PxU32 InternalIndexPool::allocateIndices(PxU32 num, const PxStrideIterator<PxU32>& indexBuffer)
{
PxU32 numAllocated = mIndexCount - mFreeList.size();
PxU32 numFree = mFreeList.capacity() - numAllocated;
PxU32 numAccepted = PxMin(num, numFree);
if (numAccepted < num)
Ps::getFoundation().error(PxErrorCode::eDEBUG_WARNING, __FILE__, __LINE__,
"PxParticleExt::IndexPool::allocateIndices: Not all requested indices allocated; maximum reached.");
PxU32 numAdded = 0;
PxU32 numToAddFromFreeList = PxMin(numAccepted, mFreeList.size());
PxU32 numToAddFromFullBlock = numAccepted - numToAddFromFreeList;
//add from free list
while (numToAddFromFreeList > 0)
{
indexBuffer[numAdded++] = mFreeList.popBack();
numToAddFromFreeList--;
}
//add from full block
while (numToAddFromFullBlock > 0)
{
indexBuffer[numAdded++] = mIndexCount++;
numToAddFromFullBlock--;
}
PX_ASSERT(numAdded == numAccepted);
return numAccepted;
}
void DeformableMesh::generateConstraintsFromTriangles()
{
PxU32 numTriangles = mPrimitives.size() / 3;
DeformableTriEdge e;
Ps::Array<DeformableTriEdge> edges PX_DEBUG_EXP("defoMeshEdges");
edges.reserve(numTriangles * 3);
const PxU32 *i0 = mPrimitives.begin();
for (PxU32 i = 0; i < numTriangles; i++, i0 += 3)
{
const PxU32 *i1 = i0+1;
const PxU32 *i2 = i0+2;
e.set(mVertexToParticleMap[*i0], mVertexToParticleMap[*i1], mVertexToParticleMap[*i2], i);
edges.pushBack(e);
e.set(mVertexToParticleMap[*i1], mVertexToParticleMap[*i2], mVertexToParticleMap[*i0], i);
edges.pushBack(e);
e.set(mVertexToParticleMap[*i2], mVertexToParticleMap[*i0], mVertexToParticleMap[*i1], i);
edges.pushBack(e);
}
quickSortEdges(edges, 0, edges.size()-1);
DeformableConstraint constraint;
constraint.particleId[4] = -1; // only used when torn
constraint.particleId[5] = -1; // only used when torn
for(PxU32 i=0; i<edges.size(); )
{
const DeformableTriEdge &e0 = edges[i];
PxU32 p0 = constraint.particleId[0] = e0.particleId[0];
PxU32 p1 = constraint.particleId[1] = e0.particleId[1];
PxVec3 d0 = mWeldedVertices[p1] - mWeldedVertices[p0];
constraint.stretchingRestLength = d0.magnitude();
constraint.particleId[2] = e0.third;
constraint.particleId[3] = -1; // for border edges -> no bending possible
constraint.bendingRestLength = 0.0f;
constraint.flags = 0;
if (++i < edges.size()) {
const DeformableTriEdge &e1 = edges[i];
if (e0 == e1) {
constraint.particleId[2] = e0.third;
constraint.particleId[3] = e1.third;
PxVec3 d1 = mWeldedVertices[e1.third] - mWeldedVertices[e0.third];
constraint.bendingRestLength = d1.magnitude();
}
while (i < edges.size() && edges[i] == e0)
++i;
}
mConstraints.pushBack(constraint);
}
}
void InternalIndexPool::freeIndices(PxU32 num, const PxStrideIterator<const PxU32>& indexBuffer)
{
PxU32 numAllocated = mIndexCount - mFreeList.size();
if (num > numAllocated)
{
Ps::getFoundation().error(PxErrorCode::eDEBUG_WARNING, __FILE__, __LINE__,
"PxParticleExt::IndexPool::freeIndices: Provided number of indices exceeds number of actually allocated indices. Call faild.");
