- fix MemcpyCommand not using the same lines for the threads as softpoly (visible as a race condition when screenblocks didn't start at top of screen)

src/polyrenderer/drawers/poly_triangle.cpp

@@ -68,17 +68,16 @@ void PolyTriangleDrawer::SetViewport(const DrawerCommandQueuePtr &queue, int x,
 	isBgraRenderTarget = dest_bgra;
 
 	int offsetx = clamp(x, 0, dest_width);
-	int offsety = clamp(y, 0, dest_height);
 	int pixelsize = dest_bgra ? 4 : 1;
 
 	int viewport_x = x - offsetx;
-	int viewport_y = y - offsety;
+	int viewport_y = y;
 	int viewport_width = width;
 	int viewport_height = height;
 
-	dest += (offsetx + offsety * dest_pitch) * pixelsize;
+	dest += offsetx * pixelsize;
 	dest_width = clamp(viewport_x + viewport_width, 0, dest_width - offsetx);
-	dest_height = clamp(viewport_y + viewport_height, 0, dest_height - offsety);
+	dest_height = clamp(viewport_y + viewport_height, 0, dest_height);
 
 	queue->Push<PolySetViewportCommand>(viewport_x, viewport_y, viewport_width, viewport_height, dest, dest_width, dest_height, dest_pitch, dest_bgra);
 }
@@ -127,13 +126,11 @@ void PolyTriangleThreadData::ClearStencil(uint8_t value)
 	int height = buffer->Height();
 	uint8_t *data = buffer->Values();
 
-	int start_y = numa_node * height / num_numa_nodes;
+	int skip = skipped_by_thread(0);
-	int end_y = (numa_node + 1) * height / num_numa_nodes;
+	int count = count_for_thread(0, height);
-	int core_skip = (num_cores - (start_y - core) % num_cores) % num_cores;
-	start_y += core_skip;
 
-	data += start_y * width;
+	data += skip * width;
-	for (int y = start_y; y < end_y; y += num_cores)
+	for (int i = 0; i < count; i++)
 	{
 		memset(data, value, width);
 		data += num_cores * width;
@@ -151,8 +148,6 @@ void PolyTriangleThreadData::SetViewport(int x, int y, int width, int height, ui
 	dest_height = new_dest_height;
 	dest_pitch = new_dest_pitch;
 	dest_bgra = new_dest_bgra;
-	numa_start_y = numa_node * screen->GetHeight() / num_numa_nodes;
-	numa_end_y = (numa_node + 1) * screen->GetHeight() / num_numa_nodes;
 	ccw = true;
 	weaponScene = false;
 }

src/polyrenderer/drawers/poly_triangle.h

@@ -71,7 +71,11 @@ public:
 	int numa_start_y;
 	int numa_end_y;
 
-	// The number of lines to skip to reach the first line to be rendered by this thread
+	bool line_skipped_by_thread(int line)
+	{
+		return line < numa_start_y || line >= numa_end_y || line % num_cores != core;
+	}
+
 	int skipped_by_thread(int first_line)
 	{
 		int clip_first_line = MAX(first_line, numa_start_y);
@@ -79,6 +83,13 @@ public:
 		return clip_first_line + core_skip - first_line;
 	}
 
+	int count_for_thread(int first_line, int count)
+	{
+		count = MIN(count, numa_end_y - first_line);
+		int c = (count - skipped_by_thread(first_line) + num_cores - 1) / num_cores;
+		return MAX(c, 0);
+	}
+
 	// Varyings
 	float worldposX[MAXWIDTH];
 	float worldposY[MAXWIDTH];
@@ -97,6 +108,8 @@ public:
 	uint8_t *dest = nullptr;
 	bool weaponScene = false;
 
+	int viewport_y = 0;
+
 private:
 	ShadedTriVertex ShadeVertex(const PolyDrawArgs &drawargs, const void *vertices, int index);
 	void DrawShadedTriangle(const ShadedTriVertex *vertices, bool ccw, TriDrawTriangleArgs *args);
@@ -105,7 +118,6 @@ private:
 	static int ClipEdge(const ShadedTriVertex *verts, ShadedTriVertex *clippedvert);
 
 	int viewport_x = 0;
-	int viewport_y = 0;
 	int viewport_width = 0;
 	int viewport_height = 0;
 	bool ccw = true;

src/polyrenderer/drawers/screen_triangle.cpp

@@ -60,73 +60,84 @@ void ScreenTriangle::Draw(const TriDrawTriangleArgs *args, PolyTriangleThreadDat
 	ShadedTriVertex *sortedVertices[3];
 	SortVertices(args, sortedVertices);
 
+	int clipleft = 0;
+	int cliptop = MAX(thread->viewport_y, thread->numa_start_y);
 	int clipright = thread->dest_width;
-	int cliptop = thread->numa_start_y;
 	int clipbottom = MIN(thread->dest_height, thread->numa_end_y);
 
 	int topY = (int)(sortedVertices[0]->y + 0.5f);
 	int midY = (int)(sortedVertices[1]->y + 0.5f);
 	int bottomY = (int)(sortedVertices[2]->y + 0.5f);
 
-	topY = MAX(topY, 0);
+	topY = MAX(topY, cliptop);
-	midY = clamp(midY, 0, clipbottom);
+	midY = MIN(midY, clipbottom);
 	bottomY = MIN(bottomY, clipbottom);
 
 	if (topY >= bottomY)
 		return;
 
+	topY += thread->skipped_by_thread(topY);
+	int num_cores = thread->num_cores;
+
 	// Find start/end X positions for each line covered by the triangle:
 
 	int16_t edges[MAXHEIGHT * 2];
 
+	int y = topY;
+
 	float longDX = sortedVertices[2]->x - sortedVertices[0]->x;
 	float longDY = sortedVertices[2]->y - sortedVertices[0]->y;
 	float longStep = longDX / longDY;
-	float longPos = sortedVertices[0]->x + longStep * (topY + 0.5f - sortedVertices[0]->y) + 0.5f;
+	float longPos = sortedVertices[0]->x + longStep * (y + 0.5f - sortedVertices[0]->y) + 0.5f;
+	longStep *= num_cores;
 
-	if (topY < midY)
+	if (y < midY)
 	{
 		float shortDX = sortedVertices[1]->x - sortedVertices[0]->x;
 		float shortDY = sortedVertices[1]->y - sortedVertices[0]->y;
 		float shortStep = shortDX / shortDY;
-		float shortPos = sortedVertices[0]->x + shortStep * (topY + 0.5f - sortedVertices[0]->y) + 0.5f;
+		float shortPos = sortedVertices[0]->x + shortStep * (y + 0.5f - sortedVertices[0]->y) + 0.5f;
+		shortStep *= num_cores;
 
-		for (int y = topY; y < midY; y++)
+		while (y < midY)
 		{
 			int x0 = (int)shortPos;
 			int x1 = (int)longPos;
 			if (x1 < x0) std::swap(x0, x1);
-			x0 = clamp(x0, 0, clipright);
+			x0 = clamp(x0, clipleft, clipright);
-			x1 = clamp(x1, 0, clipright);
+			x1 = clamp(x1, clipleft, clipright);
 
 			edges[y << 1] = x0;
 			edges[(y << 1) + 1] = x1;
 
 			shortPos += shortStep;
 			longPos += longStep;
+			y += num_cores;
 		}
 	}
 
-	if (midY < bottomY)
+	if (y < bottomY)
 	{
 		float shortDX = sortedVertices[2]->x - sortedVertices[1]->x;
 		float shortDY = sortedVertices[2]->y - sortedVertices[1]->y;
 		float shortStep = shortDX / shortDY;
-		float shortPos = sortedVertices[1]->x + shortStep * (midY + 0.5f - sortedVertices[1]->y) + 0.5f;
+		float shortPos = sortedVertices[1]->x + shortStep * (y + 0.5f - sortedVertices[1]->y) + 0.5f;
+		shortStep *= num_cores;
 
-		for (int y = midY; y < bottomY; y++)
+		while (y < bottomY)
 		{
 			int x0 = (int)shortPos;
 			int x1 = (int)longPos;
 			if (x1 < x0) std::swap(x0, x1);
-			x0 = clamp(x0, 0, clipright);
+			x0 = clamp(x0, clipleft, clipright);
-			x1 = clamp(x1, 0, clipright);
+			x1 = clamp(x1, clipleft, clipright);
 
 			edges[y << 1] = x0;
 			edges[(y << 1) + 1] = x1;
 
 			shortPos += shortStep;
 			longPos += longStep;
+			y += num_cores;
 		}
 	}
 
@@ -183,8 +194,14 @@ void DrawTriangle(const TriDrawTriangleArgs *args, PolyTriangleThreadData *threa
 	if (OptT::Flags & SWTRI_WriteStencil)
 		stencilWriteValue = args->uniforms->StencilWriteValue();
 
+	float weaponWOffset;
+	if ((OptT::Flags & SWTRI_DepthTest) || (OptT::Flags & SWTRI_WriteDepth))
+	{
+		weaponWOffset = thread->weaponScene ? 1.0f : 0.0f;
+	}
+
 	int num_cores = thread->num_cores;
-	for (int y = topY + thread->skipped_by_thread(topY); y < bottomY; y += num_cores)
+	for (int y = topY; y < bottomY; y += num_cores)
 	{
 		int x = edges[y << 1];
 		int xend = edges[(y << 1) + 1];
@@ -198,7 +215,7 @@ void DrawTriangle(const TriDrawTriangleArgs *args, PolyTriangleThreadData *threa
 
 			float startX = x + (0.5f - v1X);
 			float startY = y + (0.5f - v1Y);
-			posXW = v1W + stepXW * startX + args->gradientY.W * startY + (thread->weaponScene ? 1.0f : 0.0f);
+			posXW = v1W + stepXW * startX + args->gradientY.W * startY + weaponWOffset;
 		}
 
 #ifndef NO_SSE

src/swrenderer/drawers/r_thread.cpp

@@ -142,6 +142,11 @@ void DrawerThreads::WorkerMain(DrawerThread *thread)
 		thread->current_queue++;
 		thread->numa_start_y = thread->numa_node * screen->GetHeight() / thread->num_numa_nodes;
 		thread->numa_end_y = (thread->numa_node + 1) * screen->GetHeight() / thread->num_numa_nodes;
+		if (thread->poly)
+		{
+			thread->poly->numa_start_y = thread->numa_start_y;
+			thread->poly->numa_end_y = thread->numa_end_y;
+		}
 		start_lock.unlock();
 
 		// Do the work: