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

implemented sphere-convex (supports edge and corner-vertex cases)

Browse Source
This commit is contained in:
erwincoumans
2013-04-02 22:09:40 -07:00
parent e38c032280
commit 5c8c8e1cbb
9 changed files with 1708 additions and 303 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
build/stringify.bat
View File

@@ -13,6 +13,8 @@ premake4 --file=stringifyKernel.lua --kernelfile="../opencl/gpu_broadphase/kerne
premake4 --file=stringifyKernel.lua --kernelfile="../opencl/gpu_sat/kernels/sat.cl" --headerfile="../opencl/gpu_sat/kernels/satKernels.h" --stringname="satKernelsCL" stringify
premake4 --file=stringifyKernel.lua --kernelfile="../opencl/gpu_sat/kernels/satClipHullContacts.cl" --headerfile="../opencl/gpu_sat/kernels/satClipHullContacts.h" --stringname="satClipKernelsCL" stringify
premake4 --file=stringifyKernel.lua --kernelfile="../opencl/gpu_sat/kernels/primitiveContacts.cl" --headerfile="../opencl/gpu_sat/kernels/primitiveContacts.h" --stringname="primitiveContactsKernelsCL" stringify
premake4 --file=stringifyKernel.lua --kernelfile="../opencl/gpu_sat/kernels/bvhTraversal.cl" --headerfile="../opencl/gpu_sat/kernels/bvhTraversal.h" --stringname="bvhTraversalKernelCL" stringify

6
demo/gpudemo/GpuDemo.h
View File

@@ -38,9 +38,9 @@ public:
:useOpenCL(true),
preferredOpenCLPlatformIndex(-1),
preferredOpenCLDeviceIndex(-1),
arraySizeX(1),
arraySizeY(2),
arraySizeZ(1),
arraySizeX(25),
arraySizeY(23),
arraySizeZ(23),
m_useConcaveMesh(false),
gapX(14.3),
gapY(14.0),

33
demo/gpudemo/rigidbody/GpuSphereScene.cpp
View File

@@ -32,18 +32,19 @@ void GpuSphereScene::setupScene(const ConstructionInfo& ci)
int mask=1;
int index=0;
if (0)
if (1)
{
int shapeId = ci.m_instancingRenderer->registerShape(&cube_vertices[0],numVertices,cube_indices,numIndices);
btVector4 scaling(400,0.01,400,1);
//int colIndex = m_data->m_np->registerConvexHullShape(&cube_vertices[0],strideInBytes,numVertices, scaling);
btVector3 normal(0,1,0);
float constant=0.01;
btVector4 scaling(120,2,120,1);
int colIndex = m_data->m_np->registerConvexHullShape(&cube_vertices[0],strideInBytes,numVertices, scaling);
btVector3 normal(0,-1,0);
float constant=2;
int colIndex = m_data->m_np->registerPlaneShape(normal,constant);//>registerConvexHullShape(&cube_vertices[0],strideInBytes,numVertices, scaling);
btVector4 position(0,0,0,0);
btQuaternion orn(0,0,0,1);
//int colIndex = m_data->m_np->registerPlaneShape(normal,constant);//>registerConvexHullShape(&cube_vertices[0],strideInBytes,numVertices, scaling);
btVector4 position(0,50,0,0);
//btQuaternion orn(0,0,0,1);
btQuaternion orn(btVector3(1,0,0),0.3);
btVector4 color(0,0,1,1);
int id = ci.m_instancingRenderer->registerGraphicsInstance(shapeId,position,orn,color,scaling);
@@ -67,7 +68,7 @@ void GpuSphereScene::setupScene(const ConstructionInfo& ci)
prevGraphicsShapeIndex = ci.m_instancingRenderer->registerShape(&detailed_sphere_vertices[0],numVertices,detailed_sphere_indices,numIndices);
} else
{
bool usePointSprites = false;
bool usePointSprites = true;
if (usePointSprites)
{
int numVertices = sizeof(point_sphere_vertices)/strideInBytes;
@@ -109,7 +110,7 @@ void GpuSphereScene::setupScene(const ConstructionInfo& ci)
int curColor = 0;
float scaling[4] = {1,1,1,1};
//int colIndex = m_data->m_np->registerConvexHullShape(&cube_vertices[0],strideInBytes,numVertices, scaling);
int colIndex = m_data->m_np->registerSphereShape(radius);//>registerConvexHullShape(&cube_vertices[0],strideInBytes,numVertices, scaling);
for (int i=0;i<ci.arraySizeX;i++)
@@ -119,18 +120,16 @@ void GpuSphereScene::setupScene(const ConstructionInfo& ci)
for (int k=0;k<ci.arraySizeZ;k++)
{
float mass = 1.f;
if (j==0)
mass=0.f;
//btVector3 position((j&1)+i*2.2,2+j*2.,(j&1)+k*2.2);
btVector3 position(i*2.2,2+j*4.,k*2.2);
//btVector3 position((j&1)+i*2.2,1+j*2.,(j&1)+k*2.2);
btVector3 position(i*radius*3,j*radius*3,k*radius*3);
btQuaternion orn(0,0,0,1);
btVector4 color = colors[curColor];
curColor++;
curColor&=3;
btVector4 scaling(1,1,1,1);
btVector4 scaling(radius,radius,radius,1);
int id = ci.m_instancingRenderer->registerGraphicsInstance(prevGraphicsShapeIndex,position,orn,color,scaling);
int pid = m_data->m_rigidBodyPipeline->registerPhysicsInstance(mass,position,orn,colIndex,index);
@@ -142,7 +141,7 @@ void GpuSphereScene::setupScene(const ConstructionInfo& ci)
float camPos[4]={ci.arraySizeX,ci.arraySizeY/2,ci.arraySizeZ,0};
//float camPos[4]={1,12.5,1.5,0};
m_instancingRenderer->setCameraTargetPosition(camPos);
m_instancingRenderer->setCameraDistance(20);
m_instancingRenderer->setCameraDistance(150);
char msg[1024];

