rename gpu_sat -> gpu_narrowphase

opencl/gpu_narrowphase/host/ConvexHullContact.h

@@ -0,0 +1,97 @@
+
+#ifndef _CONVEX_HULL_CONTACT_H
+#define _CONVEX_HULL_CONTACT_H
+
+#include "parallel_primitives/host/btOpenCLArray.h"
+#include "btRigidBodyCL.h"
+#include "BulletCommon/btAlignedObjectArray.h"
+#include "btConvexUtility.h"
+#include "btConvexPolyhedronCL.h"
+#include "btCollidable.h"
+#include "btContact4.h"
+#include "parallel_primitives/host/btInt2.h"
+#include "parallel_primitives/host/btInt4.h"
+#include "btOptimizedBvh.h"
+
+//#include "../../dynamics/basic_demo/Stubs/ChNarrowPhase.h"
+
+
+struct btYetAnotherAabb
+{
+	union
+	{
+		float m_min[4];
+		int m_minIndices[4];
+	};
+	union
+	{
+		float m_max[4];
+		//int m_signedMaxIndices[4];
+		//unsigned int m_unsignedMaxIndices[4];
+	};
+};
+
+struct GpuSatCollision
+{
+	cl_context				m_context;
+	cl_device_id			m_device;
+	cl_command_queue		m_queue;
+	cl_kernel				m_findSeparatingAxisKernel;
+	cl_kernel				m_findConcaveSeparatingAxisKernel;
+	cl_kernel				m_findCompoundPairsKernel;
+	cl_kernel				m_processCompoundPairsKernel;
+
+	cl_kernel				m_clipHullHullKernel;
+	cl_kernel				m_clipCompoundsHullHullKernel;
+    
+    cl_kernel               m_clipFacesAndContactReductionKernel;
+    cl_kernel               m_findClippingFacesKernel;
+    
+	cl_kernel				m_clipHullHullConcaveConvexKernel;
+	cl_kernel				m_extractManifoldAndAddContactKernel;
+    cl_kernel               m_newContactReductionKernel;
+
+	cl_kernel				m_bvhTraversalKernel;
+	cl_kernel				m_primitiveContactsKernel;
+	cl_kernel				m_findConcaveSphereContactsKernel;
+
+	cl_kernel				m_processCompoundPairsPrimitivesKernel;
+    
+
+	btOpenCLArray<int>		m_totalContactsOut;
+
+	GpuSatCollision(cl_context ctx,cl_device_id device, cl_command_queue  q );
+	virtual ~GpuSatCollision();
+	
+
+	void computeConvexConvexContactsGPUSAT( const btOpenCLArray<btInt2>* pairs, int nPairs, 
+			const btOpenCLArray<btRigidBodyCL>* bodyBuf,
+			btOpenCLArray<btContact4>* contactOut, int& nContacts,
+			int maxContactCapacity,
+			const btOpenCLArray<btConvexPolyhedronCL>& hostConvexData,
+			const btOpenCLArray<btVector3>& vertices,
+			const btOpenCLArray<btVector3>& uniqueEdges,
+			const btOpenCLArray<btGpuFace>& faces,
+			const btOpenCLArray<int>& indices,
+			const btOpenCLArray<btCollidable>& gpuCollidables,
+			const btOpenCLArray<btGpuChildShape>& gpuChildShapes,
+
+			const btOpenCLArray<btYetAnotherAabb>& clAabbs,
+           btOpenCLArray<btVector3>& worldVertsB1GPU,
+           btOpenCLArray<btInt4>& clippingFacesOutGPU,
+           btOpenCLArray<btVector3>& worldNormalsAGPU,
+           btOpenCLArray<btVector3>& worldVertsA1GPU,
+           btOpenCLArray<btVector3>& worldVertsB2GPU,
+		   btAlignedObjectArray<class btOptimizedBvh*>& bvhData,
+		   btOpenCLArray<btQuantizedBvhNode>*	treeNodesGPU,
+			btOpenCLArray<btBvhSubtreeInfo>*	subTreesGPU,
+			int numObjects,
+			int maxTriConvexPairCapacity,
+			btOpenCLArray<btInt4>& triangleConvexPairs,
+			int& numTriConvexPairsOut
+			);
+
+
+};
+
+#endif //_CONVEX_HULL_CONTACT_H

opencl/gpu_narrowphase/host/btCollidable.h

@@ -0,0 +1,50 @@
+
+#ifndef BT_COLLIDABLE_H
+#define BT_COLLIDABLE_H
+
+enum btShapeTypes
+{
+	SHAPE_HEIGHT_FIELD=1,
+
+	SHAPE_CONVEX_HULL=3,
+	SHAPE_PLANE=4,
+	SHAPE_CONCAVE_TRIMESH=5,
+	SHAPE_COMPOUND_OF_CONVEX_HULLS=6,
+	SHAPE_SPHERE=7,
+	MAX_NUM_SHAPE_TYPES,
+};
+
+struct btCollidable
+{
+	int m_numChildShapes;
+	float m_radius;
+	int m_shapeType;
+	int m_shapeIndex;
+};
+
+struct btCollidableNew
+{
+	short int m_shapeType;
+	short int m_numShapes;
+	int m_shapeIndex;
+};
+
+struct btGpuChildShape
+{
+	float	m_childPosition[4];
+	float	m_childOrientation[4];
+	int m_shapeIndex;
+	int m_unused0;
+	int m_unused1;
+	int m_unused2;
+};
+
+struct btCompoundOverlappingPair
+{
+	int m_bodyIndexA;
+	int m_bodyIndexB;
+//	int	m_pairType;
+	int m_childShapeIndexA;
+	int m_childShapeIndexB;
+};
+#endif //BT_COLLIDABLE_H

opencl/gpu_narrowphase/host/btContact4.h

@@ -0,0 +1,42 @@
+#ifndef BT_CONTACT4_H
+#define BT_CONTACT4_H
+
+#include "BulletCommon/btVector3.h"
+
+
+ATTRIBUTE_ALIGNED16(struct) btContact4
+{
+	BT_DECLARE_ALIGNED_ALLOCATOR();
+
+	btVector3	m_worldPos[4];
+	btVector3	m_worldNormal;
+//	float m_restituitionCoeff;
+//	float m_frictionCoeff;
+	unsigned short  m_restituitionCoeffCmp;
+	unsigned short  m_frictionCoeffCmp;
+	int m_batchIdx;
+
+	int m_bodyAPtrAndSignBit;
+	int m_bodyBPtrAndSignBit;
+
+	int getBodyA()const {return abs(m_bodyAPtrAndSignBit);}
+	int getBodyB()const {return abs(m_bodyBPtrAndSignBit);}
+	bool isBodyAFixed()const { return m_bodyAPtrAndSignBit<0;}
+	bool isBodyBFixed()const { return m_bodyBPtrAndSignBit<0;}
+	//	todo. make it safer
+	int& getBatchIdx() { return m_batchIdx; }
+	const int& getBatchIdx() const { return m_batchIdx; }
+	float getRestituitionCoeff() const { return ((float)m_restituitionCoeffCmp/(float)0xffff); }
+	void setRestituitionCoeff( float c ) { btAssert( c >= 0.f && c <= 1.f ); m_restituitionCoeffCmp = (unsigned short)(c*0xffff); }
+	float getFrictionCoeff() const { return ((float)m_frictionCoeffCmp/(float)0xffff); }
+	void setFrictionCoeff( float c ) { btAssert( c >= 0.f && c <= 1.f ); m_frictionCoeffCmp = (unsigned short)(c*0xffff); }
+
+	float& getNPoints() { return m_worldNormal[3]; }
+	float getNPoints() const { return m_worldNormal[3]; }
+
+	float getPenetration(int idx) const { return m_worldPos[idx][3]; }
+
+	bool isInvalid() const { return (getBodyA()==0 || getBodyB()==0); }
+};
+
+#endif //BT_CONTACT4_H

opencl/gpu_narrowphase/host/btConvexPolyhedronCL.h

@@ -0,0 +1,64 @@
+#ifndef CONVEX_POLYHEDRON_CL
+#define CONVEX_POLYHEDRON_CL
+
+#include "BulletCommon/btTransform.h"
+
+struct btGpuFace
+{
+	btVector4 m_plane;
+	int m_indexOffset;
+	int m_numIndices;
+};
+
+ATTRIBUTE_ALIGNED16(struct) btConvexPolyhedronCL
+{
+	btVector3		m_localCenter;
+	btVector3		m_extents;
+	btVector3		mC;
+	btVector3		mE;
+
+	btScalar		m_radius;
+	int	m_faceOffset;
+	int m_numFaces;
+	int	m_numVertices;
+
+	int m_vertexOffset;
+	int	m_uniqueEdgesOffset;
+	int	m_numUniqueEdges;
+	int m_unused;	
+	
+
+
+	inline void project(const btTransform& trans, const btVector3& dir, const btAlignedObjectArray<btVector3>& vertices, btScalar& min, btScalar& max) const
+	{
+		min = FLT_MAX;
+		max = -FLT_MAX;
+		int numVerts = m_numVertices;
+
+		const btVector3 localDir = trans.getBasis().transpose()*dir;
+		const btVector3 localDi2 = quatRotate(trans.getRotation().inverse(),dir);
+		
+		btScalar offset = trans.getOrigin().dot(dir);
+
+		for(int i=0;i<numVerts;i++)
+		{
+			//btVector3 pt = trans * vertices[m_vertexOffset+i];
+			//btScalar dp = pt.dot(dir);
+			btScalar dp = vertices[m_vertexOffset+i].dot(localDir);
+			//btAssert(dp==dpL);
+			if(dp < min)	min = dp;
+			if(dp > max)	max = dp;
+		}
+		if(min>max)
+		{
+			btScalar tmp = min;
+			min = max;
+			max = tmp;
+		}
+		min += offset;
+		max += offset;
+	}
+
+};
+
+#endif //CONVEX_POLYHEDRON_CL

