GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-05-01 16:24:28 +08:00
Code Issues Packages Projects Releases Wiki Activity

Remove custom unaligned loads for SSE. They were only useful for core2 CPU.

Browse Source
This commit is contained in:
Gael Guennebaud 2016-02-08 14:29:12 +01:00
parent a4c76f8d34
commit c2bf2f56ef
1 changed files with 3 additions and 34 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

37
Eigen/src/Core/arch/SSE/PacketMath.h
View File

@ -314,58 +314,27 @@ template<> EIGEN_STRONG_INLINE Packet4i pload<Packet4i>(const int* from) { E
return _mm_loadu_ps(from);
#endif
}
template<> EIGEN_STRONG_INLINE Packet2d ploadu<Packet2d>(const double* from) { EIGEN_DEBUG_UNALIGNED_LOAD return _mm_loadu_pd(from); }
template<> EIGEN_STRONG_INLINE Packet4i ploadu<Packet4i>(const int* from) { EIGEN_DEBUG_UNALIGNED_LOAD return _mm_loadu_si128(reinterpret_cast<const __m128i*>(from)); }
#else
// NOTE: with the code below, MSVC's compiler crashes!
#if EIGEN_COMP_GNUC && (EIGEN_ARCH_i386 || (EIGEN_ARCH_x86_64 && EIGEN_GNUC_AT_LEAST(4, 8)))
// bug 195: gcc/i386 emits weird x87 fldl/fstpl instructions for _mm_load_sd
#define EIGEN_AVOID_CUSTOM_UNALIGNED_LOADS 1
#elif EIGEN_COMP_CLANG
// bug 201: Segfaults in __mm_loadh_pd with clang 2.8
#define EIGEN_AVOID_CUSTOM_UNALIGNED_LOADS 1
#else
#define EIGEN_AVOID_CUSTOM_UNALIGNED_LOADS 0
#endif
template<> EIGEN_STRONG_INLINE Packet4f ploadu<Packet4f>(const float* from)
{
EIGEN_DEBUG_UNALIGNED_LOAD
#if EIGEN_AVOID_CUSTOM_UNALIGNED_LOADS
return _mm_loadu_ps(from);
#else
__m128d res;
res = _mm_load_sd((const double*)(from)) ;
res = _mm_loadh_pd(res, (const double*)(from+2)) ;
return _mm_castpd_ps(res);
#endif
}
#endif
template<> EIGEN_STRONG_INLINE Packet2d ploadu<Packet2d>(const double* from)
{
EIGEN_DEBUG_UNALIGNED_LOAD
#if EIGEN_AVOID_CUSTOM_UNALIGNED_LOADS
return _mm_loadu_pd(from);
#else
__m128d res;
res = _mm_load_sd(from) ;
res = _mm_loadh_pd(res,from+1);
return res;
#endif
}
template<> EIGEN_STRONG_INLINE Packet4i ploadu<Packet4i>(const int* from)
{
EIGEN_DEBUG_UNALIGNED_LOAD
#if EIGEN_AVOID_CUSTOM_UNALIGNED_LOADS
return _mm_loadu_si128(reinterpret_cast<const __m128i*>(from));
#else
__m128d res;
res = _mm_load_sd((const double*)(from)) ;
res = _mm_loadh_pd(res, (const double*)(from+2)) ;
return _mm_castpd_si128(res);
#endif
}
#endif
template<> EIGEN_STRONG_INLINE Packet4f ploaddup<Packet4f>(const float* from)
{