From 3f69c2039de1c3d084ac2c9eb0aa9315490346bf Mon Sep 17 00:00:00 2001 From: bunnei Date: Tue, 21 Jul 2015 19:38:59 -0400 Subject: Shader: Define a common interface for running vertex shader programs. --- src/video_core/shader/shader.cpp | 105 +++++++++++++++++++++++++++++++++++++++ 1 file changed, 105 insertions(+) create mode 100644 src/video_core/shader/shader.cpp (limited to 'src/video_core/shader/shader.cpp') diff --git a/src/video_core/shader/shader.cpp b/src/video_core/shader/shader.cpp new file mode 100644 index 000000000..e397e8e03 --- /dev/null +++ b/src/video_core/shader/shader.cpp @@ -0,0 +1,105 @@ +// Copyright 2015 Citra Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include "common/logging/log.h" +#include "common/profiler.h" + +#include "video_core/debug_utils/debug_utils.h" +#include "video_core/pica.h" + +#include "shader.h" +#include "shader_interpreter.h" + +namespace Pica { + +namespace Shader { + +void Setup(UnitState& state) { + // TODO(bunnei): This will be used by the JIT in a subsequent commit +} + +static Common::Profiling::TimingCategory shader_category("Vertex Shader"); + +OutputVertex Run(UnitState& state, const InputVertex& input, int num_attributes) { + auto& config = g_state.regs.vs; + auto& setup = g_state.vs; + + Common::Profiling::ScopeTimer timer(shader_category); + + state.program_counter = config.main_offset; + state.debug.max_offset = 0; + state.debug.max_opdesc_id = 0; + + // Setup input register table + const auto& attribute_register_map = config.input_register_map; + + if (num_attributes > 0) state.input_registers[attribute_register_map.attribute0_register] = input.attr[0]; + if (num_attributes > 1) state.input_registers[attribute_register_map.attribute1_register] = input.attr[1]; + if (num_attributes > 2) state.input_registers[attribute_register_map.attribute2_register] = input.attr[2]; + if (num_attributes > 3) state.input_registers[attribute_register_map.attribute3_register] = input.attr[3]; + if (num_attributes > 4) state.input_registers[attribute_register_map.attribute4_register] = input.attr[4]; + if (num_attributes > 5) state.input_registers[attribute_register_map.attribute5_register] = input.attr[5]; + if (num_attributes > 6) state.input_registers[attribute_register_map.attribute6_register] = input.attr[6]; + if (num_attributes > 7) state.input_registers[attribute_register_map.attribute7_register] = input.attr[7]; + if (num_attributes > 8) state.input_registers[attribute_register_map.attribute8_register] = input.attr[8]; + if (num_attributes > 9) state.input_registers[attribute_register_map.attribute9_register] = input.attr[9]; + if (num_attributes > 10) state.input_registers[attribute_register_map.attribute10_register] = input.attr[10]; + if (num_attributes > 11) state.input_registers[attribute_register_map.attribute11_register] = input.attr[11]; + if (num_attributes > 12) state.input_registers[attribute_register_map.attribute12_register] = input.attr[12]; + if (num_attributes > 13) state.input_registers[attribute_register_map.attribute13_register] = input.attr[13]; + if (num_attributes > 14) state.input_registers[attribute_register_map.attribute14_register] = input.attr[14]; + if (num_attributes > 15) state.input_registers[attribute_register_map.attribute15_register] = input.attr[15]; + + state.conditional_code[0] = false; + state.conditional_code[1] = false; + + RunInterpreter(state); + +#if PICA_DUMP_SHADERS + DebugUtils::DumpShader(setup.program_code.data(), state.debug.max_offset, setup.swizzle_data.data(), + state.debug.max_opdesc_id, config.main_offset, + g_state.regs.vs_output_attributes); // TODO: Don't hardcode VS here +#endif + + // Setup output data + OutputVertex ret; + // TODO(neobrain): Under some circumstances, up to 16 attributes may be output. We need to + // figure out what those circumstances are and enable the remaining