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 (s->unrestricted_mv) {
214 if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 1) - 8, 0) ||
215 (unsigned)src_y > FFMAX(s->v_edge_pos - (motion_y & 1) - 8, 0)) {
216 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, src,
217 s->linesize, s->linesize,
219 src_x, src_y, s->h_edge_pos,
221 src = s->sc.edge_emu_buffer;
225 pix_op[dxy](dest, src, s->linesize, 8);
229 static av_always_inline
230 void emulated_edge_mc(MpegEncContext *s,
231 int src_x, int src_y,
232 int uvsrc_x, int uvsrc_y,
238 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, *ptr_y,
239 s->linesize, s->linesize,
240 17, 17 + field_based,
241 src_x, src_y * (1 << field_based),
242 s->h_edge_pos, s->v_edge_pos);
243 *ptr_y = s->sc.edge_emu_buffer;
244 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
245 uint8_t *uvbuf = s->sc.edge_emu_buffer + 18 * s->linesize;
246 s->vdsp.emulated_edge_mc(uvbuf, *ptr_cb,
247 s->uvlinesize, s->uvlinesize,
249 uvsrc_x, uvsrc_y * (1 << field_based),
250 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
251 s->vdsp.emulated_edge_mc(uvbuf + 16, *ptr_cr,
252 s->uvlinesize, s->uvlinesize,
254 uvsrc_x, uvsrc_y * (1 << field_based),
255 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
257 *ptr_cr = uvbuf + 16;
261 static av_always_inline
262 void mpeg_motion_internal(MpegEncContext *s,
269 uint8_t **ref_picture,
270 op_pixels_func (*pix_op)[4],
277 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
278 int dxy, uvdxy, mx, my, src_x, src_y,
279 uvsrc_x, uvsrc_y, v_edge_pos;
280 ptrdiff_t uvlinesize, linesize;
282 v_edge_pos = s->v_edge_pos >> field_based;
283 linesize = s->current_picture.f->linesize[0] << field_based;
284 uvlinesize = s->current_picture.f->linesize[1] << field_based;
286 dxy = ((motion_y & 1) << 1) | (motion_x & 1);
287 src_x = s->mb_x * 16 + (motion_x >> 1);
288 src_y = (mb_y << (4 - field_based)) + (motion_y >> 1);
290 if (!is_mpeg12 && s->out_format == FMT_H263) {
291 if ((s->workaround_bugs & FF_BUG_HPEL_CHROMA) && field_based) {
292 mx = (motion_x >> 1) | (motion_x & 1);
294 uvdxy = ((my & 1) << 1) | (mx & 1);
295 uvsrc_x = s->mb_x * 8 + (mx >> 1);
296 uvsrc_y = (mb_y << (3 - field_based)) + (my >> 1);
298 uvdxy = dxy | (motion_y & 2) | ((motion_x & 2) >> 1);
299 uvsrc_x = src_x >> 1;
300 uvsrc_y = src_y >> 1;
302 // Even chroma mv's are full pel in H261
303 } else if (!is_mpeg12 && s->out_format == FMT_H261) {
307 uvsrc_x = s->mb_x * 8 + mx;
308 uvsrc_y = mb_y * 8 + my;
310 if (s->chroma_y_shift) {
313 uvdxy = ((my & 1) << 1) | (mx & 1);
314 uvsrc_x = s->mb_x * 8 + (mx >> 1);
315 uvsrc_y = (mb_y << (3 - field_based)) + (my >> 1);
317 if (s->chroma_x_shift) {
320 uvdxy = ((motion_y & 1) << 1) | (mx & 1);
321 uvsrc_x = s->mb_x * 8 + (mx >> 1);
332 ptr_y = ref_picture[0] + src_y * linesize + src_x;
333 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
334 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
336 if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 1) - 16, 0) ||
337 (unsigned)src_y > FFMAX(v_edge_pos - (motion_y & 1) - h, 0)) {
339 s->codec_id == AV_CODEC_ID_MPEG2VIDEO ||
340 s->codec_id == AV_CODEC_ID_MPEG1VIDEO) {
341 av_log(s->avctx, AV_LOG_DEBUG,
342 "MPEG motion vector out of boundary (%d %d)\n", src_x,
