From 4fda7f1c821e1000210b1dfdf1074795b1aa9d96 Mon Sep 17 00:00:00 2001 From: ameerj <52414509+ameerj@users.noreply.github.com> Date: Mon, 23 Aug 2021 20:00:11 -0400 Subject: [PATCH] structured_control_flow: Conditionally invoke demote reorder pass This is only needed on select drivers when a fragment shader discards/demotes. --- .../frontend/maxwell/structured_control_flow.cpp | 10 ++++++---- .../frontend/maxwell/structured_control_flow.h | 9 ++++++--- .../frontend/maxwell/translate_program.cpp | 2 +- src/shader_recompiler/host_translate_info.h | 5 +++-- src/video_core/renderer_opengl/gl_device.h | 4 ++++ src/video_core/renderer_opengl/gl_shader_cache.cpp | 1 + src/video_core/renderer_vulkan/vk_pipeline_cache.cpp | 2 ++ 7 files changed, 23 insertions(+), 10 deletions(-) diff --git a/src/shader_recompiler/frontend/maxwell/structured_control_flow.cpp b/src/shader_recompiler/frontend/maxwell/structured_control_flow.cpp index f124dc8c07..77d1cd0fc3 100644 --- a/src/shader_recompiler/frontend/maxwell/structured_control_flow.cpp +++ b/src/shader_recompiler/frontend/maxwell/structured_control_flow.cpp @@ -20,6 +20,7 @@ #include "shader_recompiler/frontend/maxwell/decode.h" #include "shader_recompiler/frontend/maxwell/structured_control_flow.h" #include "shader_recompiler/frontend/maxwell/translate/translate.h" +#include "shader_recompiler/host_translate_info.h" #include "shader_recompiler/object_pool.h" namespace Shader::Maxwell { @@ -652,7 +653,7 @@ class TranslatePass { public: TranslatePass(ObjectPool& inst_pool_, ObjectPool& block_pool_, ObjectPool& stmt_pool_, Environment& env_, Statement& root_stmt, - IR::AbstractSyntaxList& syntax_list_) + IR::AbstractSyntaxList& syntax_list_, const HostTranslateInfo& host_info) : stmt_pool{stmt_pool_}, inst_pool{inst_pool_}, block_pool{block_pool_}, env{env_}, syntax_list{syntax_list_} { Visit(root_stmt, nullptr, nullptr); @@ -660,7 +661,7 @@ public: IR::Block& first_block{*syntax_list.front().data.block}; IR::IREmitter ir(first_block, first_block.begin()); ir.Prologue(); - if (uses_demote_to_helper) { + if (uses_demote_to_helper && host_info.needs_demote_reorder) { DemoteCombinationPass(); } } @@ -977,12 +978,13 @@ private: } // Anonymous namespace IR::AbstractSyntaxList BuildASL(ObjectPool& inst_pool, ObjectPool& block_pool, - Environment& env, Flow::CFG& cfg) { + Environment& env, Flow::CFG& cfg, + const HostTranslateInfo& host_info) { ObjectPool stmt_pool{64}; GotoPass goto_pass{cfg, stmt_pool}; Statement& root{goto_pass.RootStatement()}; IR::AbstractSyntaxList syntax_list; - TranslatePass{inst_pool, block_pool, stmt_pool, env, root, syntax_list}; + TranslatePass{inst_pool, block_pool, stmt_pool, env, root, syntax_list, host_info}; return syntax_list; } diff --git a/src/shader_recompiler/frontend/maxwell/structured_control_flow.h b/src/shader_recompiler/frontend/maxwell/structured_control_flow.h index 88b0836491..e38158da33 100644 --- a/src/shader_recompiler/frontend/maxwell/structured_control_flow.h +++ b/src/shader_recompiler/frontend/maxwell/structured_control_flow.h @@ -11,10 +11,13 @@ #include "shader_recompiler/frontend/maxwell/control_flow.h" #include "shader_recompiler/object_pool.h" -namespace Shader::Maxwell { +namespace Shader { +struct HostTranslateInfo; +namespace Maxwell { [[nodiscard]] IR::AbstractSyntaxList