2 * MMX optimized DSP utils
3 * Copyright (c) 2000, 2001 Fabrice Bellard
4 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
6 * MMX optimization by Nick Kurshev <nickols_k@mail.ru>
8 * This file is part of FFmpeg.
10 * FFmpeg is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU Lesser General Public
12 * License as published by the Free Software Foundation; either
13 * version 2.1 of the License, or (at your option) any later version.
15 * FFmpeg is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * Lesser General Public License for more details.
20 * You should have received a copy of the GNU Lesser General Public
21 * License along with FFmpeg; if not, write to the Free Software
22 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
25 #include "libavutil/attributes.h"
26 #include "libavutil/cpu.h"
27 #include "libavutil/x86/asm.h"
28 #include "libavutil/x86/cpu.h"
29 #include "libavcodec/dct.h"
30 #include "libavcodec/dsputil.h"
31 #include "libavcodec/mpegvideo.h"
32 #include "dsputil_x86.h"
34 void ff_get_pixels_mmx(int16_t *block, const uint8_t *pixels, int line_size);
35 void ff_get_pixels_sse2(int16_t *block, const uint8_t *pixels, int line_size);
36 void ff_diff_pixels_mmx(int16_t *block, const uint8_t *s1, const uint8_t *s2,
38 void ff_diff_pixels_sse2(int16_t *block, const uint8_t *s1, const uint8_t *s2,
40 int ff_pix_sum16_mmx(uint8_t *pix, int line_size);
41 int ff_pix_sum16_sse2(uint8_t *pix, int line_size);
42 int ff_pix_norm1_mmx(uint8_t *pix, int line_size);
43 int ff_pix_norm1_sse2(uint8_t *pix, int line_size);
44 int ff_sum_abs_dctelem_mmx(int16_t *block);
45 int ff_sum_abs_dctelem_mmxext(int16_t *block);
46 int ff_sum_abs_dctelem_sse2(int16_t *block);
47 int ff_sum_abs_dctelem_ssse3(int16_t *block);
51 static int sse8_mmx(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
59 "pxor %%mm0, %%mm0 \n" /* mm0 = 0 */
60 "pxor %%mm7, %%mm7 \n" /* mm7 holds the sum */
62 "movq (%0), %%mm1 \n" /* mm1 = pix1[0][0 - 7] */
63 "movq (%1), %%mm2 \n" /* mm2 = pix2[0][0 - 7] */
64 "movq (%0, %3), %%mm3 \n" /* mm3 = pix1[1][0 - 7] */
65 "movq (%1, %3), %%mm4 \n" /* mm4 = pix2[1][0 - 7] */
67 /* todo: mm1-mm2, mm3-mm4 */
68 /* algo: subtract mm1 from mm2 with saturation and vice versa */
69 /* OR the results to get absolute difference */
70 "movq %%mm1, %%mm5 \n"
71 "movq %%mm3, %%mm6 \n"
72 "psubusb %%mm2, %%mm1 \n"
73 "psubusb %%mm4, %%mm3 \n"
74 "psubusb %%mm5, %%mm2 \n"
75 "psubusb %%mm6, %%mm4 \n"
