only do gl2cl once

add contact stats option
re-enable a few more demos

build/premake4.lua

@@ -97,10 +97,10 @@
 --		include "../opencl/vector_add"
 		include "../opencl/basic_initialize"
 		include "../demo/gpu_initialize"
---		include "../opencl/parallel_primitives/host"
+		include "../opencl/parallel_primitives/host"
---		include "../opencl/parallel_primitives/test"
+		include "../opencl/parallel_primitives/test"
---		include "../opencl/parallel_primitives/benchmark"
+		include "../opencl/parallel_primitives/benchmark"
---		include "../opencl/lds_bank_conflict"
+		include "../opencl/lds_bank_conflict"
 --		include "../opencl/reduce"
 --		include "../opencl/gpu_broadphase/test"
 --		include "../opencl/gpu_narrowphase/test"

demo/gpudemo/main_opengl3core.cpp

@@ -67,9 +67,6 @@ GpuDemo::CreateFunc* allDemos[]=
 {
 //		ConcaveCompound2Scene::MyCreateFunc,
 
-	ConcaveCompoundScene::MyCreateFunc,
-
-	GpuCompoundPlaneScene::MyCreateFunc,
 
 
 	GpuBoxPlaneScene::MyCreateFunc,	
@@ -84,10 +81,14 @@ GpuDemo::CreateFunc* allDemos[]=
 
 	ConcaveScene::MyCreateFunc,
 	
-	GpuSphereScene::MyCreateFunc,
+
-	
 	
 
+	ConcaveCompoundScene::MyCreateFunc,
+
+	GpuCompoundPlaneScene::MyCreateFunc,
+
+	GpuSphereScene::MyCreateFunc,
 	
 	
 	PairBench::MyCreateFunc,	

demo/gpudemo/rigidbody/ConcaveScene.cpp

@@ -549,7 +549,7 @@ void ConcaveCompound2Scene::createDynamicObjects(const ConstructionInfo& ci)
 					float mass = 1;//j==0? 0.f : 1.f;
 
 					//b3Vector3 position(i*10*ci.gapX,j*ci.gapY,k*10*ci.gapZ);
-					b3Vector3 position(i*10*ci.gapX,50+j*ci.gapY,k*10*ci.gapZ);
+					b3Vector3 position(i*10*ci.gapX,10+j*ci.gapY,k*10*ci.gapZ);
 
 				//	b3Quaternion orn(0,0,0,1);
 					b3Quaternion orn(b3Vector3(0,0,1),1.8);
@@ -665,7 +665,7 @@ b3Vector3 childPositions[3] = {
 			{
 				float mass = 1;//j==0? 0.f : 1.f;
 
-				b3Vector3 position(i*ci.gapX,150+j*ci.gapY,k*ci.gapZ);
+				b3Vector3 position(i*ci.gapX,50+j*ci.gapY,k*ci.gapZ);
 				//b3Quaternion orn(0,0,0,1);
 				b3Quaternion orn(b3Vector3(1,0,0),0.7);
 				

