|
0
|
1 /* Copyright (C) 2002, 2003, 2004, 2009 Free Software Foundation, Inc.
|
|
|
2
|
|
|
3 This file is part of GCC.
|
|
|
4
|
|
|
5 GCC is free software; you can redistribute it and/or modify it
|
|
|
6 under the terms of the GNU General Public License as published
|
|
|
7 by the Free Software Foundation; either version 3, or (at your
|
|
|
8 option) any later version.
|
|
|
9
|
|
|
10 GCC is distributed in the hope that it will be useful, but WITHOUT
|
|
|
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
|
12 or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
|
13 License for more details.
|
|
|
14
|
|
|
15 Under Section 7 of GPL version 3, you are granted additional
|
|
|
16 permissions described in the GCC Runtime Library Exception, version
|
|
|
17 3.1, as published by the Free Software Foundation.
|
|
|
18
|
|
|
19 You should have received a copy of the GNU General Public License and
|
|
|
20 a copy of the GCC Runtime Library Exception along with this program;
|
|
|
21 see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
|
|
|
22 <http://www.gnu.org/licenses/>. */
|
|
|
23
|
|
|
24 #ifndef _MMINTRIN_H_INCLUDED
|
|
|
25 #define _MMINTRIN_H_INCLUDED
|
|
|
26
|
|
|
27 /* The data type intended for user use. */
|
|
|
28 typedef unsigned long long __m64, __int64;
|
|
|
29
|
|
|
30 /* Internal data types for implementing the intrinsics. */
|
|
|
31 typedef int __v2si __attribute__ ((vector_size (8)));
|
|
|
32 typedef short __v4hi __attribute__ ((vector_size (8)));
|
|
|
33 typedef char __v8qi __attribute__ ((vector_size (8)));
|
|
|
34
|
|
|
35 /* "Convert" __m64 and __int64 into each other. */
|
|
|
36 static __inline __m64
|
|
|
37 _mm_cvtsi64_m64 (__int64 __i)
|
|
|
38 {
|
|
|
39 return __i;
|
|
|
40 }
|
|
|
41
|
|
|
42 static __inline __int64
|
|
|
43 _mm_cvtm64_si64 (__m64 __i)
|
|
|
44 {
|
|
|
45 return __i;
|
|
|
46 }
|
|
|
47
|
|
|
48 static __inline int
|
|
|
49 _mm_cvtsi64_si32 (__int64 __i)
|
|
|
50 {
|
|
|
51 return __i;
|
|
|
52 }
|
|
|
53
|
|
|
54 static __inline __int64
|
|
|
55 _mm_cvtsi32_si64 (int __i)
|
|
|
56 {
|
|
|
57 return __i;
|
|
|
58 }
|
|
|
59
|
|
|
60 /* Pack the four 16-bit values from M1 into the lower four 8-bit values of
|
|
|
61 the result, and the four 16-bit values from M2 into the upper four 8-bit
|
|
|
62 values of the result, all with signed saturation. */
|
|
|
63 static __inline __m64
|
|
|
64 _mm_packs_pi16 (__m64 __m1, __m64 __m2)
|
|
|
65 {
|
|
|
66 return (__m64) __builtin_arm_wpackhss ((__v4hi)__m1, (__v4hi)__m2);
|
|
|
67 }
|
|
|
68
|
|
|
69 /* Pack the two 32-bit values from M1 in to the lower two 16-bit values of
|
|
|
70 the result, and the two 32-bit values from M2 into the upper two 16-bit
|
|
|
71 values of the result, all with signed saturation. */
|
|
|
72 static __inline __m64
|
|
|
73 _mm_packs_pi32 (__m64 __m1, __m64 __m2)
|
|
|
74 {
|
|
|
75 return (__m64) __builtin_arm_wpackwss ((__v2si)__m1, (__v2si)__m2);
|
|
|
76 }
|
|
|
77
|
|
|
78 /* Copy the 64-bit value from M1 into the lower 32-bits of the result, and
|
|
|
79 the 64-bit value from M2 into the upper 32-bits of the result, all with
|
|
|
80 signed saturation for values that do not fit exactly into 32-bits. */
|
|
|
81 static __inline __m64
|
|
|
82 _mm_packs_pi64 (__m64 __m1, __m64 __m2)
|
|
|
83 {
|
|
|
84 return (__m64) __builtin_arm_wpackdss ((long long)__m1, (long long)__m2);
|
|
|
85 }
|
|
|
86
|
|
|
87 /* Pack the four 16-bit values from M1 into the lower four 8-bit values of
|
|
|
88 the result, and the four 16-bit values from M2 into the upper four 8-bit
|
|
|
89 values of the result, all with unsigned saturation. */
|
|
|
90 static __inline __m64
|
|
|
91 _mm_packs_pu16 (__m64 __m1, __m64 __m2)
|
|
|
92 {
|
|
|
93 return (__m64) __builtin_arm_wpackhus ((__v4hi)__m1, (__v4hi)__m2);
|
|
|
94 }
|
|
|
95
|
|
|
96 /* Pack the two 32-bit values from M1 into the lower two 16-bit values of
|
|
|
97 the result, and the two 32-bit values from M2 into the upper two 16-bit
|
|
|
98 values of the result, all with unsigned saturation. */
|
|
|
99 static __inline __m64
|
|
|
100 _mm_packs_pu32 (__m64 __m1, __m64 __m2)
|
|
|
101 {
|
|
|
102 return (__m64) __builtin_arm_wpackwus ((__v2si)__m1, (__v2si)__m2);
|
|
|
103 }
|
|
|
104
|
|
|
105 /* Copy the 64-bit value from M1 into the lower 32-bits of the result, and
|
|
|
106 the 64-bit value from M2 into the upper 32-bits of the result, all with
|
|
|
107 unsigned saturation for values that do not fit exactly into 32-bits. */
|
|
|
108 static __inline __m64
|
|
|
109 _mm_packs_pu64 (__m64 __m1, __m64 __m2)
|
|
|
110 {
|
|
|
111 return (__m64) __builtin_arm_wpackdus ((long long)__m1, (long long)__m2);
|
|
|
112 }
|
|
|
113
|
|
|
114 /* Interleave the four 8-bit values from the high half of M1 with the four
|
|
|
115 8-bit values from the high half of M2. */
|
|
|
116 static __inline __m64
|
|
|
117 _mm_unpackhi_pi8 (__m64 __m1, __m64 __m2)
|
|
|
118 {
|
|
|
119 return (__m64) __builtin_arm_wunpckihb ((__v8qi)__m1, (__v8qi)__m2);
|
|
|
120 }
|
|
|
121
|
|
|
122 /* Interleave the two 16-bit values from the high half of M1 with the two
|
|
|
123 16-bit values from the high half of M2. */
|
|
|
124 static __inline __m64
|
|
|
125 _mm_unpackhi_pi16 (__m64 __m1, __m64 __m2)
|
|
|
126 {
|
|
|
127 return (__m64) __builtin_arm_wunpckihh ((__v4hi)__m1, (__v4hi)__m2);
|
|
|
128 }
|
|
|
129
|
|
|
130 /* Interleave the 32-bit value from the high half of M1 with the 32-bit
|
|
|
131 value from the high half of M2. */
|
|
|
132 static __inline __m64
|
|
|
133 _mm_unpackhi_pi32 (__m64 __m1, __m64 __m2)
|
|
|
134 {
|
|
|
135 return (__m64) __builtin_arm_wunpckihw ((__v2si)__m1, (__v2si)__m2);
|
|
|
136 }
|
|
|
137
|
|
|
138 /* Interleave the four 8-bit values from the low half of M1 with the four
|
|
|
139 8-bit values from the low half of M2. */
|
|
|
140 static __inline __m64
|
|
|
141 _mm_unpacklo_pi8 (__m64 __m1, __m64 __m2)
|
|
|
142 {
|
|
|
143 return (__m64) __builtin_arm_wunpckilb ((__v8qi)__m1, (__v8qi)__m2);
|
|
|
144 }
|
|
|
145
|
|
|
146 /* Interleave the two 16-bit values from the low half of M1 with the two
|
|
|
147 16-bit values from the low half of M2. */
|
|
|
148 static __inline __m64
|
|
|
149 _mm_unpacklo_pi16 (__m64 __m1, __m64 __m2)
|
|
|
150 {
|
|
|
151 return (__m64) __builtin_arm_wunpckilh ((__v4hi)__m1, (__v4hi)__m2);
|
|
|
152 }
|
|
|
153
|
|
|
154 /* Interleave the 32-bit value from the low half of M1 with the 32-bit
|
|
|
155 value from the low half of M2. */
|
|
|
156 static __inline __m64
|
|
|
157 _mm_unpacklo_pi32 (__m64 __m1, __m64 __m2)
|
|
|
158 {
|
|
|
159 return (__m64) __builtin_arm_wunpckilw ((__v2si)__m1, (__v2si)__m2);
|
|
|
160 }
|
|
|
161
|
|
|
162 /* Take the four 8-bit values from the low half of M1, sign extend them,
|
|
|
163 and return the result as a vector of four 16-bit quantities. */
|
|
|
164 static __inline __m64
|
|
|
165 _mm_unpackel_pi8 (__m64 __m1)
|
|
|
166 {
|
|
|
167 return (__m64) __builtin_arm_wunpckelsb ((__v8qi)__m1);
|
|
|
168 }
|
|
|
169
|
|
|
170 /* Take the two 16-bit values from the low half of M1, sign extend them,
|
|
|
171 and return the result as a vector of two 32-bit quantities. */
|
|
|
172 static __inline __m64
|
|
|
173 _mm_unpackel_pi16 (__m64 __m1)
|
|
|
174 {
|
|
|
175 return (__m64) __builtin_arm_wunpckelsh ((__v4hi)__m1);
|
|
|
176 }
|
|
|
177
|
|
|
178 /* Take the 32-bit value from the low half of M1, and return it sign extended
|
|
|
179 to 64 bits. */
|
|
|
180 static __inline __m64
|
|
|
181 _mm_unpackel_pi32 (__m64 __m1)
|
|
|
182 {
|
|
|
183 return (__m64) __builtin_arm_wunpckelsw ((__v2si)__m1);
|
|
|
184 }
|
|
|
185
|
|
|
186 /* Take the four 8-bit values from the high half of M1, sign extend them,
|
|
|
187 and return the result as a vector of four 16-bit quantities. */
|
|
|
188 static __inline __m64
|
|
|
189 _mm_unpackeh_pi8 (__m64 __m1)
|
|
|
190 {
|
|
|
191 return (__m64) __builtin_arm_wunpckehsb ((__v8qi)__m1);
|
|
|
192 }
|
|
|
193
|
|
|
194 /* Take the two 16-bit values from the high half of M1, sign extend them,
|
|
|
195 and return the result as a vector of two 32-bit quantities. */
|
|
|
196 static __inline __m64
|
|
|
197 _mm_unpackeh_pi16 (__m64 __m1)
|
|
|
198 {
|
|
|
199 return (__m64) __builtin_arm_wunpckehsh ((__v4hi)__m1);
|
|
|
200 }
|
|
|
201
|
|
|
202 /* Take the 32-bit value from the high half of M1, and return it sign extended
|
|
|
203 to 64 bits. */
|
|
|
204 static __inline __m64
|
|
|
205 _mm_unpackeh_pi32 (__m64 __m1)
|
|
|
206 {
|
|
|
207 return (__m64) __builtin_arm_wunpckehsw ((__v2si)__m1);
|
|
|
208 }
|
|
|
209
|
|
|
210 /* Take the four 8-bit values from the low half of M1, zero extend them,
|
|
|
211 and return the result as a vector of four 16-bit quantities. */
|
|
|
212 static __inline __m64
|
|
|
213 _mm_unpackel_pu8 (__m64 __m1)
|
|
|
214 {
|
|
|
215 return (__m64) __builtin_arm_wunpckelub ((__v8qi)__m1);
|
|
|
216 }
|
|
|
217
|
|
|
218 /* Take the two 16-bit values from the low half of M1, zero extend them,
|
|
|
219 and return the result as a vector of two 32-bit quantities. */
|
|
|
220 static __inline __m64
|
|
|
221 _mm_unpackel_pu16 (__m64 __m1)
|
|
|
222 {
|
|
|
223 return (__m64) __builtin_arm_wunpckeluh ((__v4hi)__m1);
|
|
|
224 }
|
|
|
225
|
|
|
226 /* Take the 32-bit value from the low half of M1, and return it zero extended
|
|
|
227 to 64 bits. */
|
|
|
228 static __inline __m64
|
|
|
229 _mm_unpackel_pu32 (__m64 __m1)
|
|
|
230 {
|
|
|
231 return (__m64) __builtin_arm_wunpckeluw ((__v2si)__m1);
|
|
|
232 }
|
|
|
233
|
|
|
234 /* Take the four 8-bit values from the high half of M1, zero extend them,
|
|
|
235 and return the result as a vector of four 16-bit quantities. */
|
|
|
236 static __inline __m64
|
|
|
237 _mm_unpackeh_pu8 (__m64 __m1)
|
|
|
238 {
|
|
|
239 return (__m64) __builtin_arm_wunpckehub ((__v8qi)__m1);
|
|
|
240 }
|
|
|
241
|
|
|
242 /* Take the two 16-bit values from the high half of M1, zero extend them,
|
|
|
243 and return the result as a vector of two 32-bit quantities. */
|
|
|
244 static __inline __m64
|
|
|
245 _mm_unpackeh_pu16 (__m64 __m1)
|
|
|
246 {
|
|
|
247 return (__m64) __builtin_arm_wunpckehuh ((__v4hi)__m1);
|
|
|
248 }
|
|
|
249
|
|
|
250 /* Take the 32-bit value from the high half of M1, and return it zero extended
|
|
|
251 to 64 bits. */
|
|
|
252 static __inline __m64
|
|
|
253 _mm_unpackeh_pu32 (__m64 __m1)
|
|
|
254 {
|
|
|
255 return (__m64) __builtin_arm_wunpckehuw ((__v2si)__m1);
|
|
|
256 }
|
|
|
257
|
|
|
258 /* Add the 8-bit values in M1 to the 8-bit values in M2. */
|
|
|
259 static __inline __m64
|
|
|
260 _mm_add_pi8 (__m64 __m1, __m64 __m2)
|
|
|
261 {
|
|
|
262 return (__m64) __builtin_arm_waddb ((__v8qi)__m1, (__v8qi)__m2);
|
|
|
263 }
|
|
|
264
|
|
|
265 /* Add the 16-bit values in M1 to the 16-bit values in M2. */
|
|
|
266 static __inline __m64
|
|
|
267 _mm_add_pi16 (__m64 __m1, __m64 __m2)
|
|
|
268 {
|
|
|
269 return (__m64) __builtin_arm_waddh ((__v4hi)__m1, (__v4hi)__m2);
|
|
|
270 }
|
|
|
271
|
|
|
272 /* Add the 32-bit values in M1 to the 32-bit values in M2. */
|
|
|
273 static __inline __m64
|
|
|
274 _mm_add_pi32 (__m64 __m1, __m64 __m2)
|
|
|
275 {
|
|
|
276 return (__m64) __builtin_arm_waddw ((__v2si)__m1, (__v2si)__m2);
|
|
|
277 }
|
|
|
278
|
|
|
279 /* Add the 8-bit values in M1 to the 8-bit values in M2 using signed
|
|
|
280 saturated arithmetic. */
|
|
|
281 static __inline __m64
|
|
|
282 _mm_adds_pi8 (__m64 __m1, __m64 __m2)
|
|
|
283 {
|
|
|
284 return (__m64) __builtin_arm_waddbss ((__v8qi)__m1, (__v8qi)__m2);
|
|
|
285 }
|
|
|
286
|
|
|
287 /* Add the 16-bit values in M1 to the 16-bit values in M2 using signed
|
|
|
288 saturated arithmetic. */
|
|
|
289 static __inline __m64
|
|
|
290 _mm_adds_pi16 (__m64 __m1, __m64 __m2)
|
|
|
291 {
|
|
|
292 return (__m64) __builtin_arm_waddhss ((__v4hi)__m1, (__v4hi)__m2);
|
|
|
293 }
|
|
|
294
|
|
|
295 /* Add the 32-bit values in M1 to the 32-bit values in M2 using signed
|
|
|
296 saturated arithmetic. */
|
|
|
297 static __inline __m64
|
|
|
298 _mm_adds_pi32 (__m64 __m1, __m64 __m2)
|
|
|
299 {
|
|
|
300 return (__m64) __builtin_arm_waddwss ((__v2si)__m1, (__v2si)__m2);
|
|
|
301 }
|
|
|
302
|
|
|
303 /* Add the 8-bit values in M1 to the 8-bit values in M2 using unsigned
|
|
|
304 saturated arithmetic. */
|
|
|
305 static __inline __m64
|
|
|
306 _mm_adds_pu8 (__m64 __m1, __m64 __m2)
|
|
|
307 {
|
|
|
308 return (__m64) __builtin_arm_waddbus ((__v8qi)__m1, (__v8qi)__m2);
