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

Introduce unified macros to identify compiler, OS, and architecture. They are all defined in util/Macros.h and prefixed with EIGEN_COMP_, EIGEN_OS_, and EIGEN_ARCH_ respectively.

Browse Source
This commit is contained in:
Gael Guennebaud
2014-11-04 21:58:52 +01:00
parent c5a3777666
commit ee06f78679
22 changed files with 357 additions and 115 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
Eigen/src/Core/arch/AVX/PacketMath.h
View File

@@ -135,7 +135,7 @@ template<> EIGEN_STRONG_INLINE Packet8i pdiv<Packet8i>(const Packet8i& /*a*/, co
#ifdef EIGEN_VECTORIZE_FMA
template<> EIGEN_STRONG_INLINE Packet8f pmadd(const Packet8f& a, const Packet8f& b, const Packet8f& c) {
#if defined(__clang__) || defined(__GNUC__)
#if EIGEN_COMP_GNUC || EIGEN_COMP_CLANG
// 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
@@ -148,7 +148,7 @@ template<> EIGEN_STRONG_INLINE Packet8f pmadd(const Packet8f& a, const Packet8f&
#endif
}
template<> EIGEN_STRONG_INLINE Packet4d pmadd(const Packet4d& a, const Packet4d& b, const Packet4d& c) {
#if defined(__clang__) || defined(__GNUC__)
#if EIGEN_COMP_GNUC || EIGEN_COMP_CLANG
// see above
Packet4d res = c;
__asm__("vfmadd231pd %[a], %[b], %[c]" : [c] "+x" (res) : [a] "x" (a), [b] "x" (b));