N-archive/suyu
1
0
Fork 0
mirror of https://git.suyu.dev/suyu/suyu.git synced 2024-11-25 08:12:44 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions

shader: Remove old shader management

Browse source
This commit is contained in:
ReinUsesLisp 2021-02-16 20:52:12 -03:00 committed by ameerj
parent 58914796c0
commit c67d64365a
83 changed files with 57 additions and 19625 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

56
CMakeModules/GenerateSCMRev.cmake
View file

@ -51,61 +51,7 @@ endif()
# The variable SRC_DIR must be passed into the script (since it uses the current build directory for all values of CMAKE_*_DIR)
set(VIDEO_CORE "${SRC_DIR}/src/video_core")
set(HASH_FILES
"${VIDEO_CORE}/renderer_opengl/gl_arb_decompiler.cpp"
"${VIDEO_CORE}/renderer_opengl/gl_arb_decompiler.h"
"${VIDEO_CORE}/renderer_opengl/gl_shader_cache.cpp"
"${VIDEO_CORE}/renderer_opengl/gl_shader_cache.h"
"${VIDEO_CORE}/renderer_opengl/gl_shader_decompiler.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}/shader/decode/arithmetic.cpp"
"${VIDEO_CORE}/shader/decode/arithmetic_half.cpp"
"${VIDEO_CORE}/shader/decode/arithmetic_half_immediate.cpp"
"${VIDEO_CORE}/shader/decode/arithmetic_immediate.cpp"
"${VIDEO_CORE}/shader/decode/arithmetic_integer.cpp"
"${VIDEO_CORE}/shader/decode/arithmetic_integer_immediate.cpp"
"${VIDEO_CORE}/shader/decode/bfe.cpp"
"${VIDEO_CORE}/shader/decode/bfi.cpp"
"${VIDEO_CORE}/shader/decode/conversion.cpp"
"${VIDEO_CORE}/shader/decode/ffma.cpp"
"${VIDEO_CORE}/shader/decode/float_set.cpp"
"${VIDEO_CORE}/shader/decode/float_set_predicate.cpp"
"${VIDEO_CORE}/shader/decode/half_set.cpp"
"${VIDEO_CORE}/shader/decode/half_set_predicate.cpp"
"${VIDEO_CORE}/shader/decode/hfma2.cpp"
"${VIDEO_CORE}/shader/decode/image.cpp"
"${VIDEO_CORE}/shader/decode/integer_set.cpp"
"${VIDEO_CORE}/shader/decode/integer_set_predicate.cpp"
"${VIDEO_CORE}/shader/decode/memory.cpp"
"${VIDEO_CORE}/shader/decode/texture.cpp"
"${VIDEO_CORE}/shader/decode/other.cpp"
"${VIDEO_CORE}/shader/decode/predicate_set_predicate.cpp"
"${VIDEO_CORE}/shader/decode/predicate_set_register.cpp"
"${VIDEO_CORE}/shader/decode/register_set_predicate.cpp"
"${VIDEO_CORE}/shader/decode/shift.cpp"
"${VIDEO_CORE}/shader/decode/video.cpp"
"${VIDEO_CORE}/shader/decode/warp.cpp"
"${VIDEO_CORE}/shader/decode/xmad.cpp"
"${VIDEO_CORE}/shader/ast.cpp"
"${VIDEO_CORE}/shader/ast.h"
"${VIDEO_CORE}/shader/compiler_settings.cpp"
"${VIDEO_CORE}/shader/compiler_settings.h"
"${VIDEO_CORE}/shader/control_flow.cpp"
"${VIDEO_CORE}/shader/control_flow.h"
"${VIDEO_CORE}/shader/decode.cpp"
"${VIDEO_CORE}/shader/expr.cpp"
"${VIDEO_CORE}/shader/expr.h"
"${VIDEO_CORE}/shader/node.h"
"${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"
"${VIDEO_CORE}/shader/transform_feedback.cpp"
"${VIDEO_CORE}/shader/transform_feedback.h"
# ...
)
set(COMBINED "")
foreach (F IN LISTS HASH_FILES)

2
externals/Vulkan-Headers vendored

@ -1 +1 @@
Subproject commit 8188e3fbbc105591064093440f88081fb957d4f0
Subproject commit 07c4a37bcf41ea50aef6e98236abdfe8089fb4c6

2
externals/sirit vendored

@ -1 +1 @@
Subproject commit 200310e8faa756b9869dd6dfc902c255246ac74a
Subproject commit a39596358a3a5488c06554c0c15184a6af71e433

56
src/common/CMakeLists.txt
View file

@ -32,61 +32,7 @@ add_custom_command(OUTPUT scm_rev.cpp
DEPENDS
# WARNING! It was too much work to try and make a common location for this list,
# so if you need to change it, please update CMakeModules/GenerateSCMRev.cmake as well
"${VIDEO_CORE}/renderer_opengl/gl_arb_decompiler.cpp"
"${VIDEO_CORE}/renderer_opengl/gl_arb_decompiler.h"
"${VIDEO_CORE}/renderer_opengl/gl_shader_cache.cpp"
"${VIDEO_CORE}/renderer_opengl/gl_shader_cache.h"
"${VIDEO_CORE}/renderer_opengl/gl_shader_decompiler.cpp"
"${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"
"${VIDEO_CORE}/shader/decode/arithmetic.cpp"
"${VIDEO_CORE}/shader/decode/arithmetic_half.cpp"
"${VIDEO_CORE}/shader/decode/arithmetic_half_immediate.cpp"
"${VIDEO_CORE}/shader/decode/arithmetic_immediate.cpp"
"${VIDEO_CORE}/shader/decode/arithmetic_integer.cpp"
"${VIDEO_CORE}/shader/decode/arithmetic_integer_immediate.cpp"
"${VIDEO_CORE}/shader/decode/bfe.cpp"
"${VIDEO_CORE}/shader/decode/bfi.cpp"
"${VIDEO_CORE}/shader/decode/conversion.cpp"
"${VIDEO_CORE}/shader/decode/ffma.cpp"
"${VIDEO_CORE}/shader/decode/float_set.cpp"
"${VIDEO_CORE}/shader/decode/float_set_predicate.cpp"
"${VIDEO_CORE}/shader/decode/half_set.cpp"
"${VIDEO_CORE}/shader/decode/half_set_predicate.cpp"
"${VIDEO_CORE}/shader/decode/hfma2.cpp"
"${VIDEO_CORE}/shader/decode/image.cpp"
"${VIDEO_CORE}/shader/decode/integer_set.cpp"
"${VIDEO_CORE}/shader/decode/integer_set_predicate.cpp"
"${VIDEO_CORE}/shader/decode/memory.cpp"
"${VIDEO_CORE}/shader/decode/texture.cpp"
"${VIDEO_CORE}/shader/decode/other.cpp"
"${VIDEO_CORE}/shader/decode/predicate_set_predicate.cpp"
"${VIDEO_CORE}/shader/decode/predicate_set_register.cpp"
"${VIDEO_CORE}/shader/decode/register_set_predicate.cpp"
"${VIDEO_CORE}/shader/decode/shift.cpp"
"${VIDEO_CORE}/shader/decode/video.cpp"
"${VIDEO_CORE}/shader/decode/warp.cpp"
"${VIDEO_CORE}/shader/decode/xmad.cpp"
"${VIDEO_CORE}/shader/ast.cpp"
"${VIDEO_CORE}/shader/ast.h"
"${VIDEO_CORE}/shader/compiler_settings.cpp"
"${VIDEO_CORE}/shader/compiler_settings.h"
"${VIDEO_CORE}/shader/control_flow.cpp"
"${VIDEO_CORE}/shader/control_flow.h"
"${VIDEO_CORE}/shader/decode.cpp"
"${VIDEO_CORE}/shader/expr.cpp"
"${VIDEO_CORE}/shader/expr.h"
"${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"
"${VIDEO_CORE}/shader/track.cpp"
"${VIDEO_CORE}/shader/transform_feedback.cpp"
"${VIDEO_CORE}/shader/transform_feedback.h"
# ...
# and also check that the scm_rev files haven't changed
"${CMAKE_CURRENT_SOURCE_DIR}/scm_rev.cpp.in"
"${CMAKE_CURRENT_SOURCE_DIR}/scm_rev.h"

64
src/video_core/CMakeLists.txt
View file

@ -29,7 +29,6 @@ add_library(video_core STATIC
dirty_flags.h
dma_pusher.cpp
dma_pusher.h
engines/const_buffer_engine_interface.h
engines/const_buffer_info.h
engines/engine_interface.h
engines/engine_upload.cpp
@ -61,8 +60,6 @@ add_library(video_core STATIC
gpu.h
gpu_thread.cpp
gpu_thread.h
guest_driver.cpp
guest_driver.h
memory_manager.cpp
memory_manager.h
query_cache.h
@ -71,8 +68,6 @@ add_library(video_core STATIC
rasterizer_interface.h
renderer_base.cpp
renderer_base.h
renderer_opengl/gl_arb_decompiler.cpp
renderer_opengl/gl_arb_decompiler.h
renderer_opengl/gl_buffer_cache.cpp
renderer_opengl/gl_buffer_cache.h
renderer_opengl/gl_device.cpp
@ -85,10 +80,6 @@ add_library(video_core STATIC
renderer_opengl/gl_resource_manager.h
renderer_opengl/gl_shader_cache.cpp
renderer_opengl/gl_shader_cache.h
renderer_opengl/gl_shader_decompiler.cpp
renderer_opengl/gl_shader_decompiler.h
renderer_opengl/gl_shader_disk_cache.cpp
renderer_opengl/gl_shader_disk_cache.h
renderer_opengl/gl_shader_manager.cpp
renderer_opengl/gl_shader_manager.h
renderer_opengl/gl_shader_util.cpp
@ -128,8 +119,6 @@ add_library(video_core STATIC
renderer_vulkan/vk_descriptor_pool.h
renderer_vulkan/vk_fence_manager.cpp
renderer_vulkan/vk_fence_manager.h
renderer_vulkan/vk_graphics_pipeline.cpp
renderer_vulkan/vk_graphics_pipeline.h
renderer_vulkan/vk_master_semaphore.cpp
renderer_vulkan/vk_master_semaphore.h
renderer_vulkan/vk_pipeline_cache.cpp
@ -142,8 +131,6 @@ add_library(video_core STATIC
renderer_vulkan/vk_resource_pool.h
renderer_vulkan/vk_scheduler.cpp
renderer_vulkan/vk_scheduler.h
renderer_vulkan/vk_shader_decompiler.cpp
renderer_vulkan/vk_shader_decompiler.h
renderer_vulkan/vk_shader_util.cpp
renderer_vulkan/vk_shader_util.h
renderer_vulkan/vk_staging_buffer_pool.cpp
@ -159,57 +146,6 @@ add_library(video_core STATIC
shader_cache.h
shader_notify.cpp
shader_notify.h
shader/decode/arithmetic.cpp
shader/decode/arithmetic_immediate.cpp
shader/decode/bfe.cpp
shader/decode/bfi.cpp
shader/decode/shift.cpp
shader/decode/arithmetic_integer.cpp
shader/decode/arithmetic_integer_immediate.cpp
shader/decode/arithmetic_half.cpp
shader/decode/arithmetic_half_immediate.cpp
shader/decode/ffma.cpp
shader/decode/hfma2.cpp
shader/decode/conversion.cpp
shader/decode/memory.cpp
shader/decode/texture.cpp
shader/decode/image.cpp
shader/decode/float_set_predicate.cpp
shader/decode/integer_set_predicate.cpp
shader/decode/half_set_predicate.cpp
shader/decode/predicate_set_register.cpp
shader/decode/predicate_set_predicate.cpp
shader/decode/register_set_predicate.cpp
shader/decode/float_set.cpp
shader/decode/integer_set.cpp
shader/decode/half_set.cpp
shader/decode/video.cpp
shader/decode/warp.cpp
shader/decode/xmad.cpp
shader/decode/other.cpp
shader/ast.cpp
shader/ast.h
shader/async_shaders.cpp
shader/async_shaders.h
shader/compiler_settings.cpp
shader/compiler_settings.h
shader/control_flow.cpp
shader/control_flow.h
shader/decode.cpp
shader/expr.cpp
shader/expr.h
shader/memory_util.cpp
shader/memory_util.h
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
shader/transform_feedback.cpp
shader/transform_feedback.h
surface.cpp
surface.h
texture_cache/accelerated_swizzle.cpp

103
src/video_core/engines/const_buffer_engine_interface.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 <type_traits>
#include "common/bit_field.h"
#include "common/common_types.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/engines/shader_type.h"
#include "video_core/guest_driver.h"
#include "video_core/textures/texture.h"
namespace Tegra::Engines {
struct SamplerDescriptor {
union {
u32 raw = 0;
BitField<0, 2, Tegra::Shader::TextureType> texture_type;
BitField<2, 3, Tegra::Texture::ComponentType> r_type;
BitField<5, 1, u32> is_array;
BitField<6, 1, u32> is_buffer;
BitField<7, 1, u32> is_shadow;
BitField<8, 3, Tegra::Texture::ComponentType> g_type;
BitField<11, 3, Tegra::Texture::ComponentType> b_type;
BitField<14, 3, Tegra::Texture::ComponentType> a_type;
BitField<17, 7, Tegra::Texture::TextureFormat> format;
};
bool operator==(const SamplerDescriptor& rhs) const noexcept {
return raw == rhs.raw;
}
bool operator!=(const SamplerDescriptor& rhs) const noexcept {
return !operator==(rhs);
}
static SamplerDescriptor FromTIC(const Tegra::Texture::TICEntry& tic) {
using Tegra::Shader::TextureType;
SamplerDescriptor result;
result.format.Assign(tic.format.Value());
result.r_type.Assign(tic.r_type.Value());
result.g_type.Assign(tic.g_type.Value());
result.b_type.Assign(tic.b_type.Value());
result.a_type.Assign(tic.a_type.Value());
switch (tic.texture_type.Value()) {
case Tegra::Texture::TextureType::Texture1D:
result.texture_type.Assign(TextureType::Texture1D);
return result;
case Tegra::Texture::TextureType::Texture2D:
result.texture_type.Assign(TextureType::Texture2D);
return result;
case Tegra::Texture::TextureType::Texture3D:
result.texture_type.Assign(TextureType::Texture3D);
return result;
case Tegra::Texture::TextureType::TextureCubemap:
result.texture_type.Assign(TextureType::TextureCube);
return result;
case Tegra::Texture::TextureType::Texture1DArray:
result.texture_type.Assign(TextureType::Texture1D);
result.is_array.Assign(1);
return result;
case Tegra::Texture::TextureType::Texture2DArray:
result.texture_type.Assign(TextureType::Texture2D);
result.is_array.Assign(1);
return result;
case Tegra::Texture::TextureType::Texture1DBuffer:
result.texture_type.Assign(TextureType::Texture1D);
result.is_buffer.Assign(1);
return result;
case Tegra::Texture::TextureType::Texture2DNoMipmap:
result.texture_type.Assign(TextureType::Texture2D);
return result;
case Tegra::Texture::TextureType::TextureCubeArray:
result.texture_type.Assign(TextureType::TextureCube);
result.is_array.Assign(1);
return result;
default:
result.texture_type.Assign(TextureType::Texture2D);
return result;
}
}
};
static_assert(std::is_trivially_copyable_v<SamplerDescriptor>);
class ConstBufferEngineInterface {
public:
virtual ~ConstBufferEngineInterface() = default;
virtual u32 AccessConstBuffer32(ShaderType stage, u64 const_buffer, u64 offset) const = 0;
virtual SamplerDescriptor AccessBoundSampler(ShaderType stage, u64 offset) const = 0;
virtual SamplerDescriptor AccessBindlessSampler(ShaderType stage, u64 const_buffer,
u64 offset) const = 0;
virtual SamplerDescriptor AccessSampler(u32 handle) const = 0;
virtual u32 GetBoundBuffer() const = 0;
virtual VideoCore::GuestDriverProfile& AccessGuestDriverProfile() = 0;
virtual const VideoCore::GuestDriverProfile& AccessGuestDriverProfile() const = 0;
};
} // namespace Tegra::Engines

44
src/video_core/engines/kepler_compute.cpp
View file

@ -57,53 +57,11 @@ void KeplerCompute::CallMultiMethod(u32 method, const u32* base_start, u32 amoun
}
}
u32 KeplerCompute::AccessConstBuffer32(ShaderType stage, u64 const_buffer, u64 offset) const {
ASSERT(stage == ShaderType::Compute);
const auto& buffer = launch_description.const_buffer_config[const_buffer];
u32 result;
std::memcpy(&result, memory_manager.GetPointer(buffer.Address() + offset), sizeof(u32));
return result;
}
SamplerDescriptor KeplerCompute::AccessBoundSampler(ShaderType stage, u64 offset) const {
return AccessBindlessSampler(stage, regs.tex_cb_index, offset * sizeof(Texture::TextureHandle));
}
SamplerDescriptor KeplerCompute::AccessBindlessSampler(ShaderType stage, u64 const_buffer,
u64 offset) const {
ASSERT(stage == ShaderType::Compute);
const auto& tex_info_buffer = launch_description.const_buffer_config[const_buffer];
const GPUVAddr tex_info_address = tex_info_buffer.Address() + offset;
return AccessSampler(memory_manager.Read<u32>(tex_info_address));
}
SamplerDescriptor KeplerCompute::AccessSampler(u32 handle) const {
const Texture::TextureHandle tex_handle{handle};
const Texture::TICEntry tic = GetTICEntry(tex_handle.tic_id);
const Texture::TSCEntry tsc = GetTSCEntry(tex_handle.tsc_id);
SamplerDescriptor result = SamplerDescriptor::FromTIC(tic);
result.is_shadow.Assign(tsc.depth_compare_enabled.Value());
return result;
}
VideoCore::GuestDriverProfile& KeplerCompute::AccessGuestDriverProfile() {
return rasterizer->AccessGuestDriverProfile();
}
const VideoCore::GuestDriverProfile& KeplerCompute::AccessGuestDriverProfile() const {
return rasterizer->AccessGuestDriverProfile();
}
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_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);
rasterizer->DispatchCompute();
}
Texture::TICEntry KeplerCompute::GetTICEntry(u32 tic_index) const {

20
src/video_core/engines/kepler_compute.h
View file

@ -10,7 +10,6 @@
#include "common/bit_field.h"
#include "common/common_funcs.h"
#include "common/common_types.h"
#include "video_core/engines/const_buffer_engine_interface.h"
#include "video_core/engines/engine_interface.h"
#include "video_core/engines/engine_upload.h"
#include "video_core/engines/shader_type.h"
@ -40,7 +39,7 @@ namespace Tegra::Engines {
#define KEPLER_COMPUTE_REG_INDEX(field_name) \
(offsetof(Tegra::Engines::KeplerCompute::Regs, field_name) / sizeof(u32))
class KeplerCompute final : public ConstBufferEngineInterface, public EngineInterface {
class KeplerCompute final : public EngineInterface {
public:
explicit KeplerCompute(Core::System& system, MemoryManager& memory_manager);
~KeplerCompute();
@ -209,23 +208,6 @@ public:
void CallMultiMethod(u32 method, const u32* base_start, u32 amount,
u32 methods_pending) override;
u32 AccessConstBuffer32(ShaderType stage, u64 const_buffer, u64 offset) const override;
SamplerDescriptor AccessBoundSampler(ShaderType stage, u64 offset) const override;
SamplerDescriptor AccessBindlessSampler(ShaderType stage, u64 const_buffer,
u64 offset) const override;
SamplerDescriptor AccessSampler(u32 handle) const override;
u32 GetBoundBuffer() const override {
return regs.tex_cb_index;
}
VideoCore::GuestDriverProfile& AccessGuestDriverProfile() override;
const VideoCore::GuestDriverProfile& AccessGuestDriverProfile() const override;
private:
void ProcessLaunch();

38
src/video_core/engines/maxwell_3d.cpp
View file

@ -670,42 +670,4 @@ void Maxwell3D::ProcessClearBuffers() {
rasterizer->Clear();
}
u32 Maxwell3D::AccessConstBuffer32(ShaderType stage, u64 const_buffer, u64 offset) const {
ASSERT(stage != ShaderType::Compute);
const auto& shader_stage = state.shader_stages[static_cast<std::size_t>(stage)];
const auto& buffer = shader_stage.const_buffers[const_buffer];
return memory_manager.Read<u32>(buffer.address + offset);
}
SamplerDescriptor Maxwell3D::AccessBoundSampler(ShaderType stage, u64 offset) const {
return AccessBindlessSampler(stage, regs.tex_cb_index, offset * sizeof(Texture::TextureHandle));
}
SamplerDescriptor Maxwell3D::AccessBindlessSampler(ShaderType stage, u64 const_buffer,
u64 offset) const {
ASSERT(stage != ShaderType::Compute);
const auto& shader = state.shader_stages[static_cast<std::size_t>(stage)];
const auto& tex_info_buffer = shader.const_buffers[const_buffer];
const GPUVAddr tex_info_address = tex_info_buffer.address + offset;
return AccessSampler(memory_manager.Read<u32>(tex_info_address));
}
SamplerDescriptor Maxwell3D::AccessSampler(u32 handle) const {
const Texture::TextureHandle tex_handle{handle};
const Texture::TICEntry tic = GetTICEntry(tex_handle.tic_id);
const Texture::TSCEntry tsc = GetTSCEntry(tex_handle.tsc_id);
SamplerDescriptor result = SamplerDescriptor::FromTIC(tic);
result.is_shadow.Assign(tsc.depth_compare_enabled.Value());
return result;
}
VideoCore::GuestDriverProfile& Maxwell3D::AccessGuestDriverProfile() {
return rasterizer->AccessGuestDriverProfile();
}
const VideoCore::GuestDriverProfile& Maxwell3D::AccessGuestDriverProfile() const {
return rasterizer->AccessGuestDriverProfile();
}
} // namespace Tegra::Engines

20
src/video_core/engines/maxwell_3d.h
View file

@ -17,7 +17,6 @@
#include "common/common_funcs.h"
#include "common/common_types.h"
#include "common/math_util.h"
#include "video_core/engines/const_buffer_engine_interface.h"
#include "video_core/engines/const_buffer_info.h"
#include "video_core/engines/engine_interface.h"
#include "video_core/engines/engine_upload.h"
@ -49,7 +48,7 @@ namespace Tegra::Engines {
#define MAXWELL3D_REG_INDEX(field_name) \
(offsetof(Tegra::Engines::Maxwell3D::Regs, field_name) / sizeof(u32))
class Maxwell3D final : public ConstBufferEngineInterface, public EngineInterface {
class Maxwell3D final : public EngineInterface {
public:
explicit Maxwell3D(Core::System& system, MemoryManager& memory_manager);
~Maxwell3D();
@ -1424,23 +1423,6 @@ public:
void FlushMMEInlineDraw();
u32 AccessConstBuffer32(ShaderType stage, u64 const_buffer, u64 offset) const override;
SamplerDescriptor AccessBoundSampler(ShaderType stage, u64 offset) const override;
SamplerDescriptor AccessBindlessSampler(ShaderType stage, u64 const_buffer,
u64 offset) const override;
SamplerDescriptor AccessSampler(u32 handle) const override;
u32 GetBoundBuffer() const override {
return regs.tex_cb_index;
}
VideoCore::GuestDriverProfile& AccessGuestDriverProfile() override;
const VideoCore::GuestDriverProfile& AccessGuestDriverProfile() const override;
bool ShouldExecute() const {
return execute_on;
}

37
src/video_core/guest_driver.cpp
View file

@ -1,37 +0,0 @@
// Copyright 2020 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#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) {
return;
}
const std::size_t size = bound_offsets.size();
if (size < 2) {
return;
}
std::sort(bound_offsets.begin(), bound_offsets.end(), std::less{});
u32 min_val = std::numeric_limits<u32>::max();
for (std::size_t i = 1; i < size; ++i) {
if (bound_offsets[i] == bound_offsets[i - 1]) {
continue;
}
const u32 new_min = bound_offsets[i] - bound_offsets[i - 1];
min_val = std::min(min_val, new_min);
}
if (min_val > 2) {
return;
}
texture_handler_size = min_texture_handler_size * min_val;
}
} // namespace VideoCore

46
src/video_core/guest_driver.h
View file

@ -1,46 +0,0 @@
// Copyright 2020 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <optional>
#include <vector>
#include "common/common_types.h"
namespace VideoCore {
/**
* The GuestDriverProfile class is used to learn about the GPU drivers behavior and collect
* information necessary for impossible to avoid HLE methods like shader tracks as they are
* Entscheidungsproblems.
*/
class GuestDriverProfile {
public:
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.value_or(default_texture_handler_size);
}
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.
static constexpr u32 default_texture_handler_size = 8;
std::optional<u32> texture_handler_size = default_texture_handler_size;
};
} // namespace VideoCore

16
src/video_core/rasterizer_interface.h
View file

@ -11,7 +11,6 @@
#include "common/common_types.h"
#include "video_core/engines/fermi_2d.h"
#include "video_core/gpu.h"
#include "video_core/guest_driver.h"
namespace Tegra {
class MemoryManager;
@ -45,7 +44,7 @@ public:
virtual void Clear() = 0;
/// Dispatches a compute shader invocation
virtual void DispatchCompute(GPUVAddr code_addr) = 0;
virtual void DispatchCompute() = 0;
/// Resets the counter of a query
virtual void ResetCounter(QueryType type) = 0;
@ -136,18 +135,5 @@ public:
/// Initialize disk cached resources for the game being emulated
virtual void LoadDiskResources(u64 title_id, std::stop_token stop_loading,
const DiskResourceLoadCallback& callback) {}
/// Grant access to the Guest Driver Profile for recording/obtaining info on the guest driver.
[[nodiscard]] GuestDriverProfile& AccessGuestDriverProfile() {
return guest_driver_profile;
}
/// Grant access to the Guest Driver Profile for recording/obtaining info on the guest driver.
[[nodiscard]] const GuestDriverProfile& AccessGuestDriverProfile() const {
return guest_driver_profile;
}
private:
GuestDriverProfile guest_driver_profile{};
};
} // namespace VideoCore

2124
src/video_core/renderer_opengl/gl_arb_decompiler.cpp
View file

File diff suppressed because it is too large Load diff

29
src/video_core/renderer_opengl/gl_arb_decompiler.h
View file

@ -1,29 +0,0 @@
// Copyright 2020 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <string>
#include <string_view>
#include "common/common_types.h"
namespace Tegra::Engines {
enum class ShaderType : u32;
}
namespace VideoCommon::Shader {
class ShaderIR;
class Registry;
} // namespace VideoCommon::Shader
namespace OpenGL {
class Device;
std::string DecompileAssemblyShader(const Device& device, const VideoCommon::Shader::ShaderIR& ir,
const VideoCommon::Shader::Registry& registry,
Tegra::Engines::ShaderType stage, std::string_view identifier);
} // namespace OpenGL

314
src/video_core/renderer_opengl/gl_rasterizer.cpp
View file

@ -54,40 +54,6 @@ namespace {
constexpr size_t NUM_SUPPORTED_VERTEX_ATTRIBUTES = 16;
struct TextureHandle {
constexpr TextureHandle(u32 data, bool via_header_index) {
const Tegra::Texture::TextureHandle handle{data};
image = handle.tic_id;
sampler = via_header_index ? image : handle.tsc_id.Value();
}
u32 image;
u32 sampler;
};
template <typename Engine, typename Entry>
TextureHandle GetTextureInfo(const Engine& engine, bool via_header_index, const Entry& entry,
ShaderType shader_type, size_t index = 0) {
if constexpr (std::is_same_v<Entry, SamplerEntry>) {
if (entry.is_separated) {
const u32 buffer_1 = entry.buffer;
const u32 buffer_2 = entry.secondary_buffer;
const u32 offset_1 = entry.offset;
const u32 offset_2 = entry.secondary_offset;
const u32 handle_1 = engine.AccessConstBuffer32(shader_type, buffer_1, offset_1);
const u32 handle_2 = engine.AccessConstBuffer32(shader_type, buffer_2, offset_2);
return TextureHandle(handle_1 | handle_2, via_header_index);
}
}
if (entry.is_bindless) {
const u32 raw = engine.AccessConstBuffer32(shader_type, entry.buffer, entry.offset);
return TextureHandle(raw, via_header_index);
}
const u32 buffer = engine.GetBoundBuffer();
const u64 offset = (entry.offset + index) * sizeof(u32);
return TextureHandle(engine.AccessConstBuffer32(shader_type, buffer, offset), via_header_index);
}
/// Translates hardware transform feedback indices
/// @param location Hardware location
/// @return Pair of ARB_transform_feedback3 token stream first and third arguments
@ -119,44 +85,6 @@ std::pair<GLint, GLint> TransformFeedbackEnum(u8 location) {
void oglEnable(GLenum cap, bool state) {
(state ? glEnable : glDisable)(cap);
}
ImageViewType ImageViewTypeFromEntry(const SamplerEntry& entry) {
if (entry.is_buffer) {
return ImageViewType::Buffer;
}
switch (entry.type) {
case Tegra::Shader::TextureType::Texture1D:
return entry.is_array ? ImageViewType::e1DArray : ImageViewType::e1D;
case Tegra::Shader::TextureType::Texture2D:
return entry.is_array ? ImageViewType::e2DArray : ImageViewType::e2D;
case Tegra::Shader::TextureType::Texture3D:
return ImageViewType::e3D;
case Tegra::Shader::TextureType::TextureCube:
return entry.is_array ? ImageViewType::CubeArray : ImageViewType::Cube;
}
UNREACHABLE();
return ImageViewType::e2D;
}
ImageViewType ImageViewTypeFromEntry(const ImageEntry& entry) {
switch (entry.type) {
case Tegra::Shader::ImageType::Texture1D:
return ImageViewType::e1D;
case Tegra::Shader::ImageType::Texture1DArray:
return ImageViewType::e1DArray;
case Tegra::Shader::ImageType::Texture2D:
return ImageViewType::e2D;
case Tegra::Shader::ImageType::Texture2DArray:
return ImageViewType::e2DArray;
case Tegra::Shader::ImageType::Texture3D:
return ImageViewType::e3D;
case Tegra::Shader::ImageType::TextureBuffer:
return ImageViewType::Buffer;
}
UNREACHABLE();
return ImageViewType::e2D;
}
} // Anonymous namespace
RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& emu_window_, Tegra::GPU& gpu_,
@ -172,12 +100,7 @@ RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& emu_window_, Tegra
buffer_cache(*this, maxwell3d, kepler_compute, gpu_memory, cpu_memory_, buffer_cache_runtime),
shader_cache(*this, emu_window_, gpu, maxwell3d, kepler_compute, gpu_memory, device),
query_cache(*this, maxwell3d, gpu_memory), accelerate_dma(buffer_cache),
fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache),
async_shaders(emu_window_) {
if (device.UseAsynchronousShaders()) {
async_shaders.AllocateWorkers();
}
}
fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache) {}
RasterizerOpenGL::~RasterizerOpenGL() = default;
@ -244,117 +167,8 @@ void RasterizerOpenGL::SyncVertexInstances() {
}
}
void RasterizerOpenGL::SetupShaders(bool is_indexed) {
u32 clip_distances = 0;
std::array<Shader*, Maxwell::MaxShaderStage> shaders{};
image_view_indices.clear();
sampler_handles.clear();
texture_cache.SynchronizeGraphicsDescriptors();
for (std::size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) {
const auto& shader_config = maxwell3d.regs.shader_config[index];
const auto program{static_cast<Maxwell::ShaderProgram>(index)};
// Skip stages that are not enabled
if (!maxwell3d.regs.IsShaderConfigEnabled(index)) {
switch (program) {
case Maxwell::ShaderProgram::Geometry:
program_manager.UseGeometryShader(0);
break;
case Maxwell::ShaderProgram::Fragment:
program_manager.UseFragmentShader(0);
break;
default:
break;
}
continue;
}
// Currently this stages are not supported in the OpenGL backend.
// TODO(Blinkhawk): Port tesselation shaders from Vulkan to OpenGL
if (program == Maxwell::ShaderProgram::TesselationControl ||
program == Maxwell::ShaderProgram::TesselationEval) {
continue;
}
Shader* const shader = shader_cache.GetStageProgram(program, async_shaders);
const GLuint program_handle = shader->IsBuilt() ? shader->GetHandle() : 0;
switch (program) {
case Maxwell::ShaderProgram::VertexA:
case Maxwell::ShaderProgram::VertexB:
program_manager.UseVertexShader(program_handle);
break;
case Maxwell::ShaderProgram::Geometry:
program_manager.UseGeometryShader(program_handle);
break;
case Maxwell::ShaderProgram::Fragment:
program_manager.UseFragmentShader(program_handle);
break;
default:
UNIMPLEMENTED_MSG("Unimplemented shader index={}, enable={}, offset=0x{:08X}", index,
shader_config.enable.Value(), shader_config.offset);
break;
}
// Stage indices are 0 - 5
const size_t stage = index == 0 ? 0 : index - 1;
shaders[stage] = shader;
SetupDrawTextures(shader, stage);
SetupDrawImages(shader, stage);
buffer_cache.SetEnabledUniformBuffers(stage, shader->GetEntries().enabled_uniform_buffers);
buffer_cache.UnbindGraphicsStorageBuffers(stage);
u32 ssbo_index = 0;
for (const auto& buffer : shader->GetEntries().global_memory_entries) {
buffer_cache.BindGraphicsStorageBuffer(stage, ssbo_index, buffer.cbuf_index,
buffer.cbuf_offset, buffer.is_written);
++ssbo_index;
}
// Workaround for Intel drivers.
// 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->GetEntries().clip_distances;
// When VertexA is enabled, we have dual vertex shaders
if (program == Maxwell::ShaderProgram::VertexA) {
// VertexB was combined with VertexA, so we skip the VertexB iteration
++index;
}
}
SyncClipEnabled(clip_distances);
maxwell3d.dirty.flags[Dirty::Shaders] = false;
buffer_cache.UpdateGraphicsBuffers(is_indexed);
const std::span indices_span(image_view_indices.data(), image_view_indices.size());
texture_cache.FillGraphicsImageViews(indices_span, image_view_ids);
buffer_cache.BindHostGeometryBuffers(is_indexed);
size_t image_view_index = 0;
size_t texture_index = 0;
size_t image_index = 0;
for (size_t stage = 0; stage < Maxwell::MaxShaderStage; ++stage) {
const Shader* const shader = shaders[stage];
if (!shader) {
continue;
}
buffer_cache.BindHostStageBuffers(stage);
const auto& base = device.GetBaseBindings(stage);
BindTextures(shader->GetEntries(), base.sampler, base.image, image_view_index,
texture_index, image_index);
}
}
void RasterizerOpenGL::LoadDiskResources(u64 title_id, std::stop_token stop_loading,
const VideoCore::DiskResourceLoadCallback& callback) {
shader_cache.LoadDiskCache(title_id, stop_loading, callback);
}
const VideoCore::DiskResourceLoadCallback& callback) {}
void RasterizerOpenGL::Clear() {
MICROPROFILE_SCOPE(OpenGL_Clears);
@ -434,7 +248,6 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) {
// Setup shaders and their used resources.
std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
SetupShaders(is_indexed);
texture_cache.UpdateRenderTargets(false);
state_tracker.BindFramebuffer(texture_cache.GetFramebuffer()->Handle());
@ -488,27 +301,8 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) {
gpu.TickWork();
}
void RasterizerOpenGL::DispatchCompute(GPUVAddr code_addr) {
Shader* const kernel = shader_cache.GetComputeKernel(code_addr);
std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
BindComputeTextures(kernel);
const auto& entries = kernel->GetEntries();
buffer_cache.SetEnabledComputeUniformBuffers(entries.enabled_uniform_buffers);
buffer_cache.UnbindComputeStorageBuffers();
u32 ssbo_index = 0;
for (const auto& buffer : entries.global_memory_entries) {
buffer_cache.BindComputeStorageBuffer(ssbo_index, buffer.cbuf_index, buffer.cbuf_offset,
buffer.is_written);
++ssbo_index;
}
buffer_cache.UpdateComputeBuffers();
buffer_cache.BindHostComputeBuffers();
const auto& launch_desc = kepler_compute.launch_description;
glDispatchCompute(launch_desc.grid_dim_x, launch_desc.grid_dim_y, launch_desc.grid_dim_z);
++num_queued_commands;
void RasterizerOpenGL::DispatchCompute() {
UNREACHABLE_MSG("Not implemented");
}
void RasterizerOpenGL::ResetCounter(VideoCore::QueryType type) {
@ -726,106 +520,6 @@ bool RasterizerOpenGL::AccelerateDisplay(const Tegra::FramebufferConfig& config,
return true;
}
void RasterizerOpenGL::BindComputeTextures(Shader* kernel) {
image_view_indices.clear();
sampler_handles.clear();
texture_cache.SynchronizeComputeDescriptors();
SetupComputeTextures(kernel);
SetupComputeImages(kernel);
const std::span indices_span(image_view_indices.data(), image_view_indices.size());
texture_cache.FillComputeImageViews(indices_span, image_view_ids);
program_manager.BindCompute(kernel->GetHandle());
size_t image_view_index = 0;
size_t texture_index = 0;
size_t image_index = 0;
BindTextures(kernel->GetEntries(), 0, 0, image_view_index, texture_index, image_index);
}
void RasterizerOpenGL::BindTextures(const ShaderEntries& entries, GLuint base_texture,
GLuint base_image, size_t& image_view_index,
size_t& texture_index, size_t& image_index) {
const GLuint* const samplers = sampler_handles.data() + texture_index;
const GLuint* const textures = texture_handles.data() + texture_index;
const GLuint* const images = image_handles.data() + image_index;
const size_t num_samplers = entries.samplers.size();
for (const auto& sampler : entries.samplers) {
for (size_t i = 0; i < sampler.size; ++i) {
const ImageViewId image_view_id = image_view_ids[image_view_index++];
const ImageView& image_view = texture_cache.GetImageView(image_view_id);
const GLuint handle = image_view.Handle(ImageViewTypeFromEntry(sampler));
texture_handles[texture_index++] = handle;
}
}
const size_t num_images = entries.images.size();
for (size_t unit = 0; unit < num_images; ++unit) {
// TODO: Mark as modified
const ImageViewId image_view_id = image_view_ids[image_view_index++];
const ImageView& image_view = texture_cache.GetImageView(image_view_id);
const GLuint handle = image_view.Handle(ImageViewTypeFromEntry(entries.images[unit]));
image_handles[image_index] = handle;
++image_index;
}
if (num_samplers > 0) {
glBindSamplers(base_texture, static_cast<GLsizei>(num_samplers), samplers);
glBindTextures(base_texture, static_cast<GLsizei>(num_samplers), textures);
}
if (num_images > 0) {
glBindImageTextures(base_image, static_cast<GLsizei>(num_images), images);
}
}
void RasterizerOpenGL::SetupDrawTextures(const Shader* shader, size_t stage_index) {
const bool via_header_index =
maxwell3d.regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex;
for (const auto& entry : shader->GetEntries().samplers) {
const auto shader_type = static_cast<ShaderType>(stage_index);
for (size_t index = 0; index < entry.size; ++index) {
const auto handle =
GetTextureInfo(maxwell3d, via_header_index, entry, shader_type, index);
const Sampler* const sampler = texture_cache.GetGraphicsSampler(handle.sampler);
sampler_handles.push_back(sampler->Handle());
image_view_indices.push_back(handle.image);
}
}
}
void RasterizerOpenGL::SetupComputeTextures(const Shader* kernel) {
const bool via_header_index = kepler_compute.launch_description.linked_tsc;
for (const auto& entry : kernel->GetEntries().samplers) {
for (size_t i = 0; i < entry.size; ++i) {
const auto handle =
GetTextureInfo(kepler_compute, via_header_index, entry, ShaderType::Compute, i);
const Sampler* const sampler = texture_cache.GetComputeSampler(handle.sampler);
sampler_handles.push_back(sampler->Handle());
image_view_indices.push_back(handle.image);
}
}
}
void RasterizerOpenGL::SetupDrawImages(const Shader* shader, size_t stage_index) {
const bool via_header_index =
maxwell3d.regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex;
for (const auto& entry : shader->GetEntries().images) {
const auto shader_type = static_cast<ShaderType>(stage_index);
const auto handle = GetTextureInfo(maxwell3d, via_header_index, entry, shader_type);
image_view_indices.push_back(handle.image);
}
}
void RasterizerOpenGL::SetupComputeImages(const Shader* shader) {
const bool via_header_index = kepler_compute.launch_description.linked_tsc;
for (const auto& entry : shader->GetEntries().images) {
const auto handle =
GetTextureInfo(kepler_compute, via_header_index, entry, ShaderType::Compute);
image_view_indices.push_back(handle.image);
}
}
void RasterizerOpenGL::SyncState() {
SyncViewport();
SyncRasterizeEnable();

33
src/video_core/renderer_opengl/gl_rasterizer.h
View file

@ -28,11 +28,9 @@
#include "video_core/renderer_opengl/gl_query_cache.h"
#include "video_core/renderer_opengl/gl_resource_manager.h"
#include "video_core/renderer_opengl/gl_shader_cache.h"
#include "video_core/renderer_opengl/gl_shader_decompiler.h"
#include "video_core/renderer_opengl/gl_shader_manager.h"
#include "video_core/renderer_opengl/gl_state_tracker.h"
#include "video_core/renderer_opengl/gl_texture_cache.h"
#include "video_core/shader/async_shaders.h"
#include "video_core/textures/texture.h"
namespace Core::Memory {
@ -81,7 +79,7 @@ public:
void Draw(bool is_indexed, bool is_instanced) override;
void Clear() override;
void DispatchCompute(GPUVAddr code_addr) override;
void DispatchCompute() override;
void ResetCounter(VideoCore::QueryType type) override;
void Query(GPUVAddr gpu_addr, VideoCore::QueryType type, std::optional<u64> timestamp) override;
void BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr, u32 size) override;
@ -118,36 +116,11 @@ public:
return num_queued_commands > 0;
}
VideoCommon::Shader::AsyncShaders& GetAsyncShaders() {
return async_shaders;
}
const VideoCommon::Shader::AsyncShaders& GetAsyncShaders() const {
return async_shaders;
}
private:
static constexpr size_t MAX_TEXTURES = 192;
static constexpr size_t MAX_IMAGES = 48;
static constexpr size_t MAX_IMAGE_VIEWS = MAX_TEXTURES + MAX_IMAGES;
void BindComputeTextures(Shader* kernel);
void BindTextures(const ShaderEntries& entries, GLuint base_texture, GLuint base_image,
size_t& image_view_index, size_t& texture_index, size_t& image_index);
/// Configures the current textures to use for the draw command.
void SetupDrawTextures(const Shader* shader, size_t stage_index);
/// Configures the textures used in a compute shader.
void SetupComputeTextures(const Shader* kernel);
/// Configures images in a graphics shader.
void SetupDrawImages(const Shader* shader, size_t stage_index);
/// Configures images in a compute shader.
void SetupComputeImages(const Shader* shader);
/// Syncs state to match guest's
void SyncState();
@ -230,8 +203,6 @@ private:
/// End a transform feedback
void EndTransformFeedback();
void SetupShaders(bool is_indexed);
Tegra::GPU& gpu;
Tegra::Engines::Maxwell3D& maxwell3d;
Tegra::Engines::KeplerCompute& kepler_compute;
@ -251,8 +222,6 @@ private:
AccelerateDMA accelerate_dma;
FenceManagerOpenGL fence_manager;
VideoCommon::Shader::AsyncShaders async_shaders;
boost::container::static_vector<u32, MAX_IMAGE_VIEWS> image_view_indices;
std::array<ImageViewId, MAX_IMAGE_VIEWS> image_view_ids;
boost::container::static_vector<GLuint, MAX_TEXTURES> sampler_handles;

564
src/video_core/renderer_opengl/gl_shader_cache.cpp
View file

@ -20,307 +20,19 @@
#include "video_core/engines/maxwell_3d.h"
#include "video_core/engines/shader_type.h"
#include "video_core/memory_manager.h"
#include "video_core/renderer_opengl/gl_arb_decompiler.h"
#include "video_core/renderer_opengl/gl_rasterizer.h"
#include "video_core/renderer_opengl/gl_resource_manager.h"
#include "video_core/renderer_opengl/gl_shader_cache.h"
#include "video_core/renderer_opengl/gl_shader_decompiler.h"
#include "video_core/renderer_opengl/gl_shader_disk_cache.h"
#include "video_core/renderer_opengl/gl_state_tracker.h"
#include "video_core/shader/memory_util.h"
#include "video_core/shader/registry.h"
#include "video_core/shader/shader_ir.h"
#include "video_core/shader_cache.h"
#include "video_core/shader_notify.h"
namespace OpenGL {
using Tegra::Engines::ShaderType;
using VideoCommon::Shader::GetShaderAddress;
using VideoCommon::Shader::GetShaderCode;
using VideoCommon::Shader::GetUniqueIdentifier;
using VideoCommon::Shader::KERNEL_MAIN_OFFSET;
using VideoCommon::Shader::ProgramCode;
using VideoCommon::Shader::Registry;
using VideoCommon::Shader::ShaderIR;
using VideoCommon::Shader::STAGE_MAIN_OFFSET;
namespace {
constexpr VideoCommon::Shader::CompilerSettings COMPILER_SETTINGS{};
/// Gets the shader type from a Maxwell program type
constexpr GLenum GetGLShaderType(ShaderType shader_type) {
switch (shader_type) {
case ShaderType::Vertex:
return GL_VERTEX_SHADER;
case ShaderType::Geometry:
return GL_GEOMETRY_SHADER;
case ShaderType::Fragment:
return GL_FRAGMENT_SHADER;
case ShaderType::Compute:
return GL_COMPUTE_SHADER;
default:
return GL_NONE;
}
}
constexpr const char* GetShaderTypeName(ShaderType shader_type) {
switch (shader_type) {
case ShaderType::Vertex:
return "VS";
case ShaderType::TesselationControl:
return "HS";
case ShaderType::TesselationEval:
return "DS";
case ShaderType::Geometry:
return "GS";
case ShaderType::Fragment:
return "FS";
case ShaderType::Compute:
return "CS";
}
return "UNK";
}
constexpr ShaderType GetShaderType(Maxwell::ShaderProgram program_type) {
switch (program_type) {
case Maxwell::ShaderProgram::VertexA:
case Maxwell::ShaderProgram::VertexB:
return ShaderType::Vertex;
case Maxwell::ShaderProgram::TesselationControl:
return ShaderType::TesselationControl;
case Maxwell::ShaderProgram::TesselationEval:
return ShaderType::TesselationEval;
case Maxwell::ShaderProgram::Geometry:
return ShaderType::Geometry;
case Maxwell::ShaderProgram::Fragment:
return ShaderType::Fragment;
}
return {};
}
constexpr GLenum AssemblyEnum(ShaderType shader_type) {
switch (shader_type) {
case ShaderType::Vertex:
return GL_VERTEX_PROGRAM_NV;
case ShaderType::TesselationControl:
return GL_TESS_CONTROL_PROGRAM_NV;
case ShaderType::TesselationEval:
return GL_TESS_EVALUATION_PROGRAM_NV;
case ShaderType::Geometry:
return GL_GEOMETRY_PROGRAM_NV;
case ShaderType::Fragment:
return GL_FRAGMENT_PROGRAM_NV;
case ShaderType::Compute:
return GL_COMPUTE_PROGRAM_NV;
}
return {};
}
std::string MakeShaderID(u64 unique_identifier, ShaderType shader_type) {
return fmt::format("{}{:016X}", GetShaderTypeName(shader_type), unique_identifier);
}
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};
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);
}
for (const auto& [offset, sampler] : entry.bound_samplers) {
registry->InsertBoundSampler(offset, sampler);
}
for (const auto& [key, sampler] : entry.bindless_samplers) {
const auto [buffer, offset] = key;
registry->InsertBindlessSampler(buffer, offset, sampler);
}
return registry;
}
std::unordered_set<GLenum> GetSupportedFormats() {
GLint num_formats;
glGetIntegerv(GL_NUM_PROGRAM_BINARY_FORMATS, &num_formats);
std::vector<GLint> formats(num_formats);
glGetIntegerv(GL_PROGRAM_BINARY_FORMATS, formats.data());
std::unordered_set<GLenum> supported_formats;
for (const GLint format : formats) {
supported_formats.insert(static_cast<GLenum>(format));
}
return supported_formats;
}
} // Anonymous namespace
ProgramSharedPtr BuildShader(const Device& device, ShaderType shader_type, u64 unique_identifier,
const ShaderIR& ir, const Registry& registry, bool hint_retrievable) {
if (device.UseDriverCache()) {
// Ignore hint retrievable if we are using the driver cache
hint_retrievable = false;
}
const std::string shader_id = MakeShaderID(unique_identifier, shader_type);
LOG_INFO(Render_OpenGL, "{}", shader_id);
auto program = std::make_shared<ProgramHandle>();
if (device.UseAssemblyShaders()) {
const std::string arb =
DecompileAssemblyShader(device, ir, registry, shader_type, shader_id);
GLuint& arb_prog = program->assembly_program.handle;
// Commented out functions signal OpenGL errors but are compatible with apitrace.
// Use them only to capture and replay on apitrace.
#if 0
glGenProgramsNV(1, &arb_prog);
glLoadProgramNV(AssemblyEnum(shader_type), arb_prog, static_cast<GLsizei>(arb.size()),
reinterpret_cast<const GLubyte*>(arb.data()));
#else
glGenProgramsARB(1, &arb_prog);
glNamedProgramStringEXT(arb_prog, AssemblyEnum(shader_type), GL_PROGRAM_FORMAT_ASCII_ARB,
static_cast<GLsizei>(arb.size()), arb.data());
#endif
const auto err = reinterpret_cast<const char*>(glGetString(GL_PROGRAM_ERROR_STRING_NV));
if (err && *err) {
LOG_CRITICAL(Render_OpenGL, "{}", err);
LOG_INFO(Render_OpenGL, "\n{}", arb);
}
} else {
const std::string glsl = DecompileShader(device, ir, registry, shader_type, shader_id);
OGLShader shader;
shader.Create(glsl.c_str(), GetGLShaderType(shader_type));
program->source_program.Create(true, hint_retrievable, shader.handle);
}
return program;
}
Shader::Shader(std::shared_ptr<Registry> registry_, ShaderEntries entries_,
ProgramSharedPtr program_, bool is_built_)
: registry{std::move(registry_)}, entries{std::move(entries_)}, program{std::move(program_)},
is_built{is_built_} {
handle = program->assembly_program.handle;
if (handle == 0) {
handle = program->source_program.handle;
}
if (is_built) {
ASSERT(handle != 0);
}
}
Shader::Shader() = default;
Shader::~Shader() = default;
GLuint Shader::GetHandle() const {
DEBUG_ASSERT(registry->IsConsistent());
return handle;
}
bool Shader::IsBuilt() const {
return is_built;
}
void Shader::AsyncOpenGLBuilt(OGLProgram new_program) {
program->source_program = std::move(new_program);
handle = program->source_program.handle;
is_built = true;
}
void Shader::AsyncGLASMBuilt(OGLAssemblyProgram new_program) {
program->assembly_program = std::move(new_program);
handle = program->assembly_program.handle;
is_built = true;
}
std::unique_ptr<Shader> Shader::CreateStageFromMemory(
const ShaderParameters& params, Maxwell::ShaderProgram program_type, ProgramCode code,
ProgramCode code_b, VideoCommon::Shader::AsyncShaders& async_shaders, VAddr cpu_addr) {
const auto shader_type = GetShaderType(program_type);
auto& gpu = params.gpu;
gpu.ShaderNotify().MarkSharderBuilding();
auto registry = std::make_shared<Registry>(shader_type, gpu.Maxwell3D());
if (!async_shaders.IsShaderAsync(gpu) || !params.device.UseAsynchronousShaders()) {
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);
// }
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));
gpu.ShaderNotify().MarkShaderComplete();
return std::unique_ptr<Shader>(new Shader(std::move(registry),
MakeEntries(params.device, ir, shader_type),
std::move(program), true));
} else {
// Required for entries
const ShaderIR ir(code, STAGE_MAIN_OFFSET, COMPILER_SETTINGS, *registry);
auto entries = MakeEntries(params.device, ir, shader_type);
async_shaders.QueueOpenGLShader(params.device, shader_type, params.unique_identifier,
std::move(code), std::move(code_b), STAGE_MAIN_OFFSET,
COMPILER_SETTINGS, *registry, cpu_addr);
auto program = std::make_shared<ProgramHandle>();
return std::unique_ptr<Shader>(
new Shader(std::move(registry), std::move(entries), std::move(program), false));
}
}
std::unique_ptr<Shader> Shader::CreateKernelFromMemory(const ShaderParameters& params,
ProgramCode code) {
auto& gpu = params.gpu;
gpu.ShaderNotify().MarkSharderBuilding();
auto registry = std::make_shared<Registry>(ShaderType::Compute, params.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));
gpu.ShaderNotify().MarkShaderComplete();
return std::unique_ptr<Shader>(new Shader(std::move(registry),
MakeEntries(params.device, ir, ShaderType::Compute),
std::move(program)));
}
std::unique_ptr<Shader> Shader::CreateFromCache(const ShaderParameters& params,
const PrecompiledShader& precompiled_shader) {
return std::unique_ptr<Shader>(new Shader(
precompiled_shader.registry, precompiled_shader.entries, precompiled_shader.program));
}
ShaderCacheOpenGL::ShaderCacheOpenGL(RasterizerOpenGL& rasterizer_,
Core::Frontend::EmuWindow& emu_window_, Tegra::GPU& gpu_,
Tegra::Engines::Maxwell3D& maxwell3d_,
@ -331,278 +43,4 @@ ShaderCacheOpenGL::ShaderCacheOpenGL(RasterizerOpenGL& rasterizer_,
ShaderCacheOpenGL::~ShaderCacheOpenGL() = default;
void ShaderCacheOpenGL::LoadDiskCache(u64 title_id, std::stop_token stop_loading,
const VideoCore::DiskResourceLoadCallback& callback) {
disk_cache.BindTitleID(title_id);
const std::optional transferable = disk_cache.LoadTransferable();
LOG_INFO(Render_OpenGL, "Total Shader Count: {}",
transferable.has_value() ? transferable->size() : 0);
if (!transferable) {
return;
}
std::vector<ShaderDiskCachePrecompiled> gl_cache;
if (!device.UseAssemblyShaders() && !device.UseDriverCache()) {
// Only load precompiled cache when we are not using assembly shaders
gl_cache = disk_cache.LoadPrecompiled();
}
const auto supported_formats = GetSupportedFormats();
// Track if precompiled cache was altered during loading to know if we have to
// serialize the virtual precompiled cache file back to the hard drive
bool precompiled_cache_altered = false;
// Inform the frontend about shader build initialization
if (callback) {
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 gl_cache_failed = false;
const auto find_precompiled = [&gl_cache](u64 id) {
return std::ranges::find(gl_cache, id, &ShaderDiskCachePrecompiled::unique_identifier);
};
const auto worker = [&](Core::Frontend::GraphicsContext* context, std::size_t begin,
std::size_t end) {
const auto scope = context->Acquire();
for (std::size_t i = begin; i < end; ++i) {
if (stop_loading.stop_requested()) {
return;
}
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;
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);
ProgramSharedPtr 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 (!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(device, ir, entry.type);
std::scoped_lock lock{mutex};
if (callback) {
callback(VideoCore::LoadCallbackStage::Build, ++built_shaders,
transferable->size());
}
runtime_cache.emplace(entry.unique_identifier, std::move(shader));
}
};
const std::size_t num_workers{std::max(1U, std::thread::hardware_concurrency())};
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 ? 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);
}
for (auto& thread : threads) {
thread.join();
}
if (gl_cache_failed) {
// Invalidate the precompiled cache if a shader dumped shader was rejected
disk_cache.InvalidatePrecompiled();
precompiled_cache_altered = true;
return;
}
if (stop_loading.stop_requested()) {
return;
}
if (device.UseAssemblyShaders() || device.UseDriverCache()) {
// Don't store precompiled binaries for assembly shaders or when using the driver cache
return;
}
// TODO(Rodrigo): Do state tracking for transferable shaders and do a dummy draw
// before precompiling them
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->source_program.handle;
disk_cache.SavePrecompiled(id, program);
precompiled_cache_altered = true;
}
}
if (precompiled_cache_altered) {
disk_cache.SaveVirtualPrecompiledFile();
}
}
ProgramSharedPtr ShaderCacheOpenGL::GeneratePrecompiledProgram(
const ShaderDiskCacheEntry& entry, const ShaderDiskCachePrecompiled& precompiled_entry,
const std::unordered_set<GLenum>& supported_formats) {
if (!supported_formats.contains(precompiled_entry.binary_format)) {
LOG_INFO(Render_OpenGL, "Precompiled cache entry with unsupported format, removing");
return {};
}
auto program = std::make_shared<ProgramHandle>();
GLuint& handle = program->source_program.handle;
handle = glCreateProgram();
glProgramParameteri(handle, GL_PROGRAM_SEPARABLE, GL_TRUE);
glProgramBinary(handle, precompiled_entry.binary_format, precompiled_entry.binary.data(),
static_cast<GLsizei>(precompiled_entry.binary.size()));
GLint link_status;
glGetProgramiv(handle, GL_LINK_STATUS, &link_status);
if (link_status == GL_FALSE) {
LOG_INFO(Render_OpenGL, "Precompiled cache rejected by the driver, removing");
return {};
}
return program;
}
Shader* ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program,
VideoCommon::Shader::AsyncShaders& async_shaders) {
if (!maxwell3d.dirty.flags[Dirty::Shaders]) {
auto* last_shader = last_shaders[static_cast<std::size_t>(program)];
if (last_shader->IsBuilt()) {
return last_shader;
}
}
const GPUVAddr address{GetShaderAddress(maxwell3d, program)};
if (device.UseAsynchronousShaders() && async_shaders.HasCompletedWork()) {
auto completed_work = async_shaders.GetCompletedWork();
for (auto& work : completed_work) {
Shader* shader = TryGet(work.cpu_address);
gpu.ShaderNotify().MarkShaderComplete();
if (shader == nullptr) {
continue;
}
using namespace VideoCommon::Shader;
if (work.backend == AsyncShaders::Backend::OpenGL) {
shader->AsyncOpenGLBuilt(std::move(work.program.opengl));
} else if (work.backend == AsyncShaders::Backend::GLASM) {
shader->AsyncGLASMBuilt(std::move(work.program.glasm));
}
auto& registry = shader->GetRegistry();
ShaderDiskCacheEntry entry;
entry.type = work.shader_type;
entry.code = std::move(work.code);
entry.code_b = std::move(work.code_b);
entry.unique_identifier = work.uid;
entry.bound_buffer = registry.GetBoundBuffer();
entry.graphics_info = registry.GetGraphicsInfo();
entry.keys = registry.GetKeys();
entry.bound_samplers = registry.GetBoundSamplers();
entry.bindless_samplers = registry.GetBindlessSamplers();
disk_cache.SaveEntry(std::move(entry));
}
}
// Look up shader in the cache based on address
const std::optional<VAddr> cpu_addr{gpu_memory.GpuToCpuAddress(address)};
if (Shader* const shader{cpu_addr ? TryGet(*cpu_addr) : null_shader.get()}) {
return last_shaders[static_cast<std::size_t>(program)] = shader;
}
const u8* const host_ptr{gpu_memory.GetPointer(address)};
// No shader found - create a new one
ProgramCode code{GetShaderCode(gpu_memory, address, host_ptr, false)};
ProgramCode code_b;
if (program == Maxwell::ShaderProgram::VertexA) {
const GPUVAddr address_b{GetShaderAddress(maxwell3d, Maxwell::ShaderProgram::VertexB)};
const u8* host_ptr_b = gpu_memory.GetPointer(address_b);
code_b = GetShaderCode(gpu_memory, address_b, host_ptr_b, false);
}
const std::size_t code_size = code.size() * sizeof(u64);
const u64 unique_identifier = GetUniqueIdentifier(
GetShaderType(program), program == Maxwell::ShaderProgram::VertexA, code, code_b);
const ShaderParameters params{gpu, maxwell3d, disk_cache, device,
*cpu_addr, host_ptr, unique_identifier};
std::unique_ptr<Shader> shader;
const auto found = runtime_cache.find(unique_identifier);
if (found == runtime_cache.end()) {
shader = Shader::CreateStageFromMemory(params, program, std::move(code), std::move(code_b),
async_shaders, cpu_addr.value_or(0));
} else {
shader = Shader::CreateFromCache(params, found->second);
}
Shader* const result = shader.get();
if (cpu_addr) {
Register(std::move(shader), *cpu_addr, code_size);
} else {
null_shader = std::move(shader);
}
return last_shaders[static_cast<std::size_t>(program)] = result;
}
Shader* ShaderCacheOpenGL::GetComputeKernel(GPUVAddr code_addr) {
const std::optional<VAddr> cpu_addr{gpu_memory.GpuToCpuAddress(code_addr)};
if (Shader* const kernel = cpu_addr ? TryGet(*cpu_addr) : null_kernel.get()) {
return kernel;
}
// No kernel found, create a new one
const u8* host_ptr{gpu_memory.GetPointer(code_addr)};
ProgramCode code{GetShaderCode(gpu_memory, code_addr, host_ptr, true)};
const std::size_t code_size{code.size() * sizeof(u64)};
const u64 unique_identifier{GetUniqueIdentifier(ShaderType::Compute, false, code)};
const ShaderParameters params{gpu, kepler_compute, disk_cache, device,
*cpu_addr, host_ptr, unique_identifier};
std::unique_ptr<Shader> kernel;
const auto found = runtime_cache.find(unique_identifier);
if (found == runtime_cache.end()) {
kernel = Shader::CreateKernelFromMemory(params, std::move(code));
} else {
kernel = Shader::CreateFromCache(params, found->second);
}
Shader* const result = kernel.get();
if (cpu_addr) {
Register(std::move(kernel), *cpu_addr, code_size);
} else {
null_kernel = std::move(kernel);
}
return result;
}
} // namespace OpenGL