demo/gpudemo/rigidbody/GpuRigidBodyDemo.cpp

@@ -158,18 +158,22 @@ void GpuRigidBodyDemo::clientMoveAndDisplay()
 	if (animate && numObjects)
 	{
 		BT_PROFILE("gl2cl");
-		GLuint vbo = m_instancingRenderer->getInternalData()->m_vbo;
+		
-		int arraySizeInBytes  = numObjects * (3)*sizeof(btVector4);
-		glBindBuffer(GL_ARRAY_BUFFER, vbo);
-		cl_bool blocking=  CL_TRUE;
-		positions=  (btVector4*)glMapBufferRange( GL_ARRAY_BUFFER,m_instancingRenderer->getMaxShapeCapacity(),arraySizeInBytes, GL_MAP_WRITE_BIT|GL_MAP_READ_BIT );//GL_READ_WRITE);//GL_WRITE_ONLY
-		GLint err = glGetError();
-		assert(err==GL_NO_ERROR);
 		if (!m_data->m_instancePosOrnColor)
 		{
+			GLuint vbo = m_instancingRenderer->getInternalData()->m_vbo;
+			int arraySizeInBytes  = numObjects * (3)*sizeof(btVector4);
+			glBindBuffer(GL_ARRAY_BUFFER, vbo);
+			cl_bool blocking=  CL_TRUE;
+			positions=  (btVector4*)glMapBufferRange( GL_ARRAY_BUFFER,m_instancingRenderer->getMaxShapeCapacity(),arraySizeInBytes, GL_MAP_READ_BIT );//GL_READ_WRITE);//GL_WRITE_ONLY
+			GLint err = glGetError();
+			assert(err==GL_NO_ERROR);
 			m_data->m_instancePosOrnColor = new btOpenCLArray<btVector4>(m_clData->m_clContext,m_clData->m_clQueue);
 			m_data->m_instancePosOrnColor->resize(3*numObjects);
 			m_data->m_instancePosOrnColor->copyFromHostPointer(positions,3*numObjects,0);
+			glUnmapBuffer( GL_ARRAY_BUFFER);
+			err = glGetError();
+			assert(err==GL_NO_ERROR);
 		}
 	}
 	
@@ -196,6 +200,13 @@ void GpuRigidBodyDemo::clientMoveAndDisplay()
 		BT_PROFILE("cl2gl_upload");
 		GLint err = glGetError();
 		assert(err==GL_NO_ERROR);
+		GLuint vbo = m_instancingRenderer->getInternalData()->m_vbo;
+		int arraySizeInBytes  = numObjects * (3)*sizeof(btVector4);
+		glBindBuffer(GL_ARRAY_BUFFER, vbo);
+		cl_bool blocking=  CL_TRUE;
+		positions=  (btVector4*)glMapBufferRange( GL_ARRAY_BUFFER,m_instancingRenderer->getMaxShapeCapacity(),arraySizeInBytes, GL_MAP_WRITE_BIT );//GL_READ_WRITE);//GL_WRITE_ONLY
+		err = glGetError();
+		assert(err==GL_NO_ERROR);
 		m_data->m_instancePosOrnColor->copyToHostPointer(positions,3*numObjects,0);
 		glUnmapBuffer( GL_ARRAY_BUFFER);
 		err = glGetError();

opencl/gpu_rigidbody/host/b3GpuBatchingPgsSolver.cpp

@@ -591,7 +591,7 @@ void b3GpuBatchingPgsSolver::solveContacts(int numBodies, cl_mem bodyBuf, cl_mem
 					if (gpuBatchContacts)
 					{
 						BT_PROFILE("gpu batchContacts");
-						maxNumBatches = 25;//250;
+						maxNumBatches = 50;//250;
 						m_data->m_solverGPU->batchContacts( m_data->m_pBufContactOutGPU, nContacts, m_data->m_solverGPU->m_numConstraints, m_data->m_solverGPU->m_offsets, csCfg.m_staticIdx );
 					} else
 					{
@@ -629,10 +629,17 @@ void b3GpuBatchingPgsSolver::solveContacts(int numBodies, cl_mem bodyBuf, cl_mem
                                 
 									
 									int simdWidth =64;//-1;//32;
+									//int numBatches = sortConstraintByBatch( &cpuContacts[0]+offset, n, simdWidth,csCfg.m_staticIdx ,numBodies);	//	on GPU
 									int numBatches = sortConstraintByBatch3( &cpuContacts[0]+offset, n, simdWidth,csCfg.m_staticIdx ,numBodies);	//	on GPU
 									
 
 									maxNumBatches = btMax(numBatches,maxNumBatches);
+									static int globalMaxBatch = 0;
+									if (maxNumBatches>globalMaxBatch )
+									{
+										globalMaxBatch  = maxNumBatches;
+										printf("maxNumBatches = %d\n",maxNumBatches);
+									}
                                 
 									clFinish(m_data->m_queue);
                                 
@@ -724,7 +731,7 @@ static bool sortfnc(const btSortData& a,const btSortData& b)
 
 
 
-b3AlignedObjectArray<int> bodyUsed;
+
 
 
 
