2 * Copyright (c) 2000,2001 Fabrice Bellard
3 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
5 * 4MV & hq & B-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
7 * This file is part of Libav.
9 * Libav is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
14 * Libav is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with Libav; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
26 #include "libavutil/internal.h"
29 #include "mpegutils.h"
30 #include "mpegvideo.h"
37 static void gmc1_motion(MpegEncContext *s,
38 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
39 uint8_t **ref_picture)
42 int src_x, src_y, motion_x, motion_y;
43 ptrdiff_t offset, linesize, uvlinesize;
46 motion_x = s->sprite_offset[0][0];
47 motion_y = s->sprite_offset[0][1];
48 src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy + 1));
49 src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy + 1));
50 motion_x <<= (3 - s->sprite_warping_accuracy);
51 motion_y <<= (3 - s->sprite_warping_accuracy);
52 src_x = av_clip(src_x, -16, s->width);
53 if (src_x == s->width)
55 src_y = av_clip(src_y, -16, s->height);
56 if (src_y == s->height)
59 linesize = s->linesize;
60 uvlinesize = s->uvlinesize;
62 ptr = ref_picture[0] + src_y * linesize + src_x;
64 if ((unsigned)src_x >= FFMAX(s->h_edge_pos - 17, 0) ||
65 (unsigned)src_y >= FFMAX(s->v_edge_pos - 17, 0)) {
66 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
70 s->h_edge_pos, s->v_edge_pos);
71 ptr = s->sc.edge_emu_buffer;
74 if ((motion_x | motion_y) & 7) {
75 s->mdsp.gmc1(dest_y, ptr, linesize, 16,
76 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
77 s->mdsp.gmc1(dest_y + 8, ptr + 8, linesize, 16,
78 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
82 dxy = ((motion_x >> 3) & 1) | ((motion_y >> 2) & 2);
84 s->hdsp.put_no_rnd_pixels_tab[0][dxy](dest_y, ptr, linesize, 16);
86 s->hdsp.put_pixels_tab[0][dxy](dest_y, ptr, linesize, 16);
90 if (CONFIG_GRAY && s->avctx->flags & AV_CODEC_FLAG_GRAY)
93 motion_x = s->sprite_offset[1][0];
94 motion_y = s->sprite_offset[1][1];
95 src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy + 1));
96 src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy + 1));
97 motion_x <<= (3 - s->sprite_warping_accuracy);
98 motion_y <<= (3 - s->sprite_warping_accuracy);
99 src_x = av_clip(src_x, -8, s->width >> 1);
100 if (src_x == s->width >> 1)
102 src_y = av_clip(src_y, -8, s->height >> 1);
103 if (src_y == s->height >> 1)
106 offset = (src_y * uvlinesize) + src_x;
107 ptr = ref_picture[1] + offset;
108 if ((unsigned)src_x >= FFMAX((s->h_edge_pos >> 1) - 9, 0) ||
109 (unsigned)src_y >= FFMAX((s->v_edge_pos >> 1) - 9, 0)) {
110 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
111 uvlinesize, uvlinesize,
114 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
115 ptr = s->sc.edge_emu_buffer;
118 s->mdsp.gmc1(dest_cb, ptr, uvlinesize, 8,
119 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
121 ptr = ref_picture[2] + offset;
123 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
124 uvlinesize, uvlinesize,
127 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