346 emulated_edge_mc(s, src_x, src_y, uvsrc_x, uvsrc_y, field_based,
347 &ptr_y, &ptr_cb, &ptr_cr);
350 /* FIXME use this for field pix too instead of the obnoxious hack which
351 * changes picture.data */
353 dest_y += s->linesize;
354 dest_cb += s->uvlinesize;
355 dest_cr += s->uvlinesize;
359 ptr_y += s->linesize;
360 ptr_cb += s->uvlinesize;
361 ptr_cr += s->uvlinesize;
364 pix_op[0][dxy](dest_y, ptr_y, linesize, h);
366 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
367 pix_op[s->chroma_x_shift][uvdxy]
368 (dest_cb, ptr_cb, uvlinesize, h >> s->chroma_y_shift);
369 pix_op[s->chroma_x_shift][uvdxy]
370 (dest_cr, ptr_cr, uvlinesize, h >> s->chroma_y_shift);
372 if (!is_mpeg12 && (CONFIG_H261_ENCODER || CONFIG_H261_DECODER) &&
373 s->out_format == FMT_H261) {
374 ff_h261_loop_filter(s);
377 /* apply one mpeg motion vector to the three components */
378 static void mpeg_motion(MpegEncContext *s,
379 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
380 int field_select, uint8_t **ref_picture,
381 op_pixels_func (*pix_op)[4],
382 int motion_x, int motion_y, int h, int mb_y)
385 if (s->out_format == FMT_MPEG1)
386 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 0, 0,
387 field_select, ref_picture, pix_op,
388 motion_x, motion_y, h, 1, mb_y);
391 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 0, 0,
392 field_select, ref_picture, pix_op,
393 motion_x, motion_y, h, 0, mb_y);
396 static void mpeg_motion_field(MpegEncContext *s, uint8_t *dest_y,
397 uint8_t *dest_cb, uint8_t *dest_cr,
398 int bottom_field, int field_select,
399 uint8_t **ref_picture,
400 op_pixels_func (*pix_op)[4],
401 int motion_x, int motion_y, int h, int mb_y)
404 if(s->out_format == FMT_MPEG1)
405 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 1,
406 bottom_field, field_select, ref_picture, pix_op,
407 motion_x, motion_y, h, 1, mb_y);
410 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 1,
411 bottom_field, field_select, ref_picture, pix_op,
412 motion_x, motion_y, h, 0, mb_y);
415 // FIXME: SIMDify, avg variant, 16x16 version
416 static inline void put_obmc(uint8_t *dst, uint8_t *src[5], int stride)
419 uint8_t *const top = src[1];
420 uint8_t *const left = src[2];
421 uint8_t *const mid = src[0];
422 uint8_t *const right = src[3];
423 uint8_t *const bottom = src[4];
424 #define OBMC_FILTER(x, t, l, m, r, b)\
425 dst[x]= (t*top[x] + l*left[x] + m*mid[x] + r*right[x] + b*bottom[x] + 4)>>3
426 #define OBMC_FILTER4(x, t, l, m, r, b)\
427 OBMC_FILTER(x , t, l, m, r, b);\
428 OBMC_FILTER(x+1 , t, l, m, r, b);\
429 OBMC_FILTER(x +stride, t, l, m, r, b);\
430 OBMC_FILTER(x+1+stride, t, l, m, r, b);
433 OBMC_FILTER (x , 2, 2, 4, 0, 0);
434 OBMC_FILTER (x + 1, 2, 1, 5, 0, 0);
435 OBMC_FILTER4(x + 2, 2, 1, 5, 0, 0);
436 OBMC_FILTER4(x + 4, 2, 0, 5, 1, 0);
437 OBMC_FILTER (x + 6, 2, 0, 5, 1, 0);
438 OBMC_FILTER (x + 7, 2, 0, 4, 2, 0);
440 OBMC_FILTER (x , 1, 2, 5, 0, 0);
441 OBMC_FILTER (x + 1, 1, 2, 5, 0, 0);
442 OBMC_FILTER (x + 6, 1, 0, 5, 2, 0);
443 OBMC_FILTER (x + 7, 1, 0, 5, 2, 0);
445 OBMC_FILTER4(x , 1, 2, 5, 0, 0);
446 OBMC_FILTER4(x + 2, 1, 1, 6, 0, 0);
447 OBMC_FILTER4(x + 4, 1, 0, 6, 1, 0);
448 OBMC_FILTER4(x + 6, 1, 0, 5, 2, 0);
450 OBMC_FILTER4(x , 0, 2, 5, 0, 1);
451 OBMC_FILTER4(x + 2, 0, 1, 6, 0, 1);
452 OBMC_FILTER4(x + 4, 0, 0, 6, 1, 1);
453 OBMC_FILTER4(x + 6, 0, 0, 5, 2, 1);
455 OBMC_FILTER (x , 0, 2, 5, 0, 1);
456 OBMC_FILTER (x + 1, 0, 2, 5, 0, 1);
457 OBMC_FILTER4(x + 2, 0, 1, 5, 0, 2);
458 OBMC_FILTER4(x + 4, 0, 0, 5, 1, 2);
459 OBMC_FILTER (x + 6, 0, 0, 5, 2, 1);
460 OBMC_FILTER (x + 7, 0, 0, 5, 2, 1);
462 OBMC_FILTER (x , 0, 2, 4, 0, 2);
463 OBMC_FILTER (x + 1, 0, 1, 5, 0, 2);
464 OBMC_FILTER (x + 6, 0, 0, 5, 1, 2);
465 OBMC_FILTER (x + 7, 0, 0, 4, 2, 2);
468 /* obmc for 1 8x8 luma block */
469 static inline void obmc_motion(MpegEncContext *s,
470 uint8_t *dest, uint8_t *src,
471 int src_x, int src_y,
472 op_pixels_func *pix_op,
473 int16_t mv[5][2] /* mid top left right bottom */)
479 assert(s->quarter_sample == 0);
481 for (i = 0; i < 5; i++) {
482 if (i && mv[i][0] == mv[MID][0] && mv[i][1] == mv[MID][1]) {
485 ptr[i] = s->sc.obmc_scratchpad + 8 * (i & 1) +
486 s->linesize * 8 * (i >> 1);
487 hpel_motion(s, ptr[i], src, src_x, src_y, pix_op,
492 put_obmc(dest, ptr, s->linesize);
495 static inline void qpel_motion(MpegEncContext *s,
499 int field_based, int bottom_field,
500 int field_select, uint8_t **ref_picture,
501 op_pixels_func (*pix_op)[4],
502 qpel_mc_func (*qpix_op)[16],
503 int motion_x, int motion_y, int h)
505 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
506 int dxy, uvdxy, mx, my, src_x, src_y, uvsrc_x, uvsrc_y, v_edge_pos;
507 ptrdiff_t linesize, uvlinesize;
509 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
511 src_x = s->mb_x * 16 + (motion_x >> 2);
512 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2);
514 v_edge_pos = s->v_edge_pos >> field_based;
515 linesize = s->linesize << field_based;
516 uvlinesize = s->uvlinesize << field_based;
521 } else if (s->workaround_bugs & FF_BUG_QPEL_CHROMA2) {
522 static const int rtab[8] = { 0, 0, 1, 1, 0, 0, 0, 1 };
523 mx = (motion_x >> 1) + rtab[motion_x & 7];
524 my = (motion_y >> 1) + rtab[motion_y & 7];
525 } else if (s->workaround_bugs & FF_BUG_QPEL_CHROMA) {
526 mx = (motion_x >> 1) | (motion_x & 1);
527 my = (motion_y >> 1) | (motion_y & 1);
532 mx = (mx >> 1) | (mx & 1);
533 my = (my >> 1) | (my & 1);
535 uvdxy = (mx & 1) | ((my & 1) << 1);
539 uvsrc_x = s->mb_x * 8 + mx;
540 uvsrc_y = s->mb_y * (8 >> field_based) + my;
542 ptr_y = ref_picture[0] + src_y * linesize + src_x;
543 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
544 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
546 if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 3) - 16, 0) ||
547 (unsigned)src_y > FFMAX(v_edge_pos - (motion_y & 3) - h, 0)) {
548 emulated_edge_mc(s, src_x, src_y, uvsrc_x, uvsrc_y, field_based,
549 &ptr_y, &ptr_cb, &ptr_cr);
553 qpix_op[0][dxy](dest_y, ptr_y, linesize);
556 dest_y += s->linesize;
557 dest_cb += s->uvlinesize;
558 dest_cr += s->uvlinesize;
562 ptr_y += s->linesize;
563 ptr_cb += s->uvlinesize;
564 ptr_cr += s->uvlinesize;
566 // damn interlaced mode
567 // FIXME boundary mirroring is not exactly correct here
568 qpix_op[1][dxy](dest_y, ptr_y, linesize);
569 qpix_op[1][dxy](dest_y + 8, ptr_y + 8, linesize);
571 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
572 pix_op[1][uvdxy](dest_cr, ptr_cr, uvlinesize, h >> 1);
573 pix_op[1][uvdxy](dest_cb, ptr_cb, uvlinesize, h >> 1);
578 * H.263 chroma 4mv motion compensation.
