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

Revert "Add AVX512 optimizations for matrix multiply"

Browse Source 
This reverts commit 25db0b4a82
This commit is contained in:
Antonio Sánchez
2022-05-13 18:50:33 +00:00
parent 25db0b4a82
commit 9b9496ad98
17 changed files with 142 additions and 1251 deletions
Download Patch File Download Diff File Expand all files Collapse all files

25
Eigen/src/Core/arch/SSE/PacketMath.h
View File

@@ -285,10 +285,6 @@ template<> EIGEN_STRONG_INLINE Packet4i padd<Packet4i>(const Packet4i& a, const
template<> EIGEN_STRONG_INLINE Packet16b padd<Packet16b>(const Packet16b& a, const Packet16b& b) { return _mm_or_si128(a,b); }
template<typename Packet> EIGEN_STRONG_INLINE Packet padds(const Packet& a, const Packet& b);
template<> EIGEN_STRONG_INLINE Packet4f padds<Packet4f>(const Packet4f& a, const Packet4f& b) { return _mm_add_ss(a,b); }
template<> EIGEN_STRONG_INLINE Packet2d padds<Packet2d>(const Packet2d& a, const Packet2d& b) { return _mm_add_sd(a,b); }
template<> EIGEN_STRONG_INLINE Packet4f psub<Packet4f>(const Packet4f& a, const Packet4f& b) { return _mm_sub_ps(a,b); }
template<> EIGEN_STRONG_INLINE Packet2d psub<Packet2d>(const Packet2d& a, const Packet2d& b) { return _mm_sub_pd(a,b); }
template<> EIGEN_STRONG_INLINE Packet4i psub<Packet4i>(const Packet4i& a, const Packet4i& b) { return _mm_sub_epi32(a,b); }
@@ -374,10 +370,6 @@ template<> EIGEN_STRONG_INLINE Packet4f pnmadd(const Packet4f& a, const Packet4f
template<> EIGEN_STRONG_INLINE Packet2d pnmadd(const Packet2d& a, const Packet2d& b, const Packet2d& c) { return _mm_fnmadd_pd(a,b,c); }
template<> EIGEN_STRONG_INLINE Packet4f pnmsub(const Packet4f& a, const Packet4f& b, const Packet4f& c) { return _mm_fnmsub_ps(a,b,c); }
template<> EIGEN_STRONG_INLINE Packet2d pnmsub(const Packet2d& a, const Packet2d& b, const Packet2d& c) { return _mm_fnmsub_pd(a,b,c); }
template<typename Packet> EIGEN_STRONG_INLINE Packet pmadds(const Packet& a, const Packet& b, const Packet& c);
template<> EIGEN_STRONG_INLINE Packet4f pmadds<Packet4f>(const Packet4f& a, const Packet4f& b, const Packet4f& c) { return _mm_fmadd_ss(a,b,c); }
template<> EIGEN_STRONG_INLINE Packet2d pmadds<Packet2d>(const Packet2d& a, const Packet2d& b, const Packet2d& c) { return _mm_fmadd_sd(a,b,c); }
#endif
#ifdef EIGEN_VECTORIZE_SSE4_1
@@ -754,15 +746,6 @@ template<> EIGEN_STRONG_INLINE Packet16b ploadu<Packet16b>(const bool* from)
return _mm_loadu_si128(reinterpret_cast<const __m128i*>(from));
}
// Load lower part of packet zero extending.
template<typename Packet> EIGEN_STRONG_INLINE Packet ploadl(const typename unpacket_traits<Packet>::type* from);
template<> EIGEN_STRONG_INLINE Packet4f ploadl<Packet4f>(const float* from) { EIGEN_DEBUG_UNALIGNED_LOAD return _mm_castpd_ps(_mm_load_sd(reinterpret_cast<const double*>(from))); }
template<> EIGEN_STRONG_INLINE Packet2d ploadl<Packet2d>(const double* from) { EIGEN_DEBUG_UNALIGNED_LOAD return _mm_load_sd(from); }
// Load scalar
template<typename Packet> EIGEN_STRONG_INLINE Packet ploads(const typename unpacket_traits<Packet>::type* from);
template<> EIGEN_STRONG_INLINE Packet4f ploads<Packet4f>(const float* from) { EIGEN_DEBUG_UNALIGNED_LOAD return _mm_load_ss(from); }
template<> EIGEN_STRONG_INLINE Packet2d ploads<Packet2d>(const double* from) { EIGEN_DEBUG_UNALIGNED_LOAD return _mm_load_sd(from); }
template<> EIGEN_STRONG_INLINE Packet4f ploaddup<Packet4f>(const float* from)
{
@@ -804,14 +787,6 @@ template<> EIGEN_STRONG_INLINE void pstoreu<float>(float* to, const Packet4f&
template<> EIGEN_STRONG_INLINE void pstoreu<int>(int* to, const Packet4i& from) { EIGEN_DEBUG_UNALIGNED_STORE _mm_storeu_si128(reinterpret_cast<__m128i*>(to), from); }
template<> EIGEN_STRONG_INLINE void pstoreu<bool>(bool* to, const Packet16b& from) { EIGEN_DEBUG_ALIGNED_STORE _mm_storeu_si128(reinterpret_cast<__m128i*>(to), from); }
template<typename Scalar, typename Packet> EIGEN_STRONG_INLINE void pstorel(Scalar* to, const Packet& from);
template<> EIGEN_STRONG_INLINE void pstorel(float* to, const Packet4f& from) { EIGEN_DEBUG_UNALIGNED_STORE _mm_storel_pi(reinterpret_cast<__m64*>(to), from); }
template<> EIGEN_STRONG_INLINE void pstorel(double* to, const Packet2d& from) { EIGEN_DEBUG_UNALIGNED_STORE _mm_storel_pd(to, from); }
template<typename Scalar, typename Packet> EIGEN_STRONG_INLINE void pstores(Scalar* to, const Packet& from);
template<> EIGEN_STRONG_INLINE void pstores(float* to, const Packet4f& from) { EIGEN_DEBUG_UNALIGNED_STORE _mm_store_ss(to, from); }
template<> EIGEN_STRONG_INLINE void pstores(double* to, const Packet2d& from) { EIGEN_DEBUG_UNALIGNED_STORE _mm_store_sd(to, from); }
template<> EIGEN_DEVICE_FUNC inline Packet4f pgather<float, Packet4f>(const float* from, Index stride)
{
return _mm_set_ps(from[3*stride], from[2*stride], from[1*stride], from[0*stride]);