3756820043f6bd42c2d895290fecfe7834f1e1cd
[libav.git] / libavcodec / mpegvideo_common.h
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 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 "avcodec.h"
34 #include "dsputil.h"
35 #include "mpegvideo.h"
36 #include "mjpegenc.h"
37 #include "msmpeg4.h"
38 #include "faandct.h"
39 #include <limits.h>
40
41 int dct_quantize_c(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow);
42 int dct_quantize_trellis_c(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow);
43 void denoise_dct_c(MpegEncContext *s, DCTELEM *block);
44
45 /**
46 * allocates a Picture
47 * The pixels are allocated/set by calling get_buffer() if shared=0
48 */
49 int alloc_picture(MpegEncContext *s, Picture *pic, int shared);
50
51 /**
52 * sets the given MpegEncContext to common defaults (same for encoding and decoding).
53 * the changed fields will not depend upon the prior state of the MpegEncContext.
54 */
55 void MPV_common_defaults(MpegEncContext *s);
56
57 static inline void gmc1_motion(MpegEncContext *s,
58 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
59 uint8_t **ref_picture)
60 {
61 uint8_t *ptr;
62 int offset, src_x, src_y, linesize, uvlinesize;
63 int motion_x, motion_y;
64 int emu=0;
65
66 motion_x= s->sprite_offset[0][0];
67 motion_y= s->sprite_offset[0][1];
68 src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy+1));
69 src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy+1));
70 motion_x<<=(3-s->sprite_warping_accuracy);
71 motion_y<<=(3-s->sprite_warping_accuracy);
72 src_x = av_clip(src_x, -16, s->width);
73 if (src_x == s->width)
74 motion_x =0;
75 src_y = av_clip(src_y, -16, s->height);
76 if (src_y == s->height)
77 motion_y =0;
78
79 linesize = s->linesize;
80 uvlinesize = s->uvlinesize;
81
82 ptr = ref_picture[0] + (src_y * linesize) + src_x;
83
84 if(s->flags&CODEC_FLAG_EMU_EDGE){
85 if( (unsigned)src_x >= s->h_edge_pos - 17
86 || (unsigned)src_y >= s->v_edge_pos - 17){
87 ff_emulated_edge_mc(s->edge_emu_buffer, ptr, linesize, 17, 17, src_x, src_y, s->h_edge_pos, s->v_edge_pos);
88 ptr= s->edge_emu_buffer;
89 }
90 }
91
92 if((motion_x|motion_y)&7){
93 s->dsp.gmc1(dest_y , ptr , linesize, 16, motion_x&15, motion_y&15, 128 - s->no_rounding);
94 s->dsp.gmc1(dest_y+8, ptr+8, linesize, 16, motion_x&15, motion_y&15, 128 - s->no_rounding);
95 }else{
96 int dxy;
97
98 dxy= ((motion_x>>3)&1) | ((motion_y>>2)&2);
99 if (s->no_rounding){
100 s->dsp.put_no_rnd_pixels_tab[0][dxy](dest_y, ptr, linesize, 16);
101 }else{
102 s->dsp.put_pixels_tab [0][dxy](dest_y, ptr, linesize, 16);
103 }
104 }
105
106 if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
107
108 motion_x= s->sprite_offset[1][0];
109 motion_y= s->sprite_offset[1][1];
110 src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy+1));
111 src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy+1));
112 motion_x<<=(3-s->sprite_warping_accuracy);
113 motion_y<<=(3-s->sprite_warping_accuracy);
114 src_x = av_clip(src_x, -8, s->width>>1);
115 if (src_x == s->width>>1)
116 motion_x =0;
117 src_y = av_clip(src_y, -8, s->height>>1);
118 if (src_y == s->height>>1)
119 motion_y =0;
120
121 offset = (src_y * uvlinesize) + src_x;
122 ptr = ref_picture[1] + offset;
123 if(s->flags&CODEC_FLAG_EMU_EDGE){
124 if( (unsigned)src_x >= (s->h_edge_pos>>1) - 9
125 || (unsigned)src_y >= (s->v_edge_pos>>1) - 9){
126 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);
127 ptr= s->edge_emu_buffer;
128 emu=1;
