TransformedAABBoxSSE move mCumulativeMatrix to the stack too

rygorous · Feb 24, 2013 · 028a108 · 028a108
1 parent 0fad7d4
commit 028a108
Show file tree

Hide file tree

Showing 5 changed files with 16 additions and 15 deletions.
diff --git a/SoftwareOcclusionCulling/AABBoxRasterizerSSE.cpp b/SoftwareOcclusionCulling/AABBoxRasterizerSSE.cpp
@@ -189,6 +189,8 @@ void AABBoxRasterizerSSE::Render(CPUTAssetSet **pAssetSet,
 	BoxTestSetup setup;
 	setup.Init(mViewMatrix, mProjMatrix, mpCamera, mOccludeeSizeThreshold);
 
+	__m128 cumulativeMatrix[4];
+
 	for(UINT assetId = 0, modelId = 0; assetId < numAssetSets; assetId++)
 	{
 		for(UINT nodeId = 0; nodeId < pAssetSet[assetId]->GetAssetCount(); nodeId++)
@@ -198,7 +200,7 @@ void AABBoxRasterizerSSE::Render(CPUTAssetSet **pAssetSet,
 			ASSERT((CPUT_SUCCESS == result), _L ("Failed getting asset by index")); 
 			if(pRenderNode->IsModel())
 			{
-				if(!mpTransformedAABBox[modelId].IsTooSmall(setup))
+				if(!mpTransformedAABBox[modelId].IsTooSmall(setup, cumulativeMatrix))
 				{
 					CPUTModelDX11* model = (CPUTModelDX11*)pRenderNode;
 					model = (CPUTModelDX11*)pRenderNode;

diff --git a/SoftwareOcclusionCulling/AABBoxRasterizerSSEMT.cpp b/SoftwareOcclusionCulling/AABBoxRasterizerSSEMT.cpp
@@ -94,6 +94,7 @@ void AABBoxRasterizerSSEMT::TransformAABBoxAndDepthTest(UINT taskId)
 	setup.Init(mViewMatrix, mProjMatrix, mpCamera, mOccludeeSizeThreshold);
 
 	__m128 xformedPos[AABB_VERTICES];
+	__m128 cumulativeMatrix[4];
 
 	static const UINT kChunkSize = 64;
 	for(UINT base = taskId*kChunkSize; base < mNumModels; base += mNumDepthTestTasks * kChunkSize)
@@ -103,9 +104,9 @@ void AABBoxRasterizerSSEMT::TransformAABBoxAndDepthTest(UINT taskId)
 		{
 			mpVisible[i] = false;
 
-			if(mpBBoxVisible[i] && !mpTransformedAABBox[i].IsTooSmall(setup))
+			if(mpBBoxVisible[i] && !mpTransformedAABBox[i].IsTooSmall(setup, cumulativeMatrix))
 			{
-				if(mpTransformedAABBox[i].TransformAABBox(xformedPos))
+				if(mpTransformedAABBox[i].TransformAABBox(xformedPos, cumulativeMatrix))
 					mpVisible[i] = mpTransformedAABBox[i].RasterizeAndDepthTestAABBox(mpRenderTargetPixels, xformedPos);
 				else
 					mpVisible[i] = true;

diff --git a/SoftwareOcclusionCulling/AABBoxRasterizerSSEST.cpp b/SoftwareOcclusionCulling/AABBoxRasterizerSSEST.cpp
@@ -51,14 +51,15 @@ void AABBoxRasterizerSSEST::TransformAABBoxAndDepthTest()
 	setup.Init(mViewMatrix, mProjMatrix, mpCamera, mOccludeeSizeThreshold);
 
 	__m128 xformedPos[AABB_VERTICES];
+	__m128 cumulativeMatrix[4];
 
 	for(UINT i = 0; i < mNumModels; i++)
 	{
 		mpVisible[i] = false;
 
