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
26 #include "libavutil/avassert.h"
27 #include "libavutil/internal.h"
30 #include "mpegutils.h"
31 #include "mpegvideo.h"
38 static void gmc1_motion(MpegEncContext *s,
39 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
40 uint8_t **ref_picture)
43 int src_x, src_y, motion_x, motion_y;
44 ptrdiff_t offset, linesize, uvlinesize;
47 motion_x = s->sprite_offset[0][0];
48 motion_y = s->sprite_offset[0][1];
49 src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy + 1));
50 src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy + 1));
51 motion_x <<= (3 - s->sprite_warping_accuracy);
52 motion_y <<= (3 - s->sprite_warping_accuracy);
53 src_x = av_clip(src_x, -16, s->width);
54 if (src_x == s->width)
56 src_y = av_clip(src_y, -16, s->height);
57 if (src_y == s->height)
60 linesize = s->linesize;
61 uvlinesize = s->uvlinesize;
63 ptr = ref_picture[0] + src_y * linesize + src_x;
65 if ((unsigned)src_x >= FFMAX(s->h_edge_pos - 17, 0) ||
66 (unsigned)src_y >= FFMAX(s->v_edge_pos - 17, 0)) {
67 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
71 s->h_edge_pos, s->v_edge_pos);
72 ptr = s->sc.edge_emu_buffer;
75 if ((motion_x | motion_y) & 7) {
76 s->mdsp.gmc1(dest_y, ptr, linesize, 16,
77 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
78 s->mdsp.gmc1(dest_y + 8, ptr + 8, linesize, 16,
79 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
83 dxy = ((motion_x >> 3) & 1) | ((motion_y >> 2) & 2);
85 s->hdsp.put_no_rnd_pixels_tab[0][dxy](dest_y, ptr, linesize, 16);
87 s->hdsp.put_pixels_tab[0][dxy](dest_y, ptr, linesize, 16);
91 if (CONFIG_GRAY && s->avctx->flags & AV_CODEC_FLAG_GRAY)
94 motion_x = s->sprite_offset[1][0];
95 motion_y = s->sprite_offset[1][1];
96 src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy + 1));
97 src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy + 1));
98 motion_x <<= (3 - s->sprite_warping_accuracy);
99 motion_y <<= (3 - s->sprite_warping_accuracy);
100 src_x = av_clip(src_x, -8, s->width >> 1);
101 if (src_x == s->width >> 1)
103 src_y = av_clip(src_y, -8, s->height >> 1);
104 if (src_y == s->height >> 1)
107 offset = (src_y * uvlinesize) + src_x;
108 ptr = ref_picture[1] + offset;
109 if ((unsigned)src_x >= FFMAX((s->h_edge_pos >> 1) - 9, 0) ||
110 (unsigned)src_y >= FFMAX((s->v_edge_pos >> 1) - 9, 0)) {
111 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
112 uvlinesize, uvlinesize,
115 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
116 ptr = s->sc.edge_emu_buffer;
119 s->mdsp.gmc1(dest_cb, ptr, uvlinesize, 8,
120 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
122 ptr = ref_picture[2] + offset;
124 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
125 uvlinesize, uvlinesize,
128 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
129 ptr = s->sc.edge_emu_buffer;
131 s->mdsp.gmc1(dest_cr, ptr, uvlinesize, 8,
132 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
135 static void gmc_motion(MpegEncContext *s,
136 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
137 uint8_t **ref_picture)
140 int linesize, uvlinesize;
141 const int a = s->sprite_warping_accuracy;
144 linesize = s->linesize;
145 uvlinesize = s->uvlinesize;
147 ptr = ref_picture[0];
149 ox = s->sprite_offset[0][0] + s->sprite_delta[0][0] * s->mb_x * 16 +
150 s->sprite_delta[0][1] * s->mb_y * 16;
151 oy = s->sprite_offset[0][1] + s->sprite_delta[1][0] * s->mb_x * 16 +
152 s->sprite_delta[1][1] * s->mb_y * 16;
154 s->mdsp.gmc(dest_y, ptr, linesize, 16,
156 s->sprite_delta[0][0], s->sprite_delta[0][1],
157 s->sprite_delta[1][0], s->sprite_delta[1][1],
158 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
159 s->h_edge_pos, s->v_edge_pos);
160 s->mdsp.gmc(dest_y + 8, ptr, linesize, 16,
161 ox + s->sprite_delta[0][0] * 8,
162 oy + s->sprite_delta[1][0] * 8,
163 s->sprite_delta[0][0], s->sprite_delta[0][1],
164 s->sprite_delta[1][0], s->sprite_delta[1][1],
165 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
166 s->h_edge_pos, s->v_edge_pos);
168 if (CONFIG_GRAY && s->avctx->flags & AV_CODEC_FLAG_GRAY)
171 ox = s->sprite_offset[1][0] + s->sprite_delta[0][0] * s->mb_x * 8 +
172 s->sprite_delta[0][1] * s->mb_y * 8;
173 oy = s->sprite_offset[1][1] + s->sprite_delta[1][0] * s->mb_x * 8 +
174 s->sprite_delta[1][1] * s->mb_y * 8;
176 ptr = ref_picture[1];
177 s->mdsp.gmc(dest_cb, ptr, uvlinesize, 8,
179 s->sprite_delta[0][0], s->sprite_delta[0][1],
180 s->sprite_delta[1][0], s->sprite_delta[1][1],
181 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
182 (s->h_edge_pos + 1) >> 1, (s->v_edge_pos + 1) >> 1);
184 ptr = ref_picture[2];
185 s->mdsp.gmc(dest_cr, ptr, uvlinesize, 8,
187 s->sprite_delta[0][0], s->sprite_delta[0][1],
188 s->sprite_delta[1][0], s->sprite_delta[1][1],
189 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
190 (s->h_edge_pos + 1) >> 1, (s->v_edge_pos + 1) >> 1);
193 static inline int hpel_motion(MpegEncContext *s,
194 uint8_t *dest, uint8_t *src,
195 int src_x, int src_y,
196 op_pixels_func *pix_op,
197 int motion_x, int motion_y)
202 src_x += motion_x >> 1;
203 src_y += motion_y >> 1;
205 /* WARNING: do no forget half pels */
206 src_x = av_clip(src_x, -16, s->width); // FIXME unneeded for emu?
207 if (src_x != s->width)
209 src_y = av_clip(src_y, -16, s->height);
210 if (src_y != s->height)
211 dxy |= (motion_y & 1) << 1;
212 src += src_y * s->linesize + src_x;
214 if ((unsigned)src_x >= FFMAX(s->h_edge_pos - (motion_x & 1) - 7, 0) ||
215 (unsigned)src_y >= FFMAX(s->v_edge_pos - (motion_y & 1) - 7, 0)) {
216 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, src,
217 s->linesize, s->linesize,
220 s->h_edge_pos, s->v_edge_pos);
221 src = s->sc.edge_emu_buffer;
224 pix_op[dxy](dest, src, s->linesize, 8);
228 static av_always_inline
229 void mpeg_motion_internal(MpegEncContext *s,
236 uint8_t **ref_picture,
237 op_pixels_func (*pix_op)[4],
244 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
245 int dxy, uvdxy, mx, my, src_x, src_y,
246 uvsrc_x, uvsrc_y, v_edge_pos;
247 ptrdiff_t uvlinesize, linesize;
250 if (s->quarter_sample) {
256 v_edge_pos = s->v_edge_pos >> field_based;
257 linesize = s->current_picture.f->linesize[0] << field_based;
258 uvlinesize = s->current_picture.f->linesize[1] << field_based;
260 dxy = ((motion_y & 1) << 1) | (motion_x & 1);
261 src_x = s->mb_x * 16 + (motion_x >> 1);
262 src_y = (mb_y << (4 - field_based)) + (motion_y >> 1);
264 if (!is_mpeg12 && s->out_format == FMT_H263) {
265 if ((s->workaround_bugs & FF_BUG_HPEL_CHROMA) && field_based) {
266 mx = (motion_x >> 1) | (motion_x & 1);
268 uvdxy = ((my & 1) << 1) | (mx & 1);
269 uvsrc_x = s->mb_x * 8 + (mx >> 1);
270 uvsrc_y = (mb_y << (3 - field_based)) + (my >> 1);
272 uvdxy = dxy | (motion_y & 2) | ((motion_x & 2) >> 1);
273 uvsrc_x = src_x >> 1;
274 uvsrc_y = src_y >> 1;
276 // Even chroma mv's are full pel in H261
277 } else if (!is_mpeg12 && s->out_format == FMT_H261) {
281 uvsrc_x = s->mb_x * 8 + mx;
282 uvsrc_y = mb_y * 8 + my;
284 if (s->chroma_y_shift) {
287 uvdxy = ((my & 1) << 1) | (mx & 1);
288 uvsrc_x = s->mb_x * 8 + (mx >> 1);
289 uvsrc_y = (mb_y << (3 - field_based)) + (my >> 1);
291 if (s->chroma_x_shift) {
294 uvdxy = ((motion_y & 1) << 1) | (mx & 1);
295 uvsrc_x = s->mb_x * 8 + (mx >> 1);
306 ptr_y = ref_picture[0] + src_y * linesize + src_x;
307 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
308 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
310 if ((unsigned)src_x >= FFMAX(s->h_edge_pos - (motion_x & 1) - 15 , 0) ||
311 (unsigned)src_y >= FFMAX( v_edge_pos - (motion_y & 1) - h + 1, 0)) {
313 s->codec_id == AV_CODEC_ID_MPEG2VIDEO ||
314 s->codec_id == AV_CODEC_ID_MPEG1VIDEO) {
315 av_log(s->avctx, AV_LOG_DEBUG,
316 "MPEG motion vector out of boundary (%d %d)\n", src_x,
320 src_y = (unsigned)src_y << field_based;
321 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr_y,
322 s->linesize, s->linesize,
323 17, 17 + field_based,
325 s->h_edge_pos, s->v_edge_pos);
326 ptr_y = s->sc.edge_emu_buffer;
327 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
328 uint8_t *ubuf = s->sc.edge_emu_buffer + 18 * s->linesize;
329 uint8_t *vbuf = ubuf + 10 * s->uvlinesize;
330 if (s->workaround_bugs & FF_BUG_IEDGE)
331 vbuf -= s->uvlinesize;
332 uvsrc_y = (unsigned)uvsrc_y << field_based;
333 s->vdsp.emulated_edge_mc(ubuf, ptr_cb,
334 s->uvlinesize, s->uvlinesize,
337 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
338 s->vdsp.emulated_edge_mc(vbuf, ptr_cr,
339 s->uvlinesize, s->uvlinesize,
342 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
348 /* FIXME use this for field pix too instead of the obnoxious hack which
349 * changes picture.data */
351 dest_y += s->linesize;
352 dest_cb += s->uvlinesize;
353 dest_cr += s->uvlinesize;
357 ptr_y += s->linesize;
358 ptr_cb += s->uvlinesize;
359 ptr_cr += s->uvlinesize;
362 pix_op[0][dxy](dest_y, ptr_y, linesize, h);
364 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
365 pix_op[s->chroma_x_shift][uvdxy]
366 (dest_cb, ptr_cb, uvlinesize, h >> s->chroma_y_shift);
367 pix_op[s->chroma_x_shift][uvdxy]
368 (dest_cr, ptr_cr, uvlinesize, h >> s->chroma_y_shift);
370 if (!is_mpeg12 && (CONFIG_H261_ENCODER || CONFIG_H261_DECODER) &&
371 s->out_format == FMT_H261) {
372 ff_h261_loop_filter(s);
375 /* apply one mpeg motion vector to the three components */
376 static void mpeg_motion(MpegEncContext *s,
377 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
378 int field_select, uint8_t **ref_picture,
