1 | /*===---- tmmintrin.h - SSSE3 intrinsics -----------------------------------=== |
---|---|
2 | * |
3 | * Permission is hereby granted, free of charge, to any person obtaining a copy |
4 | * of this software and associated documentation files (the "Software"), to deal |
5 | * in the Software without restriction, including without limitation the rights |
6 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell |
7 | * copies of the Software, and to permit persons to whom the Software is |
8 | * furnished to do so, subject to the following conditions: |
9 | * |
10 | * The above copyright notice and this permission notice shall be included in |
11 | * all copies or substantial portions of the Software. |
12 | * |
13 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
14 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
15 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE |
16 | * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
17 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, |
18 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN |
19 | * THE SOFTWARE. |
20 | * |
21 | *===-----------------------------------------------------------------------=== |
22 | */ |
23 | |
24 | #ifndef __TMMINTRIN_H |
25 | #define __TMMINTRIN_H |
26 | |
27 | #include <pmmintrin.h> |
28 | |
29 | /* Define the default attributes for the functions in this file. */ |
30 | #define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("ssse3"), __min_vector_width__(64))) |
31 | #define __DEFAULT_FN_ATTRS_MMX __attribute__((__always_inline__, __nodebug__, __target__("mmx,ssse3"), __min_vector_width__(64))) |
32 | |
33 | /// Computes the absolute value of each of the packed 8-bit signed |
34 | /// integers in the source operand and stores the 8-bit unsigned integer |
35 | /// results in the destination. |
36 | /// |
37 | /// \headerfile <x86intrin.h> |
38 | /// |
39 | /// This intrinsic corresponds to the \c PABSB instruction. |
40 | /// |
41 | /// \param __a |
42 | /// A 64-bit vector of [8 x i8]. |
43 | /// \returns A 64-bit integer vector containing the absolute values of the |
44 | /// elements in the operand. |
45 | static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
46 | _mm_abs_pi8(__m64 __a) |
47 | { |
48 | return (__m64)__builtin_ia32_pabsb((__v8qi)__a); |
49 | } |
50 | |
51 | /// Computes the absolute value of each of the packed 8-bit signed |
52 | /// integers in the source operand and stores the 8-bit unsigned integer |
53 | /// results in the destination. |
54 | /// |
55 | /// \headerfile <x86intrin.h> |
56 | /// |
57 | /// This intrinsic corresponds to the \c VPABSB instruction. |
58 | /// |
59 | /// \param __a |
60 | /// A 128-bit vector of [16 x i8]. |
61 | /// \returns A 128-bit integer vector containing the absolute values of the |
62 | /// elements in the operand. |
63 | static __inline__ __m128i __DEFAULT_FN_ATTRS |
64 | _mm_abs_epi8(__m128i __a) |
65 | { |
66 | return (__m128i)__builtin_ia32_pabsb128((__v16qi)__a); |
67 | } |
68 | |
69 | /// Computes the absolute value of each of the packed 16-bit signed |
70 | /// integers in the source operand and stores the 16-bit unsigned integer |
71 | /// results in the destination. |
72 | /// |
73 | /// \headerfile <x86intrin.h> |
74 | /// |
75 | /// This intrinsic corresponds to the \c PABSW instruction. |
76 | /// |
77 | /// \param __a |
78 | /// A 64-bit vector of [4 x i16]. |
79 | /// \returns A 64-bit integer vector containing the absolute values of the |
80 | /// elements in the operand. |
81 | static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
82 | _mm_abs_pi16(__m64 __a) |
83 | { |
84 | return (__m64)__builtin_ia32_pabsw((__v4hi)__a); |
85 | } |
86 | |
87 | /// Computes the absolute value of each of the packed 16-bit signed |
88 | /// integers in the source operand and stores the 16-bit unsigned integer |
89 | /// results in the destination. |
90 | /// |
91 | /// \headerfile <x86intrin.h> |
92 | /// |
93 | /// This intrinsic corresponds to the \c VPABSW instruction. |
94 | /// |
95 | /// \param __a |
96 | /// A 128-bit vector of [8 x i16]. |
97 | /// \returns A 128-bit integer vector containing the absolute values of the |
98 | /// elements in the operand. |
99 | static __inline__ __m128i __DEFAULT_FN_ATTRS |
100 | _mm_abs_epi16(__m128i __a) |
101 | { |
102 | return (__m128i)__builtin_ia32_pabsw128((__v8hi)__a); |
103 | } |
104 | |
105 | /// Computes the absolute value of each of the packed 32-bit signed |
106 | /// integers in the source operand and stores the 32-bit unsigned integer |
