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
27 #include "mpegvideo.h"
32 static inline void gmc1_motion(MpegEncContext *s,
33 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
34 uint8_t **ref_picture)
37 int offset, src_x, src_y, linesize, uvlinesize;
38 int motion_x, motion_y;
41 motion_x= s->sprite_offset[0][0];
42 motion_y= s->sprite_offset[0][1];
43 src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy+1));
44 src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy+1));
45 motion_x<<=(3-s->sprite_warping_accuracy);
46 motion_y<<=(3-s->sprite_warping_accuracy);
47 src_x = av_clip(src_x, -16, s->width);
48 if (src_x == s->width)
50 src_y = av_clip(src_y, -16, s->height);
51 if (src_y == s->height)
54 linesize = s->linesize;
55 uvlinesize = s->uvlinesize;
57 ptr = ref_picture[0] + (src_y * linesize) + src_x;
59 if(s->flags&CODEC_FLAG_EMU_EDGE){
60 if( (unsigned)src_x >= FFMAX(s->h_edge_pos - 17, 0)
61 || (unsigned)src_y >= FFMAX(s->v_edge_pos - 17, 0)){
62 s->dsp.emulated_edge_mc(s->edge_emu_buffer, ptr, linesize, 17, 17, src_x, src_y, s->h_edge_pos, s->v_edge_pos);
63 ptr= s->edge_emu_buffer;
67 if((motion_x|motion_y)&7){
68 s->dsp.gmc1(dest_y , ptr , linesize, 16, motion_x&15, motion_y&15, 128 - s->no_rounding);
69 s->dsp.gmc1(dest_y+8, ptr+8, linesize, 16, motion_x&15, motion_y&15, 128 - s->no_rounding);
73 dxy= ((motion_x>>3)&1) | ((motion_y>>2)&2);
75 s->dsp.put_no_rnd_pixels_tab[0][dxy](dest_y, ptr, linesize, 16);
77 s->dsp.put_pixels_tab [0][dxy](dest_y, ptr, linesize, 16);
81 if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
83 motion_x= s->sprite_offset[1][0];
84 motion_y= s->sprite_offset[1][1];
85 src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy+1));
86 src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy+1));
87 motion_x<<=(3-s->sprite_warping_accuracy);
88 motion_y<<=(3-s->sprite_warping_accuracy);
89 src_x = av_clip(src_x, -8, s->width>>1);
90 if (src_x == s->width>>1)
92 src_y = av_clip(src_y, -8, s->height>>1);
93 if (src_y == s->height>>1)
96 offset = (src_y * uvlinesize) + src_x;
97 ptr = ref_picture[1] + offset;
98 if(s->flags&CODEC_FLAG_EMU_EDGE){
99 if( (unsigned)src_x >= FFMAX((s->h_edge_pos>>1) - 9, 0)
100 || (unsigned)src_y >= FFMAX((s->v_edge_pos>>1) - 9, 0)){
101 s->dsp.emulated_edge_mc(s->edge_emu_buffer, ptr, uvlinesize, 9, 9, src_x, src_y, s->h_edge_pos>>1, s->v_edge_pos>>1);
102 ptr= s->edge_emu_buffer;
106 s->dsp.gmc1(dest_cb, ptr, uvlinesize, 8, motion_x&15, motion_y&15, 128 - s->no_rounding);
108 ptr = ref_picture[2] + offset;
