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 FFmpeg.
10 * FFmpeg 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 * FFmpeg 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 FFmpeg; 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
34 #include "libavutil/avassert.h"
37 #include "mpegvideo.h"
43 int ff_dct_quantize_c(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow);
46 * Set the given MpegEncContext to common defaults (same for encoding and decoding).
47 * The changed fields will not depend upon the prior state of the MpegEncContext.
49 void ff_MPV_common_defaults(MpegEncContext *s);
51 static inline void gmc1_motion(MpegEncContext *s,
52 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
53 uint8_t **ref_picture)
56 int offset, src_x, src_y, linesize, uvlinesize;
57 int motion_x, motion_y;
60 motion_x= s->sprite_offset[0][0];
61 motion_y= s->sprite_offset[0][1];
62 src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy+1));
63 src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy+1));
64 motion_x<<=(3-s->sprite_warping_accuracy);
65 motion_y<<=(3-s->sprite_warping_accuracy);
66 src_x = av_clip(src_x, -16, s->width);
67 if (src_x == s->width)
69 src_y = av_clip(src_y, -16, s->height);
70 if (src_y == s->height)
73 linesize = s->linesize;
74 uvlinesize = s->uvlinesize;
76 ptr = ref_picture[0] + (src_y * linesize) + src_x;
78 if(s->flags&CODEC_FLAG_EMU_EDGE){
79 if( (unsigned)src_x >= FFMAX(s->h_edge_pos - 17, 0)
80 || (unsigned)src_y >= FFMAX(s->v_edge_pos - 17, 0)){
81 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);
82 ptr= s->edge_emu_buffer;
86 if((motion_x|motion_y)&7){
87 s->dsp.gmc1(dest_y , ptr , linesize, 16, motion_x&15, motion_y&15, 128 - s->no_rounding);
88 s->dsp.gmc1(dest_y+8, ptr+8, linesize, 16, motion_x&15, motion_y&15, 128 - s->no_rounding);
92 dxy= ((motion_x>>3)&1) | ((motion_y>>2)&2);
94 s->dsp.put_no_rnd_pixels_tab[0][dxy](dest_y, ptr, linesize, 16);
96 s->dsp.put_pixels_tab [0][dxy](dest_y, ptr, linesize, 16);
100 if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
102 motion_x= s->sprite_offset[1][0];
103 motion_y= s->sprite_offset[1][1];
104 src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy+1));
105 src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy+1));
106 motion_x<<=(3-s->sprite_warping_accuracy);
107 motion_y<<=(3-s->sprite_warping_accuracy);
108 src_x = av_clip(src_x, -8, s->width>>1);
109 if (src_x == s->width>>1)
111 src_y = av_clip(src_y, -8, s->height>>1);
112 if (src_y == s->height>>1)
115 offset = (src_y * uvlinesize) + src_x;
116 ptr = ref_picture[1] + offset;
117 if(s->flags&CODEC_FLAG_EMU_EDGE){
118 if( (unsigned)src_x >= FFMAX((s->h_edge_pos>>1) - 9, 0)
119 || (unsigned)src_y >= FFMAX((s->v_edge_pos>>1) - 9, 0)){
120 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);
121 ptr= s->edge_emu_buffer;
125 s->dsp.gmc1(dest_cb, ptr, uvlinesize, 8, motion_x&15, motion_y&15, 128 - s->no_rounding);
127 ptr = ref_picture[2] + offset;
129 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);
130 ptr= s->edge_emu_buffer;
132 s->dsp.gmc1(dest_cr, ptr, uvlinesize, 8, motion_x&15, motion_y&15, 128 - s->no_rounding);
137 static inline void gmc_motion(MpegEncContext *s,
138 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
139 uint8_t **ref_picture)
142 int linesize, uvlinesize;
143 const int a= s->sprite_warping_accuracy;
146 linesize = s->linesize;
147 uvlinesize = s->uvlinesize;
149 ptr = ref_picture[0];
151 ox= s->sprite_offset[0][0] + s->sprite_delta[0][0]*s->mb_x*16 + 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 + s->sprite_delta[1][1]*s->mb_y*16;
154 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) return;
171 ox= s->sprite_offset[1][0] + s->sprite_delta[0][0]*s->mb_x*8 + 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 + s->sprite_delta[1][1]*s->mb_y*8;
174 ptr = ref_picture[1];
175 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,
187 s->sprite_delta[0][0], s->sprite_delta[0][1],
188 s->sprite_delta[1][0], s->sprite_delta[1][1],
189 a+1, (1<<(2*a+1)) - s->no_rounding,
190 s->h_edge_pos>>1, s->v_edge_pos>>1);
193 static inline int hpel_motion(MpegEncContext *s,
194 uint8_t *dest, uint8_t *src,
195 int field_based, int field_select,
196 int src_x, int src_y,
197 int width, int height, int stride,
198 int h_edge_pos, int v_edge_pos,
199 int w, int h, op_pixels_func *pix_op,
200 int motion_x, int motion_y)
205 dxy = ((motion_y & 1) << 1) | (motion_x & 1);
206 src_x += motion_x >> 1;
207 src_y += motion_y >> 1;
209 /* WARNING: do no forget half pels */
210 src_x = av_clip(src_x, -16, width); //FIXME unneeded for emu?