128 ptr = s->sc.edge_emu_buffer;
130 s->mdsp.gmc1(dest_cr, ptr, uvlinesize, 8,
131 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
134 static void gmc_motion(MpegEncContext *s,
135 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
136 uint8_t **ref_picture)
139 int linesize, uvlinesize;
140 const int a = s->sprite_warping_accuracy;
143 linesize = s->linesize;
144 uvlinesize = s->uvlinesize;
146 ptr = ref_picture[0];
148 ox = s->sprite_offset[0][0] + s->sprite_delta[0][0] * s->mb_x * 16 +
149 s->sprite_delta[0][1] * s->mb_y * 16;
150 oy = s->sprite_offset[0][1] + s->sprite_delta[1][0] * s->mb_x * 16 +
151 s->sprite_delta[1][1] * s->mb_y * 16;
153 s->mdsp.gmc(dest_y, ptr, linesize, 16,
155 s->sprite_delta[0][0], s->sprite_delta[0][1],
156 s->sprite_delta[1][0], s->sprite_delta[1][1],
157 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
158 s->h_edge_pos, s->v_edge_pos);
159 s->mdsp.gmc(dest_y + 8, ptr, linesize, 16,
160 ox + s->sprite_delta[0][0] * 8,
161 oy + s->sprite_delta[1][0] * 8,
162 s->sprite_delta[0][0], s->sprite_delta[0][1],
163 s->sprite_delta[1][0], s->sprite_delta[1][1],
164 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
165 s->h_edge_pos, s->v_edge_pos);
167 if (CONFIG_GRAY && s->avctx->flags & AV_CODEC_FLAG_GRAY)
170 ox = s->sprite_offset[1][0] + s->sprite_delta[0][0] * s->mb_x * 8 +
171 s->sprite_delta[0][1] * s->mb_y * 8;
172 oy = s->sprite_offset[1][1] + s->sprite_delta[1][0] * s->mb_x * 8 +
173 s->sprite_delta[1][1] * s->mb_y * 8;
175 ptr = ref_picture[1];
176 s->mdsp.gmc(dest_cb, ptr, uvlinesize, 8,
178 s->sprite_delta[0][0], s->sprite_delta[0][1],
179 s->sprite_delta[1][0], s->sprite_delta[1][1],
180 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
181 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
183 ptr = ref_picture[2];
184 s->mdsp.gmc(dest_cr, ptr, uvlinesize, 8,
186 s->sprite_delta[0][0], s->sprite_delta[0][1],
187 s->sprite_delta[1][0], s->sprite_delta[1][1],
188 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
189 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
192 static inline int hpel_motion(MpegEncContext *s,
193 uint8_t *dest, uint8_t *src,
194 int src_x, int src_y,
195 op_pixels_func *pix_op,
196 int motion_x, int motion_y)
201 src_x += motion_x >> 1;
202 src_y += motion_y >> 1;
204 /* WARNING: do no forget half pels */
205 src_x = av_clip(src_x, -16, s->width); // FIXME unneeded for emu?
206 if (src_x != s->width)
208 src_y = av_clip(src_y, -16, s->height);
209 if (src_y != s->height)
210 dxy |= (motion_y & 1) << 1;
211 src += src_y * s->linesize + src_x;
213 if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 1) - 8, 0) ||
214 (unsigned)src_y > FFMAX(s->v_edge_pos - (motion_y & 1) - 8, 0)) {
215 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, src,
216 s->linesize, s->linesize,
218 s->h_edge_pos, s->v_edge_pos);
219 src = s->sc.edge_emu_buffer;
222 pix_op[dxy](dest, src, s->linesize, 8);
226 static av_always_inline
227 void emulated_edge_mc(MpegEncContext *s,
228 int src_x, int src_y,
229 int uvsrc_x, int uvsrc_y,
