2 * The simplest mpeg encoder (well, it was the simplest!)
3 * Copyright (c) 2000,2001 Fabrice Bellard
4 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
6 * 4MV & hq & B-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
8 * This file is part of Libav.
10 * Libav is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU Lesser General Public
12 * License as published by the Free Software Foundation; either
13 * version 2.1 of the License, or (at your option) any later version.
15 * Libav is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * Lesser General Public License for more details.
20 * You should have received a copy of the GNU Lesser General Public
21 * License along with Libav; if not, write to the Free Software
22 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
27 * The simplest mpeg encoder (well, it was the simplest!).
30 #ifndef AVCODEC_MPEGVIDEO_COMMON_H
31 #define AVCODEC_MPEGVIDEO_COMMON_H
36 #include "mpegvideo.h"
41 int ff_dct_quantize_c(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow);
44 * Set the given MpegEncContext to common defaults (same for encoding and decoding).
45 * The changed fields will not depend upon the prior state of the MpegEncContext.
47 void ff_MPV_common_defaults(MpegEncContext *s);
49 static inline void gmc1_motion(MpegEncContext *s,
50 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
51 uint8_t **ref_picture)
54 int offset, src_x, src_y, linesize, uvlinesize;
55 int motion_x, motion_y;
58 motion_x= s->sprite_offset[0][0];
59 motion_y= s->sprite_offset[0][1];
60 src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy+1));
61 src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy+1));
62 motion_x<<=(3-s->sprite_warping_accuracy);
63 motion_y<<=(3-s->sprite_warping_accuracy);
64 src_x = av_clip(src_x, -16, s->width);
65 if (src_x == s->width)
67 src_y = av_clip(src_y, -16, s->height);
68 if (src_y == s->height)
71 linesize = s->linesize;
72 uvlinesize = s->uvlinesize;
74 ptr = ref_picture[0] + (src_y * linesize) + src_x;
76 if(s->flags&CODEC_FLAG_EMU_EDGE){
77 if( (unsigned)src_x >= FFMAX(s->h_edge_pos - 17, 0)
78 || (unsigned)src_y >= FFMAX(s->v_edge_pos - 17, 0)){
79 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);
80 ptr= s->edge_emu_buffer;
84 if((motion_x|motion_y)&7){
85 s->dsp.gmc1(dest_y , ptr , linesize, 16, motion_x&15, motion_y&15, 128 - s->no_rounding);
86 s->dsp.gmc1(dest_y+8, ptr+8, linesize, 16, motion_x&15, motion_y&15, 128 - s->no_rounding);
90 dxy= ((motion_x>>3)&1) | ((motion_y>>2)&2);
92 s->dsp.put_no_rnd_pixels_tab[0][dxy](dest_y, ptr, linesize, 16);
94 s->dsp.put_pixels_tab [0][dxy](dest_y, ptr, linesize, 16);
98 if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
100 motion_x= s->sprite_offset[1][0];
101 motion_y= s->sprite_offset[1][1];
102 src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy+1));
103 src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy+1));
104 motion_x<<=(3-s->sprite_warping_accuracy);
105 motion_y<<=(3-s->sprite_warping_accuracy);