80 /* now convert to 16-bit vectors so we can square them */
81 "movq %%mm2, %%mm1 \n"
82 "movq %%mm4, %%mm3 \n"
84 "punpckhbw %%mm0, %%mm2 \n"
85 "punpckhbw %%mm0, %%mm4 \n"
86 "punpcklbw %%mm0, %%mm1 \n" /* mm1 now spread over (mm1, mm2) */
87 "punpcklbw %%mm0, %%mm3 \n" /* mm4 now spread over (mm3, mm4) */
89 "pmaddwd %%mm2, %%mm2 \n"
90 "pmaddwd %%mm4, %%mm4 \n"
91 "pmaddwd %%mm1, %%mm1 \n"
92 "pmaddwd %%mm3, %%mm3 \n"
94 "lea (%0, %3, 2), %0 \n" /* pix1 += 2 * line_size */
95 "lea (%1, %3, 2), %1 \n" /* pix2 += 2 * line_size */
97 "paddd %%mm2, %%mm1 \n"
98 "paddd %%mm4, %%mm3 \n"
99 "paddd %%mm1, %%mm7 \n"
100 "paddd %%mm3, %%mm7 \n"
105 "movq %%mm7, %%mm1 \n"
106 "psrlq $32, %%mm7 \n" /* shift hi dword to lo */
107 "paddd %%mm7, %%mm1 \n"
109 : "+r" (pix1), "+r" (pix2), "=r" (tmp)
110 : "r" ((x86_reg) line_size), "m" (h)
116 static int sse16_mmx(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
117 int line_size, int h)
123 "pxor %%mm0, %%mm0\n" /* mm0 = 0 */
124 "pxor %%mm7, %%mm7\n" /* mm7 holds the sum */
126 "movq (%0), %%mm1\n" /* mm1 = pix1[0 - 7] */
127 "movq (%1), %%mm2\n" /* mm2 = pix2[0 - 7] */
128 "movq 8(%0), %%mm3\n" /* mm3 = pix1[8 - 15] */
129 "movq 8(%1), %%mm4\n" /* mm4 = pix2[8 - 15] */
131 /* todo: mm1-mm2, mm3-mm4 */
132 /* algo: subtract mm1 from mm2 with saturation and vice versa */
133 /* OR the results to get absolute difference */
134 "movq %%mm1, %%mm5\n"
135 "movq %%mm3, %%mm6\n"
136 "psubusb %%mm2, %%mm1\n"
137 "psubusb %%mm4, %%mm3\n"
138 "psubusb %%mm5, %%mm2\n"
139 "psubusb %%mm6, %%mm4\n"
144 /* now convert to 16-bit vectors so we can square them */
145 "movq %%mm2, %%mm1\n"
146 "movq %%mm4, %%mm3\n"
148 "punpckhbw %%mm0, %%mm2\n"
149 "punpckhbw %%mm0, %%mm4\n"
150 "punpcklbw %%mm0, %%mm1\n" /* mm1 now spread over (mm1, mm2) */
151 "punpcklbw %%mm0, %%mm3\n" /* mm4 now spread over (mm3, mm4) */
153 "pmaddwd %%mm2, %%mm2\n"
154 "pmaddwd %%mm4, %%mm4\n"
155 "pmaddwd %%mm1, %%mm1\n"
156 "pmaddwd %%mm3, %%mm3\n"
161 "paddd %%mm2, %%mm1\n"
162 "paddd %%mm4, %%mm3\n"
163 "paddd %%mm1, %%mm7\n"
164 "paddd %%mm3, %%mm7\n"
169 "movq %%mm7, %%mm1\n"
170 "psrlq $32, %%mm7\n" /* shift hi dword to lo */
171 "paddd %%mm7, %%mm1\n"
173 : "+r" (pix1), "+r" (pix2), "=r" (tmp)
174 : "r" ((x86_reg) line_size), "m" (h)
180 static int hf_noise8_mmx(uint8_t *pix1, int line_size, int h)
186 "pxor %%mm7, %%mm7\n"
187 "pxor %%mm6, %%mm6\n"
190 "movq %%mm0, %%mm1\n"
194 "movq %%mm0, %%mm2\n"
195 "movq %%mm1, %%mm3\n"
196 "punpcklbw %%mm7, %%mm0\n"
197 "punpcklbw %%mm7, %%mm1\n"
198 "punpckhbw %%mm7, %%mm2\n"
199 "punpckhbw %%mm7, %%mm3\n"
200 "psubw %%mm1, %%mm0\n"
201 "psubw %%mm3, %%mm2\n"
206 "movq %%mm4, %%mm1\n"
210 "movq %%mm4, %%mm5\n"
211 "movq %%mm1, %%mm3\n"
212 "punpcklbw %%mm7, %%mm4\n"
213 "punpcklbw %%mm7, %%mm1\n"
214 "punpckhbw %%mm7, %%mm5\n"
215 "punpckhbw %%mm7, %%mm3\n"
216 "psubw %%mm1, %%mm4\n"
217 "psubw %%mm3, %%mm5\n"
218 "psubw %%mm4, %%mm0\n"
219 "psubw %%mm5, %%mm2\n"
220 "pxor %%mm3, %%mm3\n"
221 "pxor %%mm1, %%mm1\n"
222 "pcmpgtw %%mm0, %%mm3\n\t"
223 "pcmpgtw %%mm2, %%mm1\n\t"
224 "pxor %%mm3, %%mm0\n"
225 "pxor %%mm1, %%mm2\n"
226 "psubw %%mm3, %%mm0\n"
227 "psubw %%mm1, %%mm2\n"
228 "paddw %%mm0, %%mm2\n"
229 "paddw %%mm2, %%mm6\n"
235 "movq %%mm0, %%mm1\n"
239 "movq %%mm0, %%mm2\n"
240 "movq %%mm1, %%mm3\n"
241 "punpcklbw %%mm7, %%mm0\n"
242 "punpcklbw %%mm7, %%mm1\n"
243 "punpckhbw %%mm7, %%mm2\n"
244 "punpckhbw %%mm7, %%mm3\n"
245 "psubw %%mm1, %%mm0\n"
246 "psubw %%mm3, %%mm2\n"
247 "psubw %%mm0, %%mm4\n"
248 "psubw %%mm2, %%mm5\n"
249 "pxor %%mm3, %%mm3\n"
250 "pxor %%mm1, %%mm1\n"
251 "pcmpgtw %%mm4, %%mm3\n\t"
252 "pcmpgtw %%mm5, %%mm1\n\t"
253 "pxor %%mm3, %%mm4\n"
254 "pxor %%mm1, %%mm5\n"
255 "psubw %%mm3, %%mm4\n"
256 "psubw %%mm1, %%mm5\n"
257 "paddw %%mm4, %%mm5\n"
258 "paddw %%mm5, %%mm6\n"
263 "movq %%mm4, %%mm1\n"
267 "movq %%mm4, %%mm5\n"
268 "movq %%mm1, %%mm3\n"
269 "punpcklbw %%mm7, %%mm4\n"
270 "punpcklbw %%mm7, %%mm1\n"
271 "punpckhbw %%mm7, %%mm5\n"
272 "punpckhbw %%mm7, %%mm3\n"
273 "psubw %%mm1, %%mm4\n"
274 "psubw %%mm3, %%mm5\n"
275 "psubw %%mm4, %%mm0\n"
276 "psubw %%mm5, %%mm2\n"
277 "pxor %%mm3, %%mm3\n"
278 "pxor %%mm1, %%mm1\n"
279 "pcmpgtw %%mm0, %%mm3\n\t"
280 "pcmpgtw %%mm2, %%mm1\n\t"
281 "pxor %%mm3, %%mm0\n"
282 "pxor %%mm1, %%mm2\n"
283 "psubw %%mm3, %%mm0\n"
284 "psubw %%mm1, %%mm2\n"
285 "paddw %%mm0, %%mm2\n"
286 "paddw %%mm2, %%mm6\n"
292 "movq %%mm6, %%mm0\n"
293 "punpcklwd %%mm7, %%mm0\n"
294 "punpckhwd %%mm7, %%mm6\n"
295 "paddd %%mm0, %%mm6\n"
297 "movq %%mm6, %%mm0\n"
299 "paddd %%mm6, %%mm0\n"
301 : "+r" (pix1), "=r" (tmp)
302 : "r" ((x86_reg) line_size), "g" (h - 2)
308 static int hf_noise16_mmx(uint8_t *pix1, int line_size, int h)
315 "pxor %%mm7, %%mm7\n"
316 "pxor %%mm6, %%mm6\n"
319 "movq 1(%0), %%mm1\n"
320 "movq %%mm0, %%mm2\n"
321 "movq %%mm1, %%mm3\n"
322 "punpcklbw %%mm7, %%mm0\n"
323 "punpcklbw %%mm7, %%mm1\n"
324 "punpckhbw %%mm7, %%mm2\n"
325 "punpckhbw %%mm7, %%mm3\n"
326 "psubw %%mm1, %%mm0\n"
327 "psubw %%mm3, %%mm2\n"
332 "movq 1(%0), %%mm1\n"
333 "movq %%mm4, %%mm5\n"
334 "movq %%mm1, %%mm3\n"
335 "punpcklbw %%mm7, %%mm4\n"
336 "punpcklbw %%mm7, %%mm1\n"
337 "punpckhbw %%mm7, %%mm5\n"
338 "punpckhbw %%mm7, %%mm3\n"
339 "psubw %%mm1, %%mm4\n"
340 "psubw %%mm3, %%mm5\n"
341 "psubw %%mm4, %%mm0\n"
342 "psubw %%mm5, %%mm2\n"
343 "pxor %%mm3, %%mm3\n"
344 "pxor %%mm1, %%mm1\n"
345 "pcmpgtw %%mm0, %%mm3\n\t"
346 "pcmpgtw %%mm2, %%mm1\n\t"
347 "pxor %%mm3, %%mm0\n"
348 "pxor %%mm1, %%mm2\n"
349 "psubw %%mm3, %%mm0\n"
350 "psubw %%mm1, %%mm2\n"
351 "paddw %%mm0, %%mm2\n"
352 "paddw %%mm2, %%mm6\n"
358 "movq 1(%0), %%mm1\n"
359 "movq %%mm0, %%mm2\n"
360 "movq %%mm1, %%mm3\n"
361 "punpcklbw %%mm7, %%mm0\n"
362 "punpcklbw %%mm7, %%mm1\n"
363 "punpckhbw %%mm7, %%mm2\n"
364 "punpckhbw %%mm7, %%mm3\n"
365 "psubw %%mm1, %%mm0\n"
366 "psubw %%mm3, %%mm2\n"
367 "psubw %%mm0, %%mm4\n"
368 "psubw %%mm2, %%mm5\n"
369 "pxor %%mm3, %%mm3\n"
370 "pxor %%mm1, %%mm1\n"
371 "pcmpgtw %%mm4, %%mm3\n\t"
372 "pcmpgtw %%mm5, %%mm1\n\t"
373 "pxor %%mm3, %%mm4\n"
374 "pxor %%mm1, %%mm5\n"
375 "psubw %%mm3, %%mm4\n"
376 "psubw %%mm1, %%mm5\n"
377 "paddw %%mm4, %%mm5\n"
378 "paddw %%mm5, %%mm6\n"
383 "movq 1(%0), %%mm1\n"
384 "movq %%mm4, %%mm5\n"
385 "movq %%mm1, %%mm3\n"
386 "punpcklbw %%mm7, %%mm4\n"
387 "punpcklbw %%mm7, %%mm1\n"
388 "punpckhbw %%mm7, %%mm5\n"
389 "punpckhbw %%mm7, %%mm3\n"
390 "psubw %%mm1, %%mm4\n"
391 "psubw %%mm3, %%mm5\n"
392 "psubw %%mm4, %%mm0\n"
393 "psubw %%mm5, %%mm2\n"
394 "pxor %%mm3, %%mm3\n"
395 "pxor %%mm1, %%mm1\n"
396 "pcmpgtw %%mm0, %%mm3\n\t"
397 "pcmpgtw %%mm2, %%mm1\n\t"
398 "pxor %%mm3, %%mm0\n"
399 "pxor %%mm1, %%mm2\n"
400 "psubw %%mm3, %%mm0\n"
401 "psubw %%mm1, %%mm2\n"
402 "paddw %%mm0, %%mm2\n"
403 "paddw %%mm2, %%mm6\n"
409 "movq %%mm6, %%mm0\n"
410 "punpcklwd %%mm7, %%mm0\n"
411 "punpckhwd %%mm7, %%mm6\n"
412 "paddd %%mm0, %%mm6\n"
414 "movq %%mm6, %%mm0\n"
416 "paddd %%mm6, %%mm0\n"
418 : "+r" (pix1), "=r" (tmp)
419 : "r" ((x86_reg) line_size), "g" (h - 2)
422 return tmp + hf_noise8_mmx(pix + 8, line_size, h);
425 static int nsse16_mmx(MpegEncContext *c, uint8_t *pix1, uint8_t *pix2,
426 int line_size, int h)
431 score1 = c->dsp.sse[0](c, pix1, pix2, line_size, h);
433 score1 = sse16_mmx(c, pix1, pix2, line_size, h);
434 score2 = hf_noise16_mmx(pix1, line_size, h) -
435 hf_noise16_mmx(pix2, line_size, h);
438 return score1 + FFABS(score2) * c->avctx->nsse_weight;
440 return score1 + FFABS(score2) * 8;
443 static int nsse8_mmx(MpegEncContext *c, uint8_t *pix1, uint8_t *pix2,
444 int line_size, int h)
446 int score1 = sse8_mmx(c, pix1, pix2, line_size, h);
447 int score2 = hf_noise8_mmx(pix1, line_size, h) -
448 hf_noise8_mmx(pix2, line_size, h);
451 return score1 + FFABS(score2) * c->avctx->nsse_weight;
453 return score1 + FFABS(score2) * 8;
456 static int vsad_intra16_mmx(MpegEncContext *v, uint8_t *pix, uint8_t *dummy,
457 int line_size, int h)
461 av_assert2((((int) pix) & 7) == 0);
462 av_assert2((line_size & 7) == 0);
464 #define SUM(in0, in1, out0, out1) \
465 "movq (%0), %%mm2\n" \
466 "movq 8(%0), %%mm3\n" \
468 "movq %%mm2, " #out0 "\n" \
469 "movq %%mm3, " #out1 "\n" \
470 "psubusb " #in0 ", %%mm2\n" \
471 "psubusb " #in1 ", %%mm3\n" \
472 "psubusb " #out0 ", " #in0 "\n" \
473 "psubusb " #out1 ", " #in1 "\n" \
474 "por %%mm2, " #in0 "\n" \
475 "por %%mm3, " #in1 "\n" \
476 "movq " #in0 ", %%mm2\n" \
477 "movq " #in1 ", %%mm3\n" \
478 "punpcklbw %%mm7, " #in0 "\n" \
479 "punpcklbw %%mm7, " #in1 "\n" \
480 "punpckhbw %%mm7, %%mm2\n" \
481 "punpckhbw %%mm7, %%mm3\n" \
482 "paddw " #in1 ", " #in0 "\n" \
483 "paddw %%mm3, %%mm2\n" \
484 "paddw %%mm2, " #in0 "\n" \
485 "paddw " #in0 ", %%mm6\n"
490 "pxor %%mm6, %%mm6\n"
491 "pxor %%mm7, %%mm7\n"
493 "movq 8(%0), %%mm1\n"
498 SUM(%%mm4, %%mm5, %%mm0, %%mm1)
500 SUM(%%mm0, %%mm1, %%mm4, %%mm5)
505 "movq %%mm6, %%mm0\n"
507 "paddw %%mm6, %%mm0\n"
508 "movq %%mm0, %%mm6\n"
510 "paddw %%mm6, %%mm0\n"
512 : "+r" (pix), "=r" (tmp)
513 : "r" ((x86_reg) line_size), "m" (h)
520 static int vsad_intra16_mmxext(MpegEncContext *v, uint8_t *pix, uint8_t *dummy,
521 int line_size, int h)
525 av_assert2((((int) pix) & 7) == 0);
526 av_assert2((line_size & 7) == 0);
528 #define SUM(in0, in1, out0, out1) \
529 "movq (%0), " #out0 "\n" \
530 "movq 8(%0), " #out1 "\n" \
532 "psadbw " #out0 ", " #in0 "\n" \
533 "psadbw " #out1 ", " #in1 "\n" \
534 "paddw " #in1 ", " #in0 "\n" \
535 "paddw " #in0 ", %%mm6\n"
539 "pxor %%mm6, %%mm6\n"
540 "pxor %%mm7, %%mm7\n"
542 "movq 8(%0), %%mm1\n"
547 SUM(%%mm4, %%mm5, %%mm0, %%mm1)
549 SUM(%%mm0, %%mm1, %%mm4, %%mm5)
555 : "+r" (pix), "=r" (tmp)
556 : "r" ((x86_reg) line_size), "m" (h)
563 static int vsad16_mmx(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
564 int line_size, int h)
568 av_assert2((((int) pix1) & 7) == 0);
569 av_assert2((((int) pix2) & 7) == 0);
570 av_assert2((line_size & 7) == 0);
572 #define SUM(in0, in1, out0, out1) \