102
src/video_core/renderer_opengl/gl_shader_cache.h
View file

@ -19,10 +19,6 @@
#include "common/common_types.h"
#include "video_core/engines/shader_type.h"
#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/registry.h"
#include "video_core/shader/shader_ir.h"
#include "video_core/shader_cache.h"
namespace Tegra {
@ -33,10 +29,6 @@ namespace Core::Frontend {
class EmuWindow;
}
namespace VideoCommon::Shader {
class AsyncShaders;
}
namespace OpenGL {
class Device;
@ -44,77 +36,10 @@ class RasterizerOpenGL;
using Maxwell = Tegra::Engines::Maxwell3D::Regs;
struct ProgramHandle {
OGLProgram source_program;
OGLAssemblyProgram assembly_program;
};
using ProgramSharedPtr = std::shared_ptr<ProgramHandle>;
struct PrecompiledShader {
ProgramSharedPtr program;
std::shared_ptr<VideoCommon::Shader::Registry> registry;
ShaderEntries entries;
};
struct ShaderParameters {
Tegra::GPU& gpu;
Tegra::Engines::ConstBufferEngineInterface& engine;
ShaderDiskCacheOpenGL& disk_cache;
const Device& device;
VAddr cpu_addr;
const u8* host_ptr;
u64 unique_identifier;
};
ProgramSharedPtr BuildShader(const Device& device, Tegra::Engines::ShaderType shader_type,
u64 unique_identifier, const VideoCommon::Shader::ShaderIR& ir,
const VideoCommon::Shader::Registry& registry,
bool hint_retrievable = false);
class Shader final {
class Shader {
public:
explicit Shader();
~Shader();
/// Gets the GL program handle for the shader
GLuint GetHandle() const;
bool IsBuilt() const;
/// Gets the shader entries for the shader
const ShaderEntries& GetEntries() const {
return entries;
}
const VideoCommon::Shader::Registry& GetRegistry() const {
return *registry;
}
/// Mark a OpenGL shader as built
void AsyncOpenGLBuilt(OGLProgram new_program);
/// Mark a GLASM shader as built
void AsyncGLASMBuilt(OGLAssemblyProgram new_program);
static std::unique_ptr<Shader> CreateStageFromMemory(
const ShaderParameters& params, Maxwell::ShaderProgram program_type,
ProgramCode program_code, ProgramCode program_code_b,
VideoCommon::Shader::AsyncShaders& async_shaders, VAddr cpu_addr);
static std::unique_ptr<Shader> CreateKernelFromMemory(const ShaderParameters& params,
ProgramCode code);
static std::unique_ptr<Shader> CreateFromCache(const ShaderParameters& params,
const PrecompiledShader& precompiled_shader);
private:
explicit Shader(std::shared_ptr<VideoCommon::Shader::Registry> registry, ShaderEntries entries,
ProgramSharedPtr program, bool is_built_ = true);
std::shared_ptr<VideoCommon::Shader::Registry> registry;
ShaderEntries entries;
ProgramSharedPtr program;
GLuint handle = 0;
bool is_built{};
};
class ShaderCacheOpenGL final : public VideoCommon::ShaderCache<Shader> {
@ -126,36 +51,13 @@ public:
Tegra::MemoryManager& gpu_memory_, const Device& device_);
~ShaderCacheOpenGL() override;
/// Loads disk cache for the current game
void LoadDiskCache(u64 title_id, std::stop_token stop_loading,
const VideoCore::DiskResourceLoadCallback& callback);
/// Gets the current specified shader stage program
Shader* GetStageProgram(Maxwell::ShaderProgram program,
VideoCommon::Shader::AsyncShaders& async_shaders);
/// Gets a compute kernel in the passed address
Shader* GetComputeKernel(GPUVAddr code_addr);
private:
ProgramSharedPtr GeneratePrecompiledProgram(
const ShaderDiskCacheEntry& entry, const ShaderDiskCachePrecompiled& precompiled_entry,
const std::unordered_set<GLenum>& supported_formats);
Core::Frontend::EmuWindow& emu_window;
Tegra::GPU& gpu;
Tegra::MemoryManager& gpu_memory;
Tegra::Engines::Maxwell3D& maxwell3d;
Tegra::Engines::KeplerCompute& kepler_compute;
const Device& device;
ShaderDiskCacheOpenGL disk_cache;
std::unordered_map<u64, PrecompiledShader> runtime_cache;
std::unique_ptr<Shader> null_shader;
std::unique_ptr<Shader> null_kernel;
std::array<Shader*, Maxwell::MaxShaderProgram> last_shaders{};
};
} // namespace OpenGL

2986
src/video_core/renderer_opengl/gl_shader_decompiler.cpp
View file

File diff suppressed because it is too large Load diff

69
src/video_core/renderer_opengl/gl_shader_decompiler.h
View file

@ -1,69 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#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 OpenGL {
class Device;
using Maxwell = Tegra::Engines::Maxwell3D::Regs;
using SamplerEntry = VideoCommon::Shader::SamplerEntry;
using ImageEntry = VideoCommon::Shader::ImageEntry;
class ConstBufferEntry : public VideoCommon::Shader::ConstBuffer {
public:
explicit ConstBufferEntry(u32 max_offset_, bool is_indirect_, u32 index_)
: ConstBuffer{max_offset_, is_indirect_}, index{index_} {}
u32 GetIndex() const {
return index;
}
private:
u32 index = 0;
};
struct GlobalMemoryEntry {
constexpr explicit GlobalMemoryEntry(u32 cbuf_index_, u32 cbuf_offset_, bool is_read_,
bool is_written_)
: cbuf_index{cbuf_index_}, cbuf_offset{cbuf_offset_}, is_read{is_read_}, is_written{
is_written_} {}
u32 cbuf_index = 0;
u32 cbuf_offset = 0;
bool is_read = false;
bool is_written = false;
};
struct ShaderEntries {
std::vector<ConstBufferEntry> const_buffers;
std::vector<GlobalMemoryEntry> global_memory_entries;
std::vector<SamplerEntry> samplers;
std::vector<ImageEntry> images;
std::size_t shader_length{};
u32 clip_distances{};
u32 enabled_uniform_buffers{};
};
ShaderEntries MakeEntries(const Device& device, const VideoCommon::Shader::ShaderIR& ir,
Tegra::Engines::ShaderType stage);
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 = {});
} // namespace OpenGL

482
src/video_core/renderer_opengl/gl_shader_disk_cache.cpp
View file

@ -1,482 +0,0 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <cstring>
#include <fmt/format.h>
#include "common/assert.h"
#include "common/common_types.h"
#include "common/fs/file.h"
#include "common/fs/fs.h"
#include "common/fs/path_util.h"
#include "common/logging/log.h"
#include "common/scm_rev.h"
#include "common/settings.h"
#include "common/zstd_compression.h"
#include "core/core.h"
#include "core/hle/kernel/k_process.h"
#include "video_core/engines/shader_type.h"
#include "video_core/renderer_opengl/gl_shader_cache.h"
#include "video_core/renderer_opengl/gl_shader_disk_cache.h"
namespace OpenGL {
using Tegra::Engines::ShaderType;
using VideoCommon::Shader::BindlessSamplerMap;
using VideoCommon::Shader::BoundSamplerMap;
using VideoCommon::Shader::KeyMap;
using VideoCommon::Shader::SeparateSamplerKey;
using ShaderCacheVersionHash = std::array<u8, 64>;
struct ConstBufferKey {
u32 cbuf = 0;
u32 offset = 0;
u32 value = 0;
};
struct BoundSamplerEntry {
u32 offset = 0;
Tegra::Engines::SamplerDescriptor sampler;
};
struct SeparateSamplerEntry {
u32 cbuf1 = 0;
u32 cbuf2 = 0;
u32 offset1 = 0;
u32 offset2 = 0;
Tegra::Engines::SamplerDescriptor sampler;
};
struct BindlessSamplerEntry {
u32 cbuf = 0;
u32 offset = 0;
Tegra::Engines::SamplerDescriptor sampler;
};
namespace {
constexpr u32 NativeVersion = 21;
ShaderCacheVersionHash GetShaderCacheVersionHash() {
ShaderCacheVersionHash hash{};
const std::size_t length = std::min(std::strlen(Common::g_shader_cache_version), hash.size());
std::memcpy(hash.data(), Common::g_shader_cache_version, length);
return hash;
}
} // Anonymous namespace
ShaderDiskCacheEntry::ShaderDiskCacheEntry() = default;
ShaderDiskCacheEntry::~ShaderDiskCacheEntry() = default;
bool ShaderDiskCacheEntry::Load(Common::FS::IOFile& file) {
if (!file.ReadObject(type)) {
return false;
}
u32 code_size;
u32 code_size_b;
if (!file.ReadObject(code_size) || !file.ReadObject(code_size_b)) {
return false;
}
code.resize(code_size);
code_b.resize(code_size_b);
if (file.Read(code) != code_size) {
return false;
}
if (HasProgramA() && file.Read(code_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_separate_samplers;
u32 num_bindless_samplers;
if (!file.ReadObject(unique_identifier) || !file.ReadObject(bound_buffer) ||
!file.ReadObject(is_texture_handler_size_known) ||
!file.ReadObject(texture_handler_size_value) || !file.ReadObject(graphics_info) ||
!file.ReadObject(compute_info) || !file.ReadObject(num_keys) ||
!file.ReadObject(num_bound_samplers) || !file.ReadObject(num_separate_samplers) ||
!file.ReadObject(num_bindless_samplers)) {
return false;
}
if (is_texture_handler_size_known) {
texture_handler_size = texture_handler_size_value;
}
std::vector<ConstBufferKey> flat_keys(num_keys);
std::vector<BoundSamplerEntry> flat_bound_samplers(num_bound_samplers);
std::vector<SeparateSamplerEntry> flat_separate_samplers(num_separate_samplers);
std::vector<BindlessSamplerEntry> flat_bindless_samplers(num_bindless_samplers);
if (file.Read(flat_keys) != flat_keys.size() ||
file.Read(flat_bound_samplers) != flat_bound_samplers.size() ||
file.Read(flat_separate_samplers) != flat_separate_samplers.size() ||
file.Read(flat_bindless_samplers) != flat_bindless_samplers.size()) {
return false;
}
for (const auto& entry : flat_keys) {
keys.insert({{entry.cbuf, entry.offset}, entry.value});
}
for (const auto& entry : flat_bound_samplers) {
bound_samplers.emplace(entry.offset, entry.sampler);
}
for (const auto& entry : flat_separate_samplers) {
SeparateSamplerKey key;
key.buffers = {entry.cbuf1, entry.cbuf2};
key.offsets = {entry.offset1, entry.offset2};
separate_samplers.emplace(key, entry.sampler);
}
for (const auto& entry : flat_bindless_samplers) {
bindless_samplers.insert({{entry.cbuf, entry.offset}, entry.sampler});
}
return true;
}
bool ShaderDiskCacheEntry::Save(Common::FS::IOFile& file) const {
if (!file.WriteObject(static_cast<u32>(type)) ||
!file.WriteObject(static_cast<u32>(code.size())) ||
!file.WriteObject(static_cast<u32>(code_b.size()))) {
return false;
}
if (file.Write(code) != code.size()) {
return false;
}
if (HasProgramA() && file.Write(code_b) != code_b.size()) {
return false;
}
if (!file.WriteObject(unique_identifier) || !file.WriteObject(bound_buffer) ||
!file.WriteObject(static_cast<u8>(texture_handler_size.has_value())) ||
!file.WriteObject(texture_handler_size.value_or(0)) || !file.WriteObject(graphics_info) ||
!file.WriteObject(compute_info) || !file.WriteObject(static_cast<u32>(keys.size())) ||
!file.WriteObject(static_cast<u32>(bound_samplers.size())) ||
!file.WriteObject(static_cast<u32>(separate_samplers.size())) ||
!file.WriteObject(static_cast<u32>(bindless_samplers.size()))) {
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<BoundSamplerEntry> flat_bound_samplers;
flat_bound_samplers.reserve(bound_samplers.size());
for (const auto& [address, sampler] : bound_samplers) {
flat_bound_samplers.push_back(BoundSamplerEntry{address, sampler});
}
std::vector<SeparateSamplerEntry> flat_separate_samplers;
flat_separate_samplers.reserve(separate_samplers.size());
for (const auto& [key, sampler] : separate_samplers) {
SeparateSamplerEntry entry;
std::tie(entry.cbuf1, entry.cbuf2) = key.buffers;
std::tie(entry.offset1, entry.offset2) = key.offsets;
entry.sampler = sampler;
flat_separate_samplers.push_back(entry);
}
std::vector<BindlessSamplerEntry> flat_bindless_samplers;
flat_bindless_samplers.reserve(bindless_samplers.size());
for (const auto& [address, sampler] : bindless_samplers) {
flat_bindless_samplers.push_back(
BindlessSamplerEntry{address.first, address.second, sampler});
}
return file.Write(flat_keys) == flat_keys.size() &&
file.Write(flat_bound_samplers) == flat_bound_samplers.size() &&
file.Write(flat_separate_samplers) == flat_separate_samplers.size() &&
file.Write(flat_bindless_samplers) == flat_bindless_samplers.size();
}
ShaderDiskCacheOpenGL::ShaderDiskCacheOpenGL() = default;
ShaderDiskCacheOpenGL::~ShaderDiskCacheOpenGL() = default;
void ShaderDiskCacheOpenGL::BindTitleID(u64 title_id_) {
title_id = title_id_;
}
std::optional<std::vector<ShaderDiskCacheEntry>> ShaderDiskCacheOpenGL::LoadTransferable() {
// Skip games without title id
const bool has_title_id = title_id != 0;
if (!Settings::values.use_disk_shader_cache.GetValue() || !has_title_id) {
return std::nullopt;
}
Common::FS::IOFile file{GetTransferablePath(), Common::FS::FileAccessMode::Read,
Common::FS::FileType::BinaryFile};
if (!file.IsOpen()) {
LOG_INFO(Render_OpenGL, "No transferable shader cache found");
is_usable = true;
return std::nullopt;
}
u32 version{};
if (!file.ReadObject(version)) {
LOG_ERROR(Render_OpenGL, "Failed to get transferable cache version, skipping it");
return std::nullopt;
}
if (version < NativeVersion) {
LOG_INFO(Render_OpenGL, "Transferable shader cache is old, removing");
file.Close();
InvalidateTransferable();
is_usable = true;
return std::nullopt;
}
if (version > NativeVersion) {
LOG_WARNING(Render_OpenGL, "Transferable shader cache was generated with a newer version "
"of the emulator, skipping");
return std::nullopt;
}
// Version is valid, load the shaders
std::vector<ShaderDiskCacheEntry> entries;
while (static_cast<u64>(file.Tell()) < file.GetSize()) {
ShaderDiskCacheEntry& entry = entries.emplace_back();
if (!entry.Load(file)) {
LOG_ERROR(Render_OpenGL, "Failed to load transferable raw entry, skipping");
return std::nullopt;
}
}
is_usable = true;
return {std::move(entries)};
}
std::vector<ShaderDiskCachePrecompiled> ShaderDiskCacheOpenGL::LoadPrecompiled() {
if (!is_usable) {
return {};
}
Common::FS::IOFile file{GetPrecompiledPath(), Common::FS::FileAccessMode::Read,
Common::FS::FileType::BinaryFile};
if (!file.IsOpen()) {
LOG_INFO(Render_OpenGL, "No precompiled shader cache found");
return {};
}
if (const auto result = LoadPrecompiledFile(file)) {
return *result;
}
LOG_INFO(Render_OpenGL, "Failed to load precompiled cache");
file.Close();
InvalidatePrecompiled();
return {};
}
std::optional<std::vector<ShaderDiskCachePrecompiled>> ShaderDiskCacheOpenGL::LoadPrecompiledFile(
Common::FS::IOFile& file) {
// Read compressed file from disk and decompress to virtual precompiled cache file
std::vector<u8> compressed(file.GetSize());
if (file.Read(compressed) != file.GetSize()) {
return std::nullopt;
}
const std::vector<u8> decompressed = Common::Compression::DecompressDataZSTD(compressed);
SaveArrayToPrecompiled(decompressed.data(), decompressed.size());
precompiled_cache_virtual_file_offset = 0;
ShaderCacheVersionHash file_hash{};
if (!LoadArrayFromPrecompiled(file_hash.data(), file_hash.size())) {
precompiled_cache_virtual_file_offset = 0;
return std::nullopt;
}
if (GetShaderCacheVersionHash() != file_hash) {
LOG_INFO(Render_OpenGL, "Precompiled cache is from another version of the emulator");
precompiled_cache_virtual_file_offset = 0;
return std::nullopt;
}
std::vector<ShaderDiskCachePrecompiled> entries;
while (precompiled_cache_virtual_file_offset < precompiled_cache_virtual_file.GetSize()) {
u32 binary_size;
auto& entry = entries.emplace_back();
if (!LoadObjectFromPrecompiled(entry.unique_identifier) ||
!LoadObjectFromPrecompiled(entry.binary_format) ||
!LoadObjectFromPrecompiled(binary_size)) {
return std::nullopt;
}
entry.binary.resize(binary_size);
if (!LoadArrayFromPrecompiled(entry.binary.data(), entry.binary.size())) {
return std::nullopt;
}
}
return entries;
}
void ShaderDiskCacheOpenGL::InvalidateTransferable() {
if (!Common::FS::RemoveFile(GetTransferablePath())) {
LOG_ERROR(Render_OpenGL, "Failed to invalidate transferable file={}",
Common::FS::PathToUTF8String(GetTransferablePath()));
}
InvalidatePrecompiled();
}
void ShaderDiskCacheOpenGL::InvalidatePrecompiled() {
// Clear virtaul precompiled cache file
precompiled_cache_virtual_file.Resize(0);
if (!Common::FS::RemoveFile(GetPrecompiledPath())) {
LOG_ERROR(Render_OpenGL, "Failed to invalidate precompiled file={}",
Common::FS::PathToUTF8String(GetPrecompiledPath()));
}
}
void ShaderDiskCacheOpenGL::SaveEntry(const ShaderDiskCacheEntry& entry) {
if (!is_usable) {
return;
}
const u64 id = entry.unique_identifier;
if (stored_transferable.contains(id)) {
// The shader already exists
return;
}
Common::FS::IOFile file = AppendTransferableFile();
if (!file.IsOpen()) {
return;
}
if (!entry.Save(file)) {
LOG_ERROR(Render_OpenGL, "Failed to save raw transferable cache entry, removing");
file.Close();
InvalidateTransferable();
return;
}
stored_transferable.insert(id);
}
void ShaderDiskCacheOpenGL::SavePrecompiled(u64 unique_identifier, GLuint program) {
if (!is_usable) {
return;
}
// TODO(Rodrigo): This is a design smell. I shouldn't be having to manually write the header
// when writing the dump. This should be done the moment I get access to write to the virtual
// file.
if (precompiled_cache_virtual_file.GetSize() == 0) {
SavePrecompiledHeaderToVirtualPrecompiledCache();
}
GLint binary_length;
glGetProgramiv(program, GL_PROGRAM_BINARY_LENGTH, &binary_length);
GLenum binary_format;
std::vector<u8> binary(binary_length);
glGetProgramBinary(program, binary_length, nullptr, &binary_format, binary.data());
if (!SaveObjectToPrecompiled(unique_identifier) || !SaveObjectToPrecompiled(binary_format) ||
!SaveObjectToPrecompiled(static_cast<u32>(binary.size())) ||
!SaveArrayToPrecompiled(binary.data(), binary.size())) {
LOG_ERROR(Render_OpenGL, "Failed to save binary program file in shader={:016X}, removing",
unique_identifier);
InvalidatePrecompiled();
}
}
Common::FS::IOFile ShaderDiskCacheOpenGL::AppendTransferableFile() const {
if (!EnsureDirectories()) {
return {};
}
const auto transferable_path{GetTransferablePath()};
const bool existed = Common::FS::Exists(transferable_path);
Common::FS::IOFile file{transferable_path, Common::FS::FileAccessMode::Append,
Common::FS::FileType::BinaryFile};
if (!file.IsOpen()) {
LOG_ERROR(Render_OpenGL, "Failed to open transferable cache in path={}",
Common::FS::PathToUTF8String(transferable_path));
return {};
}
if (!existed || file.GetSize() == 0) {
// If the file didn't exist, write its version
if (!file.WriteObject(NativeVersion)) {
LOG_ERROR(Render_OpenGL, "Failed to write transferable cache version in path={}",
Common::FS::PathToUTF8String(transferable_path));
return {};
}
}
return file;
}
void ShaderDiskCacheOpenGL::SavePrecompiledHeaderToVirtualPrecompiledCache() {
const auto hash{GetShaderCacheVersionHash()};
if (!SaveArrayToPrecompiled(hash.data(), hash.size())) {
LOG_ERROR(
Render_OpenGL,
"Failed to write precompiled cache version hash to virtual precompiled cache file");
}
}
void ShaderDiskCacheOpenGL::SaveVirtualPrecompiledFile() {
precompiled_cache_virtual_file_offset = 0;
const std::vector<u8> uncompressed = precompiled_cache_virtual_file.ReadAllBytes();
const std::vector<u8> compressed =
Common::Compression::CompressDataZSTDDefault(uncompressed.data(), uncompressed.size());
const auto precompiled_path = GetPrecompiledPath();
Common::FS::IOFile file{precompiled_path, Common::FS::FileAccessMode::Write,
Common::FS::FileType::BinaryFile};
if (!file.IsOpen()) {
LOG_ERROR(Render_OpenGL, "Failed to open precompiled cache in path={}",
Common::FS::PathToUTF8String(precompiled_path));
return;
}
if (file.Write(compressed) != compressed.size()) {
LOG_ERROR(Render_OpenGL, "Failed to write precompiled cache version in path={}",
Common::FS::PathToUTF8String(precompiled_path));
}
}
bool ShaderDiskCacheOpenGL::EnsureDirectories() const {
const auto CreateDir = [](const std::filesystem::path& dir) {
if (!Common::FS::CreateDir(dir)) {
LOG_ERROR(Render_OpenGL, "Failed to create directory={}",
Common::FS::PathToUTF8String(dir));
return false;
}
return true;
};
return CreateDir(Common::FS::GetYuzuPath(Common::FS::YuzuPath::ShaderDir)) &&
CreateDir(GetBaseDir()) && CreateDir(GetTransferableDir()) &&
CreateDir(GetPrecompiledDir());
}
std::filesystem::path ShaderDiskCacheOpenGL::GetTransferablePath() const {
return GetTransferableDir() / fmt::format("{}.bin", GetTitleID());
}
std::filesystem::path ShaderDiskCacheOpenGL::GetPrecompiledPath() const {
return GetPrecompiledDir() / fmt::format("{}.bin", GetTitleID());
}
std::filesystem::path ShaderDiskCacheOpenGL::GetTransferableDir() const {
return GetBaseDir() / "transferable";
}
std::filesystem::path ShaderDiskCacheOpenGL::GetPrecompiledDir() const {
return GetBaseDir() / "precompiled";
}
std::filesystem::path ShaderDiskCacheOpenGL::GetBaseDir() const {
return Common::FS::GetYuzuPath(Common::FS::YuzuPath::ShaderDir) / "opengl";
}
std::string ShaderDiskCacheOpenGL::GetTitleID() const {
return fmt::format("{:016X}", title_id);
}
} // namespace OpenGL

176
src/video_core/renderer_opengl/gl_shader_disk_cache.h
View file

@ -1,176 +0,0 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <filesystem>
#include <optional>
#include <string>
#include <tuple>
#include <type_traits>
#include <unordered_map>
#include <unordered_set>
#include <utility>
#include <vector>
#include <glad/glad.h>
#include "common/assert.h"
#include "common/common_types.h"
#include "core/file_sys/vfs_vector.h"
#include "video_core/engines/shader_type.h"
#include "video_core/shader/registry.h"
namespace Common::FS {
class IOFile;
}
namespace OpenGL {
using ProgramCode = std::vector<u64>;
/// Describes a shader and how it's used by the guest GPU
struct ShaderDiskCacheEntry {
ShaderDiskCacheEntry();
~ShaderDiskCacheEntry();
bool Load(Common::FS::IOFile& file);
bool Save(Common::FS::IOFile& file) const;
bool HasProgramA() const {
return !code.empty() && !code_b.empty();
}
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::SeparateSamplerMap separate_samplers;
VideoCommon::Shader::BindlessSamplerMap bindless_samplers;
};
/// Contains an OpenGL dumped binary program
struct ShaderDiskCachePrecompiled {
u64 unique_identifier = 0;
GLenum binary_format = 0;
std::vector<u8> binary;
};
class ShaderDiskCacheOpenGL {
public:
explicit ShaderDiskCacheOpenGL();
~ShaderDiskCacheOpenGL();
/// Binds a title ID for all future operations.
void BindTitleID(u64 title_id);
/// Loads transferable cache. If file has a old version or on failure, it deletes the file.
std::optional<std::vector<ShaderDiskCacheEntry>> LoadTransferable();
/// Loads current game's precompiled cache. Invalidates on failure.
std::vector<ShaderDiskCachePrecompiled> LoadPrecompiled();
/// Removes the transferable (and precompiled) cache file.
void InvalidateTransferable();
/// Removes the precompiled cache file and clears virtual precompiled cache file.
void InvalidatePrecompiled();
/// Saves a raw dump to the transferable file. Checks for collisions.
void SaveEntry(const ShaderDiskCacheEntry& entry);
/// Saves a dump entry to the precompiled file. Does not check for collisions.
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::vector<ShaderDiskCachePrecompiled>> LoadPrecompiledFile(
Common::FS::IOFile& file);
/// Opens current game's transferable file and write it's header if it doesn't exist
Common::FS::IOFile AppendTransferableFile() const;
/// Save precompiled header to precompiled_cache_in_memory
void SavePrecompiledHeaderToVirtualPrecompiledCache();
/// Create shader disk cache directories. Returns true on success.
bool EnsureDirectories() const;
/// Gets current game's transferable file path
std::filesystem::path GetTransferablePath() const;
/// Gets current game's precompiled file path
std::filesystem::path GetPrecompiledPath() const;
/// Get user's transferable directory path
std::filesystem::path GetTransferableDir() const;
/// Get user's precompiled directory path
std::filesystem::path GetPrecompiledDir() const;
/// Get user's shader directory path
std::filesystem::path GetBaseDir() const;
/// Get current game's title id
std::string GetTitleID() const;
template <typename T>
bool SaveArrayToPrecompiled(const T* data, std::size_t length) {
const std::size_t write_length = precompiled_cache_virtual_file.WriteArray(
data, length, precompiled_cache_virtual_file_offset);
precompiled_cache_virtual_file_offset += write_length;
return write_length == sizeof(T) * length;
}
template <typename T>
bool LoadArrayFromPrecompiled(T* data, std::size_t length) {
const std::size_t read_length = precompiled_cache_virtual_file.ReadArray(
data, length, precompiled_cache_virtual_file_offset);
precompiled_cache_virtual_file_offset += read_length;
return read_length == sizeof(T) * length;
}
template <typename T>
bool SaveObjectToPrecompiled(const T& object) {
return SaveArrayToPrecompiled(&object, 1);
}
bool SaveObjectToPrecompiled(bool object) {
const auto value = static_cast<u8>(object);
return SaveArrayToPrecompiled(&value, 1);
}
template <typename T>
bool LoadObjectFromPrecompiled(T& object) {
return LoadArrayFromPrecompiled(&object, 1);
}
// Stores whole precompiled cache which will be read from or saved to the precompiled chache
// file
FileSys::VectorVfsFile precompiled_cache_virtual_file;
// Stores the current offset of the precompiled cache file for IO purposes
std::size_t precompiled_cache_virtual_file_offset = 0;
// Stored transferable shaders
std::unordered_set<u64> stored_transferable;
/// Title ID to operate on
u64 title_id = 0;
// The cache has been loaded at boot
bool is_usable = false;
};
} // namespace OpenGL

1
src/video_core/renderer_vulkan/blit_image.cpp
View file

@ -323,7 +323,6 @@ void BindBlitState(vk::CommandBuffer cmdbuf, VkPipelineLayout layout, const Regi
cmdbuf.SetScissor(0, scissor);
cmdbuf.PushConstants(layout, VK_SHADER_STAGE_VERTEX_BIT, push_constants);
}
} // Anonymous namespace
BlitImageHelper::BlitImageHelper(const Device& device_, VKScheduler& scheduler_,

136
src/video_core/renderer_vulkan/vk_compute_pipeline.cpp
View file

@ -8,146 +8,14 @@
#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
#include "video_core/renderer_vulkan/vk_pipeline_cache.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_shader_decompiler.h"
#include "video_core/renderer_vulkan/vk_update_descriptor.h"
#include "video_core/vulkan_common/vulkan_device.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
namespace Vulkan {
VKComputePipeline::VKComputePipeline(const Device& device_, VKScheduler& scheduler_,
VKDescriptorPool& descriptor_pool_,
VKUpdateDescriptorQueue& update_descriptor_queue_,
const SPIRVShader& shader_)
: device{device_}, scheduler{scheduler_}, entries{shader_.entries},
descriptor_set_layout{CreateDescriptorSetLayout()},
descriptor_allocator{descriptor_pool_, *descriptor_set_layout},
update_descriptor_queue{update_descriptor_queue_}, layout{CreatePipelineLayout()},
descriptor_template{CreateDescriptorUpdateTemplate()},
shader_module{CreateShaderModule(shader_.code)}, pipeline{CreatePipeline()} {}
ComputePipeline::ComputePipeline() = default;
VKComputePipeline::~VKComputePipeline() = default;
VkDescriptorSet VKComputePipeline::CommitDescriptorSet() {
if (!descriptor_template) {
return {};
}
const VkDescriptorSet set = descriptor_allocator.Commit();
update_descriptor_queue.Send(*descriptor_template, set);
return set;
}
vk::DescriptorSetLayout VKComputePipeline::CreateDescriptorSetLayout() const {
std::vector<VkDescriptorSetLayoutBinding> bindings;
u32 binding = 0;
const auto add_bindings = [&](VkDescriptorType descriptor_type, std::size_t num_entries) {
// TODO(Rodrigo): Maybe make individual bindings here?
for (u32 bindpoint = 0; bindpoint < static_cast<u32>(num_entries); ++bindpoint) {
bindings.push_back({
.binding = binding++,
.descriptorType = descriptor_type,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = nullptr,
});
}
};
add_bindings(VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, entries.const_buffers.size());
add_bindings(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, entries.global_buffers.size());
add_bindings(VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER, entries.uniform_texels.size());
add_bindings(VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, entries.samplers.size());
add_bindings(VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER, entries.storage_texels.size());
add_bindings(VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, entries.images.size());
return device.GetLogical().CreateDescriptorSetLayout({
.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.bindingCount = static_cast<u32>(bindings.size()),
.pBindings = bindings.data(),
});
}
vk::PipelineLayout VKComputePipeline::CreatePipelineLayout() const {
return device.GetLogical().CreatePipelineLayout({
.sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.setLayoutCount = 1,
.pSetLayouts = descriptor_set_layout.address(),
.pushConstantRangeCount = 0,
.pPushConstantRanges = nullptr,
});
}
vk::DescriptorUpdateTemplateKHR VKComputePipeline::CreateDescriptorUpdateTemplate() const {
std::vector<VkDescriptorUpdateTemplateEntryKHR> template_entries;
u32 binding = 0;
u32 offset = 0;
FillDescriptorUpdateTemplateEntries(entries, binding, offset, template_entries);
if (template_entries.empty()) {
// If the shader doesn't use descriptor sets, skip template creation.
return {};
}
return device.GetLogical().CreateDescriptorUpdateTemplateKHR({
.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_UPDATE_TEMPLATE_CREATE_INFO_KHR,
.pNext = nullptr,
.flags = 0,
.descriptorUpdateEntryCount = static_cast<u32>(template_entries.size()),
.pDescriptorUpdateEntries = template_entries.data(),
.templateType = VK_DESCRIPTOR_UPDATE_TEMPLATE_TYPE_DESCRIPTOR_SET_KHR,
.descriptorSetLayout = *descriptor_set_layout,
.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS,
.pipelineLayout = *layout,
.set = DESCRIPTOR_SET,
});
}
vk::ShaderModule VKComputePipeline::CreateShaderModule(const std::vector<u32>& code) const {
device.SaveShader(code);
return device.GetLogical().CreateShaderModule({
.sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.codeSize = code.size() * sizeof(u32),
.pCode = code.data(),
});
}
vk::Pipeline VKComputePipeline::CreatePipeline() const {
VkComputePipelineCreateInfo ci{
.sType = VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.stage =
{
.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.stage = VK_SHADER_STAGE_COMPUTE_BIT,
.module = *shader_module,
.pName = "main",
.pSpecializationInfo = nullptr,
},
.layout = *layout,
.basePipelineHandle = nullptr,
.basePipelineIndex = 0,
};
const VkPipelineShaderStageRequiredSubgroupSizeCreateInfoEXT subgroup_size_ci{
.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_REQUIRED_SUBGROUP_SIZE_CREATE_INFO_EXT,
.pNext = nullptr,
.requiredSubgroupSize = GuestWarpSize,
};
if (entries.uses_warps && device.IsGuestWarpSizeSupported(VK_SHADER_STAGE_COMPUTE_BIT)) {
ci.stage.pNext = &subgroup_size_ci;
}
return device.GetLogical().CreateComputePipeline(ci);
}
ComputePipeline::~ComputePipeline() = default;
} // namespace Vulkan

47
src/video_core/renderer_vulkan/vk_compute_pipeline.h
View file

@ -6,7 +6,6 @@
#include "common/common_types.h"
#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
#include "video_core/renderer_vulkan/vk_shader_decompiler.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
namespace Vulkan {
@ -15,50 +14,10 @@ class Device;
class VKScheduler;
class VKUpdateDescriptorQueue;
class VKComputePipeline final {
class ComputePipeline {
public:
explicit VKComputePipeline(const Device& device_, VKScheduler& scheduler_,
VKDescriptorPool& descriptor_pool_,
VKUpdateDescriptorQueue& update_descriptor_queue_,
const SPIRVShader& shader_);
~VKComputePipeline();
VkDescriptorSet CommitDescriptorSet();
VkPipeline GetHandle() const {
return *pipeline;
}
VkPipelineLayout GetLayout() const {
return *layout;
}
const ShaderEntries& GetEntries() const {
return entries;
}
private:
vk::DescriptorSetLayout CreateDescriptorSetLayout() const;
vk::PipelineLayout CreatePipelineLayout() const;
vk::DescriptorUpdateTemplateKHR CreateDescriptorUpdateTemplate() const;
vk::ShaderModule CreateShaderModule(const std::vector<u32>& code) const;
vk::Pipeline CreatePipeline() const;
const Device& device;
VKScheduler& scheduler;
ShaderEntries entries;
vk::DescriptorSetLayout descriptor_set_layout;
DescriptorAllocator descriptor_allocator;
VKUpdateDescriptorQueue& update_descriptor_queue;
vk::PipelineLayout layout;
vk::DescriptorUpdateTemplateKHR descriptor_template;
vk::ShaderModule shader_module;
vk::Pipeline pipeline;
explicit ComputePipeline();
~ComputePipeline();
};
} // namespace Vulkan

484
src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp
View file

@ -1,484 +0,0 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <algorithm>
#include <array>
#include <cstring>
#include <vector>
#include "common/common_types.h"
#include "common/microprofile.h"
#include "video_core/renderer_vulkan/fixed_pipeline_state.h"
#include "video_core/renderer_vulkan/maxwell_to_vk.h"
#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
#include "video_core/renderer_vulkan/vk_graphics_pipeline.h"
#include "video_core/renderer_vulkan/vk_pipeline_cache.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_update_descriptor.h"
#include "video_core/vulkan_common/vulkan_device.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
namespace Vulkan {
MICROPROFILE_DECLARE(Vulkan_PipelineCache);
namespace {
template <class StencilFace>
VkStencilOpState GetStencilFaceState(const StencilFace& face) {
return {
.failOp = MaxwellToVK::StencilOp(face.ActionStencilFail()),
.passOp = MaxwellToVK::StencilOp(face.ActionDepthPass()),
.depthFailOp = MaxwellToVK::StencilOp(face.ActionDepthFail()),
.compareOp = MaxwellToVK::ComparisonOp(face.TestFunc()),
.compareMask = 0,
.writeMask = 0,
.reference = 0,
};
}
bool SupportsPrimitiveRestart(VkPrimitiveTopology topology) {
static constexpr std::array unsupported_topologies = {
VK_PRIMITIVE_TOPOLOGY_POINT_LIST,
VK_PRIMITIVE_TOPOLOGY_LINE_LIST,
VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST,
VK_PRIMITIVE_TOPOLOGY_LINE_LIST_WITH_ADJACENCY,
VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST_WITH_ADJACENCY,
VK_PRIMITIVE_TOPOLOGY_PATCH_LIST};
return std::find(std::begin(unsupported_topologies), std::end(unsupported_topologies),
topology) == std::end(unsupported_topologies);
}
VkViewportSwizzleNV UnpackViewportSwizzle(u16 swizzle) {
union Swizzle {
u32 raw;
BitField<0, 3, Maxwell::ViewportSwizzle> x;
BitField<4, 3, Maxwell::ViewportSwizzle> y;
BitField<8, 3, Maxwell::ViewportSwizzle> z;
BitField<12, 3, Maxwell::ViewportSwizzle> w;
};
const Swizzle unpacked{swizzle};
return {
.x = MaxwellToVK::ViewportSwizzle(unpacked.x),
.y = MaxwellToVK::ViewportSwizzle(unpacked.y),
.z = MaxwellToVK::ViewportSwizzle(unpacked.z),
.w = MaxwellToVK::ViewportSwizzle(unpacked.w),
};
}
VkSampleCountFlagBits ConvertMsaaMode(Tegra::Texture::MsaaMode msaa_mode) {
switch (msaa_mode) {
case Tegra::Texture::MsaaMode::Msaa1x1:
return VK_SAMPLE_COUNT_1_BIT;
case Tegra::Texture::MsaaMode::Msaa2x1:
case Tegra::Texture::MsaaMode::Msaa2x1_D3D:
return VK_SAMPLE_COUNT_2_BIT;
case Tegra::Texture::MsaaMode::Msaa2x2:
case Tegra::Texture::MsaaMode::Msaa2x2_VC4:
case Tegra::Texture::MsaaMode::Msaa2x2_VC12:
return VK_SAMPLE_COUNT_4_BIT;
case Tegra::Texture::MsaaMode::Msaa4x2:
case Tegra::Texture::MsaaMode::Msaa4x2_D3D:
case Tegra::Texture::MsaaMode::Msaa4x2_VC8:
case Tegra::Texture::MsaaMode::Msaa4x2_VC24:
return VK_SAMPLE_COUNT_8_BIT;
case Tegra::Texture::MsaaMode::Msaa4x4:
return VK_SAMPLE_COUNT_16_BIT;
default:
UNREACHABLE_MSG("Invalid msaa_mode={}", static_cast<int>(msaa_mode));
return VK_SAMPLE_COUNT_1_BIT;
}
}
} // Anonymous namespace
VKGraphicsPipeline::VKGraphicsPipeline(const Device& device_, VKScheduler& scheduler_,
VKDescriptorPool& descriptor_pool_,
VKUpdateDescriptorQueue& update_descriptor_queue_,
const GraphicsPipelineCacheKey& key,
vk::Span<VkDescriptorSetLayoutBinding> bindings,
const SPIRVProgram& program, u32 num_color_buffers)
: device{device_}, scheduler{scheduler_}, cache_key{key}, hash{cache_key.Hash()},
descriptor_set_layout{CreateDescriptorSetLayout(bindings)},
descriptor_allocator{descriptor_pool_, *descriptor_set_layout},
update_descriptor_queue{update_descriptor_queue_}, layout{CreatePipelineLayout()},
descriptor_template{CreateDescriptorUpdateTemplate(program)},
modules(CreateShaderModules(program)),
pipeline(CreatePipeline(program, cache_key.renderpass, num_color_buffers)) {}
VKGraphicsPipeline::~VKGraphicsPipeline() = default;
VkDescriptorSet VKGraphicsPipeline::CommitDescriptorSet() {
if (!descriptor_template) {
return {};
}
const VkDescriptorSet set = descriptor_allocator.Commit();
update_descriptor_queue.Send(*descriptor_template, set);
return set;
}
vk::DescriptorSetLayout VKGraphicsPipeline::CreateDescriptorSetLayout(
vk::Span<VkDescriptorSetLayoutBinding> bindings) const {
const VkDescriptorSetLayoutCreateInfo ci{
.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.bindingCount = bindings.size(),
.pBindings = bindings.data(),
};
return device.GetLogical().CreateDescriptorSetLayout(ci);
}
vk::PipelineLayout VKGraphicsPipeline::CreatePipelineLayout() const {
const VkPipelineLayoutCreateInfo ci{
.sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.setLayoutCount = 1,
.pSetLayouts = descriptor_set_layout.address(),
.pushConstantRangeCount = 0,
.pPushConstantRanges = nullptr,
};
return device.GetLogical().CreatePipelineLayout(ci);
}
vk::DescriptorUpdateTemplateKHR VKGraphicsPipeline::CreateDescriptorUpdateTemplate(
const SPIRVProgram& program) const {
std::vector<VkDescriptorUpdateTemplateEntry> template_entries;
u32 binding = 0;
u32 offset = 0;
for (const auto& stage : program) {
if (stage) {
FillDescriptorUpdateTemplateEntries(stage->entries, binding, offset, template_entries);
}
}
if (template_entries.empty()) {
// If the shader doesn't use descriptor sets, skip template creation.
return {};
}
const VkDescriptorUpdateTemplateCreateInfoKHR ci{
.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_UPDATE_TEMPLATE_CREATE_INFO_KHR,
.pNext = nullptr,
.flags = 0,
.descriptorUpdateEntryCount = static_cast<u32>(template_entries.size()),
.pDescriptorUpdateEntries = template_entries.data(),
.templateType = VK_DESCRIPTOR_UPDATE_TEMPLATE_TYPE_DESCRIPTOR_SET_KHR,
.descriptorSetLayout = *descriptor_set_layout,
.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS,
.pipelineLayout = *layout,
.set = DESCRIPTOR_SET,
};
return device.GetLogical().CreateDescriptorUpdateTemplateKHR(ci);
}
std::vector<vk::ShaderModule> VKGraphicsPipeline::CreateShaderModules(
const SPIRVProgram& program) const {
VkShaderModuleCreateInfo ci{
.sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.codeSize = 0,
.pCode = nullptr,
};
std::vector<vk::ShaderModule> shader_modules;
shader_modules.reserve(Maxwell::MaxShaderStage);
for (std::size_t i = 0; i < Maxwell::MaxShaderStage; ++i) {
const auto& stage = program[i];
if (!stage) {
continue;
}
device.SaveShader(stage->code);
ci.codeSize = stage->code.size() * sizeof(u32);
ci.pCode = stage->code.data();
shader_modules.push_back(device.GetLogical().CreateShaderModule(ci));
}
return shader_modules;
}
vk::Pipeline VKGraphicsPipeline::CreatePipeline(const SPIRVProgram& program,
VkRenderPass renderpass,
u32 num_color_buffers) const {
const auto& state = cache_key.fixed_state;
const auto& viewport_swizzles = state.viewport_swizzles;
FixedPipelineState::DynamicState dynamic;
if (device.IsExtExtendedDynamicStateSupported()) {
// Insert dummy values, as long as they are valid they don't matter as extended dynamic
// state is ignored
dynamic.raw1 = 0;
dynamic.raw2 = 0;
dynamic.vertex_strides.fill(0);
} else {
dynamic = state.dynamic_state;
}
std::vector<VkVertexInputBindingDescription> vertex_bindings;
std::vector<VkVertexInputBindingDivisorDescriptionEXT> vertex_binding_divisors;
for (std::size_t index = 0; index < Maxwell::NumVertexArrays; ++index) {
const bool instanced = state.binding_divisors[index] != 0;
const auto rate = instanced ? VK_VERTEX_INPUT_RATE_INSTANCE : VK_VERTEX_INPUT_RATE_VERTEX;
vertex_bindings.push_back({
.binding = static_cast<u32>(index),
.stride = dynamic.vertex_strides[index],
.inputRate = rate,
});
if (instanced) {
vertex_binding_divisors.push_back({
.binding = static_cast<u32>(index),
.divisor = state.binding_divisors[index],
});
}
}
std::vector<VkVertexInputAttributeDescription> vertex_attributes;
const auto& input_attributes = program[0]->entries.attributes;
for (std::size_t index = 0; index < state.attributes.size(); ++index) {
const auto& attribute = state.attributes[index];
if (!attribute.enabled) {
continue;
}
if (!input_attributes.contains(static_cast<u32>(index))) {
// Skip attributes not used by the vertex shaders.
continue;
}
vertex_attributes.push_back({
.location = static_cast<u32>(index),
.binding = attribute.buffer,
.format = MaxwellToVK::VertexFormat(attribute.Type(), attribute.Size()),
.offset = attribute.offset,
});
}
VkPipelineVertexInputStateCreateInfo vertex_input_ci{
.sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.vertexBindingDescriptionCount = static_cast<u32>(vertex_bindings.size()),
.pVertexBindingDescriptions = vertex_bindings.data(),
.vertexAttributeDescriptionCount = static_cast<u32>(vertex_attributes.size()),
.pVertexAttributeDescriptions = vertex_attributes.data(),
};
const VkPipelineVertexInputDivisorStateCreateInfoEXT input_divisor_ci{
.sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_DIVISOR_STATE_CREATE_INFO_EXT,
.pNext = nullptr,
.vertexBindingDivisorCount = static_cast<u32>(vertex_binding_divisors.size()),
.pVertexBindingDivisors = vertex_binding_divisors.data(),
};
if (!vertex_binding_divisors.empty()) {
vertex_input_ci.pNext = &input_divisor_ci;
}
const auto input_assembly_topology = MaxwellToVK::PrimitiveTopology(device, state.topology);
const VkPipelineInputAssemblyStateCreateInfo input_assembly_ci{
.sType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.topology = MaxwellToVK::PrimitiveTopology(device, state.topology),
.primitiveRestartEnable = state.primitive_restart_enable != 0 &&
SupportsPrimitiveRestart(input_assembly_topology),
};
const VkPipelineTessellationStateCreateInfo tessellation_ci{
.sType = VK_STRUCTURE_TYPE_PIPELINE_TESSELLATION_STATE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.patchControlPoints = state.patch_control_points_minus_one.Value() + 1,
};
VkPipelineViewportStateCreateInfo viewport_ci{
.sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.viewportCount = Maxwell::NumViewports,
.pViewports = nullptr,
.scissorCount = Maxwell::NumViewports,
.pScissors = nullptr,
};
std::array<VkViewportSwizzleNV, Maxwell::NumViewports> swizzles;
std::ranges::transform(viewport_swizzles, swizzles.begin(), UnpackViewportSwizzle);
VkPipelineViewportSwizzleStateCreateInfoNV swizzle_ci{
.sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_SWIZZLE_STATE_CREATE_INFO_NV,
.pNext = nullptr,
.flags = 0,
.viewportCount = Maxwell::NumViewports,
.pViewportSwizzles = swizzles.data(),
};
if (device.IsNvViewportSwizzleSupported()) {
viewport_ci.pNext = &swizzle_ci;
}
const VkPipelineRasterizationStateCreateInfo rasterization_ci{
.sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.depthClampEnable =
static_cast<VkBool32>(state.depth_clamp_disabled == 0 ? VK_TRUE : VK_FALSE),
.rasterizerDiscardEnable =
static_cast<VkBool32>(state.rasterize_enable == 0 ? VK_TRUE : VK_FALSE),
.polygonMode = VK_POLYGON_MODE_FILL,
.cullMode = static_cast<VkCullModeFlags>(
dynamic.cull_enable ? MaxwellToVK::CullFace(dynamic.CullFace()) : VK_CULL_MODE_NONE),
.frontFace = MaxwellToVK::FrontFace(dynamic.FrontFace()),
.depthBiasEnable = state.depth_bias_enable,
.depthBiasConstantFactor = 0.0f,
.depthBiasClamp = 0.0f,
.depthBiasSlopeFactor = 0.0f,
.lineWidth = 1.0f,
};
const VkPipelineMultisampleStateCreateInfo multisample_ci{
.sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.rasterizationSamples = ConvertMsaaMode(state.msaa_mode),
.sampleShadingEnable = VK_FALSE,
.minSampleShading = 0.0f,
.pSampleMask = nullptr,
.alphaToCoverageEnable = VK_FALSE,
.alphaToOneEnable = VK_FALSE,
};
const VkPipelineDepthStencilStateCreateInfo depth_stencil_ci{
.sType = VK_STRUCTURE_TYPE_PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.depthTestEnable = dynamic.depth_test_enable,
.depthWriteEnable = dynamic.depth_write_enable,
.depthCompareOp = dynamic.depth_test_enable
? MaxwellToVK::ComparisonOp(dynamic.DepthTestFunc())
: VK_COMPARE_OP_ALWAYS,
.depthBoundsTestEnable = dynamic.depth_bounds_enable,
.stencilTestEnable = dynamic.stencil_enable,
.front = GetStencilFaceState(dynamic.front),
.back = GetStencilFaceState(dynamic.back),
.minDepthBounds = 0.0f,
.maxDepthBounds = 0.0f,
};
std::array<VkPipelineColorBlendAttachmentState, Maxwell::NumRenderTargets> cb_attachments;
for (std::size_t index = 0; index < num_color_buffers; ++index) {
static constexpr std::array COMPONENT_TABLE{
VK_COLOR_COMPONENT_R_BIT,
VK_COLOR_COMPONENT_G_BIT,
VK_COLOR_COMPONENT_B_BIT,
VK_COLOR_COMPONENT_A_BIT,
};
const auto& blend = state.attachments[index];
VkColorComponentFlags color_components = 0;
for (std::size_t i = 0; i < COMPONENT_TABLE.size(); ++i) {
if (blend.Mask()[i]) {
color_components |= COMPONENT_TABLE[i];
}
}
cb_attachments[index] = {
.blendEnable = blend.enable != 0,
.srcColorBlendFactor = MaxwellToVK::BlendFactor(blend.SourceRGBFactor()),
.dstColorBlendFactor = MaxwellToVK::BlendFactor(blend.DestRGBFactor()),
.colorBlendOp = MaxwellToVK::BlendEquation(blend.EquationRGB()),
.srcAlphaBlendFactor = MaxwellToVK::BlendFactor(blend.SourceAlphaFactor()),
.dstAlphaBlendFactor = MaxwellToVK::BlendFactor(blend.DestAlphaFactor()),
.alphaBlendOp = MaxwellToVK::BlendEquation(blend.EquationAlpha()),
.colorWriteMask = color_components,
};
}
const VkPipelineColorBlendStateCreateInfo color_blend_ci{
.sType = VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.logicOpEnable = VK_FALSE,
.logicOp = VK_LOGIC_OP_COPY,
.attachmentCount = num_color_buffers,
.pAttachments = cb_attachments.data(),
.blendConstants = {},
};
std::vector dynamic_states{
VK_DYNAMIC_STATE_VIEWPORT, VK_DYNAMIC_STATE_SCISSOR,
VK_DYNAMIC_STATE_DEPTH_BIAS, VK_DYNAMIC_STATE_BLEND_CONSTANTS,
VK_DYNAMIC_STATE_DEPTH_BOUNDS, VK_DYNAMIC_STATE_STENCIL_COMPARE_MASK,
VK_DYNAMIC_STATE_STENCIL_WRITE_MASK, VK_DYNAMIC_STATE_STENCIL_REFERENCE,
};
if (device.IsExtExtendedDynamicStateSupported()) {
static constexpr std::array extended{
VK_DYNAMIC_STATE_CULL_MODE_EXT,
VK_DYNAMIC_STATE_FRONT_FACE_EXT,
VK_DYNAMIC_STATE_VERTEX_INPUT_BINDING_STRIDE_EXT,
VK_DYNAMIC_STATE_DEPTH_TEST_ENABLE_EXT,
VK_DYNAMIC_STATE_DEPTH_WRITE_ENABLE_EXT,
VK_DYNAMIC_STATE_DEPTH_COMPARE_OP_EXT,
VK_DYNAMIC_STATE_DEPTH_BOUNDS_TEST_ENABLE_EXT,
VK_DYNAMIC_STATE_STENCIL_TEST_ENABLE_EXT,
VK_DYNAMIC_STATE_STENCIL_OP_EXT,
};
dynamic_states.insert(dynamic_states.end(), extended.begin(), extended.end());
}
const VkPipelineDynamicStateCreateInfo dynamic_state_ci{
.sType = VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.dynamicStateCount = static_cast<u32>(dynamic_states.size()),
.pDynamicStates = dynamic_states.data(),
};
const VkPipelineShaderStageRequiredSubgroupSizeCreateInfoEXT subgroup_size_ci{
.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_REQUIRED_SUBGROUP_SIZE_CREATE_INFO_EXT,
.pNext = nullptr,
.requiredSubgroupSize = GuestWarpSize,
};
std::vector<VkPipelineShaderStageCreateInfo> shader_stages;
std::size_t module_index = 0;
for (std::size_t stage = 0; stage < Maxwell::MaxShaderStage; ++stage) {
if (!program[stage]) {
continue;
}
VkPipelineShaderStageCreateInfo& stage_ci = shader_stages.emplace_back();
stage_ci.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO;
stage_ci.pNext = nullptr;
stage_ci.flags = 0;
stage_ci.stage = MaxwellToVK::ShaderStage(static_cast<Tegra::Engines::ShaderType>(stage));
stage_ci.module = *modules[module_index++];
stage_ci.pName = "main";
stage_ci.pSpecializationInfo = nullptr;
if (program[stage]->entries.uses_warps && device.IsGuestWarpSizeSupported(stage_ci.stage)) {
stage_ci.pNext = &subgroup_size_ci;
}
}
return device.GetLogical().CreateGraphicsPipeline(VkGraphicsPipelineCreateInfo{
.sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.stageCount = static_cast<u32>(shader_stages.size()),
.pStages = shader_stages.data(),
.pVertexInputState = &vertex_input_ci,
.pInputAssemblyState = &input_assembly_ci,
.pTessellationState = &tessellation_ci,
.pViewportState = &viewport_ci,
.pRasterizationState = &rasterization_ci,
.pMultisampleState = &multisample_ci,
.pDepthStencilState = &depth_stencil_ci,
.pColorBlendState = &color_blend_ci,
.pDynamicState = &dynamic_state_ci,
.layout = *layout,
.renderPass = renderpass,
.subpass = 0,
.basePipelineHandle = nullptr,
.basePipelineIndex = 0,
});
}
} // namespace Vulkan

103
src/video_core/renderer_vulkan/vk_graphics_pipeline.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 <array>
#include <optional>
#include <vector>
#include "common/common_types.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/renderer_vulkan/fixed_pipeline_state.h"
#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
#include "video_core/renderer_vulkan/vk_shader_decompiler.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
namespace Vulkan {
using Maxwell = Tegra::Engines::Maxwell3D::Regs;
struct GraphicsPipelineCacheKey {
VkRenderPass renderpass;
std::array<GPUVAddr, Maxwell::MaxShaderProgram> shaders;
FixedPipelineState fixed_state;
std::size_t Hash() const noexcept;
bool operator==(const GraphicsPipelineCacheKey& rhs) const noexcept;
bool operator!=(const GraphicsPipelineCacheKey& rhs) const noexcept {
return !operator==(rhs);
}
std::size_t Size() const noexcept {
return sizeof(renderpass) + sizeof(shaders) + fixed_state.Size();
}
};
static_assert(std::has_unique_object_representations_v<GraphicsPipelineCacheKey>);
static_assert(std::is_trivially_copyable_v<GraphicsPipelineCacheKey>);
static_assert(std::is_trivially_constructible_v<GraphicsPipelineCacheKey>);
class Device;
class VKDescriptorPool;
class VKScheduler;
class VKUpdateDescriptorQueue;
using SPIRVProgram = std::array<std::optional<SPIRVShader>, Maxwell::MaxShaderStage>;
class VKGraphicsPipeline final {
public:
explicit VKGraphicsPipeline(const Device& device_, VKScheduler& scheduler_,
VKDescriptorPool& descriptor_pool,
VKUpdateDescriptorQueue& update_descriptor_queue_,
const GraphicsPipelineCacheKey& key,
vk::Span<VkDescriptorSetLayoutBinding> bindings,
const SPIRVProgram& program, u32 num_color_buffers);
~VKGraphicsPipeline();
VkDescriptorSet CommitDescriptorSet();
VkPipeline GetHandle() const {
return *pipeline;
}
VkPipelineLayout GetLayout() const {
return *layout;
}
GraphicsPipelineCacheKey GetCacheKey() const {
return cache_key;
}
private:
vk::DescriptorSetLayout CreateDescriptorSetLayout(
vk::Span<VkDescriptorSetLayoutBinding> bindings) const;
vk::PipelineLayout CreatePipelineLayout() const;
vk::DescriptorUpdateTemplateKHR CreateDescriptorUpdateTemplate(
const SPIRVProgram& program) const;
std::vector<vk::ShaderModule> CreateShaderModules(const SPIRVProgram& program) const;
vk::Pipeline CreatePipeline(const SPIRVProgram& program, VkRenderPass renderpass,
u32 num_color_buffers) const;
const Device& device;
VKScheduler& scheduler;
const GraphicsPipelineCacheKey cache_key;
const u64 hash;
vk::DescriptorSetLayout descriptor_set_layout;
DescriptorAllocator descriptor_allocator;
VKUpdateDescriptorQueue& update_descriptor_queue;
vk::PipelineLayout layout;
vk::DescriptorUpdateTemplateKHR descriptor_template;
std::vector<vk::ShaderModule> modules;
vk::Pipeline pipeline;
};
} // namespace Vulkan

