Clean up some really grotty coding in catcache.c, improve hashing

performance in catcache lookups.

src/backend/access/hash/hashfunc.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/access/hash/hashfunc.c,v 1.23 2000/01/26 05:55:55 momjian Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/access/hash/hashfunc.c,v 1.24 2000/02/21 03:36:46 tgl Exp $
  *
  * NOTES
  *	  These functions are stored in pg_amproc.	For each operator class
@@ -146,8 +146,24 @@ hashoidvector(Oid *key)
 	int			i;
 	uint32		result = 0;
 
-	for (i = 0; i < INDEX_MAX_KEYS; i++)
+	for (i = INDEX_MAX_KEYS; --i >= 0; )
-		result = result ^ (~(uint32) key[i]);
+		result = (result << 1) ^ (~(uint32) key[i]);
+	return result;
+}
+
+/*
+ * Note: hashint2vector currently can't be used as a user hash table
+ * hash function, because it has no pg_proc entry.  We only need it
+ * for catcache indexing.
+ */
+uint32
+hashint2vector(int16 *key)
+{
+	int			i;
+	uint32		result = 0;
+
+	for (i = INDEX_MAX_KEYS; --i >= 0; )
+		result = (result << 1) ^ (~(uint32) key[i]);
 	return result;
 }
 
@@ -158,13 +174,10 @@ hashoidvector(Oid *key)
 uint32
 hashchar(char key)
 {
-	int			len;
 	uint32		h;
 
-	h = 0;
-	len = sizeof(char);
 	/* Convert char to integer */
-	h = h * PRIME1 ^ (key - ' ');
+	h = (key - ' ');
 	h %= PRIME2;
 
 	return h;

src/backend/utils/adt/int.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/utils/adt/int.c,v 1.32 2000/01/26 05:57:14 momjian Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/utils/adt/int.c,v 1.33 2000/02/21 03:36:48 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -144,6 +144,17 @@ int2vectorout(int16 *int2Array)
 	return result;
 }
 
+/*
+ * We don't have a complete set of int2vector support routines,
+ * but we need int2vectoreq for catcache indexing.
+ */
+bool
+int2vectoreq(int16 *arg1, int16 *arg2)
+{
+	return (bool) (memcmp(arg1, arg2, INDEX_MAX_KEYS * sizeof(int16)) == 0);
+}
+
+
 /*
  *		int44in			- converts "num num ..." to internal form
  *
@@ -169,7 +180,7 @@ int44in(char *input_string)
 }
 
 /*
- *		int2vectorout		- converts internal form to "num num ..."
+ *		int44out		- converts internal form to "num num ..."
  */
 char *
 int44out(int32 *an_array)
@@ -489,13 +500,6 @@ int42ge(int32 arg1, int32 arg2)
 	return arg1 >= arg2;
 }
 
-
-bool
-keyfirsteq(int16 *arg1, int16 arg2)
-{
-	return *arg1 == arg2;
-}
-
 /*
  *		int[24]pl		- returns arg1 + arg2
  *		int[24]mi		- returns arg1 - arg2

src/backend/utils/cache/catcache.c

@@ -8,12 +8,14 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/utils/cache/catcache.c,v 1.61 2000/02/18 09:28:53 inoue Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/utils/cache/catcache.c,v 1.62 2000/02/21 03:36:49 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
 #include "postgres.h"
+
 #include "access/genam.h"
+#include "access/hash.h"
 #include "access/heapam.h"
 #include "access/valid.h"
 #include "catalog/pg_operator.h"
@@ -28,10 +30,11 @@
 static void CatCacheRemoveCTup(CatCache *cache, Dlelem *e);
 static Index CatalogCacheComputeHashIndex(struct catcache * cacheInP);
 static Index CatalogCacheComputeTupleHashIndex(struct catcache * cacheInOutP,
-								  Relation relation, HeapTuple tuple);
+											   Relation relation,
+											   HeapTuple tuple);
 static void CatalogCacheInitializeCache(struct catcache * cache,
-							Relation relation);
+										Relation relation);
-static long comphash(long l, char *v);
+static uint32 cc_hashname(NameData *n);
 