573 "movq (%0), %%mm2\n" \
574 "movq (%1), " #out0 "\n" \
575 "movq 8(%0), %%mm3\n" \
576 "movq 8(%1), " #out1 "\n" \
579 "psubb " #out0 ", %%mm2\n" \
580 "psubb " #out1 ", %%mm3\n" \
581 "pxor %%mm7, %%mm2\n" \
582 "pxor %%mm7, %%mm3\n" \
583 "movq %%mm2, " #out0 "\n" \
584 "movq %%mm3, " #out1 "\n" \
585 "psubusb " #in0 ", %%mm2\n" \
586 "psubusb " #in1 ", %%mm3\n" \
587 "psubusb " #out0 ", " #in0 "\n" \
588 "psubusb " #out1 ", " #in1 "\n" \
589 "por %%mm2, " #in0 "\n" \
590 "por %%mm3, " #in1 "\n" \
591 "movq " #in0 ", %%mm2\n" \
592 "movq " #in1 ", %%mm3\n" \
593 "punpcklbw %%mm7, " #in0 "\n" \
594 "punpcklbw %%mm7, " #in1 "\n" \
595 "punpckhbw %%mm7, %%mm2\n" \
596 "punpckhbw %%mm7, %%mm3\n" \
597 "paddw " #in1 ", " #in0 "\n" \
598 "paddw %%mm3, %%mm2\n" \
599 "paddw %%mm2, " #in0 "\n" \
600 "paddw " #in0 ", %%mm6\n"
605 "pxor %%mm6, %%mm6\n"
606 "pcmpeqw %%mm7, %%mm7\n"
608 "packsswb %%mm7, %%mm7\n"
611 "movq 8(%0), %%mm1\n"
612 "movq 8(%1), %%mm3\n"
615 "psubb %%mm2, %%mm0\n"
616 "psubb %%mm3, %%mm1\n"
617 "pxor %%mm7, %%mm0\n"
618 "pxor %%mm7, %%mm1\n"
622 SUM(%%mm4, %%mm5, %%mm0, %%mm1)
624 SUM(%%mm0, %%mm1, %%mm4, %%mm5)
629 "movq %%mm6, %%mm0\n"
631 "paddw %%mm6, %%mm0\n"
632 "movq %%mm0, %%mm6\n"
634 "paddw %%mm6, %%mm0\n"
636 : "+r" (pix1), "+r" (pix2), "=r" (tmp)
637 : "r" ((x86_reg) line_size), "m" (h)
644 static int vsad16_mmxext(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
645 int line_size, int h)
649 av_assert2((((int) pix1) & 7) == 0);
650 av_assert2((((int) pix2) & 7) == 0);
651 av_assert2((line_size & 7) == 0);
653 #define SUM(in0, in1, out0, out1) \
654 "movq (%0), " #out0 "\n" \
655 "movq (%1), %%mm2\n" \
656 "movq 8(%0), " #out1 "\n" \
657 "movq 8(%1), %%mm3\n" \
660 "psubb %%mm2, " #out0 "\n" \
661 "psubb %%mm3, " #out1 "\n" \
662 "pxor %%mm7, " #out0 "\n" \
663 "pxor %%mm7, " #out1 "\n" \
664 "psadbw " #out0 ", " #in0 "\n" \
665 "psadbw " #out1 ", " #in1 "\n" \
666 "paddw " #in1 ", " #in0 "\n" \
667 "paddw " #in0 ", %%mm6\n "
671 "pxor %%mm6, %%mm6\n"
672 "pcmpeqw %%mm7, %%mm7\n"
674 "packsswb %%mm7, %%mm7\n"
677 "movq 8(%0), %%mm1\n"
678 "movq 8(%1), %%mm3\n"
681 "psubb %%mm2, %%mm0\n"
682 "psubb %%mm3, %%mm1\n"
683 "pxor %%mm7, %%mm0\n"
684 "pxor %%mm7, %%mm1\n"
688 SUM(%%mm4, %%mm5, %%mm0, %%mm1)
690 SUM(%%mm0, %%mm1, %%mm4, %%mm5)
696 : "+r" (pix1), "+r" (pix2), "=r" (tmp)
697 : "r" ((x86_reg) line_size), "m" (h)
705 static int ssd_int8_vs_int16_mmx(const int8_t *pix1, const int16_t *pix2,
712 "pxor %%mm4, %%mm4 \n"
