GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-08-12 03:39:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

Introduced pscatter/pgather packet primitives. They will be used to optimize the loop peeling code of the block-panel matrix multiplication kernel.

Browse Source
This commit is contained in:
Benoit Steiner 2014-03-27 16:03:03 -07:00
parent 729363114f
commit ee86679096
2 changed files with 43 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
Eigen/src/Core/GenericPacketMath.h
View File

@ -217,6 +217,12 @@ template<typename Scalar, typename Packet> EIGEN_DEVICE_FUNC inline void pstore(
/** \internal copy the packet \a from to \a *to, (un-aligned store) */ /** \internal copy the packet \a from to \a *to, (un-aligned store) */
template<typename Scalar, typename Packet> EIGEN_DEVICE_FUNC inline void pstoreu(Scalar* to, const Packet& from) template<typename Scalar, typename Packet> EIGEN_DEVICE_FUNC inline void pstoreu(Scalar* to, const Packet& from)
{ (*to) = from; }
template<typename Scalar, typename Packet> EIGEN_DEVICE_FUNC inline Packet pgather(const Scalar* from, int /*stride*/)
{ return ploadu<Packet>(from); }
template<typename Scalar, typename Packet> EIGEN_DEVICE_FUNC inline void pscatter(Scalar* to, const Packet& from, int /*stride*/)
{ (*to) = from; } { (*to) = from; }
/** \internal tries to do cache prefetching of \a addr */ /** \internal tries to do cache prefetching of \a addr */

39
test/packetmath.cpp
View File

@ -356,8 +356,38 @@ template<typename Scalar> void packetmath_complex()
internal::pstore(pval,internal::pcplxflip(internal::pload<Packet>(data1))); internal::pstore(pval,internal::pcplxflip(internal::pload<Packet>(data1)));
VERIFY(areApprox(ref, pval, PacketSize) && "pcplxflip"); VERIFY(areApprox(ref, pval, PacketSize) && "pcplxflip");
} }
}
template<typename Scalar> void packetmath_scatter_gather() {
typedef typename internal::packet_traits<Scalar>::type Packet;
typedef typename NumTraits<Scalar>::Real RealScalar;
const int PacketSize = internal::packet_traits<Scalar>::size;
Scalar data1[PacketSize];
RealScalar refvalue = 0;
for (int i=0; i<PacketSize; ++i) {
data1[i] = internal::random<Scalar>()/RealScalar(PacketSize);
}
Scalar buffer[PacketSize*11];
memset(buffer, 0, 11*sizeof(Packet));
Packet packet = internal::pload<Packet>(data1);
internal::pscatter<Scalar, Packet>(buffer, packet, 11);
for (int i = 0; i < PacketSize*11; ++i) {
if ((i%11) == 0) {
VERIFY(isApproxAbs(buffer[i], data1[i/11], refvalue));
} else {
VERIFY(isApproxAbs(buffer[i], Scalar(0), refvalue));
}
}
for (int i=0; i<PacketSize*7; ++i) {
buffer[i] = internal::random<Scalar>()/RealScalar(PacketSize);
}
packet = internal::pgather<Scalar, Packet>(buffer, 7);
internal::pstore(data1, packet);
for (int i = 0; i < PacketSize; ++i) {
VERIFY(isApproxAbs(data1[i], buffer[i*7], refvalue));
}
} }
void test_packetmath() void test_packetmath()
@ -378,5 +408,10 @@ void test_packetmath()
CALL_SUBTEST_1( packetmath_complex<std::complex<float> >() ); CALL_SUBTEST_1( packetmath_complex<std::complex<float> >() );
CALL_SUBTEST_2( packetmath_complex<std::complex<double> >() ); CALL_SUBTEST_2( packetmath_complex<std::complex<double> >() );
CALL_SUBTEST_1( packetmath_scatter_gather<float>() );
CALL_SUBTEST_2( packetmath_scatter_gather<double>() );
CALL_SUBTEST_3( packetmath_scatter_gather<std::complex<float> >() );
CALL_SUBTEST_3( packetmath_scatter_gather<std::complex<double> >() );
} }
} }