?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Use add_*flags only after compiler-specific configuration
[FFMpeg-mirror/ordered_chapters.git] / libavcodec / mpegvideo_common.h
blob45b84e028a1c8a608f0eeb376696ddf7d84960d1
1 /*
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>
5 *
6 * 4MV & hq & B-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
7 *
8 * This file is part of FFmpeg.
9 *
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.
14 *
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.
19 *
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
23 */
24
25 /**
26 * @file libavcodec/mpegvideo_common.h
27 * The simplest mpeg encoder (well, it was the simplest!).
28 */
29
30 #ifndef AVCODEC_MPEGVIDEO_COMMON_H
31 #define AVCODEC_MPEGVIDEO_COMMON_H
32
33 #include <string.h>
34 #include "avcodec.h"
35 #include "dsputil.h"
36 #include "mpegvideo.h"
37 #include "mjpegenc.h"
38 #include "msmpeg4.h"
39 #include "faandct.h"
40 #include <limits.h>
41
42 int dct_quantize_c(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow);
43 int dct_quantize_trellis_c(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow);
44 void denoise_dct_c(MpegEncContext *s, DCTELEM *block);
45
46 /**
47 * allocates a Picture
48 * The pixels are allocated/set by calling get_buffer() if shared=0
49 */
50 int alloc_picture(MpegEncContext *s, Picture *pic, int shared);
51
52 /**
53 * sets the given MpegEncContext to common defaults (same for encoding and decoding).
54 * the changed fields will not depend upon the prior state of the MpegEncContext.
55 */
56 void MPV_common_defaults(MpegEncContext *s);
57
58 static inline void gmc1_motion(MpegEncContext *s,
59 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
60 uint8_t **ref_picture)
61 {
62 uint8_t *ptr;
63 int offset, src_x, src_y, linesize, uvlinesize;
64 int motion_x, motion_y;
65 int emu=0;
66
67 motion_x= s->sprite_offset[0][0];
68 motion_y= s->sprite_offset[0][1];
69 src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy+1));
70 src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy+1));
71 motion_x<<=(3-s->sprite_warping_accuracy);
72 motion_y<<=(3-s->sprite_warping_accuracy);
73 src_x = av_clip(src_x, -16, s->width);
74 if (src_x == s->width)
75 motion_x =0;
76 src_y = av_clip(src_y, -16, s->height);
77 if (src_y == s->height)
78 motion_y =0;
79
80 linesize = s->linesize;
81 uvlinesize = s->uvlinesize;
82
83 ptr = ref_picture[0] + (src_y * linesize) + src_x;
84
85 if(s->flags&CODEC_FLAG_EMU_EDGE){
86 if( (unsigned)src_x >= s->h_edge_pos - 17
87 || (unsigned)src_y >= s->v_edge_pos - 17){
88 ff_emulated_edge_mc(s->edge_emu_buffer, ptr, linesize, 17, 17, src_x, src_y, s->h_edge_pos, s->v_edge_pos);
89 ptr= s->edge_emu_buffer;
90 }
91 }
92
93 if((motion_x|motion_y)&7){
94 s->dsp.gmc1(dest_y , ptr , linesize, 16, motion_x&15, motion_y&15, 128 - s->no_rounding);
95 s->dsp.gmc1(dest_y+8, ptr+8, linesize, 16, motion_x&15, motion_y&15, 128 - s->no_rounding);
96 }else{
97 int dxy;
98
99 dxy= ((motion_x>>3)&1) | ((motion_y>>2)&2);
100 if (s->no_rounding){
101 s->dsp.put_no_rnd_pixels_tab[0][dxy](dest_y, ptr, linesize, 16);
102 }else{
103 s->dsp.put_pixels_tab [0][dxy](dest_y, ptr, linesize, 16);
104 }
105 }
106
107 if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
108
109 motion_x= s->sprite_offset[1][0];
110 motion_y= s->sprite_offset[1][1];
111 src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy+1));
112 src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy+1));
