Revert "Convert the pre-rotation to just the matrix multiplication."

After the previous commit, the optimization is no longer needed, and
unnecessarily bloats the uniform buffer.

This reverts commit 62bcb1785e809649ba88b49e24c5217f457fa040.

GPU/Common/ShaderUniforms.cpp

@@ -146,7 +146,8 @@ void BaseUpdateUniforms(UB_VS_FS_Base *ub, uint64_t dirtyUniforms, bool flipView
 		FlipProjMatrix(flippedMatrix);
 		ConvertProjMatrixToZeroToOneDepth(flippedMatrix);
 		CopyMatrix4x4(ub->proj, flippedMatrix.getReadPtr());
-		UpdateRotation(ub->rotation, useBufferedRendering);  // TODO: This can be done a lot less often.
+
+		ub->rotation = useBufferedRendering ? 0 : (float)g_display.rotation;
 	}
 
 	if (dirtyUniforms & DIRTY_PROJTHROUGHMATRIX) {
@@ -160,7 +161,8 @@ void BaseUpdateUniforms(UB_VS_FS_Base *ub, uint64_t dirtyUniforms, bool flipView
 		}
 
 		CopyMatrix4x4(ub->proj_through, proj_through.getReadPtr());
-		UpdateRotation(ub->rotation, useBufferedRendering);  // TODO: This can be done a lot less often.
+
+		ub->rotation = useBufferedRendering ? 0 : (float)g_display.rotation;
 	}
 
 	// Transform

GPU/Common/ShaderUniforms.h

@@ -25,7 +25,6 @@ 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];
@@ -40,11 +39,11 @@ 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 padding3;
+	float blendFixB[3]; float rotation;
 	// 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) == 496, "UB_VS_FS_Base should be 496 bytes");
+static_assert(sizeof(UB_VS_FS_Base) == 480, "UB_VS_FS_Base should be 480 bytes");
 
 static const char * const ub_baseStr =
 R"(  mat4 u_proj;
@@ -52,7 +51,6 @@ 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;
@@ -68,7 +66,7 @@ R"(  mat4 u_proj;
   vec4 u_texclamp;
   vec2 u_texclampoff; vec2 u_fogcoef;
   vec3 u_blendFixA; float u_stencilReplaceValue;
-  vec3 u_blendFixB; float pad3;
+  vec3 u_blendFixB; float u_rotation;
 )";
 
 // 512 bytes. Would like to shrink more. Some colors only have 8-bit precision and we expand

GPU/Common/VertexShaderGenerator.cpp

@@ -1264,8 +1264,12 @@ bool GenerateVertexShader(const VShaderID &id, char *buffer, const ShaderLanguag
 
 	if (compat.shaderLanguage == GLSL_VULKAN && gstate_c.Use(GPU_USE_PRE_ROTATION)) {
 		// Apply rotation from the uniform.
-		// TODO: This is only needed on platforms that support pre-rotation, and only in skip buffer effects mode.
-		WRITE(p, "  outPos.xy = mul(mat2(u_rotation), outPos.xy);\n");
+		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");
 	}
 
 	bool flipY = strlen(compat.viewportYSign) > 0;

GPU/Vulkan/TextureCacheVulkan.h

@@ -152,7 +152,6 @@ private:
 	Path cbufferPath_;
 	VulkanBuffer textureScaleCBuffer_;
 	bool cbufferInited_ = true;
-	bool cbufferFailed_ = false;
 };
 
 VkFormat getClutDestFormatVulkan(GEPaletteFormat format);