107 | /// results in the destination. |
108 | /// |
109 | /// \headerfile <x86intrin.h> |
110 | /// |
111 | /// This intrinsic corresponds to the \c PABSD instruction. |
112 | /// |
113 | /// \param __a |
114 | /// A 64-bit vector of [2 x i32]. |
115 | /// \returns A 64-bit integer vector containing the absolute values of the |
116 | /// elements in the operand. |
117 | static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
118 | _mm_abs_pi32(__m64 __a) |
119 | { |
120 | return (__m64)__builtin_ia32_pabsd((__v2si)__a); |
121 | } |
122 | |
123 | /// Computes the absolute value of each of the packed 32-bit signed |
124 | /// integers in the source operand and stores the 32-bit unsigned integer |
125 | /// results in the destination. |
126 | /// |
127 | /// \headerfile <x86intrin.h> |
128 | /// |
129 | /// This intrinsic corresponds to the \c VPABSD instruction. |
130 | /// |
131 | /// \param __a |
132 | /// A 128-bit vector of [4 x i32]. |
133 | /// \returns A 128-bit integer vector containing the absolute values of the |
134 | /// elements in the operand. |
135 | static __inline__ __m128i __DEFAULT_FN_ATTRS |
136 | _mm_abs_epi32(__m128i __a) |
137 | { |
138 | return (__m128i)__builtin_ia32_pabsd128((__v4si)__a); |
139 | } |
140 | |
141 | /// Concatenates the two 128-bit integer vector operands, and |
142 | /// right-shifts the result by the number of bytes specified in the immediate |
143 | /// operand. |
144 | /// |
145 | /// \headerfile <x86intrin.h> |
146 | /// |
147 | /// \code |
148 | /// __m128i _mm_alignr_epi8(__m128i a, __m128i b, const int n); |
149 | /// \endcode |
150 | /// |
151 | /// This intrinsic corresponds to the \c PALIGNR instruction. |
152 | /// |
153 | /// \param a |
154 | /// A 128-bit vector of [16 x i8] containing one of the source operands. |
155 | /// \param b |
156 | /// A 128-bit vector of [16 x i8] containing one of the source operands. |
157 | /// \param n |
158 | /// An immediate operand specifying how many bytes to right-shift the result. |
159 | /// \returns A 128-bit integer vector containing the concatenated right-shifted |
160 | /// value. |
161 | #define _mm_alignr_epi8(a, b, n) \ |
162 | (__m128i)__builtin_ia32_palignr128((__v16qi)(__m128i)(a), \ |
163 | (__v16qi)(__m128i)(b), (n)) |
164 | |
165 | /// Concatenates the two 64-bit integer vector operands, and right-shifts |
166 | /// the result by the number of bytes specified in the immediate operand. |
167 | /// |
168 | /// \headerfile <x86intrin.h> |
169 | /// |
170 | /// \code |
171 | /// __m64 _mm_alignr_pi8(__m64 a, __m64 b, const int n); |
172 | /// \endcode |
173 | /// |
174 | /// This intrinsic corresponds to the \c PALIGNR instruction. |
175 | /// |
176 | /// \param a |
177 | /// A 64-bit vector of [8 x i8] containing one of the source operands. |
178 | /// \param b |
179 | /// A 64-bit vector of [8 x i8] containing one of the source operands. |
180 | /// \param n |
181 | /// An immediate operand specifying how many bytes to right-shift the result. |
182 | /// \returns A 64-bit integer vector containing the concatenated right-shifted |
183 | /// value. |
184 | #define _mm_alignr_pi8(a, b, n) \ |
185 | (__m64)__builtin_ia32_palignr((__v8qi)(__m64)(a), (__v8qi)(__m64)(b), (n)) |
186 | |
187 | /// Horizontally adds the adjacent pairs of values contained in 2 packed |
188 | /// 128-bit vectors of [8 x i16]. |
189 | /// |
190 | /// \headerfile <x86intrin.h> |
191 | /// |
192 | /// This intrinsic corresponds to the \c VPHADDW instruction. |
193 | /// |
194 | /// \param __a |
195 | /// A 128-bit vector of [8 x i16] containing one of the source operands. The |
196 | /// horizontal sums of the values are stored in the lower bits of the |
197 | /// destination. |
198 | /// \param __b |
199 | /// A 128-bit vector of [8 x i16] containing one of the source operands. The |
200 | /// horizontal sums of the values are stored in the upper bits of the |
201 | /// destination. |
202 | /// \returns A 128-bit vector of [8 x i16] containing the horizontal sums of |
203 | /// both operands. |
204 | static __inline__ __m128i __DEFAULT_FN_ATTRS |
205 | _mm_hadd_epi16(__m128i __a, __m128i __b) |
206 | { |
207 | return (__m128i)__builtin_ia32_phaddw128((__v8hi)__a, (__v8hi)__b); |
208 | } |
209 | |
210 | /// Horizontally adds the adjacent pairs of values contained in 2 packed |
211 | /// 128-bit vectors of [4 x i32]. |
212 | /// |
213 | /// \headerfile <x86intrin.h> |
214 | /// |
215 | /// This intrinsic corresponds to the \c VPHADDD instruction. |