|
|
|
309 }
|
|
|
310
|
|
|
311 /* Add the 16-bit values in M1 to the 16-bit values in M2 using unsigned
|
|
|
312 saturated arithmetic. */
|
|
|
313 static __inline __m64
|
|
|
314 _mm_adds_pu16 (__m64 __m1, __m64 __m2)
|
|
|
315 {
|
|
|
316 return (__m64) __builtin_arm_waddhus ((__v4hi)__m1, (__v4hi)__m2);
|
|
|
317 }
|
|
|
318
|
|
|
319 /* Add the 32-bit values in M1 to the 32-bit values in M2 using unsigned
|
|
|
320 saturated arithmetic. */
|
|
|
321 static __inline __m64
|
|
|
322 _mm_adds_pu32 (__m64 __m1, __m64 __m2)
|
|
|
323 {
|
|
|
324 return (__m64) __builtin_arm_waddwus ((__v2si)__m1, (__v2si)__m2);
|
|
|
325 }
|
|
|
326
|
|
|
327 /* Subtract the 8-bit values in M2 from the 8-bit values in M1. */
|
|
|
328 static __inline __m64
|
|
|
329 _mm_sub_pi8 (__m64 __m1, __m64 __m2)
|
|
|
330 {
|
|
|
331 return (__m64) __builtin_arm_wsubb ((__v8qi)__m1, (__v8qi)__m2);
|
|
|
332 }
|
|
|
333
|
|
|
334 /* Subtract the 16-bit values in M2 from the 16-bit values in M1. */
|
|
|
335 static __inline __m64
|
|
|
336 _mm_sub_pi16 (__m64 __m1, __m64 __m2)
|
|
|
337 {
|
|
|
338 return (__m64) __builtin_arm_wsubh ((__v4hi)__m1, (__v4hi)__m2);
|
|
|
339 }
|
|
|
340
|
|
|
341 /* Subtract the 32-bit values in M2 from the 32-bit values in M1. */
|
|
|
342 static __inline __m64
|
|
|
343 _mm_sub_pi32 (__m64 __m1, __m64 __m2)
|
|
|
344 {
|
|
|
345 return (__m64) __builtin_arm_wsubw ((__v2si)__m1, (__v2si)__m2);
|
|
|
346 }
|
|
|
347
|
|
|
348 /* Subtract the 8-bit values in M2 from the 8-bit values in M1 using signed
|
|
|
349 saturating arithmetic. */
|
|
|
350 static __inline __m64
|
|
|
351 _mm_subs_pi8 (__m64 __m1, __m64 __m2)
|
|
|
352 {
|
|
|
353 return (__m64) __builtin_arm_wsubbss ((__v8qi)__m1, (__v8qi)__m2);
|
|
|
354 }
|
|
|
355
|
|
|
356 /* Subtract the 16-bit values in M2 from the 16-bit values in M1 using
|
|
|
357 signed saturating arithmetic. */
|
|
|
358 static __inline __m64
|
|
|
359 _mm_subs_pi16 (__m64 __m1, __m64 __m2)
|
|
|
360 {
|
|
|
361 return (__m64) __builtin_arm_wsubhss ((__v4hi)__m1, (__v4hi)__m2);
|
|
|
362 }
|
|
|
363
|
|
|
364 /* Subtract the 32-bit values in M2 from the 32-bit values in M1 using
|
|
|
365 signed saturating arithmetic. */
|
|
|
366 static __inline __m64
|
|
|
367 _mm_subs_pi32 (__m64 __m1, __m64 __m2)
|
|
|
368 {
|
|
|
369 return (__m64) __builtin_arm_wsubwss ((__v2si)__m1, (__v2si)__m2);
|
|
|
370 }
|
|
|
371
|
|
|
372 /* Subtract the 8-bit values in M2 from the 8-bit values in M1 using
|
|
|
373 unsigned saturating arithmetic. */
|
|
|
374 static __inline __m64
|
|
|
375 _mm_subs_pu8 (__m64 __m1, __m64 __m2)
|
|
|
376 {
|
|
|
377 return (__m64) __builtin_arm_wsubbus ((__v8qi)__m1, (__v8qi)__m2);
|
|
|
378 }
|
|
|
379
|
|
|
380 /* Subtract the 16-bit values in M2 from the 16-bit values in M1 using
|
|
|
381 unsigned saturating arithmetic. */
|
|
|
382 static __inline __m64
|
|
|
383 _mm_subs_pu16 (__m64 __m1, __m64 __m2)
|
|
|
384 {
|
|
|
385 return (__m64) __builtin_arm_wsubhus ((__v4hi)__m1, (__v4hi)__m2);
|
|
|
386 }
|
|
|
387
|
|
|
388 /* Subtract the 32-bit values in M2 from the 32-bit values in M1 using
|
|
|
389 unsigned saturating arithmetic. */
|
|
|
390 static __inline __m64
|
|
|
391 _mm_subs_pu32 (__m64 __m1, __m64 __m2)
|
|
|
392 {
|
|
|
393 return (__m64) __builtin_arm_wsubwus ((__v2si)__m1, (__v2si)__m2);
|
|
|
394 }
|
|
|
395
|
|
|
396 /* Multiply four 16-bit values in M1 by four 16-bit values in M2 producing
|
|
|
397 four 32-bit intermediate results, which are then summed by pairs to
|
|
|
398 produce two 32-bit results. */
|
|
|
399 static __inline __m64
|
|
|
400 _mm_madd_pi16 (__m64 __m1, __m64 __m2)
|
|
|
401 {
|
|
|
402 return (__m64) __builtin_arm_wmadds ((__v4hi)__m1, (__v4hi)__m2);
|
|
|
403 }
|
|
|
404
|
|
|
405 /* Multiply four 16-bit values in M1 by four 16-bit values in M2 producing
|
|
|
406 four 32-bit intermediate results, which are then summed by pairs to
|
|
|
407 produce two 32-bit results. */
|
|
|
408 static __inline __m64
|
|
|
409 _mm_madd_pu16 (__m64 __m1, __m64 __m2)
|
|
|
410 {
|
|
|
411 return (__m64) __builtin_arm_wmaddu ((__v4hi)__m1, (__v4hi)__m2);
|
|
|
412 }
|
|
|
413
|
|
|
414 /* Multiply four signed 16-bit values in M1 by four signed 16-bit values in
|
|
|
415 M2 and produce the high 16 bits of the 32-bit results. */
|
|
|
416 static __inline __m64
|
|
|
417 _mm_mulhi_pi16 (__m64 __m1, __m64 __m2)
|
|
|
418 {
|
|
|
419 return (__m64) __builtin_arm_wmulsm ((__v4hi)__m1, (__v4hi)__m2);
|
|
|
420 }
|
|
|
421
|
|
|
422 /* Multiply four signed 16-bit values in M1 by four signed 16-bit values in
|
|
|
423 M2 and produce the high 16 bits of the 32-bit results. */
|
|
|
424 static __inline __m64
|
|
|
425 _mm_mulhi_pu16 (__m64 __m1, __m64 __m2)
|
|
|
426 {
|
|
|
427 return (__m64) __builtin_arm_wmulum ((__v4hi)__m1, (__v4hi)__m2);
|
|
|
428 }
|
|
|
429
|
|
|
430 /* Multiply four 16-bit values in M1 by four 16-bit values in M2 and produce
|
|
|
431 the low 16 bits of the results. */
|
|
|
432 static __inline __m64
|
|
|
433 _mm_mullo_pi16 (__m64 __m1, __m64 __m2)
|
|
|
434 {
|
|
|
435 return (__m64) __builtin_arm_wmulul ((__v4hi)__m1, (__v4hi)__m2);
|
|
|
436 }
|
|
|
437
|
|
|
438 /* Shift four 16-bit values in M left by COUNT. */
|
|
|
439 static __inline __m64
|
|
|
440 _mm_sll_pi16 (__m64 __m, __m64 __count)
|
|
|
441 {
|
|
|
442 return (__m64) __builtin_arm_wsllh ((__v4hi)__m, __count);
|
|
|
443 }
|
|
|
444
|
|
|
445 static __inline __m64
|
|
|
446 _mm_slli_pi16 (__m64 __m, int __count)
|
|
|
447 {
|
|
|
448 return (__m64) __builtin_arm_wsllhi ((__v4hi)__m, __count);
|
|
|
449 }
|
|
|
450
|
|
|
451 /* Shift two 32-bit values in M left by COUNT. */
|
|
|
452 static __inline __m64
|
|
|
453 _mm_sll_pi32 (__m64 __m, __m64 __count)
|
|
|
454 {
|
|
|
455 return (__m64) __builtin_arm_wsllw ((__v2si)__m, __count);
|
|
|
456 }
|
|
|
457
|
|
|
458 static __inline __m64
|
|
|
459 _mm_slli_pi32 (__m64 __m, int __count)
|
|
|
460 {
|
|
|
461 return (__m64) __builtin_arm_wsllwi ((__v2si)__m, __count);
|
|
|
462 }
|
|
|
463
|
|
|
464 /* Shift the 64-bit value in M left by COUNT. */
|
|
|
465 static __inline __m64
|
|
|
466 _mm_sll_si64 (__m64 __m, __m64 __count)
|
|
|
467 {
|
|
