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_ZIP_LO_H |
9 | #define LIBSIMDPP_SIMDPP_DETAIL_INSN_ZIP_LO_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/shuffle_bytes16.h> |
17 | #include <simdpp/detail/null/shuffle.h> |
18 | #include <simdpp/detail/neon/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_zip16_lo(const uint8x16& a, const uint8x16& b) |
27 | { |
28 | #if SIMDPP_USE_NULL |
29 | return detail::null::zip16_lo(a, b); |
30 | #elif SIMDPP_USE_SSE2 |
31 | return _mm_unpacklo_epi8(a.native(), b.native()); |
32 | #elif SIMDPP_USE_NEON |
33 | // the compiler will optimize multiple vzip instructions if both zip_lo |
34 | // and zip_hi are used on the same arguments |
35 | return vzipq_u8(a.native(), b.native()).val[0]; |
36 | #elif SIMDPP_USE_ALTIVEC |
37 | return vec_mergeh(a.native(), b.native()); |
38 | #elif SIMDPP_USE_MSA |
39 | return (v16u8) __msa_ilvr_b((v16i8)b.native(), (v16i8)a.native()); |
40 | #endif |
41 | } |
42 | |
43 | #if SIMDPP_USE_AVX2 |
44 | static SIMDPP_INL |
45 | uint8x32 i_zip16_lo(const uint8x32& a, const uint8x32& b) |
46 | { |
47 | return _mm256_unpacklo_epi8(a.native(), b.native()); |
48 | } |
49 | #endif |
50 | |
51 | #if SIMDPP_USE_AVX512BW |
52 | SIMDPP_INL uint8<64> i_zip16_lo(const uint8<64>& a, const uint8<64>& b) |
53 | { |
54 | return _mm512_unpacklo_epi8(a.native(), b.native()); |
55 | } |
56 | #endif |
57 | |
58 | template<unsigned N> SIMDPP_INL |
59 | uint8<N> i_zip16_lo(const uint8<N>& a, const uint8<N>& b) |
60 | { |
61 | SIMDPP_VEC_ARRAY_IMPL2(uint8<N>, i_zip16_lo, a, b) |
62 | } |
63 | |
64 | // ----------------------------------------------------------------------------- |
65 | |
66 | static SIMDPP_INL |
67 | uint16x8 i_zip8_lo(const uint16x8& a, const uint16x8& b) |
68 | { |
69 | #if SIMDPP_USE_NULL |
70 | return detail::null::zip8_lo(a, b); |
71 | #elif SIMDPP_USE_SSE2 |
72 | return _mm_unpacklo_epi16(a.native(), b.native()); |
73 | #elif SIMDPP_USE_NEON |
74 | return vzipq_u16(a.native(), b.native()).val[0]; |
75 | #elif SIMDPP_USE_ALTIVEC |
76 | return vec_mergeh(a.native(), b.native()); |
77 | #elif SIMDPP_USE_MSA |
78 | return (v8u16) __msa_ilvr_h((v8i16)b.native(), (v8i16)a.native()); |
79 | #endif |
80 | } |
81 | |
82 | #if SIMDPP_USE_AVX2 |
83 | static SIMDPP_INL |
84 | uint16x16 i_zip8_lo(const uint16x16& a, const uint16x16& b) |
85 | { |
86 | return _mm256_unpacklo_epi16(a.native(), b.native()); |
87 | } |
88 | #endif |
89 | |
90 | #if SIMDPP_USE_AVX512BW |
91 | SIMDPP_INL uint16<32> i_zip8_lo(const uint16<32>& a, const uint16<32>& b) |
92 | { |
93 | return _mm512_unpacklo_epi16(a.native(), b.native()); |
94 | } |
95 | #endif |
96 | |
97 | template<unsigned N> SIMDPP_INL |
98 | uint16<N> i_zip8_lo(const uint16<N>& a, const uint16<N>& b) |
99 | { |
100 | SIMDPP_VEC_ARRAY_IMPL2(uint16<N>, i_zip8_lo, a, b) |
