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 FFmpeg.
9 * FFmpeg 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 * FFmpeg 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 FFmpeg; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 #include "libavutil/avassert.h"
25 #include "libavutil/internal.h"
26 #include "libavutil/mem_internal.h"
30 #include "mpegutils.h"
31 #include "mpegvideo.h"
35 static void gmc1_motion(MpegEncContext *s,
36 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
37 uint8_t **ref_picture)
40 int src_x, src_y, motion_x, motion_y;
41 ptrdiff_t offset, linesize, uvlinesize;
44 motion_x = s->sprite_offset[0][0];
45 motion_y = s->sprite_offset[0][1];
46 src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy + 1));
47 src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy + 1));
48 motion_x *= 1 << (3 - s->sprite_warping_accuracy);
49 motion_y *= 1 << (3 - s->sprite_warping_accuracy);
50 src_x = av_clip(src_x, -16, s->width);
51 if (src_x == s->width)
53 src_y = av_clip(src_y, -16, s->height);
54 if (src_y == s->height)
57 linesize = s->linesize;
58 uvlinesize = s->uvlinesize;
60 ptr = ref_picture[0] + src_y * linesize + src_x;
62 if ((unsigned)src_x >= FFMAX(s->h_edge_pos - 17, 0) ||
63 (unsigned)src_y >= FFMAX(s->v_edge_pos - 17, 0)) {
64 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
68 s->h_edge_pos, s->v_edge_pos);
69 ptr = s->sc.edge_emu_buffer;
72 if ((motion_x | motion_y) & 7) {
73 s->mdsp.gmc1(dest_y, ptr, linesize, 16,
74 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
75 s->mdsp.gmc1(dest_y + 8, ptr + 8, linesize, 16,
76 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
80 dxy = ((motion_x >> 3) & 1) | ((motion_y >> 2) & 2);
82 s->hdsp.put_no_rnd_pixels_tab[0][dxy](dest_y, ptr, linesize, 16);
84 s->hdsp.put_pixels_tab[0][dxy](dest_y, ptr, linesize, 16);
88 if (CONFIG_GRAY && s->avctx->flags & AV_CODEC_FLAG_GRAY)
91 motion_x = s->sprite_offset[1][0];
92 motion_y = s->sprite_offset[1][1];
93 src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy + 1));
94 src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy + 1));
95 motion_x *= 1 << (3 - s->sprite_warping_accuracy);
96 motion_y *= 1 << (3 - s->sprite_warping_accuracy);
97 src_x = av_clip(src_x, -8, s->width >> 1);
98 if (src_x == s->width >> 1)
100 src_y = av_clip(src_y, -8, s->height >> 1);
101 if (src_y == s->height >> 1)
104 offset = (src_y * uvlinesize) + src_x;
105 ptr = ref_picture[1] + offset;
106 if ((unsigned)src_x >= FFMAX((s->h_edge_pos >> 1) - 9, 0) ||
107 (unsigned)src_y >= FFMAX((s->v_edge_pos >> 1) - 9, 0)) {
108 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
109 uvlinesize, uvlinesize,
112 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
113 ptr = s->sc.edge_emu_buffer;
116 s->mdsp.gmc1(dest_cb, ptr, uvlinesize, 8,
117 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
119 ptr = ref_picture[2] + offset;
121 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
122 uvlinesize, uvlinesize,
125 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
126 ptr = s->sc.edge_emu_buffer;
128 s->mdsp.gmc1(dest_cr, ptr, uvlinesize, 8,
129 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
132 static void gmc_motion(MpegEncContext *s,
