Bart/bullet3
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Code-style consistency improvement:

Browse Source 
Apply clang-format-all.sh using the _clang-format file through all the cpp/.h files.
make sure not to apply it to certain serialization structures, since some parser expects the * as part of the name, instead of type.
This commit contains no other changes aside from adding and applying clang-format-all.sh
This commit is contained in:
erwincoumans
2018-09-23 14:17:31 -07:00
parent b73b05e9fb
commit ab8f16961e
1773 changed files with 1081087 additions and 474249 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
src/Bullet3OpenCL/RigidBody/b3GpuConstraint4.h
View File

@@ -5,14 +5,13 @@
#include "Bullet3Dynamics/shared/b3ContactConstraint4.h"
B3_ATTRIBUTE_ALIGNED16(struct) b3GpuConstraint4 : public b3ContactConstraint4
B3_ATTRIBUTE_ALIGNED16(struct)
b3GpuConstraint4 : public b3ContactConstraint4
{
B3_DECLARE_ALIGNED_ALLOCATOR();
inline void setFrictionCoeff(float value) { m_linear[3] = value; }
inline float getFrictionCoeff() const { return m_linear[3]; }
inline void setFrictionCoeff(float value) { m_linear[3] = value; }
inline float getFrictionCoeff() const { return m_linear[3]; }
};
#endif //B3_CONSTRAINT4_h
#endif //B3_CONSTRAINT4_h

75
src/Bullet3OpenCL/RigidBody/b3GpuGenericConstraint.cpp
View File

@@ -19,11 +19,11 @@ subject to the following restrictions:
#include <new>
#include "Bullet3Common/b3Transform.h"
void b3GpuGenericConstraint::getInfo1 (unsigned int* info,const b3RigidBodyData* bodies)
void b3GpuGenericConstraint::getInfo1(unsigned int* info, const b3RigidBodyData* bodies)
{
switch (m_constraintType)
{
case B3_GPU_POINT2POINT_CONSTRAINT_TYPE:
case B3_GPU_POINT2POINT_CONSTRAINT_TYPE:
{
*info = 3;
break;
@@ -35,7 +35,7 @@ void b3GpuGenericConstraint::getInfo1 (unsigned int* info,const b3RigidBodyData*
};
}
void getInfo2Point2Point(b3GpuGenericConstraint* constraint, b3GpuConstraintInfo2* info, const b3RigidBodyData* bodies)
void getInfo2Point2Point(b3GpuGenericConstraint* constraint, b3GpuConstraintInfo2* info, const b3RigidBodyData* bodies)
{
b3Transform trA;
trA.setIdentity();
@@ -47,54 +47,52 @@ void getInfo2Point2Point(b3GpuGenericConstraint* constraint, b3GpuConstraintInfo
trB.setOrigin(bodies[constraint->m_rbB].m_pos);
trB.setRotation(bodies[constraint->m_rbB].m_quat);
// anchor points in global coordinates with respect to body PORs.
// set jacobian
info->m_J1linearAxis[0] = 1;
info->m_J1linearAxis[info->rowskip+1] = 1;
info->m_J1linearAxis[2*info->rowskip+2] = 1;
// anchor points in global coordinates with respect to body PORs.
b3Vector3 a1 = trA.getBasis()*constraint->getPivotInA();
// set jacobian
info->m_J1linearAxis[0] = 1;
info->m_J1linearAxis[info->rowskip + 1] = 1;
info->m_J1linearAxis[2 * info->rowskip + 2] = 1;
b3Vector3 a1 = trA.getBasis() * constraint->getPivotInA();
//b3Vector3 a1a = b3QuatRotate(trA.getRotation(),constraint->getPivotInA());
{
b3Vector3* angular0 = (b3Vector3*)(info->m_J1angularAxis);
b3Vector3* angular1 = (b3Vector3*)(info->m_J1angularAxis+info->rowskip);
b3Vector3* angular2 = (b3Vector3*)(info->m_J1angularAxis+2*info->rowskip);
b3Vector3* angular1 = (b3Vector3*)(info->m_J1angularAxis + info->rowskip);
b3Vector3* angular2 = (b3Vector3*)(info->m_J1angularAxis + 2 * info->rowskip);
b3Vector3 a1neg = -a1;
a1neg.getSkewSymmetricMatrix(angular0,angular1,angular2);
a1neg.getSkewSymmetricMatrix(angular0, angular1, angular2);
}
if (info->m_J2linearAxis)
{
info->m_J2linearAxis[0] = -1;
info->m_J2linearAxis[info->rowskip+1] = -1;
info->m_J2linearAxis[2*info->rowskip+2] = -1;
info->m_J2linearAxis[info->rowskip + 1] = -1;
info->m_J2linearAxis[2 * info->rowskip + 2] = -1;
}
b3Vector3 a2 = trB.getBasis()*constraint->getPivotInB();
b3Vector3 a2 = trB.getBasis() * constraint->getPivotInB();
{
// b3Vector3 a2n = -a2;
// b3Vector3 a2n = -a2;
b3Vector3* angular0 = (b3Vector3*)(info->m_J2angularAxis);
b3Vector3* angular1 = (b3Vector3*)(info->m_J2angularAxis+info->rowskip);
b3Vector3* angular2 = (b3Vector3*)(info->m_J2angularAxis+2*info->rowskip);
a2.getSkewSymmetricMatrix(angular0,angular1,angular2);
b3Vector3* angular1 = (b3Vector3*)(info->m_J2angularAxis + info->rowskip);
b3Vector3* angular2 = (b3Vector3*)(info->m_J2angularAxis + 2 * info->rowskip);
a2.getSkewSymmetricMatrix(angular0, angular1, angular2);
}
// set right hand side
// b3Scalar currERP = (m_flags & B3_P2P_FLAGS_ERP) ? m_erp : info->erp;
// set right hand side
// b3Scalar currERP = (m_flags & B3_P2P_FLAGS_ERP) ? m_erp : info->erp;
b3Scalar currERP = info->erp;
b3Scalar k = info->fps * currERP;
int j;
for (j=0; j<3; j++)
{
info->m_constraintError[j*info->rowskip] = k * (a2[j] + trB.getOrigin()[j] - a1[j] - trA.getOrigin()[j]);
int j;
for (j = 0; j < 3; j++)
{
info->m_constraintError[j * info->rowskip] = k * (a2[j] + trB.getOrigin()[j] - a1[j] - trA.getOrigin()[j]);
//printf("info->m_constraintError[%d]=%f\n",j,info->m_constraintError[j]);
}
}
#if 0
if(m_flags & B3_P2P_FLAGS_CFM)
{
@@ -117,21 +115,20 @@ void getInfo2Point2Point(b3GpuGenericConstraint* constraint, b3GpuConstraintInfo
}
info->m_damping = m_setting.m_damping;
#endif
}
void b3GpuGenericConstraint::getInfo2 (b3GpuConstraintInfo2* info, const b3RigidBodyData* bodies)
void b3GpuGenericConstraint::getInfo2(b3GpuConstraintInfo2* info, const b3RigidBodyData* bodies)
{
switch (m_constraintType)
{
case B3_GPU_POINT2POINT_CONSTRAINT_TYPE:
case B3_GPU_POINT2POINT_CONSTRAINT_TYPE:
{
getInfo2Point2Point(this,info,bodies);
getInfo2Point2Point(this, info, bodies);
break;
};
default:
{
b3Assert(0);
}
{
b3Assert(0);
}
};
}

64
src/Bullet3OpenCL/RigidBody/b3GpuGenericConstraint.h
View File

@@ -20,37 +20,35 @@ subject to the following restrictions:
struct b3RigidBodyData;
enum B3_CONSTRAINT_FLAGS
{
B3_CONSTRAINT_FLAG_ENABLED=1,
B3_CONSTRAINT_FLAG_ENABLED = 1,
};
enum b3GpuGenericConstraintType
{
B3_GPU_POINT2POINT_CONSTRAINT_TYPE=3,
B3_GPU_FIXED_CONSTRAINT_TYPE=4,
// B3_HINGE_CONSTRAINT_TYPE,
// B3_CONETWIST_CONSTRAINT_TYPE,
// B3_D6_CONSTRAINT_TYPE,
// B3_SLIDER_CONSTRAINT_TYPE,
// B3_CONTACT_CONSTRAINT_TYPE,
// B3_D6_SPRING_CONSTRAINT_TYPE,
// B3_GEAR_CONSTRAINT_TYPE,
B3_GPU_POINT2POINT_CONSTRAINT_TYPE = 3,
B3_GPU_FIXED_CONSTRAINT_TYPE = 4,
// B3_HINGE_CONSTRAINT_TYPE,
// B3_CONETWIST_CONSTRAINT_TYPE,
// B3_D6_CONSTRAINT_TYPE,
// B3_SLIDER_CONSTRAINT_TYPE,
// B3_CONTACT_CONSTRAINT_TYPE,
// B3_D6_SPRING_CONSTRAINT_TYPE,
// B3_GEAR_CONSTRAINT_TYPE,
B3_GPU_MAX_CONSTRAINT_TYPE
};
struct b3GpuConstraintInfo2
struct b3GpuConstraintInfo2
{
// integrator parameters: frames per second (1/stepsize), default error
// reduction parameter (0..1).
b3Scalar fps,erp;
b3Scalar fps, erp;
// for the first and second body, pointers to two (linear and angular)
// n*3 jacobian sub matrices, stored by rows. these matrices will have
// been initialized to 0 on entry. if the second body is zero then the
// J2xx pointers may be 0.
b3Scalar *m_J1linearAxis,*m_J1angularAxis,*m_J2linearAxis,*m_J2angularAxis;
b3Scalar *m_J1linearAxis, *m_J1angularAxis, *m_J2linearAxis, *m_J2angularAxis;
// elements to jump from one row to the next in J's
int rowskip;
@@ -58,44 +56,44 @@ struct b3GpuConstraintInfo2
// right hand sides of the equation J*v = c + cfm * lambda. cfm is the
// "constraint force mixing" vector. c is set to zero on entry, cfm is
// set to a constant value (typically very small or zero) value on entry.
b3Scalar *m_constraintError,*cfm;
b3Scalar *m_constraintError, *cfm;
// lo and hi limits for variables (set to -/+ infinity on entry).
b3Scalar *m_lowerLimit,*m_upperLimit;
b3Scalar *m_lowerLimit, *m_upperLimit;
// findex vector for variables. see the LCP solver interface for a
// description of what this does. this is set to -1 on entry.
// note that the returned indexes are relative to the first index of
// the constraint.
int *findex;
int* findex;
// number of solver iterations
int m_numIterations;
//damping of the velocity
b3Scalar m_damping;
b3Scalar m_damping;
};
B3_ATTRIBUTE_ALIGNED16(struct) b3GpuGenericConstraint
B3_ATTRIBUTE_ALIGNED16(struct)
b3GpuGenericConstraint
{
int m_constraintType;
int m_rbA;
int m_rbB;
float m_breakingImpulseThreshold;
int m_constraintType;
int m_rbA;
int m_rbB;
float m_breakingImpulseThreshold;
b3Vector3 m_pivotInA;
b3Vector3 m_pivotInB;
b3Quaternion m_relTargetAB;
int m_flags;
int m_flags;
int m_uid;
int m_padding[2];
int getRigidBodyA() const
int getRigidBodyA() const
{
return m_rbA;
}
int getRigidBodyB() const
int getRigidBodyB() const
{
return m_rbB;
}
@@ -121,12 +119,10 @@ B3_ATTRIBUTE_ALIGNED16(struct) b3GpuGenericConstraint
}
///internal method used by the constraint solver, don't use them directly
void getInfo1 (unsigned int* info,const b3RigidBodyData* bodies);
void getInfo1(unsigned int* info, const b3RigidBodyData* bodies);
///internal method used by the constraint solver, don't use them directly
void getInfo2 (b3GpuConstraintInfo2* info, const b3RigidBodyData* bodies);
void getInfo2(b3GpuConstraintInfo2 * info, const b3RigidBodyData* bodies);
};
#endif //B3_GPU_GENERIC_CONSTRAINT_H
#endif //B3_GPU_GENERIC_CONSTRAINT_H

641
src/Bullet3OpenCL/RigidBody/b3GpuJacobiContactSolver.cpp
View File

File diff suppressed because it is too large Load Diff

22
src/Bullet3OpenCL/RigidBody/b3GpuJacobiContactSolver.h
View File

@@ -8,7 +8,6 @@
#include "Bullet3Collision/NarrowPhaseCollision/shared/b3Contact4Data.h"
#include "Bullet3OpenCL/ParallelPrimitives/b3OpenCLArray.h"
//struct b3InertiaData;
//b3InertiaData
@@ -21,21 +20,20 @@ struct b3JacobiSolverInfo
float m_deltaTime;
float m_positionDrift;
float m_positionConstraintCoeff;
int m_numIterations;
int m_numIterations;
b3JacobiSolverInfo()
:m_fixedBodyIndex(0),
m_deltaTime(1./60.f),
m_positionDrift( 0.005f ),
m_positionConstraintCoeff( 0.99f ),
m_numIterations(7)
: m_fixedBodyIndex(0),
m_deltaTime(1. / 60.f),
m_positionDrift(0.005f),
m_positionConstraintCoeff(0.99f),
m_numIterations(7)
{
}
};
class b3GpuJacobiContactSolver
{
protected:
struct b3GpuJacobiSolverInternalData* m_data;
cl_context m_context;
@@ -43,20 +41,16 @@ protected:
cl_command_queue m_queue;
public:
b3GpuJacobiContactSolver(cl_context ctx, cl_device_id device, cl_command_queue queue, int pairCapacity);
virtual ~b3GpuJacobiContactSolver();
void solveContacts(int numBodies, cl_mem bodyBuf, cl_mem inertiaBuf, int numContacts, cl_mem contactBuf, const struct b3Config& config, int static0Index);
void solveGroupHost(b3RigidBodyData* bodies,b3InertiaData* inertias,int numBodies,struct b3Contact4* manifoldPtr, int numManifolds,const b3JacobiSolverInfo& solverInfo);
void solveGroupHost(b3RigidBodyData* bodies, b3InertiaData* inertias, int numBodies, struct b3Contact4* manifoldPtr, int numManifolds, const b3JacobiSolverInfo& solverInfo);
//void solveGroupHost(btRigidBodyCL* bodies,b3InertiaData* inertias,int numBodies,btContact4* manifoldPtr, int numManifolds,btTypedConstraint** constraints,int numConstraints,const btJacobiSolverInfo& solverInfo);
//b3Scalar solveGroup(b3OpenCLArray<b3RigidBodyData>* gpuBodies,b3OpenCLArray<b3InertiaData>* gpuInertias, int numBodies,b3OpenCLArray<b3GpuGenericConstraint>* gpuConstraints,int numConstraints,const b3ContactSolverInfo& infoGlobal);
//void solveGroup(btOpenCLArray<btRigidBodyCL>* bodies,btOpenCLArray<btInertiaCL>* inertias,btOpenCLArray<btContact4>* manifoldPtr,const btJacobiSolverInfo& solverInfo);
//void solveGroupMixed(btOpenCLArray<btRigidBodyCL>* bodies,btOpenCLArray<btInertiaCL>* inertias,btOpenCLArray<btContact4>* manifoldPtr,const btJacobiSolverInfo& solverInfo);
};
#endif //B3_GPU_JACOBI_CONTACT_SOLVER_H
#endif //B3_GPU_JACOBI_CONTACT_SOLVER_H

752
src/Bullet3OpenCL/RigidBody/b3GpuNarrowPhase.cpp
View File

File diff suppressed because it is too large Load Diff

72
src/Bullet3OpenCL/RigidBody/b3GpuNarrowPhase.h
View File

@@ -9,11 +9,10 @@
class b3GpuNarrowPhase
{
protected:
struct b3GpuNarrowPhaseInternalData* m_data;
struct b3GpuNarrowPhaseInternalData* m_data;
int m_acceleratedCompanionShapeIndex;
int m_planeBodyIndex;
int m_static0Index;
int m_static0Index;
cl_context m_context;
cl_device_id m_device;
@@ -23,64 +22,58 @@ protected:
int registerConcaveMeshShape(b3AlignedObjectArray<b3Vector3>* vertices, b3AlignedObjectArray<int>* indices, b3Collidable& col, const float* scaling);
public:
b3GpuNarrowPhase(cl_context vtx, cl_device_id dev, cl_command_queue q, const struct b3Config& config);
virtual ~b3GpuNarrowPhase(void);
int registerSphereShape(float radius);
int registerPlaneShape(const b3Vector3& planeNormal, float planeConstant);
int registerSphereShape(float radius);
int registerPlaneShape(const b3Vector3& planeNormal, float planeConstant);
int registerCompoundShape(b3AlignedObjectArray<b3GpuChildShape>* childShapes);
int registerFace(const b3Vector3& faceNormal, float faceConstant);
int registerConcaveMesh(b3AlignedObjectArray<b3Vector3>* vertices, b3AlignedObjectArray<int>* indices,const float* scaling);
int registerConcaveMesh(b3AlignedObjectArray<b3Vector3>* vertices, b3AlignedObjectArray<int>* indices, const float* scaling);
//do they need to be merged?
int registerConvexHullShape(b3ConvexUtility* utilPtr);
int registerConvexHullShape(const float* vertices, int strideInBytes, int numVertices, const float* scaling);
int registerRigidBody(int collidableIndex, float mass, const float* position, const float* orientation, const float* aabbMin, const float* aabbMax,bool writeToGpu);
void setObjectTransform(const float* position, const float* orientation , int bodyIndex);
int registerConvexHullShape(b3ConvexUtility* utilPtr);
int registerConvexHullShape(const float* vertices, int strideInBytes, int numVertices, const float* scaling);
void writeAllBodiesToGpu();
void reset();
void readbackAllBodiesToCpu();
bool getObjectTransformFromCpu(float* position, float* orientation , int bodyIndex) const;
int registerRigidBody(int collidableIndex, float mass, const float* position, const float* orientation, const float* aabbMin, const float* aabbMax, bool writeToGpu);
void setObjectTransform(const float* position, const float* orientation, int bodyIndex);
void setObjectTransformCpu(float* position, float* orientation , int bodyIndex);
void writeAllBodiesToGpu();
void reset();
void readbackAllBodiesToCpu();
bool getObjectTransformFromCpu(float* position, float* orientation, int bodyIndex) const;
void setObjectTransformCpu(float* position, float* orientation, int bodyIndex);
void setObjectVelocityCpu(float* linVel, float* angVel, int bodyIndex);
virtual void computeContacts(cl_mem broadphasePairs, int numBroadphasePairs, cl_mem aabbsWorldSpace, int numObjects);
cl_mem getBodiesGpu();
cl_mem getBodiesGpu();
const struct b3RigidBodyData* getBodiesCpu() const;
//struct b3RigidBodyData* getBodiesCpu();
int getNumBodiesGpu() const;
int getNumBodiesGpu() const;
cl_mem getBodyInertiasGpu();
int getNumBodyInertiasGpu() const;
cl_mem getBodyInertiasGpu();
int getNumBodyInertiasGpu() const;
cl_mem getCollidablesGpu();
cl_mem getCollidablesGpu();
const struct b3Collidable* getCollidablesCpu() const;
int getNumCollidablesGpu() const;
int getNumCollidablesGpu() const;
const struct b3SapAabb* getLocalSpaceAabbsCpu() const;
const struct b3Contact4* getContactsCPU() const;
cl_mem getContactsGpu();
int getNumContactsGpu() const;
cl_mem getContactsGpu();
int getNumContactsGpu() const;
cl_mem getAabbLocalSpaceBufferGpu();
cl_mem getAabbLocalSpaceBufferGpu();
int getNumRigidBodies() const;
int allocateCollidable();
@@ -92,18 +85,17 @@ public:
b3Collidable& getCollidableCpu(int collidableIndex);
const b3Collidable& getCollidableCpu(int collidableIndex) const;
const b3GpuNarrowPhaseInternalData* getInternalData() const
const b3GpuNarrowPhaseInternalData* getInternalData() const
{
return m_data;
return m_data;
}
b3GpuNarrowPhaseInternalData* getInternalData()
b3GpuNarrowPhaseInternalData* getInternalData()
{
return m_data;
return m_data;
}
const struct b3SapAabb& getLocalSpaceAabb(int collidableIndex) const;
};
#endif //B3_GPU_NARROWPHASE_H
#endif //B3_GPU_NARROWPHASE_H

74
src/Bullet3OpenCL/RigidBody/b3GpuNarrowPhaseInternalData.h
View File

@@ -20,57 +20,53 @@
#include "Bullet3Common/shared/b3Int4.h"
#include "Bullet3Common/shared/b3Int2.h"
class b3ConvexUtility;
struct b3GpuNarrowPhaseInternalData
{
b3AlignedObjectArray<b3ConvexUtility*>* m_convexData;
b3AlignedObjectArray<b3ConvexPolyhedronData> m_convexPolyhedra;
b3AlignedObjectArray<b3Vector3> m_uniqueEdges;
b3AlignedObjectArray<b3Vector3> m_convexVertices;
b3AlignedObjectArray<int> m_convexIndices;
b3OpenCLArray<b3ConvexPolyhedronData>* m_convexPolyhedraGPU;
b3OpenCLArray<b3Vector3>* m_uniqueEdgesGPU;
b3OpenCLArray<b3Vector3>* m_convexVerticesGPU;
b3OpenCLArray<int>* m_convexIndicesGPU;
b3OpenCLArray<b3Vector3>* m_worldVertsB1GPU;
b3OpenCLArray<b3Int4>* m_clippingFacesOutGPU;
b3OpenCLArray<b3Vector3>* m_worldNormalsAGPU;
b3OpenCLArray<b3Vector3>* m_worldVertsA1GPU;
b3OpenCLArray<b3Vector3>* m_worldVertsB2GPU;
b3OpenCLArray<b3Vector3>* m_worldVertsB1GPU;
b3OpenCLArray<b3Int4>* m_clippingFacesOutGPU;
b3OpenCLArray<b3Vector3>* m_worldNormalsAGPU;
b3OpenCLArray<b3Vector3>* m_worldVertsA1GPU;
b3OpenCLArray<b3Vector3>* m_worldVertsB2GPU;
b3AlignedObjectArray<b3GpuChildShape> m_cpuChildShapes;
b3OpenCLArray<b3GpuChildShape>* m_gpuChildShapes;
b3OpenCLArray<b3GpuChildShape>* m_gpuChildShapes;
b3AlignedObjectArray<b3GpuFace> m_convexFaces;
b3OpenCLArray<b3GpuFace>* m_convexFacesGPU;
struct GpuSatCollision* m_gpuSatCollision;
b3OpenCLArray<b3Int4>* m_triangleConvexPairs;
struct GpuSatCollision* m_gpuSatCollision;
b3OpenCLArray<b3Int4>* m_triangleConvexPairs;
b3OpenCLArray<b3Contact4>* m_pBufContactBuffersGPU[2];
int m_currentContactBuffer;
int m_currentContactBuffer;
b3AlignedObjectArray<b3Contact4>* m_pBufContactOutCPU;
b3AlignedObjectArray<b3RigidBodyData>* m_bodyBufferCPU;
b3OpenCLArray<b3RigidBodyData>* m_bodyBufferGPU;
b3AlignedObjectArray<b3InertiaData>* m_inertiaBufferCPU;
b3OpenCLArray<b3InertiaData>* m_inertiaBufferGPU;
b3AlignedObjectArray<b3InertiaData>* m_inertiaBufferCPU;
b3OpenCLArray<b3InertiaData>* m_inertiaBufferGPU;
int m_numAcceleratedShapes;
int m_numAcceleratedRigidBodies;
b3AlignedObjectArray<b3Collidable> m_collidablesCPU;
b3OpenCLArray<b3Collidable>* m_collidablesGPU;
b3AlignedObjectArray<b3Collidable> m_collidablesCPU;
b3OpenCLArray<b3Collidable>* m_collidablesGPU;
b3OpenCLArray<b3SapAabb>* m_localShapeAABBGPU;
b3AlignedObjectArray<b3SapAabb>* m_localShapeAABBCPU;
@@ -78,18 +74,16 @@ struct b3GpuNarrowPhaseInternalData
b3AlignedObjectArray<class b3OptimizedBvh*> m_bvhData;
b3AlignedObjectArray<class b3TriangleIndexVertexArray*> m_meshInterfaces;
b3AlignedObjectArray<b3QuantizedBvhNode> m_treeNodesCPU;
b3AlignedObjectArray<b3BvhSubtreeInfo> m_subTreesCPU;
b3AlignedObjectArray<b3QuantizedBvhNode> m_treeNodesCPU;
b3AlignedObjectArray<b3BvhSubtreeInfo> m_subTreesCPU;
b3AlignedObjectArray<b3BvhInfo> m_bvhInfoCPU;
b3OpenCLArray<b3BvhInfo>* m_bvhInfoGPU;
b3OpenCLArray<b3QuantizedBvhNode>* m_treeNodesGPU;
b3OpenCLArray<b3BvhSubtreeInfo>* m_subTreesGPU;
b3AlignedObjectArray<b3BvhInfo> m_bvhInfoCPU;
b3OpenCLArray<b3BvhInfo>* m_bvhInfoGPU;
b3Config m_config;
b3OpenCLArray<b3QuantizedBvhNode>* m_treeNodesGPU;
b3OpenCLArray<b3BvhSubtreeInfo>* m_subTreesGPU;
b3Config m_config;
};
#endif //B3_GPU_NARROWPHASE_INTERNAL_DATA_H
#endif //B3_GPU_NARROWPHASE_INTERNAL_DATA_H

