void btConvexPolyhedron::project(const btTransform& trans, const btVector3& dir, btScalar& minProj, btScalar& maxProj, btVector3& witnesPtMin,btVector3& witnesPtMax) const
{
minProj = FLT_MAX;
maxProj = -FLT_MAX;
int numVerts = m_vertices.size();
for(int i=0;i<numVerts;i++)
{
btVector3 pt = trans * m_vertices[i];
btScalar dp = pt.dot(dir);
if(dp < minProj)
{
minProj = dp;
witnesPtMin = pt;
}
if(dp > maxProj)
{
maxProj = dp;
witnesPtMax = pt;
}
}
if(minProj>maxProj)
{
btSwap(minProj,maxProj);
btSwap(witnesPtMin,witnesPtMax);
}
}
btInternalVertexPair(short int v0,short int v1)
:m_v0(v0),
m_v1(v1)
{
if (m_v1>m_v0)
btSwap(m_v0,m_v1);
}
btBroadphasePair* btHashedOverlappingPairCache::findPair(btBroadphaseProxy* proxy0, btBroadphaseProxy* proxy1)
{
gFindPairs++;
if(proxy0->m_uniqueId>proxy1->m_uniqueId)
btSwap(proxy0,proxy1);
int proxyId1 = proxy0->getUid();
int proxyId2 = proxy1->getUid();
/*if (proxyId1 > proxyId2)
btSwap(proxyId1, proxyId2);*/
int hash = static_cast<int>(getHash(static_cast<unsigned int>(proxyId1), static_cast<unsigned int>(proxyId2)) & (m_overlappingPairArray.capacity()-1));
if (hash >= m_hashTable.size())
{
return NULL;
}
int index = m_hashTable[hash];
while (index != BT_NULL_PAIR && equalsPair(m_overlappingPairArray[index], proxyId1, proxyId2) == false)
{
index = m_next[index];
}
if (index == BT_NULL_PAIR)
{
return NULL;
}
btAssert(index < m_overlappingPairArray.size());
return &m_overlappingPairArray[index];
}
static DBVT_INLINE btDbvtNode* sort(btDbvtNode* n,btDbvtNode*& r)
{
btDbvtNode* p=n->parent;
btAssert(n->isinternal());
if(p>n)
{
const int i=indexof(n);
const int j=1-i;
btDbvtNode* s=p->childs[j];
btDbvtNode* q=p->parent;
btAssert(n==p->childs[i]);
if(q) q->childs[indexof(p)]=n; else r=n;
s->parent=n;
p->parent=n;
n->parent=q;
p->childs[0]=n->childs[0];
p->childs[1]=n->childs[1];
n->childs[0]->parent=p;
n->childs[1]->parent=p;
n->childs[i]=p;
n->childs[j]=s;
btSwap(p->volume,n->volume);
return(p);
}
return(n);
}
int4 HullLibrary::FindSimplex(btVector3 *verts,int verts_count,btAlignedObjectArray<int> &allow)
{
btVector3 basis[3];
basis[0] = btVector3( btScalar(0.01), btScalar(0.02), btScalar(1.0) );
int p0 = maxdirsterid(verts,verts_count, basis[0],allow);
int p1 = maxdirsterid(verts,verts_count,-basis[0],allow);
basis[0] = verts[p0]-verts[p1];
if(p0==p1 || basis[0]==btVector3(0,0,0))
return int4(-1,-1,-1,-1);
basis[1] = btCross(btVector3( btScalar(1),btScalar(0.02), btScalar(0)),basis[0]);
basis[2] = btCross(btVector3(btScalar(-0.02), btScalar(1), btScalar(0)),basis[0]);
if (basis[1].length() > basis[2].length())
{
basis[1].normalize();
} else {
basis[1] = basis[2];
basis[1].normalize ();
}
int p2 = maxdirsterid(verts,verts_count,basis[1],allow);
if(p2 == p0 || p2 == p1)
{
p2 = maxdirsterid(verts,verts_count,-basis[1],allow);
}
if(p2 == p0 || p2 == p1)
return int4(-1,-1,-1,-1);
basis[1] = verts[p2] - verts[p0];
basis[2] = btCross(basis[1],basis[0]).normalized();
int p3 = maxdirsterid(verts,verts_count,basis[2],allow);
if(p3==p0||p3==p1||p3==p2) p3 = maxdirsterid(verts,verts_count,-basis[2],allow);
if(p3==p0||p3==p1||p3==p2)
return int4(-1,-1,-1,-1);
btAssert(!(p0==p1||p0==p2||p0==p3||p1==p2||p1==p3||p2==p3));
if(btDot(verts[p3]-verts[p0],btCross(verts[p1]-verts[p0],verts[p2]-verts[p0])) <0) {btSwap(p2,p3);}
return int4(p0,p1,p2,p3);
}
btBroadphasePair* btHashedOverlappingPairCache::internalAddPair(btBroadphaseProxy* proxy0, btBroadphaseProxy* proxy1)
{
if(proxy0->m_uniqueId>proxy1->m_uniqueId)
btSwap(proxy0,proxy1);
int proxyId1 = proxy0->getUid();
int proxyId2 = proxy1->getUid();
/*if (proxyId1 > proxyId2)
btSwap(proxyId1, proxyId2);*/
int hash = static_cast<int>(getHash(static_cast<unsigned int>(proxyId1),static_cast<unsigned int>(proxyId2)) & (m_overlappingPairArray.capacity()-1)); // New hash value with new mask
btBroadphasePair* pair = internalFindPair(proxy0, proxy1, hash);
if (pair != NULL)
{
return pair;
}
/*for(int i=0;i<m_overlappingPairArray.size();++i)
{
if( (m_overlappingPairArray[i].m_pProxy0==proxy0)&&
(m_overlappingPairArray[i].m_pProxy1==proxy1))
{
printf("Adding duplicated %u<>%u\r\n",proxyId1,proxyId2);
internalFindPair(proxy0, proxy1, hash);
}
}*/
int count = m_overlappingPairArray.size();
int oldCapacity = m_overlappingPairArray.capacity();
void* mem = &m_overlappingPairArray.expandNonInitializing();
//this is where we add an actual pair, so also call the 'ghost'
if (m_ghostPairCallback)
m_ghostPairCallback->addOverlappingPair(proxy0,proxy1);
int newCapacity = m_overlappingPairArray.capacity();
if (oldCapacity < newCapacity)
{
growTables();
//hash with new capacity
hash = static_cast<int>(getHash(static_cast<unsigned int>(proxyId1),static_cast<unsigned int>(proxyId2)) & (m_overlappingPairArray.capacity()-1));
}
pair = new (mem) btBroadphasePair(*proxy0,*proxy1);
// pair->m_pProxy0 = proxy0;
// pair->m_pProxy1 = proxy1;
pair->m_algorithm = 0;
pair->m_internalTmpValue = 0;
m_next[count] = m_hashTable[hash];
m_hashTable[hash] = count;
return pair;
}
void btConvexHullShape::project(const btTransform& trans, const btVector3& dir, btScalar& minProj, btScalar& maxProj, btVector3& witnesPtMin,btVector3& witnesPtMax) const
{
#if 1
minProj = FLT_MAX;
maxProj = -FLT_MAX;
int numVerts = m_unscaledPoints.size();
for(int i=0;i<numVerts;i++)
{
btVector3 vtx = m_unscaledPoints[i] * m_localScaling;
btVector3 pt = trans * vtx;
btScalar dp = pt.dot(dir);
if(dp < minProj)
{
minProj = dp;
witnesPtMin = pt;
}
if(dp > maxProj)
{
maxProj = dp;
witnesPtMax=pt;
}
}
#else
btVector3 localAxis = dir*trans.getBasis();
witnesPtMin = trans(localGetSupportingVertex(localAxis));
witnesPtMax = trans(localGetSupportingVertex(-localAxis));
minProj = witnesPtMin.dot(dir);
maxProj = witnesPtMax.dot(dir);
#endif
if(minProj>maxProj)
{
btSwap(minProj,maxProj);
btSwap(witnesPtMin,witnesPtMax);
}
}
void Process(const btDbvtNode* na, const btDbvtNode* nb)
{
if (na != nb)
{
btDbvtProxy* pa = (btDbvtProxy*)na->data;
btDbvtProxy* pb = (btDbvtProxy*)nb->data;
#if DBVT_BP_SORTPAIRS
if (pa->m_uniqueId > pb->m_uniqueId)
btSwap(pa, pb);
#endif
pbp->m_paircache->addOverlappingPair(pa, pb);
++pbp->m_newpairs;
}
}
void btPolyhedralContactClipping::clipFaceAgainstHull(const btVector3& separatingNormal, const btConvexPolyhedron& hullA, const btTransform& transA, btVertexArray& worldVertsB1, const btScalar minDist, btScalar maxDist,btDiscreteCollisionDetectorInterface::Result& resultOut)
{
btVertexArray worldVertsB2;
btVertexArray* pVtxIn = &worldVertsB1;
btVertexArray* pVtxOut = &worldVertsB2;
pVtxOut->reserve(pVtxIn->size());
int closestFaceA=-1;
{
btScalar dmin = FLT_MAX;
for(int face=0;face<hullA.m_faces.size();face++)
{
