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_UNZIP_HI_H |
9 | #define LIBSIMDPP_SIMDPP_DETAIL_INSN_UNZIP_HI_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/shuffle2.h> |
17 | #include <simdpp/core/zip_hi.h> |
18 | #include <simdpp/detail/null/shuffle.h> |
19 | |
20 | namespace simdpp { |
21 | namespace SIMDPP_ARCH_NAMESPACE { |
22 | namespace detail { |
23 | namespace insn { |
24 | |
25 | static SIMDPP_INL |
26 | uint8x16 i_unzip16_hi(const uint8x16& ca, const uint8x16& cb) |
27 | { |
28 | uint8<16> a = ca, b = cb; |
29 | #if SIMDPP_USE_NULL |
30 | return detail::null::unzip16_hi(a, b); |
31 | #elif SIMDPP_USE_SSE2 |
32 | a = _mm_srai_epi16(a.native(), 8); |
33 | b = _mm_srai_epi16(b.native(), 8); |
34 | a = _mm_packs_epi16(a.native(), b.native()); |
35 | return a; |
36 | #elif SIMDPP_USE_NEON |
37 | return vuzpq_u8(a.native(), b.native()).val[1]; |
38 | #elif SIMDPP_USE_ALTIVEC |
39 | #if SIMDPP_BIG_ENDIAN |
40 | return vec_pack((__vector uint16_t) a.native(), |
41 | (__vector uint16_t) b.native()); |
42 | #else |
43 | uint8x16 mask = make_shuffle_bytes16_mask<1,3,5,7,9,11,13,15, |
44 | 17,19,21,23,25,27,29,31>(mask); |
45 | return shuffle_bytes16(a, b, mask); |
46 | #endif |
47 | #elif SIMDPP_USE_MSA |
48 | return (v16u8) __msa_pckod_b((v16i8) b.native(), (v16i8) a.native()); |
49 | #endif |
50 | } |
51 | |
52 | #if SIMDPP_USE_AVX2 |
53 | static SIMDPP_INL |
54 | uint8x32 i_unzip16_hi(const uint8x32& ca, const uint8x32& cb) |
55 | { |
56 | uint8<32> a = ca, b = cb; |
57 | a = _mm256_srai_epi16(a.native(), 8); |
58 | b = _mm256_srai_epi16(b.native(), 8); |
59 | a = _mm256_packs_epi16(a.native(), b.native()); |
60 | return a; |
61 | } |
62 | #endif |
63 | |
64 | #if SIMDPP_USE_AVX512BW |
65 | SIMDPP_INL uint8<64> i_unzip16_hi(const uint8<64>& ca, const uint8<64>& cb) |
66 | { |
67 | uint8<64> a = ca, b = cb; |
68 | a = _mm512_srai_epi16(a.native(), 8); |
69 | b = _mm512_srai_epi16(b.native(), 8); |
70 | a = _mm512_packs_epi16(a.native(), b.native()); |
71 | return a; |
72 | } |
73 | #endif |
74 | |
75 | template<unsigned N> SIMDPP_INL |
76 | uint8<N> i_unzip16_hi(const uint8<N>& a, const uint8<N>& b) |
77 | { |
78 | SIMDPP_VEC_ARRAY_IMPL2(uint8<N>, i_unzip16_hi, a, b) |
79 | } |
80 | |
81 | // ----------------------------------------------------------------------------- |
82 | |
83 | static SIMDPP_INL |
84 | uint16x8 i_unzip8_hi(const uint16x8& ca, const uint16x8& cb) |
85 | { |
86 | uint16<8> a = ca, b = cb; |
87 | #if SIMDPP_USE_NULL |
88 | return detail::null::unzip8_hi(a, b); |
89 | #elif SIMDPP_USE_SSE2 |
90 | a = _mm_srai_epi32(a.native(), 16); |
91 | b = _mm_srai_epi32(b.native(), 16); |
92 | a = _mm_packs_epi32(a.native(), b.native()); |
93 | return a; |
94 | #elif SIMDPP_USE_NEON |
95 | return vuzpq_u16(a.native(), b.native()).val[1]; |
96 | #elif SIMDPP_USE_ALTIVEC |
97 | #if SIMDPP_BIG_ENDIAN |
98 | return vec_pack((__vector uint32_t)a.native(), (__vector uint32_t)b.native()); |
99 | #else |
100 | uint16x8 mask = make_shuffle_bytes16_mask<1,3,5,7,9,11,13,15>(mask); |
101 | return shuffle_bytes16(a, b, mask); |
102 | #endif |
103 | #elif SIMDPP_USE_MSA |
104 | return (v8u16) __msa_pckod_h((v8i16) b.native(), (v8i16) a.native()); |
105 | #endif |
106 | } |
107 | |
108 | #if SIMDPP_USE_AVX2 |
109 | static SIMDPP_INL |
110 | uint16x16 i_unzip8_hi(const uint16x16& ca, const uint16x16& cb) |
111 | { |
112 | uint16<16> a = ca, b = cb; |
113 | a = _mm256_srai_epi32(a.native(), 16); |
114 | b = _mm256_srai_epi32(b.native(), 16); |
115 | a = _mm256_packs_epi32(a.native(), b.native()); |
116 | return a; |
117 | } |
118 | #endif |
119 | |
120 | #if SIMDPP_USE_AVX512BW |
121 | SIMDPP_INL uint16<32> i_unzip8_hi(const uint16<32>& ca, const uint16<32>& cb) |
