uvlinesize
[libav.git] / libavcodec / mpegvideo.c
1 /*
2 * The simplest mpeg encoder (well, it was the simplest!)
3 * Copyright (c) 2000,2001 Fabrice Bellard.
4 *
5 * This library is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU Lesser General Public
7 * License as published by the Free Software Foundation; either
8 * version 2 of the License, or (at your option) any later version.
9 *
10 * This library is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * Lesser General Public License for more details.
14 *
15 * You should have received a copy of the GNU Lesser General Public
16 * License along with this library; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18 *
19 * 4MV & hq & b-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
20 */
21 #include "avcodec.h"
22 #include "dsputil.h"
23 #include "mpegvideo.h"
24
25 #ifdef USE_FASTMEMCPY
26 #include "fastmemcpy.h"
27 #endif
28
29 static void encode_picture(MpegEncContext *s, int picture_number);
30 static void dct_unquantize_mpeg1_c(MpegEncContext *s,
31 DCTELEM *block, int n, int qscale);
32 static void dct_unquantize_mpeg2_c(MpegEncContext *s,
33 DCTELEM *block, int n, int qscale);
34 static void dct_unquantize_h263_c(MpegEncContext *s,
35 DCTELEM *block, int n, int qscale);
36 static void draw_edges_c(UINT8 *buf, int wrap, int width, int height, int w);
37 static int dct_quantize_c(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow);
38
39 int (*dct_quantize)(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow)= dct_quantize_c;
40 void (*draw_edges)(UINT8 *buf, int wrap, int width, int height, int w)= draw_edges_c;
41
42 #define EDGE_WIDTH 16
43
44 /* enable all paranoid tests for rounding, overflows, etc... */
45 //#define PARANOID
46
47 //#define DEBUG
48
49
50 /* for jpeg fast DCT */
51 #define CONST_BITS 14
52
53 static const unsigned short aanscales[64] = {
54 /* precomputed values scaled up by 14 bits */
55 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520,
56 22725, 31521, 29692, 26722, 22725, 17855, 12299, 6270,
57 21407, 29692, 27969, 25172, 21407, 16819, 11585, 5906,
58 19266, 26722, 25172, 22654, 19266, 15137, 10426, 5315,
59 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520,
60 12873, 17855, 16819, 15137, 12873, 10114, 6967, 3552,
61 8867, 12299, 11585, 10426, 8867, 6967, 4799, 2446,
62 4520, 6270, 5906, 5315, 4520, 3552, 2446, 1247
63 };
64
65 static UINT8 h263_chroma_roundtab[16] = {
66 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2,
67 };
68
69 static UINT16 default_mv_penalty[MAX_FCODE+1][MAX_MV*2+1];
70 static UINT8 default_fcode_tab[MAX_MV*2+1];
71
72 extern UINT8 zigzag_end[64];
73
74 /* default motion estimation */
75 int motion_estimation_method = ME_EPZS;
76
77 static void convert_matrix(int (*qmat)[64], uint16_t (*qmat16)[64], uint16_t (*qmat16_bias)[64],
78 const UINT16 *quant_matrix, int bias)
79 {
80 int qscale;
81
82 for(qscale=1; qscale<32; qscale++){
83 int i;
84 if (av_fdct == fdct_ifast) {
85 for(i=0;i<64;i++) {
86 const int j= block_permute_op(i);
87 /* 16 <= qscale * quant_matrix[i] <= 7905 */
88 /* 19952 <= aanscales[i] * qscale * quant_matrix[i] <= 249205026 */
89 /* (1<<36)/19952 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */
90 /* 3444240 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= 275 */
91
92 qmat[qscale][j] = (int)((UINT64_C(1) << (QMAT_SHIFT + 11)) /
93 (aanscales[i] * qscale * quant_matrix[j]));
94 }
95 } else {
96 for(i=0;i<64;i++) {
97 /* We can safely suppose that 16 <= quant_matrix[i] <= 255
98 So 16 <= qscale * quant_matrix[i] <= 7905
99 so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905
100 so 32768 >= (1<<19) / (qscale * quant_matrix[i]) >= 67
101 */
102 qmat [qscale][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]);
103 qmat16[qscale][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[block_permute_op(i)]);
104
105 if(qmat16[qscale][i]==0 || qmat16[qscale][i]==128*256) qmat16[qscale][i]=128*256-1;
106
107 qmat16_bias[qscale][i]= ROUNDED_DIV(bias<<(16-QUANT_BIAS_SHIFT), qmat16[qscale][i]);
108 }
109 }
110 }
111 }
112 // move into common.c perhaps
113 #define CHECKED_ALLOCZ(p, size)\
114 {\
115 p= av_mallocz(size);\
116 if(p==NULL){\
117 perror("malloc");\
118 goto fail;\
119 }\
120 }
121
122 /* init common structure for both encoder and decoder */
123 int MPV_common_init(MpegEncContext *s)
124 {
125 int c_size, i;
126 UINT8 *pict;
127
128 s->dct_unquantize_h263 = dct_unquantize_h263_c;
129 s->dct_unquantize_mpeg1 = dct_unquantize_mpeg1_c;
130 s->dct_unquantize_mpeg2 = dct_unquantize_mpeg2_c;
131
132 #ifdef HAVE_MMX
133 MPV_common_init_mmx(s);
134 #endif
135 #ifdef ARCH_ALPHA
136 MPV_common_init_axp(s);
137 #endif
138 //setup default unquantizers (mpeg4 might change it later)
139 if(s->out_format == FMT_H263)
140 s->dct_unquantize = s->dct_unquantize_h263;
141 else
142 s->dct_unquantize = s->dct_unquantize_mpeg1;
143
144 s->mb_width = (s->width + 15) / 16;
145 s->mb_height = (s->height + 15) / 16;
146 s->mb_num = s->mb_width * s->mb_height;
147 if(!(s->flags&CODEC_FLAG_DR1)){
148 s->linesize = s->mb_width * 16 + 2 * EDGE_WIDTH;
149 s->uvlinesize = s->mb_width * 8 + EDGE_WIDTH;
150
151 for(i=0;i<3;i++) {
152 int w, h, shift, pict_start;
153
154 w = s->linesize;
155 h = s->mb_height * 16 + 2 * EDGE_WIDTH;
156 shift = (i == 0) ? 0 : 1;
157 c_size = (s->linesize>>shift) * (h >> shift);
158 pict_start = (s->linesize>>shift) * (EDGE_WIDTH >> shift) + (EDGE_WIDTH >> shift);
159
160 CHECKED_ALLOCZ(pict, c_size)
161 s->last_picture_base[i] = pict;
162 s->last_picture[i] = pict + pict_start;
163 if(i>0) memset(s->last_picture_base[i], 128, c_size);
164
165 CHECKED_ALLOCZ(pict, c_size)
166 s->next_picture_base[i] = pict;
167 s->next_picture[i] = pict + pict_start;
168 if(i>0) memset(s->next_picture_base[i], 128, c_size);
169
170 if (s->has_b_frames || s->codec_id==CODEC_ID_MPEG4) {
171 /* Note the MPEG4 stuff is here cuz of buggy encoders which dont set the low_delay flag but
172 do low-delay encoding, so we cant allways distinguish b-frame containing streams from low_delay streams */
173 CHECKED_ALLOCZ(pict, c_size)
174 s->aux_picture_base[i] = pict;
175 s->aux_picture[i] = pict + pict_start;
176 if(i>0) memset(s->aux_picture_base[i], 128, c_size);
177 }
178 }
179 s->ip_buffer_count= 2;
180 }
181
182 CHECKED_ALLOCZ(s->edge_emu_buffer, (s->width+32)*2*17);
183
184 if (s->encoding) {
185 int j;
186 int mv_table_size= (s->mb_width+2)*(s->mb_height+2);
187
188 CHECKED_ALLOCZ(s->mb_var , s->mb_num * sizeof(INT16))
189 CHECKED_ALLOCZ(s->mc_mb_var, s->mb_num * sizeof(INT16))
190
191 /* Allocate MV tables */
192 CHECKED_ALLOCZ(s->p_mv_table , mv_table_size * 2 * sizeof(INT16))
193 CHECKED_ALLOCZ(s->b_forw_mv_table , mv_table_size * 2 * sizeof(INT16))
194 CHECKED_ALLOCZ(s->b_back_mv_table , mv_table_size * 2 * sizeof(INT16))
195 CHECKED_ALLOCZ(s->b_bidir_forw_mv_table , mv_table_size * 2 * sizeof(INT16))
196 CHECKED_ALLOCZ(s->b_bidir_back_mv_table , mv_table_size * 2 * sizeof(INT16))
197 CHECKED_ALLOCZ(s->b_direct_forw_mv_table, mv_table_size * 2 * sizeof(INT16))
198 CHECKED_ALLOCZ(s->b_direct_back_mv_table, mv_table_size * 2 * sizeof(INT16))
199 CHECKED_ALLOCZ(s->b_direct_mv_table , mv_table_size * 2 * sizeof(INT16))
200
201 CHECKED_ALLOCZ(s->me_scratchpad, s->linesize*16*3*sizeof(uint8_t))
202
203 CHECKED_ALLOCZ(s->me_map , ME_MAP_SIZE*sizeof(uint32_t))
204 CHECKED_ALLOCZ(s->me_score_map, ME_MAP_SIZE*sizeof(uint16_t))
205
206 if(s->max_b_frames){
207 for(j=0; j<REORDER_BUFFER_SIZE; j++){
208 int i;
209 for(i=0;i<3;i++) {
210 int w, h, shift;
211
212 w = s->linesize;
213 h = s->mb_height * 16;
214 shift = (i == 0) ? 0 : 1;
215 c_size = (w >> shift) * (h >> shift);
216
217 CHECKED_ALLOCZ(pict, c_size);
218 s->picture_buffer[j][i] = pict;
219 }
220 }
221 }
222
223 if(s->codec_id==CODEC_ID_MPEG4){
224 CHECKED_ALLOCZ(s->tex_pb_buffer, PB_BUFFER_SIZE);
225 CHECKED_ALLOCZ( s->pb2_buffer, PB_BUFFER_SIZE);
226 }
227 }
228
229 if (s->out_format == FMT_H263 || s->encoding) {
230 int size;
231 /* Allocate MB type table */
232 CHECKED_ALLOCZ(s->mb_type , s->mb_num * sizeof(UINT8))
233
234 /* MV prediction */
235 size = (2 * s->mb_width + 2) * (2 * s->mb_height + 2);
236 CHECKED_ALLOCZ(s->motion_val, size * 2 * sizeof(INT16));
237 }
238
239 if (s->h263_pred || s->h263_plus) {
240 int y_size, c_size, i, size;
241
242 /* dc values */
243
244 y_size = (2 * s->mb_width + 2) * (2 * s->mb_height + 2);
245 c_size = (s->mb_width + 2) * (s->mb_height + 2);
246 size = y_size + 2 * c_size;
247 CHECKED_ALLOCZ(s->dc_val[0], size * sizeof(INT16));
248 s->dc_val[1] = s->dc_val[0] + y_size;
249 s->dc_val[2] = s->dc_val[1] + c_size;
250 for(i=0;i<size;i++)
251 s->dc_val[0][i] = 1024;
252
253 /* ac values */
254 CHECKED_ALLOCZ(s->ac_val[0], size * sizeof(INT16) * 16);
255 s->ac_val[1] = s->ac_val[0] + y_size;
256 s->ac_val[2] = s->ac_val[1] + c_size;
257
258 /* cbp values */
259 CHECKED_ALLOCZ(s->coded_block, y_size);
260
261 /* which mb is a intra block */
262 CHECKED_ALLOCZ(s->mbintra_table, s->mb_num);
263 memset(s->mbintra_table, 1, s->mb_num);
264
265 /* divx501 bitstream reorder buffer */
266 CHECKED_ALLOCZ(s->bitstream_buffer, BITSTREAM_BUFFER_SIZE);
267
268 /* cbp, ac_pred, pred_dir */
269 CHECKED_ALLOCZ(s->cbp_table , s->mb_num * sizeof(UINT8))
270 CHECKED_ALLOCZ(s->pred_dir_table, s->mb_num * sizeof(UINT8))
271
272 CHECKED_ALLOCZ(s->qscale_table , s->mb_num * sizeof(UINT8))
273 }
274 /* default structure is frame */
275 s->picture_structure = PICT_FRAME;
276
277 /* init macroblock skip table */
278 CHECKED_ALLOCZ(s->mbskip_table, s->mb_num);
279
280 s->block= s->blocks[0];
281
282 s->context_initialized = 1;
283 return 0;
284 fail:
285 MPV_common_end(s);
286 return -1;
287 }
288
289
290 //extern int sads;
291
292 /* init common structure for both encoder and decoder */
293 void MPV_common_end(MpegEncContext *s)
294 {
295 int i;
296
297 av_freep(&s->mb_type);
298 av_freep(&s->mb_var);
299 av_freep(&s->mc_mb_var);
300 av_freep(&s->p_mv_table);
301 av_freep(&s->b_forw_mv_table);
302 av_freep(&s->b_back_mv_table);
303 av_freep(&s->b_bidir_forw_mv_table);
304 av_freep(&s->b_bidir_back_mv_table);
305 av_freep(&s->b_direct_forw_mv_table);
306 av_freep(&s->b_direct_back_mv_table);
307 av_freep(&s->b_direct_mv_table);
308 av_freep(&s->motion_val);
309 av_freep(&s->dc_val[0]);
310 av_freep(&s->ac_val[0]);
311 av_freep(&s->coded_block);
312 av_freep(&s->mbintra_table);
313 av_freep(&s->cbp_table);
314 av_freep(&s->pred_dir_table);
315 av_freep(&s->qscale_table);
316 av_freep(&s->me_scratchpad);
317 av_freep(&s->me_map);
318 av_freep(&s->me_score_map);
319
320 av_freep(&s->mbskip_table);
321 av_freep(&s->bitstream_buffer);
322 av_freep(&s->tex_pb_buffer);
323 av_freep(&s->pb2_buffer);
324 av_freep(&s->edge_emu_buffer);
325
326 for(i=0;i<3;i++) {
327 int j;
328 if(!(s->flags&CODEC_FLAG_DR1)){
329 av_freep(&s->last_picture_base[i]);
330 av_freep(&s->next_picture_base[i]);
331 av_freep(&s->aux_picture_base[i]);
332 }
333 s->last_picture_base[i]=
334 s->next_picture_base[i]=
335 s->aux_picture_base [i] = NULL;
336 s->last_picture[i]=
337 s->next_picture[i]=
338 s->aux_picture [i] = NULL;
339
340 for(j=0; j<REORDER_BUFFER_SIZE; j++){
341 av_freep(&s->picture_buffer[j][i]);
342 }
343 }
344 s->context_initialized = 0;
345 }
