That avoids unnecessary work during both interpreted execution, and JIT compiled expression evaluation. Both benefit from fewer expression steps needing be processed, and for interpreted execution there now is a fastpath dedicated to just fetching a value from a virtual slot. That's e.g. beneficial for hashjoins over nodes that perform projections, as the hashed columns are currently fetched individually. Author: Soumyadeep Chakraborty, Andres Freund Discussion: https://postgr.es/m/CAE-ML+9OKSN71+mHtfMD-L24oDp8dGTfaVjDU6U+j+FNAW5kRQ@mail.gmail.com
2019-10-01 01:06:16 +02:00
outslot->tts_values[resultnum] = inslot->tts_values[attnum];
outslot->tts_isnull[resultnum] = inslot->tts_isnull[attnum];
return 0;
}
/* Like ExecJustAssignInnerVar, optimized for virtual slots */
static Datum
ExecJustAssignInnerVarVirt(ExprState *state, ExprContext *econtext, bool *isnull)
{
return ExecJustAssignVarVirtImpl(state, econtext->ecxt_innertuple, isnull);
}
/* Like ExecJustAssignOuterVar, optimized for virtual slots */
static Datum
ExecJustAssignOuterVarVirt(ExprState *state, ExprContext *econtext, bool *isnull)
{
return ExecJustAssignVarVirtImpl(state, econtext->ecxt_outertuple, isnull);
}
/* Like ExecJustAssignScanVar, optimized for virtual slots */
static Datum
ExecJustAssignScanVarVirt(ExprState *state, ExprContext *econtext, bool *isnull)
{
return ExecJustAssignVarVirtImpl(state, econtext->ecxt_scantuple, isnull);
}
Perform slot validity checks in a separate pass over expression. This reduces code duplication a bit, but the primary benefit that it makes JITing expression evaluation easier. When doing so we can't, as previously done in the interpreted case, really change opcode without recompiling. Nor dow we just carry around unnecessary branches to avoid re-checking over and over. As a minor side-effect this makes ExecEvalStepOp() O(log(N)) rather than O(N). Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2017-12-29 21:38:15 +01:00
#if defined(EEO_USE_COMPUTED_GOTO)
/*
* Comparator used when building address->opcode lookup table for
* ExecEvalStepOp() in the threaded dispatch case.
*/
static int
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
dispatch_compare_ptr(const void *a, const void *b)
Perform slot validity checks in a separate pass over expression. This reduces code duplication a bit, but the primary benefit that it makes JITing expression evaluation easier. When doing so we can't, as previously done in the interpreted case, really change opcode without recompiling. Nor dow we just carry around unnecessary branches to avoid re-checking over and over. As a minor side-effect this makes ExecEvalStepOp() O(log(N)) rather than O(N). Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2017-12-29 21:38:15 +01:00
{
const ExprEvalOpLookup *la = (const ExprEvalOpLookup *) a;
const ExprEvalOpLookup *lb = (const ExprEvalOpLookup *) b;
if (la->opcode < lb->opcode)
return -1;
else if (la->opcode > lb->opcode)
return 1;
return 0;
}
#endif
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
/*
* Do one-time initialization of interpretation machinery.
*/
static void
ExecInitInterpreter(void)
{
#if defined(EEO_USE_COMPUTED_GOTO)
/* Set up externally-visible pointer to dispatch table */
if (dispatch_table == NULL)
Perform slot validity checks in a separate pass over expression. This reduces code duplication a bit, but the primary benefit that it makes JITing expression evaluation easier. When doing so we can't, as previously done in the interpreted case, really change opcode without recompiling. Nor dow we just carry around unnecessary branches to avoid re-checking over and over. As a minor side-effect this makes ExecEvalStepOp() O(log(N)) rather than O(N). Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2017-12-29 21:38:15 +01:00
{
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
dispatch_table = (const void **)
DatumGetPointer(ExecInterpExpr(NULL, NULL, NULL));
Perform slot validity checks in a separate pass over expression. This reduces code duplication a bit, but the primary benefit that it makes JITing expression evaluation easier. When doing so we can't, as previously done in the interpreted case, really change opcode without recompiling. Nor dow we just carry around unnecessary branches to avoid re-checking over and over. As a minor side-effect this makes ExecEvalStepOp() O(log(N)) rather than O(N). Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2017-12-29 21:38:15 +01:00
/* build reverse lookup table */
expression eval, jit: Minor code cleanups. This mostly consists of using C99 style for loops, moving variables into narrower scopes, and a smattering of other minor improvements. Done separately to make it easier to review patches with actual functional changes. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:04:50 +01:00
for (int i = 0; i < EEOP_LAST; i++)
Perform slot validity checks in a separate pass over expression. This reduces code duplication a bit, but the primary benefit that it makes JITing expression evaluation easier. When doing so we can't, as previously done in the interpreted case, really change opcode without recompiling. Nor dow we just carry around unnecessary branches to avoid re-checking over and over. As a minor side-effect this makes ExecEvalStepOp() O(log(N)) rather than O(N). Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2017-12-29 21:38:15 +01:00
{
reverse_dispatch_table[i].opcode = dispatch_table[i];
reverse_dispatch_table[i].op = (ExprEvalOp) i;
}
/* make it bsearch()able */
qsort(reverse_dispatch_table,
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
EEOP_LAST /* nmembers */ ,
Perform slot validity checks in a separate pass over expression. This reduces code duplication a bit, but the primary benefit that it makes JITing expression evaluation easier. When doing so we can't, as previously done in the interpreted case, really change opcode without recompiling. Nor dow we just carry around unnecessary branches to avoid re-checking over and over. As a minor side-effect this makes ExecEvalStepOp() O(log(N)) rather than O(N). Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2017-12-29 21:38:15 +01:00
sizeof(ExprEvalOpLookup),
dispatch_compare_ptr);
}
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
#endif
}
/*
* Function to return the opcode of an expression step.
*
* When direct-threading is in use, ExprState->opcode isn't easily
* decipherable. This function returns the appropriate enum member.
*/
ExprEvalOp
ExecEvalStepOp(ExprState *state, ExprEvalStep *op)
{
#if defined(EEO_USE_COMPUTED_GOTO)
if (state->flags & EEO_FLAG_DIRECT_THREADED)
{
Perform slot validity checks in a separate pass over expression. This reduces code duplication a bit, but the primary benefit that it makes JITing expression evaluation easier. When doing so we can't, as previously done in the interpreted case, really change opcode without recompiling. Nor dow we just carry around unnecessary branches to avoid re-checking over and over. As a minor side-effect this makes ExecEvalStepOp() O(log(N)) rather than O(N). Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2017-12-29 21:38:15 +01:00
ExprEvalOpLookup key;
ExprEvalOpLookup *res;
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
key.opcode = (void *) op->opcode;
Perform slot validity checks in a separate pass over expression. This reduces code duplication a bit, but the primary benefit that it makes JITing expression evaluation easier. When doing so we can't, as previously done in the interpreted case, really change opcode without recompiling. Nor dow we just carry around unnecessary branches to avoid re-checking over and over. As a minor side-effect this makes ExecEvalStepOp() O(log(N)) rather than O(N). Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2017-12-29 21:38:15 +01:00
res = bsearch(&key,
reverse_dispatch_table,
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
EEOP_LAST /* nmembers */ ,
Perform slot validity checks in a separate pass over expression. This reduces code duplication a bit, but the primary benefit that it makes JITing expression evaluation easier. When doing so we can't, as previously done in the interpreted case, really change opcode without recompiling. Nor dow we just carry around unnecessary branches to avoid re-checking over and over. As a minor side-effect this makes ExecEvalStepOp() O(log(N)) rather than O(N). Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2017-12-29 21:38:15 +01:00
sizeof(ExprEvalOpLookup),
dispatch_compare_ptr);
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
Assert(res); /* unknown ops shouldn't get looked up */
Perform slot validity checks in a separate pass over expression. This reduces code duplication a bit, but the primary benefit that it makes JITing expression evaluation easier. When doing so we can't, as previously done in the interpreted case, really change opcode without recompiling. Nor dow we just carry around unnecessary branches to avoid re-checking over and over. As a minor side-effect this makes ExecEvalStepOp() O(log(N)) rather than O(N). Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2017-12-29 21:38:15 +01:00
return res->op;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
}
#endif
return (ExprEvalOp) op->opcode;
}
/*
* Out-of-line helper functions for complex instructions.
*/
Handle EEOP_FUNCEXPR_[STRICT_]FUSAGE out of line. This isn't a very common op, and it doesn't seem worth duplicating for JIT. Author: Andres Freund
2018-03-21 01:32:21 +01:00
/*
* Evaluate EEOP_FUNCEXPR_FUSAGE
*/
void
ExecEvalFuncExprFusage(ExprState *state, ExprEvalStep *op,
ExprContext *econtext)
{
FunctionCallInfo fcinfo = op->d.func.fcinfo_data;
PgStat_FunctionCallUsage fcusage;
Datum d;
pgstat_init_function_usage(fcinfo, &fcusage);
fcinfo->isnull = false;
d = op->d.func.fn_addr(fcinfo);
*op->resvalue = d;
*op->resnull = fcinfo->isnull;
pgstat_end_function_usage(&fcusage, true);
}
/*
* Evaluate EEOP_FUNCEXPR_STRICT_FUSAGE
*/
void
ExecEvalFuncExprStrictFusage(ExprState *state, ExprEvalStep *op,
ExprContext *econtext)
{
FunctionCallInfo fcinfo = op->d.func.fcinfo_data;
PgStat_FunctionCallUsage fcusage;
Change function call information to be variable length. Before this change FunctionCallInfoData, the struct arguments etc for V1 function calls are stored in, always had space for FUNC_MAX_ARGS/100 arguments, storing datums and their nullness in two arrays. For nearly every function call 100 arguments is far more than needed, therefore wasting memory. Arg and argnull being two separate arrays also guarantees that to access a single argument, two cachelines have to be touched. Change the layout so there's a single variable-length array with pairs of value / isnull. That drastically reduces memory consumption for most function calls (on x86-64 a two argument function now uses 64bytes, previously 936 bytes), and makes it very likely that argument value and its nullness are on the same cacheline. Arguments are stored in a new NullableDatum struct, which, due to padding, needs more memory per argument than before. But as usually far fewer arguments are stored, and individual arguments are cheaper to access, that's still a clear win. It's likely that there's other places where conversion to NullableDatum arrays would make sense, e.g. TupleTableSlots, but that's for another commit. Because the function call information is now variable-length allocations have to take the number of arguments into account. For heap allocations that can be done with SizeForFunctionCallInfoData(), for on-stack allocations there's a new LOCAL_FCINFO(name, nargs) macro that helps to allocate an appropriately sized and aligned variable. Some places with stack allocation function call information don't know the number of arguments at compile time, and currently variably sized stack allocations aren't allowed in postgres. Therefore allow for FUNC_MAX_ARGS space in these cases. They're not that common, so for now that seems acceptable. Because of the need to allocate FunctionCallInfo of the appropriate size, older extensions may need to update their code. To avoid subtle breakages, the FunctionCallInfoData struct has been renamed to FunctionCallInfoBaseData. Most code only references FunctionCallInfo, so that shouldn't cause much collateral damage. This change is also a prerequisite for more efficient expression JIT compilation (by allocating the function call information on the stack, allowing LLVM to optimize it away); previously the size of the call information caused problems inside LLVM's optimizer. Author: Andres Freund Reviewed-By: Tom Lane Discussion: https://postgr.es/m/20180605172952.x34m5uz6ju6enaem@alap3.anarazel.de
2019-01-26 23:17:52 +01:00
NullableDatum *args = fcinfo->args;
expression eval, jit: Minor code cleanups. This mostly consists of using C99 style for loops, moving variables into narrower scopes, and a smattering of other minor improvements. Done separately to make it easier to review patches with actual functional changes. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:04:50 +01:00
int nargs = op->d.func.nargs;
Handle EEOP_FUNCEXPR_[STRICT_]FUSAGE out of line. This isn't a very common op, and it doesn't seem worth duplicating for JIT. Author: Andres Freund
2018-03-21 01:32:21 +01:00
Datum d;
/* strict function, so check for NULL args */
expression eval, jit: Minor code cleanups. This mostly consists of using C99 style for loops, moving variables into narrower scopes, and a smattering of other minor improvements. Done separately to make it easier to review patches with actual functional changes. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:04:50 +01:00
for (int argno = 0; argno < nargs; argno++)
Handle EEOP_FUNCEXPR_[STRICT_]FUSAGE out of line. This isn't a very common op, and it doesn't seem worth duplicating for JIT. Author: Andres Freund
2018-03-21 01:32:21 +01:00
{
Change function call information to be variable length. Before this change FunctionCallInfoData, the struct arguments etc for V1 function calls are stored in, always had space for FUNC_MAX_ARGS/100 arguments, storing datums and their nullness in two arrays. For nearly every function call 100 arguments is far more than needed, therefore wasting memory. Arg and argnull being two separate arrays also guarantees that to access a single argument, two cachelines have to be touched. Change the layout so there's a single variable-length array with pairs of value / isnull. That drastically reduces memory consumption for most function calls (on x86-64 a two argument function now uses 64bytes, previously 936 bytes), and makes it very likely that argument value and its nullness are on the same cacheline. Arguments are stored in a new NullableDatum struct, which, due to padding, needs more memory per argument than before. But as usually far fewer arguments are stored, and individual arguments are cheaper to access, that's still a clear win. It's likely that there's other places where conversion to NullableDatum arrays would make sense, e.g. TupleTableSlots, but that's for another commit. Because the function call information is now variable-length allocations have to take the number of arguments into account. For heap allocations that can be done with SizeForFunctionCallInfoData(), for on-stack allocations there's a new LOCAL_FCINFO(name, nargs) macro that helps to allocate an appropriately sized and aligned variable. Some places with stack allocation function call information don't know the number of arguments at compile time, and currently variably sized stack allocations aren't allowed in postgres. Therefore allow for FUNC_MAX_ARGS space in these cases. They're not that common, so for now that seems acceptable. Because of the need to allocate FunctionCallInfo of the appropriate size, older extensions may need to update their code. To avoid subtle breakages, the FunctionCallInfoData struct has been renamed to FunctionCallInfoBaseData. Most code only references FunctionCallInfo, so that shouldn't cause much collateral damage. This change is also a prerequisite for more efficient expression JIT compilation (by allocating the function call information on the stack, allowing LLVM to optimize it away); previously the size of the call information caused problems inside LLVM's optimizer. Author: Andres Freund Reviewed-By: Tom Lane Discussion: https://postgr.es/m/20180605172952.x34m5uz6ju6enaem@alap3.anarazel.de
2019-01-26 23:17:52 +01:00
if (args[argno].isnull)
Handle EEOP_FUNCEXPR_[STRICT_]FUSAGE out of line. This isn't a very common op, and it doesn't seem worth duplicating for JIT. Author: Andres Freund
2018-03-21 01:32:21 +01:00
{
*op->resnull = true;
return;
}
}
pgstat_init_function_usage(fcinfo, &fcusage);
fcinfo->isnull = false;
d = op->d.func.fn_addr(fcinfo);
*op->resvalue = d;
*op->resnull = fcinfo->isnull;
pgstat_end_function_usage(&fcusage, true);
}
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
/*
* Evaluate a PARAM_EXEC parameter.
*
* PARAM_EXEC params (internal executor parameters) are stored in the
* ecxt_param_exec_vals array, and can be accessed by array index.
*/
void
ExecEvalParamExec(ExprState *state, ExprEvalStep *op, ExprContext *econtext)
{
ParamExecData *prm;
prm = &(econtext->ecxt_param_exec_vals[op->d.param.paramid]);
if (unlikely(prm->execPlan != NULL))
{
/* Parameter not evaluated yet, so go do it */
ExecSetParamPlan(prm->execPlan, econtext);
/* ExecSetParamPlan should have processed this param... */
Assert(prm->execPlan == NULL);
}
*op->resvalue = prm->value;
*op->resnull = prm->isnull;
}
/*
* Evaluate a PARAM_EXTERN parameter.
*
* PARAM_EXTERN parameters must be sought in ecxt_param_list_info.
*/
void
ExecEvalParamExtern(ExprState *state, ExprEvalStep *op, ExprContext *econtext)
{
ParamListInfo paramInfo = econtext->ecxt_param_list_info;
int paramId = op->d.param.paramid;
if (likely(paramInfo &&
paramId > 0 && paramId <= paramInfo->numParams))
{
Rearrange execution of PARAM_EXTERN Params for plpgsql's benefit. This patch does three interrelated things: * Create a new expression execution step type EEOP_PARAM_CALLBACK and add the infrastructure needed for add-on modules to generate that. As discussed, the best control mechanism for that seems to be to add another hook function to ParamListInfo, which will be called by ExecInitExpr if it's supplied and a PARAM_EXTERN Param is found. For stand-alone expressions, we add a new entry point to allow the ParamListInfo to be specified directly, since it can't be retrieved from the parent plan node's EState. * Redesign the API for the ParamListInfo paramFetch hook so that the ParamExternData array can be entirely virtual. This also lets us get rid of ParamListInfo.paramMask, instead leaving it to the paramFetch hook to decide which param IDs should be accessible or not. plpgsql_param_fetch was already doing the identical masking check, so having callers do it too seemed redundant. While I was at it, I added a "speculative" flag to paramFetch that the planner can specify as TRUE to avoid unwanted failures. This solves an ancient problem for plpgsql that it couldn't provide values of non-DTYPE_VAR variables to the planner for fear of triggering premature "record not assigned yet" or "field not found" errors during planning. * Rework plpgsql to get rid of the need for "unshared" parameter lists, by dint of turning the single ParamListInfo per estate into a nearly read-only data structure that doesn't instantiate any per-variable data. Instead, the paramFetch hook controls access to per-variable data and can make the right decisions on the fly, replacing the cases that we used to need multiple ParamListInfos for. This might perhaps have been a performance loss on its own, but by using a paramCompile hook we can bypass plpgsql_param_fetch entirely during normal query execution. (It's now only called when, eg, we copy the ParamListInfo into a cursor portal. copyParamList() or SerializeParamList() effectively instantiate the virtual parameter array as a simple physical array without a paramFetch hook, which is what we want in those cases.) This allows reverting most of commit 6c82d8d1f, though I kept the cosmetic code-consolidation aspects of that (eg the assign_simple_var function). Performance testing shows this to be at worst a break-even change, and it can provide wins ranging up to 20% in test cases involving accesses to fields of "record" variables. The fact that values of such variables can now be exposed to the planner might produce wins in some situations, too, but I've not pursued that angle. In passing, remove the "parent" pointer from the arguments to ExecInitExprRec and related functions, instead storing that pointer in a transient field in ExprState. The ParamListInfo pointer for a stand-alone expression is handled the same way; we'd otherwise have had to add yet another recursively-passed-down argument in expression compilation. Discussion: https://postgr.es/m/32589.1513706441@sss.pgh.pa.us
2017-12-21 18:57:41 +01:00
ParamExternData *prm;
ParamExternData prmdata;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
/* give hook a chance in case parameter is dynamic */
Rearrange execution of PARAM_EXTERN Params for plpgsql's benefit. This patch does three interrelated things: * Create a new expression execution step type EEOP_PARAM_CALLBACK and add the infrastructure needed for add-on modules to generate that. As discussed, the best control mechanism for that seems to be to add another hook function to ParamListInfo, which will be called by ExecInitExpr if it's supplied and a PARAM_EXTERN Param is found. For stand-alone expressions, we add a new entry point to allow the ParamListInfo to be specified directly, since it can't be retrieved from the parent plan node's EState. * Redesign the API for the ParamListInfo paramFetch hook so that the ParamExternData array can be entirely virtual. This also lets us get rid of ParamListInfo.paramMask, instead leaving it to the paramFetch hook to decide which param IDs should be accessible or not. plpgsql_param_fetch was already doing the identical masking check, so having callers do it too seemed redundant. While I was at it, I added a "speculative" flag to paramFetch that the planner can specify as TRUE to avoid unwanted failures. This solves an ancient problem for plpgsql that it couldn't provide values of non-DTYPE_VAR variables to the planner for fear of triggering premature "record not assigned yet" or "field not found" errors during planning. * Rework plpgsql to get rid of the need for "unshared" parameter lists, by dint of turning the single ParamListInfo per estate into a nearly read-only data structure that doesn't instantiate any per-variable data. Instead, the paramFetch hook controls access to per-variable data and can make the right decisions on the fly, replacing the cases that we used to need multiple ParamListInfos for. This might perhaps have been a performance loss on its own, but by using a paramCompile hook we can bypass plpgsql_param_fetch entirely during normal query execution. (It's now only called when, eg, we copy the ParamListInfo into a cursor portal. copyParamList() or SerializeParamList() effectively instantiate the virtual parameter array as a simple physical array without a paramFetch hook, which is what we want in those cases.) This allows reverting most of commit 6c82d8d1f, though I kept the cosmetic code-consolidation aspects of that (eg the assign_simple_var function). Performance testing shows this to be at worst a break-even change, and it can provide wins ranging up to 20% in test cases involving accesses to fields of "record" variables. The fact that values of such variables can now be exposed to the planner might produce wins in some situations, too, but I've not pursued that angle. In passing, remove the "parent" pointer from the arguments to ExecInitExprRec and related functions, instead storing that pointer in a transient field in ExprState. The ParamListInfo pointer for a stand-alone expression is handled the same way; we'd otherwise have had to add yet another recursively-passed-down argument in expression compilation. Discussion: https://postgr.es/m/32589.1513706441@sss.pgh.pa.us
2017-12-21 18:57:41 +01:00
if (paramInfo->paramFetch != NULL)
prm = paramInfo->paramFetch(paramInfo, paramId, false, &prmdata);
else
prm = &paramInfo->params[paramId - 1];
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
if (likely(OidIsValid(prm->ptype)))
{
/* safety check in case hook did something unexpected */
if (unlikely(prm->ptype != op->d.param.paramtype))
ereport(ERROR,
(errcode(ERRCODE_DATATYPE_MISMATCH),
errmsg("type of parameter %d (%s) does not match that when preparing the plan (%s)",
paramId,
format_type_be(prm->ptype),
format_type_be(op->d.param.paramtype))));
*op->resvalue = prm->value;
*op->resnull = prm->isnull;
return;
}
}
ereport(ERROR,
(errcode(ERRCODE_UNDEFINED_OBJECT),
errmsg("no value found for parameter %d", paramId)));
}
Add back SQLValueFunction for SQL keywords This is equivalent to a revert of f193883 and fb32748, with the addition that the declaration of the SQLValueFunction node needs to gain a couple of node_attr for query jumbling. The performance impact of removing the function call inlining is proving to be too huge for some workloads where these are used. A worst-case test case of involving only simple SELECT queries with a SQL keyword is proving to lead to a reduction of 10% in TPS via pgbench and prepared queries on a high-end machine. None of the tests I ran back for this set of changes saw such a huge gap, but Alexander Lakhin and Andres Freund have found that this can be noticeable. Keeping the older performance would mean to do more inlining in the executor when using COERCE_SQL_SYNTAX for a function expression, similarly to what SQLValueFunction does. This requires more redesign work and there is little time until 16beta1 is released, so for now reverting the change is the best way forward, bringing back the previous performance. Bump catalog version. Reported-by: Alexander Lakhin Discussion: https://postgr.es/m/b32bed1b-0746-9b20-1472-4bdc9ca66d52@gmail.com
2023-05-17 03:19:17 +02:00
/*
* Evaluate a SQLValueFunction expression.
*/
void
ExecEvalSQLValueFunction(ExprState *state, ExprEvalStep *op)
{
LOCAL_FCINFO(fcinfo, 0);
SQLValueFunction *svf = op->d.sqlvaluefunction.svf;
*op->resnull = false;
/*
* Note: current_schema() can return NULL. current_user() etc currently
* cannot, but might as well code those cases the same way for safety.
*/
switch (svf->op)
{
case SVFOP_CURRENT_DATE:
*op->resvalue = DateADTGetDatum(GetSQLCurrentDate());
break;
case SVFOP_CURRENT_TIME:
case SVFOP_CURRENT_TIME_N:
*op->resvalue = TimeTzADTPGetDatum(GetSQLCurrentTime(svf->typmod));
break;
case SVFOP_CURRENT_TIMESTAMP:
case SVFOP_CURRENT_TIMESTAMP_N:
*op->resvalue = TimestampTzGetDatum(GetSQLCurrentTimestamp(svf->typmod));
break;
case SVFOP_LOCALTIME:
case SVFOP_LOCALTIME_N:
*op->resvalue = TimeADTGetDatum(GetSQLLocalTime(svf->typmod));
break;
case SVFOP_LOCALTIMESTAMP:
case SVFOP_LOCALTIMESTAMP_N:
*op->resvalue = TimestampGetDatum(GetSQLLocalTimestamp(svf->typmod));
break;
case SVFOP_CURRENT_ROLE:
case SVFOP_CURRENT_USER:
case SVFOP_USER:
InitFunctionCallInfoData(*fcinfo, NULL, 0, InvalidOid, NULL, NULL);
*op->resvalue = current_user(fcinfo);
*op->resnull = fcinfo->isnull;
break;
case SVFOP_SESSION_USER:
InitFunctionCallInfoData(*fcinfo, NULL, 0, InvalidOid, NULL, NULL);
*op->resvalue = session_user(fcinfo);
*op->resnull = fcinfo->isnull;
break;
case SVFOP_CURRENT_CATALOG:
InitFunctionCallInfoData(*fcinfo, NULL, 0, InvalidOid, NULL, NULL);
*op->resvalue = current_database(fcinfo);
*op->resnull = fcinfo->isnull;
break;
case SVFOP_CURRENT_SCHEMA:
InitFunctionCallInfoData(*fcinfo, NULL, 0, InvalidOid, NULL, NULL);
*op->resvalue = current_schema(fcinfo);
*op->resnull = fcinfo->isnull;
break;
}
}
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
/*
* Raise error if a CURRENT OF expression is evaluated.
*
* The planner should convert CURRENT OF into a TidScan qualification, or some
* other special handling in a ForeignScan node. So we have to be able to do
* ExecInitExpr on a CurrentOfExpr, but we shouldn't ever actually execute it.
* If we get here, we suppose we must be dealing with CURRENT OF on a foreign
* table whose FDW doesn't handle it, and complain accordingly.
*/
void
ExecEvalCurrentOfExpr(ExprState *state, ExprEvalStep *op)
{
ereport(ERROR,
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("WHERE CURRENT OF is not supported for this table type")));
}
Code review for NextValueExpr expression node type. Add missing infrastructure for this node type, notably in ruleutils.c where its lack could demonstrably cause EXPLAIN to fail. Add outfuncs/readfuncs support. (outfuncs support is useful today for debugging purposes. The readfuncs support may never be needed, since at present it would only matter for parallel query and NextValueExpr should never appear in a parallelizable query; but it seems like a bad idea to have a primnode type that isn't fully supported here.) Teach planner infrastructure that NextValueExpr is a volatile, parallel-unsafe, non-leaky expression node with cost cpu_operator_cost. Given its limited scope of usage, there *might* be no live bug today from the lack of that knowledge, but it's certainly going to bite us on the rear someday. Teach pg_stat_statements about the new node type, too. While at it, also teach cost_qual_eval() that MinMaxExpr, SQLValueFunction, XmlExpr, and CoerceToDomain should be charged as cpu_operator_cost. Failing to do this for SQLValueFunction was an oversight in my commit 0bb51aa96. The others are longer-standing oversights, but no time like the present to fix them. (In principle, CoerceToDomain could have cost much higher than this, but it doesn't presently seem worth trying to examine the domain's constraints here.) Modify execExprInterp.c to execute NextValueExpr as an out-of-line function; it seems quite unlikely to me that it's worth insisting that it be inlined in all expression eval methods. Besides, providing the out-of-line function doesn't stop anyone from inlining if they want to. Adjust some places where NextValueExpr support had been inserted with the aid of a dartboard rather than keeping it in the same order as elsewhere. Discussion: https://postgr.es/m/23862.1499981661@sss.pgh.pa.us
2017-07-14 21:25:43 +02:00
/*
* Evaluate NextValueExpr.
*/
void
ExecEvalNextValueExpr(ExprState *state, ExprEvalStep *op)
{
int64 newval = nextval_internal(op->d.nextvalueexpr.seqid, false);
switch (op->d.nextvalueexpr.seqtypid)
{
case INT2OID:
*op->resvalue = Int16GetDatum((int16) newval);
break;
case INT4OID:
*op->resvalue = Int32GetDatum((int32) newval);
break;
case INT8OID:
*op->resvalue = Int64GetDatum((int64) newval);
break;
default:
elog(ERROR, "unsupported sequence type %u",
op->d.nextvalueexpr.seqtypid);
}
*op->resnull = false;
}
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
/*
* Evaluate NullTest / IS NULL for rows.
*/
void
ExecEvalRowNull(ExprState *state, ExprEvalStep *op, ExprContext *econtext)
{
ExecEvalRowNullInt(state, op, econtext, true);
}
/*
* Evaluate NullTest / IS NOT NULL for rows.
*/
void
ExecEvalRowNotNull(ExprState *state, ExprEvalStep *op, ExprContext *econtext)
{
ExecEvalRowNullInt(state, op, econtext, false);
}
/* Common code for IS [NOT] NULL on a row value */
static void
ExecEvalRowNullInt(ExprState *state, ExprEvalStep *op,
ExprContext *econtext, bool checkisnull)
{
Datum value = *op->resvalue;
bool isnull = *op->resnull;
HeapTupleHeader tuple;
Oid tupType;
int32 tupTypmod;
TupleDesc tupDesc;
HeapTupleData tmptup;
*op->resnull = false;
/* NULL row variables are treated just as NULL scalar columns */
if (isnull)
{
*op->resvalue = BoolGetDatum(checkisnull);
return;
}
/*
* The SQL standard defines IS [NOT] NULL for a non-null rowtype argument
* as:
*
* "R IS NULL" is true if every field is the null value.
*
* "R IS NOT NULL" is true if no field is the null value.
*
* This definition is (apparently intentionally) not recursive; so our
* tests on the fields are primitive attisnull tests, not recursive checks
* to see if they are all-nulls or no-nulls rowtypes.
*
* The standard does not consider the possibility of zero-field rows, but
* here we consider them to vacuously satisfy both predicates.
