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_DETAIL_INSN_F_FMSUB_H |
9 | #define LIBSIMDPP_SIMDPP_DETAIL_INSN_F_FMSUB_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/not_implemented.h> |
18 | |
19 | namespace simdpp { |
20 | namespace SIMDPP_ARCH_NAMESPACE { |
21 | namespace detail { |
22 | namespace insn { |
23 | |
24 | static SIMDPP_INL |
25 | float32<4> i_fmsub(const float32<4>& a, const float32<4>& b, |
26 | const float32<4>& c) |
27 | { |
28 | #if SIMDPP_USE_NULL |
29 | return detail::null::fmsub(a, b, c); |
30 | #elif SIMDPP_USE_FMA3 |
31 | return _mm_fmsub_ps(a.native(), b.native(), c.native()); |
32 | #elif SIMDPP_USE_FMA4 |
33 | return _mm_msub_ps(a.native(), b.native(), c.native()); |
34 | #elif SIMDPP_USE_NEON64 |
35 | // FIXME: also in vfpv4 |
36 | return vfmsq_f32(a.native(), b.native(), c.native()); |
37 | #elif SIMDPP_USE_MSA |
38 | return __msa_fmsub_w(c.native(), a.native(), b.native()); |
39 | #else |
40 | return SIMDPP_NOT_IMPLEMENTED3(a, b, c); |
41 | #endif |
42 | } |
43 | |
44 | #if SIMDPP_USE_AVX |
45 | static SIMDPP_INL |
46 | float32<8> i_fmsub(const float32<8>& a, const float32<8>& b, |
47 | const float32<8>& c) |
48 | { |
49 | #if SIMDPP_USE_FMA3 |
50 | return _mm256_fmsub_ps(a.native(), b.native(), c.native()); |
51 | #elif SIMDPP_USE_FMA4 |
52 | return _mm256_msub_ps(a.native(), b.native(), c.native()); |
53 | #else |
54 | return SIMDPP_NOT_IMPLEMENTED3(a, b, c); |
55 | #endif |
56 | } |
57 | #endif |
58 | |
59 | #if SIMDPP_USE_AVX512F |
60 | static SIMDPP_INL |
61 | float32<16> i_fmsub(const float32<16>& a, const float32<16>& b, |
62 | const float32<16>& c) |
63 | { |
64 | return _mm512_fmsub_ps(a.native(), b.native(), c.native()); |
65 | } |
66 | #endif |
67 | |
68 | // ----------------------------------------------------------------------------- |
69 | |
70 | static SIMDPP_INL |
71 | float64<2> i_fmsub(const float64<2>& a, const float64<2>& b, |
72 | const float64<2>& c) |
73 | { |
74 | #if SIMDPP_USE_NULL |
75 | return detail::null::fmsub(a, b, c); |
76 | #elif SIMDPP_USE_FMA3 |
77 | return _mm_fmsub_pd(a.native(), b.native(), c.native()); |
78 | #elif SIMDPP_USE_FMA4 |
79 | return _mm_msub_pd(a.native(), b.native(), c.native()); |
80 | #elif SIMDPP_USE_NEON64 |
81 | // FIXME: also in vfpv4 |
82 | return vfmsq_f64(a.native(), b.native(), c.native()); |
83 | #elif SIMDPP_USE_MSA |
84 | return __msa_fmsub_d(c.native(), a.native(), b.native()); |
85 | #else |
86 | return SIMDPP_NOT_IMPLEMENTED3(a, b, c); |
87 | #endif |
88 | } |
89 | |
90 | #if SIMDPP_USE_AVX |
91 | static SIMDPP_INL |
92 | float64<4> i_fmsub(const float64<4>& a, const float64<4>& b, |
93 | const float64<4>& c) |
94 | { |
95 | #if SIMDPP_USE_FMA3 |
96 | return _mm256_fmsub_pd(a.native(), b.native(), c.native()); |
97 | #elif SIMDPP_USE_FMA4 |
98 | return _mm256_msub_pd(a.native(), b.native(), c.native()); |
99 | #else |
100 | return SIMDPP_NOT_IMPLEMENTED3(a, b, c); |
101 | #endif |
102 | } |
103 | #endif |
104 | |
105 | #if SIMDPP_USE_AVX512F |
106 | static SIMDPP_INL |
107 | float64<8> i_fmsub(const float64<8>& a, const float64<8>& b, |
108 | const float64<8>& c) |
109 | { |
110 | return _mm512_fmsub_pd(a.native(), b.native(), c.native()); |
111 | } |
112 | #endif |
113 | |
114 | // ----------------------------------------------------------------------------- |
115 | |
116 | template<class V> SIMDPP_INL |
117 | V i_fmsub(const V& a, const V& b, const V& c) |
118 | { |
119 | SIMDPP_VEC_ARRAY_IMPL3(V, i_fmsub, a, b, c); |
120 | } |
121 | |
122 | } // namsepace insn |
123 | } // namespace detail |
124 | } // namespace SIMDPP_ARCH_NAMESPACE |
125 | } // namespace simdpp |
126 | |
127 | #endif |
128 | |
129 | |