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"
31 #include "mpegutils.h"
32 #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->edge_emu_buffer, ptr,
70 s->h_edge_pos, s->v_edge_pos);
71 ptr = s->edge_emu_buffer;
74 if ((motion_x | motion_y) & 7) {
75 s->dsp.gmc1(dest_y, ptr, linesize, 16,
76 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
77 s->dsp.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->flags & 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->edge_emu_buffer, ptr,
111 uvlinesize, uvlinesize,
114 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
115 ptr = s->edge_emu_buffer;
118 s->dsp.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->edge_emu_buffer, ptr,
124 uvlinesize, uvlinesize,
127 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
128 ptr = s->edge_emu_buffer;
130 s->dsp.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->dsp.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->dsp.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->flags & 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->dsp.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->dsp.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->edge_emu_buffer, src,
216 s->linesize, s->linesize,
219 s->h_edge_pos, s->v_edge_pos);
220 src = s->edge_emu_buffer;
223 pix_op[dxy](dest, src, s->linesize, 8);
227 static av_always_inline
228 void mpeg_motion_internal(MpegEncContext *s,
235 uint8_t **ref_picture,
236 op_pixels_func (*pix_op)[4],
243 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
244 int dxy, uvdxy, mx, my, src_x, src_y,
245 uvsrc_x, uvsrc_y, v_edge_pos;
246 ptrdiff_t uvlinesize, linesize;
249 if (s->quarter_sample) {
255 v_edge_pos = s->v_edge_pos >> field_based;
256 linesize = s->current_picture.f->linesize[0] << field_based;
257 uvlinesize = s->current_picture.f->linesize[1] << field_based;
259 dxy = ((motion_y & 1) << 1) | (motion_x & 1);
260 src_x = s->mb_x * 16 + (motion_x >> 1);
261 src_y = (mb_y << (4 - field_based)) + (motion_y >> 1);
263 if (!is_mpeg12 && s->out_format == FMT_H263) {
264 if ((s->workaround_bugs & FF_BUG_HPEL_CHROMA) && field_based) {
265 mx = (motion_x >> 1) | (motion_x & 1);
267 uvdxy = ((my & 1) << 1) | (mx & 1);
268 uvsrc_x = s->mb_x * 8 + (mx >> 1);
269 uvsrc_y = (mb_y << (3 - field_based)) + (my >> 1);
271 uvdxy = dxy | (motion_y & 2) | ((motion_x & 2) >> 1);
272 uvsrc_x = src_x >> 1;
273 uvsrc_y = src_y >> 1;
275 // Even chroma mv's are full pel in H261
276 } else if (!is_mpeg12 && s->out_format == FMT_H261) {
280 uvsrc_x = s->mb_x * 8 + mx;
281 uvsrc_y = mb_y * 8 + my;
283 if (s->chroma_y_shift) {
286 uvdxy = ((my & 1) << 1) | (mx & 1);
287 uvsrc_x = s->mb_x * 8 + (mx >> 1);
288 uvsrc_y = (mb_y << (3 - field_based)) + (my >> 1);
290 if (s->chroma_x_shift) {
293 uvdxy = ((motion_y & 1) << 1) | (mx & 1);
294 uvsrc_x = s->mb_x * 8 + (mx >> 1);
305 ptr_y = ref_picture[0] + src_y * linesize + src_x;
306 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
307 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
309 if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 1) - 16, 0) ||
310 (unsigned)src_y > FFMAX( v_edge_pos - (motion_y & 1) - h , 0)) {
312 s->codec_id == AV_CODEC_ID_MPEG2VIDEO ||
313 s->codec_id == AV_CODEC_ID_MPEG1VIDEO) {
314 av_log(s->avctx, AV_LOG_DEBUG,
315 "MPEG motion vector out of boundary (%d %d)\n", src_x,
319 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y,
320 s->linesize, s->linesize,
321 17, 17 + field_based,
322 src_x, src_y << field_based,
323 s->h_edge_pos, s->v_edge_pos);
324 ptr_y = s->edge_emu_buffer;
325 if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
326 uint8_t *uvbuf = s->edge_emu_buffer + 18 * s->linesize;
327 s->vdsp.emulated_edge_mc(uvbuf, ptr_cb,
328 s->uvlinesize, s->uvlinesize,
330 uvsrc_x, uvsrc_y << field_based,
331 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
332 s->vdsp.emulated_edge_mc(uvbuf + 16, ptr_cr,
333 s->uvlinesize, s->uvlinesize,
335 uvsrc_x, uvsrc_y << field_based,
336 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
342 /* FIXME use this for field pix too instead of the obnoxious hack which
343 * changes picture.data */
345 dest_y += s->linesize;
346 dest_cb += s->uvlinesize;
347 dest_cr += s->uvlinesize;
351 ptr_y += s->linesize;
352 ptr_cb += s->uvlinesize;
353 ptr_cr += s->uvlinesize;
356 pix_op[0][dxy](dest_y, ptr_y, linesize, h);
358 if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
359 pix_op[s->chroma_x_shift][uvdxy]
360 (dest_cb, ptr_cb, uvlinesize, h >> s->chroma_y_shift);
361 pix_op[s->chroma_x_shift][uvdxy]
362 (dest_cr, ptr_cr, uvlinesize, h >> s->chroma_y_shift);
364 if (!is_mpeg12 && (CONFIG_H261_ENCODER || CONFIG_H261_DECODER) &&
365 s->out_format == FMT_H261) {
366 ff_h261_loop_filter(s);
369 /* apply one mpeg motion vector to the three components */
370 static void mpeg_motion(MpegEncContext *s,
371 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
372 int field_select, uint8_t **ref_picture,
373 op_pixels_func (*pix_op)[4],
374 int motion_x, int motion_y, int h, int mb_y)
377 if (s->out_format == FMT_MPEG1)
378 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 0, 0,
379 field_select, ref_picture, pix_op,
380 motion_x, motion_y, h, 1, mb_y);
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, 0, mb_y);
388 static void mpeg_motion_field(MpegEncContext *s, uint8_t *dest_y,
389 uint8_t *dest_cb, uint8_t *dest_cr,
390 int bottom_field, int field_select,
391 uint8_t **ref_picture,
392 op_pixels_func (*pix_op)[4],
393 int motion_x, int motion_y, int h, int mb_y)
396 if (s->out_format == FMT_MPEG1)
397 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 1,
398 bottom_field, field_select, ref_picture, pix_op,
399 motion_x, motion_y, h, 1, mb_y);
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, 0, mb_y);
407 // FIXME: SIMDify, avg variant, 16x16 version
408 static inline void put_obmc(uint8_t *dst, uint8_t *src[5], int stride)
411 uint8_t *const top = src[1];
412 uint8_t *const left = src[2];
413 uint8_t *const mid = src[0];
414 uint8_t *const right = src[3];
415 uint8_t *const bottom = src[4];
416 #define OBMC_FILTER(x, t, l, m, r, b)\
417 dst[x]= (t*top[x] + l*left[x] + m*mid[x] + r*right[x] + b*bottom[x] + 4)>>3
418 #define OBMC_FILTER4(x, t, l, m, r, b)\
419 OBMC_FILTER(x , t, l, m, r, b);\
420 OBMC_FILTER(x+1 , t, l, m, r, b);\
421 OBMC_FILTER(x +stride, t, l, m, r, b);\
422 OBMC_FILTER(x+1+stride, t, l, m, r, b);
425 OBMC_FILTER (x , 2, 2, 4, 0, 0);
426 OBMC_FILTER (x + 1, 2, 1, 5, 0, 0);
427 OBMC_FILTER4(x + 2, 2, 1, 5, 0, 0);
428 OBMC_FILTER4(x + 4, 2, 0, 5, 1, 0);
429 OBMC_FILTER (x + 6, 2, 0, 5, 1, 0);
430 OBMC_FILTER (x + 7, 2, 0, 4, 2, 0);
432 OBMC_FILTER (x , 1, 2, 5, 0, 0);
433 OBMC_FILTER (x + 1, 1, 2, 5, 0, 0);
434 OBMC_FILTER (x + 6, 1, 0, 5, 2, 0);
435 OBMC_FILTER (x + 7, 1, 0, 5, 2, 0);
437 OBMC_FILTER4(x , 1, 2, 5, 0, 0);
438 OBMC_FILTER4(x + 2, 1, 1, 6, 0, 0);
439 OBMC_FILTER4(x + 4, 1, 0, 6, 1, 0);
440 OBMC_FILTER4(x + 6, 1, 0, 5, 2, 0);
442 OBMC_FILTER4(x , 0, 2, 5, 0, 1);
443 OBMC_FILTER4(x + 2, 0, 1, 6, 0, 1);
444 OBMC_FILTER4(x + 4, 0, 0, 6, 1, 1);
445 OBMC_FILTER4(x + 6, 0, 0, 5, 2, 1);
447 OBMC_FILTER (x , 0, 2, 5, 0, 1);
448 OBMC_FILTER (x + 1, 0, 2, 5, 0, 1);
449 OBMC_FILTER4(x + 2, 0, 1, 5, 0, 2);
450 OBMC_FILTER4(x + 4, 0, 0, 5, 1, 2);
451 OBMC_FILTER (x + 6, 0, 0, 5, 2, 1);
452 OBMC_FILTER (x + 7, 0, 0, 5, 2, 1);
454 OBMC_FILTER (x , 0, 2, 4, 0, 2);
455 OBMC_FILTER (x + 1, 0, 1, 5, 0, 2);
456 OBMC_FILTER (x + 6, 0, 0, 5, 1, 2);
457 OBMC_FILTER (x + 7, 0, 0, 4, 2, 2);
460 /* obmc for 1 8x8 luma block */
461 static inline void obmc_motion(MpegEncContext *s,
462 uint8_t *dest, uint8_t *src,
463 int src_x, int src_y,
464 op_pixels_func *pix_op,
465 int16_t mv[5][2] /* mid top left right bottom */)
471 av_assert2(s->quarter_sample == 0);
473 for (i = 0; i < 5; i++) {
474 if (i && mv[i][0] == mv[MID][0] && mv[i][1] == mv[MID][1]) {
477 ptr[i] = s->obmc_scratchpad + 8 * (i & 1) +
478 s->linesize * 8 * (i >> 1);
479 hpel_motion(s, ptr[i], src, src_x, src_y, pix_op,
484 put_obmc(dest, ptr, s->linesize);
487 static inline void qpel_motion(MpegEncContext *s,
491 int field_based, int bottom_field,
492 int field_select, uint8_t **ref_picture,
493 op_pixels_func (*pix_op)[4],
494 qpel_mc_func (*qpix_op)[16],
495 int motion_x, int motion_y, int h)
497 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
498 int dxy, uvdxy, mx, my, src_x, src_y, uvsrc_x, uvsrc_y, v_edge_pos;
499 ptrdiff_t linesize, uvlinesize;
501 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
503 src_x = s->mb_x * 16 + (motion_x >> 2);
504 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2);
506 v_edge_pos = s->v_edge_pos >> field_based;
507 linesize = s->linesize << field_based;
508 uvlinesize = s->uvlinesize << field_based;
513 } else if (s->workaround_bugs & FF_BUG_QPEL_CHROMA2) {
514 static const int rtab[8] = { 0, 0, 1, 1, 0, 0, 0, 1 };
515 mx = (motion_x >> 1) + rtab[motion_x & 7];
516 my = (motion_y >> 1) + rtab[motion_y & 7];
517 } else if (s->workaround_bugs & FF_BUG_QPEL_CHROMA) {