133 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
134 uint8_t **ref_picture)
137 int linesize, uvlinesize;
138 const int a = s->sprite_warping_accuracy;
141 linesize = s->linesize;
142 uvlinesize = s->uvlinesize;
144 ptr = ref_picture[0];
146 ox = s->sprite_offset[0][0] + s->sprite_delta[0][0] * s->mb_x * 16 +
147 s->sprite_delta[0][1] * s->mb_y * 16;
148 oy = s->sprite_offset[0][1] + s->sprite_delta[1][0] * s->mb_x * 16 +
149 s->sprite_delta[1][1] * s->mb_y * 16;
151 s->mdsp.gmc(dest_y, ptr, linesize, 16,
153 s->sprite_delta[0][0], s->sprite_delta[0][1],
154 s->sprite_delta[1][0], s->sprite_delta[1][1],
155 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
156 s->h_edge_pos, s->v_edge_pos);
157 s->mdsp.gmc(dest_y + 8, ptr, linesize, 16,
158 ox + s->sprite_delta[0][0] * 8,
159 oy + s->sprite_delta[1][0] * 8,
160 s->sprite_delta[0][0], s->sprite_delta[0][1],
161 s->sprite_delta[1][0], s->sprite_delta[1][1],
162 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
163 s->h_edge_pos, s->v_edge_pos);
165 if (CONFIG_GRAY && s->avctx->flags & AV_CODEC_FLAG_GRAY)
168 ox = s->sprite_offset[1][0] + s->sprite_delta[0][0] * s->mb_x * 8 +
169 s->sprite_delta[0][1] * s->mb_y * 8;
170 oy = s->sprite_offset[1][1] + s->sprite_delta[1][0] * s->mb_x * 8 +
171 s->sprite_delta[1][1] * s->mb_y * 8;
173 ptr = ref_picture[1];
174 s->mdsp.gmc(dest_cb, ptr, uvlinesize, 8,
176 s->sprite_delta[0][0], s->sprite_delta[0][1],
177 s->sprite_delta[1][0], s->sprite_delta[1][1],
178 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
179 (s->h_edge_pos + 1) >> 1, (s->v_edge_pos + 1) >> 1);
181 ptr = ref_picture[2];
182 s->mdsp.gmc(dest_cr, ptr, uvlinesize, 8,
184 s->sprite_delta[0][0], s->sprite_delta[0][1],
185 s->sprite_delta[1][0], s->sprite_delta[1][1],
186 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
187 (s->h_edge_pos + 1) >> 1, (s->v_edge_pos + 1) >> 1);
190 static inline int hpel_motion(MpegEncContext *s,
191 uint8_t *dest, uint8_t *src,
192 int src_x, int src_y,
193 op_pixels_func *pix_op,
194 int motion_x, int motion_y)
199 src_x += motion_x >> 1;
200 src_y += motion_y >> 1;
202 /* WARNING: do no forget half pels */
203 src_x = av_clip(src_x, -16, s->width); // FIXME unneeded for emu?
204 if (src_x != s->width)
206 src_y = av_clip(src_y, -16, s->height);
207 if (src_y != s->height)
208 dxy |= (motion_y & 1) << 1;
209 src += src_y * s->linesize + src_x;
211 if ((unsigned)src_x >= FFMAX(s->h_edge_pos - (motion_x & 1) - 7, 0) ||
212 (unsigned)src_y >= FFMAX(s->v_edge_pos - (motion_y & 1) - 7, 0)) {
213 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, src,
214 s->linesize, s->linesize,
217 s->h_edge_pos, s->v_edge_pos);
218 src = s->sc.edge_emu_buffer;
221 pix_op[dxy](dest, src, s->linesize, 8);
225 static av_always_inline
226 void mpeg_motion_internal(MpegEncContext *s,
233 uint8_t **ref_picture,
234 op_pixels_func (*pix_op)[4],
242 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
243 int dxy, uvdxy, mx, my, src_x, src_y,
244 uvsrc_x, uvsrc_y, v_edge_pos, block_y_half;
245 ptrdiff_t uvlinesize, linesize;
247 v_edge_pos = s->v_edge_pos >> field_based;
248 linesize = s->current_picture.f->linesize[0] << field_based;