101 | } |
102 | |
103 | // ----------------------------------------------------------------------------- |
104 | |
105 | static SIMDPP_INL |
106 | uint32x4 i_zip4_lo(const uint32x4& a, const uint32x4& b) |
107 | { |
108 | #if SIMDPP_USE_NULL |
109 | return detail::null::zip4_lo(a, b); |
110 | #elif SIMDPP_USE_SSE2 |
111 | return _mm_unpacklo_epi32(a.native(), b.native()); |
112 | #elif SIMDPP_USE_NEON |
113 | return vzipq_u32(a.native(), b.native()).val[0]; |
114 | #elif SIMDPP_USE_ALTIVEC |
115 | return vec_mergeh(a.native(), b.native()); |
116 | #elif SIMDPP_USE_MSA |
117 | return (v4u32) __msa_ilvr_w((v4i32)b.native(), (v4i32)a.native()); |
118 | #endif |
119 | } |
120 | |
121 | #if SIMDPP_USE_AVX2 |
122 | static SIMDPP_INL |
123 | uint32x8 i_zip4_lo(const uint32x8& a, const uint32x8& b) |
124 | { |
125 | return _mm256_unpacklo_epi32(a.native(), b.native()); |
126 | } |
127 | #endif |
128 | |
129 | #if SIMDPP_USE_AVX512F |
130 | static SIMDPP_INL |
131 | uint32<16> i_zip4_lo(const uint32<16>& a, const uint32<16>& b) |
132 | { |
133 | return _mm512_unpacklo_epi32(a.native(), b.native()); |
134 | } |
135 | #endif |
136 | |
137 | template<unsigned N> SIMDPP_INL |
138 | uint32<N> i_zip4_lo(const uint32<N>& a, const uint32<N>& b) |
139 | { |
140 | SIMDPP_VEC_ARRAY_IMPL2(uint32<N>, i_zip4_lo, a, b) |
141 | } |
142 | |
143 | // ----------------------------------------------------------------------------- |
144 | |
145 | static SIMDPP_INL |
146 | uint64x2 i_zip2_lo(const uint64x2& a, const uint64x2& b) |
147 | { |
148 | #if SIMDPP_USE_SSE2 |
149 | return _mm_unpacklo_epi64(a.native(), b.native()); |
150 | #elif SIMDPP_USE_NEON |
151 | return neon::zip2_lo(a, b); |
152 | #elif SIMDPP_USE_VSX_207 |
153 | return vec_mergeh(a.native(), b.native()); |
154 | #elif SIMDPP_USE_MSA |
155 | return (v2u64) __msa_ilvr_d((v2i64) b.native(), (v2i64) a.native()); |
156 | #elif SIMDPP_USE_NULL || SIMDPP_USE_ALTIVEC |
157 | return detail::null::zip2_lo(a, b); |
158 | #endif |
159 | } |
160 | |
161 | #if SIMDPP_USE_AVX2 |
162 | static SIMDPP_INL |
163 | uint64x4 i_zip2_lo(const uint64x4& a, const uint64x4& b) |
164 | { |
165 | return _mm256_unpacklo_epi64(a.native(), b.native()); |
166 | } |
167 | #endif |
168 | |
169 | #if SIMDPP_USE_AVX512F |
170 | static SIMDPP_INL |
171 | uint64<8> i_zip2_lo(const uint64<8>& a, const uint64<8>& b) |
172 | { |
173 | return _mm512_unpacklo_epi64(a.native(), b.native()); |
174 | } |
175 | #endif |
176 | |
177 | template<unsigned N> SIMDPP_INL |
178 | uint64<N> i_zip2_lo(const uint64<N>& a, const uint64<N>& b) |
179 | { |
180 | SIMDPP_VEC_ARRAY_IMPL2(uint64<N>, i_zip2_lo, a, b) |
181 | } |
182 | |
183 | // ----------------------------------------------------------------------------- |
184 | |
185 | static SIMDPP_INL |
186 | float32x4 i_zip4_lo(const float32x4& a, const float32x4& b) |
