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_I_MIN_H |
9 | #define LIBSIMDPP_SIMDPP_DETAIL_INSN_I_MIN_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/blend.h> |
17 | #include <simdpp/core/cmp_lt.h> |
18 | #include <simdpp/core/bit_xor.h> |
19 | #include <simdpp/detail/not_implemented.h> |
20 | #include <simdpp/detail/null/math.h> |
21 | #include <simdpp/detail/vector_array_macros.h> |
22 | |
23 | namespace simdpp { |
24 | namespace SIMDPP_ARCH_NAMESPACE { |
25 | namespace detail { |
26 | namespace insn { |
27 | |
28 | |
29 | static SIMDPP_INL |
30 | int8x16 i_min(const int8x16& a, const int8x16& b) |
31 | { |
32 | #if SIMDPP_USE_NULL |
33 | return detail::null::min(a, b); |
34 | #elif SIMDPP_USE_SSE4_1 |
35 | return _mm_min_epi8(a.native(), b.native()); |
36 | #elif SIMDPP_USE_SSE2 |
37 | int8x16 ca = bit_xor(a, 0x80); |
38 | int8x16 cb = bit_xor(b, 0x80); |
39 | int8x16 r = _mm_min_epu8(ca.native(), cb.native()); |
40 | return bit_xor(r, 0x80); |
41 | #elif SIMDPP_USE_NEON |
42 | return vminq_s8(a.native(), b.native()); |
43 | #elif SIMDPP_USE_ALTIVEC |
44 | return vec_min(a.native(), b.native()); |
45 | #elif SIMDPP_USE_MSA |
46 | return __msa_min_s_b(a.native(), b.native()); |
47 | #endif |
48 | } |
49 | |
50 | #if SIMDPP_USE_AVX2 |
51 | static SIMDPP_INL |
52 | int8x32 i_min(const int8x32& a, const int8x32& b) |
53 | { |
54 | return _mm256_min_epi8(a.native(), b.native()); |
55 | } |
56 | #endif |
57 | |
58 | #if SIMDPP_USE_AVX512BW |
59 | SIMDPP_INL int8<64> i_min(const int8<64>& a, const int8<64>& b) |
60 | { |
61 | return _mm512_min_epi8(a.native(), b.native()); |
62 | } |
63 | #endif |
64 | |
65 | template<unsigned N> SIMDPP_INL |
66 | int8<N> i_min(const int8<N>& a, const int8<N>& b) |
67 | { |
68 | SIMDPP_VEC_ARRAY_IMPL2(int8<N>, i_min, a, b); |
69 | } |
70 | |
71 | // ----------------------------------------------------------------------------- |
72 | |
73 | static SIMDPP_INL |
74 | uint8x16 i_min(const uint8x16& a, const uint8x16& b) |
75 | { |
76 | #if SIMDPP_USE_NULL |
77 | return detail::null::min(a, b); |
78 | #elif SIMDPP_USE_SSE2 |
79 | return _mm_min_epu8(a.native(), b.native()); |
80 | #elif SIMDPP_USE_NEON |
81 | return vminq_u8(a.native(), b.native()); |
82 | #elif SIMDPP_USE_ALTIVEC |
83 | return vec_min(a.native(), b.native()); |
84 | #elif SIMDPP_USE_MSA |
85 | return __msa_min_u_b(a.native(), b.native()); |
86 | #endif |
87 | } |
88 | |
89 | #if SIMDPP_USE_AVX2 |
90 | static SIMDPP_INL |
91 | uint8x32 i_min(const uint8x32& a, const uint8x32& b) |
92 | { |
93 | return _mm256_min_epu8(a.native(), b.native()); |
94 | } |
95 | #endif |
96 | |
97 | #if SIMDPP_USE_AVX512BW |
98 | SIMDPP_INL uint8<64> i_min(const uint8<64>& a, const uint8<64>& b) |
99 | { |
100 | return _mm512_min_epu8(a.native(), b.native()); |
101 | } |
102 | #endif |
103 | |
104 | template<unsigned N> SIMDPP_INL |
