Misc improvements and cleanups

This commit is contained in:
Benoit Steiner
2014-10-13 17:02:09 -07:00
parent 4c70b0a762
commit 99d75235a9
29 changed files with 779 additions and 140 deletions

View File

@@ -359,7 +359,7 @@ pmadd(const Packet& a,
/** \internal \returns a packet version of \a *from.
* If LoadMode equals #Aligned, \a from must be 16 bytes aligned */
template<typename Packet, int LoadMode>
inline Packet ploadt(const typename unpacket_traits<Packet>::type* from)
EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Packet ploadt(const typename unpacket_traits<Packet>::type* from)
{
if(LoadMode == Aligned)
return pload<Packet>(from);
@@ -370,7 +370,7 @@ inline Packet ploadt(const typename unpacket_traits<Packet>::type* from)
/** \internal copy the packet \a from to \a *to.
* If StoreMode equals #Aligned, \a to must be 16 bytes aligned */
template<typename Scalar, typename Packet, int LoadMode>
inline void pstoret(Scalar* to, const Packet& from)
EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE void pstoret(Scalar* to, const Packet& from)
{
if(LoadMode == Aligned)
pstore(to, from);
@@ -378,6 +378,17 @@ inline void pstoret(Scalar* to, const Packet& from)
pstoreu(to, from);
}
/** \internal \returns a packet version of \a *from.
* Unlike ploadt, ploadt_ro takes advantage of the read-only memory path on the
* hardware if available to speedup the loading of data that won't be modified
* by the current computation.
*/
template<typename Packet, int LoadMode>
inline Packet ploadt_ro(const typename unpacket_traits<Packet>::type* from)
{
return ploadt<Packet, LoadMode>(from);
}
/** \internal default implementation of palign() allowing partial specialization */
template<int Offset,typename PacketType>
struct palign_impl