Merge pull request #17810 from hrydgard/bbox-cache-planes

Cache computed planes used for BBOX culling

Core/HLE/sceDisplay.cpp

@@ -594,7 +594,7 @@ void __DisplayFlip(int cyclesLate) {
 #ifndef _DEBUG
 			auto err = GetI18NCategory(I18NCat::ERRORS);
 			if (g_Config.bSoftwareRendering) {
-				g_OSD.Show(OSDType::MESSAGE_INFO, err->T("Running slow: Try turning off Software Rendering"));
+				g_OSD.Show(OSDType::MESSAGE_INFO, err->T("Running slow: Try turning off Software Rendering"), 5.0f);
 			} else {
 				g_OSD.Show(OSDType::MESSAGE_INFO, err->T("Running slow: try frameskip, sound is choppy when slow"));
 			}

GPU/Common/DrawEngineCommon.cpp

@@ -19,9 +19,9 @@
 #include <cfloat>
 
 #include "Common/Data/Convert/ColorConv.h"
-#include "Common/Math/lin/matrix4x4.h"
 #include "Common/Profiler/Profiler.h"
 #include "Common/LogReporting.h"
+#include "Common/Math/lin/matrix4x4.h"
 #include "Core/Config.h"
 #include "GPU/Common/DrawEngineCommon.h"
 #include "GPU/Common/SplineCommon.h"
@@ -136,21 +136,6 @@ std::string DrawEngineCommon::DebugGetVertexLoaderString(std::string id, DebugSh
 	return dec ? dec->GetString(stringType) : "N/A";
 }
 
-struct Plane {
-	float x, y, z, w;
-	void Set(float _x, float _y, float _z, float _w) { x = _x; y = _y; z = _z; w = _w; }
-	float Test(const float f[3]) const { return x * f[0] + y * f[1] + z * f[2] + w; }
-};
-
-static void PlanesFromMatrix(const float mtx[16], Plane planes[6]) {
-	planes[0].Set(mtx[3]-mtx[0], mtx[7]-mtx[4], mtx[11]-mtx[8], mtx[15]-mtx[12]);  // Right
-	planes[1].Set(mtx[3]+mtx[0], mtx[7]+mtx[4], mtx[11]+mtx[8], mtx[15]+mtx[12]);  // Left
-	planes[2].Set(mtx[3]+mtx[1], mtx[7]+mtx[5], mtx[11]+mtx[9], mtx[15]+mtx[13]);  // Bottom
-	planes[3].Set(mtx[3]-mtx[1], mtx[7]-mtx[5], mtx[11]-mtx[9], mtx[15]-mtx[13]);  // Top
-	planes[4].Set(mtx[3]+mtx[2], mtx[7]+mtx[6], mtx[11]+mtx[10], mtx[15]+mtx[14]); // Near
-	planes[5].Set(mtx[3]-mtx[2], mtx[7]-mtx[6], mtx[11]-mtx[10], mtx[15]-mtx[14]); // Far
-}
-
 static Vec3f ClipToScreen(const Vec4f& coords) {
 	float xScale = gstate.getViewportXScale();
 	float xCenter = gstate.getViewportXCenter();
@@ -250,6 +235,52 @@ void DrawEngineCommon::DispatchSubmitImm(GEPrimitiveType prim, TransformedVertex
 	}
 }
 