return;
}
#ifdef PX_CHECK
for (PxU32 i = 0; i < num; ++i)
{
if (indexBuffer[i] < mIndexCount)
continue;
Ps::getFoundation().error(PxErrorCode::eDEBUG_WARNING, __FILE__, __LINE__,
"PxParticleExt::IndexPool::freeIndices: Provided indices which where not actually allocated before. Call failed.");
return;
}
#endif
for (PxU32 i = 0; i < num; ++i)
mFreeList.pushBack(indexBuffer[i]);
}
static void accumulate(PvdHitType& query, Ps::Array<PvdHitType>& accumulated, const char* arrayName, Ps::Array<PvdSqHit>& dst, const SDKHitType* src, PxU32 nb, const PxQueryFilterData& fd)
{
query.mFilterFlags = fd.flags;
query.mHits = PvdReference(arrayName, dst.size(), nb);
PX_ASSERT(PxU32(-1) != nb);
for(PxU32 i=0; i<nb; i++)
dst.pushBack(PvdSqHit(src[i]));
accumulated.pushBack(query);
}
void ProfileEventHandler::reportCudaCollection(PxBufferedProfilerCallback& callback)
{
if(mCUDAEventOccurences.empty())
return;
// sort the cuda occurrences - according to eventID and startTime
Ps::sort(mCUDAEventOccurences.begin(), mCUDAEventOccurences.size(), SortCUDAProfileOccurences());
PxU16 currentEventId = mCUDAEventOccurences[0].eventId;
CUDAProfileEventOccurence currentEvent = mCUDAEventOccurences[0];
Ps::Array<CUDAProfileEventOccurence> outOccurences;
// now group the occurrences that do overlap
// put the group occurrences into out array
for (PxU32 i = 1; i < mCUDAEventOccurences.size(); i++)
{
const CUDAProfileEventOccurence& occurence = mCUDAEventOccurences[i];
// occurrences are sorted by eventId, so once it changes, we have new group occurrence
if(currentEventId != occurence.eventId)
{
outOccurences.pushBack(currentEvent);
currentEvent = occurence;
currentEventId = occurence.eventId;
}
else
{
// occurrences are sorted by start time, so we check for overlap and add occurrence or create new group
if(currentEvent.endTime >= occurence.startTime)
{
currentEvent.endTime = PxMax(currentEvent.endTime, occurence.endTime);
}
else
{
outOccurences.pushBack(currentEvent);
currentEvent = occurence;
}
}
}
outOccurences.pushBack(currentEvent);
// fire the callback with grouped occurrences
for (PxU32 i = outOccurences.size(); i--;)
{
const CUDAProfileEventOccurence& ev = outOccurences[i];
const char* name = mProfileZoneInterface->getProfileEventName(ev.eventId);
const char* profileZoneName = mProfileZoneInterface->getName();
PxBufferedProfilerEvent ce = { ev.startTime, ev.endTime, name, profileZoneName, ev.eventId, 0, 0, 0, 0 };
callback.onEvent(ce);
}
}
void ProfileEventHandler::onCUDAProfileBuffer( PxU64 , PxF32 , const PxU8* cudaData, PxU32 bufLenInBytes, PxU32 bufferVersion )
{
if(bufferVersion == 1)
{
struct currentWarpProfileEvent
{
PxU16 block;
PxU8 warp;
PxU8 mpId;
PxU8 hwWarpId;
PxU8 userDataCfg;
PxU16 eventId;
PxU32 startTime;
PxU32 endTime;
};
Ps::Array<Local::OverflowRecordPair> overflowPair;
//Run through dataset. We need to be able to correct rollover, meaning one of the timers
//runs through PxU32.MAX_VALUE and resets to zero.