692
src/Bullet3OpenCL/RigidBody/b3GpuPgsConstraintSolver.cpp
View File

File diff suppressed because it is too large Load Diff

50
src/Bullet3OpenCL/RigidBody/b3GpuPgsConstraintSolver.h
View File

@@ -19,7 +19,6 @@ subject to the following restrictions:
struct b3Contact4;
struct b3ContactPoint;
class b3Dispatcher;
#include "Bullet3Dynamics/ConstraintSolver/b3TypedConstraint.h"
@@ -38,41 +37,40 @@ class b3GpuPgsConstraintSolver
protected:
int m_staticIdx;
struct b3GpuPgsJacobiSolverInternalData* m_gpuData;
protected:
b3AlignedObjectArray<b3GpuSolverBody> m_tmpSolverBodyPool;
b3GpuConstraintArray m_tmpSolverContactConstraintPool;
b3GpuConstraintArray m_tmpSolverNonContactConstraintPool;
b3GpuConstraintArray m_tmpSolverContactFrictionConstraintPool;
b3GpuConstraintArray m_tmpSolverContactRollingFrictionConstraintPool;
protected:
b3AlignedObjectArray<b3GpuSolverBody> m_tmpSolverBodyPool;
b3GpuConstraintArray m_tmpSolverContactConstraintPool;
b3GpuConstraintArray m_tmpSolverNonContactConstraintPool;
b3GpuConstraintArray m_tmpSolverContactFrictionConstraintPool;
b3GpuConstraintArray m_tmpSolverContactRollingFrictionConstraintPool;
b3AlignedObjectArray<unsigned int> m_tmpConstraintSizesPool;
bool m_usePgs;
void averageVelocities();
bool m_usePgs;
void averageVelocities();
int m_maxOverrideNumSolverIterations;
int m_maxOverrideNumSolverIterations;
int m_numSplitImpulseRecoveries;
int m_numSplitImpulseRecoveries;
// int getOrInitSolverBody(int bodyIndex, b3RigidBodyData* bodies,b3InertiaData* inertias);
void initSolverBody(int bodyIndex, b3GpuSolverBody* solverBody, b3RigidBodyData* rb);
// int getOrInitSolverBody(int bodyIndex, b3RigidBodyData* bodies,b3InertiaData* inertias);
void initSolverBody(int bodyIndex, b3GpuSolverBody* solverBody, b3RigidBodyData* rb);
public:
b3GpuPgsConstraintSolver (cl_context ctx, cl_device_id device, cl_command_queue queue,bool usePgs);
virtual~b3GpuPgsConstraintSolver ();
b3GpuPgsConstraintSolver(cl_context ctx, cl_device_id device, cl_command_queue queue, bool usePgs);
virtual ~b3GpuPgsConstraintSolver();
virtual b3Scalar solveGroupCacheFriendlyIterations(b3OpenCLArray<b3GpuGenericConstraint>* gpuConstraints1,int numConstraints,const b3ContactSolverInfo& infoGlobal);
virtual b3Scalar solveGroupCacheFriendlySetup(b3OpenCLArray<b3RigidBodyData>* gpuBodies, b3OpenCLArray<b3InertiaData>* gpuInertias, int numBodies,b3OpenCLArray<b3GpuGenericConstraint>* gpuConstraints,int numConstraints,const b3ContactSolverInfo& infoGlobal);
b3Scalar solveGroupCacheFriendlyFinish(b3OpenCLArray<b3RigidBodyData>* gpuBodies,b3OpenCLArray<b3InertiaData>* gpuInertias,int numBodies,b3OpenCLArray<b3GpuGenericConstraint>* gpuConstraints,int numConstraints,const b3ContactSolverInfo& infoGlobal);
virtual b3Scalar solveGroupCacheFriendlyIterations(b3OpenCLArray<b3GpuGenericConstraint>* gpuConstraints1, int numConstraints, const b3ContactSolverInfo& infoGlobal);
virtual b3Scalar solveGroupCacheFriendlySetup(b3OpenCLArray<b3RigidBodyData>* gpuBodies, b3OpenCLArray<b3InertiaData>* gpuInertias, int numBodies, b3OpenCLArray<b3GpuGenericConstraint>* gpuConstraints, int numConstraints, const b3ContactSolverInfo& infoGlobal);
b3Scalar solveGroupCacheFriendlyFinish(b3OpenCLArray<b3RigidBodyData>* gpuBodies, b3OpenCLArray<b3InertiaData>* gpuInertias, int numBodies, b3OpenCLArray<b3GpuGenericConstraint>* gpuConstraints, int numConstraints, const b3ContactSolverInfo& infoGlobal);
b3Scalar solveGroup(b3OpenCLArray<b3RigidBodyData>* gpuBodies, b3OpenCLArray<b3InertiaData>* gpuInertias, int numBodies, b3OpenCLArray<b3GpuGenericConstraint>* gpuConstraints, int numConstraints, const b3ContactSolverInfo& infoGlobal);
void solveJoints(int numBodies, b3OpenCLArray<b3RigidBodyData>* gpuBodies, b3OpenCLArray<b3InertiaData>* gpuInertias,
int numConstraints, b3OpenCLArray<b3GpuGenericConstraint>* gpuConstraints);
b3Scalar solveGroup(b3OpenCLArray<b3RigidBodyData>* gpuBodies,b3OpenCLArray<b3InertiaData>* gpuInertias, int numBodies,b3OpenCLArray<b3GpuGenericConstraint>* gpuConstraints,int numConstraints,const b3ContactSolverInfo& infoGlobal);
void solveJoints(int numBodies, b3OpenCLArray<b3RigidBodyData>* gpuBodies, b3OpenCLArray<b3InertiaData>* gpuInertias,
int numConstraints, b3OpenCLArray<b3GpuGenericConstraint>* gpuConstraints);
int sortConstraintByBatch3( struct b3BatchConstraint* cs, int numConstraints, int simdWidth , int staticIdx, int numBodies);
void recomputeBatches();
int sortConstraintByBatch3(struct b3BatchConstraint* cs, int numConstraints, int simdWidth, int staticIdx, int numBodies);
void recomputeBatches();
};
#endif //B3_GPU_PGS_CONSTRAINT_SOLVER_H
#endif //B3_GPU_PGS_CONSTRAINT_SOLVER_H

1235
src/Bullet3OpenCL/RigidBody/b3GpuPgsContactSolver.cpp
View File

File diff suppressed because it is too large Load Diff

30
src/Bullet3OpenCL/RigidBody/b3GpuPgsContactSolver.h
View File

@@ -11,33 +11,27 @@
class b3GpuPgsContactSolver
{
protected:
int m_debugOutput;
struct b3GpuBatchingPgsSolverInternalData* m_data;
struct b3GpuBatchingPgsSolverInternalData* m_data;
void batchContacts( b3OpenCLArray<b3Contact4>* contacts, int nContacts, b3OpenCLArray<unsigned int>* n, b3OpenCLArray<unsigned int>* offsets, int staticIdx );
inline int sortConstraintByBatch( b3Contact4* cs, int n, int simdWidth , int staticIdx, int numBodies);
inline int sortConstraintByBatch2( b3Contact4* cs, int n, int simdWidth , int staticIdx, int numBodies);
inline int sortConstraintByBatch3( b3Contact4* cs, int n, int simdWidth , int staticIdx, int numBodies, int* batchSizes);
void batchContacts(b3OpenCLArray<b3Contact4>* contacts, int nContacts, b3OpenCLArray<unsigned int>* n, b3OpenCLArray<unsigned int>* offsets, int staticIdx);
void solveContactConstraintBatchSizes( const b3OpenCLArray<b3RigidBodyData>* bodyBuf, const b3OpenCLArray<b3InertiaData>* shapeBuf,
b3OpenCLArray<b3GpuConstraint4>* constraint, void* additionalData, int n ,int maxNumBatches, int numIterations, const b3AlignedObjectArray<int>* batchSizes);//const b3OpenCLArray<int>* gpuBatchSizes);
inline int sortConstraintByBatch(b3Contact4* cs, int n, int simdWidth, int staticIdx, int numBodies);
inline int sortConstraintByBatch2(b3Contact4* cs, int n, int simdWidth, int staticIdx, int numBodies);
inline int sortConstraintByBatch3(b3Contact4* cs, int n, int simdWidth, int staticIdx, int numBodies, int* batchSizes);
void solveContactConstraint( const b3OpenCLArray<b3RigidBodyData>* bodyBuf, const b3OpenCLArray<b3InertiaData>* shapeBuf,
b3OpenCLArray<b3GpuConstraint4>* constraint, void* additionalData, int n ,int maxNumBatches, int numIterations, const b3AlignedObjectArray<int>* batchSizes);//const b3OpenCLArray<int>* gpuBatchSizes);
void solveContactConstraintBatchSizes(const b3OpenCLArray<b3RigidBodyData>* bodyBuf, const b3OpenCLArray<b3InertiaData>* shapeBuf,
b3OpenCLArray<b3GpuConstraint4>* constraint, void* additionalData, int n, int maxNumBatches, int numIterations, const b3AlignedObjectArray<int>* batchSizes); //const b3OpenCLArray<int>* gpuBatchSizes);
void solveContactConstraint(const b3OpenCLArray<b3RigidBodyData>* bodyBuf, const b3OpenCLArray<b3InertiaData>* shapeBuf,
b3OpenCLArray<b3GpuConstraint4>* constraint, void* additionalData, int n, int maxNumBatches, int numIterations, const b3AlignedObjectArray<int>* batchSizes); //const b3OpenCLArray<int>* gpuBatchSizes);
public:
b3GpuPgsContactSolver(cl_context ctx,cl_device_id device, cl_command_queue q,int pairCapacity);
b3GpuPgsContactSolver(cl_context ctx, cl_device_id device, cl_command_queue q, int pairCapacity);
virtual ~b3GpuPgsContactSolver();
void solveContacts(int numBodies, cl_mem bodyBuf, cl_mem inertiaBuf, int numContacts, cl_mem contactBuf, const struct b3Config& config, int static0Index);
};
#endif //B3_GPU_BATCHING_PGS_SOLVER_H
#endif //B3_GPU_BATCHING_PGS_SOLVER_H

