summaryrefslogtreecommitdiff
path: root/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels
diff options
context:
space:
mode:
Diffstat (limited to 'thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels')
-rw-r--r--thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/bvhTraversal.cl283
-rw-r--r--thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/bvhTraversal.h258
-rw-r--r--thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/mpr.cl311
-rw-r--r--thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/mprKernels.h1446
-rw-r--r--thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/primitiveContacts.cl1374
-rw-r--r--thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/primitiveContacts.h1289
-rw-r--r--thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/sat.cl2018
-rw-r--r--thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/satClipHullContacts.cl1888
-rw-r--r--thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/satClipHullContacts.h2099
-rw-r--r--thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/satConcave.cl1220
-rw-r--r--thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/satConcaveKernels.h1457
-rw-r--r--thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/satKernels.h2104
12 files changed, 15747 insertions, 0 deletions
diff --git a/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/bvhTraversal.cl b/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/bvhTraversal.cl
new file mode 100644
index 0000000000..faa413441c
--- /dev/null
+++ b/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/bvhTraversal.cl
@@ -0,0 +1,283 @@
+//keep this enum in sync with the CPU version (in btCollidable.h)
+//written by Erwin Coumans
+
+#define SHAPE_CONVEX_HULL 3
+#define SHAPE_CONCAVE_TRIMESH 5
+#define TRIANGLE_NUM_CONVEX_FACES 5
+#define SHAPE_COMPOUND_OF_CONVEX_HULLS 6
+#define SHAPE_SPHERE 7
+
+typedef unsigned int u32;
+
+#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).
+typedef struct
+{
+ //12 bytes
+ unsigned short int m_quantizedAabbMin[3];
+ unsigned short int m_quantizedAabbMax[3];
+ //4 bytes
+ int m_escapeIndexOrTriangleIndex;
+} btQuantizedBvhNode;
+
+typedef struct
+{
+ float4 m_aabbMin;
+ float4 m_aabbMax;
+ float4 m_quantization;
+ int m_numNodes;
+ int m_numSubTrees;
+ int m_nodeOffset;
+ int m_subTreeOffset;
+
+} b3BvhInfo;
+
+int getTriangleIndex(const btQuantizedBvhNode* rootNode)
+{
+ 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 (rootNode->m_escapeIndexOrTriangleIndex&~(y));
+}
+
+int isLeaf(const btQuantizedBvhNode* rootNode)
+{
+ //skipindex is negative (internal node), triangleindex >=0 (leafnode)
+ return (rootNode->m_escapeIndexOrTriangleIndex >= 0)? 1 : 0;
+}
+
+int getEscapeIndex(const btQuantizedBvhNode* rootNode)
+{
+ return -rootNode->m_escapeIndexOrTriangleIndex;
+}
+
+typedef struct
+{
+ //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;
+
+///keep this in sync with btCollidable.h
+typedef struct
+{
+ int m_numChildShapes;
+ int blaat2;
+ int m_shapeType;
+ int m_shapeIndex;
+
+} btCollidableGpu;
+
+typedef struct
+{
+ float4 m_childPosition;
+ float4 m_childOrientation;
+ int m_shapeIndex;
+ int m_unused0;
+ int m_unused1;
+ int m_unused2;
+} btGpuChildShape;
+
+
+typedef struct
+{
+ float4 m_pos;
+ float4 m_quat;
+ float4 m_linVel;
+ float4 m_angVel;
+
+ u32 m_collidableIdx;
+ float m_invMass;
+ float m_restituitionCoeff;
+ float m_frictionCoeff;
+} BodyData;
+
+typedef struct
+{
+ union
+ {
+ float4 m_min;
+ float m_minElems[4];
+ int m_minIndices[4];
+ };
+ union
+ {
+ float4 m_max;
+ float m_maxElems[4];
+ int m_maxIndices[4];
+ };
+} btAabbCL;
+
+
+int testQuantizedAabbAgainstQuantizedAabb(
+ const unsigned short int* aabbMin1,
+ const unsigned short int* aabbMax1,
+ const unsigned short int* aabbMin2,
+ const unsigned short int* aabbMax2)
+{
+ //int overlap = 1;
+ if (aabbMin1[0] > aabbMax2[0])
+ return 0;
+ if (aabbMax1[0] < aabbMin2[0])
+ return 0;
+ if (aabbMin1[1] > aabbMax2[1])
+ return 0;
+ if (aabbMax1[1] < aabbMin2[1])
+ return 0;
+ if (aabbMin1[2] > aabbMax2[2])
+ return 0;
+ if (aabbMax1[2] < aabbMin2[2])
+ return 0;
+ return 1;
+ //overlap = ((aabbMin1[0] > aabbMax2[0]) || (aabbMax1[0] < aabbMin2[0])) ? 0 : overlap;
+ //overlap = ((aabbMin1[2] > aabbMax2[2]) || (aabbMax1[2] < aabbMin2[2])) ? 0 : overlap;
+ //overlap = ((aabbMin1[1] > aabbMax2[1]) || (aabbMax1[1] < aabbMin2[1])) ? 0 : overlap;
+ //return overlap;
+}
+
+
+void quantizeWithClamp(unsigned short* out, float4 point2,int isMax, float4 bvhAabbMin, float4 bvhAabbMax, float4 bvhQuantization)
+{
+ float4 clampedPoint = max(point2,bvhAabbMin);
+ clampedPoint = min (clampedPoint, bvhAabbMax);
+
+ float4 v = (clampedPoint - bvhAabbMin) * bvhQuantization;
+ if (isMax)
+ {
+ out[0] = (unsigned short) (((unsigned short)(v.x+1.f) | 1));
+ out[1] = (unsigned short) (((unsigned short)(v.y+1.f) | 1));
+ out[2] = (unsigned short) (((unsigned short)(v.z+1.f) | 1));
+ } else
+ {
+ out[0] = (unsigned short) (((unsigned short)(v.x) & 0xfffe));
+ out[1] = (unsigned short) (((unsigned short)(v.y) & 0xfffe));
+ out[2] = (unsigned short) (((unsigned short)(v.z) & 0xfffe));
+ }
+
+}
+
+
+// work-in-progress
+__kernel void bvhTraversalKernel( __global const int4* pairs,
+ __global const BodyData* rigidBodies,
+ __global const btCollidableGpu* collidables,
+ __global btAabbCL* aabbs,
+ __global int4* concavePairsOut,
+ __global volatile int* numConcavePairsOut,
+ __global const btBvhSubtreeInfo* subtreeHeadersRoot,
+ __global const btQuantizedBvhNode* quantizedNodesRoot,
+ __global const b3BvhInfo* bvhInfos,
+ int numPairs,
+ int maxNumConcavePairsCapacity)
+{
+ int id = get_global_id(0);
+ if (id>=numPairs)
+ return;
+
+ int bodyIndexA = pairs[id].x;
+ int bodyIndexB = pairs[id].y;
+ int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;
+ int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;
+
+ //once the broadphase avoids static-static pairs, we can remove this test
+ if ((rigidBodies[bodyIndexA].m_invMass==0) &&(rigidBodies[bodyIndexB].m_invMass==0))
+ {
+ return;
+ }
+
+ if (collidables[collidableIndexA].m_shapeType!=SHAPE_CONCAVE_TRIMESH)
+ return;
+
+ int shapeTypeB = collidables[collidableIndexB].m_shapeType;
+
+ if (shapeTypeB!=SHAPE_CONVEX_HULL &&
+ shapeTypeB!=SHAPE_SPHERE &&
+ shapeTypeB!=SHAPE_COMPOUND_OF_CONVEX_HULLS
+ )
+ return;
+
+ b3BvhInfo bvhInfo = bvhInfos[collidables[collidableIndexA].m_numChildShapes];
+
+ float4 bvhAabbMin = bvhInfo.m_aabbMin;
+ float4 bvhAabbMax = bvhInfo.m_aabbMax;
+ float4 bvhQuantization = bvhInfo.m_quantization;
+ int numSubtreeHeaders = bvhInfo.m_numSubTrees;
+ __global const btBvhSubtreeInfo* subtreeHeaders = &subtreeHeadersRoot[bvhInfo.m_subTreeOffset];
+ __global const btQuantizedBvhNode* quantizedNodes = &quantizedNodesRoot[bvhInfo.m_nodeOffset];
+
+
+ unsigned short int quantizedQueryAabbMin[3];
+ unsigned short int quantizedQueryAabbMax[3];
+ quantizeWithClamp(quantizedQueryAabbMin,aabbs[bodyIndexB].m_min,false,bvhAabbMin, bvhAabbMax,bvhQuantization);
+ quantizeWithClamp(quantizedQueryAabbMax,aabbs[bodyIndexB].m_max,true ,bvhAabbMin, bvhAabbMax,bvhQuantization);
+
+ for (int i=0;i<numSubtreeHeaders;i++)
+ {
+ btBvhSubtreeInfo subtree = subtreeHeaders[i];
+
+ int overlap = testQuantizedAabbAgainstQuantizedAabb(quantizedQueryAabbMin,quantizedQueryAabbMax,subtree.m_quantizedAabbMin,subtree.m_quantizedAabbMax);
+ if (overlap != 0)
+ {
+ int startNodeIndex = subtree.m_rootNodeIndex;
+ int endNodeIndex = subtree.m_rootNodeIndex+subtree.m_subtreeSize;
+ int curIndex = startNodeIndex;
+ int escapeIndex;
+ int isLeafNode;
+ int aabbOverlap;
+ while (curIndex < endNodeIndex)
+ {
+ btQuantizedBvhNode rootNode = quantizedNodes[curIndex];
+ aabbOverlap = testQuantizedAabbAgainstQuantizedAabb(quantizedQueryAabbMin,quantizedQueryAabbMax,rootNode.m_quantizedAabbMin,rootNode.m_quantizedAabbMax);
+ isLeafNode = isLeaf(&rootNode);
+ if (aabbOverlap)
+ {
+ if (isLeafNode)
+ {
+ int triangleIndex = getTriangleIndex(&rootNode);
+ if (shapeTypeB==SHAPE_COMPOUND_OF_CONVEX_HULLS)
+ {
+ int numChildrenB = collidables[collidableIndexB].m_numChildShapes;
+ int pairIdx = atomic_add(numConcavePairsOut,numChildrenB);
+ for (int b=0;b<numChildrenB;b++)
+ {
+ if ((pairIdx+b)<maxNumConcavePairsCapacity)
+ {
+ int childShapeIndexB = collidables[collidableIndexB].m_shapeIndex+b;
+ int4 newPair = (int4)(bodyIndexA,bodyIndexB,triangleIndex,childShapeIndexB);
+ concavePairsOut[pairIdx+b] = newPair;
+ }
+ }
+ } else
+ {
+ int pairIdx = atomic_inc(numConcavePairsOut);
+ if (pairIdx<maxNumConcavePairsCapacity)
+ {
+ int4 newPair = (int4)(bodyIndexA,bodyIndexB,triangleIndex,0);
+ concavePairsOut[pairIdx] = newPair;
+ }
+ }
+ }
+ curIndex++;
+ } else
+ {
+ if (isLeafNode)
+ {
+ curIndex++;
+ } else
+ {
+ escapeIndex = getEscapeIndex(&rootNode);
+ curIndex += escapeIndex;
+ }
+ }
+ }
+ }
+ }
+
+} \ No newline at end of file
diff --git a/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/bvhTraversal.h b/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/bvhTraversal.h
new file mode 100644
index 0000000000..4b3b49eae8
--- /dev/null
+++ b/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/bvhTraversal.h
@@ -0,0 +1,258 @@
+//this file is autogenerated using stringify.bat (premake --stringify) in the build folder of this project
+static const char* bvhTraversalKernelCL= \
+"//keep this enum in sync with the CPU version (in btCollidable.h)\n"
+"//written by Erwin Coumans\n"
+"#define SHAPE_CONVEX_HULL 3\n"
+"#define SHAPE_CONCAVE_TRIMESH 5\n"
+"#define TRIANGLE_NUM_CONVEX_FACES 5\n"
+"#define SHAPE_COMPOUND_OF_CONVEX_HULLS 6\n"
+"#define SHAPE_SPHERE 7\n"
+"typedef unsigned int u32;\n"
+"#define MAX_NUM_PARTS_IN_BITS 10\n"
+"///btQuantizedBvhNode is a compressed aabb node, 16 bytes.\n"
+"///Node can be used for leafnode or internal node. Leafnodes can point to 32-bit triangle index (non-negative range).\n"
+"typedef struct\n"
+"{\n"
+" //12 bytes\n"
+" unsigned short int m_quantizedAabbMin[3];\n"
+" unsigned short int m_quantizedAabbMax[3];\n"
+" //4 bytes\n"
+" int m_escapeIndexOrTriangleIndex;\n"
+"} btQuantizedBvhNode;\n"
+"typedef struct\n"
+"{\n"
+" float4 m_aabbMin;\n"
+" float4 m_aabbMax;\n"
+" float4 m_quantization;\n"
+" int m_numNodes;\n"
+" int m_numSubTrees;\n"
+" int m_nodeOffset;\n"
+" int m_subTreeOffset;\n"
+"} b3BvhInfo;\n"
+"int getTriangleIndex(const btQuantizedBvhNode* rootNode)\n"
+"{\n"
+" unsigned int x=0;\n"
+" unsigned int y = (~(x&0))<<(31-MAX_NUM_PARTS_IN_BITS);\n"
+" // Get only the lower bits where the triangle index is stored\n"
+" return (rootNode->m_escapeIndexOrTriangleIndex&~(y));\n"
+"}\n"
+"int isLeaf(const btQuantizedBvhNode* rootNode)\n"
+"{\n"
+" //skipindex is negative (internal node), triangleindex >=0 (leafnode)\n"
+" return (rootNode->m_escapeIndexOrTriangleIndex >= 0)? 1 : 0;\n"
+"}\n"
+" \n"
+"int getEscapeIndex(const btQuantizedBvhNode* rootNode)\n"
+"{\n"
+" return -rootNode->m_escapeIndexOrTriangleIndex;\n"
+"}\n"
+"typedef struct\n"
+"{\n"
+" //12 bytes\n"
+" unsigned short int m_quantizedAabbMin[3];\n"
+" unsigned short int m_quantizedAabbMax[3];\n"
+" //4 bytes, points to the root of the subtree\n"
+" int m_rootNodeIndex;\n"
+" //4 bytes\n"
+" int m_subtreeSize;\n"
+" int m_padding[3];\n"
+"} btBvhSubtreeInfo;\n"
+"///keep this in sync with btCollidable.h\n"
+"typedef struct\n"
+"{\n"
+" int m_numChildShapes;\n"
+" int blaat2;\n"
+" int m_shapeType;\n"
+" int m_shapeIndex;\n"
+" \n"
+"} btCollidableGpu;\n"
+"typedef struct\n"
+"{\n"
+" float4 m_childPosition;\n"
+" float4 m_childOrientation;\n"
+" int m_shapeIndex;\n"
+" int m_unused0;\n"
+" int m_unused1;\n"
+" int m_unused2;\n"
+"} btGpuChildShape;\n"
+"typedef struct\n"
+"{\n"
+" float4 m_pos;\n"
+" float4 m_quat;\n"
+" float4 m_linVel;\n"
+" float4 m_angVel;\n"
+" u32 m_collidableIdx;\n"
+" float m_invMass;\n"
+" float m_restituitionCoeff;\n"
+" float m_frictionCoeff;\n"
+"} BodyData;\n"
+"typedef struct \n"
+"{\n"
+" union\n"
+" {\n"
+" float4 m_min;\n"
+" float m_minElems[4];\n"
+" int m_minIndices[4];\n"
+" };\n"
+" union\n"
+" {\n"
+" float4 m_max;\n"
+" float m_maxElems[4];\n"
+" int m_maxIndices[4];\n"
+" };\n"
+"} btAabbCL;\n"
+"int testQuantizedAabbAgainstQuantizedAabb(\n"
+" const unsigned short int* aabbMin1,\n"
+" const unsigned short int* aabbMax1,\n"
+" const unsigned short int* aabbMin2,\n"
+" const unsigned short int* aabbMax2)\n"
+"{\n"
+" //int overlap = 1;\n"
+" if (aabbMin1[0] > aabbMax2[0])\n"
+" return 0;\n"
+" if (aabbMax1[0] < aabbMin2[0])\n"
+" return 0;\n"
+" if (aabbMin1[1] > aabbMax2[1])\n"
+" return 0;\n"
+" if (aabbMax1[1] < aabbMin2[1])\n"
+" return 0;\n"
+" if (aabbMin1[2] > aabbMax2[2])\n"
+" return 0;\n"
+" if (aabbMax1[2] < aabbMin2[2])\n"
+" return 0;\n"
+" return 1;\n"
+" //overlap = ((aabbMin1[0] > aabbMax2[0]) || (aabbMax1[0] < aabbMin2[0])) ? 0 : overlap;\n"
+" //overlap = ((aabbMin1[2] > aabbMax2[2]) || (aabbMax1[2] < aabbMin2[2])) ? 0 : overlap;\n"
+" //overlap = ((aabbMin1[1] > aabbMax2[1]) || (aabbMax1[1] < aabbMin2[1])) ? 0 : overlap;\n"
+" //return overlap;\n"
+"}\n"
+"void quantizeWithClamp(unsigned short* out, float4 point2,int isMax, float4 bvhAabbMin, float4 bvhAabbMax, float4 bvhQuantization)\n"
+"{\n"
+" float4 clampedPoint = max(point2,bvhAabbMin);\n"
+" clampedPoint = min (clampedPoint, bvhAabbMax);\n"
+" float4 v = (clampedPoint - bvhAabbMin) * bvhQuantization;\n"
+" if (isMax)\n"
+" {\n"
+" out[0] = (unsigned short) (((unsigned short)(v.x+1.f) | 1));\n"
+" out[1] = (unsigned short) (((unsigned short)(v.y+1.f) | 1));\n"
+" out[2] = (unsigned short) (((unsigned short)(v.z+1.f) | 1));\n"
+" } else\n"
+" {\n"
+" out[0] = (unsigned short) (((unsigned short)(v.x) & 0xfffe));\n"
+" out[1] = (unsigned short) (((unsigned short)(v.y) & 0xfffe));\n"
+" out[2] = (unsigned short) (((unsigned short)(v.z) & 0xfffe));\n"
+" }\n"
+"}\n"
+"// work-in-progress\n"
+"__kernel void bvhTraversalKernel( __global const int4* pairs, \n"
+" __global const BodyData* rigidBodies, \n"
+" __global const btCollidableGpu* collidables,\n"
+" __global btAabbCL* aabbs,\n"
+" __global int4* concavePairsOut,\n"
+" __global volatile int* numConcavePairsOut,\n"
+" __global const btBvhSubtreeInfo* subtreeHeadersRoot,\n"
+" __global const btQuantizedBvhNode* quantizedNodesRoot,\n"
+" __global const b3BvhInfo* bvhInfos,\n"
+" int numPairs,\n"
+" int maxNumConcavePairsCapacity)\n"
+"{\n"
+" int id = get_global_id(0);\n"
+" if (id>=numPairs)\n"
+" return;\n"
+" \n"
+" int bodyIndexA = pairs[id].x;\n"
+" int bodyIndexB = pairs[id].y;\n"
+" int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;\n"
+" int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;\n"
+" \n"
+" //once the broadphase avoids static-static pairs, we can remove this test\n"
+" if ((rigidBodies[bodyIndexA].m_invMass==0) &&(rigidBodies[bodyIndexB].m_invMass==0))\n"
+" {\n"
+" return;\n"
+" }\n"
+" \n"
+" if (collidables[collidableIndexA].m_shapeType!=SHAPE_CONCAVE_TRIMESH)\n"
+" return;\n"
+" int shapeTypeB = collidables[collidableIndexB].m_shapeType;\n"
+" \n"
+" if (shapeTypeB!=SHAPE_CONVEX_HULL &&\n"
+" shapeTypeB!=SHAPE_SPHERE &&\n"
+" shapeTypeB!=SHAPE_COMPOUND_OF_CONVEX_HULLS\n"
+" )\n"
+" return;\n"
+" b3BvhInfo bvhInfo = bvhInfos[collidables[collidableIndexA].m_numChildShapes];\n"
+" float4 bvhAabbMin = bvhInfo.m_aabbMin;\n"
+" float4 bvhAabbMax = bvhInfo.m_aabbMax;\n"
+" float4 bvhQuantization = bvhInfo.m_quantization;\n"
+" int numSubtreeHeaders = bvhInfo.m_numSubTrees;\n"
+" __global const btBvhSubtreeInfo* subtreeHeaders = &subtreeHeadersRoot[bvhInfo.m_subTreeOffset];\n"
+" __global const btQuantizedBvhNode* quantizedNodes = &quantizedNodesRoot[bvhInfo.m_nodeOffset];\n"
+" \n"
+" unsigned short int quantizedQueryAabbMin[3];\n"
+" unsigned short int quantizedQueryAabbMax[3];\n"
+" quantizeWithClamp(quantizedQueryAabbMin,aabbs[bodyIndexB].m_min,false,bvhAabbMin, bvhAabbMax,bvhQuantization);\n"
+" quantizeWithClamp(quantizedQueryAabbMax,aabbs[bodyIndexB].m_max,true ,bvhAabbMin, bvhAabbMax,bvhQuantization);\n"
+" \n"
+" for (int i=0;i<numSubtreeHeaders;i++)\n"
+" {\n"
+" btBvhSubtreeInfo subtree = subtreeHeaders[i];\n"
+" \n"
+" int overlap = testQuantizedAabbAgainstQuantizedAabb(quantizedQueryAabbMin,quantizedQueryAabbMax,subtree.m_quantizedAabbMin,subtree.m_quantizedAabbMax);\n"
+" if (overlap != 0)\n"
+" {\n"
+" int startNodeIndex = subtree.m_rootNodeIndex;\n"
+" int endNodeIndex = subtree.m_rootNodeIndex+subtree.m_subtreeSize;\n"
+" int curIndex = startNodeIndex;\n"
+" int escapeIndex;\n"
+" int isLeafNode;\n"
+" int aabbOverlap;\n"
+" while (curIndex < endNodeIndex)\n"
+" {\n"
+" btQuantizedBvhNode rootNode = quantizedNodes[curIndex];\n"
+" aabbOverlap = testQuantizedAabbAgainstQuantizedAabb(quantizedQueryAabbMin,quantizedQueryAabbMax,rootNode.m_quantizedAabbMin,rootNode.m_quantizedAabbMax);\n"
+" isLeafNode = isLeaf(&rootNode);\n"
+" if (aabbOverlap)\n"
+" {\n"
+" if (isLeafNode)\n"
+" {\n"
+" int triangleIndex = getTriangleIndex(&rootNode);\n"
+" if (shapeTypeB==SHAPE_COMPOUND_OF_CONVEX_HULLS)\n"
+" {\n"
+" int numChildrenB = collidables[collidableIndexB].m_numChildShapes;\n"
+" int pairIdx = atomic_add(numConcavePairsOut,numChildrenB);\n"
+" for (int b=0;b<numChildrenB;b++)\n"
+" {\n"
+" if ((pairIdx+b)<maxNumConcavePairsCapacity)\n"
+" {\n"
+" int childShapeIndexB = collidables[collidableIndexB].m_shapeIndex+b;\n"
+" int4 newPair = (int4)(bodyIndexA,bodyIndexB,triangleIndex,childShapeIndexB);\n"
+" concavePairsOut[pairIdx+b] = newPair;\n"
+" }\n"
+" }\n"
+" } else\n"
+" {\n"
+" int pairIdx = atomic_inc(numConcavePairsOut);\n"
+" if (pairIdx<maxNumConcavePairsCapacity)\n"
+" {\n"
+" int4 newPair = (int4)(bodyIndexA,bodyIndexB,triangleIndex,0);\n"
+" concavePairsOut[pairIdx] = newPair;\n"
+" }\n"
+" }\n"
+" } \n"
+" curIndex++;\n"
+" } else\n"
+" {\n"
+" if (isLeafNode)\n"
+" {\n"
+" curIndex++;\n"
+" } else\n"
+" {\n"
+" escapeIndex = getEscapeIndex(&rootNode);\n"
+" curIndex += escapeIndex;\n"
+" }\n"
+" }\n"
+" }\n"
+" }\n"
+" }\n"
+"}\n"
+;
diff --git a/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/mpr.cl b/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/mpr.cl
new file mode 100644
index 0000000000..e754f4e1da
--- /dev/null
+++ b/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/mpr.cl
@@ -0,0 +1,311 @@
+
+#include "Bullet3Collision/NarrowPhaseCollision/shared/b3MprPenetration.h"
+#include "Bullet3Collision/NarrowPhaseCollision/shared/b3Contact4Data.h"
+
+#define AppendInc(x, out) out = atomic_inc(x)
+#define GET_NPOINTS(x) (x).m_worldNormalOnB.w
+#ifdef cl_ext_atomic_counters_32
+ #pragma OPENCL EXTENSION cl_ext_atomic_counters_32 : enable
+#else
+ #define counter32_t volatile __global int*
+#endif
+
+
+__kernel void mprPenetrationKernel( __global int4* pairs,
+ __global const b3RigidBodyData_t* rigidBodies,
+ __global const b3Collidable_t* collidables,
+ __global const b3ConvexPolyhedronData_t* convexShapes,
+ __global const float4* vertices,
+ __global float4* separatingNormals,
+ __global int* hasSeparatingAxis,
+ __global struct b3Contact4Data* restrict globalContactsOut,
+ counter32_t nGlobalContactsOut,
+ int contactCapacity,
+ int numPairs)
+{
+ int i = get_global_id(0);
+ int pairIndex = i;
+ if (i<numPairs)
+ {
+ int bodyIndexA = pairs[i].x;
+ int bodyIndexB = pairs[i].y;
+
+ int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;
+ int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;
+
+ int shapeIndexA = collidables[collidableIndexA].m_shapeIndex;
+ int shapeIndexB = collidables[collidableIndexB].m_shapeIndex;
+
+
+ //once the broadphase avoids static-static pairs, we can remove this test
+ if ((rigidBodies[bodyIndexA].m_invMass==0) &&(rigidBodies[bodyIndexB].m_invMass==0))
+ {
+ return;
+ }
+
+
+ if ((collidables[collidableIndexA].m_shapeType!=SHAPE_CONVEX_HULL) ||(collidables[collidableIndexB].m_shapeType!=SHAPE_CONVEX_HULL))
+ {
+ return;
+ }
+
+ float depthOut;
+ b3Float4 dirOut;
+ b3Float4 posOut;
+
+
+ int res = b3MprPenetration(pairIndex, bodyIndexA, bodyIndexB,rigidBodies,convexShapes,collidables,vertices,separatingNormals,hasSeparatingAxis,&depthOut, &dirOut, &posOut);
+
+
+
+
+
+ if (res==0)
+ {
+ //add a contact
+
+ int dstIdx;
+ AppendInc( nGlobalContactsOut, dstIdx );
+ if (dstIdx<contactCapacity)
+ {
+ pairs[pairIndex].z = dstIdx;
+ __global struct b3Contact4Data* c = globalContactsOut + dstIdx;
+ c->m_worldNormalOnB = -dirOut;//normal;
+ c->m_restituitionCoeffCmp = (0.f*0xffff);c->m_frictionCoeffCmp = (0.7f*0xffff);
+ c->m_batchIdx = pairIndex;
+ int bodyA = pairs[pairIndex].x;
+ int bodyB = pairs[pairIndex].y;
+ c->m_bodyAPtrAndSignBit = rigidBodies[bodyA].m_invMass==0 ? -bodyA:bodyA;
+ c->m_bodyBPtrAndSignBit = rigidBodies[bodyB].m_invMass==0 ? -bodyB:bodyB;
+ c->m_childIndexA = -1;
+ c->m_childIndexB = -1;
+ //for (int i=0;i<nContacts;i++)
+ posOut.w = -depthOut;
+ c->m_worldPosB[0] = posOut;//localPoints[contactIdx[i]];
+ GET_NPOINTS(*c) = 1;//nContacts;
+ }
+ }
+
+ }
+}
+
+typedef float4 Quaternion;
+#define make_float4 (float4)
+
+__inline
+float dot3F4(float4 a, float4 b)
+{
+ float4 a1 = make_float4(a.xyz,0.f);
+ float4 b1 = make_float4(b.xyz,0.f);
+ return dot(a1, b1);
+}
+
+
+
+
+__inline
+float4 cross3(float4 a, float4 b)
+{
+ return cross(a,b);
+}
+__inline
+Quaternion qtMul(Quaternion a, Quaternion b)
+{
+ Quaternion ans;
+ ans = cross3( a, b );
+ ans += a.w*b+b.w*a;
+// ans.w = a.w*b.w - (a.x*b.x+a.y*b.y+a.z*b.z);
+ ans.w = a.w*b.w - dot3F4(a, b);
+ return ans;
+}
+
+__inline
+Quaternion qtInvert(Quaternion q)
+{
+ return (Quaternion)(-q.xyz, q.w);
+}
+
+__inline
+float4 qtRotate(Quaternion q, float4 vec)
+{
+ Quaternion qInv = qtInvert( q );
+ float4 vcpy = vec;
+ vcpy.w = 0.f;
+ float4 out = qtMul(qtMul(q,vcpy),qInv);
+ return out;
+}
+
+__inline
+float4 transform(const float4* p, const float4* translation, const Quaternion* orientation)
+{
+ return qtRotate( *orientation, *p ) + (*translation);
+}
+
+
+__inline
+float4 qtInvRotate(const Quaternion q, float4 vec)
+{
+ return qtRotate( qtInvert( q ), vec );
+}
+
+
+inline void project(__global const b3ConvexPolyhedronData_t* hull, const float4 pos, const float4 orn,
+const float4* dir, __global const float4* vertices, float* min, float* max)
+{
+ min[0] = FLT_MAX;
+ max[0] = -FLT_MAX;
+ int numVerts = hull->m_numVertices;
+
+ const float4 localDir = qtInvRotate(orn,*dir);
+ float offset = dot(pos,*dir);
+ for(int i=0;i<numVerts;i++)
+ {
+ float dp = dot(vertices[hull->m_vertexOffset+i],localDir);
+ if(dp < min[0])
+ min[0] = dp;
+ if(dp > max[0])
+ max[0] = dp;
+ }
+ if(min[0]>max[0])
+ {
+ float tmp = min[0];
+ min[0] = max[0];
+ max[0] = tmp;
+ }
+ min[0] += offset;
+ max[0] += offset;
+}
+
+
+bool findSeparatingAxisUnitSphere( __global const b3ConvexPolyhedronData_t* hullA, __global const b3ConvexPolyhedronData_t* hullB,
+ const float4 posA1,
+ const float4 ornA,
+ const float4 posB1,
+ const float4 ornB,
+ const float4 DeltaC2,
+ __global const float4* vertices,
+ __global const float4* unitSphereDirections,
+ int numUnitSphereDirections,
+ float4* sep,
+ float* dmin)
+{
+
+ float4 posA = posA1;
+ posA.w = 0.f;
+ float4 posB = posB1;
+ posB.w = 0.f;
+
+ int curPlaneTests=0;
+
+ int curEdgeEdge = 0;
+ // Test unit sphere directions
+ for (int i=0;i<numUnitSphereDirections;i++)
+ {
+
+ float4 crossje;
+ crossje = unitSphereDirections[i];
+
+ if (dot3F4(DeltaC2,crossje)>0)
+ crossje *= -1.f;
+ {
+ float dist;
+ bool result = true;
+ float Min0,Max0;
+ float Min1,Max1;
+ project(hullA,posA,ornA,&crossje,vertices, &Min0, &Max0);
+ project(hullB,posB,ornB,&crossje,vertices, &Min1, &Max1);
+
+ if(Max0<Min1 || Max1<Min0)
+ return false;
+
+ float d0 = Max0 - Min1;
+ float d1 = Max1 - Min0;
+ dist = d0<d1 ? d0:d1;
+ result = true;
+
+ if(dist<*dmin)
+ {
+ *dmin = dist;
+ *sep = crossje;
+ }
+ }
+ }
+
+
+ if((dot3F4(-DeltaC2,*sep))>0.0f)
+ {
+ *sep = -(*sep);
+ }
+ return true;
+}
+
+
+
+__kernel void findSeparatingAxisUnitSphereKernel( __global const int4* pairs,
+ __global const b3RigidBodyData_t* rigidBodies,
+ __global const b3Collidable_t* collidables,
+ __global const b3ConvexPolyhedronData_t* convexShapes,
+ __global const float4* vertices,
+ __global const float4* unitSphereDirections,
+ __global float4* separatingNormals,
+ __global int* hasSeparatingAxis,
+ __global float* dmins,
+ int numUnitSphereDirections,
+ int numPairs
+ )
+{
+
+ int i = get_global_id(0);
+
+ if (i<numPairs)
+ {
+
+ if (hasSeparatingAxis[i])
+ {
+
+ int bodyIndexA = pairs[i].x;
+ int bodyIndexB = pairs[i].y;
+
+ int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;
+ int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;
+
+ int shapeIndexA = collidables[collidableIndexA].m_shapeIndex;
+ int shapeIndexB = collidables[collidableIndexB].m_shapeIndex;
+
+
+ int numFacesA = convexShapes[shapeIndexA].m_numFaces;
+
+ float dmin = dmins[i];
+
+ float4 posA = rigidBodies[bodyIndexA].m_pos;
+ posA.w = 0.f;
+ float4 posB = rigidBodies[bodyIndexB].m_pos;
+ posB.w = 0.f;
+ float4 c0local = convexShapes[shapeIndexA].m_localCenter;
+ float4 ornA = rigidBodies[bodyIndexA].m_quat;
+ float4 c0 = transform(&c0local, &posA, &ornA);
+ float4 c1local = convexShapes[shapeIndexB].m_localCenter;
+ float4 ornB =rigidBodies[bodyIndexB].m_quat;
+ float4 c1 = transform(&c1local,&posB,&ornB);
+ const float4 DeltaC2 = c0 - c1;
+ float4 sepNormal = separatingNormals[i];
+
+ int numEdgeEdgeDirections = convexShapes[shapeIndexA].m_numUniqueEdges*convexShapes[shapeIndexB].m_numUniqueEdges;
+ if (numEdgeEdgeDirections>numUnitSphereDirections)
+ {
+ bool sepEE = findSeparatingAxisUnitSphere( &convexShapes[shapeIndexA], &convexShapes[shapeIndexB],posA,ornA,
+ posB,ornB,
+ DeltaC2,
+ vertices,unitSphereDirections,numUnitSphereDirections,&sepNormal,&dmin);
+ if (!sepEE)
+ {
+ hasSeparatingAxis[i] = 0;
+ } else
+ {
+ hasSeparatingAxis[i] = 1;
+ separatingNormals[i] = sepNormal;
+ }
+ }
+ } //if (hasSeparatingAxis[i])
+ }//(i<numPairs)
+}
diff --git a/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/mprKernels.h b/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/mprKernels.h
new file mode 100644
index 0000000000..7ed4b382c3
--- /dev/null
+++ b/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/mprKernels.h
@@ -0,0 +1,1446 @@
+//this file is autogenerated using stringify.bat (premake --stringify) in the build folder of this project
+static const char* mprKernelsCL= \
+"/***\n"
+" * ---------------------------------\n"
+" * Copyright (c)2012 Daniel Fiser <danfis@danfis.cz>\n"
+" *\n"
+" * This file was ported from mpr.c file, part of libccd.\n"
+" * The Minkoski Portal Refinement implementation was ported \n"
+" * to OpenCL by Erwin Coumans for the Bullet 3 Physics library.\n"
+" * at http://github.com/erwincoumans/bullet3\n"
+" *\n"
+" * Distributed under the OSI-approved BSD License (the \"License\");\n"
+" * see <http://www.opensource.org/licenses/bsd-license.php>.\n"
+" * This software is distributed WITHOUT ANY WARRANTY; without even the\n"
+" * implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n"
+" * See the License for more information.\n"
+" */\n"
+"#ifndef B3_MPR_PENETRATION_H\n"
+"#define B3_MPR_PENETRATION_H\n"
+"#ifndef B3_PLATFORM_DEFINITIONS_H\n"
+"#define B3_PLATFORM_DEFINITIONS_H\n"
+"struct MyTest\n"
+"{\n"
+" int bla;\n"
+"};\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+"//keep B3_LARGE_FLOAT*B3_LARGE_FLOAT < FLT_MAX\n"
+"#define B3_LARGE_FLOAT 1e18f\n"
+"#define B3_INFINITY 1e18f\n"
+"#define b3Assert(a)\n"
+"#define b3ConstArray(a) __global const a*\n"
+"#define b3AtomicInc atomic_inc\n"
+"#define b3AtomicAdd atomic_add\n"
+"#define b3Fabs fabs\n"
+"#define b3Sqrt native_sqrt\n"
+"#define b3Sin native_sin\n"
+"#define b3Cos native_cos\n"
+"#define B3_STATIC\n"
+"#endif\n"
+"#endif\n"
+"#ifndef B3_FLOAT4_H\n"
+"#define B3_FLOAT4_H\n"
+"#ifndef B3_PLATFORM_DEFINITIONS_H\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+"#endif\n"
+"#endif\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+" typedef float4 b3Float4;\n"
+" #define b3Float4ConstArg const b3Float4\n"
+" #define b3MakeFloat4 (float4)\n"
+" float b3Dot3F4(b3Float4ConstArg v0,b3Float4ConstArg v1)\n"
+" {\n"
+" float4 a1 = b3MakeFloat4(v0.xyz,0.f);\n"
+" float4 b1 = b3MakeFloat4(v1.xyz,0.f);\n"
+" return dot(a1, b1);\n"
+" }\n"
+" b3Float4 b3Cross3(b3Float4ConstArg v0,b3Float4ConstArg v1)\n"
+" {\n"
+" float4 a1 = b3MakeFloat4(v0.xyz,0.f);\n"
+" float4 b1 = b3MakeFloat4(v1.xyz,0.f);\n"
+" return cross(a1, b1);\n"
+" }\n"
+" #define b3MinFloat4 min\n"
+" #define b3MaxFloat4 max\n"
+" #define b3Normalized(a) normalize(a)\n"
+"#endif \n"
+" \n"
+"inline bool b3IsAlmostZero(b3Float4ConstArg v)\n"
+"{\n"
+" if(b3Fabs(v.x)>1e-6 || b3Fabs(v.y)>1e-6 || b3Fabs(v.z)>1e-6) \n"
+" return false;\n"
+" return true;\n"
+"}\n"
+"inline int b3MaxDot( b3Float4ConstArg vec, __global const b3Float4* vecArray, int vecLen, float* dotOut )\n"
+"{\n"
+" float maxDot = -B3_INFINITY;\n"
+" int i = 0;\n"
+" int ptIndex = -1;\n"
+" for( i = 0; i < vecLen; i++ )\n"
+" {\n"
+" float dot = b3Dot3F4(vecArray[i],vec);\n"
+" \n"
+" if( dot > maxDot )\n"
+" {\n"
+" maxDot = dot;\n"
+" ptIndex = i;\n"
+" }\n"
+" }\n"
+" b3Assert(ptIndex>=0);\n"
+" if (ptIndex<0)\n"
+" {\n"
+" ptIndex = 0;\n"
+" }\n"
+" *dotOut = maxDot;\n"
+" return ptIndex;\n"
+"}\n"
+"#endif //B3_FLOAT4_H\n"
+"#ifndef B3_RIGIDBODY_DATA_H\n"
+"#define B3_RIGIDBODY_DATA_H\n"
+"#ifndef B3_FLOAT4_H\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+"#endif \n"
+"#endif //B3_FLOAT4_H\n"
+"#ifndef B3_QUAT_H\n"
+"#define B3_QUAT_H\n"
+"#ifndef B3_PLATFORM_DEFINITIONS_H\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+"#endif\n"
+"#endif\n"
+"#ifndef B3_FLOAT4_H\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+"#endif \n"
+"#endif //B3_FLOAT4_H\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+" typedef float4 b3Quat;\n"
+" #define b3QuatConstArg const b3Quat\n"
+" \n"
+" \n"
+"inline float4 b3FastNormalize4(float4 v)\n"
+"{\n"
+" v = (float4)(v.xyz,0.f);\n"
+" return fast_normalize(v);\n"
+"}\n"
+" \n"
+"inline b3Quat b3QuatMul(b3Quat a, b3Quat b);\n"
+"inline b3Quat b3QuatNormalized(b3QuatConstArg in);\n"
+"inline b3Quat b3QuatRotate(b3QuatConstArg q, b3QuatConstArg vec);\n"
+"inline b3Quat b3QuatInvert(b3QuatConstArg q);\n"
+"inline b3Quat b3QuatInverse(b3QuatConstArg q);\n"
+"inline b3Quat b3QuatMul(b3QuatConstArg a, b3QuatConstArg b)\n"
+"{\n"
+" b3Quat ans;\n"
+" ans = b3Cross3( a, b );\n"
+" ans += a.w*b+b.w*a;\n"
+"// ans.w = a.w*b.w - (a.x*b.x+a.y*b.y+a.z*b.z);\n"
+" ans.w = a.w*b.w - b3Dot3F4(a, b);\n"
+" return ans;\n"
+"}\n"
+"inline b3Quat b3QuatNormalized(b3QuatConstArg in)\n"
+"{\n"
+" b3Quat q;\n"
+" q=in;\n"
+" //return b3FastNormalize4(in);\n"
+" float len = native_sqrt(dot(q, q));\n"
+" if(len > 0.f)\n"
+" {\n"
+" q *= 1.f / len;\n"
+" }\n"
+" else\n"
+" {\n"
+" q.x = q.y = q.z = 0.f;\n"
+" q.w = 1.f;\n"
+" }\n"
+" return q;\n"
+"}\n"
+"inline float4 b3QuatRotate(b3QuatConstArg q, b3QuatConstArg vec)\n"
+"{\n"
+" b3Quat qInv = b3QuatInvert( q );\n"
+" float4 vcpy = vec;\n"
+" vcpy.w = 0.f;\n"
+" float4 out = b3QuatMul(b3QuatMul(q,vcpy),qInv);\n"
+" return out;\n"
+"}\n"
+"inline b3Quat b3QuatInverse(b3QuatConstArg q)\n"
+"{\n"
+" return (b3Quat)(-q.xyz, q.w);\n"
+"}\n"
+"inline b3Quat b3QuatInvert(b3QuatConstArg q)\n"
+"{\n"
+" return (b3Quat)(-q.xyz, q.w);\n"
+"}\n"
+"inline float4 b3QuatInvRotate(b3QuatConstArg q, b3QuatConstArg vec)\n"
+"{\n"
+" return b3QuatRotate( b3QuatInvert( q ), vec );\n"
+"}\n"
+"inline b3Float4 b3TransformPoint(b3Float4ConstArg point, b3Float4ConstArg translation, b3QuatConstArg orientation)\n"
+"{\n"
+" return b3QuatRotate( orientation, point ) + (translation);\n"
+"}\n"
+" \n"
+"#endif \n"
+"#endif //B3_QUAT_H\n"
+"#ifndef B3_MAT3x3_H\n"
+"#define B3_MAT3x3_H\n"
+"#ifndef B3_QUAT_H\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+"#endif \n"
+"#endif //B3_QUAT_H\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+"typedef struct\n"
+"{\n"
+" b3Float4 m_row[3];\n"
+"}b3Mat3x3;\n"
+"#define b3Mat3x3ConstArg const b3Mat3x3\n"
+"#define b3GetRow(m,row) (m.m_row[row])\n"
+"inline b3Mat3x3 b3QuatGetRotationMatrix(b3Quat quat)\n"
+"{\n"
+" b3Float4 quat2 = (b3Float4)(quat.x*quat.x, quat.y*quat.y, quat.z*quat.z, 0.f);\n"
+" b3Mat3x3 out;\n"
+" out.m_row[0].x=1-2*quat2.y-2*quat2.z;\n"
+" out.m_row[0].y=2*quat.x*quat.y-2*quat.w*quat.z;\n"
+" out.m_row[0].z=2*quat.x*quat.z+2*quat.w*quat.y;\n"
+" out.m_row[0].w = 0.f;\n"
+" out.m_row[1].x=2*quat.x*quat.y+2*quat.w*quat.z;\n"
+" out.m_row[1].y=1-2*quat2.x-2*quat2.z;\n"
+" out.m_row[1].z=2*quat.y*quat.z-2*quat.w*quat.x;\n"
+" out.m_row[1].w = 0.f;\n"
+" out.m_row[2].x=2*quat.x*quat.z-2*quat.w*quat.y;\n"
+" out.m_row[2].y=2*quat.y*quat.z+2*quat.w*quat.x;\n"
+" out.m_row[2].z=1-2*quat2.x-2*quat2.y;\n"
+" out.m_row[2].w = 0.f;\n"
+" return out;\n"
+"}\n"
+"inline b3Mat3x3 b3AbsoluteMat3x3(b3Mat3x3ConstArg matIn)\n"
+"{\n"
+" b3Mat3x3 out;\n"
+" out.m_row[0] = fabs(matIn.m_row[0]);\n"
+" out.m_row[1] = fabs(matIn.m_row[1]);\n"
+" out.m_row[2] = fabs(matIn.m_row[2]);\n"
+" return out;\n"
+"}\n"
+"__inline\n"
+"b3Mat3x3 mtZero();\n"
+"__inline\n"
+"b3Mat3x3 mtIdentity();\n"
+"__inline\n"
+"b3Mat3x3 mtTranspose(b3Mat3x3 m);\n"
+"__inline\n"
+"b3Mat3x3 mtMul(b3Mat3x3 a, b3Mat3x3 b);\n"
+"__inline\n"
+"b3Float4 mtMul1(b3Mat3x3 a, b3Float4 b);\n"
+"__inline\n"
+"b3Float4 mtMul3(b3Float4 a, b3Mat3x3 b);\n"
+"__inline\n"
+"b3Mat3x3 mtZero()\n"
+"{\n"
+" b3Mat3x3 m;\n"
+" m.m_row[0] = (b3Float4)(0.f);\n"
+" m.m_row[1] = (b3Float4)(0.f);\n"
+" m.m_row[2] = (b3Float4)(0.f);\n"
+" return m;\n"
+"}\n"
+"__inline\n"
+"b3Mat3x3 mtIdentity()\n"
+"{\n"
+" b3Mat3x3 m;\n"
+" m.m_row[0] = (b3Float4)(1,0,0,0);\n"
+" m.m_row[1] = (b3Float4)(0,1,0,0);\n"
+" m.m_row[2] = (b3Float4)(0,0,1,0);\n"
+" return m;\n"
+"}\n"
+"__inline\n"
+"b3Mat3x3 mtTranspose(b3Mat3x3 m)\n"
+"{\n"
+" b3Mat3x3 out;\n"
+" out.m_row[0] = (b3Float4)(m.m_row[0].x, m.m_row[1].x, m.m_row[2].x, 0.f);\n"
+" out.m_row[1] = (b3Float4)(m.m_row[0].y, m.m_row[1].y, m.m_row[2].y, 0.f);\n"
+" out.m_row[2] = (b3Float4)(m.m_row[0].z, m.m_row[1].z, m.m_row[2].z, 0.f);\n"
+" return out;\n"
+"}\n"
+"__inline\n"
+"b3Mat3x3 mtMul(b3Mat3x3 a, b3Mat3x3 b)\n"
+"{\n"
+" b3Mat3x3 transB;\n"
+" transB = mtTranspose( b );\n"
+" b3Mat3x3 ans;\n"
+" // why this doesn't run when 0ing in the for{}\n"
+" a.m_row[0].w = 0.f;\n"
+" a.m_row[1].w = 0.f;\n"
+" a.m_row[2].w = 0.f;\n"
+" for(int i=0; i<3; i++)\n"
+" {\n"
+"// a.m_row[i].w = 0.f;\n"
+" ans.m_row[i].x = b3Dot3F4(a.m_row[i],transB.m_row[0]);\n"
+" ans.m_row[i].y = b3Dot3F4(a.m_row[i],transB.m_row[1]);\n"
+" ans.m_row[i].z = b3Dot3F4(a.m_row[i],transB.m_row[2]);\n"
+" ans.m_row[i].w = 0.f;\n"
+" }\n"
+" return ans;\n"
+"}\n"
+"__inline\n"
+"b3Float4 mtMul1(b3Mat3x3 a, b3Float4 b)\n"
+"{\n"
+" b3Float4 ans;\n"
+" ans.x = b3Dot3F4( a.m_row[0], b );\n"
+" ans.y = b3Dot3F4( a.m_row[1], b );\n"
+" ans.z = b3Dot3F4( a.m_row[2], b );\n"
+" ans.w = 0.f;\n"
+" return ans;\n"
+"}\n"
+"__inline\n"
+"b3Float4 mtMul3(b3Float4 a, b3Mat3x3 b)\n"
+"{\n"
+" b3Float4 colx = b3MakeFloat4(b.m_row[0].x, b.m_row[1].x, b.m_row[2].x, 0);\n"
+" b3Float4 coly = b3MakeFloat4(b.m_row[0].y, b.m_row[1].y, b.m_row[2].y, 0);\n"
+" b3Float4 colz = b3MakeFloat4(b.m_row[0].z, b.m_row[1].z, b.m_row[2].z, 0);\n"
+" b3Float4 ans;\n"
+" ans.x = b3Dot3F4( a, colx );\n"
+" ans.y = b3Dot3F4( a, coly );\n"
+" ans.z = b3Dot3F4( a, colz );\n"
+" return ans;\n"
+"}\n"
+"#endif\n"
+"#endif //B3_MAT3x3_H\n"
+"typedef struct b3RigidBodyData b3RigidBodyData_t;\n"
+"struct b3RigidBodyData\n"
+"{\n"
+" b3Float4 m_pos;\n"
+" b3Quat m_quat;\n"
+" b3Float4 m_linVel;\n"
+" b3Float4 m_angVel;\n"
+" int m_collidableIdx;\n"
+" float m_invMass;\n"
+" float m_restituitionCoeff;\n"
+" float m_frictionCoeff;\n"
+"};\n"
+"typedef struct b3InertiaData b3InertiaData_t;\n"
+"struct b3InertiaData\n"
+"{\n"
+" b3Mat3x3 m_invInertiaWorld;\n"
+" b3Mat3x3 m_initInvInertia;\n"
+"};\n"
+"#endif //B3_RIGIDBODY_DATA_H\n"
+" \n"
+"#ifndef B3_CONVEX_POLYHEDRON_DATA_H\n"
+"#define B3_CONVEX_POLYHEDRON_DATA_H\n"
+"#ifndef B3_FLOAT4_H\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+"#endif \n"
+"#endif //B3_FLOAT4_H\n"
+"#ifndef B3_QUAT_H\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+"#endif \n"
+"#endif //B3_QUAT_H\n"
+"typedef struct b3GpuFace b3GpuFace_t;\n"
+"struct b3GpuFace\n"
+"{\n"
+" b3Float4 m_plane;\n"
+" int m_indexOffset;\n"
+" int m_numIndices;\n"
+" int m_unusedPadding1;\n"
+" int m_unusedPadding2;\n"
+"};\n"
+"typedef struct b3ConvexPolyhedronData b3ConvexPolyhedronData_t;\n"
+"struct b3ConvexPolyhedronData\n"
+"{\n"
+" b3Float4 m_localCenter;\n"
+" b3Float4 m_extents;\n"
+" b3Float4 mC;\n"
+" b3Float4 mE;\n"
+" float m_radius;\n"
+" int m_faceOffset;\n"
+" int m_numFaces;\n"
+" int m_numVertices;\n"
+" int m_vertexOffset;\n"
+" int m_uniqueEdgesOffset;\n"
+" int m_numUniqueEdges;\n"
+" int m_unused;\n"
+"};\n"
+"#endif //B3_CONVEX_POLYHEDRON_DATA_H\n"
+"#ifndef B3_COLLIDABLE_H\n"
+"#define B3_COLLIDABLE_H\n"
+"#ifndef B3_FLOAT4_H\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+"#endif \n"
+"#endif //B3_FLOAT4_H\n"
+"#ifndef B3_QUAT_H\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+"#endif \n"
+"#endif //B3_QUAT_H\n"
+"enum b3ShapeTypes\n"
+"{\n"
+" SHAPE_HEIGHT_FIELD=1,\n"
+" SHAPE_CONVEX_HULL=3,\n"
+" SHAPE_PLANE=4,\n"
+" SHAPE_CONCAVE_TRIMESH=5,\n"
+" SHAPE_COMPOUND_OF_CONVEX_HULLS=6,\n"
+" SHAPE_SPHERE=7,\n"
+" MAX_NUM_SHAPE_TYPES,\n"
+"};\n"
+"typedef struct b3Collidable b3Collidable_t;\n"
+"struct b3Collidable\n"
+"{\n"
+" union {\n"
+" int m_numChildShapes;\n"
+" int m_bvhIndex;\n"
+" };\n"
+" union\n"
+" {\n"
+" float m_radius;\n"
+" int m_compoundBvhIndex;\n"
+" };\n"
+" int m_shapeType;\n"
+" int m_shapeIndex;\n"
+"};\n"
+"typedef struct b3GpuChildShape b3GpuChildShape_t;\n"
+"struct b3GpuChildShape\n"
+"{\n"
+" b3Float4 m_childPosition;\n"
+" b3Quat m_childOrientation;\n"
+" int m_shapeIndex;\n"
+" int m_unused0;\n"
+" int m_unused1;\n"
+" int m_unused2;\n"
+"};\n"
+"struct b3CompoundOverlappingPair\n"
+"{\n"
+" int m_bodyIndexA;\n"
+" int m_bodyIndexB;\n"
+"// int m_pairType;\n"
+" int m_childShapeIndexA;\n"
+" int m_childShapeIndexB;\n"
+"};\n"
+"#endif //B3_COLLIDABLE_H\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+"#define B3_MPR_SQRT sqrt\n"
+"#endif\n"
+"#define B3_MPR_FMIN(x, y) ((x) < (y) ? (x) : (y))\n"
+"#define B3_MPR_FABS fabs\n"
+"#define B3_MPR_TOLERANCE 1E-6f\n"
+"#define B3_MPR_MAX_ITERATIONS 1000\n"
+"struct _b3MprSupport_t \n"
+"{\n"
+" b3Float4 v; //!< Support point in minkowski sum\n"
+" b3Float4 v1; //!< Support point in obj1\n"
+" b3Float4 v2; //!< Support point in obj2\n"
+"};\n"
+"typedef struct _b3MprSupport_t b3MprSupport_t;\n"
+"struct _b3MprSimplex_t \n"
+"{\n"
+" b3MprSupport_t ps[4];\n"
+" int last; //!< index of last added point\n"
+"};\n"
+"typedef struct _b3MprSimplex_t b3MprSimplex_t;\n"
+"inline b3MprSupport_t* b3MprSimplexPointW(b3MprSimplex_t *s, int idx)\n"
+"{\n"
+" return &s->ps[idx];\n"
+"}\n"
+"inline void b3MprSimplexSetSize(b3MprSimplex_t *s, int size)\n"
+"{\n"
+" s->last = size - 1;\n"
+"}\n"
+"inline int b3MprSimplexSize(const b3MprSimplex_t *s)\n"
+"{\n"
+" return s->last + 1;\n"
+"}\n"
+"inline const b3MprSupport_t* b3MprSimplexPoint(const b3MprSimplex_t* s, int idx)\n"
+"{\n"
+" // here is no check on boundaries\n"
+" return &s->ps[idx];\n"
+"}\n"
+"inline void b3MprSupportCopy(b3MprSupport_t *d, const b3MprSupport_t *s)\n"
+"{\n"
+" *d = *s;\n"
+"}\n"
+"inline void b3MprSimplexSet(b3MprSimplex_t *s, size_t pos, const b3MprSupport_t *a)\n"
+"{\n"
+" b3MprSupportCopy(s->ps + pos, a);\n"
+"}\n"
+"inline void b3MprSimplexSwap(b3MprSimplex_t *s, size_t pos1, size_t pos2)\n"
+"{\n"
+" b3MprSupport_t supp;\n"
+" b3MprSupportCopy(&supp, &s->ps[pos1]);\n"
+" b3MprSupportCopy(&s->ps[pos1], &s->ps[pos2]);\n"
+" b3MprSupportCopy(&s->ps[pos2], &supp);\n"
+"}\n"
+"inline int b3MprIsZero(float val)\n"
+"{\n"
+" return B3_MPR_FABS(val) < FLT_EPSILON;\n"
+"}\n"
+"inline int b3MprEq(float _a, float _b)\n"
+"{\n"
+" float ab;\n"
+" float a, b;\n"
+" ab = B3_MPR_FABS(_a - _b);\n"
+" if (B3_MPR_FABS(ab) < FLT_EPSILON)\n"
+" return 1;\n"
+" a = B3_MPR_FABS(_a);\n"
+" b = B3_MPR_FABS(_b);\n"
+" if (b > a){\n"
+" return ab < FLT_EPSILON * b;\n"
+" }else{\n"
+" return ab < FLT_EPSILON * a;\n"
+" }\n"
+"}\n"
+"inline int b3MprVec3Eq(const b3Float4* a, const b3Float4 *b)\n"
+"{\n"
+" return b3MprEq((*a).x, (*b).x)\n"
+" && b3MprEq((*a).y, (*b).y)\n"
+" && b3MprEq((*a).z, (*b).z);\n"
+"}\n"
+"inline b3Float4 b3LocalGetSupportVertex(b3Float4ConstArg supportVec,__global const b3ConvexPolyhedronData_t* hull, b3ConstArray(b3Float4) verticesA)\n"
+"{\n"
+" b3Float4 supVec = b3MakeFloat4(0,0,0,0);\n"
+" float maxDot = -B3_LARGE_FLOAT;\n"
+" if( 0 < hull->m_numVertices )\n"
+" {\n"
+" const b3Float4 scaled = supportVec;\n"
+" int index = b3MaxDot(scaled, &verticesA[hull->m_vertexOffset], hull->m_numVertices, &maxDot);\n"
+" return verticesA[hull->m_vertexOffset+index];\n"
+" }\n"
+" return supVec;\n"
+"}\n"
+"B3_STATIC void b3MprConvexSupport(int pairIndex,int bodyIndex, b3ConstArray(b3RigidBodyData_t) cpuBodyBuf, \n"
+" b3ConstArray(b3ConvexPolyhedronData_t) cpuConvexData, \n"
+" b3ConstArray(b3Collidable_t) cpuCollidables,\n"
+" b3ConstArray(b3Float4) cpuVertices,\n"
+" __global b3Float4* sepAxis,\n"
+" const b3Float4* _dir, b3Float4* outp, int logme)\n"
+"{\n"
+" //dir is in worldspace, move to local space\n"
+" \n"
+" b3Float4 pos = cpuBodyBuf[bodyIndex].m_pos;\n"
+" b3Quat orn = cpuBodyBuf[bodyIndex].m_quat;\n"
+" \n"
+" b3Float4 dir = b3MakeFloat4((*_dir).x,(*_dir).y,(*_dir).z,0.f);\n"
+" \n"
+" const b3Float4 localDir = b3QuatRotate(b3QuatInverse(orn),dir);\n"
+" \n"
+" //find local support vertex\n"
+" int colIndex = cpuBodyBuf[bodyIndex].m_collidableIdx;\n"
+" \n"
+" b3Assert(cpuCollidables[colIndex].m_shapeType==SHAPE_CONVEX_HULL);\n"
+" __global const b3ConvexPolyhedronData_t* hull = &cpuConvexData[cpuCollidables[colIndex].m_shapeIndex];\n"
+" \n"
+" b3Float4 pInA;\n"
+" if (logme)\n"
+" {\n"
+" b3Float4 supVec = b3MakeFloat4(0,0,0,0);\n"
+" float maxDot = -B3_LARGE_FLOAT;\n"
+" if( 0 < hull->m_numVertices )\n"
+" {\n"
+" const b3Float4 scaled = localDir;\n"
+" int index = b3MaxDot(scaled, &cpuVertices[hull->m_vertexOffset], hull->m_numVertices, &maxDot);\n"
+" pInA = cpuVertices[hull->m_vertexOffset+index];\n"
+" \n"
+" }\n"
+" } else\n"
+" {\n"
+" pInA = b3LocalGetSupportVertex(localDir,hull,cpuVertices);\n"
+" }\n"
+" //move vertex to world space\n"
+" *outp = b3TransformPoint(pInA,pos,orn);\n"
+" \n"
+"}\n"
+"inline void b3MprSupport(int pairIndex,int bodyIndexA, int bodyIndexB, b3ConstArray(b3RigidBodyData_t) cpuBodyBuf, \n"
+" b3ConstArray(b3ConvexPolyhedronData_t) cpuConvexData, \n"
+" b3ConstArray(b3Collidable_t) cpuCollidables,\n"
+" b3ConstArray(b3Float4) cpuVertices,\n"
+" __global b3Float4* sepAxis,\n"
+" const b3Float4* _dir, b3MprSupport_t *supp)\n"
+"{\n"
+" b3Float4 dir;\n"
+" dir = *_dir;\n"
+" b3MprConvexSupport(pairIndex,bodyIndexA,cpuBodyBuf,cpuConvexData,cpuCollidables,cpuVertices,sepAxis,&dir, &supp->v1,0);\n"
+" dir = *_dir*-1.f;\n"
+" b3MprConvexSupport(pairIndex,bodyIndexB,cpuBodyBuf,cpuConvexData,cpuCollidables,cpuVertices,sepAxis,&dir, &supp->v2,0);\n"
+" supp->v = supp->v1 - supp->v2;\n"
+"}\n"
+"inline void b3FindOrigin(int bodyIndexA, int bodyIndexB, b3ConstArray(b3RigidBodyData_t) cpuBodyBuf, b3MprSupport_t *center)\n"
+"{\n"
+" center->v1 = cpuBodyBuf[bodyIndexA].m_pos;\n"
+" center->v2 = cpuBodyBuf[bodyIndexB].m_pos;\n"
+" center->v = center->v1 - center->v2;\n"
+"}\n"
+"inline void b3MprVec3Set(b3Float4 *v, float x, float y, float z)\n"
+"{\n"
+" (*v).x = x;\n"
+" (*v).y = y;\n"
+" (*v).z = z;\n"
+" (*v).w = 0.f;\n"
+"}\n"
+"inline void b3MprVec3Add(b3Float4 *v, const b3Float4 *w)\n"
+"{\n"
+" (*v).x += (*w).x;\n"
+" (*v).y += (*w).y;\n"
+" (*v).z += (*w).z;\n"
+"}\n"
+"inline void b3MprVec3Copy(b3Float4 *v, const b3Float4 *w)\n"
+"{\n"
+" *v = *w;\n"
+"}\n"
+"inline void b3MprVec3Scale(b3Float4 *d, float k)\n"
+"{\n"
+" *d *= k;\n"
+"}\n"
+"inline float b3MprVec3Dot(const b3Float4 *a, const b3Float4 *b)\n"
+"{\n"
+" float dot;\n"
+" dot = b3Dot3F4(*a,*b);\n"
+" return dot;\n"
+"}\n"
+"inline float b3MprVec3Len2(const b3Float4 *v)\n"
+"{\n"
+" return b3MprVec3Dot(v, v);\n"
+"}\n"
+"inline void b3MprVec3Normalize(b3Float4 *d)\n"
+"{\n"
+" float k = 1.f / B3_MPR_SQRT(b3MprVec3Len2(d));\n"
+" b3MprVec3Scale(d, k);\n"
+"}\n"
+"inline void b3MprVec3Cross(b3Float4 *d, const b3Float4 *a, const b3Float4 *b)\n"
+"{\n"
+" *d = b3Cross3(*a,*b);\n"
+" \n"
+"}\n"
+"inline void b3MprVec3Sub2(b3Float4 *d, const b3Float4 *v, const b3Float4 *w)\n"
+"{\n"
+" *d = *v - *w;\n"
+"}\n"
+"inline void b3PortalDir(const b3MprSimplex_t *portal, b3Float4 *dir)\n"
+"{\n"
+" b3Float4 v2v1, v3v1;\n"
+" b3MprVec3Sub2(&v2v1, &b3MprSimplexPoint(portal, 2)->v,\n"
+" &b3MprSimplexPoint(portal, 1)->v);\n"
+" b3MprVec3Sub2(&v3v1, &b3MprSimplexPoint(portal, 3)->v,\n"
+" &b3MprSimplexPoint(portal, 1)->v);\n"
+" b3MprVec3Cross(dir, &v2v1, &v3v1);\n"
+" b3MprVec3Normalize(dir);\n"
+"}\n"
+"inline int portalEncapsulesOrigin(const b3MprSimplex_t *portal,\n"
+" const b3Float4 *dir)\n"
+"{\n"
+" float dot;\n"
+" dot = b3MprVec3Dot(dir, &b3MprSimplexPoint(portal, 1)->v);\n"
+" return b3MprIsZero(dot) || dot > 0.f;\n"
+"}\n"
+"inline int portalReachTolerance(const b3MprSimplex_t *portal,\n"
+" const b3MprSupport_t *v4,\n"
+" const b3Float4 *dir)\n"
+"{\n"
+" float dv1, dv2, dv3, dv4;\n"
+" float dot1, dot2, dot3;\n"
+" // find the smallest dot product of dir and {v1-v4, v2-v4, v3-v4}\n"
+" dv1 = b3MprVec3Dot(&b3MprSimplexPoint(portal, 1)->v, dir);\n"
+" dv2 = b3MprVec3Dot(&b3MprSimplexPoint(portal, 2)->v, dir);\n"
+" dv3 = b3MprVec3Dot(&b3MprSimplexPoint(portal, 3)->v, dir);\n"
+" dv4 = b3MprVec3Dot(&v4->v, dir);\n"
+" dot1 = dv4 - dv1;\n"
+" dot2 = dv4 - dv2;\n"
+" dot3 = dv4 - dv3;\n"
+" dot1 = B3_MPR_FMIN(dot1, dot2);\n"
+" dot1 = B3_MPR_FMIN(dot1, dot3);\n"
+" return b3MprEq(dot1, B3_MPR_TOLERANCE) || dot1 < B3_MPR_TOLERANCE;\n"
+"}\n"
+"inline int portalCanEncapsuleOrigin(const b3MprSimplex_t *portal, \n"
+" const b3MprSupport_t *v4,\n"
+" const b3Float4 *dir)\n"
+"{\n"
+" float dot;\n"
+" dot = b3MprVec3Dot(&v4->v, dir);\n"
+" return b3MprIsZero(dot) || dot > 0.f;\n"
+"}\n"
+"inline void b3ExpandPortal(b3MprSimplex_t *portal,\n"
+" const b3MprSupport_t *v4)\n"
+"{\n"
+" float dot;\n"
+" b3Float4 v4v0;\n"
+" b3MprVec3Cross(&v4v0, &v4->v, &b3MprSimplexPoint(portal, 0)->v);\n"
+" dot = b3MprVec3Dot(&b3MprSimplexPoint(portal, 1)->v, &v4v0);\n"
+" if (dot > 0.f){\n"
+" dot = b3MprVec3Dot(&b3MprSimplexPoint(portal, 2)->v, &v4v0);\n"
+" if (dot > 0.f){\n"
+" b3MprSimplexSet(portal, 1, v4);\n"
+" }else{\n"
+" b3MprSimplexSet(portal, 3, v4);\n"
+" }\n"
+" }else{\n"
+" dot = b3MprVec3Dot(&b3MprSimplexPoint(portal, 3)->v, &v4v0);\n"
+" if (dot > 0.f){\n"
+" b3MprSimplexSet(portal, 2, v4);\n"
+" }else{\n"
+" b3MprSimplexSet(portal, 1, v4);\n"
+" }\n"
+" }\n"
+"}\n"
+"B3_STATIC int b3DiscoverPortal(int pairIndex, int bodyIndexA, int bodyIndexB, b3ConstArray(b3RigidBodyData_t) cpuBodyBuf, \n"
+" b3ConstArray(b3ConvexPolyhedronData_t) cpuConvexData, \n"
+" b3ConstArray(b3Collidable_t) cpuCollidables,\n"
+" b3ConstArray(b3Float4) cpuVertices,\n"
+" __global b3Float4* sepAxis,\n"
+" __global int* hasSepAxis,\n"
+" b3MprSimplex_t *portal)\n"
+"{\n"
+" b3Float4 dir, va, vb;\n"
+" float dot;\n"
+" int cont;\n"
+" \n"
+" \n"
+" // vertex 0 is center of portal\n"
+" b3FindOrigin(bodyIndexA,bodyIndexB,cpuBodyBuf, b3MprSimplexPointW(portal, 0));\n"
+" // vertex 0 is center of portal\n"
+" b3MprSimplexSetSize(portal, 1);\n"
+" \n"
+" b3Float4 zero = b3MakeFloat4(0,0,0,0);\n"
+" b3Float4* b3mpr_vec3_origin = &zero;\n"
+" if (b3MprVec3Eq(&b3MprSimplexPoint(portal, 0)->v, b3mpr_vec3_origin)){\n"
+" // Portal's center lies on origin (0,0,0) => we know that objects\n"
+" // intersect but we would need to know penetration info.\n"
+" // So move center little bit...\n"
+" b3MprVec3Set(&va, FLT_EPSILON * 10.f, 0.f, 0.f);\n"
+" b3MprVec3Add(&b3MprSimplexPointW(portal, 0)->v, &va);\n"
+" }\n"
+" // vertex 1 = support in direction of origin\n"
+" b3MprVec3Copy(&dir, &b3MprSimplexPoint(portal, 0)->v);\n"
+" b3MprVec3Scale(&dir, -1.f);\n"
+" b3MprVec3Normalize(&dir);\n"
+" b3MprSupport(pairIndex,bodyIndexA,bodyIndexB,cpuBodyBuf,cpuConvexData,cpuCollidables,cpuVertices, sepAxis,&dir, b3MprSimplexPointW(portal, 1));\n"
+" b3MprSimplexSetSize(portal, 2);\n"
+" // test if origin isn't outside of v1\n"
+" dot = b3MprVec3Dot(&b3MprSimplexPoint(portal, 1)->v, &dir);\n"
+" \n"
+" if (b3MprIsZero(dot) || dot < 0.f)\n"
+" return -1;\n"
+" // vertex 2\n"
+" b3MprVec3Cross(&dir, &b3MprSimplexPoint(portal, 0)->v,\n"
+" &b3MprSimplexPoint(portal, 1)->v);\n"
+" if (b3MprIsZero(b3MprVec3Len2(&dir))){\n"
+" if (b3MprVec3Eq(&b3MprSimplexPoint(portal, 1)->v, b3mpr_vec3_origin)){\n"
+" // origin lies on v1\n"
+" return 1;\n"
+" }else{\n"
+" // origin lies on v0-v1 segment\n"
+" return 2;\n"
+" }\n"
+" }\n"
+" b3MprVec3Normalize(&dir);\n"
+" b3MprSupport(pairIndex,bodyIndexA,bodyIndexB,cpuBodyBuf,cpuConvexData,cpuCollidables,cpuVertices, sepAxis,&dir, b3MprSimplexPointW(portal, 2));\n"
+" \n"
+" dot = b3MprVec3Dot(&b3MprSimplexPoint(portal, 2)->v, &dir);\n"
+" if (b3MprIsZero(dot) || dot < 0.f)\n"
+" return -1;\n"
+" b3MprSimplexSetSize(portal, 3);\n"
+" // vertex 3 direction\n"
+" b3MprVec3Sub2(&va, &b3MprSimplexPoint(portal, 1)->v,\n"
+" &b3MprSimplexPoint(portal, 0)->v);\n"
+" b3MprVec3Sub2(&vb, &b3MprSimplexPoint(portal, 2)->v,\n"
+" &b3MprSimplexPoint(portal, 0)->v);\n"
+" b3MprVec3Cross(&dir, &va, &vb);\n"
+" b3MprVec3Normalize(&dir);\n"
+" // it is better to form portal faces to be oriented \"outside\" origin\n"
+" dot = b3MprVec3Dot(&dir, &b3MprSimplexPoint(portal, 0)->v);\n"
+" if (dot > 0.f){\n"
+" b3MprSimplexSwap(portal, 1, 2);\n"
+" b3MprVec3Scale(&dir, -1.f);\n"
+" }\n"
+" while (b3MprSimplexSize(portal) < 4){\n"
+" b3MprSupport(pairIndex,bodyIndexA,bodyIndexB,cpuBodyBuf,cpuConvexData,cpuCollidables,cpuVertices, sepAxis,&dir, b3MprSimplexPointW(portal, 3));\n"
+" \n"
+" dot = b3MprVec3Dot(&b3MprSimplexPoint(portal, 3)->v, &dir);\n"
+" if (b3MprIsZero(dot) || dot < 0.f)\n"
+" return -1;\n"
+" cont = 0;\n"
+" // test if origin is outside (v1, v0, v3) - set v2 as v3 and\n"
+" // continue\n"
+" b3MprVec3Cross(&va, &b3MprSimplexPoint(portal, 1)->v,\n"
+" &b3MprSimplexPoint(portal, 3)->v);\n"
+" dot = b3MprVec3Dot(&va, &b3MprSimplexPoint(portal, 0)->v);\n"
+" if (dot < 0.f && !b3MprIsZero(dot)){\n"
+" b3MprSimplexSet(portal, 2, b3MprSimplexPoint(portal, 3));\n"
+" cont = 1;\n"
+" }\n"
+" if (!cont){\n"
+" // test if origin is outside (v3, v0, v2) - set v1 as v3 and\n"
+" // continue\n"
+" b3MprVec3Cross(&va, &b3MprSimplexPoint(portal, 3)->v,\n"
+" &b3MprSimplexPoint(portal, 2)->v);\n"
+" dot = b3MprVec3Dot(&va, &b3MprSimplexPoint(portal, 0)->v);\n"
+" if (dot < 0.f && !b3MprIsZero(dot)){\n"
+" b3MprSimplexSet(portal, 1, b3MprSimplexPoint(portal, 3));\n"
+" cont = 1;\n"
+" }\n"
+" }\n"
+" if (cont){\n"
+" b3MprVec3Sub2(&va, &b3MprSimplexPoint(portal, 1)->v,\n"
+" &b3MprSimplexPoint(portal, 0)->v);\n"
+" b3MprVec3Sub2(&vb, &b3MprSimplexPoint(portal, 2)->v,\n"
+" &b3MprSimplexPoint(portal, 0)->v);\n"
+" b3MprVec3Cross(&dir, &va, &vb);\n"
+" b3MprVec3Normalize(&dir);\n"
+" }else{\n"
+" b3MprSimplexSetSize(portal, 4);\n"
+" }\n"
+" }\n"
+" return 0;\n"
+"}\n"
+"B3_STATIC int b3RefinePortal(int pairIndex,int bodyIndexA, int bodyIndexB, b3ConstArray(b3RigidBodyData_t) cpuBodyBuf, \n"
+" b3ConstArray(b3ConvexPolyhedronData_t) cpuConvexData, \n"
+" b3ConstArray(b3Collidable_t) cpuCollidables,\n"
+" b3ConstArray(b3Float4) cpuVertices,\n"
+" __global b3Float4* sepAxis,\n"
+" b3MprSimplex_t *portal)\n"
+"{\n"
+" b3Float4 dir;\n"
+" b3MprSupport_t v4;\n"
+" for (int i=0;i<B3_MPR_MAX_ITERATIONS;i++)\n"
+" //while (1)\n"
+" {\n"
+" // compute direction outside the portal (from v0 throught v1,v2,v3\n"
+" // face)\n"
+" b3PortalDir(portal, &dir);\n"
+" // test if origin is inside the portal\n"
+" if (portalEncapsulesOrigin(portal, &dir))\n"
+" return 0;\n"
+" // get next support point\n"
+" \n"
+" b3MprSupport(pairIndex,bodyIndexA,bodyIndexB,cpuBodyBuf,cpuConvexData,cpuCollidables,cpuVertices, sepAxis,&dir, &v4);\n"
+" // test if v4 can expand portal to contain origin and if portal\n"
+" // expanding doesn't reach given tolerance\n"
+" if (!portalCanEncapsuleOrigin(portal, &v4, &dir)\n"
+" || portalReachTolerance(portal, &v4, &dir))\n"
+" {\n"
+" return -1;\n"
+" }\n"
+" // v1-v2-v3 triangle must be rearranged to face outside Minkowski\n"
+" // difference (direction from v0).\n"
+" b3ExpandPortal(portal, &v4);\n"
+" }\n"
+" return -1;\n"
+"}\n"
+"B3_STATIC void b3FindPos(const b3MprSimplex_t *portal, b3Float4 *pos)\n"
+"{\n"
+" b3Float4 zero = b3MakeFloat4(0,0,0,0);\n"
+" b3Float4* b3mpr_vec3_origin = &zero;\n"
+" b3Float4 dir;\n"
+" size_t i;\n"
+" float b[4], sum, inv;\n"
+" b3Float4 vec, p1, p2;\n"
+" b3PortalDir(portal, &dir);\n"
+" // use barycentric coordinates of tetrahedron to find origin\n"
+" b3MprVec3Cross(&vec, &b3MprSimplexPoint(portal, 1)->v,\n"
+" &b3MprSimplexPoint(portal, 2)->v);\n"
+" b[0] = b3MprVec3Dot(&vec, &b3MprSimplexPoint(portal, 3)->v);\n"
+" b3MprVec3Cross(&vec, &b3MprSimplexPoint(portal, 3)->v,\n"
+" &b3MprSimplexPoint(portal, 2)->v);\n"
+" b[1] = b3MprVec3Dot(&vec, &b3MprSimplexPoint(portal, 0)->v);\n"
+" b3MprVec3Cross(&vec, &b3MprSimplexPoint(portal, 0)->v,\n"
+" &b3MprSimplexPoint(portal, 1)->v);\n"
+" b[2] = b3MprVec3Dot(&vec, &b3MprSimplexPoint(portal, 3)->v);\n"
+" b3MprVec3Cross(&vec, &b3MprSimplexPoint(portal, 2)->v,\n"
+" &b3MprSimplexPoint(portal, 1)->v);\n"
+" b[3] = b3MprVec3Dot(&vec, &b3MprSimplexPoint(portal, 0)->v);\n"
+" sum = b[0] + b[1] + b[2] + b[3];\n"
+" if (b3MprIsZero(sum) || sum < 0.f){\n"
+" b[0] = 0.f;\n"
+" b3MprVec3Cross(&vec, &b3MprSimplexPoint(portal, 2)->v,\n"
+" &b3MprSimplexPoint(portal, 3)->v);\n"
+" b[1] = b3MprVec3Dot(&vec, &dir);\n"
+" b3MprVec3Cross(&vec, &b3MprSimplexPoint(portal, 3)->v,\n"
+" &b3MprSimplexPoint(portal, 1)->v);\n"
+" b[2] = b3MprVec3Dot(&vec, &dir);\n"
+" b3MprVec3Cross(&vec, &b3MprSimplexPoint(portal, 1)->v,\n"
+" &b3MprSimplexPoint(portal, 2)->v);\n"
+" b[3] = b3MprVec3Dot(&vec, &dir);\n"
+" sum = b[1] + b[2] + b[3];\n"
+" }\n"
+" inv = 1.f / sum;\n"
+" b3MprVec3Copy(&p1, b3mpr_vec3_origin);\n"
+" b3MprVec3Copy(&p2, b3mpr_vec3_origin);\n"
+" for (i = 0; i < 4; i++){\n"
+" b3MprVec3Copy(&vec, &b3MprSimplexPoint(portal, i)->v1);\n"
+" b3MprVec3Scale(&vec, b[i]);\n"
+" b3MprVec3Add(&p1, &vec);\n"
+" b3MprVec3Copy(&vec, &b3MprSimplexPoint(portal, i)->v2);\n"
+" b3MprVec3Scale(&vec, b[i]);\n"
+" b3MprVec3Add(&p2, &vec);\n"
+" }\n"
+" b3MprVec3Scale(&p1, inv);\n"
+" b3MprVec3Scale(&p2, inv);\n"
+" b3MprVec3Copy(pos, &p1);\n"
+" b3MprVec3Add(pos, &p2);\n"
+" b3MprVec3Scale(pos, 0.5);\n"
+"}\n"
+"inline float b3MprVec3Dist2(const b3Float4 *a, const b3Float4 *b)\n"
+"{\n"
+" b3Float4 ab;\n"
+" b3MprVec3Sub2(&ab, a, b);\n"
+" return b3MprVec3Len2(&ab);\n"
+"}\n"
+"inline float _b3MprVec3PointSegmentDist2(const b3Float4 *P,\n"
+" const b3Float4 *x0,\n"
+" const b3Float4 *b,\n"
+" b3Float4 *witness)\n"
+"{\n"
+" // The computation comes from solving equation of segment:\n"
+" // S(t) = x0 + t.d\n"
+" // where - x0 is initial point of segment\n"
+" // - d is direction of segment from x0 (|d| > 0)\n"
+" // - t belongs to <0, 1> interval\n"
+" // \n"
+" // Than, distance from a segment to some point P can be expressed:\n"
+" // D(t) = |x0 + t.d - P|^2\n"
+" // which is distance from any point on segment. Minimization\n"
+" // of this function brings distance from P to segment.\n"
+" // Minimization of D(t) leads to simple quadratic equation that's\n"
+" // solving is straightforward.\n"
+" //\n"
+" // Bonus of this method is witness point for free.\n"
+" float dist, t;\n"
+" b3Float4 d, a;\n"
+" // direction of segment\n"
+" b3MprVec3Sub2(&d, b, x0);\n"
+" // precompute vector from P to x0\n"
+" b3MprVec3Sub2(&a, x0, P);\n"
+" t = -1.f * b3MprVec3Dot(&a, &d);\n"
+" t /= b3MprVec3Len2(&d);\n"
+" if (t < 0.f || b3MprIsZero(t)){\n"
+" dist = b3MprVec3Dist2(x0, P);\n"
+" if (witness)\n"
+" b3MprVec3Copy(witness, x0);\n"
+" }else if (t > 1.f || b3MprEq(t, 1.f)){\n"
+" dist = b3MprVec3Dist2(b, P);\n"
+" if (witness)\n"
+" b3MprVec3Copy(witness, b);\n"
+" }else{\n"
+" if (witness){\n"
+" b3MprVec3Copy(witness, &d);\n"
+" b3MprVec3Scale(witness, t);\n"
+" b3MprVec3Add(witness, x0);\n"
+" dist = b3MprVec3Dist2(witness, P);\n"
+" }else{\n"
+" // recycling variables\n"
+" b3MprVec3Scale(&d, t);\n"
+" b3MprVec3Add(&d, &a);\n"
+" dist = b3MprVec3Len2(&d);\n"
+" }\n"
+" }\n"
+" return dist;\n"
+"}\n"
+"inline float b3MprVec3PointTriDist2(const b3Float4 *P,\n"
+" const b3Float4 *x0, const b3Float4 *B,\n"
+" const b3Float4 *C,\n"
+" b3Float4 *witness)\n"
+"{\n"
+" // Computation comes from analytic expression for triangle (x0, B, C)\n"
+" // T(s, t) = x0 + s.d1 + t.d2, where d1 = B - x0 and d2 = C - x0 and\n"
+" // Then equation for distance is:\n"
+" // D(s, t) = | T(s, t) - P |^2\n"
+" // This leads to minimization of quadratic function of two variables.\n"
+" // The solution from is taken only if s is between 0 and 1, t is\n"
+" // between 0 and 1 and t + s < 1, otherwise distance from segment is\n"
+" // computed.\n"
+" b3Float4 d1, d2, a;\n"
+" float u, v, w, p, q, r;\n"
+" float s, t, dist, dist2;\n"
+" b3Float4 witness2;\n"
+" b3MprVec3Sub2(&d1, B, x0);\n"
+" b3MprVec3Sub2(&d2, C, x0);\n"
+" b3MprVec3Sub2(&a, x0, P);\n"
+" u = b3MprVec3Dot(&a, &a);\n"
+" v = b3MprVec3Dot(&d1, &d1);\n"
+" w = b3MprVec3Dot(&d2, &d2);\n"
+" p = b3MprVec3Dot(&a, &d1);\n"
+" q = b3MprVec3Dot(&a, &d2);\n"
+" r = b3MprVec3Dot(&d1, &d2);\n"
+" s = (q * r - w * p) / (w * v - r * r);\n"
+" t = (-s * r - q) / w;\n"
+" if ((b3MprIsZero(s) || s > 0.f)\n"
+" && (b3MprEq(s, 1.f) || s < 1.f)\n"
+" && (b3MprIsZero(t) || t > 0.f)\n"
+" && (b3MprEq(t, 1.f) || t < 1.f)\n"
+" && (b3MprEq(t + s, 1.f) || t + s < 1.f)){\n"
+" if (witness){\n"
+" b3MprVec3Scale(&d1, s);\n"
+" b3MprVec3Scale(&d2, t);\n"
+" b3MprVec3Copy(witness, x0);\n"
+" b3MprVec3Add(witness, &d1);\n"
+" b3MprVec3Add(witness, &d2);\n"
+" dist = b3MprVec3Dist2(witness, P);\n"
+" }else{\n"
+" dist = s * s * v;\n"
+" dist += t * t * w;\n"
+" dist += 2.f * s * t * r;\n"
+" dist += 2.f * s * p;\n"
+" dist += 2.f * t * q;\n"
+" dist += u;\n"
+" }\n"
+" }else{\n"
+" dist = _b3MprVec3PointSegmentDist2(P, x0, B, witness);\n"
+" dist2 = _b3MprVec3PointSegmentDist2(P, x0, C, &witness2);\n"
+" if (dist2 < dist){\n"
+" dist = dist2;\n"
+" if (witness)\n"
+" b3MprVec3Copy(witness, &witness2);\n"
+" }\n"
+" dist2 = _b3MprVec3PointSegmentDist2(P, B, C, &witness2);\n"
+" if (dist2 < dist){\n"
+" dist = dist2;\n"
+" if (witness)\n"
+" b3MprVec3Copy(witness, &witness2);\n"
+" }\n"
+" }\n"
+" return dist;\n"
+"}\n"
+"B3_STATIC void b3FindPenetr(int pairIndex,int bodyIndexA, int bodyIndexB, b3ConstArray(b3RigidBodyData_t) cpuBodyBuf, \n"
+" b3ConstArray(b3ConvexPolyhedronData_t) cpuConvexData, \n"
+" b3ConstArray(b3Collidable_t) cpuCollidables,\n"
+" b3ConstArray(b3Float4) cpuVertices,\n"
+" __global b3Float4* sepAxis,\n"
+" b3MprSimplex_t *portal,\n"
+" float *depth, b3Float4 *pdir, b3Float4 *pos)\n"
+"{\n"
+" b3Float4 dir;\n"
+" b3MprSupport_t v4;\n"
+" unsigned long iterations;\n"
+" b3Float4 zero = b3MakeFloat4(0,0,0,0);\n"
+" b3Float4* b3mpr_vec3_origin = &zero;\n"
+" iterations = 1UL;\n"
+" for (int i=0;i<B3_MPR_MAX_ITERATIONS;i++)\n"
+" //while (1)\n"
+" {\n"
+" // compute portal direction and obtain next support point\n"
+" b3PortalDir(portal, &dir);\n"
+" \n"
+" b3MprSupport(pairIndex,bodyIndexA,bodyIndexB,cpuBodyBuf,cpuConvexData,cpuCollidables,cpuVertices, sepAxis,&dir, &v4);\n"
+" // reached tolerance -> find penetration info\n"
+" if (portalReachTolerance(portal, &v4, &dir)\n"
+" || iterations ==B3_MPR_MAX_ITERATIONS)\n"
+" {\n"
+" *depth = b3MprVec3PointTriDist2(b3mpr_vec3_origin,&b3MprSimplexPoint(portal, 1)->v,&b3MprSimplexPoint(portal, 2)->v,&b3MprSimplexPoint(portal, 3)->v,pdir);\n"
+" *depth = B3_MPR_SQRT(*depth);\n"
+" \n"
+" if (b3MprIsZero((*pdir).x) && b3MprIsZero((*pdir).y) && b3MprIsZero((*pdir).z))\n"
+" {\n"
+" \n"
+" *pdir = dir;\n"
+" } \n"
+" b3MprVec3Normalize(pdir);\n"
+" \n"
+" // barycentric coordinates:\n"
+" b3FindPos(portal, pos);\n"
+" return;\n"
+" }\n"
+" b3ExpandPortal(portal, &v4);\n"
+" iterations++;\n"
+" }\n"
+"}\n"
+"B3_STATIC void b3FindPenetrTouch(b3MprSimplex_t *portal,float *depth, b3Float4 *dir, b3Float4 *pos)\n"
+"{\n"
+" // Touching contact on portal's v1 - so depth is zero and direction\n"
+" // is unimportant and pos can be guessed\n"
+" *depth = 0.f;\n"
+" b3Float4 zero = b3MakeFloat4(0,0,0,0);\n"
+" b3Float4* b3mpr_vec3_origin = &zero;\n"
+" b3MprVec3Copy(dir, b3mpr_vec3_origin);\n"
+" b3MprVec3Copy(pos, &b3MprSimplexPoint(portal, 1)->v1);\n"
+" b3MprVec3Add(pos, &b3MprSimplexPoint(portal, 1)->v2);\n"
+" b3MprVec3Scale(pos, 0.5);\n"
+"}\n"
+"B3_STATIC void b3FindPenetrSegment(b3MprSimplex_t *portal,\n"
+" float *depth, b3Float4 *dir, b3Float4 *pos)\n"
+"{\n"
+" \n"
+" // Origin lies on v0-v1 segment.\n"
+" // Depth is distance to v1, direction also and position must be\n"
+" // computed\n"
+" b3MprVec3Copy(pos, &b3MprSimplexPoint(portal, 1)->v1);\n"
+" b3MprVec3Add(pos, &b3MprSimplexPoint(portal, 1)->v2);\n"
+" b3MprVec3Scale(pos, 0.5f);\n"
+" \n"
+" b3MprVec3Copy(dir, &b3MprSimplexPoint(portal, 1)->v);\n"
+" *depth = B3_MPR_SQRT(b3MprVec3Len2(dir));\n"
+" b3MprVec3Normalize(dir);\n"
+"}\n"
+"inline int b3MprPenetration(int pairIndex, int bodyIndexA, int bodyIndexB,\n"
+" b3ConstArray(b3RigidBodyData_t) cpuBodyBuf,\n"
+" b3ConstArray(b3ConvexPolyhedronData_t) cpuConvexData, \n"
+" b3ConstArray(b3Collidable_t) cpuCollidables,\n"
+" b3ConstArray(b3Float4) cpuVertices,\n"
+" __global b3Float4* sepAxis,\n"
+" __global int* hasSepAxis,\n"
+" float *depthOut, b3Float4* dirOut, b3Float4* posOut)\n"
+"{\n"
+" \n"
+" b3MprSimplex_t portal;\n"
+" \n"
+"// if (!hasSepAxis[pairIndex])\n"
+" // return -1;\n"
+" \n"
+" hasSepAxis[pairIndex] = 0;\n"
+" int res;\n"
+" // Phase 1: Portal discovery\n"
+" res = b3DiscoverPortal(pairIndex,bodyIndexA,bodyIndexB,cpuBodyBuf,cpuConvexData,cpuCollidables,cpuVertices,sepAxis,hasSepAxis, &portal);\n"
+" \n"
+" \n"
+" //sepAxis[pairIndex] = *pdir;//or -dir?\n"
+" switch (res)\n"
+" {\n"
+" case 0:\n"
+" {\n"
+" // Phase 2: Portal refinement\n"
+" \n"
+" res = b3RefinePortal(pairIndex,bodyIndexA,bodyIndexB,cpuBodyBuf,cpuConvexData,cpuCollidables,cpuVertices, sepAxis,&portal);\n"
+" if (res < 0)\n"
+" return -1;\n"
+" // Phase 3. Penetration info\n"
+" b3FindPenetr(pairIndex,bodyIndexA,bodyIndexB,cpuBodyBuf,cpuConvexData,cpuCollidables,cpuVertices, sepAxis,&portal, depthOut, dirOut, posOut);\n"
+" hasSepAxis[pairIndex] = 1;\n"
+" sepAxis[pairIndex] = -*dirOut;\n"
+" break;\n"
+" }\n"
+" case 1:\n"
+" {\n"
+" // Touching contact on portal's v1.\n"
+" b3FindPenetrTouch(&portal, depthOut, dirOut, posOut);\n"
+" break;\n"
+" }\n"
+" case 2:\n"
+" {\n"
+" \n"
+" b3FindPenetrSegment( &portal, depthOut, dirOut, posOut);\n"
+" break;\n"
+" }\n"
+" default:\n"
+" {\n"
+" hasSepAxis[pairIndex]=0;\n"
+" //if (res < 0)\n"
+" //{\n"
+" // Origin isn't inside portal - no collision.\n"
+" return -1;\n"
+" //}\n"
+" }\n"
+" };\n"
+" \n"
+" return 0;\n"
+"};\n"
+"#endif //B3_MPR_PENETRATION_H\n"
+"#ifndef B3_CONTACT4DATA_H\n"
+"#define B3_CONTACT4DATA_H\n"
+"#ifndef B3_FLOAT4_H\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+"#endif \n"
+"#endif //B3_FLOAT4_H\n"
+"typedef struct b3Contact4Data b3Contact4Data_t;\n"
+"struct b3Contact4Data\n"
+"{\n"
+" b3Float4 m_worldPosB[4];\n"
+"// b3Float4 m_localPosA[4];\n"
+"// b3Float4 m_localPosB[4];\n"
+" b3Float4 m_worldNormalOnB; // w: m_nPoints\n"
+" unsigned short m_restituitionCoeffCmp;\n"
+" unsigned short m_frictionCoeffCmp;\n"
+" int m_batchIdx;\n"
+" int m_bodyAPtrAndSignBit;//x:m_bodyAPtr, y:m_bodyBPtr\n"
+" int m_bodyBPtrAndSignBit;\n"
+" int m_childIndexA;\n"
+" int m_childIndexB;\n"
+" int m_unused1;\n"
+" int m_unused2;\n"
+"};\n"
+"inline int b3Contact4Data_getNumPoints(const struct b3Contact4Data* contact)\n"
+"{\n"
+" return (int)contact->m_worldNormalOnB.w;\n"
+"};\n"
+"inline void b3Contact4Data_setNumPoints(struct b3Contact4Data* contact, int numPoints)\n"
+"{\n"
+" contact->m_worldNormalOnB.w = (float)numPoints;\n"
+"};\n"
+"#endif //B3_CONTACT4DATA_H\n"
+"#define AppendInc(x, out) out = atomic_inc(x)\n"
+"#define GET_NPOINTS(x) (x).m_worldNormalOnB.w\n"
+"#ifdef cl_ext_atomic_counters_32\n"
+" #pragma OPENCL EXTENSION cl_ext_atomic_counters_32 : enable\n"
+"#else\n"
+" #define counter32_t volatile __global int*\n"
+"#endif\n"
+"__kernel void mprPenetrationKernel( __global int4* pairs,\n"
+" __global const b3RigidBodyData_t* rigidBodies, \n"
+" __global const b3Collidable_t* collidables,\n"
+" __global const b3ConvexPolyhedronData_t* convexShapes, \n"
+" __global const float4* vertices,\n"
+" __global float4* separatingNormals,\n"
+" __global int* hasSeparatingAxis,\n"
+" __global struct b3Contact4Data* restrict globalContactsOut,\n"
+" counter32_t nGlobalContactsOut,\n"
+" int contactCapacity,\n"
+" int numPairs)\n"
+"{\n"
+" int i = get_global_id(0);\n"
+" int pairIndex = i;\n"
+" if (i<numPairs)\n"
+" {\n"
+" int bodyIndexA = pairs[i].x;\n"
+" int bodyIndexB = pairs[i].y;\n"
+" int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;\n"
+" int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;\n"
+" \n"
+" int shapeIndexA = collidables[collidableIndexA].m_shapeIndex;\n"
+" int shapeIndexB = collidables[collidableIndexB].m_shapeIndex;\n"
+" \n"
+" \n"
+" //once the broadphase avoids static-static pairs, we can remove this test\n"
+" if ((rigidBodies[bodyIndexA].m_invMass==0) &&(rigidBodies[bodyIndexB].m_invMass==0))\n"
+" {\n"
+" return;\n"
+" }\n"
+" \n"
+" if ((collidables[collidableIndexA].m_shapeType!=SHAPE_CONVEX_HULL) ||(collidables[collidableIndexB].m_shapeType!=SHAPE_CONVEX_HULL))\n"
+" {\n"
+" return;\n"
+" }\n"
+" float depthOut;\n"
+" b3Float4 dirOut;\n"
+" b3Float4 posOut;\n"
+" int res = b3MprPenetration(pairIndex, bodyIndexA, bodyIndexB,rigidBodies,convexShapes,collidables,vertices,separatingNormals,hasSeparatingAxis,&depthOut, &dirOut, &posOut);\n"
+" \n"
+" \n"
+" \n"
+" \n"
+" if (res==0)\n"
+" {\n"
+" //add a contact\n"
+" int dstIdx;\n"
+" AppendInc( nGlobalContactsOut, dstIdx );\n"
+" if (dstIdx<contactCapacity)\n"
+" {\n"
+" pairs[pairIndex].z = dstIdx;\n"
+" __global struct b3Contact4Data* c = globalContactsOut + dstIdx;\n"
+" c->m_worldNormalOnB = -dirOut;//normal;\n"
+" c->m_restituitionCoeffCmp = (0.f*0xffff);c->m_frictionCoeffCmp = (0.7f*0xffff);\n"
+" c->m_batchIdx = pairIndex;\n"
+" int bodyA = pairs[pairIndex].x;\n"
+" int bodyB = pairs[pairIndex].y;\n"
+" c->m_bodyAPtrAndSignBit = rigidBodies[bodyA].m_invMass==0 ? -bodyA:bodyA;\n"
+" c->m_bodyBPtrAndSignBit = rigidBodies[bodyB].m_invMass==0 ? -bodyB:bodyB;\n"
+" c->m_childIndexA = -1;\n"
+" c->m_childIndexB = -1;\n"
+" //for (int i=0;i<nContacts;i++)\n"
+" posOut.w = -depthOut;\n"
+" c->m_worldPosB[0] = posOut;//localPoints[contactIdx[i]];\n"
+" GET_NPOINTS(*c) = 1;//nContacts;\n"
+" }\n"
+" }\n"
+" }\n"
+"}\n"
+"typedef float4 Quaternion;\n"
+"#define make_float4 (float4)\n"
+"__inline\n"
+"float dot3F4(float4 a, float4 b)\n"
+"{\n"
+" float4 a1 = make_float4(a.xyz,0.f);\n"
+" float4 b1 = make_float4(b.xyz,0.f);\n"
+" return dot(a1, b1);\n"
+"}\n"
+"__inline\n"
+"float4 cross3(float4 a, float4 b)\n"
+"{\n"
+" return cross(a,b);\n"
+"}\n"
+"__inline\n"
+"Quaternion qtMul(Quaternion a, Quaternion b)\n"
+"{\n"
+" Quaternion ans;\n"
+" ans = cross3( a, b );\n"
+" ans += a.w*b+b.w*a;\n"
+"// ans.w = a.w*b.w - (a.x*b.x+a.y*b.y+a.z*b.z);\n"
+" ans.w = a.w*b.w - dot3F4(a, b);\n"
+" return ans;\n"
+"}\n"
+"__inline\n"
+"Quaternion qtInvert(Quaternion q)\n"
+"{\n"
+" return (Quaternion)(-q.xyz, q.w);\n"
+"}\n"
+"__inline\n"
+"float4 qtRotate(Quaternion q, float4 vec)\n"
+"{\n"
+" Quaternion qInv = qtInvert( q );\n"
+" float4 vcpy = vec;\n"
+" vcpy.w = 0.f;\n"
+" float4 out = qtMul(qtMul(q,vcpy),qInv);\n"
+" return out;\n"
+"}\n"
+"__inline\n"
+"float4 transform(const float4* p, const float4* translation, const Quaternion* orientation)\n"
+"{\n"
+" return qtRotate( *orientation, *p ) + (*translation);\n"
+"}\n"
+"__inline\n"
+"float4 qtInvRotate(const Quaternion q, float4 vec)\n"
+"{\n"
+" return qtRotate( qtInvert( q ), vec );\n"
+"}\n"
+"inline void project(__global const b3ConvexPolyhedronData_t* hull, const float4 pos, const float4 orn, \n"
+"const float4* dir, __global const float4* vertices, float* min, float* max)\n"
+"{\n"
+" min[0] = FLT_MAX;\n"
+" max[0] = -FLT_MAX;\n"
+" int numVerts = hull->m_numVertices;\n"
+" const float4 localDir = qtInvRotate(orn,*dir);\n"
+" float offset = dot(pos,*dir);\n"
+" for(int i=0;i<numVerts;i++)\n"
+" {\n"
+" float dp = dot(vertices[hull->m_vertexOffset+i],localDir);\n"
+" if(dp < min[0]) \n"
+" min[0] = dp;\n"
+" if(dp > max[0]) \n"
+" max[0] = dp;\n"
+" }\n"
+" if(min[0]>max[0])\n"
+" {\n"
+" float tmp = min[0];\n"
+" min[0] = max[0];\n"
+" max[0] = tmp;\n"
+" }\n"
+" min[0] += offset;\n"
+" max[0] += offset;\n"
+"}\n"
+"bool findSeparatingAxisUnitSphere( __global const b3ConvexPolyhedronData_t* hullA, __global const b3ConvexPolyhedronData_t* hullB, \n"
+" const float4 posA1,\n"
+" const float4 ornA,\n"
+" const float4 posB1,\n"
+" const float4 ornB,\n"
+" const float4 DeltaC2,\n"
+" __global const float4* vertices,\n"
+" __global const float4* unitSphereDirections,\n"
+" int numUnitSphereDirections,\n"
+" float4* sep,\n"
+" float* dmin)\n"
+"{\n"
+" \n"
+" float4 posA = posA1;\n"
+" posA.w = 0.f;\n"
+" float4 posB = posB1;\n"
+" posB.w = 0.f;\n"
+" int curPlaneTests=0;\n"
+" int curEdgeEdge = 0;\n"
+" // Test unit sphere directions\n"
+" for (int i=0;i<numUnitSphereDirections;i++)\n"
+" {\n"
+" float4 crossje;\n"
+" crossje = unitSphereDirections[i]; \n"
+" if (dot3F4(DeltaC2,crossje)>0)\n"
+" crossje *= -1.f;\n"
+" {\n"
+" float dist;\n"
+" bool result = true;\n"
+" float Min0,Max0;\n"
+" float Min1,Max1;\n"
+" project(hullA,posA,ornA,&crossje,vertices, &Min0, &Max0);\n"
+" project(hullB,posB,ornB,&crossje,vertices, &Min1, &Max1);\n"
+" \n"
+" if(Max0<Min1 || Max1<Min0)\n"
+" return false;\n"
+" \n"
+" float d0 = Max0 - Min1;\n"
+" float d1 = Max1 - Min0;\n"
+" dist = d0<d1 ? d0:d1;\n"
+" result = true;\n"
+" \n"
+" if(dist<*dmin)\n"
+" {\n"
+" *dmin = dist;\n"
+" *sep = crossje;\n"
+" }\n"
+" }\n"
+" }\n"
+" \n"
+" if((dot3F4(-DeltaC2,*sep))>0.0f)\n"
+" {\n"
+" *sep = -(*sep);\n"
+" }\n"
+" return true;\n"
+"}\n"
+"__kernel void findSeparatingAxisUnitSphereKernel( __global const int4* pairs, \n"
+" __global const b3RigidBodyData_t* rigidBodies, \n"
+" __global const b3Collidable_t* collidables,\n"
+" __global const b3ConvexPolyhedronData_t* convexShapes, \n"
+" __global const float4* vertices,\n"
+" __global const float4* unitSphereDirections,\n"
+" __global float4* separatingNormals,\n"
+" __global int* hasSeparatingAxis,\n"
+" __global float* dmins,\n"
+" int numUnitSphereDirections,\n"
+" int numPairs\n"
+" )\n"
+"{\n"
+" int i = get_global_id(0);\n"
+" \n"
+" if (i<numPairs)\n"
+" {\n"
+" if (hasSeparatingAxis[i])\n"
+" {\n"
+" \n"
+" int bodyIndexA = pairs[i].x;\n"
+" int bodyIndexB = pairs[i].y;\n"
+" \n"
+" int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;\n"
+" int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;\n"
+" \n"
+" int shapeIndexA = collidables[collidableIndexA].m_shapeIndex;\n"
+" int shapeIndexB = collidables[collidableIndexB].m_shapeIndex;\n"
+" \n"
+" \n"
+" int numFacesA = convexShapes[shapeIndexA].m_numFaces;\n"
+" \n"
+" float dmin = dmins[i];\n"
+" \n"
+" float4 posA = rigidBodies[bodyIndexA].m_pos;\n"
+" posA.w = 0.f;\n"
+" float4 posB = rigidBodies[bodyIndexB].m_pos;\n"
+" posB.w = 0.f;\n"
+" float4 c0local = convexShapes[shapeIndexA].m_localCenter;\n"
+" float4 ornA = rigidBodies[bodyIndexA].m_quat;\n"
+" float4 c0 = transform(&c0local, &posA, &ornA);\n"
+" float4 c1local = convexShapes[shapeIndexB].m_localCenter;\n"
+" float4 ornB =rigidBodies[bodyIndexB].m_quat;\n"
+" float4 c1 = transform(&c1local,&posB,&ornB);\n"
+" const float4 DeltaC2 = c0 - c1;\n"
+" float4 sepNormal = separatingNormals[i];\n"
+" \n"
+" int numEdgeEdgeDirections = convexShapes[shapeIndexA].m_numUniqueEdges*convexShapes[shapeIndexB].m_numUniqueEdges;\n"
+" if (numEdgeEdgeDirections>numUnitSphereDirections)\n"
+" {\n"
+" bool sepEE = findSeparatingAxisUnitSphere( &convexShapes[shapeIndexA], &convexShapes[shapeIndexB],posA,ornA,\n"
+" posB,ornB,\n"
+" DeltaC2,\n"
+" vertices,unitSphereDirections,numUnitSphereDirections,&sepNormal,&dmin);\n"
+" if (!sepEE)\n"
+" {\n"
+" hasSeparatingAxis[i] = 0;\n"
+" } else\n"
+" {\n"
+" hasSeparatingAxis[i] = 1;\n"
+" separatingNormals[i] = sepNormal;\n"
+" }\n"
+" }\n"
+" } //if (hasSeparatingAxis[i])\n"
+" }//(i<numPairs)\n"
+"}\n"
+;
diff --git a/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/primitiveContacts.cl b/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/primitiveContacts.cl
new file mode 100644
index 0000000000..9c9e920f13
--- /dev/null
+++ b/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/primitiveContacts.cl
@@ -0,0 +1,1374 @@
+#include "Bullet3Collision/NarrowPhaseCollision/shared/b3Contact4Data.h"
+
+#define SHAPE_CONVEX_HULL 3
+#define SHAPE_PLANE 4
+#define SHAPE_CONCAVE_TRIMESH 5
+#define SHAPE_COMPOUND_OF_CONVEX_HULLS 6
+#define SHAPE_SPHERE 7
+
+
+#pragma OPENCL EXTENSION cl_amd_printf : enable
+#pragma OPENCL EXTENSION cl_khr_local_int32_base_atomics : enable
+#pragma OPENCL EXTENSION cl_khr_global_int32_base_atomics : enable
+#pragma OPENCL EXTENSION cl_khr_local_int32_extended_atomics : enable
+#pragma OPENCL EXTENSION cl_khr_global_int32_extended_atomics : enable
+
+#ifdef cl_ext_atomic_counters_32
+#pragma OPENCL EXTENSION cl_ext_atomic_counters_32 : enable
+#else
+#define counter32_t volatile __global int*
+#endif
+
+#define GET_GROUP_IDX get_group_id(0)
+#define GET_LOCAL_IDX get_local_id(0)
+#define GET_GLOBAL_IDX get_global_id(0)
+#define GET_GROUP_SIZE get_local_size(0)
+#define GET_NUM_GROUPS get_num_groups(0)
+#define GROUP_LDS_BARRIER barrier(CLK_LOCAL_MEM_FENCE)
+#define GROUP_MEM_FENCE mem_fence(CLK_LOCAL_MEM_FENCE)
+#define AtomInc(x) atom_inc(&(x))
+#define AtomInc1(x, out) out = atom_inc(&(x))
+#define AppendInc(x, out) out = atomic_inc(x)
+#define AtomAdd(x, value) atom_add(&(x), value)
+#define AtomCmpxhg(x, cmp, value) atom_cmpxchg( &(x), cmp, value )
+#define AtomXhg(x, value) atom_xchg ( &(x), value )
+
+#define max2 max
+#define min2 min
+
+typedef unsigned int u32;
+
+
+
+
+typedef struct
+{
+ union
+ {
+ float4 m_min;
+ float m_minElems[4];
+ int m_minIndices[4];
+ };
+ union
+ {
+ float4 m_max;
+ float m_maxElems[4];
+ int m_maxIndices[4];
+ };
+} btAabbCL;
+
+///keep this in sync with btCollidable.h
+typedef struct
+{
+ int m_numChildShapes;
+ float m_radius;
+ int m_shapeType;
+ int m_shapeIndex;
+
+} btCollidableGpu;
+
+typedef struct
+{
+ float4 m_childPosition;
+ float4 m_childOrientation;
+ int m_shapeIndex;
+ int m_unused0;
+ int m_unused1;
+ int m_unused2;
+} btGpuChildShape;
+
+#define GET_NPOINTS(x) (x).m_worldNormalOnB.w
+
+typedef struct
+{
+ float4 m_pos;
+ float4 m_quat;
+ float4 m_linVel;
+ float4 m_angVel;
+
+ u32 m_collidableIdx;
+ float m_invMass;
+ float m_restituitionCoeff;
+ float m_frictionCoeff;
+} BodyData;
+
+
+typedef struct
+{
+ float4 m_localCenter;
+ float4 m_extents;
+ float4 mC;
+ float4 mE;
+
+ float m_radius;
+ int m_faceOffset;
+ int m_numFaces;
+ int m_numVertices;
+
+ int m_vertexOffset;
+ int m_uniqueEdgesOffset;
+ int m_numUniqueEdges;
+ int m_unused;
+
+} ConvexPolyhedronCL;
+
+typedef struct
+{
+ float4 m_plane;
+ int m_indexOffset;
+ int m_numIndices;
+} btGpuFace;
+
+#define SELECT_UINT4( b, a, condition ) select( b,a,condition )
+
+#define make_float4 (float4)
+#define make_float2 (float2)
+#define make_uint4 (uint4)
+#define make_int4 (int4)
+#define make_uint2 (uint2)
+#define make_int2 (int2)
+
+
+__inline
+float fastDiv(float numerator, float denominator)
+{
+ return native_divide(numerator, denominator);
+// return numerator/denominator;
+}
+
+__inline
+float4 fastDiv4(float4 numerator, float4 denominator)
+{
+ return native_divide(numerator, denominator);
+}
+
+
+__inline
+float4 cross3(float4 a, float4 b)
+{
+ return cross(a,b);
+}
+
+//#define dot3F4 dot
+
+__inline
+float dot3F4(float4 a, float4 b)
+{
+ float4 a1 = make_float4(a.xyz,0.f);
+ float4 b1 = make_float4(b.xyz,0.f);
+ return dot(a1, b1);
+}
+
+__inline
+float4 fastNormalize4(float4 v)
+{
+ return fast_normalize(v);
+}
+
+
+///////////////////////////////////////
+// Quaternion
+///////////////////////////////////////
+
+typedef float4 Quaternion;
+
+__inline
+Quaternion qtMul(Quaternion a, Quaternion b);
+
+__inline
+Quaternion qtNormalize(Quaternion in);
+
+__inline
+float4 qtRotate(Quaternion q, float4 vec);
+
+__inline
+Quaternion qtInvert(Quaternion q);
+
+
+
+
+__inline
+Quaternion qtMul(Quaternion a, Quaternion b)
+{
+ Quaternion ans;
+ ans = cross3( a, b );
+ ans += a.w*b+b.w*a;
+// ans.w = a.w*b.w - (a.x*b.x+a.y*b.y+a.z*b.z);
+ ans.w = a.w*b.w - dot3F4(a, b);
+ return ans;
+}
+
+__inline
+Quaternion qtNormalize(Quaternion in)
+{
+ return fastNormalize4(in);
+// in /= length( in );
+// return in;
+}
+__inline
+float4 qtRotate(Quaternion q, float4 vec)
+{
+ Quaternion qInv = qtInvert( q );
+ float4 vcpy = vec;
+ vcpy.w = 0.f;
+ float4 out = qtMul(qtMul(q,vcpy),qInv);
+ return out;
+}
+
+__inline
+Quaternion qtInvert(Quaternion q)
+{
+ return (Quaternion)(-q.xyz, q.w);
+}
+
+__inline
+float4 qtInvRotate(const Quaternion q, float4 vec)
+{
+ return qtRotate( qtInvert( q ), vec );
+}
+
+__inline
+float4 transform(const float4* p, const float4* translation, const Quaternion* orientation)
+{
+ return qtRotate( *orientation, *p ) + (*translation);
+}
+
+void trInverse(float4 translationIn, Quaternion orientationIn,
+ float4* translationOut, Quaternion* orientationOut)
+{
+ *orientationOut = qtInvert(orientationIn);
+ *translationOut = qtRotate(*orientationOut, -translationIn);
+}
+
+void trMul(float4 translationA, Quaternion orientationA,
+ float4 translationB, Quaternion orientationB,
+ float4* translationOut, Quaternion* orientationOut)
+{
+ *orientationOut = qtMul(orientationA,orientationB);
+ *translationOut = transform(&translationB,&translationA,&orientationA);
+}
+
+
+
+__inline
+float4 normalize3(const float4 a)
+{
+ float4 n = make_float4(a.x, a.y, a.z, 0.f);
+ return fastNormalize4( n );
+}
+
+
+__inline float4 lerp3(const float4 a,const float4 b, float t)
+{
+ return make_float4( a.x + (b.x - a.x) * t,
+ a.y + (b.y - a.y) * t,
+ a.z + (b.z - a.z) * t,
+ 0.f);
+}
+
+
+float signedDistanceFromPointToPlane(float4 point, float4 planeEqn, float4* closestPointOnFace)
+{
+ float4 n = (float4)(planeEqn.x, planeEqn.y, planeEqn.z, 0);
+ float dist = dot3F4(n, point) + planeEqn.w;
+ *closestPointOnFace = point - dist * n;
+ return dist;
+}
+
+
+
+inline bool IsPointInPolygon(float4 p,
+ const btGpuFace* face,
+ __global const float4* baseVertex,
+ __global const int* convexIndices,
+ float4* out)
+{
+ float4 a;
+ float4 b;
+ float4 ab;
+ float4 ap;
+ float4 v;
+
+ float4 plane = make_float4(face->m_plane.x,face->m_plane.y,face->m_plane.z,0.f);
+
+ if (face->m_numIndices<2)
+ return false;
+
+
+ float4 v0 = baseVertex[convexIndices[face->m_indexOffset + face->m_numIndices-1]];
+
+ b = v0;
+
+ for(unsigned i=0; i != face->m_numIndices; ++i)
+ {
+ a = b;
+ float4 vi = baseVertex[convexIndices[face->m_indexOffset + i]];
+ b = vi;
+ ab = b-a;
+ ap = p-a;
+ v = cross3(ab,plane);
+
+ if (dot(ap, v) > 0.f)
+ {
+ float ab_m2 = dot(ab, ab);
+ float rt = ab_m2 != 0.f ? dot(ab, ap) / ab_m2 : 0.f;
+ if (rt <= 0.f)
+ {
+ *out = a;
+ }
+ else if (rt >= 1.f)
+ {
+ *out = b;
+ }
+ else
+ {
+ float s = 1.f - rt;
+ out[0].x = s * a.x + rt * b.x;
+ out[0].y = s * a.y + rt * b.y;
+ out[0].z = s * a.z + rt * b.z;
+ }
+ return false;
+ }
+ }
+ return true;
+}
+
+
+
+
+void computeContactSphereConvex(int pairIndex,
+ int bodyIndexA, int bodyIndexB,
+ int collidableIndexA, int collidableIndexB,
+ __global const BodyData* rigidBodies,
+ __global const btCollidableGpu* collidables,
+ __global const ConvexPolyhedronCL* convexShapes,
+ __global const float4* convexVertices,
+ __global const int* convexIndices,
+ __global const btGpuFace* faces,
+ __global struct b3Contact4Data* restrict globalContactsOut,
+ counter32_t nGlobalContactsOut,
+ int maxContactCapacity,
+ float4 spherePos2,
+ float radius,
+ float4 pos,
+ float4 quat
+ )
+{
+
+ float4 invPos;
+ float4 invOrn;
+
+ trInverse(pos,quat, &invPos,&invOrn);
+
+ float4 spherePos = transform(&spherePos2,&invPos,&invOrn);
+
+ int shapeIndex = collidables[collidableIndexB].m_shapeIndex;
+ int numFaces = convexShapes[shapeIndex].m_numFaces;
+ float4 closestPnt = (float4)(0, 0, 0, 0);
+ float4 hitNormalWorld = (float4)(0, 0, 0, 0);
+ float minDist = -1000000.f;
+ bool bCollide = true;
+
+ for ( int f = 0; f < numFaces; f++ )
+ {
+ btGpuFace face = faces[convexShapes[shapeIndex].m_faceOffset+f];
+
+ // set up a plane equation
+ float4 planeEqn;
+ float4 n1 = face.m_plane;
+ n1.w = 0.f;
+ planeEqn = n1;
+ planeEqn.w = face.m_plane.w;
+
+
+ // compute a signed distance from the vertex in cloth to the face of rigidbody.
+ float4 pntReturn;
+ float dist = signedDistanceFromPointToPlane(spherePos, planeEqn, &pntReturn);
+
+ // If the distance is positive, the plane is a separating plane.
+ if ( dist > radius )
+ {
+ bCollide = false;
+ break;
+ }
+
+
+ if (dist>0)
+ {
+ //might hit an edge or vertex
+ float4 out;
+ float4 zeroPos = make_float4(0,0,0,0);
+
+ bool isInPoly = IsPointInPolygon(spherePos,
+ &face,
+ &convexVertices[convexShapes[shapeIndex].m_vertexOffset],
+ convexIndices,
+ &out);
+ if (isInPoly)
+ {
+ if (dist>minDist)
+ {
+ minDist = dist;
+ closestPnt = pntReturn;
+ hitNormalWorld = planeEqn;
+
+ }
+ } else
+ {
+ float4 tmp = spherePos-out;
+ float l2 = dot(tmp,tmp);
+ if (l2<radius*radius)
+ {
+ dist = sqrt(l2);
+ if (dist>minDist)
+ {
+ minDist = dist;
+ closestPnt = out;
+ hitNormalWorld = tmp/dist;
+
+ }
+
+ } else
+ {
+ bCollide = false;
+ break;
+ }
+ }
+ } else
+ {
+ if ( dist > minDist )
+ {
+ minDist = dist;
+ closestPnt = pntReturn;
+ hitNormalWorld.xyz = planeEqn.xyz;
+ }
+ }
+
+ }
+
+
+
+ if (bCollide && minDist > -10000)
+ {
+ float4 normalOnSurfaceB1 = qtRotate(quat,-hitNormalWorld);
+ float4 pOnB1 = transform(&closestPnt,&pos,&quat);
+
+ float actualDepth = minDist-radius;
+ if (actualDepth<=0.f)
+ {
+
+
+ pOnB1.w = actualDepth;
+
+ int dstIdx;
+ AppendInc( nGlobalContactsOut, dstIdx );
+
+
+ if (1)//dstIdx < maxContactCapacity)
+ {
+ __global struct b3Contact4Data* c = &globalContactsOut[dstIdx];
+ c->m_worldNormalOnB = -normalOnSurfaceB1;
+ c->m_restituitionCoeffCmp = (0.f*0xffff);c->m_frictionCoeffCmp = (0.7f*0xffff);
+ c->m_batchIdx = pairIndex;
+ c->m_bodyAPtrAndSignBit = rigidBodies[bodyIndexA].m_invMass==0?-bodyIndexA:bodyIndexA;
+ c->m_bodyBPtrAndSignBit = rigidBodies[bodyIndexB].m_invMass==0?-bodyIndexB:bodyIndexB;
+ c->m_worldPosB[0] = pOnB1;
+ c->m_childIndexA = -1;
+ c->m_childIndexB = -1;
+
+ GET_NPOINTS(*c) = 1;
+ }
+
+ }
+ }//if (hasCollision)
+
+}
+
+
+
+int extractManifoldSequential(const float4* p, int nPoints, float4 nearNormal, int4* contactIdx)
+{
+ if( nPoints == 0 )
+ return 0;
+
+ if (nPoints <=4)
+ return nPoints;
+
+
+ if (nPoints >64)
+ nPoints = 64;
+
+ float4 center = make_float4(0.f);
+ {
+
+ for (int i=0;i<nPoints;i++)
+ center += p[i];
+ center /= (float)nPoints;
+ }
+
+
+
+ // sample 4 directions
+
+ float4 aVector = p[0] - center;
+ float4 u = cross3( nearNormal, aVector );
+ float4 v = cross3( nearNormal, u );
+ u = normalize3( u );
+ v = normalize3( v );
+
+
+ //keep point with deepest penetration
+ float minW= FLT_MAX;
+
+ int minIndex=-1;
+
+ float4 maxDots;
+ maxDots.x = FLT_MIN;
+ maxDots.y = FLT_MIN;
+ maxDots.z = FLT_MIN;
+ maxDots.w = FLT_MIN;
+
+ // idx, distance
+ for(int ie = 0; ie<nPoints; ie++ )
+ {
+ if (p[ie].w<minW)
+ {
+ minW = p[ie].w;
+ minIndex=ie;
+ }
+ float f;
+ float4 r = p[ie]-center;
+ f = dot3F4( u, r );
+ if (f<maxDots.x)
+ {
+ maxDots.x = f;
+ contactIdx[0].x = ie;
+ }
+
+ f = dot3F4( -u, r );
+ if (f<maxDots.y)
+ {
+ maxDots.y = f;
+ contactIdx[0].y = ie;
+ }
+
+
+ f = dot3F4( v, r );
+ if (f<maxDots.z)
+ {
+ maxDots.z = f;
+ contactIdx[0].z = ie;
+ }
+
+ f = dot3F4( -v, r );
+ if (f<maxDots.w)
+ {
+ maxDots.w = f;
+ contactIdx[0].w = ie;
+ }
+
+ }
+
+ if (contactIdx[0].x != minIndex && contactIdx[0].y != minIndex && contactIdx[0].z != minIndex && contactIdx[0].w != minIndex)
+ {
+ //replace the first contact with minimum (todo: replace contact with least penetration)
+ contactIdx[0].x = minIndex;
+ }
+
+ return 4;
+
+}
+
+#define MAX_PLANE_CONVEX_POINTS 64
+
+int computeContactPlaneConvex(int pairIndex,
+ int bodyIndexA, int bodyIndexB,
+ int collidableIndexA, int collidableIndexB,
+ __global const BodyData* rigidBodies,
+ __global const btCollidableGpu*collidables,
+ __global const ConvexPolyhedronCL* convexShapes,
+ __global const float4* convexVertices,
+ __global const int* convexIndices,
+ __global const btGpuFace* faces,
+ __global struct b3Contact4Data* restrict globalContactsOut,
+ counter32_t nGlobalContactsOut,
+ int maxContactCapacity,
+ float4 posB,
+ Quaternion ornB
+ )
+{
+ int resultIndex=-1;
+
+ int shapeIndex = collidables[collidableIndexB].m_shapeIndex;
+ __global const ConvexPolyhedronCL* hullB = &convexShapes[shapeIndex];
+
+ float4 posA;
+ posA = rigidBodies[bodyIndexA].m_pos;
+ Quaternion ornA;
+ ornA = rigidBodies[bodyIndexA].m_quat;
+
+ int numContactsOut = 0;
+ int numWorldVertsB1= 0;
+
+ float4 planeEq;
+ planeEq = faces[collidables[collidableIndexA].m_shapeIndex].m_plane;
+ float4 planeNormal = make_float4(planeEq.x,planeEq.y,planeEq.z,0.f);
+ float4 planeNormalWorld;
+ planeNormalWorld = qtRotate(ornA,planeNormal);
+ float planeConstant = planeEq.w;
+
+ float4 invPosA;Quaternion invOrnA;
+ float4 convexInPlaneTransPos1; Quaternion convexInPlaneTransOrn1;
+ {
+
+ trInverse(posA,ornA,&invPosA,&invOrnA);
+ trMul(invPosA,invOrnA,posB,ornB,&convexInPlaneTransPos1,&convexInPlaneTransOrn1);
+ }
+ float4 invPosB;Quaternion invOrnB;
+ float4 planeInConvexPos1; Quaternion planeInConvexOrn1;
+ {
+
+ trInverse(posB,ornB,&invPosB,&invOrnB);
+ trMul(invPosB,invOrnB,posA,ornA,&planeInConvexPos1,&planeInConvexOrn1);
+ }
+
+
+ float4 planeNormalInConvex = qtRotate(planeInConvexOrn1,-planeNormal);
+ float maxDot = -1e30;
+ int hitVertex=-1;
+ float4 hitVtx;
+
+
+
+ float4 contactPoints[MAX_PLANE_CONVEX_POINTS];
+ int numPoints = 0;
+
+ int4 contactIdx;
+ contactIdx=make_int4(0,1,2,3);
+
+
+ for (int i=0;i<hullB->m_numVertices;i++)
+ {
+ float4 vtx = convexVertices[hullB->m_vertexOffset+i];
+ float curDot = dot(vtx,planeNormalInConvex);
+
+
+ if (curDot>maxDot)
+ {
+ hitVertex=i;
+ maxDot=curDot;
+ hitVtx = vtx;
+ //make sure the deepest points is always included
+ if (numPoints==MAX_PLANE_CONVEX_POINTS)
+ numPoints--;
+ }
+
+ if (numPoints<MAX_PLANE_CONVEX_POINTS)
+ {
+ float4 vtxWorld = transform(&vtx, &posB, &ornB);
+ float4 vtxInPlane = transform(&vtxWorld, &invPosA, &invOrnA);//oplaneTransform.inverse()*vtxWorld;
+ float dist = dot(planeNormal,vtxInPlane)-planeConstant;
+ if (dist<0.f)
+ {
+ vtxWorld.w = dist;
+ contactPoints[numPoints] = vtxWorld;
+ numPoints++;
+ }
+ }
+
+ }
+
+ int numReducedPoints = numPoints;
+ if (numPoints>4)
+ {
+ numReducedPoints = extractManifoldSequential( contactPoints, numPoints, planeNormalInConvex, &contactIdx);
+ }
+
+ if (numReducedPoints>0)
+ {
+ int dstIdx;
+ AppendInc( nGlobalContactsOut, dstIdx );
+
+ if (dstIdx < maxContactCapacity)
+ {
+ resultIndex = dstIdx;
+ __global struct b3Contact4Data* c = &globalContactsOut[dstIdx];
+ c->m_worldNormalOnB = -planeNormalWorld;
+ //c->setFrictionCoeff(0.7);
+ //c->setRestituitionCoeff(0.f);
+ c->m_restituitionCoeffCmp = (0.f*0xffff);c->m_frictionCoeffCmp = (0.7f*0xffff);
+ c->m_batchIdx = pairIndex;
+ c->m_bodyAPtrAndSignBit = rigidBodies[bodyIndexA].m_invMass==0?-bodyIndexA:bodyIndexA;
+ c->m_bodyBPtrAndSignBit = rigidBodies[bodyIndexB].m_invMass==0?-bodyIndexB:bodyIndexB;
+ c->m_childIndexA = -1;
+ c->m_childIndexB = -1;
+
+ switch (numReducedPoints)
+ {
+ case 4:
+ c->m_worldPosB[3] = contactPoints[contactIdx.w];
+ case 3:
+ c->m_worldPosB[2] = contactPoints[contactIdx.z];
+ case 2:
+ c->m_worldPosB[1] = contactPoints[contactIdx.y];
+ case 1:
+ c->m_worldPosB[0] = contactPoints[contactIdx.x];
+ default:
+ {
+ }
+ };
+
+ GET_NPOINTS(*c) = numReducedPoints;
+ }//if (dstIdx < numPairs)
+ }
+
+ return resultIndex;
+}
+
+
+void computeContactPlaneSphere(int pairIndex,
+ int bodyIndexA, int bodyIndexB,
+ int collidableIndexA, int collidableIndexB,
+ __global const BodyData* rigidBodies,
+ __global const btCollidableGpu* collidables,
+ __global const btGpuFace* faces,
+ __global struct b3Contact4Data* restrict globalContactsOut,
+ counter32_t nGlobalContactsOut,
+ int maxContactCapacity)
+{
+ float4 planeEq = faces[collidables[collidableIndexA].m_shapeIndex].m_plane;
+ float radius = collidables[collidableIndexB].m_radius;
+ float4 posA1 = rigidBodies[bodyIndexA].m_pos;
+ float4 ornA1 = rigidBodies[bodyIndexA].m_quat;
+ float4 posB1 = rigidBodies[bodyIndexB].m_pos;
+ float4 ornB1 = rigidBodies[bodyIndexB].m_quat;
+
+ bool hasCollision = false;
+ float4 planeNormal1 = make_float4(planeEq.x,planeEq.y,planeEq.z,0.f);
+ float planeConstant = planeEq.w;
+ float4 convexInPlaneTransPos1; Quaternion convexInPlaneTransOrn1;
+ {
+ float4 invPosA;Quaternion invOrnA;
+ trInverse(posA1,ornA1,&invPosA,&invOrnA);
+ trMul(invPosA,invOrnA,posB1,ornB1,&convexInPlaneTransPos1,&convexInPlaneTransOrn1);
+ }
+ float4 planeInConvexPos1; Quaternion planeInConvexOrn1;
+ {
+ float4 invPosB;Quaternion invOrnB;
+ trInverse(posB1,ornB1,&invPosB,&invOrnB);
+ trMul(invPosB,invOrnB,posA1,ornA1,&planeInConvexPos1,&planeInConvexOrn1);
+ }
+ float4 vtx1 = qtRotate(planeInConvexOrn1,-planeNormal1)*radius;
+ float4 vtxInPlane1 = transform(&vtx1,&convexInPlaneTransPos1,&convexInPlaneTransOrn1);
+ float distance = dot3F4(planeNormal1,vtxInPlane1) - planeConstant;
+ hasCollision = distance < 0.f;//m_manifoldPtr->getContactBreakingThreshold();
+ if (hasCollision)
+ {
+ float4 vtxInPlaneProjected1 = vtxInPlane1 - distance*planeNormal1;
+ float4 vtxInPlaneWorld1 = transform(&vtxInPlaneProjected1,&posA1,&ornA1);
+ float4 normalOnSurfaceB1 = qtRotate(ornA1,planeNormal1);
+ float4 pOnB1 = vtxInPlaneWorld1+normalOnSurfaceB1*distance;
+ pOnB1.w = distance;
+
+ int dstIdx;
+ AppendInc( nGlobalContactsOut, dstIdx );
+
+ if (dstIdx < maxContactCapacity)
+ {
+ __global struct b3Contact4Data* c = &globalContactsOut[dstIdx];
+ c->m_worldNormalOnB = -normalOnSurfaceB1;
+ c->m_restituitionCoeffCmp = (0.f*0xffff);c->m_frictionCoeffCmp = (0.7f*0xffff);
+ c->m_batchIdx = pairIndex;
+ c->m_bodyAPtrAndSignBit = rigidBodies[bodyIndexA].m_invMass==0?-bodyIndexA:bodyIndexA;
+ c->m_bodyBPtrAndSignBit = rigidBodies[bodyIndexB].m_invMass==0?-bodyIndexB:bodyIndexB;
+ c->m_worldPosB[0] = pOnB1;
+ c->m_childIndexA = -1;
+ c->m_childIndexB = -1;
+ GET_NPOINTS(*c) = 1;
+ }//if (dstIdx < numPairs)
+ }//if (hasCollision)
+}
+
+
+__kernel void primitiveContactsKernel( __global int4* pairs,
+ __global const BodyData* rigidBodies,
+ __global const btCollidableGpu* collidables,
+ __global const ConvexPolyhedronCL* convexShapes,
+ __global const float4* vertices,
+ __global const float4* uniqueEdges,
+ __global const btGpuFace* faces,
+ __global const int* indices,
+ __global struct b3Contact4Data* restrict globalContactsOut,
+ counter32_t nGlobalContactsOut,
+ int numPairs, int maxContactCapacity)
+{
+
+ int i = get_global_id(0);
+ int pairIndex = i;
+
+ float4 worldVertsB1[64];
+ float4 worldVertsB2[64];
+ int capacityWorldVerts = 64;
+
+ float4 localContactsOut[64];
+ int localContactCapacity=64;
+
+ float minDist = -1e30f;
+ float maxDist = 0.02f;
+
+ if (i<numPairs)
+ {
+
+ int bodyIndexA = pairs[i].x;
+ int bodyIndexB = pairs[i].y;
+
+ int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;
+ int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;
+
+ if (collidables[collidableIndexA].m_shapeType == SHAPE_PLANE &&
+ collidables[collidableIndexB].m_shapeType == SHAPE_CONVEX_HULL)
+ {
+
+ float4 posB;
+ posB = rigidBodies[bodyIndexB].m_pos;
+ Quaternion ornB;
+ ornB = rigidBodies[bodyIndexB].m_quat;
+ int contactIndex = computeContactPlaneConvex(pairIndex, bodyIndexA, bodyIndexB, collidableIndexA, collidableIndexB,
+ rigidBodies,collidables,convexShapes,vertices,indices,
+ faces, globalContactsOut, nGlobalContactsOut,maxContactCapacity, posB,ornB);
+ if (contactIndex>=0)
+ pairs[pairIndex].z = contactIndex;
+
+ return;
+ }
+
+
+ if (collidables[collidableIndexA].m_shapeType == SHAPE_CONVEX_HULL &&
+ collidables[collidableIndexB].m_shapeType == SHAPE_PLANE)
+ {
+
+ float4 posA;
+ posA = rigidBodies[bodyIndexA].m_pos;
+ Quaternion ornA;
+ ornA = rigidBodies[bodyIndexA].m_quat;
+
+
+ int contactIndex = computeContactPlaneConvex( pairIndex, bodyIndexB,bodyIndexA, collidableIndexB,collidableIndexA,
+ rigidBodies,collidables,convexShapes,vertices,indices,
+ faces, globalContactsOut, nGlobalContactsOut,maxContactCapacity,posA,ornA);
+
+ if (contactIndex>=0)
+ pairs[pairIndex].z = contactIndex;
+
+ return;
+ }
+
+ if (collidables[collidableIndexA].m_shapeType == SHAPE_PLANE &&
+ collidables[collidableIndexB].m_shapeType == SHAPE_SPHERE)
+ {
+ computeContactPlaneSphere(pairIndex, bodyIndexA, bodyIndexB, collidableIndexA, collidableIndexB,
+ rigidBodies,collidables,faces, globalContactsOut, nGlobalContactsOut,maxContactCapacity);
+ return;
+ }
+
+
+ if (collidables[collidableIndexA].m_shapeType == SHAPE_SPHERE &&
+ collidables[collidableIndexB].m_shapeType == SHAPE_PLANE)
+ {
+
+
+ computeContactPlaneSphere( pairIndex, bodyIndexB,bodyIndexA, collidableIndexB,collidableIndexA,
+ rigidBodies,collidables,
+ faces, globalContactsOut, nGlobalContactsOut,maxContactCapacity);
+
+ return;
+ }
+
+
+
+
+ if (collidables[collidableIndexA].m_shapeType == SHAPE_SPHERE &&
+ collidables[collidableIndexB].m_shapeType == SHAPE_CONVEX_HULL)
+ {
+
+ float4 spherePos = rigidBodies[bodyIndexA].m_pos;
+ float sphereRadius = collidables[collidableIndexA].m_radius;
+ float4 convexPos = rigidBodies[bodyIndexB].m_pos;
+ float4 convexOrn = rigidBodies[bodyIndexB].m_quat;
+
+ computeContactSphereConvex(pairIndex, bodyIndexA, bodyIndexB, collidableIndexA, collidableIndexB,
+ rigidBodies,collidables,convexShapes,vertices,indices,faces, globalContactsOut, nGlobalContactsOut,maxContactCapacity,
+ spherePos,sphereRadius,convexPos,convexOrn);
+
+ return;
+ }
+
+ if (collidables[collidableIndexA].m_shapeType == SHAPE_CONVEX_HULL &&
+ collidables[collidableIndexB].m_shapeType == SHAPE_SPHERE)
+ {
+
+ float4 spherePos = rigidBodies[bodyIndexB].m_pos;
+ float sphereRadius = collidables[collidableIndexB].m_radius;
+ float4 convexPos = rigidBodies[bodyIndexA].m_pos;
+ float4 convexOrn = rigidBodies[bodyIndexA].m_quat;
+
+ computeContactSphereConvex(pairIndex, bodyIndexB, bodyIndexA, collidableIndexB, collidableIndexA,
+ rigidBodies,collidables,convexShapes,vertices,indices,faces, globalContactsOut, nGlobalContactsOut,maxContactCapacity,
+ spherePos,sphereRadius,convexPos,convexOrn);
+ return;
+ }
+
+
+
+
+
+
+ if (collidables[collidableIndexA].m_shapeType == SHAPE_SPHERE &&
+ collidables[collidableIndexB].m_shapeType == SHAPE_SPHERE)
+ {
+ //sphere-sphere
+ float radiusA = collidables[collidableIndexA].m_radius;
+ float radiusB = collidables[collidableIndexB].m_radius;
+ float4 posA = rigidBodies[bodyIndexA].m_pos;
+ float4 posB = rigidBodies[bodyIndexB].m_pos;
+
+ float4 diff = posA-posB;
+ float len = length(diff);
+
+ ///iff distance positive, don't generate a new contact
+ if ( len <= (radiusA+radiusB))
+ {
+ ///distance (negative means penetration)
+ float dist = len - (radiusA+radiusB);
+ float4 normalOnSurfaceB = make_float4(1.f,0.f,0.f,0.f);
+ if (len > 0.00001)
+ {
+ normalOnSurfaceB = diff / len;
+ }
+ float4 contactPosB = posB + normalOnSurfaceB*radiusB;
+ contactPosB.w = dist;
+
+ int dstIdx;
+ AppendInc( nGlobalContactsOut, dstIdx );
+
+ if (dstIdx < maxContactCapacity)
+ {
+ __global struct b3Contact4Data* c = &globalContactsOut[dstIdx];
+ c->m_worldNormalOnB = normalOnSurfaceB;
+ c->m_restituitionCoeffCmp = (0.f*0xffff);c->m_frictionCoeffCmp = (0.7f*0xffff);
+ c->m_batchIdx = pairIndex;
+ int bodyA = pairs[pairIndex].x;
+ int bodyB = pairs[pairIndex].y;
+ c->m_bodyAPtrAndSignBit = rigidBodies[bodyA].m_invMass==0?-bodyA:bodyA;
+ c->m_bodyBPtrAndSignBit = rigidBodies[bodyB].m_invMass==0?-bodyB:bodyB;
+ c->m_worldPosB[0] = contactPosB;
+ c->m_childIndexA = -1;
+ c->m_childIndexB = -1;
+ GET_NPOINTS(*c) = 1;
+ }//if (dstIdx < numPairs)
+ }//if ( len <= (radiusA+radiusB))
+
+ return;
+ }//SHAPE_SPHERE SHAPE_SPHERE
+
+ }// if (i<numPairs)
+
+}
+
+
+// work-in-progress
+__kernel void processCompoundPairsPrimitivesKernel( __global const int4* gpuCompoundPairs,
+ __global const BodyData* rigidBodies,
+ __global const btCollidableGpu* collidables,
+ __global const ConvexPolyhedronCL* convexShapes,
+ __global const float4* vertices,
+ __global const float4* uniqueEdges,
+ __global const btGpuFace* faces,
+ __global const int* indices,
+ __global btAabbCL* aabbs,
+ __global const btGpuChildShape* gpuChildShapes,
+ __global struct b3Contact4Data* restrict globalContactsOut,
+ counter32_t nGlobalContactsOut,
+ int numCompoundPairs, int maxContactCapacity
+ )
+{
+
+ int i = get_global_id(0);
+ if (i<numCompoundPairs)
+ {
+ int bodyIndexA = gpuCompoundPairs[i].x;
+ int bodyIndexB = gpuCompoundPairs[i].y;
+
+ int childShapeIndexA = gpuCompoundPairs[i].z;
+ int childShapeIndexB = gpuCompoundPairs[i].w;
+
+ int collidableIndexA = -1;
+ int collidableIndexB = -1;
+
+ float4 ornA = rigidBodies[bodyIndexA].m_quat;
+ float4 posA = rigidBodies[bodyIndexA].m_pos;
+
+ float4 ornB = rigidBodies[bodyIndexB].m_quat;
+ float4 posB = rigidBodies[bodyIndexB].m_pos;
+
+ if (childShapeIndexA >= 0)
+ {
+ collidableIndexA = gpuChildShapes[childShapeIndexA].m_shapeIndex;
+ float4 childPosA = gpuChildShapes[childShapeIndexA].m_childPosition;
+ float4 childOrnA = gpuChildShapes[childShapeIndexA].m_childOrientation;
+ float4 newPosA = qtRotate(ornA,childPosA)+posA;
+ float4 newOrnA = qtMul(ornA,childOrnA);
+ posA = newPosA;
+ ornA = newOrnA;
+ } else
+ {
+ collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;
+ }
+
+ if (childShapeIndexB>=0)
+ {
+ collidableIndexB = gpuChildShapes[childShapeIndexB].m_shapeIndex;
+ float4 childPosB = gpuChildShapes[childShapeIndexB].m_childPosition;
+ float4 childOrnB = gpuChildShapes[childShapeIndexB].m_childOrientation;
+ float4 newPosB = transform(&childPosB,&posB,&ornB);
+ float4 newOrnB = qtMul(ornB,childOrnB);
+ posB = newPosB;
+ ornB = newOrnB;
+ } else
+ {
+ collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;
+ }
+
+ int shapeIndexA = collidables[collidableIndexA].m_shapeIndex;
+ int shapeIndexB = collidables[collidableIndexB].m_shapeIndex;
+
+ int shapeTypeA = collidables[collidableIndexA].m_shapeType;
+ int shapeTypeB = collidables[collidableIndexB].m_shapeType;
+
+ int pairIndex = i;
+ if ((shapeTypeA == SHAPE_PLANE) && (shapeTypeB==SHAPE_CONVEX_HULL))
+ {
+
+ computeContactPlaneConvex( pairIndex, bodyIndexA,bodyIndexB, collidableIndexA,collidableIndexB,
+ rigidBodies,collidables,convexShapes,vertices,indices,
+ faces, globalContactsOut, nGlobalContactsOut,maxContactCapacity,posB,ornB);
+ return;
+ }
+
+ if ((shapeTypeA == SHAPE_CONVEX_HULL) && (shapeTypeB==SHAPE_PLANE))
+ {
+
+ computeContactPlaneConvex( pairIndex, bodyIndexB,bodyIndexA, collidableIndexB,collidableIndexA,
+ rigidBodies,collidables,convexShapes,vertices,indices,
+ faces, globalContactsOut, nGlobalContactsOut,maxContactCapacity,posA,ornA);
+ return;
+ }
+
+ if ((shapeTypeA == SHAPE_CONVEX_HULL) && (shapeTypeB == SHAPE_SPHERE))
+ {
+ float4 spherePos = rigidBodies[bodyIndexB].m_pos;
+ float sphereRadius = collidables[collidableIndexB].m_radius;
+ float4 convexPos = posA;
+ float4 convexOrn = ornA;
+
+ computeContactSphereConvex(pairIndex, bodyIndexB, bodyIndexA , collidableIndexB,collidableIndexA,
+ rigidBodies,collidables,convexShapes,vertices,indices,faces, globalContactsOut, nGlobalContactsOut,maxContactCapacity,
+ spherePos,sphereRadius,convexPos,convexOrn);
+
+ return;
+ }
+
+ if ((shapeTypeA == SHAPE_SPHERE) && (shapeTypeB == SHAPE_CONVEX_HULL))
+ {
+
+ float4 spherePos = rigidBodies[bodyIndexA].m_pos;
+ float sphereRadius = collidables[collidableIndexA].m_radius;
+ float4 convexPos = posB;
+ float4 convexOrn = ornB;
+
+
+ computeContactSphereConvex(pairIndex, bodyIndexA, bodyIndexB, collidableIndexA, collidableIndexB,
+ rigidBodies,collidables,convexShapes,vertices,indices,faces, globalContactsOut, nGlobalContactsOut,maxContactCapacity,
+ spherePos,sphereRadius,convexPos,convexOrn);
+
+ return;
+ }
+ }// if (i<numCompoundPairs)
+}
+
+
+bool pointInTriangle(const float4* vertices, const float4* normal, float4 *p )
+{
+
+ const float4* p1 = &vertices[0];
+ const float4* p2 = &vertices[1];
+ const float4* p3 = &vertices[2];
+
+ float4 edge1; edge1 = (*p2 - *p1);
+ float4 edge2; edge2 = ( *p3 - *p2 );
+ float4 edge3; edge3 = ( *p1 - *p3 );
+
+
+ float4 p1_to_p; p1_to_p = ( *p - *p1 );
+ float4 p2_to_p; p2_to_p = ( *p - *p2 );
+ float4 p3_to_p; p3_to_p = ( *p - *p3 );
+
+ float4 edge1_normal; edge1_normal = ( cross(edge1,*normal));
+ float4 edge2_normal; edge2_normal = ( cross(edge2,*normal));
+ float4 edge3_normal; edge3_normal = ( cross(edge3,*normal));
+
+
+
+ float r1, r2, r3;
+ r1 = dot(edge1_normal,p1_to_p );
+ r2 = dot(edge2_normal,p2_to_p );
+ r3 = dot(edge3_normal,p3_to_p );
+
+ if ( r1 > 0 && r2 > 0 && r3 > 0 )
+ return true;
+ if ( r1 <= 0 && r2 <= 0 && r3 <= 0 )
+ return true;
+ return false;
+
+}
+
+
+float segmentSqrDistance(float4 from, float4 to,float4 p, float4* nearest)
+{
+ float4 diff = p - from;
+ float4 v = to - from;
+ float t = dot(v,diff);
+
+ if (t > 0)
+ {
+ float dotVV = dot(v,v);
+ if (t < dotVV)
+ {
+ t /= dotVV;
+ diff -= t*v;
+ } else
+ {
+ t = 1;
+ diff -= v;
+ }
+ } else
+ {
+ t = 0;
+ }
+ *nearest = from + t*v;
+ return dot(diff,diff);
+}
+
+
+void computeContactSphereTriangle(int pairIndex,
+ int bodyIndexA, int bodyIndexB,
+ int collidableIndexA, int collidableIndexB,
+ __global const BodyData* rigidBodies,
+ __global const btCollidableGpu* collidables,
+ const float4* triangleVertices,
+ __global struct b3Contact4Data* restrict globalContactsOut,
+ counter32_t nGlobalContactsOut,
+ int maxContactCapacity,
+ float4 spherePos2,
+ float radius,
+ float4 pos,
+ float4 quat,
+ int faceIndex
+ )
+{
+
+ float4 invPos;
+ float4 invOrn;
+
+ trInverse(pos,quat, &invPos,&invOrn);
+ float4 spherePos = transform(&spherePos2,&invPos,&invOrn);
+ int numFaces = 3;
+ float4 closestPnt = (float4)(0, 0, 0, 0);
+ float4 hitNormalWorld = (float4)(0, 0, 0, 0);
+ float minDist = -1000000.f;
+ bool bCollide = false;
+
+
+ //////////////////////////////////////
+
+ float4 sphereCenter;
+ sphereCenter = spherePos;
+
+ const float4* vertices = triangleVertices;
+ float contactBreakingThreshold = 0.f;//todo?
+ float radiusWithThreshold = radius + contactBreakingThreshold;
+ float4 edge10;
+ edge10 = vertices[1]-vertices[0];
+ edge10.w = 0.f;//is this needed?
+ float4 edge20;
+ edge20 = vertices[2]-vertices[0];
+ edge20.w = 0.f;//is this needed?
