Bart/bullet3
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

added Samurai Monastry wavefront .obj

Browse Source 
added alternative batching kernel (slow)
tweaked controls a bit
added command-line options --selected_demo=<int> and --new_batching
started looking into parallel 3d sap
This commit is contained in:
erwin coumans
2013-03-23 23:00:50 -07:00
parent 11d934b63a
commit 68062bdfbd
18 changed files with 333195 additions and 76 deletions
Download Patch File Download Diff File Expand all files Collapse all files

114
opencl/gpu_broadphase/host/btGpuSapBroadphase.cpp
View File

@@ -20,7 +20,8 @@ m_smallAabbsGPU(ctx,q),
m_largeAabbsGPU(ctx,q),
m_overlappingPairs(ctx,q),
m_gpuSmallSortData(ctx,q),
m_gpuSmallSortedAabbs(ctx,q)
m_gpuSmallSortedAabbs(ctx,q),
m_currentBuffer(-1)
{
const char* sapSrc = sapCL;
const char* sapFastSrc = sapFastCL;
@@ -86,18 +87,87 @@ static bool TestAabbAgainstAabb2(const btVector3 &aabbMin1, const btVector3 &aab
return overlap;
}
void btGpuSapBroadphase::calculateOverlappingPairs(bool forceHost)
//http://stereopsis.com/radix.html
static unsigned int FloatFlip(float fl)
{
int axis = 0;//todo on GPU for now hardcode
unsigned int f = *(unsigned int*)&fl;
unsigned int mask = -(int)(f >> 31) | 0x80000000;
return f ^ mask;
};
void btGpuSapBroadphase::init3dSap()
{
if (m_currentBuffer<0)
{
m_allAabbsGPU.copyToHost(m_allAabbsCPU);
m_currentBuffer = 0;
for (int axis=0;axis<3;axis++)
{
for (int buf=0;buf<2;buf++)
{
int totalNumAabbs = m_allAabbsCPU.size();
m_sortedAxisCPU[axis][buf].resize(totalNumAabbs);
if (buf==m_currentBuffer)
{
for (int i=0;i<totalNumAabbs;i++)
{
m_sortedAxisCPU[axis][buf][i].m_key = FloatFlip(m_allAabbsCPU[i].m_minIndices[axis]);
m_sortedAxisCPU[axis][buf][i].m_value = i;
}
}
}
}
}
}
void btGpuSapBroadphase::calculateOverlappingPairsHostIncremental3Sap()
{
btAssert(m_currentBuffer>=0);
if (m_currentBuffer<0)
return;
m_allAabbsGPU.copyToHost(m_allAabbsCPU);
for (int axis=0;axis<3;axis++)
{
for (int buf=0;buf<2;buf++)
{
btAssert(m_sortedAxisCPU[axis][buf].size() == m_allAabbsCPU.size());
}
}
m_currentBuffer = 1-m_currentBuffer;
for (int axis=0;axis<3;axis++)
{
int totalNumAabbs = m_allAabbsCPU.size();
for (int i=0;i<totalNumAabbs;i++)
{
m_sortedAxisCPU[axis][m_currentBuffer][i].m_key = FloatFlip(m_allAabbsCPU[i].m_minIndices[axis]);
m_sortedAxisCPU[axis][m_currentBuffer][i].m_value = i;
}
}
}
void btGpuSapBroadphase::calculateOverlappingPairsHost()
{
//test
//if (m_currentBuffer>=0)
// calculateOverlappingPairsHostIncremental3Sap();
int axis=0;
btAssert(m_allAabbsCPU.size() == m_allAabbsGPU.size());
if (forceHost)
{
btAlignedObjectArray<btSapAabb> allHostAabbs;
m_allAabbsGPU.copyToHost(allHostAabbs);
m_allAabbsGPU.copyToHost(m_allAabbsCPU);
{
int numSmallAabbs = m_smallAabbsCPU.size();
@@ -105,7 +175,7 @@ void btGpuSapBroadphase::calculateOverlappingPairs(bool forceHost)
{
//sync aabb
int aabbIndex = m_smallAabbsCPU[j].m_signedMaxIndices[3];
m_smallAabbsCPU[j] = allHostAabbs[aabbIndex];
m_smallAabbsCPU[j] = m_allAabbsCPU[aabbIndex];
m_smallAabbsCPU[j].m_signedMaxIndices[3] = aabbIndex;
}
}
@@ -116,7 +186,7 @@ void btGpuSapBroadphase::calculateOverlappingPairs(bool forceHost)
{
//sync aabb
int aabbIndex = m_largeAabbsCPU[j].m_signedMaxIndices[3];
m_largeAabbsCPU[j] = allHostAabbs[aabbIndex];
m_largeAabbsCPU[j] = m_allAabbsCPU[aabbIndex];
m_largeAabbsCPU[j].m_signedMaxIndices[3] = aabbIndex;
}
@@ -175,8 +245,15 @@ void btGpuSapBroadphase::calculateOverlappingPairs(bool forceHost)
m_overlappingPairs.resize(0);
}
return;
}
//init3dSap();
}
void btGpuSapBroadphase::calculateOverlappingPairs()
{
int axis = 0;//todo on GPU for now hardcode
{
@@ -185,8 +262,8 @@ void btGpuSapBroadphase::calculateOverlappingPairs(bool forceHost)
if (syncOnHost)
{
BT_PROFILE("Synchronize m_smallAabbsGPU (CPU/slow)");
btAlignedObjectArray<btSapAabb> allHostAabbs;
m_allAabbsGPU.copyToHost(allHostAabbs);
m_allAabbsGPU.copyToHost(m_allAabbsCPU);
m_smallAabbsGPU.copyToHost(m_smallAabbsCPU);
{
@@ -195,7 +272,7 @@ void btGpuSapBroadphase::calculateOverlappingPairs(bool forceHost)
{
//sync aabb
int aabbIndex = m_smallAabbsCPU[j].m_signedMaxIndices[3];
m_smallAabbsCPU[j] = allHostAabbs[aabbIndex];
m_smallAabbsCPU[j] = m_allAabbsCPU[aabbIndex];
m_smallAabbsCPU[j].m_signedMaxIndices[3] = aabbIndex;
}
}
@@ -226,8 +303,8 @@ void btGpuSapBroadphase::calculateOverlappingPairs(bool forceHost)
if (syncOnHost)
{
BT_PROFILE("Synchronize m_largeAabbsGPU (CPU/slow)");
btAlignedObjectArray<btSapAabb> allHostAabbs;
m_allAabbsGPU.copyToHost(allHostAabbs);
m_allAabbsGPU.copyToHost(m_allAabbsCPU);
m_largeAabbsGPU.copyToHost(m_largeAabbsCPU);
{
@@ -236,7 +313,7 @@ void btGpuSapBroadphase::calculateOverlappingPairs(bool forceHost)
{
//sync aabb
int aabbIndex = m_largeAabbsCPU[j].m_signedMaxIndices[3];
m_largeAabbsCPU[j] = allHostAabbs[aabbIndex];
m_largeAabbsCPU[j] = m_allAabbsCPU[aabbIndex];
m_largeAabbsCPU[j].m_signedMaxIndices[3] = aabbIndex;
}
}
@@ -432,6 +509,7 @@ void btGpuSapBroadphase::calculateOverlappingPairs(bool forceHost)
}//BT_PROFILE("GPU_RADIX SORT");
}
void btGpuSapBroadphase::writeAabbsToGpu()

10
opencl/gpu_broadphase/host/btGpuSapBroadphase.h
View File

@@ -24,6 +24,10 @@ class btGpuSapBroadphase
class btRadixSort32CL* m_sorter;
///test for 3d SAP
btAlignedObjectArray<btSortData> m_sortedAxisCPU[3][2];
int m_currentBuffer;
public:
btOpenCLArray<btSapAabb> m_allAabbsGPU;
@@ -45,7 +49,11 @@ class btGpuSapBroadphase
btGpuSapBroadphase(cl_context ctx,cl_device_id device, cl_command_queue q );
virtual ~btGpuSapBroadphase();
void calculateOverlappingPairs(bool forceHost=false);
void calculateOverlappingPairs();
void calculateOverlappingPairsHost();
void init3dSap();
void calculateOverlappingPairsHostIncremental3Sap();
void createProxy(const btVector3& aabbMin, const btVector3& aabbMax, int userPtr ,short int collisionFilterGroup,short int collisionFilterMask);
void createLargeProxy(const btVector3& aabbMin, const btVector3& aabbMax, int userPtr ,short int collisionFilterGroup,short int collisionFilterMask);

42
opencl/gpu_rigidbody/host/Solver.cpp
View File

@@ -16,6 +16,8 @@ subject to the following restrictions:
#include "Solver.h"
///useNewBatchingKernel is a rewritten kernel using just a single thread of the warp, for experiments
bool useNewBatchingKernel = false;
#define SOLVER_SETUP_KERNEL_PATH "opencl/gpu_rigidbody/kernels/solverSetup.cl"
#define SOLVER_SETUP2_KERNEL_PATH "opencl/gpu_rigidbody/kernels/solverSetup2.cl"
@@ -24,6 +26,7 @@ subject to the following restrictions:
#define SOLVER_FRICTION_KERNEL_PATH "opencl/gpu_rigidbody/kernels/solveFriction.cl"
#define BATCHING_PATH "opencl/gpu_rigidbody/kernels/batchingKernels.cl"
#define BATCHING_NEW_PATH "opencl/gpu_rigidbody/kernels/batchingKernelsNew.cl"
#include "../kernels/solverSetup.h"
@@ -33,6 +36,9 @@ subject to the following restrictions:
#include "../kernels/solveFriction.h"
#include "../kernels/batchingKernels.h"
#include "../kernels/batchingKernelsNew.h"
#include "BulletCommon/btQuickprof.h"
#include "../../parallel_primitives/host/btLauncherCL.h"
#include "BulletCommon/btVector3.h"
@@ -94,7 +100,7 @@ Solver::Solver(cl_context ctx, cl_device_id device, cl_command_queue queue, int
const int sortSize = BTNEXTMULTIPLEOF( pairCapacity, 512 );
m_sortDataBuffer = new btOpenCLArray<btSortData>(ctx,queue,sortSize);
m_contactBuffer = new btOpenCLArray<btContact4>(ctx,queue);
m_contactBuffer2 = new btOpenCLArray<btContact4>(ctx,queue);
m_numConstraints = new btOpenCLArray<unsigned int>(ctx,queue,N_SPLIT*N_SPLIT );
m_numConstraints->resize(N_SPLIT*N_SPLIT);
@@ -108,6 +114,8 @@ Solver::Solver(cl_context ctx, cl_device_id device, cl_command_queue queue, int
cl_int pErrNum;
const char* batchKernelSource = batchingKernelsCL;
const char* batchKernelNewSource = batchingKernelsNewCL;
const char* solverSetupSource = solverSetupCL;
const char* solverSetup2Source = solverSetup2CL;
const char* solveContactSource = solveContactCL;
@@ -159,13 +167,20 @@ Solver::Solver(cl_context ctx, cl_device_id device, cl_command_queue queue, int
m_batchingKernel = btOpenCLUtils::compileCLKernelFromString( ctx, device, batchKernelSource, "CreateBatches", &pErrNum, batchingProg,additionalMacros );
btAssert(m_batchingKernel);
}
{
cl_program batchingNewProg = btOpenCLUtils::compileCLProgramFromString( ctx, device, batchKernelNewSource, &pErrNum,additionalMacros, BATCHING_NEW_PATH);
btAssert(batchingNewProg);
m_batchingKernelNew = btOpenCLUtils::compileCLKernelFromString( ctx, device, batchKernelNewSource, "CreateBatchesNew", &pErrNum, batchingNewProg,additionalMacros );
//m_batchingKernelNew = btOpenCLUtils::compileCLKernelFromString( ctx, device, batchKernelNewSource, "CreateBatchesBruteForce", &pErrNum, batchingNewProg,additionalMacros );
btAssert(m_batchingKernelNew);
}
}
Solver::~Solver()
{
delete m_sortDataBuffer;
delete m_contactBuffer;
delete m_contactBuffer2;
delete m_sort32;
delete m_scan;
@@ -173,6 +188,7 @@ Solver::~Solver()
clReleaseKernel(m_batchingKernel);
clReleaseKernel(m_batchingKernelNew);
clReleaseKernel( m_solveContactKernel);
clReleaseKernel( m_solveFrictionKernel);
@@ -843,7 +859,7 @@ void Solver::batchContacts( btOpenCLArray<btContact4>* contacts, int nContacts,
btBufferInfoCL bInfo[] = {
btBufferInfoCL( contacts->getBufferCL() ),
btBufferInfoCL( m_contactBuffer->getBufferCL() ),
btBufferInfoCL( m_contactBuffer2->getBufferCL()),
btBufferInfoCL( nNative->getBufferCL() ),
btBufferInfoCL( offsetsNative->getBufferCL() ),
#ifdef BATCH_DEBUG
@@ -852,10 +868,22 @@ void Solver::batchContacts( btOpenCLArray<btContact4>* contacts, int nContacts,
};
{
BT_PROFILE("batchingKernel");
btLauncherCL launcher( m_queue, m_batchingKernel);
launcher.setBuffers( bInfo, sizeof(bInfo)/sizeof(btBufferInfoCL) );
//btLauncherCL launcher( m_queue, m_batchingKernel);
cl_kernel k = useNewBatchingKernel ? m_batchingKernelNew : m_batchingKernel;
btLauncherCL launcher( m_queue, k);
if (!useNewBatchingKernel )
{
launcher.setBuffer( contacts->getBufferCL() );
}
launcher.setBuffer( m_contactBuffer2->getBufferCL() );
launcher.setBuffer( nNative->getBufferCL());
launcher.setBuffer( offsetsNative->getBufferCL());
//launcher.setConst( cdata );
launcher.setConst(staticIdx);
@@ -899,7 +927,7 @@ void Solver::batchContacts( btOpenCLArray<btContact4>* contacts, int nContacts,
}
// copy buffer to buffer
btAssert(m_contactBuffer->size()==nContacts);
//btAssert(m_contactBuffer->size()==nContacts);
//contacts->copyFromOpenCLArray( *m_contactBuffer);
//clFinish(m_queue);//needed?