+// Gated by DIRTY_CULL_PLANES
+void DrawEngineCommon::UpdatePlanes() {
+	float world[16];
+	float view[16];
+	float worldview[16];
+	float worldviewproj[16];
+	ConvertMatrix4x3To4x4(world, gstate.worldMatrix);
+	ConvertMatrix4x3To4x4(view, gstate.viewMatrix);
+	// TODO: Create a Matrix4x3ByMatrix4x3, and Matrix4x4ByMatrix4x3?
+	Matrix4ByMatrix4(worldview, world, view);
+	Matrix4ByMatrix4(worldviewproj, worldview, gstate.projMatrix);
+
+	// Next, we need to apply viewport, scissor, region, and even offset - but only for X/Y.
+	// Note that the PSP does not clip against the viewport.
+	const Vec2f baseOffset = Vec2f(gstate.getOffsetX(), gstate.getOffsetY());
+	// Region1 (rate) is used as an X1/Y1 here, matching PSP behavior.
+	minOffset_ = baseOffset + Vec2f(std::max(gstate.getRegionRateX() - 0x100, gstate.getScissorX1()), std::max(gstate.getRegionRateY() - 0x100, gstate.getScissorY1())) - Vec2f(1.0f, 1.0f);
+	maxOffset_ = baseOffset + Vec2f(std::min(gstate.getRegionX2(), gstate.getScissorX2()), std::min(gstate.getRegionY2(), gstate.getScissorY2())) + Vec2f(1.0f, 1.0f);
+
+	// Now let's apply the viewport to our scissor/region + offset range.
+	Vec2f inverseViewportScale = Vec2f(1.0f / gstate.getViewportXScale(), 1.0f / gstate.getViewportYScale());
+	Vec2f minViewport = (minOffset_ - Vec2f(gstate.getViewportXCenter(), gstate.getViewportYCenter())) * inverseViewportScale;
+	Vec2f maxViewport = (maxOffset_ - Vec2f(gstate.getViewportXCenter(), gstate.getViewportYCenter())) * inverseViewportScale;
+
+	Lin::Matrix4x4 applyViewport;
+	applyViewport.empty();
+	// Scale to the viewport's size.
+	applyViewport.xx = 2.0f / (maxViewport.x - minViewport.x);
+	applyViewport.yy = 2.0f / (maxViewport.y - minViewport.y);
+	applyViewport.zz = 1.0f;
+	applyViewport.ww = 1.0f;
+	// And offset to the viewport's centers.
+	applyViewport.wx = -(maxViewport.x + minViewport.x) / (maxViewport.x - minViewport.x);
+	applyViewport.wy = -(maxViewport.y + minViewport.y) / (maxViewport.y - minViewport.y);
+
+	float mtx[16];
+	Matrix4ByMatrix4(mtx, worldviewproj, applyViewport.m);
+
+	planes_[0].Set(mtx[3] - mtx[0], mtx[7] - mtx[4], mtx[11] - mtx[8], mtx[15] - mtx[12]);  // Right
+	planes_[1].Set(mtx[3] + mtx[0], mtx[7] + mtx[4], mtx[11] + mtx[8], mtx[15] + mtx[12]);  // Left
+	planes_[2].Set(mtx[3] + mtx[1], mtx[7] + mtx[5], mtx[11] + mtx[9], mtx[15] + mtx[13]);  // Bottom
+	planes_[3].Set(mtx[3] - mtx[1], mtx[7] - mtx[5], mtx[11] - mtx[9], mtx[15] - mtx[13]);  // Top
+	planes_[4].Set(mtx[3] + mtx[2], mtx[7] + mtx[6], mtx[11] + mtx[10], mtx[15] + mtx[14]); // Near
+	planes_[5].Set(mtx[3] - mtx[2], mtx[7] - mtx[6], mtx[11] - mtx[10], mtx[15] - mtx[14]); // Far
+}
+
 // This code has plenty of potential for optimization.
 //
 // It does the simplest and safest test possible: If all points of a bbox is outside a single of
@@ -273,7 +304,7 @@ bool DrawEngineCommon::TestBoundingBox(const void *control_points, const void *i
 			verts[i] = vtx[i] * (1.0f / 128.0f);
 		}
 	} else if ((vertType & 0xFFFFFF) == GE_VTYPE_POS_16BIT && !inds) {
-		const s16 *vtx = (const s16*)control_points;
+		const s16 *vtx = (const s16 *)control_points;
 		for (int i = 0; i < vertexCount * 3; i++) {
 			verts[i] = vtx[i] * (1.0f / 32768.0f);
 		}
@@ -302,70 +333,42 @@ bool DrawEngineCommon::TestBoundingBox(const void *control_points, const void *i
 		}
 	}
 
-	Plane planes[6];
+	// Due to world matrix updates per "thing", this isn't quite as effective as it could be if we did world transform
+	// in here as well. Though, it still does cut down on a lot of updates in Tekken 6.
+	if (gstate_c.IsDirty(DIRTY_CULL_PLANES)) {
+		UpdatePlanes();
+		gpuStats.numPlaneUpdates++;
+		gstate_c.Clean(DIRTY_CULL_PLANES);
+	}
 
