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

Remove inline assembly for FMA (AVX) and add remaining extensions as packet ops: pmsub, pnmadd, and pnmsub.

Browse Source
This commit is contained in:
Rasmus Munk Larsen
2022-01-26 04:25:41 +00:00
parent 4e629b3c1b
commit 51311ec651
5 changed files with 152 additions and 31 deletions
Download Patch File Download Diff File Expand all files Collapse all files

36
Eigen/src/Core/GenericPacketMath.h
View File

@@ -939,6 +939,35 @@ template<typename Packet> EIGEN_DEVICE_FUNC inline bool predux_any(const Packet&
* The following functions might not have to be overwritten for vectorized types
***************************************************************************/
// FMA instructions.
/** \internal \returns a * b + c (coeff-wise) */
template <typename Packet>
EIGEN_DEVICE_FUNC inline Packet pmadd(const Packet& a, const Packet& b,
const Packet& c) {
return padd(pmul(a, b), c);
}
/** \internal \returns a * b - c (coeff-wise) */
template <typename Packet>
EIGEN_DEVICE_FUNC inline Packet pmsub(const Packet& a, const Packet& b,
const Packet& c) {
return psub(pmul(a, b), c);
}
/** \internal \returns -(a * b) + c (coeff-wise) */
template <typename Packet>
EIGEN_DEVICE_FUNC inline Packet pnmadd(const Packet& a, const Packet& b,
const Packet& c) {
return padd(pnegate(pmul(a, b)), c);
}
/** \internal \returns -(a * b) - c (coeff-wise) */
template <typename Packet>
EIGEN_DEVICE_FUNC inline Packet pnmsub(const Packet& a, const Packet& b,
const Packet& c) {
return psub(pnegate(pmul(a, b)), c);
}
/** \internal copy a packet with constant coefficient \a a (e.g., [a,a,a,a]) to \a *to. \a to must be 16 bytes aligned */
// NOTE: this function must really be templated on the packet type (think about different packet types for the same scalar type)
template<typename Packet>
@@ -947,13 +976,6 @@ inline void pstore1(typename unpacket_traits<Packet>::type* to, const typename u
pstore(to, pset1<Packet>(a));
}
/** \internal \returns a * b + c (coeff-wise) */
template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
pmadd(const Packet& a,
const Packet& b,
const Packet& c)
{ return padd(pmul(a, b),c); }
/** \internal \returns a packet version of \a *from.
* The pointer \a from must be aligned on a \a Alignment bytes boundary. */
template<typename Packet, int Alignment>