346
347 /* init video encoder */
348 int MPV_encode_init(AVCodecContext *avctx)
349 {
350 MpegEncContext *s = avctx->priv_data;
351 int i;
352
353 avctx->pix_fmt = PIX_FMT_YUV420P;
354
355 s->bit_rate = avctx->bit_rate;
356 s->bit_rate_tolerance = avctx->bit_rate_tolerance;
357 s->frame_rate = avctx->frame_rate;
358 s->width = avctx->width;
359 s->height = avctx->height;
360 if(avctx->gop_size > 600){
361 fprintf(stderr, "Warning keyframe interval too large! reducing it ...\n");
362 avctx->gop_size=600;
363 }
364 s->gop_size = avctx->gop_size;
365 s->rtp_mode = avctx->rtp_mode;
366 s->rtp_payload_size = avctx->rtp_payload_size;
367 if (avctx->rtp_callback)
368 s->rtp_callback = avctx->rtp_callback;
369 s->qmin= avctx->qmin;
370 s->qmax= avctx->qmax;
371 s->max_qdiff= avctx->max_qdiff;
372 s->qcompress= avctx->qcompress;
373 s->qblur= avctx->qblur;
374 s->b_quant_factor= avctx->b_quant_factor;
375 s->b_quant_offset= avctx->b_quant_offset;
376 s->avctx = avctx;
377 s->aspect_ratio_info= avctx->aspect_ratio_info;
378 s->flags= avctx->flags;
379 s->max_b_frames= avctx->max_b_frames;
380 s->rc_strategy= avctx->rc_strategy;
381 s->b_frame_strategy= avctx->b_frame_strategy;
382 s->codec_id= avctx->codec->id;
383 s->luma_elim_threshold = avctx->luma_elim_threshold;
384 s->chroma_elim_threshold= avctx->chroma_elim_threshold;
385 s->strict_std_compliance= avctx->strict_std_compliance;
386 s->data_partitioning= avctx->flags & CODEC_FLAG_PART;
387
388 if (s->gop_size <= 1) {
389 s->intra_only = 1;
390 s->gop_size = 12;
391 } else {
392 s->intra_only = 0;
393 }
394
395 /* ME algorithm */
396 if (avctx->me_method == 0)
397 /* For compatibility */
398 s->me_method = motion_estimation_method;
399 else
400 s->me_method = avctx->me_method;
401
402 /* Fixed QSCALE */
403 s->fixed_qscale = (avctx->flags & CODEC_FLAG_QSCALE);
404
405 switch(avctx->codec->id) {
406 case CODEC_ID_MPEG1VIDEO:
407 s->out_format = FMT_MPEG1;
408 avctx->delay=0; //FIXME not sure, should check the spec
409 break;
410 case CODEC_ID_MJPEG:
411 s->out_format = FMT_MJPEG;
412 s->intra_only = 1; /* force intra only for jpeg */
413 s->mjpeg_write_tables = 1; /* write all tables */
414 s->mjpeg_data_only_frames = 0; /* write all the needed headers */
415 s->mjpeg_vsample[0] = 2; /* set up default sampling factors */
416 s->mjpeg_vsample[1] = 1; /* the only currently supported values */
417 s->mjpeg_vsample[2] = 1;
418 s->mjpeg_hsample[0] = 2;
419 s->mjpeg_hsample[1] = 1;
420 s->mjpeg_hsample[2] = 1;
421 if (mjpeg_init(s) < 0)
422 return -1;
423 avctx->delay=0;
424 break;
425 case CODEC_ID_H263:
426 if (h263_get_picture_format(s->width, s->height) == 7) {
427 printf("Input picture size isn't suitable for h263 codec! try h263+\n");
428 return -1;
429 }
430 s->out_format = FMT_H263;
431 avctx->delay=0;
432 break;
433 case CODEC_ID_H263P:
434 s->out_format = FMT_H263;
435 s->rtp_mode = 1;
436 s->rtp_payload_size = 1200;
437 s->h263_plus = 1;
438 s->unrestricted_mv = 1;
439 s->h263_aic = 1;
440
441 /* These are just to be sure */
442 s->umvplus = 0;
443 s->umvplus_dec = 0;
444 avctx->delay=0;
445 break;
446 case CODEC_ID_RV10:
447 s->out_format = FMT_H263;
448 s->h263_rv10 = 1;
449 avctx->delay=0;
450 break;
451 case CODEC_ID_MPEG4:
452 s->out_format = FMT_H263;
453 s->h263_pred = 1;
454 s->unrestricted_mv = 1;
455 s->has_b_frames= s->max_b_frames ? 1 : 0;
456 s->low_delay=0;
457 avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
458 break;
459 case CODEC_ID_MSMPEG4V1:
460 s->out_format = FMT_H263;
461 s->h263_msmpeg4 = 1;
462 s->h263_pred = 1;
463 s->unrestricted_mv = 1;
464 s->msmpeg4_version= 1;
465 avctx->delay=0;
466 break;
467 case CODEC_ID_MSMPEG4V2:
468 s->out_format = FMT_H263;
469 s->h263_msmpeg4 = 1;
470 s->h263_pred = 1;
471 s->unrestricted_mv = 1;
472 s->msmpeg4_version= 2;
473 avctx->delay=0;
474 break;
475 case CODEC_ID_MSMPEG4V3:
476 s->out_format = FMT_H263;
477 s->h263_msmpeg4 = 1;
478 s->h263_pred = 1;
479 s->unrestricted_mv = 1;
480 s->msmpeg4_version= 3;
481 avctx->delay=0;
482 break;
483 case CODEC_ID_WMV1:
484 s->out_format = FMT_H263;
485 s->h263_msmpeg4 = 1;
486 s->h263_pred = 1;
487 s->unrestricted_mv = 1;
488 s->msmpeg4_version= 4;
489 avctx->delay=0;
490 break;
491 case CODEC_ID_WMV2:
492 s->out_format = FMT_H263;
493 s->h263_msmpeg4 = 1;
494 s->h263_pred = 1;
495 s->unrestricted_mv = 1;
496 s->msmpeg4_version= 5;
497 avctx->delay=0;
498 break;
499 default:
500 return -1;
501 }
502
503 { /* set up some save defaults, some codecs might override them later */
504 static int done=0;
505 if(!done){
506 int i;
507 done=1;
508 memset(default_mv_penalty, 0, sizeof(UINT16)*(MAX_FCODE+1)*(2*MAX_MV+1));
509 memset(default_fcode_tab , 0, sizeof(UINT8)*(2*MAX_MV+1));
510
511 for(i=-16; i<16; i++){
512 default_fcode_tab[i + MAX_MV]= 1;
513 }
514 }
515 }
516 s->mv_penalty= default_mv_penalty;
517 s->fcode_tab= default_fcode_tab;
518 s->y_dc_scale_table=
519 s->c_dc_scale_table= ff_mpeg1_dc_scale_table;
520
521 if (s->out_format == FMT_H263)
522 h263_encode_init(s);
523 else if (s->out_format == FMT_MPEG1)
524 ff_mpeg1_encode_init(s);
525 if(s->msmpeg4_version)
526 ff_msmpeg4_encode_init(s);
527
528 /* dont use mv_penalty table for crap MV as it would be confused */
529 if (s->me_method < ME_EPZS) s->mv_penalty = default_mv_penalty;
530
531 s->encoding = 1;
532
533 /* init */
534 if (MPV_common_init(s) < 0)
535 return -1;
536
537 /* init default q matrix */
538 for(i=0;i<64;i++) {
539 if(s->out_format == FMT_H263)
540 s->intra_matrix[i] = ff_mpeg1_default_non_intra_matrix[i];
541 else
542 s->intra_matrix[i] = ff_mpeg1_default_intra_matrix[i];
543
544 s->inter_matrix[i] = ff_mpeg1_default_non_intra_matrix[i];
545 }
546
547 /* precompute matrix */
548 /* for mjpeg, we do include qscale in the matrix */
549 if (s->out_format != FMT_MJPEG) {
550 convert_matrix(s->q_intra_matrix, s->q_intra_matrix16, s->q_intra_matrix16_bias,
551 s->intra_matrix, s->intra_quant_bias);
552 convert_matrix(s->q_inter_matrix, s->q_inter_matrix16, s->q_inter_matrix16_bias,
553 s->inter_matrix, s->inter_quant_bias);
554 }
555
556 if(ff_rate_control_init(s) < 0)
557 return -1;
558
559 s->picture_number = 0;
560 s->picture_in_gop_number = 0;
561 s->fake_picture_number = 0;
562 /* motion detector init */
563 s->f_code = 1;
564 s->b_code = 1;
565
566 return 0;
567 }
568
569 int MPV_encode_end(AVCodecContext *avctx)
570 {
571 MpegEncContext *s = avctx->priv_data;
572
573 #ifdef STATS
574 print_stats();
575 #endif
576
577 ff_rate_control_uninit(s);
578
579 MPV_common_end(s);
580 if (s->out_format == FMT_MJPEG)
581 mjpeg_close(s);
582
583 return 0;
584 }
585
586 /* draw the edges of width 'w' of an image of size width, height */
587 static void draw_edges_c(UINT8 *buf, int wrap, int width, int height, int w)
588 {
589 UINT8 *ptr, *last_line;
590 int i;
591
592 last_line = buf + (height - 1) * wrap;
593 for(i=0;i<w;i++) {
594 /* top and bottom */
595 memcpy(buf - (i + 1) * wrap, buf, width);
596 memcpy(last_line + (i + 1) * wrap, last_line, width);
597 }
598 /* left and right */
599 ptr = buf;
600 for(i=0;i<height;i++) {
601 memset(ptr - w, ptr[0], w);
602 memset(ptr + width, ptr[width-1], w);
603 ptr += wrap;
604 }
605 /* corners */
606 for(i=0;i<w;i++) {
607 memset(buf - (i + 1) * wrap - w, buf[0], w); /* top left */
608 memset(buf - (i + 1) * wrap + width, buf[width-1], w); /* top right */
609 memset(last_line + (i + 1) * wrap - w, last_line[0], w); /* top left */
610 memset(last_line + (i + 1) * wrap + width, last_line[width-1], w); /* top right */
611 }
612 }
613
614 /* generic function for encode/decode called before a frame is coded/decoded */
615 void MPV_frame_start(MpegEncContext *s, AVCodecContext *avctx)
616 {
617 int i;
618 UINT8 *tmp;
619
620 s->mb_skiped = 0;
621 s->decoding_error=0;
622
623 if(avctx->flags&CODEC_FLAG_DR1){
624 int i;
625 avctx->get_buffer_callback(avctx, s->width, s->height, s->pict_type);
626
627 s->linesize = avctx->dr_stride;
628 s->uvlinesize= avctx->dr_uvstride;
629 s->ip_buffer_count= avctx->dr_ip_buffer_count;
630 }
631
632 if (s->pict_type == B_TYPE) {
633 for(i=0;i<3;i++) {
634 if(avctx->flags&CODEC_FLAG_DR1)
635 s->aux_picture[i]= avctx->dr_buffer[i];
636
637 s->current_picture[i] = s->aux_picture[i];
638 }
639 } else {
640 for(i=0;i<3;i++) {
641 /* swap next and last */
642 if(avctx->flags&CODEC_FLAG_DR1)
643 tmp= avctx->dr_buffer[i];
644 else
645 tmp = s->last_picture[i];
646
647 s->last_picture[i] = s->next_picture[i];
648 s->next_picture[i] = tmp;
649 s->current_picture[i] = tmp;
650
651 s->last_dr_opaque= s->next_dr_opaque;
652 s->next_dr_opaque= avctx->dr_opaque_frame;
653
654 if(s->has_b_frames && s->last_dr_opaque)
655 avctx->dr_opaque_frame= s->last_dr_opaque;
656 else
657 avctx->dr_opaque_frame= s->next_dr_opaque;
658 }
659 }
660 }
661
662 /* generic function for encode/decode called after a frame has been coded/decoded */
663 void MPV_frame_end(MpegEncContext *s)
664 {
665 // if((s->picture_number%100)==0 && s->encoding) printf("sads:%d //\n", sads);
666
667 /* draw edge for correct motion prediction if outside */
668 if (s->pict_type != B_TYPE && !s->intra_only && !(s->flags&CODEC_FLAG_EMU_EDGE)) {
669 if(s->avctx==NULL || s->avctx->codec->id!=CODEC_ID_MPEG4 || s->divx_version>=500){
670 draw_edges(s->current_picture[0], s->linesize, s->mb_width*16, s->mb_height*16, EDGE_WIDTH);
671 draw_edges(s->current_picture[1], s->uvlinesize, s->mb_width*8, s->mb_height*8, EDGE_WIDTH/2);
672 draw_edges(s->current_picture[2], s->uvlinesize, s->mb_width*8, s->mb_height*8, EDGE_WIDTH/2);
673 }else{
674 /* mpeg4? / opendivx / xvid */
675 draw_edges(s->current_picture[0], s->linesize, s->width, s->height, EDGE_WIDTH);
676 draw_edges(s->current_picture[1], s->uvlinesize, s->width/2, s->height/2, EDGE_WIDTH/2);
677 draw_edges(s->current_picture[2], s->uvlinesize, s->width/2, s->height/2, EDGE_WIDTH/2);
678 }
679 }
680 emms_c();
681
682 if(s->pict_type!=B_TYPE){
683 s->last_non_b_pict_type= s->pict_type;
684 s->last_non_b_qscale= s->qscale;
685 s->last_non_b_mc_mb_var= s->mc_mb_var_sum;
686 s->num_available_buffers++;
687 if(s->num_available_buffers>2) s->num_available_buffers= 2;
688 }
689 }
690
691 /* reorder input for encoding */
692 void reorder_input(MpegEncContext *s, AVPicture *pict)
693 {
694 int i, j, index;
695
696 if(s->max_b_frames > FF_MAX_B_FRAMES) s->max_b_frames= FF_MAX_B_FRAMES;
697
698 // delay= s->max_b_frames+1; (or 0 if no b frames cuz decoder diff)
699
700 for(j=0; j<REORDER_BUFFER_SIZE-1; j++){
701 s->coded_order[j]= s->coded_order[j+1];
702 }
703 s->coded_order[j].picture[0]= s->coded_order[j].picture[1]= s->coded_order[j].picture[2]= NULL; //catch uninitalized buffers
704 s->coded_order[j].pict_type=0;
705
706 switch(s->input_pict_type){
707 default:
708 case I_TYPE:
709 case S_TYPE:
710 case P_TYPE:
711 index= s->max_b_frames - s->b_frames_since_non_b;
712 s->b_frames_since_non_b=0;
713 break;
714 case B_TYPE:
715 index= s->max_b_frames + 1;
716 s->b_frames_since_non_b++;
717 break;
718 }
719 //printf("index:%d type:%d strides: %d %d\n", index, s->input_pict_type, pict->linesize[0], s->linesize);
720 if( (index==0 || (s->flags&CODEC_FLAG_INPUT_PRESERVED))
721 && pict->linesize[0] == s->linesize
722 && pict->linesize[1] == s->uvlinesize
723 && pict->linesize[2] == s->uvlinesize){