|
468 return (__m64) __builtin_arm_wslld (__m, __count);
|
|
|
469 }
|
|
|
470
|
|
|
471 static __inline __m64
|
|
|
472 _mm_slli_si64 (__m64 __m, int __count)
|
|
|
473 {
|
|
|
474 return (__m64) __builtin_arm_wslldi (__m, __count);
|
|
|
475 }
|
|
|
476
|
|
|
477 /* Shift four 16-bit values in M right by COUNT; shift in the sign bit. */
|
|
|
478 static __inline __m64
|
|
|
479 _mm_sra_pi16 (__m64 __m, __m64 __count)
|
|
|
480 {
|
|
|
481 return (__m64) __builtin_arm_wsrah ((__v4hi)__m, __count);
|
|
|
482 }
|
|
|
483
|
|
|
484 static __inline __m64
|
|
|
485 _mm_srai_pi16 (__m64 __m, int __count)
|
|
|
486 {
|
|
|
487 return (__m64) __builtin_arm_wsrahi ((__v4hi)__m, __count);
|
|
|
488 }
|
|
|
489
|
|
|
490 /* Shift two 32-bit values in M right by COUNT; shift in the sign bit. */
|
|
|
491 static __inline __m64
|
|
|
492 _mm_sra_pi32 (__m64 __m, __m64 __count)
|
|
|
493 {
|
|
|
494 return (__m64) __builtin_arm_wsraw ((__v2si)__m, __count);
|
|
|
495 }
|
|
|
496
|
|
|
497 static __inline __m64
|
|
|
498 _mm_srai_pi32 (__m64 __m, int __count)
|
|
|
499 {
|
|
|
500 return (__m64) __builtin_arm_wsrawi ((__v2si)__m, __count);
|
|
|
501 }
|
|
|
502
|
|
|
503 /* Shift the 64-bit value in M right by COUNT; shift in the sign bit. */
|
|
|
504 static __inline __m64
|
|
|
505 _mm_sra_si64 (__m64 __m, __m64 __count)
|
|
|
506 {
|
|
|
507 return (__m64) __builtin_arm_wsrad (__m, __count);
|
|
|
508 }
|
|
|
509
|
|
|
510 static __inline __m64
|
|
|
511 _mm_srai_si64 (__m64 __m, int __count)
|
|
|
512 {
|
|
|
513 return (__m64) __builtin_arm_wsradi (__m, __count);
|
|
|
514 }
|
|
|
515
|
|
|
516 /* Shift four 16-bit values in M right by COUNT; shift in zeros. */
|
|
|
517 static __inline __m64
|
|
|
518 _mm_srl_pi16 (__m64 __m, __m64 __count)
|
|
|
519 {
|
|
|
520 return (__m64) __builtin_arm_wsrlh ((__v4hi)__m, __count);
|
|
|
521 }
|
|
|
522
|
|
|
523 static __inline __m64
|
|
|
524 _mm_srli_pi16 (__m64 __m, int __count)
|
|
|
525 {
|
|
|
526 return (__m64) __builtin_arm_wsrlhi ((__v4hi)__m, __count);
|
|
|
527 }
|
|
|
528
|
|
|
529 /* Shift two 32-bit values in M right by COUNT; shift in zeros. */
|
|
|
530 static __inline __m64
|
|
|
531 _mm_srl_pi32 (__m64 __m, __m64 __count)
|
|
|
532 {
|
|
|
533 return (__m64) __builtin_arm_wsrlw ((__v2si)__m, __count);
|
|
|
534 }
|
|
|
535
|
|
|
536 static __inline __m64
|
|
|
537 _mm_srli_pi32 (__m64 __m, int __count)
|
|
|
538 {
|
|
|
539 return (__m64) __builtin_arm_wsrlwi ((__v2si)__m, __count);
|
|
|
540 }
|
|
|
541
|
|
|
542 /* Shift the 64-bit value in M left by COUNT; shift in zeros. */
|
|
|
543 static __inline __m64
|
|
|
544 _mm_srl_si64 (__m64 __m, __m64 __count)
|
|
|
545 {
|
|
|
546 return (__m64) __builtin_arm_wsrld (__m, __count);
|
|
|
547 }
|
|
|
548
|
|
|
549 static __inline __m64
|
|
|
550 _mm_srli_si64 (__m64 __m, int __count)
|
|
|
551 {
|
|
|
552 return (__m64) __builtin_arm_wsrldi (__m, __count);
|
|
|
553 }
|
|
|
554
|
|
|
555 /* Rotate four 16-bit values in M right by COUNT. */
|
|
|
556 static __inline __m64
|
|
|
557 _mm_ror_pi16 (__m64 __m, __m64 __count)
|
|
|
558 {
|
|
|
559 return (__m64) __builtin_arm_wrorh ((__v4hi)__m, __count);
|
|
|
560 }
|
|
|
561
|
|
|
562 static __inline __m64
|
|
|
563 _mm_rori_pi16 (__m64 __m, int __count)
|
|
|
564 {
|
|
|
565 return (__m64) __builtin_arm_wrorhi ((__v4hi)__m, __count);
|
|
|
566 }
|
|
|
567
|
|
|
568 /* Rotate two 32-bit values in M right by COUNT. */
|
|
|
569 static __inline __m64
|
|
|
570 _mm_ror_pi32 (__m64 __m, __m64 __count)
|
|
|
571 {
|
|
|
572 return (__m64) __builtin_arm_wrorw ((__v2si)__m, __count);
|
|
|
573 }
|
|
|
574
|
|
|
575 static __inline __m64
|
|
|
576 _mm_rori_pi32 (__m64 __m, int __count)
|
|
|
577 {
|
|
|
578 return (__m64) __builtin_arm_wrorwi ((__v2si)__m, __count);
|
|
|
579 }
|
|
|
580
|
|
|
581 /* Rotate two 64-bit values in M right by COUNT. */
|
|
|
582 static __inline __m64
|
|
|
583 _mm_ror_si64 (__m64 __m, __m64 __count)
|
|
|
584 {
|
|
|
585 return (__m64) __builtin_arm_wrord (__m, __count);
|
|
|
586 }
|
|
|
587
|
|
|
588 static __inline __m64
|
|
|
589 _mm_rori_si64 (__m64 __m, int __count)
|
|
|
590 {
|
|
|
591 return (__m64) __builtin_arm_wrordi (__m, __count);
|
|
|
592 }
|
|
|
593
|
|
|
594 /* Bit-wise AND the 64-bit values in M1 and M2. */
|
|
|
595 static __inline __m64
|
|
|
596 _mm_and_si64 (__m64 __m1, __m64 __m2)
|
|
|
597 {
|
|
|
598 return __builtin_arm_wand (__m1, __m2);
|
|
|
599 }
|
|
|
600
|
|
|
601 /* Bit-wise complement the 64-bit value in M1 and bit-wise AND it with the
|
|
|
602 64-bit value in M2. */
|
|
|
603 static __inline __m64
|
|
|
604 _mm_andnot_si64 (__m64 __m1, __m64 __m2)
|
|
|
605 {
|
|
|
606 return __builtin_arm_wandn (__m1, __m2);
|
|
|
607 }
|
|
|
608
|
|
|
609 /* Bit-wise inclusive OR the 64-bit values in M1 and M2. */
|
|
|
610 static __inline __m64
|
|
|
611 _mm_or_si64 (__m64 __m1, __m64 __m2)
|
|
|
612 {
|
|
|
613 return __builtin_arm_wor (__m1, __m2);
|
|
|
614 }
|
|
|
615
|
|
|
616 /* Bit-wise exclusive OR the 64-bit values in M1 and M2. */
|
|
|
617 static __inline __m64
|
|
|
618 _mm_xor_si64 (__m64 __m1, __m64 __m2)
|
|
|
619 {
|
|
|
620 return __builtin_arm_wxor (__m1, __m2);
|
|
|
621 }
|
|
|
622
|
|
|
623 /* Compare eight 8-bit values. The result of the comparison is 0xFF if the
|
|
|
624 test is true and zero if false. */
|
|
|
625 static __inline __m64
|
|
|
626 _mm_cmpeq_pi8 (__m64 __m1, __m64 __m2)
|
|
|
627 {
|
|
|
628 return (__m64) __builtin_arm_wcmpeqb ((__v8qi)__m1, (__v8qi)__m2);
|
|
|
629 }
|
|
|
630
|
|
|
631 static __inline __m64
|
|
|
632 _mm_cmpgt_pi8 (__m64 __m1, __m64 __m2)
|
|
|
633 {
|
|
|
634 return (__m64) __builtin_arm_wcmpgtsb ((__v8qi)__m1, (__v8qi)__m2);
|
|
|
635 }
|
|
|
636
|
|
|
637 static __inline __m64
|
|
|
638 _mm_cmpgt_pu8 (__m64 __m1, __m64 __m2)
|
|
|
639 {
|
|
|
640 return (__m64) __builtin_arm_wcmpgtub ((__v8qi)__m1, (__v8qi)__m2);
|
|
|
641 }
|
|
|
642
|
|
|
643 /* Compare four 16-bit values. The result of the comparison is 0xFFFF if
|
|
|
644 the test is true and zero if false. */
|
|
|
645 static __inline __m64
|
|
|
646 _mm_cmpeq_pi16 (__m64 __m1, __m64 __m2)
|
|
|
647 {
|
|
|
648 return (__m64) __builtin_arm_wcmpeqh ((__v4hi)__m1, (__v4hi)__m2);
|
|
|