opencl/gpu_narrowphase/host/btConvexUtility.cpp

@@ -0,0 +1,520 @@
+/*
+Copyright (c) 2012 Advanced Micro Devices, Inc.  
+
+This software is provided 'as-is', without any express or implied warranty.
+In no event will the authors be held liable for any damages arising from the use of this software.
+Permission is granted to anyone to use this software for any purpose, 
+including commercial applications, and to alter it and redistribute it freely, 
+subject to the following restrictions:
+
+1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
+2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
+3. This notice may not be removed or altered from any source distribution.
+*/
+//Originally written by Erwin Coumans
+
+
+#include "btConvexUtility.h"
+#include "BulletGeometry/btConvexHullComputer.h"
+#include "BulletGeometry/btGrahamScan2dConvexHull.h"
+#include "BulletCommon/btQuaternion.h"
+#include "BulletCommon/btHashMap.h"
+
+#include "btConvexPolyhedronCL.h"
+
+
+
+btConvexUtility::~btConvexUtility()
+{
+}
+
+bool	btConvexUtility::initializePolyhedralFeatures(const btVector3* orgVertices, int numPoints, bool mergeCoplanarTriangles)
+{
+	
+	
+
+	btConvexHullComputer conv;
+	conv.compute(&orgVertices[0].getX(), sizeof(btVector3),numPoints,0.f,0.f);
+
+	btAlignedObjectArray<btVector3> faceNormals;
+	int numFaces = conv.faces.size();
+	faceNormals.resize(numFaces);
+	btConvexHullComputer* convexUtil = &conv;
+
+	
+	btAlignedObjectArray<btMyFace>	tmpFaces;
+	tmpFaces.resize(numFaces);
+
+	int numVertices = convexUtil->vertices.size();
+	m_vertices.resize(numVertices);
+	for (int p=0;p<numVertices;p++)
+	{
+		m_vertices[p] = convexUtil->vertices[p];
+	}
+
+
+	for (int i=0;i<numFaces;i++)
+	{
+		int face = convexUtil->faces[i];
+		//printf("face=%d\n",face);
+		const btConvexHullComputer::Edge*  firstEdge = &convexUtil->edges[face];
+		const btConvexHullComputer::Edge*  edge = firstEdge;
+
+		btVector3 edges[3];
+		int numEdges = 0;
+		//compute face normals
+
+		do
+		{
+			
+			int src = edge->getSourceVertex();
+			tmpFaces[i].m_indices.push_back(src);
+			int targ = edge->getTargetVertex();
+			btVector3 wa = convexUtil->vertices[src];
+
+			btVector3 wb = convexUtil->vertices[targ];
+			btVector3 newEdge = wb-wa;
+			newEdge.normalize();
+			if (numEdges<2)
+				edges[numEdges++] = newEdge;
+
+			edge = edge->getNextEdgeOfFace();
+		} while (edge!=firstEdge);
+
+		btScalar planeEq = 1e30f;
+
+		
+		if (numEdges==2)
+		{
+			faceNormals[i] = edges[0].cross(edges[1]);
+			faceNormals[i].normalize();
+			tmpFaces[i].m_plane[0] = faceNormals[i].getX();
+			tmpFaces[i].m_plane[1] = faceNormals[i].getY();
+			tmpFaces[i].m_plane[2] = faceNormals[i].getZ();
+			tmpFaces[i].m_plane[3] = planeEq;
+
+		}
+		else
+		{
+			btAssert(0);//degenerate?
+			faceNormals[i].setZero();
+		}
+
+		for (int v=0;v<tmpFaces[i].m_indices.size();v++)
+		{
+			btScalar eq = m_vertices[tmpFaces[i].m_indices[v]].dot(faceNormals[i]);
+			if (planeEq>eq)
+			{
+				planeEq=eq;
+			}
+		}
+		tmpFaces[i].m_plane[3] = -planeEq;
+	}
+
+	//merge coplanar faces and copy them to m_polyhedron
+
+	btScalar faceWeldThreshold= 0.999f;
+	btAlignedObjectArray<int> todoFaces;
+	for (int i=0;i<tmpFaces.size();i++)
+		todoFaces.push_back(i);
+
+	while (todoFaces.size())
+	{
+		btAlignedObjectArray<int> coplanarFaceGroup;
+		int refFace = todoFaces[todoFaces.size()-1];
+
+		coplanarFaceGroup.push_back(refFace);
+		btMyFace& faceA = tmpFaces[refFace];
+		todoFaces.pop_back();
+
+		btVector3 faceNormalA(faceA.m_plane[0],faceA.m_plane[1],faceA.m_plane[2]);
+		for (int j=todoFaces.size()-1;j>=0;j--)
+		{
+			int i = todoFaces[j];
+			btMyFace& faceB = tmpFaces[i];
+			btVector3 faceNormalB(faceB.m_plane[0],faceB.m_plane[1],faceB.m_plane[2]);
+			if (faceNormalA.dot(faceNormalB)>faceWeldThreshold)
+			{
+				coplanarFaceGroup.push_back(i);
+				todoFaces.remove(i);
+			}
+		}
+
+
+		bool did_merge = false;
+		if (coplanarFaceGroup.size()>1)
+		{
+			//do the merge: use Graham Scan 2d convex hull
+
+			btAlignedObjectArray<GrahamVector3> orgpoints;
+			btVector3 averageFaceNormal(0,0,0);
+
+			for (int i=0;i<coplanarFaceGroup.size();i++)
+			{
+//				m_polyhedron->m_faces.push_back(tmpFaces[coplanarFaceGroup[i]]);
+
+				btMyFace& face = tmpFaces[coplanarFaceGroup[i]];
+				btVector3 faceNormal(face.m_plane[0],face.m_plane[1],face.m_plane[2]);
+				averageFaceNormal+=faceNormal;
+				for (int f=0;f<face.m_indices.size();f++)
+				{
+					int orgIndex = face.m_indices[f];
+					btVector3 pt = m_vertices[orgIndex];
+					
+					bool found = false;
+
+					for (int i=0;i<orgpoints.size();i++)
+					{
+						//if ((orgpoints[i].m_orgIndex == orgIndex) || ((rotatedPt-orgpoints[i]).length2()<0.0001))
+						if (orgpoints[i].m_orgIndex == orgIndex)
+						{
+							found=true;
+							break;
+						}
+					}
+					if (!found)
+						orgpoints.push_back(GrahamVector3(pt,orgIndex));
+				}
+			}
+
+			
+
+			btMyFace combinedFace;
+			for (int i=0;i<4;i++)
+				combinedFace.m_plane[i] = tmpFaces[coplanarFaceGroup[0]].m_plane[i];
+
+			btAlignedObjectArray<GrahamVector3> hull;
+
+			averageFaceNormal.normalize();
+			GrahamScanConvexHull2D(orgpoints,hull,averageFaceNormal);
+
+			for (int i=0;i<hull.size();i++)
+			{
+				combinedFace.m_indices.push_back(hull[i].m_orgIndex);
+				for(int k = 0; k < orgpoints.size(); k++) 
+				{
+					if(orgpoints[k].m_orgIndex == hull[i].m_orgIndex) 
+					{
+						orgpoints[k].m_orgIndex = -1; // invalidate...
+						break;
+					}
+				}
+			}
+
+			// are there rejected vertices?
+			bool reject_merge = false;
+			
+
+
+			for(int i = 0; i < orgpoints.size(); i++) {
+				if(orgpoints[i].m_orgIndex == -1)
+					continue; // this is in the hull...
+				// this vertex is rejected -- is anybody else using this vertex?
+				for(int j = 0; j < tmpFaces.size(); j++) {
+					
+					btMyFace& face = tmpFaces[j];
+					// is this a face of the current coplanar group?
+					bool is_in_current_group = false;
+					for(int k = 0; k < coplanarFaceGroup.size(); k++) {
+						if(coplanarFaceGroup[k] == j) {
+							is_in_current_group = true;
+							break;
+						}
+					}
+					if(is_in_current_group) // ignore this face...
+						continue;
+					// does this face use this rejected vertex?
+					for(int v = 0; v < face.m_indices.size(); v++) {
+						if(face.m_indices[v] == orgpoints[i].m_orgIndex) {
+							// this rejected vertex is used in another face -- reject merge
+							reject_merge = true;
+							break;
+						}
+					}
+					if(reject_merge)
+						break;
+				}
+				if(reject_merge)
+					break;
+			}
+
+			if (!reject_merge)
+			{
+				// do this merge!
+				did_merge = true;
+				m_faces.push_back(combinedFace);
+			}
+		}
+		if(!did_merge)
+		{
+			for (int i=0;i<coplanarFaceGroup.size();i++)
+			{
+				btMyFace face = tmpFaces[coplanarFaceGroup[i]];
+				m_faces.push_back(face);
+			}
+
+		} 
+
+
+
+	}
+
+	initialize();
+
+	return true;
+}
+
+
+
+
+
+
+inline bool IsAlmostZero(const btVector3& v)
+{
+	if(fabsf(v.getX())>1e-6 || fabsf(v.getY())>1e-6 || fabsf(v.getZ())>1e-6)	return false;
+	return true;
+}
+
+struct btInternalVertexPair
+{
+	btInternalVertexPair(short int v0,short int v1)
+		:m_v0(v0),
+		m_v1(v1)
+	{
+		if (m_v1>m_v0)
+			btSwap(m_v0,m_v1);
+	}
+	short int m_v0;
+	short int m_v1;
+	int getHash() const
+	{
+		return m_v0+(m_v1<<16);
+	}
+	bool equals(const btInternalVertexPair& other) const
+	{
+		return m_v0==other.m_v0 && m_v1==other.m_v1;
+	}
+};
+
+struct btInternalEdge
+{
+	btInternalEdge()
+		:m_face0(-1),
+		m_face1(-1)
+	{
+	}
+	short int m_face0;
+	short int m_face1;
+};
+
+//
+
+#ifdef TEST_INTERNAL_OBJECTS
+bool btConvexUtility::testContainment() const
+{
+	for(int p=0;p<8;p++)
+	{
+		btVector3 LocalPt;
+		if(p==0)		LocalPt = m_localCenter + btVector3(m_extents[0], m_extents[1], m_extents[2]);
+		else if(p==1)	LocalPt = m_localCenter + btVector3(m_extents[0], m_extents[1], -m_extents[2]);
+		else if(p==2)	LocalPt = m_localCenter + btVector3(m_extents[0], -m_extents[1], m_extents[2]);
+		else if(p==3)	LocalPt = m_localCenter + btVector3(m_extents[0], -m_extents[1], -m_extents[2]);
+		else if(p==4)	LocalPt = m_localCenter + btVector3(-m_extents[0], m_extents[1], m_extents[2]);
+		else if(p==5)	LocalPt = m_localCenter + btVector3(-m_extents[0], m_extents[1], -m_extents[2]);
+		else if(p==6)	LocalPt = m_localCenter + btVector3(-m_extents[0], -m_extents[1], m_extents[2]);
+		else if(p==7)	LocalPt = m_localCenter + btVector3(-m_extents[0], -m_extents[1], -m_extents[2]);
+
+		for(int i=0;i<m_faces.size();i++)
+		{
+			const btVector3 Normal(m_faces[i].m_plane[0], m_faces[i].m_plane[1], m_faces[i].m_plane[2]);
+			const btScalar d = LocalPt.dot(Normal) + m_faces[i].m_plane[3];
+			if(d>0.0f)
+				return false;
+		}
+	}
+	return true;
+}
+#endif
+
+void	btConvexUtility::initialize()
+{
+
+	btHashMap<btInternalVertexPair,btInternalEdge> edges;
+
+	btScalar TotalArea = 0.0f;
+	
+	m_localCenter.setValue(0, 0, 0);
+	for(int i=0;i<m_faces.size();i++)
+	{
+		int numVertices = m_faces[i].m_indices.size();
+		int NbTris = numVertices;
+		for(int j=0;j<NbTris;j++)
+		{
+			int k = (j+1)%numVertices;
+			btInternalVertexPair vp(m_faces[i].m_indices[j],m_faces[i].m_indices[k]);
+			btInternalEdge* edptr = edges.find(vp);
+			btVector3 edge = m_vertices[vp.m_v1]-m_vertices[vp.m_v0];
+			edge.normalize();
+
+			bool found = false;
+			btVector3 diff,diff2;
+
+			for (int p=0;p<m_uniqueEdges.size();p++)
+			{
+				diff = m_uniqueEdges[p]-edge;
+				diff2 = m_uniqueEdges[p]+edge;
+
+			//	if ((diff.length2()==0.f) || 
+				//	(diff2.length2()==0.f))
+
+				if (IsAlmostZero(diff) || 
+				IsAlmostZero(diff2))
+				{
+					found = true;
+					break;
+				}
+			}
+
+			if (!found)
+			{
+				m_uniqueEdges.push_back(edge);
+			}
+
+			if (edptr)
+			{
+					//TBD: figure out why I added this assert
+//				btAssert(edptr->m_face0>=0);
+	//			btAssert(edptr->m_face1<0);
+				edptr->m_face1 = i;
+			} else
+			{
+				btInternalEdge ed;
+				ed.m_face0 = i;
+				edges.insert(vp,ed);
+			}
+		}
+	}
+
+#ifdef USE_CONNECTED_FACES
+	for(int i=0;i<m_faces.size();i++)
+	{
+		int numVertices = m_faces[i].m_indices.size();
+		m_faces[i].m_connectedFaces.resize(numVertices);
+
+		for(int j=0;j<numVertices;j++)
+		{
+			int k = (j+1)%numVertices;
+			btInternalVertexPair vp(m_faces[i].m_indices[j],m_faces[i].m_indices[k]);
+			btInternalEdge* edptr = edges.find(vp);
+			btAssert(edptr);
+			btAssert(edptr->m_face0>=0);
+			btAssert(edptr->m_face1>=0);
+
+			int connectedFace = (edptr->m_face0==i)?edptr->m_face1:edptr->m_face0;
+			m_faces[i].m_connectedFaces[j] = connectedFace;
+		}
+	}
+#endif//USE_CONNECTED_FACES
+
+	for(int i=0;i<m_faces.size();i++)
+	{
+		int numVertices = m_faces[i].m_indices.size();
+		int NbTris = numVertices-2;
+		
+		const btVector3& p0 = m_vertices[m_faces[i].m_indices[0]];
+		for(int j=1;j<=NbTris;j++)
+		{
+			int k = (j+1)%numVertices;
+			const btVector3& p1 = m_vertices[m_faces[i].m_indices[j]];
+			const btVector3& p2 = m_vertices[m_faces[i].m_indices[k]];
+			btScalar Area = ((p0 - p1).cross(p0 - p2)).length() * 0.5f;
+			btVector3 Center = (p0+p1+p2)/3.0f;
+			m_localCenter += Area * Center;
+			TotalArea += Area;
+		}
+	}
+	m_localCenter /= TotalArea;
+
+
+
+
+#ifdef TEST_INTERNAL_OBJECTS
+	if(1)
+	{
+		m_radius = FLT_MAX;
+		for(int i=0;i<m_faces.size();i++)
+		{
+			const btVector3 Normal(m_faces[i].m_plane[0], m_faces[i].m_plane[1], m_faces[i].m_plane[2]);
+			const btScalar dist = btFabs(m_localCenter.dot(Normal) + m_faces[i].m_plane[3]);
+			if(dist<m_radius)
+				m_radius = dist;
+		}
+
+	
+		btScalar MinX = FLT_MAX;
+		btScalar MinY = FLT_MAX;
+		btScalar MinZ = FLT_MAX;
+		btScalar MaxX = -FLT_MAX;
+		btScalar MaxY = -FLT_MAX;
+		btScalar MaxZ = -FLT_MAX;
+		for(int i=0; i<m_vertices.size(); i++)
+		{
+			const btVector3& pt = m_vertices[i];
+			if(pt.getX()<MinX)	MinX = pt.getX();
+			if(pt.getX()>MaxX)	MaxX = pt.getX();
+			if(pt.getY()<MinY)	MinY = pt.getY();
+			if(pt.getY()>MaxY)	MaxY = pt.getY();
+			if(pt.getZ()<MinZ)	MinZ = pt.getZ();
+			if(pt.getZ()>MaxZ)	MaxZ = pt.getZ();
+		}
+		mC.setValue(MaxX+MinX, MaxY+MinY, MaxZ+MinZ);
+		mE.setValue(MaxX-MinX, MaxY-MinY, MaxZ-MinZ);
+
+
+
+//		const btScalar r = m_radius / sqrtf(2.0f);
+		const btScalar r = m_radius / sqrtf(3.0f);
+		const int LargestExtent = mE.maxAxis();
+		const btScalar Step = (mE[LargestExtent]*0.5f - r)/1024.0f;
+		m_extents[0] = m_extents[1] = m_extents[2] = r;
+		m_extents[LargestExtent] = mE[LargestExtent]*0.5f;
+		bool FoundBox = false;
+		for(int j=0;j<1024;j++)
+		{
+			if(testContainment())
+			{
+				FoundBox = true;
+				break;
+			}
+
+			m_extents[LargestExtent] -= Step;
+		}
+		if(!FoundBox)
+		{
+			m_extents[0] = m_extents[1] = m_extents[2] = r;
+		}
+		else
+		{
+			// Refine the box
+			const btScalar Step = (m_radius - r)/1024.0f;
+			const int e0 = (1<<LargestExtent) & 3;
+			const int e1 = (1<<e0) & 3;
+
+			for(int j=0;j<1024;j++)
+			{
+				const btScalar Saved0 = m_extents[e0];
+				const btScalar Saved1 = m_extents[e1];
+				m_extents[e0] += Step;
+				m_extents[e1] += Step;
+
+				if(!testContainment())
+				{
+					m_extents[e0] = Saved0;
+					m_extents[e1] = Saved1;
+					break;
+				}
+			}
+		}
+	}
+#endif
+}