379 op_pixels_func (*pix_op)[4],
380 int motion_x, int motion_y, int h, int mb_y)
383 if (s->out_format == FMT_MPEG1)
384 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 0, 0,
385 field_select, ref_picture, pix_op,
386 motion_x, motion_y, h, 1, mb_y);
389 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 0, 0,
390 field_select, ref_picture, pix_op,
391 motion_x, motion_y, h, 0, mb_y);
394 static void mpeg_motion_field(MpegEncContext *s, uint8_t *dest_y,
395 uint8_t *dest_cb, uint8_t *dest_cr,
396 int bottom_field, int field_select,
397 uint8_t **ref_picture,
398 op_pixels_func (*pix_op)[4],
399 int motion_x, int motion_y, int h, int mb_y)
402 if (s->out_format == FMT_MPEG1)
403 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 1,
404 bottom_field, field_select, ref_picture, pix_op,
405 motion_x, motion_y, h, 1, mb_y);
408 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 1,
409 bottom_field, field_select, ref_picture, pix_op,
410 motion_x, motion_y, h, 0, mb_y);
413 // FIXME: SIMDify, avg variant, 16x16 version
414 static inline void put_obmc(uint8_t *dst, uint8_t *src[5], int stride)
417 uint8_t *const top = src[1];
418 uint8_t *const left = src[2];
419 uint8_t *const mid = src[0];
420 uint8_t *const right = src[3];
421 uint8_t *const bottom = src[4];
422 #define OBMC_FILTER(x, t, l, m, r, b)\
423 dst[x]= (t*top[x] + l*left[x] + m*mid[x] + r*right[x] + b*bottom[x] + 4)>>3
424 #define OBMC_FILTER4(x, t, l, m, r, b)\
425 OBMC_FILTER(x , t, l, m, r, b);\
426 OBMC_FILTER(x+1 , t, l, m, r, b);\
427 OBMC_FILTER(x +stride, t, l, m, r, b);\
428 OBMC_FILTER(x+1+stride, t, l, m, r, b);
431 OBMC_FILTER (x , 2, 2, 4, 0, 0);
432 OBMC_FILTER (x + 1, 2, 1, 5, 0, 0);
433 OBMC_FILTER4(x + 2, 2, 1, 5, 0, 0);
434 OBMC_FILTER4(x + 4, 2, 0, 5, 1, 0);
435 OBMC_FILTER (x + 6, 2, 0, 5, 1, 0);
436 OBMC_FILTER (x + 7, 2, 0, 4, 2, 0);
438 OBMC_FILTER (x , 1, 2, 5, 0, 0);
439 OBMC_FILTER (x + 1, 1, 2, 5, 0, 0);
440 OBMC_FILTER (x + 6, 1, 0, 5, 2, 0);
441 OBMC_FILTER (x + 7, 1, 0, 5, 2, 0);
443 OBMC_FILTER4(x , 1, 2, 5, 0, 0);
444 OBMC_FILTER4(x + 2, 1, 1, 6, 0, 0);
445 OBMC_FILTER4(x + 4, 1, 0, 6, 1, 0);
446 OBMC_FILTER4(x + 6, 1, 0, 5, 2, 0);
448 OBMC_FILTER4(x , 0, 2, 5, 0, 1);
449 OBMC_FILTER4(x + 2, 0, 1, 6, 0, 1);
450 OBMC_FILTER4(x + 4, 0, 0, 6, 1, 1);
451 OBMC_FILTER4(x + 6, 0, 0, 5, 2, 1);
453 OBMC_FILTER (x , 0, 2, 5, 0, 1);
454 OBMC_FILTER (x + 1, 0, 2, 5, 0, 1);
455 OBMC_FILTER4(x + 2, 0, 1, 5, 0, 2);
456 OBMC_FILTER4(x + 4, 0, 0, 5, 1, 2);
457 OBMC_FILTER (x + 6, 0, 0, 5, 2, 1);
458 OBMC_FILTER (x + 7, 0, 0, 5, 2, 1);
460 OBMC_FILTER (x , 0, 2, 4, 0, 2);
461 OBMC_FILTER (x + 1, 0, 1, 5, 0, 2);
462 OBMC_FILTER (x + 6, 0, 0, 5, 1, 2);
463 OBMC_FILTER (x + 7, 0, 0, 4, 2, 2);
466 /* obmc for 1 8x8 luma block */
467 static inline void obmc_motion(MpegEncContext *s,
468 uint8_t *dest, uint8_t *src,
469 int src_x, int src_y,
470 op_pixels_func *pix_op,
471 int16_t mv[5][2] /* mid top left right bottom */)
477 av_assert2(s->quarter_sample == 0);
479 for (i = 0; i < 5; i++) {
480 if (i && mv[i][0] == mv[MID][0] && mv[i][1] == mv[MID][1]) {
483 ptr[i] = s->sc.obmc_scratchpad + 8 * (i & 1) +