216 | /// |
217 | /// \param __a |
218 | /// A 128-bit vector of [4 x i32] containing one of the source operands. The |
219 | /// horizontal sums of the values are stored in the lower bits of the |
220 | /// destination. |
221 | /// \param __b |
222 | /// A 128-bit vector of [4 x i32] containing one of the source operands. The |
223 | /// horizontal sums of the values are stored in the upper bits of the |
224 | /// destination. |
225 | /// \returns A 128-bit vector of [4 x i32] containing the horizontal sums of |
226 | /// both operands. |
227 | static __inline__ __m128i __DEFAULT_FN_ATTRS |
228 | _mm_hadd_epi32(__m128i __a, __m128i __b) |
229 | { |
230 | return (__m128i)__builtin_ia32_phaddd128((__v4si)__a, (__v4si)__b); |
231 | } |
232 | |
233 | /// Horizontally adds the adjacent pairs of values contained in 2 packed |
234 | /// 64-bit vectors of [4 x i16]. |
235 | /// |
236 | /// \headerfile <x86intrin.h> |
237 | /// |
238 | /// This intrinsic corresponds to the \c PHADDW instruction. |
239 | /// |
240 | /// \param __a |
241 | /// A 64-bit vector of [4 x i16] containing one of the source operands. The |
242 | /// horizontal sums of the values are stored in the lower bits of the |
243 | /// destination. |
244 | /// \param __b |
245 | /// A 64-bit vector of [4 x i16] containing one of the source operands. The |
246 | /// horizontal sums of the values are stored in the upper bits of the |
247 | /// destination. |
248 | /// \returns A 64-bit vector of [4 x i16] containing the horizontal sums of both |
249 | /// operands. |
250 | static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
251 | _mm_hadd_pi16(__m64 __a, __m64 __b) |
252 | { |
253 | return (__m64)__builtin_ia32_phaddw((__v4hi)__a, (__v4hi)__b); |
254 | } |
255 | |
256 | /// Horizontally adds the adjacent pairs of values contained in 2 packed |
257 | /// 64-bit vectors of [2 x i32]. |
258 | /// |
259 | /// \headerfile <x86intrin.h> |
260 | /// |
261 | /// This intrinsic corresponds to the \c PHADDD instruction. |
262 | /// |
263 | /// \param __a |
264 | /// A 64-bit vector of [2 x i32] containing one of the source operands. The |
265 | /// horizontal sums of the values are stored in the lower bits of the |
266 | /// destination. |
267 | /// \param __b |
268 | /// A 64-bit vector of [2 x i32] containing one of the source operands. The |
269 | /// horizontal sums of the values are stored in the upper bits of the |
270 | /// destination. |
271 | /// \returns A 64-bit vector of [2 x i32] containing the horizontal sums of both |
272 | /// operands. |
273 | static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
274 | _mm_hadd_pi32(__m64 __a, __m64 __b) |
275 | { |
276 | return (__m64)__builtin_ia32_phaddd((__v2si)__a, (__v2si)__b); |
277 | } |
278 | |
279 | /// Horizontally adds the adjacent pairs of values contained in 2 packed |
280 | /// 128-bit vectors of [8 x i16]. Positive sums greater than 0x7FFF are |
281 | /// saturated to 0x7FFF. Negative sums less than 0x8000 are saturated to |
282 | /// 0x8000. |
283 | /// |
284 | /// \headerfile <x86intrin.h> |
285 | /// |
286 | /// This intrinsic corresponds to the \c VPHADDSW instruction. |
287 | /// |
288 | /// \param __a |
289 | /// A 128-bit vector of [8 x i16] containing one of the source operands. The |
290 | /// horizontal sums of the values are stored in the lower bits of the |
291 | /// destination. |
292 | /// \param __b |
293 | /// A 128-bit vector of [8 x i16] containing one of the source operands. The |
294 | /// horizontal sums of the values are stored in the upper bits of the |
295 | /// destination. |
296 | /// \returns A 128-bit vector of [8 x i16] containing the horizontal saturated |
297 | /// sums of both operands. |
298 | static __inline__ __m128i __DEFAULT_FN_ATTRS |
299 | _mm_hadds_epi16(__m128i __a, __m128i __b) |
300 | { |
301 | return (__m128i)__builtin_ia32_phaddsw128((__v8hi)__a, (__v8hi)__b); |
302 | } |
303 | |
304 | /// Horizontally adds the adjacent pairs of values contained in 2 packed |
305 | /// 64-bit vectors of [4 x i16]. Positive sums greater than 0x7FFF are |
306 | /// saturated to 0x7FFF. Negative sums less than 0x8000 are saturated to |
307 | /// 0x8000. |
308 | /// |
309 | /// \headerfile <x86intrin.h> |
310 | /// |
311 | /// This intrinsic corresponds to the \c PHADDSW instruction. |
312 | /// |
313 | /// \param __a |
314 | /// A 64-bit vector of [4 x i16] containing one of the source operands. The |
315 | /// horizontal sums of the values are stored in the lower bits of the |