351
src/Bullet3OpenCL/RigidBody/b3GpuRigidBodyPipeline.cpp
View File

@@ -47,7 +47,7 @@ bool gClearPairsOnGpu = true;
#define TEST_OTHER_GPU_SOLVER 1
#ifdef TEST_OTHER_GPU_SOLVER
#include "b3GpuJacobiContactSolver.h"
#endif //TEST_OTHER_GPU_SOLVER
#endif //TEST_OTHER_GPU_SOLVER
#include "Bullet3Collision/NarrowPhaseCollision/shared/b3RigidBodyData.h"
#include "Bullet3Collision/NarrowPhaseCollision/b3Contact4.h"
@@ -59,73 +59,68 @@ bool gClearPairsOnGpu = true;
#include "Bullet3Collision/NarrowPhaseCollision/b3Config.h"
#include "Bullet3OpenCL/Raycast/b3GpuRaycast.h"
#include "Bullet3Dynamics/shared/b3IntegrateTransforms.h"
#include "Bullet3OpenCL/RigidBody/b3GpuNarrowPhaseInternalData.h"
b3GpuRigidBodyPipeline::b3GpuRigidBodyPipeline(cl_context ctx,cl_device_id device, cl_command_queue q,class b3GpuNarrowPhase* narrowphase, class b3GpuBroadphaseInterface* broadphaseSap , struct b3DynamicBvhBroadphase* broadphaseDbvt, const b3Config& config)
b3GpuRigidBodyPipeline::b3GpuRigidBodyPipeline(cl_context ctx, cl_device_id device, cl_command_queue q, class b3GpuNarrowPhase* narrowphase, class b3GpuBroadphaseInterface* broadphaseSap, struct b3DynamicBvhBroadphase* broadphaseDbvt, const b3Config& config)
{
m_data = new b3GpuRigidBodyPipelineInternalData;
m_data->m_constraintUid=0;
m_data->m_constraintUid = 0;
m_data->m_config = config;
m_data->m_context = ctx;
m_data->m_device = device;
m_data->m_queue = q;
m_data->m_solver = new b3PgsJacobiSolver(true);//new b3PgsJacobiSolver(true);
m_data->m_gpuSolver = new b3GpuPgsConstraintSolver(ctx,device,q,true);//new b3PgsJacobiSolver(true);
m_data->m_allAabbsGPU = new b3OpenCLArray<b3SapAabb>(ctx,q,config.m_maxConvexBodies);
m_data->m_overlappingPairsGPU = new b3OpenCLArray<b3BroadphasePair>(ctx,q,config.m_maxBroadphasePairs);
m_data->m_solver = new b3PgsJacobiSolver(true); //new b3PgsJacobiSolver(true);
m_data->m_gpuSolver = new b3GpuPgsConstraintSolver(ctx, device, q, true); //new b3PgsJacobiSolver(true);
m_data->m_gpuConstraints = new b3OpenCLArray<b3GpuGenericConstraint>(ctx,q);
m_data->m_allAabbsGPU = new b3OpenCLArray<b3SapAabb>(ctx, q, config.m_maxConvexBodies);
m_data->m_overlappingPairsGPU = new b3OpenCLArray<b3BroadphasePair>(ctx, q, config.m_maxBroadphasePairs);
m_data->m_gpuConstraints = new b3OpenCLArray<b3GpuGenericConstraint>(ctx, q);
#ifdef TEST_OTHER_GPU_SOLVER
m_data->m_solver3 = new b3GpuJacobiContactSolver(ctx,device,q,config.m_maxBroadphasePairs);
#endif // TEST_OTHER_GPU_SOLVER
m_data->m_solver2 = new b3GpuPgsContactSolver(ctx,device,q,config.m_maxBroadphasePairs);
m_data->m_solver3 = new b3GpuJacobiContactSolver(ctx, device, q, config.m_maxBroadphasePairs);
#endif // TEST_OTHER_GPU_SOLVER
m_data->m_raycaster = new b3GpuRaycast(ctx,device,q);
m_data->m_solver2 = new b3GpuPgsContactSolver(ctx, device, q, config.m_maxBroadphasePairs);
m_data->m_raycaster = new b3GpuRaycast(ctx, device, q);
m_data->m_broadphaseDbvt = broadphaseDbvt;
m_data->m_broadphaseSap = broadphaseSap;
m_data->m_narrowphase = narrowphase;
m_data->m_gravity.setValue(0.f,-9.8f,0.f);
m_data->m_gravity.setValue(0.f, -9.8f, 0.f);
cl_int errNum=0;
cl_int errNum = 0;
{
cl_program prog = b3OpenCLUtils::compileCLProgramFromString(m_data->m_context,m_data->m_device,integrateKernelCL,&errNum,"",B3_RIGIDBODY_INTEGRATE_PATH);
b3Assert(errNum==CL_SUCCESS);
m_data->m_integrateTransformsKernel = b3OpenCLUtils::compileCLKernelFromString(m_data->m_context, m_data->m_device,integrateKernelCL, "integrateTransformsKernel",&errNum,prog);
b3Assert(errNum==CL_SUCCESS);
cl_program prog = b3OpenCLUtils::compileCLProgramFromString(m_data->m_context, m_data->m_device, integrateKernelCL, &errNum, "", B3_RIGIDBODY_INTEGRATE_PATH);
b3Assert(errNum == CL_SUCCESS);
m_data->m_integrateTransformsKernel = b3OpenCLUtils::compileCLKernelFromString(m_data->m_context, m_data->m_device, integrateKernelCL, "integrateTransformsKernel", &errNum, prog);
b3Assert(errNum == CL_SUCCESS);
clReleaseProgram(prog);
}
{
cl_program prog = b3OpenCLUtils::compileCLProgramFromString(m_data->m_context,m_data->m_device,updateAabbsKernelCL,&errNum,"",B3_RIGIDBODY_UPDATEAABB_PATH);
b3Assert(errNum==CL_SUCCESS);
m_data->m_updateAabbsKernel = b3OpenCLUtils::compileCLKernelFromString(m_data->m_context, m_data->m_device,updateAabbsKernelCL, "initializeGpuAabbsFull",&errNum,prog);
b3Assert(errNum==CL_SUCCESS);
cl_program prog = b3OpenCLUtils::compileCLProgramFromString(m_data->m_context, m_data->m_device, updateAabbsKernelCL, &errNum, "", B3_RIGIDBODY_UPDATEAABB_PATH);
b3Assert(errNum == CL_SUCCESS);
m_data->m_updateAabbsKernel = b3OpenCLUtils::compileCLKernelFromString(m_data->m_context, m_data->m_device, updateAabbsKernelCL, "initializeGpuAabbsFull", &errNum, prog);
b3Assert(errNum == CL_SUCCESS);
m_data->m_clearOverlappingPairsKernel = b3OpenCLUtils::compileCLKernelFromString(m_data->m_context, m_data->m_device,updateAabbsKernelCL, "clearOverlappingPairsKernel",&errNum,prog);
b3Assert(errNum==CL_SUCCESS);
m_data->m_clearOverlappingPairsKernel = b3OpenCLUtils::compileCLKernelFromString(m_data->m_context, m_data->m_device, updateAabbsKernelCL, "clearOverlappingPairsKernel", &errNum, prog);
b3Assert(errNum == CL_SUCCESS);
clReleaseProgram(prog);
}
}
b3GpuRigidBodyPipeline::~b3GpuRigidBodyPipeline()
{
if (m_data->m_integrateTransformsKernel)
clReleaseKernel(m_data->m_integrateTransformsKernel);
if (m_data->m_updateAabbsKernel)
clReleaseKernel(m_data->m_updateAabbsKernel);
if (m_data->m_clearOverlappingPairsKernel)
clReleaseKernel(m_data->m_clearOverlappingPairsKernel);
delete m_data->m_raycaster;
@@ -136,15 +131,14 @@ b3GpuRigidBodyPipeline::~b3GpuRigidBodyPipeline()
#ifdef TEST_OTHER_GPU_SOLVER
delete m_data->m_solver3;
#endif //TEST_OTHER_GPU_SOLVER
#endif //TEST_OTHER_GPU_SOLVER
delete m_data->m_solver2;
delete m_data;
}
void b3GpuRigidBodyPipeline::reset()
void b3GpuRigidBodyPipeline::reset()
{
m_data->m_gpuConstraints->resize(0);
m_data->m_cpuConstraints.resize(0);
@@ -152,30 +146,28 @@ void b3GpuRigidBodyPipeline::reset()
m_data->m_allAabbsCPU.resize(0);
}
void b3GpuRigidBodyPipeline::addConstraint(b3TypedConstraint* constraint)
void b3GpuRigidBodyPipeline::addConstraint(b3TypedConstraint* constraint)
{
m_data->m_joints.push_back(constraint);
}
void b3GpuRigidBodyPipeline::removeConstraint(b3TypedConstraint* constraint)
void b3GpuRigidBodyPipeline::removeConstraint(b3TypedConstraint* constraint)
{
m_data->m_joints.remove(constraint);
}
void b3GpuRigidBodyPipeline::removeConstraintByUid(int uid)
void b3GpuRigidBodyPipeline::removeConstraintByUid(int uid)
{
m_data->m_gpuSolver->recomputeBatches();
//slow linear search
m_data->m_gpuConstraints->copyToHost(m_data->m_cpuConstraints);
//remove
for (int i=0;i<m_data->m_cpuConstraints.size();i++)
for (int i = 0; i < m_data->m_cpuConstraints.size(); i++)
{
if (m_data->m_cpuConstraints[i].m_uid == uid)
{
//m_data->m_cpuConstraints.remove(m_data->m_cpuConstraints[i]);
m_data->m_cpuConstraints.swap(i,m_data->m_cpuConstraints.size()-1);
m_data->m_cpuConstraints.swap(i, m_data->m_cpuConstraints.size() - 1);
m_data->m_cpuConstraints.pop_back();
break;
@@ -185,13 +177,13 @@ void b3GpuRigidBodyPipeline::removeConstraintByUid(int uid)
if (m_data->m_cpuConstraints.size())
{
m_data->m_gpuConstraints->copyFromHost(m_data->m_cpuConstraints);
} else
}
else
{
m_data->m_gpuConstraints->resize(0);
}
}
int b3GpuRigidBodyPipeline::createPoint2PointConstraint(int bodyA, int bodyB, const float* pivotInA, const float* pivotInB,float breakingThreshold)
int b3GpuRigidBodyPipeline::createPoint2PointConstraint(int bodyA, int bodyB, const float* pivotInA, const float* pivotInB, float breakingThreshold)
{
m_data->m_gpuSolver->recomputeBatches();
b3GpuGenericConstraint c;
@@ -200,14 +192,14 @@ int b3GpuRigidBodyPipeline::createPoint2PointConstraint(int bodyA, int bodyB, co
c.m_flags = B3_CONSTRAINT_FLAG_ENABLED;
c.m_rbA = bodyA;
c.m_rbB = bodyB;
c.m_pivotInA.setValue(pivotInA[0],pivotInA[1],pivotInA[2]);
c.m_pivotInB.setValue(pivotInB[0],pivotInB[1],pivotInB[2]);
c.m_pivotInA.setValue(pivotInA[0], pivotInA[1], pivotInA[2]);
c.m_pivotInB.setValue(pivotInB[0], pivotInB[1], pivotInB[2]);
c.m_breakingImpulseThreshold = breakingThreshold;
c.m_constraintType = B3_GPU_POINT2POINT_CONSTRAINT_TYPE;
m_data->m_cpuConstraints.push_back(c);
return c.m_uid;
}
int b3GpuRigidBodyPipeline::createFixedConstraint(int bodyA, int bodyB, const float* pivotInA, const float* pivotInB, const float* relTargetAB,float breakingThreshold)
int b3GpuRigidBodyPipeline::createFixedConstraint(int bodyA, int bodyB, const float* pivotInA, const float* pivotInB, const float* relTargetAB, float breakingThreshold)
{
m_data->m_gpuSolver->recomputeBatches();
b3GpuGenericConstraint c;
@@ -216,9 +208,9 @@ int b3GpuRigidBodyPipeline::createFixedConstraint(int bodyA, int bodyB, const fl
c.m_flags = B3_CONSTRAINT_FLAG_ENABLED;
c.m_rbA = bodyA;
c.m_rbB = bodyB;
c.m_pivotInA.setValue(pivotInA[0],pivotInA[1],pivotInA[2]);
c.m_pivotInB.setValue(pivotInB[0],pivotInB[1],pivotInB[2]);
c.m_relTargetAB.setValue(relTargetAB[0],relTargetAB[1],relTargetAB[2],relTargetAB[3]);
c.m_pivotInA.setValue(pivotInA[0], pivotInA[1], pivotInA[2]);
c.m_pivotInB.setValue(pivotInB[0], pivotInB[1], pivotInB[2]);
c.m_relTargetAB.setValue(relTargetAB[0], relTargetAB[1], relTargetAB[2], relTargetAB[3]);
c.m_breakingImpulseThreshold = breakingThreshold;
c.m_constraintType = B3_GPU_FIXED_CONSTRAINT_TYPE;
@@ -226,31 +218,28 @@ int b3GpuRigidBodyPipeline::createFixedConstraint(int bodyA, int bodyB, const fl
return c.m_uid;
}
void b3GpuRigidBodyPipeline::stepSimulation(float deltaTime)
void b3GpuRigidBodyPipeline::stepSimulation(float deltaTime)
{
//update worldspace AABBs from local AABB/worldtransform
{
B3_PROFILE("setupGpuAabbs");
setupGpuAabbsFull();
}
int numPairs =0;
int numPairs = 0;
//compute overlapping pairs
{
if (gUseDbvt)
{
{
B3_PROFILE("setAabb");
m_data->m_allAabbsGPU->copyToHost(m_data->m_allAabbsCPU);
for (int i=0;i<m_data->m_allAabbsCPU.size();i++)
for (int i = 0; i < m_data->m_allAabbsCPU.size(); i++)
{
b3Vector3 aabbMin=b3MakeVector3(m_data->m_allAabbsCPU[i].m_min[0],m_data->m_allAabbsCPU[i].m_min[1],m_data->m_allAabbsCPU[i].m_min[2]);
b3Vector3 aabbMax=b3MakeVector3(m_data->m_allAabbsCPU[i].m_max[0],m_data->m_allAabbsCPU[i].m_max[1],m_data->m_allAabbsCPU[i].m_max[2]);
m_data->m_broadphaseDbvt->setAabb(i,aabbMin,aabbMax,0);
b3Vector3 aabbMin = b3MakeVector3(m_data->m_allAabbsCPU[i].m_min[0], m_data->m_allAabbsCPU[i].m_min[1], m_data->m_allAabbsCPU[i].m_min[2]);
b3Vector3 aabbMax = b3MakeVector3(m_data->m_allAabbsCPU[i].m_max[0], m_data->m_allAabbsCPU[i].m_max[1], m_data->m_allAabbsCPU[i].m_max[2]);
m_data->m_broadphaseDbvt->setAabb(i, aabbMin, aabbMax, 0);
}
}
@@ -259,13 +248,14 @@ void b3GpuRigidBodyPipeline::stepSimulation(float deltaTime)
m_data->m_broadphaseDbvt->calculateOverlappingPairs();
}
numPairs = m_data->m_broadphaseDbvt->getOverlappingPairCache()->getNumOverlappingPairs();
} else
}
else
{
if (gUseCalculateOverlappingPairsHost)
{
m_data->m_broadphaseSap->calculateOverlappingPairsHost(m_data->m_config.m_maxBroadphasePairs);
} else
}
else
{
m_data->m_broadphaseSap->calculateOverlappingPairs(m_data->m_config.m_maxBroadphasePairs);
}
@@ -274,24 +264,24 @@ void b3GpuRigidBodyPipeline::stepSimulation(float deltaTime)
}
//compute contact points
// printf("numPairs=%d\n",numPairs);
int numContacts = 0;
// printf("numPairs=%d\n",numPairs);
int numContacts = 0;
int numBodies = m_data->m_narrowphase->getNumRigidBodies();
if (numPairs)
{
cl_mem pairs =0;
cl_mem aabbsWS =0;
cl_mem pairs = 0;
cl_mem aabbsWS = 0;
if (gUseDbvt)
{
B3_PROFILE("m_overlappingPairsGPU->copyFromHost");
m_data->m_overlappingPairsGPU->copyFromHost(m_data->m_broadphaseDbvt->getOverlappingPairCache()->getOverlappingPairArray());
pairs = m_data->m_overlappingPairsGPU->getBufferCL();
aabbsWS = m_data->m_allAabbsGPU->getBufferCL();
} else
}
else
{
pairs = m_data->m_broadphaseSap->getOverlappingPairBuffer();
aabbsWS = m_data->m_broadphaseSap->getAabbBufferWS();
@@ -302,31 +292,27 @@ void b3GpuRigidBodyPipeline::stepSimulation(float deltaTime)
//mark the contacts for each pair as 'unused'
if (numPairs)
{
b3OpenCLArray<b3BroadphasePair> gpuPairs(this->m_data->m_context,m_data->m_queue);
gpuPairs.setFromOpenCLBuffer(pairs,numPairs);
b3OpenCLArray<b3BroadphasePair> gpuPairs(this->m_data->m_context, m_data->m_queue);
gpuPairs.setFromOpenCLBuffer(pairs, numPairs);
if (gClearPairsOnGpu)
{
//b3AlignedObjectArray<b3BroadphasePair> hostPairs;//just for debugging
//gpuPairs.copyToHost(hostPairs);
b3LauncherCL launcher(m_data->m_queue,m_data->m_clearOverlappingPairsKernel,"clearOverlappingPairsKernel");
b3LauncherCL launcher(m_data->m_queue, m_data->m_clearOverlappingPairsKernel, "clearOverlappingPairsKernel");
launcher.setBuffer(pairs);
launcher.setConst(numPairs);
launcher.launch1D(numPairs);
//gpuPairs.copyToHost(hostPairs);
} else
}
else
{
b3AlignedObjectArray<b3BroadphasePair> hostPairs;
gpuPairs.copyToHost(hostPairs);
for (int i=0;i<hostPairs.size();i++)
for (int i = 0; i < hostPairs.size(); i++)
{
hostPairs[i].z = 0xffffffff;
}
@@ -335,7 +321,7 @@ void b3GpuRigidBodyPipeline::stepSimulation(float deltaTime)
}
}
m_data->m_narrowphase->computeContacts(pairs,numPairs,aabbsWS,numBodies);
m_data->m_narrowphase->computeContacts(pairs, numPairs, aabbsWS, numBodies);
numContacts = m_data->m_narrowphase->getNumContactsGpu();
if (gUseDbvt)
@@ -347,56 +333,54 @@ void b3GpuRigidBodyPipeline::stepSimulation(float deltaTime)
if (gDumpContactStats && numContacts)
{
m_data->m_narrowphase->getContactsGpu();
printf("numContacts = %d\n", numContacts);
int totalPoints = 0;
int totalPoints = 0;
const b3Contact4* contacts = m_data->m_narrowphase->getContactsCPU();
for (int i=0;i<numContacts;i++)
for (int i = 0; i < numContacts; i++)
{
totalPoints += contacts->getNPoints();
}
printf("totalPoints=%d\n",totalPoints);
printf("totalPoints=%d\n", totalPoints);
}
}
//convert contact points to contact constraints
//solve constraints
b3OpenCLArray<b3RigidBodyData> gpuBodies(m_data->m_context,m_data->m_queue,0,true);
gpuBodies.setFromOpenCLBuffer(m_data->m_narrowphase->getBodiesGpu(),m_data->m_narrowphase->getNumRigidBodies());
b3OpenCLArray<b3InertiaData> gpuInertias(m_data->m_context,m_data->m_queue,0,true);
gpuInertias.setFromOpenCLBuffer(m_data->m_narrowphase->getBodyInertiasGpu(),m_data->m_narrowphase->getNumRigidBodies());
b3OpenCLArray<b3Contact4> gpuContacts(m_data->m_context,m_data->m_queue,0,true);
gpuContacts.setFromOpenCLBuffer(m_data->m_narrowphase->getContactsGpu(),m_data->m_narrowphase->getNumContactsGpu());
b3OpenCLArray<b3RigidBodyData> gpuBodies(m_data->m_context, m_data->m_queue, 0, true);
gpuBodies.setFromOpenCLBuffer(m_data->m_narrowphase->getBodiesGpu(), m_data->m_narrowphase->getNumRigidBodies());
b3OpenCLArray<b3InertiaData> gpuInertias(m_data->m_context, m_data->m_queue, 0, true);
gpuInertias.setFromOpenCLBuffer(m_data->m_narrowphase->getBodyInertiasGpu(), m_data->m_narrowphase->getNumRigidBodies());
b3OpenCLArray<b3Contact4> gpuContacts(m_data->m_context, m_data->m_queue, 0, true);
gpuContacts.setFromOpenCLBuffer(m_data->m_narrowphase->getContactsGpu(), m_data->m_narrowphase->getNumContactsGpu());
int numJoints = m_data->m_joints.size() ? m_data->m_joints.size() : m_data->m_cpuConstraints.size();
int numJoints = m_data->m_joints.size() ? m_data->m_joints.size() : m_data->m_cpuConstraints.size();
if (useBullet2CpuSolver && numJoints)
{
// b3AlignedObjectArray<b3Contact4> hostContacts;
// b3AlignedObjectArray<b3Contact4> hostContacts;
//gpuContacts.copyToHost(hostContacts);
{
bool useGpu = m_data->m_joints.size()==0;
bool useGpu = m_data->m_joints.size() == 0;
// b3Contact4* contacts = numContacts? &hostContacts[0]: 0;
// b3Contact4* contacts = numContacts? &hostContacts[0]: 0;
//m_data->m_solver->solveContacts(m_data->m_narrowphase->getNumBodiesGpu(),&hostBodies[0],&hostInertias[0],numContacts,contacts,numJoints, joints);
if (useGpu)
{
m_data->m_gpuSolver->solveJoints(m_data->m_narrowphase->getNumRigidBodies(),&gpuBodies,&gpuInertias,numJoints, m_data->m_gpuConstraints);
} else
m_data->m_gpuSolver->solveJoints(m_data->m_narrowphase->getNumRigidBodies(), &gpuBodies, &gpuInertias, numJoints, m_data->m_gpuConstraints);
}
else
{
b3AlignedObjectArray<b3RigidBodyData> hostBodies;
gpuBodies.copyToHost(hostBodies);
b3AlignedObjectArray<b3InertiaData> hostInertias;
gpuInertias.copyToHost(hostInertias);
b3TypedConstraint** joints = numJoints? &m_data->m_joints[0] : 0;
m_data->m_solver->solveContacts(m_data->m_narrowphase->getNumRigidBodies(),&hostBodies[0],&hostInertias[0],0,0,numJoints, joints);
b3TypedConstraint** joints = numJoints ? &m_data->m_joints[0] : 0;
m_data->m_solver->solveContacts(m_data->m_narrowphase->getNumRigidBodies(), &hostBodies[0], &hostInertias[0], 0, 0, numJoints, joints);
gpuBodies.copyFromHost(hostBodies);
}
}
@@ -404,22 +388,20 @@ void b3GpuRigidBodyPipeline::stepSimulation(float deltaTime)
