summaryrefslogtreecommitdiffstats
path: root/src/video_core
diff options
context:
space:
mode:
Diffstat (limited to 'src/video_core')
-rw-r--r--src/video_core/CMakeLists.txt2
-rw-r--r--src/video_core/engines/maxwell_3d.h1
-rw-r--r--src/video_core/engines/shader_bytecode.h4
-rw-r--r--src/video_core/fence_manager.h6
-rw-r--r--src/video_core/renderer_opengl/gl_rasterizer.cpp79
-rw-r--r--src/video_core/renderer_opengl/gl_rasterizer.h3
-rw-r--r--src/video_core/renderer_opengl/gl_shader_cache.cpp82
-rw-r--r--src/video_core/renderer_opengl/gl_shader_decompiler.cpp62
-rw-r--r--src/video_core/renderer_opengl/gl_shader_decompiler.h33
-rw-r--r--src/video_core/renderer_vulkan/fixed_pipeline_state.cpp2
-rw-r--r--src/video_core/renderer_vulkan/fixed_pipeline_state.h2
-rw-r--r--src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp2
-rw-r--r--src/video_core/renderer_vulkan/vk_pipeline_cache.cpp73
-rw-r--r--src/video_core/renderer_vulkan/vk_pipeline_cache.h8
-rw-r--r--src/video_core/renderer_vulkan/vk_rasterizer.cpp17
-rw-r--r--src/video_core/renderer_vulkan/vk_shader_decompiler.cpp83
-rw-r--r--src/video_core/shader/control_flow.cpp12
-rw-r--r--src/video_core/shader/decode.cpp30
-rw-r--r--src/video_core/shader/decode/arithmetic_integer.cpp31
-rw-r--r--src/video_core/shader/decode/image.cpp18
-rw-r--r--src/video_core/shader/decode/register_set_predicate.cpp52
-rw-r--r--src/video_core/shader/decode/texture.cpp192
-rw-r--r--src/video_core/shader/memory_util.cpp77
-rw-r--r--src/video_core/shader/memory_util.h47
-rw-r--r--src/video_core/shader/node.h131
-rw-r--r--src/video_core/shader/shader_ir.h37
-rw-r--r--src/video_core/shader/track.cpp20
-rw-r--r--src/video_core/texture_cache/surface_params.cpp10
-rw-r--r--src/video_core/texture_cache/texture_cache.h70
29 files changed, 595 insertions, 591 deletions
diff --git a/src/video_core/CMakeLists.txt b/src/video_core/CMakeLists.txt
index 8ede4ba9b..ff53282c9 100644
--- a/src/video_core/CMakeLists.txt
+++ b/src/video_core/CMakeLists.txt
@@ -124,6 +124,8 @@ add_library(video_core STATIC
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
diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h
index 3dfba8197..5e522e0d2 100644
--- a/src/video_core/engines/maxwell_3d.h
+++ b/src/video_core/engines/maxwell_3d.h
@@ -1179,6 +1179,7 @@ public:
BitField<0, 1, u32> depth_range_0_1;
BitField<3, 1, u32> depth_clamp_near;
BitField<4, 1, u32> depth_clamp_far;
+ BitField<11, 1, u32> depth_clamp_disabled;
} view_volume_clip_control;
INSERT_UNION_PADDING_WORDS(0x1F);
diff --git a/src/video_core/engines/shader_bytecode.h b/src/video_core/engines/shader_bytecode.h
index cde3a26b9..8dae754d4 100644
--- a/src/video_core/engines/shader_bytecode.h
+++ b/src/video_core/engines/shader_bytecode.h
@@ -814,6 +814,10 @@ union Instruction {
} alu_integer;
union {
+ BitField<43, 1, u64> x;
+ } iadd;
+
+ union {
BitField<39, 1, u64> ftz;
BitField<32, 1, u64> saturate;
BitField<49, 2, HalfMerge> merge;
diff --git a/src/video_core/fence_manager.h b/src/video_core/fence_manager.h
index dabd1588c..8b2a6a42c 100644
--- a/src/video_core/fence_manager.h
+++ b/src/video_core/fence_manager.h
@@ -88,7 +88,8 @@ public:
}
PopAsyncFlushes();
if (current_fence->IsSemaphore()) {
- memory_manager.Write<u32>(current_fence->GetAddress(), current_fence->GetPayload());
+ memory_manager.template Write<u32>(current_fence->GetAddress(),
+ current_fence->GetPayload());
} else {
gpu.IncrementSyncPoint(current_fence->GetPayload());
}
@@ -134,7 +135,8 @@ private:
}
PopAsyncFlushes();
if (current_fence->IsSemaphore()) {
- memory_manager.Write<u32>(current_fence->GetAddress(), current_fence->GetPayload());
+ memory_manager.template Write<u32>(current_fence->GetAddress(),
+ current_fence->GetPayload());
} else {
gpu.IncrementSyncPoint(current_fence->GetPayload());
}
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp
index 6fe155bcc..725b4c32d 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.cpp
+++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp
@@ -59,14 +59,12 @@ constexpr std::size_t NumSupportedVertexAttributes = 16;
template <typename Engine, typename Entry>
Tegra::Texture::FullTextureInfo GetTextureInfo(const Engine& engine, const Entry& entry,
ShaderType shader_type, std::size_t index = 0) {
- if (entry.IsBindless()) {
- const Tegra::Texture::TextureHandle tex_handle =
- engine.AccessConstBuffer32(shader_type, entry.GetBuffer(), entry.GetOffset());
+ if (entry.is_bindless) {
+ const auto tex_handle = engine.AccessConstBuffer32(shader_type, entry.buffer, entry.offset);
return engine.GetTextureInfo(tex_handle);
}
const auto& gpu_profile = engine.AccessGuestDriverProfile();
- const u32 offset =
- entry.GetOffset() + static_cast<u32>(index * gpu_profile.GetTextureHandlerSize());
+ const u32 offset = entry.offset + static_cast<u32>(index * gpu_profile.GetTextureHandlerSize());
if constexpr (std::is_same_v<Engine, Tegra::Engines::Maxwell3D>) {
return engine.GetStageTexture(shader_type, offset);
} else {
@@ -348,7 +346,7 @@ void RasterizerOpenGL::ConfigureFramebuffers() {
texture_cache.GuardRenderTargets(true);
- View depth_surface = texture_cache.GetDepthBufferSurface();
+ View depth_surface = texture_cache.GetDepthBufferSurface(true);
const auto& regs = gpu.regs;
UNIMPLEMENTED_IF(regs.rt_separate_frag_data == 0);
@@ -357,7 +355,7 @@ void RasterizerOpenGL::ConfigureFramebuffers() {
FramebufferCacheKey key;
const auto colors_count = static_cast<std::size_t>(regs.rt_control.count);
for (std::size_t index = 0; index < colors_count; ++index) {
- View color_surface{texture_cache.GetColorBufferSurface(index)};
+ View color_surface{texture_cache.GetColorBufferSurface(index, true)};
if (!color_surface) {
continue;
}
@@ -381,28 +379,52 @@ void RasterizerOpenGL::ConfigureFramebuffers() {
glBindFramebuffer(GL_DRAW_FRAMEBUFFER, framebuffer_cache.GetFramebuffer(key));
}
-void RasterizerOpenGL::ConfigureClearFramebuffer(bool using_color_fb, bool using_depth_fb,
- bool using_stencil_fb) {
+void RasterizerOpenGL::ConfigureClearFramebuffer(bool using_color, bool using_depth_stencil) {
auto& gpu = system.GPU().Maxwell3D();
const auto& regs = gpu.regs;
texture_cache.GuardRenderTargets(true);
View color_surface;
- if (using_color_fb) {
+
+ if (using_color) {
+ // Determine if we have to preserve the contents.
+ // First we have to make sure all clear masks are enabled.
+ bool preserve_contents = !regs.clear_buffers.R || !regs.clear_buffers.G ||
+ !regs.clear_buffers.B || !regs.clear_buffers.A;
const std::size_t index = regs.clear_buffers.RT;
- color_surface = texture_cache.GetColorBufferSurface(index);
+ if (regs.clear_flags.scissor) {
+ // Then we have to confirm scissor testing clears the whole image.
+ const auto& scissor = regs.scissor_test[0];
+ preserve_contents |= scissor.min_x > 0;
+ preserve_contents |= scissor.min_y > 0;
+ preserve_contents |= scissor.max_x < regs.rt[index].width;
+ preserve_contents |= scissor.max_y < regs.rt[index].height;
+ }
+
+ color_surface = texture_cache.GetColorBufferSurface(index, preserve_contents);
texture_cache.MarkColorBufferInUse(index);
}
+
View depth_surface;
- if (using_depth_fb || using_stencil_fb) {
- depth_surface = texture_cache.GetDepthBufferSurface();
+ if (using_depth_stencil) {
+ bool preserve_contents = false;
+ if (regs.clear_flags.scissor) {
+ // For depth stencil clears we only have to confirm scissor test covers the whole image.
+ const auto& scissor = regs.scissor_test[0];
+ preserve_contents |= scissor.min_x > 0;
+ preserve_contents |= scissor.min_y > 0;
+ preserve_contents |= scissor.max_x < regs.zeta_width;
+ preserve_contents |= scissor.max_y < regs.zeta_height;
+ }
+
+ depth_surface = texture_cache.GetDepthBufferSurface(preserve_contents);
texture_cache.MarkDepthBufferInUse();
}
texture_cache.GuardRenderTargets(false);
FramebufferCacheKey key;
- key.colors[0] = color_surface;
- key.zeta = depth_surface;
+ key.colors[0] = std::move(color_surface);
+ key.zeta = std::move(depth_surface);
state_tracker.NotifyFramebuffer();
glBindFramebuffer(GL_DRAW_FRAMEBUFFER, framebuffer_cache.GetFramebuffer(key));
@@ -422,8 +444,7 @@ void RasterizerOpenGL::Clear() {
if (regs.clear_buffers.R || regs.clear_buffers.G || regs.clear_buffers.B ||
regs.clear_buffers.A) {
use_color = true;
- }
- if (use_color) {
+
state_tracker.NotifyColorMask0();
glColorMaski(0, regs.clear_buffers.R != 0, regs.clear_buffers.G != 0,
regs.clear_buffers.B != 0, regs.clear_buffers.A != 0);
@@ -461,7 +482,7 @@ void RasterizerOpenGL::Clear() {
UNIMPLEMENTED_IF(regs.clear_flags.viewport);
- ConfigureClearFramebuffer(use_color, use_depth, use_stencil);
+ ConfigureClearFramebuffer(use_color, use_depth || use_stencil);
if (use_color) {
glClearBufferfv(GL_COLOR, 0, regs.clear_color);
@@ -833,9 +854,9 @@ void RasterizerOpenGL::SetupDrawGlobalMemory(std::size_t stage_index, const Shad
u32 binding = device.GetBaseBindings(stage_index).shader_storage_buffer;
for (const auto& entry : shader->GetEntries().global_memory_entries) {
- const auto addr{cbufs.const_buffers[entry.GetCbufIndex()].address + entry.GetCbufOffset()};
- const auto gpu_addr{memory_manager.Read<u64>(addr)};
- const auto size{memory_manager.Read<u32>(addr + 8)};
+ const GPUVAddr addr{cbufs.const_buffers[entry.cbuf_index].address + entry.cbuf_offset};
+ const GPUVAddr gpu_addr{memory_manager.Read<u64>(addr)};
+ const u32 size{memory_manager.Read<u32>(addr + 8)};
SetupGlobalMemory(binding++, entry, gpu_addr, size);
}
}
@@ -847,7 +868,7 @@ void RasterizerOpenGL::SetupComputeGlobalMemory(const Shader& kernel) {
u32 binding = 0;
for (const auto& entry : kernel->GetEntries().global_memory_entries) {
- const auto addr{cbufs[entry.GetCbufIndex()].Address() + entry.GetCbufOffset()};
+ const auto addr{cbufs[entry.cbuf_index].Address() + entry.cbuf_offset};
const auto gpu_addr{memory_manager.Read<u64>(addr)};
const auto size{memory_manager.Read<u32>(addr + 8)};
SetupGlobalMemory(binding++, entry, gpu_addr, size);
@@ -858,7 +879,7 @@ void RasterizerOpenGL::SetupGlobalMemory(u32 binding, const GlobalMemoryEntry& e
GPUVAddr gpu_addr, std::size_t size) {
const auto alignment{device.GetShaderStorageBufferAlignment()};
const auto [ssbo, buffer_offset] =
- buffer_cache.UploadMemory(gpu_addr, size, alignment, entry.IsWritten());
+ buffer_cache.UploadMemory(gpu_addr, size, alignment, entry.is_written);
glBindBufferRange(GL_SHADER_STORAGE_BUFFER, binding, ssbo, buffer_offset,
static_cast<GLsizeiptr>(size));
}
@@ -869,7 +890,7 @@ void RasterizerOpenGL::SetupDrawTextures(std::size_t stage_index, const Shader&
u32 binding = device.GetBaseBindings(stage_index).sampler;
for (const auto& entry : shader->GetEntries().samplers) {
const auto shader_type = static_cast<ShaderType>(stage_index);
- for (std::size_t i = 0; i < entry.Size(); ++i) {
+ for (std::size_t i = 0; i < entry.size; ++i) {
const auto texture = GetTextureInfo(maxwell3d, entry, shader_type, i);
SetupTexture(binding++, texture, entry);
}
@@ -881,7 +902,7 @@ void RasterizerOpenGL::SetupComputeTextures(const Shader& kernel) {
const auto& compute = system.GPU().KeplerCompute();
u32 binding = 0;
for (const auto& entry : kernel->GetEntries().samplers) {
- for (std::size_t i = 0; i < entry.Size(); ++i) {
+ for (std::size_t i = 0; i < entry.size; ++i) {
const auto texture = GetTextureInfo(compute, entry, ShaderType::Compute, i);
SetupTexture(binding++, texture, entry);
}
@@ -938,7 +959,7 @@ void RasterizerOpenGL::SetupImage(u32 binding, const Tegra::Texture::TICEntry& t
if (!tic.IsBuffer()) {
view->ApplySwizzle(tic.x_source, tic.y_source, tic.z_source, tic.w_source);
}
- if (entry.IsWritten()) {
+ if (entry.is_written) {
view->MarkAsModified(texture_cache.Tick());
}
glBindImageTexture(binding, view->GetTexture(), 0, GL_TRUE, 0, GL_READ_WRITE,
@@ -999,11 +1020,7 @@ void RasterizerOpenGL::SyncDepthClamp() {
}
flags[Dirty::DepthClampEnabled] = false;
- const auto& state = gpu.regs.view_volume_clip_control;
- UNIMPLEMENTED_IF_MSG(state.depth_clamp_far != state.depth_clamp_near,
- "Unimplemented depth clamp separation!");
-
- oglEnable(GL_DEPTH_CLAMP, state.depth_clamp_far || state.depth_clamp_near);
+ oglEnable(GL_DEPTH_CLAMP, gpu.regs.view_volume_clip_control.depth_clamp_disabled == 0);
}
void RasterizerOpenGL::SyncClipEnabled(u32 clip_mask) {
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h
index ebd2173eb..87249fb6f 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.h
+++ b/src/video_core/renderer_opengl/gl_rasterizer.h
@@ -95,7 +95,8 @@ private:
/// Configures the color and depth framebuffer states.