113 motion_x<<=(3-s->sprite_warping_accuracy);
114 motion_y<<=(3-s->sprite_warping_accuracy);
115 src_x = av_clip(src_x, -8, s->width>>1);
116 if (src_x == s->width>>1)
117 motion_x =0;
118 src_y = av_clip(src_y, -8, s->height>>1);
119 if (src_y == s->height>>1)
120 motion_y =0;
121
122 offset = (src_y * uvlinesize) + src_x;
123 ptr = ref_picture[1] + offset;
124 if(s->flags&CODEC_FLAG_EMU_EDGE){
125 if( (unsigned)src_x >= (s->h_edge_pos>>1) - 9
126 || (unsigned)src_y >= (s->v_edge_pos>>1) - 9){
127 ff_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;
129 emu=1;
130 }
131 }
132 s->dsp.gmc1(dest_cb, ptr, uvlinesize, 8, motion_x&15, motion_y&15, 128 - s->no_rounding);
133
134 ptr = ref_picture[2] + offset;
135 if(emu){
136 ff_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);
137 ptr= s->edge_emu_buffer;
138 }
139 s->dsp.gmc1(dest_cr, ptr, uvlinesize, 8, motion_x&15, motion_y&15, 128 - s->no_rounding);
140
141 return;
142 }
143
144 static inline void gmc_motion(MpegEncContext *s,
145 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
146 uint8_t **ref_picture)
147 {
148 uint8_t *ptr;
149 int linesize, uvlinesize;
150 const int a= s->sprite_warping_accuracy;
151 int ox, oy;
152
153 linesize = s->linesize;
154 uvlinesize = s->uvlinesize;
155
156 ptr = ref_picture[0];
157
158 ox= s->sprite_offset[0][0] + s->sprite_delta[0][0]*s->mb_x*16 + s->sprite_delta[0][1]*s->mb_y*16;
159 oy= s->sprite_offset[0][1] + s->sprite_delta[1][0]*s->mb_x*16 + s->sprite_delta[1][1]*s->mb_y*16;
160
161 s->dsp.gmc(dest_y, ptr, linesize, 16,
162 ox,
163 oy,
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);
168 s->dsp.gmc(dest_y+8, ptr, linesize, 16,
169 ox + s->sprite_delta[0][0]*8,
170 oy + s->sprite_delta[1][0]*8,
171 s->sprite_delta[0][0], s->sprite_delta[0][1],
172 s->sprite_delta[1][0], s->sprite_delta[1][1],
173 a+1, (1<<(2*a+1)) - s->no_rounding,
174 s->h_edge_pos, s->v_edge_pos);
175
176 if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
177
178 ox= s->sprite_offset[1][0] + s->sprite_delta[0][0]*s->mb_x*8 + s->sprite_delta[0][1]*s->mb_y*8;
179 oy= s->sprite_offset[1][1] + s->sprite_delta[1][0]*s->mb_x*8 + s->sprite_delta[1][1]*s->mb_y*8;
180
181 ptr = ref_picture[1];
182 s->dsp.gmc(dest_cb, ptr, uvlinesize, 8,
183 ox,
184 oy,
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);
189
190 ptr = ref_picture[2];
191 s->dsp.gmc(dest_cr, ptr, uvlinesize, 8,
192 ox,
193 oy,
194 s->sprite_delta[0][0], s->sprite_delta[0][1],
195 s->sprite_delta[1][0], s->sprite_delta[1][1],
196 a+1, (1<<(2*a+1)) - s->no_rounding,
197 s->h_edge_pos>>1, s->v_edge_pos>>1);
198 }
199
200 static inline int hpel_motion(MpegEncContext *s,
201 uint8_t *dest, uint8_t *src,
202 int field_based, int field_select,
203 int src_x, int src_y,
204 int width, int height, int stride,
205 int h_edge_pos, int v_edge_pos,
206 int w, int h, op_pixels_func *pix_op,
207 int motion_x, int motion_y)
208 {
209 int dxy;
210 int emu=0;
211
212 dxy = ((motion_y & 1) << 1) | (motion_x & 1);
213 src_x += motion_x >> 1;
214 src_y += motion_y >> 1;
215
216 /* WARNING: do no forget half pels */
217 src_x = av_clip(src_x, -16, width); //FIXME unneeded for emu?
218 if (src_x == width)
219 dxy &= ~1;
220 src_y = av_clip(src_y, -16, height);
221 if (src_y == height)
222 dxy &= ~2;
223 src += src_y * stride + src_x;
224
225 if(s->unrestricted_mv && (s->flags&CODEC_FLAG_EMU_EDGE)){
226 if( (unsigned)src_x > h_edge_pos - (motion_x&1) - w
227 || (unsigned)src_y > v_edge_pos - (motion_y&1) - h){
