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

Optimize ploaddup for AVX

Browse Source
This commit is contained in:
Gael Guennebaud 2014-04-17 16:02:27 +02:00
parent d936ddc3d1
commit 0fa8290366
1 changed files with 9 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

19
Eigen/src/Core/arch/AVX/PacketMath.h
View File

@ -182,20 +182,19 @@ template<> EIGEN_STRONG_INLINE Packet8i ploadu<Packet8i>(const int* from) { EIGE
// Loads 4 floats from memory a returns the packet {a0, a0 a1, a1, a2, a2, a3, a3} // Loads 4 floats from memory a returns the packet {a0, a0 a1, a1, a2, a2, a3, a3}
template<> EIGEN_STRONG_INLINE Packet8f ploaddup<Packet8f>(const float* from) template<> EIGEN_STRONG_INLINE Packet8f ploaddup<Packet8f>(const float* from)
{ {
// FIXME we should only load the first 128bits // TODO try to find a way to avoid the need of a temporary register
Packet8f tmp = ploadu<Packet8f>(from); Packet8f tmp = _mm256_castps128_ps256(_mm_loadu_ps(from));
Packet8f tmp1 = _mm256_permute_ps(tmp, _MM_SHUFFLE(3,3,2,2)); tmp = _mm256_insertf128_ps(tmp, _mm_movehl_ps(_mm256_castps256_ps128(tmp),_mm256_castps256_ps128(tmp)), 1);
Packet8f tmp2 = _mm256_permute_ps(tmp, _MM_SHUFFLE(1,1,0,0)); return _mm256_unpacklo_ps(tmp,tmp);
return _mm256_blend_ps(_mm256_permute2f128_ps(tmp1,tmp1,1),tmp2,15);
} }
// Loads 2 doubles from memory a returns the packet {a0, a0 a1, a1} // Loads 2 doubles from memory a returns the packet {a0, a0 a1, a1}
template<> EIGEN_STRONG_INLINE Packet4d ploaddup<Packet4d>(const double* from) template<> EIGEN_STRONG_INLINE Packet4d ploaddup<Packet4d>(const double* from)
{ {
// FIXME we should only load the first 128bits // TODO try to find a way to avoid the need of a temporary register
Packet4d tmp = ploadu<Packet4d>(from); Packet2d tmp0 = _mm_loadu_pd(from);
Packet4d tmp1 = _mm256_permute_pd(tmp,0); Packet2d tmp1 = _mm_permute_pd(tmp0,3);
Packet4d tmp2 = _mm256_permute_pd(tmp,3); tmp0 = _mm_permute_pd(tmp0,0);
return _mm256_blend_pd(tmp1,_mm256_permute2f128_pd(tmp2,tmp2,1),12); return _mm256_insertf128_pd(_mm256_castpd128_pd256(tmp0), tmp1, 1);
} }
// Loads 2 floats from memory a returns the packet {a0, a0 a0, a0, a1, a1, a1, a1} // Loads 2 floats from memory a returns the packet {a0, a0 a0, a0, a1, a1, a1, a1}