339
opencl/gpu_sat/host/ConvexHullContact.cpp
View File

@@ -36,6 +36,7 @@ typedef btAlignedObjectArray<btVector3> btVertexArray;
#include "../kernels/satKernels.h"
#include "../kernels/satClipHullContacts.h"
#include "../kernels/bvhTraversal.h"
#include "../kernels/primitiveContacts.h"
#include "BulletGeometry/btAabbUtil2.h"
@@ -117,11 +118,20 @@ m_totalContactsOut(m_context, m_queue)
//cl_program bvhTraversalProg = btOpenCLUtils::compileCLProgramFromString(m_context,m_device,0,&errNum,"","opencl/gpu_sat/kernels/bvhTraversal.cl", true);
btAssert(errNum==CL_SUCCESS);
m_bvhTraversalKernel = btOpenCLUtils::compileCLKernelFromString(m_context, m_device,srcBvh, "bvhTraversalKernel",&errNum,bvhTraversalProg,"-g");
m_bvhTraversalKernel = btOpenCLUtils::compileCLKernelFromString(m_context, m_device,srcBvh, "bvhTraversalKernel",&errNum,bvhTraversalProg,"");
btAssert(errNum==CL_SUCCESS);
}
{
const char* primitiveContactsSrc = primitiveContactsKernelsCL;
cl_program primitiveContactsProg = btOpenCLUtils::compileCLProgramFromString(m_context,m_device,primitiveContactsSrc,&errNum,"","opencl/gpu_sat/kernels/primitiveContacts.cl");
btAssert(errNum==CL_SUCCESS);
m_primitiveContactsKernel = btOpenCLUtils::compileCLKernelFromString(m_context, m_device,primitiveContactsSrc, "primitiveContactsKernel",&errNum,primitiveContactsProg,"");
btAssert(errNum==CL_SUCCESS);
}
}
@@ -148,6 +158,8 @@ GpuSatCollision::~GpuSatCollision()
clReleaseKernel(m_clipFacesAndContactReductionKernel);
if (m_newContactReductionKernel)
clReleaseKernel(m_newContactReductionKernel);
if (m_primitiveContactsKernel)
clReleaseKernel(m_primitiveContactsKernel);
if (m_clipHullHullKernel)
clReleaseKernel(m_clipHullHullKernel);
@@ -176,6 +188,224 @@ struct MyTriangleCallback : public btNodeOverlapCallback
}
};
#define float4 btVector3
#define make_float4(x,y,z,w) btVector4(x,y,z,w)
float signedDistanceFromPointToPlane(const float4& point, const float4& planeEqn, float4* closestPointOnFace)
{
float4 n = planeEqn;
n[3] = 0.f;
float dist = dot3F4(n, point) + planeEqn[3];
*closestPointOnFace = point - dist * n;
return dist;
}
inline bool IsPointInPolygon(const btVector3& p,
const btVector3& posConvex,
const btQuaternion& ornConvex,
const btGpuFace* face,
const btVector3* baseVertex,
const int* convexIndices,
btVector3* out)
{
btVector3 a;
btVector3 b;
btVector3 ab;
btVector3 ap;
btVector3 v;
btVector3 plane (face->m_plane[0],face->m_plane[1],face->m_plane[2]);
if (face->m_numIndices<2)
return false;
btTransform tr;
tr.setIdentity();
tr.setOrigin(posConvex);
tr.setRotation(ornConvex);
float4 v0 = baseVertex[convexIndices[face->m_indexOffset + face->m_numIndices-1]];
btVector3 worldV0 = tr(v0);
b = worldV0;
for(unsigned i=0; i != face->m_numIndices; ++i)
{
a = b;
float4 vi = baseVertex[convexIndices[face->m_indexOffset + i]];
btVector3 worldVi = tr(vi);
b = worldVi;
ab = b-a;
ap = p-a;
v = ab.cross(plane);
if (btDot(ap, v) > 0.f)
{
btScalar ab_m2 = btDot(ab, ab);
btScalar s = ab_m2 != btScalar(0.0) ? btDot(ab, ap) / ab_m2 : btScalar(0.0);
if (s <= btScalar(0.0))
{
*out = a;
}
else if (s >= btScalar(1.0))
{
*out = b;
}
else
{
out->setInterpolate3(a,b,s);
}
return false;
}
}
return true;
}
void computeContactSphereConvex(int pairIndex,
int bodyIndexA, int bodyIndexB,
int collidableIndexA, int collidableIndexB,
const btRigidBodyCL* rigidBodies,
const btCollidable* collidables,
const btConvexPolyhedronCL* convexShapes,
const btVector3* convexVertices,
const int* convexIndices,
const btGpuFace* faces,
btContact4* globalContactsOut,
int& nGlobalContactsOut,
int maxContactCapacity)
{
float radius = collidables[collidableIndexA].m_radius;
float4 spherePos1 = rigidBodies[bodyIndexA].m_pos;
btQuaternion sphereOrn = rigidBodies[bodyIndexA].m_quat;
float4 pos = rigidBodies[bodyIndexB].m_pos;
float4 spherePos = spherePos1-pos;
btQuaternion quat = rigidBodies[bodyIndexB].m_quat;
int collidableIndex = rigidBodies[bodyIndexB].m_collidableIdx;
int shapeIndex = collidables[collidableIndex].m_shapeIndex;
int numFaces = convexShapes[shapeIndex].m_numFaces;
float4 closestPnt = make_float4(0, 0, 0, 0);
float4 hitNormalWorld = make_float4(0, 0, 0, 0);
float minDist = -1000000.f; // TODO: What is the largest/smallest float?
bool bCollide = true;
int region = -1;
for ( int f = 0; f < numFaces; f++ )
{
btGpuFace face = faces[convexShapes[shapeIndex].m_faceOffset+f];
float4 planeEqn;
float4 localPlaneNormal = make_float4(face.m_plane.x(),face.m_plane.y(),face.m_plane.z(),0.f);
float4 n1 = quatRotate(quat,localPlaneNormal);
planeEqn = n1;
planeEqn[3] = face.m_plane[3];
float4 pntReturn;
float dist = signedDistanceFromPointToPlane(spherePos, planeEqn, &pntReturn);
if ( dist > radius)
{
bCollide = false;
break;
}
if ( dist > 0 )
{
//might hit an edge or vertex
btVector3 out;
bool isInPoly = IsPointInPolygon(spherePos,
pos,
quat,
&face,
&convexVertices[convexShapes[shapeIndex].m_vertexOffset],
convexIndices,
&out);
if (isInPoly)
{
if (dist>minDist)
{
minDist = dist;
closestPnt = pntReturn;
hitNormalWorld = planeEqn;
region=1;
}
} else
{
btVector3 tmp = spherePos-out;
btScalar l2 = tmp.length2();
if (l2<radius*radius)
{
dist = btSqrt(l2);
if (dist>minDist)
{
minDist = dist;
closestPnt = out;
hitNormalWorld = tmp/dist;
region=2;
}
} else
{
bCollide = false;
break;
}
}
}
else
{
if ( dist > minDist )
{
minDist = dist;
closestPnt = pntReturn;
hitNormalWorld = planeEqn;
region=3;
}
}
}
if (bCollide && minDist > -100)
{
float4 normalOnSurfaceB1 = -hitNormalWorld;
float4 pOnB1 = closestPnt+pos;
//printf("dist ,%f,",minDist);
float actualDepth = minDist-radius;
//printf("actualDepth = ,%f,", actualDepth);
//printf("normalOnSurfaceB1 = ,%f,%f,%f,", normalOnSurfaceB1.getX(),normalOnSurfaceB1.getY(),normalOnSurfaceB1.getZ());
//printf("region=,%d,\n", region);
pOnB1[3] = actualDepth;
int dstIdx;
// dstIdx = nGlobalContactsOut++;//AppendInc( nGlobalContactsOut, dstIdx );
if (nGlobalContactsOut < maxContactCapacity)
{
dstIdx=nGlobalContactsOut;
nGlobalContactsOut++;
btContact4* c = &globalContactsOut[dstIdx];
c->m_worldNormal = normalOnSurfaceB1;
c->setFrictionCoeff(0.7);
c->setRestituitionCoeff(0.f);
c->m_batchIdx = pairIndex;
c->m_bodyAPtrAndSignBit = rigidBodies[bodyIndexA].m_invMass==0?-bodyIndexA:bodyIndexA;
c->m_bodyBPtrAndSignBit = rigidBodies[bodyIndexB].m_invMass==0?-bodyIndexB:bodyIndexB;
c->m_worldPos[0] = pOnB1;
int numPoints = 1;
c->m_worldNormal[3] = numPoints;
}//if (dstIdx < numPairs)
}//if (hasCollision)
}
void GpuSatCollision::computeConvexConvexContactsGPUSAT( const btOpenCLArray<btInt2>* pairs, int nPairs,
const btOpenCLArray<btRigidBodyCL>* bodyBuf,
btOpenCLArray<btContact4>* contactOut, int& nContacts,
@@ -206,6 +436,113 @@ void GpuSatCollision::computeConvexConvexContactsGPUSAT( const btOpenCLArray<btI
if (!nPairs)
return;
//#define CHECK_ON_HOST
#ifdef CHECK_ON_HOST
btAlignedObjectArray<btYetAnotherAabb> hostAabbs;
clAabbsWS.copyToHost(hostAabbs);
btAlignedObjectArray<btInt2> hostPairs;
pairs->copyToHost(hostPairs);
btAlignedObjectArray<btRigidBodyCL> hostBodyBuf;
bodyBuf->copyToHost(hostBodyBuf);
btAlignedObjectArray<btConvexPolyhedronCL> hostConvexData;
convexData.copyToHost(hostConvexData);
btAlignedObjectArray<btVector3> hostVertices;
gpuVertices.copyToHost(hostVertices);
btAlignedObjectArray<btVector3> hostUniqueEdges;
gpuUniqueEdges.copyToHost(hostUniqueEdges);
btAlignedObjectArray<btGpuFace> hostFaces;
gpuFaces.copyToHost(hostFaces);
btAlignedObjectArray<int> hostIndices;
gpuIndices.copyToHost(hostIndices);
btAlignedObjectArray<btCollidable> hostCollidables;
gpuCollidables.copyToHost(hostCollidables);
btAlignedObjectArray<btGpuChildShape> cpuChildShapes;
gpuChildShapes.copyToHost(cpuChildShapes);
btAlignedObjectArray<btInt4> hostTriangleConvexPairs;
btAlignedObjectArray<btContact4> hostContacts;
if (nContacts)
{
contactOut->copyToHost(hostContacts);
}
hostContacts.resize(nPairs);
for (int i=0;i<nPairs;i++)
{
int bodyIndexA = hostPairs[i].x;
int bodyIndexB = hostPairs[i].y;
int collidableIndexA = hostBodyBuf[bodyIndexA].m_collidableIdx;
int collidableIndexB = hostBodyBuf[bodyIndexB].m_collidableIdx;
if (hostCollidables[collidableIndexA].m_shapeType == SHAPE_SPHERE &&
hostCollidables[collidableIndexB].m_shapeType == SHAPE_CONVEX_HULL)
{
printf("sphere-convex\n");
}
if (hostCollidables[collidableIndexA].m_shapeType == SHAPE_CONVEX_HULL &&
hostCollidables[collidableIndexB].m_shapeType == SHAPE_SPHERE)
{
computeContactSphereConvex(i,bodyIndexB,bodyIndexA,collidableIndexB,collidableIndexA,&hostBodyBuf[0],
&hostCollidables[0],&hostConvexData[0],&hostVertices[0],&hostIndices[0],&hostFaces[0],&hostContacts[0],nContacts,nPairs);
//printf("convex-sphere\n");
}
}
if (nContacts)
{
hostContacts.resize(nContacts);
contactOut->copyFromHost(hostContacts);
}
#else
{
if (nPairs)
{
m_totalContactsOut.copyFromHostPointer(&nContacts,1,0,true);
BT_PROFILE("primitiveContactsKernel");
btBufferInfoCL bInfo[] = {
btBufferInfoCL( pairs->getBufferCL(), true ),
btBufferInfoCL( bodyBuf->getBufferCL(),true),
btBufferInfoCL( gpuCollidables.getBufferCL(),true),
btBufferInfoCL( convexData.getBufferCL(),true),
btBufferInfoCL( gpuVertices.getBufferCL(),true),
btBufferInfoCL( gpuUniqueEdges.getBufferCL(),true),
btBufferInfoCL( gpuFaces.getBufferCL(),true),
btBufferInfoCL( gpuIndices.getBufferCL(),true),
btBufferInfoCL( contactOut->getBufferCL()),
btBufferInfoCL( m_totalContactsOut.getBufferCL())
};
btLauncherCL launcher(m_queue, m_primitiveContactsKernel);
launcher.setBuffers( bInfo, sizeof(bInfo)/sizeof(btBufferInfoCL) );
launcher.setConst( nPairs );
int num = nPairs;
launcher.launch1D( num);
clFinish(m_queue);
nContacts = m_totalContactsOut.at(0);
contactOut->resize(nContacts);
}
}
#endif//CHECK_ON_HOST
BT_PROFILE("computeConvexConvexContactsGPUSAT");
// printf("nContacts = %d\n",nContacts);

1
opencl/gpu_sat/host/ConvexHullContact.h
View File

@@ -52,6 +52,7 @@ struct GpuSatCollision
cl_kernel m_newContactReductionKernel;
cl_kernel m_bvhTraversalKernel;
cl_kernel m_primitiveContactsKernel;
btOpenCLArray<int> m_totalContactsOut;