if (numContacts)
{
#ifdef TEST_OTHER_GPU_SOLVER
if (gUseJacobi)
{
bool useGpu = true;
if (useGpu)
{
bool forceHost = false;
if (forceHost)
{
b3AlignedObjectArray<b3RigidBodyData> hostBodies;
b3AlignedObjectArray<b3InertiaData> hostInertias;
b3AlignedObjectArray<b3Contact4> hostContacts;
{
B3_PROFILE("copyToHost");
gpuBodies.copyToHost(hostBodies);
@@ -429,25 +411,24 @@ void b3GpuRigidBodyPipeline::stepSimulation(float deltaTime)
{
b3JacobiSolverInfo solverInfo;
m_data->m_solver3->solveGroupHost(&hostBodies[0], &hostInertias[0], hostBodies.size(),&hostContacts[0],hostContacts.size(),solverInfo);
m_data->m_solver3->solveGroupHost(&hostBodies[0], &hostInertias[0], hostBodies.size(), &hostContacts[0], hostContacts.size(), solverInfo);
}
{
B3_PROFILE("copyFromHost");
gpuBodies.copyFromHost(hostBodies);
}
} else
}
else
{
int static0Index = m_data->m_narrowphase->getStatic0Index();
b3JacobiSolverInfo solverInfo;
//m_data->m_solver3->solveContacts( >solveGroup(&gpuBodies, &gpuInertias, &gpuContacts,solverInfo);
//m_data->m_solver3->solveContacts(m_data->m_narrowphase->getNumBodiesGpu(),&hostBodies[0],&hostInertias[0],numContacts,&hostContacts[0]);
m_data->m_solver3->solveContacts(numBodies, gpuBodies.getBufferCL(),gpuInertias.getBufferCL(),numContacts, gpuContacts.getBufferCL(),m_data->m_config, static0Index);
m_data->m_solver3->solveContacts(numBodies, gpuBodies.getBufferCL(), gpuInertias.getBufferCL(), numContacts, gpuContacts.getBufferCL(), m_data->m_config, static0Index);
}
} else
}
else
{
b3AlignedObjectArray<b3RigidBodyData> hostBodies;
gpuBodies.copyToHost(hostBodies);
@@ -460,17 +441,15 @@ void b3GpuRigidBodyPipeline::stepSimulation(float deltaTime)
}
gpuBodies.copyFromHost(hostBodies);
}
} else
#endif //TEST_OTHER_GPU_SOLVER
}
else
#endif //TEST_OTHER_GPU_SOLVER
{
int static0Index = m_data->m_narrowphase->getStatic0Index();
m_data->m_solver2->solveContacts(numBodies, gpuBodies.getBufferCL(),gpuInertias.getBufferCL(),numContacts, gpuContacts.getBufferCL(),m_data->m_config, static0Index);
m_data->m_solver2->solveContacts(numBodies, gpuBodies.getBufferCL(), gpuInertias.getBufferCL(), numContacts, gpuContacts.getBufferCL(), m_data->m_config, static0Index);
//m_data->m_solver4->solveContacts(m_data->m_narrowphase->getNumBodiesGpu(), gpuBodies.getBufferCL(), gpuInertias.getBufferCL(), numContacts, gpuContacts.getBufferCL());
/*m_data->m_solver3->solveContactConstraintHost(
(b3OpenCLArray<RigidBodyBase::Body>*)&gpuBodies,
(b3OpenCLArray<RigidBodyBase::Inertia>*)&gpuInertias,
@@ -481,11 +460,9 @@ void b3GpuRigidBodyPipeline::stepSimulation(float deltaTime)
}
integrate(deltaTime);
}
void b3GpuRigidBodyPipeline::integrate(float timeStep)
void b3GpuRigidBodyPipeline::integrate(float timeStep)
{
//integrate
int numBodies = m_data->m_narrowphase->getNumRigidBodies();
@@ -493,24 +470,25 @@ void b3GpuRigidBodyPipeline::integrate(float timeStep)
if (gIntegrateOnCpu)
{
if(numBodies)
if (numBodies)
{
b3GpuNarrowPhaseInternalData* npData = m_data->m_narrowphase->getInternalData();
b3GpuNarrowPhaseInternalData* npData = m_data->m_narrowphase->getInternalData();
npData->m_bodyBufferGPU->copyToHost(*npData->m_bodyBufferCPU);
b3RigidBodyData_t* bodies = &npData->m_bodyBufferCPU->at(0);
for (int nodeID=0;nodeID<numBodies;nodeID++)
for (int nodeID = 0; nodeID < numBodies; nodeID++)
{
integrateSingleTransform( bodies,nodeID, timeStep, angularDamp, m_data->m_gravity);
integrateSingleTransform(bodies, nodeID, timeStep, angularDamp, m_data->m_gravity);
}
npData->m_bodyBufferGPU->copyFromHost(*npData->m_bodyBufferCPU);
}
} else
}
else
{
b3LauncherCL launcher(m_data->m_queue,m_data->m_integrateTransformsKernel,"m_integrateTransformsKernel");
b3LauncherCL launcher(m_data->m_queue, m_data->m_integrateTransformsKernel, "m_integrateTransformsKernel");
launcher.setBuffer(m_data->m_narrowphase->getBodiesGpu());
launcher.setConst(numBodies);
launcher.setConst(timeStep);
launcher.setConst(angularDamp);
@@ -519,12 +497,9 @@ void b3GpuRigidBodyPipeline::integrate(float timeStep)
}
}
void b3GpuRigidBodyPipeline::setupGpuAabbsFull()
void b3GpuRigidBodyPipeline::setupGpuAabbsFull()
{
cl_int ciErrNum=0;
cl_int ciErrNum = 0;
int numBodies = m_data->m_narrowphase->getNumRigidBodies();
if (!numBodies)
@@ -532,34 +507,35 @@ void b3GpuRigidBodyPipeline::setupGpuAabbsFull()
if (gCalcWorldSpaceAabbOnCpu)
{
if (numBodies)
{
if (gUseDbvt)
{
m_data->m_allAabbsCPU.resize(numBodies);
m_data->m_narrowphase->readbackAllBodiesToCpu();
for (int i=0;i<numBodies;i++)
for (int i = 0; i < numBodies; i++)
{
b3ComputeWorldAabb( i, m_data->m_narrowphase->getBodiesCpu(), m_data->m_narrowphase->getCollidablesCpu(), m_data->m_narrowphase->getLocalSpaceAabbsCpu(),&m_data->m_allAabbsCPU[0]);
b3ComputeWorldAabb(i, m_data->m_narrowphase->getBodiesCpu(), m_data->m_narrowphase->getCollidablesCpu(), m_data->m_narrowphase->getLocalSpaceAabbsCpu(), &m_data->m_allAabbsCPU[0]);
}
m_data->m_allAabbsGPU->copyFromHost(m_data->m_allAabbsCPU);
} else
}
else
{
m_data->m_broadphaseSap->getAllAabbsCPU().resize(numBodies);
m_data->m_narrowphase->readbackAllBodiesToCpu();
for (int i=0;i<numBodies;i++)
for (int i = 0; i < numBodies; i++)
{
b3ComputeWorldAabb( i, m_data->m_narrowphase->getBodiesCpu(), m_data->m_narrowphase->getCollidablesCpu(), m_data->m_narrowphase->getLocalSpaceAabbsCpu(),&m_data->m_broadphaseSap->getAllAabbsCPU()[0]);
b3ComputeWorldAabb(i, m_data->m_narrowphase->getBodiesCpu(), m_data->m_narrowphase->getCollidablesCpu(), m_data->m_narrowphase->getLocalSpaceAabbsCpu(), &m_data->m_broadphaseSap->getAllAabbsCPU()[0]);
}
m_data->m_broadphaseSap->getAllAabbsGPU().copyFromHost(m_data->m_broadphaseSap->getAllAabbsCPU());
//m_data->m_broadphaseSap->writeAabbsToGpu();
}
}
} else
}
else
{
//__kernel void initializeGpuAabbsFull( const int numNodes, __global Body* gBodies,__global Collidable* collidables, __global b3AABBCL* plocalShapeAABB, __global b3AABBCL* pAABB)
b3LauncherCL launcher(m_data->m_queue,m_data->m_updateAabbsKernel,"m_updateAabbsKernel");
b3LauncherCL launcher(m_data->m_queue, m_data->m_updateAabbsKernel, "m_updateAabbsKernel");
launcher.setConst(numBodies);
cl_mem bodies = m_data->m_narrowphase->getBodiesGpu();
launcher.setBuffer(bodies);
@@ -568,17 +544,18 @@ void b3GpuRigidBodyPipeline::setupGpuAabbsFull()
cl_mem localAabbs = m_data->m_narrowphase->getAabbLocalSpaceBufferGpu();
launcher.setBuffer(localAabbs);
cl_mem worldAabbs =0;
cl_mem worldAabbs = 0;
if (gUseDbvt)
{
worldAabbs = m_data->m_allAabbsGPU->getBufferCL();
} else
}
else
{
worldAabbs = m_data->m_broadphaseSap->getAabbBufferWS();
}
launcher.setBuffer(worldAabbs);
launcher.launch1D(numBodies);
oclCHECKERROR(ciErrNum, CL_SUCCESS);
}
@@ -595,78 +572,68 @@ void b3GpuRigidBodyPipeline::setupGpuAabbsFull()
};
*/
}
cl_mem b3GpuRigidBodyPipeline::getBodyBuffer()
cl_mem b3GpuRigidBodyPipeline::getBodyBuffer()
{
return m_data->m_narrowphase->getBodiesGpu();
}
int b3GpuRigidBodyPipeline::getNumBodies() const
int b3GpuRigidBodyPipeline::getNumBodies() const
{
return m_data->m_narrowphase->getNumRigidBodies();
}
void b3GpuRigidBodyPipeline::setGravity(const float* grav)
void b3GpuRigidBodyPipeline::setGravity(const float* grav)
{
m_data->m_gravity.setValue(grav[0],grav[1],grav[2]);
m_data->m_gravity.setValue(grav[0], grav[1], grav[2]);
}
void b3GpuRigidBodyPipeline::copyConstraintsToHost()
void b3GpuRigidBodyPipeline::copyConstraintsToHost()
{
m_data->m_gpuConstraints->copyToHost(m_data->m_cpuConstraints);
}
void b3GpuRigidBodyPipeline::writeAllInstancesToGpu()
void b3GpuRigidBodyPipeline::writeAllInstancesToGpu()
{
m_data->m_allAabbsGPU->copyFromHost(m_data->m_allAabbsCPU);
m_data->m_gpuConstraints->copyFromHost(m_data->m_cpuConstraints);
}
int b3GpuRigidBodyPipeline::registerPhysicsInstance(float mass, const float* position, const float* orientation, int collidableIndex, int userIndex, bool writeInstanceToGpu)
int b3GpuRigidBodyPipeline::registerPhysicsInstance(float mass, const float* position, const float* orientation, int collidableIndex, int userIndex, bool writeInstanceToGpu)
{
b3Vector3 aabbMin=b3MakeVector3(0,0,0),aabbMax=b3MakeVector3(0,0,0);
b3Vector3 aabbMin = b3MakeVector3(0, 0, 0), aabbMax = b3MakeVector3(0, 0, 0);
if (collidableIndex>=0)
if (collidableIndex >= 0)
{
b3SapAabb localAabb = m_data->m_narrowphase->getLocalSpaceAabb(collidableIndex);
b3Vector3 localAabbMin=b3MakeVector3(localAabb.m_min[0],localAabb.m_min[1],localAabb.m_min[2]);
b3Vector3 localAabbMax=b3MakeVector3(localAabb.m_max[0],localAabb.m_max[1],localAabb.m_max[2]);
b3Vector3 localAabbMin = b3MakeVector3(localAabb.m_min[0], localAabb.m_min[1], localAabb.m_min[2]);
b3Vector3 localAabbMax = b3MakeVector3(localAabb.m_max[0], localAabb.m_max[1], localAabb.m_max[2]);
b3Scalar margin = 0.01f;
b3Transform t;
t.setIdentity();
t.setOrigin(b3MakeVector3(position[0],position[1],position[2]));
t.setRotation(b3Quaternion(orientation[0],orientation[1],orientation[2],orientation[3]));
b3TransformAabb(localAabbMin,localAabbMax, margin,t,aabbMin,aabbMax);
} else
t.setOrigin(b3MakeVector3(position[0], position[1], position[2]));
t.setRotation(b3Quaternion(orientation[0], orientation[1], orientation[2], orientation[3]));
b3TransformAabb(localAabbMin, localAabbMax, margin, t, aabbMin, aabbMax);
}
else
{
b3Error("registerPhysicsInstance using invalid collidableIndex\n");
return -1;
}
bool writeToGpu = false;
int bodyIndex = m_data->m_narrowphase->getNumRigidBodies();
bodyIndex = m_data->m_narrowphase->registerRigidBody(collidableIndex,mass,position,orientation,&aabbMin.getX(),&aabbMax.getX(),writeToGpu);
bodyIndex = m_data->m_narrowphase->registerRigidBody(collidableIndex, mass, position, orientation, &aabbMin.getX(), &aabbMax.getX(), writeToGpu);
if (bodyIndex>=0)
if (bodyIndex >= 0)
{
if (gUseDbvt)
{
m_data->m_broadphaseDbvt->createProxy(aabbMin,aabbMax,bodyIndex,0,1,1);
m_data->m_broadphaseDbvt->createProxy(aabbMin, aabbMax, bodyIndex, 0, 1, 1);
b3SapAabb aabb;
for (int i=0;i<3;i++)
for (int i = 0; i < 3; i++)
{
aabb.m_min[i] = aabbMin[i];
aabb.m_max[i] = aabbMax[i];
@@ -677,14 +644,16 @@ int b3GpuRigidBodyPipeline::registerPhysicsInstance(float mass, const float* po
{
m_data->m_allAabbsGPU->copyFromHost(m_data->m_allAabbsCPU);
}
} else
}
else
{
if (mass)
{
m_data->m_broadphaseSap->createProxy(aabbMin,aabbMax,bodyIndex,1,1);//m_dispatcher);
} else
m_data->m_broadphaseSap->createProxy(aabbMin, aabbMax, bodyIndex, 1, 1); //m_dispatcher);
}
else
{
m_data->m_broadphaseSap->createLargeProxy(aabbMin,aabbMax,bodyIndex,1,1);//m_dispatcher);
m_data->m_broadphaseSap->createLargeProxy(aabbMin, aabbMax, bodyIndex, 1, 1); //m_dispatcher);
}
}
}
@@ -699,10 +668,10 @@ int b3GpuRigidBodyPipeline::registerPhysicsInstance(float mass, const float* po
return bodyIndex;
}
void b3GpuRigidBodyPipeline::castRays(const b3AlignedObjectArray<b3RayInfo>& rays, b3AlignedObjectArray<b3RayHit>& hitResults)
void b3GpuRigidBodyPipeline::castRays(const b3AlignedObjectArray<b3RayInfo>& rays, b3AlignedObjectArray<b3RayHit>& hitResults)
{
this->m_data->m_raycaster->castRays(rays,hitResults,
getNumBodies(),this->m_data->m_narrowphase->getBodiesCpu(),
m_data->m_narrowphase->getNumCollidablesGpu(), m_data->m_narrowphase->getCollidablesCpu(),
m_data->m_narrowphase->getInternalData(), m_data->m_broadphaseSap);
this->m_data->m_raycaster->castRays(rays, hitResults,
getNumBodies(), this->m_data->m_narrowphase->getBodiesCpu(),
m_data->m_narrowphase->getNumCollidablesGpu(), m_data->m_narrowphase->getCollidablesCpu(),
m_data->m_narrowphase->getInternalData(), m_data->m_broadphaseSap);
}

42
src/Bullet3OpenCL/RigidBody/b3GpuRigidBodyPipeline.h
View File

@@ -25,50 +25,46 @@ subject to the following restrictions:
class b3GpuRigidBodyPipeline
{
protected:
struct b3GpuRigidBodyPipelineInternalData* m_data;
struct b3GpuRigidBodyPipelineInternalData* m_data;
int allocateCollidable();
public:
b3GpuRigidBodyPipeline(cl_context ctx,cl_device_id device, cl_command_queue q , class b3GpuNarrowPhase* narrowphase, class b3GpuBroadphaseInterface* broadphaseSap, struct b3DynamicBvhBroadphase* broadphaseDbvt, const b3Config& config);
b3GpuRigidBodyPipeline(cl_context ctx, cl_device_id device, cl_command_queue q, class b3GpuNarrowPhase* narrowphase, class b3GpuBroadphaseInterface* broadphaseSap, struct b3DynamicBvhBroadphase* broadphaseDbvt, const b3Config& config);
virtual ~b3GpuRigidBodyPipeline();
void stepSimulation(float deltaTime);
void integrate(float timeStep);
void setupGpuAabbsFull();
void stepSimulation(float deltaTime);
void integrate(float timeStep);
void setupGpuAabbsFull();
int registerConvexPolyhedron(class b3ConvexUtility* convex);
int registerConvexPolyhedron(class b3ConvexUtility* convex);
//int registerConvexPolyhedron(const float* vertices, int strideInBytes, int numVertices, const float* scaling);
//int registerSphereShape(float radius);
//int registerPlaneShape(const b3Vector3& planeNormal, float planeConstant);
//int registerConcaveMesh(b3AlignedObjectArray<b3Vector3>* vertices, b3AlignedObjectArray<int>* indices, const float* scaling);
//int registerCompoundShape(b3AlignedObjectArray<b3GpuChildShape>* childShapes);
int registerPhysicsInstance(float mass, const float* position, const float* orientation, int collisionShapeIndex, int userData, bool writeInstanceToGpu);
int registerPhysicsInstance(float mass, const float* position, const float* orientation, int collisionShapeIndex, int userData, bool writeInstanceToGpu);
//if you passed "writeInstanceToGpu" false in the registerPhysicsInstance method (for performance) you need to call writeAllInstancesToGpu after all instances are registered
void writeAllInstancesToGpu();
void copyConstraintsToHost();
void setGravity(const float* grav);
void writeAllInstancesToGpu();
void copyConstraintsToHost();
void setGravity(const float* grav);
void reset();
int createPoint2PointConstraint(int bodyA, int bodyB, const float* pivotInA, const float* pivotInB,float breakingThreshold);
int createPoint2PointConstraint(int bodyA, int bodyB, const float* pivotInA, const float* pivotInB, float breakingThreshold);
int createFixedConstraint(int bodyA, int bodyB, const float* pivotInA, const float* pivotInB, const float* relTargetAB, float breakingThreshold);
void removeConstraintByUid(int uid);
void addConstraint(class b3TypedConstraint* constraint);
void removeConstraint(b3TypedConstraint* constraint);
void addConstraint(class b3TypedConstraint* constraint);
void removeConstraint(b3TypedConstraint* constraint);
void castRays(const b3AlignedObjectArray<b3RayInfo>& rays, b3AlignedObjectArray<b3RayHit>& hitResults);
void castRays(const b3AlignedObjectArray<b3RayInfo>& rays, b3AlignedObjectArray<b3RayHit>& hitResults);
cl_mem getBodyBuffer();
int getNumBodies() const;
cl_mem getBodyBuffer();
int getNumBodies() const;
};
#endif //B3_GPU_RIGIDBODY_PIPELINE_H
#endif //B3_GPU_RIGIDBODY_PIPELINE_H

39
src/Bullet3OpenCL/RigidBody/b3GpuRigidBodyPipelineInternalData.h
View File

@@ -22,52 +22,47 @@ subject to the following restrictions:
#include "Bullet3OpenCL/ParallelPrimitives/b3OpenCLArray.h"
#include "Bullet3Collision/NarrowPhaseCollision/shared/b3Collidable.h"
#include "Bullet3OpenCL/BroadphaseCollision/b3SapAabb.h"
#include "Bullet3Dynamics/ConstraintSolver/b3TypedConstraint.h"
#include "Bullet3Collision/NarrowPhaseCollision/b3Config.h"
#include "Bullet3Collision/BroadPhaseCollision/b3OverlappingPair.h"
#include "Bullet3OpenCL/RigidBody/b3GpuGenericConstraint.h"
struct b3GpuRigidBodyPipelineInternalData
{
cl_context m_context;
cl_device_id m_device;
cl_command_queue m_queue;
cl_context m_context;
cl_device_id m_device;
cl_command_queue m_queue;
cl_kernel m_integrateTransformsKernel;
cl_kernel m_updateAabbsKernel;
cl_kernel m_clearOverlappingPairsKernel;
cl_kernel m_integrateTransformsKernel;
cl_kernel m_updateAabbsKernel;
cl_kernel m_clearOverlappingPairsKernel;
class b3PgsJacobiSolver* m_solver;
class b3GpuPgsConstraintSolver* m_gpuSolver;
class b3GpuPgsContactSolver* m_solver2;
class b3GpuJacobiContactSolver* m_solver3;
class b3GpuRaycast* m_raycaster;
class b3GpuBroadphaseInterface* m_broadphaseSap;
struct b3DynamicBvhBroadphase* m_broadphaseDbvt;
b3OpenCLArray<b3SapAabb>* m_allAabbsGPU;
b3AlignedObjectArray<b3SapAabb> m_allAabbsCPU;
b3OpenCLArray<b3BroadphasePair>* m_overlappingPairsGPU;
b3OpenCLArray<b3SapAabb>* m_allAabbsGPU;
b3AlignedObjectArray<b3SapAabb> m_allAabbsCPU;
b3OpenCLArray<b3BroadphasePair>* m_overlappingPairsGPU;
b3OpenCLArray<b3GpuGenericConstraint>* m_gpuConstraints;
b3AlignedObjectArray<b3GpuGenericConstraint> m_cpuConstraints;
b3AlignedObjectArray<b3TypedConstraint*> m_joints;
int m_constraintUid;
class b3GpuNarrowPhase* m_narrowphase;
b3Vector3 m_gravity;
int m_constraintUid;
class b3GpuNarrowPhase* m_narrowphase;
b3Vector3 m_gravity;
b3Config m_config;
b3Config m_config;
};
#endif //B3_GPU_RIGIDBODY_PIPELINE_INTERNAL_DATA_H
#endif //B3_GPU_RIGIDBODY_PIPELINE_INTERNAL_DATA_H

