71d571df85ea1f8b64f3905bba6beaae49ecf3d7
[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 s->linesize = s->mb_width * 16 + 2 * EDGE_WIDTH;
148
149 for(i=0;i<3;i++) {
150 int w, h, shift, pict_start;
151
152 w = s->linesize;
153 h = s->mb_height * 16 + 2 * EDGE_WIDTH;
154 shift = (i == 0) ? 0 : 1;
155 c_size = (w >> shift) * (h >> shift);
156 pict_start = (w >> shift) * (EDGE_WIDTH >> shift) + (EDGE_WIDTH >> shift);
157
158 CHECKED_ALLOCZ(pict, c_size)
159 s->last_picture_base[i] = pict;
160 s->last_picture[i] = pict + pict_start;
161 if(i>0) memset(s->last_picture_base[i], 128, c_size);
162
163 CHECKED_ALLOCZ(pict, c_size)
164 s->next_picture_base[i] = pict;
165 s->next_picture[i] = pict + pict_start;
166 if(i>0) memset(s->next_picture_base[i], 128, c_size);
167
168 if (s->has_b_frames || s->codec_id==CODEC_ID_MPEG4) {
169 /* Note the MPEG4 stuff is here cuz of buggy encoders which dont set the low_delay flag but
170 do low-delay encoding, so we cant allways distinguish b-frame containing streams from low_delay streams */
171 CHECKED_ALLOCZ(pict, c_size)
172 s->aux_picture_base[i] = pict;
173 s->aux_picture[i] = pict + pict_start;
174 if(i>0) memset(s->aux_picture_base[i], 128, c_size);
175 }
176 }
177
178 if (s->encoding) {
179 int j;
180 int mv_table_size= (s->mb_width+2)*(s->mb_height+2);
181
182 CHECKED_ALLOCZ(s->mb_var , s->mb_num * sizeof(INT16))
183 CHECKED_ALLOCZ(s->mc_mb_var, s->mb_num * sizeof(INT16))
184
185 /* Allocate MV tables */
186 CHECKED_ALLOCZ(s->p_mv_table , mv_table_size * 2 * sizeof(INT16))
187 CHECKED_ALLOCZ(s->b_forw_mv_table , mv_table_size * 2 * sizeof(INT16))
188 CHECKED_ALLOCZ(s->b_back_mv_table , mv_table_size * 2 * sizeof(INT16))
189 CHECKED_ALLOCZ(s->b_bidir_forw_mv_table , mv_table_size * 2 * sizeof(INT16))
190 CHECKED_ALLOCZ(s->b_bidir_back_mv_table , mv_table_size * 2 * sizeof(INT16))
191 CHECKED_ALLOCZ(s->b_direct_forw_mv_table, mv_table_size * 2 * sizeof(INT16))
192 CHECKED_ALLOCZ(s->b_direct_back_mv_table, mv_table_size * 2 * sizeof(INT16))
193 CHECKED_ALLOCZ(s->b_direct_mv_table , mv_table_size * 2 * sizeof(INT16))
194
195 CHECKED_ALLOCZ(s->me_scratchpad, s->linesize*16*3*sizeof(uint8_t))
196
197 CHECKED_ALLOCZ(s->me_map , ME_MAP_SIZE*sizeof(uint32_t))
198 CHECKED_ALLOCZ(s->me_score_map, ME_MAP_SIZE*sizeof(uint16_t))
199
200 if(s->max_b_frames){
201 for(j=0; j<REORDER_BUFFER_SIZE; j++){
202 int i;
203 for(i=0;i<3;i++) {
204 int w, h, shift;
205
206 w = s->linesize;
207 h = s->mb_height * 16;
208 shift = (i == 0) ? 0 : 1;
209 c_size = (w >> shift) * (h >> shift);
210
211 CHECKED_ALLOCZ(pict, c_size);
212 s->picture_buffer[j][i] = pict;
213 }
214 }
215 }
216
217 if(s->codec_id==CODEC_ID_MPEG4){
218 CHECKED_ALLOCZ(s->tex_pb_buffer, PB_BUFFER_SIZE);
219 CHECKED_ALLOCZ( s->pb2_buffer, PB_BUFFER_SIZE);
220 }
221 }
222
223 if (s->out_format == FMT_H263 || s->encoding) {
224 int size;
225 /* Allocate MB type table */
226 CHECKED_ALLOCZ(s->mb_type , s->mb_num * sizeof(UINT8))
227
228 /* MV prediction */
229 size = (2 * s->mb_width + 2) * (2 * s->mb_height + 2);
230 CHECKED_ALLOCZ(s->motion_val, size * 2 * sizeof(INT16));
231 }
232
233 if (s->h263_pred || s->h263_plus) {
234 int y_size, c_size, i, size;
235
236 /* dc values */
237
238 y_size = (2 * s->mb_width + 2) * (2 * s->mb_height + 2);
239 c_size = (s->mb_width + 2) * (s->mb_height + 2);
240 size = y_size + 2 * c_size;
241 CHECKED_ALLOCZ(s->dc_val[0], size * sizeof(INT16));
242 s->dc_val[1] = s->dc_val[0] + y_size;
243 s->dc_val[2] = s->dc_val[1] + c_size;
244 for(i=0;i<size;i++)
245 s->dc_val[0][i] = 1024;
246
247 /* ac values */
248 CHECKED_ALLOCZ(s->ac_val[0], size * sizeof(INT16) * 16);
249 s->ac_val[1] = s->ac_val[0] + y_size;
250 s->ac_val[2] = s->ac_val[1] + c_size;
251
252 /* cbp values */
253 CHECKED_ALLOCZ(s->coded_block, y_size);
254
255 /* which mb is a intra block */
256 CHECKED_ALLOCZ(s->mbintra_table, s->mb_num);
257 memset(s->mbintra_table, 1, s->mb_num);
258
259 /* divx501 bitstream reorder buffer */
260 CHECKED_ALLOCZ(s->bitstream_buffer, BITSTREAM_BUFFER_SIZE);
261
262 /* cbp, ac_pred, pred_dir */
263 CHECKED_ALLOCZ(s->cbp_table , s->mb_num * sizeof(UINT8))
264 CHECKED_ALLOCZ(s->pred_dir_table, s->mb_num * sizeof(UINT8))
265
266 CHECKED_ALLOCZ(s->qscale_table , s->mb_num * sizeof(UINT8))
267 }
268 /* default structure is frame */
269 s->picture_structure = PICT_FRAME;
270
271 /* init macroblock skip table */
272 CHECKED_ALLOCZ(s->mbskip_table, s->mb_num);
273
274 s->block= s->blocks[0];
275
276 s->context_initialized = 1;
277 return 0;
278 fail:
279 MPV_common_end(s);
280 return -1;
281 }
282
283
284 //extern int sads;
285
286 /* init common structure for both encoder and decoder */
287 void MPV_common_end(MpegEncContext *s)
288 {
289 int i;
290
291 av_freep(&s->mb_type);
292 av_freep(&s->mb_var);
293 av_freep(&s->mc_mb_var);
294 av_freep(&s->p_mv_table);
295 av_freep(&s->b_forw_mv_table);
296 av_freep(&s->b_back_mv_table);
297 av_freep(&s->b_bidir_forw_mv_table);
298 av_freep(&s->b_bidir_back_mv_table);
299 av_freep(&s->b_direct_forw_mv_table);
300 av_freep(&s->b_direct_back_mv_table);
301 av_freep(&s->b_direct_mv_table);
302 av_freep(&s->motion_val);
303 av_freep(&s->dc_val[0]);
304 av_freep(&s->ac_val[0]);
305 av_freep(&s->coded_block);
306 av_freep(&s->mbintra_table);
307 av_freep(&s->cbp_table);
308 av_freep(&s->pred_dir_table);
309 av_freep(&s->qscale_table);
310 av_freep(&s->me_scratchpad);
311 av_freep(&s->me_map);
312 av_freep(&s->me_score_map);
313
314 av_freep(&s->mbskip_table);
315 av_freep(&s->bitstream_buffer);
316 av_freep(&s->tex_pb_buffer);
317 av_freep(&s->pb2_buffer);
318 for(i=0;i<3;i++) {
319 int j;
320 av_freep(&s->last_picture_base[i]);
321 av_freep(&s->next_picture_base[i]);
322 av_freep(&s->aux_picture_base[i]);
323 for(j=0; j<REORDER_BUFFER_SIZE; j++){
324 av_freep(&s->picture_buffer[j][i]);
325 }
326 }
327 s->context_initialized = 0;
328 }
329
330 /* init video encoder */
331 int MPV_encode_init(AVCodecContext *avctx)
332 {
333 MpegEncContext *s = avctx->priv_data;
334 int i;
335
336 avctx->pix_fmt = PIX_FMT_YUV420P;
337
338 s->bit_rate = avctx->bit_rate;
339 s->bit_rate_tolerance = avctx->bit_rate_tolerance;
340 s->frame_rate = avctx->frame_rate;
341 s->width = avctx->width;
342 s->height = avctx->height;
343 if(avctx->gop_size > 600){
344 fprintf(stderr, "Warning keyframe interval too large! reducing it ...\n");
345 avctx->gop_size=600;
346 }
347 s->gop_size = avctx->gop_size;
348 s->rtp_mode = avctx->rtp_mode;
349 s->rtp_payload_size = avctx->rtp_payload_size;
350 if (avctx->rtp_callback)
351 s->rtp_callback = avctx->rtp_callback;
352 s->qmin= avctx->qmin;
353 s->qmax= avctx->qmax;
354 s->max_qdiff= avctx->max_qdiff;
355 s->qcompress= avctx->qcompress;
356 s->qblur= avctx->qblur;
357 s->b_quant_factor= avctx->b_quant_factor;
358 s->b_quant_offset= avctx->b_quant_offset;
359 s->avctx = avctx;
360 s->aspect_ratio_info= avctx->aspect_ratio_info;
361 s->flags= avctx->flags;
362 s->max_b_frames= avctx->max_b_frames;
363 s->rc_strategy= avctx->rc_strategy;
364 s->b_frame_strategy= avctx->b_frame_strategy;
365 s->codec_id= avctx->codec->id;
366 s->luma_elim_threshold = avctx->luma_elim_threshold;
367 s->chroma_elim_threshold= avctx->chroma_elim_threshold;
368 s->strict_std_compliance= avctx->strict_std_compliance;
369 s->data_partitioning= avctx->flags & CODEC_FLAG_PART;
370
371 if (s->gop_size <= 1) {
372 s->intra_only = 1;
373 s->gop_size = 12;
374 } else {
375 s->intra_only = 0;
376 }
377
378 /* ME algorithm */
379 if (avctx->me_method == 0)
380 /* For compatibility */
381 s->me_method = motion_estimation_method;
382 else
383 s->me_method = avctx->me_method;
384
385 /* Fixed QSCALE */
386 s->fixed_qscale = (avctx->flags & CODEC_FLAG_QSCALE);
387
388 switch(avctx->codec->id) {
389 case CODEC_ID_MPEG1VIDEO:
390 s->out_format = FMT_MPEG1;
391 avctx->delay=0; //FIXME not sure, should check the spec
392 break;
393 case CODEC_ID_MJPEG:
394 s->out_format = FMT_MJPEG;
395 s->intra_only = 1; /* force intra only for jpeg */
396 s->mjpeg_write_tables = 1; /* write all tables */
397 s->mjpeg_data_only_frames = 0; /* write all the needed headers */
398 s->mjpeg_vsample[0] = 2; /* set up default sampling factors */
399 s->mjpeg_vsample[1] = 1; /* the only currently supported values */
400 s->mjpeg_vsample[2] = 1;
401 s->mjpeg_hsample[0] = 2;
402 s->mjpeg_hsample[1] = 1;
403 s->mjpeg_hsample[2] = 1;
404 if (mjpeg_init(s) < 0)
405 return -1;
406 avctx->delay=0;
407 break;
408 case CODEC_ID_H263:
409 if (h263_get_picture_format(s->width, s->height) == 7) {
410 printf("Input picture size isn't suitable for h263 codec! try h263+\n");
411 return -1;
412 }
413 s->out_format = FMT_H263;
414 avctx->delay=0;
415 break;
416 case CODEC_ID_H263P:
417 s->out_format = FMT_H263;
418 s->rtp_mode = 1;
419 s->rtp_payload_size = 1200;