-		if(mpBBoxVisible[i] && !mpTransformedAABBox[i].IsTooSmall(setup))
+		if(mpBBoxVisible[i] && !mpTransformedAABBox[i].IsTooSmall(setup, cumulativeMatrix))
 		{
-			if(mpTransformedAABBox[i].TransformAABBox(xformedPos))
+			if(mpTransformedAABBox[i].TransformAABBox(xformedPos, cumulativeMatrix))
 				mpVisible[i] = mpTransformedAABBox[i].RasterizeAndDepthTestAABBox(mpRenderTargetPixels, xformedPos);
 			else
 				mpVisible[i] = true;

diff --git a/SoftwareOcclusionCulling/TransformedAABBoxSSE.cpp b/SoftwareOcclusionCulling/TransformedAABBoxSSE.cpp
@@ -65,14 +65,12 @@ TransformedAABBoxSSE::TransformedAABBoxSSE()
 {
 	mWorldMatrix = (__m128*)_aligned_malloc(sizeof(float) * 4 * 4, 16);
 	mpBBVertexList = (__m128*)_aligned_malloc(sizeof(float) * 4 * AABB_VERTICES, 16);
-	mCumulativeMatrix = (__m128*)_aligned_malloc(sizeof(float) * 4 * 4, 16); 
 }
 
 TransformedAABBoxSSE::~TransformedAABBoxSSE()
 {
 	_aligned_free(mWorldMatrix);
 	_aligned_free(mpBBVertexList);
-	_aligned_free(mCumulativeMatrix);
 }
 
 //--------------------------------------------------------------------------
@@ -111,12 +109,12 @@ void TransformedAABBoxSSE::CreateAABBVertexIndexList(CPUTModelDX11 *pModel)
 //----------------------------------------------------------------------------
 // Determine if the occluddee size is too small and if so avoid drawing it
 //----------------------------------------------------------------------------
-bool TransformedAABBoxSSE::IsTooSmall(const BoxTestSetup &setup)
+bool TransformedAABBoxSSE::IsTooSmall(const BoxTestSetup &setup, __m128 cumulativeMatrix[4])
 {
-	MatrixMultiply(mWorldMatrix, setup.mViewProjViewport, mCumulativeMatrix);
+	MatrixMultiply(mWorldMatrix, setup.mViewProjViewport, cumulativeMatrix);
 
 	__m128 center = _mm_set_ps(1.0f, mBBCenter.z, mBBCenter.y, mBBCenter.x);
-	__m128 mBBCenterOSxForm = TransformCoords(&center, mCumulativeMatrix);
+	__m128 mBBCenterOSxForm = TransformCoords(&center, cumulativeMatrix);
     float w = mBBCenterOSxForm.m128_f32[3];
 	if( w > 1.0f )
 	{
@@ -129,13 +127,13 @@ bool TransformedAABBoxSSE::IsTooSmall(const BoxTestSetup &setup)
 //----------------------------------------------------------------
 // Trasforms the AABB vertices to screen space once every frame
 //----------------------------------------------------------------
-bool TransformedAABBoxSSE::TransformAABBox(__m128 xformedPos[])
+bool TransformedAABBoxSSE::TransformAABBox(__m128 xformedPos[], const __m128 cumulativeMatrix[4])
 {
 	__m128 zAllIn = _mm_castsi128_ps(_mm_set1_epi32(~0));
 
 	for(UINT i = 0; i < AABB_VERTICES; i++)
 	{
-		__m128 vert = TransformCoords(&mpBBVertexList[i], mCumulativeMatrix);
+		__m128 vert = TransformCoords(&mpBBVertexList[i], cumulativeMatrix);
 
 		// We have inverted z; z is in front of near plane iff z <= w.
 		__m128 vertZ = _mm_shuffle_ps(vert, vert, 0xaa); // vert.zzzz

diff --git a/SoftwareOcclusionCulling/TransformedAABBoxSSE.h b/SoftwareOcclusionCulling/TransformedAABBoxSSE.h
@@ -39,17 +39,16 @@ class TransformedAABBoxSSE : public HelperSSE
 		void CreateAABBVertexIndexList(CPUTModelDX11 *pModel);
 		void TransformAABBoxAndDepthTest();
 
-		bool IsTooSmall(const BoxTestSetup &setup);
+		bool IsTooSmall(const BoxTestSetup &setup, __m128 cumulativeMatrix[4]);
 
-		bool TransformAABBox(__m128 xformedPos[]);
+		bool TransformAABBox(__m128 xformedPos[], const __m128 cumulativeMatrix[4]);
 
 		bool RasterizeAndDepthTestAABBox(UINT *pRenderTargetPixels, const __m128 xformedPos[]);
 
 	private:
 		CPUTModelDX11 *mpCPUTModel;
 		__m128 *mWorldMatrix;
 		__m128 *mpBBVertexList;
-		__m128 *mCumulativeMatrix; 
 
 		float3 mBBCenter;
 		float  mRadiusSq;