228 ff_emulated_edge_mc(s->edge_emu_buffer, src, s->linesize, w+1, (h+1)<<field_based,
229 src_x, src_y<<field_based, h_edge_pos, s->v_edge_pos);
230 src= s->edge_emu_buffer;
231 emu=1;
232 }
233 }
234 if(field_select)
235 src += s->linesize;
236 pix_op[dxy](dest, src, stride, h);
237 return emu;
238 }
239
240 static av_always_inline
241 void mpeg_motion_internal(MpegEncContext *s,
242 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
243 int field_based, int bottom_field, int field_select,
244 uint8_t **ref_picture, op_pixels_func (*pix_op)[4],
245 int motion_x, int motion_y, int h, int is_mpeg12)
246 {
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, uvlinesize, linesize;
250
251 #if 0
252 if(s->quarter_sample)
253 {
254 motion_x>>=1;
255 motion_y>>=1;
256 }
257 #endif
258
259 v_edge_pos = s->v_edge_pos >> field_based;
260 linesize = s->current_picture.linesize[0] << field_based;
261 uvlinesize = s->current_picture.linesize[1] << field_based;
262
263 dxy = ((motion_y & 1) << 1) | (motion_x & 1);
264 src_x = s->mb_x* 16 + (motion_x >> 1);
265 src_y =(s->mb_y<<(4-field_based)) + (motion_y >> 1);
266
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);
270 my = motion_y >>1;
271 uvdxy = ((my & 1) << 1) | (mx & 1);
272 uvsrc_x = s->mb_x* 8 + (mx >> 1);
273 uvsrc_y = (s->mb_y<<(3-field_based)) + (my >> 1);
274 }else{
275 uvdxy = dxy | (motion_y & 2) | ((motion_x & 2) >> 1);
276 uvsrc_x = src_x>>1;
277 uvsrc_y = src_y>>1;
278 }
279 }else if(!is_mpeg12 && s->out_format == FMT_H261){//even chroma mv's are full pel in H261
280 mx = motion_x / 4;
281 my = motion_y / 4;
282 uvdxy = 0;
283 uvsrc_x = s->mb_x*8 + mx;
284 uvsrc_y = s->mb_y*8 + my;
285 } else {
286 if(s->chroma_y_shift){
287 mx = motion_x / 2;
288 my = motion_y / 2;
289 uvdxy = ((my & 1) << 1) | (mx & 1);
290 uvsrc_x = s->mb_x* 8 + (mx >> 1);
291 uvsrc_y = (s->mb_y<<(3-field_based)) + (my >> 1);
292 } else {
293 if(s->chroma_x_shift){
294 //Chroma422
295 mx = motion_x / 2;
296 uvdxy = ((motion_y & 1) << 1) | (mx & 1);
297 uvsrc_x = s->mb_x* 8 + (mx >> 1);
298 uvsrc_y = src_y;
299 } else {
300 //Chroma444
301 uvdxy = dxy;
302 uvsrc_x = src_x;
303 uvsrc_y = src_y;
304 }
305 }
306 }
307
308 ptr_y = ref_picture[0] + src_y * linesize + src_x;
309 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
310 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
311
312 if( (unsigned)src_x > s->h_edge_pos - (motion_x&1) - 16
313 || (unsigned)src_y > v_edge_pos - (motion_y&1) - h){
314 if(is_mpeg12 || s->codec_id == CODEC_ID_MPEG2VIDEO ||
315 s->codec_id == CODEC_ID_MPEG1VIDEO){
316 av_log(s->avctx,AV_LOG_DEBUG,
317 "MPEG motion vector out of boundary\n");
318 if(!s->chroma_y_shift)
319 return;
320 }
321 ff_emulated_edge_mc(s->edge_emu_buffer, ptr_y, s->linesize,
322 17, 17+field_based,
323 src_x, src_y<<field_based,
324 s->h_edge_pos, s->v_edge_pos);
325 ptr_y = s->edge_emu_buffer;
326 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
327 uint8_t *uvbuf= s->edge_emu_buffer+18*s->linesize;
328 ff_emulated_edge_mc(uvbuf ,
329 ptr_cb, s->uvlinesize,
330 9, 9+field_based,
331 uvsrc_x, uvsrc_y<<field_based,
332 s->h_edge_pos>>1, s->v_edge_pos>>1);
333 ff_emulated_edge_mc(uvbuf+16,
334 ptr_cr, s->uvlinesize,
335 9, 9+field_based,
336 uvsrc_x, uvsrc_y<<field_based,
337 s->h_edge_pos>>1, s->v_edge_pos>>1);
338 ptr_cb= uvbuf;
339 ptr_cr= uvbuf+16;
340 }
341 }
342
343 if(bottom_field){ //FIXME use this for field pix too instead of the obnoxious hack which changes picture.data
344 dest_y += s->linesize;
345 dest_cb+= s->uvlinesize;