715 "movq (%2, %0), %%mm2 \n"
716 "movq (%3, %0, 2), %%mm0 \n"
717 "movq 8(%3, %0, 2), %%mm1 \n"
718 "punpckhbw %%mm2, %%mm3 \n"
719 "punpcklbw %%mm2, %%mm2 \n"
722 "psubw %%mm3, %%mm1 \n"
723 "psubw %%mm2, %%mm0 \n"
724 "pmaddwd %%mm1, %%mm1 \n"
725 "pmaddwd %%mm0, %%mm0 \n"
726 "paddd %%mm1, %%mm4 \n"
727 "paddd %%mm0, %%mm4 \n"
729 "movq %%mm4, %%mm3 \n"
730 "psrlq $32, %%mm3 \n"
731 "paddd %%mm3, %%mm4 \n"
733 : "+r" (i), "=r" (sum)
734 : "r" (pix1), "r" (pix2));
739 #define PHADDD(a, t) \
740 "movq " #a ", " #t " \n\t" \
741 "psrlq $32, " #a " \n\t" \
742 "paddd " #t ", " #a " \n\t"
745 * pmulhw: dst[0 - 15] = (src[0 - 15] * dst[0 - 15])[16 - 31]
746 * pmulhrw: dst[0 - 15] = (src[0 - 15] * dst[0 - 15] + 0x8000)[16 - 31]
747 * pmulhrsw: dst[0 - 15] = (src[0 - 15] * dst[0 - 15] + 0x4000)[15 - 30]
749 #define PMULHRW(x, y, s, o) \
750 "pmulhw " #s ", " #x " \n\t" \
751 "pmulhw " #s ", " #y " \n\t" \
752 "paddw " #o ", " #x " \n\t" \
753 "paddw " #o ", " #y " \n\t" \
754 "psraw $1, " #x " \n\t" \
755 "psraw $1, " #y " \n\t"
756 #define DEF(x) x ## _mmx
757 #define SET_RND MOVQ_WONE
758 #define SCALE_OFFSET 1
760 #include "dsputil_qns_template.c"
767 #define DEF(x) x ## _3dnow
769 #define SCALE_OFFSET 0
770 #define PMULHRW(x, y, s, o) \
771 "pmulhrw " #s ", " #x " \n\t" \
772 "pmulhrw " #s ", " #y " \n\t"
774 #include "dsputil_qns_template.c"
781 #if HAVE_SSSE3_INLINE
783 #define DEF(x) x ## _ssse3
785 #define SCALE_OFFSET -1
787 #define PHADDD(a, t) \
788 "pshufw $0x0E, " #a ", " #t " \n\t" \
789 /* faster than phaddd on core2 */ \
790 "paddd " #t ", " #a " \n\t"
792 #define PMULHRW(x, y, s, o) \
793 "pmulhrsw " #s ", " #x " \n\t" \
794 "pmulhrsw " #s ", " #y " \n\t"
796 #include "dsputil_qns_template.c"
803 #endif /* HAVE_SSSE3_INLINE */
805 #endif /* HAVE_INLINE_ASM */
807 int ff_sse16_sse2(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
808 int line_size, int h);
810 #define hadamard_func(cpu) \
811 int ff_hadamard8_diff_ ## cpu(MpegEncContext *s, uint8_t *src1, \
812 uint8_t *src2, int stride, int h); \
813 int ff_hadamard8_diff16_ ## cpu(MpegEncContext *s, uint8_t *src1, \
814 uint8_t *src2, int stride, int h);
817 hadamard_func(mmxext)
821 av_cold void ff_dsputilenc_init_mmx(DSPContext *c, AVCodecContext *avctx,
822 unsigned high_bit_depth)
824 int cpu_flags = av_get_cpu_flags();
825 const int dct_algo = avctx->dct_algo;
827 if (EXTERNAL_MMX(cpu_flags)) {
829 c->get_pixels = ff_get_pixels_mmx;
830 c->diff_pixels = ff_diff_pixels_mmx;