*/
tuple = DatumGetHeapTupleHeader(value);
tupType = HeapTupleHeaderGetTypeId(tuple);
tupTypmod = HeapTupleHeaderGetTypMod(tuple);
/* Lookup tupdesc if first time through or if type changes */
tupDesc = get_cached_rowtype(tupType, tupTypmod,
Redesign the caching done by get_cached_rowtype(). Previously, get_cached_rowtype() cached a pointer to a reference-counted tuple descriptor from the typcache, relying on the ExprContextCallback mechanism to release the tupdesc refcount when the expression tree using the tupdesc was destroyed. This worked fine when it was designed, but the introduction of within-DO-block COMMITs broke it. The refcount is logged in a transaction-lifespan resource owner, but plpgsql won't destroy simple expressions made within the DO block (before its first commit) until the DO block is exited. That results in a warning about a leaked tupdesc refcount when the COMMIT destroys the original resource owner, and then an error about the active resource owner not holding a matching refcount when the expression is destroyed. To fix, get rid of the need to have a shutdown callback at all, by instead caching a pointer to the relevant typcache entry. Those survive for the life of the backend, so we needn't worry about the pointer becoming stale. (For registered RECORD types, we can still cache a pointer to the tupdesc, knowing that it won't change for the life of the backend.) This mechanism has been in use in plpgsql and expandedrecord.c since commit 4b93f5799, and seems to work well. This change requires modifying the ExprEvalStep structs used by the relevant expression step types, which is slightly worrisome for back-patching. However, there seems no good reason for extensions to be familiar with the details of these particular sub-structs. Per report from Rohit Bhogate. Back-patch to v11 where within-DO-block COMMITs became a thing. Discussion: https://postgr.es/m/CAAV6ZkQRCVBh8qAY+SZiHnz+U+FqAGBBDaDTjF2yiKa2nJSLKg@mail.gmail.com
2021-04-13 19:37:07 +02:00
&op->d.nulltest_row.rowcache, NULL);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
/*
* heap_attisnull needs a HeapTuple not a bare HeapTupleHeader.
*/
tmptup.t_len = HeapTupleHeaderGetDatumLength(tuple);
tmptup.t_data = tuple;
expression eval, jit: Minor code cleanups. This mostly consists of using C99 style for loops, moving variables into narrower scopes, and a smattering of other minor improvements. Done separately to make it easier to review patches with actual functional changes. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:04:50 +01:00
for (int att = 1; att <= tupDesc->natts; att++)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
/* ignore dropped columns */
Change tupledesc->attrs[n] to TupleDescAttr(tupledesc, n). This is a mechanical change in preparation for a later commit that will change the layout of TupleDesc. Introducing a macro to abstract the details of where attributes are stored will allow us to change that in separate step and revise it in future. Author: Thomas Munro, editorialized by Andres Freund Reviewed-By: Andres Freund Discussion: https://postgr.es/m/CAEepm=0ZtQ-SpsgCyzzYpsXS6e=kZWqk3g5Ygn3MDV7A8dabUA@mail.gmail.com
2017-08-20 20:19:07 +02:00
if (TupleDescAttr(tupDesc, att - 1)->attisdropped)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
continue;
Fast ALTER TABLE ADD COLUMN with a non-NULL default Currently adding a column to a table with a non-NULL default results in a rewrite of the table. For large tables this can be both expensive and disruptive. This patch removes the need for the rewrite as long as the default value is not volatile. The default expression is evaluated at the time of the ALTER TABLE and the result stored in a new column (attmissingval) in pg_attribute, and a new column (atthasmissing) is set to true. Any existing row when fetched will be supplied with the attmissingval. New rows will have the supplied value or the default and so will never need the attmissingval. Any time the table is rewritten all the atthasmissing and attmissingval settings for the attributes are cleared, as they are no longer needed. The most visible code change from this is in heap_attisnull, which acquires a third TupleDesc argument, allowing it to detect a missing value if there is one. In many cases where it is known that there will not be any (e.g. catalog relations) NULL can be passed for this argument. Andrew Dunstan, heavily modified from an original patch from Serge Rielau. Reviewed by Tom Lane, Andres Freund, Tomas Vondra and David Rowley. Discussion: https://postgr.es/m/31e2e921-7002-4c27-59f5-51f08404c858@2ndQuadrant.com
2018-03-28 02:13:52 +02:00
if (heap_attisnull(&tmptup, att, tupDesc))
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
/* null field disproves IS NOT NULL */
if (!checkisnull)
{
*op->resvalue = BoolGetDatum(false);
return;
}
}
else
{
/* non-null field disproves IS NULL */
if (checkisnull)
{
*op->resvalue = BoolGetDatum(false);
return;
}
}
}
*op->resvalue = BoolGetDatum(true);
}
/*
* Evaluate an ARRAY[] expression.
*
* The individual array elements (or subarrays) have already been evaluated
* into op->d.arrayexpr.elemvalues[]/elemnulls[].
*/
void
ExecEvalArrayExpr(ExprState *state, ExprEvalStep *op)
{
ArrayType *result;
Oid element_type = op->d.arrayexpr.elemtype;
int nelems = op->d.arrayexpr.nelems;
int ndims = 0;
int dims[MAXDIM];
int lbs[MAXDIM];
/* Set non-null as default */
*op->resnull = false;
if (!op->d.arrayexpr.multidims)
{
/* Elements are presumably of scalar type */
Datum *dvalues = op->d.arrayexpr.elemvalues;
bool *dnulls = op->d.arrayexpr.elemnulls;
/* setup for 1-D array of the given length */
ndims = 1;
dims[0] = nelems;
lbs[0] = 1;
result = construct_md_array(dvalues, dnulls, ndims, dims, lbs,
element_type,
op->d.arrayexpr.elemlength,
op->d.arrayexpr.elembyval,
op->d.arrayexpr.elemalign);
}
else
{
/* Must be nested array expressions */
int nbytes = 0;
Fix oversights in array manipulation. The nested-arrays code path in ExecEvalArrayExpr() used palloc to allocate the result array, whereas every other array-creating function has used palloc0 since 18c0b4ecc. This mostly works, but unused bits past the end of the nulls bitmap may end up undefined. That causes valgrind complaints with -DWRITE_READ_PARSE_PLAN_TREES, and could cause planner misbehavior as cited in 18c0b4ecc. There seems no very good reason why we should strive to avoid palloc0 in just this one case, so fix it the easy way with s/palloc/palloc0/. While looking at that I noted that we also failed to check for overflow of "nbytes" and "nitems" while summing the sizes of the sub-arrays, potentially allowing a crash due to undersized output allocation. For "nbytes", follow the policy used by other array-munging code of checking for overflow after each addition. (As elsewhere, the last addition of the array's overhead space doesn't need an extra check, since palloc itself will catch a value between 1Gb and 2Gb.) For "nitems", there's no very good reason to sum the inputs at all, since we can perfectly well use ArrayGetNItems' result instead of ignoring it. Per discussion of this bug, also remove redundant zeroing of the nulls bitmap in array_set_element and array_set_slice. Patch by Alexander Lakhin and myself, per bug #17858 from Alexander Lakhin; thanks also to Richard Guo. These bugs are a dozen years old, so back-patch to all supported branches. Discussion: https://postgr.es/m/17858-8fd287fd3663d051@postgresql.org
2023-03-26 19:41:06 +02:00
int nitems;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
int outer_nelems = 0;
int elem_ndims = 0;
int *elem_dims = NULL;
int *elem_lbs = NULL;
bool firstone = true;
bool havenulls = false;
bool haveempty = false;
char **subdata;
bits8 **subbitmaps;
int *subbytes;
int *subnitems;
int32 dataoffset;
char *dat;
int iitem;
subdata = (char **) palloc(nelems * sizeof(char *));
subbitmaps = (bits8 **) palloc(nelems * sizeof(bits8 *));
subbytes = (int *) palloc(nelems * sizeof(int));
subnitems = (int *) palloc(nelems * sizeof(int));
/* loop through and get data area from each element */
expression eval, jit: Minor code cleanups. This mostly consists of using C99 style for loops, moving variables into narrower scopes, and a smattering of other minor improvements. Done separately to make it easier to review patches with actual functional changes. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:04:50 +01:00
for (int elemoff = 0; elemoff < nelems; elemoff++)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
Datum arraydatum;
bool eisnull;
ArrayType *array;
int this_ndims;
arraydatum = op->d.arrayexpr.elemvalues[elemoff];
eisnull = op->d.arrayexpr.elemnulls[elemoff];
/* temporarily ignore null subarrays */
if (eisnull)
{
haveempty = true;
continue;
}
array = DatumGetArrayTypeP(arraydatum);
/* run-time double-check on element type */
if (element_type != ARR_ELEMTYPE(array))
ereport(ERROR,
(errcode(ERRCODE_DATATYPE_MISMATCH),
errmsg("cannot merge incompatible arrays"),
errdetail("Array with element type %s cannot be "
"included in ARRAY construct with element type %s.",
format_type_be(ARR_ELEMTYPE(array)),
format_type_be(element_type))));
this_ndims = ARR_NDIM(array);
/* temporarily ignore zero-dimensional subarrays */
if (this_ndims <= 0)
{
haveempty = true;
continue;
}
if (firstone)
{
/* Get sub-array details from first member */
elem_ndims = this_ndims;
ndims = elem_ndims + 1;
if (ndims <= 0 || ndims > MAXDIM)
ereport(ERROR,
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
Avoid splitting C string literals with \-newline Using \ is unnecessary and ugly, so remove that. While at it, stitch the literals back into a single line: we've long discouraged splitting error message literals even when they go past the 80 chars line limit, to improve greppability. Leave contrib/tablefunc alone. Discussion: https://postgr.es/m/20191223195156.GA12271@alvherre.pgsql
2019-12-24 16:37:13 +01:00
errmsg("number of array dimensions (%d) exceeds the maximum allowed (%d)",
ndims, MAXDIM)));
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
elem_dims = (int *) palloc(elem_ndims * sizeof(int));
memcpy(elem_dims, ARR_DIMS(array), elem_ndims * sizeof(int));
elem_lbs = (int *) palloc(elem_ndims * sizeof(int));
memcpy(elem_lbs, ARR_LBOUND(array), elem_ndims * sizeof(int));
firstone = false;
}
else
{
/* Check other sub-arrays are compatible */
if (elem_ndims != this_ndims ||
memcmp(elem_dims, ARR_DIMS(array),
elem_ndims * sizeof(int)) != 0 ||
memcmp(elem_lbs, ARR_LBOUND(array),
elem_ndims * sizeof(int)) != 0)
ereport(ERROR,
(errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
errmsg("multidimensional arrays must have array "
"expressions with matching dimensions")));
}
subdata[outer_nelems] = ARR_DATA_PTR(array);
subbitmaps[outer_nelems] = ARR_NULLBITMAP(array);
subbytes[outer_nelems] = ARR_SIZE(array) - ARR_DATA_OFFSET(array);
nbytes += subbytes[outer_nelems];
Fix oversights in array manipulation. The nested-arrays code path in ExecEvalArrayExpr() used palloc to allocate the result array, whereas every other array-creating function has used palloc0 since 18c0b4ecc. This mostly works, but unused bits past the end of the nulls bitmap may end up undefined. That causes valgrind complaints with -DWRITE_READ_PARSE_PLAN_TREES, and could cause planner misbehavior as cited in 18c0b4ecc. There seems no very good reason why we should strive to avoid palloc0 in just this one case, so fix it the easy way with s/palloc/palloc0/. While looking at that I noted that we also failed to check for overflow of "nbytes" and "nitems" while summing the sizes of the sub-arrays, potentially allowing a crash due to undersized output allocation. For "nbytes", follow the policy used by other array-munging code of checking for overflow after each addition. (As elsewhere, the last addition of the array's overhead space doesn't need an extra check, since palloc itself will catch a value between 1Gb and 2Gb.) For "nitems", there's no very good reason to sum the inputs at all, since we can perfectly well use ArrayGetNItems' result instead of ignoring it. Per discussion of this bug, also remove redundant zeroing of the nulls bitmap in array_set_element and array_set_slice. Patch by Alexander Lakhin and myself, per bug #17858 from Alexander Lakhin; thanks also to Richard Guo. These bugs are a dozen years old, so back-patch to all supported branches. Discussion: https://postgr.es/m/17858-8fd287fd3663d051@postgresql.org
2023-03-26 19:41:06 +02:00
/* check for overflow of total request */
if (!AllocSizeIsValid(nbytes))
ereport(ERROR,
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
errmsg("array size exceeds the maximum allowed (%d)",
(int) MaxAllocSize)));
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
subnitems[outer_nelems] = ArrayGetNItems(this_ndims,
ARR_DIMS(array));
havenulls |= ARR_HASNULL(array);
outer_nelems++;
}
/*
* If all items were null or empty arrays, return an empty array;
* otherwise, if some were and some weren't, raise error. (Note: we
* must special-case this somehow to avoid trying to generate a 1-D
* array formed from empty arrays. It's not ideal...)
*/
if (haveempty)
{
if (ndims == 0) /* didn't find any nonempty array */
{
*op->resvalue = PointerGetDatum(construct_empty_array(element_type));
return;
}
ereport(ERROR,
(errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
errmsg("multidimensional arrays must have array "
"expressions with matching dimensions")));
}
/* setup for multi-D array */
dims[0] = outer_nelems;
lbs[0] = 1;
expression eval, jit: Minor code cleanups. This mostly consists of using C99 style for loops, moving variables into narrower scopes, and a smattering of other minor improvements. Done separately to make it easier to review patches with actual functional changes. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:04:50 +01:00
for (int i = 1; i < ndims; i++)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
dims[i] = elem_dims[i - 1];
lbs[i] = elem_lbs[i - 1];
}
Prevent integer overflows in array subscripting calculations. While we were (mostly) careful about ensuring that the dimensions of arrays aren't large enough to cause integer overflow, the lower bound values were generally not checked. This allows situations where lower_bound + dimension overflows an integer. It seems that that's harmless so far as array reading is concerned, except that array elements with subscripts notionally exceeding INT_MAX are inaccessible. However, it confuses various array-assignment logic, resulting in a potential for memory stomps. Fix by adding checks that array lower bounds aren't large enough to cause lower_bound + dimension to overflow. (Note: this results in disallowing cases where the last subscript position would be exactly INT_MAX. In principle we could probably allow that, but there's a lot of code that computes lower_bound + dimension and would need adjustment. It seems doubtful that it's worth the trouble/risk to allow it.) Somewhat independently of that, array_set_element() was careless about possible overflow when checking the subscript of a fixed-length array, creating a different route to memory stomps. Fix that too. Security: CVE-2021-32027
2021-05-10 16:44:38 +02:00
/* check for subscript overflow */
Fix oversights in array manipulation. The nested-arrays code path in ExecEvalArrayExpr() used palloc to allocate the result array, whereas every other array-creating function has used palloc0 since 18c0b4ecc. This mostly works, but unused bits past the end of the nulls bitmap may end up undefined. That causes valgrind complaints with -DWRITE_READ_PARSE_PLAN_TREES, and could cause planner misbehavior as cited in 18c0b4ecc. There seems no very good reason why we should strive to avoid palloc0 in just this one case, so fix it the easy way with s/palloc/palloc0/. While looking at that I noted that we also failed to check for overflow of "nbytes" and "nitems" while summing the sizes of the sub-arrays, potentially allowing a crash due to undersized output allocation. For "nbytes", follow the policy used by other array-munging code of checking for overflow after each addition. (As elsewhere, the last addition of the array's overhead space doesn't need an extra check, since palloc itself will catch a value between 1Gb and 2Gb.) For "nitems", there's no very good reason to sum the inputs at all, since we can perfectly well use ArrayGetNItems' result instead of ignoring it. Per discussion of this bug, also remove redundant zeroing of the nulls bitmap in array_set_element and array_set_slice. Patch by Alexander Lakhin and myself, per bug #17858 from Alexander Lakhin; thanks also to Richard Guo. These bugs are a dozen years old, so back-patch to all supported branches. Discussion: https://postgr.es/m/17858-8fd287fd3663d051@postgresql.org
2023-03-26 19:41:06 +02:00
nitems = ArrayGetNItems(ndims, dims);
Prevent integer overflows in array subscripting calculations. While we were (mostly) careful about ensuring that the dimensions of arrays aren't large enough to cause integer overflow, the lower bound values were generally not checked. This allows situations where lower_bound + dimension overflows an integer. It seems that that's harmless so far as array reading is concerned, except that array elements with subscripts notionally exceeding INT_MAX are inaccessible. However, it confuses various array-assignment logic, resulting in a potential for memory stomps. Fix by adding checks that array lower bounds aren't large enough to cause lower_bound + dimension to overflow. (Note: this results in disallowing cases where the last subscript position would be exactly INT_MAX. In principle we could probably allow that, but there's a lot of code that computes lower_bound + dimension and would need adjustment. It seems doubtful that it's worth the trouble/risk to allow it.) Somewhat independently of that, array_set_element() was careless about possible overflow when checking the subscript of a fixed-length array, creating a different route to memory stomps. Fix that too. Security: CVE-2021-32027
2021-05-10 16:44:38 +02:00
ArrayCheckBounds(ndims, dims, lbs);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
if (havenulls)
{
dataoffset = ARR_OVERHEAD_WITHNULLS(ndims, nitems);
nbytes += dataoffset;
}
else
{
dataoffset = 0; /* marker for no null bitmap */
nbytes += ARR_OVERHEAD_NONULLS(ndims);
}
Fix oversights in array manipulation. The nested-arrays code path in ExecEvalArrayExpr() used palloc to allocate the result array, whereas every other array-creating function has used palloc0 since 18c0b4ecc. This mostly works, but unused bits past the end of the nulls bitmap may end up undefined. That causes valgrind complaints with -DWRITE_READ_PARSE_PLAN_TREES, and could cause planner misbehavior as cited in 18c0b4ecc. There seems no very good reason why we should strive to avoid palloc0 in just this one case, so fix it the easy way with s/palloc/palloc0/. While looking at that I noted that we also failed to check for overflow of "nbytes" and "nitems" while summing the sizes of the sub-arrays, potentially allowing a crash due to undersized output allocation. For "nbytes", follow the policy used by other array-munging code of checking for overflow after each addition. (As elsewhere, the last addition of the array's overhead space doesn't need an extra check, since palloc itself will catch a value between 1Gb and 2Gb.) For "nitems", there's no very good reason to sum the inputs at all, since we can perfectly well use ArrayGetNItems' result instead of ignoring it. Per discussion of this bug, also remove redundant zeroing of the nulls bitmap in array_set_element and array_set_slice. Patch by Alexander Lakhin and myself, per bug #17858 from Alexander Lakhin; thanks also to Richard Guo. These bugs are a dozen years old, so back-patch to all supported branches. Discussion: https://postgr.es/m/17858-8fd287fd3663d051@postgresql.org
2023-03-26 19:41:06 +02:00
result = (ArrayType *) palloc0(nbytes);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
SET_VARSIZE(result, nbytes);
result->ndim = ndims;
result->dataoffset = dataoffset;
result->elemtype = element_type;
memcpy(ARR_DIMS(result), dims, ndims * sizeof(int));
memcpy(ARR_LBOUND(result), lbs, ndims * sizeof(int));
dat = ARR_DATA_PTR(result);
iitem = 0;
expression eval, jit: Minor code cleanups. This mostly consists of using C99 style for loops, moving variables into narrower scopes, and a smattering of other minor improvements. Done separately to make it easier to review patches with actual functional changes. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:04:50 +01:00
for (int i = 0; i < outer_nelems; i++)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
memcpy(dat, subdata[i], subbytes[i]);
dat += subbytes[i];
if (havenulls)
array_bitmap_copy(ARR_NULLBITMAP(result), iitem,
subbitmaps[i], 0,
subnitems[i]);
iitem += subnitems[i];
}
}
*op->resvalue = PointerGetDatum(result);
}
/*
* Evaluate an ArrayCoerceExpr expression.
*
* Source array is in step's result variable.
*/
void
Support arrays over domains. Allowing arrays with a domain type as their element type was left un-done in the original domain patch, but not for any very good reason. This omission leads to such surprising results as array_agg() not working on a domain column, because the parser can't identify a suitable output type for the polymorphic aggregate. In order to fix this, first clean up the APIs of coerce_to_domain() and some internal functions in parse_coerce.c so that we consistently pass around a CoercionContext along with CoercionForm. Previously, we sometimes passed an "isExplicit" boolean flag instead, which is strictly less information; and coerce_to_domain() didn't even get that, but instead had to reverse-engineer isExplicit from CoercionForm. That's contrary to the documentation in primnodes.h that says that CoercionForm only affects display and not semantics. I don't think this change fixes any live bugs, but it makes things more consistent. The main reason for doing it though is that now build_coercion_expression() receives ccontext, which it needs in order to be able to recursively invoke coerce_to_target_type(). Next, reimplement ArrayCoerceExpr so that the node does not directly know any details of what has to be done to the individual array elements while performing the array coercion. Instead, the per-element processing is represented by a sub-expression whose input is a source array element and whose output is a target array element. This simplifies life in parse_coerce.c, because it can build that sub-expression by a recursive invocation of coerce_to_target_type(). The executor now handles the per-element processing as a compiled expression instead of hard-wired code. The main advantage of this is that we can use a single ArrayCoerceExpr to handle as many as three successive steps per element: base type conversion, typmod coercion, and domain constraint checking. The old code used two stacked ArrayCoerceExprs to handle type + typmod coercion, which was pretty inefficient, and adding yet another array deconstruction to do domain constraint checking seemed very unappetizing. In the case where we just need a single, very simple coercion function, doing this straightforwardly leads to a noticeable increase in the per-array-element runtime cost. Hence, add an additional shortcut evalfunc in execExprInterp.c that skips unnecessary overhead for that specific form of expression. The runtime speed of simple cases is within 1% or so of where it was before, while cases that previously required two levels of array processing are significantly faster. Finally, create an implicit array type for every domain type, as we do for base types, enums, etc. Everything except the array-coercion case seems to just work without further effort. Tom Lane, reviewed by Andrew Dunstan Discussion: https://postgr.es/m/9852.1499791473@sss.pgh.pa.us
2017-09-30 19:40:56 +02:00
ExecEvalArrayCoerce(ExprState *state, ExprEvalStep *op, ExprContext *econtext)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
Datum arraydatum;
/* NULL array -> NULL result */
if (*op->resnull)
return;
arraydatum = *op->resvalue;
/*
* If it's binary-compatible, modify the element type in the array header,
* but otherwise leave the array as we received it.
*/
Support arrays over domains. Allowing arrays with a domain type as their element type was left un-done in the original domain patch, but not for any very good reason. This omission leads to such surprising results as array_agg() not working on a domain column, because the parser can't identify a suitable output type for the polymorphic aggregate. In order to fix this, first clean up the APIs of coerce_to_domain() and some internal functions in parse_coerce.c so that we consistently pass around a CoercionContext along with CoercionForm. Previously, we sometimes passed an "isExplicit" boolean flag instead, which is strictly less information; and coerce_to_domain() didn't even get that, but instead had to reverse-engineer isExplicit from CoercionForm. That's contrary to the documentation in primnodes.h that says that CoercionForm only affects display and not semantics. I don't think this change fixes any live bugs, but it makes things more consistent. The main reason for doing it though is that now build_coercion_expression() receives ccontext, which it needs in order to be able to recursively invoke coerce_to_target_type(). Next, reimplement ArrayCoerceExpr so that the node does not directly know any details of what has to be done to the individual array elements while performing the array coercion. Instead, the per-element processing is represented by a sub-expression whose input is a source array element and whose output is a target array element. This simplifies life in parse_coerce.c, because it can build that sub-expression by a recursive invocation of coerce_to_target_type(). The executor now handles the per-element processing as a compiled expression instead of hard-wired code. The main advantage of this is that we can use a single ArrayCoerceExpr to handle as many as three successive steps per element: base type conversion, typmod coercion, and domain constraint checking. The old code used two stacked ArrayCoerceExprs to handle type + typmod coercion, which was pretty inefficient, and adding yet another array deconstruction to do domain constraint checking seemed very unappetizing. In the case where we just need a single, very simple coercion function, doing this straightforwardly leads to a noticeable increase in the per-array-element runtime cost. Hence, add an additional shortcut evalfunc in execExprInterp.c that skips unnecessary overhead for that specific form of expression. The runtime speed of simple cases is within 1% or so of where it was before, while cases that previously required two levels of array processing are significantly faster. Finally, create an implicit array type for every domain type, as we do for base types, enums, etc. Everything except the array-coercion case seems to just work without further effort. Tom Lane, reviewed by Andrew Dunstan Discussion: https://postgr.es/m/9852.1499791473@sss.pgh.pa.us
2017-09-30 19:40:56 +02:00
if (op->d.arraycoerce.elemexprstate == NULL)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
/* Detoast input array if necessary, and copy in any case */
ArrayType *array = DatumGetArrayTypePCopy(arraydatum);
ARR_ELEMTYPE(array) = op->d.arraycoerce.resultelemtype;
*op->resvalue = PointerGetDatum(array);
return;
}
/*
Support arrays over domains. Allowing arrays with a domain type as their element type was left un-done in the original domain patch, but not for any very good reason. This omission leads to such surprising results as array_agg() not working on a domain column, because the parser can't identify a suitable output type for the polymorphic aggregate. In order to fix this, first clean up the APIs of coerce_to_domain() and some internal functions in parse_coerce.c so that we consistently pass around a CoercionContext along with CoercionForm. Previously, we sometimes passed an "isExplicit" boolean flag instead, which is strictly less information; and coerce_to_domain() didn't even get that, but instead had to reverse-engineer isExplicit from CoercionForm. That's contrary to the documentation in primnodes.h that says that CoercionForm only affects display and not semantics. I don't think this change fixes any live bugs, but it makes things more consistent. The main reason for doing it though is that now build_coercion_expression() receives ccontext, which it needs in order to be able to recursively invoke coerce_to_target_type(). Next, reimplement ArrayCoerceExpr so that the node does not directly know any details of what has to be done to the individual array elements while performing the array coercion. Instead, the per-element processing is represented by a sub-expression whose input is a source array element and whose output is a target array element. This simplifies life in parse_coerce.c, because it can build that sub-expression by a recursive invocation of coerce_to_target_type(). The executor now handles the per-element processing as a compiled expression instead of hard-wired code. The main advantage of this is that we can use a single ArrayCoerceExpr to handle as many as three successive steps per element: base type conversion, typmod coercion, and domain constraint checking. The old code used two stacked ArrayCoerceExprs to handle type + typmod coercion, which was pretty inefficient, and adding yet another array deconstruction to do domain constraint checking seemed very unappetizing. In the case where we just need a single, very simple coercion function, doing this straightforwardly leads to a noticeable increase in the per-array-element runtime cost. Hence, add an additional shortcut evalfunc in execExprInterp.c that skips unnecessary overhead for that specific form of expression. The runtime speed of simple cases is within 1% or so of where it was before, while cases that previously required two levels of array processing are significantly faster. Finally, create an implicit array type for every domain type, as we do for base types, enums, etc. Everything except the array-coercion case seems to just work without further effort. Tom Lane, reviewed by Andrew Dunstan Discussion: https://postgr.es/m/9852.1499791473@sss.pgh.pa.us
2017-09-30 19:40:56 +02:00
* Use array_map to apply the sub-expression to each array element.
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
*/
Support arrays over domains. Allowing arrays with a domain type as their element type was left un-done in the original domain patch, but not for any very good reason. This omission leads to such surprising results as array_agg() not working on a domain column, because the parser can't identify a suitable output type for the polymorphic aggregate. In order to fix this, first clean up the APIs of coerce_to_domain() and some internal functions in parse_coerce.c so that we consistently pass around a CoercionContext along with CoercionForm. Previously, we sometimes passed an "isExplicit" boolean flag instead, which is strictly less information; and coerce_to_domain() didn't even get that, but instead had to reverse-engineer isExplicit from CoercionForm. That's contrary to the documentation in primnodes.h that says that CoercionForm only affects display and not semantics. I don't think this change fixes any live bugs, but it makes things more consistent. The main reason for doing it though is that now build_coercion_expression() receives ccontext, which it needs in order to be able to recursively invoke coerce_to_target_type(). Next, reimplement ArrayCoerceExpr so that the node does not directly know any details of what has to be done to the individual array elements while performing the array coercion. Instead, the per-element processing is represented by a sub-expression whose input is a source array element and whose output is a target array element. This simplifies life in parse_coerce.c, because it can build that sub-expression by a recursive invocation of coerce_to_target_type(). The executor now handles the per-element processing as a compiled expression instead of hard-wired code. The main advantage of this is that we can use a single ArrayCoerceExpr to handle as many as three successive steps per element: base type conversion, typmod coercion, and domain constraint checking. The old code used two stacked ArrayCoerceExprs to handle type + typmod coercion, which was pretty inefficient, and adding yet another array deconstruction to do domain constraint checking seemed very unappetizing. In the case where we just need a single, very simple coercion function, doing this straightforwardly leads to a noticeable increase in the per-array-element runtime cost. Hence, add an additional shortcut evalfunc in execExprInterp.c that skips unnecessary overhead for that specific form of expression. The runtime speed of simple cases is within 1% or so of where it was before, while cases that previously required two levels of array processing are significantly faster. Finally, create an implicit array type for every domain type, as we do for base types, enums, etc. Everything except the array-coercion case seems to just work without further effort. Tom Lane, reviewed by Andrew Dunstan Discussion: https://postgr.es/m/9852.1499791473@sss.pgh.pa.us
2017-09-30 19:40:56 +02:00
*op->resvalue = array_map(arraydatum,
op->d.arraycoerce.elemexprstate,
econtext,
op->d.arraycoerce.resultelemtype,
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
op->d.arraycoerce.amstate);
}
/*
* Evaluate a ROW() expression.
*
* The individual columns have already been evaluated into
* op->d.row.elemvalues[]/elemnulls[].
*/
void
ExecEvalRow(ExprState *state, ExprEvalStep *op)
{
HeapTuple tuple;
/* build tuple from evaluated field values */
tuple = heap_form_tuple(op->d.row.tupdesc,
op->d.row.elemvalues,
op->d.row.elemnulls);
*op->resvalue = HeapTupleGetDatum(tuple);
*op->resnull = false;
}
/*
* Evaluate GREATEST() or LEAST() expression (note this is *not* MIN()/MAX()).
*
* All of the to-be-compared expressions have already been evaluated into
* op->d.minmax.values[]/nulls[].
