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

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

src/BulletCollision/BroadphaseCollision/btBroadphaseProxy.h

@@ -91,20 +91,16 @@ BT_DECLARE_ALIGNED_ALLOCATOR();
 	        AllFilter = -1 //all bits sets: DefaultFilter | StaticFilter | KinematicFilter | DebrisFilter | SensorTrigger
 	};
 
-	btVector3	m_aabbMin;
-	btVector3	m_aabbMax;
-
 	//Usually the client btCollisionObject or Rigidbody class
 	void*	m_clientObject;
-
 	short int m_collisionFilterGroup;
 	short int m_collisionFilterMask;
-
 	void*	m_multiSapParentProxy;		
-
-
 	int			m_uniqueId;//m_uniqueId is introduced for paircache. could get rid of this, by calculating the address offset etc.
 
+	btVector3	m_aabbMin;
+	btVector3	m_aabbMax;
+
 	SIMD_FORCE_INLINE int getUid() const
 	{
 		return m_uniqueId;
@@ -116,11 +112,11 @@ BT_DECLARE_ALIGNED_ALLOCATOR();
 	}
 
 	btBroadphaseProxy(const btVector3& aabbMin,const btVector3& aabbMax,void* userPtr,short int collisionFilterGroup, short int collisionFilterMask,void* multiSapParentProxy=0)
-		:m_aabbMin(aabbMin),
-		m_aabbMax(aabbMax),
-		m_clientObject(userPtr),
+		:m_clientObject(userPtr),
 		m_collisionFilterGroup(collisionFilterGroup),
-		m_collisionFilterMask(collisionFilterMask)
+		m_collisionFilterMask(collisionFilterMask),
+		m_aabbMin(aabbMin),
+		m_aabbMax(aabbMax)
 	{
 		m_multiSapParentProxy = multiSapParentProxy;
 	}

src/BulletCollision/BroadphaseCollision/btDbvt.cpp

@@ -500,7 +500,7 @@ void			btDbvt::update(btDbvtNode* leaf,int lookahead)
 }
 
 //
-void			btDbvt::update(btDbvtNode* leaf,const btDbvtVolume& volume)
+void			btDbvt::update(btDbvtNode* leaf,btDbvtVolume& volume)
 {
 	btDbvtNode*	root=removeleaf(this,leaf);
 	if(root)
@@ -518,7 +518,7 @@ void			btDbvt::update(btDbvtNode* leaf,const btDbvtVolume& volume)
 }
 
 //
-bool			btDbvt::update(btDbvtNode* leaf,btDbvtVolume volume,const btVector3& velocity,btScalar margin)
+bool			btDbvt::update(btDbvtNode* leaf,btDbvtVolume& volume,const btVector3& velocity,btScalar margin)
 {
 	if(leaf->volume.Contain(volume)) return(false);
 	volume.Expand(btVector3(margin,margin,margin));
@@ -528,7 +528,7 @@ bool			btDbvt::update(btDbvtNode* leaf,btDbvtVolume volume,const btVector3& velo
 }
 
 //
-bool			btDbvt::update(btDbvtNode* leaf,btDbvtVolume volume,const btVector3& velocity)
+bool			btDbvt::update(btDbvtNode* leaf,btDbvtVolume& volume,const btVector3& velocity)
 {
 	if(leaf->volume.Contain(volume)) return(false);
 	volume.SignedExpand(velocity);
@@ -537,7 +537,7 @@ bool			btDbvt::update(btDbvtNode* leaf,btDbvtVolume volume,const btVector3& velo
 }
 
 //
-bool			btDbvt::update(btDbvtNode* leaf,btDbvtVolume volume,btScalar margin)
+bool			btDbvt::update(btDbvtNode* leaf,btDbvtVolume& volume,btScalar margin)
 {
 	if(leaf->volume.Contain(volume)) return(false);
 	volume.Expand(btVector3(margin,margin,margin));

src/BulletCollision/BroadphaseCollision/btDbvt.h

@@ -276,10 +276,10 @@ struct	btDbvt
 	void			optimizeIncremental(int passes);
 	btDbvtNode*		insert(const btDbvtVolume& box,void* data);
 	void			update(btDbvtNode* leaf,int lookahead=-1);
-	void			update(btDbvtNode* leaf,const btDbvtVolume& volume);
-	bool			update(btDbvtNode* leaf,btDbvtVolume volume,const btVector3& velocity,btScalar margin);
-	bool			update(btDbvtNode* leaf,btDbvtVolume volume,const btVector3& velocity);
-	bool			update(btDbvtNode* leaf,btDbvtVolume volume,btScalar margin);	
+	void			update(btDbvtNode* leaf,btDbvtVolume& volume);
+	bool			update(btDbvtNode* leaf,btDbvtVolume& volume,const btVector3& velocity,btScalar margin);
+	bool			update(btDbvtNode* leaf,btDbvtVolume& volume,const btVector3& velocity);
+	bool			update(btDbvtNode* leaf,btDbvtVolume& volume,btScalar margin);	
 	void			remove(btDbvtNode* leaf);
 	void			write(IWriter* iwriter) const;
 	void			clone(btDbvt& dest,IClone* iclone=0) const;

src/BulletCollision/BroadphaseCollision/btQuantizedBvh.cpp

@@ -20,7 +20,9 @@ subject to the following restrictions:
 
