CCF_100/suyu-mirror
1
0
Fork 0
mirror of https://git.suyu.dev/suyu/suyu synced 2025-11-04 00:49:02 -06:00
Code Issues Releases Wiki Activity

Merge pull request #2734 from ReinUsesLisp/compute-shaders

Browse Source 
gl_rasterizer: Implement compute shaders
This commit is contained in:
bunnei
2019-07-22 11:12:55 -04:00
committed by GitHub
parent 27e10e0442 2a4044a858
commit f601f25bcc
15 changed files with 358 additions and 141 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
src/video_core/engines/kepler_compute.cpp
View File

@@ -50,13 +50,14 @@ void KeplerCompute::CallMethod(const GPU::MethodCall& method_call) {
}
void KeplerCompute::ProcessLaunch() {
const GPUVAddr launch_desc_loc = regs.launch_desc_loc.Address();
memory_manager.ReadBlockUnsafe(launch_desc_loc, &launch_description,
LaunchParams::NUM_LAUNCH_PARAMETERS * sizeof(u32));
const GPUVAddr code_loc = regs.code_loc.Address() + launch_description.program_start;
LOG_WARNING(HW_GPU, "Compute Kernel Execute at Address 0x{:016x}, STUBBED", code_loc);
const GPUVAddr code_addr = regs.code_loc.Address() + launch_description.program_start;
LOG_TRACE(HW_GPU, "Compute invocation launched at address 0x{:016x}", code_addr);
rasterizer.DispatchCompute(code_addr);
}
} // namespace Tegra::Engines

8
src/video_core/gpu.cpp
View File

@@ -50,6 +50,14 @@ const Engines::Maxwell3D& GPU::Maxwell3D() const {
return *maxwell_3d;
}
Engines::KeplerCompute& GPU::KeplerCompute() {
return *kepler_compute;
}
const Engines::KeplerCompute& GPU::KeplerCompute() const {
return *kepler_compute;
}
MemoryManager& GPU::MemoryManager() {
return *memory_manager;
}

6
src/video_core/gpu.h
View File

@@ -155,6 +155,12 @@ public:
/// Returns a const reference to the Maxwell3D GPU engine.
const Engines::Maxwell3D& Maxwell3D() const;
/// Returns a reference to the KeplerCompute GPU engine.
Engines::KeplerCompute& KeplerCompute();
/// Returns a reference to the KeplerCompute GPU engine.
const Engines::KeplerCompute& KeplerCompute() const;
/// Returns a reference to the GPU memory manager.
Tegra::MemoryManager& MemoryManager();

3
src/video_core/rasterizer_interface.h
View File

@@ -34,6 +34,9 @@ public:
/// Clear the current framebuffer
virtual void Clear() = 0;
/// Dispatches a compute shader invocation
virtual void DispatchCompute(GPUVAddr code_addr) = 0;
/// Notify rasterizer that all caches should be flushed to Switch memory
virtual void FlushAll() = 0;

