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"
30 #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 <<= (3 - s->sprite_warping_accuracy);
49 motion_y <<= (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 (s->flags & CODEC_FLAG_EMU_EDGE) {
63 if ((unsigned)src_x >= FFMAX(s->h_edge_pos - 17, 0) ||
64 (unsigned)src_y >= FFMAX(s->v_edge_pos - 17, 0)) {
65 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
69 s->h_edge_pos, s->v_edge_pos);
70 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 (s->flags & CODEC_FLAG_EMU_EDGE) {
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->edge_emu_buffer, ptr,
115 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
116 ptr = s->edge_emu_buffer;
120 s->dsp.gmc1(dest_cb, ptr, uvlinesize, 8,
121 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
123 ptr = ref_picture[2] + offset;
125 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
129 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
130 ptr = s->edge_emu_buffer;
132 s->dsp.gmc1(dest_cr, ptr, uvlinesize, 8,
133 motion_x & 15, motion_y & 15, 128 - s->no_rounding);
136 static void gmc_motion(MpegEncContext *s,
137 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
138 uint8_t **ref_picture)
141 int linesize, uvlinesize;
142 const int a = s->sprite_warping_accuracy;
145 linesize = s->linesize;
146 uvlinesize = s->uvlinesize;
148 ptr = ref_picture[0];
150 ox = s->sprite_offset[0][0] + s->sprite_delta[0][0] * s->mb_x * 16 +
151 s->sprite_delta[0][1] * s->mb_y * 16;
152 oy = s->sprite_offset[0][1] + s->sprite_delta[1][0] * s->mb_x * 16 +
153 s->sprite_delta[1][1] * s->mb_y * 16;
155 s->dsp.gmc(dest_y, ptr, linesize, 16,
157 s->sprite_delta[0][0], s->sprite_delta[0][1],
158 s->sprite_delta[1][0], s->sprite_delta[1][1],
159 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
160 s->h_edge_pos, s->v_edge_pos);
161 s->dsp.gmc(dest_y + 8, ptr, linesize, 16,
162 ox + s->sprite_delta[0][0] * 8,
163 oy + s->sprite_delta[1][0] * 8,
164 s->sprite_delta[0][0], s->sprite_delta[0][1],
165 s->sprite_delta[1][0], s->sprite_delta[1][1],
166 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
167 s->h_edge_pos, s->v_edge_pos);
169 if (CONFIG_GRAY && s->flags & CODEC_FLAG_GRAY)
172 ox = s->sprite_offset[1][0] + s->sprite_delta[0][0] * s->mb_x * 8 +
173 s->sprite_delta[0][1] * s->mb_y * 8;
174 oy = s->sprite_offset[1][1] + s->sprite_delta[1][0] * s->mb_x * 8 +
175 s->sprite_delta[1][1] * s->mb_y * 8;
177 ptr = ref_picture[1];
178 s->dsp.gmc(dest_cb, ptr, uvlinesize, 8,
180 s->sprite_delta[0][0], s->sprite_delta[0][1],
181 s->sprite_delta[1][0], s->sprite_delta[1][1],
182 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
183 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
185 ptr = ref_picture[2];
186 s->dsp.gmc(dest_cr, ptr, uvlinesize, 8,
188 s->sprite_delta[0][0], s->sprite_delta[0][1],
189 s->sprite_delta[1][0], s->sprite_delta[1][1],
190 a + 1, (1 << (2 * a + 1)) - s->no_rounding,
191 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
194 static inline int hpel_motion(MpegEncContext *s,
195 uint8_t *dest, uint8_t *src,
196 int src_x, int src_y,
197 op_pixels_func *pix_op,
198 int motion_x, int motion_y)
203 src_x += motion_x >> 1;
204 src_y += motion_y >> 1;
206 /* WARNING: do no forget half pels */
207 src_x = av_clip(src_x, -16, s->width); // FIXME unneeded for emu?
208 if (src_x != s->width)
210 src_y = av_clip(src_y, -16, s->height);
211 if (src_y != s->height)
212 dxy |= (motion_y & 1) << 1;
213 src += src_y * s->linesize + src_x;
215 if (s->unrestricted_mv && (s->flags & CODEC_FLAG_EMU_EDGE)) {
216 if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 1) - 8, 0) ||
217 (unsigned)src_y > FFMAX(s->v_edge_pos - (motion_y & 1) - 8, 0)) {
218 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, src,
221 src_x, src_y, s->h_edge_pos,
223 src = s->edge_emu_buffer;
227 pix_op[dxy](dest, src, s->linesize, 8);
231 static av_always_inline
232 void mpeg_motion_internal(MpegEncContext *s,
239 uint8_t **ref_picture,
240 op_pixels_func (*pix_op)[4],
247 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
248 int dxy, uvdxy, mx, my, src_x, src_y,
249 uvsrc_x, uvsrc_y, v_edge_pos;
250 ptrdiff_t uvlinesize, linesize;
253 if (s->quarter_sample) {
259 v_edge_pos = s->v_edge_pos >> field_based;
260 linesize = s->current_picture.f.linesize[0] << field_based;
261 uvlinesize = s->current_picture.f.linesize[1] << field_based;
263 dxy = ((motion_y & 1) << 1) | (motion_x & 1);
264 src_x = s->mb_x * 16 + (motion_x >> 1);
265 src_y = (mb_y << (4 - field_based)) + (motion_y >> 1);
267 if (!is_mpeg12 && s->out_format == FMT_H263) {
268 if ((s->workaround_bugs & FF_BUG_HPEL_CHROMA) && field_based) {
269 mx = (motion_x >> 1) | (motion_x & 1);
271 uvdxy = ((my & 1) << 1) | (mx & 1);
272 uvsrc_x = s->mb_x * 8 + (mx >> 1);
273 uvsrc_y = (mb_y << (3 - field_based)) + (my >> 1);
275 uvdxy = dxy | (motion_y & 2) | ((motion_x & 2) >> 1);
276 uvsrc_x = src_x >> 1;
277 uvsrc_y = src_y >> 1;
279 // Even chroma mv's are full pel in H261
280 } else if (!is_mpeg12 && s->out_format == FMT_H261) {
284 uvsrc_x = s->mb_x * 8 + mx;
285 uvsrc_y = mb_y * 8 + my;
287 if (s->chroma_y_shift) {
290 uvdxy = ((my & 1) << 1) | (mx & 1);
291 uvsrc_x = s->mb_x * 8 + (mx >> 1);
292 uvsrc_y = (mb_y << (3 - field_based)) + (my >> 1);
294 if (s->chroma_x_shift) {
297 uvdxy = ((motion_y & 1) << 1) | (mx & 1);
298 uvsrc_x = s->mb_x * 8 + (mx >> 1);
309 ptr_y = ref_picture[0] + src_y * linesize + src_x;
310 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
311 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
313 if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 1) - 16, 0) ||
314 (unsigned)src_y > FFMAX(v_edge_pos - (motion_y & 1) - h, 0)) {
316 s->codec_id == AV_CODEC_ID_MPEG2VIDEO ||
317 s->codec_id == AV_CODEC_ID_MPEG1VIDEO) {
318 av_log(s->avctx, AV_LOG_DEBUG,
319 "MPEG motion vector out of boundary (%d %d)\n", src_x,
323 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y,
325 17, 17 + field_based,
326 src_x, src_y << field_based,
327 s->h_edge_pos, s->v_edge_pos);
328 ptr_y = s->edge_emu_buffer;
329 if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
330 uint8_t *uvbuf = s->edge_emu_buffer + 18 * s->linesize;
331 s->vdsp.emulated_edge_mc(uvbuf, ptr_cb,
334 uvsrc_x, uvsrc_y << field_based,
335 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
336 s->vdsp.emulated_edge_mc(uvbuf + 16, ptr_cr,
339 uvsrc_x, uvsrc_y << field_based,
340 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
346 /* FIXME use this for field pix too instead of the obnoxious hack which
347 * changes picture.data */
349 dest_y += s->linesize;
350 dest_cb += s->uvlinesize;
351 dest_cr += s->uvlinesize;
355 ptr_y += s->linesize;