3
opencl/gpu_rigidbody/host/Solver.h
View File

@@ -94,6 +94,7 @@ class Solver : public SolverBase
int m_nIterations;
cl_kernel m_batchingKernel;
cl_kernel m_batchingKernelNew;
cl_kernel m_solveContactKernel;
cl_kernel m_solveFrictionKernel;
cl_kernel m_contactToConstraintKernel;
@@ -106,7 +107,7 @@ class Solver : public SolverBase
class btPrefixScanCL* m_scan;
btOpenCLArray<btSortData>* m_sortDataBuffer;
btOpenCLArray<btContact4>* m_contactBuffer;
btOpenCLArray<btContact4>* m_contactBuffer2;
enum
{

4
opencl/gpu_rigidbody/host/btConfig.h
View File

@@ -18,7 +18,7 @@ struct btConfig
int m_maxTriConvexPairCapacity;
btConfig()
:m_maxConvexBodies(64*1024),
:m_maxConvexBodies(32*1024),
m_maxConvexShapes(8192),
m_maxVerticesPerFace(64),
m_maxFacesPerShape(64),
@@ -26,7 +26,7 @@ struct btConfig
m_maxConvexIndices(8192),
m_maxConvexUniqueEdges(8192),
m_maxCompoundChildShapes(8192),
m_maxTriConvexPairCapacity(64*1024)
m_maxTriConvexPairCapacity(512*1024)
{
m_maxBroadphasePairs = 16*m_maxConvexBodies;
}