 #define RAYAABB2
 
-btQuantizedBvh::btQuantizedBvh() : m_useQuantization(false), 
+btQuantizedBvh::btQuantizedBvh() : 
+					m_bulletVersion(BT_BULLET_VERSION),
+					m_useQuantization(false), 
 					//m_traversalMode(TRAVERSAL_STACKLESS_CACHE_FRIENDLY)
 					m_traversalMode(TRAVERSAL_STACKLESS)
 					//m_traversalMode(TRAVERSAL_RECURSIVE)
@@ -1136,10 +1138,10 @@ btQuantizedBvh *btQuantizedBvh::deSerializeInPlace(void *i_alignedDataBuffer, un
 btQuantizedBvh::btQuantizedBvh(btQuantizedBvh &self, bool /* ownsMemory */) :
 m_bvhAabbMin(self.m_bvhAabbMin),
 m_bvhAabbMax(self.m_bvhAabbMax),
-m_bvhQuantization(self.m_bvhQuantization)
+m_bvhQuantization(self.m_bvhQuantization),
+m_bulletVersion(BT_BULLET_VERSION)
 {
 
-
 }
 
 

src/BulletCollision/BroadphaseCollision/btQuantizedBvh.h

@@ -158,41 +158,43 @@ typedef btAlignedObjectArray<btBvhSubtreeInfo>		BvhSubtreeInfoArray;
 ///It is recommended to use quantization for better performance and lower memory requirements.
 ATTRIBUTE_ALIGNED16(class) btQuantizedBvh
 {
-protected:
-
-	NodeArray			m_leafNodes;
-	NodeArray			m_contiguousNodes;
-
-	QuantizedNodeArray	m_quantizedLeafNodes;
-	
-	QuantizedNodeArray	m_quantizedContiguousNodes;
-	
-	int					m_curNodeIndex;
-
-
-	//quantization data
-	bool				m_useQuantization;
-	btVector3			m_bvhAabbMin;
-	btVector3			m_bvhAabbMax;
-	btVector3			m_bvhQuantization;
 public:
-	BT_DECLARE_ALIGNED_ALLOCATOR();
-
 	enum btTraversalMode
 	{
 		TRAVERSAL_STACKLESS = 0,
 		TRAVERSAL_STACKLESS_CACHE_FRIENDLY,
 		TRAVERSAL_RECURSIVE
 	};
+
 protected:
 
-	btTraversalMode	m_traversalMode;
+
+	btVector3			m_bvhAabbMin;
+	btVector3			m_bvhAabbMax;
+	btVector3			m_bvhQuantization;
+
+	int					m_bulletVersion;	//for serialization versioning. It could also be used to detect endianess.
+
+	int					m_curNodeIndex;
+	//quantization data
+	bool				m_useQuantization;
+
+
+
+	NodeArray			m_leafNodes;
+	NodeArray			m_contiguousNodes;
+	QuantizedNodeArray	m_quantizedLeafNodes;
+	QuantizedNodeArray	m_quantizedContiguousNodes;
 	
+	btTraversalMode	m_traversalMode;
 	BvhSubtreeInfoArray		m_SubtreeHeaders;
 
 	//This is only used for serialization so we don't have to add serialization directly to btAlignedObjectArray
 	int m_subtreeHeaderCount;
 
+	
+
+
 
 	///two versions, one for quantized and normal nodes. This allows code-reuse while maintaining readability (no template/macro!)
 	///this might be refactored into a virtual, it is usually not calculated at run-time
@@ -332,6 +334,9 @@ protected:
 	void	updateSubtreeHeaders(int leftChildNodexIndex,int rightChildNodexIndex);
 
 public:
+	
+	BT_DECLARE_ALIGNED_ALLOCATOR();
+
 	btQuantizedBvh();
 
 	virtual ~btQuantizedBvh();

src/BulletCollision/CollisionDispatch/btConvexConcaveCollisionAlgorithm.h

@@ -34,8 +34,8 @@ class btConvexTriangleCallback : public btTriangleCallback
 	btVector3	m_aabbMin;
 	btVector3	m_aabbMax ;
 
-	btManifoldResult* m_resultOut;
 