375
src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
View file

@ -19,49 +19,27 @@
#include "video_core/renderer_vulkan/maxwell_to_vk.h"
#include "video_core/renderer_vulkan/vk_compute_pipeline.h"
#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
#include "video_core/renderer_vulkan/vk_graphics_pipeline.h"
#include "video_core/renderer_vulkan/vk_pipeline_cache.h"
#include "video_core/renderer_vulkan/vk_rasterizer.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_update_descriptor.h"
#include "video_core/shader/compiler_settings.h"
#include "video_core/shader/memory_util.h"
#include "video_core/shader_cache.h"
#include "video_core/shader_notify.h"
#include "video_core/vulkan_common/vulkan_device.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
namespace Vulkan {
MICROPROFILE_DECLARE(Vulkan_PipelineCache);
using Tegra::Engines::ShaderType;
using VideoCommon::Shader::GetShaderAddress;
using VideoCommon::Shader::GetShaderCode;
using VideoCommon::Shader::KERNEL_MAIN_OFFSET;
using VideoCommon::Shader::ProgramCode;
using VideoCommon::Shader::STAGE_MAIN_OFFSET;
namespace {
constexpr VkDescriptorType UNIFORM_BUFFER = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
constexpr VkDescriptorType STORAGE_BUFFER = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER;
constexpr VkDescriptorType UNIFORM_TEXEL_BUFFER = VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER;
constexpr VkDescriptorType COMBINED_IMAGE_SAMPLER = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER;
constexpr VkDescriptorType STORAGE_TEXEL_BUFFER = VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER;
constexpr VkDescriptorType STORAGE_IMAGE = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE;
constexpr VideoCommon::Shader::CompilerSettings compiler_settings{
.depth = VideoCommon::Shader::CompileDepth::FullDecompile,
.disable_else_derivation = true,
};
constexpr std::size_t GetStageFromProgram(std::size_t program) {
size_t StageFromProgram(size_t program) {
return program == 0 ? 0 : program - 1;
}
constexpr ShaderType GetStageFromProgram(Maxwell::ShaderProgram program) {
return static_cast<ShaderType>(GetStageFromProgram(static_cast<std::size_t>(program)));
ShaderType StageFromProgram(Maxwell::ShaderProgram program) {
return static_cast<ShaderType>(StageFromProgram(static_cast<size_t>(program)));
}
ShaderType GetShaderType(Maxwell::ShaderProgram program) {
@ -81,165 +59,35 @@ ShaderType GetShaderType(Maxwell::ShaderProgram program) {
return ShaderType::Vertex;
}
}
template <VkDescriptorType descriptor_type, class Container>
void AddBindings(std::vector<VkDescriptorSetLayoutBinding>& bindings, u32& binding,
VkShaderStageFlags stage_flags, const Container& container) {
const u32 num_entries = static_cast<u32>(std::size(container));
for (std::size_t i = 0; i < num_entries; ++i) {
u32 count = 1;
if constexpr (descriptor_type == VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER) {
// Combined image samplers can be arrayed.
count = container[i].size;
}
bindings.push_back({
.binding = binding++,
.descriptorType = descriptor_type,
.descriptorCount = count,
.stageFlags = stage_flags,
.pImmutableSamplers = nullptr,
});
}
}
u32 FillDescriptorLayout(const ShaderEntries& entries,
std::vector<VkDescriptorSetLayoutBinding>& bindings,
Maxwell::ShaderProgram program_type, u32 base_binding) {
const ShaderType stage = GetStageFromProgram(program_type);
const VkShaderStageFlags flags = MaxwellToVK::ShaderStage(stage);
u32 binding = base_binding;
AddBindings<UNIFORM_BUFFER>(bindings, binding, flags, entries.const_buffers);
AddBindings<STORAGE_BUFFER>(bindings, binding, flags, entries.global_buffers);
AddBindings<UNIFORM_TEXEL_BUFFER>(bindings, binding, flags, entries.uniform_texels);
AddBindings<COMBINED_IMAGE_SAMPLER>(bindings, binding, flags, entries.samplers);
AddBindings<STORAGE_TEXEL_BUFFER>(bindings, binding, flags, entries.storage_texels);
AddBindings<STORAGE_IMAGE>(bindings, binding, flags, entries.images);
return binding;
}
} // Anonymous namespace
std::size_t GraphicsPipelineCacheKey::Hash() const noexcept {
const u64 hash = Common::CityHash64(reinterpret_cast<const char*>(this), Size());
return static_cast<std::size_t>(hash);
}
bool GraphicsPipelineCacheKey::operator==(const GraphicsPipelineCacheKey& rhs) const noexcept {
return std::memcmp(&rhs, this, Size()) == 0;
}
std::size_t ComputePipelineCacheKey::Hash() const noexcept {
size_t ComputePipelineCacheKey::Hash() const noexcept {
const u64 hash = Common::CityHash64(reinterpret_cast<const char*>(this), sizeof *this);
return static_cast<std::size_t>(hash);
return static_cast<size_t>(hash);
}
bool ComputePipelineCacheKey::operator==(const ComputePipelineCacheKey& rhs) const noexcept {
return std::memcmp(&rhs, this, sizeof *this) == 0;
}
Shader::Shader(Tegra::Engines::ConstBufferEngineInterface& engine_, ShaderType stage_,
GPUVAddr gpu_addr_, VAddr cpu_addr_, ProgramCode program_code_, u32 main_offset_)
: gpu_addr(gpu_addr_), program_code(std::move(program_code_)), registry(stage_, engine_),
shader_ir(program_code, main_offset_, compiler_settings, registry),
entries(GenerateShaderEntries(shader_ir)) {}
Shader::Shader() = default;
Shader::~Shader() = default;
VKPipelineCache::VKPipelineCache(RasterizerVulkan& rasterizer_, Tegra::GPU& gpu_,
Tegra::Engines::Maxwell3D& maxwell3d_,
Tegra::Engines::KeplerCompute& kepler_compute_,
Tegra::MemoryManager& gpu_memory_, const Device& device_,
VKScheduler& scheduler_, VKDescriptorPool& descriptor_pool_,
VKUpdateDescriptorQueue& update_descriptor_queue_)
PipelineCache::PipelineCache(RasterizerVulkan& rasterizer_, Tegra::GPU& gpu_,
Tegra::Engines::Maxwell3D& maxwell3d_,
Tegra::Engines::KeplerCompute& kepler_compute_,
Tegra::MemoryManager& gpu_memory_, const Device& device_,
VKScheduler& scheduler_, VKDescriptorPool& descriptor_pool_,
VKUpdateDescriptorQueue& update_descriptor_queue_)
: VideoCommon::ShaderCache<Shader>{rasterizer_}, gpu{gpu_}, maxwell3d{maxwell3d_},
kepler_compute{kepler_compute_}, gpu_memory{gpu_memory_}, device{device_},
scheduler{scheduler_}, descriptor_pool{descriptor_pool_}, update_descriptor_queue{
update_descriptor_queue_} {}
VKPipelineCache::~VKPipelineCache() = default;
PipelineCache::~PipelineCache() = default;
std::array<Shader*, Maxwell::MaxShaderProgram> VKPipelineCache::GetShaders() {
std::array<Shader*, Maxwell::MaxShaderProgram> shaders{};
for (std::size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) {
const auto program{static_cast<Maxwell::ShaderProgram>(index)};
// Skip stages that are not enabled
if (!maxwell3d.regs.IsShaderConfigEnabled(index)) {
continue;
}
const GPUVAddr gpu_addr{GetShaderAddress(maxwell3d, program)};
const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr);
ASSERT(cpu_addr);
Shader* result = cpu_addr ? TryGet(*cpu_addr) : null_shader.get();
if (!result) {
const u8* const host_ptr{gpu_memory.GetPointer(gpu_addr)};
// No shader found - create a new one
static constexpr u32 stage_offset = STAGE_MAIN_OFFSET;
const auto stage = static_cast<ShaderType>(index == 0 ? 0 : index - 1);
ProgramCode code = GetShaderCode(gpu_memory, gpu_addr, host_ptr, false);
const std::size_t size_in_bytes = code.size() * sizeof(u64);
auto shader = std::make_unique<Shader>(maxwell3d, stage, gpu_addr, *cpu_addr,
std::move(code), stage_offset);
result = shader.get();
if (cpu_addr) {
Register(std::move(shader), *cpu_addr, size_in_bytes);
} else {
null_shader = std::move(shader);
}
}
shaders[index] = result;
}
return last_shaders = shaders;
}
VKGraphicsPipeline* VKPipelineCache::GetGraphicsPipeline(
const GraphicsPipelineCacheKey& key, u32 num_color_buffers,
VideoCommon::Shader::AsyncShaders& async_shaders) {
MICROPROFILE_SCOPE(Vulkan_PipelineCache);
if (last_graphics_pipeline && last_graphics_key == key) {
return last_graphics_pipeline;
}
last_graphics_key = key;
if (device.UseAsynchronousShaders() && async_shaders.IsShaderAsync(gpu)) {
std::unique_lock lock{pipeline_cache};
const auto [pair, is_cache_miss] = graphics_cache.try_emplace(key);
if (is_cache_miss) {
gpu.ShaderNotify().MarkSharderBuilding();
LOG_INFO(Render_Vulkan, "Compile 0x{:016X}", key.Hash());
const auto [program, bindings] = DecompileShaders(key.fixed_state);
async_shaders.QueueVulkanShader(this, device, scheduler, descriptor_pool,
update_descriptor_queue, bindings, program, key,
num_color_buffers);
}
last_graphics_pipeline = pair->second.get();
return last_graphics_pipeline;
}
const auto [pair, is_cache_miss] = graphics_cache.try_emplace(key);
auto& entry = pair->second;
if (is_cache_miss) {
gpu.ShaderNotify().MarkSharderBuilding();
LOG_INFO(Render_Vulkan, "Compile 0x{:016X}", key.Hash());
const auto [program, bindings] = DecompileShaders(key.fixed_state);
entry = std::make_unique<VKGraphicsPipeline>(device, scheduler, descriptor_pool,
update_descriptor_queue, key, bindings,
program, num_color_buffers);
gpu.ShaderNotify().MarkShaderComplete();
}
last_graphics_pipeline = entry.get();
return last_graphics_pipeline;
}
VKComputePipeline& VKPipelineCache::GetComputePipeline(const ComputePipelineCacheKey& key) {
ComputePipeline& PipelineCache::GetComputePipeline(const ComputePipelineCacheKey& key) {
MICROPROFILE_SCOPE(Vulkan_PipelineCache);
const auto [pair, is_cache_miss] = compute_cache.try_emplace(key);
@ -248,200 +96,9 @@ VKComputePipeline& VKPipelineCache::GetComputePipeline(const ComputePipelineCach
return *entry;
}
LOG_INFO(Render_Vulkan, "Compile 0x{:016X}", key.Hash());
const GPUVAddr gpu_addr = key.shader;
const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr);
ASSERT(cpu_addr);
Shader* shader = cpu_addr ? TryGet(*cpu_addr) : null_kernel.get();
if (!shader) {
// No shader found - create a new one
const auto host_ptr = gpu_memory.GetPointer(gpu_addr);
ProgramCode code = GetShaderCode(gpu_memory, gpu_addr, host_ptr, true);
const std::size_t size_in_bytes = code.size() * sizeof(u64);
auto shader_info = std::make_unique<Shader>(kepler_compute, ShaderType::Compute, gpu_addr,
*cpu_addr, std::move(code), KERNEL_MAIN_OFFSET);
shader = shader_info.get();
if (cpu_addr) {
Register(std::move(shader_info), *cpu_addr, size_in_bytes);
} else {
null_kernel = std::move(shader_info);
}
}
const Specialization specialization{
.base_binding = 0,
.workgroup_size = key.workgroup_size,
.shared_memory_size = key.shared_memory_size,
.point_size = std::nullopt,
.enabled_attributes = {},
.attribute_types = {},
.ndc_minus_one_to_one = false,
};
const SPIRVShader spirv_shader{Decompile(device, shader->GetIR(), ShaderType::Compute,
shader->GetRegistry(), specialization),
shader->GetEntries()};
entry = std::make_unique<VKComputePipeline>(device, scheduler, descriptor_pool,
update_descriptor_queue, spirv_shader);
return *entry;
throw "Bad";
}
void VKPipelineCache::EmplacePipeline(std::unique_ptr<VKGraphicsPipeline> pipeline) {
gpu.ShaderNotify().MarkShaderComplete();
std::unique_lock lock{pipeline_cache};
graphics_cache.at(pipeline->GetCacheKey()) = std::move(pipeline);
}
void VKPipelineCache::OnShaderRemoval(Shader* shader) {
bool finished = false;
const auto Finish = [&] {
// TODO(Rodrigo): Instead of finishing here, wait for the fences that use this pipeline and
// flush.
if (finished) {
return;
}
finished = true;
scheduler.Finish();
};
const GPUVAddr invalidated_addr = shader->GetGpuAddr();
for (auto it = graphics_cache.begin(); it != graphics_cache.end();) {
auto& entry = it->first;
if (std::find(entry.shaders.begin(), entry.shaders.end(), invalidated_addr) ==
entry.shaders.end()) {
++it;
continue;
}
Finish();
it = graphics_cache.erase(it);
}
for (auto it = compute_cache.begin(); it != compute_cache.end();) {
auto& entry = it->first;
if (entry.shader != invalidated_addr) {
++it;
continue;
}
Finish();
it = compute_cache.erase(it);
}
}
std::pair<SPIRVProgram, std::vector<VkDescriptorSetLayoutBinding>>
VKPipelineCache::DecompileShaders(const FixedPipelineState& fixed_state) {
Specialization specialization;
if (fixed_state.topology == Maxwell::PrimitiveTopology::Points) {
float point_size;
std::memcpy(&point_size, &fixed_state.point_size, sizeof(float));
specialization.point_size = point_size;
ASSERT(point_size != 0.0f);
}
for (std::size_t i = 0; i < Maxwell::NumVertexAttributes; ++i) {
const auto& attribute = fixed_state.attributes[i];
specialization.enabled_attributes[i] = attribute.enabled.Value() != 0;
specialization.attribute_types[i] = attribute.Type();
}
specialization.ndc_minus_one_to_one = fixed_state.ndc_minus_one_to_one;
specialization.early_fragment_tests = fixed_state.early_z;
// Alpha test
specialization.alpha_test_func =
FixedPipelineState::UnpackComparisonOp(fixed_state.alpha_test_func.Value());
specialization.alpha_test_ref = Common::BitCast<float>(fixed_state.alpha_test_ref);
SPIRVProgram program;
std::vector<VkDescriptorSetLayoutBinding> bindings;
for (std::size_t index = 1; index < Maxwell::MaxShaderProgram; ++index) {
const auto program_enum = static_cast<Maxwell::ShaderProgram>(index);
// Skip stages that are not enabled
if (!maxwell3d.regs.IsShaderConfigEnabled(index)) {
continue;
}
const GPUVAddr gpu_addr = GetShaderAddress(maxwell3d, program_enum);
const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr);
Shader* const shader = cpu_addr ? TryGet(*cpu_addr) : null_shader.get();
const std::size_t stage = index == 0 ? 0 : index - 1; // Stage indices are 0 - 5
const ShaderType program_type = GetShaderType(program_enum);
const auto& entries = shader->GetEntries();
program[stage] = {
Decompile(device, shader->GetIR(), program_type, shader->GetRegistry(), specialization),
entries,
};
const u32 old_binding = specialization.base_binding;
specialization.base_binding =
FillDescriptorLayout(entries, bindings, program_enum, specialization.base_binding);
ASSERT(old_binding + entries.NumBindings() == specialization.base_binding);
}
return {std::move(program), std::move(bindings)};
}
template <VkDescriptorType descriptor_type, class Container>
void AddEntry(std::vector<VkDescriptorUpdateTemplateEntry>& template_entries, u32& binding,
u32& offset, const Container& container) {
static constexpr u32 entry_size = static_cast<u32>(sizeof(DescriptorUpdateEntry));
const u32 count = static_cast<u32>(std::size(container));
if constexpr (descriptor_type == COMBINED_IMAGE_SAMPLER) {
for (u32 i = 0; i < count; ++i) {
const u32 num_samplers = container[i].size;
template_entries.push_back({
.dstBinding = binding,
.dstArrayElement = 0,
.descriptorCount = num_samplers,
.descriptorType = descriptor_type,
.offset = offset,
.stride = entry_size,
});
++binding;
offset += num_samplers * entry_size;
}
return;
}
if constexpr (descriptor_type == UNIFORM_TEXEL_BUFFER ||
descriptor_type == STORAGE_TEXEL_BUFFER) {
// Nvidia has a bug where updating multiple texels at once causes the driver to crash.
// Note: Fixed in driver Windows 443.24, Linux 440.66.15
for (u32 i = 0; i < count; ++i) {
template_entries.push_back({
.dstBinding = binding + i,
.dstArrayElement = 0,
.descriptorCount = 1,
.descriptorType = descriptor_type,
.offset = static_cast<std::size_t>(offset + i * entry_size),
.stride = entry_size,
});
}
} else if (count > 0) {
template_entries.push_back({
.dstBinding = binding,
.dstArrayElement = 0,
.descriptorCount = count,
.descriptorType = descriptor_type,
.offset = offset,
.stride = entry_size,
});
}
offset += count * entry_size;
binding += count;
}
void FillDescriptorUpdateTemplateEntries(
const ShaderEntries& entries, u32& binding, u32& offset,
std::vector<VkDescriptorUpdateTemplateEntryKHR>& template_entries) {
AddEntry<UNIFORM_BUFFER>(template_entries, offset, binding, entries.const_buffers);
AddEntry<STORAGE_BUFFER>(template_entries, offset, binding, entries.global_buffers);
AddEntry<UNIFORM_TEXEL_BUFFER>(template_entries, offset, binding, entries.uniform_texels);
AddEntry<COMBINED_IMAGE_SAMPLER>(template_entries, offset, binding, entries.samplers);
AddEntry<STORAGE_TEXEL_BUFFER>(template_entries, offset, binding, entries.storage_texels);
AddEntry<STORAGE_IMAGE>(template_entries, offset, binding, entries.images);
}
void PipelineCache::OnShaderRemoval(Shader*) {}
} // namespace Vulkan

91
src/video_core/renderer_vulkan/vk_pipeline_cache.h
View file

@ -15,15 +15,8 @@
#include <boost/functional/hash.hpp>
#include "common/common_types.h"
#include "video_core/engines/const_buffer_engine_interface.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/renderer_vulkan/fixed_pipeline_state.h"
#include "video_core/renderer_vulkan/vk_graphics_pipeline.h"
#include "video_core/renderer_vulkan/vk_shader_decompiler.h"
#include "video_core/shader/async_shaders.h"
#include "video_core/shader/memory_util.h"
#include "video_core/shader/registry.h"
#include "video_core/shader/shader_ir.h"
#include "video_core/shader_cache.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
@ -35,7 +28,7 @@ namespace Vulkan {
class Device;
class RasterizerVulkan;
class VKComputePipeline;
class ComputePipeline;
class VKDescriptorPool;
class VKScheduler;
class VKUpdateDescriptorQueue;
@ -47,7 +40,7 @@ struct ComputePipelineCacheKey {
u32 shared_memory_size;
std::array<u32, 3> workgroup_size;
std::size_t Hash() const noexcept;
size_t Hash() const noexcept;
bool operator==(const ComputePipelineCacheKey& rhs) const noexcept;
@ -63,16 +56,9 @@ static_assert(std::is_trivially_constructible_v<ComputePipelineCacheKey>);
namespace std {
template <>
struct hash<Vulkan::GraphicsPipelineCacheKey> {
std::size_t operator()(const Vulkan::GraphicsPipelineCacheKey& k) const noexcept {
return k.Hash();
}
};
template <>
struct hash<Vulkan::ComputePipelineCacheKey> {
std::size_t operator()(const Vulkan::ComputePipelineCacheKey& k) const noexcept {
size_t operator()(const Vulkan::ComputePipelineCacheKey& k) const noexcept {
return k.Hash();
}
};
@ -83,66 +69,26 @@ namespace Vulkan {
class Shader {
public:
explicit Shader(Tegra::Engines::ConstBufferEngineInterface& engine_,
Tegra::Engines::ShaderType stage_, GPUVAddr gpu_addr, VAddr cpu_addr_,
VideoCommon::Shader::ProgramCode program_code, u32 main_offset_);
explicit Shader();
~Shader();
GPUVAddr GetGpuAddr() const {
return gpu_addr;
}
VideoCommon::Shader::ShaderIR& GetIR() {
return shader_ir;
}
const VideoCommon::Shader::ShaderIR& GetIR() const {
return shader_ir;
}
const VideoCommon::Shader::Registry& GetRegistry() const {
return registry;
}
const ShaderEntries& GetEntries() const {
return entries;
}
private:
GPUVAddr gpu_addr{};
VideoCommon::Shader::ProgramCode program_code;
VideoCommon::Shader::Registry registry;
VideoCommon::Shader::ShaderIR shader_ir;
ShaderEntries entries;
};
class VKPipelineCache final : public VideoCommon::ShaderCache<Shader> {
class PipelineCache final : public VideoCommon::ShaderCache<Shader> {
public:
explicit VKPipelineCache(RasterizerVulkan& rasterizer, Tegra::GPU& gpu,
Tegra::Engines::Maxwell3D& maxwell3d,
Tegra::Engines::KeplerCompute& kepler_compute,
Tegra::MemoryManager& gpu_memory, const Device& device,
VKScheduler& scheduler, VKDescriptorPool& descriptor_pool,
VKUpdateDescriptorQueue& update_descriptor_queue);
~VKPipelineCache() override;
explicit PipelineCache(RasterizerVulkan& rasterizer, Tegra::GPU& gpu,
Tegra::Engines::Maxwell3D& maxwell3d,
Tegra::Engines::KeplerCompute& kepler_compute,
Tegra::MemoryManager& gpu_memory, const Device& device,
VKScheduler& scheduler, VKDescriptorPool& descriptor_pool,
VKUpdateDescriptorQueue& update_descriptor_queue);
~PipelineCache() override;
std::array<Shader*, Maxwell::MaxShaderProgram> GetShaders();
VKGraphicsPipeline* GetGraphicsPipeline(const GraphicsPipelineCacheKey& key,
u32 num_color_buffers,
VideoCommon::Shader::AsyncShaders& async_shaders);
VKComputePipeline& GetComputePipeline(const ComputePipelineCacheKey& key);
void EmplacePipeline(std::unique_ptr<VKGraphicsPipeline> pipeline);
ComputePipeline& GetComputePipeline(const ComputePipelineCacheKey& key);
protected:
void OnShaderRemoval(Shader* shader) final;
private:
std::pair<SPIRVProgram, std::vector<VkDescriptorSetLayoutBinding>> DecompileShaders(
const FixedPipelineState& fixed_state);
Tegra::GPU& gpu;
Tegra::Engines::Maxwell3D& maxwell3d;
Tegra::Engines::KeplerCompute& kepler_compute;
@ -158,17 +104,8 @@ private:
std::array<Shader*, Maxwell::MaxShaderProgram> last_shaders{};
GraphicsPipelineCacheKey last_graphics_key;
VKGraphicsPipeline* last_graphics_pipeline = nullptr;
std::mutex pipeline_cache;
std::unordered_map<GraphicsPipelineCacheKey, std::unique_ptr<VKGraphicsPipeline>>
graphics_cache;
std::unordered_map<ComputePipelineCacheKey, std::unique_ptr<VKComputePipeline>> compute_cache;
std::unordered_map<ComputePipelineCacheKey, std::unique_ptr<ComputePipeline>> compute_cache;
};
void FillDescriptorUpdateTemplateEntries(
const ShaderEntries& entries, u32& binding, u32& offset,
std::vector<VkDescriptorUpdateTemplateEntryKHR>& template_entries);
} // namespace Vulkan

361
src/video_core/renderer_vulkan/vk_rasterizer.cpp
View file

@ -24,7 +24,6 @@
#include "video_core/renderer_vulkan/vk_buffer_cache.h"
#include "video_core/renderer_vulkan/vk_compute_pipeline.h"
#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
#include "video_core/renderer_vulkan/vk_graphics_pipeline.h"
#include "video_core/renderer_vulkan/vk_pipeline_cache.h"
#include "video_core/renderer_vulkan/vk_rasterizer.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
@ -97,15 +96,6 @@ VkRect2D GetScissorState(const Maxwell& regs, size_t index) {
return scissor;
}
std::array<GPUVAddr, Maxwell::MaxShaderProgram> GetShaderAddresses(
const std::array<Shader*, Maxwell::MaxShaderProgram>& shaders) {
std::array<GPUVAddr, Maxwell::MaxShaderProgram> addresses;
for (size_t i = 0; i < std::size(addresses); ++i) {
addresses[i] = shaders[i] ? shaders[i]->GetGpuAddr() : 0;
}
return addresses;
}
struct TextureHandle {
constexpr TextureHandle(u32 data, bool via_header_index) {
const Tegra::Texture::TextureHandle handle{data};
@ -117,98 +107,6 @@ struct TextureHandle {
u32 sampler;
};
template <typename Engine, typename Entry>
TextureHandle GetTextureInfo(const Engine& engine, bool via_header_index, const Entry& entry,
size_t stage, size_t index = 0) {
const auto shader_type = static_cast<Tegra::Engines::ShaderType>(stage);
if constexpr (std::is_same_v<Entry, SamplerEntry>) {
if (entry.is_separated) {
const u32 buffer_1 = entry.buffer;
const u32 buffer_2 = entry.secondary_buffer;
const u32 offset_1 = entry.offset;
const u32 offset_2 = entry.secondary_offset;
const u32 handle_1 = engine.AccessConstBuffer32(shader_type, buffer_1, offset_1);
const u32 handle_2 = engine.AccessConstBuffer32(shader_type, buffer_2, offset_2);
return TextureHandle(handle_1 | handle_2, via_header_index);
}
}
if (entry.is_bindless) {
const u32 raw = engine.AccessConstBuffer32(shader_type, entry.buffer, entry.offset);
return TextureHandle(raw, via_header_index);
}
const u32 buffer = engine.GetBoundBuffer();
const u64 offset = (entry.offset + index) * sizeof(u32);
return TextureHandle(engine.AccessConstBuffer32(shader_type, buffer, offset), via_header_index);
}
ImageViewType ImageViewTypeFromEntry(const SamplerEntry& entry) {
if (entry.is_buffer) {
return ImageViewType::e2D;
}
switch (entry.type) {
case Tegra::Shader::TextureType::Texture1D:
return entry.is_array ? ImageViewType::e1DArray : ImageViewType::e1D;
case Tegra::Shader::TextureType::Texture2D:
return entry.is_array ? ImageViewType::e2DArray : ImageViewType::e2D;
case Tegra::Shader::TextureType::Texture3D:
return ImageViewType::e3D;
case Tegra::Shader::TextureType::TextureCube:
return entry.is_array ? ImageViewType::CubeArray : ImageViewType::Cube;
}
UNREACHABLE();
return ImageViewType::e2D;
}
ImageViewType ImageViewTypeFromEntry(const ImageEntry& entry) {
switch (entry.type) {
case Tegra::Shader::ImageType::Texture1D:
return ImageViewType::e1D;
case Tegra::Shader::ImageType::Texture1DArray:
return ImageViewType::e1DArray;
case Tegra::Shader::ImageType::Texture2D:
return ImageViewType::e2D;
case Tegra::Shader::ImageType::Texture2DArray:
return ImageViewType::e2DArray;
case Tegra::Shader::ImageType::Texture3D:
return ImageViewType::e3D;
case Tegra::Shader::ImageType::TextureBuffer:
return ImageViewType::Buffer;
}
UNREACHABLE();
return ImageViewType::e2D;
}
void PushImageDescriptors(const ShaderEntries& entries, TextureCache& texture_cache,
VKUpdateDescriptorQueue& update_descriptor_queue,
ImageViewId*& image_view_id_ptr, VkSampler*& sampler_ptr) {
for ([[maybe_unused]] const auto& entry : entries.uniform_texels) {
const ImageViewId image_view_id = *image_view_id_ptr++;
const ImageView& image_view = texture_cache.GetImageView(image_view_id);
update_descriptor_queue.AddTexelBuffer(image_view.BufferView());
}
for (const auto& entry : entries.samplers) {
for (size_t i = 0; i < entry.size; ++i) {
const VkSampler sampler = *sampler_ptr++;
const ImageViewId image_view_id = *image_view_id_ptr++;
const ImageView& image_view = texture_cache.GetImageView(image_view_id);
const VkImageView handle = image_view.Handle(ImageViewTypeFromEntry(entry));
update_descriptor_queue.AddSampledImage(handle, sampler);
}
}
for ([[maybe_unused]] const auto& entry : entries.storage_texels) {
const ImageViewId image_view_id = *image_view_id_ptr++;
const ImageView& image_view = texture_cache.GetImageView(image_view_id);
update_descriptor_queue.AddTexelBuffer(image_view.BufferView());
}
for (const auto& entry : entries.images) {
// TODO: Mark as modified
const ImageViewId image_view_id = *image_view_id_ptr++;
const ImageView& image_view = texture_cache.GetImageView(image_view_id);
const VkImageView handle = image_view.Handle(ImageViewTypeFromEntry(entry));
update_descriptor_queue.AddImage(handle);
}
}
DrawParams MakeDrawParams(const Maxwell& regs, u32 num_instances, bool is_instanced,
bool is_indexed) {
DrawParams params{
@ -253,71 +151,14 @@ RasterizerVulkan::RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra
descriptor_pool, update_descriptor_queue),
query_cache{*this, maxwell3d, gpu_memory, device, scheduler}, accelerate_dma{buffer_cache},
fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache, device, scheduler),
wfi_event(device.GetLogical().CreateEvent()), async_shaders(emu_window_) {
wfi_event(device.GetLogical().CreateEvent()) {
scheduler.SetQueryCache(query_cache);
if (device.UseAsynchronousShaders()) {
async_shaders.AllocateWorkers();
}
}
RasterizerVulkan::~RasterizerVulkan() = default;
void RasterizerVulkan::Draw(bool is_indexed, bool is_instanced) {
MICROPROFILE_SCOPE(Vulkan_Drawing);
SCOPE_EXIT({ gpu.TickWork(); });
FlushWork();
query_cache.UpdateCounters();
graphics_key.fixed_state.Refresh(maxwell3d, device.IsExtExtendedDynamicStateSupported());
std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
texture_cache.SynchronizeGraphicsDescriptors();
texture_cache.UpdateRenderTargets(false);
const auto shaders = pipeline_cache.GetShaders();
graphics_key.shaders = GetShaderAddresses(shaders);
SetupShaderDescriptors(shaders, is_indexed);
const Framebuffer* const framebuffer = texture_cache.GetFramebuffer();
graphics_key.renderpass = framebuffer->RenderPass();
VKGraphicsPipeline* const pipeline = pipeline_cache.GetGraphicsPipeline(
graphics_key, framebuffer->NumColorBuffers(), async_shaders);
if (pipeline == nullptr || pipeline->GetHandle() == VK_NULL_HANDLE) {
// Async graphics pipeline was not ready.
return;
}
BeginTransformFeedback();
scheduler.RequestRenderpass(framebuffer);
scheduler.BindGraphicsPipeline(pipeline->GetHandle());
UpdateDynamicStates();
const auto& regs = maxwell3d.regs;
const u32 num_instances = maxwell3d.mme_draw.instance_count;
const DrawParams draw_params = MakeDrawParams(regs, num_instances, is_instanced, is_indexed);
const VkPipelineLayout pipeline_layout = pipeline->GetLayout();
const VkDescriptorSet descriptor_set = pipeline->CommitDescriptorSet();
scheduler.Record([pipeline_layout, descriptor_set, draw_params](vk::CommandBuffer cmdbuf) {
if (descriptor_set) {
cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline_layout,
DESCRIPTOR_SET, descriptor_set, nullptr);
}
if (draw_params.is_indexed) {
cmdbuf.DrawIndexed(draw_params.num_vertices, draw_params.num_instances, 0,
draw_params.base_vertex, draw_params.base_instance);
} else {
cmdbuf.Draw(draw_params.num_vertices, draw_params.num_instances,
draw_params.base_vertex, draw_params.base_instance);
}
});
EndTransformFeedback();
UNREACHABLE_MSG("Rendering not implemented {} {}", is_indexed, is_instanced);
}
void RasterizerVulkan::Clear() {
@ -395,73 +236,8 @@ void RasterizerVulkan::Clear() {
});
}
void RasterizerVulkan::DispatchCompute(GPUVAddr code_addr) {
MICROPROFILE_SCOPE(Vulkan_Compute);
query_cache.UpdateCounters();
const auto& launch_desc = kepler_compute.launch_description;
auto& pipeline = pipeline_cache.GetComputePipeline({
.shader = code_addr,
.shared_memory_size = launch_desc.shared_alloc,
.workgroup_size{
launch_desc.block_dim_x,
launch_desc.block_dim_y,
launch_desc.block_dim_z,
},
});
// Compute dispatches can't be executed inside a renderpass
scheduler.RequestOutsideRenderPassOperationContext();
image_view_indices.clear();
sampler_handles.clear();
std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
const auto& entries = pipeline.GetEntries();
buffer_cache.SetEnabledComputeUniformBuffers(entries.enabled_uniform_buffers);
buffer_cache.UnbindComputeStorageBuffers();
u32 ssbo_index = 0;
for (const auto& buffer : entries.global_buffers) {
buffer_cache.BindComputeStorageBuffer(ssbo_index, buffer.cbuf_index, buffer.cbuf_offset,
buffer.is_written);
++ssbo_index;
}
buffer_cache.UpdateComputeBuffers();
texture_cache.SynchronizeComputeDescriptors();
SetupComputeUniformTexels(entries);
SetupComputeTextures(entries);
SetupComputeStorageTexels(entries);
SetupComputeImages(entries);
const std::span indices_span(image_view_indices.data(), image_view_indices.size());
texture_cache.FillComputeImageViews(indices_span, image_view_ids);
update_descriptor_queue.Acquire();
buffer_cache.BindHostComputeBuffers();
ImageViewId* image_view_id_ptr = image_view_ids.data();
VkSampler* sampler_ptr = sampler_handles.data();
PushImageDescriptors(entries, texture_cache, update_descriptor_queue, image_view_id_ptr,
sampler_ptr);
const VkPipeline pipeline_handle = pipeline.GetHandle();
const VkPipelineLayout pipeline_layout = pipeline.GetLayout();
const VkDescriptorSet descriptor_set = pipeline.CommitDescriptorSet();
scheduler.Record([grid_x = launch_desc.grid_dim_x, grid_y = launch_desc.grid_dim_y,
grid_z = launch_desc.grid_dim_z, pipeline_handle, pipeline_layout,
descriptor_set](vk::CommandBuffer cmdbuf) {
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, pipeline_handle);
if (descriptor_set) {
cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, pipeline_layout,
DESCRIPTOR_SET, descriptor_set, nullptr);
}
cmdbuf.Dispatch(grid_x, grid_y, grid_z);
});
void RasterizerVulkan::DispatchCompute() {
UNREACHABLE_MSG("Not implemented");
}
void RasterizerVulkan::ResetCounter(VideoCore::QueryType type) {
@ -716,52 +492,6 @@ bool AccelerateDMA::BufferCopy(GPUVAddr src_address, GPUVAddr dest_address, u64
return buffer_cache.DMACopy(src_address, dest_address, amount);
}
void RasterizerVulkan::SetupShaderDescriptors(
const std::array<Shader*, Maxwell::MaxShaderProgram>& shaders, bool is_indexed) {
image_view_indices.clear();
sampler_handles.clear();
for (size_t stage = 0; stage < Maxwell::MaxShaderStage; ++stage) {
Shader* const shader = shaders[stage + 1];
if (!shader) {
continue;
}
const ShaderEntries& entries = shader->GetEntries();
SetupGraphicsUniformTexels(entries, stage);
SetupGraphicsTextures(entries, stage);
SetupGraphicsStorageTexels(entries, stage);
SetupGraphicsImages(entries, stage);
buffer_cache.SetEnabledUniformBuffers(stage, entries.enabled_uniform_buffers);
buffer_cache.UnbindGraphicsStorageBuffers(stage);
u32 ssbo_index = 0;
for (const auto& buffer : entries.global_buffers) {
buffer_cache.BindGraphicsStorageBuffer(stage, ssbo_index, buffer.cbuf_index,
buffer.cbuf_offset, buffer.is_written);
++ssbo_index;
}
}
const std::span indices_span(image_view_indices.data(), image_view_indices.size());
buffer_cache.UpdateGraphicsBuffers(is_indexed);
texture_cache.FillGraphicsImageViews(indices_span, image_view_ids);
buffer_cache.BindHostGeometryBuffers(is_indexed);
update_descriptor_queue.Acquire();
ImageViewId* image_view_id_ptr = image_view_ids.data();
VkSampler* sampler_ptr = sampler_handles.data();
for (size_t stage = 0; stage < Maxwell::MaxShaderStage; ++stage) {
// Skip VertexA stage
Shader* const shader = shaders[stage + 1];
if (!shader) {
continue;
}
buffer_cache.BindHostStageBuffers(stage);
PushImageDescriptors(shader->GetEntries(), texture_cache, update_descriptor_queue,
image_view_id_ptr, sampler_ptr);
}
}
void RasterizerVulkan::UpdateDynamicStates() {
auto& regs = maxwell3d.regs;
UpdateViewportsState(regs);
@ -810,89 +540,6 @@ void RasterizerVulkan::EndTransformFeedback() {
[](vk::CommandBuffer cmdbuf) { cmdbuf.EndTransformFeedbackEXT(0, 0, nullptr, nullptr); });
}
void RasterizerVulkan::SetupGraphicsUniformTexels(const ShaderEntries& entries, size_t stage) {
const auto& regs = maxwell3d.regs;
const bool via_header_index = regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex;
for (const auto& entry : entries.uniform_texels) {
const TextureHandle handle = GetTextureInfo(maxwell3d, via_header_index, entry, stage);
image_view_indices.push_back(handle.image);
}
}
void RasterizerVulkan::SetupGraphicsTextures(const ShaderEntries& entries, size_t stage) {
const auto& regs = maxwell3d.regs;
const bool via_header_index = regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex;
for (const auto& entry : entries.samplers) {
for (size_t index = 0; index < entry.size; ++index) {
const TextureHandle handle =
GetTextureInfo(maxwell3d, via_header_index, entry, stage, index);
image_view_indices.push_back(handle.image);
Sampler* const sampler = texture_cache.GetGraphicsSampler(handle.sampler);
sampler_handles.push_back(sampler->Handle());
}
}
}
void RasterizerVulkan::SetupGraphicsStorageTexels(const ShaderEntries& entries, size_t stage) {
const auto& regs = maxwell3d.regs;
const bool via_header_index = regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex;
for (const auto& entry : entries.storage_texels) {
const TextureHandle handle = GetTextureInfo(maxwell3d, via_header_index, entry, stage);
image_view_indices.push_back(handle.image);
}
}
void RasterizerVulkan::SetupGraphicsImages(const ShaderEntries& entries, size_t stage) {
const auto& regs = maxwell3d.regs;
const bool via_header_index = regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex;
for (const auto& entry : entries.images) {
const TextureHandle handle = GetTextureInfo(maxwell3d, via_header_index, entry, stage);
image_view_indices.push_back(handle.image);
}
}
void RasterizerVulkan::SetupComputeUniformTexels(const ShaderEntries& entries) {
const bool via_header_index = kepler_compute.launch_description.linked_tsc;
for (const auto& entry : entries.uniform_texels) {
const TextureHandle handle =
GetTextureInfo(kepler_compute, via_header_index, entry, COMPUTE_SHADER_INDEX);
image_view_indices.push_back(handle.image);
}
}
void RasterizerVulkan::SetupComputeTextures(const ShaderEntries& entries) {
const bool via_header_index = kepler_compute.launch_description.linked_tsc;
for (const auto& entry : entries.samplers) {
for (size_t index = 0; index < entry.size; ++index) {
const TextureHandle handle = GetTextureInfo(kepler_compute, via_header_index, entry,
COMPUTE_SHADER_INDEX, index);
image_view_indices.push_back(handle.image);
Sampler* const sampler = texture_cache.GetComputeSampler(handle.sampler);
sampler_handles.push_back(sampler->Handle());
}
}
}
void RasterizerVulkan::SetupComputeStorageTexels(const ShaderEntries& entries) {
const bool via_header_index = kepler_compute.launch_description.linked_tsc;
for (const auto& entry : entries.storage_texels) {
const TextureHandle handle =
GetTextureInfo(kepler_compute, via_header_index, entry, COMPUTE_SHADER_INDEX);
image_view_indices.push_back(handle.image);
}
}
void RasterizerVulkan::SetupComputeImages(const ShaderEntries& entries) {
const bool via_header_index = kepler_compute.launch_description.linked_tsc;
for (const auto& entry : entries.images) {
const TextureHandle handle =
GetTextureInfo(kepler_compute, via_header_index, entry, COMPUTE_SHADER_INDEX);
image_view_indices.push_back(handle.image);
}
}
void RasterizerVulkan::UpdateViewportsState(Tegra::Engines::Maxwell3D::Regs& regs) {
if (!state_tracker.TouchViewports()) {
return;

47
src/video_core/renderer_vulkan/vk_rasterizer.h
View file

@ -28,7 +28,6 @@
#include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
#include "video_core/renderer_vulkan/vk_texture_cache.h"
#include "video_core/renderer_vulkan/vk_update_descriptor.h"
#include "video_core/shader/async_shaders.h"
#include "video_core/vulkan_common/vulkan_memory_allocator.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
@ -73,7 +72,7 @@ public:
void Draw(bool is_indexed, bool is_instanced) override;
void Clear() override;
void DispatchCompute(GPUVAddr code_addr) override;
void DispatchCompute() override;
void ResetCounter(VideoCore::QueryType type) override;
void Query(GPUVAddr gpu_addr, VideoCore::QueryType type, std::optional<u64> timestamp) override;
void BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr, u32 size) override;
@ -103,19 +102,6 @@ public:
bool AccelerateDisplay(const Tegra::FramebufferConfig& config, VAddr framebuffer_addr,
u32 pixel_stride) override;
VideoCommon::Shader::AsyncShaders& GetAsyncShaders() {
return async_shaders;
}
const VideoCommon::Shader::AsyncShaders& GetAsyncShaders() const {
return async_shaders;
}
/// Maximum supported size that a constbuffer can have in bytes.
static constexpr size_t MaxConstbufferSize = 0x10000;
static_assert(MaxConstbufferSize % (4 * sizeof(float)) == 0,
"The maximum size of a constbuffer must be a multiple of the size of GLvec4");
private:
static constexpr size_t MAX_TEXTURES = 192;
static constexpr size_t MAX_IMAGES = 48;
@ -125,40 +111,12 @@ private:
void FlushWork();
/// Setup descriptors in the graphics pipeline.
void SetupShaderDescriptors(const std::array<Shader*, Maxwell::MaxShaderProgram>& shaders,
bool is_indexed);
void UpdateDynamicStates();
void BeginTransformFeedback();
void EndTransformFeedback();
/// Setup uniform texels in the graphics pipeline.
void SetupGraphicsUniformTexels(const ShaderEntries& entries, std::size_t stage);
/// Setup textures in the graphics pipeline.
void SetupGraphicsTextures(const ShaderEntries& entries, std::size_t stage);
/// Setup storage texels in the graphics pipeline.
void SetupGraphicsStorageTexels(const ShaderEntries& entries, std::size_t stage);
/// Setup images in the graphics pipeline.
void SetupGraphicsImages(const ShaderEntries& entries, std::size_t stage);
/// Setup texel buffers in the compute pipeline.
void SetupComputeUniformTexels(const ShaderEntries& entries);
/// Setup textures in the compute pipeline.
void SetupComputeTextures(const ShaderEntries& entries);
/// Setup storage texels in the compute pipeline.
void SetupComputeStorageTexels(const ShaderEntries& entries);
/// Setup images in the compute pipeline.
void SetupComputeImages(const ShaderEntries& entries);
void UpdateViewportsState(Tegra::Engines::Maxwell3D::Regs& regs);
void UpdateScissorsState(Tegra::Engines::Maxwell3D::Regs& regs);
void UpdateDepthBias(Tegra::Engines::Maxwell3D::Regs& regs);
@ -198,13 +156,12 @@ private:
TextureCache texture_cache;
BufferCacheRuntime buffer_cache_runtime;
BufferCache buffer_cache;
VKPipelineCache pipeline_cache;
PipelineCache pipeline_cache;
VKQueryCache query_cache;
AccelerateDMA accelerate_dma;
VKFenceManager fence_manager;
vk::Event wfi_event;
VideoCommon::Shader::AsyncShaders async_shaders;
boost::container::static_vector<u32, MAX_IMAGE_VIEWS> image_view_indices;
std::array<VideoCommon::ImageViewId, MAX_IMAGE_VIEWS> image_view_ids;

752
src/video_core/shader/ast.cpp
View file

@ -1,752 +0,0 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <string>
#include <string_view>
#include <fmt/format.h>
#include "common/assert.h"
#include "common/common_types.h"
#include "video_core/shader/ast.h"
#include "video_core/shader/expr.h"
namespace VideoCommon::Shader {
ASTZipper::ASTZipper() = default;
void ASTZipper::Init(const ASTNode new_first, const ASTNode parent) {
ASSERT(new_first->manager == nullptr);
first = new_first;
last = new_first;
ASTNode current = first;
while (current) {
current->manager = this;
current->parent = parent;
last = current;
current = current->next;
}
}
void ASTZipper::PushBack(const ASTNode new_node) {
ASSERT(new_node->manager == nullptr);
new_node->previous = last;
if (last) {
last->next = new_node;
}
new_node->next.reset();
last = new_node;
if (!first) {
first = new_node;
}
new_node->manager = this;
}
void ASTZipper::PushFront(const ASTNode new_node) {
ASSERT(new_node->manager == nullptr);
new_node->previous.reset();
new_node->next = first;
if (first) {
first->previous = new_node;
}
if (last == first) {
last = new_node;
}
first = new_node;
new_node->manager = this;
}
void ASTZipper::InsertAfter(const ASTNode new_node, const ASTNode at_node) {
ASSERT(new_node->manager == nullptr);
if (!at_node) {
PushFront(new_node);
return;
}
const ASTNode next = at_node->next;
if (next) {
next->previous = new_node;
}
new_node->previous = at_node;
if (at_node == last) {
last = new_node;
}
new_node->next = next;
at_node->next = new_node;
new_node->manager = this;
}
void ASTZipper::InsertBefore(const ASTNode new_node, const ASTNode at_node) {
ASSERT(new_node->manager == nullptr);
if (!at_node) {
PushBack(new_node);
return;
}
const ASTNode previous = at_node->previous;
if (previous) {
previous->next = new_node;
}
new_node->next = at_node;
if (at_node == first) {
first = new_node;
}
new_node->previous = previous;
at_node->previous = new_node;
new_node->manager = this;
}
void ASTZipper::DetachTail(ASTNode node) {
ASSERT(node->manager == this);
if (node == first) {
first.reset();
last.reset();
return;
}
last = node->previous;
last->next.reset();
node->previous.reset();
ASTNode current = std::move(node);
while (current) {
current->manager = nullptr;
current->parent.reset();
current = current->next;
}
}
void ASTZipper::DetachSegment(const ASTNode start, const ASTNode end) {
ASSERT(start->manager == this && end->manager == this);
if (start == end) {
DetachSingle(start);
return;
}
const ASTNode prev = start->previous;
const ASTNode post = end->next;
if (!prev) {
first = post;
} else {
prev->next = post;
}
if (!post) {
last = prev;
} else {
post->previous = prev;
}
start->previous.reset();
end->next.reset();
ASTNode current = start;
bool found = false;
while (current) {
current->manager = nullptr;
current->parent.reset();
found |= current == end;
current = current->next;
}
ASSERT(found);
}
void ASTZipper::DetachSingle(const ASTNode node) {
ASSERT(node->manager == this);
const ASTNode prev = node->previous;
const ASTNode post = node->next;
node->previous.reset();
node->next.reset();
if (!prev) {
first = post;
} else {
prev->next = post;
}
if (!post) {
last = prev;
} else {
post->previous = prev;
}
node->manager = nullptr;
node->parent.reset();
}
void ASTZipper::Remove(const ASTNode node) {
ASSERT(node->manager == this);
const ASTNode next = node->next;
const ASTNode previous = node->previous;
if (previous) {
previous->next = next;
}
if (next) {
next->previous = previous;
}
node->parent.reset();
node->manager = nullptr;
if (node == last) {
last = previous;
}
if (node == first) {
first = next;
}
}
class ExprPrinter final {
public:
void operator()(const ExprAnd& expr) {
inner += "( ";
std::visit(*this, *expr.operand1);
inner += " && ";
std::visit(*this, *expr.operand2);
inner += ')';
}
void operator()(const ExprOr& expr) {
inner += "( ";
std::visit(*this, *expr.operand1);
inner += " || ";
std::visit(*this, *expr.operand2);
inner += ')';
}
void operator()(const ExprNot& expr) {
inner += "!";
std::visit(*this, *expr.operand1);
}
void operator()(const ExprPredicate& expr) {
inner += fmt::format("P{}", expr.predicate);
}
void operator()(const ExprCondCode& expr) {
inner += fmt::format("CC{}", expr.cc);
}
void operator()(const ExprVar& expr) {
inner += fmt::format("V{}", expr.var_index);
}
void operator()(const ExprBoolean& expr) {
inner += expr.value ? "true" : "false";
}
void operator()(const ExprGprEqual& expr) {
inner += fmt::format("(gpr_{} == {})", expr.gpr, expr.value);
}
const std::string& GetResult() const {
return inner;
}
private:
std::string inner;
};
class ASTPrinter {
public:
void operator()(const ASTProgram& ast) {
scope++;
inner += "program {\n";
ASTNode current = ast.nodes.GetFirst();
while (current) {
Visit(current);
current = current->GetNext();
}
inner += "}\n";
scope--;
}
void operator()(const ASTIfThen& ast) {
ExprPrinter expr_parser{};
std::visit(expr_parser, *ast.condition);
inner += fmt::format("{}if ({}) {{\n", Indent(), expr_parser.GetResult());
scope++;
ASTNode current = ast.nodes.GetFirst();
while (current) {
Visit(current);
current = current->GetNext();
}
scope--;
inner += fmt::format("{}}}\n", Indent());
}
void operator()(const ASTIfElse& ast) {
inner += Indent();
inner += "else {\n";
scope++;
ASTNode current = ast.nodes.GetFirst();
while (current) {
Visit(current);
current = current->GetNext();
}
scope--;
inner += Indent();
inner += "}\n";
}
void operator()(const ASTBlockEncoded& ast) {
inner += fmt::format("{}Block({}, {});\n", Indent(), ast.start, ast.end);
}
void operator()([[maybe_unused]] const ASTBlockDecoded& ast) {
inner += Indent();
inner += "Block;\n";
}
void operator()(const ASTVarSet& ast) {
ExprPrinter expr_parser{};
std::visit(expr_parser, *ast.condition);
inner += fmt::format("{}V{} := {};\n", Indent(), ast.index, expr_parser.GetResult());
}
void operator()(const ASTLabel& ast) {
inner += fmt::format("Label_{}:\n", ast.index);
}
void operator()(const ASTGoto& ast) {
ExprPrinter expr_parser{};
std::visit(expr_parser, *ast.condition);
inner +=
fmt::format("{}({}) -> goto Label_{};\n", Indent(), expr_parser.GetResult(), ast.label);
}
void operator()(const ASTDoWhile& ast) {
ExprPrinter expr_parser{};
std::visit(expr_parser, *ast.condition);
inner += fmt::format("{}do {{\n", Indent());
scope++;
ASTNode current = ast.nodes.GetFirst();
while (current) {
Visit(current);
current = current->GetNext();
}
scope--;
inner += fmt::format("{}}} while ({});\n", Indent(), expr_parser.GetResult());
}
void operator()(const ASTReturn& ast) {
ExprPrinter expr_parser{};
std::visit(expr_parser, *ast.condition);
inner += fmt::format("{}({}) -> {};\n", Indent(), expr_parser.GetResult(),
ast.kills ? "discard" : "exit");
}
void operator()(const ASTBreak& ast) {
ExprPrinter expr_parser{};
std::visit(expr_parser, *ast.condition);
inner += fmt::format("{}({}) -> break;\n", Indent(), expr_parser.GetResult());
}
void Visit(const ASTNode& node) {
std::visit(*this, *node->GetInnerData());
}
const std::string& GetResult() const {
return inner;
}
private:
std::string_view Indent() {
if (space_segment_scope == scope) {
return space_segment;
}
// Ensure that we don't exceed our view.
ASSERT(scope * 2 < spaces.size());
space_segment = spaces.substr(0, scope * 2);
space_segment_scope = scope;
return space_segment;
}
std::string inner{};
std::string_view space_segment;
u32 scope{};
u32 space_segment_scope{};
static constexpr std::string_view spaces{" "};
};
std::string ASTManager::Print() const {
ASTPrinter printer{};
printer.Visit(main_node);
return printer.GetResult();
}
ASTManager::ASTManager(bool do_full_decompile, bool disable_else_derivation_)
: full_decompile{do_full_decompile}, disable_else_derivation{disable_else_derivation_} {}
ASTManager::~ASTManager() {
Clear();
}
void ASTManager::Init() {
main_node = ASTBase::Make<ASTProgram>(ASTNode{});
program = std::get_if<ASTProgram>(main_node->GetInnerData());
false_condition = MakeExpr<ExprBoolean>(false);
}
void ASTManager::DeclareLabel(u32 address) {
const auto pair = labels_map.emplace(address, labels_count);
if (pair.second) {
labels_count++;
labels.resize(labels_count);
}
}
void ASTManager::InsertLabel(u32 address) {
const u32 index = labels_map[address];
const ASTNode label = ASTBase::Make<ASTLabel>(main_node, index);
labels[index] = label;
program->nodes.PushBack(label);
}
void ASTManager::InsertGoto(Expr condition, u32 address) {
const u32 index = labels_map[address];
const ASTNode goto_node = ASTBase::Make<ASTGoto>(main_node, std::move(condition), index);
gotos.push_back(goto_node);
program->nodes.PushBack(goto_node);
}
void ASTManager::InsertBlock(u32 start_address, u32 end_address) {
ASTNode block = ASTBase::Make<ASTBlockEncoded>(main_node, start_address, end_address);
program->nodes.PushBack(std::move(block));
}
void ASTManager::InsertReturn(Expr condition, bool kills) {
ASTNode node = ASTBase::Make<ASTReturn>(main_node, std::move(condition), kills);
program->nodes.PushBack(std::move(node));
}
// The decompile algorithm is based on
// "Taming control flow: A structured approach to eliminating goto statements"
// by AM Erosa, LJ Hendren 1994. In general, the idea is to get gotos to be
// on the same structured level as the label which they jump to. This is done,
// through outward/inward movements and lifting. Once they are at the same
// level, you can enclose them in an "if" structure or a "do-while" structure.
void ASTManager::Decompile() {
auto it = gotos.begin();
while (it != gotos.end()) {
const ASTNode goto_node = *it;
const auto label_index = goto_node->GetGotoLabel();
if (!label_index) {
return;
}
const ASTNode label = labels[*label_index];
if (!full_decompile) {
// We only decompile backward jumps
if (!IsBackwardsJump(goto_node, label)) {
it++;
continue;
}
}
if (IndirectlyRelated(goto_node, label)) {
while (!DirectlyRelated(goto_node, label)) {
MoveOutward(goto_node);
}
}
if (DirectlyRelated(goto_node, label)) {
u32 goto_level = goto_node->GetLevel();
const u32 label_level = label->GetLevel();
while (label_level < goto_level) {
MoveOutward(goto_node);
goto_level--;
}
// TODO(Blinkhawk): Implement Lifting and Inward Movements
}
if (label->GetParent() == goto_node->GetParent()) {
bool is_loop = false;
ASTNode current = goto_node->GetPrevious();
while (current) {
if (current == label) {
is_loop = true;
break;
}
current = current->GetPrevious();
}
if (is_loop) {
EncloseDoWhile(goto_node, label);
} else {
EncloseIfThen(goto_node, label);
}
it = gotos.erase(it);
continue;
}
it++;
}
if (full_decompile) {
for (const ASTNode& label : labels) {
auto& manager = label->GetManager();
manager.Remove(label);
}
labels.clear();
} else {
auto label_it = labels.begin();
while (label_it != labels.end()) {
bool can_remove = true;
ASTNode label = *label_it;
for (const ASTNode& goto_node : gotos) {
const auto label_index = goto_node->GetGotoLabel();
if (!label_index) {
return;
}
ASTNode& glabel = labels[*label_index];
if (glabel == label) {
can_remove = false;
break;
}
}
if (can_remove) {
label->MarkLabelUnused();
}
}
}
}
bool ASTManager::IsBackwardsJump(ASTNode goto_node, ASTNode label_node) const {
u32 goto_level = goto_node->GetLevel();
u32 label_level = label_node->GetLevel();
while (goto_level > label_level) {
goto_level--;
goto_node = goto_node->GetParent();
}
while (label_level > goto_level) {
label_level--;
label_node = label_node->GetParent();
}
while (goto_node->GetParent() != label_node->GetParent()) {
goto_node = goto_node->GetParent();
label_node = label_node->GetParent();
}
ASTNode current = goto_node->GetPrevious();
while (current) {
if (current == label_node) {
return true;
}
current = current->GetPrevious();
}
return false;
}
bool ASTManager::IndirectlyRelated(const ASTNode& first, const ASTNode& second) const {
return !(first->GetParent() == second->GetParent() || DirectlyRelated(first, second));
}
bool ASTManager::DirectlyRelated(const ASTNode& first, const ASTNode& second) const {
if (first->GetParent() == second->GetParent()) {
return false;
}
const u32 first_level = first->GetLevel();
const u32 second_level = second->GetLevel();
u32 min_level;
u32 max_level;
ASTNode max;
ASTNode min;
if (first_level > second_level) {
min_level = second_level;
min = second;
max_level = first_level;
max = first;
} else {
min_level = first_level;
min = first;
max_level = second_level;
max = second;
}
while (max_level > min_level) {
max_level--;
max = max->GetParent();
}
return min->GetParent() == max->GetParent();
}
void ASTManager::ShowCurrentState(std::string_view state) const {
LOG_CRITICAL(HW_GPU, "\nState {}:\n\n{}\n", state, Print());
SanityCheck();
}
void ASTManager::SanityCheck() const {
for (const auto& label : labels) {
if (!label->GetParent()) {
LOG_CRITICAL(HW_GPU, "Sanity Check Failed");
}
}
}
void ASTManager::EncloseDoWhile(ASTNode goto_node, ASTNode label) {
ASTZipper& zipper = goto_node->GetManager();
const ASTNode loop_start = label->GetNext();
if (loop_start == goto_node) {
zipper.Remove(goto_node);
return;
}
const ASTNode parent = label->GetParent();
const Expr condition = goto_node->GetGotoCondition();
zipper.DetachSegment(loop_start, goto_node);
const ASTNode do_while_node = ASTBase::Make<ASTDoWhile>(parent, condition);
ASTZipper* sub_zipper = do_while_node->GetSubNodes();
sub_zipper->Init(loop_start, do_while_node);
zipper.InsertAfter(do_while_node, label);
sub_zipper->Remove(goto_node);
}
void ASTManager::EncloseIfThen(ASTNode goto_node, ASTNode label) {
ASTZipper& zipper = goto_node->GetManager();
const ASTNode if_end = label->GetPrevious();
if (if_end == goto_node) {
zipper.Remove(goto_node);
return;
}
const ASTNode prev = goto_node->GetPrevious();
const Expr condition = goto_node->GetGotoCondition();
bool do_else = false;
if (!disable_else_derivation && prev->IsIfThen()) {
const Expr if_condition = prev->GetIfCondition();
do_else = ExprAreEqual(if_condition, condition);
}
const ASTNode parent = label->GetParent();
zipper.DetachSegment(goto_node, if_end);
ASTNode if_node;
if (do_else) {
if_node = ASTBase::Make<ASTIfElse>(parent);
} else {
Expr neg_condition = MakeExprNot(condition);
if_node = ASTBase::Make<ASTIfThen>(parent, neg_condition);
}
ASTZipper* sub_zipper = if_node->GetSubNodes();
sub_zipper->Init(goto_node, if_node);
zipper.InsertAfter(if_node, prev);
sub_zipper->Remove(goto_node);
}
void ASTManager::MoveOutward(ASTNode goto_node) {
ASTZipper& zipper = goto_node->GetManager();
const ASTNode parent = goto_node->GetParent();
ASTZipper& zipper2 = parent->GetManager();
const ASTNode grandpa = parent->GetParent();
const bool is_loop = parent->IsLoop();
const bool is_else = parent->IsIfElse();
const bool is_if = parent->IsIfThen();
const ASTNode prev = goto_node->GetPrevious();
const ASTNode post = goto_node->GetNext();
const Expr condition = goto_node->GetGotoCondition();
zipper.DetachSingle(goto_node);
if (is_loop) {
const u32 var_index = NewVariable();
const Expr var_condition = MakeExpr<ExprVar>(var_index);
const ASTNode var_node = ASTBase::Make<ASTVarSet>(parent, var_index, condition);
const ASTNode var_node_init = ASTBase::Make<ASTVarSet>(parent, var_index, false_condition);
zipper2.InsertBefore(var_node_init, parent);
zipper.InsertAfter(var_node, prev);
goto_node->SetGotoCondition(var_condition);
const ASTNode break_node = ASTBase::Make<ASTBreak>(parent, var_condition);
zipper.InsertAfter(break_node, var_node);
} else if (is_if || is_else) {
const u32 var_index = NewVariable();
const Expr var_condition = MakeExpr<ExprVar>(var_index);
const ASTNode var_node = ASTBase::Make<ASTVarSet>(parent, var_index, condition);
const ASTNode var_node_init = ASTBase::Make<ASTVarSet>(parent, var_index, false_condition);
if (is_if) {
zipper2.InsertBefore(var_node_init, parent);
} else {
zipper2.InsertBefore(var_node_init, parent->GetPrevious());
}
zipper.InsertAfter(var_node, prev);
goto_node->SetGotoCondition(var_condition);
if (post) {
zipper.DetachTail(post);
const ASTNode if_node = ASTBase::Make<ASTIfThen>(parent, MakeExprNot(var_condition));
ASTZipper* sub_zipper = if_node->GetSubNodes();
sub_zipper->Init(post, if_node);
zipper.InsertAfter(if_node, var_node);
}
} else {
UNREACHABLE();
}
const ASTNode next = parent->GetNext();
if (is_if && next && next->IsIfElse()) {
zipper2.InsertAfter(goto_node, next);
goto_node->SetParent(grandpa);
return;
}
zipper2.InsertAfter(goto_node, parent);
goto_node->SetParent(grandpa);
}
class ASTClearer {
public:
ASTClearer() = default;
void operator()(const ASTProgram& ast) {
ASTNode current = ast.nodes.GetFirst();
while (current) {
Visit(current);
current = current->GetNext();
}
}
void operator()(const ASTIfThen& ast) {
ASTNode current = ast.nodes.GetFirst();
while (current) {
Visit(current);
current = current->GetNext();
}
}
void operator()(const ASTIfElse& ast) {
ASTNode current = ast.nodes.GetFirst();
while (current) {
Visit(current);
current = current->GetNext();
}
}
void operator()([[maybe_unused]] const ASTBlockEncoded& ast) {}
void operator()(ASTBlockDecoded& ast) {
ast.nodes.clear();
}
void operator()([[maybe_unused]] const ASTVarSet& ast) {}
void operator()([[maybe_unused]] const ASTLabel& ast) {}
void operator()([[maybe_unused]] const ASTGoto& ast) {}
void operator()(const ASTDoWhile& ast) {
ASTNode current = ast.nodes.GetFirst();
while (current) {
Visit(current);
current = current->GetNext();
}
}
void operator()([[maybe_unused]] const ASTReturn& ast) {}
void operator()([[maybe_unused]] const ASTBreak& ast) {}
void Visit(const ASTNode& node) {
std::visit(*this, *node->GetInnerData());
node->Clear();
}
};
void ASTManager::Clear() {
if (!main_node) {
return;
}
ASTClearer clearer{};
clearer.Visit(main_node);
main_node.reset();
program = nullptr;
labels_map.clear();
labels.clear();
gotos.clear();
}
} // namespace VideoCommon::Shader

398
src/video_core/shader/ast.h
View file

@ -1,398 +0,0 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <functional>
#include <list>
#include <memory>
#include <optional>
#include <string>
#include <unordered_map>
#include <vector>
#include "video_core/shader/expr.h"
#include "video_core/shader/node.h"
namespace VideoCommon::Shader {
class ASTBase;
class ASTBlockDecoded;
class ASTBlockEncoded;
class ASTBreak;
class ASTDoWhile;
class ASTGoto;
class ASTIfElse;
class ASTIfThen;
class ASTLabel;
class ASTProgram;
class ASTReturn;
class ASTVarSet;
using ASTData = std::variant<ASTProgram, ASTIfThen, ASTIfElse, ASTBlockEncoded, ASTBlockDecoded,
ASTVarSet, ASTGoto, ASTLabel, ASTDoWhile, ASTReturn, ASTBreak>;
using ASTNode = std::shared_ptr<ASTBase>;
enum class ASTZipperType : u32 {
Program,
IfThen,
IfElse,
Loop,
};
class ASTZipper final {
public:
explicit ASTZipper();
void Init(ASTNode first, ASTNode parent);
ASTNode GetFirst() const {
return first;
}
ASTNode GetLast() const {
return last;
}
void PushBack(ASTNode new_node);
void PushFront(ASTNode new_node);
void InsertAfter(ASTNode new_node, ASTNode at_node);
void InsertBefore(ASTNode new_node, ASTNode at_node);
void DetachTail(ASTNode node);
void DetachSingle(ASTNode node);
void DetachSegment(ASTNode start, ASTNode end);
void Remove(ASTNode node);
ASTNode first;
ASTNode last;
};
class ASTProgram {
public:
ASTZipper nodes{};
};
class ASTIfThen {
public:
explicit ASTIfThen(Expr condition_) : condition{std::move(condition_)} {}
Expr condition;
ASTZipper nodes{};
};
class ASTIfElse {
public:
ASTZipper nodes{};
};
class ASTBlockEncoded {
public:
explicit ASTBlockEncoded(u32 start_, u32 _) : start{start_}, end{_} {}
u32 start;
u32 end;
};
class ASTBlockDecoded {
public:
explicit ASTBlockDecoded(NodeBlock&& new_nodes_) : nodes(std::move(new_nodes_)) {}
NodeBlock nodes;
};
class ASTVarSet {
public:
explicit ASTVarSet(u32 index_, Expr condition_)
: index{index_}, condition{std::move(condition_)} {}
u32 index;
Expr condition;
};
class ASTLabel {
public:
explicit ASTLabel(u32 index_) : index{index_} {}
u32 index;
bool unused{};
};
class ASTGoto {
public:
explicit ASTGoto(Expr condition_, u32 label_)
: condition{std::move(condition_)}, label{label_} {}
Expr condition;
u32 label;
};
class ASTDoWhile {
public:
explicit ASTDoWhile(Expr condition_) : condition{std::move(condition_)} {}
Expr condition;
ASTZipper nodes{};
};
class ASTReturn {
public:
explicit ASTReturn(Expr condition_, bool kills_)
: condition{std::move(condition_)}, kills{kills_} {}
Expr condition;
bool kills;
};
class ASTBreak {
public:
explicit ASTBreak(Expr condition_) : condition{std::move(condition_)} {}
Expr condition;
};
class ASTBase {
public:
explicit ASTBase(ASTNode parent_, ASTData data_)
: data{std::move(data_)}, parent{std::move(parent_)} {}
template <class U, class... Args>
static ASTNode Make(ASTNode parent, Args&&... args) {
return std::make_shared<ASTBase>(std::move(parent),
ASTData(U(std::forward<Args>(args)...)));
}
void SetParent(ASTNode new_parent) {
parent = std::move(new_parent);
}
ASTNode& GetParent() {
return parent;
}
const ASTNode& GetParent() const {
return parent;
}
u32 GetLevel() const {
u32 level = 0;
auto next_parent = parent;
while (next_parent) {
next_parent = next_parent->GetParent();
level++;
}
return level;
}
ASTData* GetInnerData() {
return &data;
}
const ASTData* GetInnerData() const {
return &data;
}
ASTNode GetNext() const {
return next;
}
ASTNode GetPrevious() const {
return previous;
}
ASTZipper& GetManager() {
return *manager;
}
const ASTZipper& GetManager() const {
return *manager;
}
std::optional<u32> GetGotoLabel() const {
if (const auto* inner = std::get_if<ASTGoto>(&data)) {
return {inner->label};
}
return std::nullopt;
}
Expr GetGotoCondition() const {
if (const auto* inner = std::get_if<ASTGoto>(&data)) {
return inner->condition;
}
return nullptr;
}
void MarkLabelUnused() {
if (auto* inner = std::get_if<ASTLabel>(&data)) {
inner->unused = true;
}
}
bool IsLabelUnused() const {
if (const auto* inner = std::get_if<ASTLabel>(&data)) {
return inner->unused;
}
return true;
}
std::optional<u32> GetLabelIndex() const {
if (const auto* inner = std::get_if<ASTLabel>(&data)) {
return {inner->index};
}
return std::nullopt;
}
Expr GetIfCondition() const {
if (const auto* inner = std::get_if<ASTIfThen>(&data)) {
return inner->condition;
}
return nullptr;
}
void SetGotoCondition(Expr new_condition) {
if (auto* inner = std::get_if<ASTGoto>(&data)) {
inner->condition = std::move(new_condition);
}
}
bool IsIfThen() const {
return std::holds_alternative<ASTIfThen>(data);
}
bool IsIfElse() const {
return std::holds_alternative<ASTIfElse>(data);
}
bool IsBlockEncoded() const {
return std::holds_alternative<ASTBlockEncoded>(data);
}
void TransformBlockEncoded(NodeBlock&& nodes) {
data = ASTBlockDecoded(std::move(nodes));
}
bool IsLoop() const {
return std::holds_alternative<ASTDoWhile>(data);
}
ASTZipper* GetSubNodes() {
if (std::holds_alternative<ASTProgram>(data)) {
return &std::get_if<ASTProgram>(&data)->nodes;
}
if (std::holds_alternative<ASTIfThen>(data)) {
return &std::get_if<ASTIfThen>(&data)->nodes;
}
if (std::holds_alternative<ASTIfElse>(data)) {
return &std::get_if<ASTIfElse>(&data)->nodes;
}
if (std::holds_alternative<ASTDoWhile>(data)) {
return &std::get_if<ASTDoWhile>(&data)->nodes;
}
return nullptr;
}
void Clear() {
next.reset();
previous.reset();
parent.reset();
manager = nullptr;
}
private:
friend class ASTZipper;
ASTData data;
ASTNode parent;
ASTNode next;
ASTNode previous;
ASTZipper* manager{};
};
class ASTManager final {
public:
explicit ASTManager(bool do_full_decompile, bool disable_else_derivation_);
~ASTManager();
ASTManager(const ASTManager& o) = delete;
ASTManager& operator=(const ASTManager& other) = delete;
ASTManager(ASTManager&& other) noexcept = default;
ASTManager& operator=(ASTManager&& other) noexcept = default;
void Init();
void DeclareLabel(u32 address);
void InsertLabel(u32 address);
void InsertGoto(Expr condition, u32 address);
void InsertBlock(u32 start_address, u32 end_address);
void InsertReturn(Expr condition, bool kills);
std::string Print() const;
void Decompile();
void ShowCurrentState(std::string_view state) const;
void SanityCheck() const;
void Clear();
bool IsFullyDecompiled() const {
if (full_decompile) {
return gotos.empty();
}
for (ASTNode goto_node : gotos) {
auto label_index = goto_node->GetGotoLabel();
if (!label_index) {
return false;
}
ASTNode glabel = labels[*label_index];
if (IsBackwardsJump(goto_node, glabel)) {
return false;
}
}
return true;
}
ASTNode GetProgram() const {
return main_node;
}
u32 GetVariables() const {
return variables;
}
const std::vector<ASTNode>& GetLabels() const {
return labels;
}
private:
bool IsBackwardsJump(ASTNode goto_node, ASTNode label_node) const;
bool IndirectlyRelated(const ASTNode& first, const ASTNode& second) const;
bool DirectlyRelated(const ASTNode& first, const ASTNode& second) const;
void EncloseDoWhile(ASTNode goto_node, ASTNode label);
void EncloseIfThen(ASTNode goto_node, ASTNode label);
void MoveOutward(ASTNode goto_node);
u32 NewVariable() {
return variables++;
}
bool full_decompile{};
bool disable_else_derivation{};
std::unordered_map<u32, u32> labels_map{};
u32 labels_count{};
std::vector<ASTNode> labels{};
std::list<ASTNode> gotos{};
u32 variables{};
ASTProgram* program{};
ASTNode main_node{};
Expr false_condition{};
};
} // namespace VideoCommon::Shader