outputs then. + for (int i = 0; i < 7; ++i) { + const auto& output_register_map = g_state.regs.vs_output_attributes[i]; // TODO: Don't hardcode VS here + + u32 semantics[4] = { + output_register_map.map_x, output_register_map.map_y, + output_register_map.map_z, output_register_map.map_w + }; + + for (int comp = 0; comp < 4; ++comp) { + float24* out = ((float24*)&ret) + semantics[comp]; + if (semantics[comp] != Regs::VSOutputAttributes::INVALID) { + *out = state.output_registers[i][comp]; + } else { + // Zero output so that attributes which aren't output won't have denormals in them, + // which would slow us down later. + memset(out, 0, sizeof(*out)); + } + } + } + + // The hardware takes the absolute and saturates vertex colors like this, *before* doing interpolation + for (int i = 0; i < 4; ++i) { + ret.color[i] = float24::FromFloat32( + std::fmin(std::fabs(ret.color[i].ToFloat32()), 1.0f)); + } + + LOG_TRACE(Render_Software, "Output vertex: pos (%.2f, %.2f, %.2f, %.2f), col(%.2f, %.2f, %.2f, %.2f), tc0(%.2f, %.2f)", + ret.pos.x.ToFloat32(), ret.pos.y.ToFloat32(), ret.pos.z.ToFloat32(), ret.pos.w.ToFloat32(), + ret.color.x.ToFloat32(), ret.color.y.ToFloat32(), ret.color.z.ToFloat32(), ret.color.w.ToFloat32(), + ret.tc0.u().ToFloat32(), ret.tc0.v().ToFloat32()); + + return ret; +} + +} // namespace Shader + +} // namespace Pica -- cgit v1.2.3 From d67e2f78b77c0739caa8c6a915685aa8961a519c Mon Sep 17 00:00:00 2001 From: bunnei Date: Tue, 21 Jul 2015 22:54:38 -0400 Subject: Common: Added MurmurHash3 hash function for general-purpose use. --- 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 e397e8e03..13e22cb53 100644 --- a/src/video_core/shader/shader.cpp +++ b/src/video_core/shader/shader.cpp @@ -16,7 +16,7 @@ namespace Pica { namespace Shader { void Setup(UnitState& state) { - // TODO(bunnei): This will be used by the JIT in a subsequent commit + // TODO(bunnei): This will be used by the JIT in a subsequent patch } static Common::Profiling::TimingCategory shader_category("Vertex Shader"); -- cgit v1.2.3 From 094ae6fadb57883e25d412443fcae987ddf240ef Mon Sep 17 00:00:00 2001 From: bunnei Date: Wed, 22 Jul 2015 23:25:30 -0400 Subject: Shader: Initial implementation of x86_x64 JIT compiler for Pica vertex shaders. - Config: Add an option for selecting to use shader JIT or interpreter. - Qt: Add a menu option for enabling/disabling the shader JIT. --- src/video_core/shader/shader.cpp | 42 ++++++++++++++++++++++++++++++++++++++-- 1 file changed, 40 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 13e22cb53..fa1f7cafe 100644 --- a/src/video_core/shader/shader.cpp +++ b/src/video_core/shader/shader.cpp @@ -2,21 +2,52 @@ // Licensed under GPLv2 or any later version // Refer to the license.txt file included. -#include "common/logging/log.h" +#include +#include + +#include "common/hash.h" +#include "common/make_unique.h" #include "common/profiler.h" #include "video_core/debug_utils/debug_utils.h" #include "video_core/pica.h" +#include "video_core/video_core.h" #include "shader.h" #include "shader_interpreter.h" +#include "shader_jit.h" namespace Pica { namespace Shader { +#ifdef ARCHITECTURE_x86_64 + +static std::unordered_map shader_map; +static JitCompiler jit; +static CompiledShader* jit_shader; + +#endif // ARCHITECTURE_x86_64 + void Setup(UnitState& state) { - // TODO(bunnei): This will be used by the JIT in a subsequent patch +#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)) ^ + g_state.regs.vs.main_offset); + + auto iter = shader_map.find(cache_key); + if (iter != shader_map.end()) { + jit_shader = iter->second; + } else { + jit_shader = jit.Compile(); + shader_map.emplace(cache_key, jit_shader); + } + } +} + +void Shutdown() { + shader_map.clear(); } static Common::Profiling::TimingCategory shader_category("Vertex Shader"); @@ -54,7 +85,14 @@ OutputVertex Run(UnitState& state, const