From 34d581f2dcffa9f54e96af230a56cb01e8e2fccd Mon Sep 17 00:00:00 2001 From: Yuri Kunde Schlesner Date: Fri, 16 Dec 2016 21:41:38 -0800 Subject: VideoCore/Shader: Extract input vertex loading code into function --- src/video_core/shader/shader.cpp | 30 +++++++++++++----------------- 1 file changed, 13 insertions(+), 17 deletions(-) (limited to 'src/video_core/shader/shader.cpp') diff --git a/src/video_core/shader/shader.cpp b/src/video_core/shader/shader.cpp index 7ae57e619..8dca9d0cb 100644 --- a/src/video_core/shader/shader.cpp +++ b/src/video_core/shader/shader.cpp @@ -76,6 +76,17 @@ OutputVertex OutputRegisters::ToVertex(const Regs::ShaderConfig& config) const { return ret; } +void UnitState::LoadInputVertex(const InputVertex& input, int num_attributes) { + // Setup input register table + const auto& attribute_register_map = g_state.regs.vs.input_register_map; + + for (int i = 0; i < num_attributes; i++) + registers.input[attribute_register_map.GetRegisterForAttribute(i)] = input.attr[i]; + + conditional_code[0] = false; + conditional_code[1] = false; +} + #ifdef ARCHITECTURE_x86_64 static std::unordered_map> shader_map; static const JitShader* jit_shader; @@ -109,21 +120,12 @@ void ShaderSetup::Setup() { MICROPROFILE_DEFINE(GPU_Shader, "GPU", "Shader", MP_RGB(50, 50, 240)); -void ShaderSetup::Run(UnitState& state, const InputVertex& input, int num_attributes) { +void ShaderSetup::Run(UnitState& state) { auto& config = g_state.regs.vs; auto& setup = g_state.vs; MICROPROFILE_SCOPE(GPU_Shader); - // Setup input register table - const auto& attribute_register_map = config.input_register_map; - - for (int i = 0; i < num_attributes; i++) - state.registers.input[attribute_register_map.GetRegisterForAttribute(i)] = input.attr[i]; - - state.conditional_code[0] = false; - state.conditional_code[1] = false; - #ifdef ARCHITECTURE_x86_64 if (VideoCore::g_shader_jit_enabled) { jit_shader->Run(setup, state, config.main_offset); @@ -145,13 +147,7 @@ DebugData ShaderSetup::ProduceDebugInfo(const InputVertex& input, int num_ // Setup input register table boost::fill(state.registers.input, Math::Vec4::AssignToAll(float24::Zero())); - const auto& attribute_register_map = config.input_register_map; - for (int i = 0; i < num_attributes; i++) - state.registers.input[attribute_register_map.GetRegisterForAttribute(i)] = input.attr[i]; - - state.conditional_code[0] = false; - state.conditional_code[1] = false; - + state.LoadInputVertex(input, num_attributes); RunInterpreter(setup, state, debug_data, config.main_offset); return debug_data; } -- cgit v1.2.3 From e3caf669b05bc0727053885ee7e6e5c78d655df4 Mon Sep 17 00:00:00 2001 From: Yuri Kunde Schlesner Date: Fri, 16 Dec 2016 21:48:36 -0800 Subject: VideoCore/Shader: Use self instead of g_state.vs in ShaderSetup --- src/video_core/shader/shader.cpp | 16 +++++++--------- 1 file changed, 7 insertions(+), 9 deletions(-) (limited to 'src/video_core/shader/shader.cpp') diff --git a/src/video_core/shader/shader.cpp b/src/video_core/shader/shader.cpp index 8dca9d0cb..868be1360 100644 --- a/src/video_core/shader/shader.cpp +++ b/src/video_core/shader/shader.cpp @@ -102,8 +102,8 @@ void ShaderSetup::Setup() { #ifdef ARCHITECTURE_x86_64 if (VideoCore::g_shader_jit_enabled) { u64 cache_key = - Common::ComputeHash64(&g_state.vs.program_code, sizeof(g_state.vs.program_code)) ^ - Common::ComputeHash64(&g_state.vs.swizzle_data, sizeof(g_state.vs.swizzle_data)); + Common::ComputeHash64(&program_code, sizeof(program_code)) ^ + Common::ComputeHash64(&swizzle_data, sizeof(swizzle_data)); auto iter = shader_map.find(cache_key); if (iter != shader_map.end()) { @@ -122,33 +122,31 @@ MICROPROFILE_DEFINE(GPU_Shader, "GPU", "Shader", MP_RGB(50, 50, 240)); void ShaderSetup::Run(UnitState& state) { auto& config = g_state.regs.vs; - auto& setup = g_state.vs; MICROPROFILE_SCOPE(GPU_Shader); #ifdef ARCHITECTURE_x86_64 if (VideoCore::g_shader_jit_enabled) { - jit_shader->Run(setup, state, config.main_offset); + jit_shader->Run(*this, state, config.main_offset); } else { DebugData dummy_debug_data; - RunInterpreter(setup, state, dummy_debug_data, config.main_offset); + RunInterpreter(*this, state, dummy_debug_data, config.main_offset); } #else DebugData dummy_debug_data; - RunInterpreter(setup, state, dummy_debug_data, config.main_offset); + RunInterpreter(*this, state, dummy_debug_data, config.main_offset); #endif // ARCHITECTURE_x86_64 } DebugData ShaderSetup::ProduceDebugInfo(const InputVertex& input, int num_attributes, - const Regs::ShaderConfig& config, - const ShaderSetup& setup) { + const Regs::ShaderConfig& config) { UnitState state; DebugData debug_data; // Setup input register table boost::fill(state.registers.input, Math::Vec4::AssignToAll(float24::Zero())); state.LoadInputVertex(input, num_attributes); - RunInterpreter(setup, state, debug_data, config.main_offset); + RunInterpreter(*this, state, debug_data, config.main_offset); return debug_data; } -- cgit v1.2.3 From 1e1f9398176e4f1ec608f31f22a576c749a0a723 Mon Sep 17 00:00:00 2001 From: Yuri Kunde Schlesner Date: Fri, 16 Dec 2016 22:30:00 -0800 Subject: VideoCore/Shader: Use only entry_point as ShaderSetup param This removes all implicit dependency of ShaderState on global PICA state. --- src/video_core/shader/shader.cpp | 16 +++++++++------- 1 file changed, 9 insertions(+), 7 deletions(-) (limited to 'src/video_core/shader/shader.cpp') diff --git a/src/video_core/shader/shader.cpp b/src/video_core/shader/shader.cpp index 868be1360..936db0582 100644 --- a/src/video_core/shader/shader.cpp +++ b/src/video_core/shader/shader.cpp @@ -120,33 +120,35 @@ void ShaderSetup::Setup() { MICROPROFILE_DEFINE(GPU_Shader, "GPU", "Shader", MP_RGB(50, 50, 240)); -void ShaderSetup::Run(UnitState& state) { - auto& config = g_state.regs.vs; +void ShaderSetup::Run(UnitState& state, unsigned int entry_point) { + ASSERT(entry_point < 1024); MICROPROFILE_SCOPE(GPU_Shader); #ifdef ARCHITECTURE_x86_64 if (VideoCore::g_shader_jit_enabled) { - jit_shader->Run(*this, state, config.main_offset); + jit_shader->Run(*this, state, entry_point); } else { DebugData dummy_debug_data; - RunInterpreter(*this, state, dummy_debug_data, config.main_offset); + RunInterpreter(*this, state, dummy_debug_data, entry_point); } #else DebugData dummy_debug_data; - RunInterpreter(*this, state, dummy_debug_data, config.main_offset); + RunInterpreter(*this, state, dummy_debug_data, entry_point); #endif // ARCHITECTURE_x86_64 } DebugData ShaderSetup::ProduceDebugInfo(const InputVertex& input, int num_attributes, - const Regs::ShaderConfig& config) { + unsigned int entry_point) { + ASSERT(entry_point < 1024); + UnitState state; DebugData debug_data; // Setup input register table