484 s->linesize * 8 * (i >> 1);
485 hpel_motion(s, ptr[i], src, src_x, src_y, pix_op,
490 put_obmc(dest, ptr, s->linesize);
493 static inline void qpel_motion(MpegEncContext *s,
497 int field_based, int bottom_field,
498 int field_select, uint8_t **ref_picture,
499 op_pixels_func (*pix_op)[4],
500 qpel_mc_func (*qpix_op)[16],
501 int motion_x, int motion_y, int h)
503 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
504 int dxy, uvdxy, mx, my, src_x, src_y, uvsrc_x, uvsrc_y, v_edge_pos;
505 ptrdiff_t linesize, uvlinesize;
507 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
509 src_x = s->mb_x * 16 + (motion_x >> 2);
510 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2);
512 v_edge_pos = s->v_edge_pos >> field_based;
513 linesize = s->linesize << field_based;
514 uvlinesize = s->uvlinesize << field_based;
519 } else if (s->workaround_bugs & FF_BUG_QPEL_CHROMA2) {
520 static const int rtab[8] = { 0, 0, 1, 1, 0, 0, 0, 1 };
521 mx = (motion_x >> 1) + rtab[motion_x & 7];
522 my = (motion_y >> 1) + rtab[motion_y & 7];
523 } else if (s->workaround_bugs & FF_BUG_QPEL_CHROMA) {
524 mx = (motion_x >> 1) | (motion_x & 1);
525 my = (motion_y >> 1) | (motion_y & 1);
530 mx = (mx >> 1) | (mx & 1);
531 my = (my >> 1) | (my & 1);
533 uvdxy = (mx & 1) | ((my & 1) << 1);
537 uvsrc_x = s->mb_x * 8 + mx;
538 uvsrc_y = s->mb_y * (8 >> field_based) + my;
540 ptr_y = ref_picture[0] + src_y * linesize + src_x;
541 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
542 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
544 if ((unsigned)src_x >= FFMAX(s->h_edge_pos - (motion_x & 3) - 15 , 0) ||
545 (unsigned)src_y >= FFMAX( v_edge_pos - (motion_y & 3) - h + 1, 0)) {
546 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr_y,
547 s->linesize, s->linesize,
548 17, 17 + field_based,
549 src_x, src_y * (1 << field_based),
550 s->h_edge_pos, s->v_edge_pos);
551 ptr_y = s->sc.edge_emu_buffer;
552 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
553 uint8_t *ubuf = s->sc.edge_emu_buffer + 18 * s->linesize;
554 uint8_t *vbuf = ubuf + 10 * s->uvlinesize;
555 if (s->workaround_bugs & FF_BUG_IEDGE)
556 vbuf -= s->uvlinesize;
557 s->vdsp.emulated_edge_mc(ubuf, ptr_cb,
558 s->uvlinesize, s->uvlinesize,
560 uvsrc_x, uvsrc_y * (1 << field_based),
561 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
562 s->vdsp.emulated_edge_mc(vbuf, ptr_cr,
563 s->uvlinesize, s->uvlinesize,
565 uvsrc_x, uvsrc_y * (1 << field_based),
566 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
573 qpix_op[0][dxy](dest_y, ptr_y, linesize);
576 dest_y += s->linesize;
577 dest_cb += s->uvlinesize;
578 dest_cr += s->uvlinesize;
582 ptr_y += s->linesize;
583 ptr_cb += s->uvlinesize;
584 ptr_cr += s->uvlinesize;
586 // damn interlaced mode
587 // FIXME boundary mirroring is not exactly correct here
588 qpix_op[1][dxy](dest_y, ptr_y, linesize);
589 qpix_op[1][dxy](dest_y + 8, ptr_y + 8, linesize);
591 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
592 pix_op[1][uvdxy](dest_cr, ptr_cr, uvlinesize, h >> 1);
593 pix_op[1][uvdxy](dest_cb, ptr_cb, uvlinesize, h >> 1);
598 * H.263 chroma 4mv motion compensation.