213 src_y = av_clip(src_y, -16, height);
216 src += src_y * stride + src_x;
218 if(s->unrestricted_mv && (s->flags&CODEC_FLAG_EMU_EDGE)){
219 if( (unsigned)src_x > FFMAX(h_edge_pos - (motion_x&1) - w, 0)
220 || (unsigned)src_y > FFMAX(v_edge_pos - (motion_y&1) - h, 0)){
221 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src, s->linesize, w+1, (h+1)<<field_based,
222 src_x, src_y<<field_based, h_edge_pos, s->v_edge_pos);
223 src= s->edge_emu_buffer;
229 pix_op[dxy](dest, src, stride, h);
233 static av_always_inline
234 void mpeg_motion_internal(MpegEncContext *s,
235 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
236 int field_based, int bottom_field, int field_select,
237 uint8_t **ref_picture, op_pixels_func (*pix_op)[4],
238 int motion_x, int motion_y, int h, int is_mpeg12, int mb_y)
240 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
241 int dxy, uvdxy, mx, my, src_x, src_y,
242 uvsrc_x, uvsrc_y, v_edge_pos, uvlinesize, linesize;
245 if(s->quarter_sample)
252 v_edge_pos = s->v_edge_pos >> field_based;
253 linesize = s->current_picture.f.linesize[0] << field_based;
254 uvlinesize = s->current_picture.f.linesize[1] << field_based;
256 dxy = ((motion_y & 1) << 1) | (motion_x & 1);
257 src_x = s->mb_x* 16 + (motion_x >> 1);
258 src_y =( mb_y<<(4-field_based)) + (motion_y >> 1);
260 if (!is_mpeg12 && s->out_format == FMT_H263) {
261 if((s->workaround_bugs & FF_BUG_HPEL_CHROMA) && field_based){
262 mx = (motion_x>>1)|(motion_x&1);
264 uvdxy = ((my & 1) << 1) | (mx & 1);
265 uvsrc_x = s->mb_x* 8 + (mx >> 1);
266 uvsrc_y =( mb_y<<(3-field_based))+ (my >> 1);
268 uvdxy = dxy | (motion_y & 2) | ((motion_x & 2) >> 1);
272 }else if(!is_mpeg12 && s->out_format == FMT_H261){//even chroma mv's are full pel in H261
276 uvsrc_x = s->mb_x*8 + mx;
277 uvsrc_y = mb_y*8 + my;
279 if(s->chroma_y_shift){
282 uvdxy = ((my & 1) << 1) | (mx & 1);
283 uvsrc_x = s->mb_x* 8 + (mx >> 1);
284 uvsrc_y =( mb_y<<(3-field_based))+ (my >> 1);
286 if(s->chroma_x_shift){
289 uvdxy = ((motion_y & 1) << 1) | (mx & 1);
290 uvsrc_x = s->mb_x* 8 + (mx >> 1);
301 ptr_y = ref_picture[0] + src_y * linesize + src_x;
302 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
303 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
305 if( (unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x&1) - 16, 0)
306 || (unsigned)src_y > FFMAX( v_edge_pos - (motion_y&1) - h , 0)){
307 if(is_mpeg12 || s->codec_id == CODEC_ID_MPEG2VIDEO ||
308 s->codec_id == CODEC_ID_MPEG1VIDEO){
309 av_log(s->avctx,AV_LOG_DEBUG,
310 "MPEG motion vector out of boundary (%d %d)\n", src_x, src_y);
313 s->dsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y, s->linesize,
315 src_x, src_y<<field_based,
316 s->h_edge_pos, s->v_edge_pos);
317 ptr_y = s->edge_emu_buffer;
318 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
319 uint8_t *uvbuf= s->edge_emu_buffer+18*s->linesize;
320 s->dsp.emulated_edge_mc(uvbuf ,
321 ptr_cb, s->uvlinesize,
323 uvsrc_x, uvsrc_y<<field_based,
324 s->h_edge_pos>>1, s->v_edge_pos>>1);
325 s->dsp.emulated_edge_mc(uvbuf+16,
326 ptr_cr, s->uvlinesize,
328 uvsrc_x, uvsrc_y<<field_based,
329 s->h_edge_pos>>1, s->v_edge_pos>>1);
335 if(bottom_field){ //FIXME use this for field pix too instead of the obnoxious hack which changes picture.data
336 dest_y += s->linesize;
337 dest_cb+= s->uvlinesize;
338 dest_cr+= s->uvlinesize;
342 ptr_y += s->linesize;
343 ptr_cb+= s->uvlinesize;
344 ptr_cr+= s->uvlinesize;
347 pix_op[0][dxy](dest_y, ptr_y, linesize, h);
349 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
350 pix_op[s->chroma_x_shift][uvdxy]
351 (dest_cb, ptr_cb, uvlinesize, h >> s->chroma_y_shift);
352 pix_op[s->chroma_x_shift][uvdxy]
353 (dest_cr, ptr_cr, uvlinesize, h >> s->chroma_y_shift);
355 if(!is_mpeg12 && (CONFIG_H261_ENCODER || CONFIG_H261_DECODER) &&
356 s->out_format == FMT_H261){
357 ff_h261_loop_filter(s);
360 /* apply one mpeg motion vector to the three components */
361 static av_always_inline
362 void mpeg_motion(MpegEncContext *s,
363 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
364 int field_based, int bottom_field, int field_select,
365 uint8_t **ref_picture, op_pixels_func (*pix_op)[4],
366 int motion_x, int motion_y, int h, int mb_y)
369 if(s->out_format == FMT_MPEG1)
370 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, field_based,
371 bottom_field, field_select, ref_picture, pix_op,
372 motion_x, motion_y, h, 1, mb_y);
375 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, field_based,
376 bottom_field, field_select, ref_picture, pix_op,
377 motion_x, motion_y, h, 0, mb_y);
380 //FIXME move to dsputil, avg variant, 16x16 version
381 static inline void put_obmc(uint8_t *dst, uint8_t *src[5], int stride){
383 uint8_t * const top = src[1];
384 uint8_t * const left = src[2];
385 uint8_t * const mid = src[0];
386 uint8_t * const right = src[3];
387 uint8_t * const bottom= src[4];
388 #define OBMC_FILTER(x, t, l, m, r, b)\
389 dst[x]= (t*top[x] + l*left[x] + m*mid[x] + r*right[x] + b*bottom[x] + 4)>>3
390 #define OBMC_FILTER4(x, t, l, m, r, b)\
391 OBMC_FILTER(x , t, l, m, r, b);\
392 OBMC_FILTER(x+1 , t, l, m, r, b);\
393 OBMC_FILTER(x +stride, t, l, m, r, b);\
394 OBMC_FILTER(x+1+stride, t, l, m, r, b);
397 OBMC_FILTER (x , 2, 2, 4, 0, 0);
398 OBMC_FILTER (x+1, 2, 1, 5, 0, 0);
399 OBMC_FILTER4(x+2, 2, 1, 5, 0, 0);
400 OBMC_FILTER4(x+4, 2, 0, 5, 1, 0);
401 OBMC_FILTER (x+6, 2, 0, 5, 1, 0);
402 OBMC_FILTER (x+7, 2, 0, 4, 2, 0);
404 OBMC_FILTER (x , 1, 2, 5, 0, 0);
405 OBMC_FILTER (x+1, 1, 2, 5, 0, 0);
406 OBMC_FILTER (x+6, 1, 0, 5, 2, 0);
407 OBMC_FILTER (x+7, 1, 0, 5, 2, 0);
409 OBMC_FILTER4(x , 1, 2, 5, 0, 0);
410 OBMC_FILTER4(x+2, 1, 1, 6, 0, 0);
411 OBMC_FILTER4(x+4, 1, 0, 6, 1, 0);
412 OBMC_FILTER4(x+6, 1, 0, 5, 2, 0);
414 OBMC_FILTER4(x , 0, 2, 5, 0, 1);
415 OBMC_FILTER4(x+2, 0, 1, 6, 0, 1);
416 OBMC_FILTER4(x+4, 0, 0, 6, 1, 1);
417 OBMC_FILTER4(x+6, 0, 0, 5, 2, 1);
419 OBMC_FILTER (x , 0, 2, 5, 0, 1);
420 OBMC_FILTER (x+1, 0, 2, 5, 0, 1);
421 OBMC_FILTER4(x+2, 0, 1, 5, 0, 2);
422 OBMC_FILTER4(x+4, 0, 0, 5, 1, 2);
423 OBMC_FILTER (x+6, 0, 0, 5, 2, 1);
424 OBMC_FILTER (x+7, 0, 0, 5, 2, 1);
426 OBMC_FILTER (x , 0, 2, 4, 0, 2);
427 OBMC_FILTER (x+1, 0, 1, 5, 0, 2);
428 OBMC_FILTER (x+6, 0, 0, 5, 1, 2);
429 OBMC_FILTER (x+7, 0, 0, 4, 2, 2);
432 /* obmc for 1 8x8 luma block */
433 static inline void obmc_motion(MpegEncContext *s,
434 uint8_t *dest, uint8_t *src,
435 int src_x, int src_y,
436 op_pixels_func *pix_op,
437 int16_t mv[5][2]/* mid top left right bottom*/)
443 av_assert2(s->quarter_sample==0);
446 if(i && mv[i][0]==mv[MID][0] && mv[i][1]==mv[MID][1]){
449 ptr[i]= s->obmc_scratchpad + 8*(i&1) + s->linesize*8*(i>>1);
450 hpel_motion(s, ptr[i], src, 0, 0,
452 s->width, s->height, s->linesize,
453 s->h_edge_pos, s->v_edge_pos,
459 put_obmc(dest, ptr, s->linesize);
462 static inline void qpel_motion(MpegEncContext *s,
463 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
464 int field_based, int bottom_field, int field_select,
465 uint8_t **ref_picture, op_pixels_func (*pix_op)[4],
466 qpel_mc_func (*qpix_op)[16],
467 int motion_x, int motion_y, int h)
469 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
470 int dxy, uvdxy, mx, my, src_x, src_y, uvsrc_x, uvsrc_y, v_edge_pos, linesize, uvlinesize;
472 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
473 src_x = s->mb_x * 16 + (motion_x >> 2);
474 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2);
476 v_edge_pos = s->v_edge_pos >> field_based;
477 linesize = s->linesize << field_based;
478 uvlinesize = s->uvlinesize << field_based;
483 }else if(s->workaround_bugs&FF_BUG_QPEL_CHROMA2){
484 static const int rtab[8]= {0,0,1,1,0,0,0,1};
485 mx= (motion_x>>1) + rtab[motion_x&7];
486 my= (motion_y>>1) + rtab[motion_y&7];
487 }else if(s->workaround_bugs&FF_BUG_QPEL_CHROMA){
488 mx= (motion_x>>1)|(motion_x&1);
489 my= (motion_y>>1)|(motion_y&1);
497 uvdxy= (mx&1) | ((my&1)<<1);
501 uvsrc_x = s->mb_x * 8 + mx;
502 uvsrc_y = s->mb_y * (8 >> field_based) + my;
504 ptr_y = ref_picture[0] + src_y * linesize + src_x;
505 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
506 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
508 if( (unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x&3) - 16, 0)
509 || (unsigned)src_y > FFMAX( v_edge_pos - (motion_y&3) - h , 0)){
510 s->dsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y, s->linesize,
511 17, 17+field_based, src_x, src_y<<field_based,
512 s->h_edge_pos, s->v_edge_pos);
513 ptr_y= s->edge_emu_buffer;
514 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
515 uint8_t *uvbuf= s->edge_emu_buffer + 18*s->linesize;
516 s->dsp.emulated_edge_mc(uvbuf, ptr_cb, s->uvlinesize,
518 uvsrc_x, uvsrc_y<<field_based,
519 s->h_edge_pos>>1, s->v_edge_pos>>1);
520 s->dsp.emulated_edge_mc(uvbuf + 16, ptr_cr, s->uvlinesize,
522 uvsrc_x, uvsrc_y<<field_based,
523 s->h_edge_pos>>1, s->v_edge_pos>>1);
530 qpix_op[0][dxy](dest_y, ptr_y, linesize);
533 dest_y += s->linesize;
534 dest_cb+= s->uvlinesize;
535 dest_cr+= s->uvlinesize;
539 ptr_y += s->linesize;
540 ptr_cb += s->uvlinesize;
541 ptr_cr += s->uvlinesize;
543 //damn interlaced mode
544 //FIXME boundary mirroring is not exactly correct here
545 qpix_op[1][dxy](dest_y , ptr_y , linesize);
546 qpix_op[1][dxy](dest_y+8, ptr_y+8, linesize);
548 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
549 pix_op[1][uvdxy](dest_cr, ptr_cr, uvlinesize, h >> 1);
550 pix_op[1][uvdxy](dest_cb, ptr_cb, uvlinesize, h >> 1);
555 * h263 chroma 4mv motion compensation.