101
src/video_core/renderer_opengl/gl_rasterizer.cpp
View File

@@ -4,6 +4,7 @@
#include <algorithm>
#include <array>
#include <bitset>
#include <memory>
#include <string>
#include <string_view>
@@ -19,6 +20,7 @@
#include "core/core.h"
#include "core/hle/kernel/process.h"
#include "core/settings.h"
#include "video_core/engines/kepler_compute.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/memory_manager.h"
#include "video_core/renderer_opengl/gl_rasterizer.h"
@@ -326,9 +328,9 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) {
Shader shader{shader_cache.GetStageProgram(program)};
const auto stage_enum{static_cast<Maxwell::ShaderStage>(stage)};
const auto stage_enum = static_cast<Maxwell::ShaderStage>(stage);
SetupDrawConstBuffers(stage_enum, shader);
SetupGlobalRegions(stage_enum, shader);
SetupDrawGlobalMemory(stage_enum, shader);
const auto texture_buffer_usage{SetupTextures(stage_enum, shader, base_bindings)};
const ProgramVariant variant{base_bindings, primitive_mode, texture_buffer_usage};
@@ -783,6 +785,45 @@ void RasterizerOpenGL::DrawArrays() {
gpu.dirty.memory_general = false;
}
void RasterizerOpenGL::DispatchCompute(GPUVAddr code_addr) {
if (!GLAD_GL_ARB_compute_variable_group_size) {
LOG_ERROR(Render_OpenGL, "Compute is currently not supported on this device due to the "
"lack of GL_ARB_compute_variable_group_size");
return;
}
auto kernel = shader_cache.GetComputeKernel(code_addr);
const auto [program, next_bindings] = kernel->GetProgramHandle({});
state.draw.shader_program = program;
state.draw.program_pipeline = 0;
const std::size_t buffer_size =
Tegra::Engines::KeplerCompute::NumConstBuffers *
(Maxwell::MaxConstBufferSize + device.GetUniformBufferAlignment());
buffer_cache.Map(buffer_size);
bind_ubo_pushbuffer.Setup(0);
bind_ssbo_pushbuffer.Setup(0);
SetupComputeConstBuffers(kernel);
SetupComputeGlobalMemory(kernel);
// TODO(Rodrigo): Bind images and samplers
buffer_cache.Unmap();
bind_ubo_pushbuffer.Bind();
bind_ssbo_pushbuffer.Bind();
state.ApplyShaderProgram();
state.ApplyProgramPipeline();
const auto& launch_desc = system.GPU().KeplerCompute().launch_description;
glDispatchComputeGroupSizeARB(launch_desc.grid_dim_x, launch_desc.grid_dim_y,
launch_desc.grid_dim_z, launch_desc.block_dim_x,
launch_desc.block_dim_y, launch_desc.block_dim_z);
}
void RasterizerOpenGL::FlushAll() {}
void RasterizerOpenGL::FlushRegion(CacheAddr addr, u64 size) {
@@ -856,12 +897,25 @@ bool RasterizerOpenGL::AccelerateDisplay(const Tegra::FramebufferConfig& config,
void RasterizerOpenGL::SetupDrawConstBuffers(Tegra::Engines::Maxwell3D::Regs::ShaderStage stage,
const Shader& shader) {
MICROPROFILE_SCOPE(OpenGL_UBO);
const auto stage_index = static_cast<std::size_t>(stage);
const auto& shader_stage = system.GPU().Maxwell3D().state.shader_stages[stage_index];
// Upload only the enabled buffers from the 16 constbuffers of each shader stage
const auto& stages = system.GPU().Maxwell3D().state.shader_stages;
const auto& shader_stage = stages[static_cast<std::size_t>(stage)];
for (const auto& entry : shader->GetShaderEntries().const_buffers) {
SetupConstBuffer(shader_stage.const_buffers[entry.GetIndex()], entry);
const auto& buffer = shader_stage.const_buffers[entry.GetIndex()];
SetupConstBuffer(buffer, entry);
}
}
void RasterizerOpenGL::SetupComputeConstBuffers(const Shader& kernel) {
MICROPROFILE_SCOPE(OpenGL_UBO);
const auto& launch_desc = system.GPU().KeplerCompute().launch_description;
for (const auto& entry : kernel->GetShaderEntries().const_buffers) {
const auto& config = launch_desc.const_buffer_config[entry.GetIndex()];
const std::bitset<8> mask = launch_desc.memory_config.const_buffer_enable_mask.Value();
Tegra::Engines::ConstBufferInfo buffer;
buffer.address = config.Address();
buffer.size = config.size;
buffer.enabled = mask[entry.GetIndex()];
SetupConstBuffer(buffer, entry);
}
}
@@ -882,24 +936,39 @@ void RasterizerOpenGL::SetupConstBuffer(const Tegra::Engines::ConstBufferInfo& b
bind_ubo_pushbuffer.Push(cbuf, offset, size);
}
void RasterizerOpenGL::SetupGlobalRegions(Tegra::Engines::Maxwell3D::Regs::ShaderStage stage,
const Shader& shader) {
void RasterizerOpenGL::SetupDrawGlobalMemory(Tegra::Engines::Maxwell3D::Regs::ShaderStage stage,
const Shader& shader) {
auto& gpu{system.GPU()};
auto& memory_manager{gpu.MemoryManager()};
const auto cbufs{gpu.Maxwell3D().state.shader_stages[static_cast<std::size_t>(stage)]};
const auto alignment{device.GetShaderStorageBufferAlignment()};
for (const auto& entry : shader->GetShaderEntries().global_memory_entries) {
const auto addr{cbufs.const_buffers[entry.GetCbufIndex()].address + entry.GetCbufOffset()};
const auto actual_addr{memory_manager.Read<u64>(addr)};
const auto gpu_addr{memory_manager.Read<u64>(addr)};
const auto size{memory_manager.Read<u32>(addr + 8)};
const auto [ssbo, buffer_offset] =
buffer_cache.UploadMemory(actual_addr, size, alignment, true, entry.IsWritten());
bind_ssbo_pushbuffer.Push(ssbo, buffer_offset, static_cast<GLsizeiptr>(size));
SetupGlobalMemory(entry, gpu_addr, size);
}
}
void RasterizerOpenGL::SetupComputeGlobalMemory(const Shader& kernel) {
auto& gpu{system.GPU()};
auto& memory_manager{gpu.MemoryManager()};
const auto cbufs{gpu.KeplerCompute().launch_description.const_buffer_config};
for (const auto& entry : kernel->GetShaderEntries().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)};
SetupGlobalMemory(entry, gpu_addr, size);
}
}
void RasterizerOpenGL::SetupGlobalMemory(const GLShader::GlobalMemoryEntry& entry,
GPUVAddr gpu_addr, std::size_t size) {
const auto alignment{device.GetShaderStorageBufferAlignment()};
const auto [ssbo, buffer_offset] =
buffer_cache.UploadMemory(gpu_addr, size, alignment, true, entry.IsWritten());
bind_ssbo_pushbuffer.Push(ssbo, buffer_offset, static_cast<GLsizeiptr>(size));
}
TextureBufferUsage RasterizerOpenGL::SetupTextures(Maxwell::ShaderStage stage, const Shader& shader,
BaseBindings base_bindings) {
MICROPROFILE_SCOPE(OpenGL_Texture);

15
src/video_core/renderer_opengl/gl_rasterizer.h
View File

@@ -58,6 +58,7 @@ public:
void DrawArrays() override;
void Clear() override;
void DispatchCompute(GPUVAddr code_addr) override;
void FlushAll() override;
void FlushRegion(CacheAddr addr, u64 size) override;
void InvalidateRegion(CacheAddr addr, u64 size) override;
@@ -115,13 +116,23 @@ private:
void SetupDrawConstBuffers(Tegra::Engines::Maxwell3D::Regs::ShaderStage stage,
const Shader& shader);
/// Configures the current constbuffers to use for the kernel invocation.
void SetupComputeConstBuffers(const Shader& kernel);
/// Configures a constant buffer.
void SetupConstBuffer(const Tegra::Engines::ConstBufferInfo& buffer,
const GLShader::ConstBufferEntry& entry);
/// Configures the current global memory entries to use for the draw command.
void SetupGlobalRegions(Tegra::Engines::Maxwell3D::Regs::ShaderStage stage,
const Shader& shader);
void SetupDrawGlobalMemory(Tegra::Engines::Maxwell3D::Regs::ShaderStage stage,
const Shader& shader);
/// Configures the current global memory entries to use for the kernel invocation.
void SetupComputeGlobalMemory(const Shader& kernel);
/// Configures a constant buffer.
void SetupGlobalMemory(const GLShader::GlobalMemoryEntry& entry, GPUVAddr gpu_addr,
std::size_t size);
/// Configures the current textures to use for the draw command. Returns shaders texture buffer
/// usage.