234
src/video_core/shader/async_shaders.cpp
View file

@ -1,234 +0,0 @@
// Copyright 2020 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <condition_variable>
#include <mutex>
#include <thread>
#include <vector>
#include "video_core/engines/maxwell_3d.h"
#include "video_core/renderer_base.h"
#include "video_core/renderer_opengl/gl_shader_cache.h"
#include "video_core/shader/async_shaders.h"
namespace VideoCommon::Shader {
AsyncShaders::AsyncShaders(Core::Frontend::EmuWindow& emu_window_) : emu_window(emu_window_) {}
AsyncShaders::~AsyncShaders() {
KillWorkers();
}
void AsyncShaders::AllocateWorkers() {
// Use at least one thread
u32 num_workers = 1;
// Deduce how many more threads we can use
const u32 thread_count = std::thread::hardware_concurrency();
if (thread_count >= 8) {
// Increase async workers by 1 for every 2 threads >= 8
num_workers += 1 + (thread_count - 8) / 2;
}
// If we already have workers queued, ignore
if (num_workers == worker_threads.size()) {
return;
}
// If workers already exist, clear them
if (!worker_threads.empty()) {
FreeWorkers();
}
// Create workers
for (std::size_t i = 0; i < num_workers; i++) {
context_list.push_back(emu_window.CreateSharedContext());
worker_threads.emplace_back(&AsyncShaders::ShaderCompilerThread, this,
context_list[i].get());
}
}
void AsyncShaders::FreeWorkers() {
// Mark all threads to quit
is_thread_exiting.store(true);
cv.notify_all();
for (auto& thread : worker_threads) {
thread.join();
}
// Clear our shared contexts
context_list.clear();
// Clear our worker threads
worker_threads.clear();
}
void AsyncShaders::KillWorkers() {
is_thread_exiting.store(true);
cv.notify_all();
for (auto& thread : worker_threads) {
thread.detach();
}
// Clear our shared contexts
context_list.clear();
// Clear our worker threads
worker_threads.clear();
}
bool AsyncShaders::HasWorkQueued() const {
return !pending_queue.empty();
}
bool AsyncShaders::HasCompletedWork() const {
std::shared_lock lock{completed_mutex};
return !finished_work.empty();
}
bool AsyncShaders::IsShaderAsync(const Tegra::GPU& gpu) const {
const auto& regs = gpu.Maxwell3D().regs;
// If something is using depth, we can assume that games are not rendering anything which will
// be used one time.
if (regs.zeta_enable) {
return true;
}
// If games are using a small index count, we can assume these are full screen quads. Usually
// these shaders are only used once for building textures so we can assume they can't be built
// async
if (regs.index_array.count <= 6 || regs.vertex_buffer.count <= 6) {
return false;
}
return true;
}
std::vector<AsyncShaders::Result> AsyncShaders::GetCompletedWork() {
std::vector<Result> results;
{
std::unique_lock lock{completed_mutex};
results = std::move(finished_work);
finished_work.clear();
}
return results;
}
void AsyncShaders::QueueOpenGLShader(const OpenGL::Device& device,
Tegra::Engines::ShaderType shader_type, u64 uid,
std::vector<u64> code, std::vector<u64> code_b,
u32 main_offset, CompilerSettings compiler_settings,
const Registry& registry, VAddr cpu_addr) {
std::unique_lock lock(queue_mutex);
pending_queue.push({
.backend = device.UseAssemblyShaders() ? Backend::GLASM : Backend::OpenGL,
.device = &device,
.shader_type = shader_type,
.uid = uid,
.code = std::move(code),
.code_b = std::move(code_b),
.main_offset = main_offset,
.compiler_settings = compiler_settings,
.registry = registry,
.cpu_address = cpu_addr,
.pp_cache = nullptr,
.vk_device = nullptr,
.scheduler = nullptr,
.descriptor_pool = nullptr,
.update_descriptor_queue = nullptr,
.bindings{},
.program{},
.key{},
.num_color_buffers = 0,
});
cv.notify_one();
}
void AsyncShaders::QueueVulkanShader(Vulkan::VKPipelineCache* pp_cache,
const Vulkan::Device& device, Vulkan::VKScheduler& scheduler,
Vulkan::VKDescriptorPool& descriptor_pool,
Vulkan::VKUpdateDescriptorQueue& update_descriptor_queue,
std::vector<VkDescriptorSetLayoutBinding> bindings,
Vulkan::SPIRVProgram program,
Vulkan::GraphicsPipelineCacheKey key, u32 num_color_buffers) {
std::unique_lock lock(queue_mutex);
pending_queue.push({
.backend = Backend::Vulkan,
.device = nullptr,
.shader_type{},
.uid = 0,
.code{},
.code_b{},
.main_offset = 0,
.compiler_settings{},
.registry{},
.cpu_address = 0,
.pp_cache = pp_cache,
.vk_device = &device,
.scheduler = &scheduler,
.descriptor_pool = &descriptor_pool,
.update_descriptor_queue = &update_descriptor_queue,
.bindings = std::move(bindings),
.program = std::move(program),
.key = key,
.num_color_buffers = num_color_buffers,
});
cv.notify_one();
}
void AsyncShaders::ShaderCompilerThread(Core::Frontend::GraphicsContext* context) {
while (!is_thread_exiting.load(std::memory_order_relaxed)) {
std::unique_lock lock{queue_mutex};
cv.wait(lock, [this] { return HasWorkQueued() || is_thread_exiting; });
if (is_thread_exiting) {
return;
}
// Partial lock to allow all threads to read at the same time
if (!HasWorkQueued()) {
continue;
}
// Another thread beat us, just unlock and wait for the next load
if (pending_queue.empty()) {
continue;
}
// Pull work from queue
WorkerParams work = std::move(pending_queue.front());
pending_queue.pop();
lock.unlock();
if (work.backend == Backend::OpenGL || work.backend == Backend::GLASM) {
const ShaderIR ir(work.code, work.main_offset, work.compiler_settings, *work.registry);
const auto scope = context->Acquire();
auto program =
OpenGL::BuildShader(*work.device, work.shader_type, work.uid, ir, *work.registry);
Result result{};
result.backend = work.backend;
result.cpu_address = work.cpu_address;
result.uid = work.uid;
result.code = std::move(work.code);
result.code_b = std::move(work.code_b);
result.shader_type = work.shader_type;
if (work.backend == Backend::OpenGL) {
result.program.opengl = std::move(program->source_program);
} else if (work.backend == Backend::GLASM) {
result.program.glasm = std::move(program->assembly_program);
}
{
std::unique_lock complete_lock(completed_mutex);
finished_work.push_back(std::move(result));
}
} else if (work.backend == Backend::Vulkan) {
auto pipeline = std::make_unique<Vulkan::VKGraphicsPipeline>(
*work.vk_device, *work.scheduler, *work.descriptor_pool,
*work.update_descriptor_queue, work.key, work.bindings, work.program,
work.num_color_buffers);
work.pp_cache->EmplacePipeline(std::move(pipeline));
}
}
}
} // namespace VideoCommon::Shader

138
src/video_core/shader/async_shaders.h
View file

@ -1,138 +0,0 @@
// Copyright 2020 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <condition_variable>
#include <memory>
#include <shared_mutex>
#include <thread>
#include <glad/glad.h>
#include "common/common_types.h"
#include "video_core/renderer_opengl/gl_device.h"
#include "video_core/renderer_opengl/gl_resource_manager.h"
#include "video_core/renderer_opengl/gl_shader_decompiler.h"
#include "video_core/renderer_vulkan/vk_pipeline_cache.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/vulkan_common/vulkan_device.h"
namespace Core::Frontend {
class EmuWindow;
class GraphicsContext;
} // namespace Core::Frontend
namespace Tegra {
class GPU;
}
namespace Vulkan {
class VKPipelineCache;
}
namespace VideoCommon::Shader {
class AsyncShaders {
public:
enum class Backend {
OpenGL,
GLASM,
Vulkan,
};
struct ResultPrograms {
OpenGL::OGLProgram opengl;
OpenGL::OGLAssemblyProgram glasm;
};
struct Result {
u64 uid;
VAddr cpu_address;
Backend backend;
ResultPrograms program;
std::vector<u64> code;
std::vector<u64> code_b;
Tegra::Engines::ShaderType shader_type;
};
explicit AsyncShaders(Core::Frontend::EmuWindow& emu_window_);
~AsyncShaders();
/// Start up shader worker threads
void AllocateWorkers();
/// Clear the shader queue and kill all worker threads
void FreeWorkers();
// Force end all threads
void KillWorkers();
/// Check to see if any shaders have actually been compiled
[[nodiscard]] bool HasCompletedWork() const;
/// Deduce if a shader can be build on another thread of MUST be built in sync. We cannot build
/// every shader async as some shaders are only built and executed once. We try to "guess" which
/// shader would be used only once
[[nodiscard]] bool IsShaderAsync(const Tegra::GPU& gpu) const;
/// Pulls completed compiled shaders
[[nodiscard]] std::vector<Result> GetCompletedWork();
void QueueOpenGLShader(const OpenGL::Device& device, Tegra::Engines::ShaderType shader_type,
u64 uid, std::vector<u64> code, std::vector<u64> code_b, u32 main_offset,
CompilerSettings compiler_settings, const Registry& registry,
VAddr cpu_addr);
void QueueVulkanShader(Vulkan::VKPipelineCache* pp_cache, const Vulkan::Device& device,
Vulkan::VKScheduler& scheduler,
Vulkan::VKDescriptorPool& descriptor_pool,
Vulkan::VKUpdateDescriptorQueue& update_descriptor_queue,
std::vector<VkDescriptorSetLayoutBinding> bindings,
Vulkan::SPIRVProgram program, Vulkan::GraphicsPipelineCacheKey key,
u32 num_color_buffers);
private:
void ShaderCompilerThread(Core::Frontend::GraphicsContext* context);
/// Check our worker queue to see if we have any work queued already
[[nodiscard]] bool HasWorkQueued() const;
struct WorkerParams {
Backend backend;
// For OGL
const OpenGL::Device* device;
Tegra::Engines::ShaderType shader_type;
u64 uid;
std::vector<u64> code;
std::vector<u64> code_b;
u32 main_offset;
CompilerSettings compiler_settings;
std::optional<Registry> registry;
VAddr cpu_address;
// For Vulkan
Vulkan::VKPipelineCache* pp_cache;
const Vulkan::Device* vk_device;
Vulkan::VKScheduler* scheduler;
Vulkan::VKDescriptorPool* descriptor_pool;
Vulkan::VKUpdateDescriptorQueue* update_descriptor_queue;
std::vector<VkDescriptorSetLayoutBinding> bindings;
Vulkan::SPIRVProgram program;
Vulkan::GraphicsPipelineCacheKey key;
u32 num_color_buffers;
};
std::condition_variable cv;
mutable std::mutex queue_mutex;
mutable std::shared_mutex completed_mutex;
std::atomic<bool> is_thread_exiting{};
std::vector<std::unique_ptr<Core::Frontend::GraphicsContext>> context_list;
std::vector<std::thread> worker_threads;
std::queue<WorkerParams> pending_queue;
std::vector<Result> finished_work;
Core::Frontend::EmuWindow& emu_window;
};
} // namespace VideoCommon::Shader

26
src/video_core/shader/compiler_settings.cpp
View file

@ -1,26 +0,0 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "video_core/shader/compiler_settings.h"
namespace VideoCommon::Shader {
std::string CompileDepthAsString(const CompileDepth cd) {
switch (cd) {
case CompileDepth::BruteForce:
return "Brute Force Compile";
case CompileDepth::FlowStack:
return "Simple Flow Stack Mode";
case CompileDepth::NoFlowStack:
return "Remove Flow Stack";
case CompileDepth::DecompileBackwards:
return "Decompile Backward Jumps";
case CompileDepth::FullDecompile:
return "Full Decompilation";
default:
return "Unknown Compiler Process";
}
}
} // namespace VideoCommon::Shader

26
src/video_core/shader/compiler_settings.h
View file

@ -1,26 +0,0 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include "video_core/engines/shader_bytecode.h"
namespace VideoCommon::Shader {
enum class CompileDepth : u32 {
BruteForce = 0,
FlowStack = 1,
NoFlowStack = 2,
DecompileBackwards = 3,
FullDecompile = 4,
};
std::string CompileDepthAsString(CompileDepth cd);
struct CompilerSettings {
CompileDepth depth{CompileDepth::NoFlowStack};
bool disable_else_derivation{true};
};
} // namespace VideoCommon::Shader

751
src/video_core/shader/control_flow.cpp
View file

@ -1,751 +0,0 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <list>
#include <map>
#include <set>
#include <stack>
#include <unordered_map>
#include <vector>
#include "common/assert.h"
#include "common/common_types.h"
#include "video_core/shader/ast.h"
#include "video_core/shader/control_flow.h"
#include "video_core/shader/memory_util.h"
#include "video_core/shader/registry.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
namespace {
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
constexpr s32 unassigned_branch = -2;
struct Query {
u32 address{};
std::stack<u32> ssy_stack{};
std::stack<u32> pbk_stack{};
};
struct BlockStack {
BlockStack() = default;
explicit BlockStack(const Query& q) : ssy_stack{q.ssy_stack}, pbk_stack{q.pbk_stack} {}
std::stack<u32> ssy_stack{};
std::stack<u32> pbk_stack{};
};
template <typename T, typename... Args>
BlockBranchInfo MakeBranchInfo(Args&&... args) {
static_assert(std::is_convertible_v<T, BranchData>);
return std::make_shared<BranchData>(T(std::forward<Args>(args)...));
}
bool BlockBranchIsIgnored(BlockBranchInfo first) {
bool ignore = false;
if (std::holds_alternative<SingleBranch>(*first)) {
const auto branch = std::get_if<SingleBranch>(first.get());
ignore = branch->ignore;
}
return ignore;
}
struct BlockInfo {
u32 start{};
u32 end{};
bool visited{};
BlockBranchInfo branch{};
bool IsInside(const u32 address) const {
return start <= address && address <= end;
}
};
struct CFGRebuildState {
explicit CFGRebuildState(const ProgramCode& program_code_, u32 start_, Registry& registry_)
: program_code{program_code_}, registry{registry_}, start{start_} {}
const ProgramCode& program_code;
Registry& registry;
u32 start{};
std::vector<BlockInfo> block_info;
std::list<u32> inspect_queries;
std::list<Query> queries;
std::unordered_map<u32, u32> registered;
std::set<u32> labels;
std::map<u32, u32> ssy_labels;
std::map<u32, u32> pbk_labels;
std::unordered_map<u32, BlockStack> stacks;
ASTManager* manager{};
};
enum class BlockCollision : u32 { None, Found, Inside };
std::pair<BlockCollision, u32> TryGetBlock(CFGRebuildState& state, u32 address) {
const auto& blocks = state.block_info;
for (u32 index = 0; index < blocks.size(); index++) {
if (blocks[index].start == address) {
return {BlockCollision::Found, index};
}
if (blocks[index].IsInside(address)) {
return {BlockCollision::Inside, index};
}
}
return {BlockCollision::None, 0xFFFFFFFF};
}
struct ParseInfo {
BlockBranchInfo branch_info{};
u32 end_address{};
};
BlockInfo& CreateBlockInfo(CFGRebuildState& state, u32 start, u32 end) {
auto& it = state.block_info.emplace_back();
it.start = start;
it.end = end;
const u32 index = static_cast<u32>(state.block_info.size() - 1);
state.registered.insert({start, index});
return it;
}
Pred GetPredicate(u32 index, bool negated) {
return static_cast<Pred>(static_cast<u64>(index) + (negated ? 8ULL : 0ULL));
}
enum class ParseResult : u32 {
ControlCaught,
BlockEnd,
AbnormalFlow,
};
struct BranchIndirectInfo {
u32 buffer{};
u32 offset{};
u32 entries{};
s32 relative_position{};
};
struct BufferInfo {
u32 index;
u32 offset;
};
std::optional<std::pair<s32, u64>> GetBRXInfo(const CFGRebuildState& state, u32& pos) {
const Instruction instr = state.program_code[pos];
const auto opcode = OpCode::Decode(instr);
if (opcode->get().GetId() != OpCode::Id::BRX) {
return std::nullopt;
}
if (instr.brx.constant_buffer != 0) {
return std::nullopt;
}
--pos;
return std::make_pair(instr.brx.GetBranchExtend(), instr.gpr8.Value());
}
template <typename Result, typename TestCallable, typename PackCallable>
// requires std::predicate<TestCallable, Instruction, const OpCode::Matcher&>
// requires std::invocable<PackCallable, Instruction, const OpCode::Matcher&>
std::optional<Result> TrackInstruction(const CFGRebuildState& state, u32& pos, TestCallable test,
PackCallable pack) {
for (; pos >= state.start; --pos) {
if (IsSchedInstruction(pos, state.start)) {
continue;
}
const Instruction instr = state.program_code[pos];
const auto opcode = OpCode::Decode(instr);
if (!opcode) {
continue;
}
if (test(instr, opcode->get())) {
--pos;
return std::make_optional(pack(instr, opcode->get()));
}
}
return std::nullopt;
}
std::optional<std::pair<BufferInfo, u64>> TrackLDC(const CFGRebuildState& state, u32& pos,
u64 brx_tracked_register) {
return TrackInstruction<std::pair<BufferInfo, u64>>(
state, pos,
[brx_tracked_register](auto instr, const auto& opcode) {
return opcode.GetId() == OpCode::Id::LD_C &&
instr.gpr0.Value() == brx_tracked_register &&
instr.ld_c.type.Value() == Tegra::Shader::UniformType::Single;
},
[](auto instr, const auto& opcode) {
const BufferInfo info = {static_cast<u32>(instr.cbuf36.index.Value()),
static_cast<u32>(instr.cbuf36.GetOffset())};
return std::make_pair(info, instr.gpr8.Value());
});
}
std::optional<u64> TrackSHLRegister(const CFGRebuildState& state, u32& pos,
u64 ldc_tracked_register) {
return TrackInstruction<u64>(
state, pos,
[ldc_tracked_register](auto instr, const auto& opcode) {
return opcode.GetId() == OpCode::Id::SHL_IMM &&
instr.gpr0.Value() == ldc_tracked_register;
},
[](auto instr, const auto&) { return instr.gpr8.Value(); });
}
std::optional<u32> TrackIMNMXValue(const CFGRebuildState& state, u32& pos,
u64 shl_tracked_register) {
return TrackInstruction<u32>(
state, pos,
[shl_tracked_register](auto instr, const auto& opcode) {
return opcode.GetId() == OpCode::Id::IMNMX_IMM &&
instr.gpr0.Value() == shl_tracked_register;
},
[](auto instr, const auto&) {
return static_cast<u32>(instr.alu.GetSignedImm20_20() + 1);
});
}
std::optional<BranchIndirectInfo> TrackBranchIndirectInfo(const CFGRebuildState& state, u32 pos) {
const auto brx_info = GetBRXInfo(state, pos);
if (!brx_info) {
return std::nullopt;
}
const auto [relative_position, brx_tracked_register] = *brx_info;
const auto ldc_info = TrackLDC(state, pos, brx_tracked_register);
if (!ldc_info) {
return std::nullopt;
}
const auto [buffer_info, ldc_tracked_register] = *ldc_info;
const auto shl_tracked_register = TrackSHLRegister(state, pos, ldc_tracked_register);
if (!shl_tracked_register) {
return std::nullopt;
}
const auto entries = TrackIMNMXValue(state, pos, *shl_tracked_register);
if (!entries) {
return std::nullopt;
}
return BranchIndirectInfo{buffer_info.index, buffer_info.offset, *entries, relative_position};
}
std::pair<ParseResult, ParseInfo> ParseCode(CFGRebuildState& state, u32 address) {
u32 offset = static_cast<u32>(address);
const u32 end_address = static_cast<u32>(state.program_code.size());
ParseInfo parse_info{};
SingleBranch single_branch{};
const auto insert_label = [](CFGRebuildState& rebuild_state, u32 label_address) {
const auto pair = rebuild_state.labels.emplace(label_address);
if (pair.second) {
rebuild_state.inspect_queries.push_back(label_address);
}
};
while (true) {
if (offset >= end_address) {
// ASSERT_OR_EXECUTE can't be used, as it ignores the break
ASSERT_MSG(false, "Shader passed the current limit!");
single_branch.address = exit_branch;
single_branch.ignore = false;
break;
}
if (state.registered.contains(offset)) {
single_branch.address = offset;
single_branch.ignore = true;
break;
}
if (IsSchedInstruction(offset, state.start)) {
offset++;
continue;
}
const Instruction instr = {state.program_code[offset]};
const auto opcode = OpCode::Decode(instr);
if (!opcode || opcode->get().GetType() != OpCode::Type::Flow) {
offset++;
continue;
}
switch (opcode->get().GetId()) {
case OpCode::Id::EXIT: {
const auto pred_index = static_cast<u32>(instr.pred.pred_index);
single_branch.condition.predicate = GetPredicate(pred_index, instr.negate_pred != 0);
if (single_branch.condition.predicate == Pred::NeverExecute) {
offset++;
continue;
}
const ConditionCode cc = instr.flow_condition_code;
single_branch.condition.cc = cc;
if (cc == ConditionCode::F) {
offset++;
continue;
}
single_branch.address = exit_branch;
single_branch.kill = false;
single_branch.is_sync = false;
single_branch.is_brk = false;
single_branch.ignore = false;
parse_info.end_address = offset;
parse_info.branch_info = MakeBranchInfo<SingleBranch>(
single_branch.condition, single_branch.address, single_branch.kill,
single_branch.is_sync, single_branch.is_brk, single_branch.ignore);
return {ParseResult::ControlCaught, parse_info};
}
case OpCode::Id::BRA: {
if (instr.bra.constant_buffer != 0) {
return {ParseResult::AbnormalFlow, parse_info};
}
const auto pred_index = static_cast<u32>(instr.pred.pred_index);
single_branch.condition.predicate = GetPredicate(pred_index, instr.negate_pred != 0);
if (single_branch.condition.predicate == Pred::NeverExecute) {
offset++;
continue;
}
const ConditionCode cc = instr.flow_condition_code;
single_branch.condition.cc = cc;
if (cc == ConditionCode::F) {
offset++;
continue;
}
const u32 branch_offset = offset + instr.bra.GetBranchTarget();
if (branch_offset == 0) {
single_branch.address = exit_branch;
} else {
single_branch.address = branch_offset;
}
insert_label(state, branch_offset);
single_branch.kill = false;
single_branch.is_sync = false;
single_branch.is_brk = false;
single_branch.ignore = false;
parse_info.end_address = offset;
parse_info.branch_info = MakeBranchInfo<SingleBranch>(
single_branch.condition, single_branch.address, single_branch.kill,
single_branch.is_sync, single_branch.is_brk, single_branch.ignore);
return {ParseResult::ControlCaught, parse_info};
}
case OpCode::Id::SYNC: {
const auto pred_index = static_cast<u32>(instr.pred.pred_index);
single_branch.condition.predicate = GetPredicate(pred_index, instr.negate_pred != 0);
if (single_branch.condition.predicate == Pred::NeverExecute) {
offset++;
continue;
}
const ConditionCode cc = instr.flow_condition_code;
single_branch.condition.cc = cc;
if (cc == ConditionCode::F) {
offset++;
continue;
}
single_branch.address = unassigned_branch;
single_branch.kill = false;
single_branch.is_sync = true;
single_branch.is_brk = false;
single_branch.ignore = false;
parse_info.end_address = offset;
parse_info.branch_info = MakeBranchInfo<SingleBranch>(
single_branch.condition, single_branch.address, single_branch.kill,
single_branch.is_sync, single_branch.is_brk, single_branch.ignore);
return {ParseResult::ControlCaught, parse_info};
}
case OpCode::Id::BRK: {
const auto pred_index = static_cast<u32>(instr.pred.pred_index);
single_branch.condition.predicate = GetPredicate(pred_index, instr.negate_pred != 0);
if (single_branch.condition.predicate == Pred::NeverExecute) {
offset++;
continue;
}
const ConditionCode cc = instr.flow_condition_code;
single_branch.condition.cc = cc;
if (cc == ConditionCode::F) {
offset++;
continue;
}
single_branch.address = unassigned_branch;
single_branch.kill = false;
single_branch.is_sync = false;
single_branch.is_brk = true;
single_branch.ignore = false;
parse_info.end_address = offset;
parse_info.branch_info = MakeBranchInfo<SingleBranch>(
single_branch.condition, single_branch.address, single_branch.kill,
single_branch.is_sync, single_branch.is_brk, single_branch.ignore);
return {ParseResult::ControlCaught, parse_info};
}
case OpCode::Id::KIL: {
const auto pred_index = static_cast<u32>(instr.pred.pred_index);
single_branch.condition.predicate = GetPredicate(pred_index, instr.negate_pred != 0);
if (single_branch.condition.predicate == Pred::NeverExecute) {
offset++;
continue;
}
const ConditionCode cc = instr.flow_condition_code;
single_branch.condition.cc = cc;
if (cc == ConditionCode::F) {
offset++;
continue;
}
single_branch.address = exit_branch;
single_branch.kill = true;
single_branch.is_sync = false;
single_branch.is_brk = false;
single_branch.ignore = false;
parse_info.end_address = offset;
parse_info.branch_info = MakeBranchInfo<SingleBranch>(
single_branch.condition, single_branch.address, single_branch.kill,
single_branch.is_sync, single_branch.is_brk, single_branch.ignore);
return {ParseResult::ControlCaught, parse_info};
}
case OpCode::Id::SSY: {
const u32 target = offset + instr.bra.GetBranchTarget();
insert_label(state, target);
state.ssy_labels.emplace(offset, target);
break;
}
case OpCode::Id::PBK: {
const u32 target = offset + instr.bra.GetBranchTarget();
insert_label(state, target);
state.pbk_labels.emplace(offset, target);
break;
}
case OpCode::Id::BRX: {
const auto tmp = TrackBranchIndirectInfo(state, offset);
if (!tmp) {
LOG_WARNING(HW_GPU, "BRX Track Unsuccesful");
return {ParseResult::AbnormalFlow, parse_info};
}
const auto result = *tmp;
const s32 pc_target = offset + result.relative_position;
std::vector<CaseBranch> branches;
for (u32 i = 0; i < result.entries; i++) {
auto key = state.registry.ObtainKey(result.buffer, result.offset + i * 4);
if (!key) {
return {ParseResult::AbnormalFlow, parse_info};
}
u32 value = *key;
u32 target = static_cast<u32>((value >> 3) + pc_target);
insert_label(state, target);
branches.emplace_back(value, target);
}
parse_info.end_address = offset;
parse_info.branch_info = MakeBranchInfo<MultiBranch>(
static_cast<u32>(instr.gpr8.Value()), std::move(branches));
return {ParseResult::ControlCaught, parse_info};
}
default:
break;
}
offset++;
}
single_branch.kill = false;
single_branch.is_sync = false;
single_branch.is_brk = false;
parse_info.end_address = offset - 1;
parse_info.branch_info = MakeBranchInfo<SingleBranch>(
single_branch.condition, single_branch.address, single_branch.kill, single_branch.is_sync,
single_branch.is_brk, single_branch.ignore);
return {ParseResult::BlockEnd, parse_info};
}
bool TryInspectAddress(CFGRebuildState& state) {
if (state.inspect_queries.empty()) {
return false;
}
const u32 address = state.inspect_queries.front();
state.inspect_queries.pop_front();
const auto [result, block_index] = TryGetBlock(state, address);
switch (result) {
case BlockCollision::Found: {
return true;
}
case BlockCollision::Inside: {
// This case is the tricky one:
// We need to split the block into 2 separate blocks
const u32 end = state.block_info[block_index].end;
BlockInfo& new_block = CreateBlockInfo(state, address, end);
BlockInfo& current_block = state.block_info[block_index];
current_block.end = address - 1;
new_block.branch = std::move(current_block.branch);
BlockBranchInfo forward_branch = MakeBranchInfo<SingleBranch>();
const auto branch = std::get_if<SingleBranch>(forward_branch.get());
branch->address = address;
branch->ignore = true;
current_block.branch = std::move(forward_branch);
return true;
}
default:
break;
}
const auto [parse_result, parse_info] = ParseCode(state, address);
if (parse_result == ParseResult::AbnormalFlow) {
// if it's AbnormalFlow, we end it as false, ending the CFG reconstruction
return false;
}
BlockInfo& block_info = CreateBlockInfo(state, address, parse_info.end_address);
block_info.branch = parse_info.branch_info;
if (std::holds_alternative<SingleBranch>(*block_info.branch)) {
const auto branch = std::get_if<SingleBranch>(block_info.branch.get());
if (branch->condition.IsUnconditional()) {
return true;
}
const u32 fallthrough_address = parse_info.end_address + 1;
state.inspect_queries.push_front(fallthrough_address);
return true;
}
return true;
}
bool TryQuery(CFGRebuildState& state) {
const auto gather_labels = [](std::stack<u32>& cc, std::map<u32, u32>& labels,
BlockInfo& block) {
auto gather_start = labels.lower_bound(block.start);
const auto gather_end = labels.upper_bound(block.end);
while (gather_start != gather_end) {
cc.push(gather_start->second);
++gather_start;
}
};
if (state.queries.empty()) {
return false;
}
Query& q = state.queries.front();
const u32 block_index = state.registered[q.address];
BlockInfo& block = state.block_info[block_index];
// If the block is visited, check if the stacks match, else gather the ssy/pbk
// labels into the current stack and look if the branch at the end of the block
// consumes a label. Schedule new queries accordingly
if (block.visited) {
BlockStack& stack = state.stacks[q.address];
const bool all_okay = (stack.ssy_stack.empty() || q.ssy_stack == stack.ssy_stack) &&
(stack.pbk_stack.empty() || q.pbk_stack == stack.pbk_stack);
state.queries.pop_front();
return all_okay;
}
block.visited = true;
state.stacks.insert_or_assign(q.address, BlockStack{q});
Query q2(q);
state.queries.pop_front();
gather_labels(q2.ssy_stack, state.ssy_labels, block);
gather_labels(q2.pbk_stack, state.pbk_labels, block);
if (std::holds_alternative<SingleBranch>(*block.branch)) {
auto* branch = std::get_if<SingleBranch>(block.branch.get());
if (!branch->condition.IsUnconditional()) {
q2.address = block.end + 1;
state.queries.push_back(q2);
}
auto& conditional_query = state.queries.emplace_back(q2);
if (branch->is_sync) {
if (branch->address == unassigned_branch) {
branch->address = conditional_query.ssy_stack.top();
}
conditional_query.ssy_stack.pop();
}
if (branch->is_brk) {
if (branch->address == unassigned_branch) {
branch->address = conditional_query.pbk_stack.top();
}
conditional_query.pbk_stack.pop();
}
conditional_query.address = branch->address;
return true;
}
const auto* multi_branch = std::get_if<MultiBranch>(block.branch.get());
for (const auto& branch_case : multi_branch->branches) {
auto& conditional_query = state.queries.emplace_back(q2);
conditional_query.address = branch_case.address;
}
return true;
}
void InsertBranch(ASTManager& mm, const BlockBranchInfo& branch_info) {
const auto get_expr = [](const Condition& cond) -> Expr {
Expr result;
if (cond.cc != ConditionCode::T) {
result = MakeExpr<ExprCondCode>(cond.cc);
}
if (cond.predicate != Pred::UnusedIndex) {
u32 pred = static_cast<u32>(cond.predicate);
bool negate = false;
if (pred > 7) {
negate = true;
pred -= 8;
}
Expr extra = MakeExpr<ExprPredicate>(pred);
if (negate) {
extra = MakeExpr<ExprNot>(std::move(extra));
}
if (result) {
return MakeExpr<ExprAnd>(std::move(extra), std::move(result));
}
return extra;
}
if (result) {
return result;
}
return MakeExpr<ExprBoolean>(true);
};
if (std::holds_alternative<SingleBranch>(*branch_info)) {
const auto* branch = std::get_if<SingleBranch>(branch_info.get());
if (branch->address < 0) {
if (branch->kill) {
mm.InsertReturn(get_expr(branch->condition), true);
return;
}
mm.InsertReturn(get_expr(branch->condition), false);
return;
}
mm.InsertGoto(get_expr(branch->condition), branch->address);
return;
}
const auto* multi_branch = std::get_if<MultiBranch>(branch_info.get());
for (const auto& branch_case : multi_branch->branches) {
mm.InsertGoto(MakeExpr<ExprGprEqual>(multi_branch->gpr, branch_case.cmp_value),
branch_case.address);
}
}
void DecompileShader(CFGRebuildState& state) {
state.manager->Init();
for (auto label : state.labels) {
state.manager->DeclareLabel(label);
}
for (const auto& block : state.block_info) {
if (state.labels.contains(block.start)) {
state.manager->InsertLabel(block.start);
}
const bool ignore = BlockBranchIsIgnored(block.branch);
const u32 end = ignore ? block.end + 1 : block.end;
state.manager->InsertBlock(block.start, end);
if (!ignore) {
InsertBranch(*state.manager, block.branch);
}
}
state.manager->Decompile();
}
} // Anonymous namespace
std::unique_ptr<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code, u32 start_address,
const CompilerSettings& settings,
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, registry};
// Inspect Code and generate blocks
state.labels.clear();
state.labels.emplace(start_address);
state.inspect_queries.push_back(state.start);
while (!state.inspect_queries.empty()) {
if (!TryInspectAddress(state)) {
result_out->settings.depth = CompileDepth::BruteForce;
return result_out;
}
}
bool use_flow_stack = true;
bool decompiled = false;
if (settings.depth != CompileDepth::FlowStack) {
// Decompile Stacks
state.queries.push_back(Query{state.start, {}, {}});
decompiled = true;
while (!state.queries.empty()) {
if (!TryQuery(state)) {
decompiled = false;
break;
}
}
}
use_flow_stack = !decompiled;
// Sort and organize results
std::sort(state.block_info.begin(), state.block_info.end(),
[](const BlockInfo& a, const BlockInfo& b) -> bool { return a.start < b.start; });
if (decompiled && settings.depth != CompileDepth::NoFlowStack) {
ASTManager manager{settings.depth != CompileDepth::DecompileBackwards,
settings.disable_else_derivation};
state.manager = &manager;
DecompileShader(state);
decompiled = state.manager->IsFullyDecompiled();
if (!decompiled) {
if (settings.depth == CompileDepth::FullDecompile) {
LOG_CRITICAL(HW_GPU, "Failed to remove all the gotos!:");
} else {
LOG_CRITICAL(HW_GPU, "Failed to remove all backward gotos!:");
}
state.manager->ShowCurrentState("Of Shader");
state.manager->Clear();
} else {
auto characteristics = std::make_unique<ShaderCharacteristics>();
characteristics->start = start_address;
characteristics->settings.depth = settings.depth;
characteristics->manager = std::move(manager);
characteristics->end = state.block_info.back().end + 1;
return characteristics;
}
}
result_out->start = start_address;
result_out->settings.depth =
use_flow_stack ? CompileDepth::FlowStack : CompileDepth::NoFlowStack;
result_out->blocks.clear();
for (auto& block : state.block_info) {
ShaderBlock new_block{};
new_block.start = block.start;
new_block.end = block.end;
new_block.ignore_branch = BlockBranchIsIgnored(block.branch);
if (!new_block.ignore_branch) {
new_block.branch = block.branch;
}
result_out->end = std::max(result_out->end, block.end);
result_out->blocks.push_back(new_block);
}
if (!use_flow_stack) {
result_out->labels = std::move(state.labels);
return result_out;
}
auto back = result_out->blocks.begin();
auto next = std::next(back);
while (next != result_out->blocks.end()) {
if (!state.labels.contains(next->start) && next->start == back->end + 1) {
back->end = next->end;
next = result_out->blocks.erase(next);
continue;
}
back = next;
++next;
}
return result_out;
}
} // namespace VideoCommon::Shader

117
src/video_core/shader/control_flow.h
View file

@ -1,117 +0,0 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <list>
#include <optional>
#include <set>
#include <variant>
#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 {
using Tegra::Shader::ConditionCode;
using Tegra::Shader::Pred;
constexpr s32 exit_branch = -1;
struct Condition {
Pred predicate{Pred::UnusedIndex};
ConditionCode cc{ConditionCode::T};
bool IsUnconditional() const {
return predicate == Pred::UnusedIndex && cc == ConditionCode::T;
}
bool operator==(const Condition& other) const {
return std::tie(predicate, cc) == std::tie(other.predicate, other.cc);
}
bool operator!=(const Condition& other) const {
return !operator==(other);
}
};
class SingleBranch {
public:
SingleBranch() = default;
explicit SingleBranch(Condition condition_, s32 address_, bool kill_, bool is_sync_,
bool is_brk_, bool ignore_)
: condition{condition_}, address{address_}, kill{kill_}, is_sync{is_sync_}, is_brk{is_brk_},
ignore{ignore_} {}
bool operator==(const SingleBranch& b) const {
return std::tie(condition, address, kill, is_sync, is_brk, ignore) ==
std::tie(b.condition, b.address, b.kill, b.is_sync, b.is_brk, b.ignore);
}
bool operator!=(const SingleBranch& b) const {
return !operator==(b);
}
Condition condition{};
s32 address{exit_branch};
bool kill{};
bool is_sync{};
bool is_brk{};
bool ignore{};
};
struct CaseBranch {
explicit CaseBranch(u32 cmp_value_, u32 address_) : cmp_value{cmp_value_}, address{address_} {}
u32 cmp_value;
u32 address;
};
class MultiBranch {
public:
explicit MultiBranch(u32 gpr_, std::vector<CaseBranch>&& branches_)
: gpr{gpr_}, branches{std::move(branches_)} {}
u32 gpr{};
std::vector<CaseBranch> branches{};
};
using BranchData = std::variant<SingleBranch, MultiBranch>;
using BlockBranchInfo = std::shared_ptr<BranchData>;
bool BlockBranchInfoAreEqual(BlockBranchInfo first, BlockBranchInfo second);
struct ShaderBlock {
u32 start{};
u32 end{};
bool ignore_branch{};
BlockBranchInfo branch{};
bool operator==(const ShaderBlock& sb) const {
return std::tie(start, end, ignore_branch) ==
std::tie(sb.start, sb.end, sb.ignore_branch) &&
BlockBranchInfoAreEqual(branch, sb.branch);
}
bool operator!=(const ShaderBlock& sb) const {
return !operator==(sb);
}
};
struct ShaderCharacteristics {
std::list<ShaderBlock> blocks{};
std::set<u32> labels{};
u32 start{};
u32 end{};
ASTManager manager{true, true};
CompilerSettings settings{};
};
std::unique_ptr<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code, u32 start_address,
const CompilerSettings& settings,
Registry& registry);
} // namespace VideoCommon::Shader

368
src/video_core/shader/decode.cpp
View file

@ -1,368 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <cstring>
#include <limits>
#include <set>
#include <fmt/format.h>
#include "common/assert.h"
#include "common/common_types.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/engines/shader_header.h"
#include "video_core/shader/control_flow.h"
#include "video_core/shader/memory_util.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
namespace {
void DeduceTextureHandlerSize(VideoCore::GuestDriverProfile& gpu_driver,
const std::list<SamplerEntry>& used_samplers) {
if (gpu_driver.IsTextureHandlerSizeKnown() || used_samplers.size() <= 1) {
return;
}
u32 count{};
std::vector<u32> bound_offsets;
for (const auto& sampler : used_samplers) {
if (sampler.is_bindless) {
continue;
}
++count;
bound_offsets.emplace_back(sampler.offset);
}
if (count > 1) {
gpu_driver.DeduceTextureHandlerSize(std::move(bound_offsets));
}
}
std::optional<u32> TryDeduceSamplerSize(const SamplerEntry& sampler_to_deduce,
VideoCore::GuestDriverProfile& gpu_driver,
const std::list<SamplerEntry>& used_samplers) {
const u32 base_offset = sampler_to_deduce.offset;
u32 max_offset{std::numeric_limits<u32>::max()};
for (const auto& sampler : used_samplers) {
if (sampler.is_bindless) {
continue;
}
if (sampler.offset > base_offset) {
max_offset = std::min(sampler.offset, max_offset);
}
}
if (max_offset == std::numeric_limits<u32>::max()) {
return std::nullopt;
}
return ((max_offset - base_offset) * 4) / gpu_driver.GetTextureHandlerSize();
}
} // Anonymous namespace
class ASTDecoder {
public:
explicit ASTDecoder(ShaderIR& ir_) : ir(ir_) {}
void operator()(ASTProgram& ast) {
ASTNode current = ast.nodes.GetFirst();
while (current) {
Visit(current);
current = current->GetNext();
}
}
void operator()(ASTIfThen& ast) {
ASTNode current = ast.nodes.GetFirst();
while (current) {
Visit(current);
current = current->GetNext();
}
}
void operator()(ASTIfElse& ast) {
ASTNode current = ast.nodes.GetFirst();
while (current) {
Visit(current);
current = current->GetNext();
}
}
void operator()(ASTBlockEncoded& ast) {}
void operator()(ASTBlockDecoded& ast) {}
void operator()(ASTVarSet& ast) {}
void operator()(ASTLabel& ast) {}
void operator()(ASTGoto& ast) {}
void operator()(ASTDoWhile& ast) {
ASTNode current = ast.nodes.GetFirst();
while (current) {
Visit(current);
current = current->GetNext();
}
}
void operator()(ASTReturn& ast) {}
void operator()(ASTBreak& ast) {}
void Visit(ASTNode& node) {
std::visit(*this, *node->GetInnerData());
if (node->IsBlockEncoded()) {
auto block = std::get_if<ASTBlockEncoded>(node->GetInnerData());
NodeBlock bb = ir.DecodeRange(block->start, block->end);
node->TransformBlockEncoded(std::move(bb));
}
}
private:
ShaderIR& ir;
};
void ShaderIR::Decode() {
std::memcpy(&header, program_code.data(), sizeof(Tegra::Shader::Header));
decompiled = false;
auto info = ScanFlow(program_code, main_offset, settings, registry);
auto& shader_info = *info;
coverage_begin = shader_info.start;
coverage_end = shader_info.end;
switch (shader_info.settings.depth) {
case CompileDepth::FlowStack: {
for (const auto& block : shader_info.blocks) {
basic_blocks.insert({block.start, DecodeRange(block.start, block.end + 1)});
}
break;
}
case CompileDepth::NoFlowStack: {
disable_flow_stack = true;
const auto insert_block = [this](NodeBlock& nodes, u32 label) {
if (label == static_cast<u32>(exit_branch)) {
return;
}
basic_blocks.insert({label, nodes});
};
const auto& blocks = shader_info.blocks;
NodeBlock current_block;
u32 current_label = static_cast<u32>(exit_branch);
for (const auto& block : blocks) {
if (shader_info.labels.contains(block.start)) {
insert_block(current_block, current_label);
current_block.clear();
current_label = block.start;
}
if (!block.ignore_branch) {
DecodeRangeInner(current_block, block.start, block.end);
InsertControlFlow(current_block, block);
} else {
DecodeRangeInner(current_block, block.start, block.end + 1);
}
}
insert_block(current_block, current_label);
break;
}
case CompileDepth::DecompileBackwards:
case CompileDepth::FullDecompile: {
program_manager = std::move(shader_info.manager);
disable_flow_stack = true;
decompiled = true;
ASTDecoder decoder{*this};
ASTNode program = GetASTProgram();
decoder.Visit(program);
break;
}
default:
LOG_CRITICAL(HW_GPU, "Unknown decompilation mode!");
[[fallthrough]];
case CompileDepth::BruteForce: {
const auto shader_end = static_cast<u32>(program_code.size());
coverage_begin = main_offset;
coverage_end = shader_end;
for (u32 label = main_offset; label < shader_end; ++label) {
basic_blocks.insert({label, DecodeRange(label, label + 1)});
}
break;
}
}
if (settings.depth != shader_info.settings.depth) {
LOG_WARNING(
HW_GPU, "Decompiling to this setting \"{}\" failed, downgrading to this setting \"{}\"",
CompileDepthAsString(settings.depth), CompileDepthAsString(shader_info.settings.depth));
}
}
NodeBlock ShaderIR::DecodeRange(u32 begin, u32 end) {
NodeBlock basic_block;
DecodeRangeInner(basic_block, begin, end);
return basic_block;
}
void ShaderIR::DecodeRangeInner(NodeBlock& bb, u32 begin, u32 end) {
for (u32 pc = begin; pc < (begin > end ? MAX_PROGRAM_LENGTH : end);) {
pc = DecodeInstr(bb, pc);
}
}
void ShaderIR::InsertControlFlow(NodeBlock& bb, const ShaderBlock& block) {
const auto apply_conditions = [&](const Condition& cond, Node n) -> Node {
Node result = n;
if (cond.cc != ConditionCode::T) {
result = Conditional(GetConditionCode(cond.cc), {result});
}
if (cond.predicate != Pred::UnusedIndex) {
u32 pred = static_cast<u32>(cond.predicate);
const bool is_neg = pred > 7;
if (is_neg) {
pred -= 8;
}
result = Conditional(GetPredicate(pred, is_neg), {result});
}
return result;
};
if (std::holds_alternative<SingleBranch>(*block.branch)) {
auto branch = std::get_if<SingleBranch>(block.branch.get());
if (branch->address < 0) {
if (branch->kill) {
Node n = Operation(OperationCode::Discard);
n = apply_conditions(branch->condition, n);
bb.push_back(n);
global_code.push_back(n);
return;
}
Node n = Operation(OperationCode::Exit);
n = apply_conditions(branch->condition, n);
bb.push_back(n);
global_code.push_back(n);
return;
}
Node n = Operation(OperationCode::Branch, Immediate(branch->address));
n = apply_conditions(branch->condition, n);
bb.push_back(n);
global_code.push_back(n);
return;
}
auto multi_branch = std::get_if<MultiBranch>(block.branch.get());
Node op_a = GetRegister(multi_branch->gpr);
for (auto& branch_case : multi_branch->branches) {
Node n = Operation(OperationCode::Branch, Immediate(branch_case.address));
Node op_b = Immediate(branch_case.cmp_value);
Node condition =
GetPredicateComparisonInteger(Tegra::Shader::PredCondition::EQ, false, op_a, op_b);
auto result = Conditional(condition, {n});
bb.push_back(result);
global_code.push_back(result);
}
}
u32 ShaderIR::DecodeInstr(NodeBlock& bb, u32 pc) {
// Ignore sched instructions when generating code.
if (IsSchedInstruction(pc, main_offset)) {
return pc + 1;
}
const Instruction instr = {program_code[pc]};
const auto opcode = OpCode::Decode(instr);
const u32 nv_address = ConvertAddressToNvidiaSpace(pc);
// Decoding failure
if (!opcode) {
UNIMPLEMENTED_MSG("Unhandled instruction: {0:x}", instr.value);
bb.push_back(Comment(fmt::format("{:05x} Unimplemented Shader instruction (0x{:016x})",
nv_address, instr.value)));
return pc + 1;
}
bb.push_back(Comment(
fmt::format("{:05x} {} (0x{:016x})", nv_address, opcode->get().GetName(), instr.value)));
using Tegra::Shader::Pred;
UNIMPLEMENTED_IF_MSG(instr.pred.full_pred == Pred::NeverExecute,
"NeverExecute predicate not implemented");
static const std::map<OpCode::Type, u32 (ShaderIR::*)(NodeBlock&, u32)> decoders = {
{OpCode::Type::Arithmetic, &ShaderIR::DecodeArithmetic},
{OpCode::Type::ArithmeticImmediate, &ShaderIR::DecodeArithmeticImmediate},
{OpCode::Type::Bfe, &ShaderIR::DecodeBfe},
{OpCode::Type::Bfi, &ShaderIR::DecodeBfi},
{OpCode::Type::Shift, &ShaderIR::DecodeShift},
{OpCode::Type::ArithmeticInteger, &ShaderIR::DecodeArithmeticInteger},
{OpCode::Type::ArithmeticIntegerImmediate, &ShaderIR::DecodeArithmeticIntegerImmediate},
{OpCode::Type::ArithmeticHalf, &ShaderIR::DecodeArithmeticHalf},
{OpCode::Type::ArithmeticHalfImmediate, &ShaderIR::DecodeArithmeticHalfImmediate},
{OpCode::Type::Ffma, &ShaderIR::DecodeFfma},
{OpCode::Type::Hfma2, &ShaderIR::DecodeHfma2},
{OpCode::Type::Conversion, &ShaderIR::DecodeConversion},
{OpCode::Type::Warp, &ShaderIR::DecodeWarp},
{OpCode::Type::Memory, &ShaderIR::DecodeMemory},
{OpCode::Type::Texture, &ShaderIR::DecodeTexture},
{OpCode::Type::Image, &ShaderIR::DecodeImage},
{OpCode::Type::FloatSetPredicate, &ShaderIR::DecodeFloatSetPredicate},
{OpCode::Type::IntegerSetPredicate, &ShaderIR::DecodeIntegerSetPredicate},
{OpCode::Type::HalfSetPredicate, &ShaderIR::DecodeHalfSetPredicate},
{OpCode::Type::PredicateSetRegister, &ShaderIR::DecodePredicateSetRegister},
{OpCode::Type::PredicateSetPredicate, &ShaderIR::DecodePredicateSetPredicate},
{OpCode::Type::RegisterSetPredicate, &ShaderIR::DecodeRegisterSetPredicate},
{OpCode::Type::FloatSet, &ShaderIR::DecodeFloatSet},
{OpCode::Type::IntegerSet, &ShaderIR::DecodeIntegerSet},
{OpCode::Type::HalfSet, &ShaderIR::DecodeHalfSet},
{OpCode::Type::Video, &ShaderIR::DecodeVideo},
{OpCode::Type::Xmad, &ShaderIR::DecodeXmad},
};
std::vector<Node> tmp_block;
if (const auto decoder = decoders.find(opcode->get().GetType()); decoder != decoders.end()) {
pc = (this->*decoder->second)(tmp_block, pc);
} else {
pc = DecodeOther(tmp_block, pc);
}
// Some instructions (like SSY) don't have a predicate field, they are always unconditionally
// executed.
const bool can_be_predicated = OpCode::IsPredicatedInstruction(opcode->get().GetId());
const auto pred_index = static_cast<u32>(instr.pred.pred_index);
if (can_be_predicated && pred_index != static_cast<u32>(Pred::UnusedIndex)) {
const Node conditional =
Conditional(GetPredicate(pred_index, instr.negate_pred != 0), std::move(tmp_block));
global_code.push_back(conditional);
bb.push_back(conditional);
} else {
for (auto& node : tmp_block) {
global_code.push_back(node);
bb.push_back(node);
}
}
return pc + 1;
}
void ShaderIR::PostDecode() {
// Deduce texture handler size if needed
auto gpu_driver = registry.AccessGuestDriverProfile();
DeduceTextureHandlerSize(gpu_driver, used_samplers);
// Deduce Indexed Samplers
if (!uses_indexed_samplers) {
return;
}
for (auto& sampler : used_samplers) {
if (!sampler.is_indexed) {
continue;
}
if (const auto size = TryDeduceSamplerSize(sampler, gpu_driver, used_samplers)) {
sampler.size = *size;
} else {
LOG_CRITICAL(HW_GPU, "Failed to deduce size of indexed sampler");
sampler.size = 1;
}
}
}
} // namespace VideoCommon::Shader

166
src/video_core/shader/decode/arithmetic.cpp
View file

@ -1,166 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/assert.h"
#include "common/common_types.h"
#include "common/logging/log.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
using Tegra::Shader::SubOp;
u32 ShaderIR::DecodeArithmetic(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
const auto opcode = OpCode::Decode(instr);
Node op_a = GetRegister(instr.gpr8);
Node op_b = [&] {
if (instr.is_b_imm) {
return GetImmediate19(instr);
} else if (instr.is_b_gpr) {
return GetRegister(instr.gpr20);
} else {
return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset());
}
}();
switch (opcode->get().GetId()) {
case OpCode::Id::MOV_C:
case OpCode::Id::MOV_R: {
// MOV does not have neither 'abs' nor 'neg' bits.
SetRegister(bb, instr.gpr0, op_b);
break;
}
case OpCode::Id::FMUL_C:
case OpCode::Id::FMUL_R:
case OpCode::Id::FMUL_IMM: {
// FMUL does not have 'abs' bits and only the second operand has a 'neg' bit.
if (instr.fmul.tab5cb8_2 != 0) {
LOG_DEBUG(HW_GPU, "FMUL tab5cb8_2({}) is not implemented",
instr.fmul.tab5cb8_2.Value());
}
if (instr.fmul.tab5c68_0 != 1) {
LOG_DEBUG(HW_GPU, "FMUL tab5cb8_0({}) is not implemented",
instr.fmul.tab5c68_0.Value());
}
op_b = GetOperandAbsNegFloat(op_b, false, instr.fmul.negate_b);
static constexpr std::array FmulPostFactor = {
1.000f, // None
0.500f, // Divide 2
0.250f, // Divide 4
0.125f, // Divide 8
8.000f, // Mul 8
4.000f, // Mul 4
2.000f, // Mul 2
};
if (instr.fmul.postfactor != 0) {
op_a = Operation(OperationCode::FMul, NO_PRECISE, op_a,
Immediate(FmulPostFactor[instr.fmul.postfactor]));
}
// TODO(Rodrigo): Should precise be used when there's a postfactor?
Node value = Operation(OperationCode::FMul, PRECISE, op_a, op_b);
value = GetSaturatedFloat(value, instr.alu.saturate_d);
SetInternalFlagsFromFloat(bb, value, instr.generates_cc);
SetRegister(bb, instr.gpr0, value);
break;
}
case OpCode::Id::FADD_C:
case OpCode::Id::FADD_R:
case OpCode::Id::FADD_IMM: {
op_a = GetOperandAbsNegFloat(op_a, instr.alu.abs_a, instr.alu.negate_a);
op_b = GetOperandAbsNegFloat(op_b, instr.alu.abs_b, instr.alu.negate_b);
Node value = Operation(OperationCode::FAdd, PRECISE, op_a, op_b);
value = GetSaturatedFloat(value, instr.alu.saturate_d);
SetInternalFlagsFromFloat(bb, value, instr.generates_cc);
SetRegister(bb, instr.gpr0, value);
break;
}
case OpCode::Id::MUFU: {
op_a = GetOperandAbsNegFloat(op_a, instr.alu.abs_a, instr.alu.negate_a);
Node value = [&]() {
switch (instr.sub_op) {
case SubOp::Cos:
return Operation(OperationCode::FCos, PRECISE, op_a);
case SubOp::Sin:
return Operation(OperationCode::FSin, PRECISE, op_a);
case SubOp::Ex2:
return Operation(OperationCode::FExp2, PRECISE, op_a);
case SubOp::Lg2:
return Operation(OperationCode::FLog2, PRECISE, op_a);
case SubOp::Rcp:
return Operation(OperationCode::FDiv, PRECISE, Immediate(1.0f), op_a);
case SubOp::Rsq:
return Operation(OperationCode::FInverseSqrt, PRECISE, op_a);
case SubOp::Sqrt:
return Operation(OperationCode::FSqrt, PRECISE, op_a);
default:
UNIMPLEMENTED_MSG("Unhandled MUFU sub op={0:x}", instr.sub_op.Value());
return Immediate(0);
}
}();
value = GetSaturatedFloat(value, instr.alu.saturate_d);
SetRegister(bb, instr.gpr0, value);
break;
}
case OpCode::Id::FMNMX_C:
case OpCode::Id::FMNMX_R:
case OpCode::Id::FMNMX_IMM: {
op_a = GetOperandAbsNegFloat(op_a, instr.alu.abs_a, instr.alu.negate_a);
op_b = GetOperandAbsNegFloat(op_b, instr.alu.abs_b, instr.alu.negate_b);
const Node condition = GetPredicate(instr.alu.fmnmx.pred, instr.alu.fmnmx.negate_pred != 0);
const Node min = Operation(OperationCode::FMin, NO_PRECISE, op_a, op_b);
const Node max = Operation(OperationCode::FMax, NO_PRECISE, op_a, op_b);
const Node value = Operation(OperationCode::Select, NO_PRECISE, condition, min, max);
SetInternalFlagsFromFloat(bb, value, instr.generates_cc);
SetRegister(bb, instr.gpr0, value);
break;
}
case OpCode::Id::FCMP_RR:
case OpCode::Id::FCMP_RC:
case OpCode::Id::FCMP_IMMR: {
UNIMPLEMENTED_IF(instr.fcmp.ftz == 0);
Node op_c = GetRegister(instr.gpr39);
Node comp = GetPredicateComparisonFloat(instr.fcmp.cond, std::move(op_c), Immediate(0.0f));
SetRegister(
bb, instr.gpr0,
Operation(OperationCode::Select, std::move(comp), std::move(op_a), std::move(op_b)));
break;
}
case OpCode::Id::RRO_C:
case OpCode::Id::RRO_R:
case OpCode::Id::RRO_IMM: {
LOG_DEBUG(HW_GPU, "(STUBBED) RRO used");
// Currently RRO is only implemented as a register move.
op_b = GetOperandAbsNegFloat(op_b, instr.alu.abs_b, instr.alu.negate_b);
SetRegister(bb, instr.gpr0, op_b);
break;
}
default:
UNIMPLEMENTED_MSG("Unhandled arithmetic instruction: {}", opcode->get().GetName());
}
return pc;
}
} // namespace VideoCommon::Shader

101
src/video_core/shader/decode/arithmetic_half.cpp
View file

@ -1,101 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/assert.h"
#include "common/common_types.h"
#include "common/logging/log.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
using Tegra::Shader::HalfType;
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
u32 ShaderIR::DecodeArithmeticHalf(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
const auto opcode = OpCode::Decode(instr);
bool negate_a = false;
bool negate_b = false;
bool absolute_a = false;
bool absolute_b = false;
switch (opcode->get().GetId()) {
case OpCode::Id::HADD2_R:
if (instr.alu_half.ftz == 0) {
LOG_DEBUG(HW_GPU, "{} without FTZ is not implemented", opcode->get().GetName());
}
negate_a = ((instr.value >> 43) & 1) != 0;
negate_b = ((instr.value >> 31) & 1) != 0;
absolute_a = ((instr.value >> 44) & 1) != 0;
absolute_b = ((instr.value >> 30) & 1) != 0;
break;
case OpCode::Id::HADD2_C:
if (instr.alu_half.ftz == 0) {
LOG_DEBUG(HW_GPU, "{} without FTZ is not implemented", opcode->get().GetName());
}
negate_a = ((instr.value >> 43) & 1) != 0;
negate_b = ((instr.value >> 56) & 1) != 0;
absolute_a = ((instr.value >> 44) & 1) != 0;
absolute_b = ((instr.value >> 54) & 1) != 0;
break;
case OpCode::Id::HMUL2_R:
negate_a = ((instr.value >> 43) & 1) != 0;
absolute_a = ((instr.value >> 44) & 1) != 0;
absolute_b = ((instr.value >> 30) & 1) != 0;
break;
case OpCode::Id::HMUL2_C:
negate_b = ((instr.value >> 31) & 1) != 0;
absolute_a = ((instr.value >> 44) & 1) != 0;
absolute_b = ((instr.value >> 54) & 1) != 0;
break;
default:
UNREACHABLE();
break;
}
Node op_a = UnpackHalfFloat(GetRegister(instr.gpr8), instr.alu_half.type_a);
op_a = GetOperandAbsNegHalf(op_a, absolute_a, negate_a);
auto [type_b, op_b] = [this, instr, opcode]() -> std::pair<HalfType, Node> {
switch (opcode->get().GetId()) {
case OpCode::Id::HADD2_C:
case OpCode::Id::HMUL2_C:
return {HalfType::F32, GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset())};
case OpCode::Id::HADD2_R:
case OpCode::Id::HMUL2_R:
return {instr.alu_half.type_b, GetRegister(instr.gpr20)};
default:
UNREACHABLE();
return {HalfType::F32, Immediate(0)};
}
}();
op_b = UnpackHalfFloat(op_b, type_b);
op_b = GetOperandAbsNegHalf(op_b, absolute_b, negate_b);
Node value = [this, opcode, op_a, op_b = op_b] {
switch (opcode->get().GetId()) {
case OpCode::Id::HADD2_C:
case OpCode::Id::HADD2_R:
return Operation(OperationCode::HAdd, PRECISE, op_a, op_b);
case OpCode::Id::HMUL2_C:
case OpCode::Id::HMUL2_R:
return Operation(OperationCode::HMul, PRECISE, op_a, op_b);
default:
UNIMPLEMENTED_MSG("Unhandled half float instruction: {}", opcode->get().GetName());
return Immediate(0);
}
}();
value = GetSaturatedHalfFloat(value, instr.alu_half.saturate);
value = HalfMerge(GetRegister(instr.gpr0), value, instr.alu_half.merge);
SetRegister(bb, instr.gpr0, value);
return pc;
}
} // namespace VideoCommon::Shader

54
src/video_core/shader/decode/arithmetic_half_immediate.cpp
View file

@ -1,54 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/assert.h"
#include "common/common_types.h"
#include "common/logging/log.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
u32 ShaderIR::DecodeArithmeticHalfImmediate(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
const auto opcode = OpCode::Decode(instr);
if (opcode->get().GetId() == OpCode::Id::HADD2_IMM) {
if (instr.alu_half_imm.ftz == 0) {
LOG_DEBUG(HW_GPU, "{} without FTZ is not implemented", opcode->get().GetName());
}
} else {
if (instr.alu_half_imm.precision != Tegra::Shader::HalfPrecision::FTZ) {
LOG_DEBUG(HW_GPU, "{} without FTZ is not implemented", opcode->get().GetName());
}
}
Node op_a = UnpackHalfFloat(GetRegister(instr.gpr8), instr.alu_half_imm.type_a);
op_a = GetOperandAbsNegHalf(op_a, instr.alu_half_imm.abs_a, instr.alu_half_imm.negate_a);
const Node op_b = UnpackHalfImmediate(instr, true);
Node value = [&]() {
switch (opcode->get().GetId()) {
case OpCode::Id::HADD2_IMM:
return Operation(OperationCode::HAdd, PRECISE, op_a, op_b);
case OpCode::Id::HMUL2_IMM:
return Operation(OperationCode::HMul, PRECISE, op_a, op_b);
default:
UNREACHABLE();
return Immediate(0);
}
}();
value = GetSaturatedHalfFloat(value, instr.alu_half_imm.saturate);
value = HalfMerge(GetRegister(instr.gpr0), value, instr.alu_half_imm.merge);
SetRegister(bb, instr.gpr0, value);
return pc;
}
} // namespace VideoCommon::Shader