557 static inline void chroma_4mv_motion(MpegEncContext *s,
558 uint8_t *dest_cb, uint8_t *dest_cr,
559 uint8_t **ref_picture,
560 op_pixels_func *pix_op,
562 int dxy, emu=0, src_x, src_y, offset;
565 /* In case of 8X8, we construct a single chroma motion vector
566 with a special rounding */
567 mx= ff_h263_round_chroma(mx);
568 my= ff_h263_round_chroma(my);
570 dxy = ((my & 1) << 1) | (mx & 1);
574 src_x = s->mb_x * 8 + mx;
575 src_y = s->mb_y * 8 + my;
576 src_x = av_clip(src_x, -8, (s->width >> 1));
577 if (src_x == (s->width >> 1))
579 src_y = av_clip(src_y, -8, (s->height >> 1));
580 if (src_y == (s->height >> 1))
583 offset = src_y * s->uvlinesize + src_x;
584 ptr = ref_picture[1] + offset;
585 if(s->flags&CODEC_FLAG_EMU_EDGE){
586 if( (unsigned)src_x > FFMAX((s->h_edge_pos>>1) - (dxy &1) - 8, 0)
587 || (unsigned)src_y > FFMAX((s->v_edge_pos>>1) - (dxy>>1) - 8, 0)){
588 s->dsp.emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize,
590 s->h_edge_pos>>1, s->v_edge_pos>>1);
591 ptr= s->edge_emu_buffer;
595 pix_op[dxy](dest_cb, ptr, s->uvlinesize, 8);
597 ptr = ref_picture[2] + offset;
599 s->dsp.emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize,
601 s->h_edge_pos>>1, s->v_edge_pos>>1);
602 ptr= s->edge_emu_buffer;
604 pix_op[dxy](dest_cr, ptr, s->uvlinesize, 8);
607 static inline void prefetch_motion(MpegEncContext *s, uint8_t **pix, int dir){
608 /* fetch pixels for estimated mv 4 macroblocks ahead
609 * optimized for 64byte cache lines */
610 const int shift = s->quarter_sample ? 2 : 1;
611 const int mx= (s->mv[dir][0][0]>>shift) + 16*s->mb_x + 8;
612 const int my= (s->mv[dir][0][1]>>shift) + 16*s->mb_y;
613 int off= mx + (my + (s->mb_x&3)*4)*s->linesize + 64;
614 s->dsp.prefetch(pix[0]+off, s->linesize, 4);
615 off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
616 s->dsp.prefetch(pix[1]+off, pix[2]-pix[1], 2);
620 * motion compensation of a single macroblock
622 * @param dest_y luma destination pointer
623 * @param dest_cb chroma cb/u destination pointer
624 * @param dest_cr chroma cr/v destination pointer
625 * @param dir direction (0->forward, 1->backward)
626 * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
627 * @param pix_op halfpel motion compensation function (average or put normally)
628 * @param qpix_op qpel motion compensation function (average or put normally)
629 * the motion vectors are taken from s->mv and the MV type from s->mv_type
631 static av_always_inline void MPV_motion_internal(MpegEncContext *s,
632 uint8_t *dest_y, uint8_t *dest_cb,
633 uint8_t *dest_cr, int dir,
634 uint8_t **ref_picture,
635 op_pixels_func (*pix_op)[4],
636 qpel_mc_func (*qpix_op)[16], int is_mpeg12)
638 int dxy, mx, my, src_x, src_y, motion_x, motion_y;
645 prefetch_motion(s, ref_picture, dir);
647 if(!is_mpeg12 && s->obmc && s->pict_type != AV_PICTURE_TYPE_B){
648 int16_t mv_cache[4][4][2];
649 const int xy= s->mb_x + s->mb_y*s->mb_stride;
650 const int mot_stride= s->b8_stride;
651 const int mot_xy= mb_x*2 + mb_y*2*mot_stride;
653 av_assert2(!s->mb_skipped);
655 memcpy(mv_cache[1][1], s->current_picture.f.motion_val[0][mot_xy ], sizeof(int16_t) * 4);