187 | { |
188 | #if SIMDPP_USE_NULL || SIMDPP_USE_NEON_NO_FLT_SP |
189 | return detail::null::zip4_lo(a, b); |
190 | #elif SIMDPP_USE_SSE2 |
191 | return _mm_unpacklo_ps(a.native(), b.native()); |
192 | #elif SIMDPP_USE_NEON |
193 | return vzipq_f32(a.native(), b.native()).val[0]; |
194 | #elif SIMDPP_USE_ALTIVEC |
195 | return vec_mergeh(a.native(), b.native()); |
196 | #elif SIMDPP_USE_MSA |
197 | return (v4f32) __msa_ilvr_w((v4i32) b.native(), (v4i32) a.native()); |
198 | #endif |
199 | } |
200 | |
201 | #if SIMDPP_USE_AVX |
202 | static SIMDPP_INL |
203 | float32x8 i_zip4_lo(const float32x8& a, const float32x8& b) |
204 | { |
205 | return _mm256_unpacklo_ps(a.native(), b.native()); |
206 | } |
207 | #endif |
208 | |
209 | #if SIMDPP_USE_AVX512F |
210 | static SIMDPP_INL |
211 | float32<16> i_zip4_lo(const float32<16>& a, const float32<16>& b) |
212 | { |
213 | return _mm512_unpacklo_ps(a.native(), b.native()); |
214 | } |
215 | #endif |
216 | |
217 | template<unsigned N> SIMDPP_INL |
218 | float32<N> i_zip4_lo(const float32<N>& a, const float32<N>& b) |
219 | { |
220 | SIMDPP_VEC_ARRAY_IMPL2(float32<N>, i_zip4_lo, a, b) |
221 | } |
222 | |
223 | // ----------------------------------------------------------------------------- |
224 | |
225 | static SIMDPP_INL |
226 | float64x2 i_zip2_lo(const float64x2& a, const float64x2& b) |
227 | { |
228 | #if SIMDPP_USE_SSE2 |
229 | return _mm_castps_pd(_mm_movelh_ps(_mm_castpd_ps(a.native()), |
230 | _mm_castpd_ps(b.native()))); |
231 | #elif SIMDPP_USE_NEON64 |
232 | return vtrn1q_f64(a.native(), b.native()); |
233 | #elif SIMDPP_USE_VSX_206 |
234 | return (__vector double) vec_mergeh((__vector uint64_t)a.native(), |
235 | (__vector uint64_t)b.native()); |
236 | #elif SIMDPP_USE_MSA |
237 | return (v2f64) __msa_ilvr_d((v2i64) b.native(), (v2i64) a.native()); |
238 | #elif SIMDPP_USE_NULL || SIMDPP_USE_ALTIVEC || SIMDPP_USE_NEON |
239 | return detail::null::zip2_lo(a, b); |
240 | #endif |
241 | } |
242 | |
243 | #if SIMDPP_USE_AVX |
244 | static SIMDPP_INL |
245 | float64x4 i_zip2_lo(const float64x4& a, const float64x4& b) |
246 | { |
247 | return _mm256_unpacklo_pd(a.native(), b.native()); |
248 | } |
249 | #endif |
250 | |
251 | #if SIMDPP_USE_AVX512F |
252 | static SIMDPP_INL |
253 | float64<8> i_zip2_lo(const float64<8>& a, const float64<8>& b) |
254 | { |
255 | return _mm512_unpacklo_pd(a.native(), b.native()); |
256 | } |
257 | #endif |
258 | |
259 | template<unsigned N> SIMDPP_INL |
260 | float64<N> i_zip2_lo(const float64<N>& a, const float64<N>& b) |
261 | { |
262 | SIMDPP_VEC_ARRAY_IMPL2(float64<N>, i_zip2_lo, a, b) |
263 | } |
264 | |
265 | |
266 | } // namespace insn |
267 | } // namespace detail |
268 | } // namespace SIMDPP_ARCH_NAMESPACE |
269 | } // namespace simdpp |
270 | |
271 | #endif |
272 | |