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_PERMUTE_ZBYTES16_H |
9 | #define LIBSIMDPP_SIMDPP_DETAIL_INSN_PERMUTE_ZBYTES16_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/core/permute_bytes16.h> |
18 | #include <simdpp/detail/vector_array_macros.h> |
19 | |
20 | namespace simdpp { |
21 | namespace SIMDPP_ARCH_NAMESPACE { |
22 | namespace detail { |
23 | namespace insn { |
24 | |
25 | #if _MSC_VER |
26 | #pragma warning(push) |
27 | #pragma warning(disable: 4800) |
28 | #endif |
29 | |
30 | static SIMDPP_INL |
31 | uint8x16 i_permute_zbytes16(const uint8x16& a, const uint8x16& mask) |
32 | { |
33 | #if SIMDPP_USE_NULL |
34 | uint8x16 r; |
35 | |
36 | for (unsigned i = 0; i < 16; i++) { |
37 | unsigned j = mask.el(i) & 0x0f; |
38 | bool zero = mask.el(i) & 0x80; |
39 | r.el(i) = zero ? 0 : a.el(j); |
40 | } |
41 | return r; |
42 | #elif SIMDPP_USE_SSSE3 || SIMDPP_USE_NEON |
43 | return permute_bytes16(a, mask); |
44 | #elif SIMDPP_USE_ALTIVEC |
45 | int8x16 a0 = a; |
46 | int8x16 zero_mask = mask; |
47 | zero_mask = shift_r<7>(zero_mask); // shift in the sign bit |
48 | a0 = i_permute_bytes16(a0, mask); |
49 | a0 = bit_andnot(a0, zero_mask); |
50 | return a0; |
51 | #elif SIMDPP_USE_MSA |
52 | return (v16u8) __msa_vshf_b((v16i8) mask.native(), |
53 | (v16i8) a.native(), |
54 | (v16i8) a.native()); |
55 | #else |
56 | return SIMDPP_NOT_IMPLEMENTED2(a, mask); |
57 | #endif |
58 | } |
59 | |
60 | #if _MSC_VER |
61 | #pragma warning(pop) |
62 | #endif |
63 | |
64 | #if SIMDPP_USE_AVX2 |
65 | static SIMDPP_INL |
66 | uint8x32 i_permute_zbytes16(const uint8x32& a, const uint8x32& mask) |
67 | { |
68 | return _mm256_shuffle_epi8(a.native(), mask.native()); |
69 | } |
70 | #endif |
71 | |
72 | #if SIMDPP_USE_AVX512BW |
73 | SIMDPP_INL uint8<64> i_permute_zbytes16(const uint8<64>& a, const uint8<64>& mask) |
74 | { |
75 | return _mm512_shuffle_epi8(a.native(), mask.native()); |
76 | } |
77 | #endif |
78 | |
79 | template<unsigned N> SIMDPP_INL |
80 | uint8<N> i_permute_zbytes16(const uint8<N>& a, const uint8<N>& mask) |
81 | { |
82 | SIMDPP_VEC_ARRAY_IMPL2(uint8<N>, i_permute_zbytes16, a, mask); |
83 | } |
84 | template<unsigned N> SIMDPP_INL |
85 | uint16<N> i_permute_zbytes16(const uint16<N>& a, const uint16<N>& mask) |
86 | { |
87 | return (uint16<N>) i_permute_zbytes16(uint8<N*2>(a), uint8<N*2>(mask)); |
88 | } |
89 | template<unsigned N> SIMDPP_INL |
90 | uint32<N> i_permute_zbytes16(const uint32<N>& a, const uint32<N>& mask) |
91 | { |
92 | return (uint32<N>) i_permute_zbytes16(uint8<N*4>(a), uint8<N*4>(mask)); |
93 | } |
94 | template<unsigned N> SIMDPP_INL |
95 | uint64<N> i_permute_zbytes16(const uint64<N>& a, const uint64<N>& mask) |
96 | { |
97 | return (uint64<N>) i_permute_zbytes16(uint8<N*8>(a), uint8<N*8>(mask)); |
98 | } |
99 | template<unsigned N> SIMDPP_INL |
100 | float32<N> i_permute_zbytes16(const float32<N>& a, const uint32<N>& mask) |
101 | { |
102 | return float32<N>(i_permute_zbytes16(uint32<N>(a), mask)); |
103 | } |
104 | template<unsigned N> SIMDPP_INL |
105 | float64<N> i_permute_zbytes16(const float64<N>& a, const uint64<N>& mask) |
106 | { |
107 | return float64<N>(i_permute_zbytes16(uint64<N>(a), mask)); |
108 | } |
109 | |
110 | |
111 | } // namespace insn |
112 | } // namespace detail |
113 | } // namespace SIMDPP_ARCH_NAMESPACE |
114 | } // namespace simdpp |
115 | |
116 | #endif |
117 | |
118 | |