600 static void chroma_4mv_motion(MpegEncContext *s,
601 uint8_t *dest_cb, uint8_t *dest_cr,
602 uint8_t **ref_picture,
603 op_pixels_func *pix_op,
607 int src_x, src_y, dxy, emu = 0;
610 /* In case of 8X8, we construct a single chroma motion vector
611 * with a special rounding */
612 mx = ff_h263_round_chroma(mx);
613 my = ff_h263_round_chroma(my);
615 dxy = ((my & 1) << 1) | (mx & 1);
619 src_x = s->mb_x * 8 + mx;
620 src_y = s->mb_y * 8 + my;
621 src_x = av_clip(src_x, -8, (s->width >> 1));
622 if (src_x == (s->width >> 1))
624 src_y = av_clip(src_y, -8, (s->height >> 1));
625 if (src_y == (s->height >> 1))
628 offset = src_y * s->uvlinesize + src_x;
629 ptr = ref_picture[1] + offset;
630 if ((unsigned)src_x >= FFMAX((s->h_edge_pos >> 1) - (dxy & 1) - 7, 0) ||
631 (unsigned)src_y >= FFMAX((s->v_edge_pos >> 1) - (dxy >> 1) - 7, 0)) {
632 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
633 s->uvlinesize, s->uvlinesize,
635 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
636 ptr = s->sc.edge_emu_buffer;
639 pix_op[dxy](dest_cb, ptr, s->uvlinesize, 8);
641 ptr = ref_picture[2] + offset;
643 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
644 s->uvlinesize, s->uvlinesize,
646 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
647 ptr = s->sc.edge_emu_buffer;
649 pix_op[dxy](dest_cr, ptr, s->uvlinesize, 8);
652 static inline void prefetch_motion(MpegEncContext *s, uint8_t **pix, int dir)
654 /* fetch pixels for estimated mv 4 macroblocks ahead
655 * optimized for 64byte cache lines */
656 const int shift = s->quarter_sample ? 2 : 1;
657 const int mx = (s->mv[dir][0][0] >> shift) + 16 * s->mb_x + 8;
658 const int my = (s->mv[dir][0][1] >> shift) + 16 * s->mb_y;
659 int off = mx + (my + (s->mb_x & 3) * 4) * s->linesize + 64;
661 s->vdsp.prefetch(pix[0] + off, s->linesize, 4);
662 off = (mx >> 1) + ((my >> 1) + (s->mb_x & 7)) * s->uvlinesize + 64;
663 s->vdsp.prefetch(pix[1] + off, pix[2] - pix[1], 2);
666 static inline void apply_obmc(MpegEncContext *s,
670 uint8_t **ref_picture,
671 op_pixels_func (*pix_op)[4])
673 LOCAL_ALIGNED_8(int16_t, mv_cache, [4], [4][2]);
674 Picture *cur_frame = &s->current_picture;
677 const int xy = mb_x + mb_y * s->mb_stride;
678 const int mot_stride = s->b8_stride;
679 const int mot_xy = mb_x * 2 + mb_y * 2 * mot_stride;
682 av_assert2(!s->mb_skipped);
684 AV_COPY32(mv_cache[1][1], cur_frame->motion_val[0][mot_xy]);
685 AV_COPY32(mv_cache[1][2], cur_frame->motion_val[0][mot_xy + 1]);
687 AV_COPY32(mv_cache[2][1],
688 cur_frame->motion_val[0][mot_xy + mot_stride]);
689 AV_COPY32(mv_cache[2][2],
690 cur_frame->motion_val[0][mot_xy + mot_stride + 1]);
692 AV_COPY32(mv_cache[3][1],
693 cur_frame->motion_val[0][mot_xy + mot_stride]);
694 AV_COPY32(mv_cache[3][2],
695 cur_frame->motion_val[0][mot_xy + mot_stride + 1]);
697 if (mb_y == 0 || IS_INTRA(cur_frame->mb_type[xy - s->mb_stride])) {
698 AV_COPY32(mv_cache[0][1], mv_cache[1][1]);
699 AV_COPY32(mv_cache[0][2], mv_cache[1][2]);
701 AV_COPY32(mv_cache[0][1],
702 cur_frame->motion_val[0][mot_xy - mot_stride]);
703 AV_COPY32(mv_cache[0][2],
704 cur_frame->motion_val[0][mot_xy - mot_stride + 1]);
707 if (mb_x == 0 || IS_INTRA(cur_frame->mb_type[xy - 1])) {
708 AV_COPY32(mv_cache[1][0], mv_cache[1][1]);
709 AV_COPY32(mv_cache[2][0], mv_cache[2][1]);
711 AV_COPY32(mv_cache[1][0], cur_frame->motion_val[0][mot_xy - 1]);
712 AV_COPY32(mv_cache[2][0],