249 uvlinesize = s->current_picture.f->linesize[1] << field_based;
250 block_y_half = (field_based | is_16x8);
252 dxy = ((motion_y & 1) << 1) | (motion_x & 1);
253 src_x = s->mb_x * 16 + (motion_x >> 1);
254 src_y = (mb_y << (4 - block_y_half)) + (motion_y >> 1);
256 if (!is_mpeg12 && s->out_format == FMT_H263) {
257 if ((s->workaround_bugs & FF_BUG_HPEL_CHROMA) && field_based) {
258 mx = (motion_x >> 1) | (motion_x & 1);
260 uvdxy = ((my & 1) << 1) | (mx & 1);
261 uvsrc_x = s->mb_x * 8 + (mx >> 1);
262 uvsrc_y = (mb_y << (3 - block_y_half)) + (my >> 1);
264 uvdxy = dxy | (motion_y & 2) | ((motion_x & 2) >> 1);
265 uvsrc_x = src_x >> 1;
266 uvsrc_y = src_y >> 1;
268 // Even chroma mv's are full pel in H261
269 } else if (!is_mpeg12 && s->out_format == FMT_H261) {
273 uvsrc_x = s->mb_x * 8 + mx;
274 uvsrc_y = mb_y * 8 + my;
276 if (s->chroma_y_shift) {
279 uvdxy = ((my & 1) << 1) | (mx & 1);
280 uvsrc_x = s->mb_x * 8 + (mx >> 1);
281 uvsrc_y = (mb_y << (3 - block_y_half)) + (my >> 1);
283 if (s->chroma_x_shift) {
286 uvdxy = ((motion_y & 1) << 1) | (mx & 1);
287 uvsrc_x = s->mb_x * 8 + (mx >> 1);
298 ptr_y = ref_picture[0] + src_y * linesize + src_x;
299 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
300 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
302 if ((unsigned)src_x >= FFMAX(s->h_edge_pos - (motion_x & 1) - 15 , 0) ||
303 (unsigned)src_y >= FFMAX( v_edge_pos - (motion_y & 1) - h + 1, 0)) {
304 if (is_mpeg12 || (CONFIG_SMALL &&
305 (s->codec_id == AV_CODEC_ID_MPEG2VIDEO ||
306 s->codec_id == AV_CODEC_ID_MPEG1VIDEO))) {
307 av_log(s->avctx, AV_LOG_DEBUG,
308 "MPEG motion vector out of boundary (%d %d)\n", src_x,
312 src_y = (unsigned)src_y << field_based;
313 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr_y,
314 s->linesize, s->linesize,
315 17, 17 + field_based,
317 s->h_edge_pos, s->v_edge_pos);
318 ptr_y = s->sc.edge_emu_buffer;
319 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
320 uint8_t *ubuf = s->sc.edge_emu_buffer + 18 * s->linesize;
321 uint8_t *vbuf = ubuf + 10 * s->uvlinesize;
322 if (s->workaround_bugs & FF_BUG_IEDGE)
323 vbuf -= s->uvlinesize;
324 uvsrc_y = (unsigned)uvsrc_y << field_based;
325 s->vdsp.emulated_edge_mc(ubuf, ptr_cb,
326 s->uvlinesize, s->uvlinesize,
329 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
330 s->vdsp.emulated_edge_mc(vbuf, ptr_cr,
331 s->uvlinesize, s->uvlinesize,
334 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
340 /* FIXME use this for field pix too instead of the obnoxious hack which
341 * changes picture.data */
343 dest_y += s->linesize;
344 dest_cb += s->uvlinesize;
345 dest_cr += s->uvlinesize;
349 ptr_y += s->linesize;
350 ptr_cb += s->uvlinesize;
351 ptr_cr += s->uvlinesize;
354 pix_op[0][dxy](dest_y, ptr_y, linesize, h);
356 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
357 pix_op[s->chroma_x_shift][uvdxy]
358 (dest_cb, ptr_cb, uvlinesize, h >> s->chroma_y_shift);
359 pix_op[s->chroma_x_shift][uvdxy]
360 (dest_cr, ptr_cr, uvlinesize, h >> s->chroma_y_shift);
362 if (!is_mpeg12 && (CONFIG_H261_ENCODER || CONFIG_H261_DECODER) &&
363 s->out_format == FMT_H261) {
364 ff_h261_loop_filter(s);
367 /* apply one mpeg motion vector to the three components */
368 static void mpeg_motion(MpegEncContext *s,