void ConfigureFramebuffers();
- void ConfigureClearFramebuffer(bool using_color_fb, bool using_depth_fb, bool using_stencil_fb);
+ /// Configures the color and depth framebuffer for clearing.
+ void ConfigureClearFramebuffer(bool using_color, bool using_depth_stencil);
/// Configures the current constbuffers to use for the draw command.
void SetupDrawConstBuffers(std::size_t stage_index, const Shader& shader);
diff --git a/src/video_core/renderer_opengl/gl_shader_cache.cpp b/src/video_core/renderer_opengl/gl_shader_cache.cpp
index f63156b8d..9759a7078 100644
--- a/src/video_core/renderer_opengl/gl_shader_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp
@@ -10,8 +10,6 @@
#include <thread>
#include <unordered_set>
-#include <boost/functional/hash.hpp>
-
#include "common/alignment.h"
#include "common/assert.h"
#include "common/logging/log.h"
@@ -28,76 +26,26 @@
#include "video_core/renderer_opengl/gl_shader_disk_cache.h"
#include "video_core/renderer_opengl/gl_state_tracker.h"
#include "video_core/renderer_opengl/utils.h"
+#include "video_core/shader/memory_util.h"
#include "video_core/shader/registry.h"
#include "video_core/shader/shader_ir.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 u32 STAGE_MAIN_OFFSET = 10;
-constexpr u32 KERNEL_MAIN_OFFSET = 0;
-
constexpr VideoCommon::Shader::CompilerSettings COMPILER_SETTINGS{};
-/// Gets the address for the specified shader stage program
-GPUVAddr GetShaderAddress(Core::System& system, Maxwell::ShaderProgram program) {
- const auto& gpu{system.GPU().Maxwell3D()};
- const auto& shader_config{gpu.regs.shader_config[static_cast<std::size_t>(program)]};
- return gpu.regs.code_address.CodeAddress() + shader_config.offset;
-}
-
-/// Gets if the current instruction offset is a scheduler instruction
-constexpr 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;
-}
-
-/// Calculates the size of a program stream
-std::size_t CalculateProgramSize(const ProgramCode& program) {
- constexpr std::size_t start_offset = 10;
- // This is the encoded version of BRA that jumps to itself. All Nvidia
- // shaders end with one.
- constexpr u64 self_jumping_branch = 0xE2400FFFFF07000FULL;
- constexpr u64 mask = 0xFFFFFFFFFF7FFFFFULL;
- 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());
-}
-
-/// Gets the shader program code from memory for the specified address
-ProgramCode GetShaderCode(Tegra::MemoryManager& memory_manager, const GPUVAddr gpu_addr,
- const u8* host_ptr) {
- ProgramCode code(VideoCommon::Shader::MAX_PROGRAM_LENGTH);
- ASSERT_OR_EXECUTE(host_ptr != nullptr, {
- std::fill(code.begin(), code.end(), 0);
- return code;
- });
- memory_manager.ReadBlockUnsafe(gpu_addr, code.data(), code.size() * sizeof(u64));
- code.resize(CalculateProgramSize(code));
- return code;
-}
-
/// Gets the shader type from a Maxwell program type
constexpr GLenum GetGLShaderType(ShaderType shader_type) {
switch (shader_type) {
@@ -114,17 +62,6 @@ constexpr GLenum GetGLShaderType(ShaderType shader_type) {
}
}
-/// Hashes one (or two) program streams
-u64 GetUniqueIdentifier(ShaderType shader_type, bool is_a, const ProgramCode& code,
- const ProgramCode& code_b = {}) {
- u64 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 unique_identifier;
-}
-
constexpr const char* GetShaderTypeName(ShaderType shader_type) {
switch (shader_type) {
case ShaderType::Vertex:
@@ -456,11 +393,12 @@ Shader ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) {
const auto host_ptr{memory_manager.GetPointer(address)};
// No shader found - create a new one
- ProgramCode code{GetShaderCode(memory_manager, address, host_ptr)};
+ ProgramCode code{GetShaderCode(memory_manager, address, host_ptr, false)};
ProgramCode code_b;
if (program == Maxwell::ShaderProgram::VertexA) {
const GPUVAddr address_b{GetShaderAddress(system, Maxwell::ShaderProgram::VertexB)};
- code_b = GetShaderCode(memory_manager, address_b, memory_manager.GetPointer(address_b));
+ const u8* host_ptr_b = memory_manager.GetPointer(address_b);
+ code_b = GetShaderCode(memory_manager, address_b, host_ptr_b, false);
}
const auto unique_identifier = GetUniqueIdentifier(
@@ -498,7 +436,7 @@ Shader ShaderCacheOpenGL::GetComputeKernel(GPUVAddr code_addr) {
const auto host_ptr{memory_manager.GetPointer(code_addr)};
// No kernel found, create a new one
- auto code{GetShaderCode(memory_manager, code_addr, host_ptr)};
+ auto code{GetShaderCode(memory_manager, code_addr, host_ptr, true)};
const auto unique_identifier{GetUniqueIdentifier(ShaderType::Compute, false, code)};
const ShaderParameters params{system, disk_cache, device,
diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
index 0cd3ad7e1..99fd4ae2c 100644
--- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
@@ -870,13 +870,13 @@ private:
for (const auto& sampler : ir.GetSamplers()) {
const std::string name = GetSampler(sampler);
const std::string description = fmt::format("layout (binding = {}) uniform", binding);
- binding += sampler.IsIndexed() ? sampler.Size() : 1;
+ binding += sampler.is_indexed ? sampler.size : 1;
std::string sampler_type = [&]() {
- if (sampler.IsBuffer()) {
+ if (sampler.is_buffer) {
return "samplerBuffer";
}
- switch (sampler.GetType()) {
+ switch (sampler.type) {
case Tegra::Shader::TextureType::Texture1D:
return "sampler1D";
case Tegra::Shader::TextureType::Texture2D:
@@ -890,17 +890,17 @@ private:
return "sampler2D";
}
}();
- if (sampler.IsArray()) {
+ if (sampler.is_array) {
sampler_type += "Array";
}
- if (sampler.IsShadow()) {
+ if (sampler.is_shadow) {
sampler_type += "Shadow";
}
- if (!sampler.IsIndexed()) {
+ if (!sampler.is_indexed) {
code.AddLine("{} {} {};", description, sampler_type, name);
} else {
- code.AddLine("{} {} {}[{}];", description, sampler_type, name, sampler.Size());
+ code.AddLine("{} {} {}[{}];", description, sampler_type, name, sampler.size);
}
}
if (!ir.GetSamplers().empty()) {
@@ -946,14 +946,14 @@ private:
u32 binding = device.GetBaseBindings(stage).image;
for (const auto& image : ir.GetImages()) {
std::string qualifier = "coherent volatile";
- if (image.IsRead() && !image.IsWritten()) {
+ if (image.is_read && !image.is_written) {
qualifier += " readonly";
- } else if (image.IsWritten() && !image.IsRead()) {
+ } else if (image.is_written && !image.is_read) {
qualifier += " writeonly";
}
- const char* format = image.IsAtomic() ? "r32ui, " : "";
- const char* type_declaration = GetImageTypeDeclaration(image.GetType());
+ const char* format = image.is_atomic ? "r32ui, " : "";
+ const char* type_declaration = GetImageTypeDeclaration(image.type);
code.AddLine("layout ({}binding = {}) {} uniform uimage{} {};", format, binding++,
qualifier, type_declaration, GetImage(image));
}
@@ -1337,8 +1337,8 @@ private:
ASSERT(meta);
const std::size_t count = operation.GetOperandsCount();
- const bool has_array = meta->sampler.IsArray();
- const bool has_shadow = meta->sampler.IsShadow();
+ const bool has_array = meta->sampler.is_array;
+ const bool has_shadow = meta->sampler.is_shadow;
std::string expr = "texture" + function_suffix;
if (!meta->aoffi.empty()) {
@@ -1346,7 +1346,7 @@ private:
} else if (!meta->ptp.empty()) {
expr += "Offsets";
}
- if (!meta->sampler.IsIndexed()) {
+ if (!meta->sampler.is_indexed) {
expr += '(' + GetSampler(meta->sampler) + ", ";
} else {
expr += '(' + GetSampler(meta->sampler) + '[' + Visit(meta->index).AsUint() + "], ";
@@ -1870,6 +1870,14 @@ private:
return GenerateBinaryInfix(operation, ">=", Type::Bool, type, type);
}
+ Expression LogicalAddCarry(Operation operation) {
+ const std::string carry = code.GenerateTemporary();
+ code.AddLine("uint {};", carry);
+ code.AddLine("uaddCarry({}, {}, {});", VisitOperand(operation, 0).AsUint(),
+ VisitOperand(operation, 1).AsUint(), carry);
+ return {fmt::format("({} != 0)", carry), Type::Bool};
+ }
+
Expression LogicalFIsNan(Operation operation) {
return GenerateUnary(operation, "isnan", Type::Bool, Type::Float);
}
@@ -1974,7 +1982,7 @@ private:
std::string expr = GenerateTexture(
operation, "", {TextureOffset{}, TextureArgument{Type::Float, meta->bias}});
- if (meta->sampler.IsShadow()) {
+ if (meta->sampler.is_shadow) {
expr = "vec4(" + expr + ')';
}
return {expr + GetSwizzle(meta->element), Type::Float};
@@ -1986,7 +1994,7 @@ private:
std::string expr = GenerateTexture(
operation, "Lod", {TextureArgument{Type::Float, meta->lod}, TextureOffset{}});
- if (meta->sampler.IsShadow()) {
