void b3Solver::solveContactConstraint( const b3OpenCLArray<b3RigidBodyCL>* bodyBuf, const b3OpenCLArray<b3InertiaCL>* shapeBuf,
b3OpenCLArray<b3GpuConstraint4>* constraint, void* additionalData, int n ,int maxNumBatches)
{
b3Int4 cdata = b3MakeInt4( n, 0, 0, 0 );
{
const int nn = N_SPLIT*N_SPLIT;
cdata.x = 0;
cdata.y = maxNumBatches;//250;
int numWorkItems = 64*nn/N_BATCHES;
#ifdef DEBUG_ME
SolverDebugInfo* debugInfo = new SolverDebugInfo[numWorkItems];
adl::b3OpenCLArray<SolverDebugInfo> gpuDebugInfo(data->m_device,numWorkItems);
#endif
{
B3_PROFILE("m_batchSolveKernel iterations");
for(int iter=0; iter<m_nIterations; iter++)
{
for(int ib=0; ib<N_BATCHES; ib++)
{
if (verify)
{
checkConstraintBatch(bodyBuf,shapeBuf,constraint,m_numConstraints,m_offsets,ib);
}
#ifdef DEBUG_ME
memset(debugInfo,0,sizeof(SolverDebugInfo)*numWorkItems);
gpuDebugInfo.write(debugInfo,numWorkItems);
#endif
cdata.z = ib;
cdata.w = N_SPLIT;
b3LauncherCL launcher( m_queue, m_solveContactKernel );
#if 1
b3BufferInfoCL bInfo[] = {
b3BufferInfoCL( bodyBuf->getBufferCL() ),
b3BufferInfoCL( shapeBuf->getBufferCL() ),
b3BufferInfoCL( constraint->getBufferCL() ),
b3BufferInfoCL( m_numConstraints->getBufferCL() ),
b3BufferInfoCL( m_offsets->getBufferCL() )
#ifdef DEBUG_ME
, b3BufferInfoCL(&gpuDebugInfo)
#endif
};
launcher.setBuffers( bInfo, sizeof(bInfo)/sizeof(b3BufferInfoCL) );
//launcher.setConst( cdata.x );
launcher.setConst( cdata.y );
launcher.setConst( cdata.z );
launcher.setConst( cdata.w );
launcher.launch1D( numWorkItems, 64 );
#else
const char* fileName = "m_batchSolveKernel.bin";
FILE* f = fopen(fileName,"rb");
if (f)
{
int sizeInBytes=0;
if (fseek(f, 0, SEEK_END) || (sizeInBytes = ftell(f)) == EOF || fseek(f, 0, SEEK_SET))
{
printf("error, cannot get file size\n");
exit(0);
}
unsigned char* buf = (unsigned char*) malloc(sizeInBytes);
fread(buf,sizeInBytes,1,f);
int serializedBytes = launcher.deserializeArgs(buf, sizeInBytes,m_context);
int num = *(int*)&buf[serializedBytes];
launcher.launch1D( num);
//this clFinish is for testing on errors
clFinish(m_queue);
}
#endif
#ifdef DEBUG_ME
clFinish(m_queue);
gpuDebugInfo.read(debugInfo,numWorkItems);
clFinish(m_queue);
for (int i=0; i<numWorkItems; i++)
{
if (debugInfo[i].m_valInt2>0)
{
printf("debugInfo[i].m_valInt2 = %d\n",i,debugInfo[i].m_valInt2);
}
if (debugInfo[i].m_valInt3>0)
{
printf("debugInfo[i].m_valInt3 = %d\n",i,debugInfo[i].m_valInt3);
}
}
#endif //DEBUG_ME
}
}
clFinish(m_queue);
}
cdata.x = 1;
bool applyFriction=true;
if (applyFriction)
{
B3_PROFILE("m_batchSolveKernel iterations2");
for(int iter=0; iter<m_nIterations; iter++)
{
for(int ib=0; ib<N_BATCHES; ib++)
{
cdata.z = ib;
cdata.w = N_SPLIT;
b3BufferInfoCL bInfo[] = {
b3BufferInfoCL( bodyBuf->getBufferCL() ),
b3BufferInfoCL( shapeBuf->getBufferCL() ),
b3BufferInfoCL( constraint->getBufferCL() ),
b3BufferInfoCL( m_numConstraints->getBufferCL() ),
b3BufferInfoCL( m_offsets->getBufferCL() )
#ifdef DEBUG_ME
,b3BufferInfoCL(&gpuDebugInfo)
#endif //DEBUG_ME
};
b3LauncherCL launcher( m_queue, m_solveFrictionKernel );
launcher.setBuffers( bInfo, sizeof(bInfo)/sizeof(b3BufferInfoCL) );
//launcher.setConst( cdata.x );
launcher.setConst( cdata.y );
launcher.setConst( cdata.z );
launcher.setConst( cdata.w );
launcher.launch1D( 64*nn/N_BATCHES, 64 );
}
}
clFinish(m_queue);
}
#ifdef DEBUG_ME
delete[] debugInfo;
#endif //DEBUG_ME
}
}
void b3Solver::batchContacts( b3OpenCLArray<b3Contact4>* contacts, int nContacts, b3OpenCLArray<unsigned int>* nNative, b3OpenCLArray<unsigned int>* offsetsNative, int staticIdx )