-	float world[16];
-	float view[16];
-	float worldview[16];
-	float worldviewproj[16];
-	ConvertMatrix4x3To4x4(world, gstate.worldMatrix);
-	ConvertMatrix4x3To4x4(view, gstate.viewMatrix);
-	// TODO: Create a Matrix4x3ByMatrix4x3, and Matrix4x4ByMatrix4x3?
-	Matrix4ByMatrix4(worldview, world, view);
-	Matrix4ByMatrix4(worldviewproj, worldview, gstate.projMatrix);
-
-	// Next, we need to apply viewport, scissor, region, and even offset - but only for X/Y.
-	// Note that the PSP does not clip against the viewport.
-	const Vec2f baseOffset = Vec2f(gstate.getOffsetX(), gstate.getOffsetY());
-	// Region1 (rate) is used as an X1/Y1 here, matching PSP behavior.
-	Vec2f minOffset = baseOffset + Vec2f(std::max(gstate.getRegionRateX() - 0x100, gstate.getScissorX1()), std::max(gstate.getRegionRateY() - 0x100, gstate.getScissorY1())) - Vec2f(1.0f, 1.0f);
-	Vec2f maxOffset = baseOffset + Vec2f(std::min(gstate.getRegionX2(), gstate.getScissorX2()), std::min(gstate.getRegionY2(), gstate.getScissorY2())) + Vec2f(1.0f, 1.0f);
-
-	// Now let's apply the viewport to our scissor/region + offset range.
-	Vec2f inverseViewportScale = Vec2f(1.0f / gstate.getViewportXScale(), 1.0f / gstate.getViewportYScale());
-	Vec2f minViewport = (minOffset - Vec2f(gstate.getViewportXCenter(), gstate.getViewportYCenter())) * inverseViewportScale;
-	Vec2f maxViewport = (maxOffset - Vec2f(gstate.getViewportXCenter(), gstate.getViewportYCenter())) * inverseViewportScale;
-
-	Lin::Matrix4x4 applyViewport;
-	applyViewport.empty();
-	// Scale to the viewport's size.
-	applyViewport.xx = 2.0f / (maxViewport.x - minViewport.x);
-	applyViewport.yy = 2.0f / (maxViewport.y - minViewport.y);
-	applyViewport.zz = 1.0f;
-	applyViewport.ww = 1.0f;
-	// And offset to the viewport's centers.
-	applyViewport.wx = -(maxViewport.x + minViewport.x) / (maxViewport.x - minViewport.x);
-	applyViewport.wy = -(maxViewport.y + minViewport.y) / (maxViewport.y - minViewport.y);
-
-	float screenBounds[16];
-	Matrix4ByMatrix4(screenBounds, worldviewproj, applyViewport.m);
-
-	PlanesFromMatrix(screenBounds, planes);
 	// Note: near/far are not checked without clamp/clip enabled, so we skip those planes.
 	int totalPlanes = gstate.isDepthClampEnabled() ? 6 : 4;
 	for (int plane = 0; plane < totalPlanes; plane++) {
 		int inside = 0;
 		int out = 0;
 		for (int i = 0; i < vertexCount; i++) {
-			// Here we can test against the frustum planes!
-			float value = planes[plane].Test(verts + i * 3);
+			// Test against the frustum planes, and count.
+			// TODO: We should test 4 vertices at a time using SIMD.
+			// I guess could also test one vertex against 4 planes at a time, though a lot of waste at the common case of 6.
+			float value = planes_[plane].Test(verts + i * 3);
 			if (value <= -FLT_EPSILON)
 				out++;
 			else
 				inside++;
 		}
 
+		// No vertices inside this one plane? Don't need to draw.
 		if (inside == 0) {
 			// All out - but check for X and Y if the offset was near the cullbox edge.
 			bool outsideEdge = false;
 			if (plane == 1)
-				outsideEdge = minOffset.x < 1.0f;
+				outsideEdge = minOffset_.x < 1.0f;
 			if (plane == 2)
-				outsideEdge = minOffset.y < 1.0f;
+				outsideEdge = minOffset_.y < 1.0f;
 			else if (plane == 0)
-				outsideEdge = maxOffset.x >= 4096.0f;
+				outsideEdge = maxOffset_.x >= 4096.0f;
 			else if (plane == 3)
-				outsideEdge = maxOffset.y >= 4096.0f;
+				outsideEdge = maxOffset_.y >= 4096.0f;
 
