diff --git a/ChangeLog.txt b/ChangeLog.txt
index 83e4b66eb..a16dacb29 100644
--- a/ChangeLog.txt
+++ b/ChangeLog.txt
@@ -1,6 +1,10 @@
 Bullet Continuous Collision Detection and Physics Library
 Erwin Coumans
 
+
+2007 Feb 25
+	- Improved performance of convex collision shapes, cache local AABB instead of recomputation. This fixes issue with very slow performance in larger .bsp levels
+	
 2007 Feb 24
 	- Added compressed/quantized AABB tree, 16 bytes per node, while supporting 32-bit (triangle) indices. 
 	  Should be faster and smaller then original version (quantized aabb check is done in integer space)
diff --git a/Demos/ConcaveDemo/ConcavePhysicsDemo.cpp b/Demos/ConcaveDemo/ConcavePhysicsDemo.cpp
index 80fbaa256..bee786e07 100644
--- a/Demos/ConcaveDemo/ConcavePhysicsDemo.cpp
+++ b/Demos/ConcaveDemo/ConcavePhysicsDemo.cpp
@@ -153,6 +153,7 @@ void	ConcaveDemo::initPhysics()
 	btCollisionShape* trimeshShape  = new btBvhTriangleMeshShape(indexVertexArrays,useQuantizedAabbCompression);
 
 
+
 //	btCollisionShape* groundShape = new btBoxShape(btVector3(50,3,50));
 	btCollisionDispatcher* dispatcher = new btCollisionDispatcher();
 	btVector3 worldMin(-1000,-1000,-1000);
diff --git a/Demos/UserCollisionAlgorithm/UserCollisionAlgorithm.cpp b/Demos/UserCollisionAlgorithm/UserCollisionAlgorithm.cpp
index b883dcb3b..7f17b3e1c 100644
--- a/Demos/UserCollisionAlgorithm/UserCollisionAlgorithm.cpp
+++ b/Demos/UserCollisionAlgorithm/UserCollisionAlgorithm.cpp
@@ -100,7 +100,9 @@ void	UserCollisionAlgorithm::initPhysics()
 	
 	delete[] gVertices;
 
-	btCollisionShape* trimeshShape  = new btBvhTriangleMeshShape(trimesh);
+	bool useQuantizedBvhTree = false;
+//	btTriangleMesh uses a subpart for each triangle, this is not compatible with compressed quantized bvh node
+	btCollisionShape* trimeshShape  = new btBvhTriangleMeshShape(trimesh,useQuantizedBvhTree);
 		
 	//ConstraintSolver* solver = new btSequentialImpulseConstraintSolver;
 	
diff --git a/src/BulletCollision/CollisionShapes/btConvexHullShape.cpp b/src/BulletCollision/CollisionShapes/btConvexHullShape.cpp
index 23fbcaa1b..9706a340c 100644
--- a/src/BulletCollision/CollisionShapes/btConvexHullShape.cpp
+++ b/src/BulletCollision/CollisionShapes/btConvexHullShape.cpp
@@ -30,6 +30,8 @@ btConvexHullShape ::btConvexHullShape (const btScalar* points,int numPoints,int
 		btPoint3* point = (btPoint3*)(pointsBaseAddress + i*stride);
 		m_points[i] = point[0];
 	}
+
+	recalcLocalAabb();
 }
 
 btVector3	btConvexHullShape::localGetSupportingVertexWithoutMargin(const btVector3& vec0)const
diff --git a/src/BulletCollision/CollisionShapes/btConvexHullShape.h b/src/BulletCollision/CollisionShapes/btConvexHullShape.h
index 4a5271bb3..f1ac469b5 100644
--- a/src/BulletCollision/CollisionShapes/btConvexHullShape.h
+++ b/src/BulletCollision/CollisionShapes/btConvexHullShape.h
@@ -37,6 +37,7 @@ public:
 	void addPoint(const btPoint3& point)
 	{
 		m_points.push_back(point);
+		recalcLocalAabb();
 	}
 	virtual btVector3	localGetSupportingVertex(const btVector3& vec)const;
 	virtual btVector3	localGetSupportingVertexWithoutMargin(const btVector3& vec)const;
diff --git a/src/BulletCollision/CollisionShapes/btConvexShape.h b/src/BulletCollision/CollisionShapes/btConvexShape.h
index e439a33fe..bf640aa37 100644
--- a/src/BulletCollision/CollisionShapes/btConvexShape.h
+++ b/src/BulletCollision/CollisionShapes/btConvexShape.h
@@ -43,6 +43,7 @@ protected:
 	
 	btScalar	m_collisionMargin;
 