+ if (meta->sampler.is_shadow) {
expr = "vec4(" + expr + ')';
}
return {expr + GetSwizzle(meta->element), Type::Float};
@@ -1995,11 +2003,11 @@ private:
Expression TextureGather(Operation operation) {
const auto& meta = std::get<MetaTexture>(operation.GetMeta());
- const auto type = meta.sampler.IsShadow() ? Type::Float : Type::Int;
- const bool separate_dc = meta.sampler.IsShadow();
+ const auto type = meta.sampler.is_shadow ? Type::Float : Type::Int;
+ const bool separate_dc = meta.sampler.is_shadow;
std::vector<TextureIR> ir;
- if (meta.sampler.IsShadow()) {
+ if (meta.sampler.is_shadow) {
ir = {TextureOffset{}};
} else {
ir = {TextureOffset{}, TextureArgument{type, meta.component}};
@@ -2044,7 +2052,7 @@ private:
constexpr std::array constructors = {"int", "ivec2", "ivec3", "ivec4"};
const auto meta = std::get_if<MetaTexture>(&operation.GetMeta());
ASSERT(meta);
- UNIMPLEMENTED_IF(meta->sampler.IsArray());
+ UNIMPLEMENTED_IF(meta->sampler.is_array);
const std::size_t count = operation.GetOperandsCount();
std::string expr = "texelFetch(";
@@ -2065,7 +2073,7 @@ private:
}
expr += ')';
- if (meta->lod && !meta->sampler.IsBuffer()) {
+ if (meta->lod && !meta->sampler.is_buffer) {
expr += ", ";
expr += Visit(meta->lod).AsInt();
}
@@ -2076,12 +2084,10 @@ private:
}
Expression TextureGradient(Operation operation) {
- const auto meta = std::get_if<MetaTexture>(&operation.GetMeta());
- ASSERT(meta);
-
+ const auto& meta = std::get<MetaTexture>(operation.GetMeta());
std::string expr =
GenerateTexture(operation, "Grad", {TextureDerivates{}, TextureOffset{}});
- return {std::move(expr) + GetSwizzle(meta->element), Type::Float};
+ return {std::move(expr) + GetSwizzle(meta.element), Type::Float};
}
Expression ImageLoad(Operation operation) {
@@ -2441,6 +2447,8 @@ private:
&GLSLDecompiler::LogicalNotEqual<Type::Uint>,
&GLSLDecompiler::LogicalGreaterEqual<Type::Uint>,
+ &GLSLDecompiler::LogicalAddCarry,
+
&GLSLDecompiler::Logical2HLessThan<false>,
&GLSLDecompiler::Logical2HEqual<false>,
&GLSLDecompiler::Logical2HLessEqual<false>,
@@ -2598,11 +2606,11 @@ private:
}
std::string GetSampler(const Sampler& sampler) const {
- return AppendSuffix(static_cast<u32>(sampler.GetIndex()), "sampler");
+ return AppendSuffix(sampler.index, "sampler");
}
std::string GetImage(const Image& image) const {
- return AppendSuffix(static_cast<u32>(image.GetIndex()), "image");
+ return AppendSuffix(image.index, "image");
}
std::string AppendSuffix(u32 index, std::string_view name) const {
diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.h b/src/video_core/renderer_opengl/gl_shader_decompiler.h
index e7dbd810c..e8a178764 100644
--- a/src/video_core/renderer_opengl/gl_shader_decompiler.h
+++ b/src/video_core/renderer_opengl/gl_shader_decompiler.h
@@ -33,36 +33,19 @@ public:
}
private:
- u32 index{};
+ u32 index = 0;
};
-class GlobalMemoryEntry {
-public:
- explicit GlobalMemoryEntry(u32 cbuf_index, u32 cbuf_offset, bool is_read, bool is_written)
+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 GetCbufIndex() const {
- return cbuf_index;
- }
-
- u32 GetCbufOffset() const {
- return cbuf_offset;
- }
-
- bool IsRead() const {
- return is_read;
- }
-
- bool IsWritten() const {
- return is_written;
- }
-
-private:
- u32 cbuf_index{};
- u32 cbuf_offset{};
- bool is_read{};
- bool is_written{};
+ u32 cbuf_index = 0;
+ u32 cbuf_offset = 0;
+ bool is_read = false;
+ bool is_written = false;
};
struct ShaderEntries {
diff --git a/src/video_core/renderer_vulkan/fixed_pipeline_state.cpp b/src/video_core/renderer_vulkan/fixed_pipeline_state.cpp
index a7f256ff9..648b1e71b 100644
--- a/src/video_core/renderer_vulkan/fixed_pipeline_state.cpp
+++ b/src/video_core/renderer_vulkan/fixed_pipeline_state.cpp
@@ -81,7 +81,7 @@ void FixedPipelineState::Rasterizer::Fill(const Maxwell& regs) noexcept {
primitive_restart_enable.Assign(regs.primitive_restart.enabled != 0 ? 1 : 0);
cull_enable.Assign(regs.cull_test_enabled != 0 ? 1 : 0);
depth_bias_enable.Assign(enabled_lut[POLYGON_OFFSET_ENABLE_LUT[topology_index]] != 0 ? 1 : 0);
- depth_clamp_enable.Assign(clip.depth_clamp_near == 1 || clip.depth_clamp_far == 1 ? 1 : 0);
+ depth_clamp_disabled.Assign(regs.view_volume_clip_control.depth_clamp_disabled.Value());
ndc_minus_one_to_one.Assign(regs.depth_mode == Maxwell::DepthMode::MinusOneToOne ? 1 : 0);
cull_face.Assign(PackCullFace(regs.cull_face));
front_face.Assign(packed_front_face);
diff --git a/src/video_core/renderer_vulkan/fixed_pipeline_state.h b/src/video_core/renderer_vulkan/fixed_pipeline_state.h
index 77188b862..8652067a7 100644
--- a/src/video_core/renderer_vulkan/fixed_pipeline_state.h
+++ b/src/video_core/renderer_vulkan/fixed_pipeline_state.h
@@ -153,7 +153,7 @@ struct FixedPipelineState {
BitField<4, 1, u32> primitive_restart_enable;
BitField<5, 1, u32> cull_enable;
BitField<6, 1, u32> depth_bias_enable;
- BitField<7, 1, u32> depth_clamp_enable;
+ BitField<7, 1, u32> depth_clamp_disabled;
BitField<8, 1, u32> ndc_minus_one_to_one;
BitField<9, 2, u32> cull_face;
BitField<11, 1, u32> front_face;
diff --git a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp
index 45bd1fc6c..852a17a70 100644
--- a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp
+++ b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp
@@ -249,7 +249,7 @@ vk::Pipeline VKGraphicsPipeline::CreatePipeline(const RenderPassParams& renderpa
rasterization_ci.sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO;
rasterization_ci.pNext = nullptr;
rasterization_ci.flags = 0;
- rasterization_ci.depthClampEnable = rs.depth_clamp_enable;
+ rasterization_ci.depthClampEnable = rs.depth_clamp_disabled == 0 ? VK_TRUE : VK_FALSE;
rasterization_ci.rasterizerDiscardEnable = VK_FALSE;
rasterization_ci.polygonMode = VK_POLYGON_MODE_FILL;
rasterization_ci.cullMode =
diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
index e6d4adc92..8fbd63dbc 100644
--- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
@@ -27,12 +27,18 @@
#include "video_core/renderer_vulkan/vk_update_descriptor.h"
#include "video_core/renderer_vulkan/wrapper.h"
#include "video_core/shader/compiler_settings.h"
+#include "video_core/shader/memory_util.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 {
@@ -45,60 +51,6 @@ constexpr VkDescriptorType STORAGE_IMAGE = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE;
constexpr VideoCommon::Shader::CompilerSettings compiler_settings{
VideoCommon::Shader::CompileDepth::FullDecompile};
-/// Gets the address for the specified shader stage program
-GPUVAddr GetShaderAddress(Core::System& system, Maxwell::ShaderProgram program) {
- const auto& gpu{system.GPU().Maxwell3D()};
- const auto& shader_config{gpu.regs.shader_config[static_cast<std::size_t>(program)]};
- return gpu.regs.code_address.CodeAddress() + shader_config.offset;
-}
-
-/// Gets if the current instruction offset is a scheduler instruction
-constexpr 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;
-}
-
-/// Calculates the size of a program stream
-std::size_t CalculateProgramSize(const ProgramCode& program, bool is_compute) {
- const std::size_t start_offset = is_compute ? 0 : 10;
- // This is the encoded version of BRA that jumps to itself. All Nvidia
- // shaders end with one.
- constexpr u64 self_jumping_branch = 0xE2400FFFFF07000FULL;
- constexpr u64 mask = 0xFFFFFFFFFF7FFFFFULL;
- 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());
-}
-
-/// Gets the shader program code from memory for the specified address
-ProgramCode GetShaderCode(Tegra::MemoryManager& memory_manager, const GPUVAddr gpu_addr,
- const u8* host_ptr, bool is_compute) {
- ProgramCode program_code(VideoCommon::Shader::MAX_PROGRAM_LENGTH);
- ASSERT_OR_EXECUTE(host_ptr != nullptr, {
- std::fill(program_code.begin(), program_code.end(), 0);
- return program_code;
- });
- memory_manager.ReadBlockUnsafe(gpu_addr, program_code.data(),
- program_code.size() * sizeof(u64));
- program_code.resize(CalculateProgramSize(program_code, is_compute));
- return program_code;
-}
-
constexpr std::size_t GetStageFromProgram(std::size_t program) {
return program == 0 ? 0 : program - 1;
}
@@ -133,7 +85,7 @@ void AddBindings(std::vector<VkDescriptorSetLayoutBinding>& bindings, u32& bindi
u32 count = 1;
if constexpr (descriptor_type == VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER) {
// Combined image samplers can be arrayed.