118
src/Bullet3OpenCL/RigidBody/b3GpuSolverBody.h
View File

@@ -13,11 +13,9 @@ subject to the following restrictions:
*/
//Originally written by Erwin Coumans
#ifndef B3_GPU_SOLVER_BODY_H
#define B3_GPU_SOLVER_BODY_H
#include "Bullet3Common/b3Vector3.h"
#include "Bullet3Common/b3Matrix3x3.h"
@@ -27,29 +25,27 @@ subject to the following restrictions:
///Until we get other contributions, only use SIMD on Windows, when using Visual Studio 2008 or later, and not double precision
#ifdef B3_USE_SSE
#define USE_SIMD 1
#endif //
#endif //
///The b3SolverBody is an internal datastructure for the constraint solver. Only necessary data is packed to increase cache coherence/performance.
B3_ATTRIBUTE_ALIGNED16 (struct) b3GpuSolverBody
B3_ATTRIBUTE_ALIGNED16(struct)
b3GpuSolverBody
{
B3_DECLARE_ALIGNED_ALLOCATOR();
// b3Transform m_worldTransformUnused;
b3Vector3 m_deltaLinearVelocity;
b3Vector3 m_deltaAngularVelocity;
b3Vector3 m_angularFactor;
b3Vector3 m_linearFactor;
b3Vector3 m_invMass;
b3Vector3 m_pushVelocity;
b3Vector3 m_turnVelocity;
b3Vector3 m_linearVelocity;
b3Vector3 m_angularVelocity;
// b3Transform m_worldTransformUnused;
b3Vector3 m_deltaLinearVelocity;
b3Vector3 m_deltaAngularVelocity;
b3Vector3 m_angularFactor;
b3Vector3 m_linearFactor;
b3Vector3 m_invMass;
b3Vector3 m_pushVelocity;
b3Vector3 m_turnVelocity;
b3Vector3 m_linearVelocity;
b3Vector3 m_angularVelocity;
union
{
void* m_originalBody;
int m_originalBodyIndex;
union {
void* m_originalBody;
int m_originalBodyIndex;
};
int padding[3];
@@ -65,44 +61,41 @@ B3_ATTRIBUTE_ALIGNED16 (struct) b3GpuSolverBody
return m_worldTransform;
}
*/
B3_FORCE_INLINE void getVelocityInLocalPointObsolete(const b3Vector3& rel_pos, b3Vector3& velocity ) const
B3_FORCE_INLINE void getVelocityInLocalPointObsolete(const b3Vector3& rel_pos, b3Vector3& velocity) const
{
if (m_originalBody)
velocity = m_linearVelocity+m_deltaLinearVelocity + (m_angularVelocity+m_deltaAngularVelocity).cross(rel_pos);
velocity = m_linearVelocity + m_deltaLinearVelocity + (m_angularVelocity + m_deltaAngularVelocity).cross(rel_pos);
else
velocity.setValue(0,0,0);
velocity.setValue(0, 0, 0);
}
B3_FORCE_INLINE void getAngularVelocity(b3Vector3& angVel) const
B3_FORCE_INLINE void getAngularVelocity(b3Vector3 & angVel) const
{
if (m_originalBody)
angVel =m_angularVelocity+m_deltaAngularVelocity;
angVel = m_angularVelocity + m_deltaAngularVelocity;
else
angVel.setValue(0,0,0);
angVel.setValue(0, 0, 0);
}
//Optimization for the iterative solver: avoid calculating constant terms involving inertia, normal, relative position
B3_FORCE_INLINE void applyImpulse(const b3Vector3& linearComponent, const b3Vector3& angularComponent,const b3Scalar impulseMagnitude)
B3_FORCE_INLINE void applyImpulse(const b3Vector3& linearComponent, const b3Vector3& angularComponent, const b3Scalar impulseMagnitude)
{
if (m_originalBody)
{
m_deltaLinearVelocity += linearComponent*impulseMagnitude*m_linearFactor;
m_deltaAngularVelocity += angularComponent*(impulseMagnitude*m_angularFactor);
m_deltaLinearVelocity += linearComponent * impulseMagnitude * m_linearFactor;
m_deltaAngularVelocity += angularComponent * (impulseMagnitude * m_angularFactor);
}
}
B3_FORCE_INLINE void internalApplyPushImpulse(const b3Vector3& linearComponent, const b3Vector3& angularComponent,b3Scalar impulseMagnitude)
B3_FORCE_INLINE void internalApplyPushImpulse(const b3Vector3& linearComponent, const b3Vector3& angularComponent, b3Scalar impulseMagnitude)
{
if (m_originalBody)
{
m_pushVelocity += linearComponent*impulseMagnitude*m_linearFactor;
m_turnVelocity += angularComponent*(impulseMagnitude*m_angularFactor);
m_pushVelocity += linearComponent * impulseMagnitude * m_linearFactor;
m_turnVelocity += angularComponent * (impulseMagnitude * m_angularFactor);
}
}
const b3Vector3& getDeltaLinearVelocity() const
{
return m_deltaLinearVelocity;
@@ -113,20 +106,19 @@ B3_ATTRIBUTE_ALIGNED16 (struct) b3GpuSolverBody
return m_deltaAngularVelocity;
}
const b3Vector3& getPushVelocity() const
const b3Vector3& getPushVelocity() const
{
return m_pushVelocity;
}
const b3Vector3& getTurnVelocity() const
const b3Vector3& getTurnVelocity() const
{
return m_turnVelocity;
}
////////////////////////////////////////////////
///some internal methods, don't use them
b3Vector3& internalGetDeltaLinearVelocity()
{
return m_deltaLinearVelocity;
@@ -151,7 +143,7 @@ B3_ATTRIBUTE_ALIGNED16 (struct) b3GpuSolverBody
{
m_invMass = invMass;
}
b3Vector3& internalGetPushVelocity()
{
return m_pushVelocity;
@@ -162,67 +154,57 @@ B3_ATTRIBUTE_ALIGNED16 (struct) b3GpuSolverBody
return m_turnVelocity;
}
B3_FORCE_INLINE void internalGetVelocityInLocalPointObsolete(const b3Vector3& rel_pos, b3Vector3& velocity ) const
B3_FORCE_INLINE void internalGetVelocityInLocalPointObsolete(const b3Vector3& rel_pos, b3Vector3& velocity) const
{
velocity = m_linearVelocity+m_deltaLinearVelocity + (m_angularVelocity+m_deltaAngularVelocity).cross(rel_pos);
velocity = m_linearVelocity + m_deltaLinearVelocity + (m_angularVelocity + m_deltaAngularVelocity).cross(rel_pos);
}
B3_FORCE_INLINE void internalGetAngularVelocity(b3Vector3& angVel) const
B3_FORCE_INLINE void internalGetAngularVelocity(b3Vector3 & angVel) const
{
angVel = m_angularVelocity+m_deltaAngularVelocity;
angVel = m_angularVelocity + m_deltaAngularVelocity;
}
//Optimization for the iterative solver: avoid calculating constant terms involving inertia, normal, relative position
B3_FORCE_INLINE void internalApplyImpulse(const b3Vector3& linearComponent, const b3Vector3& angularComponent,const b3Scalar impulseMagnitude)
B3_FORCE_INLINE void internalApplyImpulse(const b3Vector3& linearComponent, const b3Vector3& angularComponent, const b3Scalar impulseMagnitude)
{
//if (m_originalBody)
{
m_deltaLinearVelocity += linearComponent*impulseMagnitude*m_linearFactor;
m_deltaAngularVelocity += angularComponent*(impulseMagnitude*m_angularFactor);
m_deltaLinearVelocity += linearComponent * impulseMagnitude * m_linearFactor;
m_deltaAngularVelocity += angularComponent * (impulseMagnitude * m_angularFactor);
}
}
void writebackVelocity()
void writebackVelocity()
{
//if (m_originalBody>=0)
{
m_linearVelocity +=m_deltaLinearVelocity;
m_linearVelocity += m_deltaLinearVelocity;
m_angularVelocity += m_deltaAngularVelocity;
//m_originalBody->setCompanionId(-1);
}
}
void writebackVelocityAndTransform(b3Scalar timeStep, b3Scalar splitImpulseTurnErp)
void writebackVelocityAndTransform(b3Scalar timeStep, b3Scalar splitImpulseTurnErp)
{
(void) timeStep;
(void)timeStep;
if (m_originalBody)
{
m_linearVelocity += m_deltaLinearVelocity;
m_angularVelocity += m_deltaAngularVelocity;
//correct the position/orientation based on push/turn recovery
b3Transform newTransform;
if (m_pushVelocity[0]!=0.f || m_pushVelocity[1]!=0 || m_pushVelocity[2]!=0 || m_turnVelocity[0]!=0.f || m_turnVelocity[1]!=0 || m_turnVelocity[2]!=0)
if (m_pushVelocity[0] != 0.f || m_pushVelocity[1] != 0 || m_pushVelocity[2] != 0 || m_turnVelocity[0] != 0.f || m_turnVelocity[1] != 0 || m_turnVelocity[2] != 0)
{
// b3Quaternion orn = m_worldTransform.getRotation();
// b3TransformUtil::integrateTransform(m_worldTransform,m_pushVelocity,m_turnVelocity*splitImpulseTurnErp,timeStep,newTransform);
// m_worldTransform = newTransform;
// b3Quaternion orn = m_worldTransform.getRotation();
// b3TransformUtil::integrateTransform(m_worldTransform,m_pushVelocity,m_turnVelocity*splitImpulseTurnErp,timeStep,newTransform);
// m_worldTransform = newTransform;
}
//m_worldTransform.setRotation(orn);
//m_originalBody->setCompanionId(-1);
}
}
};
#endif //B3_SOLVER_BODY_H
#endif //B3_SOLVER_BODY_H

63
src/Bullet3OpenCL/RigidBody/b3GpuSolverConstraint.h
View File

@@ -13,11 +13,9 @@ subject to the following restrictions:
3. This notice may not be removed or altered from any source distribution.
*/
#ifndef B3_GPU_SOLVER_CONSTRAINT_H
#define B3_GPU_SOLVER_CONSTRAINT_H
#include "Bullet3Common/b3Vector3.h"
#include "Bullet3Common/b3Matrix3x3.h"
//#include "b3JacobianEntry.h"
@@ -25,58 +23,51 @@ subject to the following restrictions:
//#define NO_FRICTION_TANGENTIALS 1
///1D constraint along a normal axis between bodyA and bodyB. It can be combined to solve contact and friction constraints.
B3_ATTRIBUTE_ALIGNED16 (struct) b3GpuSolverConstraint
B3_ATTRIBUTE_ALIGNED16(struct)
b3GpuSolverConstraint
{
B3_DECLARE_ALIGNED_ALLOCATOR();
b3Vector3 m_relpos1CrossNormal;
b3Vector3 m_contactNormal;
b3Vector3 m_relpos1CrossNormal;
b3Vector3 m_contactNormal;
b3Vector3 m_relpos2CrossNormal;
b3Vector3 m_relpos2CrossNormal;
//b3Vector3 m_contactNormal2;//usually m_contactNormal2 == -m_contactNormal
b3Vector3 m_angularComponentA;
b3Vector3 m_angularComponentB;
mutable b3Scalar m_appliedPushImpulse;
mutable b3Scalar m_appliedImpulse;
b3Vector3 m_angularComponentA;
b3Vector3 m_angularComponentB;
mutable b3Scalar m_appliedPushImpulse;
mutable b3Scalar m_appliedImpulse;
int m_padding1;
int m_padding2;
b3Scalar m_friction;
b3Scalar m_jacDiagABInv;
b3Scalar m_rhs;
b3Scalar m_cfm;
b3Scalar m_lowerLimit;
b3Scalar m_upperLimit;
b3Scalar m_rhsPenetration;
union
{
void* m_originalContactPoint;
int m_originalConstraintIndex;
b3Scalar m_unusedPadding4;
b3Scalar m_friction;
b3Scalar m_jacDiagABInv;
b3Scalar m_rhs;
b3Scalar m_cfm;
b3Scalar m_lowerLimit;
b3Scalar m_upperLimit;
b3Scalar m_rhsPenetration;
union {
void* m_originalContactPoint;
int m_originalConstraintIndex;
b3Scalar m_unusedPadding4;
};
int m_overrideNumSolverIterations;
int m_frictionIndex;
int m_overrideNumSolverIterations;
int m_frictionIndex;
int m_solverBodyIdA;
int m_solverBodyIdB;
enum b3SolverConstraintType
enum b3SolverConstraintType
{
B3_SOLVER_CONTACT_1D = 0,
B3_SOLVER_FRICTION_1D
};
};
typedef b3AlignedObjectArray<b3GpuSolverConstraint> b3GpuConstraintArray;
#endif //B3_GPU_SOLVER_CONSTRAINT_H
typedef b3AlignedObjectArray<b3GpuSolverConstraint> b3GpuConstraintArray;
#endif //B3_GPU_SOLVER_CONSTRAINT_H

837
src/Bullet3OpenCL/RigidBody/b3Solver.cpp
View File

File diff suppressed because it is too large Load Diff

124
src/Bullet3OpenCL/RigidBody/b3Solver.h
View File

@@ -13,7 +13,6 @@ subject to the following restrictions:
*/
//Originally written by Takahiro Harada
#ifndef __ADL_SOLVER_H
#define __ADL_SOLVER_H
@@ -29,98 +28,83 @@ subject to the following restrictions:
#include "Bullet3OpenCL/Initialize/b3OpenCLUtils.h"
#define B3NEXTMULTIPLEOF(num, alignment) (((num)/(alignment) + (((num)%(alignment)==0)?0:1))*(alignment))
#define B3NEXTMULTIPLEOF(num, alignment) (((num) / (alignment) + (((num) % (alignment) == 0) ? 0 : 1)) * (alignment))
enum
{
B3_SOLVER_N_SPLIT_X = 8,//16,//4,
B3_SOLVER_N_SPLIT_Y = 4,//16,//4,
B3_SOLVER_N_SPLIT_Z = 8,//,
B3_SOLVER_N_CELLS = B3_SOLVER_N_SPLIT_X*B3_SOLVER_N_SPLIT_Y*B3_SOLVER_N_SPLIT_Z,
B3_SOLVER_N_BATCHES = 8,//4,//8,//4,
B3_SOLVER_N_SPLIT_X = 8, //16,//4,
B3_SOLVER_N_SPLIT_Y = 4, //16,//4,
B3_SOLVER_N_SPLIT_Z = 8, //,
B3_SOLVER_N_CELLS = B3_SOLVER_N_SPLIT_X * B3_SOLVER_N_SPLIT_Y * B3_SOLVER_N_SPLIT_Z,
B3_SOLVER_N_BATCHES = 8, //4,//8,//4,
B3_MAX_NUM_BATCHES = 128,
};
class b3SolverBase
{
public:
public:
struct ConstraintCfg
{
ConstraintCfg(float dt = 0.f) : m_positionDrift(0.005f), m_positionConstraintCoeff(0.2f), m_dt(dt), m_staticIdx(-1) {}
struct ConstraintCfg
{
ConstraintCfg( float dt = 0.f ): m_positionDrift( 0.005f ), m_positionConstraintCoeff( 0.2f ), m_dt(dt), m_staticIdx(-1) {}
float m_positionDrift;
float m_positionConstraintCoeff;
float m_dt;
bool m_enableParallelSolve;
float m_batchCellSize;
int m_staticIdx;
};
float m_positionDrift;
float m_positionConstraintCoeff;
float m_dt;
bool m_enableParallelSolve;
float m_batchCellSize;
int m_staticIdx;
};
};
class b3Solver : public b3SolverBase
{
public:
public:
cl_context m_context;
cl_device_id m_device;
cl_command_queue m_queue;
cl_context m_context;
cl_device_id m_device;
cl_command_queue m_queue;
b3OpenCLArray<unsigned int>* m_numConstraints;
b3OpenCLArray<unsigned int>* m_offsets;
b3OpenCLArray<int> m_batchSizes;
b3OpenCLArray<unsigned int>* m_numConstraints;
b3OpenCLArray<unsigned int>* m_offsets;
b3OpenCLArray<int> m_batchSizes;
int m_nIterations;
cl_kernel m_batchingKernel;
cl_kernel m_batchingKernelNew;
cl_kernel m_solveContactKernel;
cl_kernel m_solveFrictionKernel;
cl_kernel m_contactToConstraintKernel;
cl_kernel m_setSortDataKernel;
cl_kernel m_reorderContactKernel;
cl_kernel m_copyConstraintKernel;
int m_nIterations;
cl_kernel m_batchingKernel;
cl_kernel m_batchingKernelNew;
cl_kernel m_solveContactKernel;
cl_kernel m_solveFrictionKernel;
cl_kernel m_contactToConstraintKernel;
cl_kernel m_setSortDataKernel;
cl_kernel m_reorderContactKernel;
cl_kernel m_copyConstraintKernel;
class b3RadixSort32CL* m_sort32;
class b3BoundSearchCL* m_search;
class b3PrefixScanCL* m_scan;
class b3RadixSort32CL* m_sort32;
class b3BoundSearchCL* m_search;
class b3PrefixScanCL* m_scan;
b3OpenCLArray<b3SortData>* m_sortDataBuffer;
b3OpenCLArray<b3Contact4>* m_contactBuffer2;
b3OpenCLArray<b3SortData>* m_sortDataBuffer;
b3OpenCLArray<b3Contact4>* m_contactBuffer2;
enum
{
DYNAMIC_CONTACT_ALLOCATION_THRESHOLD = 2000000,
};
enum
{
DYNAMIC_CONTACT_ALLOCATION_THRESHOLD = 2000000,
};
b3Solver(cl_context ctx, cl_device_id device, cl_command_queue queue, int pairCapacity);
b3Solver(cl_context ctx, cl_device_id device, cl_command_queue queue, int pairCapacity);
virtual ~b3Solver();
virtual ~b3Solver();
void solveContactConstraint( const b3OpenCLArray<b3RigidBodyData>* bodyBuf, const b3OpenCLArray<b3InertiaData>* inertiaBuf,
b3OpenCLArray<b3GpuConstraint4>* constraint, void* additionalData, int n ,int maxNumBatches);
void solveContactConstraint(const b3OpenCLArray<b3RigidBodyData>* bodyBuf, const b3OpenCLArray<b3InertiaData>* inertiaBuf,
b3OpenCLArray<b3GpuConstraint4>* constraint, void* additionalData, int n, int maxNumBatches);
void solveContactConstraintHost( b3OpenCLArray<b3RigidBodyData>* bodyBuf, b3OpenCLArray<b3InertiaData>* shapeBuf,
b3OpenCLArray<b3GpuConstraint4>* constraint, void* additionalData, int n ,int maxNumBatches, b3AlignedObjectArray<int>* batchSizes);
void solveContactConstraintHost(b3OpenCLArray<b3RigidBodyData>* bodyBuf, b3OpenCLArray<b3InertiaData>* shapeBuf,
b3OpenCLArray<b3GpuConstraint4>* constraint, void* additionalData, int n, int maxNumBatches, b3AlignedObjectArray<int>* batchSizes);
void convertToConstraints(const b3OpenCLArray<b3RigidBodyData>* bodyBuf,
const b3OpenCLArray<b3InertiaData>* shapeBuf,
b3OpenCLArray<b3Contact4>* contactsIn, b3OpenCLArray<b3GpuConstraint4>* contactCOut, void* additionalData,
int nContacts, const ConstraintCfg& cfg);
void convertToConstraints( const b3OpenCLArray<b3RigidBodyData>* bodyBuf,
const b3OpenCLArray<b3InertiaData>* shapeBuf,
b3OpenCLArray<b3Contact4>* contactsIn, b3OpenCLArray<b3GpuConstraint4>* contactCOut, void* additionalData,
int nContacts, const ConstraintCfg& cfg );
void batchContacts( b3OpenCLArray<b3Contact4>* contacts, int nContacts, b3OpenCLArray<unsigned int>* n, b3OpenCLArray<unsigned int>* offsets, int staticIdx );
void batchContacts(b3OpenCLArray<b3Contact4>* contacts, int nContacts, b3OpenCLArray<unsigned int>* n, b3OpenCLArray<unsigned int>* offsets, int staticIdx);
};
#endif //__ADL_SOLVER_H
#endif //__ADL_SOLVER_H