 /* ----------------
  *		variables, macros and other stuff
@@ -63,6 +66,7 @@ GlobalMemory CacheCxt;			/* context in which caches are allocated */
 /* ----------------
  *		EQPROC is used in CatalogCacheInitializeCache to find the equality
  *		functions for system types that are used as cache key fields.
+ *		See also GetCCHashFunc, which should support the same set of types.
  *
  *		XXX this should be replaced by catalog lookups,
  *		but that seems to pose considerable risk of circularity...
@@ -70,7 +74,7 @@ GlobalMemory CacheCxt;			/* context in which caches are allocated */
  */
 static const Oid eqproc[] = {
 	F_BOOLEQ, InvalidOid, F_CHAREQ, F_NAMEEQ, InvalidOid,
-	F_INT2EQ, F_KEYFIRSTEQ, F_INT4EQ, F_OIDEQ, F_TEXTEQ,
+	F_INT2EQ, F_INT2VECTOREQ, F_INT4EQ, F_OIDEQ, F_TEXTEQ,
 	F_OIDEQ, InvalidOid, InvalidOid, InvalidOid, F_OIDVECTOREQ
 };
 
@@ -80,6 +84,54 @@ static const Oid eqproc[] = {
  *					internal support functions
  * ----------------------------------------------------------------
  */
+
+static CCHashFunc
+GetCCHashFunc(Oid keytype)
+{
+	switch (keytype)
+	{
+		case BOOLOID:
+		case CHAROID:
+			return (CCHashFunc) hashchar;
+		case NAMEOID:
+			return (CCHashFunc) cc_hashname;
+		case INT2OID:
+			return (CCHashFunc) hashint2;
+		case INT2VECTOROID:
+			return (CCHashFunc) hashint2vector;
+		case INT4OID:
+			return (CCHashFunc) hashint4;
+		case TEXTOID:
+			return (CCHashFunc) hashtext;
+		case REGPROCOID:
+		case OIDOID:
+			return (CCHashFunc) hashoid;
+		case OIDVECTOROID:
+			return (CCHashFunc) hashoidvector;
+		default:
+			elog(FATAL, "GetCCHashFunc: type %u unsupported as catcache key",
+				 keytype);
+			return NULL;
+	}
+}
+
+static uint32
+cc_hashname(NameData *n)
+{
+	/*
+	 * We need our own variant of hashname because we want to accept
+	 * null-terminated C strings as search values for name fields.
+	 * So, we have to make sure the data is correctly padded before
+	 * we compute the hash value.
+	 */
+	NameData	my_n;
+
+	namestrcpy(&my_n, NameStr(*n));
+
+	return hashname(&my_n);
+}
+
+
 /* --------------------------------
  *		CatalogCacheInitializeCache
  * --------------------------------
@@ -190,31 +242,20 @@ CatalogCacheInitializeCache(struct catcache * cache,
 
 		if (cache->cc_key[i] > 0)
 		{
+			Oid		keytype = tupdesc->attrs[cache->cc_key[i] - 1]->atttypid;
 
-			/*
+			cache->cc_hashfunc[i] = GetCCHashFunc(keytype);
-			 * Yoiks.  The implementation of the hashing code and the
-			 * implementation of int2vector's are at loggerheads.  The right
-			 * thing to do is to throw out the implementation of int2vector's
-			 * altogether; until that happens, we do the right thing here
-			 * to guarantee that the hash key generator doesn't try to
-			 * dereference an int2 by mistake.
-			 */
 
-			if (tupdesc->attrs[cache->cc_key[i] - 1]->atttypid == INT2VECTOROID)
+			/* If GetCCHashFunc liked the type, safe to index into eqproc[] */
-				cache->cc_klen[i] = sizeof(short);
+			cache->cc_skey[i].sk_procedure = EQPROC(keytype);
-			else
-				cache->cc_klen[i] = tupdesc->attrs[cache->cc_key[i] - 1]->attlen;
-
-			cache->cc_skey[i].sk_procedure = EQPROC(tupdesc->attrs[cache->cc_key[i] - 1]->atttypid);
 
 			fmgr_info(cache->cc_skey[i].sk_procedure,
 					  &cache->cc_skey[i].sk_func);
 			cache->cc_skey[i].sk_nargs = cache->cc_skey[i].sk_func.fn_nargs;
 
-			CACHE5_elog(DEBUG, "CatalogCacheInit %s %d %d %x",
+			CACHE4_elog(DEBUG, "CatalogCacheInit %s %d %x",
 						RelationGetRelationName(relation),
 						i,
-						tupdesc->attrs[cache->cc_key[i] - 1]->attlen,
 						cache);
 		}
 	}
@@ -255,53 +296,6 @@ CatalogCacheInitializeCache(struct catcache * cache,
 	MemoryContextSwitchTo(oldcxt);
 }
 