105 | uint8<N> i_min(const uint8<N>& a, const uint8<N>& b) |
106 | { |
107 | SIMDPP_VEC_ARRAY_IMPL2(uint8<N>, i_min, a, b); |
108 | } |
109 | |
110 | // ----------------------------------------------------------------------------- |
111 | |
112 | static SIMDPP_INL |
113 | int16x8 i_min(const int16x8& a, const int16x8& b) |
114 | { |
115 | #if SIMDPP_USE_NULL |
116 | return detail::null::min(a, b); |
117 | #elif SIMDPP_USE_SSE2 |
118 | return _mm_min_epi16(a.native(), b.native()); |
119 | #elif SIMDPP_USE_NEON |
120 | return vminq_s16(a.native(), b.native()); |
121 | #elif SIMDPP_USE_ALTIVEC |
122 | return vec_min(a.native(), b.native()); |
123 | #elif SIMDPP_USE_MSA |
124 | return __msa_min_s_h(a.native(), b.native()); |
125 | #endif |
126 | } |
127 | |
128 | #if SIMDPP_USE_AVX2 |
129 | static SIMDPP_INL |
130 | int16x16 i_min(const int16x16& a, const int16x16& b) |
131 | { |
132 | return _mm256_min_epi16(a.native(), b.native()); |
133 | } |
134 | #endif |
135 | |
136 | #if SIMDPP_USE_AVX512BW |
137 | SIMDPP_INL int16<32> i_min(const int16<32>& a, const int16<32>& b) |
138 | { |
139 | return _mm512_min_epi16(a.native(), b.native()); |
140 | } |
141 | #endif |
142 | |
143 | template<unsigned N> SIMDPP_INL |
144 | int16<N> i_min(const int16<N>& a, const int16<N>& b) |
145 | { |
146 | SIMDPP_VEC_ARRAY_IMPL2(int16<N>, i_min, a, b); |
147 | } |
148 | |
149 | // ----------------------------------------------------------------------------- |
150 | |
151 | static SIMDPP_INL |
152 | uint16x8 i_min(const uint16x8& a, const uint16x8& b) |
153 | { |
154 | #if SIMDPP_USE_NULL |
155 | return detail::null::min(a, b); |
156 | #elif SIMDPP_USE_SSE4_1 |
157 | return _mm_min_epu16(a.native(), b.native()); |
158 | #elif SIMDPP_USE_SSE2 |
159 | int16x8 ca = bit_xor(a, 0x8000); |
160 | int16x8 cb = bit_xor(b, 0x8000); |
161 | int16x8 r = _mm_min_epi16(ca.native(), cb.native()); |
162 | return bit_xor(r, 0x8000); |
163 | #elif SIMDPP_USE_NEON |
164 | return vminq_u16(a.native(), b.native()); |
165 | #elif SIMDPP_USE_ALTIVEC |
166 | return vec_min(a.native(), b.native()); |
167 | #elif SIMDPP_USE_MSA |
168 | return __msa_min_u_h(a.native(), b.native()); |
169 | #endif |
170 | } |
171 | |
172 | #if SIMDPP_USE_AVX2 |
173 | static SIMDPP_INL |
174 | uint16x16 i_min(const uint16x16& a, const uint16x16& b) |
175 | { |
176 | return _mm256_min_epu16(a.native(), b.native()); |
177 | } |
178 | #endif |
179 | |
180 | #if SIMDPP_USE_AVX512BW |
181 | SIMDPP_INL uint16<32> i_min(const uint16<32>& a, const uint16<32>& b) |
182 | { |
183 | return _mm512_min_epu16(a.native(), b.native()); |
184 | } |
185 | #endif |
186 | |
187 | template<unsigned N> SIMDPP_INL |
188 | uint16<N> i_min(const uint16<N>& a, const uint16<N>& b) |
189 | { |
190 | SIMDPP_VEC_ARRAY_IMPL2(uint16<N>, i_min, a, b); |
191 | } |
192 | |
193 | // ----------------------------------------------------------------------------- |
194 | |