const btVector3 Normal(hullA.m_faces[face].m_plane[0], hullA.m_faces[face].m_plane[1], hullA.m_faces[face].m_plane[2]);
const btVector3 faceANormalWS = transA.getBasis() * Normal;
btScalar d = faceANormalWS.dot(separatingNormal);
if (d < dmin)
{
dmin = d;
closestFaceA = face;
}
}
}
if (closestFaceA<0)
return;
const btFace& polyA = hullA.m_faces[closestFaceA];
// clip polygon to back of planes of all faces of hull A that are adjacent to witness face
int numContacts = pVtxIn->size();
int numVerticesA = polyA.m_indices.size();
for(int e0=0;e0<numVerticesA;e0++)
{
const btVector3& a = hullA.m_vertices[polyA.m_indices[e0]];
const btVector3& b = hullA.m_vertices[polyA.m_indices[(e0+1)%numVerticesA]];
const btVector3 edge0 = a - b;
const btVector3 WorldEdge0 = transA.getBasis() * edge0;
btVector3 worldPlaneAnormal1 = transA.getBasis()* btVector3(polyA.m_plane[0],polyA.m_plane[1],polyA.m_plane[2]);
btVector3 planeNormalWS1 = -WorldEdge0.cross(worldPlaneAnormal1);//.cross(WorldEdge0);
btVector3 worldA1 = transA*a;
btScalar planeEqWS1 = -worldA1.dot(planeNormalWS1);
//int otherFace=0;
#ifdef BLA1
int otherFace = polyA.m_connectedFaces[e0];
btVector3 localPlaneNormal (hullA.m_faces[otherFace].m_plane[0],hullA.m_faces[otherFace].m_plane[1],hullA.m_faces[otherFace].m_plane[2]);
btScalar localPlaneEq = hullA.m_faces[otherFace].m_plane[3];
btVector3 planeNormalWS = transA.getBasis()*localPlaneNormal;
btScalar planeEqWS=localPlaneEq-planeNormalWS.dot(transA.getOrigin());
#else
btVector3 planeNormalWS = planeNormalWS1;
btScalar planeEqWS=planeEqWS1;
#endif
//clip face
clipFace(*pVtxIn, *pVtxOut,planeNormalWS,planeEqWS);
btSwap(pVtxIn,pVtxOut);
pVtxOut->resize(0);
}
//#define ONLY_REPORT_DEEPEST_POINT
btVector3 point;
// only keep points that are behind the witness face
{
btVector3 localPlaneNormal (polyA.m_plane[0],polyA.m_plane[1],polyA.m_plane[2]);
btScalar localPlaneEq = polyA.m_plane[3];
btVector3 planeNormalWS = transA.getBasis()*localPlaneNormal;
btScalar planeEqWS=localPlaneEq-planeNormalWS.dot(transA.getOrigin());
for (int i=0;i<pVtxIn->size();i++)
{
btScalar depth = planeNormalWS.dot(pVtxIn->at(i))+planeEqWS;
if (depth <=minDist)
{
// printf("clamped: depth=%f to minDist=%f\n",depth,minDist);
depth = minDist;
}
if (depth <=maxDist)
{
btVector3 point = pVtxIn->at(i);
#ifdef ONLY_REPORT_DEEPEST_POINT
curMaxDist = depth;
#else
#if 0
if (depth<-3)
{
printf("error in btPolyhedralContactClipping depth = %f\n", depth);
printf("likely wrong separatingNormal passed in\n");
}
#endif
resultOut.addContactPoint(separatingNormal,point,depth);
#endif
}
}
}
#ifdef ONLY_REPORT_DEEPEST_POINT
if (curMaxDist<maxDist)
{
resultOut.addContactPoint(separatingNormal,point,curMaxDist);
}
#endif //ONLY_REPORT_DEEPEST_POINT
}
void btDbvtBroadphase::collide(btDispatcher* dispatcher)
{
/*printf("---------------------------------------------------------\n");
printf("m_sets[0].m_leaves=%d\n",m_sets[0].m_leaves);
printf("m_sets[1].m_leaves=%d\n",m_sets[1].m_leaves);
printf("numPairs = %d\n",getOverlappingPairCache()->getNumOverlappingPairs());
{
int i;
for (i=0;i<getOverlappingPairCache()->getNumOverlappingPairs();i++)
{
printf("pair[%d]=(%d,%d),",i,getOverlappingPairCache()->getOverlappingPairArray()[i].m_pProxy0->getUid(),
getOverlappingPairCache()->getOverlappingPairArray()[i].m_pProxy1->getUid());
}
printf("\n");
}
*/
SPC(m_profiling.m_total);
/* optimize */
m_sets[0].optimizeIncremental(1 + (m_sets[0].m_leaves * m_dupdates) / 100);
if (m_fixedleft)
{
const int count = 1 + (m_sets[1].m_leaves * m_fupdates) / 100;
m_sets[1].optimizeIncremental(1 + (m_sets[1].m_leaves * m_fupdates) / 100);
m_fixedleft = btMax<int>(0, m_fixedleft - count);
}
/* dynamic -> fixed set */
m_stageCurrent = (m_stageCurrent + 1) % STAGECOUNT;
btDbvtProxy* current = m_stageRoots[m_stageCurrent];
if (current)
{
#if DBVT_BP_ACCURATESLEEPING
btDbvtTreeCollider collider(this);
#endif
do
{
btDbvtProxy* next = current->links[1];
listremove(current, m_stageRoots[current->stage]);
listappend(current, m_stageRoots[STAGECOUNT]);
#if DBVT_BP_ACCURATESLEEPING
m_paircache->removeOverlappingPairsContainingProxy(current, dispatcher);
collider.proxy = current;
btDbvt::collideTV(m_sets[0].m_root, current->aabb, collider);
btDbvt::collideTV(m_sets[1].m_root, current->aabb, collider);
#endif
m_sets[0].remove(current->leaf);
ATTRIBUTE_ALIGNED16(btDbvtVolume)
curAabb = btDbvtVolume::FromMM(current->m_aabbMin, current->m_aabbMax);
current->leaf = m_sets[1].insert(curAabb, current);
current->stage = STAGECOUNT;
current = next;
} while (current);
m_fixedleft = m_sets[1].m_leaves;
m_needcleanup = true;
}
/* collide dynamics */
{
btDbvtTreeCollider collider(this);
if (m_deferedcollide)
{
SPC(m_profiling.m_fdcollide);
m_sets[0].collideTTpersistentStack(m_sets[0].m_root, m_sets[1].m_root, collider);
}
if (m_deferedcollide)
{
SPC(m_profiling.m_ddcollide);
m_sets[0].collideTTpersistentStack(m_sets[0].m_root, m_sets[0].m_root, collider);
}
}
/* clean up */
if (m_needcleanup)
{
SPC(m_profiling.m_cleanup);
btBroadphasePairArray& pairs = m_paircache->getOverlappingPairArray();
if (pairs.size() > 0)
{
int ni = btMin(pairs.size(), btMax<int>(m_newpairs, (pairs.size() * m_cupdates) / 100));
for (int i = 0; i < ni; ++i)
{
btBroadphasePair& p = pairs[(m_cid + i) % pairs.size()];
btDbvtProxy* pa = (btDbvtProxy*)p.m_pProxy0;
btDbvtProxy* pb = (btDbvtProxy*)p.m_pProxy1;
if (!Intersect(pa->leaf->volume, pb->leaf->volume))
{
#if DBVT_BP_SORTPAIRS
if (pa->m_uniqueId > pb->m_uniqueId)
btSwap(pa, pb);
#endif
m_paircache->removeOverlappingPair(pa, pb, dispatcher);
--ni;
--i;
}
}
if (pairs.size() > 0)
m_cid = (m_cid + ni) % pairs.size();
else
m_cid = 0;
}
}
++m_pid;
m_newpairs = 1;
m_needcleanup = false;
if (m_updates_call > 0)
{
m_updates_ratio = m_updates_done / (btScalar)m_updates_call;
}
else
{
m_updates_ratio = 0;
}
m_updates_done /= 2;
m_updates_call /= 2;
}
void TinyRendererVisualShapeConverter::render(const float viewMat[16], const float projMat[16])
{
//clear the color buffer
TGAColor clearColor;
clearColor.bgra[0] = 255;
clearColor.bgra[1] = 255;
clearColor.bgra[2] = 255;
clearColor.bgra[3] = 255;
clearBuffers(clearColor);
ATTRIBUTE_ALIGNED16(btScalar modelMat[16]);
btVector3 lightDirWorld(-5,200,-40);
switch (m_data->m_upAxis)
{
case 1:
lightDirWorld = btVector3(-50.f,100,30);
break;
case 2:
lightDirWorld = btVector3(-50.f,30,100);
break;
default:{}
};
lightDirWorld.normalize();
// printf("num m_swRenderInstances = %d\n", m_data->m_swRenderInstances.size());
for (int i=0;i<m_data->m_swRenderInstances.size();i++)
{
TinyRendererObjectArray** visualArrayPtr = m_data->m_swRenderInstances.getAtIndex(i);
if (0==visualArrayPtr)
continue;//can this ever happen?