106 src_x = av_clip(src_x, -8, s->width>>1);
107 if (src_x == s->width>>1)
109 src_y = av_clip(src_y, -8, s->height>>1);
110 if (src_y == s->height>>1)
113 offset = (src_y * uvlinesize) + src_x;
114 ptr = ref_picture[1] + offset;
115 if(s->flags&CODEC_FLAG_EMU_EDGE){
116 if( (unsigned)src_x >= FFMAX((s->h_edge_pos>>1) - 9, 0)
117 || (unsigned)src_y >= FFMAX((s->v_edge_pos>>1) - 9, 0)){
118 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);
119 ptr= s->edge_emu_buffer;
123 s->dsp.gmc1(dest_cb, ptr, uvlinesize, 8, motion_x&15, motion_y&15, 128 - s->no_rounding);
125 ptr = ref_picture[2] + offset;
127 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);
128 ptr= s->edge_emu_buffer;
130 s->dsp.gmc1(dest_cr, ptr, uvlinesize, 8, motion_x&15, motion_y&15, 128 - s->no_rounding);
135 static inline void gmc_motion(MpegEncContext *s,
136 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
137 uint8_t **ref_picture)
140 int linesize, uvlinesize;
141 const int a= s->sprite_warping_accuracy;
144 linesize = s->linesize;
145 uvlinesize = s->uvlinesize;
147 ptr = ref_picture[0];
149 ox= s->sprite_offset[0][0] + s->sprite_delta[0][0]*s->mb_x*16 + 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 + s->sprite_delta[1][1]*s->mb_y*16;
152 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) return;
169 ox= s->sprite_offset[1][0] + s->sprite_delta[0][0]*s->mb_x*8 + s->sprite_delta[0][1]*s->mb_y*8;
170 oy= s->sprite_offset[1][1] + s->sprite_delta[1][0]*s->mb_x*8 + s->sprite_delta[1][1]*s->mb_y*8;
172 ptr = ref_picture[1];
173 s->dsp.gmc(dest_cb, ptr, uvlinesize, 8,
176 s->sprite_delta[0][0], s->sprite_delta[0][1],
177 s->sprite_delta[1][0], s->sprite_delta[1][1],
178 a+1, (1<<(2*a+1)) - s->no_rounding,
179 s->h_edge_pos>>1, s->v_edge_pos>>1);
181 ptr = ref_picture[2];
182 s->dsp.gmc(dest_cr, ptr, uvlinesize, 8,
185 s->sprite_delta[0][0], s->sprite_delta[0][1],
186 s->sprite_delta[1][0], s->sprite_delta[1][1],
187 a+1, (1<<(2*a+1)) - s->no_rounding,
188 s->h_edge_pos>>1, s->v_edge_pos>>1);
191 static inline int hpel_motion(MpegEncContext *s,
192 uint8_t *dest, uint8_t *src,
193 int field_based, int field_select,
194 int src_x, int src_y,
195 int width, int height, int stride,
196 int h_edge_pos, int v_edge_pos,
197 int w, int h, op_pixels_func *pix_op,
198 int motion_x, int motion_y)
203 dxy = ((motion_y & 1) << 1) | (motion_x & 1);
204 src_x += motion_x >> 1;
205 src_y += motion_y >> 1;
207 /* WARNING: do no forget half pels */
208 src_x = av_clip(src_x, -16, width); //FIXME unneeded for emu?
211 src_y = av_clip(src_y, -16, height);
214 src += src_y * stride + src_x;
216 if(s->unrestricted_mv && (s->flags&CODEC_FLAG_EMU_EDGE)){
217 if( (unsigned)src_x > FFMAX(h_edge_pos - (motion_x&1) - w, 0)
218 || (unsigned)src_y > FFMAX(v_edge_pos - (motion_y&1) - h, 0)){
219 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src, s->linesize, w+1, (h+1)<<field_based,
220 src_x, src_y<<field_based, h_edge_pos, s->v_edge_pos);
221 src= s->edge_emu_buffer;