316 | /// destination. |
317 | /// \param __b |
318 | /// A 64-bit vector of [4 x i16] containing one of the source operands. The |
319 | /// horizontal sums of the values are stored in the upper bits of the |
320 | /// destination. |
321 | /// \returns A 64-bit vector of [4 x i16] containing the horizontal saturated |
322 | /// sums of both operands. |
323 | static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
324 | _mm_hadds_pi16(__m64 __a, __m64 __b) |
325 | { |
326 | return (__m64)__builtin_ia32_phaddsw((__v4hi)__a, (__v4hi)__b); |
327 | } |
328 | |
329 | /// Horizontally subtracts the adjacent pairs of values contained in 2 |
330 | /// packed 128-bit vectors of [8 x i16]. |
331 | /// |
332 | /// \headerfile <x86intrin.h> |
333 | /// |
334 | /// This intrinsic corresponds to the \c VPHSUBW instruction. |
335 | /// |
336 | /// \param __a |
337 | /// A 128-bit vector of [8 x i16] containing one of the source operands. The |
338 | /// horizontal differences between the values are stored in the lower bits of |
339 | /// the destination. |
340 | /// \param __b |
341 | /// A 128-bit vector of [8 x i16] containing one of the source operands. The |
342 | /// horizontal differences between the values are stored in the upper bits of |
343 | /// the destination. |
344 | /// \returns A 128-bit vector of [8 x i16] containing the horizontal differences |
345 | /// of both operands. |
346 | static __inline__ __m128i __DEFAULT_FN_ATTRS |
347 | _mm_hsub_epi16(__m128i __a, __m128i __b) |
348 | { |
349 | return (__m128i)__builtin_ia32_phsubw128((__v8hi)__a, (__v8hi)__b); |
350 | } |
351 | |
352 | /// Horizontally subtracts the adjacent pairs of values contained in 2 |
353 | /// packed 128-bit vectors of [4 x i32]. |
354 | /// |
355 | /// \headerfile <x86intrin.h> |
356 | /// |
357 | /// This intrinsic corresponds to the \c VPHSUBD instruction. |
358 | /// |
359 | /// \param __a |
360 | /// A 128-bit vector of [4 x i32] containing one of the source operands. The |
361 | /// horizontal differences between the values are stored in the lower bits of |
362 | /// the destination. |
363 | /// \param __b |
364 | /// A 128-bit vector of [4 x i32] containing one of the source operands. The |
365 | /// horizontal differences between the values are stored in the upper bits of |
366 | /// the destination. |
367 | /// \returns A 128-bit vector of [4 x i32] containing the horizontal differences |
368 | /// of both operands. |
369 | static __inline__ __m128i __DEFAULT_FN_ATTRS |
370 | _mm_hsub_epi32(__m128i __a, __m128i __b) |
371 | { |
372 | return (__m128i)__builtin_ia32_phsubd128((__v4si)__a, (__v4si)__b); |
373 | } |
374 | |
375 | /// Horizontally subtracts the adjacent pairs of values contained in 2 |
376 | /// packed 64-bit vectors of [4 x i16]. |
377 | /// |
378 | /// \headerfile <x86intrin.h> |
379 | /// |
380 | /// This intrinsic corresponds to the \c PHSUBW instruction. |
381 | /// |
382 | /// \param __a |
383 | /// A 64-bit vector of [4 x i16] containing one of the source operands. The |
384 | /// horizontal differences between the values are stored in the lower bits of |
385 | /// the destination. |
386 | /// \param __b |
387 | /// A 64-bit vector of [4 x i16] containing one of the source operands. The |
388 | /// horizontal differences between the values are stored in the upper bits of |
389 | /// the destination. |
390 | /// \returns A 64-bit vector of [4 x i16] containing the horizontal differences |
391 | /// of both operands. |
392 | static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
393 | _mm_hsub_pi16(__m64 __a, __m64 __b) |
394 | { |
395 | return (__m64)__builtin_ia32_phsubw((__v4hi)__a, (__v4hi)__b); |
396 | } |
397 | |
398 | /// Horizontally subtracts the adjacent pairs of values contained in 2 |
399 | /// packed 64-bit vectors of [2 x i32]. |
400 | /// |
401 | /// \headerfile <x86intrin.h> |
402 | /// |
403 | /// This intrinsic corresponds to the \c PHSUBD instruction. |
404 | /// |
405 | /// \param __a |
406 | /// A 64-bit vector of [2 x i32] containing one of the source operands. The |
407 | /// horizontal differences between the values are stored in the lower bits of |
408 | /// the destination. |
409 | /// \param __b |
410 | /// A 64-bit vector of [2 x i32] containing one of the source operands. The |
411 | /// horizontal differences between the values are stored in the upper bits of |
412 | /// the destination. |
413 | /// \returns A 64-bit vector of [2 x i32] containing the horizontal differences |