+	btManifoldResult* m_resultOut;
 	btDispatcher*	m_dispatcher;
 	const btDispatcherInfo* m_dispatchInfoPtr;
 	btScalar m_collisionMarginTriangle;
@@ -78,6 +78,7 @@ class btConvexConcaveCollisionAlgorithm  : public btCollisionAlgorithm
 	btConvexTriangleCallback m_btConvexTriangleCallback;
 
 
+
 public:
 
 	btConvexConcaveCollisionAlgorithm( const btCollisionAlgorithmConstructionInfo& ci,btCollisionObject* body0,btCollisionObject* body1,bool isSwapped);

src/BulletCollision/CollisionDispatch/btConvexConvexAlgorithm.cpp

@@ -61,12 +61,15 @@ btConvexConvexAlgorithm::CreateFunc::~CreateFunc()
 
 btConvexConvexAlgorithm::btConvexConvexAlgorithm(btPersistentManifold* mf,const btCollisionAlgorithmConstructionInfo& ci,btCollisionObject* body0,btCollisionObject* body1,btSimplexSolverInterface* simplexSolver, btConvexPenetrationDepthSolver* pdSolver)
 : btCollisionAlgorithm(ci),
-m_gjkPairDetector(0,0,simplexSolver,pdSolver),
+m_simplexSolver(simplexSolver),
+m_pdSolver(pdSolver),
 m_ownManifold (false),
 m_manifoldPtr(mf),
-m_lowLevelOfDetail(false),
-m_sepDistance((static_cast<btConvexShape*>(body0->getCollisionShape()))->getAngularMotionDisc(),
+m_lowLevelOfDetail(false)
+#ifdef USE_SEPDISTANCE_UTIL2
+,m_sepDistance((static_cast<btConvexShape*>(body0->getCollisionShape()))->getAngularMotionDisc(),
 			  (static_cast<btConvexShape*>(body1->getCollisionShape()))->getAngularMotionDisc())
+#endif
 {
 	(void)body0;
 	(void)body1;
@@ -112,9 +115,11 @@ void btConvexConvexAlgorithm ::processCollision (btCollisionObject* body0,btColl
 	btConvexShape* min0 = static_cast<btConvexShape*>(body0->getCollisionShape());
 	btConvexShape* min1 = static_cast<btConvexShape*>(body1->getCollisionShape());
 
+#ifdef USE_SEPDISTANCE_UTIL2
 	m_sepDistance.updateSeparatingDistance(body0->getWorldTransform(),body1->getWorldTransform());
-
 	if (!dispatchInfo.m_useConvexConservativeDistanceUtil || m_sepDistance.getConservativeSeparatingDistance()<=0.f)
+#endif //USE_SEPDISTANCE_UTIL2
+
 	{
 
 #ifdef USE_BT_GJKEPA
@@ -135,9 +140,10 @@ void btConvexConvexAlgorithm ::processCollision (btCollisionObject* body0,btColl
 	
 	btGjkPairDetector::ClosestPointInput input;
 
+	btGjkPairDetector	gjkPairDetector(min0,min1,m_simplexSolver,m_pdSolver);
 	//TODO: if (dispatchInfo.m_useContinuous)
-	m_gjkPairDetector.setMinkowskiA(min0);
-	m_gjkPairDetector.setMinkowskiB(min1);
+	gjkPairDetector.setMinkowskiA(min0);
+	gjkPairDetector.setMinkowskiB(min1);
 	input.m_maximumDistanceSquared = 1e30f;//min0->getMargin() + min1->getMargin() + m_manifoldPtr->getContactBreakingThreshold();
 	//input.m_maximumDistanceSquared*= input.m_maximumDistanceSquared;
 	input.m_stackAlloc = dispatchInfo.m_stackAllocator;
@@ -147,12 +153,14 @@ void btConvexConvexAlgorithm ::processCollision (btCollisionObject* body0,btColl
 	input.m_transformA = body0->getWorldTransform();
 	input.m_transformB = body1->getWorldTransform();
 	
-	m_gjkPairDetector.getClosestPoints(input,*resultOut,dispatchInfo.m_debugDraw);
+	gjkPairDetector.getClosestPoints(input,*resultOut,dispatchInfo.m_debugDraw);
 #endif
 
-	btScalar sepDist = m_gjkPairDetector.getCachedSeparatingDistance()+dispatchInfo.m_convexConservativeDistanceThreshold;
+	btScalar sepDist = gjkPairDetector.getCachedSeparatingDistance()+dispatchInfo.m_convexConservativeDistanceThreshold;
 
+#ifdef USE_SEPDISTANCE_UTIL2
 	m_sepDistance.initSeparatingDistance(m_gjkPairDetector.getCachedSeparatingAxis(),sepDist,body0->getWorldTransform(),body1->getWorldTransform());
+#endif //USE_SEPDISTANCE_UTIL2
 