724 //printf("ptr\n");
725 for(i=0; i<3; i++){
726 s->coded_order[index].picture[i]= pict->data[i];
727 }
728 }else{
729 //printf("copy\n");
730 for(i=0; i<3; i++){
731 uint8_t *src = pict->data[i];
732 uint8_t *dest;
733 int src_wrap = pict->linesize[i];
734 int dest_wrap = s->linesize;
735 int w = s->width;
736 int h = s->height;
737
738 if(index==0) dest= s->last_picture[i]+16; //is current_picture indeed but the switch hapens after reordering
739 else dest= s->picture_buffer[s->picture_buffer_index][i];
740
741 if (i >= 1) {
742 dest_wrap >>= 1;
743 w >>= 1;
744 h >>= 1;
745 }
746
747 s->coded_order[index].picture[i]= dest;
748 for(j=0;j<h;j++) {
749 memcpy(dest, src, w);
750 dest += dest_wrap;
751 src += src_wrap;
752 }
753 }
754 if(index!=0){
755 s->picture_buffer_index++;
756 if(s->picture_buffer_index >= REORDER_BUFFER_SIZE-1) s->picture_buffer_index=0;
757 }
758 }
759 s->coded_order[index].pict_type = s->input_pict_type;
760 s->coded_order[index].qscale = s->input_qscale;
761 s->coded_order[index].force_type= s->force_input_type;
762 s->coded_order[index].picture_in_gop_number= s->input_picture_in_gop_number;
763 s->coded_order[index].picture_number= s->input_picture_number;
764
765 for(i=0; i<3; i++){
766 s->new_picture[i]= s->coded_order[0].picture[i];
767 }
768 }
769
770 int MPV_encode_picture(AVCodecContext *avctx,
771 unsigned char *buf, int buf_size, void *data)
772 {
773 MpegEncContext *s = avctx->priv_data;
774 AVPicture *pict = data;
775
776 s->input_qscale = avctx->quality;
777
778 init_put_bits(&s->pb, buf, buf_size, NULL, NULL);
779
780 if(avctx->flags&CODEC_FLAG_TYPE){
781 s->input_pict_type=
782 s->force_input_type= avctx->key_frame ? I_TYPE : P_TYPE;
783 }else if(s->flags&CODEC_FLAG_PASS2){
784 s->input_pict_type=
785 s->force_input_type= s->rc_context.entry[s->input_picture_number].new_pict_type;
786 }else{
787 s->force_input_type=0;
788 if (!s->intra_only) {
789 /* first picture of GOP is intra */
790 if (s->input_picture_in_gop_number % s->gop_size==0){
791 s->input_pict_type = I_TYPE;
792 }else if(s->max_b_frames==0){
793 s->input_pict_type = P_TYPE;
794 }else{
795 if(s->b_frames_since_non_b < s->max_b_frames) //FIXME more IQ
796 s->input_pict_type = B_TYPE;
797 else
798 s->input_pict_type = P_TYPE;
799 }
800 } else {
801 s->input_pict_type = I_TYPE;
802 }
803 }
804
805 if(s->input_pict_type==I_TYPE)
806 s->input_picture_in_gop_number=0;
807
808 reorder_input(s, pict);
809
810 /* output? */
811 if(s->coded_order[0].picture[0]){
812
813 s->pict_type= s->coded_order[0].pict_type;
814 if (s->fixed_qscale) /* the ratecontrol needs the last qscale so we dont touch it for CBR */
815 s->qscale= s->coded_order[0].qscale;
816 s->force_type= s->coded_order[0].force_type;
817 s->picture_in_gop_number= s->coded_order[0].picture_in_gop_number;
818 s->picture_number= s->coded_order[0].picture_number;
819
820 MPV_frame_start(s, avctx);
821
822 encode_picture(s, s->picture_number);
823 avctx->key_frame = (s->pict_type == I_TYPE);
824 avctx->pict_type = s->pict_type;
825 avctx->real_pict_num = s->picture_number;
826 avctx->header_bits = s->header_bits;
827 avctx->mv_bits = s->mv_bits;
828 avctx->misc_bits = s->misc_bits;
829 avctx->i_tex_bits = s->i_tex_bits;
830 avctx->p_tex_bits = s->p_tex_bits;
831 avctx->i_count = s->i_count;
832 avctx->p_count = s->p_count;
833 avctx->skip_count = s->skip_count;
834
835 MPV_frame_end(s);
836
837 if (s->out_format == FMT_MJPEG)
838 mjpeg_picture_trailer(s);
839
840 avctx->quality = s->qscale;
841
842 if(s->flags&CODEC_FLAG_PASS1)
843 ff_write_pass1_stats(s);
844
845 }
846
847 s->input_picture_number++;
848 s->input_picture_in_gop_number++;
849
850 flush_put_bits(&s->pb);
851 s->frame_bits = (pbBufPtr(&s->pb) - s->pb.buf) * 8;
852 if(s->pict_type==B_TYPE) s->pb_frame_bits+= s->frame_bits;
853 else s->pb_frame_bits= s->frame_bits;
854
855 s->total_bits += s->frame_bits;
856 avctx->frame_bits = s->frame_bits;
857 //printf("fcode: %d, type: %d, head: %d, mv: %d, misc: %d, frame: %d, itex: %d, ptex: %d\n",
858 //s->f_code, avctx->key_frame, s->header_bits, s->mv_bits, s->misc_bits, s->frame_bits, s->i_tex_bits, s->p_tex_bits);
859
860 if (avctx->get_psnr) {
861 /* At this point pict->data should have the original frame */
862 /* an s->current_picture should have the coded/decoded frame */
863 get_psnr(pict->data, s->current_picture,
864 pict->linesize, s->linesize, avctx);
865 // printf("%f\n", avctx->psnr_y);
866 }
867 return pbBufPtr(&s->pb) - s->pb.buf;
868 }
869
870 static inline void gmc1_motion(MpegEncContext *s,
871 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr,
872 int dest_offset,
873 UINT8 **ref_picture, int src_offset,
874 int h)
875 {
876 UINT8 *ptr;
877 int offset, src_x, src_y, linesize, uvlinesize;
878 int motion_x, motion_y;
879
880 if(s->real_sprite_warping_points>1) printf("more than 1 warp point isnt supported\n");
881 motion_x= s->sprite_offset[0][0];
882 motion_y= s->sprite_offset[0][1];
883 src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy+1));
884 src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy+1));
885 motion_x<<=(3-s->sprite_warping_accuracy);
886 motion_y<<=(3-s->sprite_warping_accuracy);
887 src_x = clip(src_x, -16, s->width);
888 if (src_x == s->width)
889 motion_x =0;
890 src_y = clip(src_y, -16, s->height);
891 if (src_y == s->height)
892 motion_y =0;
893
894 linesize = s->linesize;
895 uvlinesize = s->uvlinesize;
896 ptr = ref_picture[0] + (src_y * linesize) + src_x + src_offset;
897
898 dest_y+=dest_offset;
899 gmc1(dest_y , ptr , linesize, h, motion_x&15, motion_y&15, s->no_rounding);
900 gmc1(dest_y+8, ptr+8, linesize, h, motion_x&15, motion_y&15, s->no_rounding);
901
902 motion_x= s->sprite_offset[1][0];
903 motion_y= s->sprite_offset[1][1];
904 src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy+1));
905 src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy+1));
906 motion_x<<=(3-s->sprite_warping_accuracy);
907 motion_y<<=(3-s->sprite_warping_accuracy);
908 src_x = clip(src_x, -8, s->width>>1);
909 if (src_x == s->width>>1)
910 motion_x =0;
911 src_y = clip(src_y, -8, s->height>>1);
912 if (src_y == s->height>>1)
913 motion_y =0;
914
915 offset = (src_y * uvlinesize) + src_x + (src_offset>>1);
916 ptr = ref_picture[1] + offset;
917 gmc1(dest_cb + (dest_offset>>1), ptr, uvlinesize, h>>1, motion_x&15, motion_y&15, s->no_rounding);
918 ptr = ref_picture[2] + offset;
919 gmc1(dest_cr + (dest_offset>>1), ptr, uvlinesize, h>>1, motion_x&15, motion_y&15, s->no_rounding);
920
921 return;
922 }
923
924 static void emulated_edge_mc(UINT8 *buf, UINT8 *src, int linesize, int block_w, int block_h,
925 int src_x, int src_y, int w, int h){
926 int x, y;
927 int start_y, start_x, end_y, end_x;
928
929 if(src_y>= h){
930 src+= (h-1-src_y)*linesize;
931 src_y=h-1;
932 }else if(src_y<=-block_h){
933 src+= (1-block_h-src_y)*linesize;
934 src_y=1-block_h;
935 }
936 if(src_x>= w){
937 src+= (w-1-src_x);
938 src_x=w-1;
939 }else if(src_x<=-block_w){
940 src+= (1-block_w-src_x);
941 src_x=1-block_w;
942 }
943
944 start_y= MAX(0, -src_y);
945 start_x= MAX(0, -src_x);
946 end_y= MIN(block_h, h-src_y);
947 end_x= MIN(block_w, w-src_x);
948
949 // copy existing part
950 for(y=start_y; y<end_y; y++){
951 for(x=start_x; x<end_x; x++){
952 buf[x + y*linesize]= src[x + y*linesize];
953 }
954 }
955
956 //top
957 for(y=0; y<start_y; y++){
958 for(x=start_x; x<end_x; x++){
959 buf[x + y*linesize]= buf[x + start_y*linesize];
960 }
961 }
962
963 //bottom
964 for(y=end_y; y<block_h; y++){
965 for(x=start_x; x<end_x; x++){
966 buf[x + y*linesize]= buf[x + (end_y-1)*linesize];
967 }
968 }
969
970 for(y=0; y<block_h; y++){
971 //left
972 for(x=0; x<start_x; x++){
973 buf[x + y*linesize]= buf[start_x + y*linesize];
974 }
975
976 //right
977 for(x=end_x; x<block_w; x++){
978 buf[x + y*linesize]= buf[end_x - 1 + y*linesize];
979 }
980 }
981 }
982
983
984 /* apply one mpeg motion vector to the three components */
985 static inline void mpeg_motion(MpegEncContext *s,
986 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr,
987 int dest_offset,
988 UINT8 **ref_picture, int src_offset,
989 int field_based, op_pixels_func *pix_op,
990 int motion_x, int motion_y, int h)
991 {
992 UINT8 *ptr;
993 int dxy, offset, mx, my, src_x, src_y, height, linesize;
994 int emu=0;
995
996 if(s->quarter_sample)
997 {
998 motion_x>>=1;
999 motion_y>>=1;
1000 }
1001 dxy = ((motion_y & 1) << 1) | (motion_x & 1);
1002 src_x = s->mb_x * 16 + (motion_x >> 1);
1003 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 1);
1004
1005 /* WARNING: do no forget half pels */
1006 height = s->height >> field_based;
1007 src_x = clip(src_x, -16, s->width);
1008 if (src_x == s->width)
1009 dxy &= ~1;
1010 src_y = clip(src_y, -16, height);
1011 if (src_y == height)
1012 dxy &= ~2;
1013 linesize = s->linesize << field_based;
1014 ptr = ref_picture[0] + (src_y * linesize) + (src_x) + src_offset;
1015 dest_y += dest_offset;
1016
1017 if(s->flags&CODEC_FLAG_EMU_EDGE){
1018 if(src_x<0 || src_y<0 || src_x + (motion_x&1) + 16 > s->width
1019 || src_y + (motion_y&1) + h > height){
1020 emulated_edge_mc(s->edge_emu_buffer, ptr, linesize, 17, h+1, src_x, src_y, s->width, height);
1021 ptr= s->edge_emu_buffer;
1022 emu=1;
1023 }
1024 }
1025 pix_op[dxy](dest_y, ptr, linesize, h);
1026 pix_op[dxy](dest_y + 8, ptr + 8, linesize, h);
1027
1028 if(s->flags&CODEC_FLAG_GRAY) return;
1029
1030 if (s->out_format == FMT_H263) {
1031 dxy = 0;
1032 if ((motion_x & 3) != 0)
1033 dxy |= 1;
1034 if ((motion_y & 3) != 0)
1035 dxy |= 2;
1036 mx = motion_x >> 2;
1037 my = motion_y >> 2;
1038 } else {
1039 mx = motion_x / 2;
1040 my = motion_y / 2;
1041 dxy = ((my & 1) << 1) | (mx & 1);
1042 mx >>= 1;
1043 my >>= 1;
1044 }
1045
1046 src_x = s->mb_x * 8 + mx;
1047 src_y = s->mb_y * (8 >> field_based) + my;
1048 src_x = clip(src_x, -8, s->width >> 1);
1049 if (src_x == (s->width >> 1))
1050 dxy &= ~1;
1051 src_y = clip(src_y, -8, height >> 1);
1052 if (src_y == (height >> 1))
1053 dxy &= ~2;
1054
1055 offset = (src_y * s->uvlinesize) + src_x + (src_offset >> 1);
1056 ptr = ref_picture[1] + offset;
1057 if(emu){
1058 emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, 9, (h>>1)+1, src_x, src_y, s->width>>1, height>>1);
1059 ptr= s->edge_emu_buffer;
1060 }
1061 pix_op[dxy](dest_cb + (dest_offset >> 1), ptr, s->uvlinesize, h >> 1);
1062
1063 ptr = ref_picture[2] + offset;
1064 if(emu){
1065 emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, 9, (h>>1)+1, src_x, src_y, s->width>>1, height>>1);
1066 ptr= s->edge_emu_buffer;
1067 }
1068 pix_op[dxy](dest_cr + (dest_offset >> 1), ptr, s->uvlinesize, h >> 1);
1069 }
1070
1071 static inline void qpel_motion(MpegEncContext *s,
1072 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr,
1073 int dest_offset,
1074 UINT8 **ref_picture, int src_offset,
1075 int field_based, op_pixels_func *pix_op,
1076 qpel_mc_func *qpix_op,
1077 int motion_x, int motion_y, int h)
1078 {
1079 UINT8 *ptr;
1080 int dxy, offset, mx, my, src_x, src_y, height, linesize;
1081 int emu=0;
1082
1083 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
1084 src_x = s->mb_x * 16 + (motion_x >> 2);
1085 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2);
1086
1087 height = s->height >> field_based;
1088 src_x = clip(src_x, -16, s->width);
1089 if (src_x == s->width)
1090 dxy &= ~3;
1091 src_y = clip(src_y, -16, height);
1092 if (src_y == height)
1093 dxy &= ~12;
1094 linesize = s->linesize << field_based;
1095 ptr = ref_picture[0] + (src_y * linesize) + src_x + src_offset;
1096 dest_y += dest_offset;
1097 //printf("%d %d %d\n", src_x, src_y, dxy);
1098
1099 if(s->flags&CODEC_FLAG_EMU_EDGE){
1100 if(src_x<0 || src_y<0 || src_x + (motion_x&3) + 16 > s->width
1101 || src_y + (motion_y&3) + h > height){
1102 emulated_edge_mc(s->edge_emu_buffer, ptr, linesize, 17, h+1, src_x, src_y, s->width, height);
1103 ptr= s->edge_emu_buffer;
1104 emu=1;
1105 }
1106 }
1107 qpix_op[dxy](dest_y , ptr , linesize, linesize, motion_x&3, motion_y&3);
1108 qpix_op[dxy](dest_y + 8, ptr + 8, linesize, linesize, motion_x&3, motion_y&3);
1109 qpix_op[dxy](dest_y + linesize*8 , ptr + linesize*8 , linesize, linesize, motion_x&3, motion_y&3);
1110 qpix_op[dxy](dest_y + linesize*8 + 8, ptr + linesize*8 + 8, linesize, linesize, motion_x&3, motion_y&3);
1111
1112 if(s->flags&CODEC_FLAG_GRAY) return;
1113
1114 mx= (motion_x>>1) | (motion_x&1);
1115 my= (motion_y>>1) | (motion_y&1);
1116
1117 dxy = 0;
1118 if ((mx & 3) != 0)
1119 dxy |= 1;
1120 if ((my & 3) != 0)
1121 dxy |= 2;
1122 mx = mx >> 2;
1123 my = my >> 2;
1124
1125 src_x = s->mb_x * 8 + mx;
1126 src_y = s->mb_y * (8 >> field_based) + my;
1127 src_x = clip(src_x, -8, s->width >> 1);
1128 if (src_x == (s->width >> 1))
1129 dxy &= ~1;
1130 src_y = clip(src_y, -8, height >> 1);
1131 if (src_y == (height >> 1))
1132 dxy &= ~2;
1133
1134 offset = (src_y * s->uvlinesize) + src_x + (src_offset >> 1);
1135 ptr = ref_picture[1] + offset;
1136 if(emu){
1137 emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, 9, (h>>1)+1, src_x, src_y, s->width>>1, height>>1);
1138 ptr= s->edge_emu_buffer;
1139 }
1140 pix_op[dxy](dest_cb + (dest_offset >> 1), ptr, s->uvlinesize, h >> 1);
1141
1142 ptr = ref_picture[2] + offset;
1143 if(emu){
1144 emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, 9, (h>>1)+1, src_x, src_y, s->width>>1, height>>1);
1145 ptr= s->edge_emu_buffer;
1146 }
1147 pix_op[dxy](dest_cr + (dest_offset >> 1), ptr, s->uvlinesize, h >> 1);
1148 }
1149
1150
1151 static inline void MPV_motion(MpegEncContext *s,
1152 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr,
1153 int dir, UINT8 **ref_picture,
1154 op_pixels_func *pix_op, qpel_mc_func *qpix_op)
1155 {
1156 int dxy, offset, mx, my, src_x, src_y, motion_x, motion_y;
1157 int mb_x, mb_y, i;
1158 UINT8 *ptr, *dest;
1159 int emu=0;
1160
1161 mb_x = s->mb_x;
1162 mb_y = s->mb_y;
1163
1164 switch(s->mv_type) {
1165 case MV_TYPE_16X16:
1166 if(s->mcsel){
1167 #if 0
1168 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
1169 ref_picture, 0,
1170 0, pix_op,
1171 s->sprite_offset[0][0]>>3,
1172 s->sprite_offset[0][1]>>3,
1173 16);
1174 #else
1175 gmc1_motion(s, dest_y, dest_cb, dest_cr, 0,
1176 ref_picture, 0,
1177 16);
1178 #endif
1179 }else if(s->quarter_sample && dir==0){ //FIXME
1180 qpel_motion(s, dest_y, dest_cb, dest_cr, 0,
1181 ref_picture, 0,
1182 0, pix_op, qpix_op,
1183 s->mv[dir][0][0], s->mv[dir][0][1], 16);
1184 }else{
1185 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
1186 ref_picture, 0,
1187 0, pix_op,
1188 s->mv[dir][0][0], s->mv[dir][0][1], 16);
1189 }
1190 break;
1191 case MV_TYPE_8X8:
1192 for(i=0;i<4;i++) {
1193 motion_x = s->mv[dir][i][0];
1194 motion_y = s->mv[dir][i][1];
1195
1196 dxy = ((motion_y & 1) << 1) | (motion_x & 1);
1197 src_x = mb_x * 16 + (motion_x >> 1) + (i & 1) * 8;
1198 src_y = mb_y * 16 + (motion_y >> 1) + (i >>1) * 8;
1199
1200 /* WARNING: do no forget half pels */
1201 src_x = clip(src_x, -16, s->width);
1202 if (src_x == s->width)
1203 dxy &= ~1;
1204 src_y = clip(src_y, -16, s->height);
1205 if (src_y == s->height)
1206 dxy &= ~2;
1207
1208 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x);
1209 if(s->flags&CODEC_FLAG_EMU_EDGE){
1210 if(src_x<0 || src_y<0 || src_x + (motion_x&1) + 8 > s->width
1211 || src_y + (motion_y&1) + 8 > s->height){
1212 emulated_edge_mc(s->edge_emu_buffer, ptr, s->linesize, 9, 9, src_x, src_y, s->width, s->height);
1213 ptr= s->edge_emu_buffer;
1214 }
1215 }
1216 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize;
1217 pix_op[dxy](dest, ptr, s->linesize, 8);
1218 }
1219
1220 if(s->flags&CODEC_FLAG_GRAY) break;
1221 /* In case of 8X8, we construct a single chroma motion vector
1222 with a special rounding */
1223 mx = 0;
1224 my = 0;
1225 for(i=0;i<4;i++) {
1226 mx += s->mv[dir][i][0];
1227 my += s->mv[dir][i][1];
1228 }
1229 if (mx >= 0)
1230 mx = (h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1));
1231 else {
1232 mx = -mx;
1233 mx = -(h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1));
1234 }
1235 if (my >= 0)
1236 my = (h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1));
1237 else {
1238 my = -my;
1239 my = -(h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1));
1240 }
1241 dxy = ((my & 1) << 1) | (mx & 1);
1242 mx >>= 1;
1243 my >>= 1;
1244
1245 src_x = mb_x * 8 + mx;
1246 src_y = mb_y * 8 + my;
1247 src_x = clip(src_x, -8, s->width/2);
1248 if (src_x == s->width/2)
1249 dxy &= ~1;
1250 src_y = clip(src_y, -8, s->height/2);
1251 if (src_y == s->height/2)
1252 dxy &= ~2;
1253
1254 offset = (src_y * (s->uvlinesize)) + src_x;
1255 ptr = ref_picture[1] + offset;
1256 if(s->flags&CODEC_FLAG_EMU_EDGE){
1257 if(src_x<0 || src_y<0 || src_x + (dxy &1) + 8 > s->width >>1
1258 || src_y + (dxy>>1) + 8 > s->height>>1){
1259 emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, 9, 9, src_x, src_y, s->width>>1, s->height>>1);
1260 ptr= s->edge_emu_buffer;
1261 emu=1;
1262 }
1263 }
1264 pix_op[dxy](dest_cb, ptr, s->uvlinesize, 8);
1265
1266 ptr = ref_picture[2] + offset;
1267 if(emu){
1268 emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, 9, 9, src_x, src_y, s->width>>1, s->height>>1);
1269 ptr= s->edge_emu_buffer;
1270 }
1271 pix_op[dxy](dest_cr, ptr, s->uvlinesize, 8);
1272 break;
1273 case MV_TYPE_FIELD:
1274 if (s->picture_structure == PICT_FRAME) {
1275 /* top field */
1276 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
1277 ref_picture, s->field_select[dir][0] ? s->linesize : 0,
1278 1, pix_op,
1279 s->mv[dir][0][0], s->mv[dir][0][1], 8);
1280 /* bottom field */
1281 mpeg_motion(s, dest_y, dest_cb, dest_cr, s->linesize,
1282 ref_picture, s->field_select[dir][1] ? s->linesize : 0,
1283 1, pix_op,
1284 s->mv[dir][1][0], s->mv[dir][1][1], 8);
1285 } else {
1286
1287
1288 }
1289 break;
1290 }
1291 }
1292
1293
1294 /* put block[] to dest[] */
1295 static inline void put_dct(MpegEncContext *s,
1296 DCTELEM *block, int i, UINT8 *dest, int line_size)
1297 {
1298 if (!s->mpeg2)
1299 s->dct_unquantize(s, block, i, s->qscale);
1300 ff_idct_put (dest, line_size, block);
1301 }
1302
1303 /* add block[] to dest[] */
1304 static inline void add_dct(MpegEncContext *s,
1305 DCTELEM *block, int i, UINT8 *dest, int line_size)
1306 {
1307 if (s->block_last_index[i] >= 0) {
1308 ff_idct_add (dest, line_size, block);
1309 }
1310 }
1311
1312 static inline void add_dequant_dct(MpegEncContext *s,
1313 DCTELEM *block, int i, UINT8 *dest, int line_size)
1314 {
1315 if (s->block_last_index[i] >= 0) {
1316 s->dct_unquantize(s, block, i, s->qscale);
1317
1318 ff_idct_add (dest, line_size, block);
1319 }
1320 }
1321
1322 /**
1323 * cleans dc, ac, coded_block for the current non intra MB
1324 */
1325 void ff_clean_intra_table_entries(MpegEncContext *s)
1326 {
1327 int wrap = s->block_wrap[0];
1328 int xy = s->block_index[0];
1329
1330 s->dc_val[0][xy ] =
1331 s->dc_val[0][xy + 1 ] =
1332 s->dc_val[0][xy + wrap] =
1333 s->dc_val[0][xy + 1 + wrap] = 1024;
1334 /* ac pred */
1335 memset(s->ac_val[0][xy ], 0, 32 * sizeof(INT16));
1336 memset(s->ac_val[0][xy + wrap], 0, 32 * sizeof(INT16));
1337 if (s->msmpeg4_version>=3) {
1338 s->coded_block[xy ] =
1339 s->coded_block[xy + 1 ] =
1340 s->coded_block[xy + wrap] =
1341 s->coded_block[xy + 1 + wrap] = 0;
1342 }
1343 /* chroma */
1344 wrap = s->block_wrap[4];
1345 xy = s->mb_x + 1 + (s->mb_y + 1) * wrap;
1346 s->dc_val[1][xy] =
1347 s->dc_val[2][xy] = 1024;
1348 /* ac pred */
1349 memset(s->ac_val[1][xy], 0, 16 * sizeof(INT16));
1350 memset(s->ac_val[2][xy], 0, 16 * sizeof(INT16));
1351
1352 s->mbintra_table[s->mb_x + s->mb_y*s->mb_width]= 0;
1353 }
1354
1355 /* generic function called after a macroblock has been parsed by the
1356 decoder or after it has been encoded by the encoder.
1357
1358 Important variables used:
1359 s->mb_intra : true if intra macroblock
1360 s->mv_dir : motion vector direction
1361 s->mv_type : motion vector type
1362 s->mv : motion vector
1363 s->interlaced_dct : true if interlaced dct used (mpeg2)
1364 */
1365 void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64])
1366 {
1367 int mb_x, mb_y;
1368 const int mb_xy = s->mb_y * s->mb_width + s->mb_x;
1369
1370 mb_x = s->mb_x;
1371 mb_y = s->mb_y;
1372
1373 #ifdef FF_POSTPROCESS
1374 /* Obsolete. Exists for compatibility with mplayer only. */
1375 quant_store[mb_y][mb_x]=s->qscale;
1376 //printf("[%02d][%02d] %d\n",mb_x,mb_y,s->qscale);
1377 #else
1378 if(s->avctx->quant_store) s->avctx->quant_store[mb_y*s->avctx->qstride+mb_x] = s->qscale;
1379 #endif
1380
1381 /* update DC predictors for P macroblocks */
1382 if (!s->mb_intra) {
1383 if (s->h263_pred || s->h263_aic) {
1384 if(s->mbintra_table[mb_xy])
1385 ff_clean_intra_table_entries(s);
1386 } else {
1387 s->last_dc[0] =
1388 s->last_dc[1] =
1389 s->last_dc[2] = 128 << s->intra_dc_precision;
1390 }
1391 }
1392 else if (s->h263_pred || s->h263_aic)
1393 s->mbintra_table[mb_xy]=1;
1394
1395 /* update motion predictor, not for B-frames as they need the motion_val from the last P/S-Frame */
1396 if (s->out_format == FMT_H263 && s->pict_type!=B_TYPE) { //FIXME move into h263.c if possible, format specific stuff shouldnt be here
1397 int motion_x, motion_y;
1398
1399 const int wrap = s->block_wrap[0];
1400 const int xy = s->block_index[0];
1401 if (s->mb_intra) {
1402 motion_x = 0;
1403 motion_y = 0;
1404 goto motion_init;
1405 } else if (s->mv_type == MV_TYPE_16X16) {
1406 motion_x = s->mv[0][0][0];
1407 motion_y = s->mv[0][0][1];
1408 motion_init:
1409 /* no update if 8X8 because it has been done during parsing */
1410 s->motion_val[xy][0] = motion_x;
1411 s->motion_val[xy][1] = motion_y;
1412 s->motion_val[xy + 1][0] = motion_x;
1413 s->motion_val[xy + 1][1] = motion_y;
1414 s->motion_val[xy + wrap][0] = motion_x;
1415 s->motion_val[xy + wrap][1] = motion_y;
1416 s->motion_val[xy + 1 + wrap][0] = motion_x;
1417 s->motion_val[xy + 1 + wrap][1] = motion_y;
1418 }
1419 }
1420
1421 if (!(s->encoding && (s->intra_only || s->pict_type==B_TYPE))) {
1422 UINT8 *dest_y, *dest_cb, *dest_cr;
1423 int dct_linesize, dct_offset;
1424 op_pixels_func *op_pix;
1425 qpel_mc_func *op_qpix;
1426
1427 /* avoid copy if macroblock skipped in last frame too
1428 dont touch it for B-frames as they need the skip info from the next p-frame */
1429 if (s->pict_type != B_TYPE) {
1430 UINT8 *mbskip_ptr = &s->mbskip_table[mb_xy];
1431 if (s->mb_skiped) {
1432 s->mb_skiped = 0;
1433
1434 (*mbskip_ptr) ++; /* indicate that this time we skiped it */
1435 if(*mbskip_ptr >99) *mbskip_ptr= 99;
1436
1437 /* if previous was skipped too, then nothing to do !
1438 skip only during decoding as we might trash the buffers during encoding a bit */
1439 if (*mbskip_ptr >= s->ip_buffer_count && !s->encoding)
1440 goto the_end;
1441 } else {
1442 *mbskip_ptr = 0; /* not skipped */
1443 }
1444 }
1445
1446 dest_y = s->current_picture[0] + (mb_y * 16 * s->linesize) + mb_x * 16;
1447 dest_cb = s->current_picture[1] + (mb_y * 8 * (s->uvlinesize)) + mb_x * 8;
1448 dest_cr = s->current_picture[2] + (mb_y * 8 * (s->uvlinesize)) + mb_x * 8;
1449
1450 if (s->interlaced_dct) {
1451 dct_linesize = s->linesize * 2;
1452 dct_offset = s->linesize;
1453 } else {
1454 dct_linesize = s->linesize;
1455 dct_offset = s->linesize * 8;
1456 }
1457
1458 if (!s->mb_intra) {
1459 /* motion handling */
1460 /* decoding or more than one mb_type (MC was allready done otherwise) */
1461 if((!s->encoding) || (s->mb_type[mb_xy]&(s->mb_type[mb_xy]-1))){
1462 if ((!s->no_rounding) || s->pict_type==B_TYPE){
1463 op_pix = put_pixels_tab;
1464 op_qpix= qpel_mc_rnd_tab;
1465 }else{
1466 op_pix = put_no_rnd_pixels_tab;
1467 op_qpix= qpel_mc_no_rnd_tab;
1468 }
1469
1470 if (s->mv_dir & MV_DIR_FORWARD) {
1471 MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture, op_pix, op_qpix);
1472 if ((!s->no_rounding) || s->pict_type==B_TYPE)
1473 op_pix = avg_pixels_tab;
1474 else
1475 op_pix = avg_no_rnd_pixels_tab;
1476 }
1477 if (s->mv_dir & MV_DIR_BACKWARD) {
1478 MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture, op_pix, op_qpix);
1479 }
1480 }
1481
1482 /* skip dequant / idct if we are really late ;) */
1483 if(s->hurry_up>1) goto the_end;
1484
1485 /* add dct residue */
1486 if(s->encoding || !(s->mpeg2 || s->h263_msmpeg4 || s->codec_id==CODEC_ID_MPEG4)){
1487 add_dequant_dct(s, block[0], 0, dest_y, dct_linesize);
1488 add_dequant_dct(s, block[1], 1, dest_y + 8, dct_linesize);
1489 add_dequant_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize);
1490 add_dequant_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize);
1491
1492 if(!(s->flags&CODEC_FLAG_GRAY)){
1493 add_dequant_dct(s, block[4], 4, dest_cb, s->uvlinesize);
1494 add_dequant_dct(s, block[5], 5, dest_cr, s->uvlinesize);
1495 }
1496 } else {
1497 add_dct(s, block[0], 0, dest_y, dct_linesize);
1498 add_dct(s, block[1], 1, dest_y + 8, dct_linesize);
1499 add_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize);
1500 add_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize);
1501
1502 if(!(s->flags&CODEC_FLAG_GRAY)){
1503 add_dct(s, block[4], 4, dest_cb, s->uvlinesize);
1504 add_dct(s, block[5], 5, dest_cr, s->uvlinesize);
1505 }
1506 }
1507 } else {
1508 /* dct only in intra block */
1509 put_dct(s, block[0], 0, dest_y, dct_linesize);
1510 put_dct(s, block[1], 1, dest_y + 8, dct_linesize);
1511 put_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize);
1512 put_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize);
1513
1514 if(!(s->flags&CODEC_FLAG_GRAY)){
1515 put_dct(s, block[4], 4, dest_cb, s->uvlinesize);
1516 put_dct(s, block[5], 5, dest_cr, s->uvlinesize);
1517 }
1518 }
1519 }
1520 the_end:
1521 emms_c(); //FIXME remove
1522 }
1523
1524 static inline void dct_single_coeff_elimination(MpegEncContext *s, int n, int threshold, int skip_dc)
1525 {
1526 static const char tab[64]=
1527 {3,2,2,1,1,1,1,1,
1528 1,1,1,1,1,1,1,1,
1529 1,1,1,1,1,1,1,1,
1530 0,0,0,0,0,0,0,0,
1531 0,0,0,0,0,0,0,0,
1532 0,0,0,0,0,0,0,0,
1533 0,0,0,0,0,0,0,0,
1534 0,0,0,0,0,0,0,0};
1535 int score=0;
1536 int run=0;
1537 int i;
1538 DCTELEM *block= s->block[n];
1539 const int last_index= s->block_last_index[n];
1540
1541 if(skip_dc) skip_dc=1;
1542
1543 /* are all which we could set to zero are allready zero? */
1544 if(last_index<=skip_dc - 1) return;
1545
1546 for(i=0; i<=last_index; i++){
1547 const int j = zigzag_direct[i];
1548 const int level = ABS(block[j]);
1549 if(level==1){
1550 if(skip_dc && i==0) continue;
1551 score+= tab[run];
1552 run=0;
1553 }else if(level>1){
1554 return;
1555 }else{
1556 run++;
1557 }
1558 }
1559 if(score >= threshold) return;
1560 for(i=skip_dc; i<=last_index; i++){
1561 const int j = zigzag_direct[i];
1562 block[j]=0;
1563 }
1564 if(block[0]) s->block_last_index[n]= 0;
1565 else s->block_last_index[n]= -1;
1566 }
1567
1568 static inline void clip_coeffs(MpegEncContext *s, DCTELEM *block, int last_index)
1569 {
1570 int i;
1571 const int maxlevel= s->max_qcoeff;
1572 const int minlevel= s->min_qcoeff;
1573
1574 for(i=0;i<=last_index; i++){
1575 const int j = zigzag_direct[i];
1576 int level = block[j];
1577
1578 if (level>maxlevel) level=maxlevel;
1579 else if(level<minlevel) level=minlevel;
1580 block[j]= level;
1581 }
1582 }
1583
1584 static void encode_mb(MpegEncContext *s, int motion_x, int motion_y)
1585 {
1586 const int mb_x= s->mb_x;
1587 const int mb_y= s->mb_y;
1588 int i;
1589 int skip_dct[6];
1590 #if 0
1591 if (s->interlaced_dct) {
1592 dct_linesize = s->linesize * 2;
1593 dct_offset = s->linesize;
1594 } else {
1595 dct_linesize = s->linesize;
1596 dct_offset = s->linesize * 8;
1597 }
1598 #endif
1599 for(i=0; i<6; i++) skip_dct[i]=0;
1600
1601 if (s->mb_intra) {
1602 UINT8 *ptr;
1603 int wrap;
1604
1605 wrap = s->linesize;
1606 ptr = s->new_picture[0] + (mb_y * 16 * wrap) + mb_x * 16;
1607 get_pixels(s->block[0], ptr , wrap);
1608 get_pixels(s->block[1], ptr + 8, wrap);
1609 get_pixels(s->block[2], ptr + 8 * wrap , wrap);
1610 get_pixels(s->block[3], ptr + 8 * wrap + 8, wrap);
1611
1612 if(s->flags&CODEC_FLAG_GRAY){
1613 skip_dct[4]= 1;
1614 skip_dct[5]= 1;
1615 }else{
1616 wrap >>=1;
1617 ptr = s->new_picture[1] + (mb_y * 8 * wrap) + mb_x * 8;
1618 get_pixels(s->block[4], ptr, wrap);
1619
1620 ptr = s->new_picture[2] + (mb_y * 8 * wrap) + mb_x * 8;
1621 get_pixels(s->block[5], ptr, wrap);
1622 }
1623 }else{
1624 op_pixels_func *op_pix;
1625 qpel_mc_func *op_qpix;
1626 UINT8 *dest_y, *dest_cb, *dest_cr;
1627 UINT8 *ptr_y, *ptr_cb, *ptr_cr;
1628 int wrap_y, wrap_c;
1629
1630 dest_y = s->current_picture[0] + (mb_y * 16 * s->linesize ) + mb_x * 16;
1631 dest_cb = s->current_picture[1] + (mb_y * 8 * (s->uvlinesize)) + mb_x * 8;
1632 dest_cr = s->current_picture[2] + (mb_y * 8 * (s->uvlinesize)) + mb_x * 8;
1633 wrap_y = s->linesize;
1634 wrap_c = wrap_y>>1;
1635 ptr_y = s->new_picture[0] + (mb_y * 16 * wrap_y) + mb_x * 16;
1636 ptr_cb = s->new_picture[1] + (mb_y * 8 * wrap_c) + mb_x * 8;
1637 ptr_cr = s->new_picture[2] + (mb_y * 8 * wrap_c) + mb_x * 8;
1638
1639 if ((!s->no_rounding) || s->pict_type==B_TYPE){
1640 op_pix = put_pixels_tab;
1641 op_qpix= qpel_mc_rnd_tab;
1642 }else{
1643 op_pix = put_no_rnd_pixels_tab;
1644 op_qpix= qpel_mc_no_rnd_tab;
1645 }
1646
1647 if (s->mv_dir & MV_DIR_FORWARD) {
1648 MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture, op_pix, op_qpix);
1649 if ((!s->no_rounding) || s->pict_type==B_TYPE)
1650 op_pix = avg_pixels_tab;
1651 else
1652 op_pix = avg_no_rnd_pixels_tab;
1653 }
1654 if (s->mv_dir & MV_DIR_BACKWARD) {
1655 MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture, op_pix, op_qpix);
1656 }
1657
1658 diff_pixels(s->block[0], ptr_y , dest_y , wrap_y);
1659 diff_pixels(s->block[1], ptr_y + 8, dest_y + 8, wrap_y);
1660 diff_pixels(s->block[2], ptr_y + 8 * wrap_y , dest_y + 8 * wrap_y , wrap_y);
1661 diff_pixels(s->block[3], ptr_y + 8 * wrap_y + 8, dest_y + 8 * wrap_y + 8, wrap_y);
1662
1663 if(s->flags&CODEC_FLAG_GRAY){
1664 skip_dct[4]= 1;
1665 skip_dct[5]= 1;
1666 }else{
1667 diff_pixels(s->block[4], ptr_cb, dest_cb, wrap_c);
1668 diff_pixels(s->block[5], ptr_cr, dest_cr, wrap_c);
1669 }
1670
1671 /* pre quantization */
1672 if(s->mc_mb_var[s->mb_width*mb_y+ mb_x]<2*s->qscale*s->qscale){
1673 if(pix_abs8x8(ptr_y , dest_y , wrap_y) < 20*s->qscale) skip_dct[0]= 1;
1674 if(pix_abs8x8(ptr_y + 8, dest_y + 8, wrap_y) < 20*s->qscale) skip_dct[1]= 1;
1675 if(pix_abs8x8(ptr_y + 8*wrap_y , dest_y + 8*wrap_y , wrap_y) < 20*s->qscale) skip_dct[2]= 1;
1676 if(pix_abs8x8(ptr_y + 8*wrap_y + 8, dest_y + 8*wrap_y + 8, wrap_y) < 20*s->qscale) skip_dct[3]= 1;
1677 if(pix_abs8x8(ptr_cb , dest_cb , wrap_y) < 20*s->qscale) skip_dct[4]= 1;
1678 if(pix_abs8x8(ptr_cr , dest_cr , wrap_y) < 20*s->qscale) skip_dct[5]= 1;
1679 #if 0
1680 {
1681 static int stat[7];
1682 int num=0;
1683 for(i=0; i<6; i++)
1684 if(skip_dct[i]) num++;
1685 stat[num]++;
1686
1687 if(s->mb_x==0 && s->mb_y==0){
1688 for(i=0; i<7; i++){
1689 printf("%6d %1d\n", stat[i], i);
1690 }
1691 }
1692 }
1693 #endif
1694 }
1695
1696 }
1697
1698 #if 0
1699 {
1700 float adap_parm;
1701
1702 adap_parm = ((s->avg_mb_var << 1) + s->mb_var[s->mb_width*mb_y+mb_x] + 1.0) /
1703 ((s->mb_var[s->mb_width*mb_y+mb_x] << 1) + s->avg_mb_var + 1.0);
1704
1705 printf("\ntype=%c qscale=%2d adap=%0.2f dquant=%4.2f var=%4d avgvar=%4d",
1706 (s->mb_type[s->mb_width*mb_y+mb_x] > 0) ? 'I' : 'P',
1707 s->qscale, adap_parm, s->qscale*adap_parm,
1708 s->mb_var[s->mb_width*mb_y+mb_x], s->avg_mb_var);
1709 }
1710 #endif
1711 /* DCT & quantize */
1712 if(s->out_format==FMT_MJPEG){
1713 for(i=0;i<6;i++) {
1714 int overflow;
1715 s->block_last_index[i] = dct_quantize(s, s->block[i], i, 8, &overflow);
1716 if (overflow) clip_coeffs(s, s->block[i], s->block_last_index[i]);
1717 }
1718 }else{
1719 for(i=0;i<6;i++) {
1720 if(!skip_dct[i]){
1721 int overflow;
1722 s->block_last_index[i] = dct_quantize(s, s->block[i], i, s->qscale, &overflow);
1723 // FIXME we could decide to change to quantizer instead of clipping
1724 // JS: I don't think that would be a good idea it could lower quality instead
1725 // of improve it. Just INTRADC clipping deserves changes in quantizer
1726 if (overflow) clip_coeffs(s, s->block[i], s->block_last_index[i]);
1727 }else
1728 s->block_last_index[i]= -1;
1729 }
1730 if(s->luma_elim_threshold && !s->mb_intra)
1731 for(i=0; i<4; i++)
1732 dct_single_coeff_elimination(s, i, s->luma_elim_threshold, 0);
1733 if(s->chroma_elim_threshold && !s->mb_intra)
1734 for(i=4; i<6; i++)
1735 dct_single_coeff_elimination(s, i, s->chroma_elim_threshold, 1);
1736 }
1737
1738 if((s->flags&CODEC_FLAG_GRAY) && s->mb_intra){
1739 s->block_last_index[4]=
1740 s->block_last_index[5]= 0;
1741 s->block[4][0]=
1742 s->block[5][0]= 128;
1743 }
1744
1745 /* huffman encode */
1746 switch(s->out_format) {
1747 case FMT_MPEG1:
1748 mpeg1_encode_mb(s, s->block, motion_x, motion_y);
1749 break;
1750 case FMT_H263:
1751 if (s->h263_msmpeg4)
1752 msmpeg4_encode_mb(s, s->block, motion_x, motion_y);
1753 else if(s->h263_pred)
1754 mpeg4_encode_mb(s, s->block, motion_x, motion_y);
