Shader: Store AttributeBuffers in GS output buffer

This also does the output masking early at EMIT time, instead of when a
triangle is sent to the vertex handler.

src/video_core/shader/shader.cpp

@@ -99,16 +99,16 @@ GSEmitter::~GSEmitter() {
     delete handlers;
 }
 
-void GSEmitter::Emit(Math::Vec4<float24> (&vertex)[16]) {
+void GSEmitter::Emit(Math::Vec4<float24> (&output_regs)[16]) {
     ASSERT(vertex_id < 3);
-    std::copy(std::begin(vertex), std::end(vertex), buffer[vertex_id].begin());
+    // TODO: This should be merged with UnitState::WriteOutput somehow
+    CopyRegistersToOutput(output_regs, output_mask, buffer[vertex_id]);
+
     if (prim_emit) {
         if (winding)
             handlers->winding_setter();
         for (size_t i = 0; i < buffer.size(); ++i) {
-            AttributeBuffer output;
+            handlers->vertex_handler(buffer[i]);
-            CopyRegistersToOutput(buffer[i].data(), output_mask, output);
-            handlers->vertex_handler(output);
         }
     }
 }

src/video_core/shader/shader.h

@@ -72,7 +72,7 @@ static_assert(sizeof(OutputVertex) == 24 * sizeof(float), "OutputVertex has inva
  * This structure contains state information for primitive emitting in geometry shader.
  */
 struct GSEmitter {
-    std::array<std::array<Math::Vec4<float24>, 16>, 3> buffer;
+    std::array<AttributeBuffer, 3> buffer;
     u8 vertex_id;
     bool prim_emit;
     bool winding;
@@ -87,7 +87,7 @@ struct GSEmitter {
 
     GSEmitter();
     ~GSEmitter();
-    void Emit(Math::Vec4<float24> (&vertex)[16]);
+    void Emit(Math::Vec4<float24> (&output_regs)[16]);
 };
 static_assert(std::is_standard_layout<GSEmitter>::value, "GSEmitter is not standard layout type");