4 files changed, 81 insertions, 64 deletions

diff --git a/src/video_core/command_processor.cpp b/src/video_core/command_processor.cpp
index ad0da796e..c29a3fe51 100644
--- a/src/video_core/command_processor.cpp
+++ b/src/video_core/command_processor.cpp
@@ -149,7 +149,8 @@ static void WritePicaReg(u32 id, u32 value, u32 mask) {
                         // Send to vertex shader
                         if (g_debug_context)
                             g_debug_context->OnEvent(DebugContext::Event::VertexShaderInvocation, static_cast<void*>(&immediate_input));
-                        Shader::OutputVertex output = g_state.vs.Run(shader_unit, immediate_input, regs.vs.num_input_attributes+1);
+                        g_state.vs.Run(shader_unit, immediate_input, regs.vs.num_input_attributes+1);
+                        Shader::OutputVertex output_vertex = shader_unit.output_registers.ToVertex(regs.vs);
 
                         // Send to renderer
                         using Pica::Shader::OutputVertex;
@@ -157,7 +158,7 @@ static void WritePicaReg(u32 id, u32 value, u32 mask) {
                             VideoCore::g_renderer->Rasterizer()->AddTriangle(v0, v1, v2);
                         };
 
-                        g_state.primitive_assembler.SubmitVertex(output, AddTriangle);
+                        g_state.primitive_assembler.SubmitVertex(output_vertex, AddTriangle);
                     }
                 }
             }
@@ -231,7 +232,7 @@ static void WritePicaReg(u32 id, u32 value, u32 mask) {
             // The size has been tuned for optimal balance between hit-rate and the cost of lookup
             const size_t VERTEX_CACHE_SIZE = 32;
             std::array<u16, VERTEX_CACHE_SIZE> vertex_cache_ids;
-            std::array<Shader::OutputVertex, VERTEX_CACHE_SIZE> vertex_cache;
+            std::array<Shader::OutputRegisters, VERTEX_CACHE_SIZE> vertex_cache;
 
             unsigned int vertex_cache_pos = 0;
             vertex_cache_ids.fill(-1);
@@ -249,7 +250,7 @@ static void WritePicaReg(u32 id, u32 value, u32 mask) {
                 ASSERT(vertex != -1);
 
                 bool vertex_cache_hit = false;
-                Shader::OutputVertex output;
+                Shader::OutputRegisters output_registers;
 
                 if (is_indexed) {
                     if (g_debug_context && Pica::g_debug_context->recorder) {
@@ -259,7 +260,7 @@ static void WritePicaReg(u32 id, u32 value, u32 mask) {
 
                     for (unsigned int i = 0; i < VERTEX_CACHE_SIZE; ++i) {
                         if (vertex == vertex_cache_ids[i]) {
-                            output = vertex_cache[i];
+                            output_registers = vertex_cache[i];
                             vertex_cache_hit = true;
                             break;
                         }
@@ -274,15 +275,19 @@ static void WritePicaReg(u32 id, u32 value, u32 mask) {
                     // Send to vertex shader
                     if (g_debug_context)
                         g_debug_context->OnEvent(DebugContext::Event::VertexShaderInvocation, (void*)&input);
-                    output = g_state.vs.Run(shader_unit, input, loader.GetNumTotalAttributes());
+                    g_state.vs.Run(shader_unit, input, loader.GetNumTotalAttributes());
+                    output_registers = shader_unit.output_registers;
 
                     if (is_indexed) {
-                        vertex_cache[vertex_cache_pos] = output;
+                        vertex_cache[vertex_cache_pos] = output_registers;
                         vertex_cache_ids[vertex_cache_pos] = vertex;
                         vertex_cache_pos = (vertex_cache_pos + 1) % VERTEX_CACHE_SIZE;
                     }
                 }
 
+                // Retreive vertex from register data
+                Shader::OutputVertex output_vertex = output_registers.ToVertex(regs.vs);
+
                 // Send to renderer
                 using Pica::Shader::OutputVertex;
                 auto AddTriangle = [](
@@ -290,7 +295,7 @@ static void WritePicaReg(u32 id, u32 value, u32 mask) {
                     VideoCore::g_renderer->Rasterizer()->AddTriangle(v0, v1, v2);
                 };
 
