Merge pull request #17582 from hrydgard/uvscaleoffset-fix

Update the uvscale uniform a bit more conservatively on framebuffer-texture changes
hrydgard · Jun 15, 2023 · fee65be · fee65be
2 parents 4a4cd3d + def09bf
commit fee65be
Show file tree

Hide file tree

Showing 6 changed files with 53 additions and 29 deletions.
diff --git a/GPU/Common/ShaderUniforms.cpp b/GPU/Common/ShaderUniforms.cpp
@@ -215,12 +215,16 @@ void BaseUpdateUniforms(UB_VS_FS_Base *ub, uint64_t dirtyUniforms, bool flipView
 
 	// Texturing
 	if (dirtyUniforms & DIRTY_UVSCALEOFFSET) {
-		const float invW = 1.0f / (float)gstate_c.curTextureWidth;
-		const float invH = 1.0f / (float)gstate_c.curTextureHeight;
-		const int w = gstate.getTextureWidth(0);
-		const int h = gstate.getTextureHeight(0);
-		const float widthFactor = (float)w * invW;
-		const float heightFactor = (float)h * invH;
+		float widthFactor = 1.0f;
+		float heightFactor = 1.0f;
+		if (gstate_c.textureIsFramebuffer) {
+			const float invW = 1.0f / (float)gstate_c.curTextureWidth;
+			const float invH = 1.0f / (float)gstate_c.curTextureHeight;
+			const int w = gstate.getTextureWidth(0);
+			const int h = gstate.getTextureHeight(0);
+			widthFactor = (float)w * invW;
+			heightFactor = (float)h * invH;
+		}
 		if (gstate_c.submitType == SubmitType::HW_BEZIER || gstate_c.submitType == SubmitType::HW_SPLINE) {
 			// When we are generating UV coordinates through the bezier/spline, we need to apply the scaling.
 			// However, this is missing a check that we're not getting our UV:s supplied for us in the vertices.

diff --git a/GPU/Common/TextureCacheCommon.cpp b/GPU/Common/TextureCacheCommon.cpp
@@ -1141,23 +1141,26 @@ void TextureCacheCommon::SetTextureFramebuffer(const AttachCandidate &candidate)
 		bool needsDepthXSwizzle = depthUpperBits == 2;
 
 		// We need to force it, since we may have set it on a texture before attaching.
-		gstate_c.curTextureWidth = framebuffer->bufferWidth;
-		gstate_c.curTextureHeight = framebuffer->bufferHeight;
-
+		int texWidth = framebuffer->bufferWidth;
+		int texHeight = framebuffer->bufferHeight;
 		if (candidate.channel == RASTER_COLOR && gstate.getTextureFormat() == GE_TFMT_CLUT8 && framebuffer->fb_format == GE_FORMAT_5551 && PSP_CoreParameter().compat.flags().SOCOMClut8Replacement) {
 			// See #16210. UV must be adjusted as if the texture was twice the width.
-			gstate_c.curTextureWidth *= 2.0f;
+			texWidth *= 2.0f;
 		}
 
 		if (needsDepthXSwizzle) {
-			gstate_c.curTextureWidth = RoundUpToPowerOf2(gstate_c.curTextureWidth);
+			texWidth = RoundUpToPowerOf2(texWidth);
 		}
 
+		gstate_c.curTextureWidth = texWidth;
+		gstate_c.curTextureHeight = texHeight;
+		gstate_c.SetTextureIsFramebuffer(true);
+		gstate_c.SetTextureIsBGRA(false);
+
 		if ((gstate_c.curTextureXOffset == 0) != (fbInfo.xOffset == 0) || (gstate_c.curTextureYOffset == 0) != (fbInfo.yOffset == 0)) {
 			gstate_c.Dirty(DIRTY_FRAGMENTSHADER_STATE);
 		}
-		gstate_c.SetTextureIsBGRA(false);
-		gstate_c.SetTextureIsFramebuffer(true);
+
 		gstate_c.curTextureXOffset = fbInfo.xOffset;
 		gstate_c.curTextureYOffset = fbInfo.yOffset;
 		u32 texW = (u32)gstate.getTextureWidth(0);
@@ -2364,6 +2367,7 @@ void TextureCacheCommon::ApplyTextureFramebuffer(VirtualFramebuffer *framebuffer
 		gpuStats.numDepal++;
 