384
opencl/gpu_rigidbody/host/btGpuBatchingPgsSolver.cpp
View File

@@ -17,12 +17,14 @@
#define SOLVER_CONTACT_KERNEL_PATH "opencl/gpu_rigidbody/kernels/solveContact.cl"
#define SOLVER_FRICTION_KERNEL_PATH "opencl/gpu_rigidbody/kernels/solveFriction.cl"
#define BATCHING_PATH "opencl/gpu_rigidbody/kernels/batchingKernels.cl"
#define BATCHING_NEW_PATH "opencl/gpu_rigidbody/kernels/batchingKernelsNew.cl"
#include "../kernels/solverSetup.h"
#include "../kernels/solverSetup2.h"
#include "../kernels/solveContact.h"
#include "../kernels/solveFriction.h"
#include "../kernels/batchingKernels.h"
#include "../kernels/batchingKernelsNew.h"
@@ -48,13 +50,13 @@ struct btGpuBatchingPgsSolverInternalData
int m_nIterations;
btOpenCLArray<btGpuConstraint4>* m_contactCGPU;
btOpenCLArray<unsigned int>* m_numConstraints;
btOpenCLArray<unsigned int>* m_offsets;
Solver* m_solverGPU;
cl_kernel m_batchingKernel;
cl_kernel m_batchingKernelNew;
cl_kernel m_solveContactKernel;
cl_kernel m_solveFrictionKernel;
cl_kernel m_contactToConstraintKernel;
@@ -72,6 +74,11 @@ struct btGpuBatchingPgsSolverInternalData
btOpenCLArray<btRigidBodyCL>* m_bodyBufferGPU;
btOpenCLArray<btInertiaCL>* m_inertiaBufferGPU;
btOpenCLArray<btContact4>* m_pBufContactOutGPU;
btAlignedObjectArray<unsigned int> m_idxBuffer;
btAlignedObjectArray<btSortData> m_sortData;
btAlignedObjectArray<btContact4> m_old;
};
@@ -114,6 +121,7 @@ btGpuBatchingPgsSolver::btGpuBatchingPgsSolver(cl_context ctx,cl_device_id devic
cl_int pErrNum;
const char* batchKernelSource = batchingKernelsCL;
const char* batchKernelNewSource = batchingKernelsNewCL;
const char* solverSetupSource = solverSetupCL;
const char* solverSetup2Source = solverSetup2CL;
const char* solveContactSource = solveContactCL;
@@ -166,7 +174,14 @@ btGpuBatchingPgsSolver::btGpuBatchingPgsSolver(cl_context ctx,cl_device_id devic
btAssert(m_data->m_batchingKernel);
}
{
cl_program batchingNewProg = btOpenCLUtils::compileCLProgramFromString( ctx, device, batchKernelNewSource, &pErrNum,additionalMacros, BATCHING_NEW_PATH);
btAssert(batchingNewProg);
m_data->m_batchingKernelNew = btOpenCLUtils::compileCLKernelFromString( ctx, device, batchKernelNewSource, "CreateBatchesNew", &pErrNum, batchingNewProg,additionalMacros );
btAssert(m_data->m_batchingKernelNew);
}
@@ -186,6 +201,7 @@ btGpuBatchingPgsSolver::~btGpuBatchingPgsSolver()
clReleaseKernel(m_data->m_batchingKernel);
clReleaseKernel(m_data->m_batchingKernelNew);
clReleaseKernel( m_data->m_solveContactKernel);
clReleaseKernel( m_data->m_solveFrictionKernel);
@@ -406,7 +422,7 @@ void btGpuBatchingPgsSolver::solveContacts(int numBodies, cl_mem bodyBuf, cl_mem
csCfg.m_averageExtent = .2f;//@TODO m_averageObjExtent;
csCfg.m_staticIdx = 0;//m_static0Index;//m_planeBodyIndex;
btOpenCLArray<btContact4>* contactsIn = m_data->m_pBufContactOutGPU;
btOpenCLArray<btRigidBodyCL>* bodyBuf = m_data->m_bodyBufferGPU;
void* additionalData = 0;//m_data->m_frictionCGPU;
@@ -419,16 +435,17 @@ void btGpuBatchingPgsSolver::solveContacts(int numBodies, cl_mem bodyBuf, cl_mem
{
if( m_data->m_solverGPU->m_contactBuffer)
if( m_data->m_solverGPU->m_contactBuffer2)
{
m_data->m_solverGPU->m_contactBuffer->resize(nContacts);
m_data->m_solverGPU->m_contactBuffer2->resize(nContacts);
}
if( m_data->m_solverGPU->m_contactBuffer == 0 )
if( m_data->m_solverGPU->m_contactBuffer2 == 0 )
{
m_data->m_solverGPU->m_contactBuffer = new btOpenCLArray<btContact4>(m_data->m_context,m_data->m_queue, nContacts );
m_data->m_solverGPU->m_contactBuffer->resize(nContacts);
m_data->m_solverGPU->m_contactBuffer2 = new btOpenCLArray<btContact4>(m_data->m_context,m_data->m_queue, nContacts );
m_data->m_solverGPU->m_contactBuffer2->resize(nContacts);
}
clFinish(m_data->m_queue);
@@ -438,7 +455,7 @@ void btGpuBatchingPgsSolver::solveContacts(int numBodies, cl_mem bodyBuf, cl_mem
//@todo: just reserve it, without copy of original contact (unless we use warmstarting)
btOpenCLArray<btContact4>* contactNative = contactsIn;
const btOpenCLArray<btRigidBodyCL>* bodyNative = bodyBuf;
@@ -477,7 +494,7 @@ void btGpuBatchingPgsSolver::solveContacts(int numBodies, cl_mem bodyBuf, cl_mem
m_data->m_solverGPU->m_sortDataBuffer->resize(nContacts);
btBufferInfoCL bInfo[] = { btBufferInfoCL( contactNative->getBufferCL() ), btBufferInfoCL( bodyBuf->getBufferCL()), btBufferInfoCL( m_data->m_solverGPU->m_sortDataBuffer->getBufferCL()) };
btBufferInfoCL bInfo[] = { btBufferInfoCL( m_data->m_pBufContactOutGPU->getBufferCL() ), btBufferInfoCL( bodyBuf->getBufferCL()), btBufferInfoCL( m_data->m_solverGPU->m_sortDataBuffer->getBufferCL()) };
btLauncherCL launcher(m_data->m_queue, m_data->m_solverGPU->m_setSortDataKernel );
launcher.setBuffers( bInfo, sizeof(bInfo)/sizeof(btBufferInfoCL) );
launcher.setConst( cdata.m_nContacts );
@@ -536,7 +553,7 @@ void btGpuBatchingPgsSolver::solveContacts(int numBodies, cl_mem bodyBuf, cl_mem
btInt4 cdata;
cdata.x = nContacts;
btBufferInfoCL bInfo[] = { btBufferInfoCL( contactNative->getBufferCL() ), btBufferInfoCL( m_data->m_solverGPU->m_contactBuffer->getBufferCL())
btBufferInfoCL bInfo[] = { btBufferInfoCL( m_data->m_pBufContactOutGPU->getBufferCL() ), btBufferInfoCL( m_data->m_solverGPU->m_contactBuffer2->getBufferCL())
, btBufferInfoCL( m_data->m_solverGPU->m_sortDataBuffer->getBufferCL()) };
btLauncherCL launcher(m_data->m_queue,m_data->m_solverGPU->m_reorderContactKernel);
launcher.setBuffers( bInfo, sizeof(bInfo)/sizeof(btBufferInfoCL) );
@@ -554,19 +571,18 @@ void btGpuBatchingPgsSolver::solveContacts(int numBodies, cl_mem bodyBuf, cl_mem
clFinish(m_data->m_queue);
if (nContacts)
{
BT_PROFILE("gpu m_copyConstraintKernel");
btInt4 cdata; cdata.x = nContacts;
btBufferInfoCL bInfo[] = { btBufferInfoCL( m_data->m_solverGPU->m_contactBuffer->getBufferCL() ), btBufferInfoCL( contactNative->getBufferCL() ) };
btLauncherCL launcher(m_data->m_queue, m_data->m_solverGPU->m_copyConstraintKernel );
launcher.setBuffers( bInfo, sizeof(bInfo)/sizeof(btBufferInfoCL) );
launcher.setConst( cdata );
launcher.launch1D( nContacts, 64 );
clFinish(m_data->m_queue);
}
{
BT_PROFILE("gpu m_copyConstraintKernel");
btInt4 cdata; cdata.x = nContacts;
btBufferInfoCL bInfo[] = { btBufferInfoCL( m_data->m_solverGPU->m_contactBuffer2->getBufferCL() ), btBufferInfoCL( m_data->m_pBufContactOutGPU->getBufferCL() ) };
btLauncherCL launcher(m_data->m_queue, m_data->m_solverGPU->m_copyConstraintKernel );
launcher.setBuffers( bInfo, sizeof(bInfo)/sizeof(btBufferInfoCL) );
launcher.setConst( cdata );
launcher.launch1D( nContacts, 64 );
clFinish(m_data->m_queue);
}
bool compareGPU = false;
@@ -575,13 +591,13 @@ void btGpuBatchingPgsSolver::solveContacts(int numBodies, cl_mem bodyBuf, cl_mem
if (gpuBatchContacts)
{
BT_PROFILE("gpu batchContacts");
maxNumBatches = 50;
m_data->m_solverGPU->batchContacts( (btOpenCLArray<btContact4>*)contactNative, nContacts, m_data->m_solverGPU->m_numConstraints, m_data->m_solverGPU->m_offsets, csCfg.m_staticIdx );
maxNumBatches = 25;//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