235 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, *ptr_y,
236 s->linesize, s->linesize,
237 17, 17 + field_based,
238 src_x, src_y * (1 << field_based),
239 s->h_edge_pos, s->v_edge_pos);
240 *ptr_y = s->sc.edge_emu_buffer;
241 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
242 uint8_t *uvbuf = s->sc.edge_emu_buffer + 18 * s->linesize;
243 s->vdsp.emulated_edge_mc(uvbuf, *ptr_cb,
244 s->uvlinesize, s->uvlinesize,
246 uvsrc_x, uvsrc_y * (1 << field_based),
247 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
248 s->vdsp.emulated_edge_mc(uvbuf + 16, *ptr_cr,
249 s->uvlinesize, s->uvlinesize,
251 uvsrc_x, uvsrc_y * (1 << field_based),
252 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
254 *ptr_cr = uvbuf + 16;
258 static av_always_inline
259 void mpeg_motion_internal(MpegEncContext *s,
266 uint8_t **ref_picture,
267 op_pixels_func (*pix_op)[4],
274 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
275 int dxy, uvdxy, mx, my, src_x, src_y,
276 uvsrc_x, uvsrc_y, v_edge_pos;
277 ptrdiff_t uvlinesize, linesize;
279 v_edge_pos = s->v_edge_pos >> field_based;
280 linesize = s->current_picture.f->linesize[0] << field_based;
281 uvlinesize = s->current_picture.f->linesize[1] << field_based;
283 dxy = ((motion_y & 1) << 1) | (motion_x & 1);
284 src_x = s->mb_x * 16 + (motion_x >> 1);
285 src_y = (mb_y << (4 - field_based)) + (motion_y >> 1);
287 if (!is_mpeg12 && s->out_format == FMT_H263) {
288 if ((s->workaround_bugs & FF_BUG_HPEL_CHROMA) && field_based) {
289 mx = (motion_x >> 1) | (motion_x & 1);
291 uvdxy = ((my & 1) << 1) | (mx & 1);
292 uvsrc_x = s->mb_x * 8 + (mx >> 1);
293 uvsrc_y = (mb_y << (3 - field_based)) + (my >> 1);
295 uvdxy = dxy | (motion_y & 2) | ((motion_x & 2) >> 1);
296 uvsrc_x = src_x >> 1;
297 uvsrc_y = src_y >> 1;
299 // Even chroma mv's are full pel in H261
300 } else if (!is_mpeg12 && s->out_format == FMT_H261) {
304 uvsrc_x = s->mb_x * 8 + mx;
305 uvsrc_y = mb_y * 8 + my;
307 if (s->chroma_y_shift) {
310 uvdxy = ((my & 1) << 1) | (mx & 1);
311 uvsrc_x = s->mb_x * 8 + (mx >> 1);
312 uvsrc_y = (mb_y << (3 - field_based)) + (my >> 1);
314 if (s->chroma_x_shift) {
317 uvdxy = ((motion_y & 1) << 1) | (mx & 1);
318 uvsrc_x = s->mb_x * 8 + (mx >> 1);
329 ptr_y = ref_picture[0] + src_y * linesize + src_x;
330 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
331 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
333 if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 1) - 16, 0) ||
334 (unsigned)src_y > FFMAX(v_edge_pos - (motion_y & 1) - h, 0)) {
336 s->codec_id == AV_CODEC_ID_MPEG2VIDEO ||
337 s->codec_id == AV_CODEC_ID_MPEG1VIDEO) {
338 av_log(s->avctx, AV_LOG_DEBUG,
339 "MPEG motion vector out of boundary (%d %d)\n", src_x,
343 emulated_edge_mc(s, src_x, src_y, uvsrc_x, uvsrc_y, field_based,
344 &ptr_y, &ptr_cb, &ptr_cr);
347 /* FIXME use this for field pix too instead of the obnoxious hack which
348 * changes picture.data */
350 dest_y += s->linesize;
351 dest_cb += s->uvlinesize;
352 dest_cr += s->uvlinesize;
356 ptr_y += s->linesize;
357 ptr_cb += s->uvlinesize;
358 ptr_cr += s->uvlinesize;