 		gstate_c.curTextureWidth = texWidth;
+		gstate_c.Dirty(DIRTY_UVSCALEOFFSET);
 
 		draw_->BindTexture(0, nullptr);
 		framebufferManager_->RebindFramebuffer("ApplyTextureFramebuffer");
@@ -2465,6 +2469,7 @@ void TextureCacheCommon::ApplyTextureDepal(TexCacheEntry *entry) {
 	gpuStats.numDepal++;
 
 	gstate_c.curTextureWidth = texWidth;
+	gstate_c.Dirty(DIRTY_UVSCALEOFFSET);
 
 	draw_->BindTexture(0, nullptr);
 	framebufferManager_->RebindFramebuffer("ApplyTextureFramebuffer");

diff --git a/GPU/Directx9/ShaderManagerDX9.cpp b/GPU/Directx9/ShaderManagerDX9.cpp
@@ -436,12 +436,16 @@ void ShaderManagerDX9::VSUpdateUniforms(u64 dirtyUniforms) {
 
 	// Texturing
 	if (dirtyUniforms & DIRTY_UVSCALEOFFSET) {
-		const float invW = 1.0f / (float)gstate_c.curTextureWidth;
-		const float invH = 1.0f / (float)gstate_c.curTextureHeight;
-		const int w = gstate.getTextureWidth(0);
-		const int h = gstate.getTextureHeight(0);
-		const float widthFactor = (float)w * invW;
-		const float heightFactor = (float)h * invH;
+		float widthFactor = 1.0f;
+		float heightFactor = 1.0f;
+		if (gstate_c.textureIsFramebuffer) {
+			const float invW = 1.0f / (float)gstate_c.curTextureWidth;
+			const float invH = 1.0f / (float)gstate_c.curTextureHeight;
+			const int w = gstate.getTextureWidth(0);
+			const int h = gstate.getTextureHeight(0);
+			widthFactor = (float)w * invW;
+			heightFactor = (float)h * invH;
+		}
 		float uvscaleoff[4];
 		uvscaleoff[0] = widthFactor;
 		uvscaleoff[1] = heightFactor;

diff --git a/GPU/GLES/ShaderManagerGLES.cpp b/GPU/GLES/ShaderManagerGLES.cpp
@@ -504,12 +504,16 @@ void LinkedShader::UpdateUniforms(const ShaderID &vsid, bool useBufferedRenderin
 		render_->SetUniformF(&u_fogcoef, 2, fogcoef);
 	}
 	if (dirty & DIRTY_UVSCALEOFFSET) {
-		const float invW = 1.0f / (float)gstate_c.curTextureWidth;
-		const float invH = 1.0f / (float)gstate_c.curTextureHeight;
-		const int w = gstate.getTextureWidth(0);
-		const int h = gstate.getTextureHeight(0);
-		const float widthFactor = (float)w * invW;
-		const float heightFactor = (float)h * invH;
+		float widthFactor = 1.0f;
+		float heightFactor = 1.0f;
+		if (gstate_c.textureIsFramebuffer) {
+			const float invW = 1.0f / (float)gstate_c.curTextureWidth;
+			const float invH = 1.0f / (float)gstate_c.curTextureHeight;
+			const int w = gstate.getTextureWidth(0);
+			const int h = gstate.getTextureHeight(0);
+			widthFactor = (float)w * invW;
+			heightFactor = (float)h * invH;
+		}
 		float uvscaleoff[4];
 		if (gstate_c.submitType == SubmitType::HW_BEZIER || gstate_c.submitType == SubmitType::HW_SPLINE) {
 			// When we are generating UV coordinates through the bezier/spline, we need to apply the scaling.

diff --git a/GPU/GPUCommonHW.cpp b/GPU/GPUCommonHW.cpp
@@ -1192,7 +1192,8 @@ void GPUCommonHW::Execute_Bezier(u32 op, u32 diff) {
 