656 memcpy(mv_cache[2][1], s->current_picture.f.motion_val[0][mot_xy + mot_stride], sizeof(int16_t) * 4);
657 memcpy(mv_cache[3][1], s->current_picture.f.motion_val[0][mot_xy + mot_stride], sizeof(int16_t) * 4);
659 if (mb_y == 0 || IS_INTRA(s->current_picture.f.mb_type[xy - s->mb_stride])) {
660 memcpy(mv_cache[0][1], mv_cache[1][1], sizeof(int16_t)*4);
662 memcpy(mv_cache[0][1], s->current_picture.f.motion_val[0][mot_xy - mot_stride], sizeof(int16_t) * 4);
665 if (mb_x == 0 || IS_INTRA(s->current_picture.f.mb_type[xy - 1])) {
666 AV_COPY32(mv_cache[1][0], mv_cache[1][1]);
667 AV_COPY32(mv_cache[2][0], mv_cache[2][1]);
669 AV_COPY32(mv_cache[1][0], s->current_picture.f.motion_val[0][mot_xy - 1]);
670 AV_COPY32(mv_cache[2][0], s->current_picture.f.motion_val[0][mot_xy - 1 + mot_stride]);
673 if (mb_x + 1 >= s->mb_width || IS_INTRA(s->current_picture.f.mb_type[xy + 1])) {
674 AV_COPY32(mv_cache[1][3], mv_cache[1][2]);
675 AV_COPY32(mv_cache[2][3], mv_cache[2][2]);
677 AV_COPY32(mv_cache[1][3], s->current_picture.f.motion_val[0][mot_xy + 2]);
678 AV_COPY32(mv_cache[2][3], s->current_picture.f.motion_val[0][mot_xy + 2 + mot_stride]);
684 const int x= (i&1)+1;
685 const int y= (i>>1)+1;
687 {mv_cache[y][x ][0], mv_cache[y][x ][1]},
688 {mv_cache[y-1][x][0], mv_cache[y-1][x][1]},
689 {mv_cache[y][x-1][0], mv_cache[y][x-1][1]},
690 {mv_cache[y][x+1][0], mv_cache[y][x+1][1]},
691 {mv_cache[y+1][x][0], mv_cache[y+1][x][1]}};
693 obmc_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
695 mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >>1) * 8,
702 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY))
703 chroma_4mv_motion(s, dest_cb, dest_cr, ref_picture, pix_op[1], mx, my);
711 if(s->real_sprite_warping_points==1){
712 gmc1_motion(s, dest_y, dest_cb, dest_cr,
715 gmc_motion(s, dest_y, dest_cb, dest_cr,
718 }else if(!is_mpeg12 && s->quarter_sample){
719 qpel_motion(s, dest_y, dest_cb, dest_cr,
721 ref_picture, pix_op, qpix_op,
722 s->mv[dir][0][0], s->mv[dir][0][1], 16);
723 }else if(!is_mpeg12 && (CONFIG_WMV2_DECODER || CONFIG_WMV2_ENCODER) && s->mspel && s->codec_id == CODEC_ID_WMV2){
724 ff_mspel_motion(s, dest_y, dest_cb, dest_cr,
726 s->mv[dir][0][0], s->mv[dir][0][1], 16);
729 mpeg_motion(s, dest_y, dest_cb, dest_cr,
732 s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y);
739 if(s->quarter_sample){
741 motion_x = s->mv[dir][i][0];
742 motion_y = s->mv[dir][i][1];
744 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
745 src_x = mb_x * 16 + (motion_x >> 2) + (i & 1) * 8;
746 src_y = mb_y * 16 + (motion_y >> 2) + (i >>1) * 8;
748 /* WARNING: do no forget half pels */
749 src_x = av_clip(src_x, -16, s->width);
750 if (src_x == s->width)
752 src_y = av_clip(src_y, -16, s->height);
753 if (src_y == s->height)
756 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x);
757 if(s->flags&CODEC_FLAG_EMU_EDGE){
758 if( (unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x&3) - 8, 0)
759 || (unsigned)src_y > FFMAX(s->v_edge_pos - (motion_y&3) - 8, 0)){
760 s->dsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
