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

more debugging to fix Mac OSX OpenCL issue on certain GPUs.

Browse Source
This commit is contained in:
erwincoumans
2013-12-16 10:53:38 -08:00
parent dca5bbb744
commit fc474a0e62
1 changed files with 146 additions and 48 deletions
Download Patch File Download Diff File Expand all files Collapse all files

108
src/Bullet3OpenCL/NarrowphaseCollision/b3ConvexHullContact.cpp
View File

@@ -17,8 +17,10 @@ bool findSeparatingAxisOnGpu = true;
bool bvhTraversalKernelGPU = true; bool bvhTraversalKernelGPU = true;
bool findConcaveSeparatingAxisKernelGPU = true; bool findConcaveSeparatingAxisKernelGPU = true;
bool clipFacesAndFindContactsCPU = false;//true; bool clipConcaveFacesAndFindContactsCPU = false;//false;//true;
bool reduceContactsOnGPU = true;//false; bool clipConvexFacesAndFindContactsCPU = false;//false;//true;
bool reduceConcaveContactsOnGPU = true;//false;
bool reduceConvexContactsOnGPU = true;//false;
///This file was written by Erwin Coumans ///This file was written by Erwin Coumans
@@ -3583,7 +3585,7 @@ void GpuSatCollision::computeConvexConvexContactsGPUSAT( b3OpenCLArray<b3Int4>*
//clipFacesAndFindContacts //clipFacesAndFindContacts
if (clipFacesAndFindContactsCPU) if (clipConcaveFacesAndFindContactsCPU)
{ {
b3AlignedObjectArray<b3Int4> clippingFacesOutCPU; b3AlignedObjectArray<b3Int4> clippingFacesOutCPU;
@@ -3664,7 +3666,7 @@ void GpuSatCollision::computeConvexConvexContactsGPUSAT( b3OpenCLArray<b3Int4>*
{ {
int newContactCapacity=nContacts+numConcavePairs; int newContactCapacity=nContacts+numConcavePairs;
contactOut->reserve(newContactCapacity); contactOut->reserve(newContactCapacity);
if (reduceContactsOnGPU) if (reduceConcaveContactsOnGPU)
{ {
// printf("newReservation = %d\n",newReservation); // printf("newReservation = %d\n",newReservation);
{ {
@@ -3858,6 +3860,58 @@ void GpuSatCollision::computeConvexConvexContactsGPUSAT( b3OpenCLArray<b3Int4>*
///clip face B against face A, reduce contacts and append them to a global contact array ///clip face B against face A, reduce contacts and append them to a global contact array
if (1) if (1)
{
if (clipConvexFacesAndFindContactsCPU)
{
b3AlignedObjectArray<b3Vector3> hostSepNormals;
m_sepNormals.copyToHost(hostSepNormals);
b3AlignedObjectArray<int> hostHasSepAxis;
m_hasSeparatingNormals.copyToHost(hostHasSepAxis);
b3AlignedObjectArray<b3Int4> hostClippingFaces;
clippingFacesOutGPU.copyToHost(hostClippingFaces);
b3AlignedObjectArray<b3Vector3> worldVertsB2CPU;
worldVertsB2CPU.resize(vertexFaceCapacity*nPairs);
b3AlignedObjectArray<b3Vector3>worldVertsA1CPU;
worldVertsA1GPU.copyToHost(worldVertsA1CPU);
b3AlignedObjectArray<b3Vector3> worldNormalsACPU;
worldNormalsAGPU.copyToHost(worldNormalsACPU);
b3AlignedObjectArray<b3Vector3> worldVertsB1CPU;
worldVertsB1GPU.copyToHost(worldVertsB1CPU);
/*
__global const b3Float4* separatingNormals,
__global const int* hasSeparatingAxis,
__global b3Int4* clippingFacesOut,
__global b3Float4* worldVertsA1,
__global b3Float4* worldNormalsA1,
__global b3Float4* worldVertsB1,
__global b3Float4* worldVertsB2,
int vertexFaceCapacity,
