void btFillCL::execute(btOpenCLArray<btInt2> &src, const btInt2 &value, int n, int offset)
{
btAssert( n>0 );
btConstData constBuffer;
{
constBuffer.m_offset = offset;
constBuffer.m_n = n;
constBuffer.m_data = btMakeInt4( value.x, value.y, 0, 0 );
}
{
btBufferInfoCL bInfo[] = { btBufferInfoCL( src.getBufferCL() ) };
btLauncherCL launcher(m_commandQueue, m_fillKernelInt2);
launcher.setBuffers( bInfo, sizeof(bInfo)/sizeof(btBufferInfoCL) );
launcher.setConst(n);
launcher.setConst(value);
launcher.setConst(offset);
//( constBuffer );
launcher.launch1D( n );
}
}
void Solver::solveContactConstraint( const btOpenCLArray<RigidBodyBase::Body>* bodyBuf, const btOpenCLArray<RigidBodyBase::Inertia>* shapeBuf,
btOpenCLArray<Constraint4>* constraint, void* additionalData, int n ,int maxNumBatches)
{
btInt4 cdata = btMakeInt4( 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::btOpenCLArray<SolverDebugInfo> gpuDebugInfo(data->m_device,numWorkItems);
#endif
{
BT_PROFILE("m_batchSolveKernel iterations");
for(int iter=0; iter<m_nIterations; iter++)
{
for(int ib=0; ib<N_BATCHES; ib++)
{
#ifdef DEBUG_ME
memset(debugInfo,0,sizeof(SolverDebugInfo)*numWorkItems);
gpuDebugInfo.write(debugInfo,numWorkItems);
#endif
cdata.z = ib;
cdata.w = N_SPLIT;
btLauncherCL launcher( m_queue, m_solveContactKernel );
#if 1
btBufferInfoCL bInfo[] = {
btBufferInfoCL( bodyBuf->getBufferCL() ),
btBufferInfoCL( shapeBuf->getBufferCL() ),
btBufferInfoCL( constraint->getBufferCL() ),
btBufferInfoCL( m_numConstraints->getBufferCL() ),
btBufferInfoCL( m_offsets->getBufferCL() )
#ifdef DEBUG_ME
, btBufferInfoCL(&gpuDebugInfo)
#endif
};
launcher.setBuffers( bInfo, sizeof(bInfo)/sizeof(btBufferInfoCL) );
//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)
{
BT_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;
btBufferInfoCL bInfo[] = {
btBufferInfoCL( bodyBuf->getBufferCL() ),
btBufferInfoCL( shapeBuf->getBufferCL() ),
btBufferInfoCL( constraint->getBufferCL() ),
btBufferInfoCL( m_numConstraints->getBufferCL() ),
btBufferInfoCL( m_offsets->getBufferCL() )
#ifdef DEBUG_ME
,btBufferInfoCL(&gpuDebugInfo)
#endif //DEBUG_ME
};
btLauncherCL launcher( m_queue, m_solveFrictionKernel );
launcher.setBuffers( bInfo, sizeof(bInfo)/sizeof(btBufferInfoCL) );
//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
}
}