+
 public:
 	btConvexShape();
 
diff --git a/src/BulletCollision/CollisionShapes/btConvexTriangleMeshShape.cpp b/src/BulletCollision/CollisionShapes/btConvexTriangleMeshShape.cpp
index b6b68b1db..de4e93d0a 100644
--- a/src/BulletCollision/CollisionShapes/btConvexTriangleMeshShape.cpp
+++ b/src/BulletCollision/CollisionShapes/btConvexTriangleMeshShape.cpp
@@ -22,6 +22,7 @@ subject to the following restrictions:
 btConvexTriangleMeshShape ::btConvexTriangleMeshShape (btStridingMeshInterface* meshInterface)
 :m_stridingMesh(meshInterface)
 {
+	recalcLocalAabb();
 }
 
 
@@ -189,5 +190,7 @@ bool btConvexTriangleMeshShape::isInside(const btPoint3& pt,btScalar tolerance)
 void	btConvexTriangleMeshShape::setLocalScaling(const btVector3& scaling)
 {
 	m_stridingMesh->setScaling(scaling);
+	
+	recalcLocalAabb();
 }
 
diff --git a/src/BulletCollision/CollisionShapes/btCylinderShape.cpp b/src/BulletCollision/CollisionShapes/btCylinderShape.cpp
index 2e8af6820..f6c77346f 100644
--- a/src/BulletCollision/CollisionShapes/btCylinderShape.cpp
+++ b/src/BulletCollision/CollisionShapes/btCylinderShape.cpp
@@ -18,21 +18,28 @@ subject to the following restrictions:
 btCylinderShape::btCylinderShape (const btVector3& halfExtents)
 :btBoxShape(halfExtents)
 {
-
+	recalcLocalAabb();
 }
 
 
 btCylinderShapeX::btCylinderShapeX (const btVector3& halfExtents)
 :btCylinderShape(halfExtents)
 {
+	recalcLocalAabb();
 }
 
 
 btCylinderShapeZ::btCylinderShapeZ (const btVector3& halfExtents)
 :btCylinderShape(halfExtents)
 {
+	recalcLocalAabb();
 }
 
+void btCylinderShape::getAabb(const btTransform& t,btVector3& aabbMin,btVector3& aabbMax) const
+{
+	//skip the box 'getAabb'
+	btPolyhedralConvexShape::getAabb(t,aabbMin,aabbMax);
+}
 
 
 inline btVector3 CylinderLocalSupportX(const btVector3& halfExtents,const btVector3& v) 
diff --git a/src/BulletCollision/CollisionShapes/btCylinderShape.h b/src/BulletCollision/CollisionShapes/btCylinderShape.h
index 956788727..0855e86ea 100644
--- a/src/BulletCollision/CollisionShapes/btCylinderShape.h
+++ b/src/BulletCollision/CollisionShapes/btCylinderShape.h
@@ -29,10 +29,7 @@ public:
 	btCylinderShape (const btVector3& halfExtents);
 	
