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

Avoid breaking up the clipHullHull kernel, it ruins performance. Unfortunately, Mac OSX still requires it.

Browse Source 
Use indices instead of copies for small/large aabbs in broadphase (grid / sap)
This commit is contained in:
erwincoumans
2014-01-29 15:20:20 -08:00
parent ff051f87aa
commit 3e8b183587
11 changed files with 100 additions and 126 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
src/Bullet3OpenCL/BroadphaseCollision/kernels/sapKernels.h
View File

@@ -56,21 +56,22 @@ static const char* sapCL= \
" overlap = (aabb1->m_min.y > aabb2->m_max.y || aabb1->m_max.y < aabb2->m_min.y) ? false : overlap;\n"
" return overlap;\n"
"}\n"
"__kernel void computePairsKernelTwoArrays( __global const btAabbCL* unsortedAabbs, __global const int* unsortedAabbMapping, __global const btAabbCL* sortedAabbs, volatile __global int4* pairsOut,volatile __global int* pairCount, int numUnsortedAabbs, int numSortedAabbs, int axis, int maxPairs)\n"
"__kernel void computePairsKernelTwoArrays( __global const btAabbCL* unsortedAabbs, __global const int* unsortedAabbMapping, __global const int* unsortedAabbMapping2, volatile __global int4* pairsOut,volatile __global int* pairCount, int numUnsortedAabbs, int numUnSortedAabbs2, int axis, int maxPairs)\n"
"{\n"
" int i = get_global_id(0);\n"
" if (i>=numUnsortedAabbs)\n"
" return;\n"
" int j = get_global_id(1);\n"
" if (j>=numSortedAabbs)\n"
" if (j>=numUnSortedAabbs2)\n"
" return;\n"
" __global const btAabbCL* unsortedAabbPtr = &unsortedAabbs[unsortedAabbMapping[i]];\n"
" if (TestAabbAgainstAabb2GlobalGlobal(unsortedAabbPtr,&sortedAabbs[j]))\n"
" __global const btAabbCL* unsortedAabbPtr2 = &unsortedAabbs[unsortedAabbMapping2[j]];\n"
" if (TestAabbAgainstAabb2GlobalGlobal(unsortedAabbPtr,unsortedAabbPtr2))\n"
" {\n"
" int4 myPair;\n"
" \n"
" int xIndex = unsortedAabbPtr[0].m_minIndices[3];\n"
" int yIndex = sortedAabbs[j].m_minIndices[3];\n"
" int yIndex = unsortedAabbPtr2[0].m_minIndices[3];\n"
" if (xIndex>yIndex)\n"
" {\n"
" int tmp = xIndex;\n"
@@ -306,7 +307,7 @@ static const char* sapCL= \
" destAabbs[i] = allAabbs[src];\n"
" destAabbs[i].m_maxIndices[3] = src;\n"
"}\n"
"__kernel void flipFloatKernel( __global const btAabbCL* allAabbs, __global const int* smallAabbMapping, volatile __global int2* sortData, int numObjects, int axis)\n"
"__kernel void flipFloatKernel( __global const btAabbCL* allAabbs, __global const int* smallAabbMapping, __global int2* sortData, int numObjects, int axis)\n"
"{\n"
" int i = get_global_id(0);\n"
" if (i>=numObjects)\n"