{
BT_PROFILE("cpu batchContacts");
btAlignedObjectArray<btContact4> cpuContacts;
btOpenCLArray<btContact4>* contactsIn = m_data->m_pBufContactOutGPU;
btOpenCLArray<btContact4>* contactsIn = m_data->m_solverGPU->m_contactBuffer2;
contactsIn->copyToHost(cpuContacts);
btOpenCLArray<unsigned int>* countsNative = m_data->m_solverGPU->m_numConstraints;
@@ -611,8 +627,11 @@ void btGpuBatchingPgsSolver::solveContacts(int numBodies, cl_mem bodyBuf, cl_mem
numNonzeroGrid++;
//printf("cpu batch\n");
int simdWidth = 32;
int numBatches = sortConstraintByBatch( &cpuContacts[0]+offset, n, simdWidth,csCfg.m_staticIdx ,numBodies); // on GPU
int simdWidth =64;//-1;//32;
int numBatches = sortConstraintByBatch3( &cpuContacts[0]+offset, n, simdWidth,csCfg.m_staticIdx ,numBodies); // on GPU
maxNumBatches = btMax(numBatches,maxNumBatches);
clFinish(m_data->m_queue);
@@ -622,7 +641,7 @@ void btGpuBatchingPgsSolver::solveContacts(int numBodies, cl_mem bodyBuf, cl_mem
}
{
BT_PROFILE("m_contactBuffer->copyFromHost");
m_data->m_solverGPU->m_contactBuffer->copyFromHost((btAlignedObjectArray<btContact4>&)cpuContacts);
m_data->m_solverGPU->m_contactBuffer2->copyFromHost((btAlignedObjectArray<btContact4>&)cpuContacts);
}
}
@@ -636,7 +655,7 @@ void btGpuBatchingPgsSolver::solveContacts(int numBodies, cl_mem bodyBuf, cl_mem
{
//BT_PROFILE("gpu convertToConstraints");
m_data->m_solverGPU->convertToConstraints( bodyBuf,
shapeBuf, m_data->m_solverGPU->m_contactBuffer /*contactNative*/,
shapeBuf, m_data->m_solverGPU->m_contactBuffer2,
contactConstraintOut,
additionalData, nContacts,
(SolverBase::ConstraintCfg&) csCfg );
@@ -700,6 +719,13 @@ static bool sortfnc(const btSortData& a,const btSortData& b)
return (a.m_key<b.m_key);
}
btAlignedObjectArray<int> bodyUsed;
btAlignedObjectArray<unsigned int> idxBuffer;
btAlignedObjectArray<btSortData> sortData;
btAlignedObjectArray<btContact4> old;
@@ -830,3 +856,301 @@ inline int btGpuBatchingPgsSolver::sortConstraintByBatch( btContact4* cs, int n,
#endif
return batchIdx;
}
inline int btGpuBatchingPgsSolver::sortConstraintByBatch2( btContact4* cs, int numConstraints, int simdWidth , int staticIdx, int numBodies)
{
BT_PROFILE("sortConstraintByBatch");
bodyUsed.resize(2*simdWidth);
for (int q=0;q<2*simdWidth;q++)
bodyUsed[q]=0;
int curBodyUsed = 0;
int numIter = 0;
m_data->m_sortData.resize(numConstraints);
m_data->m_idxBuffer.resize(numConstraints);
m_data->m_old.resize(numConstraints);
unsigned int* idxSrc = &m_data->m_idxBuffer[0];
#if defined(_DEBUG)
for(int i=0; i<numConstraints; i++)
cs[i].getBatchIdx() = -1;
#endif
for(int i=0; i<numConstraints; i++)
idxSrc[i] = i;
int numValidConstraints = 0;
int unprocessedConstraintIndex = 0;
int batchIdx = 0;
{
BT_PROFILE("cpu batch innerloop");
while( numValidConstraints < numConstraints)
{
numIter++;
int nCurrentBatch = 0;
// clear flag
for(int i=0; i<curBodyUsed; i++)
bodyUsed[i] = 0;
curBodyUsed = 0;
for(int i=numValidConstraints; i<numConstraints; i++)
{
int idx = idxSrc[i];
btAssert( idx < numConstraints );
// check if it can go
int bodyAS = cs[idx].m_bodyAPtrAndSignBit;
int bodyBS = cs[idx].m_bodyBPtrAndSignBit;
int bodyA = abs(bodyAS);
int bodyB = abs(bodyBS);
bool aIsStatic = (bodyAS<0) || bodyAS==staticIdx;
bool bIsStatic = (bodyBS<0) || bodyBS==staticIdx;
int aUnavailable = 0;
int bUnavailable = 0;
if (!aIsStatic)
{
for (int j=0;j<curBodyUsed;j++)
{
if (bodyA == bodyUsed[j])
{
aUnavailable=1;
break;
}
}
}
if (!aUnavailable)
if (!bIsStatic)
{
for (int j=0;j<curBodyUsed;j++)
{
if (bodyB == bodyUsed[j])
{
bUnavailable=1;
break;
}
}
}
if( aUnavailable==0 && bUnavailable==0 ) // ok
{
if (!aIsStatic)
{
bodyUsed[curBodyUsed++] = bodyA;
}
if (!bIsStatic)
{
bodyUsed[curBodyUsed++] = bodyB;
}
cs[idx].getBatchIdx() = batchIdx;
m_data->m_sortData[idx].m_key = batchIdx;
m_data->m_sortData[idx].m_value = idx;
if (i!=numValidConstraints)
{
btSwap(idxSrc[i], idxSrc[numValidConstraints]);
}
numValidConstraints++;
{
nCurrentBatch++;
if( nCurrentBatch == simdWidth )
{
nCurrentBatch = 0;
for(int i=0; i<curBodyUsed; i++)
bodyUsed[i] = 0;
curBodyUsed = 0;
}
}
}
}
batchIdx ++;
}
}
{
BT_PROFILE("quickSort");
//m_data->m_sortData.quickSort(sortfnc);
}
{
BT_PROFILE("reorder");
// reorder
memcpy( &m_data->m_old[0], cs, sizeof(btContact4)*numConstraints);
for(int i=0; i<numConstraints; i++)
{
btAssert(m_data->m_sortData[idxSrc[i]].m_value == idxSrc[i]);
int idx = m_data->m_sortData[idxSrc[i]].m_value;
cs[i] = m_data->m_old[idx];
}
}
#if defined(_DEBUG)
// debugPrintf( "nBatches: %d\n", batchIdx );
for(int i=0; i<numConstraints; i++)
{
btAssert( cs[i].getBatchIdx() != -1 );
}
#endif
return batchIdx;
}
inline int btGpuBatchingPgsSolver::sortConstraintByBatch3( btContact4* cs, int numConstraints, int simdWidth , int staticIdx, int numBodies)
{
BT_PROFILE("sortConstraintByBatch");
static int maxSwaps = 0;
int numSwaps = 0;
static int maxNumConstraints = 0;
if (maxNumConstraints<numConstraints)
{
maxNumConstraints = numConstraints;
printf("maxNumConstraints = %d\n",maxNumConstraints );
}
bodyUsed.resize(2*simdWidth);
for (int q=0;q<2*simdWidth;q++)
bodyUsed[q]=0;
int curBodyUsed = 0;
int numIter = 0;
m_data->m_sortData.resize(0);
m_data->m_idxBuffer.resize(0);
m_data->m_old.resize(0);
#if defined(_DEBUG)
for(int i=0; i<numConstraints; i++)
cs[i].getBatchIdx() = -1;
#endif
int numValidConstraints = 0;
int unprocessedConstraintIndex = 0;
int batchIdx = 0;
{
BT_PROFILE("cpu batch innerloop");
while( numValidConstraints < numConstraints)
{
numIter++;
int nCurrentBatch = 0;
// clear flag
for(int i=0; i<curBodyUsed; i++)
bodyUsed[i] = 0;
curBodyUsed = 0;
for(int i=numValidConstraints; i<numConstraints; i++)
{
int idx = i;
btAssert( idx < numConstraints );
// check if it can go
int bodyAS = cs[idx].m_bodyAPtrAndSignBit;
int bodyBS = cs[idx].m_bodyBPtrAndSignBit;
int bodyA = abs(bodyAS);
int bodyB = abs(bodyBS);
bool aIsStatic = (bodyAS<0) || bodyAS==staticIdx;
bool bIsStatic = (bodyBS<0) || bodyBS==staticIdx;
int aUnavailable = 0;
int bUnavailable = 0;
if (!aIsStatic)
{
for (int j=0;j<curBodyUsed;j++)
{
if (bodyA == bodyUsed[j])
{
aUnavailable=1;
break;
}
}
}
if (!aUnavailable)
if (!bIsStatic)
{
for (int j=0;j<curBodyUsed;j++)
{
if (bodyB == bodyUsed[j])
{
bUnavailable=1;
break;
}
}
}
if( aUnavailable==0 && bUnavailable==0 ) // ok
{
if (!aIsStatic)
{
bodyUsed[curBodyUsed++] = bodyA;
}
if (!bIsStatic)
{
bodyUsed[curBodyUsed++] = bodyB;
}
cs[idx].getBatchIdx() = batchIdx;
if (i!=numValidConstraints)
{
btSwap(cs[i],cs[numValidConstraints]);
numSwaps++;
}
numValidConstraints++;
{
nCurrentBatch++;
if( nCurrentBatch == simdWidth )
{
nCurrentBatch = 0;
for(int i=0; i<curBodyUsed; i++)
bodyUsed[i] = 0;
curBodyUsed = 0;
}
}
}
}
batchIdx ++;
}
}
#if defined(_DEBUG)
// debugPrintf( "nBatches: %d\n", batchIdx );
for(int i=0; i<numConstraints; i++)
{
btAssert( cs[i].getBatchIdx() != -1 );
}
#endif
if (maxSwaps<numSwaps)
{
maxSwaps = numSwaps;
printf("maxSwaps = %d\n", maxSwaps);
}
return batchIdx;
}

