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_MOVE_R_H |
9 | #define LIBSIMDPP_SIMDPP_CORE_MOVE_R_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/insn/move_r.h> |
17 | |
18 | namespace simdpp { |
19 | namespace SIMDPP_ARCH_NAMESPACE { |
20 | |
21 | /** Moves the 8-bit elements in a vector to the right by @a shift positions. |
22 | |
23 | @code |
24 | shift: pos:| 0 1 . 14 15 | |
25 | 0 r = [ a0 a1 . a14 a15 ] |
26 | 1 r = [ 0 a0 . a13 a14 ] |
27 | 2 r = [ 0 0 . a12 a13 ] |
28 | ... .. .. .. ... .. .. .. |
29 | 14 r = [ 0 0 . a0 a1 ] |
30 | 15 r = [ 0 0 . 0 a0 ] |
31 | 16 r = [ 0 0 . 0 0 ] |
32 | @endcode |
33 | |
34 | @par 256-bit version: |
35 | The lower and higher 128-bit halves are processed as if 128-bit instruction |
36 | was applied to each of them separately. |
37 | |
38 | @icost{SSE2-AVX, NEON, ALTIVEC, 2} |
39 | */ |
40 | template<unsigned shift, unsigned N, class V> SIMDPP_INL |
41 | typename detail::get_expr_nomask<V>::empty |
42 | move16_r(const any_vec8<N,V>& a) |
43 | { |
44 | static_assert(shift <= 16, "Shift out of bounds" ); |
45 | |
46 | typename detail::get_expr_nomask_nosign<V>::type ra; |
47 | ra = a.wrapped().eval(); |
48 | return detail::insn::i_move16_r_wrapper<shift>::run(ra); |
49 | } |
50 | |
51 | /** Moves the 16-bit elements in a vector to the right by @a shift positions. |
52 | |
53 | @code |
54 | shift: pos:| 0 1 . 6 7 | |
55 | 0 r = [ a0 a1 . a6 a7 ] |
56 | 1 r = [ 0 a0 . a5 a6 ] |
57 | 2 r = [ 0 0 . a4 a5 ] |
58 | ... .. .. .. ... .. .. |
59 | 6 r = [ 0 0 . a0 a1 ] |
60 | 7 r = [ 0 0 . 0 a0 ] |
61 | 8 r = [ 0 0 . 0 0 ] |
62 | @endcode |
63 | |
64 | @par 256-bit version: |
65 | The lower and higher 128-bit halves are processed as if 128-bit instruction |
66 | was applied to each of them separately. |
67 | |
68 | @icost{SSE2-AVX, NEON, ALTIVEC, 2} |
69 | */ |
70 | template<unsigned shift, unsigned N, class V> SIMDPP_INL |
71 | typename detail::get_expr_nomask<V>::empty |
72 | move8_r(const any_vec16<N,V>& a) |
73 | { |
74 | static_assert(shift <= 8, "Shift out of bounds" ); |
75 | |
76 | typename detail::get_expr_nomask_nosign<V>::type ra; |
77 | ra = a.wrapped().eval(); |
78 | return detail::insn::i_move8_r_wrapper<shift>::run(ra); |
79 | } |
80 | |
81 | /** Moves the 32-bit elements in a vector to the right by @a shift positions. |
82 | |
83 | @code |
84 | shift: pos:| 0 1 2 3 | |
85 | 0 r = [ a0 a1 a2 a3 ] |
86 | 1 r = [ 0 a0 a1 a2 ] |
87 | 2 r = [ 0 0 a0 a1 ] |
88 | 3 r = [ 0 0 0 a0 ] |
89 | 4 r = [ 0 0 0 0 ] |
90 | @endcode |
91 | |
92 | @par 256-bit version: |
93 | The lower and higher 128-bit halves are processed as if 128-bit instruction |
94 | was applied to each of them separately. |
95 | |
96 | @icost{SSE2-AVX, NEON, ALTIVEC, 2} |
97 | */ |
98 | template<unsigned shift, unsigned N, class V> SIMDPP_INL |
99 | typename detail::get_expr_nomask<V>::empty |
100 | move4_r(const any_vec32<N,V>& a) |
101 | { |
102 | static_assert(shift <= 4, "Shift out of bounds" ); |
103 | |
104 | typename detail::get_expr_nomask_nosign<V>::type ra; |
105 | ra = a.wrapped().eval(); |
106 | return detail::insn::i_move4_r_wrapper<shift>::run(ra); |
107 | } |
108 | |
109 | |
110 | /** Moves the 64-bit elements in a vector to the right by @a shift positions. |
111 | |
112 | @code |
113 | shift: pos:| 0 1 | |
114 | 0 r = [ a0 a1 ] |
115 | 1 r = [ 0 a0 ] |
116 | 2 r = [ 0 0 ] |
117 | @endcode |
118 | |
119 | @par 256-bit version: |
120 | The lower and higher 128-bit halves are processed as if 128-bit instruction |
121 | was applied to each of them separately. |
122 | |
123 | @icost{SSE2-AVX, NEON, ALTIVEC, 2} |
124 | */ |
125 | template<unsigned shift, unsigned N, class V> SIMDPP_INL |
126 | typename detail::get_expr_nomask<V>::empty |
127 | move2_r(const any_vec64<N,V>& a) |
128 | { |
129 | static_assert(shift <= 2, "Shift out of bounds" ); |
130 | |
131 | typename detail::get_expr_nomask_nosign<V>::type ra; |
132 | ra = a.wrapped().eval(); |
133 | return detail::insn::i_move2_r_wrapper<shift>::run(ra); |
134 | } |
135 | |
136 | |
137 | } // namespace SIMDPP_ARCH_NAMESPACE |
138 | } // namespace simdpp |
139 | |
140 | #endif |
141 | |
142 | |