Merge pull request #11212 from Kelebek1/shader_stuff

Fix various misc pipeline/shader things

src/shader_recompiler/environment.h

@@ -39,7 +39,7 @@ public:
     [[nodiscard]] virtual std::optional<ReplaceConstant> GetReplaceConstBuffer(u32 bank,
                                                                                u32 offset) = 0;
 
-    virtual void Dump(u64 hash) = 0;
+    virtual void Dump(u64 pipeline_hash, u64 shader_hash) = 0;
 
     [[nodiscard]] const ProgramHeader& SPH() const noexcept {
         return sph;

src/video_core/renderer_opengl/gl_shader_cache.cpp

@@ -445,7 +445,8 @@ std::unique_ptr<GraphicsPipeline> ShaderCache::CreateGraphicsPipeline(
     ShaderContext::ShaderPools& pools, const GraphicsPipelineKey& key,
     std::span<Shader::Environment* const> envs, bool use_shader_workers,
     bool force_context_flush) try {
-    LOG_INFO(Render_OpenGL, "0x{:016x}", key.Hash());
+    auto hash = key.Hash();
+    LOG_INFO(Render_OpenGL, "0x{:016x}", hash);
     size_t env_index{};
     u32 total_storage_buffers{};
     std::array<Shader::IR::Program, Maxwell::MaxShaderProgram> programs;
@@ -474,7 +475,7 @@ std::unique_ptr<GraphicsPipeline> ShaderCache::CreateGraphicsPipeline(
         Shader::Maxwell::Flow::CFG cfg(env, pools.flow_block, cfg_offset, index == 0);
 
         if (Settings::values.dump_shaders) {
-            env.Dump(key.unique_hashes[index]);
+            env.Dump(hash, key.unique_hashes[index]);
         }
 
         if (!uses_vertex_a || index != 1) {
@@ -566,12 +567,13 @@ std::unique_ptr<ComputePipeline> ShaderCache::CreateComputePipeline(
 std::unique_ptr<ComputePipeline> ShaderCache::CreateComputePipeline(
     ShaderContext::ShaderPools& pools, const ComputePipelineKey& key, Shader::Environment& env,
     bool force_context_flush) try {
-    LOG_INFO(Render_OpenGL, "0x{:016x}", key.Hash());
+    auto hash = key.Hash();
+    LOG_INFO(Render_OpenGL, "0x{:016x}", hash);
 
     Shader::Maxwell::Flow::CFG cfg{env, pools.flow_block, env.StartAddress()};
 
     if (Settings::values.dump_shaders) {
-        env.Dump(key.Hash());
+        env.Dump(hash, key.unique_hash);
     }
 
     auto program{TranslateProgram(pools.inst, pools.block, env, cfg, host_info)};

src/video_core/renderer_vulkan/vk_pipeline_cache.cpp

@@ -584,7 +584,8 @@ std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline(
     ShaderPools& pools, const GraphicsPipelineCacheKey& key,
     std::span<Shader::Environment* const> envs, PipelineStatistics* statistics,
     bool build_in_parallel) try {
-    LOG_INFO(Render_Vulkan, "0x{:016x}", key.Hash());
+    auto hash = key.Hash();
+    LOG_INFO(Render_Vulkan, "0x{:016x}", hash);
     size_t env_index{0};
     std::array<Shader::IR::Program, Maxwell::MaxShaderProgram> programs;
     const bool uses_vertex_a{key.unique_hashes[0] != 0};
@@ -611,7 +612,7 @@ std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline(
         const u32 cfg_offset{static_cast<u32>(env.StartAddress() + sizeof(Shader::ProgramHeader))};
         Shader::Maxwell::Flow::CFG cfg(env, pools.flow_block, cfg_offset, index == 0);
         if (Settings::values.dump_shaders) {
-            env.Dump(key.unique_hashes[index]);
+            env.Dump(hash, key.unique_hashes[index]);
         }
         if (!uses_vertex_a || index != 1) {
             // Normal path
@@ -712,18 +713,19 @@ std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline(
 std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline(
     ShaderPools& pools, const ComputePipelineCacheKey& key, Shader::Environment& env,
     PipelineStatistics* statistics, bool build_in_parallel) try {
+    auto hash = key.Hash();
     if (device.HasBrokenCompute()) {
-        LOG_ERROR(Render_Vulkan, "Skipping 0x{:016x}", key.Hash());
+        LOG_ERROR(Render_Vulkan, "Skipping 0x{:016x}", hash);
         return nullptr;
     }
 
