VideoCore/Shader: Split shader uniform state and shader engine

Currently there's only a single dummy implementation, which will be
split in a following commit.

src/citra_qt/debugger/graphics/graphics_vertex_shader.cpp

@@ -518,7 +518,9 @@ void GraphicsVertexShaderWidget::Reload(bool replace_vertex_data, void* vertex_d
     info.labels.insert({entry_point, "main"});
 
     // Generate debug information
-    debug_data = shader_setup.ProduceDebugInfo(input_vertex, num_attributes, entry_point);
+    auto* shader_engine = Pica::Shader::GetEngine();
+    shader_engine->SetupBatch(&shader_setup);
+    debug_data = shader_engine->ProduceDebugInfo(input_vertex, num_attributes, entry_point);
 
     // Reload widget state
     for (int attr = 0; attr < num_attributes; ++attr) {

src/video_core/command_processor.cpp

@@ -142,15 +142,16 @@ static void WritePicaReg(u32 id, u32 value, u32 mask) {
                     MICROPROFILE_SCOPE(GPU_Drawing);
                     immediate_attribute_id = 0;
 
-                    Shader::UnitState shader_unit;
+                    auto* shader_engine = Shader::GetEngine();
-                    g_state.vs.Setup();
+                    shader_engine->SetupBatch(&g_state.vs);
 
                     // Send to vertex shader
                     if (g_debug_context)
                         g_debug_context->OnEvent(DebugContext::Event::VertexShaderInvocation,
                                                  static_cast<void*>(&immediate_input));
+                    Shader::UnitState shader_unit;
                     shader_unit.LoadInputVertex(immediate_input, regs.vs.num_input_attributes + 1);
-                    g_state.vs.Run(shader_unit, regs.vs.main_offset);
+                    shader_engine->Run(shader_unit, regs.vs.main_offset);
                     Shader::OutputVertex output_vertex =
                         shader_unit.output_registers.ToVertex(regs.vs);
 
@@ -244,8 +245,10 @@ static void WritePicaReg(u32 id, u32 value, u32 mask) {
         unsigned int vertex_cache_pos = 0;
         vertex_cache_ids.fill(-1);
 
+        auto* shader_engine = Shader::GetEngine();
         Shader::UnitState shader_unit;
-        g_state.vs.Setup();
+
+        shader_engine->SetupBatch(&g_state.vs);
 
         for (unsigned int index = 0; index < regs.num_vertices; ++index) {
             // Indexed rendering doesn't use the start offset
@@ -285,7 +288,7 @@ static void WritePicaReg(u32 id, u32 value, u32 mask) {
                     g_debug_context->OnEvent(DebugContext::Event::VertexShaderInvocation,
                                              (void*)&input);
                 shader_unit.LoadInputVertex(input, loader.GetNumTotalAttributes());
-                g_state.vs.Run(shader_unit, regs.vs.main_offset);
+                shader_engine->Run(shader_unit, regs.vs.main_offset);
 
                 // Retrieve vertex from register data
                 output_vertex = shader_unit.output_registers.ToVertex(regs.vs);

src/video_core/shader/shader.cpp

@@ -87,6 +87,17 @@ void UnitState::LoadInputVertex(const InputVertex& input, int num_attributes) {
     conditional_code[1] = false;
 }
 
+class MergedShaderEngine : public ShaderEngine {
+public:
+    void SetupBatch(const ShaderSetup* setup) override;
+    void Run(UnitState& state, unsigned int entry_point) const override;
+    DebugData<true> ProduceDebugInfo(const InputVertex& input, int num_attributes,
+        unsigned int entry_point) const override;
+
+private:
+    const ShaderSetup* setup = nullptr;
+};
+
 #ifdef ARCHITECTURE_x86_64
 static std::unordered_map<u64, std::unique_ptr<JitShader>> shader_map;
 static const JitShader* jit_shader;
@@ -98,13 +109,17 @@ void ClearCache() {
 #endif // ARCHITECTURE_x86_64
 }
 
-void ShaderSetup::Setup() {
+void MergedShaderEngine::SetupBatch(const ShaderSetup* setup_) {
+    setup = setup_;
+    if (setup == nullptr)
+        return;
+
 #ifdef ARCHITECTURE_x86_64
     if (VideoCore::g_shader_jit_enabled) {
-        u64 cache_key =
+        u64 code_hash = Common::ComputeHash64(&setup->program_code, sizeof(setup->program_code));
-            Common::ComputeHash64(&program_code, sizeof(program_code)) ^
+        u64 swizzle_hash = Common::ComputeHash64(&setup->swizzle_data, sizeof(setup->swizzle_data));
-            Common::ComputeHash64(&swizzle_data, sizeof(swizzle_data));
 