- count = container[i].Size();
+ count = container[i].size;
}
VkDescriptorSetLayoutBinding& entry = bindings.emplace_back();
entry.binding = binding++;
@@ -230,9 +182,9 @@ std::array<Shader, Maxwell::MaxShaderProgram> VKPipelineCache::GetShaders() {
const auto host_ptr{memory_manager.GetPointer(program_addr)};
// No shader found - create a new one
- constexpr u32 stage_offset = 10;
+ constexpr u32 stage_offset = STAGE_MAIN_OFFSET;
const auto stage = static_cast<Tegra::Engines::ShaderType>(index == 0 ? 0 : index - 1);
- auto code = GetShaderCode(memory_manager, program_addr, host_ptr, false);
+ ProgramCode code = GetShaderCode(memory_manager, program_addr, host_ptr, false);
shader = std::make_shared<CachedShader>(system, stage, program_addr, *cpu_addr,
std::move(code), stage_offset);
@@ -288,11 +240,10 @@ VKComputePipeline& VKPipelineCache::GetComputePipeline(const ComputePipelineCach
// No shader found - create a new one
const auto host_ptr = memory_manager.GetPointer(program_addr);
- auto code = GetShaderCode(memory_manager, program_addr, host_ptr, true);
- constexpr u32 kernel_main_offset = 0;
+ ProgramCode code = GetShaderCode(memory_manager, program_addr, host_ptr, true);
shader = std::make_shared<CachedShader>(system, Tegra::Engines::ShaderType::Compute,
program_addr, *cpu_addr, std::move(code),
- kernel_main_offset);
+ KERNEL_MAIN_OFFSET);
if (cpu_addr) {
Register(shader);
} else {
@@ -411,7 +362,7 @@ void AddEntry(std::vector<VkDescriptorUpdateTemplateEntry>& template_entries, u3
if constexpr (descriptor_type == COMBINED_IMAGE_SAMPLER) {
for (u32 i = 0; i < count; ++i) {
- const u32 num_samplers = container[i].Size();
+ const u32 num_samplers = container[i].size;
VkDescriptorUpdateTemplateEntry& entry = template_entries.emplace_back();
entry.dstBinding = binding;
entry.dstArrayElement = 0;
diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.h b/src/video_core/renderer_vulkan/vk_pipeline_cache.h
index 84d26b822..ebddafb73 100644
--- a/src/video_core/renderer_vulkan/vk_pipeline_cache.h
+++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.h
@@ -24,6 +24,7 @@
#include "video_core/renderer_vulkan/vk_resource_manager.h"
#include "video_core/renderer_vulkan/vk_shader_decompiler.h"
#include "video_core/renderer_vulkan/wrapper.h"
+#include "video_core/shader/memory_util.h"
#include "video_core/shader/registry.h"
#include "video_core/shader/shader_ir.h"
#include "video_core/surface.h"
@@ -46,8 +47,6 @@ class CachedShader;
using Shader = std::shared_ptr<CachedShader>;
using Maxwell = Tegra::Engines::Maxwell3D::Regs;
-using ProgramCode = std::vector<u64>;
-
struct GraphicsPipelineCacheKey {
FixedPipelineState fixed_state;
RenderPassParams renderpass_params;
@@ -108,7 +107,8 @@ namespace Vulkan {
class CachedShader final : public RasterizerCacheObject {
public:
explicit CachedShader(Core::System& system, Tegra::Engines::ShaderType stage, GPUVAddr gpu_addr,
- VAddr cpu_addr, ProgramCode program_code, u32 main_offset);
+ VAddr cpu_addr, VideoCommon::Shader::ProgramCode program_code,
+ u32 main_offset);
~CachedShader();
GPUVAddr GetGpuAddr() const {
@@ -140,7 +140,7 @@ private:
Tegra::Engines::ShaderType stage);
GPUVAddr gpu_addr{};
- ProgramCode program_code;
+ VideoCommon::Shader::ProgramCode program_code;
VideoCommon::Shader::Registry registry;
VideoCommon::Shader::ShaderIR shader_ir;
ShaderEntries entries;
diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
index d634aa4ea..127f77b2c 100644
--- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp
+++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
@@ -119,14 +119,13 @@ template <typename Engine, typename Entry>
Tegra::Texture::FullTextureInfo GetTextureInfo(const Engine& engine, const Entry& entry,
std::size_t stage, std::size_t index = 0) {
const auto stage_type = static_cast<Tegra::Engines::ShaderType>(stage);
- if (entry.IsBindless()) {
- const Tegra::Texture::TextureHandle tex_handle =
- engine.AccessConstBuffer32(stage_type, entry.GetBuffer(), entry.GetOffset());
+ if (entry.is_bindless) {
+ const auto tex_handle = engine.AccessConstBuffer32(stage_type, entry.buffer, entry.offset);
return engine.GetTextureInfo(tex_handle);
}
const auto& gpu_profile = engine.AccessGuestDriverProfile();
const u32 entry_offset = static_cast<u32>(index * gpu_profile.GetTextureHandlerSize());
- const u32 offset = entry.GetOffset() + entry_offset;
+ const u32 offset = entry.offset + entry_offset;
if constexpr (std::is_same_v<Engine, Tegra::Engines::Maxwell3D>) {
return engine.GetStageTexture(stage_type, offset);
} else {
@@ -656,7 +655,7 @@ RasterizerVulkan::Texceptions RasterizerVulkan::UpdateAttachments() {
Texceptions texceptions;
for (std::size_t rt = 0; rt < Maxwell::NumRenderTargets; ++rt) {
if (update_rendertargets) {
- color_attachments[rt] = texture_cache.GetColorBufferSurface(rt);
+ color_attachments[rt] = texture_cache.GetColorBufferSurface(rt, true);
}
if (color_attachments[rt] && WalkAttachmentOverlaps(*color_attachments[rt])) {
texceptions[rt] = true;
@@ -664,7 +663,7 @@ RasterizerVulkan::Texceptions RasterizerVulkan::UpdateAttachments() {
}
if (update_rendertargets) {
- zeta_attachment = texture_cache.GetDepthBufferSurface();
+ zeta_attachment = texture_cache.GetDepthBufferSurface(true);
}
if (zeta_attachment && WalkAttachmentOverlaps(*zeta_attachment)) {
texceptions[ZETA_TEXCEPTION_INDEX] = true;
@@ -974,7 +973,7 @@ void RasterizerVulkan::SetupGraphicsTextures(const ShaderEntries& entries, std::
MICROPROFILE_SCOPE(Vulkan_Textures);
const auto& gpu = system.GPU().Maxwell3D();
for (const auto& entry : entries.samplers) {
- for (std::size_t i = 0; i < entry.Size(); ++i) {
+ for (std::size_t i = 0; i < entry.size; ++i) {
const auto texture = GetTextureInfo(gpu, entry, stage, i);
SetupTexture(texture, entry);
}
@@ -1026,7 +1025,7 @@ void RasterizerVulkan::SetupComputeTextures(const ShaderEntries& entries) {
MICROPROFILE_SCOPE(Vulkan_Textures);
const auto& gpu = system.GPU().KeplerCompute();
for (const auto& entry : entries.samplers) {
- for (std::size_t i = 0; i < entry.Size(); ++i) {
+ for (std::size_t i = 0; i < entry.size; ++i) {
const auto texture = GetTextureInfo(gpu, entry, ComputeShaderIndex, i);
SetupTexture(texture, entry);
}
@@ -1108,7 +1107,7 @@ void RasterizerVulkan::SetupTexture(const Tegra::Texture::FullTextureInfo& textu
void RasterizerVulkan::SetupImage(const Tegra::Texture::TICEntry& tic, const ImageEntry& entry) {
auto view = texture_cache.GetImageSurface(tic, entry);
- if (entry.IsWritten()) {
+ if (entry.is_written) {
view->MarkAsModified(texture_cache.Tick());
}
diff --git a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
index aaa138f52..18678968c 100644
--- a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
+++ b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
@@ -103,8 +103,8 @@ struct GenericVaryingDescription {
};
spv::Dim GetSamplerDim(const Sampler& sampler) {
- ASSERT(!sampler.IsBuffer());
- switch (sampler.GetType()) {
+ ASSERT(!sampler.is_buffer);
+ switch (sampler.type) {
case Tegra::Shader::TextureType::Texture1D:
return spv::Dim::Dim1D;
case Tegra::Shader::TextureType::Texture2D:
@@ -114,13 +114,13 @@ spv::Dim GetSamplerDim(const Sampler& sampler) {
case Tegra::Shader::TextureType::TextureCube:
return spv::Dim::Cube;
default:
- UNIMPLEMENTED_MSG("Unimplemented sampler type={}", static_cast<u32>(sampler.GetType()));
+ UNIMPLEMENTED_MSG("Unimplemented sampler type={}", static_cast<int>(sampler.type));
return spv::Dim::Dim2D;
}
}
std::pair<spv::Dim, bool> GetImageDim(const Image& image) {
- switch (image.GetType()) {
+ switch (image.type) {
case Tegra::Shader::ImageType::Texture1D:
return {spv::Dim::Dim1D, false};
case Tegra::Shader::ImageType::TextureBuffer:
@@ -134,7 +134,7 @@ std::pair<spv::Dim, bool> GetImageDim(const Image& image) {
case Tegra::Shader::ImageType::Texture3D:
return {spv::Dim::Dim3D, false};
default:
- UNIMPLEMENTED_MSG("Unimplemented image type={}", static_cast<u32>(image.GetType()));
+ UNIMPLEMENTED_MSG("Unimplemented image type={}", static_cast<int>(image.type));
return {spv::Dim::Dim2D, false};
}
}
@@ -879,11 +879,11 @@ private:
u32 DeclareTexelBuffers(u32 binding) {
for (const auto& sampler : ir.GetSamplers()) {
- if (!sampler.IsBuffer()) {
+ if (!sampler.is_buffer) {
continue;
}
- ASSERT(!sampler.IsArray());
- ASSERT(!sampler.IsShadow());
+ ASSERT(!sampler.is_array);
+ ASSERT(!sampler.is_shadow);
constexpr auto dim = spv::Dim::Buffer;
constexpr int depth = 0;
@@ -894,23 +894,23 @@ private:
const Id image_type = TypeImage(t_float, dim, depth, arrayed, ms, sampled, format);
const Id pointer_type = TypePointer(spv::StorageClass::UniformConstant, image_type);
const Id id = OpVariable(pointer_type, spv::StorageClass::UniformConstant);
- AddGlobalVariable(Name(id, fmt::format("sampler_{}", sampler.GetIndex())));
+ AddGlobalVariable(Name(id, fmt::format("sampler_{}", sampler.index)));
Decorate(id, spv::Decoration::Binding, binding++);
Decorate(id, spv::Decoration::DescriptorSet, DESCRIPTOR_SET);
- texel_buffers.emplace(sampler.GetIndex(), TexelBuffer{image_type, id});
+ texel_buffers.emplace(sampler.index, TexelBuffer{image_type, id});
}
return binding;
}
u32 DeclareSamplers(u32 binding) {
for (const auto& sampler : ir.GetSamplers()) {
- if (sampler.IsBuffer()) {
+ if (sampler.is_buffer) {
continue;
}
const auto dim = GetSamplerDim(sampler);
- const int depth = sampler.IsShadow() ? 1 : 0;
- const int arrayed = sampler.IsArray() ? 1 : 0;
+ const int depth = sampler.is_shadow ? 1 : 0;
+ const int arrayed = sampler.is_array ? 1 : 0;
constexpr bool ms = false;
constexpr int sampled = 1;
constexpr auto format = spv::ImageFormat::Unknown;
@@ -918,17 +918,17 @@ private:
const Id sampler_type = TypeSampledImage(image_type);
const Id sampler_pointer_type =
TypePointer(spv::StorageClass::UniformConstant, sampler_type);
- const Id type = sampler.IsIndexed()
- ? TypeArray(sampler_type, Constant(t_uint, sampler.Size()))
+ const Id type = sampler.is_indexed
+ ? TypeArray(sampler_type, Constant(t_uint, sampler.size))
: sampler_type;
const Id pointer_type = TypePointer(spv::StorageClass::UniformConstant, type);
const Id id = OpVariable(pointer_type, spv::StorageClass::UniformConstant);
- AddGlobalVariable(Name(id, fmt::format("sampler_{}", sampler.GetIndex())));
+ AddGlobalVariable(Name(id, fmt::format("sampler_{}", sampler.index)));
Decorate(id, spv::Decoration::Binding, binding++);
Decorate(id, spv::Decoration::DescriptorSet, DESCRIPTOR_SET);
- sampled_images.emplace(sampler.GetIndex(), SampledImage{image_type, sampler_type,
- sampler_pointer_type, id});
+ sampled_images.emplace(
+ sampler.index, SampledImage{image_type, sampler_type, sampler_pointer_type, id});
}
return binding;
}
@@ -943,17 +943,17 @@ private:
const Id image_type = TypeImage(t_uint, dim, depth, arrayed, ms, sampled, format, {});
const Id pointer_type = TypePointer(spv::StorageClass::UniformConstant, image_type);
const Id id = OpVariable(pointer_type, spv::StorageClass::UniformConstant);
- AddGlobalVariable(Name(id, fmt::format("image_{}", image.GetIndex())));
+ AddGlobalVariable(Name(id, fmt::format("image_{}", image.index)));
Decorate(id, spv::Decoration::Binding, binding++);
Decorate(id, spv::Decoration::DescriptorSet, DESCRIPTOR_SET);
- if (image.IsRead() && !image.IsWritten()) {
+ if (image.is_read && !image.is_written) {
Decorate(id, spv::Decoration::NonWritable);
- } else if (image.IsWritten() && !image.IsRead()) {
+ } else if (image.is_written && !image.is_read) {
Decorate(id, spv::Decoration::NonReadable);
}
- images.emplace(static_cast<u32>(image.GetIndex()), StorageImage{image_type, id});
+ images.emplace(image.index, StorageImage{image_type, id});
}
return binding;
}
@@ -1584,6 +1584,15 @@ private:
return {OpCompositeConstruct(t_half, low, high), Type::HalfFloat};
}
+ Expression LogicalAddCarry(Operation operation) {
+ const Id op_a = AsUint(Visit(operation[0]));
+ const Id op_b = AsUint(Visit(operation[1]));
+
+ const Id result = OpIAddCarry(TypeStruct({t_uint, t_uint}), op_a, op_b);
+ const Id carry = OpCompositeExtract(t_uint, result, 1);
+ return {OpINotEqual(t_bool, carry, Constant(t_uint, 0)), Type::Bool};
+ }
+
Expression LogicalAssign(Operation operation) {
const Node& dest = operation[0];
const Node& src = operation[1];
@@ -1611,11 +1620,11 @@ private:
Id GetTextureSampler(Operation operation) {
const auto& meta = std::get<MetaTexture>(operation.GetMeta());
- ASSERT(!meta.sampler.IsBuffer());
+ ASSERT(!meta.sampler.is_buffer);
- const auto& entry = sampled_images.at(meta.sampler.GetIndex());
+ const auto& entry = sampled_images.at(meta.sampler.index);
Id sampler = entry.variable;
- if (meta.sampler.IsIndexed()) {
+ if (meta.sampler.is_indexed) {
const Id index = AsInt(Visit(meta.index));
sampler = OpAccessChain(entry.sampler_pointer_type, sampler, index);
}
@@ -1624,8 +1633,8 @@ private:
Id GetTextureImage(Operation operation) {
const auto& meta = std::get<MetaTexture>(operation.GetMeta());
- const u32 index = meta.sampler.GetIndex();
- if (meta.sampler.IsBuffer()) {
+ const u32 index = meta.sampler.index;
+ if (meta.sampler.is_buffer) {
const auto& entry = texel_buffers.at(index);
return OpLoad(entry.image_type, entry.image);
} else {
@@ -1636,7 +1645,7 @@ private:
Id GetImage(Operation operation) {
const auto& meta = std::get<MetaImage>(operation.GetMeta());
- const auto entry = images.at(meta.image.GetIndex());
+ const auto entry = images.at(meta.image.index);
return OpLoad(entry.image_type, entry.image);
}
@@ -1652,7 +1661,7 @@ private:
}
if (const auto meta = std::get_if<MetaTexture>(&operation.GetMeta())) {
// Add array coordinate for textures
- if (meta->sampler.IsArray()) {
+ if (meta->sampler.is_array) {
Id array = AsInt(Visit(meta->array));
if (type == Type::Float) {
array = OpConvertSToF(t_float, array);
@@ -1758,7 +1767,7 @@ private:
operands.push_back(GetOffsetCoordinates(operation));
}
- if (meta.sampler.IsShadow()) {
+ if (meta.sampler.is_shadow) {
const Id dref = AsFloat(Visit(meta.depth_compare));
return {OpImageSampleDrefExplicitLod(t_float, sampler, coords, dref, mask, operands),
Type::Float};
@@ -1773,7 +1782,7 @@ private:
const Id coords = GetCoordinates(operation, Type::Float);
Id texture{};
- if (meta.sampler.IsShadow()) {
+ if (meta.sampler.is_shadow) {
texture = OpImageDrefGather(t_float4, GetTextureSampler(operation), coords,
AsFloat(Visit(meta.depth_compare)));
} else {
@@ -1800,8 +1809,8 @@ private:
}
const Id lod = AsUint(Visit(operation[0]));
- const std::size_t coords_count = [&]() {
- switch (const auto type = meta.sampler.GetType(); type) {
+ const std::size_t coords_count = [&meta] {
+ switch (const auto type = meta.sampler.type) {
case Tegra::Shader::TextureType::Texture1D:
return 1;
case Tegra::Shader::TextureType::Texture2D:
@@ -1810,7 +1819,7 @@ private:
case Tegra::Shader::TextureType::Texture3D:
return 3;
default:
- UNREACHABLE_MSG("Invalid texture type={}", static_cast<u32>(type));
+ UNREACHABLE_MSG("Invalid texture type={}", static_cast<int>(type));
return 2;
}
}();
@@ -1853,7 +1862,7 @@ private:
const Id image = GetTextureImage(operation);
const Id coords = GetCoordinates(operation, Type::Int);
Id fetch;
- if (meta.lod && !meta.sampler.IsBuffer()) {
+ if (meta.lod && !meta.sampler.is_buffer) {
fetch = OpImageFetch(t_float4, image, coords, spv::ImageOperandsMask::Lod,
AsInt(Visit(meta.lod)));
} else {
@@ -2518,6 +2527,8 @@ private:
&SPIRVDecompiler::Binary<&Module::OpINotEqual, Type::Bool, Type::Uint>,
&SPIRVDecompiler::Binary<&Module::OpUGreaterThanEqual, Type::Bool, Type::Uint>,
+ &SPIRVDecompiler::LogicalAddCarry,
+
&SPIRVDecompiler::Binary<&Module::OpFOrdLessThan, Type::Bool2, Type::HalfFloat>,
&SPIRVDecompiler::Binary<&Module::OpFOrdEqual, Type::Bool2, Type::HalfFloat>,
&SPIRVDecompiler::Binary<&Module::OpFOrdLessThanEqual, Type::Bool2, Type::HalfFloat>,
@@ -2969,7 +2980,7 @@ ShaderEntries GenerateShaderEntries(const VideoCommon::Shader::ShaderIR& ir) {
entries.global_buffers.emplace_back(base.cbuf_index, base.cbuf_offset, usage.is_written);
}
for (const auto& sampler : ir.GetSamplers()) {
- if (sampler.IsBuffer()) {
+ if (sampler.is_buffer) {
entries.texel_buffers.emplace_back(sampler);
} else {
entries.samplers.emplace_back(sampler);
diff --git a/src/video_core/shader/control_flow.cpp b/src/video_core/shader/control_flow.cpp
index e00a3fb70..8d86020f6 100644
--- a/src/video_core/shader/control_flow.cpp
+++ b/src/video_core/shader/control_flow.cpp
@@ -13,6 +13,7 @@
#include "common/common_types.h"
#include "video_core/shader/ast.h"
#include "video_core/shader/control_flow.h"
+#include "video_core/shader/memory_util.h"
#include "video_core/shader/registry.h"
#include "video_core/shader/shader_ir.h"
@@ -115,17 +116,6 @@ Pred GetPredicate(u32 index, bool negated) {
return static_cast<Pred>(static_cast<u64>(index) + (negated ? 8ULL : 0ULL));
}
-/**
- * Returns whether the instruction at the specified offset is a 'sched' instruction.
- * Sched instructions always appear before a sequence of 3 instructions.
- */
-constexpr bool IsSchedInstruction(u32 offset, u32 main_offset) {
- constexpr u32 SchedPeriod = 4;
- u32 absolute_offset = offset - main_offset;
-
- return (absolute_offset % SchedPeriod) == 0;
-}
-
enum class ParseResult : u32 {
ControlCaught,
BlockEnd,
diff --git a/src/video_core/shader/decode.cpp b/src/video_core/shader/decode.cpp
index 87ac9ac6c..a75a5cc63 100644
--- a/src/video_core/shader/decode.cpp
+++ b/src/video_core/shader/decode.cpp
@@ -13,6 +13,7 @@
#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"
@@ -23,17 +24,6 @@ using Tegra::Shader::OpCode;
namespace {
-/**
- * Returns whether the instruction at the specified offset is a 'sched' instruction.
- * Sched instructions always appear before a sequence of 3 instructions.
- */
-constexpr bool IsSchedInstruction(u32 offset, u32 main_offset) {
- constexpr u32 SchedPeriod = 4;
- u32 absolute_offset = offset - main_offset;
-
- return (absolute_offset % SchedPeriod) == 0;
-}
-
void DeduceTextureHandlerSize(VideoCore::GuestDriverProfile& gpu_driver,
const std::list<Sampler>& used_samplers) {
if (gpu_driver.IsTextureHandlerSizeKnown() || used_samplers.size() <= 1) {
@@ -42,11 +32,11 @@ void DeduceTextureHandlerSize(VideoCore::GuestDriverProfile& gpu_driver,
u32 count{};
std::vector<u32> bound_offsets;
for (const auto& sampler : used_samplers) {
- if (sampler.IsBindless()) {
+ if (sampler.is_bindless) {
continue;
}
++count;
- bound_offsets.emplace_back(sampler.GetOffset());
+ bound_offsets.emplace_back(sampler.offset);
}
if (count > 1) {
gpu_driver.DeduceTextureHandlerSize(std::move(bound_offsets));
@@ -56,14 +46,14 @@ void DeduceTextureHandlerSize(VideoCore::GuestDriverProfile& gpu_driver,
std::optional<u32> TryDeduceSamplerSize(const Sampler& sampler_to_deduce,
VideoCore::GuestDriverProfile& gpu_driver,
const std::list<Sampler>& used_samplers) {
- const u32 base_offset = sampler_to_deduce.GetOffset();
+ const u32 base_offset = sampler_to_deduce.offset;
u32 max_offset{std::numeric_limits<u32>::max()};
for (const auto& sampler : used_samplers) {
- if (sampler.IsBindless()) {
+ if (sampler.is_bindless) {
continue;
}
- if (sampler.GetOffset() > base_offset) {
- max_offset = std::min(sampler.GetOffset(), max_offset);
+ if (sampler.offset > base_offset) {
+ max_offset = std::min(sampler.offset, max_offset);
}
}
if (max_offset == std::numeric_limits<u32>::max()) {
@@ -363,14 +353,14 @@ void ShaderIR::PostDecode() {
return;
}
for (auto& sampler : used_samplers) {
- if (!sampler.IsIndexed()) {
+ if (!sampler.is_indexed) {
continue;
}
if (const auto size = TryDeduceSamplerSize(sampler, gpu_driver, used_samplers)) {
- sampler.SetSize(*size);
+ sampler.size = *size;
} else {
LOG_CRITICAL(HW_GPU, "Failed to deduce size of indexed sampler");
- sampler.SetSize(1);
+ sampler.size = 1;
}
}
}
diff --git a/src/video_core/shader/decode/arithmetic_integer.cpp b/src/video_core/shader/decode/arithmetic_integer.cpp
index 9af8c606d..a041519b7 100644
--- a/src/video_core/shader/decode/arithmetic_integer.cpp
+++ b/src/video_core/shader/decode/arithmetic_integer.cpp
@@ -35,15 +35,38 @@ u32 ShaderIR::DecodeArithmeticInteger(NodeBlock& bb, u32 pc) {
case OpCode::Id::IADD_C:
case OpCode::Id::IADD_R:
case OpCode::Id::IADD_IMM: {
- UNIMPLEMENTED_IF_MSG(instr.alu.saturate_d, "IADD saturation not implemented");
+ 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);
- const Node value = Operation(OperationCode::IAdd, PRECISE, op_a, op_b);
+ Node value = Operation(OperationCode::UAdd, op_a, op_b);
- SetInternalFlagsFromInteger(bb, value, instr.generates_cc);
- SetRegister(bb, instr.gpr0, value);
+ 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:
diff --git a/src/video_core/shader/decode/image.cpp b/src/video_core/shader/decode/image.cpp
index 85ee9aa5e..60b6ad72a 100644
--- a/src/video_core/shader/decode/image.cpp
+++ b/src/video_core/shader/decode/image.cpp
@@ -485,11 +485,10 @@ u32 ShaderIR::DecodeImage(NodeBlock& bb, u32 pc) {
Image& 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 Image& entry) { return entry.GetOffset() == offset; });
+ const auto it = std::find_if(std::begin(used_images), std::end(used_images),
+ [offset](const Image& entry) { return entry.offset == offset; });
if (it != std::end(used_images)) {
- ASSERT(!it->IsBindless() && it->GetType() == it->GetType());
+ ASSERT(!it->is_bindless && it->type == type);
return *it;
}
@@ -505,13 +504,12 @@ Image& ShaderIR::GetBindlessImage(Tegra::Shader::Register reg, Tegra::Shader::Im
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 = buffer, offset = offset](const Image& entry) {
- return entry.GetBuffer() == buffer && entry.GetOffset() == offset;
- });
+ const auto it = std::find_if(std::begin(used_images), std::end(used_images),
+ [buffer, offset](const Image& entry) {
+ return entry.buffer == buffer && entry.offset == offset;
+ });
if (it != std::end(used_images)) {
- ASSERT(it->IsBindless() && it->GetType() == it->GetType());
+ ASSERT(it->is_bindless && it->type == type);
return *it;
}
diff --git a/src/video_core/shader/decode/register_set_predicate.cpp b/src/video_core/shader/decode/register_set_predicate.cpp
index 8d54cce34..6116c31aa 100644
--- a/src/video_core/shader/decode/register_set_predicate.cpp
+++ b/src/video_core/shader/decode/register_set_predicate.cpp
@@ -2,6 +2,8 @@
// 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"
@@ -10,20 +12,20 @@
namespace VideoCommon::Shader {
+using std::move;
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
namespace {
-constexpr u64 NUM_PROGRAMMABLE_PREDICATES = 7;
-}
+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);
- UNIMPLEMENTED_IF(instr.p2r_r2p.mode != Tegra::Shader::R2pMode::Pr);
-
- const Node apply_mask = [&] {
+ Node apply_mask = [this, opcode, instr] {
switch (opcode->get().GetId()) {
case OpCode::Id::R2P_IMM:
case OpCode::Id::P2R_IMM:
@@ -34,39 +36,43 @@ u32 ShaderIR::DecodeRegisterSetPredicate(NodeBlock& bb, u32 pc) {
}
}();
- const auto offset = static_cast<u32>(instr.p2r_r2p.byte) * 8;
+ 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: {
- const Node mask = GetRegister(instr.gpr8);
+ Node mask = GetRegister(instr.gpr8);
- for (u64 pred = 0; pred < NUM_PROGRAMMABLE_PREDICATES; ++pred) {
- const auto shift = static_cast<u32>(pred);
+ for (u64 entry = 0; entry < num_entries; ++entry) {
+ const u32 shift = static_cast<u32>(entry);
- const Node apply_compare = BitfieldExtract(apply_mask, shift, 1);
- const Node condition =