InputVertex& input, int num_attributes) state.conditional_code[0] = false; state.conditional_code[1] = false; +#ifdef ARCHITECTURE_x86_64 + if (VideoCore::g_shader_jit_enabled) + jit_shader(&state); + else + RunInterpreter(state); +#else RunInterpreter(state); +#endif #if PICA_DUMP_SHADERS DebugUtils::DumpShader(setup.program_code.data(), state.debug.max_offset, setup.swizzle_data.data(), -- cgit v1.2.3 From bd7e691f78d916ed6ae5396b2d646d9b3a053dd7 Mon Sep 17 00:00:00 2001 From: bunnei Date: Wed, 12 Aug 2015 00:00:44 -0400 Subject: x64: Refactor to remove fake interfaces and general cleanups. --- src/video_core/shader/shader.cpp | 9 ++++++--- 1 file changed, 6 insertions(+), 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 fa1f7cafe..f7459e2ad 100644 --- a/src/video_core/shader/shader.cpp +++ b/src/video_core/shader/shader.cpp @@ -15,7 +15,10 @@ #include "shader.h" #include "shader_interpreter.h" -#include "shader_jit.h" + +#ifdef ARCHITECTURE_X64 +#include "shader_jit_x64.h" +#endif // ARCHITECTURE_X64 namespace Pica { @@ -43,7 +46,7 @@ void Setup(UnitState& state) { jit_shader = jit.Compile(); shader_map.emplace(cache_key, jit_shader); } - } +#endif // ARCHITECTURE_X64 } void Shutdown() { @@ -92,7 +95,7 @@ OutputVertex Run(UnitState& state, const InputVertex& input, int num_attributes) RunInterpreter(state); #else RunInterpreter(state); -#endif +#endif // ARCHITECTURE_X64 #if PICA_DUMP_SHADERS DebugUtils::DumpShader(setup.program_code.data(), state.debug.max_offset, setup.swizzle_data.data(), -- cgit v1.2.3 From b39c053785abc4a94ce380885a846c5d8be89675 Mon Sep 17 00:00:00 2001 From: bunnei Date: Fri, 14 Aug 2015 22:29:08 -0400 Subject: Rename ARCHITECTURE_X64 definition to ARCHITECTURE_x86_64. --- src/video_core/shader/shader.cpp | 11 +++++------ 1 file changed, 5 insertions(+), 6 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 f7459e2ad..06c1fe653 100644 --- a/src/video_core/shader/shader.cpp +++ b/src/video_core/shader/shader.cpp @@ -16,20 +16,18 @@ #include "shader.h" #include "shader_interpreter.h" -#ifdef ARCHITECTURE_X64 +#ifdef ARCHITECTURE_x86_64 #include "shader_jit_x64.h" -#endif // ARCHITECTURE_X64 +#endif // ARCHITECTURE_x86_64 namespace Pica { namespace Shader { #ifdef ARCHITECTURE_x86_64 - static std::unordered_map shader_map; static JitCompiler jit; static CompiledShader* jit_shader; - #endif // ARCHITECTURE_x86_64 void Setup(UnitState& state) { @@ -46,7 +44,8 @@ void Setup(UnitState& state) { jit_shader = jit.Compile(); shader_map.emplace(cache_key, jit_shader); } -#endif // ARCHITECTURE_X64 + } +#endif // ARCHITECTURE_x86_64 } void Shutdown() { @@ -95,7 +94,7 @@ OutputVertex Run(UnitState& state, const InputVertex& input, int num_attributes) RunInterpreter(state); #else RunInterpreter(state); -#endif // ARCHITECTURE_X64 +#endif // ARCHITECTURE_x86_64 #if PICA_DUMP_SHADERS DebugUtils::DumpShader(setup.program_code.data(), state.debug.max_offset, setup.swizzle_data.data(), -- cgit v1.2.3 From db97090cad236eeeb0909eb1d35cbece15e1f0a5 Mon Sep 17 00:00:00 2001 From: bunnei Date: Sat, 15 Aug 2015 16:51:32 -0400 Subject: Shader: Use a POD struct for registers. --- src/video_core/shader/shader.cpp | 36 ++++++++++++++++++------------------ 1 file changed, 18 insertions(+), 18 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 06c1fe653..6a27a8015 100644 --- a/src/video_core/shader/shader.cpp +++ b/src/video_core/shader/shader.cpp @@ -67,29 +67,29 @@ OutputVertex Run(UnitState& state, const InputVertex& input, int num_attributes) // Setup input register table const auto& attribute_register_map = config.input_register_map; - if (num_attributes > 0) state.input_registers[attribute_register_map.attribute0_register] = input.attr[0]; - if (num_attributes > 1) state.input_registers[attribute_register_map.attribute1_register] = input.attr[1]; - if (num_attributes > 