369 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
370 int field_select, uint8_t **ref_picture,
371 op_pixels_func (*pix_op)[4],
372 int motion_x, int motion_y, int h, int is_16x8, int mb_y)
375 if (s->out_format == FMT_MPEG1)
376 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 0, 0,
377 field_select, ref_picture, pix_op,
378 motion_x, motion_y, h, 1, is_16x8, mb_y);
381 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 0, 0,
382 field_select, ref_picture, pix_op,
383 motion_x, motion_y, h, 0, is_16x8, mb_y);
386 static void mpeg_motion_field(MpegEncContext *s, uint8_t *dest_y,
387 uint8_t *dest_cb, uint8_t *dest_cr,
388 int bottom_field, int field_select,
389 uint8_t **ref_picture,
390 op_pixels_func (*pix_op)[4],
391 int motion_x, int motion_y, int h, int mb_y)
394 if (s->out_format == FMT_MPEG1)
395 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 1,
396 bottom_field, field_select, ref_picture, pix_op,
397 motion_x, motion_y, h, 1, 0, mb_y);
400 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 1,
401 bottom_field, field_select, ref_picture, pix_op,
402 motion_x, motion_y, h, 0, 0, mb_y);
405 // FIXME: SIMDify, avg variant, 16x16 version
406 static inline void put_obmc(uint8_t *dst, uint8_t *src[5], int stride)
409 uint8_t *const top = src[1];
410 uint8_t *const left = src[2];
411 uint8_t *const mid = src[0];
412 uint8_t *const right = src[3];
413 uint8_t *const bottom = src[4];
414 #define OBMC_FILTER(x, t, l, m, r, b)\
415 dst[x]= (t*top[x] + l*left[x] + m*mid[x] + r*right[x] + b*bottom[x] + 4)>>3
416 #define OBMC_FILTER4(x, t, l, m, r, b)\
417 OBMC_FILTER(x , t, l, m, r, b);\
418 OBMC_FILTER(x+1 , t, l, m, r, b);\
419 OBMC_FILTER(x +stride, t, l, m, r, b);\
420 OBMC_FILTER(x+1+stride, t, l, m, r, b);
423 OBMC_FILTER (x , 2, 2, 4, 0, 0);
424 OBMC_FILTER (x + 1, 2, 1, 5, 0, 0);
425 OBMC_FILTER4(x + 2, 2, 1, 5, 0, 0);
426 OBMC_FILTER4(x + 4, 2, 0, 5, 1, 0);
427 OBMC_FILTER (x + 6, 2, 0, 5, 1, 0);
428 OBMC_FILTER (x + 7, 2, 0, 4, 2, 0);
430 OBMC_FILTER (x , 1, 2, 5, 0, 0);
431 OBMC_FILTER (x + 1, 1, 2, 5, 0, 0);
432 OBMC_FILTER (x + 6, 1, 0, 5, 2, 0);
433 OBMC_FILTER (x + 7, 1, 0, 5, 2, 0);
435 OBMC_FILTER4(x , 1, 2, 5, 0, 0);
436 OBMC_FILTER4(x + 2, 1, 1, 6, 0, 0);
437 OBMC_FILTER4(x + 4, 1, 0, 6, 1, 0);
438 OBMC_FILTER4(x + 6, 1, 0, 5, 2, 0);
440 OBMC_FILTER4(x , 0, 2, 5, 0, 1);
441 OBMC_FILTER4(x + 2, 0, 1, 6, 0, 1);
442 OBMC_FILTER4(x + 4, 0, 0, 6, 1, 1);
443 OBMC_FILTER4(x + 6, 0, 0, 5, 2, 1);
445 OBMC_FILTER (x , 0, 2, 5, 0, 1);
446 OBMC_FILTER (x + 1, 0, 2, 5, 0, 1);
447 OBMC_FILTER4(x + 2, 0, 1, 5, 0, 2);
448 OBMC_FILTER4(x + 4, 0, 0, 5, 1, 2);
449 OBMC_FILTER (x + 6, 0, 0, 5, 2, 1);
450 OBMC_FILTER (x + 7, 0, 0, 5, 2, 1);
452 OBMC_FILTER (x , 0, 2, 4, 0, 2);
453 OBMC_FILTER (x + 1, 0, 1, 5, 0, 2);
454 OBMC_FILTER (x + 6, 0, 0, 5, 1, 2);
455 OBMC_FILTER (x + 7, 0, 0, 4, 2, 2);
458 /* obmc for 1 8x8 luma block */
459 static inline void obmc_motion(MpegEncContext *s,
460 uint8_t *dest, uint8_t *src,
461 int src_x, int src_y,
462 op_pixels_func *pix_op,
463 int16_t mv[5][2] /* mid top left right bottom */)
469 av_assert2(s->quarter_sample == 0);
471 for (i = 0; i < 5; i++) {
472 if (i && mv[i][0] == mv[MID][0] && mv[i][1] == mv[MID][1]) {