1755 else
1756 h263_encode_mb(s, s->block, motion_x, motion_y);
1757 break;
1758 case FMT_MJPEG:
1759 mjpeg_encode_mb(s, s->block);
1760 break;
1761 }
1762 }
1763
1764 void ff_copy_bits(PutBitContext *pb, UINT8 *src, int length)
1765 {
1766 int bytes= length>>4;
1767 int bits= length&15;
1768 int i;
1769
1770 if(length==0) return;
1771
1772 for(i=0; i<bytes; i++) put_bits(pb, 16, be2me_16(((uint16_t*)src)[i]));
1773 put_bits(pb, bits, be2me_16(((uint16_t*)src)[i])>>(16-bits));
1774 }
1775
1776 static inline void copy_context_before_encode(MpegEncContext *d, MpegEncContext *s, int type){
1777 int i;
1778
1779 memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
1780
1781 /* mpeg1 */
1782 d->mb_incr= s->mb_incr;
1783 for(i=0; i<3; i++)
1784 d->last_dc[i]= s->last_dc[i];
1785
1786 /* statistics */
1787 d->mv_bits= s->mv_bits;
1788 d->i_tex_bits= s->i_tex_bits;
1789 d->p_tex_bits= s->p_tex_bits;
1790 d->i_count= s->i_count;
1791 d->p_count= s->p_count;
1792 d->skip_count= s->skip_count;
1793 d->misc_bits= s->misc_bits;
1794 d->last_bits= 0;
1795
1796 d->mb_skiped= s->mb_skiped;
1797 }
1798
1799 static inline void copy_context_after_encode(MpegEncContext *d, MpegEncContext *s, int type){
1800 int i;
1801
1802 memcpy(d->mv, s->mv, 2*4*2*sizeof(int));
1803 memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
1804
1805 /* mpeg1 */
1806 d->mb_incr= s->mb_incr;
1807 for(i=0; i<3; i++)
1808 d->last_dc[i]= s->last_dc[i];
1809
1810 /* statistics */
1811 d->mv_bits= s->mv_bits;
1812 d->i_tex_bits= s->i_tex_bits;
1813 d->p_tex_bits= s->p_tex_bits;
1814 d->i_count= s->i_count;
1815 d->p_count= s->p_count;
1816 d->skip_count= s->skip_count;
1817 d->misc_bits= s->misc_bits;
1818
1819 d->mb_intra= s->mb_intra;
1820 d->mb_skiped= s->mb_skiped;
1821 d->mv_type= s->mv_type;
1822 d->mv_dir= s->mv_dir;
1823 d->pb= s->pb;
1824 if(s->data_partitioning){
1825 d->pb2= s->pb2;
1826 d->tex_pb= s->tex_pb;
1827 }
1828 d->block= s->block;
1829 for(i=0; i<6; i++)
1830 d->block_last_index[i]= s->block_last_index[i];
1831 }
1832
1833 static inline void encode_mb_hq(MpegEncContext *s, MpegEncContext *backup, MpegEncContext *best, int type,
1834 PutBitContext pb[2], PutBitContext pb2[2], PutBitContext tex_pb[2],
1835 int *dmin, int *next_block, int motion_x, int motion_y)
1836 {
1837 int bits_count;
1838
1839 copy_context_before_encode(s, backup, type);
1840
1841 s->block= s->blocks[*next_block];
1842 s->pb= pb[*next_block];
1843 if(s->data_partitioning){
1844 s->pb2 = pb2 [*next_block];
1845 s->tex_pb= tex_pb[*next_block];
1846 }
1847
1848 encode_mb(s, motion_x, motion_y);
1849
1850 bits_count= get_bit_count(&s->pb);
1851 if(s->data_partitioning){
1852 bits_count+= get_bit_count(&s->pb2);
1853 bits_count+= get_bit_count(&s->tex_pb);
1854 }
1855
1856 if(bits_count<*dmin){
1857 *dmin= bits_count;
1858 *next_block^=1;
1859
1860 copy_context_after_encode(best, s, type);
1861 }
1862 }
1863
1864 static void encode_picture(MpegEncContext *s, int picture_number)
1865 {
1866 int mb_x, mb_y, last_gob, pdif = 0;
1867 int i;
1868 int bits;
1869 MpegEncContext best_s, backup_s;
1870 UINT8 bit_buf[2][3000];
1871 UINT8 bit_buf2[2][3000];
1872 UINT8 bit_buf_tex[2][3000];
1873 PutBitContext pb[2], pb2[2], tex_pb[2];
1874
1875 for(i=0; i<2; i++){
1876 init_put_bits(&pb [i], bit_buf [i], 3000, NULL, NULL);
1877 init_put_bits(&pb2 [i], bit_buf2 [i], 3000, NULL, NULL);
1878 init_put_bits(&tex_pb[i], bit_buf_tex[i], 3000, NULL, NULL);
1879 }
1880
1881 s->picture_number = picture_number;
1882
1883 s->block_wrap[0]=
1884 s->block_wrap[1]=
1885 s->block_wrap[2]=
1886 s->block_wrap[3]= s->mb_width*2 + 2;
1887 s->block_wrap[4]=
1888 s->block_wrap[5]= s->mb_width + 2;
1889
1890 /* Reset the average MB variance */
1891 s->mb_var_sum = 0;
1892 s->mc_mb_var_sum = 0;
1893
1894 /* we need to initialize some time vars before we can encode b-frames */
1895 if (s->h263_pred && !s->h263_msmpeg4)
1896 ff_set_mpeg4_time(s, s->picture_number);
1897
1898 /* Estimate motion for every MB */
1899 if(s->pict_type != I_TYPE){
1900 for(mb_y=0; mb_y < s->mb_height; mb_y++) {
1901 s->block_index[0]= s->block_wrap[0]*(mb_y*2 + 1) - 1;
1902 s->block_index[1]= s->block_wrap[0]*(mb_y*2 + 1);
1903 s->block_index[2]= s->block_wrap[0]*(mb_y*2 + 2) - 1;
1904 s->block_index[3]= s->block_wrap[0]*(mb_y*2 + 2);
1905 for(mb_x=0; mb_x < s->mb_width; mb_x++) {
1906 s->mb_x = mb_x;
1907 s->mb_y = mb_y;
1908 s->block_index[0]+=2;
1909 s->block_index[1]+=2;
1910 s->block_index[2]+=2;
1911 s->block_index[3]+=2;
1912
1913 /* compute motion vector & mb_type and store in context */
1914 if(s->pict_type==B_TYPE)
1915 ff_estimate_b_frame_motion(s, mb_x, mb_y);
1916 else
1917 ff_estimate_p_frame_motion(s, mb_x, mb_y);
1918 // s->mb_type[mb_y*s->mb_width + mb_x]=MB_TYPE_INTER;
1919 }
1920 }
1921 emms_c();
1922 }else /* if(s->pict_type == I_TYPE) */{
1923 /* I-Frame */
1924 //FIXME do we need to zero them?
1925 memset(s->motion_val[0], 0, sizeof(INT16)*(s->mb_width*2 + 2)*(s->mb_height*2 + 2)*2);
1926 memset(s->p_mv_table , 0, sizeof(INT16)*(s->mb_width+2)*(s->mb_height+2)*2);
1927 memset(s->mb_type , MB_TYPE_INTRA, sizeof(UINT8)*s->mb_width*s->mb_height);
1928 }
1929
1930 if(s->mb_var_sum < s->mc_mb_var_sum && s->pict_type == P_TYPE){ //FIXME subtract MV bits
1931 s->pict_type= I_TYPE;
1932 memset(s->mb_type , MB_TYPE_INTRA, sizeof(UINT8)*s->mb_width*s->mb_height);
1933 if(s->max_b_frames==0){
1934 s->input_pict_type= I_TYPE;
1935 s->input_picture_in_gop_number=0;
1936 }
1937 //printf("Scene change detected, encoding as I Frame\n");
1938 }
1939
1940 if(s->pict_type==P_TYPE || s->pict_type==S_TYPE)
1941 s->f_code= ff_get_best_fcode(s, s->p_mv_table, MB_TYPE_INTER);
1942 ff_fix_long_p_mvs(s);
1943 if(s->pict_type==B_TYPE){
1944 s->f_code= ff_get_best_fcode(s, s->b_forw_mv_table, MB_TYPE_FORWARD);
1945 s->b_code= ff_get_best_fcode(s, s->b_back_mv_table, MB_TYPE_BACKWARD);
1946
1947 ff_fix_long_b_mvs(s, s->b_forw_mv_table, s->f_code, MB_TYPE_FORWARD);
1948 ff_fix_long_b_mvs(s, s->b_back_mv_table, s->b_code, MB_TYPE_BACKWARD);
1949 ff_fix_long_b_mvs(s, s->b_bidir_forw_mv_table, s->f_code, MB_TYPE_BIDIR);
1950 ff_fix_long_b_mvs(s, s->b_bidir_back_mv_table, s->b_code, MB_TYPE_BIDIR);
1951 }
1952
1953 //printf("f_code %d ///\n", s->f_code);
1954
1955 // printf("%d %d\n", s->avg_mb_var, s->mc_mb_var);
1956
1957 if(s->flags&CODEC_FLAG_PASS2)
1958 s->qscale = ff_rate_estimate_qscale_pass2(s);
1959 else if (!s->fixed_qscale)
1960 s->qscale = ff_rate_estimate_qscale(s);
1961
1962 if (s->out_format == FMT_MJPEG) {
1963 /* for mjpeg, we do include qscale in the matrix */
1964 s->intra_matrix[0] = ff_mpeg1_default_intra_matrix[0];
1965 for(i=1;i<64;i++)
1966 s->intra_matrix[i] = CLAMP_TO_8BIT((ff_mpeg1_default_intra_matrix[i] * s->qscale) >> 3);
1967 convert_matrix(s->q_intra_matrix, s->q_intra_matrix16,
1968 s->q_intra_matrix16_bias, s->intra_matrix, s->intra_quant_bias);
1969 }
1970
1971 s->last_bits= get_bit_count(&s->pb);
1972 switch(s->out_format) {
1973 case FMT_MJPEG:
1974 mjpeg_picture_header(s);
1975 break;
1976 case FMT_H263:
1977 if (s->h263_msmpeg4)
1978 msmpeg4_encode_picture_header(s, picture_number);
1979 else if (s->h263_pred)
1980 mpeg4_encode_picture_header(s, picture_number);
1981 else if (s->h263_rv10)
1982 rv10_encode_picture_header(s, picture_number);
1983 else
1984 h263_encode_picture_header(s, picture_number);
1985 break;
1986 case FMT_MPEG1:
1987 mpeg1_encode_picture_header(s, picture_number);
1988 break;
1989 }
1990 bits= get_bit_count(&s->pb);
1991 s->header_bits= bits - s->last_bits;
1992 s->last_bits= bits;
1993 s->mv_bits=0;
1994 s->misc_bits=0;
1995 s->i_tex_bits=0;
1996 s->p_tex_bits=0;
1997 s->i_count=0;
1998 s->p_count=0;
1999 s->skip_count=0;
2000
2001 /* init last dc values */
2002 /* note: quant matrix value (8) is implied here */
2003 s->last_dc[0] = 128;
2004 s->last_dc[1] = 128;
2005 s->last_dc[2] = 128;
2006 s->mb_incr = 1;
2007 s->last_mv[0][0][0] = 0;
2008 s->last_mv[0][0][1] = 0;
2009
2010 /* Get the GOB height based on picture height */
2011 if (s->out_format == FMT_H263 && !s->h263_pred && !s->h263_msmpeg4) {
2012 if (s->height <= 400)
2013 s->gob_index = 1;
2014 else if (s->height <= 800)
2015 s->gob_index = 2;
2016 else
2017 s->gob_index = 4;
2018 }else if(s->codec_id==CODEC_ID_MPEG4){
2019 s->gob_index = 1;
2020 }
2021
2022 if(s->codec_id==CODEC_ID_MPEG4 && s->data_partitioning && s->pict_type!=B_TYPE)
2023 ff_mpeg4_init_partitions(s);
2024
2025 s->resync_mb_x=0;
2026 s->resync_mb_y=0;
2027 for(mb_y=0; mb_y < s->mb_height; mb_y++) {
2028 /* Put GOB header based on RTP MTU for formats which support it per line (H263*)*/
2029 /* TODO: Put all this stuff in a separate generic function */
2030 if (s->rtp_mode) {
2031 if (!mb_y) {
2032 s->ptr_lastgob = s->pb.buf;
2033 s->ptr_last_mb_line = s->pb.buf;
2034 } else if (s->out_format == FMT_H263 && !s->h263_pred && !s->h263_msmpeg4 && !(mb_y % s->gob_index)) {
2035 // MN: we could move the space check from h263 -> here, as its not h263 specific
2036 last_gob = h263_encode_gob_header(s, mb_y);
2037 if (last_gob) {
2038 s->first_slice_line = 1;
2039 }else{
2040 /*MN: we reset it here instead at the end of each line cuz mpeg4 can have
2041 slice lines starting & ending in the middle*/
2042 s->first_slice_line = 0;
2043 }
2044 }
2045 }
2046
2047 s->y_dc_scale= s->y_dc_scale_table[ s->qscale ];
2048 s->c_dc_scale= s->c_dc_scale_table[ s->qscale ];
2049
2050 s->block_index[0]= s->block_wrap[0]*(mb_y*2 + 1) - 1;
2051 s->block_index[1]= s->block_wrap[0]*(mb_y*2 + 1);
2052 s->block_index[2]= s->block_wrap[0]*(mb_y*2 + 2) - 1;
2053 s->block_index[3]= s->block_wrap[0]*(mb_y*2 + 2);
2054 s->block_index[4]= s->block_wrap[4]*(mb_y + 1) + s->block_wrap[0]*(s->mb_height*2 + 2);
2055 s->block_index[5]= s->block_wrap[4]*(mb_y + 1 + s->mb_height + 2) + s->block_wrap[0]*(s->mb_height*2 + 2);
2056 for(mb_x=0; mb_x < s->mb_width; mb_x++) {
2057 const int mb_type= s->mb_type[mb_y * s->mb_width + mb_x];
2058 const int xy= (mb_y+1) * (s->mb_width+2) + mb_x + 1;
2059 // int d;
2060 int dmin=10000000;
2061
2062 s->mb_x = mb_x;
2063 s->mb_y = mb_y;
2064 s->block_index[0]+=2;
2065 s->block_index[1]+=2;
2066 s->block_index[2]+=2;
2067 s->block_index[3]+=2;
2068 s->block_index[4]++;
2069 s->block_index[5]++;
2070
2071 /* write gob / video packet header for formats which support it at any MB (MPEG4) */
2072 if(s->rtp_mode && s->mb_y>0 && s->codec_id==CODEC_ID_MPEG4){