+ float4 normal = cross3(edge10,edge20);
+ normal = normalize(normal);
+ float4 p1ToCenter;
+ p1ToCenter = sphereCenter - vertices[0];
+
+ float distanceFromPlane = dot(p1ToCenter,normal);
+
+ if (distanceFromPlane < 0.f)
+ {
+ //triangle facing the other way
+ distanceFromPlane *= -1.f;
+ normal *= -1.f;
+ }
+ hitNormalWorld = normal;
+
+ bool isInsideContactPlane = distanceFromPlane < radiusWithThreshold;
+
+ // Check for contact / intersection
+ bool hasContact = false;
+ float4 contactPoint;
+ if (isInsideContactPlane)
+ {
+
+ if (pointInTriangle(vertices,&normal, &sphereCenter))
+ {
+ // Inside the contact wedge - touches a point on the shell plane
+ hasContact = true;
+ contactPoint = sphereCenter - normal*distanceFromPlane;
+
+ } else {
+ // Could be inside one of the contact capsules
+ float contactCapsuleRadiusSqr = radiusWithThreshold*radiusWithThreshold;
+ float4 nearestOnEdge;
+ int numEdges = 3;
+ for (int i = 0; i < numEdges; i++)
+ {
+ float4 pa =vertices[i];
+ float4 pb = vertices[(i+1)%3];
+
+ float distanceSqr = segmentSqrDistance(pa,pb,sphereCenter, &nearestOnEdge);
+ if (distanceSqr < contactCapsuleRadiusSqr)
+ {
+ // Yep, we're inside a capsule
+ hasContact = true;
+ contactPoint = nearestOnEdge;
+
+ }
+
+ }
+ }
+ }
+
+ if (hasContact)
+ {
+
+ closestPnt = contactPoint;
+ float4 contactToCenter = sphereCenter - contactPoint;
+ minDist = length(contactToCenter);
+ if (minDist>FLT_EPSILON)
+ {
+ hitNormalWorld = normalize(contactToCenter);//*(1./minDist);
+ bCollide = true;
+ }
+
+ }
+
+
+ /////////////////////////////////////
+
+ if (bCollide && minDist > -10000)
+ {
+
+ float4 normalOnSurfaceB1 = qtRotate(quat,-hitNormalWorld);
+ float4 pOnB1 = transform(&closestPnt,&pos,&quat);
+ float actualDepth = minDist-radius;
+
+
+ if (actualDepth<=0.f)
+ {
+ pOnB1.w = actualDepth;
+ int dstIdx;
+
+
+ float lenSqr = dot3F4(normalOnSurfaceB1,normalOnSurfaceB1);
+ if (lenSqr>FLT_EPSILON)
+ {
+ AppendInc( nGlobalContactsOut, dstIdx );
+
+ if (dstIdx < maxContactCapacity)
+ {
+ __global struct b3Contact4Data* c = &globalContactsOut[dstIdx];
+ c->m_worldNormalOnB = -normalOnSurfaceB1;
+ c->m_restituitionCoeffCmp = (0.f*0xffff);c->m_frictionCoeffCmp = (0.7f*0xffff);
+ c->m_batchIdx = pairIndex;
+ c->m_bodyAPtrAndSignBit = rigidBodies[bodyIndexA].m_invMass==0?-bodyIndexA:bodyIndexA;
+ c->m_bodyBPtrAndSignBit = rigidBodies[bodyIndexB].m_invMass==0?-bodyIndexB:bodyIndexB;
+ c->m_worldPosB[0] = pOnB1;
+
+ c->m_childIndexA = -1;
+ c->m_childIndexB = faceIndex;
+
+ GET_NPOINTS(*c) = 1;
+ }
+ }
+
+ }
+ }//if (hasCollision)
+
+}
+
+
+
+// work-in-progress
+__kernel void findConcaveSphereContactsKernel( __global int4* concavePairs,
+ __global const BodyData* rigidBodies,
+ __global const btCollidableGpu* collidables,
+ __global const ConvexPolyhedronCL* convexShapes,
+ __global const float4* vertices,
+ __global const float4* uniqueEdges,
+ __global const btGpuFace* faces,
+ __global const int* indices,
+ __global btAabbCL* aabbs,
+ __global struct b3Contact4Data* restrict globalContactsOut,
+ counter32_t nGlobalContactsOut,
+ int numConcavePairs, int maxContactCapacity
+ )
+{
+
+ int i = get_global_id(0);
+ if (i>=numConcavePairs)
+ return;
+ int pairIdx = i;
+
+ int bodyIndexA = concavePairs[i].x;
+ int bodyIndexB = concavePairs[i].y;
+
+ int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;
+ int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;
+
+ int shapeIndexA = collidables[collidableIndexA].m_shapeIndex;
+ int shapeIndexB = collidables[collidableIndexB].m_shapeIndex;
+
+ if (collidables[collidableIndexB].m_shapeType==SHAPE_SPHERE)
+ {
+ int f = concavePairs[i].z;
+ btGpuFace face = faces[convexShapes[shapeIndexA].m_faceOffset+f];
+
+ float4 verticesA[3];
+ for (int i=0;i<3;i++)
+ {
+ int index = indices[face.m_indexOffset+i];
+ float4 vert = vertices[convexShapes[shapeIndexA].m_vertexOffset+index];
+ verticesA[i] = vert;
+ }
+
+ float4 spherePos = rigidBodies[bodyIndexB].m_pos;
+ float sphereRadius = collidables[collidableIndexB].m_radius;
+ float4 convexPos = rigidBodies[bodyIndexA].m_pos;
+ float4 convexOrn = rigidBodies[bodyIndexA].m_quat;
+
+ computeContactSphereTriangle(i, bodyIndexB, bodyIndexA, collidableIndexB, collidableIndexA,
+ rigidBodies,collidables,
+ verticesA,
+ globalContactsOut, nGlobalContactsOut,maxContactCapacity,
+ spherePos,sphereRadius,convexPos,convexOrn, f);
+
+ return;
+ }
+} \ No newline at end of file
diff --git a/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/primitiveContacts.h b/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/primitiveContacts.h
new file mode 100644
index 0000000000..b0103fe674
--- /dev/null
+++ b/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/primitiveContacts.h
@@ -0,0 +1,1289 @@
+//this file is autogenerated using stringify.bat (premake --stringify) in the build folder of this project
+static const char* primitiveContactsKernelsCL= \
+"#ifndef B3_CONTACT4DATA_H\n"
+"#define B3_CONTACT4DATA_H\n"
+"#ifndef B3_FLOAT4_H\n"
+"#define B3_FLOAT4_H\n"
+"#ifndef B3_PLATFORM_DEFINITIONS_H\n"
+"#define B3_PLATFORM_DEFINITIONS_H\n"
+"struct MyTest\n"
+"{\n"
+" int bla;\n"
+"};\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+"//keep B3_LARGE_FLOAT*B3_LARGE_FLOAT < FLT_MAX\n"
+"#define B3_LARGE_FLOAT 1e18f\n"
+"#define B3_INFINITY 1e18f\n"
+"#define b3Assert(a)\n"
+"#define b3ConstArray(a) __global const a*\n"
+"#define b3AtomicInc atomic_inc\n"
+"#define b3AtomicAdd atomic_add\n"
+"#define b3Fabs fabs\n"
+"#define b3Sqrt native_sqrt\n"
+"#define b3Sin native_sin\n"
+"#define b3Cos native_cos\n"
+"#define B3_STATIC\n"
+"#endif\n"
+"#endif\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+" typedef float4 b3Float4;\n"
+" #define b3Float4ConstArg const b3Float4\n"
+" #define b3MakeFloat4 (float4)\n"
+" float b3Dot3F4(b3Float4ConstArg v0,b3Float4ConstArg v1)\n"
+" {\n"
+" float4 a1 = b3MakeFloat4(v0.xyz,0.f);\n"
+" float4 b1 = b3MakeFloat4(v1.xyz,0.f);\n"
+" return dot(a1, b1);\n"
+" }\n"
+" b3Float4 b3Cross3(b3Float4ConstArg v0,b3Float4ConstArg v1)\n"
+" {\n"
+" float4 a1 = b3MakeFloat4(v0.xyz,0.f);\n"
+" float4 b1 = b3MakeFloat4(v1.xyz,0.f);\n"
+" return cross(a1, b1);\n"
+" }\n"
+" #define b3MinFloat4 min\n"
+" #define b3MaxFloat4 max\n"
+" #define b3Normalized(a) normalize(a)\n"
+"#endif \n"
+" \n"
+"inline bool b3IsAlmostZero(b3Float4ConstArg v)\n"
+"{\n"
+" if(b3Fabs(v.x)>1e-6 || b3Fabs(v.y)>1e-6 || b3Fabs(v.z)>1e-6) \n"
+" return false;\n"
+" return true;\n"
+"}\n"
+"inline int b3MaxDot( b3Float4ConstArg vec, __global const b3Float4* vecArray, int vecLen, float* dotOut )\n"
+"{\n"
+" float maxDot = -B3_INFINITY;\n"
+" int i = 0;\n"
+" int ptIndex = -1;\n"
+" for( i = 0; i < vecLen; i++ )\n"
+" {\n"
+" float dot = b3Dot3F4(vecArray[i],vec);\n"
+" \n"
+" if( dot > maxDot )\n"
+" {\n"
+" maxDot = dot;\n"
+" ptIndex = i;\n"
+" }\n"
+" }\n"
+" b3Assert(ptIndex>=0);\n"
+" if (ptIndex<0)\n"
+" {\n"
+" ptIndex = 0;\n"
+" }\n"
+" *dotOut = maxDot;\n"
+" return ptIndex;\n"
+"}\n"
+"#endif //B3_FLOAT4_H\n"
+"typedef struct b3Contact4Data b3Contact4Data_t;\n"
+"struct b3Contact4Data\n"
+"{\n"
+" b3Float4 m_worldPosB[4];\n"
+"// b3Float4 m_localPosA[4];\n"
+"// b3Float4 m_localPosB[4];\n"
+" b3Float4 m_worldNormalOnB; // w: m_nPoints\n"
+" unsigned short m_restituitionCoeffCmp;\n"
+" unsigned short m_frictionCoeffCmp;\n"
+" int m_batchIdx;\n"
+" int m_bodyAPtrAndSignBit;//x:m_bodyAPtr, y:m_bodyBPtr\n"
+" int m_bodyBPtrAndSignBit;\n"
+" int m_childIndexA;\n"
+" int m_childIndexB;\n"
+" int m_unused1;\n"
+" int m_unused2;\n"
+"};\n"
+"inline int b3Contact4Data_getNumPoints(const struct b3Contact4Data* contact)\n"
+"{\n"
+" return (int)contact->m_worldNormalOnB.w;\n"
+"};\n"
+"inline void b3Contact4Data_setNumPoints(struct b3Contact4Data* contact, int numPoints)\n"
+"{\n"
+" contact->m_worldNormalOnB.w = (float)numPoints;\n"
+"};\n"
+"#endif //B3_CONTACT4DATA_H\n"
+"#define SHAPE_CONVEX_HULL 3\n"
+"#define SHAPE_PLANE 4\n"
+"#define SHAPE_CONCAVE_TRIMESH 5\n"
+"#define SHAPE_COMPOUND_OF_CONVEX_HULLS 6\n"
+"#define SHAPE_SPHERE 7\n"
+"#pragma OPENCL EXTENSION cl_amd_printf : enable\n"
+"#pragma OPENCL EXTENSION cl_khr_local_int32_base_atomics : enable\n"
+"#pragma OPENCL EXTENSION cl_khr_global_int32_base_atomics : enable\n"
+"#pragma OPENCL EXTENSION cl_khr_local_int32_extended_atomics : enable\n"
+"#pragma OPENCL EXTENSION cl_khr_global_int32_extended_atomics : enable\n"
+"#ifdef cl_ext_atomic_counters_32\n"
+"#pragma OPENCL EXTENSION cl_ext_atomic_counters_32 : enable\n"
+"#else\n"
+"#define counter32_t volatile __global int*\n"
+"#endif\n"
+"#define GET_GROUP_IDX get_group_id(0)\n"
+"#define GET_LOCAL_IDX get_local_id(0)\n"
+"#define GET_GLOBAL_IDX get_global_id(0)\n"
+"#define GET_GROUP_SIZE get_local_size(0)\n"
+"#define GET_NUM_GROUPS get_num_groups(0)\n"
+"#define GROUP_LDS_BARRIER barrier(CLK_LOCAL_MEM_FENCE)\n"
+"#define GROUP_MEM_FENCE mem_fence(CLK_LOCAL_MEM_FENCE)\n"
+"#define AtomInc(x) atom_inc(&(x))\n"
+"#define AtomInc1(x, out) out = atom_inc(&(x))\n"
+"#define AppendInc(x, out) out = atomic_inc(x)\n"
+"#define AtomAdd(x, value) atom_add(&(x), value)\n"
+"#define AtomCmpxhg(x, cmp, value) atom_cmpxchg( &(x), cmp, value )\n"
+"#define AtomXhg(x, value) atom_xchg ( &(x), value )\n"
+"#define max2 max\n"
+"#define min2 min\n"
+"typedef unsigned int u32;\n"
+"typedef struct \n"
+"{\n"
+" union\n"
+" {\n"
+" float4 m_min;\n"
+" float m_minElems[4];\n"
+" int m_minIndices[4];\n"
+" };\n"
+" union\n"
+" {\n"
+" float4 m_max;\n"
+" float m_maxElems[4];\n"
+" int m_maxIndices[4];\n"
+" };\n"
+"} btAabbCL;\n"
+"///keep this in sync with btCollidable.h\n"
+"typedef struct\n"
+"{\n"
+" int m_numChildShapes;\n"
+" float m_radius;\n"
+" int m_shapeType;\n"
+" int m_shapeIndex;\n"
+" \n"
+"} btCollidableGpu;\n"
+"typedef struct\n"
+"{\n"
+" float4 m_childPosition;\n"
+" float4 m_childOrientation;\n"
+" int m_shapeIndex;\n"
+" int m_unused0;\n"
+" int m_unused1;\n"
+" int m_unused2;\n"
+"} btGpuChildShape;\n"
+"#define GET_NPOINTS(x) (x).m_worldNormalOnB.w\n"
+"typedef struct\n"
+"{\n"
+" float4 m_pos;\n"
+" float4 m_quat;\n"
+" float4 m_linVel;\n"
+" float4 m_angVel;\n"
+" u32 m_collidableIdx; \n"
+" float m_invMass;\n"
+" float m_restituitionCoeff;\n"
+" float m_frictionCoeff;\n"
+"} BodyData;\n"
+"typedef struct \n"
+"{\n"
+" float4 m_localCenter;\n"
+" float4 m_extents;\n"
+" float4 mC;\n"
+" float4 mE;\n"
+" \n"
+" float m_radius;\n"
+" int m_faceOffset;\n"
+" int m_numFaces;\n"
+" int m_numVertices;\n"
+" \n"
+" int m_vertexOffset;\n"
+" int m_uniqueEdgesOffset;\n"
+" int m_numUniqueEdges;\n"
+" int m_unused;\n"
+"} ConvexPolyhedronCL;\n"
+"typedef struct\n"
+"{\n"
+" float4 m_plane;\n"
+" int m_indexOffset;\n"
+" int m_numIndices;\n"
+"} btGpuFace;\n"
+"#define SELECT_UINT4( b, a, condition ) select( b,a,condition )\n"
+"#define make_float4 (float4)\n"
+"#define make_float2 (float2)\n"
+"#define make_uint4 (uint4)\n"
+"#define make_int4 (int4)\n"
+"#define make_uint2 (uint2)\n"
+"#define make_int2 (int2)\n"
+"__inline\n"
+"float fastDiv(float numerator, float denominator)\n"
+"{\n"
+" return native_divide(numerator, denominator); \n"
+"// return numerator/denominator; \n"
+"}\n"
+"__inline\n"
+"float4 fastDiv4(float4 numerator, float4 denominator)\n"
+"{\n"
+" return native_divide(numerator, denominator); \n"
+"}\n"
+"__inline\n"
+"float4 cross3(float4 a, float4 b)\n"
+"{\n"
+" return cross(a,b);\n"
+"}\n"
+"//#define dot3F4 dot\n"
+"__inline\n"
+"float dot3F4(float4 a, float4 b)\n"
+"{\n"
+" float4 a1 = make_float4(a.xyz,0.f);\n"
+" float4 b1 = make_float4(b.xyz,0.f);\n"
+" return dot(a1, b1);\n"
+"}\n"
+"__inline\n"
+"float4 fastNormalize4(float4 v)\n"
+"{\n"
+" return fast_normalize(v);\n"
+"}\n"
+"///////////////////////////////////////\n"
+"// Quaternion\n"
+"///////////////////////////////////////\n"
+"typedef float4 Quaternion;\n"
+"__inline\n"
+"Quaternion qtMul(Quaternion a, Quaternion b);\n"
+"__inline\n"
+"Quaternion qtNormalize(Quaternion in);\n"
+"__inline\n"
+"float4 qtRotate(Quaternion q, float4 vec);\n"
+"__inline\n"
+"Quaternion qtInvert(Quaternion q);\n"
+"__inline\n"
+"Quaternion qtMul(Quaternion a, Quaternion b)\n"
+"{\n"
+" Quaternion ans;\n"
+" ans = cross3( a, b );\n"
+" ans += a.w*b+b.w*a;\n"
+"// ans.w = a.w*b.w - (a.x*b.x+a.y*b.y+a.z*b.z);\n"
+" ans.w = a.w*b.w - dot3F4(a, b);\n"
+" return ans;\n"
+"}\n"
+"__inline\n"
+"Quaternion qtNormalize(Quaternion in)\n"
+"{\n"
+" return fastNormalize4(in);\n"
+"// in /= length( in );\n"
+"// return in;\n"
+"}\n"
+"__inline\n"
+"float4 qtRotate(Quaternion q, float4 vec)\n"
+"{\n"
+" Quaternion qInv = qtInvert( q );\n"
+" float4 vcpy = vec;\n"
+" vcpy.w = 0.f;\n"
+" float4 out = qtMul(qtMul(q,vcpy),qInv);\n"
+" return out;\n"
+"}\n"
+"__inline\n"
+"Quaternion qtInvert(Quaternion q)\n"
+"{\n"
+" return (Quaternion)(-q.xyz, q.w);\n"
+"}\n"
+"__inline\n"
+"float4 qtInvRotate(const Quaternion q, float4 vec)\n"
+"{\n"
+" return qtRotate( qtInvert( q ), vec );\n"
+"}\n"
+"__inline\n"
+"float4 transform(const float4* p, const float4* translation, const Quaternion* orientation)\n"
+"{\n"
+" return qtRotate( *orientation, *p ) + (*translation);\n"
+"}\n"
+"void trInverse(float4 translationIn, Quaternion orientationIn,\n"
+" float4* translationOut, Quaternion* orientationOut)\n"
+"{\n"
+" *orientationOut = qtInvert(orientationIn);\n"
+" *translationOut = qtRotate(*orientationOut, -translationIn);\n"
+"}\n"
+"void trMul(float4 translationA, Quaternion orientationA,\n"
+" float4 translationB, Quaternion orientationB,\n"
+" float4* translationOut, Quaternion* orientationOut)\n"
+"{\n"
+" *orientationOut = qtMul(orientationA,orientationB);\n"
+" *translationOut = transform(&translationB,&translationA,&orientationA);\n"
+"}\n"
+"__inline\n"
+"float4 normalize3(const float4 a)\n"
+"{\n"
+" float4 n = make_float4(a.x, a.y, a.z, 0.f);\n"
+" return fastNormalize4( n );\n"
+"}\n"
+"__inline float4 lerp3(const float4 a,const float4 b, float t)\n"
+"{\n"
+" return make_float4( a.x + (b.x - a.x) * t,\n"
+" a.y + (b.y - a.y) * t,\n"
+" a.z + (b.z - a.z) * t,\n"
+" 0.f);\n"
+"}\n"
+"float signedDistanceFromPointToPlane(float4 point, float4 planeEqn, float4* closestPointOnFace)\n"
+"{\n"
+" float4 n = (float4)(planeEqn.x, planeEqn.y, planeEqn.z, 0);\n"
+" float dist = dot3F4(n, point) + planeEqn.w;\n"
+" *closestPointOnFace = point - dist * n;\n"
+" return dist;\n"
+"}\n"
+"inline bool IsPointInPolygon(float4 p, \n"
+" const btGpuFace* face,\n"
+" __global const float4* baseVertex,\n"
+" __global const int* convexIndices,\n"
+" float4* out)\n"
+"{\n"
+" float4 a;\n"
+" float4 b;\n"
+" float4 ab;\n"
+" float4 ap;\n"
+" float4 v;\n"
+" float4 plane = make_float4(face->m_plane.x,face->m_plane.y,face->m_plane.z,0.f);\n"
+" \n"
+" if (face->m_numIndices<2)\n"
+" return false;\n"
+" \n"
+" float4 v0 = baseVertex[convexIndices[face->m_indexOffset + face->m_numIndices-1]];\n"
+" \n"
+" b = v0;\n"
+" for(unsigned i=0; i != face->m_numIndices; ++i)\n"
+" {\n"
+" a = b;\n"
+" float4 vi = baseVertex[convexIndices[face->m_indexOffset + i]];\n"
+" b = vi;\n"
+" ab = b-a;\n"
+" ap = p-a;\n"
+" v = cross3(ab,plane);\n"
+" if (dot(ap, v) > 0.f)\n"
+" {\n"
+" float ab_m2 = dot(ab, ab);\n"
+" float rt = ab_m2 != 0.f ? dot(ab, ap) / ab_m2 : 0.f;\n"
+" if (rt <= 0.f)\n"
+" {\n"
+" *out = a;\n"
+" }\n"
+" else if (rt >= 1.f) \n"
+" {\n"
+" *out = b;\n"
+" }\n"
+" else\n"
+" {\n"
+" float s = 1.f - rt;\n"
+" out[0].x = s * a.x + rt * b.x;\n"
+" out[0].y = s * a.y + rt * b.y;\n"
+" out[0].z = s * a.z + rt * b.z;\n"
+" }\n"
+" return false;\n"
+" }\n"
+" }\n"
+" return true;\n"
+"}\n"
+"void computeContactSphereConvex(int pairIndex,\n"
+" int bodyIndexA, int bodyIndexB, \n"
+" int collidableIndexA, int collidableIndexB, \n"
+" __global const BodyData* rigidBodies, \n"
+" __global const btCollidableGpu* collidables,\n"
+" __global const ConvexPolyhedronCL* convexShapes,\n"
+" __global const float4* convexVertices,\n"
+" __global const int* convexIndices,\n"
+" __global const btGpuFace* faces,\n"
+" __global struct b3Contact4Data* restrict globalContactsOut,\n"
+" counter32_t nGlobalContactsOut,\n"
+" int maxContactCapacity,\n"
+" float4 spherePos2,\n"
+" float radius,\n"
+" float4 pos,\n"
+" float4 quat\n"
+" )\n"
+"{\n"
+" float4 invPos;\n"
+" float4 invOrn;\n"
+" trInverse(pos,quat, &invPos,&invOrn);\n"
+" float4 spherePos = transform(&spherePos2,&invPos,&invOrn);\n"
+" int shapeIndex = collidables[collidableIndexB].m_shapeIndex;\n"
+" int numFaces = convexShapes[shapeIndex].m_numFaces;\n"
+" float4 closestPnt = (float4)(0, 0, 0, 0);\n"
+" float4 hitNormalWorld = (float4)(0, 0, 0, 0);\n"
+" float minDist = -1000000.f;\n"
+" bool bCollide = true;\n"
+" for ( int f = 0; f < numFaces; f++ )\n"
+" {\n"
+" btGpuFace face = faces[convexShapes[shapeIndex].m_faceOffset+f];\n"
+" // set up a plane equation \n"
+" float4 planeEqn;\n"
+" float4 n1 = face.m_plane;\n"
+" n1.w = 0.f;\n"
+" planeEqn = n1;\n"
+" planeEqn.w = face.m_plane.w;\n"
+" \n"
+" \n"
+" // compute a signed distance from the vertex in cloth to the face of rigidbody.\n"
+" float4 pntReturn;\n"
+" float dist = signedDistanceFromPointToPlane(spherePos, planeEqn, &pntReturn);\n"
+" // If the distance is positive, the plane is a separating plane. \n"
+" if ( dist > radius )\n"
+" {\n"
+" bCollide = false;\n"
+" break;\n"
+" }\n"
+" if (dist>0)\n"
+" {\n"
+" //might hit an edge or vertex\n"
+" float4 out;\n"
+" float4 zeroPos = make_float4(0,0,0,0);\n"
+" bool isInPoly = IsPointInPolygon(spherePos,\n"
+" &face,\n"
+" &convexVertices[convexShapes[shapeIndex].m_vertexOffset],\n"
+" convexIndices,\n"
+" &out);\n"
+" if (isInPoly)\n"
+" {\n"
+" if (dist>minDist)\n"
+" {\n"
+" minDist = dist;\n"
+" closestPnt = pntReturn;\n"
+" hitNormalWorld = planeEqn;\n"
+" \n"
+" }\n"
+" } else\n"
+" {\n"
+" float4 tmp = spherePos-out;\n"
+" float l2 = dot(tmp,tmp);\n"
+" if (l2<radius*radius)\n"
+" {\n"
+" dist = sqrt(l2);\n"
+" if (dist>minDist)\n"
+" {\n"
+" minDist = dist;\n"
+" closestPnt = out;\n"
+" hitNormalWorld = tmp/dist;\n"
+" \n"
+" }\n"
+" \n"
+" } else\n"
+" {\n"
+" bCollide = false;\n"
+" break;\n"
+" }\n"
+" }\n"
+" } else\n"
+" {\n"
+" if ( dist > minDist )\n"
+" {\n"
+" minDist = dist;\n"
+" closestPnt = pntReturn;\n"
+" hitNormalWorld.xyz = planeEqn.xyz;\n"
+" }\n"
+" }\n"
+" \n"
+" }\n"
+" \n"
+" if (bCollide && minDist > -10000)\n"
+" {\n"
+" float4 normalOnSurfaceB1 = qtRotate(quat,-hitNormalWorld);\n"
+" float4 pOnB1 = transform(&closestPnt,&pos,&quat);\n"
+" \n"
+" float actualDepth = minDist-radius;\n"
+" if (actualDepth<=0.f)\n"
+" {\n"
+" \n"
+" pOnB1.w = actualDepth;\n"
+" int dstIdx;\n"
+" AppendInc( nGlobalContactsOut, dstIdx );\n"
+" \n"
+" \n"
+" if (1)//dstIdx < maxContactCapacity)\n"
+" {\n"
+" __global struct b3Contact4Data* c = &globalContactsOut[dstIdx];\n"
+" c->m_worldNormalOnB = -normalOnSurfaceB1;\n"
+" c->m_restituitionCoeffCmp = (0.f*0xffff);c->m_frictionCoeffCmp = (0.7f*0xffff);\n"
+" c->m_batchIdx = pairIndex;\n"
+" c->m_bodyAPtrAndSignBit = rigidBodies[bodyIndexA].m_invMass==0?-bodyIndexA:bodyIndexA;\n"
+" c->m_bodyBPtrAndSignBit = rigidBodies[bodyIndexB].m_invMass==0?-bodyIndexB:bodyIndexB;\n"
+" c->m_worldPosB[0] = pOnB1;\n"
+" c->m_childIndexA = -1;\n"
+" c->m_childIndexB = -1;\n"
+" GET_NPOINTS(*c) = 1;\n"
+" } \n"
+" }\n"
+" }//if (hasCollision)\n"
+"}\n"
+" \n"
+"int extractManifoldSequential(const float4* p, int nPoints, float4 nearNormal, int4* contactIdx)\n"
+"{\n"
+" if( nPoints == 0 )\n"
+" return 0;\n"
+" \n"
+" if (nPoints <=4)\n"
+" return nPoints;\n"
+" \n"
+" \n"
+" if (nPoints >64)\n"
+" nPoints = 64;\n"
+" \n"
+" float4 center = make_float4(0.f);\n"
+" {\n"
+" \n"
+" for (int i=0;i<nPoints;i++)\n"
+" center += p[i];\n"
+" center /= (float)nPoints;\n"
+" }\n"
+" \n"
+" \n"
+" \n"
+" // sample 4 directions\n"
+" \n"
+" float4 aVector = p[0] - center;\n"
+" float4 u = cross3( nearNormal, aVector );\n"
+" float4 v = cross3( nearNormal, u );\n"
+" u = normalize3( u );\n"
+" v = normalize3( v );\n"
+" \n"
+" \n"
+" //keep point with deepest penetration\n"
+" float minW= FLT_MAX;\n"
+" \n"
+" int minIndex=-1;\n"
+" \n"
+" float4 maxDots;\n"
+" maxDots.x = FLT_MIN;\n"
+" maxDots.y = FLT_MIN;\n"
+" maxDots.z = FLT_MIN;\n"
+" maxDots.w = FLT_MIN;\n"
+" \n"
+" // idx, distance\n"
+" for(int ie = 0; ie<nPoints; ie++ )\n"
+" {\n"
+" if (p[ie].w<minW)\n"
+" {\n"
+" minW = p[ie].w;\n"
+" minIndex=ie;\n"
+" }\n"
+" float f;\n"
+" float4 r = p[ie]-center;\n"
+" f = dot3F4( u, r );\n"
+" if (f<maxDots.x)\n"
+" {\n"
+" maxDots.x = f;\n"
+" contactIdx[0].x = ie;\n"
+" }\n"
+" \n"
+" f = dot3F4( -u, r );\n"
+" if (f<maxDots.y)\n"
+" {\n"
+" maxDots.y = f;\n"
+" contactIdx[0].y = ie;\n"
+" }\n"
+" \n"
+" \n"
+" f = dot3F4( v, r );\n"
+" if (f<maxDots.z)\n"
+" {\n"
+" maxDots.z = f;\n"
+" contactIdx[0].z = ie;\n"
+" }\n"
+" \n"
+" f = dot3F4( -v, r );\n"
+" if (f<maxDots.w)\n"
+" {\n"
+" maxDots.w = f;\n"
+" contactIdx[0].w = ie;\n"
+" }\n"
+" \n"
+" }\n"
+" \n"
+" if (contactIdx[0].x != minIndex && contactIdx[0].y != minIndex && contactIdx[0].z != minIndex && contactIdx[0].w != minIndex)\n"
+" {\n"
+" //replace the first contact with minimum (todo: replace contact with least penetration)\n"
+" contactIdx[0].x = minIndex;\n"
+" }\n"
+" \n"
+" return 4;\n"
+" \n"
+"}\n"
+"#define MAX_PLANE_CONVEX_POINTS 64\n"
+"int computeContactPlaneConvex(int pairIndex,\n"
+" int bodyIndexA, int bodyIndexB, \n"
+" int collidableIndexA, int collidableIndexB, \n"
+" __global const BodyData* rigidBodies, \n"
+" __global const btCollidableGpu*collidables,\n"
+" __global const ConvexPolyhedronCL* convexShapes,\n"
+" __global const float4* convexVertices,\n"
+" __global const int* convexIndices,\n"
+" __global const btGpuFace* faces,\n"
+" __global struct b3Contact4Data* restrict globalContactsOut,\n"
+" counter32_t nGlobalContactsOut,\n"
+" int maxContactCapacity,\n"
+" float4 posB,\n"
+" Quaternion ornB\n"
+" )\n"
+"{\n"
+" int resultIndex=-1;\n"
+" int shapeIndex = collidables[collidableIndexB].m_shapeIndex;\n"
+" __global const ConvexPolyhedronCL* hullB = &convexShapes[shapeIndex];\n"
+" \n"
+" float4 posA;\n"
+" posA = rigidBodies[bodyIndexA].m_pos;\n"
+" Quaternion ornA;\n"
+" ornA = rigidBodies[bodyIndexA].m_quat;\n"
+" int numContactsOut = 0;\n"
+" int numWorldVertsB1= 0;\n"
+" float4 planeEq;\n"
+" planeEq = faces[collidables[collidableIndexA].m_shapeIndex].m_plane;\n"
+" float4 planeNormal = make_float4(planeEq.x,planeEq.y,planeEq.z,0.f);\n"
+" float4 planeNormalWorld;\n"
+" planeNormalWorld = qtRotate(ornA,planeNormal);\n"
+" float planeConstant = planeEq.w;\n"
+" \n"
+" float4 invPosA;Quaternion invOrnA;\n"
+" float4 convexInPlaneTransPos1; Quaternion convexInPlaneTransOrn1;\n"
+" {\n"
+" \n"
+" trInverse(posA,ornA,&invPosA,&invOrnA);\n"
+" trMul(invPosA,invOrnA,posB,ornB,&convexInPlaneTransPos1,&convexInPlaneTransOrn1);\n"
+" }\n"
+" float4 invPosB;Quaternion invOrnB;\n"
+" float4 planeInConvexPos1; Quaternion planeInConvexOrn1;\n"
+" {\n"
+" \n"
+" trInverse(posB,ornB,&invPosB,&invOrnB);\n"
+" trMul(invPosB,invOrnB,posA,ornA,&planeInConvexPos1,&planeInConvexOrn1); \n"
+" }\n"
+" \n"
+" float4 planeNormalInConvex = qtRotate(planeInConvexOrn1,-planeNormal);\n"
+" float maxDot = -1e30;\n"
+" int hitVertex=-1;\n"
+" float4 hitVtx;\n"
+" float4 contactPoints[MAX_PLANE_CONVEX_POINTS];\n"
+" int numPoints = 0;\n"
+" int4 contactIdx;\n"
+" contactIdx=make_int4(0,1,2,3);\n"
+" \n"
+" \n"
+" for (int i=0;i<hullB->m_numVertices;i++)\n"
+" {\n"
+" float4 vtx = convexVertices[hullB->m_vertexOffset+i];\n"
+" float curDot = dot(vtx,planeNormalInConvex);\n"
+" if (curDot>maxDot)\n"
+" {\n"
+" hitVertex=i;\n"
+" maxDot=curDot;\n"
+" hitVtx = vtx;\n"
+" //make sure the deepest points is always included\n"
+" if (numPoints==MAX_PLANE_CONVEX_POINTS)\n"
+" numPoints--;\n"
+" }\n"
+" if (numPoints<MAX_PLANE_CONVEX_POINTS)\n"
+" {\n"
+" float4 vtxWorld = transform(&vtx, &posB, &ornB);\n"
+" float4 vtxInPlane = transform(&vtxWorld, &invPosA, &invOrnA);//oplaneTransform.inverse()*vtxWorld;\n"
+" float dist = dot(planeNormal,vtxInPlane)-planeConstant;\n"
+" if (dist<0.f)\n"
+" {\n"
+" vtxWorld.w = dist;\n"
+" contactPoints[numPoints] = vtxWorld;\n"
+" numPoints++;\n"
+" }\n"
+" }\n"
+" }\n"
+" int numReducedPoints = numPoints;\n"
+" if (numPoints>4)\n"
+" {\n"
+" numReducedPoints = extractManifoldSequential( contactPoints, numPoints, planeNormalInConvex, &contactIdx);\n"
+" }\n"
+" if (numReducedPoints>0)\n"
+" {\n"
+" int dstIdx;\n"
+" AppendInc( nGlobalContactsOut, dstIdx );\n"
+" if (dstIdx < maxContactCapacity)\n"
+" {\n"
+" resultIndex = dstIdx;\n"
+" __global struct b3Contact4Data* c = &globalContactsOut[dstIdx];\n"
+" c->m_worldNormalOnB = -planeNormalWorld;\n"
+" //c->setFrictionCoeff(0.7);\n"
+" //c->setRestituitionCoeff(0.f);\n"
+" c->m_restituitionCoeffCmp = (0.f*0xffff);c->m_frictionCoeffCmp = (0.7f*0xffff);\n"
+" c->m_batchIdx = pairIndex;\n"
+" c->m_bodyAPtrAndSignBit = rigidBodies[bodyIndexA].m_invMass==0?-bodyIndexA:bodyIndexA;\n"
+" c->m_bodyBPtrAndSignBit = rigidBodies[bodyIndexB].m_invMass==0?-bodyIndexB:bodyIndexB;\n"
+" c->m_childIndexA = -1;\n"
+" c->m_childIndexB = -1;\n"
+" switch (numReducedPoints)\n"
+" {\n"
+" case 4:\n"
+" c->m_worldPosB[3] = contactPoints[contactIdx.w];\n"
+" case 3:\n"
+" c->m_worldPosB[2] = contactPoints[contactIdx.z];\n"
+" case 2:\n"
+" c->m_worldPosB[1] = contactPoints[contactIdx.y];\n"
+" case 1:\n"
+" c->m_worldPosB[0] = contactPoints[contactIdx.x];\n"
+" default:\n"
+" {\n"
+" }\n"
+" };\n"
+" \n"
+" GET_NPOINTS(*c) = numReducedPoints;\n"
+" }//if (dstIdx < numPairs)\n"
+" } \n"
+" return resultIndex;\n"
+"}\n"
+"void computeContactPlaneSphere(int pairIndex,\n"
+" int bodyIndexA, int bodyIndexB, \n"
+" int collidableIndexA, int collidableIndexB, \n"
+" __global const BodyData* rigidBodies, \n"
+" __global const btCollidableGpu* collidables,\n"
+" __global const btGpuFace* faces,\n"
+" __global struct b3Contact4Data* restrict globalContactsOut,\n"
+" counter32_t nGlobalContactsOut,\n"
+" int maxContactCapacity)\n"
+"{\n"
+" float4 planeEq = faces[collidables[collidableIndexA].m_shapeIndex].m_plane;\n"
+" float radius = collidables[collidableIndexB].m_radius;\n"
+" float4 posA1 = rigidBodies[bodyIndexA].m_pos;\n"
+" float4 ornA1 = rigidBodies[bodyIndexA].m_quat;\n"
+" float4 posB1 = rigidBodies[bodyIndexB].m_pos;\n"
+" float4 ornB1 = rigidBodies[bodyIndexB].m_quat;\n"
+" \n"
+" bool hasCollision = false;\n"
+" float4 planeNormal1 = make_float4(planeEq.x,planeEq.y,planeEq.z,0.f);\n"
+" float planeConstant = planeEq.w;\n"
+" float4 convexInPlaneTransPos1; Quaternion convexInPlaneTransOrn1;\n"
+" {\n"
+" float4 invPosA;Quaternion invOrnA;\n"
+" trInverse(posA1,ornA1,&invPosA,&invOrnA);\n"
+" trMul(invPosA,invOrnA,posB1,ornB1,&convexInPlaneTransPos1,&convexInPlaneTransOrn1);\n"
+" }\n"
+" float4 planeInConvexPos1; Quaternion planeInConvexOrn1;\n"
+" {\n"
+" float4 invPosB;Quaternion invOrnB;\n"
+" trInverse(posB1,ornB1,&invPosB,&invOrnB);\n"
+" trMul(invPosB,invOrnB,posA1,ornA1,&planeInConvexPos1,&planeInConvexOrn1); \n"
+" }\n"
+" float4 vtx1 = qtRotate(planeInConvexOrn1,-planeNormal1)*radius;\n"
+" float4 vtxInPlane1 = transform(&vtx1,&convexInPlaneTransPos1,&convexInPlaneTransOrn1);\n"
+" float distance = dot3F4(planeNormal1,vtxInPlane1) - planeConstant;\n"
+" hasCollision = distance < 0.f;//m_manifoldPtr->getContactBreakingThreshold();\n"
+" if (hasCollision)\n"
+" {\n"
+" float4 vtxInPlaneProjected1 = vtxInPlane1 - distance*planeNormal1;\n"
+" float4 vtxInPlaneWorld1 = transform(&vtxInPlaneProjected1,&posA1,&ornA1);\n"
+" float4 normalOnSurfaceB1 = qtRotate(ornA1,planeNormal1);\n"
+" float4 pOnB1 = vtxInPlaneWorld1+normalOnSurfaceB1*distance;\n"
+" pOnB1.w = distance;\n"
+" int dstIdx;\n"
+" AppendInc( nGlobalContactsOut, dstIdx );\n"
+" \n"
+" if (dstIdx < maxContactCapacity)\n"
+" {\n"
+" __global struct b3Contact4Data* c = &globalContactsOut[dstIdx];\n"
+" c->m_worldNormalOnB = -normalOnSurfaceB1;\n"
+" c->m_restituitionCoeffCmp = (0.f*0xffff);c->m_frictionCoeffCmp = (0.7f*0xffff);\n"
+" c->m_batchIdx = pairIndex;\n"
+" c->m_bodyAPtrAndSignBit = rigidBodies[bodyIndexA].m_invMass==0?-bodyIndexA:bodyIndexA;\n"
+" c->m_bodyBPtrAndSignBit = rigidBodies[bodyIndexB].m_invMass==0?-bodyIndexB:bodyIndexB;\n"
+" c->m_worldPosB[0] = pOnB1;\n"
+" c->m_childIndexA = -1;\n"
+" c->m_childIndexB = -1;\n"
+" GET_NPOINTS(*c) = 1;\n"
+" }//if (dstIdx < numPairs)\n"
+" }//if (hasCollision)\n"
+"}\n"
+"__kernel void primitiveContactsKernel( __global int4* pairs, \n"
+" __global const BodyData* rigidBodies, \n"
+" __global const btCollidableGpu* collidables,\n"
+" __global const ConvexPolyhedronCL* convexShapes, \n"
+" __global const float4* vertices,\n"
+" __global const float4* uniqueEdges,\n"
+" __global const btGpuFace* faces,\n"
+" __global const int* indices,\n"
+" __global struct b3Contact4Data* restrict globalContactsOut,\n"
+" counter32_t nGlobalContactsOut,\n"
+" int numPairs, int maxContactCapacity)\n"
+"{\n"
+" int i = get_global_id(0);\n"
+" int pairIndex = i;\n"
+" \n"
+" float4 worldVertsB1[64];\n"
+" float4 worldVertsB2[64];\n"
+" int capacityWorldVerts = 64; \n"
+" float4 localContactsOut[64];\n"
+" int localContactCapacity=64;\n"
+" \n"
+" float minDist = -1e30f;\n"
+" float maxDist = 0.02f;\n"
+" if (i<numPairs)\n"
+" {\n"
+" int bodyIndexA = pairs[i].x;\n"
+" int bodyIndexB = pairs[i].y;\n"
+" \n"
+" int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;\n"
+" int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;\n"
+" \n"
+" if (collidables[collidableIndexA].m_shapeType == SHAPE_PLANE &&\n"
+" collidables[collidableIndexB].m_shapeType == SHAPE_CONVEX_HULL)\n"
+" {\n"
+" float4 posB;\n"
+" posB = rigidBodies[bodyIndexB].m_pos;\n"
+" Quaternion ornB;\n"
+" ornB = rigidBodies[bodyIndexB].m_quat;\n"
+" int contactIndex = computeContactPlaneConvex(pairIndex, bodyIndexA, bodyIndexB, collidableIndexA, collidableIndexB, \n"
+" rigidBodies,collidables,convexShapes,vertices,indices,\n"
+" faces, globalContactsOut, nGlobalContactsOut,maxContactCapacity, posB,ornB);\n"
+" if (contactIndex>=0)\n"
+" pairs[pairIndex].z = contactIndex;\n"
+" return;\n"
+" }\n"
+" if (collidables[collidableIndexA].m_shapeType == SHAPE_CONVEX_HULL &&\n"
+" collidables[collidableIndexB].m_shapeType == SHAPE_PLANE)\n"
+" {\n"
+" float4 posA;\n"
+" posA = rigidBodies[bodyIndexA].m_pos;\n"
+" Quaternion ornA;\n"
+" ornA = rigidBodies[bodyIndexA].m_quat;\n"
+" int contactIndex = computeContactPlaneConvex( pairIndex, bodyIndexB,bodyIndexA, collidableIndexB,collidableIndexA, \n"
+" rigidBodies,collidables,convexShapes,vertices,indices,\n"
+" faces, globalContactsOut, nGlobalContactsOut,maxContactCapacity,posA,ornA);\n"
+" if (contactIndex>=0)\n"
+" pairs[pairIndex].z = contactIndex;\n"
+" return;\n"
+" }\n"
+" if (collidables[collidableIndexA].m_shapeType == SHAPE_PLANE &&\n"
+" collidables[collidableIndexB].m_shapeType == SHAPE_SPHERE)\n"
+" {\n"
+" computeContactPlaneSphere(pairIndex, bodyIndexA, bodyIndexB, collidableIndexA, collidableIndexB, \n"
+" rigidBodies,collidables,faces, globalContactsOut, nGlobalContactsOut,maxContactCapacity);\n"
+" return;\n"
+" }\n"
+" if (collidables[collidableIndexA].m_shapeType == SHAPE_SPHERE &&\n"
+" collidables[collidableIndexB].m_shapeType == SHAPE_PLANE)\n"
+" {\n"
+" computeContactPlaneSphere( pairIndex, bodyIndexB,bodyIndexA, collidableIndexB,collidableIndexA, \n"
+" rigidBodies,collidables,\n"
+" faces, globalContactsOut, nGlobalContactsOut,maxContactCapacity);\n"
+" return;\n"
+" }\n"
+" \n"
+" \n"
+" if (collidables[collidableIndexA].m_shapeType == SHAPE_SPHERE &&\n"
+" collidables[collidableIndexB].m_shapeType == SHAPE_CONVEX_HULL)\n"
+" {\n"
+" \n"
+" float4 spherePos = rigidBodies[bodyIndexA].m_pos;\n"
+" float sphereRadius = collidables[collidableIndexA].m_radius;\n"
+" float4 convexPos = rigidBodies[bodyIndexB].m_pos;\n"
+" float4 convexOrn = rigidBodies[bodyIndexB].m_quat;\n"
+" computeContactSphereConvex(pairIndex, bodyIndexA, bodyIndexB, collidableIndexA, collidableIndexB, \n"
+" rigidBodies,collidables,convexShapes,vertices,indices,faces, globalContactsOut, nGlobalContactsOut,maxContactCapacity,\n"
+" spherePos,sphereRadius,convexPos,convexOrn);\n"
+" return;\n"
+" }\n"
+" if (collidables[collidableIndexA].m_shapeType == SHAPE_CONVEX_HULL &&\n"
+" collidables[collidableIndexB].m_shapeType == SHAPE_SPHERE)\n"
+" {\n"
+" \n"
+" float4 spherePos = rigidBodies[bodyIndexB].m_pos;\n"
+" float sphereRadius = collidables[collidableIndexB].m_radius;\n"
+" float4 convexPos = rigidBodies[bodyIndexA].m_pos;\n"
+" float4 convexOrn = rigidBodies[bodyIndexA].m_quat;\n"
+" computeContactSphereConvex(pairIndex, bodyIndexB, bodyIndexA, collidableIndexB, collidableIndexA, \n"
+" rigidBodies,collidables,convexShapes,vertices,indices,faces, globalContactsOut, nGlobalContactsOut,maxContactCapacity,\n"
+" spherePos,sphereRadius,convexPos,convexOrn);\n"
+" return;\n"
+" }\n"
+" \n"
+" \n"
+" \n"
+" \n"
+" \n"
+" \n"
+" if (collidables[collidableIndexA].m_shapeType == SHAPE_SPHERE &&\n"
+" collidables[collidableIndexB].m_shapeType == SHAPE_SPHERE)\n"
+" {\n"
+" //sphere-sphere\n"
+" float radiusA = collidables[collidableIndexA].m_radius;\n"
+" float radiusB = collidables[collidableIndexB].m_radius;\n"
+" float4 posA = rigidBodies[bodyIndexA].m_pos;\n"
+" float4 posB = rigidBodies[bodyIndexB].m_pos;\n"
+" float4 diff = posA-posB;\n"
+" float len = length(diff);\n"
+" \n"
+" ///iff distance positive, don't generate a new contact\n"
+" if ( len <= (radiusA+radiusB))\n"
+" {\n"
+" ///distance (negative means penetration)\n"
+" float dist = len - (radiusA+radiusB);\n"
+" float4 normalOnSurfaceB = make_float4(1.f,0.f,0.f,0.f);\n"
+" if (len > 0.00001)\n"
+" {\n"
+" normalOnSurfaceB = diff / len;\n"
+" }\n"
+" float4 contactPosB = posB + normalOnSurfaceB*radiusB;\n"
+" contactPosB.w = dist;\n"
+" \n"
+" int dstIdx;\n"
+" AppendInc( nGlobalContactsOut, dstIdx );\n"
+" \n"
+" if (dstIdx < maxContactCapacity)\n"
+" {\n"
+" __global struct b3Contact4Data* c = &globalContactsOut[dstIdx];\n"
+" c->m_worldNormalOnB = normalOnSurfaceB;\n"
+" c->m_restituitionCoeffCmp = (0.f*0xffff);c->m_frictionCoeffCmp = (0.7f*0xffff);\n"
+" c->m_batchIdx = pairIndex;\n"
+" int bodyA = pairs[pairIndex].x;\n"
+" int bodyB = pairs[pairIndex].y;\n"
+" c->m_bodyAPtrAndSignBit = rigidBodies[bodyA].m_invMass==0?-bodyA:bodyA;\n"
+" c->m_bodyBPtrAndSignBit = rigidBodies[bodyB].m_invMass==0?-bodyB:bodyB;\n"
+" c->m_worldPosB[0] = contactPosB;\n"
+" c->m_childIndexA = -1;\n"
+" c->m_childIndexB = -1;\n"
+" GET_NPOINTS(*c) = 1;\n"
+" }//if (dstIdx < numPairs)\n"
+" }//if ( len <= (radiusA+radiusB))\n"
+" return;\n"
+" }//SHAPE_SPHERE SHAPE_SPHERE\n"
+" }// if (i<numPairs)\n"
+"}\n"
+"// work-in-progress\n"
+"__kernel void processCompoundPairsPrimitivesKernel( __global const int4* gpuCompoundPairs,\n"
+" __global const BodyData* rigidBodies, \n"
+" __global const btCollidableGpu* collidables,\n"
+" __global const ConvexPolyhedronCL* convexShapes, \n"
+" __global const float4* vertices,\n"
+" __global const float4* uniqueEdges,\n"
+" __global const btGpuFace* faces,\n"
+" __global const int* indices,\n"
+" __global btAabbCL* aabbs,\n"
+" __global const btGpuChildShape* gpuChildShapes,\n"
+" __global struct b3Contact4Data* restrict globalContactsOut,\n"
+" counter32_t nGlobalContactsOut,\n"
+" int numCompoundPairs, int maxContactCapacity\n"
+" )\n"
+"{\n"
+" int i = get_global_id(0);\n"
+" if (i<numCompoundPairs)\n"
+" {\n"
+" int bodyIndexA = gpuCompoundPairs[i].x;\n"
+" int bodyIndexB = gpuCompoundPairs[i].y;\n"
+" int childShapeIndexA = gpuCompoundPairs[i].z;\n"
+" int childShapeIndexB = gpuCompoundPairs[i].w;\n"
+" \n"
+" int collidableIndexA = -1;\n"
+" int collidableIndexB = -1;\n"
+" \n"
+" float4 ornA = rigidBodies[bodyIndexA].m_quat;\n"
+" float4 posA = rigidBodies[bodyIndexA].m_pos;\n"
+" \n"
+" float4 ornB = rigidBodies[bodyIndexB].m_quat;\n"
+" float4 posB = rigidBodies[bodyIndexB].m_pos;\n"
+" \n"
+" if (childShapeIndexA >= 0)\n"
+" {\n"
+" collidableIndexA = gpuChildShapes[childShapeIndexA].m_shapeIndex;\n"
+" float4 childPosA = gpuChildShapes[childShapeIndexA].m_childPosition;\n"
+" float4 childOrnA = gpuChildShapes[childShapeIndexA].m_childOrientation;\n"
+" float4 newPosA = qtRotate(ornA,childPosA)+posA;\n"
+" float4 newOrnA = qtMul(ornA,childOrnA);\n"
+" posA = newPosA;\n"
+" ornA = newOrnA;\n"
+" } else\n"
+" {\n"
+" collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;\n"
+" }\n"
+" \n"
+" if (childShapeIndexB>=0)\n"
+" {\n"
+" collidableIndexB = gpuChildShapes[childShapeIndexB].m_shapeIndex;\n"
+" float4 childPosB = gpuChildShapes[childShapeIndexB].m_childPosition;\n"
+" float4 childOrnB = gpuChildShapes[childShapeIndexB].m_childOrientation;\n"
+" float4 newPosB = transform(&childPosB,&posB,&ornB);\n"
+" float4 newOrnB = qtMul(ornB,childOrnB);\n"
+" posB = newPosB;\n"
+" ornB = newOrnB;\n"
+" } else\n"
+" {\n"
+" collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx; \n"
+" }\n"
+" \n"
+" int shapeIndexA = collidables[collidableIndexA].m_shapeIndex;\n"
+" int shapeIndexB = collidables[collidableIndexB].m_shapeIndex;\n"
+" \n"
+" int shapeTypeA = collidables[collidableIndexA].m_shapeType;\n"
+" int shapeTypeB = collidables[collidableIndexB].m_shapeType;\n"
+" int pairIndex = i;\n"
+" if ((shapeTypeA == SHAPE_PLANE) && (shapeTypeB==SHAPE_CONVEX_HULL))\n"
+" {\n"
+" computeContactPlaneConvex( pairIndex, bodyIndexA,bodyIndexB, collidableIndexA,collidableIndexB, \n"
+" rigidBodies,collidables,convexShapes,vertices,indices,\n"
+" faces, globalContactsOut, nGlobalContactsOut,maxContactCapacity,posB,ornB);\n"
+" return;\n"
+" }\n"
+" if ((shapeTypeA == SHAPE_CONVEX_HULL) && (shapeTypeB==SHAPE_PLANE))\n"
+" {\n"
+" computeContactPlaneConvex( pairIndex, bodyIndexB,bodyIndexA, collidableIndexB,collidableIndexA, \n"
+" rigidBodies,collidables,convexShapes,vertices,indices,\n"
+" faces, globalContactsOut, nGlobalContactsOut,maxContactCapacity,posA,ornA);\n"
+" return;\n"
+" }\n"
+" if ((shapeTypeA == SHAPE_CONVEX_HULL) && (shapeTypeB == SHAPE_SPHERE))\n"
+" {\n"
+" float4 spherePos = rigidBodies[bodyIndexB].m_pos;\n"
+" float sphereRadius = collidables[collidableIndexB].m_radius;\n"
+" float4 convexPos = posA;\n"
+" float4 convexOrn = ornA;\n"
+" \n"
+" computeContactSphereConvex(pairIndex, bodyIndexB, bodyIndexA , collidableIndexB,collidableIndexA, \n"
+" rigidBodies,collidables,convexShapes,vertices,indices,faces, globalContactsOut, nGlobalContactsOut,maxContactCapacity,\n"
+" spherePos,sphereRadius,convexPos,convexOrn);\n"
+" \n"
+" return;\n"
+" }\n"
+" if ((shapeTypeA == SHAPE_SPHERE) && (shapeTypeB == SHAPE_CONVEX_HULL))\n"
+" {\n"
+" float4 spherePos = rigidBodies[bodyIndexA].m_pos;\n"
+" float sphereRadius = collidables[collidableIndexA].m_radius;\n"
+" float4 convexPos = posB;\n"
+" float4 convexOrn = ornB;\n"
+" \n"
+" computeContactSphereConvex(pairIndex, bodyIndexA, bodyIndexB, collidableIndexA, collidableIndexB, \n"
+" rigidBodies,collidables,convexShapes,vertices,indices,faces, globalContactsOut, nGlobalContactsOut,maxContactCapacity,\n"
+" spherePos,sphereRadius,convexPos,convexOrn);\n"
+" \n"
+" return;\n"
+" }\n"
+" }// if (i<numCompoundPairs)\n"
+"}\n"
+"bool pointInTriangle(const float4* vertices, const float4* normal, float4 *p )\n"
+"{\n"
+" const float4* p1 = &vertices[0];\n"
+" const float4* p2 = &vertices[1];\n"
+" const float4* p3 = &vertices[2];\n"
+" float4 edge1; edge1 = (*p2 - *p1);\n"
+" float4 edge2; edge2 = ( *p3 - *p2 );\n"
+" float4 edge3; edge3 = ( *p1 - *p3 );\n"
+" \n"
+" float4 p1_to_p; p1_to_p = ( *p - *p1 );\n"
+" float4 p2_to_p; p2_to_p = ( *p - *p2 );\n"
+" float4 p3_to_p; p3_to_p = ( *p - *p3 );\n"
+" float4 edge1_normal; edge1_normal = ( cross(edge1,*normal));\n"
+" float4 edge2_normal; edge2_normal = ( cross(edge2,*normal));\n"
+" float4 edge3_normal; edge3_normal = ( cross(edge3,*normal));\n"
+" \n"
+" \n"
+" float r1, r2, r3;\n"
+" r1 = dot(edge1_normal,p1_to_p );\n"
+" r2 = dot(edge2_normal,p2_to_p );\n"
+" r3 = dot(edge3_normal,p3_to_p );\n"
+" \n"
+" if ( r1 > 0 && r2 > 0 && r3 > 0 )\n"
+" return true;\n"
+" if ( r1 <= 0 && r2 <= 0 && r3 <= 0 ) \n"
+" return true;\n"
+" return false;\n"
+"}\n"
+"float segmentSqrDistance(float4 from, float4 to,float4 p, float4* nearest) \n"
+"{\n"
+" float4 diff = p - from;\n"
+" float4 v = to - from;\n"
+" float t = dot(v,diff);\n"
+" \n"
+" if (t > 0) \n"
+" {\n"
+" float dotVV = dot(v,v);\n"
+" if (t < dotVV) \n"
+" {\n"
+" t /= dotVV;\n"
+" diff -= t*v;\n"
+" } else \n"
+" {\n"
+" t = 1;\n"
+" diff -= v;\n"
+" }\n"
+" } else\n"
+" {\n"
+" t = 0;\n"
+" }\n"
+" *nearest = from + t*v;\n"
+" return dot(diff,diff); \n"
+"}\n"
+"void computeContactSphereTriangle(int pairIndex,\n"
+" int bodyIndexA, int bodyIndexB,\n"
+" int collidableIndexA, int collidableIndexB, \n"
+" __global const BodyData* rigidBodies, \n"
+" __global const btCollidableGpu* collidables,\n"
+" const float4* triangleVertices,\n"
+" __global struct b3Contact4Data* restrict globalContactsOut,\n"
+" counter32_t nGlobalContactsOut,\n"
+" int maxContactCapacity,\n"
+" float4 spherePos2,\n"
+" float radius,\n"
+" float4 pos,\n"
+" float4 quat,\n"
+" int faceIndex\n"
+" )\n"
+"{\n"
+" float4 invPos;\n"
+" float4 invOrn;\n"
+" trInverse(pos,quat, &invPos,&invOrn);\n"
+" float4 spherePos = transform(&spherePos2,&invPos,&invOrn);\n"
+" int numFaces = 3;\n"
+" float4 closestPnt = (float4)(0, 0, 0, 0);\n"
+" float4 hitNormalWorld = (float4)(0, 0, 0, 0);\n"
+" float minDist = -1000000.f;\n"
+" bool bCollide = false;\n"
+" \n"
+" //////////////////////////////////////\n"
+" float4 sphereCenter;\n"
+" sphereCenter = spherePos;\n"
+" const float4* vertices = triangleVertices;\n"
+" float contactBreakingThreshold = 0.f;//todo?\n"
+" float radiusWithThreshold = radius + contactBreakingThreshold;\n"
+" float4 edge10;\n"
+" edge10 = vertices[1]-vertices[0];\n"
+" edge10.w = 0.f;//is this needed?\n"
+" float4 edge20;\n"
+" edge20 = vertices[2]-vertices[0];\n"
+" edge20.w = 0.f;//is this needed?\n"
+" float4 normal = cross3(edge10,edge20);\n"
+" normal = normalize(normal);\n"
+" float4 p1ToCenter;\n"
+" p1ToCenter = sphereCenter - vertices[0];\n"
+" \n"
+" float distanceFromPlane = dot(p1ToCenter,normal);\n"
+" if (distanceFromPlane < 0.f)\n"
+" {\n"
+" //triangle facing the other way\n"
+" distanceFromPlane *= -1.f;\n"
+" normal *= -1.f;\n"
+" }\n"
+" hitNormalWorld = normal;\n"
+" bool isInsideContactPlane = distanceFromPlane < radiusWithThreshold;\n"
+" \n"
+" // Check for contact / intersection\n"
+" bool hasContact = false;\n"
+" float4 contactPoint;\n"
+" if (isInsideContactPlane) \n"
+" {\n"
+" \n"
+" if (pointInTriangle(vertices,&normal, &sphereCenter)) \n"
+" {\n"
+" // Inside the contact wedge - touches a point on the shell plane\n"
+" hasContact = true;\n"
+" contactPoint = sphereCenter - normal*distanceFromPlane;\n"
+" \n"
+" } else {\n"
+" // Could be inside one of the contact capsules\n"
+" float contactCapsuleRadiusSqr = radiusWithThreshold*radiusWithThreshold;\n"
+" float4 nearestOnEdge;\n"
+" int numEdges = 3;\n"
+" for (int i = 0; i < numEdges; i++) \n"
+" {\n"
+" float4 pa =vertices[i];\n"
+" float4 pb = vertices[(i+1)%3];\n"
+" float distanceSqr = segmentSqrDistance(pa,pb,sphereCenter, &nearestOnEdge);\n"
+" if (distanceSqr < contactCapsuleRadiusSqr) \n"
+" {\n"
+" // Yep, we're inside a capsule\n"
+" hasContact = true;\n"
+" contactPoint = nearestOnEdge;\n"
+" \n"
+" }\n"
+" \n"
+" }\n"
+" }\n"
+" }\n"
+" if (hasContact) \n"
+" {\n"
+" closestPnt = contactPoint;\n"
+" float4 contactToCenter = sphereCenter - contactPoint;\n"
+" minDist = length(contactToCenter);\n"
+" if (minDist>FLT_EPSILON)\n"
+" {\n"
+" hitNormalWorld = normalize(contactToCenter);//*(1./minDist);\n"
+" bCollide = true;\n"
+" }\n"
+" \n"
+" }\n"
+" /////////////////////////////////////\n"
+" if (bCollide && minDist > -10000)\n"
+" {\n"
+" \n"
+" float4 normalOnSurfaceB1 = qtRotate(quat,-hitNormalWorld);\n"
+" float4 pOnB1 = transform(&closestPnt,&pos,&quat);\n"
+" float actualDepth = minDist-radius;\n"
+" \n"
+" if (actualDepth<=0.f)\n"
+" {\n"
+" pOnB1.w = actualDepth;\n"
+" int dstIdx;\n"
+" \n"
+" float lenSqr = dot3F4(normalOnSurfaceB1,normalOnSurfaceB1);\n"
+" if (lenSqr>FLT_EPSILON)\n"
+" {\n"
+" AppendInc( nGlobalContactsOut, dstIdx );\n"
+" \n"
+" if (dstIdx < maxContactCapacity)\n"
+" {\n"
+" __global struct b3Contact4Data* c = &globalContactsOut[dstIdx];\n"
+" c->m_worldNormalOnB = -normalOnSurfaceB1;\n"
+" c->m_restituitionCoeffCmp = (0.f*0xffff);c->m_frictionCoeffCmp = (0.7f*0xffff);\n"
+" c->m_batchIdx = pairIndex;\n"
+" c->m_bodyAPtrAndSignBit = rigidBodies[bodyIndexA].m_invMass==0?-bodyIndexA:bodyIndexA;\n"
+" c->m_bodyBPtrAndSignBit = rigidBodies[bodyIndexB].m_invMass==0?-bodyIndexB:bodyIndexB;\n"
+" c->m_worldPosB[0] = pOnB1;\n"
+" c->m_childIndexA = -1;\n"
+" c->m_childIndexB = faceIndex;\n"
+" GET_NPOINTS(*c) = 1;\n"
+" } \n"
+" }\n"
+" }\n"
+" }//if (hasCollision)\n"
+"}\n"
+"// work-in-progress\n"
+"__kernel void findConcaveSphereContactsKernel( __global int4* concavePairs,\n"
+" __global const BodyData* rigidBodies,\n"
+" __global const btCollidableGpu* collidables,\n"
+" __global const ConvexPolyhedronCL* convexShapes, \n"
+" __global const float4* vertices,\n"
+" __global const float4* uniqueEdges,\n"
+" __global const btGpuFace* faces,\n"
+" __global const int* indices,\n"
+" __global btAabbCL* aabbs,\n"
+" __global struct b3Contact4Data* restrict globalContactsOut,\n"
+" counter32_t nGlobalContactsOut,\n"
+" int numConcavePairs, int maxContactCapacity\n"
+" )\n"
+"{\n"
+" int i = get_global_id(0);\n"
+" if (i>=numConcavePairs)\n"
+" return;\n"
+" int pairIdx = i;\n"
+" int bodyIndexA = concavePairs[i].x;\n"
+" int bodyIndexB = concavePairs[i].y;\n"
+" int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;\n"
+" int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;\n"
+" int shapeIndexA = collidables[collidableIndexA].m_shapeIndex;\n"
+" int shapeIndexB = collidables[collidableIndexB].m_shapeIndex;\n"
+" if (collidables[collidableIndexB].m_shapeType==SHAPE_SPHERE)\n"
+" {\n"
+" int f = concavePairs[i].z;\n"
+" btGpuFace face = faces[convexShapes[shapeIndexA].m_faceOffset+f];\n"
+" \n"
+" float4 verticesA[3];\n"
+" for (int i=0;i<3;i++)\n"
+" {\n"
+" int index = indices[face.m_indexOffset+i];\n"
+" float4 vert = vertices[convexShapes[shapeIndexA].m_vertexOffset+index];\n"
+" verticesA[i] = vert;\n"
+" }\n"
+" float4 spherePos = rigidBodies[bodyIndexB].m_pos;\n"
+" float sphereRadius = collidables[collidableIndexB].m_radius;\n"
+" float4 convexPos = rigidBodies[bodyIndexA].m_pos;\n"
+" float4 convexOrn = rigidBodies[bodyIndexA].m_quat;\n"
+" computeContactSphereTriangle(i, bodyIndexB, bodyIndexA, collidableIndexB, collidableIndexA, \n"
+" rigidBodies,collidables,\n"
+" verticesA,\n"
+" globalContactsOut, nGlobalContactsOut,maxContactCapacity,\n"
+" spherePos,sphereRadius,convexPos,convexOrn, f);\n"
+" return;\n"
+" }\n"
+"}\n"
+;
diff --git a/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/sat.cl b/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/sat.cl
new file mode 100644
index 0000000000..a6565fd6fa
--- /dev/null
+++ b/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/sat.cl
@@ -0,0 +1,2018 @@
+//keep this enum in sync with the CPU version (in btCollidable.h)
+//written by Erwin Coumans
+
+
+#define SHAPE_CONVEX_HULL 3
+#define SHAPE_CONCAVE_TRIMESH 5
+#define TRIANGLE_NUM_CONVEX_FACES 5
+#define SHAPE_COMPOUND_OF_CONVEX_HULLS 6
+
+#define B3_MAX_STACK_DEPTH 256
+
+
+typedef unsigned int u32;
+
+///keep this in sync with btCollidable.h
+typedef struct
+{
+ union {
+ int m_numChildShapes;
+ int m_bvhIndex;
+ };
+ union
+ {
+ float m_radius;
+ int m_compoundBvhIndex;
+ };
+
+ int m_shapeType;
+ int m_shapeIndex;
+
+} btCollidableGpu;
+
+#define MAX_NUM_PARTS_IN_BITS 10
+
+///b3QuantizedBvhNode 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).
+typedef struct
+{
+ //12 bytes
+ unsigned short int m_quantizedAabbMin[3];
+ unsigned short int m_quantizedAabbMax[3];
+ //4 bytes
+ int m_escapeIndexOrTriangleIndex;
+} b3QuantizedBvhNode;
+
+typedef struct
+{
+ float4 m_aabbMin;
+ float4 m_aabbMax;
+ float4 m_quantization;
+ int m_numNodes;
+ int m_numSubTrees;
+ int m_nodeOffset;
+ int m_subTreeOffset;
+
+} b3BvhInfo;
+
+
+int getTriangleIndex(const b3QuantizedBvhNode* rootNode)
+{
+ 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 (rootNode->m_escapeIndexOrTriangleIndex&~(y));
+}
+
+int getTriangleIndexGlobal(__global const b3QuantizedBvhNode* rootNode)
+{
+ 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 (rootNode->m_escapeIndexOrTriangleIndex&~(y));
+}
+
+int isLeafNode(const b3QuantizedBvhNode* rootNode)
+{
+ //skipindex is negative (internal node), triangleindex >=0 (leafnode)
+ return (rootNode->m_escapeIndexOrTriangleIndex >= 0)? 1 : 0;
+}
+
+int isLeafNodeGlobal(__global const b3QuantizedBvhNode* rootNode)
+{
+ //skipindex is negative (internal node), triangleindex >=0 (leafnode)
+ return (rootNode->m_escapeIndexOrTriangleIndex >= 0)? 1 : 0;
+}
+
+int getEscapeIndex(const b3QuantizedBvhNode* rootNode)
+{
+ return -rootNode->m_escapeIndexOrTriangleIndex;
+}
+
+int getEscapeIndexGlobal(__global const b3QuantizedBvhNode* rootNode)
+{
+ return -rootNode->m_escapeIndexOrTriangleIndex;
+}
+
+
+typedef struct
+{
+ //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];
+} b3BvhSubtreeInfo;
+
+
+
+
+
+
+
+typedef struct
+{
+ float4 m_childPosition;
+ float4 m_childOrientation;
+ int m_shapeIndex;
+ int m_unused0;
+ int m_unused1;
+ int m_unused2;
+} btGpuChildShape;
+
+
+typedef struct
+{
+ float4 m_pos;
+ float4 m_quat;
+ float4 m_linVel;
+ float4 m_angVel;
+
+ u32 m_collidableIdx;
+ float m_invMass;
+ float m_restituitionCoeff;
+ float m_frictionCoeff;
+} BodyData;
+
+
+typedef struct
+{
+ float4 m_localCenter;
+ float4 m_extents;
+ float4 mC;
+ float4 mE;
+
+ float m_radius;
+ int m_faceOffset;
+ int m_numFaces;
+ int m_numVertices;
+
+ int m_vertexOffset;
+ int m_uniqueEdgesOffset;
+ int m_numUniqueEdges;
+ int m_unused;
+} ConvexPolyhedronCL;
+
+typedef struct
+{
+ union
+ {
+ float4 m_min;
+ float m_minElems[4];
+ int m_minIndices[4];
+ };
+ union
+ {
+ float4 m_max;
+ float m_maxElems[4];
+ int m_maxIndices[4];
+ };
+} btAabbCL;
+
+#include "Bullet3Collision/BroadPhaseCollision/shared/b3Aabb.h"
+#include "Bullet3Common/shared/b3Int2.h"
+
+
+
+typedef struct
+{
+ float4 m_plane;
+ int m_indexOffset;
+ int m_numIndices;
+} btGpuFace;
+
+#define make_float4 (float4)
+
+
+__inline
+float4 cross3(float4 a, float4 b)
+{
+ return cross(a,b);
+
+
+// float4 a1 = make_float4(a.xyz,0.f);
+// float4 b1 = make_float4(b.xyz,0.f);
+
+// return cross(a1,b1);
+
+//float4 c = make_float4(a.y*b.z - a.z*b.y,a.z*b.x - a.x*b.z,a.x*b.y - a.y*b.x,0.f);
+
+ // float4 c = make_float4(a.y*b.z - a.z*b.y,1.f,a.x*b.y - a.y*b.x,0.f);
+
+ //return c;
+}
+
+__inline
+float dot3F4(float4 a, float4 b)
+{
+ float4 a1 = make_float4(a.xyz,0.f);
+ float4 b1 = make_float4(b.xyz,0.f);
+ return dot(a1, b1);
+}
+
+__inline
+float4 fastNormalize4(float4 v)
+{
+ v = make_float4(v.xyz,0.f);
+ return fast_normalize(v);
+}
+
+
+///////////////////////////////////////
+// Quaternion
+///////////////////////////////////////
+
+typedef float4 Quaternion;
+
+__inline
+Quaternion qtMul(Quaternion a, Quaternion b);
+
+__inline
+Quaternion qtNormalize(Quaternion in);
+
+__inline
+float4 qtRotate(Quaternion q, float4 vec);
+
+__inline
+Quaternion qtInvert(Quaternion q);
+
+
+
+
+__inline
+Quaternion qtMul(Quaternion a, Quaternion b)
+{
+ Quaternion ans;
+ ans = cross3( a, b );
+ ans += a.w*b+b.w*a;
+// ans.w = a.w*b.w - (a.x*b.x+a.y*b.y+a.z*b.z);
+ ans.w = a.w*b.w - dot3F4(a, b);
+ return ans;
+}
+
+__inline
+Quaternion qtNormalize(Quaternion in)
+{
+ return fastNormalize4(in);
+// in /= length( in );
+// return in;
+}
+__inline
+float4 qtRotate(Quaternion q, float4 vec)
+{
+ Quaternion qInv = qtInvert( q );
+ float4 vcpy = vec;
+ vcpy.w = 0.f;
+ float4 out = qtMul(qtMul(q,vcpy),qInv);
+ return out;
+}
+
+__inline
+Quaternion qtInvert(Quaternion q)
+{
+ return (Quaternion)(-q.xyz, q.w);
+}
+
+__inline
+float4 qtInvRotate(const Quaternion q, float4 vec)
+{
+ return qtRotate( qtInvert( q ), vec );
+}
+
+__inline
+float4 transform(const float4* p, const float4* translation, const Quaternion* orientation)
+{
+ return qtRotate( *orientation, *p ) + (*translation);
+}
+
+
+
+__inline
+float4 normalize3(const float4 a)
+{
+ float4 n = make_float4(a.x, a.y, a.z, 0.f);
+ return fastNormalize4( n );
+}
+
+inline void projectLocal(const ConvexPolyhedronCL* hull, const float4 pos, const float4 orn,
+const float4* dir, const float4* vertices, float* min, float* max)
+{
+ min[0] = FLT_MAX;
+ max[0] = -FLT_MAX;
+ int numVerts = hull->m_numVertices;
+
+ const float4 localDir = qtInvRotate(orn,*dir);
+ float offset = dot(pos,*dir);
+ for(int i=0;i<numVerts;i++)
+ {
+ float dp = dot(vertices[hull->m_vertexOffset+i],localDir);
+ if(dp < min[0])
+ min[0] = dp;
+ if(dp > max[0])
+ max[0] = dp;
+ }
+ if(min[0]>max[0])
+ {
+ float tmp = min[0];
+ min[0] = max[0];
+ max[0] = tmp;
+ }
+ min[0] += offset;
+ max[0] += offset;
+}
+
+inline void project(__global const ConvexPolyhedronCL* hull, const float4 pos, const float4 orn,
+const float4* dir, __global const float4* vertices, float* min, float* max)
+{
+ min[0] = FLT_MAX;
+ max[0] = -FLT_MAX;
+ int numVerts = hull->m_numVertices;
+
+ const float4 localDir = qtInvRotate(orn,*dir);
+ float offset = dot(pos,*dir);
+ for(int i=0;i<numVerts;i++)
+ {
+ float dp = dot(vertices[hull->m_vertexOffset+i],localDir);
+ if(dp < min[0])
+ min[0] = dp;
+ if(dp > max[0])
+ max[0] = dp;
+ }
+ if(min[0]>max[0])
+ {
+ float tmp = min[0];
+ min[0] = max[0];
+ max[0] = tmp;
+ }
+ min[0] += offset;
+ max[0] += offset;
+}
+
+inline bool TestSepAxisLocalA(const ConvexPolyhedronCL* hullA, __global const ConvexPolyhedronCL* hullB,
+ const float4 posA,const float4 ornA,
+ const float4 posB,const float4 ornB,
+ float4* sep_axis, const float4* verticesA, __global const float4* verticesB,float* depth)
+{
+ float Min0,Max0;
+ float Min1,Max1;
+ projectLocal(hullA,posA,ornA,sep_axis,verticesA, &Min0, &Max0);
+ project(hullB,posB,ornB, sep_axis,verticesB, &Min1, &Max1);
+
+ if(Max0<Min1 || Max1<Min0)
+ return false;
+
+ float d0 = Max0 - Min1;
+ float d1 = Max1 - Min0;
+ *depth = d0<d1 ? d0:d1;
+ return true;
+}
+
+
+
+
+inline bool IsAlmostZero(const float4 v)
+{
+ if(fabs(v.x)>1e-6f || fabs(v.y)>1e-6f || fabs(v.z)>1e-6f)
+ return false;
+ return true;
+}
+
+
+
+bool findSeparatingAxisLocalA( const ConvexPolyhedronCL* hullA, __global const ConvexPolyhedronCL* hullB,
+ const float4 posA1,
+ const float4 ornA,
+ const float4 posB1,
+ const float4 ornB,
+ const float4 DeltaC2,
+
+ const float4* verticesA,
+ const float4* uniqueEdgesA,
+ const btGpuFace* facesA,
+ const int* indicesA,
+
+ __global const float4* verticesB,
+ __global const float4* uniqueEdgesB,
+ __global const btGpuFace* facesB,
+ __global const int* indicesB,
+ float4* sep,
+ float* dmin)
+{
+
+
+ float4 posA = posA1;
+ posA.w = 0.f;
+ float4 posB = posB1;
+ posB.w = 0.f;
+ int curPlaneTests=0;
+ {
+ int numFacesA = hullA->m_numFaces;
+ // Test normals from hullA
+ for(int i=0;i<numFacesA;i++)
+ {
+ const float4 normal = facesA[hullA->m_faceOffset+i].m_plane;
+ float4 faceANormalWS = qtRotate(ornA,normal);
+ if (dot3F4(DeltaC2,faceANormalWS)<0)
+ faceANormalWS*=-1.f;
+ curPlaneTests++;
+ float d;
+ if(!TestSepAxisLocalA( hullA, hullB, posA,ornA,posB,ornB,&faceANormalWS, verticesA, verticesB,&d))
+ return false;
+ if(d<*dmin)
+ {
+ *dmin = d;
+ *sep = faceANormalWS;
+ }
+ }
+ }
+ if((dot3F4(-DeltaC2,*sep))>0.0f)
+ {
+ *sep = -(*sep);
+ }
+ return true;
+}
+
+bool findSeparatingAxisLocalB( __global const ConvexPolyhedronCL* hullA, const ConvexPolyhedronCL* hullB,
+ const float4 posA1,
+ const float4 ornA,
+ const float4 posB1,
+ const float4 ornB,
+ const float4 DeltaC2,
+ __global const float4* verticesA,
+ __global const float4* uniqueEdgesA,
+ __global const btGpuFace* facesA,
+ __global const int* indicesA,
+ const float4* verticesB,
+ const float4* uniqueEdgesB,
+ const btGpuFace* facesB,
+ const int* indicesB,
+ float4* sep,
+ float* dmin)
+{
+
+
+ float4 posA = posA1;
+ posA.w = 0.f;
+ float4 posB = posB1;
+ posB.w = 0.f;
+ int curPlaneTests=0;
+ {
+ int numFacesA = hullA->m_numFaces;
+ // Test normals from hullA
+ for(int i=0;i<numFacesA;i++)
+ {
+ const float4 normal = facesA[hullA->m_faceOffset+i].m_plane;
+ float4 faceANormalWS = qtRotate(ornA,normal);
+ if (dot3F4(DeltaC2,faceANormalWS)<0)
+ faceANormalWS *= -1.f;
+ curPlaneTests++;
+ float d;
+ if(!TestSepAxisLocalA( hullB, hullA, posB,ornB,posA,ornA, &faceANormalWS, verticesB,verticesA, &d))
+ return false;
+ if(d<*dmin)
+ {
+ *dmin = d;
+ *sep = faceANormalWS;
+ }
+ }
+ }
+ if((dot3F4(-DeltaC2,*sep))>0.0f)
+ {
+ *sep = -(*sep);
+ }
+ return true;
+}
+
+
+
+bool findSeparatingAxisEdgeEdgeLocalA( const ConvexPolyhedronCL* hullA, __global const ConvexPolyhedronCL* hullB,
+ const float4 posA1,
+ const float4 ornA,
+ const float4 posB1,
+ const float4 ornB,
+ const float4 DeltaC2,
+ const float4* verticesA,
+ const float4* uniqueEdgesA,
+ const btGpuFace* facesA,
+ const int* indicesA,
+ __global const float4* verticesB,
+ __global const float4* uniqueEdgesB,
+ __global const btGpuFace* facesB,
+ __global const int* indicesB,
+ float4* sep,
+ float* dmin)
+{
+
+
+ float4 posA = posA1;
+ posA.w = 0.f;
+ float4 posB = posB1;
+ posB.w = 0.f;
+
+ int curPlaneTests=0;
+
+ int curEdgeEdge = 0;
+ // Test edges
+ for(int e0=0;e0<hullA->m_numUniqueEdges;e0++)
+ {
+ const float4 edge0 = uniqueEdgesA[hullA->m_uniqueEdgesOffset+e0];
+ float4 edge0World = qtRotate(ornA,edge0);
+
+ for(int e1=0;e1<hullB->m_numUniqueEdges;e1++)
+ {
+ const float4 edge1 = uniqueEdgesB[hullB->m_uniqueEdgesOffset+e1];
+ float4 edge1World = qtRotate(ornB,edge1);
+
+
+ float4 crossje = cross3(edge0World,edge1World);
+
+ curEdgeEdge++;
+ if(!IsAlmostZero(crossje))
+ {
+ crossje = normalize3(crossje);
+ if (dot3F4(DeltaC2,crossje)<0)
+ crossje *= -1.f;
+
+ float dist;
+ bool result = true;
+ {
+ float Min0,Max0;
+ float Min1,Max1;
+ projectLocal(hullA,posA,ornA,&crossje,verticesA, &Min0, &Max0);
+ project(hullB,posB,ornB,&crossje,verticesB, &Min1, &Max1);
+
+ if(Max0<Min1 || Max1<Min0)
+ result = false;
+
+ float d0 = Max0 - Min1;
+ float d1 = Max1 - Min0;
+ dist = d0<d1 ? d0:d1;
+ result = true;
+
+ }
+
+
+ if(dist<*dmin)
+ {
+ *dmin = dist;
+ *sep = crossje;
+ }
+ }
+ }
+
+ }
+
+
+ if((dot3F4(-DeltaC2,*sep))>0.0f)
+ {
+ *sep = -(*sep);
+ }
+ return true;
+}
+
+
+inline bool TestSepAxis(__global const ConvexPolyhedronCL* hullA, __global const ConvexPolyhedronCL* hullB,
+ const float4 posA,const float4 ornA,
+ const float4 posB,const float4 ornB,
+ float4* sep_axis, __global const float4* vertices,float* depth)
+{
+ float Min0,Max0;
+ float Min1,Max1;
+ project(hullA,posA,ornA,sep_axis,vertices, &Min0, &Max0);
+ project(hullB,posB,ornB, sep_axis,vertices, &Min1, &Max1);
+
+ if(Max0<Min1 || Max1<Min0)
+ return false;
+
+ float d0 = Max0 - Min1;
+ float d1 = Max1 - Min0;
+ *depth = d0<d1 ? d0:d1;
+ return true;
+}
+
+
+bool findSeparatingAxis( __global const ConvexPolyhedronCL* hullA, __global const ConvexPolyhedronCL* hullB,
+ const float4 posA1,
+ const float4 ornA,
+ const float4 posB1,
+ const float4 ornB,
+ const float4 DeltaC2,
+ __global const float4* vertices,
+ __global const float4* uniqueEdges,
+ __global const btGpuFace* faces,
+ __global const int* indices,
+ float4* sep,
+ float* dmin)
+{
+
+
+ float4 posA = posA1;
+ posA.w = 0.f;
+ float4 posB = posB1;
+ posB.w = 0.f;
+
+ int curPlaneTests=0;
+
+ {
+ int numFacesA = hullA->m_numFaces;
+ // Test normals from hullA
+ for(int i=0;i<numFacesA;i++)
+ {
+ const float4 normal = faces[hullA->m_faceOffset+i].m_plane;
+ float4 faceANormalWS = qtRotate(ornA,normal);
+
+ if (dot3F4(DeltaC2,faceANormalWS)<0)
+ faceANormalWS*=-1.f;
+
+ curPlaneTests++;
+
+ float d;
+ if(!TestSepAxis( hullA, hullB, posA,ornA,posB,ornB,&faceANormalWS, vertices,&d))
+ return false;
+
+ if(d<*dmin)
+ {
+ *dmin = d;
+ *sep = faceANormalWS;
+ }
+ }
+ }
+
+
+ if((dot3F4(-DeltaC2,*sep))>0.0f)
+ {
+ *sep = -(*sep);
+ }
+
+ return true;
+}
+
+
+
+
+bool findSeparatingAxisUnitSphere( __global const ConvexPolyhedronCL* hullA, __global const ConvexPolyhedronCL* hullB,
+ const float4 posA1,
+ const float4 ornA,
+ const float4 posB1,
+ const float4 ornB,
+ const float4 DeltaC2,
+ __global const float4* vertices,
+ __global const float4* unitSphereDirections,
+ int numUnitSphereDirections,
+ float4* sep,
+ float* dmin)
+{
+
+ float4 posA = posA1;
+ posA.w = 0.f;
+ float4 posB = posB1;
+ posB.w = 0.f;
+
+ int curPlaneTests=0;
+
+ int curEdgeEdge = 0;
+ // Test unit sphere directions
+ for (int i=0;i<numUnitSphereDirections;i++)
+ {
+
+ float4 crossje;
+ crossje = unitSphereDirections[i];
+
+ if (dot3F4(DeltaC2,crossje)>0)
+ crossje *= -1.f;
+ {
+ float dist;
+ bool result = true;
+ float Min0,Max0;
+ float Min1,Max1;
+ project(hullA,posA,ornA,&crossje,vertices, &Min0, &Max0);
+ project(hullB,posB,ornB,&crossje,vertices, &Min1, &Max1);
+
+ if(Max0<Min1 || Max1<Min0)
+ return false;
+
+ float d0 = Max0 - Min1;
+ float d1 = Max1 - Min0;
+ dist = d0<d1 ? d0:d1;
+ result = true;
+
+ if(dist<*dmin)
+ {
+ *dmin = dist;
+ *sep = crossje;
+ }
+ }
+ }
+
+
+ if((dot3F4(-DeltaC2,*sep))>0.0f)
+ {
+ *sep = -(*sep);
+ }
+ return true;
+}
+
+
+bool findSeparatingAxisEdgeEdge( __global const ConvexPolyhedronCL* hullA, __global const ConvexPolyhedronCL* hullB,
+ const float4 posA1,
+ const float4 ornA,
+ const float4 posB1,
+ const float4 ornB,
+ const float4 DeltaC2,
+ __global const float4* vertices,
+ __global const float4* uniqueEdges,
+ __global const btGpuFace* faces,
+ __global const int* indices,
+ float4* sep,
+ float* dmin)
+{
+
+
+ float4 posA = posA1;
+ posA.w = 0.f;
+ float4 posB = posB1;
+ posB.w = 0.f;
+
+ int curPlaneTests=0;
+
+ int curEdgeEdge = 0;
+ // Test edges
+ for(int e0=0;e0<hullA->m_numUniqueEdges;e0++)
+ {
+ const float4 edge0 = uniqueEdges[hullA->m_uniqueEdgesOffset+e0];
+ float4 edge0World = qtRotate(ornA,edge0);
+
+ for(int e1=0;e1<hullB->m_numUniqueEdges;e1++)
+ {
+ const float4 edge1 = uniqueEdges[hullB->m_uniqueEdgesOffset+e1];
+ float4 edge1World = qtRotate(ornB,edge1);
+
+
+ float4 crossje = cross3(edge0World,edge1World);
+
+ curEdgeEdge++;
+ if(!IsAlmostZero(crossje))
+ {
+ crossje = normalize3(crossje);
+ if (dot3F4(DeltaC2,crossje)<0)
+ crossje*=-1.f;
+
+ float dist;
+ bool result = true;
+ {
+ float Min0,Max0;
+ float Min1,Max1;
+ project(hullA,posA,ornA,&crossje,vertices, &Min0, &Max0);
+ project(hullB,posB,ornB,&crossje,vertices, &Min1, &Max1);
+
+ if(Max0<Min1 || Max1<Min0)
+ return false;
+
+ float d0 = Max0 - Min1;
+ float d1 = Max1 - Min0;
+ dist = d0<d1 ? d0:d1;
+ result = true;
+
+ }
+
+
+ if(dist<*dmin)
+ {
+ *dmin = dist;
+ *sep = crossje;
+ }
+ }
+ }
+
+ }
+
+
+ if((dot3F4(-DeltaC2,*sep))>0.0f)
+ {
+ *sep = -(*sep);
+ }
+ return true;
+}
+
+
+// work-in-progress
+__kernel void processCompoundPairsKernel( __global const int4* gpuCompoundPairs,
+ __global const BodyData* rigidBodies,
+ __global const btCollidableGpu* collidables,
+ __global const ConvexPolyhedronCL* convexShapes,
+ __global const float4* vertices,
+ __global const float4* uniqueEdges,
+ __global const btGpuFace* faces,
+ __global const int* indices,
+ __global btAabbCL* aabbs,
+ __global const btGpuChildShape* gpuChildShapes,
+ __global volatile float4* gpuCompoundSepNormalsOut,
+ __global volatile int* gpuHasCompoundSepNormalsOut,
+ int numCompoundPairs
+ )
+{
+
+ int i = get_global_id(0);
+ if (i<numCompoundPairs)
+ {
+ int bodyIndexA = gpuCompoundPairs[i].x;
+ int bodyIndexB = gpuCompoundPairs[i].y;
+
+ int childShapeIndexA = gpuCompoundPairs[i].z;
+ int childShapeIndexB = gpuCompoundPairs[i].w;
+
+ int collidableIndexA = -1;
+ int collidableIndexB = -1;
+
+ float4 ornA = rigidBodies[bodyIndexA].m_quat;
+ float4 posA = rigidBodies[bodyIndexA].m_pos;
+
+ float4 ornB = rigidBodies[bodyIndexB].m_quat;
+ float4 posB = rigidBodies[bodyIndexB].m_pos;
+
+ if (childShapeIndexA >= 0)
+ {
+ collidableIndexA = gpuChildShapes[childShapeIndexA].m_shapeIndex;
+ float4 childPosA = gpuChildShapes[childShapeIndexA].m_childPosition;
+ float4 childOrnA = gpuChildShapes[childShapeIndexA].m_childOrientation;
+ float4 newPosA = qtRotate(ornA,childPosA)+posA;
+ float4 newOrnA = qtMul(ornA,childOrnA);
+ posA = newPosA;
+ ornA = newOrnA;
+ } else
+ {
+ collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;
+ }
+
+ if (childShapeIndexB>=0)
+ {
+ collidableIndexB = gpuChildShapes[childShapeIndexB].m_shapeIndex;
+ float4 childPosB = gpuChildShapes[childShapeIndexB].m_childPosition;
+ float4 childOrnB = gpuChildShapes[childShapeIndexB].m_childOrientation;
+ float4 newPosB = transform(&childPosB,&posB,&ornB);
+ float4 newOrnB = qtMul(ornB,childOrnB);
+ posB = newPosB;
+ ornB = newOrnB;
+ } else
+ {
+ collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;
+ }
+
+ gpuHasCompoundSepNormalsOut[i] = 0;
+
+ int shapeIndexA = collidables[collidableIndexA].m_shapeIndex;
+ int shapeIndexB = collidables[collidableIndexB].m_shapeIndex;
+
+ int shapeTypeA = collidables[collidableIndexA].m_shapeType;
+ int shapeTypeB = collidables[collidableIndexB].m_shapeType;
+
+
+ if ((shapeTypeA != SHAPE_CONVEX_HULL) || (shapeTypeB != SHAPE_CONVEX_HULL))
+ {
+ return;
+ }
+
+ int hasSeparatingAxis = 5;
+
+ int numFacesA = convexShapes[shapeIndexA].m_numFaces;
+ float dmin = FLT_MAX;
+ posA.w = 0.f;
+ posB.w = 0.f;
+ float4 c0local = convexShapes[shapeIndexA].m_localCenter;
+ float4 c0 = transform(&c0local, &posA, &ornA);
+ float4 c1local = convexShapes[shapeIndexB].m_localCenter;
+ float4 c1 = transform(&c1local,&posB,&ornB);
+ const float4 DeltaC2 = c0 - c1;
+ float4 sepNormal = make_float4(1,0,0,0);
+ bool sepA = findSeparatingAxis( &convexShapes[shapeIndexA], &convexShapes[shapeIndexB],posA,ornA,posB,ornB,DeltaC2,vertices,uniqueEdges,faces,indices,&sepNormal,&dmin);
+ hasSeparatingAxis = 4;
+ if (!sepA)
+ {
+ hasSeparatingAxis = 0;
+ } else
+ {
+ bool sepB = findSeparatingAxis( &convexShapes[shapeIndexB],&convexShapes[shapeIndexA],posB,ornB,posA,ornA,DeltaC2,vertices,uniqueEdges,faces,indices,&sepNormal,&dmin);
+
+ if (!sepB)
+ {
+ hasSeparatingAxis = 0;
+ } else//(!sepB)
+ {
+ bool sepEE = findSeparatingAxisEdgeEdge( &convexShapes[shapeIndexA], &convexShapes[shapeIndexB],posA,ornA,posB,ornB,DeltaC2,vertices,uniqueEdges,faces,indices,&sepNormal,&dmin);
+ if (sepEE)
+ {
+ gpuCompoundSepNormalsOut[i] = sepNormal;//fastNormalize4(sepNormal);
+ gpuHasCompoundSepNormalsOut[i] = 1;
+ }//sepEE
+ }//(!sepB)
+ }//(!sepA)
+
+
+ }
+
+}
+
+
+inline b3Float4 MyUnQuantize(const unsigned short* vecIn, b3Float4 quantization, b3Float4 bvhAabbMin)
+{
+ b3Float4 vecOut;
+ vecOut = b3MakeFloat4(
+ (float)(vecIn[0]) / (quantization.x),
+ (float)(vecIn[1]) / (quantization.y),
+ (float)(vecIn[2]) / (quantization.z),
+ 0.f);
+
+ vecOut += bvhAabbMin;
+ return vecOut;
+}
+
+inline b3Float4 MyUnQuantizeGlobal(__global const unsigned short* vecIn, b3Float4 quantization, b3Float4 bvhAabbMin)
+{
+ b3Float4 vecOut;
+ vecOut = b3MakeFloat4(
+ (float)(vecIn[0]) / (quantization.x),
+ (float)(vecIn[1]) / (quantization.y),
+ (float)(vecIn[2]) / (quantization.z),
+ 0.f);
+
+ vecOut += bvhAabbMin;
+ return vecOut;
+}
+
+
+// work-in-progress
+__kernel void findCompoundPairsKernel( __global const int4* pairs,
+ __global const BodyData* rigidBodies,
+ __global const btCollidableGpu* collidables,
+ __global const ConvexPolyhedronCL* convexShapes,
+ __global const float4* vertices,
+ __global const float4* uniqueEdges,
+ __global const btGpuFace* faces,
+ __global const int* indices,
+ __global b3Aabb_t* aabbLocalSpace,
+ __global const btGpuChildShape* gpuChildShapes,
+ __global volatile int4* gpuCompoundPairsOut,
+ __global volatile int* numCompoundPairsOut,
+ __global const b3BvhSubtreeInfo* subtrees,
+ __global const b3QuantizedBvhNode* quantizedNodes,
+ __global const b3BvhInfo* bvhInfos,
+ int numPairs,
+ int maxNumCompoundPairsCapacity
+ )
+{
+
+ int i = get_global_id(0);
+
+ if (i<numPairs)
+ {
+ int bodyIndexA = pairs[i].x;
+ int bodyIndexB = pairs[i].y;
+
+ int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;
+ int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;
+
+ int shapeIndexA = collidables[collidableIndexA].m_shapeIndex;
+ int shapeIndexB = collidables[collidableIndexB].m_shapeIndex;
+
+
+ //once the broadphase avoids static-static pairs, we can remove this test
+ if ((rigidBodies[bodyIndexA].m_invMass==0) &&(rigidBodies[bodyIndexB].m_invMass==0))
+ {
+ return;
+ }
+
+ if ((collidables[collidableIndexA].m_shapeType==SHAPE_COMPOUND_OF_CONVEX_HULLS) &&(collidables[collidableIndexB].m_shapeType==SHAPE_COMPOUND_OF_CONVEX_HULLS))
+ {
+ int bvhA = collidables[collidableIndexA].m_compoundBvhIndex;
+ int bvhB = collidables[collidableIndexB].m_compoundBvhIndex;
+ int numSubTreesA = bvhInfos[bvhA].m_numSubTrees;
+ int subTreesOffsetA = bvhInfos[bvhA].m_subTreeOffset;
+ int subTreesOffsetB = bvhInfos[bvhB].m_subTreeOffset;
+
+
+ int numSubTreesB = bvhInfos[bvhB].m_numSubTrees;
+
+ float4 posA = rigidBodies[bodyIndexA].m_pos;
+ b3Quat ornA = rigidBodies[bodyIndexA].m_quat;
+
+ b3Quat ornB = rigidBodies[bodyIndexB].m_quat;
+ float4 posB = rigidBodies[bodyIndexB].m_pos;
+
+
+ for (int p=0;p<numSubTreesA;p++)
+ {
+ b3BvhSubtreeInfo subtreeA = subtrees[subTreesOffsetA+p];
+ //bvhInfos[bvhA].m_quantization
+ b3Float4 treeAminLocal = MyUnQuantize(subtreeA.m_quantizedAabbMin,bvhInfos[bvhA].m_quantization,bvhInfos[bvhA].m_aabbMin);
+ b3Float4 treeAmaxLocal = MyUnQuantize(subtreeA.m_quantizedAabbMax,bvhInfos[bvhA].m_quantization,bvhInfos[bvhA].m_aabbMin);
+
+ b3Float4 aabbAMinOut,aabbAMaxOut;
+ float margin=0.f;
+ b3TransformAabb2(treeAminLocal,treeAmaxLocal, margin,posA,ornA,&aabbAMinOut,&aabbAMaxOut);
+
+ for (int q=0;q<numSubTreesB;q++)
+ {
+ b3BvhSubtreeInfo subtreeB = subtrees[subTreesOffsetB+q];
+
+ b3Float4 treeBminLocal = MyUnQuantize(subtreeB.m_quantizedAabbMin,bvhInfos[bvhB].m_quantization,bvhInfos[bvhB].m_aabbMin);
+ b3Float4 treeBmaxLocal = MyUnQuantize(subtreeB.m_quantizedAabbMax,bvhInfos[bvhB].m_quantization,bvhInfos[bvhB].m_aabbMin);
+
+ b3Float4 aabbBMinOut,aabbBMaxOut;
+ float margin=0.f;
+ b3TransformAabb2(treeBminLocal,treeBmaxLocal, margin,posB,ornB,&aabbBMinOut,&aabbBMaxOut);
+
+
+
+ bool aabbOverlap = b3TestAabbAgainstAabb(aabbAMinOut,aabbAMaxOut,aabbBMinOut,aabbBMaxOut);
+ if (aabbOverlap)
+ {
+
+ int startNodeIndexA = subtreeA.m_rootNodeIndex+bvhInfos[bvhA].m_nodeOffset;
+ int endNodeIndexA = startNodeIndexA+subtreeA.m_subtreeSize;
+
+ int startNodeIndexB = subtreeB.m_rootNodeIndex+bvhInfos[bvhB].m_nodeOffset;
+ int endNodeIndexB = startNodeIndexB+subtreeB.m_subtreeSize;
+
+
+ b3Int2 nodeStack[B3_MAX_STACK_DEPTH];
+ b3Int2 node0;
+ node0.x = startNodeIndexA;
+ node0.y = startNodeIndexB;
+ int maxStackDepth = B3_MAX_STACK_DEPTH;
+ int depth=0;
+ nodeStack[depth++]=node0;
+
+ do
+ {
+ b3Int2 node = nodeStack[--depth];
+
+ b3Float4 aMinLocal = MyUnQuantizeGlobal(quantizedNodes[node.x].m_quantizedAabbMin,bvhInfos[bvhA].m_quantization,bvhInfos[bvhA].m_aabbMin);
+ b3Float4 aMaxLocal = MyUnQuantizeGlobal(quantizedNodes[node.x].m_quantizedAabbMax,bvhInfos[bvhA].m_quantization,bvhInfos[bvhA].m_aabbMin);
+
+ b3Float4 bMinLocal = MyUnQuantizeGlobal(quantizedNodes[node.y].m_quantizedAabbMin,bvhInfos[bvhB].m_quantization,bvhInfos[bvhB].m_aabbMin);
+ b3Float4 bMaxLocal = MyUnQuantizeGlobal(quantizedNodes[node.y].m_quantizedAabbMax,bvhInfos[bvhB].m_quantization,bvhInfos[bvhB].m_aabbMin);
+
+ float margin=0.f;
+ b3Float4 aabbAMinOut,aabbAMaxOut;
+ b3TransformAabb2(aMinLocal,aMaxLocal, margin,posA,ornA,&aabbAMinOut,&aabbAMaxOut);
+
+ b3Float4 aabbBMinOut,aabbBMaxOut;
+ b3TransformAabb2(bMinLocal,bMaxLocal, margin,posB,ornB,&aabbBMinOut,&aabbBMaxOut);
+
+
+ bool nodeOverlap = b3TestAabbAgainstAabb(aabbAMinOut,aabbAMaxOut,aabbBMinOut,aabbBMaxOut);
+ if (nodeOverlap)
+ {
+ bool isLeafA = isLeafNodeGlobal(&quantizedNodes[node.x]);
+ bool isLeafB = isLeafNodeGlobal(&quantizedNodes[node.y]);
+ bool isInternalA = !isLeafA;
+ bool isInternalB = !isLeafB;
+
+ //fail, even though it might hit two leaf nodes
+ if (depth+4>maxStackDepth && !(isLeafA && isLeafB))
+ {
+ //printf("Error: traversal exceeded maxStackDepth");
+ continue;
+ }
+
+ if(isInternalA)
+ {
+ int nodeAleftChild = node.x+1;
+ bool isNodeALeftChildLeaf = isLeafNodeGlobal(&quantizedNodes[node.x+1]);
+ int nodeArightChild = isNodeALeftChildLeaf? node.x+2 : node.x+1 + getEscapeIndexGlobal(&quantizedNodes[node.x+1]);
+
+ if(isInternalB)
+ {
+ int nodeBleftChild = node.y+1;
+ bool isNodeBLeftChildLeaf = isLeafNodeGlobal(&quantizedNodes[node.y+1]);
+ int nodeBrightChild = isNodeBLeftChildLeaf? node.y+2 : node.y+1 + getEscapeIndexGlobal(&quantizedNodes[node.y+1]);
+
+ nodeStack[depth++] = b3MakeInt2(nodeAleftChild, nodeBleftChild);
+ nodeStack[depth++] = b3MakeInt2(nodeArightChild, nodeBleftChild);
+ nodeStack[depth++] = b3MakeInt2(nodeAleftChild, nodeBrightChild);
+ nodeStack[depth++] = b3MakeInt2(nodeArightChild, nodeBrightChild);
+ }
+ else
+ {
+ nodeStack[depth++] = b3MakeInt2(nodeAleftChild,node.y);
+ nodeStack[depth++] = b3MakeInt2(nodeArightChild,node.y);
+ }
+ }
+ else
+ {
+ if(isInternalB)
+ {
+ int nodeBleftChild = node.y+1;
+ bool isNodeBLeftChildLeaf = isLeafNodeGlobal(&quantizedNodes[node.y+1]);
+ int nodeBrightChild = isNodeBLeftChildLeaf? node.y+2 : node.y+1 + getEscapeIndexGlobal(&quantizedNodes[node.y+1]);
+ nodeStack[depth++] = b3MakeInt2(node.x,nodeBleftChild);
+ nodeStack[depth++] = b3MakeInt2(node.x,nodeBrightChild);
+ }
+ else
+ {
+ int compoundPairIdx = atomic_inc(numCompoundPairsOut);
+ if (compoundPairIdx<maxNumCompoundPairsCapacity)
+ {
+ int childShapeIndexA = getTriangleIndexGlobal(&quantizedNodes[node.x]);
+ int childShapeIndexB = getTriangleIndexGlobal(&quantizedNodes[node.y]);
+ gpuCompoundPairsOut[compoundPairIdx] = (int4)(bodyIndexA,bodyIndexB,childShapeIndexA,childShapeIndexB);
+ }
+ }
+ }
+ }
+ } while (depth);
+ }
+ }
+ }
+
+ return;
+ }
+
+
+
+
+
+ if ((collidables[collidableIndexA].m_shapeType==SHAPE_COMPOUND_OF_CONVEX_HULLS) ||(collidables[collidableIndexB].m_shapeType==SHAPE_COMPOUND_OF_CONVEX_HULLS))
+ {
+
+ if (collidables[collidableIndexA].m_shapeType==SHAPE_COMPOUND_OF_CONVEX_HULLS)
+ {
+
+ int numChildrenA = collidables[collidableIndexA].m_numChildShapes;
+ for (int c=0;c<numChildrenA;c++)
+ {
+ int childShapeIndexA = collidables[collidableIndexA].m_shapeIndex+c;
+ int childColIndexA = gpuChildShapes[childShapeIndexA].m_shapeIndex;
+
+ float4 posA = rigidBodies[bodyIndexA].m_pos;
+ float4 ornA = rigidBodies[bodyIndexA].m_quat;
+ float4 childPosA = gpuChildShapes[childShapeIndexA].m_childPosition;
+ float4 childOrnA = gpuChildShapes[childShapeIndexA].m_childOrientation;
+ float4 newPosA = qtRotate(ornA,childPosA)+posA;
+ float4 newOrnA = qtMul(ornA,childOrnA);
+
+ int shapeIndexA = collidables[childColIndexA].m_shapeIndex;
+ b3Aabb_t aabbAlocal = aabbLocalSpace[shapeIndexA];
+ float margin = 0.f;
+
+ b3Float4 aabbAMinWS;
+ b3Float4 aabbAMaxWS;
+
+ b3TransformAabb2(aabbAlocal.m_minVec,aabbAlocal.m_maxVec,margin,
+ newPosA,
+ newOrnA,
+ &aabbAMinWS,&aabbAMaxWS);
+
+
+ if (collidables[collidableIndexB].m_shapeType==SHAPE_COMPOUND_OF_CONVEX_HULLS)
+ {
+ int numChildrenB = collidables[collidableIndexB].m_numChildShapes;
+ for (int b=0;b<numChildrenB;b++)
+ {
+ int childShapeIndexB = collidables[collidableIndexB].m_shapeIndex+b;
+ int childColIndexB = gpuChildShapes[childShapeIndexB].m_shapeIndex;
+ float4 ornB = rigidBodies[bodyIndexB].m_quat;
+ float4 posB = rigidBodies[bodyIndexB].m_pos;
+ float4 childPosB = gpuChildShapes[childShapeIndexB].m_childPosition;
+ float4 childOrnB = gpuChildShapes[childShapeIndexB].m_childOrientation;
+ float4 newPosB = transform(&childPosB,&posB,&ornB);
+ float4 newOrnB = qtMul(ornB,childOrnB);
+
+ int shapeIndexB = collidables[childColIndexB].m_shapeIndex;
+ b3Aabb_t aabbBlocal = aabbLocalSpace[shapeIndexB];
+
+ b3Float4 aabbBMinWS;
+ b3Float4 aabbBMaxWS;
+
+ b3TransformAabb2(aabbBlocal.m_minVec,aabbBlocal.m_maxVec,margin,
+ newPosB,
+ newOrnB,
+ &aabbBMinWS,&aabbBMaxWS);
+
+
+
+ bool aabbOverlap = b3TestAabbAgainstAabb(aabbAMinWS,aabbAMaxWS,aabbBMinWS,aabbBMaxWS);
+ if (aabbOverlap)
+ {
+ int numFacesA = convexShapes[shapeIndexA].m_numFaces;
+ float dmin = FLT_MAX;
+ float4 posA = newPosA;
+ posA.w = 0.f;
+ float4 posB = newPosB;
+ posB.w = 0.f;
+ float4 c0local = convexShapes[shapeIndexA].m_localCenter;
+ float4 ornA = newOrnA;
+ float4 c0 = transform(&c0local, &posA, &ornA);
+ float4 c1local = convexShapes[shapeIndexB].m_localCenter;
+ float4 ornB =newOrnB;
+ float4 c1 = transform(&c1local,&posB,&ornB);
+ const float4 DeltaC2 = c0 - c1;
+
+ {//
+ int compoundPairIdx = atomic_inc(numCompoundPairsOut);
+ if (compoundPairIdx<maxNumCompoundPairsCapacity)
+ {
+ gpuCompoundPairsOut[compoundPairIdx] = (int4)(bodyIndexA,bodyIndexB,childShapeIndexA,childShapeIndexB);
+ }
+ }//
+ }//fi(1)
+ } //for (int b=0
+ }//if (collidables[collidableIndexB].