-                primitive_assembler.SubmitVertex(output, AddTriangle);
+                primitive_assembler.SubmitVertex(output_vertex, AddTriangle);
             }
 
             for (auto& range : memory_accesses.ranges) {
diff --git a/src/video_core/shader/shader.cpp b/src/video_core/shader/shader.cpp
index 161097610..f565e2c91 100644
--- a/src/video_core/shader/shader.cpp
+++ b/src/video_core/shader/shader.cpp
@@ -30,6 +30,58 @@ namespace Pica {
 
 namespace Shader {
 
+OutputVertex OutputRegisters::ToVertex(const Regs::ShaderConfig& config) {
+    // 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;
+
+        if ((config.output_mask & (1 << i)) == 0)
+            continue;
+
+        const auto& output_register_map = g_state.regs.vs_output_attributes[index];
+
+        u32 semantics[4] = {
+            output_register_map.map_x, output_register_map.map_y,
+            output_register_map.map_z, output_register_map.map_w
+        };
+
+        for (unsigned comp = 0; comp < 4; ++comp) {
+            float24* out = ((float24*)&ret) + semantics[comp];
+            if (semantics[comp] != Regs::VSOutputAttributes::INVALID) {
+                *out = value[i][comp];
+            } else {
+                // Zero output so that attributes which aren't output won't have denormals in them,
+                // which would slow us down later.
+                memset(out, 0, sizeof(*out));
+            }
+        }
+
+        index++;
+    }
+
+    // The hardware takes the absolute and saturates vertex colors like this, *before* doing interpolation
+    for (unsigned i = 0; i < 4; ++i) {
+        ret.color[i] = float24::FromFloat32(
+            std::fmin(std::fabs(ret.color[i].ToFloat32()), 1.0f));
+    }
+
+    LOG_TRACE(HW_GPU, "Output vertex: pos(%.2f, %.2f, %.2f, %.2f), quat(%.2f, %.2f, %.2f, %.2f), "
+        "col(%.2f, %.2f, %.2f, %.2f), tc0(%.2f, %.2f), view(%.2f, %.2f, %.2f)",
+        ret.pos.x.ToFloat32(), ret.pos.y.ToFloat32(), ret.pos.z.ToFloat32(), ret.pos.w.ToFloat32(),
+        ret.quat.x.ToFloat32(), ret.quat.y.ToFloat32(), ret.quat.z.ToFloat32(), ret.quat.w.ToFloat32(),
+        ret.color.x.ToFloat32(), ret.color.y.ToFloat32(), ret.color.z.ToFloat32(), ret.color.w.ToFloat32(),
+        ret.tc0.u().ToFloat32(), ret.tc0.v().ToFloat32(),
+        ret.view.x.ToFloat32(), ret.view.y.ToFloat32(), ret.view.z.ToFloat32());
+
+    return ret;
+}
+
 #ifdef ARCHITECTURE_x86_64
 static std::unordered_map<u64, std::unique_ptr<JitShader>> shader_map;
 static const JitShader* jit_shader;
@@ -62,7 +114,7 @@ void ShaderSetup::Setup() {
 
 MICROPROFILE_DEFINE(GPU_Shader, "GPU", "Shader", MP_RGB(50, 50, 240));
 
-OutputVertex ShaderSetup::Run(UnitState<false>& state, const InputVertex& input, int num_attributes) {
+void ShaderSetup::Run(UnitState<false>& state, const InputVertex& input, int num_attributes) {
     auto& config = g_state.regs.vs;
     auto& setup = g_state.vs;
 
@@ -89,55 +141,6 @@ OutputVertex ShaderSetup::Run(UnitState<false>& state, const InputVertex& input,
     RunInterpreter(setup, state, config.main_offset);
 #endif // ARCHITECTURE_x86_64
 
-    // 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;
-
-        if ((g_state.regs.vs.output_mask & (1 << i)) == 0)
-            continue;
-
-        const auto& output_register_map = g_state.regs.vs_output_attributes[index]; // TODO: Don't hardcode VS here
-
-        u32 semantics[4] = {
-            output_register_map.map_x, output_register_map.map_y,
-            output_register_map.map_z, output_register_map.map_w
-        };
-
-        for (unsigned comp = 0; comp < 4; ++comp) {
-            float24* out = ((float24*)&ret) + semantics[comp];
-            if (semantics[comp] != Regs::VSOutputAttributes::INVALID) {
-                *out = state.registers.output[i][comp];
-            } else {
-                // Zero output so that attributes which aren't output won't have denormals in them,
-                // which would slow us down later.
-                memset(out, 0, sizeof(*out));
-            }
-        }
-
-        index++;
-    }
-
-    // The hardware takes the absolute and saturates vertex colors like this, *before* doing interpolation
-    for (unsigned i = 0; i < 4; ++i) {
-        ret.color[i] = float24::FromFloat32(
-            std::fmin(std::fabs(ret.color[i].ToFloat32()), 1.0f));
-    }
-
-    LOG_TRACE(HW_GPU, "Output vertex: pos(%.2f, %.2f, %.2f, %.2f), quat(%.2f, %.2f, %.2f, %.2f), "
-        "col(%.2f, %.2f, %.2f, %.2f), tc0(%.2f, %.2f), view(%.2f, %.2f, %.2f)",
-        ret.pos.x.ToFloat32(), ret.pos.y.ToFloat32(), ret.pos.z.ToFloat32(), ret.pos.w.ToFloat32(),
-        ret.quat.x.ToFloat32(), ret.quat.y.ToFloat32(), ret.quat.z.ToFloat32(), ret.quat.w.ToFloat32(),
-        ret.color.x.ToFloat32(), ret.color.y.ToFloat32(), ret.color.z.ToFloat32(), ret.color.w.ToFloat32(),
-        ret.tc0.u().ToFloat32(), ret.tc0.v().ToFloat32(),
-        ret.view.x.ToFloat32(), ret.view.y.ToFloat32(), ret.view.z.ToFloat32());
-
-    return ret;
 }
 
