| 1 | /* Copyright (C) 2011-2014 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_F_MAX_H |
| 9 | #define LIBSIMDPP_SIMDPP_DETAIL_INSN_F_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/null/math.h> |
| 17 | #include <simdpp/detail/vector_array_macros.h> |
| 18 | |
| 19 | namespace simdpp { |
| 20 | namespace SIMDPP_ARCH_NAMESPACE { |
| 21 | namespace detail { |
| 22 | namespace insn { |
| 23 | |
| 24 | |
| 25 | static SIMDPP_INL |
| 26 | float32x4 i_max(const float32x4& a, const float32x4& b) |
| 27 | { |
| 28 | #if SIMDPP_USE_NULL || SIMDPP_USE_NEON_NO_FLT_SP |
| 29 | return detail::null::max(a, b); |
| 30 | #elif SIMDPP_USE_SSE2 |
| 31 | return _mm_max_ps(a.native(), b.native()); |
| 32 | #elif SIMDPP_USE_NEON |
| 33 | return vmaxq_f32(a.native(), b.native()); |
| 34 | #elif SIMDPP_USE_ALTIVEC |
| 35 | return vec_max(a.native(), b.native()); |
| 36 | #elif SIMDPP_USE_MSA |
| 37 | return __msa_fmax_w(a.native(), b.native()); |
| 38 | #endif |
| 39 | } |
| 40 | |
| 41 | #if SIMDPP_USE_AVX |
| 42 | static SIMDPP_INL |
| 43 | float32x8 i_max(const float32x8& a, const float32x8& b) |
| 44 | { |
| 45 | return _mm256_max_ps(a.native(), b.native()); |
| 46 | } |
| 47 | #endif |
| 48 | |
| 49 | #if SIMDPP_USE_AVX512F |
| 50 | static SIMDPP_INL |
| 51 | float32<16> i_max(const float32<16>& a, const float32<16>& b) |
| 52 | { |
| 53 | return _mm512_max_ps(a.native(), b.native()); |
| 54 | } |
| 55 | #endif |
| 56 | |
| 57 | template<unsigned N> SIMDPP_INL |
| 58 | float32<N> i_max(const float32<N>& a, const float32<N>& b) |
| 59 | { |
| 60 | SIMDPP_VEC_ARRAY_IMPL2(float32<N>, i_max, a, b); |
| 61 | } |
| 62 | |
| 63 | // ----------------------------------------------------------------------------- |
| 64 | |
| 65 | static SIMDPP_INL |
| 66 | float64x2 i_max(const float64x2& a, const float64x2& b) |
| 67 | { |
| 68 | #if SIMDPP_USE_SSE2 |
| 69 | return _mm_max_pd(a.native(), b.native()); |
| 70 | #elif SIMDPP_USE_NEON64 |
| 71 | return vmaxq_f64(a.native(), b.native()); |
| 72 | #elif SIMDPP_USE_VSX_206 |
| 73 | return vec_max(a.native(), b.native()); |
| 74 | #elif SIMDPP_USE_MSA |
| 75 | return __msa_fmax_d(a.native(), b.native()); |
| 76 | #elif SIMDPP_USE_NULL || SIMDPP_USE_NEON32 || SIMDPP_USE_ALTIVEC |
| 77 | return detail::null::max(a, b); |
| 78 | #endif |
| 79 | } |
| 80 | |
| 81 | #if SIMDPP_USE_AVX |
| 82 | static SIMDPP_INL |
| 83 | float64x4 i_max(const float64x4& a, const float64x4& b) |
| 84 | { |
| 85 | return _mm256_max_pd(a.native(), b.native()); |
| 86 | } |
| 87 | #endif |
| 88 | |
| 89 | #if SIMDPP_USE_AVX512F |
| 90 | static SIMDPP_INL |
| 91 | float64<8> i_max(const float64<8>& a, const float64<8>& b) |
| 92 | { |
| 93 | return _mm512_max_pd(a.native(), b.native()); |
| 94 | } |
| 95 | #endif |
| 96 | |
| 97 | template<unsigned N> SIMDPP_INL |
| 98 | float64<N> i_max(const float64<N>& a, const float64<N>& b) |
| 99 | { |
| 100 | SIMDPP_VEC_ARRAY_IMPL2(float64<N>, i_max, a, b); |
| 101 | } |
| 102 | |
| 103 | } // namespace insn |
| 104 | } // namespace detail |
| 105 | } // namespace SIMDPP_ARCH_NAMESPACE |
| 106 | } // namespace simdpp |
| 107 | |
| 108 | #endif |
| 109 | |
| 110 | |