2017-08-01 14:30:58 +02:00
//this file is autogenerated using stringify.bat (premake --stringify) in the build folder of this project
2019-01-03 14:26:51 +01:00
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 " ;