diff options
author | Gael Guennebaud <g.gael@free.fr> | 2014-04-25 10:56:18 +0200 |
---|---|---|
committer | Gael Guennebaud <g.gael@free.fr> | 2014-04-25 10:56:18 +0200 |
commit | 3d8d0f6269478a06f4fcbd4b838c8e9b9d7e9d62 (patch) | |
tree | 9f55e81498c1acc61ea0c10bce0ead69a71875d1 /Eigen/src/Core/arch/SSE | |
parent | b0e19db1cf462a07e25429d4f04f7d8e858f670f (diff) |
Enable vectorization of pack_rhs with a column-major RHS.
Rename and generalize Kernel<*> to PacketBlock<*,N>.
Diffstat (limited to 'Eigen/src/Core/arch/SSE')
-rw-r--r-- | Eigen/src/Core/arch/SSE/Complex.h | 4 | ||||
-rwxr-xr-x | Eigen/src/Core/arch/SSE/PacketMath.h | 12 |
2 files changed, 8 insertions, 8 deletions
diff --git a/Eigen/src/Core/arch/SSE/Complex.h b/Eigen/src/Core/arch/SSE/Complex.h index d0c080c4f..758183c18 100644 --- a/Eigen/src/Core/arch/SSE/Complex.h +++ b/Eigen/src/Core/arch/SSE/Complex.h @@ -462,8 +462,8 @@ EIGEN_STRONG_INLINE Packet1cd pcplxflip/*<Packet1cd>*/(const Packet1cd& x) return Packet1cd(preverse(Packet2d(x.v))); } -template<> EIGEN_DEVICE_FUNC inline void -ptranspose(Kernel<Packet2cf>& kernel) { +EIGEN_DEVICE_FUNC inline void +ptranspose(PacketBlock<Packet2cf,2>& kernel) { __m128d w1 = _mm_castps_pd(kernel.packet[0].v); __m128d w2 = _mm_castps_pd(kernel.packet[1].v); diff --git a/Eigen/src/Core/arch/SSE/PacketMath.h b/Eigen/src/Core/arch/SSE/PacketMath.h index 4f9d8c4fd..ad935d5f1 100755 --- a/Eigen/src/Core/arch/SSE/PacketMath.h +++ b/Eigen/src/Core/arch/SSE/PacketMath.h @@ -784,20 +784,20 @@ struct palign_impl<Offset,Packet2d> }; #endif -template<> EIGEN_DEVICE_FUNC inline void -ptranspose(Kernel<Packet4f>& kernel) { +EIGEN_DEVICE_FUNC inline void +ptranspose(PacketBlock<Packet4f,4>& kernel) { _MM_TRANSPOSE4_PS(kernel.packet[0], kernel.packet[1], kernel.packet[2], kernel.packet[3]); } -template<> EIGEN_DEVICE_FUNC inline void -ptranspose(Kernel<Packet2d>& kernel) { +EIGEN_DEVICE_FUNC inline void +ptranspose(PacketBlock<Packet2d,2>& kernel) { __m128d tmp = _mm_unpackhi_pd(kernel.packet[0], kernel.packet[1]); kernel.packet[0] = _mm_unpacklo_pd(kernel.packet[0], kernel.packet[1]); kernel.packet[1] = tmp; } -template<> EIGEN_DEVICE_FUNC inline void -ptranspose(Kernel<Packet4i>& kernel) { +EIGEN_DEVICE_FUNC inline void +ptranspose(PacketBlock<Packet4i,4>& kernel) { __m128i T0 = _mm_unpacklo_epi32(kernel.packet[0], kernel.packet[1]); __m128i T1 = _mm_unpacklo_epi32(kernel.packet[2], kernel.packet[3]); __m128i T2 = _mm_unpackhi_epi32(kernel.packet[0], kernel.packet[1]); |