{
int numWorkItems = 64*N_SPLIT*N_SPLIT;
{
B3_PROFILE("batch generation");
b3Int4 cdata;
cdata.x = nContacts;
cdata.y = 0;
cdata.z = staticIdx;
#ifdef BATCH_DEBUG
SolverDebugInfo* debugInfo = new SolverDebugInfo[numWorkItems];
adl::b3OpenCLArray<SolverDebugInfo> gpuDebugInfo(data->m_device,numWorkItems);
memset(debugInfo,0,sizeof(SolverDebugInfo)*numWorkItems);
gpuDebugInfo.write(debugInfo,numWorkItems);
#endif
b3BufferInfoCL bInfo[] = {
b3BufferInfoCL( contacts->getBufferCL() ),
b3BufferInfoCL( m_contactBuffer2->getBufferCL()),
b3BufferInfoCL( nNative->getBufferCL() ),
b3BufferInfoCL( offsetsNative->getBufferCL() ),
#ifdef BATCH_DEBUG
, b3BufferInfoCL(&gpuDebugInfo)
#endif
};
{
B3_PROFILE("batchingKernel");
//b3LauncherCL launcher( m_queue, m_batchingKernel);
cl_kernel k = useNewBatchingKernel ? m_batchingKernelNew : m_batchingKernel;
b3LauncherCL 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);
launcher.launch1D( numWorkItems, 64 );
clFinish(m_queue);
}
#ifdef BATCH_DEBUG
aaaa
b3Contact4* hostContacts = new b3Contact4[nContacts];
m_contactBuffer->read(hostContacts,nContacts);
clFinish(m_queue);
gpuDebugInfo.read(debugInfo,numWorkItems);
clFinish(m_queue);
for (int i=0; i<numWorkItems; i++)
{
if (debugInfo[i].m_valInt1>0)
{
printf("catch\n");
}
if (debugInfo[i].m_valInt2>0)
{
printf("catch22\n");
}
if (debugInfo[i].m_valInt3>0)
{
printf("catch666\n");
}
if (debugInfo[i].m_valInt4>0)
{
printf("catch777\n");
}
}
delete[] debugInfo;
#endif //BATCH_DEBUG
}
// copy buffer to buffer
//b3Assert(m_contactBuffer->size()==nContacts);
//contacts->copyFromOpenCLArray( *m_contactBuffer);
//clFinish(m_queue);//needed?
}
void Solver::batchContacts( btOpenCLArray<Contact4>* contacts, int nContacts, btOpenCLArray<u32>* nNative, btOpenCLArray<u32>* offsetsNative, int staticIdx )
{
{
BT_PROFILE("batch generation");
btInt4 cdata;
cdata.x = nContacts;
cdata.y = 0;
cdata.z = staticIdx;
int numWorkItems = 64*N_SPLIT*N_SPLIT;
#ifdef BATCH_DEBUG
SolverDebugInfo* debugInfo = new SolverDebugInfo[numWorkItems];
adl::btOpenCLArray<SolverDebugInfo> gpuDebugInfo(data->m_device,numWorkItems);
memset(debugInfo,0,sizeof(SolverDebugInfo)*numWorkItems);
gpuDebugInfo.write(debugInfo,numWorkItems);
#endif
btBufferInfoCL bInfo[] = {
btBufferInfoCL( contacts->getBufferCL() ),
btBufferInfoCL( m_contactBuffer->getBufferCL() ),
btBufferInfoCL( nNative->getBufferCL() ),
btBufferInfoCL( offsetsNative->getBufferCL() )
#ifdef BATCH_DEBUG
, btBufferInfoCL(&gpuDebugInfo)
#endif
};
{
BT_PROFILE("batchingKernel");
btLauncherCL launcher( m_queue, m_batchingKernel);
launcher.setBuffers( bInfo, sizeof(bInfo)/sizeof(btBufferInfoCL) );
//launcher.setConst( cdata );
launcher.setConst(staticIdx);
launcher.launch1D( numWorkItems, 64 );
clFinish(m_queue);
}
#ifdef BATCH_DEBUG
aaaa
Contact4* hostContacts = new Contact4[nContacts];
m_contactBuffer->read(hostContacts,nContacts);
clFinish(m_queue);
gpuDebugInfo.read(debugInfo,numWorkItems);
clFinish(m_queue);
for (int i=0; i<numWorkItems; i++)
{
if (debugInfo[i].m_valInt1>0)
{
printf("catch\n");
}
if (debugInfo[i].m_valInt2>0)
{
printf("catch22\n");
}
if (debugInfo[i].m_valInt3>0)
{
printf("catch666\n");
}
if (debugInfo[i].m_valInt4>0)
{
printf("catch777\n");
}
}
delete[] debugInfo;
#endif //BATCH_DEBUG
}
// copy buffer to buffer
btAssert(m_contactBuffer->size()==nContacts);
//contacts->copyFromOpenCLArray( *m_contactBuffer);
//clFinish(m_queue);//needed?
}