227 pix_op[dxy](dest, src, stride, h);
231 static av_always_inline
232 void mpeg_motion_internal(MpegEncContext *s,
233 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
234 int field_based, int bottom_field, int field_select,
235 uint8_t **ref_picture, op_pixels_func (*pix_op)[4],
236 int motion_x, int motion_y, int h, int is_mpeg12, int mb_y)
238 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
239 int dxy, uvdxy, mx, my, src_x, src_y,
240 uvsrc_x, uvsrc_y, v_edge_pos, uvlinesize, linesize;
243 if(s->quarter_sample)
250 v_edge_pos = s->v_edge_pos >> field_based;
251 linesize = s->current_picture.f.linesize[0] << field_based;
252 uvlinesize = s->current_picture.f.linesize[1] << field_based;
254 dxy = ((motion_y & 1) << 1) | (motion_x & 1);
255 src_x = s->mb_x* 16 + (motion_x >> 1);
256 src_y =( mb_y<<(4-field_based)) + (motion_y >> 1);
258 if (!is_mpeg12 && s->out_format == FMT_H263) {
259 if((s->workaround_bugs & FF_BUG_HPEL_CHROMA) && field_based){
260 mx = (motion_x>>1)|(motion_x&1);
262 uvdxy = ((my & 1) << 1) | (mx & 1);
263 uvsrc_x = s->mb_x* 8 + (mx >> 1);
264 uvsrc_y =( mb_y<<(3-field_based))+ (my >> 1);
266 uvdxy = dxy | (motion_y & 2) | ((motion_x & 2) >> 1);
270 }else if(!is_mpeg12 && s->out_format == FMT_H261){//even chroma mv's are full pel in H261
274 uvsrc_x = s->mb_x*8 + mx;
275 uvsrc_y = mb_y*8 + my;
277 if(s->chroma_y_shift){
280 uvdxy = ((my & 1) << 1) | (mx & 1);
281 uvsrc_x = s->mb_x* 8 + (mx >> 1);
282 uvsrc_y =( mb_y<<(3-field_based))+ (my >> 1);
284 if(s->chroma_x_shift){
287 uvdxy = ((motion_y & 1) << 1) | (mx & 1);
288 uvsrc_x = s->mb_x* 8 + (mx >> 1);
299 ptr_y = ref_picture[0] + src_y * linesize + src_x;
300 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
301 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
303 if( (unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x&1) - 16, 0)
304 || (unsigned)src_y > FFMAX( v_edge_pos - (motion_y&1) - h , 0)){
305 if(is_mpeg12 || s->codec_id == CODEC_ID_MPEG2VIDEO ||
306 s->codec_id == CODEC_ID_MPEG1VIDEO){
307 av_log(s->avctx,AV_LOG_DEBUG,
308 "MPEG motion vector out of boundary (%d %d)\n", src_x, src_y);
311 s->dsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y, s->linesize,
313 src_x, src_y<<field_based,
314 s->h_edge_pos, s->v_edge_pos);
315 ptr_y = s->edge_emu_buffer;
316 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
317 uint8_t *uvbuf= s->edge_emu_buffer+18*s->linesize;
318 s->dsp.emulated_edge_mc(uvbuf ,
319 ptr_cb, s->uvlinesize,
321 uvsrc_x, uvsrc_y<<field_based,
322 s->h_edge_pos>>1, s->v_edge_pos>>1);
323 s->dsp.emulated_edge_mc(uvbuf+16,
324 ptr_cr, s->uvlinesize,
326 uvsrc_x, uvsrc_y<<field_based,
327 s->h_edge_pos>>1, s->v_edge_pos>>1);
333 if(bottom_field){ //FIXME use this for field pix too instead of the obnoxious hack which changes picture.data
334 dest_y += s->linesize;
335 dest_cb+= s->uvlinesize;
336 dest_cr+= s->uvlinesize;
340 ptr_y += s->linesize;
341 ptr_cb+= s->uvlinesize;
342 ptr_cr+= s->uvlinesize;
345 pix_op[0][dxy](dest_y, ptr_y, linesize, h);