129 }
130 }
131 s->dsp.gmc1(dest_cb, ptr, uvlinesize, 8, motion_x&15, motion_y&15, 128 - s->no_rounding);
132
133 ptr = ref_picture[2] + offset;
134 if(emu){
135 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);
136 ptr= s->edge_emu_buffer;
137 }
138 s->dsp.gmc1(dest_cr, ptr, uvlinesize, 8, motion_x&15, motion_y&15, 128 - s->no_rounding);
139
140 return;
141 }
142
143 static inline void gmc_motion(MpegEncContext *s,
144 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
145 uint8_t **ref_picture)
146 {
147 uint8_t *ptr;
148 int linesize, uvlinesize;
149 const int a= s->sprite_warping_accuracy;
150 int ox, oy;
151
152 linesize = s->linesize;
153 uvlinesize = s->uvlinesize;
154
155 ptr = ref_picture[0];
156
157 ox= s->sprite_offset[0][0] + s->sprite_delta[0][0]*s->mb_x*16 + s->sprite_delta[0][1]*s->mb_y*16;
158 oy= s->sprite_offset[0][1] + s->sprite_delta[1][0]*s->mb_x*16 + s->sprite_delta[1][1]*s->mb_y*16;
159
160 s->dsp.gmc(dest_y, ptr, linesize, 16,
161 ox,
162 oy,
163 s->sprite_delta[0][0], s->sprite_delta[0][1],
164 s->sprite_delta[1][0], s->sprite_delta[1][1],
165 a+1, (1<<(2*a+1)) - s->no_rounding,
166 s->h_edge_pos, s->v_edge_pos);
167 s->dsp.gmc(dest_y+8, ptr, linesize, 16,
168 ox + s->sprite_delta[0][0]*8,
169 oy + s->sprite_delta[1][0]*8,
170 s->sprite_delta[0][0], s->sprite_delta[0][1],
171 s->sprite_delta[1][0], s->sprite_delta[1][1],
172 a+1, (1<<(2*a+1)) - s->no_rounding,
173 s->h_edge_pos, s->v_edge_pos);
174
175 if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
176
177 ox= s->sprite_offset[1][0] + s->sprite_delta[0][0]*s->mb_x*8 + s->sprite_delta[0][1]*s->mb_y*8;
178 oy= s->sprite_offset[1][1] + s->sprite_delta[1][0]*s->mb_x*8 + s->sprite_delta[1][1]*s->mb_y*8;
179
180 ptr = ref_picture[1];
181 s->dsp.gmc(dest_cb, ptr, uvlinesize, 8,
182 ox,
183 oy,
184 s->sprite_delta[0][0], s->sprite_delta[0][1],
185 s->sprite_delta[1][0], s->sprite_delta[1][1],
186 a+1, (1<<(2*a+1)) - s->no_rounding,
187 s->h_edge_pos>>1, s->v_edge_pos>>1);
188
189 ptr = ref_picture[2];
190 s->dsp.gmc(dest_cr, ptr, uvlinesize, 8,
191 ox,
192 oy,
193 s->sprite_delta[0][0], s->sprite_delta[0][1],
194 s->sprite_delta[1][0], s->sprite_delta[1][1],
195 a+1, (1<<(2*a+1)) - s->no_rounding,
196 s->h_edge_pos>>1, s->v_edge_pos>>1);
197 }
198
199 static inline int hpel_motion(MpegEncContext *s,
200 uint8_t *dest, uint8_t *src,
201 int field_based, int field_select,
202 int src_x, int src_y,
203 int width, int height, int stride,
204 int h_edge_pos, int v_edge_pos,
205 int w, int h, op_pixels_func *pix_op,
206 int motion_x, int motion_y)
207 {
208 int dxy;
209 int emu=0;
210
211 dxy = ((motion_y & 1) << 1) | (motion_x & 1);
212 src_x += motion_x >> 1;
213 src_y += motion_y >> 1;
214
215 /* WARNING: do no forget half pels */
216 src_x = av_clip(src_x, -16, width); //FIXME unneeded for emu?
217 if (src_x == width)
218 dxy &= ~1;
219 src_y = av_clip(src_y, -16, height);
220 if (src_y == height)
221 dxy &= ~2;
222 src += src_y * stride + src_x;
223
224 if(s->unrestricted_mv && (s->flags&CODEC_FLAG_EMU_EDGE)){
225 if( (unsigned)src_x > h_edge_pos - (motion_x&1) - w
226 || (unsigned)src_y > v_edge_pos - (motion_y&1) - h){
227 ff_emulated_edge_mc(s->edge_emu_buffer, src, s->linesize, w+1, (h+1)<<field_based,
228 src_x, src_y<<field_based, h_edge_pos, s->v_edge_pos);
229 src= s->edge_emu_buffer;
230 emu=1;
231 }
232 }
233 if(field_select)
234 src += s->linesize;
235 pix_op[dxy](dest, src, stride, h);
236 return emu;