667
opencl/gpu_sat/kernels/primitiveContacts.cl Normal file
View File

@@ -0,0 +1,667 @@
#define TRIANGLE_NUM_CONVEX_FACES 5
#define SHAPE_CONVEX_HULL 3
#define SHAPE_PLANE 4
#define SHAPE_CONCAVE_TRIMESH 5
#define SHAPE_COMPOUND_OF_CONVEX_HULLS 6
#define SHAPE_SPHERE 7
#pragma OPENCL EXTENSION cl_amd_printf : enable
#pragma OPENCL EXTENSION cl_khr_local_int32_base_atomics : enable
#pragma OPENCL EXTENSION cl_khr_global_int32_base_atomics : enable
#pragma OPENCL EXTENSION cl_khr_local_int32_extended_atomics : enable
#pragma OPENCL EXTENSION cl_khr_global_int32_extended_atomics : enable
#ifdef cl_ext_atomic_counters_32
#pragma OPENCL EXTENSION cl_ext_atomic_counters_32 : enable
#else
#define counter32_t volatile __global int*
#endif
#define GET_GROUP_IDX get_group_id(0)
#define GET_LOCAL_IDX get_local_id(0)
#define GET_GLOBAL_IDX get_global_id(0)
#define GET_GROUP_SIZE get_local_size(0)
#define GET_NUM_GROUPS get_num_groups(0)
#define GROUP_LDS_BARRIER barrier(CLK_LOCAL_MEM_FENCE)
#define GROUP_MEM_FENCE mem_fence(CLK_LOCAL_MEM_FENCE)
#define AtomInc(x) atom_inc(&(x))
#define AtomInc1(x, out) out = atom_inc(&(x))
#define AppendInc(x, out) out = atomic_inc(x)
#define AtomAdd(x, value) atom_add(&(x), value)
#define AtomCmpxhg(x, cmp, value) atom_cmpxchg( &(x), cmp, value )
#define AtomXhg(x, value) atom_xchg ( &(x), value )
#define max2 max
#define min2 min
typedef unsigned int u32;
typedef struct
{
float4 m_worldPos[4];
float4 m_worldNormal; // w: m_nPoints
u32 m_coeffs;
u32 m_batchIdx;
int m_bodyAPtrAndSignBit;//x:m_bodyAPtr, y:m_bodyBPtr
int m_bodyBPtrAndSignBit;
} Contact4;
///keep this in sync with btCollidable.h
typedef struct
{
int m_numChildShapes;
float m_radius;
int m_shapeType;
int m_shapeIndex;
} btCollidableGpu;
typedef struct
{
float4 m_childPosition;
float4 m_childOrientation;
int m_shapeIndex;
int m_unused0;
int m_unused1;
int m_unused2;
} btGpuChildShape;
#define GET_NPOINTS(x) (x).m_worldNormal.w
typedef struct
{
float4 m_pos;
float4 m_quat;
float4 m_linVel;
float4 m_angVel;
u32 m_collidableIdx;
float m_invMass;
float m_restituitionCoeff;
float m_frictionCoeff;
} BodyData;
typedef struct
{
float4 m_localCenter;
float4 m_extents;
float4 mC;
float4 mE;
float m_radius;
int m_faceOffset;
int m_numFaces;
int m_numVertices;
int m_vertexOffset;
int m_uniqueEdgesOffset;
int m_numUniqueEdges;
int m_unused;
} ConvexPolyhedronCL;
typedef struct
{
float4 m_plane;
int m_indexOffset;
int m_numIndices;
} btGpuFace;
#define SELECT_UINT4( b, a, condition ) select( b,a,condition )
#define make_float4 (float4)
#define make_float2 (float2)
#define make_uint4 (uint4)
#define make_int4 (int4)
#define make_uint2 (uint2)
#define make_int2 (int2)
__inline
float fastDiv(float numerator, float denominator)
{
return native_divide(numerator, denominator);
// return numerator/denominator;
}
__inline
float4 fastDiv4(float4 numerator, float4 denominator)
{
return native_divide(numerator, denominator);
}
__inline
float4 cross3(float4 a, float4 b)
{
return cross(a,b);
}
//#define dot3F4 dot
__inline
float dot3F4(float4 a, float4 b)
{
float4 a1 = make_float4(a.xyz,0.f);
float4 b1 = make_float4(b.xyz,0.f);
return dot(a1, b1);
}
__inline
float4 fastNormalize4(float4 v)
{
return fast_normalize(v);
}
///////////////////////////////////////
// Quaternion
///////////////////////////////////////
typedef float4 Quaternion;
__inline
Quaternion qtMul(Quaternion a, Quaternion b);
__inline
Quaternion qtNormalize(Quaternion in);
__inline
float4 qtRotate(Quaternion q, float4 vec);
__inline
Quaternion qtInvert(Quaternion q);
__inline
Quaternion qtMul(Quaternion a, Quaternion b)
{
Quaternion ans;
ans = cross3( a, b );
ans += a.w*b+b.w*a;
// ans.w = a.w*b.w - (a.x*b.x+a.y*b.y+a.z*b.z);
ans.w = a.w*b.w - dot3F4(a, b);
return ans;
}
__inline
Quaternion qtNormalize(Quaternion in)
{
return fastNormalize4(in);
// in /= length( in );
// return in;
}
__inline
float4 qtRotate(Quaternion q, float4 vec)
{
Quaternion qInv = qtInvert( q );
float4 vcpy = vec;
vcpy.w = 0.f;
float4 out = qtMul(qtMul(q,vcpy),qInv);
return out;
}
__inline
Quaternion qtInvert(Quaternion q)
{
return (Quaternion)(-q.xyz, q.w);
}
__inline
float4 qtInvRotate(const Quaternion q, float4 vec)
{
return qtRotate( qtInvert( q ), vec );
}
__inline
float4 transform(const float4* p, const float4* translation, const Quaternion* orientation)
{
return qtRotate( *orientation, *p ) + (*translation);
}
void trInverse(float4 translationIn, Quaternion orientationIn,
float4* translationOut, Quaternion* orientationOut)
{
*orientationOut = qtInvert(orientationIn);
*translationOut = qtRotate(*orientationOut, -translationIn);
}
void trMul(float4 translationA, Quaternion orientationA,
float4 translationB, Quaternion orientationB,
float4* translationOut, Quaternion* orientationOut)
{
*orientationOut = qtMul(orientationA,orientationB);
*translationOut = transform(&translationB,&translationA,&orientationA);
}
__inline
float4 normalize3(const float4 a)
{
float4 n = make_float4(a.x, a.y, a.z, 0.f);
return fastNormalize4( n );
}
__inline float4 lerp3(const float4 a,const float4 b, float t)
{
return make_float4( a.x + (b.x - a.x) * t,
a.y + (b.y - a.y) * t,
a.z + (b.z - a.z) * t,
0.f);
}
float signedDistanceFromPointToPlane(float4 point, float4 planeEqn, float4* closestPointOnFace)
{
float4 n = (float4)(planeEqn.x, planeEqn.y, planeEqn.z, 0);
float dist = dot3F4(n, point) + planeEqn.w;
*closestPointOnFace = point - dist * n;
return dist;
}
inline bool IsPointInPolygon(float4 p,
float4 posConvex,
float4 ornConvex,
const btGpuFace* face,
__global const float4* baseVertex,
__global const int* convexIndices,
float4* out)
{
float4 a;
float4 b;
float4 ab;
float4 ap;
float4 v;
float4 plane = make_float4(face->m_plane.x,face->m_plane.y,face->m_plane.z,0.f);
if (face->m_numIndices<2)
return false;
float4 v0 = baseVertex[convexIndices[face->m_indexOffset + face->m_numIndices-1]];
float4 worldV0 = transform(&v0, &posConvex, &ornConvex);
b = worldV0;
for(unsigned i=0; i != face->m_numIndices; ++i)
{
a = b;
float4 vi = baseVertex[convexIndices[face->m_indexOffset + i]];
float4 worldVi = transform(&vi, &posConvex, &ornConvex);
b = worldVi;
ab = b-a;
ap = p-a;
v = cross3(ab,plane);
if (dot(ap, v) > 0.f)
{
float ab_m2 = dot(ab, ab);
float rt = ab_m2 != 0.f ? dot(ab, ap) / ab_m2 : 0.f;
if (rt <= 0.f)
{
*out = a;
}
else if (rt >= 1.f)
{
*out = b;
}
else
{
float s = 1.f - rt;
out[0].x = s * a.x + rt * b.x;
out[0].y = s * a.y + rt * b.y;
out[0].z = s * a.z + rt * b.z;
}
return false;
}
}
return true;
}
void computeContactSphereConvex(int pairIndex,
int bodyIndexA, int bodyIndexB,
int collidableIndexA, int collidableIndexB,
__global const BodyData* rigidBodies,
__global const btCollidableGpu* collidables,
__global const ConvexPolyhedronCL* convexShapes,
__global const float4* convexVertices,
__global const int* convexIndices,
__global const btGpuFace* faces,
__global Contact4* restrict globalContactsOut,
counter32_t nGlobalContactsOut,
int numPairs)
{
float radius = collidables[collidableIndexA].m_radius;
float4 spherePos1 = rigidBodies[bodyIndexA].m_pos;
float4 sphereOrn = rigidBodies[bodyIndexA].m_quat;
float4 pos = rigidBodies[bodyIndexB].m_pos;
float4 quat = rigidBodies[bodyIndexB].m_quat;
float4 spherePos = spherePos1 - pos;
int collidableIndex = rigidBodies[bodyIndexB].m_collidableIdx;
int shapeIndex = collidables[collidableIndex].m_shapeIndex;
int numFaces = convexShapes[shapeIndex].m_numFaces;
float4 closestPnt = (float4)(0, 0, 0, 0);
float4 hitNormalWorld = (float4)(0, 0, 0, 0);
float minDist = -1000000.f;
bool bCollide = true;
for ( int f = 0; f < numFaces; f++ )
{
btGpuFace face = faces[convexShapes[shapeIndex].m_faceOffset+f];