*/
void
ExecEvalMinMax(ExprState *state, ExprEvalStep *op)
{
Datum *values = op->d.minmax.values;
bool *nulls = op->d.minmax.nulls;
FunctionCallInfo fcinfo = op->d.minmax.fcinfo_data;
MinMaxOp operator = op->d.minmax.op;
/* set at initialization */
Change function call information to be variable length. Before this change FunctionCallInfoData, the struct arguments etc for V1 function calls are stored in, always had space for FUNC_MAX_ARGS/100 arguments, storing datums and their nullness in two arrays. For nearly every function call 100 arguments is far more than needed, therefore wasting memory. Arg and argnull being two separate arrays also guarantees that to access a single argument, two cachelines have to be touched. Change the layout so there's a single variable-length array with pairs of value / isnull. That drastically reduces memory consumption for most function calls (on x86-64 a two argument function now uses 64bytes, previously 936 bytes), and makes it very likely that argument value and its nullness are on the same cacheline. Arguments are stored in a new NullableDatum struct, which, due to padding, needs more memory per argument than before. But as usually far fewer arguments are stored, and individual arguments are cheaper to access, that's still a clear win. It's likely that there's other places where conversion to NullableDatum arrays would make sense, e.g. TupleTableSlots, but that's for another commit. Because the function call information is now variable-length allocations have to take the number of arguments into account. For heap allocations that can be done with SizeForFunctionCallInfoData(), for on-stack allocations there's a new LOCAL_FCINFO(name, nargs) macro that helps to allocate an appropriately sized and aligned variable. Some places with stack allocation function call information don't know the number of arguments at compile time, and currently variably sized stack allocations aren't allowed in postgres. Therefore allow for FUNC_MAX_ARGS space in these cases. They're not that common, so for now that seems acceptable. Because of the need to allocate FunctionCallInfo of the appropriate size, older extensions may need to update their code. To avoid subtle breakages, the FunctionCallInfoData struct has been renamed to FunctionCallInfoBaseData. Most code only references FunctionCallInfo, so that shouldn't cause much collateral damage. This change is also a prerequisite for more efficient expression JIT compilation (by allocating the function call information on the stack, allowing LLVM to optimize it away); previously the size of the call information caused problems inside LLVM's optimizer. Author: Andres Freund Reviewed-By: Tom Lane Discussion: https://postgr.es/m/20180605172952.x34m5uz6ju6enaem@alap3.anarazel.de
2019-01-26 23:17:52 +01:00
Assert(fcinfo->args[0].isnull == false);
Assert(fcinfo->args[1].isnull == false);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
/* default to null result */
*op->resnull = true;
expression eval, jit: Minor code cleanups. This mostly consists of using C99 style for loops, moving variables into narrower scopes, and a smattering of other minor improvements. Done separately to make it easier to review patches with actual functional changes. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:04:50 +01:00
for (int off = 0; off < op->d.minmax.nelems; off++)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
/* ignore NULL inputs */
if (nulls[off])
continue;
if (*op->resnull)
{
/* first nonnull input, adopt value */
*op->resvalue = values[off];
*op->resnull = false;
}
else
{
int cmpresult;
/* apply comparison function */
Change function call information to be variable length. Before this change FunctionCallInfoData, the struct arguments etc for V1 function calls are stored in, always had space for FUNC_MAX_ARGS/100 arguments, storing datums and their nullness in two arrays. For nearly every function call 100 arguments is far more than needed, therefore wasting memory. Arg and argnull being two separate arrays also guarantees that to access a single argument, two cachelines have to be touched. Change the layout so there's a single variable-length array with pairs of value / isnull. That drastically reduces memory consumption for most function calls (on x86-64 a two argument function now uses 64bytes, previously 936 bytes), and makes it very likely that argument value and its nullness are on the same cacheline. Arguments are stored in a new NullableDatum struct, which, due to padding, needs more memory per argument than before. But as usually far fewer arguments are stored, and individual arguments are cheaper to access, that's still a clear win. It's likely that there's other places where conversion to NullableDatum arrays would make sense, e.g. TupleTableSlots, but that's for another commit. Because the function call information is now variable-length allocations have to take the number of arguments into account. For heap allocations that can be done with SizeForFunctionCallInfoData(), for on-stack allocations there's a new LOCAL_FCINFO(name, nargs) macro that helps to allocate an appropriately sized and aligned variable. Some places with stack allocation function call information don't know the number of arguments at compile time, and currently variably sized stack allocations aren't allowed in postgres. Therefore allow for FUNC_MAX_ARGS space in these cases. They're not that common, so for now that seems acceptable. Because of the need to allocate FunctionCallInfo of the appropriate size, older extensions may need to update their code. To avoid subtle breakages, the FunctionCallInfoData struct has been renamed to FunctionCallInfoBaseData. Most code only references FunctionCallInfo, so that shouldn't cause much collateral damage. This change is also a prerequisite for more efficient expression JIT compilation (by allocating the function call information on the stack, allowing LLVM to optimize it away); previously the size of the call information caused problems inside LLVM's optimizer. Author: Andres Freund Reviewed-By: Tom Lane Discussion: https://postgr.es/m/20180605172952.x34m5uz6ju6enaem@alap3.anarazel.de
2019-01-26 23:17:52 +01:00
fcinfo->args[0].value = *op->resvalue;
fcinfo->args[1].value = values[off];
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
fcinfo->isnull = false;
cmpresult = DatumGetInt32(FunctionCallInvoke(fcinfo));
if (fcinfo->isnull) /* probably should not happen */
continue;
if (cmpresult > 0 && operator == IS_LEAST)
*op->resvalue = values[off];
else if (cmpresult < 0 && operator == IS_GREATEST)
*op->resvalue = values[off];
}
}
}
/*
* Evaluate a FieldSelect node.
*
* Source record is in step's result variable.
*/
void
ExecEvalFieldSelect(ExprState *state, ExprEvalStep *op, ExprContext *econtext)
{
AttrNumber fieldnum = op->d.fieldselect.fieldnum;
Datum tupDatum;
HeapTupleHeader tuple;
Oid tupType;
int32 tupTypmod;
TupleDesc tupDesc;
Form_pg_attribute attr;
HeapTupleData tmptup;
/* NULL record -> NULL result */
if (*op->resnull)
return;
tupDatum = *op->resvalue;
Make plpgsql use its DTYPE_REC code paths for composite-type variables. Formerly, DTYPE_REC was used only for variables declared as "record"; variables of named composite types used DTYPE_ROW, which is faster for some purposes but much less flexible. In particular, the ROW code paths are entirely incapable of dealing with DDL-caused changes to the number or data types of the columns of a row variable, once a particular plpgsql function has been parsed for the first time in a session. And, since the stored representation of a ROW isn't a tuple, there wasn't any easy way to deal with variables of domain-over-composite types, since the domain constraint checking code would expect the value to be checked to be a tuple. A lesser, but still real, annoyance is that ROW format cannot represent a true NULL composite value, only a row of per-field NULL values, which is not exactly the same thing. Hence, switch to using DTYPE_REC for all composite-typed variables, whether "record", named composite type, or domain over named composite type. DTYPE_ROW remains but is used only for its native purpose, to represent a fixed-at-compile-time list of variables, for instance the targets of an INTO clause. To accomplish this without taking significant performance losses, introduce infrastructure that allows storing composite-type variables as "expanded objects", similar to the "expanded array" infrastructure introduced in commit 1dc5ebc90. A composite variable's value is thereby kept (most of the time) in the form of separate Datums, so that field accesses and updates are not much more expensive than they were in the ROW format. This holds the line, more or less, on performance of variables of named composite types in field-access-intensive microbenchmarks, and makes variables declared "record" perform much better than before in similar tests. In addition, the logic involved with enforcing composite-domain constraints against updates of individual fields is in the expanded record infrastructure not plpgsql proper, so that it might be reusable for other purposes. In further support of this, introduce a typcache feature for assigning a unique-within-process identifier to each distinct tuple descriptor of interest; in particular, DDL alterations on composite types result in a new identifier for that type. This allows very cheap detection of the need to refresh tupdesc-dependent data. This improves on the "tupDescSeqNo" idea I had in commit 687f096ea: that assigned identifying sequence numbers to successive versions of individual composite types, but the numbers were not unique across different types, nor was there support for assigning numbers to registered record types. In passing, allow plpgsql functions to accept as well as return type "record". There was no good reason for the old restriction, and it was out of step with most of the other PLs. Tom Lane, reviewed by Pavel Stehule Discussion: https://postgr.es/m/8962.1514399547@sss.pgh.pa.us
2018-02-14 00:52:21 +01:00
/* We can special-case expanded records for speed */
if (VARATT_IS_EXTERNAL_EXPANDED(DatumGetPointer(tupDatum)))
{
ExpandedRecordHeader *erh = (ExpandedRecordHeader *) DatumGetEOHP(tupDatum);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
Make plpgsql use its DTYPE_REC code paths for composite-type variables. Formerly, DTYPE_REC was used only for variables declared as "record"; variables of named composite types used DTYPE_ROW, which is faster for some purposes but much less flexible. In particular, the ROW code paths are entirely incapable of dealing with DDL-caused changes to the number or data types of the columns of a row variable, once a particular plpgsql function has been parsed for the first time in a session. And, since the stored representation of a ROW isn't a tuple, there wasn't any easy way to deal with variables of domain-over-composite types, since the domain constraint checking code would expect the value to be checked to be a tuple. A lesser, but still real, annoyance is that ROW format cannot represent a true NULL composite value, only a row of per-field NULL values, which is not exactly the same thing. Hence, switch to using DTYPE_REC for all composite-typed variables, whether "record", named composite type, or domain over named composite type. DTYPE_ROW remains but is used only for its native purpose, to represent a fixed-at-compile-time list of variables, for instance the targets of an INTO clause. To accomplish this without taking significant performance losses, introduce infrastructure that allows storing composite-type variables as "expanded objects", similar to the "expanded array" infrastructure introduced in commit 1dc5ebc90. A composite variable's value is thereby kept (most of the time) in the form of separate Datums, so that field accesses and updates are not much more expensive than they were in the ROW format. This holds the line, more or less, on performance of variables of named composite types in field-access-intensive microbenchmarks, and makes variables declared "record" perform much better than before in similar tests. In addition, the logic involved with enforcing composite-domain constraints against updates of individual fields is in the expanded record infrastructure not plpgsql proper, so that it might be reusable for other purposes. In further support of this, introduce a typcache feature for assigning a unique-within-process identifier to each distinct tuple descriptor of interest; in particular, DDL alterations on composite types result in a new identifier for that type. This allows very cheap detection of the need to refresh tupdesc-dependent data. This improves on the "tupDescSeqNo" idea I had in commit 687f096ea: that assigned identifying sequence numbers to successive versions of individual composite types, but the numbers were not unique across different types, nor was there support for assigning numbers to registered record types. In passing, allow plpgsql functions to accept as well as return type "record". There was no good reason for the old restriction, and it was out of step with most of the other PLs. Tom Lane, reviewed by Pavel Stehule Discussion: https://postgr.es/m/8962.1514399547@sss.pgh.pa.us
2018-02-14 00:52:21 +01:00
Assert(erh->er_magic == ER_MAGIC);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
Make plpgsql use its DTYPE_REC code paths for composite-type variables. Formerly, DTYPE_REC was used only for variables declared as "record"; variables of named composite types used DTYPE_ROW, which is faster for some purposes but much less flexible. In particular, the ROW code paths are entirely incapable of dealing with DDL-caused changes to the number or data types of the columns of a row variable, once a particular plpgsql function has been parsed for the first time in a session. And, since the stored representation of a ROW isn't a tuple, there wasn't any easy way to deal with variables of domain-over-composite types, since the domain constraint checking code would expect the value to be checked to be a tuple. A lesser, but still real, annoyance is that ROW format cannot represent a true NULL composite value, only a row of per-field NULL values, which is not exactly the same thing. Hence, switch to using DTYPE_REC for all composite-typed variables, whether "record", named composite type, or domain over named composite type. DTYPE_ROW remains but is used only for its native purpose, to represent a fixed-at-compile-time list of variables, for instance the targets of an INTO clause. To accomplish this without taking significant performance losses, introduce infrastructure that allows storing composite-type variables as "expanded objects", similar to the "expanded array" infrastructure introduced in commit 1dc5ebc90. A composite variable's value is thereby kept (most of the time) in the form of separate Datums, so that field accesses and updates are not much more expensive than they were in the ROW format. This holds the line, more or less, on performance of variables of named composite types in field-access-intensive microbenchmarks, and makes variables declared "record" perform much better than before in similar tests. In addition, the logic involved with enforcing composite-domain constraints against updates of individual fields is in the expanded record infrastructure not plpgsql proper, so that it might be reusable for other purposes. In further support of this, introduce a typcache feature for assigning a unique-within-process identifier to each distinct tuple descriptor of interest; in particular, DDL alterations on composite types result in a new identifier for that type. This allows very cheap detection of the need to refresh tupdesc-dependent data. This improves on the "tupDescSeqNo" idea I had in commit 687f096ea: that assigned identifying sequence numbers to successive versions of individual composite types, but the numbers were not unique across different types, nor was there support for assigning numbers to registered record types. In passing, allow plpgsql functions to accept as well as return type "record". There was no good reason for the old restriction, and it was out of step with most of the other PLs. Tom Lane, reviewed by Pavel Stehule Discussion: https://postgr.es/m/8962.1514399547@sss.pgh.pa.us
2018-02-14 00:52:21 +01:00
/* Extract record's TupleDesc */
tupDesc = expanded_record_get_tupdesc(erh);
/*
* Find field's attr record. Note we don't support system columns
* here: a datum tuple doesn't have valid values for most of the
* interesting system columns anyway.
*/
if (fieldnum <= 0) /* should never happen */
elog(ERROR, "unsupported reference to system column %d in FieldSelect",
fieldnum);
if (fieldnum > tupDesc->natts) /* should never happen */
elog(ERROR, "attribute number %d exceeds number of columns %d",
fieldnum, tupDesc->natts);
attr = TupleDescAttr(tupDesc, fieldnum - 1);
/* Check for dropped column, and force a NULL result if so */
if (attr->attisdropped)
{
*op->resnull = true;
return;
}
/* Check for type mismatch --- possible after ALTER COLUMN TYPE? */
/* As in CheckVarSlotCompatibility, we should but can't check typmod */
if (op->d.fieldselect.resulttype != attr->atttypid)
ereport(ERROR,
(errcode(ERRCODE_DATATYPE_MISMATCH),
errmsg("attribute %d has wrong type", fieldnum),
errdetail("Table has type %s, but query expects %s.",
format_type_be(attr->atttypid),
format_type_be(op->d.fieldselect.resulttype))));
/* extract the field */
*op->resvalue = expanded_record_get_field(erh, fieldnum,
op->resnull);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
}
Make plpgsql use its DTYPE_REC code paths for composite-type variables. Formerly, DTYPE_REC was used only for variables declared as "record"; variables of named composite types used DTYPE_ROW, which is faster for some purposes but much less flexible. In particular, the ROW code paths are entirely incapable of dealing with DDL-caused changes to the number or data types of the columns of a row variable, once a particular plpgsql function has been parsed for the first time in a session. And, since the stored representation of a ROW isn't a tuple, there wasn't any easy way to deal with variables of domain-over-composite types, since the domain constraint checking code would expect the value to be checked to be a tuple. A lesser, but still real, annoyance is that ROW format cannot represent a true NULL composite value, only a row of per-field NULL values, which is not exactly the same thing. Hence, switch to using DTYPE_REC for all composite-typed variables, whether "record", named composite type, or domain over named composite type. DTYPE_ROW remains but is used only for its native purpose, to represent a fixed-at-compile-time list of variables, for instance the targets of an INTO clause. To accomplish this without taking significant performance losses, introduce infrastructure that allows storing composite-type variables as "expanded objects", similar to the "expanded array" infrastructure introduced in commit 1dc5ebc90. A composite variable's value is thereby kept (most of the time) in the form of separate Datums, so that field accesses and updates are not much more expensive than they were in the ROW format. This holds the line, more or less, on performance of variables of named composite types in field-access-intensive microbenchmarks, and makes variables declared "record" perform much better than before in similar tests. In addition, the logic involved with enforcing composite-domain constraints against updates of individual fields is in the expanded record infrastructure not plpgsql proper, so that it might be reusable for other purposes. In further support of this, introduce a typcache feature for assigning a unique-within-process identifier to each distinct tuple descriptor of interest; in particular, DDL alterations on composite types result in a new identifier for that type. This allows very cheap detection of the need to refresh tupdesc-dependent data. This improves on the "tupDescSeqNo" idea I had in commit 687f096ea: that assigned identifying sequence numbers to successive versions of individual composite types, but the numbers were not unique across different types, nor was there support for assigning numbers to registered record types. In passing, allow plpgsql functions to accept as well as return type "record". There was no good reason for the old restriction, and it was out of step with most of the other PLs. Tom Lane, reviewed by Pavel Stehule Discussion: https://postgr.es/m/8962.1514399547@sss.pgh.pa.us
2018-02-14 00:52:21 +01:00
else
{
/* Get the composite datum and extract its type fields */
tuple = DatumGetHeapTupleHeader(tupDatum);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
Make plpgsql use its DTYPE_REC code paths for composite-type variables. Formerly, DTYPE_REC was used only for variables declared as "record"; variables of named composite types used DTYPE_ROW, which is faster for some purposes but much less flexible. In particular, the ROW code paths are entirely incapable of dealing with DDL-caused changes to the number or data types of the columns of a row variable, once a particular plpgsql function has been parsed for the first time in a session. And, since the stored representation of a ROW isn't a tuple, there wasn't any easy way to deal with variables of domain-over-composite types, since the domain constraint checking code would expect the value to be checked to be a tuple. A lesser, but still real, annoyance is that ROW format cannot represent a true NULL composite value, only a row of per-field NULL values, which is not exactly the same thing. Hence, switch to using DTYPE_REC for all composite-typed variables, whether "record", named composite type, or domain over named composite type. DTYPE_ROW remains but is used only for its native purpose, to represent a fixed-at-compile-time list of variables, for instance the targets of an INTO clause. To accomplish this without taking significant performance losses, introduce infrastructure that allows storing composite-type variables as "expanded objects", similar to the "expanded array" infrastructure introduced in commit 1dc5ebc90. A composite variable's value is thereby kept (most of the time) in the form of separate Datums, so that field accesses and updates are not much more expensive than they were in the ROW format. This holds the line, more or less, on performance of variables of named composite types in field-access-intensive microbenchmarks, and makes variables declared "record" perform much better than before in similar tests. In addition, the logic involved with enforcing composite-domain constraints against updates of individual fields is in the expanded record infrastructure not plpgsql proper, so that it might be reusable for other purposes. In further support of this, introduce a typcache feature for assigning a unique-within-process identifier to each distinct tuple descriptor of interest; in particular, DDL alterations on composite types result in a new identifier for that type. This allows very cheap detection of the need to refresh tupdesc-dependent data. This improves on the "tupDescSeqNo" idea I had in commit 687f096ea: that assigned identifying sequence numbers to successive versions of individual composite types, but the numbers were not unique across different types, nor was there support for assigning numbers to registered record types. In passing, allow plpgsql functions to accept as well as return type "record". There was no good reason for the old restriction, and it was out of step with most of the other PLs. Tom Lane, reviewed by Pavel Stehule Discussion: https://postgr.es/m/8962.1514399547@sss.pgh.pa.us
2018-02-14 00:52:21 +01:00
tupType = HeapTupleHeaderGetTypeId(tuple);
tupTypmod = HeapTupleHeaderGetTypMod(tuple);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
Make plpgsql use its DTYPE_REC code paths for composite-type variables. Formerly, DTYPE_REC was used only for variables declared as "record"; variables of named composite types used DTYPE_ROW, which is faster for some purposes but much less flexible. In particular, the ROW code paths are entirely incapable of dealing with DDL-caused changes to the number or data types of the columns of a row variable, once a particular plpgsql function has been parsed for the first time in a session. And, since the stored representation of a ROW isn't a tuple, there wasn't any easy way to deal with variables of domain-over-composite types, since the domain constraint checking code would expect the value to be checked to be a tuple. A lesser, but still real, annoyance is that ROW format cannot represent a true NULL composite value, only a row of per-field NULL values, which is not exactly the same thing. Hence, switch to using DTYPE_REC for all composite-typed variables, whether "record", named composite type, or domain over named composite type. DTYPE_ROW remains but is used only for its native purpose, to represent a fixed-at-compile-time list of variables, for instance the targets of an INTO clause. To accomplish this without taking significant performance losses, introduce infrastructure that allows storing composite-type variables as "expanded objects", similar to the "expanded array" infrastructure introduced in commit 1dc5ebc90. A composite variable's value is thereby kept (most of the time) in the form of separate Datums, so that field accesses and updates are not much more expensive than they were in the ROW format. This holds the line, more or less, on performance of variables of named composite types in field-access-intensive microbenchmarks, and makes variables declared "record" perform much better than before in similar tests. In addition, the logic involved with enforcing composite-domain constraints against updates of individual fields is in the expanded record infrastructure not plpgsql proper, so that it might be reusable for other purposes. In further support of this, introduce a typcache feature for assigning a unique-within-process identifier to each distinct tuple descriptor of interest; in particular, DDL alterations on composite types result in a new identifier for that type. This allows very cheap detection of the need to refresh tupdesc-dependent data. This improves on the "tupDescSeqNo" idea I had in commit 687f096ea: that assigned identifying sequence numbers to successive versions of individual composite types, but the numbers were not unique across different types, nor was there support for assigning numbers to registered record types. In passing, allow plpgsql functions to accept as well as return type "record". There was no good reason for the old restriction, and it was out of step with most of the other PLs. Tom Lane, reviewed by Pavel Stehule Discussion: https://postgr.es/m/8962.1514399547@sss.pgh.pa.us
2018-02-14 00:52:21 +01:00
/* Lookup tupdesc if first time through or if type changes */
tupDesc = get_cached_rowtype(tupType, tupTypmod,
Redesign the caching done by get_cached_rowtype(). Previously, get_cached_rowtype() cached a pointer to a reference-counted tuple descriptor from the typcache, relying on the ExprContextCallback mechanism to release the tupdesc refcount when the expression tree using the tupdesc was destroyed. This worked fine when it was designed, but the introduction of within-DO-block COMMITs broke it. The refcount is logged in a transaction-lifespan resource owner, but plpgsql won't destroy simple expressions made within the DO block (before its first commit) until the DO block is exited. That results in a warning about a leaked tupdesc refcount when the COMMIT destroys the original resource owner, and then an error about the active resource owner not holding a matching refcount when the expression is destroyed. To fix, get rid of the need to have a shutdown callback at all, by instead caching a pointer to the relevant typcache entry. Those survive for the life of the backend, so we needn't worry about the pointer becoming stale. (For registered RECORD types, we can still cache a pointer to the tupdesc, knowing that it won't change for the life of the backend.) This mechanism has been in use in plpgsql and expandedrecord.c since commit 4b93f5799, and seems to work well. This change requires modifying the ExprEvalStep structs used by the relevant expression step types, which is slightly worrisome for back-patching. However, there seems no good reason for extensions to be familiar with the details of these particular sub-structs. Per report from Rohit Bhogate. Back-patch to v11 where within-DO-block COMMITs became a thing. Discussion: https://postgr.es/m/CAAV6ZkQRCVBh8qAY+SZiHnz+U+FqAGBBDaDTjF2yiKa2nJSLKg@mail.gmail.com
2021-04-13 19:37:07 +02:00
&op->d.fieldselect.rowcache, NULL);
Make plpgsql use its DTYPE_REC code paths for composite-type variables. Formerly, DTYPE_REC was used only for variables declared as "record"; variables of named composite types used DTYPE_ROW, which is faster for some purposes but much less flexible. In particular, the ROW code paths are entirely incapable of dealing with DDL-caused changes to the number or data types of the columns of a row variable, once a particular plpgsql function has been parsed for the first time in a session. And, since the stored representation of a ROW isn't a tuple, there wasn't any easy way to deal with variables of domain-over-composite types, since the domain constraint checking code would expect the value to be checked to be a tuple. A lesser, but still real, annoyance is that ROW format cannot represent a true NULL composite value, only a row of per-field NULL values, which is not exactly the same thing. Hence, switch to using DTYPE_REC for all composite-typed variables, whether "record", named composite type, or domain over named composite type. DTYPE_ROW remains but is used only for its native purpose, to represent a fixed-at-compile-time list of variables, for instance the targets of an INTO clause. To accomplish this without taking significant performance losses, introduce infrastructure that allows storing composite-type variables as "expanded objects", similar to the "expanded array" infrastructure introduced in commit 1dc5ebc90. A composite variable's value is thereby kept (most of the time) in the form of separate Datums, so that field accesses and updates are not much more expensive than they were in the ROW format. This holds the line, more or less, on performance of variables of named composite types in field-access-intensive microbenchmarks, and makes variables declared "record" perform much better than before in similar tests. In addition, the logic involved with enforcing composite-domain constraints against updates of individual fields is in the expanded record infrastructure not plpgsql proper, so that it might be reusable for other purposes. In further support of this, introduce a typcache feature for assigning a unique-within-process identifier to each distinct tuple descriptor of interest; in particular, DDL alterations on composite types result in a new identifier for that type. This allows very cheap detection of the need to refresh tupdesc-dependent data. This improves on the "tupDescSeqNo" idea I had in commit 687f096ea: that assigned identifying sequence numbers to successive versions of individual composite types, but the numbers were not unique across different types, nor was there support for assigning numbers to registered record types. In passing, allow plpgsql functions to accept as well as return type "record". There was no good reason for the old restriction, and it was out of step with most of the other PLs. Tom Lane, reviewed by Pavel Stehule Discussion: https://postgr.es/m/8962.1514399547@sss.pgh.pa.us
2018-02-14 00:52:21 +01:00
/*
* Find field's attr record. Note we don't support system columns
* here: a datum tuple doesn't have valid values for most of the
* interesting system columns anyway.
*/
if (fieldnum <= 0) /* should never happen */
elog(ERROR, "unsupported reference to system column %d in FieldSelect",
fieldnum);
if (fieldnum > tupDesc->natts) /* should never happen */
elog(ERROR, "attribute number %d exceeds number of columns %d",
fieldnum, tupDesc->natts);
attr = TupleDescAttr(tupDesc, fieldnum - 1);
/* Check for dropped column, and force a NULL result if so */
if (attr->attisdropped)
{
*op->resnull = true;
return;
}
/* Check for type mismatch --- possible after ALTER COLUMN TYPE? */
/* As in CheckVarSlotCompatibility, we should but can't check typmod */
if (op->d.fieldselect.resulttype != attr->atttypid)
ereport(ERROR,
(errcode(ERRCODE_DATATYPE_MISMATCH),
errmsg("attribute %d has wrong type", fieldnum),
errdetail("Table has type %s, but query expects %s.",
format_type_be(attr->atttypid),
format_type_be(op->d.fieldselect.resulttype))));
/* heap_getattr needs a HeapTuple not a bare HeapTupleHeader */
tmptup.t_len = HeapTupleHeaderGetDatumLength(tuple);
tmptup.t_data = tuple;
/* extract the field */
*op->resvalue = heap_getattr(&tmptup,
fieldnum,
tupDesc,
op->resnull);
}
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
}
/*
* Deform source tuple, filling in the step's values/nulls arrays, before
* evaluating individual new values as part of a FieldStore expression.
* Subsequent steps will overwrite individual elements of the values/nulls
* arrays with the new field values, and then FIELDSTORE_FORM will build the
* new tuple value.
*
* Source record is in step's result variable.
*/
void
ExecEvalFieldStoreDeForm(ExprState *state, ExprEvalStep *op, ExprContext *econtext)
{
TupleDesc tupDesc;
Redesign the caching done by get_cached_rowtype(). Previously, get_cached_rowtype() cached a pointer to a reference-counted tuple descriptor from the typcache, relying on the ExprContextCallback mechanism to release the tupdesc refcount when the expression tree using the tupdesc was destroyed. This worked fine when it was designed, but the introduction of within-DO-block COMMITs broke it. The refcount is logged in a transaction-lifespan resource owner, but plpgsql won't destroy simple expressions made within the DO block (before its first commit) until the DO block is exited. That results in a warning about a leaked tupdesc refcount when the COMMIT destroys the original resource owner, and then an error about the active resource owner not holding a matching refcount when the expression is destroyed. To fix, get rid of the need to have a shutdown callback at all, by instead caching a pointer to the relevant typcache entry. Those survive for the life of the backend, so we needn't worry about the pointer becoming stale. (For registered RECORD types, we can still cache a pointer to the tupdesc, knowing that it won't change for the life of the backend.) This mechanism has been in use in plpgsql and expandedrecord.c since commit 4b93f5799, and seems to work well. This change requires modifying the ExprEvalStep structs used by the relevant expression step types, which is slightly worrisome for back-patching. However, there seems no good reason for extensions to be familiar with the details of these particular sub-structs. Per report from Rohit Bhogate. Back-patch to v11 where within-DO-block COMMITs became a thing. Discussion: https://postgr.es/m/CAAV6ZkQRCVBh8qAY+SZiHnz+U+FqAGBBDaDTjF2yiKa2nJSLKg@mail.gmail.com
2021-04-13 19:37:07 +02:00
/* Lookup tupdesc if first time through or if type changes */
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
tupDesc = get_cached_rowtype(op->d.fieldstore.fstore->resulttype, -1,
Redesign the caching done by get_cached_rowtype(). Previously, get_cached_rowtype() cached a pointer to a reference-counted tuple descriptor from the typcache, relying on the ExprContextCallback mechanism to release the tupdesc refcount when the expression tree using the tupdesc was destroyed. This worked fine when it was designed, but the introduction of within-DO-block COMMITs broke it. The refcount is logged in a transaction-lifespan resource owner, but plpgsql won't destroy simple expressions made within the DO block (before its first commit) until the DO block is exited. That results in a warning about a leaked tupdesc refcount when the COMMIT destroys the original resource owner, and then an error about the active resource owner not holding a matching refcount when the expression is destroyed. To fix, get rid of the need to have a shutdown callback at all, by instead caching a pointer to the relevant typcache entry. Those survive for the life of the backend, so we needn't worry about the pointer becoming stale. (For registered RECORD types, we can still cache a pointer to the tupdesc, knowing that it won't change for the life of the backend.) This mechanism has been in use in plpgsql and expandedrecord.c since commit 4b93f5799, and seems to work well. This change requires modifying the ExprEvalStep structs used by the relevant expression step types, which is slightly worrisome for back-patching. However, there seems no good reason for extensions to be familiar with the details of these particular sub-structs. Per report from Rohit Bhogate. Back-patch to v11 where within-DO-block COMMITs became a thing. Discussion: https://postgr.es/m/CAAV6ZkQRCVBh8qAY+SZiHnz+U+FqAGBBDaDTjF2yiKa2nJSLKg@mail.gmail.com
2021-04-13 19:37:07 +02:00
op->d.fieldstore.rowcache, NULL);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
/* Check that current tupdesc doesn't have more fields than we allocated */
if (unlikely(tupDesc->natts > op->d.fieldstore.ncolumns))
elog(ERROR, "too many columns in composite type %u",
op->d.fieldstore.fstore->resulttype);
if (*op->resnull)
{
/* Convert null input tuple into an all-nulls row */
memset(op->d.fieldstore.nulls, true,
op->d.fieldstore.ncolumns * sizeof(bool));
}
else
{
/*
* heap_deform_tuple needs a HeapTuple not a bare HeapTupleHeader. We
* set all the fields in the struct just in case.