346 dest_cr+= s->uvlinesize;
347 }
348
349 if(field_select){
350 ptr_y += s->linesize;
351 ptr_cb+= s->uvlinesize;
352 ptr_cr+= s->uvlinesize;
353 }
354
355 pix_op[0][dxy](dest_y, ptr_y, linesize, h);
356
357 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
358 pix_op[s->chroma_x_shift][uvdxy]
359 (dest_cb, ptr_cb, uvlinesize, h >> s->chroma_y_shift);
360 pix_op[s->chroma_x_shift][uvdxy]
361 (dest_cr, ptr_cr, uvlinesize, h >> s->chroma_y_shift);
362 }
363 if(!is_mpeg12 && (CONFIG_H261_ENCODER || CONFIG_H261_DECODER) &&
364 s->out_format == FMT_H261){
365 ff_h261_loop_filter(s);
366 }
367 }
368 /* apply one mpeg motion vector to the three components */
369 static av_always_inline
370 void mpeg_motion(MpegEncContext *s,
371 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
372 int field_based, int bottom_field, int field_select,
373 uint8_t **ref_picture, op_pixels_func (*pix_op)[4],
374 int motion_x, int motion_y, int h)
375 {
376 #if !CONFIG_SMALL
377 if(s->out_format == FMT_MPEG1)
378 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, field_based,
379 bottom_field, field_select, ref_picture, pix_op,
380 motion_x, motion_y, h, 1);
381 else
382 #endif
383 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, field_based,
384 bottom_field, field_select, ref_picture, pix_op,
385 motion_x, motion_y, h, 0);
386 }
387
388 //FIXME move to dsputil, avg variant, 16x16 version
389 static inline void put_obmc(uint8_t *dst, uint8_t *src[5], int stride){
390 int x;
391 uint8_t * const top = src[1];
392 uint8_t * const left = src[2];
393 uint8_t * const mid = src[0];
394 uint8_t * const right = src[3];
395 uint8_t * const bottom= src[4];
396 #define OBMC_FILTER(x, t, l, m, r, b)\
397 dst[x]= (t*top[x] + l*left[x] + m*mid[x] + r*right[x] + b*bottom[x] + 4)>>3
398 #define OBMC_FILTER4(x, t, l, m, r, b)\
399 OBMC_FILTER(x , t, l, m, r, b);\
400 OBMC_FILTER(x+1 , t, l, m, r, b);\
401 OBMC_FILTER(x +stride, t, l, m, r, b);\
402 OBMC_FILTER(x+1+stride, t, l, m, r, b);
403
404 x=0;
405 OBMC_FILTER (x , 2, 2, 4, 0, 0);
406 OBMC_FILTER (x+1, 2, 1, 5, 0, 0);
407 OBMC_FILTER4(x+2, 2, 1, 5, 0, 0);
408 OBMC_FILTER4(x+4, 2, 0, 5, 1, 0);
409 OBMC_FILTER (x+6, 2, 0, 5, 1, 0);
410 OBMC_FILTER (x+7, 2, 0, 4, 2, 0);
411 x+= stride;
412 OBMC_FILTER (x , 1, 2, 5, 0, 0);
413 OBMC_FILTER (x+1, 1, 2, 5, 0, 0);
414 OBMC_FILTER (x+6, 1, 0, 5, 2, 0);
415 OBMC_FILTER (x+7, 1, 0, 5, 2, 0);
416 x+= stride;
417 OBMC_FILTER4(x , 1, 2, 5, 0, 0);
418 OBMC_FILTER4(x+2, 1, 1, 6, 0, 0);
419 OBMC_FILTER4(x+4, 1, 0, 6, 1, 0);
420 OBMC_FILTER4(x+6, 1, 0, 5, 2, 0);
421 x+= 2*stride;
422 OBMC_FILTER4(x , 0, 2, 5, 0, 1);
423 OBMC_FILTER4(x+2, 0, 1, 6, 0, 1);
424 OBMC_FILTER4(x+4, 0, 0, 6, 1, 1);
425 OBMC_FILTER4(x+6, 0, 0, 5, 2, 1);
426 x+= 2*stride;
427 OBMC_FILTER (x , 0, 2, 5, 0, 1);
428 OBMC_FILTER (x+1, 0, 2, 5, 0, 1);
429 OBMC_FILTER4(x+2, 0, 1, 5, 0, 2);
430 OBMC_FILTER4(x+4, 0, 0, 5, 1, 2);
431 OBMC_FILTER (x+6, 0, 0, 5, 2, 1);
432 OBMC_FILTER (x+7, 0, 0, 5, 2, 1);
433 x+= stride;
434 OBMC_FILTER (x , 0, 2, 4, 0, 2);
435 OBMC_FILTER (x+1, 0, 1, 5, 0, 2);
436 OBMC_FILTER (x+6, 0, 0, 5, 1, 2);
437 OBMC_FILTER (x+7, 0, 0, 4, 2, 2);
438 }
439
440 /* obmc for 1 8x8 luma block */
441 static inline void obmc_motion(MpegEncContext *s,
442 uint8_t *dest, uint8_t *src,
443 int src_x, int src_y,
444 op_pixels_func *pix_op,
445 int16_t mv[5][2]/* mid top left right bottom*/)
446 #define MID 0
447 {
448 int i;
449 uint8_t *ptr[5];
450
451 assert(s->quarter_sample==0);
452
