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_ISNAN_H |
9 | #define LIBSIMDPP_SIMDPP_DETAIL_INSN_ISNAN_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 | mask_float32x4 i_isnan(const float32x4& a) |
27 | { |
28 | #if SIMDPP_USE_NULL || SIMDPP_USE_NEON_NO_FLT_SP |
29 | return detail::null::isnan(a); |
30 | #elif SIMDPP_USE_AVX512VL |
31 | return _mm_cmp_ps_mask(a.native(), a.native(), _CMP_UNORD_Q); |
32 | #elif SIMDPP_USE_AVX |
33 | return _mm_cmp_ps(a.native(), a.native(), _CMP_UNORD_Q); |
34 | #elif SIMDPP_USE_SSE2 |
35 | return (mask_float32x4) _mm_cmpunord_ps(a.native(), a.native()); |
36 | #elif SIMDPP_USE_NEON || SIMDPP_USE_ALTIVEC |
37 | return cmp_neq(a, a); |
38 | #elif SIMDPP_USE_MSA |
39 | return (v4f32) __msa_fcun_w(a.native(), a.native()); |
40 | #endif |
41 | } |
42 | |
43 | #if SIMDPP_USE_AVX |
44 | static SIMDPP_INL |
45 | mask_float32x8 i_isnan(const float32x8& a) |
46 | { |
47 | #if SIMDPP_USE_AVX512VL |
48 | return _mm256_cmp_ps_mask(a.native(), a.native(), _CMP_UNORD_Q); |
49 | #else |
50 | return _mm256_cmp_ps(a.native(), a.native(), _CMP_UNORD_Q); |
51 | #endif |
52 | } |
53 | #endif |
54 | |
55 | #if SIMDPP_USE_AVX512F |
56 | static SIMDPP_INL |
57 | mask_float32<16> i_isnan(const float32<16>& a) |
58 | { |
59 | return _mm512_cmp_ps_mask(a.native(), a.native(), _CMP_UNORD_Q); |
60 | } |
61 | #endif |
62 | |
63 | // ----------------------------------------------------------------------------- |
64 | |
65 | static SIMDPP_INL |
66 | mask_float64x2 i_isnan(const float64x2& a) |
67 | { |
68 | #if SIMDPP_USE_AVX512VL |
69 | return _mm_cmp_pd_mask(a.native(), a.native(), _CMP_UNORD_Q); |
70 | #elif SIMDPP_USE_AVX |
71 | return _mm_cmp_pd(a.native(), a.native(), _CMP_UNORD_Q); |
72 | #elif SIMDPP_USE_SSE2 |
73 | return _mm_cmpunord_pd(a.native(), a.native()); |
74 | #elif SIMDPP_USE_NEON64 || SIMDPP_USE_VSX_206 |
75 | return cmp_neq(a, a); |
76 | #elif SIMDPP_USE_MSA |
77 | return (v2f64) __msa_fcun_d(a.native(), a.native()); |
78 | #elif SIMDPP_USE_NULL || SIMDPP_USE_NEON32 || SIMDPP_USE_ALTIVEC |
79 | return detail::null::isnan(a); |
80 | #endif |
81 | } |
82 | |
83 | #if SIMDPP_USE_AVX |
84 | static SIMDPP_INL |
85 | mask_float64x4 i_isnan(const float64x4& a) |
86 | { |
87 | #if SIMDPP_USE_AVX512VL |
88 | return _mm256_cmp_pd_mask(a.native(), a.native(), _CMP_UNORD_Q); |
89 | #else |
90 | return _mm256_cmp_pd(a.native(), a.native(), _CMP_UNORD_Q); |
91 | #endif |
92 | } |
93 | #endif |
94 | |
95 | #if SIMDPP_USE_AVX512F |
96 | static SIMDPP_INL |
97 | mask_float64<8> i_isnan(const float64<8>& a) |
98 | { |
99 | return _mm512_cmp_pd_mask(a.native(), a.native(), _CMP_UNORD_Q); |
100 | } |
101 | #endif |
102 | |
103 | // ----------------------------------------------------------------------------- |
104 | |
105 | template<class V> SIMDPP_INL |
106 | typename V::mask_vector_type i_isnan(const V& a) |
107 | { |
108 | SIMDPP_VEC_ARRAY_IMPL1(typename V::mask_vector_type, i_isnan, a); |
109 | } |
110 | |
111 | } // namespace insn |
112 | } // namespace detail |
113 | } // namespace SIMDPP_ARCH_NAMESPACE |
114 | } // namespace simdpp |
115 | |
116 | #endif |
117 | |
118 | |