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

Improve performance for Power10 MMA bfloat16 GEMM

Browse Source
This commit is contained in:
Chip Kerchner
2023-01-06 23:08:37 +00:00
committed by Rasmus Munk Larsen
parent fe7f527787
commit d20fe21ae4
3 changed files with 536 additions and 228 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
Eigen/src/Core/util/BlasUtil.h
View File

@@ -380,11 +380,6 @@ public:
EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE void store(const blas_data_mapper<Scalar, Index, StorageOrder, AlignmentType, Incr>* sup, Index i, Index j, const PacketBlock<SubPacket, n>& block) const {
spbh.store(sup, i,j,block);
sup->template storePacket<SubPacket>(i, j+idx, block.packet[idx]);
//for(int l = 0; l < unpacket_traits<SubPacket>::size; l++)
//{
// Scalar_ *v = &sup->operator()(i+l, j+idx);
// *v = *reinterpret_cast<Scalar_ *>(&block.packet[idx][l]);
//}
}
};