-/* ----------------
- * comphash
- *		Compute a hash value, somehow.
- *
- * XXX explain algorithm here.
- *
- * l is length of the attribute value, v
- * v is the attribute value ("Datum")
- * ----------------
- */
-static long
-comphash(long l, char *v)
-{
-	long		i;
-	NameData	n;
-
-	CACHE3_elog(DEBUG, "comphash (%d,%x)", l, v);
-
-	switch (l)
-	{
-		case 1:
-		case 2:
-		case 4:
-			return (long) v;
-	}
-
-	if (l == NAMEDATALEN)
-	{
-
-		/*
-		 * if it's a name, make sure that the values are null-padded.
-		 *
-		 * Note that this other fixed-length types can also have the same
-		 * typelen so this may break them	  - XXX
-		 */
-		namestrcpy(&n, v);
-		v = NameStr(n);
-	}
-	else if (l < 0)
-		l = VARSIZE(v);
-
-	i = 0;
-	while (l--)
-		i += *v++;
-	return i;
-}
-
 /* --------------------------------
  *		CatalogCacheComputeHashIndex
  * --------------------------------
@@ -309,40 +303,37 @@ comphash(long l, char *v)
 static Index
 CatalogCacheComputeHashIndex(struct catcache * cacheInP)
 {
-	Index		hashIndex;
+	uint32		hashIndex = 0;
 
-	hashIndex = 0x0;
+	CACHE4_elog(DEBUG, "CatalogCacheComputeHashIndex %s %d %x",
-	CACHE6_elog(DEBUG, "CatalogCacheComputeHashIndex %s %d %d %d %x",
 				cacheInP->cc_relname,
 				cacheInP->cc_nkeys,
-				cacheInP->cc_klen[0],
-				cacheInP->cc_klen[1],
 				cacheInP);
 
 	switch (cacheInP->cc_nkeys)
 	{
 		case 4:
-			hashIndex ^= comphash(cacheInP->cc_klen[3],
+			hashIndex ^=
-						 (char *) cacheInP->cc_skey[3].sk_argument) << 9;
+				(*cacheInP->cc_hashfunc[3])(cacheInP->cc_skey[3].sk_argument) << 9;
 			/* FALLTHROUGH */
 		case 3:
-			hashIndex ^= comphash(cacheInP->cc_klen[2],
+			hashIndex ^=
-						 (char *) cacheInP->cc_skey[2].sk_argument) << 6;
+				(*cacheInP->cc_hashfunc[2])(cacheInP->cc_skey[2].sk_argument) << 6;
 			/* FALLTHROUGH */
 		case 2:
-			hashIndex ^= comphash(cacheInP->cc_klen[1],
+			hashIndex ^=
-						 (char *) cacheInP->cc_skey[1].sk_argument) << 3;
+				(*cacheInP->cc_hashfunc[1])(cacheInP->cc_skey[1].sk_argument) << 3;
 			/* FALLTHROUGH */
 		case 1:
-			hashIndex ^= comphash(cacheInP->cc_klen[0],
+			hashIndex ^=
-							  (char *) cacheInP->cc_skey[0].sk_argument);
+				(*cacheInP->cc_hashfunc[0])(cacheInP->cc_skey[0].sk_argument);
 			break;
 		default:
 			elog(FATAL, "CCComputeHashIndex: %d cc_nkeys", cacheInP->cc_nkeys);
 			break;
 	}
-	hashIndex %= cacheInP->cc_size;
+	hashIndex %= (uint32) cacheInP->cc_size;
-	return hashIndex;
+	return (Index) hashIndex;
 }
 
 /* --------------------------------
@@ -645,8 +636,8 @@ do { \
 		cp->relationId, cp->id, cp->cc_nkeys, cp->cc_size); \
 	for (i = 0; i < nkeys; i += 1) \
 	{ \
-		elog(DEBUG, "InitSysCache: key=%d len=%d skey=[%d %d %d %d]\n", \
+		elog(DEBUG, "InitSysCache: key=%d skey=[%d %d %d %d]\n", \
-			 cp->cc_key[i], cp->cc_klen[i], \
+			 cp->cc_key[i], \
 			 cp->cc_skey[i].sk_flags, \
 			 cp->cc_skey[i].sk_attno, \
 			 cp->cc_skey[i].sk_procedure, \
@@ -742,7 +733,8 @@ InitSysCache(char *relname,
 	cp->cc_iscanfunc = iScanfuncP;
 