150
src/video_core/renderer_opengl/gl_shader_cache.cpp
View File

@@ -23,13 +23,13 @@ namespace OpenGL {
using VideoCommon::Shader::ProgramCode;
// One UBO is always reserved for emulation values
constexpr u32 RESERVED_UBOS = 1;
// One UBO is always reserved for emulation values on staged shaders
constexpr u32 STAGE_RESERVED_UBOS = 1;
struct UnspecializedShader {
std::string code;
GLShader::ShaderEntries entries;
Maxwell::ShaderProgram program_type;
ProgramType program_type;
};
namespace {
@@ -55,15 +55,17 @@ ProgramCode GetShaderCode(Tegra::MemoryManager& memory_manager, const GPUVAddr g
}
/// Gets the shader type from a Maxwell program type
constexpr GLenum GetShaderType(Maxwell::ShaderProgram program_type) {
constexpr GLenum GetShaderType(ProgramType program_type) {
switch (program_type) {
case Maxwell::ShaderProgram::VertexA:
case Maxwell::ShaderProgram::VertexB:
case ProgramType::VertexA:
case ProgramType::VertexB:
return GL_VERTEX_SHADER;
case Maxwell::ShaderProgram::Geometry:
case ProgramType::Geometry:
return GL_GEOMETRY_SHADER;
case Maxwell::ShaderProgram::Fragment:
case ProgramType::Fragment:
return GL_FRAGMENT_SHADER;
case ProgramType::Compute:
return GL_COMPUTE_SHADER;
default:
return GL_NONE;
}
@@ -100,6 +102,25 @@ constexpr std::tuple<const char*, const char*, u32> GetPrimitiveDescription(GLen
}
}
ProgramType GetProgramType(Maxwell::ShaderProgram program) {
switch (program) {
case Maxwell::ShaderProgram::VertexA:
return ProgramType::VertexA;
case Maxwell::ShaderProgram::VertexB:
return ProgramType::VertexB;
case Maxwell::ShaderProgram::TesselationControl:
return ProgramType::TessellationControl;
case Maxwell::ShaderProgram::TesselationEval:
return ProgramType::TessellationEval;
case Maxwell::ShaderProgram::Geometry:
return ProgramType::Geometry;
case Maxwell::ShaderProgram::Fragment:
return ProgramType::Fragment;
}
UNREACHABLE();
return {};
}
/// Calculates the size of a program stream
std::size_t CalculateProgramSize(const GLShader::ProgramCode& program) {
constexpr std::size_t start_offset = 10;
@@ -128,13 +149,13 @@ std::size_t CalculateProgramSize(const GLShader::ProgramCode& program) {
}
/// Hashes one (or two) program streams
u64 GetUniqueIdentifier(Maxwell::ShaderProgram program_type, const ProgramCode& code,
u64 GetUniqueIdentifier(ProgramType program_type, const ProgramCode& code,
const ProgramCode& code_b, std::size_t size_a = 0, std::size_t size_b = 0) {
if (size_a == 0) {
size_a = CalculateProgramSize(code);
}
u64 unique_identifier = Common::CityHash64(reinterpret_cast<const char*>(code.data()), size_a);
if (program_type != Maxwell::ShaderProgram::VertexA) {
if (program_type != ProgramType::VertexA) {
return unique_identifier;
}
// VertexA programs include two programs
@@ -152,12 +173,12 @@ u64 GetUniqueIdentifier(Maxwell::ShaderProgram program_type, const ProgramCode&
}
/// Creates an unspecialized program from code streams
GLShader::ProgramResult CreateProgram(const Device& device, Maxwell::ShaderProgram program_type,
GLShader::ProgramResult CreateProgram(const Device& device, ProgramType program_type,
ProgramCode program_code, ProgramCode program_code_b) {
GLShader::ShaderSetup setup(program_code);
setup.program.size_a = CalculateProgramSize(program_code);
setup.program.size_b = 0;
if (program_type == Maxwell::ShaderProgram::VertexA) {
if (program_type == ProgramType::VertexA) {
// VertexB is always enabled, so when VertexA is enabled, we have two vertex shaders.
// Conventional HW does not support this, so we combine VertexA and VertexB into one
// stage here.
@@ -168,22 +189,23 @@ GLShader::ProgramResult CreateProgram(const Device& device, Maxwell::ShaderProgr
program_type, program_code, program_code_b, setup.program.size_a, setup.program.size_b);
switch (program_type) {
case Maxwell::ShaderProgram::VertexA:
case Maxwell::ShaderProgram::VertexB:
case ProgramType::VertexA:
case ProgramType::VertexB:
return GLShader::GenerateVertexShader(device, setup);
case Maxwell::ShaderProgram::Geometry:
case ProgramType::Geometry:
return GLShader::GenerateGeometryShader(device, setup);
case Maxwell::ShaderProgram::Fragment:
case ProgramType::Fragment:
return GLShader::GenerateFragmentShader(device, setup);
case ProgramType::Compute:
return GLShader::GenerateComputeShader(device, setup);
default:
LOG_CRITICAL(HW_GPU, "Unimplemented program_type={}", static_cast<u32>(program_type));
UNREACHABLE();
UNIMPLEMENTED_MSG("Unimplemented program_type={}", static_cast<u32>(program_type));
return {};
}
}
CachedProgram SpecializeShader(const std::string& code, const GLShader::ShaderEntries& entries,
Maxwell::ShaderProgram program_type, const ProgramVariant& variant,
ProgramType program_type, const ProgramVariant& variant,
bool hint_retrievable = false) {
auto base_bindings{variant.base_bindings};
const auto primitive_mode{variant.primitive_mode};
@@ -194,7 +216,14 @@ CachedProgram SpecializeShader(const std::string& code, const GLShader::ShaderEn
if (entries.shader_viewport_layer_array) {
source += "#extension GL_ARB_shader_viewport_layer_array : enable\n";
}
source += fmt::format("\n#define EMULATION_UBO_BINDING {}\n", base_bindings.cbuf++);
if (program_type == ProgramType::Compute) {
source += "#extension GL_ARB_compute_variable_group_size : require\n";
}
source += '\n';
if (program_type != ProgramType::Compute) {
source += fmt::format("#define EMULATION_UBO_BINDING {}\n", base_bindings.cbuf++);
}
for (const auto& cbuf : entries.const_buffers) {
source +=
@@ -221,13 +250,16 @@ CachedProgram SpecializeShader(const std::string& code, const GLShader::ShaderEn
source += fmt::format("#define SAMPLER_{}_IS_BUFFER", i);
}
if (program_type == Maxwell::ShaderProgram::Geometry) {
if (program_type == ProgramType::Geometry) {
const auto [glsl_topology, debug_name, max_vertices] =
GetPrimitiveDescription(primitive_mode);
source += "layout (" + std::string(glsl_topology) + ") in;\n";