237 }
238
239 static av_always_inline
240 void mpeg_motion_internal(MpegEncContext *s,
241 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
242 int field_based, int bottom_field, int field_select,
243 uint8_t **ref_picture, op_pixels_func (*pix_op)[4],
244 int motion_x, int motion_y, int h, int is_mpeg12)
245 {
246 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
247 int dxy, uvdxy, mx, my, src_x, src_y,
248 uvsrc_x, uvsrc_y, v_edge_pos, uvlinesize, linesize;
249
250 #if 0
251 if(s->quarter_sample)
252 {
253 motion_x>>=1;
254 motion_y>>=1;
255 }
256 #endif
257
258 v_edge_pos = s->v_edge_pos >> field_based;
259 linesize = s->current_picture.linesize[0] << field_based;
260 uvlinesize = s->current_picture.linesize[1] << field_based;
261
262 dxy = ((motion_y & 1) << 1) | (motion_x & 1);
263 src_x = s->mb_x* 16 + (motion_x >> 1);
264 src_y =(s->mb_y<<(4-field_based)) + (motion_y >> 1);
265
266 if (!is_mpeg12 && s->out_format == FMT_H263) {
267 if((s->workaround_bugs & FF_BUG_HPEL_CHROMA) && field_based){
268 mx = (motion_x>>1)|(motion_x&1);
269 my = motion_y >>1;
270 uvdxy = ((my & 1) << 1) | (mx & 1);
271 uvsrc_x = s->mb_x* 8 + (mx >> 1);
272 uvsrc_y = (s->mb_y<<(3-field_based)) + (my >> 1);
273 }else{
274 uvdxy = dxy | (motion_y & 2) | ((motion_x & 2) >> 1);
275 uvsrc_x = src_x>>1;
276 uvsrc_y = src_y>>1;
277 }
278 }else if(!is_mpeg12 && s->out_format == FMT_H261){//even chroma mv's are full pel in H261
279 mx = motion_x / 4;
280 my = motion_y / 4;
281 uvdxy = 0;
282 uvsrc_x = s->mb_x*8 + mx;
283 uvsrc_y = s->mb_y*8 + my;
284 } else {
285 if(s->chroma_y_shift){
286 mx = motion_x / 2;
287 my = motion_y / 2;
288 uvdxy = ((my & 1) << 1) | (mx & 1);
289 uvsrc_x = s->mb_x* 8 + (mx >> 1);
290 uvsrc_y = (s->mb_y<<(3-field_based)) + (my >> 1);
291 } else {
292 if(s->chroma_x_shift){
293 //Chroma422
294 mx = motion_x / 2;
295 uvdxy = ((motion_y & 1) << 1) | (mx & 1);
296 uvsrc_x = s->mb_x* 8 + (mx >> 1);
297 uvsrc_y = src_y;
298 } else {
299 //Chroma444
300 uvdxy = dxy;
301 uvsrc_x = src_x;
302 uvsrc_y = src_y;
303 }
304 }
305 }
306
307 ptr_y = ref_picture[0] + src_y * linesize + src_x;
308 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
309 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
310
311 if( (unsigned)src_x > s->h_edge_pos - (motion_x&1) - 16
312 || (unsigned)src_y > v_edge_pos - (motion_y&1) - h){
313 if(is_mpeg12 || s->codec_id == CODEC_ID_MPEG2VIDEO ||
314 s->codec_id == CODEC_ID_MPEG1VIDEO){
315 av_log(s->avctx,AV_LOG_DEBUG,
316 "MPEG motion vector out of boundary\n");
317 return ;
318 }
319 ff_emulated_edge_mc(s->edge_emu_buffer, ptr_y, s->linesize,
320 17, 17+field_based,
321 src_x, src_y<<field_based,
322 s->h_edge_pos, s->v_edge_pos);
323 ptr_y = s->edge_emu_buffer;
324 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
325 uint8_t *uvbuf= s->edge_emu_buffer+18*s->linesize;
326 ff_emulated_edge_mc(uvbuf ,
327 ptr_cb, s->uvlinesize,
328 9, 9+field_based,
329 uvsrc_x, uvsrc_y<<field_based,
330 s->h_edge_pos>>1, s->v_edge_pos>>1);
331 ff_emulated_edge_mc(uvbuf+16,
332 ptr_cr, s->uvlinesize,
333 9, 9+field_based,
334 uvsrc_x, uvsrc_y<<field_based,
335 s->h_edge_pos>>1, s->v_edge_pos>>1);
336 ptr_cb= uvbuf;
337 ptr_cr= uvbuf+16;
338 }
339 }
340
341 if(bottom_field){ //FIXME use this for field pix too instead of the obnoxious hack which changes picture.data
342 dest_y += s->linesize;
343 dest_cb+= s->uvlinesize;
344 dest_cr+= s->uvlinesize;
345 }
346
347 if(field_select){
