Preparation for compiling the shader epilogue as well.

src/video_core/shader/shader.cpp

@@ -40,7 +40,8 @@ void Setup() {
         u64 cache_key =
             Common::ComputeHash64(&g_state.vs.program_code, sizeof(g_state.vs.program_code)) ^
             Common::ComputeHash64(&g_state.vs.swizzle_data, sizeof(g_state.vs.swizzle_data)) ^
-            Common::ComputeHash64(&attribute_register_map, sizeof(attribute_register_map));
+            Common::ComputeHash64(&attribute_register_map, sizeof(attribute_register_map)) ^
+            Common::ComputeHash64(&g_state.regs.vs_output_attributes, sizeof(g_state.regs.vs_output_attributes));
 
         auto iter = shader_map.find(cache_key);
         if (iter != shader_map.end()) {
@@ -64,25 +65,25 @@ void Shutdown() {
 MICROPROFILE_DEFINE(GPU_VertexShader, "GPU", "Vertex Shader", MP_RGB(50, 50, 240));
 
 OutputVertex Run(UnitState<false>& state, const InputVertex& input, int num_attributes) {
-    auto& config = g_state.regs.vs;
-
     MICROPROFILE_SCOPE(GPU_VertexShader);
 
-    state.program_counter = config.main_offset;
     state.debug.max_offset = 0;
     state.debug.max_opdesc_id = 0;
 
-    // Setup input register table
+    // Setup output data
-
+    OutputVertex ret;
-    state.conditional_code[0] = false;
-    state.conditional_code[1] = false;
 
 #ifdef ARCHITECTURE_x86_64
     if (VideoCore::g_shader_jit_enabled) {
-        jit_shader->Run(&state.registers, &input.attr[0], g_state.regs.vs.main_offset);
+        jit_shader->Run(&state.registers, &input.attr[0], g_state.regs.vs.main_offset, &ret);
     } else
 #endif // ARCHITECTURE_x86_64
     {
+        auto& config = g_state.regs.vs;
+        // Setup input register table
+        state.program_counter = config.main_offset;
+        state.conditional_code[0] = false;
+        state.conditional_code[1] = false;
         const auto& attribute_register_map = config.input_register_map;
         // TODO: Instead of this cumbersome logic, just load the input data directly like
         // for (int attr = 0; attr < num_attributes; ++attr) { input_attr[0] = state.registers.input[attribute_register_map.attribute0_register]; }
@@ -106,13 +107,10 @@ OutputVertex Run(UnitState<false>& state, const InputVertex& input, int num_attr
         RunInterpreter(state);
     }
 
-    // Setup output data
-    OutputVertex ret;
     // TODO(neobrain): Under some circumstances, up to 16 attributes may be output. We need to
     // figure out what those circumstances are and enable the remaining outputs then.
     unsigned index = 0;
     for (unsigned i = 0; i < 7; ++i) {
-
         if (index >= g_state.regs.vs_output_total)
             break;
 

src/video_core/shader/shader_jit_x64.cpp

@@ -581,7 +581,6 @@ void JitShader::Compile_NOP(Instruction instr) {
 }
 
 void JitShader::Compile_END(Instruction instr) {
-    ABI_PopRegistersAndAdjustStack(ABI_ALL_CALLEE_SAVED, 8);
     RET();
 }
 
@@ -825,7 +824,7 @@ void JitShader::Compile() {
     // The stack pointer is 8 modulo 16 at the entry of a procedure
     ABI_PushRegistersAndAdjustStack(ABI_ALL_CALLEE_SAVED, 8);
 
-    // Load inputs
+    // Prologue: Scatter inputs into registers according to map
     MOV(PTRBITS, R(REGISTERS), R(ABI_PARAM1));
     MOV(PTRBITS, R(INPUT), R(ABI_PARAM2));
     for (int i = 0; i < num_attributes; i++) {
@@ -849,9 +848,12 @@ void JitShader::Compile() {
     static const __m128 neg = { -0.f, -0.f, -0.f, -0.f };
     MOV(PTRBITS, R(RAX), ImmPtr(&neg));
     MOVAPS(NEGBIT, MatR(RAX));
+    // Call the start of the shader program.
+    CALLptr(R(ABI_PARAM3));
+    // Alright, back from the program. Now we can do the epilogue.
 
-    // Jump to start of the shader program
+    ABI_PopRegistersAndAdjustStack(ABI_ALL_CALLEE_SAVED, 8);
-    JMPptr(R(ABI_PARAM3));
+    RET();
 
     // Compile entire program
     Compile_Block(static_cast<unsigned>(g_state.vs.program_code.size()));

src/video_core/shader/shader_jit_x64.h

@@ -33,8 +33,8 @@ class JitShader : public Gen::XCodeBlock {
 public:
     JitShader();
 
-    void Run(void* registers, const void *input, unsigned offset) const {
+    void Run(void* registers, const void *input, unsigned int offset, void *output) const {
-        program(registers, input, code_ptr[offset]);
+        program(registers, input, code_ptr[offset], output);
     }
 
     void Compile();
@@ -114,7 +114,7 @@ private:
     /// Branches that need to be fixed up once the entire shader program is compiled
     std::vector<std::pair<Gen::FixupBranch, unsigned>> fixup_branches;
 
-    using CompiledShader = void(void* registers, const void *input, const u8* start_addr);
+    using CompiledShader = void(void* registers, const void *input, const u8* start_addr, void *output);
     CompiledShader* program = nullptr;
 };