 	///getAabb's default implementation is brute force, expected derived classes to implement a fast dedicated version
-	void getAabb(const btTransform& t,btVector3& aabbMin,btVector3& aabbMax) const
-	{
-		getAabbSlow(t,aabbMin,aabbMax);
-	}
+	void getAabb(const btTransform& t,btVector3& aabbMin,btVector3& aabbMax) const;
 
 	virtual btVector3	localGetSupportingVertexWithoutMargin(const btVector3& vec)const;
 
diff --git a/src/BulletCollision/CollisionShapes/btPolyhedralConvexShape.cpp b/src/BulletCollision/CollisionShapes/btPolyhedralConvexShape.cpp
index 6aac93509..58306dce4 100644
--- a/src/BulletCollision/CollisionShapes/btPolyhedralConvexShape.cpp
+++ b/src/BulletCollision/CollisionShapes/btPolyhedralConvexShape.cpp
@@ -16,7 +16,10 @@ subject to the following restrictions:
 #include <BulletCollision/CollisionShapes/btPolyhedralConvexShape.h>
 
 btPolyhedralConvexShape::btPolyhedralConvexShape()
-:m_optionalHull(0)
+:m_optionalHull(0),
+m_localAabbMin(1,1,1),
+m_localAabbMax(-1,-1,-1),
+m_isLocalAabbValid(false)
 {
 
 }
@@ -116,3 +119,48 @@ void	btPolyhedralConvexShape::calculateLocalInertia(btScalar mass,btVector3& ine
 
 }
 
+void btPolyhedralConvexShape::getAabb(const btTransform& trans,btVector3& aabbMin,btVector3& aabbMax) const
+{
+
+	//lazy evaluation of local aabb
+	btAssert(m_isLocalAabbValid);
+
+	btAssert(m_localAabbMin.getX() <= m_localAabbMax.getX());
+	btAssert(m_localAabbMin.getY() <= m_localAabbMax.getY());
+	btAssert(m_localAabbMin.getZ() <= m_localAabbMax.getZ());
+
+
+	btVector3 localHalfExtents = btScalar(0.5)*(m_localAabbMax-m_localAabbMin);
+	btVector3 localCenter = btScalar(0.5)*(m_localAabbMax+m_localAabbMin);
+	
+	btMatrix3x3 abs_b = trans.getBasis().absolute();  
+
+	btPoint3 center = trans(localCenter);
+
+	btVector3 extent = btVector3(abs_b[0].dot(localHalfExtents),
+		   abs_b[1].dot(localHalfExtents),
+		  abs_b[2].dot(localHalfExtents));
+	extent += btVector3(getMargin(),getMargin(),getMargin());
+
+	aabbMin = center - extent;
+	aabbMax = center + extent;
+
+	
+}
+
+void	btPolyhedralConvexShape::recalcLocalAabb()
+{
+	m_isLocalAabbValid = true;
+
+	for (int i=0;i<3;i++)
+	{
+		btVector3 vec(btScalar(0.),btScalar(0.),btScalar(0.));
+		vec[i] = btScalar(1.);
+		btVector3 tmp = localGetSupportingVertex(vec);
+		m_localAabbMax[i] = tmp[i]+m_collisionMargin;
+		vec[i] = btScalar(-1.);
+		tmp = localGetSupportingVertex(vec);
+		m_localAabbMin[i] = tmp[i]-m_collisionMargin;
+	}
+}
+	
\ No newline at end of file
diff --git a/src/BulletCollision/CollisionShapes/btPolyhedralConvexShape.h b/src/BulletCollision/CollisionShapes/btPolyhedralConvexShape.h
index 18e1750a2..f1d40f3e2 100644
--- a/src/BulletCollision/CollisionShapes/btPolyhedralConvexShape.h
+++ b/src/BulletCollision/CollisionShapes/btPolyhedralConvexShape.h
@@ -25,6 +25,10 @@ subject to the following restrictions:
 class btPolyhedralConvexShape : public btConvexShape
 {
 
+	btVector3	m_localAabbMin;
+	btVector3	m_localAabbMax;
+	bool		m_isLocalAabbValid;
+
 public:
 
 	btPolyhedralConvexShape();
@@ -36,6 +40,8 @@ public:
 	virtual void	calculateLocalInertia(btScalar mass,btVector3& inertia);
 
 
+	virtual void getAabb(const btTransform& t,btVector3& aabbMin,btVector3& aabbMax) const;
+	void	recalcLocalAabb();
 
 	virtual int	getNumVertices() const = 0 ;
 	virtual int getNumEdges() const = 0;
diff --git a/src/BulletCollision/CollisionShapes/btTetrahedronShape.cpp b/src/BulletCollision/CollisionShapes/btTetrahedronShape.cpp
index 7cb40c4fa..e2092416a 100644
--- a/src/BulletCollision/CollisionShapes/btTetrahedronShape.cpp
+++ b/src/BulletCollision/CollisionShapes/btTetrahedronShape.cpp
@@ -58,6 +58,8 @@ btBU_Simplex1to4::btBU_Simplex1to4(const btPoint3& pt0,const btPoint3& pt1,const
 void btBU_Simplex1to4::addVertex(const btPoint3& pt)
 {
 	m_vertices[m_numVertices++] = pt;
+
+	recalcLocalAabb();
 }
 
 
diff --git a/src/BulletCollision/CollisionShapes/btTriangleShape.h b/src/BulletCollision/CollisionShapes/btTriangleShape.h
index e7dbae117..14a7c93ab 100644
--- a/src/BulletCollision/CollisionShapes/btTriangleShape.h
+++ b/src/BulletCollision/CollisionShapes/btTriangleShape.h
@@ -57,6 +57,7 @@ public:
 		getVertex((i+1)%3,pb);
 	}
 
+
 	virtual void getAabb(const btTransform& t,btVector3& aabbMin,btVector3& aabbMax)const 
 	{
 //		btAssert(0);
diff --git a/src/BulletDynamics/ConstraintSolver/btSequentialImpulseConstraintSolver.cpp b/src/BulletDynamics/ConstraintSolver/btSequentialImpulseConstraintSolver.cpp
index ea072e131..cab15203a 100644
--- a/src/BulletDynamics/ConstraintSolver/btSequentialImpulseConstraintSolver.cpp
+++ b/src/BulletDynamics/ConstraintSolver/btSequentialImpulseConstraintSolver.cpp
@@ -123,7 +123,8 @@ btSequentialImpulseConstraintSolver::btSequentialImpulseConstraintSolver()
 /// btSequentialImpulseConstraintSolver Sequentially applies impulses
 btScalar btSequentialImpulseConstraintSolver::solveGroup(btPersistentManifold** manifoldPtr, int numManifolds,btTypedConstraint** constraints,int numConstraints,const btContactSolverInfo& infoGlobal,btIDebugDraw* debugDrawer)
 {
-	
+
+
 	btContactSolverInfo info = infoGlobal;
 
 	int numiter = infoGlobal.m_numIterations;
diff --git a/src/LinearMath/btMatrix3x3.h b/src/LinearMath/btMatrix3x3.h
index 97fc1417e..44899640f 100644
--- a/src/LinearMath/btMatrix3x3.h
+++ b/src/LinearMath/btMatrix3x3.h
@@ -45,12 +45,12 @@ class btMatrix3x3 {
 					 zx, zy, zz);
 		}
 		
-		btVector3 getColumn(int i) const
+		SIMD_FORCE_INLINE btVector3 getColumn(int i) const
 		{
 			return btVector3(m_el[0][i],m_el[1][i],m_el[2][i]);
 		}
 
-		const btVector3& getRow(int i) const
+		SIMD_FORCE_INLINE const btVector3& getRow(int i) const
 		{
 			return m_el[i];
 		}
@@ -58,13 +58,13 @@ class btMatrix3x3 {
 
 		SIMD_FORCE_INLINE btVector3&  operator[](int i)
 		{ 
-			assert(0 <= i && i < 3);
+			btFullAssert(0 <= i && i < 3);
 			return m_el[i]; 
 		}
 		
-		const btVector3& operator[](int i) const
+		SIMD_FORCE_INLINE const btVector3& operator[](int i) const
 		{
-			assert(0 <= i && i < 3);
+			btFullAssert(0 <= i && i < 3);
 			return m_el[i]; 
 		}
 		