475 ptr[i] = s->sc.obmc_scratchpad + 8 * (i & 1) +
476 s->linesize * 8 * (i >> 1);
477 hpel_motion(s, ptr[i], src, src_x, src_y, pix_op,
482 put_obmc(dest, ptr, s->linesize);
485 static inline void qpel_motion(MpegEncContext *s,
489 int field_based, int bottom_field,
490 int field_select, uint8_t **ref_picture,
491 op_pixels_func (*pix_op)[4],
492 qpel_mc_func (*qpix_op)[16],
493 int motion_x, int motion_y, int h)
495 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
496 int dxy, uvdxy, mx, my, src_x, src_y, uvsrc_x, uvsrc_y, v_edge_pos;
497 ptrdiff_t linesize, uvlinesize;
499 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
501 src_x = s->mb_x * 16 + (motion_x >> 2);
502 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2);
504 v_edge_pos = s->v_edge_pos >> field_based;
505 linesize = s->linesize << field_based;
506 uvlinesize = s->uvlinesize << field_based;
511 } else if (s->workaround_bugs & FF_BUG_QPEL_CHROMA2) {
512 static const int rtab[8] = { 0, 0, 1, 1, 0, 0, 0, 1 };
513 mx = (motion_x >> 1) + rtab[motion_x & 7];
514 my = (motion_y >> 1) + rtab[motion_y & 7];
515 } else if (s->workaround_bugs & FF_BUG_QPEL_CHROMA) {
516 mx = (motion_x >> 1) | (motion_x & 1);
517 my = (motion_y >> 1) | (motion_y & 1);
522 mx = (mx >> 1) | (mx & 1);
523 my = (my >> 1) | (my & 1);
525 uvdxy = (mx & 1) | ((my & 1) << 1);
529 uvsrc_x = s->mb_x * 8 + mx;
530 uvsrc_y = s->mb_y * (8 >> field_based) + my;
532 ptr_y = ref_picture[0] + src_y * linesize + src_x;
533 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
534 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
536 if ((unsigned)src_x >= FFMAX(s->h_edge_pos - (motion_x & 3) - 15 , 0) ||
537 (unsigned)src_y >= FFMAX( v_edge_pos - (motion_y & 3) - h + 1, 0)) {
538 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr_y,
539 s->linesize, s->linesize,
540 17, 17 + field_based,
541 src_x, src_y * (1 << field_based),
542 s->h_edge_pos, s->v_edge_pos);
543 ptr_y = s->sc.edge_emu_buffer;
544 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
545 uint8_t *ubuf = s->sc.edge_emu_buffer + 18 * s->linesize;
546 uint8_t *vbuf = ubuf + 10 * s->uvlinesize;
547 if (s->workaround_bugs & FF_BUG_IEDGE)
548 vbuf -= s->uvlinesize;
549 s->vdsp.emulated_edge_mc(ubuf, ptr_cb,
550 s->uvlinesize, s->uvlinesize,
552 uvsrc_x, uvsrc_y * (1 << field_based),
553 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
554 s->vdsp.emulated_edge_mc(vbuf, ptr_cr,
555 s->uvlinesize, s->uvlinesize,
557 uvsrc_x, uvsrc_y * (1 << field_based),
558 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
565 qpix_op[0][dxy](dest_y, ptr_y, linesize);
568 dest_y += s->linesize;
569 dest_cb += s->uvlinesize;
570 dest_cr += s->uvlinesize;
574 ptr_y += s->linesize;
575 ptr_cb += s->uvlinesize;
576 ptr_cr += s->uvlinesize;
578 // damn interlaced mode
579 // FIXME boundary mirroring is not exactly correct here
580 qpix_op[1][dxy](dest_y, ptr_y, linesize);
581 qpix_op[1][dxy](dest_y + 8, ptr_y + 8, linesize);
583 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
584 pix_op[1][uvdxy](dest_cr, ptr_cr, uvlinesize, h >> 1);
585 pix_op[1][uvdxy](dest_cb, ptr_cb, uvlinesize, h >> 1);
590 * H.263 chroma 4mv motion compensation.
