GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-07-30 00:32:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

bug #1270: bypass custom asm for pmadd and recent clang version

Browse Source
This commit is contained in:
Gael Guennebaud 2016-08-22 15:38:03 +02:00
parent 49b005181a
commit 2d5731e40a
1 changed files with 2 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

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

@ -153,7 +153,7 @@ template<> EIGEN_STRONG_INLINE Packet8i pdiv<Packet8i>(const Packet8i& /*a*/, co
#ifdef __FMA__ #ifdef __FMA__
template<> EIGEN_STRONG_INLINE Packet8f pmadd(const Packet8f& a, const Packet8f& b, const Packet8f& c) { template<> EIGEN_STRONG_INLINE Packet8f pmadd(const Packet8f& a, const Packet8f& b, const Packet8f& c) {
#if EIGEN_COMP_GNUC || EIGEN_COMP_CLANG #if ( EIGEN_COMP_GNUC_STRICT || (EIGEN_COMP_CLANG && (EIGEN_COMP_CLANG<308)) )
// clang stupidly generates a vfmadd213ps instruction plus some vmovaps on registers, // clang stupidly generates a vfmadd213ps instruction plus some vmovaps on registers,
// and gcc stupidly generates a vfmadd132ps instruction, // 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 // so let's enforce it to generate a vfmadd231ps instruction since the most common use case is to accumulate
@ -166,7 +166,7 @@ template<> EIGEN_STRONG_INLINE Packet8f pmadd(const Packet8f& a, const Packet8f&
#endif #endif
} }
template<> EIGEN_STRONG_INLINE Packet4d pmadd(const Packet4d& a, const Packet4d& b, const Packet4d& c) { template<> EIGEN_STRONG_INLINE Packet4d pmadd(const Packet4d& a, const Packet4d& b, const Packet4d& c) {
#if EIGEN_COMP_GNUC || EIGEN_COMP_CLANG #if ( EIGEN_COMP_GNUC_STRICT || (EIGEN_COMP_CLANG && (EIGEN_COMP_CLANG<308)) )
// see above // see above
Packet4d res = c; Packet4d res = c;
__asm__("vfmadd231pd %[a], %[b], %[c]" : [c] "+x" (res) : [a] "x" (a), [b] "x" (b)); __asm__("vfmadd231pd %[a], %[b], %[c]" : [c] "+x" (res) : [a] "x" (a), [b] "x" (b));