356 ptr_cb += s->uvlinesize;
357 ptr_cr += s->uvlinesize;
360 pix_op[0][dxy](dest_y, ptr_y, linesize, h);
362 if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
363 pix_op[s->chroma_x_shift][uvdxy]
364 (dest_cb, ptr_cb, uvlinesize, h >> s->chroma_y_shift);
365 pix_op[s->chroma_x_shift][uvdxy]
366 (dest_cr, ptr_cr, uvlinesize, h >> s->chroma_y_shift);
368 if (!is_mpeg12 && (CONFIG_H261_ENCODER || CONFIG_H261_DECODER) &&
369 s->out_format == FMT_H261) {
370 ff_h261_loop_filter(s);
373 /* apply one mpeg motion vector to the three components */
374 static void mpeg_motion(MpegEncContext *s,
375 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
376 int field_select, uint8_t **ref_picture,
377 op_pixels_func (*pix_op)[4],
378 int motion_x, int motion_y, int h, int mb_y)
381 if (s->out_format == FMT_MPEG1)
382 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 0, 0,
383 field_select, ref_picture, pix_op,
384 motion_x, motion_y, h, 1, mb_y);
387 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 0, 0,
388 field_select, ref_picture, pix_op,
389 motion_x, motion_y, h, 0, mb_y);
392 static void mpeg_motion_field(MpegEncContext *s, uint8_t *dest_y,
393 uint8_t *dest_cb, uint8_t *dest_cr,
394 int bottom_field, int field_select,
395 uint8_t **ref_picture,
396 op_pixels_func (*pix_op)[4],
397 int motion_x, int motion_y, int h, int mb_y)
400 if(s->out_format == FMT_MPEG1)
401 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 1,
402 bottom_field, field_select, ref_picture, pix_op,
403 motion_x, motion_y, h, 1, mb_y);
406 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 1,
407 bottom_field, field_select, ref_picture, pix_op,
408 motion_x, motion_y, h, 0, mb_y);
411 // FIXME move to dsputil, avg variant, 16x16 version
412 static inline void put_obmc(uint8_t *dst, uint8_t *src[5], int stride)
415 uint8_t *const top = src[1];
416 uint8_t *const left = src[2];
417 uint8_t *const mid = src[0];
418 uint8_t *const right = src[3];
419 uint8_t *const bottom = src[4];
420 #define OBMC_FILTER(x, t, l, m, r, b)\
421 dst[x]= (t*top[x] + l*left[x] + m*mid[x] + r*right[x] + b*bottom[x] + 4)>>3
422 #define OBMC_FILTER4(x, t, l, m, r, b)\
423 OBMC_FILTER(x , t, l, m, r, b);\
424 OBMC_FILTER(x+1 , t, l, m, r, b);\
425 OBMC_FILTER(x +stride, t, l, m, r, b);\
426 OBMC_FILTER(x+1+stride, t, l, m, r, b);
429 OBMC_FILTER (x , 2, 2, 4, 0, 0);
430 OBMC_FILTER (x + 1, 2, 1, 5, 0, 0);
431 OBMC_FILTER4(x + 2, 2, 1, 5, 0, 0);
432 OBMC_FILTER4(x + 4, 2, 0, 5, 1, 0);
433 OBMC_FILTER (x + 6, 2, 0, 5, 1, 0);
434 OBMC_FILTER (x + 7, 2, 0, 4, 2, 0);
436 OBMC_FILTER (x , 1, 2, 5, 0, 0);
437 OBMC_FILTER (x + 1, 1, 2, 5, 0, 0);
438 OBMC_FILTER (x + 6, 1, 0, 5, 2, 0);
439 OBMC_FILTER (x + 7, 1, 0, 5, 2, 0);
441 OBMC_FILTER4(x , 1, 2, 5, 0, 0);
442 OBMC_FILTER4(x + 2, 1, 1, 6, 0, 0);
443 OBMC_FILTER4(x + 4, 1, 0, 6, 1, 0);
444 OBMC_FILTER4(x + 6, 1, 0, 5, 2, 0);
446 OBMC_FILTER4(x , 0, 2, 5, 0, 1);
447 OBMC_FILTER4(x + 2, 0, 1, 6, 0, 1);
448 OBMC_FILTER4(x + 4, 0, 0, 6, 1, 1);
449 OBMC_FILTER4(x + 6, 0, 0, 5, 2, 1);
451 OBMC_FILTER (x , 0, 2, 5, 0, 1);
452 OBMC_FILTER (x + 1, 0, 2, 5, 0, 1);
453 OBMC_FILTER4(x + 2, 0, 1, 5, 0, 2);
454 OBMC_FILTER4(x + 4, 0, 0, 5, 1, 2);
455 OBMC_FILTER (x + 6, 0, 0, 5, 2, 1);
456 OBMC_FILTER (x + 7, 0, 0, 5, 2, 1);
458 OBMC_FILTER (x , 0, 2, 4, 0, 2);
459 OBMC_FILTER (x + 1, 0, 1, 5, 0, 2);