opencl/gpu_narrowphase/host/btConvexUtility.h

@@ -0,0 +1,62 @@
+
+/*
+Copyright (c) 2012 Advanced Micro Devices, Inc.  
+
+This software is provided 'as-is', without any express or implied warranty.
+In no event will the authors be held liable for any damages arising from the use of this software.
+Permission is granted to anyone to use this software for any purpose, 
+including commercial applications, and to alter it and redistribute it freely, 
+subject to the following restrictions:
+
+1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
+2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
+3. This notice may not be removed or altered from any source distribution.
+*/
+//Originally written by Erwin Coumans
+
+#ifndef _BT_CONVEX_UTILITY_H
+#define _BT_CONVEX_UTILITY_H
+
+#include "BulletCommon/btAlignedObjectArray.h"
+#include "BulletCommon/btTransform.h"
+
+#include "btConvexPolyhedronCL.h"
+
+
+struct btMyFace
+{
+	btAlignedObjectArray<int>	m_indices;
+	btScalar	m_plane[4];
+};
+
+ATTRIBUTE_ALIGNED16(class) btConvexUtility
+{
+	public:
+	BT_DECLARE_ALIGNED_ALLOCATOR();
+
+	btVector3		m_localCenter;
+	btVector3		m_extents;
+	btVector3		mC;
+	btVector3		mE;
+	btScalar		m_radius;
+	
+	btAlignedObjectArray<btVector3>	m_vertices;
+	btAlignedObjectArray<btMyFace>	m_faces;
+	btAlignedObjectArray<btVector3> m_uniqueEdges;
+
+		
+	btConvexUtility()
+	{
+	}
+	virtual ~btConvexUtility();
+
+	bool	initializePolyhedralFeatures(const btVector3* orgVertices, int numVertices, bool mergeCoplanarTriangles=true);
+		
+	void	initialize();
+	bool testContainment() const;
+
+
+
+};
+#endif
+	

opencl/gpu_narrowphase/host/btOptimizedBvh.cpp

@@ -0,0 +1,391 @@
+/*
+Bullet Continuous Collision Detection and Physics Library
+Copyright (c) 2003-2009 Erwin Coumans  http://bulletphysics.org
+
+This software is provided 'as-is', without any express or implied warranty.
+In no event will the authors be held liable for any damages arising from the use of this software.
+Permission is granted to anyone to use this software for any purpose, 
+including commercial applications, and to alter it and redistribute it freely, 
+subject to the following restrictions:
+
+1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
+2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
+3. This notice may not be removed or altered from any source distribution.
+*/
+
+
+#include "btOptimizedBvh.h"
+#include "btStridingMeshInterface.h"
+#include "BulletGeometry/btAabbUtil2.h"
+#include "BulletCommon/btIDebugDraw.h"
+
+
+btOptimizedBvh::btOptimizedBvh()
+{ 
+}
+
+btOptimizedBvh::~btOptimizedBvh()
+{
+}
+
+
+void btOptimizedBvh::build(btStridingMeshInterface* triangles, bool useQuantizedAabbCompression, const btVector3& bvhAabbMin, const btVector3& bvhAabbMax)
+{
+	m_useQuantization = useQuantizedAabbCompression;
+
+
+	// NodeArray	triangleNodes;
+
+	struct	NodeTriangleCallback : public btInternalTriangleIndexCallback
+	{
+
+		NodeArray&	m_triangleNodes;
+
+		NodeTriangleCallback& operator=(NodeTriangleCallback& other)
+		{
+			m_triangleNodes.copyFromArray(other.m_triangleNodes);
+			return *this;
+		}
+		
+		NodeTriangleCallback(NodeArray&	triangleNodes)
+			:m_triangleNodes(triangleNodes)
+		{
+		}
+
+		virtual void internalProcessTriangleIndex(btVector3* triangle,int partId,int  triangleIndex)
+		{
+			btOptimizedBvhNode node;
+			btVector3	aabbMin,aabbMax;
+			aabbMin.setValue(btScalar(BT_LARGE_FLOAT),btScalar(BT_LARGE_FLOAT),btScalar(BT_LARGE_FLOAT));
+			aabbMax.setValue(btScalar(-BT_LARGE_FLOAT),btScalar(-BT_LARGE_FLOAT),btScalar(-BT_LARGE_FLOAT)); 
+			aabbMin.setMin(triangle[0]);
+			aabbMax.setMax(triangle[0]);
+			aabbMin.setMin(triangle[1]);
+			aabbMax.setMax(triangle[1]);
+			aabbMin.setMin(triangle[2]);
+			aabbMax.setMax(triangle[2]);
+
+			//with quantization?
+			node.m_aabbMinOrg = aabbMin;
+			node.m_aabbMaxOrg = aabbMax;
+
+			node.m_escapeIndex = -1;
+	
+			//for child nodes
+			node.m_subPart = partId;
+			node.m_triangleIndex = triangleIndex;
+			m_triangleNodes.push_back(node);
+		}
+	};
+	struct	QuantizedNodeTriangleCallback : public btInternalTriangleIndexCallback
+	{
+		QuantizedNodeArray&	m_triangleNodes;
+		const btQuantizedBvh* m_optimizedTree; // for quantization
+
+		QuantizedNodeTriangleCallback& operator=(QuantizedNodeTriangleCallback& other)
+		{
+			m_triangleNodes.copyFromArray(other.m_triangleNodes);
+			m_optimizedTree = other.m_optimizedTree;
+			return *this;
+		}
+
+		QuantizedNodeTriangleCallback(QuantizedNodeArray&	triangleNodes,const btQuantizedBvh* tree)
+			:m_triangleNodes(triangleNodes),m_optimizedTree(tree)
+		{
+		}
+
+		virtual void internalProcessTriangleIndex(btVector3* triangle,int partId,int  triangleIndex)
+		{
+			// The partId and triangle index must fit in the same (positive) integer
+			btAssert(partId < (1<<MAX_NUM_PARTS_IN_BITS));
+			btAssert(triangleIndex < (1<<(31-MAX_NUM_PARTS_IN_BITS)));
+			//negative indices are reserved for escapeIndex
+			btAssert(triangleIndex>=0);
+
+			btQuantizedBvhNode node;
+			btVector3	aabbMin,aabbMax;
+			aabbMin.setValue(btScalar(BT_LARGE_FLOAT),btScalar(BT_LARGE_FLOAT),btScalar(BT_LARGE_FLOAT));
+			aabbMax.setValue(btScalar(-BT_LARGE_FLOAT),btScalar(-BT_LARGE_FLOAT),btScalar(-BT_LARGE_FLOAT)); 
+			aabbMin.setMin(triangle[0]);
+			aabbMax.setMax(triangle[0]);
+			aabbMin.setMin(triangle[1]);
+			aabbMax.setMax(triangle[1]);
+			aabbMin.setMin(triangle[2]);
+			aabbMax.setMax(triangle[2]);
+
+			//PCK: add these checks for zero dimensions of aabb
+			const btScalar MIN_AABB_DIMENSION = btScalar(0.002);
+			const btScalar MIN_AABB_HALF_DIMENSION = btScalar(0.001);
+			if (aabbMax.getX() - aabbMin.getX() < MIN_AABB_DIMENSION)
+			{
+				aabbMax.setX(aabbMax.getX() + MIN_AABB_HALF_DIMENSION);
+				aabbMin.setX(aabbMin.getX() - MIN_AABB_HALF_DIMENSION);
+			}
+			if (aabbMax.getY() - aabbMin.getY() < MIN_AABB_DIMENSION)
+			{
+				aabbMax.setY(aabbMax.getY() + MIN_AABB_HALF_DIMENSION);
+				aabbMin.setY(aabbMin.getY() - MIN_AABB_HALF_DIMENSION);
+			}
+			if (aabbMax.getZ() - aabbMin.getZ() < MIN_AABB_DIMENSION)
+			{
+				aabbMax.setZ(aabbMax.getZ() + MIN_AABB_HALF_DIMENSION);
+				aabbMin.setZ(aabbMin.getZ() - MIN_AABB_HALF_DIMENSION);
+			}
+
+			m_optimizedTree->quantize(&node.m_quantizedAabbMin[0],aabbMin,0);
+			m_optimizedTree->quantize(&node.m_quantizedAabbMax[0],aabbMax,1);
+
+			node.m_escapeIndexOrTriangleIndex = (partId<<(31-MAX_NUM_PARTS_IN_BITS)) | triangleIndex;
+
+			m_triangleNodes.push_back(node);
+		}
+	};
+	
+
+
+	int numLeafNodes = 0;
+
+	
+	if (m_useQuantization)
+	{
+
+		//initialize quantization values
+		setQuantizationValues(bvhAabbMin,bvhAabbMax);
+
+		QuantizedNodeTriangleCallback	callback(m_quantizedLeafNodes,this);
+
+	
+		triangles->InternalProcessAllTriangles(&callback,m_bvhAabbMin,m_bvhAabbMax);
+
+		//now we have an array of leafnodes in m_leafNodes
+		numLeafNodes = m_quantizedLeafNodes.size();
+
+
+		m_quantizedContiguousNodes.resize(2*numLeafNodes);
+
+
+	} else
+	{
+		NodeTriangleCallback	callback(m_leafNodes);
+
+		btVector3 aabbMin(btScalar(-BT_LARGE_FLOAT),btScalar(-BT_LARGE_FLOAT),btScalar(-BT_LARGE_FLOAT));
+		btVector3 aabbMax(btScalar(BT_LARGE_FLOAT),btScalar(BT_LARGE_FLOAT),btScalar(BT_LARGE_FLOAT));
+
+		triangles->InternalProcessAllTriangles(&callback,aabbMin,aabbMax);
+
+		//now we have an array of leafnodes in m_leafNodes
+		numLeafNodes = m_leafNodes.size();
+
+		m_contiguousNodes.resize(2*numLeafNodes);
+	}
+
+	m_curNodeIndex = 0;
+
+	buildTree(0,numLeafNodes);
+
+	///if the entire tree is small then subtree size, we need to create a header info for the tree
+	if(m_useQuantization && !m_SubtreeHeaders.size())
+	{
+		btBvhSubtreeInfo& subtree = m_SubtreeHeaders.expand();
+		subtree.setAabbFromQuantizeNode(m_quantizedContiguousNodes[0]);
+		subtree.m_rootNodeIndex = 0;
+		subtree.m_subtreeSize = m_quantizedContiguousNodes[0].isLeafNode() ? 1 : m_quantizedContiguousNodes[0].getEscapeIndex();
+	}
+
+	//PCK: update the copy of the size
+	m_subtreeHeaderCount = m_SubtreeHeaders.size();
+
+	//PCK: clear m_quantizedLeafNodes and m_leafNodes, they are temporary
+	m_quantizedLeafNodes.clear();
+	m_leafNodes.clear();
+}
+
+
+
+
+void	btOptimizedBvh::refit(btStridingMeshInterface* meshInterface,const btVector3& aabbMin,const btVector3& aabbMax)
+{
+	if (m_useQuantization)
+	{
+
+		setQuantizationValues(aabbMin,aabbMax);
+
+		updateBvhNodes(meshInterface,0,m_curNodeIndex,0);
+
+		///now update all subtree headers
+
+		int i;
+		for (i=0;i<m_SubtreeHeaders.size();i++)
+		{
+			btBvhSubtreeInfo& subtree = m_SubtreeHeaders[i];
+			subtree.setAabbFromQuantizeNode(m_quantizedContiguousNodes[subtree.m_rootNodeIndex]);
+		}
+
+	} else
+	{
+
+	}
+}
+
+
+
+
+void	btOptimizedBvh::refitPartial(btStridingMeshInterface* meshInterface,const btVector3& aabbMin,const btVector3& aabbMax)
+{
+	//incrementally initialize quantization values
+	btAssert(m_useQuantization);
+
+	btAssert(aabbMin.getX() > m_bvhAabbMin.getX());
+	btAssert(aabbMin.getY() > m_bvhAabbMin.getY());
+	btAssert(aabbMin.getZ() > m_bvhAabbMin.getZ());
+
+	btAssert(aabbMax.getX() < m_bvhAabbMax.getX());
+	btAssert(aabbMax.getY() < m_bvhAabbMax.getY());
+	btAssert(aabbMax.getZ() < m_bvhAabbMax.getZ());
+
+	///we should update all quantization values, using updateBvhNodes(meshInterface);
+	///but we only update chunks that overlap the given aabb
+	
+	unsigned short	quantizedQueryAabbMin[3];
+	unsigned short	quantizedQueryAabbMax[3];
+
+	quantize(&quantizedQueryAabbMin[0],aabbMin,0);
+	quantize(&quantizedQueryAabbMax[0],aabbMax,1);
+
+	int i;
+	for (i=0;i<this->m_SubtreeHeaders.size();i++)
+	{
+		btBvhSubtreeInfo& subtree = m_SubtreeHeaders[i];
+
+		//PCK: unsigned instead of bool
+		unsigned overlap = testQuantizedAabbAgainstQuantizedAabb(quantizedQueryAabbMin,quantizedQueryAabbMax,subtree.m_quantizedAabbMin,subtree.m_quantizedAabbMax);
+		if (overlap != 0)
+		{
+			updateBvhNodes(meshInterface,subtree.m_rootNodeIndex,subtree.m_rootNodeIndex+subtree.m_subtreeSize,i);
+
+			subtree.setAabbFromQuantizeNode(m_quantizedContiguousNodes[subtree.m_rootNodeIndex]);
+		}
+	}
+	
+}
+
+void	btOptimizedBvh::updateBvhNodes(btStridingMeshInterface* meshInterface,int firstNode,int endNode,int index)
+{
+	(void)index;
+
+	btAssert(m_useQuantization);
+
+	int curNodeSubPart=-1;
+
+	//get access info to trianglemesh data
+		const unsigned char *vertexbase = 0;
+		int numverts = 0;
+		PHY_ScalarType type = PHY_INTEGER;
+		int stride = 0;
+		const unsigned char *indexbase = 0;
+		int indexstride = 0;
+		int numfaces = 0;
+		PHY_ScalarType indicestype = PHY_INTEGER;
+
+		btVector3	triangleVerts[3];
+		btVector3	aabbMin,aabbMax;
+		const btVector3& meshScaling = meshInterface->getScaling();
+		
+		int i;
+		for (i=endNode-1;i>=firstNode;i--)
+		{
+
+
+			btQuantizedBvhNode& curNode = m_quantizedContiguousNodes[i];
+			if (curNode.isLeafNode())
+			{
+				//recalc aabb from triangle data
+				int nodeSubPart = curNode.getPartId();
+				int nodeTriangleIndex = curNode.getTriangleIndex();
+				if (nodeSubPart != curNodeSubPart)
+				{
+					if (curNodeSubPart >= 0)
+						meshInterface->unLockReadOnlyVertexBase(curNodeSubPart);
+					meshInterface->getLockedReadOnlyVertexIndexBase(&vertexbase,numverts,	type,stride,&indexbase,indexstride,numfaces,indicestype,nodeSubPart);
+
+					curNodeSubPart = nodeSubPart;
+					btAssert(indicestype==PHY_INTEGER||indicestype==PHY_SHORT);
+				}
+				//triangles->getLockedReadOnlyVertexIndexBase(vertexBase,numVerts,
+
+				unsigned int* gfxbase = (unsigned int*)(indexbase+nodeTriangleIndex*indexstride);
+				
+				
+				for (int j=2;j>=0;j--)
+				{
+					
+					int graphicsindex = indicestype==PHY_SHORT?((unsigned short*)gfxbase)[j]:gfxbase[j];
+					if (type == PHY_FLOAT)
+					{
+						float* graphicsbase = (float*)(vertexbase+graphicsindex*stride);
+						triangleVerts[j] = btVector3(
+							graphicsbase[0]*meshScaling.getX(),
+							graphicsbase[1]*meshScaling.getY(),
+							graphicsbase[2]*meshScaling.getZ());
+					}
+					else
+					{
+						double* graphicsbase = (double*)(vertexbase+graphicsindex*stride);
+						triangleVerts[j] = btVector3( btScalar(graphicsbase[0]*meshScaling.getX()), btScalar(graphicsbase[1]*meshScaling.getY()), btScalar(graphicsbase[2]*meshScaling.getZ()));
+					}
+				}
+
+
+				
+				aabbMin.setValue(btScalar(BT_LARGE_FLOAT),btScalar(BT_LARGE_FLOAT),btScalar(BT_LARGE_FLOAT));
+				aabbMax.setValue(btScalar(-BT_LARGE_FLOAT),btScalar(-BT_LARGE_FLOAT),btScalar(-BT_LARGE_FLOAT)); 
+				aabbMin.setMin(triangleVerts[0]);
+				aabbMax.setMax(triangleVerts[0]);
+				aabbMin.setMin(triangleVerts[1]);
+				aabbMax.setMax(triangleVerts[1]);
+				aabbMin.setMin(triangleVerts[2]);
+				aabbMax.setMax(triangleVerts[2]);
+
+				quantize(&curNode.m_quantizedAabbMin[0],aabbMin,0);
+				quantize(&curNode.m_quantizedAabbMax[0],aabbMax,1);
+				
+			} else
+			{
+				//combine aabb from both children
+
+				btQuantizedBvhNode* leftChildNode = &m_quantizedContiguousNodes[i+1];
+				
+				btQuantizedBvhNode* rightChildNode = leftChildNode->isLeafNode() ? &m_quantizedContiguousNodes[i+2] :
+					&m_quantizedContiguousNodes[i+1+leftChildNode->getEscapeIndex()];
+				
+
+				{
+					for (int i=0;i<3;i++)
+					{
+						curNode.m_quantizedAabbMin[i] = leftChildNode->m_quantizedAabbMin[i];
+						if (curNode.m_quantizedAabbMin[i]>rightChildNode->m_quantizedAabbMin[i])
+							curNode.m_quantizedAabbMin[i]=rightChildNode->m_quantizedAabbMin[i];
+
+						curNode.m_quantizedAabbMax[i] = leftChildNode->m_quantizedAabbMax[i];
+						if (curNode.m_quantizedAabbMax[i] < rightChildNode->m_quantizedAabbMax[i])
+							curNode.m_quantizedAabbMax[i] = rightChildNode->m_quantizedAabbMax[i];
+					}
+				}
+			}
+
+		}
+
+		if (curNodeSubPart >= 0)
+			meshInterface->unLockReadOnlyVertexBase(curNodeSubPart);
+
+		
+}
+
+///deSerializeInPlace loads and initializes a BVH from a buffer in memory 'in place'
+btOptimizedBvh* btOptimizedBvh::deSerializeInPlace(void *i_alignedDataBuffer, unsigned int i_dataBufferSize, bool i_swapEndian)
+{
+	btQuantizedBvh* bvh = btQuantizedBvh::deSerializeInPlace(i_alignedDataBuffer,i_dataBufferSize,i_swapEndian);
+	
+	//we don't add additional data so just do a static upcast
+	return static_cast<btOptimizedBvh*>(bvh);
+}