414 | /// of both operands. |
415 | static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
416 | _mm_hsub_pi32(__m64 __a, __m64 __b) |
417 | { |
418 | return (__m64)__builtin_ia32_phsubd((__v2si)__a, (__v2si)__b); |
419 | } |
420 | |
421 | /// Horizontally subtracts the adjacent pairs of values contained in 2 |
422 | /// packed 128-bit vectors of [8 x i16]. Positive differences greater than |
423 | /// 0x7FFF are saturated to 0x7FFF. Negative differences less than 0x8000 are |
424 | /// saturated to 0x8000. |
425 | /// |
426 | /// \headerfile <x86intrin.h> |
427 | /// |
428 | /// This intrinsic corresponds to the \c VPHSUBSW instruction. |
429 | /// |
430 | /// \param __a |
431 | /// A 128-bit vector of [8 x i16] containing one of the source operands. The |
432 | /// horizontal differences between the values are stored in the lower bits of |
433 | /// the destination. |
434 | /// \param __b |
435 | /// A 128-bit vector of [8 x i16] containing one of the source operands. The |
436 | /// horizontal differences between the values are stored in the upper bits of |
437 | /// the destination. |
438 | /// \returns A 128-bit vector of [8 x i16] containing the horizontal saturated |
439 | /// differences of both operands. |
440 | static __inline__ __m128i __DEFAULT_FN_ATTRS |
441 | _mm_hsubs_epi16(__m128i __a, __m128i __b) |
442 | { |
443 | return (__m128i)__builtin_ia32_phsubsw128((__v8hi)__a, (__v8hi)__b); |
444 | } |
445 | |
446 | /// Horizontally subtracts the adjacent pairs of values contained in 2 |
447 | /// packed 64-bit vectors of [4 x i16]. Positive differences greater than |
448 | /// 0x7FFF are saturated to 0x7FFF. Negative differences less than 0x8000 are |
449 | /// saturated to 0x8000. |
450 | /// |
451 | /// \headerfile <x86intrin.h> |
452 | /// |
453 | /// This intrinsic corresponds to the \c PHSUBSW instruction. |
454 | /// |
455 | /// \param __a |
456 | /// A 64-bit vector of [4 x i16] containing one of the source operands. The |
457 | /// horizontal differences between the values are stored in the lower bits of |
458 | /// the destination. |
459 | /// \param __b |
460 | /// A 64-bit vector of [4 x i16] containing one of the source operands. The |
461 | /// horizontal differences between the values are stored in the upper bits of |
462 | /// the destination. |
463 | /// \returns A 64-bit vector of [4 x i16] containing the horizontal saturated |
464 | /// differences of both operands. |
465 | static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
466 | _mm_hsubs_pi16(__m64 __a, __m64 __b) |
467 | { |
468 | return (__m64)__builtin_ia32_phsubsw((__v4hi)__a, (__v4hi)__b); |
469 | } |
470 | |
471 | /// Multiplies corresponding pairs of packed 8-bit unsigned integer |
472 | /// values contained in the first source operand and packed 8-bit signed |
473 | /// integer values contained in the second source operand, adds pairs of |
474 | /// contiguous products with signed saturation, and writes the 16-bit sums to |
475 | /// the corresponding bits in the destination. |
476 | /// |
477 | /// For example, bits [7:0] of both operands are multiplied, bits [15:8] of |
478 | /// both operands are multiplied, and the sum of both results is written to |
479 | /// bits [15:0] of the destination. |
480 | /// |
481 | /// \headerfile <x86intrin.h> |
482 | /// |
483 | /// This intrinsic corresponds to the \c VPMADDUBSW instruction. |
484 | /// |
485 | /// \param __a |
486 | /// A 128-bit integer vector containing the first source operand. |
487 | /// \param __b |
488 | /// A 128-bit integer vector containing the second source operand. |
489 | /// \returns A 128-bit integer vector containing the sums of products of both |
490 | /// operands: \n |
491 | /// \a R0 := (\a __a0 * \a __b0) + (\a __a1 * \a __b1) \n |
492 | /// \a R1 := (\a __a2 * \a __b2) + (\a __a3 * \a __b3) \n |
493 | /// \a R2 := (\a __a4 * \a __b4) + (\a __a5 * \a __b5) \n |
494 | /// \a R3 := (\a __a6 * \a __b6) + (\a __a7 * \a __b7) \n |
495 | /// \a R4 := (\a __a8 * \a __b8) + (\a __a9 * \a __b9) \n |
496 | /// \a R5 := (\a __a10 * \a __b10) + (\a __a11 * \a __b11) \n |
497 | /// \a R6 := (\a __a12 * \a __b12) + (\a __a13 * \a __b13) \n |
498 | /// \a R7 := (\a __a14 * \a __b14) + (\a __a15 * \a __b15) |
499 | static __inline__ __m128i __DEFAULT_FN_ATTRS |
500 | _mm_maddubs_epi16(__m128i __a, __m128i __b) |
501 | { |
502 | return (__m128i)__builtin_ia32_pmaddubsw128((__v16qi)__a, (__v16qi)__b); |
503 | } |
504 | |