@@ -736,11 +743,7 @@ b3AlignedObjectArray<b3Contact4> old;
 
 inline int b3GpuBatchingPgsSolver::sortConstraintByBatch( b3Contact4* cs, int n, int simdWidth , int staticIdx, int numBodies)
 {
-	b3AlignedObjectArray<int> bodyUsed;
+	
-	bodyUsed.resize(numBodies);
-	for (int q=0;q<numBodies;q++)
-		bodyUsed[q]=0;
-
 	BT_PROFILE("sortConstraintByBatch");
 	int numIter = 0;
     
@@ -759,7 +762,8 @@ inline int b3GpuBatchingPgsSolver::sortConstraintByBatch( b3Contact4* cs, int n,
 	for(int i=0; i<n; i++)
 		cs[i].getBatchIdx() = -1;
 #endif
-	for(int i=0; i<n; i++) idxSrc[i] = i;
+	for(int i=0; i<n; i++) 
+		idxSrc[i] = i;
 	nIdxSrc = n;
     
 	int batchIdx = 0;
@@ -803,7 +807,7 @@ inline int b3GpuBatchingPgsSolver::sortConstraintByBatch( b3Contact4* cs, int n,
                 
 				if( aUnavailable==0 && bUnavailable==0 ) // ok
 				{
-					if (!!aIsStatic)
+					if (!aIsStatic)
 						flg[ aIdx/32 ] |= (1<<(aIdx&31));
 					if (!bIsStatic)
 						flg[ bIdx/32 ] |= (1<<(bIdx&31));
@@ -861,17 +865,19 @@ inline int b3GpuBatchingPgsSolver::sortConstraintByBatch( b3Contact4* cs, int n,
 }
 
 
+b3AlignedObjectArray<int> bodyUsed2;
+
 inline int b3GpuBatchingPgsSolver::sortConstraintByBatch2( b3Contact4* cs, int numConstraints, int simdWidth , int staticIdx, int numBodies)
 {
 	
-	BT_PROFILE("sortConstraintByBatch");
+	BT_PROFILE("sortConstraintByBatch2");
 	
 
-
+	
-	bodyUsed.resize(2*simdWidth);
+	bodyUsed2.resize(2*simdWidth);
 
 	for (int q=0;q<2*simdWidth;q++)
-		bodyUsed[q]=0;
+		bodyUsed2[q]=0;
 
 	int curBodyUsed = 0;
 
@@ -905,7 +911,7 @@ inline int b3GpuBatchingPgsSolver::sortConstraintByBatch2( b3Contact4* cs, int n
 			int nCurrentBatch = 0;
 			//	clear flag
 			for(int i=0; i<curBodyUsed; i++) 
-				bodyUsed[i] = 0;
+				bodyUsed2[i] = 0;
             curBodyUsed = 0;
 
 			for(int i=numValidConstraints; i<numConstraints; i++)
@@ -925,7 +931,7 @@ inline int b3GpuBatchingPgsSolver::sortConstraintByBatch2( b3Contact4* cs, int n
 				{
 					for (int j=0;j<curBodyUsed;j++)
 					{
-						if (bodyA == bodyUsed[j])
+						if (bodyA == bodyUsed2[j])
 						{
 							aUnavailable=1;
 							break;
@@ -937,7 +943,7 @@ inline int b3GpuBatchingPgsSolver::sortConstraintByBatch2( b3Contact4* cs, int n
 				{
 					for (int j=0;j<curBodyUsed;j++)
 					{
-						if (bodyB == bodyUsed[j])
+						if (bodyB == bodyUsed2[j])
 						{
 							bUnavailable=1;
 							break;
@@ -949,11 +955,11 @@ inline int b3GpuBatchingPgsSolver::sortConstraintByBatch2( b3Contact4* cs, int n
 				{
 					if (!aIsStatic)
 					{
-						bodyUsed[curBodyUsed++] = bodyA;
+						bodyUsed2[curBodyUsed++] = bodyA;
 					}
 					if (!bIsStatic)
 					{
-						bodyUsed[curBodyUsed++] = bodyB;
+						bodyUsed2[curBodyUsed++] = bodyB;
 					}
 
 					cs[idx].getBatchIdx() = batchIdx;
@@ -972,7 +978,7 @@ inline int b3GpuBatchingPgsSolver::sortConstraintByBatch2( b3Contact4* cs, int n
 						{
 							nCurrentBatch = 0;
 							for(int i=0; i<curBodyUsed; i++) 
-								bodyUsed[i] = 0;
+								bodyUsed2[i] = 0;
 
 							
 							curBodyUsed = 0;
@@ -1016,26 +1022,34 @@ inline int b3GpuBatchingPgsSolver::sortConstraintByBatch2( b3Contact4* cs, int n
 }
 