713 cur_frame->motion_val[0][mot_xy - 1 + mot_stride]);
716 if (mb_x + 1 >= s->mb_width || IS_INTRA(cur_frame->mb_type[xy + 1])) {
717 AV_COPY32(mv_cache[1][3], mv_cache[1][2]);
718 AV_COPY32(mv_cache[2][3], mv_cache[2][2]);
720 AV_COPY32(mv_cache[1][3], cur_frame->motion_val[0][mot_xy + 2]);
721 AV_COPY32(mv_cache[2][3],
722 cur_frame->motion_val[0][mot_xy + 2 + mot_stride]);
727 for (i = 0; i < 4; i++) {
728 const int x = (i & 1) + 1;
729 const int y = (i >> 1) + 1;
731 { mv_cache[y][x][0], mv_cache[y][x][1] },
732 { mv_cache[y - 1][x][0], mv_cache[y - 1][x][1] },
733 { mv_cache[y][x - 1][0], mv_cache[y][x - 1][1] },
734 { mv_cache[y][x + 1][0], mv_cache[y][x + 1][1] },
735 { mv_cache[y + 1][x][0], mv_cache[y + 1][x][1] }
738 obmc_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
740 mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >> 1) * 8,
747 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY))
748 chroma_4mv_motion(s, dest_cb, dest_cr,
749 ref_picture, pix_op[1],
753 static inline void apply_8x8(MpegEncContext *s,
758 uint8_t **ref_picture,
759 qpel_mc_func (*qpix_op)[16],
760 op_pixels_func (*pix_op)[4])
762 int dxy, mx, my, src_x, src_y;
770 if (s->quarter_sample) {
771 for (i = 0; i < 4; i++) {
772 int motion_x = s->mv[dir][i][0];
773 int motion_y = s->mv[dir][i][1];
775 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
776 src_x = mb_x * 16 + (motion_x >> 2) + (i & 1) * 8;
777 src_y = mb_y * 16 + (motion_y >> 2) + (i >> 1) * 8;
779 /* WARNING: do no forget half pels */
780 src_x = av_clip(src_x, -16, s->width);
781 if (src_x == s->width)
783 src_y = av_clip(src_y, -16, s->height);
784 if (src_y == s->height)
787 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x);
788 if ((unsigned)src_x >= FFMAX(s->h_edge_pos - (motion_x & 3) - 7, 0) ||
789 (unsigned)src_y >= FFMAX(s->v_edge_pos - (motion_y & 3) - 7, 0)) {
790 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
791 s->linesize, s->linesize,
796 ptr = s->sc.edge_emu_buffer;
798 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize;
799 qpix_op[1][dxy](dest, ptr, s->linesize);
801 mx += s->mv[dir][i][0] / 2;
802 my += s->mv[dir][i][1] / 2;
805 for (i = 0; i < 4; i++) {
807 dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
809 mb_x * 16 + (i & 1) * 8,
810 mb_y * 16 + (i >> 1) * 8,
815 mx += s->mv[dir][i][0];
816 my += s->mv[dir][i][1];
820 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY))
821 chroma_4mv_motion(s, dest_cb, dest_cr,
822 ref_picture, pix_op[1], mx, my);
826 * motion compensation of a single macroblock
828 * @param dest_y luma destination pointer
829 * @param dest_cb chroma cb/u destination pointer
830 * @param dest_cr chroma cr/v destination pointer
831 * @param dir direction (0->forward, 1->backward)
832 * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
833 * @param pix_op halfpel motion compensation function (average or put normally)
834 * @param qpix_op qpel motion compensation function (average or put normally)
835 * the motion vectors are taken from s->mv and the MV type from s->mv_type
837 static av_always_inline void mpv_motion_internal(MpegEncContext *s,
842 uint8_t **ref_picture,
843 op_pixels_func (*pix_op)[4],
844 qpel_mc_func (*qpix_op)[16],
850 prefetch_motion(s, ref_picture, dir);
852 if (!is_mpeg12 && s->obmc && s->pict_type != AV_PICTURE_TYPE_B) {
853 apply_obmc(s, dest_y, dest_cb, dest_cr, ref_picture, pix_op);
857 switch (s->mv_type) {