 	/* ----------------
-	 *	initialize the cache's key information
+	 *	partially initialize the cache's key information
+	 *	CatalogCacheInitializeCache() will do the rest
 	 * ----------------
 	 */
 	for (i = 0; i < nkeys; ++i)
@@ -756,15 +748,7 @@ InitSysCache(char *relname,
 				elog(FATAL, "InitSysCache: called with %d key[%d]", key[i], i);
 			else
 			{
-				cp->cc_klen[i] = sizeof(Oid);
+				cp->cc_hashfunc[i] = GetCCHashFunc(OIDOID);
-
-				/*
-				 * ScanKeyEntryData and struct skey are equivalent. It
-				 * looks like a move was made to obsolete struct skey, but
-				 * it didn't reach this file.  Someday we should clean up
-				 * this code and consolidate to ScanKeyEntry - mer 10 Nov
-				 * 1991
-				 */
 				ScanKeyEntryInitialize(&cp->cc_skey[i],
 									   (bits16) 0,
 									   (AttrNumber) key[i],

src/include/access/hash.h

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2000, PostgreSQL, Inc
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: hash.h,v 1.30 2000/01/26 05:57:50 momjian Exp $
+ * $Id: hash.h,v 1.31 2000/02/21 03:36:51 tgl Exp $
  *
  * NOTES
  *		modeled after Margo Seltzer's hash implementation for unix.
@@ -270,6 +270,7 @@ extern uint32 hashfloat4(float32 keyp);
 extern uint32 hashfloat8(float64 keyp);
 extern uint32 hashoid(Oid key);
 extern uint32 hashoidvector(Oid *key);
+extern uint32 hashint2vector(int16 *key);
 extern uint32 hashchar(char key);
 extern uint32 hashtext(struct varlena * key);
 extern uint32 hashname(NameData *n);

src/include/catalog/pg_proc.h

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2000, PostgreSQL, Inc
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: pg_proc.h,v 1.123 2000/02/17 03:39:48 tgl Exp $
+ * $Id: pg_proc.h,v 1.124 2000/02/21 03:36:55 tgl Exp $
  *
  * NOTES
  *	  The script catalog/genbki.sh reads this file and generates .bki
@@ -661,8 +661,8 @@ DATA(insert OID = 313 (  i2toi4			   PGUID 11 f t t 1 f  23  "21" 100 0 0 100  i
 DESCR("convert int2 to int4");
 DATA(insert OID = 314 (  i4toi2			   PGUID 11 f t t 1 f  21  "23" 100 0 0 100  i4toi2 - ));
 DESCR("convert int4 to int2");
-DATA(insert OID = 315 (  keyfirsteq		   PGUID 11 f t f 2 f  16	"0 21" 100 0 0 100	keyfirsteq - ));
+DATA(insert OID = 315 (  int2vectoreq	   PGUID 11 f t t 2 f  16  "22 22" 100 0 0 100	int2vectoreq - ));
-DESCR("");
+DESCR("equal");
 DATA(insert OID = 316 (  i4tod			   PGUID 11 f t t 1 f 701  "23" 100 0 0 100  i4tod - ));
 DESCR("convert int4 to float8");
 DATA(insert OID = 317 (  dtoi4			   PGUID 11 f t t 1 f  23 "701" 100 0 0 100  dtoi4 - ));

src/include/catalog/pg_type.h

@@ -8,7 +8,7 @@
  * Portions Copyright (c) 1996-2000, PostgreSQL, Inc
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: pg_type.h,v 1.82 2000/02/20 06:28:41 tgl Exp $
+ * $Id: pg_type.h,v 1.83 2000/02/21 03:36:57 tgl Exp $
  *
  * NOTES
  *	  the genbki.sh script reads this file and generates .bki
@@ -177,13 +177,6 @@ DESCR("-32 thousand to 32 thousand, 2-byte storage");
 