2) state.input_registers[attribute_register_map.attribute2_register] = input.attr[2]; - if (num_attributes > 3) state.input_registers[attribute_register_map.attribute3_register] = input.attr[3]; - if (num_attributes > 4) state.input_registers[attribute_register_map.attribute4_register] = input.attr[4]; - if (num_attributes > 5) state.input_registers[attribute_register_map.attribute5_register] = input.attr[5]; - if (num_attributes > 6) state.input_registers[attribute_register_map.attribute6_register] = input.attr[6]; - if (num_attributes > 7) state.input_registers[attribute_register_map.attribute7_register] = input.attr[7]; - if (num_attributes > 8) state.input_registers[attribute_register_map.attribute8_register] = input.attr[8]; - if (num_attributes > 9) state.input_registers[attribute_register_map.attribute9_register] = input.attr[9]; - if (num_attributes > 10) state.input_registers[attribute_register_map.attribute10_register] = input.attr[10]; - if (num_attributes > 11) state.input_registers[attribute_register_map.attribute11_register] = input.attr[11]; - if (num_attributes > 12) state.input_registers[attribute_register_map.attribute12_register] = input.attr[12]; - if (num_attributes > 13) state.input_registers[attribute_register_map.attribute13_register] = input.attr[13]; - if (num_attributes > 14) state.input_registers[attribute_register_map.attribute14_register] = input.attr[14]; - if (num_attributes > 15) state.input_registers[attribute_register_map.attribute15_register] = input.attr[15]; + if (num_attributes > 0) state.registers.input[attribute_register_map.attribute0_register] = input.attr[0]; + if (num_attributes > 1) state.registers.input[attribute_register_map.attribute1_register] = input.attr[1]; + if (num_attributes > 2) state.registers.input[attribute_register_map.attribute2_register] = input.attr[2]; + if (num_attributes > 3) state.registers.input[attribute_register_map.attribute3_register] = input.attr[3]; + if (num_attributes > 4) state.registers.input[attribute_register_map.attribute4_register] = input.attr[4]; + if (num_attributes > 5) state.registers.input[attribute_register_map.attribute5_register] = input.attr[5]; + if (num_attributes > 6) state.registers.input[attribute_register_map.attribute6_register] = input.attr[6]; + if (num_attributes > 7) state.registers.input[attribute_register_map.attribute7_register] = input.attr[7]; + if (num_attributes > 8) state.registers.input[attribute_register_map.attribute8_register] = input.attr[8]; + if (num_attributes > 9) state.registers.input[attribute_register_map.attribute9_register] = input.attr[9]; + if (num_attributes > 10) state.registers.input[attribute_register_map.attribute10_register] = input.attr[10]; + if (num_attributes > 11) state.registers.input[attribute_register_map.attribute11_register] = input.attr[11]; + if (num_attributes > 12) state.registers.input[attribute_register_map.attribute12_register] = input.attr[12]; + if (num_attributes > 13) state.registers.input[attribute_register_map.attribute13_register] = input.attr[13]; + if (num_attributes > 14) state.registers.input[attribute_register_map.attribute14_register] = input.attr[14]; + if (num_attributes > 15) state.registers.input[attribute_register_map.attribute15_register] = input.attr[15]; state.conditional_code[0] = false; state.conditional_code[1] = false; #ifdef ARCHITECTURE_x86_64 if (VideoCore::g_shader_jit_enabled) - jit_shader(&state); + jit_shader(&state.registers); else RunInterpreter(state); #else @@ -117,7 +117,7 @@ OutputVertex Run(UnitState& state, const InputVertex& input, int num_attributes) for (int comp = 0; comp < 4; ++comp) { float24* out = ((float24*)&ret) + semantics[comp]; if (semantics[comp] != Regs::VSOutputAttributes::INVALID) { - *out = state.output_registers[i][comp]; + *out = state.registers.output[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