53
src/video_core/shader/decode/arithmetic_immediate.cpp
View file

@ -1,53 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/assert.h"
#include "common/common_types.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
u32 ShaderIR::DecodeArithmeticImmediate(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
const auto opcode = OpCode::Decode(instr);
switch (opcode->get().GetId()) {
case OpCode::Id::MOV32_IMM: {
SetRegister(bb, instr.gpr0, GetImmediate32(instr));
break;
}
case OpCode::Id::FMUL32_IMM: {
Node value =
Operation(OperationCode::FMul, PRECISE, GetRegister(instr.gpr8), GetImmediate32(instr));
value = GetSaturatedFloat(value, instr.fmul32.saturate);
SetInternalFlagsFromFloat(bb, value, instr.op_32.generates_cc);
SetRegister(bb, instr.gpr0, value);
break;
}
case OpCode::Id::FADD32I: {
const Node op_a = GetOperandAbsNegFloat(GetRegister(instr.gpr8), instr.fadd32i.abs_a,
instr.fadd32i.negate_a);
const Node op_b = GetOperandAbsNegFloat(GetImmediate32(instr), instr.fadd32i.abs_b,
instr.fadd32i.negate_b);
const Node value = Operation(OperationCode::FAdd, PRECISE, op_a, op_b);
SetInternalFlagsFromFloat(bb, value, instr.op_32.generates_cc);
SetRegister(bb, instr.gpr0, value);
break;
}
default:
UNIMPLEMENTED_MSG("Unhandled arithmetic immediate instruction: {}",
opcode->get().GetName());
}
return pc;
}
} // namespace VideoCommon::Shader

375
src/video_core/shader/decode/arithmetic_integer.cpp
View file

@ -1,375 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/assert.h"
#include "common/common_types.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
using Tegra::Shader::IAdd3Height;
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
using Tegra::Shader::Pred;
using Tegra::Shader::Register;
u32 ShaderIR::DecodeArithmeticInteger(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
const auto opcode = OpCode::Decode(instr);
Node op_a = GetRegister(instr.gpr8);
Node op_b = [&]() {
if (instr.is_b_imm) {
return Immediate(instr.alu.GetSignedImm20_20());
} else if (instr.is_b_gpr) {
return GetRegister(instr.gpr20);
} else {
return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset());
}
}();
switch (opcode->get().GetId()) {
case OpCode::Id::IADD_C:
case OpCode::Id::IADD_R:
case OpCode::Id::IADD_IMM: {
UNIMPLEMENTED_IF_MSG(instr.alu.saturate_d, "IADD.SAT");
UNIMPLEMENTED_IF_MSG(instr.iadd.x && instr.generates_cc, "IADD.X Rd.CC");
op_a = GetOperandAbsNegInteger(op_a, false, instr.alu_integer.negate_a, true);
op_b = GetOperandAbsNegInteger(op_b, false, instr.alu_integer.negate_b, true);
Node value = Operation(OperationCode::UAdd, op_a, op_b);
if (instr.iadd.x) {
Node carry = GetInternalFlag(InternalFlag::Carry);
Node x = Operation(OperationCode::Select, std::move(carry), Immediate(1), Immediate(0));
value = Operation(OperationCode::UAdd, std::move(value), std::move(x));
}
if (instr.generates_cc) {
const Node i0 = Immediate(0);
Node zero = Operation(OperationCode::LogicalIEqual, value, i0);
Node sign = Operation(OperationCode::LogicalILessThan, value, i0);
Node carry = Operation(OperationCode::LogicalAddCarry, op_a, op_b);
Node pos_a = Operation(OperationCode::LogicalIGreaterThan, op_a, i0);
Node pos_b = Operation(OperationCode::LogicalIGreaterThan, op_b, i0);
Node pos = Operation(OperationCode::LogicalAnd, std::move(pos_a), std::move(pos_b));
Node overflow = Operation(OperationCode::LogicalAnd, pos, sign);
SetInternalFlag(bb, InternalFlag::Zero, std::move(zero));
SetInternalFlag(bb, InternalFlag::Sign, std::move(sign));
SetInternalFlag(bb, InternalFlag::Carry, std::move(carry));
SetInternalFlag(bb, InternalFlag::Overflow, std::move(overflow));
}
SetRegister(bb, instr.gpr0, std::move(value));
break;
}
case OpCode::Id::IADD3_C:
case OpCode::Id::IADD3_R:
case OpCode::Id::IADD3_IMM: {
Node op_c = GetRegister(instr.gpr39);
const auto ApplyHeight = [&](IAdd3Height height, Node value) {
switch (height) {
case IAdd3Height::None:
return value;
case IAdd3Height::LowerHalfWord:
return BitfieldExtract(value, 0, 16);
case IAdd3Height::UpperHalfWord:
return BitfieldExtract(value, 16, 16);
default:
UNIMPLEMENTED_MSG("Unhandled IADD3 height: {}", height);
return Immediate(0);
}
};
if (opcode->get().GetId() == OpCode::Id::IADD3_R) {
op_a = ApplyHeight(instr.iadd3.height_a, op_a);
op_b = ApplyHeight(instr.iadd3.height_b, op_b);
op_c = ApplyHeight(instr.iadd3.height_c, op_c);
}
op_a = GetOperandAbsNegInteger(op_a, false, instr.iadd3.neg_a, true);
op_b = GetOperandAbsNegInteger(op_b, false, instr.iadd3.neg_b, true);
op_c = GetOperandAbsNegInteger(op_c, false, instr.iadd3.neg_c, true);
const Node value = [&] {
Node add_ab = Operation(OperationCode::IAdd, NO_PRECISE, op_a, op_b);
if (opcode->get().GetId() != OpCode::Id::IADD3_R) {
return Operation(OperationCode::IAdd, NO_PRECISE, add_ab, op_c);
}
const Node shifted = [&] {
switch (instr.iadd3.mode) {
case Tegra::Shader::IAdd3Mode::RightShift:
// TODO(tech4me): According to
// https://envytools.readthedocs.io/en/latest/hw/graph/maxwell/cuda/int.html?highlight=iadd3
// The addition between op_a and op_b should be done in uint33, more
// investigation required
return Operation(OperationCode::ILogicalShiftRight, NO_PRECISE, add_ab,
Immediate(16));
case Tegra::Shader::IAdd3Mode::LeftShift:
return Operation(OperationCode::ILogicalShiftLeft, NO_PRECISE, add_ab,
Immediate(16));
default:
return add_ab;
}
}();
return Operation(OperationCode::IAdd, NO_PRECISE, shifted, op_c);
}();
SetInternalFlagsFromInteger(bb, value, instr.generates_cc);
SetRegister(bb, instr.gpr0, value);
break;
}
case OpCode::Id::ISCADD_C:
case OpCode::Id::ISCADD_R:
case OpCode::Id::ISCADD_IMM: {
UNIMPLEMENTED_IF_MSG(instr.generates_cc,
"Condition codes generation in ISCADD is not implemented");
op_a = GetOperandAbsNegInteger(op_a, false, instr.alu_integer.negate_a, true);
op_b = GetOperandAbsNegInteger(op_b, false, instr.alu_integer.negate_b, true);
const Node shift = Immediate(static_cast<u32>(instr.alu_integer.shift_amount));
const Node shifted_a = Operation(OperationCode::ILogicalShiftLeft, NO_PRECISE, op_a, shift);
const Node value = Operation(OperationCode::IAdd, NO_PRECISE, shifted_a, op_b);
SetInternalFlagsFromInteger(bb, value, instr.generates_cc);
SetRegister(bb, instr.gpr0, value);
break;
}
case OpCode::Id::POPC_C:
case OpCode::Id::POPC_R:
case OpCode::Id::POPC_IMM: {
if (instr.popc.invert) {
op_b = Operation(OperationCode::IBitwiseNot, NO_PRECISE, op_b);
}
const Node value = Operation(OperationCode::IBitCount, PRECISE, op_b);
SetRegister(bb, instr.gpr0, value);
break;
}
case OpCode::Id::FLO_R:
case OpCode::Id::FLO_C:
case OpCode::Id::FLO_IMM: {
Node value;
if (instr.flo.invert) {
op_b = Operation(OperationCode::IBitwiseNot, NO_PRECISE, std::move(op_b));
}
if (instr.flo.is_signed) {
value = Operation(OperationCode::IBitMSB, NO_PRECISE, std::move(op_b));
} else {
value = Operation(OperationCode::UBitMSB, NO_PRECISE, std::move(op_b));
}
if (instr.flo.sh) {
value =
Operation(OperationCode::UBitwiseXor, NO_PRECISE, std::move(value), Immediate(31));
}
SetRegister(bb, instr.gpr0, std::move(value));
break;
}
case OpCode::Id::SEL_C:
case OpCode::Id::SEL_R:
case OpCode::Id::SEL_IMM: {
const Node condition = GetPredicate(instr.sel.pred, instr.sel.neg_pred != 0);
const Node value = Operation(OperationCode::Select, PRECISE, condition, op_a, op_b);
SetRegister(bb, instr.gpr0, value);
break;
}
case OpCode::Id::ICMP_CR:
case OpCode::Id::ICMP_R:
case OpCode::Id::ICMP_RC:
case OpCode::Id::ICMP_IMM: {
const Node zero = Immediate(0);
const auto [op_rhs, test] = [&]() -> std::pair<Node, Node> {
switch (opcode->get().GetId()) {
case OpCode::Id::ICMP_CR:
return {GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset()),
GetRegister(instr.gpr39)};
case OpCode::Id::ICMP_R:
return {GetRegister(instr.gpr20), GetRegister(instr.gpr39)};
case OpCode::Id::ICMP_RC:
return {GetRegister(instr.gpr39),
GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset())};
case OpCode::Id::ICMP_IMM:
return {Immediate(instr.alu.GetSignedImm20_20()), GetRegister(instr.gpr39)};
default:
UNREACHABLE();
return {zero, zero};
}
}();
const Node op_lhs = GetRegister(instr.gpr8);
const Node comparison =
GetPredicateComparisonInteger(instr.icmp.cond, instr.icmp.is_signed != 0, test, zero);
SetRegister(bb, instr.gpr0, Operation(OperationCode::Select, comparison, op_lhs, op_rhs));
break;
}
case OpCode::Id::LOP_C:
case OpCode::Id::LOP_R:
case OpCode::Id::LOP_IMM: {
if (instr.alu.lop.invert_a)
op_a = Operation(OperationCode::IBitwiseNot, NO_PRECISE, op_a);
if (instr.alu.lop.invert_b)
op_b = Operation(OperationCode::IBitwiseNot, NO_PRECISE, op_b);
WriteLogicOperation(bb, instr.gpr0, instr.alu.lop.operation, op_a, op_b,
instr.alu.lop.pred_result_mode, instr.alu.lop.pred48,
instr.generates_cc);
break;
}
case OpCode::Id::LOP3_C:
case OpCode::Id::LOP3_R:
case OpCode::Id::LOP3_IMM: {
const Node op_c = GetRegister(instr.gpr39);
const Node lut = [&]() {
if (opcode->get().GetId() == OpCode::Id::LOP3_R) {
return Immediate(instr.alu.lop3.GetImmLut28());
} else {
return Immediate(instr.alu.lop3.GetImmLut48());
}
}();
WriteLop3Instruction(bb, instr.gpr0, op_a, op_b, op_c, lut, instr.generates_cc);
break;
}
case OpCode::Id::IMNMX_C:
case OpCode::Id::IMNMX_R:
case OpCode::Id::IMNMX_IMM: {
UNIMPLEMENTED_IF(instr.imnmx.exchange != Tegra::Shader::IMinMaxExchange::None);
const bool is_signed = instr.imnmx.is_signed;
const Node condition = GetPredicate(instr.imnmx.pred, instr.imnmx.negate_pred != 0);
const Node min = SignedOperation(OperationCode::IMin, is_signed, NO_PRECISE, op_a, op_b);
const Node max = SignedOperation(OperationCode::IMax, is_signed, NO_PRECISE, op_a, op_b);
const Node value = Operation(OperationCode::Select, NO_PRECISE, condition, min, max);
SetInternalFlagsFromInteger(bb, value, instr.generates_cc);
SetRegister(bb, instr.gpr0, value);
break;
}
case OpCode::Id::LEA_R2:
case OpCode::Id::LEA_R1:
case OpCode::Id::LEA_IMM:
case OpCode::Id::LEA_RZ:
case OpCode::Id::LEA_HI: {
auto [op_a_, op_b_, op_c_] = [&]() -> std::tuple<Node, Node, Node> {
switch (opcode->get().GetId()) {
case OpCode::Id::LEA_R2: {
return {GetRegister(instr.gpr20), GetRegister(instr.gpr39),
Immediate(static_cast<u32>(instr.lea.r2.entry_a))};
}
case OpCode::Id::LEA_R1: {
const bool neg = instr.lea.r1.neg != 0;
return {GetOperandAbsNegInteger(GetRegister(instr.gpr8), false, neg, true),
GetRegister(instr.gpr20),
Immediate(static_cast<u32>(instr.lea.r1.entry_a))};
}
case OpCode::Id::LEA_IMM: {
const bool neg = instr.lea.imm.neg != 0;
return {GetOperandAbsNegInteger(GetRegister(instr.gpr8), false, neg, true),
Immediate(static_cast<u32>(instr.lea.imm.entry_a)),
Immediate(static_cast<u32>(instr.lea.imm.entry_b))};
}
case OpCode::Id::LEA_RZ: {
const bool neg = instr.lea.rz.neg != 0;
return {GetConstBuffer(instr.lea.rz.cb_index, instr.lea.rz.cb_offset),
GetOperandAbsNegInteger(GetRegister(instr.gpr8), false, neg, true),
Immediate(static_cast<u32>(instr.lea.rz.entry_a))};
}
case OpCode::Id::LEA_HI:
default:
UNIMPLEMENTED_MSG("Unhandled LEA subinstruction: {}", opcode->get().GetName());
return {Immediate(static_cast<u32>(instr.lea.imm.entry_a)), GetRegister(instr.gpr8),
Immediate(static_cast<u32>(instr.lea.imm.entry_b))};
}
}();
UNIMPLEMENTED_IF_MSG(instr.lea.pred48 != static_cast<u64>(Pred::UnusedIndex),
"Unhandled LEA Predicate");
Node value =
Operation(OperationCode::ILogicalShiftLeft, std::move(op_a_), std::move(op_c_));
value = Operation(OperationCode::IAdd, std::move(op_b_), std::move(value));
SetRegister(bb, instr.gpr0, std::move(value));
break;
}
default:
UNIMPLEMENTED_MSG("Unhandled ArithmeticInteger instruction: {}", opcode->get().GetName());
}
return pc;
}
void ShaderIR::WriteLop3Instruction(NodeBlock& bb, Register dest, Node op_a, Node op_b, Node op_c,
Node imm_lut, bool sets_cc) {
const Node lop3_fast = [&](const Node na, const Node nb, const Node nc, const Node ttbl) {
Node value = Immediate(0);
const ImmediateNode imm = std::get<ImmediateNode>(*ttbl);
if (imm.GetValue() & 0x01) {
const Node a = Operation(OperationCode::IBitwiseNot, na);
const Node b = Operation(OperationCode::IBitwiseNot, nb);
const Node c = Operation(OperationCode::IBitwiseNot, nc);
Node r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, a, b);
r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, r, c);
value = Operation(OperationCode::IBitwiseOr, value, r);
}
if (imm.GetValue() & 0x02) {
const Node a = Operation(OperationCode::IBitwiseNot, na);
const Node b = Operation(OperationCode::IBitwiseNot, nb);
Node r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, a, b);
r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, r, nc);
value = Operation(OperationCode::IBitwiseOr, value, r);
}
if (imm.GetValue() & 0x04) {
const Node a = Operation(OperationCode::IBitwiseNot, na);
const Node c = Operation(OperationCode::IBitwiseNot, nc);
Node r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, a, nb);
r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, r, c);
value = Operation(OperationCode::IBitwiseOr, value, r);
}
if (imm.GetValue() & 0x08) {
const Node a = Operation(OperationCode::IBitwiseNot, na);
Node r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, a, nb);
r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, r, nc);
value = Operation(OperationCode::IBitwiseOr, value, r);
}
if (imm.GetValue() & 0x10) {
const Node b = Operation(OperationCode::IBitwiseNot, nb);
const Node c = Operation(OperationCode::IBitwiseNot, nc);
Node r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, na, b);
r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, r, c);
value = Operation(OperationCode::IBitwiseOr, value, r);
}
if (imm.GetValue() & 0x20) {
const Node b = Operation(OperationCode::IBitwiseNot, nb);
Node r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, na, b);
r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, r, nc);
value = Operation(OperationCode::IBitwiseOr, value, r);
}
if (imm.GetValue() & 0x40) {
const Node c = Operation(OperationCode::IBitwiseNot, nc);
Node r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, na, nb);
r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, r, c);
value = Operation(OperationCode::IBitwiseOr, value, r);
}
if (imm.GetValue() & 0x80) {
Node r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, na, nb);
r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, r, nc);
value = Operation(OperationCode::IBitwiseOr, value, r);
}
return value;
}(op_a, op_b, op_c, imm_lut);
SetInternalFlagsFromInteger(bb, lop3_fast, sets_cc);
SetRegister(bb, dest, lop3_fast);
}
} // namespace VideoCommon::Shader

99
src/video_core/shader/decode/arithmetic_integer_immediate.cpp
View file

@ -1,99 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/assert.h"
#include "common/common_types.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
using Tegra::Shader::Instruction;
using Tegra::Shader::LogicOperation;
using Tegra::Shader::OpCode;
using Tegra::Shader::Pred;
using Tegra::Shader::PredicateResultMode;
using Tegra::Shader::Register;
u32 ShaderIR::DecodeArithmeticIntegerImmediate(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
const auto opcode = OpCode::Decode(instr);
Node op_a = GetRegister(instr.gpr8);
Node op_b = Immediate(static_cast<s32>(instr.alu.imm20_32));
switch (opcode->get().GetId()) {
case OpCode::Id::IADD32I: {
UNIMPLEMENTED_IF_MSG(instr.iadd32i.saturate, "IADD32I saturation is not implemented");
op_a = GetOperandAbsNegInteger(std::move(op_a), false, instr.iadd32i.negate_a != 0, true);
Node value = Operation(OperationCode::IAdd, PRECISE, std::move(op_a), std::move(op_b));
SetInternalFlagsFromInteger(bb, value, instr.op_32.generates_cc != 0);
SetRegister(bb, instr.gpr0, std::move(value));
break;
}
case OpCode::Id::LOP32I: {
if (instr.alu.lop32i.invert_a) {
op_a = Operation(OperationCode::IBitwiseNot, NO_PRECISE, std::move(op_a));
}
if (instr.alu.lop32i.invert_b) {
op_b = Operation(OperationCode::IBitwiseNot, NO_PRECISE, std::move(op_b));
}
WriteLogicOperation(bb, instr.gpr0, instr.alu.lop32i.operation, std::move(op_a),
std::move(op_b), PredicateResultMode::None, Pred::UnusedIndex,
instr.op_32.generates_cc != 0);
break;
}
default:
UNIMPLEMENTED_MSG("Unhandled ArithmeticIntegerImmediate instruction: {}",
opcode->get().GetName());
}
return pc;
}
void ShaderIR::WriteLogicOperation(NodeBlock& bb, Register dest, LogicOperation logic_op, Node op_a,
Node op_b, PredicateResultMode predicate_mode, Pred predicate,
bool sets_cc) {
Node result = [&] {
switch (logic_op) {
case LogicOperation::And:
return Operation(OperationCode::IBitwiseAnd, PRECISE, std::move(op_a), std::move(op_b));
case LogicOperation::Or:
return Operation(OperationCode::IBitwiseOr, PRECISE, std::move(op_a), std::move(op_b));
case LogicOperation::Xor:
return Operation(OperationCode::IBitwiseXor, PRECISE, std::move(op_a), std::move(op_b));
case LogicOperation::PassB:
return op_b;
default:
UNIMPLEMENTED_MSG("Unimplemented logic operation={}", logic_op);
return Immediate(0);
}
}();
SetInternalFlagsFromInteger(bb, result, sets_cc);
SetRegister(bb, dest, result);
// Write the predicate value depending on the predicate mode.
switch (predicate_mode) {
case PredicateResultMode::None:
// Do nothing.
return;
case PredicateResultMode::NotZero: {
// Set the predicate to true if the result is not zero.
Node compare = Operation(OperationCode::LogicalINotEqual, std::move(result), Immediate(0));
SetPredicate(bb, static_cast<u64>(predicate), std::move(compare));
break;
}
default:
UNIMPLEMENTED_MSG("Unimplemented predicate result mode: {}", predicate_mode);
}
}
} // namespace VideoCommon::Shader

77
src/video_core/shader/decode/bfe.cpp
View file

@ -1,77 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/assert.h"
#include "common/common_types.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
u32 ShaderIR::DecodeBfe(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
const auto opcode = OpCode::Decode(instr);
Node op_a = GetRegister(instr.gpr8);
Node op_b = [&] {
switch (opcode->get().GetId()) {
case OpCode::Id::BFE_R:
return GetRegister(instr.gpr20);
case OpCode::Id::BFE_C:
return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset());
case OpCode::Id::BFE_IMM:
return Immediate(instr.alu.GetSignedImm20_20());
default:
UNREACHABLE();
return Immediate(0);
}
}();
UNIMPLEMENTED_IF_MSG(instr.bfe.rd_cc, "Condition codes in BFE is not implemented");
const bool is_signed = instr.bfe.is_signed;
// using reverse parallel method in
// https://graphics.stanford.edu/~seander/bithacks.html#ReverseParallel
// note for later if possible to implement faster method.
if (instr.bfe.brev) {
const auto swap = [&](u32 s, u32 mask) {
Node v1 =
SignedOperation(OperationCode::ILogicalShiftRight, is_signed, op_a, Immediate(s));
if (mask != 0) {
v1 = SignedOperation(OperationCode::IBitwiseAnd, is_signed, std::move(v1),
Immediate(mask));
}
Node v2 = op_a;
if (mask != 0) {
v2 = SignedOperation(OperationCode::IBitwiseAnd, is_signed, std::move(v2),
Immediate(mask));
}
v2 = SignedOperation(OperationCode::ILogicalShiftLeft, is_signed, std::move(v2),
Immediate(s));
return SignedOperation(OperationCode::IBitwiseOr, is_signed, std::move(v1),
std::move(v2));
};
op_a = swap(1, 0x55555555U);
op_a = swap(2, 0x33333333U);
op_a = swap(4, 0x0F0F0F0FU);
op_a = swap(8, 0x00FF00FFU);
op_a = swap(16, 0);
}
const auto offset = SignedOperation(OperationCode::IBitfieldExtract, is_signed, op_b,
Immediate(0), Immediate(8));
const auto bits = SignedOperation(OperationCode::IBitfieldExtract, is_signed, op_b,
Immediate(8), Immediate(8));
auto result = SignedOperation(OperationCode::IBitfieldExtract, is_signed, op_a, offset, bits);
SetRegister(bb, instr.gpr0, std::move(result));
return pc;
}
} // namespace VideoCommon::Shader

45
src/video_core/shader/decode/bfi.cpp
View file

@ -1,45 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/assert.h"
#include "common/common_types.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
u32 ShaderIR::DecodeBfi(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
const auto opcode = OpCode::Decode(instr);
const auto [packed_shift, base] = [&]() -> std::pair<Node, Node> {
switch (opcode->get().GetId()) {
case OpCode::Id::BFI_RC:
return {GetRegister(instr.gpr39),
GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset())};
case OpCode::Id::BFI_IMM_R:
return {Immediate(instr.alu.GetSignedImm20_20()), GetRegister(instr.gpr39)};
default:
UNREACHABLE();
return {Immediate(0), Immediate(0)};
}
}();
const Node insert = GetRegister(instr.gpr8);
const Node offset = BitfieldExtract(packed_shift, 0, 8);
const Node bits = BitfieldExtract(packed_shift, 8, 8);
const Node value =
Operation(OperationCode::UBitfieldInsert, PRECISE, base, insert, offset, bits);
SetInternalFlagsFromInteger(bb, value, instr.generates_cc);
SetRegister(bb, instr.gpr0, value);
return pc;
}
} // namespace VideoCommon::Shader

321
src/video_core/shader/decode/conversion.cpp
View file

@ -1,321 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <limits>
#include <optional>
#include <utility>
#include "common/assert.h"
#include "common/common_types.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
using Tegra::Shader::Register;
namespace {
constexpr OperationCode GetFloatSelector(u64 selector) {
return selector == 0 ? OperationCode::FCastHalf0 : OperationCode::FCastHalf1;
}
constexpr u32 SizeInBits(Register::Size size) {
switch (size) {
case Register::Size::Byte:
return 8;
case Register::Size::Short:
return 16;
case Register::Size::Word:
return 32;
case Register::Size::Long:
return 64;
}
return 0;
}
constexpr std::optional<std::pair<s32, s32>> IntegerSaturateBounds(Register::Size src_size,
Register::Size dst_size,
bool src_signed,
bool dst_signed) {
const u32 dst_bits = SizeInBits(dst_size);
if (src_size == Register::Size::Word && dst_size == Register::Size::Word) {
if (src_signed == dst_signed) {
return std::nullopt;
}
return std::make_pair(0, std::numeric_limits<s32>::max());
}
if (dst_signed) {
// Signed destination, clamp to [-128, 127] for instance
return std::make_pair(-(1 << (dst_bits - 1)), (1 << (dst_bits - 1)) - 1);
} else {
// Unsigned destination
if (dst_bits == 32) {
// Avoid shifting by 32, that is undefined behavior
return std::make_pair(0, s32(std::numeric_limits<u32>::max()));
}
return std::make_pair(0, (1 << dst_bits) - 1);
}
}
} // Anonymous namespace
u32 ShaderIR::DecodeConversion(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
const auto opcode = OpCode::Decode(instr);
switch (opcode->get().GetId()) {
case OpCode::Id::I2I_R:
case OpCode::Id::I2I_C:
case OpCode::Id::I2I_IMM: {
const bool src_signed = instr.conversion.is_input_signed;
const bool dst_signed = instr.conversion.is_output_signed;
const Register::Size src_size = instr.conversion.src_size;
const Register::Size dst_size = instr.conversion.dst_size;
const u32 selector = static_cast<u32>(instr.conversion.int_src.selector);
Node value = [this, instr, opcode] {
switch (opcode->get().GetId()) {
case OpCode::Id::I2I_R:
return GetRegister(instr.gpr20);
case OpCode::Id::I2I_C:
return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset());
case OpCode::Id::I2I_IMM:
return Immediate(instr.alu.GetSignedImm20_20());
default:
UNREACHABLE();
return Immediate(0);
}
}();
// Ensure the source selector is valid
switch (instr.conversion.src_size) {
case Register::Size::Byte:
break;
case Register::Size::Short:
ASSERT(selector == 0 || selector == 2);
break;
default:
ASSERT(selector == 0);
break;
}
if (src_size != Register::Size::Word || selector != 0) {
value = SignedOperation(OperationCode::IBitfieldExtract, src_signed, std::move(value),
Immediate(selector * 8), Immediate(SizeInBits(src_size)));
}
value = GetOperandAbsNegInteger(std::move(value), instr.conversion.abs_a,
instr.conversion.negate_a, src_signed);
if (instr.alu.saturate_d) {
if (src_signed && !dst_signed) {
Node is_negative = Operation(OperationCode::LogicalUGreaterEqual, value,
Immediate(1 << (SizeInBits(src_size) - 1)));
value = Operation(OperationCode::Select, std::move(is_negative), Immediate(0),
std::move(value));
// Simplify generated expressions, this can be removed without semantic impact
SetTemporary(bb, 0, std::move(value));
value = GetTemporary(0);
if (dst_size != Register::Size::Word) {
const Node limit = Immediate((1 << SizeInBits(dst_size)) - 1);
Node is_large =
Operation(OperationCode::LogicalUGreaterThan, std::move(value), limit);
value = Operation(OperationCode::Select, std::move(is_large), limit,
std::move(value));
}
} else if (const std::optional bounds =
IntegerSaturateBounds(src_size, dst_size, src_signed, dst_signed)) {
value = SignedOperation(OperationCode::IMax, src_signed, std::move(value),
Immediate(bounds->first));
value = SignedOperation(OperationCode::IMin, src_signed, std::move(value),
Immediate(bounds->second));
}
} else if (dst_size != Register::Size::Word) {
// No saturation, we only have to mask the result
Node mask = Immediate((1 << SizeInBits(dst_size)) - 1);
value = Operation(OperationCode::UBitwiseAnd, std::move(value), std::move(mask));
}
SetInternalFlagsFromInteger(bb, value, instr.generates_cc);
SetRegister(bb, instr.gpr0, std::move(value));
break;
}
case OpCode::Id::I2F_R:
case OpCode::Id::I2F_C:
case OpCode::Id::I2F_IMM: {
UNIMPLEMENTED_IF(instr.conversion.dst_size == Register::Size::Long);
UNIMPLEMENTED_IF_MSG(instr.generates_cc,
"Condition codes generation in I2F is not implemented");
Node value = [&] {
switch (opcode->get().GetId()) {
case OpCode::Id::I2F_R:
return GetRegister(instr.gpr20);
case OpCode::Id::I2F_C:
return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset());
case OpCode::Id::I2F_IMM:
return Immediate(instr.alu.GetSignedImm20_20());
default:
UNREACHABLE();
return Immediate(0);
}
}();
const bool input_signed = instr.conversion.is_input_signed;
if (const u32 offset = static_cast<u32>(instr.conversion.int_src.selector); offset > 0) {
ASSERT(instr.conversion.src_size == Register::Size::Byte ||
instr.conversion.src_size == Register::Size::Short);
if (instr.conversion.src_size == Register::Size::Short) {
ASSERT(offset == 0 || offset == 2);
}
value = SignedOperation(OperationCode::ILogicalShiftRight, input_signed,
std::move(value), Immediate(offset * 8));
}
value = ConvertIntegerSize(value, instr.conversion.src_size, input_signed);
value = GetOperandAbsNegInteger(value, instr.conversion.abs_a, false, input_signed);
value = SignedOperation(OperationCode::FCastInteger, input_signed, PRECISE, value);
value = GetOperandAbsNegFloat(value, false, instr.conversion.negate_a);
SetInternalFlagsFromFloat(bb, value, instr.generates_cc);
if (instr.conversion.dst_size == Register::Size::Short) {
value = Operation(OperationCode::HCastFloat, PRECISE, value);
}
SetRegister(bb, instr.gpr0, value);
break;
}
case OpCode::Id::F2F_R:
case OpCode::Id::F2F_C:
case OpCode::Id::F2F_IMM: {
UNIMPLEMENTED_IF(instr.conversion.dst_size == Register::Size::Long);
UNIMPLEMENTED_IF(instr.conversion.src_size == Register::Size::Long);
UNIMPLEMENTED_IF_MSG(instr.generates_cc,
"Condition codes generation in F2F is not implemented");
Node value = [&]() {
switch (opcode->get().GetId()) {
case OpCode::Id::F2F_R:
return GetRegister(instr.gpr20);
case OpCode::Id::F2F_C:
return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset());
case OpCode::Id::F2F_IMM:
return GetImmediate19(instr);
default:
UNREACHABLE();
return Immediate(0);
}
}();
if (instr.conversion.src_size == Register::Size::Short) {
value = Operation(GetFloatSelector(instr.conversion.float_src.selector), NO_PRECISE,
std::move(value));
} else {
ASSERT(instr.conversion.float_src.selector == 0);
}
value = GetOperandAbsNegFloat(value, instr.conversion.abs_a, instr.conversion.negate_a);
value = [&] {
if (instr.conversion.src_size != instr.conversion.dst_size) {
// Rounding operations only matter when the source and destination conversion size
// is the same.
return value;
}
switch (instr.conversion.f2f.GetRoundingMode()) {
case Tegra::Shader::F2fRoundingOp::None:
return value;
case Tegra::Shader::F2fRoundingOp::Round:
return Operation(OperationCode::FRoundEven, value);
case Tegra::Shader::F2fRoundingOp::Floor:
return Operation(OperationCode::FFloor, value);
case Tegra::Shader::F2fRoundingOp::Ceil:
return Operation(OperationCode::FCeil, value);
case Tegra::Shader::F2fRoundingOp::Trunc:
return Operation(OperationCode::FTrunc, value);
default:
UNIMPLEMENTED_MSG("Unimplemented F2F rounding mode {}",
instr.conversion.f2f.rounding.Value());
return value;
}
}();
value = GetSaturatedFloat(value, instr.alu.saturate_d);
SetInternalFlagsFromFloat(bb, value, instr.generates_cc);
if (instr.conversion.dst_size == Register::Size::Short) {
value = Operation(OperationCode::HCastFloat, PRECISE, value);
}
SetRegister(bb, instr.gpr0, value);
break;
}
case OpCode::Id::F2I_R:
case OpCode::Id::F2I_C:
case OpCode::Id::F2I_IMM: {
UNIMPLEMENTED_IF(instr.conversion.src_size == Register::Size::Long);
UNIMPLEMENTED_IF_MSG(instr.generates_cc,
"Condition codes generation in F2I is not implemented");
Node value = [&]() {
switch (opcode->get().GetId()) {
case OpCode::Id::F2I_R:
return GetRegister(instr.gpr20);
case OpCode::Id::F2I_C:
return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset());
case OpCode::Id::F2I_IMM:
return GetImmediate19(instr);
default:
UNREACHABLE();
return Immediate(0);
}
}();
if (instr.conversion.src_size == Register::Size::Short) {
value = Operation(GetFloatSelector(instr.conversion.float_src.selector), NO_PRECISE,
std::move(value));
} else {
ASSERT(instr.conversion.float_src.selector == 0);
}
value = GetOperandAbsNegFloat(value, instr.conversion.abs_a, instr.conversion.negate_a);
value = [&]() {
switch (instr.conversion.f2i.rounding) {
case Tegra::Shader::F2iRoundingOp::RoundEven:
return Operation(OperationCode::FRoundEven, PRECISE, value);
case Tegra::Shader::F2iRoundingOp::Floor:
return Operation(OperationCode::FFloor, PRECISE, value);
case Tegra::Shader::F2iRoundingOp::Ceil:
return Operation(OperationCode::FCeil, PRECISE, value);
case Tegra::Shader::F2iRoundingOp::Trunc:
return Operation(OperationCode::FTrunc, PRECISE, value);
default:
UNIMPLEMENTED_MSG("Unimplemented F2I rounding mode {}",
instr.conversion.f2i.rounding.Value());
return Immediate(0);
}
}();
const bool is_signed = instr.conversion.is_output_signed;
value = SignedOperation(OperationCode::ICastFloat, is_signed, PRECISE, value);
value = ConvertIntegerSize(value, instr.conversion.dst_size, is_signed);
SetRegister(bb, instr.gpr0, value);
break;
}
default:
UNIMPLEMENTED_MSG("Unhandled conversion instruction: {}", opcode->get().GetName());
}
return pc;
}
} // namespace VideoCommon::Shader

62
src/video_core/shader/decode/ffma.cpp
View file

@ -1,62 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/assert.h"
#include "common/common_types.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
u32 ShaderIR::DecodeFfma(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
const auto opcode = OpCode::Decode(instr);
UNIMPLEMENTED_IF_MSG(instr.ffma.cc != 0, "FFMA cc not implemented");
if (instr.ffma.tab5980_0 != 1) {
LOG_DEBUG(HW_GPU, "FFMA tab5980_0({}) not implemented", instr.ffma.tab5980_0.Value());
}
if (instr.ffma.tab5980_1 != 0) {
LOG_DEBUG(HW_GPU, "FFMA tab5980_1({}) not implemented", instr.ffma.tab5980_1.Value());
}
const Node op_a = GetRegister(instr.gpr8);
auto [op_b, op_c] = [&]() -> std::tuple<Node, Node> {
switch (opcode->get().GetId()) {
case OpCode::Id::FFMA_CR: {
return {GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset()),
GetRegister(instr.gpr39)};
}
case OpCode::Id::FFMA_RR:
return {GetRegister(instr.gpr20), GetRegister(instr.gpr39)};
case OpCode::Id::FFMA_RC: {
return {GetRegister(instr.gpr39),
GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset())};
}
case OpCode::Id::FFMA_IMM:
return {GetImmediate19(instr), GetRegister(instr.gpr39)};
default:
UNIMPLEMENTED_MSG("Unhandled FFMA instruction: {}", opcode->get().GetName());
return {Immediate(0), Immediate(0)};
}
}();
op_b = GetOperandAbsNegFloat(op_b, false, instr.ffma.negate_b);
op_c = GetOperandAbsNegFloat(op_c, false, instr.ffma.negate_c);
Node value = Operation(OperationCode::FFma, PRECISE, op_a, op_b, op_c);
value = GetSaturatedFloat(value, instr.alu.saturate_d);
SetInternalFlagsFromFloat(bb, value, instr.generates_cc);
SetRegister(bb, instr.gpr0, value);
return pc;
}
} // namespace VideoCommon::Shader

58
src/video_core/shader/decode/float_set.cpp
View file

@ -1,58 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/assert.h"
#include "common/common_types.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
u32 ShaderIR::DecodeFloatSet(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
const Node op_a = GetOperandAbsNegFloat(GetRegister(instr.gpr8), instr.fset.abs_a != 0,
instr.fset.neg_a != 0);
Node op_b = [&]() {
if (instr.is_b_imm) {
return GetImmediate19(instr);
} else if (instr.is_b_gpr) {
return GetRegister(instr.gpr20);
} else {
return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset());
}
}();
op_b = GetOperandAbsNegFloat(op_b, instr.fset.abs_b != 0, instr.fset.neg_b != 0);
// The fset instruction sets a register to 1.0 or -1 (depending on the bf bit) if the
// condition is true, and to 0 otherwise.
const Node second_pred = GetPredicate(instr.fset.pred39, instr.fset.neg_pred != 0);
const OperationCode combiner = GetPredicateCombiner(instr.fset.op);
const Node first_pred = GetPredicateComparisonFloat(instr.fset.cond, op_a, op_b);
const Node predicate = Operation(combiner, first_pred, second_pred);
const Node true_value = instr.fset.bf ? Immediate(1.0f) : Immediate(-1);
const Node false_value = instr.fset.bf ? Immediate(0.0f) : Immediate(0);
const Node value =
Operation(OperationCode::Select, PRECISE, predicate, true_value, false_value);
if (instr.fset.bf) {
SetInternalFlagsFromFloat(bb, value, instr.generates_cc);
} else {
SetInternalFlagsFromInteger(bb, value, instr.generates_cc);
}
SetRegister(bb, instr.gpr0, value);
return pc;
}
} // namespace VideoCommon::Shader

57
src/video_core/shader/decode/float_set_predicate.cpp
View file

@ -1,57 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/assert.h"
#include "common/common_types.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
using Tegra::Shader::Pred;
u32 ShaderIR::DecodeFloatSetPredicate(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
Node op_a = GetOperandAbsNegFloat(GetRegister(instr.gpr8), instr.fsetp.abs_a != 0,
instr.fsetp.neg_a != 0);
Node op_b = [&]() {
if (instr.is_b_imm) {
return GetImmediate19(instr);
} else if (instr.is_b_gpr) {
return GetRegister(instr.gpr20);
} else {
return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset());
}
}();
op_b = GetOperandAbsNegFloat(std::move(op_b), instr.fsetp.abs_b, instr.fsetp.neg_b);
// We can't use the constant predicate as destination.
ASSERT(instr.fsetp.pred3 != static_cast<u64>(Pred::UnusedIndex));
const Node predicate =
GetPredicateComparisonFloat(instr.fsetp.cond, std::move(op_a), std::move(op_b));
const Node second_pred = GetPredicate(instr.fsetp.pred39, instr.fsetp.neg_pred != 0);
const OperationCode combiner = GetPredicateCombiner(instr.fsetp.op);
const Node value = Operation(combiner, predicate, second_pred);
// Set the primary predicate to the result of Predicate OP SecondPredicate
SetPredicate(bb, instr.fsetp.pred3, value);
if (instr.fsetp.pred0 != static_cast<u64>(Pred::UnusedIndex)) {
// Set the secondary predicate to the result of !Predicate OP SecondPredicate,
// if enabled
const Node negated_pred = Operation(OperationCode::LogicalNegate, predicate);
const Node second_value = Operation(combiner, negated_pred, second_pred);
SetPredicate(bb, instr.fsetp.pred0, second_value);
}
return pc;
}
} // namespace VideoCommon::Shader

115
src/video_core/shader/decode/half_set.cpp
View file

@ -1,115 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <array>
#include "common/assert.h"
#include "common/common_types.h"
#include "common/logging/log.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
using std::move;
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
using Tegra::Shader::PredCondition;
u32 ShaderIR::DecodeHalfSet(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
const auto opcode = OpCode::Decode(instr);
PredCondition cond{};
bool bf = false;
bool ftz = false;
bool neg_a = false;
bool abs_a = false;
bool neg_b = false;
bool abs_b = false;
switch (opcode->get().GetId()) {
case OpCode::Id::HSET2_C:
case OpCode::Id::HSET2_IMM:
cond = instr.hsetp2.cbuf_and_imm.cond;
bf = instr.Bit(53);
ftz = instr.Bit(54);
neg_a = instr.Bit(43);
abs_a = instr.Bit(44);
neg_b = instr.Bit(56);
abs_b = instr.Bit(54);
break;
case OpCode::Id::HSET2_R:
cond = instr.hsetp2.reg.cond;
bf = instr.Bit(49);
ftz = instr.Bit(50);
neg_a = instr.Bit(43);
abs_a = instr.Bit(44);
neg_b = instr.Bit(31);
abs_b = instr.Bit(30);
break;
default:
UNREACHABLE();
}
Node op_b = [this, instr, opcode] {
switch (opcode->get().GetId()) {
case OpCode::Id::HSET2_C:
// Inform as unimplemented as this is not tested.
UNIMPLEMENTED_MSG("HSET2_C is not implemented");
return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset());
case OpCode::Id::HSET2_R:
return GetRegister(instr.gpr20);
case OpCode::Id::HSET2_IMM:
return UnpackHalfImmediate(instr, true);
default:
UNREACHABLE();
return Node{};
}
}();
if (!ftz) {
LOG_DEBUG(HW_GPU, "{} without FTZ is not implemented", opcode->get().GetName());
}
Node op_a = UnpackHalfFloat(GetRegister(instr.gpr8), instr.hset2.type_a);
op_a = GetOperandAbsNegHalf(op_a, abs_a, neg_a);
switch (opcode->get().GetId()) {
case OpCode::Id::HSET2_R:
op_b = GetOperandAbsNegHalf(move(op_b), abs_b, neg_b);
[[fallthrough]];
case OpCode::Id::HSET2_C:
op_b = UnpackHalfFloat(move(op_b), instr.hset2.type_b);
break;
default:
break;
}
Node second_pred = GetPredicate(instr.hset2.pred39, instr.hset2.neg_pred);
Node comparison_pair = GetPredicateComparisonHalf(cond, op_a, op_b);
const OperationCode combiner = GetPredicateCombiner(instr.hset2.op);
// HSET2 operates on each half float in the pack.
std::array<Node, 2> values;
for (u32 i = 0; i < 2; ++i) {
const u32 raw_value = bf ? 0x3c00 : 0xffff;
Node true_value = Immediate(raw_value << (i * 16));
Node false_value = Immediate(0);
Node comparison = Operation(OperationCode::LogicalPick2, comparison_pair, Immediate(i));
Node predicate = Operation(combiner, comparison, second_pred);
values[i] =
Operation(OperationCode::Select, predicate, move(true_value), move(false_value));
}
Node value = Operation(OperationCode::UBitwiseOr, values[0], values[1]);
SetRegister(bb, instr.gpr0, move(value));
return pc;
}
} // namespace VideoCommon::Shader

80
src/video_core/shader/decode/half_set_predicate.cpp
View file

@ -1,80 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/assert.h"
#include "common/common_types.h"
#include "common/logging/log.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
using Tegra::Shader::Pred;
u32 ShaderIR::DecodeHalfSetPredicate(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
const auto opcode = OpCode::Decode(instr);
if (instr.hsetp2.ftz != 0) {
LOG_DEBUG(HW_GPU, "{} without FTZ is not implemented", opcode->get().GetName());
}
Node op_a = UnpackHalfFloat(GetRegister(instr.gpr8), instr.hsetp2.type_a);
op_a = GetOperandAbsNegHalf(op_a, instr.hsetp2.abs_a, instr.hsetp2.negate_a);
Tegra::Shader::PredCondition cond{};
bool h_and{};
Node op_b{};
switch (opcode->get().GetId()) {
case OpCode::Id::HSETP2_C:
cond = instr.hsetp2.cbuf_and_imm.cond;
h_and = instr.hsetp2.cbuf_and_imm.h_and;
op_b = GetOperandAbsNegHalf(GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset()),
instr.hsetp2.cbuf.abs_b, instr.hsetp2.cbuf.negate_b);
// F32 is hardcoded in hardware
op_b = UnpackHalfFloat(std::move(op_b), Tegra::Shader::HalfType::F32);
break;
case OpCode::Id::HSETP2_IMM:
cond = instr.hsetp2.cbuf_and_imm.cond;
h_and = instr.hsetp2.cbuf_and_imm.h_and;
op_b = UnpackHalfImmediate(instr, true);
break;
case OpCode::Id::HSETP2_R:
cond = instr.hsetp2.reg.cond;
h_and = instr.hsetp2.reg.h_and;
op_b =
GetOperandAbsNegHalf(UnpackHalfFloat(GetRegister(instr.gpr20), instr.hsetp2.reg.type_b),
instr.hsetp2.reg.abs_b, instr.hsetp2.reg.negate_b);
break;
default:
UNREACHABLE();
op_b = Immediate(0);
}
const OperationCode combiner = GetPredicateCombiner(instr.hsetp2.op);
const Node combined_pred = GetPredicate(instr.hsetp2.pred39, instr.hsetp2.neg_pred);
const auto Write = [&](u64 dest, Node src) {
SetPredicate(bb, dest, Operation(combiner, std::move(src), combined_pred));
};
const Node comparison = GetPredicateComparisonHalf(cond, op_a, op_b);
const u64 first = instr.hsetp2.pred3;
const u64 second = instr.hsetp2.pred0;
if (h_and) {
Node joined = Operation(OperationCode::LogicalAnd2, comparison);
Write(first, joined);
Write(second, Operation(OperationCode::LogicalNegate, std::move(joined)));
} else {
Write(first, Operation(OperationCode::LogicalPick2, comparison, Immediate(0U)));
Write(second, Operation(OperationCode::LogicalPick2, comparison, Immediate(1U)));
}
return pc;
}
} // namespace VideoCommon::Shader

73
src/video_core/shader/decode/hfma2.cpp
View file

@ -1,73 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <tuple>
#include "common/assert.h"
#include "common/common_types.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
using Tegra::Shader::HalfPrecision;
using Tegra::Shader::HalfType;
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
u32 ShaderIR::DecodeHfma2(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
const auto opcode = OpCode::Decode(instr);
if (opcode->get().GetId() == OpCode::Id::HFMA2_RR) {
DEBUG_ASSERT(instr.hfma2.rr.precision == HalfPrecision::None);
} else {
DEBUG_ASSERT(instr.hfma2.precision == HalfPrecision::None);
}
constexpr auto identity = HalfType::H0_H1;
bool neg_b{}, neg_c{};
auto [saturate, type_b, op_b, type_c,
op_c] = [&]() -> std::tuple<bool, HalfType, Node, HalfType, Node> {
switch (opcode->get().GetId()) {
case OpCode::Id::HFMA2_CR:
neg_b = instr.hfma2.negate_b;
neg_c = instr.hfma2.negate_c;
return {instr.hfma2.saturate, HalfType::F32,
GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset()),
instr.hfma2.type_reg39, GetRegister(instr.gpr39)};
case OpCode::Id::HFMA2_RC:
neg_b = instr.hfma2.negate_b;
neg_c = instr.hfma2.negate_c;
return {instr.hfma2.saturate, instr.hfma2.type_reg39, GetRegister(instr.gpr39),
HalfType::F32, GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset())};
case OpCode::Id::HFMA2_RR:
neg_b = instr.hfma2.rr.negate_b;
neg_c = instr.hfma2.rr.negate_c;
return {instr.hfma2.rr.saturate, instr.hfma2.type_b, GetRegister(instr.gpr20),
instr.hfma2.rr.type_c, GetRegister(instr.gpr39)};
case OpCode::Id::HFMA2_IMM_R:
neg_c = instr.hfma2.negate_c;
return {instr.hfma2.saturate, identity, UnpackHalfImmediate(instr, true),
instr.hfma2.type_reg39, GetRegister(instr.gpr39)};
default:
return {false, identity, Immediate(0), identity, Immediate(0)};
}
}();
const Node op_a = UnpackHalfFloat(GetRegister(instr.gpr8), instr.hfma2.type_a);
op_b = GetOperandAbsNegHalf(UnpackHalfFloat(op_b, type_b), false, neg_b);
op_c = GetOperandAbsNegHalf(UnpackHalfFloat(op_c, type_c), false, neg_c);
Node value = Operation(OperationCode::HFma, PRECISE, op_a, op_b, op_c);
value = GetSaturatedHalfFloat(value, saturate);
value = HalfMerge(GetRegister(instr.gpr0), value, instr.hfma2.merge);
SetRegister(bb, instr.gpr0, value);
return pc;
}
} // namespace VideoCommon::Shader

536
src/video_core/shader/decode/image.cpp
View file

@ -1,536 +0,0 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <algorithm>
#include <vector>
#include <fmt/format.h>
#include "common/assert.h"
#include "common/bit_field.h"
#include "common/common_types.h"
#include "common/logging/log.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/shader_ir.h"
#include "video_core/textures/texture.h"
namespace VideoCommon::Shader {
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
using Tegra::Shader::PredCondition;
using Tegra::Shader::StoreType;
using Tegra::Texture::ComponentType;
using Tegra::Texture::TextureFormat;
using Tegra::Texture::TICEntry;
namespace {
ComponentType GetComponentType(Tegra::Engines::SamplerDescriptor descriptor,
std::size_t component) {
const TextureFormat format{descriptor.format};
switch (format) {
case TextureFormat::R16G16B16A16:
case TextureFormat::R32G32B32A32:
case TextureFormat::R32G32B32:
case TextureFormat::R32G32:
case TextureFormat::R16G16:
case TextureFormat::R32:
case TextureFormat::R16:
case TextureFormat::R8:
case TextureFormat::R1:
if (component == 0) {
return descriptor.r_type;
}
if (component == 1) {
return descriptor.g_type;
}
if (component == 2) {
return descriptor.b_type;
}
if (component == 3) {
return descriptor.a_type;
}
break;
case TextureFormat::A8R8G8B8:
if (component == 0) {
return descriptor.a_type;
}
if (component == 1) {
return descriptor.r_type;
}
if (component == 2) {
return descriptor.g_type;
}
if (component == 3) {
return descriptor.b_type;
}
break;
case TextureFormat::A2B10G10R10:
case TextureFormat::A4B4G4R4:
case TextureFormat::A5B5G5R1:
case TextureFormat::A1B5G5R5:
if (component == 0) {
return descriptor.a_type;
}
if (component == 1) {
return descriptor.b_type;
}
if (component == 2) {
return descriptor.g_type;
}
if (component == 3) {
return descriptor.r_type;
}
break;
case TextureFormat::R32_B24G8:
if (component == 0) {
return descriptor.r_type;
}
if (component == 1) {
return descriptor.b_type;
}
if (component == 2) {
return descriptor.g_type;
}
break;
case TextureFormat::B5G6R5:
case TextureFormat::B6G5R5:
case TextureFormat::B10G11R11:
if (component == 0) {
return descriptor.b_type;
}
if (component == 1) {
return descriptor.g_type;
}
if (component == 2) {
return descriptor.r_type;
}
break;
case TextureFormat::R24G8:
case TextureFormat::R8G24:
case TextureFormat::R8G8:
case TextureFormat::G4R4:
if (component == 0) {
return descriptor.g_type;
}
if (component == 1) {
return descriptor.r_type;
}
break;
default:
break;
}
UNIMPLEMENTED_MSG("Texture format not implemented={}", format);
return ComponentType::FLOAT;
}
bool IsComponentEnabled(std::size_t component_mask, std::size_t component) {
constexpr u8 R = 0b0001;
constexpr u8 G = 0b0010;
constexpr u8 B = 0b0100;
constexpr u8 A = 0b1000;
constexpr std::array<u8, 16> mask = {
0, (R), (G), (R | G), (B), (R | B), (G | B), (R | G | B),
(A), (R | A), (G | A), (R | G | A), (B | A), (R | B | A), (G | B | A), (R | G | B | A)};
return std::bitset<4>{mask.at(component_mask)}.test(component);
}
u32 GetComponentSize(TextureFormat format, std::size_t component) {
switch (format) {
case TextureFormat::R32G32B32A32:
return 32;
case TextureFormat::R16G16B16A16:
return 16;
case TextureFormat::R32G32B32:
return component <= 2 ? 32 : 0;
case TextureFormat::R32G32:
return component <= 1 ? 32 : 0;
case TextureFormat::R16G16:
return component <= 1 ? 16 : 0;
case TextureFormat::R32:
return component == 0 ? 32 : 0;
case TextureFormat::R16:
return component == 0 ? 16 : 0;
case TextureFormat::R8:
return component == 0 ? 8 : 0;
case TextureFormat::R1:
return component == 0 ? 1 : 0;
case TextureFormat::A8R8G8B8:
return 8;
case TextureFormat::A2B10G10R10:
return (component == 3 || component == 2 || component == 1) ? 10 : 2;
case TextureFormat::A4B4G4R4:
return 4;
case TextureFormat::A5B5G5R1:
return (component == 0 || component == 1 || component == 2) ? 5 : 1;
case TextureFormat::A1B5G5R5:
return (component == 1 || component == 2 || component == 3) ? 5 : 1;
case TextureFormat::R32_B24G8:
if (component == 0) {
return 32;
}
if (component == 1) {
return 24;
}
if (component == 2) {
return 8;
}
return 0;
case TextureFormat::B5G6R5:
if (component == 0 || component == 2) {
return 5;
}
if (component == 1) {
return 6;
}
return 0;
case TextureFormat::B6G5R5:
if (component == 1 || component == 2) {
return 5;
}
if (component == 0) {
return 6;
}
return 0;
case TextureFormat::B10G11R11:
if (component == 1 || component == 2) {
return 11;
}
if (component == 0) {
return 10;
}
return 0;
case TextureFormat::R24G8:
if (component == 0) {
return 8;
}
if (component == 1) {
return 24;
}
return 0;
case TextureFormat::R8G24:
if (component == 0) {
return 24;
}
if (component == 1) {
return 8;
}
return 0;
case TextureFormat::R8G8:
return (component == 0 || component == 1) ? 8 : 0;
case TextureFormat::G4R4:
return (component == 0 || component == 1) ? 4 : 0;
default:
UNIMPLEMENTED_MSG("Texture format not implemented={}", format);
return 0;
}
}
std::size_t GetImageComponentMask(TextureFormat format) {
constexpr u8 R = 0b0001;
constexpr u8 G = 0b0010;
constexpr u8 B = 0b0100;
constexpr u8 A = 0b1000;
switch (format) {
case TextureFormat::R32G32B32A32:
case TextureFormat::R16G16B16A16:
case TextureFormat::A8R8G8B8:
case TextureFormat::A2B10G10R10:
case TextureFormat::A4B4G4R4:
case TextureFormat::A5B5G5R1:
case TextureFormat::A1B5G5R5:
return std::size_t{R | G | B | A};
case TextureFormat::R32G32B32:
case TextureFormat::R32_B24G8:
case TextureFormat::B5G6R5:
case TextureFormat::B6G5R5:
case TextureFormat::B10G11R11:
return std::size_t{R | G | B};
case TextureFormat::R32G32:
case TextureFormat::R16G16:
case TextureFormat::R24G8:
case TextureFormat::R8G24:
case TextureFormat::R8G8:
case TextureFormat::G4R4:
return std::size_t{R | G};
case TextureFormat::R32:
case TextureFormat::R16:
case TextureFormat::R8:
case TextureFormat::R1:
return std::size_t{R};
default:
UNIMPLEMENTED_MSG("Texture format not implemented={}", format);
return std::size_t{R | G | B | A};
}
}
std::size_t GetImageTypeNumCoordinates(Tegra::Shader::ImageType image_type) {
switch (image_type) {
case Tegra::Shader::ImageType::Texture1D:
case Tegra::Shader::ImageType::TextureBuffer:
return 1;
case Tegra::Shader::ImageType::Texture1DArray:
case Tegra::Shader::ImageType::Texture2D:
return 2;
case Tegra::Shader::ImageType::Texture2DArray:
case Tegra::Shader::ImageType::Texture3D:
return 3;
}
UNREACHABLE();
return 1;
}
} // Anonymous namespace
std::pair<Node, bool> ShaderIR::GetComponentValue(ComponentType component_type, u32 component_size,
Node original_value) {
switch (component_type) {
case ComponentType::SNORM: {
// range [-1.0, 1.0]
auto cnv_value = Operation(OperationCode::FMul, original_value,
Immediate(static_cast<float>(1 << component_size) / 2.f - 1.f));
cnv_value = Operation(OperationCode::ICastFloat, std::move(cnv_value));
return {BitfieldExtract(std::move(cnv_value), 0, component_size), true};
}
case ComponentType::SINT:
case ComponentType::UNORM: {
bool is_signed = component_type == ComponentType::SINT;
// range [0.0, 1.0]
auto cnv_value = Operation(OperationCode::FMul, original_value,
Immediate(static_cast<float>(1 << component_size) - 1.f));
return {SignedOperation(OperationCode::ICastFloat, is_signed, std::move(cnv_value)),
is_signed};
}
case ComponentType::UINT: // range [0, (1 << component_size) - 1]
return {std::move(original_value), false};
case ComponentType::FLOAT:
if (component_size == 16) {
return {Operation(OperationCode::HCastFloat, original_value), true};
} else {
return {std::move(original_value), true};
}
default:
UNIMPLEMENTED_MSG("Unimplemented component type={}", component_type);
return {std::move(original_value), true};
}
}
u32 ShaderIR::DecodeImage(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
const auto opcode = OpCode::Decode(instr);
const auto GetCoordinates = [this, instr](Tegra::Shader::ImageType image_type) {
std::vector<Node> coords;
const std::size_t num_coords{GetImageTypeNumCoordinates(image_type)};
coords.reserve(num_coords);
for (std::size_t i = 0; i < num_coords; ++i) {
coords.push_back(GetRegister(instr.gpr8.Value() + i));
}
return coords;
};
switch (opcode->get().GetId()) {
case OpCode::Id::SULD: {
UNIMPLEMENTED_IF(instr.suldst.out_of_bounds_store !=
Tegra::Shader::OutOfBoundsStore::Ignore);
const auto type{instr.suldst.image_type};
auto& image{instr.suldst.is_immediate ? GetImage(instr.image, type)
: GetBindlessImage(instr.gpr39, type)};
image.MarkRead();
if (instr.suldst.mode == Tegra::Shader::SurfaceDataMode::P) {
u32 indexer = 0;
for (u32 element = 0; element < 4; ++element) {
if (!instr.suldst.IsComponentEnabled(element)) {
continue;
}
MetaImage meta{image, {}, element};
Node value = Operation(OperationCode::ImageLoad, meta, GetCoordinates(type));
SetTemporary(bb, indexer++, std::move(value));
}
for (u32 i = 0; i < indexer; ++i) {
SetRegister(bb, instr.gpr0.Value() + i, GetTemporary(i));
}
} else if (instr.suldst.mode == Tegra::Shader::SurfaceDataMode::D_BA) {
UNIMPLEMENTED_IF(instr.suldst.GetStoreDataLayout() != StoreType::Bits32 &&
instr.suldst.GetStoreDataLayout() != StoreType::Bits64);
auto descriptor = [this, instr] {
std::optional<Tegra::Engines::SamplerDescriptor> sampler_descriptor;
if (instr.suldst.is_immediate) {
sampler_descriptor =
registry.ObtainBoundSampler(static_cast<u32>(instr.image.index.Value()));
} else {
const Node image_register = GetRegister(instr.gpr39);
const auto result = TrackCbuf(image_register, global_code,
static_cast<s64>(global_code.size()));
const auto buffer = std::get<1>(result);
const auto offset = std::get<2>(result);
sampler_descriptor = registry.ObtainBindlessSampler(buffer, offset);
}
if (!sampler_descriptor) {
UNREACHABLE_MSG("Failed to obtain image descriptor");
}
return *sampler_descriptor;
}();
const auto comp_mask = GetImageComponentMask(descriptor.format);
switch (instr.suldst.GetStoreDataLayout()) {
case StoreType::Bits32:
case StoreType::Bits64: {
u32 indexer = 0;
u32 shifted_counter = 0;
Node value = Immediate(0);
for (u32 element = 0; element < 4; ++element) {
if (!IsComponentEnabled(comp_mask, element)) {
continue;
}
const auto component_type = GetComponentType(descriptor, element);
const auto component_size = GetComponentSize(descriptor.format, element);
MetaImage meta{image, {}, element};
auto [converted_value, is_signed] = GetComponentValue(
component_type, component_size,
Operation(OperationCode::ImageLoad, meta, GetCoordinates(type)));
// shift element to correct position
const auto shifted = shifted_counter;
if (shifted > 0) {
converted_value =
SignedOperation(OperationCode::ILogicalShiftLeft, is_signed,
std::move(converted_value), Immediate(shifted));
}
shifted_counter += component_size;
// add value into result
value = Operation(OperationCode::UBitwiseOr, value, std::move(converted_value));
// if we shifted enough for 1 byte -> we save it into temp
if (shifted_counter >= 32) {
SetTemporary(bb, indexer++, std::move(value));
// reset counter and value to prepare pack next byte
value = Immediate(0);
shifted_counter = 0;
}
}
for (u32 i = 0; i < indexer; ++i) {
SetRegister(bb, instr.gpr0.Value() + i, GetTemporary(i));
}
break;
}
default:
UNREACHABLE();
break;
}
}
break;
}
case OpCode::Id::SUST: {
UNIMPLEMENTED_IF(instr.suldst.mode != Tegra::Shader::SurfaceDataMode::P);
UNIMPLEMENTED_IF(instr.suldst.out_of_bounds_store !=
Tegra::Shader::OutOfBoundsStore::Ignore);
UNIMPLEMENTED_IF(instr.suldst.component_mask_selector != 0xf); // Ensure we have RGBA
std::vector<Node> values;
constexpr std::size_t hardcoded_size{4};
for (std::size_t i = 0; i < hardcoded_size; ++i) {
values.push_back(GetRegister(instr.gpr0.Value() + i));
}
const auto type{instr.suldst.image_type};
auto& image{instr.suldst.is_immediate ? GetImage(instr.image, type)
: GetBindlessImage(instr.gpr39, type)};
image.MarkWrite();
MetaImage meta{image, std::move(values)};
bb.push_back(Operation(OperationCode::ImageStore, meta, GetCoordinates(type)));
break;
}
case OpCode::Id::SUATOM: {
UNIMPLEMENTED_IF(instr.suatom_d.is_ba != 0);
const OperationCode operation_code = [instr] {
switch (instr.suatom_d.operation_type) {
case Tegra::Shader::ImageAtomicOperationType::S32:
case Tegra::Shader::ImageAtomicOperationType::U32:
switch (instr.suatom_d.operation) {
case Tegra::Shader::ImageAtomicOperation::Add:
return OperationCode::AtomicImageAdd;
case Tegra::Shader::ImageAtomicOperation::And:
return OperationCode::AtomicImageAnd;
case Tegra::Shader::ImageAtomicOperation::Or:
return OperationCode::AtomicImageOr;
case Tegra::Shader::ImageAtomicOperation::Xor:
return OperationCode::AtomicImageXor;
case Tegra::Shader::ImageAtomicOperation::Exch:
return OperationCode::AtomicImageExchange;
default:
break;
}
break;
default:
break;
}
UNIMPLEMENTED_MSG("Unimplemented operation={}, type={}",
static_cast<u64>(instr.suatom_d.operation.Value()),
static_cast<u64>(instr.suatom_d.operation_type.Value()));
return OperationCode::AtomicImageAdd;
}();
Node value = GetRegister(instr.gpr0);
const auto type = instr.suatom_d.image_type;
auto& image = GetImage(instr.image, type);
image.MarkAtomic();
MetaImage meta{image, {std::move(value)}};
SetRegister(bb, instr.gpr0, Operation(operation_code, meta, GetCoordinates(type)));
break;
}
default:
UNIMPLEMENTED_MSG("Unhandled image instruction: {}", opcode->get().GetName());
}
return pc;
}
ImageEntry& ShaderIR::GetImage(Tegra::Shader::Image image, Tegra::Shader::ImageType type) {
const auto offset = static_cast<u32>(image.index.Value());
const auto it =
std::find_if(std::begin(used_images), std::end(used_images),
[offset](const ImageEntry& entry) { return entry.offset == offset; });
if (it != std::end(used_images)) {
ASSERT(!it->is_bindless && it->type == type);
return *it;
}
const auto next_index = static_cast<u32>(used_images.size());
return used_images.emplace_back(next_index, offset, type);
}
ImageEntry& ShaderIR::GetBindlessImage(Tegra::Shader::Register reg, Tegra::Shader::ImageType type) {
const Node image_register = GetRegister(reg);
const auto result =
TrackCbuf(image_register, global_code, static_cast<s64>(global_code.size()));
const auto buffer = std::get<1>(result);
const auto offset = std::get<2>(result);
const auto it = std::find_if(std::begin(used_images), std::end(used_images),
[buffer, offset](const ImageEntry& entry) {
return entry.buffer == buffer && entry.offset == offset;
});
if (it != std::end(used_images)) {
ASSERT(it->is_bindless && it->type == type);
return *it;
}
const auto next_index = static_cast<u32>(used_images.size());
return used_images.emplace_back(next_index, offset, buffer, type);
}
} // namespace VideoCommon::Shader