763 s->h_edge_pos, s->v_edge_pos);
764 ptr= s->edge_emu_buffer;
767 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize;
768 qpix_op[1][dxy](dest, ptr, s->linesize);
770 mx += s->mv[dir][i][0]/2;
771 my += s->mv[dir][i][1]/2;
775 hpel_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
776 ref_picture[0], 0, 0,
777 mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >>1) * 8,
778 s->width, s->height, s->linesize,
779 s->h_edge_pos, s->v_edge_pos,
781 s->mv[dir][i][0], s->mv[dir][i][1]);
783 mx += s->mv[dir][i][0];
784 my += s->mv[dir][i][1];
788 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY))
789 chroma_4mv_motion(s, dest_cb, dest_cr, ref_picture, pix_op[1], mx, my);
793 if (s->picture_structure == PICT_FRAME) {
794 if(!is_mpeg12 && s->quarter_sample){
796 qpel_motion(s, dest_y, dest_cb, dest_cr,
797 1, i, s->field_select[dir][i],
798 ref_picture, pix_op, qpix_op,
799 s->mv[dir][i][0], s->mv[dir][i][1], 8);
803 mpeg_motion(s, dest_y, dest_cb, dest_cr,
804 1, 0, s->field_select[dir][0],
806 s->mv[dir][0][0], s->mv[dir][0][1], 8, mb_y);
808 mpeg_motion(s, dest_y, dest_cb, dest_cr,
809 1, 1, s->field_select[dir][1],
811 s->mv[dir][1][0], s->mv[dir][1][1], 8, mb_y);
814 if(s->picture_structure != s->field_select[dir][0] + 1 && s->pict_type != AV_PICTURE_TYPE_B && !s->first_field){
815 ref_picture = s->current_picture_ptr->f.data;
818 mpeg_motion(s, dest_y, dest_cb, dest_cr,
819 0, 0, s->field_select[dir][0],
821 s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y>>1);
826 uint8_t ** ref2picture;
828 if(s->picture_structure == s->field_select[dir][i] + 1
829 || s->pict_type == AV_PICTURE_TYPE_B || s->first_field){
830 ref2picture= ref_picture;
832 ref2picture = s->current_picture_ptr->f.data;
835 mpeg_motion(s, dest_y, dest_cb, dest_cr,
836 0, 0, s->field_select[dir][i],
838 s->mv[dir][i][0], s->mv[dir][i][1] + 16*i, 8, mb_y>>1);
840 dest_y += 16*s->linesize;
841 dest_cb+= (16>>s->chroma_y_shift)*s->uvlinesize;
842 dest_cr+= (16>>s->chroma_y_shift)*s->uvlinesize;
846 if(s->picture_structure == PICT_FRAME){
850 mpeg_motion(s, dest_y, dest_cb, dest_cr,
853 s->mv[dir][2*i + j][0], s->mv[dir][2*i + j][1], 8, mb_y);
855 pix_op = s->dsp.avg_pixels_tab;
859 mpeg_motion(s, dest_y, dest_cb, dest_cr,
860 0, 0, s->picture_structure != i+1,
862 s->mv[dir][2*i][0],s->mv[dir][2*i][1],16, mb_y>>1);
864 // after put we make avg of the same block
865 pix_op=s->dsp.avg_pixels_tab;
867 //opposite parity is always in the same frame if this is second field
869 ref_picture = s->current_picture_ptr->f.data;
874 default: av_assert2(0);
878 static inline void MPV_motion(MpegEncContext *s,
879 uint8_t *dest_y, uint8_t *dest_cb,
880 uint8_t *dest_cr, int dir,
881 uint8_t **ref_picture,
882 op_pixels_func (*pix_op)[4],
883 qpel_mc_func (*qpix_op)[16])
886 if(s->out_format == FMT_MPEG1)
887 MPV_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
888 ref_picture, pix_op, qpix_op, 1);
891 MPV_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
892 ref_picture, pix_op, qpix_op, 0);
894 #endif /* AVCODEC_MPEGVIDEO_COMMON_H */