Exverge/suyu
3
0
Fork 0
forked from suyu/suyu
Code Pull requests Activity

Merge pull request #3473 from ReinUsesLisp/shader-purge

Browse source 
gl_shader_cache: Rework shader cache and store texture arrays
This commit is contained in:
Fernando Sahmkow 2020-03-13 16:26:24 -04:00 committed by GitHub
commit 666d431ad8
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
34 changed files with 1006 additions and 1435 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

6
CMakeModules/GenerateSCMRev.cmake
View file

@ -57,8 +57,6 @@ set(HASH_FILES
"${VIDEO_CORE}/renderer_opengl/gl_shader_decompiler.h"
"${VIDEO_CORE}/renderer_opengl/gl_shader_disk_cache.cpp"
"${VIDEO_CORE}/renderer_opengl/gl_shader_disk_cache.h"
"${VIDEO_CORE}/renderer_opengl/gl_shader_gen.cpp"
"${VIDEO_CORE}/renderer_opengl/gl_shader_gen.h"
"${VIDEO_CORE}/shader/decode/arithmetic.cpp"
"${VIDEO_CORE}/shader/decode/arithmetic_half.cpp"
"${VIDEO_CORE}/shader/decode/arithmetic_half_immediate.cpp"
@ -91,8 +89,6 @@ set(HASH_FILES
"${VIDEO_CORE}/shader/ast.h"
"${VIDEO_CORE}/shader/compiler_settings.cpp"
"${VIDEO_CORE}/shader/compiler_settings.h"
"${VIDEO_CORE}/shader/const_buffer_locker.cpp"
"${VIDEO_CORE}/shader/const_buffer_locker.h"
"${VIDEO_CORE}/shader/control_flow.cpp"
"${VIDEO_CORE}/shader/control_flow.h"
"${VIDEO_CORE}/shader/decode.cpp"
@ -101,6 +97,8 @@ set(HASH_FILES
"${VIDEO_CORE}/shader/node.h"
"${VIDEO_CORE}/shader/node_helper.cpp"
"${VIDEO_CORE}/shader/node_helper.h"
"${VIDEO_CORE}/shader/registry.cpp"
"${VIDEO_CORE}/shader/registry.h"
"${VIDEO_CORE}/shader/shader_ir.cpp"
"${VIDEO_CORE}/shader/shader_ir.h"
"${VIDEO_CORE}/shader/track.cpp"

6
src/common/CMakeLists.txt
View file

@ -38,8 +38,6 @@ add_custom_command(OUTPUT scm_rev.cpp
"${VIDEO_CORE}/renderer_opengl/gl_shader_decompiler.h"
"${VIDEO_CORE}/renderer_opengl/gl_shader_disk_cache.cpp"
"${VIDEO_CORE}/renderer_opengl/gl_shader_disk_cache.h"
"${VIDEO_CORE}/renderer_opengl/gl_shader_gen.cpp"
"${VIDEO_CORE}/renderer_opengl/gl_shader_gen.h"
"${VIDEO_CORE}/shader/decode/arithmetic.cpp"
"${VIDEO_CORE}/shader/decode/arithmetic_half.cpp"
"${VIDEO_CORE}/shader/decode/arithmetic_half_immediate.cpp"
@ -72,8 +70,6 @@ add_custom_command(OUTPUT scm_rev.cpp
"${VIDEO_CORE}/shader/ast.h"
"${VIDEO_CORE}/shader/compiler_settings.cpp"
"${VIDEO_CORE}/shader/compiler_settings.h"
"${VIDEO_CORE}/shader/const_buffer_locker.cpp"
"${VIDEO_CORE}/shader/const_buffer_locker.h"
"${VIDEO_CORE}/shader/control_flow.cpp"
"${VIDEO_CORE}/shader/control_flow.h"
"${VIDEO_CORE}/shader/decode.cpp"
@ -82,6 +78,8 @@ add_custom_command(OUTPUT scm_rev.cpp
"${VIDEO_CORE}/shader/node.h"
"${VIDEO_CORE}/shader/node_helper.cpp"
"${VIDEO_CORE}/shader/node_helper.h"
"${VIDEO_CORE}/shader/registry.cpp"
"${VIDEO_CORE}/shader/registry.h"
"${VIDEO_CORE}/shader/shader_ir.cpp"
"${VIDEO_CORE}/shader/shader_ir.h"
"${VIDEO_CORE}/shader/track.cpp"

6
src/video_core/CMakeLists.txt
View file

@ -65,8 +65,6 @@ add_library(video_core STATIC
renderer_opengl/gl_shader_decompiler.h
renderer_opengl/gl_shader_disk_cache.cpp
renderer_opengl/gl_shader_disk_cache.h
renderer_opengl/gl_shader_gen.cpp
renderer_opengl/gl_shader_gen.h
renderer_opengl/gl_shader_manager.cpp
renderer_opengl/gl_shader_manager.h
renderer_opengl/gl_shader_util.cpp
@ -118,8 +116,6 @@ add_library(video_core STATIC
shader/ast.h
shader/compiler_settings.cpp
shader/compiler_settings.h
shader/const_buffer_locker.cpp
shader/const_buffer_locker.h
shader/control_flow.cpp
shader/control_flow.h
shader/decode.cpp
@ -128,6 +124,8 @@ add_library(video_core STATIC
shader/node_helper.cpp
shader/node_helper.h
shader/node.h
shader/registry.cpp
shader/registry.h
shader/shader_ir.cpp
shader/shader_ir.h
shader/track.cpp

67
src/video_core/engines/const_buffer_engine_interface.h
View file

@ -16,11 +16,12 @@ namespace Tegra::Engines {
struct SamplerDescriptor {
union {
BitField<0, 20, Tegra::Shader::TextureType> texture_type;
BitField<20, 1, u32> is_array;
BitField<21, 1, u32> is_buffer;
BitField<22, 1, u32> is_shadow;
u32 raw{};
u32 raw = 0;
BitField<0, 2, Tegra::Shader::TextureType> texture_type;
BitField<2, 3, Tegra::Texture::ComponentType> component_type;
BitField<5, 1, u32> is_array;
BitField<6, 1, u32> is_buffer;
BitField<7, 1, u32> is_shadow;
};
bool operator==(const SamplerDescriptor& rhs) const noexcept {
@ -31,68 +32,48 @@ struct SamplerDescriptor {
return !operator==(rhs);
}
static SamplerDescriptor FromTicTexture(Tegra::Texture::TextureType tic_texture_type) {
static SamplerDescriptor FromTIC(const Tegra::Texture::TICEntry& tic) {
using Tegra::Shader::TextureType;
SamplerDescriptor result;
switch (tic_texture_type) {
// This is going to be used to determine the shading language type.
// Because of that we don't care about all component types on color textures.
result.component_type.Assign(tic.r_type.Value());
switch (tic.texture_type.Value()) {
case Tegra::Texture::TextureType::Texture1D:
result.texture_type.Assign(Tegra::Shader::TextureType::Texture1D);
result.is_array.Assign(0);
result.is_buffer.Assign(0);
result.is_shadow.Assign(0);
result.texture_type.Assign(TextureType::Texture1D);
return result;
case Tegra::Texture::TextureType::Texture2D:
result.texture_type.Assign(Tegra::Shader::TextureType::Texture2D);
result.is_array.Assign(0);
result.is_buffer.Assign(0);
result.is_shadow.Assign(0);
result.texture_type.Assign(TextureType::Texture2D);
return result;
case Tegra::Texture::TextureType::Texture3D:
result.texture_type.Assign(Tegra::Shader::TextureType::Texture3D);
result.is_array.Assign(0);
result.is_buffer.Assign(0);
result.is_shadow.Assign(0);
result.texture_type.Assign(TextureType::Texture3D);
return result;
case Tegra::Texture::TextureType::TextureCubemap:
result.texture_type.Assign(Tegra::Shader::TextureType::TextureCube);
result.is_array.Assign(0);
result.is_buffer.Assign(0);
result.is_shadow.Assign(0);
result.texture_type.Assign(TextureType::TextureCube);
return result;
case Tegra::Texture::TextureType::Texture1DArray:
result.texture_type.Assign(Tegra::Shader::TextureType::Texture1D);
result.texture_type.Assign(TextureType::Texture1D);
result.is_array.Assign(1);
result.is_buffer.Assign(0);
result.is_shadow.Assign(0);
return result;
case Tegra::Texture::TextureType::Texture2DArray:
result.texture_type.Assign(Tegra::Shader::TextureType::Texture2D);
result.texture_type.Assign(TextureType::Texture2D);
result.is_array.Assign(1);
result.is_buffer.Assign(0);
result.is_shadow.Assign(0);
return result;
case Tegra::Texture::TextureType::Texture1DBuffer:
result.texture_type.Assign(Tegra::Shader::TextureType::Texture1D);
result.is_array.Assign(0);
result.texture_type.Assign(TextureType::Texture1D);
result.is_buffer.Assign(1);
result.is_shadow.Assign(0);
return result;
case Tegra::Texture::TextureType::Texture2DNoMipmap:
result.texture_type.Assign(Tegra::Shader::TextureType::Texture2D);
result.is_array.Assign(0);
result.is_buffer.Assign(0);
result.is_shadow.Assign(0);
result.texture_type.Assign(TextureType::Texture2D);
return result;
case Tegra::Texture::TextureType::TextureCubeArray:
result.texture_type.Assign(Tegra::Shader::TextureType::TextureCube);
result.texture_type.Assign(TextureType::TextureCube);
result.is_array.Assign(1);
result.is_buffer.Assign(0);
result.is_shadow.Assign(0);
return result;
default:
result.texture_type.Assign(Tegra::Shader::TextureType::Texture2D);
result.is_array.Assign(0);
result.is_buffer.Assign(0);
result.is_shadow.Assign(0);
result.texture_type.Assign(TextureType::Texture2D);
return result;
}
}

2
src/video_core/engines/kepler_compute.cpp
View file

@ -89,7 +89,7 @@ SamplerDescriptor KeplerCompute::AccessBindlessSampler(ShaderType stage, u64 con
const Texture::TextureHandle tex_handle{memory_manager.Read<u32>(tex_info_address)};
const Texture::FullTextureInfo tex_info = GetTextureInfo(tex_handle);
SamplerDescriptor result = SamplerDescriptor::FromTicTexture(tex_info.tic.texture_type.Value());
SamplerDescriptor result = SamplerDescriptor::FromTIC(tex_info.tic);
result.is_shadow.Assign(tex_info.tsc.depth_compare_enabled.Value());
return result;
}

2
src/video_core/engines/maxwell_3d.cpp
View file

@ -638,7 +638,7 @@ SamplerDescriptor Maxwell3D::AccessBindlessSampler(ShaderType stage, u64 const_b
const Texture::TextureHandle tex_handle{memory_manager.Read<u32>(tex_info_address)};
const Texture::FullTextureInfo tex_info = GetTextureInfo(tex_handle);
SamplerDescriptor result = SamplerDescriptor::FromTicTexture(tex_info.tic.texture_type.Value());
SamplerDescriptor result = SamplerDescriptor::FromTIC(tex_info.tic);
result.is_shadow.Assign(tex_info.tsc.depth_compare_enabled.Value());
return result;
}

34
src/video_core/engines/maxwell_3d.h
View file

@ -67,6 +67,7 @@ public:
static constexpr std::size_t NumVaryings = 31;
static constexpr std::size_t NumImages = 8; // TODO(Rodrigo): Investigate this number
static constexpr std::size_t NumClipDistances = 8;
static constexpr std::size_t NumTransformFeedbackBuffers = 4;
static constexpr std::size_t MaxShaderProgram = 6;
static constexpr std::size_t MaxShaderStage = 5;
// Maximum number of const buffers per shader stage.
@ -627,6 +628,22 @@ public:
float depth_range_far;
};
struct alignas(32) TransformFeedbackBinding {
u32 buffer_enable;
u32 address_high;
u32 address_low;
s32 buffer_size;
s32 buffer_offset;
};
static_assert(sizeof(TransformFeedbackBinding) == 32);
struct alignas(16) TransformFeedbackLayout {
u32 stream;
u32 varying_count;
u32 stride;
};
static_assert(sizeof(TransformFeedbackLayout) == 16);
bool IsShaderConfigEnabled(std::size_t index) const {
// The VertexB is always enabled.
if (index == static_cast<std::size_t>(Regs::ShaderProgram::VertexB)) {
@ -683,7 +700,13 @@ public:
u32 rasterize_enable;
INSERT_UNION_PADDING_WORDS(0xF1);
std::array<TransformFeedbackBinding, NumTransformFeedbackBuffers> tfb_bindings;
INSERT_UNION_PADDING_WORDS(0xC0);
std::array<TransformFeedbackLayout, NumTransformFeedbackBuffers> tfb_layouts;
INSERT_UNION_PADDING_WORDS(0x1);
u32 tfb_enabled;
@ -1202,7 +1225,11 @@ public:
u32 tex_cb_index;
INSERT_UNION_PADDING_WORDS(0x395);
INSERT_UNION_PADDING_WORDS(0x7D);
std::array<std::array<u8, 128>, NumTransformFeedbackBuffers> tfb_varying_locs;
INSERT_UNION_PADDING_WORDS(0x298);
struct {
/// Compressed address of a buffer that holds information about bound SSBOs.
@ -1428,6 +1455,8 @@ ASSERT_REG_POSITION(tess_mode, 0xC8);
ASSERT_REG_POSITION(tess_level_outer, 0xC9);
ASSERT_REG_POSITION(tess_level_inner, 0xCD);
ASSERT_REG_POSITION(rasterize_enable, 0xDF);
ASSERT_REG_POSITION(tfb_bindings, 0xE0);
ASSERT_REG_POSITION(tfb_layouts, 0x1C0);
ASSERT_REG_POSITION(tfb_enabled, 0x1D1);
ASSERT_REG_POSITION(rt, 0x200);
ASSERT_REG_POSITION(viewport_transform, 0x280);
@ -1526,6 +1555,7 @@ ASSERT_REG_POSITION(firmware, 0x8C0);
ASSERT_REG_POSITION(const_buffer, 0x8E0);
ASSERT_REG_POSITION(cb_bind[0], 0x904);
ASSERT_REG_POSITION(tex_cb_index, 0x982);
ASSERT_REG_POSITION(tfb_varying_locs, 0xA00);
ASSERT_REG_POSITION(ssbo_info, 0xD18);
ASSERT_REG_POSITION(tex_info_buffers.address[0], 0xD2A);
ASSERT_REG_POSITION(tex_info_buffers.size[0], 0xD2F);

7
src/video_core/guest_driver.cpp
View file

@ -4,13 +4,15 @@
#include <algorithm>
#include <limits>
#include <vector>
#include "common/common_types.h"
#include "video_core/guest_driver.h"
namespace VideoCore {
void GuestDriverProfile::DeduceTextureHandlerSize(std::vector<u32>&& bound_offsets) {
if (texture_handler_size_deduced) {
void GuestDriverProfile::DeduceTextureHandlerSize(std::vector<u32> bound_offsets) {
if (texture_handler_size) {
return;
}
const std::size_t size = bound_offsets.size();
@ -29,7 +31,6 @@ void GuestDriverProfile::DeduceTextureHandlerSize(std::vector<u32>&& bound_offse
if (min_val > 2) {
return;
}
texture_handler_size_deduced = true;
texture_handler_size = min_texture_handler_size * min_val;
}

21
src/video_core/guest_driver.h
View file

@ -4,6 +4,7 @@
#pragma once
#include <optional>
#include <vector>
#include "common/common_types.h"
@ -17,25 +18,29 @@ namespace VideoCore {
*/
class GuestDriverProfile {
public:
void DeduceTextureHandlerSize(std::vector<u32>&& bound_offsets);
explicit GuestDriverProfile() = default;
explicit GuestDriverProfile(std::optional<u32> texture_handler_size)
: texture_handler_size{texture_handler_size} {}
void DeduceTextureHandlerSize(std::vector<u32> bound_offsets);
u32 GetTextureHandlerSize() const {
return texture_handler_size;
return texture_handler_size.value_or(default_texture_handler_size);
}
bool TextureHandlerSizeKnown() const {
return texture_handler_size_deduced;
bool IsTextureHandlerSizeKnown() const {
return texture_handler_size.has_value();
}
private:
// Minimum size of texture handler any driver can use.
static constexpr u32 min_texture_handler_size = 4;
// This goes with Vulkan and OpenGL standards but Nvidia GPUs can easily
// use 4 bytes instead. Thus, certain drivers may squish the size.
// This goes with Vulkan and OpenGL standards but Nvidia GPUs can easily use 4 bytes instead.
// Thus, certain drivers may squish the size.
static constexpr u32 default_texture_handler_size = 8;
u32 texture_handler_size = default_texture_handler_size;
bool texture_handler_size_deduced = false;
std::optional<u32> texture_handler_size = default_texture_handler_size;
};
} // namespace VideoCore

1
src/video_core/rasterizer_interface.h
View file

@ -25,7 +25,6 @@ constexpr std::size_t NumQueryTypes = 1;
enum class LoadCallbackStage {
Prepare,
Decompile,
Build,
Complete,
};