49
src/video_core/shader/decode/integer_set.cpp
View file

@ -1,49 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/common_types.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
u32 ShaderIR::DecodeIntegerSet(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
const Node op_a = GetRegister(instr.gpr8);
const Node op_b = [&]() {
if (instr.is_b_imm) {
return Immediate(instr.alu.GetSignedImm20_20());
} else if (instr.is_b_gpr) {
return GetRegister(instr.gpr20);
} else {
return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset());
}
}();
// The iset instruction sets a register to 1.0 or -1 (depending on the bf bit) if the condition
// is true, and to 0 otherwise.
const Node second_pred = GetPredicate(instr.iset.pred39, instr.iset.neg_pred != 0);
const Node first_pred =
GetPredicateComparisonInteger(instr.iset.cond, instr.iset.is_signed, op_a, op_b);
const OperationCode combiner = GetPredicateCombiner(instr.iset.op);
const Node predicate = Operation(combiner, first_pred, second_pred);
const Node true_value = instr.iset.bf ? Immediate(1.0f) : Immediate(-1);
const Node false_value = instr.iset.bf ? Immediate(0.0f) : Immediate(0);
const Node value =
Operation(OperationCode::Select, PRECISE, predicate, true_value, false_value);
SetRegister(bb, instr.gpr0, value);
return pc;
}
} // namespace VideoCommon::Shader

53
src/video_core/shader/decode/integer_set_predicate.cpp
View file

@ -1,53 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/assert.h"
#include "common/common_types.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
using Tegra::Shader::Pred;
u32 ShaderIR::DecodeIntegerSetPredicate(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
const Node op_a = GetRegister(instr.gpr8);
const Node op_b = [&]() {
if (instr.is_b_imm) {
return Immediate(instr.alu.GetSignedImm20_20());
} else if (instr.is_b_gpr) {
return GetRegister(instr.gpr20);
} else {
return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset());
}
}();
// We can't use the constant predicate as destination.
ASSERT(instr.isetp.pred3 != static_cast<u64>(Pred::UnusedIndex));
const Node second_pred = GetPredicate(instr.isetp.pred39, instr.isetp.neg_pred != 0);
const Node predicate =
GetPredicateComparisonInteger(instr.isetp.cond, instr.isetp.is_signed, op_a, op_b);
// Set the primary predicate to the result of Predicate OP SecondPredicate
const OperationCode combiner = GetPredicateCombiner(instr.isetp.op);
const Node value = Operation(combiner, predicate, second_pred);
SetPredicate(bb, instr.isetp.pred3, value);
if (instr.isetp.pred0 != static_cast<u64>(Pred::UnusedIndex)) {
// Set the secondary predicate to the result of !Predicate OP SecondPredicate, if enabled
const Node negated_pred = Operation(OperationCode::LogicalNegate, predicate);
SetPredicate(bb, instr.isetp.pred0, Operation(combiner, negated_pred, second_pred));
}
return pc;
}
} // namespace VideoCommon::Shader

493
src/video_core/shader/decode/memory.cpp
View file

@ -1,493 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <algorithm>
#include <utility>
#include <vector>
#include <fmt/format.h>
#include "common/alignment.h"
#include "common/assert.h"
#include "common/common_types.h"
#include "common/logging/log.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
using std::move;
using Tegra::Shader::AtomicOp;
using Tegra::Shader::AtomicType;
using Tegra::Shader::Attribute;
using Tegra::Shader::GlobalAtomicType;
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
using Tegra::Shader::Register;
using Tegra::Shader::StoreType;
namespace {
OperationCode GetAtomOperation(AtomicOp op) {
switch (op) {
case AtomicOp::Add:
return OperationCode::AtomicIAdd;
case AtomicOp::Min:
return OperationCode::AtomicIMin;
case AtomicOp::Max:
return OperationCode::AtomicIMax;
case AtomicOp::And:
return OperationCode::AtomicIAnd;
case AtomicOp::Or:
return OperationCode::AtomicIOr;
case AtomicOp::Xor:
return OperationCode::AtomicIXor;
case AtomicOp::Exch:
return OperationCode::AtomicIExchange;
default:
UNIMPLEMENTED_MSG("op={}", op);
return OperationCode::AtomicIAdd;
}
}
bool IsUnaligned(Tegra::Shader::UniformType uniform_type) {
return uniform_type == Tegra::Shader::UniformType::UnsignedByte ||
uniform_type == Tegra::Shader::UniformType::UnsignedShort;
}
u32 GetUnalignedMask(Tegra::Shader::UniformType uniform_type) {
switch (uniform_type) {
case Tegra::Shader::UniformType::UnsignedByte:
return 0b11;
case Tegra::Shader::UniformType::UnsignedShort:
return 0b10;
default:
UNREACHABLE();
return 0;
}
}
u32 GetMemorySize(Tegra::Shader::UniformType uniform_type) {
switch (uniform_type) {
case Tegra::Shader::UniformType::UnsignedByte:
return 8;
case Tegra::Shader::UniformType::UnsignedShort:
return 16;
case Tegra::Shader::UniformType::Single:
return 32;
case Tegra::Shader::UniformType::Double:
return 64;
case Tegra::Shader::UniformType::Quad:
case Tegra::Shader::UniformType::UnsignedQuad:
return 128;
default:
UNIMPLEMENTED_MSG("Unimplemented size={}!", uniform_type);
return 32;
}
}
Node ExtractUnaligned(Node value, Node address, u32 mask, u32 size) {
Node offset = Operation(OperationCode::UBitwiseAnd, address, Immediate(mask));
offset = Operation(OperationCode::ULogicalShiftLeft, move(offset), Immediate(3));
return Operation(OperationCode::UBitfieldExtract, move(value), move(offset), Immediate(size));
}
Node InsertUnaligned(Node dest, Node value, Node address, u32 mask, u32 size) {
Node offset = Operation(OperationCode::UBitwiseAnd, move(address), Immediate(mask));
offset = Operation(OperationCode::ULogicalShiftLeft, move(offset), Immediate(3));
return Operation(OperationCode::UBitfieldInsert, move(dest), move(value), move(offset),
Immediate(size));
}
Node Sign16Extend(Node value) {
Node sign = Operation(OperationCode::UBitwiseAnd, value, Immediate(1U << 15));
Node is_sign = Operation(OperationCode::LogicalUEqual, move(sign), Immediate(1U << 15));
Node extend = Operation(OperationCode::Select, is_sign, Immediate(0xFFFF0000), Immediate(0));
return Operation(OperationCode::UBitwiseOr, move(value), move(extend));
}
} // Anonymous namespace
u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
const auto opcode = OpCode::Decode(instr);
switch (opcode->get().GetId()) {
case OpCode::Id::LD_A: {
// Note: Shouldn't this be interp mode flat? As in no interpolation made.
UNIMPLEMENTED_IF_MSG(instr.gpr8.Value() != Register::ZeroIndex,
"Indirect attribute loads are not supported");
UNIMPLEMENTED_IF_MSG((instr.attribute.fmt20.immediate.Value() % sizeof(u32)) != 0,
"Unaligned attribute loads are not supported");
UNIMPLEMENTED_IF_MSG(instr.attribute.fmt20.IsPhysical() &&
instr.attribute.fmt20.size != Tegra::Shader::AttributeSize::Word,
"Non-32 bits PHYS reads are not implemented");
const Node buffer{GetRegister(instr.gpr39)};
u64 next_element = instr.attribute.fmt20.element;
auto next_index = static_cast<u64>(instr.attribute.fmt20.index.Value());
const auto LoadNextElement = [&](u32 reg_offset) {
const Node attribute{instr.attribute.fmt20.IsPhysical()
? GetPhysicalInputAttribute(instr.gpr8, buffer)
: GetInputAttribute(static_cast<Attribute::Index>(next_index),
next_element, buffer)};
SetRegister(bb, instr.gpr0.Value() + reg_offset, attribute);
// Load the next attribute element into the following register. If the element
// to load goes beyond the vec4 size, load the first element of the next
// attribute.
next_element = (next_element + 1) % 4;
next_index = next_index + (next_element == 0 ? 1 : 0);
};
const u32 num_words = static_cast<u32>(instr.attribute.fmt20.size.Value()) + 1;
for (u32 reg_offset = 0; reg_offset < num_words; ++reg_offset) {
LoadNextElement(reg_offset);
}
break;
}
case OpCode::Id::LD_C: {
UNIMPLEMENTED_IF(instr.ld_c.unknown != 0);
Node index = GetRegister(instr.gpr8);
const Node op_a =
GetConstBufferIndirect(instr.cbuf36.index, instr.cbuf36.GetOffset() + 0, index);
switch (instr.ld_c.type.Value()) {
case Tegra::Shader::UniformType::Single:
SetRegister(bb, instr.gpr0, op_a);
break;
case Tegra::Shader::UniformType::Double: {
const Node op_b =
GetConstBufferIndirect(instr.cbuf36.index, instr.cbuf36.GetOffset() + 4, index);
SetTemporary(bb, 0, op_a);
SetTemporary(bb, 1, op_b);
SetRegister(bb, instr.gpr0, GetTemporary(0));
SetRegister(bb, instr.gpr0.Value() + 1, GetTemporary(1));
break;
}
default:
UNIMPLEMENTED_MSG("Unhandled type: {}", instr.ld_c.type.Value());
}
break;
}
case OpCode::Id::LD_L:
LOG_DEBUG(HW_GPU, "LD_L cache management mode: {}", instr.ld_l.unknown);
[[fallthrough]];
case OpCode::Id::LD_S: {
const auto GetAddress = [&](s32 offset) {
ASSERT(offset % 4 == 0);
const Node immediate_offset = Immediate(static_cast<s32>(instr.smem_imm) + offset);
return Operation(OperationCode::IAdd, GetRegister(instr.gpr8), immediate_offset);
};
const auto GetMemory = [&](s32 offset) {
return opcode->get().GetId() == OpCode::Id::LD_S ? GetSharedMemory(GetAddress(offset))
: GetLocalMemory(GetAddress(offset));
};
switch (instr.ldst_sl.type.Value()) {
case StoreType::Signed16:
SetRegister(bb, instr.gpr0,
Sign16Extend(ExtractUnaligned(GetMemory(0), GetAddress(0), 0b10, 16)));
break;
case StoreType::Bits32:
case StoreType::Bits64:
case StoreType::Bits128: {
const u32 count = [&] {
switch (instr.ldst_sl.type.Value()) {
case StoreType::Bits32:
return 1;
case StoreType::Bits64:
return 2;
case StoreType::Bits128:
return 4;
default:
UNREACHABLE();
return 0;
}
}();
for (u32 i = 0; i < count; ++i) {
SetTemporary(bb, i, GetMemory(i * 4));
}
for (u32 i = 0; i < count; ++i) {
SetRegister(bb, instr.gpr0.Value() + i, GetTemporary(i));
}
break;
}
default:
UNIMPLEMENTED_MSG("{} Unhandled type: {}", opcode->get().GetName(),
instr.ldst_sl.type.Value());
}
break;
}
case OpCode::Id::LD:
case OpCode::Id::LDG: {
const auto type = [instr, &opcode]() -> Tegra::Shader::UniformType {
switch (opcode->get().GetId()) {
case OpCode::Id::LD:
UNIMPLEMENTED_IF_MSG(!instr.generic.extended, "Unextended LD is not implemented");
return instr.generic.type;
case OpCode::Id::LDG:
return instr.ldg.type;
default:
UNREACHABLE();
return {};
}
}();
const auto [real_address_base, base_address, descriptor] =
TrackGlobalMemory(bb, instr, true, false);
const u32 size = GetMemorySize(type);
const u32 count = Common::AlignUp(size, 32) / 32;
if (!real_address_base || !base_address) {
// Tracking failed, load zeroes.
for (u32 i = 0; i < count; ++i) {
SetRegister(bb, instr.gpr0.Value() + i, Immediate(0.0f));
}
break;
}
for (u32 i = 0; i < count; ++i) {
const Node it_offset = Immediate(i * 4);
const Node real_address = Operation(OperationCode::UAdd, real_address_base, it_offset);
Node gmem = MakeNode<GmemNode>(real_address, base_address, descriptor);
// To handle unaligned loads get the bytes used to dereference global memory and extract
// those bytes from the loaded u32.
if (IsUnaligned(type)) {
gmem = ExtractUnaligned(gmem, real_address, GetUnalignedMask(type), size);
}
SetTemporary(bb, i, gmem);
}
for (u32 i = 0; i < count; ++i) {
SetRegister(bb, instr.gpr0.Value() + i, GetTemporary(i));
}
break;
}
case OpCode::Id::ST_A: {
UNIMPLEMENTED_IF_MSG(instr.gpr8.Value() != Register::ZeroIndex,
"Indirect attribute loads are not supported");
UNIMPLEMENTED_IF_MSG((instr.attribute.fmt20.immediate.Value() % sizeof(u32)) != 0,
"Unaligned attribute loads are not supported");
u64 element = instr.attribute.fmt20.element;
auto index = static_cast<u64>(instr.attribute.fmt20.index.Value());
const u32 num_words = static_cast<u32>(instr.attribute.fmt20.size.Value()) + 1;
for (u32 reg_offset = 0; reg_offset < num_words; ++reg_offset) {
Node dest;
if (instr.attribute.fmt20.patch) {
const u32 offset = static_cast<u32>(index) * 4 + static_cast<u32>(element);
dest = MakeNode<PatchNode>(offset);
} else {
dest = GetOutputAttribute(static_cast<Attribute::Index>(index), element,
GetRegister(instr.gpr39));
}
const auto src = GetRegister(instr.gpr0.Value() + reg_offset);
bb.push_back(Operation(OperationCode::Assign, dest, src));
// Load the next attribute element into the following register. If the element to load
// goes beyond the vec4 size, load the first element of the next attribute.
element = (element + 1) % 4;
index = index + (element == 0 ? 1 : 0);
}
break;
}
case OpCode::Id::ST_L:
LOG_DEBUG(HW_GPU, "ST_L cache management mode: {}", instr.st_l.cache_management.Value());
[[fallthrough]];
case OpCode::Id::ST_S: {
const auto GetAddress = [&](s32 offset) {
ASSERT(offset % 4 == 0);
const Node immediate = Immediate(static_cast<s32>(instr.smem_imm) + offset);
return Operation(OperationCode::IAdd, NO_PRECISE, GetRegister(instr.gpr8), immediate);
};
const bool is_local = opcode->get().GetId() == OpCode::Id::ST_L;
const auto set_memory = is_local ? &ShaderIR::SetLocalMemory : &ShaderIR::SetSharedMemory;
const auto get_memory = is_local ? &ShaderIR::GetLocalMemory : &ShaderIR::GetSharedMemory;
switch (instr.ldst_sl.type.Value()) {
case StoreType::Bits128:
(this->*set_memory)(bb, GetAddress(12), GetRegister(instr.gpr0.Value() + 3));
(this->*set_memory)(bb, GetAddress(8), GetRegister(instr.gpr0.Value() + 2));
[[fallthrough]];
case StoreType::Bits64:
(this->*set_memory)(bb, GetAddress(4), GetRegister(instr.gpr0.Value() + 1));
[[fallthrough]];
case StoreType::Bits32:
(this->*set_memory)(bb, GetAddress(0), GetRegister(instr.gpr0));
break;
case StoreType::Unsigned16:
case StoreType::Signed16: {
Node address = GetAddress(0);
Node memory = (this->*get_memory)(address);
(this->*set_memory)(
bb, address, InsertUnaligned(memory, GetRegister(instr.gpr0), address, 0b10, 16));
break;
}
default:
UNIMPLEMENTED_MSG("{} unhandled type: {}", opcode->get().GetName(),
instr.ldst_sl.type.Value());
}
break;
}
case OpCode::Id::ST:
case OpCode::Id::STG: {
const auto type = [instr, &opcode]() -> Tegra::Shader::UniformType {
switch (opcode->get().GetId()) {
case OpCode::Id::ST:
UNIMPLEMENTED_IF_MSG(!instr.generic.extended, "Unextended ST is not implemented");
return instr.generic.type;
case OpCode::Id::STG:
return instr.stg.type;
default:
UNREACHABLE();
return {};
}
}();
// For unaligned reads we have to read memory too.
const bool is_read = IsUnaligned(type);
const auto [real_address_base, base_address, descriptor] =
TrackGlobalMemory(bb, instr, is_read, true);
if (!real_address_base || !base_address) {
// Tracking failed, skip the store.
break;
}
const u32 size = GetMemorySize(type);
const u32 count = Common::AlignUp(size, 32) / 32;
for (u32 i = 0; i < count; ++i) {
const Node it_offset = Immediate(i * 4);
const Node real_address = Operation(OperationCode::UAdd, real_address_base, it_offset);
const Node gmem = MakeNode<GmemNode>(real_address, base_address, descriptor);
Node value = GetRegister(instr.gpr0.Value() + i);
if (IsUnaligned(type)) {
const u32 mask = GetUnalignedMask(type);
value = InsertUnaligned(gmem, move(value), real_address, mask, size);
}
bb.push_back(Operation(OperationCode::Assign, gmem, value));
}
break;
}
case OpCode::Id::RED: {
UNIMPLEMENTED_IF_MSG(instr.red.type != GlobalAtomicType::U32, "type={}",
instr.red.type.Value());
const auto [real_address, base_address, descriptor] =
TrackGlobalMemory(bb, instr, true, true);
if (!real_address || !base_address) {
// Tracking failed, skip atomic.
break;
}
Node gmem = MakeNode<GmemNode>(real_address, base_address, descriptor);
Node value = GetRegister(instr.gpr0);
bb.push_back(Operation(GetAtomOperation(instr.red.operation), move(gmem), move(value)));
break;
}
case OpCode::Id::ATOM: {
UNIMPLEMENTED_IF_MSG(instr.atom.operation == AtomicOp::Inc ||
instr.atom.operation == AtomicOp::Dec ||
instr.atom.operation == AtomicOp::SafeAdd,
"operation={}", instr.atom.operation.Value());
UNIMPLEMENTED_IF_MSG(instr.atom.type == GlobalAtomicType::S64 ||
instr.atom.type == GlobalAtomicType::U64 ||
instr.atom.type == GlobalAtomicType::F16x2_FTZ_RN ||
instr.atom.type == GlobalAtomicType::F32_FTZ_RN,
"type={}", instr.atom.type.Value());
const auto [real_address, base_address, descriptor] =
TrackGlobalMemory(bb, instr, true, true);
if (!real_address || !base_address) {
// Tracking failed, skip atomic.
break;
}
const bool is_signed =
instr.atom.type == GlobalAtomicType::S32 || instr.atom.type == GlobalAtomicType::S64;
Node gmem = MakeNode<GmemNode>(real_address, base_address, descriptor);
SetRegister(bb, instr.gpr0,
SignedOperation(GetAtomOperation(instr.atom.operation), is_signed, gmem,
GetRegister(instr.gpr20)));
break;
}
case OpCode::Id::ATOMS: {
UNIMPLEMENTED_IF_MSG(instr.atoms.operation == AtomicOp::Inc ||
instr.atoms.operation == AtomicOp::Dec,
"operation={}", instr.atoms.operation.Value());
UNIMPLEMENTED_IF_MSG(instr.atoms.type == AtomicType::S64 ||
instr.atoms.type == AtomicType::U64,
"type={}", instr.atoms.type.Value());
const bool is_signed =
instr.atoms.type == AtomicType::S32 || instr.atoms.type == AtomicType::S64;
const s32 offset = instr.atoms.GetImmediateOffset();
Node address = GetRegister(instr.gpr8);
address = Operation(OperationCode::IAdd, move(address), Immediate(offset));
SetRegister(bb, instr.gpr0,
SignedOperation(GetAtomOperation(instr.atoms.operation), is_signed,
GetSharedMemory(move(address)), GetRegister(instr.gpr20)));
break;
}
case OpCode::Id::AL2P: {
// Ignore al2p.direction since we don't care about it.
// Calculate emulation fake physical address.
const Node fixed_address{Immediate(static_cast<u32>(instr.al2p.address))};
const Node reg{GetRegister(instr.gpr8)};
const Node fake_address{Operation(OperationCode::IAdd, NO_PRECISE, reg, fixed_address)};
// Set the fake address to target register.
SetRegister(bb, instr.gpr0, fake_address);
// Signal the shader IR to declare all possible attributes and varyings
uses_physical_attributes = true;
break;
}
default:
UNIMPLEMENTED_MSG("Unhandled memory instruction: {}", opcode->get().GetName());
}
return pc;
}
std::tuple<Node, Node, GlobalMemoryBase> ShaderIR::TrackGlobalMemory(NodeBlock& bb,
Instruction instr,
bool is_read, bool is_write) {
const auto addr_register{GetRegister(instr.gmem.gpr)};
const auto immediate_offset{static_cast<u32>(instr.gmem.offset)};
const auto [base_address, index, offset] =
TrackCbuf(addr_register, global_code, static_cast<s64>(global_code.size()));
ASSERT_OR_EXECUTE_MSG(
base_address != nullptr, { return std::make_tuple(nullptr, nullptr, GlobalMemoryBase{}); },
"Global memory tracking failed");
bb.push_back(Comment(fmt::format("Base address is c[0x{:x}][0x{:x}]", index, offset)));
const GlobalMemoryBase descriptor{index, offset};
const auto& entry = used_global_memory.try_emplace(descriptor).first;
auto& usage = entry->second;
usage.is_written |= is_write;
usage.is_read |= is_read;
const auto real_address =
Operation(OperationCode::UAdd, NO_PRECISE, Immediate(immediate_offset), addr_register);
return {real_address, base_address, descriptor};
}
} // namespace VideoCommon::Shader

322
src/video_core/shader/decode/other.cpp
View file

@ -1,322 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/assert.h"
#include "common/common_types.h"
#include "common/logging/log.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
using std::move;
using Tegra::Shader::ConditionCode;
using Tegra::Shader::Instruction;
using Tegra::Shader::IpaInterpMode;
using Tegra::Shader::OpCode;
using Tegra::Shader::PixelImap;
using Tegra::Shader::Register;
using Tegra::Shader::SystemVariable;
using Index = Tegra::Shader::Attribute::Index;
u32 ShaderIR::DecodeOther(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
const auto opcode = OpCode::Decode(instr);
switch (opcode->get().GetId()) {
case OpCode::Id::NOP: {
UNIMPLEMENTED_IF(instr.nop.cc != Tegra::Shader::ConditionCode::T);
UNIMPLEMENTED_IF(instr.nop.trigger != 0);
// With the previous preconditions, this instruction is a no-operation.
break;
}
case OpCode::Id::EXIT: {
const ConditionCode cc = instr.flow_condition_code;
UNIMPLEMENTED_IF_MSG(cc != ConditionCode::T, "EXIT condition code used: {}", cc);
switch (instr.flow.cond) {
case Tegra::Shader::FlowCondition::Always:
bb.push_back(Operation(OperationCode::Exit));
if (instr.pred.pred_index == static_cast<u64>(Pred::UnusedIndex)) {
// If this is an unconditional exit then just end processing here,
// otherwise we have to account for the possibility of the condition
// not being met, so continue processing the next instruction.
pc = MAX_PROGRAM_LENGTH - 1;
}
break;
case Tegra::Shader::FlowCondition::Fcsm_Tr:
// TODO(bunnei): What is this used for? If we assume this conditon is not
// satisifed, dual vertex shaders in Farming Simulator make more sense
UNIMPLEMENTED_MSG("Skipping unknown FlowCondition::Fcsm_Tr");
break;
default:
UNIMPLEMENTED_MSG("Unhandled flow condition: {}", instr.flow.cond.Value());
}
break;
}
case OpCode::Id::KIL: {
UNIMPLEMENTED_IF(instr.flow.cond != Tegra::Shader::FlowCondition::Always);
const ConditionCode cc = instr.flow_condition_code;
UNIMPLEMENTED_IF_MSG(cc != ConditionCode::T, "KIL condition code used: {}", cc);
bb.push_back(Operation(OperationCode::Discard));
break;
}
case OpCode::Id::S2R: {
const Node value = [this, instr] {
switch (instr.sys20) {
case SystemVariable::LaneId:
return Operation(OperationCode::ThreadId);
case SystemVariable::InvocationId:
return Operation(OperationCode::InvocationId);
case SystemVariable::Ydirection:
uses_y_negate = true;
return Operation(OperationCode::YNegate);
case SystemVariable::InvocationInfo:
LOG_WARNING(HW_GPU, "S2R instruction with InvocationInfo is incomplete");
return Immediate(0x00ff'0000U);
case SystemVariable::WscaleFactorXY:
UNIMPLEMENTED_MSG("S2R WscaleFactorXY is not implemented");
return Immediate(0U);
case SystemVariable::WscaleFactorZ:
UNIMPLEMENTED_MSG("S2R WscaleFactorZ is not implemented");
return Immediate(0U);
case SystemVariable::Tid: {
Node val = Immediate(0);
val = BitfieldInsert(val, Operation(OperationCode::LocalInvocationIdX), 0, 9);
val = BitfieldInsert(val, Operation(OperationCode::LocalInvocationIdY), 16, 9);
val = BitfieldInsert(val, Operation(OperationCode::LocalInvocationIdZ), 26, 5);
return val;
}
case SystemVariable::TidX:
return Operation(OperationCode::LocalInvocationIdX);
case SystemVariable::TidY:
return Operation(OperationCode::LocalInvocationIdY);
case SystemVariable::TidZ:
return Operation(OperationCode::LocalInvocationIdZ);
case SystemVariable::CtaIdX:
return Operation(OperationCode::WorkGroupIdX);
case SystemVariable::CtaIdY:
return Operation(OperationCode::WorkGroupIdY);
case SystemVariable::CtaIdZ:
return Operation(OperationCode::WorkGroupIdZ);
case SystemVariable::EqMask:
case SystemVariable::LtMask:
case SystemVariable::LeMask:
case SystemVariable::GtMask:
case SystemVariable::GeMask:
uses_warps = true;
switch (instr.sys20) {
case SystemVariable::EqMask:
return Operation(OperationCode::ThreadEqMask);
case SystemVariable::LtMask:
return Operation(OperationCode::ThreadLtMask);
case SystemVariable::LeMask:
return Operation(OperationCode::ThreadLeMask);
case SystemVariable::GtMask:
return Operation(OperationCode::ThreadGtMask);
case SystemVariable::GeMask:
return Operation(OperationCode::ThreadGeMask);
default:
UNREACHABLE();
return Immediate(0u);
}
default:
UNIMPLEMENTED_MSG("Unhandled system move: {}", instr.sys20.Value());
return Immediate(0u);
}
}();
SetRegister(bb, instr.gpr0, value);
break;
}
case OpCode::Id::BRA: {
Node branch;
if (instr.bra.constant_buffer == 0) {
const u32 target = pc + instr.bra.GetBranchTarget();
branch = Operation(OperationCode::Branch, Immediate(target));
} else {
const u32 target = pc + 1;
const Node op_a = GetConstBuffer(instr.cbuf36.index, instr.cbuf36.GetOffset());
const Node convert = SignedOperation(OperationCode::IArithmeticShiftRight, true,
PRECISE, op_a, Immediate(3));
const Node operand =
Operation(OperationCode::IAdd, PRECISE, convert, Immediate(target));
branch = Operation(OperationCode::BranchIndirect, operand);
}
const Tegra::Shader::ConditionCode cc = instr.flow_condition_code;
if (cc != Tegra::Shader::ConditionCode::T) {
bb.push_back(Conditional(GetConditionCode(cc), {branch}));
} else {
bb.push_back(branch);
}
break;
}
case OpCode::Id::BRX: {
Node operand;
if (instr.brx.constant_buffer != 0) {
const s32 target = pc + 1;
const Node index = GetRegister(instr.gpr8);
const Node op_a =
GetConstBufferIndirect(instr.cbuf36.index, instr.cbuf36.GetOffset() + 0, index);
const Node convert = SignedOperation(OperationCode::IArithmeticShiftRight, true,
PRECISE, op_a, Immediate(3));
operand = Operation(OperationCode::IAdd, PRECISE, convert, Immediate(target));
} else {
const s32 target = pc + instr.brx.GetBranchExtend();
const Node op_a = GetRegister(instr.gpr8);
const Node convert = SignedOperation(OperationCode::IArithmeticShiftRight, true,
PRECISE, op_a, Immediate(3));
operand = Operation(OperationCode::IAdd, PRECISE, convert, Immediate(target));
}
const Node branch = Operation(OperationCode::BranchIndirect, operand);
const ConditionCode cc = instr.flow_condition_code;
if (cc != ConditionCode::T) {
bb.push_back(Conditional(GetConditionCode(cc), {branch}));
} else {
bb.push_back(branch);
}
break;
}
case OpCode::Id::SSY: {
UNIMPLEMENTED_IF_MSG(instr.bra.constant_buffer != 0,
"Constant buffer flow is not supported");
if (disable_flow_stack) {
break;
}
// The SSY opcode tells the GPU where to re-converge divergent execution paths with SYNC.
const u32 target = pc + instr.bra.GetBranchTarget();
bb.push_back(
Operation(OperationCode::PushFlowStack, MetaStackClass::Ssy, Immediate(target)));
break;
}
case OpCode::Id::PBK: {
UNIMPLEMENTED_IF_MSG(instr.bra.constant_buffer != 0,
"Constant buffer PBK is not supported");
if (disable_flow_stack) {
break;
}
// PBK pushes to a stack the address where BRK will jump to.
const u32 target = pc + instr.bra.GetBranchTarget();
bb.push_back(
Operation(OperationCode::PushFlowStack, MetaStackClass::Pbk, Immediate(target)));
break;
}
case OpCode::Id::SYNC: {
const ConditionCode cc = instr.flow_condition_code;
UNIMPLEMENTED_IF_MSG(cc != ConditionCode::T, "SYNC condition code used: {}", cc);
if (decompiled) {
break;
}
// The SYNC opcode jumps to the address previously set by the SSY opcode
bb.push_back(Operation(OperationCode::PopFlowStack, MetaStackClass::Ssy));
break;
}
case OpCode::Id::BRK: {
const ConditionCode cc = instr.flow_condition_code;
UNIMPLEMENTED_IF_MSG(cc != ConditionCode::T, "BRK condition code used: {}", cc);
if (decompiled) {
break;
}
// The BRK opcode jumps to the address previously set by the PBK opcode
bb.push_back(Operation(OperationCode::PopFlowStack, MetaStackClass::Pbk));
break;
}
case OpCode::Id::IPA: {
const bool is_physical = instr.ipa.idx && instr.gpr8.Value() != 0xff;
const auto attribute = instr.attribute.fmt28;
const Index index = attribute.index;
Node value = is_physical ? GetPhysicalInputAttribute(instr.gpr8)
: GetInputAttribute(index, attribute.element);
// Code taken from Ryujinx.
if (index >= Index::Attribute_0 && index <= Index::Attribute_31) {
const u32 location = static_cast<u32>(index) - static_cast<u32>(Index::Attribute_0);
if (header.ps.GetPixelImap(location) == PixelImap::Perspective) {
Node position_w = GetInputAttribute(Index::Position, 3);
value = Operation(OperationCode::FMul, move(value), move(position_w));
}
}
if (instr.ipa.interp_mode == IpaInterpMode::Multiply) {
value = Operation(OperationCode::FMul, move(value), GetRegister(instr.gpr20));
}
value = GetSaturatedFloat(move(value), instr.ipa.saturate);
SetRegister(bb, instr.gpr0, move(value));
break;
}
case OpCode::Id::OUT_R: {
UNIMPLEMENTED_IF_MSG(instr.gpr20.Value() != Register::ZeroIndex,
"Stream buffer is not supported");
if (instr.out.emit) {
// gpr0 is used to store the next address and gpr8 contains the address to emit.
// Hardware uses pointers here but we just ignore it
bb.push_back(Operation(OperationCode::EmitVertex));
SetRegister(bb, instr.gpr0, Immediate(0));
}
if (instr.out.cut) {
bb.push_back(Operation(OperationCode::EndPrimitive));
}
break;
}
case OpCode::Id::ISBERD: {
UNIMPLEMENTED_IF(instr.isberd.o != 0);
UNIMPLEMENTED_IF(instr.isberd.skew != 0);
UNIMPLEMENTED_IF(instr.isberd.shift != Tegra::Shader::IsberdShift::None);
UNIMPLEMENTED_IF(instr.isberd.mode != Tegra::Shader::IsberdMode::None);
LOG_WARNING(HW_GPU, "ISBERD instruction is incomplete");
SetRegister(bb, instr.gpr0, GetRegister(instr.gpr8));
break;
}
case OpCode::Id::BAR: {
UNIMPLEMENTED_IF_MSG(instr.value != 0xF0A81B8000070000ULL, "BAR is not BAR.SYNC 0x0");
bb.push_back(Operation(OperationCode::Barrier));
break;
}
case OpCode::Id::MEMBAR: {
UNIMPLEMENTED_IF(instr.membar.unknown != Tegra::Shader::MembarUnknown::Default);
const OperationCode type = [instr] {
switch (instr.membar.type) {
case Tegra::Shader::MembarType::CTA:
return OperationCode::MemoryBarrierGroup;
case Tegra::Shader::MembarType::GL:
return OperationCode::MemoryBarrierGlobal;
default:
UNIMPLEMENTED_MSG("MEMBAR type={}", instr.membar.type.Value());
return OperationCode::MemoryBarrierGlobal;
}
}();
bb.push_back(Operation(type));
break;
}
case OpCode::Id::DEPBAR: {
LOG_DEBUG(HW_GPU, "DEPBAR instruction is stubbed");
break;
}
default:
UNIMPLEMENTED_MSG("Unhandled instruction: {}", opcode->get().GetName());
}
return pc;
}
} // namespace VideoCommon::Shader

68
src/video_core/shader/decode/predicate_set_predicate.cpp
View file

@ -1,68 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/assert.h"
#include "common/common_types.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
using Tegra::Shader::Pred;
u32 ShaderIR::DecodePredicateSetPredicate(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
const auto opcode = OpCode::Decode(instr);
switch (opcode->get().GetId()) {
case OpCode::Id::PSETP: {
const Node op_a = GetPredicate(instr.psetp.pred12, instr.psetp.neg_pred12 != 0);
const Node op_b = GetPredicate(instr.psetp.pred29, instr.psetp.neg_pred29 != 0);
// We can't use the constant predicate as destination.
ASSERT(instr.psetp.pred3 != static_cast<u64>(Pred::UnusedIndex));
const Node second_pred = GetPredicate(instr.psetp.pred39, instr.psetp.neg_pred39 != 0);
const OperationCode combiner = GetPredicateCombiner(instr.psetp.op);
const Node predicate = Operation(combiner, op_a, op_b);
// Set the primary predicate to the result of Predicate OP SecondPredicate
SetPredicate(bb, instr.psetp.pred3, Operation(combiner, predicate, second_pred));
if (instr.psetp.pred0 != static_cast<u64>(Pred::UnusedIndex)) {
// Set the secondary predicate to the result of !Predicate OP SecondPredicate, if
// enabled
SetPredicate(bb, instr.psetp.pred0,
Operation(combiner, Operation(OperationCode::LogicalNegate, predicate),
second_pred));
}
break;
}
case OpCode::Id::CSETP: {
const Node pred = GetPredicate(instr.csetp.pred39, instr.csetp.neg_pred39 != 0);
const Node condition_code = GetConditionCode(instr.csetp.cc);
const OperationCode combiner = GetPredicateCombiner(instr.csetp.op);
if (instr.csetp.pred3 != static_cast<u64>(Pred::UnusedIndex)) {
SetPredicate(bb, instr.csetp.pred3, Operation(combiner, condition_code, pred));
}
if (instr.csetp.pred0 != static_cast<u64>(Pred::UnusedIndex)) {
const Node neg_cc = Operation(OperationCode::LogicalNegate, condition_code);
SetPredicate(bb, instr.csetp.pred0, Operation(combiner, neg_cc, pred));
}
break;
}
default:
UNIMPLEMENTED_MSG("Unhandled predicate instruction: {}", opcode->get().GetName());
}
return pc;
}
} // namespace VideoCommon::Shader

46
src/video_core/shader/decode/predicate_set_register.cpp
View file

@ -1,46 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/assert.h"
#include "common/common_types.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
u32 ShaderIR::DecodePredicateSetRegister(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
UNIMPLEMENTED_IF_MSG(instr.generates_cc,
"Condition codes generation in PSET is not implemented");
const Node op_a = GetPredicate(instr.pset.pred12, instr.pset.neg_pred12 != 0);
const Node op_b = GetPredicate(instr.pset.pred29, instr.pset.neg_pred29 != 0);
const Node first_pred = Operation(GetPredicateCombiner(instr.pset.cond), op_a, op_b);
const Node second_pred = GetPredicate(instr.pset.pred39, instr.pset.neg_pred39 != 0);
const OperationCode combiner = GetPredicateCombiner(instr.pset.op);
const Node predicate = Operation(combiner, first_pred, second_pred);
const Node true_value = instr.pset.bf ? Immediate(1.0f) : Immediate(0xffffffff);
const Node false_value = instr.pset.bf ? Immediate(0.0f) : Immediate(0);
const Node value =
Operation(OperationCode::Select, PRECISE, predicate, true_value, false_value);
if (instr.pset.bf) {
SetInternalFlagsFromFloat(bb, value, instr.generates_cc);
} else {
SetInternalFlagsFromInteger(bb, value, instr.generates_cc);
}
SetRegister(bb, instr.gpr0, value);
return pc;
}
} // namespace VideoCommon::Shader

86
src/video_core/shader/decode/register_set_predicate.cpp
View file

@ -1,86 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <utility>
#include "common/assert.h"
#include "common/common_types.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
using std::move;
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
namespace {
constexpr u64 NUM_CONDITION_CODES = 4;
constexpr u64 NUM_PREDICATES = 7;
} // namespace
u32 ShaderIR::DecodeRegisterSetPredicate(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
const auto opcode = OpCode::Decode(instr);
Node apply_mask = [this, opcode, instr] {
switch (opcode->get().GetId()) {
case OpCode::Id::R2P_IMM:
case OpCode::Id::P2R_IMM:
return Immediate(static_cast<u32>(instr.p2r_r2p.immediate_mask));
default:
UNREACHABLE();
return Immediate(0);
}
}();
const u32 offset = static_cast<u32>(instr.p2r_r2p.byte) * 8;
const bool cc = instr.p2r_r2p.mode == Tegra::Shader::R2pMode::Cc;
const u64 num_entries = cc ? NUM_CONDITION_CODES : NUM_PREDICATES;
const auto get_entry = [this, cc](u64 entry) {
return cc ? GetInternalFlag(static_cast<InternalFlag>(entry)) : GetPredicate(entry);
};
switch (opcode->get().GetId()) {
case OpCode::Id::R2P_IMM: {
Node mask = GetRegister(instr.gpr8);
for (u64 entry = 0; entry < num_entries; ++entry) {
const u32 shift = static_cast<u32>(entry);
Node apply = BitfieldExtract(apply_mask, shift, 1);
Node condition = Operation(OperationCode::LogicalUNotEqual, apply, Immediate(0));
Node compare = BitfieldExtract(mask, offset + shift, 1);
Node value = Operation(OperationCode::LogicalUNotEqual, move(compare), Immediate(0));
Node code = Operation(OperationCode::LogicalAssign, get_entry(entry), move(value));
bb.push_back(Conditional(condition, {move(code)}));
}
break;
}
case OpCode::Id::P2R_IMM: {
Node value = Immediate(0);
for (u64 entry = 0; entry < num_entries; ++entry) {
Node bit = Operation(OperationCode::Select, get_entry(entry), Immediate(1U << entry),
Immediate(0));
value = Operation(OperationCode::UBitwiseOr, move(value), move(bit));
}
value = Operation(OperationCode::UBitwiseAnd, move(value), apply_mask);
value = BitfieldInsert(GetRegister(instr.gpr8), move(value), offset, 8);
SetRegister(bb, instr.gpr0, move(value));
break;
}
default:
UNIMPLEMENTED_MSG("Unhandled P2R/R2R instruction: {}", opcode->get().GetName());
break;
}
return pc;
}
} // namespace VideoCommon::Shader

153
src/video_core/shader/decode/shift.cpp
View file

@ -1,153 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/assert.h"
#include "common/common_types.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
using std::move;
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
using Tegra::Shader::ShfType;
using Tegra::Shader::ShfXmode;
namespace {
Node IsFull(Node shift) {
return Operation(OperationCode::LogicalIEqual, move(shift), Immediate(32));
}
Node Shift(OperationCode opcode, Node value, Node shift) {
Node shifted = Operation(opcode, move(value), shift);
return Operation(OperationCode::Select, IsFull(move(shift)), Immediate(0), move(shifted));
}
Node ClampShift(Node shift, s32 size = 32) {
shift = Operation(OperationCode::IMax, move(shift), Immediate(0));
return Operation(OperationCode::IMin, move(shift), Immediate(size));
}
Node WrapShift(Node shift, s32 size = 32) {
return Operation(OperationCode::UBitwiseAnd, move(shift), Immediate(size - 1));
}
Node ShiftRight(Node low, Node high, Node shift, Node low_shift, ShfType type) {
// These values are used when the shift value is less than 32
Node less_low = Shift(OperationCode::ILogicalShiftRight, low, shift);
Node less_high = Shift(OperationCode::ILogicalShiftLeft, high, low_shift);
Node less = Operation(OperationCode::IBitwiseOr, move(less_high), move(less_low));
if (type == ShfType::Bits32) {
// On 32 bit shifts we are either full (shifting 32) or shifting less than 32 bits
return Operation(OperationCode::Select, IsFull(move(shift)), move(high), move(less));
}
// And these when it's larger than or 32
const bool is_signed = type == ShfType::S64;
const auto opcode = SignedToUnsignedCode(OperationCode::IArithmeticShiftRight, is_signed);
Node reduced = Operation(OperationCode::IAdd, shift, Immediate(-32));
Node greater = Shift(opcode, high, move(reduced));
Node is_less = Operation(OperationCode::LogicalILessThan, shift, Immediate(32));
Node is_zero = Operation(OperationCode::LogicalIEqual, move(shift), Immediate(0));
Node value = Operation(OperationCode::Select, move(is_less), move(less), move(greater));
return Operation(OperationCode::Select, move(is_zero), move(high), move(value));
}
Node ShiftLeft(Node low, Node high, Node shift, Node low_shift, ShfType type) {
// These values are used when the shift value is less than 32
Node less_low = Operation(OperationCode::ILogicalShiftRight, low, low_shift);
Node less_high = Operation(OperationCode::ILogicalShiftLeft, high, shift);
Node less = Operation(OperationCode::IBitwiseOr, move(less_low), move(less_high));
if (type == ShfType::Bits32) {
// On 32 bit shifts we are either full (shifting 32) or shifting less than 32 bits
return Operation(OperationCode::Select, IsFull(move(shift)), move(low), move(less));
}
// And these when it's larger than or 32
Node reduced = Operation(OperationCode::IAdd, shift, Immediate(-32));
Node greater = Shift(OperationCode::ILogicalShiftLeft, move(low), move(reduced));
Node is_less = Operation(OperationCode::LogicalILessThan, shift, Immediate(32));
Node is_zero = Operation(OperationCode::LogicalIEqual, move(shift), Immediate(0));
Node value = Operation(OperationCode::Select, move(is_less), move(less), move(greater));
return Operation(OperationCode::Select, move(is_zero), move(high), move(value));
}
} // Anonymous namespace
u32 ShaderIR::DecodeShift(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
const auto opcode = OpCode::Decode(instr);
Node op_a = GetRegister(instr.gpr8);
Node op_b = [this, instr] {
if (instr.is_b_imm) {
return Immediate(instr.alu.GetSignedImm20_20());
} else if (instr.is_b_gpr) {
return GetRegister(instr.gpr20);
} else {
return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset());
}
}();
switch (const auto opid = opcode->get().GetId(); opid) {
case OpCode::Id::SHR_C:
case OpCode::Id::SHR_R:
case OpCode::Id::SHR_IMM: {
op_b = instr.shr.wrap ? WrapShift(move(op_b)) : ClampShift(move(op_b));
Node value = SignedOperation(OperationCode::IArithmeticShiftRight, instr.shift.is_signed,
move(op_a), move(op_b));
SetInternalFlagsFromInteger(bb, value, instr.generates_cc);
SetRegister(bb, instr.gpr0, move(value));
break;
}
case OpCode::Id::SHL_C:
case OpCode::Id::SHL_R:
case OpCode::Id::SHL_IMM: {
Node value = Operation(OperationCode::ILogicalShiftLeft, op_a, op_b);
SetInternalFlagsFromInteger(bb, value, instr.generates_cc);
SetRegister(bb, instr.gpr0, move(value));
break;
}
case OpCode::Id::SHF_RIGHT_R:
case OpCode::Id::SHF_RIGHT_IMM:
case OpCode::Id::SHF_LEFT_R:
case OpCode::Id::SHF_LEFT_IMM: {
UNIMPLEMENTED_IF(instr.generates_cc);
UNIMPLEMENTED_IF_MSG(instr.shf.xmode != ShfXmode::None, "xmode={}",
instr.shf.xmode.Value());
if (instr.is_b_imm) {
op_b = Immediate(static_cast<u32>(instr.shf.immediate));
}
const s32 size = instr.shf.type == ShfType::Bits32 ? 32 : 64;
Node shift = instr.shf.wrap ? WrapShift(move(op_b), size) : ClampShift(move(op_b), size);
Node negated_shift = Operation(OperationCode::INegate, shift);
Node low_shift = Operation(OperationCode::IAdd, move(negated_shift), Immediate(32));
const bool is_right = opid == OpCode::Id::SHF_RIGHT_R || opid == OpCode::Id::SHF_RIGHT_IMM;
Node value = (is_right ? ShiftRight : ShiftLeft)(
move(op_a), GetRegister(instr.gpr39), move(shift), move(low_shift), instr.shf.type);
SetRegister(bb, instr.gpr0, move(value));
break;
}
default:
UNIMPLEMENTED_MSG("Unhandled shift instruction: {}", opcode->get().GetName());
}
return pc;
}
} // namespace VideoCommon::Shader

935
src/video_core/shader/decode/texture.cpp
View file

@ -1,935 +0,0 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <algorithm>
#include <vector>
#include <fmt/format.h>
#include "common/assert.h"
#include "common/bit_field.h"
#include "common/common_types.h"
#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 {
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
using Tegra::Shader::Register;
using Tegra::Shader::TextureMiscMode;
using Tegra::Shader::TextureProcessMode;
using Tegra::Shader::TextureType;
static std::size_t GetCoordCount(TextureType texture_type) {
switch (texture_type) {
case TextureType::Texture1D:
return 1;
case TextureType::Texture2D:
return 2;
case TextureType::Texture3D:
case TextureType::TextureCube:
return 3;
default:
UNIMPLEMENTED_MSG("Unhandled texture type: {}", texture_type);
return 0;
}
}
u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
const auto opcode = OpCode::Decode(instr);
bool is_bindless = false;
switch (opcode->get().GetId()) {
case OpCode::Id::TEX: {
const TextureType texture_type{instr.tex.texture_type};
const bool is_array = instr.tex.array != 0;
const bool is_aoffi = instr.tex.UsesMiscMode(TextureMiscMode::AOFFI);
const bool depth_compare = instr.tex.UsesMiscMode(TextureMiscMode::DC);
const auto process_mode = instr.tex.GetTextureProcessMode();
WriteTexInstructionFloat(
bb, instr,
GetTexCode(instr, texture_type, process_mode, depth_compare, is_array, is_aoffi, {}));
break;
}
case OpCode::Id::TEX_B: {
UNIMPLEMENTED_IF_MSG(instr.tex.UsesMiscMode(TextureMiscMode::AOFFI),
"AOFFI is not implemented");
const TextureType texture_type{instr.tex_b.texture_type};
const bool is_array = instr.tex_b.array != 0;
const bool is_aoffi = instr.tex.UsesMiscMode(TextureMiscMode::AOFFI);
const bool depth_compare = instr.tex_b.UsesMiscMode(TextureMiscMode::DC);
const auto process_mode = instr.tex_b.GetTextureProcessMode();
WriteTexInstructionFloat(bb, instr,
GetTexCode(instr, texture_type, process_mode, depth_compare,
is_array, is_aoffi, {instr.gpr20}));
break;
}
case OpCode::Id::TEXS: {
const TextureType texture_type{instr.texs.GetTextureType()};
const bool is_array{instr.texs.IsArrayTexture()};
const bool depth_compare = instr.texs.UsesMiscMode(TextureMiscMode::DC);
const auto process_mode = instr.texs.GetTextureProcessMode();
const Node4 components =
GetTexsCode(instr, texture_type, process_mode, depth_compare, is_array);
if (instr.texs.fp32_flag) {
WriteTexsInstructionFloat(bb, instr, components);
} else {
WriteTexsInstructionHalfFloat(bb, instr, components);
}
break;
}
case OpCode::Id::TLD4_B: {
is_bindless = true;
[[fallthrough]];
}
case OpCode::Id::TLD4: {
UNIMPLEMENTED_IF_MSG(instr.tld4.UsesMiscMode(TextureMiscMode::NDV),
"NDV is not implemented");
const auto texture_type = instr.tld4.texture_type.Value();
const bool depth_compare = is_bindless ? instr.tld4_b.UsesMiscMode(TextureMiscMode::DC)
: instr.tld4.UsesMiscMode(TextureMiscMode::DC);
const bool is_array = instr.tld4.array != 0;
const bool is_aoffi = is_bindless ? instr.tld4_b.UsesMiscMode(TextureMiscMode::AOFFI)
: instr.tld4.UsesMiscMode(TextureMiscMode::AOFFI);
const bool is_ptp = is_bindless ? instr.tld4_b.UsesMiscMode(TextureMiscMode::PTP)
: instr.tld4.UsesMiscMode(TextureMiscMode::PTP);
WriteTexInstructionFloat(bb, instr,
GetTld4Code(instr, texture_type, depth_compare, is_array, is_aoffi,
is_ptp, is_bindless));
break;
}
case OpCode::Id::TLD4S: {
constexpr std::size_t num_coords = 2;
const bool is_aoffi = instr.tld4s.UsesMiscMode(TextureMiscMode::AOFFI);
const bool is_depth_compare = instr.tld4s.UsesMiscMode(TextureMiscMode::DC);
const Node op_a = GetRegister(instr.gpr8);
const Node op_b = GetRegister(instr.gpr20);
// TODO(Subv): Figure out how the sampler type is encoded in the TLD4S instruction.
std::vector<Node> coords;
std::vector<Node> aoffi;
Node depth_compare;
if (is_depth_compare) {
// Note: TLD4S coordinate encoding works just like TEXS's
const Node op_y = GetRegister(instr.gpr8.Value() + 1);
coords.push_back(op_a);
coords.push_back(op_y);
if (is_aoffi) {
aoffi = GetAoffiCoordinates(op_b, num_coords, true);
depth_compare = GetRegister(instr.gpr20.Value() + 1);
} else {
depth_compare = op_b;
}
} else {
// There's no depth compare
coords.push_back(op_a);
if (is_aoffi) {
coords.push_back(GetRegister(instr.gpr8.Value() + 1));
aoffi = GetAoffiCoordinates(op_b, num_coords, true);
} else {
coords.push_back(op_b);
}
}
const Node component = Immediate(static_cast<u32>(instr.tld4s.component));
SamplerInfo info;
info.is_shadow = is_depth_compare;
const std::optional<SamplerEntry> sampler = GetSampler(instr.sampler, info);
Node4 values;
for (u32 element = 0; element < values.size(); ++element) {
MetaTexture meta{*sampler, {}, depth_compare, aoffi, {}, {},
{}, {}, component, element, {}};
values[element] = Operation(OperationCode::TextureGather, meta, coords);
}
if (instr.tld4s.fp16_flag) {
WriteTexsInstructionHalfFloat(bb, instr, values, true);
} else {
WriteTexsInstructionFloat(bb, instr, values, true);
}
break;
}
case OpCode::Id::TXD_B:
is_bindless = true;
[[fallthrough]];
case OpCode::Id::TXD: {
UNIMPLEMENTED_IF_MSG(instr.txd.UsesMiscMode(TextureMiscMode::AOFFI),
"AOFFI is not implemented");
const bool is_array = instr.txd.is_array != 0;
const auto derivate_reg = instr.gpr20.Value();
const auto texture_type = instr.txd.texture_type.Value();
const auto coord_count = GetCoordCount(texture_type);
u64 base_reg = instr.gpr8.Value();
Node index_var;
SamplerInfo info;
info.type = texture_type;
info.is_array = is_array;
const std::optional<SamplerEntry> sampler =
is_bindless ? GetBindlessSampler(base_reg, info, index_var)
: GetSampler(instr.sampler, info);
Node4 values;
if (!sampler) {
std::generate(values.begin(), values.end(), [this] { return Immediate(0); });
WriteTexInstructionFloat(bb, instr, values);
break;
}
if (is_bindless) {
base_reg++;
}
std::vector<Node> coords;
std::vector<Node> derivates;
for (std::size_t i = 0; i < coord_count; ++i) {
coords.push_back(GetRegister(base_reg + i));
const std::size_t derivate = i * 2;
derivates.push_back(GetRegister(derivate_reg + derivate));
derivates.push_back(GetRegister(derivate_reg + derivate + 1));
}
Node array_node = {};
if (is_array) {
const Node info_reg = GetRegister(base_reg + coord_count);
array_node = BitfieldExtract(info_reg, 0, 16);
}
for (u32 element = 0; element < values.size(); ++element) {
MetaTexture meta{*sampler, array_node, {}, {}, {}, derivates,
{}, {}, {}, element, index_var};
values[element] = Operation(OperationCode::TextureGradient, std::move(meta), coords);
}
WriteTexInstructionFloat(bb, instr, values);
break;
}
case OpCode::Id::TXQ_B:
is_bindless = true;
[[fallthrough]];
case OpCode::Id::TXQ: {
Node index_var;
const std::optional<SamplerEntry> sampler =
is_bindless ? GetBindlessSampler(instr.gpr8, {}, index_var)
: GetSampler(instr.sampler, {});
if (!sampler) {
u32 indexer = 0;
for (u32 element = 0; element < 4; ++element) {
if (!instr.txq.IsComponentEnabled(element)) {
continue;
}
const Node value = Immediate(0);
SetTemporary(bb, indexer++, value);
}
for (u32 i = 0; i < indexer; ++i) {
SetRegister(bb, instr.gpr0.Value() + i, GetTemporary(i));
}
break;
}
u32 indexer = 0;
switch (instr.txq.query_type) {
case Tegra::Shader::TextureQueryType::Dimension: {
for (u32 element = 0; element < 4; ++element) {
if (!instr.txq.IsComponentEnabled(element)) {
continue;
}
MetaTexture meta{*sampler, {}, {}, {}, {}, {}, {}, {}, {}, element, index_var};
const Node value =
Operation(OperationCode::TextureQueryDimensions, meta,
GetRegister(instr.gpr8.Value() + (is_bindless ? 1 : 0)));
SetTemporary(bb, indexer++, value);
}
for (u32 i = 0; i < indexer; ++i) {
SetRegister(bb, instr.gpr0.Value() + i, GetTemporary(i));
}
break;
}
default:
UNIMPLEMENTED_MSG("Unhandled texture query type: {}", instr.txq.query_type.Value());
}
break;
}
case OpCode::Id::TMML_B:
is_bindless = true;
[[fallthrough]];
case OpCode::Id::TMML: {
UNIMPLEMENTED_IF_MSG(instr.tmml.UsesMiscMode(Tegra::Shader::TextureMiscMode::NDV),
"NDV is not implemented");
const auto texture_type = instr.tmml.texture_type.Value();
const bool is_array = instr.tmml.array != 0;
SamplerInfo info;
info.type = texture_type;
info.is_array = is_array;
Node index_var;
const std::optional<SamplerEntry> sampler =
is_bindless ? GetBindlessSampler(instr.gpr20, info, index_var)
: GetSampler(instr.sampler, info);
if (!sampler) {
u32 indexer = 0;
for (u32 element = 0; element < 2; ++element) {
if (!instr.tmml.IsComponentEnabled(element)) {
continue;
}
const Node value = Immediate(0);
SetTemporary(bb, indexer++, value);
}
for (u32 i = 0; i < indexer; ++i) {
SetRegister(bb, instr.gpr0.Value() + i, GetTemporary(i));
}
break;
}
const u64 base_index = is_array ? 1 : 0;
const u64 num_components = [texture_type] {
switch (texture_type) {
case TextureType::Texture1D:
return 1;
case TextureType::Texture2D:
return 2;
case TextureType::TextureCube:
return 3;
default:
UNIMPLEMENTED_MSG("Unhandled texture type {}", texture_type);
return 2;
}
}();
// TODO: What's the array component used for?
std::vector<Node> coords;
coords.reserve(num_components);
for (u64 component = 0; component < num_components; ++component) {
coords.push_back(GetRegister(instr.gpr8.Value() + base_index + component));
}
u32 indexer = 0;
for (u32 element = 0; element < 2; ++element) {
if (!instr.tmml.IsComponentEnabled(element)) {
continue;
}
MetaTexture meta{*sampler, {}, {}, {}, {}, {}, {}, {}, {}, element, index_var};
Node value = Operation(OperationCode::TextureQueryLod, meta, coords);
SetTemporary(bb, indexer++, std::move(value));
}
for (u32 i = 0; i < indexer; ++i) {
SetRegister(bb, instr.gpr0.Value() + i, GetTemporary(i));
}
break;
}
case OpCode::Id::TLD: {
UNIMPLEMENTED_IF_MSG(instr.tld.aoffi, "AOFFI is not implemented");
UNIMPLEMENTED_IF_MSG(instr.tld.ms, "MS is not implemented");
UNIMPLEMENTED_IF_MSG(instr.tld.cl, "CL is not implemented");
WriteTexInstructionFloat(bb, instr, GetTldCode(instr));
break;
}
case OpCode::Id::TLDS: {
const TextureType texture_type{instr.tlds.GetTextureType()};
const bool is_array{instr.tlds.IsArrayTexture()};
UNIMPLEMENTED_IF_MSG(instr.tlds.UsesMiscMode(TextureMiscMode::AOFFI),
"AOFFI is not implemented");
UNIMPLEMENTED_IF_MSG(instr.tlds.UsesMiscMode(TextureMiscMode::MZ), "MZ is not implemented");
const Node4 components = GetTldsCode(instr, texture_type, is_array);
if (instr.tlds.fp32_flag) {
WriteTexsInstructionFloat(bb, instr, components);
} else {
WriteTexsInstructionHalfFloat(bb, instr, components);
}
break;
}
default:
UNIMPLEMENTED_MSG("Unhandled memory instruction: {}", opcode->get().GetName());
}
return pc;
}
ShaderIR::SamplerInfo ShaderIR::GetSamplerInfo(
SamplerInfo info, std::optional<Tegra::Engines::SamplerDescriptor> sampler) {
if (info.IsComplete()) {
return info;
}
if (!sampler) {
LOG_WARNING(HW_GPU, "Unknown sampler info");
info.type = info.type.value_or(Tegra::Shader::TextureType::Texture2D);
info.is_array = info.is_array.value_or(false);
info.is_shadow = info.is_shadow.value_or(false);
info.is_buffer = info.is_buffer.value_or(false);
return info;
}
info.type = info.type.value_or(sampler->texture_type);
info.is_array = info.is_array.value_or(sampler->is_array != 0);
info.is_shadow = info.is_shadow.value_or(sampler->is_shadow != 0);
info.is_buffer = info.is_buffer.value_or(sampler->is_buffer != 0);
return info;
}
std::optional<SamplerEntry> ShaderIR::GetSampler(Tegra::Shader::Sampler sampler,
SamplerInfo sampler_info) {
const u32 offset = static_cast<u32>(sampler.index.Value());
const auto info = GetSamplerInfo(sampler_info, registry.ObtainBoundSampler(offset));
// If this sampler has already been used, return the existing mapping.
const auto it =
std::find_if(used_samplers.begin(), used_samplers.end(),
[offset](const SamplerEntry& entry) { return entry.offset == offset; });
if (it != used_samplers.end()) {
ASSERT(!it->is_bindless && it->type == info.type && it->is_array == info.is_array &&
it->is_shadow == info.is_shadow && it->is_buffer == info.is_buffer);
return *it;
}
// Otherwise create a new mapping for this sampler
const auto next_index = static_cast<u32>(used_samplers.size());
return used_samplers.emplace_back(next_index, offset, *info.type, *info.is_array,
*info.is_shadow, *info.is_buffer, false);
}
std::optional<SamplerEntry> ShaderIR::GetBindlessSampler(Tegra::Shader::Register reg,
SamplerInfo info, Node& index_var) {
const Node sampler_register = GetRegister(reg);
const auto [base_node, tracked_sampler_info] =
TrackBindlessSampler(sampler_register, global_code, static_cast<s64>(global_code.size()));
if (!base_node) {
UNREACHABLE();
return std::nullopt;
}
if (const auto sampler_info = std::get_if<BindlessSamplerNode>(&*tracked_sampler_info)) {
const u32 buffer = sampler_info->index;
const u32 offset = sampler_info->offset;
info = GetSamplerInfo(info, registry.ObtainBindlessSampler(buffer, offset));
// If this sampler has already been used, return the existing mapping.
const auto it = std::find_if(used_samplers.begin(), used_samplers.end(),
[buffer, offset](const SamplerEntry& entry) {
return entry.buffer == buffer && entry.offset == offset;
});
if (it != used_samplers.end()) {
ASSERT(it->is_bindless && it->type == info.type && it->is_array == info.is_array &&
it->is_shadow == info.is_shadow);
return *it;
}
// Otherwise create a new mapping for this sampler
const auto next_index = static_cast<u32>(used_samplers.size());
return used_samplers.emplace_back(next_index, offset, buffer, *info.type, *info.is_array,
*info.is_shadow, *info.is_buffer, false);
}
if (const auto sampler_info = std::get_if<SeparateSamplerNode>(&*tracked_sampler_info)) {
const std::pair indices = sampler_info->indices;
const std::pair offsets = sampler_info->offsets;
info = GetSamplerInfo(info, registry.ObtainSeparateSampler(indices, offsets));
// Try to use an already created sampler if it exists
const auto it =
std::find_if(used_samplers.begin(), used_samplers.end(),
[indices, offsets](const SamplerEntry& entry) {
return offsets == std::pair{entry.offset, entry.secondary_offset} &&
indices == std::pair{entry.buffer, entry.secondary_buffer};
});
if (it != used_samplers.end()) {
ASSERT(it->is_separated && it->type == info.type && it->is_array == info.is_array &&
it->is_shadow == info.is_shadow && it->is_buffer == info.is_buffer);
return *it;
}
// Otherwise create a new mapping for this sampler
const u32 next_index = static_cast<u32>(used_samplers.size());
return used_samplers.emplace_back(next_index, offsets, indices, *info.type, *info.is_array,
*info.is_shadow, *info.is_buffer);
}
if (const auto sampler_info = std::get_if<ArraySamplerNode>(&*tracked_sampler_info)) {
const u32 base_offset = sampler_info->base_offset / 4;
index_var = GetCustomVariable(sampler_info->bindless_var);
info = GetSamplerInfo(info, registry.ObtainBoundSampler(base_offset));
// If this sampler has already been used, return the existing mapping.
const auto it = std::find_if(
used_samplers.begin(), used_samplers.end(),
[base_offset](const SamplerEntry& entry) { return entry.offset == base_offset; });
if (it != used_samplers.end()) {
ASSERT(!it->is_bindless && it->type == info.type && it->is_array == info.is_array &&
it->is_shadow == info.is_shadow && it->is_buffer == info.is_buffer &&
it->is_indexed);
return *it;
}
uses_indexed_samplers = true;
// Otherwise create a new mapping for this sampler
const auto next_index = static_cast<u32>(used_samplers.size());
return used_samplers.emplace_back(next_index, base_offset, *info.type, *info.is_array,
*info.is_shadow, *info.is_buffer, true);
}
return std::nullopt;
}
void ShaderIR::WriteTexInstructionFloat(NodeBlock& bb, Instruction instr, const Node4& components) {
u32 dest_elem = 0;
for (u32 elem = 0; elem < 4; ++elem) {
if (!instr.tex.IsComponentEnabled(elem)) {
// Skip disabled components
continue;
}
SetTemporary(bb, dest_elem++, components[elem]);
}
// After writing values in temporals, move them to the real registers
for (u32 i = 0; i < dest_elem; ++i) {
SetRegister(bb, instr.gpr0.Value() + i, GetTemporary(i));
}
}
void ShaderIR::WriteTexsInstructionFloat(NodeBlock& bb, Instruction instr, const Node4& components,
bool ignore_mask) {
// TEXS has two destination registers and a swizzle. The first two elements in the swizzle
// go into gpr0+0 and gpr0+1, and the rest goes into gpr28+0 and gpr28+1
u32 dest_elem = 0;
for (u32 component = 0; component < 4; ++component) {
if (!instr.texs.IsComponentEnabled(component) && !ignore_mask)
continue;
SetTemporary(bb, dest_elem++, components[component]);
}
for (u32 i = 0; i < dest_elem; ++i) {
if (i < 2) {
// Write the first two swizzle components to gpr0 and gpr0+1
SetRegister(bb, instr.gpr0.Value() + i % 2, GetTemporary(i));
} else {
ASSERT(instr.texs.HasTwoDestinations());
// Write the rest of the swizzle components to gpr28 and gpr28+1
SetRegister(bb, instr.gpr28.Value() + i % 2, GetTemporary(i));
}
}
}
void ShaderIR::WriteTexsInstructionHalfFloat(NodeBlock& bb, Instruction instr,
const Node4& components, bool ignore_mask) {
// TEXS.F16 destionation registers are packed in two registers in pairs (just like any half
// float instruction).
Node4 values;
u32 dest_elem = 0;
for (u32 component = 0; component < 4; ++component) {
if (!instr.texs.IsComponentEnabled(component) && !ignore_mask)
continue;
values[dest_elem++] = components[component];
}
if (dest_elem == 0)
return;
std::generate(values.begin() + dest_elem, values.end(), [&]() { return Immediate(0); });
const Node first_value = Operation(OperationCode::HPack2, values[0], values[1]);
if (dest_elem <= 2) {
SetRegister(bb, instr.gpr0, first_value);
return;
}
SetTemporary(bb, 0, first_value);
SetTemporary(bb, 1, Operation(OperationCode::HPack2, values[2], values[3]));
SetRegister(bb, instr.gpr0, GetTemporary(0));
SetRegister(bb, instr.gpr28, GetTemporary(1));
}
Node4 ShaderIR::GetTextureCode(Instruction instr, TextureType texture_type,
TextureProcessMode process_mode, std::vector<Node> coords,
Node array, Node depth_compare, u32 bias_offset,
std::vector<Node> aoffi,
std::optional<Tegra::Shader::Register> bindless_reg) {
const bool is_array = array != nullptr;
const bool is_shadow = depth_compare != nullptr;
const bool is_bindless = bindless_reg.has_value();
ASSERT_MSG(texture_type != TextureType::Texture3D || !is_array || !is_shadow,
"Illegal texture type");
SamplerInfo info;
info.type = texture_type;
info.is_array = is_array;
info.is_shadow = is_shadow;
info.is_buffer = false;
Node index_var;
const std::optional<SamplerEntry> sampler =
is_bindless ? GetBindlessSampler(*bindless_reg, info, index_var)
: GetSampler(instr.sampler, info);
if (!sampler) {
return {Immediate(0), Immediate(0), Immediate(0), Immediate(0)};
}
const bool lod_needed = process_mode == TextureProcessMode::LZ ||
process_mode == TextureProcessMode::LL ||
process_mode == TextureProcessMode::LLA;
const OperationCode opcode = lod_needed ? OperationCode::TextureLod : OperationCode::Texture;
Node bias;
Node lod;
switch (process_mode) {
case TextureProcessMode::None:
break;
case TextureProcessMode::LZ:
lod = Immediate(0.0f);
break;
case TextureProcessMode::LB:
// If present, lod or bias are always stored in the register indexed by the gpr20 field with
// an offset depending on the usage of the other registers.
bias = GetRegister(instr.gpr20.Value() + bias_offset);
break;
case TextureProcessMode::LL:
lod = GetRegister(instr.gpr20.Value() + bias_offset);
break;
default:
UNIMPLEMENTED_MSG("Unimplemented process mode={}", process_mode);
break;
}
Node4 values;
for (u32 element = 0; element < values.size(); ++element) {
MetaTexture meta{*sampler, array, depth_compare, aoffi, {}, {}, bias,
lod, {}, element, index_var};
values[element] = Operation(opcode, meta, coords);
}
return values;
}
Node4 ShaderIR::GetTexCode(Instruction instr, TextureType texture_type,
TextureProcessMode process_mode, bool depth_compare, bool is_array,
bool is_aoffi, std::optional<Tegra::Shader::Register> bindless_reg) {
const bool lod_bias_enabled{
(process_mode != TextureProcessMode::None && process_mode != TextureProcessMode::LZ)};
const bool is_bindless = bindless_reg.has_value();
u64 parameter_register = instr.gpr20.Value();
if (is_bindless) {
++parameter_register;
}
const u32 bias_lod_offset = (is_bindless ? 1 : 0);
if (lod_bias_enabled) {
++parameter_register;
}
const auto coord_counts = ValidateAndGetCoordinateElement(texture_type, depth_compare, is_array,
lod_bias_enabled, 4, 5);
const auto coord_count = std::get<0>(coord_counts);
// If enabled arrays index is always stored in the gpr8 field
const u64 array_register = instr.gpr8.Value();
// First coordinate index is the gpr8 or gpr8 + 1 when arrays are used
const u64 coord_register = array_register + (is_array ? 1 : 0);
std::vector<Node> coords;
for (std::size_t i = 0; i < coord_count; ++i) {
coords.push_back(GetRegister(coord_register + i));
}
// 1D.DC in OpenGL the 2nd component is ignored.
if (depth_compare && !is_array && texture_type == TextureType::Texture1D) {
coords.push_back(Immediate(0.0f));
}
const Node array = is_array ? GetRegister(array_register) : nullptr;
std::vector<Node> aoffi;
if (is_aoffi) {
aoffi = GetAoffiCoordinates(GetRegister(parameter_register++), coord_count, false);
}
Node dc;
if (depth_compare) {
// Depth is always stored in the register signaled by gpr20 or in the next register if lod
// or bias are used
dc = GetRegister(parameter_register++);
}
return GetTextureCode(instr, texture_type, process_mode, coords, array, dc, bias_lod_offset,
aoffi, bindless_reg);
}
Node4 ShaderIR::GetTexsCode(Instruction instr, TextureType texture_type,
TextureProcessMode process_mode, bool depth_compare, bool is_array) {
const bool lod_bias_enabled =
(process_mode != TextureProcessMode::None && process_mode != TextureProcessMode::LZ);
const auto coord_counts = ValidateAndGetCoordinateElement(texture_type, depth_compare, is_array,
lod_bias_enabled, 4, 4);
const auto coord_count = std::get<0>(coord_counts);
// If enabled arrays index is always stored in the gpr8 field
const u64 array_register = instr.gpr8.Value();
// First coordinate index is stored in gpr8 field or (gpr8 + 1) when arrays are used
const u64 coord_register = array_register + (is_array ? 1 : 0);
const u64 last_coord_register =
(is_array || !(lod_bias_enabled || depth_compare) || (coord_count > 2))
? static_cast<u64>(instr.gpr20.Value())
: coord_register + 1;
const u32 bias_offset = coord_count > 2 ? 1 : 0;
std::vector<Node> coords;
for (std::size_t i = 0; i < coord_count; ++i) {
const bool last = (i == (coord_count - 1)) && (coord_count > 1);
coords.push_back(GetRegister(last ? last_coord_register : coord_register + i));
}
const Node array = is_array ? GetRegister(array_register) : nullptr;
Node dc;
if (depth_compare) {
// Depth is always stored in the register signaled by gpr20 or in the next register if lod
// or bias are used
const u64 depth_register = instr.gpr20.Value() + (lod_bias_enabled ? 1 : 0);
dc = GetRegister(depth_register);
}
return GetTextureCode(instr, texture_type, process_mode, coords, array, dc, bias_offset, {},
{});
}
Node4 ShaderIR::GetTld4Code(Instruction instr, TextureType texture_type, bool depth_compare,
bool is_array, bool is_aoffi, bool is_ptp, bool is_bindless) {
ASSERT_MSG(!(is_aoffi && is_ptp), "AOFFI and PTP can't be enabled at the same time");
const std::size_t coord_count = GetCoordCount(texture_type);
// If enabled arrays index is always stored in the gpr8 field
const u64 array_register = instr.gpr8.Value();
// First coordinate index is the gpr8 or gpr8 + 1 when arrays are used
const u64 coord_register = array_register + (is_array ? 1 : 0);
std::vector<Node> coords;
for (std::size_t i = 0; i < coord_count; ++i) {
coords.push_back(GetRegister(coord_register + i));
}
u64 parameter_register = instr.gpr20.Value();
SamplerInfo info;
info.type = texture_type;
info.is_array = is_array;
info.is_shadow = depth_compare;
Node index_var;
const std::optional<SamplerEntry> sampler =
is_bindless ? GetBindlessSampler(parameter_register++, info, index_var)
: GetSampler(instr.sampler, info);
Node4 values;
if (!sampler) {
for (u32 element = 0; element < values.size(); ++element) {
values[element] = Immediate(0);
}
return values;
}
std::vector<Node> aoffi, ptp;
if (is_aoffi) {
aoffi = GetAoffiCoordinates(GetRegister(parameter_register++), coord_count, true);
} else if (is_ptp) {
ptp = GetPtpCoordinates(
{GetRegister(parameter_register++), GetRegister(parameter_register++)});
}
Node dc;
if (depth_compare) {
dc = GetRegister(parameter_register++);
}
const Node component = is_bindless ? Immediate(static_cast<u32>(instr.tld4_b.component))
: Immediate(static_cast<u32>(instr.tld4.component));
for (u32 element = 0; element < values.size(); ++element) {
auto coords_copy = coords;
MetaTexture meta{
*sampler, GetRegister(array_register), dc, aoffi, ptp, {}, {}, {}, component, element,
index_var};
values[element] = Operation(OperationCode::TextureGather, meta, std::move(coords_copy));
}
return values;
}
Node4 ShaderIR::GetTldCode(Tegra::Shader::Instruction instr) {
const auto texture_type{instr.tld.texture_type};
const bool is_array{instr.tld.is_array != 0};
const bool lod_enabled{instr.tld.GetTextureProcessMode() == TextureProcessMode::LL};
const std::size_t coord_count{GetCoordCount(texture_type)};
u64 gpr8_cursor{instr.gpr8.Value()};
const Node array_register{is_array ? GetRegister(gpr8_cursor++) : nullptr};
std::vector<Node> coords;
coords.reserve(coord_count);
for (std::size_t i = 0; i < coord_count; ++i) {
coords.push_back(GetRegister(gpr8_cursor++));
}
u64 gpr20_cursor{instr.gpr20.Value()};
// const Node bindless_register{is_bindless ? GetRegister(gpr20_cursor++) : nullptr};
const Node lod{lod_enabled ? GetRegister(gpr20_cursor++) : Immediate(0u)};
// const Node aoffi_register{is_aoffi ? GetRegister(gpr20_cursor++) : nullptr};
// const Node multisample{is_multisample ? GetRegister(gpr20_cursor++) : nullptr};
const std::optional<SamplerEntry> sampler = GetSampler(instr.sampler, {});
Node4 values;
for (u32 element = 0; element < values.size(); ++element) {
auto coords_copy = coords;
MetaTexture meta{*sampler, array_register, {}, {}, {}, {}, {}, lod, {}, element, {}};
values[element] = Operation(OperationCode::TexelFetch, meta, std::move(coords_copy));
}
return values;
}
Node4 ShaderIR::GetTldsCode(Instruction instr, TextureType texture_type, bool is_array) {
SamplerInfo info;
info.type = texture_type;
info.is_array = is_array;
info.is_shadow = false;
const std::optional<SamplerEntry> sampler = GetSampler(instr.sampler, info);
const std::size_t type_coord_count = GetCoordCount(texture_type);
const bool lod_enabled = instr.tlds.GetTextureProcessMode() == TextureProcessMode::LL;
const bool aoffi_enabled = instr.tlds.UsesMiscMode(TextureMiscMode::AOFFI);
// If enabled arrays index is always stored in the gpr8 field
const u64 array_register = instr.gpr8.Value();
// if is array gpr20 is used
const u64 coord_register = is_array ? instr.gpr20.Value() : instr.gpr8.Value();
const u64 last_coord_register =
((type_coord_count > 2) || (type_coord_count == 2 && !lod_enabled)) && !is_array
? static_cast<u64>(instr.gpr20.Value())
: coord_register + 1;
std::vector<Node> coords;
for (std::size_t i = 0; i < type_coord_count; ++i) {
const bool last = (i == (type_coord_count - 1)) && (type_coord_count > 1);
coords.push_back(
GetRegister(last && !aoffi_enabled ? last_coord_register : coord_register + i));
}
const Node array = is_array ? GetRegister(array_register) : nullptr;
// When lod is used always is in gpr20
const Node lod = lod_enabled ? GetRegister(instr.gpr20) : Immediate(0);
std::vector<Node> aoffi;
if (aoffi_enabled) {
aoffi = GetAoffiCoordinates(GetRegister(instr.gpr20), type_coord_count, false);
}
Node4 values;
for (u32 element = 0; element < values.size(); ++element) {
auto coords_copy = coords;
MetaTexture meta{*sampler, array, {}, aoffi, {}, {}, {}, lod, {}, element, {}};
values[element] = Operation(OperationCode::TexelFetch, meta, std::move(coords_copy));
}
return values;
}
std::tuple<std::size_t, std::size_t> ShaderIR::ValidateAndGetCoordinateElement(
TextureType texture_type, bool depth_compare, bool is_array, bool lod_bias_enabled,
std::size_t max_coords, std::size_t max_inputs) {
const std::size_t coord_count = GetCoordCount(texture_type);
std::size_t total_coord_count = coord_count + (is_array ? 1 : 0) + (depth_compare ? 1 : 0);
const std::size_t total_reg_count = total_coord_count + (lod_bias_enabled ? 1 : 0);
if (total_coord_count > max_coords || total_reg_count > max_inputs) {
UNIMPLEMENTED_MSG("Unsupported Texture operation");
total_coord_count = std::min(total_coord_count, max_coords);
}
// 1D.DC OpenGL is using a vec3 but 2nd component is ignored later.
total_coord_count +=
(depth_compare && !is_array && texture_type == TextureType::Texture1D) ? 1 : 0;
return {coord_count, total_coord_count};
}
std::vector<Node> ShaderIR::GetAoffiCoordinates(Node aoffi_reg, std::size_t coord_count,
bool is_tld4) {
const std::array coord_offsets = is_tld4 ? std::array{0U, 8U, 16U} : std::array{0U, 4U, 8U};
const u32 size = is_tld4 ? 6 : 4;
const s32 wrap_value = is_tld4 ? 32 : 8;
const s32 diff_value = is_tld4 ? 64 : 16;
const u32 mask = (1U << size) - 1;
std::vector<Node> aoffi;
aoffi.reserve(coord_count);
const auto aoffi_immediate{
TrackImmediate(aoffi_reg, global_code, static_cast<s64>(global_code.size()))};
if (!aoffi_immediate) {
// Variable access, not supported on AMD.
LOG_WARNING(HW_GPU,
"AOFFI constant folding failed, some hardware might have graphical issues");
for (std::size_t coord = 0; coord < coord_count; ++coord) {
const Node value = BitfieldExtract(aoffi_reg, coord_offsets[coord], size);
const Node condition =
Operation(OperationCode::LogicalIGreaterEqual, value, Immediate(wrap_value));
const Node negative = Operation(OperationCode::IAdd, value, Immediate(-diff_value));
aoffi.push_back(Operation(OperationCode::Select, condition, negative, value));
}
return aoffi;
}
for (std::size_t coord = 0; coord < coord_count; ++coord) {
s32 value = (*aoffi_immediate >> coord_offsets[coord]) & mask;
if (value >= wrap_value) {
value -= diff_value;
}
aoffi.push_back(Immediate(value));
}
return aoffi;
}
std::vector<Node> ShaderIR::GetPtpCoordinates(std::array<Node, 2> ptp_regs) {
static constexpr u32 num_entries = 8;
std::vector<Node> ptp;
ptp.reserve(num_entries);
const auto global_size = static_cast<s64>(global_code.size());
const std::optional low = TrackImmediate(ptp_regs[0], global_code, global_size);
const std::optional high = TrackImmediate(ptp_regs[1], global_code, global_size);
if (!low || !high) {
for (u32 entry = 0; entry < num_entries; ++entry) {
const u32 reg = entry / 4;
const u32 offset = entry % 4;
const Node value = BitfieldExtract(ptp_regs[reg], offset * 8, 6);
const Node condition =
Operation(OperationCode::LogicalIGreaterEqual, value, Immediate(32));
const Node negative = Operation(OperationCode::IAdd, value, Immediate(-64));
ptp.push_back(Operation(OperationCode::Select, condition, negative, value));
}
return ptp;
}
const u64 immediate = (static_cast<u64>(*high) << 32) | static_cast<u64>(*low);
for (u32 entry = 0; entry < num_entries; ++entry) {
s32 value = (immediate >> (entry * 8)) & 0b111111;
if (value >= 32) {
value -= 64;
}
ptp.push_back(Immediate(value));
}
return ptp;
}
} // namespace VideoCommon::Shader

169
src/video_core/shader/decode/video.cpp
View file

@ -1,169 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/assert.h"
#include "common/common_types.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
using std::move;
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
using Tegra::Shader::Pred;
using Tegra::Shader::VideoType;
using Tegra::Shader::VmadShr;
using Tegra::Shader::VmnmxOperation;
using Tegra::Shader::VmnmxType;
u32 ShaderIR::DecodeVideo(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
const auto opcode = OpCode::Decode(instr);
if (opcode->get().GetId() == OpCode::Id::VMNMX) {
DecodeVMNMX(bb, instr);
return pc;
}
const Node op_a =
GetVideoOperand(GetRegister(instr.gpr8), instr.video.is_byte_chunk_a, instr.video.signed_a,
instr.video.type_a, instr.video.byte_height_a);
const Node op_b = [this, instr] {
if (instr.video.use_register_b) {
return GetVideoOperand(GetRegister(instr.gpr20), instr.video.is_byte_chunk_b,
instr.video.signed_b, instr.video.type_b,
instr.video.byte_height_b);
}
if (instr.video.signed_b) {
const auto imm = static_cast<s16>(instr.alu.GetImm20_16());
return Immediate(static_cast<u32>(imm));
} else {
return Immediate(instr.alu.GetImm20_16());
}
}();
switch (opcode->get().GetId()) {
case OpCode::Id::VMAD: {
const bool result_signed = instr.video.signed_a == 1 || instr.video.signed_b == 1;
const Node op_c = GetRegister(instr.gpr39);
Node value = SignedOperation(OperationCode::IMul, result_signed, NO_PRECISE, op_a, op_b);
value = SignedOperation(OperationCode::IAdd, result_signed, NO_PRECISE, value, op_c);
if (instr.vmad.shr == VmadShr::Shr7 || instr.vmad.shr == VmadShr::Shr15) {
const Node shift = Immediate(instr.vmad.shr == VmadShr::Shr7 ? 7 : 15);
value =
SignedOperation(OperationCode::IArithmeticShiftRight, result_signed, value, shift);
}
SetInternalFlagsFromInteger(bb, value, instr.generates_cc);
SetRegister(bb, instr.gpr0, value);
break;
}
case OpCode::Id::VSETP: {
// We can't use the constant predicate as destination.
ASSERT(instr.vsetp.pred3 != static_cast<u64>(Pred::UnusedIndex));
const bool sign = instr.video.signed_a == 1 || instr.video.signed_b == 1;
const Node first_pred = GetPredicateComparisonInteger(instr.vsetp.cond, sign, op_a, op_b);
const Node second_pred = GetPredicate(instr.vsetp.pred39, false);
const OperationCode combiner = GetPredicateCombiner(instr.vsetp.op);
// Set the primary predicate to the result of Predicate OP SecondPredicate
SetPredicate(bb, instr.vsetp.pred3, Operation(combiner, first_pred, second_pred));
if (instr.vsetp.pred0 != static_cast<u64>(Pred::UnusedIndex)) {
// Set the secondary predicate to the result of !Predicate OP SecondPredicate,
// if enabled
const Node negate_pred = Operation(OperationCode::LogicalNegate, first_pred);
SetPredicate(bb, instr.vsetp.pred0, Operation(combiner, negate_pred, second_pred));
}
break;
}
default:
UNIMPLEMENTED_MSG("Unhandled video instruction: {}", opcode->get().GetName());
}
return pc;
}
Node ShaderIR::GetVideoOperand(Node op, bool is_chunk, bool is_signed, VideoType type,
u64 byte_height) {
if (!is_chunk) {
return BitfieldExtract(op, static_cast<u32>(byte_height * 8), 8);
}
switch (type) {
case VideoType::Size16_Low:
return BitfieldExtract(op, 0, 16);
case VideoType::Size16_High:
return BitfieldExtract(op, 16, 16);
case VideoType::Size32:
// TODO(Rodrigo): From my hardware tests it becomes a bit "mad" when this type is used
// (1 * 1 + 0 == 0x5b800000). Until a better explanation is found: abort.
UNIMPLEMENTED();
return Immediate(0);
case VideoType::Invalid:
UNREACHABLE_MSG("Invalid instruction encoding");
return Immediate(0);
default:
UNREACHABLE();
return Immediate(0);
}
}
void ShaderIR::DecodeVMNMX(NodeBlock& bb, Tegra::Shader::Instruction instr) {
UNIMPLEMENTED_IF(!instr.vmnmx.is_op_b_register);
UNIMPLEMENTED_IF(instr.vmnmx.SourceFormatA() != VmnmxType::Bits32);
UNIMPLEMENTED_IF(instr.vmnmx.SourceFormatB() != VmnmxType::Bits32);
UNIMPLEMENTED_IF(instr.vmnmx.is_src_a_signed != instr.vmnmx.is_src_b_signed);
UNIMPLEMENTED_IF(instr.vmnmx.sat);
UNIMPLEMENTED_IF(instr.generates_cc);
Node op_a = GetRegister(instr.gpr8);
Node op_b = GetRegister(instr.gpr20);
Node op_c = GetRegister(instr.gpr39);
const bool is_oper1_signed = instr.vmnmx.is_src_a_signed; // Stubbed
const bool is_oper2_signed = instr.vmnmx.is_dest_signed;
const auto operation_a = instr.vmnmx.mx ? OperationCode::IMax : OperationCode::IMin;
Node value = SignedOperation(operation_a, is_oper1_signed, move(op_a), move(op_b));
switch (instr.vmnmx.operation) {
case VmnmxOperation::Mrg_16H:
value = BitfieldInsert(move(op_c), move(value), 16, 16);
break;
case VmnmxOperation::Mrg_16L:
value = BitfieldInsert(move(op_c), move(value), 0, 16);
break;
case VmnmxOperation::Mrg_8B0:
value = BitfieldInsert(move(op_c), move(value), 0, 8);
break;
case VmnmxOperation::Mrg_8B2:
value = BitfieldInsert(move(op_c), move(value), 16, 8);
break;
case VmnmxOperation::Acc:
value = Operation(OperationCode::IAdd, move(value), move(op_c));
break;
case VmnmxOperation::Min:
value = SignedOperation(OperationCode::IMin, is_oper2_signed, move(value), move(op_c));
break;
case VmnmxOperation::Max:
value = SignedOperation(OperationCode::IMax, is_oper2_signed, move(value), move(op_c));
break;
case VmnmxOperation::Nop:
break;
default:
UNREACHABLE();
break;
}
SetRegister(bb, instr.gpr0, move(value));
}
} // namespace VideoCommon::Shader

117
src/video_core/shader/decode/warp.cpp
View file

@ -1,117 +0,0 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/assert.h"
#include "common/common_types.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
using Tegra::Shader::Pred;
using Tegra::Shader::ShuffleOperation;
using Tegra::Shader::VoteOperation;
namespace {
OperationCode GetOperationCode(VoteOperation vote_op) {
switch (vote_op) {
case VoteOperation::All:
return OperationCode::VoteAll;
case VoteOperation::Any:
return OperationCode::VoteAny;
case VoteOperation::Eq:
return OperationCode::VoteEqual;
default:
UNREACHABLE_MSG("Invalid vote operation={}", vote_op);
return OperationCode::VoteAll;
}
}
} // Anonymous namespace
u32 ShaderIR::DecodeWarp(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
const auto opcode = OpCode::Decode(instr);
// Signal the backend that this shader uses warp instructions.
uses_warps = true;
switch (opcode->get().GetId()) {
case OpCode::Id::VOTE: {
const Node value = GetPredicate(instr.vote.value, instr.vote.negate_value != 0);
const Node active = Operation(OperationCode::BallotThread, value);
const Node vote = Operation(GetOperationCode(instr.vote.operation), value);
SetRegister(bb, instr.gpr0, active);
SetPredicate(bb, instr.vote.dest_pred, vote);
break;
}
case OpCode::Id::SHFL: {
Node mask = instr.shfl.is_mask_imm ? Immediate(static_cast<u32>(instr.shfl.mask_imm))
: GetRegister(instr.gpr39);
Node index = instr.shfl.is_index_imm ? Immediate(static_cast<u32>(instr.shfl.index_imm))
: GetRegister(instr.gpr20);
Node thread_id = Operation(OperationCode::ThreadId);
Node clamp = Operation(OperationCode::IBitwiseAnd, mask, Immediate(0x1FU));
Node seg_mask = BitfieldExtract(mask, 8, 16);
Node neg_seg_mask = Operation(OperationCode::IBitwiseNot, seg_mask);
Node min_thread_id = Operation(OperationCode::IBitwiseAnd, thread_id, seg_mask);
Node max_thread_id = Operation(OperationCode::IBitwiseOr, min_thread_id,
Operation(OperationCode::IBitwiseAnd, clamp, neg_seg_mask));
Node src_thread_id = [instr, index, neg_seg_mask, min_thread_id, thread_id] {
switch (instr.shfl.operation) {
case ShuffleOperation::Idx:
return Operation(OperationCode::IBitwiseOr,
Operation(OperationCode::IBitwiseAnd, index, neg_seg_mask),
min_thread_id);
case ShuffleOperation::Down:
return Operation(OperationCode::IAdd, thread_id, index);
case ShuffleOperation::Up:
return Operation(OperationCode::IAdd, thread_id,
Operation(OperationCode::INegate, index));
case ShuffleOperation::Bfly:
return Operation(OperationCode::IBitwiseXor, thread_id, index);
}
UNREACHABLE();
return Immediate(0U);
}();
Node in_bounds = [instr, src_thread_id, min_thread_id, max_thread_id] {
if (instr.shfl.operation == ShuffleOperation::Up) {
return Operation(OperationCode::LogicalIGreaterEqual, src_thread_id, min_thread_id);
} else {
return Operation(OperationCode::LogicalILessEqual, src_thread_id, max_thread_id);
}
}();
SetPredicate(bb, instr.shfl.pred48, in_bounds);
SetRegister(
bb, instr.gpr0,
Operation(OperationCode::ShuffleIndexed, GetRegister(instr.gpr8), src_thread_id));
break;
}
case OpCode::Id::FSWZADD: {
UNIMPLEMENTED_IF(instr.fswzadd.ndv);
Node op_a = GetRegister(instr.gpr8);
Node op_b = GetRegister(instr.gpr20);
Node mask = Immediate(static_cast<u32>(instr.fswzadd.swizzle));
SetRegister(bb, instr.gpr0, Operation(OperationCode::FSwizzleAdd, op_a, op_b, mask));
break;
}
default:
UNIMPLEMENTED_MSG("Unhandled warp instruction: {}", opcode->get().GetName());
break;
}
return pc;
}
} // namespace VideoCommon::Shader

156
src/video_core/shader/decode/xmad.cpp
View file

@ -1,156 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/assert.h"
#include "common/common_types.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
using Tegra::Shader::PredCondition;
u32 ShaderIR::DecodeXmad(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
const auto opcode = OpCode::Decode(instr);
UNIMPLEMENTED_IF(instr.xmad.sign_a);
UNIMPLEMENTED_IF(instr.xmad.sign_b);
UNIMPLEMENTED_IF_MSG(instr.generates_cc,
"Condition codes generation in XMAD is not implemented");
Node op_a = GetRegister(instr.gpr8);
// TODO(bunnei): Needs to be fixed once op_a or op_b is signed
UNIMPLEMENTED_IF(instr.xmad.sign_a != instr.xmad.sign_b);
const bool is_signed_a = instr.xmad.sign_a == 1;
const bool is_signed_b = instr.xmad.sign_b == 1;
const bool is_signed_c = is_signed_a;
auto [is_merge, is_psl, is_high_b, mode, op_b_binding,
op_c] = [&]() -> std::tuple<bool, bool, bool, Tegra::Shader::XmadMode, Node, Node> {
switch (opcode->get().GetId()) {
case OpCode::Id::XMAD_CR:
return {instr.xmad.merge_56,
instr.xmad.product_shift_left_second,
instr.xmad.high_b,
instr.xmad.mode_cbf,
GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset()),
GetRegister(instr.gpr39)};
case OpCode::Id::XMAD_RR:
return {instr.xmad.merge_37, instr.xmad.product_shift_left, instr.xmad.high_b_rr,
instr.xmad.mode, GetRegister(instr.gpr20), GetRegister(instr.gpr39)};
case OpCode::Id::XMAD_RC:
return {false,
false,
instr.xmad.high_b,
instr.xmad.mode_cbf,
GetRegister(instr.gpr39),
GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset())};
case OpCode::Id::XMAD_IMM:
return {instr.xmad.merge_37,
instr.xmad.product_shift_left,
false,
instr.xmad.mode,
Immediate(static_cast<u32>(instr.xmad.imm20_16)),
GetRegister(instr.gpr39)};
default:
UNIMPLEMENTED_MSG("Unhandled XMAD instruction: {}", opcode->get().GetName());
return {false, false, false, Tegra::Shader::XmadMode::None, Immediate(0), Immediate(0)};
}
}();
op_a = SignedOperation(OperationCode::IBitfieldExtract, is_signed_a, std::move(op_a),
instr.xmad.high_a ? Immediate(16) : Immediate(0), Immediate(16));
const Node original_b = op_b_binding;
const Node op_b =
SignedOperation(OperationCode::IBitfieldExtract, is_signed_b, std::move(op_b_binding),
is_high_b ? Immediate(16) : Immediate(0), Immediate(16));
// we already check sign_a and sign_b is difference or not before so just use one in here.
Node product = SignedOperation(OperationCode::IMul, is_signed_a, op_a, op_b);
if (is_psl) {
product =
SignedOperation(OperationCode::ILogicalShiftLeft, is_signed_a, product, Immediate(16));
}
SetTemporary(bb, 0, product);
product = GetTemporary(0);
Node original_c = op_c;
const Tegra::Shader::XmadMode set_mode = mode; // Workaround to clang compile error
op_c = [&] {
switch (set_mode) {
case Tegra::Shader::XmadMode::None:
return original_c;
case Tegra::Shader::XmadMode::CLo:
return BitfieldExtract(std::move(original_c), 0, 16);
case Tegra::Shader::XmadMode::CHi:
return BitfieldExtract(std::move(original_c), 16, 16);
case Tegra::Shader::XmadMode::CBcc: {
Node shifted_b = SignedOperation(OperationCode::ILogicalShiftLeft, is_signed_b,
original_b, Immediate(16));
return SignedOperation(OperationCode::IAdd, is_signed_c, std::move(original_c),
std::move(shifted_b));
}
case Tegra::Shader::XmadMode::CSfu: {
const Node comp_a =
GetPredicateComparisonInteger(PredCondition::EQ, is_signed_a, op_a, Immediate(0));
const Node comp_b =
GetPredicateComparisonInteger(PredCondition::EQ, is_signed_b, op_b, Immediate(0));
const Node comp = Operation(OperationCode::LogicalOr, comp_a, comp_b);
const Node comp_minus_a = GetPredicateComparisonInteger(
PredCondition::NE, is_signed_a,
SignedOperation(OperationCode::IBitwiseAnd, is_signed_a, op_a,
Immediate(0x80000000)),
Immediate(0));
const Node comp_minus_b = GetPredicateComparisonInteger(
PredCondition::NE, is_signed_b,
SignedOperation(OperationCode::IBitwiseAnd, is_signed_b, op_b,
Immediate(0x80000000)),
Immediate(0));
Node new_c = Operation(
OperationCode::Select, comp_minus_a,
SignedOperation(OperationCode::IAdd, is_signed_c, original_c, Immediate(-65536)),
original_c);
new_c = Operation(
OperationCode::Select, comp_minus_b,
SignedOperation(OperationCode::IAdd, is_signed_c, new_c, Immediate(-65536)),
std::move(new_c));
return Operation(OperationCode::Select, comp, original_c, std::move(new_c));
}
default:
UNREACHABLE();
return Immediate(0);
}
}();
SetTemporary(bb, 1, op_c);
op_c = GetTemporary(1);
// TODO(Rodrigo): Use an appropiate sign for this operation
Node sum = SignedOperation(OperationCode::IAdd, is_signed_a, product, std::move(op_c));
SetTemporary(bb, 2, sum);
sum = GetTemporary(2);
if (is_merge) {
const Node a = SignedOperation(OperationCode::IBitfieldExtract, is_signed_a, std::move(sum),
Immediate(0), Immediate(16));
const Node b = SignedOperation(OperationCode::ILogicalShiftLeft, is_signed_b, original_b,
Immediate(16));
sum = SignedOperation(OperationCode::IBitwiseOr, is_signed_a, a, b);
}
SetInternalFlagsFromInteger(bb, sum, instr.generates_cc);
SetRegister(bb, instr.gpr0, std::move(sum));
return pc;
}
} // namespace VideoCommon::Shader

93
src/video_core/shader/expr.cpp
View file

@ -1,93 +0,0 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <memory>
#include <variant>
#include "video_core/shader/expr.h"
namespace VideoCommon::Shader {
namespace {
bool ExprIsBoolean(const Expr& expr) {
return std::holds_alternative<ExprBoolean>(*expr);
}
bool ExprBooleanGet(const Expr& expr) {
return std::get_if<ExprBoolean>(expr.get())->value;
}
} // Anonymous namespace
bool ExprAnd::operator==(const ExprAnd& b) const {
return (*operand1 == *b.operand1) && (*operand2 == *b.operand2);
}
bool ExprAnd::operator!=(const ExprAnd& b) const {
return !operator==(b);
}
bool ExprOr::operator==(const ExprOr& b) const {
return (*operand1 == *b.operand1) && (*operand2 == *b.operand2);
}
bool ExprOr::operator!=(const ExprOr& b) const {
return !operator==(b);
}
bool ExprNot::operator==(const ExprNot& b) const {
return *operand1 == *b.operand1;
}
bool ExprNot::operator!=(const ExprNot& b) const {
return !operator==(b);
}
Expr MakeExprNot(Expr first) {
if (std::holds_alternative<ExprNot>(*first)) {
return std::get_if<ExprNot>(first.get())->operand1;
}
return MakeExpr<ExprNot>(std::move(first));
}
Expr MakeExprAnd(Expr first, Expr second) {
if (ExprIsBoolean(first)) {
return ExprBooleanGet(first) ? second : first;
}
if (ExprIsBoolean(second)) {
return ExprBooleanGet(second) ? first : second;
}
return MakeExpr<ExprAnd>(std::move(first), std::move(second));
}
Expr MakeExprOr(Expr first, Expr second) {
if (ExprIsBoolean(first)) {
return ExprBooleanGet(first) ? first : second;
}
if (ExprIsBoolean(second)) {
return ExprBooleanGet(second) ? second : first;
}
return MakeExpr<ExprOr>(std::move(first), std::move(second));
}
bool ExprAreEqual(const Expr& first, const Expr& second) {
return (*first) == (*second);
}
bool ExprAreOpposite(const Expr& first, const Expr& second) {
if (std::holds_alternative<ExprNot>(*first)) {
return ExprAreEqual(std::get_if<ExprNot>(first.get())->operand1, second);
}
if (std::holds_alternative<ExprNot>(*second)) {
return ExprAreEqual(std::get_if<ExprNot>(second.get())->operand1, first);
}
return false;
}
bool ExprIsTrue(const Expr& first) {
if (ExprIsBoolean(first)) {
return ExprBooleanGet(first);
}
return false;
}
} // namespace VideoCommon::Shader

156
src/video_core/shader/expr.h
View file

@ -1,156 +0,0 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <memory>
#include <variant>
#include "video_core/engines/shader_bytecode.h"
namespace VideoCommon::Shader {
using Tegra::Shader::ConditionCode;
using Tegra::Shader::Pred;
class ExprAnd;
class ExprBoolean;
class ExprCondCode;
class ExprGprEqual;
class ExprNot;
class ExprOr;
class ExprPredicate;
class ExprVar;
using ExprData = std::variant<ExprVar, ExprCondCode, ExprPredicate, ExprNot, ExprOr, ExprAnd,
ExprBoolean, ExprGprEqual>;
using Expr = std::shared_ptr<ExprData>;
class ExprAnd final {
public:
explicit ExprAnd(Expr a, Expr b) : operand1{std::move(a)}, operand2{std::move(b)} {}
bool operator==(const ExprAnd& b) const;
bool operator!=(const ExprAnd& b) const;
Expr operand1;
Expr operand2;
};
class ExprOr final {
public:
explicit ExprOr(Expr a, Expr b) : operand1{std::move(a)}, operand2{std::move(b)} {}
bool operator==(const ExprOr& b) const;
bool operator!=(const ExprOr& b) const;
Expr operand1;
Expr operand2;
};
class ExprNot final {
public:
explicit ExprNot(Expr a) : operand1{std::move(a)} {}
bool operator==(const ExprNot& b) const;
bool operator!=(const ExprNot& b) const;
Expr operand1;
};
class ExprVar final {
public:
explicit ExprVar(u32 index) : var_index{index} {}
bool operator==(const ExprVar& b) const {
return var_index == b.var_index;
}
bool operator!=(const ExprVar& b) const {
return !operator==(b);
}
u32 var_index;
};
class ExprPredicate final {
public:
explicit ExprPredicate(u32 predicate_) : predicate{predicate_} {}
bool operator==(const ExprPredicate& b) const {
return predicate == b.predicate;
}
bool operator!=(const ExprPredicate& b) const {
return !operator==(b);
}
u32 predicate;
};
class ExprCondCode final {
public:
explicit ExprCondCode(ConditionCode condition_code) : cc{condition_code} {}
bool operator==(const ExprCondCode& b) const {
return cc == b.cc;
}
bool operator!=(const ExprCondCode& b) const {
return !operator==(b);
}
ConditionCode cc;
};
class ExprBoolean final {
public:
explicit ExprBoolean(bool val) : value{val} {}
bool operator==(const ExprBoolean& b) const {
return value == b.value;
}
bool operator!=(const ExprBoolean& b) const {
return !operator==(b);
}
bool value;
};
class ExprGprEqual final {
public:
explicit ExprGprEqual(u32 gpr_, u32 value_) : gpr{gpr_}, value{value_} {}
bool operator==(const ExprGprEqual& b) const {
return gpr == b.gpr && value == b.value;
}
bool operator!=(const ExprGprEqual& b) const {
return !operator==(b);
}
u32 gpr;
u32 value;
};
template <typename T, typename... Args>
Expr MakeExpr(Args&&... args) {
static_assert(std::is_convertible_v<T, ExprData>);
return std::make_shared<ExprData>(T(std::forward<Args>(args)...));
}
bool ExprAreEqual(const Expr& first, const Expr& second);
bool ExprAreOpposite(const Expr& first, const Expr& second);
Expr MakeExprNot(Expr first);
Expr MakeExprAnd(Expr first, Expr second);
Expr MakeExprOr(Expr first, Expr second);
bool ExprIsTrue(const Expr& first);
} // namespace VideoCommon::Shader

76
src/video_core/shader/memory_util.cpp
View file

@ -1,76 +0,0 @@
// Copyright 2020 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <algorithm>
#include <cstddef>
#include <boost/container_hash/hash.hpp>
#include "common/common_types.h"
#include "core/core.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/memory_manager.h"
#include "video_core/shader/memory_util.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
GPUVAddr GetShaderAddress(Tegra::Engines::Maxwell3D& maxwell3d,
Tegra::Engines::Maxwell3D::Regs::ShaderProgram program) {
const auto& shader_config{maxwell3d.regs.shader_config[static_cast<std::size_t>(program)]};
return maxwell3d.regs.code_address.CodeAddress() + shader_config.offset;
}
bool IsSchedInstruction(std::size_t offset, std::size_t main_offset) {
// Sched instructions appear once every 4 instructions.
constexpr std::size_t SchedPeriod = 4;
const std::size_t absolute_offset = offset - main_offset;
return (absolute_offset % SchedPeriod) == 0;
}
std::size_t CalculateProgramSize(const ProgramCode& program, bool is_compute) {
// This is the encoded version of BRA that jumps to itself. All Nvidia
// shaders end with one.
static constexpr u64 SELF_JUMPING_BRANCH = 0xE2400FFFFF07000FULL;
static constexpr u64 MASK = 0xFFFFFFFFFF7FFFFFULL;
const std::size_t start_offset = is_compute ? KERNEL_MAIN_OFFSET : STAGE_MAIN_OFFSET;
std::size_t offset = start_offset;
while (offset < program.size()) {
const u64 instruction = program[offset];
if (!IsSchedInstruction(offset, start_offset)) {
if ((instruction & MASK) == SELF_JUMPING_BRANCH) {
// End on Maxwell's "nop" instruction
break;
}
if (instruction == 0) {
break;
}
}
++offset;
}
// The last instruction is included in the program size
return std::min(offset + 1, program.size());
}
ProgramCode GetShaderCode(Tegra::MemoryManager& memory_manager, GPUVAddr gpu_addr,
const u8* host_ptr, bool is_compute) {
ProgramCode code(VideoCommon::Shader::MAX_PROGRAM_LENGTH);
ASSERT_OR_EXECUTE(host_ptr != nullptr, { return code; });
memory_manager.ReadBlockUnsafe(gpu_addr, code.data(), code.size() * sizeof(u64));
code.resize(CalculateProgramSize(code, is_compute));
return code;
}
u64 GetUniqueIdentifier(Tegra::Engines::ShaderType shader_type, bool is_a, const ProgramCode& code,
const ProgramCode& code_b) {
size_t unique_identifier = boost::hash_value(code);
if (is_a) {
// VertexA programs include two programs
boost::hash_combine(unique_identifier, boost::hash_value(code_b));
}
return static_cast<u64>(unique_identifier);
}
} // namespace VideoCommon::Shader

43
src/video_core/shader/memory_util.h
View file

@ -1,43 +0,0 @@
// Copyright 2020 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <cstddef>
#include <vector>
#include "common/common_types.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/engines/shader_type.h"
namespace Tegra {
class MemoryManager;
}
namespace VideoCommon::Shader {
using ProgramCode = std::vector<u64>;
constexpr u32 STAGE_MAIN_OFFSET = 10;
constexpr u32 KERNEL_MAIN_OFFSET = 0;
/// Gets the address for the specified shader stage program
GPUVAddr GetShaderAddress(Tegra::Engines::Maxwell3D& maxwell3d,
Tegra::Engines::Maxwell3D::Regs::ShaderProgram program);
/// Gets if the current instruction offset is a scheduler instruction
bool IsSchedInstruction(std::size_t offset, std::size_t main_offset);
/// Calculates the size of a program stream
std::size_t CalculateProgramSize(const ProgramCode& program, bool is_compute);
/// Gets the shader program code from memory for the specified address
ProgramCode GetShaderCode(Tegra::MemoryManager& memory_manager, GPUVAddr gpu_addr,
const u8* host_ptr, bool is_compute);
/// Hashes one (or two) program streams
u64 GetUniqueIdentifier(Tegra::Engines::ShaderType shader_type, bool is_a, const ProgramCode& code,
const ProgramCode& code_b = {});
} // namespace VideoCommon::Shader

701
src/video_core/shader/node.h
View file

@ -1,701 +0,0 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <array>
#include <cstddef>
#include <memory>
#include <optional>
#include <string>
#include <tuple>
#include <utility>
#include <variant>
#include <vector>
#include "common/common_types.h"
#include "video_core/engines/shader_bytecode.h"
namespace VideoCommon::Shader {
enum class OperationCode {
Assign, /// (float& dest, float src) -> void
Select, /// (MetaArithmetic, bool pred, float a, float b) -> float
FAdd, /// (MetaArithmetic, float a, float b) -> float
FMul, /// (MetaArithmetic, float a, float b) -> float
FDiv, /// (MetaArithmetic, float a, float b) -> float
FFma, /// (MetaArithmetic, float a, float b, float c) -> float
FNegate, /// (MetaArithmetic, float a) -> float
FAbsolute, /// (MetaArithmetic, float a) -> float
FClamp, /// (MetaArithmetic, float value, float min, float max) -> float
FCastHalf0, /// (MetaArithmetic, f16vec2 a) -> float
FCastHalf1, /// (MetaArithmetic, f16vec2 a) -> float
FMin, /// (MetaArithmetic, float a, float b) -> float
FMax, /// (MetaArithmetic, float a, float b) -> float
FCos, /// (MetaArithmetic, float a) -> float
FSin, /// (MetaArithmetic, float a) -> float
FExp2, /// (MetaArithmetic, float a) -> float
FLog2, /// (MetaArithmetic, float a) -> float
FInverseSqrt, /// (MetaArithmetic, float a) -> float
FSqrt, /// (MetaArithmetic, float a) -> float
FRoundEven, /// (MetaArithmetic, float a) -> float
FFloor, /// (MetaArithmetic, float a) -> float
FCeil, /// (MetaArithmetic, float a) -> float
FTrunc, /// (MetaArithmetic, float a) -> float
FCastInteger, /// (MetaArithmetic, int a) -> float
FCastUInteger, /// (MetaArithmetic, uint a) -> float
FSwizzleAdd, /// (float a, float b, uint mask) -> float
IAdd, /// (MetaArithmetic, int a, int b) -> int
IMul, /// (MetaArithmetic, int a, int b) -> int
IDiv, /// (MetaArithmetic, int a, int b) -> int
INegate, /// (MetaArithmetic, int a) -> int
IAbsolute, /// (MetaArithmetic, int a) -> int
IMin, /// (MetaArithmetic, int a, int b) -> int
IMax, /// (MetaArithmetic, int a, int b) -> int
ICastFloat, /// (MetaArithmetic, float a) -> int
ICastUnsigned, /// (MetaArithmetic, uint a) -> int
ILogicalShiftLeft, /// (MetaArithmetic, int a, uint b) -> int
ILogicalShiftRight, /// (MetaArithmetic, int a, uint b) -> int
IArithmeticShiftRight, /// (MetaArithmetic, int a, uint b) -> int
IBitwiseAnd, /// (MetaArithmetic, int a, int b) -> int
IBitwiseOr, /// (MetaArithmetic, int a, int b) -> int
IBitwiseXor, /// (MetaArithmetic, int a, int b) -> int
IBitwiseNot, /// (MetaArithmetic, int a) -> int
IBitfieldInsert, /// (MetaArithmetic, int base, int insert, int offset, int bits) -> int
IBitfieldExtract, /// (MetaArithmetic, int value, int offset, int offset) -> int
IBitCount, /// (MetaArithmetic, int) -> int
IBitMSB, /// (MetaArithmetic, int) -> int
UAdd, /// (MetaArithmetic, uint a, uint b) -> uint
UMul, /// (MetaArithmetic, uint a, uint b) -> uint
UDiv, /// (MetaArithmetic, uint a, uint b) -> uint
UMin, /// (MetaArithmetic, uint a, uint b) -> uint
UMax, /// (MetaArithmetic, uint a, uint b) -> uint
UCastFloat, /// (MetaArithmetic, float a) -> uint
UCastSigned, /// (MetaArithmetic, int a) -> uint
ULogicalShiftLeft, /// (MetaArithmetic, uint a, uint b) -> uint
ULogicalShiftRight, /// (MetaArithmetic, uint a, uint b) -> uint
UArithmeticShiftRight, /// (MetaArithmetic, uint a, uint b) -> uint
UBitwiseAnd, /// (MetaArithmetic, uint a, uint b) -> uint
UBitwiseOr, /// (MetaArithmetic, uint a, uint b) -> uint
UBitwiseXor, /// (MetaArithmetic, uint a, uint b) -> uint
UBitwiseNot, /// (MetaArithmetic, uint a) -> uint
UBitfieldInsert, /// (MetaArithmetic, uint base, uint insert, int offset, int bits) -> uint
UBitfieldExtract, /// (MetaArithmetic, uint value, int offset, int offset) -> uint
UBitCount, /// (MetaArithmetic, uint) -> uint
UBitMSB, /// (MetaArithmetic, uint) -> uint
HAdd, /// (MetaArithmetic, f16vec2 a, f16vec2 b) -> f16vec2
HMul, /// (MetaArithmetic, f16vec2 a, f16vec2 b) -> f16vec2
HFma, /// (MetaArithmetic, f16vec2 a, f16vec2 b, f16vec2 c) -> f16vec2
HAbsolute, /// (f16vec2 a) -> f16vec2
HNegate, /// (f16vec2 a, bool first, bool second) -> f16vec2
HClamp, /// (f16vec2 src, float min, float max) -> f16vec2
HCastFloat, /// (MetaArithmetic, float a) -> f16vec2
HUnpack, /// (Tegra::Shader::HalfType, T value) -> f16vec2
HMergeF32, /// (f16vec2 src) -> float
HMergeH0, /// (f16vec2 dest, f16vec2 src) -> f16vec2
HMergeH1, /// (f16vec2 dest, f16vec2 src) -> f16vec2
HPack2, /// (float a, float b) -> f16vec2
LogicalAssign, /// (bool& dst, bool src) -> void
LogicalAnd, /// (bool a, bool b) -> bool
LogicalOr, /// (bool a, bool b) -> bool
LogicalXor, /// (bool a, bool b) -> bool
LogicalNegate, /// (bool a) -> bool
LogicalPick2, /// (bool2 pair, uint index) -> bool
LogicalAnd2, /// (bool2 a) -> bool
LogicalFOrdLessThan, /// (float a, float b) -> bool
LogicalFOrdEqual, /// (float a, float b) -> bool
LogicalFOrdLessEqual, /// (float a, float b) -> bool
LogicalFOrdGreaterThan, /// (float a, float b) -> bool
LogicalFOrdNotEqual, /// (float a, float b) -> bool
LogicalFOrdGreaterEqual, /// (float a, float b) -> bool
LogicalFOrdered, /// (float a, float b) -> bool
LogicalFUnordered, /// (float a, float b) -> bool
LogicalFUnordLessThan, /// (float a, float b) -> bool
LogicalFUnordEqual, /// (float a, float b) -> bool
LogicalFUnordLessEqual, /// (float a, float b) -> bool
LogicalFUnordGreaterThan, /// (float a, float b) -> bool
LogicalFUnordNotEqual, /// (float a, float b) -> bool
LogicalFUnordGreaterEqual, /// (float a, float b) -> bool
LogicalILessThan, /// (int a, int b) -> bool
LogicalIEqual, /// (int a, int b) -> bool
LogicalILessEqual, /// (int a, int b) -> bool
LogicalIGreaterThan, /// (int a, int b) -> bool
LogicalINotEqual, /// (int a, int b) -> bool
LogicalIGreaterEqual, /// (int a, int b) -> bool
LogicalULessThan, /// (uint a, uint b) -> bool
LogicalUEqual, /// (uint a, uint b) -> bool
LogicalULessEqual, /// (uint a, uint b) -> bool
LogicalUGreaterThan, /// (uint a, uint b) -> bool
LogicalUNotEqual, /// (uint a, uint b) -> bool
LogicalUGreaterEqual, /// (uint a, uint b) -> bool
LogicalAddCarry, /// (uint a, uint b) -> bool
Logical2HLessThan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2
Logical2HEqual, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2
Logical2HLessEqual, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2
Logical2HGreaterThan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2
Logical2HNotEqual, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2
Logical2HGreaterEqual, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2
Logical2HLessThanWithNan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2
Logical2HEqualWithNan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2
Logical2HLessEqualWithNan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2
Logical2HGreaterThanWithNan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2
Logical2HNotEqualWithNan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2
Logical2HGreaterEqualWithNan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2
Texture, /// (MetaTexture, float[N] coords) -> float4
TextureLod, /// (MetaTexture, float[N] coords) -> float4
TextureGather, /// (MetaTexture, float[N] coords) -> float4
TextureQueryDimensions, /// (MetaTexture, float a) -> float4
TextureQueryLod, /// (MetaTexture, float[N] coords) -> float4
TexelFetch, /// (MetaTexture, int[N], int) -> float4
TextureGradient, /// (MetaTexture, float[N] coords, float[N*2] derivates) -> float4
ImageLoad, /// (MetaImage, int[N] coords) -> void
ImageStore, /// (MetaImage, int[N] coords) -> void
AtomicImageAdd, /// (MetaImage, int[N] coords) -> void
AtomicImageAnd, /// (MetaImage, int[N] coords) -> void
AtomicImageOr, /// (MetaImage, int[N] coords) -> void
AtomicImageXor, /// (MetaImage, int[N] coords) -> void
AtomicImageExchange, /// (MetaImage, int[N] coords) -> void
AtomicUExchange, /// (memory, uint) -> uint
AtomicUAdd, /// (memory, uint) -> uint
AtomicUMin, /// (memory, uint) -> uint
AtomicUMax, /// (memory, uint) -> uint
AtomicUAnd, /// (memory, uint) -> uint
AtomicUOr, /// (memory, uint) -> uint
AtomicUXor, /// (memory, uint) -> uint
AtomicIExchange, /// (memory, int) -> int
AtomicIAdd, /// (memory, int) -> int
AtomicIMin, /// (memory, int) -> int
AtomicIMax, /// (memory, int) -> int
AtomicIAnd, /// (memory, int) -> int
AtomicIOr, /// (memory, int) -> int
AtomicIXor, /// (memory, int) -> int
ReduceUAdd, /// (memory, uint) -> void
ReduceUMin, /// (memory, uint) -> void
ReduceUMax, /// (memory, uint) -> void
ReduceUAnd, /// (memory, uint) -> void
ReduceUOr, /// (memory, uint) -> void
ReduceUXor, /// (memory, uint) -> void
ReduceIAdd, /// (memory, int) -> void
ReduceIMin, /// (memory, int) -> void
ReduceIMax, /// (memory, int) -> void
ReduceIAnd, /// (memory, int) -> void
ReduceIOr, /// (memory, int) -> void
ReduceIXor, /// (memory, int) -> void
Branch, /// (uint branch_target) -> void
BranchIndirect, /// (uint branch_target) -> void
PushFlowStack, /// (uint branch_target) -> void
PopFlowStack, /// () -> void
Exit, /// () -> void
Discard, /// () -> void
EmitVertex, /// () -> void
EndPrimitive, /// () -> void
InvocationId, /// () -> int
YNegate, /// () -> float
LocalInvocationIdX, /// () -> uint
LocalInvocationIdY, /// () -> uint
LocalInvocationIdZ, /// () -> uint
WorkGroupIdX, /// () -> uint
WorkGroupIdY, /// () -> uint
WorkGroupIdZ, /// () -> uint
BallotThread, /// (bool) -> uint
VoteAll, /// (bool) -> bool
VoteAny, /// (bool) -> bool
VoteEqual, /// (bool) -> bool
ThreadId, /// () -> uint
ThreadEqMask, /// () -> uint
ThreadGeMask, /// () -> uint
ThreadGtMask, /// () -> uint
ThreadLeMask, /// () -> uint
ThreadLtMask, /// () -> uint
ShuffleIndexed, /// (uint value, uint index) -> uint
Barrier, /// () -> void
MemoryBarrierGroup, /// () -> void
MemoryBarrierGlobal, /// () -> void
Amount,
};
enum class InternalFlag {
Zero = 0,
Sign = 1,
Carry = 2,
Overflow = 3,
Amount = 4,
};
enum class MetaStackClass {
Ssy,
Pbk,
};
class OperationNode;
class ConditionalNode;
class GprNode;
class CustomVarNode;
class ImmediateNode;
class InternalFlagNode;
class PredicateNode;
class AbufNode;
class CbufNode;
class LmemNode;
class PatchNode;
class SmemNode;
class GmemNode;
class CommentNode;
using NodeData = std::variant<OperationNode, ConditionalNode, GprNode, CustomVarNode, ImmediateNode,
InternalFlagNode, PredicateNode, AbufNode, PatchNode, CbufNode,
LmemNode, SmemNode, GmemNode, CommentNode>;
using Node = std::shared_ptr<NodeData>;
using Node4 = std::array<Node, 4>;
using NodeBlock = std::vector<Node>;
struct ArraySamplerNode;
struct BindlessSamplerNode;
struct SeparateSamplerNode;
using TrackSamplerData = std::variant<BindlessSamplerNode, SeparateSamplerNode, ArraySamplerNode>;
using TrackSampler = std::shared_ptr<TrackSamplerData>;
struct SamplerEntry {
/// Bound samplers constructor
explicit SamplerEntry(u32 index_, u32 offset_, Tegra::Shader::TextureType type_, bool is_array_,
bool is_shadow_, bool is_buffer_, bool is_indexed_)
: index{index_}, offset{offset_}, type{type_}, is_array{is_array_}, is_shadow{is_shadow_},
is_buffer{is_buffer_}, is_indexed{is_indexed_} {}
/// Separate sampler constructor
explicit SamplerEntry(u32 index_, std::pair<u32, u32> offsets, std::pair<u32, u32> buffers,
Tegra::Shader::TextureType type_, bool is_array_, bool is_shadow_,
bool is_buffer_)
: index{index_}, offset{offsets.first}, secondary_offset{offsets.second},
buffer{buffers.first}, secondary_buffer{buffers.second}, type{type_}, is_array{is_array_},
is_shadow{is_shadow_}, is_buffer{is_buffer_}, is_separated{true} {}
/// Bindless samplers constructor
explicit SamplerEntry(u32 index_, u32 offset_, u32 buffer_, Tegra::Shader::TextureType type_,
bool is_array_, bool is_shadow_, bool is_buffer_, bool is_indexed_)
: index{index_}, offset{offset_}, buffer{buffer_}, type{type_}, is_array{is_array_},
is_shadow{is_shadow_}, is_buffer{is_buffer_}, is_bindless{true}, is_indexed{is_indexed_} {
}
u32 index = 0; ///< Emulated index given for the this sampler.
u32 offset = 0; ///< Offset in the const buffer from where the sampler is being read.
u32 secondary_offset = 0; ///< Secondary offset in the const buffer.
u32 buffer = 0; ///< Buffer where the bindless sampler is read.
u32 secondary_buffer = 0; ///< Secondary buffer where the bindless sampler is read.
u32 size = 1; ///< Size of the sampler.
Tegra::Shader::TextureType type{}; ///< The type used to sample this texture (Texture2D, etc)
bool is_array = false; ///< Whether the texture is being sampled as an array texture or not.
bool is_shadow = false; ///< Whether the texture is being sampled as a depth texture or not.
bool is_buffer = false; ///< Whether the texture is a texture buffer without sampler.
bool is_bindless = false; ///< Whether this sampler belongs to a bindless texture or not.
bool is_indexed = false; ///< Whether this sampler is an indexed array of textures.
bool is_separated = false; ///< Whether the image and sampler is separated or not.
};
/// Represents a tracked bindless sampler into a direct const buffer
struct ArraySamplerNode {
u32 index;
u32 base_offset;
u32 bindless_var;
};
/// Represents a tracked separate sampler image pair that was folded statically
struct SeparateSamplerNode {
std::pair<u32, u32> indices;
std::pair<u32, u32> offsets;
};
/// Represents a tracked bindless sampler into a direct const buffer
struct BindlessSamplerNode {
u32 index;
u32 offset;
};
struct ImageEntry {
public:
/// Bound images constructor
explicit ImageEntry(u32 index_, u32 offset_, Tegra::Shader::ImageType type_)
: index{index_}, offset{offset_}, type{type_} {}
/// Bindless samplers constructor
explicit ImageEntry(u32 index_, u32 offset_, u32 buffer_, Tegra::Shader::ImageType type_)
: index{index_}, offset{offset_}, buffer{buffer_}, type{type_}, is_bindless{true} {}
void MarkWrite() {
is_written = true;
}
void MarkRead() {
is_read = true;
}
void MarkAtomic() {
MarkWrite();
MarkRead();
is_atomic = true;
}
u32 index = 0;
u32 offset = 0;
u32 buffer = 0;
Tegra::Shader::ImageType type{};
bool is_bindless = false;
bool is_written = false;
bool is_read = false;
bool is_atomic = false;
};
struct GlobalMemoryBase {
u32 cbuf_index = 0;
u32 cbuf_offset = 0;
[[nodiscard]] bool operator<(const GlobalMemoryBase& rhs) const {
return std::tie(cbuf_index, cbuf_offset) < std::tie(rhs.cbuf_index, rhs.cbuf_offset);
}
};
/// Parameters describing an arithmetic operation
struct MetaArithmetic {
bool precise{}; ///< Whether the operation can be constraint or not
};
/// Parameters describing a texture sampler
struct MetaTexture {
SamplerEntry sampler;
Node array;
Node depth_compare;
std::vector<Node> aoffi;
std::vector<Node> ptp;
std::vector<Node> derivates;
Node bias;
Node lod;
Node component;
u32 element{};
Node index;
};
struct MetaImage {
const ImageEntry& image;
std::vector<Node> values;
u32 element{};
};
/// Parameters that modify an operation but are not part of any particular operand
using Meta =
std::variant<MetaArithmetic, MetaTexture, MetaImage, MetaStackClass, Tegra::Shader::HalfType>;
class AmendNode {
public:
[[nodiscard]] std::optional<std::size_t> GetAmendIndex() const {
if (amend_index == amend_null_index) {
return std::nullopt;
}
return {amend_index};
}
void SetAmendIndex(std::size_t index) {
amend_index = index;
}
void ClearAmend() {
amend_index = amend_null_index;
}
private:
static constexpr std::size_t amend_null_index = 0xFFFFFFFFFFFFFFFFULL;
std::size_t amend_index{amend_null_index};
};
/// Holds any kind of operation that can be done in the IR
class OperationNode final : public AmendNode {
public:
explicit OperationNode(OperationCode code_) : OperationNode(code_, Meta{}) {}
explicit OperationNode(OperationCode code_, Meta meta_)
: OperationNode(code_, std::move(meta_), std::vector<Node>{}) {}
explicit OperationNode(OperationCode code_, std::vector<Node> operands_)
: OperationNode(code_, Meta{}, std::move(operands_)) {}
explicit OperationNode(OperationCode code_, Meta meta_, std::vector<Node> operands_)
: code{code_}, meta{std::move(meta_)}, operands{std::move(operands_)} {}
template <typename... Args>
explicit OperationNode(OperationCode code_, Meta meta_, Args&&... operands_)
: code{code_}, meta{std::move(meta_)}, operands{operands_...} {}
[[nodiscard]] OperationCode GetCode() const {
return code;
}
[[nodiscard]] const Meta& GetMeta() const {
return meta;
}
[[nodiscard]] std::size_t GetOperandsCount() const {
return operands.size();
}
[[nodiscard]] const Node& operator[](std::size_t operand_index) const {
return operands.at(operand_index);
}
private:
OperationCode code{};
Meta meta{};
std::vector<Node> operands;
};
/// Encloses inside any kind of node that returns a boolean conditionally-executed code
class ConditionalNode final : public AmendNode {
public:
explicit ConditionalNode(Node condition_, std::vector<Node>&& code_)
: condition{std::move(condition_)}, code{std::move(code_)} {}
[[nodiscard]] const Node& GetCondition() const {
return condition;
}
[[nodiscard]] const std::vector<Node>& GetCode() const {
return code;
}
private:
Node condition; ///< Condition to be satisfied
std::vector<Node> code; ///< Code to execute
};
/// A general purpose register
class GprNode final {
public:
explicit constexpr GprNode(Tegra::Shader::Register index_) : index{index_} {}
[[nodiscard]] constexpr u32 GetIndex() const {
return static_cast<u32>(index);
}
private:
Tegra::Shader::Register index{};
};
/// A custom variable
class CustomVarNode final {
public:
explicit constexpr CustomVarNode(u32 index_) : index{index_} {}
[[nodiscard]] constexpr u32 GetIndex() const {
return index;
}
private:
u32 index{};
};
/// A 32-bits value that represents an immediate value
class ImmediateNode final {
public:
explicit constexpr ImmediateNode(u32 value_) : value{value_} {}
[[nodiscard]] constexpr u32 GetValue() const {
return value;
}
private:
u32 value{};
};
/// One of Maxwell's internal flags
class InternalFlagNode final {
public:
explicit constexpr InternalFlagNode(InternalFlag flag_) : flag{flag_} {}
[[nodiscard]] constexpr InternalFlag GetFlag() const {
return flag;
}
private:
InternalFlag flag{};
};
/// A predicate register, it can be negated without additional nodes
class PredicateNode final {
public:
explicit constexpr PredicateNode(Tegra::Shader::Pred index_, bool negated_)
: index{index_}, negated{negated_} {}
[[nodiscard]] constexpr Tegra::Shader::Pred GetIndex() const {
return index;
}
[[nodiscard]] constexpr bool IsNegated() const {
return negated;
}
private:
Tegra::Shader::Pred index{};
bool negated{};
};
/// Attribute buffer memory (known as attributes or varyings in GLSL terms)
class AbufNode final {
public:
// Initialize for standard attributes (index is explicit).
explicit AbufNode(Tegra::Shader::Attribute::Index index_, u32 element_, Node buffer_ = {})
: buffer{std::move(buffer_)}, index{index_}, element{element_} {}
// Initialize for physical attributes (index is a variable value).
explicit AbufNode(Node physical_address_, Node buffer_ = {})
: physical_address{std::move(physical_address_)}, buffer{std::move(buffer_)} {}
[[nodiscard]] Tegra::Shader::Attribute::Index GetIndex() const {
return index;
}
[[nodiscard]] u32 GetElement() const {
return element;
}
[[nodiscard]] const Node& GetBuffer() const {
return buffer;
}
[[nodiscard]] bool IsPhysicalBuffer() const {
return static_cast<bool>(physical_address);
}
[[nodiscard]] const Node& GetPhysicalAddress() const {
return physical_address;
}
private:
Node physical_address;
Node buffer;
Tegra::Shader::Attribute::Index index{};
u32 element{};
};
/// Patch memory (used to communicate tessellation stages).
class PatchNode final {
public:
explicit constexpr PatchNode(u32 offset_) : offset{offset_} {}
[[nodiscard]] constexpr u32 GetOffset() const {
return offset;
}
private:
u32 offset{};
};
/// Constant buffer node, usually mapped to uniform buffers in GLSL
class CbufNode final {
public:
explicit CbufNode(u32 index_, Node offset_) : index{index_}, offset{std::move(offset_)} {}
[[nodiscard]] u32 GetIndex() const {
return index;
}
[[nodiscard]] const Node& GetOffset() const {
return offset;
}
private:
u32 index{};
Node offset;
};
/// Local memory node
class LmemNode final {
public:
explicit LmemNode(Node address_) : address{std::move(address_)} {}
[[nodiscard]] const Node& GetAddress() const {
return address;
}
private:
Node address;
};
/// Shared memory node
class SmemNode final {
public:
explicit SmemNode(Node address_) : address{std::move(address_)} {}
[[nodiscard]] const Node& GetAddress() const {
return address;
}
private:
Node address;
};
/// Global memory node
class GmemNode final {
public:
explicit GmemNode(Node real_address_, Node base_address_, const GlobalMemoryBase& descriptor_)
: real_address{std::move(real_address_)}, base_address{std::move(base_address_)},
descriptor{descriptor_} {}
[[nodiscard]] const Node& GetRealAddress() const {
return real_address;
}
[[nodiscard]] const Node& GetBaseAddress() const {
return base_address;
}
[[nodiscard]] const GlobalMemoryBase& GetDescriptor() const {
return descriptor;
}
private:
Node real_address;
Node base_address;
GlobalMemoryBase descriptor;
};
/// Commentary, can be dropped
class CommentNode final {
public:
explicit CommentNode(std::string text_) : text{std::move(text_)} {}
[[nodiscard]] const std::string& GetText() const {
return text;
}
private:
std::string text;
};
} // namespace VideoCommon::Shader