453 for(i=0; i<5; i++){
454 if(i && mv[i][0]==mv[MID][0] && mv[i][1]==mv[MID][1]){
455 ptr[i]= ptr[MID];
456 }else{
457 ptr[i]= s->obmc_scratchpad + 8*(i&1) + s->linesize*8*(i>>1);
458 hpel_motion(s, ptr[i], src, 0, 0,
459 src_x, src_y,
460 s->width, s->height, s->linesize,
461 s->h_edge_pos, s->v_edge_pos,
462 8, 8, pix_op,
463 mv[i][0], mv[i][1]);
464 }
465 }
466
467 put_obmc(dest, ptr, s->linesize);
468 }
469
470 static inline void qpel_motion(MpegEncContext *s,
471 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
472 int field_based, int bottom_field, int field_select,
473 uint8_t **ref_picture, op_pixels_func (*pix_op)[4],
474 qpel_mc_func (*qpix_op)[16],
475 int motion_x, int motion_y, int h)
476 {
477 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
478 int dxy, uvdxy, mx, my, src_x, src_y, uvsrc_x, uvsrc_y, v_edge_pos, linesize, uvlinesize;
479
480 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
481 src_x = s->mb_x * 16 + (motion_x >> 2);
482 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2);
483
484 v_edge_pos = s->v_edge_pos >> field_based;
485 linesize = s->linesize << field_based;
486 uvlinesize = s->uvlinesize << field_based;
487
488 if(field_based){
489 mx= motion_x/2;
490 my= motion_y>>1;
491 }else if(s->workaround_bugs&FF_BUG_QPEL_CHROMA2){
492 static const int rtab[8]= {0,0,1,1,0,0,0,1};
493 mx= (motion_x>>1) + rtab[motion_x&7];
494 my= (motion_y>>1) + rtab[motion_y&7];
495 }else if(s->workaround_bugs&FF_BUG_QPEL_CHROMA){
496 mx= (motion_x>>1)|(motion_x&1);
497 my= (motion_y>>1)|(motion_y&1);
498 }else{
499 mx= motion_x/2;
500 my= motion_y/2;
501 }
502 mx= (mx>>1)|(mx&1);
503 my= (my>>1)|(my&1);
504
505 uvdxy= (mx&1) | ((my&1)<<1);
506 mx>>=1;
507 my>>=1;
508
509 uvsrc_x = s->mb_x * 8 + mx;
510 uvsrc_y = s->mb_y * (8 >> field_based) + my;
511
512 ptr_y = ref_picture[0] + src_y * linesize + src_x;
513 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
514 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
515
516 if( (unsigned)src_x > s->h_edge_pos - (motion_x&3) - 16
517 || (unsigned)src_y > v_edge_pos - (motion_y&3) - h ){
518 ff_emulated_edge_mc(s->edge_emu_buffer, ptr_y, s->linesize,
519 17, 17+field_based, src_x, src_y<<field_based,
520 s->h_edge_pos, s->v_edge_pos);
521 ptr_y= s->edge_emu_buffer;
522 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
523 uint8_t *uvbuf= s->edge_emu_buffer + 18*s->linesize;
524 ff_emulated_edge_mc(uvbuf, ptr_cb, s->uvlinesize,
525 9, 9 + field_based,
526 uvsrc_x, uvsrc_y<<field_based,
527 s->h_edge_pos>>1, s->v_edge_pos>>1);
528 ff_emulated_edge_mc(uvbuf + 16, ptr_cr, s->uvlinesize,
529 9, 9 + field_based,
530 uvsrc_x, uvsrc_y<<field_based,
531 s->h_edge_pos>>1, s->v_edge_pos>>1);
532 ptr_cb= uvbuf;
533 ptr_cr= uvbuf + 16;
534 }
535 }
536
537 if(!field_based)
538 qpix_op[0][dxy](dest_y, ptr_y, linesize);
539 else{
540 if(bottom_field){
541 dest_y += s->linesize;
542 dest_cb+= s->uvlinesize;
543 dest_cr+= s->uvlinesize;
544 }
545
546 if(field_select){
547 ptr_y += s->linesize;
548 ptr_cb += s->uvlinesize;
549 ptr_cr += s->uvlinesize;
550 }
551 //damn interlaced mode
552 //FIXME boundary mirroring is not exactly correct here
553 qpix_op[1][dxy](dest_y , ptr_y , linesize);
554 qpix_op[1][dxy](dest_y+8, ptr_y+8, linesize);
555 }
556 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
557 pix_op[1][uvdxy](dest_cr, ptr_cr, uvlinesize, h >> 1);
558 pix_op[1][uvdxy](dest_cb, ptr_cb, uvlinesize, h >> 1);
559 }
560 }
561
562 /**
563 * h263 chroma 4mv motion compensation.
564 */
565 static inline void chroma_4mv_motion(MpegEncContext *s,
566 uint8_t *dest_cb, uint8_t *dest_cr,
567 uint8_t **ref_picture,
568 op_pixels_func *pix_op,
569 int mx, int my){
570 int dxy, emu=0, src_x, src_y, offset;
571 uint8_t *ptr;
572
573 /* In case of 8X8, we construct a single chroma motion vector
574 with a special rounding */
575 mx= ff_h263_round_chroma(mx);
576 my= ff_h263_round_chroma(my);
577
578 dxy = ((my & 1) << 1) | (mx & 1);
579 mx >>= 1;
580 my >>= 1;
581
582 src_x = s->mb_x * 8 + mx;
583 src_y = s->mb_y * 8 + my;
584 src_x = av_clip(src_x, -8, s->width/2);
585 if (src_x == s->width/2)
586 dxy &= ~1;
587 src_y = av_clip(src_y, -8, s->height/2);
588 if (src_y == s->height/2)
589 dxy &= ~2;
590
591 offset = (src_y * (s->uvlinesize)) + src_x;
592 ptr = ref_picture[1] + offset;
593 if(s->flags&CODEC_FLAG_EMU_EDGE){
594 if( (unsigned)src_x > (s->h_edge_pos>>1) - (dxy &1) - 8
595 || (unsigned)src_y > (s->v_edge_pos>>1) - (dxy>>1) - 8){
596 ff_emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize,
597 9, 9, src_x, src_y,
598 s->h_edge_pos>>1, s->v_edge_pos>>1);
599 ptr= s->edge_emu_buffer;
600 emu=1;
601 }
602 }
603 pix_op[dxy](dest_cb, ptr, s->uvlinesize, 8);
604
605 ptr = ref_picture[2] + offset;
606 if(emu){
607 ff_emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize,
608 9, 9, src_x, src_y,
609 s->h_edge_pos>>1, s->v_edge_pos>>1);
610 ptr= s->edge_emu_buffer;
611 }
612 pix_op[dxy](dest_cr, ptr, s->uvlinesize, 8);
613 }
614
615 static inline void prefetch_motion(MpegEncContext *s, uint8_t **pix, int dir){
616 /* fetch pixels for estimated mv 4 macroblocks ahead
617 * optimized for 64byte cache lines */
618 const int shift = s->quarter_sample ? 2 : 1;
619 const int mx= (s->mv[dir][0][0]>>shift) + 16*s->mb_x + 8;
620 const int my= (s->mv[dir][0][1]>>shift) + 16*s->mb_y;
621 int off= mx + (my + (s->mb_x&3)*4)*s->linesize + 64;
622 s->dsp.prefetch(pix[0]+off, s->linesize, 4);
623 off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
624 s->dsp.prefetch(pix[1]+off, pix[2]-pix[1], 2);
625 }
626
627 /**
628 * motion compensation of a single macroblock
629 * @param s context
630 * @param dest_y luma destination pointer
631 * @param dest_cb chroma cb/u destination pointer
632 * @param dest_cr chroma cr/v destination pointer
633 * @param dir direction (0->forward, 1->backward)
634 * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
635 * @param pic_op halfpel motion compensation function (average or put normally)
636 * @param pic_op qpel motion compensation function (average or put normally)
637 * the motion vectors are taken from s->mv and the MV type from s->mv_type
638 */
639 static av_always_inline void MPV_motion_internal(MpegEncContext *s,
640 uint8_t *dest_y, uint8_t *dest_cb,
641 uint8_t *dest_cr, int dir,
642 uint8_t **ref_picture,
643 op_pixels_func (*pix_op)[4],
644 qpel_mc_func (*qpix_op)[16], int is_mpeg12)
645 {
646 int dxy, mx, my, src_x, src_y, motion_x, motion_y;
647 int mb_x, mb_y, i;
648 uint8_t *ptr, *dest;
649
650 mb_x = s->mb_x;
651 mb_y = s->mb_y;
652
653 prefetch_motion(s, ref_picture, dir);
654
655 if(!is_mpeg12 && s->obmc && s->pict_type != FF_B_TYPE){
656 int16_t mv_cache[4][4][2];
657 const int xy= s->mb_x + s->mb_y*s->mb_stride;
658 const int mot_stride= s->b8_stride;
659 const int mot_xy= mb_x*2 + mb_y*2*mot_stride;
660
661 assert(!s->mb_skipped);
662
663 memcpy(mv_cache[1][1], s->current_picture.motion_val[0][mot_xy ], sizeof(int16_t)*4);
664 memcpy(mv_cache[2][1], s->current_picture.motion_val[0][mot_xy+mot_stride], sizeof(int16_t)*4);
665 memcpy(mv_cache[3][1], s->current_picture.motion_val[0][mot_xy+mot_stride], sizeof(int16_t)*4);
666
667 if(mb_y==0 || IS_INTRA(s->current_picture.mb_type[xy-s->mb_stride])){
668 memcpy(mv_cache[0][1], mv_cache[1][1], sizeof(int16_t)*4);
669 }else{
670 memcpy(mv_cache[0][1], s->current_picture.motion_val[0][mot_xy-mot_stride], sizeof(int16_t)*4);
671 }
672
673 if(mb_x==0 || IS_INTRA(s->current_picture.mb_type[xy-1])){
674 *(int32_t*)mv_cache[1][0]= *(int32_t*)mv_cache[1][1];
675 *(int32_t*)mv_cache[2][0]= *(int32_t*)mv_cache[2][1];
676 }else{
677 *(int32_t*)mv_cache[1][0]= *(int32_t*)s->current_picture.motion_val[0][mot_xy-1];
678 *(int32_t*)mv_cache[2][0]= *(int32_t*)s->current_picture.motion_val[0][mot_xy-1+mot_stride];
679 }
680
681 if(mb_x+1>=s->mb_width || IS_INTRA(s->current_picture.mb_type[xy+1])){
682 *(int32_t*)mv_cache[1][3]= *(int32_t*)mv_cache[1][2];
683 *(int32_t*)mv_cache[2][3]= *(int32_t*)mv_cache[2][2];
684 }else{
685 *(int32_t*)mv_cache[1][3]= *(int32_t*)s->current_picture.motion_val[0][mot_xy+2];
686 *(int32_t*)mv_cache[2][3]= *(int32_t*)s->current_picture.motion_val[0][mot_xy+2+mot_stride];
687 }
688
689 mx = 0;
690 my = 0;
691 for(i=0;i<4;i++) {
692 const int x= (i&1)+1;
693 const int y= (i>>1)+1;
694 int16_t mv[5][2]= {
695 {mv_cache[y][x ][0], mv_cache[y][x ][1]},
696 {mv_cache[y-1][x][0], mv_cache[y-1][x][1]},
697 {mv_cache[y][x-1][0], mv_cache[y][x-1][1]},
698 {mv_cache[y][x+1][0], mv_cache[y][x+1][1]},
699 {mv_cache[y+1][x][0], mv_cache[y+1][x][1]}};
700 //FIXME cleanup
701 obmc_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
702 ref_picture[0],
703 mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >>1) * 8,
704 pix_op[1],
705 mv);
706
707 mx += mv[0][0];
708 my += mv[0][1];
709 }
710 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY))
711 chroma_4mv_motion(s, dest_cb, dest_cr, ref_picture, pix_op[1], mx, my);
712
713 return;
714 }
715
716 switch(s->mv_type) {
717 case MV_TYPE_16X16:
718 if(s->mcsel){
719 if(s->real_sprite_warping_points==1){
720 gmc1_motion(s, dest_y, dest_cb, dest_cr,
721 ref_picture);
722 }else{
723 gmc_motion(s, dest_y, dest_cb, dest_cr,
724 ref_picture);
725 }
726 }else if(!is_mpeg12 && s->quarter_sample){
727 qpel_motion(s, dest_y, dest_cb, dest_cr,
728 0, 0, 0,
729 ref_picture, pix_op, qpix_op,
730 s->mv[dir][0][0], s->mv[dir][0][1], 16);
731 }else if(!is_mpeg12 && CONFIG_WMV2 && s->mspel){
732 ff_mspel_motion(s, dest_y, dest_cb, dest_cr,
733 ref_picture, pix_op,
734 s->mv[dir][0][0], s->mv[dir][0][1], 16);
735 }else
736 {
737 mpeg_motion(s, dest_y, dest_cb, dest_cr,
738 0, 0, 0,
739 ref_picture, pix_op,
740 s->mv[dir][0][0], s->mv[dir][0][1], 16);
741 }
742 break;
743 case MV_TYPE_8X8:
744 if (!is_mpeg12) {
745 mx = 0;
746 my = 0;
747 if(s->quarter_sample){
748 for(i=0;i<4;i++) {
749 motion_x = s->mv[dir][i][0];
750 motion_y = s->mv[dir][i][1];
751
752 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
753 src_x = mb_x * 16 + (motion_x >> 2) + (i & 1) * 8;
754 src_y = mb_y * 16 + (motion_y >> 2) + (i >>1) * 8;
755
756 /* WARNING: do no forget half pels */
757 src_x = av_clip(src_x, -16, s->width);
758 if (src_x == s->width)
759 dxy &= ~3;
760 src_y = av_clip(src_y, -16, s->height);
761 if (src_y == s->height)
762 dxy &= ~12;
763
764 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x);
765 if(s->flags&CODEC_FLAG_EMU_EDGE){
766 if( (unsigned)src_x > s->h_edge_pos - (motion_x&3) - 8
767 || (unsigned)src_y > s->v_edge_pos - (motion_y&3) - 8 ){
768 ff_emulated_edge_mc(s->edge_emu_buffer, ptr,
769 s->linesize, 9, 9,
770 src_x, src_y,
771 s->h_edge_pos, s->v_edge_pos);
772 ptr= s->edge_emu_buffer;
773 }
774 }
775 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize;
776 qpix_op[1][dxy](dest, ptr, s->linesize);