122 | { |
123 | uint16<32> a = ca, b = cb; |
124 | a = _mm512_srai_epi32(a.native(), 16); |
125 | b = _mm512_srai_epi32(b.native(), 16); |
126 | a = _mm512_packs_epi32(a.native(), b.native()); |
127 | return a; |
128 | } |
129 | #endif |
130 | |
131 | template<unsigned N> SIMDPP_INL |
132 | uint16<N> i_unzip8_hi(const uint16<N>& a, const uint16<N>& b) |
133 | { |
134 | SIMDPP_VEC_ARRAY_IMPL2(uint16<N>, i_unzip8_hi, a, b) |
135 | } |
136 | |
137 | // ----------------------------------------------------------------------------- |
138 | |
139 | static SIMDPP_INL |
140 | uint32x4 i_unzip4_hi(const uint32x4& a, const uint32x4& b) |
141 | { |
142 | #if SIMDPP_USE_NULL |
143 | return detail::null::unzip4_hi(a, b); |
144 | #elif SIMDPP_USE_SSE2 || SIMDPP_USE_ALTIVEC |
145 | return shuffle2<1,3,1,3>(a, b); |
146 | #elif SIMDPP_USE_NEON |
147 | return vuzpq_u32(a.native(), b.native()).val[1]; |
148 | #elif SIMDPP_USE_MSA |
149 | return (v4u32) __msa_pckod_w((v4i32) b.native(), (v4i32) a.native()); |
150 | #endif |
151 | } |
152 | |
153 | #if SIMDPP_USE_AVX2 |
154 | static SIMDPP_INL |
155 | uint32x8 i_unzip4_hi(const uint32x8& a, const uint32x8& b) |
156 | { |
157 | return shuffle2<1,3,1,3>(a, b); |
158 | } |
159 | #endif |
160 | |
161 | #if SIMDPP_USE_AVX512F |
162 | static SIMDPP_INL |
163 | uint32<16> i_unzip4_hi(const uint32<16>& a, const uint32<16>& b) |
164 | { |
165 | return shuffle2<1,3,1,3>(a, b); |
166 | } |
167 | #endif |
168 | |
169 | template<unsigned N> SIMDPP_INL |
170 | uint32<N> i_unzip4_hi(const uint32<N>& a, const uint32<N>& b) |
171 | { |
172 | SIMDPP_VEC_ARRAY_IMPL2(uint32<N>, i_unzip4_hi, a, b) |
173 | } |
174 | |
175 | // ----------------------------------------------------------------------------- |
176 | |
177 | template<unsigned N> SIMDPP_INL |
178 | uint64<N> i_unzip2_hi(const uint64<N>& a, const uint64<N>& b) |
179 | { |
180 | return i_zip2_hi(a, b); |
181 | } |
182 | |
183 | // ----------------------------------------------------------------------------- |
184 | |
185 | static SIMDPP_INL |
186 | float32x4 i_unzip4_hi(const float32x4& a, const float32x4& b) |
187 | { |
188 | #if SIMDPP_USE_NULL || SIMDPP_USE_NEON_NO_FLT_SP |
189 | return detail::null::unzip4_hi(a, b); |
190 | #elif SIMDPP_USE_SSE2 |
191 | return shuffle2<1,3,1,3>(a,b); |
192 | #elif SIMDPP_USE_NEON |
193 | return vuzpq_f32(a.native(), b.native()).val[1]; |
194 | #elif SIMDPP_USE_ALTIVEC |
195 | return float32x4(i_unzip4_hi((uint32x4)a, (uint32x4)b)); |
196 | #elif SIMDPP_USE_MSA |
197 | return (v4f32) __msa_pckod_w((v4i32) b.native(), (v4i32) a.native()); |
198 | #endif |
199 | } |
200 | |
201 | #if SIMDPP_USE_AVX |
202 | static SIMDPP_INL |
203 | float32x8 i_unzip4_hi(const float32x8& a, const float32x8& b) |
204 | { |
205 | return shuffle2<1,3,1,3>(a, b); |
206 | } |
207 | #endif |
208 | |
209 | #if SIMDPP_USE_AVX512F |
210 | static SIMDPP_INL |
211 | float32<16> i_unzip4_hi(const float32<16>& a, const float32<16>& b) |
212 | { |
213 | return shuffle2<1,3,1,3>(a, b); |
214 | } |
215 | #endif |
216 | |
217 | template<unsigned N> SIMDPP_INL |
218 | float32<N> i_unzip4_hi(const float32<N>& a, const float32<N>& b) |
219 | { |
220 | SIMDPP_VEC_ARRAY_IMPL2(float32<N>, i_unzip4_hi, a, b) |
221 | } |
222 | |
223 | // ----------------------------------------------------------------------------- |
224 | |
225 | template<unsigned N> SIMDPP_INL |
226 | float64<N> i_unzip2_hi(const float64<N>& a, const float64<N>& b) |
227 | { |
228 | return i_zip2_hi(a, b); |
229 | } |
230 | |
231 | } // namespace insn |
232 | } // namespace detail |
233 | } // namespace SIMDPP_ARCH_NAMESPACE |
234 | } // namespace simdpp |
235 | |
236 | #endif |
237 | |
238 | |