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

Add support for Power10 (AltiVec) MMA instructions for bfloat16.

Browse Source
This commit is contained in:
Pedro Caldeira
2022-11-30 23:33:37 +00:00
committed by Rasmus Munk Larsen
parent dcb042a87d
commit 31ab62d347
8 changed files with 438 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
Eigen/src/Core/util/BlasUtil.h
View File

@@ -379,11 +379,12 @@ public:
storePacketBlock_helper<SubPacket, Scalar_, n, idx-1> spbh;
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);
for(int l = 0; l < unpacket_traits<SubPacket>::size; l++)
{
Scalar_ *v = &sup->operator()(i+l, j+idx);
*v = block.packet[idx][l];
}
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]);
//}
}
};
@@ -393,12 +394,7 @@ public:
storePacketBlock_helper<SubPacket, std::complex<float>, n, idx-1> spbh;
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);
for(int l = 0; l < unpacket_traits<SubPacket>::size; l++)
{
std::complex<float> *v = &sup->operator()(i+l, j+idx);
v->real(block.packet[idx].v[2*l+0]);
v->imag(block.packet[idx].v[2*l+1]);
}
sup->template storePacket<SubPacket>(i, j+idx, block.packet[idx]);
}
};