 DebugData<true> ShaderSetup::ProduceDebugInfo(const InputVertex& input, int num_attributes, const Regs::ShaderConfig& config, const ShaderSetup& setup) {
diff --git a/src/video_core/shader/shader.h b/src/video_core/shader/shader.h
index 84898f21c..fee16df62 100644
--- a/src/video_core/shader/shader.h
+++ b/src/video_core/shader/shader.h
@@ -84,6 +84,15 @@ struct OutputVertex {
 static_assert(std::is_pod<OutputVertex>::value, "Structure is not POD");
 static_assert(sizeof(OutputVertex) == 32 * sizeof(float), "OutputVertex has invalid size");
 
+struct OutputRegisters {
+    OutputRegisters() = default;
+
+    alignas(16) Math::Vec4<float24> value[16];
+
+    OutputVertex ToVertex(const Regs::ShaderConfig& config);
+};
+static_assert(std::is_pod<OutputRegisters>::value, "Structure is not POD");
+
 // Helper structure used to keep track of data useful for inspection of shader emulation
 template<bool full_debugging>
 struct DebugData;
@@ -267,11 +276,12 @@ struct UnitState {
         // The registers are accessed by the shader JIT using SSE instructions, and are therefore
         // required to be 16-byte aligned.
         alignas(16) Math::Vec4<float24> input[16];
-        alignas(16) Math::Vec4<float24> output[16];
         alignas(16) Math::Vec4<float24> temporary[16];
     } registers;
     static_assert(std::is_pod<Registers>::value, "Structure is not POD");
 
+    OutputRegisters output_registers;
+
     bool conditional_code[2];
 
     // Two Address registers and one loop counter
@@ -297,7 +307,7 @@ struct UnitState {
     static size_t OutputOffset(const DestRegister& reg) {
         switch (reg.GetRegisterType()) {
         case RegisterType::Output:
-            return offsetof(UnitState, registers.output) + reg.GetIndex()*sizeof(Math::Vec4<float24>);
+            return offsetof(UnitState, output_registers.value) + reg.GetIndex()*sizeof(Math::Vec4<float24>);
 
         case RegisterType::Temporary:
             return offsetof(UnitState, registers.temporary) + reg.GetIndex()*sizeof(Math::Vec4<float24>);
@@ -354,9 +364,8 @@ struct ShaderSetup {
      * @param state Shader unit state, must be setup per shader and per shader unit
      * @param input Input vertex into the shader
      * @param num_attributes The number of vertex shader attributes
-     * @return The output vertex, after having been processed by the vertex shader
      */
-    OutputVertex Run(UnitState<false>& state, const InputVertex& input, int num_attributes);
+    void Run(UnitState<false>& state, const InputVertex& input, int num_attributes);
 
     /**
      * Produce debug information based on the given shader and input vertex
diff --git a/src/video_core/shader/shader_interpreter.cpp b/src/video_core/shader/shader_interpreter.cpp
index 714e8bfd5..b1eadc071 100644
--- a/src/video_core/shader/shader_interpreter.cpp
+++ b/src/video_core/shader/shader_interpreter.cpp
@@ -144,7 +144,7 @@ void RunInterpreter(const ShaderSetup& setup, UnitState<Debug>& state, unsigned
                 src2[3] = src2[3] * float24::FromFloat32(-1);
             }
 
-            float24* dest = (instr.common.dest.Value() < 0x10) ? &state.registers.output[instr.common.dest.Value().GetIndex()][0]
+            float24* dest = (instr.common.dest.Value() < 0x10) ? &state.output_registers.value[instr.common.dest.Value().GetIndex()][0]
                         : (instr.common.dest.Value() < 0x20) ? &state.registers.temporary[instr.common.dest.Value().GetIndex()][0]
                         : dummy_vec4_float24;
 
@@ -483,7 +483,7 @@ void RunInterpreter(const ShaderSetup& setup, UnitState<Debug>& state, unsigned
                     src3[3] = src3[3] * float24::FromFloat32(-1);
                 }
 
-                float24* dest = (instr.mad.dest.Value() < 0x10) ? &state.registers.output[instr.mad.dest.Value().GetIndex()][0]
+                float24* dest = (instr.mad.dest.Value() < 0x10) ? &state.output_registers.value[instr.mad.dest.Value().GetIndex()][0]
                             : (instr.mad.dest.Value() < 0x20) ? &state.registers.temporary[instr.mad.dest.Value().GetIndex()][0]
                             : dummy_vec4_float24;