 DATA(insert OID = 22 (	int2vector PGUID INDEX_MAX_KEYS*2 -1 f b t \054 0  21 int2vectorin int2vectorout int2vectorin int2vectorout i _null_ ));
 DESCR("array of INDEX_MAX_KEYS int2 integers, used in system tables");
-/*
- * XXX -- the implementation of int2vector's in postgres is a hack, and will
- *		  go away someday.	until that happens, there is a case (in the
- *		  catalog cache management code) where we need to step gingerly
- *		  over piles of int2vector's on the sidewalk.  in order to do so, we
- *		  need the OID of the int2vector row from pg_type.
- */
 #define INT2VECTOROID	22
 
 DATA(insert OID = 23 (	int4	   PGUID  4  10 t b t \054 0   0 int4in int4out int4in int4out i _null_ ));
@@ -216,6 +209,8 @@ DESCR("command identifier type, sequence in transaction id");
 
 DATA(insert OID = 30 (	oidvector  PGUID INDEX_MAX_KEYS*4 -1 f b t \054 0  26 oidvectorin oidvectorout oidvectorin oidvectorout i _null_ ));
 DESCR("array of INDEX_MAX_KEYS oids, used in system tables");
+#define OIDVECTOROID	30
+
 DATA(insert OID = 32 (	SET		   PGUID -1  -1 f b t \054 0   0 textin textout textin textout i _null_ ));
 DESCR("set of tuples");
 

src/include/utils/builtins.h

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2000, PostgreSQL, Inc
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: builtins.h,v 1.102 2000/02/16 17:26:25 thomas Exp $
+ * $Id: builtins.h,v 1.103 2000/02/21 03:36:59 tgl Exp $
  *
  * NOTES
  *	  This should normally only be included by fmgr.h.
@@ -75,6 +75,7 @@ extern int32 int2in(char *num);
 extern char *int2out(int16 sh);
 extern int16 *int2vectorin(char *shs);
 extern char *int2vectorout(int16 *shs);
+extern bool int2vectoreq(int16 *arg1, int16 *arg2);
 extern int32 *int44in(char *input_string);
 extern char *int44out(int32 *an_array);
 extern int32 int4in(char *num);
@@ -109,7 +110,6 @@ extern bool int42lt(int32 arg1, int32 arg2);
 extern bool int42le(int32 arg1, int32 arg2);
 extern bool int42gt(int32 arg1, int32 arg2);
 extern bool int42ge(int32 arg1, int32 arg2);
-extern bool keyfirsteq(int16 *arg1, int16 arg2);
 extern int32 int4um(int32 arg);
 extern int32 int4pl(int32 arg1, int32 arg2);
 extern int32 int4mi(int32 arg1, int32 arg2);

src/include/utils/catcache.h

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2000, PostgreSQL, Inc
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: catcache.h,v 1.19 2000/01/26 05:58:37 momjian Exp $
+ * $Id: catcache.h,v 1.20 2000/02/21 03:36:59 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -38,6 +38,8 @@ typedef struct catctup
 #define NCCBUCK 500				/* CatCache buckets */
 #define MAXTUP 300				/* Maximum # of tuples cached per cache */
 
+typedef uint32 (* CCHashFunc) (Datum);
+
 typedef struct catcache
 {
 	Oid			relationId;
@@ -52,8 +54,8 @@ typedef struct catcache
 	short		cc_maxtup;		/* max # of tuples allowed (LRU) */
 	short		cc_nkeys;
 	short		cc_size;
-	short		cc_key[4];
+	short		cc_key[4];		/* AttrNumber of each key */
-	short		cc_klen[4];
+	CCHashFunc	cc_hashfunc[4];	/* hash function to use for each key */
 	ScanKeyData cc_skey[4];
 	struct catcache *cc_next;
 	Dllist	   *cc_lrulist;		/* LRU list, most recent first */

src/include/utils/int8.h

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2000, PostgreSQL, Inc
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: int8.h,v 1.17 2000/01/26 05:58:38 momjian Exp $
+ * $Id: int8.h,v 1.18 2000/02/21 03:36:59 tgl Exp $
  *
  * NOTES
  * These data types are supported on all 64-bit architectures, and may
@@ -93,9 +93,7 @@ extern int64 *int48(int32 val);
 extern int32 int84(int64 *val);
 
 #ifdef NOT_USED
-extern int64 *int2vector (int16 val);
 extern int16 int82(int64 *val);
-
 #endif
 
 extern float64 i8tod(int64 *val);