41
src/video_core/renderer_opengl/gl_rasterizer.cpp
View file

@ -28,7 +28,6 @@
#include "video_core/renderer_opengl/gl_query_cache.h"
#include "video_core/renderer_opengl/gl_rasterizer.h"
#include "video_core/renderer_opengl/gl_shader_cache.h"
#include "video_core/renderer_opengl/gl_shader_gen.h"
#include "video_core/renderer_opengl/maxwell_to_gl.h"
#include "video_core/renderer_opengl/renderer_opengl.h"
@ -76,7 +75,7 @@ Tegra::Texture::FullTextureInfo GetTextureInfo(const Engine& engine, const Entry
}
std::size_t GetConstBufferSize(const Tegra::Engines::ConstBufferInfo& buffer,
const GLShader::ConstBufferEntry& entry) {
const ConstBufferEntry& entry) {
if (!entry.IsIndirect()) {
return entry.GetSize();
}
@ -272,9 +271,7 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) {
SetupDrawTextures(stage, shader);
SetupDrawImages(stage, shader);
const ProgramVariant variant(primitive_mode);
const auto program_handle = shader->GetHandle(variant);
const GLuint program_handle = shader->GetHandle();
switch (program) {
case Maxwell::ShaderProgram::VertexA:
case Maxwell::ShaderProgram::VertexB:
@ -295,7 +292,7 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) {
// When a clip distance is enabled but not set in the shader it crops parts of the screen
// (sometimes it's half the screen, sometimes three quarters). To avoid this, enable the
// clip distances only when it's written by a shader stage.
clip_distances |= shader->GetShaderEntries().clip_distances;
clip_distances |= shader->GetEntries().clip_distances;
// When VertexA is enabled, we have dual vertex shaders
if (program == Maxwell::ShaderProgram::VertexA) {
@ -623,12 +620,7 @@ void RasterizerOpenGL::DispatchCompute(GPUVAddr code_addr) {
auto kernel = shader_cache.GetComputeKernel(code_addr);
SetupComputeTextures(kernel);
SetupComputeImages(kernel);
const auto& launch_desc = system.GPU().KeplerCompute().launch_description;
const ProgramVariant variant(launch_desc.block_dim_x, launch_desc.block_dim_y,
launch_desc.block_dim_z, launch_desc.shared_alloc,
launch_desc.local_pos_alloc);
program_manager.BindComputeShader(kernel->GetHandle(variant));
program_manager.BindComputeShader(kernel->GetHandle());
const std::size_t buffer_size =
Tegra::Engines::KeplerCompute::NumConstBuffers *
@ -646,6 +638,7 @@ void RasterizerOpenGL::DispatchCompute(GPUVAddr code_addr) {
bind_ubo_pushbuffer.Bind();
bind_ssbo_pushbuffer.Bind();
const auto& launch_desc = system.GPU().KeplerCompute().launch_description;
glDispatchCompute(launch_desc.grid_dim_x, launch_desc.grid_dim_y, launch_desc.grid_dim_z);
++num_queued_commands;
}
@ -750,7 +743,7 @@ void RasterizerOpenGL::SetupDrawConstBuffers(std::size_t stage_index, const Shad
const auto& shader_stage = stages[stage_index];
u32 binding = device.GetBaseBindings(stage_index).uniform_buffer;
for (const auto& entry : shader->GetShaderEntries().const_buffers) {
for (const auto& entry : shader->GetEntries().const_buffers) {
const auto& buffer = shader_stage.const_buffers[entry.GetIndex()];
SetupConstBuffer(binding++, buffer, entry);
}
@ -761,7 +754,7 @@ void RasterizerOpenGL::SetupComputeConstBuffers(const Shader& kernel) {
const auto& launch_desc = system.GPU().KeplerCompute().launch_description;
u32 binding = 0;
for (const auto& entry : kernel->GetShaderEntries().const_buffers) {
for (const auto& entry : kernel->GetEntries().const_buffers) {
const auto& config = launch_desc.const_buffer_config[entry.GetIndex()];
const std::bitset<8> mask = launch_desc.const_buffer_enable_mask.Value();
Tegra::Engines::ConstBufferInfo buffer;
@ -773,7 +766,7 @@ void RasterizerOpenGL::SetupComputeConstBuffers(const Shader& kernel) {
}
void RasterizerOpenGL::SetupConstBuffer(u32 binding, const Tegra::Engines::ConstBufferInfo& buffer,
const GLShader::ConstBufferEntry& entry) {
const ConstBufferEntry& entry) {
if (!buffer.enabled) {
// Set values to zero to unbind buffers
bind_ubo_pushbuffer.Push(binding, buffer_cache.GetEmptyBuffer(sizeof(float)), 0,
@ -797,7 +790,7 @@ void RasterizerOpenGL::SetupDrawGlobalMemory(std::size_t stage_index, const Shad
const auto cbufs{gpu.Maxwell3D().state.shader_stages[stage_index]};
u32 binding = device.GetBaseBindings(stage_index).shader_storage_buffer;
for (const auto& entry : shader->GetShaderEntries().global_memory_entries) {
for (const auto& entry : shader->GetEntries().global_memory_entries) {
const auto addr{cbufs.const_buffers[entry.GetCbufIndex()].address + entry.GetCbufOffset()};
const auto gpu_addr{memory_manager.Read<u64>(addr)};
const auto size{memory_manager.Read<u32>(addr + 8)};
@ -811,7 +804,7 @@ void RasterizerOpenGL::SetupComputeGlobalMemory(const Shader& kernel) {
const auto cbufs{gpu.KeplerCompute().launch_description.const_buffer_config};
u32 binding = 0;
for (const auto& entry : kernel->GetShaderEntries().global_memory_entries) {
for (const auto& entry : kernel->GetEntries().global_memory_entries) {
const auto addr{cbufs[entry.GetCbufIndex()].Address() + entry.GetCbufOffset()};
const auto gpu_addr{memory_manager.Read<u64>(addr)};
const auto size{memory_manager.Read<u32>(addr + 8)};
@ -819,7 +812,7 @@ void RasterizerOpenGL::SetupComputeGlobalMemory(const Shader& kernel) {
}
}
void RasterizerOpenGL::SetupGlobalMemory(u32 binding, const GLShader::GlobalMemoryEntry& entry,
void RasterizerOpenGL::SetupGlobalMemory(u32 binding, const GlobalMemoryEntry& entry,
GPUVAddr gpu_addr, std::size_t size) {
const auto alignment{device.GetShaderStorageBufferAlignment()};
const auto [ssbo, buffer_offset] =
@ -831,7 +824,7 @@ void RasterizerOpenGL::SetupDrawTextures(std::size_t stage_index, const Shader&
MICROPROFILE_SCOPE(OpenGL_Texture);
const auto& maxwell3d = system.GPU().Maxwell3D();
u32 binding = device.GetBaseBindings(stage_index).sampler;
for (const auto& entry : shader->GetShaderEntries().samplers) {
for (const auto& entry : shader->GetEntries().samplers) {
const auto shader_type = static_cast<ShaderType>(stage_index);
for (std::size_t i = 0; i < entry.Size(); ++i) {
const auto texture = GetTextureInfo(maxwell3d, entry, shader_type, i);
@ -844,7 +837,7 @@ void RasterizerOpenGL::SetupComputeTextures(const Shader& kernel) {
MICROPROFILE_SCOPE(OpenGL_Texture);
const auto& compute = system.GPU().KeplerCompute();
u32 binding = 0;
for (const auto& entry : kernel->GetShaderEntries().samplers) {
for (const auto& entry : kernel->GetEntries().samplers) {
for (std::size_t i = 0; i < entry.Size(); ++i) {
const auto texture = GetTextureInfo(compute, entry, ShaderType::Compute, i);
SetupTexture(binding++, texture, entry);
@ -853,7 +846,7 @@ void RasterizerOpenGL::SetupComputeTextures(const Shader& kernel) {
}
void RasterizerOpenGL::SetupTexture(u32 binding, const Tegra::Texture::FullTextureInfo& texture,
const GLShader::SamplerEntry& entry) {
const SamplerEntry& entry) {
const auto view = texture_cache.GetTextureSurface(texture.tic, entry);
if (!view) {
// Can occur when texture addr is null or its memory is unmapped/invalid
@ -876,7 +869,7 @@ void RasterizerOpenGL::SetupTexture(u32 binding, const Tegra::Texture::FullTextu
void RasterizerOpenGL::SetupDrawImages(std::size_t stage_index, const Shader& shader) {
const auto& maxwell3d = system.GPU().Maxwell3D();
u32 binding = device.GetBaseBindings(stage_index).image;
for (const auto& entry : shader->GetShaderEntries().images) {
for (const auto& entry : shader->GetEntries().images) {
const auto shader_type = static_cast<Tegra::Engines::ShaderType>(stage_index);
const auto tic = GetTextureInfo(maxwell3d, entry, shader_type).tic;
SetupImage(binding++, tic, entry);
@ -886,14 +879,14 @@ void RasterizerOpenGL::SetupDrawImages(std::size_t stage_index, const Shader& sh
void RasterizerOpenGL::SetupComputeImages(const Shader& shader) {
const auto& compute = system.GPU().KeplerCompute();
u32 binding = 0;
for (const auto& entry : shader->GetShaderEntries().images) {
for (const auto& entry : shader->GetEntries().images) {
const auto tic = GetTextureInfo(compute, entry, Tegra::Engines::ShaderType::Compute).tic;
SetupImage(binding++, tic, entry);
}
}
void RasterizerOpenGL::SetupImage(u32 binding, const Tegra::Texture::TICEntry& tic,
const GLShader::ImageEntry& entry) {
const ImageEntry& entry) {
const auto view = texture_cache.GetImageSurface(tic, entry);
if (!view) {
glBindImageTexture(binding, 0, 0, GL_FALSE, 0, GL_READ_ONLY, GL_R8);

9
src/video_core/renderer_opengl/gl_rasterizer.h
View file

@ -98,7 +98,7 @@ private:
/// Configures a constant buffer.
void SetupConstBuffer(u32 binding, const Tegra::Engines::ConstBufferInfo& buffer,
const GLShader::ConstBufferEntry& entry);
const ConstBufferEntry& entry);
/// Configures the current global memory entries to use for the draw command.
void SetupDrawGlobalMemory(std::size_t stage_index, const Shader& shader);
@ -107,7 +107,7 @@ private:
void SetupComputeGlobalMemory(const Shader& kernel);
/// Configures a constant buffer.
void SetupGlobalMemory(u32 binding, const GLShader::GlobalMemoryEntry& entry, GPUVAddr gpu_addr,
void SetupGlobalMemory(u32 binding, const GlobalMemoryEntry& entry, GPUVAddr gpu_addr,
std::size_t size);
/// Configures the current textures to use for the draw command.
@ -118,7 +118,7 @@ private:
/// Configures a texture.
void SetupTexture(u32 binding, const Tegra::Texture::FullTextureInfo& texture,
const GLShader::SamplerEntry& entry);
const SamplerEntry& entry);
/// Configures images in a graphics shader.
void SetupDrawImages(std::size_t stage_index, const Shader& shader);
@ -127,8 +127,7 @@ private:
void SetupComputeImages(const Shader& shader);
/// Configures an image.
void SetupImage(u32 binding, const Tegra::Texture::TICEntry& tic,
const GLShader::ImageEntry& entry);
void SetupImage(u32 binding, const Tegra::Texture::TICEntry& tic, const ImageEntry& entry);
/// Syncs the viewport and depth range to match the guest state
void SyncViewport();