195 | static SIMDPP_INL |
196 | int32x4 i_min(const int32x4& a, const int32x4& b) |
197 | { |
198 | #if SIMDPP_USE_NULL |
199 | return detail::null::min(a, b); |
200 | #elif SIMDPP_USE_SSE4_1 |
201 | return _mm_min_epi32(a.native(), b.native()); |
202 | #elif SIMDPP_USE_SSE2 |
203 | mask_int32x4 mask = cmp_lt(a, b); |
204 | return blend(a, b, mask); |
205 | #elif SIMDPP_USE_NEON |
206 | return vminq_s32(a.native(), b.native()); |
207 | #elif SIMDPP_USE_ALTIVEC |
208 | return vec_min(a.native(), b.native()); |
209 | #elif SIMDPP_USE_MSA |
210 | return __msa_min_s_w(a.native(), b.native()); |
211 | #endif |
212 | } |
213 | |
214 | #if SIMDPP_USE_AVX2 |
215 | static SIMDPP_INL |
216 | int32x8 i_min(const int32x8& a, const int32x8& b) |
217 | { |
218 | return _mm256_min_epi32(a.native(), b.native()); |
219 | } |
220 | #endif |
221 | |
222 | #if SIMDPP_USE_AVX512F |
223 | static SIMDPP_INL |
224 | int32<16> i_min(const int32<16>& a, const int32<16>& b) |
225 | { |
226 | return _mm512_min_epi32(a.native(), b.native()); |
227 | } |
228 | #endif |
229 | |
230 | template<unsigned N> SIMDPP_INL |
231 | int32<N> i_min(const int32<N>& a, const int32<N>& b) |
232 | { |
233 | SIMDPP_VEC_ARRAY_IMPL2(int32<N>, i_min, a, b); |
234 | } |
235 | |
236 | // ----------------------------------------------------------------------------- |
237 | |
238 | static SIMDPP_INL |
239 | uint32x4 i_min(const uint32x4& a, const uint32x4& b) |
240 | { |
241 | #if SIMDPP_USE_NULL |
242 | return detail::null::min(a, b); |
243 | #elif SIMDPP_USE_SSE4_1 |
244 | return _mm_min_epu32(a.native(), b.native()); |
245 | #elif SIMDPP_USE_SSE2 |
246 | mask_int32x4 mask = cmp_lt(a, b); |
247 | return blend(a, b, mask); |
248 | #elif SIMDPP_USE_NEON |
249 | return vminq_u32(a.native(), b.native()); |
250 | #elif SIMDPP_USE_ALTIVEC |
251 | return vec_min(a.native(), b.native()); |
252 | #elif SIMDPP_USE_MSA |
253 | return __msa_min_u_w(a.native(), b.native()); |
254 | #endif |
255 | } |
256 | |
257 | #if SIMDPP_USE_AVX2 |
258 | static SIMDPP_INL |
259 | uint32x8 i_min(const uint32x8& a, const uint32x8& b) |
260 | { |
261 | return _mm256_min_epu32(a.native(), b.native()); |
262 | } |
263 | #endif |
264 | |
265 | #if SIMDPP_USE_AVX512F |
266 | static SIMDPP_INL |
267 | uint32<16> i_min(const uint32<16>& a, const uint32<16>& b) |
268 | { |
269 | return _mm512_min_epu32(a.native(), b.native()); |
270 | } |
271 | #endif |
272 | |
273 | template<unsigned N> SIMDPP_INL |
274 | uint32<N> i_min(const uint32<N>& a, const uint32<N>& b) |
275 | { |
276 | SIMDPP_VEC_ARRAY_IMPL2(uint32<N>, i_min, a, b); |
277 | } |
278 | |
279 | // ----------------------------------------------------------------------------- |
280 | |
281 | static SIMDPP_INL |
282 | int64x2 i_min(const int64x2& a, const int64x2& b) |
283 | { |
284 | #if SIMDPP_USE_AVX512VL |
285 | return _mm_min_epi64(a.native(), b.native()); |