 
+b3AlignedObjectArray<int> bodyUsed;
+b3AlignedObjectArray<int> curUsed;
+
+
 inline int b3GpuBatchingPgsSolver::sortConstraintByBatch3( b3Contact4* cs, int numConstraints, int simdWidth , int staticIdx, int numBodies)
 {
 	
-	BT_PROFILE("sortConstraintByBatch");
+	BT_PROFILE("sortConstraintByBatch3");
 	
 	static int maxSwaps = 0;
 	int numSwaps = 0;
 
+	curUsed.resize(2*simdWidth);
+
 	static int maxNumConstraints = 0;
 	if (maxNumConstraints<numConstraints)
 	{
 		maxNumConstraints = numConstraints;
-		printf("maxNumConstraints  = %d\n",maxNumConstraints );
+		//printf("maxNumConstraints  = %d\n",maxNumConstraints );
 	}
 
-	bodyUsed.resize(2*simdWidth);
+	int numUsedArray = numBodies/32+1;
+	bodyUsed.resize(numUsedArray);
 
-	for (int q=0;q<2*simdWidth;q++)
+	for (int q=0;q<numUsedArray;q++)
 		bodyUsed[q]=0;
 
+	
 	int curBodyUsed = 0;
 
 	int numIter = 0;
@@ -1065,7 +1079,8 @@ inline int b3GpuBatchingPgsSolver::sortConstraintByBatch3( b3Contact4* cs, int n
 			int nCurrentBatch = 0;
 			//	clear flag
 			for(int i=0; i<curBodyUsed; i++) 
-				bodyUsed[i] = 0;
+				bodyUsed[curUsed[i]/32] = 0;
+
             curBodyUsed = 0;
 
 			for(int i=numValidConstraints; i<numConstraints; i++)
@@ -1083,37 +1098,25 @@ inline int b3GpuBatchingPgsSolver::sortConstraintByBatch3( b3Contact4* cs, int n
 				int bUnavailable = 0;
 				if (!aIsStatic)
 				{
-					for (int j=0;j<curBodyUsed;j++)
+					aUnavailable = bodyUsed[ bodyA/32 ] & (1<<(bodyA&31));
-					{
-						if (bodyA == bodyUsed[j])
-						{
-							aUnavailable=1;
-							break;
-						}
-					}
 				}
 				if (!aUnavailable)
 				if (!bIsStatic)
 				{
-					for (int j=0;j<curBodyUsed;j++)
+					bUnavailable = bodyUsed[ bodyB/32 ] & (1<<(bodyB&31));
-					{
-						if (bodyB == bodyUsed[j])
-						{
-							bUnavailable=1;
-							break;
-						}
-					}
 				}
                 
 				if( aUnavailable==0 && bUnavailable==0 ) // ok
 				{
 					if (!aIsStatic)
 					{
-						bodyUsed[curBodyUsed++] = bodyA;
+						bodyUsed[ bodyA/32 ] |= (1<<(bodyA&31));
+						curUsed[curBodyUsed++]=bodyA;
 					}
 					if (!bIsStatic)
 					{
-						bodyUsed[curBodyUsed++] = bodyB;
+						bodyUsed[ bodyB/32 ] |= (1<<(bodyB&31));
+						curUsed[curBodyUsed++]=bodyB;
 					}
 