502
src/video_core/renderer_opengl/gl_shader_cache.cpp
View file

@ -2,12 +2,16 @@
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <atomic>
#include <functional>
#include <mutex>
#include <optional>
#include <string>
#include <thread>
#include <unordered_set>
#include <boost/functional/hash.hpp>
#include "common/alignment.h"
#include "common/assert.h"
#include "common/logging/log.h"
@ -24,13 +28,14 @@
#include "video_core/renderer_opengl/gl_shader_disk_cache.h"
#include "video_core/renderer_opengl/gl_state_tracker.h"
#include "video_core/renderer_opengl/utils.h"
#include "video_core/shader/registry.h"
#include "video_core/shader/shader_ir.h"
namespace OpenGL {
using Tegra::Engines::ShaderType;
using VideoCommon::Shader::ConstBufferLocker;
using VideoCommon::Shader::ProgramCode;
using VideoCommon::Shader::Registry;
using VideoCommon::Shader::ShaderIR;
namespace {
@ -56,7 +61,7 @@ constexpr bool IsSchedInstruction(std::size_t offset, std::size_t main_offset) {
}
/// Calculates the size of a program stream
std::size_t CalculateProgramSize(const GLShader::ProgramCode& program) {
std::size_t CalculateProgramSize(const ProgramCode& program) {
constexpr std::size_t start_offset = 10;
// This is the encoded version of BRA that jumps to itself. All Nvidia
// shaders end with one.
@ -109,32 +114,9 @@ constexpr GLenum GetGLShaderType(ShaderType shader_type) {
}
}
/// Describes primitive behavior on geometry shaders
constexpr std::pair<const char*, u32> GetPrimitiveDescription(GLenum primitive_mode) {
switch (primitive_mode) {
case GL_POINTS:
return {"points", 1};
case GL_LINES:
case GL_LINE_STRIP:
return {"lines", 2};
case GL_LINES_ADJACENCY:
case GL_LINE_STRIP_ADJACENCY:
return {"lines_adjacency", 4};
case GL_TRIANGLES:
case GL_TRIANGLE_STRIP:
case GL_TRIANGLE_FAN:
return {"triangles", 3};
case GL_TRIANGLES_ADJACENCY:
case GL_TRIANGLE_STRIP_ADJACENCY:
return {"triangles_adjacency", 6};
default:
return {"points", 1};
}
}
/// Hashes one (or two) program streams
u64 GetUniqueIdentifier(ShaderType shader_type, bool is_a, const ProgramCode& code,
const ProgramCode& code_b) {
const ProgramCode& code_b = {}) {
u64 unique_identifier = boost::hash_value(code);
if (is_a) {
// VertexA programs include two programs
@ -143,24 +125,6 @@ u64 GetUniqueIdentifier(ShaderType shader_type, bool is_a, const ProgramCode& co
return unique_identifier;
}
/// Creates an unspecialized program from code streams
std::string GenerateGLSL(const Device& device, ShaderType shader_type, const ShaderIR& ir,
const std::optional<ShaderIR>& ir_b) {
switch (shader_type) {
case ShaderType::Vertex:
return GLShader::GenerateVertexShader(device, ir, ir_b ? &*ir_b : nullptr);
case ShaderType::Geometry:
return GLShader::GenerateGeometryShader(device, ir);
case ShaderType::Fragment:
return GLShader::GenerateFragmentShader(device, ir);
case ShaderType::Compute:
return GLShader::GenerateComputeShader(device, ir);
default:
UNIMPLEMENTED_MSG("Unimplemented shader_type={}", static_cast<u32>(shader_type));
return {};
}
}
constexpr const char* GetShaderTypeName(ShaderType shader_type) {
switch (shader_type) {
case ShaderType::Vertex:
@ -196,102 +160,38 @@ constexpr ShaderType GetShaderType(Maxwell::ShaderProgram program_type) {
return {};
}
std::string GetShaderId(u64 unique_identifier, ShaderType shader_type) {
std::string MakeShaderID(u64 unique_identifier, ShaderType shader_type) {
return fmt::format("{}{:016X}", GetShaderTypeName(shader_type), unique_identifier);
}
Tegra::Engines::ConstBufferEngineInterface& GetConstBufferEngineInterface(Core::System& system,
ShaderType shader_type) {
if (shader_type == ShaderType::Compute) {
return system.GPU().KeplerCompute();
} else {
return system.GPU().Maxwell3D();
std::shared_ptr<Registry> MakeRegistry(const ShaderDiskCacheEntry& entry) {
const VideoCore::GuestDriverProfile guest_profile{entry.texture_handler_size};
const VideoCommon::Shader::SerializedRegistryInfo info{guest_profile, entry.bound_buffer,
entry.graphics_info, entry.compute_info};
const auto registry = std::make_shared<Registry>(entry.type, info);
for (const auto& [address, value] : entry.keys) {
const auto [buffer, offset] = address;
registry->InsertKey(buffer, offset, value);
}
}
std::unique_ptr<ConstBufferLocker> MakeLocker(Core::System& system, ShaderType shader_type) {
return std::make_unique<ConstBufferLocker>(shader_type,
GetConstBufferEngineInterface(system, shader_type));
}
void FillLocker(ConstBufferLocker& locker, const ShaderDiskCacheUsage& usage) {
locker.SetBoundBuffer(usage.bound_buffer);
for (const auto& key : usage.keys) {
const auto [buffer, offset] = key.first;
locker.InsertKey(buffer, offset, key.second);
for (const auto& [offset, sampler] : entry.bound_samplers) {
registry->InsertBoundSampler(offset, sampler);
}
for (const auto& [offset, sampler] : usage.bound_samplers) {
locker.InsertBoundSampler(offset, sampler);
}
for (const auto& [key, sampler] : usage.bindless_samplers) {
for (const auto& [key, sampler] : entry.bindless_samplers) {
const auto [buffer, offset] = key;
locker.InsertBindlessSampler(buffer, offset, sampler);
registry->InsertBindlessSampler(buffer, offset, sampler);
}
return registry;
}
CachedProgram BuildShader(const Device& device, u64 unique_identifier, ShaderType shader_type,
const ProgramCode& code, const ProgramCode& code_b,
ConstBufferLocker& locker, const ProgramVariant& variant,
bool hint_retrievable = false) {
LOG_INFO(Render_OpenGL, "called. {}", GetShaderId(unique_identifier, shader_type));
const bool is_compute = shader_type == ShaderType::Compute;
const u32 main_offset = is_compute ? KERNEL_MAIN_OFFSET : STAGE_MAIN_OFFSET;
const ShaderIR ir(code, main_offset, COMPILER_SETTINGS, locker);
std::optional<ShaderIR> ir_b;
if (!code_b.empty()) {
ir_b.emplace(code_b, main_offset, COMPILER_SETTINGS, locker);
}
std::string source = fmt::format(R"(// {}
#version 430 core
#extension GL_ARB_separate_shader_objects : enable
)",
GetShaderId(unique_identifier, shader_type));
if (device.HasShaderBallot()) {
source += "#extension GL_ARB_shader_ballot : require\n";
}
if (device.HasVertexViewportLayer()) {
source += "#extension GL_ARB_shader_viewport_layer_array : require\n";
}
if (device.HasImageLoadFormatted()) {
source += "#extension GL_EXT_shader_image_load_formatted : require\n";
}
if (device.HasWarpIntrinsics()) {
source += "#extension GL_NV_gpu_shader5 : require\n"
"#extension GL_NV_shader_thread_group : require\n"
"#extension GL_NV_shader_thread_shuffle : require\n";
}
// This pragma stops Nvidia's driver from over optimizing math (probably using fp16 operations)
// on places where we don't want to.
// Thanks to Ryujinx for finding this workaround.
source += "#pragma optionNV(fastmath off)\n";
if (shader_type == ShaderType::Geometry) {
const auto [glsl_topology, max_vertices] = GetPrimitiveDescription(variant.primitive_mode);
source += fmt::format("#define MAX_VERTEX_INPUT {}\n", max_vertices);
source += fmt::format("layout ({}) in;\n", glsl_topology);
}
if (shader_type == ShaderType::Compute) {
if (variant.local_memory_size > 0) {
source += fmt::format("#define LOCAL_MEMORY_SIZE {}\n",
Common::AlignUp(variant.local_memory_size, 4) / 4);
}
source +=
fmt::format("layout (local_size_x = {}, local_size_y = {}, local_size_z = {}) in;\n",
variant.block_x, variant.block_y, variant.block_z);
if (variant.shared_memory_size > 0) {
// shared_memory_size is described in number of words
source += fmt::format("shared uint smem[{}];\n", variant.shared_memory_size);
}
}
source += '\n';
source += GenerateGLSL(device, shader_type, ir, ir_b);
std::shared_ptr<OGLProgram> BuildShader(const Device& device, ShaderType shader_type,
u64 unique_identifier, const ShaderIR& ir,
const Registry& registry, bool hint_retrievable = false) {
const std::string shader_id = MakeShaderID(unique_identifier, shader_type);
LOG_INFO(Render_OpenGL, "{}", shader_id);
const std::string glsl = DecompileShader(device, ir, registry, shader_type, shader_id);
OGLShader shader;
shader.Create(source.c_str(), GetGLShaderType(shader_type));
shader.Create(glsl.c_str(), GetGLShaderType(shader_type));
auto program = std::make_shared<OGLProgram>();
program->Create(true, hint_retrievable, shader.handle);
@ -299,7 +199,7 @@ CachedProgram BuildShader(const Device& device, u64 unique_identifier, ShaderTyp
}
std::unordered_set<GLenum> GetSupportedFormats() {
GLint num_formats{};
GLint num_formats;
glGetIntegerv(GL_NUM_PROGRAM_BINARY_FORMATS, &num_formats);
std::vector<GLint> formats(num_formats);
@ -314,115 +214,82 @@ std::unordered_set<GLenum> GetSupportedFormats() {
} // Anonymous namespace
CachedShader::CachedShader(const ShaderParameters& params, ShaderType shader_type,
GLShader::ShaderEntries entries, ProgramCode code, ProgramCode code_b)
: RasterizerCacheObject{params.host_ptr}, system{params.system},
disk_cache{params.disk_cache}, device{params.device}, cpu_addr{params.cpu_addr},
unique_identifier{params.unique_identifier}, shader_type{shader_type},
entries{std::move(entries)}, code{std::move(code)}, code_b{std::move(code_b)} {
if (!params.precompiled_variants) {
return;
}
for (const auto& pair : *params.precompiled_variants) {
auto locker = MakeLocker(system, shader_type);
const auto& usage = pair->first;
FillLocker(*locker, usage);
CachedShader::CachedShader(const u8* host_ptr, VAddr cpu_addr, std::size_t size_in_bytes,
std::shared_ptr<VideoCommon::Shader::Registry> registry,
ShaderEntries entries, std::shared_ptr<OGLProgram> program)
: RasterizerCacheObject{host_ptr}, registry{std::move(registry)}, entries{std::move(entries)},
cpu_addr{cpu_addr}, size_in_bytes{size_in_bytes}, program{std::move(program)} {}
std::unique_ptr<LockerVariant>* locker_variant = nullptr;
const auto it =
std::find_if(locker_variants.begin(), locker_variants.end(), [&](const auto& variant) {
return variant->locker->HasEqualKeys(*locker);
});
if (it == locker_variants.end()) {
locker_variant = &locker_variants.emplace_back();
*locker_variant = std::make_unique<LockerVariant>();
locker_variant->get()->locker = std::move(locker);
} else {
locker_variant = &*it;
}
locker_variant->get()->programs.emplace(usage.variant, pair->second);
}
CachedShader::~CachedShader() = default;
GLuint CachedShader::GetHandle() const {
DEBUG_ASSERT(registry->IsConsistent());
return program->handle;
}
Shader CachedShader::CreateStageFromMemory(const ShaderParameters& params,
Maxwell::ShaderProgram program_type, ProgramCode code,
ProgramCode code_b) {
const auto shader_type = GetShaderType(program_type);
params.disk_cache.SaveRaw(
ShaderDiskCacheRaw(params.unique_identifier, shader_type, code, code_b));
const std::size_t size_in_bytes = code.size() * sizeof(u64);
ConstBufferLocker locker(shader_type, params.system.GPU().Maxwell3D());
const ShaderIR ir(code, STAGE_MAIN_OFFSET, COMPILER_SETTINGS, locker);
auto registry = std::make_shared<Registry>(shader_type, params.system.GPU().Maxwell3D());
const ShaderIR ir(code, STAGE_MAIN_OFFSET, COMPILER_SETTINGS, *registry);
// TODO(Rodrigo): Handle VertexA shaders
// std::optional<ShaderIR> ir_b;
// if (!code_b.empty()) {
// ir_b.emplace(code_b, STAGE_MAIN_OFFSET);
// }
return std::shared_ptr<CachedShader>(new CachedShader(
params, shader_type, GLShader::GetEntries(ir), std::move(code), std::move(code_b)));
auto program = BuildShader(params.device, shader_type, params.unique_identifier, ir, *registry);
ShaderDiskCacheEntry entry;
entry.type = shader_type;
entry.code = std::move(code);
entry.code_b = std::move(code_b);
entry.unique_identifier = params.unique_identifier;
entry.bound_buffer = registry->GetBoundBuffer();
entry.graphics_info = registry->GetGraphicsInfo();
entry.keys = registry->GetKeys();
entry.bound_samplers = registry->GetBoundSamplers();
entry.bindless_samplers = registry->GetBindlessSamplers();
params.disk_cache.SaveEntry(std::move(entry));
return std::shared_ptr<CachedShader>(new CachedShader(params.host_ptr, params.cpu_addr,
size_in_bytes, std::move(registry),
MakeEntries(ir), std::move(program)));
}
Shader CachedShader::CreateKernelFromMemory(const ShaderParameters& params, ProgramCode code) {
params.disk_cache.SaveRaw(
ShaderDiskCacheRaw(params.unique_identifier, ShaderType::Compute, code));
const std::size_t size_in_bytes = code.size() * sizeof(u64);
ConstBufferLocker locker(Tegra::Engines::ShaderType::Compute,
params.system.GPU().KeplerCompute());
const ShaderIR ir(code, KERNEL_MAIN_OFFSET, COMPILER_SETTINGS, locker);
return std::shared_ptr<CachedShader>(new CachedShader(
params, ShaderType::Compute, GLShader::GetEntries(ir), std::move(code), {}));
auto& engine = params.system.GPU().KeplerCompute();
auto registry = std::make_shared<Registry>(ShaderType::Compute, engine);
const ShaderIR ir(code, KERNEL_MAIN_OFFSET, COMPILER_SETTINGS, *registry);
const u64 uid = params.unique_identifier;
auto program = BuildShader(params.device, ShaderType::Compute, uid, ir, *registry);
ShaderDiskCacheEntry entry;
entry.type = ShaderType::Compute;
entry.code = std::move(code);
entry.unique_identifier = uid;
entry.bound_buffer = registry->GetBoundBuffer();
entry.compute_info = registry->GetComputeInfo();
entry.keys = registry->GetKeys();
entry.bound_samplers = registry->GetBoundSamplers();
entry.bindless_samplers = registry->GetBindlessSamplers();
params.disk_cache.SaveEntry(std::move(entry));
return std::shared_ptr<CachedShader>(new CachedShader(params.host_ptr, params.cpu_addr,
size_in_bytes, std::move(registry),
MakeEntries(ir), std::move(program)));
}
Shader CachedShader::CreateFromCache(const ShaderParameters& params,
const UnspecializedShader& unspecialized) {
return std::shared_ptr<CachedShader>(new CachedShader(params, unspecialized.type,
unspecialized.entries, unspecialized.code,
unspecialized.code_b));
}
GLuint CachedShader::GetHandle(const ProgramVariant& variant) {
EnsureValidLockerVariant();
const auto [entry, is_cache_miss] = curr_locker_variant->programs.try_emplace(variant);
auto& program = entry->second;
if (!is_cache_miss) {
return program->handle;
}
program = BuildShader(device, unique_identifier, shader_type, code, code_b,
*curr_locker_variant->locker, variant);
disk_cache.SaveUsage(GetUsage(variant, *curr_locker_variant->locker));
LabelGLObject(GL_PROGRAM, program->handle, cpu_addr);
return program->handle;
}
bool CachedShader::EnsureValidLockerVariant() {
const auto previous_variant = curr_locker_variant;
if (curr_locker_variant && !curr_locker_variant->locker->IsConsistent()) {
curr_locker_variant = nullptr;
}
if (!curr_locker_variant) {
for (auto& variant : locker_variants) {
if (variant->locker->IsConsistent()) {
curr_locker_variant = variant.get();
}
}
}
if (!curr_locker_variant) {
auto& new_variant = locker_variants.emplace_back();
new_variant = std::make_unique<LockerVariant>();
new_variant->locker = MakeLocker(system, shader_type);
curr_locker_variant = new_variant.get();
}
return previous_variant == curr_locker_variant;
}
ShaderDiskCacheUsage CachedShader::GetUsage(const ProgramVariant& variant,
const ConstBufferLocker& locker) const {
return ShaderDiskCacheUsage{unique_identifier, variant,
locker.GetBoundBuffer(), locker.GetKeys(),
locker.GetBoundSamplers(), locker.GetBindlessSamplers()};
const PrecompiledShader& precompiled_shader,
std::size_t size_in_bytes) {
return std::shared_ptr<CachedShader>(new CachedShader(
params.host_ptr, params.cpu_addr, size_in_bytes, precompiled_shader.registry,
precompiled_shader.entries, precompiled_shader.program));
}
ShaderCacheOpenGL::ShaderCacheOpenGL(RasterizerOpenGL& rasterizer, Core::System& system,
@ -432,16 +299,12 @@ ShaderCacheOpenGL::ShaderCacheOpenGL(RasterizerOpenGL& rasterizer, Core::System&
void ShaderCacheOpenGL::LoadDiskCache(const std::atomic_bool& stop_loading,
const VideoCore::DiskResourceLoadCallback& callback) {
const auto transferable = disk_cache.LoadTransferable();
const std::optional transferable = disk_cache.LoadTransferable();
if (!transferable) {
return;
}
const auto [raws, shader_usages] = *transferable;
if (!GenerateUnspecializedShaders(stop_loading, callback, raws) || stop_loading) {
return;
}
const auto dumps = disk_cache.LoadPrecompiled();
const std::vector gl_cache = disk_cache.LoadPrecompiled();
const auto supported_formats = GetSupportedFormats();
// Track if precompiled cache was altered during loading to know if we have to
@ -450,77 +313,82 @@ void ShaderCacheOpenGL::LoadDiskCache(const std::atomic_bool& stop_loading,
// Inform the frontend about shader build initialization
if (callback) {
callback(VideoCore::LoadCallbackStage::Build, 0, shader_usages.size());
callback(VideoCore::LoadCallbackStage::Build, 0, transferable->size());
}
std::mutex mutex;
std::size_t built_shaders = 0; // It doesn't have be atomic since it's used behind a mutex
std::atomic_bool compilation_failed = false;
std::atomic_bool gl_cache_failed = false;
const auto Worker = [&](Core::Frontend::GraphicsContext* context, std::size_t begin,
std::size_t end, const std::vector<ShaderDiskCacheUsage>& shader_usages,
const ShaderDumpsMap& dumps) {
const auto find_precompiled = [&gl_cache](u64 id) {
return std::find_if(gl_cache.begin(), gl_cache.end(),
[id](const auto& entry) { return entry.unique_identifier == id; });
};
const auto worker = [&](Core::Frontend::GraphicsContext* context, std::size_t begin,
std::size_t end) {
context->MakeCurrent();
SCOPE_EXIT({ return context->DoneCurrent(); });
for (std::size_t i = begin; i < end; ++i) {
if (stop_loading || compilation_failed) {
if (stop_loading) {
return;
}
const auto& usage{shader_usages[i]};
const auto& unspecialized{unspecialized_shaders.at(usage.unique_identifier)};
const auto dump{dumps.find(usage)};
const auto& entry = (*transferable)[i];
const u64 uid = entry.unique_identifier;
const auto it = find_precompiled(uid);
const auto precompiled_entry = it != gl_cache.end() ? &*it : nullptr;
CachedProgram shader;
if (dump != dumps.end()) {
// If the shader is dumped, attempt to load it with
shader = GeneratePrecompiledProgram(dump->second, supported_formats);
if (!shader) {
compilation_failed = true;
return;
const bool is_compute = entry.type == ShaderType::Compute;
const u32 main_offset = is_compute ? KERNEL_MAIN_OFFSET : STAGE_MAIN_OFFSET;
auto registry = MakeRegistry(entry);
const ShaderIR ir(entry.code, main_offset, COMPILER_SETTINGS, *registry);
std::shared_ptr<OGLProgram> program;
if (precompiled_entry) {
// If the shader is precompiled, attempt to load it with
program = GeneratePrecompiledProgram(entry, *precompiled_entry, supported_formats);
if (!program) {
gl_cache_failed = true;
}
}
if (!shader) {
auto locker{MakeLocker(system, unspecialized.type)};
FillLocker(*locker, usage);
shader = BuildShader(device, usage.unique_identifier, unspecialized.type,
unspecialized.code, unspecialized.code_b, *locker,
usage.variant, true);
if (!program) {
// Otherwise compile it from GLSL
program = BuildShader(device, entry.type, uid, ir, *registry, true);
}
PrecompiledShader shader;
shader.program = std::move(program);
shader.registry = std::move(registry);
shader.entries = MakeEntries(ir);
std::scoped_lock lock{mutex};
if (callback) {
callback(VideoCore::LoadCallbackStage::Build, ++built_shaders,
shader_usages.size());
transferable->size());
}
precompiled_programs.emplace(usage, std::move(shader));
// TODO(Rodrigo): Is there a better way to do this?
precompiled_variants[usage.unique_identifier].push_back(
precompiled_programs.find(usage));
runtime_cache.emplace(entry.unique_identifier, std::move(shader));
}
};
const auto num_workers{static_cast<std::size_t>(std::thread::hardware_concurrency() + 1ULL)};
const std::size_t bucket_size{shader_usages.size() / num_workers};
const std::size_t bucket_size{transferable->size() / num_workers};
std::vector<std::unique_ptr<Core::Frontend::GraphicsContext>> contexts(num_workers);
std::vector<std::thread> threads(num_workers);
for (std::size_t i = 0; i < num_workers; ++i) {
const bool is_last_worker = i + 1 == num_workers;
const std::size_t start{bucket_size * i};
const std::size_t end{is_last_worker ? shader_usages.size() : start + bucket_size};
const std::size_t end{is_last_worker ? transferable->size() : start + bucket_size};
// On some platforms the shared context has to be created from the GUI thread
contexts[i] = emu_window.CreateSharedContext();
threads[i] = std::thread(Worker, contexts[i].get(), start, end, shader_usages, dumps);
threads[i] = std::thread(worker, contexts[i].get(), start, end);
}
for (auto& thread : threads) {
thread.join();
}
if (compilation_failed) {
if (gl_cache_failed) {
// Invalidate the precompiled cache if a shader dumped shader was rejected
disk_cache.InvalidatePrecompiled();
precompiled_cache_altered = true;
@ -533,11 +401,12 @@ void ShaderCacheOpenGL::LoadDiskCache(const std::atomic_bool& stop_loading,
// TODO(Rodrigo): Do state tracking for transferable shaders and do a dummy draw
// before precompiling them
for (std::size_t i = 0; i < shader_usages.size(); ++i) {
const auto& usage{shader_usages[i]};
if (dumps.find(usage) == dumps.end()) {
const auto& program{precompiled_programs.at(usage)};
disk_cache.SaveDump(usage, program->handle);
for (std::size_t i = 0; i < transferable->size(); ++i) {
const u64 id = (*transferable)[i].unique_identifier;
const auto it = find_precompiled(id);
if (it == gl_cache.end()) {
const GLuint program = runtime_cache.at(id).program->handle;
disk_cache.SavePrecompiled(id, program);
precompiled_cache_altered = true;
}
}
@ -547,80 +416,29 @@ void ShaderCacheOpenGL::LoadDiskCache(const std::atomic_bool& stop_loading,
}
}
const PrecompiledVariants* ShaderCacheOpenGL::GetPrecompiledVariants(u64 unique_identifier) const {
const auto it = precompiled_variants.find(unique_identifier);
return it == precompiled_variants.end() ? nullptr : &it->second;
}
CachedProgram ShaderCacheOpenGL::GeneratePrecompiledProgram(
const ShaderDiskCacheDump& dump, const std::unordered_set<GLenum>& supported_formats) {
if (supported_formats.find(dump.binary_format) == supported_formats.end()) {
LOG_INFO(Render_OpenGL, "Precompiled cache entry with unsupported format - removing");
std::shared_ptr<OGLProgram> ShaderCacheOpenGL::GeneratePrecompiledProgram(
const ShaderDiskCacheEntry& entry, const ShaderDiskCachePrecompiled& precompiled_entry,
const std::unordered_set<GLenum>& supported_formats) {
if (supported_formats.find(precompiled_entry.binary_format) == supported_formats.end()) {
LOG_INFO(Render_OpenGL, "Precompiled cache entry with unsupported format, removing");
return {};
}
CachedProgram shader = std::make_shared<OGLProgram>();
shader->handle = glCreateProgram();
glProgramParameteri(shader->handle, GL_PROGRAM_SEPARABLE, GL_TRUE);
glProgramBinary(shader->handle, dump.binary_format, dump.binary.data(),
static_cast<GLsizei>(dump.binary.size()));
auto program = std::make_shared<OGLProgram>();
program->handle = glCreateProgram();
glProgramParameteri(program->handle, GL_PROGRAM_SEPARABLE, GL_TRUE);
glProgramBinary(program->handle, precompiled_entry.binary_format,
precompiled_entry.binary.data(),
static_cast<GLsizei>(precompiled_entry.binary.size()));
GLint link_status{};
glGetProgramiv(shader->handle, GL_LINK_STATUS, &link_status);
GLint link_status;
glGetProgramiv(program->handle, GL_LINK_STATUS, &link_status);
if (link_status == GL_FALSE) {
LOG_INFO(Render_OpenGL, "Precompiled cache rejected by the driver - removing");
LOG_INFO(Render_OpenGL, "Precompiled cache rejected by the driver, removing");
return {};
}
return shader;
}
bool ShaderCacheOpenGL::GenerateUnspecializedShaders(
const std::atomic_bool& stop_loading, const VideoCore::DiskResourceLoadCallback& callback,
const std::vector<ShaderDiskCacheRaw>& raws) {
if (callback) {
callback(VideoCore::LoadCallbackStage::Decompile, 0, raws.size());
}
for (std::size_t i = 0; i < raws.size(); ++i) {
if (stop_loading) {
return false;
}
const auto& raw{raws[i]};
const u64 unique_identifier{raw.GetUniqueIdentifier()};
const u64 calculated_hash{
GetUniqueIdentifier(raw.GetType(), raw.HasProgramA(), raw.GetCode(), raw.GetCodeB())};
if (unique_identifier != calculated_hash) {
LOG_ERROR(Render_OpenGL,
"Invalid hash in entry={:016x} (obtained hash={:016x}) - "
"removing shader cache",
raw.GetUniqueIdentifier(), calculated_hash);
disk_cache.InvalidateTransferable();
return false;
}
const u32 main_offset =
raw.GetType() == ShaderType::Compute ? KERNEL_MAIN_OFFSET : STAGE_MAIN_OFFSET;
ConstBufferLocker locker(raw.GetType());
const ShaderIR ir(raw.GetCode(), main_offset, COMPILER_SETTINGS, locker);
// TODO(Rodrigo): Handle VertexA shaders
// std::optional<ShaderIR> ir_b;
// if (raw.HasProgramA()) {
// ir_b.emplace(raw.GetProgramCodeB(), main_offset);
// }
UnspecializedShader unspecialized;
unspecialized.entries = GLShader::GetEntries(ir);
unspecialized.type = raw.GetType();
unspecialized.code = raw.GetCode();
unspecialized.code_b = raw.GetCodeB();
unspecialized_shaders.emplace(raw.GetUniqueIdentifier(), unspecialized);
if (callback) {
callback(VideoCore::LoadCallbackStage::Decompile, i, raws.size());
}
}
return true;
return program;
}
Shader ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) {
@ -648,17 +466,17 @@ Shader ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) {
const auto unique_identifier = GetUniqueIdentifier(
GetShaderType(program), program == Maxwell::ShaderProgram::VertexA, code, code_b);
const auto precompiled_variants = GetPrecompiledVariants(unique_identifier);
const auto cpu_addr{*memory_manager.GpuToCpuAddress(address)};
const ShaderParameters params{system, disk_cache, precompiled_variants, device,
const ShaderParameters params{system, disk_cache, device,
cpu_addr, host_ptr, unique_identifier};
const auto found = unspecialized_shaders.find(unique_identifier);
if (found == unspecialized_shaders.end()) {
const auto found = runtime_cache.find(unique_identifier);
if (found == runtime_cache.end()) {
shader = CachedShader::CreateStageFromMemory(params, program, std::move(code),
std::move(code_b));
} else {
shader = CachedShader::CreateFromCache(params, found->second);
const std::size_t size_in_bytes = code.size() * sizeof(u64);
shader = CachedShader::CreateFromCache(params, found->second, size_in_bytes);
}
Register(shader);
@ -673,19 +491,19 @@ Shader ShaderCacheOpenGL::GetComputeKernel(GPUVAddr code_addr) {
return kernel;
}
// No kernel found - create a new one
// No kernel found, create a new one
auto code{GetShaderCode(memory_manager, code_addr, host_ptr)};
const auto unique_identifier{GetUniqueIdentifier(ShaderType::Compute, false, code, {})};
const auto precompiled_variants = GetPrecompiledVariants(unique_identifier);
const auto unique_identifier{GetUniqueIdentifier(ShaderType::Compute, false, code)};
const auto cpu_addr{*memory_manager.GpuToCpuAddress(code_addr)};
const ShaderParameters params{system, disk_cache, precompiled_variants, device,
const ShaderParameters params{system, disk_cache, device,
cpu_addr, host_ptr, unique_identifier};
const auto found = unspecialized_shaders.find(unique_identifier);
if (found == unspecialized_shaders.end()) {
const auto found = runtime_cache.find(unique_identifier);
if (found == runtime_cache.end()) {
kernel = CachedShader::CreateKernelFromMemory(params, std::move(code));
} else {
kernel = CachedShader::CreateFromCache(params, found->second);
const std::size_t size_in_bytes = code.size() * sizeof(u64);
kernel = CachedShader::CreateFromCache(params, found->second, size_in_bytes);
}
Register(kernel);