-    LOG_INFO(Render_Vulkan, "0x{:016x}", key.Hash());
+    LOG_INFO(Render_Vulkan, "0x{:016x}", hash);
 
     Shader::Maxwell::Flow::CFG cfg{env, pools.flow_block, env.StartAddress()};
 
     // Dump it before error.
     if (Settings::values.dump_shaders) {
-        env.Dump(key.Hash());
+        env.Dump(hash, key.unique_hash);
     }
 
     auto program{TranslateProgram(pools.inst, pools.block, env, cfg, host_info)};

src/video_core/shader_cache.cpp

@@ -51,6 +51,11 @@ bool ShaderCache::RefreshStages(std::array<u64, 6>& unique_hashes) {
         }
         const auto& shader_config{maxwell3d->regs.pipelines[index]};
         const auto program{static_cast<Tegra::Engines::Maxwell3D::Regs::ShaderType>(index)};
+        if (program == Tegra::Engines::Maxwell3D::Regs::ShaderType::Pixel &&
+            !maxwell3d->regs.rasterize_enable) {
+            unique_hashes[index] = 0;
+            continue;
+        }
         const GPUVAddr shader_addr{base_addr + shader_config.offset};
         const std::optional<VAddr> cpu_shader_addr{gpu_memory->GpuToCpuAddress(shader_addr)};
         if (!cpu_shader_addr) {

src/video_core/shader_cache.h

@@ -70,7 +70,7 @@ public:
 protected:
     struct GraphicsEnvironments {
         std::array<GraphicsEnvironment, NUM_PROGRAMS> envs;
-        std::array<Shader::Environment*, NUM_PROGRAMS> env_ptrs;
+        std::array<Shader::Environment*, NUM_PROGRAMS> env_ptrs{};
 
         std::span<Shader::Environment* const> Span() const noexcept {
             return std::span(env_ptrs.begin(), std::ranges::find(env_ptrs, nullptr));

src/video_core/shader_environment.cpp

@@ -102,7 +102,8 @@ static std::string_view StageToPrefix(Shader::Stage stage) {
     }
 }
 
-static void DumpImpl(u64 hash, const u64* code, u32 read_highest, u32 read_lowest,
+static void DumpImpl(u64 pipeline_hash, u64 shader_hash, std::span<const u64> code,
+                     [[maybe_unused]] u32 read_highest, [[maybe_unused]] u32 read_lowest,
                      u32 initial_offset, Shader::Stage stage) {
     const auto shader_dir{Common::FS::GetYuzuPath(Common::FS::YuzuPath::DumpDir)};
     const auto base_dir{shader_dir / "shaders"};
@@ -111,13 +112,18 @@ static void DumpImpl(u64 hash, const u64* code, u32 read_highest, u32 read_lowes
         return;
     }
     const auto prefix = StageToPrefix(stage);
-    const auto name{base_dir / fmt::format("{}{:016x}.ash", prefix, hash)};
+    const auto name{base_dir /
-    const size_t real_size = read_highest - read_lowest + initial_offset;
+                    fmt::format("{:016x}_{}_{:016x}.ash", pipeline_hash, prefix, shader_hash)};
-    const size_t padding_needed = ((32 - (real_size % 32)) % 32);
     std::fstream shader_file(name, std::ios::out | std::ios::binary);
+    ASSERT(initial_offset % sizeof(u64) == 0);
     const size_t jump_index = initial_offset / sizeof(u64);
-    shader_file.write(reinterpret_cast<const char*>(code + jump_index), real_size);
+    const size_t code_size = code.size_bytes() - initial_offset;
-    for (size_t i = 0; i < padding_needed; i++) {
+    shader_file.write(reinterpret_cast<const char*>(&code[jump_index]), code_size);
+
+    // + 1 instruction, due to the fact that we skip the final self branch instruction in the code,
+    // but we need to consider it for padding, otherwise nvdisasm rages.
+    const size_t padding_needed = (32 - ((code_size + INST_SIZE) % 32)) % 32;
+    for (size_t i = 0; i < INST_SIZE + padding_needed; i++) {
         shader_file.put(0);
     }
 }
@@ -197,8 +203,8 @@ u64 GenericEnvironment::CalculateHash() const {
     return Common::CityHash64(data.get(), size);
 }
 