518 mx = (motion_x >> 1) | (motion_x & 1);
519 my = (motion_y >> 1) | (motion_y & 1);
524 mx = (mx >> 1) | (mx & 1);
525 my = (my >> 1) | (my & 1);
527 uvdxy = (mx & 1) | ((my & 1) << 1);
531 uvsrc_x = s->mb_x * 8 + mx;
532 uvsrc_y = s->mb_y * (8 >> field_based) + my;
534 ptr_y = ref_picture[0] + src_y * linesize + src_x;
535 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
536 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
538 if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 3) - 16, 0) ||
539 (unsigned)src_y > FFMAX( v_edge_pos - (motion_y & 3) - h, 0)) {
540 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y,
541 s->linesize, s->linesize,
542 17, 17 + field_based,
543 src_x, src_y << field_based,
544 s->h_edge_pos, s->v_edge_pos);
545 ptr_y = s->edge_emu_buffer;
546 if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
547 uint8_t *uvbuf = s->edge_emu_buffer + 18 * s->linesize;
548 s->vdsp.emulated_edge_mc(uvbuf, ptr_cb,
549 s->uvlinesize, s->uvlinesize,
551 uvsrc_x, uvsrc_y << field_based,
552 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
553 s->vdsp.emulated_edge_mc(uvbuf + 16, ptr_cr,
554 s->uvlinesize, s->uvlinesize,
556 uvsrc_x, uvsrc_y << field_based,
557 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
564 qpix_op[0][dxy](dest_y, ptr_y, linesize);
567 dest_y += s->linesize;
568 dest_cb += s->uvlinesize;
569 dest_cr += s->uvlinesize;
573 ptr_y += s->linesize;
574 ptr_cb += s->uvlinesize;
575 ptr_cr += s->uvlinesize;
577 // damn interlaced mode
578 // FIXME boundary mirroring is not exactly correct here
579 qpix_op[1][dxy](dest_y, ptr_y, linesize);
580 qpix_op[1][dxy](dest_y + 8, ptr_y + 8, linesize);
582 if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
583 pix_op[1][uvdxy](dest_cr, ptr_cr, uvlinesize, h >> 1);
584 pix_op[1][uvdxy](dest_cb, ptr_cb, uvlinesize, h >> 1);
589 * h263 chroma 4mv motion compensation.
591 static void chroma_4mv_motion(MpegEncContext *s,
592 uint8_t *dest_cb, uint8_t *dest_cr,
593 uint8_t **ref_picture,
594 op_pixels_func *pix_op,
598 int src_x, src_y, dxy, emu = 0;
601 /* In case of 8X8, we construct a single chroma motion vector
602 * with a special rounding */
603 mx = ff_h263_round_chroma(mx);
604 my = ff_h263_round_chroma(my);
606 dxy = ((my & 1) << 1) | (mx & 1);
610 src_x = s->mb_x * 8 + mx;
611 src_y = s->mb_y * 8 + my;
612 src_x = av_clip(src_x, -8, (s->width >> 1));
613 if (src_x == (s->width >> 1))
615 src_y = av_clip(src_y, -8, (s->height >> 1));
616 if (src_y == (s->height >> 1))
619 offset = src_y * s->uvlinesize + src_x;
620 ptr = ref_picture[1] + offset;
621 if ((unsigned)src_x > FFMAX((s->h_edge_pos >> 1) - (dxy & 1) - 8, 0) ||
622 (unsigned)src_y > FFMAX((s->v_edge_pos >> 1) - (dxy >> 1) - 8, 0)) {
623 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
624 s->uvlinesize, s->uvlinesize,
626 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
627 ptr = s->edge_emu_buffer;
630 pix_op[dxy](dest_cb, ptr, s->uvlinesize, 8);
632 ptr = ref_picture[2] + offset;
634 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