777
778 mx += s->mv[dir][i][0]/2;
779 my += s->mv[dir][i][1]/2;
780 }
781 }else{
782 for(i=0;i<4;i++) {
783 hpel_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
784 ref_picture[0], 0, 0,
785 mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >>1) * 8,
786 s->width, s->height, s->linesize,
787 s->h_edge_pos, s->v_edge_pos,
788 8, 8, pix_op[1],
789 s->mv[dir][i][0], s->mv[dir][i][1]);
790
791 mx += s->mv[dir][i][0];
792 my += s->mv[dir][i][1];
793 }
794 }
795
796 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY))
797 chroma_4mv_motion(s, dest_cb, dest_cr, ref_picture, pix_op[1], mx, my);
798 }
799 break;
800 case MV_TYPE_FIELD:
801 if (s->picture_structure == PICT_FRAME) {
802 if(!is_mpeg12 && s->quarter_sample){
803 for(i=0; i<2; i++){
804 qpel_motion(s, dest_y, dest_cb, dest_cr,
805 1, i, s->field_select[dir][i],
806 ref_picture, pix_op, qpix_op,
807 s->mv[dir][i][0], s->mv[dir][i][1], 8);
808 }
809 }else{
810 /* top field */
811 mpeg_motion(s, dest_y, dest_cb, dest_cr,
812 1, 0, s->field_select[dir][0],
813 ref_picture, pix_op,
814 s->mv[dir][0][0], s->mv[dir][0][1], 8);
815 /* bottom field */
816 mpeg_motion(s, dest_y, dest_cb, dest_cr,
817 1, 1, s->field_select[dir][1],
818 ref_picture, pix_op,
819 s->mv[dir][1][0], s->mv[dir][1][1], 8);
820 }
821 } else {
822 if(s->picture_structure != s->field_select[dir][0] + 1 && s->pict_type != FF_B_TYPE && !s->first_field){
823 ref_picture= s->current_picture_ptr->data;
824 }
825
826 mpeg_motion(s, dest_y, dest_cb, dest_cr,
827 0, 0, s->field_select[dir][0],
828 ref_picture, pix_op,
829 s->mv[dir][0][0], s->mv[dir][0][1], 16);
830 }
831 break;
832 case MV_TYPE_16X8:
833 for(i=0; i<2; i++){
834 uint8_t ** ref2picture;
835
836 if(s->picture_structure == s->field_select[dir][i] + 1
837 || s->pict_type == FF_B_TYPE || s->first_field){
838 ref2picture= ref_picture;
839 }else{
840 ref2picture= s->current_picture_ptr->data;
841 }
842
843 mpeg_motion(s, dest_y, dest_cb, dest_cr,
844 0, 0, s->field_select[dir][i],
845 ref2picture, pix_op,
846 s->mv[dir][i][0], s->mv[dir][i][1] + 16*i, 8);
847
848 dest_y += 16*s->linesize;
849 dest_cb+= (16>>s->chroma_y_shift)*s->uvlinesize;
850 dest_cr+= (16>>s->chroma_y_shift)*s->uvlinesize;
851 }
852 break;
853 case MV_TYPE_DMV:
854 if(s->picture_structure == PICT_FRAME){
855 for(i=0; i<2; i++){
856 int j;
857 for(j=0; j<2; j++){
858 mpeg_motion(s, dest_y, dest_cb, dest_cr,
859 1, j, j^i,
860 ref_picture, pix_op,
861 s->mv[dir][2*i + j][0], s->mv[dir][2*i + j][1], 8);
862 }
863 pix_op = s->dsp.avg_pixels_tab;
864 }
865 }else{
866 for(i=0; i<2; i++){
867 mpeg_motion(s, dest_y, dest_cb, dest_cr,
868 0, 0, s->picture_structure != i+1,
869 ref_picture, pix_op,
870 s->mv[dir][2*i][0],s->mv[dir][2*i][1],16);
871
872 // after put we make avg of the same block
873 pix_op=s->dsp.avg_pixels_tab;
874
875 //opposite parity is always in the same frame if this is second field
876 if(!s->first_field){
877 ref_picture = s->current_picture_ptr->data;
878 }
879 }
880 }
881 break;
882 default: assert(0);
883 }
884 }
885
886 static inline void MPV_motion(MpegEncContext *s,
887 uint8_t *dest_y, uint8_t *dest_cb,
888 uint8_t *dest_cr, int dir,
889 uint8_t **ref_picture,
890 op_pixels_func (*pix_op)[4],
891 qpel_mc_func (*qpix_op)[16])
892 {
893 #if !CONFIG_SMALL
894 if(s->out_format == FMT_MPEG1)
895 MPV_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
896 ref_picture, pix_op, qpix_op, 1);
897 else
898 #endif
899 MPV_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
900 ref_picture, pix_op, qpix_op, 0);
901 }
902 #endif /* AVCODEC_MPEGVIDEO_COMMON_H */
ffmpeg with support for matroska ordered chapters