2073 int pdif= pbBufPtr(&s->pb) - s->ptr_lastgob;
2074
2075 //the *2 is there so we stay below the requested size
2076 if(pdif + s->mb_line_avgsize/s->mb_width >= s->rtp_payload_size){
2077 if(s->codec_id==CODEC_ID_MPEG4){
2078 if(s->data_partitioning && s->pict_type!=B_TYPE){
2079 ff_mpeg4_merge_partitions(s);
2080 ff_mpeg4_init_partitions(s);
2081 }
2082 ff_mpeg4_encode_video_packet_header(s);
2083
2084 if(s->flags&CODEC_FLAG_PASS1){
2085 int bits= get_bit_count(&s->pb);
2086 s->misc_bits+= bits - s->last_bits;
2087 s->last_bits= bits;
2088 }
2089 ff_mpeg4_clean_buffers(s);
2090 }
2091 s->ptr_lastgob = pbBufPtr(&s->pb);
2092 s->first_slice_line=1;
2093 s->resync_mb_x=mb_x;
2094 s->resync_mb_y=mb_y;
2095 }
2096
2097 if( (s->resync_mb_x == s->mb_x)
2098 && s->resync_mb_y+1 == s->mb_y){
2099 s->first_slice_line=0;
2100 }
2101 }
2102
2103 if(mb_type & (mb_type-1)){ // more than 1 MB type possible
2104 int next_block=0;
2105 int pb_bits_count, pb2_bits_count, tex_pb_bits_count;
2106
2107 copy_context_before_encode(&backup_s, s, -1);
2108 backup_s.pb= s->pb;
2109 best_s.data_partitioning= s->data_partitioning;
2110 if(s->data_partitioning){
2111 backup_s.pb2= s->pb2;
2112 backup_s.tex_pb= s->tex_pb;
2113 }
2114
2115 if(mb_type&MB_TYPE_INTER){
2116 s->mv_dir = MV_DIR_FORWARD;
2117 s->mv_type = MV_TYPE_16X16;
2118 s->mb_intra= 0;
2119 s->mv[0][0][0] = s->p_mv_table[xy][0];
2120 s->mv[0][0][1] = s->p_mv_table[xy][1];
2121 encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_INTER, pb, pb2, tex_pb,
2122 &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2123 }
2124 if(mb_type&MB_TYPE_INTER4V){
2125 s->mv_dir = MV_DIR_FORWARD;
2126 s->mv_type = MV_TYPE_8X8;
2127 s->mb_intra= 0;
2128 for(i=0; i<4; i++){
2129 s->mv[0][i][0] = s->motion_val[s->block_index[i]][0];
2130 s->mv[0][i][1] = s->motion_val[s->block_index[i]][1];
2131 }
2132 encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_INTER4V, pb, pb2, tex_pb,
2133 &dmin, &next_block, 0, 0);
2134 }
2135 if(mb_type&MB_TYPE_FORWARD){
2136 s->mv_dir = MV_DIR_FORWARD;
2137 s->mv_type = MV_TYPE_16X16;
2138 s->mb_intra= 0;
2139 s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
2140 s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2141 encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_FORWARD, pb, pb2, tex_pb,
2142 &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2143 }
2144 if(mb_type&MB_TYPE_BACKWARD){
2145 s->mv_dir = MV_DIR_BACKWARD;
2146 s->mv_type = MV_TYPE_16X16;
2147 s->mb_intra= 0;
2148 s->mv[1][0][0] = s->b_back_mv_table[xy][0];
2149 s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2150 encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_BACKWARD, pb, pb2, tex_pb,
2151 &dmin, &next_block, s->mv[1][0][0], s->mv[1][0][1]);
2152 }
2153 if(mb_type&MB_TYPE_BIDIR){
2154 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2155 s->mv_type = MV_TYPE_16X16;
2156 s->mb_intra= 0;
2157 s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
2158 s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
2159 s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
2160 s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2161 encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_BIDIR, pb, pb2, tex_pb,
2162 &dmin, &next_block, 0, 0);
2163 }
2164 if(mb_type&MB_TYPE_DIRECT){
2165 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2166 s->mv_type = MV_TYPE_16X16; //FIXME
2167 s->mb_intra= 0;
2168 s->mv[0][0][0] = s->b_direct_forw_mv_table[xy][0];
2169 s->mv[0][0][1] = s->b_direct_forw_mv_table[xy][1];
2170 s->mv[1][0][0] = s->b_direct_back_mv_table[xy][0];
2171 s->mv[1][0][1] = s->b_direct_back_mv_table[xy][1];
2172 encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_DIRECT, pb, pb2, tex_pb,
2173 &dmin, &next_block, s->b_direct_mv_table[xy][0], s->b_direct_mv_table[xy][1]);
2174 }
2175 if(mb_type&MB_TYPE_INTRA){
2176 s->mv_dir = MV_DIR_FORWARD;
2177 s->mv_type = MV_TYPE_16X16;
2178 s->mb_intra= 1;
2179 s->mv[0][0][0] = 0;
2180 s->mv[0][0][1] = 0;
2181 encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_INTRA, pb, pb2, tex_pb,
2182 &dmin, &next_block, 0, 0);
2183 /* force cleaning of ac/dc pred stuff if needed ... */
2184 if(s->h263_pred || s->h263_aic)
2185 s->mbintra_table[mb_x + mb_y*s->mb_width]=1;
2186 }
2187 copy_context_after_encode(s, &best_s, -1);
2188
2189 pb_bits_count= get_bit_count(&s->pb);
2190 flush_put_bits(&s->pb);
2191 ff_copy_bits(&backup_s.pb, bit_buf[next_block^1], pb_bits_count);
2192 s->pb= backup_s.pb;
2193
2194 if(s->data_partitioning){
2195 pb2_bits_count= get_bit_count(&s->pb2);
2196 flush_put_bits(&s->pb2);
2197 ff_copy_bits(&backup_s.pb2, bit_buf2[next_block^1], pb2_bits_count);
2198 s->pb2= backup_s.pb2;
2199
2200 tex_pb_bits_count= get_bit_count(&s->tex_pb);
2201 flush_put_bits(&s->tex_pb);
2202 ff_copy_bits(&backup_s.tex_pb, bit_buf_tex[next_block^1], tex_pb_bits_count);
2203 s->tex_pb= backup_s.tex_pb;
2204 }
2205 s->last_bits= get_bit_count(&s->pb);
2206 } else {
2207 int motion_x, motion_y;
2208 s->mv_type=MV_TYPE_16X16;
2209 // only one MB-Type possible
2210 switch(mb_type){
2211 case MB_TYPE_INTRA:
2212 s->mv_dir = MV_DIR_FORWARD;
2213 s->mb_intra= 1;
2214 motion_x= s->mv[0][0][0] = 0;
2215 motion_y= s->mv[0][0][1] = 0;
2216 break;
2217 case MB_TYPE_INTER:
2218 s->mv_dir = MV_DIR_FORWARD;
2219 s->mb_intra= 0;
2220 motion_x= s->mv[0][0][0] = s->p_mv_table[xy][0];
2221 motion_y= s->mv[0][0][1] = s->p_mv_table[xy][1];
2222 break;
2223 case MB_TYPE_INTER4V:
2224 s->mv_dir = MV_DIR_FORWARD;
2225 s->mv_type = MV_TYPE_8X8;
2226 s->mb_intra= 0;
2227 for(i=0; i<4; i++){
2228 s->mv[0][i][0] = s->motion_val[s->block_index[i]][0];
2229 s->mv[0][i][1] = s->motion_val[s->block_index[i]][1];
2230 }
2231 motion_x= motion_y= 0;
2232 break;
2233 case MB_TYPE_DIRECT:
2234 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2235 s->mb_intra= 0;
2236 motion_x=s->b_direct_mv_table[xy][0];
2237 motion_y=s->b_direct_mv_table[xy][1];
2238 s->mv[0][0][0] = s->b_direct_forw_mv_table[xy][0];
2239 s->mv[0][0][1] = s->b_direct_forw_mv_table[xy][1];
2240 s->mv[1][0][0] = s->b_direct_back_mv_table[xy][0];
2241 s->mv[1][0][1] = s->b_direct_back_mv_table[xy][1];
2242 break;
2243 case MB_TYPE_BIDIR:
2244 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2245 s->mb_intra= 0;
2246 motion_x=0;
2247 motion_y=0;
2248 s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
2249 s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
2250 s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
2251 s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2252 break;
2253 case MB_TYPE_BACKWARD:
2254 s->mv_dir = MV_DIR_BACKWARD;
2255 s->mb_intra= 0;
2256 motion_x= s->mv[1][0][0] = s->b_back_mv_table[xy][0];
2257 motion_y= s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2258 break;
2259 case MB_TYPE_FORWARD:
2260 s->mv_dir = MV_DIR_FORWARD;
2261 s->mb_intra= 0;
2262 motion_x= s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
2263 motion_y= s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2264 // printf(" %d %d ", motion_x, motion_y);
2265 break;
2266 default:
2267 motion_x=motion_y=0; //gcc warning fix
2268 printf("illegal MB type\n");
2269 }
2270 encode_mb(s, motion_x, motion_y);
2271 }
2272 /* clean the MV table in IPS frames for direct mode in B frames */
2273 if(s->mb_intra /* && I,P,S_TYPE */){
2274 s->p_mv_table[xy][0]=0;
2275 s->p_mv_table[xy][1]=0;
2276 }
2277
2278 MPV_decode_mb(s, s->block);
2279 //printf("MB %d %d bits\n", s->mb_x+s->mb_y*s->mb_width, get_bit_count(&s->pb));
2280 }
2281
2282
2283 /* Obtain average GOB size for RTP */
2284 if (s->rtp_mode) {
2285 if (!mb_y)
2286 s->mb_line_avgsize = pbBufPtr(&s->pb) - s->ptr_last_mb_line;
2287 else if (!(mb_y % s->gob_index)) {
2288 s->mb_line_avgsize = (s->mb_line_avgsize + pbBufPtr(&s->pb) - s->ptr_last_mb_line) >> 1;
2289 s->ptr_last_mb_line = pbBufPtr(&s->pb);
2290 }
2291 //fprintf(stderr, "\nMB line: %d\tSize: %u\tAvg. Size: %u", s->mb_y,
2292 // (s->pb.buf_ptr - s->ptr_last_mb_line), s->mb_line_avgsize);
2293 if(s->codec_id!=CODEC_ID_MPEG4) s->first_slice_line = 0; //FIXME clean
2294 }
2295 }
2296 emms_c();
2297
2298 if(s->codec_id==CODEC_ID_MPEG4 && s->data_partitioning && s->pict_type!=B_TYPE)
2299 ff_mpeg4_merge_partitions(s);
2300
2301 if (s->msmpeg4_version && s->msmpeg4_version<4 && s->pict_type == I_TYPE)
2302 msmpeg4_encode_ext_header(s);
2303
2304 if(s->codec_id==CODEC_ID_MPEG4)
2305 ff_mpeg4_stuffing(&s->pb);
2306
2307 //if (s->gob_number)
2308 // fprintf(stderr,"\nNumber of GOB: %d", s->gob_number);
2309
2310 /* Send the last GOB if RTP */
2311 if (s->rtp_mode) {
2312 flush_put_bits(&s->pb);
2313 pdif = pbBufPtr(&s->pb) - s->ptr_lastgob;
2314 /* Call the RTP callback to send the last GOB */
2315 if (s->rtp_callback)
2316 s->rtp_callback(s->ptr_lastgob, pdif, s->gob_number);
2317 s->ptr_lastgob = pbBufPtr(&s->pb);
2318 //fprintf(stderr,"\nGOB: %2d size: %d (last)", s->gob_number, pdif);
2319 }
2320 }
2321
2322 static int dct_quantize_c(MpegEncContext *s,
2323 DCTELEM *block, int n,
2324 int qscale, int *overflow)
2325 {
2326 int i, j, level, last_non_zero, q;
2327 const int *qmat;
2328 int bias;
2329 int max=0;
2330 unsigned int threshold1, threshold2;
2331
2332 av_fdct (block);
2333
2334 /* we need this permutation so that we correct the IDCT
2335 permutation. will be moved into DCT code */
2336 block_permute(block);
2337
2338 if (s->mb_intra) {
2339 if (!s->h263_aic) {
2340 if (n < 4)
2341 q = s->y_dc_scale;
2342 else
2343 q = s->c_dc_scale;
2344 q = q << 3;
2345 } else
2346 /* For AIC we skip quant/dequant of INTRADC */
2347 q = 1 << 3;
2348
2349 /* note: block[0] is assumed to be positive */
2350 block[0] = (block[0] + (q >> 1)) / q;
2351 i = 1;
2352 last_non_zero = 0;
2353 qmat = s->q_intra_matrix[qscale];
2354 bias= s->intra_quant_bias<<(QMAT_SHIFT - 3 - QUANT_BIAS_SHIFT);
2355 } else {
2356 i = 0;
2357 last_non_zero = -1;
2358 qmat = s->q_inter_matrix[qscale];
2359 bias= s->inter_quant_bias<<(QMAT_SHIFT - 3 - QUANT_BIAS_SHIFT);
2360 }
2361 threshold1= (1<<(QMAT_SHIFT - 3)) - bias - 1;
2362 threshold2= threshold1<<1;
2363
2364 for(;i<64;i++) {
2365 j = zigzag_direct[i];
2366 level = block[j];
2367 level = level * qmat[j];
2368
2369 // if( bias+level >= (1<<(QMAT_SHIFT - 3))
2370 // || bias-level >= (1<<(QMAT_SHIFT - 3))){
2371 if(((unsigned)(level+threshold1))>threshold2){
2372 if(level>0){
2373 level= (bias + level)>>(QMAT_SHIFT - 3);