+ else//if (collidables[collidableIndexB].m_shapeType==SHAPE_COMPOUND_OF_CONVEX_HULLS)
+ {
+ if (1)
+ {
+ int numFacesA = convexShapes[shapeIndexA].m_numFaces;
+ float dmin = FLT_MAX;
+ float4 posA = newPosA;
+ posA.w = 0.f;
+ float4 posB = rigidBodies[bodyIndexB].m_pos;
+ posB.w = 0.f;
+ float4 c0local = convexShapes[shapeIndexA].m_localCenter;
+ float4 ornA = newOrnA;
+ float4 c0 = transform(&c0local, &posA, &ornA);
+ float4 c1local = convexShapes[shapeIndexB].m_localCenter;
+ float4 ornB = rigidBodies[bodyIndexB].m_quat;
+ float4 c1 = transform(&c1local,&posB,&ornB);
+ const float4 DeltaC2 = c0 - c1;
+
+ {
+ int compoundPairIdx = atomic_inc(numCompoundPairsOut);
+ if (compoundPairIdx<maxNumCompoundPairsCapacity)
+ {
+ gpuCompoundPairsOut[compoundPairIdx] = (int4)(bodyIndexA,bodyIndexB,childShapeIndexA,-1);
+ }//if (compoundPairIdx<maxNumCompoundPairsCapacity)
+ }//
+ }//fi (1)
+ }//if (collidables[collidableIndexB].m_shapeType==SHAPE_COMPOUND_OF_CONVEX_HULLS)
+ }//for (int b=0;b<numChildrenB;b++)
+ return;
+ }//if (collidables[collidableIndexB].m_shapeType==SHAPE_COMPOUND_OF_CONVEX_HULLS)
+ if ((collidables[collidableIndexA].m_shapeType!=SHAPE_CONCAVE_TRIMESH)
+ && (collidables[collidableIndexB].m_shapeType==SHAPE_COMPOUND_OF_CONVEX_HULLS))
+ {
+ int numChildrenB = collidables[collidableIndexB].m_numChildShapes;
+ for (int b=0;b<numChildrenB;b++)
+ {
+ int childShapeIndexB = collidables[collidableIndexB].m_shapeIndex+b;
+ int childColIndexB = gpuChildShapes[childShapeIndexB].m_shapeIndex;
+ float4 ornB = rigidBodies[bodyIndexB].m_quat;
+ float4 posB = rigidBodies[bodyIndexB].m_pos;
+ float4 childPosB = gpuChildShapes[childShapeIndexB].m_childPosition;
+ float4 childOrnB = gpuChildShapes[childShapeIndexB].m_childOrientation;
+ float4 newPosB = qtRotate(ornB,childPosB)+posB;
+ float4 newOrnB = qtMul(ornB,childOrnB);
+
+ int shapeIndexB = collidables[childColIndexB].m_shapeIndex;
+
+
+ //////////////////////////////////////
+
+ if (1)
+ {
+ int numFacesA = convexShapes[shapeIndexA].m_numFaces;
+ float dmin = FLT_MAX;
+ float4 posA = rigidBodies[bodyIndexA].m_pos;
+ posA.w = 0.f;
+ float4 posB = newPosB;
+ posB.w = 0.f;
+ float4 c0local = convexShapes[shapeIndexA].m_localCenter;
+ float4 ornA = rigidBodies[bodyIndexA].m_quat;
+ float4 c0 = transform(&c0local, &posA, &ornA);
+ float4 c1local = convexShapes[shapeIndexB].m_localCenter;
+ float4 ornB =newOrnB;
+ float4 c1 = transform(&c1local,&posB,&ornB);
+ const float4 DeltaC2 = c0 - c1;
+ {//
+ int compoundPairIdx = atomic_inc(numCompoundPairsOut);
+ if (compoundPairIdx<maxNumCompoundPairsCapacity)
+ {
+ gpuCompoundPairsOut[compoundPairIdx] = (int4)(bodyIndexA,bodyIndexB,-1,childShapeIndexB);
+ }//fi (compoundPairIdx<maxNumCompoundPairsCapacity)
+ }//
+ }//fi (1)
+ }//for (int b=0;b<numChildrenB;b++)
+ return;
+ }//if (collidables[collidableIndexB].m_shapeType==SHAPE_COMPOUND_OF_CONVEX_HULLS)
+ return;
+ }//fi ((collidables[collidableIndexA].m_shapeType==SHAPE_COMPOUND_OF_CONVEX_HULLS) ||(collidables[collidableIndexB].m_shapeType==SHAPE_COMPOUND_OF_CONVEX_HULLS))
+ }//i<numPairs
+}
+
+// work-in-progress
+__kernel void findSeparatingAxisKernel( __global const int4* pairs,
+ __global const BodyData* rigidBodies,
+ __global const btCollidableGpu* collidables,
+ __global const ConvexPolyhedronCL* convexShapes,
+ __global const float4* vertices,
+ __global const float4* uniqueEdges,
+ __global const btGpuFace* faces,
+ __global const int* indices,
+ __global btAabbCL* aabbs,
+ __global volatile float4* separatingNormals,
+ __global volatile int* hasSeparatingAxis,
+ int numPairs
+ )
+{
+
+ int i = get_global_id(0);
+
+ if (i<numPairs)
+ {
+
+
+ int bodyIndexA = pairs[i].x;
+ int bodyIndexB = pairs[i].y;
+
+ int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;
+ int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;
+
+ int shapeIndexA = collidables[collidableIndexA].m_shapeIndex;
+ int shapeIndexB = collidables[collidableIndexB].m_shapeIndex;
+
+
+ //once the broadphase avoids static-static pairs, we can remove this test
+ if ((rigidBodies[bodyIndexA].m_invMass==0) &&(rigidBodies[bodyIndexB].m_invMass==0))
+ {
+ hasSeparatingAxis[i] = 0;
+ return;
+ }
+
+
+ if ((collidables[collidableIndexA].m_shapeType!=SHAPE_CONVEX_HULL) ||(collidables[collidableIndexB].m_shapeType!=SHAPE_CONVEX_HULL))
+ {
+ hasSeparatingAxis[i] = 0;
+ return;
+ }
+
+ if ((collidables[collidableIndexA].m_shapeType==SHAPE_CONCAVE_TRIMESH))
+ {
+ hasSeparatingAxis[i] = 0;
+ return;
+ }
+
+ int numFacesA = convexShapes[shapeIndexA].m_numFaces;
+
+ float dmin = FLT_MAX;
+
+ float4 posA = rigidBodies[bodyIndexA].m_pos;
+ posA.w = 0.f;
+ float4 posB = rigidBodies[bodyIndexB].m_pos;
+ posB.w = 0.f;
+ float4 c0local = convexShapes[shapeIndexA].m_localCenter;
+ float4 ornA = rigidBodies[bodyIndexA].m_quat;
+ float4 c0 = transform(&c0local, &posA, &ornA);
+ float4 c1local = convexShapes[shapeIndexB].m_localCenter;
+ float4 ornB =rigidBodies[bodyIndexB].m_quat;
+ float4 c1 = transform(&c1local,&posB,&ornB);
+ const float4 DeltaC2 = c0 - c1;
+ float4 sepNormal;
+
+ bool sepA = findSeparatingAxis( &convexShapes[shapeIndexA], &convexShapes[shapeIndexB],posA,ornA,
+ posB,ornB,
+ DeltaC2,
+ vertices,uniqueEdges,faces,
+ indices,&sepNormal,&dmin);
+ hasSeparatingAxis[i] = 4;
+ if (!sepA)
+ {
+ hasSeparatingAxis[i] = 0;
+ } else
+ {
+ bool sepB = findSeparatingAxis( &convexShapes[shapeIndexB],&convexShapes[shapeIndexA],posB,ornB,
+ posA,ornA,
+ DeltaC2,
+ vertices,uniqueEdges,faces,
+ indices,&sepNormal,&dmin);
+
+ if (!sepB)
+ {
+ hasSeparatingAxis[i] = 0;
+ } else
+ {
+ bool sepEE = findSeparatingAxisEdgeEdge( &convexShapes[shapeIndexA], &convexShapes[shapeIndexB],posA,ornA,
+ posB,ornB,
+ DeltaC2,
+ vertices,uniqueEdges,faces,
+ indices,&sepNormal,&dmin);
+ if (!sepEE)
+ {
+ hasSeparatingAxis[i] = 0;
+ } else
+ {
+ hasSeparatingAxis[i] = 1;
+ separatingNormals[i] = sepNormal;
+ }
+ }
+ }
+
+ }
+
+}
+
+
+__kernel void findSeparatingAxisVertexFaceKernel( __global const int4* pairs,
+ __global const BodyData* rigidBodies,
+ __global const btCollidableGpu* collidables,
+ __global const ConvexPolyhedronCL* convexShapes,
+ __global const float4* vertices,
+ __global const float4* uniqueEdges,
+ __global const btGpuFace* faces,
+ __global const int* indices,
+ __global btAabbCL* aabbs,
+ __global volatile float4* separatingNormals,
+ __global volatile int* hasSeparatingAxis,
+ __global float* dmins,
+ int numPairs
+ )
+{
+
+ int i = get_global_id(0);
+
+ if (i<numPairs)
+ {
+
+
+ int bodyIndexA = pairs[i].x;
+ int bodyIndexB = pairs[i].y;
+
+ int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;
+ int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;
+
+ int shapeIndexA = collidables[collidableIndexA].m_shapeIndex;
+ int shapeIndexB = collidables[collidableIndexB].m_shapeIndex;
+
+ hasSeparatingAxis[i] = 0;
+
+ //once the broadphase avoids static-static pairs, we can remove this test
+ if ((rigidBodies[bodyIndexA].m_invMass==0) &&(rigidBodies[bodyIndexB].m_invMass==0))
+ {
+ return;
+ }
+
+
+ if ((collidables[collidableIndexA].m_shapeType!=SHAPE_CONVEX_HULL) ||(collidables[collidableIndexB].m_shapeType!=SHAPE_CONVEX_HULL))
+ {
+ return;
+ }
+
+
+ int numFacesA = convexShapes[shapeIndexA].m_numFaces;
+
+ float dmin = FLT_MAX;
+
+ dmins[i] = dmin;
+
+ float4 posA = rigidBodies[bodyIndexA].m_pos;
+ posA.w = 0.f;
+ float4 posB = rigidBodies[bodyIndexB].m_pos;
+ posB.w = 0.f;
+ float4 c0local = convexShapes[shapeIndexA].m_localCenter;
+ float4 ornA = rigidBodies[bodyIndexA].m_quat;
+ float4 c0 = transform(&c0local, &posA, &ornA);
+ float4 c1local = convexShapes[shapeIndexB].m_localCenter;
+ float4 ornB =rigidBodies[bodyIndexB].m_quat;
+ float4 c1 = transform(&c1local,&posB,&ornB);
+ const float4 DeltaC2 = c0 - c1;
+ float4 sepNormal;
+
+ bool sepA = findSeparatingAxis( &convexShapes[shapeIndexA], &convexShapes[shapeIndexB],posA,ornA,
+ posB,ornB,
+ DeltaC2,
+ vertices,uniqueEdges,faces,
+ indices,&sepNormal,&dmin);
+ hasSeparatingAxis[i] = 4;
+ if (!sepA)
+ {
+ hasSeparatingAxis[i] = 0;
+ } else
+ {
+ bool sepB = findSeparatingAxis( &convexShapes[shapeIndexB],&convexShapes[shapeIndexA],posB,ornB,
+ posA,ornA,
+ DeltaC2,
+ vertices,uniqueEdges,faces,
+ indices,&sepNormal,&dmin);
+
+ if (sepB)
+ {
+ dmins[i] = dmin;
+ hasSeparatingAxis[i] = 1;
+ separatingNormals[i] = sepNormal;
+ }
+ }
+
+ }
+
+}
+
+
+__kernel void findSeparatingAxisEdgeEdgeKernel( __global const int4* pairs,
+ __global const BodyData* rigidBodies,
+ __global const btCollidableGpu* collidables,
+ __global const ConvexPolyhedronCL* convexShapes,
+ __global const float4* vertices,
+ __global const float4* uniqueEdges,
+ __global const btGpuFace* faces,
+ __global const int* indices,
+ __global btAabbCL* aabbs,
+ __global float4* separatingNormals,
+ __global int* hasSeparatingAxis,
+ __global float* dmins,
+ __global const float4* unitSphereDirections,
+ int numUnitSphereDirections,
+ int numPairs
+ )
+{
+
+ int i = get_global_id(0);
+
+ if (i<numPairs)
+ {
+
+ if (hasSeparatingAxis[i])
+ {
+
+ int bodyIndexA = pairs[i].x;
+ int bodyIndexB = pairs[i].y;
+
+ int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;
+ int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;
+
+ int shapeIndexA = collidables[collidableIndexA].m_shapeIndex;
+ int shapeIndexB = collidables[collidableIndexB].m_shapeIndex;
+
+
+ int numFacesA = convexShapes[shapeIndexA].m_numFaces;
+
+ float dmin = dmins[i];
+
+ float4 posA = rigidBodies[bodyIndexA].m_pos;
+ posA.w = 0.f;
+ float4 posB = rigidBodies[bodyIndexB].m_pos;
+ posB.w = 0.f;
+ float4 c0local = convexShapes[shapeIndexA].m_localCenter;
+ float4 ornA = rigidBodies[bodyIndexA].m_quat;
+ float4 c0 = transform(&c0local, &posA, &ornA);
+ float4 c1local = convexShapes[shapeIndexB].m_localCenter;
+ float4 ornB =rigidBodies[bodyIndexB].m_quat;
+ float4 c1 = transform(&c1local,&posB,&ornB);
+ const float4 DeltaC2 = c0 - c1;
+ float4 sepNormal = separatingNormals[i];
+
+
+
+ bool sepEE = false;
+ int numEdgeEdgeDirections = convexShapes[shapeIndexA].m_numUniqueEdges*convexShapes[shapeIndexB].m_numUniqueEdges;
+ if (numEdgeEdgeDirections<=numUnitSphereDirections)
+ {
+ sepEE = findSeparatingAxisEdgeEdge( &convexShapes[shapeIndexA], &convexShapes[shapeIndexB],posA,ornA,
+ posB,ornB,
+ DeltaC2,
+ vertices,uniqueEdges,faces,
+ indices,&sepNormal,&dmin);
+
+ if (!sepEE)
+ {
+ hasSeparatingAxis[i] = 0;
+ } else
+ {
+ hasSeparatingAxis[i] = 1;
+ separatingNormals[i] = sepNormal;
+ }
+ }
+ /*
+ ///else case is a separate kernel, to make Mac OSX OpenCL compiler happy
+ else
+ {
+ sepEE = findSeparatingAxisUnitSphere(&convexShapes[shapeIndexA], &convexShapes[shapeIndexB],posA,ornA,
+ posB,ornB,
+ DeltaC2,
+ vertices,unitSphereDirections,numUnitSphereDirections,
+ &sepNormal,&dmin);
+ if (!sepEE)
+ {
+ hasSeparatingAxis[i] = 0;
+ } else
+ {
+ hasSeparatingAxis[i] = 1;
+ separatingNormals[i] = sepNormal;
+ }
+ }
+ */
+ } //if (hasSeparatingAxis[i])
+ }//(i<numPairs)
+}
+
+
+
+
+
+inline int findClippingFaces(const float4 separatingNormal,
+ const ConvexPolyhedronCL* hullA,
+ __global const ConvexPolyhedronCL* hullB,
+ const float4 posA, const Quaternion ornA,const float4 posB, const Quaternion ornB,
+ __global float4* worldVertsA1,
+ __global float4* worldNormalsA1,
+ __global float4* worldVertsB1,
+ int capacityWorldVerts,
+ const float minDist, float maxDist,
+ const float4* verticesA,
+ const btGpuFace* facesA,
+ const int* indicesA,
+ __global const float4* verticesB,
+ __global const btGpuFace* facesB,
+ __global const int* indicesB,
+ __global int4* clippingFaces, int pairIndex)
+{
+ int numContactsOut = 0;
+ int numWorldVertsB1= 0;
+
+
+ int closestFaceB=0;
+ float dmax = -FLT_MAX;
+
+ {
+ for(int face=0;face<hullB->m_numFaces;face++)
+ {
+ const float4 Normal = make_float4(facesB[hullB->m_faceOffset+face].m_plane.x,
+ facesB[hullB->m_faceOffset+face].m_plane.y, facesB[hullB->m_faceOffset+face].m_plane.z,0.f);
+ const float4 WorldNormal = qtRotate(ornB, Normal);
+ float d = dot3F4(WorldNormal,separatingNormal);
+ if (d > dmax)
+ {
+ dmax = d;
+ closestFaceB = face;
+ }
+ }
+ }
+
+ {
+ const btGpuFace polyB = facesB[hullB->m_faceOffset+closestFaceB];
+ int numVertices = polyB.m_numIndices;
+ if (numVertices>capacityWorldVerts)
+ numVertices = capacityWorldVerts;
+
+ for(int e0=0;e0<numVertices;e0++)
+ {
+ if (e0<capacityWorldVerts)
+ {
+ const float4 b = verticesB[hullB->m_vertexOffset+indicesB[polyB.m_indexOffset+e0]];
+ worldVertsB1[pairIndex*capacityWorldVerts+numWorldVertsB1++] = transform(&b,&posB,&ornB);
+ }
+ }
+ }
+
+ int closestFaceA=0;
+ {
+ float dmin = FLT_MAX;
+ for(int face=0;face<hullA->m_numFaces;face++)
+ {
+ const float4 Normal = make_float4(
+ facesA[hullA->m_faceOffset+face].m_plane.x,
+ facesA[hullA->m_faceOffset+face].m_plane.y,
+ facesA[hullA->m_faceOffset+face].m_plane.z,
+ 0.f);
+ const float4 faceANormalWS = qtRotate(ornA,Normal);
+
+ float d = dot3F4(faceANormalWS,separatingNormal);
+ if (d < dmin)
+ {
+ dmin = d;
+ closestFaceA = face;
+ worldNormalsA1[pairIndex] = faceANormalWS;
+ }
+ }
+ }
+
+ int numVerticesA = facesA[hullA->m_faceOffset+closestFaceA].m_numIndices;
+ if (numVerticesA>capacityWorldVerts)
+ numVerticesA = capacityWorldVerts;
+
+ for(int e0=0;e0<numVerticesA;e0++)
+ {
+ if (e0<capacityWorldVerts)
+ {
+ const float4 a = verticesA[hullA->m_vertexOffset+indicesA[facesA[hullA->m_faceOffset+closestFaceA].m_indexOffset+e0]];
+ worldVertsA1[pairIndex*capacityWorldVerts+e0] = transform(&a, &posA,&ornA);
+ }
+ }
+
+ clippingFaces[pairIndex].x = closestFaceA;
+ clippingFaces[pairIndex].y = closestFaceB;
+ clippingFaces[pairIndex].z = numVerticesA;
+ clippingFaces[pairIndex].w = numWorldVertsB1;
+
+
+ return numContactsOut;
+}
+
+
+
+
+// work-in-progress
+__kernel void findConcaveSeparatingAxisKernel( __global int4* concavePairs,
+ __global const BodyData* rigidBodies,
+ __global const btCollidableGpu* collidables,
+ __global const ConvexPolyhedronCL* convexShapes,
+ __global const float4* vertices,
+ __global const float4* uniqueEdges,
+ __global const btGpuFace* faces,
+ __global const int* indices,
+ __global const btGpuChildShape* gpuChildShapes,
+ __global btAabbCL* aabbs,
+ __global float4* concaveSeparatingNormalsOut,
+ __global int* concaveHasSeparatingNormals,
+ __global int4* clippingFacesOut,
+ __global float4* worldVertsA1GPU,
+ __global float4* worldNormalsAGPU,
+ __global float4* worldVertsB1GPU,
+ int vertexFaceCapacity,
+ int numConcavePairs
+ )
+{
+
+ int i = get_global_id(0);
+ if (i>=numConcavePairs)
+ return;
+
+ concaveHasSeparatingNormals[i] = 0;
+
+ int pairIdx = i;
+
+ int bodyIndexA = concavePairs[i].x;
+ int bodyIndexB = concavePairs[i].y;
+
+ int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;
+ int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;
+
+ int shapeIndexA = collidables[collidableIndexA].m_shapeIndex;
+ int shapeIndexB = collidables[collidableIndexB].m_shapeIndex;
+
+ if (collidables[collidableIndexB].m_shapeType!=SHAPE_CONVEX_HULL&&
+ collidables[collidableIndexB].m_shapeType!=SHAPE_COMPOUND_OF_CONVEX_HULLS)
+ {
+ concavePairs[pairIdx].w = -1;
+ return;
+ }
+
+
+
+ int numFacesA = convexShapes[shapeIndexA].m_numFaces;
+ int numActualConcaveConvexTests = 0;
+
+ int f = concavePairs[i].z;
+
+ bool overlap = false;
+
+ ConvexPolyhedronCL convexPolyhedronA;
+
+ //add 3 vertices of the triangle
+ convexPolyhedronA.m_numVertices = 3;
+ convexPolyhedronA.m_vertexOffset = 0;
+ float4 localCenter = make_float4(0.f,0.f,0.f,0.f);
+
+ btGpuFace face = faces[convexShapes[shapeIndexA].m_faceOffset+f];
+ float4 triMinAabb, triMaxAabb;
+ btAabbCL triAabb;
+ triAabb.m_min = make_float4(1e30f,1e30f,1e30f,0.f);
+ triAabb.m_max = make_float4(-1e30f,-1e30f,-1e30f,0.f);
+
+ float4 verticesA[3];
+ for (int i=0;i<3;i++)
+ {
+ int index = indices[face.m_indexOffset+i];
+ float4 vert = vertices[convexShapes[shapeIndexA].m_vertexOffset+index];
+ verticesA[i] = vert;
+ localCenter += vert;
+
+ triAabb.m_min = min(triAabb.m_min,vert);
+ triAabb.m_max = max(triAabb.m_max,vert);
+
+ }
+
+ overlap = true;
+ overlap = (triAabb.m_min.x > aabbs[bodyIndexB].m_max.x || triAabb.m_max.x < aabbs[bodyIndexB].m_min.x) ? false : overlap;
+ overlap = (triAabb.m_min.z > aabbs[bodyIndexB].m_max.z || triAabb.m_max.z < aabbs[bodyIndexB].m_min.z) ? false : overlap;
+ overlap = (triAabb.m_min.y > aabbs[bodyIndexB].m_max.y || triAabb.m_max.y < aabbs[bodyIndexB].m_min.y) ? false : overlap;
+
+ if (overlap)
+ {
+ float dmin = FLT_MAX;
+ int hasSeparatingAxis=5;
+ float4 sepAxis=make_float4(1,2,3,4);
+
+ int localCC=0;
+ numActualConcaveConvexTests++;
+
+ //a triangle has 3 unique edges
+ convexPolyhedronA.m_numUniqueEdges = 3;
+ convexPolyhedronA.m_uniqueEdgesOffset = 0;
+ float4 uniqueEdgesA[3];
+
+ uniqueEdgesA[0] = (verticesA[1]-verticesA[0]);
+ uniqueEdgesA[1] = (verticesA[2]-verticesA[1]);
+ uniqueEdgesA[2] = (verticesA[0]-verticesA[2]);
+
+
+ convexPolyhedronA.m_faceOffset = 0;
+
+ float4 normal = make_float4(face.m_plane.x,face.m_plane.y,face.m_plane.z,0.f);
+
+ btGpuFace facesA[TRIANGLE_NUM_CONVEX_FACES];
+ int indicesA[3+3+2+2+2];
+ int curUsedIndices=0;
+ int fidx=0;
+
+ //front size of triangle
+ {
+ facesA[fidx].m_indexOffset=curUsedIndices;
+ indicesA[0] = 0;
+ indicesA[1] = 1;
+ indicesA[2] = 2;
+ curUsedIndices+=3;
+ float c = face.m_plane.w;
+ facesA[fidx].m_plane.x = normal.x;
+ facesA[fidx].m_plane.y = normal.y;
+ facesA[fidx].m_plane.z = normal.z;
+ facesA[fidx].m_plane.w = c;
+ facesA[fidx].m_numIndices=3;
+ }
+ fidx++;
+ //back size of triangle
+ {
+ facesA[fidx].m_indexOffset=curUsedIndices;
+ indicesA[3]=2;
+ indicesA[4]=1;
+ indicesA[5]=0;
+ curUsedIndices+=3;
+ float c = dot(normal,verticesA[0]);
+ float c1 = -face.m_plane.w;
+ facesA[fidx].m_plane.x = -normal.x;
+ facesA[fidx].m_plane.y = -normal.y;
+ facesA[fidx].m_plane.z = -normal.z;
+ facesA[fidx].m_plane.w = c;
+ facesA[fidx].m_numIndices=3;
+ }
+ fidx++;
+
+ bool addEdgePlanes = true;
+ if (addEdgePlanes)
+ {
+ int numVertices=3;
+ int prevVertex = numVertices-1;
+ for (int i=0;i<numVertices;i++)
+ {
+ float4 v0 = verticesA[i];
+ float4 v1 = verticesA[prevVertex];
+
+ float4 edgeNormal = normalize(cross(normal,v1-v0));
+ float c = -dot(edgeNormal,v0);
+
+ facesA[fidx].m_numIndices = 2;
+ facesA[fidx].m_indexOffset=curUsedIndices;
+ indicesA[curUsedIndices++]=i;
+ indicesA[curUsedIndices++]=prevVertex;
+
+ facesA[fidx].m_plane.x = edgeNormal.x;
+ facesA[fidx].m_plane.y = edgeNormal.y;
+ facesA[fidx].m_plane.z = edgeNormal.z;
+ facesA[fidx].m_plane.w = c;
+ fidx++;
+ prevVertex = i;
+ }
+ }
+ convexPolyhedronA.m_numFaces = TRIANGLE_NUM_CONVEX_FACES;
+ convexPolyhedronA.m_localCenter = localCenter*(1.f/3.f);
+
+
+ float4 posA = rigidBodies[bodyIndexA].m_pos;
+ posA.w = 0.f;
+ float4 posB = rigidBodies[bodyIndexB].m_pos;
+ posB.w = 0.f;
+
+ float4 ornA = rigidBodies[bodyIndexA].m_quat;
+ float4 ornB =rigidBodies[bodyIndexB].m_quat;
+
+
+
+
+ ///////////////////
+ ///compound shape support
+
+ if (collidables[collidableIndexB].m_shapeType==SHAPE_COMPOUND_OF_CONVEX_HULLS)
+ {
+ int compoundChild = concavePairs[pairIdx].w;
+ int childShapeIndexB = compoundChild;//collidables[collidableIndexB].m_shapeIndex+compoundChild;
+ int childColIndexB = gpuChildShapes[childShapeIndexB].m_shapeIndex;
+ float4 childPosB = gpuChildShapes[childShapeIndexB].m_childPosition;
+ float4 childOrnB = gpuChildShapes[childShapeIndexB].m_childOrientation;
+ float4 newPosB = transform(&childPosB,&posB,&ornB);
+ float4 newOrnB = qtMul(ornB,childOrnB);
+ posB = newPosB;
+ ornB = newOrnB;
+ shapeIndexB = collidables[childColIndexB].m_shapeIndex;
+ }
+ //////////////////
+
+ float4 c0local = convexPolyhedronA.m_localCenter;
+ float4 c0 = transform(&c0local, &posA, &ornA);
+ float4 c1local = convexShapes[shapeIndexB].m_localCenter;
+ float4 c1 = transform(&c1local,&posB,&ornB);
+ const float4 DeltaC2 = c0 - c1;
+
+
+ bool sepA = findSeparatingAxisLocalA( &convexPolyhedronA, &convexShapes[shapeIndexB],
+ posA,ornA,
+ posB,ornB,
+ DeltaC2,
+ verticesA,uniqueEdgesA,facesA,indicesA,
+ vertices,uniqueEdges,faces,indices,
+ &sepAxis,&dmin);
+ hasSeparatingAxis = 4;
+ if (!sepA)
+ {
+ hasSeparatingAxis = 0;
+ } else
+ {
+ bool sepB = findSeparatingAxisLocalB( &convexShapes[shapeIndexB],&convexPolyhedronA,
+ posB,ornB,
+ posA,ornA,
+ DeltaC2,
+ vertices,uniqueEdges,faces,indices,
+ verticesA,uniqueEdgesA,facesA,indicesA,
+ &sepAxis,&dmin);
+
+ if (!sepB)
+ {
+ hasSeparatingAxis = 0;
+ } else
+ {
+ bool sepEE = findSeparatingAxisEdgeEdgeLocalA( &convexPolyhedronA, &convexShapes[shapeIndexB],
+ posA,ornA,
+ posB,ornB,
+ DeltaC2,
+ verticesA,uniqueEdgesA,facesA,indicesA,
+ vertices,uniqueEdges,faces,indices,
+ &sepAxis,&dmin);
+
+ if (!sepEE)
+ {
+ hasSeparatingAxis = 0;
+ } else
+ {
+ hasSeparatingAxis = 1;
+ }
+ }
+ }
+
+ if (hasSeparatingAxis)
+ {
+ sepAxis.w = dmin;
+ concaveSeparatingNormalsOut[pairIdx]=sepAxis;
+ concaveHasSeparatingNormals[i]=1;
+
+
+ float minDist = -1e30f;
+ float maxDist = 0.02f;
+
+
+
+ findClippingFaces(sepAxis,
+ &convexPolyhedronA,
+ &convexShapes[shapeIndexB],
+ posA,ornA,
+ posB,ornB,
+ worldVertsA1GPU,
+ worldNormalsAGPU,
+ worldVertsB1GPU,
+ vertexFaceCapacity,
+ minDist, maxDist,
+ verticesA,
+ facesA,
+ indicesA,
+ vertices,
+ faces,
+ indices,
+ clippingFacesOut, pairIdx);
+
+
+ } else
+ {
+ //mark this pair as in-active
+ concavePairs[pairIdx].w = -1;
+ }
+ }
+ else
+ {
+ //mark this pair as in-active
+ concavePairs[pairIdx].w = -1;
+ }
+
+ concavePairs[pairIdx].z = -1;//now z is used for existing/persistent contacts
+}
+
+
+
diff --git a/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/satClipHullContacts.cl b/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/satClipHullContacts.cl
new file mode 100644
index 0000000000..f433971741
--- /dev/null
+++ b/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/satClipHullContacts.cl
@@ -0,0 +1,1888 @@
+
+#define TRIANGLE_NUM_CONVEX_FACES 5
+
+
+
+#pragma OPENCL EXTENSION cl_amd_printf : enable
+#pragma OPENCL EXTENSION cl_khr_local_int32_base_atomics : enable
+#pragma OPENCL EXTENSION cl_khr_global_int32_base_atomics : enable
+#pragma OPENCL EXTENSION cl_khr_local_int32_extended_atomics : enable
+#pragma OPENCL EXTENSION cl_khr_global_int32_extended_atomics : enable
+
+#ifdef cl_ext_atomic_counters_32
+#pragma OPENCL EXTENSION cl_ext_atomic_counters_32 : enable
+#else
+#define counter32_t volatile __global int*
+#endif
+
+#define GET_GROUP_IDX get_group_id(0)
+#define GET_LOCAL_IDX get_local_id(0)
+#define GET_GLOBAL_IDX get_global_id(0)
+#define GET_GROUP_SIZE get_local_size(0)
+#define GET_NUM_GROUPS get_num_groups(0)
+#define GROUP_LDS_BARRIER barrier(CLK_LOCAL_MEM_FENCE)
+#define GROUP_MEM_FENCE mem_fence(CLK_LOCAL_MEM_FENCE)
+#define AtomInc(x) atom_inc(&(x))
+#define AtomInc1(x, out) out = atom_inc(&(x))
+#define AppendInc(x, out) out = atomic_inc(x)
+#define AtomAdd(x, value) atom_add(&(x), value)
+#define AtomCmpxhg(x, cmp, value) atom_cmpxchg( &(x), cmp, value )
+#define AtomXhg(x, value) atom_xchg ( &(x), value )
+
+#define max2 max
+#define min2 min
+
+typedef unsigned int u32;
+
+
+
+#include "Bullet3Collision/NarrowPhaseCollision/shared/b3Contact4Data.h"
+#include "Bullet3Collision/NarrowPhaseCollision/shared/b3ConvexPolyhedronData.h"
+#include "Bullet3Collision/NarrowPhaseCollision/shared/b3Collidable.h"
+#include "Bullet3Collision/NarrowPhaseCollision/shared/b3RigidBodyData.h"
+
+
+
+#define GET_NPOINTS(x) (x).m_worldNormalOnB.w
+
+
+
+#define SELECT_UINT4( b, a, condition ) select( b,a,condition )
+
+#define make_float4 (float4)
+#define make_float2 (float2)
+#define make_uint4 (uint4)
+#define make_int4 (int4)
+#define make_uint2 (uint2)
+#define make_int2 (int2)
+
+
+__inline
+float fastDiv(float numerator, float denominator)
+{
+ return native_divide(numerator, denominator);
+// return numerator/denominator;
+}
+
+__inline
+float4 fastDiv4(float4 numerator, float4 denominator)
+{
+ return native_divide(numerator, denominator);
+}
+
+
+__inline
+float4 cross3(float4 a, float4 b)
+{
+ return cross(a,b);
+}
+
+//#define dot3F4 dot
+
+__inline
+float dot3F4(float4 a, float4 b)
+{
+ float4 a1 = make_float4(a.xyz,0.f);
+ float4 b1 = make_float4(b.xyz,0.f);
+ return dot(a1, b1);
+}
+
+__inline
+float4 fastNormalize4(float4 v)
+{
+ return fast_normalize(v);
+}
+
+
+///////////////////////////////////////
+// Quaternion
+///////////////////////////////////////
+
+typedef float4 Quaternion;
+
+__inline
+Quaternion qtMul(Quaternion a, Quaternion b);
+
+__inline
+Quaternion qtNormalize(Quaternion in);
+
+__inline
+float4 qtRotate(Quaternion q, float4 vec);
+
+__inline
+Quaternion qtInvert(Quaternion q);
+
+
+
+
+__inline
+Quaternion qtMul(Quaternion a, Quaternion b)
+{
+ Quaternion ans;
+ ans = cross3( a, b );
+ ans += a.w*b+b.w*a;
+// ans.w = a.w*b.w - (a.x*b.x+a.y*b.y+a.z*b.z);
+ ans.w = a.w*b.w - dot3F4(a, b);
+ return ans;
+}
+
+__inline
+Quaternion qtNormalize(Quaternion in)
+{
+ return fastNormalize4(in);
+// in /= length( in );
+// return in;
+}
+__inline
+float4 qtRotate(Quaternion q, float4 vec)
+{
+ Quaternion qInv = qtInvert( q );
+ float4 vcpy = vec;
+ vcpy.w = 0.f;
+ float4 out = qtMul(qtMul(q,vcpy),qInv);
+ return out;
+}
+
+__inline
+Quaternion qtInvert(Quaternion q)
+{
+ return (Quaternion)(-q.xyz, q.w);
+}
+
+__inline
+float4 qtInvRotate(const Quaternion q, float4 vec)
+{
+ return qtRotate( qtInvert( q ), vec );
+}
+
+__inline
+float4 transform(const float4* p, const float4* translation, const Quaternion* orientation)
+{
+ return qtRotate( *orientation, *p ) + (*translation);
+}
+
+
+
+__inline
+float4 normalize3(const float4 a)
+{
+ float4 n = make_float4(a.x, a.y, a.z, 0.f);
+ return fastNormalize4( n );
+}
+
+
+__inline float4 lerp3(const float4 a,const float4 b, float t)
+{
+ return make_float4( a.x + (b.x - a.x) * t,
+ a.y + (b.y - a.y) * t,
+ a.z + (b.z - a.z) * t,
+ 0.f);
+}
+
+
+
+// Clips a face to the back of a plane, return the number of vertices out, stored in ppVtxOut
+int clipFaceGlobal(__global const float4* pVtxIn, int numVertsIn, float4 planeNormalWS,float planeEqWS, __global float4* ppVtxOut)
+{
+
+ int ve;
+ float ds, de;
+ int numVertsOut = 0;
+ //double-check next test
+ if (numVertsIn < 2)
+ return 0;
+
+ float4 firstVertex=pVtxIn[numVertsIn-1];
+ float4 endVertex = pVtxIn[0];
+
+ ds = dot3F4(planeNormalWS,firstVertex)+planeEqWS;
+
+ for (ve = 0; ve < numVertsIn; ve++)
+ {
+ endVertex=pVtxIn[ve];
+ de = dot3F4(planeNormalWS,endVertex)+planeEqWS;
+ if (ds<0)
+ {
+ if (de<0)
+ {
+ // Start < 0, end < 0, so output endVertex
+ ppVtxOut[numVertsOut++] = endVertex;
+ }
+ else
+ {
+ // Start < 0, end >= 0, so output intersection
+ ppVtxOut[numVertsOut++] = lerp3(firstVertex, endVertex,(ds * 1.f/(ds - de)) );
+ }
+ }
+ else
+ {
+ if (de<0)
+ {
+ // Start >= 0, end < 0 so output intersection and end
+ ppVtxOut[numVertsOut++] = lerp3(firstVertex, endVertex,(ds * 1.f/(ds - de)) );
+ ppVtxOut[numVertsOut++] = endVertex;
+ }
+ }
+ firstVertex = endVertex;
+ ds = de;
+ }
+ return numVertsOut;
+}
+
+
+
+// Clips a face to the back of a plane, return the number of vertices out, stored in ppVtxOut
+int clipFace(const float4* pVtxIn, int numVertsIn, float4 planeNormalWS,float planeEqWS, float4* ppVtxOut)
+{
+
+ int ve;
+ float ds, de;
+ int numVertsOut = 0;
+//double-check next test
+ if (numVertsIn < 2)
+ return 0;
+
+ float4 firstVertex=pVtxIn[numVertsIn-1];
+ float4 endVertex = pVtxIn[0];
+
+ ds = dot3F4(planeNormalWS,firstVertex)+planeEqWS;
+
+ for (ve = 0; ve < numVertsIn; ve++)
+ {
+ endVertex=pVtxIn[ve];
+
+ de = dot3F4(planeNormalWS,endVertex)+planeEqWS;
+
+ if (ds<0)
+ {
+ if (de<0)
+ {
+ // Start < 0, end < 0, so output endVertex
+ ppVtxOut[numVertsOut++] = endVertex;
+ }
+ else
+ {
+ // Start < 0, end >= 0, so output intersection
+ ppVtxOut[numVertsOut++] = lerp3(firstVertex, endVertex,(ds * 1.f/(ds - de)) );
+ }
+ }
+ else
+ {
+ if (de<0)
+ {
+ // Start >= 0, end < 0 so output intersection and end
+ ppVtxOut[numVertsOut++] = lerp3(firstVertex, endVertex,(ds * 1.f/(ds - de)) );
+ ppVtxOut[numVertsOut++] = endVertex;
+ }
+ }
+ firstVertex = endVertex;
+ ds = de;
+ }
+ return numVertsOut;
+}
+
+
+int clipFaceAgainstHull(const float4 separatingNormal, __global const b3ConvexPolyhedronData_t* hullA,
+ const float4 posA, const Quaternion ornA, float4* worldVertsB1, int numWorldVertsB1,
+ float4* worldVertsB2, int capacityWorldVertsB2,
+ const float minDist, float maxDist,
+ __global const float4* vertices,
+ __global const b3GpuFace_t* faces,
+ __global const int* indices,
+ float4* contactsOut,
+ int contactCapacity)
+{
+ int numContactsOut = 0;
+
+ float4* pVtxIn = worldVertsB1;
+ float4* pVtxOut = worldVertsB2;
+
+ int numVertsIn = numWorldVertsB1;
+ int numVertsOut = 0;
+
+ int closestFaceA=-1;
+ {
+ float dmin = FLT_MAX;
+ for(int face=0;face<hullA->m_numFaces;face++)
+ {
+ const float4 Normal = make_float4(
+ faces[hullA->m_faceOffset+face].m_plane.x,
+ faces[hullA->m_faceOffset+face].m_plane.y,
+ faces[hullA->m_faceOffset+face].m_plane.z,0.f);
+ const float4 faceANormalWS = qtRotate(ornA,Normal);
+
+ float d = dot3F4(faceANormalWS,separatingNormal);
+ if (d < dmin)
+ {
+ dmin = d;
+ closestFaceA = face;
+ }
+ }
+ }
+ if (closestFaceA<0)
+ return numContactsOut;
+
+ b3GpuFace_t polyA = faces[hullA->m_faceOffset+closestFaceA];
+
+ // clip polygon to back of planes of all faces of hull A that are adjacent to witness face
+ int numVerticesA = polyA.m_numIndices;
+ for(int e0=0;e0<numVerticesA;e0++)
+ {
+ const float4 a = vertices[hullA->m_vertexOffset+indices[polyA.m_indexOffset+e0]];
+ const float4 b = vertices[hullA->m_vertexOffset+indices[polyA.m_indexOffset+((e0+1)%numVerticesA)]];
+ const float4 edge0 = a - b;
+ const float4 WorldEdge0 = qtRotate(ornA,edge0);
+ float4 planeNormalA = make_float4(polyA.m_plane.x,polyA.m_plane.y,polyA.m_plane.z,0.f);
+ float4 worldPlaneAnormal1 = qtRotate(ornA,planeNormalA);
+
+ float4 planeNormalWS1 = -cross3(WorldEdge0,worldPlaneAnormal1);
+ float4 worldA1 = transform(&a,&posA,&ornA);
+ float planeEqWS1 = -dot3F4(worldA1,planeNormalWS1);
+
+ float4 planeNormalWS = planeNormalWS1;
+ float planeEqWS=planeEqWS1;
+
+ //clip face
+ //clipFace(*pVtxIn, *pVtxOut,planeNormalWS,planeEqWS);
+ numVertsOut = clipFace(pVtxIn, numVertsIn, planeNormalWS,planeEqWS, pVtxOut);
+
+ //btSwap(pVtxIn,pVtxOut);
+ float4* tmp = pVtxOut;
+ pVtxOut = pVtxIn;
+ pVtxIn = tmp;
+ numVertsIn = numVertsOut;
+ numVertsOut = 0;
+ }
+
+
+ // only keep points that are behind the witness face
+ {
+ float4 localPlaneNormal = make_float4(polyA.m_plane.x,polyA.m_plane.y,polyA.m_plane.z,0.f);
+ float localPlaneEq = polyA.m_plane.w;
+ float4 planeNormalWS = qtRotate(ornA,localPlaneNormal);
+ float planeEqWS=localPlaneEq-dot3F4(planeNormalWS,posA);
+ for (int i=0;i<numVertsIn;i++)
+ {
+ float depth = dot3F4(planeNormalWS,pVtxIn[i])+planeEqWS;
+ if (depth <=minDist)
+ {
+ depth = minDist;
+ }
+
+ if (depth <=maxDist)
+ {
+ float4 pointInWorld = pVtxIn[i];
+ //resultOut.addContactPoint(separatingNormal,point,depth);
+ contactsOut[numContactsOut++] = make_float4(pointInWorld.x,pointInWorld.y,pointInWorld.z,depth);
+ }
+ }
+ }
+
+ return numContactsOut;
+}
+
+
+
+int clipFaceAgainstHullLocalA(const float4 separatingNormal, const b3ConvexPolyhedronData_t* hullA,
+ const float4 posA, const Quaternion ornA, float4* worldVertsB1, int numWorldVertsB1,
+ float4* worldVertsB2, int capacityWorldVertsB2,
+ const float minDist, float maxDist,
+ const float4* verticesA,
+ const b3GpuFace_t* facesA,
+ const int* indicesA,
+ __global const float4* verticesB,
+ __global const b3GpuFace_t* facesB,
+ __global const int* indicesB,
+ float4* contactsOut,
+ int contactCapacity)
+{
+ int numContactsOut = 0;
+
+ float4* pVtxIn = worldVertsB1;
+ float4* pVtxOut = worldVertsB2;
+
+ int numVertsIn = numWorldVertsB1;
+ int numVertsOut = 0;
+
+ int closestFaceA=-1;
+ {
+ float dmin = FLT_MAX;
+ for(int face=0;face<hullA->m_numFaces;face++)
+ {
+ const float4 Normal = make_float4(
+ facesA[hullA->m_faceOffset+face].m_plane.x,
+ facesA[hullA->m_faceOffset+face].m_plane.y,
+ facesA[hullA->m_faceOffset+face].m_plane.z,0.f);
+ const float4 faceANormalWS = qtRotate(ornA,Normal);
+
+ float d = dot3F4(faceANormalWS,separatingNormal);
+ if (d < dmin)
+ {
+ dmin = d;
+ closestFaceA = face;
+ }
+ }
+ }
+ if (closestFaceA<0)
+ return numContactsOut;
+
+ b3GpuFace_t polyA = facesA[hullA->m_faceOffset+closestFaceA];
+
+ // clip polygon to back of planes of all faces of hull A that are adjacent to witness face
+ int numVerticesA = polyA.m_numIndices;
+ for(int e0=0;e0<numVerticesA;e0++)
+ {
+ const float4 a = verticesA[hullA->m_vertexOffset+indicesA[polyA.m_indexOffset+e0]];
+ const float4 b = verticesA[hullA->m_vertexOffset+indicesA[polyA.m_indexOffset+((e0+1)%numVerticesA)]];
+ const float4 edge0 = a - b;
+ const float4 WorldEdge0 = qtRotate(ornA,edge0);
+ float4 planeNormalA = make_float4(polyA.m_plane.x,polyA.m_plane.y,polyA.m_plane.z,0.f);
+ float4 worldPlaneAnormal1 = qtRotate(ornA,planeNormalA);
+
+ float4 planeNormalWS1 = -cross3(WorldEdge0,worldPlaneAnormal1);
+ float4 worldA1 = transform(&a,&posA,&ornA);
+ float planeEqWS1 = -dot3F4(worldA1,planeNormalWS1);
+
+ float4 planeNormalWS = planeNormalWS1;
+ float planeEqWS=planeEqWS1;
+
+ //clip face
+ //clipFace(*pVtxIn, *pVtxOut,planeNormalWS,planeEqWS);
+ numVertsOut = clipFace(pVtxIn, numVertsIn, planeNormalWS,planeEqWS, pVtxOut);
+
+ //btSwap(pVtxIn,pVtxOut);
+ float4* tmp = pVtxOut;
+ pVtxOut = pVtxIn;
+ pVtxIn = tmp;
+ numVertsIn = numVertsOut;
+ numVertsOut = 0;
+ }
+
+
+ // only keep points that are behind the witness face
+ {
+ float4 localPlaneNormal = make_float4(polyA.m_plane.x,polyA.m_plane.y,polyA.m_plane.z,0.f);
+ float localPlaneEq = polyA.m_plane.w;
+ float4 planeNormalWS = qtRotate(ornA,localPlaneNormal);
+ float planeEqWS=localPlaneEq-dot3F4(planeNormalWS,posA);
+ for (int i=0;i<numVertsIn;i++)
+ {
+ float depth = dot3F4(planeNormalWS,pVtxIn[i])+planeEqWS;
+ if (depth <=minDist)
+ {
+ depth = minDist;
+ }
+
+ if (depth <=maxDist)
+ {
+ float4 pointInWorld = pVtxIn[i];
+ //resultOut.addContactPoint(separatingNormal,point,depth);
+ contactsOut[numContactsOut++] = make_float4(pointInWorld.x,pointInWorld.y,pointInWorld.z,depth);
+ }
+ }
+ }
+
+ return numContactsOut;
+}
+
+int clipHullAgainstHull(const float4 separatingNormal,
+ __global const b3ConvexPolyhedronData_t* hullA, __global const b3ConvexPolyhedronData_t* hullB,
+ const float4 posA, const Quaternion ornA,const float4 posB, const Quaternion ornB,
+ float4* worldVertsB1, float4* worldVertsB2, int capacityWorldVerts,
+ const float minDist, float maxDist,
+ __global const float4* vertices,
+ __global const b3GpuFace_t* faces,
+ __global const int* indices,
+ float4* localContactsOut,
+ int localContactCapacity)
+{
+ int numContactsOut = 0;
+ int numWorldVertsB1= 0;
+
+
+ int closestFaceB=-1;
+ float dmax = -FLT_MAX;
+
+ {
+ for(int face=0;face<hullB->m_numFaces;face++)
+ {
+ const float4 Normal = make_float4(faces[hullB->m_faceOffset+face].m_plane.x,
+ faces[hullB->m_faceOffset+face].m_plane.y, faces[hullB->m_faceOffset+face].m_plane.z,0.f);
+ const float4 WorldNormal = qtRotate(ornB, Normal);
+ float d = dot3F4(WorldNormal,separatingNormal);
+ if (d > dmax)
+ {
+ dmax = d;
+ closestFaceB = face;
+ }
+ }
+ }
+
+ {
+ const b3GpuFace_t polyB = faces[hullB->m_faceOffset+closestFaceB];
+ const int numVertices = polyB.m_numIndices;
+ for(int e0=0;e0<numVertices;e0++)
+ {
+ const float4 b = vertices[hullB->m_vertexOffset+indices[polyB.m_indexOffset+e0]];
+ worldVertsB1[numWorldVertsB1++] = transform(&b,&posB,&ornB);
+ }
+ }
+
+ if (closestFaceB>=0)
+ {
+ numContactsOut = clipFaceAgainstHull(separatingNormal, hullA,
+ posA,ornA,
+ worldVertsB1,numWorldVertsB1,worldVertsB2,capacityWorldVerts, minDist, maxDist,vertices,
+ faces,
+ indices,localContactsOut,localContactCapacity);
+ }
+
+ return numContactsOut;
+}
+
+
+int clipHullAgainstHullLocalA(const float4 separatingNormal,
+ const b3ConvexPolyhedronData_t* hullA, __global const b3ConvexPolyhedronData_t* hullB,
+ const float4 posA, const Quaternion ornA,const float4 posB, const Quaternion ornB,
+ float4* worldVertsB1, float4* worldVertsB2, int capacityWorldVerts,
+ const float minDist, float maxDist,
+ const float4* verticesA,
+ const b3GpuFace_t* facesA,
+ const int* indicesA,
+ __global const float4* verticesB,
+ __global const b3GpuFace_t* facesB,
+ __global const int* indicesB,
+ float4* localContactsOut,
+ int localContactCapacity)
+{
+ int numContactsOut = 0;
+ int numWorldVertsB1= 0;
+
+
+ int closestFaceB=-1;
+ float dmax = -FLT_MAX;
+
+ {
+ for(int face=0;face<hullB->m_numFaces;face++)
+ {
+ const float4 Normal = make_float4(facesB[hullB->m_faceOffset+face].m_plane.x,
+ facesB[hullB->m_faceOffset+face].m_plane.y, facesB[hullB->m_faceOffset+face].m_plane.z,0.f);
+ const float4 WorldNormal = qtRotate(ornB, Normal);
+ float d = dot3F4(WorldNormal,separatingNormal);
+ if (d > dmax)
+ {
+ dmax = d;
+ closestFaceB = face;
+ }
+ }
+ }
+
+ {
+ const b3GpuFace_t polyB = facesB[hullB->m_faceOffset+closestFaceB];
+ const int numVertices = polyB.m_numIndices;
+ for(int e0=0;e0<numVertices;e0++)
+ {
+ const float4 b = verticesB[hullB->m_vertexOffset+indicesB[polyB.m_indexOffset+e0]];
+ worldVertsB1[numWorldVertsB1++] = transform(&b,&posB,&ornB);
+ }
+ }
+
+ if (closestFaceB>=0)
+ {
+ numContactsOut = clipFaceAgainstHullLocalA(separatingNormal, hullA,
+ posA,ornA,
+ worldVertsB1,numWorldVertsB1,worldVertsB2,capacityWorldVerts, minDist, maxDist,
+ verticesA,facesA,indicesA,
+ verticesB,facesB,indicesB,
+ localContactsOut,localContactCapacity);
+ }
+
+ return numContactsOut;
+}
+
+#define PARALLEL_SUM(v, n) for(int j=1; j<n; j++) v[0] += v[j];
+#define PARALLEL_DO(execution, n) for(int ie=0; ie<n; ie++){execution;}
+#define REDUCE_MAX(v, n) {int i=0;\
+for(int offset=0; offset<n; offset++) v[i] = (v[i].y > v[i+offset].y)? v[i]: v[i+offset]; }
+#define REDUCE_MIN(v, n) {int i=0;\
+for(int offset=0; offset<n; offset++) v[i] = (v[i].y < v[i+offset].y)? v[i]: v[i+offset]; }
+
+int extractManifoldSequentialGlobal(__global const float4* p, int nPoints, float4 nearNormal, int4* contactIdx)
+{
+ if( nPoints == 0 )
+ return 0;
+
+ if (nPoints <=4)
+ return nPoints;
+
+
+ if (nPoints >64)
+ nPoints = 64;
+
+ float4 center = make_float4(0.f);
+ {
+
+ for (int i=0;i<nPoints;i++)
+ center += p[i];
+ center /= (float)nPoints;
+ }
+
+
+
+ // sample 4 directions
+
+ float4 aVector = p[0] - center;
+ float4 u = cross3( nearNormal, aVector );
+ float4 v = cross3( nearNormal, u );
+ u = normalize3( u );
+ v = normalize3( v );
+
+
+ //keep point with deepest penetration
+ float minW= FLT_MAX;
+
+ int minIndex=-1;
+
+ float4 maxDots;
+ maxDots.x = FLT_MIN;
+ maxDots.y = FLT_MIN;
+ maxDots.z = FLT_MIN;
+ maxDots.w = FLT_MIN;
+
+ // idx, distance
+ for(int ie = 0; ie<nPoints; ie++ )
+ {
+ if (p[ie].w<minW)
+ {
+ minW = p[ie].w;
+ minIndex=ie;
+ }
+ float f;
+ float4 r = p[ie]-center;
+ f = dot3F4( u, r );
+ if (f<maxDots.x)
+ {
+ maxDots.x = f;
+ contactIdx[0].x = ie;
+ }
+
+ f = dot3F4( -u, r );
+ if (f<maxDots.y)
+ {
+ maxDots.y = f;
+ contactIdx[0].y = ie;
+ }
+
+
+ f = dot3F4( v, r );
+ if (f<maxDots.z)
+ {
+ maxDots.z = f;
+ contactIdx[0].z = ie;
+ }
+
+ f = dot3F4( -v, r );
+ if (f<maxDots.w)
+ {
+ maxDots.w = f;
+ contactIdx[0].w = ie;
+ }
+
+ }
+
+ if (contactIdx[0].x != minIndex && contactIdx[0].y != minIndex && contactIdx[0].z != minIndex && contactIdx[0].w != minIndex)
+ {
+ //replace the first contact with minimum (todo: replace contact with least penetration)
+ contactIdx[0].x = minIndex;
+ }
+
+ return 4;
+
+}
+
+
+int extractManifoldSequentialGlobalFake(__global const float4* p, int nPoints, float4 nearNormal, int* contactIdx)
+{
+ contactIdx[0] = 0;
+ contactIdx[1] = 1;
+ contactIdx[2] = 2;
+ contactIdx[3] = 3;
+
+ if( nPoints == 0 ) return 0;
+
+ nPoints = min2( nPoints, 4 );
+ return nPoints;
+
+}
+
+
+
+int extractManifoldSequential(const float4* p, int nPoints, float4 nearNormal, int* contactIdx)
+{
+ if( nPoints == 0 ) return 0;
+
+ nPoints = min2( nPoints, 64 );
+
+ float4 center = make_float4(0.f);
+ {
+ float4 v[64];
+ for (int i=0;i<nPoints;i++)
+ v[i] = p[i];
+ //memcpy( v, p, nPoints*sizeof(float4) );
+ PARALLEL_SUM( v, nPoints );
+ center = v[0]/(float)nPoints;
+ }
+
+
+
+ { // sample 4 directions
+ if( nPoints < 4 )
+ {
+ for(int i=0; i<nPoints; i++)
+ contactIdx[i] = i;
+ return nPoints;
+ }
+
+ float4 aVector = p[0] - center;
+ float4 u = cross3( nearNormal, aVector );
+ float4 v = cross3( nearNormal, u );
+ u = normalize3( u );
+ v = normalize3( v );
+
+ int idx[4];
+
+ float2 max00 = make_float2(0,FLT_MAX);
+ {
+ // idx, distance
+ {
+ {
+ int4 a[64];
+ for(int ie = 0; ie<nPoints; ie++ )
+ {
+
+
+ float f;
+ float4 r = p[ie]-center;
+ f = dot3F4( u, r );
+ a[ie].x = ((*(u32*)&f) & 0xffffff00) | (0xff & ie);
+
+ f = dot3F4( -u, r );
+ a[ie].y = ((*(u32*)&f) & 0xffffff00) | (0xff & ie);
+
+ f = dot3F4( v, r );
+ a[ie].z = ((*(u32*)&f) & 0xffffff00) | (0xff & ie);
+
+ f = dot3F4( -v, r );
+ a[ie].w = ((*(u32*)&f) & 0xffffff00) | (0xff & ie);
+ }
+
+ for(int ie=0; ie<nPoints; ie++)
+ {
+ a[0].x = (a[0].x > a[ie].x )? a[0].x: a[ie].x;
+ a[0].y = (a[0].y > a[ie].y )? a[0].y: a[ie].y;
+ a[0].z = (a[0].z > a[ie].z )? a[0].z: a[ie].z;
+ a[0].w = (a[0].w > a[ie].w )? a[0].w: a[ie].w;
+ }
+
+ idx[0] = (int)a[0].x & 0xff;
+ idx[1] = (int)a[0].y & 0xff;
+ idx[2] = (int)a[0].z & 0xff;
+ idx[3] = (int)a[0].w & 0xff;
+ }
+ }
+
+ {
+ float2 h[64];
+ PARALLEL_DO( h[ie] = make_float2((float)ie, p[ie].w), nPoints );
+ REDUCE_MIN( h, nPoints );
+ max00 = h[0];
+ }
+ }
+
+ contactIdx[0] = idx[0];
+ contactIdx[1] = idx[1];
+ contactIdx[2] = idx[2];
+ contactIdx[3] = idx[3];
+
+
+ return 4;
+ }
+}
+
+
+
+__kernel void extractManifoldAndAddContactKernel(__global const int4* pairs,
+ __global const b3RigidBodyData_t* rigidBodies,
+ __global const float4* closestPointsWorld,
+ __global const float4* separatingNormalsWorld,
+ __global const int* contactCounts,
+ __global const int* contactOffsets,
+ __global struct b3Contact4Data* restrict contactsOut,
+ counter32_t nContactsOut,
+ int contactCapacity,
+ int numPairs,
+ int pairIndex
+ )
+{
+ int idx = get_global_id(0);
+
+ if (idx<numPairs)
+ {
+ float4 normal = separatingNormalsWorld[idx];
+ int nPoints = contactCounts[idx];
+ __global const float4* pointsIn = &closestPointsWorld[contactOffsets[idx]];
+ float4 localPoints[64];
+ for (int i=0;i<nPoints;i++)
+ {
+ localPoints[i] = pointsIn[i];
+ }
+
+ int contactIdx[4];// = {-1,-1,-1,-1};
+ contactIdx[0] = -1;
+ contactIdx[1] = -1;
+ contactIdx[2] = -1;
+ contactIdx[3] = -1;
+
+ int nContacts = extractManifoldSequential(localPoints, nPoints, normal, contactIdx);
+
+ int dstIdx;
+ AppendInc( nContactsOut, dstIdx );
+ if (dstIdx<contactCapacity)
+ {
+ __global struct b3Contact4Data* c = contactsOut + dstIdx;
+ c->m_worldNormalOnB = -normal;
+ c->m_restituitionCoeffCmp = (0.f*0xffff);c->m_frictionCoeffCmp = (0.7f*0xffff);
+ c->m_batchIdx = idx;
+ int bodyA = pairs[pairIndex].x;
+ int bodyB = pairs[pairIndex].y;
+ c->m_bodyAPtrAndSignBit = rigidBodies[bodyA].m_invMass==0 ? -bodyA:bodyA;
+ c->m_bodyBPtrAndSignBit = rigidBodies[bodyB].m_invMass==0 ? -bodyB:bodyB;
+ c->m_childIndexA = -1;
+ c->m_childIndexB = -1;
+ for (int i=0;i<nContacts;i++)
+ {
+ c->m_worldPosB[i] = localPoints[contactIdx[i]];
+ }
+ GET_NPOINTS(*c) = nContacts;
+ }
+ }
+}
+
+
+void trInverse(float4 translationIn, Quaternion orientationIn,
+ float4* translationOut, Quaternion* orientationOut)
+{
+ *orientationOut = qtInvert(orientationIn);
+ *translationOut = qtRotate(*orientationOut, -translationIn);
+}
+
+void trMul(float4 translationA, Quaternion orientationA,
+ float4 translationB, Quaternion orientationB,
+ float4* translationOut, Quaternion* orientationOut)
+{
+ *orientationOut = qtMul(orientationA,orientationB);
+ *translationOut = transform(&translationB,&translationA,&orientationA);
+}
+
+
+
+
+__kernel void clipHullHullKernel( __global int4* pairs,
+ __global const b3RigidBodyData_t* rigidBodies,
+ __global const b3Collidable_t* collidables,
+ __global const b3ConvexPolyhedronData_t* convexShapes,
+ __global const float4* vertices,
+ __global const float4* uniqueEdges,
+ __global const b3GpuFace_t* faces,
+ __global const int* indices,
+ __global const float4* separatingNormals,
+ __global const int* hasSeparatingAxis,
+ __global struct b3Contact4Data* restrict globalContactsOut,
+ counter32_t nGlobalContactsOut,
+ int numPairs,
+ int contactCapacity)
+{
+
+ int i = get_global_id(0);
+ int pairIndex = i;
+
+ float4 worldVertsB1[64];
+ float4 worldVertsB2[64];
+ int capacityWorldVerts = 64;
+
+ float4 localContactsOut[64];
+ int localContactCapacity=64;
+
+ float minDist = -1e30f;
+ float maxDist = 0.02f;
+
+ if (i<numPairs)
+ {
+
+ int bodyIndexA = pairs[i].x;
+ int bodyIndexB = pairs[i].y;
+
+ int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;
+ int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;
+
+ if (hasSeparatingAxis[i])
+ {
+
+
+ int shapeIndexA = collidables[collidableIndexA].m_shapeIndex;
+ int shapeIndexB = collidables[collidableIndexB].m_shapeIndex;
+
+
+
+
+ int numLocalContactsOut = clipHullAgainstHull(separatingNormals[i],
+ &convexShapes[shapeIndexA], &convexShapes[shapeIndexB],
+ rigidBodies[bodyIndexA].m_pos,rigidBodies[bodyIndexA].m_quat,
+ rigidBodies[bodyIndexB].m_pos,rigidBodies[bodyIndexB].m_quat,
+ worldVertsB1,worldVertsB2,capacityWorldVerts,
+ minDist, maxDist,
+ vertices,faces,indices,
+ localContactsOut,localContactCapacity);
+
+ if (numLocalContactsOut>0)
+ {
+ float4 normal = -separatingNormals[i];
+ int nPoints = numLocalContactsOut;
+ float4* pointsIn = localContactsOut;
+ int contactIdx[4];// = {-1,-1,-1,-1};
+
+ contactIdx[0] = -1;
+ contactIdx[1] = -1;
+ contactIdx[2] = -1;
+ contactIdx[3] = -1;
+
+ int nReducedContacts = extractManifoldSequential(pointsIn, nPoints, normal, contactIdx);
+
+
+ int mprContactIndex = pairs[pairIndex].z;
+
+ int dstIdx = mprContactIndex;
+ if (dstIdx<0)
+ {
+ AppendInc( nGlobalContactsOut, dstIdx );
+ }
+
+ if (dstIdx<contactCapacity)
+ {
+ pairs[pairIndex].z = dstIdx;
+
+ __global struct b3Contact4Data* c = globalContactsOut+ dstIdx;
+ c->m_worldNormalOnB = -normal;
+ c->m_restituitionCoeffCmp = (0.f*0xffff);c->m_frictionCoeffCmp = (0.7f*0xffff);
+ c->m_batchIdx = pairIndex;
+ int bodyA = pairs[pairIndex].x;
+ int bodyB = pairs[pairIndex].y;
+ c->m_bodyAPtrAndSignBit = rigidBodies[bodyA].m_invMass==0?-bodyA:bodyA;
+ c->m_bodyBPtrAndSignBit = rigidBodies[bodyB].m_invMass==0?-bodyB:bodyB;
+ c->m_childIndexA = -1;
+ c->m_childIndexB = -1;
+
+ for (int i=0;i<nReducedContacts;i++)
+ {
+ //this condition means: overwrite contact point, unless at index i==0 we have a valid 'mpr' contact
+ if (i>0||(mprContactIndex<0))
+ {
+ c->m_worldPosB[i] = pointsIn[contactIdx[i]];
+ }
+ }
+ GET_NPOINTS(*c) = nReducedContacts;
+ }
+
+ }// if (numContactsOut>0)
+ }// if (hasSeparatingAxis[i])
+ }// if (i<numPairs)
+
+}
+
+
+__kernel void clipCompoundsHullHullKernel( __global const int4* gpuCompoundPairs,
+ __global const b3RigidBodyData_t* rigidBodies,
+ __global const b3Collidable_t* collidables,
+ __global const b3ConvexPolyhedronData_t* convexShapes,
+ __global const float4* vertices,
+ __global const float4* uniqueEdges,
+ __global const b3GpuFace_t* faces,
+ __global const int* indices,
+ __global const b3GpuChildShape_t* gpuChildShapes,
+ __global const float4* gpuCompoundSepNormalsOut,
+ __global const int* gpuHasCompoundSepNormalsOut,
+ __global struct b3Contact4Data* restrict globalContactsOut,
+ counter32_t nGlobalContactsOut,
+ int numCompoundPairs, int maxContactCapacity)
+{
+
+ int i = get_global_id(0);
+ int pairIndex = i;
+
+ float4 worldVertsB1[64];
+ float4 worldVertsB2[64];
+ int capacityWorldVerts = 64;
+
+ float4 localContactsOut[64];
+ int localContactCapacity=64;
+
+ float minDist = -1e30f;
+ float maxDist = 0.02f;
+
+ if (i<numCompoundPairs)
+ {
+
+ if (gpuHasCompoundSepNormalsOut[i])
+ {
+
+ int bodyIndexA = gpuCompoundPairs[i].x;
+ int bodyIndexB = gpuCompoundPairs[i].y;
+
+ int childShapeIndexA = gpuCompoundPairs[i].z;
+ int childShapeIndexB = gpuCompoundPairs[i].w;
+
+ int collidableIndexA = -1;
+ int collidableIndexB = -1;
+
+ float4 ornA = rigidBodies[bodyIndexA].m_quat;
+ float4 posA = rigidBodies[bodyIndexA].m_pos;
+
+ float4 ornB = rigidBodies[bodyIndexB].m_quat;
+ float4 posB = rigidBodies[bodyIndexB].m_pos;
+
+ if (childShapeIndexA >= 0)
+ {
+ collidableIndexA = gpuChildShapes[childShapeIndexA].m_shapeIndex;
+ float4 childPosA = gpuChildShapes[childShapeIndexA].m_childPosition;
+ float4 childOrnA = gpuChildShapes[childShapeIndexA].m_childOrientation;
+ float4 newPosA = qtRotate(ornA,childPosA)+posA;
+ float4 newOrnA = qtMul(ornA,childOrnA);
+ posA = newPosA;
+ ornA = newOrnA;
+ } else
+ {
+ collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;
+ }
+
+ if (childShapeIndexB>=0)
+ {
+ collidableIndexB = gpuChildShapes[childShapeIndexB].m_shapeIndex;
+ float4 childPosB = gpuChildShapes[childShapeIndexB].m_childPosition;
+ float4 childOrnB = gpuChildShapes[childShapeIndexB].m_childOrientation;
+ float4 newPosB = transform(&childPosB,&posB,&ornB);
+ float4 newOrnB = qtMul(ornB,childOrnB);
+ posB = newPosB;
+ ornB = newOrnB;
+ } else
+ {
+ collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;
+ }
+
+ int shapeIndexA = collidables[collidableIndexA].m_shapeIndex;
+ int shapeIndexB = collidables[collidableIndexB].m_shapeIndex;
+
+ int numLocalContactsOut = clipHullAgainstHull(gpuCompoundSepNormalsOut[i],
+ &convexShapes[shapeIndexA], &convexShapes[shapeIndexB],
+ posA,ornA,
+ posB,ornB,
+ worldVertsB1,worldVertsB2,capacityWorldVerts,
+ minDist, maxDist,
+ vertices,faces,indices,
+ localContactsOut,localContactCapacity);
+
+ if (numLocalContactsOut>0)
+ {
+ float4 normal = -gpuCompoundSepNormalsOut[i];
+ int nPoints = numLocalContactsOut;
+ float4* pointsIn = localContactsOut;
+ int contactIdx[4];// = {-1,-1,-1,-1};
+
+ contactIdx[0] = -1;
+ contactIdx[1] = -1;
+ contactIdx[2] = -1;
+ contactIdx[3] = -1;
+
+ int nReducedContacts = extractManifoldSequential(pointsIn, nPoints, normal, contactIdx);
+
+ int dstIdx;
+ AppendInc( nGlobalContactsOut, dstIdx );
+ if ((dstIdx+nReducedContacts) < maxContactCapacity)
+ {
+ __global struct b3Contact4Data* c = globalContactsOut+ dstIdx;
+ c->m_worldNormalOnB = -normal;
+ c->m_restituitionCoeffCmp = (0.f*0xffff);c->m_frictionCoeffCmp = (0.7f*0xffff);
+ c->m_batchIdx = pairIndex;
+ int bodyA = gpuCompoundPairs[pairIndex].x;
+ int bodyB = gpuCompoundPairs[pairIndex].y;
+ c->m_bodyAPtrAndSignBit = rigidBodies[bodyA].m_invMass==0?-bodyA:bodyA;
+ c->m_bodyBPtrAndSignBit = rigidBodies[bodyB].m_invMass==0?-bodyB:bodyB;
+ c->m_childIndexA = childShapeIndexA;
+ c->m_childIndexB = childShapeIndexB;
+ for (int i=0;i<nReducedContacts;i++)
+ {
+ c->m_worldPosB[i] = pointsIn[contactIdx[i]];
+ }
+ GET_NPOINTS(*c) = nReducedContacts;
+ }
+
+ }// if (numContactsOut>0)
+ }// if (gpuHasCompoundSepNormalsOut[i])
+ }// if (i<numCompoundPairs)
+
+}
+
+
+
+__kernel void sphereSphereCollisionKernel( __global const int4* pairs,
+ __global const b3RigidBodyData_t* rigidBodies,
+ __global const b3Collidable_t* collidables,
+ __global const float4* separatingNormals,
+ __global const int* hasSeparatingAxis,
+ __global struct b3Contact4Data* restrict globalContactsOut,
+ counter32_t nGlobalContactsOut,
+ int contactCapacity,
+ int numPairs)
+{
+
+ int i = get_global_id(0);
+ int pairIndex = i;
+
+ if (i<numPairs)
+ {
+ int bodyIndexA = pairs[i].x;
+ int bodyIndexB = pairs[i].y;
+
+ int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;
+ int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;
+
+ if (collidables[collidableIndexA].m_shapeType == SHAPE_SPHERE &&
+ collidables[collidableIndexB].m_shapeType == SHAPE_SPHERE)
+ {
+ //sphere-sphere
+ float radiusA = collidables[collidableIndexA].m_radius;
+ float radiusB = collidables[collidableIndexB].m_radius;
+ float4 posA = rigidBodies[bodyIndexA].m_pos;
+ float4 posB = rigidBodies[bodyIndexB].m_pos;
+
+ float4 diff = posA-posB;
+ float len = length(diff);
+
+ ///iff distance positive, don't generate a new contact
+ if ( len <= (radiusA+radiusB))
+ {
+ ///distance (negative means penetration)
+ float dist = len - (radiusA+radiusB);
+ float4 normalOnSurfaceB = make_float4(1.f,0.f,0.f,0.f);
+ if (len > 0.00001)
+ {
+ normalOnSurfaceB = diff / len;
+ }
+ float4 contactPosB = posB + normalOnSurfaceB*radiusB;
+ contactPosB.w = dist;
+
+ int dstIdx;
+ AppendInc( nGlobalContactsOut, dstIdx );
+ if (dstIdx < contactCapacity)
+ {
+ __global struct b3Contact4Data* c = &globalContactsOut[dstIdx];
+ c->m_worldNormalOnB = -normalOnSurfaceB;
+ c->m_restituitionCoeffCmp = (0.f*0xffff);c->m_frictionCoeffCmp = (0.7f*0xffff);
+ c->m_batchIdx = pairIndex;
+ int bodyA = pairs[pairIndex].x;
+ int bodyB = pairs[pairIndex].y;
+ c->m_bodyAPtrAndSignBit = rigidBodies[bodyA].m_invMass==0?-bodyA:bodyA;
+ c->m_bodyBPtrAndSignBit = rigidBodies[bodyB].m_invMass==0?-bodyB:bodyB;
+ c->m_worldPosB[0] = contactPosB;
+ c->m_childIndexA = -1;
+ c->m_childIndexB = -1;
+
+ GET_NPOINTS(*c) = 1;
+ }//if (dstIdx < numPairs)
+ }//if ( len <= (radiusA+radiusB))
+ }//SHAPE_SPHERE SHAPE_SPHERE
+ }//if (i<numPairs)
+}
+
+__kernel void clipHullHullConcaveConvexKernel( __global int4* concavePairsIn,
+ __global const b3RigidBodyData_t* rigidBodies,
+ __global const b3Collidable_t* collidables,
+ __global const b3ConvexPolyhedronData_t* convexShapes,
+ __global const float4* vertices,
+ __global const float4* uniqueEdges,
+ __global const b3GpuFace_t* faces,
+ __global const int* indices,
+ __global const b3GpuChildShape_t* gpuChildShapes,
+ __global const float4* separatingNormals,
+ __global struct b3Contact4Data* restrict globalContactsOut,
+ counter32_t nGlobalContactsOut,
+ int contactCapacity,
+ int numConcavePairs)
+{
+
+ int i = get_global_id(0);
+ int pairIndex = i;
+
+ float4 worldVertsB1[64];
+ float4 worldVertsB2[64];
+ int capacityWorldVerts = 64;
+
+ float4 localContactsOut[64];
+ int localContactCapacity=64;
+
+ float minDist = -1e30f;
+ float maxDist = 0.02f;
+
+ if (i<numConcavePairs)
+ {
+ //negative value means that the pair is invalid
+ if (concavePairsIn[i].w<0)
+ return;
+
+ int bodyIndexA = concavePairsIn[i].x;
+ int bodyIndexB = concavePairsIn[i].y;
+ int f = concavePairsIn[i].z;
+ int childShapeIndexA = f;
+
+ int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;
+ int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;
+
+ int shapeIndexA = collidables[collidableIndexA].m_shapeIndex;
+ int shapeIndexB = collidables[collidableIndexB].m_shapeIndex;
+
+ ///////////////////////////////////////////////////////////////
+
+
+ bool overlap = false;
+
+ b3ConvexPolyhedronData_t convexPolyhedronA;
+
+ //add 3 vertices of the triangle
+ convexPolyhedronA.m_numVertices = 3;
+ convexPolyhedronA.m_vertexOffset = 0;
+ float4 localCenter = make_float4(0.f,0.f,0.f,0.f);
+
+ b3GpuFace_t face = faces[convexShapes[shapeIndexA].m_faceOffset+f];
+
+ float4 verticesA[3];
+ for (int i=0;i<3;i++)
+ {
+ int index = indices[face.m_indexOffset+i];
+ float4 vert = vertices[convexShapes[shapeIndexA].m_vertexOffset+index];
+ verticesA[i] = vert;
+ localCenter += vert;
+ }
+
+ float dmin = FLT_MAX;
+
+ int localCC=0;
+
+ //a triangle has 3 unique edges
+ convexPolyhedronA.m_numUniqueEdges = 3;
+ convexPolyhedronA.m_uniqueEdgesOffset = 0;
+ float4 uniqueEdgesA[3];
+
+ uniqueEdgesA[0] = (verticesA[1]-verticesA[0]);
+ uniqueEdgesA[1] = (verticesA[2]-verticesA[1]);
+ uniqueEdgesA[2] = (verticesA[0]-verticesA[2]);
+
+
+ convexPolyhedronA.m_faceOffset = 0;
+
+ float4 normal = make_float4(face.m_plane.x,face.m_plane.y,face.m_plane.z,0.f);
+
+ b3GpuFace_t facesA[TRIANGLE_NUM_CONVEX_FACES];
+ int indicesA[3+3+2+2+2];
+ int curUsedIndices=0;
+ int fidx=0;
+
+ //front size of triangle
+ {
+ facesA[fidx].m_indexOffset=curUsedIndices;
+ indicesA[0] = 0;
+ indicesA[1] = 1;
+ indicesA[2] = 2;
+ curUsedIndices+=3;
+ float c = face.m_plane.w;
+ facesA[fidx].m_plane.x = normal.x;
+ facesA[fidx].m_plane.y = normal.y;
+ facesA[fidx].m_plane.z = normal.z;
+ facesA[fidx].m_plane.w = c;
+ facesA[fidx].m_numIndices=3;
+ }
+ fidx++;
+ //back size of triangle
+ {
+ facesA[fidx].m_indexOffset=curUsedIndices;
+ indicesA[3]=2;
+ indicesA[4]=1;
+ indicesA[5]=0;
+ curUsedIndices+=3;
+ float c = dot3F4(normal,verticesA[0]);
+ float c1 = -face.m_plane.w;
+ facesA[fidx].m_plane.x = -normal.x;
+ facesA[fidx].m_plane.y = -normal.y;
+ facesA[fidx].m_plane.z = -normal.z;
+ facesA[fidx].m_plane.w = c;
+ facesA[fidx].m_numIndices=3;
+ }
+ fidx++;
+
+ bool addEdgePlanes = true;
+ if (addEdgePlanes)
+ {
+ int numVertices=3;
+ int prevVertex = numVertices-1;
+ for (int i=0;i<numVertices;i++)
+ {
+ float4 v0 = verticesA[i];
+ float4 v1 = verticesA[prevVertex];
+
+ float4 edgeNormal = normalize(cross(normal,v1-v0));
+ float c = -dot3F4(edgeNormal,v0);
+
+ facesA[fidx].m_numIndices = 2;
+ facesA[fidx].m_indexOffset=curUsedIndices;
+ indicesA[curUsedIndices++]=i;
+ indicesA[curUsedIndices++]=prevVertex;
+
+ facesA[fidx].m_plane.x = edgeNormal.x;
+ facesA[fidx].m_plane.y = edgeNormal.y;
+ facesA[fidx].m_plane.z = edgeNormal.z;
+ facesA[fidx].m_plane.w = c;
+ fidx++;
+ prevVertex = i;
+ }
+ }
+ convexPolyhedronA.m_numFaces = TRIANGLE_NUM_CONVEX_FACES;
+ convexPolyhedronA.m_localCenter = localCenter*(1.f/3.f);
+
+
+ float4 posA = rigidBodies[bodyIndexA].m_pos;
+ posA.w = 0.f;
+ float4 posB = rigidBodies[bodyIndexB].m_pos;
+ posB.w = 0.f;
+ float4 ornA = rigidBodies[bodyIndexA].m_quat;
+ float4 ornB =rigidBodies[bodyIndexB].m_quat;
+
+
+ float4 sepAxis = separatingNormals[i];
+
+ int shapeTypeB = collidables[collidableIndexB].m_shapeType;
+ int childShapeIndexB =-1;
+ if (shapeTypeB==SHAPE_COMPOUND_OF_CONVEX_HULLS)
+ {
+ ///////////////////
+ ///compound shape support
+
+ childShapeIndexB = concavePairsIn[pairIndex].w;
+ int childColIndexB = gpuChildShapes[childShapeIndexB].m_shapeIndex;
+ shapeIndexB = collidables[childColIndexB].m_shapeIndex;
+ float4 childPosB = gpuChildShapes[childShapeIndexB].m_childPosition;
+ float4 childOrnB = gpuChildShapes[childShapeIndexB].m_childOrientation;
+ float4 newPosB = transform(&childPosB,&posB,&ornB);
+ float4 newOrnB = qtMul(ornB,childOrnB);
+ posB = newPosB;
+ ornB = newOrnB;
+
+ }
+
+ ////////////////////////////////////////
+
+
+
+ int numLocalContactsOut = clipHullAgainstHullLocalA(sepAxis,
+ &convexPolyhedronA, &convexShapes[shapeIndexB],
+ posA,ornA,
+ posB,ornB,
+ worldVertsB1,worldVertsB2,capacityWorldVerts,
+ minDist, maxDist,
+ &verticesA,&facesA,&indicesA,
+ vertices,faces,indices,
+ localContactsOut,localContactCapacity);
+
+ if (numLocalContactsOut>0)
+ {
+ float4 normal = -separatingNormals[i];
+ int nPoints = numLocalContactsOut;
+ float4* pointsIn = localContactsOut;
+ int contactIdx[4];// = {-1,-1,-1,-1};
+
+ contactIdx[0] = -1;
+ contactIdx[1] = -1;
+ contactIdx[2] = -1;
+ contactIdx[3] = -1;
+
+ int nReducedContacts = extractManifoldSequential(pointsIn, nPoints, normal, contactIdx);
+
+ int dstIdx;
+ AppendInc( nGlobalContactsOut, dstIdx );
+ if (dstIdx<contactCapacity)
+ {
+ __global struct b3Contact4Data* c = globalContactsOut+ dstIdx;
+ c->m_worldNormalOnB = -normal;
+ c->m_restituitionCoeffCmp = (0.f*0xffff);c->m_frictionCoeffCmp = (0.7f*0xffff);
+ c->m_batchIdx = pairIndex;
+ int bodyA = concavePairsIn[pairIndex].x;
+ int bodyB = concavePairsIn[pairIndex].y;
+ c->m_bodyAPtrAndSignBit = rigidBodies[bodyA].m_invMass==0?-bodyA:bodyA;
+ c->m_bodyBPtrAndSignBit = rigidBodies[bodyB].m_invMass==0?-bodyB:bodyB;
+ c->m_childIndexA = childShapeIndexA;
+ c->m_childIndexB = childShapeIndexB;
+ for (int i=0;i<nReducedContacts;i++)
+ {
+ c->m_worldPosB[i] = pointsIn[contactIdx[i]];
+ }
+ GET_NPOINTS(*c) = nReducedContacts;
+ }
+
+ }// if (numContactsOut>0)
+ }// if (i<numPairs)
+}
+
+
+
+
+
+
+int findClippingFaces(const float4 separatingNormal,
+ __global const b3ConvexPolyhedronData_t* hullA, __global const b3ConvexPolyhedronData_t* hullB,
+ const float4 posA, const Quaternion ornA,const float4 posB, const Quaternion ornB,
+ __global float4* worldVertsA1,
+ __global float4* worldNormalsA1,
+ __global float4* worldVertsB1,
+ int capacityWorldVerts,
+ const float minDist, float maxDist,
+ __global const float4* vertices,
+ __global const b3GpuFace_t* faces,
+ __global const int* indices,
+ __global int4* clippingFaces, int pairIndex)
+{
+ int numContactsOut = 0;
+ int numWorldVertsB1= 0;
+
+
+ int closestFaceB=-1;
+ float dmax = -FLT_MAX;
+
+ {
+ for(int face=0;face<hullB->m_numFaces;face++)
+ {
+ const float4 Normal = make_float4(faces[hullB->m_faceOffset+face].m_plane.x,
+ faces[hullB->m_faceOffset+face].m_plane.y, faces[hullB->m_faceOffset+face].m_plane.z,0.f);
+ const float4 WorldNormal = qtRotate(ornB, Normal);
+ float d = dot3F4(WorldNormal,separatingNormal);
+ if (d > dmax)
+ {
+ dmax = d;
+ closestFaceB = face;
+ }
+ }
+ }
+
+ {
+ const b3GpuFace_t polyB = faces[hullB->m_faceOffset+closestFaceB];
+ const int numVertices = polyB.m_numIndices;
+ for(int e0=0;e0<numVertices;e0++)
+ {
+ const float4 b = vertices[hullB->m_vertexOffset+indices[polyB.m_indexOffset+e0]];
+ worldVertsB1[pairIndex*capacityWorldVerts+numWorldVertsB1++] = transform(&b,&posB,&ornB);
+ }
+ }
+
+ int closestFaceA=-1;
+ {
+ float dmin = FLT_MAX;
+ for(int face=0;face<hullA->m_numFaces;face++)
+ {
+ const float4 Normal = make_float4(
+ faces[hullA->m_faceOffset+face].m_plane.x,
+ faces[hullA->m_faceOffset+face].m_plane.y,
+ faces[hullA->m_faceOffset+face].m_plane.z,
+ 0.f);
+ const float4 faceANormalWS = qtRotate(ornA,Normal);
+
+ float d = dot3F4(faceANormalWS,separatingNormal);
+ if (d < dmin)
+ {
+ dmin = d;
+ closestFaceA = face;
+ worldNormalsA1[pairIndex] = faceANormalWS;
+ }
+ }
+ }
+
+ int numVerticesA = faces[hullA->m_faceOffset+closestFaceA].m_numIndices;
+ for(int e0=0;e0<numVerticesA;e0++)
+ {
+ const float4 a = vertices[hullA->m_vertexOffset+indices[faces[hullA->m_faceOffset+closestFaceA].m_indexOffset+e0]];
+ worldVertsA1[pairIndex*capacityWorldVerts+e0] = transform(&a, &posA,&ornA);
+ }
+
+ clippingFaces[pairIndex].x = closestFaceA;
+ clippingFaces[pairIndex].y = closestFaceB;
+ clippingFaces[pairIndex].z = numVerticesA;
+ clippingFaces[pairIndex].w = numWorldVertsB1;
+
+
+ return numContactsOut;
+}
+
+
+
+int clipFaces(__global float4* worldVertsA1,
+ __global float4* worldNormalsA1,
+ __global float4* worldVertsB1,
+ __global float4* worldVertsB2,
+ int capacityWorldVertsB2,
+ const float minDist, float maxDist,
+ __global int4* clippingFaces,
+ int pairIndex)
+{
+ int numContactsOut = 0;
+
+ int closestFaceA = clippingFaces[pairIndex].x;
+ int closestFaceB = clippingFaces[pairIndex].y;
+ int numVertsInA = clippingFaces[pairIndex].z;
+ int numVertsInB = clippingFaces[pairIndex].w;
+
+ int numVertsOut = 0;
+
+ if (closestFaceA<0)
+ return numContactsOut;
+
+ __global float4* pVtxIn = &worldVertsB1[pairIndex*capacityWorldVertsB2];
+ __global float4* pVtxOut = &worldVertsB2[pairIndex*capacityWorldVertsB2];
+
+
+
+ // clip polygon to back of planes of all faces of hull A that are adjacent to witness face
+
+ for(int e0=0;e0<numVertsInA;e0++)
+ {
+ const float4 aw = worldVertsA1[pairIndex*capacityWorldVertsB2+e0];
+ const float4 bw = worldVertsA1[pairIndex*capacityWorldVertsB2+((e0+1)%numVertsInA)];
+ const float4 WorldEdge0 = aw - bw;
+ float4 worldPlaneAnormal1 = worldNormalsA1[pairIndex];
+ float4 planeNormalWS1 = -cross3(WorldEdge0,worldPlaneAnormal1);
+ float4 worldA1 = aw;
+ float planeEqWS1 = -dot3F4(worldA1,planeNormalWS1);
+ float4 planeNormalWS = planeNormalWS1;
+ float planeEqWS=planeEqWS1;
+ numVertsOut = clipFaceGlobal(pVtxIn, numVertsInB, planeNormalWS,planeEqWS, pVtxOut);
+ __global float4* tmp = pVtxOut;
+ pVtxOut = pVtxIn;
+ pVtxIn = tmp;
+ numVertsInB = numVertsOut;
+ numVertsOut = 0;
+ }
+
+ //float4 planeNormalWS = worldNormalsA1[pairIndex];
+ //float planeEqWS=-dot3F4(planeNormalWS,worldVertsA1[pairIndex*capacityWorldVertsB2]);
+
+
+
+ /*for (int i=0;i<numVertsInB;i++)
+ {
+ pVtxOut[i] = pVtxIn[i];
+ }*/
+
+
+
+
+ //numVertsInB=0;
+
+ float4 planeNormalWS = worldNormalsA1[pairIndex];
+ float planeEqWS=-dot3F4(planeNormalWS,worldVertsA1[pairIndex*capacityWorldVertsB2]);
+
+ for (int i=0;i<numVertsInB;i++)
+ {
+ float depth = dot3F4(planeNormalWS,pVtxIn[i])+planeEqWS;
+ if (depth <=minDist)
+ {
+ depth = minDist;
+ }
+
+ if (depth <=maxDist)
+ {
+ float4 pointInWorld = pVtxIn[i];
+ pVtxOut[numContactsOut++] = make_float4(pointInWorld.x,pointInWorld.y,pointInWorld.z,depth);
+ }
+ }
+
+ clippingFaces[pairIndex].w =numContactsOut;
+
+
+ return numContactsOut;
+
+}
+
+
+
+
+__kernel void findClippingFacesKernel( __global const int4* pairs,
+ __global const b3RigidBodyData_t* rigidBodies,
+ __global const b3Collidable_t* collidables,
+ __global const b3ConvexPolyhedronData_t* convexShapes,
+ __global const float4* vertices,
+ __global const float4* uniqueEdges,
+ __global const b3GpuFace_t* faces,
+ __global const int* indices,
+ __global const float4* separatingNormals,
+ __global const int* hasSeparatingAxis,
+ __global int4* clippingFacesOut,
+ __global float4* worldVertsA1,
+ __global float4* worldNormalsA1,
+ __global float4* worldVertsB1,
+ int capacityWorldVerts,
+ int numPairs
+ )
+{
+
+ int i = get_global_id(0);
+ int pairIndex = i;
+
+
+ float minDist = -1e30f;
+ float maxDist = 0.02f;
+
+ if (i<numPairs)
+ {
+
+ if (hasSeparatingAxis[i])
+ {
+
+ int bodyIndexA = pairs[i].x;
+ int bodyIndexB = pairs[i].y;
+
+ int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;
+ int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;
+
+ int shapeIndexA = collidables[collidableIndexA].m_shapeIndex;
+ int shapeIndexB = collidables[collidableIndexB].m_shapeIndex;
+
+
+
+ int numLocalContactsOut = findClippingFaces(separatingNormals[i],
+ &convexShapes[shapeIndexA], &convexShapes[shapeIndexB],
+ rigidBodies[bodyIndexA].m_pos,rigidBodies[bodyIndexA].m_quat,
+ rigidBodies[bodyIndexB].m_pos,rigidBodies[bodyIndexB].m_quat,
+ worldVertsA1,
+ worldNormalsA1,
+ worldVertsB1,capacityWorldVerts,
+ minDist, maxDist,
+ vertices,faces,indices,
+ clippingFacesOut,i);
+
+
+ }// if (hasSeparatingAxis[i])
+ }// if (i<numPairs)
+
+}
+
+
+
+
+__kernel void clipFacesAndFindContactsKernel( __global const float4* separatingNormals,
+ __global const int* hasSeparatingAxis,
+ __global int4* clippingFacesOut,
+ __global float4* worldVertsA1,
+ __global float4* worldNormalsA1,
+ __global float4* worldVertsB1,
+ __global float4* worldVertsB2,
+ int vertexFaceCapacity,
+ int numPairs,
+ int debugMode
+ )
+{
+ int i = get_global_id(0);
+ int pairIndex = i;
+
+
+ float minDist = -1e30f;
+ float maxDist = 0.02f;
+
+ if (i<numPairs)
+ {
+
+ if (hasSeparatingAxis[i])
+ {
+
+// int bodyIndexA = pairs[i].x;
+ // int bodyIndexB = pairs[i].y;
+
+ int numLocalContactsOut = 0;
+
+ int capacityWorldVertsB2 = vertexFaceCapacity;
+
+ __global float4* pVtxIn = &worldVertsB1[pairIndex*capacityWorldVertsB2];
+ __global float4* pVtxOut = &worldVertsB2[pairIndex*capacityWorldVertsB2];
+
+
+ {
+ __global int4* clippingFaces = clippingFacesOut;
+
+
+ int closestFaceA = clippingFaces[pairIndex].x;
+ int closestFaceB = clippingFaces[pairIndex].y;
+ int numVertsInA = clippingFaces[pairIndex].z;
+ int numVertsInB = clippingFaces[pairIndex].w;
+
+ int numVertsOut = 0;
+
+ if (closestFaceA>=0)
+ {
+
+
+
+ // clip polygon to back of planes of all faces of hull A that are adjacent to witness face
+
+ for(int e0=0;e0<numVertsInA;e0++)
+ {
+ const float4 aw = worldVertsA1[pairIndex*capacityWorldVertsB2+e0];
+ const float4 bw = worldVertsA1[pairIndex*capacityWorldVertsB2+((e0+1)%numVertsInA)];
+ const float4 WorldEdge0 = aw - bw;
+ float4 worldPlaneAnormal1 = worldNormalsA1[pairIndex];
+ float4 planeNormalWS1 = -cross3(WorldEdge0,worldPlaneAnormal1);
+ float4 worldA1 = aw;
+ float planeEqWS1 = -dot3F4(worldA1,planeNormalWS1);
+ float4 planeNormalWS = planeNormalWS1;
+ float planeEqWS=planeEqWS1;
+ numVertsOut = clipFaceGlobal(pVtxIn, numVertsInB, planeNormalWS,planeEqWS, pVtxOut);
+ __global float4* tmp = pVtxOut;
+ pVtxOut = pVtxIn;
+ pVtxIn = tmp;
+ numVertsInB = numVertsOut;
+ numVertsOut = 0;
+ }
+
+ float4 planeNormalWS = worldNormalsA1[pairIndex];
+ float planeEqWS=-dot3F4(planeNormalWS,worldVertsA1[pairIndex*capacityWorldVertsB2]);
+
+ for (int i=0;i<numVertsInB;i++)
+ {
+ float depth = dot3F4(planeNormalWS,pVtxIn[i])+planeEqWS;
+ if (depth <=minDist)
+ {
+ depth = minDist;
+ }
+
+ if (depth <=maxDist)
+ {
+ float4 pointInWorld = pVtxIn[i];
+ pVtxOut[numLocalContactsOut++] = make_float4(pointInWorld.x,pointInWorld.y,pointInWorld.z,depth);
+ }
+ }
+
+ }
+ clippingFaces[pairIndex].w =numLocalContactsOut;
+
+
+ }
+
+ for (int i=0;i<numLocalContactsOut;i++)
+ pVtxIn[i] = pVtxOut[i];
+
+ }// if (hasSeparatingAxis[i])
+ }// if (i<numPairs)
+
+}
+
+
+
+
+
+__kernel void newContactReductionKernel( __global int4* pairs,
+ __global const b3RigidBodyData_t* rigidBodies,
+ __global const float4* separatingNormals,
+ __global const int* hasSeparatingAxis,
+ __global struct b3Contact4Data* globalContactsOut,
+ __global int4* clippingFaces,
+ __global float4* worldVertsB2,
+ volatile __global int* nGlobalContactsOut,
+ int vertexFaceCapacity,
+ int contactCapacity,
+ int numPairs
+ )
+{
+ int i = get_global_id(0);
+ int pairIndex = i;
+
+ int4 contactIdx;
+ contactIdx=make_int4(0,1,2,3);
+
+ if (i<numPairs)
+ {
+
+ if (hasSeparatingAxis[i])
+ {
+
+
+
+
+ int nPoints = clippingFaces[pairIndex].w;
+
+ if (nPoints>0)
+ {
+
+ __global float4* pointsIn = &worldVertsB2[pairIndex*vertexFaceCapacity];
+ float4 normal = -separatingNormals[i];
+
+ int nReducedContacts = extractManifoldSequentialGlobal(pointsIn, nPoints, normal, &contactIdx);
+
+ int mprContactIndex = pairs[pairIndex].z;
+
+ int dstIdx = mprContactIndex;
+
+ if (dstIdx<0)
+ {
+ AppendInc( nGlobalContactsOut, dstIdx );
+ }
+//#if 0
+
+ if (dstIdx < contactCapacity)
+ {
+
+ __global struct b3Contact4Data* c = &globalContactsOut[dstIdx];
+ c->m_worldNormalOnB = -normal;
+ c->m_restituitionCoeffCmp = (0.f*0xffff);c->m_frictionCoeffCmp = (0.7f*0xffff);
+ c->m_batchIdx = pairIndex;
+ int bodyA = pairs[pairIndex].x;
+ int bodyB = pairs[pairIndex].y;
+
+ pairs[pairIndex].w = dstIdx;
+
+ c->m_bodyAPtrAndSignBit = rigidBodies[bodyA].m_invMass==0?-bodyA:bodyA;
+ c->m_bodyBPtrAndSignBit = rigidBodies[bodyB].m_invMass==0?-bodyB:bodyB;
+ c->m_childIndexA =-1;
+ c->m_childIndexB =-1;
+
+ switch (nReducedContacts)
+ {
+ case 4:
+ c->m_worldPosB[3] = pointsIn[contactIdx.w];
+ case 3:
+ c->m_worldPosB[2] = pointsIn[contactIdx.z];
+ case 2:
+ c->m_worldPosB[1] = pointsIn[contactIdx.y];
+ case 1:
+ if (mprContactIndex<0)//test
+ c->m_worldPosB[0] = pointsIn[contactIdx.x];
+ default:
+ {
+ }
+ };
+
+ GET_NPOINTS(*c) = nReducedContacts;
+
+ }
+
+
+//#endif
+
+ }// if (numContactsOut>0)
+ }// if (hasSeparatingAxis[i])
+ }// if (i<numPairs)
+
+
+
+}
diff --git a/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/satClipHullContacts.h b/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/satClipHullContacts.h
new file mode 100644
index 0000000000..f0ecfc7851
--- /dev/null
+++ b/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/satClipHullContacts.h
@@ -0,0 +1,2099 @@
+//this file is autogenerated using stringify.bat (premake --stringify) in the build folder of this project
+static const char* satClipKernelsCL= \
+"#define TRIANGLE_NUM_CONVEX_FACES 5\n"
+"#pragma OPENCL EXTENSION cl_amd_printf : enable\n"
+"#pragma OPENCL EXTENSION cl_khr_local_int32_base_atomics : enable\n"
+"#pragma OPENCL EXTENSION cl_khr_global_int32_base_atomics : enable\n"
+"#pragma OPENCL EXTENSION cl_khr_local_int32_extended_atomics : enable\n"
+"#pragma OPENCL EXTENSION cl_khr_global_int32_extended_atomics : enable\n"
+"#ifdef cl_ext_atomic_counters_32\n"
+"#pragma OPENCL EXTENSION cl_ext_atomic_counters_32 : enable\n"
+"#else\n"
+"#define counter32_t volatile __global int*\n"
+"#endif\n"
+"#define GET_GROUP_IDX get_group_id(0)\n"
+"#define GET_LOCAL_IDX get_local_id(0)\n"
+"#define GET_GLOBAL_IDX get_global_id(0)\n"
+"#define GET_GROUP_SIZE get_local_size(0)\n"
+"#define GET_NUM_GROUPS get_num_groups(0)\n"
+"#define GROUP_LDS_BARRIER barrier(CLK_LOCAL_MEM_FENCE)\n"
+"#define GROUP_MEM_FENCE mem_fence(CLK_LOCAL_MEM_FENCE)\n"
+"#define AtomInc(x) atom_inc(&(x))\n"
+"#define AtomInc1(x, out) out = atom_inc(&(x))\n"
+"#define AppendInc(x, out) out = atomic_inc(x)\n"
+"#define AtomAdd(x, value) atom_add(&(x), value)\n"
+"#define AtomCmpxhg(x, cmp, value) atom_cmpxchg( &(x), cmp, value )\n"
+"#define AtomXhg(x, value) atom_xchg ( &(x), value )\n"
+"#define max2 max\n"
+"#define min2 min\n"
+"typedef unsigned int u32;\n"
+"#ifndef B3_CONTACT4DATA_H\n"
+"#define B3_CONTACT4DATA_H\n"
+"#ifndef B3_FLOAT4_H\n"
+"#define B3_FLOAT4_H\n"
+"#ifndef B3_PLATFORM_DEFINITIONS_H\n"
+"#define B3_PLATFORM_DEFINITIONS_H\n"
+"struct MyTest\n"
+"{\n"
+" int bla;\n"
+"};\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+"//keep B3_LARGE_FLOAT*B3_LARGE_FLOAT < FLT_MAX\n"
+"#define B3_LARGE_FLOAT 1e18f\n"
+"#define B3_INFINITY 1e18f\n"
+"#define b3Assert(a)\n"
+"#define b3ConstArray(a) __global const a*\n"
+"#define b3AtomicInc atomic_inc\n"
+"#define b3AtomicAdd atomic_add\n"
+"#define b3Fabs fabs\n"
+"#define b3Sqrt native_sqrt\n"
+"#define b3Sin native_sin\n"
+"#define b3Cos native_cos\n"
+"#define B3_STATIC\n"
+"#endif\n"
+"#endif\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+" typedef float4 b3Float4;\n"
+" #define b3Float4ConstArg const b3Float4\n"
+" #define b3MakeFloat4 (float4)\n"
+" float b3Dot3F4(b3Float4ConstArg v0,b3Float4ConstArg v1)\n"
+" {\n"
+" float4 a1 = b3MakeFloat4(v0.xyz,0.f);\n"
+" float4 b1 = b3MakeFloat4(v1.xyz,0.f);\n"
+" return dot(a1, b1);\n"
+" }\n"
+" b3Float4 b3Cross3(b3Float4ConstArg v0,b3Float4ConstArg v1)\n"
+" {\n"
+" float4 a1 = b3MakeFloat4(v0.xyz,0.f);\n"
+" float4 b1 = b3MakeFloat4(v1.xyz,0.f);\n"
+" return cross(a1, b1);\n"
+" }\n"
+" #define b3MinFloat4 min\n"
+" #define b3MaxFloat4 max\n"
+" #define b3Normalized(a) normalize(a)\n"
+"#endif \n"
+" \n"
+"inline bool b3IsAlmostZero(b3Float4ConstArg v)\n"
+"{\n"
+" if(b3Fabs(v.x)>1e-6 || b3Fabs(v.y)>1e-6 || b3Fabs(v.z)>1e-6) \n"
+" return false;\n"
+" return true;\n"
+"}\n"
+"inline int b3MaxDot( b3Float4ConstArg vec, __global const b3Float4* vecArray, int vecLen, float* dotOut )\n"
+"{\n"
+" float maxDot = -B3_INFINITY;\n"
+" int i = 0;\n"
+" int ptIndex = -1;\n"
+" for( i = 0; i < vecLen; i++ )\n"
+" {\n"
+" float dot = b3Dot3F4(vecArray[i],vec);\n"
+" \n"
+" if( dot > maxDot )\n"
+" {\n"
+" maxDot = dot;\n"
+" ptIndex = i;\n"
+" }\n"
+" }\n"
+" b3Assert(ptIndex>=0);\n"
+" if (ptIndex<0)\n"
+" {\n"
+" ptIndex = 0;\n"
+" }\n"
+" *dotOut = maxDot;\n"
+" return ptIndex;\n"
+"}\n"
+"#endif //B3_FLOAT4_H\n"
+"typedef struct b3Contact4Data b3Contact4Data_t;\n"
+"struct b3Contact4Data\n"
+"{\n"
+" b3Float4 m_worldPosB[4];\n"
+"// b3Float4 m_localPosA[4];\n"
+"// b3Float4 m_localPosB[4];\n"
+" b3Float4 m_worldNormalOnB; // w: m_nPoints\n"
+" unsigned short m_restituitionCoeffCmp;\n"
+" unsigned short m_frictionCoeffCmp;\n"
+" int m_batchIdx;\n"
+" int m_bodyAPtrAndSignBit;//x:m_bodyAPtr, y:m_bodyBPtr\n"
+" int m_bodyBPtrAndSignBit;\n"
+" int m_childIndexA;\n"
+" int m_childIndexB;\n"
+" int m_unused1;\n"
+" int m_unused2;\n"
+"};\n"
+"inline int b3Contact4Data_getNumPoints(const struct b3Contact4Data* contact)\n"
+"{\n"
+" return (int)contact->m_worldNormalOnB.w;\n"
+"};\n"
+"inline void b3Contact4Data_setNumPoints(struct b3Contact4Data* contact, int numPoints)\n"
+"{\n"
+" contact->m_worldNormalOnB.w = (float)numPoints;\n"
+"};\n"
+"#endif //B3_CONTACT4DATA_H\n"
+"#ifndef B3_CONVEX_POLYHEDRON_DATA_H\n"
+"#define B3_CONVEX_POLYHEDRON_DATA_H\n"
+"#ifndef B3_FLOAT4_H\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+"#endif \n"
+"#endif //B3_FLOAT4_H\n"
+"#ifndef B3_QUAT_H\n"
+"#define B3_QUAT_H\n"
+"#ifndef B3_PLATFORM_DEFINITIONS_H\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+"#endif\n"
+"#endif\n"
+"#ifndef B3_FLOAT4_H\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+"#endif \n"
+"#endif //B3_FLOAT4_H\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+" typedef float4 b3Quat;\n"
+" #define b3QuatConstArg const b3Quat\n"
+" \n"
+" \n"
+"inline float4 b3FastNormalize4(float4 v)\n"
+"{\n"
+" v = (float4)(v.xyz,0.f);\n"
+" return fast_normalize(v);\n"
+"}\n"
+" \n"
+"inline b3Quat b3QuatMul(b3Quat a, b3Quat b);\n"
+"inline b3Quat b3QuatNormalized(b3QuatConstArg in);\n"
+"inline b3Quat b3QuatRotate(b3QuatConstArg q, b3QuatConstArg vec);\n"
+"inline b3Quat b3QuatInvert(b3QuatConstArg q);\n"
+"inline b3Quat b3QuatInverse(b3QuatConstArg q);\n"
+"inline b3Quat b3QuatMul(b3QuatConstArg a, b3QuatConstArg b)\n"
+"{\n"
+" b3Quat ans;\n"
+" ans = b3Cross3( a, b );\n"
+" ans += a.w*b+b.w*a;\n"
+"// ans.w = a.w*b.w - (a.x*b.x+a.y*b.y+a.z*b.z);\n"
+" ans.w = a.w*b.w - b3Dot3F4(a, b);\n"
+" return ans;\n"
+"}\n"
+"inline b3Quat b3QuatNormalized(b3QuatConstArg in)\n"
+"{\n"
+" b3Quat q;\n"
+" q=in;\n"
+" //return b3FastNormalize4(in);\n"
+" float len = native_sqrt(dot(q, q));\n"
+" if(len > 0.f)\n"
+" {\n"
+" q *= 1.f / len;\n"
+" }\n"
+" else\n"
+" {\n"
+" q.x = q.y = q.z = 0.f;\n"
+" q.w = 1.f;\n"
+" }\n"
+" return q;\n"
+"}\n"
+"inline float4 b3QuatRotate(b3QuatConstArg q, b3QuatConstArg vec)\n"
+"{\n"
+" b3Quat qInv = b3QuatInvert( q );\n"
+" float4 vcpy = vec;\n"
+" vcpy.w = 0.f;\n"
+" float4 out = b3QuatMul(b3QuatMul(q,vcpy),qInv);\n"
+" return out;\n"
+"}\n"
+"inline b3Quat b3QuatInverse(b3QuatConstArg q)\n"
+"{\n"
+" return (b3Quat)(-q.xyz, q.w);\n"
+"}\n"
+"inline b3Quat b3QuatInvert(b3QuatConstArg q)\n"
+"{\n"
+" return (b3Quat)(-q.xyz, q.w);\n"
+"}\n"
+"inline float4 b3QuatInvRotate(b3QuatConstArg q, b3QuatConstArg vec)\n"
+"{\n"
+" return b3QuatRotate( b3QuatInvert( q ), vec );\n"
+"}\n"
+"inline b3Float4 b3TransformPoint(b3Float4ConstArg point, b3Float4ConstArg translation, b3QuatConstArg orientation)\n"
+"{\n"
+" return b3QuatRotate( orientation, point ) + (translation);\n"
+"}\n"
+" \n"
+"#endif \n"
+"#endif //B3_QUAT_H\n"
+"typedef struct b3GpuFace b3GpuFace_t;\n"
+"struct b3GpuFace\n"
+"{\n"
+" b3Float4 m_plane;\n"
+" int m_indexOffset;\n"
+" int m_numIndices;\n"
+" int m_unusedPadding1;\n"
+" int m_unusedPadding2;\n"
+"};\n"
+"typedef struct b3ConvexPolyhedronData b3ConvexPolyhedronData_t;\n"
+"struct b3ConvexPolyhedronData\n"
+"{\n"
+" b3Float4 m_localCenter;\n"
+" b3Float4 m_extents;\n"
+" b3Float4 mC;\n"
+" b3Float4 mE;\n"
+" float m_radius;\n"
+" int m_faceOffset;\n"
+" int m_numFaces;\n"
+" int m_numVertices;\n"
+" int m_vertexOffset;\n"
+" int m_uniqueEdgesOffset;\n"
+" int m_numUniqueEdges;\n"
+" int m_unused;\n"
+"};\n"
+"#endif //B3_CONVEX_POLYHEDRON_DATA_H\n"
+"#ifndef B3_COLLIDABLE_H\n"
+"#define B3_COLLIDABLE_H\n"
+"#ifndef B3_FLOAT4_H\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+"#endif \n"
+"#endif //B3_FLOAT4_H\n"
+"#ifndef B3_QUAT_H\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+"#endif \n"
+"#endif //B3_QUAT_H\n"
+"enum b3ShapeTypes\n"
+"{\n"
+" SHAPE_HEIGHT_FIELD=1,\n"
+" SHAPE_CONVEX_HULL=3,\n"
+" SHAPE_PLANE=4,\n"
+" SHAPE_CONCAVE_TRIMESH=5,\n"
+" SHAPE_COMPOUND_OF_CONVEX_HULLS=6,\n"
+" SHAPE_SPHERE=7,\n"
+" MAX_NUM_SHAPE_TYPES,\n"
+"};\n"
+"typedef struct b3Collidable b3Collidable_t;\n"
+"struct b3Collidable\n"
+"{\n"
+" union {\n"
+" int m_numChildShapes;\n"
+" int m_bvhIndex;\n"
+" };\n"
+" union\n"
+" {\n"
+" float m_radius;\n"
+" int m_compoundBvhIndex;\n"
+" };\n"
+" int m_shapeType;\n"
+" int m_shapeIndex;\n"
+"};\n"
+"typedef struct b3GpuChildShape b3GpuChildShape_t;\n"
+"struct b3GpuChildShape\n"
+"{\n"
+" b3Float4 m_childPosition;\n"
+" b3Quat m_childOrientation;\n"
+" int m_shapeIndex;\n"
+" int m_unused0;\n"
+" int m_unused1;\n"
+" int m_unused2;\n"
+"};\n"
+"struct b3CompoundOverlappingPair\n"
+"{\n"
+" int m_bodyIndexA;\n"
+" int m_bodyIndexB;\n"
+"// int m_pairType;\n"
+" int m_childShapeIndexA;\n"
+" int m_childShapeIndexB;\n"
+"};\n"
+"#endif //B3_COLLIDABLE_H\n"
+"#ifndef B3_RIGIDBODY_DATA_H\n"
+"#define B3_RIGIDBODY_DATA_H\n"
+"#ifndef B3_FLOAT4_H\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+"#endif \n"
+"#endif //B3_FLOAT4_H\n"
+"#ifndef B3_QUAT_H\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+"#endif \n"
+"#endif //B3_QUAT_H\n"
+"#ifndef B3_MAT3x3_H\n"
+"#define B3_MAT3x3_H\n"
+"#ifndef B3_QUAT_H\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+"#endif \n"
+"#endif //B3_QUAT_H\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+"typedef struct\n"
+"{\n"
+" b3Float4 m_row[3];\n"
+"}b3Mat3x3;\n"
+"#define b3Mat3x3ConstArg const b3Mat3x3\n"
+"#define b3GetRow(m,row) (m.m_row[row])\n"
+"inline b3Mat3x3 b3QuatGetRotationMatrix(b3Quat quat)\n"
+"{\n"
+" b3Float4 quat2 = (b3Float4)(quat.x*quat.x, quat.y*quat.y, quat.z*quat.z, 0.f);\n"
+" b3Mat3x3 out;\n"
+" out.m_row[0].x=1-2*quat2.y-2*quat2.z;\n"
+" out.m_row[0].y=2*quat.x*quat.y-2*quat.w*quat.z;\n"
+" out.m_row[0].z=2*quat.x*quat.z+2*quat.w*quat.y;\n"
+" out.m_row[0].w = 0.f;\n"
+" out.m_row[1].x=2*quat.x*quat.y+2*quat.w*quat.z;\n"
+" out.m_row[1].y=1-2*quat2.x-2*quat2.z;\n"
+" out.m_row[1].z=2*quat.y*quat.z-2*quat.w*quat.x;\n"
+" out.m_row[1].w = 0.f;\n"
+" out.m_row[2].x=2*quat.x*quat.z-2*quat.w*quat.y;\n"
+" out.m_row[2].y=2*quat.y*quat.z+2*quat.w*quat.x;\n"
+" out.m_row[2].z=1-2*quat2.x-2*quat2.y;\n"
+" out.m_row[2].w = 0.f;\n"
+" return out;\n"
+"}\n"
+"inline b3Mat3x3 b3AbsoluteMat3x3(b3Mat3x3ConstArg matIn)\n"
+"{\n"
+" b3Mat3x3 out;\n"
+" out.m_row[0] = fabs(matIn.m_row[0]);\n"
+" out.m_row[1] = fabs(matIn.m_row[1]);\n"
+" out.m_row[2] = fabs(matIn.m_row[2]);\n"
+" return out;\n"
+"}\n"
+"__inline\n"
+"b3Mat3x3 mtZero();\n"
+"__inline\n"
+"b3Mat3x3 mtIdentity();\n"
+"__inline\n"
+"b3Mat3x3 mtTranspose(b3Mat3x3 m);\n"
+"__inline\n"
+"b3Mat3x3 mtMul(b3Mat3x3 a, b3Mat3x3 b);\n"
+"__inline\n"
+"b3Float4 mtMul1(b3Mat3x3 a, b3Float4 b);\n"
+"__inline\n"
+"b3Float4 mtMul3(b3Float4 a, b3Mat3x3 b);\n"
+"__inline\n"
+"b3Mat3x3 mtZero()\n"
+"{\n"
+" b3Mat3x3 m;\n"
+" m.m_row[0] = (b3Float4)(0.f);\n"
+" m.m_row[1] = (b3Float4)(0.f);\n"
+" m.m_row[2] = (b3Float4)(0.f);\n"
+" return m;\n"
+"}\n"
+"__inline\n"
+"b3Mat3x3 mtIdentity()\n"
+"{\n"
+" b3Mat3x3 m;\n"
+" m.m_row[0] = (b3Float4)(1,0,0,0);\n"
+" m.m_row[1] = (b3Float4)(0,1,0,0);\n"
+" m.m_row[2] = (b3Float4)(0,0,1,0);\n"
+" return m;\n"
+"}\n"
+"__inline\n"
+"b3Mat3x3 mtTranspose(b3Mat3x3 m)\n"
+"{\n"
+" b3Mat3x3 out;\n"
+" out.m_row[0] = (b3Float4)(m.m_row[0].x, m.m_row[1].x, m.m_row[2].x, 0.f);\n"
+" out.m_row[1] = (b3Float4)(m.m_row[0].y, m.m_row[1].y, m.m_row[2].y, 0.f);\n"
+" out.m_row[2] = (b3Float4)(m.m_row[0].z, m.m_row[1].z, m.m_row[2].z, 0.f);\n"
+" return out;\n"
+"}\n"
+"__inline\n"
+"b3Mat3x3 mtMul(b3Mat3x3 a, b3Mat3x3 b)\n"
+"{\n"
+" b3Mat3x3 transB;\n"
+" transB = mtTranspose( b );\n"
+" b3Mat3x3 ans;\n"
+" // why this doesn't run when 0ing in the for{}\n"
+" a.m_row[0].w = 0.f;\n"
+" a.m_row[1].w = 0.f;\n"
+" a.m_row[2].w = 0.f;\n"
+" for(int i=0; i<3; i++)\n"
+" {\n"
+"// a.m_row[i].w = 0.f;\n"
+" ans.m_row[i].x = b3Dot3F4(a.m_row[i],transB.m_row[0]);\n"
+" ans.m_row[i].y = b3Dot3F4(a.m_row[i],transB.m_row[1]);\n"
+" ans.m_row[i].z = b3Dot3F4(a.m_row[i],transB.m_row[2]);\n"
+" ans.m_row[i].w = 0.f;\n"
+" }\n"
+" return ans;\n"
+"}\n"
+"__inline\n"
+"b3Float4 mtMul1(b3Mat3x3 a, b3Float4 b)\n"
+"{\n"
+" b3Float4 ans;\n"
+" ans.x = b3Dot3F4( a.m_row[0], b );\n"
+" ans.y = b3Dot3F4( a.m_row[1], b );\n"
+" ans.z = b3Dot3F4( a.m_row[2], b );\n"
+" ans.w = 0.f;\n"
+" return ans;\n"
+"}\n"
+"__inline\n"
+"b3Float4 mtMul3(b3Float4 a, b3Mat3x3 b)\n"
+"{\n"
+" b3Float4 colx = b3MakeFloat4(b.m_row[0].x, b.m_row[1].x, b.m_row[2].x, 0);\n"
+" b3Float4 coly = b3MakeFloat4(b.m_row[0].y, b.m_row[1].y, b.m_row[2].y, 0);\n"
+" b3Float4 colz = b3MakeFloat4(b.m_row[0].z, b.m_row[1].z, b.m_row[2].z, 0);\n"
+" b3Float4 ans;\n"
+" ans.x = b3Dot3F4( a, colx );\n"
+" ans.y = b3Dot3F4( a, coly );\n"
+" ans.z = b3Dot3F4( a, colz );\n"
+" return ans;\n"
+"}\n"
+"#endif\n"
+"#endif //B3_MAT3x3_H\n"
+"typedef struct b3RigidBodyData b3RigidBodyData_t;\n"
+"struct b3RigidBodyData\n"
+"{\n"
+" b3Float4 m_pos;\n"
+" b3Quat m_quat;\n"
+" b3Float4 m_linVel;\n"
+" b3Float4 m_angVel;\n"
+" int m_collidableIdx;\n"
+" float m_invMass;\n"
+" float m_restituitionCoeff;\n"
+" float m_frictionCoeff;\n"
+"};\n"
+"typedef struct b3InertiaData b3InertiaData_t;\n"
+"struct b3InertiaData\n"
+"{\n"
+" b3Mat3x3 m_invInertiaWorld;\n"
+" b3Mat3x3 m_initInvInertia;\n"
+"};\n"
+"#endif //B3_RIGIDBODY_DATA_H\n"
+" \n"
+"#define GET_NPOINTS(x) (x).m_worldNormalOnB.w\n"
+"#define SELECT_UINT4( b, a, condition ) select( b,a,condition )\n"
+"#define make_float4 (float4)\n"
+"#define make_float2 (float2)\n"
+"#define make_uint4 (uint4)\n"
+"#define make_int4 (int4)\n"
+"#define make_uint2 (uint2)\n"
+"#define make_int2 (int2)\n"
+"__inline\n"
+"float fastDiv(float numerator, float denominator)\n"
+"{\n"
+" return native_divide(numerator, denominator); \n"
+"// return numerator/denominator; \n"
+"}\n"
+"__inline\n"
+"float4 fastDiv4(float4 numerator, float4 denominator)\n"
+"{\n"
+" return native_divide(numerator, denominator); \n"
+"}\n"
+"__inline\n"
+"float4 cross3(float4 a, float4 b)\n"
+"{\n"
+" return cross(a,b);\n"
+"}\n"
+"//#define dot3F4 dot\n"
+"__inline\n"
+"float dot3F4(float4 a, float4 b)\n"
+"{\n"
+" float4 a1 = make_float4(a.xyz,0.f);\n"
+" float4 b1 = make_float4(b.xyz,0.f);\n"
+" return dot(a1, b1);\n"
+"}\n"
+"__inline\n"
+"float4 fastNormalize4(float4 v)\n"
+"{\n"
+" return fast_normalize(v);\n"
+"}\n"
+"///////////////////////////////////////\n"
+"// Quaternion\n"
+"///////////////////////////////////////\n"
+"typedef float4 Quaternion;\n"
+"__inline\n"
+"Quaternion qtMul(Quaternion a, Quaternion b);\n"
+"__inline\n"
+"Quaternion qtNormalize(Quaternion in);\n"
+"__inline\n"
+"float4 qtRotate(Quaternion q, float4 vec);\n"
+"__inline\n"
+"Quaternion qtInvert(Quaternion q);\n"
+"__inline\n"
+"Quaternion qtMul(Quaternion a, Quaternion b)\n"
+"{\n"
+" Quaternion ans;\n"
+" ans = cross3( a, b );\n"
+" ans += a.w*b+b.w*a;\n"
+"// ans.w = a.w*b.w - (a.x*b.x+a.y*b.y+a.z*b.z);\n"
+" ans.w = a.w*b.w - dot3F4(a, b);\n"
+" return ans;\n"
+"}\n"
+"__inline\n"
+"Quaternion qtNormalize(Quaternion in)\n"
+"{\n"
+" return fastNormalize4(in);\n"
+"// in /= length( in );\n"
+"// return in;\n"
+"}\n"
+"__inline\n"
+"float4 qtRotate(Quaternion q, float4 vec)\n"
+"{\n"
+" Quaternion qInv = qtInvert( q );\n"
+" float4 vcpy = vec;\n"
+" vcpy.w = 0.f;\n"
+" float4 out = qtMul(qtMul(q,vcpy),qInv);\n"
+" return out;\n"
+"}\n"
+"__inline\n"
+"Quaternion qtInvert(Quaternion q)\n"
+"{\n"
+" return (Quaternion)(-q.xyz, q.w);\n"
+"}\n"
+"__inline\n"
+"float4 qtInvRotate(const Quaternion q, float4 vec)\n"
+"{\n"
+" return qtRotate( qtInvert( q ), vec );\n"
+"}\n"
+"__inline\n"
+"float4 transform(const float4* p, const float4* translation, const Quaternion* orientation)\n"
+"{\n"
+" return qtRotate( *orientation, *p ) + (*translation);\n"
+"}\n"
+"__inline\n"
+"float4 normalize3(const float4 a)\n"
+"{\n"
+" float4 n = make_float4(a.x, a.y, a.z, 0.f);\n"
+" return fastNormalize4( n );\n"
+"}\n"
+"__inline float4 lerp3(const float4 a,const float4 b, float t)\n"
+"{\n"
+" return make_float4( a.x + (b.x - a.x) * t,\n"
+" a.y + (b.y - a.y) * t,\n"
+" a.z + (b.z - a.z) * t,\n"
+" 0.f);\n"
+"}\n"
+"// Clips a face to the back of a plane, return the number of vertices out, stored in ppVtxOut\n"
+"int clipFaceGlobal(__global const float4* pVtxIn, int numVertsIn, float4 planeNormalWS,float planeEqWS, __global float4* ppVtxOut)\n"
+"{\n"
+" \n"
+" int ve;\n"
+" float ds, de;\n"
+" int numVertsOut = 0;\n"
+" //double-check next test\n"
+" if (numVertsIn < 2)\n"
+" return 0;\n"
+" \n"
+" float4 firstVertex=pVtxIn[numVertsIn-1];\n"
+" float4 endVertex = pVtxIn[0];\n"
+" \n"
+" ds = dot3F4(planeNormalWS,firstVertex)+planeEqWS;\n"
+" \n"
+" for (ve = 0; ve < numVertsIn; ve++)\n"
+" {\n"
+" endVertex=pVtxIn[ve];\n"
+" de = dot3F4(planeNormalWS,endVertex)+planeEqWS;\n"
+" if (ds<0)\n"
+" {\n"
+" if (de<0)\n"
+" {\n"
+" // Start < 0, end < 0, so output endVertex\n"
+" ppVtxOut[numVertsOut++] = endVertex;\n"
+" }\n"
+" else\n"
+" {\n"
+" // Start < 0, end >= 0, so output intersection\n"
+" ppVtxOut[numVertsOut++] = lerp3(firstVertex, endVertex,(ds * 1.f/(ds - de)) );\n"
+" }\n"
+" }\n"
+" else\n"
+" {\n"
+" if (de<0)\n"
+" {\n"
+" // Start >= 0, end < 0 so output intersection and end\n"
+" ppVtxOut[numVertsOut++] = lerp3(firstVertex, endVertex,(ds * 1.f/(ds - de)) );\n"
+" ppVtxOut[numVertsOut++] = endVertex;\n"
+" }\n"
+" }\n"
+" firstVertex = endVertex;\n"
+" ds = de;\n"
+" }\n"
+" return numVertsOut;\n"
+"}\n"
+"// Clips a face to the back of a plane, return the number of vertices out, stored in ppVtxOut\n"
+"int clipFace(const float4* pVtxIn, int numVertsIn, float4 planeNormalWS,float planeEqWS, float4* ppVtxOut)\n"
+"{\n"
+" \n"
+" int ve;\n"
+" float ds, de;\n"
+" int numVertsOut = 0;\n"
+"//double-check next test\n"
+" if (numVertsIn < 2)\n"
+" return 0;\n"
+" float4 firstVertex=pVtxIn[numVertsIn-1];\n"
+" float4 endVertex = pVtxIn[0];\n"
+" \n"
+" ds = dot3F4(planeNormalWS,firstVertex)+planeEqWS;\n"
+" for (ve = 0; ve < numVertsIn; ve++)\n"
+" {\n"
+" endVertex=pVtxIn[ve];\n"
+" de = dot3F4(planeNormalWS,endVertex)+planeEqWS;\n"
+" if (ds<0)\n"
+" {\n"
+" if (de<0)\n"
+" {\n"
+" // Start < 0, end < 0, so output endVertex\n"
+" ppVtxOut[numVertsOut++] = endVertex;\n"
+" }\n"
+" else\n"
+" {\n"
+" // Start < 0, end >= 0, so output intersection\n"
+" ppVtxOut[numVertsOut++] = lerp3(firstVertex, endVertex,(ds * 1.f/(ds - de)) );\n"
+" }\n"
+" }\n"
+" else\n"
+" {\n"
+" if (de<0)\n"
+" {\n"
+" // Start >= 0, end < 0 so output intersection and end\n"
+" ppVtxOut[numVertsOut++] = lerp3(firstVertex, endVertex,(ds * 1.f/(ds - de)) );\n"
+" ppVtxOut[numVertsOut++] = endVertex;\n"
+" }\n"
+" }\n"
+" firstVertex = endVertex;\n"
+" ds = de;\n"
+" }\n"
+" return numVertsOut;\n"
+"}\n"
+"int clipFaceAgainstHull(const float4 separatingNormal, __global const b3ConvexPolyhedronData_t* hullA, \n"
+" const float4 posA, const Quaternion ornA, float4* worldVertsB1, int numWorldVertsB1,\n"
+" float4* worldVertsB2, int capacityWorldVertsB2,\n"
+" const float minDist, float maxDist,\n"
+" __global const float4* vertices,\n"
+" __global const b3GpuFace_t* faces,\n"
+" __global const int* indices,\n"
+" float4* contactsOut,\n"
+" int contactCapacity)\n"
+"{\n"
+" int numContactsOut = 0;\n"
+" float4* pVtxIn = worldVertsB1;\n"
+" float4* pVtxOut = worldVertsB2;\n"
+" \n"
+" int numVertsIn = numWorldVertsB1;\n"
+" int numVertsOut = 0;\n"
+" int closestFaceA=-1;\n"
+" {\n"
+" float dmin = FLT_MAX;\n"
+" for(int face=0;face<hullA->m_numFaces;face++)\n"
+" {\n"
+" const float4 Normal = make_float4(\n"
+" faces[hullA->m_faceOffset+face].m_plane.x, \n"
+" faces[hullA->m_faceOffset+face].m_plane.y, \n"
+" faces[hullA->m_faceOffset+face].m_plane.z,0.f);\n"
+" const float4 faceANormalWS = qtRotate(ornA,Normal);\n"
+" \n"
+" float d = dot3F4(faceANormalWS,separatingNormal);\n"
+" if (d < dmin)\n"
+" {\n"
+" dmin = d;\n"
+" closestFaceA = face;\n"
+" }\n"
+" }\n"
+" }\n"
+" if (closestFaceA<0)\n"
+" return numContactsOut;\n"
+" b3GpuFace_t polyA = faces[hullA->m_faceOffset+closestFaceA];\n"
+" // clip polygon to back of planes of all faces of hull A that are adjacent to witness face\n"
+" int numVerticesA = polyA.m_numIndices;\n"
+" for(int e0=0;e0<numVerticesA;e0++)\n"
+" {\n"
+" const float4 a = vertices[hullA->m_vertexOffset+indices[polyA.m_indexOffset+e0]];\n"
+" const float4 b = vertices[hullA->m_vertexOffset+indices[polyA.m_indexOffset+((e0+1)%numVerticesA)]];\n"
+" const float4 edge0 = a - b;\n"
+" const float4 WorldEdge0 = qtRotate(ornA,edge0);\n"
+" float4 planeNormalA = make_float4(polyA.m_plane.x,polyA.m_plane.y,polyA.m_plane.z,0.f);\n"
+" float4 worldPlaneAnormal1 = qtRotate(ornA,planeNormalA);\n"
+" float4 planeNormalWS1 = -cross3(WorldEdge0,worldPlaneAnormal1);\n"
+" float4 worldA1 = transform(&a,&posA,&ornA);\n"
+" float planeEqWS1 = -dot3F4(worldA1,planeNormalWS1);\n"
+" \n"
+" float4 planeNormalWS = planeNormalWS1;\n"
+" float planeEqWS=planeEqWS1;\n"
+" \n"
+" //clip face\n"
+" //clipFace(*pVtxIn, *pVtxOut,planeNormalWS,planeEqWS);\n"
+" numVertsOut = clipFace(pVtxIn, numVertsIn, planeNormalWS,planeEqWS, pVtxOut);\n"
+" //btSwap(pVtxIn,pVtxOut);\n"
+" float4* tmp = pVtxOut;\n"
+" pVtxOut = pVtxIn;\n"
+" pVtxIn = tmp;\n"
+" numVertsIn = numVertsOut;\n"
+" numVertsOut = 0;\n"
+" }\n"
+" \n"
+" // only keep points that are behind the witness face\n"
+" {\n"
+" float4 localPlaneNormal = make_float4(polyA.m_plane.x,polyA.m_plane.y,polyA.m_plane.z,0.f);\n"
+" float localPlaneEq = polyA.m_plane.w;\n"
+" float4 planeNormalWS = qtRotate(ornA,localPlaneNormal);\n"
+" float planeEqWS=localPlaneEq-dot3F4(planeNormalWS,posA);\n"
+" for (int i=0;i<numVertsIn;i++)\n"
+" {\n"
+" float depth = dot3F4(planeNormalWS,pVtxIn[i])+planeEqWS;\n"
+" if (depth <=minDist)\n"
+" {\n"
+" depth = minDist;\n"
+" }\n"
+" if (depth <=maxDist)\n"
+" {\n"
+" float4 pointInWorld = pVtxIn[i];\n"
+" //resultOut.addContactPoint(separatingNormal,point,depth);\n"
+" contactsOut[numContactsOut++] = make_float4(pointInWorld.x,pointInWorld.y,pointInWorld.z,depth);\n"
+" }\n"
+" }\n"
+" }\n"
+" return numContactsOut;\n"
+"}\n"
+"int clipFaceAgainstHullLocalA(const float4 separatingNormal, const b3ConvexPolyhedronData_t* hullA, \n"
+" const float4 posA, const Quaternion ornA, float4* worldVertsB1, int numWorldVertsB1,\n"
+" float4* worldVertsB2, int capacityWorldVertsB2,\n"
+" const float minDist, float maxDist,\n"
+" const float4* verticesA,\n"
+" const b3GpuFace_t* facesA,\n"
+" const int* indicesA,\n"
+" __global const float4* verticesB,\n"
+" __global const b3GpuFace_t* facesB,\n"
+" __global const int* indicesB,\n"
+" float4* contactsOut,\n"
+" int contactCapacity)\n"
+"{\n"
+" int numContactsOut = 0;\n"
+" float4* pVtxIn = worldVertsB1;\n"
+" float4* pVtxOut = worldVertsB2;\n"
+" \n"
+" int numVertsIn = numWorldVertsB1;\n"
+" int numVertsOut = 0;\n"
+" int closestFaceA=-1;\n"
+" {\n"
+" float dmin = FLT_MAX;\n"
+" for(int face=0;face<hullA->m_numFaces;face++)\n"
+" {\n"
+" const float4 Normal = make_float4(\n"
+" facesA[hullA->m_faceOffset+face].m_plane.x, \n"
+" facesA[hullA->m_faceOffset+face].m_plane.y, \n"
+" facesA[hullA->m_faceOffset+face].m_plane.z,0.f);\n"
+" const float4 faceANormalWS = qtRotate(ornA,Normal);\n"
+" \n"
+" float d = dot3F4(faceANormalWS,separatingNormal);\n"
+" if (d < dmin)\n"
+" {\n"
+" dmin = d;\n"
+" closestFaceA = face;\n"
+" }\n"
+" }\n"
+" }\n"
+" if (closestFaceA<0)\n"
+" return numContactsOut;\n"
+" b3GpuFace_t polyA = facesA[hullA->m_faceOffset+closestFaceA];\n"
+" // clip polygon to back of planes of all faces of hull A that are adjacent to witness face\n"
+" int numVerticesA = polyA.m_numIndices;\n"
+" for(int e0=0;e0<numVerticesA;e0++)\n"
+" {\n"
+" const float4 a = verticesA[hullA->m_vertexOffset+indicesA[polyA.m_indexOffset+e0]];\n"
+" const float4 b = verticesA[hullA->m_vertexOffset+indicesA[polyA.m_indexOffset+((e0+1)%numVerticesA)]];\n"
+" const float4 edge0 = a - b;\n"
+" const float4 WorldEdge0 = qtRotate(ornA,edge0);\n"
+" float4 planeNormalA = make_float4(polyA.m_plane.x,polyA.m_plane.y,polyA.m_plane.z,0.f);\n"
+" float4 worldPlaneAnormal1 = qtRotate(ornA,planeNormalA);\n"
+" float4 planeNormalWS1 = -cross3(WorldEdge0,worldPlaneAnormal1);\n"
+" float4 worldA1 = transform(&a,&posA,&ornA);\n"
+" float planeEqWS1 = -dot3F4(worldA1,planeNormalWS1);\n"
+" \n"
+" float4 planeNormalWS = planeNormalWS1;\n"
+" float planeEqWS=planeEqWS1;\n"
+" \n"
+" //clip face\n"
+" //clipFace(*pVtxIn, *pVtxOut,planeNormalWS,planeEqWS);\n"
+" numVertsOut = clipFace(pVtxIn, numVertsIn, planeNormalWS,planeEqWS, pVtxOut);\n"
+" //btSwap(pVtxIn,pVtxOut);\n"
+" float4* tmp = pVtxOut;\n"
+" pVtxOut = pVtxIn;\n"
+" pVtxIn = tmp;\n"
+" numVertsIn = numVertsOut;\n"
+" numVertsOut = 0;\n"
+" }\n"
+" \n"
+" // only keep points that are behind the witness face\n"
+" {\n"
+" float4 localPlaneNormal = make_float4(polyA.m_plane.x,polyA.m_plane.y,polyA.m_plane.z,0.f);\n"
+" float localPlaneEq = polyA.m_plane.w;\n"
+" float4 planeNormalWS = qtRotate(ornA,localPlaneNormal);\n"
+" float planeEqWS=localPlaneEq-dot3F4(planeNormalWS,posA);\n"
+" for (int i=0;i<numVertsIn;i++)\n"
+" {\n"
+" float depth = dot3F4(planeNormalWS,pVtxIn[i])+planeEqWS;\n"
+" if (depth <=minDist)\n"
+" {\n"
+" depth = minDist;\n"
+" }\n"
+" if (depth <=maxDist)\n"
+" {\n"
+" float4 pointInWorld = pVtxIn[i];\n"
+" //resultOut.addContactPoint(separatingNormal,point,depth);\n"
+" contactsOut[numContactsOut++] = make_float4(pointInWorld.x,pointInWorld.y,pointInWorld.z,depth);\n"
+" }\n"
+" }\n"
+" }\n"
+" return numContactsOut;\n"
+"}\n"
+"int clipHullAgainstHull(const float4 separatingNormal,\n"
+" __global const b3ConvexPolyhedronData_t* hullA, __global const b3ConvexPolyhedronData_t* hullB, \n"
+" const float4 posA, const Quaternion ornA,const float4 posB, const Quaternion ornB, \n"
+" float4* worldVertsB1, float4* worldVertsB2, int capacityWorldVerts,\n"
+" const float minDist, float maxDist,\n"
+" __global const float4* vertices,\n"
+" __global const b3GpuFace_t* faces,\n"
+" __global const int* indices,\n"
+" float4* localContactsOut,\n"
+" int localContactCapacity)\n"
+"{\n"
+" int numContactsOut = 0;\n"
+" int numWorldVertsB1= 0;\n"
+" int closestFaceB=-1;\n"
+" float dmax = -FLT_MAX;\n"
+" {\n"
+" for(int face=0;face<hullB->m_numFaces;face++)\n"
+" {\n"
+" const float4 Normal = make_float4(faces[hullB->m_faceOffset+face].m_plane.x, \n"
+" faces[hullB->m_faceOffset+face].m_plane.y, faces[hullB->m_faceOffset+face].m_plane.z,0.f);\n"
+" const float4 WorldNormal = qtRotate(ornB, Normal);\n"
+" float d = dot3F4(WorldNormal,separatingNormal);\n"
+" if (d > dmax)\n"
+" {\n"
+" dmax = d;\n"
+" closestFaceB = face;\n"
+" }\n"
+" }\n"
+" }\n"
+" {\n"
+" const b3GpuFace_t polyB = faces[hullB->m_faceOffset+closestFaceB];\n"
+" const int numVertices = polyB.m_numIndices;\n"
+" for(int e0=0;e0<numVertices;e0++)\n"
+" {\n"
+" const float4 b = vertices[hullB->m_vertexOffset+indices[polyB.m_indexOffset+e0]];\n"
+" worldVertsB1[numWorldVertsB1++] = transform(&b,&posB,&ornB);\n"
+" }\n"
+" }\n"
+" if (closestFaceB>=0)\n"
+" {\n"
+" numContactsOut = clipFaceAgainstHull(separatingNormal, hullA, \n"
+" posA,ornA,\n"
+" worldVertsB1,numWorldVertsB1,worldVertsB2,capacityWorldVerts, minDist, maxDist,vertices,\n"
+" faces,\n"
+" indices,localContactsOut,localContactCapacity);\n"
+" }\n"
+" return numContactsOut;\n"
+"}\n"
+"int clipHullAgainstHullLocalA(const float4 separatingNormal,\n"
+" const b3ConvexPolyhedronData_t* hullA, __global const b3ConvexPolyhedronData_t* hullB, \n"
+" const float4 posA, const Quaternion ornA,const float4 posB, const Quaternion ornB, \n"
+" float4* worldVertsB1, float4* worldVertsB2, int capacityWorldVerts,\n"
+" const float minDist, float maxDist,\n"
+" const float4* verticesA,\n"
+" const b3GpuFace_t* facesA,\n"
+" const int* indicesA,\n"
+" __global const float4* verticesB,\n"
+" __global const b3GpuFace_t* facesB,\n"
+" __global const int* indicesB,\n"
+" float4* localContactsOut,\n"
+" int localContactCapacity)\n"
+"{\n"
+" int numContactsOut = 0;\n"
+" int numWorldVertsB1= 0;\n"
+" int closestFaceB=-1;\n"
+" float dmax = -FLT_MAX;\n"
+" {\n"
+" for(int face=0;face<hullB->m_numFaces;face++)\n"
+" {\n"
+" const float4 Normal = make_float4(facesB[hullB->m_faceOffset+face].m_plane.x, \n"
+" facesB[hullB->m_faceOffset+face].m_plane.y, facesB[hullB->m_faceOffset+face].m_plane.z,0.f);\n"
+" const float4 WorldNormal = qtRotate(ornB, Normal);\n"
+" float d = dot3F4(WorldNormal,separatingNormal);\n"
+" if (d > dmax)\n"
+" {\n"
+" dmax = d;\n"
+" closestFaceB = face;\n"
+" }\n"
+" }\n"
+" }\n"
+" {\n"
+" const b3GpuFace_t polyB = facesB[hullB->m_faceOffset+closestFaceB];\n"
+" const int numVertices = polyB.m_numIndices;\n"
+" for(int e0=0;e0<numVertices;e0++)\n"
+" {\n"
+" const float4 b = verticesB[hullB->m_vertexOffset+indicesB[polyB.m_indexOffset+e0]];\n"
+" worldVertsB1[numWorldVertsB1++] = transform(&b,&posB,&ornB);\n"
+" }\n"
+" }\n"
+" if (closestFaceB>=0)\n"
+" {\n"
+" numContactsOut = clipFaceAgainstHullLocalA(separatingNormal, hullA, \n"
+" posA,ornA,\n"
+" worldVertsB1,numWorldVertsB1,worldVertsB2,capacityWorldVerts, minDist, maxDist,\n"
+" verticesA,facesA,indicesA,\n"
+" verticesB,facesB,indicesB,\n"
+" localContactsOut,localContactCapacity);\n"
+" }\n"
+" return numContactsOut;\n"
+"}\n"
+"#define PARALLEL_SUM(v, n) for(int j=1; j<n; j++) v[0] += v[j];\n"
+"#define PARALLEL_DO(execution, n) for(int ie=0; ie<n; ie++){execution;}\n"
+"#define REDUCE_MAX(v, n) {int i=0; for(int offset=0; offset<n; offset++) v[i] = (v[i].y > v[i+offset].y)? v[i]: v[i+offset]; }\n"
+"#define REDUCE_MIN(v, n) {int i=0; for(int offset=0; offset<n; offset++) v[i] = (v[i].y < v[i+offset].y)? v[i]: v[i+offset]; }\n"
+"int extractManifoldSequentialGlobal(__global const float4* p, int nPoints, float4 nearNormal, int4* contactIdx)\n"
+"{\n"
+" if( nPoints == 0 )\n"
+" return 0;\n"
+" \n"
+" if (nPoints <=4)\n"
+" return nPoints;\n"
+" \n"
+" \n"
+" if (nPoints >64)\n"
+" nPoints = 64;\n"
+" \n"
+" float4 center = make_float4(0.f);\n"
+" {\n"
+" \n"
+" for (int i=0;i<nPoints;i++)\n"
+" center += p[i];\n"
+" center /= (float)nPoints;\n"
+" }\n"
+" \n"
+" \n"
+" \n"
+" // sample 4 directions\n"
+" \n"
+" float4 aVector = p[0] - center;\n"
+" float4 u = cross3( nearNormal, aVector );\n"
+" float4 v = cross3( nearNormal, u );\n"
+" u = normalize3( u );\n"
+" v = normalize3( v );\n"
+" \n"
+" \n"
+" //keep point with deepest penetration\n"
+" float minW= FLT_MAX;\n"
+" \n"
+" int minIndex=-1;\n"
+" \n"
+" float4 maxDots;\n"
+" maxDots.x = FLT_MIN;\n"
+" maxDots.y = FLT_MIN;\n"
+" maxDots.z = FLT_MIN;\n"
+" maxDots.w = FLT_MIN;\n"
+" \n"
+" // idx, distance\n"
+" for(int ie = 0; ie<nPoints; ie++ )\n"
+" {\n"
+" if (p[ie].w<minW)\n"
+" {\n"
+" minW = p[ie].w;\n"
+" minIndex=ie;\n"
+" }\n"
+" float f;\n"
+" float4 r = p[ie]-center;\n"
+" f = dot3F4( u, r );\n"
+" if (f<maxDots.x)\n"
+" {\n"
+" maxDots.x = f;\n"
+" contactIdx[0].x = ie;\n"
+" }\n"
+" \n"
+" f = dot3F4( -u, r );\n"
+" if (f<maxDots.y)\n"
+" {\n"
+" maxDots.y = f;\n"
+" contactIdx[0].y = ie;\n"
+" }\n"
+" \n"
+" \n"
+" f = dot3F4( v, r );\n"
+" if (f<maxDots.z)\n"
+" {\n"
+" maxDots.z = f;\n"
+" contactIdx[0].z = ie;\n"
+" }\n"
+" \n"
+" f = dot3F4( -v, r );\n"
+" if (f<maxDots.w)\n"
+" {\n"
+" maxDots.w = f;\n"
+" contactIdx[0].w = ie;\n"
+" }\n"
+" \n"
+" }\n"
+" \n"
+" if (contactIdx[0].x != minIndex && contactIdx[0].y != minIndex && contactIdx[0].z != minIndex && contactIdx[0].w != minIndex)\n"
+" {\n"
+" //replace the first contact with minimum (todo: replace contact with least penetration)\n"
+" contactIdx[0].x = minIndex;\n"
+" }\n"
+" \n"
+" return 4;\n"
+" \n"
+"}\n"
+"int extractManifoldSequentialGlobalFake(__global const float4* p, int nPoints, float4 nearNormal, int* contactIdx)\n"
+"{\n"
+" contactIdx[0] = 0;\n"
+" contactIdx[1] = 1;\n"
+" contactIdx[2] = 2;\n"
+" contactIdx[3] = 3;\n"
+" \n"
+" if( nPoints == 0 ) return 0;\n"
+" \n"
+" nPoints = min2( nPoints, 4 );\n"
+" return nPoints;\n"
+" \n"
+"}\n"
+"int extractManifoldSequential(const float4* p, int nPoints, float4 nearNormal, int* contactIdx)\n"
+"{\n"
+" if( nPoints == 0 ) return 0;\n"
+" nPoints = min2( nPoints, 64 );\n"
+" float4 center = make_float4(0.f);\n"
+" {\n"
+" float4 v[64];\n"
+" for (int i=0;i<nPoints;i++)\n"
+" v[i] = p[i];\n"
+" //memcpy( v, p, nPoints*sizeof(float4) );\n"
+" PARALLEL_SUM( v, nPoints );\n"
+" center = v[0]/(float)nPoints;\n"
+" }\n"
+" \n"
+" { // sample 4 directions\n"
+" if( nPoints < 4 )\n"
+" {\n"
+" for(int i=0; i<nPoints; i++) \n"
+" contactIdx[i] = i;\n"
+" return nPoints;\n"
+" }\n"
+" float4 aVector = p[0] - center;\n"
+" float4 u = cross3( nearNormal, aVector );\n"
+" float4 v = cross3( nearNormal, u );\n"
+" u = normalize3( u );\n"
+" v = normalize3( v );\n"
+" int idx[4];\n"
+" float2 max00 = make_float2(0,FLT_MAX);\n"
+" {\n"
+" // idx, distance\n"
+" {\n"
+" {\n"
+" int4 a[64];\n"
+" for(int ie = 0; ie<nPoints; ie++ )\n"
+" {\n"
+" \n"
+" \n"
+" float f;\n"
+" float4 r = p[ie]-center;\n"
+" f = dot3F4( u, r );\n"
+" a[ie].x = ((*(u32*)&f) & 0xffffff00) | (0xff & ie);\n"
+" f = dot3F4( -u, r );\n"
+" a[ie].y = ((*(u32*)&f) & 0xffffff00) | (0xff & ie);\n"
+" f = dot3F4( v, r );\n"
+" a[ie].z = ((*(u32*)&f) & 0xffffff00) | (0xff & ie);\n"
+" f = dot3F4( -v, r );\n"
+" a[ie].w = ((*(u32*)&f) & 0xffffff00) | (0xff & ie);\n"
+" }\n"
+" for(int ie=0; ie<nPoints; ie++)\n"
+" {\n"
+" a[0].x = (a[0].x > a[ie].x )? a[0].x: a[ie].x;\n"
+" a[0].y = (a[0].y > a[ie].y )? a[0].y: a[ie].y;\n"
+" a[0].z = (a[0].z > a[ie].z )? a[0].z: a[ie].z;\n"
+" a[0].w = (a[0].w > a[ie].w )? a[0].w: a[ie].w;\n"
+" }\n"
+" idx[0] = (int)a[0].x & 0xff;\n"
+" idx[1] = (int)a[0].y & 0xff;\n"
+" idx[2] = (int)a[0].z & 0xff;\n"
+" idx[3] = (int)a[0].w & 0xff;\n"
+" }\n"
+" }\n"
+" {\n"
+" float2 h[64];\n"
+" PARALLEL_DO( h[ie] = make_float2((float)ie, p[ie].w), nPoints );\n"
+" REDUCE_MIN( h, nPoints );\n"
+" max00 = h[0];\n"
+" }\n"
+" }\n"
+" contactIdx[0] = idx[0];\n"
+" contactIdx[1] = idx[1];\n"
+" contactIdx[2] = idx[2];\n"
+" contactIdx[3] = idx[3];\n"
+" return 4;\n"
+" }\n"
+"}\n"
+"__kernel void extractManifoldAndAddContactKernel(__global const int4* pairs, \n"
+" __global const b3RigidBodyData_t* rigidBodies, \n"
+" __global const float4* closestPointsWorld,\n"
+" __global const float4* separatingNormalsWorld,\n"
+" __global const int* contactCounts,\n"
+" __global const int* contactOffsets,\n"
+" __global struct b3Contact4Data* restrict contactsOut,\n"
+" counter32_t nContactsOut,\n"
+" int contactCapacity,\n"
+" int numPairs,\n"
+" int pairIndex\n"
+" )\n"
+"{\n"
+" int idx = get_global_id(0);\n"
+" \n"
+" if (idx<numPairs)\n"
+" {\n"
+" float4 normal = separatingNormalsWorld[idx];\n"
+" int nPoints = contactCounts[idx];\n"
+" __global const float4* pointsIn = &closestPointsWorld[contactOffsets[idx]];\n"
+" float4 localPoints[64];\n"
+" for (int i=0;i<nPoints;i++)\n"
+" {\n"
+" localPoints[i] = pointsIn[i];\n"
+" }\n"
+" int contactIdx[4];// = {-1,-1,-1,-1};\n"
+" contactIdx[0] = -1;\n"
+" contactIdx[1] = -1;\n"
+" contactIdx[2] = -1;\n"
+" contactIdx[3] = -1;\n"
+" int nContacts = extractManifoldSequential(localPoints, nPoints, normal, contactIdx);\n"
+" int dstIdx;\n"
+" AppendInc( nContactsOut, dstIdx );\n"
+" if (dstIdx<contactCapacity)\n"
+" {\n"
+" __global struct b3Contact4Data* c = contactsOut + dstIdx;\n"
+" c->m_worldNormalOnB = -normal;\n"
+" c->m_restituitionCoeffCmp = (0.f*0xffff);c->m_frictionCoeffCmp = (0.7f*0xffff);\n"
+" c->m_batchIdx = idx;\n"
+" int bodyA = pairs[pairIndex].x;\n"
+" int bodyB = pairs[pairIndex].y;\n"
+" c->m_bodyAPtrAndSignBit = rigidBodies[bodyA].m_invMass==0 ? -bodyA:bodyA;\n"
+" c->m_bodyBPtrAndSignBit = rigidBodies[bodyB].m_invMass==0 ? -bodyB:bodyB;\n"
+" c->m_childIndexA = -1;\n"
+" c->m_childIndexB = -1;\n"
+" for (int i=0;i<nContacts;i++)\n"
+" {\n"
+" c->m_worldPosB[i] = localPoints[contactIdx[i]];\n"
+" }\n"
+" GET_NPOINTS(*c) = nContacts;\n"
+" }\n"
+" }\n"
+"}\n"
+"void trInverse(float4 translationIn, Quaternion orientationIn,\n"
+" float4* translationOut, Quaternion* orientationOut)\n"
+"{\n"
+" *orientationOut = qtInvert(orientationIn);\n"
+" *translationOut = qtRotate(*orientationOut, -translationIn);\n"
+"}\n"
+"void trMul(float4 translationA, Quaternion orientationA,\n"
+" float4 translationB, Quaternion orientationB,\n"
+" float4* translationOut, Quaternion* orientationOut)\n"
+"{\n"
+" *orientationOut = qtMul(orientationA,orientationB);\n"
+" *translationOut = transform(&translationB,&translationA,&orientationA);\n"
+"}\n"
+"__kernel void clipHullHullKernel( __global int4* pairs, \n"
+" __global const b3RigidBodyData_t* rigidBodies, \n"
+" __global const b3Collidable_t* collidables,\n"
+" __global const b3ConvexPolyhedronData_t* convexShapes, \n"
+" __global const float4* vertices,\n"
+" __global const float4* uniqueEdges,\n"
+" __global const b3GpuFace_t* faces,\n"
+" __global const int* indices,\n"
+" __global const float4* separatingNormals,\n"
+" __global const int* hasSeparatingAxis,\n"
+" __global struct b3Contact4Data* restrict globalContactsOut,\n"
+" counter32_t nGlobalContactsOut,\n"
+" int numPairs,\n"
+" int contactCapacity)\n"
+"{\n"
+" int i = get_global_id(0);\n"
+" int pairIndex = i;\n"
+" \n"
+" float4 worldVertsB1[64];\n"
+" float4 worldVertsB2[64];\n"
+" int capacityWorldVerts = 64; \n"
+" float4 localContactsOut[64];\n"
+" int localContactCapacity=64;\n"
+" \n"
+" float minDist = -1e30f;\n"
+" float maxDist = 0.02f;\n"
+" if (i<numPairs)\n"
+" {\n"
+" int bodyIndexA = pairs[i].x;\n"
+" int bodyIndexB = pairs[i].y;\n"
+" \n"
+" int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;\n"
+" int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;\n"
+" if (hasSeparatingAxis[i])\n"
+" {\n"
+" \n"
+" int shapeIndexA = collidables[collidableIndexA].m_shapeIndex;\n"
+" int shapeIndexB = collidables[collidableIndexB].m_shapeIndex;\n"
+" \n"
+" \n"
+" int numLocalContactsOut = clipHullAgainstHull(separatingNormals[i],\n"
+" &convexShapes[shapeIndexA], &convexShapes[shapeIndexB],\n"
+" rigidBodies[bodyIndexA].m_pos,rigidBodies[bodyIndexA].m_quat,\n"
+" rigidBodies[bodyIndexB].m_pos,rigidBodies[bodyIndexB].m_quat,\n"
+" worldVertsB1,worldVertsB2,capacityWorldVerts,\n"
+" minDist, maxDist,\n"
+" vertices,faces,indices,\n"
+" localContactsOut,localContactCapacity);\n"
+" \n"
+" if (numLocalContactsOut>0)\n"
+" {\n"
+" float4 normal = -separatingNormals[i];\n"
+" int nPoints = numLocalContactsOut;\n"
+" float4* pointsIn = localContactsOut;\n"
+" int contactIdx[4];// = {-1,-1,-1,-1};\n"
+" contactIdx[0] = -1;\n"
+" contactIdx[1] = -1;\n"
+" contactIdx[2] = -1;\n"
+" contactIdx[3] = -1;\n"
+" \n"
+" int nReducedContacts = extractManifoldSequential(pointsIn, nPoints, normal, contactIdx);\n"
+" \n"
+" \n"
+" int mprContactIndex = pairs[pairIndex].z;\n"
+" int dstIdx = mprContactIndex;\n"
+" if (dstIdx<0)\n"
+" {\n"
+" AppendInc( nGlobalContactsOut, dstIdx );\n"
+" }\n"
+" if (dstIdx<contactCapacity)\n"
+" {\n"
+" pairs[pairIndex].z = dstIdx;\n"
+" __global struct b3Contact4Data* c = globalContactsOut+ dstIdx;\n"
+" c->m_worldNormalOnB = -normal;\n"
+" c->m_restituitionCoeffCmp = (0.f*0xffff);c->m_frictionCoeffCmp = (0.7f*0xffff);\n"
+" c->m_batchIdx = pairIndex;\n"
+" int bodyA = pairs[pairIndex].x;\n"
+" int bodyB = pairs[pairIndex].y;\n"
+" c->m_bodyAPtrAndSignBit = rigidBodies[bodyA].m_invMass==0?-bodyA:bodyA;\n"
+" c->m_bodyBPtrAndSignBit = rigidBodies[bodyB].m_invMass==0?-bodyB:bodyB;\n"
+" c->m_childIndexA = -1;\n"
+" c->m_childIndexB = -1;\n"
+" for (int i=0;i<nReducedContacts;i++)\n"
+" {\n"
+" //this condition means: overwrite contact point, unless at index i==0 we have a valid 'mpr' contact\n"
+" if (i>0||(mprContactIndex<0))\n"
+" {\n"
+" c->m_worldPosB[i] = pointsIn[contactIdx[i]];\n"
+" }\n"
+" }\n"
+" GET_NPOINTS(*c) = nReducedContacts;\n"
+" }\n"
+" \n"
+" }// if (numContactsOut>0)\n"
+" }// if (hasSeparatingAxis[i])\n"
+" }// if (i<numPairs)\n"
+"}\n"
+"__kernel void clipCompoundsHullHullKernel( __global const int4* gpuCompoundPairs, \n"
+" __global const b3RigidBodyData_t* rigidBodies, \n"
+" __global const b3Collidable_t* collidables,\n"
+" __global const b3ConvexPolyhedronData_t* convexShapes, \n"
+" __global const float4* vertices,\n"
+" __global const float4* uniqueEdges,\n"
+" __global const b3GpuFace_t* faces,\n"
+" __global const int* indices,\n"
+" __global const b3GpuChildShape_t* gpuChildShapes,\n"
+" __global const float4* gpuCompoundSepNormalsOut,\n"
+" __global const int* gpuHasCompoundSepNormalsOut,\n"
+" __global struct b3Contact4Data* restrict globalContactsOut,\n"
+" counter32_t nGlobalContactsOut,\n"
+" int numCompoundPairs, int maxContactCapacity)\n"
+"{\n"
+" int i = get_global_id(0);\n"
+" int pairIndex = i;\n"
+" \n"
+" float4 worldVertsB1[64];\n"
+" float4 worldVertsB2[64];\n"
+" int capacityWorldVerts = 64; \n"
+" float4 localContactsOut[64];\n"
+" int localContactCapacity=64;\n"
+" \n"
+" float minDist = -1e30f;\n"
+" float maxDist = 0.02f;\n"
+" if (i<numCompoundPairs)\n"
+" {\n"
+" if (gpuHasCompoundSepNormalsOut[i])\n"
+" {\n"