773
src/Bullet3OpenCL/RigidBody/kernels/batchingKernels.h
View File

@@ -1,388 +1,387 @@
//this file is autogenerated using stringify.bat (premake --stringify) in the build folder of this project
static const char* batchingKernelsCL= \
"/*\n"
"Copyright (c) 2012 Advanced Micro Devices, Inc. \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"
"//Originally written by Takahiro Harada\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"
"#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"
"typedef unsigned int u32;\n"
"typedef unsigned short u16;\n"
"typedef unsigned char u8;\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 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"
"#define max2 max\n"
"#define min2 min\n"
"#define WG_SIZE 64\n"
"typedef struct \n"
"{\n"
" int m_n;\n"
" int m_start;\n"
" int m_staticIdx;\n"
" int m_paddings[1];\n"
"} ConstBuffer;\n"
"typedef struct \n"
"{\n"
" int m_a;\n"
" int m_b;\n"
" u32 m_idx;\n"
"}Elem;\n"
"#define STACK_SIZE (WG_SIZE*10)\n"
"//#define STACK_SIZE (WG_SIZE)\n"
"#define RING_SIZE 1024\n"
"#define RING_SIZE_MASK (RING_SIZE-1)\n"
"#define CHECK_SIZE (WG_SIZE)\n"
"#define GET_RING_CAPACITY (RING_SIZE - ldsRingEnd)\n"
"#define RING_END ldsTmp\n"
"u32 readBuf(__local u32* buff, int idx)\n"
"{\n"
" idx = idx % (32*CHECK_SIZE);\n"
" int bitIdx = idx%32;\n"
" int bufIdx = idx/32;\n"
" return buff[bufIdx] & (1<<bitIdx);\n"
"}\n"
"void writeBuf(__local u32* buff, int idx)\n"
"{\n"
" idx = idx % (32*CHECK_SIZE);\n"
" int bitIdx = idx%32;\n"
" int bufIdx = idx/32;\n"
"// buff[bufIdx] |= (1<<bitIdx);\n"
" atom_or( &buff[bufIdx], (1<<bitIdx) );\n"
"}\n"
"u32 tryWrite(__local u32* buff, int idx)\n"
"{\n"
" idx = idx % (32*CHECK_SIZE);\n"
" int bitIdx = idx%32;\n"
" int bufIdx = idx/32;\n"
" u32 ans = (u32)atom_or( &buff[bufIdx], (1<<bitIdx) );\n"
" return ((ans >> bitIdx)&1) == 0;\n"
"}\n"
"// batching on the GPU\n"
"__kernel void CreateBatches( __global const struct b3Contact4Data* gConstraints, __global struct b3Contact4Data* gConstraintsOut,\n"
" __global const u32* gN, __global const u32* gStart, __global int* batchSizes, \n"
" int m_staticIdx )\n"
"{\n"
" __local u32 ldsStackIdx[STACK_SIZE];\n"
" __local u32 ldsStackEnd;\n"
" __local Elem ldsRingElem[RING_SIZE];\n"
" __local u32 ldsRingEnd;\n"
" __local u32 ldsTmp;\n"
" __local u32 ldsCheckBuffer[CHECK_SIZE];\n"
" __local u32 ldsFixedBuffer[CHECK_SIZE];\n"
" __local u32 ldsGEnd;\n"
" __local u32 ldsDstEnd;\n"
" int wgIdx = GET_GROUP_IDX;\n"
" int lIdx = GET_LOCAL_IDX;\n"
" \n"
" const int m_n = gN[wgIdx];\n"
" const int m_start = gStart[wgIdx];\n"
" \n"
" if( lIdx == 0 )\n"
" {\n"
" ldsRingEnd = 0;\n"
" ldsGEnd = 0;\n"
" ldsStackEnd = 0;\n"
" ldsDstEnd = m_start;\n"
" }\n"
" \n"
" \n"
" \n"
"// while(1)\n"
"//was 250\n"
" int ie=0;\n"
" int maxBatch = 0;\n"
" for(ie=0; ie<50; ie++)\n"
" {\n"
" ldsFixedBuffer[lIdx] = 0;\n"
" for(int giter=0; giter<4; giter++)\n"
" {\n"
" int ringCap = GET_RING_CAPACITY;\n"
" \n"
" // 1. fill ring\n"
" if( ldsGEnd < m_n )\n"
" {\n"
" while( ringCap > WG_SIZE )\n"
" {\n"
" if( ldsGEnd >= m_n ) break;\n"
" if( lIdx < ringCap - WG_SIZE )\n"
" {\n"
" int srcIdx;\n"
" AtomInc1( ldsGEnd, srcIdx );\n"
" if( srcIdx < m_n )\n"
" {\n"
" int dstIdx;\n"
" AtomInc1( ldsRingEnd, dstIdx );\n"
" \n"
" int a = gConstraints[m_start+srcIdx].m_bodyAPtrAndSignBit;\n"
" int b = gConstraints[m_start+srcIdx].m_bodyBPtrAndSignBit;\n"
" ldsRingElem[dstIdx].m_a = (a>b)? b:a;\n"
" ldsRingElem[dstIdx].m_b = (a>b)? a:b;\n"
" ldsRingElem[dstIdx].m_idx = srcIdx;\n"
" }\n"
" }\n"
" ringCap = GET_RING_CAPACITY;\n"
" }\n"
" }\n"
" GROUP_LDS_BARRIER;\n"
" \n"
" // 2. fill stack\n"
" __local Elem* dst = ldsRingElem;\n"
" if( lIdx == 0 ) RING_END = 0;\n"
" int srcIdx=lIdx;\n"
" int end = ldsRingEnd;\n"
" {\n"
" for(int ii=0; ii<end; ii+=WG_SIZE, srcIdx+=WG_SIZE)\n"
" {\n"
" Elem e;\n"
" if(srcIdx<end) e = ldsRingElem[srcIdx];\n"
" bool done = (srcIdx<end)?false:true;\n"
" for(int i=lIdx; i<CHECK_SIZE; i+=WG_SIZE) ldsCheckBuffer[lIdx] = 0;\n"
" \n"
" if( !done )\n"
" {\n"
" int aUsed = readBuf( ldsFixedBuffer, abs(e.m_a));\n"
" int bUsed = readBuf( ldsFixedBuffer, abs(e.m_b));\n"
" if( aUsed==0 && bUsed==0 )\n"
" {\n"
" int aAvailable=1;\n"
" int bAvailable=1;\n"
" int ea = abs(e.m_a);\n"
" int eb = abs(e.m_b);\n"
" bool aStatic = (e.m_a<0) ||(ea==m_staticIdx);\n"
" bool bStatic = (e.m_b<0) ||(eb==m_staticIdx);\n"
" \n"
" if (!aStatic)\n"
" aAvailable = tryWrite( ldsCheckBuffer, ea );\n"
" if (!bStatic)\n"
" bAvailable = tryWrite( ldsCheckBuffer, eb );\n"
" \n"
" //aAvailable = aStatic? 1: aAvailable;\n"
" //bAvailable = bStatic? 1: bAvailable;\n"
" bool success = (aAvailable && bAvailable);\n"
" if(success)\n"
" {\n"
" \n"
" if (!aStatic)\n"
" writeBuf( ldsFixedBuffer, ea );\n"
" if (!bStatic)\n"
" writeBuf( ldsFixedBuffer, eb );\n"
" }\n"
" done = success;\n"
" }\n"
" }\n"
" // put it aside\n"
" if(srcIdx<end)\n"
" {\n"
" if( done )\n"
" {\n"
" int dstIdx; AtomInc1( ldsStackEnd, dstIdx );\n"
" if( dstIdx < STACK_SIZE )\n"
" ldsStackIdx[dstIdx] = e.m_idx;\n"
" else{\n"
" done = false;\n"
" AtomAdd( ldsStackEnd, -1 );\n"
" }\n"
" }\n"
" if( !done )\n"
" {\n"
" int dstIdx; AtomInc1( RING_END, dstIdx );\n"
" dst[dstIdx] = e;\n"
" }\n"
" }\n"
" // if filled, flush\n"
" if( ldsStackEnd == STACK_SIZE )\n"
" {\n"
" for(int i=lIdx; i<STACK_SIZE; i+=WG_SIZE)\n"
" {\n"
" int idx = m_start + ldsStackIdx[i];\n"
" int dstIdx; AtomInc1( ldsDstEnd, dstIdx );\n"
" gConstraintsOut[ dstIdx ] = gConstraints[ idx ];\n"
" gConstraintsOut[ dstIdx ].m_batchIdx = ie;\n"
" }\n"
" if( lIdx == 0 ) ldsStackEnd = 0;\n"
" //for(int i=lIdx; i<CHECK_SIZE; i+=WG_SIZE) \n"
" ldsFixedBuffer[lIdx] = 0;\n"
" }\n"
" }\n"
" }\n"
" if( lIdx == 0 ) ldsRingEnd = RING_END;\n"
" }\n"
" GROUP_LDS_BARRIER;\n"
" for(int i=lIdx; i<ldsStackEnd; i+=WG_SIZE)\n"
" {\n"
" int idx = m_start + ldsStackIdx[i];\n"
" int dstIdx; AtomInc1( ldsDstEnd, dstIdx );\n"
" gConstraintsOut[ dstIdx ] = gConstraints[ idx ];\n"
" gConstraintsOut[ dstIdx ].m_batchIdx = ie;\n"
" }\n"
" // in case it couldn't consume any pair. Flush them\n"
" // todo. Serial batch worth while?\n"
" if( ldsStackEnd == 0 )\n"
" {\n"
" for(int i=lIdx; i<ldsRingEnd; i+=WG_SIZE)\n"
" {\n"
" int idx = m_start + ldsRingElem[i].m_idx;\n"
" int dstIdx; AtomInc1( ldsDstEnd, dstIdx );\n"
" gConstraintsOut[ dstIdx ] = gConstraints[ idx ];\n"
" int curBatch = 100+i;\n"
" if (maxBatch < curBatch)\n"
" maxBatch = curBatch;\n"
" \n"
" gConstraintsOut[ dstIdx ].m_batchIdx = curBatch;\n"
" \n"
" }\n"
" GROUP_LDS_BARRIER;\n"
" if( lIdx == 0 ) ldsRingEnd = 0;\n"
" }\n"
" if( lIdx == 0 ) ldsStackEnd = 0;\n"
" GROUP_LDS_BARRIER;\n"
" // termination\n"
" if( ldsGEnd == m_n && ldsRingEnd == 0 )\n"
" break;\n"
" }\n"
" if( lIdx == 0 )\n"
" {\n"
" if (maxBatch < ie)\n"
" maxBatch=ie;\n"
" batchSizes[wgIdx]=maxBatch;\n"
" }\n"
"}\n"
;
static const char* batchingKernelsCL =
"/*\n"
"Copyright (c) 2012 Advanced Micro Devices, Inc. \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"
"//Originally written by Takahiro Harada\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"
"#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"
"typedef unsigned int u32;\n"
"typedef unsigned short u16;\n"
"typedef unsigned char u8;\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 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"
"#define max2 max\n"
"#define min2 min\n"
"#define WG_SIZE 64\n"
"typedef struct \n"
"{\n"
" int m_n;\n"
" int m_start;\n"
" int m_staticIdx;\n"
" int m_paddings[1];\n"
"} ConstBuffer;\n"
"typedef struct \n"
"{\n"
" int m_a;\n"
" int m_b;\n"
" u32 m_idx;\n"
"}Elem;\n"
"#define STACK_SIZE (WG_SIZE*10)\n"
"//#define STACK_SIZE (WG_SIZE)\n"
"#define RING_SIZE 1024\n"
"#define RING_SIZE_MASK (RING_SIZE-1)\n"
"#define CHECK_SIZE (WG_SIZE)\n"
"#define GET_RING_CAPACITY (RING_SIZE - ldsRingEnd)\n"
"#define RING_END ldsTmp\n"
"u32 readBuf(__local u32* buff, int idx)\n"
"{\n"
" idx = idx % (32*CHECK_SIZE);\n"
" int bitIdx = idx%32;\n"
" int bufIdx = idx/32;\n"
" return buff[bufIdx] & (1<<bitIdx);\n"
"}\n"
"void writeBuf(__local u32* buff, int idx)\n"
"{\n"
" idx = idx % (32*CHECK_SIZE);\n"
" int bitIdx = idx%32;\n"
" int bufIdx = idx/32;\n"
"// buff[bufIdx] |= (1<<bitIdx);\n"
" atom_or( &buff[bufIdx], (1<<bitIdx) );\n"
"}\n"
"u32 tryWrite(__local u32* buff, int idx)\n"
"{\n"
" idx = idx % (32*CHECK_SIZE);\n"
" int bitIdx = idx%32;\n"
" int bufIdx = idx/32;\n"
" u32 ans = (u32)atom_or( &buff[bufIdx], (1<<bitIdx) );\n"
" return ((ans >> bitIdx)&1) == 0;\n"
"}\n"
"// batching on the GPU\n"
"__kernel void CreateBatches( __global const struct b3Contact4Data* gConstraints, __global struct b3Contact4Data* gConstraintsOut,\n"
" __global const u32* gN, __global const u32* gStart, __global int* batchSizes, \n"
" int m_staticIdx )\n"
"{\n"
" __local u32 ldsStackIdx[STACK_SIZE];\n"
" __local u32 ldsStackEnd;\n"
" __local Elem ldsRingElem[RING_SIZE];\n"
" __local u32 ldsRingEnd;\n"
" __local u32 ldsTmp;\n"
" __local u32 ldsCheckBuffer[CHECK_SIZE];\n"
" __local u32 ldsFixedBuffer[CHECK_SIZE];\n"
" __local u32 ldsGEnd;\n"
" __local u32 ldsDstEnd;\n"
" int wgIdx = GET_GROUP_IDX;\n"
" int lIdx = GET_LOCAL_IDX;\n"
" \n"
" const int m_n = gN[wgIdx];\n"
" const int m_start = gStart[wgIdx];\n"
" \n"
" if( lIdx == 0 )\n"
" {\n"
" ldsRingEnd = 0;\n"
" ldsGEnd = 0;\n"
" ldsStackEnd = 0;\n"
" ldsDstEnd = m_start;\n"
" }\n"
" \n"
" \n"
" \n"
"// while(1)\n"
"//was 250\n"
" int ie=0;\n"
" int maxBatch = 0;\n"
" for(ie=0; ie<50; ie++)\n"
" {\n"
" ldsFixedBuffer[lIdx] = 0;\n"
" for(int giter=0; giter<4; giter++)\n"
" {\n"
" int ringCap = GET_RING_CAPACITY;\n"
" \n"
" // 1. fill ring\n"
" if( ldsGEnd < m_n )\n"
" {\n"
" while( ringCap > WG_SIZE )\n"
" {\n"
" if( ldsGEnd >= m_n ) break;\n"
" if( lIdx < ringCap - WG_SIZE )\n"
" {\n"
" int srcIdx;\n"
" AtomInc1( ldsGEnd, srcIdx );\n"
" if( srcIdx < m_n )\n"
" {\n"
" int dstIdx;\n"
" AtomInc1( ldsRingEnd, dstIdx );\n"
" \n"
" int a = gConstraints[m_start+srcIdx].m_bodyAPtrAndSignBit;\n"
" int b = gConstraints[m_start+srcIdx].m_bodyBPtrAndSignBit;\n"
" ldsRingElem[dstIdx].m_a = (a>b)? b:a;\n"
" ldsRingElem[dstIdx].m_b = (a>b)? a:b;\n"
" ldsRingElem[dstIdx].m_idx = srcIdx;\n"
" }\n"
" }\n"
" ringCap = GET_RING_CAPACITY;\n"
" }\n"
" }\n"
" GROUP_LDS_BARRIER;\n"
" \n"
" // 2. fill stack\n"
" __local Elem* dst = ldsRingElem;\n"
" if( lIdx == 0 ) RING_END = 0;\n"
" int srcIdx=lIdx;\n"
" int end = ldsRingEnd;\n"
" {\n"
" for(int ii=0; ii<end; ii+=WG_SIZE, srcIdx+=WG_SIZE)\n"
" {\n"
" Elem e;\n"
" if(srcIdx<end) e = ldsRingElem[srcIdx];\n"
" bool done = (srcIdx<end)?false:true;\n"
" for(int i=lIdx; i<CHECK_SIZE; i+=WG_SIZE) ldsCheckBuffer[lIdx] = 0;\n"
" \n"
" if( !done )\n"
" {\n"
" int aUsed = readBuf( ldsFixedBuffer, abs(e.m_a));\n"
" int bUsed = readBuf( ldsFixedBuffer, abs(e.m_b));\n"
" if( aUsed==0 && bUsed==0 )\n"
" {\n"
" int aAvailable=1;\n"
" int bAvailable=1;\n"
" int ea = abs(e.m_a);\n"
" int eb = abs(e.m_b);\n"
" bool aStatic = (e.m_a<0) ||(ea==m_staticIdx);\n"
" bool bStatic = (e.m_b<0) ||(eb==m_staticIdx);\n"
" \n"
" if (!aStatic)\n"
" aAvailable = tryWrite( ldsCheckBuffer, ea );\n"
" if (!bStatic)\n"
" bAvailable = tryWrite( ldsCheckBuffer, eb );\n"
" \n"
" //aAvailable = aStatic? 1: aAvailable;\n"
" //bAvailable = bStatic? 1: bAvailable;\n"
" bool success = (aAvailable && bAvailable);\n"
" if(success)\n"
" {\n"
" \n"
" if (!aStatic)\n"
" writeBuf( ldsFixedBuffer, ea );\n"
" if (!bStatic)\n"
" writeBuf( ldsFixedBuffer, eb );\n"
" }\n"
" done = success;\n"
" }\n"
" }\n"
" // put it aside\n"
" if(srcIdx<end)\n"
" {\n"
" if( done )\n"
" {\n"
" int dstIdx; AtomInc1( ldsStackEnd, dstIdx );\n"
" if( dstIdx < STACK_SIZE )\n"
" ldsStackIdx[dstIdx] = e.m_idx;\n"
" else{\n"
" done = false;\n"
" AtomAdd( ldsStackEnd, -1 );\n"
" }\n"
" }\n"
" if( !done )\n"
" {\n"
" int dstIdx; AtomInc1( RING_END, dstIdx );\n"
" dst[dstIdx] = e;\n"
" }\n"
" }\n"
" // if filled, flush\n"
" if( ldsStackEnd == STACK_SIZE )\n"
" {\n"
" for(int i=lIdx; i<STACK_SIZE; i+=WG_SIZE)\n"
" {\n"
" int idx = m_start + ldsStackIdx[i];\n"
" int dstIdx; AtomInc1( ldsDstEnd, dstIdx );\n"
" gConstraintsOut[ dstIdx ] = gConstraints[ idx ];\n"
" gConstraintsOut[ dstIdx ].m_batchIdx = ie;\n"
" }\n"
" if( lIdx == 0 ) ldsStackEnd = 0;\n"
" //for(int i=lIdx; i<CHECK_SIZE; i+=WG_SIZE) \n"
" ldsFixedBuffer[lIdx] = 0;\n"
" }\n"
" }\n"
" }\n"
" if( lIdx == 0 ) ldsRingEnd = RING_END;\n"
" }\n"
" GROUP_LDS_BARRIER;\n"
" for(int i=lIdx; i<ldsStackEnd; i+=WG_SIZE)\n"
" {\n"
" int idx = m_start + ldsStackIdx[i];\n"
" int dstIdx; AtomInc1( ldsDstEnd, dstIdx );\n"
" gConstraintsOut[ dstIdx ] = gConstraints[ idx ];\n"
" gConstraintsOut[ dstIdx ].m_batchIdx = ie;\n"
" }\n"
" // in case it couldn't consume any pair. Flush them\n"
" // todo. Serial batch worth while?\n"
" if( ldsStackEnd == 0 )\n"
" {\n"
" for(int i=lIdx; i<ldsRingEnd; i+=WG_SIZE)\n"
" {\n"
" int idx = m_start + ldsRingElem[i].m_idx;\n"
" int dstIdx; AtomInc1( ldsDstEnd, dstIdx );\n"
" gConstraintsOut[ dstIdx ] = gConstraints[ idx ];\n"
" int curBatch = 100+i;\n"
" if (maxBatch < curBatch)\n"
" maxBatch = curBatch;\n"
" \n"
" gConstraintsOut[ dstIdx ].m_batchIdx = curBatch;\n"
" \n"
" }\n"
" GROUP_LDS_BARRIER;\n"
" if( lIdx == 0 ) ldsRingEnd = 0;\n"
" }\n"
" if( lIdx == 0 ) ldsStackEnd = 0;\n"
" GROUP_LDS_BARRIER;\n"
" // termination\n"
" if( ldsGEnd == m_n && ldsRingEnd == 0 )\n"
" break;\n"
" }\n"
" if( lIdx == 0 )\n"
" {\n"
" if (maxBatch < ie)\n"
" maxBatch=ie;\n"
" batchSizes[wgIdx]=maxBatch;\n"
" }\n"
"}\n";

579
src/Bullet3OpenCL/RigidBody/kernels/batchingKernelsNew.h
View File

@@ -1,291 +1,290 @@
//this file is autogenerated using stringify.bat (premake --stringify) in the build folder of this project
static const char* batchingKernelsNewCL= \
"/*\n"
"Copyright (c) 2012 Advanced Micro Devices, Inc. \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"
"//Originally written by Erwin Coumans\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"
"#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 SIMD_WIDTH 64\n"
"typedef unsigned int u32;\n"
"typedef unsigned short u16;\n"
"typedef unsigned char u8;\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 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"
"#define max2 max\n"
"#define min2 min\n"
"#define WG_SIZE 64\n"
"typedef struct \n"
"{\n"
" int m_n;\n"
" int m_start;\n"
" int m_staticIdx;\n"
" int m_paddings[1];\n"
"} ConstBuffer;\n"
"typedef struct \n"
"{\n"
" int m_a;\n"
" int m_b;\n"
" u32 m_idx;\n"
"}Elem;\n"
"// batching on the GPU\n"
"__kernel void CreateBatchesBruteForce( __global struct b3Contact4Data* gConstraints, __global const u32* gN, __global const u32* gStart, int m_staticIdx )\n"
"{\n"
" int wgIdx = GET_GROUP_IDX;\n"
" int lIdx = GET_LOCAL_IDX;\n"
" \n"
" const int m_n = gN[wgIdx];\n"
" const int m_start = gStart[wgIdx];\n"
" \n"
" if( lIdx == 0 )\n"
" {\n"
" for (int i=0;i<m_n;i++)\n"
" {\n"
" int srcIdx = i+m_start;\n"
" int batchIndex = i;\n"
" gConstraints[ srcIdx ].m_batchIdx = batchIndex; \n"
" }\n"
" }\n"
"}\n"
"#define CHECK_SIZE (WG_SIZE)\n"
"u32 readBuf(__local u32* buff, int idx)\n"
"{\n"
" idx = idx % (32*CHECK_SIZE);\n"
" int bitIdx = idx%32;\n"
" int bufIdx = idx/32;\n"
" return buff[bufIdx] & (1<<bitIdx);\n"
"}\n"
"void writeBuf(__local u32* buff, int idx)\n"
"{\n"
" idx = idx % (32*CHECK_SIZE);\n"
" int bitIdx = idx%32;\n"
" int bufIdx = idx/32;\n"
" buff[bufIdx] |= (1<<bitIdx);\n"
" //atom_or( &buff[bufIdx], (1<<bitIdx) );\n"
"}\n"
"u32 tryWrite(__local u32* buff, int idx)\n"
"{\n"
" idx = idx % (32*CHECK_SIZE);\n"
" int bitIdx = idx%32;\n"
" int bufIdx = idx/32;\n"
" u32 ans = (u32)atom_or( &buff[bufIdx], (1<<bitIdx) );\n"
" return ((ans >> bitIdx)&1) == 0;\n"
"}\n"
"// batching on the GPU\n"
"__kernel void CreateBatchesNew( __global struct b3Contact4Data* gConstraints, __global const u32* gN, __global const u32* gStart, __global int* batchSizes, int staticIdx )\n"
"{\n"
" int wgIdx = GET_GROUP_IDX;\n"
" int lIdx = GET_LOCAL_IDX;\n"
" const int numConstraints = gN[wgIdx];\n"
" const int m_start = gStart[wgIdx];\n"
" b3Contact4Data_t tmp;\n"
" \n"
" __local u32 ldsFixedBuffer[CHECK_SIZE];\n"
" \n"
" \n"
" \n"
" \n"
" \n"
" if( lIdx == 0 )\n"
" {\n"
" \n"
" \n"
" __global struct b3Contact4Data* cs = &gConstraints[m_start]; \n"
" \n"
" \n"
" int numValidConstraints = 0;\n"
" int batchIdx = 0;\n"
" while( numValidConstraints < numConstraints)\n"
" {\n"
" int nCurrentBatch = 0;\n"
" // clear flag\n"
" \n"
" for(int i=0; i<CHECK_SIZE; i++) \n"
" ldsFixedBuffer[i] = 0; \n"
" for(int i=numValidConstraints; i<numConstraints; i++)\n"
" {\n"
" int bodyAS = cs[i].m_bodyAPtrAndSignBit;\n"
" int bodyBS = cs[i].m_bodyBPtrAndSignBit;\n"
" int bodyA = abs(bodyAS);\n"
" int bodyB = abs(bodyBS);\n"
" bool aIsStatic = (bodyAS<0) || bodyAS==staticIdx;\n"
" bool bIsStatic = (bodyBS<0) || bodyBS==staticIdx;\n"
" int aUnavailable = aIsStatic ? 0 : readBuf( ldsFixedBuffer, bodyA);\n"
" int bUnavailable = bIsStatic ? 0 : readBuf( ldsFixedBuffer, bodyB);\n"
" \n"
" if( aUnavailable==0 && bUnavailable==0 ) // ok\n"
" {\n"
" if (!aIsStatic)\n"
" {\n"
" writeBuf( ldsFixedBuffer, bodyA );\n"
" }\n"
" if (!bIsStatic)\n"
" {\n"
" writeBuf( ldsFixedBuffer, bodyB );\n"
" }\n"
" cs[i].m_batchIdx = batchIdx;\n"
" if (i!=numValidConstraints)\n"
" {\n"
" tmp = cs[i];\n"
" cs[i] = cs[numValidConstraints];\n"
" cs[numValidConstraints] = tmp;\n"
" }\n"
" numValidConstraints++;\n"
" \n"
" nCurrentBatch++;\n"
" if( nCurrentBatch == SIMD_WIDTH)\n"
" {\n"
" nCurrentBatch = 0;\n"
" for(int i=0; i<CHECK_SIZE; i++) \n"
" ldsFixedBuffer[i] = 0;\n"
" \n"
" }\n"
" }\n"
" }//for\n"
" batchIdx ++;\n"
" }//while\n"
" \n"
" batchSizes[wgIdx] = batchIdx;\n"
" }//if( lIdx == 0 )\n"
" \n"
" //return batchIdx;\n"
"}\n"
;
static const char* batchingKernelsNewCL =
"/*\n"
"Copyright (c) 2012 Advanced Micro Devices, Inc. \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"
"//Originally written by Erwin Coumans\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"
"#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 SIMD_WIDTH 64\n"
"typedef unsigned int u32;\n"
"typedef unsigned short u16;\n"
"typedef unsigned char u8;\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 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"
"#define max2 max\n"
"#define min2 min\n"
"#define WG_SIZE 64\n"
"typedef struct \n"
"{\n"
" int m_n;\n"
" int m_start;\n"
" int m_staticIdx;\n"
" int m_paddings[1];\n"
"} ConstBuffer;\n"
"typedef struct \n"
"{\n"
" int m_a;\n"
" int m_b;\n"
" u32 m_idx;\n"
"}Elem;\n"
"// batching on the GPU\n"
"__kernel void CreateBatchesBruteForce( __global struct b3Contact4Data* gConstraints, __global const u32* gN, __global const u32* gStart, int m_staticIdx )\n"
"{\n"
" int wgIdx = GET_GROUP_IDX;\n"
" int lIdx = GET_LOCAL_IDX;\n"
" \n"
" const int m_n = gN[wgIdx];\n"
" const int m_start = gStart[wgIdx];\n"
" \n"
" if( lIdx == 0 )\n"
" {\n"
" for (int i=0;i<m_n;i++)\n"
" {\n"
" int srcIdx = i+m_start;\n"
" int batchIndex = i;\n"
" gConstraints[ srcIdx ].m_batchIdx = batchIndex; \n"
" }\n"
" }\n"
"}\n"
"#define CHECK_SIZE (WG_SIZE)\n"
"u32 readBuf(__local u32* buff, int idx)\n"
"{\n"
" idx = idx % (32*CHECK_SIZE);\n"
" int bitIdx = idx%32;\n"
" int bufIdx = idx/32;\n"
" return buff[bufIdx] & (1<<bitIdx);\n"
"}\n"
"void writeBuf(__local u32* buff, int idx)\n"
"{\n"
" idx = idx % (32*CHECK_SIZE);\n"
" int bitIdx = idx%32;\n"
" int bufIdx = idx/32;\n"
" buff[bufIdx] |= (1<<bitIdx);\n"
" //atom_or( &buff[bufIdx], (1<<bitIdx) );\n"
"}\n"
"u32 tryWrite(__local u32* buff, int idx)\n"
"{\n"
" idx = idx % (32*CHECK_SIZE);\n"
" int bitIdx = idx%32;\n"
" int bufIdx = idx/32;\n"
" u32 ans = (u32)atom_or( &buff[bufIdx], (1<<bitIdx) );\n"
" return ((ans >> bitIdx)&1) == 0;\n"
"}\n"
"// batching on the GPU\n"
"__kernel void CreateBatchesNew( __global struct b3Contact4Data* gConstraints, __global const u32* gN, __global const u32* gStart, __global int* batchSizes, int staticIdx )\n"
"{\n"
" int wgIdx = GET_GROUP_IDX;\n"
" int lIdx = GET_LOCAL_IDX;\n"
" const int numConstraints = gN[wgIdx];\n"
" const int m_start = gStart[wgIdx];\n"
" b3Contact4Data_t tmp;\n"
" \n"
" __local u32 ldsFixedBuffer[CHECK_SIZE];\n"
" \n"
" \n"
" \n"
" \n"
" \n"
" if( lIdx == 0 )\n"
" {\n"
" \n"
" \n"
" __global struct b3Contact4Data* cs = &gConstraints[m_start]; \n"
" \n"
" \n"
" int numValidConstraints = 0;\n"
" int batchIdx = 0;\n"
" while( numValidConstraints < numConstraints)\n"
" {\n"
" int nCurrentBatch = 0;\n"
" // clear flag\n"
" \n"
" for(int i=0; i<CHECK_SIZE; i++) \n"
" ldsFixedBuffer[i] = 0; \n"
" for(int i=numValidConstraints; i<numConstraints; i++)\n"
" {\n"
" int bodyAS = cs[i].m_bodyAPtrAndSignBit;\n"
" int bodyBS = cs[i].m_bodyBPtrAndSignBit;\n"
" int bodyA = abs(bodyAS);\n"
" int bodyB = abs(bodyBS);\n"
" bool aIsStatic = (bodyAS<0) || bodyAS==staticIdx;\n"
" bool bIsStatic = (bodyBS<0) || bodyBS==staticIdx;\n"
" int aUnavailable = aIsStatic ? 0 : readBuf( ldsFixedBuffer, bodyA);\n"
" int bUnavailable = bIsStatic ? 0 : readBuf( ldsFixedBuffer, bodyB);\n"
" \n"
" if( aUnavailable==0 && bUnavailable==0 ) // ok\n"
" {\n"
" if (!aIsStatic)\n"
" {\n"
" writeBuf( ldsFixedBuffer, bodyA );\n"
" }\n"
" if (!bIsStatic)\n"
" {\n"
" writeBuf( ldsFixedBuffer, bodyB );\n"
" }\n"
" cs[i].m_batchIdx = batchIdx;\n"
" if (i!=numValidConstraints)\n"
" {\n"
" tmp = cs[i];\n"
" cs[i] = cs[numValidConstraints];\n"
" cs[numValidConstraints] = tmp;\n"
" }\n"
" numValidConstraints++;\n"
" \n"
" nCurrentBatch++;\n"
" if( nCurrentBatch == SIMD_WIDTH)\n"
" {\n"
" nCurrentBatch = 0;\n"
" for(int i=0; i<CHECK_SIZE; i++) \n"
" ldsFixedBuffer[i] = 0;\n"
" \n"
" }\n"
" }\n"
" }//for\n"
" batchIdx ++;\n"
" }//while\n"
" \n"
" batchSizes[wgIdx] = batchIdx;\n"
" }//if( lIdx == 0 )\n"
" \n"
" //return batchIdx;\n"
"}\n";