BuildASL(ObjectPool& inst_pool, ObjectPool& block_pool, Environment& env, - Flow::CFG& cfg); + Flow::CFG& cfg, const HostTranslateInfo& host_info); -} // namespace Shader::Maxwell +} // namespace Maxwell +} // namespace Shader diff --git a/src/shader_recompiler/frontend/maxwell/translate_program.cpp b/src/shader_recompiler/frontend/maxwell/translate_program.cpp index c067d459cc..012d553574 100644 --- a/src/shader_recompiler/frontend/maxwell/translate_program.cpp +++ b/src/shader_recompiler/frontend/maxwell/translate_program.cpp @@ -130,7 +130,7 @@ void AddNVNStorageBuffers(IR::Program& program) { IR::Program TranslateProgram(ObjectPool& inst_pool, ObjectPool& block_pool, Environment& env, Flow::CFG& cfg, const HostTranslateInfo& host_info) { IR::Program program; - program.syntax_list = BuildASL(inst_pool, block_pool, env, cfg); + program.syntax_list = BuildASL(inst_pool, block_pool, env, cfg, host_info); program.blocks = GenerateBlocks(program.syntax_list); program.post_order_blocks = PostOrder(program.syntax_list.front()); program.stage = env.ShaderStage(); diff --git a/src/shader_recompiler/host_translate_info.h b/src/shader_recompiler/host_translate_info.h index 94a584219c..96468b2e79 100644 --- a/src/shader_recompiler/host_translate_info.h +++ b/src/shader_recompiler/host_translate_info.h @@ -11,8 +11,9 @@ namespace Shader { /// Misc information about the host struct HostTranslateInfo { - bool support_float16{}; ///< True when the device supports 16-bit floats - bool support_int64{}; ///< True when the device supports 64-bit integers + bool support_float16{}; ///< True when the device supports 16-bit floats + bool support_int64{}; ///< True when the device supports 64-bit integers + bool needs_demote_reorder{}; ///< True when the device needs DemoteToHelperInvocation reordered }; } // namespace Shader diff --git a/src/video_core/renderer_opengl/gl_device.h b/src/video_core/renderer_opengl/gl_device.h index ee992aed4a..de9e41659d 100644 --- a/src/video_core/renderer_opengl/gl_device.h +++ b/src/video_core/renderer_opengl/gl_device.h @@ -156,6 +156,10 @@ public: return shader_backend; } + bool IsAmd() const { + return vendor_name == "ATI Technologies Inc."; + } + private: static bool TestVariableAoffi(); static bool TestPreciseBug(); diff --git a/src/video_core/renderer_opengl/gl_shader_cache.cpp b/src/video_core/renderer_opengl/gl_shader_cache.cpp index 1f4dda17ef..b0e14182e3 100644 --- a/src/video_core/renderer_opengl/gl_shader_cache.cpp +++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp @@ -219,6 +219,7 @@ ShaderCache::ShaderCache(RasterizerOpenGL& rasterizer_, Core::Frontend::EmuWindo host_info{ .support_float16 = false, .support_int64 = device.HasShaderInt64(), + .needs_demote_reorder = device.IsAmd(), } { if (use_asynchronous_shaders) { workers = CreateWorkers(); diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp index f316c4f922..31bfbcb064 100644 --- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp @@ -325,6 +325,8 @@ PipelineCache::PipelineCache(RasterizerVulkan& rasterizer_, Tegra::Engines::Maxw host_info = Shader::HostTranslateInfo{ .support_float16 = device.IsFloat16Supported(), .support_int64 = device.IsShaderInt64Supported(), + .needs_demote_reorder = driver_id == VK_DRIVER_ID_AMD_PROPRIETARY_KHR || + driver_id == VK_DRIVER_ID_AMD_OPEN_SOURCE_KHR, }; }