opencl/gpu_narrowphase/host/btOptimizedBvh.h

@@ -0,0 +1,65 @@
+/*
+Bullet Continuous Collision Detection and Physics Library
+Copyright (c) 2003-2009 Erwin Coumans  http://bulletphysics.org
+
+This software is provided 'as-is', without any express or implied warranty.
+In no event will the authors be held liable for any damages arising from the use of this software.
+Permission is granted to anyone to use this software for any purpose, 
+including commercial applications, and to alter it and redistribute it freely, 
+subject to the following restrictions:
+
+1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
+2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
+3. This notice may not be removed or altered from any source distribution.
+*/
+
+///Contains contributions from Disney Studio's
+
+#ifndef BT_OPTIMIZED_BVH_H
+#define BT_OPTIMIZED_BVH_H
+
+#include "btQuantizedBvh.h"
+
+class btStridingMeshInterface;
+
+
+///The btOptimizedBvh extends the btQuantizedBvh to create AABB tree for triangle meshes, through the btStridingMeshInterface.
+ATTRIBUTE_ALIGNED16(class) btOptimizedBvh : public btQuantizedBvh
+{
+	
+public:
+	BT_DECLARE_ALIGNED_ALLOCATOR();
+
+protected:
+
+public:
+
+	btOptimizedBvh();
+
+	virtual ~btOptimizedBvh();
+
+	void	build(btStridingMeshInterface* triangles,bool useQuantizedAabbCompression, const btVector3& bvhAabbMin, const btVector3& bvhAabbMax);
+
+	void	refit(btStridingMeshInterface* triangles,const btVector3& aabbMin,const btVector3& aabbMax);
+
+	void	refitPartial(btStridingMeshInterface* triangles,const btVector3& aabbMin, const btVector3& aabbMax);
+
+	void	updateBvhNodes(btStridingMeshInterface* meshInterface,int firstNode,int endNode,int index);
+
+	/// Data buffer MUST be 16 byte aligned
+	virtual bool serializeInPlace(void *o_alignedDataBuffer, unsigned i_dataBufferSize, bool i_swapEndian) const
+	{
+		return btQuantizedBvh::serialize(o_alignedDataBuffer,i_dataBufferSize,i_swapEndian);
+
+	}
+
+	///deSerializeInPlace loads and initializes a BVH from a buffer in memory 'in place'
+	static btOptimizedBvh *deSerializeInPlace(void *i_alignedDataBuffer, unsigned int i_dataBufferSize, bool i_swapEndian);
+
+
+};
+
+
+#endif //BT_OPTIMIZED_BVH_H
+
+

