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

26
src/Bullet3OpenCL/BroadphaseCollision/kernels/gridBroadphase.cl
View File

@@ -22,15 +22,15 @@ int4 getGridPos(float4 worldPos, __global float4* pParams)
// calculate grid hash value for each body using its AABB
__kernel void kCalcHashAABB(int numObjects, __global float4* pAABB, __global int2* pHash, __global float4* pParams )
__kernel void kCalcHashAABB(int numObjects, __global float4* allpAABB, __global const int* smallAabbMapping, __global int2* pHash, __global float4* pParams )
{
int index = get_global_id(0);
if(index >= numObjects)
{
return;
}
float4 bbMin = pAABB[index*2];
float4 bbMax = pAABB[index*2 + 1];
float4 bbMin = allpAABB[smallAabbMapping[index]*2];
float4 bbMax = allpAABB[smallAabbMapping[index]*2 + 1];
float4 pos;
pos.x = (bbMin.x + bbMax.x) * 0.5f;
pos.y = (bbMin.y + bbMax.y) * 0.5f;
@@ -102,7 +102,8 @@ void findPairsInCell( int numObjects,
int index,
__global int2* pHash,
__global int* pCellStart,
__global float4* pAABB,
__global float4* allpAABB,
__global const int* smallAabbMapping,
__global float4* pParams,
volatile __global int* pairCount,
__global int4* pPairBuff2,
@@ -121,8 +122,8 @@ void findPairsInCell( int numObjects,
// iterate over bodies in this cell
int2 sortedData = pHash[index];
int unsorted_indx = sortedData.y;
float4 min0 = pAABB[unsorted_indx*2 + 0];
float4 max0 = pAABB[unsorted_indx*2 + 1];
float4 min0 = allpAABB[smallAabbMapping[unsorted_indx]*2 + 0];
float4 max0 = allpAABB[smallAabbMapping[unsorted_indx]*2 + 1];
int handleIndex = as_int(min0.w);
int bucketEnd = bucketStart + maxBodiesPerCell;
@@ -138,8 +139,8 @@ void findPairsInCell( int numObjects,
//if (unsorted_indx2 < unsorted_indx) // check not colliding with self
if (unsorted_indx2 != unsorted_indx) // check not colliding with self
{
float4 min1 = pAABB[unsorted_indx2*2 + 0];
float4 max1 = pAABB[unsorted_indx2*2 + 1];
float4 min1 = allpAABB[smallAabbMapping[unsorted_indx2]*2 + 0];
float4 max1 = allpAABB[smallAabbMapping[unsorted_indx2]*2 + 1];
if(testAABBOverlap(min0, max0, min1, max1))
{
if (pairCount)
@@ -166,7 +167,8 @@ void findPairsInCell( int numObjects,
}
__kernel void kFindOverlappingPairs( int numObjects,
__global float4* pAABB,
__global float4* allpAABB,
__global const int* smallAabbMapping,
__global int2* pHash,
__global int* pCellStart,
__global float4* pParams ,
@@ -183,8 +185,8 @@ __kernel void kFindOverlappingPairs( int numObjects,
}
int2 sortedData = pHash[index];
int unsorted_indx = sortedData.y;
float4 bbMin = pAABB[unsorted_indx*2 + 0];
float4 bbMax = pAABB[unsorted_indx*2 + 1];
float4 bbMin = allpAABB[smallAabbMapping[unsorted_indx]*2 + 0];
float4 bbMax = allpAABB[smallAabbMapping[unsorted_indx]*2 + 1];
float4 pos;
pos.x = (bbMin.x + bbMax.x) * 0.5f;
pos.y = (bbMin.y + bbMax.y) * 0.5f;
@@ -202,7 +204,7 @@ __kernel void kFindOverlappingPairs( int numObjects,
for(int x=-1; x<=1; x++)
{
gridPosB.x = gridPosA.x + x;
findPairsInCell(numObjects, gridPosB, index, pHash, pCellStart, pAABB, pParams, pairCount,pPairBuff2, maxPairs);
findPairsInCell(numObjects, gridPosB, index, pHash, pCellStart, allpAABB,smallAabbMapping, pParams, pairCount,pPairBuff2, maxPairs);
}
}
}

26
src/Bullet3OpenCL/BroadphaseCollision/kernels/gridBroadphaseKernels.h
View File