460 OBMC_FILTER (x + 6, 0, 0, 5, 1, 2);
461 OBMC_FILTER (x + 7, 0, 0, 4, 2, 2);
464 /* obmc for 1 8x8 luma block */
465 static inline void obmc_motion(MpegEncContext *s,
466 uint8_t *dest, uint8_t *src,
467 int src_x, int src_y,
468 op_pixels_func *pix_op,
469 int16_t mv[5][2] /* mid top left right bottom */)
475 assert(s->quarter_sample == 0);
477 for (i = 0; i < 5; i++) {
478 if (i && mv[i][0] == mv[MID][0] && mv[i][1] == mv[MID][1]) {
481 ptr[i] = s->obmc_scratchpad + 8 * (i & 1) +
482 s->linesize * 8 * (i >> 1);
483 hpel_motion(s, ptr[i], src, src_x, src_y, pix_op,
488 put_obmc(dest, ptr, s->linesize);
491 static inline void qpel_motion(MpegEncContext *s,
495 int field_based, int bottom_field,
496 int field_select, uint8_t **ref_picture,
497 op_pixels_func (*pix_op)[4],
498 qpel_mc_func (*qpix_op)[16],
499 int motion_x, int motion_y, int h)
501 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
502 int dxy, uvdxy, mx, my, src_x, src_y, uvsrc_x, uvsrc_y, v_edge_pos;
503 ptrdiff_t linesize, uvlinesize;
505 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
507 src_x = s->mb_x * 16 + (motion_x >> 2);
508 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2);
510 v_edge_pos = s->v_edge_pos >> field_based;
511 linesize = s->linesize << field_based;
512 uvlinesize = s->uvlinesize << field_based;
517 } else if (s->workaround_bugs & FF_BUG_QPEL_CHROMA2) {
518 static const int rtab[8] = { 0, 0, 1, 1, 0, 0, 0, 1 };
519 mx = (motion_x >> 1) + rtab[motion_x & 7];
520 my = (motion_y >> 1) + rtab[motion_y & 7];
521 } else if (s->workaround_bugs & FF_BUG_QPEL_CHROMA) {
522 mx = (motion_x >> 1) | (motion_x & 1);
523 my = (motion_y >> 1) | (motion_y & 1);
528 mx = (mx >> 1) | (mx & 1);
529 my = (my >> 1) | (my & 1);
531 uvdxy = (mx & 1) | ((my & 1) << 1);
535 uvsrc_x = s->mb_x * 8 + mx;
536 uvsrc_y = s->mb_y * (8 >> field_based) + my;
538 ptr_y = ref_picture[0] + src_y * linesize + src_x;
539 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
540 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
542 if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 3) - 16, 0) ||
543 (unsigned)src_y > FFMAX(v_edge_pos - (motion_y & 3) - h, 0)) {
544 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y,
546 17, 17 + field_based,
547 src_x, src_y << field_based,
548 s->h_edge_pos, s->v_edge_pos);
549 ptr_y = s->edge_emu_buffer;
550 if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
551 uint8_t *uvbuf = s->edge_emu_buffer + 18 * s->linesize;
552 s->vdsp.emulated_edge_mc(uvbuf, ptr_cb,
555 uvsrc_x, uvsrc_y << field_based,
556 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
557 s->vdsp.emulated_edge_mc(uvbuf + 16, ptr_cr,
560 uvsrc_x, uvsrc_y << field_based,
561 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
568 qpix_op[0][dxy](dest_y, ptr_y, linesize);
571 dest_y += s->linesize;
572 dest_cb += s->uvlinesize;
573 dest_cr += s->uvlinesize;
577 ptr_y += s->linesize;
578 ptr_cb += s->uvlinesize;
579 ptr_cr += s->uvlinesize;
581 // damn interlaced mode
582 // FIXME boundary mirroring is not exactly correct here
583 qpix_op[1][dxy](dest_y, ptr_y, linesize);
584 qpix_op[1][dxy](dest_y + 8, ptr_y + 8, linesize);
586 if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
587 pix_op[1][uvdxy](dest_cr, ptr_cr, uvlinesize, h >> 1);
588 pix_op[1][uvdxy](dest_cb, ptr_cb, uvlinesize, h >> 1);
593 * h263 chroma 4mv motion compensation.