TinyRendererObjectArray* visualArray = *visualArrayPtr;
btHashPtr colObjHash = m_data->m_swRenderInstances.getKeyAtIndex(i);
const btCollisionObject* colObj = (btCollisionObject*) colObjHash.getPointer();
for (int v=0;v<visualArray->m_renderObjects.size();v++)
{
TinyRenderObjectData* renderObj = visualArray->m_renderObjects[v];
//sync the object transform
const btTransform& tr = colObj->getWorldTransform();
tr.getOpenGLMatrix(modelMat);
for (int i=0;i<4;i++)
{
for (int j=0;j<4;j++)
{
renderObj->m_projectionMatrix[i][j] = projMat[i+4*j];
renderObj->m_modelMatrix[i][j] = modelMat[i+4*j];
renderObj->m_viewMatrix[i][j] = viewMat[i+4*j];
renderObj->m_localScaling = colObj->getCollisionShape()->getLocalScaling();
renderObj->m_lightDirWorld = lightDirWorld;
}
}
TinyRenderer::renderObject(*renderObj);
}
}
//printf("write tga \n");
//m_data->m_rgbColorBuffer.write_tga_file("camera.tga");
// printf("flipped!\n");
m_data->m_rgbColorBuffer.flip_vertically();
//flip z-buffer
{
int half = m_data->m_swHeight>>1;
for (int j=0; j<half; j++)
{
unsigned long l1 = j*m_data->m_swWidth;
unsigned long l2 = (m_data->m_swHeight-1-j)*m_data->m_swWidth;
for (int i=0;i<m_data->m_swWidth;i++)
{
btSwap(m_data->m_depthBuffer[l1+i],m_data->m_depthBuffer[l2+i]);
}
}
}
}
void* btHashedOverlappingPairCache::removeOverlappingPair(btBroadphaseProxy* proxy0, btBroadphaseProxy* proxy1,btDispatcher* dispatcher)
{
gRemovePairs++;
if(proxy0->m_uniqueId>proxy1->m_uniqueId)
btSwap(proxy0,proxy1);
int proxyId1 = proxy0->getUid();
int proxyId2 = proxy1->getUid();
/*if (proxyId1 > proxyId2)
btSwap(proxyId1, proxyId2);*/
int hash = static_cast<int>(getHash(static_cast<unsigned int>(proxyId1),static_cast<unsigned int>(proxyId2)) & (m_overlappingPairArray.capacity()-1));
btBroadphasePair* pair = internalFindPair(proxy0, proxy1, hash);
if (pair == NULL)
{
return 0;
}
cleanOverlappingPair(*pair,dispatcher);
void* userData = pair->m_internalInfo1;
btAssert(pair->m_pProxy0->getUid() == proxyId1);
btAssert(pair->m_pProxy1->getUid() == proxyId2);
int pairIndex = int(pair - &m_overlappingPairArray[0]);
btAssert(pairIndex < m_overlappingPairArray.size());
// Remove the pair from the hash table.
int index = m_hashTable[hash];
btAssert(index != BT_NULL_PAIR);
int previous = BT_NULL_PAIR;
while (index != pairIndex)
{
previous = index;
index = m_next[index];
}
if (previous != BT_NULL_PAIR)
{
btAssert(m_next[previous] == pairIndex);
m_next[previous] = m_next[pairIndex];
}
else
{
m_hashTable[hash] = m_next[pairIndex];
}
// We now move the last pair into spot of the
// pair being removed. We need to fix the hash
// table indices to support the move.
int lastPairIndex = m_overlappingPairArray.size() - 1;
if (m_ghostPairCallback)
m_ghostPairCallback->removeOverlappingPair(proxy0, proxy1,dispatcher);
// If the removed pair is the last pair, we are done.
if (lastPairIndex == pairIndex)
{
m_overlappingPairArray.pop_back();
return userData;
}
// Remove the last pair from the hash table.
const btBroadphasePair* last = &m_overlappingPairArray[lastPairIndex];
/* missing swap here too, Nat. */
int lastHash = static_cast<int>(getHash(static_cast<unsigned int>(last->m_pProxy0->getUid()), static_cast<unsigned int>(last->m_pProxy1->getUid())) & (m_overlappingPairArray.capacity()-1));
index = m_hashTable[lastHash];
btAssert(index != BT_NULL_PAIR);
previous = BT_NULL_PAIR;
while (index != lastPairIndex)
{
previous = index;
index = m_next[index];
}
if (previous != BT_NULL_PAIR)
{
btAssert(m_next[previous] == lastPairIndex);
m_next[previous] = m_next[lastPairIndex];
}
else
{
m_hashTable[lastHash] = m_next[lastPairIndex];
}
// Copy the last pair into the remove pair's spot.
m_overlappingPairArray[pairIndex] = m_overlappingPairArray[lastPairIndex];
// Insert the last pair into the hash table
m_next[pairIndex] = m_hashTable[lastHash];
m_hashTable[lastHash] = pairIndex;
m_overlappingPairArray.pop_back();
return userData;
}
void btDbvt::benchmark()
{
static const btScalar cfgVolumeCenterScale = 100;
static const btScalar cfgVolumeExentsBase = 1;
static const btScalar cfgVolumeExentsScale = 4;
static const int cfgLeaves = 8192;
static const bool cfgEnable = true;
//[1] btDbvtVolume intersections
bool cfgBenchmark1_Enable = cfgEnable;
static const int cfgBenchmark1_Iterations = 8;
static const int cfgBenchmark1_Reference = 3499;
//[2] btDbvtVolume merges
bool cfgBenchmark2_Enable = cfgEnable;
static const int cfgBenchmark2_Iterations = 4;
static const int cfgBenchmark2_Reference = 1945;
//[3] btDbvt::collideTT
bool cfgBenchmark3_Enable = cfgEnable;
static const int cfgBenchmark3_Iterations = 512;
static const int cfgBenchmark3_Reference = 5485;
//[4] btDbvt::collideTT self
bool cfgBenchmark4_Enable = cfgEnable;
static const int cfgBenchmark4_Iterations = 512;
static const int cfgBenchmark4_Reference = 2814;
//[5] btDbvt::collideTT xform
bool cfgBenchmark5_Enable = cfgEnable;
static const int cfgBenchmark5_Iterations = 512;
static const btScalar cfgBenchmark5_OffsetScale = 2;
static const int cfgBenchmark5_Reference = 7379;
//[6] btDbvt::collideTT xform,self
bool cfgBenchmark6_Enable = cfgEnable;
static const int cfgBenchmark6_Iterations = 512;
static const btScalar cfgBenchmark6_OffsetScale = 2;
static const int cfgBenchmark6_Reference = 7270;
//[7] btDbvt::rayTest
bool cfgBenchmark7_Enable = cfgEnable;
static const int cfgBenchmark7_Passes = 32;
static const int cfgBenchmark7_Iterations = 65536;
static const int cfgBenchmark7_Reference = 6307;
//[8] insert/remove
bool cfgBenchmark8_Enable = cfgEnable;
static const int cfgBenchmark8_Passes = 32;
static const int cfgBenchmark8_Iterations = 65536;
static const int cfgBenchmark8_Reference = 2105;
//[9] updates (teleport)
bool cfgBenchmark9_Enable = cfgEnable;
static const int cfgBenchmark9_Passes = 32;
static const int cfgBenchmark9_Iterations = 65536;
static const int cfgBenchmark9_Reference = 1879;
//[10] updates (jitter)
bool cfgBenchmark10_Enable = cfgEnable;
static const btScalar cfgBenchmark10_Scale = cfgVolumeCenterScale/10000;
static const int cfgBenchmark10_Passes = 32;
static const int cfgBenchmark10_Iterations = 65536;
static const int cfgBenchmark10_Reference = 1244;
//[11] optimize (incremental)
bool cfgBenchmark11_Enable = cfgEnable;
static const int cfgBenchmark11_Passes = 64;
static const int cfgBenchmark11_Iterations = 65536;
static const int cfgBenchmark11_Reference = 2510;
//[12] btDbvtVolume notequal
bool cfgBenchmark12_Enable = cfgEnable;
static const int cfgBenchmark12_Iterations = 32;
static const int cfgBenchmark12_Reference = 3677;
//[13] culling(OCL+fullsort)
bool cfgBenchmark13_Enable = cfgEnable;
static const int cfgBenchmark13_Iterations = 1024;
static const int cfgBenchmark13_Reference = 2231;
//[14] culling(OCL+qsort)
bool cfgBenchmark14_Enable = cfgEnable;
static const int cfgBenchmark14_Iterations = 8192;
static const int cfgBenchmark14_Reference = 3500;
//[15] culling(KDOP+qsort)
bool cfgBenchmark15_Enable = cfgEnable;
static const int cfgBenchmark15_Iterations = 8192;
static const int cfgBenchmark15_Reference = 1151;
//[16] insert/remove batch
bool cfgBenchmark16_Enable = cfgEnable;
static const int cfgBenchmark16_BatchCount = 256;
static const int cfgBenchmark16_Passes = 16384;
static const int cfgBenchmark16_Reference = 5138;
//[17] select
bool cfgBenchmark17_Enable = cfgEnable;
static const int cfgBenchmark17_Iterations = 4;
static const int cfgBenchmark17_Reference = 3390;
btClock wallclock;
printf("Benchmarking dbvt...\r\n");
printf("\tWorld scale: %f\r\n",cfgVolumeCenterScale);
printf("\tExtents base: %f\r\n",cfgVolumeExentsBase);
printf("\tExtents range: %f\r\n",cfgVolumeExentsScale);
printf("\tLeaves: %u\r\n",cfgLeaves);
printf("\tsizeof(btDbvtVolume): %u bytes\r\n",sizeof(btDbvtVolume));
printf("\tsizeof(btDbvtNode): %u bytes\r\n",sizeof(btDbvtNode));
if(cfgBenchmark1_Enable)
{// Benchmark 1
srand(380843);
btAlignedObjectArray<btDbvtVolume> volumes;
btAlignedObjectArray<bool> results;
volumes.resize(cfgLeaves);
results.resize(cfgLeaves);
for(int i=0;i<cfgLeaves;++i)
{
volumes[i]=btDbvtBenchmark::RandVolume(cfgVolumeCenterScale,cfgVolumeExentsBase,cfgVolumeExentsScale);
}
printf("[1] btDbvtVolume intersections: ");
wallclock.reset();
for(int i=0;i<cfgBenchmark1_Iterations;++i)
{
for(int j=0;j<cfgLeaves;++j)
{
for(int k=0;k<cfgLeaves;++k)
{
results[k]=Intersect(volumes[j],volumes[k]);
}
}
}
const int time=(int)wallclock.getTimeMilliseconds();
printf("%u ms (%i%%)\r\n",time,(time-cfgBenchmark1_Reference)*100/time);
}
if(cfgBenchmark2_Enable)
{// Benchmark 2
srand(380843);
btAlignedObjectArray<btDbvtVolume> volumes;
btAlignedObjectArray<btDbvtVolume> results;
volumes.resize(cfgLeaves);
results.resize(cfgLeaves);
for(int i=0;i<cfgLeaves;++i)
{
volumes[i]=btDbvtBenchmark::RandVolume(cfgVolumeCenterScale,cfgVolumeExentsBase,cfgVolumeExentsScale);
}
printf("[2] btDbvtVolume merges: ");
wallclock.reset();
for(int i=0;i<cfgBenchmark2_Iterations;++i)
{
for(int j=0;j<cfgLeaves;++j)
{
for(int k=0;k<cfgLeaves;++k)
{
Merge(volumes[j],volumes[k],results[k]);
}
}
}
const int time=(int)wallclock.getTimeMilliseconds();
printf("%u ms (%i%%)\r\n",time,(time-cfgBenchmark2_Reference)*100/time);
}
if(cfgBenchmark3_Enable)
{// Benchmark 3
srand(380843);
btDbvt dbvt[2];
btDbvtBenchmark::NilPolicy policy;
btDbvtBenchmark::RandTree(cfgVolumeCenterScale,cfgVolumeExentsBase,cfgVolumeExentsScale,cfgLeaves,dbvt[0]);
btDbvtBenchmark::RandTree(cfgVolumeCenterScale,cfgVolumeExentsBase,cfgVolumeExentsScale,cfgLeaves,dbvt[1]);
dbvt[0].optimizeTopDown();
dbvt[1].optimizeTopDown();
printf("[3] btDbvt::collideTT: ");
wallclock.reset();
for(int i=0;i<cfgBenchmark3_Iterations;++i)
{