*/
Datum tupDatum = *op->resvalue;
HeapTupleHeader tuphdr;
HeapTupleData tmptup;
tuphdr = DatumGetHeapTupleHeader(tupDatum);
tmptup.t_len = HeapTupleHeaderGetDatumLength(tuphdr);
ItemPointerSetInvalid(&(tmptup.t_self));
tmptup.t_tableOid = InvalidOid;
tmptup.t_data = tuphdr;
heap_deform_tuple(&tmptup, tupDesc,
op->d.fieldstore.values,
op->d.fieldstore.nulls);
}
}
/*
* Compute the new composite datum after each individual field value of a
* FieldStore expression has been evaluated.
*/
void
ExecEvalFieldStoreForm(ExprState *state, ExprEvalStep *op, ExprContext *econtext)
{
Redesign the caching done by get_cached_rowtype(). Previously, get_cached_rowtype() cached a pointer to a reference-counted tuple descriptor from the typcache, relying on the ExprContextCallback mechanism to release the tupdesc refcount when the expression tree using the tupdesc was destroyed. This worked fine when it was designed, but the introduction of within-DO-block COMMITs broke it. The refcount is logged in a transaction-lifespan resource owner, but plpgsql won't destroy simple expressions made within the DO block (before its first commit) until the DO block is exited. That results in a warning about a leaked tupdesc refcount when the COMMIT destroys the original resource owner, and then an error about the active resource owner not holding a matching refcount when the expression is destroyed. To fix, get rid of the need to have a shutdown callback at all, by instead caching a pointer to the relevant typcache entry. Those survive for the life of the backend, so we needn't worry about the pointer becoming stale. (For registered RECORD types, we can still cache a pointer to the tupdesc, knowing that it won't change for the life of the backend.) This mechanism has been in use in plpgsql and expandedrecord.c since commit 4b93f5799, and seems to work well. This change requires modifying the ExprEvalStep structs used by the relevant expression step types, which is slightly worrisome for back-patching. However, there seems no good reason for extensions to be familiar with the details of these particular sub-structs. Per report from Rohit Bhogate. Back-patch to v11 where within-DO-block COMMITs became a thing. Discussion: https://postgr.es/m/CAAV6ZkQRCVBh8qAY+SZiHnz+U+FqAGBBDaDTjF2yiKa2nJSLKg@mail.gmail.com
2021-04-13 19:37:07 +02:00
TupleDesc tupDesc;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
HeapTuple tuple;
Redesign the caching done by get_cached_rowtype(). Previously, get_cached_rowtype() cached a pointer to a reference-counted tuple descriptor from the typcache, relying on the ExprContextCallback mechanism to release the tupdesc refcount when the expression tree using the tupdesc was destroyed. This worked fine when it was designed, but the introduction of within-DO-block COMMITs broke it. The refcount is logged in a transaction-lifespan resource owner, but plpgsql won't destroy simple expressions made within the DO block (before its first commit) until the DO block is exited. That results in a warning about a leaked tupdesc refcount when the COMMIT destroys the original resource owner, and then an error about the active resource owner not holding a matching refcount when the expression is destroyed. To fix, get rid of the need to have a shutdown callback at all, by instead caching a pointer to the relevant typcache entry. Those survive for the life of the backend, so we needn't worry about the pointer becoming stale. (For registered RECORD types, we can still cache a pointer to the tupdesc, knowing that it won't change for the life of the backend.) This mechanism has been in use in plpgsql and expandedrecord.c since commit 4b93f5799, and seems to work well. This change requires modifying the ExprEvalStep structs used by the relevant expression step types, which is slightly worrisome for back-patching. However, there seems no good reason for extensions to be familiar with the details of these particular sub-structs. Per report from Rohit Bhogate. Back-patch to v11 where within-DO-block COMMITs became a thing. Discussion: https://postgr.es/m/CAAV6ZkQRCVBh8qAY+SZiHnz+U+FqAGBBDaDTjF2yiKa2nJSLKg@mail.gmail.com
2021-04-13 19:37:07 +02:00
/* Lookup tupdesc (should be valid already) */
tupDesc = get_cached_rowtype(op->d.fieldstore.fstore->resulttype, -1,
op->d.fieldstore.rowcache, NULL);
tuple = heap_form_tuple(tupDesc,
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
op->d.fieldstore.values,
op->d.fieldstore.nulls);
*op->resvalue = HeapTupleGetDatum(tuple);
*op->resnull = false;
}
/*
* Evaluate a rowtype coercion operation.
* This may require rearranging field positions.
*
* Source record is in step's result variable.
*/
void
ExecEvalConvertRowtype(ExprState *state, ExprEvalStep *op, ExprContext *econtext)
{
HeapTuple result;
Datum tupDatum;
HeapTupleHeader tuple;
HeapTupleData tmptup;
TupleDesc indesc,
outdesc;
Redesign the caching done by get_cached_rowtype(). Previously, get_cached_rowtype() cached a pointer to a reference-counted tuple descriptor from the typcache, relying on the ExprContextCallback mechanism to release the tupdesc refcount when the expression tree using the tupdesc was destroyed. This worked fine when it was designed, but the introduction of within-DO-block COMMITs broke it. The refcount is logged in a transaction-lifespan resource owner, but plpgsql won't destroy simple expressions made within the DO block (before its first commit) until the DO block is exited. That results in a warning about a leaked tupdesc refcount when the COMMIT destroys the original resource owner, and then an error about the active resource owner not holding a matching refcount when the expression is destroyed. To fix, get rid of the need to have a shutdown callback at all, by instead caching a pointer to the relevant typcache entry. Those survive for the life of the backend, so we needn't worry about the pointer becoming stale. (For registered RECORD types, we can still cache a pointer to the tupdesc, knowing that it won't change for the life of the backend.) This mechanism has been in use in plpgsql and expandedrecord.c since commit 4b93f5799, and seems to work well. This change requires modifying the ExprEvalStep structs used by the relevant expression step types, which is slightly worrisome for back-patching. However, there seems no good reason for extensions to be familiar with the details of these particular sub-structs. Per report from Rohit Bhogate. Back-patch to v11 where within-DO-block COMMITs became a thing. Discussion: https://postgr.es/m/CAAV6ZkQRCVBh8qAY+SZiHnz+U+FqAGBBDaDTjF2yiKa2nJSLKg@mail.gmail.com
2021-04-13 19:37:07 +02:00
bool changed = false;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
/* NULL in -> NULL out */
if (*op->resnull)
return;
tupDatum = *op->resvalue;
tuple = DatumGetHeapTupleHeader(tupDatum);
Redesign the caching done by get_cached_rowtype(). Previously, get_cached_rowtype() cached a pointer to a reference-counted tuple descriptor from the typcache, relying on the ExprContextCallback mechanism to release the tupdesc refcount when the expression tree using the tupdesc was destroyed. This worked fine when it was designed, but the introduction of within-DO-block COMMITs broke it. The refcount is logged in a transaction-lifespan resource owner, but plpgsql won't destroy simple expressions made within the DO block (before its first commit) until the DO block is exited. That results in a warning about a leaked tupdesc refcount when the COMMIT destroys the original resource owner, and then an error about the active resource owner not holding a matching refcount when the expression is destroyed. To fix, get rid of the need to have a shutdown callback at all, by instead caching a pointer to the relevant typcache entry. Those survive for the life of the backend, so we needn't worry about the pointer becoming stale. (For registered RECORD types, we can still cache a pointer to the tupdesc, knowing that it won't change for the life of the backend.) This mechanism has been in use in plpgsql and expandedrecord.c since commit 4b93f5799, and seems to work well. This change requires modifying the ExprEvalStep structs used by the relevant expression step types, which is slightly worrisome for back-patching. However, there seems no good reason for extensions to be familiar with the details of these particular sub-structs. Per report from Rohit Bhogate. Back-patch to v11 where within-DO-block COMMITs became a thing. Discussion: https://postgr.es/m/CAAV6ZkQRCVBh8qAY+SZiHnz+U+FqAGBBDaDTjF2yiKa2nJSLKg@mail.gmail.com
2021-04-13 19:37:07 +02:00
/*
* Lookup tupdescs if first time through or if type changes. We'd better
* pin them since type conversion functions could do catalog lookups and
* hence cause cache invalidation.
*/
indesc = get_cached_rowtype(op->d.convert_rowtype.inputtype, -1,
op->d.convert_rowtype.incache,
&changed);
IncrTupleDescRefCount(indesc);
outdesc = get_cached_rowtype(op->d.convert_rowtype.outputtype, -1,
op->d.convert_rowtype.outcache,
&changed);
IncrTupleDescRefCount(outdesc);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
/*
* We used to be able to assert that incoming tuples are marked with
* exactly the rowtype of indesc. However, now that ExecEvalWholeRowVar
* might change the tuples' marking to plain RECORD due to inserting
* aliases, we can only make this weak test:
*/
Assert(HeapTupleHeaderGetTypeId(tuple) == indesc->tdtypeid ||
HeapTupleHeaderGetTypeId(tuple) == RECORDOID);
Redesign the caching done by get_cached_rowtype(). Previously, get_cached_rowtype() cached a pointer to a reference-counted tuple descriptor from the typcache, relying on the ExprContextCallback mechanism to release the tupdesc refcount when the expression tree using the tupdesc was destroyed. This worked fine when it was designed, but the introduction of within-DO-block COMMITs broke it. The refcount is logged in a transaction-lifespan resource owner, but plpgsql won't destroy simple expressions made within the DO block (before its first commit) until the DO block is exited. That results in a warning about a leaked tupdesc refcount when the COMMIT destroys the original resource owner, and then an error about the active resource owner not holding a matching refcount when the expression is destroyed. To fix, get rid of the need to have a shutdown callback at all, by instead caching a pointer to the relevant typcache entry. Those survive for the life of the backend, so we needn't worry about the pointer becoming stale. (For registered RECORD types, we can still cache a pointer to the tupdesc, knowing that it won't change for the life of the backend.) This mechanism has been in use in plpgsql and expandedrecord.c since commit 4b93f5799, and seems to work well. This change requires modifying the ExprEvalStep structs used by the relevant expression step types, which is slightly worrisome for back-patching. However, there seems no good reason for extensions to be familiar with the details of these particular sub-structs. Per report from Rohit Bhogate. Back-patch to v11 where within-DO-block COMMITs became a thing. Discussion: https://postgr.es/m/CAAV6ZkQRCVBh8qAY+SZiHnz+U+FqAGBBDaDTjF2yiKa2nJSLKg@mail.gmail.com
2021-04-13 19:37:07 +02:00
/* if first time through, or after change, initialize conversion map */
if (changed)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
MemoryContext old_cxt;
/* allocate map in long-lived memory context */
old_cxt = MemoryContextSwitchTo(econtext->ecxt_per_query_memory);
/* prepare map from old to new attribute numbers */
Remove 'msg' parameter from convert_tuples_by_name The message was included as a parameter when this function was added in dcb2bda9b704, but I don't think it has ever served any useful purpose. Let's stop spreading it pointlessly. Reviewed by Amit Langote and Peter Eisentraut. Discussion: https://postgr.es/m/20190806224728.GA17233@alvherre.pgsql
2019-09-03 20:47:29 +02:00
op->d.convert_rowtype.map = convert_tuples_by_name(indesc, outdesc);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
MemoryContextSwitchTo(old_cxt);
}
Clean up after insufficiently-researched optimization of tuple conversions. tupconvert.c's functions formerly considered that an explicit tuple conversion was necessary if the input and output tupdescs contained different type OIDs. The point of that was to make sure that a composite datum resulting from the conversion would contain the destination rowtype OID in its composite-datum header. However, commit 3838074f8 entirely misunderstood what that check was for, thinking that it had something to do with presence or absence of an OID column within the tuple. Removal of the check broke the no-op conversion path in ExecEvalConvertRowtype, as reported by Ashutosh Bapat. It turns out that of the dozen or so call sites for tupconvert.c functions, ExecEvalConvertRowtype is the only one that cares about the composite-datum header fields in the output tuple. In all the rest, we'd much rather avoid an unnecessary conversion whenever the tuples are physically compatible. Moreover, the comments in tupconvert.c only promise physical compatibility not a metadata match. So, let's accept the removal of the guarantee about the output tuple's rowtype marking, recognizing that this is a API change that could conceivably break third-party callers of tupconvert.c. (So, let's remember to mention it in the v10 release notes.) However, commit 3838074f8 did have a bit of a point here, in that two tuples mustn't be considered physically compatible if one has HEAP_HASOID set and the other doesn't. (Some of the callers of tupconvert.c might not really care about that, but we can't assume it in general.) The previous check accidentally covered that issue, because no RECORD types ever have OIDs, while if two tupdescs have the same named composite type OID then, a fortiori, they have the same tdhasoid setting. If we're removing the type OID match check then we'd better include tdhasoid match as part of the physical compatibility check. Without that hack in tupconvert.c, we need ExecEvalConvertRowtype to take responsibility for inserting the correct rowtype OID label whenever tupconvert.c decides it need not do anything. This is easily done with heap_copy_tuple_as_datum, which will be considerably faster than a tuple disassembly and reassembly anyway; so from a performance standpoint this change is a win all around compared to what happened in earlier branches. It just means a couple more lines of code in ExecEvalConvertRowtype. Ashutosh Bapat and Tom Lane Discussion: https://postgr.es/m/CAFjFpRfvHABV6+oVvGcshF8rHn+1LfRUhj7Jz1CDZ4gPUwehBg@mail.gmail.com
2017-04-07 03:10:09 +02:00
/* Following steps need a HeapTuple not a bare HeapTupleHeader */
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
tmptup.t_len = HeapTupleHeaderGetDatumLength(tuple);
tmptup.t_data = tuple;
Clean up after insufficiently-researched optimization of tuple conversions. tupconvert.c's functions formerly considered that an explicit tuple conversion was necessary if the input and output tupdescs contained different type OIDs. The point of that was to make sure that a composite datum resulting from the conversion would contain the destination rowtype OID in its composite-datum header. However, commit 3838074f8 entirely misunderstood what that check was for, thinking that it had something to do with presence or absence of an OID column within the tuple. Removal of the check broke the no-op conversion path in ExecEvalConvertRowtype, as reported by Ashutosh Bapat. It turns out that of the dozen or so call sites for tupconvert.c functions, ExecEvalConvertRowtype is the only one that cares about the composite-datum header fields in the output tuple. In all the rest, we'd much rather avoid an unnecessary conversion whenever the tuples are physically compatible. Moreover, the comments in tupconvert.c only promise physical compatibility not a metadata match. So, let's accept the removal of the guarantee about the output tuple's rowtype marking, recognizing that this is a API change that could conceivably break third-party callers of tupconvert.c. (So, let's remember to mention it in the v10 release notes.) However, commit 3838074f8 did have a bit of a point here, in that two tuples mustn't be considered physically compatible if one has HEAP_HASOID set and the other doesn't. (Some of the callers of tupconvert.c might not really care about that, but we can't assume it in general.) The previous check accidentally covered that issue, because no RECORD types ever have OIDs, while if two tupdescs have the same named composite type OID then, a fortiori, they have the same tdhasoid setting. If we're removing the type OID match check then we'd better include tdhasoid match as part of the physical compatibility check. Without that hack in tupconvert.c, we need ExecEvalConvertRowtype to take responsibility for inserting the correct rowtype OID label whenever tupconvert.c decides it need not do anything. This is easily done with heap_copy_tuple_as_datum, which will be considerably faster than a tuple disassembly and reassembly anyway; so from a performance standpoint this change is a win all around compared to what happened in earlier branches. It just means a couple more lines of code in ExecEvalConvertRowtype. Ashutosh Bapat and Tom Lane Discussion: https://postgr.es/m/CAFjFpRfvHABV6+oVvGcshF8rHn+1LfRUhj7Jz1CDZ4gPUwehBg@mail.gmail.com
2017-04-07 03:10:09 +02:00
if (op->d.convert_rowtype.map != NULL)
{
/* Full conversion with attribute rearrangement needed */
Use slots more widely in tuple mapping code and make naming more consistent. It's inefficient to use a single slot for mapping between tuple descriptors for multiple tuples, as previously done when using ConvertPartitionTupleSlot(), as that means the slot's tuple descriptors change for every tuple. Previously we also, via ConvertPartitionTupleSlot(), built new tuples after the mapping even in cases where we, immediately afterwards, access individual columns again. Refactor the code so one slot, on demand, is used for each partition. That avoids having to change the descriptor (and allows to use the more efficient "fixed" tuple slots). Then use slot->slot mapping, to avoid unnecessarily forming a tuple. As the naming between the tuple and slot mapping functions wasn't consistent, rename them to execute_attr_map_{tuple,slot}. It's likely that we'll also rename convert_tuples_by_* to denote that these functions "only" build a map, but that's left for later. Author: Amit Khandekar and Amit Langote, editorialized by me Reviewed-By: Amit Langote, Amit Khandekar, Andres Freund Discussion: https://postgr.es/m/CAJ3gD9fR0wRNeAE8VqffNTyONS_UfFPRpqxhnD9Q42vZB+Jvpg@mail.gmail.com https://postgr.es/m/e4f9d743-cd4b-efb0-7574-da21d86a7f36%40lab.ntt.co.jp Backpatch: -
2018-10-02 20:14:26 +02:00
result = execute_attr_map_tuple(&tmptup, op->d.convert_rowtype.map);
Clean up after insufficiently-researched optimization of tuple conversions. tupconvert.c's functions formerly considered that an explicit tuple conversion was necessary if the input and output tupdescs contained different type OIDs. The point of that was to make sure that a composite datum resulting from the conversion would contain the destination rowtype OID in its composite-datum header. However, commit 3838074f8 entirely misunderstood what that check was for, thinking that it had something to do with presence or absence of an OID column within the tuple. Removal of the check broke the no-op conversion path in ExecEvalConvertRowtype, as reported by Ashutosh Bapat. It turns out that of the dozen or so call sites for tupconvert.c functions, ExecEvalConvertRowtype is the only one that cares about the composite-datum header fields in the output tuple. In all the rest, we'd much rather avoid an unnecessary conversion whenever the tuples are physically compatible. Moreover, the comments in tupconvert.c only promise physical compatibility not a metadata match. So, let's accept the removal of the guarantee about the output tuple's rowtype marking, recognizing that this is a API change that could conceivably break third-party callers of tupconvert.c. (So, let's remember to mention it in the v10 release notes.) However, commit 3838074f8 did have a bit of a point here, in that two tuples mustn't be considered physically compatible if one has HEAP_HASOID set and the other doesn't. (Some of the callers of tupconvert.c might not really care about that, but we can't assume it in general.) The previous check accidentally covered that issue, because no RECORD types ever have OIDs, while if two tupdescs have the same named composite type OID then, a fortiori, they have the same tdhasoid setting. If we're removing the type OID match check then we'd better include tdhasoid match as part of the physical compatibility check. Without that hack in tupconvert.c, we need ExecEvalConvertRowtype to take responsibility for inserting the correct rowtype OID label whenever tupconvert.c decides it need not do anything. This is easily done with heap_copy_tuple_as_datum, which will be considerably faster than a tuple disassembly and reassembly anyway; so from a performance standpoint this change is a win all around compared to what happened in earlier branches. It just means a couple more lines of code in ExecEvalConvertRowtype. Ashutosh Bapat and Tom Lane Discussion: https://postgr.es/m/CAFjFpRfvHABV6+oVvGcshF8rHn+1LfRUhj7Jz1CDZ4gPUwehBg@mail.gmail.com
2017-04-07 03:10:09 +02:00
/* Result already has appropriate composite-datum header fields */
*op->resvalue = HeapTupleGetDatum(result);
}
else
{
/*
* The tuple is physically compatible as-is, but we need to insert the
* destination rowtype OID in its composite-datum header field, so we
* have to copy it anyway. heap_copy_tuple_as_datum() is convenient
* for this since it will both make the physical copy and insert the
* correct composite header fields. Note that we aren't expecting to
* have to flatten any toasted fields: the input was a composite
* datum, so it shouldn't contain any. So heap_copy_tuple_as_datum()
* is overkill here, but its check for external fields is cheap.
*/
*op->resvalue = heap_copy_tuple_as_datum(&tmptup, outdesc);
}
Redesign the caching done by get_cached_rowtype(). Previously, get_cached_rowtype() cached a pointer to a reference-counted tuple descriptor from the typcache, relying on the ExprContextCallback mechanism to release the tupdesc refcount when the expression tree using the tupdesc was destroyed. This worked fine when it was designed, but the introduction of within-DO-block COMMITs broke it. The refcount is logged in a transaction-lifespan resource owner, but plpgsql won't destroy simple expressions made within the DO block (before its first commit) until the DO block is exited. That results in a warning about a leaked tupdesc refcount when the COMMIT destroys the original resource owner, and then an error about the active resource owner not holding a matching refcount when the expression is destroyed. To fix, get rid of the need to have a shutdown callback at all, by instead caching a pointer to the relevant typcache entry. Those survive for the life of the backend, so we needn't worry about the pointer becoming stale. (For registered RECORD types, we can still cache a pointer to the tupdesc, knowing that it won't change for the life of the backend.) This mechanism has been in use in plpgsql and expandedrecord.c since commit 4b93f5799, and seems to work well. This change requires modifying the ExprEvalStep structs used by the relevant expression step types, which is slightly worrisome for back-patching. However, there seems no good reason for extensions to be familiar with the details of these particular sub-structs. Per report from Rohit Bhogate. Back-patch to v11 where within-DO-block COMMITs became a thing. Discussion: https://postgr.es/m/CAAV6ZkQRCVBh8qAY+SZiHnz+U+FqAGBBDaDTjF2yiKa2nJSLKg@mail.gmail.com
2021-04-13 19:37:07 +02:00
DecrTupleDescRefCount(indesc);
DecrTupleDescRefCount(outdesc);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
}
/*
* Evaluate "scalar op ANY/ALL (array)".
*
* Source array is in our result area, scalar arg is already evaluated into
Change function call information to be variable length. Before this change FunctionCallInfoData, the struct arguments etc for V1 function calls are stored in, always had space for FUNC_MAX_ARGS/100 arguments, storing datums and their nullness in two arrays. For nearly every function call 100 arguments is far more than needed, therefore wasting memory. Arg and argnull being two separate arrays also guarantees that to access a single argument, two cachelines have to be touched. Change the layout so there's a single variable-length array with pairs of value / isnull. That drastically reduces memory consumption for most function calls (on x86-64 a two argument function now uses 64bytes, previously 936 bytes), and makes it very likely that argument value and its nullness are on the same cacheline. Arguments are stored in a new NullableDatum struct, which, due to padding, needs more memory per argument than before. But as usually far fewer arguments are stored, and individual arguments are cheaper to access, that's still a clear win. It's likely that there's other places where conversion to NullableDatum arrays would make sense, e.g. TupleTableSlots, but that's for another commit. Because the function call information is now variable-length allocations have to take the number of arguments into account. For heap allocations that can be done with SizeForFunctionCallInfoData(), for on-stack allocations there's a new LOCAL_FCINFO(name, nargs) macro that helps to allocate an appropriately sized and aligned variable. Some places with stack allocation function call information don't know the number of arguments at compile time, and currently variably sized stack allocations aren't allowed in postgres. Therefore allow for FUNC_MAX_ARGS space in these cases. They're not that common, so for now that seems acceptable. Because of the need to allocate FunctionCallInfo of the appropriate size, older extensions may need to update their code. To avoid subtle breakages, the FunctionCallInfoData struct has been renamed to FunctionCallInfoBaseData. Most code only references FunctionCallInfo, so that shouldn't cause much collateral damage. This change is also a prerequisite for more efficient expression JIT compilation (by allocating the function call information on the stack, allowing LLVM to optimize it away); previously the size of the call information caused problems inside LLVM's optimizer. Author: Andres Freund Reviewed-By: Tom Lane Discussion: https://postgr.es/m/20180605172952.x34m5uz6ju6enaem@alap3.anarazel.de
2019-01-26 23:17:52 +01:00
* fcinfo->args[0].
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
*
* The operator always yields boolean, and we combine the results across all
* array elements using OR and AND (for ANY and ALL respectively). Of course
* we short-circuit as soon as the result is known.
*/
void
ExecEvalScalarArrayOp(ExprState *state, ExprEvalStep *op)
{
FunctionCallInfo fcinfo = op->d.scalararrayop.fcinfo_data;
bool useOr = op->d.scalararrayop.useOr;
bool strictfunc = op->d.scalararrayop.finfo->fn_strict;
ArrayType *arr;
int nitems;
Datum result;
bool resultnull;
int16 typlen;
bool typbyval;
char typalign;
char *s;
bits8 *bitmap;
int bitmask;
/*
* If the array is NULL then we return NULL --- it's not very meaningful
* to do anything else, even if the operator isn't strict.
*/
if (*op->resnull)
return;
/* Else okay to fetch and detoast the array */
arr = DatumGetArrayTypeP(*op->resvalue);
/*
* If the array is empty, we return either FALSE or TRUE per the useOr
* flag. This is correct even if the scalar is NULL; since we would
* evaluate the operator zero times, it matters not whether it would want
* to return NULL.
*/
nitems = ArrayGetNItems(ARR_NDIM(arr), ARR_DIMS(arr));
if (nitems <= 0)
{
*op->resvalue = BoolGetDatum(!useOr);
*op->resnull = false;
return;
}
/*
* If the scalar is NULL, and the function is strict, return NULL; no
* point in iterating the loop.
*/
Change function call information to be variable length. Before this change FunctionCallInfoData, the struct arguments etc for V1 function calls are stored in, always had space for FUNC_MAX_ARGS/100 arguments, storing datums and their nullness in two arrays. For nearly every function call 100 arguments is far more than needed, therefore wasting memory. Arg and argnull being two separate arrays also guarantees that to access a single argument, two cachelines have to be touched. Change the layout so there's a single variable-length array with pairs of value / isnull. That drastically reduces memory consumption for most function calls (on x86-64 a two argument function now uses 64bytes, previously 936 bytes), and makes it very likely that argument value and its nullness are on the same cacheline. Arguments are stored in a new NullableDatum struct, which, due to padding, needs more memory per argument than before. But as usually far fewer arguments are stored, and individual arguments are cheaper to access, that's still a clear win. It's likely that there's other places where conversion to NullableDatum arrays would make sense, e.g. TupleTableSlots, but that's for another commit. Because the function call information is now variable-length allocations have to take the number of arguments into account. For heap allocations that can be done with SizeForFunctionCallInfoData(), for on-stack allocations there's a new LOCAL_FCINFO(name, nargs) macro that helps to allocate an appropriately sized and aligned variable. Some places with stack allocation function call information don't know the number of arguments at compile time, and currently variably sized stack allocations aren't allowed in postgres. Therefore allow for FUNC_MAX_ARGS space in these cases. They're not that common, so for now that seems acceptable. Because of the need to allocate FunctionCallInfo of the appropriate size, older extensions may need to update their code. To avoid subtle breakages, the FunctionCallInfoData struct has been renamed to FunctionCallInfoBaseData. Most code only references FunctionCallInfo, so that shouldn't cause much collateral damage. This change is also a prerequisite for more efficient expression JIT compilation (by allocating the function call information on the stack, allowing LLVM to optimize it away); previously the size of the call information caused problems inside LLVM's optimizer. Author: Andres Freund Reviewed-By: Tom Lane Discussion: https://postgr.es/m/20180605172952.x34m5uz6ju6enaem@alap3.anarazel.de
2019-01-26 23:17:52 +01:00
if (fcinfo->args[0].isnull && strictfunc)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
*op->resnull = true;
return;
}
/*
* We arrange to look up info about the element type only once per series
* of calls, assuming the element type doesn't change underneath us.