8
opencl/gpu_rigidbody/host/btGpuBatchingPgsSolver.h
View File

@@ -12,10 +12,18 @@ class btGpuBatchingPgsSolver
{
protected:
struct btGpuBatchingPgsSolverInternalData* m_data;
void batchContacts( btOpenCLArray<btContact4>* contacts, int nContacts, btOpenCLArray<unsigned int>* n, btOpenCLArray<unsigned int>* offsets, int staticIdx );
inline int sortConstraintByBatch( btContact4* cs, int n, int simdWidth , int staticIdx, int numBodies);
inline int sortConstraintByBatch2( btContact4* cs, int n, int simdWidth , int staticIdx, int numBodies);
inline int sortConstraintByBatch3( btContact4* cs, int n, int simdWidth , int staticIdx, int numBodies);
void solveContactConstraint( const btOpenCLArray<btRigidBodyCL>* bodyBuf, const btOpenCLArray<btInertiaCL>* shapeBuf,
btOpenCLArray<btGpuConstraint4>* constraint, void* additionalData, int n ,int maxNumBatches, int numIterations);

9
opencl/gpu_rigidbody/host/btGpuRigidBodyPipeline.cpp
View File

@@ -63,10 +63,15 @@ void btGpuRigidBodyPipeline::stepSimulation(float deltaTime)
{
//update worldspace AABBs from local AABB/worldtransform
setupGpuAabbsFull();
{
setupGpuAabbsFull();
}
//compute overlapping pairs
m_data->m_broadphaseSap->calculateOverlappingPairs();
{
//m_data->m_broadphaseSap->calculateOverlappingPairsHost();
m_data->m_broadphaseSap->calculateOverlappingPairs();
}
//compute contact points