348 ptr_y += s->linesize;
349 ptr_cb+= s->uvlinesize;
350 ptr_cr+= s->uvlinesize;
351 }
352
353 pix_op[0][dxy](dest_y, ptr_y, linesize, h);
354
355 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
356 pix_op[s->chroma_x_shift][uvdxy]
357 (dest_cb, ptr_cb, uvlinesize, h >> s->chroma_y_shift);
358 pix_op[s->chroma_x_shift][uvdxy]
359 (dest_cr, ptr_cr, uvlinesize, h >> s->chroma_y_shift);
360 }
361 if(!is_mpeg12 && (CONFIG_H261_ENCODER || CONFIG_H261_DECODER) &&
362 s->out_format == FMT_H261){
363 ff_h261_loop_filter(s);
364 }
365 }
366 /* apply one mpeg motion vector to the three components */
367 static av_always_inline
368 void mpeg_motion(MpegEncContext *s,
369 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
370 int field_based, int bottom_field, int field_select,
371 uint8_t **ref_picture, op_pixels_func (*pix_op)[4],
372 int motion_x, int motion_y, int h)
373 {
374 #if !CONFIG_SMALL
375 if(s->out_format == FMT_MPEG1)
376 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, field_based,
377 bottom_field, field_select, ref_picture, pix_op,
378 motion_x, motion_y, h, 1);
379 else
380 #endif
381 mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, field_based,
382 bottom_field, field_select, ref_picture, pix_op,
383 motion_x, motion_y, h, 0);
384 }
385
386 //FIXME move to dsputil, avg variant, 16x16 version
387 static inline void put_obmc(uint8_t *dst, uint8_t *src[5], int stride){
388 int x;
389 uint8_t * const top = src[1];
390 uint8_t * const left = src[2];
391 uint8_t * const mid = src[0];
392 uint8_t * const right = src[3];
393 uint8_t * const bottom= src[4];
394 #define OBMC_FILTER(x, t, l, m, r, b)\
395 dst[x]= (t*top[x] + l*left[x] + m*mid[x] + r*right[x] + b*bottom[x] + 4)>>3
396 #define OBMC_FILTER4(x, t, l, m, r, b)\
397 OBMC_FILTER(x , t, l, m, r, b);\
398 OBMC_FILTER(x+1 , t, l, m, r, b);\
399 OBMC_FILTER(x +stride, t, l, m, r, b);\
400 OBMC_FILTER(x+1+stride, t, l, m, r, b);
401
402 x=0;
403 OBMC_FILTER (x , 2, 2, 4, 0, 0);
404 OBMC_FILTER (x+1, 2, 1, 5, 0, 0);
405 OBMC_FILTER4(x+2, 2, 1, 5, 0, 0);
406 OBMC_FILTER4(x+4, 2, 0, 5, 1, 0);
407 OBMC_FILTER (x+6, 2, 0, 5, 1, 0);
408 OBMC_FILTER (x+7, 2, 0, 4, 2, 0);
409 x+= stride;
410 OBMC_FILTER (x , 1, 2, 5, 0, 0);
411 OBMC_FILTER (x+1, 1, 2, 5, 0, 0);
412 OBMC_FILTER (x+6, 1, 0, 5, 2, 0);
413 OBMC_FILTER (x+7, 1, 0, 5, 2, 0);
414 x+= stride;
415 OBMC_FILTER4(x , 1, 2, 5, 0, 0);
416 OBMC_FILTER4(x+2, 1, 1, 6, 0, 0);
417 OBMC_FILTER4(x+4, 1, 0, 6, 1, 0);
418 OBMC_FILTER4(x+6, 1, 0, 5, 2, 0);
419 x+= 2*stride;
420 OBMC_FILTER4(x , 0, 2, 5, 0, 1);
421 OBMC_FILTER4(x+2, 0, 1, 6, 0, 1);
422 OBMC_FILTER4(x+4, 0, 0, 6, 1, 1);
423 OBMC_FILTER4(x+6, 0, 0, 5, 2, 1);
424 x+= 2*stride;
425 OBMC_FILTER (x , 0, 2, 5, 0, 1);
426 OBMC_FILTER (x+1, 0, 2, 5, 0, 1);
427 OBMC_FILTER4(x+2, 0, 1, 5, 0, 2);
428 OBMC_FILTER4(x+4, 0, 0, 5, 1, 2);
429 OBMC_FILTER (x+6, 0, 0, 5, 2, 1);
430 OBMC_FILTER (x+7, 0, 0, 5, 2, 1);
431 x+= stride;
432 OBMC_FILTER (x , 0, 2, 4, 0, 2);
433 OBMC_FILTER (x+1, 0, 1, 5, 0, 2);
434 OBMC_FILTER (x+6, 0, 0, 5, 1, 2);
435 OBMC_FILTER (x+7, 0, 0, 4, 2, 2);
436 }
437
438 /* obmc for 1 8x8 luma block */
439 static inline void obmc_motion(MpegEncContext *s,
440 uint8_t *dest, uint8_t *src,
441 int src_x, int src_y,
442 op_pixels_func *pix_op,
443 int16_t mv[5][2]/* mid top left right bottom*/)
444 #define MID 0
445 {
446 int i;
447 uint8_t *ptr[5];
448
449 assert(s->quarter_sample==0);
450
451 for(i=0; i<5; i++){