420 s->h263_plus = 1;
421 s->unrestricted_mv = 1;
422 s->h263_aic = 1;
423
424 /* These are just to be sure */
425 s->umvplus = 0;
426 s->umvplus_dec = 0;
427 avctx->delay=0;
428 break;
429 case CODEC_ID_RV10:
430 s->out_format = FMT_H263;
431 s->h263_rv10 = 1;
432 avctx->delay=0;
433 break;
434 case CODEC_ID_MPEG4:
435 s->out_format = FMT_H263;
436 s->h263_pred = 1;
437 s->unrestricted_mv = 1;
438 s->has_b_frames= s->max_b_frames ? 1 : 0;
439 s->low_delay=0;
440 avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
441 break;
442 case CODEC_ID_MSMPEG4V1:
443 s->out_format = FMT_H263;
444 s->h263_msmpeg4 = 1;
445 s->h263_pred = 1;
446 s->unrestricted_mv = 1;
447 s->msmpeg4_version= 1;
448 avctx->delay=0;
449 break;
450 case CODEC_ID_MSMPEG4V2:
451 s->out_format = FMT_H263;
452 s->h263_msmpeg4 = 1;
453 s->h263_pred = 1;
454 s->unrestricted_mv = 1;
455 s->msmpeg4_version= 2;
456 avctx->delay=0;
457 break;
458 case CODEC_ID_MSMPEG4V3:
459 s->out_format = FMT_H263;
460 s->h263_msmpeg4 = 1;
461 s->h263_pred = 1;
462 s->unrestricted_mv = 1;
463 s->msmpeg4_version= 3;
464 avctx->delay=0;
465 break;
466 case CODEC_ID_WMV1:
467 s->out_format = FMT_H263;
468 s->h263_msmpeg4 = 1;
469 s->h263_pred = 1;
470 s->unrestricted_mv = 1;
471 s->msmpeg4_version= 4;
472 avctx->delay=0;
473 break;
474 case CODEC_ID_WMV2:
475 s->out_format = FMT_H263;
476 s->h263_msmpeg4 = 1;
477 s->h263_pred = 1;
478 s->unrestricted_mv = 1;
479 s->msmpeg4_version= 5;
480 avctx->delay=0;
481 break;
482 default:
483 return -1;
484 }
485
486 { /* set up some save defaults, some codecs might override them later */
487 static int done=0;
488 if(!done){
489 int i;
490 done=1;
491 memset(default_mv_penalty, 0, sizeof(UINT16)*(MAX_FCODE+1)*(2*MAX_MV+1));
492 memset(default_fcode_tab , 0, sizeof(UINT8)*(2*MAX_MV+1));
493
494 for(i=-16; i<16; i++){
495 default_fcode_tab[i + MAX_MV]= 1;
496 }
497 }
498 }
499 s->mv_penalty= default_mv_penalty;
500 s->fcode_tab= default_fcode_tab;
501 s->y_dc_scale_table=
502 s->c_dc_scale_table= ff_mpeg1_dc_scale_table;
503
504 if (s->out_format == FMT_H263)
505 h263_encode_init(s);
506 else if (s->out_format == FMT_MPEG1)
507 ff_mpeg1_encode_init(s);
508 if(s->msmpeg4_version)
509 ff_msmpeg4_encode_init(s);
510
511 /* dont use mv_penalty table for crap MV as it would be confused */
512 if (s->me_method < ME_EPZS) s->mv_penalty = default_mv_penalty;
513
514 s->encoding = 1;
515
516 /* init */
517 if (MPV_common_init(s) < 0)
518 return -1;
519
520 /* init default q matrix */
521 for(i=0;i<64;i++) {
522 if(s->out_format == FMT_H263)
523 s->intra_matrix[i] = ff_mpeg1_default_non_intra_matrix[i];
524 else
525 s->intra_matrix[i] = ff_mpeg1_default_intra_matrix[i];
526
527 s->inter_matrix[i] = ff_mpeg1_default_non_intra_matrix[i];
528 }
529
530 /* precompute matrix */
531 /* for mjpeg, we do include qscale in the matrix */
532 if (s->out_format != FMT_MJPEG) {
533 convert_matrix(s->q_intra_matrix, s->q_intra_matrix16, s->q_intra_matrix16_bias,
534 s->intra_matrix, s->intra_quant_bias);
535 convert_matrix(s->q_inter_matrix, s->q_inter_matrix16, s->q_inter_matrix16_bias,
536 s->inter_matrix, s->inter_quant_bias);
537 }
538
539 if(ff_rate_control_init(s) < 0)
540 return -1;
541
542 s->picture_number = 0;
543 s->picture_in_gop_number = 0;
544 s->fake_picture_number = 0;
545 /* motion detector init */
546 s->f_code = 1;
547 s->b_code = 1;
548
549 return 0;
550 }
551
552 int MPV_encode_end(AVCodecContext *avctx)
553 {
554 MpegEncContext *s = avctx->priv_data;
555
556 #ifdef STATS
557 print_stats();
558 #endif
559
560 ff_rate_control_uninit(s);
561
562 MPV_common_end(s);
563 if (s->out_format == FMT_MJPEG)
564 mjpeg_close(s);
565
566 return 0;
567 }
568
569 /* draw the edges of width 'w' of an image of size width, height */
570 static void draw_edges_c(UINT8 *buf, int wrap, int width, int height, int w)
571 {
572 UINT8 *ptr, *last_line;
573 int i;
574
575 last_line = buf + (height - 1) * wrap;
576 for(i=0;i<w;i++) {
577 /* top and bottom */
578 memcpy(buf - (i + 1) * wrap, buf, width);
579 memcpy(last_line + (i + 1) * wrap, last_line, width);
580 }
581 /* left and right */
582 ptr = buf;
583 for(i=0;i<height;i++) {
584 memset(ptr - w, ptr[0], w);
585 memset(ptr + width, ptr[width-1], w);
586 ptr += wrap;
587 }
588 /* corners */
589 for(i=0;i<w;i++) {
590 memset(buf - (i + 1) * wrap - w, buf[0], w); /* top left */
591 memset(buf - (i + 1) * wrap + width, buf[width-1], w); /* top right */
592 memset(last_line + (i + 1) * wrap - w, last_line[0], w); /* top left */
593 memset(last_line + (i + 1) * wrap + width, last_line[width-1], w); /* top right */
594 }
595 }
596
597 /* generic function for encode/decode called before a frame is coded/decoded */
598 void MPV_frame_start(MpegEncContext *s)
599 {
600 int i;
601 UINT8 *tmp;
602
603 s->mb_skiped = 0;
604 s->decoding_error=0;
605
606 if (s->pict_type == B_TYPE) {
607 for(i=0;i<3;i++) {
608 s->current_picture[i] = s->aux_picture[i];
609 }
610 } else {
611 for(i=0;i<3;i++) {
612 /* swap next and last */
613 tmp = s->last_picture[i];
614 s->last_picture[i] = s->next_picture[i];
615 s->next_picture[i] = tmp;
616 s->current_picture[i] = tmp;
617 }
618 }
619 }
620
621 /* generic function for encode/decode called after a frame has been coded/decoded */
622 void MPV_frame_end(MpegEncContext *s)
623 {
624 // if((s->picture_number%100)==0 && s->encoding) printf("sads:%d //\n", sads);
625
626 /* draw edge for correct motion prediction if outside */
627 if (s->pict_type != B_TYPE && !s->intra_only) {
628 if(s->avctx==NULL || s->avctx->codec->id!=CODEC_ID_MPEG4 || s->divx_version>=500){
629 draw_edges(s->current_picture[0], s->linesize, s->mb_width*16, s->mb_height*16, EDGE_WIDTH);
630 draw_edges(s->current_picture[1], s->linesize/2, s->mb_width*8, s->mb_height*8, EDGE_WIDTH/2);
631 draw_edges(s->current_picture[2], s->linesize/2, s->mb_width*8, s->mb_height*8, EDGE_WIDTH/2);
632 }else{
633 /* mpeg4? / opendivx / xvid */
634 draw_edges(s->current_picture[0], s->linesize, s->width, s->height, EDGE_WIDTH);
635 draw_edges(s->current_picture[1], s->linesize/2, s->width/2, s->height/2, EDGE_WIDTH/2);
636 draw_edges(s->current_picture[2], s->linesize/2, s->width/2, s->height/2, EDGE_WIDTH/2);
637 }
638 }
639 emms_c();
640
641 if(s->pict_type!=B_TYPE){
642 s->last_non_b_pict_type= s->pict_type;
643 s->last_non_b_qscale= s->qscale;
644 s->last_non_b_mc_mb_var= s->mc_mb_var_sum;
645 s->num_available_buffers++;
646 if(s->num_available_buffers>2) s->num_available_buffers= 2;
647 }
648 }
649
650 /* reorder input for encoding */
651 void reorder_input(MpegEncContext *s, AVPicture *pict)
652 {
653 int i, j, index;
654
655 if(s->max_b_frames > FF_MAX_B_FRAMES) s->max_b_frames= FF_MAX_B_FRAMES;
656
657 // delay= s->max_b_frames+1; (or 0 if no b frames cuz decoder diff)
658
659 for(j=0; j<REORDER_BUFFER_SIZE-1; j++){
660 s->coded_order[j]= s->coded_order[j+1];
661 }
662 s->coded_order[j].picture[0]= s->coded_order[j].picture[1]= s->coded_order[j].picture[2]= NULL; //catch uninitalized buffers
663 s->coded_order[j].pict_type=0;
664
665 switch(s->input_pict_type){
666 default:
667 case I_TYPE:
668 case S_TYPE:
669 case P_TYPE:
670 index= s->max_b_frames - s->b_frames_since_non_b;
671 s->b_frames_since_non_b=0;
672 break;
673 case B_TYPE:
674 index= s->max_b_frames + 1;
675 s->b_frames_since_non_b++;
676 break;
677 }
678 //printf("index:%d type:%d strides: %d %d\n", index, s->input_pict_type, pict->linesize[0], s->linesize);
679 if( (index==0 || (s->flags&CODEC_FLAG_INPUT_PRESERVED))
680 && pict->linesize[0] == s->linesize
681 && pict->linesize[1] == s->linesize>>1
682 && pict->linesize[2] == s->linesize>>1){
683 //printf("ptr\n");
684 for(i=0; i<3; i++){
685 s->coded_order[index].picture[i]= pict->data[i];
686 }
687 }else{
688 //printf("copy\n");
689 for(i=0; i<3; i++){
690 uint8_t *src = pict->data[i];
691 uint8_t *dest;
692 int src_wrap = pict->linesize[i];
693 int dest_wrap = s->linesize;
694 int w = s->width;
695 int h = s->height;
696
697 if(index==0) dest= s->last_picture[i]+16; //is current_picture indeed but the switch hapens after reordering
698 else dest= s->picture_buffer[s->picture_buffer_index][i];
699
700 if (i >= 1) {
701 dest_wrap >>= 1;
702 w >>= 1;
703 h >>= 1;
704 }
705
706 s->coded_order[index].picture[i]= dest;
707 for(j=0;j<h;j++) {
708 memcpy(dest, src, w);
709 dest += dest_wrap;
710 src += src_wrap;
711 }
712 }
713 if(index!=0){
714 s->picture_buffer_index++;
715 if(s->picture_buffer_index >= REORDER_BUFFER_SIZE-1) s->picture_buffer_index=0;
716 }
717 }
718 s->coded_order[index].pict_type = s->input_pict_type;
719 s->coded_order[index].qscale = s->input_qscale;
720 s->coded_order[index].force_type= s->force_input_type;
721 s->coded_order[index].picture_in_gop_number= s->input_picture_in_gop_number;