 					cs[idx].getBatchIdx() = batchIdx;
@@ -1131,7 +1134,7 @@ inline int b3GpuBatchingPgsSolver::sortConstraintByBatch3( b3Contact4* cs, int n
 						{
 							nCurrentBatch = 0;
 							for(int i=0; i<curBodyUsed; i++) 
-								bodyUsed[i] = 0;
+								bodyUsed[curUsed[i]/32] = 0;
 							curBodyUsed = 0;
 						}
 					}
@@ -1152,7 +1155,7 @@ inline int b3GpuBatchingPgsSolver::sortConstraintByBatch3( b3Contact4* cs, int n
 	if (maxSwaps<numSwaps)
 	{
 		maxSwaps = numSwaps;
-		printf("maxSwaps = %d\n", maxSwaps);
+		//printf("maxSwaps = %d\n", maxSwaps);
 	}
 	
 	return batchIdx;

opencl/gpu_rigidbody/host/b3GpuNarrowPhase.cpp

@@ -698,6 +698,11 @@ cl_mem b3GpuNarrowPhase::getContactsGpu()
 	return m_data->m_pBufContactOutGPU->getBufferCL();
 }
 
+const b3Contact4* b3GpuNarrowPhase::getContactsCPU() const
+{
+	m_data->m_pBufContactOutGPU->copyToHost(*m_data->m_pBufContactOutCPU);
+	return &m_data->m_pBufContactOutCPU->at(0);
+}
 
 void b3GpuNarrowPhase::computeContacts(cl_mem broadphasePairs, int numBroadphasePairs, cl_mem aabbsWS, int numObjects)
 {

opencl/gpu_rigidbody/host/b3GpuNarrowPhase.h

@@ -64,6 +64,9 @@ public:
 	cl_mem	getCollidablesGpu();
 	int		getNumCollidablesGpu() const;
 
+
+	const struct b3Contact4* getContactsCPU() const;
+
 	cl_mem	getContactsGpu();
 	int	getNumContactsGpu() const;
 

opencl/gpu_rigidbody/host/b3GpuRigidBodyPipeline.cpp

@@ -23,6 +23,8 @@
 #include "Bullet3Common/b3Quickprof.h"
 #include "b3Config.h"
 
+bool dumpContactStats = false;
+
 b3GpuRigidBodyPipeline::b3GpuRigidBodyPipeline(cl_context ctx,cl_device_id device, cl_command_queue  q,class b3GpuNarrowPhase* narrowphase, class b3GpuSapBroadphase* broadphaseSap )
 {
 	m_data = new b3GpuRigidBodyPipelineInternalData;
@@ -95,6 +97,7 @@ void	b3GpuRigidBodyPipeline::stepSimulation(float deltaTime)
 	int numPairs = m_data->m_broadphaseSap->getNumOverlap();
 	int numContacts  = 0;
 
+
 	int numBodies = m_data->m_narrowphase->getNumBodiesGpu();
 
 	if (numPairs)
@@ -105,8 +108,23 @@ void	b3GpuRigidBodyPipeline::stepSimulation(float deltaTime)
 
 		m_data->m_narrowphase->computeContacts(pairs,numPairs,aabbsWS,numBodies);
 		numContacts = m_data->m_narrowphase->getNumContactsGpu();
-		//if (numContacts)
+
-		//	printf("numContacts = %d\n", numContacts);
+		if (dumpContactStats && numContacts)
+		{
+			m_data->m_narrowphase->getContactsGpu();
+			
+			printf("numContacts = %d\n", numContacts);
+
+			int totalPoints  = 0;
+			const b3Contact4* contacts = m_data->m_narrowphase->getContactsCPU();
+
+			for (int i=0;i<numContacts;i++)
+			{
+				totalPoints += contacts->getNPoints();
+			}
+			printf("totalPoints=%d\n",totalPoints);
+
+		}
 	}
 	
 

opencl/parallel_primitives/benchmark/test_large_problem_sorting.cpp

@@ -641,7 +641,7 @@ extern bool gDebugSkipLoadingBinary;
 
 int main( int argc, char** argv) 
 {
-	gDebugSkipLoadingBinary = true;
+	//gDebugSkipLoadingBinary = true;
 
 	cl_int ciErrNum;
 	b3CommandLineArgs args(argc,argv);