236
opencl/gpu_rigidbody/kernels/batchingKernelsNew.cl Normal file
View File

@@ -0,0 +1,236 @@
/*
Copyright (c) 2012 Advanced Micro Devices, Inc.
This software is provided 'as-is', without any express or implied warranty.
In no event will the authors be held liable for any damages arising from the use of this software.
Permission is granted to anyone to use this software for any purpose,
including commercial applications, and to alter it and redistribute it freely,
subject to the following restrictions:
1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
3. This notice may not be removed or altered from any source distribution.
*/
//Originally written by Erwin Coumans
#pragma OPENCL EXTENSION cl_amd_printf : enable
#pragma OPENCL EXTENSION cl_khr_local_int32_base_atomics : enable
#pragma OPENCL EXTENSION cl_khr_global_int32_base_atomics : enable
#pragma OPENCL EXTENSION cl_khr_local_int32_extended_atomics : enable
#pragma OPENCL EXTENSION cl_khr_global_int32_extended_atomics : enable
#ifdef cl_ext_atomic_counters_32
#pragma OPENCL EXTENSION cl_ext_atomic_counters_32 : enable
#else
#define counter32_t volatile __global int*
#endif
#define SIMD_WIDTH 64
typedef unsigned int u32;
typedef unsigned short u16;
typedef unsigned char u8;
#define GET_GROUP_IDX get_group_id(0)
#define GET_LOCAL_IDX get_local_id(0)
#define GET_GLOBAL_IDX get_global_id(0)
#define GET_GROUP_SIZE get_local_size(0)
#define GET_NUM_GROUPS get_num_groups(0)
#define GROUP_LDS_BARRIER barrier(CLK_LOCAL_MEM_FENCE)
#define GROUP_MEM_FENCE mem_fence(CLK_LOCAL_MEM_FENCE)
#define AtomInc(x) atom_inc(&(x))
#define AtomInc1(x, out) out = atom_inc(&(x))
#define AppendInc(x, out) out = atomic_inc(x)
#define AtomAdd(x, value) atom_add(&(x), value)
#define AtomCmpxhg(x, cmp, value) atom_cmpxchg( &(x), cmp, value )
#define AtomXhg(x, value) atom_xchg ( &(x), value )
#define SELECT_UINT4( b, a, condition ) select( b,a,condition )
#define make_float4 (float4)
#define make_float2 (float2)
#define make_uint4 (uint4)
#define make_int4 (int4)
#define make_uint2 (uint2)
#define make_int2 (int2)
#define max2 max
#define min2 min
#define WG_SIZE 64
typedef struct
{
float4 m_worldPos[4];
float4 m_worldNormal;
u32 m_coeffs;
int m_batchIdx;
int m_bodyAPtrAndSignBit;//sign bit set for fixed objects
int m_bodyBPtrAndSignBit;
}Contact4;
typedef struct
{
int m_n;
int m_start;
int m_staticIdx;
int m_paddings[1];
} ConstBuffer;
typedef struct
{
int m_a;
int m_b;
u32 m_idx;
}Elem;
// batching on the GPU
__kernel void CreateBatchesBruteForce( __global Contact4* gConstraints, __global const u32* gN, __global const u32* gStart, int m_staticIdx )
{
int wgIdx = GET_GROUP_IDX;
int lIdx = GET_LOCAL_IDX;
const int m_n = gN[wgIdx];
const int m_start = gStart[wgIdx];
if( lIdx == 0 )
{
for (int i=0;i<m_n;i++)
{
int srcIdx = i+m_start;
int batchIndex = i;
gConstraints[ srcIdx ].m_batchIdx = batchIndex;
}
}
}
#define CHECK_SIZE (WG_SIZE)
u32 readBuf(__local u32* buff, int idx)
{
idx = idx % (32*CHECK_SIZE);
int bitIdx = idx%32;
int bufIdx = idx/32;
return buff[bufIdx] & (1<<bitIdx);
}
void writeBuf(__local u32* buff, int idx)
{
idx = idx % (32*CHECK_SIZE);
int bitIdx = idx%32;
int bufIdx = idx/32;
buff[bufIdx] |= (1<<bitIdx);
//atom_or( &buff[bufIdx], (1<<bitIdx) );
}
u32 tryWrite(__local u32* buff, int idx)
{
idx = idx % (32*CHECK_SIZE);
int bitIdx = idx%32;
int bufIdx = idx/32;
u32 ans = (u32)atom_or( &buff[bufIdx], (1<<bitIdx) );
return ((ans >> bitIdx)&1) == 0;
}
// batching on the GPU
__kernel void CreateBatchesNew( __global Contact4* gConstraints, __global const u32* gN, __global const u32* gStart, int staticIdx )
{
int wgIdx = GET_GROUP_IDX;
int lIdx = GET_LOCAL_IDX;
const int numConstraints = gN[wgIdx];
const int m_start = gStart[wgIdx];
__local u32 ldsFixedBuffer[CHECK_SIZE];
if( lIdx == 0 )
{
__global Contact4* cs = &gConstraints[m_start];
int numValidConstraints = 0;
int batchIdx = 0;
while( numValidConstraints < numConstraints)
{
int nCurrentBatch = 0;
// clear flag
for(int i=0; i<CHECK_SIZE; i++)
ldsFixedBuffer[i] = 0;
for(int i=numValidConstraints; i<numConstraints; i++)
{
int bodyAS = cs[i].m_bodyAPtrAndSignBit;
int bodyBS = cs[i].m_bodyBPtrAndSignBit;
int bodyA = abs(bodyAS);
int bodyB = abs(bodyBS);
bool aIsStatic = (bodyAS<0) || bodyAS==staticIdx;
bool bIsStatic = (bodyBS<0) || bodyBS==staticIdx;
int aUnavailable = aIsStatic ? 0 : readBuf( ldsFixedBuffer, bodyA);
int bUnavailable = bIsStatic ? 0 : readBuf( ldsFixedBuffer, bodyB);
if( aUnavailable==0 && bUnavailable==0 ) // ok
{
if (!aIsStatic)
{
writeBuf( ldsFixedBuffer, bodyA );
}
if (!bIsStatic)
{
writeBuf( ldsFixedBuffer, bodyB );
}
cs[i].m_batchIdx = batchIdx;
if (i!=numValidConstraints)
{
//btSwap(cs[i],cs[numValidConstraints]);
Contact4 tmp = cs[i];
cs[i] = cs[numValidConstraints];
cs[numValidConstraints] = tmp;
}
numValidConstraints++;
nCurrentBatch++;
if( nCurrentBatch == SIMD_WIDTH)
{
nCurrentBatch = 0;
for(int i=0; i<CHECK_SIZE; i++)
ldsFixedBuffer[i] = 0;
}
}
}//for
batchIdx ++;
}//while
}//if( lIdx == 0 )
//return batchIdx;
}

