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_LOAD_PACKED4_H |
9 | #define LIBSIMDPP_SIMDPP_DETAIL_INSN_LOAD_PACKED4_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/insn/mem_unpack.h> |
17 | #include <simdpp/core/load.h> |
18 | #include <simdpp/core/transpose.h> |
19 | #include <simdpp/detail/null/memory.h> |
20 | |
21 | namespace simdpp { |
22 | namespace SIMDPP_ARCH_NAMESPACE { |
23 | namespace detail { |
24 | namespace insn { |
25 | |
26 | |
27 | // collect some boilerplate |
28 | template<class V> SIMDPP_INL |
29 | void v128_load_packed4(V& a, V& b, V& c, V& d, const char* p); |
30 | template<class V> SIMDPP_INL |
31 | void v256_load_packed4(V& a, V& b, V& c, V& d, const char* p); |
32 | template<class V> SIMDPP_INL |
33 | void v512_load_packed4(V& a, V& b, V& c, V& d, const char* p); |
34 | |
35 | // ----------------------------------------------------------------------------- |
36 | |
37 | static SIMDPP_INL |
38 | void i_load_packed4(uint8x16& a, uint8x16& b, uint8x16& c, uint8x16& d, |
39 | const char* p) |
40 | { |
41 | p = detail::assume_aligned(p, 16); |
42 | #if SIMDPP_USE_NULL |
43 | detail::null::load_packed4(a, b, c, d, p); |
44 | #elif SIMDPP_USE_SSE2 || SIMDPP_USE_ALTIVEC || SIMDPP_USE_MSA |
45 | v128_load_packed4(a, b, c, d, p); |
46 | #elif SIMDPP_USE_NEON |
47 | auto r = vld4q_u8(reinterpret_cast<const uint8_t*>(p)); |
48 | a = r.val[0]; |
49 | b = r.val[1]; |
50 | c = r.val[2]; |
51 | d = r.val[3]; |
52 | #endif |
53 | } |
54 | |
55 | #if SIMDPP_USE_AVX2 |
56 | static SIMDPP_INL |
57 | void i_load_packed4(uint8x32& a, uint8x32& b, uint8x32& c, uint8x32& d, |
58 | const char* p) |
59 | { |
60 | v256_load_packed4(a, b, c, d, p); |
61 | } |
62 | #endif |
63 | |
64 | #if SIMDPP_USE_AVX512BW |
65 | static SIMDPP_INL |
66 | void i_load_packed4(uint8<64>& a, uint8<64>& b, uint8<64>& c, uint8<64>& d, |
67 | const char* p) |
68 | { |
69 | v512_load_packed4(a, b, c, d, p); |
70 | } |
71 | #endif |
72 | |
73 | // ----------------------------------------------------------------------------- |
74 | |
75 | static SIMDPP_INL |
76 | void i_load_packed4(uint16x8& a, uint16x8& b, uint16x8& c, uint16x8& d, |
77 | const char* p) |
78 | { |
79 | p = detail::assume_aligned(p, 16); |
80 | #if SIMDPP_USE_NULL |
81 | detail::null::load_packed4(a, b, c, d, p); |
82 | #elif SIMDPP_USE_SSE2 || SIMDPP_USE_ALTIVEC || SIMDPP_USE_MSA |
83 | v128_load_packed4(a, b, c, d, p); |
84 | #elif SIMDPP_USE_NEON |
85 | auto r = vld4q_u16(reinterpret_cast<const uint16_t*>(p)); |
86 | a = r.val[0]; |
87 | b = r.val[1]; |
88 | c = r.val[2]; |
89 | d = r.val[3]; |
90 | #endif |
91 | } |
92 | |
93 | #if SIMDPP_USE_AVX2 |
94 | static SIMDPP_INL |
95 | void i_load_packed4(uint16x16& a, uint16x16& b, uint16x16& c, uint16x16& d, |
96 | const char* p) |
97 | { |
98 | v256_load_packed4(a, b, c, d, p); |
99 | } |
100 | #endif |
101 | |
102 | #if SIMDPP_USE_AVX512BW |
103 | static SIMDPP_INL |
104 | void i_load_packed4(uint16<32>& a, uint16<32>& b, uint16<32>& c, uint16<32>& d, |
105 | const char* p) |
106 | { |
107 | v512_load_packed4(a, b, c, d, p); |
108 | } |
109 | #endif |
110 | |
111 | // ----------------------------------------------------------------------------- |
