| 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_CORE_I_MAX_H |
| 9 | #define LIBSIMDPP_SIMDPP_CORE_I_MAX_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/detail/insn/i_max.h> |
| 17 | #include <simdpp/core/detail/scalar_arg_impl.h> |
| 18 | |
| 19 | namespace simdpp { |
| 20 | namespace SIMDPP_ARCH_NAMESPACE { |
| 21 | |
| 22 | |
| 23 | /** Computes maximum of the signed 8-bit values. |
| 24 | |
| 25 | @code |
| 26 | r0 = max(a0, b0) |
| 27 | ... |
| 28 | rN = max(aN, bN) |
| 29 | @endcode |
| 30 | |
| 31 | @par 128-bit version: |
| 32 | @icost{SSE2-SSSE3, 4} |
| 33 | |
| 34 | @par 256-bit version: |
| 35 | @icost{SSE2-SSSE3, 8} |
| 36 | @icost{SSE4.1-AVX, NEON, ALTIVEC, 2} |
| 37 | */ |
| 38 | template<unsigned N, class E1, class E2> SIMDPP_INL |
| 39 | int8<N,expr_empty> max(const int8<N,E1>& a, const int8<N,E2>& b) |
| 40 | { |
| 41 | return detail::insn::i_max(a.eval(), b.eval()); |
| 42 | } |
| 43 | |
| 44 | SIMDPP_SCALAR_ARG_IMPL_VEC(max, int8, int8) |
| 45 | |
| 46 | /** Computes maximum of the unsigned 8-bit values. |
| 47 | |
| 48 | @code |
| 49 | r0 = max(a0, b0) |
| 50 | ... |
| 51 | rN = max(aN, bN) |
| 52 | @endcode |
| 53 | |
| 54 | @par 256-bit version: |
| 55 | @icost{SSE2-AVX, NEON, ALTIVEC, 2} |
| 56 | */ |
| 57 | template<unsigned N, class E1, class E2> SIMDPP_INL |
| 58 | uint8<N,expr_empty> max(const uint8<N,E1>& a, const uint8<N,E2>& b) |
| 59 | { |
| 60 | return detail::insn::i_max(a.eval(), b.eval()); |
| 61 | } |
| 62 | |
| 63 | SIMDPP_SCALAR_ARG_IMPL_VEC(max, uint8, uint8) |
| 64 | |
| 65 | /** Computes maximum of the signed 16-bit values. |
| 66 | |
| 67 | @code |
| 68 | r0 = max(a0, b0) |
| 69 | ... |
| 70 | rN = max(aN, bN) |
| 71 | @endcode |
| 72 | |
| 73 | @par 256-bit version: |
| 74 | @icost{SSE2-AVX, NEON, ALTIVEC, 2} |
| 75 | */ |
| 76 | template<unsigned N, class E1, class E2> SIMDPP_INL |
| 77 | int16<N,expr_empty> max(const int16<N,E1>& a, const int16<N,E2>& b) |
| 78 | { |
| 79 | return detail::insn::i_max(a.eval(), b.eval()); |
| 80 | } |
| 81 | |
| 82 | SIMDPP_SCALAR_ARG_IMPL_VEC(max, int16, int16) |
| 83 | |
| 84 | /** Computes maximum of the unsigned 16-bit values. |
| 85 | |
| 86 | @code |
| 87 | r0 = max(a0, b0) |
| 88 | ... |
| 89 | rN = max(aN, bN) |
| 90 | @endcode |
| 91 | |
| 92 | @par 128-bit version: |
| 93 | @icost{SSE2-SSSE3, 6-7} |
| 94 | |
| 95 | @par 256-bit version: |
| 96 | @icost{SSE2-SSSE3, 12-13} |
| 97 | @icost{SSE4.1-AVX, NEON, ALTIVEC, 2} |
| 98 | */ |
| 99 | template<unsigned N, class E1, class E2> SIMDPP_INL |
| 100 | uint16<N,expr_empty> max(const uint16<N,E1>& a, const uint16<N,E2>& b) |
| 101 | { |
| 102 | return detail::insn::i_max(a.eval(), b.eval()); |
| 103 | } |