source += "#define MAX_VERTEX_INPUT " + std::to_string(max_vertices) + '\n';
}
if (program_type == ProgramType::Compute) {
source += "layout (local_size_variable) in;\n";
}
source += code;
@@ -255,7 +287,7 @@ std::set<GLenum> GetSupportedFormats() {
} // Anonymous namespace
CachedShader::CachedShader(const ShaderParameters& params, Maxwell::ShaderProgram program_type,
CachedShader::CachedShader(const ShaderParameters& params, ProgramType program_type,
GLShader::ProgramResult result)
: RasterizerCacheObject{params.host_ptr}, host_ptr{params.host_ptr}, cpu_addr{params.cpu_addr},
unique_identifier{params.unique_identifier}, program_type{program_type},
@@ -268,29 +300,50 @@ Shader CachedShader::CreateStageFromMemory(const ShaderParameters& params,
ProgramCode&& program_code_b) {
const auto code_size{CalculateProgramSize(program_code)};
const auto code_size_b{CalculateProgramSize(program_code_b)};
auto result{CreateProgram(params.device, program_type, program_code, program_code_b)};
auto result{
CreateProgram(params.device, GetProgramType(program_type), program_code, program_code_b)};
if (result.first.empty()) {
// TODO(Rodrigo): Unimplemented shader stages hit here, avoid using these for now
return {};
}
params.disk_cache.SaveRaw(ShaderDiskCacheRaw(
params.unique_identifier, program_type, static_cast<u32>(code_size / sizeof(u64)),
static_cast<u32>(code_size_b / sizeof(u64)), std::move(program_code),
std::move(program_code_b)));
params.unique_identifier, GetProgramType(program_type),
static_cast<u32>(code_size / sizeof(u64)), static_cast<u32>(code_size_b / sizeof(u64)),
std::move(program_code), std::move(program_code_b)));
return std::shared_ptr<CachedShader>(new CachedShader(params, program_type, std::move(result)));
return std::shared_ptr<CachedShader>(
new CachedShader(params, GetProgramType(program_type), std::move(result)));
}
Shader CachedShader::CreateStageFromCache(const ShaderParameters& params,
Maxwell::ShaderProgram program_type,
GLShader::ProgramResult result) {
return std::shared_ptr<CachedShader>(new CachedShader(params, program_type, std::move(result)));
return std::shared_ptr<CachedShader>(
new CachedShader(params, GetProgramType(program_type), std::move(result)));
}
Shader CachedShader::CreateKernelFromMemory(const ShaderParameters& params, ProgramCode&& code) {
auto result{CreateProgram(params.device, ProgramType::Compute, code, {})};
const auto code_size{CalculateProgramSize(code)};
params.disk_cache.SaveRaw(ShaderDiskCacheRaw(params.unique_identifier, ProgramType::Compute,
static_cast<u32>(code_size / sizeof(u64)), 0,
std::move(code), {}));
return std::shared_ptr<CachedShader>(
new CachedShader(params, ProgramType::Compute, std::move(result)));
}
Shader CachedShader::CreateKernelFromCache(const ShaderParameters& params,
GLShader::ProgramResult result) {
return std::shared_ptr<CachedShader>(
new CachedShader(params, ProgramType::Compute, std::move(result)));
}
std::tuple<GLuint, BaseBindings> CachedShader::GetProgramHandle(const ProgramVariant& variant) {
GLuint handle{};
if (program_type == Maxwell::ShaderProgram::Geometry) {
if (program_type == ProgramType::Geometry) {
handle = GetGeometryShader(variant);
} else {
const auto [entry, is_cache_miss] = programs.try_emplace(variant);
@@ -308,8 +361,11 @@ std::tuple<GLuint, BaseBindings> CachedShader::GetProgramHandle(const ProgramVar
handle = program->handle;
}
auto base_bindings{variant.base_bindings};
base_bindings.cbuf += static_cast<u32>(entries.const_buffers.size()) + RESERVED_UBOS;
auto base_bindings = variant.base_bindings;
base_bindings.cbuf += static_cast<u32>(entries.const_buffers.size());
if (program_type != ProgramType::Compute) {
base_bindings.cbuf += STAGE_RESERVED_UBOS;
}
base_bindings.gmem += static_cast<u32>(entries.global_memory_entries.size());
base_bindings.sampler += static_cast<u32>(entries.samplers.size());
@@ -589,13 +645,15 @@ Shader ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) {
// No shader found - create a new one
ProgramCode program_code{GetShaderCode(memory_manager, program_addr, host_ptr)};
ProgramCode program_code_b;
if (program == Maxwell::ShaderProgram::VertexA) {
const bool is_program_a{program == Maxwell::ShaderProgram::VertexA};
if (is_program_a) {
const GPUVAddr program_addr_b{GetShaderAddress(system, Maxwell::ShaderProgram::VertexB)};
program_code_b = GetShaderCode(memory_manager, program_addr_b,
memory_manager.GetPointer(program_addr_b));
}
const auto unique_identifier = GetUniqueIdentifier(program, program_code, program_code_b);
const auto unique_identifier =
GetUniqueIdentifier(GetProgramType(program), program_code, program_code_b);
const auto cpu_addr{*memory_manager.GpuToCpuAddress(program_addr)};
const ShaderParameters params{disk_cache, precompiled_programs, device, cpu_addr,
host_ptr, unique_identifier};
@@ -612,4 +670,30 @@ Shader ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) {
return last_shaders[static_cast<std::size_t>(program)] = shader;
}
Shader ShaderCacheOpenGL::GetComputeKernel(GPUVAddr code_addr) {
auto& memory_manager{system.GPU().MemoryManager()};
const auto host_ptr{memory_manager.GetPointer(code_addr)};
auto kernel = TryGet(host_ptr);
if (kernel) {
return kernel;
}
// No kernel found - create a new one
auto code{GetShaderCode(memory_manager, code_addr, host_ptr)};
const auto unique_identifier{GetUniqueIdentifier(ProgramType::Compute, code, {})};
const auto cpu_addr{*memory_manager.GpuToCpuAddress(code_addr)};
const ShaderParameters params{disk_cache, precompiled_programs, device, cpu_addr,
host_ptr, unique_identifier};
const auto found = precompiled_shaders.find(unique_identifier);
if (found == precompiled_shaders.end()) {
kernel = CachedShader::CreateKernelFromMemory(params, std::move(code));
} else {
kernel = CachedShader::CreateKernelFromCache(params, found->second);
}
Register(kernel);
return kernel;
}
} // namespace OpenGL