PxU32 numEvents = bufLenInBytes/sizeof(currentWarpProfileEvent);
const currentWarpProfileEvent* cudaEvents = reinterpret_cast<const currentWarpProfileEvent*> (cudaData);
for (PxU32 i = 0; i < numEvents; i++)
{
const currentWarpProfileEvent& cudaEvent = cudaEvents[i];
Local::OverflowRecord* record = NULL;
for (PxU32 j = 0; j < overflowPair.size(); j++)
{
if(overflowPair[j].mpId == cudaEvent.mpId)
{
record = &overflowPair[j].mOverflowRecord;
break;
}
}
if(!record)
{
Local::OverflowRecordPair pair;
pair.mpId = cudaEvent.mpId;
overflowPair.pushBack(pair);
record = &overflowPair.back().mOverflowRecord;
}
//account for overflow
PxU64 startTime = record->NextValue(cudaEvent.startTime);
PxU64 endTime = record->NextValue(cudaEvent.endTime);
CUDAProfileEventOccurence cudaEventOccurence = { cudaEvent.eventId, startTime, endTime };
mCUDAEventOccurences.pushBack(cudaEventOccurence);
}
}
}
void ProfileEventHandler::reportEvents(Ps::Array<PxBufferedProfilerCallback*>& callbacks)
{
PX_ASSERT(mProfileZoneInterface);
for (PxU32 callbackIndex = callbacks.size(); callbackIndex--; )
{
PxBufferedProfilerCallback& callback = *callbacks[callbackIndex];
reportCollection(callback, mCrossThreadCollection);
for (PxU32 i = mThreadCollections.size(); i--;)
{
reportCollection(callback, mThreadCollections[i]);
}
reportCudaCollection(callback);
}
clear();
}
static void collectBatchedHits(const QueryResultT* results, Ps::Array<PvdHitType>& accumulated, Ps::Array<PvdSqHit>& pvdSqHits, PxU32 nb, PxU32 startIdx, const char* arrayName)
{
for(PxU32 i=0; i<nb; i++)
{
const QueryResultT& result = results[i];
if(result.queryStatus != PxBatchQueryStatus::eSUCCESS)
continue;
PvdHitType& query = accumulated[startIdx + i];
const PxU32 nbAnyHits = result.getNbAnyHits();
if(query.mHits.mCount != nbAnyHits)
{
query.mHits = PvdReference(arrayName, pvdSqHits.size(), nbAnyHits);
for(PxU32 j=0; j<nbAnyHits; j++)
pvdSqHits.pushBack(PvdSqHit(result.getAnyHit(j)));
}
}
}
void CharacterControllerManager::computeInteractions(PxF32 elapsedTime)
{
PxU32 nbControllers = mControllers.size();
Controller** controllers = mControllers.begin();
PxBounds3* boxes = new PxBounds3[nbControllers]; // PT: TODO: get rid of alloc
PxBounds3* runningBoxes = boxes;
while(nbControllers--)
{
Controller* current = *controllers++;
PxExtendedBounds3 extBox;
current->getWorldBox(extBox);
*runningBoxes++ = PxBounds3(toVec3(extBox.minimum), toVec3(extBox.maximum)); // ### LOSS OF ACCURACY
}
//
const PxU32 nbEntities = runningBoxes - boxes;
Ps::Array<PxU32> pairs; // PT: TODO: get rid of alloc
CompleteBoxPruning(boxes, nbEntities, pairs, Gu::Axes(physx::Gu::AXES_XZY)); // PT: TODO: revisit for variable up axis
PxU32 nbPairs = pairs.size()>>1;
const PxU32* indices = pairs.begin();
while(nbPairs--)
{
const PxU32 index0 = *indices++;
const PxU32 index1 = *indices++;
Controller* ctrl0 = mControllers[index0];
Controller* ctrl1 = mControllers[index1];
InteractionCharacterCharacter(ctrl0, ctrl1, elapsedTime);
}
delete [] boxes;
}
PxU32 PxParticleExt::buildBoundsHash(PxU32* sortedParticleIndices,
ParticleBounds* particleBounds,
const PxStrideIterator<const PxVec3>& positionBuffer,
const PxU32 validParticleRange,
const PxU32* validParticleBitmap,
const PxU32 hashSize,
const PxU32 maxBounds,
const PxReal gridSpacing)
{
// test if hash size is a multiple of 2
PX_ASSERT((((hashSize - 1) ^ hashSize) + 1) == (2 * hashSize));
PX_ASSERT(maxBounds <= hashSize);
PxReal cellSizeInv = 1.0f / gridSpacing;
Ps::Array<PxU32> particleToCellMap PX_DEBUG_EXP("buildBoundsHashCellMap");
particleToCellMap.resize(validParticleRange);
// initialize cells
Ps::Array<Cell> cells PX_DEBUG_EXP("buildBoundsCells");
cells.resize(hashSize);
PxMemSet(cells.begin(), sInvalidIndex, sizeof(Cell) * hashSize);
// count number of particles per cell
PxU32 entryCounter = 0;
if (validParticleRange > 0)
{
for (PxU32 w = 0; w <= (validParticleRange-1) >> 5; w++)
for (PxU32 b = validParticleBitmap[w]; b; b &= b-1)
{
PxU32 index = (w<<5|Ps::lowestSetBit(b));
const PxVec3& position = positionBuffer[index];
PxU32& cellIndex = particleToCellMap[index];
cellIndex = sInvalidIndex; // initialize to invalid in case we reach maxBounds
if (entryCounter < maxBounds)
{
CellCoords particleCoords;
particleCoords.set(position, cellSizeInv);
cellIndex = getEntry(particleCoords, hashSize, cells.begin());
PX_ASSERT(cellIndex != sInvalidIndex);
Cell& cell = cells[cellIndex];
if (cell.size == sInvalidIndex)
{
// this is the first particle in this cell
cell.coords = particleCoords;
cell.aabb = PxBounds3(position, position);
cell.size = 1;
++entryCounter;
}
else
{
// the cell is already occupied
cell.aabb.include(position);
++cell.size;
}
}
}
}
// accumulate start indices from cell size histogram and write to the user's particleBounds buffer
PxU32 numBounds = 0;
for (PxU32 i = 0, counter = 0; i < cells.size(); i++)
{
Cell& cell = cells[i];
if (cell.size != sInvalidIndex)
{
cell.start = counter;
counter += cell.size;
PxParticleExt::ParticleBounds& cellBounds = particleBounds[numBounds++];
PX_ASSERT(cell.aabb.isValid());
cellBounds.bounds = cell.aabb;
cellBounds.firstParticle = cell.start;
cellBounds.numParticles = cell.size;
cell.size = 0;
}
}
// sort output particle indices by cell
if (validParticleRange > 0)
{
for (PxU32 w = 0; w <= (validParticleRange-1) >> 5; w++)
for (PxU32 b = validParticleBitmap[w]; b; b &= b-1)
{
PxU32 index = (w<<5|Ps::lowestSetBit(b));
PxU32 cellIndex = particleToCellMap[index];
if (cellIndex != sInvalidIndex)
{
Cell& cell = cells[cellIndex];
PX_ASSERT(cell.start != sInvalidIndex && cell.size != sInvalidIndex);
sortedParticleIndices[cell.start + cell.size] = index;
++cell.size;
}
}
}
return numBounds;
}
template<class Type> static void pushBackT(Ps::Array<Type>& array, const Type& item, PvdReference& ref, const char* arrayName)
{
ref = PvdReference(arrayName, array.size(), 1);
array.pushBack(item);
}
void DeformableMesh::calculateInvMasses(Ps::Array<PxReal>& invMasses, PxReal mass) const
{
PX_ASSERT(mVertexMasses.empty());
PX_ASSERT(invMasses.size() == mVertexPositions.size());
// clear output array
for (PxU32 i = 0; i < invMasses.size(); i++)
invMasses[i] = 0.0f;
switch (mPrimitiveType)
{
case PxDeformablePrimitiveType::eTRIANGLE:
{
