Merge pull request #11287 from unknownbrackets/gpu-clipping

Improve clipping/culling handling in software renderer

GPU/Common/GPUStateUtils.cpp

@@ -676,7 +676,7 @@ void ConvertViewportAndScissor(bool useBufferedRendering, float renderWidth, flo
 		float minz = gstate.getDepthRangeMin();
 		float maxz = gstate.getDepthRangeMax();
 
-		if (gstate.isClippingEnabled() && (minz == 0 || maxz == 65535)) {
+		if (gstate.isDepthClampEnabled() && (minz == 0 || maxz == 65535)) {
 			// Here, we should "clamp."  But clamping per fragment would be slow.
 			// So, instead, we just increase the available range and hope.
 			// If depthSliceFactor is 4, it means (75% / 2) of the depth lies in each direction.

GPU/GPUState.h

@@ -58,7 +58,7 @@ struct GPUgstate {
 				region2,
 				lightingEnable,
 				lightEnable[4],
-				clipEnable,
+				depthClampEnable,
 				cullfaceEnable,
 				textureMapEnable,  // 0x1E GE_CMD_TEXTUREMAPENABLE
 				fogEnable,
@@ -392,8 +392,9 @@ struct GPUgstate {
 	int getRegionX2() const { return (region2 & 0x3FF); }
 	int getRegionY2() const { return (region2 >> 10) & 0x3FF; }
 
+	bool isDepthClampEnabled() const { return depthClampEnable & 1; }
+
 	// Note that the X1/Y1/Z1 here does not mean the upper-left corner, but half the dimensions. X2/Y2/Z2 are the center.
-	bool isClippingEnabled() const { return clipEnable & 1; }
 	float getViewportXScale() const { return getFloat24(viewportxscale); }
 	float getViewportYScale() const { return getFloat24(viewportyscale); }
 	float getViewportZScale() const { return getFloat24(viewportzscale); }

GPU/Software/Clipper.cpp

@@ -39,6 +39,7 @@ enum {
 static inline int CalcClipMask(const ClipCoords& v)
 {
 	int mask = 0;
+	// This checks `x / w` compared to 1 or -1, skipping the division.
 	if (v.x > v.w) mask |= CLIP_POS_X_BIT;
 	if (v.x < -v.w) mask |= CLIP_NEG_X_BIT;
 	if (v.y > v.w) mask |= CLIP_POS_Y_BIT;
@@ -255,11 +256,7 @@ void ProcessLine(VertexData& v0, VertexData& v1)
 		return;
 	}
 
-	if (mask && gstate.isClippingEnabled()) {
-		// discard if any vertex is outside the near clipping plane
-		if (mask & CLIP_NEG_Z_BIT)
-			return;
-
+	if (mask) {
 		CLIP_LINE(CLIP_POS_X_BIT, -1,  0,  0, 1);
 		CLIP_LINE(CLIP_NEG_X_BIT,  1,  0,  0, 1);
 		CLIP_LINE(CLIP_POS_Y_BIT,  0, -1,  0, 1);
@@ -303,11 +300,7 @@ void ProcessTriangle(VertexData& v0, VertexData& v1, VertexData& v2)
 	mask |= CalcClipMask(v1.clippos);
 	mask |= CalcClipMask(v2.clippos);
 
-	if (mask && gstate.isClippingEnabled()) {
-		// discard if any vertex is outside the near clipping plane
-		if (mask & CLIP_NEG_Z_BIT)
-			return;
-
+	if (mask) {
 		for (int i = 0; i < 3; i += 3) {
 			int vlist[2][2*6+1];
 			int *inlist = vlist[0], *outlist = vlist[1];

GPU/Software/TransformUnit.cpp

@@ -104,16 +104,23 @@ static inline ScreenCoords ClipToScreenInternal(const ClipCoords& coords, bool *
 	float y = coords.y * yScale / coords.w + yCenter;
 	float z = coords.z * zScale / coords.w + zCenter;
 
