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

Fix pfrexp/pldexp for half.

Browse Source 
The recent addition of vectorized pow (!330) relies on `pfrexp` and
`pldexp`.  This was missing for `Eigen::half` and `Eigen::bfloat16`.
Adding tests for these packet ops also exposed an issue with handling
negative values in `pfrexp`, returning an incorrect exponent.

Added the missing implementations, corrected the exponent in `pfrexp1`,
and added `packetmath` tests.
This commit is contained in:
Antonio Sanchez
2021-01-20 19:00:09 -08:00
committed by Rasmus Munk Larsen
parent 25d8498f8b
commit b2126fd6b5
7 changed files with 112 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
Eigen/src/Core/arch/NEON/MathFunctions.h
View File

@@ -44,6 +44,18 @@ BF16_PACKET_FUNCTION(Packet4f, Packet4bf, plog)
BF16_PACKET_FUNCTION(Packet4f, Packet4bf, pexp)
BF16_PACKET_FUNCTION(Packet4f, Packet4bf, ptanh)
template <>
EIGEN_STRONG_INLINE Packet4bf pfrexp(const Packet4bf& a, Packet4bf& exponent) {
Packet4f fexponent;
const Packet4bf out = F32ToBf16(pfrexp<Packet4f>(Bf16ToF32(a), fexponent));
exponent = F32ToBf16(fexponent);
return out;
}
template <>
EIGEN_STRONG_INLINE Packet4bf pldexp(const Packet4bf& a, const Packet4bf& exponent) {
return F32ToBf16(pldexp<Packet4f>(Bf16ToF32(a), Bf16ToF32(exponent)));
}
//---------- double ----------