// first, we compute area associated with each vertex and temporarily store it in invMasses array
PxReal totalArea = 0.0f;
const PxU32* i0 = mPrimitives.begin();
for (PxU32 i = 0; i < mPrimitives.size() / 3; ++i)
{
const PxU32* i1 = i0 + 1;
const PxU32* i2 = i0 + 2;
PxVec3 d0(mVertexPositions[*i0].x, mVertexPositions[*i0].y, mVertexPositions[*i0].z);
PxVec3 d1(mVertexPositions[*i1].x, mVertexPositions[*i1].y, mVertexPositions[*i1].z);
PxVec3 d2(mVertexPositions[*i2].x, mVertexPositions[*i2].y, mVertexPositions[*i2].z);
d1 = d1 - d0;
d2 = d2 - d0;
PxVec3 n = d1.cross(d2);
PxReal area = 0.5f * n.magnitude();
invMasses[*i0] += area / 3.0f;
invMasses[*i1] += area / 3.0f;
invMasses[*i2] += area / 3.0f;
totalArea += area;
i0 += 3;
}
// distribute overall mass by associated area
for (PxU32 i = 0; i < invMasses.size(); i++)
{
PxReal area = invMasses[i];
invMasses[i] = (1.0f / mass) * (totalArea / area);
}
}
break;
case PxDeformablePrimitiveType::eTETRAHEDRON:
{
// first, we compute volume associated with each vertex and temporarily store it in invMasses array
PxReal totalVolume = 0.0f;
const PxU32* i0 = mPrimitives.begin();
for (PxU32 i = 0; i < mPrimitives.size() / 4; i++)
{
const PxU32 *i1 = i0 + 1;
const PxU32 *i2 = i0 + 2;
const PxU32 *i3 = i0 + 3;
PxVec3 d0(mVertexPositions[*i0].x, mVertexPositions[*i0].y, mVertexPositions[*i0].z);
PxVec3 d1(mVertexPositions[*i1].x, mVertexPositions[*i1].y, mVertexPositions[*i1].z);
PxVec3 d2(mVertexPositions[*i2].x, mVertexPositions[*i2].y, mVertexPositions[*i2].z);
PxVec3 d3(mVertexPositions[*i3].x, mVertexPositions[*i3].y, mVertexPositions[*i3].z);
d1 = d1 - d0;
d2 = d2 - d0;
d3 = d3 - d0;
PxVec3 n = d1.cross(d2);
PxReal volume = n.dot(d3) / 6.0f;
invMasses[*i0] += volume / 4.0f;
invMasses[*i1] += volume / 4.0f;
invMasses[*i2] += volume / 4.0f;
invMasses[*i3] += volume / 4.0f;
totalVolume += volume;
i0 += 4;
}
// distribute overall mass by associated volume
for (PxU32 i = 0; i < invMasses.size(); i++)
{
PxReal volume = invMasses[i];
invMasses[i] = (1.0f / mass) * (totalVolume / volume);
}
}
break;
default:
PX_ASSERT(0);
break;
}
}
void DeformableMesh::generateConstraintsFromTetrahedra()
{
PxU32 edgeIndices[6][2] = { {0,1}, {0,2}, {0,3}, {1,2}, {1,3}, {2,3} };
PxU32 *tetIndices;
// - tetrahedra are assumed to be unique (thus no parent code here)
PxU32 numTetrahedra = mPrimitives.size() / 4;
DeformableTetraEdge e;
Ps::Array<DeformableTetraEdge> edges PX_DEBUG_EXP("defoMeshEdges2");
edges.reserve(numTetrahedra * 6);
tetIndices = mPrimitives.begin();
PxU32 i, j;
for (i = 0; i < numTetrahedra; i++, tetIndices += 4)
{
for(j = 0; j < 6; j++)
{
PxU32 e0 = mVertexToParticleMap[tetIndices[edgeIndices[j][0]]];
PxU32 e1 = mVertexToParticleMap[tetIndices[edgeIndices[j][1]]];
e.set(e0, e1, i); edges.pushBack(e);
}
}
quickSortTetraEdges(edges, 0, edges.size()-1);
mConstraints.resize(numTetrahedra);
DeformableConstraint constraint;
DeformableTetraEdge *tetEdges[6];
tetIndices = mPrimitives.begin();
bool warningIssued = false;
for (i = 0; i < numTetrahedra; i++, tetIndices += 4)
{
for (j = 0; j < 6; j++)
{
PxU32 e0 = mVertexToParticleMap[tetIndices[edgeIndices[j][0]]];