110 s->dsp.emulated_edge_mc(s->edge_emu_buffer, ptr, uvlinesize, 9, 9, src_x, src_y, s->h_edge_pos>>1, s->v_edge_pos>>1);
111 ptr= s->edge_emu_buffer;
113 s->dsp.gmc1(dest_cr, ptr, uvlinesize, 8, motion_x&15, motion_y&15, 128 - s->no_rounding);
118 static inline void gmc_motion(MpegEncContext *s,
119 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
120 uint8_t **ref_picture)
123 int linesize, uvlinesize;
124 const int a= s->sprite_warping_accuracy;
127 linesize = s->linesize;
128 uvlinesize = s->uvlinesize;
130 ptr = ref_picture[0];
132 ox= s->sprite_offset[0][0] + s->sprite_delta[0][0]*s->mb_x*16 + s->sprite_delta[0][1]*s->mb_y*16;
133 oy= s->sprite_offset[0][1] + s->sprite_delta[1][0]*s->mb_x*16 + s->sprite_delta[1][1]*s->mb_y*16;
135 s->dsp.gmc(dest_y, ptr, linesize, 16,
138 s->sprite_delta[0][0], s->sprite_delta[0][1],
139 s->sprite_delta[1][0], s->sprite_delta[1][1],
140 a+1, (1<<(2*a+1)) - s->no_rounding,
141 s->h_edge_pos, s->v_edge_pos);
142 s->dsp.gmc(dest_y+8, ptr, linesize, 16,
143 ox + s->sprite_delta[0][0]*8,
144 oy + s->sprite_delta[1][0]*8,
145 s->sprite_delta[0][0], s->sprite_delta[0][1],
146 s->sprite_delta[1][0], s->sprite_delta[1][1],
147 a+1, (1<<(2*a+1)) - s->no_rounding,
148 s->h_edge_pos, s->v_edge_pos);
150 if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
152 ox= s->sprite_offset[1][0] + s->sprite_delta[0][0]*s->mb_x*8 + s->sprite_delta[0][1]*s->mb_y*8;
153 oy= s->sprite_offset[1][1] + s->sprite_delta[1][0]*s->mb_x*8 + s->sprite_delta[1][1]*s->mb_y*8;
155 ptr = ref_picture[1];
156 s->dsp.gmc(dest_cb, ptr, uvlinesize, 8,
159 s->sprite_delta[0][0], s->sprite_delta[0][1],
160 s->sprite_delta[1][0], s->sprite_delta[1][1],
161 a+1, (1<<(2*a+1)) - s->no_rounding,
162 s->h_edge_pos>>1, s->v_edge_pos>>1);
164 ptr = ref_picture[2];
165 s->dsp.gmc(dest_cr, ptr, uvlinesize, 8,
168 s->sprite_delta[0][0], s->sprite_delta[0][1],
169 s->sprite_delta[1][0], s->sprite_delta[1][1],
170 a+1, (1<<(2*a+1)) - s->no_rounding,
171 s->h_edge_pos>>1, s->v_edge_pos>>1);
174 static inline int hpel_motion(MpegEncContext *s,
175 uint8_t *dest, uint8_t *src,
176 int field_based, int field_select,
177 int src_x, int src_y,
178 int width, int height, int stride,
179 int h_edge_pos, int v_edge_pos,
180 int w, int h, op_pixels_func *pix_op,
181 int motion_x, int motion_y)
186 dxy = ((motion_y & 1) << 1) | (motion_x & 1);
187 src_x += motion_x >> 1;
188 src_y += motion_y >> 1;
190 /* WARNING: do no forget half pels */
191 src_x = av_clip(src_x, -16, width); //FIXME unneeded for emu?
