diff options
| author |  Benoit Steiner <benoit.steiner.goog@gmail.com> | 2014-04-17 12:28:01 -0700 |
|---|---|---|
| committer | Benoit Steiner <benoit.steiner.goog@gmail.com> | 2014-04-17 12:28:01 -0700 |
| commit | 6d6df90c9a813287d18fe045b78a7f4e3996ee5f (patch) | |
| tree | c29c47aac3d9c1325aebca966df9ff4a6a59d88d /Eigen/src/Core/arch/NEON/Complex.h | |
| parent | d936ddc3d129162f096d8eb39084ea1a47ff6f29 (diff) | |
Implemented the pgather/pscatter packet primitives for the arm/NEON architecture
Diffstat (limited to 'Eigen/src/Core/arch/NEON/Complex.h')
| -rw-r--r-- | Eigen/src/Core/arch/NEON/Complex.h | 16 |
1 files changed, 16 insertions, 0 deletions
diff --git a/Eigen/src/Core/arch/NEON/Complex.h b/Eigen/src/Core/arch/NEON/Complex.h index 7ca76714f..7b94733ab 100644 --- a/Eigen/src/Core/arch/NEON/Complex.h +++ b/Eigen/src/Core/arch/NEON/Complex.h @@ -111,6 +111,22 @@ template<> EIGEN_STRONG_INLINE Packet2cf ploaddup<Packet2cf>(const std::complex< template<> EIGEN_STRONG_INLINE void pstore <std::complex<float> >(std::complex<float> * to, const Packet2cf& from) { EIGEN_DEBUG_ALIGNED_STORE pstore((float*)to, from.v); } template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<float> >(std::complex<float> * to, const Packet2cf& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu((float*)to, from.v); } +template<> EIGEN_DEVICE_FUNC inline Packet2cf pgather<std::complex<float>, Packet2cf>(const std::complex<float>* from, int stride) +{ + Packet4f res; + res = vsetq_lane_f32(std::real(from[0*stride]), res, 0); + res = vsetq_lane_f32(std::imag(from[0*stride]), res, 1); + res = vsetq_lane_f32(std::real(from[1*stride]), res, 2); + res = vsetq_lane_f32(std::imag(from[1*stride]), res, 3); + return Packet2cf(res); +} + +template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<float>, Packet2cf>(std::complex<float>* to, const Packet2cf& from, int stride) +{ + to[stride*0] = std::complex<float>(vgetq_lane_f32(from.v, 0), vgetq_lane_f32(from.v, 1)); + to[stride*1] = std::complex<float>(vgetq_lane_f32(from.v, 2), vgetq_lane_f32(from.v, 3)); +} + template<> EIGEN_STRONG_INLINE void prefetch<std::complex<float> >(const std::complex<float> * addr) { EIGEN_ARM_PREFETCH((float *)addr); } template<> EIGEN_STRONG_INLINE std::complex<float> pfirst<Packet2cf>(const Packet2cf& a) |