592 static void chroma_4mv_motion(MpegEncContext *s,
593 uint8_t *dest_cb, uint8_t *dest_cr,
594 uint8_t **ref_picture,
595 op_pixels_func *pix_op,
599 int src_x, src_y, dxy, emu = 0;
602 /* In case of 8X8, we construct a single chroma motion vector
603 * with a special rounding */
604 mx = ff_h263_round_chroma(mx);
605 my = ff_h263_round_chroma(my);
607 dxy = ((my & 1) << 1) | (mx & 1);
611 src_x = s->mb_x * 8 + mx;
612 src_y = s->mb_y * 8 + my;
613 src_x = av_clip(src_x, -8, (s->width >> 1));
614 if (src_x == (s->width >> 1))
616 src_y = av_clip(src_y, -8, (s->height >> 1));
617 if (src_y == (s->height >> 1))
620 offset = src_y * s->uvlinesize + src_x;
621 ptr = ref_picture[1] + offset;
622 if ((unsigned)src_x >= FFMAX((s->h_edge_pos >> 1) - (dxy & 1) - 7, 0) ||
623 (unsigned)src_y >= FFMAX((s->v_edge_pos >> 1) - (dxy >> 1) - 7, 0)) {
624 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
625 s->uvlinesize, s->uvlinesize,
627 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
628 ptr = s->sc.edge_emu_buffer;
631 pix_op[dxy](dest_cb, ptr, s->uvlinesize, 8);
633 ptr = ref_picture[2] + offset;
635 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
636 s->uvlinesize, s->uvlinesize,
638 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
639 ptr = s->sc.edge_emu_buffer;
641 pix_op[dxy](dest_cr, ptr, s->uvlinesize, 8);
644 static inline void prefetch_motion(MpegEncContext *s, uint8_t **pix, int dir)
646 /* fetch pixels for estimated mv 4 macroblocks ahead
647 * optimized for 64byte cache lines */
648 const int shift = s->quarter_sample ? 2 : 1;
649 const int mx = (s->mv[dir][0][0] >> shift) + 16 * s->mb_x + 8;
650 const int my = (s->mv[dir][0][1] >> shift) + 16 * s->mb_y;
651 int off = mx + (my + (s->mb_x & 3) * 4) * s->linesize + 64;
653 s->vdsp.prefetch(pix[0] + off, s->linesize, 4);
654 off = (mx >> 1) + ((my >> 1) + (s->mb_x & 7)) * s->uvlinesize + 64;
655 s->vdsp.prefetch(pix[1] + off, pix[2] - pix[1], 2);
658 static inline void apply_obmc(MpegEncContext *s,
662 uint8_t **ref_picture,
663 op_pixels_func (*pix_op)[4])
665 LOCAL_ALIGNED_8(int16_t, mv_cache, [4], [4][2]);
666 Picture *cur_frame = &s->current_picture;
669 const int xy = mb_x + mb_y * s->mb_stride;
670 const int mot_stride = s->b8_stride;
671 const int mot_xy = mb_x * 2 + mb_y * 2 * mot_stride;
674 av_assert2(!s->mb_skipped);
676 AV_COPY32(mv_cache[1][1], cur_frame->motion_val[0][mot_xy]);
677 AV_COPY32(mv_cache[1][2], cur_frame->motion_val[0][mot_xy + 1]);
679 AV_COPY32(mv_cache[2][1],
680 cur_frame->motion_val[0][mot_xy + mot_stride]);
681 AV_COPY32(mv_cache[2][2],
682 cur_frame->motion_val[0][mot_xy + mot_stride + 1]);
684 AV_COPY32(mv_cache[3][1],
685 cur_frame->motion_val[0][mot_xy + mot_stride]);
686 AV_COPY32(mv_cache[3][2],
687 cur_frame->motion_val[0][mot_xy + mot_stride + 1]);
689 if (mb_y == 0 || IS_INTRA(cur_frame->mb_type[xy - s->mb_stride])) {
690 AV_COPY32(mv_cache[0][1], mv_cache[1][1]);
691 AV_COPY32(mv_cache[0][2], mv_cache[1][2]);
693 AV_COPY32(mv_cache[0][1],
694 cur_frame->motion_val[0][mot_xy - mot_stride]);
695 AV_COPY32(mv_cache[0][2],
696 cur_frame->motion_val[0][mot_xy - mot_stride + 1]);
699 if (mb_x == 0 || IS_INTRA(cur_frame->mb_type[xy - 1])) {
700 AV_COPY32(mv_cache[1][0], mv_cache[1][1]);
701 AV_COPY32(mv_cache[2][0], mv_cache[2][1]);
703 AV_COPY32(mv_cache[1][0], cur_frame->motion_val[0][mot_xy - 1]);
704 AV_COPY32(mv_cache[2][0],
705 cur_frame->motion_val[0][mot_xy - 1 + mot_stride]);
708 if (mb_x + 1 >= s->mb_width || IS_INTRA(cur_frame->mb_type[xy + 1])) {
709 AV_COPY32(mv_cache[1][3], mv_cache[1][2]);
710 AV_COPY32(mv_cache[2][3], mv_cache[2][2]);
712 AV_COPY32(mv_cache[1][3], cur_frame->motion_val[0][mot_xy + 2]);
713 AV_COPY32(mv_cache[2][3],
714 cur_frame->motion_val[0][mot_xy + 2 + mot_stride]);
719 for (i = 0; i < 4; i++) {
720 const int x = (i & 1) + 1;
721 const int y = (i >> 1) + 1;
723 { mv_cache[y][x][0], mv_cache[y][x][1] },
724 { mv_cache[y - 1][x][0], mv_cache[y - 1][x][1] },
725 { mv_cache[y][x - 1][0], mv_cache[y][x - 1][1] },
726 { mv_cache[y][x + 1][0], mv_cache[y][x + 1][1] },
727 { mv_cache[y + 1][x][0], mv_cache[y + 1][x][1] }
730 obmc_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
732 mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >> 1) * 8,
739 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY))
740 chroma_4mv_motion(s, dest_cb, dest_cr,
741 ref_picture, pix_op[1],
745 static inline void apply_8x8(MpegEncContext *s,
750 uint8_t **ref_picture,
751 qpel_mc_func (*qpix_op)[16],
752 op_pixels_func (*pix_op)[4])
754 int dxy, mx, my, src_x, src_y;
762 if (s->quarter_sample) {
763 for (i = 0; i < 4; i++) {
764 int motion_x = s->mv[dir][i][0];
765 int motion_y = s->mv[dir][i][1];
767 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
768 src_x = mb_x * 16 + (motion_x >> 2) + (i & 1) * 8;
769 src_y = mb_y * 16 + (motion_y >> 2) + (i >> 1) * 8;
771 /* WARNING: do no forget half pels */
772 src_x = av_clip(src_x, -16, s->width);
773 if (src_x == s->width)
775 src_y = av_clip(src_y, -16, s->height);
776 if (src_y == s->height)
779 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x);
780 if ((unsigned)src_x >= FFMAX(s->h_edge_pos - (motion_x & 3) - 7, 0) ||
781 (unsigned)src_y >= FFMAX(s->v_edge_pos - (motion_y & 3) - 7, 0)) {
782 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
783 s->linesize, s->linesize,
788 ptr = s->sc.edge_emu_buffer;
790 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize;
791 qpix_op[1][dxy](dest, ptr, s->linesize);
793 mx += s->mv[dir][i][0] / 2;
794 my += s->mv[dir][i][1] / 2;
797 for (i = 0; i < 4; i++) {
799 dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
801 mb_x * 16 + (i & 1) * 8,
802 mb_y * 16 + (i >> 1) * 8,
807 mx += s->mv[dir][i][0];
808 my += s->mv[dir][i][1];
812 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY))
813 chroma_4mv_motion(s, dest_cb, dest_cr,
814 ref_picture, pix_op[1], mx, my);
818 * motion compensation of a single macroblock
820 * @param dest_y luma destination pointer
821 * @param dest_cb chroma cb/u destination pointer
822 * @param dest_cr chroma cr/v destination pointer
823 * @param dir direction (0->forward, 1->backward)
824 * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
825 * @param pix_op halfpel motion compensation function (average or put normally)
826 * @param qpix_op qpel motion compensation function (average or put normally)
827 * the motion vectors are taken from s->mv and the MV type from s->mv_type
829 static av_always_inline void mpv_motion_internal(MpegEncContext *s,
834 uint8_t **ref_picture,
835 op_pixels_func (*pix_op)[4],
836 qpel_mc_func (*qpix_op)[16],
842 prefetch_motion(s, ref_picture, dir);
844 if (!is_mpeg12 && s->obmc && s->pict_type != AV_PICTURE_TYPE_B) {
845 apply_obmc(s, dest_y, dest_cb, dest_cr, ref_picture, pix_op);
849 switch (s->mv_type) {
851 if (!is_mpeg12 && s->mcsel) {
852 if (s->real_sprite_warping_points == 1) {
853 gmc1_motion(s, dest_y, dest_cb, dest_cr,
856 gmc_motion(s, dest_y, dest_cb, dest_cr,
859 } else if (!is_mpeg12 && s->quarter_sample) {
860 qpel_motion(s, dest_y, dest_cb, dest_cr,
862 ref_picture, pix_op, qpix_op,
863 s->mv[dir][0][0], s->mv[dir][0][1], 16);
864 } else if (!is_mpeg12 && (CONFIG_WMV2_DECODER || CONFIG_WMV2_ENCODER) &&
865 s->mspel && s->codec_id == AV_CODEC_ID_WMV2) {
866 ff_mspel_motion(s, dest_y, dest_cb, dest_cr,
868 s->mv[dir][0][0], s->mv[dir][0][1], 16);
870 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
872 s->mv[dir][0][0], s->mv[dir][0][1], 16, 0, mb_y);
877 apply_8x8(s, dest_y, dest_cb, dest_cr,
878 dir, ref_picture, qpix_op, pix_op);
881 if (s->picture_structure == PICT_FRAME) {
882 if (!is_mpeg12 && s->quarter_sample) {
883 for (i = 0; i < 2; i++)
884 qpel_motion(s, dest_y, dest_cb, dest_cr,
885 1, i, s->field_select[dir][i],
886 ref_picture, pix_op, qpix_op,
887 s->mv[dir][i][0], s->mv[dir][i][1], 8);
890 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
891 0, s->field_select[dir][0],
893 s->mv[dir][0][0], s->mv[dir][0][1], 8, mb_y);
895 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
896 1, s->field_select[dir][1],
898 s->mv[dir][1][0], s->mv[dir][1][1], 8, mb_y);
901 if ( s->picture_structure != s->field_select[dir][0] + 1 && s->pict_type != AV_PICTURE_TYPE_B && !s->first_field
902 || !ref_picture[0]) {
903 ref_picture = s->current_picture_ptr->f->data;
906 mpeg_motion(s, dest_y, dest_cb, dest_cr,
907 s->field_select[dir][0],
909 s->mv[dir][0][0], s->mv[dir][0][1], 16, 0, mb_y >> 1);
913 if (CONFIG_SMALL || is_mpeg12) {
914 for (i = 0; i < 2; i++) {
915 uint8_t **ref2picture;
917 if ((s->picture_structure == s->field_select[dir][i] + 1 ||
918 s->pict_type == AV_PICTURE_TYPE_B || s->first_field) &&
920 ref2picture = ref_picture;
922 ref2picture = s->current_picture_ptr->f->data;
925 mpeg_motion(s, dest_y, dest_cb, dest_cr,
926 s->field_select[dir][i],
928 s->mv[dir][i][0], s->mv[dir][i][1],
929 8, 1, (mb_y & ~1) + i);
931 dest_y += 16 * s->linesize;
932 dest_cb += (16 >> s->chroma_y_shift) * s->uvlinesize;
933 dest_cr += (16 >> s->chroma_y_shift) * s->uvlinesize;
938 if (CONFIG_SMALL || is_mpeg12) {
939 if (s->picture_structure == PICT_FRAME) {
940 for (i = 0; i < 2; i++) {
941 for (int j = 0; j < 2; j++)
942 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
943 j, j ^ i, ref_picture, pix_op,
944 s->mv[dir][2 * i + j][0],
945 s->mv[dir][2 * i + j][1], 8, mb_y);
946 pix_op = s->hdsp.avg_pixels_tab;
949 if (!ref_picture[0]) {
950 ref_picture = s->current_picture_ptr->f->data;
952 for (i = 0; i < 2; i++) {
953 mpeg_motion(s, dest_y, dest_cb, dest_cr,
954 s->picture_structure != i + 1,
956 s->mv[dir][2 * i][0], s->mv[dir][2 * i][1],
959 // after put we make avg of the same block
960 pix_op = s->hdsp.avg_pixels_tab;
962 /* opposite parity is always in the same frame if this is
965 ref_picture = s->current_picture_ptr->f->data;
970 default: av_assert2(0);
974 void ff_mpv_motion(MpegEncContext *s,
975 uint8_t *dest_y, uint8_t *dest_cb,
976 uint8_t *dest_cr, int dir,
977 uint8_t **ref_picture,
978 op_pixels_func (*pix_op)[4],
979 qpel_mc_func (*qpix_op)[16])
982 if (s->out_format == FMT_MPEG1)
983 mpv_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
984 ref_picture, pix_op, qpix_op, 1);
987 mpv_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
988 ref_picture, pix_op, qpix_op, 0);