opencl/gpu_narrowphase/host/btQuantizedBvh.h

@@ -0,0 +1,582 @@
+/*
+Bullet Continuous Collision Detection and Physics Library
+Copyright (c) 2003-2006 Erwin Coumans  http://continuousphysics.com/Bullet/
+
+This software is provided 'as-is', without any express or implied warranty.
+In no event will the authors be held liable for any damages arising from the use of this software.
+Permission is granted to anyone to use this software for any purpose, 
+including commercial applications, and to alter it and redistribute it freely, 
+subject to the following restrictions:
+
+1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
+2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
+3. This notice may not be removed or altered from any source distribution.
+*/
+
+#ifndef BT_QUANTIZED_BVH_H
+#define BT_QUANTIZED_BVH_H
+
+class btSerializer;
+
+//#define DEBUG_CHECK_DEQUANTIZATION 1
+#ifdef DEBUG_CHECK_DEQUANTIZATION
+#ifdef __SPU__
+#define printf spu_printf
+#endif //__SPU__
+
+#include <stdio.h>
+#include <stdlib.h>
+#endif //DEBUG_CHECK_DEQUANTIZATION
+
+#include "BulletCommon/btVector3.h"
+#include "BulletCommon/btAlignedAllocator.h"
+
+#ifdef BT_USE_DOUBLE_PRECISION
+#define btQuantizedBvhData btQuantizedBvhDoubleData
+#define btOptimizedBvhNodeData btOptimizedBvhNodeDoubleData
+#define btQuantizedBvhDataName "btQuantizedBvhDoubleData"
+#else
+#define btQuantizedBvhData btQuantizedBvhFloatData
+#define btOptimizedBvhNodeData btOptimizedBvhNodeFloatData
+#define btQuantizedBvhDataName "btQuantizedBvhFloatData"
+#endif
+
+
+
+//http://msdn.microsoft.com/library/default.asp?url=/library/en-us/vclang/html/vclrf__m128.asp
+
+
+//Note: currently we have 16 bytes per quantized node
+#define MAX_SUBTREE_SIZE_IN_BYTES  2048
+
+// 10 gives the potential for 1024 parts, with at most 2^21 (2097152) (minus one
+// actually) triangles each (since the sign bit is reserved
+#define MAX_NUM_PARTS_IN_BITS 10
+
+///btQuantizedBvhNode is a compressed aabb node, 16 bytes.
+///Node can be used for leafnode or internal node. Leafnodes can point to 32-bit triangle index (non-negative range).
+ATTRIBUTE_ALIGNED16	(struct) btQuantizedBvhNode
+{
+	BT_DECLARE_ALIGNED_ALLOCATOR();
+
+	//12 bytes
+	unsigned short int	m_quantizedAabbMin[3];
+	unsigned short int	m_quantizedAabbMax[3];
+	//4 bytes
+	int	m_escapeIndexOrTriangleIndex;
+
+	bool isLeafNode() const
+	{
+		//skipindex is negative (internal node), triangleindex >=0 (leafnode)
+		return (m_escapeIndexOrTriangleIndex >= 0);
+	}
+	int getEscapeIndex() const
+	{
+		btAssert(!isLeafNode());
+		return -m_escapeIndexOrTriangleIndex;
+	}
+	int	getTriangleIndex() const
+	{
+		btAssert(isLeafNode());
+		unsigned int x=0;
+		unsigned int y = (~(x&0))<<(31-MAX_NUM_PARTS_IN_BITS);
+		// Get only the lower bits where the triangle index is stored
+		return (m_escapeIndexOrTriangleIndex&~(y));
+	}
+	int	getPartId() const
+	{
+		btAssert(isLeafNode());
+		// Get only the highest bits where the part index is stored
+		return (m_escapeIndexOrTriangleIndex>>(31-MAX_NUM_PARTS_IN_BITS));
+	}
+}
+;
+
+/// btOptimizedBvhNode contains both internal and leaf node information.
+/// Total node size is 44 bytes / node. You can use the compressed version of 16 bytes.
+ATTRIBUTE_ALIGNED16 (struct) btOptimizedBvhNode
+{
+	BT_DECLARE_ALIGNED_ALLOCATOR();
+
+	//32 bytes
+	btVector3	m_aabbMinOrg;
+	btVector3	m_aabbMaxOrg;
+
+	//4
+	int	m_escapeIndex;
+
+	//8
+	//for child nodes
+	int	m_subPart;
+	int	m_triangleIndex;
+
+//pad the size to 64 bytes
+	char	m_padding[20];
+};
+
+
+///btBvhSubtreeInfo provides info to gather a subtree of limited size
+ATTRIBUTE_ALIGNED16(class) btBvhSubtreeInfo
+{
+public:
+	BT_DECLARE_ALIGNED_ALLOCATOR();
+
+	//12 bytes
+	unsigned short int	m_quantizedAabbMin[3];
+	unsigned short int	m_quantizedAabbMax[3];
+	//4 bytes, points to the root of the subtree
+	int			m_rootNodeIndex;
+	//4 bytes
+	int			m_subtreeSize;
+	int			m_padding[3];
+
+	btBvhSubtreeInfo()
+	{
+		//memset(&m_padding[0], 0, sizeof(m_padding));
+	}
+
+
+	void	setAabbFromQuantizeNode(const btQuantizedBvhNode& quantizedNode)
+	{
+		m_quantizedAabbMin[0] = quantizedNode.m_quantizedAabbMin[0];
+		m_quantizedAabbMin[1] = quantizedNode.m_quantizedAabbMin[1];
+		m_quantizedAabbMin[2] = quantizedNode.m_quantizedAabbMin[2];
+		m_quantizedAabbMax[0] = quantizedNode.m_quantizedAabbMax[0];
+		m_quantizedAabbMax[1] = quantizedNode.m_quantizedAabbMax[1];
+		m_quantizedAabbMax[2] = quantizedNode.m_quantizedAabbMax[2];
+	}
+}
+;
+
+
+class btNodeOverlapCallback
+{
+public:
+	virtual ~btNodeOverlapCallback() {};
+
+	virtual void processNode(int subPart, int triangleIndex) = 0;
+};
+
+#include "BulletCommon/btAlignedAllocator.h"
+#include "BulletCommon/btAlignedObjectArray.h"
+
+
+
+///for code readability:
+typedef btAlignedObjectArray<btOptimizedBvhNode>	NodeArray;
+typedef btAlignedObjectArray<btQuantizedBvhNode>	QuantizedNodeArray;
+typedef btAlignedObjectArray<btBvhSubtreeInfo>		BvhSubtreeInfoArray;
+
+
+///The btQuantizedBvh class stores an AABB tree that can be quickly traversed on CPU and Cell SPU.
+///It is used by the btBvhTriangleMeshShape as midphase, and by the btMultiSapBroadphase.
+///It is recommended to use quantization for better performance and lower memory requirements.
+ATTRIBUTE_ALIGNED16(class) btQuantizedBvh
+{
+public:
+	enum btTraversalMode
+	{
+		TRAVERSAL_STACKLESS = 0,
+		TRAVERSAL_STACKLESS_CACHE_FRIENDLY,
+		TRAVERSAL_RECURSIVE
+	};
+
+
+
+
+	btVector3			m_bvhAabbMin;
+	btVector3			m_bvhAabbMax;
+	btVector3			m_bvhQuantization;
+
+protected:
+	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
+	mutable 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
+	void	setInternalNodeAabbMin(int nodeIndex, const btVector3& aabbMin)
+	{
+		if (m_useQuantization)
+		{
+			quantize(&m_quantizedContiguousNodes[nodeIndex].m_quantizedAabbMin[0] ,aabbMin,0);
+		} else
+		{
+			m_contiguousNodes[nodeIndex].m_aabbMinOrg = aabbMin;
+
+		}
+	}
+	void	setInternalNodeAabbMax(int nodeIndex,const btVector3& aabbMax)
+	{
+		if (m_useQuantization)
+		{
+			quantize(&m_quantizedContiguousNodes[nodeIndex].m_quantizedAabbMax[0],aabbMax,1);
+		} else
+		{
+			m_contiguousNodes[nodeIndex].m_aabbMaxOrg = aabbMax;
+		}
+	}
+
+	btVector3 getAabbMin(int nodeIndex) const
+	{
+		if (m_useQuantization)
+		{
+			return unQuantize(&m_quantizedLeafNodes[nodeIndex].m_quantizedAabbMin[0]);
+		}
+		//non-quantized
+		return m_leafNodes[nodeIndex].m_aabbMinOrg;
+
+	}
+	btVector3 getAabbMax(int nodeIndex) const
+	{
+		if (m_useQuantization)
+		{
+			return unQuantize(&m_quantizedLeafNodes[nodeIndex].m_quantizedAabbMax[0]);
+		} 
+		//non-quantized
+		return m_leafNodes[nodeIndex].m_aabbMaxOrg;
+		
+	}
+
+	
+	void	setInternalNodeEscapeIndex(int nodeIndex, int escapeIndex)
+	{
+		if (m_useQuantization)
+		{
+			m_quantizedContiguousNodes[nodeIndex].m_escapeIndexOrTriangleIndex = -escapeIndex;
+		} 
+		else
+		{
+			m_contiguousNodes[nodeIndex].m_escapeIndex = escapeIndex;
+		}
+
+	}
+
+	void mergeInternalNodeAabb(int nodeIndex,const btVector3& newAabbMin,const btVector3& newAabbMax) 
+	{
+		if (m_useQuantization)
+		{
+			unsigned short int quantizedAabbMin[3];
+			unsigned short int quantizedAabbMax[3];
+			quantize(quantizedAabbMin,newAabbMin,0);
+			quantize(quantizedAabbMax,newAabbMax,1);
+			for (int i=0;i<3;i++)
+			{
+				if (m_quantizedContiguousNodes[nodeIndex].m_quantizedAabbMin[i] > quantizedAabbMin[i])
+					m_quantizedContiguousNodes[nodeIndex].m_quantizedAabbMin[i] = quantizedAabbMin[i];
+
+				if (m_quantizedContiguousNodes[nodeIndex].m_quantizedAabbMax[i] < quantizedAabbMax[i])
+					m_quantizedContiguousNodes[nodeIndex].m_quantizedAabbMax[i] = quantizedAabbMax[i];
+
+			}
+		} else
+		{
+			//non-quantized
+			m_contiguousNodes[nodeIndex].m_aabbMinOrg.setMin(newAabbMin);
+			m_contiguousNodes[nodeIndex].m_aabbMaxOrg.setMax(newAabbMax);		
+		}
+	}
+
+	void	swapLeafNodes(int firstIndex,int secondIndex);
+
+	void	assignInternalNodeFromLeafNode(int internalNode,int leafNodeIndex);
+
+protected:
+
+	
+
+	void	buildTree	(int startIndex,int endIndex);
+
+	int	calcSplittingAxis(int startIndex,int endIndex);
+
+	int	sortAndCalcSplittingIndex(int startIndex,int endIndex,int splitAxis);
+	
+	void	walkStacklessTree(btNodeOverlapCallback* nodeCallback,const btVector3& aabbMin,const btVector3& aabbMax) const;
+
+	void	walkStacklessQuantizedTreeAgainstRay(btNodeOverlapCallback* nodeCallback, const btVector3& raySource, const btVector3& rayTarget, const btVector3& aabbMin, const btVector3& aabbMax, int startNodeIndex,int endNodeIndex) const;
+	void	walkStacklessQuantizedTree(btNodeOverlapCallback* nodeCallback,unsigned short int* quantizedQueryAabbMin,unsigned short int* quantizedQueryAabbMax,int startNodeIndex,int endNodeIndex) const;
+	void	walkStacklessTreeAgainstRay(btNodeOverlapCallback* nodeCallback, const btVector3& raySource, const btVector3& rayTarget, const btVector3& aabbMin, const btVector3& aabbMax, int startNodeIndex,int endNodeIndex) const;
+
+	///tree traversal designed for small-memory processors like PS3 SPU