649 }
|
|
|
650
|
|
|
651 static __inline __m64
|
|
|
652 _mm_cmpgt_pi16 (__m64 __m1, __m64 __m2)
|
|
|
653 {
|
|
|
654 return (__m64) __builtin_arm_wcmpgtsh ((__v4hi)__m1, (__v4hi)__m2);
|
|
|
655 }
|
|
|
656
|
|
|
657 static __inline __m64
|
|
|
658 _mm_cmpgt_pu16 (__m64 __m1, __m64 __m2)
|
|
|
659 {
|
|
|
660 return (__m64) __builtin_arm_wcmpgtuh ((__v4hi)__m1, (__v4hi)__m2);
|
|
|
661 }
|
|
|
662
|
|
|
663 /* Compare two 32-bit values. The result of the comparison is 0xFFFFFFFF if
|
|
|
664 the test is true and zero if false. */
|
|
|
665 static __inline __m64
|
|
|
666 _mm_cmpeq_pi32 (__m64 __m1, __m64 __m2)
|
|
|
667 {
|
|
|
668 return (__m64) __builtin_arm_wcmpeqw ((__v2si)__m1, (__v2si)__m2);
|
|
|
669 }
|
|
|
670
|
|
|
671 static __inline __m64
|
|
|
672 _mm_cmpgt_pi32 (__m64 __m1, __m64 __m2)
|
|
|
673 {
|
|
|
674 return (__m64) __builtin_arm_wcmpgtsw ((__v2si)__m1, (__v2si)__m2);
|
|
|
675 }
|
|
|
676
|
|
|
677 static __inline __m64
|
|
|
678 _mm_cmpgt_pu32 (__m64 __m1, __m64 __m2)
|
|
|
679 {
|
|
|
680 return (__m64) __builtin_arm_wcmpgtuw ((__v2si)__m1, (__v2si)__m2);
|
|
|
681 }
|
|
|
682
|
|
|
683 /* Element-wise multiplication of unsigned 16-bit values __B and __C, followed
|
|
|
684 by accumulate across all elements and __A. */
|
|
|
685 static __inline __m64
|
|
|
686 _mm_mac_pu16 (__m64 __A, __m64 __B, __m64 __C)
|
|
|
687 {
|
|
|
688 return __builtin_arm_wmacu (__A, (__v4hi)__B, (__v4hi)__C);
|
|
|
689 }
|
|
|
690
|
|
|
691 /* Element-wise multiplication of signed 16-bit values __B and __C, followed
|
|
|
692 by accumulate across all elements and __A. */
|
|
|
693 static __inline __m64
|
|
|
694 _mm_mac_pi16 (__m64 __A, __m64 __B, __m64 __C)
|
|
|
695 {
|
|
|
696 return __builtin_arm_wmacs (__A, (__v4hi)__B, (__v4hi)__C);
|
|
|
697 }
|
|
|
698
|
|
|
699 /* Element-wise multiplication of unsigned 16-bit values __B and __C, followed
|
|
|
700 by accumulate across all elements. */
|
|
|
701 static __inline __m64
|
|
|
702 _mm_macz_pu16 (__m64 __A, __m64 __B)
|
|
|
703 {
|
|
|
704 return __builtin_arm_wmacuz ((__v4hi)__A, (__v4hi)__B);
|
|
|
705 }
|
|
|
706
|
|
|
707 /* Element-wise multiplication of signed 16-bit values __B and __C, followed
|
|
|
708 by accumulate across all elements. */
|
|
|
709 static __inline __m64
|
|
|
710 _mm_macz_pi16 (__m64 __A, __m64 __B)
|
|
|
711 {
|
|
|
712 return __builtin_arm_wmacsz ((__v4hi)__A, (__v4hi)__B);
|
|
|
713 }
|
|
|
714
|
|
|
715 /* Accumulate across all unsigned 8-bit values in __A. */
|
|
|
716 static __inline __m64
|
|
|
717 _mm_acc_pu8 (__m64 __A)
|
|
|
718 {
|
|
|
719 return __builtin_arm_waccb ((__v8qi)__A);
|
|
|
720 }
|
|
|
721
|
|
|
722 /* Accumulate across all unsigned 16-bit values in __A. */
|
|
|
723 static __inline __m64
|
|
|
724 _mm_acc_pu16 (__m64 __A)
|
|
|
725 {
|
|
|
726 return __builtin_arm_wacch ((__v4hi)__A);
|
|
|
727 }
|
|
|
728
|
|
|
729 /* Accumulate across all unsigned 32-bit values in __A. */
|
|
|
730 static __inline __m64
|
|
|
731 _mm_acc_pu32 (__m64 __A)
|
|
|
732 {
|
|
|
733 return __builtin_arm_waccw ((__v2si)__A);
|
|
|
734 }
|
|
|
735
|
|
|
736 static __inline __m64
|
|
|
737 _mm_mia_si64 (__m64 __A, int __B, int __C)
|
|
|
738 {
|
|
|
739 return __builtin_arm_tmia (__A, __B, __C);
|
|
|
740 }
|
|
|
741
|
|
|
742 static __inline __m64
|
|
|
743 _mm_miaph_si64 (__m64 __A, int __B, int __C)
|
|
|
744 {
|
|
|
745 return __builtin_arm_tmiaph (__A, __B, __C);
|
|
|
746 }
|
|
|
747
|
|
|
748 static __inline __m64
|
|
|
749 _mm_miabb_si64 (__m64 __A, int __B, int __C)
|
|
|
750 {
|
|
|
751 return __builtin_arm_tmiabb (__A, __B, __C);
|
|
|
752 }
|
|
|
753
|
|
|
754 static __inline __m64
|
|
|
755 _mm_miabt_si64 (__m64 __A, int __B, int __C)
|
|
|
756 {
|
|
|
757 return __builtin_arm_tmiabt (__A, __B, __C);
|
|
|
758 }
|
|
|
759
|
|
|
760 static __inline __m64
|
|
|
761 _mm_miatb_si64 (__m64 __A, int __B, int __C)
|
|
|
762 {
|
|
|
763 return __builtin_arm_tmiatb (__A, __B, __C);
|
|
|
764 }
|
|
|
765
|
|
|
766 static __inline __m64
|
|
|
767 _mm_miatt_si64 (__m64 __A, int __B, int __C)
|
|
|
768 {
|
|
|
769 return __builtin_arm_tmiatt (__A, __B, __C);
|
|
|
770 }
|
|
|
771
|
|
|
772 /* Extract one of the elements of A and sign extend. The selector N must
|
|
|
773 be immediate. */
|
|
|
774 #define _mm_extract_pi8(A, N) __builtin_arm_textrmsb ((__v8qi)(A), (N))
|
|
|
775 #define _mm_extract_pi16(A, N) __builtin_arm_textrmsh ((__v4hi)(A), (N))
|
|
|
776 #define _mm_extract_pi32(A, N) __builtin_arm_textrmsw ((__v2si)(A), (N))
|
|
|
777
|
|
|
778 /* Extract one of the elements of A and zero extend. The selector N must
|
|
|
779 be immediate. */
|
|
|
780 #define _mm_extract_pu8(A, N) __builtin_arm_textrmub ((__v8qi)(A), (N))
|
|
|
781 #define _mm_extract_pu16(A, N) __builtin_arm_textrmuh ((__v4hi)(A), (N))
|
|
|
782 #define _mm_extract_pu32(A, N) __builtin_arm_textrmuw ((__v2si)(A), (N))
|
|
|
783
|
|
|
784 /* Inserts word D into one of the elements of A. The selector N must be
|
|
|
785 immediate. */
|
|
|
786 #define _mm_insert_pi8(A, D, N) \
|
|
|
787 ((__m64) __builtin_arm_tinsrb ((__v8qi)(A), (D), (N)))
|
|
|
788 #define _mm_insert_pi16(A, D, N) \
|
|
|
789 ((__m64) __builtin_arm_tinsrh ((__v4hi)(A), (D), (N)))
|
|
|
790 #define _mm_insert_pi32(A, D, N) \
|
|
|
791 ((__m64) __builtin_arm_tinsrw ((__v2si)(A), (D), (N)))
|
|
|
792
|
|
|
793 /* Compute the element-wise maximum of signed 8-bit values. */
|
|
|
794 static __inline __m64
|
|
|
795 _mm_max_pi8 (__m64 __A, __m64 __B)
|
|
|
796 {
|
|
|
797 return (__m64) __builtin_arm_wmaxsb ((__v8qi)__A, (__v8qi)__B);
|
|
|
798 }
|
|
|
799
|
|
|
800 /* Compute the element-wise maximum of signed 16-bit values. */
|
|
|
801 static __inline __m64
|
|
|
802 _mm_max_pi16 (__m64 __A, __m64 __B)
|
|
|
803 {
|
|
|
804 return (__m64) __builtin_arm_wmaxsh ((__v4hi)__A, (__v4hi)__B);
|
|
|
805 }
|
|
|
806
|
|
|
807 /* Compute the element-wise maximum of signed 32-bit values. */
|
|
|
808 static __inline __m64
|
|
|
809 _mm_max_pi32 (__m64 __A, __m64 __B)
|
|
|
810 {
|
|
|
811 return (__m64) __builtin_arm_wmaxsw ((__v2si)__A, (__v2si)__B);
|
|
|
812 }
|
|
|
813
|
|
|
814 /* Compute the element-wise maximum of unsigned 8-bit values. */
|
|
|
815 static __inline __m64
