mirror/eigen
2
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2024-12-15 07:10:37 +08:00
Code Issues Packages Projects Releases Wiki Activity

Reintroduced the if-clause for MSVC ei_ploadu via _loadu_.

Browse Source
This commit is contained in:
Hauke Heibel 2010-03-07 14:05:26 +01:00
parent aeea00a9cf
commit 9fe040ad29
1 changed files with 7 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
Eigen/src/Core/arch/SSE/PacketMath.h
View File

@ -184,17 +184,17 @@ template<> EIGEN_STRONG_INLINE Packet4f ei_pload<float>(const float* from) {
template<> EIGEN_STRONG_INLINE Packet2d ei_pload<double>(const double* from) { EIGEN_DEBUG_ALIGNED_LOAD return _mm_load_pd(from); }
template<> EIGEN_STRONG_INLINE Packet4i ei_pload<int>(const int* from) { EIGEN_DEBUG_ALIGNED_LOAD return _mm_load_si128(reinterpret_cast<const Packet4i*>(from)); }
// #if (!defined __GNUC__) && (!defined __ICC)
// template<> EIGEN_STRONG_INLINE Packet4f ei_ploadu(const float* from) { EIGEN_DEBUG_UNALIGNED_LOAD return _mm_loadu_ps(from); }
// template<> EIGEN_STRONG_INLINE Packet2d ei_ploadu<double>(const double* from) { EIGEN_DEBUG_UNALIGNED_LOAD return _mm_loadu_pd(from); }
// template<> EIGEN_STRONG_INLINE Packet4i ei_ploadu<int>(const int* from) { EIGEN_DEBUG_UNALIGNED_LOAD return _mm_loadu_si128(reinterpret_cast<const Packet4i*>(from)); }
// #else
#if defined(_MSC_VER)
template<> EIGEN_STRONG_INLINE Packet4f ei_ploadu(const float* from) { EIGEN_DEBUG_UNALIGNED_LOAD return _mm_loadu_ps(from); }
template<> EIGEN_STRONG_INLINE Packet2d ei_ploadu<double>(const double* from) { EIGEN_DEBUG_UNALIGNED_LOAD return _mm_loadu_pd(from); }
template<> EIGEN_STRONG_INLINE Packet4i ei_ploadu<int>(const int* from) { EIGEN_DEBUG_UNALIGNED_LOAD return _mm_loadu_si128(reinterpret_cast<const Packet4i*>(from)); }
#else
// 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
// because of the strict aliasing rule. The "dummy" stuff are required to enforce
// a correct instruction dependency.
// TODO: do the same for MSVC (ICC is compatible)
// NOTE: with the code below, MSVC's compiler crashes!
template<> EIGEN_STRONG_INLINE Packet4f ei_ploadu(const float* from)
{
EIGEN_DEBUG_UNALIGNED_LOAD
@ -219,6 +219,7 @@ template<> EIGEN_STRONG_INLINE Packet4i ei_ploadu(const int* from)
res = _mm_loadh_pd(res, (const double*)(from+2)) ;
return _mm_castpd_si128(res);
}
#endif
template<> EIGEN_STRONG_INLINE void ei_pstore<float>(float* to, const Packet4f& from) { EIGEN_DEBUG_ALIGNED_STORE _mm_store_ps(to, from); }
template<> EIGEN_STRONG_INLINE void ei_pstore<double>(double* to, const Packet2d& from) { EIGEN_DEBUG_ALIGNED_STORE _mm_store_pd(to, from); }