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_PACKED2_H |
9 | #define LIBSIMDPP_SIMDPP_DETAIL_INSN_LOAD_PACKED2_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/align.h> |
17 | #include <simdpp/detail/insn/mem_unpack.h> |
18 | #include <simdpp/core/load.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_packed2(V& a, V& b, const char* p); |
30 | template<class V> SIMDPP_INL |
31 | void v256_load_packed2(V& a, V& b, const char* p); |
32 | template<class V> SIMDPP_INL |
33 | void v512_load_packed2(V& a, V& b, const char* p); |
34 | |
35 | // ----------------------------------------------------------------------------- |
36 | |
37 | static SIMDPP_INL |
38 | void i_load_packed2(uint8x16& a, uint8x16& b, const char* p) |
39 | { |
40 | p = detail::assume_aligned(p, 16); |
41 | #if SIMDPP_USE_NULL |
42 | detail::null::load_packed2(a, b, p); |
43 | #elif SIMDPP_USE_SSE2 || SIMDPP_USE_ALTIVEC || SIMDPP_USE_MSA |
44 | v128_load_packed2(a, b, p); |
45 | #elif SIMDPP_USE_NEON |
46 | auto r = vld2q_u8(reinterpret_cast<const uint8_t*>(p)); |
47 | a = r.val[0]; |
48 | b = r.val[1]; |
49 | #endif |
50 | } |
51 | |
52 | #if SIMDPP_USE_AVX2 |
53 | static SIMDPP_INL |
54 | void i_load_packed2(uint8x32& a, uint8x32& b, const char* p) |
55 | { |
56 | v256_load_packed2(a, b, p); |
57 | } |
58 | #endif |
59 | |
60 | #if SIMDPP_USE_AVX512BW |
61 | SIMDPP_INL void i_load_packed2(uint8<64>& a, uint8<64>& b, const char* p) |
62 | { |
63 | v512_load_packed2(a, b, p); |
64 | } |
65 | #endif |
66 | |
67 | // ----------------------------------------------------------------------------- |
68 | |
69 | static SIMDPP_INL |
70 | void i_load_packed2(uint16x8& a, uint16x8& b, const char* p) |
71 | { |
72 | p = detail::assume_aligned(p, 16); |
73 | #if SIMDPP_USE_NULL |
74 | detail::null::load_packed2(a, b, p); |
75 | #elif SIMDPP_USE_SSE2 || SIMDPP_USE_ALTIVEC || SIMDPP_USE_MSA |
76 | v128_load_packed2(a, b, p); |
77 | #elif SIMDPP_USE_NEON |
78 | auto r = vld2q_u16(reinterpret_cast<const uint16_t*>(p)); |
79 | a = r.val[0]; |
80 | b = r.val[1]; |
81 | #endif |
82 | } |
83 | |
84 | #if SIMDPP_USE_AVX2 |
85 | static SIMDPP_INL |
86 | void i_load_packed2(uint16x16& a, uint16x16& b, const char* p) |
87 | { |
88 | v256_load_packed2(a, b, p); |
89 | } |
90 | #endif |
91 | |
92 | #if SIMDPP_USE_AVX512BW |
93 | SIMDPP_INL void i_load_packed2(uint16<32>& a, uint16<32>& b, const char* p) |
94 | { |
95 | v512_load_packed2(a, b, p); |
96 | } |
97 | #endif |
98 | |
99 | // ----------------------------------------------------------------------------- |
100 | |
101 | static SIMDPP_INL |
102 | void i_load_packed2(uint32x4& a, uint32x4& b, const char* p) |
103 | { |
104 | p = detail::assume_aligned(p, 16); |
105 | #if SIMDPP_USE_NULL |
106 | detail::null::load_packed2(a, b, p); |
107 | #elif SIMDPP_USE_SSE2 || SIMDPP_USE_ALTIVEC || SIMDPP_USE_MSA |
108 | v128_load_packed2(a, b, p); |
109 | #elif SIMDPP_USE_NEON |
110 | auto r = vld2q_u32(reinterpret_cast<const uint32_t*>(p)); |
111 | a = r.val[0]; |
112 | b = r.val[1]; |
113 | #endif |
114 | } |
115 | |
116 | #if SIMDPP_USE_AVX2 |
117 | static SIMDPP_INL |
118 | void i_load_packed2(uint32x8& a, uint32x8& b, const char* p) |
119 | { |
120 | v256_load_packed2(a, b, p); |
121 | } |
122 | #endif |
123 | |
124 | #if SIMDPP_USE_AVX512F |
125 | static SIMDPP_INL |
126 | void i_load_packed2(uint32<16>& a, uint32<16>& b, const char* p) |
127 | { |
128 | v512_load_packed2(a, b, p); |
129 | } |
130 | #endif |
131 | |
132 | // ----------------------------------------------------------------------------- |
133 | |
134 | static SIMDPP_INL |
135 | void i_load_packed2(uint64x2& a, uint64x2& b, const char* p) |
136 | { |
137 | p = detail::assume_aligned(p, 16); |
138 | #if SIMDPP_USE_NEON64 |
139 | auto r = vld2q_u64(reinterpret_cast<const uint64_t*>(p)); |
140 | a = r.val[0]; |
141 | b = r.val[1]; |
142 | #elif SIMDPP_USE_SSE2 || SIMDPP_USE_NEON || SIMDPP_USE_VSX_207 || SIMDPP_USE_MSA |
143 | v128_load_packed2(a, b, p); |
144 | #elif SIMDPP_USE_NULL || SIMDPP_USE_ALTIVEC |
145 | detail::null::load_packed2(a, b, p); |
146 | #endif |
147 | } |
148 | |
149 | #if SIMDPP_USE_AVX2 |
150 | static SIMDPP_INL |
151 | void i_load_packed2(uint64x4& a, uint64x4& b, const char* p) |
152 | { |
153 | v256_load_packed2(a, b, p); |
154 | } |
155 | #endif |
156 | |
157 | #if SIMDPP_USE_AVX512F |
158 | static SIMDPP_INL |
159 | void i_load_packed2(uint64<8>& a, uint64<8>& b, const char* p) |
160 | { |
161 | v512_load_packed2(a, b, p); |
162 | } |
163 | #endif |
164 | |
165 | // ----------------------------------------------------------------------------- |
166 | |
167 | static SIMDPP_INL |
168 | void i_load_packed2(float32x4& a, float32x4& b, const char* p) |
169 | { |
170 | p = detail::assume_aligned(p, 16); |
171 | #if SIMDPP_USE_NULL || SIMDPP_USE_NEON_NO_FLT_SP |
172 | detail::null::load_packed2(a, b, p); |
173 | #elif SIMDPP_USE_SSE2 || SIMDPP_USE_ALTIVEC || SIMDPP_USE_MSA |
174 | v128_load_packed2(a, b, p); |
175 | #elif SIMDPP_USE_NEON |
176 | auto r = vld2q_f32(reinterpret_cast<const float*>(p)); |
177 | a = r.val[0]; |
178 | b = r.val[1]; |
179 | #endif |
180 | } |
181 | |
182 | #if SIMDPP_USE_AVX |
183 | static SIMDPP_INL |
184 | void i_load_packed2(float32x8& a, float32x8& b, const char* p) |
185 | { |
186 | v256_load_packed2(a, b, p); |
187 | } |
188 | #endif |
189 | |
190 | #if SIMDPP_USE_AVX512F |
191 | static SIMDPP_INL |
192 | void i_load_packed2(float32<16>& a, float32<16>& b, const char* p) |
193 | { |
194 | v512_load_packed2(a, b, p); |
195 | } |
196 | #endif |
197 | |
198 | // ----------------------------------------------------------------------------- |
199 | |
200 | static SIMDPP_INL |
201 | void i_load_packed2(float64x2& a, float64x2& b, const char* p) |
202 | { |
203 | p = detail::assume_aligned(p, 16); |
204 | #if SIMDPP_USE_NEON64 |
205 | auto r = vld2q_f64(reinterpret_cast<const double*>(p)); |
206 | a = r.val[0]; |
207 | b = r.val[1]; |
208 | #elif SIMDPP_USE_SSE2 || SIMDPP_USE_VSX_206 || SIMDPP_USE_MSA |
209 | v128_load_packed2(a, b, p); |
210 | #elif SIMDPP_USE_NULL || SIMDPP_USE_NEON32 || SIMDPP_USE_ALTIVEC |
211 | detail::null::load_packed2(a, b, p); |
212 | #endif |
213 | } |
214 | |
215 | #if SIMDPP_USE_AVX |
216 | static SIMDPP_INL |
217 | void i_load_packed2(float64x4& a, float64x4& b, const char* p) |
218 | { |
219 | v256_load_packed2(a, b, p); |
220 | } |
221 | #endif |
222 | |
223 | #if SIMDPP_USE_AVX512F |
224 | static SIMDPP_INL |
225 | void i_load_packed2(float64<8>& a, float64<8>& b, const char* p) |
226 | { |
227 | v512_load_packed2(a, b, p); |
228 | } |
229 | #endif |
230 | |
231 | // ----------------------------------------------------------------------------- |
232 | |
233 | template<class V> SIMDPP_INL |
234 | void v128_load_packed2(V& a, V& b, const char* p) |
235 | { |
236 | p = detail::assume_aligned(p, 16); |
237 | a = load(p); |
238 | b = load(p + 16); |
239 | mem_unpack2(a, b); |
240 | } |
241 | |
242 | template<class V> SIMDPP_INL |
243 | void v256_load_packed2(V& a, V& b, const char* p) |
244 | { |
245 | p = detail::assume_aligned(p, 32); |
246 | a = load(p); |
247 | b = load(p + 32); |
248 | mem_unpack2(a, b); |
249 | } |
250 | |
251 | template<class V> SIMDPP_INL |
252 | void v512_load_packed2(V& a, V& b, const char* p) |
253 | { |
254 | p = detail::assume_aligned(p, 64); |
255 | a = load(p); |
256 | b = load(p + 64); |
257 | mem_unpack2(a, b); |
258 | } |
259 | |
260 | template<class V> SIMDPP_INL |
261 | void i_load_packed2(V& a, V& b, const char* p) |
262 | { |
263 | const unsigned veclen = V::base_vector_type::length_bytes; |
264 | |
265 | p = detail::assume_aligned(p, veclen); |
266 | for (unsigned i = 0; i < V::vec_length; ++i) { |
267 | i_load_packed2(a.vec(i), b.vec(i), p); |
268 | p += veclen*2; |
269 | } |
270 | } |
271 | |
272 | |
273 | } // namespace insn |
274 | } // namespace detail |
275 | } // namespace SIMDPP_ARCH_NAMESPACE |
276 | } // namespace simdpp |
277 | |
278 | #endif |
279 | |
280 | |