505 | /// Multiplies corresponding pairs of packed 8-bit unsigned integer |
506 | /// values contained in the first source operand and packed 8-bit signed |
507 | /// integer values contained in the second source operand, adds pairs of |
508 | /// contiguous products with signed saturation, and writes the 16-bit sums to |
509 | /// the corresponding bits in the destination. |
510 | /// |
511 | /// For example, bits [7:0] of both operands are multiplied, bits [15:8] of |
512 | /// both operands are multiplied, and the sum of both results is written to |
513 | /// bits [15:0] of the destination. |
514 | /// |
515 | /// \headerfile <x86intrin.h> |
516 | /// |
517 | /// This intrinsic corresponds to the \c PMADDUBSW instruction. |
518 | /// |
519 | /// \param __a |
520 | /// A 64-bit integer vector containing the first source operand. |
521 | /// \param __b |
522 | /// A 64-bit integer vector containing the second source operand. |
523 | /// \returns A 64-bit integer vector containing the sums of products of both |
524 | /// operands: \n |
525 | /// \a R0 := (\a __a0 * \a __b0) + (\a __a1 * \a __b1) \n |
526 | /// \a R1 := (\a __a2 * \a __b2) + (\a __a3 * \a __b3) \n |
527 | /// \a R2 := (\a __a4 * \a __b4) + (\a __a5 * \a __b5) \n |
528 | /// \a R3 := (\a __a6 * \a __b6) + (\a __a7 * \a __b7) |
529 | static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
530 | _mm_maddubs_pi16(__m64 __a, __m64 __b) |
531 | { |
532 | return (__m64)__builtin_ia32_pmaddubsw((__v8qi)__a, (__v8qi)__b); |
533 | } |
534 | |
535 | /// Multiplies packed 16-bit signed integer values, truncates the 32-bit |
536 | /// products to the 18 most significant bits by right-shifting, rounds the |
537 | /// truncated value by adding 1, and writes bits [16:1] to the destination. |
538 | /// |
539 | /// \headerfile <x86intrin.h> |
540 | /// |
541 | /// This intrinsic corresponds to the \c VPMULHRSW instruction. |
542 | /// |
543 | /// \param __a |
544 | /// A 128-bit vector of [8 x i16] containing one of the source operands. |
545 | /// \param __b |
546 | /// A 128-bit vector of [8 x i16] containing one of the source operands. |
547 | /// \returns A 128-bit vector of [8 x i16] containing the rounded and scaled |
548 | /// products of both operands. |
549 | static __inline__ __m128i __DEFAULT_FN_ATTRS |
550 | _mm_mulhrs_epi16(__m128i __a, __m128i __b) |
551 | { |
552 | return (__m128i)__builtin_ia32_pmulhrsw128((__v8hi)__a, (__v8hi)__b); |
553 | } |
554 | |
555 | /// Multiplies packed 16-bit signed integer values, truncates the 32-bit |
556 | /// products to the 18 most significant bits by right-shifting, rounds the |
557 | /// truncated value by adding 1, and writes bits [16:1] to the destination. |
558 | /// |
559 | /// \headerfile <x86intrin.h> |
560 | /// |
561 | /// This intrinsic corresponds to the \c PMULHRSW instruction. |
562 | /// |
563 | /// \param __a |
564 | /// A 64-bit vector of [4 x i16] containing one of the source operands. |
565 | /// \param __b |
566 | /// A 64-bit vector of [4 x i16] containing one of the source operands. |
567 | /// \returns A 64-bit vector of [4 x i16] containing the rounded and scaled |
568 | /// products of both operands. |
569 | static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
570 | _mm_mulhrs_pi16(__m64 __a, __m64 __b) |
571 | { |
572 | return (__m64)__builtin_ia32_pmulhrsw((__v4hi)__a, (__v4hi)__b); |
573 | } |
574 | |
575 | /// Copies the 8-bit integers from a 128-bit integer vector to the |
576 | /// destination or clears 8-bit values in the destination, as specified by |
577 | /// the second source operand. |
578 | /// |
579 | /// \headerfile <x86intrin.h> |
580 | /// |
581 | /// This intrinsic corresponds to the \c VPSHUFB instruction. |
582 | /// |
583 | /// \param __a |
584 | /// A 128-bit integer vector containing the values to be copied. |
585 | /// \param __b |
586 | /// A 128-bit integer vector containing control bytes corresponding to |
587 | /// positions in the destination: |
588 | /// Bit 7: \n |
589 | /// 1: Clear the corresponding byte in the destination. \n |
590 | /// 0: Copy the selected source byte to the corresponding byte in the |
591 | /// destination. \n |
592 | /// Bits [6:4] Reserved. \n |
593 | /// Bits [3:0] select the source byte to be copied. |
594 | /// \returns A 128-bit integer vector containing the copied or cleared values. |
595 | static __inline__ __m128i __DEFAULT_FN_ATTRS |
596 | _mm_shuffle_epi8(__m128i __a, __m128i __b) |
597 | { |
598 | return (__m128i)__builtin_ia32_pshufb128((__v16qi)__a, (__v16qi)__b); |