- Operation(OperationCode::LogicalUNotEqual, apply_compare, Immediate(0));
+ Node apply = BitfieldExtract(apply_mask, shift, 1);
+ Node condition = Operation(OperationCode::LogicalUNotEqual, apply, Immediate(0));
- const Node value_compare = BitfieldExtract(mask, offset + shift, 1);
- const Node value =
- Operation(OperationCode::LogicalUNotEqual, value_compare, Immediate(0));
+ Node compare = BitfieldExtract(mask, offset + shift, 1);
+ Node value = Operation(OperationCode::LogicalUNotEqual, move(compare), Immediate(0));
- const Node code = Operation(OperationCode::LogicalAssign, GetPredicate(pred), value);
- bb.push_back(Conditional(condition, {code}));
+ 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 pred = 0; pred < NUM_PROGRAMMABLE_PREDICATES; ++pred) {
- Node bit = Operation(OperationCode::Select, GetPredicate(pred), Immediate(1U << pred),
+ 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, std::move(value), std::move(bit));
+ value = Operation(OperationCode::UBitwiseOr, move(value), move(bit));
}
- value = Operation(OperationCode::UBitwiseAnd, std::move(value), apply_mask);
- value = BitfieldInsert(GetRegister(instr.gpr8), std::move(value), offset, 8);
+ value = Operation(OperationCode::UBitwiseAnd, move(value), apply_mask);
+ value = BitfieldInsert(GetRegister(instr.gpr8), move(value), offset, 8);
- SetRegister(bb, instr.gpr0, std::move(value));
+ SetRegister(bb, instr.gpr0, move(value));
break;
}
default:
diff --git a/src/video_core/shader/decode/texture.cpp b/src/video_core/shader/decode/texture.cpp
index e68f1d305..8f0bb996e 100644
--- a/src/video_core/shader/decode/texture.cpp
+++ b/src/video_core/shader/decode/texture.cpp
@@ -139,15 +139,15 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
}
const Node component = Immediate(static_cast<u32>(instr.tld4s.component));
- const SamplerInfo info{TextureType::Texture2D, false, is_depth_compare, false};
- const Sampler& sampler = *GetSampler(instr.sampler, info);
+ SamplerInfo info;
+ info.is_shadow = is_depth_compare;
+ const std::optional<Sampler> sampler = GetSampler(instr.sampler, info);
Node4 values;
for (u32 element = 0; element < values.size(); ++element) {
- auto coords_copy = coords;
- MetaTexture meta{sampler, {}, depth_compare, aoffi, {}, {},
- {}, {}, component, element, {}};
- values[element] = Operation(OperationCode::TextureGather, meta, std::move(coords_copy));
+ MetaTexture meta{*sampler, {}, depth_compare, aoffi, {}, {},
+ {}, {}, component, element, {}};
+ values[element] = Operation(OperationCode::TextureGather, meta, coords);
}
if (instr.tld4s.fp16_flag) {
@@ -165,18 +165,20 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
"AOFFI is not implemented");
const bool is_array = instr.txd.is_array != 0;
- u64 base_reg = instr.gpr8.Value();
const auto derivate_reg = instr.gpr20.Value();
const auto texture_type = instr.txd.texture_type.Value();
const auto coord_count = GetCoordCount(texture_type);
- Node index_var{};
- const Sampler* sampler =
- is_bindless
- ? GetBindlessSampler(base_reg, index_var, {{texture_type, is_array, false, false}})
- : GetSampler(instr.sampler, {{texture_type, is_array, false, false}});
+ u64 base_reg = instr.gpr8.Value();
+ Node index_var;
+ SamplerInfo info;
+ info.type = texture_type;
+ info.is_array = is_array;
+ const std::optional<Sampler> sampler = is_bindless
+ ? GetBindlessSampler(base_reg, info, index_var)
+ : GetSampler(instr.sampler, info);
Node4 values;
- if (sampler == nullptr) {
- std::generate(values.begin(), values.end(), [] { return Immediate(0); });
+ if (!sampler) {
+ std::generate(values.begin(), values.end(), [this] { return Immediate(0); });
WriteTexInstructionFloat(bb, instr, values);
break;
}
@@ -214,14 +216,12 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
is_bindless = true;
[[fallthrough]];
case OpCode::Id::TXQ: {
- // TODO: The new commits on the texture refactor, change the way samplers work.
- // Sadly, not all texture instructions specify the type of texture their sampler
- // uses. This must be fixed at a later instance.
- Node index_var{};
- const Sampler* sampler =
- is_bindless ? GetBindlessSampler(instr.gpr8, index_var) : GetSampler(instr.sampler);
-
- if (sampler == nullptr) {
+ Node index_var;
+ const std::optional<Sampler> 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)) {
@@ -267,12 +267,17 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
UNIMPLEMENTED_IF_MSG(instr.tmml.UsesMiscMode(Tegra::Shader::TextureMiscMode::NDV),
"NDV is not implemented");
- auto texture_type = instr.tmml.texture_type.Value();
- Node index_var{};
- const Sampler* sampler =
- is_bindless ? GetBindlessSampler(instr.gpr20, index_var) : GetSampler(instr.sampler);
-
- if (sampler == nullptr) {
+ 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<Sampler> 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)) {
@@ -299,12 +304,11 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
coords.push_back(GetRegister(instr.gpr8.Value() + 1));
break;
default:
- UNIMPLEMENTED_MSG("Unhandled texture type {}", static_cast<u32>(texture_type));
+ UNIMPLEMENTED_MSG("Unhandled texture type {}", static_cast<int>(texture_type));
// Fallback to interpreting as a 2D texture for now
coords.push_back(GetRegister(instr.gpr8.Value() + 0));
coords.push_back(GetRegister(instr.gpr8.Value() + 1));
- texture_type = TextureType::Texture2D;
}
u32 indexer = 0;
for (u32 element = 0; element < 2; ++element) {
@@ -353,98 +357,103 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
return pc;
}
-ShaderIR::SamplerInfo ShaderIR::GetSamplerInfo(std::optional<SamplerInfo> sampler_info, u32 offset,
+ShaderIR::SamplerInfo ShaderIR::GetSamplerInfo(SamplerInfo info, u32 offset,
std::optional<u32> buffer) {
- if (sampler_info) {
- return *sampler_info;
+ if (info.IsComplete()) {
+ return info;
}
const auto sampler = buffer ? registry.ObtainBindlessSampler(*buffer, offset)
: registry.ObtainBoundSampler(offset);
if (!sampler) {
LOG_WARNING(HW_GPU, "Unknown sampler info");
- return SamplerInfo{TextureType::Texture2D, false, false, false};
- }
- return SamplerInfo{sampler->texture_type, sampler->is_array != 0, sampler->is_shadow != 0,
- sampler->is_buffer != 0};
+ 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;
}
-const Sampler* ShaderIR::GetSampler(const Tegra::Shader::Sampler& sampler,
- std::optional<SamplerInfo> sampler_info) {
+std::optional<Sampler> ShaderIR::GetSampler(Tegra::Shader::Sampler sampler,
+ SamplerInfo sampler_info) {
const auto offset = static_cast<u32>(sampler.index.Value());
const auto info = GetSamplerInfo(sampler_info, 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 Sampler& entry) { return entry.GetOffset() == offset; });
+ const auto it = std::find_if(used_samplers.begin(), used_samplers.end(),
+ [offset](const Sampler& entry) { return entry.offset == offset; });
if (it != used_samplers.end()) {
- ASSERT(!it->IsBindless() && it->GetType() == info.type && it->IsArray() == info.is_array &&
- it->IsShadow() == info.is_shadow && it->IsBuffer() == info.is_buffer);
- return &*it;
+ 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);
+ return used_samplers.emplace_back(next_index, offset, *info.type, *info.is_array,
+ *info.is_shadow, *info.is_buffer, false);
}
-const Sampler* ShaderIR::GetBindlessSampler(Tegra::Shader::Register reg, Node& index_var,
- std::optional<SamplerInfo> sampler_info) {
+std::optional<Sampler> 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()));
ASSERT(base_node != nullptr);
if (base_node == nullptr) {
- return nullptr;
+ return std::nullopt;
}
if (const auto bindless_sampler_info =
std::get_if<BindlessSamplerNode>(&*tracked_sampler_info)) {
const u32 buffer = bindless_sampler_info->GetIndex();
const u32 offset = bindless_sampler_info->GetOffset();
- const auto info = GetSamplerInfo(sampler_info, offset, buffer);
+ info = GetSamplerInfo(info, offset, buffer);
// If this sampler has already been used, return the existing mapping.
- const auto it =
- std::find_if(used_samplers.begin(), used_samplers.end(),
- [buffer = buffer, offset = offset](const Sampler& entry) {
- return entry.GetBuffer() == buffer && entry.GetOffset() == offset;
- });
+ const auto it = std::find_if(used_samplers.begin(), used_samplers.end(),
+ [buffer = buffer, offset = offset](const Sampler& entry) {
+ return entry.buffer == buffer && entry.offset == offset;
+ });
if (it != used_samplers.end()) {
- ASSERT(it->IsBindless() && it->GetType() == info.type &&
- it->IsArray() == info.is_array && it->IsShadow() == info.is_shadow);
- return &*it;
+ 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);
- } else if (const auto array_sampler_info =
- std::get_if<ArraySamplerNode>(&*tracked_sampler_info)) {
+ return used_samplers.emplace_back(next_index, offset, buffer, *info.type, *info.is_array,
+ *info.is_shadow, *info.is_buffer, false);
+ }
+ if (const auto array_sampler_info = std::get_if<ArraySamplerNode>(&*tracked_sampler_info)) {
const u32 base_offset = array_sampler_info->GetBaseOffset() / 4;
index_var = GetCustomVariable(array_sampler_info->GetIndexVar());
- const auto info = GetSamplerInfo(sampler_info, base_offset);
+ info = GetSamplerInfo(info, 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 Sampler& entry) { return entry.GetOffset() == base_offset; });
+ [base_offset](const Sampler& entry) { return entry.offset == base_offset; });
if (it != used_samplers.end()) {
- ASSERT(!it->IsBindless() && it->GetType() == info.type &&
- it->IsArray() == info.is_array && it->IsShadow() == info.is_shadow &&
- it->IsBuffer() == info.is_buffer && it->IsIndexed());
- return &*it;
+ 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 used_samplers.emplace_back(next_index, base_offset, *info.type, *info.is_array,
+ *info.is_shadow, *info.is_buffer, true);
}
- return nullptr;
+ return std::nullopt;
}
void ShaderIR::WriteTexInstructionFloat(NodeBlock& bb, Instruction instr, const Node4& components) {
@@ -529,10 +538,16 @@ Node4 ShaderIR::GetTextureCode(Instruction instr, TextureType texture_type,
ASSERT_MSG(texture_type != TextureType::Texture3D || !is_array || !is_shadow,
"Illegal texture type");
- const SamplerInfo info{texture_type, is_array, is_shadow, false};
+ SamplerInfo info;
+ info.type = texture_type;
+ info.is_array = is_array;
+ info.is_shadow = is_shadow;
+ info.is_buffer = false;
+
Node index_var;
- const Sampler* sampler = is_bindless ? GetBindlessSampler(*bindless_reg, index_var, info)
- : GetSampler(instr.sampler, info);
+ const std::optional<Sampler> sampler = is_bindless
+ ? GetBindlessSampler(*bindless_reg, info, index_var)
+ : GetSampler(instr.sampler, info);
if (!sampler) {
return {Immediate(0), Immediate(0), Immediate(0), Immediate(0)};
}
@@ -683,12 +698,17 @@ Node4 ShaderIR::GetTld4Code(Instruction instr, TextureType texture_type, bool de
u64 parameter_register = instr.gpr20.Value();
- const SamplerInfo info{texture_type, is_array, depth_compare, false};
- Node index_var{};
- const Sampler* sampler = is_bindless ? GetBindlessSampler(parameter_register++, index_var, info)
- : GetSampler(instr.sampler, info);
+ SamplerInfo info;
+ info.type = texture_type;
+ info.is_array = is_array;
+ info.is_shadow = depth_compare;
+
+ Node index_var;
+ const std::optional<Sampler> sampler =
+ is_bindless ? GetBindlessSampler(parameter_register++, info, index_var)
+ : GetSampler(instr.sampler, info);
Node4 values;
- if (sampler == nullptr) {
+ if (!sampler) {
for (u32 element = 0; element < values.size(); ++element) {
values[element] = Immediate(0);
}
@@ -743,12 +763,12 @@ Node4 ShaderIR::GetTldCode(Tegra::Shader::Instruction instr) {
// const Node aoffi_register{is_aoffi ? GetRegister(gpr20_cursor++) : nullptr};
// const Node multisample{is_multisample ? GetRegister(gpr20_cursor++) : nullptr};