 			// Only consider this outside if offset + scissor/region is fully inside the cullbox.
 			if (!outsideEdge)

GPU/Common/DrawEngineCommon.h

@@ -22,6 +22,7 @@
 #include "Common/CommonTypes.h"
 #include "Common/Data/Collections/Hashmaps.h"
 
+#include "GPU/Math3D.h"
 #include "GPU/GPUState.h"
 #include "GPU/Common/GPUStateUtils.h"
 #include "GPU/Common/GPUDebugInterface.h"
@@ -68,6 +69,13 @@ public:
 	virtual void SendDataToShader(const SimpleVertex *const *points, int size_u, int size_v, u32 vertType, const Spline::Weight2D &weights) = 0;
 };
 
+// Culling plane.
+struct Plane {
+	float x, y, z, w;
+	void Set(float _x, float _y, float _z, float _w) { x = _x; y = _y; z = _z; w = _w; }
+	float Test(const float f[3]) const { return x * f[0] + y * f[1] + z * f[2] + w; }
+};
+
 class DrawEngineCommon {
 public:
 	DrawEngineCommon();
@@ -131,6 +139,7 @@ public:
 
 protected:
 	virtual bool UpdateUseHWTessellation(bool enabled) const { return enabled; }
+	void UpdatePlanes();
 
 	int ComputeNumVertsToDecode() const;
 	void DecodeVerts(u8 *dest);
@@ -236,4 +245,9 @@ protected:
 
 	// Hardware tessellation
 	TessellationDataTransfer *tessDataTransfer;
+
+	// Culling
+	Plane planes_[6];
+	Vec2f minOffset_;
+	Vec2f maxOffset_;
 };

GPU/Common/ShaderCommon.h

@@ -93,7 +93,7 @@ enum : uint64_t {
 	DIRTY_LIGHT_CONTROL = 1ULL << 38,
 	DIRTY_TEX_ALPHA_MUL = 1ULL << 39,
 
-	// Bits 40-43 are free for new uniforms. Then we're really out and need to start merging.
+	// Bits 40-42 are free for new uniforms. Then we're really out and need to start merging.
 	// Don't forget to update DIRTY_ALL_UNIFORMS when you start using them.
 
 	DIRTY_BONE_UNIFORMS = 0xFF000000ULL,
@@ -101,6 +101,7 @@ enum : uint64_t {
 	DIRTY_ALL_UNIFORMS = 0x0FFFFFFFFFFULL,
 
 	// Other dirty elements that aren't uniforms
+	DIRTY_CULL_PLANES = 1ULL << 43,
 	DIRTY_FRAMEBUF = 1ULL << 44,
 	DIRTY_TEXTURE_IMAGE = 1ULL << 45,  // Means that the definition of the texture image has changed (address, stride etc), and we need to look up again.
 	DIRTY_TEXTURE_PARAMS = 1ULL << 46,

GPU/GPU.h

@@ -66,6 +66,7 @@ inline unsigned int toFloat24(float f) {
 	return i >> 8;
 }
 
+// The ToString function lives in GPUCommonHW.cpp.
 struct GPUStatistics {
 	void Reset() {
 		ResetFrame();
@@ -84,10 +85,10 @@ struct GPUStatistics {
 		numTextureInvalidations = 0;
 		numTextureInvalidationsByFramebuffer = 0;
 		numTexturesHashed = 0;
-		numTextureSwitches = 0;
 		numTextureDataBytesHashed = 0;
-		numShaderSwitches = 0;
 		numFlushes = 0;
+		numBBOXJumps = 0;
+		numPlaneUpdates = 0;
 		numTexturesDecoded = 0;
 		numFramebufferEvaluations = 0;
 		numBlockingReadbacks = 0;
@@ -114,6 +115,8 @@ struct GPUStatistics {
 	int numListSyncs;
 	int numCachedDrawCalls;
 	int numFlushes;
+	int numBBOXJumps;
+	int numPlaneUpdates;
 	int numVertsSubmitted;
 	int numCachedVertsDrawn;
 	int numUncachedVertsDrawn;
@@ -122,8 +125,6 @@ struct GPUStatistics {
 	int numTextureInvalidationsByFramebuffer;
 	int numTexturesHashed;
 	int numTextureDataBytesHashed;
-	int numTextureSwitches;
-	int numShaderSwitches;
 	int numTexturesDecoded;
 	int numFramebufferEvaluations;
 	int numBlockingReadbacks;