863
src/Bullet3OpenCL/RigidBody/kernels/integrateKernel.h
View File

@@ -1,433 +1,432 @@
//this file is autogenerated using stringify.bat (premake --stringify) in the build folder of this project
static const char* integrateKernelCL= \
"/*\n"
"Copyright (c) 2013 Advanced Micro Devices, Inc. \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"
"//Originally written by Erwin Coumans\n"
"#ifndef B3_RIGIDBODY_DATA_H\n"
"#define B3_RIGIDBODY_DATA_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_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_RIGIDBODY_DATA_H\n"
"#endif //B3_RIGIDBODY_DATA_H\n"
" \n"
"inline void integrateSingleTransform( __global b3RigidBodyData_t* bodies,int nodeID, float timeStep, float angularDamping, b3Float4ConstArg gravityAcceleration)\n"
"{\n"
" \n"
" if (bodies[nodeID].m_invMass != 0.f)\n"
" {\n"
" float BT_GPU_ANGULAR_MOTION_THRESHOLD = (0.25f * 3.14159254f);\n"
" //angular velocity\n"
" {\n"
" b3Float4 axis;\n"
" //add some hardcoded angular damping\n"
" bodies[nodeID].m_angVel.x *= angularDamping;\n"
" bodies[nodeID].m_angVel.y *= angularDamping;\n"
" bodies[nodeID].m_angVel.z *= angularDamping;\n"
" \n"
" b3Float4 angvel = bodies[nodeID].m_angVel;\n"
" float fAngle = b3Sqrt(b3Dot3F4(angvel, angvel));\n"
" \n"
" //limit the angular motion\n"
" if(fAngle*timeStep > BT_GPU_ANGULAR_MOTION_THRESHOLD)\n"
" {\n"
" fAngle = BT_GPU_ANGULAR_MOTION_THRESHOLD / timeStep;\n"
" }\n"
" if(fAngle < 0.001f)\n"
" {\n"
" // use Taylor's expansions of sync function\n"
" axis = angvel * (0.5f*timeStep-(timeStep*timeStep*timeStep)*0.020833333333f * fAngle * fAngle);\n"
" }\n"
" else\n"
" {\n"
" // sync(fAngle) = sin(c*fAngle)/t\n"
" axis = angvel * ( b3Sin(0.5f * fAngle * timeStep) / fAngle);\n"
" }\n"
" \n"
" b3Quat dorn;\n"
" dorn.x = axis.x;\n"
" dorn.y = axis.y;\n"
" dorn.z = axis.z;\n"
" dorn.w = b3Cos(fAngle * timeStep * 0.5f);\n"
" b3Quat orn0 = bodies[nodeID].m_quat;\n"
" b3Quat predictedOrn = b3QuatMul(dorn, orn0);\n"
" predictedOrn = b3QuatNormalized(predictedOrn);\n"
" bodies[nodeID].m_quat=predictedOrn;\n"
" }\n"
" //linear velocity \n"
" bodies[nodeID].m_pos += bodies[nodeID].m_linVel * timeStep;\n"
" \n"
" //apply gravity\n"
" bodies[nodeID].m_linVel += gravityAcceleration * timeStep;\n"
" \n"
" }\n"
" \n"
"}\n"
"inline void b3IntegrateTransform( __global b3RigidBodyData_t* body, float timeStep, float angularDamping, b3Float4ConstArg gravityAcceleration)\n"
"{\n"
" float BT_GPU_ANGULAR_MOTION_THRESHOLD = (0.25f * 3.14159254f);\n"
" \n"
" if( (body->m_invMass != 0.f))\n"
" {\n"
" //angular velocity\n"
" {\n"
" b3Float4 axis;\n"
" //add some hardcoded angular damping\n"
" body->m_angVel.x *= angularDamping;\n"
" body->m_angVel.y *= angularDamping;\n"
" body->m_angVel.z *= angularDamping;\n"
" \n"
" b3Float4 angvel = body->m_angVel;\n"
" float fAngle = b3Sqrt(b3Dot3F4(angvel, angvel));\n"
" //limit the angular motion\n"
" if(fAngle*timeStep > BT_GPU_ANGULAR_MOTION_THRESHOLD)\n"
" {\n"
" fAngle = BT_GPU_ANGULAR_MOTION_THRESHOLD / timeStep;\n"
" }\n"
" if(fAngle < 0.001f)\n"
" {\n"
" // use Taylor's expansions of sync function\n"
" axis = angvel * (0.5f*timeStep-(timeStep*timeStep*timeStep)*0.020833333333f * fAngle * fAngle);\n"
" }\n"
" else\n"
" {\n"
" // sync(fAngle) = sin(c*fAngle)/t\n"
" axis = angvel * ( b3Sin(0.5f * fAngle * timeStep) / fAngle);\n"
" }\n"
" b3Quat dorn;\n"
" dorn.x = axis.x;\n"
" dorn.y = axis.y;\n"
" dorn.z = axis.z;\n"
" dorn.w = b3Cos(fAngle * timeStep * 0.5f);\n"
" b3Quat orn0 = body->m_quat;\n"
" b3Quat predictedOrn = b3QuatMul(dorn, orn0);\n"
" predictedOrn = b3QuatNormalized(predictedOrn);\n"
" body->m_quat=predictedOrn;\n"
" }\n"
" //apply gravity\n"
" body->m_linVel += gravityAcceleration * timeStep;\n"
" //linear velocity \n"
" body->m_pos += body->m_linVel * timeStep;\n"
" \n"
" }\n"
" \n"
"}\n"
"__kernel void \n"
" integrateTransformsKernel( __global b3RigidBodyData_t* bodies,const int numNodes, float timeStep, float angularDamping, float4 gravityAcceleration)\n"
"{\n"
" int nodeID = get_global_id(0);\n"
" \n"
" if( nodeID < numNodes)\n"
" {\n"
" integrateSingleTransform(bodies,nodeID, timeStep, angularDamping,gravityAcceleration);\n"
" }\n"
"}\n"
;
static const char* integrateKernelCL =
"/*\n"
"Copyright (c) 2013 Advanced Micro Devices, Inc. \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"
"//Originally written by Erwin Coumans\n"
"#ifndef B3_RIGIDBODY_DATA_H\n"
"#define B3_RIGIDBODY_DATA_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_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_RIGIDBODY_DATA_H\n"
"#endif //B3_RIGIDBODY_DATA_H\n"
" \n"
"inline void integrateSingleTransform( __global b3RigidBodyData_t* bodies,int nodeID, float timeStep, float angularDamping, b3Float4ConstArg gravityAcceleration)\n"
"{\n"
" \n"
" if (bodies[nodeID].m_invMass != 0.f)\n"
" {\n"
" float BT_GPU_ANGULAR_MOTION_THRESHOLD = (0.25f * 3.14159254f);\n"
" //angular velocity\n"
" {\n"
" b3Float4 axis;\n"
" //add some hardcoded angular damping\n"
" bodies[nodeID].m_angVel.x *= angularDamping;\n"
" bodies[nodeID].m_angVel.y *= angularDamping;\n"
" bodies[nodeID].m_angVel.z *= angularDamping;\n"
" \n"
" b3Float4 angvel = bodies[nodeID].m_angVel;\n"
" float fAngle = b3Sqrt(b3Dot3F4(angvel, angvel));\n"
" \n"
" //limit the angular motion\n"
" if(fAngle*timeStep > BT_GPU_ANGULAR_MOTION_THRESHOLD)\n"
" {\n"
" fAngle = BT_GPU_ANGULAR_MOTION_THRESHOLD / timeStep;\n"
" }\n"
" if(fAngle < 0.001f)\n"
" {\n"
" // use Taylor's expansions of sync function\n"
" axis = angvel * (0.5f*timeStep-(timeStep*timeStep*timeStep)*0.020833333333f * fAngle * fAngle);\n"
" }\n"
" else\n"
" {\n"
" // sync(fAngle) = sin(c*fAngle)/t\n"
" axis = angvel * ( b3Sin(0.5f * fAngle * timeStep) / fAngle);\n"
" }\n"
" \n"
" b3Quat dorn;\n"
" dorn.x = axis.x;\n"
" dorn.y = axis.y;\n"
" dorn.z = axis.z;\n"
" dorn.w = b3Cos(fAngle * timeStep * 0.5f);\n"
" b3Quat orn0 = bodies[nodeID].m_quat;\n"
" b3Quat predictedOrn = b3QuatMul(dorn, orn0);\n"
" predictedOrn = b3QuatNormalized(predictedOrn);\n"
" bodies[nodeID].m_quat=predictedOrn;\n"
" }\n"
" //linear velocity \n"
" bodies[nodeID].m_pos += bodies[nodeID].m_linVel * timeStep;\n"
" \n"
" //apply gravity\n"
" bodies[nodeID].m_linVel += gravityAcceleration * timeStep;\n"
" \n"
" }\n"
" \n"
"}\n"
"inline void b3IntegrateTransform( __global b3RigidBodyData_t* body, float timeStep, float angularDamping, b3Float4ConstArg gravityAcceleration)\n"
"{\n"
" float BT_GPU_ANGULAR_MOTION_THRESHOLD = (0.25f * 3.14159254f);\n"
" \n"
" if( (body->m_invMass != 0.f))\n"
" {\n"
" //angular velocity\n"
" {\n"
" b3Float4 axis;\n"
" //add some hardcoded angular damping\n"
" body->m_angVel.x *= angularDamping;\n"
" body->m_angVel.y *= angularDamping;\n"
" body->m_angVel.z *= angularDamping;\n"
" \n"
" b3Float4 angvel = body->m_angVel;\n"
" float fAngle = b3Sqrt(b3Dot3F4(angvel, angvel));\n"
" //limit the angular motion\n"
" if(fAngle*timeStep > BT_GPU_ANGULAR_MOTION_THRESHOLD)\n"
" {\n"
" fAngle = BT_GPU_ANGULAR_MOTION_THRESHOLD / timeStep;\n"
" }\n"
" if(fAngle < 0.001f)\n"
" {\n"
" // use Taylor's expansions of sync function\n"
" axis = angvel * (0.5f*timeStep-(timeStep*timeStep*timeStep)*0.020833333333f * fAngle * fAngle);\n"
" }\n"
" else\n"
" {\n"
" // sync(fAngle) = sin(c*fAngle)/t\n"
" axis = angvel * ( b3Sin(0.5f * fAngle * timeStep) / fAngle);\n"
" }\n"
" b3Quat dorn;\n"
" dorn.x = axis.x;\n"
" dorn.y = axis.y;\n"
" dorn.z = axis.z;\n"
" dorn.w = b3Cos(fAngle * timeStep * 0.5f);\n"
" b3Quat orn0 = body->m_quat;\n"
" b3Quat predictedOrn = b3QuatMul(dorn, orn0);\n"
" predictedOrn = b3QuatNormalized(predictedOrn);\n"
" body->m_quat=predictedOrn;\n"
" }\n"
" //apply gravity\n"
" body->m_linVel += gravityAcceleration * timeStep;\n"
" //linear velocity \n"
" body->m_pos += body->m_linVel * timeStep;\n"
" \n"
" }\n"
" \n"
"}\n"
"__kernel void \n"
" integrateTransformsKernel( __global b3RigidBodyData_t* bodies,const int numNodes, float timeStep, float angularDamping, float4 gravityAcceleration)\n"
"{\n"
" int nodeID = get_global_id(0);\n"
" \n"
" if( nodeID < numNodes)\n"
" {\n"
" integrateSingleTransform(bodies,nodeID, timeStep, angularDamping,gravityAcceleration);\n"
" }\n"
"}\n";

1439
src/Bullet3OpenCL/RigidBody/kernels/jointSolver.h
View File

File diff suppressed because it is too large Load Diff

783
src/Bullet3OpenCL/RigidBody/kernels/solveContact.h
View File

@@ -1,393 +1,392 @@
//this file is autogenerated using stringify.bat (premake --stringify) in the build folder of this project
static const char* solveContactCL= \
"/*\n"
"Copyright (c) 2012 Advanced Micro Devices, Inc. \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"
"//Originally written by Takahiro Harada\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"
"typedef unsigned int u32;\n"
"typedef unsigned short u16;\n"
"typedef unsigned char u8;\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 SELECT_UINT4( b, a, condition ) select( b,a,condition )\n"
"#define mymake_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"
"#define max2 max\n"
"#define min2 min\n"
"///////////////////////////////////////\n"
"// Vector\n"
"///////////////////////////////////////\n"
"__inline\n"
"float4 fastNormalize4(float4 v)\n"
"{\n"
" return fast_normalize(v);\n"
"}\n"
"__inline\n"
"float4 cross3(float4 a, float4 b)\n"
"{\n"
" return cross(a,b);\n"
"}\n"
"__inline\n"
"float dot3F4(float4 a, float4 b)\n"
"{\n"
" float4 a1 = mymake_float4(a.xyz,0.f);\n"
" float4 b1 = mymake_float4(b.xyz,0.f);\n"
" return dot(a1, b1);\n"
"}\n"
"__inline\n"
"float4 normalize3(const float4 a)\n"
"{\n"
" float4 n = mymake_float4(a.x, a.y, a.z, 0.f);\n"
" return fastNormalize4( n );\n"
"// float length = sqrtf(dot3F4(a, a));\n"
"// return 1.f/length * a;\n"
"}\n"
"///////////////////////////////////////\n"
"// Matrix3x3\n"
"///////////////////////////////////////\n"
"typedef struct\n"
"{\n"
" float4 m_row[3];\n"
"}Matrix3x3;\n"
"__inline\n"
"float4 mtMul1(Matrix3x3 a, float4 b);\n"
"__inline\n"
"float4 mtMul3(float4 a, Matrix3x3 b);\n"
"__inline\n"
"float4 mtMul1(Matrix3x3 a, float4 b)\n"
"{\n"
" float4 ans;\n"
" ans.x = dot3F4( a.m_row[0], b );\n"
" ans.y = dot3F4( a.m_row[1], b );\n"
" ans.z = dot3F4( a.m_row[2], b );\n"
" ans.w = 0.f;\n"
" return ans;\n"
"}\n"
"__inline\n"
"float4 mtMul3(float4 a, Matrix3x3 b)\n"
"{\n"
" float4 colx = mymake_float4(b.m_row[0].x, b.m_row[1].x, b.m_row[2].x, 0);\n"
" float4 coly = mymake_float4(b.m_row[0].y, b.m_row[1].y, b.m_row[2].y, 0);\n"
" float4 colz = mymake_float4(b.m_row[0].z, b.m_row[1].z, b.m_row[2].z, 0);\n"
" float4 ans;\n"
" ans.x = dot3F4( a, colx );\n"
" ans.y = dot3F4( a, coly );\n"
" ans.z = dot3F4( a, colz );\n"
" return ans;\n"
"}\n"
"///////////////////////////////////////\n"
"// Quaternion\n"
"///////////////////////////////////////\n"
"typedef float4 Quaternion;\n"
"#define WG_SIZE 64\n"
"typedef struct\n"
"{\n"
" float4 m_pos;\n"
" Quaternion m_quat;\n"
" float4 m_linVel;\n"
" float4 m_angVel;\n"
" u32 m_shapeIdx;\n"
" float m_invMass;\n"
" float m_restituitionCoeff;\n"
" float m_frictionCoeff;\n"
"} Body;\n"
"typedef struct\n"
"{\n"
" Matrix3x3 m_invInertia;\n"
" Matrix3x3 m_initInvInertia;\n"
"} Shape;\n"
"typedef struct\n"
"{\n"
" float4 m_linear;\n"
" float4 m_worldPos[4];\n"
" float4 m_center; \n"
" float m_jacCoeffInv[4];\n"
" float m_b[4];\n"
" float m_appliedRambdaDt[4];\n"
" float m_fJacCoeffInv[2]; \n"
" float m_fAppliedRambdaDt[2]; \n"
" u32 m_bodyA;\n"
" u32 m_bodyB;\n"
" int m_batchIdx;\n"
" u32 m_paddings[1];\n"
"} Constraint4;\n"
"typedef struct\n"
"{\n"
" int m_nConstraints;\n"
" int m_start;\n"
" int m_batchIdx;\n"
" int m_nSplit;\n"
"// int m_paddings[1];\n"
"} ConstBuffer;\n"
"typedef struct\n"
"{\n"
" int m_solveFriction;\n"
" int m_maxBatch; // long batch really kills the performance\n"
" int m_batchIdx;\n"
" int m_nSplit;\n"
"// int m_paddings[1];\n"
"} ConstBufferBatchSolve;\n"
"void setLinearAndAngular( float4 n, float4 r0, float4 r1, float4* linear, float4* angular0, float4* angular1);\n"
"void setLinearAndAngular( float4 n, float4 r0, float4 r1, float4* linear, float4* angular0, float4* angular1)\n"
"{\n"
" *linear = mymake_float4(-n.xyz,0.f);\n"
" *angular0 = -cross3(r0, n);\n"
" *angular1 = cross3(r1, n);\n"
"}\n"
"float calcRelVel( float4 l0, float4 l1, float4 a0, float4 a1, float4 linVel0, float4 angVel0, float4 linVel1, float4 angVel1 );\n"
"float calcRelVel( float4 l0, float4 l1, float4 a0, float4 a1, float4 linVel0, float4 angVel0, float4 linVel1, float4 angVel1 )\n"
"{\n"
" return dot3F4(l0, linVel0) + dot3F4(a0, angVel0) + dot3F4(l1, linVel1) + dot3F4(a1, angVel1);\n"
"}\n"
"float calcJacCoeff(const float4 linear0, const float4 linear1, const float4 angular0, const float4 angular1,\n"
" float invMass0, const Matrix3x3* invInertia0, float invMass1, const Matrix3x3* invInertia1);\n"
"float calcJacCoeff(const float4 linear0, const float4 linear1, const float4 angular0, const float4 angular1,\n"
" float invMass0, const Matrix3x3* invInertia0, float invMass1, const Matrix3x3* invInertia1)\n"
"{\n"
" // linear0,1 are normlized\n"
" float jmj0 = invMass0;//dot3F4(linear0, linear0)*invMass0;\n"
" float jmj1 = dot3F4(mtMul3(angular0,*invInertia0), angular0);\n"
" float jmj2 = invMass1;//dot3F4(linear1, linear1)*invMass1;\n"
" float jmj3 = dot3F4(mtMul3(angular1,*invInertia1), angular1);\n"
" return -1.f/(jmj0+jmj1+jmj2+jmj3);\n"
"}\n"
"void solveContact(__global Constraint4* cs,\n"
" float4 posA, float4* linVelA, float4* angVelA, float invMassA, Matrix3x3 invInertiaA,\n"
" float4 posB, float4* linVelB, float4* angVelB, float invMassB, Matrix3x3 invInertiaB);\n"
"void solveContact(__global Constraint4* cs,\n"
" float4 posA, float4* linVelA, float4* angVelA, float invMassA, Matrix3x3 invInertiaA,\n"
" float4 posB, float4* linVelB, float4* angVelB, float invMassB, Matrix3x3 invInertiaB)\n"
"{\n"
" float minRambdaDt = 0;\n"
" float maxRambdaDt = FLT_MAX;\n"
" for(int ic=0; ic<4; ic++)\n"
" {\n"
" if( cs->m_jacCoeffInv[ic] == 0.f ) continue;\n"
" float4 angular0, angular1, linear;\n"
" float4 r0 = cs->m_worldPos[ic] - posA;\n"
" float4 r1 = cs->m_worldPos[ic] - posB;\n"
" setLinearAndAngular( -cs->m_linear, r0, r1, &linear, &angular0, &angular1 );\n"
" float rambdaDt = calcRelVel( cs->m_linear, -cs->m_linear, angular0, angular1, \n"
" *linVelA, *angVelA, *linVelB, *angVelB ) + cs->m_b[ic];\n"
" rambdaDt *= cs->m_jacCoeffInv[ic];\n"
" {\n"
" float prevSum = cs->m_appliedRambdaDt[ic];\n"
" float updated = prevSum;\n"
" updated += rambdaDt;\n"
" updated = max2( updated, minRambdaDt );\n"
" updated = min2( updated, maxRambdaDt );\n"
" rambdaDt = updated - prevSum;\n"
" cs->m_appliedRambdaDt[ic] = updated;\n"
" }\n"
" float4 linImp0 = invMassA*linear*rambdaDt;\n"
" float4 linImp1 = invMassB*(-linear)*rambdaDt;\n"
" float4 angImp0 = mtMul1(invInertiaA, angular0)*rambdaDt;\n"
" float4 angImp1 = mtMul1(invInertiaB, angular1)*rambdaDt;\n"
" *linVelA += linImp0;\n"
" *angVelA += angImp0;\n"
" *linVelB += linImp1;\n"
" *angVelB += angImp1;\n"
" }\n"
"}\n"
"void btPlaneSpace1 (const float4* n, float4* p, float4* q);\n"
" void btPlaneSpace1 (const float4* n, float4* p, float4* q)\n"
"{\n"
" if (fabs(n[0].z) > 0.70710678f) {\n"
" // choose p in y-z plane\n"
" float a = n[0].y*n[0].y + n[0].z*n[0].z;\n"
" float k = 1.f/sqrt(a);\n"
" p[0].x = 0;\n"
" p[0].y = -n[0].z*k;\n"
" p[0].z = n[0].y*k;\n"
" // set q = n x p\n"
" q[0].x = a*k;\n"
" q[0].y = -n[0].x*p[0].z;\n"
" q[0].z = n[0].x*p[0].y;\n"
" }\n"
" else {\n"
" // choose p in x-y plane\n"
" float a = n[0].x*n[0].x + n[0].y*n[0].y;\n"
" float k = 1.f/sqrt(a);\n"
" p[0].x = -n[0].y*k;\n"
" p[0].y = n[0].x*k;\n"
" p[0].z = 0;\n"
" // set q = n x p\n"
" q[0].x = -n[0].z*p[0].y;\n"
" q[0].y = n[0].z*p[0].x;\n"
" q[0].z = a*k;\n"
" }\n"
"}\n"
"void solveContactConstraint(__global Body* gBodies, __global Shape* gShapes, __global Constraint4* ldsCs);\n"
"void solveContactConstraint(__global Body* gBodies, __global Shape* gShapes, __global Constraint4* ldsCs)\n"
"{\n"
" //float frictionCoeff = ldsCs[0].m_linear.w;\n"
" int aIdx = ldsCs[0].m_bodyA;\n"
" int bIdx = ldsCs[0].m_bodyB;\n"
" float4 posA = gBodies[aIdx].m_pos;\n"
" float4 linVelA = gBodies[aIdx].m_linVel;\n"
" float4 angVelA = gBodies[aIdx].m_angVel;\n"
" float invMassA = gBodies[aIdx].m_invMass;\n"
" Matrix3x3 invInertiaA = gShapes[aIdx].m_invInertia;\n"
" float4 posB = gBodies[bIdx].m_pos;\n"
" float4 linVelB = gBodies[bIdx].m_linVel;\n"
" float4 angVelB = gBodies[bIdx].m_angVel;\n"
" float invMassB = gBodies[bIdx].m_invMass;\n"
" Matrix3x3 invInertiaB = gShapes[bIdx].m_invInertia;\n"
" solveContact( ldsCs, posA, &linVelA, &angVelA, invMassA, invInertiaA,\n"
" posB, &linVelB, &angVelB, invMassB, invInertiaB );\n"
" if (gBodies[aIdx].m_invMass)\n"
" {\n"
" gBodies[aIdx].m_linVel = linVelA;\n"
" gBodies[aIdx].m_angVel = angVelA;\n"
" } else\n"
" {\n"
" gBodies[aIdx].m_linVel = mymake_float4(0,0,0,0);\n"
" gBodies[aIdx].m_angVel = mymake_float4(0,0,0,0);\n"
" \n"
" }\n"
" if (gBodies[bIdx].m_invMass)\n"
" {\n"
" gBodies[bIdx].m_linVel = linVelB;\n"
" gBodies[bIdx].m_angVel = angVelB;\n"
" } else\n"
" {\n"
" gBodies[bIdx].m_linVel = mymake_float4(0,0,0,0);\n"
" gBodies[bIdx].m_angVel = mymake_float4(0,0,0,0);\n"
" \n"
" }\n"
"}\n"
"typedef struct \n"
"{\n"
" int m_valInt0;\n"
" int m_valInt1;\n"
" int m_valInt2;\n"
" int m_valInt3;\n"
" float m_val0;\n"
" float m_val1;\n"
" float m_val2;\n"
" float m_val3;\n"
"} SolverDebugInfo;\n"
"__kernel\n"
"__attribute__((reqd_work_group_size(WG_SIZE,1,1)))\n"
"void BatchSolveKernelContact(__global Body* gBodies,\n"
" __global Shape* gShapes,\n"
" __global Constraint4* gConstraints,\n"
" __global int* gN,\n"
" __global int* gOffsets,\n"
" __global int* batchSizes,\n"
" int maxBatch1,\n"
" int cellBatch,\n"
" int4 nSplit\n"
" )\n"
"{\n"
" //__local int ldsBatchIdx[WG_SIZE+1];\n"
" __local int ldsCurBatch;\n"
" __local int ldsNextBatch;\n"
" __local int ldsStart;\n"
" int lIdx = GET_LOCAL_IDX;\n"
" int wgIdx = GET_GROUP_IDX;\n"
"// int gIdx = GET_GLOBAL_IDX;\n"
"// debugInfo[gIdx].m_valInt0 = gIdx;\n"
" //debugInfo[gIdx].m_valInt1 = GET_GROUP_SIZE;\n"
" \n"
" \n"
" int zIdx = (wgIdx/((nSplit.x*nSplit.y)/4))*2+((cellBatch&4)>>2);\n"
" int remain= (wgIdx%((nSplit.x*nSplit.y)/4));\n"
" int yIdx = (remain/(nSplit.x/2))*2 + ((cellBatch&2)>>1);\n"
" int xIdx = (remain%(nSplit.x/2))*2 + (cellBatch&1);\n"
" int cellIdx = xIdx+yIdx*nSplit.x+zIdx*(nSplit.x*nSplit.y);\n"
" //int xIdx = (wgIdx/(nSplit/2))*2 + (bIdx&1);\n"
" //int yIdx = (wgIdx%(nSplit/2))*2 + (bIdx>>1);\n"
" //int cellIdx = xIdx+yIdx*nSplit;\n"
" \n"
" if( gN[cellIdx] == 0 ) \n"
" return;\n"
" int maxBatch = batchSizes[cellIdx];\n"
" \n"
" \n"
" const int start = gOffsets[cellIdx];\n"
" const int end = start + gN[cellIdx];\n"
" \n"
" \n"
" \n"
" if( lIdx == 0 )\n"
" {\n"
" ldsCurBatch = 0;\n"
" ldsNextBatch = 0;\n"
" ldsStart = start;\n"
" }\n"
" GROUP_LDS_BARRIER;\n"
" int idx=ldsStart+lIdx;\n"
" while (ldsCurBatch < maxBatch)\n"
" {\n"
" for(; idx<end; )\n"
" {\n"
" if (gConstraints[idx].m_batchIdx == ldsCurBatch)\n"
" {\n"
" solveContactConstraint( gBodies, gShapes, &gConstraints[idx] );\n"
" idx+=64;\n"
" } else\n"
" {\n"
" break;\n"
" }\n"
" }\n"
" GROUP_LDS_BARRIER;\n"
" \n"
" if( lIdx == 0 )\n"
" {\n"
" ldsCurBatch++;\n"
" }\n"
" GROUP_LDS_BARRIER;\n"
" }\n"
" \n"
" \n"
"}\n"
"__kernel void solveSingleContactKernel(__global Body* gBodies,\n"
" __global Shape* gShapes,\n"
" __global Constraint4* gConstraints,\n"
" int cellIdx,\n"
" int batchOffset,\n"
" int numConstraintsInBatch\n"
" )\n"
"{\n"
" int index = get_global_id(0);\n"
" if (index < numConstraintsInBatch)\n"
" {\n"
" int idx=batchOffset+index;\n"
" solveContactConstraint( gBodies, gShapes, &gConstraints[idx] );\n"
" } \n"
"}\n"
;
static const char* solveContactCL =
"/*\n"
"Copyright (c) 2012 Advanced Micro Devices, Inc. \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"
"//Originally written by Takahiro Harada\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"
"typedef unsigned int u32;\n"
"typedef unsigned short u16;\n"
"typedef unsigned char u8;\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 SELECT_UINT4( b, a, condition ) select( b,a,condition )\n"
"#define mymake_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"
"#define max2 max\n"
"#define min2 min\n"
"///////////////////////////////////////\n"
"// Vector\n"
"///////////////////////////////////////\n"
"__inline\n"
"float4 fastNormalize4(float4 v)\n"
"{\n"
" return fast_normalize(v);\n"
"}\n"
"__inline\n"
"float4 cross3(float4 a, float4 b)\n"
"{\n"
" return cross(a,b);\n"
"}\n"
"__inline\n"
"float dot3F4(float4 a, float4 b)\n"
"{\n"
" float4 a1 = mymake_float4(a.xyz,0.f);\n"
" float4 b1 = mymake_float4(b.xyz,0.f);\n"
" return dot(a1, b1);\n"
"}\n"
"__inline\n"
"float4 normalize3(const float4 a)\n"
"{\n"
" float4 n = mymake_float4(a.x, a.y, a.z, 0.f);\n"
" return fastNormalize4( n );\n"
"// float length = sqrtf(dot3F4(a, a));\n"
"// return 1.f/length * a;\n"
"}\n"
"///////////////////////////////////////\n"
"// Matrix3x3\n"
"///////////////////////////////////////\n"
"typedef struct\n"
"{\n"
" float4 m_row[3];\n"
"}Matrix3x3;\n"
"__inline\n"
"float4 mtMul1(Matrix3x3 a, float4 b);\n"
"__inline\n"
"float4 mtMul3(float4 a, Matrix3x3 b);\n"
"__inline\n"
"float4 mtMul1(Matrix3x3 a, float4 b)\n"
"{\n"
" float4 ans;\n"
" ans.x = dot3F4( a.m_row[0], b );\n"
" ans.y = dot3F4( a.m_row[1], b );\n"
" ans.z = dot3F4( a.m_row[2], b );\n"
" ans.w = 0.f;\n"
" return ans;\n"
"}\n"
"__inline\n"
"float4 mtMul3(float4 a, Matrix3x3 b)\n"
"{\n"
" float4 colx = mymake_float4(b.m_row[0].x, b.m_row[1].x, b.m_row[2].x, 0);\n"
" float4 coly = mymake_float4(b.m_row[0].y, b.m_row[1].y, b.m_row[2].y, 0);\n"
" float4 colz = mymake_float4(b.m_row[0].z, b.m_row[1].z, b.m_row[2].z, 0);\n"
" float4 ans;\n"
" ans.x = dot3F4( a, colx );\n"
" ans.y = dot3F4( a, coly );\n"
" ans.z = dot3F4( a, colz );\n"
" return ans;\n"
"}\n"
"///////////////////////////////////////\n"
"// Quaternion\n"
"///////////////////////////////////////\n"
"typedef float4 Quaternion;\n"
"#define WG_SIZE 64\n"
"typedef struct\n"
"{\n"
" float4 m_pos;\n"
" Quaternion m_quat;\n"
" float4 m_linVel;\n"
" float4 m_angVel;\n"
" u32 m_shapeIdx;\n"
" float m_invMass;\n"
" float m_restituitionCoeff;\n"
" float m_frictionCoeff;\n"
"} Body;\n"
"typedef struct\n"
"{\n"
" Matrix3x3 m_invInertia;\n"
" Matrix3x3 m_initInvInertia;\n"
"} Shape;\n"
"typedef struct\n"
"{\n"
" float4 m_linear;\n"
" float4 m_worldPos[4];\n"
" float4 m_center; \n"
" float m_jacCoeffInv[4];\n"
" float m_b[4];\n"
" float m_appliedRambdaDt[4];\n"
" float m_fJacCoeffInv[2]; \n"
" float m_fAppliedRambdaDt[2]; \n"
" u32 m_bodyA;\n"
" u32 m_bodyB;\n"
" int m_batchIdx;\n"
" u32 m_paddings[1];\n"
"} Constraint4;\n"
"typedef struct\n"
"{\n"
" int m_nConstraints;\n"
" int m_start;\n"
" int m_batchIdx;\n"
" int m_nSplit;\n"
"// int m_paddings[1];\n"
"} ConstBuffer;\n"
"typedef struct\n"
"{\n"
" int m_solveFriction;\n"
" int m_maxBatch; // long batch really kills the performance\n"
" int m_batchIdx;\n"
" int m_nSplit;\n"
"// int m_paddings[1];\n"
"} ConstBufferBatchSolve;\n"
"void setLinearAndAngular( float4 n, float4 r0, float4 r1, float4* linear, float4* angular0, float4* angular1);\n"
"void setLinearAndAngular( float4 n, float4 r0, float4 r1, float4* linear, float4* angular0, float4* angular1)\n"
"{\n"
" *linear = mymake_float4(-n.xyz,0.f);\n"
" *angular0 = -cross3(r0, n);\n"
" *angular1 = cross3(r1, n);\n"
"}\n"
"float calcRelVel( float4 l0, float4 l1, float4 a0, float4 a1, float4 linVel0, float4 angVel0, float4 linVel1, float4 angVel1 );\n"
"float calcRelVel( float4 l0, float4 l1, float4 a0, float4 a1, float4 linVel0, float4 angVel0, float4 linVel1, float4 angVel1 )\n"
"{\n"
" return dot3F4(l0, linVel0) + dot3F4(a0, angVel0) + dot3F4(l1, linVel1) + dot3F4(a1, angVel1);\n"
"}\n"
"float calcJacCoeff(const float4 linear0, const float4 linear1, const float4 angular0, const float4 angular1,\n"
" float invMass0, const Matrix3x3* invInertia0, float invMass1, const Matrix3x3* invInertia1);\n"
"float calcJacCoeff(const float4 linear0, const float4 linear1, const float4 angular0, const float4 angular1,\n"
" float invMass0, const Matrix3x3* invInertia0, float invMass1, const Matrix3x3* invInertia1)\n"
"{\n"
" // linear0,1 are normlized\n"
" float jmj0 = invMass0;//dot3F4(linear0, linear0)*invMass0;\n"
" float jmj1 = dot3F4(mtMul3(angular0,*invInertia0), angular0);\n"
" float jmj2 = invMass1;//dot3F4(linear1, linear1)*invMass1;\n"
" float jmj3 = dot3F4(mtMul3(angular1,*invInertia1), angular1);\n"
" return -1.f/(jmj0+jmj1+jmj2+jmj3);\n"
"}\n"
"void solveContact(__global Constraint4* cs,\n"
" float4 posA, float4* linVelA, float4* angVelA, float invMassA, Matrix3x3 invInertiaA,\n"
" float4 posB, float4* linVelB, float4* angVelB, float invMassB, Matrix3x3 invInertiaB);\n"
"void solveContact(__global Constraint4* cs,\n"
" float4 posA, float4* linVelA, float4* angVelA, float invMassA, Matrix3x3 invInertiaA,\n"
" float4 posB, float4* linVelB, float4* angVelB, float invMassB, Matrix3x3 invInertiaB)\n"
"{\n"
" float minRambdaDt = 0;\n"
" float maxRambdaDt = FLT_MAX;\n"
" for(int ic=0; ic<4; ic++)\n"
" {\n"
" if( cs->m_jacCoeffInv[ic] == 0.f ) continue;\n"
" float4 angular0, angular1, linear;\n"
" float4 r0 = cs->m_worldPos[ic] - posA;\n"
" float4 r1 = cs->m_worldPos[ic] - posB;\n"
" setLinearAndAngular( -cs->m_linear, r0, r1, &linear, &angular0, &angular1 );\n"
" float rambdaDt = calcRelVel( cs->m_linear, -cs->m_linear, angular0, angular1, \n"
" *linVelA, *angVelA, *linVelB, *angVelB ) + cs->m_b[ic];\n"
" rambdaDt *= cs->m_jacCoeffInv[ic];\n"
" {\n"
" float prevSum = cs->m_appliedRambdaDt[ic];\n"
" float updated = prevSum;\n"
" updated += rambdaDt;\n"
" updated = max2( updated, minRambdaDt );\n"
" updated = min2( updated, maxRambdaDt );\n"
" rambdaDt = updated - prevSum;\n"
" cs->m_appliedRambdaDt[ic] = updated;\n"
" }\n"
" float4 linImp0 = invMassA*linear*rambdaDt;\n"
" float4 linImp1 = invMassB*(-linear)*rambdaDt;\n"
" float4 angImp0 = mtMul1(invInertiaA, angular0)*rambdaDt;\n"
" float4 angImp1 = mtMul1(invInertiaB, angular1)*rambdaDt;\n"
" *linVelA += linImp0;\n"
" *angVelA += angImp0;\n"
" *linVelB += linImp1;\n"
" *angVelB += angImp1;\n"
" }\n"
"}\n"
"void btPlaneSpace1 (const float4* n, float4* p, float4* q);\n"
" void btPlaneSpace1 (const float4* n, float4* p, float4* q)\n"
"{\n"
" if (fabs(n[0].z) > 0.70710678f) {\n"
" // choose p in y-z plane\n"
" float a = n[0].y*n[0].y + n[0].z*n[0].z;\n"
" float k = 1.f/sqrt(a);\n"
" p[0].x = 0;\n"
" p[0].y = -n[0].z*k;\n"
" p[0].z = n[0].y*k;\n"
" // set q = n x p\n"
" q[0].x = a*k;\n"
" q[0].y = -n[0].x*p[0].z;\n"
" q[0].z = n[0].x*p[0].y;\n"
" }\n"
" else {\n"
" // choose p in x-y plane\n"
" float a = n[0].x*n[0].x + n[0].y*n[0].y;\n"
" float k = 1.f/sqrt(a);\n"
" p[0].x = -n[0].y*k;\n"
" p[0].y = n[0].x*k;\n"
" p[0].z = 0;\n"
" // set q = n x p\n"
" q[0].x = -n[0].z*p[0].y;\n"
" q[0].y = n[0].z*p[0].x;\n"
" q[0].z = a*k;\n"
" }\n"
"}\n"
"void solveContactConstraint(__global Body* gBodies, __global Shape* gShapes, __global Constraint4* ldsCs);\n"
"void solveContactConstraint(__global Body* gBodies, __global Shape* gShapes, __global Constraint4* ldsCs)\n"
"{\n"
" //float frictionCoeff = ldsCs[0].m_linear.w;\n"
" int aIdx = ldsCs[0].m_bodyA;\n"
" int bIdx = ldsCs[0].m_bodyB;\n"
" float4 posA = gBodies[aIdx].m_pos;\n"
" float4 linVelA = gBodies[aIdx].m_linVel;\n"
" float4 angVelA = gBodies[aIdx].m_angVel;\n"
" float invMassA = gBodies[aIdx].m_invMass;\n"
" Matrix3x3 invInertiaA = gShapes[aIdx].m_invInertia;\n"
" float4 posB = gBodies[bIdx].m_pos;\n"
" float4 linVelB = gBodies[bIdx].m_linVel;\n"
" float4 angVelB = gBodies[bIdx].m_angVel;\n"
" float invMassB = gBodies[bIdx].m_invMass;\n"
" Matrix3x3 invInertiaB = gShapes[bIdx].m_invInertia;\n"
" solveContact( ldsCs, posA, &linVelA, &angVelA, invMassA, invInertiaA,\n"
" posB, &linVelB, &angVelB, invMassB, invInertiaB );\n"
" if (gBodies[aIdx].m_invMass)\n"
" {\n"
" gBodies[aIdx].m_linVel = linVelA;\n"
" gBodies[aIdx].m_angVel = angVelA;\n"
" } else\n"
" {\n"
" gBodies[aIdx].m_linVel = mymake_float4(0,0,0,0);\n"
" gBodies[aIdx].m_angVel = mymake_float4(0,0,0,0);\n"
" \n"
" }\n"
" if (gBodies[bIdx].m_invMass)\n"
" {\n"
" gBodies[bIdx].m_linVel = linVelB;\n"
" gBodies[bIdx].m_angVel = angVelB;\n"
" } else\n"
" {\n"
" gBodies[bIdx].m_linVel = mymake_float4(0,0,0,0);\n"
" gBodies[bIdx].m_angVel = mymake_float4(0,0,0,0);\n"
" \n"
" }\n"
"}\n"
"typedef struct \n"
"{\n"
" int m_valInt0;\n"
" int m_valInt1;\n"
" int m_valInt2;\n"
" int m_valInt3;\n"
" float m_val0;\n"
" float m_val1;\n"
" float m_val2;\n"
" float m_val3;\n"
"} SolverDebugInfo;\n"
"__kernel\n"
"__attribute__((reqd_work_group_size(WG_SIZE,1,1)))\n"
"void BatchSolveKernelContact(__global Body* gBodies,\n"
" __global Shape* gShapes,\n"
" __global Constraint4* gConstraints,\n"
" __global int* gN,\n"
" __global int* gOffsets,\n"
" __global int* batchSizes,\n"
" int maxBatch1,\n"
" int cellBatch,\n"
" int4 nSplit\n"
" )\n"
"{\n"
" //__local int ldsBatchIdx[WG_SIZE+1];\n"
" __local int ldsCurBatch;\n"
" __local int ldsNextBatch;\n"
" __local int ldsStart;\n"
" int lIdx = GET_LOCAL_IDX;\n"
" int wgIdx = GET_GROUP_IDX;\n"
"// int gIdx = GET_GLOBAL_IDX;\n"
"// debugInfo[gIdx].m_valInt0 = gIdx;\n"
" //debugInfo[gIdx].m_valInt1 = GET_GROUP_SIZE;\n"
" \n"
" \n"
" int zIdx = (wgIdx/((nSplit.x*nSplit.y)/4))*2+((cellBatch&4)>>2);\n"
" int remain= (wgIdx%((nSplit.x*nSplit.y)/4));\n"
" int yIdx = (remain/(nSplit.x/2))*2 + ((cellBatch&2)>>1);\n"
" int xIdx = (remain%(nSplit.x/2))*2 + (cellBatch&1);\n"
" int cellIdx = xIdx+yIdx*nSplit.x+zIdx*(nSplit.x*nSplit.y);\n"
" //int xIdx = (wgIdx/(nSplit/2))*2 + (bIdx&1);\n"
" //int yIdx = (wgIdx%(nSplit/2))*2 + (bIdx>>1);\n"
" //int cellIdx = xIdx+yIdx*nSplit;\n"
" \n"
" if( gN[cellIdx] == 0 ) \n"
" return;\n"
" int maxBatch = batchSizes[cellIdx];\n"
" \n"
" \n"
" const int start = gOffsets[cellIdx];\n"
" const int end = start + gN[cellIdx];\n"
" \n"
" \n"
" \n"
" if( lIdx == 0 )\n"
" {\n"
" ldsCurBatch = 0;\n"
" ldsNextBatch = 0;\n"
" ldsStart = start;\n"
" }\n"
" GROUP_LDS_BARRIER;\n"
" int idx=ldsStart+lIdx;\n"
" while (ldsCurBatch < maxBatch)\n"
" {\n"
" for(; idx<end; )\n"
" {\n"
" if (gConstraints[idx].m_batchIdx == ldsCurBatch)\n"
" {\n"
" solveContactConstraint( gBodies, gShapes, &gConstraints[idx] );\n"
" idx+=64;\n"
" } else\n"
" {\n"
" break;\n"
" }\n"
" }\n"
" GROUP_LDS_BARRIER;\n"
" \n"
" if( lIdx == 0 )\n"
" {\n"
" ldsCurBatch++;\n"
" }\n"
" GROUP_LDS_BARRIER;\n"
" }\n"
" \n"
" \n"
"}\n"
"__kernel void solveSingleContactKernel(__global Body* gBodies,\n"
" __global Shape* gShapes,\n"
" __global Constraint4* gConstraints,\n"
" int cellIdx,\n"
" int batchOffset,\n"
" int numConstraintsInBatch\n"
" )\n"
"{\n"
" int index = get_global_id(0);\n"
" if (index < numConstraintsInBatch)\n"
" {\n"
" int idx=batchOffset+index;\n"
" solveContactConstraint( gBodies, gShapes, &gConstraints[idx] );\n"
" } \n"
"}\n";