+        u64 cache_key = code_hash ^ swizzle_hash;
         auto iter = shader_map.find(cache_key);
         if (iter != shader_map.end()) {
             jit_shader = iter->second.get();
@@ -120,26 +135,28 @@ void ShaderSetup::Setup() {
 
 MICROPROFILE_DEFINE(GPU_Shader, "GPU", "Shader", MP_RGB(50, 50, 240));
 
-void ShaderSetup::Run(UnitState& state, unsigned int entry_point) const {
+void MergedShaderEngine::Run(UnitState& state, unsigned int entry_point) const {
+    ASSERT(setup != nullptr);
     ASSERT(entry_point < 1024);
 
     MICROPROFILE_SCOPE(GPU_Shader);
 
 #ifdef ARCHITECTURE_x86_64
     if (VideoCore::g_shader_jit_enabled) {
-        jit_shader->Run(*this, state, entry_point);
+        jit_shader->Run(*setup, state, entry_point);
     } else {
         DebugData<false> dummy_debug_data;
-        RunInterpreter(*this, state, dummy_debug_data, entry_point);
+        RunInterpreter(*setup, state, dummy_debug_data, entry_point);
     }
 #else
     DebugData<false> dummy_debug_data;
-    RunInterpreter(*this, state, dummy_debug_data, entry_point);
+    RunInterpreter(*setup, state, dummy_debug_data, entry_point);
 #endif // ARCHITECTURE_x86_64
 }
 
-DebugData<true> ShaderSetup::ProduceDebugInfo(const InputVertex& input, int num_attributes,
+DebugData<true> MergedShaderEngine::ProduceDebugInfo(const InputVertex& input, int num_attributes,
                                                      unsigned int entry_point) const {
+    ASSERT(setup != nullptr);
     ASSERT(entry_point < 1024);
 
     UnitState state;
@@ -148,10 +165,15 @@ DebugData<true> ShaderSetup::ProduceDebugInfo(const InputVertex& input, int num_
     // Setup input register table
     boost::fill(state.registers.input, Math::Vec4<float24>::AssignToAll(float24::Zero()));
     state.LoadInputVertex(input, num_attributes);
-    RunInterpreter(*this, state, debug_data, entry_point);
+    RunInterpreter(*setup, state, debug_data, entry_point);
     return debug_data;
 }
 
+ShaderEngine* GetEngine() {
+    static MergedShaderEngine merged_engine;
+    return &merged_engine;
+}
+
 } // namespace Shader
 
 } // namespace Pica

src/video_core/shader/shader.h

@@ -156,7 +156,6 @@ struct UnitState {
 void ClearCache();
 
 struct ShaderSetup {
-
     struct {
         // The float uniforms are accessed by the shader JIT using SSE instructions, and are
         // therefore required to be 16-byte aligned.
@@ -180,18 +179,23 @@ struct ShaderSetup {
 
     std::array<u32, 1024> program_code;
     std::array<u32, 1024> swizzle_data;
+};
+
+class ShaderEngine {
+public:
+    virtual ~ShaderEngine() = default;
 
     /**
      * Performs any shader unit setup that only needs to happen once per shader (as opposed to once
      * per vertex, which would happen within the `Run` function).
      */
-    void Setup();
+    virtual void SetupBatch(const ShaderSetup* setup) = 0;
 
     /**
      * Runs the currently setup shader
      * @param state Shader unit state, must be setup per shader and per shader unit
      */
-    void Run(UnitState& state, unsigned int entry_point) const;
+    virtual void Run(UnitState& state, unsigned int entry_point) const = 0;
 
     /**
      * Produce debug information based on the given shader and input vertex
@@ -200,10 +204,13 @@ struct ShaderSetup {
      * @param config Configuration object for the shader pipeline
      * @return Debug information for this shader with regards to the given vertex
      */
-    DebugData<true> ProduceDebugInfo(const InputVertex& input, int num_attributes,
+    virtual DebugData<true> ProduceDebugInfo(const InputVertex& input, int num_attributes,
-                                     unsigned int entry_point) const;
+                                             unsigned int entry_point) const = 0;
 };
 
+// TODO(yuriks): Remove and make it non-global state somewhere
+ShaderEngine* GetEngine();
+
 } // namespace Shader
 
 } // namespace Pica

src/video_core/shader/shader_interpreter.h

@@ -8,6 +8,7 @@ namespace Pica {
 
 namespace Shader {
 
+struct ShaderSetup;
 struct UnitState;
 
 template <bool Debug>