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_CORE_F_FMSUB_H |
9 | #define LIBSIMDPP_SIMDPP_CORE_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/expr/f_fmsub.h> |
17 | |
18 | namespace simdpp { |
19 | namespace SIMDPP_ARCH_NAMESPACE { |
20 | |
21 | /** Performs a fused multiply-sutract operation |
22 | |
23 | @code |
24 | r0 = a0 * b0 - c0 |
25 | ... |
26 | rN = aN * bN - cN |
27 | @endcode |
28 | |
29 | Implemented only on architectures with either @c X86_FMA3 or @c X86_FMA4 |
30 | support. |
31 | */ |
32 | template<unsigned N, class E1, class E2, class E3> SIMDPP_INL |
33 | float32<N, expr_fmsub<float32<N,E1>, |
34 | float32<N,E2>, |
35 | float32<N,E3>>> fmsub(const float32<N,E1>& a, |
36 | const float32<N,E2>& b, |
37 | const float32<N,E3>& c) |
38 | { |
39 | return { { a, b, c } }; |
40 | } |
41 | |
42 | template<unsigned N, class E1, class E2, class E3> SIMDPP_INL |
43 | float64<N, expr_fmsub<float64<N,E1>, |
44 | float64<N,E2>, |
45 | float64<N,E3>>> fmsub(const float64<N,E1>& a, |
46 | const float64<N,E2>& b, |
47 | const float64<N,E3>& c) |
48 | { |
49 | return { { a, b, c } }; |
50 | } |
51 | |
52 | } // namespace SIMDPP_ARCH_NAMESPACE |
53 | } // namespace simdpp |
54 | |
55 | #endif |
56 | |
57 | |