btDbvt::collideTT(dbvt[0].m_root,dbvt[1].m_root,policy);
}
const int time=(int)wallclock.getTimeMilliseconds();
printf("%u ms (%i%%)\r\n",time,(time-cfgBenchmark3_Reference)*100/time);
}
if(cfgBenchmark4_Enable)
{// Benchmark 4
srand(380843);
btDbvt dbvt;
btDbvtBenchmark::NilPolicy policy;
btDbvtBenchmark::RandTree(cfgVolumeCenterScale,cfgVolumeExentsBase,cfgVolumeExentsScale,cfgLeaves,dbvt);
dbvt.optimizeTopDown();
printf("[4] btDbvt::collideTT self: ");
wallclock.reset();
for(int i=0;i<cfgBenchmark4_Iterations;++i)
{
btDbvt::collideTT(dbvt.m_root,dbvt.m_root,policy);
}
const int time=(int)wallclock.getTimeMilliseconds();
printf("%u ms (%i%%)\r\n",time,(time-cfgBenchmark4_Reference)*100/time);
}
if(cfgBenchmark5_Enable)
{// Benchmark 5
srand(380843);
btDbvt dbvt[2];
btAlignedObjectArray<btTransform> transforms;
btDbvtBenchmark::NilPolicy policy;
transforms.resize(cfgBenchmark5_Iterations);
for(int i=0;i<transforms.size();++i)
{
transforms[i]=btDbvtBenchmark::RandTransform(cfgVolumeCenterScale*cfgBenchmark5_OffsetScale);
}
btDbvtBenchmark::RandTree(cfgVolumeCenterScale,cfgVolumeExentsBase,cfgVolumeExentsScale,cfgLeaves,dbvt[0]);
btDbvtBenchmark::RandTree(cfgVolumeCenterScale,cfgVolumeExentsBase,cfgVolumeExentsScale,cfgLeaves,dbvt[1]);
dbvt[0].optimizeTopDown();
dbvt[1].optimizeTopDown();
printf("[5] btDbvt::collideTT xform: ");
wallclock.reset();
for(int i=0;i<cfgBenchmark5_Iterations;++i)
{
btDbvt::collideTT(dbvt[0].m_root,dbvt[1].m_root,transforms[i],policy);
}
const int time=(int)wallclock.getTimeMilliseconds();
printf("%u ms (%i%%)\r\n",time,(time-cfgBenchmark5_Reference)*100/time);
}
if(cfgBenchmark6_Enable)
{// Benchmark 6
srand(380843);
btDbvt dbvt;
btAlignedObjectArray<btTransform> transforms;
btDbvtBenchmark::NilPolicy policy;
transforms.resize(cfgBenchmark6_Iterations);
for(int i=0;i<transforms.size();++i)
{
transforms[i]=btDbvtBenchmark::RandTransform(cfgVolumeCenterScale*cfgBenchmark6_OffsetScale);
}
btDbvtBenchmark::RandTree(cfgVolumeCenterScale,cfgVolumeExentsBase,cfgVolumeExentsScale,cfgLeaves,dbvt);
dbvt.optimizeTopDown();
printf("[6] btDbvt::collideTT xform,self: ");
wallclock.reset();
for(int i=0;i<cfgBenchmark6_Iterations;++i)
{
btDbvt::collideTT(dbvt.m_root,dbvt.m_root,transforms[i],policy);
}
const int time=(int)wallclock.getTimeMilliseconds();
printf("%u ms (%i%%)\r\n",time,(time-cfgBenchmark6_Reference)*100/time);
}
if(cfgBenchmark7_Enable)
{// Benchmark 7
srand(380843);
btDbvt dbvt;
btAlignedObjectArray<btVector3> rayorg;
btAlignedObjectArray<btVector3> raydir;
btDbvtBenchmark::NilPolicy policy;
rayorg.resize(cfgBenchmark7_Iterations);
raydir.resize(cfgBenchmark7_Iterations);
for(int i=0;i<rayorg.size();++i)
{
rayorg[i]=btDbvtBenchmark::RandVector3(cfgVolumeCenterScale*2);
raydir[i]=btDbvtBenchmark::RandVector3(cfgVolumeCenterScale*2);
}
btDbvtBenchmark::RandTree(cfgVolumeCenterScale,cfgVolumeExentsBase,cfgVolumeExentsScale,cfgLeaves,dbvt);
dbvt.optimizeTopDown();
printf("[7] btDbvt::rayTest: ");
wallclock.reset();
for(int i=0;i<cfgBenchmark7_Passes;++i)
{
for(int j=0;j<cfgBenchmark7_Iterations;++j)
{
btDbvt::rayTest(dbvt.m_root,rayorg[j],rayorg[j]+raydir[j],policy);
}
}
const int time=(int)wallclock.getTimeMilliseconds();
unsigned rays=cfgBenchmark7_Passes*cfgBenchmark7_Iterations;
printf("%u ms (%i%%),(%u r/s)\r\n",time,(time-cfgBenchmark7_Reference)*100/time,(rays*1000)/time);
}
if(cfgBenchmark8_Enable)
{// Benchmark 8
srand(380843);
btDbvt dbvt;
btDbvtBenchmark::RandTree(cfgVolumeCenterScale,cfgVolumeExentsBase,cfgVolumeExentsScale,cfgLeaves,dbvt);
dbvt.optimizeTopDown();
printf("[8] insert/remove: ");
wallclock.reset();
for(int i=0;i<cfgBenchmark8_Passes;++i)
{
for(int j=0;j<cfgBenchmark8_Iterations;++j)
{
dbvt.remove(dbvt.insert(btDbvtBenchmark::RandVolume(cfgVolumeCenterScale,cfgVolumeExentsBase,cfgVolumeExentsScale),0));
}
}
const int time=(int)wallclock.getTimeMilliseconds();
const int ir=cfgBenchmark8_Passes*cfgBenchmark8_Iterations;
printf("%u ms (%i%%),(%u ir/s)\r\n",time,(time-cfgBenchmark8_Reference)*100/time,ir*1000/time);
}
if(cfgBenchmark9_Enable)
{// Benchmark 9
srand(380843);
btDbvt dbvt;
btAlignedObjectArray<const btDbvtNode*> leaves;
btDbvtBenchmark::RandTree(cfgVolumeCenterScale,cfgVolumeExentsBase,cfgVolumeExentsScale,cfgLeaves,dbvt);
dbvt.optimizeTopDown();
dbvt.extractLeaves(dbvt.m_root,leaves);
printf("[9] updates (teleport): ");
wallclock.reset();
for(int i=0;i<cfgBenchmark9_Passes;++i)
{
for(int j=0;j<cfgBenchmark9_Iterations;++j)
{
dbvt.update(const_cast<btDbvtNode*>(leaves[rand()%cfgLeaves]),
btDbvtBenchmark::RandVolume(cfgVolumeCenterScale,cfgVolumeExentsBase,cfgVolumeExentsScale));
}
}
const int time=(int)wallclock.getTimeMilliseconds();
const int up=cfgBenchmark9_Passes*cfgBenchmark9_Iterations;
printf("%u ms (%i%%),(%u u/s)\r\n",time,(time-cfgBenchmark9_Reference)*100/time,up*1000/time);
}
if(cfgBenchmark10_Enable)
{// Benchmark 10
srand(380843);
btDbvt dbvt;
btAlignedObjectArray<const btDbvtNode*> leaves;
btAlignedObjectArray<btVector3> vectors;
vectors.resize(cfgBenchmark10_Iterations);
for(int i=0;i<vectors.size();++i)
{
vectors[i]=(btDbvtBenchmark::RandVector3()*2-btVector3(1,1,1))*cfgBenchmark10_Scale;
}
btDbvtBenchmark::RandTree(cfgVolumeCenterScale,cfgVolumeExentsBase,cfgVolumeExentsScale,cfgLeaves,dbvt);
dbvt.optimizeTopDown();
dbvt.extractLeaves(dbvt.m_root,leaves);
printf("[10] updates (jitter): ");
wallclock.reset();
for(int i=0;i<cfgBenchmark10_Passes;++i)
{
for(int j=0;j<cfgBenchmark10_Iterations;++j)
{
const btVector3& d=vectors[j];
btDbvtNode* l=const_cast<btDbvtNode*>(leaves[rand()%cfgLeaves]);
btDbvtVolume v=btDbvtVolume::FromMM(l->volume.Mins()+d,l->volume.Maxs()+d);
dbvt.update(l,v);
}
}
const int time=(int)wallclock.getTimeMilliseconds();
const int up=cfgBenchmark10_Passes*cfgBenchmark10_Iterations;
printf("%u ms (%i%%),(%u u/s)\r\n",time,(time-cfgBenchmark10_Reference)*100/time,up*1000/time);
}
if(cfgBenchmark11_Enable)
{// Benchmark 11
srand(380843);
btDbvt dbvt;
