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

Some performance improvements and fixes related to btVector3 being aligned on SPU.

Browse Source 
btQuantizedBvh has a version number, memory layout might be different now (due to aligned btVector3)
reorganized some data members of some classes, to reduce memory footprint
This commit is contained in:
erwin.coumans
2008-10-29 02:44:50 +00:00
parent ce5df7cf47
commit 55b29e2355
10 changed files with 241 additions and 110 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
src/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuCollisionShapes.h
View File

@@ -64,7 +64,7 @@ struct bvhMeshShape_LocalStoreMemory
btVector3 localGetSupportingVertexWithoutMargin(int shapeType, void* shape, const btVector3& localDir,struct SpuConvexPolyhedronVertexData* convexVertexData);//, int *featureIndex)
void computeAabb (btVector3& aabbMin, btVector3& aabbMax, btConvexInternalShape* convexShape, ppu_address_t convexShapePtr, int shapeType, btTransform xform);
void computeAabb (btVector3& aabbMin, btVector3& aabbMax, btConvexInternalShape* convexShape, ppu_address_t convexShapePtr, int shapeType, const btTransform& xform);
void dmaBvhShapeData (bvhMeshShape_LocalStoreMemory* bvhMeshShape, btBvhTriangleMeshShape* triMeshShape);
void dmaBvhIndexedMesh (btIndexedMesh* IndexMesh, IndexedMeshArray& indexArray, int index, uint32_t dmaTag);
void dmaBvhSubTreeHeaders (btBvhSubtreeInfo* subTreeHeaders, ppu_address_t subTreePtr, int batchSize, uint32_t dmaTag);
@@ -76,18 +76,27 @@ void dmaCollisionShape (void* collisionShapeLocation, ppu_address_t collisionSha
void dmaCompoundShapeInfo (CompoundShape_LocalStoreMemory* compoundShapeLocation, btCompoundShape* spuCompoundShape, uint32_t dmaTag);
void dmaCompoundSubShapes (CompoundShape_LocalStoreMemory* compoundShapeLocation, btCompoundShape* spuCompoundShape, uint32_t dmaTag);
#define USE_BRANCHFREE_TEST 1
#ifdef USE_BRANCHFREE_TEST
SIMD_FORCE_INLINE unsigned int spuTestQuantizedAabbAgainstQuantizedAabb(unsigned short int* aabbMin1,unsigned short int* aabbMax1,const unsigned short int* aabbMin2,const unsigned short int* aabbMax2)
{
#if defined(__CELLOS_LV2__) && defined (__SPU__)
vec_ushort8 vecMin = {aabbMin1[0],aabbMin2[0],aabbMin1[2],aabbMin2[2],aabbMin1[1],aabbMin2[1],0,0};
vec_ushort8 vecMax = {aabbMax2[0],aabbMax1[0],aabbMax2[2],aabbMax1[2],aabbMax2[1],aabbMax1[1],0,0};
vec_ushort8 isGt = spu_cmpgt(vecMin,vecMax);
return spu_extract(spu_gather(isGt),0)==0;
#else
return btSelect((unsigned)((aabbMin1[0] <= aabbMax2[0]) & (aabbMax1[0] >= aabbMin2[0])
& (aabbMin1[2] <= aabbMax2[2]) & (aabbMax1[2] >= aabbMin2[2])
& (aabbMin1[1] <= aabbMax2[1]) & (aabbMax1[1] >= aabbMin2[1])),
1, 0);
#endif
}
#else
unsigned int spuTestQuantizedAabbAgainstQuantizedAabb(const unsigned short int* aabbMin1,const unsigned short int* aabbMax1,const unsigned short int* aabbMin2,const unsigned short int* aabbMax2)
SIMD_FORCE_INLINE unsigned int spuTestQuantizedAabbAgainstQuantizedAabb(const unsigned short int* aabbMin1,const unsigned short int* aabbMax1,const unsigned short int* aabbMin2,const unsigned short int* aabbMax2)
{
unsigned int overlap = 1;
overlap = (aabbMin1[0] > aabbMax2[0] || aabbMax1[0] < aabbMin2[0]) ? 0 : overlap;