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_MUL_H |
9 | #define LIBSIMDPP_SIMDPP_DETAIL_INSN_F_MUL_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 | #if SIMDPP_USE_NULL || SIMDPP_USE_NEON |
17 | #include <simdpp/detail/null/math.h> |
18 | #endif |
19 | #include <simdpp/detail/vector_array_macros.h> |
20 | |
21 | namespace simdpp { |
22 | namespace SIMDPP_ARCH_NAMESPACE { |
23 | namespace detail { |
24 | namespace insn { |
25 | |
26 | static SIMDPP_INL |
27 | float32<4> i_fmul(const float32<4>& a, const float32<4>& b) |
28 | { |
29 | #if SIMDPP_USE_NULL || SIMDPP_USE_NEON_NO_FLT_SP |
30 | return detail::null::mul(a, b); |
31 | #elif SIMDPP_USE_SSE2 |
32 | return _mm_mul_ps(a.native(), b.native()); |
33 | #elif SIMDPP_USE_NEON_FLT_SP |
34 | return vmulq_f32(a.native(), b.native()); |
35 | #elif SIMDPP_USE_VSX_206 |
36 | return vec_mul(a.native(), b.native()); |
37 | #elif SIMDPP_USE_ALTIVEC |
38 | float32<4> zero = make_zero(); |
39 | return vec_madd(a.native(), b.native(), zero.native()); |
40 | #elif SIMDPP_USE_MSA |
41 | return __msa_fmul_w(a.native(), b.native()); |
42 | #endif |
43 | } |
44 | |
45 | #if SIMDPP_USE_AVX |
46 | static SIMDPP_INL |
47 | float32<8> i_fmul(const float32<8>& a, const float32<8>& b) |
48 | { |
49 | return _mm256_mul_ps(a.native(), b.native()); |
50 | } |
51 | #endif |
52 | |
53 | #if SIMDPP_USE_AVX512F |
54 | static SIMDPP_INL |
55 | float32<16> i_fmul(const float32<16>& a, const float32<16>& b) |
56 | { |
57 | return _mm512_mul_ps(a.native(), b.native()); |
58 | } |
59 | #endif |
60 | |
61 | // ----------------------------------------------------------------------------- |
62 | |
63 | static SIMDPP_INL |
64 | float64<2> i_fmul(const float64<2>& a, const float64<2>& b) |
65 | { |
66 | #if SIMDPP_USE_SSE2 |
67 | return _mm_mul_pd(a.native(), b.native()); |
68 | #elif SIMDPP_USE_NEON64 |
69 | return vmulq_f64(a.native(), b.native()); |
70 | #elif SIMDPP_USE_VSX_206 |
71 | return vec_mul(a.native(), b.native()); |
72 | #elif SIMDPP_USE_MSA |
73 | return __msa_fmul_d(a.native(), b.native()); |
74 | #elif SIMDPP_USE_NULL || SIMDPP_USE_NEON32 || SIMDPP_USE_ALTIVEC |
75 | return detail::null::mul(a, b); |
76 | #endif |
77 | } |
78 | |
79 | #if SIMDPP_USE_AVX |
80 | static SIMDPP_INL |
81 | float64<4> i_fmul(const float64<4>& a, const float64<4>& b) |
82 | { |
83 | return _mm256_mul_pd(a.native(), b.native()); |
84 | } |
85 | #endif |
86 | |
87 | #if SIMDPP_USE_AVX512F |
88 | static SIMDPP_INL |
89 | float64<8> i_fmul(const float64<8>& a, const float64<8>& b) |
90 | { |
91 | return _mm512_mul_pd(a.native(), b.native()); |
92 | } |
93 | #endif |
94 | |
95 | // ----------------------------------------------------------------------------- |
96 | |
97 | template<class V> SIMDPP_INL |
98 | V i_fmul(const V& a, const V& b) |
99 | { |
100 | SIMDPP_VEC_ARRAY_IMPL2(V, i_fmul, a, b) |
101 | } |
102 | |
103 | } // namespace insn |
104 | } // namespace detail |
105 | } // namespace SIMDPP_ARCH_NAMESPACE |
106 | } // namespace simdpp |
107 | |
108 | #endif |
109 | |
110 | |