int pairIndex
*/
for (int i=0;i<nPairs;i++)
{
clipFacesAndFindContactsKernel(
&hostSepNormals.at(0),
&hostHasSepAxis.at(0),
&hostClippingFaces.at(0),
&worldVertsA1CPU.at(0),
&worldNormalsACPU.at(0),
&worldVertsB1CPU.at(0),
&worldVertsB2CPU.at(0),
vertexFaceCapacity,
i);
}
clippingFacesOutGPU.copyFromHost(hostClippingFaces);
worldVertsB2GPU.copyFromHost(worldVertsB2CPU);
} else
{ {
B3_PROFILE("clipFacesAndFindContacts"); B3_PROFILE("clipFacesAndFindContacts");
//nContacts = m_totalContactsOut.at(0); //nContacts = m_totalContactsOut.at(0);
@@ -3883,6 +3937,7 @@ void GpuSatCollision::computeConvexConvexContactsGPUSAT( b3OpenCLArray<b3Int4>*
int num = nPairs; int num = nPairs;
launcher.launch1D( num); launcher.launch1D( num);
clFinish(m_queue); clFinish(m_queue);
}
{ {
// nContacts = m_totalContactsOut.at(0); // nContacts = m_totalContactsOut.at(0);
@@ -3891,6 +3946,8 @@ void GpuSatCollision::computeConvexConvexContactsGPUSAT( b3OpenCLArray<b3Int4>*
int newContactCapacity = nContacts+nPairs; int newContactCapacity = nContacts+nPairs;
contactOut->reserve(newContactCapacity); contactOut->reserve(newContactCapacity);
if (reduceConvexContactsOnGPU)
{
{ {
B3_PROFILE("newContactReductionKernel"); B3_PROFILE("newContactReductionKernel");
b3BufferInfoCL bInfo[] = b3BufferInfoCL bInfo[] =
@@ -3916,9 +3973,50 @@ void GpuSatCollision::computeConvexConvexContactsGPUSAT( b3OpenCLArray<b3Int4>*
} }
nContacts = m_totalContactsOut.at(0); nContacts = m_totalContactsOut.at(0);
contactOut->resize(nContacts); contactOut->resize(nContacts);
} else
{
volatile int nGlobalContactsOut = nContacts;
b3AlignedObjectArray<b3Int4> hostPairs;
pairs->copyToHost(hostPairs);
b3AlignedObjectArray<b3RigidBodyCL> hostBodyBuf;
bodyBuf->copyToHost(hostBodyBuf);
b3AlignedObjectArray<b3Vector3> hostSepNormals;
m_sepNormals.copyToHost(hostSepNormals);
b3AlignedObjectArray<int> hostHasSepAxis;
m_hasSeparatingNormals.copyToHost(hostHasSepAxis);
b3AlignedObjectArray<b3Contact4> hostContactsOut;
contactOut->copyToHost(hostContactsOut);
hostContactsOut.resize(newContactCapacity);
b3AlignedObjectArray<b3Int4> hostClippingFaces;
clippingFacesOutGPU.copyToHost(hostClippingFaces);
b3AlignedObjectArray<b3Vector3> worldVertsB2CPU;
worldVertsB2GPU.copyToHost(worldVertsB2CPU);
for (int i=0;i<nPairs;i++)
{
b3NewContactReductionKernel(&hostPairs.at(0),
&hostBodyBuf.at(0),
&hostSepNormals.at(0),
&hostHasSepAxis.at(0),
&hostContactsOut.at(0),
&hostClippingFaces.at(0),
&worldVertsB2CPU.at(0),
&nGlobalContactsOut,
vertexFaceCapacity,
newContactCapacity,
nPairs,
i);
}
nContacts = nGlobalContactsOut;
m_totalContactsOut.copyFromHostPointer(&nContacts,1,0,true);
hostContactsOut.resize(nContacts);
//printf("contactOut4 (after newContactReductionKernel) = %d\n",nContacts);
contactOut->copyFromHost(hostContactsOut);
}
// b3Contact4 pt = contactOut->at(0); // b3Contact4 pt = contactOut->at(0);
// printf("nContacts = %d\n",nContacts); // printf("nContacts = %d\n",nContacts);
} }
} }