839
src/Bullet3OpenCL/RigidBody/kernels/solveFriction.h
View File

@@ -1,421 +1,420 @@
//this file is autogenerated using stringify.bat (premake --stringify) in the build folder of this project
static const char* solveFrictionCL= \
"/*\n"
"Copyright (c) 2012 Advanced Micro Devices, Inc. \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"
"//Originally written by Takahiro Harada\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"
"typedef unsigned int u32;\n"
"typedef unsigned short u16;\n"
"typedef unsigned char u8;\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 SELECT_UINT4( b, a, condition ) select( b,a,condition )\n"
"#define mymake_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"
"#define max2 max\n"
"#define min2 min\n"
"///////////////////////////////////////\n"
"// Vector\n"
"///////////////////////////////////////\n"
"__inline\n"
"float4 fastNormalize4(float4 v)\n"
"{\n"
" return fast_normalize(v);\n"
"}\n"
"__inline\n"
"float4 cross3(float4 a, float4 b)\n"
"{\n"
" return cross(a,b);\n"
"}\n"
"__inline\n"
"float dot3F4(float4 a, float4 b)\n"
"{\n"
" float4 a1 = mymake_float4(a.xyz,0.f);\n"
" float4 b1 = mymake_float4(b.xyz,0.f);\n"
" return dot(a1, b1);\n"
"}\n"
"__inline\n"
"float4 normalize3(const float4 a)\n"
"{\n"
" float4 n = mymake_float4(a.x, a.y, a.z, 0.f);\n"
" return fastNormalize4( n );\n"
"// float length = sqrtf(dot3F4(a, a));\n"
"// return 1.f/length * a;\n"
"}\n"
"///////////////////////////////////////\n"
"// Matrix3x3\n"
"///////////////////////////////////////\n"
"typedef struct\n"
"{\n"
" float4 m_row[3];\n"
"}Matrix3x3;\n"
"__inline\n"
"float4 mtMul1(Matrix3x3 a, float4 b);\n"
"__inline\n"
"float4 mtMul3(float4 a, Matrix3x3 b);\n"
"__inline\n"
"float4 mtMul1(Matrix3x3 a, float4 b)\n"
"{\n"
" float4 ans;\n"
" ans.x = dot3F4( a.m_row[0], b );\n"
" ans.y = dot3F4( a.m_row[1], b );\n"
" ans.z = dot3F4( a.m_row[2], b );\n"
" ans.w = 0.f;\n"
" return ans;\n"
"}\n"
"__inline\n"
"float4 mtMul3(float4 a, Matrix3x3 b)\n"
"{\n"
" float4 colx = mymake_float4(b.m_row[0].x, b.m_row[1].x, b.m_row[2].x, 0);\n"
" float4 coly = mymake_float4(b.m_row[0].y, b.m_row[1].y, b.m_row[2].y, 0);\n"
" float4 colz = mymake_float4(b.m_row[0].z, b.m_row[1].z, b.m_row[2].z, 0);\n"
" float4 ans;\n"
" ans.x = dot3F4( a, colx );\n"
" ans.y = dot3F4( a, coly );\n"
" ans.z = dot3F4( a, colz );\n"
" return ans;\n"
"}\n"
"///////////////////////////////////////\n"
"// Quaternion\n"
"///////////////////////////////////////\n"
"typedef float4 Quaternion;\n"
"#define WG_SIZE 64\n"
"typedef struct\n"
"{\n"
" float4 m_pos;\n"
" Quaternion m_quat;\n"
" float4 m_linVel;\n"
" float4 m_angVel;\n"
" u32 m_shapeIdx;\n"
" float m_invMass;\n"
" float m_restituitionCoeff;\n"
" float m_frictionCoeff;\n"
"} Body;\n"
"typedef struct\n"
"{\n"
" Matrix3x3 m_invInertia;\n"
" Matrix3x3 m_initInvInertia;\n"
"} Shape;\n"
"typedef struct\n"
"{\n"
" float4 m_linear;\n"
" float4 m_worldPos[4];\n"
" float4 m_center; \n"
" float m_jacCoeffInv[4];\n"
" float m_b[4];\n"
" float m_appliedRambdaDt[4];\n"
" float m_fJacCoeffInv[2]; \n"
" float m_fAppliedRambdaDt[2]; \n"
" u32 m_bodyA;\n"
" u32 m_bodyB;\n"
" int m_batchIdx;\n"
" u32 m_paddings[1];\n"
"} Constraint4;\n"
"typedef struct\n"
"{\n"
" int m_nConstraints;\n"
" int m_start;\n"
" int m_batchIdx;\n"
" int m_nSplit;\n"
"// int m_paddings[1];\n"
"} ConstBuffer;\n"
"typedef struct\n"
"{\n"
" int m_solveFriction;\n"
" int m_maxBatch; // long batch really kills the performance\n"
" int m_batchIdx;\n"
" int m_nSplit;\n"
"// int m_paddings[1];\n"
"} ConstBufferBatchSolve;\n"
"void setLinearAndAngular( float4 n, float4 r0, float4 r1, float4* linear, float4* angular0, float4* angular1);\n"
"void setLinearAndAngular( float4 n, float4 r0, float4 r1, float4* linear, float4* angular0, float4* angular1)\n"
"{\n"
" *linear = mymake_float4(-n.xyz,0.f);\n"
" *angular0 = -cross3(r0, n);\n"
" *angular1 = cross3(r1, n);\n"
"}\n"
"float calcRelVel( float4 l0, float4 l1, float4 a0, float4 a1, float4 linVel0, float4 angVel0, float4 linVel1, float4 angVel1 );\n"
"float calcRelVel( float4 l0, float4 l1, float4 a0, float4 a1, float4 linVel0, float4 angVel0, float4 linVel1, float4 angVel1 )\n"
"{\n"
" return dot3F4(l0, linVel0) + dot3F4(a0, angVel0) + dot3F4(l1, linVel1) + dot3F4(a1, angVel1);\n"
"}\n"
"float calcJacCoeff(const float4 linear0, const float4 linear1, const float4 angular0, const float4 angular1,\n"
" float invMass0, const Matrix3x3* invInertia0, float invMass1, const Matrix3x3* invInertia1);\n"
"float calcJacCoeff(const float4 linear0, const float4 linear1, const float4 angular0, const float4 angular1,\n"
" float invMass0, const Matrix3x3* invInertia0, float invMass1, const Matrix3x3* invInertia1)\n"
"{\n"
" // linear0,1 are normlized\n"
" float jmj0 = invMass0;//dot3F4(linear0, linear0)*invMass0;\n"
" float jmj1 = dot3F4(mtMul3(angular0,*invInertia0), angular0);\n"
" float jmj2 = invMass1;//dot3F4(linear1, linear1)*invMass1;\n"
" float jmj3 = dot3F4(mtMul3(angular1,*invInertia1), angular1);\n"
" return -1.f/(jmj0+jmj1+jmj2+jmj3);\n"
"}\n"
"void btPlaneSpace1 (const float4* n, float4* p, float4* q);\n"
" void btPlaneSpace1 (const float4* n, float4* p, float4* q)\n"
"{\n"
" if (fabs(n[0].z) > 0.70710678f) {\n"
" // choose p in y-z plane\n"
" float a = n[0].y*n[0].y + n[0].z*n[0].z;\n"
" float k = 1.f/sqrt(a);\n"
" p[0].x = 0;\n"
" p[0].y = -n[0].z*k;\n"
" p[0].z = n[0].y*k;\n"
" // set q = n x p\n"
" q[0].x = a*k;\n"
" q[0].y = -n[0].x*p[0].z;\n"
" q[0].z = n[0].x*p[0].y;\n"
" }\n"
" else {\n"
" // choose p in x-y plane\n"
" float a = n[0].x*n[0].x + n[0].y*n[0].y;\n"
" float k = 1.f/sqrt(a);\n"
" p[0].x = -n[0].y*k;\n"
" p[0].y = n[0].x*k;\n"
" p[0].z = 0;\n"
" // set q = n x p\n"
" q[0].x = -n[0].z*p[0].y;\n"
" q[0].y = n[0].z*p[0].x;\n"
" q[0].z = a*k;\n"
" }\n"
"}\n"
"void solveFrictionConstraint(__global Body* gBodies, __global Shape* gShapes, __global Constraint4* ldsCs);\n"
"void solveFrictionConstraint(__global Body* gBodies, __global Shape* gShapes, __global Constraint4* ldsCs)\n"
"{\n"
" float frictionCoeff = ldsCs[0].m_linear.w;\n"
" int aIdx = ldsCs[0].m_bodyA;\n"
" int bIdx = ldsCs[0].m_bodyB;\n"
" float4 posA = gBodies[aIdx].m_pos;\n"
" float4 linVelA = gBodies[aIdx].m_linVel;\n"
" float4 angVelA = gBodies[aIdx].m_angVel;\n"
" float invMassA = gBodies[aIdx].m_invMass;\n"
" Matrix3x3 invInertiaA = gShapes[aIdx].m_invInertia;\n"
" float4 posB = gBodies[bIdx].m_pos;\n"
" float4 linVelB = gBodies[bIdx].m_linVel;\n"
" float4 angVelB = gBodies[bIdx].m_angVel;\n"
" float invMassB = gBodies[bIdx].m_invMass;\n"
" Matrix3x3 invInertiaB = gShapes[bIdx].m_invInertia;\n"
" \n"
" {\n"
" float maxRambdaDt[4] = {FLT_MAX,FLT_MAX,FLT_MAX,FLT_MAX};\n"
" float minRambdaDt[4] = {0.f,0.f,0.f,0.f};\n"
" float sum = 0;\n"
" for(int j=0; j<4; j++)\n"
" {\n"
" sum +=ldsCs[0].m_appliedRambdaDt[j];\n"
" }\n"
" frictionCoeff = 0.7f;\n"
" for(int j=0; j<4; j++)\n"
" {\n"
" maxRambdaDt[j] = frictionCoeff*sum;\n"
" minRambdaDt[j] = -maxRambdaDt[j];\n"
" }\n"
" \n"
"// solveFriction( ldsCs, posA, &linVelA, &angVelA, invMassA, invInertiaA,\n"
"// posB, &linVelB, &angVelB, invMassB, invInertiaB, maxRambdaDt, minRambdaDt );\n"
" \n"
" \n"
" {\n"
" \n"
" __global Constraint4* cs = ldsCs;\n"
" \n"
" if( cs->m_fJacCoeffInv[0] == 0 && cs->m_fJacCoeffInv[0] == 0 ) return;\n"
" const float4 center = cs->m_center;\n"
" \n"
" float4 n = -cs->m_linear;\n"
" \n"
" float4 tangent[2];\n"
" btPlaneSpace1(&n,&tangent[0],&tangent[1]);\n"
" float4 angular0, angular1, linear;\n"
" float4 r0 = center - posA;\n"
" float4 r1 = center - posB;\n"
" for(int i=0; i<2; i++)\n"
" {\n"
" setLinearAndAngular( tangent[i], r0, r1, &linear, &angular0, &angular1 );\n"
" float rambdaDt = calcRelVel(linear, -linear, angular0, angular1,\n"
" linVelA, angVelA, linVelB, angVelB );\n"
" rambdaDt *= cs->m_fJacCoeffInv[i];\n"
" \n"
" {\n"
" float prevSum = cs->m_fAppliedRambdaDt[i];\n"
" float updated = prevSum;\n"
" updated += rambdaDt;\n"
" updated = max2( updated, minRambdaDt[i] );\n"
" updated = min2( updated, maxRambdaDt[i] );\n"
" rambdaDt = updated - prevSum;\n"
" cs->m_fAppliedRambdaDt[i] = updated;\n"
" }\n"
" \n"
" float4 linImp0 = invMassA*linear*rambdaDt;\n"
" float4 linImp1 = invMassB*(-linear)*rambdaDt;\n"
" float4 angImp0 = mtMul1(invInertiaA, angular0)*rambdaDt;\n"
" float4 angImp1 = mtMul1(invInertiaB, angular1)*rambdaDt;\n"
" \n"
" linVelA += linImp0;\n"
" angVelA += angImp0;\n"
" linVelB += linImp1;\n"
" angVelB += angImp1;\n"
" }\n"
" { // angular damping for point constraint\n"
" float4 ab = normalize3( posB - posA );\n"
" float4 ac = normalize3( center - posA );\n"
" if( dot3F4( ab, ac ) > 0.95f || (invMassA == 0.f || invMassB == 0.f))\n"
" {\n"
" float angNA = dot3F4( n, angVelA );\n"
" float angNB = dot3F4( n, angVelB );\n"
" \n"
" angVelA -= (angNA*0.1f)*n;\n"
" angVelB -= (angNB*0.1f)*n;\n"
" }\n"
" }\n"
" }\n"
" \n"
" \n"
" }\n"
" if (gBodies[aIdx].m_invMass)\n"
" {\n"
" gBodies[aIdx].m_linVel = linVelA;\n"
" gBodies[aIdx].m_angVel = angVelA;\n"
" } else\n"
" {\n"
" gBodies[aIdx].m_linVel = mymake_float4(0,0,0,0);\n"
" gBodies[aIdx].m_angVel = mymake_float4(0,0,0,0);\n"
" }\n"
" if (gBodies[bIdx].m_invMass)\n"
" {\n"
" gBodies[bIdx].m_linVel = linVelB;\n"
" gBodies[bIdx].m_angVel = angVelB;\n"
" } else\n"
" {\n"
" gBodies[bIdx].m_linVel = mymake_float4(0,0,0,0);\n"
" gBodies[bIdx].m_angVel = mymake_float4(0,0,0,0);\n"
" }\n"
" \n"
"}\n"
"typedef struct \n"
"{\n"
" int m_valInt0;\n"
" int m_valInt1;\n"
" int m_valInt2;\n"
" int m_valInt3;\n"
" float m_val0;\n"
" float m_val1;\n"
" float m_val2;\n"
" float m_val3;\n"
"} SolverDebugInfo;\n"
"__kernel\n"
"__attribute__((reqd_work_group_size(WG_SIZE,1,1)))\n"
"void BatchSolveKernelFriction(__global Body* gBodies,\n"
" __global Shape* gShapes,\n"
" __global Constraint4* gConstraints,\n"
" __global int* gN,\n"
" __global int* gOffsets,\n"
" __global int* batchSizes,\n"
" int maxBatch1,\n"
" int cellBatch,\n"
" int4 nSplit\n"
" )\n"
"{\n"
" //__local int ldsBatchIdx[WG_SIZE+1];\n"
" __local int ldsCurBatch;\n"
" __local int ldsNextBatch;\n"
" __local int ldsStart;\n"
" int lIdx = GET_LOCAL_IDX;\n"
" int wgIdx = GET_GROUP_IDX;\n"
"// int gIdx = GET_GLOBAL_IDX;\n"
"// debugInfo[gIdx].m_valInt0 = gIdx;\n"
" //debugInfo[gIdx].m_valInt1 = GET_GROUP_SIZE;\n"
" int zIdx = (wgIdx/((nSplit.x*nSplit.y)/4))*2+((cellBatch&4)>>2);\n"
" int remain= (wgIdx%((nSplit.x*nSplit.y)/4));\n"
" int yIdx = (remain/(nSplit.x/2))*2 + ((cellBatch&2)>>1);\n"
" int xIdx = (remain%(nSplit.x/2))*2 + (cellBatch&1);\n"
" int cellIdx = xIdx+yIdx*nSplit.x+zIdx*(nSplit.x*nSplit.y);\n"
" \n"
" if( gN[cellIdx] == 0 ) \n"
" return;\n"
" int maxBatch = batchSizes[cellIdx];\n"
" const int start = gOffsets[cellIdx];\n"
" const int end = start + gN[cellIdx];\n"
" \n"
" if( lIdx == 0 )\n"
" {\n"
" ldsCurBatch = 0;\n"
" ldsNextBatch = 0;\n"
" ldsStart = start;\n"
" }\n"
" GROUP_LDS_BARRIER;\n"
" int idx=ldsStart+lIdx;\n"
" while (ldsCurBatch < maxBatch)\n"
" {\n"
" for(; idx<end; )\n"
" {\n"
" if (gConstraints[idx].m_batchIdx == ldsCurBatch)\n"
" {\n"
" solveFrictionConstraint( gBodies, gShapes, &gConstraints[idx] );\n"
" idx+=64;\n"
" } else\n"
" {\n"
" break;\n"
" }\n"
" }\n"
" GROUP_LDS_BARRIER;\n"
" if( lIdx == 0 )\n"
" {\n"
" ldsCurBatch++;\n"
" }\n"
" GROUP_LDS_BARRIER;\n"
" }\n"
" \n"
" \n"
"}\n"
"__kernel void solveSingleFrictionKernel(__global Body* gBodies,\n"
" __global Shape* gShapes,\n"
" __global Constraint4* gConstraints,\n"
" int cellIdx,\n"
" int batchOffset,\n"
" int numConstraintsInBatch\n"
" )\n"
"{\n"
" int index = get_global_id(0);\n"
" if (index < numConstraintsInBatch)\n"
" {\n"
" \n"
" int idx=batchOffset+index;\n"
" \n"
" solveFrictionConstraint( gBodies, gShapes, &gConstraints[idx] );\n"
" } \n"
"}\n"
;
static const char* solveFrictionCL =
"/*\n"
"Copyright (c) 2012 Advanced Micro Devices, Inc. \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"
"//Originally written by Takahiro Harada\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"
"typedef unsigned int u32;\n"
"typedef unsigned short u16;\n"
"typedef unsigned char u8;\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 SELECT_UINT4( b, a, condition ) select( b,a,condition )\n"
"#define mymake_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"
"#define max2 max\n"
"#define min2 min\n"
"///////////////////////////////////////\n"
"// Vector\n"
"///////////////////////////////////////\n"
"__inline\n"
"float4 fastNormalize4(float4 v)\n"
"{\n"
" return fast_normalize(v);\n"
"}\n"
"__inline\n"
"float4 cross3(float4 a, float4 b)\n"
"{\n"
" return cross(a,b);\n"
"}\n"
"__inline\n"
"float dot3F4(float4 a, float4 b)\n"
"{\n"
" float4 a1 = mymake_float4(a.xyz,0.f);\n"
" float4 b1 = mymake_float4(b.xyz,0.f);\n"
" return dot(a1, b1);\n"
"}\n"
"__inline\n"
"float4 normalize3(const float4 a)\n"
"{\n"
" float4 n = mymake_float4(a.x, a.y, a.z, 0.f);\n"
" return fastNormalize4( n );\n"
"// float length = sqrtf(dot3F4(a, a));\n"
"// return 1.f/length * a;\n"
"}\n"
"///////////////////////////////////////\n"
"// Matrix3x3\n"
"///////////////////////////////////////\n"
"typedef struct\n"
"{\n"
" float4 m_row[3];\n"
"}Matrix3x3;\n"
"__inline\n"
"float4 mtMul1(Matrix3x3 a, float4 b);\n"
"__inline\n"
"float4 mtMul3(float4 a, Matrix3x3 b);\n"
"__inline\n"
"float4 mtMul1(Matrix3x3 a, float4 b)\n"
"{\n"
" float4 ans;\n"
" ans.x = dot3F4( a.m_row[0], b );\n"
" ans.y = dot3F4( a.m_row[1], b );\n"
" ans.z = dot3F4( a.m_row[2], b );\n"
" ans.w = 0.f;\n"
" return ans;\n"
"}\n"
"__inline\n"
"float4 mtMul3(float4 a, Matrix3x3 b)\n"
"{\n"
" float4 colx = mymake_float4(b.m_row[0].x, b.m_row[1].x, b.m_row[2].x, 0);\n"
" float4 coly = mymake_float4(b.m_row[0].y, b.m_row[1].y, b.m_row[2].y, 0);\n"
" float4 colz = mymake_float4(b.m_row[0].z, b.m_row[1].z, b.m_row[2].z, 0);\n"
" float4 ans;\n"
" ans.x = dot3F4( a, colx );\n"
" ans.y = dot3F4( a, coly );\n"
" ans.z = dot3F4( a, colz );\n"
" return ans;\n"
"}\n"
"///////////////////////////////////////\n"
"// Quaternion\n"
"///////////////////////////////////////\n"
"typedef float4 Quaternion;\n"
"#define WG_SIZE 64\n"
"typedef struct\n"
"{\n"
" float4 m_pos;\n"
" Quaternion m_quat;\n"
" float4 m_linVel;\n"
" float4 m_angVel;\n"
" u32 m_shapeIdx;\n"
" float m_invMass;\n"
" float m_restituitionCoeff;\n"
" float m_frictionCoeff;\n"
"} Body;\n"
"typedef struct\n"
"{\n"
" Matrix3x3 m_invInertia;\n"
" Matrix3x3 m_initInvInertia;\n"
"} Shape;\n"
"typedef struct\n"
"{\n"
" float4 m_linear;\n"
" float4 m_worldPos[4];\n"
" float4 m_center; \n"
" float m_jacCoeffInv[4];\n"
" float m_b[4];\n"
" float m_appliedRambdaDt[4];\n"
" float m_fJacCoeffInv[2]; \n"
" float m_fAppliedRambdaDt[2]; \n"
" u32 m_bodyA;\n"
" u32 m_bodyB;\n"
" int m_batchIdx;\n"
" u32 m_paddings[1];\n"
"} Constraint4;\n"
"typedef struct\n"
"{\n"
" int m_nConstraints;\n"
" int m_start;\n"
" int m_batchIdx;\n"
" int m_nSplit;\n"
"// int m_paddings[1];\n"
"} ConstBuffer;\n"
"typedef struct\n"
"{\n"
" int m_solveFriction;\n"
" int m_maxBatch; // long batch really kills the performance\n"
" int m_batchIdx;\n"
" int m_nSplit;\n"
"// int m_paddings[1];\n"
"} ConstBufferBatchSolve;\n"
"void setLinearAndAngular( float4 n, float4 r0, float4 r1, float4* linear, float4* angular0, float4* angular1);\n"
"void setLinearAndAngular( float4 n, float4 r0, float4 r1, float4* linear, float4* angular0, float4* angular1)\n"
"{\n"
" *linear = mymake_float4(-n.xyz,0.f);\n"
" *angular0 = -cross3(r0, n);\n"
" *angular1 = cross3(r1, n);\n"
"}\n"
"float calcRelVel( float4 l0, float4 l1, float4 a0, float4 a1, float4 linVel0, float4 angVel0, float4 linVel1, float4 angVel1 );\n"
"float calcRelVel( float4 l0, float4 l1, float4 a0, float4 a1, float4 linVel0, float4 angVel0, float4 linVel1, float4 angVel1 )\n"
"{\n"
" return dot3F4(l0, linVel0) + dot3F4(a0, angVel0) + dot3F4(l1, linVel1) + dot3F4(a1, angVel1);\n"
"}\n"
"float calcJacCoeff(const float4 linear0, const float4 linear1, const float4 angular0, const float4 angular1,\n"
" float invMass0, const Matrix3x3* invInertia0, float invMass1, const Matrix3x3* invInertia1);\n"
"float calcJacCoeff(const float4 linear0, const float4 linear1, const float4 angular0, const float4 angular1,\n"
" float invMass0, const Matrix3x3* invInertia0, float invMass1, const Matrix3x3* invInertia1)\n"
"{\n"
" // linear0,1 are normlized\n"
" float jmj0 = invMass0;//dot3F4(linear0, linear0)*invMass0;\n"
" float jmj1 = dot3F4(mtMul3(angular0,*invInertia0), angular0);\n"
" float jmj2 = invMass1;//dot3F4(linear1, linear1)*invMass1;\n"
" float jmj3 = dot3F4(mtMul3(angular1,*invInertia1), angular1);\n"
" return -1.f/(jmj0+jmj1+jmj2+jmj3);\n"
"}\n"
"void btPlaneSpace1 (const float4* n, float4* p, float4* q);\n"
" void btPlaneSpace1 (const float4* n, float4* p, float4* q)\n"
"{\n"
" if (fabs(n[0].z) > 0.70710678f) {\n"
" // choose p in y-z plane\n"
" float a = n[0].y*n[0].y + n[0].z*n[0].z;\n"
" float k = 1.f/sqrt(a);\n"
" p[0].x = 0;\n"
" p[0].y = -n[0].z*k;\n"
" p[0].z = n[0].y*k;\n"
" // set q = n x p\n"
" q[0].x = a*k;\n"
" q[0].y = -n[0].x*p[0].z;\n"
" q[0].z = n[0].x*p[0].y;\n"
" }\n"
" else {\n"
" // choose p in x-y plane\n"
" float a = n[0].x*n[0].x + n[0].y*n[0].y;\n"
" float k = 1.f/sqrt(a);\n"
" p[0].x = -n[0].y*k;\n"
" p[0].y = n[0].x*k;\n"
" p[0].z = 0;\n"
" // set q = n x p\n"
" q[0].x = -n[0].z*p[0].y;\n"
" q[0].y = n[0].z*p[0].x;\n"
" q[0].z = a*k;\n"
" }\n"
"}\n"
"void solveFrictionConstraint(__global Body* gBodies, __global Shape* gShapes, __global Constraint4* ldsCs);\n"
"void solveFrictionConstraint(__global Body* gBodies, __global Shape* gShapes, __global Constraint4* ldsCs)\n"
"{\n"
" float frictionCoeff = ldsCs[0].m_linear.w;\n"
" int aIdx = ldsCs[0].m_bodyA;\n"
" int bIdx = ldsCs[0].m_bodyB;\n"
" float4 posA = gBodies[aIdx].m_pos;\n"
" float4 linVelA = gBodies[aIdx].m_linVel;\n"
" float4 angVelA = gBodies[aIdx].m_angVel;\n"
" float invMassA = gBodies[aIdx].m_invMass;\n"
" Matrix3x3 invInertiaA = gShapes[aIdx].m_invInertia;\n"
" float4 posB = gBodies[bIdx].m_pos;\n"
" float4 linVelB = gBodies[bIdx].m_linVel;\n"
" float4 angVelB = gBodies[bIdx].m_angVel;\n"
" float invMassB = gBodies[bIdx].m_invMass;\n"
" Matrix3x3 invInertiaB = gShapes[bIdx].m_invInertia;\n"
" \n"
" {\n"
" float maxRambdaDt[4] = {FLT_MAX,FLT_MAX,FLT_MAX,FLT_MAX};\n"
" float minRambdaDt[4] = {0.f,0.f,0.f,0.f};\n"
" float sum = 0;\n"
" for(int j=0; j<4; j++)\n"
" {\n"
" sum +=ldsCs[0].m_appliedRambdaDt[j];\n"
" }\n"
" frictionCoeff = 0.7f;\n"
" for(int j=0; j<4; j++)\n"
" {\n"
" maxRambdaDt[j] = frictionCoeff*sum;\n"
" minRambdaDt[j] = -maxRambdaDt[j];\n"
" }\n"
" \n"
"// solveFriction( ldsCs, posA, &linVelA, &angVelA, invMassA, invInertiaA,\n"
"// posB, &linVelB, &angVelB, invMassB, invInertiaB, maxRambdaDt, minRambdaDt );\n"
" \n"
" \n"
" {\n"
" \n"
" __global Constraint4* cs = ldsCs;\n"
" \n"
" if( cs->m_fJacCoeffInv[0] == 0 && cs->m_fJacCoeffInv[0] == 0 ) return;\n"
" const float4 center = cs->m_center;\n"
" \n"
" float4 n = -cs->m_linear;\n"
" \n"
" float4 tangent[2];\n"
" btPlaneSpace1(&n,&tangent[0],&tangent[1]);\n"
" float4 angular0, angular1, linear;\n"
" float4 r0 = center - posA;\n"
" float4 r1 = center - posB;\n"
" for(int i=0; i<2; i++)\n"
" {\n"
" setLinearAndAngular( tangent[i], r0, r1, &linear, &angular0, &angular1 );\n"
" float rambdaDt = calcRelVel(linear, -linear, angular0, angular1,\n"
" linVelA, angVelA, linVelB, angVelB );\n"
" rambdaDt *= cs->m_fJacCoeffInv[i];\n"
" \n"
" {\n"
" float prevSum = cs->m_fAppliedRambdaDt[i];\n"
" float updated = prevSum;\n"
" updated += rambdaDt;\n"
" updated = max2( updated, minRambdaDt[i] );\n"
" updated = min2( updated, maxRambdaDt[i] );\n"
" rambdaDt = updated - prevSum;\n"
" cs->m_fAppliedRambdaDt[i] = updated;\n"
" }\n"
" \n"
" float4 linImp0 = invMassA*linear*rambdaDt;\n"
" float4 linImp1 = invMassB*(-linear)*rambdaDt;\n"
" float4 angImp0 = mtMul1(invInertiaA, angular0)*rambdaDt;\n"
" float4 angImp1 = mtMul1(invInertiaB, angular1)*rambdaDt;\n"
" \n"
" linVelA += linImp0;\n"
" angVelA += angImp0;\n"
" linVelB += linImp1;\n"
" angVelB += angImp1;\n"
" }\n"
" { // angular damping for point constraint\n"
" float4 ab = normalize3( posB - posA );\n"
" float4 ac = normalize3( center - posA );\n"
" if( dot3F4( ab, ac ) > 0.95f || (invMassA == 0.f || invMassB == 0.f))\n"
" {\n"
" float angNA = dot3F4( n, angVelA );\n"
" float angNB = dot3F4( n, angVelB );\n"
" \n"
" angVelA -= (angNA*0.1f)*n;\n"
" angVelB -= (angNB*0.1f)*n;\n"
" }\n"
" }\n"
" }\n"
" \n"
" \n"
" }\n"
" if (gBodies[aIdx].m_invMass)\n"
" {\n"
" gBodies[aIdx].m_linVel = linVelA;\n"
" gBodies[aIdx].m_angVel = angVelA;\n"
" } else\n"
" {\n"
" gBodies[aIdx].m_linVel = mymake_float4(0,0,0,0);\n"
" gBodies[aIdx].m_angVel = mymake_float4(0,0,0,0);\n"
" }\n"
" if (gBodies[bIdx].m_invMass)\n"
" {\n"
" gBodies[bIdx].m_linVel = linVelB;\n"
" gBodies[bIdx].m_angVel = angVelB;\n"
" } else\n"
" {\n"
" gBodies[bIdx].m_linVel = mymake_float4(0,0,0,0);\n"
" gBodies[bIdx].m_angVel = mymake_float4(0,0,0,0);\n"
" }\n"
" \n"
"}\n"
"typedef struct \n"
"{\n"
" int m_valInt0;\n"
" int m_valInt1;\n"
" int m_valInt2;\n"
" int m_valInt3;\n"
" float m_val0;\n"
" float m_val1;\n"
" float m_val2;\n"
" float m_val3;\n"
"} SolverDebugInfo;\n"
"__kernel\n"
"__attribute__((reqd_work_group_size(WG_SIZE,1,1)))\n"
"void BatchSolveKernelFriction(__global Body* gBodies,\n"
" __global Shape* gShapes,\n"
" __global Constraint4* gConstraints,\n"
" __global int* gN,\n"
" __global int* gOffsets,\n"
" __global int* batchSizes,\n"
" int maxBatch1,\n"
" int cellBatch,\n"
" int4 nSplit\n"
" )\n"
"{\n"
" //__local int ldsBatchIdx[WG_SIZE+1];\n"
" __local int ldsCurBatch;\n"
" __local int ldsNextBatch;\n"
" __local int ldsStart;\n"
" int lIdx = GET_LOCAL_IDX;\n"
" int wgIdx = GET_GROUP_IDX;\n"
"// int gIdx = GET_GLOBAL_IDX;\n"
"// debugInfo[gIdx].m_valInt0 = gIdx;\n"
" //debugInfo[gIdx].m_valInt1 = GET_GROUP_SIZE;\n"
" int zIdx = (wgIdx/((nSplit.x*nSplit.y)/4))*2+((cellBatch&4)>>2);\n"
" int remain= (wgIdx%((nSplit.x*nSplit.y)/4));\n"
" int yIdx = (remain/(nSplit.x/2))*2 + ((cellBatch&2)>>1);\n"
" int xIdx = (remain%(nSplit.x/2))*2 + (cellBatch&1);\n"
" int cellIdx = xIdx+yIdx*nSplit.x+zIdx*(nSplit.x*nSplit.y);\n"
" \n"
" if( gN[cellIdx] == 0 ) \n"
" return;\n"
" int maxBatch = batchSizes[cellIdx];\n"
" const int start = gOffsets[cellIdx];\n"
" const int end = start + gN[cellIdx];\n"
" \n"
" if( lIdx == 0 )\n"
" {\n"
" ldsCurBatch = 0;\n"
" ldsNextBatch = 0;\n"
" ldsStart = start;\n"
" }\n"
" GROUP_LDS_BARRIER;\n"
" int idx=ldsStart+lIdx;\n"
" while (ldsCurBatch < maxBatch)\n"
" {\n"
" for(; idx<end; )\n"
" {\n"
" if (gConstraints[idx].m_batchIdx == ldsCurBatch)\n"
" {\n"
" solveFrictionConstraint( gBodies, gShapes, &gConstraints[idx] );\n"
" idx+=64;\n"
" } else\n"
" {\n"
" break;\n"
" }\n"
" }\n"
" GROUP_LDS_BARRIER;\n"
" if( lIdx == 0 )\n"
" {\n"
" ldsCurBatch++;\n"
" }\n"
" GROUP_LDS_BARRIER;\n"
" }\n"
" \n"
" \n"
"}\n"
"__kernel void solveSingleFrictionKernel(__global Body* gBodies,\n"
" __global Shape* gShapes,\n"
" __global Constraint4* gConstraints,\n"
" int cellIdx,\n"
" int batchOffset,\n"
" int numConstraintsInBatch\n"
" )\n"
"{\n"
" int index = get_global_id(0);\n"
" if (index < numConstraintsInBatch)\n"
" {\n"
" \n"
" int idx=batchOffset+index;\n"
" \n"
" solveFrictionConstraint( gBodies, gShapes, &gConstraints[idx] );\n"
" } \n"
"}\n";

1403
src/Bullet3OpenCL/RigidBody/kernels/solverSetup.h
View File

File diff suppressed because it is too large Load Diff

1199
src/Bullet3OpenCL/RigidBody/kernels/solverSetup2.h
View File

File diff suppressed because it is too large Load Diff

1815
src/Bullet3OpenCL/RigidBody/kernels/solverUtils.h
View File

File diff suppressed because it is too large Load Diff

963
src/Bullet3OpenCL/RigidBody/kernels/updateAabbsKernel.h
View File

@@ -1,483 +1,482 @@
//this file is autogenerated using stringify.bat (premake --stringify) in the build folder of this project
static const char* updateAabbsKernelCL= \
"#ifndef B3_UPDATE_AABBS_H\n"
"#define B3_UPDATE_AABBS_H\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"
"#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"
" union\n"
" {\n"
" int m_shapeIndex;\n"
" float m_height;\n"
" };\n"
"};\n"
"typedef struct b3GpuChildShape b3GpuChildShape_t;\n"
"struct b3GpuChildShape\n"
"{\n"
" b3Float4 m_childPosition;\n"
" b3Quat m_childOrientation;\n"
" union\n"
" {\n"
" int m_shapeIndex;//used for SHAPE_COMPOUND_OF_CONVEX_HULLS\n"
" int m_capsuleAxis;\n"
" };\n"
" union \n"
" {\n"
" float m_radius;//used for childshape of SHAPE_COMPOUND_OF_SPHERES or SHAPE_COMPOUND_OF_CAPSULES\n"
" int m_numChildShapes;//used for compound shape\n"
" };\n"
" union \n"
" {\n"
" float m_height;//used for childshape of SHAPE_COMPOUND_OF_CAPSULES\n"
" int m_collidableShapeIndex;\n"
" };\n"
" int m_shapeType;\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"
"#ifdef __cplusplus\n"
"#else\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"
"void b3ComputeWorldAabb( int bodyId, __global const b3RigidBodyData_t* bodies, __global const b3Collidable_t* collidables, __global const b3Aabb_t* localShapeAABB, __global b3Aabb_t* worldAabbs)\n"
"{\n"
" __global const b3RigidBodyData_t* body = &bodies[bodyId];\n"
" b3Float4 position = body->m_pos;\n"
" b3Quat orientation = body->m_quat;\n"
" \n"
" int collidableIndex = body->m_collidableIdx;\n"
" int shapeIndex = collidables[collidableIndex].m_shapeIndex;\n"
" \n"
" if (shapeIndex>=0)\n"
" {\n"
" \n"
" b3Aabb_t localAabb = localShapeAABB[collidableIndex];\n"
" b3Aabb_t worldAabb;\n"
" \n"
" b3Float4 aabbAMinOut,aabbAMaxOut; \n"
" float margin = 0.f;\n"
" b3TransformAabb2(localAabb.m_minVec,localAabb.m_maxVec,margin,position,orientation,&aabbAMinOut,&aabbAMaxOut);\n"
" \n"
" worldAabb.m_minVec =aabbAMinOut;\n"
" worldAabb.m_minIndices[3] = bodyId;\n"
" worldAabb.m_maxVec = aabbAMaxOut;\n"
" worldAabb.m_signedMaxIndices[3] = body[bodyId].m_invMass==0.f? 0 : 1;\n"
" worldAabbs[bodyId] = worldAabb;\n"
" }\n"
"}\n"
"#endif //B3_UPDATE_AABBS_H\n"
"__kernel void initializeGpuAabbsFull( const int numNodes, __global b3RigidBodyData_t* gBodies,__global b3Collidable_t* collidables, __global b3Aabb_t* plocalShapeAABB, __global b3Aabb_t* pAABB)\n"
"{\n"
" int nodeID = get_global_id(0);\n"
" if( nodeID < numNodes )\n"
" {\n"
" b3ComputeWorldAabb(nodeID, gBodies, collidables, plocalShapeAABB,pAABB);\n"
" }\n"
"}\n"
"__kernel void clearOverlappingPairsKernel( __global int4* pairs, int numPairs)\n"
"{\n"
" int pairId = get_global_id(0);\n"
" if( pairId< numPairs )\n"
" {\n"
" pairs[pairId].z = 0xffffffff;\n"
" }\n"
"}\n"
;
static const char* updateAabbsKernelCL =
"#ifndef B3_UPDATE_AABBS_H\n"
"#define B3_UPDATE_AABBS_H\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"
"#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"
" union\n"
" {\n"
" int m_shapeIndex;\n"
" float m_height;\n"
" };\n"
"};\n"
"typedef struct b3GpuChildShape b3GpuChildShape_t;\n"
"struct b3GpuChildShape\n"
"{\n"
" b3Float4 m_childPosition;\n"
" b3Quat m_childOrientation;\n"
" union\n"
" {\n"
" int m_shapeIndex;//used for SHAPE_COMPOUND_OF_CONVEX_HULLS\n"
" int m_capsuleAxis;\n"
" };\n"
" union \n"
" {\n"
" float m_radius;//used for childshape of SHAPE_COMPOUND_OF_SPHERES or SHAPE_COMPOUND_OF_CAPSULES\n"
" int m_numChildShapes;//used for compound shape\n"
" };\n"
" union \n"
" {\n"
" float m_height;//used for childshape of SHAPE_COMPOUND_OF_CAPSULES\n"
" int m_collidableShapeIndex;\n"
" };\n"
" int m_shapeType;\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"
"#ifdef __cplusplus\n"
"#else\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"
"void b3ComputeWorldAabb( int bodyId, __global const b3RigidBodyData_t* bodies, __global const b3Collidable_t* collidables, __global const b3Aabb_t* localShapeAABB, __global b3Aabb_t* worldAabbs)\n"
"{\n"
" __global const b3RigidBodyData_t* body = &bodies[bodyId];\n"
" b3Float4 position = body->m_pos;\n"
" b3Quat orientation = body->m_quat;\n"
" \n"
" int collidableIndex = body->m_collidableIdx;\n"
" int shapeIndex = collidables[collidableIndex].m_shapeIndex;\n"
" \n"
" if (shapeIndex>=0)\n"
" {\n"
" \n"
" b3Aabb_t localAabb = localShapeAABB[collidableIndex];\n"
" b3Aabb_t worldAabb;\n"
" \n"
" b3Float4 aabbAMinOut,aabbAMaxOut; \n"
" float margin = 0.f;\n"
" b3TransformAabb2(localAabb.m_minVec,localAabb.m_maxVec,margin,position,orientation,&aabbAMinOut,&aabbAMaxOut);\n"
" \n"
" worldAabb.m_minVec =aabbAMinOut;\n"
" worldAabb.m_minIndices[3] = bodyId;\n"
" worldAabb.m_maxVec = aabbAMaxOut;\n"
" worldAabb.m_signedMaxIndices[3] = body[bodyId].m_invMass==0.f? 0 : 1;\n"
" worldAabbs[bodyId] = worldAabb;\n"
" }\n"
"}\n"
"#endif //B3_UPDATE_AABBS_H\n"
"__kernel void initializeGpuAabbsFull( const int numNodes, __global b3RigidBodyData_t* gBodies,__global b3Collidable_t* collidables, __global b3Aabb_t* plocalShapeAABB, __global b3Aabb_t* pAABB)\n"
"{\n"
" int nodeID = get_global_id(0);\n"
" if( nodeID < numNodes )\n"
" {\n"
" b3ComputeWorldAabb(nodeID, gBodies, collidables, plocalShapeAABB,pAABB);\n"
" }\n"
"}\n"
"__kernel void clearOverlappingPairsKernel( __global int4* pairs, int numPairs)\n"
"{\n"
" int pairId = get_global_id(0);\n"
" if( pairId< numPairs )\n"
" {\n"
" pairs[pairId].z = 0xffffffff;\n"
" }\n"
"}\n";