580 static void chroma_4mv_motion(MpegEncContext *s,
581 uint8_t *dest_cb, uint8_t *dest_cr,
582 uint8_t **ref_picture,
583 op_pixels_func *pix_op,
587 int src_x, src_y, dxy, emu = 0;
590 /* In case of 8X8, we construct a single chroma motion vector
591 * with a special rounding */
592 mx = ff_h263_round_chroma(mx);
593 my = ff_h263_round_chroma(my);
595 dxy = ((my & 1) << 1) | (mx & 1);
599 src_x = s->mb_x * 8 + mx;
600 src_y = s->mb_y * 8 + my;
601 src_x = av_clip(src_x, -8, (s->width >> 1));
602 if (src_x == (s->width >> 1))
604 src_y = av_clip(src_y, -8, (s->height >> 1));
605 if (src_y == (s->height >> 1))
608 offset = src_y * s->uvlinesize + src_x;
609 ptr = ref_picture[1] + offset;
610 if ((unsigned)src_x > FFMAX((s->h_edge_pos >> 1) - (dxy & 1) - 8, 0) ||
611 (unsigned)src_y > FFMAX((s->v_edge_pos >> 1) - (dxy >> 1) - 8, 0)) {
612 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
613 s->uvlinesize, s->uvlinesize,
615 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
616 ptr = s->sc.edge_emu_buffer;
619 pix_op[dxy](dest_cb, ptr, s->uvlinesize, 8);
621 ptr = ref_picture[2] + offset;
623 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
624 s->uvlinesize, s->uvlinesize,
626 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
627 ptr = s->sc.edge_emu_buffer;
629 pix_op[dxy](dest_cr, ptr, s->uvlinesize, 8);
632 static inline void prefetch_motion(MpegEncContext *s, uint8_t **pix, int dir)
634 /* fetch pixels for estimated mv 4 macroblocks ahead
635 * optimized for 64byte cache lines */
636 const int shift = s->quarter_sample ? 2 : 1;
637 const int mx = (s->mv[dir][0][0] >> shift) + 16 * s->mb_x + 8;
638 const int my = (s->mv[dir][0][1] >> shift) + 16 * s->mb_y;
639 int off = mx + (my + (s->mb_x & 3) * 4) * s->linesize + 64;
641 s->vdsp.prefetch(pix[0] + off, s->linesize, 4);
642 off = (mx >> 1) + ((my >> 1) + (s->mb_x & 7)) * s->uvlinesize + 64;
643 s->vdsp.prefetch(pix[1] + off, pix[2] - pix[1], 2);
646 static inline void apply_obmc(MpegEncContext *s,
650 uint8_t **ref_picture,
651 op_pixels_func (*pix_op)[4])
653 LOCAL_ALIGNED_8(int16_t, mv_cache, [4], [4][2]);
654 Picture *cur_frame = &s->current_picture;
657 const int xy = mb_x + mb_y * s->mb_stride;
658 const int mot_stride = s->b8_stride;
659 const int mot_xy = mb_x * 2 + mb_y * 2 * mot_stride;
662 assert(!s->mb_skipped);
664 AV_COPY32(mv_cache[1][1], cur_frame->motion_val[0][mot_xy]);
665 AV_COPY32(mv_cache[1][2], cur_frame->motion_val[0][mot_xy + 1]);
667 AV_COPY32(mv_cache[2][1],
668 cur_frame->motion_val[0][mot_xy + mot_stride]);
669 AV_COPY32(mv_cache[2][2],
670 cur_frame->motion_val[0][mot_xy + mot_stride + 1]);
672 AV_COPY32(mv_cache[3][1],
673 cur_frame->motion_val[0][mot_xy + mot_stride]);
674 AV_COPY32(mv_cache[3][2],
675 cur_frame->motion_val[0][mot_xy + mot_stride + 1]);
677 if (mb_y == 0 || IS_INTRA(cur_frame->mb_type[xy - s->mb_stride])) {
678 AV_COPY32(mv_cache[0][1], mv_cache[1][1]);
679 AV_COPY32(mv_cache[0][2], mv_cache[1][2]);
681 AV_COPY32(mv_cache[0][1],
682 cur_frame->motion_val[0][mot_xy - mot_stride]);
683 AV_COPY32(mv_cache[0][2],
684 cur_frame->motion_val[0][mot_xy - mot_stride + 1]);
687 if (mb_x == 0 || IS_INTRA(cur_frame->mb_type[xy - 1])) {
688 AV_COPY32(mv_cache[1][0], mv_cache[1][1]);
689 AV_COPY32(mv_cache[2][0], mv_cache[2][1]);
691 AV_COPY32(mv_cache[1][0], cur_frame->motion_val[0][mot_xy - 1]);