+	void	walkStacklessQuantizedTreeCacheFriendly(btNodeOverlapCallback* nodeCallback,unsigned short int* quantizedQueryAabbMin,unsigned short int* quantizedQueryAabbMax) const;
+
+	///use the 16-byte stackless 'skipindex' node tree to do a recursive traversal
+	void	walkRecursiveQuantizedTreeAgainstQueryAabb(const btQuantizedBvhNode* currentNode,btNodeOverlapCallback* nodeCallback,unsigned short int* quantizedQueryAabbMin,unsigned short int* quantizedQueryAabbMax) const;
+
+	///use the 16-byte stackless 'skipindex' node tree to do a recursive traversal
+	void	walkRecursiveQuantizedTreeAgainstQuantizedTree(const btQuantizedBvhNode* treeNodeA,const btQuantizedBvhNode* treeNodeB,btNodeOverlapCallback* nodeCallback) const;
+	
+
+
+
+	void	updateSubtreeHeaders(int leftChildNodexIndex,int rightChildNodexIndex);
+
+public:
+	
+	BT_DECLARE_ALIGNED_ALLOCATOR();
+
+	btQuantizedBvh();
+
+	virtual ~btQuantizedBvh();
+
+	
+	///***************************************** expert/internal use only *************************
+	void	setQuantizationValues(const btVector3& bvhAabbMin,const btVector3& bvhAabbMax,btScalar quantizationMargin=btScalar(1.0));
+	QuantizedNodeArray&	getLeafNodeArray() {			return	m_quantizedLeafNodes;	}
+	///buildInternal is expert use only: assumes that setQuantizationValues and LeafNodeArray are initialized
+	void	buildInternal();
+	///***************************************** expert/internal use only *************************
+
+	void	reportAabbOverlappingNodex(btNodeOverlapCallback* nodeCallback,const btVector3& aabbMin,const btVector3& aabbMax) const;
+	void	reportRayOverlappingNodex (btNodeOverlapCallback* nodeCallback, const btVector3& raySource, const btVector3& rayTarget) const;
+	void	reportBoxCastOverlappingNodex(btNodeOverlapCallback* nodeCallback, const btVector3& raySource, const btVector3& rayTarget, const btVector3& aabbMin,const btVector3& aabbMax) const;
+
+		SIMD_FORCE_INLINE void quantize(unsigned short* out, const btVector3& point,int isMax) const
+	{
+
+		btAssert(m_useQuantization);
+
+		btAssert(point.getX() <= m_bvhAabbMax.getX());
+		btAssert(point.getY() <= m_bvhAabbMax.getY());
+		btAssert(point.getZ() <= m_bvhAabbMax.getZ());
+
+		btAssert(point.getX() >= m_bvhAabbMin.getX());
+		btAssert(point.getY() >= m_bvhAabbMin.getY());
+		btAssert(point.getZ() >= m_bvhAabbMin.getZ());
+
+		btVector3 v = (point - m_bvhAabbMin) * m_bvhQuantization;
+		///Make sure rounding is done in a way that unQuantize(quantizeWithClamp(...)) is conservative
+		///end-points always set the first bit, so that they are sorted properly (so that neighbouring AABBs overlap properly)
+		///@todo: double-check this
+		if (isMax)
+		{
+			out[0] = (unsigned short) (((unsigned short)(v.getX()+btScalar(1.)) | 1));
+			out[1] = (unsigned short) (((unsigned short)(v.getY()+btScalar(1.)) | 1));
+			out[2] = (unsigned short) (((unsigned short)(v.getZ()+btScalar(1.)) | 1));
+		} else
+		{
+			out[0] = (unsigned short) (((unsigned short)(v.getX()) & 0xfffe));
+			out[1] = (unsigned short) (((unsigned short)(v.getY()) & 0xfffe));
+			out[2] = (unsigned short) (((unsigned short)(v.getZ()) & 0xfffe));
+		}
+
+
+#ifdef DEBUG_CHECK_DEQUANTIZATION
+		btVector3 newPoint = unQuantize(out);
+		if (isMax)
+		{
+			if (newPoint.getX() < point.getX())
+			{
+				printf("unconservative X, diffX = %f, oldX=%f,newX=%f\n",newPoint.getX()-point.getX(), newPoint.getX(),point.getX());
+			}
+			if (newPoint.getY() < point.getY())
+			{
+				printf("unconservative Y, diffY = %f, oldY=%f,newY=%f\n",newPoint.getY()-point.getY(), newPoint.getY(),point.getY());
+			}
+			if (newPoint.getZ() < point.getZ())
+			{
+
+				printf("unconservative Z, diffZ = %f, oldZ=%f,newZ=%f\n",newPoint.getZ()-point.getZ(), newPoint.getZ(),point.getZ());
+			}
+		} else
+		{
+			if (newPoint.getX() > point.getX())
+			{
+				printf("unconservative X, diffX = %f, oldX=%f,newX=%f\n",newPoint.getX()-point.getX(), newPoint.getX(),point.getX());
+			}
+			if (newPoint.getY() > point.getY())
+			{
+				printf("unconservative Y, diffY = %f, oldY=%f,newY=%f\n",newPoint.getY()-point.getY(), newPoint.getY(),point.getY());
+			}
+			if (newPoint.getZ() > point.getZ())
+			{
+				printf("unconservative Z, diffZ = %f, oldZ=%f,newZ=%f\n",newPoint.getZ()-point.getZ(), newPoint.getZ(),point.getZ());
+			}
+		}
+#endif //DEBUG_CHECK_DEQUANTIZATION
+
+	}
+
+
+	SIMD_FORCE_INLINE void quantizeWithClamp(unsigned short* out, const btVector3& point2,int isMax) const
+	{
+
+		btAssert(m_useQuantization);
+
+		btVector3 clampedPoint(point2);
+		clampedPoint.setMax(m_bvhAabbMin);
+		clampedPoint.setMin(m_bvhAabbMax);
+
+		quantize(out,clampedPoint,isMax);
+
+	}
+	
+	SIMD_FORCE_INLINE btVector3	unQuantize(const unsigned short* vecIn) const
+	{
+			btVector3	vecOut;
+			vecOut.setValue(
+			(btScalar)(vecIn[0]) / (m_bvhQuantization.getX()),
+			(btScalar)(vecIn[1]) / (m_bvhQuantization.getY()),
+			(btScalar)(vecIn[2]) / (m_bvhQuantization.getZ()));
+			vecOut += m_bvhAabbMin;
+			return vecOut;
+	}
+
+	///setTraversalMode let's you choose between stackless, recursive or stackless cache friendly tree traversal. Note this is only implemented for quantized trees.
+	void	setTraversalMode(btTraversalMode	traversalMode)
+	{
+		m_traversalMode = traversalMode;
+	}
+
+
+	SIMD_FORCE_INLINE QuantizedNodeArray&	getQuantizedNodeArray()
+	{	
+		return	m_quantizedContiguousNodes;
+	}
+
+
+	SIMD_FORCE_INLINE BvhSubtreeInfoArray&	getSubtreeInfoArray()
+	{
+		return m_SubtreeHeaders;
+	}
+
+////////////////////////////////////////////////////////////////////
+
+	/////Calculate space needed to store BVH for serialization
+	unsigned calculateSerializeBufferSize() const;
+
+	/// Data buffer MUST be 16 byte aligned
+	virtual bool serialize(void *o_alignedDataBuffer, unsigned i_dataBufferSize, bool i_swapEndian) const;
+
+	///deSerializeInPlace loads and initializes a BVH from a buffer in memory 'in place'
+	static btQuantizedBvh *deSerializeInPlace(void *i_alignedDataBuffer, unsigned int i_dataBufferSize, bool i_swapEndian);
+
+	static unsigned int getAlignmentSerializationPadding();
+//////////////////////////////////////////////////////////////////////
+
+	
+	virtual	int	calculateSerializeBufferSizeNew() const;
+
+	///fills the dataBuffer and returns the struct name (and 0 on failure)
+	virtual	const char*	serialize(void* dataBuffer, btSerializer* serializer) const;
+
+	virtual	void deSerializeFloat(struct btQuantizedBvhFloatData& quantizedBvhFloatData);
+
+	virtual	void deSerializeDouble(struct btQuantizedBvhDoubleData& quantizedBvhDoubleData);
+
+
+////////////////////////////////////////////////////////////////////
+
+	SIMD_FORCE_INLINE bool isQuantized()
+	{
+		return m_useQuantization;
+	}
+
+private:
+	// Special "copy" constructor that allows for in-place deserialization
+	// Prevents btVector3's default constructor from being called, but doesn't inialize much else
+	// ownsMemory should most likely be false if deserializing, and if you are not, don't call this (it also changes the function signature, which we need)
+	btQuantizedBvh(btQuantizedBvh &other, bool ownsMemory);
+
+}
+;
+
+
+struct	btBvhSubtreeInfoData
+{
+	int			m_rootNodeIndex;
+	int			m_subtreeSize;
+	unsigned short m_quantizedAabbMin[3];
+	unsigned short m_quantizedAabbMax[3];
+};
+
+struct btOptimizedBvhNodeFloatData
+{
+	btVector3FloatData	m_aabbMinOrg;
+	btVector3FloatData	m_aabbMaxOrg;
+	int	m_escapeIndex;
+	int	m_subPart;
+	int	m_triangleIndex;
+	char m_pad[4];
+};
+
+struct btOptimizedBvhNodeDoubleData
+{
+	btVector3DoubleData	m_aabbMinOrg;
+	btVector3DoubleData	m_aabbMaxOrg;
+	int	m_escapeIndex;
+	int	m_subPart;
+	int	m_triangleIndex;
+	char	m_pad[4];
+};
+
+
+struct btQuantizedBvhNodeData
+{
+	unsigned short m_quantizedAabbMin[3];
+	unsigned short m_quantizedAabbMax[3];
+	int	m_escapeIndexOrTriangleIndex;
+};
+
+struct	btQuantizedBvhFloatData
+{
+	btVector3FloatData			m_bvhAabbMin;
+	btVector3FloatData			m_bvhAabbMax;
+	btVector3FloatData			m_bvhQuantization;
+	int					m_curNodeIndex;
+	int					m_useQuantization;
+	int					m_numContiguousLeafNodes;
+	int					m_numQuantizedContiguousNodes;
+	btOptimizedBvhNodeFloatData	*m_contiguousNodesPtr;
+	btQuantizedBvhNodeData		*m_quantizedContiguousNodesPtr;
+	btBvhSubtreeInfoData	*m_subTreeInfoPtr;
+	int					m_traversalMode;
+	int					m_numSubtreeHeaders;
+	
+};
+
+struct	btQuantizedBvhDoubleData
+{
+	btVector3DoubleData			m_bvhAabbMin;
+	btVector3DoubleData			m_bvhAabbMax;
+	btVector3DoubleData			m_bvhQuantization;
+	int							m_curNodeIndex;
+	int							m_useQuantization;
+	int							m_numContiguousLeafNodes;
+	int							m_numQuantizedContiguousNodes;
+	btOptimizedBvhNodeDoubleData	*m_contiguousNodesPtr;
+	btQuantizedBvhNodeData			*m_quantizedContiguousNodesPtr;
+
+	int							m_traversalMode;
+	int							m_numSubtreeHeaders;
+	btBvhSubtreeInfoData		*m_subTreeInfoPtr;
+};
+
+
+SIMD_FORCE_INLINE	int	btQuantizedBvh::calculateSerializeBufferSizeNew() const
+{
+	return sizeof(btQuantizedBvhData);
+}
+
+
+
+#endif //BT_QUANTIZED_BVH_H

opencl/gpu_narrowphase/host/btRigidBodyCL.h

@@ -0,0 +1,35 @@
+#ifndef BT_RIGID_BODY_CL
+#define BT_RIGID_BODY_CL
+
+#include "BulletCommon/btScalar.h"
+#include "BulletCommon/btMatrix3x3.h"
+
+ATTRIBUTE_ALIGNED16(struct) btRigidBodyCL
+{
+	BT_DECLARE_ALIGNED_ALLOCATOR();
+
+	btVector3 		m_pos;
+	btQuaternion 	m_quat;
+	btVector3			m_linVel;
+	btVector3			m_angVel;
+
+	int 					m_collidableIdx;
+	float 				m_invMass;
+	float 				m_restituitionCoeff;
+	float 				m_frictionCoeff;
+
+	float	getInvMass() const
+	{
+			return m_invMass;
+	}
+};
+
+
+struct btInertiaCL
+{
+	btMatrix3x3 m_invInertiaWorld;
+	btMatrix3x3 m_initInvInertia;
+};
+
+
+#endif//BT_RIGID_BODY_CL

opencl/gpu_narrowphase/host/btStridingMeshInterface.cpp

@@ -0,0 +1,214 @@
+/*
+Bullet Continuous Collision Detection and Physics Library
+Copyright (c) 2003-2009 Erwin Coumans  http://bulletphysics.org
+
+This software is provided 'as-is', without any express or implied warranty.
+In no event will the authors be held liable for any damages arising from the use of this software.
+Permission is granted to anyone to use this software for any purpose, 
+including commercial applications, and to alter it and redistribute it freely, 
+subject to the following restrictions:
+
+1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
+2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
+3. This notice may not be removed or altered from any source distribution.
+*/
+
+#include "btStridingMeshInterface.h"
+
+
+btStridingMeshInterface::~btStridingMeshInterface()
+{
+
+}
+
+
+void	btStridingMeshInterface::InternalProcessAllTriangles(btInternalTriangleIndexCallback* callback,const btVector3& aabbMin,const btVector3& aabbMax) const
+{
+	(void)aabbMin;
+	(void)aabbMax;
+	int numtotalphysicsverts = 0;
+	int part,graphicssubparts = getNumSubParts();
+	const unsigned char * vertexbase;
+	const unsigned char * indexbase;
+	int indexstride;
+	PHY_ScalarType type;
+	PHY_ScalarType gfxindextype;
+	int stride,numverts,numtriangles;
+	int gfxindex;
+	btVector3 triangle[3];
+
+	btVector3 meshScaling = getScaling();
+
+	///if the number of parts is big, the performance might drop due to the innerloop switch on indextype
+	for (part=0;part<graphicssubparts ;part++)
+	{
+		getLockedReadOnlyVertexIndexBase(&vertexbase,numverts,type,stride,&indexbase,indexstride,numtriangles,gfxindextype,part);
+		numtotalphysicsverts+=numtriangles*3; //upper bound
+
+		///unlike that developers want to pass in double-precision meshes in single-precision Bullet build
+		///so disable this feature by default
+		///see patch http://code.google.com/p/bullet/issues/detail?id=213
+
+		switch (type)
+		{
+		case PHY_FLOAT:
+		 {
+
+			 float* graphicsbase;
+
+			 switch (gfxindextype)
+			 {
+			 case PHY_INTEGER:
+				 {
+					 for (gfxindex=0;gfxindex<numtriangles;gfxindex++)
+					 {
+						 unsigned int* tri_indices= (unsigned int*)(indexbase+gfxindex*indexstride);
+						 graphicsbase = (float*)(vertexbase+tri_indices[0]*stride);
+						 triangle[0].setValue(graphicsbase[0]*meshScaling.getX(),graphicsbase[1]*meshScaling.getY(),graphicsbase[2]*meshScaling.getZ());
+						 graphicsbase = (float*)(vertexbase+tri_indices[1]*stride);
+						 triangle[1].setValue(graphicsbase[0]*meshScaling.getX(),graphicsbase[1]*meshScaling.getY(),	graphicsbase[2]*meshScaling.getZ());
+						 graphicsbase = (float*)(vertexbase+tri_indices[2]*stride);
+						 triangle[2].setValue(graphicsbase[0]*meshScaling.getX(),graphicsbase[1]*meshScaling.getY(),	graphicsbase[2]*meshScaling.getZ());
+						 callback->internalProcessTriangleIndex(triangle,part,gfxindex);
+					 }
+					 break;
+				 }
+			 case PHY_SHORT:
+				 {
+					 for (gfxindex=0;gfxindex<numtriangles;gfxindex++)
+					 {
+						 unsigned short int* tri_indices= (unsigned short int*)(indexbase+gfxindex*indexstride);
+						 graphicsbase = (float*)(vertexbase+tri_indices[0]*stride);
+						 triangle[0].setValue(graphicsbase[0]*meshScaling.getX(),graphicsbase[1]*meshScaling.getY(),graphicsbase[2]*meshScaling.getZ());
+						 graphicsbase = (float*)(vertexbase+tri_indices[1]*stride);
+						 triangle[1].setValue(graphicsbase[0]*meshScaling.getX(),graphicsbase[1]*meshScaling.getY(),	graphicsbase[2]*meshScaling.getZ());
+						 graphicsbase = (float*)(vertexbase+tri_indices[2]*stride);
+						 triangle[2].setValue(graphicsbase[0]*meshScaling.getX(),graphicsbase[1]*meshScaling.getY(),	graphicsbase[2]*meshScaling.getZ());
+						 callback->internalProcessTriangleIndex(triangle,part,gfxindex);
+					 }
+					 break;
+				 }
+			case PHY_UCHAR:
+				 {
+					 for (gfxindex=0;gfxindex<numtriangles;gfxindex++)
+					 {
+						 unsigned char* tri_indices= (unsigned char*)(indexbase+gfxindex*indexstride);
+						 graphicsbase = (float*)(vertexbase+tri_indices[0]*stride);
+						 triangle[0].setValue(graphicsbase[0]*meshScaling.getX(),graphicsbase[1]*meshScaling.getY(),graphicsbase[2]*meshScaling.getZ());
+						 graphicsbase = (float*)(vertexbase+tri_indices[1]*stride);
+						 triangle[1].setValue(graphicsbase[0]*meshScaling.getX(),graphicsbase[1]*meshScaling.getY(),	graphicsbase[2]*meshScaling.getZ());
+						 graphicsbase = (float*)(vertexbase+tri_indices[2]*stride);
+						 triangle[2].setValue(graphicsbase[0]*meshScaling.getX(),graphicsbase[1]*meshScaling.getY(),	graphicsbase[2]*meshScaling.getZ());
+						 callback->internalProcessTriangleIndex(triangle,part,gfxindex);
+					 }
+					 break;
+				 }
+			 default:
+				 btAssert((gfxindextype == PHY_INTEGER) || (gfxindextype == PHY_SHORT));
+			 }
+			 break;
+		 }
+
+		case PHY_DOUBLE:
+			{
+				double* graphicsbase;
+
+				switch (gfxindextype)
+				{
+				case PHY_INTEGER:
+					{
+						for (gfxindex=0;gfxindex<numtriangles;gfxindex++)
+						{
+							unsigned int* tri_indices= (unsigned int*)(indexbase+gfxindex*indexstride);
+							graphicsbase = (double*)(vertexbase+tri_indices[0]*stride);
+							triangle[0].setValue((btScalar)graphicsbase[0]*meshScaling.getX(),(btScalar)graphicsbase[1]*meshScaling.getY(),(btScalar)graphicsbase[2]*meshScaling.getZ());
+							graphicsbase = (double*)(vertexbase+tri_indices[1]*stride);
+							triangle[1].setValue((btScalar)graphicsbase[0]*meshScaling.getX(),(btScalar)graphicsbase[1]*meshScaling.getY(),  (btScalar)graphicsbase[2]*meshScaling.getZ());
+							graphicsbase = (double*)(vertexbase+tri_indices[2]*stride);
+							triangle[2].setValue((btScalar)graphicsbase[0]*meshScaling.getX(),(btScalar)graphicsbase[1]*meshScaling.getY(),  (btScalar)graphicsbase[2]*meshScaling.getZ());
+							callback->internalProcessTriangleIndex(triangle,part,gfxindex);
+						}
+						break;
+					}
+				case PHY_SHORT:
+					{
+						for (gfxindex=0;gfxindex<numtriangles;gfxindex++)
+						{
+							unsigned short int* tri_indices= (unsigned short int*)(indexbase+gfxindex*indexstride);
+							graphicsbase = (double*)(vertexbase+tri_indices[0]*stride);
+							triangle[0].setValue((btScalar)graphicsbase[0]*meshScaling.getX(),(btScalar)graphicsbase[1]*meshScaling.getY(),(btScalar)graphicsbase[2]*meshScaling.getZ());
+							graphicsbase = (double*)(vertexbase+tri_indices[1]*stride);
+							triangle[1].setValue((btScalar)graphicsbase[0]*meshScaling.getX(),(btScalar)graphicsbase[1]*meshScaling.getY(),  (btScalar)graphicsbase[2]*meshScaling.getZ());
+							graphicsbase = (double*)(vertexbase+tri_indices[2]*stride);
+							triangle[2].setValue((btScalar)graphicsbase[0]*meshScaling.getX(),(btScalar)graphicsbase[1]*meshScaling.getY(),  (btScalar)graphicsbase[2]*meshScaling.getZ());
+							callback->internalProcessTriangleIndex(triangle,part,gfxindex);
+						}
+						break;
+					}
+				case PHY_UCHAR:
+					{
+						for (gfxindex=0;gfxindex<numtriangles;gfxindex++)
+						{
+							unsigned char* tri_indices= (unsigned char*)(indexbase+gfxindex*indexstride);
+							graphicsbase = (double*)(vertexbase+tri_indices[0]*stride);
+							triangle[0].setValue((btScalar)graphicsbase[0]*meshScaling.getX(),(btScalar)graphicsbase[1]*meshScaling.getY(),(btScalar)graphicsbase[2]*meshScaling.getZ());
+							graphicsbase = (double*)(vertexbase+tri_indices[1]*stride);
+							triangle[1].setValue((btScalar)graphicsbase[0]*meshScaling.getX(),(btScalar)graphicsbase[1]*meshScaling.getY(),  (btScalar)graphicsbase[2]*meshScaling.getZ());
+							graphicsbase = (double*)(vertexbase+tri_indices[2]*stride);
+							triangle[2].setValue((btScalar)graphicsbase[0]*meshScaling.getX(),(btScalar)graphicsbase[1]*meshScaling.getY(),  (btScalar)graphicsbase[2]*meshScaling.getZ());
+							callback->internalProcessTriangleIndex(triangle,part,gfxindex);
+						}
+						break;
+					}
+				default:
+					btAssert((gfxindextype == PHY_INTEGER) || (gfxindextype == PHY_SHORT));
+				}
+				break;
+			}
+		default:
+			btAssert((type == PHY_FLOAT) || (type == PHY_DOUBLE));
+		}
+
+		unLockReadOnlyVertexBase(part);
+	}
+}
+
+void	btStridingMeshInterface::calculateAabbBruteForce(btVector3& aabbMin,btVector3& aabbMax)
+{
+
+	struct	AabbCalculationCallback : public btInternalTriangleIndexCallback
+	{
+		btVector3	m_aabbMin;
+		btVector3	m_aabbMax;
+
+		AabbCalculationCallback()
+		{
+			m_aabbMin.setValue(btScalar(BT_LARGE_FLOAT),btScalar(BT_LARGE_FLOAT),btScalar(BT_LARGE_FLOAT));
+			m_aabbMax.setValue(btScalar(-BT_LARGE_FLOAT),btScalar(-BT_LARGE_FLOAT),btScalar(-BT_LARGE_FLOAT));
+		}
+
+		virtual void internalProcessTriangleIndex(btVector3* triangle,int partId,int  triangleIndex)
+		{
+			(void)partId;
+			(void)triangleIndex;
+
+			m_aabbMin.setMin(triangle[0]);
+			m_aabbMax.setMax(triangle[0]);
+			m_aabbMin.setMin(triangle[1]);
+			m_aabbMax.setMax(triangle[1]);
+			m_aabbMin.setMin(triangle[2]);
+			m_aabbMax.setMax(triangle[2]);
+		}
+	};
+
+	//first calculate the total aabb for all triangles
+	AabbCalculationCallback	aabbCallback;
+	aabbMin.setValue(btScalar(-BT_LARGE_FLOAT),btScalar(-BT_LARGE_FLOAT),btScalar(-BT_LARGE_FLOAT));
+	aabbMax.setValue(btScalar(BT_LARGE_FLOAT),btScalar(BT_LARGE_FLOAT),btScalar(BT_LARGE_FLOAT));
+	InternalProcessAllTriangles(&aabbCallback,aabbMin,aabbMax);
+
+	aabbMin = aabbCallback.m_aabbMin;
+	aabbMax = aabbCallback.m_aabbMax;
+}
+
+

opencl/gpu_narrowphase/host/btStridingMeshInterface.h

@@ -0,0 +1,167 @@
+/*
+Bullet Continuous Collision Detection and Physics Library
+Copyright (c) 2003-2009 Erwin Coumans  http://bulletphysics.org
+
+This software is provided 'as-is', without any express or implied warranty.
+In no event will the authors be held liable for any damages arising from the use of this software.
+Permission is granted to anyone to use this software for any purpose, 
+including commercial applications, and to alter it and redistribute it freely, 
+subject to the following restrictions:
+
+1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
+2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
+3. This notice may not be removed or altered from any source distribution.
+*/
+
+#ifndef BT_STRIDING_MESHINTERFACE_H
+#define BT_STRIDING_MESHINTERFACE_H
+
+#include "BulletCommon/btVector3.h"
+#include "btTriangleCallback.h"
+//#include "btConcaveShape.h"
+
+
+enum  	PHY_ScalarType { 
+  PHY_FLOAT, PHY_DOUBLE, PHY_INTEGER, PHY_SHORT, 
+  PHY_FIXEDPOINT88, PHY_UCHAR 
+};
+
+
+///	The btStridingMeshInterface is the interface class for high performance generic access to triangle meshes, used in combination with btBvhTriangleMeshShape and some other collision shapes.
+/// Using index striding of 3*sizeof(integer) it can use triangle arrays, using index striding of 1*sizeof(integer) it can handle triangle strips.
+/// It allows for sharing graphics and collision meshes. Also it provides locking/unlocking of graphics meshes that are in gpu memory.
+ATTRIBUTE_ALIGNED16(class ) btStridingMeshInterface
+{
+	protected:
+	
+		btVector3 m_scaling;
+
+	public:
+		BT_DECLARE_ALIGNED_ALLOCATOR();
+		
+		btStridingMeshInterface() :m_scaling(btScalar(1.),btScalar(1.),btScalar(1.))
+		{
+
+		}
+
+		virtual ~btStridingMeshInterface();
+
+
+
+		virtual void	InternalProcessAllTriangles(btInternalTriangleIndexCallback* callback,const btVector3& aabbMin,const btVector3& aabbMax) const;
+
+		///brute force method to calculate aabb
+		void	calculateAabbBruteForce(btVector3& aabbMin,btVector3& aabbMax);
+
+		/// get read and write access to a subpart of a triangle mesh
+		/// this subpart has a continuous array of vertices and indices
+		/// in this way the mesh can be handled as chunks of memory with striding
+		/// very similar to OpenGL vertexarray support
+		/// make a call to unLockVertexBase when the read and write access is finished	
+		virtual void	getLockedVertexIndexBase(unsigned char **vertexbase, int& numverts,PHY_ScalarType& type, int& stride,unsigned char **indexbase,int & indexstride,int& numfaces,PHY_ScalarType& indicestype,int subpart=0)=0;
+		
+		virtual void	getLockedReadOnlyVertexIndexBase(const unsigned char **vertexbase, int& numverts,PHY_ScalarType& type, int& stride,const unsigned char **indexbase,int & indexstride,int& numfaces,PHY_ScalarType& indicestype,int subpart=0) const=0;
+	
+		/// unLockVertexBase finishes the access to a subpart of the triangle mesh
+		/// make a call to unLockVertexBase when the read and write access (using getLockedVertexIndexBase) is finished
+		virtual void	unLockVertexBase(int subpart)=0;
+
+		virtual void	unLockReadOnlyVertexBase(int subpart) const=0;
+
+
+		/// getNumSubParts returns the number of seperate subparts
+		/// each subpart has a continuous array of vertices and indices
+		virtual int		getNumSubParts() const=0;
+
+		virtual void	preallocateVertices(int numverts)=0;
+		virtual void	preallocateIndices(int numindices)=0;
+
+		virtual bool	hasPremadeAabb() const { return false; }
+		virtual void	setPremadeAabb(const btVector3& aabbMin, const btVector3& aabbMax ) const
+                {
+                        (void) aabbMin;
+                        (void) aabbMax;
+                }
+		virtual void	getPremadeAabb(btVector3* aabbMin, btVector3* aabbMax ) const
+        {
+            (void) aabbMin;
+            (void) aabbMax;
+        }
+
+		const btVector3&	getScaling() const {
+			return m_scaling;
+		}
+		void	setScaling(const btVector3& scaling)
+		{
+			m_scaling = scaling;
+		}
+
+		virtual	int	calculateSerializeBufferSize() const;
+
+		///fills the dataBuffer and returns the struct name (and 0 on failure)
+		//virtual	const char*	serialize(void* dataBuffer, btSerializer* serializer) const;
+
+
+};
+
+struct	btIntIndexData
+{
+	int	m_value;
+};
+
+struct	btShortIntIndexData
+{
+	short m_value;
+	char m_pad[2];
+};
+
+struct	btShortIntIndexTripletData
+{
+	short	m_values[3];
+	char	m_pad[2];
+};
+
+struct	btCharIndexTripletData
+{
+	unsigned char m_values[3];
+	char	m_pad;
+};
+
+
+///do not change those serialization structures, it requires an updated sBulletDNAstr/sBulletDNAstr64
+struct	btMeshPartData
+{
+	btVector3FloatData			*m_vertices3f;
+	btVector3DoubleData			*m_vertices3d;
+
+	btIntIndexData				*m_indices32;
+	btShortIntIndexTripletData	*m_3indices16;
+	btCharIndexTripletData		*m_3indices8;
+
+	btShortIntIndexData			*m_indices16;//backwards compatibility
+
+	int                     m_numTriangles;//length of m_indices = m_numTriangles
+	int                     m_numVertices;
+};
+
+
+///do not change those serialization structures, it requires an updated sBulletDNAstr/sBulletDNAstr64
+struct	btStridingMeshInterfaceData
+{
+	btMeshPartData	*m_meshPartsPtr;
+	btVector3FloatData	m_scaling;
+	int	m_numMeshParts;
+	char m_padding[4];
+};
+
+
+
+
+SIMD_FORCE_INLINE	int	btStridingMeshInterface::calculateSerializeBufferSize() const
+{
+	return sizeof(btStridingMeshInterfaceData);
+}
+
+
+
+#endif //BT_STRIDING_MESHINTERFACE_H