109
src/video_core/renderer_opengl/gl_shader_cache.h
View file

@ -22,7 +22,7 @@
#include "video_core/renderer_opengl/gl_resource_manager.h"
#include "video_core/renderer_opengl/gl_shader_decompiler.h"
#include "video_core/renderer_opengl/gl_shader_disk_cache.h"
#include "video_core/shader/const_buffer_locker.h"
#include "video_core/shader/registry.h"
#include "video_core/shader/shader_ir.h"
namespace Core {
@ -41,22 +41,17 @@ class RasterizerOpenGL;
struct UnspecializedShader;
using Shader = std::shared_ptr<CachedShader>;
using CachedProgram = std::shared_ptr<OGLProgram>;
using Maxwell = Tegra::Engines::Maxwell3D::Regs;
using PrecompiledPrograms = std::unordered_map<ShaderDiskCacheUsage, CachedProgram>;
using PrecompiledVariants = std::vector<PrecompiledPrograms::iterator>;
struct UnspecializedShader {
GLShader::ShaderEntries entries;
Tegra::Engines::ShaderType type;
ProgramCode code;
ProgramCode code_b;
struct PrecompiledShader {
std::shared_ptr<OGLProgram> program;
std::shared_ptr<VideoCommon::Shader::Registry> registry;
ShaderEntries entries;
};
struct ShaderParameters {
Core::System& system;
ShaderDiskCacheOpenGL& disk_cache;
const PrecompiledVariants* precompiled_variants;
const Device& device;
VAddr cpu_addr;
u8* host_ptr;
@ -65,61 +60,45 @@ struct ShaderParameters {
class CachedShader final : public RasterizerCacheObject {
public:
~CachedShader();
/// Gets the GL program handle for the shader
GLuint GetHandle() const;
/// Returns the guest CPU address of the shader
VAddr GetCpuAddr() const override {
return cpu_addr;
}
/// Returns the size in bytes of the shader
std::size_t GetSizeInBytes() const override {
return size_in_bytes;
}
/// Gets the shader entries for the shader
const ShaderEntries& GetEntries() const {
return entries;
}
static Shader CreateStageFromMemory(const ShaderParameters& params,
Maxwell::ShaderProgram program_type,
ProgramCode program_code, ProgramCode program_code_b);
static Shader CreateKernelFromMemory(const ShaderParameters& params, ProgramCode code);
static Shader CreateFromCache(const ShaderParameters& params,
const UnspecializedShader& unspecialized);
VAddr GetCpuAddr() const override {
return cpu_addr;
}
std::size_t GetSizeInBytes() const override {
return code.size() * sizeof(u64);
}
/// Gets the shader entries for the shader
const GLShader::ShaderEntries& GetShaderEntries() const {
return entries;
}
/// Gets the GL program handle for the shader
GLuint GetHandle(const ProgramVariant& variant);
const PrecompiledShader& precompiled_shader,
std::size_t size_in_bytes);
private:
struct LockerVariant {
std::unique_ptr<VideoCommon::Shader::ConstBufferLocker> locker;
std::unordered_map<ProgramVariant, CachedProgram> programs;
};
explicit CachedShader(const u8* host_ptr, VAddr cpu_addr, std::size_t size_in_bytes,
std::shared_ptr<VideoCommon::Shader::Registry> registry,
ShaderEntries entries, std::shared_ptr<OGLProgram> program);
explicit CachedShader(const ShaderParameters& params, Tegra::Engines::ShaderType shader_type,
GLShader::ShaderEntries entries, ProgramCode program_code,
ProgramCode program_code_b);
bool EnsureValidLockerVariant();
ShaderDiskCacheUsage GetUsage(const ProgramVariant& variant,
const VideoCommon::Shader::ConstBufferLocker& locker) const;
Core::System& system;
ShaderDiskCacheOpenGL& disk_cache;
const Device& device;
VAddr cpu_addr{};
u64 unique_identifier{};
Tegra::Engines::ShaderType shader_type{};
GLShader::ShaderEntries entries;
ProgramCode code;
ProgramCode code_b;
LockerVariant* curr_locker_variant = nullptr;
std::vector<std::unique_ptr<LockerVariant>> locker_variants;
std::shared_ptr<VideoCommon::Shader::Registry> registry;
ShaderEntries entries;
VAddr cpu_addr = 0;
std::size_t size_in_bytes = 0;
std::shared_ptr<OGLProgram> program;
};
class ShaderCacheOpenGL final : public RasterizerCache<Shader> {
@ -142,25 +121,15 @@ protected:
void FlushObjectInner(const Shader& object) override {}
private:
bool GenerateUnspecializedShaders(const std::atomic_bool& stop_loading,
const VideoCore::DiskResourceLoadCallback& callback,
const std::vector<ShaderDiskCacheRaw>& raws);
CachedProgram GeneratePrecompiledProgram(const ShaderDiskCacheDump& dump,
const std::unordered_set<GLenum>& supported_formats);
const PrecompiledVariants* GetPrecompiledVariants(u64 unique_identifier) const;
std::shared_ptr<OGLProgram> GeneratePrecompiledProgram(
const ShaderDiskCacheEntry& entry, const ShaderDiskCachePrecompiled& precompiled_entry,
const std::unordered_set<GLenum>& supported_formats);
Core::System& system;
Core::Frontend::EmuWindow& emu_window;
const Device& device;
ShaderDiskCacheOpenGL disk_cache;
PrecompiledPrograms precompiled_programs;
std::unordered_map<u64, PrecompiledVariants> precompiled_variants;
std::unordered_map<u64, UnspecializedShader> unspecialized_shaders;
std::unordered_map<u64, PrecompiledShader> runtime_cache;
std::array<Shader, Maxwell::MaxShaderProgram> last_shaders;
};

278
src/video_core/renderer_opengl/gl_shader_decompiler.cpp
View file

@ -24,7 +24,7 @@
#include "video_core/shader/node.h"
#include "video_core/shader/shader_ir.h"
namespace OpenGL::GLShader {
namespace OpenGL {
namespace {
@ -36,6 +36,7 @@ using Tegra::Shader::IpaInterpMode;
using Tegra::Shader::IpaMode;
using Tegra::Shader::IpaSampleMode;
using Tegra::Shader::Register;
using VideoCommon::Shader::Registry;
using namespace std::string_literals;
using namespace VideoCommon::Shader;
@ -56,6 +57,25 @@ using TextureIR = std::variant<TextureOffset, TextureDerivates, TextureArgument>
constexpr u32 MAX_CONSTBUFFER_ELEMENTS =
static_cast<u32>(Maxwell::MaxConstBufferSize) / (4 * sizeof(float));
constexpr std::string_view CommonDeclarations = R"(#define ftoi floatBitsToInt
#define ftou floatBitsToUint
#define itof intBitsToFloat
#define utof uintBitsToFloat
bvec2 HalfFloatNanComparison(bvec2 comparison, vec2 pair1, vec2 pair2) {{
bvec2 is_nan1 = isnan(pair1);
bvec2 is_nan2 = isnan(pair2);
return bvec2(comparison.x || is_nan1.x || is_nan2.x, comparison.y || is_nan1.y || is_nan2.y);
}}
const float fswzadd_modifiers_a[] = float[4](-1.0f, 1.0f, -1.0f, 0.0f );
const float fswzadd_modifiers_b[] = float[4](-1.0f, -1.0f, 1.0f, -1.0f );
layout (std140, binding = {}) uniform vs_config {{
float y_direction;
}};
)";
class ShaderWriter final {
public:
void AddExpression(std::string_view text) {
@ -269,12 +289,41 @@ const char* GetImageTypeDeclaration(Tegra::Shader::ImageType image_type) {
}
}
/// Describes primitive behavior on geometry shaders
std::pair<const char*, u32> GetPrimitiveDescription(Maxwell::PrimitiveTopology topology) {
switch (topology) {
case Maxwell::PrimitiveTopology::Points:
return {"points", 1};
case Maxwell::PrimitiveTopology::Lines:
case Maxwell::PrimitiveTopology::LineStrip:
return {"lines", 2};
case Maxwell::PrimitiveTopology::LinesAdjacency:
case Maxwell::PrimitiveTopology::LineStripAdjacency:
return {"lines_adjacency", 4};
case Maxwell::PrimitiveTopology::Triangles:
case Maxwell::PrimitiveTopology::TriangleStrip:
case Maxwell::PrimitiveTopology::TriangleFan:
return {"triangles", 3};
case Maxwell::PrimitiveTopology::TrianglesAdjacency:
case Maxwell::PrimitiveTopology::TriangleStripAdjacency:
return {"triangles_adjacency", 6};
default:
UNIMPLEMENTED_MSG("topology={}", static_cast<int>(topology));
return {"points", 1};
}
}
/// Generates code to use for a swizzle operation.
constexpr const char* GetSwizzle(u32 element) {
constexpr const char* GetSwizzle(std::size_t element) {
constexpr std::array swizzle = {".x", ".y", ".z", ".w"};
return swizzle.at(element);
}
constexpr const char* GetColorSwizzle(std::size_t element) {
constexpr std::array swizzle = {".r", ".g", ".b", ".a"};
return swizzle.at(element);
}
/// Translate topology
std::string GetTopologyName(Tegra::Shader::OutputTopology topology) {
switch (topology) {
@ -343,9 +392,54 @@ std::string FlowStackTopName(MetaStackClass stack) {
class GLSLDecompiler final {
public:
explicit GLSLDecompiler(const Device& device, const ShaderIR& ir, ShaderType stage,
std::string suffix)
: device{device}, ir{ir}, stage{stage}, suffix{suffix}, header{ir.GetHeader()} {}
explicit GLSLDecompiler(const Device& device, const ShaderIR& ir, const Registry& registry,
ShaderType stage, std::string_view identifier, std::string_view suffix)
: device{device}, ir{ir}, registry{registry}, stage{stage},
identifier{identifier}, suffix{suffix}, header{ir.GetHeader()} {}
void Decompile() {
DeclareHeader();
DeclareVertex();
DeclareGeometry();
DeclareFragment();
DeclareCompute();
DeclareRegisters();
DeclareCustomVariables();
DeclarePredicates();
DeclareLocalMemory();
DeclareInternalFlags();
DeclareInputAttributes();
DeclareOutputAttributes();
DeclareConstantBuffers();
DeclareGlobalMemory();
DeclareSamplers();
DeclareImages();
DeclarePhysicalAttributeReader();
code.AddLine("void main() {{");
++code.scope;
if (stage == ShaderType::Vertex) {
code.AddLine("gl_Position = vec4(0.0f, 0.0f, 0.0f, 1.0f);");
}
if (ir.IsDecompiled()) {
DecompileAST();
} else {
DecompileBranchMode();
}
--code.scope;
code.AddLine("}}");
}
std::string GetResult() {
return code.GetResult();
}
private:
friend class ASTDecompiler;
friend class ExprDecompiler;
void DecompileBranchMode() {
// VM's program counter
@ -387,43 +481,36 @@ public:
void DecompileAST();
void Decompile() {
DeclareVertex();
DeclareGeometry();
DeclareRegisters();
DeclareCustomVariables();
DeclarePredicates();
DeclareLocalMemory();
DeclareInternalFlags();
DeclareInputAttributes();
DeclareOutputAttributes();
DeclareConstantBuffers();
DeclareGlobalMemory();
DeclareSamplers();
DeclareImages();
DeclarePhysicalAttributeReader();
code.AddLine("void execute_{}() {{", suffix);
++code.scope;
if (ir.IsDecompiled()) {
DecompileAST();
} else {
DecompileBranchMode();
void DeclareHeader() {
if (!identifier.empty()) {
code.AddLine("// {}", identifier);
}
code.AddLine("#version 430 core");
code.AddLine("#extension GL_ARB_separate_shader_objects : enable");
if (device.HasShaderBallot()) {
code.AddLine("#extension GL_ARB_shader_ballot : require");
}
if (device.HasVertexViewportLayer()) {
code.AddLine("#extension GL_ARB_shader_viewport_layer_array : require");
}
if (device.HasImageLoadFormatted()) {
code.AddLine("#extension GL_EXT_shader_image_load_formatted : require");
}
if (device.HasWarpIntrinsics()) {
code.AddLine("#extension GL_NV_gpu_shader5 : require");
code.AddLine("#extension GL_NV_shader_thread_group : require");
code.AddLine("#extension GL_NV_shader_thread_shuffle : require");
}
// This pragma stops Nvidia's driver from over optimizing math (probably using fp16
// operations) on places where we don't want to.
// Thanks to Ryujinx for finding this workaround.
code.AddLine("#pragma optionNV(fastmath off)");
--code.scope;
code.AddLine("}}");
code.AddNewLine();
code.AddLine(CommonDeclarations, EmulationUniformBlockBinding);
}
std::string GetResult() {
return code.GetResult();
}
private:
friend class ASTDecompiler;
friend class ExprDecompiler;
void DeclareVertex() {
if (!IsVertexShader(stage))
return;
@ -436,9 +523,15 @@ private:
return;
}
const auto& info = registry.GetGraphicsInfo();
const auto input_topology = info.primitive_topology;
const auto [glsl_topology, max_vertices] = GetPrimitiveDescription(input_topology);
max_input_vertices = max_vertices;
code.AddLine("layout ({}) in;", glsl_topology);
const auto topology = GetTopologyName(header.common3.output_topology);
const auto max_vertices = header.common4.max_output_vertices.Value();
code.AddLine("layout ({}, max_vertices = {}) out;", topology, max_vertices);
const auto max_output_vertices = header.common4.max_output_vertices.Value();
code.AddLine("layout ({}, max_vertices = {}) out;", topology, max_output_vertices);
code.AddNewLine();
code.AddLine("in gl_PerVertex {{");
@ -450,6 +543,29 @@ private:
DeclareVertexRedeclarations();
}
void DeclareFragment() {
if (stage != ShaderType::Fragment) {
return;
}
for (u32 rt = 0; rt < Maxwell::NumRenderTargets; ++rt) {
code.AddLine("layout (location = {}) out vec4 frag_color{};", rt, rt);
}
}
void DeclareCompute() {
if (stage != ShaderType::Compute) {
return;
}
const auto& info = registry.GetComputeInfo();
if (const u32 size = info.shared_memory_size_in_words; size > 0) {
code.AddLine("shared uint smem[{}];", size);
code.AddNewLine();
}
code.AddLine("layout (local_size_x = {}, local_size_y = {}, local_size_z = {}) in;",
info.workgroup_size[0], info.workgroup_size[1], info.workgroup_size[2]);
code.AddNewLine();
}
void DeclareVertexRedeclarations() {
code.AddLine("out gl_PerVertex {{");
++code.scope;
@ -525,18 +641,16 @@ private:
}
void DeclareLocalMemory() {
u64 local_memory_size = 0;
if (stage == ShaderType::Compute) {
code.AddLine("#ifdef LOCAL_MEMORY_SIZE");
code.AddLine("uint {}[LOCAL_MEMORY_SIZE];", GetLocalMemory());
code.AddLine("#endif");
return;
local_memory_size = registry.GetComputeInfo().local_memory_size_in_words * 4ULL;
} else {
local_memory_size = header.GetLocalMemorySize();
}
const u64 local_memory_size = header.GetLocalMemorySize();
if (local_memory_size == 0) {
return;
}
const auto element_count = Common::AlignUp(local_memory_size, 4) / 4;
const u64 element_count = Common::AlignUp(local_memory_size, 4) / 4;
code.AddLine("uint {}[{}];", GetLocalMemory(), element_count);
code.AddNewLine();
}
@ -925,7 +1039,8 @@ private:
// TODO(Rodrigo): Guard geometry inputs against out of bound reads. Some games
// set an 0x80000000 index for those and the shader fails to build. Find out why
// this happens and what's its intent.
return fmt::format("gs_{}[{} % MAX_VERTEX_INPUT]", name, Visit(buffer).AsUint());
return fmt::format("gs_{}[{} % {}]", name, Visit(buffer).AsUint(),
max_input_vertices.value());
}
return std::string(name);
};
@ -1945,7 +2060,7 @@ private:
// TODO(Subv): Figure out how dual-source blending is configured in the Switch.
for (u32 component = 0; component < 4; ++component) {
if (header.ps.IsColorComponentOutputEnabled(render_target, component)) {
code.AddLine("FragColor{}[{}] = {};", render_target, component,
code.AddLine("frag_color{}{} = {};", render_target, GetColorSwizzle(component),
SafeGetRegister(current_reg).AsFloat());
++current_reg;
}
@ -2298,7 +2413,11 @@ private:
}
std::string GetLocalMemory() const {
return "lmem_" + suffix;
if (suffix.empty()) {
return "lmem";
} else {
return "lmem_" + std::string{suffix};
}
}
std::string GetInternalFlag(InternalFlag flag) const {
@ -2307,7 +2426,11 @@ private:
const auto index = static_cast<u32>(flag);
ASSERT(index < static_cast<u32>(InternalFlag::Amount));
return fmt::format("{}_{}", InternalFlagNames[index], suffix);
if (suffix.empty()) {
return InternalFlagNames[index];
} else {
return fmt::format("{}_{}", InternalFlagNames[index], suffix);
}
}
std::string GetSampler(const Sampler& sampler) const {
@ -2319,7 +2442,11 @@ private:
}
std::string GetDeclarationWithSuffix(u32 index, std::string_view name) const {
return fmt::format("{}_{}_{}", name, index, suffix);
if (suffix.empty()) {
return fmt::format("{}{}", name, index);
} else {
return fmt::format("{}{}_{}", name, index, suffix);
}
}
u32 GetNumPhysicalInputAttributes() const {
@ -2334,17 +2461,30 @@ private:
return std::min<u32>(device.GetMaxVaryings(), Maxwell::NumVaryings);
}
bool IsRenderTargetEnabled(u32 render_target) const {
for (u32 component = 0; component < 4; ++component) {
if (header.ps.IsColorComponentOutputEnabled(render_target, component)) {
return true;
}
}
return false;
}
const Device& device;
const ShaderIR& ir;
const Registry& registry;
const ShaderType stage;
const std::string suffix;
const std::string_view identifier;
const std::string_view suffix;
const Header header;
ShaderWriter code;
std::optional<u32> max_input_vertices;
};
std::string GetFlowVariable(u32 i) {
return fmt::format("flow_var_{}", i);
std::string GetFlowVariable(u32 index) {
return fmt::format("flow_var{}", index);
}
class ExprDecompiler {
@ -2531,7 +2671,7 @@ void GLSLDecompiler::DecompileAST() {
} // Anonymous namespace
ShaderEntries GetEntries(const VideoCommon::Shader::ShaderIR& ir) {
ShaderEntries MakeEntries(const VideoCommon::Shader::ShaderIR& ir) {
ShaderEntries entries;
for (const auto& cbuf : ir.GetConstantBuffers()) {
entries.const_buffers.emplace_back(cbuf.second.GetMaxOffset(), cbuf.second.IsIndirect(),
@ -2555,28 +2695,12 @@ ShaderEntries GetEntries(const VideoCommon::Shader::ShaderIR& ir) {
return entries;
}
std::string GetCommonDeclarations() {
return R"(#define ftoi floatBitsToInt
#define ftou floatBitsToUint
#define itof intBitsToFloat
#define utof uintBitsToFloat
bvec2 HalfFloatNanComparison(bvec2 comparison, vec2 pair1, vec2 pair2) {
bvec2 is_nan1 = isnan(pair1);
bvec2 is_nan2 = isnan(pair2);
return bvec2(comparison.x || is_nan1.x || is_nan2.x, comparison.y || is_nan1.y || is_nan2.y);
}
const float fswzadd_modifiers_a[] = float[4](-1.0f, 1.0f, -1.0f, 0.0f );
const float fswzadd_modifiers_b[] = float[4](-1.0f, -1.0f, 1.0f, -1.0f );
)";
}
std::string Decompile(const Device& device, const ShaderIR& ir, ShaderType stage,
const std::string& suffix) {
GLSLDecompiler decompiler(device, ir, stage, suffix);
std::string DecompileShader(const Device& device, const ShaderIR& ir, const Registry& registry,
ShaderType stage, std::string_view identifier,
std::string_view suffix) {
GLSLDecompiler decompiler(device, ir, registry, stage, identifier, suffix);
decompiler.Decompile();
return decompiler.GetResult();
}
} // namespace OpenGL::GLShader
} // namespace OpenGL