347 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
348 pix_op[s->chroma_x_shift][uvdxy]
349 (dest_cb, ptr_cb, uvlinesize, h >> s->chroma_y_shift);
350 pix_op[s->chroma_x_shift][uvdxy]
351 (dest_cr, ptr_cr, uvlinesize, h >> s->chroma_y_shift);
353 if(!is_mpeg12 && (CONFIG_H261_ENCODER || CONFIG_H261_DECODER) &&
354 s->out_format == FMT_H261){
355 ff_h261_loop_filter(s);
358 /* apply one mpeg motion vector to the three components */
359 static av_always_inline
360 void mpeg_motion(MpegEncContext *s,
361 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
362 int field_based, int bottom_field, int field_select,
363 uint8_t **ref_picture, op_pixels_func (*pix_op)[4],
364 int motion_x, int motion_y, int h, int mb_y)
367 if(s->out_format == FMT_MPEG1)
368 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, field_based,
369 bottom_field, field_select, ref_picture, pix_op,
370 motion_x, motion_y, h, 1, mb_y);
373 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, field_based,
374 bottom_field, field_select, ref_picture, pix_op,
375 motion_x, motion_y, h, 0, mb_y);
378 //FIXME move to dsputil, avg variant, 16x16 version
379 static inline void put_obmc(uint8_t *dst, uint8_t *src[5], int stride){
381 uint8_t * const top = src[1];
382 uint8_t * const left = src[2];
383 uint8_t * const mid = src[0];
384 uint8_t * const right = src[3];
385 uint8_t * const bottom= src[4];
386 #define OBMC_FILTER(x, t, l, m, r, b)\
387 dst[x]= (t*top[x] + l*left[x] + m*mid[x] + r*right[x] + b*bottom[x] + 4)>>3
388 #define OBMC_FILTER4(x, t, l, m, r, b)\
389 OBMC_FILTER(x , t, l, m, r, b);\
390 OBMC_FILTER(x+1 , t, l, m, r, b);\
391 OBMC_FILTER(x +stride, t, l, m, r, b);\
392 OBMC_FILTER(x+1+stride, t, l, m, r, b);
395 OBMC_FILTER (x , 2, 2, 4, 0, 0);
396 OBMC_FILTER (x+1, 2, 1, 5, 0, 0);
397 OBMC_FILTER4(x+2, 2, 1, 5, 0, 0);
398 OBMC_FILTER4(x+4, 2, 0, 5, 1, 0);
399 OBMC_FILTER (x+6, 2, 0, 5, 1, 0);
400 OBMC_FILTER (x+7, 2, 0, 4, 2, 0);
402 OBMC_FILTER (x , 1, 2, 5, 0, 0);
403 OBMC_FILTER (x+1, 1, 2, 5, 0, 0);
404 OBMC_FILTER (x+6, 1, 0, 5, 2, 0);
405 OBMC_FILTER (x+7, 1, 0, 5, 2, 0);
407 OBMC_FILTER4(x , 1, 2, 5, 0, 0);
408 OBMC_FILTER4(x+2, 1, 1, 6, 0, 0);
409 OBMC_FILTER4(x+4, 1, 0, 6, 1, 0);
410 OBMC_FILTER4(x+6, 1, 0, 5, 2, 0);
412 OBMC_FILTER4(x , 0, 2, 5, 0, 1);
413 OBMC_FILTER4(x+2, 0, 1, 6, 0, 1);
414 OBMC_FILTER4(x+4, 0, 0, 6, 1, 1);
415 OBMC_FILTER4(x+6, 0, 0, 5, 2, 1);
417 OBMC_FILTER (x , 0, 2, 5, 0, 1);
418 OBMC_FILTER (x+1, 0, 2, 5, 0, 1);
419 OBMC_FILTER4(x+2, 0, 1, 5, 0, 2);
420 OBMC_FILTER4(x+4, 0, 0, 5, 1, 2);
421 OBMC_FILTER (x+6, 0, 0, 5, 2, 1);
422 OBMC_FILTER (x+7, 0, 0, 5, 2, 1);
424 OBMC_FILTER (x , 0, 2, 4, 0, 2);
425 OBMC_FILTER (x+1, 0, 1, 5, 0, 2);
426 OBMC_FILTER (x+6, 0, 0, 5, 1, 2);
427 OBMC_FILTER (x+7, 0, 0, 4, 2, 2);
430 /* obmc for 1 8x8 luma block */
431 static inline void obmc_motion(MpegEncContext *s,
432 uint8_t *dest, uint8_t *src,
433 int src_x, int src_y,
434 op_pixels_func *pix_op,
435 int16_t mv[5][2]/* mid top left right bottom*/)