240
opencl/gpu_rigidbody/kernels/batchingKernelsNew.h Normal file
View File

@@ -0,0 +1,240 @@
//this file is autogenerated using stringify.bat (premake --stringify) in the build folder of this project
static const char* batchingKernelsNewCL= \
"/*\n"
"Copyright (c) 2012 Advanced Micro Devices, Inc. \n"
"\n"
"This software is provided 'as-is', without any express or implied warranty.\n"
"In no event will the authors be held liable for any damages arising from the use of this software.\n"
"Permission is granted to anyone to use this software for any purpose, \n"
"including commercial applications, and to alter it and redistribute it freely, \n"
"subject to the following restrictions:\n"
"\n"
"1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.\n"
"2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.\n"
"3. This notice may not be removed or altered from any source distribution.\n"
"*/\n"
"//Originally written by Erwin Coumans\n"
"\n"
"\n"
"#pragma OPENCL EXTENSION cl_amd_printf : enable\n"
"#pragma OPENCL EXTENSION cl_khr_local_int32_base_atomics : enable\n"
"#pragma OPENCL EXTENSION cl_khr_global_int32_base_atomics : enable\n"
"#pragma OPENCL EXTENSION cl_khr_local_int32_extended_atomics : enable\n"
"#pragma OPENCL EXTENSION cl_khr_global_int32_extended_atomics : enable\n"
"\n"
"#ifdef cl_ext_atomic_counters_32\n"
"#pragma OPENCL EXTENSION cl_ext_atomic_counters_32 : enable\n"
"#else\n"
"#define counter32_t volatile __global int*\n"
"#endif\n"
"\n"
"#define SIMD_WIDTH 64\n"
"\n"
"typedef unsigned int u32;\n"
"typedef unsigned short u16;\n"
"typedef unsigned char u8;\n"
"\n"
"#define GET_GROUP_IDX get_group_id(0)\n"
"#define GET_LOCAL_IDX get_local_id(0)\n"
"#define GET_GLOBAL_IDX get_global_id(0)\n"
"#define GET_GROUP_SIZE get_local_size(0)\n"
"#define GET_NUM_GROUPS get_num_groups(0)\n"
"#define GROUP_LDS_BARRIER barrier(CLK_LOCAL_MEM_FENCE)\n"
"#define GROUP_MEM_FENCE mem_fence(CLK_LOCAL_MEM_FENCE)\n"
"#define AtomInc(x) atom_inc(&(x))\n"
"#define AtomInc1(x, out) out = atom_inc(&(x))\n"
"#define AppendInc(x, out) out = atomic_inc(x)\n"
"#define AtomAdd(x, value) atom_add(&(x), value)\n"
"#define AtomCmpxhg(x, cmp, value) atom_cmpxchg( &(x), cmp, value )\n"
"#define AtomXhg(x, value) atom_xchg ( &(x), value )\n"
"\n"
"\n"
"#define SELECT_UINT4( b, a, condition ) select( b,a,condition )\n"
"\n"
"#define make_float4 (float4)\n"
"#define make_float2 (float2)\n"
"#define make_uint4 (uint4)\n"
"#define make_int4 (int4)\n"
"#define make_uint2 (uint2)\n"
"#define make_int2 (int2)\n"
"\n"
"\n"
"#define max2 max\n"
"#define min2 min\n"
"\n"
"\n"
"#define WG_SIZE 64\n"
"\n"
"\n"
"\n"
"typedef struct \n"
"{\n"
" float4 m_worldPos[4];\n"
" float4 m_worldNormal;\n"
" u32 m_coeffs;\n"
" int m_batchIdx;\n"
"\n"
" int m_bodyAPtrAndSignBit;//sign bit set for fixed objects\n"
" int m_bodyBPtrAndSignBit;\n"
"}Contact4;\n"
"\n"
"typedef struct \n"
"{\n"
" int m_n;\n"
" int m_start;\n"
" int m_staticIdx;\n"
" int m_paddings[1];\n"
"} ConstBuffer;\n"
"\n"
"typedef struct \n"
"{\n"
" int m_a;\n"
" int m_b;\n"
" u32 m_idx;\n"
"}Elem;\n"
"\n"
"\n"
"\n"
"\n"
"\n"
"// batching on the GPU\n"
"__kernel void CreateBatchesBruteForce( __global Contact4* gConstraints, __global const u32* gN, __global const u32* gStart, int m_staticIdx )\n"
"{\n"
" int wgIdx = GET_GROUP_IDX;\n"
" int lIdx = GET_LOCAL_IDX;\n"
" \n"
" const int m_n = gN[wgIdx];\n"
" const int m_start = gStart[wgIdx];\n"
" \n"
" if( lIdx == 0 )\n"
" {\n"
" for (int i=0;i<m_n;i++)\n"
" {\n"
" int srcIdx = i+m_start;\n"
" int batchIndex = i;\n"
" gConstraints[ srcIdx ].m_batchIdx = batchIndex; \n"
" }\n"
" }\n"
"}\n"
"\n"
"\n"
"#define CHECK_SIZE (WG_SIZE)\n"
"\n"
"\n"
"\n"
"\n"
"u32 readBuf(__local u32* buff, int idx)\n"
"{\n"
" idx = idx % (32*CHECK_SIZE);\n"
" int bitIdx = idx%32;\n"
" int bufIdx = idx/32;\n"
" return buff[bufIdx] & (1<<bitIdx);\n"
"}\n"
"\n"
"void writeBuf(__local u32* buff, int idx)\n"
"{\n"
" idx = idx % (32*CHECK_SIZE);\n"
" int bitIdx = idx%32;\n"
" int bufIdx = idx/32;\n"
" buff[bufIdx] |= (1<<bitIdx);\n"
" //atom_or( &buff[bufIdx], (1<<bitIdx) );\n"
"}\n"
"\n"
"u32 tryWrite(__local u32* buff, int idx)\n"
"{\n"
" idx = idx % (32*CHECK_SIZE);\n"
" int bitIdx = idx%32;\n"
" int bufIdx = idx/32;\n"
" u32 ans = (u32)atom_or( &buff[bufIdx], (1<<bitIdx) );\n"
" return ((ans >> bitIdx)&1) == 0;\n"
"}\n"
"\n"
"\n"
"// batching on the GPU\n"
"__kernel void CreateBatchesNew( __global Contact4* gConstraints, __global const u32* gN, __global const u32* gStart, int staticIdx )\n"
"{\n"
" int wgIdx = GET_GROUP_IDX;\n"
" int lIdx = GET_LOCAL_IDX;\n"
" const int numConstraints = gN[wgIdx];\n"
" const int m_start = gStart[wgIdx];\n"
" \n"
" \n"
" __local u32 ldsFixedBuffer[CHECK_SIZE];\n"
" \n"
" \n"
" \n"
" \n"
" \n"
" if( lIdx == 0 )\n"
" {\n"
" \n"
" \n"
" __global Contact4* cs = &gConstraints[m_start]; \n"
" \n"
" \n"
" int numValidConstraints = 0;\n"
" int batchIdx = 0;\n"
"\n"
" while( numValidConstraints < numConstraints)\n"
" {\n"
" int nCurrentBatch = 0;\n"
" // clear flag\n"
" \n"
" for(int i=0; i<CHECK_SIZE; i++) \n"
" ldsFixedBuffer[i] = 0; \n"
"\n"
" for(int i=numValidConstraints; i<numConstraints; i++)\n"
" {\n"
"\n"
" int bodyAS = cs[i].m_bodyAPtrAndSignBit;\n"
" int bodyBS = cs[i].m_bodyBPtrAndSignBit;\n"
" int bodyA = abs(bodyAS);\n"
" int bodyB = abs(bodyBS);\n"
" bool aIsStatic = (bodyAS<0) || bodyAS==staticIdx;\n"
" bool bIsStatic = (bodyBS<0) || bodyBS==staticIdx;\n"
" int aUnavailable = aIsStatic ? 0 : readBuf( ldsFixedBuffer, bodyA);\n"
" int bUnavailable = bIsStatic ? 0 : readBuf( ldsFixedBuffer, bodyB);\n"
" \n"
" if( aUnavailable==0 && bUnavailable==0 ) // ok\n"
" {\n"
" if (!aIsStatic)\n"
" {\n"
" writeBuf( ldsFixedBuffer, bodyA );\n"
" }\n"
" if (!bIsStatic)\n"
" {\n"
" writeBuf( ldsFixedBuffer, bodyB );\n"
" }\n"
"\n"
" cs[i].m_batchIdx = batchIdx;\n"
"\n"
" if (i!=numValidConstraints)\n"
" {\n"
" //btSwap(cs[i],cs[numValidConstraints]);\n"
" \n"
" Contact4 tmp = cs[i];\n"
" cs[i] = cs[numValidConstraints];\n"
" cs[numValidConstraints] = tmp;\n"
" \n"
" }\n"
"\n"
" numValidConstraints++;\n"
" \n"
" nCurrentBatch++;\n"
" if( nCurrentBatch == SIMD_WIDTH)\n"
" {\n"
" nCurrentBatch = 0;\n"
" for(int i=0; i<CHECK_SIZE; i++) \n"
" ldsFixedBuffer[i] = 0;\n"
" \n"
" }\n"
" }\n"
" }//for\n"
" batchIdx ++;\n"
" }//while\n"
" }//if( lIdx == 0 )\n"
" \n"
" //return batchIdx;\n"
"}\n"
"\n"
;