PxU32 e1 = mVertexToParticleMap[tetIndices[edgeIndices[j][1]]];
DeformableTetraEdge goalEdge;
goalEdge.set(e0, e1, i);
tetEdges[j] = binarySearchTetraEdge(edges, goalEdge);
}
for (j = 0; j < 4; j++)
constraint.particleId[j] = mVertexToParticleMap[tetIndices[j]];
PxVec3 groundArea = (mWeldedVertices[tetIndices[1]] - mWeldedVertices[tetIndices[0]]).cross(mWeldedVertices[tetIndices[2]] - mWeldedVertices[tetIndices[0]]);
constraint.restVolume = groundArea.dot(mWeldedVertices[tetIndices[3]] - mWeldedVertices[tetIndices[0]]);
constraint.flags = 0;
if (!warningIssued && constraint.restVolume < 0.0f)
{
Ps::getFoundation().error(PxErrorCode::eDEBUG_WARNING, __FILE__, __LINE__, "Soft body mesh tetrahedron %d has illegal winding order.", i);
warningIssued = true;
}
for (j = 0; j < 6; j++)
{
PxU32 e0 = mVertexToParticleMap[tetIndices[edgeIndices[j][0]]];
PxU32 e1 = mVertexToParticleMap[tetIndices[edgeIndices[j][1]]];
PxVec3 edgeVec = mWeldedVertices[e1] - mWeldedVertices[e0];
if(tetEdges[j])
{
PX_ASSERT(tetEdges[j]->tetrahedron == i);
constraint.restEdgeLengths[j] = edgeVec.magnitude();
}
else
constraint.restEdgeLengths[j] = -edgeVec.magnitude();
}
mConstraints[i] = constraint;
}
}
void DeformableMesh::weldMesh()
{
mVertexToParticleMap.resize(mVertexPositions.size());
if (mPrimitiveType == PxDeformablePrimitiveType::eTRIANGLE && (mFlags & PxDeformableMeshFlag::eWELD_VERTICES) != 0)
{
Ps::Array<int> order PX_DEBUG_EXP("defoMeshOrder");
order.resize(mVertexPositions.size());
for (PxU32 i = 0; i < order.size(); i++)
order[i] = i;
// sort vertices by their x coordinate
Ps::sort(order.begin(), order.size(), WeldComparator(mVertexPositions));
// generate permutation table which welds similar vertices
for (PxU32 i = 0; i < mVertexPositions.size(); i++)
mVertexToParticleMap[i] = PX_MAX_U32;
int newNr = 0;
PxReal mWeldingDistanceSq = mWeldingDistance * mWeldingDistance;
for (PxU32 i = 0; i < mVertexPositions.size(); i++)
{
int oldNr = order[i];
if (mVertexToParticleMap[oldNr] != PX_MAX_U32)
continue;
mVertexToParticleMap[oldNr] = newNr;
PxVec3 vi = mVertexPositions[oldNr];
PxU32 j = i+1;
while (j < mVertexPositions.size() && PxAbs(vi.x - mVertexPositions[order[j]].x) <= mWeldingDistance)
{
oldNr = order[j];
if ((vi-mVertexPositions[oldNr]).magnitudeSquared() <= mWeldingDistanceSq)
{
if (mVertexToParticleMap[oldNr] == PX_MAX_U32)
mVertexToParticleMap[oldNr] = newNr;
}
j++;
}
newNr++;
}
mNumWeldedVertices = newNr;
}
else
{
// Welding not enabled. We use an identity permutation.
for (PxU32 i = 0; i < mVertexToParticleMap.size(); i++)
mVertexToParticleMap[i] = i;
#if 0
Ps::Array<int> order;
order.resize(mVertexPositions.size());
for (PxU32 i = 0; i < order.size(); i++)
order[i] = i;
Ps::sort(order.begin(), order.size(), WeldComparator(mVertexPositions));
for (PxU32 i = 0; i < mVertexToParticleMap.size(); i++)
mVertexToParticleMap[i] = order[i];
#endif
mNumWeldedVertices = mVertexPositions.size();
}
mWeldedVertices.resize(mNumWeldedVertices);
for (PxU32 i = 0; i < mVertexPositions.size(); i++)
mWeldedVertices[mVertexToParticleMap[i]] = mVertexPositions[i];
}