// set up a plane equation
float4 planeEqn;
float4 n1 = qtRotate(quat, (float4)(face.m_plane.xyz, 0));
planeEqn = n1;
planeEqn.w = face.m_plane.w;
// compute a signed distance from the vertex in cloth to the face of rigidbody.
float4 pntReturn;
float dist = signedDistanceFromPointToPlane(spherePos, planeEqn, &pntReturn);
// If the distance is positive, the plane is a separating plane.
if ( dist > radius )
{
bCollide = false;
break;
}
if (dist>0)
{
//might hit an edge or vertex
float4 out;
bool isInPoly = IsPointInPolygon(spherePos,
pos,
quat,
&face,
&convexVertices[convexShapes[shapeIndex].m_vertexOffset],
convexIndices,
&out);
if (isInPoly)
{
if (dist>minDist)
{
minDist = dist;
closestPnt = pntReturn;
hitNormalWorld = planeEqn;
}
} else
{
float4 tmp = spherePos-out;
float l2 = dot(tmp,tmp);
if (l2<radius*radius)
{
dist = sqrt(l2);
if (dist>minDist)
{
minDist = dist;
closestPnt = out;
hitNormalWorld = tmp/dist;
}
} else
{
bCollide = false;
break;
}
}
} else
{
if ( dist > minDist )
{
minDist = dist;
closestPnt = pntReturn;
hitNormalWorld.xyz = planeEqn.xyz;
}
}
}
if (bCollide)
{
float4 normalOnSurfaceB1 = -hitNormalWorld;
float4 pOnB1 = closestPnt+pos;
float actualDepth = minDist-radius;
pOnB1.w = actualDepth;
int dstIdx;
AppendInc( nGlobalContactsOut, dstIdx );
if (dstIdx < numPairs)
{
__global Contact4* c = &globalContactsOut[dstIdx];
c->m_worldNormal = normalOnSurfaceB1;
c->m_coeffs = (u32)(0.f*0xffff) | ((u32)(0.7f*0xffff)<<16);
c->m_batchIdx = pairIndex;
c->m_bodyAPtrAndSignBit = rigidBodies[bodyIndexA].m_invMass==0?-bodyIndexA:bodyIndexA;
c->m_bodyBPtrAndSignBit = rigidBodies[bodyIndexB].m_invMass==0?-bodyIndexB:bodyIndexB;
c->m_worldPos[0] = pOnB1;
GET_NPOINTS(*c) = 1;
}//if (dstIdx < numPairs)
}//if (hasCollision)
}
void computeContactPlaneConvex(int pairIndex,
int bodyIndexA, int bodyIndexB,
int collidableIndexA, int collidableIndexB,
__global const BodyData* rigidBodies,
__global const btCollidableGpu* collidables,
__global const btGpuFace* faces,
__global Contact4* restrict globalContactsOut,
counter32_t nGlobalContactsOut,
int numPairs)
{
float4 planeEq = faces[collidables[collidableIndexA].m_shapeIndex].m_plane;
float radius = collidables[collidableIndexB].m_radius;
float4 posA1 = rigidBodies[bodyIndexA].m_pos;
float4 ornA1 = rigidBodies[bodyIndexA].m_quat;
float4 posB1 = rigidBodies[bodyIndexB].m_pos;
float4 ornB1 = rigidBodies[bodyIndexB].m_quat;
bool hasCollision = false;
float4 planeNormal1 = make_float4(planeEq.x,planeEq.y,planeEq.z,0.f);
float planeConstant = planeEq.w;
float4 convexInPlaneTransPos1; Quaternion convexInPlaneTransOrn1;
{
float4 invPosA;Quaternion invOrnA;
trInverse(posA1,ornA1,&invPosA,&invOrnA);
trMul(invPosA,invOrnA,posB1,ornB1,&convexInPlaneTransPos1,&convexInPlaneTransOrn1);
}
float4 planeInConvexPos1; Quaternion planeInConvexOrn1;
{
float4 invPosB;Quaternion invOrnB;
trInverse(posB1,ornB1,&invPosB,&invOrnB);
trMul(invPosB,invOrnB,posA1,ornA1,&planeInConvexPos1,&planeInConvexOrn1);
}
float4 vtx1 = qtRotate(planeInConvexOrn1,-planeNormal1)*radius;
float4 vtxInPlane1 = transform(&vtx1,&convexInPlaneTransPos1,&convexInPlaneTransOrn1);
float distance = dot3F4(planeNormal1,vtxInPlane1) - planeConstant;
hasCollision = distance < 0.f;//m_manifoldPtr->getContactBreakingThreshold();
if (hasCollision)
{
float4 vtxInPlaneProjected1 = vtxInPlane1 - distance*planeNormal1;
float4 vtxInPlaneWorld1 = transform(&vtxInPlaneProjected1,&posA1,&ornA1);
float4 normalOnSurfaceB1 = qtRotate(ornA1,planeNormal1);
float4 pOnB1 = vtxInPlaneWorld1+normalOnSurfaceB1*distance;
pOnB1.w = distance;
int dstIdx;
AppendInc( nGlobalContactsOut, dstIdx );
if (dstIdx < numPairs)
{
__global Contact4* c = &globalContactsOut[dstIdx];
c->m_worldNormal = normalOnSurfaceB1;
c->m_coeffs = (u32)(0.f*0xffff) | ((u32)(0.7f*0xffff)<<16);
c->m_batchIdx = pairIndex;
c->m_bodyAPtrAndSignBit = rigidBodies[bodyIndexA].m_invMass==0?-bodyIndexA:bodyIndexA;
c->m_bodyBPtrAndSignBit = rigidBodies[bodyIndexB].m_invMass==0?-bodyIndexB:bodyIndexB;
c->m_worldPos[0] = pOnB1;
GET_NPOINTS(*c) = 1;
}//if (dstIdx < numPairs)
}//if (hasCollision)
}
__kernel void primitiveContactsKernel( __global const int2* pairs,
__global const BodyData* rigidBodies,
__global const btCollidableGpu* collidables,
__global const ConvexPolyhedronCL* convexShapes,
__global const float4* vertices,
__global const float4* uniqueEdges,
__global const btGpuFace* faces,
__global const int* indices,
__global Contact4* restrict globalContactsOut,
counter32_t nGlobalContactsOut,
int numPairs)
{
int i = get_global_id(0);
int pairIndex = i;
float4 worldVertsB1[64];
float4 worldVertsB2[64];
int capacityWorldVerts = 64;
float4 localContactsOut[64];
int localContactCapacity=64;
float minDist = -1e30f;
float maxDist = 0.02f;
if (i<numPairs)
{
int bodyIndexA = pairs[i].x;
int bodyIndexB = pairs[i].y;
int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;
int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;
if (collidables[collidableIndexA].m_shapeType == SHAPE_SPHERE &&
collidables[collidableIndexB].m_shapeType == SHAPE_PLANE)
{
computeContactPlaneConvex( pairIndex, bodyIndexB,bodyIndexA, collidableIndexB,collidableIndexA,
rigidBodies,collidables,faces, globalContactsOut, nGlobalContactsOut,numPairs);
return;
}
if (collidables[collidableIndexA].m_shapeType == SHAPE_PLANE &&
collidables[collidableIndexB].m_shapeType == SHAPE_SPHERE)
{
computeContactPlaneConvex(pairIndex, bodyIndexA, bodyIndexB, collidableIndexA, collidableIndexB,
rigidBodies,collidables,faces, globalContactsOut, nGlobalContactsOut,numPairs);
return;
}
if (collidables[collidableIndexA].m_shapeType == SHAPE_SPHERE &&
collidables[collidableIndexB].m_shapeType == SHAPE_CONVEX_HULL)
{
computeContactSphereConvex(pairIndex, bodyIndexA, bodyIndexB, collidableIndexA, collidableIndexB,
rigidBodies,collidables,convexShapes,vertices,indices,faces, globalContactsOut, nGlobalContactsOut,numPairs);
return;
}
if (collidables[collidableIndexA].m_shapeType == SHAPE_CONVEX_HULL &&
collidables[collidableIndexB].m_shapeType == SHAPE_SPHERE)
{
computeContactSphereConvex(pairIndex, bodyIndexB, bodyIndexA, collidableIndexB, collidableIndexA,
rigidBodies,collidables,convexShapes,vertices,indices,faces, globalContactsOut, nGlobalContactsOut,numPairs);
return;
}
if (collidables[collidableIndexA].m_shapeType == SHAPE_SPHERE &&
collidables[collidableIndexB].m_shapeType == SHAPE_SPHERE)
{
//sphere-sphere
float radiusA = collidables[collidableIndexA].m_radius;
float radiusB = collidables[collidableIndexB].m_radius;
float4 posA = rigidBodies[bodyIndexA].m_pos;
float4 posB = rigidBodies[bodyIndexB].m_pos;
float4 diff = posA-posB;
float len = length(diff);
///iff distance positive, don't generate a new contact
if ( len <= (radiusA+radiusB))
{
///distance (negative means penetration)
float dist = len - (radiusA+radiusB);
float4 normalOnSurfaceB = make_float4(1.f,0.f,0.f,0.f);
if (len > 0.00001)
{
normalOnSurfaceB = diff / len;
}
float4 contactPosB = posB + normalOnSurfaceB*radiusB;
contactPosB.w = dist;
int dstIdx;
AppendInc( nGlobalContactsOut, dstIdx );
if (dstIdx < numPairs)
{
__global Contact4* c = &globalContactsOut[dstIdx];
c->m_worldNormal = -normalOnSurfaceB;
c->m_coeffs = (u32)(0.f*0xffff) | ((u32)(0.7f*0xffff)<<16);
c->m_batchIdx = pairIndex;
int bodyA = pairs[pairIndex].x;
int bodyB = pairs[pairIndex].y;
c->m_bodyAPtrAndSignBit = rigidBodies[bodyA].m_invMass==0?-bodyA:bodyA;
c->m_bodyBPtrAndSignBit = rigidBodies[bodyB].m_invMass==0?-bodyB:bodyB;
c->m_worldPos[0] = contactPosB;
GET_NPOINTS(*c) = 1;
}//if (dstIdx < numPairs)
}//if ( len <= (radiusA+radiusB))
return;
}//SHAPE_SPHERE SHAPE_SPHERE
}// if (i<numPairs)
}

