mixing types in product step 2:

* pload* and pset1 are now templated on the packet type
* gemv routines are now embeded into a structure with
  a consistent API with respect to gemm
* some configurations of vector * matrix and matrix * matrix works fine,
  some need more work...
This commit is contained in:
Gael Guennebaud
2010-07-11 15:48:30 +02:00
parent 4161b8be67
commit ff96c94043
20 changed files with 600 additions and 549 deletions

View File

@@ -89,15 +89,15 @@ template<> EIGEN_STRONG_INLINE Packet2cf ei_por <Packet2cf>(const Packet2cf&
template<> EIGEN_STRONG_INLINE Packet2cf ei_pxor <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(_mm_xor_ps(a.v,b.v)); }
template<> EIGEN_STRONG_INLINE Packet2cf ei_pandnot<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(_mm_andnot_ps(a.v,b.v)); }
template<> EIGEN_STRONG_INLINE Packet2cf ei_pload <std::complex<float> >(const std::complex<float>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet2cf(ei_pload(&ei_real_ref(*from))); }
template<> EIGEN_STRONG_INLINE Packet2cf ei_ploadu<std::complex<float> >(const std::complex<float>* from) { EIGEN_DEBUG_UNALIGNED_LOAD return Packet2cf(ei_ploadu(&ei_real_ref(*from))); }
template<> EIGEN_STRONG_INLINE Packet2cf ei_pload <Packet2cf>(const std::complex<float>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet2cf(ei_pload<Packet4f>(&ei_real_ref(*from))); }
template<> EIGEN_STRONG_INLINE Packet2cf ei_ploadu<Packet2cf>(const std::complex<float>* from) { EIGEN_DEBUG_UNALIGNED_LOAD return Packet2cf(ei_ploadu<Packet4f>(&ei_real_ref(*from))); }
template<> EIGEN_STRONG_INLINE void ei_pstore <std::complex<float> >(std::complex<float> * to, const Packet2cf& from) { EIGEN_DEBUG_ALIGNED_STORE ei_pstore(&ei_real_ref(*to), from.v); }
template<> EIGEN_STRONG_INLINE void ei_pstoreu<std::complex<float> >(std::complex<float> * to, const Packet2cf& from) { EIGEN_DEBUG_UNALIGNED_STORE ei_pstoreu(&ei_real_ref(*to), from.v); }
template<> EIGEN_STRONG_INLINE void ei_prefetch<std::complex<float> >(const std::complex<float> * addr) { _mm_prefetch((const char*)(addr), _MM_HINT_T0); }
template<> EIGEN_STRONG_INLINE Packet2cf ei_pset1<std::complex<float> >(const std::complex<float>& from)
template<> EIGEN_STRONG_INLINE Packet2cf ei_pset1<Packet2cf>(const std::complex<float>& from)
{
Packet2cf res;
res.v = _mm_loadl_pi(res.v, (const __m64*)&from);
@@ -276,10 +276,12 @@ template<> EIGEN_STRONG_INLINE Packet1cd ei_pxor <Packet1cd>(const Packet1cd&
template<> EIGEN_STRONG_INLINE Packet1cd ei_pandnot<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(_mm_andnot_pd(a.v,b.v)); }
// 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 Packet1cd ei_pload <Packet1cd>(const std::complex<double>* from)
{ EIGEN_DEBUG_ALIGNED_LOAD return Packet1cd(ei_ploadu<Packet2d>((const double*)from)); }
template<> EIGEN_STRONG_INLINE Packet1cd ei_ploadu<Packet1cd>(const std::complex<double>* from)
{ EIGEN_DEBUG_UNALIGNED_LOAD return Packet1cd(ei_ploadu<Packet2d>((const double*)from)); }
template<> EIGEN_STRONG_INLINE Packet1cd ei_pset1<Packet1cd>(const std::complex<double>& from)
{ /* here we really have to use unaligned loads :( */ return ei_ploadu<Packet1cd>(&from); }
// 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); }
@@ -387,6 +389,15 @@ template<> struct ei_conj_helper<Packet2d, Packet1cd, false,false>
{ return Packet1cd(ei_pmul(x, y.v)); }
};
template<> struct ei_conj_helper<Packet1cd, Packet2d, false,false>
{
EIGEN_STRONG_INLINE Packet1cd pmadd(const Packet1cd& x, const Packet2d& y, const Packet1cd& c) const
{ return ei_padd(c, pmul(x,y)); }
EIGEN_STRONG_INLINE Packet1cd pmul(const Packet1cd& x, const Packet2d& y) const
{ return Packet1cd(ei_pmul(x.v, y)); }
};
template<> EIGEN_STRONG_INLINE Packet1cd ei_pdiv<Packet1cd>(const Packet1cd& a, const Packet1cd& b)
{
// TODO optimize it for SSE3 and 4