GPU/GPUCommon.cpp

@@ -931,6 +931,7 @@ void GPUCommon::Execute_BJump(u32 op, u32 diff) {
 	if (!currentList->bboxResult) {
 		// bounding box jump.
 		const u32 target = gstate_c.getRelativeAddress(op & 0x00FFFFFC);
+		gpuStats.numBBOXJumps++;
 		if (Memory::IsValidAddress(target)) {
 			UpdatePC(currentList->pc, target - 4);
 			currentList->pc = target - 4; // pc will be increased after we return, counteract that

GPU/GPUCommonHW.cpp

@@ -191,12 +191,12 @@ const CommonCommandTableEntry commonCommandTable[] = {
 	{ GE_CMD_ANTIALIASENABLE, FLAG_FLUSHBEFOREONCHANGE },
 
 	// Viewport.
-	{ GE_CMD_OFFSETX, FLAG_FLUSHBEFOREONCHANGE, DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_CULLRANGE },
-	{ GE_CMD_OFFSETY, FLAG_FLUSHBEFOREONCHANGE, DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_CULLRANGE },
-	{ GE_CMD_VIEWPORTXSCALE, FLAG_FLUSHBEFOREONCHANGE,  DIRTY_FRAMEBUF | DIRTY_TEXTURE_PARAMS | DIRTY_CULLRANGE | DIRTY_PROJMATRIX | DIRTY_VIEWPORTSCISSOR_STATE },
-	{ GE_CMD_VIEWPORTYSCALE, FLAG_FLUSHBEFOREONCHANGE,  DIRTY_FRAMEBUF | DIRTY_TEXTURE_PARAMS | DIRTY_CULLRANGE | DIRTY_PROJMATRIX | DIRTY_VIEWPORTSCISSOR_STATE },
-	{ GE_CMD_VIEWPORTXCENTER, FLAG_FLUSHBEFOREONCHANGE, DIRTY_FRAMEBUF | DIRTY_TEXTURE_PARAMS | DIRTY_CULLRANGE | DIRTY_PROJMATRIX | DIRTY_VIEWPORTSCISSOR_STATE },
-	{ GE_CMD_VIEWPORTYCENTER, FLAG_FLUSHBEFOREONCHANGE, DIRTY_FRAMEBUF | DIRTY_TEXTURE_PARAMS | DIRTY_CULLRANGE | DIRTY_PROJMATRIX | DIRTY_VIEWPORTSCISSOR_STATE },
+	{ GE_CMD_OFFSETX, FLAG_FLUSHBEFOREONCHANGE, DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_CULLRANGE | DIRTY_CULL_PLANES },
+	{ GE_CMD_OFFSETY, FLAG_FLUSHBEFOREONCHANGE, DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_CULLRANGE | DIRTY_CULL_PLANES },
+	{ GE_CMD_VIEWPORTXSCALE, FLAG_FLUSHBEFOREONCHANGE,  DIRTY_FRAMEBUF | DIRTY_TEXTURE_PARAMS | DIRTY_CULLRANGE | DIRTY_PROJMATRIX | DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_CULL_PLANES },
+	{ GE_CMD_VIEWPORTYSCALE, FLAG_FLUSHBEFOREONCHANGE,  DIRTY_FRAMEBUF | DIRTY_TEXTURE_PARAMS | DIRTY_CULLRANGE | DIRTY_PROJMATRIX | DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_CULL_PLANES },
+	{ GE_CMD_VIEWPORTXCENTER, FLAG_FLUSHBEFOREONCHANGE, DIRTY_FRAMEBUF | DIRTY_TEXTURE_PARAMS | DIRTY_CULLRANGE | DIRTY_PROJMATRIX | DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_CULL_PLANES },
+	{ GE_CMD_VIEWPORTYCENTER, FLAG_FLUSHBEFOREONCHANGE, DIRTY_FRAMEBUF | DIRTY_TEXTURE_PARAMS | DIRTY_CULLRANGE | DIRTY_PROJMATRIX | DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_CULL_PLANES },
 	{ GE_CMD_VIEWPORTZSCALE, FLAG_FLUSHBEFOREONCHANGE,  DIRTY_FRAMEBUF | DIRTY_TEXTURE_PARAMS | DIRTY_CULLRANGE | DIRTY_DEPTHRANGE | DIRTY_PROJMATRIX | DIRTY_VIEWPORTSCISSOR_STATE },
 	{ GE_CMD_VIEWPORTZCENTER, FLAG_FLUSHBEFOREONCHANGE, DIRTY_FRAMEBUF | DIRTY_TEXTURE_PARAMS | DIRTY_CULLRANGE | DIRTY_DEPTHRANGE | DIRTY_PROJMATRIX | DIRTY_VIEWPORTSCISSOR_STATE },
 	{ GE_CMD_DEPTHCLAMPENABLE, FLAG_FLUSHBEFOREONCHANGE, DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_CULLRANGE | DIRTY_RASTER_STATE },
@@ -206,12 +206,12 @@ const CommonCommandTableEntry commonCommandTable[] = {
 	{ GE_CMD_MAXZ, FLAG_FLUSHBEFOREONCHANGE, DIRTY_DEPTHRANGE | DIRTY_RASTER_STATE | DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_CULLRANGE },
 