722 s->coded_order[index].picture_number= s->input_picture_number;
723
724 for(i=0; i<3; i++){
725 s->new_picture[i]= s->coded_order[0].picture[i];
726 }
727 }
728
729 int MPV_encode_picture(AVCodecContext *avctx,
730 unsigned char *buf, int buf_size, void *data)
731 {
732 MpegEncContext *s = avctx->priv_data;
733 AVPicture *pict = data;
734
735 s->input_qscale = avctx->quality;
736
737 init_put_bits(&s->pb, buf, buf_size, NULL, NULL);
738
739 if(avctx->flags&CODEC_FLAG_TYPE){
740 s->input_pict_type=
741 s->force_input_type= avctx->key_frame ? I_TYPE : P_TYPE;
742 }else if(s->flags&CODEC_FLAG_PASS2){
743 s->input_pict_type=
744 s->force_input_type= s->rc_context.entry[s->input_picture_number].new_pict_type;
745 }else{
746 s->force_input_type=0;
747 if (!s->intra_only) {
748 /* first picture of GOP is intra */
749 if (s->input_picture_in_gop_number % s->gop_size==0){
750 s->input_pict_type = I_TYPE;
751 }else if(s->max_b_frames==0){
752 s->input_pict_type = P_TYPE;
753 }else{
754 if(s->b_frames_since_non_b < s->max_b_frames) //FIXME more IQ
755 s->input_pict_type = B_TYPE;
756 else
757 s->input_pict_type = P_TYPE;
758 }
759 } else {
760 s->input_pict_type = I_TYPE;
761 }
762 }
763
764 if(s->input_pict_type==I_TYPE)
765 s->input_picture_in_gop_number=0;
766
767 reorder_input(s, pict);
768
769 /* output? */
770 if(s->coded_order[0].picture[0]){
771
772 s->pict_type= s->coded_order[0].pict_type;
773 if (s->fixed_qscale) /* the ratecontrol needs the last qscale so we dont touch it for CBR */
774 s->qscale= s->coded_order[0].qscale;
775 s->force_type= s->coded_order[0].force_type;
776 s->picture_in_gop_number= s->coded_order[0].picture_in_gop_number;
777 s->picture_number= s->coded_order[0].picture_number;
778
779 MPV_frame_start(s);
780
781 encode_picture(s, s->picture_number);
782 avctx->key_frame = (s->pict_type == I_TYPE);
783 avctx->pict_type = s->pict_type;
784 avctx->real_pict_num = s->picture_number;
785 avctx->header_bits = s->header_bits;
786 avctx->mv_bits = s->mv_bits;
787 avctx->misc_bits = s->misc_bits;
788 avctx->i_tex_bits = s->i_tex_bits;
789 avctx->p_tex_bits = s->p_tex_bits;
790 avctx->i_count = s->i_count;
791 avctx->p_count = s->p_count;
792 avctx->skip_count = s->skip_count;
793
794 MPV_frame_end(s);
795
796 if (s->out_format == FMT_MJPEG)
797 mjpeg_picture_trailer(s);
798
799 avctx->quality = s->qscale;
800
801 if(s->flags&CODEC_FLAG_PASS1)
802 ff_write_pass1_stats(s);
803
804 }
805
806 s->input_picture_number++;
807 s->input_picture_in_gop_number++;
808
809 flush_put_bits(&s->pb);
810 s->frame_bits = (pbBufPtr(&s->pb) - s->pb.buf) * 8;
811 if(s->pict_type==B_TYPE) s->pb_frame_bits+= s->frame_bits;
812 else s->pb_frame_bits= s->frame_bits;
813
814 s->total_bits += s->frame_bits;
815 avctx->frame_bits = s->frame_bits;
816 //printf("fcode: %d, type: %d, head: %d, mv: %d, misc: %d, frame: %d, itex: %d, ptex: %d\n",
817 //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);
818
819 if (avctx->get_psnr) {
820 /* At this point pict->data should have the original frame */
821 /* an s->current_picture should have the coded/decoded frame */
822 get_psnr(pict->data, s->current_picture,
823 pict->linesize, s->linesize, avctx);
824 // printf("%f\n", avctx->psnr_y);
825 }
826 return pbBufPtr(&s->pb) - s->pb.buf;
827 }
828
829 static inline void gmc1_motion(MpegEncContext *s,
830 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr,
831 int dest_offset,
832 UINT8 **ref_picture, int src_offset,
833 int h)
834 {
835 UINT8 *ptr;
836 int offset, src_x, src_y, linesize;
837 int motion_x, motion_y;
838
839 if(s->real_sprite_warping_points>1) printf("more than 1 warp point isnt supported\n");
840 motion_x= s->sprite_offset[0][0];
841 motion_y= s->sprite_offset[0][1];
842 src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy+1));
843 src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy+1));
844 motion_x<<=(3-s->sprite_warping_accuracy);
845 motion_y<<=(3-s->sprite_warping_accuracy);
846 src_x = clip(src_x, -16, s->width);
847 if (src_x == s->width)
848 motion_x =0;
849 src_y = clip(src_y, -16, s->height);
850 if (src_y == s->height)
851 motion_y =0;
852
853 linesize = s->linesize;
854 ptr = ref_picture[0] + (src_y * linesize) + src_x + src_offset;
855
856 dest_y+=dest_offset;
857 gmc1(dest_y , ptr , linesize, h, motion_x&15, motion_y&15, s->no_rounding);
858 gmc1(dest_y+8, ptr+8, linesize, h, motion_x&15, motion_y&15, s->no_rounding);
859
860 motion_x= s->sprite_offset[1][0];
861 motion_y= s->sprite_offset[1][1];
862 src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy+1));
863 src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy+1));
864 motion_x<<=(3-s->sprite_warping_accuracy);
865 motion_y<<=(3-s->sprite_warping_accuracy);
866 src_x = clip(src_x, -8, s->width>>1);
867 if (src_x == s->width>>1)
868 motion_x =0;
869 src_y = clip(src_y, -8, s->height>>1);
870 if (src_y == s->height>>1)
871 motion_y =0;
872
873 offset = (src_y * linesize>>1) + src_x + (src_offset>>1);
874 ptr = ref_picture[1] + offset;
875 gmc1(dest_cb + (dest_offset>>1), ptr, linesize>>1, h>>1, motion_x&15, motion_y&15, s->no_rounding);
876 ptr = ref_picture[2] + offset;
877 gmc1(dest_cr + (dest_offset>>1), ptr, linesize>>1, h>>1, motion_x&15, motion_y&15, s->no_rounding);
878
879 return;
880 }
881
882 /* apply one mpeg motion vector to the three components */
883 static inline void mpeg_motion(MpegEncContext *s,
884 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr,
885 int dest_offset,
886 UINT8 **ref_picture, int src_offset,
887 int field_based, op_pixels_func *pix_op,
888 int motion_x, int motion_y, int h)
889 {
890 UINT8 *ptr;
891 int dxy, offset, mx, my, src_x, src_y, height, linesize;
892 if(s->quarter_sample)
893 {
894 motion_x>>=1;
895 motion_y>>=1;
896 }
897 dxy = ((motion_y & 1) << 1) | (motion_x & 1);
898 src_x = s->mb_x * 16 + (motion_x >> 1);
899 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 1);
900
901 /* WARNING: do no forget half pels */
902 height = s->height >> field_based;
903 src_x = clip(src_x, -16, s->width);
904 if (src_x == s->width)
905 dxy &= ~1;
906 src_y = clip(src_y, -16, height);
907 if (src_y == height)
908 dxy &= ~2;
909 linesize = s->linesize << field_based;
910 ptr = ref_picture[0] + (src_y * linesize) + (src_x) + src_offset;
911 dest_y += dest_offset;
912 pix_op[dxy](dest_y, ptr, linesize, h);
913 pix_op[dxy](dest_y + 8, ptr + 8, linesize, h);
914
915 if(s->flags&CODEC_FLAG_GRAY) return;
916
917 if (s->out_format == FMT_H263) {
918 dxy = 0;
919 if ((motion_x & 3) != 0)
920 dxy |= 1;
921 if ((motion_y & 3) != 0)
922 dxy |= 2;
923 mx = motion_x >> 2;
924 my = motion_y >> 2;
925 } else {
926 mx = motion_x / 2;
927 my = motion_y / 2;
928 dxy = ((my & 1) << 1) | (mx & 1);
929 mx >>= 1;
930 my >>= 1;
931 }
932
933 src_x = s->mb_x * 8 + mx;
934 src_y = s->mb_y * (8 >> field_based) + my;
935 src_x = clip(src_x, -8, s->width >> 1);
936 if (src_x == (s->width >> 1))
937 dxy &= ~1;
938 src_y = clip(src_y, -8, height >> 1);
939 if (src_y == (height >> 1))
940 dxy &= ~2;
941
942 offset = (src_y * (linesize >> 1)) + src_x + (src_offset >> 1);
943 ptr = ref_picture[1] + offset;
944 pix_op[dxy](dest_cb + (dest_offset >> 1), ptr, linesize >> 1, h >> 1);
945 ptr = ref_picture[2] + offset;
946 pix_op[dxy](dest_cr + (dest_offset >> 1), ptr, linesize >> 1, h >> 1);
947 }
948
949 static inline void qpel_motion(MpegEncContext *s,
950 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr,
951 int dest_offset,
952 UINT8 **ref_picture, int src_offset,
953 int field_based, op_pixels_func *pix_op,
954 qpel_mc_func *qpix_op,
955 int motion_x, int motion_y, int h)
956 {
957 UINT8 *ptr;
958 int dxy, offset, mx, my, src_x, src_y, height, linesize;
959
960 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
961 src_x = s->mb_x * 16 + (motion_x >> 2);
962 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2);
963
964 height = s->height >> field_based;
965 src_x = clip(src_x, -16, s->width);
966 if (src_x == s->width)
967 dxy &= ~3;
968 src_y = clip(src_y, -16, height);
969 if (src_y == height)
970 dxy &= ~12;
971 linesize = s->linesize << field_based;
972 ptr = ref_picture[0] + (src_y * linesize) + src_x + src_offset;
973 dest_y += dest_offset;
974 //printf("%d %d %d\n", src_x, src_y, dxy);
975 qpix_op[dxy](dest_y , ptr , linesize, linesize, motion_x&3, motion_y&3);
976 qpix_op[dxy](dest_y + 8, ptr + 8, linesize, linesize, motion_x&3, motion_y&3);
977 qpix_op[dxy](dest_y + linesize*8 , ptr + linesize*8 , linesize, linesize, motion_x&3, motion_y&3);
978 qpix_op[dxy](dest_y + linesize*8 + 8, ptr + linesize*8 + 8, linesize, linesize, motion_x&3, motion_y&3);
979
980 if(s->flags&CODEC_FLAG_GRAY) return;
981
982 mx= (motion_x>>1) | (motion_x&1);
983 my= (motion_y>>1) | (motion_y&1);
984
985 dxy = 0;
986 if ((mx & 3) != 0)
987 dxy |= 1;