12
src/video_core/renderer_opengl/gl_shader_cache.h
View File

@@ -61,6 +61,11 @@ public:
Maxwell::ShaderProgram program_type,
GLShader::ProgramResult result);
static Shader CreateKernelFromMemory(const ShaderParameters& params, ProgramCode&& code);
static Shader CreateKernelFromCache(const ShaderParameters& params,
GLShader::ProgramResult result);
VAddr GetCpuAddr() const override {
return cpu_addr;
}
@@ -78,7 +83,7 @@ public:
std::tuple<GLuint, BaseBindings> GetProgramHandle(const ProgramVariant& variant);
private:
explicit CachedShader(const ShaderParameters& params, Maxwell::ShaderProgram program_type,
explicit CachedShader(const ShaderParameters& params, ProgramType program_type,
GLShader::ProgramResult result);
// Geometry programs. These are needed because GLSL needs an input topology but it's not
@@ -104,7 +109,7 @@ private:
u8* host_ptr{};
VAddr cpu_addr{};
u64 unique_identifier{};
Maxwell::ShaderProgram program_type{};
ProgramType program_type{};
ShaderDiskCacheOpenGL& disk_cache;
const PrecompiledPrograms& precompiled_programs;
@@ -132,6 +137,9 @@ public:
/// Gets the current specified shader stage program
Shader GetStageProgram(Maxwell::ShaderProgram program);
/// Gets a compute kernel in the passed address
Shader GetComputeKernel(GPUVAddr code_addr);
protected:
// We do not have to flush this cache as things in it are never modified by us.
void FlushObjectInner(const Shader& object) override {}