 
 	}

src/BulletCollision/CollisionDispatch/btConvexConvexAlgorithm.h

@@ -27,19 +27,24 @@ subject to the following restrictions:
 
 class btConvexPenetrationDepthSolver;
 
+//#define USE_SEPDISTANCE_UTIL2 1
+
 ///ConvexConvexAlgorithm collision algorithm implements time of impact, convex closest points and penetration depth calculations.
 class btConvexConvexAlgorithm : public btCollisionAlgorithm
 {
-	btGjkPairDetector m_gjkPairDetector;
-public:
+#ifdef USE_SEPDISTANCE_UTIL2
+	btConvexSeparatingDistanceUtil	m_sepDistance;
+#endif
+	btSimplexSolverInterface*		m_simplexSolver;
+	btConvexPenetrationDepthSolver* m_pdSolver;
 
+	
 	bool	m_ownManifold;
 	btPersistentManifold*	m_manifoldPtr;
 	bool			m_lowLevelOfDetail;
 	
 	///cache separating vector to speedup collision detection
-	btConvexSeparatingDistanceUtil	m_sepDistance;
-
+	
 
 public:
 

src/BulletCollision/CollisionDispatch/btDefaultCollisionConfiguration.cpp

@@ -17,6 +17,7 @@ subject to the following restrictions:
 
 #include "BulletCollision/CollisionDispatch/btConvexConvexAlgorithm.h"
 #include "BulletCollision/CollisionDispatch/btEmptyCollisionAlgorithm.h"
+#include "BulletMultiThreaded/SpuContactManifoldCollisionAlgorithm.h"
 #include "BulletCollision/CollisionDispatch/btConvexConcaveCollisionAlgorithm.h"
 #include "BulletCollision/CollisionDispatch/btCompoundCollisionAlgorithm.h"
 #include "BulletCollision/CollisionDispatch/btConvexPlaneCollisionAlgorithm.h"
@@ -100,8 +101,9 @@ btDefaultCollisionConfiguration::btDefaultCollisionConfiguration(const btDefault
 	int maxSize = sizeof(btConvexConvexAlgorithm);
 	int maxSize2 = sizeof(btConvexConcaveCollisionAlgorithm);
 	int maxSize3 = sizeof(btCompoundCollisionAlgorithm);
-	int maxSize4 = sizeof(btEmptyAlgorithm);
-	
+	int maxSize4 = sizeof(SpuContactManifoldCollisionAlgorithm);
+	int sl = sizeof(btConvexSeparatingDistanceUtil);
+	sl = sizeof(btGjkPairDetector);
 	int	collisionAlgorithmMaxElementSize = btMax(maxSize,maxSize2);
 	collisionAlgorithmMaxElementSize = btMax(collisionAlgorithmMaxElementSize,maxSize3);
 	collisionAlgorithmMaxElementSize = btMax(collisionAlgorithmMaxElementSize,maxSize4);

src/BulletCollision/CollisionShapes/btScaledBvhTriangleMeshShape.cpp

@@ -15,7 +15,7 @@ subject to the following restrictions:
 
 #include "btScaledBvhTriangleMeshShape.h"
 
-btScaledBvhTriangleMeshShape::btScaledBvhTriangleMeshShape(btBvhTriangleMeshShape* childShape,btVector3 localScaling)
+btScaledBvhTriangleMeshShape::btScaledBvhTriangleMeshShape(btBvhTriangleMeshShape* childShape,const btVector3& localScaling)
 :m_localScaling(localScaling),m_bvhTriMeshShape(childShape)
 {
 	m_shapeType = SCALED_TRIANGLE_MESH_SHAPE_PROXYTYPE;
@@ -34,7 +34,7 @@ class btScaledTriangleCallback : public btTriangleCallback
 
 public:
 
-	btScaledTriangleCallback(btTriangleCallback* originalCallback,btVector3 localScaling)
+	btScaledTriangleCallback(btTriangleCallback* originalCallback,const btVector3& localScaling)
 		:m_originalCallback(originalCallback),
 		m_localScaling(localScaling)
 	{

src/BulletCollision/CollisionShapes/btScaledBvhTriangleMeshShape.h

@@ -32,7 +32,7 @@ ATTRIBUTE_ALIGNED16(class) btScaledBvhTriangleMeshShape : public btConcaveShape
 public:
 
 
-	btScaledBvhTriangleMeshShape(btBvhTriangleMeshShape* childShape,btVector3 localScaling);
+	btScaledBvhTriangleMeshShape(btBvhTriangleMeshShape* childShape,const btVector3& localScaling);
 
 	virtual ~btScaledBvhTriangleMeshShape();