+" int bodyIndexA = gpuCompoundPairs[i].x;\n"
+" int bodyIndexB = gpuCompoundPairs[i].y;\n"
+" \n"
+" int childShapeIndexA = gpuCompoundPairs[i].z;\n"
+" int childShapeIndexB = gpuCompoundPairs[i].w;\n"
+" \n"
+" int collidableIndexA = -1;\n"
+" int collidableIndexB = -1;\n"
+" \n"
+" float4 ornA = rigidBodies[bodyIndexA].m_quat;\n"
+" float4 posA = rigidBodies[bodyIndexA].m_pos;\n"
+" \n"
+" float4 ornB = rigidBodies[bodyIndexB].m_quat;\n"
+" float4 posB = rigidBodies[bodyIndexB].m_pos;\n"
+" \n"
+" if (childShapeIndexA >= 0)\n"
+" {\n"
+" collidableIndexA = gpuChildShapes[childShapeIndexA].m_shapeIndex;\n"
+" float4 childPosA = gpuChildShapes[childShapeIndexA].m_childPosition;\n"
+" float4 childOrnA = gpuChildShapes[childShapeIndexA].m_childOrientation;\n"
+" float4 newPosA = qtRotate(ornA,childPosA)+posA;\n"
+" float4 newOrnA = qtMul(ornA,childOrnA);\n"
+" posA = newPosA;\n"
+" ornA = newOrnA;\n"
+" } else\n"
+" {\n"
+" collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;\n"
+" }\n"
+" \n"
+" if (childShapeIndexB>=0)\n"
+" {\n"
+" collidableIndexB = gpuChildShapes[childShapeIndexB].m_shapeIndex;\n"
+" float4 childPosB = gpuChildShapes[childShapeIndexB].m_childPosition;\n"
+" float4 childOrnB = gpuChildShapes[childShapeIndexB].m_childOrientation;\n"
+" float4 newPosB = transform(&childPosB,&posB,&ornB);\n"
+" float4 newOrnB = qtMul(ornB,childOrnB);\n"
+" posB = newPosB;\n"
+" ornB = newOrnB;\n"
+" } else\n"
+" {\n"
+" collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx; \n"
+" }\n"
+" \n"
+" int shapeIndexA = collidables[collidableIndexA].m_shapeIndex;\n"
+" int shapeIndexB = collidables[collidableIndexB].m_shapeIndex;\n"
+" \n"
+" int numLocalContactsOut = clipHullAgainstHull(gpuCompoundSepNormalsOut[i],\n"
+" &convexShapes[shapeIndexA], &convexShapes[shapeIndexB],\n"
+" posA,ornA,\n"
+" posB,ornB,\n"
+" worldVertsB1,worldVertsB2,capacityWorldVerts,\n"
+" minDist, maxDist,\n"
+" vertices,faces,indices,\n"
+" localContactsOut,localContactCapacity);\n"
+" \n"
+" if (numLocalContactsOut>0)\n"
+" {\n"
+" float4 normal = -gpuCompoundSepNormalsOut[i];\n"
+" int nPoints = numLocalContactsOut;\n"
+" float4* pointsIn = localContactsOut;\n"
+" int contactIdx[4];// = {-1,-1,-1,-1};\n"
+" contactIdx[0] = -1;\n"
+" contactIdx[1] = -1;\n"
+" contactIdx[2] = -1;\n"
+" contactIdx[3] = -1;\n"
+" \n"
+" int nReducedContacts = extractManifoldSequential(pointsIn, nPoints, normal, contactIdx);\n"
+" \n"
+" int dstIdx;\n"
+" AppendInc( nGlobalContactsOut, dstIdx );\n"
+" if ((dstIdx+nReducedContacts) < maxContactCapacity)\n"
+" {\n"
+" __global struct b3Contact4Data* c = globalContactsOut+ dstIdx;\n"
+" c->m_worldNormalOnB = -normal;\n"
+" c->m_restituitionCoeffCmp = (0.f*0xffff);c->m_frictionCoeffCmp = (0.7f*0xffff);\n"
+" c->m_batchIdx = pairIndex;\n"
+" int bodyA = gpuCompoundPairs[pairIndex].x;\n"
+" int bodyB = gpuCompoundPairs[pairIndex].y;\n"
+" c->m_bodyAPtrAndSignBit = rigidBodies[bodyA].m_invMass==0?-bodyA:bodyA;\n"
+" c->m_bodyBPtrAndSignBit = rigidBodies[bodyB].m_invMass==0?-bodyB:bodyB;\n"
+" c->m_childIndexA = childShapeIndexA;\n"
+" c->m_childIndexB = childShapeIndexB;\n"
+" for (int i=0;i<nReducedContacts;i++)\n"
+" {\n"
+" c->m_worldPosB[i] = pointsIn[contactIdx[i]];\n"
+" }\n"
+" GET_NPOINTS(*c) = nReducedContacts;\n"
+" }\n"
+" \n"
+" }// if (numContactsOut>0)\n"
+" }// if (gpuHasCompoundSepNormalsOut[i])\n"
+" }// if (i<numCompoundPairs)\n"
+"}\n"
+"__kernel void sphereSphereCollisionKernel( __global const int4* pairs, \n"
+" __global const b3RigidBodyData_t* rigidBodies, \n"
+" __global const b3Collidable_t* collidables,\n"
+" __global const float4* separatingNormals,\n"
+" __global const int* hasSeparatingAxis,\n"
+" __global struct b3Contact4Data* restrict globalContactsOut,\n"
+" counter32_t nGlobalContactsOut,\n"
+" int contactCapacity,\n"
+" int numPairs)\n"
+"{\n"
+" int i = get_global_id(0);\n"
+" int pairIndex = i;\n"
+" \n"
+" if (i<numPairs)\n"
+" {\n"
+" int bodyIndexA = pairs[i].x;\n"
+" int bodyIndexB = pairs[i].y;\n"
+" \n"
+" int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;\n"
+" int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;\n"
+" if (collidables[collidableIndexA].m_shapeType == SHAPE_SPHERE &&\n"
+" collidables[collidableIndexB].m_shapeType == SHAPE_SPHERE)\n"
+" {\n"
+" //sphere-sphere\n"
+" float radiusA = collidables[collidableIndexA].m_radius;\n"
+" float radiusB = collidables[collidableIndexB].m_radius;\n"
+" float4 posA = rigidBodies[bodyIndexA].m_pos;\n"
+" float4 posB = rigidBodies[bodyIndexB].m_pos;\n"
+" float4 diff = posA-posB;\n"
+" float len = length(diff);\n"
+" \n"
+" ///iff distance positive, don't generate a new contact\n"
+" if ( len <= (radiusA+radiusB))\n"
+" {\n"
+" ///distance (negative means penetration)\n"
+" float dist = len - (radiusA+radiusB);\n"
+" float4 normalOnSurfaceB = make_float4(1.f,0.f,0.f,0.f);\n"
+" if (len > 0.00001)\n"
+" {\n"
+" normalOnSurfaceB = diff / len;\n"
+" }\n"
+" float4 contactPosB = posB + normalOnSurfaceB*radiusB;\n"
+" contactPosB.w = dist;\n"
+" \n"
+" int dstIdx;\n"
+" AppendInc( nGlobalContactsOut, dstIdx );\n"
+" if (dstIdx < contactCapacity)\n"
+" {\n"
+" __global struct b3Contact4Data* c = &globalContactsOut[dstIdx];\n"
+" c->m_worldNormalOnB = -normalOnSurfaceB;\n"
+" c->m_restituitionCoeffCmp = (0.f*0xffff);c->m_frictionCoeffCmp = (0.7f*0xffff);\n"
+" c->m_batchIdx = pairIndex;\n"
+" int bodyA = pairs[pairIndex].x;\n"
+" int bodyB = pairs[pairIndex].y;\n"
+" c->m_bodyAPtrAndSignBit = rigidBodies[bodyA].m_invMass==0?-bodyA:bodyA;\n"
+" c->m_bodyBPtrAndSignBit = rigidBodies[bodyB].m_invMass==0?-bodyB:bodyB;\n"
+" c->m_worldPosB[0] = contactPosB;\n"
+" c->m_childIndexA = -1;\n"
+" c->m_childIndexB = -1;\n"
+" GET_NPOINTS(*c) = 1;\n"
+" }//if (dstIdx < numPairs)\n"
+" }//if ( len <= (radiusA+radiusB))\n"
+" }//SHAPE_SPHERE SHAPE_SPHERE\n"
+" }//if (i<numPairs)\n"
+"} \n"
+"__kernel void clipHullHullConcaveConvexKernel( __global int4* concavePairsIn,\n"
+" __global const b3RigidBodyData_t* rigidBodies, \n"
+" __global const b3Collidable_t* collidables,\n"
+" __global const b3ConvexPolyhedronData_t* convexShapes, \n"
+" __global const float4* vertices,\n"
+" __global const float4* uniqueEdges,\n"
+" __global const b3GpuFace_t* faces,\n"
+" __global const int* indices,\n"
+" __global const b3GpuChildShape_t* gpuChildShapes,\n"
+" __global const float4* separatingNormals,\n"
+" __global struct b3Contact4Data* restrict globalContactsOut,\n"
+" counter32_t nGlobalContactsOut,\n"
+" int contactCapacity,\n"
+" int numConcavePairs)\n"
+"{\n"
+" int i = get_global_id(0);\n"
+" int pairIndex = i;\n"
+" \n"
+" float4 worldVertsB1[64];\n"
+" float4 worldVertsB2[64];\n"
+" int capacityWorldVerts = 64; \n"
+" float4 localContactsOut[64];\n"
+" int localContactCapacity=64;\n"
+" \n"
+" float minDist = -1e30f;\n"
+" float maxDist = 0.02f;\n"
+" if (i<numConcavePairs)\n"
+" {\n"
+" //negative value means that the pair is invalid\n"
+" if (concavePairsIn[i].w<0)\n"
+" return;\n"
+" int bodyIndexA = concavePairsIn[i].x;\n"
+" int bodyIndexB = concavePairsIn[i].y;\n"
+" int f = concavePairsIn[i].z;\n"
+" int childShapeIndexA = f;\n"
+" \n"
+" int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;\n"
+" int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;\n"
+" \n"
+" int shapeIndexA = collidables[collidableIndexA].m_shapeIndex;\n"
+" int shapeIndexB = collidables[collidableIndexB].m_shapeIndex;\n"
+" \n"
+" ///////////////////////////////////////////////////////////////\n"
+" \n"
+" \n"
+" bool overlap = false;\n"
+" \n"
+" b3ConvexPolyhedronData_t convexPolyhedronA;\n"
+" //add 3 vertices of the triangle\n"
+" convexPolyhedronA.m_numVertices = 3;\n"
+" convexPolyhedronA.m_vertexOffset = 0;\n"
+" float4 localCenter = make_float4(0.f,0.f,0.f,0.f);\n"
+" b3GpuFace_t face = faces[convexShapes[shapeIndexA].m_faceOffset+f];\n"
+" \n"
+" float4 verticesA[3];\n"
+" for (int i=0;i<3;i++)\n"
+" {\n"
+" int index = indices[face.m_indexOffset+i];\n"
+" float4 vert = vertices[convexShapes[shapeIndexA].m_vertexOffset+index];\n"
+" verticesA[i] = vert;\n"
+" localCenter += vert;\n"
+" }\n"
+" float dmin = FLT_MAX;\n"
+" int localCC=0;\n"
+" //a triangle has 3 unique edges\n"
+" convexPolyhedronA.m_numUniqueEdges = 3;\n"
+" convexPolyhedronA.m_uniqueEdgesOffset = 0;\n"
+" float4 uniqueEdgesA[3];\n"
+" \n"
+" uniqueEdgesA[0] = (verticesA[1]-verticesA[0]);\n"
+" uniqueEdgesA[1] = (verticesA[2]-verticesA[1]);\n"
+" uniqueEdgesA[2] = (verticesA[0]-verticesA[2]);\n"
+" convexPolyhedronA.m_faceOffset = 0;\n"
+" \n"
+" float4 normal = make_float4(face.m_plane.x,face.m_plane.y,face.m_plane.z,0.f);\n"
+" \n"
+" b3GpuFace_t facesA[TRIANGLE_NUM_CONVEX_FACES];\n"
+" int indicesA[3+3+2+2+2];\n"
+" int curUsedIndices=0;\n"
+" int fidx=0;\n"
+" //front size of triangle\n"
+" {\n"
+" facesA[fidx].m_indexOffset=curUsedIndices;\n"
+" indicesA[0] = 0;\n"
+" indicesA[1] = 1;\n"
+" indicesA[2] = 2;\n"
+" curUsedIndices+=3;\n"
+" float c = face.m_plane.w;\n"
+" facesA[fidx].m_plane.x = normal.x;\n"
+" facesA[fidx].m_plane.y = normal.y;\n"
+" facesA[fidx].m_plane.z = normal.z;\n"
+" facesA[fidx].m_plane.w = c;\n"
+" facesA[fidx].m_numIndices=3;\n"
+" }\n"
+" fidx++;\n"
+" //back size of triangle\n"
+" {\n"
+" facesA[fidx].m_indexOffset=curUsedIndices;\n"
+" indicesA[3]=2;\n"
+" indicesA[4]=1;\n"
+" indicesA[5]=0;\n"
+" curUsedIndices+=3;\n"
+" float c = dot3F4(normal,verticesA[0]);\n"
+" float c1 = -face.m_plane.w;\n"
+" facesA[fidx].m_plane.x = -normal.x;\n"
+" facesA[fidx].m_plane.y = -normal.y;\n"
+" facesA[fidx].m_plane.z = -normal.z;\n"
+" facesA[fidx].m_plane.w = c;\n"
+" facesA[fidx].m_numIndices=3;\n"
+" }\n"
+" fidx++;\n"
+" bool addEdgePlanes = true;\n"
+" if (addEdgePlanes)\n"
+" {\n"
+" int numVertices=3;\n"
+" int prevVertex = numVertices-1;\n"
+" for (int i=0;i<numVertices;i++)\n"
+" {\n"
+" float4 v0 = verticesA[i];\n"
+" float4 v1 = verticesA[prevVertex];\n"
+" \n"
+" float4 edgeNormal = normalize(cross(normal,v1-v0));\n"
+" float c = -dot3F4(edgeNormal,v0);\n"
+" facesA[fidx].m_numIndices = 2;\n"
+" facesA[fidx].m_indexOffset=curUsedIndices;\n"
+" indicesA[curUsedIndices++]=i;\n"
+" indicesA[curUsedIndices++]=prevVertex;\n"
+" \n"
+" facesA[fidx].m_plane.x = edgeNormal.x;\n"
+" facesA[fidx].m_plane.y = edgeNormal.y;\n"
+" facesA[fidx].m_plane.z = edgeNormal.z;\n"
+" facesA[fidx].m_plane.w = c;\n"
+" fidx++;\n"
+" prevVertex = i;\n"
+" }\n"
+" }\n"
+" convexPolyhedronA.m_numFaces = TRIANGLE_NUM_CONVEX_FACES;\n"
+" convexPolyhedronA.m_localCenter = localCenter*(1.f/3.f);\n"
+" float4 posA = rigidBodies[bodyIndexA].m_pos;\n"
+" posA.w = 0.f;\n"
+" float4 posB = rigidBodies[bodyIndexB].m_pos;\n"
+" posB.w = 0.f;\n"
+" float4 ornA = rigidBodies[bodyIndexA].m_quat;\n"
+" float4 ornB =rigidBodies[bodyIndexB].m_quat;\n"
+" float4 sepAxis = separatingNormals[i];\n"
+" \n"
+" int shapeTypeB = collidables[collidableIndexB].m_shapeType;\n"
+" int childShapeIndexB =-1;\n"
+" if (shapeTypeB==SHAPE_COMPOUND_OF_CONVEX_HULLS)\n"
+" {\n"
+" ///////////////////\n"
+" ///compound shape support\n"
+" \n"
+" childShapeIndexB = concavePairsIn[pairIndex].w;\n"
+" int childColIndexB = gpuChildShapes[childShapeIndexB].m_shapeIndex;\n"
+" shapeIndexB = collidables[childColIndexB].m_shapeIndex;\n"
+" float4 childPosB = gpuChildShapes[childShapeIndexB].m_childPosition;\n"
+" float4 childOrnB = gpuChildShapes[childShapeIndexB].m_childOrientation;\n"
+" float4 newPosB = transform(&childPosB,&posB,&ornB);\n"
+" float4 newOrnB = qtMul(ornB,childOrnB);\n"
+" posB = newPosB;\n"
+" ornB = newOrnB;\n"
+" \n"
+" }\n"
+" \n"
+" ////////////////////////////////////////\n"
+" \n"
+" \n"
+" \n"
+" int numLocalContactsOut = clipHullAgainstHullLocalA(sepAxis,\n"
+" &convexPolyhedronA, &convexShapes[shapeIndexB],\n"
+" posA,ornA,\n"
+" posB,ornB,\n"
+" worldVertsB1,worldVertsB2,capacityWorldVerts,\n"
+" minDist, maxDist,\n"
+" &verticesA,&facesA,&indicesA,\n"
+" vertices,faces,indices,\n"
+" localContactsOut,localContactCapacity);\n"
+" \n"
+" if (numLocalContactsOut>0)\n"
+" {\n"
+" float4 normal = -separatingNormals[i];\n"
+" int nPoints = numLocalContactsOut;\n"
+" float4* pointsIn = localContactsOut;\n"
+" int contactIdx[4];// = {-1,-1,-1,-1};\n"
+" contactIdx[0] = -1;\n"
+" contactIdx[1] = -1;\n"
+" contactIdx[2] = -1;\n"
+" contactIdx[3] = -1;\n"
+" \n"
+" int nReducedContacts = extractManifoldSequential(pointsIn, nPoints, normal, contactIdx);\n"
+" \n"
+" int dstIdx;\n"
+" AppendInc( nGlobalContactsOut, dstIdx );\n"
+" if (dstIdx<contactCapacity)\n"
+" {\n"
+" __global struct b3Contact4Data* c = globalContactsOut+ dstIdx;\n"
+" c->m_worldNormalOnB = -normal;\n"
+" c->m_restituitionCoeffCmp = (0.f*0xffff);c->m_frictionCoeffCmp = (0.7f*0xffff);\n"
+" c->m_batchIdx = pairIndex;\n"
+" int bodyA = concavePairsIn[pairIndex].x;\n"
+" int bodyB = concavePairsIn[pairIndex].y;\n"
+" c->m_bodyAPtrAndSignBit = rigidBodies[bodyA].m_invMass==0?-bodyA:bodyA;\n"
+" c->m_bodyBPtrAndSignBit = rigidBodies[bodyB].m_invMass==0?-bodyB:bodyB;\n"
+" c->m_childIndexA = childShapeIndexA;\n"
+" c->m_childIndexB = childShapeIndexB;\n"
+" for (int i=0;i<nReducedContacts;i++)\n"
+" {\n"
+" c->m_worldPosB[i] = pointsIn[contactIdx[i]];\n"
+" }\n"
+" GET_NPOINTS(*c) = nReducedContacts;\n"
+" }\n"
+" \n"
+" }// if (numContactsOut>0)\n"
+" }// if (i<numPairs)\n"
+"}\n"
+"int findClippingFaces(const float4 separatingNormal,\n"
+" __global const b3ConvexPolyhedronData_t* hullA, __global const b3ConvexPolyhedronData_t* hullB,\n"
+" const float4 posA, const Quaternion ornA,const float4 posB, const Quaternion ornB,\n"
+" __global float4* worldVertsA1,\n"
+" __global float4* worldNormalsA1,\n"
+" __global float4* worldVertsB1,\n"
+" int capacityWorldVerts,\n"
+" const float minDist, float maxDist,\n"
+" __global const float4* vertices,\n"
+" __global const b3GpuFace_t* faces,\n"
+" __global const int* indices,\n"
+" __global int4* clippingFaces, int pairIndex)\n"
+"{\n"
+" int numContactsOut = 0;\n"
+" int numWorldVertsB1= 0;\n"
+" \n"
+" \n"
+" int closestFaceB=-1;\n"
+" float dmax = -FLT_MAX;\n"
+" \n"
+" {\n"
+" for(int face=0;face<hullB->m_numFaces;face++)\n"
+" {\n"
+" const float4 Normal = make_float4(faces[hullB->m_faceOffset+face].m_plane.x,\n"
+" faces[hullB->m_faceOffset+face].m_plane.y, faces[hullB->m_faceOffset+face].m_plane.z,0.f);\n"
+" const float4 WorldNormal = qtRotate(ornB, Normal);\n"
+" float d = dot3F4(WorldNormal,separatingNormal);\n"
+" if (d > dmax)\n"
+" {\n"
+" dmax = d;\n"
+" closestFaceB = face;\n"
+" }\n"
+" }\n"
+" }\n"
+" \n"
+" {\n"
+" const b3GpuFace_t polyB = faces[hullB->m_faceOffset+closestFaceB];\n"
+" const int numVertices = polyB.m_numIndices;\n"
+" for(int e0=0;e0<numVertices;e0++)\n"
+" {\n"
+" const float4 b = vertices[hullB->m_vertexOffset+indices[polyB.m_indexOffset+e0]];\n"
+" worldVertsB1[pairIndex*capacityWorldVerts+numWorldVertsB1++] = transform(&b,&posB,&ornB);\n"
+" }\n"
+" }\n"
+" \n"
+" int closestFaceA=-1;\n"
+" {\n"
+" float dmin = FLT_MAX;\n"
+" for(int face=0;face<hullA->m_numFaces;face++)\n"
+" {\n"
+" const float4 Normal = make_float4(\n"
+" faces[hullA->m_faceOffset+face].m_plane.x,\n"
+" faces[hullA->m_faceOffset+face].m_plane.y,\n"
+" faces[hullA->m_faceOffset+face].m_plane.z,\n"
+" 0.f);\n"
+" const float4 faceANormalWS = qtRotate(ornA,Normal);\n"
+" \n"
+" float d = dot3F4(faceANormalWS,separatingNormal);\n"
+" if (d < dmin)\n"
+" {\n"
+" dmin = d;\n"
+" closestFaceA = face;\n"
+" worldNormalsA1[pairIndex] = faceANormalWS;\n"
+" }\n"
+" }\n"
+" }\n"
+" \n"
+" int numVerticesA = faces[hullA->m_faceOffset+closestFaceA].m_numIndices;\n"
+" for(int e0=0;e0<numVerticesA;e0++)\n"
+" {\n"
+" const float4 a = vertices[hullA->m_vertexOffset+indices[faces[hullA->m_faceOffset+closestFaceA].m_indexOffset+e0]];\n"
+" worldVertsA1[pairIndex*capacityWorldVerts+e0] = transform(&a, &posA,&ornA);\n"
+" }\n"
+" \n"
+" clippingFaces[pairIndex].x = closestFaceA;\n"
+" clippingFaces[pairIndex].y = closestFaceB;\n"
+" clippingFaces[pairIndex].z = numVerticesA;\n"
+" clippingFaces[pairIndex].w = numWorldVertsB1;\n"
+" \n"
+" \n"
+" return numContactsOut;\n"
+"}\n"
+"int clipFaces(__global float4* worldVertsA1,\n"
+" __global float4* worldNormalsA1,\n"
+" __global float4* worldVertsB1,\n"
+" __global float4* worldVertsB2, \n"
+" int capacityWorldVertsB2,\n"
+" const float minDist, float maxDist,\n"
+" __global int4* clippingFaces,\n"
+" int pairIndex)\n"
+"{\n"
+" int numContactsOut = 0;\n"
+" \n"
+" int closestFaceA = clippingFaces[pairIndex].x;\n"
+" int closestFaceB = clippingFaces[pairIndex].y;\n"
+" int numVertsInA = clippingFaces[pairIndex].z;\n"
+" int numVertsInB = clippingFaces[pairIndex].w;\n"
+" \n"
+" int numVertsOut = 0;\n"
+" \n"
+" if (closestFaceA<0)\n"
+" return numContactsOut;\n"
+" \n"
+" __global float4* pVtxIn = &worldVertsB1[pairIndex*capacityWorldVertsB2];\n"
+" __global float4* pVtxOut = &worldVertsB2[pairIndex*capacityWorldVertsB2];\n"
+" \n"
+" \n"
+" \n"
+" // clip polygon to back of planes of all faces of hull A that are adjacent to witness face\n"
+" \n"
+" for(int e0=0;e0<numVertsInA;e0++)\n"
+" {\n"
+" const float4 aw = worldVertsA1[pairIndex*capacityWorldVertsB2+e0];\n"
+" const float4 bw = worldVertsA1[pairIndex*capacityWorldVertsB2+((e0+1)%numVertsInA)];\n"
+" const float4 WorldEdge0 = aw - bw;\n"
+" float4 worldPlaneAnormal1 = worldNormalsA1[pairIndex];\n"
+" float4 planeNormalWS1 = -cross3(WorldEdge0,worldPlaneAnormal1);\n"
+" float4 worldA1 = aw;\n"
+" float planeEqWS1 = -dot3F4(worldA1,planeNormalWS1);\n"
+" float4 planeNormalWS = planeNormalWS1;\n"
+" float planeEqWS=planeEqWS1;\n"
+" numVertsOut = clipFaceGlobal(pVtxIn, numVertsInB, planeNormalWS,planeEqWS, pVtxOut);\n"
+" __global float4* tmp = pVtxOut;\n"
+" pVtxOut = pVtxIn;\n"
+" pVtxIn = tmp;\n"
+" numVertsInB = numVertsOut;\n"
+" numVertsOut = 0;\n"
+" }\n"
+" \n"
+" //float4 planeNormalWS = worldNormalsA1[pairIndex];\n"
+" //float planeEqWS=-dot3F4(planeNormalWS,worldVertsA1[pairIndex*capacityWorldVertsB2]);\n"
+" \n"
+" /*for (int i=0;i<numVertsInB;i++)\n"
+" {\n"
+" pVtxOut[i] = pVtxIn[i];\n"
+" }*/\n"
+" \n"
+" \n"
+" \n"
+" \n"
+" //numVertsInB=0;\n"
+" \n"
+" float4 planeNormalWS = worldNormalsA1[pairIndex];\n"
+" float planeEqWS=-dot3F4(planeNormalWS,worldVertsA1[pairIndex*capacityWorldVertsB2]);\n"
+" for (int i=0;i<numVertsInB;i++)\n"
+" {\n"
+" float depth = dot3F4(planeNormalWS,pVtxIn[i])+planeEqWS;\n"
+" if (depth <=minDist)\n"
+" {\n"
+" depth = minDist;\n"
+" }\n"
+" \n"
+" if (depth <=maxDist)\n"
+" {\n"
+" float4 pointInWorld = pVtxIn[i];\n"
+" pVtxOut[numContactsOut++] = make_float4(pointInWorld.x,pointInWorld.y,pointInWorld.z,depth);\n"
+" }\n"
+" }\n"
+" \n"
+" clippingFaces[pairIndex].w =numContactsOut;\n"
+" \n"
+" \n"
+" return numContactsOut;\n"
+"}\n"
+"__kernel void findClippingFacesKernel( __global const int4* pairs,\n"
+" __global const b3RigidBodyData_t* rigidBodies,\n"
+" __global const b3Collidable_t* collidables,\n"
+" __global const b3ConvexPolyhedronData_t* convexShapes,\n"
+" __global const float4* vertices,\n"
+" __global const float4* uniqueEdges,\n"
+" __global const b3GpuFace_t* faces,\n"
+" __global const int* indices,\n"
+" __global const float4* separatingNormals,\n"
+" __global const int* hasSeparatingAxis,\n"
+" __global int4* clippingFacesOut,\n"
+" __global float4* worldVertsA1,\n"
+" __global float4* worldNormalsA1,\n"
+" __global float4* worldVertsB1,\n"
+" int capacityWorldVerts,\n"
+" int numPairs\n"
+" )\n"
+"{\n"
+" \n"
+" int i = get_global_id(0);\n"
+" int pairIndex = i;\n"
+" \n"
+" \n"
+" float minDist = -1e30f;\n"
+" float maxDist = 0.02f;\n"
+" \n"
+" if (i<numPairs)\n"
+" {\n"
+" \n"
+" if (hasSeparatingAxis[i])\n"
+" {\n"
+" \n"
+" int bodyIndexA = pairs[i].x;\n"
+" int bodyIndexB = pairs[i].y;\n"
+" \n"
+" int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;\n"
+" int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;\n"
+" \n"
+" int shapeIndexA = collidables[collidableIndexA].m_shapeIndex;\n"
+" int shapeIndexB = collidables[collidableIndexB].m_shapeIndex;\n"
+" \n"
+" \n"
+" \n"
+" int numLocalContactsOut = findClippingFaces(separatingNormals[i],\n"
+" &convexShapes[shapeIndexA], &convexShapes[shapeIndexB],\n"
+" rigidBodies[bodyIndexA].m_pos,rigidBodies[bodyIndexA].m_quat,\n"
+" rigidBodies[bodyIndexB].m_pos,rigidBodies[bodyIndexB].m_quat,\n"
+" worldVertsA1,\n"
+" worldNormalsA1,\n"
+" worldVertsB1,capacityWorldVerts,\n"
+" minDist, maxDist,\n"
+" vertices,faces,indices,\n"
+" clippingFacesOut,i);\n"
+" \n"
+" \n"
+" }// if (hasSeparatingAxis[i])\n"
+" }// if (i<numPairs)\n"
+" \n"
+"}\n"
+"__kernel void clipFacesAndFindContactsKernel( __global const float4* separatingNormals,\n"
+" __global const int* hasSeparatingAxis,\n"
+" __global int4* clippingFacesOut,\n"
+" __global float4* worldVertsA1,\n"
+" __global float4* worldNormalsA1,\n"
+" __global float4* worldVertsB1,\n"
+" __global float4* worldVertsB2,\n"
+" int vertexFaceCapacity,\n"
+" int numPairs,\n"
+" int debugMode\n"
+" )\n"
+"{\n"
+" int i = get_global_id(0);\n"
+" int pairIndex = i;\n"
+" \n"
+" \n"
+" float minDist = -1e30f;\n"
+" float maxDist = 0.02f;\n"
+" \n"
+" if (i<numPairs)\n"
+" {\n"
+" \n"
+" if (hasSeparatingAxis[i])\n"
+" {\n"
+" \n"
+"// int bodyIndexA = pairs[i].x;\n"
+" // int bodyIndexB = pairs[i].y;\n"
+" \n"
+" int numLocalContactsOut = 0;\n"
+" int capacityWorldVertsB2 = vertexFaceCapacity;\n"
+" \n"
+" __global float4* pVtxIn = &worldVertsB1[pairIndex*capacityWorldVertsB2];\n"
+" __global float4* pVtxOut = &worldVertsB2[pairIndex*capacityWorldVertsB2];\n"
+" \n"
+" {\n"
+" __global int4* clippingFaces = clippingFacesOut;\n"
+" \n"
+" \n"
+" int closestFaceA = clippingFaces[pairIndex].x;\n"
+" int closestFaceB = clippingFaces[pairIndex].y;\n"
+" int numVertsInA = clippingFaces[pairIndex].z;\n"
+" int numVertsInB = clippingFaces[pairIndex].w;\n"
+" \n"
+" int numVertsOut = 0;\n"
+" \n"
+" if (closestFaceA>=0)\n"
+" {\n"
+" \n"
+" \n"
+" \n"
+" // clip polygon to back of planes of all faces of hull A that are adjacent to witness face\n"
+" \n"
+" for(int e0=0;e0<numVertsInA;e0++)\n"
+" {\n"
+" const float4 aw = worldVertsA1[pairIndex*capacityWorldVertsB2+e0];\n"
+" const float4 bw = worldVertsA1[pairIndex*capacityWorldVertsB2+((e0+1)%numVertsInA)];\n"
+" const float4 WorldEdge0 = aw - bw;\n"
+" float4 worldPlaneAnormal1 = worldNormalsA1[pairIndex];\n"
+" float4 planeNormalWS1 = -cross3(WorldEdge0,worldPlaneAnormal1);\n"
+" float4 worldA1 = aw;\n"
+" float planeEqWS1 = -dot3F4(worldA1,planeNormalWS1);\n"
+" float4 planeNormalWS = planeNormalWS1;\n"
+" float planeEqWS=planeEqWS1;\n"
+" numVertsOut = clipFaceGlobal(pVtxIn, numVertsInB, planeNormalWS,planeEqWS, pVtxOut);\n"
+" __global float4* tmp = pVtxOut;\n"
+" pVtxOut = pVtxIn;\n"
+" pVtxIn = tmp;\n"
+" numVertsInB = numVertsOut;\n"
+" numVertsOut = 0;\n"
+" }\n"
+" \n"
+" float4 planeNormalWS = worldNormalsA1[pairIndex];\n"
+" float planeEqWS=-dot3F4(planeNormalWS,worldVertsA1[pairIndex*capacityWorldVertsB2]);\n"
+" \n"
+" for (int i=0;i<numVertsInB;i++)\n"
+" {\n"
+" float depth = dot3F4(planeNormalWS,pVtxIn[i])+planeEqWS;\n"
+" if (depth <=minDist)\n"
+" {\n"
+" depth = minDist;\n"
+" }\n"
+" \n"
+" if (depth <=maxDist)\n"
+" {\n"
+" float4 pointInWorld = pVtxIn[i];\n"
+" pVtxOut[numLocalContactsOut++] = make_float4(pointInWorld.x,pointInWorld.y,pointInWorld.z,depth);\n"
+" }\n"
+" }\n"
+" \n"
+" }\n"
+" clippingFaces[pairIndex].w =numLocalContactsOut;\n"
+" \n"
+" }\n"
+" \n"
+" for (int i=0;i<numLocalContactsOut;i++)\n"
+" pVtxIn[i] = pVtxOut[i];\n"
+" \n"
+" }// if (hasSeparatingAxis[i])\n"
+" }// if (i<numPairs)\n"
+" \n"
+"}\n"
+"__kernel void newContactReductionKernel( __global int4* pairs,\n"
+" __global const b3RigidBodyData_t* rigidBodies,\n"
+" __global const float4* separatingNormals,\n"
+" __global const int* hasSeparatingAxis,\n"
+" __global struct b3Contact4Data* globalContactsOut,\n"
+" __global int4* clippingFaces,\n"
+" __global float4* worldVertsB2,\n"
+" volatile __global int* nGlobalContactsOut,\n"
+" int vertexFaceCapacity,\n"
+" int contactCapacity,\n"
+" int numPairs\n"
+" )\n"
+"{\n"
+" int i = get_global_id(0);\n"
+" int pairIndex = i;\n"
+" \n"
+" int4 contactIdx;\n"
+" contactIdx=make_int4(0,1,2,3);\n"
+" \n"
+" if (i<numPairs)\n"
+" {\n"
+" \n"
+" if (hasSeparatingAxis[i])\n"
+" {\n"
+" \n"
+" \n"
+" \n"
+" \n"
+" int nPoints = clippingFaces[pairIndex].w;\n"
+" \n"
+" if (nPoints>0)\n"
+" {\n"
+" __global float4* pointsIn = &worldVertsB2[pairIndex*vertexFaceCapacity];\n"
+" float4 normal = -separatingNormals[i];\n"
+" \n"
+" int nReducedContacts = extractManifoldSequentialGlobal(pointsIn, nPoints, normal, &contactIdx);\n"
+" \n"
+" int mprContactIndex = pairs[pairIndex].z;\n"
+" int dstIdx = mprContactIndex;\n"
+" if (dstIdx<0)\n"
+" {\n"
+" AppendInc( nGlobalContactsOut, dstIdx );\n"
+" }\n"
+"//#if 0\n"
+" \n"
+" if (dstIdx < contactCapacity)\n"
+" {\n"
+" __global struct b3Contact4Data* c = &globalContactsOut[dstIdx];\n"
+" c->m_worldNormalOnB = -normal;\n"
+" c->m_restituitionCoeffCmp = (0.f*0xffff);c->m_frictionCoeffCmp = (0.7f*0xffff);\n"
+" c->m_batchIdx = pairIndex;\n"
+" int bodyA = pairs[pairIndex].x;\n"
+" int bodyB = pairs[pairIndex].y;\n"
+" pairs[pairIndex].w = dstIdx;\n"
+" c->m_bodyAPtrAndSignBit = rigidBodies[bodyA].m_invMass==0?-bodyA:bodyA;\n"
+" c->m_bodyBPtrAndSignBit = rigidBodies[bodyB].m_invMass==0?-bodyB:bodyB;\n"
+" c->m_childIndexA =-1;\n"
+" c->m_childIndexB =-1;\n"
+" switch (nReducedContacts)\n"
+" {\n"
+" case 4:\n"
+" c->m_worldPosB[3] = pointsIn[contactIdx.w];\n"
+" case 3:\n"
+" c->m_worldPosB[2] = pointsIn[contactIdx.z];\n"
+" case 2:\n"
+" c->m_worldPosB[1] = pointsIn[contactIdx.y];\n"
+" case 1:\n"
+" if (mprContactIndex<0)//test\n"
+" c->m_worldPosB[0] = pointsIn[contactIdx.x];\n"
+" default:\n"
+" {\n"
+" }\n"
+" };\n"
+" \n"
+" GET_NPOINTS(*c) = nReducedContacts;\n"
+" \n"
+" }\n"
+" \n"
+" \n"
+"//#endif\n"
+" \n"
+" }// if (numContactsOut>0)\n"
+" }// if (hasSeparatingAxis[i])\n"
+" }// if (i<numPairs)\n"
+" \n"
+" \n"
+"}\n"
+;
diff --git a/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/satConcave.cl b/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/satConcave.cl
new file mode 100644
index 0000000000..31ca43b8cd
--- /dev/null
+++ b/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/satConcave.cl
@@ -0,0 +1,1220 @@
+
+//keep this enum in sync with the CPU version (in btCollidable.h)
+//written by Erwin Coumans
+
+
+#define SHAPE_CONVEX_HULL 3
+#define SHAPE_CONCAVE_TRIMESH 5
+#define TRIANGLE_NUM_CONVEX_FACES 5
+#define SHAPE_COMPOUND_OF_CONVEX_HULLS 6
+
+#define B3_MAX_STACK_DEPTH 256
+
+
+typedef unsigned int u32;
+
+///keep this in sync with btCollidable.h
+typedef struct
+{
+ union {
+ int m_numChildShapes;
+ int m_bvhIndex;
+ };
+ union
+ {
+ float m_radius;
+ int m_compoundBvhIndex;
+ };
+
+ int m_shapeType;
+ int m_shapeIndex;
+
+} btCollidableGpu;
+
+#define MAX_NUM_PARTS_IN_BITS 10
+
+///b3QuantizedBvhNode 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).
+typedef struct
+{
+ //12 bytes
+ unsigned short int m_quantizedAabbMin[3];
+ unsigned short int m_quantizedAabbMax[3];
+ //4 bytes
+ int m_escapeIndexOrTriangleIndex;
+} b3QuantizedBvhNode;
+
+typedef struct
+{
+ float4 m_aabbMin;
+ float4 m_aabbMax;
+ float4 m_quantization;
+ int m_numNodes;
+ int m_numSubTrees;
+ int m_nodeOffset;
+ int m_subTreeOffset;
+
+} b3BvhInfo;
+
+
+int getTriangleIndex(const b3QuantizedBvhNode* rootNode)
+{
+ 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 (rootNode->m_escapeIndexOrTriangleIndex&~(y));
+}
+
+int getTriangleIndexGlobal(__global const b3QuantizedBvhNode* rootNode)
+{
+ 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 (rootNode->m_escapeIndexOrTriangleIndex&~(y));
+}
+
+int isLeafNode(const b3QuantizedBvhNode* rootNode)
+{
+ //skipindex is negative (internal node), triangleindex >=0 (leafnode)
+ return (rootNode->m_escapeIndexOrTriangleIndex >= 0)? 1 : 0;
+}
+
+int isLeafNodeGlobal(__global const b3QuantizedBvhNode* rootNode)
+{
+ //skipindex is negative (internal node), triangleindex >=0 (leafnode)
+ return (rootNode->m_escapeIndexOrTriangleIndex >= 0)? 1 : 0;
+}
+
+int getEscapeIndex(const b3QuantizedBvhNode* rootNode)
+{
+ return -rootNode->m_escapeIndexOrTriangleIndex;
+}
+
+int getEscapeIndexGlobal(__global const b3QuantizedBvhNode* rootNode)
+{
+ return -rootNode->m_escapeIndexOrTriangleIndex;
+}
+
+
+typedef struct
+{
+ //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];
+} b3BvhSubtreeInfo;
+
+
+
+
+
+
+
+typedef struct
+{
+ float4 m_childPosition;
+ float4 m_childOrientation;
+ int m_shapeIndex;
+ int m_unused0;
+ int m_unused1;
+ int m_unused2;
+} btGpuChildShape;
+
+
+typedef struct
+{
+ float4 m_pos;
+ float4 m_quat;
+ float4 m_linVel;
+ float4 m_angVel;
+
+ u32 m_collidableIdx;
+ float m_invMass;
+ float m_restituitionCoeff;
+ float m_frictionCoeff;
+} BodyData;
+
+
+typedef struct
+{
+ float4 m_localCenter;
+ float4 m_extents;
+ float4 mC;
+ float4 mE;
+
+ float m_radius;
+ int m_faceOffset;
+ int m_numFaces;
+ int m_numVertices;
+
+ int m_vertexOffset;
+ int m_uniqueEdgesOffset;
+ int m_numUniqueEdges;
+ int m_unused;
+} ConvexPolyhedronCL;
+
+typedef struct
+{
+ union
+ {
+ float4 m_min;
+ float m_minElems[4];
+ int m_minIndices[4];
+ };
+ union
+ {
+ float4 m_max;
+ float m_maxElems[4];
+ int m_maxIndices[4];
+ };
+} btAabbCL;
+
+#include "Bullet3Collision/BroadPhaseCollision/shared/b3Aabb.h"
+#include "Bullet3Common/shared/b3Int2.h"
+
+
+
+typedef struct
+{
+ float4 m_plane;
+ int m_indexOffset;
+ int m_numIndices;
+} btGpuFace;
+
+#define make_float4 (float4)
+
+
+__inline
+float4 cross3(float4 a, float4 b)
+{
+ return cross(a,b);
+
+
+// float4 a1 = make_float4(a.xyz,0.f);
+// float4 b1 = make_float4(b.xyz,0.f);
+
+// return cross(a1,b1);
+
+//float4 c = make_float4(a.y*b.z - a.z*b.y,a.z*b.x - a.x*b.z,a.x*b.y - a.y*b.x,0.f);
+
+ // float4 c = make_float4(a.y*b.z - a.z*b.y,1.f,a.x*b.y - a.y*b.x,0.f);
+
+ //return c;
+}
+
+__inline
+float dot3F4(float4 a, float4 b)
+{
+ float4 a1 = make_float4(a.xyz,0.f);
+ float4 b1 = make_float4(b.xyz,0.f);
+ return dot(a1, b1);
+}
+
+__inline
+float4 fastNormalize4(float4 v)
+{
+ v = make_float4(v.xyz,0.f);
+ return fast_normalize(v);
+}
+
+
+///////////////////////////////////////
+// Quaternion
+///////////////////////////////////////
+
+typedef float4 Quaternion;
+
+__inline
+Quaternion qtMul(Quaternion a, Quaternion b);
+
+__inline
+Quaternion qtNormalize(Quaternion in);
+
+__inline
+float4 qtRotate(Quaternion q, float4 vec);
+
+__inline
+Quaternion qtInvert(Quaternion q);
+
+
+
+
+__inline
+Quaternion qtMul(Quaternion a, Quaternion b)
+{
+ Quaternion ans;
+ ans = cross3( a, b );
+ ans += a.w*b+b.w*a;
+// ans.w = a.w*b.w - (a.x*b.x+a.y*b.y+a.z*b.z);
+ ans.w = a.w*b.w - dot3F4(a, b);
+ return ans;
+}
+
+__inline
+Quaternion qtNormalize(Quaternion in)
+{
+ return fastNormalize4(in);
+// in /= length( in );
+// return in;
+}
+__inline
+float4 qtRotate(Quaternion q, float4 vec)
+{
+ Quaternion qInv = qtInvert( q );
+ float4 vcpy = vec;
+ vcpy.w = 0.f;
+ float4 out = qtMul(qtMul(q,vcpy),qInv);
+ return out;
+}
+
+__inline
+Quaternion qtInvert(Quaternion q)
+{
+ return (Quaternion)(-q.xyz, q.w);
+}
+
+__inline
+float4 qtInvRotate(const Quaternion q, float4 vec)
+{
+ return qtRotate( qtInvert( q ), vec );
+}
+
+__inline
+float4 transform(const float4* p, const float4* translation, const Quaternion* orientation)
+{
+ return qtRotate( *orientation, *p ) + (*translation);
+}
+
+
+
+__inline
+float4 normalize3(const float4 a)
+{
+ float4 n = make_float4(a.x, a.y, a.z, 0.f);
+ return fastNormalize4( n );
+}
+
+inline void projectLocal(const ConvexPolyhedronCL* hull, const float4 pos, const float4 orn,
+const float4* dir, const float4* vertices, float* min, float* max)
+{
+ min[0] = FLT_MAX;
+ max[0] = -FLT_MAX;
+ int numVerts = hull->m_numVertices;
+
+ const float4 localDir = qtInvRotate(orn,*dir);
+ float offset = dot(pos,*dir);
+ for(int i=0;i<numVerts;i++)
+ {
+ float dp = dot(vertices[hull->m_vertexOffset+i],localDir);
+ if(dp < min[0])
+ min[0] = dp;
+ if(dp > max[0])
+ max[0] = dp;
+ }
+ if(min[0]>max[0])
+ {
+ float tmp = min[0];
+ min[0] = max[0];
+ max[0] = tmp;
+ }
+ min[0] += offset;
+ max[0] += offset;
+}
+
+inline void project(__global const ConvexPolyhedronCL* hull, const float4 pos, const float4 orn,
+const float4* dir, __global const float4* vertices, float* min, float* max)
+{
+ min[0] = FLT_MAX;
+ max[0] = -FLT_MAX;
+ int numVerts = hull->m_numVertices;
+
+ const float4 localDir = qtInvRotate(orn,*dir);
+ float offset = dot(pos,*dir);
+ for(int i=0;i<numVerts;i++)
+ {
+ float dp = dot(vertices[hull->m_vertexOffset+i],localDir);
+ if(dp < min[0])
+ min[0] = dp;
+ if(dp > max[0])
+ max[0] = dp;
+ }
+ if(min[0]>max[0])
+ {
+ float tmp = min[0];
+ min[0] = max[0];
+ max[0] = tmp;
+ }
+ min[0] += offset;
+ max[0] += offset;
+}
+
+inline bool TestSepAxisLocalA(const ConvexPolyhedronCL* hullA, __global const ConvexPolyhedronCL* hullB,
+ const float4 posA,const float4 ornA,
+ const float4 posB,const float4 ornB,
+ float4* sep_axis, const float4* verticesA, __global const float4* verticesB,float* depth)
+{
+ float Min0,Max0;
+ float Min1,Max1;
+ projectLocal(hullA,posA,ornA,sep_axis,verticesA, &Min0, &Max0);
+ project(hullB,posB,ornB, sep_axis,verticesB, &Min1, &Max1);
+
+ if(Max0<Min1 || Max1<Min0)
+ return false;
+
+ float d0 = Max0 - Min1;
+ float d1 = Max1 - Min0;
+ *depth = d0<d1 ? d0:d1;
+ return true;
+}
+
+
+
+
+inline bool IsAlmostZero(const float4 v)
+{
+ if(fabs(v.x)>1e-6f || fabs(v.y)>1e-6f || fabs(v.z)>1e-6f)
+ return false;
+ return true;
+}
+
+
+
+bool findSeparatingAxisLocalA( const ConvexPolyhedronCL* hullA, __global const ConvexPolyhedronCL* hullB,
+ const float4 posA1,
+ const float4 ornA,
+ const float4 posB1,
+ const float4 ornB,
+ const float4 DeltaC2,
+
+ const float4* verticesA,
+ const float4* uniqueEdgesA,
+ const btGpuFace* facesA,
+ const int* indicesA,
+
+ __global const float4* verticesB,
+ __global const float4* uniqueEdgesB,
+ __global const btGpuFace* facesB,
+ __global const int* indicesB,
+ float4* sep,
+ float* dmin)
+{
+
+
+ float4 posA = posA1;
+ posA.w = 0.f;
+ float4 posB = posB1;
+ posB.w = 0.f;
+ int curPlaneTests=0;
+ {
+ int numFacesA = hullA->m_numFaces;
+ // Test normals from hullA
+ for(int i=0;i<numFacesA;i++)
+ {
+ const float4 normal = facesA[hullA->m_faceOffset+i].m_plane;
+ float4 faceANormalWS = qtRotate(ornA,normal);
+ if (dot3F4(DeltaC2,faceANormalWS)<0)
+ faceANormalWS*=-1.f;
+ curPlaneTests++;
+ float d;
+ if(!TestSepAxisLocalA( hullA, hullB, posA,ornA,posB,ornB,&faceANormalWS, verticesA, verticesB,&d))
+ return false;
+ if(d<*dmin)
+ {
+ *dmin = d;
+ *sep = faceANormalWS;
+ }
+ }
+ }
+ if((dot3F4(-DeltaC2,*sep))>0.0f)
+ {
+ *sep = -(*sep);
+ }
+ return true;
+}
+
+bool findSeparatingAxisLocalB( __global const ConvexPolyhedronCL* hullA, const ConvexPolyhedronCL* hullB,
+ const float4 posA1,
+ const float4 ornA,
+ const float4 posB1,
+ const float4 ornB,
+ const float4 DeltaC2,
+ __global const float4* verticesA,
+ __global const float4* uniqueEdgesA,
+ __global const btGpuFace* facesA,
+ __global const int* indicesA,
+ const float4* verticesB,
+ const float4* uniqueEdgesB,
+ const btGpuFace* facesB,
+ const int* indicesB,
+ float4* sep,
+ float* dmin)
+{
+
+
+ float4 posA = posA1;
+ posA.w = 0.f;
+ float4 posB = posB1;
+ posB.w = 0.f;
+ int curPlaneTests=0;
+ {
+ int numFacesA = hullA->m_numFaces;
+ // Test normals from hullA
+ for(int i=0;i<numFacesA;i++)
+ {
+ const float4 normal = facesA[hullA->m_faceOffset+i].m_plane;
+ float4 faceANormalWS = qtRotate(ornA,normal);
+ if (dot3F4(DeltaC2,faceANormalWS)<0)
+ faceANormalWS *= -1.f;
+ curPlaneTests++;
+ float d;
+ if(!TestSepAxisLocalA( hullB, hullA, posB,ornB,posA,ornA, &faceANormalWS, verticesB,verticesA, &d))
+ return false;
+ if(d<*dmin)
+ {
+ *dmin = d;
+ *sep = faceANormalWS;
+ }
+ }
+ }
+ if((dot3F4(-DeltaC2,*sep))>0.0f)
+ {
+ *sep = -(*sep);
+ }
+ return true;
+}
+
+
+
+bool findSeparatingAxisEdgeEdgeLocalA( const ConvexPolyhedronCL* hullA, __global const ConvexPolyhedronCL* hullB,
+ const float4 posA1,
+ const float4 ornA,
+ const float4 posB1,
+ const float4 ornB,
+ const float4 DeltaC2,
+ const float4* verticesA,
+ const float4* uniqueEdgesA,
+ const btGpuFace* facesA,
+ const int* indicesA,
+ __global const float4* verticesB,
+ __global const float4* uniqueEdgesB,
+ __global const btGpuFace* facesB,
+ __global const int* indicesB,
+ float4* sep,
+ float* dmin)
+{
+
+
+ float4 posA = posA1;
+ posA.w = 0.f;
+ float4 posB = posB1;
+ posB.w = 0.f;
+
+ int curPlaneTests=0;
+
+ int curEdgeEdge = 0;
+ // Test edges
+ for(int e0=0;e0<hullA->m_numUniqueEdges;e0++)
+ {
+ const float4 edge0 = uniqueEdgesA[hullA->m_uniqueEdgesOffset+e0];
+ float4 edge0World = qtRotate(ornA,edge0);
+
+ for(int e1=0;e1<hullB->m_numUniqueEdges;e1++)
+ {
+ const float4 edge1 = uniqueEdgesB[hullB->m_uniqueEdgesOffset+e1];
+ float4 edge1World = qtRotate(ornB,edge1);
+
+
+ float4 crossje = cross3(edge0World,edge1World);
+
+ curEdgeEdge++;
+ if(!IsAlmostZero(crossje))
+ {
+ crossje = normalize3(crossje);
+ if (dot3F4(DeltaC2,crossje)<0)
+ crossje *= -1.f;
+
+ float dist;
+ bool result = true;
+ {
+ float Min0,Max0;
+ float Min1,Max1;
+ projectLocal(hullA,posA,ornA,&crossje,verticesA, &Min0, &Max0);
+ project(hullB,posB,ornB,&crossje,verticesB, &Min1, &Max1);
+
+ if(Max0<Min1 || Max1<Min0)
+ result = false;
+
+ float d0 = Max0 - Min1;
+ float d1 = Max1 - Min0;
+ dist = d0<d1 ? d0:d1;
+ result = true;
+
+ }
+
+
+ if(dist<*dmin)
+ {
+ *dmin = dist;
+ *sep = crossje;
+ }
+ }
+ }
+
+ }
+
+
+ if((dot3F4(-DeltaC2,*sep))>0.0f)
+ {
+ *sep = -(*sep);
+ }
+ return true;
+}
+
+
+
+inline int findClippingFaces(const float4 separatingNormal,
+ const ConvexPolyhedronCL* hullA,
+ __global const ConvexPolyhedronCL* hullB,
+ const float4 posA, const Quaternion ornA,const float4 posB, const Quaternion ornB,
+ __global float4* worldVertsA1,
+ __global float4* worldNormalsA1,
+ __global float4* worldVertsB1,
+ int capacityWorldVerts,
+ const float minDist, float maxDist,
+ const float4* verticesA,
+ const btGpuFace* facesA,
+ const int* indicesA,
+ __global const float4* verticesB,
+ __global const btGpuFace* facesB,
+ __global const int* indicesB,
+ __global int4* clippingFaces, int pairIndex)
+{
+ int numContactsOut = 0;
+ int numWorldVertsB1= 0;
+
+
+ int closestFaceB=0;
+ float dmax = -FLT_MAX;
+
+ {
+ for(int face=0;face<hullB->m_numFaces;face++)
+ {
+ const float4 Normal = make_float4(facesB[hullB->m_faceOffset+face].m_plane.x,
+ facesB[hullB->m_faceOffset+face].m_plane.y, facesB[hullB->m_faceOffset+face].m_plane.z,0.f);
+ const float4 WorldNormal = qtRotate(ornB, Normal);
+ float d = dot3F4(WorldNormal,separatingNormal);
+ if (d > dmax)
+ {
+ dmax = d;
+ closestFaceB = face;
+ }
+ }
+ }
+
+ {
+ const btGpuFace polyB = facesB[hullB->m_faceOffset+closestFaceB];
+ int numVertices = polyB.m_numIndices;
+ if (numVertices>capacityWorldVerts)
+ numVertices = capacityWorldVerts;
+ if (numVertices<0)
+ numVertices = 0;
+
+ for(int e0=0;e0<numVertices;e0++)
+ {
+ if (e0<capacityWorldVerts)
+ {
+ const float4 b = verticesB[hullB->m_vertexOffset+indicesB[polyB.m_indexOffset+e0]];
+ worldVertsB1[pairIndex*capacityWorldVerts+numWorldVertsB1++] = transform(&b,&posB,&ornB);
+ }
+ }
+ }
+
+ int closestFaceA=0;
+ {
+ float dmin = FLT_MAX;
+ for(int face=0;face<hullA->m_numFaces;face++)
+ {
+ const float4 Normal = make_float4(
+ facesA[hullA->m_faceOffset+face].m_plane.x,
+ facesA[hullA->m_faceOffset+face].m_plane.y,
+ facesA[hullA->m_faceOffset+face].m_plane.z,
+ 0.f);
+ const float4 faceANormalWS = qtRotate(ornA,Normal);
+
+ float d = dot3F4(faceANormalWS,separatingNormal);
+ if (d < dmin)
+ {
+ dmin = d;
+ closestFaceA = face;
+ worldNormalsA1[pairIndex] = faceANormalWS;
+ }
+ }
+ }
+
+ int numVerticesA = facesA[hullA->m_faceOffset+closestFaceA].m_numIndices;
+ if (numVerticesA>capacityWorldVerts)
+ numVerticesA = capacityWorldVerts;
+ if (numVerticesA<0)
+ numVerticesA=0;
+
+ for(int e0=0;e0<numVerticesA;e0++)
+ {
+ if (e0<capacityWorldVerts)
+ {
+ const float4 a = verticesA[hullA->m_vertexOffset+indicesA[facesA[hullA->m_faceOffset+closestFaceA].m_indexOffset+e0]];
+ worldVertsA1[pairIndex*capacityWorldVerts+e0] = transform(&a, &posA,&ornA);
+ }
+ }
+
+ clippingFaces[pairIndex].x = closestFaceA;
+ clippingFaces[pairIndex].y = closestFaceB;
+ clippingFaces[pairIndex].z = numVerticesA;
+ clippingFaces[pairIndex].w = numWorldVertsB1;
+
+
+ return numContactsOut;
+}
+
+
+
+
+// work-in-progress
+__kernel void findConcaveSeparatingAxisVertexFaceKernel( __global int4* concavePairs,
+ __global const BodyData* rigidBodies,
+ __global const btCollidableGpu* collidables,
+ __global const ConvexPolyhedronCL* convexShapes,
+ __global const float4* vertices,
+ __global const float4* uniqueEdges,
+ __global const btGpuFace* faces,
+ __global const int* indices,
+ __global const btGpuChildShape* gpuChildShapes,
+ __global btAabbCL* aabbs,
+ __global float4* concaveSeparatingNormalsOut,
+ __global int* concaveHasSeparatingNormals,
+ __global int4* clippingFacesOut,
+ __global float4* worldVertsA1GPU,
+ __global float4* worldNormalsAGPU,
+ __global float4* worldVertsB1GPU,
+ __global float* dmins,
+ int vertexFaceCapacity,
+ int numConcavePairs
+ )
+{
+
+ int i = get_global_id(0);
+ if (i>=numConcavePairs)
+ return;
+
+ concaveHasSeparatingNormals[i] = 0;
+
+ int pairIdx = i;
+
+ int bodyIndexA = concavePairs[i].x;
+ int bodyIndexB = concavePairs[i].y;
+
+ int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;
+ int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;
+
+ int shapeIndexA = collidables[collidableIndexA].m_shapeIndex;
+ int shapeIndexB = collidables[collidableIndexB].m_shapeIndex;
+
+ if (collidables[collidableIndexB].m_shapeType!=SHAPE_CONVEX_HULL&&
+ collidables[collidableIndexB].m_shapeType!=SHAPE_COMPOUND_OF_CONVEX_HULLS)
+ {
+ concavePairs[pairIdx].w = -1;
+ return;
+ }
+
+
+
+ int numFacesA = convexShapes[shapeIndexA].m_numFaces;
+ int numActualConcaveConvexTests = 0;
+
+ int f = concavePairs[i].z;
+
+ bool overlap = false;
+
+ ConvexPolyhedronCL convexPolyhedronA;
+
+ //add 3 vertices of the triangle
+ convexPolyhedronA.m_numVertices = 3;
+ convexPolyhedronA.m_vertexOffset = 0;
+ float4 localCenter = make_float4(0.f,0.f,0.f,0.f);
+
+ btGpuFace face = faces[convexShapes[shapeIndexA].m_faceOffset+f];
+ float4 triMinAabb, triMaxAabb;
+ btAabbCL triAabb;
+ triAabb.m_min = make_float4(1e30f,1e30f,1e30f,0.f);
+ triAabb.m_max = make_float4(-1e30f,-1e30f,-1e30f,0.f);
+
+ float4 verticesA[3];
+ for (int i=0;i<3;i++)
+ {
+ int index = indices[face.m_indexOffset+i];
+ float4 vert = vertices[convexShapes[shapeIndexA].m_vertexOffset+index];
+ verticesA[i] = vert;
+ localCenter += vert;
+
+ triAabb.m_min = min(triAabb.m_min,vert);
+ triAabb.m_max = max(triAabb.m_max,vert);
+
+ }
+
+ overlap = true;
+ overlap = (triAabb.m_min.x > aabbs[bodyIndexB].m_max.x || triAabb.m_max.x < aabbs[bodyIndexB].m_min.x) ? false : overlap;
+ overlap = (triAabb.m_min.z > aabbs[bodyIndexB].m_max.z || triAabb.m_max.z < aabbs[bodyIndexB].m_min.z) ? false : overlap;
+ overlap = (triAabb.m_min.y > aabbs[bodyIndexB].m_max.y || triAabb.m_max.y < aabbs[bodyIndexB].m_min.y) ? false : overlap;
+
+ if (overlap)
+ {
+ float dmin = FLT_MAX;
+ int hasSeparatingAxis=5;
+ float4 sepAxis=make_float4(1,2,3,4);
+
+ int localCC=0;
+ numActualConcaveConvexTests++;
+
+ //a triangle has 3 unique edges
+ convexPolyhedronA.m_numUniqueEdges = 3;
+ convexPolyhedronA.m_uniqueEdgesOffset = 0;
+ float4 uniqueEdgesA[3];
+
+ uniqueEdgesA[0] = (verticesA[1]-verticesA[0]);
+ uniqueEdgesA[1] = (verticesA[2]-verticesA[1]);
+ uniqueEdgesA[2] = (verticesA[0]-verticesA[2]);
+
+
+ convexPolyhedronA.m_faceOffset = 0;
+
+ float4 normal = make_float4(face.m_plane.x,face.m_plane.y,face.m_plane.z,0.f);
+
+ btGpuFace facesA[TRIANGLE_NUM_CONVEX_FACES];
+ int indicesA[3+3+2+2+2];
+ int curUsedIndices=0;
+ int fidx=0;
+
+ //front size of triangle
+ {
+ facesA[fidx].m_indexOffset=curUsedIndices;
+ indicesA[0] = 0;
+ indicesA[1] = 1;
+ indicesA[2] = 2;
+ curUsedIndices+=3;
+ float c = face.m_plane.w;
+ facesA[fidx].m_plane.x = normal.x;
+ facesA[fidx].m_plane.y = normal.y;
+ facesA[fidx].m_plane.z = normal.z;
+ facesA[fidx].m_plane.w = c;
+ facesA[fidx].m_numIndices=3;
+ }
+ fidx++;
+ //back size of triangle
+ {
+ facesA[fidx].m_indexOffset=curUsedIndices;
+ indicesA[3]=2;
+ indicesA[4]=1;
+ indicesA[5]=0;
+ curUsedIndices+=3;
+ float c = dot(normal,verticesA[0]);
+ float c1 = -face.m_plane.w;
+ facesA[fidx].m_plane.x = -normal.x;
+ facesA[fidx].m_plane.y = -normal.y;
+ facesA[fidx].m_plane.z = -normal.z;
+ facesA[fidx].m_plane.w = c;
+ facesA[fidx].m_numIndices=3;
+ }
+ fidx++;
+
+ bool addEdgePlanes = true;
+ if (addEdgePlanes)
+ {
+ int numVertices=3;
+ int prevVertex = numVertices-1;
+ for (int i=0;i<numVertices;i++)
+ {
+ float4 v0 = verticesA[i];
+ float4 v1 = verticesA[prevVertex];
+
+ float4 edgeNormal = normalize(cross(normal,v1-v0));
+ float c = -dot(edgeNormal,v0);
+
+ facesA[fidx].m_numIndices = 2;
+ facesA[fidx].m_indexOffset=curUsedIndices;
+ indicesA[curUsedIndices++]=i;
+ indicesA[curUsedIndices++]=prevVertex;
+
+ facesA[fidx].m_plane.x = edgeNormal.x;
+ facesA[fidx].m_plane.y = edgeNormal.y;
+ facesA[fidx].m_plane.z = edgeNormal.z;
+ facesA[fidx].m_plane.w = c;
+ fidx++;
+ prevVertex = i;
+ }
+ }
+ convexPolyhedronA.m_numFaces = TRIANGLE_NUM_CONVEX_FACES;
+ convexPolyhedronA.m_localCenter = localCenter*(1.f/3.f);
+
+
+ float4 posA = rigidBodies[bodyIndexA].m_pos;
+ posA.w = 0.f;
+ float4 posB = rigidBodies[bodyIndexB].m_pos;
+ posB.w = 0.f;
+
+ float4 ornA = rigidBodies[bodyIndexA].m_quat;
+ float4 ornB =rigidBodies[bodyIndexB].m_quat;
+
+
+
+
+ ///////////////////
+ ///compound shape support
+
+ if (collidables[collidableIndexB].m_shapeType==SHAPE_COMPOUND_OF_CONVEX_HULLS)
+ {
+ int compoundChild = concavePairs[pairIdx].w;
+ int childShapeIndexB = compoundChild;//collidables[collidableIndexB].m_shapeIndex+compoundChild;
+ int childColIndexB = gpuChildShapes[childShapeIndexB].m_shapeIndex;
+ float4 childPosB = gpuChildShapes[childShapeIndexB].m_childPosition;
+ float4 childOrnB = gpuChildShapes[childShapeIndexB].m_childOrientation;
+ float4 newPosB = transform(&childPosB,&posB,&ornB);
+ float4 newOrnB = qtMul(ornB,childOrnB);
+ posB = newPosB;
+ ornB = newOrnB;
+ shapeIndexB = collidables[childColIndexB].m_shapeIndex;
+ }
+ //////////////////
+
+ float4 c0local = convexPolyhedronA.m_localCenter;
+ float4 c0 = transform(&c0local, &posA, &ornA);
+ float4 c1local = convexShapes[shapeIndexB].m_localCenter;
+ float4 c1 = transform(&c1local,&posB,&ornB);
+ const float4 DeltaC2 = c0 - c1;
+
+
+ bool sepA = findSeparatingAxisLocalA( &convexPolyhedronA, &convexShapes[shapeIndexB],
+ posA,ornA,
+ posB,ornB,
+ DeltaC2,
+ verticesA,uniqueEdgesA,facesA,indicesA,
+ vertices,uniqueEdges,faces,indices,
+ &sepAxis,&dmin);
+ hasSeparatingAxis = 4;
+ if (!sepA)
+ {
+ hasSeparatingAxis = 0;
+ } else
+ {
+ bool sepB = findSeparatingAxisLocalB( &convexShapes[shapeIndexB],&convexPolyhedronA,
+ posB,ornB,
+ posA,ornA,
+ DeltaC2,
+ vertices,uniqueEdges,faces,indices,
+ verticesA,uniqueEdgesA,facesA,indicesA,
+ &sepAxis,&dmin);
+
+ if (!sepB)
+ {
+ hasSeparatingAxis = 0;
+ } else
+ {
+ hasSeparatingAxis = 1;
+ }
+ }
+
+ if (hasSeparatingAxis)
+ {
+ dmins[i] = dmin;
+ concaveSeparatingNormalsOut[pairIdx]=sepAxis;
+ concaveHasSeparatingNormals[i]=1;
+
+ } else
+ {
+ //mark this pair as in-active
+ concavePairs[pairIdx].w = -1;
+ }
+ }
+ else
+ {
+ //mark this pair as in-active
+ concavePairs[pairIdx].w = -1;
+ }
+}
+
+
+
+
+// work-in-progress
+__kernel void findConcaveSeparatingAxisEdgeEdgeKernel( __global int4* concavePairs,
+ __global const BodyData* rigidBodies,
+ __global const btCollidableGpu* collidables,
+ __global const ConvexPolyhedronCL* convexShapes,
+ __global const float4* vertices,
+ __global const float4* uniqueEdges,
+ __global const btGpuFace* faces,
+ __global const int* indices,
+ __global const btGpuChildShape* gpuChildShapes,
+ __global btAabbCL* aabbs,
+ __global float4* concaveSeparatingNormalsOut,
+ __global int* concaveHasSeparatingNormals,
+ __global int4* clippingFacesOut,
+ __global float4* worldVertsA1GPU,
+ __global float4* worldNormalsAGPU,
+ __global float4* worldVertsB1GPU,
+ __global float* dmins,
+ int vertexFaceCapacity,
+ int numConcavePairs
+ )
+{
+
+ int i = get_global_id(0);
+ if (i>=numConcavePairs)
+ return;
+
+ if (!concaveHasSeparatingNormals[i])
+ return;
+
+ int pairIdx = i;
+
+ int bodyIndexA = concavePairs[i].x;
+ int bodyIndexB = concavePairs[i].y;
+
+ int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;
+ int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;
+
+ int shapeIndexA = collidables[collidableIndexA].m_shapeIndex;
+ int shapeIndexB = collidables[collidableIndexB].m_shapeIndex;
+
+
+ int numFacesA = convexShapes[shapeIndexA].m_numFaces;
+ int numActualConcaveConvexTests = 0;
+
+ int f = concavePairs[i].z;
+
+ bool overlap = false;
+
+ ConvexPolyhedronCL convexPolyhedronA;
+
+ //add 3 vertices of the triangle
+ convexPolyhedronA.m_numVertices = 3;
+ convexPolyhedronA.m_vertexOffset = 0;
+ float4 localCenter = make_float4(0.f,0.f,0.f,0.f);
+
+ btGpuFace face = faces[convexShapes[shapeIndexA].m_faceOffset+f];
+ float4 triMinAabb, triMaxAabb;
+ btAabbCL triAabb;
+ triAabb.m_min = make_float4(1e30f,1e30f,1e30f,0.f);
+ triAabb.m_max = make_float4(-1e30f,-1e30f,-1e30f,0.f);
+
+ float4 verticesA[3];
+ for (int i=0;i<3;i++)
+ {
+ int index = indices[face.m_indexOffset+i];
+ float4 vert = vertices[convexShapes[shapeIndexA].m_vertexOffset+index];
+ verticesA[i] = vert;
+ localCenter += vert;
+
+ triAabb.m_min = min(triAabb.m_min,vert);
+ triAabb.m_max = max(triAabb.m_max,vert);
+
+ }
+
+ overlap = true;
+ overlap = (triAabb.m_min.x > aabbs[bodyIndexB].m_max.x || triAabb.m_max.x < aabbs[bodyIndexB].m_min.x) ? false : overlap;
+ overlap = (triAabb.m_min.z > aabbs[bodyIndexB].m_max.z || triAabb.m_max.z < aabbs[bodyIndexB].m_min.z) ? false : overlap;
+ overlap = (triAabb.m_min.y > aabbs[bodyIndexB].m_max.y || triAabb.m_max.y < aabbs[bodyIndexB].m_min.y) ? false : overlap;
+
+ if (overlap)
+ {
+ float dmin = dmins[i];
+ int hasSeparatingAxis=5;
+ float4 sepAxis=make_float4(1,2,3,4);
+ sepAxis = concaveSeparatingNormalsOut[pairIdx];
+
+ int localCC=0;
+ numActualConcaveConvexTests++;
+
+ //a triangle has 3 unique edges
+ convexPolyhedronA.m_numUniqueEdges = 3;
+ convexPolyhedronA.m_uniqueEdgesOffset = 0;
+ float4 uniqueEdgesA[3];
+
+ uniqueEdgesA[0] = (verticesA[1]-verticesA[0]);
+ uniqueEdgesA[1] = (verticesA[2]-verticesA[1]);
+ uniqueEdgesA[2] = (verticesA[0]-verticesA[2]);
+
+
+ convexPolyhedronA.m_faceOffset = 0;
+
+ float4 normal = make_float4(face.m_plane.x,face.m_plane.y,face.m_plane.z,0.f);
+
+ btGpuFace facesA[TRIANGLE_NUM_CONVEX_FACES];
+ int indicesA[3+3+2+2+2];
+ int curUsedIndices=0;
+ int fidx=0;
+
+ //front size of triangle
+ {
+ facesA[fidx].m_indexOffset=curUsedIndices;
+ indicesA[0] = 0;
+ indicesA[1] = 1;
+ indicesA[2] = 2;
+ curUsedIndices+=3;
+ float c = face.m_plane.w;
+ facesA[fidx].m_plane.x = normal.x;
+ facesA[fidx].m_plane.y = normal.y;
+ facesA[fidx].m_plane.z = normal.z;
+ facesA[fidx].m_plane.w = c;
+ facesA[fidx].m_numIndices=3;
+ }
+ fidx++;
+ //back size of triangle
+ {
+ facesA[fidx].m_indexOffset=curUsedIndices;
+ indicesA[3]=2;
+ indicesA[4]=1;
+ indicesA[5]=0;
+ curUsedIndices+=3;
+ float c = dot(normal,verticesA[0]);
+ float c1 = -face.m_plane.w;
+ facesA[fidx].m_plane.x = -normal.x;
+ facesA[fidx].m_plane.y = -normal.y;
+ facesA[fidx].m_plane.z = -normal.z;
+ facesA[fidx].m_plane.w = c;
+ facesA[fidx].m_numIndices=3;
+ }
+ fidx++;
+
+ bool addEdgePlanes = true;
+ if (addEdgePlanes)
+ {
+ int numVertices=3;
+ int prevVertex = numVertices-1;
+ for (int i=0;i<numVertices;i++)
+ {
+ float4 v0 = verticesA[i];
+ float4 v1 = verticesA[prevVertex];
+
+ float4 edgeNormal = normalize(cross(normal,v1-v0));
+ float c = -dot(edgeNormal,v0);
+
+ facesA[fidx].m_numIndices = 2;
+ facesA[fidx].m_indexOffset=curUsedIndices;
+ indicesA[curUsedIndices++]=i;
+ indicesA[curUsedIndices++]=prevVertex;
+
+ facesA[fidx].m_plane.x = edgeNormal.x;
+ facesA[fidx].m_plane.y = edgeNormal.y;
+ facesA[fidx].m_plane.z = edgeNormal.z;
+ facesA[fidx].m_plane.w = c;
+ fidx++;
+ prevVertex = i;
+ }
+ }
+ convexPolyhedronA.m_numFaces = TRIANGLE_NUM_CONVEX_FACES;
+ convexPolyhedronA.m_localCenter = localCenter*(1.f/3.f);
+
+
+ float4 posA = rigidBodies[bodyIndexA].m_pos;
+ posA.w = 0.f;
+ float4 posB = rigidBodies[bodyIndexB].m_pos;
+ posB.w = 0.f;
+
+ float4 ornA = rigidBodies[bodyIndexA].m_quat;
+ float4 ornB =rigidBodies[bodyIndexB].m_quat;
+
+
+
+
+ ///////////////////
+ ///compound shape support
+
+ if (collidables[collidableIndexB].m_shapeType==SHAPE_COMPOUND_OF_CONVEX_HULLS)
+ {
+ int compoundChild = concavePairs[pairIdx].w;
+ int childShapeIndexB = compoundChild;//collidables[collidableIndexB].m_shapeIndex+compoundChild;
+ int childColIndexB = gpuChildShapes[childShapeIndexB].m_shapeIndex;
+ float4 childPosB = gpuChildShapes[childShapeIndexB].m_childPosition;
+ float4 childOrnB = gpuChildShapes[childShapeIndexB].m_childOrientation;
+ float4 newPosB = transform(&childPosB,&posB,&ornB);
+ float4 newOrnB = qtMul(ornB,childOrnB);
+ posB = newPosB;
+ ornB = newOrnB;
+ shapeIndexB = collidables[childColIndexB].m_shapeIndex;
+ }
+ //////////////////
+
+ float4 c0local = convexPolyhedronA.m_localCenter;
+ float4 c0 = transform(&c0local, &posA, &ornA);
+ float4 c1local = convexShapes[shapeIndexB].m_localCenter;
+ float4 c1 = transform(&c1local,&posB,&ornB);
+ const float4 DeltaC2 = c0 - c1;
+
+
+ {
+ bool sepEE = findSeparatingAxisEdgeEdgeLocalA( &convexPolyhedronA, &convexShapes[shapeIndexB],
+ posA,ornA,
+ posB,ornB,
+ DeltaC2,
+ verticesA,uniqueEdgesA,facesA,indicesA,
+ vertices,uniqueEdges,faces,indices,
+ &sepAxis,&dmin);
+
+ if (!sepEE)
+ {
+ hasSeparatingAxis = 0;
+ } else
+ {
+ hasSeparatingAxis = 1;
+ }
+ }
+
+
+ if (hasSeparatingAxis)
+ {
+ sepAxis.w = dmin;
+ dmins[i] = dmin;
+ concaveSeparatingNormalsOut[pairIdx]=sepAxis;
+ concaveHasSeparatingNormals[i]=1;
+
+ float minDist = -1e30f;
+ float maxDist = 0.02f;
+
+
+ findClippingFaces(sepAxis,
+ &convexPolyhedronA,
+ &convexShapes[shapeIndexB],
+ posA,ornA,
+ posB,ornB,
+ worldVertsA1GPU,
+ worldNormalsAGPU,
+ worldVertsB1GPU,
+ vertexFaceCapacity,
+ minDist, maxDist,
+ verticesA,
+ facesA,
+ indicesA,
+ vertices,
+ faces,
+ indices,
+ clippingFacesOut, pairIdx);
+
+
+ } else
+ {
+ //mark this pair as in-active
+ concavePairs[pairIdx].w = -1;
+ }
+ }
+ else
+ {
+ //mark this pair as in-active
+ concavePairs[pairIdx].w = -1;
+ }
+
+ concavePairs[i].z = -1;//for the next stage, z is used to determine existing contact points
+}
+
diff --git a/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/satConcaveKernels.h b/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/satConcaveKernels.h
new file mode 100644
index 0000000000..611569cacf
--- /dev/null
+++ b/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/satConcaveKernels.h
@@ -0,0 +1,1457 @@
+//this file is autogenerated using stringify.bat (premake --stringify) in the build folder of this project
+static const char* satConcaveKernelsCL= \
+"//keep this enum in sync with the CPU version (in btCollidable.h)\n"
+"//written by Erwin Coumans\n"
+"#define SHAPE_CONVEX_HULL 3\n"
+"#define SHAPE_CONCAVE_TRIMESH 5\n"
+"#define TRIANGLE_NUM_CONVEX_FACES 5\n"
+"#define SHAPE_COMPOUND_OF_CONVEX_HULLS 6\n"
+"#define B3_MAX_STACK_DEPTH 256\n"
+"typedef unsigned int u32;\n"
+"///keep this in sync with btCollidable.h\n"
+"typedef struct\n"
+"{\n"
+" union {\n"
+" int m_numChildShapes;\n"
+" int m_bvhIndex;\n"
+" };\n"
+" union\n"
+" {\n"
+" float m_radius;\n"
+" int m_compoundBvhIndex;\n"
+" };\n"
+" \n"
+" int m_shapeType;\n"
+" int m_shapeIndex;\n"
+" \n"
+"} btCollidableGpu;\n"
+"#define MAX_NUM_PARTS_IN_BITS 10\n"
+"///b3QuantizedBvhNode is a compressed aabb node, 16 bytes.\n"
+"///Node can be used for leafnode or internal node. Leafnodes can point to 32-bit triangle index (non-negative range).\n"
+"typedef struct\n"
+"{\n"
