Disable the use of tbb::parallel_for in Blender

This commit is contained in:
Brad Davis 2019-05-06 12:22:39 -07:00
parent 2886e94cdc
commit 2b1267ffeb

View file

@ -1741,15 +1741,23 @@ Blender::Blender(ModelPointer model, int blendNumber, const Geometry::WeakPointe
_blendshapeCoefficients(blendshapeCoefficients) {
}
#define BLENDER_USE_NONE 0
#define BLENDER_USE_TBB 1
#define BLENDER_TBB_CHUNK_SIZE 512
#define BLENDER_USE_OPENMP 2
#define BLENDER_PARALLELISM BLENDER_USE_NONE
void Blender::run() {
QVector<BlendshapeOffset> blendshapeOffsets;
QVector<int> blendedMeshSizes;
if (_model && _model->isLoaded()) {
DETAILED_PROFILE_RANGE_EX(simulation_animation, __FUNCTION__, 0xFFFF0000, 0, { { "url", _model->getURL().toString() } });
int offset = 0;
auto meshes = _model->getHFMModel().meshes;
const auto& meshes = _model->getHFMModel().meshes;
int meshIndex = 0;
foreach(const HFMMesh& mesh, meshes) {
for(const HFMMesh& mesh : meshes) {
auto modelMeshBlendshapeOffsets = _model->_blendshapeOffsets.find(meshIndex++);
if (mesh.blendshapes.isEmpty() || modelMeshBlendshapeOffsets == _model->_blendshapeOffsets.end()) {
// Not blendshaped or not initialized
@ -1780,9 +1788,18 @@ void Blender::run() {
float normalCoefficient = vertexCoefficient * NORMAL_COEFFICIENT_SCALE;
const HFMBlendshape& blendshape = mesh.blendshapes.at(i);
tbb::parallel_for(tbb::blocked_range<int>(0, blendshape.indices.size()), [&](const tbb::blocked_range<int>& range) {
#if (BLENDER_PARALLELISM == BLENDER_USE_TBB)
tbb::parallel_for(tbb::blocked_range<int>(0, blendshape.indices.size(), BLENDER_TBB_CHUNK_SIZE), [&](const tbb::blocked_range<int>& range) {
for (auto j = range.begin(); j < range.end(); j++) {
#elif (BLENDER_PARALLELISM == BLENDER_USE_OPENMP)
{
#pragma omp parallel for
for (int j = 0; j < (int)blendshape.indices.size(); ++j) {
#else
{
for (int j = 0; j < blendshape.indices.size(); ++j) {
#endif
int index = blendshape.indices.at(j);
auto& currentBlendshapeOffset = unpackedBlendshapeOffsets[index];
@ -1793,20 +1810,41 @@ void Blender::run() {
currentBlendshapeOffset.tangentOffset += blendshape.tangents.at(j) * normalCoefficient;
}
}
#if (BLENDER_PARALLELISM == BLENDER_USE_TBB)
});
#else
}
#endif
}
// Blendshape offsets are generrated, now let's pack it on its way to gpu
tbb::parallel_for(tbb::blocked_range<int>(0, (int) unpackedBlendshapeOffsets.size()), [&](const tbb::blocked_range<int>& range) {
#if (BLENDER_PARALLELISM == BLENDER_USE_TBB)
tbb::parallel_for(tbb::blocked_range<int>(0, (int) unpackedBlendshapeOffsets.size(), BLENDER_TBB_CHUNK_SIZE), [&](const tbb::blocked_range<int>& range) {
auto unpacked = unpackedBlendshapeOffsets.data() + range.begin();
auto packed = meshBlendshapeOffsets + range.begin();
for (auto j = range.begin(); j < range.end(); j++) {
#elif (BLENDER_PARALLELISM == BLENDER_USE_OPENMP)
{
auto unpacked = unpackedBlendshapeOffsets.data();
auto packed = meshBlendshapeOffsets;
#pragma omp parallel for
for (int j = 0; j < (int)unpackedBlendshapeOffsets.size(); ++j) {
#else
{
auto unpacked = unpackedBlendshapeOffsets.data();
auto packed = meshBlendshapeOffsets;
for (int j = 0; j < (int)unpackedBlendshapeOffsets.size(); ++j) {
#endif
//for (auto j = range.begin(); j < range.end(); j++) {
packBlendshapeOffsetTo_Pos_F32_3xSN10_Nor_3xSN10_Tan_3xSN10((*packed).packedPosNorTan, (*unpacked));
unpacked++;
packed++;
++unpacked;
++packed;
}
#if (BLENDER_PARALLELISM == BLENDER_USE_TBB)
});
#else
}
#endif
}
}
// post the result to the ModelBlender, which will dispatch to the model if still alive