GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-05-19 08:07:36 +08:00
Code Issues Packages Projects Releases Wiki Activity

AVX512: implement faster ploadquad<Packet16f> thus speeding up GEMM

Browse Source
This commit is contained in:
Gael Guennebaud 2019-02-21 17:18:28 +01:00
parent 1c09ee8541
commit cca6c207f4
1 changed files with 4 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
Eigen/src/Core/arch/AVX512/PacketMath.h
View File

@ -526,13 +526,11 @@ EIGEN_STRONG_INLINE Packet8d ploaddup<Packet8d>(const double* from) {
// {a0, a0 a0, a0, a1, a1, a1, a1, a2, a2, a2, a2, a3, a3, a3, a3} // {a0, a0 a0, a0, a1, a1, a1, a1, a2, a2, a2, a2, a3, a3, a3, a3}
template <> template <>
EIGEN_STRONG_INLINE Packet16f ploadquad<Packet16f>(const float* from) { EIGEN_STRONG_INLINE Packet16f ploadquad<Packet16f>(const float* from) {
Packet16f tmp = _mm512_undefined_ps(); Packet16f tmp = _mm512_castps128_ps512(pload<Packet4f>(from));
tmp = _mm512_insertf32x4(tmp, _mm_load_ps1(from), 0); const Packet16i scatter_mask = _mm512_set_epi32(3,3,3,3, 2,2,2,2, 1,1,1,1, 0,0,0,0);
tmp = _mm512_insertf32x4(tmp, _mm_load_ps1(from + 1), 1); return _mm512_permutexvar_ps(scatter_mask, tmp);
tmp = _mm512_insertf32x4(tmp, _mm_load_ps1(from + 2), 2);
tmp = _mm512_insertf32x4(tmp, _mm_load_ps1(from + 3), 3);
return tmp;
} }
// Loads 2 doubles from memory a returns the packet // Loads 2 doubles from memory a returns the packet
// {a0, a0 a0, a0, a1, a1, a1, a1} // {a0, a0 a0, a0, a1, a1, a1, a1}
template <> template <>