635 s->uvlinesize, s->uvlinesize,
637 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
638 ptr = s->edge_emu_buffer;
640 pix_op[dxy](dest_cr, ptr, s->uvlinesize, 8);
643 static inline void prefetch_motion(MpegEncContext *s, uint8_t **pix, int dir)
645 /* fetch pixels for estimated mv 4 macroblocks ahead
646 * optimized for 64byte cache lines */
647 const int shift = s->quarter_sample ? 2 : 1;
648 const int mx = (s->mv[dir][0][0] >> shift) + 16 * s->mb_x + 8;
649 const int my = (s->mv[dir][0][1] >> shift) + 16 * s->mb_y;
650 int off = mx + (my + (s->mb_x & 3) * 4) * s->linesize + 64;
652 s->vdsp.prefetch(pix[0] + off, s->linesize, 4);
653 off = (mx >> 1) + ((my >> 1) + (s->mb_x & 7)) * s->uvlinesize + 64;
654 s->vdsp.prefetch(pix[1] + off, pix[2] - pix[1], 2);
657 static inline void apply_obmc(MpegEncContext *s,
661 uint8_t **ref_picture,
662 op_pixels_func (*pix_op)[4])
664 LOCAL_ALIGNED_8(int16_t, mv_cache, [4], [4][2]);
665 Picture *cur_frame = &s->current_picture;
668 const int xy = mb_x + mb_y * s->mb_stride;
669 const int mot_stride = s->b8_stride;
670 const int mot_xy = mb_x * 2 + mb_y * 2 * mot_stride;
673 av_assert2(!s->mb_skipped);
675 AV_COPY32(mv_cache[1][1], cur_frame->motion_val[0][mot_xy]);
676 AV_COPY32(mv_cache[1][2], cur_frame->motion_val[0][mot_xy + 1]);
678 AV_COPY32(mv_cache[2][1],
679 cur_frame->motion_val[0][mot_xy + mot_stride]);
680 AV_COPY32(mv_cache[2][2],
681 cur_frame->motion_val[0][mot_xy + mot_stride + 1]);
683 AV_COPY32(mv_cache[3][1],
684 cur_frame->motion_val[0][mot_xy + mot_stride]);
685 AV_COPY32(mv_cache[3][2],
686 cur_frame->motion_val[0][mot_xy + mot_stride + 1]);
688 if (mb_y == 0 || IS_INTRA(cur_frame->mb_type[xy - s->mb_stride])) {
689 AV_COPY32(mv_cache[0][1], mv_cache[1][1]);
690 AV_COPY32(mv_cache[0][2], mv_cache[1][2]);
692 AV_COPY32(mv_cache[0][1],
693 cur_frame->motion_val[0][mot_xy - mot_stride]);
694 AV_COPY32(mv_cache[0][2],
695 cur_frame->motion_val[0][mot_xy - mot_stride + 1]);
698 if (mb_x == 0 || IS_INTRA(cur_frame->mb_type[xy - 1])) {
699 AV_COPY32(mv_cache[1][0], mv_cache[1][1]);
700 AV_COPY32(mv_cache[2][0], mv_cache[2][1]);
702 AV_COPY32(mv_cache[1][0], cur_frame->motion_val[0][mot_xy - 1]);
703 AV_COPY32(mv_cache[2][0],
704 cur_frame->motion_val[0][mot_xy - 1 + mot_stride]);
707 if (mb_x + 1 >= s->mb_width || IS_INTRA(cur_frame->mb_type[xy + 1])) {
708 AV_COPY32(mv_cache[1][3], mv_cache[1][2]);
709 AV_COPY32(mv_cache[2][3], mv_cache[2][2]);
711 AV_COPY32(mv_cache[1][3], cur_frame->motion_val[0][mot_xy + 2]);
712 AV_COPY32(mv_cache[2][3],
713 cur_frame->motion_val[0][mot_xy + 2 + mot_stride]);
718 for (i = 0; i < 4; i++) {
719 const int x = (i & 1) + 1;
720 const int y = (i >> 1) + 1;
722 { mv_cache[y][x][0], mv_cache[y][x][1] },
723 { mv_cache[y - 1][x][0], mv_cache[y - 1][x][1] },
724 { mv_cache[y][x - 1][0], mv_cache[y][x - 1][1] },
725 { mv_cache[y][x + 1][0], mv_cache[y][x + 1][1] },
726 { mv_cache[y + 1][x][0], mv_cache[y + 1][x][1] }
729 obmc_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