452 if(i && mv[i][0]==mv[MID][0] && mv[i][1]==mv[MID][1]){
453 ptr[i]= ptr[MID];
454 }else{
455 ptr[i]= s->obmc_scratchpad + 8*(i&1) + s->linesize*8*(i>>1);
456 hpel_motion(s, ptr[i], src, 0, 0,
457 src_x, src_y,
458 s->width, s->height, s->linesize,
459 s->h_edge_pos, s->v_edge_pos,
460 8, 8, pix_op,
461 mv[i][0], mv[i][1]);
462 }
463 }
464
465 put_obmc(dest, ptr, s->linesize);
466 }
467
468 static inline void qpel_motion(MpegEncContext *s,
469 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
470 int field_based, int bottom_field, int field_select,
471 uint8_t **ref_picture, op_pixels_func (*pix_op)[4],
472 qpel_mc_func (*qpix_op)[16],
473 int motion_x, int motion_y, int h)
474 {
475 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
476 int dxy, uvdxy, mx, my, src_x, src_y, uvsrc_x, uvsrc_y, v_edge_pos, linesize, uvlinesize;
477
478 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
479 src_x = s->mb_x * 16 + (motion_x >> 2);
480 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2);
481
482 v_edge_pos = s->v_edge_pos >> field_based;
483 linesize = s->linesize << field_based;
484 uvlinesize = s->uvlinesize << field_based;
485
486 if(field_based){
487 mx= motion_x/2;
488 my= motion_y>>1;
489 }else if(s->workaround_bugs&FF_BUG_QPEL_CHROMA2){
490 static const int rtab[8]= {0,0,1,1,0,0,0,1};
491 mx= (motion_x>>1) + rtab[motion_x&7];
492 my= (motion_y>>1) + rtab[motion_y&7];
493 }else if(s->workaround_bugs&FF_BUG_QPEL_CHROMA){
494 mx= (motion_x>>1)|(motion_x&1);
495 my= (motion_y>>1)|(motion_y&1);
496 }else{
497 mx= motion_x/2;
498 my= motion_y/2;
499 }
500 mx= (mx>>1)|(mx&1);
501 my= (my>>1)|(my&1);
502
503 uvdxy= (mx&1) | ((my&1)<<1);
504 mx>>=1;
505 my>>=1;
506
507 uvsrc_x = s->mb_x * 8 + mx;
508 uvsrc_y = s->mb_y * (8 >> field_based) + my;
509
510 ptr_y = ref_picture[0] + src_y * linesize + src_x;
511 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
512 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
513
514 if( (unsigned)src_x > s->h_edge_pos - (motion_x&3) - 16
515 || (unsigned)src_y > v_edge_pos - (motion_y&3) - h ){
516 ff_emulated_edge_mc(s->edge_emu_buffer, ptr_y, s->linesize,
517 17, 17+field_based, src_x, src_y<<field_based,
518 s->h_edge_pos, s->v_edge_pos);
519 ptr_y= s->edge_emu_buffer;
520 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
521 uint8_t *uvbuf= s->edge_emu_buffer + 18*s->linesize;
522 ff_emulated_edge_mc(uvbuf, ptr_cb, s->uvlinesize,
523 9, 9 + field_based,
524 uvsrc_x, uvsrc_y<<field_based,
525 s->h_edge_pos>>1, s->v_edge_pos>>1);
526 ff_emulated_edge_mc(uvbuf + 16, ptr_cr, s->uvlinesize,
527 9, 9 + field_based,
528 uvsrc_x, uvsrc_y<<field_based,
529 s->h_edge_pos>>1, s->v_edge_pos>>1);
530 ptr_cb= uvbuf;
531 ptr_cr= uvbuf + 16;
532 }
533 }
534
535 if(!field_based)
536 qpix_op[0][dxy](dest_y, ptr_y, linesize);
537 else{
538 if(bottom_field){
539 dest_y += s->linesize;
540 dest_cb+= s->uvlinesize;
541 dest_cr+= s->uvlinesize;
542 }
543
544 if(field_select){
545 ptr_y += s->linesize;
546 ptr_cb += s->uvlinesize;
547 ptr_cr += s->uvlinesize;
548 }
549 //damn interlaced mode
550 //FIXME boundary mirroring is not exactly correct here
551 qpix_op[1][dxy](dest_y , ptr_y , linesize);
552 qpix_op[1][dxy](dest_y+8, ptr_y+8, linesize);
553 }
554 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
555 pix_op[1][uvdxy](dest_cr, ptr_cr, uvlinesize, h >> 1);
556 pix_op[1][uvdxy](dest_cb, ptr_cb, uvlinesize, h >> 1);
557 }
558 }
559
560 /**
561 * h263 chroma 4mv motion compensation.
