Merge pull request #2810 from ReinUsesLisp/mme-opt

maxwell_3d: Avoid moving macro_params

src/video_core/engines/maxwell_3d.cpp

@@ -244,7 +244,7 @@ void Maxwell3D::InitDirtySettings() {
     dirty_pointers[MAXWELL3D_REG_INDEX(polygon_offset_clamp)] = polygon_offset_dirty_reg;
 }
 
-void Maxwell3D::CallMacroMethod(u32 method, std::vector<u32> parameters) {
+void Maxwell3D::CallMacroMethod(u32 method, std::size_t num_parameters, const u32* parameters) {
     // Reset the current macro.
     executing_macro = 0;
 
@@ -252,7 +252,7 @@ void Maxwell3D::CallMacroMethod(u32 method, std::vector<u32> parameters) {
     const u32 entry = ((method - MacroRegistersStart) >> 1) % macro_positions.size();
 
     // Execute the current macro.
-    macro_interpreter.Execute(macro_positions[entry], std::move(parameters));
+    macro_interpreter.Execute(macro_positions[entry], num_parameters, parameters);
 }
 
 void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) {
@@ -289,7 +289,8 @@ void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) {
 
         // Call the macro when there are no more parameters in the command buffer
         if (method_call.IsLastCall()) {
-            CallMacroMethod(executing_macro, std::move(macro_params));
+            CallMacroMethod(executing_macro, macro_params.size(), macro_params.data());
+            macro_params.clear();
         }
         return;
     }

src/video_core/engines/maxwell_3d.h

@@ -1308,9 +1308,10 @@ private:
     /**
      * Call a macro on this engine.
      * @param method Method to call
+     * @param num_parameters Number of arguments
      * @param parameters Arguments to the method call
      */
-    void CallMacroMethod(u32 method, std::vector<u32> parameters);
+    void CallMacroMethod(u32 method, std::size_t num_parameters, const u32* parameters);
 
     /// Handles writes to the macro uploading register.
     void ProcessMacroUpload(u32 data);

src/video_core/macro_interpreter.cpp

@@ -14,11 +14,18 @@ namespace Tegra {
 
 MacroInterpreter::MacroInterpreter(Engines::Maxwell3D& maxwell3d) : maxwell3d(maxwell3d) {}
 
-void MacroInterpreter::Execute(u32 offset, std::vector<u32> parameters) {
+void MacroInterpreter::Execute(u32 offset, std::size_t num_parameters, const u32* parameters) {
     MICROPROFILE_SCOPE(MacroInterp);
     Reset();
+
     registers[1] = parameters[0];
-    this->parameters = std::move(parameters);
+
+    if (num_parameters > parameters_capacity) {
+        parameters_capacity = num_parameters;
+        this->parameters = std::make_unique<u32[]>(num_parameters);
+    }
+    std::memcpy(this->parameters.get(), parameters, num_parameters * sizeof(u32));
+    this->num_parameters = num_parameters;
 
     // Execute the code until we hit an exit condition.
     bool keep_executing = true;
@@ -27,7 +34,7 @@ void MacroInterpreter::Execute(u32 offset, std::vector<u32> parameters) {
     }
 
     // Assert the the macro used all the input parameters
-    ASSERT(next_parameter_index == this->parameters.size());
+    ASSERT(next_parameter_index == num_parameters);
 }
 
 void MacroInterpreter::Reset() {
@@ -35,7 +42,7 @@ void MacroInterpreter::Reset() {
     pc = 0;
     delayed_pc = {};
     method_address.raw = 0;
-    parameters.clear();
+    num_parameters = 0;
     // The next parameter index starts at 1, because $r1 already has the value of the first
     // parameter.
     next_parameter_index = 1;
@@ -229,7 +236,8 @@ void MacroInterpreter::ProcessResult(ResultOperation operation, u32 reg, u32 res
 }
 
 u32 MacroInterpreter::FetchParameter() {
-    return parameters.at(next_parameter_index++);
+    ASSERT(next_parameter_index < num_parameters);
+    return parameters[next_parameter_index++];
 }
 
 u32 MacroInterpreter::GetRegister(u32 register_id) const {

src/video_core/macro_interpreter.h

@@ -25,7 +25,7 @@ public:
      * @param offset Offset to start execution at.
      * @param parameters The parameters of the macro.
      */
-    void Execute(u32 offset, std::vector<u32> parameters);
+    void Execute(u32 offset, std::size_t num_parameters, const u32* parameters);
 
 private:
     enum class Operation : u32 {
@@ -162,10 +162,12 @@ private:
     MethodAddress method_address = {};
 
     /// Input parameters of the current macro.
-    std::vector<u32> parameters;
+    std::unique_ptr<u32[]> parameters;
+    std::size_t num_parameters = 0;
+    std::size_t parameters_capacity = 0;
     /// Index of the next parameter that will be fetched by the 'parm' instruction.
     u32 next_parameter_index = 0;
 
-    bool carry_flag{};
+    bool carry_flag = false;
 };
 } // namespace Tegra