731 mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >> 1) * 8,
738 if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY))
739 chroma_4mv_motion(s, dest_cb, dest_cr,
740 ref_picture, pix_op[1],
744 static inline void apply_8x8(MpegEncContext *s,
749 uint8_t **ref_picture,
750 qpel_mc_func (*qpix_op)[16],
751 op_pixels_func (*pix_op)[4])
753 int dxy, mx, my, src_x, src_y;
761 if (s->quarter_sample) {
762 for (i = 0; i < 4; i++) {
763 int motion_x = s->mv[dir][i][0];
764 int motion_y = s->mv[dir][i][1];
766 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
767 src_x = mb_x * 16 + (motion_x >> 2) + (i & 1) * 8;
768 src_y = mb_y * 16 + (motion_y >> 2) + (i >> 1) * 8;
770 /* WARNING: do no forget half pels */
771 src_x = av_clip(src_x, -16, s->width);
772 if (src_x == s->width)
774 src_y = av_clip(src_y, -16, s->height);
775 if (src_y == s->height)
778 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x);
779 if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 3) - 8, 0) ||
780 (unsigned)src_y > FFMAX(s->v_edge_pos - (motion_y & 3) - 8, 0)) {
781 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
782 s->linesize, s->linesize,
787 ptr = s->edge_emu_buffer;
789 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize;
790 qpix_op[1][dxy](dest, ptr, s->linesize);
792 mx += s->mv[dir][i][0] / 2;
793 my += s->mv[dir][i][1] / 2;
796 for (i = 0; i < 4; i++) {
798 dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
800 mb_x * 16 + (i & 1) * 8,
801 mb_y * 16 + (i >> 1) * 8,
806 mx += s->mv[dir][i][0];
807 my += s->mv[dir][i][1];
811 if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY))
812 chroma_4mv_motion(s, dest_cb, dest_cr,
813 ref_picture, pix_op[1], mx, my);
817 * motion compensation of a single macroblock
819 * @param dest_y luma destination pointer
820 * @param dest_cb chroma cb/u destination pointer
821 * @param dest_cr chroma cr/v destination pointer
822 * @param dir direction (0->forward, 1->backward)
823 * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
824 * @param pix_op halfpel motion compensation function (average or put normally)
825 * @param qpix_op qpel motion compensation function (average or put normally)
826 * the motion vectors are taken from s->mv and the MV type from s->mv_type
828 static av_always_inline void MPV_motion_internal(MpegEncContext *s,
833 uint8_t **ref_picture,
834 op_pixels_func (*pix_op)[4],
835 qpel_mc_func (*qpix_op)[16],
841 prefetch_motion(s, ref_picture, dir);
843 if (!is_mpeg12 && s->obmc && s->pict_type != AV_PICTURE_TYPE_B) {
844 apply_obmc(s, dest_y, dest_cb, dest_cr, ref_picture, pix_op);
848 switch (s->mv_type) {
851 if (s->real_sprite_warping_points == 1) {
852 gmc1_motion(s, dest_y, dest_cb, dest_cr,
855 gmc_motion(s, dest_y, dest_cb, dest_cr,
858 } else if (!is_mpeg12 && s->quarter_sample) {
859 qpel_motion(s, dest_y, dest_cb, dest_cr,
861 ref_picture, pix_op, qpix_op,
862 s->mv[dir][0][0], s->mv[dir][0][1], 16);
863 } else if (!is_mpeg12 && (CONFIG_WMV2_DECODER || CONFIG_WMV2_ENCODER) &&