671
opencl/gpu_sat/kernels/primitiveContacts.h Normal file
View File

@@ -0,0 +1,671 @@
//this file is autogenerated using stringify.bat (premake --stringify) in the build folder of this project
static const char* primitiveContactsKernelsCL= \
"#define TRIANGLE_NUM_CONVEX_FACES 5\n"
"\n"
"#define SHAPE_CONVEX_HULL 3\n"
"#define SHAPE_PLANE 4\n"
"#define SHAPE_CONCAVE_TRIMESH 5\n"
"#define SHAPE_COMPOUND_OF_CONVEX_HULLS 6\n"
"#define SHAPE_SPHERE 7\n"
"\n"
"\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"
"\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"
"\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"
"\n"
"#define max2 max\n"
"#define min2 min\n"
"\n"
"typedef unsigned int u32;\n"
"\n"
"\n"
"\n"
"typedef struct\n"
"{\n"
" float4 m_worldPos[4];\n"
" float4 m_worldNormal; // w: m_nPoints\n"
" u32 m_coeffs;\n"
" u32 m_batchIdx;\n"
"\n"
" int m_bodyAPtrAndSignBit;//x:m_bodyAPtr, y:m_bodyBPtr\n"
" int m_bodyBPtrAndSignBit;\n"
"} Contact4;\n"
"\n"
"\n"
"///keep this in sync with btCollidable.h\n"
"typedef struct\n"
"{\n"
" int m_numChildShapes;\n"
" float m_radius;\n"
" int m_shapeType;\n"
" int m_shapeIndex;\n"
" \n"
"} btCollidableGpu;\n"
"\n"
"typedef struct\n"
"{\n"
" float4 m_childPosition;\n"
" float4 m_childOrientation;\n"
" int m_shapeIndex;\n"
" int m_unused0;\n"
" int m_unused1;\n"
" int m_unused2;\n"
"} btGpuChildShape;\n"
"\n"
"#define GET_NPOINTS(x) (x).m_worldNormal.w\n"
"\n"
"typedef struct\n"
"{\n"
" float4 m_pos;\n"
" float4 m_quat;\n"
" float4 m_linVel;\n"
" float4 m_angVel;\n"
"\n"
" u32 m_collidableIdx; \n"
" float m_invMass;\n"
" float m_restituitionCoeff;\n"
" float m_frictionCoeff;\n"
"} BodyData;\n"
"\n"
"\n"
"typedef struct \n"
"{\n"
" float4 m_localCenter;\n"
" float4 m_extents;\n"
" float4 mC;\n"
" float4 mE;\n"
" \n"
" float m_radius;\n"
" int m_faceOffset;\n"
" int m_numFaces;\n"
" int m_numVertices;\n"
" \n"
" int m_vertexOffset;\n"
" int m_uniqueEdgesOffset;\n"
" int m_numUniqueEdges;\n"
" int m_unused;\n"
"\n"
"} ConvexPolyhedronCL;\n"
"\n"
"typedef struct\n"
"{\n"
" float4 m_plane;\n"
" int m_indexOffset;\n"
" int m_numIndices;\n"
"} btGpuFace;\n"
"\n"
"#define SELECT_UINT4( b, a, condition ) select( b,a,condition )\n"
"\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"
"\n"
"\n"
"__inline\n"
"float fastDiv(float numerator, float denominator)\n"
"{\n"
" return native_divide(numerator, denominator); \n"
"// return numerator/denominator; \n"
"}\n"
"\n"
"__inline\n"
"float4 fastDiv4(float4 numerator, float4 denominator)\n"
"{\n"
" return native_divide(numerator, denominator); \n"
"}\n"
"\n"
"\n"
"__inline\n"
"float4 cross3(float4 a, float4 b)\n"
"{\n"
" return cross(a,b);\n"
"}\n"
"\n"
"//#define dot3F4 dot\n"
"\n"
"__inline\n"
"float dot3F4(float4 a, float4 b)\n"
"{\n"
" float4 a1 = make_float4(a.xyz,0.f);\n"
" float4 b1 = make_float4(b.xyz,0.f);\n"
" return dot(a1, b1);\n"
"}\n"
"\n"
"__inline\n"
"float4 fastNormalize4(float4 v)\n"
"{\n"
" return fast_normalize(v);\n"
"}\n"
"\n"
"\n"
"///////////////////////////////////////\n"
"// Quaternion\n"
"///////////////////////////////////////\n"
"\n"
"typedef float4 Quaternion;\n"
"\n"
"__inline\n"
"Quaternion qtMul(Quaternion a, Quaternion b);\n"
"\n"
"__inline\n"
"Quaternion qtNormalize(Quaternion in);\n"
"\n"
"__inline\n"
"float4 qtRotate(Quaternion q, float4 vec);\n"
"\n"
"__inline\n"
"Quaternion qtInvert(Quaternion q);\n"
"\n"
"\n"
"\n"
"\n"
"__inline\n"
"Quaternion qtMul(Quaternion a, Quaternion b)\n"
"{\n"
" Quaternion ans;\n"
" ans = cross3( a, b );\n"
" ans += a.w*b+b.w*a;\n"
"// ans.w = a.w*b.w - (a.x*b.x+a.y*b.y+a.z*b.z);\n"
" ans.w = a.w*b.w - dot3F4(a, b);\n"
" return ans;\n"
"}\n"
"\n"
"__inline\n"
"Quaternion qtNormalize(Quaternion in)\n"
"{\n"
" return fastNormalize4(in);\n"
"// in /= length( in );\n"
"// return in;\n"
"}\n"
"__inline\n"
"float4 qtRotate(Quaternion q, float4 vec)\n"
"{\n"
" Quaternion qInv = qtInvert( q );\n"
" float4 vcpy = vec;\n"
" vcpy.w = 0.f;\n"
" float4 out = qtMul(qtMul(q,vcpy),qInv);\n"
" return out;\n"
"}\n"
"\n"
"__inline\n"
"Quaternion qtInvert(Quaternion q)\n"
"{\n"
" return (Quaternion)(-q.xyz, q.w);\n"
"}\n"
"\n"
"__inline\n"
"float4 qtInvRotate(const Quaternion q, float4 vec)\n"
"{\n"
" return qtRotate( qtInvert( q ), vec );\n"
"}\n"
"\n"
"__inline\n"
"float4 transform(const float4* p, const float4* translation, const Quaternion* orientation)\n"
"{\n"
" return qtRotate( *orientation, *p ) + (*translation);\n"
"}\n"
"\n"
"void trInverse(float4 translationIn, Quaternion orientationIn,\n"
" float4* translationOut, Quaternion* orientationOut)\n"
"{\n"
" *orientationOut = qtInvert(orientationIn);\n"
" *translationOut = qtRotate(*orientationOut, -translationIn);\n"
"}\n"
"\n"
"void trMul(float4 translationA, Quaternion orientationA,\n"
" float4 translationB, Quaternion orientationB,\n"
" float4* translationOut, Quaternion* orientationOut)\n"
"{\n"
" *orientationOut = qtMul(orientationA,orientationB);\n"
" *translationOut = transform(&translationB,&translationA,&orientationA);\n"
"}\n"
"\n"
"\n"
"\n"
"__inline\n"
"float4 normalize3(const float4 a)\n"
"{\n"
" float4 n = make_float4(a.x, a.y, a.z, 0.f);\n"
" return fastNormalize4( n );\n"
"}\n"
"\n"
"\n"
"__inline float4 lerp3(const float4 a,const float4 b, float t)\n"
"{\n"
" return make_float4( a.x + (b.x - a.x) * t,\n"
" a.y + (b.y - a.y) * t,\n"
" a.z + (b.z - a.z) * t,\n"
" 0.f);\n"
"}\n"
"\n"
"\n"
"float signedDistanceFromPointToPlane(float4 point, float4 planeEqn, float4* closestPointOnFace)\n"
"{\n"
" float4 n = (float4)(planeEqn.x, planeEqn.y, planeEqn.z, 0);\n"
" float dist = dot3F4(n, point) + planeEqn.w;\n"
" *closestPointOnFace = point - dist * n;\n"
" return dist;\n"
"}\n"
"\n"
"\n"
"\n"
"inline bool IsPointInPolygon(float4 p, \n"
" float4 posConvex,\n"
" float4 ornConvex,\n"
" const btGpuFace* face,\n"
" __global const float4* baseVertex,\n"
" __global const int* convexIndices,\n"
" float4* out)\n"
"{\n"
" float4 a;\n"
" float4 b;\n"
" float4 ab;\n"
" float4 ap;\n"
" float4 v;\n"
"\n"
" float4 plane = make_float4(face->m_plane.x,face->m_plane.y,face->m_plane.z,0.f);\n"
" \n"
" if (face->m_numIndices<2)\n"
" return false;\n"
"\n"
" \n"
" float4 v0 = baseVertex[convexIndices[face->m_indexOffset + face->m_numIndices-1]];\n"
" float4 worldV0 = transform(&v0, &posConvex, &ornConvex);\n"
" \n"
" b = worldV0;\n"
"\n"
" for(unsigned i=0; i != face->m_numIndices; ++i)\n"
" {\n"
" a = b;\n"
" float4 vi = baseVertex[convexIndices[face->m_indexOffset + i]];\n"
" float4 worldVi = transform(&vi, &posConvex, &ornConvex);\n"
" b = worldVi;\n"
" ab = b-a;\n"
" ap = p-a;\n"
" v = cross3(ab,plane);\n"
"\n"
" if (dot(ap, v) > 0.f)\n"
" {\n"
" float ab_m2 = dot(ab, ab);\n"
" float rt = ab_m2 != 0.f ? dot(ab, ap) / ab_m2 : 0.f;\n"
" if (rt <= 0.f)\n"
" {\n"
" *out = a;\n"
" }\n"
" else if (rt >= 1.f) \n"
" {\n"
" *out = b;\n"
" }\n"
" else\n"
" {\n"
" float s = 1.f - rt;\n"
" out[0].x = s * a.x + rt * b.x;\n"
" out[0].y = s * a.y + rt * b.y;\n"
" out[0].z = s * a.z + rt * b.z;\n"
" }\n"
" return false;\n"
" }\n"
" }\n"
" return true;\n"
"}\n"
"\n"
"\n"
"\n"
"\n"
"void computeContactSphereConvex(int pairIndex,\n"
" int bodyIndexA, int bodyIndexB, \n"
" int collidableIndexA, int collidableIndexB, \n"
" __global const BodyData* rigidBodies, \n"
" __global const btCollidableGpu* collidables,\n"
" __global const ConvexPolyhedronCL* convexShapes,\n"
" __global const float4* convexVertices,\n"
" __global const int* convexIndices,\n"
" __global const btGpuFace* faces,\n"
" __global Contact4* restrict globalContactsOut,\n"
" counter32_t nGlobalContactsOut,\n"
" int numPairs)\n"
"{\n"
"\n"
" float radius = collidables[collidableIndexA].m_radius;\n"
" float4 spherePos1 = rigidBodies[bodyIndexA].m_pos;\n"
" float4 sphereOrn = rigidBodies[bodyIndexA].m_quat;\n"
"\n"
"\n"
"\n"
" float4 pos = rigidBodies[bodyIndexB].m_pos;\n"
" float4 quat = rigidBodies[bodyIndexB].m_quat;\n"
"\n"
" float4 spherePos = spherePos1 - pos;\n"
"\n"
" int collidableIndex = rigidBodies[bodyIndexB].m_collidableIdx;\n"
" int shapeIndex = collidables[collidableIndex].m_shapeIndex;\n"
" int numFaces = convexShapes[shapeIndex].m_numFaces;\n"
" float4 closestPnt = (float4)(0, 0, 0, 0);\n"
" float4 hitNormalWorld = (float4)(0, 0, 0, 0);\n"
" float minDist = -1000000.f;\n"
" bool bCollide = true;\n"
"\n"
" for ( int f = 0; f < numFaces; f++ )\n"
" {\n"
" btGpuFace face = faces[convexShapes[shapeIndex].m_faceOffset+f];\n"
"\n"
" // set up a plane equation \n"
" float4 planeEqn;\n"
" float4 n1 = qtRotate(quat, (float4)(face.m_plane.xyz, 0));\n"
" planeEqn = n1;\n"
" planeEqn.w = face.m_plane.w;\n"
" \n"
" \n"
" // compute a signed distance from the vertex in cloth to the face of rigidbody.\n"
" float4 pntReturn;\n"
" float dist = signedDistanceFromPointToPlane(spherePos, planeEqn, &pntReturn);\n"