988 if ((my & 3) != 0)
989 dxy |= 2;
990 mx = mx >> 2;
991 my = my >> 2;
992
993 src_x = s->mb_x * 8 + mx;
994 src_y = s->mb_y * (8 >> field_based) + my;
995 src_x = clip(src_x, -8, s->width >> 1);
996 if (src_x == (s->width >> 1))
997 dxy &= ~1;
998 src_y = clip(src_y, -8, height >> 1);
999 if (src_y == (height >> 1))
1000 dxy &= ~2;
1001
1002 offset = (src_y * (linesize >> 1)) + src_x + (src_offset >> 1);
1003 ptr = ref_picture[1] + offset;
1004 pix_op[dxy](dest_cb + (dest_offset >> 1), ptr, linesize >> 1, h >> 1);
1005 ptr = ref_picture[2] + offset;
1006 pix_op[dxy](dest_cr + (dest_offset >> 1), ptr, linesize >> 1, h >> 1);
1007 }
1008
1009
1010 static inline void MPV_motion(MpegEncContext *s,
1011 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr,
1012 int dir, UINT8 **ref_picture,
1013 op_pixels_func *pix_op, qpel_mc_func *qpix_op)
1014 {
1015 int dxy, offset, mx, my, src_x, src_y, motion_x, motion_y;
1016 int mb_x, mb_y, i;
1017 UINT8 *ptr, *dest;
1018
1019 mb_x = s->mb_x;
1020 mb_y = s->mb_y;
1021
1022 switch(s->mv_type) {
1023 case MV_TYPE_16X16:
1024 if(s->mcsel){
1025 #if 0
1026 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
1027 ref_picture, 0,
1028 0, pix_op,
1029 s->sprite_offset[0][0]>>3,
1030 s->sprite_offset[0][1]>>3,
1031 16);
1032 #else
1033 gmc1_motion(s, dest_y, dest_cb, dest_cr, 0,
1034 ref_picture, 0,
1035 16);
1036 #endif
1037 }else if(s->quarter_sample && dir==0){ //FIXME
1038 qpel_motion(s, dest_y, dest_cb, dest_cr, 0,
1039 ref_picture, 0,
1040 0, pix_op, qpix_op,
1041 s->mv[dir][0][0], s->mv[dir][0][1], 16);
1042 }else{
1043 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
1044 ref_picture, 0,
1045 0, pix_op,
1046 s->mv[dir][0][0], s->mv[dir][0][1], 16);
1047 }
1048 break;
1049 case MV_TYPE_8X8:
1050 for(i=0;i<4;i++) {
1051 motion_x = s->mv[dir][i][0];
1052 motion_y = s->mv[dir][i][1];
1053
1054 dxy = ((motion_y & 1) << 1) | (motion_x & 1);
1055 src_x = mb_x * 16 + (motion_x >> 1) + (i & 1) * 8;
1056 src_y = mb_y * 16 + (motion_y >> 1) + (i >>1) * 8;
1057
1058 /* WARNING: do no forget half pels */
1059 src_x = clip(src_x, -16, s->width);
1060 if (src_x == s->width)
1061 dxy &= ~1;
1062 src_y = clip(src_y, -16, s->height);
1063 if (src_y == s->height)
1064 dxy &= ~2;
1065
1066 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x);
1067 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize;
1068 pix_op[dxy](dest, ptr, s->linesize, 8);
1069 }
1070
1071 if(s->flags&CODEC_FLAG_GRAY) break;
1072 /* In case of 8X8, we construct a single chroma motion vector
1073 with a special rounding */
1074 mx = 0;
1075 my = 0;
1076 for(i=0;i<4;i++) {
1077 mx += s->mv[dir][i][0];
1078 my += s->mv[dir][i][1];
1079 }
1080 if (mx >= 0)
1081 mx = (h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1));
1082 else {
1083 mx = -mx;
1084 mx = -(h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1));
1085 }
1086 if (my >= 0)
1087 my = (h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1));
1088 else {
1089 my = -my;
1090 my = -(h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1));
1091 }
1092 dxy = ((my & 1) << 1) | (mx & 1);
1093 mx >>= 1;
1094 my >>= 1;
1095
1096 src_x = mb_x * 8 + mx;
1097 src_y = mb_y * 8 + my;
1098 src_x = clip(src_x, -8, s->width/2);
1099 if (src_x == s->width/2)
1100 dxy &= ~1;
1101 src_y = clip(src_y, -8, s->height/2);
1102 if (src_y == s->height/2)
1103 dxy &= ~2;
1104
1105 offset = (src_y * (s->linesize >> 1)) + src_x;
1106 ptr = ref_picture[1] + offset;
1107 pix_op[dxy](dest_cb, ptr, s->linesize >> 1, 8);
1108 ptr = ref_picture[2] + offset;
1109 pix_op[dxy](dest_cr, ptr, s->linesize >> 1, 8);
1110 break;
1111 case MV_TYPE_FIELD:
1112 if (s->picture_structure == PICT_FRAME) {
1113 /* top field */
1114 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
1115 ref_picture, s->field_select[dir][0] ? s->linesize : 0,
1116 1, pix_op,
1117 s->mv[dir][0][0], s->mv[dir][0][1], 8);
1118 /* bottom field */
1119 mpeg_motion(s, dest_y, dest_cb, dest_cr, s->linesize,
1120 ref_picture, s->field_select[dir][1] ? s->linesize : 0,
1121 1, pix_op,
1122 s->mv[dir][1][0], s->mv[dir][1][1], 8);
1123 } else {
1124
1125
1126 }
1127 break;
1128 }
1129 }
1130
1131
1132 /* put block[] to dest[] */
1133 static inline void put_dct(MpegEncContext *s,
1134 DCTELEM *block, int i, UINT8 *dest, int line_size)
1135 {
1136 if (!s->mpeg2)
1137 s->dct_unquantize(s, block, i, s->qscale);
1138 ff_idct_put (dest, line_size, block);
1139 }
1140
1141 /* add block[] to dest[] */
1142 static inline void add_dct(MpegEncContext *s,
1143 DCTELEM *block, int i, UINT8 *dest, int line_size)
1144 {
1145 if (s->block_last_index[i] >= 0) {
1146 ff_idct_add (dest, line_size, block);
1147 }
1148 }
1149
1150 static inline void add_dequant_dct(MpegEncContext *s,
1151 DCTELEM *block, int i, UINT8 *dest, int line_size)
1152 {
1153 if (s->block_last_index[i] >= 0) {
1154 s->dct_unquantize(s, block, i, s->qscale);
1155
1156 ff_idct_add (dest, line_size, block);
1157 }
1158 }
1159
1160 /**
1161 * cleans dc, ac, coded_block for the current non intra MB
1162 */
1163 void ff_clean_intra_table_entries(MpegEncContext *s)
1164 {
1165 int wrap = s->block_wrap[0];
1166 int xy = s->block_index[0];
1167
1168 s->dc_val[0][xy ] =
1169 s->dc_val[0][xy + 1 ] =
1170 s->dc_val[0][xy + wrap] =
1171 s->dc_val[0][xy + 1 + wrap] = 1024;
1172 /* ac pred */
1173 memset(s->ac_val[0][xy ], 0, 32 * sizeof(INT16));
1174 memset(s->ac_val[0][xy + wrap], 0, 32 * sizeof(INT16));
1175 if (s->msmpeg4_version>=3) {
1176 s->coded_block[xy ] =
1177 s->coded_block[xy + 1 ] =
1178 s->coded_block[xy + wrap] =
1179 s->coded_block[xy + 1 + wrap] = 0;
1180 }
1181 /* chroma */
1182 wrap = s->block_wrap[4];
1183 xy = s->mb_x + 1 + (s->mb_y + 1) * wrap;
1184 s->dc_val[1][xy] =
1185 s->dc_val[2][xy] = 1024;
1186 /* ac pred */
1187 memset(s->ac_val[1][xy], 0, 16 * sizeof(INT16));
1188 memset(s->ac_val[2][xy], 0, 16 * sizeof(INT16));
1189
1190 s->mbintra_table[s->mb_x + s->mb_y*s->mb_width]= 0;
1191 }
1192
1193 /* generic function called after a macroblock has been parsed by the
1194 decoder or after it has been encoded by the encoder.
1195
1196 Important variables used:
1197 s->mb_intra : true if intra macroblock
1198 s->mv_dir : motion vector direction
1199 s->mv_type : motion vector type
1200 s->mv : motion vector
1201 s->interlaced_dct : true if interlaced dct used (mpeg2)
1202 */
1203 void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64])
1204 {
1205 int mb_x, mb_y;
1206 const int mb_xy = s->mb_y * s->mb_width + s->mb_x;
1207
1208 mb_x = s->mb_x;
1209 mb_y = s->mb_y;
1210
1211 #ifdef FF_POSTPROCESS
1212 /* Obsolete. Exists for compatibility with mplayer only. */
1213 quant_store[mb_y][mb_x]=s->qscale;
1214 //printf("[%02d][%02d] %d\n",mb_x,mb_y,s->qscale);
1215 #else
1216 if(s->avctx->quant_store) s->avctx->quant_store[mb_y*s->avctx->qstride+mb_x] = s->qscale;
1217 #endif
1218
1219 /* update DC predictors for P macroblocks */
1220 if (!s->mb_intra) {
1221 if (s->h263_pred || s->h263_aic) {
1222 if(s->mbintra_table[mb_xy])
1223 ff_clean_intra_table_entries(s);
1224 } else {
1225 s->last_dc[0] =
1226 s->last_dc[1] =
1227 s->last_dc[2] = 128 << s->intra_dc_precision;
1228 }
1229 }
1230 else if (s->h263_pred || s->h263_aic)
1231 s->mbintra_table[mb_xy]=1;
1232
1233 /* update motion predictor, not for B-frames as they need the motion_val from the last P/S-Frame */
1234 if (s->out_format == FMT_H263 && s->pict_type!=B_TYPE) { //FIXME move into h263.c if possible, format specific stuff shouldnt be here
1235 int motion_x, motion_y;
1236
1237 const int wrap = s->block_wrap[0];
1238 const int xy = s->block_index[0];
1239 if (s->mb_intra) {
1240 motion_x = 0;
1241 motion_y = 0;
1242 goto motion_init;
1243 } else if (s->mv_type == MV_TYPE_16X16) {
1244 motion_x = s->mv[0][0][0];
1245 motion_y = s->mv[0][0][1];
1246 motion_init:
1247 /* no update if 8X8 because it has been done during parsing */
1248 s->motion_val[xy][0] = motion_x;
1249 s->motion_val[xy][1] = motion_y;
1250 s->motion_val[xy + 1][0] = motion_x;
1251 s->motion_val[xy + 1][1] = motion_y;
1252 s->motion_val[xy + wrap][0] = motion_x;
1253 s->motion_val[xy + wrap][1] = motion_y;
1254 s->motion_val[xy + 1 + wrap][0] = motion_x;
1255 s->motion_val[xy + 1 + wrap][1] = motion_y;
1256 }
1257 }
1258
1259 if (!(s->encoding && (s->intra_only || s->pict_type==B_TYPE))) {
1260 UINT8 *dest_y, *dest_cb, *dest_cr;
1261 int dct_linesize, dct_offset;
1262 op_pixels_func *op_pix;
1263 qpel_mc_func *op_qpix;
1264
1265 /* avoid copy if macroblock skipped in last frame too
1266 dont touch it for B-frames as they need the skip info from the next p-frame */
1267 if (s->pict_type != B_TYPE) {
1268 UINT8 *mbskip_ptr = &s->mbskip_table[mb_xy];
1269 if (s->mb_skiped) {
1270 s->mb_skiped = 0;
1271 /* if previous was skipped too, then nothing to do !