692 AV_COPY32(mv_cache[2][0],
693 cur_frame->motion_val[0][mot_xy - 1 + mot_stride]);
696 if (mb_x + 1 >= s->mb_width || IS_INTRA(cur_frame->mb_type[xy + 1])) {
697 AV_COPY32(mv_cache[1][3], mv_cache[1][2]);
698 AV_COPY32(mv_cache[2][3], mv_cache[2][2]);
700 AV_COPY32(mv_cache[1][3], cur_frame->motion_val[0][mot_xy + 2]);
701 AV_COPY32(mv_cache[2][3],
702 cur_frame->motion_val[0][mot_xy + 2 + mot_stride]);
707 for (i = 0; i < 4; i++) {
708 const int x = (i & 1) + 1;
709 const int y = (i >> 1) + 1;
711 { mv_cache[y][x][0], mv_cache[y][x][1] },
712 { mv_cache[y - 1][x][0], mv_cache[y - 1][x][1] },
713 { mv_cache[y][x - 1][0], mv_cache[y][x - 1][1] },
714 { mv_cache[y][x + 1][0], mv_cache[y][x + 1][1] },
715 { mv_cache[y + 1][x][0], mv_cache[y + 1][x][1] }
718 obmc_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
720 mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >> 1) * 8,
727 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY))
728 chroma_4mv_motion(s, dest_cb, dest_cr,
729 ref_picture, pix_op[1],
733 static inline void apply_8x8(MpegEncContext *s,
738 uint8_t **ref_picture,
739 qpel_mc_func (*qpix_op)[16],
740 op_pixels_func (*pix_op)[4])
742 int dxy, mx, my, src_x, src_y;
750 if (s->quarter_sample) {
751 for (i = 0; i < 4; i++) {
752 int motion_x = s->mv[dir][i][0];
753 int motion_y = s->mv[dir][i][1];
755 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
756 src_x = mb_x * 16 + (motion_x >> 2) + (i & 1) * 8;
757 src_y = mb_y * 16 + (motion_y >> 2) + (i >> 1) * 8;
759 /* WARNING: do no forget half pels */
760 src_x = av_clip(src_x, -16, s->width);
761 if (src_x == s->width)
763 src_y = av_clip(src_y, -16, s->height);
764 if (src_y == s->height)
767 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x);
768 if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 3) - 8, 0) ||
769 (unsigned)src_y > FFMAX(s->v_edge_pos - (motion_y & 3) - 8, 0)) {
770 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
771 s->linesize, s->linesize,
776 ptr = s->sc.edge_emu_buffer;
778 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize;
779 qpix_op[1][dxy](dest, ptr, s->linesize);
781 mx += s->mv[dir][i][0] / 2;
782 my += s->mv[dir][i][1] / 2;
785 for (i = 0; i < 4; i++) {
787 dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
789 mb_x * 16 + (i & 1) * 8,
790 mb_y * 16 + (i >> 1) * 8,
795 mx += s->mv[dir][i][0];
796 my += s->mv[dir][i][1];
800 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY))
801 chroma_4mv_motion(s, dest_cb, dest_cr,
802 ref_picture, pix_op[1], mx, my);
806 * motion compensation of a single macroblock
808 * @param dest_y luma destination pointer
809 * @param dest_cb chroma cb/u destination pointer
810 * @param dest_cr chroma cr/v destination pointer
811 * @param dir direction (0->forward, 1->backward)
812 * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
813 * @param pix_op halfpel motion compensation function (average or put normally)
814 * @param qpix_op qpel motion compensation function (average or put normally)
815 * the motion vectors are taken from s->mv and the MV type from s->mv_type
817 static av_always_inline void mpv_motion_internal(MpegEncContext *s,
822 uint8_t **ref_picture,
823 op_pixels_func (*pix_op)[4],
824 qpel_mc_func (*qpix_op)[16],
830 prefetch_motion(s, ref_picture, dir);
832 if (!is_mpeg12 && s->obmc && s->pict_type != AV_PICTURE_TYPE_B) {