btDbvtBenchmark::RandTree(cfgVolumeCenterScale,cfgVolumeExentsBase,cfgVolumeExentsScale,cfgLeaves,dbvt);
dbvt.optimizeTopDown();
printf("[11] optimize (incremental): ");
wallclock.reset();
for(int i=0;i<cfgBenchmark11_Passes;++i)
{
dbvt.optimizeIncremental(cfgBenchmark11_Iterations);
}
const int time=(int)wallclock.getTimeMilliseconds();
const int op=cfgBenchmark11_Passes*cfgBenchmark11_Iterations;
printf("%u ms (%i%%),(%u o/s)\r\n",time,(time-cfgBenchmark11_Reference)*100/time,op/time*1000);
}
if(cfgBenchmark12_Enable)
{// Benchmark 12
srand(380843);
btAlignedObjectArray<btDbvtVolume> volumes;
btAlignedObjectArray<bool> results;
volumes.resize(cfgLeaves);
results.resize(cfgLeaves);
for(int i=0;i<cfgLeaves;++i)
{
volumes[i]=btDbvtBenchmark::RandVolume(cfgVolumeCenterScale,cfgVolumeExentsBase,cfgVolumeExentsScale);
}
printf("[12] btDbvtVolume notequal: ");
wallclock.reset();
for(int i=0;i<cfgBenchmark12_Iterations;++i)
{
for(int j=0;j<cfgLeaves;++j)
{
for(int k=0;k<cfgLeaves;++k)
{
results[k]=NotEqual(volumes[j],volumes[k]);
}
}
}
const int time=(int)wallclock.getTimeMilliseconds();
printf("%u ms (%i%%)\r\n",time,(time-cfgBenchmark12_Reference)*100/time);
}
if(cfgBenchmark13_Enable)
{// Benchmark 13
srand(380843);
btDbvt dbvt;
btAlignedObjectArray<btVector3> vectors;
btDbvtBenchmark::NilPolicy policy;
vectors.resize(cfgBenchmark13_Iterations);
for(int i=0;i<vectors.size();++i)
{
vectors[i]=(btDbvtBenchmark::RandVector3()*2-btVector3(1,1,1)).normalized();
}
btDbvtBenchmark::RandTree(cfgVolumeCenterScale,cfgVolumeExentsBase,cfgVolumeExentsScale,cfgLeaves,dbvt);
dbvt.optimizeTopDown();
printf("[13] culling(OCL+fullsort): ");
wallclock.reset();
for(int i=0;i<cfgBenchmark13_Iterations;++i)
{
static const btScalar offset=0;
policy.m_depth=-SIMD_INFINITY;
dbvt.collideOCL(dbvt.m_root,&vectors[i],&offset,vectors[i],1,policy);
}
const int time=(int)wallclock.getTimeMilliseconds();
const int t=cfgBenchmark13_Iterations;
printf("%u ms (%i%%),(%u t/s)\r\n",time,(time-cfgBenchmark13_Reference)*100/time,(t*1000)/time);
}
if(cfgBenchmark14_Enable)
{// Benchmark 14
srand(380843);
btDbvt dbvt;
btAlignedObjectArray<btVector3> vectors;
btDbvtBenchmark::P14 policy;
vectors.resize(cfgBenchmark14_Iterations);
for(int i=0;i<vectors.size();++i)
{
vectors[i]=(btDbvtBenchmark::RandVector3()*2-btVector3(1,1,1)).normalized();
}
btDbvtBenchmark::RandTree(cfgVolumeCenterScale,cfgVolumeExentsBase,cfgVolumeExentsScale,cfgLeaves,dbvt);
dbvt.optimizeTopDown();
policy.m_nodes.reserve(cfgLeaves);
printf("[14] culling(OCL+qsort): ");
wallclock.reset();
for(int i=0;i<cfgBenchmark14_Iterations;++i)
{
static const btScalar offset=0;
policy.m_nodes.resize(0);
dbvt.collideOCL(dbvt.m_root,&vectors[i],&offset,vectors[i],1,policy,false);
policy.m_nodes.quickSort(btDbvtBenchmark::P14::sortfnc);
}
const int time=(int)wallclock.getTimeMilliseconds();
const int t=cfgBenchmark14_Iterations;
printf("%u ms (%i%%),(%u t/s)\r\n",time,(time-cfgBenchmark14_Reference)*100/time,(t*1000)/time);
}
if(cfgBenchmark15_Enable)
{// Benchmark 15
srand(380843);
btDbvt dbvt;
btAlignedObjectArray<btVector3> vectors;
btDbvtBenchmark::P15 policy;
vectors.resize(cfgBenchmark15_Iterations);
for(int i=0;i<vectors.size();++i)
{
vectors[i]=(btDbvtBenchmark::RandVector3()*2-btVector3(1,1,1)).normalized();
}
btDbvtBenchmark::RandTree(cfgVolumeCenterScale,cfgVolumeExentsBase,cfgVolumeExentsScale,cfgLeaves,dbvt);
dbvt.optimizeTopDown();
policy.m_nodes.reserve(cfgLeaves);
printf("[15] culling(KDOP+qsort): ");
wallclock.reset();
for(int i=0;i<cfgBenchmark15_Iterations;++i)
{
static const btScalar offset=0;
policy.m_nodes.resize(0);
policy.m_axis=vectors[i];
dbvt.collideKDOP(dbvt.m_root,&vectors[i],&offset,1,policy);
policy.m_nodes.quickSort(btDbvtBenchmark::P15::sortfnc);
}
const int time=(int)wallclock.getTimeMilliseconds();
const int t=cfgBenchmark15_Iterations;
printf("%u ms (%i%%),(%u t/s)\r\n",time,(time-cfgBenchmark15_Reference)*100/time,(t*1000)/time);
}
if(cfgBenchmark16_Enable)
{// Benchmark 16
srand(380843);
btDbvt dbvt;
btAlignedObjectArray<btDbvtNode*> batch;
btDbvtBenchmark::RandTree(cfgVolumeCenterScale,cfgVolumeExentsBase,cfgVolumeExentsScale,cfgLeaves,dbvt);
dbvt.optimizeTopDown();
batch.reserve(cfgBenchmark16_BatchCount);
printf("[16] insert/remove batch(%u): ",cfgBenchmark16_BatchCount);
wallclock.reset();
for(int i=0;i<cfgBenchmark16_Passes;++i)
{
for(int j=0;j<cfgBenchmark16_BatchCount;++j)
{
batch.push_back(dbvt.insert(btDbvtBenchmark::RandVolume(cfgVolumeCenterScale,cfgVolumeExentsBase,cfgVolumeExentsScale),0));
}
for(int j=0;j<cfgBenchmark16_BatchCount;++j)
{
dbvt.remove(batch[j]);
}
batch.resize(0);
}
const int time=(int)wallclock.getTimeMilliseconds();
const int ir=cfgBenchmark16_Passes*cfgBenchmark16_BatchCount;
printf("%u ms (%i%%),(%u bir/s)\r\n",time,(time-cfgBenchmark16_Reference)*100/time,int(ir*1000.0/time));
}
if(cfgBenchmark17_Enable)
{// Benchmark 17
srand(380843);
btAlignedObjectArray<btDbvtVolume> volumes;
btAlignedObjectArray<int> results;
btAlignedObjectArray<int> indices;
volumes.resize(cfgLeaves);
results.resize(cfgLeaves);
indices.resize(cfgLeaves);
for(int i=0;i<cfgLeaves;++i)
{
indices[i]=i;
volumes[i]=btDbvtBenchmark::RandVolume(cfgVolumeCenterScale,cfgVolumeExentsBase,cfgVolumeExentsScale);
}
for(int i=0;i<cfgLeaves;++i)
{
btSwap(indices[i],indices[rand()%cfgLeaves]);
}
printf("[17] btDbvtVolume select: ");
wallclock.reset();
for(int i=0;i<cfgBenchmark17_Iterations;++i)
{
for(int j=0;j<cfgLeaves;++j)
{
for(int k=0;k<cfgLeaves;++k)
{
const int idx=indices[k];
results[idx]=Select(volumes[idx],volumes[j],volumes[k]);
}
}
}
const int time=(int)wallclock.getTimeMilliseconds();
printf("%u ms (%i%%)\r\n",time,(time-cfgBenchmark17_Reference)*100/time);
}
printf("\r\n\r\n");
}
void btRadixSort32CL::executeHost(btAlignedObjectArray<btSortData>& inout, int sortBits /* = 32 */)
{
int n = inout.size();
const int BITS_PER_PASS = 8;
const int NUM_TABLES = (1<<BITS_PER_PASS);
int tables[NUM_TABLES];
int counter[NUM_TABLES];