115
src/video_core/shader/node_helper.cpp
View file

@ -1,115 +0,0 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <cstring>
#include <vector>
#include "common/common_types.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
Node Conditional(Node condition, std::vector<Node> code) {
return MakeNode<ConditionalNode>(std::move(condition), std::move(code));
}
Node Comment(std::string text) {
return MakeNode<CommentNode>(std::move(text));
}
Node Immediate(u32 value) {
return MakeNode<ImmediateNode>(value);
}
Node Immediate(s32 value) {
return Immediate(static_cast<u32>(value));
}
Node Immediate(f32 value) {
u32 integral;
std::memcpy(&integral, &value, sizeof(u32));
return Immediate(integral);
}
OperationCode SignedToUnsignedCode(OperationCode operation_code, bool is_signed) {
if (is_signed) {
return operation_code;
}
switch (operation_code) {
case OperationCode::FCastInteger:
return OperationCode::FCastUInteger;
case OperationCode::IAdd:
return OperationCode::UAdd;
case OperationCode::IMul:
return OperationCode::UMul;
case OperationCode::IDiv:
return OperationCode::UDiv;
case OperationCode::IMin:
return OperationCode::UMin;
case OperationCode::IMax:
return OperationCode::UMax;
case OperationCode::ICastFloat:
return OperationCode::UCastFloat;
case OperationCode::ICastUnsigned:
return OperationCode::UCastSigned;
case OperationCode::ILogicalShiftLeft:
return OperationCode::ULogicalShiftLeft;
case OperationCode::ILogicalShiftRight:
return OperationCode::ULogicalShiftRight;
case OperationCode::IArithmeticShiftRight:
return OperationCode::UArithmeticShiftRight;
case OperationCode::IBitwiseAnd:
return OperationCode::UBitwiseAnd;
case OperationCode::IBitwiseOr:
return OperationCode::UBitwiseOr;
case OperationCode::IBitwiseXor:
return OperationCode::UBitwiseXor;
case OperationCode::IBitwiseNot:
return OperationCode::UBitwiseNot;
case OperationCode::IBitfieldExtract:
return OperationCode::UBitfieldExtract;
case OperationCode::IBitfieldInsert:
return OperationCode::UBitfieldInsert;
case OperationCode::IBitCount:
return OperationCode::UBitCount;
case OperationCode::LogicalILessThan:
return OperationCode::LogicalULessThan;
case OperationCode::LogicalIEqual:
return OperationCode::LogicalUEqual;
case OperationCode::LogicalILessEqual:
return OperationCode::LogicalULessEqual;
case OperationCode::LogicalIGreaterThan:
return OperationCode::LogicalUGreaterThan;
case OperationCode::LogicalINotEqual:
return OperationCode::LogicalUNotEqual;
case OperationCode::LogicalIGreaterEqual:
return OperationCode::LogicalUGreaterEqual;
case OperationCode::AtomicIExchange:
return OperationCode::AtomicUExchange;
case OperationCode::AtomicIAdd:
return OperationCode::AtomicUAdd;
case OperationCode::AtomicIMin:
return OperationCode::AtomicUMin;
case OperationCode::AtomicIMax:
return OperationCode::AtomicUMax;
case OperationCode::AtomicIAnd:
return OperationCode::AtomicUAnd;
case OperationCode::AtomicIOr:
return OperationCode::AtomicUOr;
case OperationCode::AtomicIXor:
return OperationCode::AtomicUXor;
case OperationCode::INegate:
UNREACHABLE_MSG("Can't negate an unsigned integer");
return {};
case OperationCode::IAbsolute:
UNREACHABLE_MSG("Can't apply absolute to an unsigned integer");
return {};
default:
UNREACHABLE_MSG("Unknown signed operation with code={}", operation_code);
return {};
}
}
} // namespace VideoCommon::Shader

71
src/video_core/shader/node_helper.h
View file

@ -1,71 +0,0 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <memory>
#include <string>
#include <tuple>
#include <type_traits>
#include <utility>
#include <vector>
#include "common/common_types.h"
#include "video_core/shader/node.h"
namespace VideoCommon::Shader {
/// This arithmetic operation cannot be constraint
inline constexpr MetaArithmetic PRECISE = {true};
/// This arithmetic operation can be optimized away
inline constexpr MetaArithmetic NO_PRECISE = {false};
/// Creates a conditional node
Node Conditional(Node condition, std::vector<Node> code);
/// Creates a commentary node
Node Comment(std::string text);
/// Creates an u32 immediate
Node Immediate(u32 value);
/// Creates a s32 immediate
Node Immediate(s32 value);
/// Creates a f32 immediate
Node Immediate(f32 value);
/// Converts an signed operation code to an unsigned operation code
OperationCode SignedToUnsignedCode(OperationCode operation_code, bool is_signed);
template <typename T, typename... Args>
Node MakeNode(Args&&... args) {
static_assert(std::is_convertible_v<T, NodeData>);
return std::make_shared<NodeData>(T(std::forward<Args>(args)...));
}
template <typename T, typename... Args>
TrackSampler MakeTrackSampler(Args&&... args) {
static_assert(std::is_convertible_v<T, TrackSamplerData>);
return std::make_shared<TrackSamplerData>(T{std::forward<Args>(args)...});
}
template <typename... Args>
Node Operation(OperationCode code, Args&&... args) {
if constexpr (sizeof...(args) == 0) {
return MakeNode<OperationNode>(code);
} else if constexpr (std::is_convertible_v<std::tuple_element_t<0, std::tuple<Args...>>,
Meta>) {
return MakeNode<OperationNode>(code, std::forward<Args>(args)...);
} else {
return MakeNode<OperationNode>(code, Meta{}, std::forward<Args>(args)...);
}
}
template <typename... Args>
Node SignedOperation(OperationCode code, bool is_signed, Args&&... args) {
return Operation(SignedToUnsignedCode(code, is_signed), std::forward<Args>(args)...);
}
} // namespace VideoCommon::Shader

181
src/video_core/shader/registry.cpp
View file

@ -1,181 +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/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 = dynamic_cast<Tegra::Engines::Maxwell3D&>(engine);
return {
.tfb_layouts = graphics.regs.tfb_layouts,
.tfb_varying_locs = graphics.regs.tfb_varying_locs,
.primitive_topology = graphics.regs.draw.topology,
.tessellation_primitive = graphics.regs.tess_mode.prim,
.tessellation_spacing = graphics.regs.tess_mode.spacing,
.tfb_enabled = graphics.regs.tfb_enabled != 0,
.tessellation_clockwise = graphics.regs.tess_mode.cw.Value() != 0,
};
}
ComputeInfo MakeComputeInfo(ShaderType shader_stage, ConstBufferEngineInterface& engine) {
if (shader_stage != ShaderType::Compute) {
return {};
}
auto& compute = dynamic_cast<Tegra::Engines::KeplerCompute&>(engine);
const auto& launch = compute.launch_description;
return {
.workgroup_size = {launch.block_dim_x, launch.block_dim_y, launch.block_dim_z},
.shared_memory_size_in_words = launch.shared_alloc,
.local_memory_size_in_words = launch.local_pos_alloc,
};
}
} // Anonymous namespace
Registry::Registry(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(ShaderType shader_stage, 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::ObtainSeparateSampler(
std::pair<u32, u32> buffers, std::pair<u32, u32> offsets) {
SeparateSamplerKey key;
key.buffers = buffers;
key.offsets = offsets;
const auto iter = separate_samplers.find(key);
if (iter != separate_samplers.end()) {
return iter->second;
}
if (!engine) {
return std::nullopt;
}
const u32 handle_1 = engine->AccessConstBuffer32(stage, key.buffers.first, key.offsets.first);
const u32 handle_2 = engine->AccessConstBuffer32(stage, key.buffers.second, key.offsets.second);
const SamplerDescriptor value = engine->AccessSampler(handle_1 | handle_2);
separate_samplers.emplace(key, value);
return value;
}
std::optional<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

172
src/video_core/shader/registry.h
View file

@ -1,172 +0,0 @@
// 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 {
struct SeparateSamplerKey {
std::pair<u32, u32> buffers;
std::pair<u32, u32> offsets;
};
} // namespace VideoCommon::Shader
namespace std {
template <>
struct hash<VideoCommon::Shader::SeparateSamplerKey> {
std::size_t operator()(const VideoCommon::Shader::SeparateSamplerKey& key) const noexcept {
return std::hash<u32>{}(key.buffers.first ^ key.buffers.second ^ key.offsets.first ^
key.offsets.second);
}
};
template <>
struct equal_to<VideoCommon::Shader::SeparateSamplerKey> {
bool operator()(const VideoCommon::Shader::SeparateSamplerKey& lhs,
const VideoCommon::Shader::SeparateSamplerKey& rhs) const noexcept {
return lhs.buffers == rhs.buffers && lhs.offsets == rhs.offsets;
}
};
} // namespace std
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 SeparateSamplerMap =
std::unordered_map<SeparateSamplerKey, 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> ObtainSeparateSampler(
std::pair<u32, u32> buffers, std::pair<u32, u32> offsets);
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;
SeparateSamplerMap separate_samplers;
BindlessSamplerMap bindless_samplers;
u32 bound_buffer;
GraphicsInfo graphics_info;
ComputeInfo compute_info;
};
} // namespace VideoCommon::Shader

464
src/video_core/shader/shader_ir.cpp
View file

@ -1,464 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <algorithm>
#include <array>
#include <cmath>
#include "common/assert.h"
#include "common/common_types.h"
#include "common/logging/log.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/node.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/registry.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
using Tegra::Shader::Attribute;
using Tegra::Shader::Instruction;
using Tegra::Shader::IpaMode;
using Tegra::Shader::Pred;
using Tegra::Shader::PredCondition;
using Tegra::Shader::PredOperation;
using Tegra::Shader::Register;
ShaderIR::ShaderIR(const ProgramCode& program_code_, u32 main_offset_, CompilerSettings settings_,
Registry& registry_)
: program_code{program_code_}, main_offset{main_offset_}, settings{settings_}, registry{
registry_} {
Decode();
PostDecode();
}
ShaderIR::~ShaderIR() = default;
Node ShaderIR::GetRegister(Register reg) {
if (reg != Register::ZeroIndex) {
used_registers.insert(static_cast<u32>(reg));
}
return MakeNode<GprNode>(reg);
}
Node ShaderIR::GetCustomVariable(u32 id) {
return MakeNode<CustomVarNode>(id);
}
Node ShaderIR::GetImmediate19(Instruction instr) {
return Immediate(instr.alu.GetImm20_19());
}
Node ShaderIR::GetImmediate32(Instruction instr) {
return Immediate(instr.alu.GetImm20_32());
}
Node ShaderIR::GetConstBuffer(u64 index_, u64 offset_) {
const auto index = static_cast<u32>(index_);
const auto offset = static_cast<u32>(offset_);
used_cbufs.try_emplace(index).first->second.MarkAsUsed(offset);
return MakeNode<CbufNode>(index, Immediate(offset));
}
Node ShaderIR::GetConstBufferIndirect(u64 index_, u64 offset_, Node node) {
const auto index = static_cast<u32>(index_);
const auto offset = static_cast<u32>(offset_);
used_cbufs.try_emplace(index).first->second.MarkAsUsedIndirect();
Node final_offset = [&] {
// Attempt to inline constant buffer without a variable offset. This is done to allow
// tracking LDC calls.
if (const auto gpr = std::get_if<GprNode>(&*node)) {
if (gpr->GetIndex() == Register::ZeroIndex) {
return Immediate(offset);
}
}
return Operation(OperationCode::UAdd, NO_PRECISE, std::move(node), Immediate(offset));
}();
return MakeNode<CbufNode>(index, std::move(final_offset));
}
Node ShaderIR::GetPredicate(u64 pred_, bool negated) {
const auto pred = static_cast<Pred>(pred_);
if (pred != Pred::UnusedIndex && pred != Pred::NeverExecute) {
used_predicates.insert(pred);
}
return MakeNode<PredicateNode>(pred, negated);
}
Node ShaderIR::GetPredicate(bool immediate) {
return GetPredicate(static_cast<u64>(immediate ? Pred::UnusedIndex : Pred::NeverExecute));
}
Node ShaderIR::GetInputAttribute(Attribute::Index index, u64 element, Node buffer) {
MarkAttributeUsage(index, element);
used_input_attributes.emplace(index);
return MakeNode<AbufNode>(index, static_cast<u32>(element), std::move(buffer));
}
Node ShaderIR::GetPhysicalInputAttribute(Tegra::Shader::Register physical_address, Node buffer) {
uses_physical_attributes = true;
return MakeNode<AbufNode>(GetRegister(physical_address), buffer);
}
Node ShaderIR::GetOutputAttribute(Attribute::Index index, u64 element, Node buffer) {
MarkAttributeUsage(index, element);
used_output_attributes.insert(index);
return MakeNode<AbufNode>(index, static_cast<u32>(element), std::move(buffer));
}
Node ShaderIR::GetInternalFlag(InternalFlag flag, bool negated) const {
Node node = MakeNode<InternalFlagNode>(flag);
if (negated) {
return Operation(OperationCode::LogicalNegate, std::move(node));
}
return node;
}
Node ShaderIR::GetLocalMemory(Node address) {
return MakeNode<LmemNode>(std::move(address));
}
Node ShaderIR::GetSharedMemory(Node address) {
return MakeNode<SmemNode>(std::move(address));
}
Node ShaderIR::GetTemporary(u32 id) {
return GetRegister(Register::ZeroIndex + 1 + id);
}
Node ShaderIR::GetOperandAbsNegFloat(Node value, bool absolute, bool negate) {
if (absolute) {
value = Operation(OperationCode::FAbsolute, NO_PRECISE, std::move(value));
}
if (negate) {
value = Operation(OperationCode::FNegate, NO_PRECISE, std::move(value));
}
return value;
}
Node ShaderIR::GetSaturatedFloat(Node value, bool saturate) {
if (!saturate) {
return value;
}
Node positive_zero = Immediate(std::copysignf(0, 1));
Node positive_one = Immediate(1.0f);
return Operation(OperationCode::FClamp, NO_PRECISE, std::move(value), std::move(positive_zero),
std::move(positive_one));
}
Node ShaderIR::ConvertIntegerSize(Node value, Register::Size size, bool is_signed) {
switch (size) {
case Register::Size::Byte:
value = SignedOperation(OperationCode::ILogicalShiftLeft, is_signed, NO_PRECISE,
std::move(value), Immediate(24));
value = SignedOperation(OperationCode::IArithmeticShiftRight, is_signed, NO_PRECISE,
std::move(value), Immediate(24));
return value;
case Register::Size::Short:
value = SignedOperation(OperationCode::ILogicalShiftLeft, is_signed, NO_PRECISE,
std::move(value), Immediate(16));
value = SignedOperation(OperationCode::IArithmeticShiftRight, is_signed, NO_PRECISE,
std::move(value), Immediate(16));
return value;
case Register::Size::Word:
// Default - do nothing
return value;
default:
UNREACHABLE_MSG("Unimplemented conversion size: {}", size);
return value;
}
}
Node ShaderIR::GetOperandAbsNegInteger(Node value, bool absolute, bool negate, bool is_signed) {
if (!is_signed) {
// Absolute or negate on an unsigned is pointless
return value;
}
if (absolute) {
value = Operation(OperationCode::IAbsolute, NO_PRECISE, std::move(value));
}
if (negate) {
value = Operation(OperationCode::INegate, NO_PRECISE, std::move(value));
}
return value;
}
Node ShaderIR::UnpackHalfImmediate(Instruction instr, bool has_negation) {
Node value = Immediate(instr.half_imm.PackImmediates());
if (!has_negation) {
return value;
}
Node first_negate = GetPredicate(instr.half_imm.first_negate != 0);
Node second_negate = GetPredicate(instr.half_imm.second_negate != 0);
return Operation(OperationCode::HNegate, NO_PRECISE, std::move(value), std::move(first_negate),
std::move(second_negate));
}
Node ShaderIR::UnpackHalfFloat(Node value, Tegra::Shader::HalfType type) {
return Operation(OperationCode::HUnpack, type, std::move(value));
}
Node ShaderIR::HalfMerge(Node dest, Node src, Tegra::Shader::HalfMerge merge) {
switch (merge) {
case Tegra::Shader::HalfMerge::H0_H1:
return src;
case Tegra::Shader::HalfMerge::F32:
return Operation(OperationCode::HMergeF32, std::move(src));
case Tegra::Shader::HalfMerge::Mrg_H0:
return Operation(OperationCode::HMergeH0, std::move(dest), std::move(src));
case Tegra::Shader::HalfMerge::Mrg_H1:
return Operation(OperationCode::HMergeH1, std::move(dest), std::move(src));
}
UNREACHABLE();
return src;
}
Node ShaderIR::GetOperandAbsNegHalf(Node value, bool absolute, bool negate) {
if (absolute) {
value = Operation(OperationCode::HAbsolute, NO_PRECISE, std::move(value));
}
if (negate) {
value = Operation(OperationCode::HNegate, NO_PRECISE, std::move(value), GetPredicate(true),
GetPredicate(true));
}
return value;
}
Node ShaderIR::GetSaturatedHalfFloat(Node value, bool saturate) {
if (!saturate) {
return value;
}
Node positive_zero = Immediate(std::copysignf(0, 1));
Node positive_one = Immediate(1.0f);
return Operation(OperationCode::HClamp, NO_PRECISE, std::move(value), std::move(positive_zero),
std::move(positive_one));
}
Node ShaderIR::GetPredicateComparisonFloat(PredCondition condition, Node op_a, Node op_b) {
if (condition == PredCondition::T) {
return GetPredicate(true);
} else if (condition == PredCondition::F) {
return GetPredicate(false);
}
static constexpr std::array comparison_table{
OperationCode(0),
OperationCode::LogicalFOrdLessThan, // LT
OperationCode::LogicalFOrdEqual, // EQ
OperationCode::LogicalFOrdLessEqual, // LE
OperationCode::LogicalFOrdGreaterThan, // GT
OperationCode::LogicalFOrdNotEqual, // NE
OperationCode::LogicalFOrdGreaterEqual, // GE
OperationCode::LogicalFOrdered, // NUM
OperationCode::LogicalFUnordered, // NAN
OperationCode::LogicalFUnordLessThan, // LTU
OperationCode::LogicalFUnordEqual, // EQU
OperationCode::LogicalFUnordLessEqual, // LEU
OperationCode::LogicalFUnordGreaterThan, // GTU
OperationCode::LogicalFUnordNotEqual, // NEU
OperationCode::LogicalFUnordGreaterEqual, // GEU
};
const std::size_t index = static_cast<std::size_t>(condition);
ASSERT_MSG(index < std::size(comparison_table), "Invalid condition={}", index);
return Operation(comparison_table[index], op_a, op_b);
}
Node ShaderIR::GetPredicateComparisonInteger(PredCondition condition, bool is_signed, Node op_a,
Node op_b) {
static constexpr std::array comparison_table{
std::pair{PredCondition::LT, OperationCode::LogicalILessThan},
std::pair{PredCondition::EQ, OperationCode::LogicalIEqual},
std::pair{PredCondition::LE, OperationCode::LogicalILessEqual},
std::pair{PredCondition::GT, OperationCode::LogicalIGreaterThan},
std::pair{PredCondition::NE, OperationCode::LogicalINotEqual},
std::pair{PredCondition::GE, OperationCode::LogicalIGreaterEqual},
};
const auto comparison =
std::find_if(comparison_table.cbegin(), comparison_table.cend(),
[condition](const auto entry) { return condition == entry.first; });
UNIMPLEMENTED_IF_MSG(comparison == comparison_table.cend(),
"Unknown predicate comparison operation");
return SignedOperation(comparison->second, is_signed, NO_PRECISE, std::move(op_a),
std::move(op_b));
}
Node ShaderIR::GetPredicateComparisonHalf(Tegra::Shader::PredCondition condition, Node op_a,
Node op_b) {
static constexpr std::array comparison_table{
std::pair{PredCondition::LT, OperationCode::Logical2HLessThan},
std::pair{PredCondition::EQ, OperationCode::Logical2HEqual},
std::pair{PredCondition::LE, OperationCode::Logical2HLessEqual},
std::pair{PredCondition::GT, OperationCode::Logical2HGreaterThan},
std::pair{PredCondition::NE, OperationCode::Logical2HNotEqual},
std::pair{PredCondition::GE, OperationCode::Logical2HGreaterEqual},
std::pair{PredCondition::LTU, OperationCode::Logical2HLessThanWithNan},
std::pair{PredCondition::LEU, OperationCode::Logical2HLessEqualWithNan},
std::pair{PredCondition::GTU, OperationCode::Logical2HGreaterThanWithNan},
std::pair{PredCondition::NEU, OperationCode::Logical2HNotEqualWithNan},
std::pair{PredCondition::GEU, OperationCode::Logical2HGreaterEqualWithNan},
};
const auto comparison =
std::find_if(comparison_table.cbegin(), comparison_table.cend(),
[condition](const auto entry) { return condition == entry.first; });
UNIMPLEMENTED_IF_MSG(comparison == comparison_table.cend(),
"Unknown predicate comparison operation");
return Operation(comparison->second, NO_PRECISE, std::move(op_a), std::move(op_b));
}
OperationCode ShaderIR::GetPredicateCombiner(PredOperation operation) {
static constexpr std::array operation_table{
OperationCode::LogicalAnd,
OperationCode::LogicalOr,
OperationCode::LogicalXor,
};
const auto index = static_cast<std::size_t>(operation);
if (index >= operation_table.size()) {
UNIMPLEMENTED_MSG("Unknown predicate operation.");
return {};
}
return operation_table[index];
}
Node ShaderIR::GetConditionCode(ConditionCode cc) const {
switch (cc) {
case ConditionCode::NEU:
return GetInternalFlag(InternalFlag::Zero, true);
case ConditionCode::FCSM_TR:
UNIMPLEMENTED_MSG("EXIT.FCSM_TR is not implemented");
return MakeNode<PredicateNode>(Pred::NeverExecute, false);
default:
UNIMPLEMENTED_MSG("Unimplemented condition code: {}", cc);
return MakeNode<PredicateNode>(Pred::NeverExecute, false);
}
}
void ShaderIR::SetRegister(NodeBlock& bb, Register dest, Node src) {
bb.push_back(Operation(OperationCode::Assign, GetRegister(dest), std::move(src)));
}
void ShaderIR::SetPredicate(NodeBlock& bb, u64 dest, Node src) {
bb.push_back(Operation(OperationCode::LogicalAssign, GetPredicate(dest), std::move(src)));
}
void ShaderIR::SetInternalFlag(NodeBlock& bb, InternalFlag flag, Node value) {
bb.push_back(Operation(OperationCode::LogicalAssign, GetInternalFlag(flag), std::move(value)));
}
void ShaderIR::SetLocalMemory(NodeBlock& bb, Node address, Node value) {
bb.push_back(
Operation(OperationCode::Assign, GetLocalMemory(std::move(address)), std::move(value)));
}
void ShaderIR::SetSharedMemory(NodeBlock& bb, Node address, Node value) {
bb.push_back(
Operation(OperationCode::Assign, GetSharedMemory(std::move(address)), std::move(value)));
}
void ShaderIR::SetTemporary(NodeBlock& bb, u32 id, Node value) {
SetRegister(bb, Register::ZeroIndex + 1 + id, std::move(value));
}
void ShaderIR::SetInternalFlagsFromFloat(NodeBlock& bb, Node value, bool sets_cc) {
if (!sets_cc) {
return;
}
Node zerop = Operation(OperationCode::LogicalFOrdEqual, std::move(value), Immediate(0.0f));
SetInternalFlag(bb, InternalFlag::Zero, std::move(zerop));
LOG_WARNING(HW_GPU, "Condition codes implementation is incomplete");
}
void ShaderIR::SetInternalFlagsFromInteger(NodeBlock& bb, Node value, bool sets_cc) {
if (!sets_cc) {
return;
}
Node zerop = Operation(OperationCode::LogicalIEqual, std::move(value), Immediate(0));
SetInternalFlag(bb, InternalFlag::Zero, std::move(zerop));
LOG_WARNING(HW_GPU, "Condition codes implementation is incomplete");
}
Node ShaderIR::BitfieldExtract(Node value, u32 offset, u32 bits) {
return Operation(OperationCode::UBitfieldExtract, NO_PRECISE, std::move(value),
Immediate(offset), Immediate(bits));
}
Node ShaderIR::BitfieldInsert(Node base, Node insert, u32 offset, u32 bits) {
return Operation(OperationCode::UBitfieldInsert, NO_PRECISE, base, insert, Immediate(offset),
Immediate(bits));
}
void ShaderIR::MarkAttributeUsage(Attribute::Index index, u64 element) {
switch (index) {
case Attribute::Index::LayerViewportPointSize:
switch (element) {
case 0:
UNIMPLEMENTED();
break;
case 1:
uses_layer = true;
break;
case 2:
uses_viewport_index = true;
break;
case 3:
uses_point_size = true;
break;
}
break;
case Attribute::Index::TessCoordInstanceIDVertexID:
switch (element) {
case 2:
uses_instance_id = true;
break;
case 3:
uses_vertex_id = true;
break;
}
break;
case Attribute::Index::ClipDistances0123:
case Attribute::Index::ClipDistances4567: {
const u64 clip_index = (index == Attribute::Index::ClipDistances4567 ? 4 : 0) + element;
used_clip_distances.at(clip_index) = true;
break;
}
case Attribute::Index::FrontColor:
case Attribute::Index::FrontSecondaryColor:
case Attribute::Index::BackColor:
case Attribute::Index::BackSecondaryColor:
uses_legacy_varyings = true;
break;
default:
if (index >= Attribute::Index::TexCoord_0 && index <= Attribute::Index::TexCoord_7) {
uses_legacy_varyings = true;
}
break;
}
}
std::size_t ShaderIR::DeclareAmend(Node new_amend) {
const auto id = amend_code.size();
amend_code.push_back(std::move(new_amend));
return id;
}
u32 ShaderIR::NewCustomVariable() {
return num_custom_variables++;
}
} // namespace VideoCommon::Shader

479
src/video_core/shader/shader_ir.h
View file

@ -1,479 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <array>
#include <list>
#include <map>
#include <optional>
#include <set>
#include <tuple>
#include <vector>
#include "common/common_types.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/engines/shader_header.h"
#include "video_core/shader/ast.h"
#include "video_core/shader/compiler_settings.h"
#include "video_core/shader/memory_util.h"
#include "video_core/shader/node.h"
#include "video_core/shader/registry.h"
namespace VideoCommon::Shader {
struct ShaderBlock;
constexpr u32 MAX_PROGRAM_LENGTH = 0x1000;
struct ConstBuffer {
constexpr explicit ConstBuffer(u32 max_offset_, bool is_indirect_)
: max_offset{max_offset_}, is_indirect{is_indirect_} {}
constexpr ConstBuffer() = default;
void MarkAsUsed(u64 offset) {
max_offset = std::max(max_offset, static_cast<u32>(offset));
}
void MarkAsUsedIndirect() {
is_indirect = true;
}
bool IsIndirect() const {
return is_indirect;
}
u32 GetSize() const {
return max_offset + static_cast<u32>(sizeof(float));
}
u32 GetMaxOffset() const {
return max_offset;
}
private:
u32 max_offset = 0;
bool is_indirect = false;
};
struct GlobalMemoryUsage {
bool is_read{};
bool is_written{};
};
class ShaderIR final {
public:
explicit ShaderIR(const ProgramCode& program_code_, u32 main_offset_,
CompilerSettings settings_, Registry& registry_);
~ShaderIR();
const std::map<u32, NodeBlock>& GetBasicBlocks() const {
return basic_blocks;
}
const std::set<u32>& GetRegisters() const {
return used_registers;
}
const std::set<Tegra::Shader::Pred>& GetPredicates() const {
return used_predicates;
}
const std::set<Tegra::Shader::Attribute::Index>& GetInputAttributes() const {
return used_input_attributes;
}
const std::set<Tegra::Shader::Attribute::Index>& GetOutputAttributes() const {
return used_output_attributes;
}
const std::map<u32, ConstBuffer>& GetConstantBuffers() const {
return used_cbufs;
}
const std::list<SamplerEntry>& GetSamplers() const {
return used_samplers;
}
const std::list<ImageEntry>& GetImages() const {
return used_images;
}
const std::array<bool, Tegra::Engines::Maxwell3D::Regs::NumClipDistances>& GetClipDistances()
const {
return used_clip_distances;
}
const std::map<GlobalMemoryBase, GlobalMemoryUsage>& GetGlobalMemory() const {
return used_global_memory;
}
std::size_t GetLength() const {
return static_cast<std::size_t>(coverage_end * sizeof(u64));
}
bool UsesLayer() const {
return uses_layer;
}
bool UsesViewportIndex() const {
return uses_viewport_index;
}
bool UsesPointSize() const {
return uses_point_size;
}
bool UsesInstanceId() const {
return uses_instance_id;
}
bool UsesVertexId() const {
return uses_vertex_id;
}
bool UsesLegacyVaryings() const {
return uses_legacy_varyings;
}
bool UsesYNegate() const {
return uses_y_negate;
}
bool UsesWarps() const {
return uses_warps;
}
bool HasPhysicalAttributes() const {
return uses_physical_attributes;
}
const Tegra::Shader::Header& GetHeader() const {
return header;
}
bool IsFlowStackDisabled() const {
return disable_flow_stack;
}
bool IsDecompiled() const {
return decompiled;
}
const ASTManager& GetASTManager() const {
return program_manager;
}
ASTNode GetASTProgram() const {
return program_manager.GetProgram();
}
u32 GetASTNumVariables() const {
return program_manager.GetVariables();
}
u32 ConvertAddressToNvidiaSpace(u32 address) const {
return (address - main_offset) * static_cast<u32>(sizeof(Tegra::Shader::Instruction));
}
/// Returns a condition code evaluated from internal flags
Node GetConditionCode(Tegra::Shader::ConditionCode cc) const;
const Node& GetAmendNode(std::size_t index) const {
return amend_code[index];
}
u32 GetNumCustomVariables() const {
return num_custom_variables;
}
private:
friend class ASTDecoder;
struct SamplerInfo {
std::optional<Tegra::Shader::TextureType> type;
std::optional<bool> is_array;
std::optional<bool> is_shadow;
std::optional<bool> is_buffer;
constexpr bool IsComplete() const noexcept {
return type && is_array && is_shadow && is_buffer;
}
};
void Decode();
void PostDecode();
NodeBlock DecodeRange(u32 begin, u32 end);
void DecodeRangeInner(NodeBlock& bb, u32 begin, u32 end);
void InsertControlFlow(NodeBlock& bb, const ShaderBlock& block);
/**
* Decodes a single instruction from Tegra to IR.
* @param bb Basic block where the nodes will be written to.
* @param pc Program counter. Offset to decode.
* @return Next address to decode.
*/
u32 DecodeInstr(NodeBlock& bb, u32 pc);
u32 DecodeArithmetic(NodeBlock& bb, u32 pc);
u32 DecodeArithmeticImmediate(NodeBlock& bb, u32 pc);
u32 DecodeBfe(NodeBlock& bb, u32 pc);
u32 DecodeBfi(NodeBlock& bb, u32 pc);
u32 DecodeShift(NodeBlock& bb, u32 pc);
u32 DecodeArithmeticInteger(NodeBlock& bb, u32 pc);
u32 DecodeArithmeticIntegerImmediate(NodeBlock& bb, u32 pc);
u32 DecodeArithmeticHalf(NodeBlock& bb, u32 pc);
u32 DecodeArithmeticHalfImmediate(NodeBlock& bb, u32 pc);
u32 DecodeFfma(NodeBlock& bb, u32 pc);
u32 DecodeHfma2(NodeBlock& bb, u32 pc);
u32 DecodeConversion(NodeBlock& bb, u32 pc);
u32 DecodeWarp(NodeBlock& bb, u32 pc);
u32 DecodeMemory(NodeBlock& bb, u32 pc);
u32 DecodeTexture(NodeBlock& bb, u32 pc);
u32 DecodeImage(NodeBlock& bb, u32 pc);
u32 DecodeFloatSetPredicate(NodeBlock& bb, u32 pc);
u32 DecodeIntegerSetPredicate(NodeBlock& bb, u32 pc);
u32 DecodeHalfSetPredicate(NodeBlock& bb, u32 pc);
u32 DecodePredicateSetRegister(NodeBlock& bb, u32 pc);
u32 DecodePredicateSetPredicate(NodeBlock& bb, u32 pc);
u32 DecodeRegisterSetPredicate(NodeBlock& bb, u32 pc);
u32 DecodeFloatSet(NodeBlock& bb, u32 pc);
u32 DecodeIntegerSet(NodeBlock& bb, u32 pc);
u32 DecodeHalfSet(NodeBlock& bb, u32 pc);
u32 DecodeVideo(NodeBlock& bb, u32 pc);
u32 DecodeXmad(NodeBlock& bb, u32 pc);
u32 DecodeOther(NodeBlock& bb, u32 pc);
/// Generates a node for a passed register.
Node GetRegister(Tegra::Shader::Register reg);
/// Generates a node for a custom variable
Node GetCustomVariable(u32 id);
/// Generates a node representing a 19-bit immediate value
Node GetImmediate19(Tegra::Shader::Instruction instr);
/// Generates a node representing a 32-bit immediate value
Node GetImmediate32(Tegra::Shader::Instruction instr);
/// Generates a node representing a constant buffer
Node GetConstBuffer(u64 index, u64 offset);
/// Generates a node representing a constant buffer with a variadic offset
Node GetConstBufferIndirect(u64 index, u64 offset, Node node);
/// Generates a node for a passed predicate. It can be optionally negated
Node GetPredicate(u64 pred, bool negated = false);
/// Generates a predicate node for an immediate true or false value
Node GetPredicate(bool immediate);
/// Generates a node representing an input attribute. Keeps track of used attributes.
Node GetInputAttribute(Tegra::Shader::Attribute::Index index, u64 element, Node buffer = {});
/// Generates a node representing a physical input attribute.
Node GetPhysicalInputAttribute(Tegra::Shader::Register physical_address, Node buffer = {});
/// Generates a node representing an output attribute. Keeps track of used attributes.
Node GetOutputAttribute(Tegra::Shader::Attribute::Index index, u64 element, Node buffer);
/// Generates a node representing an internal flag
Node GetInternalFlag(InternalFlag flag, bool negated = false) const;
/// Generates a node representing a local memory address
Node GetLocalMemory(Node address);
/// Generates a node representing a shared memory address
Node GetSharedMemory(Node address);
/// Generates a temporary, internally it uses a post-RZ register
Node GetTemporary(u32 id);
/// Sets a register. src value must be a number-evaluated node.
void SetRegister(NodeBlock& bb, Tegra::Shader::Register dest, Node src);
/// Sets a predicate. src value must be a bool-evaluated node
void SetPredicate(NodeBlock& bb, u64 dest, Node src);
/// Sets an internal flag. src value must be a bool-evaluated node
void SetInternalFlag(NodeBlock& bb, InternalFlag flag, Node value);
/// Sets a local memory address with a value.
void SetLocalMemory(NodeBlock& bb, Node address, Node value);
/// Sets a shared memory address with a value.
void SetSharedMemory(NodeBlock& bb, Node address, Node value);
/// Sets a temporary. Internally it uses a post-RZ register
void SetTemporary(NodeBlock& bb, u32 id, Node value);
/// Sets internal flags from a float
void SetInternalFlagsFromFloat(NodeBlock& bb, Node value, bool sets_cc = true);
/// Sets internal flags from an integer
void SetInternalFlagsFromInteger(NodeBlock& bb, Node value, bool sets_cc = true);
/// Conditionally absolute/negated float. Absolute is applied first
Node GetOperandAbsNegFloat(Node value, bool absolute, bool negate);
/// Conditionally saturates a float
Node GetSaturatedFloat(Node value, bool saturate = true);
/// Converts an integer to different sizes.
Node ConvertIntegerSize(Node value, Tegra::Shader::Register::Size size, bool is_signed);
/// Conditionally absolute/negated integer. Absolute is applied first
Node GetOperandAbsNegInteger(Node value, bool absolute, bool negate, bool is_signed);
/// Unpacks a half immediate from an instruction
Node UnpackHalfImmediate(Tegra::Shader::Instruction instr, bool has_negation);
/// Unpacks a binary value into a half float pair with a type format
Node UnpackHalfFloat(Node value, Tegra::Shader::HalfType type);
/// Merges a half pair into another value
Node HalfMerge(Node dest, Node src, Tegra::Shader::HalfMerge merge);
/// Conditionally absolute/negated half float pair. Absolute is applied first
Node GetOperandAbsNegHalf(Node value, bool absolute, bool negate);
/// Conditionally saturates a half float pair
Node GetSaturatedHalfFloat(Node value, bool saturate = true);
/// Get image component value by type and size
std::pair<Node, bool> GetComponentValue(Tegra::Texture::ComponentType component_type,
u32 component_size, Node original_value);
/// Returns a predicate comparing two floats
Node GetPredicateComparisonFloat(Tegra::Shader::PredCondition condition, Node op_a, Node op_b);
/// Returns a predicate comparing two integers
Node GetPredicateComparisonInteger(Tegra::Shader::PredCondition condition, bool is_signed,
Node op_a, Node op_b);
/// Returns a predicate comparing two half floats. meta consumes how both pairs will be compared
Node GetPredicateComparisonHalf(Tegra::Shader::PredCondition condition, Node op_a, Node op_b);
/// Returns a predicate combiner operation
OperationCode GetPredicateCombiner(Tegra::Shader::PredOperation operation);
/// Queries the missing sampler info from the execution context.
SamplerInfo GetSamplerInfo(SamplerInfo info,
std::optional<Tegra::Engines::SamplerDescriptor> sampler);
/// Accesses a texture sampler.
std::optional<SamplerEntry> GetSampler(Tegra::Shader::Sampler sampler, SamplerInfo info);
/// Accesses a texture sampler for a bindless texture.
std::optional<SamplerEntry> GetBindlessSampler(Tegra::Shader::Register reg, SamplerInfo info,
Node& index_var);
/// Accesses an image.
ImageEntry& GetImage(Tegra::Shader::Image image, Tegra::Shader::ImageType type);
/// Access a bindless image sampler.
ImageEntry& GetBindlessImage(Tegra::Shader::Register reg, Tegra::Shader::ImageType type);
/// Extracts a sequence of bits from a node
Node BitfieldExtract(Node value, u32 offset, u32 bits);
/// Inserts a sequence of bits from a node
Node BitfieldInsert(Node base, Node insert, u32 offset, u32 bits);
/// Marks the usage of a input or output attribute.
void MarkAttributeUsage(Tegra::Shader::Attribute::Index index, u64 element);
/// Decodes VMNMX instruction and inserts its code into the passed basic block.
void DecodeVMNMX(NodeBlock& bb, Tegra::Shader::Instruction instr);
void WriteTexInstructionFloat(NodeBlock& bb, Tegra::Shader::Instruction instr,
const Node4& components);
void WriteTexsInstructionFloat(NodeBlock& bb, Tegra::Shader::Instruction instr,
const Node4& components, bool ignore_mask = false);
void WriteTexsInstructionHalfFloat(NodeBlock& bb, Tegra::Shader::Instruction instr,
const Node4& components, bool ignore_mask = false);
Node4 GetTexCode(Tegra::Shader::Instruction instr, Tegra::Shader::TextureType texture_type,
Tegra::Shader::TextureProcessMode process_mode, bool depth_compare,
bool is_array, bool is_aoffi,
std::optional<Tegra::Shader::Register> bindless_reg);
Node4 GetTexsCode(Tegra::Shader::Instruction instr, Tegra::Shader::TextureType texture_type,
Tegra::Shader::TextureProcessMode process_mode, bool depth_compare,
bool is_array);
Node4 GetTld4Code(Tegra::Shader::Instruction instr, Tegra::Shader::TextureType texture_type,
bool depth_compare, bool is_array, bool is_aoffi, bool is_ptp,
bool is_bindless);
Node4 GetTldCode(Tegra::Shader::Instruction instr);
Node4 GetTldsCode(Tegra::Shader::Instruction instr, Tegra::Shader::TextureType texture_type,
bool is_array);
std::tuple<std::size_t, std::size_t> ValidateAndGetCoordinateElement(
Tegra::Shader::TextureType texture_type, bool depth_compare, bool is_array,
bool lod_bias_enabled, std::size_t max_coords, std::size_t max_inputs);
std::vector<Node> GetAoffiCoordinates(Node aoffi_reg, std::size_t coord_count, bool is_tld4);
std::vector<Node> GetPtpCoordinates(std::array<Node, 2> ptp_regs);
Node4 GetTextureCode(Tegra::Shader::Instruction instr, Tegra::Shader::TextureType texture_type,
Tegra::Shader::TextureProcessMode process_mode, std::vector<Node> coords,
Node array, Node depth_compare, u32 bias_offset, std::vector<Node> aoffi,
std::optional<Tegra::Shader::Register> bindless_reg);
Node GetVideoOperand(Node op, bool is_chunk, bool is_signed, Tegra::Shader::VideoType type,
u64 byte_height);
void WriteLogicOperation(NodeBlock& bb, Tegra::Shader::Register dest,
Tegra::Shader::LogicOperation logic_op, Node op_a, Node op_b,
Tegra::Shader::PredicateResultMode predicate_mode,
Tegra::Shader::Pred predicate, bool sets_cc);
void WriteLop3Instruction(NodeBlock& bb, Tegra::Shader::Register dest, Node op_a, Node op_b,
Node op_c, Node imm_lut, bool sets_cc);
std::tuple<Node, u32, u32> TrackCbuf(Node tracked, const NodeBlock& code, s64 cursor) const;
std::pair<Node, TrackSampler> TrackBindlessSampler(Node tracked, const NodeBlock& code,
s64 cursor);
std::pair<Node, TrackSampler> HandleBindlessIndirectRead(const CbufNode& cbuf,
const OperationNode& operation,
Node gpr, Node base_offset,
Node tracked, const NodeBlock& code,
s64 cursor);
std::optional<u32> TrackImmediate(Node tracked, const NodeBlock& code, s64 cursor) const;
std::pair<Node, s64> TrackRegister(const GprNode* tracked, const NodeBlock& code,
s64 cursor) const;
std::tuple<Node, Node, GlobalMemoryBase> TrackGlobalMemory(NodeBlock& bb,
Tegra::Shader::Instruction instr,
bool is_read, bool is_write);
/// Register new amending code and obtain the reference id.
std::size_t DeclareAmend(Node new_amend);
u32 NewCustomVariable();
const ProgramCode& program_code;
const u32 main_offset;
const CompilerSettings settings;
Registry& registry;
bool decompiled{};
bool disable_flow_stack{};
u32 coverage_begin{};
u32 coverage_end{};
std::map<u32, NodeBlock> basic_blocks;
NodeBlock global_code;
ASTManager program_manager{true, true};
std::vector<Node> amend_code;
u32 num_custom_variables{};
std::set<u32> used_registers;
std::set<Tegra::Shader::Pred> used_predicates;
std::set<Tegra::Shader::Attribute::Index> used_input_attributes;
std::set<Tegra::Shader::Attribute::Index> used_output_attributes;
std::map<u32, ConstBuffer> used_cbufs;
std::list<SamplerEntry> used_samplers;
std::list<ImageEntry> used_images;
std::array<bool, Tegra::Engines::Maxwell3D::Regs::NumClipDistances> used_clip_distances{};
std::map<GlobalMemoryBase, GlobalMemoryUsage> used_global_memory;
bool uses_layer{};
bool uses_viewport_index{};
bool uses_point_size{};
bool uses_physical_attributes{}; // Shader uses AL2P or physical attribute read/writes
bool uses_instance_id{};
bool uses_vertex_id{};
bool uses_legacy_varyings{};
bool uses_y_negate{};
bool uses_warps{};
bool uses_indexed_samplers{};
Tegra::Shader::Header header;
};
} // namespace VideoCommon::Shader

236
src/video_core/shader/track.cpp
View file

@ -1,236 +0,0 @@
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <algorithm>
#include <utility>
#include <variant>
#include "common/common_types.h"
#include "video_core/shader/node.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
namespace {
std::pair<Node, s64> FindOperation(const NodeBlock& code, s64 cursor,
OperationCode operation_code) {
for (; cursor >= 0; --cursor) {
Node node = code.at(cursor);
if (const auto operation = std::get_if<OperationNode>(&*node)) {
if (operation->GetCode() == operation_code) {
return {std::move(node), cursor};
}
}
if (const auto conditional = std::get_if<ConditionalNode>(&*node)) {
const auto& conditional_code = conditional->GetCode();
auto result = FindOperation(
conditional_code, static_cast<s64>(conditional_code.size() - 1), operation_code);
auto& found = result.first;
if (found) {
return {std::move(found), cursor};
}
}
}
return {};
}
std::optional<std::pair<Node, Node>> DecoupleIndirectRead(const OperationNode& operation) {
if (operation.GetCode() != OperationCode::UAdd) {
return std::nullopt;
}
Node gpr;
Node offset;
ASSERT(operation.GetOperandsCount() == 2);
for (std::size_t i = 0; i < operation.GetOperandsCount(); i++) {
Node operand = operation[i];
if (std::holds_alternative<ImmediateNode>(*operand)) {
offset = operation[i];
} else if (std::holds_alternative<GprNode>(*operand)) {
gpr = operation[i];
}
}
if (offset && gpr) {
return std::make_pair(gpr, offset);
}
return std::nullopt;
}
bool AmendNodeCv(std::size_t amend_index, Node node) {
if (const auto operation = std::get_if<OperationNode>(&*node)) {
operation->SetAmendIndex(amend_index);
return true;
}
if (const auto conditional = std::get_if<ConditionalNode>(&*node)) {
conditional->SetAmendIndex(amend_index);
return true;
}
return false;
}
} // Anonymous namespace
std::pair<Node, TrackSampler> ShaderIR::TrackBindlessSampler(Node tracked, const NodeBlock& code,
s64 cursor) {
if (const auto cbuf = std::get_if<CbufNode>(&*tracked)) {
const u32 cbuf_index = cbuf->GetIndex();
// Constant buffer found, test if it's an immediate
const auto& offset = cbuf->GetOffset();
if (const auto immediate = std::get_if<ImmediateNode>(&*offset)) {
auto track = MakeTrackSampler<BindlessSamplerNode>(cbuf_index, immediate->GetValue());
return {tracked, track};
}
if (const auto operation = std::get_if<OperationNode>(&*offset)) {
const u32 bound_buffer = registry.GetBoundBuffer();
if (bound_buffer != cbuf_index) {
return {};
}
if (const std::optional pair = DecoupleIndirectRead(*operation)) {
auto [gpr, base_offset] = *pair;
return HandleBindlessIndirectRead(*cbuf, *operation, gpr, base_offset, tracked,
code, cursor);
}
}
return {};
}
if (const auto gpr = std::get_if<GprNode>(&*tracked)) {
if (gpr->GetIndex() == Tegra::Shader::Register::ZeroIndex) {
return {};
}
// Reduce the cursor in one to avoid infinite loops when the instruction sets the same
// register that it uses as operand
const auto [source, new_cursor] = TrackRegister(gpr, code, cursor - 1);
if (!source) {
return {};
}
return TrackBindlessSampler(source, code, new_cursor);
}
if (const auto operation = std::get_if<OperationNode>(&*tracked)) {
const OperationNode& op = *operation;
const OperationCode opcode = operation->GetCode();
if (opcode == OperationCode::IBitwiseOr || opcode == OperationCode::UBitwiseOr) {
ASSERT(op.GetOperandsCount() == 2);
auto [node_a, index_a, offset_a] = TrackCbuf(op[0], code, cursor);
auto [node_b, index_b, offset_b] = TrackCbuf(op[1], code, cursor);
if (node_a && node_b) {
auto track = MakeTrackSampler<SeparateSamplerNode>(std::pair{index_a, index_b},
std::pair{offset_a, offset_b});
return {tracked, std::move(track)};
}
}
std::size_t i = op.GetOperandsCount();
while (i--) {
if (auto found = TrackBindlessSampler(op[i - 1], code, cursor); std::get<0>(found)) {
// Constant buffer found in operand.
return found;
}
}
return {};
}
if (const auto conditional = std::get_if<ConditionalNode>(&*tracked)) {
const auto& conditional_code = conditional->GetCode();
return TrackBindlessSampler(tracked, conditional_code,
static_cast<s64>(conditional_code.size()));
}
return {};
}
std::pair<Node, TrackSampler> ShaderIR::HandleBindlessIndirectRead(
const CbufNode& cbuf, const OperationNode& operation, Node gpr, Node base_offset, Node tracked,
const NodeBlock& code, s64 cursor) {
const auto offset_imm = std::get<ImmediateNode>(*base_offset);
const auto& gpu_driver = registry.AccessGuestDriverProfile();
const u32 bindless_cv = NewCustomVariable();
const u32 texture_handler_size = gpu_driver.GetTextureHandlerSize();
Node op = Operation(OperationCode::UDiv, gpr, Immediate(texture_handler_size));
Node cv_node = GetCustomVariable(bindless_cv);
Node amend_op = Operation(OperationCode::Assign, std::move(cv_node), std::move(op));
const std::size_t amend_index = DeclareAmend(std::move(amend_op));
AmendNodeCv(amend_index, code[cursor]);
// TODO: Implement bindless index custom variable
auto track =
MakeTrackSampler<ArraySamplerNode>(cbuf.GetIndex(), offset_imm.GetValue(), bindless_cv);
return {tracked, track};
}
std::tuple<Node, u32, u32> ShaderIR::TrackCbuf(Node tracked, const NodeBlock& code,
s64 cursor) const {
if (const auto cbuf = std::get_if<CbufNode>(&*tracked)) {
// Constant buffer found, test if it's an immediate
const auto& offset = cbuf->GetOffset();
if (const auto immediate = std::get_if<ImmediateNode>(&*offset)) {
return {tracked, cbuf->GetIndex(), immediate->GetValue()};
}
return {};
}
if (const auto gpr = std::get_if<GprNode>(&*tracked)) {
if (gpr->GetIndex() == Tegra::Shader::Register::ZeroIndex) {
return {};
}
// Reduce the cursor in one to avoid infinite loops when the instruction sets the same
// register that it uses as operand
const auto [source, new_cursor] = TrackRegister(gpr, code, cursor - 1);
if (!source) {
return {};
}
return TrackCbuf(source, code, new_cursor);
}
if (const auto operation = std::get_if<OperationNode>(&*tracked)) {
for (std::size_t i = operation->GetOperandsCount(); i > 0; --i) {
if (auto found = TrackCbuf((*operation)[i - 1], code, cursor); std::get<0>(found)) {
// Cbuf found in operand.
return found;
}
}
return {};
}
if (const auto conditional = std::get_if<ConditionalNode>(&*tracked)) {
const auto& conditional_code = conditional->GetCode();
return TrackCbuf(tracked, conditional_code, static_cast<s64>(conditional_code.size()));
}
return {};
}
std::optional<u32> ShaderIR::TrackImmediate(Node tracked, const NodeBlock& code, s64 cursor) const {
// Reduce the cursor in one to avoid infinite loops when the instruction sets the same register
// that it uses as operand
const auto result = TrackRegister(&std::get<GprNode>(*tracked), code, cursor - 1);
const auto& found = result.first;
if (!found) {
return std::nullopt;
}
if (const auto immediate = std::get_if<ImmediateNode>(&*found)) {
return immediate->GetValue();
}
return std::nullopt;
}
std::pair<Node, s64> ShaderIR::TrackRegister(const GprNode* tracked, const NodeBlock& code,
s64 cursor) const {
for (; cursor >= 0; --cursor) {
const auto [found_node, new_cursor] = FindOperation(code, cursor, OperationCode::Assign);
if (!found_node) {
return {};
}
const auto operation = std::get_if<OperationNode>(&*found_node);
ASSERT(operation);
const auto& target = (*operation)[0];
if (const auto gpr_target = std::get_if<GprNode>(&*target)) {
if (gpr_target->GetIndex() == tracked->GetIndex()) {
return {(*operation)[1], new_cursor};
}
}
}
return {};
}
} // namespace VideoCommon::Shader

115
src/video_core/shader/transform_feedback.cpp
View file

@ -1,115 +0,0 @@
// Copyright 2020 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <algorithm>
#include <array>
#include <unordered_map>
#include "common/assert.h"
#include "common/common_types.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/shader/registry.h"
#include "video_core/shader/transform_feedback.h"
namespace VideoCommon::Shader {
namespace {
using Maxwell = Tegra::Engines::Maxwell3D::Regs;
// TODO(Rodrigo): Change this to constexpr std::unordered_set in C++20
/// Attribute offsets that describe a vector
constexpr std::array VECTORS = {
28, // gl_Position
32, // Generic 0
36, // Generic 1
40, // Generic 2
44, // Generic 3
48, // Generic 4
52, // Generic 5
56, // Generic 6
60, // Generic 7
64, // Generic 8
68, // Generic 9
72, // Generic 10
76, // Generic 11
80, // Generic 12
84, // Generic 13
88, // Generic 14
92, // Generic 15
96, // Generic 16
100, // Generic 17
104, // Generic 18
108, // Generic 19
112, // Generic 20
116, // Generic 21
120, // Generic 22
124, // Generic 23
128, // Generic 24
132, // Generic 25
136, // Generic 26
140, // Generic 27
144, // Generic 28
148, // Generic 29
152, // Generic 30
156, // Generic 31
160, // gl_FrontColor
164, // gl_FrontSecondaryColor
160, // gl_BackColor
164, // gl_BackSecondaryColor
192, // gl_TexCoord[0]
196, // gl_TexCoord[1]
200, // gl_TexCoord[2]
204, // gl_TexCoord[3]
208, // gl_TexCoord[4]
212, // gl_TexCoord[5]
216, // gl_TexCoord[6]
220, // gl_TexCoord[7]
};
} // namespace
std::unordered_map<u8, VaryingTFB> BuildTransformFeedback(const GraphicsInfo& info) {
std::unordered_map<u8, VaryingTFB> tfb;
for (std::size_t buffer = 0; buffer < Maxwell::NumTransformFeedbackBuffers; ++buffer) {
const auto& locations = info.tfb_varying_locs[buffer];
const auto& layout = info.tfb_layouts[buffer];
const std::size_t varying_count = layout.varying_count;
std::size_t highest = 0;
for (std::size_t offset = 0; offset < varying_count; ++offset) {
const std::size_t base_offset = offset;
const u8 location = locations[offset];
VaryingTFB varying;
varying.buffer = layout.stream;
varying.stride = layout.stride;
varying.offset = offset * sizeof(u32);
varying.components = 1;
if (std::find(VECTORS.begin(), VECTORS.end(), location / 4 * 4) != VECTORS.end()) {
UNIMPLEMENTED_IF_MSG(location % 4 != 0, "Unaligned TFB");
const u8 base_index = location / 4;
while (offset + 1 < varying_count && base_index == locations[offset + 1] / 4) {
++offset;
++varying.components;
}
}
[[maybe_unused]] const bool inserted = tfb.emplace(location, varying).second;
UNIMPLEMENTED_IF_MSG(!inserted, "Varying already stored");
highest = std::max(highest, (base_offset + varying.components) * sizeof(u32));
}
UNIMPLEMENTED_IF(highest != layout.stride);
}
return tfb;
}
} // namespace VideoCommon::Shader

23
src/video_core/shader/transform_feedback.h
View file

@ -1,23 +0,0 @@
// Copyright 2020 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <unordered_map>
#include "common/common_types.h"
#include "video_core/shader/registry.h"
namespace VideoCommon::Shader {
struct VaryingTFB {
std::size_t buffer;
std::size_t stride;
std::size_t offset;
std::size_t components;
};
std::unordered_map<u8, VaryingTFB> BuildTransformFeedback(const GraphicsInfo& info);
} // namespace VideoCommon::Shader