|
|
|
816 _mm_max_pu8 (__m64 __A, __m64 __B)
|
|
|
817 {
|
|
|
818 return (__m64) __builtin_arm_wmaxub ((__v8qi)__A, (__v8qi)__B);
|
|
|
819 }
|
|
|
820
|
|
|
821 /* Compute the element-wise maximum of unsigned 16-bit values. */
|
|
|
822 static __inline __m64
|
|
|
823 _mm_max_pu16 (__m64 __A, __m64 __B)
|
|
|
824 {
|
|
|
825 return (__m64) __builtin_arm_wmaxuh ((__v4hi)__A, (__v4hi)__B);
|
|
|
826 }
|
|
|
827
|
|
|
828 /* Compute the element-wise maximum of unsigned 32-bit values. */
|
|
|
829 static __inline __m64
|
|
|
830 _mm_max_pu32 (__m64 __A, __m64 __B)
|
|
|
831 {
|
|
|
832 return (__m64) __builtin_arm_wmaxuw ((__v2si)__A, (__v2si)__B);
|
|
|
833 }
|
|
|
834
|
|
|
835 /* Compute the element-wise minimum of signed 16-bit values. */
|
|
|
836 static __inline __m64
|
|
|
837 _mm_min_pi8 (__m64 __A, __m64 __B)
|
|
|
838 {
|
|
|
839 return (__m64) __builtin_arm_wminsb ((__v8qi)__A, (__v8qi)__B);
|
|
|
840 }
|
|
|
841
|
|
|
842 /* Compute the element-wise minimum of signed 16-bit values. */
|
|
|
843 static __inline __m64
|
|
|
844 _mm_min_pi16 (__m64 __A, __m64 __B)
|
|
|
845 {
|
|
|
846 return (__m64) __builtin_arm_wminsh ((__v4hi)__A, (__v4hi)__B);
|
|
|
847 }
|
|
|
848
|
|
|
849 /* Compute the element-wise minimum of signed 32-bit values. */
|
|
|
850 static __inline __m64
|
|
|
851 _mm_min_pi32 (__m64 __A, __m64 __B)
|
|
|
852 {
|
|
|
853 return (__m64) __builtin_arm_wminsw ((__v2si)__A, (__v2si)__B);
|
|
|
854 }
|
|
|
855
|
|
|
856 /* Compute the element-wise minimum of unsigned 16-bit values. */
|
|
|
857 static __inline __m64
|
|
|
858 _mm_min_pu8 (__m64 __A, __m64 __B)
|
|
|
859 {
|
|
|
860 return (__m64) __builtin_arm_wminub ((__v8qi)__A, (__v8qi)__B);
|
|
|
861 }
|
|
|
862
|
|
|
863 /* Compute the element-wise minimum of unsigned 16-bit values. */
|
|
|
864 static __inline __m64
|
|
|
865 _mm_min_pu16 (__m64 __A, __m64 __B)
|
|
|
866 {
|
|
|
867 return (__m64) __builtin_arm_wminuh ((__v4hi)__A, (__v4hi)__B);
|
|
|
868 }
|
|
|
869
|
|
|
870 /* Compute the element-wise minimum of unsigned 32-bit values. */
|
|
|
871 static __inline __m64
|
|
|
872 _mm_min_pu32 (__m64 __A, __m64 __B)
|
|
|
873 {
|
|
|
874 return (__m64) __builtin_arm_wminuw ((__v2si)__A, (__v2si)__B);
|
|
|
875 }
|
|
|
876
|
|
|
877 /* Create an 8-bit mask of the signs of 8-bit values. */
|
|
|
878 static __inline int
|
|
|
879 _mm_movemask_pi8 (__m64 __A)
|
|
|
880 {
|
|
|
881 return __builtin_arm_tmovmskb ((__v8qi)__A);
|
|
|
882 }
|
|
|
883
|
|
|
884 /* Create an 8-bit mask of the signs of 16-bit values. */
|
|
|
885 static __inline int
|
|
|
886 _mm_movemask_pi16 (__m64 __A)
|
|
|
887 {
|
|
|
888 return __builtin_arm_tmovmskh ((__v4hi)__A);
|
|
|
889 }
|
|
|
890
|
|
|
891 /* Create an 8-bit mask of the signs of 32-bit values. */
|
|
|
892 static __inline int
|
|
|
893 _mm_movemask_pi32 (__m64 __A)
|
|
|
894 {
|
|
|
895 return __builtin_arm_tmovmskw ((__v2si)__A);
|
|
|
896 }
|
|
|
897
|
|
|
898 /* Return a combination of the four 16-bit values in A. The selector
|
|
|
899 must be an immediate. */
|
|
|
900 #define _mm_shuffle_pi16(A, N) \
|
|
|
901 ((__m64) __builtin_arm_wshufh ((__v4hi)(A), (N)))
|
|
|
902
|
|
|
903
|
|
|
904 /* Compute the rounded averages of the unsigned 8-bit values in A and B. */
|
|
|
905 static __inline __m64
|
|
|
906 _mm_avg_pu8 (__m64 __A, __m64 __B)
|
|
|
907 {
|
|
|
908 return (__m64) __builtin_arm_wavg2br ((__v8qi)__A, (__v8qi)__B);
|
|
|
909 }
|
|
|
910
|
|
|
911 /* Compute the rounded averages of the unsigned 16-bit values in A and B. */
|
|
|
912 static __inline __m64
|
|
|
913 _mm_avg_pu16 (__m64 __A, __m64 __B)
|
|
|
914 {
|
|
|
915 return (__m64) __builtin_arm_wavg2hr ((__v4hi)__A, (__v4hi)__B);
|
|
|
916 }
|
|
|
917
|
|
|
918 /* Compute the averages of the unsigned 8-bit values in A and B. */
|
|
|
919 static __inline __m64
|
|
|
920 _mm_avg2_pu8 (__m64 __A, __m64 __B)
|
|
|
921 {
|
|
|
922 return (__m64) __builtin_arm_wavg2b ((__v8qi)__A, (__v8qi)__B);
|
|
|
923 }
|
|
|
924
|
|
|
925 /* Compute the averages of the unsigned 16-bit values in A and B. */
|
|
|
926 static __inline __m64
|
|
|
927 _mm_avg2_pu16 (__m64 __A, __m64 __B)
|
|
|
928 {
|
|
|
929 return (__m64) __builtin_arm_wavg2h ((__v4hi)__A, (__v4hi)__B);
|
|
|
930 }
|
|
|
931
|
|
|
932 /* Compute the sum of the absolute differences of the unsigned 8-bit
|
|
|
933 values in A and B. Return the value in the lower 16-bit word; the
|
|
|
934 upper words are cleared. */
|
|
|
935 static __inline __m64
|
|
|
936 _mm_sad_pu8 (__m64 __A, __m64 __B)
|
|
|
937 {
|
|
|
938 return (__m64) __builtin_arm_wsadb ((__v8qi)__A, (__v8qi)__B);
|
|
|
939 }
|
|
|
940
|
|
|
941 /* Compute the sum of the absolute differences of the unsigned 16-bit
|
|
|
942 values in A and B. Return the value in the lower 32-bit word; the
|
|
|
943 upper words are cleared. */
|
|
|
944 static __inline __m64
|
|
|
945 _mm_sad_pu16 (__m64 __A, __m64 __B)
|
|
|
946 {
|
|
|
947 return (__m64) __builtin_arm_wsadh ((__v4hi)__A, (__v4hi)__B);
|
|
|
948 }
|
|
|
949
|
|
|
950 /* Compute the sum of the absolute differences of the unsigned 8-bit
|
|
|
951 values in A and B. Return the value in the lower 16-bit word; the
|
|
|
952 upper words are cleared. */
|
|
|
953 static __inline __m64
|
|
|
954 _mm_sadz_pu8 (__m64 __A, __m64 __B)
|
|
|
955 {
|
|
|
956 return (__m64) __builtin_arm_wsadbz ((__v8qi)__A, (__v8qi)__B);
|
|
|
957 }
|
|
|
958
|
|
|
959 /* Compute the sum of the absolute differences of the unsigned 16-bit
|
|
|
960 values in A and B. Return the value in the lower 32-bit word; the
|
|
|
961 upper words are cleared. */
|
|
|
962 static __inline __m64
|
|
|
963 _mm_sadz_pu16 (__m64 __A, __m64 __B)
|
|
|
964 {
|
|
|
965 return (__m64) __builtin_arm_wsadhz ((__v4hi)__A, (__v4hi)__B);
|
|
|
966 }
|
|
|
967
|
|
|
968 static __inline __m64
|
|
|
969 _mm_align_si64 (__m64 __A, __m64 __B, int __C)
|
|
|
970 {
|
|
|
971 return (__m64) __builtin_arm_walign ((__v8qi)__A, (__v8qi)__B, __C);
|
|
|
972 }
|
|
|
973
|
|
|
974 /* Creates a 64-bit zero. */
|
|
|
975 static __inline __m64
|
|
|
976 _mm_setzero_si64 (void)
|
|
|
977 {
|
|
|
978 return __builtin_arm_wzero ();
|
|
|
979 }
|
|
|
980
|
|
|
981 /* Set and Get arbitrary iWMMXt Control registers.
