GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-04-22 17:49:36 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix bug #195: fast unaligned load for integer using _mm_load_sd failed when the value interpreted as a NaN

Browse Source
This commit is contained in:
Benoit Jacob 2011-02-24 10:31:57 -05:00
parent 2064c59878
commit 5dfae4524b
1 changed files with 5 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

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

@ -237,7 +237,6 @@ 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
@ -261,16 +260,13 @@ template<> EIGEN_STRONG_INLINE Packet2d ploadu<Packet2d>(const double* from)
res = _mm_loadh_pd(res,from+1); res = _mm_loadh_pd(res,from+1);
return res; return res;
} }
template<> EIGEN_STRONG_INLINE Packet4i ploadu<Packet4i>(const int* from)
{
EIGEN_DEBUG_UNALIGNED_LOAD
__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
// 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);