1272 skip only during decoding as we might trash the buffers during encoding a bit */
1273 if (*mbskip_ptr != 0 && !s->encoding)
1274 goto the_end;
1275 *mbskip_ptr = 1; /* indicate that this time we skiped it */
1276 } else {
1277 *mbskip_ptr = 0; /* not skipped */
1278 }
1279 }
1280
1281 dest_y = s->current_picture[0] + (mb_y * 16 * s->linesize) + mb_x * 16;
1282 dest_cb = s->current_picture[1] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8;
1283 dest_cr = s->current_picture[2] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8;
1284
1285 if (s->interlaced_dct) {
1286 dct_linesize = s->linesize * 2;
1287 dct_offset = s->linesize;
1288 } else {
1289 dct_linesize = s->linesize;
1290 dct_offset = s->linesize * 8;
1291 }
1292
1293 if (!s->mb_intra) {
1294 /* motion handling */
1295 /* decoding or more than one mb_type (MC was allready done otherwise) */
1296 if((!s->encoding) || (s->mb_type[mb_xy]&(s->mb_type[mb_xy]-1))){
1297 if ((!s->no_rounding) || s->pict_type==B_TYPE){
1298 op_pix = put_pixels_tab;
1299 op_qpix= qpel_mc_rnd_tab;
1300 }else{
1301 op_pix = put_no_rnd_pixels_tab;
1302 op_qpix= qpel_mc_no_rnd_tab;
1303 }
1304
1305 if (s->mv_dir & MV_DIR_FORWARD) {
1306 MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture, op_pix, op_qpix);
1307 if ((!s->no_rounding) || s->pict_type==B_TYPE)
1308 op_pix = avg_pixels_tab;
1309 else
1310 op_pix = avg_no_rnd_pixels_tab;
1311 }
1312 if (s->mv_dir & MV_DIR_BACKWARD) {
1313 MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture, op_pix, op_qpix);
1314 }
1315 }
1316
1317 /* skip dequant / idct if we are really late ;) */
1318 if(s->hurry_up>1) goto the_end;
1319
1320 /* add dct residue */
1321 if(s->encoding || !(s->mpeg2 || s->h263_msmpeg4 || s->codec_id==CODEC_ID_MPEG4)){
1322 add_dequant_dct(s, block[0], 0, dest_y, dct_linesize);
1323 add_dequant_dct(s, block[1], 1, dest_y + 8, dct_linesize);
1324 add_dequant_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize);
1325 add_dequant_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize);
1326
1327 if(!(s->flags&CODEC_FLAG_GRAY)){
1328 add_dequant_dct(s, block[4], 4, dest_cb, s->linesize >> 1);
1329 add_dequant_dct(s, block[5], 5, dest_cr, s->linesize >> 1);
1330 }
1331 } else {
1332 add_dct(s, block[0], 0, dest_y, dct_linesize);
1333 add_dct(s, block[1], 1, dest_y + 8, dct_linesize);
1334 add_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize);
1335 add_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize);
1336
1337 if(!(s->flags&CODEC_FLAG_GRAY)){
1338 add_dct(s, block[4], 4, dest_cb, s->linesize >> 1);
1339 add_dct(s, block[5], 5, dest_cr, s->linesize >> 1);
1340 }
1341 }
1342 } else {
1343 /* dct only in intra block */
1344 put_dct(s, block[0], 0, dest_y, dct_linesize);
1345 put_dct(s, block[1], 1, dest_y + 8, dct_linesize);
1346 put_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize);
1347 put_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize);
1348
1349 if(!(s->flags&CODEC_FLAG_GRAY)){
1350 put_dct(s, block[4], 4, dest_cb, s->linesize >> 1);
1351 put_dct(s, block[5], 5, dest_cr, s->linesize >> 1);
1352 }
1353 }
1354 }
1355 the_end:
1356 emms_c(); //FIXME remove
1357 }
1358
1359 static inline void dct_single_coeff_elimination(MpegEncContext *s, int n, int threshold, int skip_dc)
1360 {
1361 static const char tab[64]=
1362 {3,2,2,1,1,1,1,1,
1363 1,1,1,1,1,1,1,1,
1364 1,1,1,1,1,1,1,1,
1365 0,0,0,0,0,0,0,0,
1366 0,0,0,0,0,0,0,0,
1367 0,0,0,0,0,0,0,0,
1368 0,0,0,0,0,0,0,0,
1369 0,0,0,0,0,0,0,0};
1370 int score=0;
1371 int run=0;
1372 int i;
1373 DCTELEM *block= s->block[n];
1374 const int last_index= s->block_last_index[n];
1375
1376 if(skip_dc) skip_dc=1;
1377
1378 /* are all which we could set to zero are allready zero? */
1379 if(last_index<=skip_dc - 1) return;
1380
1381 for(i=0; i<=last_index; i++){
1382 const int j = zigzag_direct[i];
1383 const int level = ABS(block[j]);
1384 if(level==1){
1385 if(skip_dc && i==0) continue;
1386 score+= tab[run];
1387 run=0;
1388 }else if(level>1){
1389 return;
1390 }else{
1391 run++;
1392 }
1393 }
1394 if(score >= threshold) return;
1395 for(i=skip_dc; i<=last_index; i++){
1396 const int j = zigzag_direct[i];
1397 block[j]=0;
1398 }
1399 if(block[0]) s->block_last_index[n]= 0;
1400 else s->block_last_index[n]= -1;
1401 }
1402
1403 static inline void clip_coeffs(MpegEncContext *s, DCTELEM *block, int last_index)
1404 {
1405 int i;
1406 const int maxlevel= s->max_qcoeff;
1407 const int minlevel= s->min_qcoeff;
1408
1409 for(i=0;i<=last_index; i++){
1410 const int j = zigzag_direct[i];
1411 int level = block[j];
1412
1413 if (level>maxlevel) level=maxlevel;
1414 else if(level<minlevel) level=minlevel;
1415 block[j]= level;
1416 }
1417 }
1418
1419 static void encode_mb(MpegEncContext *s, int motion_x, int motion_y)
1420 {
1421 const int mb_x= s->mb_x;
1422 const int mb_y= s->mb_y;
1423 int i;
1424 int skip_dct[6];
1425 #if 0
1426 if (s->interlaced_dct) {
1427 dct_linesize = s->linesize * 2;
1428 dct_offset = s->linesize;
1429 } else {
1430 dct_linesize = s->linesize;
1431 dct_offset = s->linesize * 8;
1432 }
1433 #endif
1434 for(i=0; i<6; i++) skip_dct[i]=0;
1435
1436 if (s->mb_intra) {
1437 UINT8 *ptr;
1438 int wrap;
1439
1440 wrap = s->linesize;
1441 ptr = s->new_picture[0] + (mb_y * 16 * wrap) + mb_x * 16;
1442 get_pixels(s->block[0], ptr , wrap);
1443 get_pixels(s->block[1], ptr + 8, wrap);
1444 get_pixels(s->block[2], ptr + 8 * wrap , wrap);
1445 get_pixels(s->block[3], ptr + 8 * wrap + 8, wrap);
1446
1447 if(s->flags&CODEC_FLAG_GRAY){
1448 skip_dct[4]= 1;
1449 skip_dct[5]= 1;
1450 }else{
1451 wrap >>=1;
1452 ptr = s->new_picture[1] + (mb_y * 8 * wrap) + mb_x * 8;
1453 get_pixels(s->block[4], ptr, wrap);
1454
1455 ptr = s->new_picture[2] + (mb_y * 8 * wrap) + mb_x * 8;
1456 get_pixels(s->block[5], ptr, wrap);
1457 }
1458 }else{
1459 op_pixels_func *op_pix;
1460 qpel_mc_func *op_qpix;
1461 UINT8 *dest_y, *dest_cb, *dest_cr;
1462 UINT8 *ptr_y, *ptr_cb, *ptr_cr;
1463 int wrap_y, wrap_c;
1464
1465 dest_y = s->current_picture[0] + (mb_y * 16 * s->linesize ) + mb_x * 16;
1466 dest_cb = s->current_picture[1] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8;
1467 dest_cr = s->current_picture[2] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8;
1468 wrap_y = s->linesize;
1469 wrap_c = wrap_y>>1;
1470 ptr_y = s->new_picture[0] + (mb_y * 16 * wrap_y) + mb_x * 16;
1471 ptr_cb = s->new_picture[1] + (mb_y * 8 * wrap_c) + mb_x * 8;
1472 ptr_cr = s->new_picture[2] + (mb_y * 8 * wrap_c) + mb_x * 8;
1473
1474 if ((!s->no_rounding) || s->pict_type==B_TYPE){
1475 op_pix = put_pixels_tab;
1476 op_qpix= qpel_mc_rnd_tab;
1477 }else{
1478 op_pix = put_no_rnd_pixels_tab;
1479 op_qpix= qpel_mc_no_rnd_tab;
1480 }
1481
1482 if (s->mv_dir & MV_DIR_FORWARD) {
1483 MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture, op_pix, op_qpix);
1484 if ((!s->no_rounding) || s->pict_type==B_TYPE)
1485 op_pix = avg_pixels_tab;
1486 else
1487 op_pix = avg_no_rnd_pixels_tab;
1488 }
1489 if (s->mv_dir & MV_DIR_BACKWARD) {
1490 MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture, op_pix, op_qpix);
1491 }
1492
1493 diff_pixels(s->block[0], ptr_y , dest_y , wrap_y);
1494 diff_pixels(s->block[1], ptr_y + 8, dest_y + 8, wrap_y);
1495 diff_pixels(s->block[2], ptr_y + 8 * wrap_y , dest_y + 8 * wrap_y , wrap_y);
1496 diff_pixels(s->block[3], ptr_y + 8 * wrap_y + 8, dest_y + 8 * wrap_y + 8, wrap_y);
1497
1498 if(s->flags&CODEC_FLAG_GRAY){
1499 skip_dct[4]= 1;
1500 skip_dct[5]= 1;
1501 }else{
1502 diff_pixels(s->block[4], ptr_cb, dest_cb, wrap_c);
1503 diff_pixels(s->block[5], ptr_cr, dest_cr, wrap_c);
1504 }
1505
1506 /* pre quantization */
1507 if(s->mc_mb_var[s->mb_width*mb_y+ mb_x]<2*s->qscale*s->qscale){
1508 if(pix_abs8x8(ptr_y , dest_y , wrap_y) < 20*s->qscale) skip_dct[0]= 1;
1509 if(pix_abs8x8(ptr_y + 8, dest_y + 8, wrap_y) < 20*s->qscale) skip_dct[1]= 1;
1510 if(pix_abs8x8(ptr_y + 8*wrap_y , dest_y + 8*wrap_y , wrap_y) < 20*s->qscale) skip_dct[2]= 1;
1511 if(pix_abs8x8(ptr_y + 8*wrap_y + 8, dest_y + 8*wrap_y + 8, wrap_y) < 20*s->qscale) skip_dct[3]= 1;
1512 if(pix_abs8x8(ptr_cb , dest_cb , wrap_y) < 20*s->qscale) skip_dct[4]= 1;
1513 if(pix_abs8x8(ptr_cr , dest_cr , wrap_y) < 20*s->qscale) skip_dct[5]= 1;
1514 #if 0
1515 {
1516 static int stat[7];
1517 int num=0;
1518 for(i=0; i<6; i++)
1519 if(skip_dct[i]) num++;
1520 stat[num]++;
1521
1522 if(s->mb_x==0 && s->mb_y==0){
1523 for(i=0; i<7; i++){
1524 printf("%6d %1d\n", stat[i], i);
1525 }
1526 }
1527 }
1528 #endif
1529 }
1530
1531 }
1532
1533 #if 0
1534 {
1535 float adap_parm;
1536
1537 adap_parm = ((s->avg_mb_var << 1) + s->mb_var[s->mb_width*mb_y+mb_x] + 1.0) /
1538 ((s->mb_var[s->mb_width*mb_y+mb_x] << 1) + s->avg_mb_var + 1.0);
1539
1540 printf("\ntype=%c qscale=%2d adap=%0.2f dquant=%4.2f var=%4d avgvar=%4d",
1541 (s->mb_type[s->mb_width*mb_y+mb_x] > 0) ? 'I' : 'P',
1542 s->qscale, adap_parm, s->qscale*adap_parm,
1543 s->mb_var[s->mb_width*mb_y+mb_x], s->avg_mb_var);
1544 }
1545 #endif
1546 /* DCT & quantize */
1547 if(s->out_format==FMT_MJPEG){
1548 for(i=0;i<6;i++) {
1549 int overflow;
1550 s->block_last_index[i] = dct_quantize(s, s->block[i], i, 8, &overflow);
1551 if (overflow) clip_coeffs(s, s->block[i], s->block_last_index[i]);
1552 }
1553 }else{
1554 for(i=0;i<6;i++) {
1555 if(!skip_dct[i]){
1556 int overflow;
1557 s->block_last_index[i] = dct_quantize(s, s->block[i], i, s->qscale, &overflow);
1558 // FIXME we could decide to change to quantizer instead of clipping