|
|
|
982 Note only registers 0-3 and 8-11 are currently defined,
|
|
|
983 the rest are reserved. */
|
|
|
984
|
|
|
985 static __inline void
|
|
|
986 _mm_setwcx (const int __value, const int __regno)
|
|
|
987 {
|
|
|
988 switch (__regno)
|
|
|
989 {
|
|
|
990 case 0: __builtin_arm_setwcx (__value, 0); break;
|
|
|
991 case 1: __builtin_arm_setwcx (__value, 1); break;
|
|
|
992 case 2: __builtin_arm_setwcx (__value, 2); break;
|
|
|
993 case 3: __builtin_arm_setwcx (__value, 3); break;
|
|
|
994 case 8: __builtin_arm_setwcx (__value, 8); break;
|
|
|
995 case 9: __builtin_arm_setwcx (__value, 9); break;
|
|
|
996 case 10: __builtin_arm_setwcx (__value, 10); break;
|
|
|
997 case 11: __builtin_arm_setwcx (__value, 11); break;
|
|
|
998 default: break;
|
|
|
999 }
|
|
|
1000 }
|
|
|
1001
|
|
|
1002 static __inline int
|
|
|
1003 _mm_getwcx (const int __regno)
|
|
|
1004 {
|
|
|
1005 switch (__regno)
|
|
|
1006 {
|
|
|
1007 case 0: return __builtin_arm_getwcx (0);
|
|
|
1008 case 1: return __builtin_arm_getwcx (1);
|
|
|
1009 case 2: return __builtin_arm_getwcx (2);
|
|
|
1010 case 3: return __builtin_arm_getwcx (3);
|
|
|
1011 case 8: return __builtin_arm_getwcx (8);
|
|
|
1012 case 9: return __builtin_arm_getwcx (9);
|
|
|
1013 case 10: return __builtin_arm_getwcx (10);
|
|
|
1014 case 11: return __builtin_arm_getwcx (11);
|
|
|
1015 default: return 0;
|
|
|
1016 }
|
|
|
1017 }
|
|
|
1018
|
|
|
1019 /* Creates a vector of two 32-bit values; I0 is least significant. */
|
|
|
1020 static __inline __m64
|
|
|
1021 _mm_set_pi32 (int __i1, int __i0)
|
|
|
1022 {
|
|
|
1023 union {
|
|
|
1024 __m64 __q;
|
|
|
1025 struct {
|
|
|
1026 unsigned int __i0;
|
|
|
1027 unsigned int __i1;
|
|
|
1028 } __s;
|
|
|
1029 } __u;
|
|
|
1030
|
|
|
1031 __u.__s.__i0 = __i0;
|
|
|
1032 __u.__s.__i1 = __i1;
|
|
|
1033
|
|
|
1034 return __u.__q;
|
|
|
1035 }
|
|
|
1036
|
|
|
1037 /* Creates a vector of four 16-bit values; W0 is least significant. */
|
|
|
1038 static __inline __m64
|
|
|
1039 _mm_set_pi16 (short __w3, short __w2, short __w1, short __w0)
|
|
|
1040 {
|
|
|
1041 unsigned int __i1 = (unsigned short)__w3 << 16 | (unsigned short)__w2;
|
|
|
1042 unsigned int __i0 = (unsigned short)__w1 << 16 | (unsigned short)__w0;
|
|
|
1043 return _mm_set_pi32 (__i1, __i0);
|
|
|
1044
|
|
|
1045 }
|
|
|
1046
|
|
|
1047 /* Creates a vector of eight 8-bit values; B0 is least significant. */
|
|
|
1048 static __inline __m64
|
|
|
1049 _mm_set_pi8 (char __b7, char __b6, char __b5, char __b4,
|
|
|
1050 char __b3, char __b2, char __b1, char __b0)
|
|
|
1051 {
|
|
|
1052 unsigned int __i1, __i0;
|
|
|
1053
|
|
|
1054 __i1 = (unsigned char)__b7;
|
|
|
1055 __i1 = __i1 << 8 | (unsigned char)__b6;
|
|
|
1056 __i1 = __i1 << 8 | (unsigned char)__b5;
|
|
|
1057 __i1 = __i1 << 8 | (unsigned char)__b4;
|
|
|
1058
|
|
|
1059 __i0 = (unsigned char)__b3;
|
|
|
1060 __i0 = __i0 << 8 | (unsigned char)__b2;
|
|
|
1061 __i0 = __i0 << 8 | (unsigned char)__b1;
|
|
|
1062 __i0 = __i0 << 8 | (unsigned char)__b0;
|
|
|
1063
|
|
|
1064 return _mm_set_pi32 (__i1, __i0);
|
|
|
1065 }
|
|
|
1066
|
|
|
1067 /* Similar, but with the arguments in reverse order. */
|
|
|
1068 static __inline __m64
|
|
|
1069 _mm_setr_pi32 (int __i0, int __i1)
|
|
|
1070 {
|
|
|
1071 return _mm_set_pi32 (__i1, __i0);
|
|
|
1072 }
|
|
|
1073
|
|
|
1074 static __inline __m64
|
|
|
1075 _mm_setr_pi16 (short __w0, short __w1, short __w2, short __w3)
|
|
|
1076 {
|
|
|
1077 return _mm_set_pi16 (__w3, __w2, __w1, __w0);
|
|
|
1078 }
|
|
|
1079
|
|
|
1080 static __inline __m64
|
|
|
1081 _mm_setr_pi8 (char __b0, char __b1, char __b2, char __b3,
|
|
|
1082 char __b4, char __b5, char __b6, char __b7)
|
|
|
1083 {
|
|
|
1084 return _mm_set_pi8 (__b7, __b6, __b5, __b4, __b3, __b2, __b1, __b0);
|
|
|
1085 }
|
|
|
1086
|
|
|
1087 /* Creates a vector of two 32-bit values, both elements containing I. */
|
|
|
1088 static __inline __m64
|
|
|
1089 _mm_set1_pi32 (int __i)
|
|
|
1090 {
|
|
|
1091 return _mm_set_pi32 (__i, __i);
|
|
|
1092 }
|
|
|
1093
|
|
|
1094 /* Creates a vector of four 16-bit values, all elements containing W. */
|
|
|
1095 static __inline __m64
|
|
|
1096 _mm_set1_pi16 (short __w)
|
|
|
1097 {
|
|
|
1098 unsigned int __i = (unsigned short)__w << 16 | (unsigned short)__w;
|
|
|
1099 return _mm_set1_pi32 (__i);
|
|
|
1100 }
|
|
|
1101
|
|
|
1102 /* Creates a vector of four 16-bit values, all elements containing B. */
|
|
|
1103 static __inline __m64
|
|
|
1104 _mm_set1_pi8 (char __b)
|
|
|
1105 {
|
|
|
1106 unsigned int __w = (unsigned char)__b << 8 | (unsigned char)__b;
|
|
|
1107 unsigned int __i = __w << 16 | __w;
|
|
|
1108 return _mm_set1_pi32 (__i);
|
|
|
1109 }
|
|
|
1110
|
|
|
1111 /* Convert an integer to a __m64 object. */
|
|
|
1112 static __inline __m64
|
|
|
1113 _m_from_int (int __a)
|
|
|
1114 {
|
|
|
1115 return (__m64)__a;
|
|
|
1116 }
|
|
|
1117
|
|
|
1118 #define _m_packsswb _mm_packs_pi16
|
|
|
1119 #define _m_packssdw _mm_packs_pi32
|
|
|
1120 #define _m_packuswb _mm_packs_pu16
|
|
|
1121 #define _m_packusdw _mm_packs_pu32
|
|
|
1122 #define _m_packssqd _mm_packs_pi64
|
|
|
1123 #define _m_packusqd _mm_packs_pu64
|
|
|
1124 #define _mm_packs_si64 _mm_packs_pi64
|
|
|
1125 #define _mm_packs_su64 _mm_packs_pu64
|
|
|
1126 #define _m_punpckhbw _mm_unpackhi_pi8
|
|
|
1127 #define _m_punpckhwd _mm_unpackhi_pi16
|
|
|
1128 #define _m_punpckhdq _mm_unpackhi_pi32
|
|
|
1129 #define _m_punpcklbw _mm_unpacklo_pi8
|
|
|