 	// Region
-	{ GE_CMD_REGION1, FLAG_FLUSHBEFOREONCHANGE, DIRTY_FRAMEBUF | DIRTY_TEXTURE_PARAMS | DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_CULLRANGE },
-	{ GE_CMD_REGION2, FLAG_FLUSHBEFOREONCHANGE, DIRTY_FRAMEBUF | DIRTY_TEXTURE_PARAMS | DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_CULLRANGE },
+	{ GE_CMD_REGION1, FLAG_FLUSHBEFOREONCHANGE, DIRTY_FRAMEBUF | DIRTY_TEXTURE_PARAMS | DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_CULLRANGE | DIRTY_CULL_PLANES },
+	{ GE_CMD_REGION2, FLAG_FLUSHBEFOREONCHANGE, DIRTY_FRAMEBUF | DIRTY_TEXTURE_PARAMS | DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_CULLRANGE | DIRTY_CULL_PLANES },
 
 	// Scissor
-	{ GE_CMD_SCISSOR1, FLAG_FLUSHBEFOREONCHANGE, DIRTY_FRAMEBUF | DIRTY_TEXTURE_PARAMS | DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_CULLRANGE },
-	{ GE_CMD_SCISSOR2, FLAG_FLUSHBEFOREONCHANGE, DIRTY_FRAMEBUF | DIRTY_TEXTURE_PARAMS | DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_CULLRANGE },
+	{ GE_CMD_SCISSOR1, FLAG_FLUSHBEFOREONCHANGE, DIRTY_FRAMEBUF | DIRTY_TEXTURE_PARAMS | DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_CULLRANGE | DIRTY_CULL_PLANES },
+	{ GE_CMD_SCISSOR2, FLAG_FLUSHBEFOREONCHANGE, DIRTY_FRAMEBUF | DIRTY_TEXTURE_PARAMS | DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_CULLRANGE | DIRTY_CULL_PLANES },
 