 	SetDrawType(DRAW_BEZIER, PatchPrimToPrim(surface.primType));
 
-	gstate_c.Dirty(DIRTY_RASTER_STATE | DIRTY_VERTEXSHADER_STATE | DIRTY_GEOMETRYSHADER_STATE);
+	// We need to dirty UVSCALEOFFSET here because we look at the submit type when setting that uniform.
+	gstate_c.Dirty(DIRTY_RASTER_STATE | DIRTY_VERTEXSHADER_STATE | DIRTY_GEOMETRYSHADER_STATE | DIRTY_UVSCALEOFFSET);
 	if (drawEngineCommon_->CanUseHardwareTessellation(surface.primType)) {
 		gstate_c.submitType = SubmitType::HW_BEZIER;
 		if (gstate_c.spline_num_points_u != surface.num_points_u) {
@@ -1207,7 +1208,7 @@ void GPUCommonHW::Execute_Bezier(u32 op, u32 diff) {
 	UpdateUVScaleOffset();
 	drawEngineCommon_->SubmitCurve(control_points, indices, surface, gstate.vertType, &bytesRead, "bezier");
 
-	gstate_c.Dirty(DIRTY_RASTER_STATE | DIRTY_VERTEXSHADER_STATE | DIRTY_GEOMETRYSHADER_STATE);
+	gstate_c.Dirty(DIRTY_RASTER_STATE | DIRTY_VERTEXSHADER_STATE | DIRTY_GEOMETRYSHADER_STATE | DIRTY_UVSCALEOFFSET);
 	gstate_c.submitType = SubmitType::DRAW;
 
 	// After drawing, we advance pointers - see SubmitPrim which does the same.
@@ -1265,7 +1266,8 @@ void GPUCommonHW::Execute_Spline(u32 op, u32 diff) {
 
 	SetDrawType(DRAW_SPLINE, PatchPrimToPrim(surface.primType));
 
-	gstate_c.Dirty(DIRTY_RASTER_STATE | DIRTY_VERTEXSHADER_STATE | DIRTY_GEOMETRYSHADER_STATE);
+	// We need to dirty UVSCALEOFFSET here because we look at the submit type when setting that uniform.
+	gstate_c.Dirty(DIRTY_RASTER_STATE | DIRTY_VERTEXSHADER_STATE | DIRTY_GEOMETRYSHADER_STATE | DIRTY_UVSCALEOFFSET);
 	if (drawEngineCommon_->CanUseHardwareTessellation(surface.primType)) {
 		gstate_c.submitType = SubmitType::HW_SPLINE;
 		if (gstate_c.spline_num_points_u != surface.num_points_u) {
@@ -1280,7 +1282,7 @@ void GPUCommonHW::Execute_Spline(u32 op, u32 diff) {
 	UpdateUVScaleOffset();
 	drawEngineCommon_->SubmitCurve(control_points, indices, surface, gstate.vertType, &bytesRead, "spline");
 
-	gstate_c.Dirty(DIRTY_RASTER_STATE | DIRTY_VERTEXSHADER_STATE | DIRTY_GEOMETRYSHADER_STATE);
+	gstate_c.Dirty(DIRTY_RASTER_STATE | DIRTY_VERTEXSHADER_STATE | DIRTY_GEOMETRYSHADER_STATE | DIRTY_UVSCALEOFFSET);
 	gstate_c.submitType = SubmitType::DRAW;
 
 	// After drawing, we advance pointers - see SubmitPrim which does the same.

diff --git a/GPU/GPUState.h b/GPU/GPUState.h
@@ -584,6 +584,11 @@ struct GPUStateCache {
 		if (textureIsFramebuffer != isFramebuffer) {
 			textureIsFramebuffer = isFramebuffer;
 			Dirty(DIRTY_UVSCALEOFFSET);
+		} else if (isFramebuffer) {
+			// Always dirty if it's a framebuffer, since the uniform value depends both
+			// on the specified texture size and the bound texture size. Makes things easier.
+			// TODO: Look at this again later.
+			Dirty(DIRTY_UVSCALEOFFSET);
 		}
 	}
 	void SetUseFlags(u32 newFlags) {