595 static void chroma_4mv_motion(MpegEncContext *s,
596 uint8_t *dest_cb, uint8_t *dest_cr,
597 uint8_t **ref_picture,
598 op_pixels_func *pix_op,
602 int src_x, src_y, dxy, emu = 0;
605 /* In case of 8X8, we construct a single chroma motion vector
606 * with a special rounding */
607 mx = ff_h263_round_chroma(mx);
608 my = ff_h263_round_chroma(my);
610 dxy = ((my & 1) << 1) | (mx & 1);
614 src_x = s->mb_x * 8 + mx;
615 src_y = s->mb_y * 8 + my;
616 src_x = av_clip(src_x, -8, (s->width >> 1));
617 if (src_x == (s->width >> 1))
619 src_y = av_clip(src_y, -8, (s->height >> 1));
620 if (src_y == (s->height >> 1))
623 offset = src_y * s->uvlinesize + src_x;
624 ptr = ref_picture[1] + offset;
625 if (s->flags & CODEC_FLAG_EMU_EDGE) {
626 if ((unsigned)src_x > FFMAX((s->h_edge_pos >> 1) - (dxy & 1) - 8, 0) ||
627 (unsigned)src_y > FFMAX((s->v_edge_pos >> 1) - (dxy >> 1) - 8, 0)) {
628 s->vdsp.emulated_edge_mc(s->edge_emu_buffer,
631 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
632 ptr = s->edge_emu_buffer;
636 pix_op[dxy](dest_cb, ptr, s->uvlinesize, 8);
638 ptr = ref_picture[2] + offset;
640 s->vdsp.emulated_edge_mc(s->edge_emu_buffer,
643 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
644 ptr = s->edge_emu_buffer;
646 pix_op[dxy](dest_cr, ptr, s->uvlinesize, 8);
649 static inline void prefetch_motion(MpegEncContext *s, uint8_t **pix, int dir)
651 /* fetch pixels for estimated mv 4 macroblocks ahead
652 * optimized for 64byte cache lines */
653 const int shift = s->quarter_sample ? 2 : 1;
654 const int mx = (s->mv[dir][0][0] >> shift) + 16 * s->mb_x + 8;
655 const int my = (s->mv[dir][0][1] >> shift) + 16 * s->mb_y;
656 int off = mx + (my + (s->mb_x & 3) * 4) * s->linesize + 64;
658 s->vdsp.prefetch(pix[0] + off, s->linesize, 4);
659 off = (mx >> 1) + ((my >> 1) + (s->mb_x & 7)) * s->uvlinesize + 64;
660 s->vdsp.prefetch(pix[1] + off, pix[2] - pix[1], 2);
663 static inline void apply_obmc(MpegEncContext *s,
667 uint8_t **ref_picture,
668 op_pixels_func (*pix_op)[4])
670 LOCAL_ALIGNED_8(int16_t, mv_cache, [4], [4][2]);
671 Picture *cur_frame = &s->current_picture;
674 const int xy = mb_x + mb_y * s->mb_stride;
675 const int mot_stride = s->b8_stride;
676 const int mot_xy = mb_x * 2 + mb_y * 2 * mot_stride;
679 assert(!s->mb_skipped);
681 AV_COPY32(mv_cache[1][1], cur_frame->motion_val[0][mot_xy]);
682 AV_COPY32(mv_cache[1][2], cur_frame->motion_val[0][mot_xy + 1]);
684 AV_COPY32(mv_cache[2][1],
685 cur_frame->motion_val[0][mot_xy + mot_stride]);
686 AV_COPY32(mv_cache[2][2],
687 cur_frame->motion_val[0][mot_xy + mot_stride + 1]);
689 AV_COPY32(mv_cache[3][1],
690 cur_frame->motion_val[0][mot_xy + mot_stride]);
691 AV_COPY32(mv_cache[3][2],
692 cur_frame->motion_val[0][mot_xy + mot_stride + 1]);
694 if (mb_y == 0 || IS_INTRA(cur_frame->mb_type[xy - s->mb_stride])) {
695 AV_COPY32(mv_cache[0][1], mv_cache[1][1]);
696 AV_COPY32(mv_cache[0][2], mv_cache[1][2]);
698 AV_COPY32(mv_cache[0][1],
699 cur_frame->motion_val[0][mot_xy - mot_stride]);
700 AV_COPY32(mv_cache[0][2],
701 cur_frame->motion_val[0][mot_xy - mot_stride + 1]);
704 if (mb_x == 0 || IS_INTRA(cur_frame->mb_type[xy - 1])) {
705 AV_COPY32(mv_cache[1][0], mv_cache[1][1]);
706 AV_COPY32(mv_cache[2][0], mv_cache[2][1]);
708 AV_COPY32(mv_cache[1][0], cur_frame->motion_val[0][mot_xy - 1]);