833 apply_obmc(s, dest_y, dest_cb, dest_cr, ref_picture, pix_op);
837 switch (s->mv_type) {
840 if (s->real_sprite_warping_points == 1) {
841 gmc1_motion(s, dest_y, dest_cb, dest_cr,
844 gmc_motion(s, dest_y, dest_cb, dest_cr,
847 } else if (!is_mpeg12 && s->quarter_sample) {
848 qpel_motion(s, dest_y, dest_cb, dest_cr,
850 ref_picture, pix_op, qpix_op,
851 s->mv[dir][0][0], s->mv[dir][0][1], 16);
852 } else if (!is_mpeg12 && (CONFIG_WMV2_DECODER || CONFIG_WMV2_ENCODER) &&
853 s->mspel && s->codec_id == AV_CODEC_ID_WMV2) {
854 ff_mspel_motion(s, dest_y, dest_cb, dest_cr,
856 s->mv[dir][0][0], s->mv[dir][0][1], 16);
858 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
860 s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y);
865 apply_8x8(s, dest_y, dest_cb, dest_cr,
866 dir, ref_picture, qpix_op, pix_op);
869 if (s->picture_structure == PICT_FRAME) {
870 if (!is_mpeg12 && s->quarter_sample) {
871 for (i = 0; i < 2; i++)
872 qpel_motion(s, dest_y, dest_cb, dest_cr,
873 1, i, s->field_select[dir][i],
874 ref_picture, pix_op, qpix_op,
875 s->mv[dir][i][0], s->mv[dir][i][1], 8);
878 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
879 0, s->field_select[dir][0],
881 s->mv[dir][0][0], s->mv[dir][0][1], 8, mb_y);
883 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
884 1, s->field_select[dir][1],
886 s->mv[dir][1][0], s->mv[dir][1][1], 8, mb_y);
889 if (s->picture_structure != s->field_select[dir][0] + 1 &&
890 s->pict_type != AV_PICTURE_TYPE_B && !s->first_field) {
891 ref_picture = s->current_picture_ptr->f->data;
894 mpeg_motion(s, dest_y, dest_cb, dest_cr,
895 s->field_select[dir][0],
897 s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y >> 1);
901 for (i = 0; i < 2; i++) {
902 uint8_t **ref2picture;
904 if (s->picture_structure == s->field_select[dir][i] + 1
905 || s->pict_type == AV_PICTURE_TYPE_B || s->first_field) {
906 ref2picture = ref_picture;
908 ref2picture = s->current_picture_ptr->f->data;
911 mpeg_motion(s, dest_y, dest_cb, dest_cr,
912 s->field_select[dir][i],
914 s->mv[dir][i][0], s->mv[dir][i][1] + 16 * i,
917 dest_y += 16 * s->linesize;
918 dest_cb += (16 >> s->chroma_y_shift) * s->uvlinesize;
919 dest_cr += (16 >> s->chroma_y_shift) * s->uvlinesize;
923 if (s->picture_structure == PICT_FRAME) {
924 for (i = 0; i < 2; i++) {
926 for (j = 0; j < 2; j++)
927 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
928 j, j ^ i, ref_picture, pix_op,
929 s->mv[dir][2 * i + j][0],
930 s->mv[dir][2 * i + j][1], 8, mb_y);
931 pix_op = s->hdsp.avg_pixels_tab;
934 for (i = 0; i < 2; i++) {
935 mpeg_motion(s, dest_y, dest_cb, dest_cr,
936 s->picture_structure != i + 1,
938 s->mv[dir][2 * i][0], s->mv[dir][2 * i][1],
941 // after put we make avg of the same block
942 pix_op = s->hdsp.avg_pixels_tab;
944 /* opposite parity is always in the same frame if this is
946 if (!s->first_field) {
947 ref_picture = s->current_picture_ptr->f->data;
956 void ff_mpv_motion(MpegEncContext *s,
957 uint8_t *dest_y, uint8_t *dest_cb,
958 uint8_t *dest_cr, int dir,
959 uint8_t **ref_picture,
960 op_pixels_func (*pix_op)[4],
961 qpel_mc_func (*qpix_op)[16])
964 if (s->out_format == FMT_MPEG1)
965 mpv_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
966 ref_picture, pix_op, qpix_op, 1);
969 mpv_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
970 ref_picture, pix_op, qpix_op, 0);