22
src/video_core/renderer_opengl/gl_shader_decompiler.h
View file

@ -6,22 +6,18 @@
#include <array>
#include <string>
#include <string_view>
#include <utility>
#include <vector>
#include "common/common_types.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/engines/shader_type.h"
#include "video_core/shader/registry.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
class ShaderIR;
}
namespace OpenGL {
class Device;
}
namespace OpenGL::GLShader {
class Device;
using Maxwell = Tegra::Engines::Maxwell3D::Regs;
using SamplerEntry = VideoCommon::Shader::Sampler;
@ -78,11 +74,11 @@ struct ShaderEntries {
std::size_t shader_length{};
};
ShaderEntries GetEntries(const VideoCommon::Shader::ShaderIR& ir);
ShaderEntries MakeEntries(const VideoCommon::Shader::ShaderIR& ir);
std::string GetCommonDeclarations();
std::string DecompileShader(const Device& device, const VideoCommon::Shader::ShaderIR& ir,
const VideoCommon::Shader::Registry& registry,
Tegra::Engines::ShaderType stage, std::string_view identifier,
std::string_view suffix = {});
std::string Decompile(const Device& device, const VideoCommon::Shader::ShaderIR& ir,
Tegra::Engines::ShaderType stage, const std::string& suffix);
} // namespace OpenGL::GLShader
} // namespace OpenGL