"\n"
" // If the distance is positive, the plane is a separating plane. \n"
" if ( dist > radius )\n"
" {\n"
" bCollide = false;\n"
" break;\n"
" }\n"
"\n"
"\n"
" if (dist>0)\n"
" {\n"
" //might hit an edge or vertex\n"
" float4 out;\n"
" bool isInPoly = IsPointInPolygon(spherePos,\n"
" pos,\n"
" quat,\n"
" &face,\n"
" &convexVertices[convexShapes[shapeIndex].m_vertexOffset],\n"
" convexIndices,\n"
" &out);\n"
" if (isInPoly)\n"
" {\n"
" if (dist>minDist)\n"
" {\n"
" minDist = dist;\n"
" closestPnt = pntReturn;\n"
" hitNormalWorld = planeEqn;\n"
" \n"
" }\n"
" } else\n"
" {\n"
" float4 tmp = spherePos-out;\n"
" float l2 = dot(tmp,tmp);\n"
" if (l2<radius*radius)\n"
" {\n"
" dist = sqrt(l2);\n"
" if (dist>minDist)\n"
" {\n"
" minDist = dist;\n"
" closestPnt = out;\n"
" hitNormalWorld = tmp/dist;\n"
" \n"
" }\n"
" \n"
" } else\n"
" {\n"
" bCollide = false;\n"
" break;\n"
" }\n"
" }\n"
" } else\n"
" {\n"
" if ( dist > minDist )\n"
" {\n"
" minDist = dist;\n"
" closestPnt = pntReturn;\n"
" hitNormalWorld.xyz = planeEqn.xyz;\n"
" }\n"
" }\n"
" \n"
" }\n"
"\n"
" \n"
"\n"
" if (bCollide)\n"
" {\n"
" float4 normalOnSurfaceB1 = -hitNormalWorld;\n"
" float4 pOnB1 = closestPnt+pos;\n"
" float actualDepth = minDist-radius;\n"
" pOnB1.w = actualDepth;\n"
"\n"
" int dstIdx;\n"
" AppendInc( nGlobalContactsOut, dstIdx );\n"
" \n"
" if (dstIdx < numPairs)\n"
" {\n"
" __global Contact4* c = &globalContactsOut[dstIdx];\n"
" c->m_worldNormal = normalOnSurfaceB1;\n"
" c->m_coeffs = (u32)(0.f*0xffff) | ((u32)(0.7f*0xffff)<<16);\n"
" c->m_batchIdx = pairIndex;\n"
" c->m_bodyAPtrAndSignBit = rigidBodies[bodyIndexA].m_invMass==0?-bodyIndexA:bodyIndexA;\n"
" c->m_bodyBPtrAndSignBit = rigidBodies[bodyIndexB].m_invMass==0?-bodyIndexB:bodyIndexB;\n"
" c->m_worldPos[0] = pOnB1;\n"
" GET_NPOINTS(*c) = 1;\n"
" }//if (dstIdx < numPairs)\n"
" }//if (hasCollision)\n"
"\n"
"}\n"
" \n"
"\n"
" \n"
"void computeContactPlaneConvex(int pairIndex,\n"
" int bodyIndexA, int bodyIndexB, \n"
" int collidableIndexA, int collidableIndexB, \n"
" __global const BodyData* rigidBodies, \n"
" __global const btCollidableGpu* collidables,\n"
" __global const btGpuFace* faces,\n"
" __global Contact4* restrict globalContactsOut,\n"
" counter32_t nGlobalContactsOut,\n"
" int numPairs)\n"
"{\n"
" float4 planeEq = faces[collidables[collidableIndexA].m_shapeIndex].m_plane;\n"
" float radius = collidables[collidableIndexB].m_radius;\n"
" float4 posA1 = rigidBodies[bodyIndexA].m_pos;\n"
" float4 ornA1 = rigidBodies[bodyIndexA].m_quat;\n"
" float4 posB1 = rigidBodies[bodyIndexB].m_pos;\n"
" float4 ornB1 = rigidBodies[bodyIndexB].m_quat;\n"
" \n"
" bool hasCollision = false;\n"
" float4 planeNormal1 = make_float4(planeEq.x,planeEq.y,planeEq.z,0.f);\n"
" float planeConstant = planeEq.w;\n"
" float4 convexInPlaneTransPos1; Quaternion convexInPlaneTransOrn1;\n"
" {\n"
" float4 invPosA;Quaternion invOrnA;\n"
" trInverse(posA1,ornA1,&invPosA,&invOrnA);\n"
" trMul(invPosA,invOrnA,posB1,ornB1,&convexInPlaneTransPos1,&convexInPlaneTransOrn1);\n"
" }\n"
" float4 planeInConvexPos1; Quaternion planeInConvexOrn1;\n"
" {\n"
" float4 invPosB;Quaternion invOrnB;\n"
" trInverse(posB1,ornB1,&invPosB,&invOrnB);\n"
" trMul(invPosB,invOrnB,posA1,ornA1,&planeInConvexPos1,&planeInConvexOrn1); \n"
" }\n"
" float4 vtx1 = qtRotate(planeInConvexOrn1,-planeNormal1)*radius;\n"
" float4 vtxInPlane1 = transform(&vtx1,&convexInPlaneTransPos1,&convexInPlaneTransOrn1);\n"
" float distance = dot3F4(planeNormal1,vtxInPlane1) - planeConstant;\n"
" hasCollision = distance < 0.f;//m_manifoldPtr->getContactBreakingThreshold();\n"
" if (hasCollision)\n"
" {\n"
" float4 vtxInPlaneProjected1 = vtxInPlane1 - distance*planeNormal1;\n"
" float4 vtxInPlaneWorld1 = transform(&vtxInPlaneProjected1,&posA1,&ornA1);\n"
" float4 normalOnSurfaceB1 = qtRotate(ornA1,planeNormal1);\n"
" float4 pOnB1 = vtxInPlaneWorld1+normalOnSurfaceB1*distance;\n"
" pOnB1.w = distance;\n"
"\n"
" int dstIdx;\n"
" AppendInc( nGlobalContactsOut, dstIdx );\n"
" \n"
" if (dstIdx < numPairs)\n"
" {\n"
" __global Contact4* c = &globalContactsOut[dstIdx];\n"
" c->m_worldNormal = normalOnSurfaceB1;\n"
" c->m_coeffs = (u32)(0.f*0xffff) | ((u32)(0.7f*0xffff)<<16);\n"
" c->m_batchIdx = pairIndex;\n"
" c->m_bodyAPtrAndSignBit = rigidBodies[bodyIndexA].m_invMass==0?-bodyIndexA:bodyIndexA;\n"
" c->m_bodyBPtrAndSignBit = rigidBodies[bodyIndexB].m_invMass==0?-bodyIndexB:bodyIndexB;\n"
" c->m_worldPos[0] = pOnB1;\n"
" GET_NPOINTS(*c) = 1;\n"
" }//if (dstIdx < numPairs)\n"
" }//if (hasCollision)\n"
"}\n"
"\n"
"\n"
"\n"
"\n"
"__kernel void primitiveContactsKernel( __global const int2* pairs, \n"
" __global const BodyData* rigidBodies, \n"
" __global const btCollidableGpu* collidables,\n"
" __global const ConvexPolyhedronCL* convexShapes, \n"
" __global const float4* vertices,\n"
" __global const float4* uniqueEdges,\n"
" __global const btGpuFace* faces,\n"
" __global const int* indices,\n"
" __global Contact4* restrict globalContactsOut,\n"
" counter32_t nGlobalContactsOut,\n"
" int numPairs)\n"
"{\n"
"\n"
" int i = get_global_id(0);\n"
" int pairIndex = i;\n"
" \n"
" float4 worldVertsB1[64];\n"
" float4 worldVertsB2[64];\n"
" int capacityWorldVerts = 64; \n"
"\n"
" float4 localContactsOut[64];\n"
" int localContactCapacity=64;\n"
" \n"
" float minDist = -1e30f;\n"
" float maxDist = 0.02f;\n"
"\n"
" if (i<numPairs)\n"
" {\n"
"\n"
" int bodyIndexA = pairs[i].x;\n"
" int bodyIndexB = pairs[i].y;\n"
" \n"
" int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;\n"
" int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;\n"
" \n"
"\n"
" if (collidables[collidableIndexA].m_shapeType == SHAPE_SPHERE &&\n"
" collidables[collidableIndexB].m_shapeType == SHAPE_PLANE)\n"
" {\n"
"\n"
"\n"
" computeContactPlaneConvex( pairIndex, bodyIndexB,bodyIndexA, collidableIndexB,collidableIndexA, \n"
" rigidBodies,collidables,faces, globalContactsOut, nGlobalContactsOut,numPairs);\n"
" return;\n"
" }\n"
"\n"
" if (collidables[collidableIndexA].m_shapeType == SHAPE_PLANE &&\n"
" collidables[collidableIndexB].m_shapeType == SHAPE_SPHERE)\n"
" {\n"
"\n"
"\n"
" computeContactPlaneConvex(pairIndex, bodyIndexA, bodyIndexB, collidableIndexA, collidableIndexB, \n"
" rigidBodies,collidables,faces, globalContactsOut, nGlobalContactsOut,numPairs);\n"
" return;\n"
" \n"
" }\n"
" \n"
" if (collidables[collidableIndexA].m_shapeType == SHAPE_SPHERE &&\n"
" collidables[collidableIndexB].m_shapeType == SHAPE_CONVEX_HULL)\n"
" {\n"
" \n"
" computeContactSphereConvex(pairIndex, bodyIndexA, bodyIndexB, collidableIndexA, collidableIndexB, \n"
" rigidBodies,collidables,convexShapes,vertices,indices,faces, globalContactsOut, nGlobalContactsOut,numPairs);\n"
" return;\n"
" }\n"
"\n"
" if (collidables[collidableIndexA].m_shapeType == SHAPE_CONVEX_HULL &&\n"
" collidables[collidableIndexB].m_shapeType == SHAPE_SPHERE)\n"
" {\n"
" \n"
" computeContactSphereConvex(pairIndex, bodyIndexB, bodyIndexA, collidableIndexB, collidableIndexA, \n"
" rigidBodies,collidables,convexShapes,vertices,indices,faces, globalContactsOut, nGlobalContactsOut,numPairs);\n"
" return;\n"
" }\n"
" \n"
" \n"
" \n"
" \n"
" \n"
" if (collidables[collidableIndexA].m_shapeType == SHAPE_SPHERE &&\n"
" collidables[collidableIndexB].m_shapeType == SHAPE_SPHERE)\n"
" {\n"
" //sphere-sphere\n"
" float radiusA = collidables[collidableIndexA].m_radius;\n"
" float radiusB = collidables[collidableIndexB].m_radius;\n"
" float4 posA = rigidBodies[bodyIndexA].m_pos;\n"
" float4 posB = rigidBodies[bodyIndexB].m_pos;\n"
"\n"
" float4 diff = posA-posB;\n"
" float len = length(diff);\n"
" \n"
" ///iff distance positive, don't generate a new contact\n"
" if ( len <= (radiusA+radiusB))\n"
" {\n"
" ///distance (negative means penetration)\n"
" float dist = len - (radiusA+radiusB);\n"
" float4 normalOnSurfaceB = make_float4(1.f,0.f,0.f,0.f);\n"
" if (len > 0.00001)\n"
" {\n"
" normalOnSurfaceB = diff / len;\n"
" }\n"
" float4 contactPosB = posB + normalOnSurfaceB*radiusB;\n"
" contactPosB.w = dist;\n"
" \n"
" int dstIdx;\n"
" AppendInc( nGlobalContactsOut, dstIdx );\n"
" \n"
" if (dstIdx < numPairs)\n"
" {\n"
" __global Contact4* c = &globalContactsOut[dstIdx];\n"
" c->m_worldNormal = -normalOnSurfaceB;\n"
" c->m_coeffs = (u32)(0.f*0xffff) | ((u32)(0.7f*0xffff)<<16);\n"
" c->m_batchIdx = pairIndex;\n"
" int bodyA = pairs[pairIndex].x;\n"
" int bodyB = pairs[pairIndex].y;\n"
" c->m_bodyAPtrAndSignBit = rigidBodies[bodyA].m_invMass==0?-bodyA:bodyA;\n"
" c->m_bodyBPtrAndSignBit = rigidBodies[bodyB].m_invMass==0?-bodyB:bodyB;\n"
" c->m_worldPos[0] = contactPosB;\n"
" GET_NPOINTS(*c) = 1;\n"
" }//if (dstIdx < numPairs)\n"
" }//if ( len <= (radiusA+radiusB))\n"
"\n"
" return;\n"
" }//SHAPE_SPHERE SHAPE_SPHERE\n"
"\n"
" }// if (i<numPairs)\n"
"\n"
"}\n"
"\n"
;