562 */
563 static inline void chroma_4mv_motion(MpegEncContext *s,
564 uint8_t *dest_cb, uint8_t *dest_cr,
565 uint8_t **ref_picture,
566 op_pixels_func *pix_op,
567 int mx, int my){
568 int dxy, emu=0, src_x, src_y, offset;
569 uint8_t *ptr;
570
571 /* In case of 8X8, we construct a single chroma motion vector
572 with a special rounding */
573 mx= ff_h263_round_chroma(mx);
574 my= ff_h263_round_chroma(my);
575
576 dxy = ((my & 1) << 1) | (mx & 1);
577 mx >>= 1;
578 my >>= 1;
579
580 src_x = s->mb_x * 8 + mx;
581 src_y = s->mb_y * 8 + my;
582 src_x = av_clip(src_x, -8, s->width/2);
583 if (src_x == s->width/2)
584 dxy &= ~1;
585 src_y = av_clip(src_y, -8, s->height/2);
586 if (src_y == s->height/2)
587 dxy &= ~2;
588
589 offset = (src_y * (s->uvlinesize)) + src_x;
590 ptr = ref_picture[1] + offset;
591 if(s->flags&CODEC_FLAG_EMU_EDGE){
592 if( (unsigned)src_x > (s->h_edge_pos>>1) - (dxy &1) - 8
593 || (unsigned)src_y > (s->v_edge_pos>>1) - (dxy>>1) - 8){
594 ff_emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize,
595 9, 9, src_x, src_y,
596 s->h_edge_pos>>1, s->v_edge_pos>>1);
597 ptr= s->edge_emu_buffer;
598 emu=1;
599 }
600 }
601 pix_op[dxy](dest_cb, ptr, s->uvlinesize, 8);
602
603 ptr = ref_picture[2] + offset;
604 if(emu){
605 ff_emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize,
606 9, 9, src_x, src_y,
607 s->h_edge_pos>>1, s->v_edge_pos>>1);
608 ptr= s->edge_emu_buffer;
609 }
610 pix_op[dxy](dest_cr, ptr, s->uvlinesize, 8);
611 }
612
613 static inline void prefetch_motion(MpegEncContext *s, uint8_t **pix, int dir){
614 /* fetch pixels for estimated mv 4 macroblocks ahead
615 * optimized for 64byte cache lines */
616 const int shift = s->quarter_sample ? 2 : 1;
617 const int mx= (s->mv[dir][0][0]>>shift) + 16*s->mb_x + 8;
618 const int my= (s->mv[dir][0][1]>>shift) + 16*s->mb_y;
619 int off= mx + (my + (s->mb_x&3)*4)*s->linesize + 64;
620 s->dsp.prefetch(pix[0]+off, s->linesize, 4);
621 off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
622 s->dsp.prefetch(pix[1]+off, pix[2]-pix[1], 2);
623 }
624
625 /**
626 * motion compensation of a single macroblock
627 * @param s context
628 * @param dest_y luma destination pointer
629 * @param dest_cb chroma cb/u destination pointer
630 * @param dest_cr chroma cr/v destination pointer
631 * @param dir direction (0->forward, 1->backward)
632 * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
633 * @param pic_op halfpel motion compensation function (average or put normally)
634 * @param pic_op qpel motion compensation function (average or put normally)
635 * the motion vectors are taken from s->mv and the MV type from s->mv_type
636 */
637 static av_always_inline void MPV_motion_internal(MpegEncContext *s,
638 uint8_t *dest_y, uint8_t *dest_cb,
639 uint8_t *dest_cr, int dir,
640 uint8_t **ref_picture,
641 op_pixels_func (*pix_op)[4],
642 qpel_mc_func (*qpix_op)[16], int is_mpeg12)
643 {
644 int dxy, mx, my, src_x, src_y, motion_x, motion_y;
645 int mb_x, mb_y, i;
646 uint8_t *ptr, *dest;
647
648 mb_x = s->mb_x;
649 mb_y = s->mb_y;
650
651 prefetch_motion(s, ref_picture, dir);
652
653 if(!is_mpeg12 && s->obmc && s->pict_type != FF_B_TYPE){
654 int16_t mv_cache[4][4][2];
655 const int xy= s->mb_x + s->mb_y*s->mb_stride;
656 const int mot_stride= s->b8_stride;
657 const int mot_xy= mb_x*2 + mb_y*2*mot_stride;
658
659 assert(!s->mb_skipped);
660
661 memcpy(mv_cache[1][1], s->current_picture.motion_val[0][mot_xy ], sizeof(int16_t)*4);
662 memcpy(mv_cache[2][1], s->current_picture.motion_val[0][mot_xy+mot_stride], sizeof(int16_t)*4);
663 memcpy(mv_cache[3][1], s->current_picture.motion_val[0][mot_xy+mot_stride], sizeof(int16_t)*4);
