mirror/eigen
2
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-01-30 17:40:05 +08:00
Code Issues Packages Projects Releases Wiki Activity

bug #195 - fix this once and for all: just never use _mm_load_sd on gcc/i386, it generates redundant x87 ops

Browse Source
This commit is contained in:
Benoit Jacob 2011-02-27 17:26:59 -05:00
parent a8f5ef9388
commit b3544ce2ae
1 changed files with 24 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

29
Eigen/src/Core/arch/SSE/PacketMath.h
View File

@ -237,6 +237,7 @@ template<> EIGEN_STRONG_INLINE Packet4i pload<Packet4i>(const int* from) { E
#endif #endif
} }
template<> EIGEN_STRONG_INLINE Packet2d ploadu<Packet2d>(const double* from) { EIGEN_DEBUG_UNALIGNED_LOAD return _mm_loadu_pd(from); } 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 Packet4i*>(from)); }
#else #else
// Fast unaligned loads. Note that here we cannot directly use intrinsics: this would // Fast unaligned loads. Note that here we cannot directly use intrinsics: this would
// require pointer casting to incompatible pointer types and leads to invalid code // require pointer casting to incompatible pointer types and leads to invalid code
@ -247,26 +248,44 @@ template<> EIGEN_STRONG_INLINE Packet4i pload<Packet4i>(const int* from) { E
template<> EIGEN_STRONG_INLINE Packet4f ploadu<Packet4f>(const float* from) template<> EIGEN_STRONG_INLINE Packet4f ploadu<Packet4f>(const float* from)
{ {
EIGEN_DEBUG_UNALIGNED_LOAD EIGEN_DEBUG_UNALIGNED_LOAD
#if defined(__GNUC__) && defined(__i386__)
// bug 195: gcc/i386 emits weird x87 fldl/fstpl instructions for _mm_load_sd
return _mm_loadu_ps(from);
#else
__m128d res; __m128d res;
res = _mm_load_sd((const double*)(from)) ; res = _mm_load_sd((const double*)(from)) ;
res = _mm_loadh_pd(res, (const double*)(from+2)) ; res = _mm_loadh_pd(res, (const double*)(from+2)) ;
return _mm_castpd_ps(res); return _mm_castpd_ps(res);
#endif
} }
template<> EIGEN_STRONG_INLINE Packet2d ploadu<Packet2d>(const double* from) template<> EIGEN_STRONG_INLINE Packet2d ploadu<Packet2d>(const double* from)
{ {
EIGEN_DEBUG_UNALIGNED_LOAD EIGEN_DEBUG_UNALIGNED_LOAD
#if defined(__GNUC__) && defined(__i386__)
// bug 195: gcc/i386 emits weird x87 fldl/fstpl instructions for _mm_load_sd
return _mm_loadu_pd(from);
#else
__m128d res; __m128d res;
res = _mm_load_sd(from) ; res = _mm_load_sd(from) ;
res = _mm_loadh_pd(res,from+1); res = _mm_loadh_pd(res,from+1);
return res; return res;
#endif
}
template<> EIGEN_STRONG_INLINE Packet4i ploadu<Packet4i>(const int* from)
{
EIGEN_DEBUG_UNALIGNED_LOAD
#if defined(__GNUC__) && defined(__i386__)
// bug 195: gcc/i386 emits weird x87 fldl/fstpl instructions for _mm_load_sd
return _mm_loadu_si128(reinterpret_cast<const Packet4i*>(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 #endif
// bug 195: we used to have an optimized ploadu using _mm_load_sd/_mm_loadh_pd but that gave wrong results when some 64bit value,
// interpreted as double, was a NaN
template<> EIGEN_STRONG_INLINE Packet4i ploadu<Packet4i>(const int* from) { EIGEN_DEBUG_UNALIGNED_LOAD return _mm_loadu_si128(reinterpret_cast<const Packet4i*>(from)); }
template<> EIGEN_STRONG_INLINE Packet4f ploaddup<Packet4f>(const float* from) template<> EIGEN_STRONG_INLINE Packet4f ploaddup<Packet4f>(const float* from)
{ {
return vec4f_swizzle1(_mm_castpd_ps(_mm_load_sd((const double*)from)), 0, 0, 1, 1); return vec4f_swizzle1(_mm_castpd_ps(_mm_load_sd((const double*)from)), 0, 0, 1, 1);