74
src/video_core/renderer_opengl/gl_shader_decompiler.cpp
View File

@@ -37,7 +37,6 @@ using namespace std::string_literals;
using namespace VideoCommon::Shader;
using Maxwell = Tegra::Engines::Maxwell3D::Regs;
using ShaderStage = Tegra::Engines::Maxwell3D::Regs::ShaderStage;
using Operation = const OperationNode&;
enum class Type { Bool, Bool2, Float, Int, Uint, HalfFloat };
@@ -162,9 +161,13 @@ std::string FlowStackTopName(MetaStackClass stack) {
return fmt::format("{}_flow_stack_top", GetFlowStackPrefix(stack));
}
constexpr bool IsVertexShader(ProgramType stage) {
return stage == ProgramType::VertexA || stage == ProgramType::VertexB;
}
class GLSLDecompiler final {
public:
explicit GLSLDecompiler(const Device& device, const ShaderIR& ir, ShaderStage stage,
explicit GLSLDecompiler(const Device& device, const ShaderIR& ir, ProgramType stage,
std::string suffix)
: device{device}, ir{ir}, stage{stage}, suffix{suffix}, header{ir.GetHeader()} {}
@@ -248,21 +251,21 @@ public:
}
entries.clip_distances = ir.GetClipDistances();
entries.shader_viewport_layer_array =
stage == ShaderStage::Vertex && (ir.UsesLayer() || ir.UsesViewportIndex());
IsVertexShader(stage) && (ir.UsesLayer() || ir.UsesViewportIndex());
entries.shader_length = ir.GetLength();
return entries;
}
private:
void DeclareVertex() {
if (stage != ShaderStage::Vertex)
if (!IsVertexShader(stage))
return;
DeclareVertexRedeclarations();
}
void DeclareGeometry() {
if (stage != ShaderStage::Geometry) {
if (stage != ProgramType::Geometry) {
return;
}
@@ -293,14 +296,14 @@ private:
break;
}
}
if (stage != ShaderStage::Vertex || device.HasVertexViewportLayer()) {
if (!IsVertexShader(stage) || device.HasVertexViewportLayer()) {
if (ir.UsesLayer()) {
code.AddLine("int gl_Layer;");
}
if (ir.UsesViewportIndex()) {
code.AddLine("int gl_ViewportIndex;");
}
} else if ((ir.UsesLayer() || ir.UsesViewportIndex()) && stage == ShaderStage::Vertex &&
} else if ((ir.UsesLayer() || ir.UsesViewportIndex()) && IsVertexShader(stage) &&
!device.HasVertexViewportLayer()) {
LOG_ERROR(
Render_OpenGL,
@@ -337,11 +340,16 @@ private:
}
void DeclareLocalMemory() {
if (const u64 local_memory_size = header.GetLocalMemorySize(); local_memory_size > 0) {
const auto element_count = Common::AlignUp(local_memory_size, 4) / 4;
code.AddLine("float {}[{}];", GetLocalMemory(), element_count);
code.AddNewLine();
// TODO(Rodrigo): Unstub kernel local memory size and pass it from a register at
// specialization time.
const u64 local_memory_size =
stage == ProgramType::Compute ? 0x400 : header.GetLocalMemorySize();
if (local_memory_size == 0) {
return;
}
const auto element_count = Common::AlignUp(local_memory_size, 4) / 4;
code.AddLine("float {}[{}];", GetLocalMemory(), element_count);
code.AddNewLine();
}
void DeclareInternalFlags() {
@@ -395,12 +403,12 @@ private:
const u32 location{GetGenericAttributeIndex(index)};
std::string name{GetInputAttribute(index)};
if (stage == ShaderStage::Geometry) {
if (stage == ProgramType::Geometry) {
name = "gs_" + name + "[]";
}
std::string suffix;
if (stage == ShaderStage::Fragment) {
if (stage == ProgramType::Fragment) {
const auto input_mode{header.ps.GetAttributeUse(location)};
if (skip_unused && input_mode == AttributeUse::Unused) {
return;
@@ -412,7 +420,7 @@ private:
}
void DeclareOutputAttributes() {
if (ir.HasPhysicalAttributes() && stage != ShaderStage::Fragment) {
if (ir.HasPhysicalAttributes() && stage != ProgramType::Fragment) {
for (u32 i = 0; i < GetNumPhysicalVaryings(); ++i) {
DeclareOutputAttribute(ToGenericAttribute(i));
}
@@ -534,7 +542,7 @@ private:
constexpr u32 element_stride{4};
const u32 address{generic_base + index * generic_stride + element * element_stride};
const bool declared{stage != ShaderStage::Fragment ||
const bool declared{stage != ProgramType::Fragment ||
header.ps.GetAttributeUse(index) != AttributeUse::Unused};
const std::string value{declared ? ReadAttribute(attribute, element) : "0"};
code.AddLine("case 0x{:x}: return {};", address, value);
@@ -638,7 +646,7 @@ private:
}
if (const auto abuf = std::get_if<AbufNode>(&*node)) {
UNIMPLEMENTED_IF_MSG(abuf->IsPhysicalBuffer() && stage == ShaderStage::Geometry,
UNIMPLEMENTED_IF_MSG(abuf->IsPhysicalBuffer() && stage == ProgramType::Geometry,
"Physical attributes in geometry shaders are not implemented");
if (abuf->IsPhysicalBuffer()) {
return fmt::format("readPhysicalAttribute(ftou({}))",
@@ -693,6 +701,9 @@ private:
}
if (const auto lmem = std::get_if<LmemNode>(&*node)) {
if (stage == ProgramType::Compute) {
LOG_WARNING(Render_OpenGL, "Local memory is stubbed on compute shaders");
}
return fmt::format("{}[ftou({}) / 4]", GetLocalMemory(), Visit(lmem->GetAddress()));
}
@@ -722,7 +733,7 @@ private:
std::string ReadAttribute(Attribute::Index attribute, u32 element, const Node& buffer = {}) {
const auto GeometryPass = [&](std::string_view name) {
if (stage == ShaderStage::Geometry && buffer) {
if (stage == ProgramType::Geometry && buffer) {
// 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.
@@ -734,10 +745,10 @@ private:
switch (attribute) {
case Attribute::Index::Position:
switch (stage) {
case ShaderStage::Geometry:
case ProgramType::Geometry:
return fmt::format("gl_in[ftou({})].gl_Position{}", Visit(buffer),
GetSwizzle(element));
case ShaderStage::Fragment:
case ProgramType::Fragment:
return element == 3 ? "1.0f" : ("gl_FragCoord"s + GetSwizzle(element));
default:
UNREACHABLE();
@@ -758,7 +769,7 @@ private:
// TODO(Subv): Find out what the values are for the first two elements when inside a
// vertex shader, and what's the value of the fourth element when inside a Tess Eval
// shader.
ASSERT(stage == ShaderStage::Vertex);
ASSERT(IsVertexShader(stage));
switch (element) {
case 2:
// Config pack's first value is instance_id.
@@ -770,7 +781,7 @@ private:
return "0";
case Attribute::Index::FrontFacing:
// TODO(Subv): Find out what the values are for the other elements.
ASSERT(stage == ShaderStage::Fragment);
ASSERT(stage == ProgramType::Fragment);
switch (element) {
case 3:
return "itof(gl_FrontFacing ? -1 : 0)";
@@ -792,7 +803,7 @@ private:
return value;
}
// There's a bug in NVidia's proprietary drivers that makes precise fail on fragment shaders
const std::string precise = stage != ShaderStage::Fragment ? "precise " : "";
const std::string precise = stage != ProgramType::Fragment ? "precise " : "";
const std::string temporary = code.GenerateTemporary();
code.AddLine("{}float {} = {};", precise, temporary, value);
@@ -827,12 +838,12 @@ private:
UNIMPLEMENTED();
return {};
case 1:
if (stage == ShaderStage::Vertex && !device.HasVertexViewportLayer()) {
if (IsVertexShader(stage) && !device.HasVertexViewportLayer()) {
return {};
}
return std::make_pair("gl_Layer", true);
case 2:
if (stage == ShaderStage::Vertex && !device.HasVertexViewportLayer()) {
if (IsVertexShader(stage) && !device.HasVertexViewportLayer()) {
return {};
}
return std::make_pair("gl_ViewportIndex", true);
@@ -1069,6 +1080,9 @@ private:
target = result->first;
is_integer = result->second;
} else if (const auto lmem = std::get_if<LmemNode>(&*dest)) {
if (stage == ProgramType::Compute) {
LOG_WARNING(Render_OpenGL, "Local memory is stubbed on compute shaders");
}
target = fmt::format("{}[ftou({}) / 4]", GetLocalMemory(), Visit(lmem->GetAddress()));
} else if (const auto gmem = std::get_if<GmemNode>(&*dest)) {
const std::string real = Visit(gmem->GetRealAddress());
@@ -1622,7 +1636,7 @@ private:
}
std::string Exit(Operation operation) {
if (stage != ShaderStage::Fragment) {
if (stage != ProgramType::Fragment) {
code.AddLine("return;");
return {};
}
@@ -1673,7 +1687,7 @@ private:
}
std::string EmitVertex(Operation operation) {
ASSERT_MSG(stage == ShaderStage::Geometry,
ASSERT_MSG(stage == ProgramType::Geometry,
"EmitVertex is expected to be used in a geometry shader.");
// If a geometry shader is attached, it will always flip (it's the last stage before
@@ -1684,7 +1698,7 @@ private:
}
std::string EndPrimitive(Operation operation) {
ASSERT_MSG(stage == ShaderStage::Geometry,
ASSERT_MSG(stage == ProgramType::Geometry,
"EndPrimitive is expected to be used in a geometry shader.");
code.AddLine("EndPrimitive();");
@@ -1919,7 +1933,7 @@ private:
}
u32 GetNumPhysicalInputAttributes() const {
return stage == ShaderStage::Vertex ? GetNumPhysicalAttributes() : GetNumPhysicalVaryings();
return IsVertexShader(stage) ? GetNumPhysicalAttributes() : GetNumPhysicalVaryings();
}
u32 GetNumPhysicalAttributes() const {
@@ -1932,7 +1946,7 @@ private:
const Device& device;
const ShaderIR& ir;
const ShaderStage stage;
const ProgramType stage;
const std::string suffix;
const Header header;
@@ -1963,7 +1977,7 @@ std::string GetCommonDeclarations() {
MAX_CONSTBUFFER_ELEMENTS);
}
ProgramResult Decompile(const Device& device, const ShaderIR& ir, Maxwell::ShaderStage stage,
ProgramResult Decompile(const Device& device, const ShaderIR& ir, ProgramType stage,
const std::string& suffix) {
GLSLDecompiler decompiler(device, ir, stage, suffix);
decompiler.Decompile();