864 s->mspel && s->codec_id == AV_CODEC_ID_WMV2) {
865 ff_mspel_motion(s, dest_y, dest_cb, dest_cr,
867 s->mv[dir][0][0], s->mv[dir][0][1], 16);
869 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
871 s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y);
876 apply_8x8(s, dest_y, dest_cb, dest_cr,
877 dir, ref_picture, qpix_op, pix_op);
880 if (s->picture_structure == PICT_FRAME) {
881 if (!is_mpeg12 && s->quarter_sample) {
882 for (i = 0; i < 2; i++)
883 qpel_motion(s, dest_y, dest_cb, dest_cr,
884 1, i, s->field_select[dir][i],
885 ref_picture, pix_op, qpix_op,
886 s->mv[dir][i][0], s->mv[dir][i][1], 8);
889 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
890 0, s->field_select[dir][0],
892 s->mv[dir][0][0], s->mv[dir][0][1], 8, mb_y);
894 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
895 1, s->field_select[dir][1],
897 s->mv[dir][1][0], s->mv[dir][1][1], 8, mb_y);
900 if ( s->picture_structure != s->field_select[dir][0] + 1 && s->pict_type != AV_PICTURE_TYPE_B && !s->first_field
901 || !ref_picture[0]) {
902 ref_picture = s->current_picture_ptr->f->data;
905 mpeg_motion(s, dest_y, dest_cb, dest_cr,
906 s->field_select[dir][0],
908 s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y >> 1);
912 for (i = 0; i < 2; i++) {
913 uint8_t **ref2picture;
915 if ((s->picture_structure == s->field_select[dir][i] + 1
916 || s->pict_type == AV_PICTURE_TYPE_B || s->first_field) && ref_picture[0]) {
917 ref2picture = ref_picture;
919 ref2picture = s->current_picture_ptr->f->data;
922 mpeg_motion(s, dest_y, dest_cb, dest_cr,
923 s->field_select[dir][i],
925 s->mv[dir][i][0], s->mv[dir][i][1] + 16 * i,
928 dest_y += 16 * s->linesize;
929 dest_cb += (16 >> s->chroma_y_shift) * s->uvlinesize;
930 dest_cr += (16 >> s->chroma_y_shift) * s->uvlinesize;
934 if (s->picture_structure == PICT_FRAME) {
935 for (i = 0; i < 2; i++) {
937 for (j = 0; j < 2; j++)
938 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
939 j, j ^ i, ref_picture, pix_op,
940 s->mv[dir][2 * i + j][0],
941 s->mv[dir][2 * i + j][1], 8, mb_y);
942 pix_op = s->hdsp.avg_pixels_tab;
945 if (!ref_picture[0]) {
946 ref_picture = s->current_picture_ptr->f->data;
948 for (i = 0; i < 2; i++) {
949 mpeg_motion(s, dest_y, dest_cb, dest_cr,
950 s->picture_structure != i + 1,
952 s->mv[dir][2 * i][0], s->mv[dir][2 * i][1],
955 // after put we make avg of the same block
956 pix_op = s->hdsp.avg_pixels_tab;
958 /* opposite parity is always in the same frame if this is
960 if (!s->first_field) {
961 ref_picture = s->current_picture_ptr->f->data;
966 default: av_assert2(0);
970 void ff_MPV_motion(MpegEncContext *s,
971 uint8_t *dest_y, uint8_t *dest_cb,
972 uint8_t *dest_cr, int dir,
973 uint8_t **ref_picture,
974 op_pixels_func (*pix_op)[4],
975 qpel_mc_func (*qpix_op)[16])
978 if (s->out_format == FMT_MPEG1)
979 MPV_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
980 ref_picture, pix_op, qpix_op, 1);
983 MPV_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
984 ref_picture, pix_op, qpix_op, 0);