boost::fill(state.registers.input, Math::Vec4::AssignToAll(float24::Zero())); state.LoadInputVertex(input, num_attributes); - RunInterpreter(*this, state, debug_data, config.main_offset); + RunInterpreter(*this, state, debug_data, entry_point); return debug_data; } -- cgit v1.2.3 From bd82cffd0bf5d61eab8c7c856bcc284b1a77e33b Mon Sep 17 00:00:00 2001 From: Yuri Kunde Schlesner Date: Fri, 16 Dec 2016 22:32:35 -0800 Subject: VideoCore/Shader: Add constness to methods --- src/video_core/shader/shader.cpp | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'src/video_core/shader/shader.cpp') diff --git a/src/video_core/shader/shader.cpp b/src/video_core/shader/shader.cpp index 936db0582..ae696533f 100644 --- a/src/video_core/shader/shader.cpp +++ b/src/video_core/shader/shader.cpp @@ -120,7 +120,7 @@ void ShaderSetup::Setup() { MICROPROFILE_DEFINE(GPU_Shader, "GPU", "Shader", MP_RGB(50, 50, 240)); -void ShaderSetup::Run(UnitState& state, unsigned int entry_point) { +void ShaderSetup::Run(UnitState& state, unsigned int entry_point) const { ASSERT(entry_point < 1024); MICROPROFILE_SCOPE(GPU_Shader); @@ -139,7 +139,7 @@ void ShaderSetup::Run(UnitState& state, unsigned int entry_point) { } DebugData ShaderSetup::ProduceDebugInfo(const InputVertex& input, int num_attributes, - unsigned int entry_point) { + unsigned int entry_point) const { ASSERT(entry_point < 1024); UnitState state; -- cgit v1.2.3 From dd4a1672a77830a53de61cf0554b34e9e17a2905 Mon Sep 17 00:00:00 2001 From: Yuri Kunde Schlesner Date: Fri, 16 Dec 2016 23:21:26 -0800 Subject: VideoCore/Shader: Split shader uniform state and shader engine Currently there's only a single dummy implementation, which will be split in a following commit. --- src/video_core/shader/shader.cpp | 44 ++++++++++++++++++++++++++++++---------- 1 file changed, 33 insertions(+), 11 deletions(-) (limited to 'src/video_core/shader/shader.cpp') diff --git a/src/video_core/shader/shader.cpp b/src/video_core/shader/shader.cpp index ae696533f..d276a1221 100644 --- a/src/video_core/shader/shader.cpp +++ b/src/video_core/shader/shader.cpp @@ -87,6 +87,17 @@ void UnitState::LoadInputVertex(const InputVertex& input, int num_attributes) { conditional_code[1] = false; } +class MergedShaderEngine : public ShaderEngine { +public: + void SetupBatch(const ShaderSetup* setup) override; + void Run(UnitState& state, unsigned int entry_point) const override; + DebugData ProduceDebugInfo(const InputVertex& input, int num_attributes, + unsigned int entry_point) const override; + +private: + const ShaderSetup* setup = nullptr; +}; + #ifdef ARCHITECTURE_x86_64 static std::unordered_map> shader_map; static const JitShader* jit_shader; @@ -98,13 +109,17 @@ void ClearCache() { #endif // ARCHITECTURE_x86_64 } -void ShaderSetup::Setup() { +void MergedShaderEngine::SetupBatch(const ShaderSetup* setup_) { + setup = setup_; + if (setup == nullptr) + return; + #ifdef ARCHITECTURE_x86_64 if (VideoCore::g_shader_jit_enabled) { - u64 cache_key = - Common::ComputeHash64(&program_code, sizeof(program_code)) ^ - Common::ComputeHash64(&swizzle_data, sizeof(swizzle_data)); + u64 code_hash = Common::ComputeHash64(&setup->program_code, sizeof(setup->program_code)); + u64 swizzle_hash = Common::ComputeHash64(&setup->swizzle_data, sizeof(setup->swizzle_data)); + u64 cache_key = code_hash ^ swizzle_hash; auto iter = shader_map.find(cache_key); if (iter != shader_map.end()) { jit_shader = iter->second.get(); @@ -120,26 +135,28 @@ void ShaderSetup::Setup() { MICROPROFILE_DEFINE(GPU_Shader, "GPU", "Shader", MP_RGB(50, 50, 240)); -void ShaderSetup::Run(UnitState& state, unsigned int entry_point) const { +void MergedShaderEngine::Run(UnitState& state, unsigned int entry_point) const { + ASSERT(setup != nullptr); ASSERT(entry_point < 1024); MICROPROFILE_SCOPE(GPU_Shader); #ifdef ARCHITECTURE_x86_64 if (VideoCore::g_shader_jit_enabled) { - jit_shader->Run(*this, state, entry_point); + jit_shader->Run(*setup, state, entry_point); } else { DebugData dummy_debug_data; - RunInterpreter(*this, state, dummy_debug_data, entry_point); + RunInterpreter(*setup, state, dummy_debug_data, entry_point); } #else DebugData dummy_debug_data; - RunInterpreter(*this, state, dummy_debug_data, entry_point); + RunInterpreter(*setup, state, dummy_debug_data, entry_point); #endif // ARCHITECTURE_x86_64 } -DebugData ShaderSetup::ProduceDebugInfo(const InputVertex& input, int num_attributes, - unsigned int entry_point) const { +DebugData MergedShaderEngine::ProduceDebugInfo(const InputVertex& input, int num_attributes, + unsigned int entry_point) const { + ASSERT(setup != nullptr); ASSERT(entry_point < 1024); UnitState state; @@ -148,10 +165,15 @@ DebugData ShaderSetup::ProduceDebugInfo(const InputVertex& input, int num_ // Setup input register table boost::fill(state.registers.input, Math::Vec4::AssignToAll(float24::Zero())); state.LoadInputVertex(input, num_attributes); - RunInterpreter(*this, state, debug_data, entry_point); + RunInterpreter(*setup, state, debug_data, entry_point); return debug_data; } +ShaderEngine* GetEngine() { + static MergedShaderEngine merged_engine; + return &merged_engine; +} + } // namespace Shader } // namespace Pica -- cgit v1.2.3 From 8eefc62833bc8c3052c23f4f0d01d8b60a01925c Mon Sep 17 00:00:00 2001 From: Yuri Kunde Schlesner Date: Sat, 17 Dec 2016 00:06:23 -0800 Subject: VideoCore/Shader: Rename shader_jit_x64{ => _compiler}.{cpp,h} --- src/video_core/shader/shader.cpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'src/video_core/shader/shader.cpp') diff --git a/src/video_core/shader/shader.cpp b/src/video_core/shader/shader.cpp index d276a1221..97c6519d6 100644 --- a/src/video_core/shader/shader.cpp +++ b/src/video_core/shader/shader.cpp @@ -17,7 +17,7 @@ #include "video_core/shader/shader.h" #include "video_core/shader/shader_interpreter.h" #ifdef ARCHITECTURE_x86_64 -#include "video_core/shader/shader_jit_x64.h" +#include "video_core/shader/shader_jit_x64_compiler.h" #endif // ARCHITECTURE_x86_64 #include "video_core/video_core.h" -- cgit v1.2.3 From 114d6b2f97eb62c7d8c958ebb391b70b026130f9 Mon Sep 17 00:00:00 2001 From: Yuri Kunde Schlesner Date: Sat, 17 Dec 2016 01:21:16 -0800 Subject: VideoCore/Shader: Split interpreter and JIT into separate ShaderEngines --- src/video_core/shader/shader.cpp | 92 ++++++---------------------------------- 1 file changed, 13 insertions(+), 79 deletions(-) (limited to 'src/video_core/shader/shader.cpp') diff --git a/src/video_core/shader/shader.cpp b/src/video_core/shader/shader.cpp index 97c6519d6..b30dae476 100644 --- a/src/video_core/shader/shader.cpp +++ b/src/video_core/shader/shader.cpp @@ -2,14 +2,8 @@ // Licensed under GPLv2 or any later version // Refer to the license.txt file included. -#include #include #include -#include -#include -#include -#include "common/bit_field.h" -#include "common/hash.h" #include "common/logging/log.h" #include "common/microprofile.h" #include "video_core/pica.h" @@ -17,7 +11,7 @@ #include "video_core/shader/shader.h" #include "video_core/shader/shader_interpreter.h" #ifdef ARCHITECTURE_x86_64 -#include "video_core/shader/shader_jit_x64_compiler.h" +#include "video_core/shader/shader_jit_x64.h" #endif // ARCHITECTURE_x86_64 #include "video_core/video_core.h" @@ -87,93 +81,33 @@ void UnitState::LoadInputVertex(const InputVertex& input, int num_attributes) { conditional_code[1] = false; } -class MergedShaderEngine : public ShaderEngine { -public: - void SetupBatch(const ShaderSetup* setup) override; - void Run(UnitState& state, unsigned int entry_point) const override; - DebugData ProduceDebugInfo(const InputVertex& input, int num_attributes, - unsigned int entry_point) const override; - -private: - const ShaderSetup* setup = nullptr; -}; - -#ifdef ARCHITECTURE_x86_64 -static std::unordered_map> shader_map; -static const JitShader* jit_shader; -#endif // ARCHITECTURE_x86_64 +MICROPROFILE_DEFINE(GPU_Shader, "GPU", "Shader", MP_RGB(50, 50, 240)); -void ClearCache() { #ifdef ARCHITECTURE_x86_64 - shader_map.clear(); +static std::unique_ptr jit_engine; #endif // ARCHITECTURE_x86_64 -} - -void MergedShaderEngine::SetupBatch(const ShaderSetup* setup_) { - setup = setup_; - if (setup == nullptr) - return; +static InterpreterEngine interpreter_engine; +ShaderEngine* GetEngine() { #ifdef ARCHITECTURE_x86_64 + // TODO(yuriks): Re-initialize on each change rather than being persistent if (VideoCore::g_shader_jit_enabled) { - u64 code_hash = Common::ComputeHash64(&setup->program_code, sizeof(setup->program_code)); - u64 swizzle_hash = Common::ComputeHash64(&setup->swizzle_data, sizeof(setup->swizzle_data)); - - u64 cache_key = code_hash ^ swizzle_hash; - auto iter = shader_map.find(cache_key); - if (iter != shader_map.end()) { - jit_shader = iter->second.get(); - } else { - auto shader = std::make_unique(); - shader->Compile(); - jit_shader = shader.get(); - shader_map[cache_key] = std::move(shader); + if (jit_engine == nullptr) { + jit_engine = std::make_unique(); } + return jit_engine.get(); } #endif // ARCHITECTURE_x86_64 -} - -MICROPROFILE_DEFINE(GPU_Shader, "GPU", "Shader", MP_RGB(50, 50, 240)); - -void MergedShaderEngine::Run(UnitState& state, unsigned int entry_point) const { - ASSERT(setup != nullptr); - ASSERT(entry_point < 1024); - MICROPROFILE_SCOPE(GPU_Shader); + return &interpreter_engine; +} +void Shutdown() { #ifdef ARCHITECTURE_x86_64 - if (VideoCore::g_shader_jit_enabled) { - jit_shader->Run(*setup, state, entry_point); - } else { - DebugData dummy_debug_data; - RunInterpreter(*setup, state, dummy_debug_data, entry_point); - } -#else - DebugData dummy_debug_data; - RunInterpreter(*setup, state, dummy_debug_data, entry_point); + jit_engine = nullptr; #endif // ARCHITECTURE_x86_64 } -DebugData MergedShaderEngine::ProduceDebugInfo(const InputVertex& input, int num_attributes, - unsigned int entry_point) const { - ASSERT(setup != nullptr); - ASSERT(entry_point < 1024); - - UnitState state; - DebugData debug_data; - - // Setup input register table - boost::fill(state.registers.input, Math::Vec4::AssignToAll(float24::Zero())); - state.LoadInputVertex(input, num_attributes); - RunInterpreter(*setup, state, debug_data, entry_point); - return debug_data; -} - -ShaderEngine* GetEngine() { - static MergedShaderEngine merged_engine; - return &merged_engine; -} - } // namespace Shader } // namespace Pica -- cgit v1.2.3 From 9ea5eacf919c8c257f8c5fda65e5fac2b6adee07 Mon Sep 17 00:00:00 2001 From: Yuri Kunde Schlesner Date: Sat, 17 Dec 2016 14:09:02 -0800 Subject: Shader: Initialize conditional_code in interpreter This doesn't belong in LoadInputVertex because it also happens for non-VS invocations. Since it's not used by the JIT it seems adequate to initialize it in the interpreter which is the only thing that cares about them. --- src/video_core/shader/shader.cpp | 3 --- 1 file changed, 3 deletions(-) (limited to 'src/video_core/shader/shader.cpp') diff --git a/src/video_core/shader/shader.cpp b/src/video_core/shader/shader.cpp index b30dae476..1662b5d38 100644 --- a/src/video_core/shader/shader.cpp +++ b/src/video_core/shader/shader.cpp @@ -76,9 +76,6 @@ void UnitState::LoadInputVertex(const InputVertex& input, int num_attributes) { for (int i = 0; i < num_attributes; i++) registers.input[attribute_register_map.GetRegisterForAttribute(i)] = input.attr[i]; - - conditional_code[0] = false; - conditional_code[1] = false; } MICROPROFILE_DEFINE(GPU_Shader, "GPU", "Shader", MP_RGB(50, 50, 240)); -- cgit v1.2.3 From 6fa3687afc97685101f9ee5c65cf98f505980695 Mon Sep 17 00:00:00 2001 From: Yuri Kunde Schlesner Date: Sat, 17 Dec 2016 14:38:03 -0800 Subject: Shader: Remove OutputRegisters struct --- src/video_core/shader/shader.cpp | 11 ++++++----- 1 file changed, 6 insertions(+), 5 deletions(-) (limited to 'src/video_core/shader/shader.cpp') diff --git a/src/video_core/shader/shader.cpp b/src/video_core/shader/shader.cpp index 1662b5d38..2da50bd62 100644 --- a/src/video_core/shader/shader.cpp +++ b/src/video_core/shader/shader.cpp @@ -19,7 +19,8 @@ namespace Pica { namespace Shader { -OutputVertex OutputRegisters::ToVertex(const Regs::ShaderConfig& config) const { +OutputVertex OutputVertex::FromRegisters(Math::Vec4 output_regs[16], const Regs& regs, + u32 output_mask) { // Setup output data OutputVertex ret; // TODO(neobrain): Under some circumstances, up to 16 attributes may be output. We need to @@ -27,13 +28,13 @@ OutputVertex OutputRegisters::ToVertex(const Regs::ShaderConfig& config) const { unsigned index = 0; for (unsigned i = 0; i < 7; ++i) { - if (index >= g_state.regs.vs_output_total) + if (index >= regs.vs_output_total) break; - if ((config.output_mask & (1 << i)) == 0) + if ((output_mask & (1 << i)) == 0) continue; - const auto& output_register_map = g_state.regs.vs_output_attributes[index]; + const auto& output_register_map = regs.vs_output_attributes[index]; u32 semantics[4] = {output_register_map.map_x, output_register_map.map_y, output_register_map.map_z, output_register_map.map_w}; @@ -41,7 +42,7 @@ OutputVertex OutputRegisters::ToVertex(const Regs::ShaderConfig& config) const { for (unsigned comp = 0; comp < 4; ++comp) { float24* out = ((float24*)&ret) + semantics[comp]; if (semantics[comp] != Regs::VSOutputAttributes::INVALID) { - *out = value[i][comp]; + *out = output_regs[i][comp]; } else { // Zero output so that attributes which aren't output won't have denormals in them, // which would slow us down later. -- cgit v1.2.3