441 assert(s->quarter_sample==0);
444 if(i && mv[i][0]==mv[MID][0] && mv[i][1]==mv[MID][1]){
447 ptr[i]= s->obmc_scratchpad + 8*(i&1) + s->linesize*8*(i>>1);
448 hpel_motion(s, ptr[i], src, 0, 0,
450 s->width, s->height, s->linesize,
451 s->h_edge_pos, s->v_edge_pos,
457 put_obmc(dest, ptr, s->linesize);
460 static inline void qpel_motion(MpegEncContext *s,
461 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
462 int field_based, int bottom_field, int field_select,
463 uint8_t **ref_picture, op_pixels_func (*pix_op)[4],
464 qpel_mc_func (*qpix_op)[16],
465 int motion_x, int motion_y, int h)
467 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
468 int dxy, uvdxy, mx, my, src_x, src_y, uvsrc_x, uvsrc_y, v_edge_pos, linesize, uvlinesize;
470 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
471 src_x = s->mb_x * 16 + (motion_x >> 2);
472 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2);
474 v_edge_pos = s->v_edge_pos >> field_based;
475 linesize = s->linesize << field_based;
476 uvlinesize = s->uvlinesize << field_based;
481 }else if(s->workaround_bugs&FF_BUG_QPEL_CHROMA2){
482 static const int rtab[8]= {0,0,1,1,0,0,0,1};
483 mx= (motion_x>>1) + rtab[motion_x&7];
484 my= (motion_y>>1) + rtab[motion_y&7];
485 }else if(s->workaround_bugs&FF_BUG_QPEL_CHROMA){
486 mx= (motion_x>>1)|(motion_x&1);
487 my= (motion_y>>1)|(motion_y&1);
495 uvdxy= (mx&1) | ((my&1)<<1);
499 uvsrc_x = s->mb_x * 8 + mx;
500 uvsrc_y = s->mb_y * (8 >> field_based) + my;
502 ptr_y = ref_picture[0] + src_y * linesize + src_x;
503 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
504 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
506 if( (unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x&3) - 16, 0)
507 || (unsigned)src_y > FFMAX( v_edge_pos - (motion_y&3) - h , 0)){
508 s->dsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y, s->linesize,
509 17, 17+field_based, src_x, src_y<<field_based,
510 s->h_edge_pos, s->v_edge_pos);
511 ptr_y= s->edge_emu_buffer;
512 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
513 uint8_t *uvbuf= s->edge_emu_buffer + 18*s->linesize;
514 s->dsp.emulated_edge_mc(uvbuf, ptr_cb, s->uvlinesize,
516 uvsrc_x, uvsrc_y<<field_based,
517 s->h_edge_pos>>1, s->v_edge_pos>>1);
518 s->dsp.emulated_edge_mc(uvbuf + 16, ptr_cr, s->uvlinesize,
520 uvsrc_x, uvsrc_y<<field_based,
521 s->h_edge_pos>>1, s->v_edge_pos>>1);
528 qpix_op[0][dxy](dest_y, ptr_y, linesize);
531 dest_y += s->linesize;
532 dest_cb+= s->uvlinesize;
533 dest_cr+= s->uvlinesize;
537 ptr_y += s->linesize;
538 ptr_cb += s->uvlinesize;
539 ptr_cr += s->uvlinesize;
541 //damn interlaced mode
542 //FIXME boundary mirroring is not exactly correct here
543 qpix_op[1][dxy](dest_y , ptr_y , linesize);
544 qpix_op[1][dxy](dest_y+8, ptr_y+8, linesize);
546 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
547 pix_op[1][uvdxy](dest_cr, ptr_cr, uvlinesize, h >> 1);
548 pix_op[1][uvdxy](dest_cb, ptr_cb, uvlinesize, h >> 1);
553 * h263 chroma 4mv motion compensation.