408
src/video_core/renderer_opengl/gl_shader_disk_cache.cpp
View file

@ -31,32 +31,24 @@ namespace {
using ShaderCacheVersionHash = std::array<u8, 64>;
enum class TransferableEntryKind : u32 {
Raw,
Usage,
};
struct ConstBufferKey {
u32 cbuf{};
u32 offset{};
u32 value{};
u32 cbuf = 0;
u32 offset = 0;
u32 value = 0;
};
struct BoundSamplerKey {
u32 offset{};
Tegra::Engines::SamplerDescriptor sampler{};
u32 offset = 0;
Tegra::Engines::SamplerDescriptor sampler;
};
struct BindlessSamplerKey {
u32 cbuf{};
u32 offset{};
Tegra::Engines::SamplerDescriptor sampler{};
u32 cbuf = 0;
u32 offset = 0;
Tegra::Engines::SamplerDescriptor sampler;
};
constexpr u32 NativeVersion = 12;
// Making sure sizes doesn't change by accident
static_assert(sizeof(ProgramVariant) == 20);
constexpr u32 NativeVersion = 20;
ShaderCacheVersionHash GetShaderCacheVersionHash() {
ShaderCacheVersionHash hash{};
@ -67,61 +59,124 @@ ShaderCacheVersionHash GetShaderCacheVersionHash() {
} // Anonymous namespace
ShaderDiskCacheRaw::ShaderDiskCacheRaw(u64 unique_identifier, ShaderType type, ProgramCode code,
ProgramCode code_b)
: unique_identifier{unique_identifier}, type{type}, code{std::move(code)}, code_b{std::move(
code_b)} {}
ShaderDiskCacheEntry::ShaderDiskCacheEntry() = default;
ShaderDiskCacheRaw::ShaderDiskCacheRaw() = default;
ShaderDiskCacheEntry::~ShaderDiskCacheEntry() = default;
ShaderDiskCacheRaw::~ShaderDiskCacheRaw() = default;
bool ShaderDiskCacheRaw::Load(FileUtil::IOFile& file) {
if (file.ReadBytes(&unique_identifier, sizeof(u64)) != sizeof(u64) ||
file.ReadBytes(&type, sizeof(u32)) != sizeof(u32)) {
bool ShaderDiskCacheEntry::Load(FileUtil::IOFile& file) {
if (file.ReadBytes(&type, sizeof(u32)) != sizeof(u32)) {
return false;
}
u32 code_size{};
u32 code_size_b{};
u32 code_size;
u32 code_size_b;
if (file.ReadBytes(&code_size, sizeof(u32)) != sizeof(u32) ||
file.ReadBytes(&code_size_b, sizeof(u32)) != sizeof(u32)) {
return false;
}
code.resize(code_size);
code_b.resize(code_size_b);
if (file.ReadArray(code.data(), code_size) != code_size)
if (file.ReadArray(code.data(), code_size) != code_size) {
return false;
}
if (HasProgramA() && file.ReadArray(code_b.data(), code_size_b) != code_size_b) {
return false;
}
u8 is_texture_handler_size_known;
u32 texture_handler_size_value;
u32 num_keys;
u32 num_bound_samplers;
u32 num_bindless_samplers;
if (file.ReadArray(&unique_identifier, 1) != 1 || file.ReadArray(&bound_buffer, 1) != 1 ||
file.ReadArray(&is_texture_handler_size_known, 1) != 1 ||
file.ReadArray(&texture_handler_size_value, 1) != 1 ||
file.ReadArray(&graphics_info, 1) != 1 || file.ReadArray(&compute_info, 1) != 1 ||
file.ReadArray(&num_keys, 1) != 1 || file.ReadArray(&num_bound_samplers, 1) != 1 ||
file.ReadArray(&num_bindless_samplers, 1) != 1) {
return false;
}
if (is_texture_handler_size_known) {
texture_handler_size = texture_handler_size_value;
}
std::vector<ConstBufferKey> flat_keys(num_keys);
std::vector<BoundSamplerKey> flat_bound_samplers(num_bound_samplers);
std::vector<BindlessSamplerKey> flat_bindless_samplers(num_bindless_samplers);
if (file.ReadArray(flat_keys.data(), flat_keys.size()) != flat_keys.size() ||
file.ReadArray(flat_bound_samplers.data(), flat_bound_samplers.size()) !=
flat_bound_samplers.size() ||
file.ReadArray(flat_bindless_samplers.data(), flat_bindless_samplers.size()) !=
flat_bindless_samplers.size()) {
return false;
}
for (const auto& key : flat_keys) {
keys.insert({{key.cbuf, key.offset}, key.value});
}
for (const auto& key : flat_bound_samplers) {
bound_samplers.emplace(key.offset, key.sampler);
}
for (const auto& key : flat_bindless_samplers) {
bindless_samplers.insert({{key.cbuf, key.offset}, key.sampler});
}
return true;
}
bool ShaderDiskCacheRaw::Save(FileUtil::IOFile& file) const {
if (file.WriteObject(unique_identifier) != 1 || file.WriteObject(static_cast<u32>(type)) != 1 ||
bool ShaderDiskCacheEntry::Save(FileUtil::IOFile& file) const {
if (file.WriteObject(static_cast<u32>(type)) != 1 ||
file.WriteObject(static_cast<u32>(code.size())) != 1 ||
file.WriteObject(static_cast<u32>(code_b.size())) != 1) {
return false;
}
if (file.WriteArray(code.data(), code.size()) != code.size())
if (file.WriteArray(code.data(), code.size()) != code.size()) {
return false;
}
if (HasProgramA() && file.WriteArray(code_b.data(), code_b.size()) != code_b.size()) {
return false;
}
return true;
if (file.WriteObject(unique_identifier) != 1 || file.WriteObject(bound_buffer) != 1 ||
file.WriteObject(static_cast<u8>(texture_handler_size.has_value())) != 1 ||
file.WriteObject(texture_handler_size.value_or(0)) != 1 ||
file.WriteObject(graphics_info) != 1 || file.WriteObject(compute_info) != 1 ||
file.WriteObject(static_cast<u32>(keys.size())) != 1 ||
file.WriteObject(static_cast<u32>(bound_samplers.size())) != 1 ||
file.WriteObject(static_cast<u32>(bindless_samplers.size())) != 1) {
return false;
}
std::vector<ConstBufferKey> flat_keys;
flat_keys.reserve(keys.size());
for (const auto& [address, value] : keys) {
flat_keys.push_back(ConstBufferKey{address.first, address.second, value});
}
std::vector<BoundSamplerKey> flat_bound_samplers;
flat_bound_samplers.reserve(bound_samplers.size());
for (const auto& [address, sampler] : bound_samplers) {
flat_bound_samplers.push_back(BoundSamplerKey{address, sampler});
}
std::vector<BindlessSamplerKey> flat_bindless_samplers;
flat_bindless_samplers.reserve(bindless_samplers.size());
for (const auto& [address, sampler] : bindless_samplers) {
flat_bindless_samplers.push_back(
BindlessSamplerKey{address.first, address.second, sampler});
}
return file.WriteArray(flat_keys.data(), flat_keys.size()) == flat_keys.size() &&
file.WriteArray(flat_bound_samplers.data(), flat_bound_samplers.size()) ==
flat_bound_samplers.size() &&
file.WriteArray(flat_bindless_samplers.data(), flat_bindless_samplers.size()) ==
flat_bindless_samplers.size();
}
ShaderDiskCacheOpenGL::ShaderDiskCacheOpenGL(Core::System& system) : system{system} {}
ShaderDiskCacheOpenGL::~ShaderDiskCacheOpenGL() = default;
std::optional<std::pair<std::vector<ShaderDiskCacheRaw>, std::vector<ShaderDiskCacheUsage>>>
ShaderDiskCacheOpenGL::LoadTransferable() {
std::optional<std::vector<ShaderDiskCacheEntry>> ShaderDiskCacheOpenGL::LoadTransferable() {
// Skip games without title id
const bool has_title_id = system.CurrentProcess()->GetTitleID() != 0;
if (!Settings::values.use_disk_shader_cache || !has_title_id) {
@ -130,17 +185,14 @@ ShaderDiskCacheOpenGL::LoadTransferable() {
FileUtil::IOFile file(GetTransferablePath(), "rb");
if (!file.IsOpen()) {
LOG_INFO(Render_OpenGL, "No transferable shader cache found for game with title id={}",
GetTitleID());
LOG_INFO(Render_OpenGL, "No transferable shader cache found");
is_usable = true;
return {};
}
u32 version{};
if (file.ReadBytes(&version, sizeof(version)) != sizeof(version)) {
LOG_ERROR(Render_OpenGL,
"Failed to get transferable cache version for title id={}, skipping",
GetTitleID());
LOG_ERROR(Render_OpenGL, "Failed to get transferable cache version, skipping it");
return {};
}
@ -158,105 +210,42 @@ ShaderDiskCacheOpenGL::LoadTransferable() {
}
// Version is valid, load the shaders
constexpr const char error_loading[] = "Failed to load transferable raw entry, skipping";
std::vector<ShaderDiskCacheRaw> raws;
std::vector<ShaderDiskCacheUsage> usages;
std::vector<ShaderDiskCacheEntry> entries;
while (file.Tell() < file.GetSize()) {
TransferableEntryKind kind{};
if (file.ReadBytes(&kind, sizeof(u32)) != sizeof(u32)) {
LOG_ERROR(Render_OpenGL, "Failed to read transferable file, skipping");
return {};
}
switch (kind) {
case TransferableEntryKind::Raw: {
ShaderDiskCacheRaw entry;
if (!entry.Load(file)) {
LOG_ERROR(Render_OpenGL, error_loading);
return {};
}
transferable.insert({entry.GetUniqueIdentifier(), {}});
raws.push_back(std::move(entry));
break;
}
case TransferableEntryKind::Usage: {
ShaderDiskCacheUsage usage;
u32 num_keys{};
u32 num_bound_samplers{};
u32 num_bindless_samplers{};
if (file.ReadArray(&usage.unique_identifier, 1) != 1 ||
file.ReadArray(&usage.variant, 1) != 1 ||
file.ReadArray(&usage.bound_buffer, 1) != 1 || file.ReadArray(&num_keys, 1) != 1 ||
file.ReadArray(&num_bound_samplers, 1) != 1 ||
file.ReadArray(&num_bindless_samplers, 1) != 1) {
LOG_ERROR(Render_OpenGL, error_loading);
return {};
}
std::vector<ConstBufferKey> keys(num_keys);
std::vector<BoundSamplerKey> bound_samplers(num_bound_samplers);
std::vector<BindlessSamplerKey> bindless_samplers(num_bindless_samplers);
if (file.ReadArray(keys.data(), keys.size()) != keys.size() ||
file.ReadArray(bound_samplers.data(), bound_samplers.size()) !=
bound_samplers.size() ||
file.ReadArray(bindless_samplers.data(), bindless_samplers.size()) !=
bindless_samplers.size()) {
LOG_ERROR(Render_OpenGL, error_loading);
return {};
}
for (const auto& key : keys) {
usage.keys.insert({{key.cbuf, key.offset}, key.value});
}
for (const auto& key : bound_samplers) {
usage.bound_samplers.emplace(key.offset, key.sampler);
}
for (const auto& key : bindless_samplers) {
usage.bindless_samplers.insert({{key.cbuf, key.offset}, key.sampler});
}
usages.push_back(std::move(usage));
break;
}
default:
LOG_ERROR(Render_OpenGL, "Unknown transferable shader cache entry kind={}, skipping",
static_cast<u32>(kind));
ShaderDiskCacheEntry& entry = entries.emplace_back();
if (!entry.Load(file)) {
LOG_ERROR(Render_OpenGL, "Failed to load transferable raw entry, skipping");
return {};
}
}
is_usable = true;
return {{std::move(raws), std::move(usages)}};
return {std::move(entries)};
}
std::unordered_map<ShaderDiskCacheUsage, ShaderDiskCacheDump>
ShaderDiskCacheOpenGL::LoadPrecompiled() {
std::vector<ShaderDiskCachePrecompiled> ShaderDiskCacheOpenGL::LoadPrecompiled() {
if (!is_usable) {
return {};
}
std::string path = GetPrecompiledPath();
FileUtil::IOFile file(path, "rb");
FileUtil::IOFile file(GetPrecompiledPath(), "rb");
if (!file.IsOpen()) {
LOG_INFO(Render_OpenGL, "No precompiled shader cache found for game with title id={}",
GetTitleID());
LOG_INFO(Render_OpenGL, "No precompiled shader cache found");
return {};
}
const auto result = LoadPrecompiledFile(file);
if (!result) {
LOG_INFO(Render_OpenGL,
"Failed to load precompiled cache for game with title id={}, removing",
GetTitleID());
file.Close();
InvalidatePrecompiled();
return {};
if (const auto result = LoadPrecompiledFile(file)) {
return *result;
}
return *result;
LOG_INFO(Render_OpenGL, "Failed to load precompiled cache");
file.Close();
InvalidatePrecompiled();
return {};
}
std::optional<std::unordered_map<ShaderDiskCacheUsage, ShaderDiskCacheDump>>
ShaderDiskCacheOpenGL::LoadPrecompiledFile(FileUtil::IOFile& file) {
std::optional<std::vector<ShaderDiskCachePrecompiled>> ShaderDiskCacheOpenGL::LoadPrecompiledFile(
FileUtil::IOFile& file) {
// Read compressed file from disk and decompress to virtual precompiled cache file
std::vector<u8> compressed(file.GetSize());
file.ReadBytes(compressed.data(), compressed.size());
@ -275,58 +264,22 @@ ShaderDiskCacheOpenGL::LoadPrecompiledFile(FileUtil::IOFile& file) {
return {};
}
ShaderDumpsMap dumps;
std::vector<ShaderDiskCachePrecompiled> entries;
while (precompiled_cache_virtual_file_offset < precompiled_cache_virtual_file.GetSize()) {
u32 num_keys{};
u32 num_bound_samplers{};
u32 num_bindless_samplers{};
ShaderDiskCacheUsage usage;
if (!LoadObjectFromPrecompiled(usage.unique_identifier) ||
!LoadObjectFromPrecompiled(usage.variant) ||
!LoadObjectFromPrecompiled(usage.bound_buffer) ||
!LoadObjectFromPrecompiled(num_keys) ||
!LoadObjectFromPrecompiled(num_bound_samplers) ||
!LoadObjectFromPrecompiled(num_bindless_samplers)) {
return {};
}
std::vector<ConstBufferKey> keys(num_keys);
std::vector<BoundSamplerKey> bound_samplers(num_bound_samplers);
std::vector<BindlessSamplerKey> bindless_samplers(num_bindless_samplers);
if (!LoadArrayFromPrecompiled(keys.data(), keys.size()) ||
!LoadArrayFromPrecompiled(bound_samplers.data(), bound_samplers.size()) !=
bound_samplers.size() ||
!LoadArrayFromPrecompiled(bindless_samplers.data(), bindless_samplers.size()) !=
bindless_samplers.size()) {
return {};
}
for (const auto& key : keys) {
usage.keys.insert({{key.cbuf, key.offset}, key.value});
}
for (const auto& key : bound_samplers) {
usage.bound_samplers.emplace(key.offset, key.sampler);
}
for (const auto& key : bindless_samplers) {
usage.bindless_samplers.insert({{key.cbuf, key.offset}, key.sampler});
}
ShaderDiskCacheDump dump;
if (!LoadObjectFromPrecompiled(dump.binary_format)) {
u32 binary_size;
auto& entry = entries.emplace_back();
if (!LoadObjectFromPrecompiled(entry.unique_identifier) ||
!LoadObjectFromPrecompiled(entry.binary_format) ||
!LoadObjectFromPrecompiled(binary_size)) {
return {};
}
u32 binary_length{};
if (!LoadObjectFromPrecompiled(binary_length)) {
entry.binary.resize(binary_size);
if (!LoadArrayFromPrecompiled(entry.binary.data(), entry.binary.size())) {
return {};
}
dump.binary.resize(binary_length);
if (!LoadArrayFromPrecompiled(dump.binary.data(), dump.binary.size())) {
return {};
}
dumps.emplace(std::move(usage), dump);
}
return dumps;
return entries;
}
void ShaderDiskCacheOpenGL::InvalidateTransferable() {
@ -346,13 +299,13 @@ void ShaderDiskCacheOpenGL::InvalidatePrecompiled() {
}
}
void ShaderDiskCacheOpenGL::SaveRaw(const ShaderDiskCacheRaw& entry) {
void ShaderDiskCacheOpenGL::SaveEntry(const ShaderDiskCacheEntry& entry) {
if (!is_usable) {
return;
}
const u64 id = entry.GetUniqueIdentifier();
if (transferable.find(id) != transferable.end()) {
const u64 id = entry.unique_identifier;
if (stored_transferable.find(id) != stored_transferable.end()) {
// The shader already exists
return;
}
@ -361,71 +314,17 @@ void ShaderDiskCacheOpenGL::SaveRaw(const ShaderDiskCacheRaw& entry) {
if (!file.IsOpen()) {
return;
}
if (file.WriteObject(TransferableEntryKind::Raw) != 1 || !entry.Save(file)) {
if (!entry.Save(file)) {
LOG_ERROR(Render_OpenGL, "Failed to save raw transferable cache entry, removing");
file.Close();
InvalidateTransferable();
return;
}
transferable.insert({id, {}});
stored_transferable.insert(id);
}
void ShaderDiskCacheOpenGL::SaveUsage(const ShaderDiskCacheUsage& usage) {
if (!is_usable) {
return;
}
const auto it = transferable.find(usage.unique_identifier);
ASSERT_MSG(it != transferable.end(), "Saving shader usage without storing raw previously");
auto& usages{it->second};
if (usages.find(usage) != usages.end()) {
// Skip this variant since the shader is already stored.
return;
}
usages.insert(usage);
FileUtil::IOFile file = AppendTransferableFile();
if (!file.IsOpen())
return;
const auto Close = [&] {
LOG_ERROR(Render_OpenGL, "Failed to save usage transferable cache entry, removing");
file.Close();
InvalidateTransferable();
};
if (file.WriteObject(TransferableEntryKind::Usage) != 1 ||
file.WriteObject(usage.unique_identifier) != 1 || file.WriteObject(usage.variant) != 1 ||
file.WriteObject(usage.bound_buffer) != 1 ||
file.WriteObject(static_cast<u32>(usage.keys.size())) != 1 ||
file.WriteObject(static_cast<u32>(usage.bound_samplers.size())) != 1 ||
file.WriteObject(static_cast<u32>(usage.bindless_samplers.size())) != 1) {
Close();
return;
}
for (const auto& [pair, value] : usage.keys) {
const auto [cbuf, offset] = pair;
if (file.WriteObject(ConstBufferKey{cbuf, offset, value}) != 1) {
Close();
return;
}
}
for (const auto& [offset, sampler] : usage.bound_samplers) {
if (file.WriteObject(BoundSamplerKey{offset, sampler}) != 1) {
Close();
return;
}
}
for (const auto& [pair, sampler] : usage.bindless_samplers) {
const auto [cbuf, offset] = pair;
if (file.WriteObject(BindlessSamplerKey{cbuf, offset, sampler}) != 1) {
Close();
return;
}
}
}
void ShaderDiskCacheOpenGL::SaveDump(const ShaderDiskCacheUsage& usage, GLuint program) {
void ShaderDiskCacheOpenGL::SavePrecompiled(u64 unique_identifier, GLuint program) {
if (!is_usable) {
return;
}
@ -437,51 +336,19 @@ void ShaderDiskCacheOpenGL::SaveDump(const ShaderDiskCacheUsage& usage, GLuint p
SavePrecompiledHeaderToVirtualPrecompiledCache();
}
GLint binary_length{};
GLint binary_length;
glGetProgramiv(program, GL_PROGRAM_BINARY_LENGTH, &binary_length);
GLenum binary_format{};
GLenum binary_format;
std::vector<u8> binary(binary_length);
glGetProgramBinary(program, binary_length, nullptr, &binary_format, binary.data());
const auto Close = [&] {
LOG_ERROR(Render_OpenGL, "Failed to save binary program file in shader={:016X}, removing",
usage.unique_identifier);
InvalidatePrecompiled();
};
if (!SaveObjectToPrecompiled(usage.unique_identifier) ||
!SaveObjectToPrecompiled(usage.variant) || !SaveObjectToPrecompiled(usage.bound_buffer) ||
!SaveObjectToPrecompiled(static_cast<u32>(usage.keys.size())) ||
!SaveObjectToPrecompiled(static_cast<u32>(usage.bound_samplers.size())) ||
!SaveObjectToPrecompiled(static_cast<u32>(usage.bindless_samplers.size()))) {
Close();
return;
}
for (const auto& [pair, value] : usage.keys) {
const auto [cbuf, offset] = pair;
if (SaveObjectToPrecompiled(ConstBufferKey{cbuf, offset, value}) != 1) {
Close();
return;
}
}
for (const auto& [offset, sampler] : usage.bound_samplers) {
if (SaveObjectToPrecompiled(BoundSamplerKey{offset, sampler}) != 1) {
Close();
return;
}
}
for (const auto& [pair, sampler] : usage.bindless_samplers) {
const auto [cbuf, offset] = pair;
if (SaveObjectToPrecompiled(BindlessSamplerKey{cbuf, offset, sampler}) != 1) {
Close();
return;
}
}
if (!SaveObjectToPrecompiled(static_cast<u32>(binary_format)) ||
!SaveObjectToPrecompiled(static_cast<u32>(binary_length)) ||
if (!SaveObjectToPrecompiled(unique_identifier) || !SaveObjectToPrecompiled(binary_format) ||
!SaveObjectToPrecompiled(static_cast<u32>(binary.size())) ||
!SaveArrayToPrecompiled(binary.data(), binary.size())) {
Close();
LOG_ERROR(Render_OpenGL, "Failed to save binary program file in shader={:016X}, removing",
unique_identifier);
InvalidatePrecompiled();
}
}
@ -534,7 +401,6 @@ void ShaderDiskCacheOpenGL::SaveVirtualPrecompiledFile() {
if (file.WriteBytes(compressed.data(), compressed.size()) != compressed.size()) {
LOG_ERROR(Render_OpenGL, "Failed to write precompiled cache version in path={}",
precompiled_path);
return;
}
}

153
src/video_core/renderer_opengl/gl_shader_disk_cache.h
View file

@ -19,8 +19,7 @@
#include "common/common_types.h"
#include "core/file_sys/vfs_vector.h"
#include "video_core/engines/shader_type.h"
#include "video_core/renderer_opengl/gl_shader_gen.h"
#include "video_core/shader/const_buffer_locker.h"
#include "video_core/shader/registry.h"
namespace Core {
class System;
@ -32,139 +31,39 @@ class IOFile;
namespace OpenGL {
struct ShaderDiskCacheUsage;
struct ShaderDiskCacheDump;
using ProgramCode = std::vector<u64>;
using ShaderDumpsMap = std::unordered_map<ShaderDiskCacheUsage, ShaderDiskCacheDump>;
/// Describes the different variants a program can be compiled with.
struct ProgramVariant final {
ProgramVariant() = default;
/// Graphics constructor.
explicit constexpr ProgramVariant(GLenum primitive_mode) noexcept
: primitive_mode{primitive_mode} {}
/// Compute constructor.
explicit constexpr ProgramVariant(u32 block_x, u32 block_y, u32 block_z, u32 shared_memory_size,
u32 local_memory_size) noexcept
: block_x{block_x}, block_y{static_cast<u16>(block_y)}, block_z{static_cast<u16>(block_z)},
shared_memory_size{shared_memory_size}, local_memory_size{local_memory_size} {}
// Graphics specific parameters.
GLenum primitive_mode{};
// Compute specific parameters.
u32 block_x{};
u16 block_y{};
u16 block_z{};
u32 shared_memory_size{};
u32 local_memory_size{};
bool operator==(const ProgramVariant& rhs) const noexcept {
return std::tie(primitive_mode, block_x, block_y, block_z, shared_memory_size,
local_memory_size) == std::tie(rhs.primitive_mode, rhs.block_x, rhs.block_y,
rhs.block_z, rhs.shared_memory_size,
rhs.local_memory_size);
}
bool operator!=(const ProgramVariant& rhs) const noexcept {
return !operator==(rhs);
}
};
static_assert(std::is_trivially_copyable_v<ProgramVariant>);
/// Describes how a shader is used.
struct ShaderDiskCacheUsage {
u64 unique_identifier{};
ProgramVariant variant;
u32 bound_buffer{};
VideoCommon::Shader::KeyMap keys;
VideoCommon::Shader::BoundSamplerMap bound_samplers;
VideoCommon::Shader::BindlessSamplerMap bindless_samplers;
bool operator==(const ShaderDiskCacheUsage& rhs) const {
return std::tie(unique_identifier, variant, keys, bound_samplers, bindless_samplers) ==
std::tie(rhs.unique_identifier, rhs.variant, rhs.keys, rhs.bound_samplers,
rhs.bindless_samplers);
}
bool operator!=(const ShaderDiskCacheUsage& rhs) const {
return !operator==(rhs);
}
};
} // namespace OpenGL
namespace std {
template <>
struct hash<OpenGL::ProgramVariant> {
std::size_t operator()(const OpenGL::ProgramVariant& variant) const noexcept {
return (static_cast<std::size_t>(variant.primitive_mode) << 6) ^
static_cast<std::size_t>(variant.block_x) ^
(static_cast<std::size_t>(variant.block_y) << 32) ^
(static_cast<std::size_t>(variant.block_z) << 48) ^
(static_cast<std::size_t>(variant.shared_memory_size) << 16) ^
(static_cast<std::size_t>(variant.local_memory_size) << 36);
}
};
template <>
struct hash<OpenGL::ShaderDiskCacheUsage> {
std::size_t operator()(const OpenGL::ShaderDiskCacheUsage& usage) const noexcept {
return static_cast<std::size_t>(usage.unique_identifier) ^
std::hash<OpenGL::ProgramVariant>{}(usage.variant);
}
};
} // namespace std
namespace OpenGL {
/// Describes a shader how it's used by the guest GPU
class ShaderDiskCacheRaw {
public:
explicit ShaderDiskCacheRaw(u64 unique_identifier, Tegra::Engines::ShaderType type,
ProgramCode code, ProgramCode code_b = {});
ShaderDiskCacheRaw();
~ShaderDiskCacheRaw();
/// Describes a shader and how it's used by the guest GPU
struct ShaderDiskCacheEntry {
ShaderDiskCacheEntry();
~ShaderDiskCacheEntry();
bool Load(FileUtil::IOFile& file);
bool Save(FileUtil::IOFile& file) const;
u64 GetUniqueIdentifier() const {
return unique_identifier;
}
bool HasProgramA() const {
return !code.empty() && !code_b.empty();
}
Tegra::Engines::ShaderType GetType() const {
return type;
}
const ProgramCode& GetCode() const {
return code;
}
const ProgramCode& GetCodeB() const {
return code_b;
}
private:
u64 unique_identifier{};
Tegra::Engines::ShaderType type{};
ProgramCode code;
ProgramCode code_b;
u64 unique_identifier = 0;
std::optional<u32> texture_handler_size;
u32 bound_buffer = 0;
VideoCommon::Shader::GraphicsInfo graphics_info;
VideoCommon::Shader::ComputeInfo compute_info;
VideoCommon::Shader::KeyMap keys;
VideoCommon::Shader::BoundSamplerMap bound_samplers;
VideoCommon::Shader::BindlessSamplerMap bindless_samplers;
};
/// Contains an OpenGL dumped binary program
struct ShaderDiskCacheDump {
GLenum binary_format{};
struct ShaderDiskCachePrecompiled {
u64 unique_identifier = 0;
GLenum binary_format = 0;
std::vector<u8> binary;
};
@ -174,11 +73,10 @@ public:
~ShaderDiskCacheOpenGL();
/// Loads transferable cache. If file has a old version or on failure, it deletes the file.
std::optional<std::pair<std::vector<ShaderDiskCacheRaw>, std::vector<ShaderDiskCacheUsage>>>
LoadTransferable();
std::optional<std::vector<ShaderDiskCacheEntry>> LoadTransferable();
/// Loads current game's precompiled cache. Invalidates on failure.
std::unordered_map<ShaderDiskCacheUsage, ShaderDiskCacheDump> LoadPrecompiled();
std::vector<ShaderDiskCachePrecompiled> LoadPrecompiled();
/// Removes the transferable (and precompiled) cache file.
void InvalidateTransferable();
@ -187,21 +85,18 @@ public:
void InvalidatePrecompiled();
/// Saves a raw dump to the transferable file. Checks for collisions.
void SaveRaw(const ShaderDiskCacheRaw& entry);
/// Saves shader usage to the transferable file. Does not check for collisions.
void SaveUsage(const ShaderDiskCacheUsage& usage);
void SaveEntry(const ShaderDiskCacheEntry& entry);
/// Saves a dump entry to the precompiled file. Does not check for collisions.
void SaveDump(const ShaderDiskCacheUsage& usage, GLuint program);
void SavePrecompiled(u64 unique_identifier, GLuint program);
/// Serializes virtual precompiled shader cache file to real file
void SaveVirtualPrecompiledFile();
private:
/// Loads the transferable cache. Returns empty on failure.
std::optional<std::unordered_map<ShaderDiskCacheUsage, ShaderDiskCacheDump>>
LoadPrecompiledFile(FileUtil::IOFile& file);
std::optional<std::vector<ShaderDiskCachePrecompiled>> LoadPrecompiledFile(
FileUtil::IOFile& file);
/// Opens current game's transferable file and write it's header if it doesn't exist
FileUtil::IOFile AppendTransferableFile() const;
@ -270,7 +165,7 @@ private:
std::size_t precompiled_cache_virtual_file_offset = 0;
// Stored transferable shaders
std::unordered_map<u64, std::unordered_set<ShaderDiskCacheUsage>> transferable;
std::unordered_set<u64> stored_transferable;
// The cache has been loaded at boot
bool is_usable{};

