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

Various fixes for packet ops.

Browse Source 
1. Fix buggy pcmp_eq and unit test for half types.
2. Add unit test for pselect and add specializations for SSE 4.1, AVX512, and half types.
3. Get rid of FIXME: Implement faster pnegate for half by XOR'ing with a sign bit mask.
This commit is contained in:
Rasmus Munk Larsen
2019-06-20 11:47:49 -07:00
parent e0be7f30e1
commit 988f24b730
4 changed files with 89 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
Eigen/src/Core/arch/SSE/PacketMath.h
View File

@@ -273,6 +273,12 @@ template<> EIGEN_STRONG_INLINE Packet4f pmadd(const Packet4f& a, const Packet4f&
template<> EIGEN_STRONG_INLINE Packet2d pmadd(const Packet2d& a, const Packet2d& b, const Packet2d& c) { return _mm_fmadd_pd(a,b,c); }
#endif
#ifdef EIGEN_VECTORIZE_SSE4_1
template<> EIGEN_DEVICE_FUNC inline Packet4f pselect(const Packet4f& mask, const Packet4f& a, const Packet4f& b) { return _mm_blendv_ps(b,a,mask); }
template<> EIGEN_DEVICE_FUNC inline Packet2d pselect(const Packet2d& mask, const Packet2d& a, const Packet2d& b) { return _mm_blendv_pd(b,a,mask); }
#endif
template<> EIGEN_STRONG_INLINE Packet4f pmin<Packet4f>(const Packet4f& a, const Packet4f& b) {
#if EIGEN_COMP_GNUC && EIGEN_COMP_GNUC < 63
// There appears to be a bug in GCC, by which the optimizer may