555 static inline void chroma_4mv_motion(MpegEncContext *s,
556 uint8_t *dest_cb, uint8_t *dest_cr,
557 uint8_t **ref_picture,
558 op_pixels_func *pix_op,
560 int dxy, emu=0, src_x, src_y, offset;
563 /* In case of 8X8, we construct a single chroma motion vector
564 with a special rounding */
565 mx= ff_h263_round_chroma(mx);
566 my= ff_h263_round_chroma(my);
568 dxy = ((my & 1) << 1) | (mx & 1);
572 src_x = s->mb_x * 8 + mx;
573 src_y = s->mb_y * 8 + my;
574 src_x = av_clip(src_x, -8, (s->width >> 1));
575 if (src_x == (s->width >> 1))
577 src_y = av_clip(src_y, -8, (s->height >> 1));
578 if (src_y == (s->height >> 1))
581 offset = src_y * s->uvlinesize + src_x;
582 ptr = ref_picture[1] + offset;
583 if(s->flags&CODEC_FLAG_EMU_EDGE){
584 if( (unsigned)src_x > FFMAX((s->h_edge_pos>>1) - (dxy &1) - 8, 0)
585 || (unsigned)src_y > FFMAX((s->v_edge_pos>>1) - (dxy>>1) - 8, 0)){
586 s->dsp.emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize,
588 s->h_edge_pos>>1, s->v_edge_pos>>1);
589 ptr= s->edge_emu_buffer;
593 pix_op[dxy](dest_cb, ptr, s->uvlinesize, 8);
595 ptr = ref_picture[2] + offset;
597 s->dsp.emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize,
599 s->h_edge_pos>>1, s->v_edge_pos>>1);
600 ptr= s->edge_emu_buffer;
602 pix_op[dxy](dest_cr, ptr, s->uvlinesize, 8);
605 static inline void prefetch_motion(MpegEncContext *s, uint8_t **pix, int dir){
606 /* fetch pixels for estimated mv 4 macroblocks ahead
607 * optimized for 64byte cache lines */
608 const int shift = s->quarter_sample ? 2 : 1;
609 const int mx= (s->mv[dir][0][0]>>shift) + 16*s->mb_x + 8;
610 const int my= (s->mv[dir][0][1]>>shift) + 16*s->mb_y;
611 int off= mx + (my + (s->mb_x&3)*4)*s->linesize + 64;
612 s->dsp.prefetch(pix[0]+off, s->linesize, 4);
613 off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
614 s->dsp.prefetch(pix[1]+off, pix[2]-pix[1], 2);
618 * motion compensation of a single macroblock
620 * @param dest_y luma destination pointer
621 * @param dest_cb chroma cb/u destination pointer
622 * @param dest_cr chroma cr/v destination pointer
623 * @param dir direction (0->forward, 1->backward)
624 * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
625 * @param pix_op halfpel motion compensation function (average or put normally)
626 * @param qpix_op qpel motion compensation function (average or put normally)
627 * the motion vectors are taken from s->mv and the MV type from s->mv_type
629 static av_always_inline void MPV_motion_internal(MpegEncContext *s,
630 uint8_t *dest_y, uint8_t *dest_cb,
631 uint8_t *dest_cr, int dir,
632 uint8_t **ref_picture,
633 op_pixels_func (*pix_op)[4],
634 qpel_mc_func (*qpix_op)[16], int is_mpeg12)
636 int dxy, mx, my, src_x, src_y, motion_x, motion_y;
643 prefetch_motion(s, ref_picture, dir);
645 if(!is_mpeg12 && s->obmc && s->pict_type != AV_PICTURE_TYPE_B){
646 int16_t mv_cache[4][4][2];
647 const int xy= s->mb_x + s->mb_y*s->mb_stride;
648 const int mot_stride= s->b8_stride;
649 const int mot_xy= mb_x*2 + mb_y*2*mot_stride;
651 assert(!s->mb_skipped);
653 memcpy(mv_cache[1][1], s->current_picture.f.motion_val[0][mot_xy ], sizeof(int16_t) * 4);
