devtools/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2026-04-10 11:34:33 +08:00
Code Issues Packages Projects Releases Wiki Activity

Do not use std::vector in getResourceRequirements

Browse Source
This commit is contained in:
Eugene Zhulenev
2019-12-09 16:19:38 -08:00
parent 8056a05b54
commit 2918f85ba9
17 changed files with 153 additions and 151 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
unsupported/Eigen/CXX11/src/Tensor/TensorBroadcasting.h
View File

@@ -616,17 +616,16 @@ struct TensorEvaluator<const TensorBroadcastingOp<Broadcast, ArgType>, Device>
TensorOpCost(0, 0, compute_cost, vectorized, PacketSize);
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void getResourceRequirements(
std::vector<internal::TensorOpResourceRequirements>* resources) const {
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
internal::TensorBlockV2ResourceRequirements getResourceRequirements() const {
// TODO(wuke): Targeting L1 size is 30% faster than targeting L{-1} on large
// tensors. But this might need further tuning.
Eigen::Index block_total_size_max = numext::maxi<Eigen::Index>(
const size_t target_block_size = numext::maxi<size_t>(
1, m_device.firstLevelCacheSize() / sizeof(Scalar));
resources->push_back(internal::TensorOpResourceRequirements(
internal::kSkewedInnerDims, block_total_size_max));
m_impl.getResourceRequirements(resources);
return internal::TensorBlockV2ResourceRequirements::merge(
{internal::TensorBlockV2ShapeType::kSkewedInnerDims, target_block_size},
m_impl.getResourceRequirements());
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorBlockV2