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

Collapsed revision

Browse Source 
* Add packet up "pones". Write pnot(a) as pxor(pones(a), a).
* Collapsed revision
* Simplify a bit.
* Undo useless diffs.
* Fix typo.
This commit is contained in:
Rasmus Munk Larsen
2019-01-09 16:34:23 -08:00
parent cb3c059fa4
commit e15bb785ad
9 changed files with 104 additions and 53 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
Eigen/src/Core/arch/GPU/PacketMathHalf.h
View File

@@ -143,6 +143,10 @@ template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE half2 pabs<half2>(const half2&
return result;
}
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE half2 pones<half2>(const half2& a) {
half2 result;
*(reinterpret_cast<unsigned*>(&(result))) = 0xffffffffu;
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void
ptranspose(PacketBlock<half2,2>& kernel) {
@@ -640,6 +644,14 @@ EIGEN_STRONG_INLINE Packet16h float2half(const Packet16f& a) {
#endif
}
template<> EIGEN_STRONG_INLINE Packet16h pnot(const Packet16h& a) {
Packet16h r; r.x = _mm256_xor_si256(a.x, pcmp_eq(a.x, a.x)); return r;
}
template<> EIGEN_STRONG_INLINE Packet16h pones(const Packet16h& a) {
Packet16h r; r.x = Packet8i(pones(a.x)); return r;
}
template<> EIGEN_STRONG_INLINE Packet16h por(const Packet16h& a,const Packet16h& b) {
// in some cases Packet8i is a wrapper around __m256i, so we need to
// cast to Packet8i to call the correct overload.
@@ -1085,6 +1097,10 @@ EIGEN_STRONG_INLINE Packet8h float2half(const Packet8f& a) {
#endif
}
template<> EIGEN_STRONG_INLINE Packet8h pones(const Packet8h& a) {
Packet8h r; r.x = _mm_cmpeq_epi32(a.x, a.x); return r;
}
template<> EIGEN_STRONG_INLINE Packet8h por(const Packet8h& a,const Packet8h& b) {
// in some cases Packet4i is a wrapper around __m128i, so we either need to
// cast to Packet4i to directly call the intrinsics as below: