| 1 | /* Copyright (C) 2011-2012 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_DETAIL_NULL_MEMORY_H |
| 9 | #define LIBSIMDPP_DETAIL_NULL_MEMORY_H |
| 10 | #if SIMDPP_USE_NULL || SIMDPP_USE_NEON || SIMDPP_USE_ALTIVEC |
| 11 | |
| 12 | #ifndef LIBSIMDPP_SIMD_H |
| 13 | #error "This file must be included through simd.h" |
| 14 | #endif |
| 15 | |
| 16 | #include <simdpp/types.h> |
| 17 | |
| 18 | namespace simdpp { |
| 19 | namespace SIMDPP_ARCH_NAMESPACE { |
| 20 | namespace detail { |
| 21 | namespace null { |
| 22 | |
| 23 | template<class V> SIMDPP_INL |
| 24 | void load(V& a, const char* p) |
| 25 | { |
| 26 | using T = typename V::element_type; |
| 27 | const T* pt = reinterpret_cast<const T*>(p); |
| 28 | for (unsigned i = 0; i < V::length; i++) { |
| 29 | a.el(i) = *pt++; |
| 30 | } |
| 31 | } |
| 32 | |
| 33 | template<class V> SIMDPP_INL |
| 34 | void load_packed2(V& a, V& b, const char* p) |
| 35 | { |
| 36 | using T = typename V::element_type; |
| 37 | const T* pt = reinterpret_cast<const T*>(p); |
| 38 | for (unsigned i = 0; i < V::length; i++) { |
| 39 | a.el(i) = *pt++; |
| 40 | b.el(i) = *pt++; |
| 41 | } |
| 42 | } |
| 43 | |
| 44 | template<class V> SIMDPP_INL |
| 45 | void load_packed3(V& a, V& b, V& c, const char* p) |
| 46 | { |
| 47 | using T = typename V::element_type; |
| 48 | const T* pt = reinterpret_cast<const T*>(p); |
| 49 | for (unsigned i = 0; i < V::length; i++) { |
| 50 | a.el(i) = *pt++; |
| 51 | b.el(i) = *pt++; |
| 52 | c.el(i) = *pt++; |
| 53 | } |
| 54 | } |
| 55 | |
| 56 | template<class V> SIMDPP_INL |
| 57 | void load_packed4(V& a, V& b, V& c, V& d, const char* p) |
| 58 | { |
| 59 | using T = typename V::element_type; |
| 60 | const T* pt = reinterpret_cast<const T*>(p); |
| 61 | for (unsigned i = 0; i < V::length; i++) { |
| 62 | a.el(i) = *pt++; |
| 63 | b.el(i) = *pt++; |
| 64 | c.el(i) = *pt++; |
| 65 | d.el(i) = *pt++; |
| 66 | } |
| 67 | } |
| 68 | |
| 69 | template<class V> SIMDPP_INL |
| 70 | void store(char* p, const V& a) |
| 71 | { |
| 72 | using T = typename V::element_type; |
| 73 | T* pt = reinterpret_cast<T*>(p); |
| 74 | for (unsigned i = 0; i < V::length; i++) { |
| 75 | *pt++ = a.el(i); |
| 76 | } |
| 77 | } |
| 78 | |
| 79 | template<class V, class M> SIMDPP_INL |
| 80 | void store_masked(char* p, const V& a, const M& mask) |
| 81 | { |
| 82 | using T = typename V::element_type; |
| 83 | T* pt = reinterpret_cast<T*>(p); |
| 84 | for (unsigned i = 0; i < a.length; ++i) { |
| 85 | if (mask.el(i)) |
| 86 | *pt = a.el(i); |
| 87 | pt++; |
| 88 | } |
| 89 | } |
| 90 | |
| 91 | template<class V> SIMDPP_INL |
| 92 | void store_first(char* p, const V& a, unsigned n) |
| 93 | { |
| 94 | using T = typename V::element_type; |
| 95 | T* pt = reinterpret_cast<T*>(p); |
| 96 | for (unsigned i = 0; i < V::length && i < n; i++) { |
| 97 | *pt++ = a.el(i); |
| 98 | } |
| 99 | } |
| 100 | |
| 101 | template<class V> SIMDPP_INL |
| 102 | void store_last(char* p, const V& a, unsigned n) |
| 103 | { |
| 104 | using T = typename V::element_type; |
| 105 | T* pt = reinterpret_cast<T*>(p); |
| 106 | pt += V::length - n; |
| 107 | for (unsigned i = V::length - n; i < V::length; i++) { |
| 108 | *pt++ = a.el(i); |
| 109 | } |
| 110 | } |
| 111 | |
| 112 | template<class V> SIMDPP_INL |
| 113 | void store_packed2(char* p, const V& a, const V& b) |
| 114 | { |
| 115 | using T = typename V::element_type; |
| 116 | T* pt = reinterpret_cast<T*>(p); |
| 117 | for (unsigned i = 0; i < V::length; i++) { |
| 118 | *pt++ = a.el(i); |
| 119 | *pt++ = b.el(i); |
| 120 | } |
| 121 | } |
| 122 | |
| 123 | template<class V> SIMDPP_INL |
| 124 | void store_packed3(char* p, const V& a, const V& b, const V& c) |
| 125 | { |
| 126 | using T = typename V::element_type; |
| 127 | T* pt = reinterpret_cast<T*>(p); |
| 128 | for (unsigned i = 0; i < V::length; i++) { |
| 129 | *pt++ = a.el(i); |
| 130 | *pt++ = b.el(i); |
| 131 | *pt++ = c.el(i); |
| 132 | } |
| 133 | } |
| 134 | |
| 135 | template<class V> SIMDPP_INL |
| 136 | void store_packed4(char* p, const V& a, const V& b, const V& c, const V& d) |
| 137 | { |
| 138 | using T = typename V::element_type; |
| 139 | T* pt = reinterpret_cast<T*>(p); |
| 140 | for (unsigned i = 0; i < V::length; i++) { |
| 141 | *pt++ = a.el(i); |
| 142 | *pt++ = b.el(i); |
| 143 | *pt++ = c.el(i); |
| 144 | *pt++ = d.el(i); |
| 145 | } |
| 146 | } |
| 147 | |
| 148 | } // namespace null |
| 149 | } // namespace detail |
| 150 | } // namespace SIMDPP_ARCH_NAMESPACE |
| 151 | } // namespace simdpp |
| 152 | |
| 153 | #endif |
| 154 | #endif |
| 155 | |