Measure the available time for "background depth raster". The idea is viable.

GPU/Common/DrawEngineCommon.cpp

@@ -1042,6 +1042,10 @@ void DrawEngineCommon::DepthRasterTransform(GEPrimitiveType prim, VertexDecoder
 	depthIndexCount_ += vertexCount;
 	depthVertexCount_ += numDecoded;
 
+	if (depthDraws_.empty()) {
+		rasterTimeStart_ = time_now_d();
+	}
+
 	depthDraws_.push_back(draw);
 
 	// FlushQueuedDepth();
@@ -1081,11 +1085,19 @@ void DrawEngineCommon::DepthRasterPredecoded(GEPrimitiveType prim, const void *i
 
 	depthDraws_.push_back(draw);
 
+	if (depthDraws_.empty()) {
+		rasterTimeStart_ = time_now_d();
+	}
 	// FlushQueuedDepth();
 }
 
 void DrawEngineCommon::FlushQueuedDepth() {
-	TimeCollector collectStat(&gpuStats.msRasterizeDepth, coreCollectDebugStats);
+	if (rasterTimeStart_ != 0.0) {
+		gpuStats.msRasterTimeAvailable += time_now_d() - rasterTimeStart_;
+		rasterTimeStart_ = 0.0;
+	}
+
+	const bool collectStats = coreCollectDebugStats;
 
 	for (const auto &draw : depthDraws_) {
 		int *tx = depthScreenVerts_;
@@ -1097,19 +1109,24 @@ void DrawEngineCommon::FlushQueuedDepth() {
 		const float *vertices = depthTransformed_ + 4 * draw.vertexOffset;
 		const uint16_t *indices = depthIndices_ + draw.indexOffset;
 
-		switch (draw.prim) {
-		case GE_PRIM_RECTANGLES:
-			outVertCount = DepthRasterClipIndexedRectangles(tx, ty, tz, vertices, indices, draw);
-			break;
-		case GE_PRIM_TRIANGLES:
-			outVertCount = DepthRasterClipIndexedTriangles(tx, ty, tz, vertices, indices, draw);
-			break;
-		default:
-			_dbg_assert_(false);
-			break;
+		{
+			TimeCollector collectStat(&gpuStats.msCullDepth, collectStats);
+			switch (draw.prim) {
+			case GE_PRIM_RECTANGLES:
+				outVertCount = DepthRasterClipIndexedRectangles(tx, ty, tz, vertices, indices, draw);
+				break;
+			case GE_PRIM_TRIANGLES:
+				outVertCount = DepthRasterClipIndexedTriangles(tx, ty, tz, vertices, indices, draw);
+				break;
+			default:
+				_dbg_assert_(false);
+				break;
+			}
+		}
+		{
+			TimeCollector collectStat(&gpuStats.msRasterizeDepth, collectStats);
+			DepthRasterScreenVerts((uint16_t *)Memory::GetPointerWrite(draw.depthAddr), draw.depthStride, tx, ty, tz, outVertCount, draw);
 		}
-		// TODO: Could potentially split into tasks here!
-		DepthRasterScreenVerts((uint16_t *)Memory::GetPointerWrite(draw.depthAddr), draw.depthStride, tx, ty, tz, outVertCount, draw);
 	}
 
 	// Reset queue

GPU/Common/DrawEngineCommon.h

@@ -367,4 +367,6 @@ protected:
 	int depthVertexCount_ = 0;
 	int depthIndexCount_ = 0;
 	std::vector<DepthDraw> depthDraws_;
+
+	double rasterTimeStart_ = 0.0;
 };

GPU/GPU.h

@@ -108,8 +108,10 @@ struct GPUStatistics {
 		numCachedReplacedTextures = 0;
 		numClutTextures = 0;
 		msProcessingDisplayLists = 0;
-		msPrepareDepth = 0.0f;
-		msRasterizeDepth = 0.0f;
+		msPrepareDepth = 0.0;
+		msCullDepth = 0.0;
+		msRasterizeDepth = 0.0;
+		msRasterTimeAvailable = 0.0;
 		numDepthRasterPrims = 0;
 		numDepthRasterEarlySize = 0;
 		numDepthRasterNoPixels = 0;
@@ -156,7 +158,9 @@ struct GPUStatistics {
 	int numClutTextures;
 	double msProcessingDisplayLists;
 	double msPrepareDepth;
+	double msCullDepth;
 	double msRasterizeDepth;
+	double msRasterTimeAvailable;
 	int vertexGPUCycles;
 	int otherGPUCycles;
 	int numDepthRasterPrims;

GPU/GPUCommonHW.cpp

@@ -1763,7 +1763,7 @@ void GPUCommonHW::Execute_BoneMtxData(u32 op, u32 diff) {
 			gstate_c.Dirty(DIRTY_BONEMATRIX0 << (num / 12));
 		} else {
 			gstate_c.deferredVertTypeDirty |= DIRTY_BONEMATRIX0 << (num / 12);
-		}
+		} 
 		((u32 *)gstate.boneMatrix)[num] = newVal;
 	}
 	num++;
@@ -1801,7 +1801,8 @@ size_t GPUCommonHW::FormatGPUStatsCommon(char *buffer, size_t size) {
 		"replacer: tracks %d references, %d unique textures\n"
 		"Cpy: depth %d, color %d, reint %d, blend %d, self %d\n"
 		"GPU cycles: %d (%0.1f per vertex)\n"
-		"Z-rast: %0.2f/%0.2f ms, %d prim, %d nopix, %d small, %d earlysize, %d zcull, %d box\n%s",
+		"Z-rast: %0.2f+%0.2f+%0.2f (total %0.2f/%0.2f) ms\n"
+		"Z-rast: %d prim, %d nopix, %d small, %d earlysize, %d zcull, %d box\n%s",
 		gpuStats.msProcessingDisplayLists * 1000.0f,
 		gpuStats.numDrawSyncs,
 		gpuStats.numListSyncs,
@@ -1839,7 +1840,10 @@ size_t GPUCommonHW::FormatGPUStatsCommon(char *buffer, size_t size) {
 		gpuStats.vertexGPUCycles + gpuStats.otherGPUCycles,
 		vertexAverageCycles,
 		gpuStats.msPrepareDepth * 1000.0,
+		gpuStats.msCullDepth * 1000.0,
 		gpuStats.msRasterizeDepth * 1000.0,
+		(gpuStats.msPrepareDepth + gpuStats.msCullDepth + gpuStats.msRasterizeDepth) * 1000.0,
+		gpuStats.msRasterTimeAvailable * 1000.0,
 		gpuStats.numDepthRasterPrims,
 		gpuStats.numDepthRasterNoPixels,
 		gpuStats.numDepthRasterTooSmall,