1559 // JS: I don't think that would be a good idea it could lower quality instead
1560 // of improve it. Just INTRADC clipping deserves changes in quantizer
1561 if (overflow) clip_coeffs(s, s->block[i], s->block_last_index[i]);
1562 }else
1563 s->block_last_index[i]= -1;
1564 }
1565 if(s->luma_elim_threshold && !s->mb_intra)
1566 for(i=0; i<4; i++)
1567 dct_single_coeff_elimination(s, i, s->luma_elim_threshold, 0);
1568 if(s->chroma_elim_threshold && !s->mb_intra)
1569 for(i=4; i<6; i++)
1570 dct_single_coeff_elimination(s, i, s->chroma_elim_threshold, 1);
1571 }
1572
1573 if((s->flags&CODEC_FLAG_GRAY) && s->mb_intra){
1574 s->block_last_index[4]=
1575 s->block_last_index[5]= 0;
1576 s->block[4][0]=
1577 s->block[5][0]= 128;
1578 }
1579
1580 /* huffman encode */
1581 switch(s->out_format) {
1582 case FMT_MPEG1:
1583 mpeg1_encode_mb(s, s->block, motion_x, motion_y);
1584 break;
1585 case FMT_H263:
1586 if (s->h263_msmpeg4)
1587 msmpeg4_encode_mb(s, s->block, motion_x, motion_y);
1588 else if(s->h263_pred)
1589 mpeg4_encode_mb(s, s->block, motion_x, motion_y);
1590 else
1591 h263_encode_mb(s, s->block, motion_x, motion_y);
1592 break;
1593 case FMT_MJPEG:
1594 mjpeg_encode_mb(s, s->block);
1595 break;
1596 }
1597 }
1598
1599 void ff_copy_bits(PutBitContext *pb, UINT8 *src, int length)
1600 {
1601 int bytes= length>>4;
1602 int bits= length&15;
1603 int i;
1604
1605 if(length==0) return;
1606
1607 for(i=0; i<bytes; i++) put_bits(pb, 16, be2me_16(((uint16_t*)src)[i]));
1608 put_bits(pb, bits, be2me_16(((uint16_t*)src)[i])>>(16-bits));
1609 }
1610
1611 static inline void copy_context_before_encode(MpegEncContext *d, MpegEncContext *s, int type){
1612 int i;
1613
1614 memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
1615
1616 /* mpeg1 */
1617 d->mb_incr= s->mb_incr;
1618 for(i=0; i<3; i++)
1619 d->last_dc[i]= s->last_dc[i];
1620
1621 /* statistics */
1622 d->mv_bits= s->mv_bits;
1623 d->i_tex_bits= s->i_tex_bits;
1624 d->p_tex_bits= s->p_tex_bits;
1625 d->i_count= s->i_count;
1626 d->p_count= s->p_count;
1627 d->skip_count= s->skip_count;
1628 d->misc_bits= s->misc_bits;
1629 d->last_bits= 0;
1630
1631 d->mb_skiped= s->mb_skiped;
1632 }
1633
1634 static inline void copy_context_after_encode(MpegEncContext *d, MpegEncContext *s, int type){
1635 int i;
1636
1637 memcpy(d->mv, s->mv, 2*4*2*sizeof(int));
1638 memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
1639
1640 /* mpeg1 */
1641 d->mb_incr= s->mb_incr;
1642 for(i=0; i<3; i++)
1643 d->last_dc[i]= s->last_dc[i];
1644
1645 /* statistics */
1646 d->mv_bits= s->mv_bits;
1647 d->i_tex_bits= s->i_tex_bits;
1648 d->p_tex_bits= s->p_tex_bits;
1649 d->i_count= s->i_count;
1650 d->p_count= s->p_count;
1651 d->skip_count= s->skip_count;
1652 d->misc_bits= s->misc_bits;
1653
1654 d->mb_intra= s->mb_intra;
1655 d->mb_skiped= s->mb_skiped;
1656 d->mv_type= s->mv_type;
1657 d->mv_dir= s->mv_dir;
1658 d->pb= s->pb;
1659 if(s->data_partitioning){
1660 d->pb2= s->pb2;
1661 d->tex_pb= s->tex_pb;
1662 }
1663 d->block= s->block;
1664 for(i=0; i<6; i++)
1665 d->block_last_index[i]= s->block_last_index[i];
1666 }
1667
1668 static inline void encode_mb_hq(MpegEncContext *s, MpegEncContext *backup, MpegEncContext *best, int type,
1669 PutBitContext pb[2], PutBitContext pb2[2], PutBitContext tex_pb[2],
1670 int *dmin, int *next_block, int motion_x, int motion_y)
1671 {
1672 int bits_count;
1673
1674 copy_context_before_encode(s, backup, type);
1675
1676 s->block= s->blocks[*next_block];
1677 s->pb= pb[*next_block];
1678 if(s->data_partitioning){
1679 s->pb2 = pb2 [*next_block];
1680 s->tex_pb= tex_pb[*next_block];
1681 }
1682
1683 encode_mb(s, motion_x, motion_y);
1684
1685 bits_count= get_bit_count(&s->pb);
1686 if(s->data_partitioning){
1687 bits_count+= get_bit_count(&s->pb2);
1688 bits_count+= get_bit_count(&s->tex_pb);
1689 }
1690
1691 if(bits_count<*dmin){
1692 *dmin= bits_count;
1693 *next_block^=1;
1694
1695 copy_context_after_encode(best, s, type);
1696 }
1697 }
1698
1699 static void encode_picture(MpegEncContext *s, int picture_number)
1700 {
1701 int mb_x, mb_y, last_gob, pdif = 0;
1702 int i;
1703 int bits;
1704 MpegEncContext best_s, backup_s;
1705 UINT8 bit_buf[2][3000];
1706 UINT8 bit_buf2[2][3000];
1707 UINT8 bit_buf_tex[2][3000];
1708 PutBitContext pb[2], pb2[2], tex_pb[2];
1709
1710 for(i=0; i<2; i++){
1711 init_put_bits(&pb [i], bit_buf [i], 3000, NULL, NULL);
1712 init_put_bits(&pb2 [i], bit_buf2 [i], 3000, NULL, NULL);
1713 init_put_bits(&tex_pb[i], bit_buf_tex[i], 3000, NULL, NULL);
1714 }
1715
1716 s->picture_number = picture_number;
1717
1718 s->block_wrap[0]=
1719 s->block_wrap[1]=
1720 s->block_wrap[2]=
1721 s->block_wrap[3]= s->mb_width*2 + 2;
1722 s->block_wrap[4]=
1723 s->block_wrap[5]= s->mb_width + 2;
1724
1725 /* Reset the average MB variance */
1726 s->mb_var_sum = 0;
1727 s->mc_mb_var_sum = 0;
1728
1729 /* we need to initialize some time vars before we can encode b-frames */
1730 if (s->h263_pred && !s->h263_msmpeg4)
1731 ff_set_mpeg4_time(s, s->picture_number);
1732
1733 /* Estimate motion for every MB */
1734 if(s->pict_type != I_TYPE){
1735 for(mb_y=0; mb_y < s->mb_height; mb_y++) {
1736 s->block_index[0]= s->block_wrap[0]*(mb_y*2 + 1) - 1;
1737 s->block_index[1]= s->block_wrap[0]*(mb_y*2 + 1);
1738 s->block_index[2]= s->block_wrap[0]*(mb_y*2 + 2) - 1;
1739 s->block_index[3]= s->block_wrap[0]*(mb_y*2 + 2);
1740 for(mb_x=0; mb_x < s->mb_width; mb_x++) {
1741 s->mb_x = mb_x;
1742 s->mb_y = mb_y;
1743 s->block_index[0]+=2;
1744 s->block_index[1]+=2;
1745 s->block_index[2]+=2;
1746 s->block_index[3]+=2;
1747
1748 /* compute motion vector & mb_type and store in context */
1749 if(s->pict_type==B_TYPE)
1750 ff_estimate_b_frame_motion(s, mb_x, mb_y);
1751 else
1752 ff_estimate_p_frame_motion(s, mb_x, mb_y);
1753 // s->mb_type[mb_y*s->mb_width + mb_x]=MB_TYPE_INTER;
1754 }
1755 }
1756 emms_c();
1757 }else /* if(s->pict_type == I_TYPE) */{
1758 /* I-Frame */
1759 //FIXME do we need to zero them?
1760 memset(s->motion_val[0], 0, sizeof(INT16)*(s->mb_width*2 + 2)*(s->mb_height*2 + 2)*2);
1761 memset(s->p_mv_table , 0, sizeof(INT16)*(s->mb_width+2)*(s->mb_height+2)*2);
1762 memset(s->mb_type , MB_TYPE_INTRA, sizeof(UINT8)*s->mb_width*s->mb_height);
1763 }
1764
1765 if(s->mb_var_sum < s->mc_mb_var_sum && s->pict_type == P_TYPE){ //FIXME subtract MV bits
1766 s->pict_type= I_TYPE;
1767 memset(s->mb_type , MB_TYPE_INTRA, sizeof(UINT8)*s->mb_width*s->mb_height);
1768 if(s->max_b_frames==0){
1769 s->input_pict_type= I_TYPE;
1770 s->input_picture_in_gop_number=0;
1771 }
1772 //printf("Scene change detected, encoding as I Frame\n");
1773 }
1774
1775 if(s->pict_type==P_TYPE || s->pict_type==S_TYPE)
1776 s->f_code= ff_get_best_fcode(s, s->p_mv_table, MB_TYPE_INTER);
1777 ff_fix_long_p_mvs(s);
1778 if(s->pict_type==B_TYPE){
1779 s->f_code= ff_get_best_fcode(s, s->b_forw_mv_table, MB_TYPE_FORWARD);
1780 s->b_code= ff_get_best_fcode(s, s->b_back_mv_table, MB_TYPE_BACKWARD);
1781
1782 ff_fix_long_b_mvs(s, s->b_forw_mv_table, s->f_code, MB_TYPE_FORWARD);
1783 ff_fix_long_b_mvs(s, s->b_back_mv_table, s->b_code, MB_TYPE_BACKWARD);
1784 ff_fix_long_b_mvs(s, s->b_bidir_forw_mv_table, s->f_code, MB_TYPE_BIDIR);
1785 ff_fix_long_b_mvs(s, s->b_bidir_back_mv_table, s->b_code, MB_TYPE_BIDIR);
1786 }
1787
1788 //printf("f_code %d ///\n", s->f_code);
1789
1790 // printf("%d %d\n", s->avg_mb_var, s->mc_mb_var);
1791
1792 if(s->flags&CODEC_FLAG_PASS2)
1793 s->qscale = ff_rate_estimate_qscale_pass2(s);
1794 else if (!s->fixed_qscale)
1795 s->qscale = ff_rate_estimate_qscale(s);
1796
1797 if (s->out_format == FMT_MJPEG) {
1798 /* for mjpeg, we do include qscale in the matrix */
1799 s->intra_matrix[0] = ff_mpeg1_default_intra_matrix[0];
1800 for(i=1;i<64;i++)
1801 s->intra_matrix[i] = CLAMP_TO_8BIT((ff_mpeg1_default_intra_matrix[i] * s->qscale) >> 3);
1802 convert_matrix(s->q_intra_matrix, s->q_intra_matrix16,
1803 s->q_intra_matrix16_bias, s->intra_matrix, s->intra_quant_bias);
1804 }
1805
1806 s->last_bits= get_bit_count(&s->pb);
1807 switch(s->out_format) {
1808 case FMT_MJPEG:
1809 mjpeg_picture_header(s);
1810 break;
1811 case FMT_H263:
1812 if (s->h263_msmpeg4)
1813 msmpeg4_encode_picture_header(s, picture_number);
1814 else if (s->h263_pred)
1815 mpeg4_encode_picture_header(s, picture_number);
1816 else if (s->h263_rv10)
1817 rv10_encode_picture_header(s, picture_number);
1818 else
1819 h263_encode_picture_header(s, picture_number);
1820 break;
1821 case FMT_MPEG1:
1822 mpeg1_encode_picture_header(s, picture_number);
1823 break;
1824 }
1825 bits= get_bit_count(&s->pb);
1826 s->header_bits= bits - s->last_bits;
1827 s->last_bits= bits;
1828 s->mv_bits=0;
1829 s->misc_bits=0;
1830 s->i_tex_bits=0;
1831 s->p_tex_bits=0;
1832 s->i_count=0;
1833 s->p_count=0;
1834 s->skip_count=0;
1835
1836 /* init last dc values */
1837 /* note: quant matrix value (8) is implied here */
1838 s->last_dc[0] = 128;
1839 s->last_dc[1] = 128;
1840 s->last_dc[2] = 128;
1841 s->mb_incr = 1;
1842 s->last_mv[0][0][0] = 0;
1843 s->last_mv[0][0][1] = 0;
1844
1845 /* Get the GOB height based on picture height */
1846 if (s->out_format == FMT_H263 && !s->h263_pred && !s->h263_msmpeg4) {
1847 if (s->height <= 400)
1848 s->gob_index = 1;
1849 else if (s->height <= 800)
1850 s->gob_index = 2;
1851 else
1852 s->gob_index = 4;
1853 }else if(s->codec_id==CODEC_ID_MPEG4){
1854 s->gob_index = 1;
1855 }
1856