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

@@ -12,14 +12,26 @@
#include "video_core/engines/maxwell_3d.h"
#include "video_core/shader/shader_ir.h"
namespace OpenGL {
class Device;
}
namespace VideoCommon::Shader {
class ShaderIR;
}
namespace OpenGL {
class Device;
enum class ProgramType : u32 {
VertexA = 0,
VertexB = 1,
TessellationControl = 2,
TessellationEval = 3,
Geometry = 4,
Fragment = 5,
Compute = 6
};
} // namespace OpenGL
namespace OpenGL::GLShader {
struct ShaderEntries;
@@ -85,6 +97,6 @@ struct ShaderEntries {
std::string GetCommonDeclarations();
ProgramResult Decompile(const Device& device, const VideoCommon::Shader::ShaderIR& ir,
Maxwell::ShaderStage stage, const std::string& suffix);
ProgramType stage, const std::string& suffix);
} // namespace OpenGL::GLShader

2
src/video_core/renderer_opengl/gl_shader_disk_cache.cpp
View File

@@ -51,7 +51,7 @@ ShaderCacheVersionHash GetShaderCacheVersionHash() {
} // namespace
ShaderDiskCacheRaw::ShaderDiskCacheRaw(u64 unique_identifier, Maxwell::ShaderProgram program_type,
ShaderDiskCacheRaw::ShaderDiskCacheRaw(u64 unique_identifier, ProgramType program_type,
u32 program_code_size, u32 program_code_size_b,
ProgramCode program_code, ProgramCode program_code_b)
: unique_identifier{unique_identifier}, program_type{program_type},

33
src/video_core/renderer_opengl/gl_shader_disk_cache.h
View File

@@ -18,7 +18,6 @@
#include "common/assert.h"
#include "common/common_types.h"
#include "core/file_sys/vfs_vector.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/renderer_opengl/gl_shader_gen.h"
namespace Core {
@@ -34,14 +33,11 @@ namespace OpenGL {
struct ShaderDiskCacheUsage;
struct ShaderDiskCacheDump;
using ShaderDumpsMap = std::unordered_map<ShaderDiskCacheUsage, ShaderDiskCacheDump>;
using ProgramCode = std::vector<u64>;
using Maxwell = Tegra::Engines::Maxwell3D::Regs;
using ShaderDumpsMap = std::unordered_map<ShaderDiskCacheUsage, ShaderDiskCacheDump>;
using TextureBufferUsage = std::bitset<64>;
/// Allocated bindings used by an OpenGL shader program.
/// Allocated bindings used by an OpenGL shader program
struct BaseBindings {
u32 cbuf{};
u32 gmem{};
@@ -126,7 +122,7 @@ namespace OpenGL {
/// Describes a shader how it's used by the guest GPU
class ShaderDiskCacheRaw {
public:
explicit ShaderDiskCacheRaw(u64 unique_identifier, Maxwell::ShaderProgram program_type,
explicit ShaderDiskCacheRaw(u64 unique_identifier, ProgramType program_type,
u32 program_code_size, u32 program_code_size_b,
ProgramCode program_code, ProgramCode program_code_b);
ShaderDiskCacheRaw();
@@ -141,30 +137,13 @@ public:
}
bool HasProgramA() const {
return program_type == Maxwell::ShaderProgram::VertexA;
return program_type == ProgramType::VertexA;
}
Maxwell::ShaderProgram GetProgramType() const {
ProgramType GetProgramType() const {
return program_type;
}
Maxwell::ShaderStage GetProgramStage() const {
switch (program_type) {
case Maxwell::ShaderProgram::VertexA:
case Maxwell::ShaderProgram::VertexB:
return Maxwell::ShaderStage::Vertex;
case Maxwell::ShaderProgram::TesselationControl:
return Maxwell::ShaderStage::TesselationControl;
case Maxwell::ShaderProgram::TesselationEval:
return Maxwell::ShaderStage::TesselationEval;
case Maxwell::ShaderProgram::Geometry:
return Maxwell::ShaderStage::Geometry;
case Maxwell::ShaderProgram::Fragment:
return Maxwell::ShaderStage::Fragment;
}
UNREACHABLE();
}
const ProgramCode& GetProgramCode() const {
return program_code;
}
@@ -175,7 +154,7 @@ public:
private:
u64 unique_identifier{};
Maxwell::ShaderProgram program_type{};
ProgramType program_type{};
u32 program_code_size{};
u32 program_code_size_b{};

