| 1 | /* Copyright (C) 2013-2014 Povilas Kanapickas <povilas@radix.lt> |
| 2 | |
| 3 | Distributed under the Boost Software License, Version 1.0. |
| 4 | (See accompanying file LICENSE_1_0.txt or copy at |
| 5 | http://www.boost.org/LICENSE_1_0.txt) |
| 6 | */ |
| 7 | |
| 8 | #ifndef LIBSIMDPP_SIMDPP_DETAIL_INSN_PERMUTE2_H |
| 9 | #define LIBSIMDPP_SIMDPP_DETAIL_INSN_PERMUTE2_H |
| 10 | |
| 11 | #ifndef LIBSIMDPP_SIMD_H |
| 12 | #error "This file must be included through simd.h" |
| 13 | #endif |
| 14 | |
| 15 | #include <simdpp/types.h> |
| 16 | #include <simdpp/core/make_shuffle_bytes_mask.h> |
| 17 | #include <simdpp/core/permute4.h> |
| 18 | #include <simdpp/detail/null/shuffle.h> |
| 19 | #include <simdpp/detail/shuffle/neon_int16x8.h> |
| 20 | #include <simdpp/detail/shuffle/neon_int32x4.h> |
| 21 | #include <simdpp/detail/shuffle/neon_int64x2.h> |
| 22 | #include <simdpp/detail/shuffle/shuffle_mask.h> |
| 23 | #include <simdpp/detail/vector_array_macros.h> |
| 24 | |
| 25 | namespace simdpp { |
| 26 | namespace SIMDPP_ARCH_NAMESPACE { |
| 27 | namespace detail { |
| 28 | namespace insn { |
| 29 | |
| 30 | |
| 31 | template<unsigned s0, unsigned s1, unsigned N> SIMDPP_INL |
| 32 | uint16<N> i_permute2(const uint16<N>& a) |
| 33 | { |
| 34 | static_assert(s0 < 2 && s1 < 2, "Selector out of range" ); |
| 35 | return i_permute4<s0,s1,s0+2,s1+2>(a); |
| 36 | } |
| 37 | |
| 38 | |
| 39 | template<unsigned s0, unsigned s1, unsigned N> SIMDPP_INL |
| 40 | uint32<N> i_permute2(const uint32<N>& a) |
| 41 | { |
| 42 | static_assert(s0 < 2 && s1 < 2, "Selector out of range" ); |
| 43 | return i_permute4<s0,s1,s0+2,s1+2>(a); |
| 44 | } |
| 45 | |
| 46 | |
| 47 | template<unsigned s0, unsigned s1, unsigned N> SIMDPP_INL |
| 48 | float32<N> i_permute2(const float32<N>& a) |
| 49 | { |
| 50 | static_assert(s0 < 2 && s1 < 2, "Selector out of range" ); |
| 51 | return i_permute4<s0,s1,s0+2,s1+2>(a); |
| 52 | } |
| 53 | |
| 54 | |
| 55 | template<unsigned s0, unsigned s1> SIMDPP_INL |
| 56 | uint64x2 i_permute2(const uint64x2& a) |
| 57 | { |
| 58 | static_assert(s0 < 2 && s1 < 2, "Selector out of range" ); |
| 59 | #if SIMDPP_USE_SSE2 || SIMDPP_USE_MSA |
| 60 | return (uint64x2) i_permute4<s0*2, s0*2+1, s1*2, s1*2+1>(int32x4(a)); |
| 61 | #elif SIMDPP_USE_NEON |
| 62 | return detail::neon_shuffle_int64x2::permute2<s0,s1>(a); |
| 63 | #elif SIMDPP_USE_VSX_207 |
| 64 | return vec_xxpermdi(a.native(), a.native(), |
| 65 | SIMDPP_VSX_SHUFFLE_MASK_2x2(s0, s1)); |
| 66 | #elif SIMDPP_USE_NULL || SIMDPP_USE_ALTIVEC |
| 67 | return detail::null::permute<s0,s1>(a); |
| 68 | #endif |
| 69 | } |
| 70 | |
| 71 | #if SIMDPP_USE_AVX2 |
| 72 | template<unsigned s0, unsigned s1> SIMDPP_INL |
| 73 | uint64x4 i_permute2(const uint64x4& a) |
| 74 | { |
| 75 | static_assert(s0 < 2 && s1 < 2, "Selector out of range" ); |
| 76 | return i_permute4<s0,s1,s0+2,s1+2>(a); |
| 77 | } |
| 78 | #endif |
| 79 | |
| 80 | #if SIMDPP_USE_AVX512F |
| 81 | template<unsigned s0, unsigned s1> SIMDPP_INL |
| 82 | uint64<8> i_permute2(const uint64<8>& a) |
| 83 | { |
| 84 | static_assert(s0 < 2 && s1 < 2, "Selector out of range" ); |
| 85 | return i_permute4<s0,s1,s0+2,s1+2>(a); |
| 86 | } |
| 87 | #endif |
| 88 | |
| 89 | template<unsigned s0, unsigned s1, unsigned N> SIMDPP_INL |
| 90 | uint64<N> i_permute2(const uint64<N>& a) |
| 91 | { |
| 92 | static_assert(s0 < 2 && s1 < 2, "Selector out of range" ); |
| 93 | SIMDPP_VEC_ARRAY_IMPL1(uint64<N>, (i_permute2<s0,s1>), a); |
| 94 | } |
| 95 | |
| 96 | // ----------------------------------------------------------------------------- |
| 97 | |
| 98 | template<unsigned s0, unsigned s1> SIMDPP_INL |
| 99 | float64x2 i_permute2(const float64x2& a) |
| 100 | { |
| 101 | static_assert(s0 < 2 && s1 < 2, "Selector out of range" ); |
| 102 | #if SIMDPP_USE_SSE2 |
| 103 | return _mm_shuffle_pd(a.native(), a.native(), SIMDPP_SHUFFLE_MASK_2x2(s0, s1)); |
| 104 | #elif SIMDPP_USE_VSX_206 |
| 105 | return vec_xxpermdi(a.native(), a.native(), |
| 106 | SIMDPP_VSX_SHUFFLE_MASK_2x2(s0, s1)); |
| 107 | #elif SIMDPP_USE_NEON64 || SIMDPP_USE_MSA |
| 108 | return float64x2(i_permute2<s0,s1>(int64x2(a))); |
| 109 | #elif SIMDPP_USE_NULL || SIMDPP_USE_NEON || SIMDPP_USE_ALTIVEC |
| 110 | return detail::null::permute<s0,s1>(a); |
| 111 | #endif |
| 112 | } |
| 113 | |
| 114 | #if SIMDPP_USE_AVX |
| 115 | template<unsigned s0, unsigned s1> SIMDPP_INL |
| 116 | float64x4 i_permute2(const float64x4& a) |
| 117 | { |
| 118 | static_assert(s0 < 2 && s1 < 2, "Selector out of range" ); |
| 119 | #if SIMDPP_USE_AVX2 |
| 120 | return _mm256_permute4x64_pd(a.native(), s0 | s1<<2 | (s0+2)<<4 | (s1+2)<<6); |
| 121 | #else // SIMDPP_USE_AVX |
| 122 | return _mm256_permute_pd(a.native(), s0 | s1<<1 | s0<<2 | s1<<3); |
| 123 | #endif |
| 124 | } |
| 125 | #endif |
| 126 | |
| 127 | #if SIMDPP_USE_AVX512F |
| 128 | template<unsigned s0, unsigned s1> SIMDPP_INL |
| 129 | float64<8> i_permute2(const float64<8>& a) |
| 130 | { |
| 131 | static_assert(s0 < 2 && s1 < 2, "Selector out of range" ); |
| 132 | return _mm512_permute_pd(a.native(), s0 | s1<<1 | s0<<2 | s1<<3 | s0<<4 | s1<<5 | s0<<6 | s1<<7); |
| 133 | } |
| 134 | #endif |
| 135 | |
| 136 | template<unsigned s0, unsigned s1, unsigned N> SIMDPP_INL |
| 137 | float64<N> i_permute2(const float64<N>& a) |
| 138 | { |
| 139 | static_assert(s0 < 2 && s1 < 2, "Selector out of range" ); |
| 140 | SIMDPP_VEC_ARRAY_IMPL1(float64<N>, (i_permute2<s0,s1>), a); |
| 141 | } |
| 142 | |
| 143 | } // namespace insn |
| 144 | } // namespace detail |
| 145 | } // namespace SIMDPP_ARCH_NAMESPACE |
| 146 | } // namespace simdpp |
| 147 | |
| 148 | #endif |
| 149 | |
| 150 | |