831 c->pix_sum = ff_pix_sum16_mmx;
832 c->pix_norm1 = ff_pix_norm1_mmx;
835 if (EXTERNAL_SSE2(cpu_flags))
837 c->get_pixels = ff_get_pixels_sse2;
840 if (INLINE_MMX(cpu_flags)) {
841 if (!high_bit_depth &&
842 (dct_algo == FF_DCT_AUTO || dct_algo == FF_DCT_MMX))
843 c->fdct = ff_fdct_mmx;
845 c->sse[0] = sse16_mmx;
846 c->sse[1] = sse8_mmx;
847 c->vsad[4] = vsad_intra16_mmx;
849 c->nsse[0] = nsse16_mmx;
850 c->nsse[1] = nsse8_mmx;
851 if (!(avctx->flags & CODEC_FLAG_BITEXACT)) {
852 c->vsad[0] = vsad16_mmx;
853 c->try_8x8basis = try_8x8basis_mmx;
855 c->add_8x8basis = add_8x8basis_mmx;
857 c->ssd_int8_vs_int16 = ssd_int8_vs_int16_mmx;
860 if (INLINE_AMD3DNOW(cpu_flags)) {
861 if (!(avctx->flags & CODEC_FLAG_BITEXACT)) {
862 c->try_8x8basis = try_8x8basis_3dnow;
864 c->add_8x8basis = add_8x8basis_3dnow;
867 if (INLINE_MMXEXT(cpu_flags)) {
868 if (!high_bit_depth &&
869 (dct_algo == FF_DCT_AUTO || dct_algo == FF_DCT_MMX))
870 c->fdct = ff_fdct_mmxext;
872 c->vsad[4] = vsad_intra16_mmxext;
874 if (!(avctx->flags & CODEC_FLAG_BITEXACT)) {
875 c->vsad[0] = vsad16_mmxext;
879 if (INLINE_SSE2(cpu_flags)) {
880 if (!high_bit_depth &&
881 (dct_algo == FF_DCT_AUTO || dct_algo == FF_DCT_MMX))
882 c->fdct = ff_fdct_sse2;
885 #if HAVE_SSSE3_INLINE
886 if (INLINE_SSSE3(cpu_flags)) {
887 if (!(avctx->flags & CODEC_FLAG_BITEXACT)) {
888 c->try_8x8basis = try_8x8basis_ssse3;
890 c->add_8x8basis = add_8x8basis_ssse3;
893 #endif /* HAVE_INLINE_ASM */
895 if (EXTERNAL_MMX(cpu_flags)) {
896 c->hadamard8_diff[0] = ff_hadamard8_diff16_mmx;
897 c->hadamard8_diff[1] = ff_hadamard8_diff_mmx;
898 c->sum_abs_dctelem = ff_sum_abs_dctelem_mmx;
901 if (EXTERNAL_MMXEXT(cpu_flags)) {
902 c->hadamard8_diff[0] = ff_hadamard8_diff16_mmxext;
903 c->hadamard8_diff[1] = ff_hadamard8_diff_mmxext;
904 c->sum_abs_dctelem = ff_sum_abs_dctelem_mmxext;
907 if (EXTERNAL_SSE2(cpu_flags)) {
908 c->sse[0] = ff_sse16_sse2;
909 c->sum_abs_dctelem = ff_sum_abs_dctelem_sse2;
910 c->diff_pixels = ff_diff_pixels_sse2;
911 c->pix_sum = ff_pix_sum16_sse2;
912 c->pix_norm1 = ff_pix_norm1_sse2;
914 #if HAVE_ALIGNED_STACK
915 c->hadamard8_diff[0] = ff_hadamard8_diff16_sse2;
916 c->hadamard8_diff[1] = ff_hadamard8_diff_sse2;
920 if (EXTERNAL_SSSE3(cpu_flags)) {
921 c->sum_abs_dctelem = ff_sum_abs_dctelem_ssse3;
922 #if HAVE_ALIGNED_STACK
923 c->hadamard8_diff[0] = ff_hadamard8_diff16_ssse3;
924 c->hadamard8_diff[1] = ff_hadamard8_diff_ssse3;
928 ff_dsputil_init_pix_mmx(c, avctx);