1857 if(s->codec_id==CODEC_ID_MPEG4 && s->data_partitioning && s->pict_type!=B_TYPE)
1858 ff_mpeg4_init_partitions(s);
1859
1860 s->resync_mb_x=0;
1861 s->resync_mb_y=0;
1862 for(mb_y=0; mb_y < s->mb_height; mb_y++) {
1863 /* Put GOB header based on RTP MTU for formats which support it per line (H263*)*/
1864 /* TODO: Put all this stuff in a separate generic function */
1865 if (s->rtp_mode) {
1866 if (!mb_y) {
1867 s->ptr_lastgob = s->pb.buf;
1868 s->ptr_last_mb_line = s->pb.buf;
1869 } else if (s->out_format == FMT_H263 && !s->h263_pred && !s->h263_msmpeg4 && !(mb_y % s->gob_index)) {
1870 // MN: we could move the space check from h263 -> here, as its not h263 specific
1871 last_gob = h263_encode_gob_header(s, mb_y);
1872 if (last_gob) {
1873 s->first_slice_line = 1;
1874 }else{
1875 /*MN: we reset it here instead at the end of each line cuz mpeg4 can have
1876 slice lines starting & ending in the middle*/
1877 s->first_slice_line = 0;
1878 }
1879 }
1880 }
1881
1882 s->y_dc_scale= s->y_dc_scale_table[ s->qscale ];
1883 s->c_dc_scale= s->c_dc_scale_table[ s->qscale ];
1884
1885 s->block_index[0]= s->block_wrap[0]*(mb_y*2 + 1) - 1;
1886 s->block_index[1]= s->block_wrap[0]*(mb_y*2 + 1);
1887 s->block_index[2]= s->block_wrap[0]*(mb_y*2 + 2) - 1;
1888 s->block_index[3]= s->block_wrap[0]*(mb_y*2 + 2);
1889 s->block_index[4]= s->block_wrap[4]*(mb_y + 1) + s->block_wrap[0]*(s->mb_height*2 + 2);
1890 s->block_index[5]= s->block_wrap[4]*(mb_y + 1 + s->mb_height + 2) + s->block_wrap[0]*(s->mb_height*2 + 2);
1891 for(mb_x=0; mb_x < s->mb_width; mb_x++) {
1892 const int mb_type= s->mb_type[mb_y * s->mb_width + mb_x];
1893 const int xy= (mb_y+1) * (s->mb_width+2) + mb_x + 1;
1894 // int d;
1895 int dmin=10000000;
1896
1897 s->mb_x = mb_x;
1898 s->mb_y = mb_y;
1899 s->block_index[0]+=2;
1900 s->block_index[1]+=2;
1901 s->block_index[2]+=2;
1902 s->block_index[3]+=2;
1903 s->block_index[4]++;
1904 s->block_index[5]++;
1905
1906 /* write gob / video packet header for formats which support it at any MB (MPEG4) */
1907 if(s->rtp_mode && s->mb_y>0 && s->codec_id==CODEC_ID_MPEG4){
1908 int pdif= pbBufPtr(&s->pb) - s->ptr_lastgob;
1909
1910 //the *2 is there so we stay below the requested size
1911 if(pdif + s->mb_line_avgsize/s->mb_width >= s->rtp_payload_size){
1912 if(s->codec_id==CODEC_ID_MPEG4){
1913 if(s->data_partitioning && s->pict_type!=B_TYPE){
1914 ff_mpeg4_merge_partitions(s);
1915 ff_mpeg4_init_partitions(s);
1916 }
1917 ff_mpeg4_encode_video_packet_header(s);
1918
1919 if(s->flags&CODEC_FLAG_PASS1){
1920 int bits= get_bit_count(&s->pb);
1921 s->misc_bits+= bits - s->last_bits;
1922 s->last_bits= bits;
1923 }
1924 ff_mpeg4_clean_buffers(s);
1925 }
1926 s->ptr_lastgob = pbBufPtr(&s->pb);
1927 s->first_slice_line=1;
1928 s->resync_mb_x=mb_x;
1929 s->resync_mb_y=mb_y;
1930 }
1931
1932 if( (s->resync_mb_x == s->mb_x)
1933 && s->resync_mb_y+1 == s->mb_y){
1934 s->first_slice_line=0;
1935 }
1936 }
1937
1938 if(mb_type & (mb_type-1)){ // more than 1 MB type possible
1939 int next_block=0;
1940 int pb_bits_count, pb2_bits_count, tex_pb_bits_count;
1941
1942 copy_context_before_encode(&backup_s, s, -1);
1943 backup_s.pb= s->pb;
1944 best_s.data_partitioning= s->data_partitioning;
1945 if(s->data_partitioning){
1946 backup_s.pb2= s->pb2;
1947 backup_s.tex_pb= s->tex_pb;
1948 }
1949
1950 if(mb_type&MB_TYPE_INTER){
1951 s->mv_dir = MV_DIR_FORWARD;
1952 s->mv_type = MV_TYPE_16X16;
1953 s->mb_intra= 0;
1954 s->mv[0][0][0] = s->p_mv_table[xy][0];
1955 s->mv[0][0][1] = s->p_mv_table[xy][1];
1956 encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_INTER, pb, pb2, tex_pb,
1957 &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
1958 }
1959 if(mb_type&MB_TYPE_INTER4V){
1960 s->mv_dir = MV_DIR_FORWARD;
1961 s->mv_type = MV_TYPE_8X8;
1962 s->mb_intra= 0;
1963 for(i=0; i<4; i++){
1964 s->mv[0][i][0] = s->motion_val[s->block_index[i]][0];
1965 s->mv[0][i][1] = s->motion_val[s->block_index[i]][1];
1966 }
1967 encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_INTER4V, pb, pb2, tex_pb,
1968 &dmin, &next_block, 0, 0);
1969 }
1970 if(mb_type&MB_TYPE_FORWARD){
1971 s->mv_dir = MV_DIR_FORWARD;
1972 s->mv_type = MV_TYPE_16X16;
1973 s->mb_intra= 0;
1974 s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
1975 s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
1976 encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_FORWARD, pb, pb2, tex_pb,
1977 &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
1978 }
1979 if(mb_type&MB_TYPE_BACKWARD){
1980 s->mv_dir = MV_DIR_BACKWARD;
1981 s->mv_type = MV_TYPE_16X16;
1982 s->mb_intra= 0;
1983 s->mv[1][0][0] = s->b_back_mv_table[xy][0];
1984 s->mv[1][0][1] = s->b_back_mv_table[xy][1];
1985 encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_BACKWARD, pb, pb2, tex_pb,
1986 &dmin, &next_block, s->mv[1][0][0], s->mv[1][0][1]);
1987 }
1988 if(mb_type&MB_TYPE_BIDIR){
1989 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
1990 s->mv_type = MV_TYPE_16X16;
1991 s->mb_intra= 0;
1992 s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
1993 s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
1994 s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
1995 s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
1996 encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_BIDIR, pb, pb2, tex_pb,
1997 &dmin, &next_block, 0, 0);
1998 }
1999 if(mb_type&MB_TYPE_DIRECT){
2000 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2001 s->mv_type = MV_TYPE_16X16; //FIXME
2002 s->mb_intra= 0;
2003 s->mv[0][0][0] = s->b_direct_forw_mv_table[xy][0];
2004 s->mv[0][0][1] = s->b_direct_forw_mv_table[xy][1];
2005 s->mv[1][0][0] = s->b_direct_back_mv_table[xy][0];
2006 s->mv[1][0][1] = s->b_direct_back_mv_table[xy][1];
2007 encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_DIRECT, pb, pb2, tex_pb,
2008 &dmin, &next_block, s->b_direct_mv_table[xy][0], s->b_direct_mv_table[xy][1]);
2009 }
2010 if(mb_type&MB_TYPE_INTRA){
2011 s->mv_dir = MV_DIR_FORWARD;
2012 s->mv_type = MV_TYPE_16X16;
2013 s->mb_intra= 1;
2014 s->mv[0][0][0] = 0;
2015 s->mv[0][0][1] = 0;
2016 encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_INTRA, pb, pb2, tex_pb,
2017 &dmin, &next_block, 0, 0);
2018 /* force cleaning of ac/dc pred stuff if needed ... */
2019 if(s->h263_pred || s->h263_aic)
2020 s->mbintra_table[mb_x + mb_y*s->mb_width]=1;
2021 }
2022 copy_context_after_encode(s, &best_s, -1);
2023
2024 pb_bits_count= get_bit_count(&s->pb);
2025 flush_put_bits(&s->pb);
2026 ff_copy_bits(&backup_s.pb, bit_buf[next_block^1], pb_bits_count);
2027 s->pb= backup_s.pb;
2028
2029 if(s->data_partitioning){
2030 pb2_bits_count= get_bit_count(&s->pb2);
2031 flush_put_bits(&s->pb2);
2032 ff_copy_bits(&backup_s.pb2, bit_buf2[next_block^1], pb2_bits_count);
2033 s->pb2= backup_s.pb2;
2034
2035 tex_pb_bits_count= get_bit_count(&s->tex_pb);
2036 flush_put_bits(&s->tex_pb);
2037 ff_copy_bits(&backup_s.tex_pb, bit_buf_tex[next_block^1], tex_pb_bits_count);
2038 s->tex_pb= backup_s.tex_pb;
2039 }
2040 s->last_bits= get_bit_count(&s->pb);
2041 } else {
2042 int motion_x, motion_y;
2043 s->mv_type=MV_TYPE_16X16;
2044 // only one MB-Type possible
2045 switch(mb_type){
2046 case MB_TYPE_INTRA:
2047 s->mv_dir = MV_DIR_FORWARD;
2048 s->mb_intra= 1;
2049 motion_x= s->mv[0][0][0] = 0;
2050 motion_y= s->mv[0][0][1] = 0;
2051 break;
2052 case MB_TYPE_INTER:
2053 s->mv_dir = MV_DIR_FORWARD;
2054 s->mb_intra= 0;
2055 motion_x= s->mv[0][0][0] = s->p_mv_table[xy][0];
2056 motion_y= s->mv[0][0][1] = s->p_mv_table[xy][1];
2057 break;
2058 case MB_TYPE_INTER4V:
2059 s->mv_dir = MV_DIR_FORWARD;
2060 s->mv_type = MV_TYPE_8X8;
2061 s->mb_intra= 0;
2062 for(i=0; i<4; i++){
2063 s->mv[0][i][0] = s->motion_val[s->block_index[i]][0];
2064 s->mv[0][i][1] = s->motion_val[s->block_index[i]][1];
2065 }
2066 motion_x= motion_y= 0;
2067 break;
2068 case MB_TYPE_DIRECT:
2069 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2070 s->mb_intra= 0;
2071 motion_x=s->b_direct_mv_table[xy][0];
2072 motion_y=s->b_direct_mv_table[xy][1];
2073 s->mv[0][0][0] = s->b_direct_forw_mv_table[xy][0];
2074 s->mv[0][0][1] = s->b_direct_forw_mv_table[xy][1];
2075 s->mv[1][0][0] = s->b_direct_back_mv_table[xy][0];
2076 s->mv[1][0][1] = s->b_direct_back_mv_table[xy][1];
2077 break;
2078 case MB_TYPE_BIDIR:
2079 s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2080 s->mb_intra= 0;
2081 motion_x=0;
2082 motion_y=0;
2083 s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
2084 s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
2085 s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
2086 s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2087 break;
2088 case MB_TYPE_BACKWARD:
2089 s->mv_dir = MV_DIR_BACKWARD;
2090 s->mb_intra= 0;
2091 motion_x= s->mv[1][0][0] = s->b_back_mv_table[xy][0];
2092 motion_y= s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2093 break;
2094 case MB_TYPE_FORWARD:
2095 s->mv_dir = MV_DIR_FORWARD;
2096 s->mb_intra= 0;
2097 motion_x= s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
2098 motion_y= s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2099 // printf(" %d %d ", motion_x, motion_y);
2100 break;
2101 default:
2102 motion_x=motion_y=0; //gcc warning fix
2103 printf("illegal MB type\n");
2104 }
2105 encode_mb(s, motion_x, motion_y);
2106 }
2107 /* clean the MV table in IPS frames for direct mode in B frames */
2108 if(s->mb_intra /* && I,P,S_TYPE */){
2109 s->p_mv_table[xy][0]=0;