opencl/gpu_narrowphase/host/btTriangleCallback.cpp

@@ -0,0 +1,28 @@
+/*
+Bullet Continuous Collision Detection and Physics Library
+Copyright (c) 2003-2009 Erwin Coumans  http://bulletphysics.org
+
+This software is provided 'as-is', without any express or implied warranty.
+In no event will the authors be held liable for any damages arising from the use of this software.
+Permission is granted to anyone to use this software for any purpose, 
+including commercial applications, and to alter it and redistribute it freely, 
+subject to the following restrictions:
+
+1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
+2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
+3. This notice may not be removed or altered from any source distribution.
+*/
+
+#include "btTriangleCallback.h"
+
+btTriangleCallback::~btTriangleCallback()
+{
+
+}
+
+
+btInternalTriangleIndexCallback::~btInternalTriangleIndexCallback()
+{
+
+}
+

opencl/gpu_narrowphase/host/btTriangleCallback.h

@@ -0,0 +1,42 @@
+/*
+Bullet Continuous Collision Detection and Physics Library
+Copyright (c) 2003-2009 Erwin Coumans  http://bulletphysics.org
+
+This software is provided 'as-is', without any express or implied warranty.
+In no event will the authors be held liable for any damages arising from the use of this software.
+Permission is granted to anyone to use this software for any purpose, 
+including commercial applications, and to alter it and redistribute it freely, 
+subject to the following restrictions:
+
+1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
+2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
+3. This notice may not be removed or altered from any source distribution.
+*/
+
+#ifndef BT_TRIANGLE_CALLBACK_H
+#define BT_TRIANGLE_CALLBACK_H
+
+#include "BulletCommon/btVector3.h"
+
+
+///The btTriangleCallback provides a callback for each overlapping triangle when calling processAllTriangles.
+///This callback is called by processAllTriangles for all btConcaveShape derived class, such as  btBvhTriangleMeshShape, btStaticPlaneShape and btHeightfieldTerrainShape.
+class btTriangleCallback
+{
+public:
+
+	virtual ~btTriangleCallback();
+	virtual void processTriangle(btVector3* triangle, int partId, int triangleIndex) = 0;
+};
+
+class btInternalTriangleIndexCallback
+{
+public:
+
+	virtual ~btInternalTriangleIndexCallback();
+	virtual void internalProcessTriangleIndex(btVector3* triangle,int partId,int  triangleIndex) = 0;
+};
+
+
+
+#endif //BT_TRIANGLE_CALLBACK_H

opencl/gpu_narrowphase/host/btTriangleIndexVertexArray.cpp

@@ -0,0 +1,95 @@
+/*
+Bullet Continuous Collision Detection and Physics Library
+Copyright (c) 2003-2009 Erwin Coumans  http://bulletphysics.org
+
+This software is provided 'as-is', without any express or implied warranty.
+In no event will the authors be held liable for any damages arising from the use of this software.
+Permission is granted to anyone to use this software for any purpose, 
+including commercial applications, and to alter it and redistribute it freely, 
+subject to the following restrictions:
+
+1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
+2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
+3. This notice may not be removed or altered from any source distribution.
+*/
+
+#include "btTriangleIndexVertexArray.h"
+
+btTriangleIndexVertexArray::btTriangleIndexVertexArray(int numTriangles,int* triangleIndexBase,int triangleIndexStride,int numVertices,btScalar* vertexBase,int vertexStride)
+: m_hasAabb(0)
+{
+	btIndexedMesh mesh;
+
+	mesh.m_numTriangles = numTriangles;
+	mesh.m_triangleIndexBase = (const unsigned char *)triangleIndexBase;
+	mesh.m_triangleIndexStride = triangleIndexStride;
+	mesh.m_numVertices = numVertices;
+	mesh.m_vertexBase = (const unsigned char *)vertexBase;
+	mesh.m_vertexStride = vertexStride;
+
+	addIndexedMesh(mesh);
+
+}
+
+btTriangleIndexVertexArray::~btTriangleIndexVertexArray()
+{
+
+}
+
+void	btTriangleIndexVertexArray::getLockedVertexIndexBase(unsigned char **vertexbase, int& numverts,PHY_ScalarType& type, int& vertexStride,unsigned char **indexbase,int & indexstride,int& numfaces,PHY_ScalarType& indicestype,int subpart)
+{
+	btAssert(subpart< getNumSubParts() );
+
+	btIndexedMesh& mesh = m_indexedMeshes[subpart];
+
+	numverts = mesh.m_numVertices;
+	(*vertexbase) = (unsigned char *) mesh.m_vertexBase;
+
+   type = mesh.m_vertexType;
+
+	vertexStride = mesh.m_vertexStride;
+
+	numfaces = mesh.m_numTriangles;
+
+	(*indexbase) = (unsigned char *)mesh.m_triangleIndexBase;
+	indexstride = mesh.m_triangleIndexStride;
+	indicestype = mesh.m_indexType;
+}
+
+void	btTriangleIndexVertexArray::getLockedReadOnlyVertexIndexBase(const unsigned char **vertexbase, int& numverts,PHY_ScalarType& type, int& vertexStride,const unsigned char **indexbase,int & indexstride,int& numfaces,PHY_ScalarType& indicestype,int subpart) const
+{
+	const btIndexedMesh& mesh = m_indexedMeshes[subpart];
+
+	numverts = mesh.m_numVertices;
+	(*vertexbase) = (const unsigned char *)mesh.m_vertexBase;
+
+   type = mesh.m_vertexType;
+   
+	vertexStride = mesh.m_vertexStride;
+
+	numfaces = mesh.m_numTriangles;
+	(*indexbase) = (const unsigned char *)mesh.m_triangleIndexBase;
+	indexstride = mesh.m_triangleIndexStride;
+	indicestype = mesh.m_indexType;
+}
+
+bool	btTriangleIndexVertexArray::hasPremadeAabb() const
+{
+	return (m_hasAabb == 1);
+}
+
+
+void	btTriangleIndexVertexArray::setPremadeAabb(const btVector3& aabbMin, const btVector3& aabbMax ) const
+{
+	m_aabbMin = aabbMin;
+	m_aabbMax = aabbMax;
+	m_hasAabb = 1; // this is intentionally an int see notes in header
+}
+
+void	btTriangleIndexVertexArray::getPremadeAabb(btVector3* aabbMin, btVector3* aabbMax ) const
+{
+	*aabbMin = m_aabbMin;
+	*aabbMax = m_aabbMax;
+}
+
+

opencl/gpu_narrowphase/host/btTriangleIndexVertexArray.h

@@ -0,0 +1,133 @@
+/*
+Bullet Continuous Collision Detection and Physics Library
+Copyright (c) 2003-2009 Erwin Coumans  http://bulletphysics.org
+
+This software is provided 'as-is', without any express or implied warranty.
+In no event will the authors be held liable for any damages arising from the use of this software.
+Permission is granted to anyone to use this software for any purpose, 
+including commercial applications, and to alter it and redistribute it freely, 
+subject to the following restrictions:
+
+1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
+2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
+3. This notice may not be removed or altered from any source distribution.
+*/
+
+#ifndef BT_TRIANGLE_INDEX_VERTEX_ARRAY_H
+#define BT_TRIANGLE_INDEX_VERTEX_ARRAY_H
+
+#include "btStridingMeshInterface.h"
+#include "BulletCommon/btAlignedObjectArray.h"
+#include "BulletCommon/btScalar.h"
+
+
+///The btIndexedMesh indexes a single vertex and index array. Multiple btIndexedMesh objects can be passed into a btTriangleIndexVertexArray using addIndexedMesh.
+///Instead of the number of indices, we pass the number of triangles.
+ATTRIBUTE_ALIGNED16( struct)	btIndexedMesh
+{
+	BT_DECLARE_ALIGNED_ALLOCATOR();
+
+   int                     m_numTriangles;
+   const unsigned char *   m_triangleIndexBase;
+   // Size in byte of the indices for one triangle (3*sizeof(index_type) if the indices are tightly packed)
+   int                     m_triangleIndexStride;
+   int                     m_numVertices;
+   const unsigned char *   m_vertexBase;
+   // Size of a vertex, in bytes
+   int                     m_vertexStride;
+
+   // The index type is set when adding an indexed mesh to the
+   // btTriangleIndexVertexArray, do not set it manually
+   PHY_ScalarType m_indexType;
+
+   // The vertex type has a default type similar to Bullet's precision mode (float or double)
+   // but can be set manually if you for example run Bullet with double precision but have
+   // mesh data in single precision..
+   PHY_ScalarType m_vertexType;
+
+
+   btIndexedMesh()
+	   :m_indexType(PHY_INTEGER),
+#ifdef BT_USE_DOUBLE_PRECISION
+      m_vertexType(PHY_DOUBLE)
+#else // BT_USE_DOUBLE_PRECISION
+      m_vertexType(PHY_FLOAT)
+#endif // BT_USE_DOUBLE_PRECISION
+      {
+      }
+}
+;
+
+
+typedef btAlignedObjectArray<btIndexedMesh>	IndexedMeshArray;
+
+///The btTriangleIndexVertexArray allows to access multiple triangle meshes, by indexing into existing triangle/index arrays.
+///Additional meshes can be added using addIndexedMesh
+///No duplcate is made of the vertex/index data, it only indexes into external vertex/index arrays.
+///So keep those arrays around during the lifetime of this btTriangleIndexVertexArray.
+ATTRIBUTE_ALIGNED16( class) btTriangleIndexVertexArray : public btStridingMeshInterface
+{
+protected:
+	IndexedMeshArray	m_indexedMeshes;
+	int m_pad[2];
+	mutable int m_hasAabb; // using int instead of bool to maintain alignment
+	mutable btVector3 m_aabbMin;
+	mutable btVector3 m_aabbMax;
+
+public:
+
+	BT_DECLARE_ALIGNED_ALLOCATOR();
+
+	btTriangleIndexVertexArray() : m_hasAabb(0)
+	{
+	}
+
+	virtual ~btTriangleIndexVertexArray();
+
+	//just to be backwards compatible
+	btTriangleIndexVertexArray(int numTriangles,int* triangleIndexBase,int triangleIndexStride,int numVertices,btScalar* vertexBase,int vertexStride);
+	
+	void	addIndexedMesh(const btIndexedMesh& mesh, PHY_ScalarType indexType = PHY_INTEGER)
+	{
+		m_indexedMeshes.push_back(mesh);
+		m_indexedMeshes[m_indexedMeshes.size()-1].m_indexType = indexType;
+	}
+	
+	
+	virtual void	getLockedVertexIndexBase(unsigned char **vertexbase, int& numverts,PHY_ScalarType& type, int& vertexStride,unsigned char **indexbase,int & indexstride,int& numfaces,PHY_ScalarType& indicestype,int subpart=0);
+
+	virtual void	getLockedReadOnlyVertexIndexBase(const unsigned char **vertexbase, int& numverts,PHY_ScalarType& type, int& vertexStride,const unsigned char **indexbase,int & indexstride,int& numfaces,PHY_ScalarType& indicestype,int subpart=0) const;
+
+	/// unLockVertexBase finishes the access to a subpart of the triangle mesh
+	/// make a call to unLockVertexBase when the read and write access (using getLockedVertexIndexBase) is finished
+	virtual void	unLockVertexBase(int subpart) {(void)subpart;}
+
+	virtual void	unLockReadOnlyVertexBase(int subpart) const {(void)subpart;}
+
+	/// getNumSubParts returns the number of seperate subparts
+	/// each subpart has a continuous array of vertices and indices
+	virtual int		getNumSubParts() const { 
+		return (int)m_indexedMeshes.size();
+	}
+
+	IndexedMeshArray&	getIndexedMeshArray()
+	{
+		return m_indexedMeshes;
+	}
+
+	const IndexedMeshArray&	getIndexedMeshArray() const
+	{
+		return m_indexedMeshes;
+	}
+
+	virtual void	preallocateVertices(int numverts){(void) numverts;}
+	virtual void	preallocateIndices(int numindices){(void) numindices;}
+
+	virtual bool	hasPremadeAabb() const;
+	virtual void	setPremadeAabb(const btVector3& aabbMin, const btVector3& aabbMax ) const;
+	virtual void	getPremadeAabb(btVector3* aabbMin, btVector3* aabbMax ) const;
+
+}
+;
+
+#endif //BT_TRIANGLE_INDEX_VERTEX_ARRAY_H