286 | #elif SIMDPP_USE_AVX2 || SIMDPP_USE_NEON64 |
287 | mask_int64x2 mask = cmp_lt(a, b); |
288 | return blend(a, b, mask); |
289 | #elif SIMDPP_USE_VSX_207 |
290 | return vec_min(a.native(), b.native()); |
291 | #elif SIMDPP_USE_MSA |
292 | return __msa_min_s_d(a.native(), b.native()); |
293 | #elif SIMDPP_USE_NULL || SIMDPP_USE_ALTIVEC |
294 | return detail::null::min(a, b); |
295 | #else |
296 | return SIMDPP_NOT_IMPLEMENTED2(a, b); |
297 | #endif |
298 | } |
299 | |
300 | #if SIMDPP_USE_AVX2 |
301 | static SIMDPP_INL |
302 | int64x4 i_min(const int64x4& a, const int64x4& b) |
303 | { |
304 | #if SIMDPP_USE_AVX512VL |
305 | return _mm256_min_epi64(a.native(), b.native()); |
306 | #else |
307 | mask_int64x4 mask = cmp_lt(a, b); |
308 | return blend(a, b, mask); |
309 | #endif |
310 | } |
311 | #endif |
312 | |
313 | #if SIMDPP_USE_AVX512F |
314 | static SIMDPP_INL |
315 | int64<8> i_min(const int64<8>& a, const int64<8>& b) |
316 | { |
317 | return _mm512_min_epi64(a.native(), b.native()); |
318 | } |
319 | #endif |
320 | |
321 | template<unsigned N> SIMDPP_INL |
322 | int64<N> i_min(const int64<N>& a, const int64<N>& b) |
323 | { |
324 | SIMDPP_VEC_ARRAY_IMPL2(int64<N>, i_min, a, b); |
325 | } |
326 | |
327 | // ----------------------------------------------------------------------------- |
328 | |
329 | static SIMDPP_INL |
330 | uint64x2 i_min(const uint64x2& a, const uint64x2& b) |
331 | { |
332 | #if SIMDPP_USE_AVX512VL |
333 | return _mm_min_epu64(a.native(), b.native()); |
334 | #elif SIMDPP_USE_AVX2 || SIMDPP_USE_NEON64 |
335 | mask_int64x2 mask = cmp_lt(a, b); |
336 | return blend(a, b, mask); |
337 | #elif SIMDPP_USE_VSX_207 |
338 | return vec_min(a.native(), b.native()); |
339 | #elif SIMDPP_USE_MSA |
340 | return __msa_min_u_d(a.native(), b.native()); |
341 | #elif SIMDPP_USE_NULL || SIMDPP_USE_ALTIVEC |
342 | return detail::null::min(a, b); |
343 | #else |
344 | return SIMDPP_NOT_IMPLEMENTED2(a, b); |
345 | #endif |
346 | } |
347 | |
348 | #if SIMDPP_USE_AVX2 |
349 | static SIMDPP_INL |
350 | uint64x4 i_min(const uint64x4& a, const uint64x4& b) |
351 | { |
352 | #if SIMDPP_USE_AVX512VL |
353 | return _mm256_min_epu64(a.native(), b.native()); |
354 | #else |
355 | mask_int64x4 mask = cmp_lt(a, b); |
356 | return blend(a, b, mask); |
357 | #endif |
358 | } |
359 | #endif |
360 | |
361 | #if SIMDPP_USE_AVX512F |
362 | static SIMDPP_INL |
363 | uint64<8> i_min(const uint64<8>& a, const uint64<8>& b) |
364 | { |
365 | return _mm512_min_epu64(a.native(), b.native()); |
366 | } |
367 | #endif |
368 | |
369 | template<unsigned N> SIMDPP_INL |
370 | uint64<N> i_min(const uint64<N>& a, const uint64<N>& b) |
371 | { |
372 | SIMDPP_VEC_ARRAY_IMPL2(uint64<N>, i_min, a, b); |
373 | } |
374 | |
375 | } // namespace insn |
376 | } // namespace detail |
377 | } // namespace SIMDPP_ARCH_NAMESPACE |
378 | } // namespace simdpp |
379 | |
380 | #endif |
381 | |
382 | |