@@ -102,7 +102,7 @@ class btMatrix3x3 {
 		void setRotation(const btQuaternion& q) 
 		{
 			btScalar d = q.length2();
-			assert(d != btScalar(0.0));
+			btFullAssert(d != btScalar(0.0));
 			btScalar s = btScalar(2.0) / d;
 			btScalar xs = q[0] * s,   ys = q[1] * s,   zs = q[2] * s;
 			btScalar wx = q[3] * xs,  wy = q[3] * ys,  wz = q[3] * zs;
@@ -323,7 +323,7 @@ class btMatrix3x3 {
 	{
 		btVector3 co(cofac(1, 1, 2, 2), cofac(1, 2, 2, 0), cofac(1, 0, 2, 1));
 		btScalar det = (*this)[0].dot(co);
-		assert(det != btScalar(0.0));
+		btFullAssert(det != btScalar(0.0));
 		btScalar s = btScalar(1.0) / det;
 		return btMatrix3x3(co[0] * s, cofac(0, 2, 2, 1) * s, cofac(0, 1, 1, 2) * s,
 								 co[1] * s, cofac(0, 0, 2, 2) * s, cofac(0, 2, 1, 0) * s,
diff --git a/src/LinearMath/btScalar.h b/src/LinearMath/btScalar.h
index ef99310e8..4e6952860 100644
--- a/src/LinearMath/btScalar.h
+++ b/src/LinearMath/btScalar.h
@@ -38,6 +38,8 @@ subject to the following restrictions:
 
 		#include <assert.h>
 		#define btAssert assert
+		//btFullAssert is optional, slows down a lot
+		#define btFullAssert
 #else
 	
 	//non-windows systems
@@ -48,6 +50,8 @@ subject to the following restrictions:
 		#include <assert.h>
 		#endif
 		#define btAssert assert
+		//btFullAssert is optional, slows down a lot
+		#define btFullAssert
 #endif
 
 /// older compilers (gcc 3.x) and Sun needs double version of sqrt etc.
diff --git a/src/LinearMath/btTransformUtil.h b/src/LinearMath/btTransformUtil.h
index 0b93b3d16..d3b219c3a 100644
--- a/src/LinearMath/btTransformUtil.h
+++ b/src/LinearMath/btTransformUtil.h
@@ -74,9 +74,9 @@ public:
 		predictedTransform.setOrigin(curTrans.getOrigin() + linvel * timeStep);
 //	#define QUATERNION_DERIVATIVE
 	#ifdef QUATERNION_DERIVATIVE
-		btQuaternion orn = curTrans.getRotation();
-		orn += (angvel * orn) * (timeStep * btScalar(0.5));
-		orn.normalize();
+		btQuaternion predictedOrn = curTrans.getRotation();
+		predictedOrn += (angvel * predictedOrn) * (timeStep * btScalar(0.5));
+		predictedOrn.normalize();
 	#else
 		//exponential map
 		btVector3 axis;
@@ -101,6 +101,7 @@ public:
 		btQuaternion orn0 = curTrans.getRotation();
 
 		btQuaternion predictedOrn = dorn * orn0;
+		predictedOrn.normalize();
 	#endif
 		predictedTransform.setRotation(predictedOrn);
 	}
diff --git a/src/LinearMath/btVector3.h b/src/LinearMath/btVector3.h
index 8d8d396ef..4222e713f 100644
--- a/src/LinearMath/btVector3.h
+++ b/src/LinearMath/btVector3.h
@@ -64,7 +64,7 @@ public:
 
 	SIMD_FORCE_INLINE btVector3& operator/=(const btScalar& s) 
 	{
-		assert(s != btScalar(0.0));
+		btFullAssert(s != btScalar(0.0));
 		return *this *= btScalar(1.0) / s;
 	}
 
@@ -99,7 +99,7 @@ public:
 	SIMD_FORCE_INLINE btScalar angle(const btVector3& v) const 
 	{
 		btScalar s = btSqrt(length2() * v.length2());
-		assert(s != btScalar(0.0));
+		btFullAssert(s != btScalar(0.0));
 		return btAcos(dot(v) / s);
 	}
 
@@ -213,7 +213,7 @@ operator*(const btScalar& s, const btVector3& v)
 SIMD_FORCE_INLINE btVector3
 operator/(const btVector3& v, const btScalar& s)
 {
-	assert(s != btScalar(0.0));
+	btFullAssert(s != btScalar(0.0));
 	return v * (btScalar(1.0) / s);
 }