devtools/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2026-04-10 11:34:33 +08:00
Code Issues Packages Projects Releases Wiki Activity

Add optional offset parameter to ploadu_partial and pstoreu_partial

Browse Source
This commit is contained in:
Chip Kerchner
2023-06-23 19:53:05 +00:00
committed by Charles Schlosser
parent 44c20bbbe3
commit 211c5dfc67
3 changed files with 85 additions and 65 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
Eigen/src/Core/GenericPacketMath.h
View File

@@ -730,14 +730,14 @@ ploadu(const typename unpacket_traits<Packet>::type* from) { return *from; }
/** \internal \returns n elements of a packet version of \a *from, (un-aligned load)
* All elements after the last element loaded will initialized with zero */
template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
ploadu_partial(const typename unpacket_traits<Packet>::type* from, const Index n)
ploadu_partial(const typename unpacket_traits<Packet>::type* from, const Index n, const Index offset = 0)
{
const Index packet_size = unpacket_traits<Packet>::size;
eigen_assert(n <= packet_size && "number of elements will read past end of packet");
eigen_assert(n + offset <= packet_size && "number of elements plus offset will read past end of packet");
typedef typename unpacket_traits<Packet>::type Scalar;
EIGEN_ALIGN_MAX Scalar elements[packet_size] = { Scalar(0) };
for (Index i = 0; i < numext::mini(n,packet_size); i++) {
elements[i] = from[i];
for (Index i = offset; i < numext::mini(n+offset,packet_size); i++) {
elements[i] = from[i-offset];
}
return pload<Packet>(elements);
}
@@ -855,14 +855,14 @@ template<typename Scalar, typename Packet> EIGEN_DEVICE_FUNC inline void pstoreu
{ (*to) = from; }
/** \internal copy n elements of the packet \a from to \a *to, (un-aligned store) */
template<typename Scalar, typename Packet> EIGEN_DEVICE_FUNC inline void pstoreu_partial(Scalar* to, const Packet& from, const Index n)
template<typename Scalar, typename Packet> EIGEN_DEVICE_FUNC inline void pstoreu_partial(Scalar* to, const Packet& from, const Index n, const Index offset = 0)
{
const Index packet_size = unpacket_traits<Packet>::size;
eigen_assert(n <= packet_size && "number of elements will write past end of packet");
eigen_assert(n + offset <= packet_size && "number of elements plus offset will write past end of packet");
EIGEN_ALIGN_MAX Scalar elements[packet_size];
pstore<Scalar>(elements, from);
for (Index i = 0; i < numext::mini(n,packet_size); i++) {
to[i] = elements[i];
for (Index i = 0; i < numext::mini(n,packet_size-offset); i++) {
to[i] = elements[i + offset];
}
}
@@ -1201,7 +1201,7 @@ EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Packet ploadt_partial(const typename unpac
if(Alignment >= unpacket_traits<Packet>::alignment)
return pload_partial<Packet>(from, n, offset);
else
return ploadu_partial<Packet>(from, n);
return ploadu_partial<Packet>(from, n, offset);
}
/** \internal copy the packet \a from to \a *to.
@@ -1223,7 +1223,7 @@ EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE void pstoret_partial(Scalar* to, const Pac
if(Alignment >= unpacket_traits<Packet>::alignment)
pstore_partial(to, from, n, offset);
else
pstoreu_partial(to, from, n);
pstoreu_partial(to, from, n, offset);
}
/** \internal \returns a packet version of \a *from.