194 src_y = av_clip(src_y, -16, height);
197 src += src_y * stride + src_x;
199 if(s->unrestricted_mv && (s->flags&CODEC_FLAG_EMU_EDGE)){
200 if( (unsigned)src_x > FFMAX(h_edge_pos - (motion_x&1) - w, 0)
201 || (unsigned)src_y > FFMAX(v_edge_pos - (motion_y&1) - h, 0)){
202 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src, s->linesize, w+1, (h+1)<<field_based,
203 src_x, src_y<<field_based, h_edge_pos, s->v_edge_pos);
204 src= s->edge_emu_buffer;
210 pix_op[dxy](dest, src, stride, h);
214 static av_always_inline
215 void mpeg_motion_internal(MpegEncContext *s,
216 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
217 int field_based, int bottom_field, int field_select,
218 uint8_t **ref_picture, op_pixels_func (*pix_op)[4],
219 int motion_x, int motion_y, int h, int is_mpeg12, int mb_y)
221 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
222 int dxy, uvdxy, mx, my, src_x, src_y,
223 uvsrc_x, uvsrc_y, v_edge_pos, uvlinesize, linesize;
226 if(s->quarter_sample)
233 v_edge_pos = s->v_edge_pos >> field_based;
234 linesize = s->current_picture.f.linesize[0] << field_based;
235 uvlinesize = s->current_picture.f.linesize[1] << field_based;
237 dxy = ((motion_y & 1) << 1) | (motion_x & 1);
238 src_x = s->mb_x* 16 + (motion_x >> 1);
239 src_y =( mb_y<<(4-field_based)) + (motion_y >> 1);
241 if (!is_mpeg12 && s->out_format == FMT_H263) {
242 if((s->workaround_bugs & FF_BUG_HPEL_CHROMA) && field_based){
243 mx = (motion_x>>1)|(motion_x&1);
245 uvdxy = ((my & 1) << 1) | (mx & 1);
246 uvsrc_x = s->mb_x* 8 + (mx >> 1);
247 uvsrc_y =( mb_y<<(3-field_based))+ (my >> 1);
249 uvdxy = dxy | (motion_y & 2) | ((motion_x & 2) >> 1);
253 }else if(!is_mpeg12 && s->out_format == FMT_H261){//even chroma mv's are full pel in H261
257 uvsrc_x = s->mb_x*8 + mx;
258 uvsrc_y = mb_y*8 + my;
260 if(s->chroma_y_shift){
263 uvdxy = ((my & 1) << 1) | (mx & 1);
264 uvsrc_x = s->mb_x* 8 + (mx >> 1);
265 uvsrc_y =( mb_y<<(3-field_based))+ (my >> 1);
267 if(s->chroma_x_shift){
270 uvdxy = ((motion_y & 1) << 1) | (mx & 1);
271 uvsrc_x = s->mb_x* 8 + (mx >> 1);
282 ptr_y = ref_picture[0] + src_y * linesize + src_x;
283 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
284 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
286 if( (unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x&1) - 16, 0)
287 || (unsigned)src_y > FFMAX( v_edge_pos - (motion_y&1) - h , 0)){
288 if(is_mpeg12 || s->codec_id == AV_CODEC_ID_MPEG2VIDEO ||
289 s->codec_id == AV_CODEC_ID_MPEG1VIDEO){
290 av_log(s->avctx,AV_LOG_DEBUG,
291 "MPEG motion vector out of boundary (%d %d)\n", src_x, src_y);
294 s->dsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y, s->linesize,
296 src_x, src_y<<field_based,
297 s->h_edge_pos, s->v_edge_pos);
298 ptr_y = s->edge_emu_buffer;
299 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
300 uint8_t *uvbuf= s->edge_emu_buffer+18*s->linesize;
301 s->dsp.emulated_edge_mc(uvbuf ,
302 ptr_cb, s->uvlinesize,
304 uvsrc_x, uvsrc_y<<field_based,
305 s->h_edge_pos>>1, s->v_edge_pos>>1);
306 s->dsp.emulated_edge_mc(uvbuf+16,
307 ptr_cr, s->uvlinesize,
309 uvsrc_x, uvsrc_y<<field_based,
310 s->h_edge_pos>>1, s->v_edge_pos>>1);