 	// Lighting base colors
 	{ GE_CMD_AMBIENTCOLOR, FLAG_FLUSHBEFOREONCHANGE, DIRTY_AMBIENT },
@@ -1368,7 +1368,7 @@ void GPUCommonHW::Execute_WorldMtxNum(u32 op, u32 diff) {
 			if (dst[i] != newVal) {
 				Flush();
 				dst[i] = newVal;
-				gstate_c.Dirty(DIRTY_WORLDMATRIX);
+				gstate_c.Dirty(DIRTY_WORLDMATRIX | DIRTY_CULL_PLANES);
 			}
 			if (++i >= end) {
 				break;
@@ -1391,7 +1391,7 @@ void GPUCommonHW::Execute_WorldMtxData(u32 op, u32 diff) {
 	if (num < 12 && newVal != ((const u32 *)gstate.worldMatrix)[num]) {
 		Flush();
 		((u32 *)gstate.worldMatrix)[num] = newVal;
-		gstate_c.Dirty(DIRTY_WORLDMATRIX);
+		gstate_c.Dirty(DIRTY_WORLDMATRIX | DIRTY_CULL_PLANES);
 	}
 	num++;
 	gstate.worldmtxnum = (GE_CMD_WORLDMATRIXNUMBER << 24) | (num & 0x00FFFFFF);
@@ -1421,7 +1421,7 @@ void GPUCommonHW::Execute_ViewMtxNum(u32 op, u32 diff) {
 			if (dst[i] != newVal) {
 				Flush();
 				dst[i] = newVal;
-				gstate_c.Dirty(DIRTY_VIEWMATRIX);
+				gstate_c.Dirty(DIRTY_VIEWMATRIX | DIRTY_CULL_PLANES);
 			}
 			if (++i >= end) {
 				break;
@@ -1444,7 +1444,7 @@ void GPUCommonHW::Execute_ViewMtxData(u32 op, u32 diff) {
 	if (num < 12 && newVal != ((const u32 *)gstate.viewMatrix)[num]) {
 		Flush();
 		((u32 *)gstate.viewMatrix)[num] = newVal;
-		gstate_c.Dirty(DIRTY_VIEWMATRIX);
+		gstate_c.Dirty(DIRTY_VIEWMATRIX | DIRTY_CULL_PLANES);
 	}
 	num++;
 	gstate.viewmtxnum = (GE_CMD_VIEWMATRIXNUMBER << 24) | (num & 0x00FFFFFF);
@@ -1474,7 +1474,7 @@ void GPUCommonHW::Execute_ProjMtxNum(u32 op, u32 diff) {
 			if (dst[i] != newVal) {
 				Flush();
 				dst[i] = newVal;
-				gstate_c.Dirty(DIRTY_PROJMATRIX);
+				gstate_c.Dirty(DIRTY_PROJMATRIX | DIRTY_CULL_PLANES);
 			}
 			if (++i >= end) {
 				break;
@@ -1497,7 +1497,7 @@ void GPUCommonHW::Execute_ProjMtxData(u32 op, u32 diff) {
 	if (num < 16 && newVal != ((const u32 *)gstate.projMatrix)[num]) {
 		Flush();
 		((u32 *)gstate.projMatrix)[num] = newVal;
-		gstate_c.Dirty(DIRTY_PROJMATRIX);
+		gstate_c.Dirty(DIRTY_PROJMATRIX | DIRTY_CULL_PLANES);
 	}
 	num++;
 	if (num <= 16)
@@ -1647,8 +1647,8 @@ size_t GPUCommonHW::FormatGPUStatsCommon(char *buffer, size_t size) {
 	float vertexAverageCycles = gpuStats.numVertsSubmitted > 0 ? (float)gpuStats.vertexGPUCycles / (float)gpuStats.numVertsSubmitted : 0.0f;
 	return snprintf(buffer, size,
 		"DL processing time: %0.2f ms, %d drawsync, %d listsync\n"
-		"Draw calls: %d, flushes %d, clears %d (cached: %d)\n"
-		"Num Tracked Vertex Arrays: %d\n"
+		"Draw calls: %d, flushes %d, clears %d, bbox jumps %d (%d updates)\n"
+		"Cached draws: %d (tracked: %d)\n"
 		"Vertices: %d cached: %d uncached: %d\n"
 		"FBOs active: %d (evaluations: %d)\n"
 		"Textures: %d, dec: %d, invalidated: %d, hashed: %d kB\n"
@@ -1662,6 +1662,8 @@ size_t GPUCommonHW::FormatGPUStatsCommon(char *buffer, size_t size) {
 		gpuStats.numDrawCalls,
 		gpuStats.numFlushes,
 		gpuStats.numClears,
+		gpuStats.numBBOXJumps,
+		gpuStats.numPlaneUpdates,
 		gpuStats.numCachedDrawCalls,
 		gpuStats.numTrackedVertexArrays,
 		gpuStats.numVertsSubmitted,