112 | |
113 | static SIMDPP_INL |
114 | void i_load_packed4(uint32x4& a, uint32x4& b, uint32x4& c, uint32x4& d, |
115 | const char* p) |
116 | { |
117 | p = detail::assume_aligned(p, 16); |
118 | #if SIMDPP_USE_NULL |
119 | detail::null::load_packed4(a, b, c, d, p); |
120 | #elif SIMDPP_USE_SSE2 || SIMDPP_USE_ALTIVEC || SIMDPP_USE_MSA |
121 | v128_load_packed4(a, b, c, d, p); |
122 | #elif SIMDPP_USE_NEON |
123 | auto r = vld4q_u32(reinterpret_cast<const uint32_t*>(p)); |
124 | a = r.val[0]; |
125 | b = r.val[1]; |
126 | c = r.val[2]; |
127 | d = r.val[3]; |
128 | #endif |
129 | } |
130 | |
131 | #if SIMDPP_USE_AVX2 |
132 | static SIMDPP_INL |
133 | void i_load_packed4(uint32x8& a, uint32x8& b, uint32x8& c, uint32x8& d, |
134 | const char* p) |
135 | { |
136 | v256_load_packed4(a, b, c, d, p); |
137 | } |
138 | #endif |
139 | |
140 | #if SIMDPP_USE_AVX512F |
141 | static SIMDPP_INL |
142 | void i_load_packed4(uint32<16>& a, uint32<16>& b, uint32<16>& c, uint32<16>& d, |
143 | const char* p) |
144 | { |
145 | v512_load_packed4(a, b, c, d, p); |
146 | } |
147 | #endif |
148 | |
149 | // ----------------------------------------------------------------------------- |
150 | |
151 | static SIMDPP_INL |
152 | void i_load_packed4(uint64x2& a, uint64x2& b, uint64x2& c, uint64x2& d, |
153 | const char* p) |
154 | { |
155 | #if SIMDPP_USE_NEON64 |
156 | auto r = vld4q_u64(reinterpret_cast<const uint64_t*>(p)); |
157 | a = r.val[0]; |
158 | b = r.val[1]; |
159 | c = r.val[2]; |
160 | d = r.val[3]; |
161 | #elif SIMDPP_USE_SSE2 || SIMDPP_USE_NEON || SIMDPP_USE_VSX_207 || SIMDPP_USE_MSA |
162 | v128_load_packed4(a, b, c, d, p); |
163 | #elif SIMDPP_USE_NULL || SIMDPP_USE_ALTIVEC |
164 | detail::null::load_packed4(a, b, c, d, p); |
165 | #endif |
166 | } |
167 | |
168 | #if SIMDPP_USE_AVX2 |
169 | static SIMDPP_INL |
170 | void i_load_packed4(uint64x4& a, uint64x4& b, uint64x4& c, uint64x4& d, |
171 | const char* p) |
172 | { |
173 | v256_load_packed4(a, b, c, d, p); |
174 | } |
175 | #endif |
176 | |
177 | #if SIMDPP_USE_AVX512F |
178 | static SIMDPP_INL |
179 | void i_load_packed4(uint64<8>& a, uint64<8>& b, uint64<8>& c, uint64<8>& d, |
180 | const char* p) |
181 | { |
182 | v512_load_packed4(a, b, c, d, p); |
183 | } |
184 | #endif |
185 | |
186 | // ----------------------------------------------------------------------------- |
187 | |
188 | static SIMDPP_INL |
189 | void i_load_packed4(float32x4& a, float32x4& b, float32x4& c, float32x4& d, |
190 | const char* p) |
191 | { |
192 | p = detail::assume_aligned(p, 16); |
193 | #if SIMDPP_USE_NULL || SIMDPP_USE_NEON_NO_FLT_SP |
194 | detail::null::load_packed4(a, b, c, d, p); |
195 | #elif SIMDPP_USE_SSE2 || SIMDPP_USE_ALTIVEC || SIMDPP_USE_MSA |
196 | v128_load_packed4(a, b, c, d, p); |
197 | #elif SIMDPP_USE_NEON |
198 | auto r = vld4q_f32(reinterpret_cast<const float*>(p)); |
199 | a = r.val[0]; |
200 | b = r.val[1]; |
201 | c = r.val[2]; |
202 | d = r.val[3]; |
203 | #endif |
204 | } |
205 | |
206 | #if SIMDPP_USE_AVX |
207 | static SIMDPP_INL |
208 | void i_load_packed4(float32x8& a, float32x8& b, float32x8& c, float32x8& d, |
209 | const char* p) |
210 | { |
