diff options
-rw-r--r-- | Eigen/src/Core/arch/NEON/Complex.h | 16 | ||||
-rw-r--r-- | Eigen/src/Core/arch/NEON/PacketMath.h | 34 |
2 files changed, 50 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) diff --git a/Eigen/src/Core/arch/NEON/PacketMath.h b/Eigen/src/Core/arch/NEON/PacketMath.h index 83150507a..7f3301f51 100644 --- a/Eigen/src/Core/arch/NEON/PacketMath.h +++ b/Eigen/src/Core/arch/NEON/PacketMath.h @@ -221,6 +221,40 @@ template<> EIGEN_STRONG_INLINE void pstore<int>(int* to, const Packet4i& f template<> EIGEN_STRONG_INLINE void pstoreu<float>(float* to, const Packet4f& from) { EIGEN_DEBUG_UNALIGNED_STORE vst1q_f32(to, from); } template<> EIGEN_STRONG_INLINE void pstoreu<int>(int* to, const Packet4i& from) { EIGEN_DEBUG_UNALIGNED_STORE vst1q_s32(to, from); } +template<> EIGEN_DEVICE_FUNC inline Packet4f pgather<float, Packet4f>(const float* from, int stride) +{ + Packet4f res; + res = vsetq_lane_f32(from[0*stride], res, 0); + res = vsetq_lane_f32(from[1*stride], res, 1); + res = vsetq_lane_f32(from[2*stride], res, 2); + res = vsetq_lane_f32(from[3*stride], res, 3); + return res; +} +template<> EIGEN_DEVICE_FUNC inline Packet4i pgather<int, Packet4i>(const int* from, int stride) +{ + Packet4i res; + res = vsetq_lane_s32(from[0*stride], res, 0); + res = vsetq_lane_s32(from[1*stride], res, 1); + res = vsetq_lane_s32(from[2*stride], res, 2); + res = vsetq_lane_s32(from[3*stride], res, 3); + return res; +} + +template<> EIGEN_DEVICE_FUNC inline void pscatter<float, Packet4f>(float* to, const Packet4f& from, int stride) +{ + to[stride*0] = vgetq_lane_f32(from, 0); + to[stride*1] = vgetq_lane_f32(from, 1); + to[stride*2] = vgetq_lane_f32(from, 2); + to[stride*3] = vgetq_lane_f32(from, 3); +} +template<> EIGEN_DEVICE_FUNC inline void pscatter<int, Packet4i>(int* to, const Packet4i& from, int stride) +{ + to[stride*0] = vgetq_lane_s32(from, 0); + to[stride*1] = vgetq_lane_s32(from, 1); + to[stride*2] = vgetq_lane_s32(from, 2); + to[stride*3] = vgetq_lane_s32(from, 3); +} + template<> EIGEN_STRONG_INLINE void prefetch<float>(const float* addr) { EIGEN_ARM_PREFETCH(addr); } template<> EIGEN_STRONG_INLINE void prefetch<int>(const int* addr) { EIGEN_ARM_PREFETCH(addr); } |