GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-08-12 03:39:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

Merged latest changes from parent.

Browse Source
This commit is contained in:
Benoit Steiner 2014-03-27 18:32:15 -07:00
commit 51e85c936d
1 changed files with 13 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
Eigen/src/Core/arch/AVX/PacketMath.h
View File

@ -124,7 +124,19 @@ template<> EIGEN_STRONG_INLINE Packet8i pdiv<Packet8i>(const Packet8i& /*a*/, co
} }
#ifdef EIGEN_VECTORIZE_FMA #ifdef EIGEN_VECTORIZE_FMA
template<> EIGEN_STRONG_INLINE Packet8f pmadd(const Packet8f& a, const Packet8f& b, const Packet8f& c) { return _mm256_fmadd_ps(a,b,c); } template<> EIGEN_STRONG_INLINE Packet8f pmadd(const Packet8f& a, const Packet8f& b, const Packet8f& c) {
#if defined(__clang__) || defined(__GNUC__)
// clang stupidly generates a vfmadd213ps instruction plus some vmovaps on registers,
// and gcc stupidly generates a vfmadd132ps instruction,
// so let's enforce it to generate a vfmadd231ps instruction since the most common use case is to accumulate
// the result of the product.
Packet8f res = c;
asm("vfmadd231ps %[a], %[b], %[c]" : [c] "+x" (res) : [a] "x" (a), [b] "x" (b));
return res;
#else
return _mm256_fmadd_ps(a,b,c);
#endif
}
template<> EIGEN_STRONG_INLINE Packet4d pmadd(const Packet4d& a, const Packet4d& b, const Packet4d& c) { return _mm256_fmadd_pd(a,b,c); } template<> EIGEN_STRONG_INLINE Packet4d pmadd(const Packet4d& a, const Packet4d& b, const Packet4d& c) { return _mm256_fmadd_pd(a,b,c); }
#endif #endif