GPU/GPUState.cpp

@@ -128,6 +128,8 @@ void GPUgstate::Reset() {
 	memset(gstate.boneMatrix, 0, sizeof(gstate.boneMatrix));
 
 	savedContextVersion = 1;
+
+	gstate_c.Dirty(DIRTY_CULL_PLANES);
 }
 
 void GPUgstate::Save(u32_le *ptr) {
@@ -258,6 +260,8 @@ void GPUgstate::Restore(const u32_le *ptr) {
 
 	if (gpu)
 		gpu->ResetMatrices();
+
+	gstate_c.Dirty(DIRTY_CULL_PLANES);
 }
 
 bool vertTypeIsSkinningEnabled(u32 vertType) {
@@ -366,6 +370,9 @@ void GPUStateCache::DoState(PointerWrap &p) {
 	} else {
 		Do(p, savedContextVersion);
 	}
+
+	if (p.GetMode() == PointerWrap::MODE_READ)
+		gstate_c.Dirty(DIRTY_CULL_PLANES);
 }
 
 static const char *const gpuUseFlagNames[32] = {

GPU/Software/SoftGpu.cpp

@@ -81,7 +81,7 @@ const SoftwareCommandTableEntry softgpuCommandTable[] = {
 	{ GE_CMD_VADDR, FLAG_EXECUTE, SoftDirty::NONE, &GPUCommon::Execute_Vaddr },
 	{ GE_CMD_IADDR, FLAG_EXECUTE, SoftDirty::NONE, &GPUCommon::Execute_Iaddr },
 	{ GE_CMD_BJUMP, FLAG_EXECUTE | FLAG_READS_PC | FLAG_WRITES_PC, SoftDirty::NONE, &GPUCommon::Execute_BJump },
-	{ GE_CMD_BOUNDINGBOX, FLAG_EXECUTE, SoftDirty::NONE, &GPUCommon::Execute_BoundingBox },
+	{ GE_CMD_BOUNDINGBOX, FLAG_EXECUTE, SoftDirty::NONE, &SoftGPU::Execute_BoundingBox },
 
 	{ GE_CMD_PRIM, FLAG_EXECUTE, SoftDirty::NONE, &SoftGPU::Execute_Prim },
 	{ GE_CMD_BEZIER, FLAG_EXECUTE, SoftDirty::NONE, &SoftGPU::Execute_Bezier },
@@ -1031,6 +1031,11 @@ void SoftGPU::Execute_FramebufFormat(u32 op, u32 diff) {
 		drawEngine_->transformUnit.Flush("framebuf");
 }
 
+void SoftGPU::Execute_BoundingBox(u32 op, u32 diff) {
+	gstate_c.Dirty(DIRTY_CULL_PLANES);
+	GPUCommon::Execute_BoundingBox(op, diff);
+}
+
 void SoftGPU::Execute_ZbufPtr(u32 op, u32 diff) {
 	// We assume depthbuf.data won't change while we're drawing.
 	if (diff) {
@@ -1081,6 +1086,7 @@ void SoftGPU::Execute_WorldMtxData(u32 op, u32 diff) {
 		if (newVal != *target) {
 			*target = newVal;
 			dirtyFlags_ |= SoftDirty::TRANSFORM_MATRIX;
+			gstate_c.Dirty(DIRTY_CULL_PLANES);
 		}
 	}
 
@@ -1101,6 +1107,7 @@ void SoftGPU::Execute_ViewMtxData(u32 op, u32 diff) {
 		if (newVal != *target) {
 			*target = newVal;
 			dirtyFlags_ |= SoftDirty::TRANSFORM_MATRIX;
+			gstate_c.Dirty(DIRTY_CULL_PLANES);
 		}
 	}
 
@@ -1121,6 +1128,7 @@ void SoftGPU::Execute_ProjMtxData(u32 op, u32 diff) {
 		if (newVal != *target) {
 			*target = newVal;
 			dirtyFlags_ |= SoftDirty::TRANSFORM_MATRIX;
+			gstate_c.Dirty(DIRTY_CULL_PLANES);
 		}
 	}
 

GPU/Software/SoftGpu.h

@@ -187,6 +187,9 @@ public:
 	// Overridden to change flushing behavior.
 	void Execute_Call(u32 op, u32 diff);
 
+	// Overridden for a dirty flag change.
+	void Execute_BoundingBox(u32 op, u32 diff);
+
 	void Execute_WorldMtxNum(u32 op, u32 diff);
 	void Execute_ViewMtxNum(u32 op, u32 diff);
 	void Execute_ProjMtxNum(u32 op, u32 diff);