860 if (s->real_sprite_warping_points == 1) {
861 gmc1_motion(s, dest_y, dest_cb, dest_cr,
864 gmc_motion(s, dest_y, dest_cb, dest_cr,
867 } else if (!is_mpeg12 && s->quarter_sample) {
868 qpel_motion(s, dest_y, dest_cb, dest_cr,
870 ref_picture, pix_op, qpix_op,
871 s->mv[dir][0][0], s->mv[dir][0][1], 16);
872 } else if (!is_mpeg12 && (CONFIG_WMV2_DECODER || CONFIG_WMV2_ENCODER) &&
873 s->mspel && s->codec_id == AV_CODEC_ID_WMV2) {
874 ff_mspel_motion(s, dest_y, dest_cb, dest_cr,
876 s->mv[dir][0][0], s->mv[dir][0][1], 16);
878 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
880 s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y);
885 apply_8x8(s, dest_y, dest_cb, dest_cr,
886 dir, ref_picture, qpix_op, pix_op);
889 if (s->picture_structure == PICT_FRAME) {
890 if (!is_mpeg12 && s->quarter_sample) {
891 for (i = 0; i < 2; i++)
892 qpel_motion(s, dest_y, dest_cb, dest_cr,
893 1, i, s->field_select[dir][i],
894 ref_picture, pix_op, qpix_op,
895 s->mv[dir][i][0], s->mv[dir][i][1], 8);
898 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
899 0, s->field_select[dir][0],
901 s->mv[dir][0][0], s->mv[dir][0][1], 8, mb_y);
903 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
904 1, s->field_select[dir][1],
906 s->mv[dir][1][0], s->mv[dir][1][1], 8, mb_y);
909 if ( s->picture_structure != s->field_select[dir][0] + 1 && s->pict_type != AV_PICTURE_TYPE_B && !s->first_field
910 || !ref_picture[0]) {
911 ref_picture = s->current_picture_ptr->f->data;
914 mpeg_motion(s, dest_y, dest_cb, dest_cr,
915 s->field_select[dir][0],
917 s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y >> 1);
921 for (i = 0; i < 2; i++) {
922 uint8_t **ref2picture;
924 if ((s->picture_structure == s->field_select[dir][i] + 1
925 || s->pict_type == AV_PICTURE_TYPE_B || s->first_field) && ref_picture[0]) {
926 ref2picture = ref_picture;
928 ref2picture = s->current_picture_ptr->f->data;
931 mpeg_motion(s, dest_y, dest_cb, dest_cr,
932 s->field_select[dir][i],
934 s->mv[dir][i][0], s->mv[dir][i][1] + 16 * i,
937 dest_y += 16 * s->linesize;
938 dest_cb += (16 >> s->chroma_y_shift) * s->uvlinesize;
939 dest_cr += (16 >> s->chroma_y_shift) * s->uvlinesize;
943 if (s->picture_structure == PICT_FRAME) {
944 for (i = 0; i < 2; i++) {
946 for (j = 0; j < 2; j++)
947 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
948 j, j ^ i, ref_picture, pix_op,
949 s->mv[dir][2 * i + j][0],
950 s->mv[dir][2 * i + j][1], 8, mb_y);
951 pix_op = s->hdsp.avg_pixels_tab;
954 if (!ref_picture[0]) {
955 ref_picture = s->current_picture_ptr->f->data;
957 for (i = 0; i < 2; i++) {
958 mpeg_motion(s, dest_y, dest_cb, dest_cr,
959 s->picture_structure != i + 1,
961 s->mv[dir][2 * i][0], s->mv[dir][2 * i][1],
964 // after put we make avg of the same block
965 pix_op = s->hdsp.avg_pixels_tab;
967 /* opposite parity is always in the same frame if this is
969 if (!s->first_field) {
970 ref_picture = s->current_picture_ptr->f->data;
975 default: av_assert2(0);
979 void ff_mpv_motion(MpegEncContext *s,
980 uint8_t *dest_y, uint8_t *dest_cb,
981 uint8_t *dest_cr, int dir,
982 uint8_t **ref_picture,
983 op_pixels_func (*pix_op)[4],
984 qpel_mc_func (*qpix_op)[16])
987 if (s->out_format == FMT_MPEG1)
988 mpv_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
989 ref_picture, pix_op, qpix_op, 1);
992 mpv_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
993 ref_picture, pix_op, qpix_op, 0);