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_PERMUTE_BYTES16_H |
9 | #define LIBSIMDPP_SIMDPP_DETAIL_INSN_PERMUTE_BYTES16_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/not_implemented.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 | static SIMDPP_INL |
25 | uint8x16 i_permute_bytes16(const uint8x16& a, const uint8x16& mask) |
26 | { |
27 | #if SIMDPP_USE_NULL |
28 | uint8x16 r; |
29 | |
30 | for (unsigned i = 0; i < 16; i++) { |
31 | unsigned j = mask.el(i) & 0x0f; |
32 | r.el(i) = a.el(j); |
33 | } |
34 | return r; |
35 | #elif SIMDPP_USE_SSSE3 |
36 | return _mm_shuffle_epi8(a.native(), mask.native()); |
37 | #elif SIMDPP_USE_NEON32 |
38 | uint8x8x2_t table = {{vget_low_u8(a.native()), vget_high_u8(a.native())}}; |
39 | uint8x8_t lo = vtbl2_u8(table, vget_low_u8(mask.native())); |
40 | uint8x8_t hi = vtbl2_u8(table, vget_high_u8(mask.native())); |
41 | return vcombine_u8(lo, hi); |
42 | #elif SIMDPP_USE_NEON64 |
43 | return vqtbl1q_u8(a.native(), mask.native()); |
44 | #elif SIMDPP_USE_ALTIVEC |
45 | return vec_perm(a.native(), a.native(), mask.native()); |
46 | #elif SIMDPP_USE_MSA |
47 | return (v16u8) __msa_vshf_b((v16i8)mask.native(), |
48 | (v16i8)a.native(), |
49 | (v16i8)a.native()); |
50 | #else |
51 | return SIMDPP_NOT_IMPLEMENTED2(a, mask); |
52 | #endif |
53 | } |
54 | |
55 | #if SIMDPP_USE_AVX2 |
56 | static SIMDPP_INL |
57 | uint8x32 i_permute_bytes16(const uint8x32& a, const uint8x32& mask) |
58 | { |
59 | return _mm256_shuffle_epi8(a.native(), mask.native()); |
60 | } |
61 | #endif |
62 | |
63 | #if SIMDPP_USE_AVX512BW |
64 | SIMDPP_INL uint8<64> i_permute_bytes16(const uint8<64>& a, const uint8<64>& mask) |
65 | { |
66 | return _mm512_shuffle_epi8(a.native(), mask.native()); |
67 | } |
68 | #endif |
69 | |
70 | template<unsigned N> SIMDPP_INL |
71 | uint8<N> i_permute_bytes16(const uint8<N>& a, const uint8<N>& mask) |
72 | { |
73 | SIMDPP_VEC_ARRAY_IMPL2(uint8<N>, i_permute_bytes16, a, mask) |
74 | } |
75 | template<unsigned N> SIMDPP_INL |
76 | uint16<N> i_permute_bytes16(const uint16<N>& a, const uint16<N>& mask) |
77 | { |
78 | return (uint16<N>) i_permute_bytes16(uint8<N*2>(a), uint8<N*2>(mask)); |
79 | } |
80 | template<unsigned N> SIMDPP_INL |
81 | uint32<N> i_permute_bytes16(const uint32<N>& a, const uint32<N>& mask) |
82 | { |
83 | return (uint32<N>) i_permute_bytes16(uint8<N*4>(a), uint8<N*4>(mask)); |
84 | } |
85 | template<unsigned N> SIMDPP_INL |
86 | uint64<N> i_permute_bytes16(const uint64<N>& a, const uint64<N>& mask) |
87 | { |
88 | return (uint64<N>) i_permute_bytes16(uint8<N*8>(a), uint8<N*8>(mask)); |
89 | } |
90 | template<unsigned N> SIMDPP_INL |
91 | float32<N> i_permute_bytes16(const float32<N>& a, const uint32<N>& mask) |
92 | { |
93 | return float32<N>(i_permute_bytes16(uint32<N>(a), mask)); |
94 | } |
95 | template<unsigned N> SIMDPP_INL |
96 | float64<N> i_permute_bytes16(const float64<N>& a, const uint64<N>& mask) |
97 | { |
98 | return float64<N>(i_permute_bytes16(uint64<N>(a), mask)); |
99 | } |
100 | |
101 | |
102 | } // namespace insn |
103 | } // namespace detail |
104 | } // namespace SIMDPP_ARCH_NAMESPACE |
105 | } // namespace simdpp |
106 | |
107 | #endif |
108 | |
109 | |