709 AV_COPY32(mv_cache[2][0],
710 cur_frame->motion_val[0][mot_xy - 1 + mot_stride]);
713 if (mb_x + 1 >= s->mb_width || IS_INTRA(cur_frame->mb_type[xy + 1])) {
714 AV_COPY32(mv_cache[1][3], mv_cache[1][2]);
715 AV_COPY32(mv_cache[2][3], mv_cache[2][2]);
717 AV_COPY32(mv_cache[1][3], cur_frame->motion_val[0][mot_xy + 2]);
718 AV_COPY32(mv_cache[2][3],
719 cur_frame->motion_val[0][mot_xy + 2 + mot_stride]);
724 for (i = 0; i < 4; i++) {
725 const int x = (i & 1) + 1;
726 const int y = (i >> 1) + 1;
728 { mv_cache[y][x][0], mv_cache[y][x][1] },
729 { mv_cache[y - 1][x][0], mv_cache[y - 1][x][1] },
730 { mv_cache[y][x - 1][0], mv_cache[y][x - 1][1] },
731 { mv_cache[y][x + 1][0], mv_cache[y][x + 1][1] },
732 { mv_cache[y + 1][x][0], mv_cache[y + 1][x][1] }
735 obmc_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
737 mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >> 1) * 8,
744 if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY))
745 chroma_4mv_motion(s, dest_cb, dest_cr,
746 ref_picture, pix_op[1],
751 * motion compensation of a single macroblock
753 * @param dest_y luma destination pointer
754 * @param dest_cb chroma cb/u destination pointer
755 * @param dest_cr chroma cr/v destination pointer
756 * @param dir direction (0->forward, 1->backward)
757 * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
758 * @param pix_op halfpel motion compensation function (average or put normally)
759 * @param qpix_op qpel motion compensation function (average or put normally)
760 * the motion vectors are taken from s->mv and the MV type from s->mv_type
762 static av_always_inline void MPV_motion_internal(MpegEncContext *s,
767 uint8_t **ref_picture,
768 op_pixels_func (*pix_op)[4],
769 qpel_mc_func (*qpix_op)[16],
772 int dxy, mx, my, src_x, src_y, motion_x, motion_y;
779 prefetch_motion(s, ref_picture, dir);
781 if (!is_mpeg12 && s->obmc && s->pict_type != AV_PICTURE_TYPE_B) {
782 apply_obmc(s, dest_y, dest_cb, dest_cr, ref_picture, pix_op);
786 switch (s->mv_type) {
789 if (s->real_sprite_warping_points == 1) {
790 gmc1_motion(s, dest_y, dest_cb, dest_cr,
793 gmc_motion(s, dest_y, dest_cb, dest_cr,
796 } else if (!is_mpeg12 && s->quarter_sample) {
797 qpel_motion(s, dest_y, dest_cb, dest_cr,
799 ref_picture, pix_op, qpix_op,
800 s->mv[dir][0][0], s->mv[dir][0][1], 16);
801 } else if (!is_mpeg12 && (CONFIG_WMV2_DECODER || CONFIG_WMV2_ENCODER) &&
802 s->mspel && s->codec_id == AV_CODEC_ID_WMV2) {
803 ff_mspel_motion(s, dest_y, dest_cb, dest_cr,
805 s->mv[dir][0][0], s->mv[dir][0][1], 16);
807 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
809 s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y);
816 if (s->quarter_sample) {
817 for (i = 0; i < 4; i++) {
818 motion_x = s->mv[dir][i][0];
819 motion_y = s->mv[dir][i][1];
821 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
822 src_x = mb_x * 16 + (motion_x >> 2) + (i & 1) * 8;
823 src_y = mb_y * 16 + (motion_y >> 2) + (i >> 1) * 8;
825 /* WARNING: do no forget half pels */
826 src_x = av_clip(src_x, -16, s->width);
827 if (src_x == s->width)
829 src_y = av_clip(src_y, -16, s->height);
830 if (src_y == s->height)
833 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x);
834 if (s->flags & CODEC_FLAG_EMU_EDGE) {