211 | v256_load_packed4(a, b, c, d, p); |
212 | } |
213 | #endif |
214 | |
215 | #if SIMDPP_USE_AVX512F |
216 | static SIMDPP_INL |
217 | void i_load_packed4(float32<16>& a, float32<16>& b, float32<16>& c, float32<16>& d, |
218 | const char* p) |
219 | { |
220 | v512_load_packed4(a, b, c, d, p); |
221 | } |
222 | #endif |
223 | |
224 | // ----------------------------------------------------------------------------- |
225 | |
226 | static SIMDPP_INL |
227 | void i_load_packed4(float64x2& a, float64x2& b, float64x2& c, float64x2& d, |
228 | const char* p) |
229 | { |
230 | p = detail::assume_aligned(p, 16); |
231 | #if SIMDPP_USE_SSE2 || SIMDPP_USE_VSX_206 || SIMDPP_USE_MSA |
232 | v128_load_packed4(a, b, c, d, p); |
233 | #elif SIMDPP_USE_NEON64 |
234 | auto r = vld4q_f64(reinterpret_cast<const double*>(p)); |
235 | a = r.val[0]; |
236 | b = r.val[1]; |
237 | c = r.val[2]; |
238 | d = r.val[3]; |
239 | #elif SIMDPP_USE_NULL || SIMDPP_USE_NEON || SIMDPP_USE_ALTIVEC |
240 | detail::null::load_packed4(a, b, c, d, p); |
241 | #endif |
242 | } |
243 | |
244 | #if SIMDPP_USE_AVX |
245 | static SIMDPP_INL |
246 | void i_load_packed4(float64x4& a, float64x4& b, float64x4& c, float64x4& d, |
247 | const char* p) |
248 | { |
249 | v256_load_packed4(a, b, c, d, p); |
250 | } |
251 | #endif |
252 | |
253 | #if SIMDPP_USE_AVX512F |
254 | static SIMDPP_INL |
255 | void i_load_packed4(float64<8>& a, float64<8>& b, float64<8>& c, float64<8>& d, |
256 | const char* p) |
257 | { |
258 | v512_load_packed4(a, b, c, d, p); |
259 | } |
260 | #endif |
261 | |
262 | // ----------------------------------------------------------------------------- |
263 | |
264 | template<class V> SIMDPP_INL |
265 | void v128_load_packed4(V& a, V& b, V& c, V& d, const char* p) |
266 | { |
267 | p = detail::assume_aligned(p, 16); |
268 | a = load(p); |
269 | b = load(p + 16); |
270 | c = load(p + 32); |
271 | d = load(p + 48); |
272 | mem_unpack4(a, b, c, d); |
273 | } |
274 | |
275 | template<class V> SIMDPP_INL |
276 | void v256_load_packed4(V& a, V& b, V& c, V& d, const char* p) |
277 | { |
278 | p = detail::assume_aligned(p, 32); |
279 | a = load(p); |
280 | b = load(p + 32); |
281 | c = load(p + 64); |
282 | d = load(p + 96); |
283 | mem_unpack4(a, b, c, d); |
284 | } |
285 | |
286 | template<class V> SIMDPP_INL |
287 | void v512_load_packed4(V& a, V& b, V& c, V& d, const char* p) |
288 | { |
289 | p = detail::assume_aligned(p, 64); |
290 | a = load(p); |
291 | b = load(p + 64); |
292 | c = load(p + 128); |
293 | d = load(p + 192); |
294 | mem_unpack4(a, b, c, d); |
295 | } |
296 | |
297 | template<class V> SIMDPP_INL |
298 | void i_load_packed4(V& a, V& b, V& c, V& d, const char* p) |
299 | { |
300 | const unsigned veclen = V::base_vector_type::length_bytes; |
301 | |
302 | p = detail::assume_aligned(p, veclen); |
303 | for (unsigned i = 0; i < V::vec_length; ++i) { |
304 | i_load_packed4(a.vec(i), b.vec(i), c.vec(i), d.vec(i), p); |
305 | p += veclen*4; |
306 | } |
307 | } |
308 | |
309 | |
310 | } // namespace insn |
311 | } // namespace detail |
312 | } // namespace SIMDPP_ARCH_NAMESPACE |
313 | } // namespace simdpp |
314 | |
315 | #endif |
316 | |
317 | |