+	// Account for rounding for X and Y.
+	const float SCREEN_BOUND = 4095.0f + (15.0f / 16.0f) + 0.375f;
+	// TODO: Validate actual rounding range.
+	const float DEPTH_BOUND = 65535.5f;
+
 	// This matches hardware tests - depth is clamped when this flag is on.
-	if (gstate.isClippingEnabled()) {
+	if (gstate.isDepthClampEnabled()) {
+		// Note: if the depth is clamped, the outside_range_flag should NOT be set, even for x and y.
 		if (z < 0.f)
 			z = 0.f;
-		if (z > 65535.f)
-			z = 65535.f;
-	}
-
-	if (outside_range_flag && (x > 4095.9375f || y > 4095.9375f || x < 0 || y < 0 || z < 0 || z > 65535.f))
+		else if (z > 65535.0f)
+			z = 65535.0f;
+		else if (outside_range_flag && (x >= SCREEN_BOUND || y >= SCREEN_BOUND || x < 0 || y < 0))
+			*outside_range_flag = true;
+	} else if (outside_range_flag && (x > 4095.9675f || y >= SCREEN_BOUND || x < 0 || y < 0 || z < 0 || z >= DEPTH_BOUND)) {
 		*outside_range_flag = true;
+	}
 
 	// 16 = 0xFFFF / 4095.9375
 	// Round up at 0.625 to the nearest subpixel.

GPU/Vulkan/StateMappingVulkan.cpp

@@ -247,7 +247,7 @@ void DrawEngineVulkan::ConvertStateToVulkanKey(FramebufferManagerVulkan &fbManag
 			// Set cull
 			bool wantCull = !gstate.isModeThrough() && prim != GE_PRIM_RECTANGLES && gstate.isCullEnabled();
 			key.cullMode = wantCull ? (gstate.getCullMode() ? VK_CULL_MODE_FRONT_BIT : VK_CULL_MODE_BACK_BIT) : VK_CULL_MODE_NONE;
-			key.depthClampEnable = gstate.isClippingEnabled() && gstate_c.Supports(GPU_SUPPORTS_DEPTH_CLAMP);
+			key.depthClampEnable = gstate.isDepthClampEnabled() && gstate_c.Supports(GPU_SUPPORTS_DEPTH_CLAMP);
 		}
 	}
 

Windows/GEDebugger/TabVertices.cpp

@@ -369,7 +369,7 @@ void CtrlMatrixList::GetColumnText(wchar_t *dest, int row, int col) {
 	if (row >= MATRIXLIST_ROW_BONE_0_0) {
 		int b = (row - MATRIXLIST_ROW_BONE_0_0) / 3;
 		int r = (row - MATRIXLIST_ROW_BONE_0_0) % 3;
-		int offset = (row - MATRIXLIST_ROW_BONE_0_0) * 4 + col - 1;
+		int offset = b * 12 + r + (col - 1) * 3;
 
 		switch (col) {
 		case MATRIXLIST_COL_NAME:
@@ -382,7 +382,7 @@ void CtrlMatrixList::GetColumnText(wchar_t *dest, int row, int col) {
 		}
 	} else if (row >= MATRIXLIST_ROW_TGEN_0) {
 		int r = row - MATRIXLIST_ROW_TGEN_0;
-		int offset = r * 4 + col - 1;
+		int offset = r + (col - 1) * 4;
 
 		switch (col) {
 		case MATRIXLIST_COL_NAME:
@@ -395,7 +395,7 @@ void CtrlMatrixList::GetColumnText(wchar_t *dest, int row, int col) {
 		}
 	} else if (row >= MATRIXLIST_ROW_PROJ_0) {
 		int r = row - MATRIXLIST_ROW_PROJ_0;
-		int offset = r * 4 + col - 1;
+		int offset = r + (col - 1) * 4;
 
 		switch (col) {
 		case MATRIXLIST_COL_NAME:
@@ -408,7 +408,7 @@ void CtrlMatrixList::GetColumnText(wchar_t *dest, int row, int col) {
 		}
 	} else if (row >= MATRIXLIST_ROW_VIEW_0) {
 		int r = row - MATRIXLIST_ROW_VIEW_0;
-		int offset = r * 4 + col - 1;
+		int offset = r + (col - 1) * 3;
 
 		switch (col) {
 		case MATRIXLIST_COL_NAME:
@@ -421,7 +421,7 @@ void CtrlMatrixList::GetColumnText(wchar_t *dest, int row, int col) {
 		}
 	} else {
 		int r = row - MATRIXLIST_ROW_WORLD_0;
-		int offset = r * 4 + col - 1;
+		int offset = r + (col - 1) * 3;
 
 		switch (col) {
 		case MATRIXLIST_COL_NAME:

Windows/GEDebugger/VertexPreview.cpp

@@ -494,7 +494,7 @@ void CGEDebugger::UpdatePrimPreview(u32 op, int which) {
 		}
 
 		if (texPreviewVao == 0) {
-			glDisableVertexAttribArray(previewProgram->a_position);
+			glDisableVertexAttribArray(texPreviewProgram->a_position);
 		}
 
 		secondWindow->End();