btSortData* src = &inout[0];
btAlignedObjectArray<btSortData> workbuffer;
workbuffer.resize(inout.size());
btSortData* dst = &workbuffer[0];
int count=0;
for(int startBit=0; startBit<sortBits; startBit+=BITS_PER_PASS)
{
for(int i=0; i<NUM_TABLES; i++)
{
tables[i] = 0;
}
for(int i=0; i<n; i++)
{
int tableIdx = (src[i].m_key >> startBit) & (NUM_TABLES-1);
tables[tableIdx]++;
}
//#define TEST
#ifdef TEST
printf("histogram size=%d\n",NUM_TABLES);
for (int i=0;i<NUM_TABLES;i++)
{
if (tables[i]!=0)
{
printf("tables[%d]=%d]\n",i,tables[i]);
}
}
#endif //TEST
// prefix scan
int sum = 0;
for(int i=0; i<NUM_TABLES; i++)
{
int iData = tables[i];
tables[i] = sum;
sum += iData;
counter[i] = 0;
}
// distribute
for(int i=0; i<n; i++)
{
int tableIdx = (src[i].m_key >> startBit) & (NUM_TABLES-1);
dst[tables[tableIdx] + counter[tableIdx]] = src[i];
counter[tableIdx] ++;
}
btSwap( src, dst );
count++;
}
if (count&1)
{
btAssert(0);//need to copy
}
}
void btRadixSort32CL::execute(btOpenCLArray<btSortData>& keyValuesInOut, int sortBits /* = 32 */)
{
int originalSize = keyValuesInOut.size();
int workingSize = originalSize;
int dataAlignment = DATA_ALIGNMENT;
#ifdef DEBUG_RADIXSORT2
btAlignedObjectArray<btSortData> test2;
keyValuesInOut.copyToHost(test2);
printf("numElem = %d\n",test2.size());
for (int i=0;i<test2.size();i++)
{
printf("test2[%d].m_key=%d\n",i,test2[i].m_key);
printf("test2[%d].m_value=%d\n",i,test2[i].m_value);
}
#endif //DEBUG_RADIXSORT2
btOpenCLArray<btSortData>* src = 0;
if (workingSize%dataAlignment)
{
workingSize += dataAlignment-(workingSize%dataAlignment);
m_workBuffer4->copyFromOpenCLArray(keyValuesInOut);
m_workBuffer4->resize(workingSize);
btSortData fillValue;
fillValue.m_key = 0xffffffff;
fillValue.m_value = 0xffffffff;
#define USE_BTFILL
#ifdef USE_BTFILL
m_fill->execute((btOpenCLArray<btInt2>&)*m_workBuffer4,(btInt2&)fillValue,workingSize-originalSize,originalSize);
#else
//fill the remaining bits (very slow way, todo: fill on GPU/OpenCL side)
for (int i=originalSize; i<workingSize;i++)
{
m_workBuffer4->copyFromHostPointer(&fillValue,1,i);
}
#endif//USE_BTFILL
src = m_workBuffer4;
} else
{
src = &keyValuesInOut;
m_workBuffer4->resize(0);
}
btAssert( workingSize%DATA_ALIGNMENT == 0 );
int minCap = NUM_BUCKET*NUM_WGS;
int n = workingSize;
m_workBuffer1->resize(minCap);
m_workBuffer3->resize(workingSize);
// ADLASSERT( ELEMENTS_PER_WORK_ITEM == 4 );
btAssert( BITS_PER_PASS == 4 );
btAssert( WG_SIZE == 64 );
btAssert( (sortBits&0x3) == 0 );
btOpenCLArray<btSortData>* dst = m_workBuffer3;
btOpenCLArray<unsigned int>* srcHisto = m_workBuffer1;
btOpenCLArray<unsigned int>* destHisto = m_workBuffer2;
int nWGs = NUM_WGS;
btConstData cdata;
{
int blockSize = ELEMENTS_PER_WORK_ITEM*WG_SIZE;//set at 256
int nBlocks = (n+blockSize-1)/(blockSize);
cdata.m_n = n;
cdata.m_nWGs = NUM_WGS;
cdata.m_startBit = 0;
cdata.m_nBlocksPerWG = (nBlocks + cdata.m_nWGs - 1)/cdata.m_nWGs;
if( nBlocks < NUM_WGS )
{
cdata.m_nBlocksPerWG = 1;
nWGs = nBlocks;
}
}
int count=0;
for(int ib=0; ib<sortBits; ib+=4)
{
#ifdef DEBUG_RADIXSORT2
keyValuesInOut.copyToHost(test2);
printf("numElem = %d\n",test2.size());
for (int i=0;i<test2.size();i++)
{
if (test2[i].m_key != test2[i].m_value)
{
printf("test2[%d].m_key=%d\n",i,test2[i].m_key);
printf("test2[%d].m_value=%d\n",i,test2[i].m_value);
}
}
#endif //DEBUG_RADIXSORT2
cdata.m_startBit = ib;
{
btBufferInfoCL bInfo[] = { btBufferInfoCL( src->getBufferCL(), true ), btBufferInfoCL( srcHisto->getBufferCL() ) };
btLauncherCL launcher(m_commandQueue, m_streamCountSortDataKernel);
launcher.setBuffers( bInfo, sizeof(bInfo)/sizeof(btBufferInfoCL) );
launcher.setConst( cdata );
int num = NUM_WGS*WG_SIZE;
launcher.launch1D( num, WG_SIZE );
}
#ifdef DEBUG_RADIXSORT
btAlignedObjectArray<unsigned int> testHist;
srcHisto->copyToHost(testHist);
printf("ib = %d, testHist size = %d, non zero elements:\n",ib, testHist.size());
for (int i=0;i<testHist.size();i++)
{
if (testHist[i]!=0)
printf("testHist[%d]=%d\n",i,testHist[i]);
}
#endif //DEBUG_RADIXSORT
//fast prefix scan is not working properly on Mac OSX yet
#ifdef _WIN32
bool fastScan=true;
#else
bool fastScan=false;
#endif
if (fastScan)
{// prefix scan group histogram
btBufferInfoCL bInfo[] = { btBufferInfoCL( srcHisto->getBufferCL() ) };
btLauncherCL launcher( m_commandQueue, m_prefixScanKernel );
launcher.setBuffers( bInfo, sizeof(bInfo)/sizeof(btBufferInfoCL) );
launcher.setConst( cdata );
launcher.launch1D( 128, 128 );
destHisto = srcHisto;
}else
{
//unsigned int sum; //for debugging
m_scan->execute(*srcHisto,*destHisto,1920,0);//,&sum);
}
#ifdef DEBUG_RADIXSORT
destHisto->copyToHost(testHist);
printf("ib = %d, testHist size = %d, non zero elements:\n",ib, testHist.size());
for (int i=0;i<testHist.size();i++)
{
if (testHist[i]!=0)
printf("testHist[%d]=%d\n",i,testHist[i]);
}
for (int i=0;i<testHist.size();i+=NUM_WGS)
{
printf("testHist[%d]=%d\n",i/NUM_WGS,testHist[i]);
}
#endif //DEBUG_RADIXSORT
#define USE_GPU
#ifdef USE_GPU
{// local sort and distribute
btBufferInfoCL bInfo[] = { btBufferInfoCL( src->getBufferCL(), true ), btBufferInfoCL( destHisto->getBufferCL(), true ), btBufferInfoCL( dst->getBufferCL() )};
btLauncherCL launcher( m_commandQueue, m_sortAndScatterSortDataKernel );
launcher.setBuffers( bInfo, sizeof(bInfo)/sizeof(btBufferInfoCL) );
launcher.setConst( cdata );
launcher.launch1D( nWGs*WG_SIZE, WG_SIZE );
}
#else
{
#define NUM_TABLES 16
//#define SEQUENTIAL
#ifdef SEQUENTIAL
int counter2[NUM_TABLES]={0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0};
int tables[NUM_TABLES];
int startBit = ib;
destHisto->copyToHost(testHist);
btAlignedObjectArray<btSortData> srcHost;
btAlignedObjectArray<btSortData> dstHost;
dstHost.resize(src->size());
src->copyToHost(srcHost);
for (int i=0;i<NUM_TABLES;i++)
{
tables[i] = testHist[i*NUM_WGS];
}
// distribute
for(int i=0; i<n; i++)
{
int tableIdx = (srcHost[i].m_key >> startBit) & (NUM_TABLES-1);
dstHost[tables[tableIdx] + counter2[tableIdx]] = srcHost[i];