654 memcpy(mv_cache[2][1], s->current_picture.f.motion_val[0][mot_xy + mot_stride], sizeof(int16_t) * 4);
655 memcpy(mv_cache[3][1], s->current_picture.f.motion_val[0][mot_xy + mot_stride], sizeof(int16_t) * 4);
657 if (mb_y == 0 || IS_INTRA(s->current_picture.f.mb_type[xy - s->mb_stride])) {
658 memcpy(mv_cache[0][1], mv_cache[1][1], sizeof(int16_t)*4);
660 memcpy(mv_cache[0][1], s->current_picture.f.motion_val[0][mot_xy - mot_stride], sizeof(int16_t) * 4);
663 if (mb_x == 0 || IS_INTRA(s->current_picture.f.mb_type[xy - 1])) {
664 AV_COPY32(mv_cache[1][0], mv_cache[1][1]);
665 AV_COPY32(mv_cache[2][0], mv_cache[2][1]);
667 AV_COPY32(mv_cache[1][0], s->current_picture.f.motion_val[0][mot_xy - 1]);
668 AV_COPY32(mv_cache[2][0], s->current_picture.f.motion_val[0][mot_xy - 1 + mot_stride]);
671 if (mb_x + 1 >= s->mb_width || IS_INTRA(s->current_picture.f.mb_type[xy + 1])) {
672 AV_COPY32(mv_cache[1][3], mv_cache[1][2]);
673 AV_COPY32(mv_cache[2][3], mv_cache[2][2]);
675 AV_COPY32(mv_cache[1][3], s->current_picture.f.motion_val[0][mot_xy + 2]);
676 AV_COPY32(mv_cache[2][3], s->current_picture.f.motion_val[0][mot_xy + 2 + mot_stride]);
682 const int x= (i&1)+1;
683 const int y= (i>>1)+1;
685 {mv_cache[y][x ][0], mv_cache[y][x ][1]},
686 {mv_cache[y-1][x][0], mv_cache[y-1][x][1]},
687 {mv_cache[y][x-1][0], mv_cache[y][x-1][1]},
688 {mv_cache[y][x+1][0], mv_cache[y][x+1][1]},
689 {mv_cache[y+1][x][0], mv_cache[y+1][x][1]}};
691 obmc_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
693 mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >>1) * 8,
700 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY))
701 chroma_4mv_motion(s, dest_cb, dest_cr, ref_picture, pix_op[1], mx, my);
709 if(s->real_sprite_warping_points==1){
710 gmc1_motion(s, dest_y, dest_cb, dest_cr,
713 gmc_motion(s, dest_y, dest_cb, dest_cr,
716 }else if(!is_mpeg12 && s->quarter_sample){
717 qpel_motion(s, dest_y, dest_cb, dest_cr,
719 ref_picture, pix_op, qpix_op,
720 s->mv[dir][0][0], s->mv[dir][0][1], 16);
721 } else if (!is_mpeg12 && (CONFIG_WMV2_DECODER || CONFIG_WMV2_ENCODER) &&
722 s->mspel && s->codec_id == CODEC_ID_WMV2) {
723 ff_mspel_motion(s, dest_y, dest_cb, dest_cr,
725 s->mv[dir][0][0], s->mv[dir][0][1], 16);
728 mpeg_motion(s, dest_y, dest_cb, dest_cr,
731 s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y);
738 if(s->quarter_sample){
740 motion_x = s->mv[dir][i][0];
741 motion_y = s->mv[dir][i][1];
743 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
744 src_x = mb_x * 16 + (motion_x >> 2) + (i & 1) * 8;
745 src_y = mb_y * 16 + (motion_y >> 2) + (i >>1) * 8;
747 /* WARNING: do no forget half pels */
748 src_x = av_clip(src_x, -16, s->width);
749 if (src_x == s->width)
751 src_y = av_clip(src_y, -16, s->height);
752 if (src_y == s->height)
755 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x);
756 if(s->flags&CODEC_FLAG_EMU_EDGE){
757 if( (unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x&3) - 8, 0)
758 || (unsigned)src_y > FFMAX(s->v_edge_pos - (motion_y&3) - 8, 0)){
759 s->dsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
762 s->h_edge_pos, s->v_edge_pos);
763 ptr= s->edge_emu_buffer;
766 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize;
767 qpix_op[1][dxy](dest, ptr, s->linesize);
769 mx += s->mv[dir][i][0]/2;
770 my += s->mv[dir][i][1]/2;
774 hpel_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
775 ref_picture[0], 0, 0,
776 mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >>1) * 8,
777 s->width, s->height, s->linesize,
778 s->h_edge_pos, s->v_edge_pos,
780 s->mv[dir][i][0], s->mv[dir][i][1]);
782 mx += s->mv[dir][i][0];
783 my += s->mv[dir][i][1];
787 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY))
788 chroma_4mv_motion(s, dest_cb, dest_cr, ref_picture, pix_op[1], mx, my);
792 if (s->picture_structure == PICT_FRAME) {
793 if(!is_mpeg12 && s->quarter_sample){
795 qpel_motion(s, dest_y, dest_cb, dest_cr,
796 1, i, s->field_select[dir][i],
797 ref_picture, pix_op, qpix_op,
798 s->mv[dir][i][0], s->mv[dir][i][1], 8);
802 mpeg_motion(s, dest_y, dest_cb, dest_cr,
803 1, 0, s->field_select[dir][0],
805 s->mv[dir][0][0], s->mv[dir][0][1], 8, mb_y);
807 mpeg_motion(s, dest_y, dest_cb, dest_cr,
808 1, 1, s->field_select[dir][1],
810 s->mv[dir][1][0], s->mv[dir][1][1], 8, mb_y);
813 if(s->picture_structure != s->field_select[dir][0] + 1 && s->pict_type != AV_PICTURE_TYPE_B && !s->first_field){
814 ref_picture = s->current_picture_ptr->f.data;
817 mpeg_motion(s, dest_y, dest_cb, dest_cr,
818 0, 0, s->field_select[dir][0],
820 s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y>>1);
825 uint8_t ** ref2picture;
827 if(s->picture_structure == s->field_select[dir][i] + 1
828 || s->pict_type == AV_PICTURE_TYPE_B || s->first_field){
829 ref2picture= ref_picture;
831 ref2picture = s->current_picture_ptr->f.data;
834 mpeg_motion(s, dest_y, dest_cb, dest_cr,
835 0, 0, s->field_select[dir][i],
837 s->mv[dir][i][0], s->mv[dir][i][1] + 16*i, 8, mb_y>>1);
839 dest_y += 16*s->linesize;
840 dest_cb+= (16>>s->chroma_y_shift)*s->uvlinesize;
841 dest_cr+= (16>>s->chroma_y_shift)*s->uvlinesize;
845 if(s->picture_structure == PICT_FRAME){
849 mpeg_motion(s, dest_y, dest_cb, dest_cr,
852 s->mv[dir][2*i + j][0], s->mv[dir][2*i + j][1], 8, mb_y);
854 pix_op = s->dsp.avg_pixels_tab;
858 mpeg_motion(s, dest_y, dest_cb, dest_cr,
859 0, 0, s->picture_structure != i+1,
861 s->mv[dir][2*i][0],s->mv[dir][2*i][1],16, mb_y>>1);
863 // after put we make avg of the same block
864 pix_op=s->dsp.avg_pixels_tab;
866 //opposite parity is always in the same frame if this is second field
868 ref_picture = s->current_picture_ptr->f.data;
877 static inline void MPV_motion(MpegEncContext *s,
878 uint8_t *dest_y, uint8_t *dest_cb,
879 uint8_t *dest_cr, int dir,
880 uint8_t **ref_picture,
881 op_pixels_func (*pix_op)[4],
882 qpel_mc_func (*qpix_op)[16])
885 if(s->out_format == FMT_MPEG1)
886 MPV_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
887 ref_picture, pix_op, qpix_op, 1);
890 MPV_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
891 ref_picture, pix_op, qpix_op, 0);
893 #endif /* AVCODEC_MPEGVIDEO_COMMON_H */