2374 block[j]= level;
2375 }else{
2376 level= (bias - level)>>(QMAT_SHIFT - 3);
2377 block[j]= -level;
2378 }
2379 max |=level;
2380 last_non_zero = i;
2381 }else{
2382 block[j]=0;
2383 }
2384 }
2385 *overflow= s->max_qcoeff < max; //overflow might have happend
2386
2387 return last_non_zero;
2388 }
2389
2390 static void dct_unquantize_mpeg1_c(MpegEncContext *s,
2391 DCTELEM *block, int n, int qscale)
2392 {
2393 int i, level, nCoeffs;
2394 const UINT16 *quant_matrix;
2395
2396 if(s->alternate_scan) nCoeffs= 64;
2397 else nCoeffs= s->block_last_index[n]+1;
2398
2399 if (s->mb_intra) {
2400 if (n < 4)
2401 block[0] = block[0] * s->y_dc_scale;
2402 else
2403 block[0] = block[0] * s->c_dc_scale;
2404 /* XXX: only mpeg1 */
2405 quant_matrix = s->intra_matrix;
2406 for(i=1;i<nCoeffs;i++) {
2407 int j= zigzag_direct[i];
2408 level = block[j];
2409 if (level) {
2410 if (level < 0) {
2411 level = -level;
2412 level = (int)(level * qscale * quant_matrix[j]) >> 3;
2413 level = (level - 1) | 1;
2414 level = -level;
2415 } else {
2416 level = (int)(level * qscale * quant_matrix[j]) >> 3;
2417 level = (level - 1) | 1;
2418 }
2419 #ifdef PARANOID
2420 if (level < -2048 || level > 2047)
2421 fprintf(stderr, "unquant error %d %d\n", i, level);
2422 #endif
2423 block[j] = level;
2424 }
2425 }
2426 } else {
2427 i = 0;
2428 quant_matrix = s->inter_matrix;
2429 for(;i<nCoeffs;i++) {
2430 int j= zigzag_direct[i];
2431 level = block[j];
2432 if (level) {
2433 if (level < 0) {
2434 level = -level;
2435 level = (((level << 1) + 1) * qscale *
2436 ((int) (quant_matrix[j]))) >> 4;
2437 level = (level - 1) | 1;
2438 level = -level;
2439 } else {
2440 level = (((level << 1) + 1) * qscale *
2441 ((int) (quant_matrix[j]))) >> 4;
2442 level = (level - 1) | 1;
2443 }
2444 #ifdef PARANOID
2445 if (level < -2048 || level > 2047)
2446 fprintf(stderr, "unquant error %d %d\n", i, level);
2447 #endif
2448 block[j] = level;
2449 }
2450 }
2451 }
2452 }
2453
2454 static void dct_unquantize_mpeg2_c(MpegEncContext *s,
2455 DCTELEM *block, int n, int qscale)
2456 {
2457 int i, level, nCoeffs;
2458 const UINT16 *quant_matrix;
2459
2460 if(s->alternate_scan) nCoeffs= 64;
2461 else nCoeffs= s->block_last_index[n]+1;
2462
2463 if (s->mb_intra) {
2464 if (n < 4)
2465 block[0] = block[0] * s->y_dc_scale;
2466 else
2467 block[0] = block[0] * s->c_dc_scale;
2468 quant_matrix = s->intra_matrix;
2469 for(i=1;i<nCoeffs;i++) {
2470 int j= zigzag_direct[i];
2471 level = block[j];
2472 if (level) {
2473 if (level < 0) {
2474 level = -level;
2475 level = (int)(level * qscale * quant_matrix[j]) >> 3;
2476 level = -level;
2477 } else {
2478 level = (int)(level * qscale * quant_matrix[j]) >> 3;
2479 }
2480 #ifdef PARANOID
2481 if (level < -2048 || level > 2047)
2482 fprintf(stderr, "unquant error %d %d\n", i, level);
2483 #endif
2484 block[j] = level;
2485 }
2486 }
2487 } else {
2488 int sum=-1;
2489 i = 0;
2490 quant_matrix = s->inter_matrix;
2491 for(;i<nCoeffs;i++) {
2492 int j= zigzag_direct[i];
2493 level = block[j];
2494 if (level) {
2495 if (level < 0) {
2496 level = -level;
2497 level = (((level << 1) + 1) * qscale *
2498 ((int) (quant_matrix[j]))) >> 4;
2499 level = -level;
2500 } else {
2501 level = (((level << 1) + 1) * qscale *
2502 ((int) (quant_matrix[j]))) >> 4;
2503 }
2504 #ifdef PARANOID
2505 if (level < -2048 || level > 2047)
2506 fprintf(stderr, "unquant error %d %d\n", i, level);
2507 #endif
2508 block[j] = level;
2509 sum+=level;
2510 }
2511 }
2512 block[63]^=sum&1;
2513 }
2514 }
2515
2516
2517 static void dct_unquantize_h263_c(MpegEncContext *s,
2518 DCTELEM *block, int n, int qscale)
2519 {
2520 int i, level, qmul, qadd;
2521 int nCoeffs;
2522
2523 if (s->mb_intra) {
2524 if (!s->h263_aic) {
2525 if (n < 4)
2526 block[0] = block[0] * s->y_dc_scale;
2527 else
2528 block[0] = block[0] * s->c_dc_scale;
2529 }
2530 i = 1;
2531 nCoeffs= 64; //does not allways use zigzag table
2532 } else {
2533 i = 0;
2534 nCoeffs= zigzag_end[ s->block_last_index[n] ];
2535 }
2536
2537 qmul = s->qscale << 1;
2538 if (s->h263_aic && s->mb_intra)
2539 qadd = 0;
2540 else
2541 qadd = (s->qscale - 1) | 1;
2542
2543 for(;i<nCoeffs;i++) {
2544 level = block[i];
2545 if (level) {
2546 if (level < 0) {
2547 level = level * qmul - qadd;
2548 } else {
2549 level = level * qmul + qadd;
2550 }
2551 #ifdef PARANOID
2552 if (level < -2048 || level > 2047)
2553 fprintf(stderr, "unquant error %d %d\n", i, level);
2554 #endif
2555 block[i] = level;
2556 }
2557 }
2558 }
2559
2560 static void remove_ac(MpegEncContext *s, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, int mb_x, int mb_y)
2561 {
2562 int dc, dcb, dcr, y, i;
2563 for(i=0; i<4; i++){
2564 dc= s->dc_val[0][mb_x*2+1 + (i&1) + (mb_y*2+1 + (i>>1))*(s->mb_width*2+2)];
2565 for(y=0; y<8; y++){
2566 int x;
2567 for(x=0; x<8; x++){
2568 dest_y[x + (i&1)*8 + (y + (i>>1)*8)*s->linesize]= dc/8;
2569 }
2570 }
2571 }
2572 dcb = s->dc_val[1][mb_x+1 + (mb_y+1)*(s->mb_width+2)];
2573 dcr= s->dc_val[2][mb_x+1 + (mb_y+1)*(s->mb_width+2)];
2574 for(y=0; y<8; y++){
2575 int x;
2576 for(x=0; x<8; x++){
2577 dest_cb[x + y*(s->uvlinesize)]= dcb/8;
2578 dest_cr[x + y*(s->uvlinesize)]= dcr/8;
2579 }
2580 }
2581 }
2582
2583 /**
2584 * will conceal past errors, and allso drop b frames if needed
2585 *
2586 */
2587 void ff_conceal_past_errors(MpegEncContext *s, int unknown_pos)
2588 {
2589 int mb_x= s->mb_x;
2590 int mb_y= s->mb_y;
2591 int mb_dist=0;
2592 int i, intra_count=0, inter_count=0;
2593 int intra_conceal= s->msmpeg4_version ? 50 : 50; //FIXME finetune
2594 int inter_conceal= s->msmpeg4_version ? 50 : 50;
2595
2596 // for last block
2597 if(mb_x>=s->mb_width) mb_x= s->mb_width -1;
2598 if(mb_y>=s->mb_height) mb_y= s->mb_height-1;
2599
2600 if(s->decoding_error==0 && unknown_pos){
2601 if(s->data_partitioning && s->pict_type!=B_TYPE)
2602 s->decoding_error= DECODING_AC_LOST;
2603 else
2604 s->decoding_error= DECODING_DESYNC;
2605 }
2606
2607 if(s->decoding_error==DECODING_DESYNC && s->pict_type!=B_TYPE) s->next_p_frame_damaged=1;
2608
2609 for(i=mb_x + mb_y*s->mb_width; i>=0; i--){
2610 if(s->mbintra_table[i]) intra_count++;
2611 else inter_count++;
2612 }
2613
2614 if(s->decoding_error==DECODING_AC_LOST){
2615 intra_conceal*=2;
2616 inter_conceal*=2;
2617 }else if(s->decoding_error==DECODING_ACDC_LOST){
2618 intra_conceal*=2;
2619 inter_conceal*=2;
2620 }
2621
2622 if(unknown_pos && (intra_count<inter_count)){
2623 intra_conceal= inter_conceal= s->mb_num;
2624 // printf("%d %d\n",intra_count, inter_count);
2625 }
2626
2627 fprintf(stderr, "concealing errors\n");
2628
2629 /* for all MBs from the current one back until the last resync marker */
2630 for(; mb_y>=0 && mb_y>=s->resync_mb_y; mb_y--){
2631 for(; mb_x>=0; mb_x--){
2632 uint8_t *dest_y = s->current_picture[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
2633 uint8_t *dest_cb = s->current_picture[1] + (mb_y * 8 * (s->uvlinesize)) + mb_x * 8;
2634 uint8_t *dest_cr = s->current_picture[2] + (mb_y * 8 * (s->uvlinesize)) + mb_x * 8;
2635 int mb_x_backup= s->mb_x; //FIXME pass xy to mpeg_motion
2636 int mb_y_backup= s->mb_y;
2637 s->mb_x=mb_x;
2638 s->mb_y=mb_y;
2639 if(s->mbintra_table[mb_y*s->mb_width + mb_x] && mb_dist<intra_conceal){
2640 if(s->decoding_error==DECODING_AC_LOST){
2641 remove_ac(s, dest_y, dest_cb, dest_cr, mb_x, mb_y);
2642 // printf("remove ac to %d %d\n", mb_x, mb_y);
2643 }else{
2644 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
2645 s->last_picture, 0, 0, put_pixels_tab,
2646 0/*mx*/, 0/*my*/, 16);
2647 }
2648 }
2649 else if(!s->mbintra_table[mb_y*s->mb_width + mb_x] && mb_dist<inter_conceal){
2650 int mx=0;
2651 int my=0;
2652
2653 if(s->decoding_error!=DECODING_DESYNC){
2654 int xy= mb_x*2+1 + (mb_y*2+1)*(s->mb_width*2+2);
2655 mx= s->motion_val[ xy ][0];
2656 my= s->motion_val[ xy ][1];
2657 }
2658
2659 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
2660 s->last_picture, 0, 0, put_pixels_tab,
2661 mx, my, 16);
2662 }
2663 s->mb_x= mb_x_backup;
2664 s->mb_y= mb_y_backup;
2665
2666 if(mb_x== s->resync_mb_x && mb_y== s->resync_mb_y) return;
2667 if(!s->mbskip_table[mb_x + mb_y*s->mb_width]) mb_dist++;
2668 }
2669 mb_x=s->mb_width-1;
2670 }
2671 }
2672
2673 AVCodec mpeg1video_encoder = {
2674 "mpeg1video",
2675 CODEC_TYPE_VIDEO,
2676 CODEC_ID_MPEG1VIDEO,
2677 sizeof(MpegEncContext),
2678 MPV_encode_init,
2679 MPV_encode_picture,
2680 MPV_encode_end,
2681 };
2682
2683 AVCodec h263_encoder = {
2684 "h263",
2685 CODEC_TYPE_VIDEO,
2686 CODEC_ID_H263,
2687 sizeof(MpegEncContext),
2688 MPV_encode_init,
2689 MPV_encode_picture,
2690 MPV_encode_end,
2691 };
2692
2693 AVCodec h263p_encoder = {
2694 "h263p",
2695 CODEC_TYPE_VIDEO,
2696 CODEC_ID_H263P,
2697 sizeof(MpegEncContext),
2698 MPV_encode_init,
2699 MPV_encode_picture,
2700 MPV_encode_end,
2701 };
2702
2703 AVCodec rv10_encoder = {
2704 "rv10",
2705 CODEC_TYPE_VIDEO,
2706 CODEC_ID_RV10,
2707 sizeof(MpegEncContext),
2708 MPV_encode_init,
2709 MPV_encode_picture,
2710 MPV_encode_end,
2711 };
2712
2713 AVCodec mjpeg_encoder = {
2714 "mjpeg",
2715 CODEC_TYPE_VIDEO,
2716 CODEC_ID_MJPEG,
2717 sizeof(MpegEncContext),
2718 MPV_encode_init,
2719 MPV_encode_picture,
2720 MPV_encode_end,
2721 };
2722
2723 AVCodec mpeg4_encoder = {
2724 "mpeg4",
2725 CODEC_TYPE_VIDEO,
2726 CODEC_ID_MPEG4,
2727 sizeof(MpegEncContext),
2728 MPV_encode_init,
2729 MPV_encode_picture,
2730 MPV_encode_end,
2731 };
2732
2733 AVCodec msmpeg4v1_encoder = {
2734 "msmpeg4v1",
2735 CODEC_TYPE_VIDEO,
2736 CODEC_ID_MSMPEG4V1,
2737 sizeof(MpegEncContext),
2738 MPV_encode_init,
2739 MPV_encode_picture,
2740 MPV_encode_end,
2741 };
2742
2743 AVCodec msmpeg4v2_encoder = {
2744 "msmpeg4v2",
2745 CODEC_TYPE_VIDEO,
2746 CODEC_ID_MSMPEG4V2,
2747 sizeof(MpegEncContext),
2748 MPV_encode_init,
2749 MPV_encode_picture,
2750 MPV_encode_end,
2751 };
2752
2753 AVCodec msmpeg4v3_encoder = {
2754 "msmpeg4",
2755 CODEC_TYPE_VIDEO,
2756 CODEC_ID_MSMPEG4V3,
2757 sizeof(MpegEncContext),
2758 MPV_encode_init,
2759 MPV_encode_picture,
2760 MPV_encode_end,
2761 };
2762
2763 AVCodec wmv1_encoder = {
2764 "wmv1",
2765 CODEC_TYPE_VIDEO,
2766 CODEC_ID_WMV1,
2767 sizeof(MpegEncContext),
2768 MPV_encode_init,
2769 MPV_encode_picture,
2770 MPV_encode_end,
2771 };
2772
2773 AVCodec wmv2_encoder = {
2774 "wmv2",
2775 CODEC_TYPE_VIDEO,
2776 CODEC_ID_WMV2,
2777 sizeof(MpegEncContext),
2778 MPV_encode_init,
2779 MPV_encode_picture,
2780 MPV_encode_end,
2781 };