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

Rename predux_downto4 to be more accurate on its semantic.

Browse Source
This commit is contained in:
Gael Guennebaud
2018-04-03 14:28:38 +02:00
parent 67bac6368c
commit 584951ca4d
5 changed files with 10 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
Eigen/src/Core/GenericPacketMath.h
View File

@@ -324,13 +324,13 @@ preduxp(const Packet* vecs) { return vecs[0]; }
template<typename Packet> EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux(const Packet& a) template<typename Packet> EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux(const Packet& a)
{ return a; } { return a; }
/** \internal \returns the sum of the elements of \a a by block of 4 elements. /** \internal \returns the sum of the elements of upper and lower half of \a a if \a a is larger than 4.
* For a packet {a0, a1, a2, a3, a4, a5, a6, a7}, it returns a half packet {a0+a4, a1+a5, a2+a6, a3+a7} * For a packet {a0, a1, a2, a3, a4, a5, a6, a7}, it returns a half packet {a0+a4, a1+a5, a2+a6, a3+a7}
* For packet-size smaller or equal to 4, this boils down to a noop. * For packet-size smaller or equal to 4, this boils down to a noop.
*/ */
template<typename Packet> EIGEN_DEVICE_FUNC inline template<typename Packet> EIGEN_DEVICE_FUNC inline
typename conditional<(unpacket_traits<Packet>::size%8)==0,typename unpacket_traits<Packet>::half,Packet>::type typename conditional<(unpacket_traits<Packet>::size%8)==0,typename unpacket_traits<Packet>::half,Packet>::type
predux_downto4(const Packet& a) predux_half_dowto4(const Packet& a)
{ return a; } { return a; }
/** \internal \returns the product of the elements of \a a*/ /** \internal \returns the product of the elements of \a a*/

2
Eigen/src/Core/arch/AVX/PacketMath.h
View File

@@ -412,7 +412,7 @@ template<> EIGEN_STRONG_INLINE double predux<Packet4d>(const Packet4d& a)
return predux(Packet2d(_mm_add_pd(_mm256_castpd256_pd128(a),_mm256_extractf128_pd(a,1)))); return predux(Packet2d(_mm_add_pd(_mm256_castpd256_pd128(a),_mm256_extractf128_pd(a,1))));
} }
template<> EIGEN_STRONG_INLINE Packet4f predux_downto4<Packet8f>(const Packet8f& a) template<> EIGEN_STRONG_INLINE Packet4f predux_half_dowto4<Packet8f>(const Packet8f& a)
{ {
return _mm_add_ps(_mm256_castps256_ps128(a),_mm256_extractf128_ps(a,1)); return _mm_add_ps(_mm256_castps256_ps128(a),_mm256_extractf128_ps(a,1));
} }

4
Eigen/src/Core/arch/AVX512/PacketMath.h
View File

@@ -888,7 +888,7 @@ EIGEN_STRONG_INLINE double predux<Packet8d>(const Packet8d& a) {
} }
template <> template <>
EIGEN_STRONG_INLINE Packet8f predux_downto4<Packet16f>(const Packet16f& a) { EIGEN_STRONG_INLINE Packet8f predux_half_dowto4<Packet16f>(const Packet16f& a) {
#ifdef EIGEN_VECTORIZE_AVX512DQ #ifdef EIGEN_VECTORIZE_AVX512DQ
__m256 lane0 = _mm512_extractf32x8_ps(a, 0); __m256 lane0 = _mm512_extractf32x8_ps(a, 0);
__m256 lane1 = _mm512_extractf32x8_ps(a, 1); __m256 lane1 = _mm512_extractf32x8_ps(a, 1);
@@ -904,7 +904,7 @@ EIGEN_STRONG_INLINE Packet8f predux_downto4<Packet16f>(const Packet16f& a) {
#endif #endif
} }
template <> template <>
EIGEN_STRONG_INLINE Packet4d predux_downto4<Packet8d>(const Packet8d& a) { EIGEN_STRONG_INLINE Packet4d predux_half_dowto4<Packet8d>(const Packet8d& a) {
__m256d lane0 = _mm512_extractf64x4_pd(a, 0); __m256d lane0 = _mm512_extractf64x4_pd(a, 0);
__m256d lane1 = _mm512_extractf64x4_pd(a, 1); __m256d lane1 = _mm512_extractf64x4_pd(a, 1);
__m256d res = _mm256_add_pd(lane0, lane1); __m256d res = _mm256_add_pd(lane0, lane1);

6
Eigen/src/Core/products/GeneralBlockPanelKernel.h
View File

@@ -580,7 +580,7 @@ DoublePacket<Packet> padd(const DoublePacket<Packet> &a, const DoublePacket<Pack
} }
template<typename Packet> template<typename Packet>
const DoublePacket<Packet>& predux_downto4(const DoublePacket<Packet> &a) const DoublePacket<Packet>& predux_half_dowto4(const DoublePacket<Packet> &a)
{ {
return a; return a;
} }
@@ -1596,13 +1596,13 @@ void gebp_kernel<LhsScalar,RhsScalar,Index,DataMapper,mr,nr,ConjugateLhs,Conjuga
SRhsPacketHalf b0; SRhsPacketHalf b0;
straits.loadLhsUnaligned(blB, a0); straits.loadLhsUnaligned(blB, a0);
straits.loadRhs(blA, b0); straits.loadRhs(blA, b0);
SAccPacketHalf c0 = predux_downto4(C0); SAccPacketHalf c0 = predux_half_dowto4(C0);
straits.madd(a0,b0,c0,b0); straits.madd(a0,b0,c0,b0);
straits.acc(c0, alphav, R); straits.acc(c0, alphav, R);
} }
else else
{ {
straits.acc(predux_downto4(C0), alphav, R); straits.acc(predux_half_dowto4(C0), alphav, R);
} }
res.scatterPacket(i, j2, R); res.scatterPacket(i, j2, R);
} }

4
test/packetmath.cpp
View File

@@ -253,8 +253,8 @@ template<typename Scalar> void packetmath()
ref[i] = 0; ref[i] = 0;
for (int i=0; i<PacketSize; ++i) for (int i=0; i<PacketSize; ++i)
ref[i%HalfPacketSize] += data1[i]; ref[i%HalfPacketSize] += data1[i];
internal::pstore(data2, internal::predux_downto4(internal::pload<Packet>(data1))); internal::pstore(data2, internal::predux_half_dowto4(internal::pload<Packet>(data1)));
VERIFY(areApprox(ref, data2, HalfPacketSize) && "internal::predux_downto4"); VERIFY(areApprox(ref, data2, HalfPacketSize) && "internal::predux_half_dowto4");
} }
ref[0] = 1; ref[0] = 1;