This commit is contained in:
Gael Guennebaud
2010-07-10 22:58:51 +02:00
54 changed files with 1191 additions and 99 deletions

View File

@@ -275,21 +275,26 @@ template<> EIGEN_STRONG_INLINE Packet1cd ei_por <Packet1cd>(const Packet1cd&
template<> EIGEN_STRONG_INLINE Packet1cd ei_pxor <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(_mm_xor_pd(a.v,b.v)); }
template<> EIGEN_STRONG_INLINE Packet1cd ei_pandnot<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(_mm_andnot_pd(a.v,b.v)); }
template<> EIGEN_STRONG_INLINE Packet1cd ei_pload <std::complex<double> >(const std::complex<double>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet1cd(_mm_load_pd((const double*)from)); }
// FIXME force unaligned load, this is a temporary fix
template<> EIGEN_STRONG_INLINE Packet1cd ei_pload <std::complex<double> >(const std::complex<double>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet1cd(ei_ploadu((const double*)from)); }
template<> EIGEN_STRONG_INLINE Packet1cd ei_ploadu<std::complex<double> >(const std::complex<double>* from) { EIGEN_DEBUG_UNALIGNED_LOAD return Packet1cd(ei_ploadu((const double*)from)); }
template<> EIGEN_STRONG_INLINE Packet1cd ei_pset1<std::complex<double> >(const std::complex<double>& from)
{ /* here we really have to use unaligned loads :( */ return ei_ploadu(&from); }
template<> EIGEN_STRONG_INLINE void ei_pstore <std::complex<double> >(std::complex<double> * to, const Packet1cd& from) { EIGEN_DEBUG_ALIGNED_STORE _mm_store_pd((double*)to, from.v); }
// FIXME force unaligned store, this is a temporary fix
template<> EIGEN_STRONG_INLINE void ei_pstore <std::complex<double> >(std::complex<double> * to, const Packet1cd& from) { EIGEN_DEBUG_ALIGNED_STORE ei_pstoreu((double*)to, from.v); }
template<> EIGEN_STRONG_INLINE void ei_pstoreu<std::complex<double> >(std::complex<double> * to, const Packet1cd& from) { EIGEN_DEBUG_UNALIGNED_STORE ei_pstoreu((double*)to, from.v); }
template<> EIGEN_STRONG_INLINE void ei_prefetch<std::complex<double> >(const std::complex<double> * addr) { _mm_prefetch((const char*)(addr), _MM_HINT_T0); }
template<> EIGEN_STRONG_INLINE std::complex<double> ei_pfirst<Packet1cd>(const Packet1cd& a)
{
EIGEN_ALIGN16 double res[2];
_mm_store_pd(res, a.v);
return *(std::complex<double>*)res;
// EIGEN_ALIGN16 double res[2];
// _mm_store_pd(res, a.v);
// return *(std::complex<double>*)res;
EIGEN_ALIGN16 std::complex<double> res;
ei_pstore(&res, a);
return res;
}
template<> EIGEN_STRONG_INLINE Packet1cd ei_preverse(const Packet1cd& a) { return a; }

View File

@@ -112,8 +112,12 @@ template<> EIGEN_STRONG_INLINE Packet4f ei_pset1<float>(const float& from) {
return ei_vec4f_swizzle1(res,0,0,0,0);
}
template<> EIGEN_STRONG_INLINE Packet2d ei_pset1<double>(const double& from) {
#ifdef EIGEN_VECTORIZE_SSE3
return _mm_loaddup_pd(&from);
#else
Packet2d res = _mm_set_sd(from);
return ei_vec2d_swizzle1(res, 0, 0);
#endif
}
#else
template<> EIGEN_STRONG_INLINE Packet4f ei_pset1<float>(const float& from) { return _mm_set1_ps(from); }
@@ -267,13 +271,13 @@ template<> EIGEN_STRONG_INLINE void ei_prefetch<float>(const float* addr) { _m
template<> EIGEN_STRONG_INLINE void ei_prefetch<double>(const double* addr) { _mm_prefetch((const char*)(addr), _MM_HINT_T0); }
template<> EIGEN_STRONG_INLINE void ei_prefetch<int>(const int* addr) { _mm_prefetch((const char*)(addr), _MM_HINT_T0); }
#if defined(_MSC_VER) && (_MSC_VER <= 1500) && defined(_WIN64)
#if defined(_MSC_VER) && (_MSC_VER <= 1500) && defined(_WIN64) && !defined(__INTEL_COMPILER)
// The temporary variable fixes an internal compilation error.
// Direct of the struct members fixed bug #62.
template<> EIGEN_STRONG_INLINE float ei_pfirst<Packet4f>(const Packet4f& a) { return a.m128_f32[0]; }
template<> EIGEN_STRONG_INLINE double ei_pfirst<Packet2d>(const Packet2d& a) { return a.m128d_f64[0]; }
template<> EIGEN_STRONG_INLINE int ei_pfirst<Packet4i>(const Packet4i& a) { int x = _mm_cvtsi128_si32(a); return x; }
#elif defined(_MSC_VER) && (_MSC_VER <= 1500)
#elif defined(_MSC_VER) && (_MSC_VER <= 1500) && !defined(__INTEL_COMPILER)
// The temporary variable fixes an internal compilation error.
template<> EIGEN_STRONG_INLINE float ei_pfirst<Packet4f>(const Packet4f& a) { float x = _mm_cvtss_f32(a); return x; }
template<> EIGEN_STRONG_INLINE double ei_pfirst<Packet2d>(const Packet2d& a) { double x = _mm_cvtsd_f64(a); return x; }