*/
if (op->d.scalararrayop.element_type != ARR_ELEMTYPE(arr))
{
get_typlenbyvalalign(ARR_ELEMTYPE(arr),
&op->d.scalararrayop.typlen,
&op->d.scalararrayop.typbyval,
&op->d.scalararrayop.typalign);
op->d.scalararrayop.element_type = ARR_ELEMTYPE(arr);
}
typlen = op->d.scalararrayop.typlen;
typbyval = op->d.scalararrayop.typbyval;
typalign = op->d.scalararrayop.typalign;
/* Initialize result appropriately depending on useOr */
result = BoolGetDatum(!useOr);
resultnull = false;
/* Loop over the array elements */
s = (char *) ARR_DATA_PTR(arr);
bitmap = ARR_NULLBITMAP(arr);
bitmask = 1;
expression eval, jit: Minor code cleanups. This mostly consists of using C99 style for loops, moving variables into narrower scopes, and a smattering of other minor improvements. Done separately to make it easier to review patches with actual functional changes. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:04:50 +01:00
for (int i = 0; i < nitems; i++)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
Datum elt;
Datum thisresult;
/* Get array element, checking for NULL */
if (bitmap && (*bitmap & bitmask) == 0)
{
Change function call information to be variable length. Before this change FunctionCallInfoData, the struct arguments etc for V1 function calls are stored in, always had space for FUNC_MAX_ARGS/100 arguments, storing datums and their nullness in two arrays. For nearly every function call 100 arguments is far more than needed, therefore wasting memory. Arg and argnull being two separate arrays also guarantees that to access a single argument, two cachelines have to be touched. Change the layout so there's a single variable-length array with pairs of value / isnull. That drastically reduces memory consumption for most function calls (on x86-64 a two argument function now uses 64bytes, previously 936 bytes), and makes it very likely that argument value and its nullness are on the same cacheline. Arguments are stored in a new NullableDatum struct, which, due to padding, needs more memory per argument than before. But as usually far fewer arguments are stored, and individual arguments are cheaper to access, that's still a clear win. It's likely that there's other places where conversion to NullableDatum arrays would make sense, e.g. TupleTableSlots, but that's for another commit. Because the function call information is now variable-length allocations have to take the number of arguments into account. For heap allocations that can be done with SizeForFunctionCallInfoData(), for on-stack allocations there's a new LOCAL_FCINFO(name, nargs) macro that helps to allocate an appropriately sized and aligned variable. Some places with stack allocation function call information don't know the number of arguments at compile time, and currently variably sized stack allocations aren't allowed in postgres. Therefore allow for FUNC_MAX_ARGS space in these cases. They're not that common, so for now that seems acceptable. Because of the need to allocate FunctionCallInfo of the appropriate size, older extensions may need to update their code. To avoid subtle breakages, the FunctionCallInfoData struct has been renamed to FunctionCallInfoBaseData. Most code only references FunctionCallInfo, so that shouldn't cause much collateral damage. This change is also a prerequisite for more efficient expression JIT compilation (by allocating the function call information on the stack, allowing LLVM to optimize it away); previously the size of the call information caused problems inside LLVM's optimizer. Author: Andres Freund Reviewed-By: Tom Lane Discussion: https://postgr.es/m/20180605172952.x34m5uz6ju6enaem@alap3.anarazel.de
2019-01-26 23:17:52 +01:00
fcinfo->args[1].value = (Datum) 0;
fcinfo->args[1].isnull = true;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
}
else
{
elt = fetch_att(s, typbyval, typlen);
s = att_addlength_pointer(s, typlen, s);
s = (char *) att_align_nominal(s, typalign);
Change function call information to be variable length. Before this change FunctionCallInfoData, the struct arguments etc for V1 function calls are stored in, always had space for FUNC_MAX_ARGS/100 arguments, storing datums and their nullness in two arrays. For nearly every function call 100 arguments is far more than needed, therefore wasting memory. Arg and argnull being two separate arrays also guarantees that to access a single argument, two cachelines have to be touched. Change the layout so there's a single variable-length array with pairs of value / isnull. That drastically reduces memory consumption for most function calls (on x86-64 a two argument function now uses 64bytes, previously 936 bytes), and makes it very likely that argument value and its nullness are on the same cacheline. Arguments are stored in a new NullableDatum struct, which, due to padding, needs more memory per argument than before. But as usually far fewer arguments are stored, and individual arguments are cheaper to access, that's still a clear win. It's likely that there's other places where conversion to NullableDatum arrays would make sense, e.g. TupleTableSlots, but that's for another commit. Because the function call information is now variable-length allocations have to take the number of arguments into account. For heap allocations that can be done with SizeForFunctionCallInfoData(), for on-stack allocations there's a new LOCAL_FCINFO(name, nargs) macro that helps to allocate an appropriately sized and aligned variable. Some places with stack allocation function call information don't know the number of arguments at compile time, and currently variably sized stack allocations aren't allowed in postgres. Therefore allow for FUNC_MAX_ARGS space in these cases. They're not that common, so for now that seems acceptable. Because of the need to allocate FunctionCallInfo of the appropriate size, older extensions may need to update their code. To avoid subtle breakages, the FunctionCallInfoData struct has been renamed to FunctionCallInfoBaseData. Most code only references FunctionCallInfo, so that shouldn't cause much collateral damage. This change is also a prerequisite for more efficient expression JIT compilation (by allocating the function call information on the stack, allowing LLVM to optimize it away); previously the size of the call information caused problems inside LLVM's optimizer. Author: Andres Freund Reviewed-By: Tom Lane Discussion: https://postgr.es/m/20180605172952.x34m5uz6ju6enaem@alap3.anarazel.de
2019-01-26 23:17:52 +01:00
fcinfo->args[1].value = elt;
fcinfo->args[1].isnull = false;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
}
/* Call comparison function */
Change function call information to be variable length. Before this change FunctionCallInfoData, the struct arguments etc for V1 function calls are stored in, always had space for FUNC_MAX_ARGS/100 arguments, storing datums and their nullness in two arrays. For nearly every function call 100 arguments is far more than needed, therefore wasting memory. Arg and argnull being two separate arrays also guarantees that to access a single argument, two cachelines have to be touched. Change the layout so there's a single variable-length array with pairs of value / isnull. That drastically reduces memory consumption for most function calls (on x86-64 a two argument function now uses 64bytes, previously 936 bytes), and makes it very likely that argument value and its nullness are on the same cacheline. Arguments are stored in a new NullableDatum struct, which, due to padding, needs more memory per argument than before. But as usually far fewer arguments are stored, and individual arguments are cheaper to access, that's still a clear win. It's likely that there's other places where conversion to NullableDatum arrays would make sense, e.g. TupleTableSlots, but that's for another commit. Because the function call information is now variable-length allocations have to take the number of arguments into account. For heap allocations that can be done with SizeForFunctionCallInfoData(), for on-stack allocations there's a new LOCAL_FCINFO(name, nargs) macro that helps to allocate an appropriately sized and aligned variable. Some places with stack allocation function call information don't know the number of arguments at compile time, and currently variably sized stack allocations aren't allowed in postgres. Therefore allow for FUNC_MAX_ARGS space in these cases. They're not that common, so for now that seems acceptable. Because of the need to allocate FunctionCallInfo of the appropriate size, older extensions may need to update their code. To avoid subtle breakages, the FunctionCallInfoData struct has been renamed to FunctionCallInfoBaseData. Most code only references FunctionCallInfo, so that shouldn't cause much collateral damage. This change is also a prerequisite for more efficient expression JIT compilation (by allocating the function call information on the stack, allowing LLVM to optimize it away); previously the size of the call information caused problems inside LLVM's optimizer. Author: Andres Freund Reviewed-By: Tom Lane Discussion: https://postgr.es/m/20180605172952.x34m5uz6ju6enaem@alap3.anarazel.de
2019-01-26 23:17:52 +01:00
if (fcinfo->args[1].isnull && strictfunc)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
fcinfo->isnull = true;
thisresult = (Datum) 0;
}
else
{
fcinfo->isnull = false;
Reduce excessive dereferencing of function pointers It is equivalent in ANSI C to write (*funcptr) () and funcptr(). These two styles have been applied inconsistently. After discussion, we'll use the more verbose style for plain function pointer variables, to make it clear that it's a variable, and the shorter style when the function pointer is in a struct (s.func() or s->func()), because then it's clear that it's not a plain function name, and otherwise the excessive punctuation makes some of those invocations hard to read. Discussion: https://www.postgresql.org/message-id/f52c16db-14ed-757d-4b48-7ef360b1631d@2ndquadrant.com
2017-09-07 18:06:23 +02:00
thisresult = op->d.scalararrayop.fn_addr(fcinfo);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
}
/* Combine results per OR or AND semantics */
if (fcinfo->isnull)
resultnull = true;
else if (useOr)
{
if (DatumGetBool(thisresult))
{
result = BoolGetDatum(true);
resultnull = false;
break; /* needn't look at any more elements */
}
}
else
{
if (!DatumGetBool(thisresult))
{
result = BoolGetDatum(false);
resultnull = false;
break; /* needn't look at any more elements */
}
}
/* advance bitmap pointer if any */
if (bitmap)
{
bitmask <<= 1;
if (bitmask == 0x100)
{
bitmap++;
bitmask = 1;
}
}
}
*op->resvalue = result;
*op->resnull = resultnull;
}
Speedup ScalarArrayOpExpr evaluation ScalarArrayOpExprs with "useOr=true" and a set of Consts on the righthand side have traditionally been evaluated by using a linear search over the array. When these arrays contain large numbers of elements then this linear search could become a significant part of execution time. Here we add a new method of evaluating ScalarArrayOpExpr expressions to allow them to be evaluated by first building a hash table containing each element, then on subsequent evaluations, we just probe that hash table to determine if there is a match. The planner is in charge of determining when this optimization is possible and it enables it by setting hashfuncid in the ScalarArrayOpExpr. The executor will only perform the hash table evaluation when the hashfuncid is set. This means that not all cases are optimized. For example CHECK constraints containing an IN clause won't go through the planner, so won't get the hashfuncid set. We could maybe do something about that at some later date. The reason we're not doing it now is from fear that we may slow down cases where the expression is evaluated only once. Those cases can be common, for example, a single row INSERT to a table with a CHECK constraint containing an IN clause. In the planner, we enable this when there are suitable hash functions for the ScalarArrayOpExpr's operator and only when there is at least MIN_ARRAY_SIZE_FOR_HASHED_SAOP elements in the array. The threshold is currently set to 9. Author: James Coleman, David Rowley Reviewed-by: David Rowley, Tomas Vondra, Heikki Linnakangas Discussion: https://postgr.es/m/CAAaqYe8x62+=wn0zvNKCj55tPpg-JBHzhZFFc6ANovdqFw7-dA@mail.gmail.com
2021-04-08 13:51:22 +02:00
/*
* Hash function for scalar array hash op elements.
*
* We use the element type's default hash opclass, and the column collation
* if the type is collation-sensitive.
*/
static uint32
saop_element_hash(struct saophash_hash *tb, Datum key)
{
ScalarArrayOpExprHashTable *elements_tab = (ScalarArrayOpExprHashTable *) tb->private_data;
Remove size increase in ExprEvalStep caused by hashed saops 50e17ad28 increased the size of ExprEvalStep from 64 bytes up to 88 bytes. Lots of effort was spent during the development of the current expression evaluation code to make an instance of this struct as small as possible. Making this struct larger than needed reduces CPU cache efficiency during expression evaluation which causes noticeable slowdowns during query execution. In order to reduce the size of the struct, here we remove the fn_addr field. The values from this field can be obtained via fcinfo, just with some extra pointer dereferencing. The extra indirection does not seem to cause any noticeable slowdowns. Various other fields have been moved into the ScalarArrayOpExprHashTable struct. These fields are only used when the ScalarArrayOpExprHashTable pointer has already been dereferenced, so no additional pointer dereferences occur for these. Here we also make hash_fcinfo_data the last field in ScalarArrayOpExprHashTable so that we can avoid a further pointer dereference to get the FunctionCallInfoBaseData. This also saves a call to palloc(). 50e17ad28 was added in 14, but it's too late to adjust the size of the ExprEvalStep in that version, so here we just backpatch to 15, which is currently in beta. Author: Andres Freund, David Rowley Discussion: https://postgr.es/m/20220616233130.rparivafipt6doj3@alap3.anarazel.de Backpatch-through: 15
2022-07-06 09:40:32 +02:00
FunctionCallInfo fcinfo = &elements_tab->hash_fcinfo_data;
Speedup ScalarArrayOpExpr evaluation ScalarArrayOpExprs with "useOr=true" and a set of Consts on the righthand side have traditionally been evaluated by using a linear search over the array. When these arrays contain large numbers of elements then this linear search could become a significant part of execution time. Here we add a new method of evaluating ScalarArrayOpExpr expressions to allow them to be evaluated by first building a hash table containing each element, then on subsequent evaluations, we just probe that hash table to determine if there is a match. The planner is in charge of determining when this optimization is possible and it enables it by setting hashfuncid in the ScalarArrayOpExpr. The executor will only perform the hash table evaluation when the hashfuncid is set. This means that not all cases are optimized. For example CHECK constraints containing an IN clause won't go through the planner, so won't get the hashfuncid set. We could maybe do something about that at some later date. The reason we're not doing it now is from fear that we may slow down cases where the expression is evaluated only once. Those cases can be common, for example, a single row INSERT to a table with a CHECK constraint containing an IN clause. In the planner, we enable this when there are suitable hash functions for the ScalarArrayOpExpr's operator and only when there is at least MIN_ARRAY_SIZE_FOR_HASHED_SAOP elements in the array. The threshold is currently set to 9. Author: James Coleman, David Rowley Reviewed-by: David Rowley, Tomas Vondra, Heikki Linnakangas Discussion: https://postgr.es/m/CAAaqYe8x62+=wn0zvNKCj55tPpg-JBHzhZFFc6ANovdqFw7-dA@mail.gmail.com
2021-04-08 13:51:22 +02:00
Datum hash;
fcinfo->args[0].value = key;
fcinfo->args[0].isnull = false;
Remove size increase in ExprEvalStep caused by hashed saops 50e17ad28 increased the size of ExprEvalStep from 64 bytes up to 88 bytes. Lots of effort was spent during the development of the current expression evaluation code to make an instance of this struct as small as possible. Making this struct larger than needed reduces CPU cache efficiency during expression evaluation which causes noticeable slowdowns during query execution. In order to reduce the size of the struct, here we remove the fn_addr field. The values from this field can be obtained via fcinfo, just with some extra pointer dereferencing. The extra indirection does not seem to cause any noticeable slowdowns. Various other fields have been moved into the ScalarArrayOpExprHashTable struct. These fields are only used when the ScalarArrayOpExprHashTable pointer has already been dereferenced, so no additional pointer dereferences occur for these. Here we also make hash_fcinfo_data the last field in ScalarArrayOpExprHashTable so that we can avoid a further pointer dereference to get the FunctionCallInfoBaseData. This also saves a call to palloc(). 50e17ad28 was added in 14, but it's too late to adjust the size of the ExprEvalStep in that version, so here we just backpatch to 15, which is currently in beta. Author: Andres Freund, David Rowley Discussion: https://postgr.es/m/20220616233130.rparivafipt6doj3@alap3.anarazel.de Backpatch-through: 15
2022-07-06 09:40:32 +02:00
hash = elements_tab->hash_finfo.fn_addr(fcinfo);
Speedup ScalarArrayOpExpr evaluation ScalarArrayOpExprs with "useOr=true" and a set of Consts on the righthand side have traditionally been evaluated by using a linear search over the array. When these arrays contain large numbers of elements then this linear search could become a significant part of execution time. Here we add a new method of evaluating ScalarArrayOpExpr expressions to allow them to be evaluated by first building a hash table containing each element, then on subsequent evaluations, we just probe that hash table to determine if there is a match. The planner is in charge of determining when this optimization is possible and it enables it by setting hashfuncid in the ScalarArrayOpExpr. The executor will only perform the hash table evaluation when the hashfuncid is set. This means that not all cases are optimized. For example CHECK constraints containing an IN clause won't go through the planner, so won't get the hashfuncid set. We could maybe do something about that at some later date. The reason we're not doing it now is from fear that we may slow down cases where the expression is evaluated only once. Those cases can be common, for example, a single row INSERT to a table with a CHECK constraint containing an IN clause. In the planner, we enable this when there are suitable hash functions for the ScalarArrayOpExpr's operator and only when there is at least MIN_ARRAY_SIZE_FOR_HASHED_SAOP elements in the array. The threshold is currently set to 9. Author: James Coleman, David Rowley Reviewed-by: David Rowley, Tomas Vondra, Heikki Linnakangas Discussion: https://postgr.es/m/CAAaqYe8x62+=wn0zvNKCj55tPpg-JBHzhZFFc6ANovdqFw7-dA@mail.gmail.com
2021-04-08 13:51:22 +02:00
return DatumGetUInt32(hash);
}
/*
* Matching function for scalar array hash op elements, to be used in hashtable
* lookups.
*/
static bool
saop_hash_element_match(struct saophash_hash *tb, Datum key1, Datum key2)
{
Datum result;
ScalarArrayOpExprHashTable *elements_tab = (ScalarArrayOpExprHashTable *) tb->private_data;
FunctionCallInfo fcinfo = elements_tab->op->d.hashedscalararrayop.fcinfo_data;
fcinfo->args[0].value = key1;
fcinfo->args[0].isnull = false;
fcinfo->args[1].value = key2;
fcinfo->args[1].isnull = false;
Remove size increase in ExprEvalStep caused by hashed saops 50e17ad28 increased the size of ExprEvalStep from 64 bytes up to 88 bytes. Lots of effort was spent during the development of the current expression evaluation code to make an instance of this struct as small as possible. Making this struct larger than needed reduces CPU cache efficiency during expression evaluation which causes noticeable slowdowns during query execution. In order to reduce the size of the struct, here we remove the fn_addr field. The values from this field can be obtained via fcinfo, just with some extra pointer dereferencing. The extra indirection does not seem to cause any noticeable slowdowns. Various other fields have been moved into the ScalarArrayOpExprHashTable struct. These fields are only used when the ScalarArrayOpExprHashTable pointer has already been dereferenced, so no additional pointer dereferences occur for these. Here we also make hash_fcinfo_data the last field in ScalarArrayOpExprHashTable so that we can avoid a further pointer dereference to get the FunctionCallInfoBaseData. This also saves a call to palloc(). 50e17ad28 was added in 14, but it's too late to adjust the size of the ExprEvalStep in that version, so here we just backpatch to 15, which is currently in beta. Author: Andres Freund, David Rowley Discussion: https://postgr.es/m/20220616233130.rparivafipt6doj3@alap3.anarazel.de Backpatch-through: 15
2022-07-06 09:40:32 +02:00
result = elements_tab->op->d.hashedscalararrayop.finfo->fn_addr(fcinfo);
Speedup ScalarArrayOpExpr evaluation ScalarArrayOpExprs with "useOr=true" and a set of Consts on the righthand side have traditionally been evaluated by using a linear search over the array. When these arrays contain large numbers of elements then this linear search could become a significant part of execution time. Here we add a new method of evaluating ScalarArrayOpExpr expressions to allow them to be evaluated by first building a hash table containing each element, then on subsequent evaluations, we just probe that hash table to determine if there is a match. The planner is in charge of determining when this optimization is possible and it enables it by setting hashfuncid in the ScalarArrayOpExpr. The executor will only perform the hash table evaluation when the hashfuncid is set. This means that not all cases are optimized. For example CHECK constraints containing an IN clause won't go through the planner, so won't get the hashfuncid set. We could maybe do something about that at some later date. The reason we're not doing it now is from fear that we may slow down cases where the expression is evaluated only once. Those cases can be common, for example, a single row INSERT to a table with a CHECK constraint containing an IN clause. In the planner, we enable this when there are suitable hash functions for the ScalarArrayOpExpr's operator and only when there is at least MIN_ARRAY_SIZE_FOR_HASHED_SAOP elements in the array. The threshold is currently set to 9. Author: James Coleman, David Rowley Reviewed-by: David Rowley, Tomas Vondra, Heikki Linnakangas Discussion: https://postgr.es/m/CAAaqYe8x62+=wn0zvNKCj55tPpg-JBHzhZFFc6ANovdqFw7-dA@mail.gmail.com
2021-04-08 13:51:22 +02:00
return DatumGetBool(result);
}
/*
* Evaluate "scalar op ANY (const array)".
*
* Similar to ExecEvalScalarArrayOp, but optimized for faster repeat lookups
* by building a hashtable on the first lookup. This hashtable will be reused
* by subsequent lookups. Unlike ExecEvalScalarArrayOp, this version only
* supports OR semantics.
*
* Source array is in our result area, scalar arg is already evaluated into
* fcinfo->args[0].
*
* The operator always yields boolean.
*/
void
ExecEvalHashedScalarArrayOp(ExprState *state, ExprEvalStep *op, ExprContext *econtext)
{
ScalarArrayOpExprHashTable *elements_tab = op->d.hashedscalararrayop.elements_tab;
FunctionCallInfo fcinfo = op->d.hashedscalararrayop.fcinfo_data;
Use a hash table to speed up NOT IN(values) Similar to 50e17ad28, which allowed hash tables to be used for IN clauses with a set of constants, here we add the same feature for NOT IN clauses. NOT IN evaluates the same as: WHERE a <> v1 AND a <> v2 AND a <> v3. Obviously, if we're using a hash table we must be exactly equivalent to that and return the same result taking into account that either side of the condition could contain a NULL. This requires a little bit of special handling to make work with the hash table version. When processing NOT IN, the ScalarArrayOpExpr's operator will be the <> operator. To be able to build and lookup a hash table we must use the <>'s negator operator. The planner checks if that exists and is hashable and sets the relevant fields in ScalarArrayOpExpr to instruct the executor to use hashing. Author: David Rowley, James Coleman Reviewed-by: James Coleman, Zhihong Yu Discussion: https://postgr.es/m/CAApHDvoF1mum_FRk6D621edcB6KSHBi2+GAgWmioj5AhOu2vwQ@mail.gmail.com
2021-07-07 06:29:17 +02:00
bool inclause = op->d.hashedscalararrayop.inclause;
Speedup ScalarArrayOpExpr evaluation ScalarArrayOpExprs with "useOr=true" and a set of Consts on the righthand side have traditionally been evaluated by using a linear search over the array. When these arrays contain large numbers of elements then this linear search could become a significant part of execution time. Here we add a new method of evaluating ScalarArrayOpExpr expressions to allow them to be evaluated by first building a hash table containing each element, then on subsequent evaluations, we just probe that hash table to determine if there is a match. The planner is in charge of determining when this optimization is possible and it enables it by setting hashfuncid in the ScalarArrayOpExpr. The executor will only perform the hash table evaluation when the hashfuncid is set. This means that not all cases are optimized. For example CHECK constraints containing an IN clause won't go through the planner, so won't get the hashfuncid set. We could maybe do something about that at some later date. The reason we're not doing it now is from fear that we may slow down cases where the expression is evaluated only once. Those cases can be common, for example, a single row INSERT to a table with a CHECK constraint containing an IN clause. In the planner, we enable this when there are suitable hash functions for the ScalarArrayOpExpr's operator and only when there is at least MIN_ARRAY_SIZE_FOR_HASHED_SAOP elements in the array. The threshold is currently set to 9. Author: James Coleman, David Rowley Reviewed-by: David Rowley, Tomas Vondra, Heikki Linnakangas Discussion: https://postgr.es/m/CAAaqYe8x62+=wn0zvNKCj55tPpg-JBHzhZFFc6ANovdqFw7-dA@mail.gmail.com
2021-04-08 13:51:22 +02:00
bool strictfunc = op->d.hashedscalararrayop.finfo->fn_strict;
Datum scalar = fcinfo->args[0].value;
bool scalar_isnull = fcinfo->args[0].isnull;
Datum result;
bool resultnull;
bool hashfound;
/* We don't setup a hashed scalar array op if the array const is null. */
Assert(!*op->resnull);
/*
* If the scalar is NULL, and the function is strict, return NULL; no
* point in executing the search.
*/
if (fcinfo->args[0].isnull && strictfunc)
{
*op->resnull = true;
return;
}
/* Build the hash table on first evaluation */
if (elements_tab == NULL)
{
Remove size increase in ExprEvalStep caused by hashed saops 50e17ad28 increased the size of ExprEvalStep from 64 bytes up to 88 bytes. Lots of effort was spent during the development of the current expression evaluation code to make an instance of this struct as small as possible. Making this struct larger than needed reduces CPU cache efficiency during expression evaluation which causes noticeable slowdowns during query execution. In order to reduce the size of the struct, here we remove the fn_addr field. The values from this field can be obtained via fcinfo, just with some extra pointer dereferencing. The extra indirection does not seem to cause any noticeable slowdowns. Various other fields have been moved into the ScalarArrayOpExprHashTable struct. These fields are only used when the ScalarArrayOpExprHashTable pointer has already been dereferenced, so no additional pointer dereferences occur for these. Here we also make hash_fcinfo_data the last field in ScalarArrayOpExprHashTable so that we can avoid a further pointer dereference to get the FunctionCallInfoBaseData. This also saves a call to palloc(). 50e17ad28 was added in 14, but it's too late to adjust the size of the ExprEvalStep in that version, so here we just backpatch to 15, which is currently in beta. Author: Andres Freund, David Rowley Discussion: https://postgr.es/m/20220616233130.rparivafipt6doj3@alap3.anarazel.de Backpatch-through: 15
2022-07-06 09:40:32 +02:00
ScalarArrayOpExpr *saop;
Speedup ScalarArrayOpExpr evaluation ScalarArrayOpExprs with "useOr=true" and a set of Consts on the righthand side have traditionally been evaluated by using a linear search over the array. When these arrays contain large numbers of elements then this linear search could become a significant part of execution time. Here we add a new method of evaluating ScalarArrayOpExpr expressions to allow them to be evaluated by first building a hash table containing each element, then on subsequent evaluations, we just probe that hash table to determine if there is a match. The planner is in charge of determining when this optimization is possible and it enables it by setting hashfuncid in the ScalarArrayOpExpr. The executor will only perform the hash table evaluation when the hashfuncid is set. This means that not all cases are optimized. For example CHECK constraints containing an IN clause won't go through the planner, so won't get the hashfuncid set. We could maybe do something about that at some later date. The reason we're not doing it now is from fear that we may slow down cases where the expression is evaluated only once. Those cases can be common, for example, a single row INSERT to a table with a CHECK constraint containing an IN clause. In the planner, we enable this when there are suitable hash functions for the ScalarArrayOpExpr's operator and only when there is at least MIN_ARRAY_SIZE_FOR_HASHED_SAOP elements in the array. The threshold is currently set to 9. Author: James Coleman, David Rowley Reviewed-by: David Rowley, Tomas Vondra, Heikki Linnakangas Discussion: https://postgr.es/m/CAAaqYe8x62+=wn0zvNKCj55tPpg-JBHzhZFFc6ANovdqFw7-dA@mail.gmail.com
2021-04-08 13:51:22 +02:00
int16 typlen;
bool typbyval;
char typalign;
int nitems;
bool has_nulls = false;
char *s;
bits8 *bitmap;
int bitmask;
MemoryContext oldcontext;
ArrayType *arr;
Remove size increase in ExprEvalStep caused by hashed saops 50e17ad28 increased the size of ExprEvalStep from 64 bytes up to 88 bytes. Lots of effort was spent during the development of the current expression evaluation code to make an instance of this struct as small as possible. Making this struct larger than needed reduces CPU cache efficiency during expression evaluation which causes noticeable slowdowns during query execution. In order to reduce the size of the struct, here we remove the fn_addr field. The values from this field can be obtained via fcinfo, just with some extra pointer dereferencing. The extra indirection does not seem to cause any noticeable slowdowns. Various other fields have been moved into the ScalarArrayOpExprHashTable struct. These fields are only used when the ScalarArrayOpExprHashTable pointer has already been dereferenced, so no additional pointer dereferences occur for these. Here we also make hash_fcinfo_data the last field in ScalarArrayOpExprHashTable so that we can avoid a further pointer dereference to get the FunctionCallInfoBaseData. This also saves a call to palloc(). 50e17ad28 was added in 14, but it's too late to adjust the size of the ExprEvalStep in that version, so here we just backpatch to 15, which is currently in beta. Author: Andres Freund, David Rowley Discussion: https://postgr.es/m/20220616233130.rparivafipt6doj3@alap3.anarazel.de Backpatch-through: 15
2022-07-06 09:40:32 +02:00
saop = op->d.hashedscalararrayop.saop;
Speedup ScalarArrayOpExpr evaluation ScalarArrayOpExprs with "useOr=true" and a set of Consts on the righthand side have traditionally been evaluated by using a linear search over the array. When these arrays contain large numbers of elements then this linear search could become a significant part of execution time. Here we add a new method of evaluating ScalarArrayOpExpr expressions to allow them to be evaluated by first building a hash table containing each element, then on subsequent evaluations, we just probe that hash table to determine if there is a match. The planner is in charge of determining when this optimization is possible and it enables it by setting hashfuncid in the ScalarArrayOpExpr. The executor will only perform the hash table evaluation when the hashfuncid is set. This means that not all cases are optimized. For example CHECK constraints containing an IN clause won't go through the planner, so won't get the hashfuncid set. We could maybe do something about that at some later date. The reason we're not doing it now is from fear that we may slow down cases where the expression is evaluated only once. Those cases can be common, for example, a single row INSERT to a table with a CHECK constraint containing an IN clause. In the planner, we enable this when there are suitable hash functions for the ScalarArrayOpExpr's operator and only when there is at least MIN_ARRAY_SIZE_FOR_HASHED_SAOP elements in the array. The threshold is currently set to 9. Author: James Coleman, David Rowley Reviewed-by: David Rowley, Tomas Vondra, Heikki Linnakangas Discussion: https://postgr.es/m/CAAaqYe8x62+=wn0zvNKCj55tPpg-JBHzhZFFc6ANovdqFw7-dA@mail.gmail.com
2021-04-08 13:51:22 +02:00
arr = DatumGetArrayTypeP(*op->resvalue);
nitems = ArrayGetNItems(ARR_NDIM(arr), ARR_DIMS(arr));
get_typlenbyvalalign(ARR_ELEMTYPE(arr),
&typlen,
&typbyval,
&typalign);
oldcontext = MemoryContextSwitchTo(econtext->ecxt_per_query_memory);
elements_tab = (ScalarArrayOpExprHashTable *)
Remove size increase in ExprEvalStep caused by hashed saops 50e17ad28 increased the size of ExprEvalStep from 64 bytes up to 88 bytes. Lots of effort was spent during the development of the current expression evaluation code to make an instance of this struct as small as possible. Making this struct larger than needed reduces CPU cache efficiency during expression evaluation which causes noticeable slowdowns during query execution. In order to reduce the size of the struct, here we remove the fn_addr field. The values from this field can be obtained via fcinfo, just with some extra pointer dereferencing. The extra indirection does not seem to cause any noticeable slowdowns. Various other fields have been moved into the ScalarArrayOpExprHashTable struct. These fields are only used when the ScalarArrayOpExprHashTable pointer has already been dereferenced, so no additional pointer dereferences occur for these. Here we also make hash_fcinfo_data the last field in ScalarArrayOpExprHashTable so that we can avoid a further pointer dereference to get the FunctionCallInfoBaseData. This also saves a call to palloc(). 50e17ad28 was added in 14, but it's too late to adjust the size of the ExprEvalStep in that version, so here we just backpatch to 15, which is currently in beta. Author: Andres Freund, David Rowley Discussion: https://postgr.es/m/20220616233130.rparivafipt6doj3@alap3.anarazel.de Backpatch-through: 15
2022-07-06 09:40:32 +02:00
palloc0(offsetof(ScalarArrayOpExprHashTable, hash_fcinfo_data) +
SizeForFunctionCallInfo(1));
Speedup ScalarArrayOpExpr evaluation ScalarArrayOpExprs with "useOr=true" and a set of Consts on the righthand side have traditionally been evaluated by using a linear search over the array. When these arrays contain large numbers of elements then this linear search could become a significant part of execution time. Here we add a new method of evaluating ScalarArrayOpExpr expressions to allow them to be evaluated by first building a hash table containing each element, then on subsequent evaluations, we just probe that hash table to determine if there is a match. The planner is in charge of determining when this optimization is possible and it enables it by setting hashfuncid in the ScalarArrayOpExpr. The executor will only perform the hash table evaluation when the hashfuncid is set. This means that not all cases are optimized. For example CHECK constraints containing an IN clause won't go through the planner, so won't get the hashfuncid set. We could maybe do something about that at some later date. The reason we're not doing it now is from fear that we may slow down cases where the expression is evaluated only once. Those cases can be common, for example, a single row INSERT to a table with a CHECK constraint containing an IN clause. In the planner, we enable this when there are suitable hash functions for the ScalarArrayOpExpr's operator and only when there is at least MIN_ARRAY_SIZE_FOR_HASHED_SAOP elements in the array. The threshold is currently set to 9. Author: James Coleman, David Rowley Reviewed-by: David Rowley, Tomas Vondra, Heikki Linnakangas Discussion: https://postgr.es/m/CAAaqYe8x62+=wn0zvNKCj55tPpg-JBHzhZFFc6ANovdqFw7-dA@mail.gmail.com
2021-04-08 13:51:22 +02:00
op->d.hashedscalararrayop.elements_tab = elements_tab;
elements_tab->op = op;
Remove size increase in ExprEvalStep caused by hashed saops 50e17ad28 increased the size of ExprEvalStep from 64 bytes up to 88 bytes. Lots of effort was spent during the development of the current expression evaluation code to make an instance of this struct as small as possible. Making this struct larger than needed reduces CPU cache efficiency during expression evaluation which causes noticeable slowdowns during query execution. In order to reduce the size of the struct, here we remove the fn_addr field. The values from this field can be obtained via fcinfo, just with some extra pointer dereferencing. The extra indirection does not seem to cause any noticeable slowdowns. Various other fields have been moved into the ScalarArrayOpExprHashTable struct. These fields are only used when the ScalarArrayOpExprHashTable pointer has already been dereferenced, so no additional pointer dereferences occur for these. Here we also make hash_fcinfo_data the last field in ScalarArrayOpExprHashTable so that we can avoid a further pointer dereference to get the FunctionCallInfoBaseData. This also saves a call to palloc(). 50e17ad28 was added in 14, but it's too late to adjust the size of the ExprEvalStep in that version, so here we just backpatch to 15, which is currently in beta. Author: Andres Freund, David Rowley Discussion: https://postgr.es/m/20220616233130.rparivafipt6doj3@alap3.anarazel.de Backpatch-through: 15
2022-07-06 09:40:32 +02:00
fmgr_info(saop->hashfuncid, &elements_tab->hash_finfo);
fmgr_info_set_expr((Node *) saop, &elements_tab->hash_finfo);
InitFunctionCallInfoData(elements_tab->hash_fcinfo_data,
&elements_tab->hash_finfo,
1,
saop->inputcollid,
NULL,
NULL);
Speedup ScalarArrayOpExpr evaluation ScalarArrayOpExprs with "useOr=true" and a set of Consts on the righthand side have traditionally been evaluated by using a linear search over the array. When these arrays contain large numbers of elements then this linear search could become a significant part of execution time. Here we add a new method of evaluating ScalarArrayOpExpr expressions to allow them to be evaluated by first building a hash table containing each element, then on subsequent evaluations, we just probe that hash table to determine if there is a match. The planner is in charge of determining when this optimization is possible and it enables it by setting hashfuncid in the ScalarArrayOpExpr. The executor will only perform the hash table evaluation when the hashfuncid is set. This means that not all cases are optimized. For example CHECK constraints containing an IN clause won't go through the planner, so won't get the hashfuncid set. We could maybe do something about that at some later date. The reason we're not doing it now is from fear that we may slow down cases where the expression is evaluated only once. Those cases can be common, for example, a single row INSERT to a table with a CHECK constraint containing an IN clause. In the planner, we enable this when there are suitable hash functions for the ScalarArrayOpExpr's operator and only when there is at least MIN_ARRAY_SIZE_FOR_HASHED_SAOP elements in the array. The threshold is currently set to 9. Author: James Coleman, David Rowley Reviewed-by: David Rowley, Tomas Vondra, Heikki Linnakangas Discussion: https://postgr.es/m/CAAaqYe8x62+=wn0zvNKCj55tPpg-JBHzhZFFc6ANovdqFw7-dA@mail.gmail.com
2021-04-08 13:51:22 +02:00
/*
* Create the hash table sizing it according to the number of elements
* in the array. This does assume that the array has no duplicates.
* If the array happens to contain many duplicate values then it'll
* just mean that we sized the table a bit on the large side.
*/
elements_tab->hashtab = saophash_create(CurrentMemoryContext, nitems,
elements_tab);
MemoryContextSwitchTo(oldcontext);
s = (char *) ARR_DATA_PTR(arr);
bitmap = ARR_NULLBITMAP(arr);
bitmask = 1;
for (int i = 0; i < nitems; i++)
{
/* Get array element, checking for NULL. */
if (bitmap && (*bitmap & bitmask) == 0)
{
has_nulls = true;
}
else
{
Datum element;
element = fetch_att(s, typbyval, typlen);
s = att_addlength_pointer(s, typlen, s);
s = (char *) att_align_nominal(s, typalign);
saophash_insert(elements_tab->hashtab, element, &hashfound);
}
/* Advance bitmap pointer if any. */
if (bitmap)
{
bitmask <<= 1;
if (bitmask == 0x100)
{
bitmap++;
bitmask = 1;
}
}
}
/*
* Remember if we had any nulls so that we know if we need to execute
* non-strict functions with a null lhs value if no match is found.
*/
op->d.hashedscalararrayop.has_nulls = has_nulls;
}
/* Check the hash to see if we have a match. */
hashfound = NULL != saophash_lookup(elements_tab->hashtab, scalar);
Use a hash table to speed up NOT IN(values) Similar to 50e17ad28, which allowed hash tables to be used for IN clauses with a set of constants, here we add the same feature for NOT IN clauses. NOT IN evaluates the same as: WHERE a <> v1 AND a <> v2 AND a <> v3. Obviously, if we're using a hash table we must be exactly equivalent to that and return the same result taking into account that either side of the condition could contain a NULL. This requires a little bit of special handling to make work with the hash table version. When processing NOT IN, the ScalarArrayOpExpr's operator will be the <> operator. To be able to build and lookup a hash table we must use the <>'s negator operator. The planner checks if that exists and is hashable and sets the relevant fields in ScalarArrayOpExpr to instruct the executor to use hashing. Author: David Rowley, James Coleman Reviewed-by: James Coleman, Zhihong Yu Discussion: https://postgr.es/m/CAApHDvoF1mum_FRk6D621edcB6KSHBi2+GAgWmioj5AhOu2vwQ@mail.gmail.com
2021-07-07 06:29:17 +02:00
/* the result depends on if the clause is an IN or NOT IN clause */
if (inclause)
result = BoolGetDatum(hashfound); /* IN */
else
result = BoolGetDatum(!hashfound); /* NOT IN */
Speedup ScalarArrayOpExpr evaluation ScalarArrayOpExprs with "useOr=true" and a set of Consts on the righthand side have traditionally been evaluated by using a linear search over the array. When these arrays contain large numbers of elements then this linear search could become a significant part of execution time. Here we add a new method of evaluating ScalarArrayOpExpr expressions to allow them to be evaluated by first building a hash table containing each element, then on subsequent evaluations, we just probe that hash table to determine if there is a match. The planner is in charge of determining when this optimization is possible and it enables it by setting hashfuncid in the ScalarArrayOpExpr. The executor will only perform the hash table evaluation when the hashfuncid is set. This means that not all cases are optimized. For example CHECK constraints containing an IN clause won't go through the planner, so won't get the hashfuncid set. We could maybe do something about that at some later date. The reason we're not doing it now is from fear that we may slow down cases where the expression is evaluated only once. Those cases can be common, for example, a single row INSERT to a table with a CHECK constraint containing an IN clause. In the planner, we enable this when there are suitable hash functions for the ScalarArrayOpExpr's operator and only when there is at least MIN_ARRAY_SIZE_FOR_HASHED_SAOP elements in the array. The threshold is currently set to 9. Author: James Coleman, David Rowley Reviewed-by: David Rowley, Tomas Vondra, Heikki Linnakangas Discussion: https://postgr.es/m/CAAaqYe8x62+=wn0zvNKCj55tPpg-JBHzhZFFc6ANovdqFw7-dA@mail.gmail.com
2021-04-08 13:51:22 +02:00
resultnull = false;
/*
* If we didn't find a match in the array, we still might need to handle
* the possibility of null values. We didn't put any NULLs into the
* hashtable, but instead marked if we found any when building the table
* in has_nulls.
*/
Use a hash table to speed up NOT IN(values) Similar to 50e17ad28, which allowed hash tables to be used for IN clauses with a set of constants, here we add the same feature for NOT IN clauses. NOT IN evaluates the same as: WHERE a <> v1 AND a <> v2 AND a <> v3. Obviously, if we're using a hash table we must be exactly equivalent to that and return the same result taking into account that either side of the condition could contain a NULL. This requires a little bit of special handling to make work with the hash table version. When processing NOT IN, the ScalarArrayOpExpr's operator will be the <> operator. To be able to build and lookup a hash table we must use the <>'s negator operator. The planner checks if that exists and is hashable and sets the relevant fields in ScalarArrayOpExpr to instruct the executor to use hashing. Author: David Rowley, James Coleman Reviewed-by: James Coleman, Zhihong Yu Discussion: https://postgr.es/m/CAApHDvoF1mum_FRk6D621edcB6KSHBi2+GAgWmioj5AhOu2vwQ@mail.gmail.com
2021-07-07 06:29:17 +02:00
if (!hashfound && op->d.hashedscalararrayop.has_nulls)
Speedup ScalarArrayOpExpr evaluation ScalarArrayOpExprs with "useOr=true" and a set of Consts on the righthand side have traditionally been evaluated by using a linear search over the array. When these arrays contain large numbers of elements then this linear search could become a significant part of execution time. Here we add a new method of evaluating ScalarArrayOpExpr expressions to allow them to be evaluated by first building a hash table containing each element, then on subsequent evaluations, we just probe that hash table to determine if there is a match. The planner is in charge of determining when this optimization is possible and it enables it by setting hashfuncid in the ScalarArrayOpExpr. The executor will only perform the hash table evaluation when the hashfuncid is set. This means that not all cases are optimized. For example CHECK constraints containing an IN clause won't go through the planner, so won't get the hashfuncid set. We could maybe do something about that at some later date. The reason we're not doing it now is from fear that we may slow down cases where the expression is evaluated only once. Those cases can be common, for example, a single row INSERT to a table with a CHECK constraint containing an IN clause. In the planner, we enable this when there are suitable hash functions for the ScalarArrayOpExpr's operator and only when there is at least MIN_ARRAY_SIZE_FOR_HASHED_SAOP elements in the array. The threshold is currently set to 9. Author: James Coleman, David Rowley Reviewed-by: David Rowley, Tomas Vondra, Heikki Linnakangas Discussion: https://postgr.es/m/CAAaqYe8x62+=wn0zvNKCj55tPpg-JBHzhZFFc6ANovdqFw7-dA@mail.gmail.com
2021-04-08 13:51:22 +02:00
{
if (strictfunc)
{
/*
* We have nulls in the array so a non-null lhs and no match must
* yield NULL.
*/
result = (Datum) 0;
resultnull = true;
}
else
{
/*
* Execute function will null rhs just once.
*
* The hash lookup path will have scribbled on the lhs argument so
* we need to set it up also (even though we entered this function
* with it already set).
*/
fcinfo->args[0].value = scalar;
fcinfo->args[0].isnull = scalar_isnull;
fcinfo->args[1].value = (Datum) 0;
fcinfo->args[1].isnull = true;
Remove size increase in ExprEvalStep caused by hashed saops 50e17ad28 increased the size of ExprEvalStep from 64 bytes up to 88 bytes. Lots of effort was spent during the development of the current expression evaluation code to make an instance of this struct as small as possible. Making this struct larger than needed reduces CPU cache efficiency during expression evaluation which causes noticeable slowdowns during query execution. In order to reduce the size of the struct, here we remove the fn_addr field. The values from this field can be obtained via fcinfo, just with some extra pointer dereferencing. The extra indirection does not seem to cause any noticeable slowdowns. Various other fields have been moved into the ScalarArrayOpExprHashTable struct. These fields are only used when the ScalarArrayOpExprHashTable pointer has already been dereferenced, so no additional pointer dereferences occur for these. Here we also make hash_fcinfo_data the last field in ScalarArrayOpExprHashTable so that we can avoid a further pointer dereference to get the FunctionCallInfoBaseData. This also saves a call to palloc(). 50e17ad28 was added in 14, but it's too late to adjust the size of the ExprEvalStep in that version, so here we just backpatch to 15, which is currently in beta. Author: Andres Freund, David Rowley Discussion: https://postgr.es/m/20220616233130.rparivafipt6doj3@alap3.anarazel.de Backpatch-through: 15
2022-07-06 09:40:32 +02:00
result = op->d.hashedscalararrayop.finfo->fn_addr(fcinfo);
Speedup ScalarArrayOpExpr evaluation ScalarArrayOpExprs with "useOr=true" and a set of Consts on the righthand side have traditionally been evaluated by using a linear search over the array. When these arrays contain large numbers of elements then this linear search could become a significant part of execution time. Here we add a new method of evaluating ScalarArrayOpExpr expressions to allow them to be evaluated by first building a hash table containing each element, then on subsequent evaluations, we just probe that hash table to determine if there is a match. The planner is in charge of determining when this optimization is possible and it enables it by setting hashfuncid in the ScalarArrayOpExpr. The executor will only perform the hash table evaluation when the hashfuncid is set. This means that not all cases are optimized. For example CHECK constraints containing an IN clause won't go through the planner, so won't get the hashfuncid set. We could maybe do something about that at some later date. The reason we're not doing it now is from fear that we may slow down cases where the expression is evaluated only once. Those cases can be common, for example, a single row INSERT to a table with a CHECK constraint containing an IN clause. In the planner, we enable this when there are suitable hash functions for the ScalarArrayOpExpr's operator and only when there is at least MIN_ARRAY_SIZE_FOR_HASHED_SAOP elements in the array. The threshold is currently set to 9. Author: James Coleman, David Rowley Reviewed-by: David Rowley, Tomas Vondra, Heikki Linnakangas Discussion: https://postgr.es/m/CAAaqYe8x62+=wn0zvNKCj55tPpg-JBHzhZFFc6ANovdqFw7-dA@mail.gmail.com
2021-04-08 13:51:22 +02:00
resultnull = fcinfo->isnull;
Use a hash table to speed up NOT IN(values) Similar to 50e17ad28, which allowed hash tables to be used for IN clauses with a set of constants, here we add the same feature for NOT IN clauses. NOT IN evaluates the same as: WHERE a <> v1 AND a <> v2 AND a <> v3. Obviously, if we're using a hash table we must be exactly equivalent to that and return the same result taking into account that either side of the condition could contain a NULL. This requires a little bit of special handling to make work with the hash table version. When processing NOT IN, the ScalarArrayOpExpr's operator will be the <> operator. To be able to build and lookup a hash table we must use the <>'s negator operator. The planner checks if that exists and is hashable and sets the relevant fields in ScalarArrayOpExpr to instruct the executor to use hashing. Author: David Rowley, James Coleman Reviewed-by: James Coleman, Zhihong Yu Discussion: https://postgr.es/m/CAApHDvoF1mum_FRk6D621edcB6KSHBi2+GAgWmioj5AhOu2vwQ@mail.gmail.com
2021-07-07 06:29:17 +02:00
/*
* Reverse the result for NOT IN clauses since the above function
* is the equality function and we need not-equals.
*/
if (!inclause)
result = !result;
Speedup ScalarArrayOpExpr evaluation ScalarArrayOpExprs with "useOr=true" and a set of Consts on the righthand side have traditionally been evaluated by using a linear search over the array. When these arrays contain large numbers of elements then this linear search could become a significant part of execution time. Here we add a new method of evaluating ScalarArrayOpExpr expressions to allow them to be evaluated by first building a hash table containing each element, then on subsequent evaluations, we just probe that hash table to determine if there is a match. The planner is in charge of determining when this optimization is possible and it enables it by setting hashfuncid in the ScalarArrayOpExpr. The executor will only perform the hash table evaluation when the hashfuncid is set. This means that not all cases are optimized. For example CHECK constraints containing an IN clause won't go through the planner, so won't get the hashfuncid set. We could maybe do something about that at some later date. The reason we're not doing it now is from fear that we may slow down cases where the expression is evaluated only once. Those cases can be common, for example, a single row INSERT to a table with a CHECK constraint containing an IN clause. In the planner, we enable this when there are suitable hash functions for the ScalarArrayOpExpr's operator and only when there is at least MIN_ARRAY_SIZE_FOR_HASHED_SAOP elements in the array. The threshold is currently set to 9. Author: James Coleman, David Rowley Reviewed-by: David Rowley, Tomas Vondra, Heikki Linnakangas Discussion: https://postgr.es/m/CAAaqYe8x62+=wn0zvNKCj55tPpg-JBHzhZFFc6ANovdqFw7-dA@mail.gmail.com
2021-04-08 13:51:22 +02:00
}
}
*op->resvalue = result;
*op->resnull = resultnull;
}
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
/*
* Evaluate a NOT NULL domain constraint.
*/
void
ExecEvalConstraintNotNull(ExprState *state, ExprEvalStep *op)
{
if (*op->resnull)
ereport(ERROR,
(errcode(ERRCODE_NOT_NULL_VIOLATION),
errmsg("domain %s does not allow null values",
format_type_be(op->d.domaincheck.resulttype)),
errdatatype(op->d.domaincheck.resulttype)));
}
/*
* Evaluate a CHECK domain constraint.
*/
void
ExecEvalConstraintCheck(ExprState *state, ExprEvalStep *op)
{
if (!*op->d.domaincheck.checknull &&
!DatumGetBool(*op->d.domaincheck.checkvalue))
ereport(ERROR,
(errcode(ERRCODE_CHECK_VIOLATION),
errmsg("value for domain %s violates check constraint \"%s\"",
format_type_be(op->d.domaincheck.resulttype),
op->d.domaincheck.constraintname),
errdomainconstraint(op->d.domaincheck.resulttype,
op->d.domaincheck.constraintname)));
}
/*
* Evaluate the various forms of XmlExpr.
*
* Arguments have been evaluated into named_argvalue/named_argnull
* and/or argvalue/argnull arrays.
*/
void
ExecEvalXmlExpr(ExprState *state, ExprEvalStep *op)
{
XmlExpr *xexpr = op->d.xmlexpr.xexpr;
Datum value;
*op->resnull = true; /* until we get a result */
*op->resvalue = (Datum) 0;
switch (xexpr->op)
{
case IS_XMLCONCAT:
{
Datum *argvalue = op->d.xmlexpr.argvalue;
bool *argnull = op->d.xmlexpr.argnull;
List *values = NIL;
expression eval, jit: Minor code cleanups. This mostly consists of using C99 style for loops, moving variables into narrower scopes, and a smattering of other minor improvements. Done separately to make it easier to review patches with actual functional changes. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:04:50 +01:00
for (int i = 0; i < list_length(xexpr->args); i++)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
if (!argnull[i])
values = lappend(values, DatumGetPointer(argvalue[i]));
}
if (values != NIL)
{
*op->resvalue = PointerGetDatum(xmlconcat(values));
*op->resnull = false;
}
}
break;
case IS_XMLFOREST:
{
Datum *argvalue = op->d.xmlexpr.named_argvalue;
bool *argnull = op->d.xmlexpr.named_argnull;
StringInfoData buf;
ListCell *lc;
ListCell *lc2;
expression eval, jit: Minor code cleanups. This mostly consists of using C99 style for loops, moving variables into narrower scopes, and a smattering of other minor improvements. Done separately to make it easier to review patches with actual functional changes. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:04:50 +01:00
int i;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
initStringInfo(&buf);
i = 0;
forboth(lc, xexpr->named_args, lc2, xexpr->arg_names)
{
Expr *e = (Expr *) lfirst(lc);
char *argname = strVal(lfirst(lc2));
if (!argnull[i])
{
value = argvalue[i];
appendStringInfo(&buf, "<%s>%s</%s>",
argname,
map_sql_value_to_xml_value(value,
exprType((Node *) e), true),
argname);
*op->resnull = false;
}
i++;
}
if (!*op->resnull)
{
text *result;
result = cstring_to_text_with_len(buf.data, buf.len);
*op->resvalue = PointerGetDatum(result);
}
pfree(buf.data);
}
break;
case IS_XMLELEMENT:
*op->resvalue = PointerGetDatum(xmlelement(xexpr,
op->d.xmlexpr.named_argvalue,
op->d.xmlexpr.named_argnull,
op->d.xmlexpr.argvalue,
op->d.xmlexpr.argnull));
*op->resnull = false;
break;
case IS_XMLPARSE:
{
Datum *argvalue = op->d.xmlexpr.argvalue;
bool *argnull = op->d.xmlexpr.argnull;
text *data;
bool preserve_whitespace;
/* arguments are known to be text, bool */
Assert(list_length(xexpr->args) == 2);
if (argnull[0])
return;
value = argvalue[0];
data = DatumGetTextPP(value);
if (argnull[1]) /* probably can't happen */
return;
value = argvalue[1];
preserve_whitespace = DatumGetBool(value);
*op->resvalue = PointerGetDatum(xmlparse(data,
xexpr->xmloption,
preserve_whitespace));
*op->resnull = false;
}
break;
case IS_XMLPI:
{
text *arg;
bool isnull;
/* optional argument is known to be text */
Assert(list_length(xexpr->args) <= 1);
if (xexpr->args)
{
isnull = op->d.xmlexpr.argnull[0];
if (isnull)
arg = NULL;
else
arg = DatumGetTextPP(op->d.xmlexpr.argvalue[0]);
}
else
{
arg = NULL;
isnull = false;
}
*op->resvalue = PointerGetDatum(xmlpi(xexpr->name,
arg,
isnull,
op->resnull));
}
break;
case IS_XMLROOT:
{
Datum *argvalue = op->d.xmlexpr.argvalue;
bool *argnull = op->d.xmlexpr.argnull;
xmltype *data;
text *version;
int standalone;
/* arguments are known to be xml, text, int */
Assert(list_length(xexpr->args) == 3);
if (argnull[0])
return;
data = DatumGetXmlP(argvalue[0]);
if (argnull[1])
version = NULL;
else
version = DatumGetTextPP(argvalue[1]);
Assert(!argnull[2]); /* always present */
standalone = DatumGetInt32(argvalue[2]);
*op->resvalue = PointerGetDatum(xmlroot(data,
version,
standalone));
*op->resnull = false;
}
break;
case IS_XMLSERIALIZE:
{
Datum *argvalue = op->d.xmlexpr.argvalue;
bool *argnull = op->d.xmlexpr.argnull;
/* argument type is known to be xml */
Assert(list_length(xexpr->args) == 1);
if (argnull[0])
return;
value = argvalue[0];
Support [NO] INDENT option in XMLSERIALIZE(). This adds the ability to pretty-print XML documents ... according to libxml's somewhat idiosyncratic notions of what's pretty, anyway. One notable divergence from a strict reading of the spec is that libxml is willing to collapse empty nodes "<node></node>" to just "<node/>", whereas SQL and the underlying XML spec say that this option should only result in whitespace tweaks. Nonetheless, it seems close enough to justify using the SQL-standard syntax. Jim Jones, reviewed by Peter Smith and myself Discussion: https://postgr.es/m/2f5df461-dad8-6d7d-4568-08e10608a69b@uni-muenster.de
2023-03-15 21:58:59 +01:00
*op->resvalue =
PointerGetDatum(xmltotext_with_options(DatumGetXmlP(value),
xexpr->xmloption,
xexpr->indent));
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
*op->resnull = false;
}
break;
case IS_DOCUMENT:
{
Datum *argvalue = op->d.xmlexpr.argvalue;
bool *argnull = op->d.xmlexpr.argnull;
/* optional argument is known to be xml */
Assert(list_length(xexpr->args) == 1);
if (argnull[0])
return;
value = argvalue[0];
*op->resvalue =
BoolGetDatum(xml_is_document(DatumGetXmlP(value)));
*op->resnull = false;
}
break;
default:
elog(ERROR, "unrecognized XML operation");
break;
}
}
Move ExecEvalJsonConstructor new function to a more natural place Commit 7081ac46ace8 put it at the end of the file, but that doesn't look very nice.
2023-03-31 12:55:25 +02:00
/*
* Evaluate a JSON constructor expression.
*/
void
ExecEvalJsonConstructor(ExprState *state, ExprEvalStep *op,
ExprContext *econtext)
{
Datum res;
JsonConstructorExprState *jcstate = op->d.json_constructor.jcstate;
JsonConstructorExpr *ctor = jcstate->constructor;
bool is_jsonb = ctor->returning->format->format_type == JS_FORMAT_JSONB;
bool isnull = false;
if (ctor->type == JSCTOR_JSON_ARRAY)
res = (is_jsonb ?
jsonb_build_array_worker :
json_build_array_worker) (jcstate->nargs,
jcstate->arg_values,
jcstate->arg_nulls,
jcstate->arg_types,
jcstate->constructor->absent_on_null);
else if (ctor->type == JSCTOR_JSON_OBJECT)
res = (is_jsonb ?
jsonb_build_object_worker :
json_build_object_worker) (jcstate->nargs,
jcstate->arg_values,
jcstate->arg_nulls,
jcstate->arg_types,
jcstate->constructor->absent_on_null,
jcstate->constructor->unique);
else
elog(ERROR, "invalid JsonConstructorExpr type %d", ctor->type);
*op->resvalue = res;
*op->resnull = isnull;
}
SQL/JSON: support the IS JSON predicate This patch introduces the SQL standard IS JSON predicate. It operates on text and bytea values representing JSON, as well as on the json and jsonb types. Each test has IS and IS NOT variants and supports a WITH UNIQUE KEYS flag. The tests are: IS JSON [VALUE] IS JSON ARRAY IS JSON OBJECT IS JSON SCALAR These should be self-explanatory. The WITH UNIQUE KEYS flag makes these return false when duplicate keys exist in any object within the value, not necessarily directly contained in the outermost object. Author: Nikita Glukhov <n.gluhov@postgrespro.ru> Author: Teodor Sigaev <teodor@sigaev.ru> Author: Oleg Bartunov <obartunov@gmail.com> Author: Alexander Korotkov <aekorotkov@gmail.com> Author: Amit Langote <amitlangote09@gmail.com> Author: Andrew Dunstan <andrew@dunslane.net> Reviewers have included (in no particular order) Andres Freund, Alexander Korotkov, Pavel Stehule, Andrew Alsup, Erik Rijkers, Zihong Yu, Himanshu Upadhyaya, Daniel Gustafsson, Justin Pryzby. Discussion: https://postgr.es/m/CAF4Au4w2x-5LTnN_bxky-mq4=WOqsGsxSpENCzHRAzSnEd8+WQ@mail.gmail.com Discussion: https://postgr.es/m/cd0bb935-0158-78a7-08b5-904886deac4b@postgrespro.ru Discussion: https://postgr.es/m/20220616233130.rparivafipt6doj3@alap3.anarazel.de Discussion: https://postgr.es/m/abd9b83b-aa66-f230-3d6d-734817f0995d%40postgresql.org
2023-03-31 22:34:04 +02:00
/*
* Evaluate a IS JSON predicate.
*/
void
ExecEvalJsonIsPredicate(ExprState *state, ExprEvalStep *op)
{
JsonIsPredicate *pred = op->d.is_json.pred;
Datum js = *op->resvalue;
Oid exprtype;
bool res;
if (*op->resnull)
{
*op->resvalue = BoolGetDatum(false);
return;
}
exprtype = exprType(pred->expr);
if (exprtype == TEXTOID || exprtype == JSONOID)
{
text *json = DatumGetTextP(js);
if (pred->item_type == JS_TYPE_ANY)
res = true;
else
{
switch (json_get_first_token(json, false))
{
case JSON_TOKEN_OBJECT_START:
res = pred->item_type == JS_TYPE_OBJECT;
break;
case JSON_TOKEN_ARRAY_START:
res = pred->item_type == JS_TYPE_ARRAY;
break;
case JSON_TOKEN_STRING:
case JSON_TOKEN_NUMBER:
case JSON_TOKEN_TRUE:
case JSON_TOKEN_FALSE:
case JSON_TOKEN_NULL:
res = pred->item_type == JS_TYPE_SCALAR;
break;
default:
res = false;
break;
}
}
/*
* Do full parsing pass only for uniqueness check or for JSON text
* validation.
*/
if (res && (pred->unique_keys || exprtype == TEXTOID))
res = json_validate(json, pred->unique_keys, false);
}
else if (exprtype == JSONBOID)
{
if (pred->item_type == JS_TYPE_ANY)
res = true;
else
{
Jsonb *jb = DatumGetJsonbP(js);
switch (pred->item_type)
{
case JS_TYPE_OBJECT:
res = JB_ROOT_IS_OBJECT(jb);
break;
case JS_TYPE_ARRAY:
res = JB_ROOT_IS_ARRAY(jb) && !JB_ROOT_IS_SCALAR(jb);
break;
case JS_TYPE_SCALAR:
res = JB_ROOT_IS_ARRAY(jb) && JB_ROOT_IS_SCALAR(jb);
break;
default:
res = false;
break;
}
}
/* Key uniqueness check is redundant for jsonb */
}
else
res = false;
*op->resvalue = BoolGetDatum(res);
}
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
/*
* ExecEvalGroupingFunc
*
* Computes a bitmask with a bit for each (unevaluated) argument expression
* (rightmost arg is least significant bit).
*
* A bit is set if the corresponding expression is NOT part of the set of
* grouping expressions in the current grouping set.
*/
void
ExecEvalGroupingFunc(ExprState *state, ExprEvalStep *op)
{
expression eval: Don't redundantly keep track of AggState. It's already tracked via ExprState->parent, so we don't need to also include it in ExprEvalStep. When that code originally was written ExprState->parent didn't exist, but it since has been introduced in 6719b238e8f. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:06:16 +01:00
AggState *aggstate = castNode(AggState, state->parent);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
int result = 0;
expression eval: Don't redundantly keep track of AggState. It's already tracked via ExprState->parent, so we don't need to also include it in ExprEvalStep. When that code originally was written ExprState->parent didn't exist, but it since has been introduced in 6719b238e8f. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:06:16 +01:00
Bitmapset *grouped_cols = aggstate->grouped_cols;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
ListCell *lc;
foreach(lc, op->d.grouping_func.clauses)
{
int attnum = lfirst_int(lc);
result <<= 1;
if (!bms_is_member(attnum, grouped_cols))
result |= 1;
}
*op->resvalue = Int32GetDatum(result);
*op->resnull = false;
}
/*
* Hand off evaluation of a subplan to nodeSubplan.c
*/
void
ExecEvalSubPlan(ExprState *state, ExprEvalStep *op, ExprContext *econtext)
{
SubPlanState *sstate = op->d.subplan.sstate;
/* could potentially be nested, so make sure there's enough stack */
check_stack_depth();
*op->resvalue = ExecSubPlan(sstate, econtext, op->resnull);
}
/*
* Evaluate a wholerow Var expression.