- const auto& sampler = *GetSampler(instr.sampler);
+ const std::optional<Sampler> 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, {}};
+ MetaTexture meta{*sampler, array_register, {}, {}, {}, {}, {}, lod, {}, element, {}};
values[element] = Operation(OperationCode::TexelFetch, meta, std::move(coords_copy));
}
@@ -756,7 +776,11 @@ Node4 ShaderIR::GetTldCode(Tegra::Shader::Instruction instr) {
}
Node4 ShaderIR::GetTldsCode(Instruction instr, TextureType texture_type, bool is_array) {
- const Sampler& sampler = *GetSampler(instr.sampler);
+ SamplerInfo info;
+ info.type = texture_type;
+ info.is_array = is_array;
+ info.is_shadow = false;
+ const std::optional<Sampler> sampler = GetSampler(instr.sampler, info);
const std::size_t type_coord_count = GetCoordCount(texture_type);
const bool lod_enabled = instr.tlds.GetTextureProcessMode() == TextureProcessMode::LL;
@@ -784,7 +808,7 @@ Node4 ShaderIR::GetTldsCode(Instruction instr, TextureType texture_type, bool is
Node4 values;
for (u32 element = 0; element < values.size(); ++element) {
auto coords_copy = coords;
- MetaTexture meta{sampler, array, {}, {}, {}, {}, {}, lod, {}, element, {}};
+ MetaTexture meta{*sampler, array, {}, {}, {}, {}, {}, lod, {}, element, {}};
values[element] = Operation(OperationCode::TexelFetch, meta, std::move(coords_copy));
}
return values;
diff --git a/src/video_core/shader/memory_util.cpp b/src/video_core/shader/memory_util.cpp
new file mode 100644
index 000000000..074f21691
--- /dev/null
+++ b/src/video_core/shader/memory_util.cpp
@@ -0,0 +1,77 @@
+// 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(Core::System& system,
+ Tegra::Engines::Maxwell3D::Regs::ShaderProgram program) {
+ const auto& gpu{system.GPU().Maxwell3D()};
+ const auto& shader_config{gpu.regs.shader_config[static_cast<std::size_t>(program)]};
+ return gpu.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) {
+ u64 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 unique_identifier;
+}
+
+} // namespace VideoCommon::Shader
diff --git a/src/video_core/shader/memory_util.h b/src/video_core/shader/memory_util.h
new file mode 100644
index 000000000..be90d24fd
--- /dev/null
+++ b/src/video_core/shader/memory_util.h
@@ -0,0 +1,47 @@
+// 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 Core {
+class System;
+}
+
+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(Core::System& system,
+ 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
diff --git a/src/video_core/shader/node.h b/src/video_core/shader/node.h
index 3eee961f5..601c822d2 100644
--- a/src/video_core/shader/node.h
+++ b/src/video_core/shader/node.h
@@ -132,6 +132,8 @@ enum class OperationCode {
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
@@ -265,76 +267,30 @@ class ArraySamplerNode;
using TrackSamplerData = std::variant<BindlessSamplerNode, ArraySamplerNode>;
using TrackSampler = std::shared_ptr<TrackSamplerData>;
-class Sampler {
-public:
- /// This constructor is for bound samplers
+struct Sampler {
+ /// Bound samplers constructor
constexpr explicit Sampler(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} {}
- /// This constructor is for bindless samplers
+ /// Bindless samplers constructor
constexpr explicit Sampler(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} {}
- constexpr u32 GetIndex() const {
- return index;
- }
-
- constexpr u32 GetOffset() const {
- return offset;
- }
-
- constexpr u32 GetBuffer() const {
- return buffer;
- }
-
- constexpr Tegra::Shader::TextureType GetType() const {
- return type;
- }
-
- constexpr bool IsArray() const {
- return is_array;
- }
-
- constexpr bool IsShadow() const {
- return is_shadow;
- }
-
- constexpr bool IsBuffer() const {
- return is_buffer;
- }
-
- constexpr bool IsBindless() const {
- return is_bindless;
- }
-
- constexpr bool IsIndexed() const {
- return is_indexed;
- }
-
- constexpr u32 Size() const {
- return size;
- }
-
- constexpr void SetSize(u32 new_size) {
- size = new_size;
- }
-
-private:
- u32 index{}; ///< Emulated index given for the this sampler.
- u32 offset{}; ///< Offset in the const buffer from where the sampler is being read.
- u32 buffer{}; ///< Buffer where the bindless sampler is being read (unused on bound samplers).
- u32 size{1}; ///< Size of the sampler.
+ 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 buffer = 0; ///< Buffer where the bindless sampler is being read (unused on bound samplers).
+ u32 size = 1; ///< Size of the sampler.
Tegra::Shader::TextureType type{}; ///< The type used to sample this texture (Texture2D, etc)
- bool is_array{}; ///< Whether the texture is being sampled as an array texture or not.
- bool is_shadow{}; ///< Whether the texture is being sampled as a depth texture or not.
- bool is_buffer{}; ///< Whether the texture is a texture buffer without sampler.
- bool is_bindless{}; ///< Whether this sampler belongs to a bindless texture or not.
- bool is_indexed{}; ///< Whether this sampler is an indexed array of textures.
+ 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.
};
/// Represents a tracked bindless sampler into a direct const buffer
@@ -379,13 +335,13 @@ private:
u32 offset;
};
-class Image final {
+struct Image {
public:
- /// This constructor is for bound images
+ /// Bound images constructor
constexpr explicit Image(u32 index, u32 offset, Tegra::Shader::ImageType type)
: index{index}, offset{offset}, type{type} {}
- /// This constructor is for bindless samplers
+ /// Bindless samplers constructor
constexpr explicit Image(u32 index, u32 offset, u32 buffer, Tegra::Shader::ImageType type)
: index{index}, offset{offset}, buffer{buffer}, type{type}, is_bindless{true} {}
@@ -403,53 +359,20 @@ public:
is_atomic = true;
}
- constexpr u32 GetIndex() const {
- return index;
- }
-
- constexpr u32 GetOffset() const {
- return offset;
- }
-
- constexpr u32 GetBuffer() const {
- return buffer;
- }
-
- constexpr Tegra::Shader::ImageType GetType() const {
- return type;
- }
-
- constexpr bool IsBindless() const {
- return is_bindless;
- }
-
- constexpr bool IsWritten() const {
- return is_written;
- }
-
- constexpr bool IsRead() const {
- return is_read;
- }
-
- constexpr bool IsAtomic() const {
- return is_atomic;
- }
-
-private:
- u32 index{};
- u32 offset{};
- u32 buffer{};
+ u32 index = 0;
+ u32 offset = 0;
+ u32 buffer = 0;
Tegra::Shader::ImageType type{};
- bool is_bindless{};
- bool is_written{};
- bool is_read{};
- bool is_atomic{};
+ bool is_bindless = false;
+ bool is_written = false;
+ bool is_read = false;
+ bool is_atomic = false;
};
struct GlobalMemoryBase {
- u32 cbuf_index{};
- u32 cbuf_offset{};
+ u32 cbuf_index = 0;
+ u32 cbuf_offset = 0;
bool operator<(const GlobalMemoryBase& rhs) const {
return std::tie(cbuf_index, cbuf_offset) < std::tie(rhs.cbuf_index, rhs.cbuf_offset);
@@ -463,7 +386,7 @@ struct MetaArithmetic {
/// Parameters describing a texture sampler
struct MetaTexture {
- const Sampler& sampler;
+ Sampler sampler;
Node array;
Node depth_compare;
std::vector<Node> aoffi;
diff --git a/src/video_core/shader/shader_ir.h b/src/video_core/shader/shader_ir.h
index c6e7bdf50..15ae152f2 100644
--- a/src/video_core/shader/shader_ir.h
+++ b/src/video_core/shader/shader_ir.h
@@ -18,6 +18,7 @@
#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"
@@ -25,16 +26,13 @@ namespace VideoCommon::Shader {
struct ShaderBlock;
-using ProgramCode = std::vector<u64>;
-
constexpr u32 MAX_PROGRAM_LENGTH = 0x1000;
-class ConstBuffer {
-public:
- explicit ConstBuffer(u32 max_offset, bool is_indirect)
+struct ConstBuffer {
+ constexpr explicit ConstBuffer(u32 max_offset, bool is_indirect)
: max_offset{max_offset}, is_indirect{is_indirect} {}
- ConstBuffer() = default;
+ constexpr ConstBuffer() = default;
void MarkAsUsed(u64 offset) {
max_offset = std::max(max_offset, static_cast<u32>(offset));
@@ -57,8 +55,8 @@ public:
}
private:
- u32 max_offset{};
- bool is_indirect{};
+ u32 max_offset = 0;
+ bool is_indirect = false;
};
struct GlobalMemoryUsage {
@@ -192,10 +190,14 @@ private:
friend class ASTDecoder;
struct SamplerInfo {
- Tegra::Shader::TextureType type;
- bool is_array;
- bool is_shadow;
- bool is_buffer;
+ 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();
@@ -328,16 +330,15 @@ private:
OperationCode GetPredicateCombiner(Tegra::Shader::PredOperation operation);
/// Queries the missing sampler info from the execution context.
- SamplerInfo GetSamplerInfo(std::optional<SamplerInfo> sampler_info, u32 offset,
+ SamplerInfo GetSamplerInfo(SamplerInfo info, u32 offset,
std::optional<u32> buffer = std::nullopt);
- /// Accesses a texture sampler
- const Sampler* GetSampler(const Tegra::Shader::Sampler& sampler,
- std::optional<SamplerInfo> sampler_info = std::nullopt);
+ /// Accesses a texture sampler.
+ std::optional<Sampler> GetSampler(Tegra::Shader::Sampler sampler, SamplerInfo info);
/// Accesses a texture sampler for a bindless texture.
- const Sampler* GetBindlessSampler(Tegra::Shader::Register reg, Node& index_var,
- std::optional<SamplerInfo> sampler_info = std::nullopt);
+ std::optional<Sampler> GetBindlessSampler(Tegra::Shader::Register reg, SamplerInfo info,
+ Node& index_var);
/// Accesses an image.
Image& GetImage(Tegra::Shader::Image image, Tegra::Shader::ImageType type);
diff --git a/src/video_core/shader/track.cpp b/src/video_core/shader/track.cpp
index 513e9bf49..eb97bfd41 100644
--- a/src/video_core/shader/track.cpp
+++ b/src/video_core/shader/track.cpp
@@ -153,21 +153,13 @@ std::tuple<Node, u32, u32> ShaderIR::TrackCbuf(Node tracked, const NodeBlock& co
if (gpr->GetIndex() == Tegra::Shader::Register::ZeroIndex) {
return {};
}
- s64 current_cursor = cursor;
- while (current_cursor > 0) {
- // 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, current_cursor - 1);
- current_cursor = new_cursor;
- if (!source) {
- continue;
- }
- const auto [base_address, index, offset] = TrackCbuf(source, code, current_cursor);
- if (base_address != nullptr) {
- return {base_address, index, offset};
- }
+ // 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 {};
+ 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) {
diff --git a/src/video_core/texture_cache/surface_params.cpp b/src/video_core/texture_cache/surface_params.cpp
index 0de499946..884fabffe 100644
--- a/src/video_core/texture_cache/surface_params.cpp
+++ b/src/video_core/texture_cache/surface_params.cpp
@@ -81,7 +81,7 @@ SurfaceParams SurfaceParams::CreateForTexture(const FormatLookupTable& lookup_ta
params.pixel_format = lookup_table.GetPixelFormat(
tic.format, params.srgb_conversion, tic.r_type, tic.g_type, tic.b_type, tic.a_type);
params.type = GetFormatType(params.pixel_format);
- if (entry.IsShadow() && params.type == SurfaceType::ColorTexture) {
+ if (entry.is_shadow && params.type == SurfaceType::ColorTexture) {
switch (params.pixel_format) {
case PixelFormat::R16U:
case PixelFormat::R16F:
@@ -108,7 +108,7 @@ SurfaceParams SurfaceParams::CreateForTexture(const FormatLookupTable& lookup_ta
params.emulated_levels = 1;
params.is_layered = false;
} else {
- params.target = TextureTypeToSurfaceTarget(entry.GetType(), entry.IsArray());
+ params.target = TextureTypeToSurfaceTarget(entry.type, entry.is_array);
params.width = tic.Width();
params.height = tic.Height();
params.depth = tic.Depth();
@@ -138,7 +138,7 @@ SurfaceParams SurfaceParams::CreateForImage(const FormatLookupTable& lookup_tabl
tic.format, params.srgb_conversion, tic.r_type, tic.g_type, tic.b_type, tic.a_type);
params.type = GetFormatType(params.pixel_format);
params.type = GetFormatType(params.pixel_format);
- params.target = ImageTypeToSurfaceTarget(entry.GetType());
+ params.target = ImageTypeToSurfaceTarget(entry.type);
// TODO: on 1DBuffer we should use the tic info.