39
src/video_core/renderer_opengl/gl_shader_gen.cpp
View File

@@ -14,7 +14,8 @@ using Tegra::Engines::Maxwell3D;
using VideoCommon::Shader::ProgramCode;
using VideoCommon::Shader::ShaderIR;
static constexpr u32 PROGRAM_OFFSET{10};
static constexpr u32 PROGRAM_OFFSET = 10;
static constexpr u32 COMPUTE_OFFSET = 0;
ProgramResult GenerateVertexShader(const Device& device, const ShaderSetup& setup) {
const std::string id = fmt::format("{:016x}", setup.program.unique_identifier);
@@ -29,17 +30,15 @@ layout (std140, binding = EMULATION_UBO_BINDING) uniform vs_config {
};
)";
const ShaderIR program_ir(setup.program.code, PROGRAM_OFFSET, setup.program.size_a);
ProgramResult program =
Decompile(device, program_ir, Maxwell3D::Regs::ShaderStage::Vertex, "vertex");
const ShaderIR program_ir(setup.program.code, PROGRAM_OFFSET, setup.program.size_a);
const auto stage = setup.IsDualProgram() ? ProgramType::VertexA : ProgramType::VertexB;
ProgramResult program = Decompile(device, program_ir, stage, "vertex");
out += program.first;
if (setup.IsDualProgram()) {
const ShaderIR program_ir_b(setup.program.code_b, PROGRAM_OFFSET, setup.program.size_b);
ProgramResult program_b =
Decompile(device, program_ir_b, Maxwell3D::Regs::ShaderStage::Vertex, "vertex_b");
ProgramResult program_b = Decompile(device, program_ir_b, ProgramType::VertexB, "vertex_b");
out += program_b.first;
}
@@ -80,9 +79,9 @@ layout (std140, binding = EMULATION_UBO_BINDING) uniform gs_config {
};
)";
const ShaderIR program_ir(setup.program.code, PROGRAM_OFFSET, setup.program.size_a);
ProgramResult program =
Decompile(device, program_ir, Maxwell3D::Regs::ShaderStage::Geometry, "geometry");
ProgramResult program = Decompile(device, program_ir, ProgramType::Geometry, "geometry");
out += program.first;
out += R"(
@@ -116,9 +115,7 @@ layout (std140, binding = EMULATION_UBO_BINDING) uniform fs_config {
)";
const ShaderIR program_ir(setup.program.code, PROGRAM_OFFSET, setup.program.size_a);
ProgramResult program =
Decompile(device, program_ir, Maxwell3D::Regs::ShaderStage::Fragment, "fragment");
ProgramResult program = Decompile(device, program_ir, ProgramType::Fragment, "fragment");
out += program.first;
out += R"(
@@ -130,4 +127,22 @@ void main() {
return {std::move(out), std::move(program.second)};
}
ProgramResult GenerateComputeShader(const Device& device, const ShaderSetup& setup) {
const std::string id = fmt::format("{:016x}", setup.program.unique_identifier);
std::string out = "// Shader Unique Id: CS" + id + "\n\n";
out += GetCommonDeclarations();
const ShaderIR program_ir(setup.program.code, COMPUTE_OFFSET, setup.program.size_a);
ProgramResult program = Decompile(device, program_ir, ProgramType::Compute, "compute");
out += program.first;
out += R"(
void main() {
execute_compute();
}
)";
return {std::move(out), std::move(program.second)};
}
} // namespace OpenGL::GLShader

3
src/video_core/renderer_opengl/gl_shader_gen.h
View File

@@ -54,4 +54,7 @@ ProgramResult GenerateGeometryShader(const Device& device, const ShaderSetup& se
/// Generates the GLSL fragment shader program source code for the given FS program
ProgramResult GenerateFragmentShader(const Device& device, const ShaderSetup& setup);
/// Generates the GLSL compute shader program source code for the given CS program
ProgramResult GenerateComputeShader(const Device& device, const ShaderSetup& setup);
} // namespace OpenGL::GLShader

24
src/video_core/renderer_opengl/gl_shader_util.cpp
View File

@@ -10,21 +10,25 @@
namespace OpenGL::GLShader {
GLuint LoadShader(const char* source, GLenum type) {
const char* debug_type;
namespace {
const char* GetStageDebugName(GLenum type) {
switch (type) {
case GL_VERTEX_SHADER:
debug_type = "vertex";
break;
return "vertex";
case GL_GEOMETRY_SHADER:
debug_type = "geometry";
break;
return "geometry";
case GL_FRAGMENT_SHADER:
debug_type = "fragment";
break;
default:
UNREACHABLE();
return "fragment";
case GL_COMPUTE_SHADER:
return "compute";
}
UNIMPLEMENTED();
return "unknown";
}
} // Anonymous namespace
GLuint LoadShader(const char* source, GLenum type) {
const char* debug_type = GetStageDebugName(type);
const GLuint shader_id = glCreateShader(type);
glShaderSource(shader_id, 1, &source, nullptr);
LOG_DEBUG(Render_OpenGL, "Compiling {} shader...", debug_type);