+" //12 bytes\n"
+" unsigned short int m_quantizedAabbMin[3];\n"
+" unsigned short int m_quantizedAabbMax[3];\n"
+" //4 bytes\n"
+" int m_escapeIndexOrTriangleIndex;\n"
+"} b3QuantizedBvhNode;\n"
+"typedef struct\n"
+"{\n"
+" float4 m_aabbMin;\n"
+" float4 m_aabbMax;\n"
+" float4 m_quantization;\n"
+" int m_numNodes;\n"
+" int m_numSubTrees;\n"
+" int m_nodeOffset;\n"
+" int m_subTreeOffset;\n"
+"} b3BvhInfo;\n"
+"int getTriangleIndex(const b3QuantizedBvhNode* rootNode)\n"
+"{\n"
+" unsigned int x=0;\n"
+" unsigned int y = (~(x&0))<<(31-MAX_NUM_PARTS_IN_BITS);\n"
+" // Get only the lower bits where the triangle index is stored\n"
+" return (rootNode->m_escapeIndexOrTriangleIndex&~(y));\n"
+"}\n"
+"int getTriangleIndexGlobal(__global const b3QuantizedBvhNode* rootNode)\n"
+"{\n"
+" unsigned int x=0;\n"
+" unsigned int y = (~(x&0))<<(31-MAX_NUM_PARTS_IN_BITS);\n"
+" // Get only the lower bits where the triangle index is stored\n"
+" return (rootNode->m_escapeIndexOrTriangleIndex&~(y));\n"
+"}\n"
+"int isLeafNode(const b3QuantizedBvhNode* rootNode)\n"
+"{\n"
+" //skipindex is negative (internal node), triangleindex >=0 (leafnode)\n"
+" return (rootNode->m_escapeIndexOrTriangleIndex >= 0)? 1 : 0;\n"
+"}\n"
+"int isLeafNodeGlobal(__global const b3QuantizedBvhNode* rootNode)\n"
+"{\n"
+" //skipindex is negative (internal node), triangleindex >=0 (leafnode)\n"
+" return (rootNode->m_escapeIndexOrTriangleIndex >= 0)? 1 : 0;\n"
+"}\n"
+" \n"
+"int getEscapeIndex(const b3QuantizedBvhNode* rootNode)\n"
+"{\n"
+" return -rootNode->m_escapeIndexOrTriangleIndex;\n"
+"}\n"
+"int getEscapeIndexGlobal(__global const b3QuantizedBvhNode* rootNode)\n"
+"{\n"
+" return -rootNode->m_escapeIndexOrTriangleIndex;\n"
+"}\n"
+"typedef struct\n"
+"{\n"
+" //12 bytes\n"
+" unsigned short int m_quantizedAabbMin[3];\n"
+" unsigned short int m_quantizedAabbMax[3];\n"
+" //4 bytes, points to the root of the subtree\n"
+" int m_rootNodeIndex;\n"
+" //4 bytes\n"
+" int m_subtreeSize;\n"
+" int m_padding[3];\n"
+"} b3BvhSubtreeInfo;\n"
+"typedef struct\n"
+"{\n"
+" float4 m_childPosition;\n"
+" float4 m_childOrientation;\n"
+" int m_shapeIndex;\n"
+" int m_unused0;\n"
+" int m_unused1;\n"
+" int m_unused2;\n"
+"} btGpuChildShape;\n"
+"typedef struct\n"
+"{\n"
+" float4 m_pos;\n"
+" float4 m_quat;\n"
+" float4 m_linVel;\n"
+" float4 m_angVel;\n"
+" u32 m_collidableIdx;\n"
+" float m_invMass;\n"
+" float m_restituitionCoeff;\n"
+" float m_frictionCoeff;\n"
+"} BodyData;\n"
+"typedef struct \n"
+"{\n"
+" float4 m_localCenter;\n"
+" float4 m_extents;\n"
+" float4 mC;\n"
+" float4 mE;\n"
+" \n"
+" float m_radius;\n"
+" int m_faceOffset;\n"
+" int m_numFaces;\n"
+" int m_numVertices;\n"
+" int m_vertexOffset;\n"
+" int m_uniqueEdgesOffset;\n"
+" int m_numUniqueEdges;\n"
+" int m_unused;\n"
+"} ConvexPolyhedronCL;\n"
+"typedef struct \n"
+"{\n"
+" union\n"
+" {\n"
+" float4 m_min;\n"
+" float m_minElems[4];\n"
+" int m_minIndices[4];\n"
+" };\n"
+" union\n"
+" {\n"
+" float4 m_max;\n"
+" float m_maxElems[4];\n"
+" int m_maxIndices[4];\n"
+" };\n"
+"} btAabbCL;\n"
+"#ifndef B3_AABB_H\n"
+"#define B3_AABB_H\n"
+"#ifndef B3_FLOAT4_H\n"
+"#define B3_FLOAT4_H\n"
+"#ifndef B3_PLATFORM_DEFINITIONS_H\n"
+"#define B3_PLATFORM_DEFINITIONS_H\n"
+"struct MyTest\n"
+"{\n"
+" int bla;\n"
+"};\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+"//keep B3_LARGE_FLOAT*B3_LARGE_FLOAT < FLT_MAX\n"
+"#define B3_LARGE_FLOAT 1e18f\n"
+"#define B3_INFINITY 1e18f\n"
+"#define b3Assert(a)\n"
+"#define b3ConstArray(a) __global const a*\n"
+"#define b3AtomicInc atomic_inc\n"
+"#define b3AtomicAdd atomic_add\n"
+"#define b3Fabs fabs\n"
+"#define b3Sqrt native_sqrt\n"
+"#define b3Sin native_sin\n"
+"#define b3Cos native_cos\n"
+"#define B3_STATIC\n"
+"#endif\n"
+"#endif\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+" typedef float4 b3Float4;\n"
+" #define b3Float4ConstArg const b3Float4\n"
+" #define b3MakeFloat4 (float4)\n"
+" float b3Dot3F4(b3Float4ConstArg v0,b3Float4ConstArg v1)\n"
+" {\n"
+" float4 a1 = b3MakeFloat4(v0.xyz,0.f);\n"
+" float4 b1 = b3MakeFloat4(v1.xyz,0.f);\n"
+" return dot(a1, b1);\n"
+" }\n"
+" b3Float4 b3Cross3(b3Float4ConstArg v0,b3Float4ConstArg v1)\n"
+" {\n"
+" float4 a1 = b3MakeFloat4(v0.xyz,0.f);\n"
+" float4 b1 = b3MakeFloat4(v1.xyz,0.f);\n"
+" return cross(a1, b1);\n"
+" }\n"
+" #define b3MinFloat4 min\n"
+" #define b3MaxFloat4 max\n"
+" #define b3Normalized(a) normalize(a)\n"
+"#endif \n"
+" \n"
+"inline bool b3IsAlmostZero(b3Float4ConstArg v)\n"
+"{\n"
+" if(b3Fabs(v.x)>1e-6 || b3Fabs(v.y)>1e-6 || b3Fabs(v.z)>1e-6) \n"
+" return false;\n"
+" return true;\n"
+"}\n"
+"inline int b3MaxDot( b3Float4ConstArg vec, __global const b3Float4* vecArray, int vecLen, float* dotOut )\n"
+"{\n"
+" float maxDot = -B3_INFINITY;\n"
+" int i = 0;\n"
+" int ptIndex = -1;\n"
+" for( i = 0; i < vecLen; i++ )\n"
+" {\n"
+" float dot = b3Dot3F4(vecArray[i],vec);\n"
+" \n"
+" if( dot > maxDot )\n"
+" {\n"
+" maxDot = dot;\n"
+" ptIndex = i;\n"
+" }\n"
+" }\n"
+" b3Assert(ptIndex>=0);\n"
+" if (ptIndex<0)\n"
+" {\n"
+" ptIndex = 0;\n"
+" }\n"
+" *dotOut = maxDot;\n"
+" return ptIndex;\n"
+"}\n"
+"#endif //B3_FLOAT4_H\n"
+"#ifndef B3_MAT3x3_H\n"
+"#define B3_MAT3x3_H\n"
+"#ifndef B3_QUAT_H\n"
+"#define B3_QUAT_H\n"
+"#ifndef B3_PLATFORM_DEFINITIONS_H\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+"#endif\n"
+"#endif\n"
+"#ifndef B3_FLOAT4_H\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+"#endif \n"
+"#endif //B3_FLOAT4_H\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+" typedef float4 b3Quat;\n"
+" #define b3QuatConstArg const b3Quat\n"
+" \n"
+" \n"
+"inline float4 b3FastNormalize4(float4 v)\n"
+"{\n"
+" v = (float4)(v.xyz,0.f);\n"
+" return fast_normalize(v);\n"
+"}\n"
+" \n"
+"inline b3Quat b3QuatMul(b3Quat a, b3Quat b);\n"
+"inline b3Quat b3QuatNormalized(b3QuatConstArg in);\n"
+"inline b3Quat b3QuatRotate(b3QuatConstArg q, b3QuatConstArg vec);\n"
+"inline b3Quat b3QuatInvert(b3QuatConstArg q);\n"
+"inline b3Quat b3QuatInverse(b3QuatConstArg q);\n"
+"inline b3Quat b3QuatMul(b3QuatConstArg a, b3QuatConstArg b)\n"
+"{\n"
+" b3Quat ans;\n"
+" ans = b3Cross3( a, b );\n"
+" ans += a.w*b+b.w*a;\n"
+"// ans.w = a.w*b.w - (a.x*b.x+a.y*b.y+a.z*b.z);\n"
+" ans.w = a.w*b.w - b3Dot3F4(a, b);\n"
+" return ans;\n"
+"}\n"
+"inline b3Quat b3QuatNormalized(b3QuatConstArg in)\n"
+"{\n"
+" b3Quat q;\n"
+" q=in;\n"
+" //return b3FastNormalize4(in);\n"
+" float len = native_sqrt(dot(q, q));\n"
+" if(len > 0.f)\n"
+" {\n"
+" q *= 1.f / len;\n"
+" }\n"
+" else\n"
+" {\n"
+" q.x = q.y = q.z = 0.f;\n"
+" q.w = 1.f;\n"
+" }\n"
+" return q;\n"
+"}\n"
+"inline float4 b3QuatRotate(b3QuatConstArg q, b3QuatConstArg vec)\n"
+"{\n"
+" b3Quat qInv = b3QuatInvert( q );\n"
+" float4 vcpy = vec;\n"
+" vcpy.w = 0.f;\n"
+" float4 out = b3QuatMul(b3QuatMul(q,vcpy),qInv);\n"
+" return out;\n"
+"}\n"
+"inline b3Quat b3QuatInverse(b3QuatConstArg q)\n"
+"{\n"
+" return (b3Quat)(-q.xyz, q.w);\n"
+"}\n"
+"inline b3Quat b3QuatInvert(b3QuatConstArg q)\n"
+"{\n"
+" return (b3Quat)(-q.xyz, q.w);\n"
+"}\n"
+"inline float4 b3QuatInvRotate(b3QuatConstArg q, b3QuatConstArg vec)\n"
+"{\n"
+" return b3QuatRotate( b3QuatInvert( q ), vec );\n"
+"}\n"
+"inline b3Float4 b3TransformPoint(b3Float4ConstArg point, b3Float4ConstArg translation, b3QuatConstArg orientation)\n"
+"{\n"
+" return b3QuatRotate( orientation, point ) + (translation);\n"
+"}\n"
+" \n"
+"#endif \n"
+"#endif //B3_QUAT_H\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+"typedef struct\n"
+"{\n"
+" b3Float4 m_row[3];\n"
+"}b3Mat3x3;\n"
+"#define b3Mat3x3ConstArg const b3Mat3x3\n"
+"#define b3GetRow(m,row) (m.m_row[row])\n"
+"inline b3Mat3x3 b3QuatGetRotationMatrix(b3Quat quat)\n"
+"{\n"
+" b3Float4 quat2 = (b3Float4)(quat.x*quat.x, quat.y*quat.y, quat.z*quat.z, 0.f);\n"
+" b3Mat3x3 out;\n"
+" out.m_row[0].x=1-2*quat2.y-2*quat2.z;\n"
+" out.m_row[0].y=2*quat.x*quat.y-2*quat.w*quat.z;\n"
+" out.m_row[0].z=2*quat.x*quat.z+2*quat.w*quat.y;\n"
+" out.m_row[0].w = 0.f;\n"
+" out.m_row[1].x=2*quat.x*quat.y+2*quat.w*quat.z;\n"
+" out.m_row[1].y=1-2*quat2.x-2*quat2.z;\n"
+" out.m_row[1].z=2*quat.y*quat.z-2*quat.w*quat.x;\n"
+" out.m_row[1].w = 0.f;\n"
+" out.m_row[2].x=2*quat.x*quat.z-2*quat.w*quat.y;\n"
+" out.m_row[2].y=2*quat.y*quat.z+2*quat.w*quat.x;\n"
+" out.m_row[2].z=1-2*quat2.x-2*quat2.y;\n"
+" out.m_row[2].w = 0.f;\n"
+" return out;\n"
+"}\n"
+"inline b3Mat3x3 b3AbsoluteMat3x3(b3Mat3x3ConstArg matIn)\n"
+"{\n"
+" b3Mat3x3 out;\n"
+" out.m_row[0] = fabs(matIn.m_row[0]);\n"
+" out.m_row[1] = fabs(matIn.m_row[1]);\n"
+" out.m_row[2] = fabs(matIn.m_row[2]);\n"
+" return out;\n"
+"}\n"
+"__inline\n"
+"b3Mat3x3 mtZero();\n"
+"__inline\n"
+"b3Mat3x3 mtIdentity();\n"
+"__inline\n"
+"b3Mat3x3 mtTranspose(b3Mat3x3 m);\n"
+"__inline\n"
+"b3Mat3x3 mtMul(b3Mat3x3 a, b3Mat3x3 b);\n"
+"__inline\n"
+"b3Float4 mtMul1(b3Mat3x3 a, b3Float4 b);\n"
+"__inline\n"
+"b3Float4 mtMul3(b3Float4 a, b3Mat3x3 b);\n"
+"__inline\n"
+"b3Mat3x3 mtZero()\n"
+"{\n"
+" b3Mat3x3 m;\n"
+" m.m_row[0] = (b3Float4)(0.f);\n"
+" m.m_row[1] = (b3Float4)(0.f);\n"
+" m.m_row[2] = (b3Float4)(0.f);\n"
+" return m;\n"
+"}\n"
+"__inline\n"
+"b3Mat3x3 mtIdentity()\n"
+"{\n"
+" b3Mat3x3 m;\n"
+" m.m_row[0] = (b3Float4)(1,0,0,0);\n"
+" m.m_row[1] = (b3Float4)(0,1,0,0);\n"
+" m.m_row[2] = (b3Float4)(0,0,1,0);\n"
+" return m;\n"
+"}\n"
+"__inline\n"
+"b3Mat3x3 mtTranspose(b3Mat3x3 m)\n"
+"{\n"
+" b3Mat3x3 out;\n"
+" out.m_row[0] = (b3Float4)(m.m_row[0].x, m.m_row[1].x, m.m_row[2].x, 0.f);\n"
+" out.m_row[1] = (b3Float4)(m.m_row[0].y, m.m_row[1].y, m.m_row[2].y, 0.f);\n"
+" out.m_row[2] = (b3Float4)(m.m_row[0].z, m.m_row[1].z, m.m_row[2].z, 0.f);\n"
+" return out;\n"
+"}\n"
+"__inline\n"
+"b3Mat3x3 mtMul(b3Mat3x3 a, b3Mat3x3 b)\n"
+"{\n"
+" b3Mat3x3 transB;\n"
+" transB = mtTranspose( b );\n"
+" b3Mat3x3 ans;\n"
+" // why this doesn't run when 0ing in the for{}\n"
+" a.m_row[0].w = 0.f;\n"
+" a.m_row[1].w = 0.f;\n"
+" a.m_row[2].w = 0.f;\n"
+" for(int i=0; i<3; i++)\n"
+" {\n"
+"// a.m_row[i].w = 0.f;\n"
+" ans.m_row[i].x = b3Dot3F4(a.m_row[i],transB.m_row[0]);\n"
+" ans.m_row[i].y = b3Dot3F4(a.m_row[i],transB.m_row[1]);\n"
+" ans.m_row[i].z = b3Dot3F4(a.m_row[i],transB.m_row[2]);\n"
+" ans.m_row[i].w = 0.f;\n"
+" }\n"
+" return ans;\n"
+"}\n"
+"__inline\n"
+"b3Float4 mtMul1(b3Mat3x3 a, b3Float4 b)\n"
+"{\n"
+" b3Float4 ans;\n"
+" ans.x = b3Dot3F4( a.m_row[0], b );\n"
+" ans.y = b3Dot3F4( a.m_row[1], b );\n"
+" ans.z = b3Dot3F4( a.m_row[2], b );\n"
+" ans.w = 0.f;\n"
+" return ans;\n"
+"}\n"
+"__inline\n"
+"b3Float4 mtMul3(b3Float4 a, b3Mat3x3 b)\n"
+"{\n"
+" b3Float4 colx = b3MakeFloat4(b.m_row[0].x, b.m_row[1].x, b.m_row[2].x, 0);\n"
+" b3Float4 coly = b3MakeFloat4(b.m_row[0].y, b.m_row[1].y, b.m_row[2].y, 0);\n"
+" b3Float4 colz = b3MakeFloat4(b.m_row[0].z, b.m_row[1].z, b.m_row[2].z, 0);\n"
+" b3Float4 ans;\n"
+" ans.x = b3Dot3F4( a, colx );\n"
+" ans.y = b3Dot3F4( a, coly );\n"
+" ans.z = b3Dot3F4( a, colz );\n"
+" return ans;\n"
+"}\n"
+"#endif\n"
+"#endif //B3_MAT3x3_H\n"
+"typedef struct b3Aabb b3Aabb_t;\n"
+"struct b3Aabb\n"
+"{\n"
+" union\n"
+" {\n"
+" float m_min[4];\n"
+" b3Float4 m_minVec;\n"
+" int m_minIndices[4];\n"
+" };\n"
+" union\n"
+" {\n"
+" float m_max[4];\n"
+" b3Float4 m_maxVec;\n"
+" int m_signedMaxIndices[4];\n"
+" };\n"
+"};\n"
+"inline void b3TransformAabb2(b3Float4ConstArg localAabbMin,b3Float4ConstArg localAabbMax, float margin,\n"
+" b3Float4ConstArg pos,\n"
+" b3QuatConstArg orn,\n"
+" b3Float4* aabbMinOut,b3Float4* aabbMaxOut)\n"
+"{\n"
+" b3Float4 localHalfExtents = 0.5f*(localAabbMax-localAabbMin);\n"
+" localHalfExtents+=b3MakeFloat4(margin,margin,margin,0.f);\n"
+" b3Float4 localCenter = 0.5f*(localAabbMax+localAabbMin);\n"
+" b3Mat3x3 m;\n"
+" m = b3QuatGetRotationMatrix(orn);\n"
+" b3Mat3x3 abs_b = b3AbsoluteMat3x3(m);\n"
+" b3Float4 center = b3TransformPoint(localCenter,pos,orn);\n"
+" \n"
+" b3Float4 extent = b3MakeFloat4(b3Dot3F4(localHalfExtents,b3GetRow(abs_b,0)),\n"
+" b3Dot3F4(localHalfExtents,b3GetRow(abs_b,1)),\n"
+" b3Dot3F4(localHalfExtents,b3GetRow(abs_b,2)),\n"
+" 0.f);\n"
+" *aabbMinOut = center-extent;\n"
+" *aabbMaxOut = center+extent;\n"
+"}\n"
+"/// conservative test for overlap between two aabbs\n"
+"inline bool b3TestAabbAgainstAabb(b3Float4ConstArg aabbMin1,b3Float4ConstArg aabbMax1,\n"
+" b3Float4ConstArg aabbMin2, b3Float4ConstArg aabbMax2)\n"
+"{\n"
+" bool overlap = true;\n"
+" overlap = (aabbMin1.x > aabbMax2.x || aabbMax1.x < aabbMin2.x) ? false : overlap;\n"
+" overlap = (aabbMin1.z > aabbMax2.z || aabbMax1.z < aabbMin2.z) ? false : overlap;\n"
+" overlap = (aabbMin1.y > aabbMax2.y || aabbMax1.y < aabbMin2.y) ? false : overlap;\n"
+" return overlap;\n"
+"}\n"
+"#endif //B3_AABB_H\n"
+"/*\n"
+"Bullet Continuous Collision Detection and Physics Library\n"
+"Copyright (c) 2003-2013 Erwin Coumans http://bulletphysics.org\n"
+"This software is provided 'as-is', without any express or implied warranty.\n"
+"In no event will the authors be held liable for any damages arising from the use of this software.\n"
+"Permission is granted to anyone to use this software for any purpose,\n"
+"including commercial applications, and to alter it and redistribute it freely,\n"
+"subject to the following restrictions:\n"
+"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.\n"
+"2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.\n"
+"3. This notice may not be removed or altered from any source distribution.\n"
+"*/\n"
+"#ifndef B3_INT2_H\n"
+"#define B3_INT2_H\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+"#define b3UnsignedInt2 uint2\n"
+"#define b3Int2 int2\n"
+"#define b3MakeInt2 (int2)\n"
+"#endif //__cplusplus\n"
+"#endif\n"
+"typedef struct\n"
+"{\n"
+" float4 m_plane;\n"
+" int m_indexOffset;\n"
+" int m_numIndices;\n"
+"} btGpuFace;\n"
+"#define make_float4 (float4)\n"
+"__inline\n"
+"float4 cross3(float4 a, float4 b)\n"
+"{\n"
+" return cross(a,b);\n"
+" \n"
+"// float4 a1 = make_float4(a.xyz,0.f);\n"
+"// float4 b1 = make_float4(b.xyz,0.f);\n"
+"// return cross(a1,b1);\n"
+"//float4 c = make_float4(a.y*b.z - a.z*b.y,a.z*b.x - a.x*b.z,a.x*b.y - a.y*b.x,0.f);\n"
+" \n"
+" // float4 c = make_float4(a.y*b.z - a.z*b.y,1.f,a.x*b.y - a.y*b.x,0.f);\n"
+" \n"
+" //return c;\n"
+"}\n"
+"__inline\n"
+"float dot3F4(float4 a, float4 b)\n"
+"{\n"
+" float4 a1 = make_float4(a.xyz,0.f);\n"
+" float4 b1 = make_float4(b.xyz,0.f);\n"
+" return dot(a1, b1);\n"
+"}\n"
+"__inline\n"
+"float4 fastNormalize4(float4 v)\n"
+"{\n"
+" v = make_float4(v.xyz,0.f);\n"
+" return fast_normalize(v);\n"
+"}\n"
+"///////////////////////////////////////\n"
+"// Quaternion\n"
+"///////////////////////////////////////\n"
+"typedef float4 Quaternion;\n"
+"__inline\n"
+"Quaternion qtMul(Quaternion a, Quaternion b);\n"
+"__inline\n"
+"Quaternion qtNormalize(Quaternion in);\n"
+"__inline\n"
+"float4 qtRotate(Quaternion q, float4 vec);\n"
+"__inline\n"
+"Quaternion qtInvert(Quaternion q);\n"
+"__inline\n"
+"Quaternion qtMul(Quaternion a, Quaternion b)\n"
+"{\n"
+" Quaternion ans;\n"
+" ans = cross3( a, b );\n"
+" ans += a.w*b+b.w*a;\n"
+"// ans.w = a.w*b.w - (a.x*b.x+a.y*b.y+a.z*b.z);\n"
+" ans.w = a.w*b.w - dot3F4(a, b);\n"
+" return ans;\n"
+"}\n"
+"__inline\n"
+"Quaternion qtNormalize(Quaternion in)\n"
+"{\n"
+" return fastNormalize4(in);\n"
+"// in /= length( in );\n"
+"// return in;\n"
+"}\n"
+"__inline\n"
+"float4 qtRotate(Quaternion q, float4 vec)\n"
+"{\n"
+" Quaternion qInv = qtInvert( q );\n"
+" float4 vcpy = vec;\n"
+" vcpy.w = 0.f;\n"
+" float4 out = qtMul(qtMul(q,vcpy),qInv);\n"
+" return out;\n"
+"}\n"
+"__inline\n"
+"Quaternion qtInvert(Quaternion q)\n"
+"{\n"
+" return (Quaternion)(-q.xyz, q.w);\n"
+"}\n"
+"__inline\n"
+"float4 qtInvRotate(const Quaternion q, float4 vec)\n"
+"{\n"
+" return qtRotate( qtInvert( q ), vec );\n"
+"}\n"
+"__inline\n"
+"float4 transform(const float4* p, const float4* translation, const Quaternion* orientation)\n"
+"{\n"
+" return qtRotate( *orientation, *p ) + (*translation);\n"
+"}\n"
+"__inline\n"
+"float4 normalize3(const float4 a)\n"
+"{\n"
+" float4 n = make_float4(a.x, a.y, a.z, 0.f);\n"
+" return fastNormalize4( n );\n"
+"}\n"
+"inline void projectLocal(const ConvexPolyhedronCL* hull, const float4 pos, const float4 orn, \n"
+"const float4* dir, const float4* vertices, float* min, float* max)\n"
+"{\n"
+" min[0] = FLT_MAX;\n"
+" max[0] = -FLT_MAX;\n"
+" int numVerts = hull->m_numVertices;\n"
+" const float4 localDir = qtInvRotate(orn,*dir);\n"
+" float offset = dot(pos,*dir);\n"
+" for(int i=0;i<numVerts;i++)\n"
+" {\n"
+" float dp = dot(vertices[hull->m_vertexOffset+i],localDir);\n"
+" if(dp < min[0]) \n"
+" min[0] = dp;\n"
+" if(dp > max[0]) \n"
+" max[0] = dp;\n"
+" }\n"
+" if(min[0]>max[0])\n"
+" {\n"
+" float tmp = min[0];\n"
+" min[0] = max[0];\n"
+" max[0] = tmp;\n"
+" }\n"
+" min[0] += offset;\n"
+" max[0] += offset;\n"
+"}\n"
+"inline void project(__global const ConvexPolyhedronCL* hull, const float4 pos, const float4 orn, \n"
+"const float4* dir, __global const float4* vertices, float* min, float* max)\n"
+"{\n"
+" min[0] = FLT_MAX;\n"
+" max[0] = -FLT_MAX;\n"
+" int numVerts = hull->m_numVertices;\n"
+" const float4 localDir = qtInvRotate(orn,*dir);\n"
+" float offset = dot(pos,*dir);\n"
+" for(int i=0;i<numVerts;i++)\n"
+" {\n"
+" float dp = dot(vertices[hull->m_vertexOffset+i],localDir);\n"
+" if(dp < min[0]) \n"
+" min[0] = dp;\n"
+" if(dp > max[0]) \n"
+" max[0] = dp;\n"
+" }\n"
+" if(min[0]>max[0])\n"
+" {\n"
+" float tmp = min[0];\n"
+" min[0] = max[0];\n"
+" max[0] = tmp;\n"
+" }\n"
+" min[0] += offset;\n"
+" max[0] += offset;\n"
+"}\n"
+"inline bool TestSepAxisLocalA(const ConvexPolyhedronCL* hullA, __global const ConvexPolyhedronCL* hullB, \n"
+" const float4 posA,const float4 ornA,\n"
+" const float4 posB,const float4 ornB,\n"
+" float4* sep_axis, const float4* verticesA, __global const float4* verticesB,float* depth)\n"
+"{\n"
+" float Min0,Max0;\n"
+" float Min1,Max1;\n"
+" projectLocal(hullA,posA,ornA,sep_axis,verticesA, &Min0, &Max0);\n"
+" project(hullB,posB,ornB, sep_axis,verticesB, &Min1, &Max1);\n"
+" if(Max0<Min1 || Max1<Min0)\n"
+" return false;\n"
+" float d0 = Max0 - Min1;\n"
+" float d1 = Max1 - Min0;\n"
+" *depth = d0<d1 ? d0:d1;\n"
+" return true;\n"
+"}\n"
+"inline bool IsAlmostZero(const float4 v)\n"
+"{\n"
+" if(fabs(v.x)>1e-6f || fabs(v.y)>1e-6f || fabs(v.z)>1e-6f)\n"
+" return false;\n"
+" return true;\n"
+"}\n"
+"bool findSeparatingAxisLocalA( const ConvexPolyhedronCL* hullA, __global const ConvexPolyhedronCL* hullB, \n"
+" const float4 posA1,\n"
+" const float4 ornA,\n"
+" const float4 posB1,\n"
+" const float4 ornB,\n"
+" const float4 DeltaC2,\n"
+" \n"
+" const float4* verticesA, \n"
+" const float4* uniqueEdgesA, \n"
+" const btGpuFace* facesA,\n"
+" const int* indicesA,\n"
+" __global const float4* verticesB, \n"
+" __global const float4* uniqueEdgesB, \n"
+" __global const btGpuFace* facesB,\n"
+" __global const int* indicesB,\n"
+" float4* sep,\n"
+" float* dmin)\n"
+"{\n"
+" \n"
+" float4 posA = posA1;\n"
+" posA.w = 0.f;\n"
+" float4 posB = posB1;\n"
+" posB.w = 0.f;\n"
+" int curPlaneTests=0;\n"
+" {\n"
+" int numFacesA = hullA->m_numFaces;\n"
+" // Test normals from hullA\n"
+" for(int i=0;i<numFacesA;i++)\n"
+" {\n"
+" const float4 normal = facesA[hullA->m_faceOffset+i].m_plane;\n"
+" float4 faceANormalWS = qtRotate(ornA,normal);\n"
+" if (dot3F4(DeltaC2,faceANormalWS)<0)\n"
+" faceANormalWS*=-1.f;\n"
+" curPlaneTests++;\n"
+" float d;\n"
+" if(!TestSepAxisLocalA( hullA, hullB, posA,ornA,posB,ornB,&faceANormalWS, verticesA, verticesB,&d))\n"
+" return false;\n"
+" if(d<*dmin)\n"
+" {\n"
+" *dmin = d;\n"
+" *sep = faceANormalWS;\n"
+" }\n"
+" }\n"
+" }\n"
+" if((dot3F4(-DeltaC2,*sep))>0.0f)\n"
+" {\n"
+" *sep = -(*sep);\n"
+" }\n"
+" return true;\n"
+"}\n"
+"bool findSeparatingAxisLocalB( __global const ConvexPolyhedronCL* hullA, const ConvexPolyhedronCL* hullB, \n"
+" const float4 posA1,\n"
+" const float4 ornA,\n"
+" const float4 posB1,\n"
+" const float4 ornB,\n"
+" const float4 DeltaC2,\n"
+" __global const float4* verticesA, \n"
+" __global const float4* uniqueEdgesA, \n"
+" __global const btGpuFace* facesA,\n"
+" __global const int* indicesA,\n"
+" const float4* verticesB,\n"
+" const float4* uniqueEdgesB, \n"
+" const btGpuFace* facesB,\n"
+" const int* indicesB,\n"
+" float4* sep,\n"
+" float* dmin)\n"
+"{\n"
+" float4 posA = posA1;\n"
+" posA.w = 0.f;\n"
+" float4 posB = posB1;\n"
+" posB.w = 0.f;\n"
+" int curPlaneTests=0;\n"
+" {\n"
+" int numFacesA = hullA->m_numFaces;\n"
+" // Test normals from hullA\n"
+" for(int i=0;i<numFacesA;i++)\n"
+" {\n"
+" const float4 normal = facesA[hullA->m_faceOffset+i].m_plane;\n"
+" float4 faceANormalWS = qtRotate(ornA,normal);\n"
+" if (dot3F4(DeltaC2,faceANormalWS)<0)\n"
+" faceANormalWS *= -1.f;\n"
+" curPlaneTests++;\n"
+" float d;\n"
+" if(!TestSepAxisLocalA( hullB, hullA, posB,ornB,posA,ornA, &faceANormalWS, verticesB,verticesA, &d))\n"
+" return false;\n"
+" if(d<*dmin)\n"
+" {\n"
+" *dmin = d;\n"
+" *sep = faceANormalWS;\n"
+" }\n"
+" }\n"
+" }\n"
+" if((dot3F4(-DeltaC2,*sep))>0.0f)\n"
+" {\n"
+" *sep = -(*sep);\n"
+" }\n"
+" return true;\n"
+"}\n"
+"bool findSeparatingAxisEdgeEdgeLocalA( const ConvexPolyhedronCL* hullA, __global const ConvexPolyhedronCL* hullB, \n"
+" const float4 posA1,\n"
+" const float4 ornA,\n"
+" const float4 posB1,\n"
+" const float4 ornB,\n"
+" const float4 DeltaC2,\n"
+" const float4* verticesA, \n"
+" const float4* uniqueEdgesA, \n"
+" const btGpuFace* facesA,\n"
+" const int* indicesA,\n"
+" __global const float4* verticesB, \n"
+" __global const float4* uniqueEdgesB, \n"
+" __global const btGpuFace* facesB,\n"
+" __global const int* indicesB,\n"
+" float4* sep,\n"
+" float* dmin)\n"
+"{\n"
+" float4 posA = posA1;\n"
+" posA.w = 0.f;\n"
+" float4 posB = posB1;\n"
+" posB.w = 0.f;\n"
+" int curPlaneTests=0;\n"
+" int curEdgeEdge = 0;\n"
+" // Test edges\n"
+" for(int e0=0;e0<hullA->m_numUniqueEdges;e0++)\n"
+" {\n"
+" const float4 edge0 = uniqueEdgesA[hullA->m_uniqueEdgesOffset+e0];\n"
+" float4 edge0World = qtRotate(ornA,edge0);\n"
+" for(int e1=0;e1<hullB->m_numUniqueEdges;e1++)\n"
+" {\n"
+" const float4 edge1 = uniqueEdgesB[hullB->m_uniqueEdgesOffset+e1];\n"
+" float4 edge1World = qtRotate(ornB,edge1);\n"
+" float4 crossje = cross3(edge0World,edge1World);\n"
+" curEdgeEdge++;\n"
+" if(!IsAlmostZero(crossje))\n"
+" {\n"
+" crossje = normalize3(crossje);\n"
+" if (dot3F4(DeltaC2,crossje)<0)\n"
+" crossje *= -1.f;\n"
+" float dist;\n"
+" bool result = true;\n"
+" {\n"
+" float Min0,Max0;\n"
+" float Min1,Max1;\n"
+" projectLocal(hullA,posA,ornA,&crossje,verticesA, &Min0, &Max0);\n"
+" project(hullB,posB,ornB,&crossje,verticesB, &Min1, &Max1);\n"
+" \n"
+" if(Max0<Min1 || Max1<Min0)\n"
+" result = false;\n"
+" \n"
+" float d0 = Max0 - Min1;\n"
+" float d1 = Max1 - Min0;\n"
+" dist = d0<d1 ? d0:d1;\n"
+" result = true;\n"
+" }\n"
+" \n"
+" if(dist<*dmin)\n"
+" {\n"
+" *dmin = dist;\n"
+" *sep = crossje;\n"
+" }\n"
+" }\n"
+" }\n"
+" }\n"
+" \n"
+" if((dot3F4(-DeltaC2,*sep))>0.0f)\n"
+" {\n"
+" *sep = -(*sep);\n"
+" }\n"
+" return true;\n"
+"}\n"
+"inline int findClippingFaces(const float4 separatingNormal,\n"
+" const ConvexPolyhedronCL* hullA, \n"
+" __global const ConvexPolyhedronCL* hullB,\n"
+" const float4 posA, const Quaternion ornA,const float4 posB, const Quaternion ornB,\n"
+" __global float4* worldVertsA1,\n"
+" __global float4* worldNormalsA1,\n"
+" __global float4* worldVertsB1,\n"
+" int capacityWorldVerts,\n"
+" const float minDist, float maxDist,\n"
+" const float4* verticesA,\n"
+" const btGpuFace* facesA,\n"
+" const int* indicesA,\n"
+" __global const float4* verticesB,\n"
+" __global const btGpuFace* facesB,\n"
+" __global const int* indicesB,\n"
+" __global int4* clippingFaces, int pairIndex)\n"
+"{\n"
+" int numContactsOut = 0;\n"
+" int numWorldVertsB1= 0;\n"
+" \n"
+" \n"
+" int closestFaceB=0;\n"
+" float dmax = -FLT_MAX;\n"
+" \n"
+" {\n"
+" for(int face=0;face<hullB->m_numFaces;face++)\n"
+" {\n"
+" const float4 Normal = make_float4(facesB[hullB->m_faceOffset+face].m_plane.x,\n"
+" facesB[hullB->m_faceOffset+face].m_plane.y, facesB[hullB->m_faceOffset+face].m_plane.z,0.f);\n"
+" const float4 WorldNormal = qtRotate(ornB, Normal);\n"
+" float d = dot3F4(WorldNormal,separatingNormal);\n"
+" if (d > dmax)\n"
+" {\n"
+" dmax = d;\n"
+" closestFaceB = face;\n"
+" }\n"
+" }\n"
+" }\n"
+" \n"
+" {\n"
+" const btGpuFace polyB = facesB[hullB->m_faceOffset+closestFaceB];\n"
+" int numVertices = polyB.m_numIndices;\n"
+" if (numVertices>capacityWorldVerts)\n"
+" numVertices = capacityWorldVerts;\n"
+" if (numVertices<0)\n"
+" numVertices = 0;\n"
+" \n"
+" for(int e0=0;e0<numVertices;e0++)\n"
+" {\n"
+" if (e0<capacityWorldVerts)\n"
+" {\n"
+" const float4 b = verticesB[hullB->m_vertexOffset+indicesB[polyB.m_indexOffset+e0]];\n"
+" worldVertsB1[pairIndex*capacityWorldVerts+numWorldVertsB1++] = transform(&b,&posB,&ornB);\n"
+" }\n"
+" }\n"
+" }\n"
+" \n"
+" int closestFaceA=0;\n"
+" {\n"
+" float dmin = FLT_MAX;\n"
+" for(int face=0;face<hullA->m_numFaces;face++)\n"
+" {\n"
+" const float4 Normal = make_float4(\n"
+" facesA[hullA->m_faceOffset+face].m_plane.x,\n"
+" facesA[hullA->m_faceOffset+face].m_plane.y,\n"
+" facesA[hullA->m_faceOffset+face].m_plane.z,\n"
+" 0.f);\n"
+" const float4 faceANormalWS = qtRotate(ornA,Normal);\n"
+" \n"
+" float d = dot3F4(faceANormalWS,separatingNormal);\n"
+" if (d < dmin)\n"
+" {\n"
+" dmin = d;\n"
+" closestFaceA = face;\n"
+" worldNormalsA1[pairIndex] = faceANormalWS;\n"
+" }\n"
+" }\n"
+" }\n"
+" \n"
+" int numVerticesA = facesA[hullA->m_faceOffset+closestFaceA].m_numIndices;\n"
+" if (numVerticesA>capacityWorldVerts)\n"
+" numVerticesA = capacityWorldVerts;\n"
+" if (numVerticesA<0)\n"
+" numVerticesA=0;\n"
+" \n"
+" for(int e0=0;e0<numVerticesA;e0++)\n"
+" {\n"
+" if (e0<capacityWorldVerts)\n"
+" {\n"
+" const float4 a = verticesA[hullA->m_vertexOffset+indicesA[facesA[hullA->m_faceOffset+closestFaceA].m_indexOffset+e0]];\n"
+" worldVertsA1[pairIndex*capacityWorldVerts+e0] = transform(&a, &posA,&ornA);\n"
+" }\n"
+" }\n"
+" \n"
+" clippingFaces[pairIndex].x = closestFaceA;\n"
+" clippingFaces[pairIndex].y = closestFaceB;\n"
+" clippingFaces[pairIndex].z = numVerticesA;\n"
+" clippingFaces[pairIndex].w = numWorldVertsB1;\n"
+" \n"
+" \n"
+" return numContactsOut;\n"
+"}\n"
+"// work-in-progress\n"
+"__kernel void findConcaveSeparatingAxisVertexFaceKernel( __global int4* concavePairs,\n"
+" __global const BodyData* rigidBodies,\n"
+" __global const btCollidableGpu* collidables,\n"
+" __global const ConvexPolyhedronCL* convexShapes,\n"
+" __global const float4* vertices,\n"
+" __global const float4* uniqueEdges,\n"
+" __global const btGpuFace* faces,\n"
+" __global const int* indices,\n"
+" __global const btGpuChildShape* gpuChildShapes,\n"
+" __global btAabbCL* aabbs,\n"
+" __global float4* concaveSeparatingNormalsOut,\n"
+" __global int* concaveHasSeparatingNormals,\n"
+" __global int4* clippingFacesOut,\n"
+" __global float4* worldVertsA1GPU,\n"
+" __global float4* worldNormalsAGPU,\n"
+" __global float4* worldVertsB1GPU,\n"
+" __global float* dmins,\n"
+" int vertexFaceCapacity,\n"
+" int numConcavePairs\n"
+" )\n"
+"{\n"
+" \n"
+" int i = get_global_id(0);\n"
+" if (i>=numConcavePairs)\n"
+" return;\n"
+" \n"
+" concaveHasSeparatingNormals[i] = 0;\n"
+" \n"
+" int pairIdx = i;\n"
+" \n"
+" int bodyIndexA = concavePairs[i].x;\n"
+" int bodyIndexB = concavePairs[i].y;\n"
+" \n"
+" int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;\n"
+" int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;\n"
+" \n"
+" int shapeIndexA = collidables[collidableIndexA].m_shapeIndex;\n"
+" int shapeIndexB = collidables[collidableIndexB].m_shapeIndex;\n"
+" \n"
+" if (collidables[collidableIndexB].m_shapeType!=SHAPE_CONVEX_HULL&&\n"
+" collidables[collidableIndexB].m_shapeType!=SHAPE_COMPOUND_OF_CONVEX_HULLS)\n"
+" {\n"
+" concavePairs[pairIdx].w = -1;\n"
+" return;\n"
+" }\n"
+" \n"
+" \n"
+" \n"
+" int numFacesA = convexShapes[shapeIndexA].m_numFaces;\n"
+" int numActualConcaveConvexTests = 0;\n"
+" \n"
+" int f = concavePairs[i].z;\n"
+" \n"
+" bool overlap = false;\n"
+" \n"
+" ConvexPolyhedronCL convexPolyhedronA;\n"
+" \n"
+" //add 3 vertices of the triangle\n"
+" convexPolyhedronA.m_numVertices = 3;\n"
+" convexPolyhedronA.m_vertexOffset = 0;\n"
+" float4 localCenter = make_float4(0.f,0.f,0.f,0.f);\n"
+" \n"
+" btGpuFace face = faces[convexShapes[shapeIndexA].m_faceOffset+f];\n"
+" float4 triMinAabb, triMaxAabb;\n"
+" btAabbCL triAabb;\n"
+" triAabb.m_min = make_float4(1e30f,1e30f,1e30f,0.f);\n"
+" triAabb.m_max = make_float4(-1e30f,-1e30f,-1e30f,0.f);\n"
+" \n"
+" float4 verticesA[3];\n"
+" for (int i=0;i<3;i++)\n"
+" {\n"
+" int index = indices[face.m_indexOffset+i];\n"
+" float4 vert = vertices[convexShapes[shapeIndexA].m_vertexOffset+index];\n"
+" verticesA[i] = vert;\n"
+" localCenter += vert;\n"
+" \n"
+" triAabb.m_min = min(triAabb.m_min,vert);\n"
+" triAabb.m_max = max(triAabb.m_max,vert);\n"
+" \n"
+" }\n"
+" \n"
+" overlap = true;\n"
+" overlap = (triAabb.m_min.x > aabbs[bodyIndexB].m_max.x || triAabb.m_max.x < aabbs[bodyIndexB].m_min.x) ? false : overlap;\n"
+" overlap = (triAabb.m_min.z > aabbs[bodyIndexB].m_max.z || triAabb.m_max.z < aabbs[bodyIndexB].m_min.z) ? false : overlap;\n"
+" overlap = (triAabb.m_min.y > aabbs[bodyIndexB].m_max.y || triAabb.m_max.y < aabbs[bodyIndexB].m_min.y) ? false : overlap;\n"
+" \n"
+" if (overlap)\n"
+" {\n"
+" float dmin = FLT_MAX;\n"
+" int hasSeparatingAxis=5;\n"
+" float4 sepAxis=make_float4(1,2,3,4);\n"
+" \n"
+" int localCC=0;\n"
+" numActualConcaveConvexTests++;\n"
+" \n"
+" //a triangle has 3 unique edges\n"
+" convexPolyhedronA.m_numUniqueEdges = 3;\n"
+" convexPolyhedronA.m_uniqueEdgesOffset = 0;\n"
+" float4 uniqueEdgesA[3];\n"
+" \n"
+" uniqueEdgesA[0] = (verticesA[1]-verticesA[0]);\n"
+" uniqueEdgesA[1] = (verticesA[2]-verticesA[1]);\n"
+" uniqueEdgesA[2] = (verticesA[0]-verticesA[2]);\n"
+" \n"
+" \n"
+" convexPolyhedronA.m_faceOffset = 0;\n"
+" \n"
+" float4 normal = make_float4(face.m_plane.x,face.m_plane.y,face.m_plane.z,0.f);\n"
+" \n"
+" btGpuFace facesA[TRIANGLE_NUM_CONVEX_FACES];\n"
+" int indicesA[3+3+2+2+2];\n"
+" int curUsedIndices=0;\n"
+" int fidx=0;\n"
+" \n"
+" //front size of triangle\n"
+" {\n"
+" facesA[fidx].m_indexOffset=curUsedIndices;\n"
+" indicesA[0] = 0;\n"
+" indicesA[1] = 1;\n"
+" indicesA[2] = 2;\n"
+" curUsedIndices+=3;\n"
+" float c = face.m_plane.w;\n"
+" facesA[fidx].m_plane.x = normal.x;\n"
+" facesA[fidx].m_plane.y = normal.y;\n"
+" facesA[fidx].m_plane.z = normal.z;\n"
+" facesA[fidx].m_plane.w = c;\n"
+" facesA[fidx].m_numIndices=3;\n"
+" }\n"
+" fidx++;\n"
+" //back size of triangle\n"
+" {\n"
+" facesA[fidx].m_indexOffset=curUsedIndices;\n"
+" indicesA[3]=2;\n"
+" indicesA[4]=1;\n"
+" indicesA[5]=0;\n"
+" curUsedIndices+=3;\n"
+" float c = dot(normal,verticesA[0]);\n"
+" float c1 = -face.m_plane.w;\n"
+" facesA[fidx].m_plane.x = -normal.x;\n"
+" facesA[fidx].m_plane.y = -normal.y;\n"
+" facesA[fidx].m_plane.z = -normal.z;\n"
+" facesA[fidx].m_plane.w = c;\n"
+" facesA[fidx].m_numIndices=3;\n"
+" }\n"
+" fidx++;\n"
+" \n"
+" bool addEdgePlanes = true;\n"
+" if (addEdgePlanes)\n"
+" {\n"
+" int numVertices=3;\n"
+" int prevVertex = numVertices-1;\n"
+" for (int i=0;i<numVertices;i++)\n"
+" {\n"
+" float4 v0 = verticesA[i];\n"
+" float4 v1 = verticesA[prevVertex];\n"
+" \n"
+" float4 edgeNormal = normalize(cross(normal,v1-v0));\n"
+" float c = -dot(edgeNormal,v0);\n"
+" \n"
+" facesA[fidx].m_numIndices = 2;\n"
+" facesA[fidx].m_indexOffset=curUsedIndices;\n"
+" indicesA[curUsedIndices++]=i;\n"
+" indicesA[curUsedIndices++]=prevVertex;\n"
+" \n"
+" facesA[fidx].m_plane.x = edgeNormal.x;\n"
+" facesA[fidx].m_plane.y = edgeNormal.y;\n"
+" facesA[fidx].m_plane.z = edgeNormal.z;\n"
+" facesA[fidx].m_plane.w = c;\n"
+" fidx++;\n"
+" prevVertex = i;\n"
+" }\n"
+" }\n"
+" convexPolyhedronA.m_numFaces = TRIANGLE_NUM_CONVEX_FACES;\n"
+" convexPolyhedronA.m_localCenter = localCenter*(1.f/3.f);\n"
+" \n"
+" \n"
+" float4 posA = rigidBodies[bodyIndexA].m_pos;\n"
+" posA.w = 0.f;\n"
+" float4 posB = rigidBodies[bodyIndexB].m_pos;\n"
+" posB.w = 0.f;\n"
+" \n"
+" float4 ornA = rigidBodies[bodyIndexA].m_quat;\n"
+" float4 ornB =rigidBodies[bodyIndexB].m_quat;\n"
+" \n"
+" \n"
+" \n"
+" \n"
+" ///////////////////\n"
+" ///compound shape support\n"
+" \n"
+" if (collidables[collidableIndexB].m_shapeType==SHAPE_COMPOUND_OF_CONVEX_HULLS)\n"
+" {\n"
+" int compoundChild = concavePairs[pairIdx].w;\n"
+" int childShapeIndexB = compoundChild;//collidables[collidableIndexB].m_shapeIndex+compoundChild;\n"
+" int childColIndexB = gpuChildShapes[childShapeIndexB].m_shapeIndex;\n"
+" float4 childPosB = gpuChildShapes[childShapeIndexB].m_childPosition;\n"
+" float4 childOrnB = gpuChildShapes[childShapeIndexB].m_childOrientation;\n"
+" float4 newPosB = transform(&childPosB,&posB,&ornB);\n"
+" float4 newOrnB = qtMul(ornB,childOrnB);\n"
+" posB = newPosB;\n"
+" ornB = newOrnB;\n"
+" shapeIndexB = collidables[childColIndexB].m_shapeIndex;\n"
+" }\n"
+" //////////////////\n"
+" \n"
+" float4 c0local = convexPolyhedronA.m_localCenter;\n"
+" float4 c0 = transform(&c0local, &posA, &ornA);\n"
+" float4 c1local = convexShapes[shapeIndexB].m_localCenter;\n"
+" float4 c1 = transform(&c1local,&posB,&ornB);\n"
+" const float4 DeltaC2 = c0 - c1;\n"
+" \n"
+" \n"
+" bool sepA = findSeparatingAxisLocalA( &convexPolyhedronA, &convexShapes[shapeIndexB],\n"
+" posA,ornA,\n"
+" posB,ornB,\n"
+" DeltaC2,\n"
+" verticesA,uniqueEdgesA,facesA,indicesA,\n"
+" vertices,uniqueEdges,faces,indices,\n"
+" &sepAxis,&dmin);\n"
+" hasSeparatingAxis = 4;\n"
+" if (!sepA)\n"
+" {\n"
+" hasSeparatingAxis = 0;\n"
+" } else\n"
+" {\n"
+" bool sepB = findSeparatingAxisLocalB( &convexShapes[shapeIndexB],&convexPolyhedronA,\n"
+" posB,ornB,\n"
+" posA,ornA,\n"
+" DeltaC2,\n"
+" vertices,uniqueEdges,faces,indices,\n"
+" verticesA,uniqueEdgesA,facesA,indicesA,\n"
+" &sepAxis,&dmin);\n"
+" \n"
+" if (!sepB)\n"
+" {\n"
+" hasSeparatingAxis = 0;\n"
+" } else\n"
+" {\n"
+" hasSeparatingAxis = 1;\n"
+" }\n"
+" } \n"
+" \n"
+" if (hasSeparatingAxis)\n"
+" {\n"
+" dmins[i] = dmin;\n"
+" concaveSeparatingNormalsOut[pairIdx]=sepAxis;\n"
+" concaveHasSeparatingNormals[i]=1;\n"
+" \n"
+" } else\n"
+" { \n"
+" //mark this pair as in-active\n"
+" concavePairs[pairIdx].w = -1;\n"
+" }\n"
+" }\n"
+" else\n"
+" { \n"
+" //mark this pair as in-active\n"
+" concavePairs[pairIdx].w = -1;\n"
+" }\n"
+"}\n"
+"// work-in-progress\n"
+"__kernel void findConcaveSeparatingAxisEdgeEdgeKernel( __global int4* concavePairs,\n"
+" __global const BodyData* rigidBodies,\n"
+" __global const btCollidableGpu* collidables,\n"
+" __global const ConvexPolyhedronCL* convexShapes,\n"
+" __global const float4* vertices,\n"
+" __global const float4* uniqueEdges,\n"
+" __global const btGpuFace* faces,\n"
+" __global const int* indices,\n"
+" __global const btGpuChildShape* gpuChildShapes,\n"
+" __global btAabbCL* aabbs,\n"
+" __global float4* concaveSeparatingNormalsOut,\n"
+" __global int* concaveHasSeparatingNormals,\n"
+" __global int4* clippingFacesOut,\n"
+" __global float4* worldVertsA1GPU,\n"
+" __global float4* worldNormalsAGPU,\n"
+" __global float4* worldVertsB1GPU,\n"
+" __global float* dmins,\n"
+" int vertexFaceCapacity,\n"
+" int numConcavePairs\n"
+" )\n"
+"{\n"
+" \n"
+" int i = get_global_id(0);\n"
+" if (i>=numConcavePairs)\n"
+" return;\n"
+" \n"
+" if (!concaveHasSeparatingNormals[i])\n"
+" return;\n"
+" \n"
+" int pairIdx = i;\n"
+" \n"
+" int bodyIndexA = concavePairs[i].x;\n"
+" int bodyIndexB = concavePairs[i].y;\n"
+" \n"
+" int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;\n"
+" int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;\n"
+" \n"
+" int shapeIndexA = collidables[collidableIndexA].m_shapeIndex;\n"
+" int shapeIndexB = collidables[collidableIndexB].m_shapeIndex;\n"
+" \n"
+" \n"
+" int numFacesA = convexShapes[shapeIndexA].m_numFaces;\n"
+" int numActualConcaveConvexTests = 0;\n"
+" \n"
+" int f = concavePairs[i].z;\n"
+" \n"
+" bool overlap = false;\n"
+" \n"
+" ConvexPolyhedronCL convexPolyhedronA;\n"
+" \n"
+" //add 3 vertices of the triangle\n"
+" convexPolyhedronA.m_numVertices = 3;\n"
+" convexPolyhedronA.m_vertexOffset = 0;\n"
+" float4 localCenter = make_float4(0.f,0.f,0.f,0.f);\n"
+" \n"
+" btGpuFace face = faces[convexShapes[shapeIndexA].m_faceOffset+f];\n"
+" float4 triMinAabb, triMaxAabb;\n"
+" btAabbCL triAabb;\n"
+" triAabb.m_min = make_float4(1e30f,1e30f,1e30f,0.f);\n"
+" triAabb.m_max = make_float4(-1e30f,-1e30f,-1e30f,0.f);\n"
+" \n"
+" float4 verticesA[3];\n"
+" for (int i=0;i<3;i++)\n"
+" {\n"
+" int index = indices[face.m_indexOffset+i];\n"
+" float4 vert = vertices[convexShapes[shapeIndexA].m_vertexOffset+index];\n"
+" verticesA[i] = vert;\n"
+" localCenter += vert;\n"
+" \n"
+" triAabb.m_min = min(triAabb.m_min,vert);\n"
+" triAabb.m_max = max(triAabb.m_max,vert);\n"
+" \n"
+" }\n"
+" \n"
+" overlap = true;\n"
+" overlap = (triAabb.m_min.x > aabbs[bodyIndexB].m_max.x || triAabb.m_max.x < aabbs[bodyIndexB].m_min.x) ? false : overlap;\n"
+" overlap = (triAabb.m_min.z > aabbs[bodyIndexB].m_max.z || triAabb.m_max.z < aabbs[bodyIndexB].m_min.z) ? false : overlap;\n"
+" overlap = (triAabb.m_min.y > aabbs[bodyIndexB].m_max.y || triAabb.m_max.y < aabbs[bodyIndexB].m_min.y) ? false : overlap;\n"
+" \n"
+" if (overlap)\n"
+" {\n"
+" float dmin = dmins[i];\n"
+" int hasSeparatingAxis=5;\n"
+" float4 sepAxis=make_float4(1,2,3,4);\n"
+" sepAxis = concaveSeparatingNormalsOut[pairIdx];\n"
+" \n"
+" int localCC=0;\n"
+" numActualConcaveConvexTests++;\n"
+" \n"
+" //a triangle has 3 unique edges\n"
+" convexPolyhedronA.m_numUniqueEdges = 3;\n"
+" convexPolyhedronA.m_uniqueEdgesOffset = 0;\n"
+" float4 uniqueEdgesA[3];\n"
+" \n"
+" uniqueEdgesA[0] = (verticesA[1]-verticesA[0]);\n"
+" uniqueEdgesA[1] = (verticesA[2]-verticesA[1]);\n"
+" uniqueEdgesA[2] = (verticesA[0]-verticesA[2]);\n"
+" \n"
+" \n"
+" convexPolyhedronA.m_faceOffset = 0;\n"
+" \n"
+" float4 normal = make_float4(face.m_plane.x,face.m_plane.y,face.m_plane.z,0.f);\n"
+" \n"
+" btGpuFace facesA[TRIANGLE_NUM_CONVEX_FACES];\n"
+" int indicesA[3+3+2+2+2];\n"
+" int curUsedIndices=0;\n"
+" int fidx=0;\n"
+" \n"
+" //front size of triangle\n"
+" {\n"
+" facesA[fidx].m_indexOffset=curUsedIndices;\n"
+" indicesA[0] = 0;\n"
+" indicesA[1] = 1;\n"
+" indicesA[2] = 2;\n"
+" curUsedIndices+=3;\n"
+" float c = face.m_plane.w;\n"
+" facesA[fidx].m_plane.x = normal.x;\n"
+" facesA[fidx].m_plane.y = normal.y;\n"
+" facesA[fidx].m_plane.z = normal.z;\n"
+" facesA[fidx].m_plane.w = c;\n"
+" facesA[fidx].m_numIndices=3;\n"
+" }\n"
+" fidx++;\n"
+" //back size of triangle\n"
+" {\n"
+" facesA[fidx].m_indexOffset=curUsedIndices;\n"
+" indicesA[3]=2;\n"
+" indicesA[4]=1;\n"
+" indicesA[5]=0;\n"
+" curUsedIndices+=3;\n"
+" float c = dot(normal,verticesA[0]);\n"
+" float c1 = -face.m_plane.w;\n"
+" facesA[fidx].m_plane.x = -normal.x;\n"
+" facesA[fidx].m_plane.y = -normal.y;\n"
+" facesA[fidx].m_plane.z = -normal.z;\n"
+" facesA[fidx].m_plane.w = c;\n"
+" facesA[fidx].m_numIndices=3;\n"
+" }\n"
+" fidx++;\n"
+" \n"
+" bool addEdgePlanes = true;\n"
+" if (addEdgePlanes)\n"
+" {\n"
+" int numVertices=3;\n"
+" int prevVertex = numVertices-1;\n"
+" for (int i=0;i<numVertices;i++)\n"
+" {\n"
+" float4 v0 = verticesA[i];\n"
+" float4 v1 = verticesA[prevVertex];\n"
+" \n"
+" float4 edgeNormal = normalize(cross(normal,v1-v0));\n"
+" float c = -dot(edgeNormal,v0);\n"
+" \n"
+" facesA[fidx].m_numIndices = 2;\n"
+" facesA[fidx].m_indexOffset=curUsedIndices;\n"
+" indicesA[curUsedIndices++]=i;\n"
+" indicesA[curUsedIndices++]=prevVertex;\n"
+" \n"
+" facesA[fidx].m_plane.x = edgeNormal.x;\n"
+" facesA[fidx].m_plane.y = edgeNormal.y;\n"
+" facesA[fidx].m_plane.z = edgeNormal.z;\n"
+" facesA[fidx].m_plane.w = c;\n"
+" fidx++;\n"
+" prevVertex = i;\n"
+" }\n"
+" }\n"
+" convexPolyhedronA.m_numFaces = TRIANGLE_NUM_CONVEX_FACES;\n"
+" convexPolyhedronA.m_localCenter = localCenter*(1.f/3.f);\n"
+" \n"
+" \n"
+" float4 posA = rigidBodies[bodyIndexA].m_pos;\n"
+" posA.w = 0.f;\n"
+" float4 posB = rigidBodies[bodyIndexB].m_pos;\n"
+" posB.w = 0.f;\n"
+" \n"
+" float4 ornA = rigidBodies[bodyIndexA].m_quat;\n"
+" float4 ornB =rigidBodies[bodyIndexB].m_quat;\n"
+" \n"
+" \n"
+" \n"
+" \n"
+" ///////////////////\n"
+" ///compound shape support\n"
+" \n"
+" if (collidables[collidableIndexB].m_shapeType==SHAPE_COMPOUND_OF_CONVEX_HULLS)\n"
+" {\n"
+" int compoundChild = concavePairs[pairIdx].w;\n"
+" int childShapeIndexB = compoundChild;//collidables[collidableIndexB].m_shapeIndex+compoundChild;\n"
+" int childColIndexB = gpuChildShapes[childShapeIndexB].m_shapeIndex;\n"
+" float4 childPosB = gpuChildShapes[childShapeIndexB].m_childPosition;\n"
+" float4 childOrnB = gpuChildShapes[childShapeIndexB].m_childOrientation;\n"
+" float4 newPosB = transform(&childPosB,&posB,&ornB);\n"
+" float4 newOrnB = qtMul(ornB,childOrnB);\n"
+" posB = newPosB;\n"
+" ornB = newOrnB;\n"
+" shapeIndexB = collidables[childColIndexB].m_shapeIndex;\n"
+" }\n"
+" //////////////////\n"
+" \n"
+" float4 c0local = convexPolyhedronA.m_localCenter;\n"
+" float4 c0 = transform(&c0local, &posA, &ornA);\n"
+" float4 c1local = convexShapes[shapeIndexB].m_localCenter;\n"
+" float4 c1 = transform(&c1local,&posB,&ornB);\n"
+" const float4 DeltaC2 = c0 - c1;\n"
+" \n"
+" \n"
+" {\n"
+" bool sepEE = findSeparatingAxisEdgeEdgeLocalA( &convexPolyhedronA, &convexShapes[shapeIndexB],\n"
+" posA,ornA,\n"
+" posB,ornB,\n"
+" DeltaC2,\n"
+" verticesA,uniqueEdgesA,facesA,indicesA,\n"
+" vertices,uniqueEdges,faces,indices,\n"
+" &sepAxis,&dmin);\n"
+" \n"
+" if (!sepEE)\n"
+" {\n"
+" hasSeparatingAxis = 0;\n"
+" } else\n"
+" {\n"
+" hasSeparatingAxis = 1;\n"
+" }\n"
+" }\n"
+" \n"
+" \n"
+" if (hasSeparatingAxis)\n"
+" {\n"
+" sepAxis.w = dmin;\n"
+" dmins[i] = dmin;\n"
+" concaveSeparatingNormalsOut[pairIdx]=sepAxis;\n"
+" concaveHasSeparatingNormals[i]=1;\n"
+" \n"
+" float minDist = -1e30f;\n"
+" float maxDist = 0.02f;\n"
+" \n"
+" findClippingFaces(sepAxis,\n"
+" &convexPolyhedronA,\n"
+" &convexShapes[shapeIndexB],\n"
+" posA,ornA,\n"
+" posB,ornB,\n"
+" worldVertsA1GPU,\n"
+" worldNormalsAGPU,\n"
+" worldVertsB1GPU,\n"
+" vertexFaceCapacity,\n"
+" minDist, maxDist,\n"
+" verticesA,\n"
+" facesA,\n"
+" indicesA,\n"
+" vertices,\n"
+" faces,\n"
+" indices,\n"
+" clippingFacesOut, pairIdx);\n"
+" \n"
+" \n"
+" } else\n"
+" { \n"
+" //mark this pair as in-active\n"
+" concavePairs[pairIdx].w = -1;\n"
+" }\n"
+" }\n"
+" else\n"
+" { \n"
+" //mark this pair as in-active\n"
+" concavePairs[pairIdx].w = -1;\n"
+" }\n"
+" \n"
+" concavePairs[i].z = -1;//for the next stage, z is used to determine existing contact points\n"
+"}\n"
+;
diff --git a/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/satKernels.h b/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/satKernels.h
new file mode 100644
index 0000000000..6f8b0a90db
--- /dev/null
+++ b/thirdparty/bullet/Bullet3OpenCL/NarrowphaseCollision/kernels/satKernels.h
@@ -0,0 +1,2104 @@
+//this file is autogenerated using stringify.bat (premake --stringify) in the build folder of this project
+static const char* satKernelsCL= \
+"//keep this enum in sync with the CPU version (in btCollidable.h)\n"
+"//written by Erwin Coumans\n"
+"#define SHAPE_CONVEX_HULL 3\n"
+"#define SHAPE_CONCAVE_TRIMESH 5\n"
+"#define TRIANGLE_NUM_CONVEX_FACES 5\n"
+"#define SHAPE_COMPOUND_OF_CONVEX_HULLS 6\n"
+"#define B3_MAX_STACK_DEPTH 256\n"
+"typedef unsigned int u32;\n"
+"///keep this in sync with btCollidable.h\n"
+"typedef struct\n"
+"{\n"
+" union {\n"
+" int m_numChildShapes;\n"
+" int m_bvhIndex;\n"
+" };\n"
+" union\n"
+" {\n"
+" float m_radius;\n"
+" int m_compoundBvhIndex;\n"
+" };\n"
+" \n"
+" int m_shapeType;\n"
+" int m_shapeIndex;\n"
+" \n"
+"} btCollidableGpu;\n"
+"#define MAX_NUM_PARTS_IN_BITS 10\n"
+"///b3QuantizedBvhNode is a compressed aabb node, 16 bytes.\n"
+"///Node can be used for leafnode or internal node. Leafnodes can point to 32-bit triangle index (non-negative range).\n"
+"typedef struct\n"
+"{\n"
+" //12 bytes\n"
+" unsigned short int m_quantizedAabbMin[3];\n"
+" unsigned short int m_quantizedAabbMax[3];\n"
+" //4 bytes\n"
+" int m_escapeIndexOrTriangleIndex;\n"
+"} b3QuantizedBvhNode;\n"
+"typedef struct\n"
+"{\n"
+" float4 m_aabbMin;\n"
+" float4 m_aabbMax;\n"
+" float4 m_quantization;\n"
+" int m_numNodes;\n"
+" int m_numSubTrees;\n"
+" int m_nodeOffset;\n"
+" int m_subTreeOffset;\n"
+"} b3BvhInfo;\n"
+"int getTriangleIndex(const b3QuantizedBvhNode* rootNode)\n"
+"{\n"
+" unsigned int x=0;\n"
+" unsigned int y = (~(x&0))<<(31-MAX_NUM_PARTS_IN_BITS);\n"
+" // Get only the lower bits where the triangle index is stored\n"
+" return (rootNode->m_escapeIndexOrTriangleIndex&~(y));\n"
+"}\n"
+"int getTriangleIndexGlobal(__global const b3QuantizedBvhNode* rootNode)\n"
+"{\n"
+" unsigned int x=0;\n"
+" unsigned int y = (~(x&0))<<(31-MAX_NUM_PARTS_IN_BITS);\n"
+" // Get only the lower bits where the triangle index is stored\n"
+" return (rootNode->m_escapeIndexOrTriangleIndex&~(y));\n"
+"}\n"
+"int isLeafNode(const b3QuantizedBvhNode* rootNode)\n"
+"{\n"
+" //skipindex is negative (internal node), triangleindex >=0 (leafnode)\n"
+" return (rootNode->m_escapeIndexOrTriangleIndex >= 0)? 1 : 0;\n"
+"}\n"
+"int isLeafNodeGlobal(__global const b3QuantizedBvhNode* rootNode)\n"
+"{\n"
+" //skipindex is negative (internal node), triangleindex >=0 (leafnode)\n"
+" return (rootNode->m_escapeIndexOrTriangleIndex >= 0)? 1 : 0;\n"
+"}\n"
+" \n"
+"int getEscapeIndex(const b3QuantizedBvhNode* rootNode)\n"
+"{\n"
+" return -rootNode->m_escapeIndexOrTriangleIndex;\n"
+"}\n"
+"int getEscapeIndexGlobal(__global const b3QuantizedBvhNode* rootNode)\n"
+"{\n"
+" return -rootNode->m_escapeIndexOrTriangleIndex;\n"
+"}\n"
+"typedef struct\n"
+"{\n"
+" //12 bytes\n"
+" unsigned short int m_quantizedAabbMin[3];\n"
+" unsigned short int m_quantizedAabbMax[3];\n"
+" //4 bytes, points to the root of the subtree\n"
+" int m_rootNodeIndex;\n"
+" //4 bytes\n"
+" int m_subtreeSize;\n"
+" int m_padding[3];\n"
+"} b3BvhSubtreeInfo;\n"
+"typedef struct\n"
+"{\n"
+" float4 m_childPosition;\n"
+" float4 m_childOrientation;\n"
+" int m_shapeIndex;\n"
+" int m_unused0;\n"
+" int m_unused1;\n"
+" int m_unused2;\n"
+"} btGpuChildShape;\n"
+"typedef struct\n"
+"{\n"
+" float4 m_pos;\n"
+" float4 m_quat;\n"
+" float4 m_linVel;\n"
+" float4 m_angVel;\n"
+" u32 m_collidableIdx;\n"
+" float m_invMass;\n"
+" float m_restituitionCoeff;\n"
+" float m_frictionCoeff;\n"
+"} BodyData;\n"
+"typedef struct \n"
+"{\n"
+" float4 m_localCenter;\n"
+" float4 m_extents;\n"
+" float4 mC;\n"
+" float4 mE;\n"
+" \n"
+" float m_radius;\n"
+" int m_faceOffset;\n"
+" int m_numFaces;\n"
+" int m_numVertices;\n"
+" int m_vertexOffset;\n"
+" int m_uniqueEdgesOffset;\n"
+" int m_numUniqueEdges;\n"
+" int m_unused;\n"
+"} ConvexPolyhedronCL;\n"
+"typedef struct \n"
+"{\n"
+" union\n"
+" {\n"
+" float4 m_min;\n"
+" float m_minElems[4];\n"
+" int m_minIndices[4];\n"
+" };\n"
+" union\n"
+" {\n"
+" float4 m_max;\n"
+" float m_maxElems[4];\n"
+" int m_maxIndices[4];\n"
+" };\n"
+"} btAabbCL;\n"
+"#ifndef B3_AABB_H\n"
+"#define B3_AABB_H\n"
+"#ifndef B3_FLOAT4_H\n"
+"#define B3_FLOAT4_H\n"
+"#ifndef B3_PLATFORM_DEFINITIONS_H\n"
+"#define B3_PLATFORM_DEFINITIONS_H\n"
+"struct MyTest\n"
+"{\n"
+" int bla;\n"
+"};\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+"//keep B3_LARGE_FLOAT*B3_LARGE_FLOAT < FLT_MAX\n"
+"#define B3_LARGE_FLOAT 1e18f\n"
+"#define B3_INFINITY 1e18f\n"
+"#define b3Assert(a)\n"
+"#define b3ConstArray(a) __global const a*\n"
+"#define b3AtomicInc atomic_inc\n"
+"#define b3AtomicAdd atomic_add\n"
+"#define b3Fabs fabs\n"
+"#define b3Sqrt native_sqrt\n"
+"#define b3Sin native_sin\n"
+"#define b3Cos native_cos\n"
+"#define B3_STATIC\n"
+"#endif\n"
+"#endif\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+" typedef float4 b3Float4;\n"
+" #define b3Float4ConstArg const b3Float4\n"
+" #define b3MakeFloat4 (float4)\n"
+" float b3Dot3F4(b3Float4ConstArg v0,b3Float4ConstArg v1)\n"
+" {\n"
+" float4 a1 = b3MakeFloat4(v0.xyz,0.f);\n"
+" float4 b1 = b3MakeFloat4(v1.xyz,0.f);\n"
+" return dot(a1, b1);\n"
+" }\n"
+" b3Float4 b3Cross3(b3Float4ConstArg v0,b3Float4ConstArg v1)\n"
+" {\n"
+" float4 a1 = b3MakeFloat4(v0.xyz,0.f);\n"
+" float4 b1 = b3MakeFloat4(v1.xyz,0.f);\n"
+" return cross(a1, b1);\n"
+" }\n"
+" #define b3MinFloat4 min\n"
+" #define b3MaxFloat4 max\n"
+" #define b3Normalized(a) normalize(a)\n"
+"#endif \n"
+" \n"
+"inline bool b3IsAlmostZero(b3Float4ConstArg v)\n"
+"{\n"
+" if(b3Fabs(v.x)>1e-6 || b3Fabs(v.y)>1e-6 || b3Fabs(v.z)>1e-6) \n"
+" return false;\n"
+" return true;\n"
+"}\n"
+"inline int b3MaxDot( b3Float4ConstArg vec, __global const b3Float4* vecArray, int vecLen, float* dotOut )\n"
+"{\n"
+" float maxDot = -B3_INFINITY;\n"
+" int i = 0;\n"
+" int ptIndex = -1;\n"
+" for( i = 0; i < vecLen; i++ )\n"
+" {\n"
+" float dot = b3Dot3F4(vecArray[i],vec);\n"
+" \n"
+" if( dot > maxDot )\n"
+" {\n"
+" maxDot = dot;\n"
+" ptIndex = i;\n"
+" }\n"
+" }\n"
+" b3Assert(ptIndex>=0);\n"
+" if (ptIndex<0)\n"
+" {\n"
+" ptIndex = 0;\n"
+" }\n"
+" *dotOut = maxDot;\n"
+" return ptIndex;\n"
+"}\n"
+"#endif //B3_FLOAT4_H\n"
+"#ifndef B3_MAT3x3_H\n"
+"#define B3_MAT3x3_H\n"
+"#ifndef B3_QUAT_H\n"
+"#define B3_QUAT_H\n"
+"#ifndef B3_PLATFORM_DEFINITIONS_H\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+"#endif\n"
+"#endif\n"
+"#ifndef B3_FLOAT4_H\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+"#endif \n"
+"#endif //B3_FLOAT4_H\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+" typedef float4 b3Quat;\n"
+" #define b3QuatConstArg const b3Quat\n"
+" \n"
+" \n"
+"inline float4 b3FastNormalize4(float4 v)\n"
+"{\n"
+" v = (float4)(v.xyz,0.f);\n"
+" return fast_normalize(v);\n"
+"}\n"
+" \n"
+"inline b3Quat b3QuatMul(b3Quat a, b3Quat b);\n"
+"inline b3Quat b3QuatNormalized(b3QuatConstArg in);\n"
+"inline b3Quat b3QuatRotate(b3QuatConstArg q, b3QuatConstArg vec);\n"
+"inline b3Quat b3QuatInvert(b3QuatConstArg q);\n"
+"inline b3Quat b3QuatInverse(b3QuatConstArg q);\n"
+"inline b3Quat b3QuatMul(b3QuatConstArg a, b3QuatConstArg b)\n"
+"{\n"
+" b3Quat ans;\n"
+" ans = b3Cross3( a, b );\n"
+" ans += a.w*b+b.w*a;\n"
+"// ans.w = a.w*b.w - (a.x*b.x+a.y*b.y+a.z*b.z);\n"
+" ans.w = a.w*b.w - b3Dot3F4(a, b);\n"
+" return ans;\n"
+"}\n"
+"inline b3Quat b3QuatNormalized(b3QuatConstArg in)\n"
+"{\n"
+" b3Quat q;\n"
+" q=in;\n"
+" //return b3FastNormalize4(in);\n"
+" float len = native_sqrt(dot(q, q));\n"
+" if(len > 0.f)\n"
+" {\n"
+" q *= 1.f / len;\n"
+" }\n"
+" else\n"
+" {\n"
+" q.x = q.y = q.z = 0.f;\n"
+" q.w = 1.f;\n"
+" }\n"
+" return q;\n"
+"}\n"
+"inline float4 b3QuatRotate(b3QuatConstArg q, b3QuatConstArg vec)\n"
+"{\n"
+" b3Quat qInv = b3QuatInvert( q );\n"
+" float4 vcpy = vec;\n"
+" vcpy.w = 0.f;\n"
+" float4 out = b3QuatMul(b3QuatMul(q,vcpy),qInv);\n"
+" return out;\n"
+"}\n"
+"inline b3Quat b3QuatInverse(b3QuatConstArg q)\n"
+"{\n"
+" return (b3Quat)(-q.xyz, q.w);\n"
+"}\n"
+"inline b3Quat b3QuatInvert(b3QuatConstArg q)\n"
+"{\n"
+" return (b3Quat)(-q.xyz, q.w);\n"
+"}\n"
+"inline float4 b3QuatInvRotate(b3QuatConstArg q, b3QuatConstArg vec)\n"
+"{\n"
+" return b3QuatRotate( b3QuatInvert( q ), vec );\n"
+"}\n"
+"inline b3Float4 b3TransformPoint(b3Float4ConstArg point, b3Float4ConstArg translation, b3QuatConstArg orientation)\n"
+"{\n"
+" return b3QuatRotate( orientation, point ) + (translation);\n"
+"}\n"
+" \n"
+"#endif \n"
+"#endif //B3_QUAT_H\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+"typedef struct\n"
+"{\n"
+" b3Float4 m_row[3];\n"
+"}b3Mat3x3;\n"
+"#define b3Mat3x3ConstArg const b3Mat3x3\n"
+"#define b3GetRow(m,row) (m.m_row[row])\n"
+"inline b3Mat3x3 b3QuatGetRotationMatrix(b3Quat quat)\n"
+"{\n"
+" b3Float4 quat2 = (b3Float4)(quat.x*quat.x, quat.y*quat.y, quat.z*quat.z, 0.f);\n"
+" b3Mat3x3 out;\n"
+" out.m_row[0].x=1-2*quat2.y-2*quat2.z;\n"
+" out.m_row[0].y=2*quat.x*quat.y-2*quat.w*quat.z;\n"
+" out.m_row[0].z=2*quat.x*quat.z+2*quat.w*quat.y;\n"
+" out.m_row[0].w = 0.f;\n"
+" out.m_row[1].x=2*quat.x*quat.y+2*quat.w*quat.z;\n"
+" out.m_row[1].y=1-2*quat2.x-2*quat2.z;\n"
+" out.m_row[1].z=2*quat.y*quat.z-2*quat.w*quat.x;\n"
+" out.m_row[1].w = 0.f;\n"
+" out.m_row[2].x=2*quat.x*quat.z-2*quat.w*quat.y;\n"
+" out.m_row[2].y=2*quat.y*quat.z+2*quat.w*quat.x;\n"
+" out.m_row[2].z=1-2*quat2.x-2*quat2.y;\n"
+" out.m_row[2].w = 0.f;\n"
+" return out;\n"
+"}\n"
+"inline b3Mat3x3 b3AbsoluteMat3x3(b3Mat3x3ConstArg matIn)\n"
+"{\n"
+" b3Mat3x3 out;\n"
+" out.m_row[0] = fabs(matIn.m_row[0]);\n"
+" out.m_row[1] = fabs(matIn.m_row[1]);\n"
+" out.m_row[2] = fabs(matIn.m_row[2]);\n"
+" return out;\n"
+"}\n"
+"__inline\n"
+"b3Mat3x3 mtZero();\n"
+"__inline\n"
+"b3Mat3x3 mtIdentity();\n"
+"__inline\n"
+"b3Mat3x3 mtTranspose(b3Mat3x3 m);\n"
+"__inline\n"
+"b3Mat3x3 mtMul(b3Mat3x3 a, b3Mat3x3 b);\n"
+"__inline\n"
+"b3Float4 mtMul1(b3Mat3x3 a, b3Float4 b);\n"
+"__inline\n"
+"b3Float4 mtMul3(b3Float4 a, b3Mat3x3 b);\n"
+"__inline\n"
+"b3Mat3x3 mtZero()\n"
+"{\n"
+" b3Mat3x3 m;\n"
+" m.m_row[0] = (b3Float4)(0.f);\n"
+" m.m_row[1] = (b3Float4)(0.f);\n"
+" m.m_row[2] = (b3Float4)(0.f);\n"
+" return m;\n"
+"}\n"
+"__inline\n"
+"b3Mat3x3 mtIdentity()\n"
+"{\n"
+" b3Mat3x3 m;\n"
+" m.m_row[0] = (b3Float4)(1,0,0,0);\n"
+" m.m_row[1] = (b3Float4)(0,1,0,0);\n"
+" m.m_row[2] = (b3Float4)(0,0,1,0);\n"
+" return m;\n"
+"}\n"
+"__inline\n"
+"b3Mat3x3 mtTranspose(b3Mat3x3 m)\n"
+"{\n"
+" b3Mat3x3 out;\n"
+" out.m_row[0] = (b3Float4)(m.m_row[0].x, m.m_row[1].x, m.m_row[2].x, 0.f);\n"
+" out.m_row[1] = (b3Float4)(m.m_row[0].y, m.m_row[1].y, m.m_row[2].y, 0.f);\n"
+" out.m_row[2] = (b3Float4)(m.m_row[0].z, m.m_row[1].z, m.m_row[2].z, 0.f);\n"
+" return out;\n"
+"}\n"
+"__inline\n"
+"b3Mat3x3 mtMul(b3Mat3x3 a, b3Mat3x3 b)\n"
+"{\n"
+" b3Mat3x3 transB;\n"
+" transB = mtTranspose( b );\n"
+" b3Mat3x3 ans;\n"
+" // why this doesn't run when 0ing in the for{}\n"
+" a.m_row[0].w = 0.f;\n"
+" a.m_row[1].w = 0.f;\n"
+" a.m_row[2].w = 0.f;\n"
+" for(int i=0; i<3; i++)\n"
+" {\n"
+"// a.m_row[i].w = 0.f;\n"
+" ans.m_row[i].x = b3Dot3F4(a.m_row[i],transB.m_row[0]);\n"
+" ans.m_row[i].y = b3Dot3F4(a.m_row[i],transB.m_row[1]);\n"
+" ans.m_row[i].z = b3Dot3F4(a.m_row[i],transB.m_row[2]);\n"
+" ans.m_row[i].w = 0.f;\n"
+" }\n"
+" return ans;\n"
+"}\n"
+"__inline\n"
+"b3Float4 mtMul1(b3Mat3x3 a, b3Float4 b)\n"
+"{\n"
+" b3Float4 ans;\n"
+" ans.x = b3Dot3F4( a.m_row[0], b );\n"
+" ans.y = b3Dot3F4( a.m_row[1], b );\n"
+" ans.z = b3Dot3F4( a.m_row[2], b );\n"
+" ans.w = 0.f;\n"
+" return ans;\n"
+"}\n"
+"__inline\n"
+"b3Float4 mtMul3(b3Float4 a, b3Mat3x3 b)\n"
+"{\n"
+" b3Float4 colx = b3MakeFloat4(b.m_row[0].x, b.m_row[1].x, b.m_row[2].x, 0);\n"
+" b3Float4 coly = b3MakeFloat4(b.m_row[0].y, b.m_row[1].y, b.m_row[2].y, 0);\n"
+" b3Float4 colz = b3MakeFloat4(b.m_row[0].z, b.m_row[1].z, b.m_row[2].z, 0);\n"
+" b3Float4 ans;\n"
+" ans.x = b3Dot3F4( a, colx );\n"
+" ans.y = b3Dot3F4( a, coly );\n"
+" ans.z = b3Dot3F4( a, colz );\n"
+" return ans;\n"
+"}\n"
+"#endif\n"
+"#endif //B3_MAT3x3_H\n"
+"typedef struct b3Aabb b3Aabb_t;\n"
+"struct b3Aabb\n"
+"{\n"
+" union\n"
+" {\n"
+" float m_min[4];\n"
+" b3Float4 m_minVec;\n"
+" int m_minIndices[4];\n"
+" };\n"
+" union\n"
+" {\n"
+" float m_max[4];\n"
+" b3Float4 m_maxVec;\n"
+" int m_signedMaxIndices[4];\n"
+" };\n"
+"};\n"
+"inline void b3TransformAabb2(b3Float4ConstArg localAabbMin,b3Float4ConstArg localAabbMax, float margin,\n"
+" b3Float4ConstArg pos,\n"
+" b3QuatConstArg orn,\n"
+" b3Float4* aabbMinOut,b3Float4* aabbMaxOut)\n"
+"{\n"
+" b3Float4 localHalfExtents = 0.5f*(localAabbMax-localAabbMin);\n"
+" localHalfExtents+=b3MakeFloat4(margin,margin,margin,0.f);\n"
+" b3Float4 localCenter = 0.5f*(localAabbMax+localAabbMin);\n"
+" b3Mat3x3 m;\n"
+" m = b3QuatGetRotationMatrix(orn);\n"
+" b3Mat3x3 abs_b = b3AbsoluteMat3x3(m);\n"
+" b3Float4 center = b3TransformPoint(localCenter,pos,orn);\n"
+" \n"
+" b3Float4 extent = b3MakeFloat4(b3Dot3F4(localHalfExtents,b3GetRow(abs_b,0)),\n"