1130 #define _m_punpcklwd _mm_unpacklo_pi16
|
|
|
1131 #define _m_punpckldq _mm_unpacklo_pi32
|
|
|
1132 #define _m_punpckehsbw _mm_unpackeh_pi8
|
|
|
1133 #define _m_punpckehswd _mm_unpackeh_pi16
|
|
|
1134 #define _m_punpckehsdq _mm_unpackeh_pi32
|
|
|
1135 #define _m_punpckehubw _mm_unpackeh_pu8
|
|
|
1136 #define _m_punpckehuwd _mm_unpackeh_pu16
|
|
|
1137 #define _m_punpckehudq _mm_unpackeh_pu32
|
|
|
1138 #define _m_punpckelsbw _mm_unpackel_pi8
|
|
|
1139 #define _m_punpckelswd _mm_unpackel_pi16
|
|
|
1140 #define _m_punpckelsdq _mm_unpackel_pi32
|
|
|
1141 #define _m_punpckelubw _mm_unpackel_pu8
|
|
|
1142 #define _m_punpckeluwd _mm_unpackel_pu16
|
|
|
1143 #define _m_punpckeludq _mm_unpackel_pu32
|
|
|
1144 #define _m_paddb _mm_add_pi8
|
|
|
1145 #define _m_paddw _mm_add_pi16
|
|
|
1146 #define _m_paddd _mm_add_pi32
|
|
|
1147 #define _m_paddsb _mm_adds_pi8
|
|
|
1148 #define _m_paddsw _mm_adds_pi16
|
|
|
1149 #define _m_paddsd _mm_adds_pi32
|
|
|
1150 #define _m_paddusb _mm_adds_pu8
|
|
|
1151 #define _m_paddusw _mm_adds_pu16
|
|
|
1152 #define _m_paddusd _mm_adds_pu32
|
|
|
1153 #define _m_psubb _mm_sub_pi8
|
|
|
1154 #define _m_psubw _mm_sub_pi16
|
|
|
1155 #define _m_psubd _mm_sub_pi32
|
|
|
1156 #define _m_psubsb _mm_subs_pi8
|
|
|
1157 #define _m_psubsw _mm_subs_pi16
|
|
|
1158 #define _m_psubuw _mm_subs_pi32
|
|
|
1159 #define _m_psubusb _mm_subs_pu8
|
|
|
1160 #define _m_psubusw _mm_subs_pu16
|
|
|
1161 #define _m_psubusd _mm_subs_pu32
|
|
|
1162 #define _m_pmaddwd _mm_madd_pi16
|
|
|
1163 #define _m_pmadduwd _mm_madd_pu16
|
|
|
1164 #define _m_pmulhw _mm_mulhi_pi16
|
|
|
1165 #define _m_pmulhuw _mm_mulhi_pu16
|
|
|
1166 #define _m_pmullw _mm_mullo_pi16
|
|
|
1167 #define _m_pmacsw _mm_mac_pi16
|
|
|
1168 #define _m_pmacuw _mm_mac_pu16
|
|
|
1169 #define _m_pmacszw _mm_macz_pi16
|
|
|
1170 #define _m_pmacuzw _mm_macz_pu16
|
|
|
1171 #define _m_paccb _mm_acc_pu8
|
|
|
1172 #define _m_paccw _mm_acc_pu16
|
|
|
1173 #define _m_paccd _mm_acc_pu32
|
|
|
1174 #define _m_pmia _mm_mia_si64
|
|
|
1175 #define _m_pmiaph _mm_miaph_si64
|
|
|
1176 #define _m_pmiabb _mm_miabb_si64
|
|
|
1177 #define _m_pmiabt _mm_miabt_si64
|
|
|
1178 #define _m_pmiatb _mm_miatb_si64
|
|
|
1179 #define _m_pmiatt _mm_miatt_si64
|
|
|
1180 #define _m_psllw _mm_sll_pi16
|
|
|
1181 #define _m_psllwi _mm_slli_pi16
|
|
|
1182 #define _m_pslld _mm_sll_pi32
|
|
|
1183 #define _m_pslldi _mm_slli_pi32
|
|
|
1184 #define _m_psllq _mm_sll_si64
|
|
|
1185 #define _m_psllqi _mm_slli_si64
|
|
|
1186 #define _m_psraw _mm_sra_pi16
|
|
|
1187 #define _m_psrawi _mm_srai_pi16
|
|
|
1188 #define _m_psrad _mm_sra_pi32
|
|
|
1189 #define _m_psradi _mm_srai_pi32
|
|
|
1190 #define _m_psraq _mm_sra_si64
|
|
|
1191 #define _m_psraqi _mm_srai_si64
|
|
|
1192 #define _m_psrlw _mm_srl_pi16
|
|
|
1193 #define _m_psrlwi _mm_srli_pi16
|
|
|
1194 #define _m_psrld _mm_srl_pi32
|
|
|
1195 #define _m_psrldi _mm_srli_pi32
|
|
|
1196 #define _m_psrlq _mm_srl_si64
|
|
|
1197 #define _m_psrlqi _mm_srli_si64
|
|
|
1198 #define _m_prorw _mm_ror_pi16
|
|
|
1199 #define _m_prorwi _mm_rori_pi16
|
|
|
1200 #define _m_prord _mm_ror_pi32
|
|
|
1201 #define _m_prordi _mm_rori_pi32
|
|
|
1202 #define _m_prorq _mm_ror_si64
|
|
|
1203 #define _m_prorqi _mm_rori_si64
|
|
|
1204 #define _m_pand _mm_and_si64
|
|
|
1205 #define _m_pandn _mm_andnot_si64
|
|
|
1206 #define _m_por _mm_or_si64
|
|
|
1207 #define _m_pxor _mm_xor_si64
|
|
|
1208 #define _m_pcmpeqb _mm_cmpeq_pi8
|
|
|
1209 #define _m_pcmpeqw _mm_cmpeq_pi16
|
|
|
1210 #define _m_pcmpeqd _mm_cmpeq_pi32
|
|
|
1211 #define _m_pcmpgtb _mm_cmpgt_pi8
|
|
|
1212 #define _m_pcmpgtub _mm_cmpgt_pu8
|
|
|
1213 #define _m_pcmpgtw _mm_cmpgt_pi16
|
|
|
1214 #define _m_pcmpgtuw _mm_cmpgt_pu16
|
|
|
1215 #define _m_pcmpgtd _mm_cmpgt_pi32
|
|
|
1216 #define _m_pcmpgtud _mm_cmpgt_pu32
|
|
|
1217 #define _m_pextrb _mm_extract_pi8
|
|
|
1218 #define _m_pextrw _mm_extract_pi16
|
|
|
1219 #define _m_pextrd _mm_extract_pi32
|
|
|
1220 #define _m_pextrub _mm_extract_pu8
|
|
|
1221 #define _m_pextruw _mm_extract_pu16
|
|
|
1222 #define _m_pextrud _mm_extract_pu32
|
|
|
1223 #define _m_pinsrb _mm_insert_pi8
|
|
|
1224 #define _m_pinsrw _mm_insert_pi16
|
|
|
1225 #define _m_pinsrd _mm_insert_pi32
|
|
|
1226 #define _m_pmaxsb _mm_max_pi8
|
|
|
1227 #define _m_pmaxsw _mm_max_pi16
|
|
|
1228 #define _m_pmaxsd _mm_max_pi32
|
|
|
1229 #define _m_pmaxub _mm_max_pu8
|
|
|
1230 #define _m_pmaxuw _mm_max_pu16
|
|
|
1231 #define _m_pmaxud _mm_max_pu32
|
|
|
1232 #define _m_pminsb _mm_min_pi8
|
|
|
1233 #define _m_pminsw _mm_min_pi16
|
|
|
1234 #define _m_pminsd _mm_min_pi32
|
|
|
1235 #define _m_pminub _mm_min_pu8
|
|
|
1236 #define _m_pminuw _mm_min_pu16
|
|
|
1237 #define _m_pminud _mm_min_pu32
|
|
|
1238 #define _m_pmovmskb _mm_movemask_pi8
|
|
|
1239 #define _m_pmovmskw _mm_movemask_pi16
|
|
|
1240 #define _m_pmovmskd _mm_movemask_pi32
|
|
|
1241 #define _m_pshufw _mm_shuffle_pi16
|
|
|
1242 #define _m_pavgb _mm_avg_pu8
|
|
|
1243 #define _m_pavgw _mm_avg_pu16
|
|
|
1244 #define _m_pavg2b _mm_avg2_pu8
|
|
|
1245 #define _m_pavg2w _mm_avg2_pu16
|
|
|
1246 #define _m_psadbw _mm_sad_pu8
|
|
|
1247 #define _m_psadwd _mm_sad_pu16
|
|
|
1248 #define _m_psadzbw _mm_sadz_pu8
|
|
|
1249 #define _m_psadzwd _mm_sadz_pu16
|
|
|
1250 #define _m_paligniq _mm_align_si64
|
|
|
1251 #define _m_cvt_si2pi _mm_cvtsi64_m64
|
|
|
1252 #define _m_cvt_pi2si _mm_cvtm64_si64
|
|
|
1253
|
|
|
1254 #endif /* _MMINTRIN_H_INCLUDED */
|