599 | } |
600 | |
601 | /// Copies the 8-bit integers from a 64-bit integer vector to the |
602 | /// destination or clears 8-bit values in the destination, as specified by |
603 | /// the second source operand. |
604 | /// |
605 | /// \headerfile <x86intrin.h> |
606 | /// |
607 | /// This intrinsic corresponds to the \c PSHUFB instruction. |
608 | /// |
609 | /// \param __a |
610 | /// A 64-bit integer vector containing the values to be copied. |
611 | /// \param __b |
612 | /// A 64-bit integer vector containing control bytes corresponding to |
613 | /// positions in the destination: |
614 | /// Bit 7: \n |
615 | /// 1: Clear the corresponding byte in the destination. \n |
616 | /// 0: Copy the selected source byte to the corresponding byte in the |
617 | /// destination. \n |
618 | /// Bits [3:0] select the source byte to be copied. |
619 | /// \returns A 64-bit integer vector containing the copied or cleared values. |
620 | static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
621 | _mm_shuffle_pi8(__m64 __a, __m64 __b) |
622 | { |
623 | return (__m64)__builtin_ia32_pshufb((__v8qi)__a, (__v8qi)__b); |
624 | } |
625 | |
626 | /// For each 8-bit integer in the first source operand, perform one of |
627 | /// the following actions as specified by the second source operand. |
628 | /// |
629 | /// If the byte in the second source is negative, calculate the two's |
630 | /// complement of the corresponding byte in the first source, and write that |
631 | /// value to the destination. If the byte in the second source is positive, |
632 | /// copy the corresponding byte from the first source to the destination. If |
633 | /// the byte in the second source is zero, clear the corresponding byte in |
634 | /// the destination. |
635 | /// |
636 | /// \headerfile <x86intrin.h> |
637 | /// |
638 | /// This intrinsic corresponds to the \c VPSIGNB instruction. |
639 | /// |
640 | /// \param __a |
641 | /// A 128-bit integer vector containing the values to be copied. |
642 | /// \param __b |
643 | /// A 128-bit integer vector containing control bytes corresponding to |
644 | /// positions in the destination. |
645 | /// \returns A 128-bit integer vector containing the resultant values. |
646 | static __inline__ __m128i __DEFAULT_FN_ATTRS |
647 | _mm_sign_epi8(__m128i __a, __m128i __b) |
648 | { |
649 | return (__m128i)__builtin_ia32_psignb128((__v16qi)__a, (__v16qi)__b); |
650 | } |
651 | |
652 | /// For each 16-bit integer in the first source operand, perform one of |
653 | /// the following actions as specified by the second source operand. |
654 | /// |
655 | /// If the word in the second source is negative, calculate the two's |
656 | /// complement of the corresponding word in the first source, and write that |
657 | /// value to the destination. If the word in the second source is positive, |
658 | /// copy the corresponding word from the first source to the destination. If |
659 | /// the word in the second source is zero, clear the corresponding word in |
660 | /// the destination. |
661 | /// |
662 | /// \headerfile <x86intrin.h> |
663 | /// |
664 | /// This intrinsic corresponds to the \c VPSIGNW instruction. |
665 | /// |
666 | /// \param __a |
667 | /// A 128-bit integer vector containing the values to be copied. |
668 | /// \param __b |
669 | /// A 128-bit integer vector containing control words corresponding to |
670 | /// positions in the destination. |
671 | /// \returns A 128-bit integer vector containing the resultant values. |
672 | static __inline__ __m128i __DEFAULT_FN_ATTRS |
673 | _mm_sign_epi16(__m128i __a, __m128i __b) |
674 | { |
675 | return (__m128i)__builtin_ia32_psignw128((__v8hi)__a, (__v8hi)__b); |
676 | } |
677 | |
678 | /// For each 32-bit integer in the first source operand, perform one of |
679 | /// the following actions as specified by the second source operand. |
680 | /// |
681 | /// If the doubleword in the second source is negative, calculate the two's |
682 | /// complement of the corresponding word in the first source, and write that |
683 | /// value to the destination. If the doubleword in the second source is |
684 | /// positive, copy the corresponding word from the first source to the |
685 | /// destination. If the doubleword in the second source is zero, clear the |
686 | /// corresponding word in the destination. |
687 | /// |
688 | /// \headerfile <x86intrin.h> |
689 | /// |
690 | /// This intrinsic corresponds to the \c VPSIGND instruction. |
691 | /// |
692 | /// \param __a |