@@ -19,15 +19,15 @@ static const char* gridBroadphaseCL= \
" return gridPos;\n"
"}\n"
"// calculate grid hash value for each body using its AABB\n"
"__kernel void kCalcHashAABB(int numObjects, __global float4* pAABB, __global int2* pHash, __global float4* pParams )\n"
"__kernel void kCalcHashAABB(int numObjects, __global float4* allpAABB, __global const int* smallAabbMapping, __global int2* pHash, __global float4* pParams )\n"
"{\n"
" int index = get_global_id(0);\n"
" if(index >= numObjects)\n"
" {\n"
" return;\n"
" }\n"
" float4 bbMin = pAABB[index*2];\n"
" float4 bbMax = pAABB[index*2 + 1];\n"
" float4 bbMin = allpAABB[smallAabbMapping[index]*2];\n"
" float4 bbMax = allpAABB[smallAabbMapping[index]*2 + 1];\n"
" float4 pos;\n"
" pos.x = (bbMin.x + bbMax.x) * 0.5f;\n"
" pos.y = (bbMin.y + bbMax.y) * 0.5f;\n"
@@ -91,7 +91,8 @@ static const char* gridBroadphaseCL= \
" int index,\n"
" __global int2* pHash,\n"
" __global int* pCellStart,\n"
" __global float4* pAABB, \n"
" __global float4* allpAABB, \n"
" __global const int* smallAabbMapping,\n"
" __global float4* pParams,\n"
" volatile __global int* pairCount,\n"
" __global int4* pPairBuff2,\n"
@@ -110,8 +111,8 @@ static const char* gridBroadphaseCL= \
" // iterate over bodies in this cell\n"
" int2 sortedData = pHash[index];\n"
" int unsorted_indx = sortedData.y;\n"
" float4 min0 = pAABB[unsorted_indx*2 + 0]; \n"
" float4 max0 = pAABB[unsorted_indx*2 + 1];\n"
" float4 min0 = allpAABB[smallAabbMapping[unsorted_indx]*2 + 0]; \n"
" float4 max0 = allpAABB[smallAabbMapping[unsorted_indx]*2 + 1];\n"
" int handleIndex = as_int(min0.w);\n"
" \n"
" int bucketEnd = bucketStart + maxBodiesPerCell;\n"
@@ -127,8 +128,8 @@ static const char* gridBroadphaseCL= \
" //if (unsorted_indx2 < unsorted_indx) // check not colliding with self\n"
" if (unsorted_indx2 != unsorted_indx) // check not colliding with self\n"
" { \n"
" float4 min1 = pAABB[unsorted_indx2*2 + 0];\n"
" float4 max1 = pAABB[unsorted_indx2*2 + 1];\n"
" float4 min1 = allpAABB[smallAabbMapping[unsorted_indx2]*2 + 0];\n"
" float4 max1 = allpAABB[smallAabbMapping[unsorted_indx2]*2 + 1];\n"
" if(testAABBOverlap(min0, max0, min1, max1))\n"
" {\n"
" if (pairCount)\n"
@@ -154,7 +155,8 @@ static const char* gridBroadphaseCL= \
" }\n"
"}\n"
"__kernel void kFindOverlappingPairs( int numObjects,\n"
" __global float4* pAABB, \n"
" __global float4* allpAABB, \n"
" __global const int* smallAabbMapping,\n"
" __global int2* pHash, \n"
" __global int* pCellStart, \n"
" __global float4* pParams ,\n"
@@ -170,8 +172,8 @@ static const char* gridBroadphaseCL= \
" }\n"
" int2 sortedData = pHash[index];\n"
" int unsorted_indx = sortedData.y;\n"
" float4 bbMin = pAABB[unsorted_indx*2 + 0];\n"
" float4 bbMax = pAABB[unsorted_indx*2 + 1];\n"
" float4 bbMin = allpAABB[smallAabbMapping[unsorted_indx]*2 + 0];\n"
" float4 bbMax = allpAABB[smallAabbMapping[unsorted_indx]*2 + 1];\n"
" float4 pos;\n"
" pos.x = (bbMin.x + bbMax.x) * 0.5f;\n"
" pos.y = (bbMin.y + bbMax.y) * 0.5f;\n"
@@ -189,7 +191,7 @@ static const char* gridBroadphaseCL= \
" for(int x=-1; x<=1; x++) \n"
" {\n"
" gridPosB.x = gridPosA.x + x;\n"
" findPairsInCell(numObjects, gridPosB, index, pHash, pCellStart, pAABB, pParams, pairCount,pPairBuff2, maxPairs);\n"
" findPairsInCell(numObjects, gridPosB, index, pHash, pCellStart, allpAABB,smallAabbMapping, pParams, pairCount,pPairBuff2, maxPairs);\n"
" }\n"
" }\n"
" }\n"

11
src/Bullet3OpenCL/BroadphaseCollision/kernels/sap.cl
View File

@@ -63,25 +63,26 @@ bool TestAabbAgainstAabb2Global(const btAabbCL* aabb1, __global const btAabbCL*
}
__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)
__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)
{
int i = get_global_id(0);
if (i>=numUnsortedAabbs)
return;
int j = get_global_id(1);
if (j>=numSortedAabbs)
if (j>=numUnSortedAabbs2)
return;
__global const btAabbCL* unsortedAabbPtr = &unsortedAabbs[unsortedAabbMapping[i]];
__global const btAabbCL* unsortedAabbPtr2 = &unsortedAabbs[unsortedAabbMapping2[j]];
if (TestAabbAgainstAabb2GlobalGlobal(unsortedAabbPtr,&sortedAabbs[j]))
if (TestAabbAgainstAabb2GlobalGlobal(unsortedAabbPtr,unsortedAabbPtr2))
{
int4 myPair;
int xIndex = unsortedAabbPtr[0].m_minIndices[3];
int yIndex = sortedAabbs[j].m_minIndices[3];
int yIndex = unsortedAabbPtr2[0].m_minIndices[3];
if (xIndex>yIndex)
{
int tmp = xIndex;
@@ -349,7 +350,7 @@ __kernel void copyAabbsKernel( __global const btAabbCL* allAabbs, __global btA
}
__kernel void flipFloatKernel( __global const btAabbCL* allAabbs, __global const int* smallAabbMapping, volatile __global int2* sortData, int numObjects, int axis)
__kernel void flipFloatKernel( __global const btAabbCL* allAabbs, __global const int* smallAabbMapping, __global int2* sortData, int numObjects, int axis)
{
int i = get_global_id(0);
if (i>=numObjects)

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"