Convert the pre-rotation to just the matrix multiplication.

Common/GPU/Vulkan/VulkanContext.h

@@ -441,6 +441,10 @@ public:
 		return winsys_;
 	}
 
+	bool SupportsPreRotation() const {
+		return surfCapabilities_.supportedTransforms != VK_SURFACE_TRANSFORM_IDENTITY_BIT_KHR;
+	}
+
 private:
 	bool ChooseQueue();
 

GPU/Common/ShaderUniforms.cpp

@@ -72,6 +72,22 @@ void CalcCullRange(float minValues[4], float maxValues[4], bool flipViewport, bo
 	maxValues[3] = NAN;
 }
 
+void UpdateRotation(float rotMatrix[4], bool useBufferedRendering) {
+	if (useBufferedRendering) {
+		rotMatrix[0] = 1.0f;
+		rotMatrix[1] = 0.0f;
+		rotMatrix[2] = 0.0f;
+		rotMatrix[3] = 1.0f;
+	} else {
+		const DisplayRotation rotation = g_display.rotation;
+		// The others are ROTATE_90 and so on.
+		rotMatrix[0] = rotation == DisplayRotation::ROTATE_0 ? 1.0 : (rotation == DisplayRotation::ROTATE_180 ? -1.0 : 0.0);
+		rotMatrix[1] = rotation == DisplayRotation::ROTATE_90 ? 1.0 : (rotation == DisplayRotation::ROTATE_270 ? -1.0 : 0.0);
+		rotMatrix[2] = rotation == DisplayRotation::ROTATE_270 ? 1.0 : (rotation == DisplayRotation::ROTATE_90 ? -1.0 : 0.0);
+		rotMatrix[3] = rotation == DisplayRotation::ROTATE_0 ? 1.0 : (rotation == DisplayRotation::ROTATE_180 ? -1.0 : 0.0);
+	}
+}
+
 void BaseUpdateUniforms(UB_VS_FS_Base *ub, uint64_t dirtyUniforms, bool flipViewport, bool useBufferedRendering) {
 	if (dirtyUniforms & DIRTY_TEXENV) {
 		Uint8x3ToFloat3(ub->texEnvColor, gstate.texenvcolor);
@@ -136,8 +152,7 @@ void BaseUpdateUniforms(UB_VS_FS_Base *ub, uint64_t dirtyUniforms, bool flipView
 		}
 
 		CopyMatrix4x4(ub->proj, flippedMatrix.getReadPtr());
-
-		ub->rotation = useBufferedRendering ? 0 : (float)g_display.rotation;
+		UpdateRotation(ub->rotation, useBufferedRendering);  // TODO: This can be done a lot less often.
 	}
 
 	if (dirtyUniforms & DIRTY_PROJTHROUGHMATRIX) {
@@ -155,8 +170,7 @@ void BaseUpdateUniforms(UB_VS_FS_Base *ub, uint64_t dirtyUniforms, bool flipView
 		}
 
 		CopyMatrix4x4(ub->proj_through, proj_through.getReadPtr());
-
-		ub->rotation = useBufferedRendering ? 0 : (float)g_display.rotation;
+		UpdateRotation(ub->rotation, useBufferedRendering);  // TODO: This can be done a lot less often.
 	}
 
 	// Transform

GPU/Common/ShaderUniforms.h

@@ -25,6 +25,7 @@ struct alignas(16) UB_VS_FS_Base {
 	float view[12];
 	float world[12];
 	float tex[12];
+	float rotation[4];
 	float uvScaleOffset[4];
 	float depthRange[4];
 	float matAmbient[4];
@@ -39,11 +40,11 @@ struct alignas(16) UB_VS_FS_Base {
 	float texClamp[4];
 	float texClampOffset[2]; float fogCoef[2];
 	float blendFixA[3]; float stencilReplaceValue;
-	float blendFixB[3]; float rotation;
+	float blendFixB[3]; float padding3;
 	// VR stuff is to go here, later. For normal drawing, we can then get away
 	// with just uploading the first 448 bytes of the struct (up to and including fogCoef).
 };
-static_assert(sizeof(UB_VS_FS_Base) == 480, "UB_VS_FS_Base should be 496 bytes");
+static_assert(sizeof(UB_VS_FS_Base) == 496, "UB_VS_FS_Base should be 496 bytes");
 
 static const char * const ub_baseStr =
 R"(  mat4 u_proj;
@@ -51,6 +52,7 @@ R"(  mat4 u_proj;
   mat3x4 u_view;
   mat3x4 u_world;
   mat3x4 u_texmtx;
+  vec4 u_rotation;
   vec4 u_uvscaleoffset;
   vec4 u_depthRange;
   vec4 u_matambientalpha;
@@ -66,7 +68,7 @@ R"(  mat4 u_proj;
   vec4 u_texclamp;
   vec2 u_texclampoff; vec2 u_fogcoef;
   vec3 u_blendFixA; float u_stencilReplaceValue;
-  vec3 u_blendFixB; float u_rotation;
+  vec3 u_blendFixB; float pad3;
 )";
 
 // 512 bytes. Would like to shrink more. Some colors only have 8-bit precision and we expand

GPU/Common/VertexShaderGenerator.cpp

@@ -1264,12 +1264,8 @@ bool GenerateVertexShader(const VShaderID &id, char *buffer, const ShaderLanguag
 
 	if (compat.shaderLanguage == GLSL_VULKAN) {
 		// Apply rotation from the uniform.
-		WRITE(p, "  mat2 displayRotation = mat2(\n");
-		WRITE(p, "    u_rotation == 0.0 ? 1.0 : (u_rotation == 2.0 ? -1.0 : 0.0), u_rotation == 1.0 ? 1.0 : (u_rotation == 3.0 ? -1.0 : 0.0),\n");
-		WRITE(p, "    u_rotation == 3.0 ? 1.0 : (u_rotation == 1.0 ? -1.0 : 0.0), u_rotation == 0.0 ? 1.0 : (u_rotation == 2.0 ? -1.0 : 0.0)\n");
-		WRITE(p, "  );\n");
-
-		WRITE(p, "  outPos.xy = mul(displayRotation, outPos.xy);\n");
+		// NOTE: This is only needed on platforms that support pre-rotation.
+		WRITE(p, "  outPos.xy = mul(mat2(u_rotation), outPos.xy);\n");
 	}
 
 	// We've named the output gl_Position in HLSL as well.