| 104 | |
| 105 | SIMDPP_SCALAR_ARG_IMPL_VEC(max, uint16, uint16) |
| 106 | |
| 107 | /** Computes maximum of the signed 32-bit values. |
| 108 | |
| 109 | @code |
| 110 | r0 = max(a0, b0) |
| 111 | ... |
| 112 | rN = max(aN, bN) |
| 113 | @endcode |
| 114 | |
| 115 | @par 128-bit version: |
| 116 | @icost{SSE2-SSSE3, 4} |
| 117 | |
| 118 | @par 256-bit version: |
| 119 | @icost{SSE2-SSSE3, 8} |
| 120 | @icost{SSE4.1-AVX, NEON, ALTIVEC, 2} |
| 121 | */ |
| 122 | template<unsigned N, class E1, class E2> SIMDPP_INL |
| 123 | int32<N,expr_empty> max(const int32<N,E1>& a, const int32<N,E2>& b) |
| 124 | { |
| 125 | return detail::insn::i_max(a.eval(), b.eval()); |
| 126 | } |
| 127 | |
| 128 | SIMDPP_SCALAR_ARG_IMPL_VEC(max, int32, int32) |
| 129 | |
| 130 | /** Computes maximum of the unsigned 32-bit values. |
| 131 | |
| 132 | @code |
| 133 | r0 = max(a0, b0) |
| 134 | ... |
| 135 | rN = max(aN, bN) |
| 136 | @endcode |
| 137 | |
| 138 | @par 128-bit version: |
| 139 | @icost{SSE2-SSSE3, 6-7} |
| 140 | |
| 141 | @par 256-bit version: |
| 142 | @icost{SSE2-SSSE3, 12-13} |
| 143 | @icost{SSE4.1-AVX, NEON, ALTIVEC, 2} |
| 144 | */ |
| 145 | template<unsigned N, class E1, class E2> SIMDPP_INL |
| 146 | uint32<N,expr_empty> max(const uint32<N,E1>& a, const uint32<N,E2>& b) |
| 147 | { |
| 148 | return detail::insn::i_max(a.eval(), b.eval()); |
| 149 | } |
| 150 | |
| 151 | SIMDPP_SCALAR_ARG_IMPL_VEC(max, uint32, uint32) |
| 152 | |
| 153 | /** Computes maximum of the signed 64-bit values. |
| 154 | |
| 155 | @code |
| 156 | r0 = max(a0, b0) |
| 157 | ... |
| 158 | rN = max(aN, bN) |
| 159 | @endcode |
| 160 | |
| 161 | Supported since AVX2, NEON64. Not supported on ALTIVEC. |
| 162 | */ |
| 163 | template<unsigned N, class E1, class E2> SIMDPP_INL |
| 164 | int64<N,expr_empty> max(const int64<N,E1>& a, const int64<N,E2>& b) |
| 165 | { |
| 166 | return detail::insn::i_max(a.eval(), b.eval()); |
| 167 | } |
| 168 | |
| 169 | SIMDPP_SCALAR_ARG_IMPL_VEC(max, int64, int64) |
| 170 | |
| 171 | /** Computes maximum of the unsigned 64-bit values. |
| 172 | |
| 173 | @code |
| 174 | r0 = max(a0, b0) |
| 175 | ... |
| 176 | rN = max(aN, bN) |
| 177 | @endcode |
| 178 | |
| 179 | Supported since AVX2, NEON64. Not supported on ALTIVEC. |
| 180 | */ |
| 181 | template<unsigned N, class E1, class E2> SIMDPP_INL |
| 182 | uint64<N,expr_empty> max(const uint64<N,E1>& a, const uint64<N,E2>& b) |
| 183 | { |
| 184 | return detail::insn::i_max(a.eval(), b.eval()); |
| 185 | } |
| 186 | |
| 187 | SIMDPP_SCALAR_ARG_IMPL_VEC(max, uint64, uint64) |
| 188 | |
| 189 | } // namespace SIMDPP_ARCH_NAMESPACE |
| 190 | } // namespace simdpp |
| 191 | |
| 192 | #endif |
| 193 | |
| 194 | |