693 | /// A 128-bit integer vector containing the values to be copied. |
694 | /// \param __b |
695 | /// A 128-bit integer vector containing control doublewords corresponding to |
696 | /// positions in the destination. |
697 | /// \returns A 128-bit integer vector containing the resultant values. |
698 | static __inline__ __m128i __DEFAULT_FN_ATTRS |
699 | _mm_sign_epi32(__m128i __a, __m128i __b) |
700 | { |
701 | return (__m128i)__builtin_ia32_psignd128((__v4si)__a, (__v4si)__b); |
702 | } |
703 | |
704 | /// For each 8-bit integer in the first source operand, perform one of |
705 | /// the following actions as specified by the second source operand. |
706 | /// |
707 | /// If the byte in the second source is negative, calculate the two's |
708 | /// complement of the corresponding byte in the first source, and write that |
709 | /// value to the destination. If the byte in the second source is positive, |
710 | /// copy the corresponding byte from the first source to the destination. If |
711 | /// the byte in the second source is zero, clear the corresponding byte in |
712 | /// the destination. |
713 | /// |
714 | /// \headerfile <x86intrin.h> |
715 | /// |
716 | /// This intrinsic corresponds to the \c PSIGNB instruction. |
717 | /// |
718 | /// \param __a |
719 | /// A 64-bit integer vector containing the values to be copied. |
720 | /// \param __b |
721 | /// A 64-bit integer vector containing control bytes corresponding to |
722 | /// positions in the destination. |
723 | /// \returns A 64-bit integer vector containing the resultant values. |
724 | static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
725 | _mm_sign_pi8(__m64 __a, __m64 __b) |
726 | { |
727 | return (__m64)__builtin_ia32_psignb((__v8qi)__a, (__v8qi)__b); |
728 | } |
729 | |
730 | /// For each 16-bit integer in the first source operand, perform one of |
731 | /// the following actions as specified by the second source operand. |
732 | /// |
733 | /// If the word in the second source is negative, calculate the two's |
734 | /// complement of the corresponding word in the first source, and write that |
735 | /// value to the destination. If the word in the second source is positive, |
736 | /// copy the corresponding word from the first source to the destination. If |
737 | /// the word in the second source is zero, clear the corresponding word in |
738 | /// the destination. |
739 | /// |
740 | /// \headerfile <x86intrin.h> |
741 | /// |
742 | /// This intrinsic corresponds to the \c PSIGNW instruction. |
743 | /// |
744 | /// \param __a |
745 | /// A 64-bit integer vector containing the values to be copied. |
746 | /// \param __b |
747 | /// A 64-bit integer vector containing control words corresponding to |
748 | /// positions in the destination. |
749 | /// \returns A 64-bit integer vector containing the resultant values. |
750 | static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
751 | _mm_sign_pi16(__m64 __a, __m64 __b) |
752 | { |
753 | return (__m64)__builtin_ia32_psignw((__v4hi)__a, (__v4hi)__b); |
754 | } |
755 | |
756 | /// For each 32-bit integer in the first source operand, perform one of |
757 | /// the following actions as specified by the second source operand. |
758 | /// |
759 | /// If the doubleword in the second source is negative, calculate the two's |
760 | /// complement of the corresponding doubleword in the first source, and |
761 | /// write that value to the destination. If the doubleword in the second |
762 | /// source is positive, copy the corresponding doubleword from the first |
763 | /// source to the destination. If the doubleword in the second source is |
764 | /// zero, clear the corresponding doubleword in the destination. |
765 | /// |
766 | /// \headerfile <x86intrin.h> |
767 | /// |
768 | /// This intrinsic corresponds to the \c PSIGND instruction. |
769 | /// |
770 | /// \param __a |
771 | /// A 64-bit integer vector containing the values to be copied. |
772 | /// \param __b |
773 | /// A 64-bit integer vector containing two control doublewords corresponding |
774 | /// to positions in the destination. |
775 | /// \returns A 64-bit integer vector containing the resultant values. |
776 | static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
777 | _mm_sign_pi32(__m64 __a, __m64 __b) |
778 | { |
779 | return (__m64)__builtin_ia32_psignd((__v2si)__a, (__v2si)__b); |
780 | } |
781 | |
782 | #undef __DEFAULT_FN_ATTRS |
783 | #undef __DEFAULT_FN_ATTRS_MMX |
784 | |
785 | #endif /* __TMMINTRIN_H */ |
786 |