109
src/video_core/renderer_opengl/gl_shader_gen.cpp
View file

@ -1,109 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <string>
#include <fmt/format.h>
#include "video_core/engines/maxwell_3d.h"
#include "video_core/engines/shader_type.h"
#include "video_core/renderer_opengl/gl_device.h"
#include "video_core/renderer_opengl/gl_shader_decompiler.h"
#include "video_core/renderer_opengl/gl_shader_gen.h"
#include "video_core/shader/shader_ir.h"
namespace OpenGL::GLShader {
using Tegra::Engines::Maxwell3D;
using Tegra::Engines::ShaderType;
using VideoCommon::Shader::CompileDepth;
using VideoCommon::Shader::CompilerSettings;
using VideoCommon::Shader::ProgramCode;
using VideoCommon::Shader::ShaderIR;
std::string GenerateVertexShader(const Device& device, const ShaderIR& ir, const ShaderIR* ir_b) {
std::string out = GetCommonDeclarations();
out += fmt::format(R"(
layout (std140, binding = {}) uniform vs_config {{
float y_direction;
}};
)",
EmulationUniformBlockBinding);
out += Decompile(device, ir, ShaderType::Vertex, "vertex");
if (ir_b) {
out += Decompile(device, *ir_b, ShaderType::Vertex, "vertex_b");
}
out += R"(
void main() {
gl_Position = vec4(0.0f, 0.0f, 0.0f, 1.0f);
execute_vertex();
)";
if (ir_b) {
out += " execute_vertex_b();";
}
out += "}\n";
return out;
}
std::string GenerateGeometryShader(const Device& device, const ShaderIR& ir) {
std::string out = GetCommonDeclarations();
out += fmt::format(R"(
layout (std140, binding = {}) uniform gs_config {{
float y_direction;
}};
)",
EmulationUniformBlockBinding);
out += Decompile(device, ir, ShaderType::Geometry, "geometry");
out += R"(
void main() {
execute_geometry();
}
)";
return out;
}
std::string GenerateFragmentShader(const Device& device, const ShaderIR& ir) {
std::string out = GetCommonDeclarations();
out += fmt::format(R"(
layout (location = 0) out vec4 FragColor0;
layout (location = 1) out vec4 FragColor1;
layout (location = 2) out vec4 FragColor2;
layout (location = 3) out vec4 FragColor3;
layout (location = 4) out vec4 FragColor4;
layout (location = 5) out vec4 FragColor5;
layout (location = 6) out vec4 FragColor6;
layout (location = 7) out vec4 FragColor7;
layout (std140, binding = {}) uniform fs_config {{
float y_direction;
}};
)",
EmulationUniformBlockBinding);
out += Decompile(device, ir, ShaderType::Fragment, "fragment");
out += R"(
void main() {
execute_fragment();
}
)";
return out;
}
std::string GenerateComputeShader(const Device& device, const ShaderIR& ir) {
std::string out = GetCommonDeclarations();
out += Decompile(device, ir, ShaderType::Compute, "compute");
out += R"(
void main() {
execute_compute();
}
)";
return out;
}
} // namespace OpenGL::GLShader

34
src/video_core/renderer_opengl/gl_shader_gen.h
View file

@ -1,34 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <vector>
#include "common/common_types.h"
#include "video_core/renderer_opengl/gl_shader_decompiler.h"
#include "video_core/shader/shader_ir.h"
namespace OpenGL {
class Device;
}
namespace OpenGL::GLShader {
using VideoCommon::Shader::ProgramCode;
using VideoCommon::Shader::ShaderIR;
/// Generates the GLSL vertex shader program source code for the given VS program
std::string GenerateVertexShader(const Device& device, const ShaderIR& ir, const ShaderIR* ir_b);
/// Generates the GLSL geometry shader program source code for the given GS program
std::string GenerateGeometryShader(const Device& device, const ShaderIR& ir);
/// Generates the GLSL fragment shader program source code for the given FS program
std::string GenerateFragmentShader(const Device& device, const ShaderIR& ir);
/// Generates the GLSL compute shader program source code for the given CS program
std::string GenerateComputeShader(const Device& device, const ShaderIR& ir);
} // namespace OpenGL::GLShader

4
src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
View file

@ -161,8 +161,8 @@ CachedShader::CachedShader(Core::System& system, Tegra::Engines::ShaderType stag
GPUVAddr gpu_addr, VAddr cpu_addr, u8* host_ptr,
ProgramCode program_code, u32 main_offset)
: RasterizerCacheObject{host_ptr}, gpu_addr{gpu_addr}, cpu_addr{cpu_addr},
program_code{std::move(program_code)}, locker{stage, GetEngine(system, stage)},
shader_ir{this->program_code, main_offset, compiler_settings, locker},
program_code{std::move(program_code)}, registry{stage, GetEngine(system, stage)},
shader_ir{this->program_code, main_offset, compiler_settings, registry},
entries{GenerateShaderEntries(shader_ir)} {}
CachedShader::~CachedShader() = default;

4
src/video_core/renderer_vulkan/vk_pipeline_cache.h
View file

@ -25,7 +25,7 @@
#include "video_core/renderer_vulkan/vk_renderpass_cache.h"
#include "video_core/renderer_vulkan/vk_resource_manager.h"
#include "video_core/renderer_vulkan/vk_shader_decompiler.h"
#include "video_core/shader/const_buffer_locker.h"
#include "video_core/shader/registry.h"
#include "video_core/shader/shader_ir.h"
#include "video_core/surface.h"
@ -147,7 +147,7 @@ private:
GPUVAddr gpu_addr{};
VAddr cpu_addr{};
ProgramCode program_code;
VideoCommon::Shader::ConstBufferLocker locker;
VideoCommon::Shader::Registry registry;
VideoCommon::Shader::ShaderIR shader_ir;
ShaderEntries entries;
};

126
src/video_core/shader/const_buffer_locker.cpp
View file

@ -1,126 +0,0 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <algorithm>
#include <tuple>
#include "common/common_types.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/engines/shader_type.h"
#include "video_core/shader/const_buffer_locker.h"
namespace VideoCommon::Shader {
using Tegra::Engines::SamplerDescriptor;
ConstBufferLocker::ConstBufferLocker(Tegra::Engines::ShaderType shader_stage)
: stage{shader_stage} {}
ConstBufferLocker::ConstBufferLocker(Tegra::Engines::ShaderType shader_stage,
Tegra::Engines::ConstBufferEngineInterface& engine)
: stage{shader_stage}, engine{&engine} {}
ConstBufferLocker::~ConstBufferLocker() = default;
std::optional<u32> ConstBufferLocker::ObtainKey(u32 buffer, u32 offset) {
const std::pair<u32, u32> key = {buffer, offset};
const auto iter = keys.find(key);
if (iter != keys.end()) {
return iter->second;
}
if (!engine) {
return std::nullopt;
}
const u32 value = engine->AccessConstBuffer32(stage, buffer, offset);
keys.emplace(key, value);
return value;
}
std::optional<SamplerDescriptor> ConstBufferLocker::ObtainBoundSampler(u32 offset) {
const u32 key = offset;
const auto iter = bound_samplers.find(key);
if (iter != bound_samplers.end()) {
return iter->second;
}
if (!engine) {
return std::nullopt;
}
const SamplerDescriptor value = engine->AccessBoundSampler(stage, offset);
bound_samplers.emplace(key, value);
return value;
}
std::optional<Tegra::Engines::SamplerDescriptor> ConstBufferLocker::ObtainBindlessSampler(
u32 buffer, u32 offset) {
const std::pair key = {buffer, offset};
const auto iter = bindless_samplers.find(key);
if (iter != bindless_samplers.end()) {
return iter->second;
}
if (!engine) {
return std::nullopt;
}
const SamplerDescriptor value = engine->AccessBindlessSampler(stage, buffer, offset);
bindless_samplers.emplace(key, value);
return value;
}
std::optional<u32> ConstBufferLocker::ObtainBoundBuffer() {
if (bound_buffer_saved) {
return bound_buffer;
}
if (!engine) {
return std::nullopt;
}
bound_buffer_saved = true;
bound_buffer = engine->GetBoundBuffer();
return bound_buffer;
}
void ConstBufferLocker::InsertKey(u32 buffer, u32 offset, u32 value) {
keys.insert_or_assign({buffer, offset}, value);
}
void ConstBufferLocker::InsertBoundSampler(u32 offset, SamplerDescriptor sampler) {
bound_samplers.insert_or_assign(offset, sampler);
}
void ConstBufferLocker::InsertBindlessSampler(u32 buffer, u32 offset, SamplerDescriptor sampler) {
bindless_samplers.insert_or_assign({buffer, offset}, sampler);
}
void ConstBufferLocker::SetBoundBuffer(u32 buffer) {
bound_buffer_saved = true;
bound_buffer = buffer;
}
bool ConstBufferLocker::IsConsistent() const {
if (!engine) {
return false;
}
return std::all_of(keys.begin(), keys.end(),
[this](const auto& pair) {
const auto [cbuf, offset] = pair.first;
const auto value = pair.second;
return value == engine->AccessConstBuffer32(stage, cbuf, offset);
}) &&
std::all_of(bound_samplers.begin(), bound_samplers.end(),
[this](const auto& sampler) {
const auto [key, value] = sampler;
return value == engine->AccessBoundSampler(stage, key);
}) &&
std::all_of(bindless_samplers.begin(), bindless_samplers.end(),
[this](const auto& sampler) {
const auto [cbuf, offset] = sampler.first;
const auto value = sampler.second;
return value == engine->AccessBindlessSampler(stage, cbuf, offset);
});
}
bool ConstBufferLocker::HasEqualKeys(const ConstBufferLocker& rhs) const {
return std::tie(keys, bound_samplers, bindless_samplers) ==
std::tie(rhs.keys, rhs.bound_samplers, rhs.bindless_samplers);
}
} // namespace VideoCommon::Shader

103
src/video_core/shader/const_buffer_locker.h
View file

@ -1,103 +0,0 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <optional>
#include <unordered_map>
#include "common/common_types.h"
#include "common/hash.h"
#include "video_core/engines/const_buffer_engine_interface.h"
#include "video_core/engines/shader_type.h"
#include "video_core/guest_driver.h"
namespace VideoCommon::Shader {
using KeyMap = std::unordered_map<std::pair<u32, u32>, u32, Common::PairHash>;
using BoundSamplerMap = std::unordered_map<u32, Tegra::Engines::SamplerDescriptor>;
using BindlessSamplerMap =
std::unordered_map<std::pair<u32, u32>, Tegra::Engines::SamplerDescriptor, Common::PairHash>;
/**
* The ConstBufferLocker is a class use to interface the 3D and compute engines with the shader
* compiler. with it, the shader can obtain required data from GPU state and store it for disk
* shader compilation.
*/
class ConstBufferLocker {
public:
explicit ConstBufferLocker(Tegra::Engines::ShaderType shader_stage);
explicit ConstBufferLocker(Tegra::Engines::ShaderType shader_stage,
Tegra::Engines::ConstBufferEngineInterface& engine);
~ConstBufferLocker();
/// Retrieves a key from the locker, if it's registered, it will give the registered value, if
/// not it will obtain it from maxwell3d and register it.
std::optional<u32> ObtainKey(u32 buffer, u32 offset);
std::optional<Tegra::Engines::SamplerDescriptor> ObtainBoundSampler(u32 offset);
std::optional<Tegra::Engines::SamplerDescriptor> ObtainBindlessSampler(u32 buffer, u32 offset);
std::optional<u32> ObtainBoundBuffer();
/// Inserts a key.
void InsertKey(u32 buffer, u32 offset, u32 value);
/// Inserts a bound sampler key.
void InsertBoundSampler(u32 offset, Tegra::Engines::SamplerDescriptor sampler);
/// Inserts a bindless sampler key.
void InsertBindlessSampler(u32 buffer, u32 offset, Tegra::Engines::SamplerDescriptor sampler);
/// Set the bound buffer for this locker.
void SetBoundBuffer(u32 buffer);
/// Checks keys and samplers against engine's current const buffers. Returns true if they are
/// the same value, false otherwise;
bool IsConsistent() const;
/// Returns true if the keys are equal to the other ones in the locker.
bool HasEqualKeys(const ConstBufferLocker& rhs) const;
/// Gives an getter to the const buffer keys in the database.
const KeyMap& GetKeys() const {
return keys;
}
/// Gets samplers database.
const BoundSamplerMap& GetBoundSamplers() const {
return bound_samplers;
}
/// Gets bindless samplers database.
const BindlessSamplerMap& GetBindlessSamplers() const {
return bindless_samplers;
}
/// Gets bound buffer used on this shader
u32 GetBoundBuffer() const {
return bound_buffer;
}
/// Obtains access to the guest driver's profile.
VideoCore::GuestDriverProfile* AccessGuestDriverProfile() const {
if (engine) {
return &engine->AccessGuestDriverProfile();
}
return nullptr;
}
private:
const Tegra::Engines::ShaderType stage;
Tegra::Engines::ConstBufferEngineInterface* engine = nullptr;
KeyMap keys;
BoundSamplerMap bound_samplers;
BindlessSamplerMap bindless_samplers;
bool bound_buffer_saved{};
u32 bound_buffer{};
};
} // namespace VideoCommon::Shader

13
src/video_core/shader/control_flow.cpp
View file

@ -13,6 +13,7 @@
#include "common/common_types.h"
#include "video_core/shader/ast.h"
#include "video_core/shader/control_flow.h"
#include "video_core/shader/registry.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
@ -64,11 +65,11 @@ struct BlockInfo {
};
struct CFGRebuildState {
explicit CFGRebuildState(const ProgramCode& program_code, u32 start, ConstBufferLocker& locker)
: program_code{program_code}, locker{locker}, start{start} {}
explicit CFGRebuildState(const ProgramCode& program_code, u32 start, Registry& registry)
: program_code{program_code}, registry{registry}, start{start} {}
const ProgramCode& program_code;
ConstBufferLocker& locker;
Registry& registry;
u32 start{};
std::vector<BlockInfo> block_info;
std::list<u32> inspect_queries;
@ -438,7 +439,7 @@ std::pair<ParseResult, ParseInfo> ParseCode(CFGRebuildState& state, u32 address)
const s32 pc_target = offset + result.relative_position;
std::vector<CaseBranch> branches;
for (u32 i = 0; i < result.entries; i++) {
auto key = state.locker.ObtainKey(result.buffer, result.offset + i * 4);
auto key = state.registry.ObtainKey(result.buffer, result.offset + i * 4);
if (!key) {
return {ParseResult::AbnormalFlow, parse_info};
}
@ -656,14 +657,14 @@ void DecompileShader(CFGRebuildState& state) {
std::unique_ptr<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code, u32 start_address,
const CompilerSettings& settings,
ConstBufferLocker& locker) {
Registry& registry) {
auto result_out = std::make_unique<ShaderCharacteristics>();
if (settings.depth == CompileDepth::BruteForce) {
result_out->settings.depth = CompileDepth::BruteForce;
return result_out;
}
CFGRebuildState state{program_code, start_address, locker};
CFGRebuildState state{program_code, start_address, registry};
// Inspect Code and generate blocks
state.labels.clear();
state.labels.emplace(start_address);

3
src/video_core/shader/control_flow.h
View file

@ -12,6 +12,7 @@
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/ast.h"
#include "video_core/shader/compiler_settings.h"
#include "video_core/shader/registry.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
@ -111,6 +112,6 @@ struct ShaderCharacteristics {
std::unique_ptr<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code, u32 start_address,
const CompilerSettings& settings,
ConstBufferLocker& locker);
Registry& registry);
} // namespace VideoCommon::Shader

