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

Special function implementations for half/bfloat16 packets.

Browse Source 
Current implementations fail to consider half-float packets, only
half-float scalars.  Added specializations for packets on AVX, AVX512 and
NEON.  Added tests to `special_packetmath`.

The current `special_functions` tests would fail for half and bfloat16 due to
lack of precision. The NEON tests also fail with precision issues and
due to different handling of `sqrt(inf)`, so special functions bessel, ndtri
have been disabled.

Tested with AVX, AVX512.
This commit is contained in:
Antonio Sanchez
2020-12-02 14:00:57 -08:00
parent 305b8bd277
commit e2f21465fe
17 changed files with 418 additions and 159 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
unsupported/Eigen/src/SpecialFunctions/SpecialFunctionsImpl.h
View File

@@ -348,7 +348,7 @@ EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T generic_fast_erf_float(const T& a_x) {
template <typename T>
struct erf_impl {
EIGEN_DEVICE_FUNC
static EIGEN_STRONG_INLINE T run(const T x) {
static EIGEN_STRONG_INLINE T run(const T& x) {
return generic_fast_erf_float(x);
}
};
@@ -490,7 +490,8 @@ struct erfc_impl<double> {
template<typename T>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T flipsign(
const T& should_flipsign, const T& x) {
const T sign_mask = pset1<T>(-0.0);
typedef typename unpacket_traits<T>::type Scalar;
const T sign_mask = pset1<T>(Scalar(-0.0));
T sign_bit = pand<T>(should_flipsign, sign_mask);
return pxor<T>(sign_bit, x);
}