664
665 if(mb_y==0 || IS_INTRA(s->current_picture.mb_type[xy-s->mb_stride])){
666 memcpy(mv_cache[0][1], mv_cache[1][1], sizeof(int16_t)*4);
667 }else{
668 memcpy(mv_cache[0][1], s->current_picture.motion_val[0][mot_xy-mot_stride], sizeof(int16_t)*4);
669 }
670
671 if(mb_x==0 || IS_INTRA(s->current_picture.mb_type[xy-1])){
672 *(int32_t*)mv_cache[1][0]= *(int32_t*)mv_cache[1][1];
673 *(int32_t*)mv_cache[2][0]= *(int32_t*)mv_cache[2][1];
674 }else{
675 *(int32_t*)mv_cache[1][0]= *(int32_t*)s->current_picture.motion_val[0][mot_xy-1];
676 *(int32_t*)mv_cache[2][0]= *(int32_t*)s->current_picture.motion_val[0][mot_xy-1+mot_stride];
677 }
678
679 if(mb_x+1>=s->mb_width || IS_INTRA(s->current_picture.mb_type[xy+1])){
680 *(int32_t*)mv_cache[1][3]= *(int32_t*)mv_cache[1][2];
681 *(int32_t*)mv_cache[2][3]= *(int32_t*)mv_cache[2][2];
682 }else{
683 *(int32_t*)mv_cache[1][3]= *(int32_t*)s->current_picture.motion_val[0][mot_xy+2];
684 *(int32_t*)mv_cache[2][3]= *(int32_t*)s->current_picture.motion_val[0][mot_xy+2+mot_stride];
685 }
686
687 mx = 0;
688 my = 0;
689 for(i=0;i<4;i++) {
690 const int x= (i&1)+1;
691 const int y= (i>>1)+1;
692 int16_t mv[5][2]= {
693 {mv_cache[y][x ][0], mv_cache[y][x ][1]},
694 {mv_cache[y-1][x][0], mv_cache[y-1][x][1]},
695 {mv_cache[y][x-1][0], mv_cache[y][x-1][1]},
696 {mv_cache[y][x+1][0], mv_cache[y][x+1][1]},
697 {mv_cache[y+1][x][0], mv_cache[y+1][x][1]}};
698 //FIXME cleanup
699 obmc_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
700 ref_picture[0],
701 mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >>1) * 8,
702 pix_op[1],
703 mv);
704
705 mx += mv[0][0];
706 my += mv[0][1];
707 }
708 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY))
709 chroma_4mv_motion(s, dest_cb, dest_cr, ref_picture, pix_op[1], mx, my);
710
711 return;
712 }
713
714 switch(s->mv_type) {
715 case MV_TYPE_16X16:
716 if(s->mcsel){
717 if(s->real_sprite_warping_points==1){
718 gmc1_motion(s, dest_y, dest_cb, dest_cr,
719 ref_picture);
720 }else{
721 gmc_motion(s, dest_y, dest_cb, dest_cr,
722 ref_picture);
723 }
724 }else if(!is_mpeg12 && s->quarter_sample){
725 qpel_motion(s, dest_y, dest_cb, dest_cr,
726 0, 0, 0,
727 ref_picture, pix_op, qpix_op,
728 s->mv[dir][0][0], s->mv[dir][0][1], 16);
729 }else if(!is_mpeg12 && CONFIG_WMV2 && s->mspel){
730 ff_mspel_motion(s, dest_y, dest_cb, dest_cr,
731 ref_picture, pix_op,
732 s->mv[dir][0][0], s->mv[dir][0][1], 16);
733 }else
734 {
735 mpeg_motion(s, dest_y, dest_cb, dest_cr,
736 0, 0, 0,
737 ref_picture, pix_op,
738 s->mv[dir][0][0], s->mv[dir][0][1], 16);
739 }
740 break;
741 case MV_TYPE_8X8:
742 if (!is_mpeg12) {
743 mx = 0;
744 my = 0;
745 if(s->quarter_sample){
746 for(i=0;i<4;i++) {
747 motion_x = s->mv[dir][i][0];
748 motion_y = s->mv[dir][i][1];
749
750 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
751 src_x = mb_x * 16 + (motion_x >> 2) + (i & 1) * 8;
752 src_y = mb_y * 16 + (motion_y >> 2) + (i >>1) * 8;
753
754 /* WARNING: do no forget half pels */
755 src_x = av_clip(src_x, -16, s->width);
756 if (src_x == s->width)
757 dxy &= ~3;
758 src_y = av_clip(src_y, -16, s->height);
759 if (src_y == s->height)
760 dxy &= ~12;
761
762 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x);
763 if(s->flags&CODEC_FLAG_EMU_EDGE){
764 if( (unsigned)src_x > s->h_edge_pos - (motion_x&3) - 8
765 || (unsigned)src_y > s->v_edge_pos - (motion_y&3) - 8 ){
766 ff_emulated_edge_mc(s->edge_emu_buffer, ptr,
767 s->linesize, 9, 9,
768 src_x, src_y,
769 s->h_edge_pos, s->v_edge_pos);
770 ptr= s->edge_emu_buffer;
771 }
772 }
773 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize;