22
src/video_core/shader/decode.cpp
View file

@ -34,13 +34,9 @@ constexpr bool IsSchedInstruction(u32 offset, u32 main_offset) {
return (absolute_offset % SchedPeriod) == 0;
}
void DeduceTextureHandlerSize(VideoCore::GuestDriverProfile* gpu_driver,
void DeduceTextureHandlerSize(VideoCore::GuestDriverProfile& gpu_driver,
const std::list<Sampler>& used_samplers) {
if (gpu_driver == nullptr) {
LOG_CRITICAL(HW_GPU, "GPU driver profile has not been created yet");
return;
}
if (gpu_driver->TextureHandlerSizeKnown() || used_samplers.size() <= 1) {
if (gpu_driver.IsTextureHandlerSizeKnown() || used_samplers.size() <= 1) {
return;
}
u32 count{};
@ -53,17 +49,13 @@ void DeduceTextureHandlerSize(VideoCore::GuestDriverProfile* gpu_driver,
bound_offsets.emplace_back(sampler.GetOffset());
}
if (count > 1) {
gpu_driver->DeduceTextureHandlerSize(std::move(bound_offsets));
gpu_driver.DeduceTextureHandlerSize(std::move(bound_offsets));
}
}
std::optional<u32> TryDeduceSamplerSize(const Sampler& sampler_to_deduce,
VideoCore::GuestDriverProfile* gpu_driver,
VideoCore::GuestDriverProfile& gpu_driver,
const std::list<Sampler>& used_samplers) {
if (gpu_driver == nullptr) {
LOG_CRITICAL(HW_GPU, "GPU Driver profile has not been created yet");
return std::nullopt;
}
const u32 base_offset = sampler_to_deduce.GetOffset();
u32 max_offset{std::numeric_limits<u32>::max()};
for (const auto& sampler : used_samplers) {
@ -77,7 +69,7 @@ std::optional<u32> TryDeduceSamplerSize(const Sampler& sampler_to_deduce,
if (max_offset == std::numeric_limits<u32>::max()) {
return std::nullopt;
}
return ((max_offset - base_offset) * 4) / gpu_driver->GetTextureHandlerSize();
return ((max_offset - base_offset) * 4) / gpu_driver.GetTextureHandlerSize();
}
} // Anonymous namespace
@ -149,7 +141,7 @@ void ShaderIR::Decode() {
std::memcpy(&header, program_code.data(), sizeof(Tegra::Shader::Header));
decompiled = false;
auto info = ScanFlow(program_code, main_offset, settings, locker);
auto info = ScanFlow(program_code, main_offset, settings, registry);
auto& shader_info = *info;
coverage_begin = shader_info.start;
coverage_end = shader_info.end;
@ -364,7 +356,7 @@ u32 ShaderIR::DecodeInstr(NodeBlock& bb, u32 pc) {
void ShaderIR::PostDecode() {
// Deduce texture handler size if needed
auto gpu_driver = locker.AccessGuestDriverProfile();
auto gpu_driver = registry.AccessGuestDriverProfile();
DeduceTextureHandlerSize(gpu_driver, used_samplers);
// Deduce Indexed Samplers
if (!uses_indexed_samplers) {

5
src/video_core/shader/decode/texture.cpp
View file

@ -12,6 +12,7 @@
#include "common/logging/log.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/registry.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
@ -359,8 +360,8 @@ ShaderIR::SamplerInfo ShaderIR::GetSamplerInfo(std::optional<SamplerInfo> sample
if (sampler_info) {
return *sampler_info;
}
const auto sampler =
buffer ? locker.ObtainBindlessSampler(*buffer, offset) : locker.ObtainBoundSampler(offset);
const auto sampler = buffer ? registry.ObtainBindlessSampler(*buffer, offset)
: registry.ObtainBoundSampler(offset);
if (!sampler) {
LOG_WARNING(HW_GPU, "Unknown sampler info");
return SamplerInfo{TextureType::Texture2D, false, false, false};

161
src/video_core/shader/registry.cpp Normal file
View file

@ -0,0 +1,161 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <algorithm>
#include <tuple>
#include "common/assert.h"
#include "common/common_types.h"
#include "video_core/engines/kepler_compute.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/engines/shader_type.h"
#include "video_core/shader/registry.h"
namespace VideoCommon::Shader {
using Tegra::Engines::ConstBufferEngineInterface;
using Tegra::Engines::SamplerDescriptor;
using Tegra::Engines::ShaderType;
namespace {
GraphicsInfo MakeGraphicsInfo(ShaderType shader_stage, ConstBufferEngineInterface& engine) {
if (shader_stage == ShaderType::Compute) {
return {};
}
auto& graphics = static_cast<Tegra::Engines::Maxwell3D&>(engine);
GraphicsInfo info;
info.tfb_layouts = graphics.regs.tfb_layouts;
info.tfb_varying_locs = graphics.regs.tfb_varying_locs;
info.primitive_topology = graphics.regs.draw.topology;
info.tessellation_primitive = graphics.regs.tess_mode.prim;
info.tessellation_spacing = graphics.regs.tess_mode.spacing;
info.tfb_enabled = graphics.regs.tfb_enabled;
info.tessellation_clockwise = graphics.regs.tess_mode.cw;
return info;
}
ComputeInfo MakeComputeInfo(ShaderType shader_stage, ConstBufferEngineInterface& engine) {
if (shader_stage != ShaderType::Compute) {
return {};
}
auto& compute = static_cast<Tegra::Engines::KeplerCompute&>(engine);
const auto& launch = compute.launch_description;
ComputeInfo info;
info.workgroup_size = {launch.block_dim_x, launch.block_dim_y, launch.block_dim_z};
info.local_memory_size_in_words = launch.local_pos_alloc;
info.shared_memory_size_in_words = launch.shared_alloc;
return info;
}
} // Anonymous namespace
Registry::Registry(Tegra::Engines::ShaderType shader_stage, const SerializedRegistryInfo& info)
: stage{shader_stage}, stored_guest_driver_profile{info.guest_driver_profile},
bound_buffer{info.bound_buffer}, graphics_info{info.graphics}, compute_info{info.compute} {}
Registry::Registry(Tegra::Engines::ShaderType shader_stage,
Tegra::Engines::ConstBufferEngineInterface& engine)
: stage{shader_stage}, engine{&engine}, bound_buffer{engine.GetBoundBuffer()},
graphics_info{MakeGraphicsInfo(shader_stage, engine)}, compute_info{MakeComputeInfo(
shader_stage, engine)} {}
Registry::~Registry() = default;
std::optional<u32> Registry::ObtainKey(u32 buffer, u32 offset) {
const std::pair<u32, u32> key = {buffer, offset};
const auto iter = keys.find(key);
if (iter != keys.end()) {
return iter->second;
}
if (!engine) {
return std::nullopt;
}
const u32 value = engine->AccessConstBuffer32(stage, buffer, offset);
keys.emplace(key, value);
return value;
}
std::optional<SamplerDescriptor> Registry::ObtainBoundSampler(u32 offset) {
const u32 key = offset;
const auto iter = bound_samplers.find(key);
if (iter != bound_samplers.end()) {
return iter->second;
}
if (!engine) {
return std::nullopt;
}
const SamplerDescriptor value = engine->AccessBoundSampler(stage, offset);
bound_samplers.emplace(key, value);
return value;
}
std::optional<Tegra::Engines::SamplerDescriptor> Registry::ObtainBindlessSampler(u32 buffer,
u32 offset) {
const std::pair key = {buffer, offset};
const auto iter = bindless_samplers.find(key);
if (iter != bindless_samplers.end()) {
return iter->second;
}
if (!engine) {
return std::nullopt;
}
const SamplerDescriptor value = engine->AccessBindlessSampler(stage, buffer, offset);
bindless_samplers.emplace(key, value);
return value;
}
void Registry::InsertKey(u32 buffer, u32 offset, u32 value) {
keys.insert_or_assign({buffer, offset}, value);
}
void Registry::InsertBoundSampler(u32 offset, SamplerDescriptor sampler) {
bound_samplers.insert_or_assign(offset, sampler);
}
void Registry::InsertBindlessSampler(u32 buffer, u32 offset, SamplerDescriptor sampler) {
bindless_samplers.insert_or_assign({buffer, offset}, sampler);
}
bool Registry::IsConsistent() const {
if (!engine) {
return true;
}
return std::all_of(keys.begin(), keys.end(),
[this](const auto& pair) {
const auto [cbuf, offset] = pair.first;
const auto value = pair.second;
return value == engine->AccessConstBuffer32(stage, cbuf, offset);
}) &&
std::all_of(bound_samplers.begin(), bound_samplers.end(),
[this](const auto& sampler) {
const auto [key, value] = sampler;
return value == engine->AccessBoundSampler(stage, key);
}) &&
std::all_of(bindless_samplers.begin(), bindless_samplers.end(),
[this](const auto& sampler) {
const auto [cbuf, offset] = sampler.first;
const auto value = sampler.second;
return value == engine->AccessBindlessSampler(stage, cbuf, offset);
});
}
bool Registry::HasEqualKeys(const Registry& rhs) const {
return std::tie(keys, bound_samplers, bindless_samplers) ==
std::tie(rhs.keys, rhs.bound_samplers, rhs.bindless_samplers);
}
const GraphicsInfo& Registry::GetGraphicsInfo() const {
ASSERT(stage != Tegra::Engines::ShaderType::Compute);
return graphics_info;
}
const ComputeInfo& Registry::GetComputeInfo() const {
ASSERT(stage == Tegra::Engines::ShaderType::Compute);
return compute_info;
}
} // namespace VideoCommon::Shader

137
src/video_core/shader/registry.h Normal file
View file

@ -0,0 +1,137 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <array>
#include <optional>
#include <type_traits>
#include <unordered_map>
#include <utility>
#include "common/common_types.h"
#include "common/hash.h"
#include "video_core/engines/const_buffer_engine_interface.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/engines/shader_type.h"
#include "video_core/guest_driver.h"
namespace VideoCommon::Shader {
using KeyMap = std::unordered_map<std::pair<u32, u32>, u32, Common::PairHash>;
using BoundSamplerMap = std::unordered_map<u32, Tegra::Engines::SamplerDescriptor>;
using BindlessSamplerMap =
std::unordered_map<std::pair<u32, u32>, Tegra::Engines::SamplerDescriptor, Common::PairHash>;
struct GraphicsInfo {
using Maxwell = Tegra::Engines::Maxwell3D::Regs;
std::array<Maxwell::TransformFeedbackLayout, Maxwell::NumTransformFeedbackBuffers>
tfb_layouts{};
std::array<std::array<u8, 128>, Maxwell::NumTransformFeedbackBuffers> tfb_varying_locs{};
Maxwell::PrimitiveTopology primitive_topology{};
Maxwell::TessellationPrimitive tessellation_primitive{};
Maxwell::TessellationSpacing tessellation_spacing{};
bool tfb_enabled = false;
bool tessellation_clockwise = false;
};
static_assert(std::is_trivially_copyable_v<GraphicsInfo> &&
std::is_standard_layout_v<GraphicsInfo>);
struct ComputeInfo {
std::array<u32, 3> workgroup_size{};
u32 shared_memory_size_in_words = 0;
u32 local_memory_size_in_words = 0;
};
static_assert(std::is_trivially_copyable_v<ComputeInfo> && std::is_standard_layout_v<ComputeInfo>);
struct SerializedRegistryInfo {
VideoCore::GuestDriverProfile guest_driver_profile;
u32 bound_buffer = 0;
GraphicsInfo graphics;
ComputeInfo compute;
};
/**
* The Registry is a class use to interface the 3D and compute engines with the shader compiler.
* With it, the shader can obtain required data from GPU state and store it for disk shader
* compilation.
*/
class Registry {
public:
explicit Registry(Tegra::Engines::ShaderType shader_stage, const SerializedRegistryInfo& info);
explicit Registry(Tegra::Engines::ShaderType shader_stage,
Tegra::Engines::ConstBufferEngineInterface& engine);
~Registry();
/// Retrieves a key from the registry, if it's registered, it will give the registered value, if
/// not it will obtain it from maxwell3d and register it.
std::optional<u32> ObtainKey(u32 buffer, u32 offset);
std::optional<Tegra::Engines::SamplerDescriptor> ObtainBoundSampler(u32 offset);
std::optional<Tegra::Engines::SamplerDescriptor> ObtainBindlessSampler(u32 buffer, u32 offset);
/// Inserts a key.
void InsertKey(u32 buffer, u32 offset, u32 value);
/// Inserts a bound sampler key.
void InsertBoundSampler(u32 offset, Tegra::Engines::SamplerDescriptor sampler);
/// Inserts a bindless sampler key.
void InsertBindlessSampler(u32 buffer, u32 offset, Tegra::Engines::SamplerDescriptor sampler);
/// Checks keys and samplers against engine's current const buffers.
/// Returns true if they are the same value, false otherwise.
bool IsConsistent() const;
/// Returns true if the keys are equal to the other ones in the registry.
bool HasEqualKeys(const Registry& rhs) const;
/// Returns graphics information from this shader
const GraphicsInfo& GetGraphicsInfo() const;
/// Returns compute information from this shader
const ComputeInfo& GetComputeInfo() const;
/// Gives an getter to the const buffer keys in the database.
const KeyMap& GetKeys() const {
return keys;
}
/// Gets samplers database.
const BoundSamplerMap& GetBoundSamplers() const {
return bound_samplers;
}
/// Gets bindless samplers database.
const BindlessSamplerMap& GetBindlessSamplers() const {
return bindless_samplers;
}
/// Gets bound buffer used on this shader
u32 GetBoundBuffer() const {
return bound_buffer;
}
/// Obtains access to the guest driver's profile.
VideoCore::GuestDriverProfile& AccessGuestDriverProfile() {
return engine ? engine->AccessGuestDriverProfile() : stored_guest_driver_profile;
}
private:
const Tegra::Engines::ShaderType stage;
VideoCore::GuestDriverProfile stored_guest_driver_profile;
Tegra::Engines::ConstBufferEngineInterface* engine = nullptr;
KeyMap keys;
BoundSamplerMap bound_samplers;
BindlessSamplerMap bindless_samplers;
u32 bound_buffer;
GraphicsInfo graphics_info;
ComputeInfo compute_info;
};
} // namespace VideoCommon::Shader

5
src/video_core/shader/shader_ir.cpp
View file

@ -11,6 +11,7 @@
#include "common/logging/log.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/registry.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
@ -24,8 +25,8 @@ using Tegra::Shader::PredOperation;
using Tegra::Shader::Register;
ShaderIR::ShaderIR(const ProgramCode& program_code, u32 main_offset, CompilerSettings settings,
ConstBufferLocker& locker)
: program_code{program_code}, main_offset{main_offset}, settings{settings}, locker{locker} {
Registry& registry)
: program_code{program_code}, main_offset{main_offset}, settings{settings}, registry{registry} {
Decode();
PostDecode();
}

6
src/video_core/shader/shader_ir.h
View file

@ -18,8 +18,8 @@
#include "video_core/engines/shader_header.h"
#include "video_core/shader/ast.h"
#include "video_core/shader/compiler_settings.h"
#include "video_core/shader/const_buffer_locker.h"
#include "video_core/shader/node.h"
#include "video_core/shader/registry.h"
namespace VideoCommon::Shader {
@ -69,7 +69,7 @@ struct GlobalMemoryUsage {
class ShaderIR final {
public:
explicit ShaderIR(const ProgramCode& program_code, u32 main_offset, CompilerSettings settings,
ConstBufferLocker& locker);
Registry& registry);
~ShaderIR();
const std::map<u32, NodeBlock>& GetBasicBlocks() const {
@ -414,7 +414,7 @@ private:
const ProgramCode& program_code;
const u32 main_offset;
const CompilerSettings settings;
ConstBufferLocker& locker;
Registry& registry;
bool decompiled{};
bool disable_flow_stack{};

18
src/video_core/shader/track.cpp
View file

@ -81,26 +81,20 @@ std::tuple<Node, TrackSampler> ShaderIR::TrackBindlessSampler(Node tracked, cons
MakeTrackSampler<BindlessSamplerNode>(cbuf->GetIndex(), immediate->GetValue());
return {tracked, track};
} else if (const auto operation = std::get_if<OperationNode>(&*offset)) {
auto bound_buffer = locker.ObtainBoundBuffer();
if (!bound_buffer) {
const u32 bound_buffer = registry.GetBoundBuffer();
if (bound_buffer != cbuf->GetIndex()) {
return {};
}
if (*bound_buffer != cbuf->GetIndex()) {
return {};
}
auto pair = DecoupleIndirectRead(*operation);
const auto pair = DecoupleIndirectRead(*operation);
if (!pair) {
return {};
}
auto [gpr, base_offset] = *pair;
const auto offset_inm = std::get_if<ImmediateNode>(&*base_offset);
auto gpu_driver = locker.AccessGuestDriverProfile();
if (gpu_driver == nullptr) {
return {};
}
const auto& gpu_driver = registry.AccessGuestDriverProfile();
const u32 bindless_cv = NewCustomVariable();
const Node op = Operation(OperationCode::UDiv, NO_PRECISE, gpr,
Immediate(gpu_driver->GetTextureHandlerSize()));
const Node op =
Operation(OperationCode::UDiv, gpr, Immediate(gpu_driver.GetTextureHandlerSize()));
const Node cv_node = GetCustomVariable(bindless_cv);
Node amend_op = Operation(OperationCode::Assign, cv_node, std::move(op));

17
src/yuzu/loading_screen.cpp
View file

@ -34,18 +34,6 @@ constexpr char PROGRESSBAR_STYLE_PREPARE[] = R"(
QProgressBar {}
QProgressBar::chunk {})";
constexpr char PROGRESSBAR_STYLE_DECOMPILE[] = R"(
QProgressBar {
background-color: black;
border: 2px solid white;
border-radius: 4px;
padding: 2px;
}
QProgressBar::chunk {
background-color: #0ab9e6;
width: 1px;
})";
constexpr char PROGRESSBAR_STYLE_BUILD[] = R"(
QProgressBar {
background-color: black;
@ -100,13 +88,11 @@ LoadingScreen::LoadingScreen(QWidget* parent)
stage_translations = {
{VideoCore::LoadCallbackStage::Prepare, tr("Loading...")},
{VideoCore::LoadCallbackStage::Decompile, tr("Preparing Shaders %1 / %2")},
{VideoCore::LoadCallbackStage::Build, tr("Loading Shaders %1 / %2")},
{VideoCore::LoadCallbackStage::Complete, tr("Launching...")},
};
progressbar_style = {
{VideoCore::LoadCallbackStage::Prepare, PROGRESSBAR_STYLE_PREPARE},
{VideoCore::LoadCallbackStage::Decompile, PROGRESSBAR_STYLE_DECOMPILE},
{VideoCore::LoadCallbackStage::Build, PROGRESSBAR_STYLE_BUILD},
{VideoCore::LoadCallbackStage::Complete, PROGRESSBAR_STYLE_COMPLETE},
};
@ -192,8 +178,7 @@ void LoadingScreen::OnLoadProgress(VideoCore::LoadCallbackStage stage, std::size
}
// update labels and progress bar
if (stage == VideoCore::LoadCallbackStage::Decompile ||
stage == VideoCore::LoadCallbackStage::Build) {
if (stage == VideoCore::LoadCallbackStage::Build) {
ui->stage->setText(stage_translations[stage].arg(value).arg(total));
} else {
ui->stage->setText(stage_translations[stage]);