+" b3Dot3F4(localHalfExtents,b3GetRow(abs_b,1)),\n"
+" b3Dot3F4(localHalfExtents,b3GetRow(abs_b,2)),\n"
+" 0.f);\n"
+" *aabbMinOut = center-extent;\n"
+" *aabbMaxOut = center+extent;\n"
+"}\n"
+"/// conservative test for overlap between two aabbs\n"
+"inline bool b3TestAabbAgainstAabb(b3Float4ConstArg aabbMin1,b3Float4ConstArg aabbMax1,\n"
+" b3Float4ConstArg aabbMin2, b3Float4ConstArg aabbMax2)\n"
+"{\n"
+" bool overlap = true;\n"
+" overlap = (aabbMin1.x > aabbMax2.x || aabbMax1.x < aabbMin2.x) ? false : overlap;\n"
+" overlap = (aabbMin1.z > aabbMax2.z || aabbMax1.z < aabbMin2.z) ? false : overlap;\n"
+" overlap = (aabbMin1.y > aabbMax2.y || aabbMax1.y < aabbMin2.y) ? false : overlap;\n"
+" return overlap;\n"
+"}\n"
+"#endif //B3_AABB_H\n"
+"/*\n"
+"Bullet Continuous Collision Detection and Physics Library\n"
+"Copyright (c) 2003-2013 Erwin Coumans http://bulletphysics.org\n"
+"This software is provided 'as-is', without any express or implied warranty.\n"
+"In no event will the authors be held liable for any damages arising from the use of this software.\n"
+"Permission is granted to anyone to use this software for any purpose,\n"
+"including commercial applications, and to alter it and redistribute it freely,\n"
+"subject to the following restrictions:\n"
+"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.\n"
+"2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.\n"
+"3. This notice may not be removed or altered from any source distribution.\n"
+"*/\n"
+"#ifndef B3_INT2_H\n"
+"#define B3_INT2_H\n"
+"#ifdef __cplusplus\n"
+"#else\n"
+"#define b3UnsignedInt2 uint2\n"
+"#define b3Int2 int2\n"
+"#define b3MakeInt2 (int2)\n"
+"#endif //__cplusplus\n"
+"#endif\n"
+"typedef struct\n"
+"{\n"
+" float4 m_plane;\n"
+" int m_indexOffset;\n"
+" int m_numIndices;\n"
+"} btGpuFace;\n"
+"#define make_float4 (float4)\n"
+"__inline\n"
+"float4 cross3(float4 a, float4 b)\n"
+"{\n"
+" return cross(a,b);\n"
+" \n"
+"// float4 a1 = make_float4(a.xyz,0.f);\n"
+"// float4 b1 = make_float4(b.xyz,0.f);\n"
+"// return cross(a1,b1);\n"
+"//float4 c = make_float4(a.y*b.z - a.z*b.y,a.z*b.x - a.x*b.z,a.x*b.y - a.y*b.x,0.f);\n"
+" \n"
+" // float4 c = make_float4(a.y*b.z - a.z*b.y,1.f,a.x*b.y - a.y*b.x,0.f);\n"
+" \n"
+" //return c;\n"
+"}\n"
+"__inline\n"
+"float dot3F4(float4 a, float4 b)\n"
+"{\n"
+" float4 a1 = make_float4(a.xyz,0.f);\n"
+" float4 b1 = make_float4(b.xyz,0.f);\n"
+" return dot(a1, b1);\n"
+"}\n"
+"__inline\n"
+"float4 fastNormalize4(float4 v)\n"
+"{\n"
+" v = make_float4(v.xyz,0.f);\n"
+" return fast_normalize(v);\n"
+"}\n"
+"///////////////////////////////////////\n"
+"// Quaternion\n"
+"///////////////////////////////////////\n"
+"typedef float4 Quaternion;\n"
+"__inline\n"
+"Quaternion qtMul(Quaternion a, Quaternion b);\n"
+"__inline\n"
+"Quaternion qtNormalize(Quaternion in);\n"
+"__inline\n"
+"float4 qtRotate(Quaternion q, float4 vec);\n"
+"__inline\n"
+"Quaternion qtInvert(Quaternion q);\n"
+"__inline\n"
+"Quaternion qtMul(Quaternion a, Quaternion b)\n"
+"{\n"
+" Quaternion ans;\n"
+" ans = cross3( a, b );\n"
+" ans += a.w*b+b.w*a;\n"
+"// ans.w = a.w*b.w - (a.x*b.x+a.y*b.y+a.z*b.z);\n"
+" ans.w = a.w*b.w - dot3F4(a, b);\n"
+" return ans;\n"
+"}\n"
+"__inline\n"
+"Quaternion qtNormalize(Quaternion in)\n"
+"{\n"
+" return fastNormalize4(in);\n"
+"// in /= length( in );\n"
+"// return in;\n"
+"}\n"
+"__inline\n"
+"float4 qtRotate(Quaternion q, float4 vec)\n"
+"{\n"
+" Quaternion qInv = qtInvert( q );\n"
+" float4 vcpy = vec;\n"
+" vcpy.w = 0.f;\n"
+" float4 out = qtMul(qtMul(q,vcpy),qInv);\n"
+" return out;\n"
+"}\n"
+"__inline\n"
+"Quaternion qtInvert(Quaternion q)\n"
+"{\n"
+" return (Quaternion)(-q.xyz, q.w);\n"
+"}\n"
+"__inline\n"
+"float4 qtInvRotate(const Quaternion q, float4 vec)\n"
+"{\n"
+" return qtRotate( qtInvert( q ), vec );\n"
+"}\n"
+"__inline\n"
+"float4 transform(const float4* p, const float4* translation, const Quaternion* orientation)\n"
+"{\n"
+" return qtRotate( *orientation, *p ) + (*translation);\n"
+"}\n"
+"__inline\n"
+"float4 normalize3(const float4 a)\n"
+"{\n"
+" float4 n = make_float4(a.x, a.y, a.z, 0.f);\n"
+" return fastNormalize4( n );\n"
+"}\n"
+"inline void projectLocal(const ConvexPolyhedronCL* hull, const float4 pos, const float4 orn, \n"
+"const float4* dir, const float4* vertices, float* min, float* max)\n"
+"{\n"
+" min[0] = FLT_MAX;\n"
+" max[0] = -FLT_MAX;\n"
+" int numVerts = hull->m_numVertices;\n"
+" const float4 localDir = qtInvRotate(orn,*dir);\n"
+" float offset = dot(pos,*dir);\n"
+" for(int i=0;i<numVerts;i++)\n"
+" {\n"
+" float dp = dot(vertices[hull->m_vertexOffset+i],localDir);\n"
+" if(dp < min[0]) \n"
+" min[0] = dp;\n"
+" if(dp > max[0]) \n"
+" max[0] = dp;\n"
+" }\n"
+" if(min[0]>max[0])\n"
+" {\n"
+" float tmp = min[0];\n"
+" min[0] = max[0];\n"
+" max[0] = tmp;\n"
+" }\n"
+" min[0] += offset;\n"
+" max[0] += offset;\n"
+"}\n"
+"inline void project(__global const ConvexPolyhedronCL* hull, const float4 pos, const float4 orn, \n"
+"const float4* dir, __global const float4* vertices, float* min, float* max)\n"
+"{\n"
+" min[0] = FLT_MAX;\n"
+" max[0] = -FLT_MAX;\n"
+" int numVerts = hull->m_numVertices;\n"
+" const float4 localDir = qtInvRotate(orn,*dir);\n"
+" float offset = dot(pos,*dir);\n"
+" for(int i=0;i<numVerts;i++)\n"
+" {\n"
+" float dp = dot(vertices[hull->m_vertexOffset+i],localDir);\n"
+" if(dp < min[0]) \n"
+" min[0] = dp;\n"
+" if(dp > max[0]) \n"
+" max[0] = dp;\n"
+" }\n"
+" if(min[0]>max[0])\n"
+" {\n"
+" float tmp = min[0];\n"
+" min[0] = max[0];\n"
+" max[0] = tmp;\n"
+" }\n"
+" min[0] += offset;\n"
+" max[0] += offset;\n"
+"}\n"
+"inline bool TestSepAxisLocalA(const ConvexPolyhedronCL* hullA, __global const ConvexPolyhedronCL* hullB, \n"
+" const float4 posA,const float4 ornA,\n"
+" const float4 posB,const float4 ornB,\n"
+" float4* sep_axis, const float4* verticesA, __global const float4* verticesB,float* depth)\n"
+"{\n"
+" float Min0,Max0;\n"
+" float Min1,Max1;\n"
+" projectLocal(hullA,posA,ornA,sep_axis,verticesA, &Min0, &Max0);\n"
+" project(hullB,posB,ornB, sep_axis,verticesB, &Min1, &Max1);\n"
+" if(Max0<Min1 || Max1<Min0)\n"
+" return false;\n"
+" float d0 = Max0 - Min1;\n"
+" float d1 = Max1 - Min0;\n"
+" *depth = d0<d1 ? d0:d1;\n"
+" return true;\n"
+"}\n"
+"inline bool IsAlmostZero(const float4 v)\n"
+"{\n"
+" if(fabs(v.x)>1e-6f || fabs(v.y)>1e-6f || fabs(v.z)>1e-6f)\n"
+" return false;\n"
+" return true;\n"
+"}\n"
+"bool findSeparatingAxisLocalA( const ConvexPolyhedronCL* hullA, __global const ConvexPolyhedronCL* hullB, \n"
+" const float4 posA1,\n"
+" const float4 ornA,\n"
+" const float4 posB1,\n"
+" const float4 ornB,\n"
+" const float4 DeltaC2,\n"
+" \n"
+" const float4* verticesA, \n"
+" const float4* uniqueEdgesA, \n"
+" const btGpuFace* facesA,\n"
+" const int* indicesA,\n"
+" __global const float4* verticesB, \n"
+" __global const float4* uniqueEdgesB, \n"
+" __global const btGpuFace* facesB,\n"
+" __global const int* indicesB,\n"
+" float4* sep,\n"
+" float* dmin)\n"
+"{\n"
+" \n"
+" float4 posA = posA1;\n"
+" posA.w = 0.f;\n"
+" float4 posB = posB1;\n"
+" posB.w = 0.f;\n"
+" int curPlaneTests=0;\n"
+" {\n"
+" int numFacesA = hullA->m_numFaces;\n"
+" // Test normals from hullA\n"
+" for(int i=0;i<numFacesA;i++)\n"
+" {\n"
+" const float4 normal = facesA[hullA->m_faceOffset+i].m_plane;\n"
+" float4 faceANormalWS = qtRotate(ornA,normal);\n"
+" if (dot3F4(DeltaC2,faceANormalWS)<0)\n"
+" faceANormalWS*=-1.f;\n"
+" curPlaneTests++;\n"
+" float d;\n"
+" if(!TestSepAxisLocalA( hullA, hullB, posA,ornA,posB,ornB,&faceANormalWS, verticesA, verticesB,&d))\n"
+" return false;\n"
+" if(d<*dmin)\n"
+" {\n"
+" *dmin = d;\n"
+" *sep = faceANormalWS;\n"
+" }\n"
+" }\n"
+" }\n"
+" if((dot3F4(-DeltaC2,*sep))>0.0f)\n"
+" {\n"
+" *sep = -(*sep);\n"
+" }\n"
+" return true;\n"
+"}\n"
+"bool findSeparatingAxisLocalB( __global const ConvexPolyhedronCL* hullA, const ConvexPolyhedronCL* hullB, \n"
+" const float4 posA1,\n"
+" const float4 ornA,\n"
+" const float4 posB1,\n"
+" const float4 ornB,\n"
+" const float4 DeltaC2,\n"
+" __global const float4* verticesA, \n"
+" __global const float4* uniqueEdgesA, \n"
+" __global const btGpuFace* facesA,\n"
+" __global const int* indicesA,\n"
+" const float4* verticesB,\n"
+" const float4* uniqueEdgesB, \n"
+" const btGpuFace* facesB,\n"
+" const int* indicesB,\n"
+" float4* sep,\n"
+" float* dmin)\n"
+"{\n"
+" float4 posA = posA1;\n"
+" posA.w = 0.f;\n"
+" float4 posB = posB1;\n"
+" posB.w = 0.f;\n"
+" int curPlaneTests=0;\n"
+" {\n"
+" int numFacesA = hullA->m_numFaces;\n"
+" // Test normals from hullA\n"
+" for(int i=0;i<numFacesA;i++)\n"
+" {\n"
+" const float4 normal = facesA[hullA->m_faceOffset+i].m_plane;\n"
+" float4 faceANormalWS = qtRotate(ornA,normal);\n"
+" if (dot3F4(DeltaC2,faceANormalWS)<0)\n"
+" faceANormalWS *= -1.f;\n"
+" curPlaneTests++;\n"
+" float d;\n"
+" if(!TestSepAxisLocalA( hullB, hullA, posB,ornB,posA,ornA, &faceANormalWS, verticesB,verticesA, &d))\n"
+" return false;\n"
+" if(d<*dmin)\n"
+" {\n"
+" *dmin = d;\n"
+" *sep = faceANormalWS;\n"
+" }\n"
+" }\n"
+" }\n"
+" if((dot3F4(-DeltaC2,*sep))>0.0f)\n"
+" {\n"
+" *sep = -(*sep);\n"
+" }\n"
+" return true;\n"
+"}\n"
+"bool findSeparatingAxisEdgeEdgeLocalA( const ConvexPolyhedronCL* hullA, __global const ConvexPolyhedronCL* hullB, \n"
+" const float4 posA1,\n"
+" const float4 ornA,\n"
+" const float4 posB1,\n"
+" const float4 ornB,\n"
+" const float4 DeltaC2,\n"
+" const float4* verticesA, \n"
+" const float4* uniqueEdgesA, \n"
+" const btGpuFace* facesA,\n"
+" const int* indicesA,\n"
+" __global const float4* verticesB, \n"
+" __global const float4* uniqueEdgesB, \n"
+" __global const btGpuFace* facesB,\n"
+" __global const int* indicesB,\n"
+" float4* sep,\n"
+" float* dmin)\n"
+"{\n"
+" float4 posA = posA1;\n"
+" posA.w = 0.f;\n"
+" float4 posB = posB1;\n"
+" posB.w = 0.f;\n"
+" int curPlaneTests=0;\n"
+" int curEdgeEdge = 0;\n"
+" // Test edges\n"
+" for(int e0=0;e0<hullA->m_numUniqueEdges;e0++)\n"
+" {\n"
+" const float4 edge0 = uniqueEdgesA[hullA->m_uniqueEdgesOffset+e0];\n"
+" float4 edge0World = qtRotate(ornA,edge0);\n"
+" for(int e1=0;e1<hullB->m_numUniqueEdges;e1++)\n"
+" {\n"
+" const float4 edge1 = uniqueEdgesB[hullB->m_uniqueEdgesOffset+e1];\n"
+" float4 edge1World = qtRotate(ornB,edge1);\n"
+" float4 crossje = cross3(edge0World,edge1World);\n"
+" curEdgeEdge++;\n"
+" if(!IsAlmostZero(crossje))\n"
+" {\n"
+" crossje = normalize3(crossje);\n"
+" if (dot3F4(DeltaC2,crossje)<0)\n"
+" crossje *= -1.f;\n"
+" float dist;\n"
+" bool result = true;\n"
+" {\n"
+" float Min0,Max0;\n"
+" float Min1,Max1;\n"
+" projectLocal(hullA,posA,ornA,&crossje,verticesA, &Min0, &Max0);\n"
+" project(hullB,posB,ornB,&crossje,verticesB, &Min1, &Max1);\n"
+" \n"
+" if(Max0<Min1 || Max1<Min0)\n"
+" result = false;\n"
+" \n"
+" float d0 = Max0 - Min1;\n"
+" float d1 = Max1 - Min0;\n"
+" dist = d0<d1 ? d0:d1;\n"
+" result = true;\n"
+" }\n"
+" \n"
+" if(dist<*dmin)\n"
+" {\n"
+" *dmin = dist;\n"
+" *sep = crossje;\n"
+" }\n"
+" }\n"
+" }\n"
+" }\n"
+" \n"
+" if((dot3F4(-DeltaC2,*sep))>0.0f)\n"
+" {\n"
+" *sep = -(*sep);\n"
+" }\n"
+" return true;\n"
+"}\n"
+"inline bool TestSepAxis(__global const ConvexPolyhedronCL* hullA, __global const ConvexPolyhedronCL* hullB, \n"
+" const float4 posA,const float4 ornA,\n"
+" const float4 posB,const float4 ornB,\n"
+" float4* sep_axis, __global const float4* vertices,float* depth)\n"
+"{\n"
+" float Min0,Max0;\n"
+" float Min1,Max1;\n"
+" project(hullA,posA,ornA,sep_axis,vertices, &Min0, &Max0);\n"
+" project(hullB,posB,ornB, sep_axis,vertices, &Min1, &Max1);\n"
+" if(Max0<Min1 || Max1<Min0)\n"
+" return false;\n"
+" float d0 = Max0 - Min1;\n"
+" float d1 = Max1 - Min0;\n"
+" *depth = d0<d1 ? d0:d1;\n"
+" return true;\n"
+"}\n"
+"bool findSeparatingAxis( __global const ConvexPolyhedronCL* hullA, __global const ConvexPolyhedronCL* hullB, \n"
+" const float4 posA1,\n"
+" const float4 ornA,\n"
+" const float4 posB1,\n"
+" const float4 ornB,\n"
+" const float4 DeltaC2,\n"
+" __global const float4* vertices, \n"
+" __global const float4* uniqueEdges, \n"
+" __global const btGpuFace* faces,\n"
+" __global const int* indices,\n"
+" float4* sep,\n"
+" float* dmin)\n"
+"{\n"
+" \n"
+" float4 posA = posA1;\n"
+" posA.w = 0.f;\n"
+" float4 posB = posB1;\n"
+" posB.w = 0.f;\n"
+" \n"
+" int curPlaneTests=0;\n"
+" {\n"
+" int numFacesA = hullA->m_numFaces;\n"
+" // Test normals from hullA\n"
+" for(int i=0;i<numFacesA;i++)\n"
+" {\n"
+" const float4 normal = faces[hullA->m_faceOffset+i].m_plane;\n"
+" float4 faceANormalWS = qtRotate(ornA,normal);\n"
+" \n"
+" if (dot3F4(DeltaC2,faceANormalWS)<0)\n"
+" faceANormalWS*=-1.f;\n"
+" \n"
+" curPlaneTests++;\n"
+" \n"
+" float d;\n"
+" if(!TestSepAxis( hullA, hullB, posA,ornA,posB,ornB,&faceANormalWS, vertices,&d))\n"
+" return false;\n"
+" \n"
+" if(d<*dmin)\n"
+" {\n"
+" *dmin = d;\n"
+" *sep = faceANormalWS;\n"
+" }\n"
+" }\n"
+" }\n"
+" if((dot3F4(-DeltaC2,*sep))>0.0f)\n"
+" {\n"
+" *sep = -(*sep);\n"
+" }\n"
+" \n"
+" return true;\n"
+"}\n"
+"bool findSeparatingAxisUnitSphere( __global const ConvexPolyhedronCL* hullA, __global const ConvexPolyhedronCL* hullB, \n"
+" const float4 posA1,\n"
+" const float4 ornA,\n"
+" const float4 posB1,\n"
+" const float4 ornB,\n"
+" const float4 DeltaC2,\n"
+" __global const float4* vertices,\n"
+" __global const float4* unitSphereDirections,\n"
+" int numUnitSphereDirections,\n"
+" float4* sep,\n"
+" float* dmin)\n"
+"{\n"
+" \n"
+" float4 posA = posA1;\n"
+" posA.w = 0.f;\n"
+" float4 posB = posB1;\n"
+" posB.w = 0.f;\n"
+" int curPlaneTests=0;\n"
+" int curEdgeEdge = 0;\n"
+" // Test unit sphere directions\n"
+" for (int i=0;i<numUnitSphereDirections;i++)\n"
+" {\n"
+" float4 crossje;\n"
+" crossje = unitSphereDirections[i]; \n"
+" if (dot3F4(DeltaC2,crossje)>0)\n"
+" crossje *= -1.f;\n"
+" {\n"
+" float dist;\n"
+" bool result = true;\n"
+" float Min0,Max0;\n"
+" float Min1,Max1;\n"
+" project(hullA,posA,ornA,&crossje,vertices, &Min0, &Max0);\n"
+" project(hullB,posB,ornB,&crossje,vertices, &Min1, &Max1);\n"
+" \n"
+" if(Max0<Min1 || Max1<Min0)\n"
+" return false;\n"
+" \n"
+" float d0 = Max0 - Min1;\n"
+" float d1 = Max1 - Min0;\n"
+" dist = d0<d1 ? d0:d1;\n"
+" result = true;\n"
+" \n"
+" if(dist<*dmin)\n"
+" {\n"
+" *dmin = dist;\n"
+" *sep = crossje;\n"
+" }\n"
+" }\n"
+" }\n"
+" \n"
+" if((dot3F4(-DeltaC2,*sep))>0.0f)\n"
+" {\n"
+" *sep = -(*sep);\n"
+" }\n"
+" return true;\n"
+"}\n"
+"bool findSeparatingAxisEdgeEdge( __global const ConvexPolyhedronCL* hullA, __global const ConvexPolyhedronCL* hullB, \n"
+" const float4 posA1,\n"
+" const float4 ornA,\n"
+" const float4 posB1,\n"
+" const float4 ornB,\n"
+" const float4 DeltaC2,\n"
+" __global const float4* vertices, \n"
+" __global const float4* uniqueEdges, \n"
+" __global const btGpuFace* faces,\n"
+" __global const int* indices,\n"
+" float4* sep,\n"
+" float* dmin)\n"
+"{\n"
+" \n"
+" float4 posA = posA1;\n"
+" posA.w = 0.f;\n"
+" float4 posB = posB1;\n"
+" posB.w = 0.f;\n"
+" int curPlaneTests=0;\n"
+" int curEdgeEdge = 0;\n"
+" // Test edges\n"
+" for(int e0=0;e0<hullA->m_numUniqueEdges;e0++)\n"
+" {\n"
+" const float4 edge0 = uniqueEdges[hullA->m_uniqueEdgesOffset+e0];\n"
+" float4 edge0World = qtRotate(ornA,edge0);\n"
+" for(int e1=0;e1<hullB->m_numUniqueEdges;e1++)\n"
+" {\n"
+" const float4 edge1 = uniqueEdges[hullB->m_uniqueEdgesOffset+e1];\n"
+" float4 edge1World = qtRotate(ornB,edge1);\n"
+" float4 crossje = cross3(edge0World,edge1World);\n"
+" curEdgeEdge++;\n"
+" if(!IsAlmostZero(crossje))\n"
+" {\n"
+" crossje = normalize3(crossje);\n"
+" if (dot3F4(DeltaC2,crossje)<0)\n"
+" crossje*=-1.f;\n"
+" \n"
+" float dist;\n"
+" bool result = true;\n"
+" {\n"
+" float Min0,Max0;\n"
+" float Min1,Max1;\n"
+" project(hullA,posA,ornA,&crossje,vertices, &Min0, &Max0);\n"
+" project(hullB,posB,ornB,&crossje,vertices, &Min1, &Max1);\n"
+" \n"
+" if(Max0<Min1 || Max1<Min0)\n"
+" return false;\n"
+" \n"
+" float d0 = Max0 - Min1;\n"
+" float d1 = Max1 - Min0;\n"
+" dist = d0<d1 ? d0:d1;\n"
+" result = true;\n"
+" }\n"
+" \n"
+" if(dist<*dmin)\n"
+" {\n"
+" *dmin = dist;\n"
+" *sep = crossje;\n"
+" }\n"
+" }\n"
+" }\n"
+" }\n"
+" \n"
+" if((dot3F4(-DeltaC2,*sep))>0.0f)\n"
+" {\n"
+" *sep = -(*sep);\n"
+" }\n"
+" return true;\n"
+"}\n"
+"// work-in-progress\n"
+"__kernel void processCompoundPairsKernel( __global const int4* gpuCompoundPairs,\n"
+" __global const BodyData* rigidBodies, \n"
+" __global const btCollidableGpu* collidables,\n"
+" __global const ConvexPolyhedronCL* convexShapes, \n"
+" __global const float4* vertices,\n"
+" __global const float4* uniqueEdges,\n"
+" __global const btGpuFace* faces,\n"
+" __global const int* indices,\n"
+" __global btAabbCL* aabbs,\n"
+" __global const btGpuChildShape* gpuChildShapes,\n"
+" __global volatile float4* gpuCompoundSepNormalsOut,\n"
+" __global volatile int* gpuHasCompoundSepNormalsOut,\n"
+" int numCompoundPairs\n"
+" )\n"
+"{\n"
+" int i = get_global_id(0);\n"
+" if (i<numCompoundPairs)\n"
+" {\n"
+" int bodyIndexA = gpuCompoundPairs[i].x;\n"
+" int bodyIndexB = gpuCompoundPairs[i].y;\n"
+" int childShapeIndexA = gpuCompoundPairs[i].z;\n"
+" int childShapeIndexB = gpuCompoundPairs[i].w;\n"
+" \n"
+" int collidableIndexA = -1;\n"
+" int collidableIndexB = -1;\n"
+" \n"
+" float4 ornA = rigidBodies[bodyIndexA].m_quat;\n"
+" float4 posA = rigidBodies[bodyIndexA].m_pos;\n"
+" \n"
+" float4 ornB = rigidBodies[bodyIndexB].m_quat;\n"
+" float4 posB = rigidBodies[bodyIndexB].m_pos;\n"
+" \n"
+" if (childShapeIndexA >= 0)\n"
+" {\n"
+" collidableIndexA = gpuChildShapes[childShapeIndexA].m_shapeIndex;\n"
+" float4 childPosA = gpuChildShapes[childShapeIndexA].m_childPosition;\n"
+" float4 childOrnA = gpuChildShapes[childShapeIndexA].m_childOrientation;\n"
+" float4 newPosA = qtRotate(ornA,childPosA)+posA;\n"
+" float4 newOrnA = qtMul(ornA,childOrnA);\n"
+" posA = newPosA;\n"
+" ornA = newOrnA;\n"
+" } else\n"
+" {\n"
+" collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;\n"
+" }\n"
+" \n"
+" if (childShapeIndexB>=0)\n"
+" {\n"
+" collidableIndexB = gpuChildShapes[childShapeIndexB].m_shapeIndex;\n"
+" float4 childPosB = gpuChildShapes[childShapeIndexB].m_childPosition;\n"
+" float4 childOrnB = gpuChildShapes[childShapeIndexB].m_childOrientation;\n"
+" float4 newPosB = transform(&childPosB,&posB,&ornB);\n"
+" float4 newOrnB = qtMul(ornB,childOrnB);\n"
+" posB = newPosB;\n"
+" ornB = newOrnB;\n"
+" } else\n"
+" {\n"
+" collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx; \n"
+" }\n"
+" \n"
+" gpuHasCompoundSepNormalsOut[i] = 0;\n"
+" \n"
+" int shapeIndexA = collidables[collidableIndexA].m_shapeIndex;\n"
+" int shapeIndexB = collidables[collidableIndexB].m_shapeIndex;\n"
+" \n"
+" int shapeTypeA = collidables[collidableIndexA].m_shapeType;\n"
+" int shapeTypeB = collidables[collidableIndexB].m_shapeType;\n"
+" \n"
+" if ((shapeTypeA != SHAPE_CONVEX_HULL) || (shapeTypeB != SHAPE_CONVEX_HULL))\n"
+" {\n"
+" return;\n"
+" }\n"
+" int hasSeparatingAxis = 5;\n"
+" \n"
+" int numFacesA = convexShapes[shapeIndexA].m_numFaces;\n"
+" float dmin = FLT_MAX;\n"
+" posA.w = 0.f;\n"
+" posB.w = 0.f;\n"
+" float4 c0local = convexShapes[shapeIndexA].m_localCenter;\n"
+" float4 c0 = transform(&c0local, &posA, &ornA);\n"
+" float4 c1local = convexShapes[shapeIndexB].m_localCenter;\n"
+" float4 c1 = transform(&c1local,&posB,&ornB);\n"
+" const float4 DeltaC2 = c0 - c1;\n"
+" float4 sepNormal = make_float4(1,0,0,0);\n"
+" bool sepA = findSeparatingAxis( &convexShapes[shapeIndexA], &convexShapes[shapeIndexB],posA,ornA,posB,ornB,DeltaC2,vertices,uniqueEdges,faces,indices,&sepNormal,&dmin);\n"
+" hasSeparatingAxis = 4;\n"
+" if (!sepA)\n"
+" {\n"
+" hasSeparatingAxis = 0;\n"
+" } else\n"
+" {\n"
+" bool sepB = findSeparatingAxis( &convexShapes[shapeIndexB],&convexShapes[shapeIndexA],posB,ornB,posA,ornA,DeltaC2,vertices,uniqueEdges,faces,indices,&sepNormal,&dmin);\n"
+" if (!sepB)\n"
+" {\n"
+" hasSeparatingAxis = 0;\n"
+" } else//(!sepB)\n"
+" {\n"
+" bool sepEE = findSeparatingAxisEdgeEdge( &convexShapes[shapeIndexA], &convexShapes[shapeIndexB],posA,ornA,posB,ornB,DeltaC2,vertices,uniqueEdges,faces,indices,&sepNormal,&dmin);\n"
+" if (sepEE)\n"
+" {\n"
+" gpuCompoundSepNormalsOut[i] = sepNormal;//fastNormalize4(sepNormal);\n"
+" gpuHasCompoundSepNormalsOut[i] = 1;\n"
+" }//sepEE\n"
+" }//(!sepB)\n"
+" }//(!sepA)\n"
+" \n"
+" \n"
+" }\n"
+" \n"
+"}\n"
+"inline b3Float4 MyUnQuantize(const unsigned short* vecIn, b3Float4 quantization, b3Float4 bvhAabbMin)\n"
+"{\n"
+" b3Float4 vecOut;\n"
+" vecOut = b3MakeFloat4(\n"
+" (float)(vecIn[0]) / (quantization.x),\n"
+" (float)(vecIn[1]) / (quantization.y),\n"
+" (float)(vecIn[2]) / (quantization.z),\n"
+" 0.f);\n"
+" vecOut += bvhAabbMin;\n"
+" return vecOut;\n"
+"}\n"
+"inline b3Float4 MyUnQuantizeGlobal(__global const unsigned short* vecIn, b3Float4 quantization, b3Float4 bvhAabbMin)\n"
+"{\n"
+" b3Float4 vecOut;\n"
+" vecOut = b3MakeFloat4(\n"
+" (float)(vecIn[0]) / (quantization.x),\n"
+" (float)(vecIn[1]) / (quantization.y),\n"
+" (float)(vecIn[2]) / (quantization.z),\n"
+" 0.f);\n"
+" vecOut += bvhAabbMin;\n"
+" return vecOut;\n"
+"}\n"
+"// work-in-progress\n"
+"__kernel void findCompoundPairsKernel( __global const int4* pairs, \n"
+" __global const BodyData* rigidBodies, \n"
+" __global const btCollidableGpu* collidables,\n"
+" __global const ConvexPolyhedronCL* convexShapes, \n"
+" __global const float4* vertices,\n"
+" __global const float4* uniqueEdges,\n"
+" __global const btGpuFace* faces,\n"
+" __global const int* indices,\n"
+" __global b3Aabb_t* aabbLocalSpace,\n"
+" __global const btGpuChildShape* gpuChildShapes,\n"
+" __global volatile int4* gpuCompoundPairsOut,\n"
+" __global volatile int* numCompoundPairsOut,\n"
+" __global const b3BvhSubtreeInfo* subtrees,\n"
+" __global const b3QuantizedBvhNode* quantizedNodes,\n"
+" __global const b3BvhInfo* bvhInfos,\n"
+" int numPairs,\n"
+" int maxNumCompoundPairsCapacity\n"
+" )\n"
+"{\n"
+" int i = get_global_id(0);\n"
+" if (i<numPairs)\n"
+" {\n"
+" int bodyIndexA = pairs[i].x;\n"
+" int bodyIndexB = pairs[i].y;\n"
+" int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;\n"
+" int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;\n"
+" int shapeIndexA = collidables[collidableIndexA].m_shapeIndex;\n"
+" int shapeIndexB = collidables[collidableIndexB].m_shapeIndex;\n"
+" //once the broadphase avoids static-static pairs, we can remove this test\n"
+" if ((rigidBodies[bodyIndexA].m_invMass==0) &&(rigidBodies[bodyIndexB].m_invMass==0))\n"
+" {\n"
+" return;\n"
+" }\n"
+" if ((collidables[collidableIndexA].m_shapeType==SHAPE_COMPOUND_OF_CONVEX_HULLS) &&(collidables[collidableIndexB].m_shapeType==SHAPE_COMPOUND_OF_CONVEX_HULLS))\n"
+" {\n"
+" int bvhA = collidables[collidableIndexA].m_compoundBvhIndex;\n"
+" int bvhB = collidables[collidableIndexB].m_compoundBvhIndex;\n"
+" int numSubTreesA = bvhInfos[bvhA].m_numSubTrees;\n"
+" int subTreesOffsetA = bvhInfos[bvhA].m_subTreeOffset;\n"
+" int subTreesOffsetB = bvhInfos[bvhB].m_subTreeOffset;\n"
+" int numSubTreesB = bvhInfos[bvhB].m_numSubTrees;\n"
+" \n"
+" float4 posA = rigidBodies[bodyIndexA].m_pos;\n"
+" b3Quat ornA = rigidBodies[bodyIndexA].m_quat;\n"
+" b3Quat ornB = rigidBodies[bodyIndexB].m_quat;\n"
+" float4 posB = rigidBodies[bodyIndexB].m_pos;\n"
+" \n"
+" for (int p=0;p<numSubTreesA;p++)\n"
+" {\n"
+" b3BvhSubtreeInfo subtreeA = subtrees[subTreesOffsetA+p];\n"
+" //bvhInfos[bvhA].m_quantization\n"
+" b3Float4 treeAminLocal = MyUnQuantize(subtreeA.m_quantizedAabbMin,bvhInfos[bvhA].m_quantization,bvhInfos[bvhA].m_aabbMin);\n"
+" b3Float4 treeAmaxLocal = MyUnQuantize(subtreeA.m_quantizedAabbMax,bvhInfos[bvhA].m_quantization,bvhInfos[bvhA].m_aabbMin);\n"
+" b3Float4 aabbAMinOut,aabbAMaxOut;\n"
+" float margin=0.f;\n"
+" b3TransformAabb2(treeAminLocal,treeAmaxLocal, margin,posA,ornA,&aabbAMinOut,&aabbAMaxOut);\n"
+" \n"
+" for (int q=0;q<numSubTreesB;q++)\n"
+" {\n"
+" b3BvhSubtreeInfo subtreeB = subtrees[subTreesOffsetB+q];\n"
+" b3Float4 treeBminLocal = MyUnQuantize(subtreeB.m_quantizedAabbMin,bvhInfos[bvhB].m_quantization,bvhInfos[bvhB].m_aabbMin);\n"
+" b3Float4 treeBmaxLocal = MyUnQuantize(subtreeB.m_quantizedAabbMax,bvhInfos[bvhB].m_quantization,bvhInfos[bvhB].m_aabbMin);\n"
+" b3Float4 aabbBMinOut,aabbBMaxOut;\n"
+" float margin=0.f;\n"
+" b3TransformAabb2(treeBminLocal,treeBmaxLocal, margin,posB,ornB,&aabbBMinOut,&aabbBMaxOut);\n"
+" \n"
+" \n"
+" bool aabbOverlap = b3TestAabbAgainstAabb(aabbAMinOut,aabbAMaxOut,aabbBMinOut,aabbBMaxOut);\n"
+" if (aabbOverlap)\n"
+" {\n"
+" \n"
+" int startNodeIndexA = subtreeA.m_rootNodeIndex+bvhInfos[bvhA].m_nodeOffset;\n"
+" int endNodeIndexA = startNodeIndexA+subtreeA.m_subtreeSize;\n"
+" int startNodeIndexB = subtreeB.m_rootNodeIndex+bvhInfos[bvhB].m_nodeOffset;\n"
+" int endNodeIndexB = startNodeIndexB+subtreeB.m_subtreeSize;\n"
+" b3Int2 nodeStack[B3_MAX_STACK_DEPTH];\n"
+" b3Int2 node0;\n"
+" node0.x = startNodeIndexA;\n"
+" node0.y = startNodeIndexB;\n"
+" int maxStackDepth = B3_MAX_STACK_DEPTH;\n"
+" int depth=0;\n"
+" nodeStack[depth++]=node0;\n"
+" do\n"
+" {\n"
+" b3Int2 node = nodeStack[--depth];\n"
+" b3Float4 aMinLocal = MyUnQuantizeGlobal(quantizedNodes[node.x].m_quantizedAabbMin,bvhInfos[bvhA].m_quantization,bvhInfos[bvhA].m_aabbMin);\n"
+" b3Float4 aMaxLocal = MyUnQuantizeGlobal(quantizedNodes[node.x].m_quantizedAabbMax,bvhInfos[bvhA].m_quantization,bvhInfos[bvhA].m_aabbMin);\n"
+" b3Float4 bMinLocal = MyUnQuantizeGlobal(quantizedNodes[node.y].m_quantizedAabbMin,bvhInfos[bvhB].m_quantization,bvhInfos[bvhB].m_aabbMin);\n"
+" b3Float4 bMaxLocal = MyUnQuantizeGlobal(quantizedNodes[node.y].m_quantizedAabbMax,bvhInfos[bvhB].m_quantization,bvhInfos[bvhB].m_aabbMin);\n"
+" float margin=0.f;\n"
+" b3Float4 aabbAMinOut,aabbAMaxOut;\n"
+" b3TransformAabb2(aMinLocal,aMaxLocal, margin,posA,ornA,&aabbAMinOut,&aabbAMaxOut);\n"
+" b3Float4 aabbBMinOut,aabbBMaxOut;\n"
+" b3TransformAabb2(bMinLocal,bMaxLocal, margin,posB,ornB,&aabbBMinOut,&aabbBMaxOut);\n"
+" \n"
+" bool nodeOverlap = b3TestAabbAgainstAabb(aabbAMinOut,aabbAMaxOut,aabbBMinOut,aabbBMaxOut);\n"
+" if (nodeOverlap)\n"
+" {\n"
+" bool isLeafA = isLeafNodeGlobal(&quantizedNodes[node.x]);\n"
+" bool isLeafB = isLeafNodeGlobal(&quantizedNodes[node.y]);\n"
+" bool isInternalA = !isLeafA;\n"
+" bool isInternalB = !isLeafB;\n"
+" //fail, even though it might hit two leaf nodes\n"
+" if (depth+4>maxStackDepth && !(isLeafA && isLeafB))\n"
+" {\n"
+" //printf(\"Error: traversal exceeded maxStackDepth\");\n"
+" continue;\n"
+" }\n"
+" if(isInternalA)\n"
+" {\n"
+" int nodeAleftChild = node.x+1;\n"
+" bool isNodeALeftChildLeaf = isLeafNodeGlobal(&quantizedNodes[node.x+1]);\n"
+" int nodeArightChild = isNodeALeftChildLeaf? node.x+2 : node.x+1 + getEscapeIndexGlobal(&quantizedNodes[node.x+1]);\n"
+" if(isInternalB)\n"
+" { \n"
+" int nodeBleftChild = node.y+1;\n"
+" bool isNodeBLeftChildLeaf = isLeafNodeGlobal(&quantizedNodes[node.y+1]);\n"
+" int nodeBrightChild = isNodeBLeftChildLeaf? node.y+2 : node.y+1 + getEscapeIndexGlobal(&quantizedNodes[node.y+1]);\n"
+" nodeStack[depth++] = b3MakeInt2(nodeAleftChild, nodeBleftChild);\n"
+" nodeStack[depth++] = b3MakeInt2(nodeArightChild, nodeBleftChild);\n"
+" nodeStack[depth++] = b3MakeInt2(nodeAleftChild, nodeBrightChild);\n"
+" nodeStack[depth++] = b3MakeInt2(nodeArightChild, nodeBrightChild);\n"
+" }\n"
+" else\n"
+" {\n"
+" nodeStack[depth++] = b3MakeInt2(nodeAleftChild,node.y);\n"
+" nodeStack[depth++] = b3MakeInt2(nodeArightChild,node.y);\n"
+" }\n"
+" }\n"
+" else\n"
+" {\n"
+" if(isInternalB)\n"
+" {\n"
+" int nodeBleftChild = node.y+1;\n"
+" bool isNodeBLeftChildLeaf = isLeafNodeGlobal(&quantizedNodes[node.y+1]);\n"
+" int nodeBrightChild = isNodeBLeftChildLeaf? node.y+2 : node.y+1 + getEscapeIndexGlobal(&quantizedNodes[node.y+1]);\n"
+" nodeStack[depth++] = b3MakeInt2(node.x,nodeBleftChild);\n"
+" nodeStack[depth++] = b3MakeInt2(node.x,nodeBrightChild);\n"
+" }\n"
+" else\n"
+" {\n"
+" int compoundPairIdx = atomic_inc(numCompoundPairsOut);\n"
+" if (compoundPairIdx<maxNumCompoundPairsCapacity)\n"
+" {\n"
+" int childShapeIndexA = getTriangleIndexGlobal(&quantizedNodes[node.x]);\n"
+" int childShapeIndexB = getTriangleIndexGlobal(&quantizedNodes[node.y]);\n"
+" gpuCompoundPairsOut[compoundPairIdx] = (int4)(bodyIndexA,bodyIndexB,childShapeIndexA,childShapeIndexB);\n"
+" }\n"
+" }\n"
+" }\n"
+" }\n"
+" } while (depth);\n"
+" }\n"
+" }\n"
+" }\n"
+" \n"
+" return;\n"
+" }\n"
+" if ((collidables[collidableIndexA].m_shapeType==SHAPE_COMPOUND_OF_CONVEX_HULLS) ||(collidables[collidableIndexB].m_shapeType==SHAPE_COMPOUND_OF_CONVEX_HULLS))\n"
+" {\n"
+" if (collidables[collidableIndexA].m_shapeType==SHAPE_COMPOUND_OF_CONVEX_HULLS) \n"
+" {\n"
+" int numChildrenA = collidables[collidableIndexA].m_numChildShapes;\n"
+" for (int c=0;c<numChildrenA;c++)\n"
+" {\n"
+" int childShapeIndexA = collidables[collidableIndexA].m_shapeIndex+c;\n"
+" int childColIndexA = gpuChildShapes[childShapeIndexA].m_shapeIndex;\n"
+" float4 posA = rigidBodies[bodyIndexA].m_pos;\n"
+" float4 ornA = rigidBodies[bodyIndexA].m_quat;\n"
+" float4 childPosA = gpuChildShapes[childShapeIndexA].m_childPosition;\n"
+" float4 childOrnA = gpuChildShapes[childShapeIndexA].m_childOrientation;\n"
+" float4 newPosA = qtRotate(ornA,childPosA)+posA;\n"
+" float4 newOrnA = qtMul(ornA,childOrnA);\n"
+" int shapeIndexA = collidables[childColIndexA].m_shapeIndex;\n"
+" b3Aabb_t aabbAlocal = aabbLocalSpace[shapeIndexA];\n"
+" float margin = 0.f;\n"
+" \n"
+" b3Float4 aabbAMinWS;\n"
+" b3Float4 aabbAMaxWS;\n"
+" \n"
+" b3TransformAabb2(aabbAlocal.m_minVec,aabbAlocal.m_maxVec,margin,\n"
+" newPosA,\n"
+" newOrnA,\n"
+" &aabbAMinWS,&aabbAMaxWS);\n"
+" \n"
+" \n"
+" if (collidables[collidableIndexB].m_shapeType==SHAPE_COMPOUND_OF_CONVEX_HULLS)\n"
+" {\n"
+" int numChildrenB = collidables[collidableIndexB].m_numChildShapes;\n"
+" for (int b=0;b<numChildrenB;b++)\n"
+" {\n"
+" int childShapeIndexB = collidables[collidableIndexB].m_shapeIndex+b;\n"
+" int childColIndexB = gpuChildShapes[childShapeIndexB].m_shapeIndex;\n"
+" float4 ornB = rigidBodies[bodyIndexB].m_quat;\n"
+" float4 posB = rigidBodies[bodyIndexB].m_pos;\n"
+" float4 childPosB = gpuChildShapes[childShapeIndexB].m_childPosition;\n"
+" float4 childOrnB = gpuChildShapes[childShapeIndexB].m_childOrientation;\n"
+" float4 newPosB = transform(&childPosB,&posB,&ornB);\n"
+" float4 newOrnB = qtMul(ornB,childOrnB);\n"
+" int shapeIndexB = collidables[childColIndexB].m_shapeIndex;\n"
+" b3Aabb_t aabbBlocal = aabbLocalSpace[shapeIndexB];\n"
+" \n"
+" b3Float4 aabbBMinWS;\n"
+" b3Float4 aabbBMaxWS;\n"
+" \n"
+" b3TransformAabb2(aabbBlocal.m_minVec,aabbBlocal.m_maxVec,margin,\n"
+" newPosB,\n"
+" newOrnB,\n"
+" &aabbBMinWS,&aabbBMaxWS);\n"
+" \n"
+" \n"
+" \n"
+" bool aabbOverlap = b3TestAabbAgainstAabb(aabbAMinWS,aabbAMaxWS,aabbBMinWS,aabbBMaxWS);\n"
+" if (aabbOverlap)\n"
+" {\n"
+" int numFacesA = convexShapes[shapeIndexA].m_numFaces;\n"
+" float dmin = FLT_MAX;\n"
+" float4 posA = newPosA;\n"
+" posA.w = 0.f;\n"
+" float4 posB = newPosB;\n"
+" posB.w = 0.f;\n"
+" float4 c0local = convexShapes[shapeIndexA].m_localCenter;\n"
+" float4 ornA = newOrnA;\n"
+" float4 c0 = transform(&c0local, &posA, &ornA);\n"
+" float4 c1local = convexShapes[shapeIndexB].m_localCenter;\n"
+" float4 ornB =newOrnB;\n"
+" float4 c1 = transform(&c1local,&posB,&ornB);\n"
+" const float4 DeltaC2 = c0 - c1;\n"
+" {//\n"
+" int compoundPairIdx = atomic_inc(numCompoundPairsOut);\n"
+" if (compoundPairIdx<maxNumCompoundPairsCapacity)\n"
+" {\n"
+" gpuCompoundPairsOut[compoundPairIdx] = (int4)(bodyIndexA,bodyIndexB,childShapeIndexA,childShapeIndexB);\n"
+" }\n"
+" }//\n"
+" }//fi(1)\n"
+" } //for (int b=0\n"
+" }//if (collidables[collidableIndexB].\n"
+" else//if (collidables[collidableIndexB].m_shapeType==SHAPE_COMPOUND_OF_CONVEX_HULLS)\n"
+" {\n"
+" if (1)\n"
+" {\n"
+" int numFacesA = convexShapes[shapeIndexA].m_numFaces;\n"
+" float dmin = FLT_MAX;\n"
+" float4 posA = newPosA;\n"
+" posA.w = 0.f;\n"
+" float4 posB = rigidBodies[bodyIndexB].m_pos;\n"
+" posB.w = 0.f;\n"
+" float4 c0local = convexShapes[shapeIndexA].m_localCenter;\n"
+" float4 ornA = newOrnA;\n"
+" float4 c0 = transform(&c0local, &posA, &ornA);\n"
+" float4 c1local = convexShapes[shapeIndexB].m_localCenter;\n"
+" float4 ornB = rigidBodies[bodyIndexB].m_quat;\n"
+" float4 c1 = transform(&c1local,&posB,&ornB);\n"
+" const float4 DeltaC2 = c0 - c1;\n"
+" {\n"
+" int compoundPairIdx = atomic_inc(numCompoundPairsOut);\n"
+" if (compoundPairIdx<maxNumCompoundPairsCapacity)\n"
+" {\n"
+" gpuCompoundPairsOut[compoundPairIdx] = (int4)(bodyIndexA,bodyIndexB,childShapeIndexA,-1);\n"
+" }//if (compoundPairIdx<maxNumCompoundPairsCapacity)\n"
+" }//\n"
+" }//fi (1)\n"
+" }//if (collidables[collidableIndexB].m_shapeType==SHAPE_COMPOUND_OF_CONVEX_HULLS)\n"
+" }//for (int b=0;b<numChildrenB;b++) \n"
+" return;\n"
+" }//if (collidables[collidableIndexB].m_shapeType==SHAPE_COMPOUND_OF_CONVEX_HULLS)\n"
+" if ((collidables[collidableIndexA].m_shapeType!=SHAPE_CONCAVE_TRIMESH) \n"
+" && (collidables[collidableIndexB].m_shapeType==SHAPE_COMPOUND_OF_CONVEX_HULLS))\n"
+" {\n"
+" int numChildrenB = collidables[collidableIndexB].m_numChildShapes;\n"
+" for (int b=0;b<numChildrenB;b++)\n"
+" {\n"
+" int childShapeIndexB = collidables[collidableIndexB].m_shapeIndex+b;\n"
+" int childColIndexB = gpuChildShapes[childShapeIndexB].m_shapeIndex;\n"
+" float4 ornB = rigidBodies[bodyIndexB].m_quat;\n"
+" float4 posB = rigidBodies[bodyIndexB].m_pos;\n"
+" float4 childPosB = gpuChildShapes[childShapeIndexB].m_childPosition;\n"
+" float4 childOrnB = gpuChildShapes[childShapeIndexB].m_childOrientation;\n"
+" float4 newPosB = qtRotate(ornB,childPosB)+posB;\n"
+" float4 newOrnB = qtMul(ornB,childOrnB);\n"
+" int shapeIndexB = collidables[childColIndexB].m_shapeIndex;\n"
+" //////////////////////////////////////\n"
+" if (1)\n"
+" {\n"
+" int numFacesA = convexShapes[shapeIndexA].m_numFaces;\n"
+" float dmin = FLT_MAX;\n"
+" float4 posA = rigidBodies[bodyIndexA].m_pos;\n"
+" posA.w = 0.f;\n"
+" float4 posB = newPosB;\n"
+" posB.w = 0.f;\n"
+" float4 c0local = convexShapes[shapeIndexA].m_localCenter;\n"
+" float4 ornA = rigidBodies[bodyIndexA].m_quat;\n"
+" float4 c0 = transform(&c0local, &posA, &ornA);\n"
+" float4 c1local = convexShapes[shapeIndexB].m_localCenter;\n"
+" float4 ornB =newOrnB;\n"
+" float4 c1 = transform(&c1local,&posB,&ornB);\n"
+" const float4 DeltaC2 = c0 - c1;\n"
+" {//\n"
+" int compoundPairIdx = atomic_inc(numCompoundPairsOut);\n"
+" if (compoundPairIdx<maxNumCompoundPairsCapacity)\n"
+" {\n"
+" gpuCompoundPairsOut[compoundPairIdx] = (int4)(bodyIndexA,bodyIndexB,-1,childShapeIndexB);\n"
+" }//fi (compoundPairIdx<maxNumCompoundPairsCapacity)\n"
+" }//\n"
+" }//fi (1) \n"
+" }//for (int b=0;b<numChildrenB;b++)\n"
+" return;\n"
+" }//if (collidables[collidableIndexB].m_shapeType==SHAPE_COMPOUND_OF_CONVEX_HULLS)\n"
+" return;\n"
+" }//fi ((collidables[collidableIndexA].m_shapeType==SHAPE_COMPOUND_OF_CONVEX_HULLS) ||(collidables[collidableIndexB].m_shapeType==SHAPE_COMPOUND_OF_CONVEX_HULLS))\n"
+" }//i<numPairs\n"
+"}\n"
+"// work-in-progress\n"
+"__kernel void findSeparatingAxisKernel( __global const int4* pairs, \n"
+" __global const BodyData* rigidBodies, \n"
+" __global const btCollidableGpu* collidables,\n"
+" __global const ConvexPolyhedronCL* convexShapes, \n"
+" __global const float4* vertices,\n"
+" __global const float4* uniqueEdges,\n"
+" __global const btGpuFace* faces,\n"
+" __global const int* indices,\n"
+" __global btAabbCL* aabbs,\n"
+" __global volatile float4* separatingNormals,\n"
+" __global volatile int* hasSeparatingAxis,\n"
+" int numPairs\n"
+" )\n"
+"{\n"
+" int i = get_global_id(0);\n"
+" \n"
+" if (i<numPairs)\n"
+" {\n"
+" \n"
+" int bodyIndexA = pairs[i].x;\n"
+" int bodyIndexB = pairs[i].y;\n"
+" int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;\n"
+" int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;\n"
+" \n"
+" int shapeIndexA = collidables[collidableIndexA].m_shapeIndex;\n"
+" int shapeIndexB = collidables[collidableIndexB].m_shapeIndex;\n"
+" \n"
+" \n"
+" //once the broadphase avoids static-static pairs, we can remove this test\n"
+" if ((rigidBodies[bodyIndexA].m_invMass==0) &&(rigidBodies[bodyIndexB].m_invMass==0))\n"
+" {\n"
+" hasSeparatingAxis[i] = 0;\n"
+" return;\n"
+" }\n"
+" \n"
+" if ((collidables[collidableIndexA].m_shapeType!=SHAPE_CONVEX_HULL) ||(collidables[collidableIndexB].m_shapeType!=SHAPE_CONVEX_HULL))\n"
+" {\n"
+" hasSeparatingAxis[i] = 0;\n"
+" return;\n"
+" }\n"
+" \n"
+" if ((collidables[collidableIndexA].m_shapeType==SHAPE_CONCAVE_TRIMESH))\n"
+" {\n"
+" hasSeparatingAxis[i] = 0;\n"
+" return;\n"
+" }\n"
+" int numFacesA = convexShapes[shapeIndexA].m_numFaces;\n"
+" float dmin = FLT_MAX;\n"
+" float4 posA = rigidBodies[bodyIndexA].m_pos;\n"
+" posA.w = 0.f;\n"
+" float4 posB = rigidBodies[bodyIndexB].m_pos;\n"
+" posB.w = 0.f;\n"
+" float4 c0local = convexShapes[shapeIndexA].m_localCenter;\n"
+" float4 ornA = rigidBodies[bodyIndexA].m_quat;\n"
+" float4 c0 = transform(&c0local, &posA, &ornA);\n"
+" float4 c1local = convexShapes[shapeIndexB].m_localCenter;\n"
+" float4 ornB =rigidBodies[bodyIndexB].m_quat;\n"
+" float4 c1 = transform(&c1local,&posB,&ornB);\n"
+" const float4 DeltaC2 = c0 - c1;\n"
+" float4 sepNormal;\n"
+" \n"
+" bool sepA = findSeparatingAxis( &convexShapes[shapeIndexA], &convexShapes[shapeIndexB],posA,ornA,\n"
+" posB,ornB,\n"
+" DeltaC2,\n"
+" vertices,uniqueEdges,faces,\n"
+" indices,&sepNormal,&dmin);\n"
+" hasSeparatingAxis[i] = 4;\n"
+" if (!sepA)\n"
+" {\n"
+" hasSeparatingAxis[i] = 0;\n"
+" } else\n"
+" {\n"
+" bool sepB = findSeparatingAxis( &convexShapes[shapeIndexB],&convexShapes[shapeIndexA],posB,ornB,\n"
+" posA,ornA,\n"
+" DeltaC2,\n"
+" vertices,uniqueEdges,faces,\n"
+" indices,&sepNormal,&dmin);\n"
+" if (!sepB)\n"
+" {\n"
+" hasSeparatingAxis[i] = 0;\n"
+" } else\n"
+" {\n"
+" bool sepEE = findSeparatingAxisEdgeEdge( &convexShapes[shapeIndexA], &convexShapes[shapeIndexB],posA,ornA,\n"
+" posB,ornB,\n"
+" DeltaC2,\n"
+" vertices,uniqueEdges,faces,\n"
+" indices,&sepNormal,&dmin);\n"
+" if (!sepEE)\n"
+" {\n"
+" hasSeparatingAxis[i] = 0;\n"
+" } else\n"
+" {\n"
+" hasSeparatingAxis[i] = 1;\n"
+" separatingNormals[i] = sepNormal;\n"
+" }\n"
+" }\n"
+" }\n"
+" \n"
+" }\n"
+"}\n"
+"__kernel void findSeparatingAxisVertexFaceKernel( __global const int4* pairs, \n"
+" __global const BodyData* rigidBodies, \n"
+" __global const btCollidableGpu* collidables,\n"
+" __global const ConvexPolyhedronCL* convexShapes, \n"
+" __global const float4* vertices,\n"
+" __global const float4* uniqueEdges,\n"
+" __global const btGpuFace* faces,\n"
+" __global const int* indices,\n"
+" __global btAabbCL* aabbs,\n"
+" __global volatile float4* separatingNormals,\n"
+" __global volatile int* hasSeparatingAxis,\n"
+" __global float* dmins,\n"
+" int numPairs\n"
+" )\n"
+"{\n"
+" int i = get_global_id(0);\n"
+" \n"
+" if (i<numPairs)\n"
+" {\n"
+" \n"
+" int bodyIndexA = pairs[i].x;\n"
+" int bodyIndexB = pairs[i].y;\n"
+" int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;\n"
+" int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;\n"
+" \n"
+" int shapeIndexA = collidables[collidableIndexA].m_shapeIndex;\n"
+" int shapeIndexB = collidables[collidableIndexB].m_shapeIndex;\n"
+" \n"
+" hasSeparatingAxis[i] = 0; \n"
+" \n"
+" //once the broadphase avoids static-static pairs, we can remove this test\n"
+" if ((rigidBodies[bodyIndexA].m_invMass==0) &&(rigidBodies[bodyIndexB].m_invMass==0))\n"
+" {\n"
+" return;\n"
+" }\n"
+" \n"
+" if ((collidables[collidableIndexA].m_shapeType!=SHAPE_CONVEX_HULL) ||(collidables[collidableIndexB].m_shapeType!=SHAPE_CONVEX_HULL))\n"
+" {\n"
+" return;\n"
+" }\n"
+" \n"
+" int numFacesA = convexShapes[shapeIndexA].m_numFaces;\n"
+" float dmin = FLT_MAX;\n"
+" dmins[i] = dmin;\n"
+" \n"
+" float4 posA = rigidBodies[bodyIndexA].m_pos;\n"
+" posA.w = 0.f;\n"
+" float4 posB = rigidBodies[bodyIndexB].m_pos;\n"
+" posB.w = 0.f;\n"
+" float4 c0local = convexShapes[shapeIndexA].m_localCenter;\n"
+" float4 ornA = rigidBodies[bodyIndexA].m_quat;\n"
+" float4 c0 = transform(&c0local, &posA, &ornA);\n"
+" float4 c1local = convexShapes[shapeIndexB].m_localCenter;\n"
+" float4 ornB =rigidBodies[bodyIndexB].m_quat;\n"
+" float4 c1 = transform(&c1local,&posB,&ornB);\n"
+" const float4 DeltaC2 = c0 - c1;\n"
+" float4 sepNormal;\n"
+" \n"
+" bool sepA = findSeparatingAxis( &convexShapes[shapeIndexA], &convexShapes[shapeIndexB],posA,ornA,\n"
+" posB,ornB,\n"
+" DeltaC2,\n"
+" vertices,uniqueEdges,faces,\n"
+" indices,&sepNormal,&dmin);\n"
+" hasSeparatingAxis[i] = 4;\n"
+" if (!sepA)\n"
+" {\n"
+" hasSeparatingAxis[i] = 0;\n"
+" } else\n"
+" {\n"
+" bool sepB = findSeparatingAxis( &convexShapes[shapeIndexB],&convexShapes[shapeIndexA],posB,ornB,\n"
+" posA,ornA,\n"
+" DeltaC2,\n"
+" vertices,uniqueEdges,faces,\n"
+" indices,&sepNormal,&dmin);\n"
+" if (sepB)\n"
+" {\n"
+" dmins[i] = dmin;\n"
+" hasSeparatingAxis[i] = 1;\n"
+" separatingNormals[i] = sepNormal;\n"
+" }\n"
+" }\n"
+" \n"
+" }\n"
+"}\n"
+"__kernel void findSeparatingAxisEdgeEdgeKernel( __global const int4* pairs, \n"
+" __global const BodyData* rigidBodies, \n"
+" __global const btCollidableGpu* collidables,\n"
+" __global const ConvexPolyhedronCL* convexShapes, \n"
+" __global const float4* vertices,\n"
+" __global const float4* uniqueEdges,\n"
+" __global const btGpuFace* faces,\n"
+" __global const int* indices,\n"
+" __global btAabbCL* aabbs,\n"
+" __global float4* separatingNormals,\n"
+" __global int* hasSeparatingAxis,\n"
+" __global float* dmins,\n"
+" __global const float4* unitSphereDirections,\n"
+" int numUnitSphereDirections,\n"
+" int numPairs\n"
+" )\n"
+"{\n"
+" int i = get_global_id(0);\n"
+" \n"
+" if (i<numPairs)\n"
+" {\n"
+" if (hasSeparatingAxis[i])\n"
+" {\n"
+" \n"
+" int bodyIndexA = pairs[i].x;\n"
+" int bodyIndexB = pairs[i].y;\n"
+" \n"
+" int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;\n"
+" int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;\n"
+" \n"
+" int shapeIndexA = collidables[collidableIndexA].m_shapeIndex;\n"
+" int shapeIndexB = collidables[collidableIndexB].m_shapeIndex;\n"
+" \n"
+" \n"
+" int numFacesA = convexShapes[shapeIndexA].m_numFaces;\n"
+" \n"
+" float dmin = dmins[i];\n"
+" \n"
+" float4 posA = rigidBodies[bodyIndexA].m_pos;\n"
+" posA.w = 0.f;\n"
+" float4 posB = rigidBodies[bodyIndexB].m_pos;\n"
+" posB.w = 0.f;\n"
+" float4 c0local = convexShapes[shapeIndexA].m_localCenter;\n"
+" float4 ornA = rigidBodies[bodyIndexA].m_quat;\n"
+" float4 c0 = transform(&c0local, &posA, &ornA);\n"
+" float4 c1local = convexShapes[shapeIndexB].m_localCenter;\n"
+" float4 ornB =rigidBodies[bodyIndexB].m_quat;\n"
+" float4 c1 = transform(&c1local,&posB,&ornB);\n"
+" const float4 DeltaC2 = c0 - c1;\n"
+" float4 sepNormal = separatingNormals[i];\n"
+" \n"
+" \n"
+" \n"
+" bool sepEE = false;\n"
+" int numEdgeEdgeDirections = convexShapes[shapeIndexA].m_numUniqueEdges*convexShapes[shapeIndexB].m_numUniqueEdges;\n"
+" if (numEdgeEdgeDirections<=numUnitSphereDirections)\n"
+" {\n"
+" sepEE = findSeparatingAxisEdgeEdge( &convexShapes[shapeIndexA], &convexShapes[shapeIndexB],posA,ornA,\n"
+" posB,ornB,\n"
+" DeltaC2,\n"
+" vertices,uniqueEdges,faces,\n"
+" indices,&sepNormal,&dmin);\n"
+" \n"
+" if (!sepEE)\n"
+" {\n"
+" hasSeparatingAxis[i] = 0;\n"
+" } else\n"
+" {\n"
+" hasSeparatingAxis[i] = 1;\n"
+" separatingNormals[i] = sepNormal;\n"
+" }\n"
+" }\n"
+" /*\n"
+" ///else case is a separate kernel, to make Mac OSX OpenCL compiler happy\n"
+" else\n"
+" {\n"
+" sepEE = findSeparatingAxisUnitSphere(&convexShapes[shapeIndexA], &convexShapes[shapeIndexB],posA,ornA,\n"
+" posB,ornB,\n"
+" DeltaC2,\n"
+" vertices,unitSphereDirections,numUnitSphereDirections,\n"
+" &sepNormal,&dmin);\n"
+" if (!sepEE)\n"
+" {\n"
+" hasSeparatingAxis[i] = 0;\n"
+" } else\n"
+" {\n"
+" hasSeparatingAxis[i] = 1;\n"
+" separatingNormals[i] = sepNormal;\n"
+" }\n"
+" }\n"
+" */\n"
+" } //if (hasSeparatingAxis[i])\n"
+" }//(i<numPairs)\n"
+"}\n"
+"inline int findClippingFaces(const float4 separatingNormal,\n"
+" const ConvexPolyhedronCL* hullA, \n"
+" __global const ConvexPolyhedronCL* hullB,\n"
+" const float4 posA, const Quaternion ornA,const float4 posB, const Quaternion ornB,\n"
+" __global float4* worldVertsA1,\n"
+" __global float4* worldNormalsA1,\n"
+" __global float4* worldVertsB1,\n"
+" int capacityWorldVerts,\n"
+" const float minDist, float maxDist,\n"
+" const float4* verticesA,\n"
+" const btGpuFace* facesA,\n"
+" const int* indicesA,\n"
+" __global const float4* verticesB,\n"
+" __global const btGpuFace* facesB,\n"
+" __global const int* indicesB,\n"
+" __global int4* clippingFaces, int pairIndex)\n"
+"{\n"
+" int numContactsOut = 0;\n"
+" int numWorldVertsB1= 0;\n"
+" \n"
+" \n"
+" int closestFaceB=0;\n"
+" float dmax = -FLT_MAX;\n"
+" \n"
+" {\n"
+" for(int face=0;face<hullB->m_numFaces;face++)\n"
+" {\n"
+" const float4 Normal = make_float4(facesB[hullB->m_faceOffset+face].m_plane.x,\n"
+" facesB[hullB->m_faceOffset+face].m_plane.y, facesB[hullB->m_faceOffset+face].m_plane.z,0.f);\n"
+" const float4 WorldNormal = qtRotate(ornB, Normal);\n"
+" float d = dot3F4(WorldNormal,separatingNormal);\n"
+" if (d > dmax)\n"
+" {\n"
+" dmax = d;\n"
+" closestFaceB = face;\n"
+" }\n"
+" }\n"
+" }\n"
+" \n"
+" {\n"
+" const btGpuFace polyB = facesB[hullB->m_faceOffset+closestFaceB];\n"
+" int numVertices = polyB.m_numIndices;\n"
+" if (numVertices>capacityWorldVerts)\n"
+" numVertices = capacityWorldVerts;\n"
+" \n"
+" for(int e0=0;e0<numVertices;e0++)\n"
+" {\n"
+" if (e0<capacityWorldVerts)\n"
+" {\n"
+" const float4 b = verticesB[hullB->m_vertexOffset+indicesB[polyB.m_indexOffset+e0]];\n"
+" worldVertsB1[pairIndex*capacityWorldVerts+numWorldVertsB1++] = transform(&b,&posB,&ornB);\n"
+" }\n"
+" }\n"
+" }\n"
+" \n"
+" int closestFaceA=0;\n"
+" {\n"
+" float dmin = FLT_MAX;\n"
+" for(int face=0;face<hullA->m_numFaces;face++)\n"
+" {\n"
+" const float4 Normal = make_float4(\n"
+" facesA[hullA->m_faceOffset+face].m_plane.x,\n"
+" facesA[hullA->m_faceOffset+face].m_plane.y,\n"
+" facesA[hullA->m_faceOffset+face].m_plane.z,\n"
+" 0.f);\n"
+" const float4 faceANormalWS = qtRotate(ornA,Normal);\n"
+" \n"
+" float d = dot3F4(faceANormalWS,separatingNormal);\n"
+" if (d < dmin)\n"
+" {\n"
+" dmin = d;\n"
+" closestFaceA = face;\n"
+" worldNormalsA1[pairIndex] = faceANormalWS;\n"
+" }\n"
+" }\n"
+" }\n"
+" \n"
+" int numVerticesA = facesA[hullA->m_faceOffset+closestFaceA].m_numIndices;\n"
+" if (numVerticesA>capacityWorldVerts)\n"
+" numVerticesA = capacityWorldVerts;\n"
+" \n"
+" for(int e0=0;e0<numVerticesA;e0++)\n"
+" {\n"
+" if (e0<capacityWorldVerts)\n"
+" {\n"
+" const float4 a = verticesA[hullA->m_vertexOffset+indicesA[facesA[hullA->m_faceOffset+closestFaceA].m_indexOffset+e0]];\n"
+" worldVertsA1[pairIndex*capacityWorldVerts+e0] = transform(&a, &posA,&ornA);\n"
+" }\n"
+" }\n"
+" \n"
+" clippingFaces[pairIndex].x = closestFaceA;\n"
+" clippingFaces[pairIndex].y = closestFaceB;\n"
+" clippingFaces[pairIndex].z = numVerticesA;\n"
+" clippingFaces[pairIndex].w = numWorldVertsB1;\n"
+" \n"
+" \n"
+" return numContactsOut;\n"
+"}\n"
+"// work-in-progress\n"
+"__kernel void findConcaveSeparatingAxisKernel( __global int4* concavePairs,\n"
+" __global const BodyData* rigidBodies,\n"
+" __global const btCollidableGpu* collidables,\n"
+" __global const ConvexPolyhedronCL* convexShapes, \n"
+" __global const float4* vertices,\n"
+" __global const float4* uniqueEdges,\n"
+" __global const btGpuFace* faces,\n"
+" __global const int* indices,\n"
+" __global const btGpuChildShape* gpuChildShapes,\n"
+" __global btAabbCL* aabbs,\n"
+" __global float4* concaveSeparatingNormalsOut,\n"
+" __global int* concaveHasSeparatingNormals,\n"
+" __global int4* clippingFacesOut,\n"
+" __global float4* worldVertsA1GPU,\n"
+" __global float4* worldNormalsAGPU,\n"
+" __global float4* worldVertsB1GPU,\n"
+" int vertexFaceCapacity,\n"
+" int numConcavePairs\n"
+" )\n"
+"{\n"
+" int i = get_global_id(0);\n"
+" if (i>=numConcavePairs)\n"
+" return;\n"
+" concaveHasSeparatingNormals[i] = 0;\n"
+" int pairIdx = i;\n"
+" int bodyIndexA = concavePairs[i].x;\n"
+" int bodyIndexB = concavePairs[i].y;\n"
+" int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;\n"
+" int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;\n"
+" int shapeIndexA = collidables[collidableIndexA].m_shapeIndex;\n"
+" int shapeIndexB = collidables[collidableIndexB].m_shapeIndex;\n"
+" if (collidables[collidableIndexB].m_shapeType!=SHAPE_CONVEX_HULL&&\n"
+" collidables[collidableIndexB].m_shapeType!=SHAPE_COMPOUND_OF_CONVEX_HULLS)\n"
+" {\n"
+" concavePairs[pairIdx].w = -1;\n"
+" return;\n"
+" }\n"
+" int numFacesA = convexShapes[shapeIndexA].m_numFaces;\n"
+" int numActualConcaveConvexTests = 0;\n"
+" \n"
+" int f = concavePairs[i].z;\n"
+" \n"
+" bool overlap = false;\n"
+" \n"
+" ConvexPolyhedronCL convexPolyhedronA;\n"
+" //add 3 vertices of the triangle\n"
+" convexPolyhedronA.m_numVertices = 3;\n"
+" convexPolyhedronA.m_vertexOffset = 0;\n"
+" float4 localCenter = make_float4(0.f,0.f,0.f,0.f);\n"
+" btGpuFace face = faces[convexShapes[shapeIndexA].m_faceOffset+f];\n"
+" float4 triMinAabb, triMaxAabb;\n"
+" btAabbCL triAabb;\n"
+" triAabb.m_min = make_float4(1e30f,1e30f,1e30f,0.f);\n"
+" triAabb.m_max = make_float4(-1e30f,-1e30f,-1e30f,0.f);\n"
+" \n"
+" float4 verticesA[3];\n"
+" for (int i=0;i<3;i++)\n"
+" {\n"
+" int index = indices[face.m_indexOffset+i];\n"
+" float4 vert = vertices[convexShapes[shapeIndexA].m_vertexOffset+index];\n"
+" verticesA[i] = vert;\n"
+" localCenter += vert;\n"
+" \n"
+" triAabb.m_min = min(triAabb.m_min,vert); \n"
+" triAabb.m_max = max(triAabb.m_max,vert); \n"
+" }\n"
+" overlap = true;\n"
+" overlap = (triAabb.m_min.x > aabbs[bodyIndexB].m_max.x || triAabb.m_max.x < aabbs[bodyIndexB].m_min.x) ? false : overlap;\n"
+" overlap = (triAabb.m_min.z > aabbs[bodyIndexB].m_max.z || triAabb.m_max.z < aabbs[bodyIndexB].m_min.z) ? false : overlap;\n"
+" overlap = (triAabb.m_min.y > aabbs[bodyIndexB].m_max.y || triAabb.m_max.y < aabbs[bodyIndexB].m_min.y) ? false : overlap;\n"
+" \n"
+" if (overlap)\n"
+" {\n"
+" float dmin = FLT_MAX;\n"
+" int hasSeparatingAxis=5;\n"
+" float4 sepAxis=make_float4(1,2,3,4);\n"
+" int localCC=0;\n"
+" numActualConcaveConvexTests++;\n"
+" //a triangle has 3 unique edges\n"
+" convexPolyhedronA.m_numUniqueEdges = 3;\n"
+" convexPolyhedronA.m_uniqueEdgesOffset = 0;\n"
+" float4 uniqueEdgesA[3];\n"
+" \n"
+" uniqueEdgesA[0] = (verticesA[1]-verticesA[0]);\n"
+" uniqueEdgesA[1] = (verticesA[2]-verticesA[1]);\n"
+" uniqueEdgesA[2] = (verticesA[0]-verticesA[2]);\n"
+" convexPolyhedronA.m_faceOffset = 0;\n"
+" \n"
+" float4 normal = make_float4(face.m_plane.x,face.m_plane.y,face.m_plane.z,0.f);\n"
+" \n"
+" btGpuFace facesA[TRIANGLE_NUM_CONVEX_FACES];\n"
+" int indicesA[3+3+2+2+2];\n"
+" int curUsedIndices=0;\n"
+" int fidx=0;\n"
+" //front size of triangle\n"
+" {\n"
+" facesA[fidx].m_indexOffset=curUsedIndices;\n"
+" indicesA[0] = 0;\n"
+" indicesA[1] = 1;\n"
+" indicesA[2] = 2;\n"
+" curUsedIndices+=3;\n"
+" float c = face.m_plane.w;\n"
+" facesA[fidx].m_plane.x = normal.x;\n"
+" facesA[fidx].m_plane.y = normal.y;\n"
+" facesA[fidx].m_plane.z = normal.z;\n"
+" facesA[fidx].m_plane.w = c;\n"
+" facesA[fidx].m_numIndices=3;\n"
+" }\n"
+" fidx++;\n"
+" //back size of triangle\n"
+" {\n"
+" facesA[fidx].m_indexOffset=curUsedIndices;\n"
+" indicesA[3]=2;\n"
+" indicesA[4]=1;\n"
+" indicesA[5]=0;\n"
+" curUsedIndices+=3;\n"
+" float c = dot(normal,verticesA[0]);\n"
+" float c1 = -face.m_plane.w;\n"
+" facesA[fidx].m_plane.x = -normal.x;\n"
+" facesA[fidx].m_plane.y = -normal.y;\n"
+" facesA[fidx].m_plane.z = -normal.z;\n"
+" facesA[fidx].m_plane.w = c;\n"
+" facesA[fidx].m_numIndices=3;\n"
+" }\n"
+" fidx++;\n"
+" bool addEdgePlanes = true;\n"
+" if (addEdgePlanes)\n"
+" {\n"
+" int numVertices=3;\n"
+" int prevVertex = numVertices-1;\n"
+" for (int i=0;i<numVertices;i++)\n"
+" {\n"
+" float4 v0 = verticesA[i];\n"
+" float4 v1 = verticesA[prevVertex];\n"
+" \n"
+" float4 edgeNormal = normalize(cross(normal,v1-v0));\n"
+" float c = -dot(edgeNormal,v0);\n"
+" facesA[fidx].m_numIndices = 2;\n"
+" facesA[fidx].m_indexOffset=curUsedIndices;\n"
+" indicesA[curUsedIndices++]=i;\n"
+" indicesA[curUsedIndices++]=prevVertex;\n"
+" \n"
+" facesA[fidx].m_plane.x = edgeNormal.x;\n"
+" facesA[fidx].m_plane.y = edgeNormal.y;\n"
+" facesA[fidx].m_plane.z = edgeNormal.z;\n"
+" facesA[fidx].m_plane.w = c;\n"
+" fidx++;\n"
+" prevVertex = i;\n"
+" }\n"
+" }\n"
+" convexPolyhedronA.m_numFaces = TRIANGLE_NUM_CONVEX_FACES;\n"
+" convexPolyhedronA.m_localCenter = localCenter*(1.f/3.f);\n"
+" float4 posA = rigidBodies[bodyIndexA].m_pos;\n"
+" posA.w = 0.f;\n"
+" float4 posB = rigidBodies[bodyIndexB].m_pos;\n"
+" posB.w = 0.f;\n"
+" float4 ornA = rigidBodies[bodyIndexA].m_quat;\n"
+" float4 ornB =rigidBodies[bodyIndexB].m_quat;\n"
+" \n"
+" ///////////////////\n"
+" ///compound shape support\n"
+" if (collidables[collidableIndexB].m_shapeType==SHAPE_COMPOUND_OF_CONVEX_HULLS)\n"
+" {\n"
+" int compoundChild = concavePairs[pairIdx].w;\n"
+" int childShapeIndexB = compoundChild;//collidables[collidableIndexB].m_shapeIndex+compoundChild;\n"
+" int childColIndexB = gpuChildShapes[childShapeIndexB].m_shapeIndex;\n"
+" float4 childPosB = gpuChildShapes[childShapeIndexB].m_childPosition;\n"
+" float4 childOrnB = gpuChildShapes[childShapeIndexB].m_childOrientation;\n"
+" float4 newPosB = transform(&childPosB,&posB,&ornB);\n"
+" float4 newOrnB = qtMul(ornB,childOrnB);\n"
+" posB = newPosB;\n"
+" ornB = newOrnB;\n"
+" shapeIndexB = collidables[childColIndexB].m_shapeIndex;\n"
+" }\n"
+" //////////////////\n"
+" float4 c0local = convexPolyhedronA.m_localCenter;\n"
+" float4 c0 = transform(&c0local, &posA, &ornA);\n"
+" float4 c1local = convexShapes[shapeIndexB].m_localCenter;\n"
+" float4 c1 = transform(&c1local,&posB,&ornB);\n"
+" const float4 DeltaC2 = c0 - c1;\n"
+" bool sepA = findSeparatingAxisLocalA( &convexPolyhedronA, &convexShapes[shapeIndexB],\n"
+" posA,ornA,\n"
+" posB,ornB,\n"
+" DeltaC2,\n"
+" verticesA,uniqueEdgesA,facesA,indicesA,\n"
+" vertices,uniqueEdges,faces,indices,\n"
+" &sepAxis,&dmin);\n"
+" hasSeparatingAxis = 4;\n"
+" if (!sepA)\n"
+" {\n"
+" hasSeparatingAxis = 0;\n"
+" } else\n"
+" {\n"
+" bool sepB = findSeparatingAxisLocalB( &convexShapes[shapeIndexB],&convexPolyhedronA,\n"
+" posB,ornB,\n"
+" posA,ornA,\n"
+" DeltaC2,\n"
+" vertices,uniqueEdges,faces,indices,\n"
+" verticesA,uniqueEdgesA,facesA,indicesA,\n"
+" &sepAxis,&dmin);\n"
+" if (!sepB)\n"
+" {\n"
+" hasSeparatingAxis = 0;\n"
+" } else\n"
+" {\n"
+" bool sepEE = findSeparatingAxisEdgeEdgeLocalA( &convexPolyhedronA, &convexShapes[shapeIndexB],\n"
+" posA,ornA,\n"
+" posB,ornB,\n"
+" DeltaC2,\n"
+" verticesA,uniqueEdgesA,facesA,indicesA,\n"
+" vertices,uniqueEdges,faces,indices,\n"
+" &sepAxis,&dmin);\n"
+" \n"
+" if (!sepEE)\n"
+" {\n"
+" hasSeparatingAxis = 0;\n"
+" } else\n"
+" {\n"
+" hasSeparatingAxis = 1;\n"
+" }\n"
+" }\n"
+" } \n"
+" \n"
+" if (hasSeparatingAxis)\n"
+" {\n"
+" sepAxis.w = dmin;\n"
+" concaveSeparatingNormalsOut[pairIdx]=sepAxis;\n"
+" concaveHasSeparatingNormals[i]=1;\n"
+" float minDist = -1e30f;\n"
+" float maxDist = 0.02f;\n"
+" \n"
+" findClippingFaces(sepAxis,\n"
+" &convexPolyhedronA,\n"
+" &convexShapes[shapeIndexB],\n"
+" posA,ornA,\n"
+" posB,ornB,\n"
+" worldVertsA1GPU,\n"
+" worldNormalsAGPU,\n"
+" worldVertsB1GPU,\n"
+" vertexFaceCapacity,\n"
+" minDist, maxDist,\n"
+" verticesA,\n"
+" facesA,\n"
+" indicesA,\n"
+" vertices,\n"
+" faces,\n"
+" indices,\n"
+" clippingFacesOut, pairIdx);\n"
+" } else\n"
+" { \n"
+" //mark this pair as in-active\n"
+" concavePairs[pairIdx].w = -1;\n"
+" }\n"
+" }\n"
+" else\n"
+" { \n"
+" //mark this pair as in-active\n"
+" concavePairs[pairIdx].w = -1;\n"
+" }\n"
+" \n"
+" concavePairs[pairIdx].z = -1;//now z is used for existing/persistent contacts\n"
+"}\n"
+;