counter2[tableIdx] ++;
}
#else
int counter2[NUM_TABLES]={0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0};
int tables[NUM_TABLES];
btAlignedObjectArray<btSortData> dstHostOK;
dstHostOK.resize(src->size());
destHisto->copyToHost(testHist);
btAlignedObjectArray<btSortData> srcHost;
src->copyToHost(srcHost);
int blockSize = 256;
int nBlocksPerWG = cdata.m_nBlocksPerWG;
int startBit = ib;
{
for (int i=0;i<NUM_TABLES;i++)
{
tables[i] = testHist[i*NUM_WGS];
}
// distribute
for(int i=0; i<n; i++)
{
int tableIdx = (srcHost[i].m_key >> startBit) & (NUM_TABLES-1);
dstHostOK[tables[tableIdx] + counter2[tableIdx]] = srcHost[i];
counter2[tableIdx] ++;
}
}
btAlignedObjectArray<btSortData> dstHost;
dstHost.resize(src->size());
int counter[NUM_TABLES]={0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0};
for (int wgIdx=0;wgIdx<NUM_WGS;wgIdx++)
{
int counter[NUM_TABLES]={0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0};
int nBlocks = (n)/blockSize - nBlocksPerWG*wgIdx;
for(int iblock=0; iblock<btMin(cdata.m_nBlocksPerWG, nBlocks); iblock++)
{
for (int lIdx = 0;lIdx < 64;lIdx++)
{
int addr = iblock*blockSize + blockSize*cdata.m_nBlocksPerWG*wgIdx + ELEMENTS_PER_WORK_ITEM*lIdx;
// MY_HISTOGRAM( localKeys.x ) ++ is much expensive than atomic add as it requires read and write while atomics can just add on AMD
// Using registers didn't perform well. It seems like use localKeys to address requires a lot of alu ops
// AMD: AtomInc performs better while NV prefers ++
for(int j=0; j<ELEMENTS_PER_WORK_ITEM; j++)
{
if( addr+j < n )
{
// printf ("addr+j=%d\n", addr+j);
int i = addr+j;
int tableIdx = (srcHost[i].m_key >> startBit) & (NUM_TABLES-1);
int destIndex = testHist[tableIdx*NUM_WGS+wgIdx] + counter[tableIdx];
btSortData ok = dstHostOK[destIndex];
if (ok.m_key != srcHost[i].m_key)
{
printf("ok.m_key = %d, srcHost[i].m_key = %d\n", ok.m_key,srcHost[i].m_key );
printf("(ok.m_value = %d, srcHost[i].m_value = %d)\n", ok.m_value,srcHost[i].m_value );
}
if (ok.m_value != srcHost[i].m_value)
{
printf("ok.m_value = %d, srcHost[i].m_value = %d\n", ok.m_value,srcHost[i].m_value );
printf("(ok.m_key = %d, srcHost[i].m_key = %d)\n", ok.m_key,srcHost[i].m_key );
}
dstHost[destIndex] = srcHost[i];
counter[tableIdx] ++;
}
}
}
}
}
#endif //SEQUENTIAL
dst->copyFromHost(dstHost);
}
#endif//USE_GPU
#ifdef DEBUG_RADIXSORT
destHisto->copyToHost(testHist);
printf("ib = %d, testHist size = %d, non zero elements:\n",ib, testHist.size());
for (int i=0;i<testHist.size();i++)
{
if (testHist[i]!=0)
printf("testHist[%d]=%d\n",i,testHist[i]);
}
#endif //DEBUG_RADIXSORT
btSwap(src, dst );
btSwap(srcHisto,destHisto);
#ifdef DEBUG_RADIXSORT2
keyValuesInOut.copyToHost(test2);
printf("numElem = %d\n",test2.size());
for (int i=0;i<test2.size();i++)
{
if (test2[i].m_key != test2[i].m_value)
{
printf("test2[%d].m_key=%d\n",i,test2[i].m_key);
printf("test2[%d].m_value=%d\n",i,test2[i].m_value);
}
}
#endif //DEBUG_RADIXSORT2
count++;
}
bool btPolyhedralConvexShape::initializePolyhedralFeatures()
{
if (m_polyhedron)
btAlignedFree(m_polyhedron);
void* mem = btAlignedAlloc(sizeof(btConvexPolyhedron),16);
m_polyhedron = new (mem) btConvexPolyhedron;
btAlignedObjectArray<btVector3> tmpVertices;
for (int i=0;i<getNumVertices();i++)
{
btVector3& newVertex = tmpVertices.expand();
getVertex(i,newVertex);
}
btConvexHullComputer conv;
conv.compute(&tmpVertices[0].getX(), sizeof(btVector3),tmpVertices.size(),0.f,0.f);
btAlignedObjectArray<btVector3> faceNormals;
int numFaces = conv.faces.size();
faceNormals.resize(numFaces);
btConvexHullComputer* convexUtil = &conv;
m_polyhedron->m_faces.resize(numFaces);
int numVertices = convexUtil->vertices.size();
m_polyhedron->m_vertices.resize(numVertices);
for (int p=0;p<numVertices;p++)
{
m_polyhedron->m_vertices[p] = convexUtil->vertices[p];
}
for (int i=0;i<numFaces;i++)
{
int face = convexUtil->faces[i];
//printf("face=%d\n",face);
const btConvexHullComputer::Edge* firstEdge = &convexUtil->edges[face];
const btConvexHullComputer::Edge* edge = firstEdge;
btVector3 edges[3];
int numEdges = 0;
//compute face normals
//btScalar maxCross2 = 0.f;
//int chosenEdge = -1;
do
{
int src = edge->getSourceVertex();
m_polyhedron->m_faces[i].m_indices.push_back(src);
int targ = edge->getTargetVertex();
btVector3 wa = convexUtil->vertices[src];
btVector3 wb = convexUtil->vertices[targ];
btVector3 newEdge = wb-wa;
newEdge.normalize();
if (numEdges<2)
edges[numEdges++] = newEdge;
edge = edge->getNextEdgeOfFace();
} while (edge!=firstEdge);
btScalar planeEq = 1e30f;
if (numEdges==2)
{
faceNormals[i] = edges[0].cross(edges[1]);
faceNormals[i].normalize();
m_polyhedron->m_faces[i].m_plane[0] = -faceNormals[i].getX();
m_polyhedron->m_faces[i].m_plane[1] = -faceNormals[i].getY();
m_polyhedron->m_faces[i].m_plane[2] = -faceNormals[i].getZ();
m_polyhedron->m_faces[i].m_plane[3] = planeEq;
}
else
{
btAssert(0);//degenerate?
faceNormals[i].setZero();
}
for (int v=0;v<m_polyhedron->m_faces[i].m_indices.size();v++)
{
btScalar eq = m_polyhedron->m_vertices[m_polyhedron->m_faces[i].m_indices[v]].dot(faceNormals[i]);
if (planeEq>eq)
{
planeEq=eq;
}
}
m_polyhedron->m_faces[i].m_plane[3] = planeEq;
}
if (m_polyhedron->m_faces.size() && conv.vertices.size())
{
for (int f=0;f<m_polyhedron->m_faces.size();f++)
{
btVector3 planeNormal(m_polyhedron->m_faces[f].m_plane[0],m_polyhedron->m_faces[f].m_plane[1],m_polyhedron->m_faces[f].m_plane[2]);
btScalar planeEq = m_polyhedron->m_faces[f].m_plane[3];
btVector3 supVec = localGetSupportingVertex(-planeNormal);
if (supVec.dot(planeNormal)<planeEq)
{
m_polyhedron->m_faces[f].m_plane[0] *= -1;
m_polyhedron->m_faces[f].m_plane[1] *= -1;
m_polyhedron->m_faces[f].m_plane[2] *= -1;
m_polyhedron->m_faces[f].m_plane[3] *= -1;
int numVerts = m_polyhedron->m_faces[f].m_indices.size();
for (int v=0;v<numVerts/2;v++)
{
btSwap(m_polyhedron->m_faces[f].m_indices[v],m_polyhedron->m_faces[f].m_indices[numVerts-1-v]);
}
}
}
}
m_polyhedron->initialize();
return true;
}
