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

pybullet/shared memory API: add some more check for out-of-bounds indices when setting up a command.

Browse Source 
avoid many run-time memory allocations in btCompoundCompoundCollisionAlgorithm
This commit is contained in:
erwincoumans
2017-02-04 12:32:42 -08:00
parent b342f08f86
commit 88fdffba04
2 changed files with 31 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
src/BulletCollision/CollisionDispatch/btCompoundCompoundCollisionAlgorithm.cpp
View File

@@ -24,6 +24,8 @@ subject to the following restrictions:
#include "BulletCollision/CollisionDispatch/btManifoldResult.h"
#include "BulletCollision/CollisionDispatch/btCollisionObjectWrapper.h"
//USE_LOCAL_STACK will avoid most (often all) dynamic memory allocations due to resizing in processCollision and MycollideTT
#define USE_LOCAL_STACK 1
btShapePairCallback gCompoundCompoundChildShapePairCallback = 0;
@@ -251,7 +253,12 @@ static inline void MycollideTT( const btDbvtNode* root0,
int depth=1;
int treshold=btDbvt::DOUBLE_STACKSIZE-4;
btAlignedObjectArray<btDbvt::sStkNN> stkStack;
#ifdef USE_LOCAL_STACK
ATTRIBUTE_ALIGNED16(btDbvt::sStkNN localStack[btDbvt::DOUBLE_STACKSIZE]);
stkStack.initializeFromBuffer(&localStack,btDbvt::DOUBLE_STACKSIZE,btDbvt::DOUBLE_STACKSIZE);
#else
stkStack.resize(btDbvt::DOUBLE_STACKSIZE);
#endif
stkStack[0]=btDbvt::sStkNN(root0,root1);
do {
btDbvt::sStkNN p=stkStack[--depth];
@@ -329,6 +336,10 @@ void btCompoundCompoundCollisionAlgorithm::processCollision (const btCollisionOb
{
int i;
btManifoldArray manifoldArray;
#ifdef USE_LOCAL_STACK
btPersistentManifold localManifolds[4];
manifoldArray.initializeFromBuffer(&localManifolds,0,4);
#endif
btSimplePairArray& pairs = m_childCollisionAlgorithmCache->getOverlappingPairArray();
for (i=0;i<pairs.size();i++)
{