361 pix_op[0][dxy](dest_y, ptr_y, linesize, h);
363 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
364 pix_op[s->chroma_x_shift][uvdxy]
365 (dest_cb, ptr_cb, uvlinesize, h >> s->chroma_y_shift);
366 pix_op[s->chroma_x_shift][uvdxy]
367 (dest_cr, ptr_cr, uvlinesize, h >> s->chroma_y_shift);
369 if (!is_mpeg12 && (CONFIG_H261_ENCODER || CONFIG_H261_DECODER) &&
370 s->out_format == FMT_H261) {
371 ff_h261_loop_filter(s);
374 /* apply one mpeg motion vector to the three components */
375 static void mpeg_motion(MpegEncContext *s,
376 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
377 int field_select, uint8_t **ref_picture,
378 op_pixels_func (*pix_op)[4],
379 int motion_x, int motion_y, int h, int mb_y)
382 if (s->out_format == FMT_MPEG1)
383 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 0, 0,
384 field_select, ref_picture, pix_op,
385 motion_x, motion_y, h, 1, mb_y);
388 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 0, 0,
389 field_select, ref_picture, pix_op,
390 motion_x, motion_y, h, 0, mb_y);
393 static void mpeg_motion_field(MpegEncContext *s, uint8_t *dest_y,
394 uint8_t *dest_cb, uint8_t *dest_cr,
395 int bottom_field, int field_select,
396 uint8_t **ref_picture,
397 op_pixels_func (*pix_op)[4],
398 int motion_x, int motion_y, int h, int mb_y)
401 if(s->out_format == FMT_MPEG1)
402 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 1,
403 bottom_field, field_select, ref_picture, pix_op,
404 motion_x, motion_y, h, 1, mb_y);
407 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 1,
408 bottom_field, field_select, ref_picture, pix_op,
409 motion_x, motion_y, h, 0, mb_y);
412 // FIXME: SIMDify, avg variant, 16x16 version
413 static inline void put_obmc(uint8_t *dst, uint8_t *src[5], int stride)
416 uint8_t *const top = src[1];
417 uint8_t *const left = src[2];
418 uint8_t *const mid = src[0];
419 uint8_t *const right = src[3];
420 uint8_t *const bottom = src[4];
421 #define OBMC_FILTER(x, t, l, m, r, b)\
422 dst[x]= (t*top[x] + l*left[x] + m*mid[x] + r*right[x] + b*bottom[x] + 4)>>3
423 #define OBMC_FILTER4(x, t, l, m, r, b)\
424 OBMC_FILTER(x , t, l, m, r, b);\
425 OBMC_FILTER(x+1 , t, l, m, r, b);\
426 OBMC_FILTER(x +stride, t, l, m, r, b);\
427 OBMC_FILTER(x+1+stride, t, l, m, r, b);
430 OBMC_FILTER (x , 2, 2, 4, 0, 0);
431 OBMC_FILTER (x + 1, 2, 1, 5, 0, 0);
432 OBMC_FILTER4(x + 2, 2, 1, 5, 0, 0);
433 OBMC_FILTER4(x + 4, 2, 0, 5, 1, 0);
434 OBMC_FILTER (x + 6, 2, 0, 5, 1, 0);
435 OBMC_FILTER (x + 7, 2, 0, 4, 2, 0);
437 OBMC_FILTER (x , 1, 2, 5, 0, 0);
438 OBMC_FILTER (x + 1, 1, 2, 5, 0, 0);
439 OBMC_FILTER (x + 6, 1, 0, 5, 2, 0);
440 OBMC_FILTER (x + 7, 1, 0, 5, 2, 0);
442 OBMC_FILTER4(x , 1, 2, 5, 0, 0);
443 OBMC_FILTER4(x + 2, 1, 1, 6, 0, 0);
444 OBMC_FILTER4(x + 4, 1, 0, 6, 1, 0);
445 OBMC_FILTER4(x + 6, 1, 0, 5, 2, 0);
447 OBMC_FILTER4(x , 0, 2, 5, 0, 1);
448 OBMC_FILTER4(x + 2, 0, 1, 6, 0, 1);
449 OBMC_FILTER4(x + 4, 0, 0, 6, 1, 1);
450 OBMC_FILTER4(x + 6, 0, 0, 5, 2, 1);
452 OBMC_FILTER (x , 0, 2, 5, 0, 1);
453 OBMC_FILTER (x + 1, 0, 2, 5, 0, 1);
454 OBMC_FILTER4(x + 2, 0, 1, 5, 0, 2);
455 OBMC_FILTER4(x + 4, 0, 0, 5, 1, 2);
456 OBMC_FILTER (x + 6, 0, 0, 5, 2, 1);
457 OBMC_FILTER (x + 7, 0, 0, 5, 2, 1);
459 OBMC_FILTER (x , 0, 2, 4, 0, 2);
460 OBMC_FILTER (x + 1, 0, 1, 5, 0, 2);
461 OBMC_FILTER (x + 6, 0, 0, 5, 1, 2);
462 OBMC_FILTER (x + 7, 0, 0, 4, 2, 2);
465 /* obmc for 1 8x8 luma block */
466 static inline void obmc_motion(MpegEncContext *s,
467 uint8_t *dest, uint8_t *src,
468 int src_x, int src_y,
469 op_pixels_func *pix_op,
470 int16_t mv[5][2] /* mid top left right bottom */)
476 assert(s->quarter_sample == 0);
478 for (i = 0; i < 5; i++) {
479 if (i && mv[i][0] == mv[MID][0] && mv[i][1] == mv[MID][1]) {
482 ptr[i] = s->sc.obmc_scratchpad + 8 * (i & 1) +
483 s->linesize * 8 * (i >> 1);
484 hpel_motion(s, ptr[i], src, src_x, src_y, pix_op,
489 put_obmc(dest, ptr, s->linesize);
492 static inline void qpel_motion(MpegEncContext *s,
496 int field_based, int bottom_field,
497 int field_select, uint8_t **ref_picture,
498 op_pixels_func (*pix_op)[4],
499 qpel_mc_func (*qpix_op)[16],
500 int motion_x, int motion_y, int h)
502 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
503 int dxy, uvdxy, mx, my, src_x, src_y, uvsrc_x, uvsrc_y, v_edge_pos;
504 ptrdiff_t linesize, uvlinesize;
506 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
508 src_x = s->mb_x * 16 + (motion_x >> 2);
509 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2);
511 v_edge_pos = s->v_edge_pos >> field_based;
512 linesize = s->linesize << field_based;
513 uvlinesize = s->uvlinesize << field_based;
518 } else if (s->workaround_bugs & FF_BUG_QPEL_CHROMA2) {
519 static const int rtab[8] = { 0, 0, 1, 1, 0, 0, 0, 1 };
520 mx = (motion_x >> 1) + rtab[motion_x & 7];
521 my = (motion_y >> 1) + rtab[motion_y & 7];
522 } else if (s->workaround_bugs & FF_BUG_QPEL_CHROMA) {
523 mx = (motion_x >> 1) | (motion_x & 1);
524 my = (motion_y >> 1) | (motion_y & 1);
529 mx = (mx >> 1) | (mx & 1);
530 my = (my >> 1) | (my & 1);
532 uvdxy = (mx & 1) | ((my & 1) << 1);
536 uvsrc_x = s->mb_x * 8 + mx;
537 uvsrc_y = s->mb_y * (8 >> field_based) + my;
539 ptr_y = ref_picture[0] + src_y * linesize + src_x;
540 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
541 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
543 if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 3) - 16, 0) ||
544 (unsigned)src_y > FFMAX(v_edge_pos - (motion_y & 3) - h, 0)) {
545 emulated_edge_mc(s, src_x, src_y, uvsrc_x, uvsrc_y, field_based,
546 &ptr_y, &ptr_cb, &ptr_cr);
550 qpix_op[0][dxy](dest_y, ptr_y, linesize);
553 dest_y += s->linesize;
554 dest_cb += s->uvlinesize;
555 dest_cr += s->uvlinesize;
559 ptr_y += s->linesize;
560 ptr_cb += s->uvlinesize;
561 ptr_cr += s->uvlinesize;
563 // damn interlaced mode
564 // FIXME boundary mirroring is not exactly correct here
565 qpix_op[1][dxy](dest_y, ptr_y, linesize);
566 qpix_op[1][dxy](dest_y + 8, ptr_y + 8, linesize);
568 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
569 pix_op[1][uvdxy](dest_cr, ptr_cr, uvlinesize, h >> 1);
570 pix_op[1][uvdxy](dest_cb, ptr_cb, uvlinesize, h >> 1);
575 * H.263 chroma 4mv motion compensation.
577 static void chroma_4mv_motion(MpegEncContext *s,
578 uint8_t *dest_cb, uint8_t *dest_cr,
579 uint8_t **ref_picture,
580 op_pixels_func *pix_op,
584 int src_x, src_y, dxy, emu = 0;
587 /* In case of 8X8, we construct a single chroma motion vector
588 * with a special rounding */
589 mx = ff_h263_round_chroma(mx);
590 my = ff_h263_round_chroma(my);
592 dxy = ((my & 1) << 1) | (mx & 1);
596 src_x = s->mb_x * 8 + mx;
597 src_y = s->mb_y * 8 + my;
598 src_x = av_clip(src_x, -8, (s->width >> 1));
599 if (src_x == (s->width >> 1))
601 src_y = av_clip(src_y, -8, (s->height >> 1));
602 if (src_y == (s->height >> 1))
605 offset = src_y * s->uvlinesize + src_x;
606 ptr = ref_picture[1] + offset;
607 if ((unsigned)src_x > FFMAX((s->h_edge_pos >> 1) - (dxy & 1) - 8, 0) ||
608 (unsigned)src_y > FFMAX((s->v_edge_pos >> 1) - (dxy >> 1) - 8, 0)) {
609 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
610 s->uvlinesize, s->uvlinesize,
612 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
613 ptr = s->sc.edge_emu_buffer;
616 pix_op[dxy](dest_cb, ptr, s->uvlinesize, 8);
618 ptr = ref_picture[2] + offset;
620 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
621 s->uvlinesize, s->uvlinesize,
623 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
624 ptr = s->sc.edge_emu_buffer;
626 pix_op[dxy](dest_cr, ptr, s->uvlinesize, 8);
629 static inline void prefetch_motion(MpegEncContext *s, uint8_t **pix, int dir)
631 /* fetch pixels for estimated mv 4 macroblocks ahead
632 * optimized for 64byte cache lines */
633 const int shift = s->quarter_sample ? 2 : 1;
634 const int mx = (s->mv[dir][0][0] >> shift) + 16 * s->mb_x + 8;
635 const int my = (s->mv[dir][0][1] >> shift) + 16 * s->mb_y;
636 int off = mx + (my + (s->mb_x & 3) * 4) * s->linesize + 64;
638 s->vdsp.prefetch(pix[0] + off, s->linesize, 4);
639 off = (mx >> 1) + ((my >> 1) + (s->mb_x & 7)) * s->uvlinesize + 64;
640 s->vdsp.prefetch(pix[1] + off, pix[2] - pix[1], 2);
643 static inline void apply_obmc(MpegEncContext *s,
647 uint8_t **ref_picture,
648 op_pixels_func (*pix_op)[4])
650 LOCAL_ALIGNED_8(int16_t, mv_cache, [4], [4][2]);
651 Picture *cur_frame = &s->current_picture;
654 const int xy = mb_x + mb_y * s->mb_stride;
655 const int mot_stride = s->b8_stride;
656 const int mot_xy = mb_x * 2 + mb_y * 2 * mot_stride;
659 assert(!s->mb_skipped);
661 AV_COPY32(mv_cache[1][1], cur_frame->motion_val[0][mot_xy]);
662 AV_COPY32(mv_cache[1][2], cur_frame->motion_val[0][mot_xy + 1]);
664 AV_COPY32(mv_cache[2][1],
665 cur_frame->motion_val[0][mot_xy + mot_stride]);
666 AV_COPY32(mv_cache[2][2],
667 cur_frame->motion_val[0][mot_xy + mot_stride + 1]);
669 AV_COPY32(mv_cache[3][1],
670 cur_frame->motion_val[0][mot_xy + mot_stride]);
671 AV_COPY32(mv_cache[3][2],
672 cur_frame->motion_val[0][mot_xy + mot_stride + 1]);
674 if (mb_y == 0 || IS_INTRA(cur_frame->mb_type[xy - s->mb_stride])) {
675 AV_COPY32(mv_cache[0][1], mv_cache[1][1]);
676 AV_COPY32(mv_cache[0][2], mv_cache[1][2]);
678 AV_COPY32(mv_cache[0][1],
679 cur_frame->motion_val[0][mot_xy - mot_stride]);
680 AV_COPY32(mv_cache[0][2],
681 cur_frame->motion_val[0][mot_xy - mot_stride + 1]);
684 if (mb_x == 0 || IS_INTRA(cur_frame->mb_type[xy - 1])) {
685 AV_COPY32(mv_cache[1][0], mv_cache[1][1]);
686 AV_COPY32(mv_cache[2][0], mv_cache[2][1]);
688 AV_COPY32(mv_cache[1][0], cur_frame->motion_val[0][mot_xy - 1]);
689 AV_COPY32(mv_cache[2][0],
690 cur_frame->motion_val[0][mot_xy - 1 + mot_stride]);
693 if (mb_x + 1 >= s->mb_width || IS_INTRA(cur_frame->mb_type[xy + 1])) {
694 AV_COPY32(mv_cache[1][3], mv_cache[1][2]);
695 AV_COPY32(mv_cache[2][3], mv_cache[2][2]);
697 AV_COPY32(mv_cache[1][3], cur_frame->motion_val[0][mot_xy + 2]);
698 AV_COPY32(mv_cache[2][3],
699 cur_frame->motion_val[0][mot_xy + 2 + mot_stride]);
704 for (i = 0; i < 4; i++) {
705 const int x = (i & 1) + 1;
706 const int y = (i >> 1) + 1;
708 { mv_cache[y][x][0], mv_cache[y][x][1] },
709 { mv_cache[y - 1][x][0], mv_cache[y - 1][x][1] },
710 { mv_cache[y][x - 1][0], mv_cache[y][x - 1][1] },
711 { mv_cache[y][x + 1][0], mv_cache[y][x + 1][1] },
712 { mv_cache[y + 1][x][0], mv_cache[y + 1][x][1] }
715 obmc_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
717 mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >> 1) * 8,
724 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY))
725 chroma_4mv_motion(s, dest_cb, dest_cr,
726 ref_picture, pix_op[1],
730 static inline void apply_8x8(MpegEncContext *s,
735 uint8_t **ref_picture,
736 qpel_mc_func (*qpix_op)[16],
737 op_pixels_func (*pix_op)[4])
739 int dxy, mx, my, src_x, src_y;
747 if (s->quarter_sample) {
748 for (i = 0; i < 4; i++) {
749 int motion_x = s->mv[dir][i][0];
750 int motion_y = s->mv[dir][i][1];
752 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
753 src_x = mb_x * 16 + (motion_x >> 2) + (i & 1) * 8;
754 src_y = mb_y * 16 + (motion_y >> 2) + (i >> 1) * 8;
756 /* WARNING: do no forget half pels */
757 src_x = av_clip(src_x, -16, s->width);
758 if (src_x == s->width)
760 src_y = av_clip(src_y, -16, s->height);
761 if (src_y == s->height)
764 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x);
765 if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 3) - 8, 0) ||
766 (unsigned)src_y > FFMAX(s->v_edge_pos - (motion_y & 3) - 8, 0)) {
767 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
768 s->linesize, s->linesize,
773 ptr = s->sc.edge_emu_buffer;
775 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize;
776 qpix_op[1][dxy](dest, ptr, s->linesize);
778 mx += s->mv[dir][i][0] / 2;
779 my += s->mv[dir][i][1] / 2;
782 for (i = 0; i < 4; i++) {
784 dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
786 mb_x * 16 + (i & 1) * 8,
787 mb_y * 16 + (i >> 1) * 8,
792 mx += s->mv[dir][i][0];
793 my += s->mv[dir][i][1];
797 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY))
798 chroma_4mv_motion(s, dest_cb, dest_cr,
799 ref_picture, pix_op[1], mx, my);
803 * motion compensation of a single macroblock
805 * @param dest_y luma destination pointer
806 * @param dest_cb chroma cb/u destination pointer
807 * @param dest_cr chroma cr/v destination pointer
808 * @param dir direction (0->forward, 1->backward)
809 * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
810 * @param pix_op halfpel motion compensation function (average or put normally)
811 * @param qpix_op qpel motion compensation function (average or put normally)
812 * the motion vectors are taken from s->mv and the MV type from s->mv_type
814 static av_always_inline void mpv_motion_internal(MpegEncContext *s,
819 uint8_t **ref_picture,
820 op_pixels_func (*pix_op)[4],
821 qpel_mc_func (*qpix_op)[16],
827 prefetch_motion(s, ref_picture, dir);
829 if (!is_mpeg12 && s->obmc && s->pict_type != AV_PICTURE_TYPE_B) {
830 apply_obmc(s, dest_y, dest_cb, dest_cr, ref_picture, pix_op);
834 switch (s->mv_type) {
837 if (s->real_sprite_warping_points == 1) {
838 gmc1_motion(s, dest_y, dest_cb, dest_cr,
841 gmc_motion(s, dest_y, dest_cb, dest_cr,
844 } else if (!is_mpeg12 && s->quarter_sample) {
845 qpel_motion(s, dest_y, dest_cb, dest_cr,
847 ref_picture, pix_op, qpix_op,
848 s->mv[dir][0][0], s->mv[dir][0][1], 16);
849 } else if (!is_mpeg12 && (CONFIG_WMV2_DECODER || CONFIG_WMV2_ENCODER) &&
850 s->mspel && s->codec_id == AV_CODEC_ID_WMV2) {
851 ff_mspel_motion(s, dest_y, dest_cb, dest_cr,
853 s->mv[dir][0][0], s->mv[dir][0][1], 16);
855 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
857 s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y);
862 apply_8x8(s, dest_y, dest_cb, dest_cr,
863 dir, ref_picture, qpix_op, pix_op);
866 if (s->picture_structure == PICT_FRAME) {
867 if (!is_mpeg12 && s->quarter_sample) {
868 for (i = 0; i < 2; i++)
869 qpel_motion(s, dest_y, dest_cb, dest_cr,
870 1, i, s->field_select[dir][i],
871 ref_picture, pix_op, qpix_op,
872 s->mv[dir][i][0], s->mv[dir][i][1], 8);
875 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
876 0, s->field_select[dir][0],
878 s->mv[dir][0][0], s->mv[dir][0][1], 8, mb_y);
880 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
881 1, s->field_select[dir][1],
883 s->mv[dir][1][0], s->mv[dir][1][1], 8, mb_y);
886 if (s->picture_structure != s->field_select[dir][0] + 1 &&
887 s->pict_type != AV_PICTURE_TYPE_B && !s->first_field) {
888 ref_picture = s->current_picture_ptr->f->data;
891 mpeg_motion(s, dest_y, dest_cb, dest_cr,
892 s->field_select[dir][0],
894 s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y >> 1);
898 for (i = 0; i < 2; i++) {
899 uint8_t **ref2picture;
901 if (s->picture_structure == s->field_select[dir][i] + 1
902 || s->pict_type == AV_PICTURE_TYPE_B || s->first_field) {
903 ref2picture = ref_picture;
905 ref2picture = s->current_picture_ptr->f->data;
908 mpeg_motion(s, dest_y, dest_cb, dest_cr,
909 s->field_select[dir][i],
911 s->mv[dir][i][0], s->mv[dir][i][1] + 16 * i,
914 dest_y += 16 * s->linesize;
915 dest_cb += (16 >> s->chroma_y_shift) * s->uvlinesize;
916 dest_cr += (16 >> s->chroma_y_shift) * s->uvlinesize;
920 if (s->picture_structure == PICT_FRAME) {
921 for (i = 0; i < 2; i++) {
923 for (j = 0; j < 2; j++)
924 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
925 j, j ^ i, ref_picture, pix_op,
926 s->mv[dir][2 * i + j][0],
927 s->mv[dir][2 * i + j][1], 8, mb_y);
928 pix_op = s->hdsp.avg_pixels_tab;
931 for (i = 0; i < 2; i++) {
932 mpeg_motion(s, dest_y, dest_cb, dest_cr,
933 s->picture_structure != i + 1,
935 s->mv[dir][2 * i][0], s->mv[dir][2 * i][1],
938 // after put we make avg of the same block
939 pix_op = s->hdsp.avg_pixels_tab;
941 /* opposite parity is always in the same frame if this is
943 if (!s->first_field) {
944 ref_picture = s->current_picture_ptr->f->data;
953 void ff_mpv_motion(MpegEncContext *s,
954 uint8_t *dest_y, uint8_t *dest_cb,
955 uint8_t *dest_cr, int dir,
956 uint8_t **ref_picture,
957 op_pixels_func (*pix_op)[4],
958 qpel_mc_func (*qpix_op)[16])
961 if (s->out_format == FMT_MPEG1)
962 mpv_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
963 ref_picture, pix_op, qpix_op, 1);
966 mpv_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
967 ref_picture, pix_op, qpix_op, 0);