146
opencl/gpu_sat/kernels/satClipHullContacts.cl
View File

@@ -1,9 +1,13 @@
#define TRIANGLE_NUM_CONVEX_FACES 5
#define SHAPE_CONVEX_HULL 3
#define SHAPE_PLANE 4
#define SHAPE_CONCAVE_TRIMESH 5
#define SHAPE_COMPOUND_OF_CONVEX_HULLS 6
#define SHAPE_SPHERE 7
#pragma OPENCL EXTENSION cl_amd_printf : enable
#pragma OPENCL EXTENSION cl_khr_local_int32_base_atomics : enable
#pragma OPENCL EXTENSION cl_khr_global_int32_base_atomics : enable
@@ -41,12 +45,8 @@ typedef struct
{
float4 m_worldPos[4];
float4 m_worldNormal; // w: m_nPoints
// float m_restituitionCoeff;
// float m_frictionCoeff;
u32 m_coeffs;
u32 m_batchIdx;
// int m_nPoints;
// int m_padding0;
int m_bodyAPtrAndSignBit;//x:m_bodyAPtr, y:m_bodyBPtr
int m_bodyBPtrAndSignBit;
@@ -872,11 +872,6 @@ int extractManifoldSequential(const float4* p, int nPoints, float4 nearNormal, i
contactIdx[2] = idx[2];
contactIdx[3] = idx[3];
// if( max00.y < 0.0f )
// contactIdx[0] = (int)max00.x;
//does this sort happen on GPU too?
//std::sort( contactIdx, contactIdx+4 );
return 4;
}
@@ -908,7 +903,7 @@ __kernel void extractManifoldAndAddContactKernel(__global const int2* pairs,
{
localPoints[i] = pointsIn[i];
}
// int contactIdx[4] = {-1,-1,-1,-1};
int contactIdx[4];// = {-1,-1,-1,-1};
contactIdx[0] = -1;
contactIdx[1] = -1;
@@ -954,66 +949,7 @@ void trMul(float4 translationA, Quaternion orientationA,
*translationOut = transform(&translationB,&translationA,&orientationA);
}
void computeContactPlaneConvex(int pairIndex,
int bodyIndexA, int bodyIndexB,
int collidableIndexA, int collidableIndexB,
__global const BodyData* rigidBodies,
__global const btCollidableGpu* collidables,
__global const btGpuFace* faces,
__global Contact4* restrict globalContactsOut,
counter32_t nGlobalContactsOut,
int numPairs)
{
float4 planeEq = faces[collidables[collidableIndexA].m_shapeIndex].m_plane;
float radius = collidables[collidableIndexB].m_radius;
float4 posA1 = rigidBodies[bodyIndexA].m_pos;
float4 ornA1 = rigidBodies[bodyIndexA].m_quat;
float4 posB1 = rigidBodies[bodyIndexB].m_pos;
float4 ornB1 = rigidBodies[bodyIndexB].m_quat;
bool hasCollision = false;
float4 planeNormal1 = make_float4(planeEq.x,planeEq.y,planeEq.z,0.f);
float planeConstant = planeEq.w;
float4 convexInPlaneTransPos1; Quaternion convexInPlaneTransOrn1;
{
float4 invPosA;Quaternion invOrnA;
trInverse(posA1,ornA1,&invPosA,&invOrnA);
trMul(invPosA,invOrnA,posB1,ornB1,&convexInPlaneTransPos1,&convexInPlaneTransOrn1);
}
float4 planeInConvexPos1; Quaternion planeInConvexOrn1;
{
float4 invPosB;Quaternion invOrnB;
trInverse(posB1,ornB1,&invPosB,&invOrnB);
trMul(invPosB,invOrnB,posA1,ornA1,&planeInConvexPos1,&planeInConvexOrn1);
}
float4 vtx1 = qtRotate(planeInConvexOrn1,-planeNormal1)*radius;
float4 vtxInPlane1 = transform(&vtx1,&convexInPlaneTransPos1,&convexInPlaneTransOrn1);
float distance = dot3F4(planeNormal1,vtxInPlane1) - planeConstant;
hasCollision = distance < 0.f;//m_manifoldPtr->getContactBreakingThreshold();
if (hasCollision)
{
float4 vtxInPlaneProjected1 = vtxInPlane1 - distance*planeNormal1;
float4 vtxInPlaneWorld1 = transform(&vtxInPlaneProjected1,&posA1,&ornA1);
float4 normalOnSurfaceB1 = qtRotate(ornA1,planeNormal1);
float4 pOnB1 = vtxInPlaneWorld1+normalOnSurfaceB1*distance;
pOnB1.w = distance;
int dstIdx;
AppendInc( nGlobalContactsOut, dstIdx );
if (dstIdx < numPairs)
{
__global Contact4* c = &globalContactsOut[dstIdx];
c->m_worldNormal = normalOnSurfaceB1;
c->m_coeffs = (u32)(0.f*0xffff) | ((u32)(0.7f*0xffff)<<16);
c->m_batchIdx = pairIndex;
c->m_bodyAPtrAndSignBit = rigidBodies[bodyIndexA].m_invMass==0?-bodyIndexA:bodyIndexA;
c->m_bodyBPtrAndSignBit = rigidBodies[bodyIndexB].m_invMass==0?-bodyIndexB:bodyIndexB;
c->m_worldPos[0] = pOnB1;
GET_NPOINTS(*c) = 1;
}//if (dstIdx < numPairs)
}//if (hasCollision)
}
__kernel void clipHullHullKernel( __global const int2* pairs,
@@ -1052,74 +988,6 @@ __kernel void clipHullHullKernel( __global const int2* pairs,
int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;
int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;
if (collidables[collidableIndexA].m_shapeType == SHAPE_SPHERE &&
collidables[collidableIndexB].m_shapeType == SHAPE_PLANE)
{
computeContactPlaneConvex( pairIndex, bodyIndexB,bodyIndexA, collidableIndexB,collidableIndexA,
rigidBodies,collidables,faces, globalContactsOut, nGlobalContactsOut,numPairs);
return;
}
if (collidables[collidableIndexA].m_shapeType == SHAPE_PLANE &&
collidables[collidableIndexB].m_shapeType == SHAPE_SPHERE)
{
computeContactPlaneConvex(pairIndex, bodyIndexA, bodyIndexB, collidableIndexA, collidableIndexB,
rigidBodies,collidables,faces, globalContactsOut, nGlobalContactsOut,numPairs);
return;
}
if (collidables[collidableIndexA].m_shapeType == SHAPE_SPHERE &&
collidables[collidableIndexB].m_shapeType == SHAPE_SPHERE)
{
//sphere-sphere
float radiusA = collidables[collidableIndexA].m_radius;
float radiusB = collidables[collidableIndexB].m_radius;
float4 posA = rigidBodies[bodyIndexA].m_pos;
float4 posB = rigidBodies[bodyIndexB].m_pos;
float4 diff = posA-posB;
float len = length(diff);
///iff distance positive, don't generate a new contact
if ( len <= (radiusA+radiusB))
{
///distance (negative means penetration)
float dist = len - (radiusA+radiusB);
float4 normalOnSurfaceB = make_float4(1.f,0.f,0.f,0.f);
if (len > 0.00001)
{
normalOnSurfaceB = diff / len;
}
float4 contactPosB = posB + normalOnSurfaceB*radiusB;
contactPosB.w = dist;
int dstIdx;
AppendInc( nGlobalContactsOut, dstIdx );
if (dstIdx < numPairs)
{
__global Contact4* c = &globalContactsOut[dstIdx];
c->m_worldNormal = -normalOnSurfaceB;
c->m_coeffs = (u32)(0.f*0xffff) | ((u32)(0.7f*0xffff)<<16);
c->m_batchIdx = pairIndex;
int bodyA = pairs[pairIndex].x;
int bodyB = pairs[pairIndex].y;
c->m_bodyAPtrAndSignBit = rigidBodies[bodyA].m_invMass==0?-bodyA:bodyA;
c->m_bodyBPtrAndSignBit = rigidBodies[bodyB].m_invMass==0?-bodyB:bodyB;
c->m_worldPos[0] = contactPosB;
GET_NPOINTS(*c) = 1;
}//if (dstIdx < numPairs)
}//if ( len <= (radiusA+radiusB))
return;
}//SHAPE_SPHERE SHAPE_SPHERE
if (hasSeparatingAxis[i])
{
@@ -1261,10 +1129,6 @@ __kernel void clipCompoundsHullHullKernel( __global const int4* gpuCompoundPai
int shapeIndexA = collidables[collidableIndexA].m_shapeIndex;
int shapeIndexB = collidables[collidableIndexB].m_shapeIndex;
int numLocalContactsOut = clipHullAgainstHull(gpuCompoundSepNormalsOut[i],
&convexShapes[shapeIndexA], &convexShapes[shapeIndexB],

146
opencl/gpu_sat/kernels/satClipHullContacts.h
View File

@@ -3,9 +3,13 @@ static const char* satClipKernelsCL= \
"\n"
"#define TRIANGLE_NUM_CONVEX_FACES 5\n"
"\n"
"#define SHAPE_CONVEX_HULL 3\n"
"#define SHAPE_PLANE 4\n"
"#define SHAPE_CONCAVE_TRIMESH 5\n"
"#define SHAPE_COMPOUND_OF_CONVEX_HULLS 6\n"
"#define SHAPE_SPHERE 7\n"
"\n"
"\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"
@@ -43,12 +47,8 @@ static const char* satClipKernelsCL= \
"{\n"
" float4 m_worldPos[4];\n"
" float4 m_worldNormal; // w: m_nPoints\n"
"// float m_restituitionCoeff;\n"
"// float m_frictionCoeff;\n"
" u32 m_coeffs;\n"
" u32 m_batchIdx;\n"
"// int m_nPoints;\n"
"// int m_padding0;\n"
"\n"
" int m_bodyAPtrAndSignBit;//x:m_bodyAPtr, y:m_bodyBPtr\n"
" int m_bodyBPtrAndSignBit;\n"
@@ -874,11 +874,6 @@ static const char* satClipKernelsCL= \
" contactIdx[2] = idx[2];\n"
" contactIdx[3] = idx[3];\n"
"\n"
"// if( max00.y < 0.0f )\n"
"// contactIdx[0] = (int)max00.x;\n"
"\n"
" //does this sort happen on GPU too?\n"
" //std::sort( contactIdx, contactIdx+4 );\n"
"\n"
" return 4;\n"
" }\n"
@@ -910,7 +905,7 @@ static const char* satClipKernelsCL= \
" {\n"
" localPoints[i] = pointsIn[i];\n"
" }\n"
"// int contactIdx[4] = {-1,-1,-1,-1};\n"
"\n"
" int contactIdx[4];// = {-1,-1,-1,-1};\n"
" contactIdx[0] = -1;\n"
" contactIdx[1] = -1;\n"
@@ -956,66 +951,7 @@ static const char* satClipKernelsCL= \
" *translationOut = transform(&translationB,&translationA,&orientationA);\n"
"}\n"
"\n"
"void computeContactPlaneConvex(int pairIndex,\n"
" int bodyIndexA, int bodyIndexB, \n"
" int collidableIndexA, int collidableIndexB, \n"
" __global const BodyData* rigidBodies, \n"
" __global const btCollidableGpu* collidables,\n"
" __global const btGpuFace* faces,\n"
" __global Contact4* restrict globalContactsOut,\n"
" counter32_t nGlobalContactsOut,\n"
" int numPairs)\n"
"{\n"
" float4 planeEq = faces[collidables[collidableIndexA].m_shapeIndex].m_plane;\n"
" float radius = collidables[collidableIndexB].m_radius;\n"
" float4 posA1 = rigidBodies[bodyIndexA].m_pos;\n"
" float4 ornA1 = rigidBodies[bodyIndexA].m_quat;\n"
" float4 posB1 = rigidBodies[bodyIndexB].m_pos;\n"
" float4 ornB1 = rigidBodies[bodyIndexB].m_quat;\n"
" \n"
" bool hasCollision = false;\n"
" float4 planeNormal1 = make_float4(planeEq.x,planeEq.y,planeEq.z,0.f);\n"
" float planeConstant = planeEq.w;\n"
" float4 convexInPlaneTransPos1; Quaternion convexInPlaneTransOrn1;\n"
" {\n"
" float4 invPosA;Quaternion invOrnA;\n"
" trInverse(posA1,ornA1,&invPosA,&invOrnA);\n"
" trMul(invPosA,invOrnA,posB1,ornB1,&convexInPlaneTransPos1,&convexInPlaneTransOrn1);\n"
" }\n"
" float4 planeInConvexPos1; Quaternion planeInConvexOrn1;\n"
" {\n"
" float4 invPosB;Quaternion invOrnB;\n"
" trInverse(posB1,ornB1,&invPosB,&invOrnB);\n"
" trMul(invPosB,invOrnB,posA1,ornA1,&planeInConvexPos1,&planeInConvexOrn1); \n"
" }\n"
" float4 vtx1 = qtRotate(planeInConvexOrn1,-planeNormal1)*radius;\n"
" float4 vtxInPlane1 = transform(&vtx1,&convexInPlaneTransPos1,&convexInPlaneTransOrn1);\n"
" float distance = dot3F4(planeNormal1,vtxInPlane1) - planeConstant;\n"
" hasCollision = distance < 0.f;//m_manifoldPtr->getContactBreakingThreshold();\n"
" if (hasCollision)\n"
" {\n"
" float4 vtxInPlaneProjected1 = vtxInPlane1 - distance*planeNormal1;\n"
" float4 vtxInPlaneWorld1 = transform(&vtxInPlaneProjected1,&posA1,&ornA1);\n"
" float4 normalOnSurfaceB1 = qtRotate(ornA1,planeNormal1);\n"
" float4 pOnB1 = vtxInPlaneWorld1+normalOnSurfaceB1*distance;\n"
" pOnB1.w = distance;\n"
"\n"
" int dstIdx;\n"
" AppendInc( nGlobalContactsOut, dstIdx );\n"
" \n"
" if (dstIdx < numPairs)\n"
" {\n"
" __global Contact4* c = &globalContactsOut[dstIdx];\n"
" c->m_worldNormal = normalOnSurfaceB1;\n"
" c->m_coeffs = (u32)(0.f*0xffff) | ((u32)(0.7f*0xffff)<<16);\n"
" c->m_batchIdx = pairIndex;\n"
" c->m_bodyAPtrAndSignBit = rigidBodies[bodyIndexA].m_invMass==0?-bodyIndexA:bodyIndexA;\n"
" c->m_bodyBPtrAndSignBit = rigidBodies[bodyIndexB].m_invMass==0?-bodyIndexB:bodyIndexB;\n"
" c->m_worldPos[0] = pOnB1;\n"
" GET_NPOINTS(*c) = 1;\n"
" }//if (dstIdx < numPairs)\n"
" }//if (hasCollision)\n"
"}\n"
"\n"
"\n"
"__kernel void clipHullHullKernel( __global const int2* pairs, \n"
@@ -1054,74 +990,6 @@ static const char* satClipKernelsCL= \
" \n"
" int collidableIndexA = rigidBodies[bodyIndexA].m_collidableIdx;\n"
" int collidableIndexB = rigidBodies[bodyIndexB].m_collidableIdx;\n"
" \n"
"\n"
" if (collidables[collidableIndexA].m_shapeType == SHAPE_SPHERE &&\n"
" collidables[collidableIndexB].m_shapeType == SHAPE_PLANE)\n"
" {\n"
"\n"
"\n"
" computeContactPlaneConvex( pairIndex, bodyIndexB,bodyIndexA, collidableIndexB,collidableIndexA, \n"
" rigidBodies,collidables,faces, globalContactsOut, nGlobalContactsOut,numPairs);\n"
" return;\n"
" }\n"
"\n"
" if (collidables[collidableIndexA].m_shapeType == SHAPE_PLANE &&\n"
" collidables[collidableIndexB].m_shapeType == SHAPE_SPHERE)\n"
" {\n"
"\n"
"\n"
" computeContactPlaneConvex(pairIndex, bodyIndexA, bodyIndexB, collidableIndexA, collidableIndexB, \n"
" rigidBodies,collidables,faces, globalContactsOut, nGlobalContactsOut,numPairs);\n"
" return;\n"
" \n"
" }\n"
" \n"
" if (collidables[collidableIndexA].m_shapeType == SHAPE_SPHERE &&\n"
" collidables[collidableIndexB].m_shapeType == SHAPE_SPHERE)\n"
" {\n"
" //sphere-sphere\n"
" float radiusA = collidables[collidableIndexA].m_radius;\n"
" float radiusB = collidables[collidableIndexB].m_radius;\n"
" float4 posA = rigidBodies[bodyIndexA].m_pos;\n"
" float4 posB = rigidBodies[bodyIndexB].m_pos;\n"
"\n"
" float4 diff = posA-posB;\n"
" float len = length(diff);\n"
" \n"
" ///iff distance positive, don't generate a new contact\n"
" if ( len <= (radiusA+radiusB))\n"
" {\n"
" ///distance (negative means penetration)\n"
" float dist = len - (radiusA+radiusB);\n"
" float4 normalOnSurfaceB = make_float4(1.f,0.f,0.f,0.f);\n"
" if (len > 0.00001)\n"
" {\n"
" normalOnSurfaceB = diff / len;\n"
" }\n"
" float4 contactPosB = posB + normalOnSurfaceB*radiusB;\n"
" contactPosB.w = dist;\n"
" \n"
" int dstIdx;\n"
" AppendInc( nGlobalContactsOut, dstIdx );\n"
" \n"
" if (dstIdx < numPairs)\n"
" {\n"
" __global Contact4* c = &globalContactsOut[dstIdx];\n"
" c->m_worldNormal = -normalOnSurfaceB;\n"
" c->m_coeffs = (u32)(0.f*0xffff) | ((u32)(0.7f*0xffff)<<16);\n"
" c->m_batchIdx = pairIndex;\n"
" int bodyA = pairs[pairIndex].x;\n"
" int bodyB = pairs[pairIndex].y;\n"
" c->m_bodyAPtrAndSignBit = rigidBodies[bodyA].m_invMass==0?-bodyA:bodyA;\n"
" c->m_bodyBPtrAndSignBit = rigidBodies[bodyB].m_invMass==0?-bodyB:bodyB;\n"
" c->m_worldPos[0] = contactPosB;\n"
" GET_NPOINTS(*c) = 1;\n"
" }//if (dstIdx < numPairs)\n"
" }//if ( len <= (radiusA+radiusB))\n"
"\n"
" return;\n"
" }//SHAPE_SPHERE SHAPE_SPHERE\n"
"\n"
" if (hasSeparatingAxis[i])\n"
" {\n"
@@ -1263,10 +1131,6 @@ static const char* satClipKernelsCL= \
" \n"
" int shapeIndexA = collidables[collidableIndexA].m_shapeIndex;\n"
" int shapeIndexB = collidables[collidableIndexB].m_shapeIndex;\n"
"\n"
" \n"
"\n"
"\n"
" \n"
" int numLocalContactsOut = clipHullAgainstHull(gpuCompoundSepNormalsOut[i],\n"
" &convexShapes[shapeIndexA], &convexShapes[shapeIndexB],\n"