316 if(bottom_field){ //FIXME use this for field pix too instead of the obnoxious hack which changes picture.data
317 dest_y += s->linesize;
318 dest_cb+= s->uvlinesize;
319 dest_cr+= s->uvlinesize;
323 ptr_y += s->linesize;
324 ptr_cb+= s->uvlinesize;
325 ptr_cr+= s->uvlinesize;
328 pix_op[0][dxy](dest_y, ptr_y, linesize, h);
330 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
331 pix_op[s->chroma_x_shift][uvdxy]
332 (dest_cb, ptr_cb, uvlinesize, h >> s->chroma_y_shift);
333 pix_op[s->chroma_x_shift][uvdxy]
334 (dest_cr, ptr_cr, uvlinesize, h >> s->chroma_y_shift);
336 if(!is_mpeg12 && (CONFIG_H261_ENCODER || CONFIG_H261_DECODER) &&
337 s->out_format == FMT_H261){
338 ff_h261_loop_filter(s);
341 /* apply one mpeg motion vector to the three components */
342 static av_always_inline
343 void mpeg_motion(MpegEncContext *s,
344 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
345 int field_based, int bottom_field, int field_select,
346 uint8_t **ref_picture, op_pixels_func (*pix_op)[4],
347 int motion_x, int motion_y, int h, int mb_y)
350 if(s->out_format == FMT_MPEG1)
351 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, field_based,
352 bottom_field, field_select, ref_picture, pix_op,
353 motion_x, motion_y, h, 1, mb_y);
356 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, field_based,
357 bottom_field, field_select, ref_picture, pix_op,
358 motion_x, motion_y, h, 0, mb_y);
361 //FIXME move to dsputil, avg variant, 16x16 version
362 static inline void put_obmc(uint8_t *dst, uint8_t *src[5], int stride){
364 uint8_t * const top = src[1];
365 uint8_t * const left = src[2];
366 uint8_t * const mid = src[0];
367 uint8_t * const right = src[3];
368 uint8_t * const bottom= src[4];
369 #define OBMC_FILTER(x, t, l, m, r, b)\
370 dst[x]= (t*top[x] + l*left[x] + m*mid[x] + r*right[x] + b*bottom[x] + 4)>>3
371 #define OBMC_FILTER4(x, t, l, m, r, b)\
372 OBMC_FILTER(x , t, l, m, r, b);\
373 OBMC_FILTER(x+1 , t, l, m, r, b);\
374 OBMC_FILTER(x +stride, t, l, m, r, b);\
375 OBMC_FILTER(x+1+stride, t, l, m, r, b);
378 OBMC_FILTER (x , 2, 2, 4, 0, 0);
379 OBMC_FILTER (x+1, 2, 1, 5, 0, 0);
380 OBMC_FILTER4(x+2, 2, 1, 5, 0, 0);
381 OBMC_FILTER4(x+4, 2, 0, 5, 1, 0);
382 OBMC_FILTER (x+6, 2, 0, 5, 1, 0);
383 OBMC_FILTER (x+7, 2, 0, 4, 2, 0);
385 OBMC_FILTER (x , 1, 2, 5, 0, 0);
386 OBMC_FILTER (x+1, 1, 2, 5, 0, 0);
387 OBMC_FILTER (x+6, 1, 0, 5, 2, 0);
388 OBMC_FILTER (x+7, 1, 0, 5, 2, 0);
390 OBMC_FILTER4(x , 1, 2, 5, 0, 0);
391 OBMC_FILTER4(x+2, 1, 1, 6, 0, 0);
392 OBMC_FILTER4(x+4, 1, 0, 6, 1, 0);
393 OBMC_FILTER4(x+6, 1, 0, 5, 2, 0);
395 OBMC_FILTER4(x , 0, 2, 5, 0, 1);
396 OBMC_FILTER4(x+2, 0, 1, 6, 0, 1);
397 OBMC_FILTER4(x+4, 0, 0, 6, 1, 1);
398 OBMC_FILTER4(x+6, 0, 0, 5, 2, 1);
400 OBMC_FILTER (x , 0, 2, 5, 0, 1);
401 OBMC_FILTER (x+1, 0, 2, 5, 0, 1);
402 OBMC_FILTER4(x+2, 0, 1, 5, 0, 2);
403 OBMC_FILTER4(x+4, 0, 0, 5, 1, 2);
404 OBMC_FILTER (x+6, 0, 0, 5, 2, 1);
405 OBMC_FILTER (x+7, 0, 0, 5, 2, 1);
407 OBMC_FILTER (x , 0, 2, 4, 0, 2);
408 OBMC_FILTER (x+1, 0, 1, 5, 0, 2);
409 OBMC_FILTER (x+6, 0, 0, 5, 1, 2);
410 OBMC_FILTER (x+7, 0, 0, 4, 2, 2);
413 /* obmc for 1 8x8 luma block */
414 static inline void obmc_motion(MpegEncContext *s,
415 uint8_t *dest, uint8_t *src,
416 int src_x, int src_y,
417 op_pixels_func *pix_op,
418 int16_t mv[5][2]/* mid top left right bottom*/)
424 assert(s->quarter_sample==0);
427 if(i && mv[i][0]==mv[MID][0] && mv[i][1]==mv[MID][1]){
430 ptr[i]= s->obmc_scratchpad + 8*(i&1) + s->linesize*8*(i>>1);
431 hpel_motion(s, ptr[i], src, 0, 0,
433 s->width, s->height, s->linesize,
434 s->h_edge_pos, s->v_edge_pos,
440 put_obmc(dest, ptr, s->linesize);
443 static inline void qpel_motion(MpegEncContext *s,
444 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
445 int field_based, int bottom_field, int field_select,
446 uint8_t **ref_picture, op_pixels_func (*pix_op)[4],
447 qpel_mc_func (*qpix_op)[16],
448 int motion_x, int motion_y, int h)
450 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
451 int dxy, uvdxy, mx, my, src_x, src_y, uvsrc_x, uvsrc_y, v_edge_pos, linesize, uvlinesize;
453 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
454 src_x = s->mb_x * 16 + (motion_x >> 2);
455 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2);
457 v_edge_pos = s->v_edge_pos >> field_based;
458 linesize = s->linesize << field_based;
459 uvlinesize = s->uvlinesize << field_based;
464 }else if(s->workaround_bugs&FF_BUG_QPEL_CHROMA2){
465 static const int rtab[8]= {0,0,1,1,0,0,0,1};
466 mx= (motion_x>>1) + rtab[motion_x&7];
467 my= (motion_y>>1) + rtab[motion_y&7];
468 }else if(s->workaround_bugs&FF_BUG_QPEL_CHROMA){
469 mx= (motion_x>>1)|(motion_x&1);
470 my= (motion_y>>1)|(motion_y&1);
478 uvdxy= (mx&1) | ((my&1)<<1);
482 uvsrc_x = s->mb_x * 8 + mx;
483 uvsrc_y = s->mb_y * (8 >> field_based) + my;
485 ptr_y = ref_picture[0] + src_y * linesize + src_x;
486 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
487 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
489 if( (unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x&3) - 16, 0)
490 || (unsigned)src_y > FFMAX( v_edge_pos - (motion_y&3) - h , 0)){
491 s->dsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y, s->linesize,
492 17, 17+field_based, src_x, src_y<<field_based,
493 s->h_edge_pos, s->v_edge_pos);
494 ptr_y= s->edge_emu_buffer;
495 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
496 uint8_t *uvbuf= s->edge_emu_buffer + 18*s->linesize;
497 s->dsp.emulated_edge_mc(uvbuf, ptr_cb, s->uvlinesize,
499 uvsrc_x, uvsrc_y<<field_based,
500 s->h_edge_pos>>1, s->v_edge_pos>>1);
501 s->dsp.emulated_edge_mc(uvbuf + 16, ptr_cr, s->uvlinesize,
503 uvsrc_x, uvsrc_y<<field_based,
504 s->h_edge_pos>>1, s->v_edge_pos>>1);
511 qpix_op[0][dxy](dest_y, ptr_y, linesize);
514 dest_y += s->linesize;
515 dest_cb+= s->uvlinesize;
516 dest_cr+= s->uvlinesize;
520 ptr_y += s->linesize;
521 ptr_cb += s->uvlinesize;
522 ptr_cr += s->uvlinesize;
524 //damn interlaced mode
525 //FIXME boundary mirroring is not exactly correct here
526 qpix_op[1][dxy](dest_y , ptr_y , linesize);
527 qpix_op[1][dxy](dest_y+8, ptr_y+8, linesize);
529 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
530 pix_op[1][uvdxy](dest_cr, ptr_cr, uvlinesize, h >> 1);
531 pix_op[1][uvdxy](dest_cb, ptr_cb, uvlinesize, h >> 1);
536 * h263 chroma 4mv motion compensation.
538 static inline void chroma_4mv_motion(MpegEncContext *s,
539 uint8_t *dest_cb, uint8_t *dest_cr,
540 uint8_t **ref_picture,
541 op_pixels_func *pix_op,
543 int dxy, emu=0, src_x, src_y, offset;
546 /* In case of 8X8, we construct a single chroma motion vector
547 with a special rounding */
548 mx= ff_h263_round_chroma(mx);
549 my= ff_h263_round_chroma(my);
551 dxy = ((my & 1) << 1) | (mx & 1);
555 src_x = s->mb_x * 8 + mx;
556 src_y = s->mb_y * 8 + my;
557 src_x = av_clip(src_x, -8, (s->width >> 1));
558 if (src_x == (s->width >> 1))
560 src_y = av_clip(src_y, -8, (s->height >> 1));
561 if (src_y == (s->height >> 1))
564 offset = src_y * s->uvlinesize + src_x;
565 ptr = ref_picture[1] + offset;
566 if(s->flags&CODEC_FLAG_EMU_EDGE){
567 if( (unsigned)src_x > FFMAX((s->h_edge_pos>>1) - (dxy &1) - 8, 0)
568 || (unsigned)src_y > FFMAX((s->v_edge_pos>>1) - (dxy>>1) - 8, 0)){
569 s->dsp.emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize,
571 s->h_edge_pos>>1, s->v_edge_pos>>1);
572 ptr= s->edge_emu_buffer;
576 pix_op[dxy](dest_cb, ptr, s->uvlinesize, 8);
578 ptr = ref_picture[2] + offset;
580 s->dsp.emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize,
582 s->h_edge_pos>>1, s->v_edge_pos>>1);
583 ptr= s->edge_emu_buffer;
585 pix_op[dxy](dest_cr, ptr, s->uvlinesize, 8);
588 static inline void prefetch_motion(MpegEncContext *s, uint8_t **pix, int dir){
589 /* fetch pixels for estimated mv 4 macroblocks ahead
590 * optimized for 64byte cache lines */
591 const int shift = s->quarter_sample ? 2 : 1;
592 const int mx= (s->mv[dir][0][0]>>shift) + 16*s->mb_x + 8;
593 const int my= (s->mv[dir][0][1]>>shift) + 16*s->mb_y;
594 int off= mx + (my + (s->mb_x&3)*4)*s->linesize + 64;
595 s->dsp.prefetch(pix[0]+off, s->linesize, 4);
596 off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
597 s->dsp.prefetch(pix[1]+off, pix[2]-pix[1], 2);
601 * motion compensation of a single macroblock
603 * @param dest_y luma destination pointer
604 * @param dest_cb chroma cb/u destination pointer
605 * @param dest_cr chroma cr/v destination pointer
606 * @param dir direction (0->forward, 1->backward)
607 * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
608 * @param pix_op halfpel motion compensation function (average or put normally)
609 * @param qpix_op qpel motion compensation function (average or put normally)
610 * the motion vectors are taken from s->mv and the MV type from s->mv_type
612 static av_always_inline void MPV_motion_internal(MpegEncContext *s,
613 uint8_t *dest_y, uint8_t *dest_cb,
614 uint8_t *dest_cr, int dir,
615 uint8_t **ref_picture,
616 op_pixels_func (*pix_op)[4],
617 qpel_mc_func (*qpix_op)[16], int is_mpeg12)
619 int dxy, mx, my, src_x, src_y, motion_x, motion_y;
626 prefetch_motion(s, ref_picture, dir);
628 if(!is_mpeg12 && s->obmc && s->pict_type != AV_PICTURE_TYPE_B){
629 int16_t mv_cache[4][4][2];
630 const int xy= s->mb_x + s->mb_y*s->mb_stride;
631 const int mot_stride= s->b8_stride;
632 const int mot_xy= mb_x*2 + mb_y*2*mot_stride;
634 assert(!s->mb_skipped);
636 memcpy(mv_cache[1][1], s->current_picture.f.motion_val[0][mot_xy ], sizeof(int16_t) * 4);
637 memcpy(mv_cache[2][1], s->current_picture.f.motion_val[0][mot_xy + mot_stride], sizeof(int16_t) * 4);
638 memcpy(mv_cache[3][1], s->current_picture.f.motion_val[0][mot_xy + mot_stride], sizeof(int16_t) * 4);
640 if (mb_y == 0 || IS_INTRA(s->current_picture.f.mb_type[xy - s->mb_stride])) {
641 memcpy(mv_cache[0][1], mv_cache[1][1], sizeof(int16_t)*4);
643 memcpy(mv_cache[0][1], s->current_picture.f.motion_val[0][mot_xy - mot_stride], sizeof(int16_t) * 4);
646 if (mb_x == 0 || IS_INTRA(s->current_picture.f.mb_type[xy - 1])) {
647 AV_COPY32(mv_cache[1][0], mv_cache[1][1]);
648 AV_COPY32(mv_cache[2][0], mv_cache[2][1]);
650 AV_COPY32(mv_cache[1][0], s->current_picture.f.motion_val[0][mot_xy - 1]);
651 AV_COPY32(mv_cache[2][0], s->current_picture.f.motion_val[0][mot_xy - 1 + mot_stride]);
654 if (mb_x + 1 >= s->mb_width || IS_INTRA(s->current_picture.f.mb_type[xy + 1])) {
655 AV_COPY32(mv_cache[1][3], mv_cache[1][2]);
656 AV_COPY32(mv_cache[2][3], mv_cache[2][2]);
658 AV_COPY32(mv_cache[1][3], s->current_picture.f.motion_val[0][mot_xy + 2]);
659 AV_COPY32(mv_cache[2][3], s->current_picture.f.motion_val[0][mot_xy + 2 + mot_stride]);
665 const int x= (i&1)+1;
666 const int y= (i>>1)+1;
668 {mv_cache[y][x ][0], mv_cache[y][x ][1]},
669 {mv_cache[y-1][x][0], mv_cache[y-1][x][1]},
670 {mv_cache[y][x-1][0], mv_cache[y][x-1][1]},
671 {mv_cache[y][x+1][0], mv_cache[y][x+1][1]},
672 {mv_cache[y+1][x][0], mv_cache[y+1][x][1]}};
674 obmc_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
676 mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >>1) * 8,
683 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY))
684 chroma_4mv_motion(s, dest_cb, dest_cr, ref_picture, pix_op[1], mx, my);
692 if(s->real_sprite_warping_points==1){
693 gmc1_motion(s, dest_y, dest_cb, dest_cr,
696 gmc_motion(s, dest_y, dest_cb, dest_cr,
699 }else if(!is_mpeg12 && s->quarter_sample){
700 qpel_motion(s, dest_y, dest_cb, dest_cr,
702 ref_picture, pix_op, qpix_op,
703 s->mv[dir][0][0], s->mv[dir][0][1], 16);
704 } else if (!is_mpeg12 && (CONFIG_WMV2_DECODER || CONFIG_WMV2_ENCODER) &&
705 s->mspel && s->codec_id == AV_CODEC_ID_WMV2) {
706 ff_mspel_motion(s, dest_y, dest_cb, dest_cr,
708 s->mv[dir][0][0], s->mv[dir][0][1], 16);
711 mpeg_motion(s, dest_y, dest_cb, dest_cr,
714 s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y);
721 if(s->quarter_sample){
723 motion_x = s->mv[dir][i][0];
724 motion_y = s->mv[dir][i][1];
726 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
727 src_x = mb_x * 16 + (motion_x >> 2) + (i & 1) * 8;
728 src_y = mb_y * 16 + (motion_y >> 2) + (i >>1) * 8;
730 /* WARNING: do no forget half pels */
731 src_x = av_clip(src_x, -16, s->width);
732 if (src_x == s->width)
734 src_y = av_clip(src_y, -16, s->height);
735 if (src_y == s->height)
738 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x);
739 if(s->flags&CODEC_FLAG_EMU_EDGE){
740 if( (unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x&3) - 8, 0)
741 || (unsigned)src_y > FFMAX(s->v_edge_pos - (motion_y&3) - 8, 0)){
742 s->dsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
745 s->h_edge_pos, s->v_edge_pos);
746 ptr= s->edge_emu_buffer;
749 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize;
750 qpix_op[1][dxy](dest, ptr, s->linesize);
752 mx += s->mv[dir][i][0]/2;
753 my += s->mv[dir][i][1]/2;
757 hpel_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
758 ref_picture[0], 0, 0,
759 mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >>1) * 8,
760 s->width, s->height, s->linesize,
761 s->h_edge_pos, s->v_edge_pos,
763 s->mv[dir][i][0], s->mv[dir][i][1]);
765 mx += s->mv[dir][i][0];
766 my += s->mv[dir][i][1];
770 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY))
771 chroma_4mv_motion(s, dest_cb, dest_cr, ref_picture, pix_op[1], mx, my);
775 if (s->picture_structure == PICT_FRAME) {
776 if(!is_mpeg12 && s->quarter_sample){
778 qpel_motion(s, dest_y, dest_cb, dest_cr,
779 1, i, s->field_select[dir][i],
780 ref_picture, pix_op, qpix_op,
781 s->mv[dir][i][0], s->mv[dir][i][1], 8);
785 mpeg_motion(s, dest_y, dest_cb, dest_cr,
786 1, 0, s->field_select[dir][0],
788 s->mv[dir][0][0], s->mv[dir][0][1], 8, mb_y);
790 mpeg_motion(s, dest_y, dest_cb, dest_cr,
791 1, 1, s->field_select[dir][1],
793 s->mv[dir][1][0], s->mv[dir][1][1], 8, mb_y);
796 if(s->picture_structure != s->field_select[dir][0] + 1 && s->pict_type != AV_PICTURE_TYPE_B && !s->first_field){
797 ref_picture = s->current_picture_ptr->f.data;
800 mpeg_motion(s, dest_y, dest_cb, dest_cr,
801 0, 0, s->field_select[dir][0],
803 s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y>>1);
808 uint8_t ** ref2picture;
810 if(s->picture_structure == s->field_select[dir][i] + 1
811 || s->pict_type == AV_PICTURE_TYPE_B || s->first_field){
812 ref2picture= ref_picture;
814 ref2picture = s->current_picture_ptr->f.data;
817 mpeg_motion(s, dest_y, dest_cb, dest_cr,
818 0, 0, s->field_select[dir][i],
820 s->mv[dir][i][0], s->mv[dir][i][1] + 16*i, 8, mb_y>>1);
822 dest_y += 16*s->linesize;
823 dest_cb+= (16>>s->chroma_y_shift)*s->uvlinesize;
824 dest_cr+= (16>>s->chroma_y_shift)*s->uvlinesize;
828 if(s->picture_structure == PICT_FRAME){
832 mpeg_motion(s, dest_y, dest_cb, dest_cr,
835 s->mv[dir][2*i + j][0], s->mv[dir][2*i + j][1], 8, mb_y);
837 pix_op = s->dsp.avg_pixels_tab;
841 mpeg_motion(s, dest_y, dest_cb, dest_cr,
842 0, 0, s->picture_structure != i+1,
844 s->mv[dir][2*i][0],s->mv[dir][2*i][1],16, mb_y>>1);
846 // after put we make avg of the same block
847 pix_op=s->dsp.avg_pixels_tab;
849 //opposite parity is always in the same frame if this is second field
851 ref_picture = s->current_picture_ptr->f.data;
860 void ff_MPV_motion(MpegEncContext *s,
861 uint8_t *dest_y, uint8_t *dest_cb,
862 uint8_t *dest_cr, int dir,
863 uint8_t **ref_picture,
864 op_pixels_func (*pix_op)[4],
865 qpel_mc_func (*qpix_op)[16])
868 if(s->out_format == FMT_MPEG1)
869 MPV_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
870 ref_picture, pix_op, qpix_op, 1);
873 MPV_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
874 ref_picture, pix_op, qpix_op, 0);