| 1 | /* Copyright (C) 2013-2017 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_NEG_H |
| 9 | #define LIBSIMDPP_SIMDPP_DETAIL_INSN_F_NEG_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/bit_xor.h> |
| 17 | #include <simdpp/detail/null/math.h> |
| 18 | #include <simdpp/detail/vector_array_macros.h> |
| 19 | |
| 20 | namespace simdpp { |
| 21 | namespace SIMDPP_ARCH_NAMESPACE { |
| 22 | namespace detail { |
| 23 | namespace insn { |
| 24 | |
| 25 | static SIMDPP_INL |
| 26 | float32<4> i_fneg(const float32<4>& a) |
| 27 | { |
| 28 | #if SIMDPP_USE_NULL || SIMDPP_USE_NEON_NO_FLT_SP |
| 29 | return detail::null::neg(a); |
| 30 | #elif SIMDPP_USE_SSE2 || SIMDPP_USE_ALTIVEC || SIMDPP_USE_MSA |
| 31 | // reversion of the sign bit required even for NaNs and zeros |
| 32 | return bit_xor(a, 0x80000000); |
| 33 | #elif SIMDPP_USE_NEON_FLT_SP |
| 34 | return vnegq_f32(a.native()); |
| 35 | #endif |
| 36 | } |
| 37 | |
| 38 | #if SIMDPP_USE_AVX |
| 39 | static SIMDPP_INL |
| 40 | float32<8> i_fneg(const float32<8>& a) |
| 41 | { |
| 42 | return bit_xor(a, 0x80000000); |
| 43 | } |
| 44 | #endif |
| 45 | |
| 46 | #if SIMDPP_USE_AVX512F |
| 47 | static SIMDPP_INL |
| 48 | float32<16> i_fneg(const float32<16>& a) |
| 49 | { |
| 50 | // FIXME: check whether we can simply use sub |
| 51 | return bit_xor(a, 0x80000000); |
| 52 | } |
| 53 | #endif |
| 54 | |
| 55 | // ----------------------------------------------------------------------------- |
| 56 | |
| 57 | static SIMDPP_INL |
| 58 | float64x2 i_fneg(const float64<2>& a) |
| 59 | { |
| 60 | #if SIMDPP_USE_SSE2 || SIMDPP_USE_VSX_206 || SIMDPP_USE_MSA |
| 61 | return bit_xor(a, 0x8000000000000000); |
| 62 | #elif SIMDPP_USE_NEON64 |
| 63 | return vnegq_f64(a.native()); |
| 64 | #elif SIMDPP_USE_NULL || SIMDPP_USE_NEON32 || SIMDPP_USE_ALTIVEC |
| 65 | return detail::null::neg(a); |
| 66 | #endif |
| 67 | } |
| 68 | |
| 69 | #if SIMDPP_USE_AVX |
| 70 | static SIMDPP_INL |
| 71 | float64x4 i_fneg(const float64<4>& a) |
| 72 | { |
| 73 | return bit_xor(a, 0x8000000000000000); |
| 74 | } |
| 75 | #endif |
| 76 | |
| 77 | #if SIMDPP_USE_AVX512F |
| 78 | static SIMDPP_INL |
| 79 | float64<8> i_fneg(const float64<8>& a) |
| 80 | { |
| 81 | // FIXME: check whether we can simply use sub |
| 82 | return bit_xor(a, 0x8000000000000000); |
| 83 | } |
| 84 | #endif |
| 85 | |
| 86 | // ----------------------------------------------------------------------------- |
| 87 | |
| 88 | template<class V> SIMDPP_INL |
| 89 | V i_fneg(const V& a) |
| 90 | { |
| 91 | SIMDPP_VEC_ARRAY_IMPL1(V, i_fneg, a) |
| 92 | } |
| 93 | |
| 94 | } // namespace insn |
| 95 | } // namespace detail |
| 96 | } // namespace SIMDPP_ARCH_NAMESPACE |
| 97 | } // namespace simdpp |
| 98 | |
| 99 | #endif |
| 100 | |
| 101 | |