if (tic.IsBuffer()) {
params.target = SurfaceTarget::TextureBuffer;
@@ -248,12 +248,12 @@ SurfaceParams SurfaceParams::CreateForFermiCopySurface(
VideoCore::Surface::SurfaceTarget SurfaceParams::ExpectedTarget(
const VideoCommon::Shader::Sampler& entry) {
- return TextureTypeToSurfaceTarget(entry.GetType(), entry.IsArray());
+ return TextureTypeToSurfaceTarget(entry.type, entry.is_array);
}
VideoCore::Surface::SurfaceTarget SurfaceParams::ExpectedTarget(
const VideoCommon::Shader::Image& entry) {
- return ImageTypeToSurfaceTarget(entry.GetType());
+ return ImageTypeToSurfaceTarget(entry.type);
}
bool SurfaceParams::IsLayered() const {
diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h
index cf6bd005a..d6efc34b2 100644
--- a/src/video_core/texture_cache/texture_cache.h
+++ b/src/video_core/texture_cache/texture_cache.h
@@ -143,7 +143,7 @@ public:
}
const auto params{SurfaceParams::CreateForTexture(format_lookup_table, tic, entry)};
- const auto [surface, view] = GetSurface(gpu_addr, *cpu_addr, params, false);
+ const auto [surface, view] = GetSurface(gpu_addr, *cpu_addr, params, true, false);
if (guard_samplers) {
sampled_textures.push_back(surface);
}
@@ -163,7 +163,7 @@ public:
return GetNullSurface(SurfaceParams::ExpectedTarget(entry));
}
const auto params{SurfaceParams::CreateForImage(format_lookup_table, tic, entry)};
- const auto [surface, view] = GetSurface(gpu_addr, *cpu_addr, params, false);
+ const auto [surface, view] = GetSurface(gpu_addr, *cpu_addr, params, true, false);
if (guard_samplers) {
sampled_textures.push_back(surface);
}
@@ -178,7 +178,7 @@ public:
return any_rt;
}
- TView GetDepthBufferSurface() {
+ TView GetDepthBufferSurface(bool preserve_contents) {
std::lock_guard lock{mutex};
auto& maxwell3d = system.GPU().Maxwell3D();
if (!maxwell3d.dirty.flags[VideoCommon::Dirty::ZetaBuffer]) {
@@ -199,7 +199,7 @@ public:
return {};
}
const auto depth_params{SurfaceParams::CreateForDepthBuffer(system)};
- auto surface_view = GetSurface(gpu_addr, *cpu_addr, depth_params, true);
+ auto surface_view = GetSurface(gpu_addr, *cpu_addr, depth_params, preserve_contents, true);
if (depth_buffer.target)
depth_buffer.target->MarkAsRenderTarget(false, NO_RT);
depth_buffer.target = surface_view.first;
@@ -209,7 +209,7 @@ public:
return surface_view.second;
}
- TView GetColorBufferSurface(std::size_t index) {
+ TView GetColorBufferSurface(std::size_t index, bool preserve_contents) {
std::lock_guard lock{mutex};
ASSERT(index < Tegra::Engines::Maxwell3D::Regs::NumRenderTargets);
auto& maxwell3d = system.GPU().Maxwell3D();
@@ -239,8 +239,9 @@ public:
return {};
}
- auto surface_view = GetSurface(gpu_addr, *cpu_addr,
- SurfaceParams::CreateForFramebuffer(system, index), true);
+ auto surface_view =
+ GetSurface(gpu_addr, *cpu_addr, SurfaceParams::CreateForFramebuffer(system, index),
+ preserve_contents, true);
if (render_targets[index].target) {
auto& surface = render_targets[index].target;
surface->MarkAsRenderTarget(false, NO_RT);
@@ -300,9 +301,9 @@ public:
const std::optional<VAddr> src_cpu_addr =
system.GPU().MemoryManager().GpuToCpuAddress(src_gpu_addr);
std::pair<TSurface, TView> dst_surface =
- GetSurface(dst_gpu_addr, *dst_cpu_addr, dst_params, false);
+ GetSurface(dst_gpu_addr, *dst_cpu_addr, dst_params, true, false);
std::pair<TSurface, TView> src_surface =
- GetSurface(src_gpu_addr, *src_cpu_addr, src_params, false);
+ GetSurface(src_gpu_addr, *src_cpu_addr, src_params, true, false);
ImageBlit(src_surface.second, dst_surface.second, copy_config);
dst_surface.first->MarkAsModified(true, Tick());
}
@@ -532,18 +533,22 @@ private:
* @param overlaps The overlapping surfaces registered in the cache.
* @param params The parameters for the new surface.
* @param gpu_addr The starting address of the new surface.
+ * @param preserve_contents Indicates that the new surface should be loaded from memory or left
+ * blank.
* @param untopological Indicates to the recycler that the texture has no way to match the
* overlaps due to topological reasons.
**/
std::pair<TSurface, TView> RecycleSurface(std::vector<TSurface>& overlaps,
const SurfaceParams& params, const GPUVAddr gpu_addr,
+ const bool preserve_contents,
const MatchTopologyResult untopological) {
+ const bool do_load = preserve_contents && Settings::IsGPULevelExtreme();
for (auto& surface : overlaps) {
Unregister(surface);
}
switch (PickStrategy(overlaps, params, gpu_addr, untopological)) {
case RecycleStrategy::Ignore: {
- return InitializeSurface(gpu_addr, params, Settings::IsGPULevelExtreme());
+ return InitializeSurface(gpu_addr, params, do_load);
}
case RecycleStrategy::Flush: {
std::sort(overlaps.begin(), overlaps.end(),
@@ -553,7 +558,7 @@ private:
for (auto& surface : overlaps) {
FlushSurface(surface);
}
- return InitializeSurface(gpu_addr, params);
+ return InitializeSurface(gpu_addr, params, preserve_contents);
}
case RecycleStrategy::BufferCopy: {
auto new_surface = GetUncachedSurface(gpu_addr, params);
@@ -562,7 +567,7 @@ private:
}
default: {
UNIMPLEMENTED_MSG("Unimplemented Texture Cache Recycling Strategy!");
- return InitializeSurface(gpu_addr, params);
+ return InitializeSurface(gpu_addr, params, do_load);
}
}
}
@@ -700,11 +705,14 @@ private:
* @param params The parameters on the new surface.
* @param gpu_addr The starting address of the new surface.
* @param cpu_addr The starting address of the new surface on physical memory.
+ * @param preserve_contents Indicates that the new surface should be loaded from memory or
+ * left blank.
*/
std::optional<std::pair<TSurface, TView>> Manage3DSurfaces(std::vector<TSurface>& overlaps,
const SurfaceParams& params,
const GPUVAddr gpu_addr,
- const VAddr cpu_addr) {
+ const VAddr cpu_addr,
+ bool preserve_contents) {
if (params.target == SurfaceTarget::Texture3D) {
bool failed = false;
if (params.num_levels > 1) {
@@ -754,7 +762,7 @@ private:
return std::nullopt;
}
Unregister(surface);
- return InitializeSurface(gpu_addr, params);
+ return InitializeSurface(gpu_addr, params, preserve_contents);
}
return std::nullopt;
}
@@ -765,7 +773,7 @@ private:
return {{surface, surface->GetMainView()}};
}
}
- return InitializeSurface(gpu_addr, params);
+ return InitializeSurface(gpu_addr, params, preserve_contents);
}
}
@@ -788,10 +796,13 @@ private:
*
* @param gpu_addr The starting address of the candidate surface.
* @param params The parameters on the candidate surface.
+ * @param preserve_contents Indicates that the new surface should be loaded from memory or
+ * left blank.
* @param is_render Whether or not the surface is a render target.
**/
std::pair<TSurface, TView> GetSurface(const GPUVAddr gpu_addr, const VAddr cpu_addr,
- const SurfaceParams& params, bool is_render) {
+ const SurfaceParams& params, bool preserve_contents,
+ bool is_render) {
// Step 1
// Check Level 1 Cache for a fast structural match. If candidate surface
// matches at certain level we are pretty much done.
@@ -800,7 +811,8 @@ private:
const auto topological_result = current_surface->MatchesTopology(params);
if (topological_result != MatchTopologyResult::FullMatch) {
std::vector<TSurface> overlaps{current_surface};
- return RecycleSurface(overlaps, params, gpu_addr, topological_result);
+ return RecycleSurface(overlaps, params, gpu_addr, preserve_contents,
+ topological_result);
}
const auto struct_result = current_surface->MatchesStructure(params);
@@ -825,7 +837,7 @@ private:
// If none are found, we are done. we just load the surface and create it.
if (overlaps.empty()) {
- return InitializeSurface(gpu_addr, params);
+ return InitializeSurface(gpu_addr, params, preserve_contents);
}
// Step 3
@@ -835,13 +847,15 @@ private:
for (const auto& surface : overlaps) {
const auto topological_result = surface->MatchesTopology(params);
if (topological_result != MatchTopologyResult::FullMatch) {
- return RecycleSurface(overlaps, params, gpu_addr, topological_result);
+ return RecycleSurface(overlaps, params, gpu_addr, preserve_contents,
+ topological_result);
}
}
// Check if it's a 3D texture
if (params.block_depth > 0) {
- auto surface = Manage3DSurfaces(overlaps, params, gpu_addr, cpu_addr);
+ auto surface =
+ Manage3DSurfaces(overlaps, params, gpu_addr, cpu_addr, preserve_contents);
if (surface) {
return *surface;
}
@@ -861,7 +875,8 @@ private:
return *view;
}
}
- return RecycleSurface(overlaps, params, gpu_addr, MatchTopologyResult::FullMatch);
+ return RecycleSurface(overlaps, params, gpu_addr, preserve_contents,
+ MatchTopologyResult::FullMatch);
}
// Now we check if the candidate is a mipmap/layer of the overlap
std::optional<TView> view =
@@ -885,7 +900,7 @@ private:
pair.first->EmplaceView(params, gpu_addr, candidate_size);
if (mirage_view)
return {pair.first, *mirage_view};
- return RecycleSurface(overlaps, params, gpu_addr,
+ return RecycleSurface(overlaps, params, gpu_addr, preserve_contents,
MatchTopologyResult::FullMatch);
}
return {current_surface, *view};
@@ -901,7 +916,8 @@ private:
}
}
// We failed all the tests, recycle the overlaps into a new texture.
- return RecycleSurface(overlaps, params, gpu_addr, MatchTopologyResult::FullMatch);
+ return RecycleSurface(overlaps, params, gpu_addr, preserve_contents,
+ MatchTopologyResult::FullMatch);
}
/**
@@ -1059,10 +1075,10 @@ private:
}
std::pair<TSurface, TView> InitializeSurface(GPUVAddr gpu_addr, const SurfaceParams& params,
- bool do_load = true) {
+ bool preserve_contents) {
auto new_surface{GetUncachedSurface(gpu_addr, params)};
Register(new_surface);
- if (do_load) {
+ if (preserve_contents) {
LoadSurface(new_surface);
}
return {new_surface, new_surface->GetMainView()};
@@ -1156,7 +1172,7 @@ private:
/// Returns true the shader sampler entry is compatible with the TIC texture type.
static bool IsTypeCompatible(Tegra::Texture::TextureType tic_type,
const VideoCommon::Shader::Sampler& entry) {
- const auto shader_type = entry.GetType();
+ const auto shader_type = entry.type;
switch (tic_type) {
case Tegra::Texture::TextureType::Texture1D:
case Tegra::Texture::TextureType::Texture1DArray:
@@ -1177,7 +1193,7 @@ private:
if (shader_type == Tegra::Shader::TextureType::TextureCube) {
return true;
}
- return shader_type == Tegra::Shader::TextureType::Texture2D && entry.IsArray();
+ return shader_type == Tegra::Shader::TextureType::Texture2D && entry.is_array;
}
UNREACHABLE();
return true;