-void GenericEnvironment::Dump(u64 hash) {
+void GenericEnvironment::Dump(u64 pipeline_hash, u64 shader_hash) {
-    DumpImpl(hash, code.data(), read_highest, read_lowest, initial_offset, stage);
+    DumpImpl(pipeline_hash, shader_hash, code, read_highest, read_lowest, initial_offset, stage);
 }
 
 void GenericEnvironment::Serialize(std::ofstream& file) const {
@@ -282,6 +288,7 @@ std::optional<u64> GenericEnvironment::TryFindSize() {
 Tegra::Texture::TICEntry GenericEnvironment::ReadTextureInfo(GPUVAddr tic_addr, u32 tic_limit,
                                                              bool via_header_index, u32 raw) {
     const auto handle{Tegra::Texture::TexturePair(raw, via_header_index)};
+    ASSERT(handle.first <= tic_limit);
     const GPUVAddr descriptor_addr{tic_addr + handle.first * sizeof(Tegra::Texture::TICEntry)};
     Tegra::Texture::TICEntry entry;
     gpu_memory->ReadBlock(descriptor_addr, &entry, sizeof(entry));
@@ -465,8 +472,8 @@ void FileEnvironment::Deserialize(std::ifstream& file) {
         .read(reinterpret_cast<char*>(&read_highest), sizeof(read_highest))
         .read(reinterpret_cast<char*>(&viewport_transform_state), sizeof(viewport_transform_state))
         .read(reinterpret_cast<char*>(&stage), sizeof(stage));
-    code = std::make_unique<u64[]>(Common::DivCeil(code_size, sizeof(u64)));
+    code.resize(Common::DivCeil(code_size, sizeof(u64)));
-    file.read(reinterpret_cast<char*>(code.get()), code_size);
+    file.read(reinterpret_cast<char*>(code.data()), code_size);
     for (size_t i = 0; i < num_texture_types; ++i) {
         u32 key;
         Shader::TextureType type;
@@ -509,8 +516,8 @@ void FileEnvironment::Deserialize(std::ifstream& file) {
     is_propietary_driver = texture_bound == 2;
 }
 
-void FileEnvironment::Dump(u64 hash) {
+void FileEnvironment::Dump(u64 pipeline_hash, u64 shader_hash) {
-    DumpImpl(hash, code.get(), read_highest, read_lowest, initial_offset, stage);
+    DumpImpl(pipeline_hash, shader_hash, code, read_highest, read_lowest, initial_offset, stage);
 }
 
 u64 FileEnvironment::ReadInstruction(u32 address) {

src/video_core/shader_environment.h

@@ -58,7 +58,7 @@ public:
 
     [[nodiscard]] u64 CalculateHash() const;
 
-    void Dump(u64 hash) override;
+    void Dump(u64 pipeline_hash, u64 shader_hash) override;
 
     void Serialize(std::ofstream& file) const;
 
@@ -188,10 +188,10 @@ public:
         return cbuf_replacements.size() != 0;
     }
 
-    void Dump(u64 hash) override;
+    void Dump(u64 pipeline_hash, u64 shader_hash) override;
 
 private:
-    std::unique_ptr<u64[]> code;
+    std::vector<u64> code;
     std::unordered_map<u32, Shader::TextureType> texture_types;
     std::unordered_map<u32, Shader::TexturePixelFormat> texture_pixel_formats;
     std::unordered_map<u64, u32> cbuf_values;