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_CONV_EXTEND_TO_INT16_H |
9 | #define LIBSIMDPP_SIMDPP_DETAIL_INSN_CONV_EXTEND_TO_INT16_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/core/combine.h> |
17 | #include <simdpp/detail/not_implemented.h> |
18 | #include <simdpp/core/i_shift_r.h> |
19 | #include <simdpp/core/move_l.h> |
20 | #include <simdpp/core/zip_hi.h> |
21 | #include <simdpp/core/zip_lo.h> |
22 | #include <simdpp/detail/vector_array_conv_macros.h> |
23 | |
24 | namespace simdpp { |
25 | namespace SIMDPP_ARCH_NAMESPACE { |
26 | namespace detail { |
27 | namespace insn { |
28 | |
29 | SIMDPP_INL uint16<16> i_to_uint16(const uint8<16>& a) |
30 | { |
31 | #if SIMDPP_USE_NULL |
32 | uint16x16 r; |
33 | for (unsigned i = 0; i < 16; i++) { |
34 | r.vec(i/8).el(i%8) = uint16_t(a.el(i)); |
35 | } |
36 | return r; |
37 | #elif SIMDPP_USE_AVX2 |
38 | return _mm256_cvtepu8_epi16(a.native()); |
39 | #elif SIMDPP_USE_SSE4_1 |
40 | uint16x8 r1, r2; |
41 | r1 = _mm_cvtepu8_epi16(a.native()); |
42 | r2 = _mm_cvtepu8_epi16(move16_l<8>(a).eval().native()); |
43 | return combine(r1, r2); |
44 | #elif SIMDPP_USE_SSE2 || (SIMDPP_USE_ALTIVEC && SIMDPP_LITTLE_ENDIAN) || SIMDPP_USE_MSA |
45 | uint16x8 r1, r2; |
46 | r1 = zip16_lo(a, (uint8x16) make_zero()); |
47 | r2 = zip16_hi(a, (uint8x16) make_zero()); |
48 | return combine(r1, r2); |
49 | #elif SIMDPP_USE_NEON |
50 | uint16x16 r; |
51 | r.vec(0) = vmovl_u8(vget_low_u8(a.native())); |
52 | r.vec(1) = vmovl_u8(vget_high_u8(a.native())); |
53 | return r; |
54 | #elif (SIMDPP_USE_ALTIVEC && SIMDPP_BIG_ENDIAN) |
55 | uint16x8 r1, r2; |
56 | r1 = zip16_lo((uint8x16) make_zero(), a); |
57 | r2 = zip16_hi((uint8x16) make_zero(), a); |
58 | return combine(r1, r2); |
59 | #endif |
60 | } |
61 | |
62 | #if SIMDPP_USE_AVX2 |
63 | SIMDPP_INL uint16<32> i_to_uint16(const uint8<32>& a) |
64 | { |
65 | #if SIMDPP_USE_AVX512BW |
66 | return _mm512_cvtepu8_epi16(a.native()); |
67 | #else |
68 | uint16<16> r0, r1; |
69 | uint8<16> a0, a1; |
70 | split(a, a0, a1); |
71 | r0 = _mm256_cvtepu8_epi16(a0.native()); |
72 | r1 = _mm256_cvtepu8_epi16(a1.native()); |
73 | return combine(r0, r1); |
74 | #endif |
75 | } |
76 | #endif |
77 | |
78 | #if SIMDPP_USE_AVX512BW |
79 | SIMDPP_INL uint16<64> i_to_uint16(const uint8<64>& a) |
80 | { |
81 | uint16<32> r0, r1; |
82 | uint8<32> a0, a1; |
83 | split(a, a0, a1); |
84 | r0 = _mm512_cvtepu8_epi16(a0.native()); |
85 | r1 = _mm512_cvtepu8_epi16(a1.native()); |
86 | return combine(r0, r1); |
87 | } |
88 | #endif |
89 | |
90 | template<unsigned N> SIMDPP_INL |
91 | uint16<N> i_to_uint16(const uint8<N>& a) |
92 | { |