*
* Returns a Datum whose value is the value of a whole-row range variable
* with respect to given expression context.
*/
void
ExecEvalWholeRowVar(ExprState *state, ExprEvalStep *op, ExprContext *econtext)
{
Var *variable = op->d.wholerow.var;
TupleTableSlot *slot;
TupleDesc output_tupdesc;
MemoryContext oldcontext;
HeapTupleHeader dtuple;
HeapTuple tuple;
/* This was checked by ExecInitExpr */
Assert(variable->varattno == InvalidAttrNumber);
/* Get the input slot we want */
switch (variable->varno)
{
case INNER_VAR:
/* get the tuple from the inner node */
slot = econtext->ecxt_innertuple;
break;
case OUTER_VAR:
/* get the tuple from the outer node */
slot = econtext->ecxt_outertuple;
break;
/* INDEX_VAR is handled by default case */
default:
/* get the tuple from the relation being scanned */
slot = econtext->ecxt_scantuple;
break;
}
/* Apply the junkfilter if any */
if (op->d.wholerow.junkFilter != NULL)
slot = ExecFilterJunk(op->d.wholerow.junkFilter, slot);
/*
* If first time through, obtain tuple descriptor and check compatibility.
*
* XXX: It'd be great if this could be moved to the expression
* initialization phase, but due to using slots that's currently not
* feasible.
*/
if (op->d.wholerow.first)
{
/* optimistically assume we don't need slow path */
op->d.wholerow.slow = false;
/*
* If the Var identifies a named composite type, we must check that
* the actual tuple type is compatible with it.
*/
if (variable->vartype != RECORDOID)
{
TupleDesc var_tupdesc;
TupleDesc slot_tupdesc;
/*
* We really only care about numbers of attributes and data types.
* Also, we can ignore type mismatch on columns that are dropped
* in the destination type, so long as (1) the physical storage
* matches or (2) the actual column value is NULL. Case (1) is
* helpful in some cases involving out-of-date cached plans, while
* case (2) is expected behavior in situations such as an INSERT
* into a table with dropped columns (the planner typically
* generates an INT4 NULL regardless of the dropped column type).
* If we find a dropped column and cannot verify that case (1)
* holds, we have to use the slow path to check (2) for each row.
Un-break whole-row Vars referencing domain-over-composite types. In commit ec62cb0aa, I foolishly replaced ExecEvalWholeRowVar's lookup_rowtype_tupdesc_domain call with just lookup_rowtype_tupdesc, because I didn't see how a domain could be involved there, and there were no regression test cases to jog my memory. But the existing code was correct, so revert that change and add a test case showing why it's necessary. (Note: per comment in struct DatumTupleFields, it is correct to produce an output tuple that's labeled with the base composite type, not the domain; hence just blindly looking through the domain is correct here.) Per bug #17515 from Dan Kubb. Back-patch to v11 where domains over composites became a thing. Discussion: https://postgr.es/m/17515-a24737438363aca0@postgresql.org
2022-06-10 16:35:57 +02:00
*
* If vartype is a domain over composite, just look through that
* to the base composite type.
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
*/
Un-break whole-row Vars referencing domain-over-composite types. In commit ec62cb0aa, I foolishly replaced ExecEvalWholeRowVar's lookup_rowtype_tupdesc_domain call with just lookup_rowtype_tupdesc, because I didn't see how a domain could be involved there, and there were no regression test cases to jog my memory. But the existing code was correct, so revert that change and add a test case showing why it's necessary. (Note: per comment in struct DatumTupleFields, it is correct to produce an output tuple that's labeled with the base composite type, not the domain; hence just blindly looking through the domain is correct here.) Per bug #17515 from Dan Kubb. Back-patch to v11 where domains over composites became a thing. Discussion: https://postgr.es/m/17515-a24737438363aca0@postgresql.org
2022-06-10 16:35:57 +02:00
var_tupdesc = lookup_rowtype_tupdesc_domain(variable->vartype,
-1, false);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
slot_tupdesc = slot->tts_tupleDescriptor;
if (var_tupdesc->natts != slot_tupdesc->natts)
ereport(ERROR,
(errcode(ERRCODE_DATATYPE_MISMATCH),
errmsg("table row type and query-specified row type do not match"),
errdetail_plural("Table row contains %d attribute, but query expects %d.",
"Table row contains %d attributes, but query expects %d.",
slot_tupdesc->natts,
slot_tupdesc->natts,
var_tupdesc->natts)));
expression eval, jit: Minor code cleanups. This mostly consists of using C99 style for loops, moving variables into narrower scopes, and a smattering of other minor improvements. Done separately to make it easier to review patches with actual functional changes. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:04:50 +01:00
for (int i = 0; i < var_tupdesc->natts; i++)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
Change tupledesc->attrs[n] to TupleDescAttr(tupledesc, n). This is a mechanical change in preparation for a later commit that will change the layout of TupleDesc. Introducing a macro to abstract the details of where attributes are stored will allow us to change that in separate step and revise it in future. Author: Thomas Munro, editorialized by Andres Freund Reviewed-By: Andres Freund Discussion: https://postgr.es/m/CAEepm=0ZtQ-SpsgCyzzYpsXS6e=kZWqk3g5Ygn3MDV7A8dabUA@mail.gmail.com
2017-08-20 20:19:07 +02:00
Form_pg_attribute vattr = TupleDescAttr(var_tupdesc, i);
Form_pg_attribute sattr = TupleDescAttr(slot_tupdesc, i);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
if (vattr->atttypid == sattr->atttypid)
continue; /* no worries */
if (!vattr->attisdropped)
ereport(ERROR,
(errcode(ERRCODE_DATATYPE_MISMATCH),
errmsg("table row type and query-specified row type do not match"),
errdetail("Table has type %s at ordinal position %d, but query expects %s.",
format_type_be(sattr->atttypid),
i + 1,
format_type_be(vattr->atttypid))));
if (vattr->attlen != sattr->attlen ||
vattr->attalign != sattr->attalign)
op->d.wholerow.slow = true; /* need to check for nulls */
}
/*
* Use the variable's declared rowtype as the descriptor for the
Revert applying column aliases to the output of whole-row Vars. In commit bf7ca1587, I had the bright idea that we could make the result of a whole-row Var (that is, foo.*) track any column aliases that had been applied to the FROM entry the Var refers to. However, that's not terribly logically consistent, because now the output of the Var is no longer of the named composite type that the Var claims to emit. bf7ca1587 tried to handle that by changing the output tuple values to be labeled with a blessed RECORD type, but that's really pretty disastrous: we can wind up storing such tuples onto disk, whereupon they're not readable by other sessions. The only practical fix I can see is to give up on what bf7ca1587 tried to do, and say that the column names of tuples produced by a whole-row Var are always those of the underlying named composite type, query aliases or no. While this introduces some inconsistencies, it removes others, so it's not that awful in the abstract. What *is* kind of awful is to make such a behavioral change in a back-patched bug fix. But corrupt data is worse, so back-patched it will be. (A workaround available to anyone who's unhappy about this is to introduce an extra level of sub-SELECT, so that the whole-row Var is referring to the sub-SELECT's output and not to a named table type. Then the Var is of type RECORD to begin with and there's no issue.) Per report from Miles Delahunty. The faulty commit dates to 9.5, so back-patch to all supported branches. Discussion: https://postgr.es/m/2950001.1638729947@sss.pgh.pa.us
2022-03-17 23:18:05 +01:00
* output values. In particular, we *must* absorb any
* attisdropped markings.
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
*/
oldcontext = MemoryContextSwitchTo(econtext->ecxt_per_query_memory);
output_tupdesc = CreateTupleDescCopy(var_tupdesc);
MemoryContextSwitchTo(oldcontext);
ReleaseTupleDesc(var_tupdesc);
}
else
{
/*
* In the RECORD case, we use the input slot's rowtype as the
* descriptor for the output values, modulo possibly assigning new
* column names below.
*/
oldcontext = MemoryContextSwitchTo(econtext->ecxt_per_query_memory);
output_tupdesc = CreateTupleDescCopy(slot->tts_tupleDescriptor);
MemoryContextSwitchTo(oldcontext);
Revert applying column aliases to the output of whole-row Vars. In commit bf7ca1587, I had the bright idea that we could make the result of a whole-row Var (that is, foo.*) track any column aliases that had been applied to the FROM entry the Var refers to. However, that's not terribly logically consistent, because now the output of the Var is no longer of the named composite type that the Var claims to emit. bf7ca1587 tried to handle that by changing the output tuple values to be labeled with a blessed RECORD type, but that's really pretty disastrous: we can wind up storing such tuples onto disk, whereupon they're not readable by other sessions. The only practical fix I can see is to give up on what bf7ca1587 tried to do, and say that the column names of tuples produced by a whole-row Var are always those of the underlying named composite type, query aliases or no. While this introduces some inconsistencies, it removes others, so it's not that awful in the abstract. What *is* kind of awful is to make such a behavioral change in a back-patched bug fix. But corrupt data is worse, so back-patched it will be. (A workaround available to anyone who's unhappy about this is to introduce an extra level of sub-SELECT, so that the whole-row Var is referring to the sub-SELECT's output and not to a named table type. Then the Var is of type RECORD to begin with and there's no issue.) Per report from Miles Delahunty. The faulty commit dates to 9.5, so back-patch to all supported branches. Discussion: https://postgr.es/m/2950001.1638729947@sss.pgh.pa.us
2022-03-17 23:18:05 +01:00
/*
* It's possible that the input slot is a relation scan slot and
* so is marked with that relation's rowtype. But we're supposed
* to be returning RECORD, so reset to that.
*/
output_tupdesc->tdtypeid = RECORDOID;
output_tupdesc->tdtypmod = -1;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
Revert applying column aliases to the output of whole-row Vars. In commit bf7ca1587, I had the bright idea that we could make the result of a whole-row Var (that is, foo.*) track any column aliases that had been applied to the FROM entry the Var refers to. However, that's not terribly logically consistent, because now the output of the Var is no longer of the named composite type that the Var claims to emit. bf7ca1587 tried to handle that by changing the output tuple values to be labeled with a blessed RECORD type, but that's really pretty disastrous: we can wind up storing such tuples onto disk, whereupon they're not readable by other sessions. The only practical fix I can see is to give up on what bf7ca1587 tried to do, and say that the column names of tuples produced by a whole-row Var are always those of the underlying named composite type, query aliases or no. While this introduces some inconsistencies, it removes others, so it's not that awful in the abstract. What *is* kind of awful is to make such a behavioral change in a back-patched bug fix. But corrupt data is worse, so back-patched it will be. (A workaround available to anyone who's unhappy about this is to introduce an extra level of sub-SELECT, so that the whole-row Var is referring to the sub-SELECT's output and not to a named table type. Then the Var is of type RECORD to begin with and there's no issue.) Per report from Miles Delahunty. The faulty commit dates to 9.5, so back-patch to all supported branches. Discussion: https://postgr.es/m/2950001.1638729947@sss.pgh.pa.us
2022-03-17 23:18:05 +01:00
/*
* We already got the correct physical datatype info above, but
* now we should try to find the source RTE and adopt its column
* aliases, since it's unlikely that the input slot has the
* desired names.
*
* If we can't locate the RTE, assume the column names we've got
* are OK. (As of this writing, the only cases where we can't
* locate the RTE are in execution of trigger WHEN clauses, and
* then the Var will have the trigger's relation's rowtype, so its
* names are fine.) Also, if the creator of the RTE didn't bother
* to fill in an eref field, assume our column names are OK. (This
* happens in COPY, and perhaps other places.)
*/
if (econtext->ecxt_estate &&
variable->varno <= econtext->ecxt_estate->es_range_table_size)
{
RangeTblEntry *rte = exec_rt_fetch(variable->varno,
econtext->ecxt_estate);
if (rte->eref)
ExecTypeSetColNames(output_tupdesc, rte->eref->colnames);
}
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
}
/* Bless the tupdesc if needed, and save it in the execution state */
op->d.wholerow.tupdesc = BlessTupleDesc(output_tupdesc);
op->d.wholerow.first = false;
}
/*
* Make sure all columns of the slot are accessible in the slot's
* Datum/isnull arrays.
*/
slot_getallattrs(slot);
if (op->d.wholerow.slow)
{
/* Check to see if any dropped attributes are non-null */
TupleDesc tupleDesc = slot->tts_tupleDescriptor;
TupleDesc var_tupdesc = op->d.wholerow.tupdesc;
Assert(var_tupdesc->natts == tupleDesc->natts);
expression eval, jit: Minor code cleanups. This mostly consists of using C99 style for loops, moving variables into narrower scopes, and a smattering of other minor improvements. Done separately to make it easier to review patches with actual functional changes. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de
2020-02-07 04:04:50 +01:00
for (int i = 0; i < var_tupdesc->natts; i++)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
{
Change tupledesc->attrs[n] to TupleDescAttr(tupledesc, n). This is a mechanical change in preparation for a later commit that will change the layout of TupleDesc. Introducing a macro to abstract the details of where attributes are stored will allow us to change that in separate step and revise it in future. Author: Thomas Munro, editorialized by Andres Freund Reviewed-By: Andres Freund Discussion: https://postgr.es/m/CAEepm=0ZtQ-SpsgCyzzYpsXS6e=kZWqk3g5Ygn3MDV7A8dabUA@mail.gmail.com
2017-08-20 20:19:07 +02:00
Form_pg_attribute vattr = TupleDescAttr(var_tupdesc, i);
Form_pg_attribute sattr = TupleDescAttr(tupleDesc, i);
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
if (!vattr->attisdropped)
continue; /* already checked non-dropped cols */
if (slot->tts_isnull[i])
continue; /* null is always okay */
if (vattr->attlen != sattr->attlen ||
vattr->attalign != sattr->attalign)
ereport(ERROR,
(errcode(ERRCODE_DATATYPE_MISMATCH),
errmsg("table row type and query-specified row type do not match"),
errdetail("Physical storage mismatch on dropped attribute at ordinal position %d.",
i + 1)));
}
}
/*
Improve performance of ExecEvalWholeRowVar. In commit b8d7f053c, we needed to fix ExecEvalWholeRowVar to not change the state of the slot it's copying. The initial quick hack at that required two rounds of tuple construction, which is not very nice. To fix, add another primitive to tuptoaster.c that does precisely what we need. (I initially tried to do this by refactoring one of the existing functions into two pieces; but it looked like that might hurt performance for the existing case, and the amount of code that could be shared is not very large, so I gave up on that.) Discussion: https://postgr.es/m/26088.1490315792@sss.pgh.pa.us
2017-03-27 01:14:47 +02:00
* Build a composite datum, making sure any toasted fields get detoasted.
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
*
Improve performance of ExecEvalWholeRowVar. In commit b8d7f053c, we needed to fix ExecEvalWholeRowVar to not change the state of the slot it's copying. The initial quick hack at that required two rounds of tuple construction, which is not very nice. To fix, add another primitive to tuptoaster.c that does precisely what we need. (I initially tried to do this by refactoring one of the existing functions into two pieces; but it looked like that might hurt performance for the existing case, and the amount of code that could be shared is not very large, so I gave up on that.) Discussion: https://postgr.es/m/26088.1490315792@sss.pgh.pa.us
2017-03-27 01:14:47 +02:00
* (Note: it is critical that we not change the slot's state here.)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
*/
Improve performance of ExecEvalWholeRowVar. In commit b8d7f053c, we needed to fix ExecEvalWholeRowVar to not change the state of the slot it's copying. The initial quick hack at that required two rounds of tuple construction, which is not very nice. To fix, add another primitive to tuptoaster.c that does precisely what we need. (I initially tried to do this by refactoring one of the existing functions into two pieces; but it looked like that might hurt performance for the existing case, and the amount of code that could be shared is not very large, so I gave up on that.) Discussion: https://postgr.es/m/26088.1490315792@sss.pgh.pa.us
2017-03-27 01:14:47 +02:00
tuple = toast_build_flattened_tuple(slot->tts_tupleDescriptor,
slot->tts_values,
slot->tts_isnull);
dtuple = tuple->t_data;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
/*
* Label the datum with the composite type info we identified before.
Improve performance of ExecEvalWholeRowVar. In commit b8d7f053c, we needed to fix ExecEvalWholeRowVar to not change the state of the slot it's copying. The initial quick hack at that required two rounds of tuple construction, which is not very nice. To fix, add another primitive to tuptoaster.c that does precisely what we need. (I initially tried to do this by refactoring one of the existing functions into two pieces; but it looked like that might hurt performance for the existing case, and the amount of code that could be shared is not very large, so I gave up on that.) Discussion: https://postgr.es/m/26088.1490315792@sss.pgh.pa.us
2017-03-27 01:14:47 +02:00
*
* (Note: we could skip doing this by passing op->d.wholerow.tupdesc to
* the tuple build step; but that seems a tad risky so let's not.)
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
*/
HeapTupleHeaderSetTypeId(dtuple, op->d.wholerow.tupdesc->tdtypeid);
HeapTupleHeaderSetTypMod(dtuple, op->d.wholerow.tupdesc->tdtypmod);
*op->resvalue = PointerGetDatum(dtuple);
Improve performance of ExecEvalWholeRowVar. In commit b8d7f053c, we needed to fix ExecEvalWholeRowVar to not change the state of the slot it's copying. The initial quick hack at that required two rounds of tuple construction, which is not very nice. To fix, add another primitive to tuptoaster.c that does precisely what we need. (I initially tried to do this by refactoring one of the existing functions into two pieces; but it looked like that might hurt performance for the existing case, and the amount of code that could be shared is not very large, so I gave up on that.) Discussion: https://postgr.es/m/26088.1490315792@sss.pgh.pa.us
2017-03-27 01:14:47 +02:00
*op->resnull = false;
Faster expression evaluation and targetlist projection. This replaces the old, recursive tree-walk based evaluation, with non-recursive, opcode dispatch based, expression evaluation. Projection is now implemented as part of expression evaluation. This both leads to significant performance improvements, and makes future just-in-time compilation of expressions easier. The speed gains primarily come from: - non-recursive implementation reduces stack usage / overhead - simple sub-expressions are implemented with a single jump, without function calls - sharing some state between different sub-expressions - reduced amount of indirect/hard to predict memory accesses by laying out operation metadata sequentially; including the avoidance of nearly all of the previously used linked lists - more code has been moved to expression initialization, avoiding constant re-checks at evaluation time Future just-in-time compilation (JIT) has become easier, as demonstrated by released patches intended to be merged in a later release, for primarily two reasons: Firstly, due to a stricter split between expression initialization and evaluation, less code has to be handled by the JIT. Secondly, due to the non-recursive nature of the generated "instructions", less performance-critical code-paths can easily be shared between interpreted and compiled evaluation. The new framework allows for significant future optimizations. E.g.: - basic infrastructure for to later reduce the per executor-startup overhead of expression evaluation, by caching state in prepared statements. That'd be helpful in OLTPish scenarios where initialization overhead is measurable. - optimizing the generated "code". A number of proposals for potential work has already been made. - optimizing the interpreter. Similarly a number of proposals have been made here too. The move of logic into the expression initialization step leads to some backward-incompatible changes: - Function permission checks are now done during expression initialization, whereas previously they were done during execution. In edge cases this can lead to errors being raised that previously wouldn't have been, e.g. a NULL array being coerced to a different array type previously didn't perform checks. - The set of domain constraints to be checked, is now evaluated once during expression initialization, previously it was re-built every time a domain check was evaluated. For normal queries this doesn't change much, but e.g. for plpgsql functions, which caches ExprStates, the old set could stick around longer. The behavior around might still change. Author: Andres Freund, with significant changes by Tom Lane, changes by Heikki Linnakangas Reviewed-By: Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
2017-03-14 23:45:36 +01:00
}
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
Move EEOP_*_SYSVAR evaluation out of line. This mainly de-duplicates code. As evaluating a system variable isn't the hottest path and the current inline implementation ends up calling out to an external function anyway, this is OK from a performance POV. The main motivation for de-duplicating is the upcoming slot abstraction work, after which there's not guaranteed to be a HeapTuple backing the slot. Author: Andres Freund, Amit Khandekar Discussion: https://postgr.es/m/20181105210039.hh4vvi4vwoq5ba2q@alap3.anarazel.de
2018-11-07 20:08:45 +01:00
void
ExecEvalSysVar(ExprState *state, ExprEvalStep *op, ExprContext *econtext,
TupleTableSlot *slot)
{
Make TupleTableSlots extensible, finish split of existing slot type. This commit completes the work prepared in 1a0586de36, splitting the old TupleTableSlot implementation (which could store buffer, heap, minimal and virtual slots) into four different slot types. As described in the aforementioned commit, this is done with the goal of making tuple table slots extensible, to allow for pluggable table access methods. To achieve runtime extensibility for TupleTableSlots, operations on slots that can differ between types of slots are performed using the TupleTableSlotOps struct provided at slot creation time. That includes information from the size of TupleTableSlot struct to be allocated, initialization, deforming etc. See the struct's definition for more detailed information about callbacks TupleTableSlotOps. I decided to rename TTSOpsBufferTuple to TTSOpsBufferHeapTuple and ExecCopySlotTuple to ExecCopySlotHeapTuple, as that seems more consistent with other naming introduced in recent patches. There's plenty optimization potential in the slot implementation, but according to benchmarking the state after this commit has similar performance characteristics to before this set of changes, which seems sufficient. There's a few changes in execReplication.c that currently need to poke through the slot abstraction, that'll be repaired once the pluggable storage patchset provides the necessary infrastructure. Author: Andres Freund and Ashutosh Bapat, with changes by Amit Khandekar Discussion: https://postgr.es/m/20181105210039.hh4vvi4vwoq5ba2q@alap3.anarazel.de
2018-11-17 01:35:11 +01:00
Datum d;
Move EEOP_*_SYSVAR evaluation out of line. This mainly de-duplicates code. As evaluating a system variable isn't the hottest path and the current inline implementation ends up calling out to an external function anyway, this is OK from a performance POV. The main motivation for de-duplicating is the upcoming slot abstraction work, after which there's not guaranteed to be a HeapTuple backing the slot. Author: Andres Freund, Amit Khandekar Discussion: https://postgr.es/m/20181105210039.hh4vvi4vwoq5ba2q@alap3.anarazel.de
2018-11-07 20:08:45 +01:00
/* slot_getsysattr has sufficient defenses against bad attnums */
Make TupleTableSlots extensible, finish split of existing slot type. This commit completes the work prepared in 1a0586de36, splitting the old TupleTableSlot implementation (which could store buffer, heap, minimal and virtual slots) into four different slot types. As described in the aforementioned commit, this is done with the goal of making tuple table slots extensible, to allow for pluggable table access methods. To achieve runtime extensibility for TupleTableSlots, operations on slots that can differ between types of slots are performed using the TupleTableSlotOps struct provided at slot creation time. That includes information from the size of TupleTableSlot struct to be allocated, initialization, deforming etc. See the struct's definition for more detailed information about callbacks TupleTableSlotOps. I decided to rename TTSOpsBufferTuple to TTSOpsBufferHeapTuple and ExecCopySlotTuple to ExecCopySlotHeapTuple, as that seems more consistent with other naming introduced in recent patches. There's plenty optimization potential in the slot implementation, but according to benchmarking the state after this commit has similar performance characteristics to before this set of changes, which seems sufficient. There's a few changes in execReplication.c that currently need to poke through the slot abstraction, that'll be repaired once the pluggable storage patchset provides the necessary infrastructure. Author: Andres Freund and Ashutosh Bapat, with changes by Amit Khandekar Discussion: https://postgr.es/m/20181105210039.hh4vvi4vwoq5ba2q@alap3.anarazel.de
2018-11-17 01:35:11 +01:00
d = slot_getsysattr(slot,
op->d.var.attnum,
op->resnull);
*op->resvalue = d;
Move EEOP_*_SYSVAR evaluation out of line. This mainly de-duplicates code. As evaluating a system variable isn't the hottest path and the current inline implementation ends up calling out to an external function anyway, this is OK from a performance POV. The main motivation for de-duplicating is the upcoming slot abstraction work, after which there's not guaranteed to be a HeapTuple backing the slot. Author: Andres Freund, Amit Khandekar Discussion: https://postgr.es/m/20181105210039.hh4vvi4vwoq5ba2q@alap3.anarazel.de
2018-11-07 20:08:45 +01:00
/* this ought to be unreachable, but it's cheap enough to check */
Make TupleTableSlots extensible, finish split of existing slot type. This commit completes the work prepared in 1a0586de36, splitting the old TupleTableSlot implementation (which could store buffer, heap, minimal and virtual slots) into four different slot types. As described in the aforementioned commit, this is done with the goal of making tuple table slots extensible, to allow for pluggable table access methods. To achieve runtime extensibility for TupleTableSlots, operations on slots that can differ between types of slots are performed using the TupleTableSlotOps struct provided at slot creation time. That includes information from the size of TupleTableSlot struct to be allocated, initialization, deforming etc. See the struct's definition for more detailed information about callbacks TupleTableSlotOps. I decided to rename TTSOpsBufferTuple to TTSOpsBufferHeapTuple and ExecCopySlotTuple to ExecCopySlotHeapTuple, as that seems more consistent with other naming introduced in recent patches. There's plenty optimization potential in the slot implementation, but according to benchmarking the state after this commit has similar performance characteristics to before this set of changes, which seems sufficient. There's a few changes in execReplication.c that currently need to poke through the slot abstraction, that'll be repaired once the pluggable storage patchset provides the necessary infrastructure. Author: Andres Freund and Ashutosh Bapat, with changes by Amit Khandekar Discussion: https://postgr.es/m/20181105210039.hh4vvi4vwoq5ba2q@alap3.anarazel.de
2018-11-17 01:35:11 +01:00
if (unlikely(*op->resnull))
Move EEOP_*_SYSVAR evaluation out of line. This mainly de-duplicates code. As evaluating a system variable isn't the hottest path and the current inline implementation ends up calling out to an external function anyway, this is OK from a performance POV. The main motivation for de-duplicating is the upcoming slot abstraction work, after which there's not guaranteed to be a HeapTuple backing the slot. Author: Andres Freund, Amit Khandekar Discussion: https://postgr.es/m/20181105210039.hh4vvi4vwoq5ba2q@alap3.anarazel.de
2018-11-07 20:08:45 +01:00
elog(ERROR, "failed to fetch attribute from slot");
}
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
/*
* Transition value has not been initialized. This is the first non-NULL input
* value for a group. We use it as the initial value for transValue.
*/
void
expression eval: Reduce number of steps for agg transition invocations. Do so by combining the various steps that are part of aggregate transition function invocation into one larger step. As some of the current steps are only necessary for some aggregates, have one variant of the aggregate transition step for each possible combination. To avoid further manual copies of code in the different transition step implementations, move most of the code into helper functions marked as "always inline". The benefit of this change is an increase in performance when aggregating lots of rows. This comes in part due to the reduced number of indirect jumps due to the reduced number of steps, and in part by reducing redundant setup code across steps. This mainly benefits interpreted execution, but the code generated by JIT is also improved a bit. As a nice side-effect it also ends up making the code a bit simpler. A small additional optimization is removing the need to set aggstate->curaggcontext before calling ExecAggInitGroup, choosing to instead passign curaggcontext as an argument. It was, in contrast to other aggregate related functions, only needed to fetch a memory context to copy the transition value into. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de https://postgr.es/m/5c371df7cee903e8cd4c685f90c6c72086d3a2dc.camel@j-davis.com
2020-02-24 23:39:22 +01:00
ExecAggInitGroup(AggState *aggstate, AggStatePerTrans pertrans, AggStatePerGroup pergroup,
ExprContext *aggcontext)
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
{
Change function call information to be variable length. Before this change FunctionCallInfoData, the struct arguments etc for V1 function calls are stored in, always had space for FUNC_MAX_ARGS/100 arguments, storing datums and their nullness in two arrays. For nearly every function call 100 arguments is far more than needed, therefore wasting memory. Arg and argnull being two separate arrays also guarantees that to access a single argument, two cachelines have to be touched. Change the layout so there's a single variable-length array with pairs of value / isnull. That drastically reduces memory consumption for most function calls (on x86-64 a two argument function now uses 64bytes, previously 936 bytes), and makes it very likely that argument value and its nullness are on the same cacheline. Arguments are stored in a new NullableDatum struct, which, due to padding, needs more memory per argument than before. But as usually far fewer arguments are stored, and individual arguments are cheaper to access, that's still a clear win. It's likely that there's other places where conversion to NullableDatum arrays would make sense, e.g. TupleTableSlots, but that's for another commit. Because the function call information is now variable-length allocations have to take the number of arguments into account. For heap allocations that can be done with SizeForFunctionCallInfoData(), for on-stack allocations there's a new LOCAL_FCINFO(name, nargs) macro that helps to allocate an appropriately sized and aligned variable. Some places with stack allocation function call information don't know the number of arguments at compile time, and currently variably sized stack allocations aren't allowed in postgres. Therefore allow for FUNC_MAX_ARGS space in these cases. They're not that common, so for now that seems acceptable. Because of the need to allocate FunctionCallInfo of the appropriate size, older extensions may need to update their code. To avoid subtle breakages, the FunctionCallInfoData struct has been renamed to FunctionCallInfoBaseData. Most code only references FunctionCallInfo, so that shouldn't cause much collateral damage. This change is also a prerequisite for more efficient expression JIT compilation (by allocating the function call information on the stack, allowing LLVM to optimize it away); previously the size of the call information caused problems inside LLVM's optimizer. Author: Andres Freund Reviewed-By: Tom Lane Discussion: https://postgr.es/m/20180605172952.x34m5uz6ju6enaem@alap3.anarazel.de
2019-01-26 23:17:52 +01:00
FunctionCallInfo fcinfo = pertrans->transfn_fcinfo;
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
MemoryContext oldContext;
/*
* We must copy the datum into aggcontext if it is pass-by-ref. We do not
* need to pfree the old transValue, since it's NULL. (We already checked
* that the agg's input type is binary-compatible with its transtype, so
* straight copy here is OK.)