774 qpix_op[1][dxy](dest, ptr, s->linesize);
775
776 mx += s->mv[dir][i][0]/2;
777 my += s->mv[dir][i][1]/2;
778 }
779 }else{
780 for(i=0;i<4;i++) {
781 hpel_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
782 ref_picture[0], 0, 0,
783 mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >>1) * 8,
784 s->width, s->height, s->linesize,
785 s->h_edge_pos, s->v_edge_pos,
786 8, 8, pix_op[1],
787 s->mv[dir][i][0], s->mv[dir][i][1]);
788
789 mx += s->mv[dir][i][0];
790 my += s->mv[dir][i][1];
791 }
792 }
793
794 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY))
795 chroma_4mv_motion(s, dest_cb, dest_cr, ref_picture, pix_op[1], mx, my);
796 }
797 break;
798 case MV_TYPE_FIELD:
799 if (s->picture_structure == PICT_FRAME) {
800 if(!is_mpeg12 && s->quarter_sample){
801 for(i=0; i<2; i++){
802 qpel_motion(s, dest_y, dest_cb, dest_cr,
803 1, i, s->field_select[dir][i],
804 ref_picture, pix_op, qpix_op,
805 s->mv[dir][i][0], s->mv[dir][i][1], 8);
806 }
807 }else{
808 /* top field */
809 mpeg_motion(s, dest_y, dest_cb, dest_cr,
810 1, 0, s->field_select[dir][0],
811 ref_picture, pix_op,
812 s->mv[dir][0][0], s->mv[dir][0][1], 8);
813 /* bottom field */
814 mpeg_motion(s, dest_y, dest_cb, dest_cr,
815 1, 1, s->field_select[dir][1],
816 ref_picture, pix_op,
817 s->mv[dir][1][0], s->mv[dir][1][1], 8);
818 }
819 } else {
820 if(s->picture_structure != s->field_select[dir][0] + 1 && s->pict_type != FF_B_TYPE && !s->first_field){
821 ref_picture= s->current_picture_ptr->data;
822 }
823
824 mpeg_motion(s, dest_y, dest_cb, dest_cr,
825 0, 0, s->field_select[dir][0],
826 ref_picture, pix_op,
827 s->mv[dir][0][0], s->mv[dir][0][1], 16);
828 }
829 break;
830 case MV_TYPE_16X8:
831 for(i=0; i<2; i++){
832 uint8_t ** ref2picture;
833
834 if(s->picture_structure == s->field_select[dir][i] + 1
835 || s->pict_type == FF_B_TYPE || s->first_field){
836 ref2picture= ref_picture;
837 }else{
838 ref2picture= s->current_picture_ptr->data;
839 }
840
841 mpeg_motion(s, dest_y, dest_cb, dest_cr,
842 0, 0, s->field_select[dir][i],
843 ref2picture, pix_op,
844 s->mv[dir][i][0], s->mv[dir][i][1] + 16*i, 8);
845
846 dest_y += 16*s->linesize;
847 dest_cb+= (16>>s->chroma_y_shift)*s->uvlinesize;
848 dest_cr+= (16>>s->chroma_y_shift)*s->uvlinesize;
849 }
850 break;
851 case MV_TYPE_DMV:
852 if(s->picture_structure == PICT_FRAME){
853 for(i=0; i<2; i++){
854 int j;
855 for(j=0; j<2; j++){
856 mpeg_motion(s, dest_y, dest_cb, dest_cr,
857 1, j, j^i,
858 ref_picture, pix_op,
859 s->mv[dir][2*i + j][0], s->mv[dir][2*i + j][1], 8);
860 }
861 pix_op = s->dsp.avg_pixels_tab;
862 }
863 }else{
864 for(i=0; i<2; i++){
865 mpeg_motion(s, dest_y, dest_cb, dest_cr,
866 0, 0, s->picture_structure != i+1,
867 ref_picture, pix_op,
868 s->mv[dir][2*i][0],s->mv[dir][2*i][1],16);
869
870 // after put we make avg of the same block
871 pix_op=s->dsp.avg_pixels_tab;
872
873 //opposite parity is always in the same frame if this is second field
874 if(!s->first_field){
875 ref_picture = s->current_picture_ptr->data;
876 }
877 }
878 }
879 break;
880 default: assert(0);
881 }
882 }
883
884 static inline void MPV_motion(MpegEncContext *s,
885 uint8_t *dest_y, uint8_t *dest_cb,
886 uint8_t *dest_cr, int dir,
887 uint8_t **ref_picture,
888 op_pixels_func (*pix_op)[4],
889 qpel_mc_func (*qpix_op)[16])
890 {
891 #if !CONFIG_SMALL
892 if(s->out_format == FMT_MPEG1)
893 MPV_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
894 ref_picture, pix_op, qpix_op, 1);
895 else
896 #endif
897 MPV_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
898 ref_picture, pix_op, qpix_op, 0);
899 }
900 #endif /* AVCODEC_MPEGVIDEO_COMMON_H */