93 | SIMDPP_VEC_ARRAY_IMPL_CONV_INSERT(uint16<N>, i_to_uint16, a) |
94 | } |
95 | |
96 | // ----------------------------------------------------------------------------- |
97 | |
98 | SIMDPP_INL int16x16 i_to_int16(const int8x16& a) |
99 | { |
100 | #if SIMDPP_USE_NULL |
101 | int16x16 r; |
102 | for (unsigned i = 0; i < 16; i++) { |
103 | r.vec(i/8).el(i%8) = int16_t(a.el(i)); |
104 | } |
105 | return r; |
106 | #elif SIMDPP_USE_AVX2 |
107 | return _mm256_cvtepi8_epi16(a.native()); |
108 | #elif SIMDPP_USE_SSE4_1 |
109 | int16x8 r1, r2; |
110 | r1 = _mm_cvtepi8_epi16(a.native()); |
111 | r2 = _mm_cvtepi8_epi16(move16_l<8>(a).eval().native()); |
112 | return combine(r1, r2); |
113 | #elif SIMDPP_USE_SSE2 |
114 | int16x8 r1, r2; |
115 | r1 = zip16_lo((int8x16) make_zero(), a); |
116 | r1 = shift_r<8>(r1); |
117 | r2 = zip16_hi((int8x16) make_zero(), a); |
118 | r2 = shift_r<8>(r2); |
119 | return combine(r1, r2); |
120 | #elif SIMDPP_USE_NEON |
121 | int16x16 r; |
122 | r.vec(0) = vmovl_s8(vget_low_s8(a.native())); |
123 | r.vec(1) = vmovl_s8(vget_high_s8(a.native())); |
124 | return r; |
125 | #elif SIMDPP_USE_MSA |
126 | int8x16 sign = shift_r<7>(a); |
127 | int16x8 lo, hi; |
128 | lo = zip16_lo(a, sign); |
129 | hi = zip16_hi(a, sign); |
130 | return combine(lo, hi); |
131 | #elif SIMDPP_USE_ALTIVEC |
132 | int16x16 r; |
133 | r.vec(0) = vec_unpackh(a.vec(0).native()); |
134 | r.vec(1) = vec_unpackl(a.vec(0).native()); |
135 | return r; |
136 | #endif |
137 | } |
138 | |
139 | #if SIMDPP_USE_AVX2 |
140 | SIMDPP_INL int16<32> i_to_int16(const int8<32>& a) |
141 | { |
142 | #if SIMDPP_USE_AVX512BW |
143 | return _mm512_cvtepi8_epi16(a.native()); |
144 | #else |
145 | int16<16> r0, r1; |
146 | int8<16> a0, a1; |
147 | split(a, a0, a1); |
148 | r0 = _mm256_cvtepi8_epi16(a0.native()); |
149 | r1 = _mm256_cvtepi8_epi16(a1.native()); |
150 | return combine(r0, r1); |
151 | #endif |
152 | } |
153 | #endif |
154 | |
155 | #if SIMDPP_USE_AVX512BW |
156 | SIMDPP_INL int16<64> i_to_int16(const int8<64>& a) |
157 | { |
158 | int16<32> r0, r1; |
159 | int8<32> a0, a1; |
160 | split(a, a0, a1); |
161 | r0 = _mm512_cvtepi8_epi16(a0.native()); |
162 | r1 = _mm512_cvtepi8_epi16(a1.native()); |
163 | return combine(r0, r1); |
164 | } |
165 | #endif |
166 | |
167 | template<unsigned N> SIMDPP_INL |
168 | int16<N> i_to_int16(const int8<N>& a) |
169 | { |
170 | SIMDPP_VEC_ARRAY_IMPL_CONV_INSERT(int16<N>, i_to_int16, a) |
171 | } |
172 | |
173 | |
174 | } // namespace insn |
175 | } // namespace detail |
176 | } // namespace SIMDPP_ARCH_NAMESPACE |
177 | } // namespace simdpp |
178 | |
179 | #endif |
180 | |
181 | |
182 | |