*/
expression eval: Reduce number of steps for agg transition invocations. Do so by combining the various steps that are part of aggregate transition function invocation into one larger step. As some of the current steps are only necessary for some aggregates, have one variant of the aggregate transition step for each possible combination. To avoid further manual copies of code in the different transition step implementations, move most of the code into helper functions marked as "always inline". The benefit of this change is an increase in performance when aggregating lots of rows. This comes in part due to the reduced number of indirect jumps due to the reduced number of steps, and in part by reducing redundant setup code across steps. This mainly benefits interpreted execution, but the code generated by JIT is also improved a bit. As a nice side-effect it also ends up making the code a bit simpler. A small additional optimization is removing the need to set aggstate->curaggcontext before calling ExecAggInitGroup, choosing to instead passign curaggcontext as an argument. It was, in contrast to other aggregate related functions, only needed to fetch a memory context to copy the transition value into. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de https://postgr.es/m/5c371df7cee903e8cd4c685f90c6c72086d3a2dc.camel@j-davis.com
2020-02-24 23:39:22 +01:00
oldContext = MemoryContextSwitchTo(aggcontext->ecxt_per_tuple_memory);
Change function call information to be variable length. Before this change FunctionCallInfoData, the struct arguments etc for V1 function calls are stored in, always had space for FUNC_MAX_ARGS/100 arguments, storing datums and their nullness in two arrays. For nearly every function call 100 arguments is far more than needed, therefore wasting memory. Arg and argnull being two separate arrays also guarantees that to access a single argument, two cachelines have to be touched. Change the layout so there's a single variable-length array with pairs of value / isnull. That drastically reduces memory consumption for most function calls (on x86-64 a two argument function now uses 64bytes, previously 936 bytes), and makes it very likely that argument value and its nullness are on the same cacheline. Arguments are stored in a new NullableDatum struct, which, due to padding, needs more memory per argument than before. But as usually far fewer arguments are stored, and individual arguments are cheaper to access, that's still a clear win. It's likely that there's other places where conversion to NullableDatum arrays would make sense, e.g. TupleTableSlots, but that's for another commit. Because the function call information is now variable-length allocations have to take the number of arguments into account. For heap allocations that can be done with SizeForFunctionCallInfoData(), for on-stack allocations there's a new LOCAL_FCINFO(name, nargs) macro that helps to allocate an appropriately sized and aligned variable. Some places with stack allocation function call information don't know the number of arguments at compile time, and currently variably sized stack allocations aren't allowed in postgres. Therefore allow for FUNC_MAX_ARGS space in these cases. They're not that common, so for now that seems acceptable. Because of the need to allocate FunctionCallInfo of the appropriate size, older extensions may need to update their code. To avoid subtle breakages, the FunctionCallInfoData struct has been renamed to FunctionCallInfoBaseData. Most code only references FunctionCallInfo, so that shouldn't cause much collateral damage. This change is also a prerequisite for more efficient expression JIT compilation (by allocating the function call information on the stack, allowing LLVM to optimize it away); previously the size of the call information caused problems inside LLVM's optimizer. Author: Andres Freund Reviewed-By: Tom Lane Discussion: https://postgr.es/m/20180605172952.x34m5uz6ju6enaem@alap3.anarazel.de
2019-01-26 23:17:52 +01:00
pergroup->transValue = datumCopy(fcinfo->args[1].value,
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
pertrans->transtypeByVal,
pertrans->transtypeLen);
pergroup->transValueIsNull = false;
pergroup->noTransValue = false;
MemoryContextSwitchTo(oldContext);
}
/*
Rename ExecAggTransReparent, and improve its documentation. The name of this function suggests that it ought to reparent R/W expanded objects to be children of the persistent aggcontext, instead of copying them. In fact it does no such thing, and if you try to make it do so you will see multiple regression failures. Rename it to the less-misleading ExecAggCopyTransValue, and add commentary about why that attractive-sounding optimization won't work. Also adjust comments at call sites, some of which were describing logic that has since been moved into ExecAggCopyTransValue. Discussion: https://postgr.es/m/3004282.1681930251@sss.pgh.pa.us
2023-04-24 19:01:33 +02:00
* Ensure that the new transition value is stored in the aggcontext,
* rather than the per-tuple context. This should be invoked only when
* we know (a) the transition data type is pass-by-reference, and (b)
* the newValue is distinct from the oldValue.
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
*
* NB: This can change the current memory context.
Rename ExecAggTransReparent, and improve its documentation. The name of this function suggests that it ought to reparent R/W expanded objects to be children of the persistent aggcontext, instead of copying them. In fact it does no such thing, and if you try to make it do so you will see multiple regression failures. Rename it to the less-misleading ExecAggCopyTransValue, and add commentary about why that attractive-sounding optimization won't work. Also adjust comments at call sites, some of which were describing logic that has since been moved into ExecAggCopyTransValue. Discussion: https://postgr.es/m/3004282.1681930251@sss.pgh.pa.us
2023-04-24 19:01:33 +02:00
*
* We copy the presented newValue into the aggcontext, except when the datum
* points to a R/W expanded object that is already a child of the aggcontext,
* in which case we need not copy. We then delete the oldValue, if not null.
*
* If the presented datum points to a R/W expanded object that is a child of
* some other context, ideally we would just reparent it under the aggcontext.
* Unfortunately, that doesn't work easily, and it wouldn't help anyway for
* aggregate-aware transfns. We expect that a transfn that deals in expanded
* objects and is aware of the memory management conventions for aggregate
* transition values will (1) on first call, return a R/W expanded object that
* is already in the right context, allowing us to do nothing here, and (2) on
* subsequent calls, modify and return that same object, so that control
* doesn't even reach here. However, if we have a generic transfn that
* returns a new R/W expanded object (probably in the per-tuple context),
* reparenting that result would cause problems. We'd pass that R/W object to
* the next invocation of the transfn, and then it would be at liberty to
* change or delete that object, and if it deletes it then our own attempt to
* delete the now-old transvalue afterwards would be a double free. We avoid
* this problem by forcing the stored transvalue to always be a flat
* non-expanded object unless the transfn is visibly doing aggregate-aware
* memory management. This is somewhat inefficient, but the best answer to
* that is to write a smarter transfn.
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
*/
Datum
Rename ExecAggTransReparent, and improve its documentation. The name of this function suggests that it ought to reparent R/W expanded objects to be children of the persistent aggcontext, instead of copying them. In fact it does no such thing, and if you try to make it do so you will see multiple regression failures. Rename it to the less-misleading ExecAggCopyTransValue, and add commentary about why that attractive-sounding optimization won't work. Also adjust comments at call sites, some of which were describing logic that has since been moved into ExecAggCopyTransValue. Discussion: https://postgr.es/m/3004282.1681930251@sss.pgh.pa.us
2023-04-24 19:01:33 +02:00
ExecAggCopyTransValue(AggState *aggstate, AggStatePerTrans pertrans,
Datum newValue, bool newValueIsNull,
Datum oldValue, bool oldValueIsNull)
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
{
Fix edge case leading to agg transitions skipping ExecAggTransReparent() calls. The code checking whether an aggregate transition value needs to be reparented into the current context has always only compared the transition return value with the previous transition value by datum, i.e. without regard for NULLness. This normally works, because when the transition function returns NULL (via fcinfo->isnull), it'll return a value that won't be the same as its input value. But there's no hard requirement that that's the case. And it turns out, it's possible to hit this case (see discussion or reproducers), leading to a non-null transition value not being reparented, followed by a crash caused by that. Instead of adding another comparison of NULLness, instead have ExecAggTransReparent() ensure that pergroup->transValue ends up as 0 when the new transition value is NULL. That avoids having to add an additional branch to the much more common cases of the transition function returning the old transition value (which is a pointer in this case), and when the new value is different, but not NULL. In branches since 69c3936a149, also deduplicate the reparenting code between the expression evaluation based transitions, and the path for ordered aggregates. Reported-By: Teodor Sigaev, Nikita Glukhov Author: Andres Freund Discussion: https://postgr.es/m/bd34e930-cfec-ea9b-3827-a8bc50891393@sigaev.ru Backpatch: 9.4-, this issue has existed since at least 7.4
2020-01-21 08:26:51 +01:00
Assert(newValue != oldValue);
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
if (!newValueIsNull)
{
MemoryContextSwitchTo(aggstate->curaggcontext->ecxt_per_tuple_memory);
if (DatumIsReadWriteExpandedObject(newValue,
false,
pertrans->transtypeLen) &&
MemoryContextGetParent(DatumGetEOHP(newValue)->eoh_context) == CurrentMemoryContext)
/* do nothing */ ;
else
newValue = datumCopy(newValue,
pertrans->transtypeByVal,
pertrans->transtypeLen);
}
Fix edge case leading to agg transitions skipping ExecAggTransReparent() calls. The code checking whether an aggregate transition value needs to be reparented into the current context has always only compared the transition return value with the previous transition value by datum, i.e. without regard for NULLness. This normally works, because when the transition function returns NULL (via fcinfo->isnull), it'll return a value that won't be the same as its input value. But there's no hard requirement that that's the case. And it turns out, it's possible to hit this case (see discussion or reproducers), leading to a non-null transition value not being reparented, followed by a crash caused by that. Instead of adding another comparison of NULLness, instead have ExecAggTransReparent() ensure that pergroup->transValue ends up as 0 when the new transition value is NULL. That avoids having to add an additional branch to the much more common cases of the transition function returning the old transition value (which is a pointer in this case), and when the new value is different, but not NULL. In branches since 69c3936a149, also deduplicate the reparenting code between the expression evaluation based transitions, and the path for ordered aggregates. Reported-By: Teodor Sigaev, Nikita Glukhov Author: Andres Freund Discussion: https://postgr.es/m/bd34e930-cfec-ea9b-3827-a8bc50891393@sigaev.ru Backpatch: 9.4-, this issue has existed since at least 7.4
2020-01-21 08:26:51 +01:00
else
{
/*
* Ensure that AggStatePerGroup->transValue ends up being 0, so
* callers can safely compare newValue/oldValue without having to
* check their respective nullness.
*/
newValue = (Datum) 0;
}
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
if (!oldValueIsNull)
{
if (DatumIsReadWriteExpandedObject(oldValue,
false,
pertrans->transtypeLen))
DeleteExpandedObject(oldValue);
else
pfree(DatumGetPointer(oldValue));
}
return newValue;
}
Improve performance of ORDER BY / DISTINCT aggregates ORDER BY / DISTINCT aggreagtes have, since implemented in Postgres, been executed by always performing a sort in nodeAgg.c to sort the tuples in the current group into the correct order before calling the transition function on the sorted tuples. This was not great as often there might be an index that could have provided pre-sorted input and allowed the transition functions to be called as the rows come in, rather than having to store them in a tuplestore in order to sort them once all the tuples for the group have arrived. Here we change the planner so it requests a path with a sort order which supports the most amount of ORDER BY / DISTINCT aggregate functions and add new code to the executor to allow it to support the processing of ORDER BY / DISTINCT aggregates where the tuples are already sorted in the correct order. Since there can be many ORDER BY / DISTINCT aggregates in any given query level, it's very possible that we can't find an order that suits all of these aggregates. The sort order that the planner chooses is simply the one that suits the most aggregate functions. We take the most strictly sorted variation of each order and see how many aggregate functions can use that, then we try again with the order of the remaining aggregates to see if another order would suit more aggregate functions. For example: SELECT agg(a ORDER BY a),agg2(a ORDER BY a,b) ... would request the sort order to be {a, b} because {a} is a subset of the sort order of {a,b}, but; SELECT agg(a ORDER BY a),agg2(a ORDER BY c) ... would just pick a plan ordered by {a} (we give precedence to aggregates which are earlier in the targetlist). SELECT agg(a ORDER BY a),agg2(a ORDER BY b),agg3(a ORDER BY b) ... would choose to order by {b} since two aggregates suit that vs just one that requires input ordered by {a}. Author: David Rowley Reviewed-by: Ronan Dunklau, James Coleman, Ranier Vilela, Richard Guo, Tom Lane Discussion: https://postgr.es/m/CAApHDvpHzfo92%3DR4W0%2BxVua3BUYCKMckWAmo-2t_KiXN-wYH%3Dw%40mail.gmail.com
2022-08-02 13:11:45 +02:00
/*
* ExecEvalPreOrderedDistinctSingle
* Returns true when the aggregate transition value Datum is distinct
* from the previous input Datum and returns false when the input Datum
* matches the previous input Datum.
*/
bool
ExecEvalPreOrderedDistinctSingle(AggState *aggstate, AggStatePerTrans pertrans)
{
Datum value = pertrans->transfn_fcinfo->args[1].value;
bool isnull = pertrans->transfn_fcinfo->args[1].isnull;
if (!pertrans->haslast ||
pertrans->lastisnull != isnull ||
Fix incorrect presorted DISTINCT aggregate if condition Here we fix a faulty "if" condition which failed to correctly handle two or more consecutive NULL transition values when checking if the new value is DISTINCT from the old value for presorted aggregates. Given a suitably non-strict aggregate transition function, a byref aggregate could cause a crash due to calling the type's equality function and passing along a (Datum) 0 value to test for equality, the equality function would then try to dereference that 0 Datum and segfault. For byval types, there'd have been no crash and the equality function would have seen that the two 0 Datums matched, which (only by chance) meant the calling code would have worked correctly. Here we ensure that we only call the equality function when neither of the input values are NULL. This code is all new as of 1349d2790, so no backpatch needed. Reported-by: Fujii Masao Discussion: https://postgr.es/m/860c6d6f-a3c5-3ae9-9da2-827177bede06@oss.nttdata.com
2023-02-13 08:38:37 +01:00
(!isnull && !DatumGetBool(FunctionCall2Coll(&pertrans->equalfnOne,
pertrans->aggCollation,
pertrans->lastdatum, value))))
Improve performance of ORDER BY / DISTINCT aggregates ORDER BY / DISTINCT aggreagtes have, since implemented in Postgres, been executed by always performing a sort in nodeAgg.c to sort the tuples in the current group into the correct order before calling the transition function on the sorted tuples. This was not great as often there might be an index that could have provided pre-sorted input and allowed the transition functions to be called as the rows come in, rather than having to store them in a tuplestore in order to sort them once all the tuples for the group have arrived. Here we change the planner so it requests a path with a sort order which supports the most amount of ORDER BY / DISTINCT aggregate functions and add new code to the executor to allow it to support the processing of ORDER BY / DISTINCT aggregates where the tuples are already sorted in the correct order. Since there can be many ORDER BY / DISTINCT aggregates in any given query level, it's very possible that we can't find an order that suits all of these aggregates. The sort order that the planner chooses is simply the one that suits the most aggregate functions. We take the most strictly sorted variation of each order and see how many aggregate functions can use that, then we try again with the order of the remaining aggregates to see if another order would suit more aggregate functions. For example: SELECT agg(a ORDER BY a),agg2(a ORDER BY a,b) ... would request the sort order to be {a, b} because {a} is a subset of the sort order of {a,b}, but; SELECT agg(a ORDER BY a),agg2(a ORDER BY c) ... would just pick a plan ordered by {a} (we give precedence to aggregates which are earlier in the targetlist). SELECT agg(a ORDER BY a),agg2(a ORDER BY b),agg3(a ORDER BY b) ... would choose to order by {b} since two aggregates suit that vs just one that requires input ordered by {a}. Author: David Rowley Reviewed-by: Ronan Dunklau, James Coleman, Ranier Vilela, Richard Guo, Tom Lane Discussion: https://postgr.es/m/CAApHDvpHzfo92%3DR4W0%2BxVua3BUYCKMckWAmo-2t_KiXN-wYH%3Dw%40mail.gmail.com
2022-08-02 13:11:45 +02:00
{
Fix pfree issue in presorted DISTINCT aggregate code The logic in this area was recently changed in 7da51590e, however, in that commit, I neglected to consider that the conditions in which we should pfree the old Datum needed to be updated after that change. This could result in trying to pfree a NULL value, as was demonstrated by Alexander Lakhin. Reported-by: Alexander Lakhin Discussion: https://postgr.es/m/4103db46-d888-6d1d-e88d-87c21ed99472@gmail.com
2023-02-13 11:38:21 +01:00
if (pertrans->haslast && !pertrans->inputtypeByVal &&
!pertrans->lastisnull)
Improve performance of ORDER BY / DISTINCT aggregates ORDER BY / DISTINCT aggreagtes have, since implemented in Postgres, been executed by always performing a sort in nodeAgg.c to sort the tuples in the current group into the correct order before calling the transition function on the sorted tuples. This was not great as often there might be an index that could have provided pre-sorted input and allowed the transition functions to be called as the rows come in, rather than having to store them in a tuplestore in order to sort them once all the tuples for the group have arrived. Here we change the planner so it requests a path with a sort order which supports the most amount of ORDER BY / DISTINCT aggregate functions and add new code to the executor to allow it to support the processing of ORDER BY / DISTINCT aggregates where the tuples are already sorted in the correct order. Since there can be many ORDER BY / DISTINCT aggregates in any given query level, it's very possible that we can't find an order that suits all of these aggregates. The sort order that the planner chooses is simply the one that suits the most aggregate functions. We take the most strictly sorted variation of each order and see how many aggregate functions can use that, then we try again with the order of the remaining aggregates to see if another order would suit more aggregate functions. For example: SELECT agg(a ORDER BY a),agg2(a ORDER BY a,b) ... would request the sort order to be {a, b} because {a} is a subset of the sort order of {a,b}, but; SELECT agg(a ORDER BY a),agg2(a ORDER BY c) ... would just pick a plan ordered by {a} (we give precedence to aggregates which are earlier in the targetlist). SELECT agg(a ORDER BY a),agg2(a ORDER BY b),agg3(a ORDER BY b) ... would choose to order by {b} since two aggregates suit that vs just one that requires input ordered by {a}. Author: David Rowley Reviewed-by: Ronan Dunklau, James Coleman, Ranier Vilela, Richard Guo, Tom Lane Discussion: https://postgr.es/m/CAApHDvpHzfo92%3DR4W0%2BxVua3BUYCKMckWAmo-2t_KiXN-wYH%3Dw%40mail.gmail.com
2022-08-02 13:11:45 +02:00
pfree(DatumGetPointer(pertrans->lastdatum));
pertrans->haslast = true;
if (!isnull)
{
MemoryContext oldContext;
oldContext = MemoryContextSwitchTo(aggstate->curaggcontext->ecxt_per_tuple_memory);
pertrans->lastdatum = datumCopy(value, pertrans->inputtypeByVal,
pertrans->inputtypeLen);
MemoryContextSwitchTo(oldContext);
}
else
pertrans->lastdatum = (Datum) 0;
pertrans->lastisnull = isnull;
return true;
}
return false;
}
/*
* ExecEvalPreOrderedDistinctMulti
* Returns true when the aggregate input is distinct from the previous
* input and returns false when the input matches the previous input.
*/
bool
ExecEvalPreOrderedDistinctMulti(AggState *aggstate, AggStatePerTrans pertrans)
{
ExprContext *tmpcontext = aggstate->tmpcontext;
for (int i = 0; i < pertrans->numTransInputs; i++)
{
pertrans->sortslot->tts_values[i] = pertrans->transfn_fcinfo->args[i + 1].value;
pertrans->sortslot->tts_isnull[i] = pertrans->transfn_fcinfo->args[i + 1].isnull;
}
ExecClearTuple(pertrans->sortslot);
pertrans->sortslot->tts_nvalid = pertrans->numInputs;
ExecStoreVirtualTuple(pertrans->sortslot);
tmpcontext->ecxt_outertuple = pertrans->sortslot;
tmpcontext->ecxt_innertuple = pertrans->uniqslot;
if (!pertrans->haslast ||
!ExecQual(pertrans->equalfnMulti, tmpcontext))
{
if (pertrans->haslast)
ExecClearTuple(pertrans->uniqslot);
pertrans->haslast = true;
ExecCopySlot(pertrans->uniqslot, pertrans->sortslot);
return true;
}
return false;
}
Expression evaluation based aggregate transition invocation. Previously aggregate transition and combination functions were invoked by special case code in nodeAgg.c, evaluating input and filters separately using the expression evaluation machinery. That turns out to not be great for performance for several reasons: - repeated expression evaluations have some cost - the transition functions invocations are poorly predicted, as commonly there are multiple aggregates in a query, resulting in the same call-stack invoking different functions. - filter and input computation had to be done separately - the special case code made it hard to implement JITing of the whole transition function invocation Address this by building one large expression that computes input, evaluates filters, and invokes transition functions. This leads to moderate speedups in queries bottlenecked by aggregate computations, and enables large speedups for similar cases once JITing is done. There's potential for further improvement: - It'd be nice if we could simplify the somewhat expensive aggstate->all_pergroups lookups. - right now there's still an advance_transition_function invocation in nodeAgg.c, leading to some code duplication. Author: Andres Freund Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
2018-01-09 22:25:38 +01:00
/*
* Invoke ordered transition function, with a datum argument.
*/
void
ExecEvalAggOrderedTransDatum(ExprState *state, ExprEvalStep *op,
ExprContext *econtext)
{
AggStatePerTrans pertrans = op->d.agg_trans.pertrans;
int setno = op->d.agg_trans.setno;
tuplesort_putdatum(pertrans->sortstates[setno],
*op->resvalue, *op->resnull);
}
/*
* Invoke ordered transition function, with a tuple argument.
*/
void
ExecEvalAggOrderedTransTuple(ExprState *state, ExprEvalStep *op,
ExprContext *econtext)
{
AggStatePerTrans pertrans = op->d.agg_trans.pertrans;
int setno = op->d.agg_trans.setno;
ExecClearTuple(pertrans->sortslot);
pertrans->sortslot->tts_nvalid = pertrans->numInputs;
ExecStoreVirtualTuple(pertrans->sortslot);
tuplesort_puttupleslot(pertrans->sortstates[setno], pertrans->sortslot);
}
expression eval: Reduce number of steps for agg transition invocations. Do so by combining the various steps that are part of aggregate transition function invocation into one larger step. As some of the current steps are only necessary for some aggregates, have one variant of the aggregate transition step for each possible combination. To avoid further manual copies of code in the different transition step implementations, move most of the code into helper functions marked as "always inline". The benefit of this change is an increase in performance when aggregating lots of rows. This comes in part due to the reduced number of indirect jumps due to the reduced number of steps, and in part by reducing redundant setup code across steps. This mainly benefits interpreted execution, but the code generated by JIT is also improved a bit. As a nice side-effect it also ends up making the code a bit simpler. A small additional optimization is removing the need to set aggstate->curaggcontext before calling ExecAggInitGroup, choosing to instead passign curaggcontext as an argument. It was, in contrast to other aggregate related functions, only needed to fetch a memory context to copy the transition value into. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de https://postgr.es/m/5c371df7cee903e8cd4c685f90c6c72086d3a2dc.camel@j-davis.com
2020-02-24 23:39:22 +01:00
/* implementation of transition function invocation for byval types */
static pg_attribute_always_inline void
ExecAggPlainTransByVal(AggState *aggstate, AggStatePerTrans pertrans,
AggStatePerGroup pergroup,
ExprContext *aggcontext, int setno)
{
FunctionCallInfo fcinfo = pertrans->transfn_fcinfo;
MemoryContext oldContext;
Datum newVal;
/* cf. select_current_set() */
aggstate->curaggcontext = aggcontext;
aggstate->current_set = setno;
/* set up aggstate->curpertrans for AggGetAggref() */
aggstate->curpertrans = pertrans;
/* invoke transition function in per-tuple context */
oldContext = MemoryContextSwitchTo(aggstate->tmpcontext->ecxt_per_tuple_memory);
fcinfo->args[0].value = pergroup->transValue;
fcinfo->args[0].isnull = pergroup->transValueIsNull;
fcinfo->isnull = false; /* just in case transfn doesn't set it */
newVal = FunctionCallInvoke(fcinfo);
pergroup->transValue = newVal;
pergroup->transValueIsNull = fcinfo->isnull;
MemoryContextSwitchTo(oldContext);
}
/* implementation of transition function invocation for byref types */
static pg_attribute_always_inline void
ExecAggPlainTransByRef(AggState *aggstate, AggStatePerTrans pertrans,
AggStatePerGroup pergroup,
ExprContext *aggcontext, int setno)
{
FunctionCallInfo fcinfo = pertrans->transfn_fcinfo;
MemoryContext oldContext;
Datum newVal;
/* cf. select_current_set() */
aggstate->curaggcontext = aggcontext;
aggstate->current_set = setno;
/* set up aggstate->curpertrans for AggGetAggref() */
aggstate->curpertrans = pertrans;
/* invoke transition function in per-tuple context */
oldContext = MemoryContextSwitchTo(aggstate->tmpcontext->ecxt_per_tuple_memory);
fcinfo->args[0].value = pergroup->transValue;
fcinfo->args[0].isnull = pergroup->transValueIsNull;
fcinfo->isnull = false; /* just in case transfn doesn't set it */
newVal = FunctionCallInvoke(fcinfo);
/*
* For pass-by-ref datatype, must copy the new value into aggcontext and
* free the prior transValue. But if transfn returned a pointer to its
Rename ExecAggTransReparent, and improve its documentation. The name of this function suggests that it ought to reparent R/W expanded objects to be children of the persistent aggcontext, instead of copying them. In fact it does no such thing, and if you try to make it do so you will see multiple regression failures. Rename it to the less-misleading ExecAggCopyTransValue, and add commentary about why that attractive-sounding optimization won't work. Also adjust comments at call sites, some of which were describing logic that has since been moved into ExecAggCopyTransValue. Discussion: https://postgr.es/m/3004282.1681930251@sss.pgh.pa.us
2023-04-24 19:01:33 +02:00
* first input, we don't need to do anything.
expression eval: Reduce number of steps for agg transition invocations. Do so by combining the various steps that are part of aggregate transition function invocation into one larger step. As some of the current steps are only necessary for some aggregates, have one variant of the aggregate transition step for each possible combination. To avoid further manual copies of code in the different transition step implementations, move most of the code into helper functions marked as "always inline". The benefit of this change is an increase in performance when aggregating lots of rows. This comes in part due to the reduced number of indirect jumps due to the reduced number of steps, and in part by reducing redundant setup code across steps. This mainly benefits interpreted execution, but the code generated by JIT is also improved a bit. As a nice side-effect it also ends up making the code a bit simpler. A small additional optimization is removing the need to set aggstate->curaggcontext before calling ExecAggInitGroup, choosing to instead passign curaggcontext as an argument. It was, in contrast to other aggregate related functions, only needed to fetch a memory context to copy the transition value into. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de https://postgr.es/m/5c371df7cee903e8cd4c685f90c6c72086d3a2dc.camel@j-davis.com
2020-02-24 23:39:22 +01:00
*
* It's safe to compare newVal with pergroup->transValue without regard
Rename ExecAggTransReparent, and improve its documentation. The name of this function suggests that it ought to reparent R/W expanded objects to be children of the persistent aggcontext, instead of copying them. In fact it does no such thing, and if you try to make it do so you will see multiple regression failures. Rename it to the less-misleading ExecAggCopyTransValue, and add commentary about why that attractive-sounding optimization won't work. Also adjust comments at call sites, some of which were describing logic that has since been moved into ExecAggCopyTransValue. Discussion: https://postgr.es/m/3004282.1681930251@sss.pgh.pa.us
2023-04-24 19:01:33 +02:00
* for either being NULL, because ExecAggCopyTransValue takes care to set
expression eval: Reduce number of steps for agg transition invocations. Do so by combining the various steps that are part of aggregate transition function invocation into one larger step. As some of the current steps are only necessary for some aggregates, have one variant of the aggregate transition step for each possible combination. To avoid further manual copies of code in the different transition step implementations, move most of the code into helper functions marked as "always inline". The benefit of this change is an increase in performance when aggregating lots of rows. This comes in part due to the reduced number of indirect jumps due to the reduced number of steps, and in part by reducing redundant setup code across steps. This mainly benefits interpreted execution, but the code generated by JIT is also improved a bit. As a nice side-effect it also ends up making the code a bit simpler. A small additional optimization is removing the need to set aggstate->curaggcontext before calling ExecAggInitGroup, choosing to instead passign curaggcontext as an argument. It was, in contrast to other aggregate related functions, only needed to fetch a memory context to copy the transition value into. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de https://postgr.es/m/5c371df7cee903e8cd4c685f90c6c72086d3a2dc.camel@j-davis.com
2020-02-24 23:39:22 +01:00
* transValue to 0 when NULL. Otherwise we could end up accidentally not
* reparenting, when the transValue has the same numerical value as
* newValue, despite being NULL. This is a somewhat hot path, making it
* undesirable to instead solve this with another branch for the common
* case of the transition function returning its (modified) input
* argument.
*/
if (DatumGetPointer(newVal) != DatumGetPointer(pergroup->transValue))
Rename ExecAggTransReparent, and improve its documentation. The name of this function suggests that it ought to reparent R/W expanded objects to be children of the persistent aggcontext, instead of copying them. In fact it does no such thing, and if you try to make it do so you will see multiple regression failures. Rename it to the less-misleading ExecAggCopyTransValue, and add commentary about why that attractive-sounding optimization won't work. Also adjust comments at call sites, some of which were describing logic that has since been moved into ExecAggCopyTransValue. Discussion: https://postgr.es/m/3004282.1681930251@sss.pgh.pa.us
2023-04-24 19:01:33 +02:00
newVal = ExecAggCopyTransValue(aggstate, pertrans,
newVal, fcinfo->isnull,
pergroup->transValue,
pergroup->transValueIsNull);
expression eval: Reduce number of steps for agg transition invocations. Do so by combining the various steps that are part of aggregate transition function invocation into one larger step. As some of the current steps are only necessary for some aggregates, have one variant of the aggregate transition step for each possible combination. To avoid further manual copies of code in the different transition step implementations, move most of the code into helper functions marked as "always inline". The benefit of this change is an increase in performance when aggregating lots of rows. This comes in part due to the reduced number of indirect jumps due to the reduced number of steps, and in part by reducing redundant setup code across steps. This mainly benefits interpreted execution, but the code generated by JIT is also improved a bit. As a nice side-effect it also ends up making the code a bit simpler. A small additional optimization is removing the need to set aggstate->curaggcontext before calling ExecAggInitGroup, choosing to instead passign curaggcontext as an argument. It was, in contrast to other aggregate related functions, only needed to fetch a memory context to copy the transition value into. Author: Andres Freund Discussion: https://postgr.es/m/20191023163849.sosqbfs5yenocez3@alap3.anarazel.de https://postgr.es/m/5c371df7cee903e8cd4c685f90c6c72086d3a2dc.camel@j-davis.com
2020-02-24 23:39:22 +01:00
pergroup->transValue = newVal;
pergroup->transValueIsNull = fcinfo->isnull;
MemoryContextSwitchTo(oldContext);
}