835 if ((unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x & 3) - 8, 0) ||
836 (unsigned)src_y > FFMAX(s->v_edge_pos - (motion_y & 3) - 8, 0)) {
837 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
843 ptr = s->edge_emu_buffer;
846 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize;
847 qpix_op[1][dxy](dest, ptr, s->linesize);
849 mx += s->mv[dir][i][0] / 2;
850 my += s->mv[dir][i][1] / 2;
853 for (i = 0; i < 4; i++) {
855 dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
857 mb_x * 16 + (i & 1) * 8,
858 mb_y * 16 + (i >> 1) * 8,
863 mx += s->mv[dir][i][0];
864 my += s->mv[dir][i][1];
868 if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY))
869 chroma_4mv_motion(s, dest_cb, dest_cr,
870 ref_picture, pix_op[1], mx, my);
874 if (s->picture_structure == PICT_FRAME) {
875 if (!is_mpeg12 && s->quarter_sample) {
876 for (i = 0; i < 2; i++)
877 qpel_motion(s, dest_y, dest_cb, dest_cr,
878 1, i, s->field_select[dir][i],
879 ref_picture, pix_op, qpix_op,
880 s->mv[dir][i][0], s->mv[dir][i][1], 8);
883 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
884 0, s->field_select[dir][0],
886 s->mv[dir][0][0], s->mv[dir][0][1], 8, mb_y);
888 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
889 1, s->field_select[dir][1],
891 s->mv[dir][1][0], s->mv[dir][1][1], 8, mb_y);
894 if (s->picture_structure != s->field_select[dir][0] + 1 &&
895 s->pict_type != AV_PICTURE_TYPE_B && !s->first_field) {
896 ref_picture = s->current_picture_ptr->f.data;
899 mpeg_motion(s, dest_y, dest_cb, dest_cr,
900 s->field_select[dir][0],
902 s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y >> 1);
906 for (i = 0; i < 2; i++) {
907 uint8_t **ref2picture;
909 if (s->picture_structure == s->field_select[dir][i] + 1
910 || s->pict_type == AV_PICTURE_TYPE_B || s->first_field) {
911 ref2picture = ref_picture;
913 ref2picture = s->current_picture_ptr->f.data;
916 mpeg_motion(s, dest_y, dest_cb, dest_cr,
917 s->field_select[dir][i],
919 s->mv[dir][i][0], s->mv[dir][i][1] + 16 * i,
922 dest_y += 16 * s->linesize;
923 dest_cb += (16 >> s->chroma_y_shift) * s->uvlinesize;
924 dest_cr += (16 >> s->chroma_y_shift) * s->uvlinesize;
928 if (s->picture_structure == PICT_FRAME) {
929 for (i = 0; i < 2; i++) {
931 for (j = 0; j < 2; j++)
932 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
933 j, j ^ i, ref_picture, pix_op,
934 s->mv[dir][2 * i + j][0],
935 s->mv[dir][2 * i + j][1], 8, mb_y);
936 pix_op = s->hdsp.avg_pixels_tab;
939 for (i = 0; i < 2; i++) {
940 mpeg_motion(s, dest_y, dest_cb, dest_cr,
941 s->picture_structure != i + 1,
943 s->mv[dir][2 * i][0], s->mv[dir][2 * i][1],
946 // after put we make avg of the same block
947 pix_op = s->hdsp.avg_pixels_tab;
949 /* opposite parity is always in the same frame if this is
951 if (!s->first_field) {
952 ref_picture = s->current_picture_ptr->f.data;
961 void ff_MPV_motion(MpegEncContext *s,
962 uint8_t *dest_y, uint8_t *dest_cb,
963 uint8_t *dest_cr, int dir,
964 uint8_t **ref_picture,
965 op_pixels_func (*pix_op)[4],
966 qpel_mc_func (*qpix_op)[16])
969 if (s->out_format == FMT_MPEG1)
970 MPV_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
971 ref_picture, pix_op, qpix_op, 1);
974 MPV_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
975 ref_picture, pix_op, qpix_op, 0);