2110 s->p_mv_table[xy][1]=0;
2111 }
2112
2113 MPV_decode_mb(s, s->block);
2114 //printf("MB %d %d bits\n", s->mb_x+s->mb_y*s->mb_width, get_bit_count(&s->pb));
2115 }
2116
2117
2118 /* Obtain average GOB size for RTP */
2119 if (s->rtp_mode) {
2120 if (!mb_y)
2121 s->mb_line_avgsize = pbBufPtr(&s->pb) - s->ptr_last_mb_line;
2122 else if (!(mb_y % s->gob_index)) {
2123 s->mb_line_avgsize = (s->mb_line_avgsize + pbBufPtr(&s->pb) - s->ptr_last_mb_line) >> 1;
2124 s->ptr_last_mb_line = pbBufPtr(&s->pb);
2125 }
2126 //fprintf(stderr, "\nMB line: %d\tSize: %u\tAvg. Size: %u", s->mb_y,
2127 // (s->pb.buf_ptr - s->ptr_last_mb_line), s->mb_line_avgsize);
2128 if(s->codec_id!=CODEC_ID_MPEG4) s->first_slice_line = 0; //FIXME clean
2129 }
2130 }
2131 emms_c();
2132
2133 if(s->codec_id==CODEC_ID_MPEG4 && s->data_partitioning && s->pict_type!=B_TYPE)
2134 ff_mpeg4_merge_partitions(s);
2135
2136 if (s->msmpeg4_version && s->msmpeg4_version<4 && s->pict_type == I_TYPE)
2137 msmpeg4_encode_ext_header(s);
2138
2139 if(s->codec_id==CODEC_ID_MPEG4)
2140 ff_mpeg4_stuffing(&s->pb);
2141
2142 //if (s->gob_number)
2143 // fprintf(stderr,"\nNumber of GOB: %d", s->gob_number);
2144
2145 /* Send the last GOB if RTP */
2146 if (s->rtp_mode) {
2147 flush_put_bits(&s->pb);
2148 pdif = pbBufPtr(&s->pb) - s->ptr_lastgob;
2149 /* Call the RTP callback to send the last GOB */
2150 if (s->rtp_callback)
2151 s->rtp_callback(s->ptr_lastgob, pdif, s->gob_number);
2152 s->ptr_lastgob = pbBufPtr(&s->pb);
2153 //fprintf(stderr,"\nGOB: %2d size: %d (last)", s->gob_number, pdif);
2154 }
2155 }
2156
2157 static int dct_quantize_c(MpegEncContext *s,
2158 DCTELEM *block, int n,
2159 int qscale, int *overflow)
2160 {
2161 int i, j, level, last_non_zero, q;
2162 const int *qmat;
2163 int bias;
2164 int max=0;
2165 unsigned int threshold1, threshold2;
2166
2167 av_fdct (block);
2168
2169 /* we need this permutation so that we correct the IDCT
2170 permutation. will be moved into DCT code */
2171 block_permute(block);
2172
2173 if (s->mb_intra) {
2174 if (!s->h263_aic) {
2175 if (n < 4)
2176 q = s->y_dc_scale;
2177 else
2178 q = s->c_dc_scale;
2179 q = q << 3;
2180 } else
2181 /* For AIC we skip quant/dequant of INTRADC */
2182 q = 1 << 3;
2183
2184 /* note: block[0] is assumed to be positive */
2185 block[0] = (block[0] + (q >> 1)) / q;
2186 i = 1;
2187 last_non_zero = 0;
2188 qmat = s->q_intra_matrix[qscale];
2189 bias= s->intra_quant_bias<<(QMAT_SHIFT - 3 - QUANT_BIAS_SHIFT);
2190 } else {
2191 i = 0;
2192 last_non_zero = -1;
2193 qmat = s->q_inter_matrix[qscale];
2194 bias= s->inter_quant_bias<<(QMAT_SHIFT - 3 - QUANT_BIAS_SHIFT);
2195 }
2196 threshold1= (1<<(QMAT_SHIFT - 3)) - bias - 1;
2197 threshold2= threshold1<<1;
2198
2199 for(;i<64;i++) {
2200 j = zigzag_direct[i];
2201 level = block[j];
2202 level = level * qmat[j];
2203
2204 // if( bias+level >= (1<<(QMAT_SHIFT - 3))
2205 // || bias-level >= (1<<(QMAT_SHIFT - 3))){
2206 if(((unsigned)(level+threshold1))>threshold2){
2207 if(level>0){
2208 level= (bias + level)>>(QMAT_SHIFT - 3);
2209 block[j]= level;
2210 }else{
2211 level= (bias - level)>>(QMAT_SHIFT - 3);
2212 block[j]= -level;
2213 }
2214 max |=level;
2215 last_non_zero = i;
2216 }else{
2217 block[j]=0;
2218 }
2219 }
2220 *overflow= s->max_qcoeff < max; //overflow might have happend
2221
2222 return last_non_zero;
2223 }
2224
2225 static void dct_unquantize_mpeg1_c(MpegEncContext *s,
2226 DCTELEM *block, int n, int qscale)
2227 {
2228 int i, level, nCoeffs;
2229 const UINT16 *quant_matrix;
2230
2231 if(s->alternate_scan) nCoeffs= 64;
2232 else nCoeffs= s->block_last_index[n]+1;
2233
2234 if (s->mb_intra) {
2235 if (n < 4)
2236 block[0] = block[0] * s->y_dc_scale;
2237 else
2238 block[0] = block[0] * s->c_dc_scale;
2239 /* XXX: only mpeg1 */
2240 quant_matrix = s->intra_matrix;
2241 for(i=1;i<nCoeffs;i++) {
2242 int j= zigzag_direct[i];
2243 level = block[j];
2244 if (level) {
2245 if (level < 0) {
2246 level = -level;
2247 level = (int)(level * qscale * quant_matrix[j]) >> 3;
2248 level = (level - 1) | 1;
2249 level = -level;
2250 } else {
2251 level = (int)(level * qscale * quant_matrix[j]) >> 3;
2252 level = (level - 1) | 1;
2253 }
2254 #ifdef PARANOID
2255 if (level < -2048 || level > 2047)
2256 fprintf(stderr, "unquant error %d %d\n", i, level);
2257 #endif
2258 block[j] = level;
2259 }
2260 }
2261 } else {
2262 i = 0;
2263 quant_matrix = s->inter_matrix;
2264 for(;i<nCoeffs;i++) {
2265 int j= zigzag_direct[i];
2266 level = block[j];
2267 if (level) {
2268 if (level < 0) {
2269 level = -level;
2270 level = (((level << 1) + 1) * qscale *
2271 ((int) (quant_matrix[j]))) >> 4;
2272 level = (level - 1) | 1;
2273 level = -level;
2274 } else {
2275 level = (((level << 1) + 1) * qscale *
2276 ((int) (quant_matrix[j]))) >> 4;
2277 level = (level - 1) | 1;
2278 }
2279 #ifdef PARANOID
2280 if (level < -2048 || level > 2047)
2281 fprintf(stderr, "unquant error %d %d\n", i, level);
2282 #endif
2283 block[j] = level;
2284 }
2285 }
2286 }
2287 }
2288
2289 static void dct_unquantize_mpeg2_c(MpegEncContext *s,
2290 DCTELEM *block, int n, int qscale)
2291 {
2292 int i, level, nCoeffs;
2293 const UINT16 *quant_matrix;
2294
2295 if(s->alternate_scan) nCoeffs= 64;
2296 else nCoeffs= s->block_last_index[n]+1;
2297
2298 if (s->mb_intra) {
2299 if (n < 4)
2300 block[0] = block[0] * s->y_dc_scale;
2301 else
2302 block[0] = block[0] * s->c_dc_scale;
2303 quant_matrix = s->intra_matrix;
2304 for(i=1;i<nCoeffs;i++) {
2305 int j= zigzag_direct[i];
2306 level = block[j];
2307 if (level) {
2308 if (level < 0) {
2309 level = -level;
2310 level = (int)(level * qscale * quant_matrix[j]) >> 3;
2311 level = -level;
2312 } else {
2313 level = (int)(level * qscale * quant_matrix[j]) >> 3;
2314 }
2315 #ifdef PARANOID
2316 if (level < -2048 || level > 2047)
2317 fprintf(stderr, "unquant error %d %d\n", i, level);
2318 #endif
2319 block[j] = level;
2320 }
2321 }
2322 } else {
2323 int sum=-1;
2324 i = 0;
2325 quant_matrix = s->inter_matrix;
2326 for(;i<nCoeffs;i++) {
2327 int j= zigzag_direct[i];
2328 level = block[j];
2329 if (level) {
2330 if (level < 0) {
2331 level = -level;
2332 level = (((level << 1) + 1) * qscale *
2333 ((int) (quant_matrix[j]))) >> 4;
2334 level = -level;
2335 } else {
2336 level = (((level << 1) + 1) * qscale *
2337 ((int) (quant_matrix[j]))) >> 4;
2338 }
2339 #ifdef PARANOID
2340 if (level < -2048 || level > 2047)
2341 fprintf(stderr, "unquant error %d %d\n", i, level);
2342 #endif
2343 block[j] = level;
2344 sum+=level;
2345 }
2346 }
2347 block[63]^=sum&1;
2348 }
2349 }
2350
2351
2352 static void dct_unquantize_h263_c(MpegEncContext *s,
2353 DCTELEM *block, int n, int qscale)
2354 {
2355 int i, level, qmul, qadd;
2356 int nCoeffs;
2357
2358 if (s->mb_intra) {
2359 if (!s->h263_aic) {
2360 if (n < 4)
2361 block[0] = block[0] * s->y_dc_scale;
2362 else
2363 block[0] = block[0] * s->c_dc_scale;
2364 }
2365 i = 1;
2366 nCoeffs= 64; //does not allways use zigzag table
2367 } else {
2368 i = 0;
2369 nCoeffs= zigzag_end[ s->block_last_index[n] ];
2370 }
2371
2372 qmul = s->qscale << 1;
2373 if (s->h263_aic && s->mb_intra)
2374 qadd = 0;
2375 else
2376 qadd = (s->qscale - 1) | 1;
2377
2378 for(;i<nCoeffs;i++) {
2379 level = block[i];
2380 if (level) {
2381 if (level < 0) {
2382 level = level * qmul - qadd;
2383 } else {
2384 level = level * qmul + qadd;
2385 }
2386 #ifdef PARANOID
2387 if (level < -2048 || level > 2047)
2388 fprintf(stderr, "unquant error %d %d\n", i, level);
2389 #endif
2390 block[i] = level;
2391 }
2392 }
2393 }
2394
2395 static void remove_ac(MpegEncContext *s, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, int mb_x, int mb_y)
2396 {
2397 int dc, dcb, dcr, y, i;
2398 for(i=0; i<4; i++){
2399 dc= s->dc_val[0][mb_x*2+1 + (i&1) + (mb_y*2+1 + (i>>1))*(s->mb_width*2+2)];
2400 for(y=0; y<8; y++){
2401 int x;
2402 for(x=0; x<8; x++){
2403 dest_y[x + (i&1)*8 + (y + (i>>1)*8)*s->linesize]= dc/8;
2404 }
2405 }
2406 }
2407 dcb = s->dc_val[1][mb_x+1 + (mb_y+1)*(s->mb_width+2)];
2408 dcr= s->dc_val[2][mb_x+1 + (mb_y+1)*(s->mb_width+2)];
2409 for(y=0; y<8; y++){
2410 int x;
2411 for(x=0; x<8; x++){
2412 dest_cb[x + y*(s->linesize>>1)]= dcb/8;
2413 dest_cr[x + y*(s->linesize>>1)]= dcr/8;
2414 }
2415 }
2416 }
2417
2418 /**
2419 * will conceal past errors, and allso drop b frames if needed
2420 *
2421 */
2422 void ff_conceal_past_errors(MpegEncContext *s, int unknown_pos)
2423 {
2424 int mb_x= s->mb_x;
2425 int mb_y= s->mb_y;
2426 int mb_dist=0;
2427 int i, intra_count=0, inter_count=0;
2428 int intra_conceal= s->msmpeg4_version ? 50 : 50; //FIXME finetune
2429 int inter_conceal= s->msmpeg4_version ? 50 : 50;
2430
2431 // for last block
2432 if(mb_x>=s->mb_width) mb_x= s->mb_width -1;
2433 if(mb_y>=s->mb_height) mb_y= s->mb_height-1;
2434
2435 if(s->decoding_error==0 && unknown_pos){
2436 if(s->data_partitioning && s->pict_type!=B_TYPE)
2437 s->decoding_error= DECODING_AC_LOST;
2438 else
2439 s->decoding_error= DECODING_DESYNC;
2440 }
2441
2442 if(s->decoding_error==DECODING_DESYNC && s->pict_type!=B_TYPE) s->next_p_frame_damaged=1;
2443
2444 for(i=mb_x + mb_y*s->mb_width; i>=0; i--){
2445 if(s->mbintra_table[i]) intra_count++;
2446 else inter_count++;
2447 }
2448
2449 if(s->decoding_error==DECODING_AC_LOST){
2450 intra_conceal*=2;
2451 inter_conceal*=2;
2452 }else if(s->decoding_error==DECODING_ACDC_LOST){
2453 intra_conceal*=2;
2454 inter_conceal*=2;
2455 }
2456
2457 if(unknown_pos && (intra_count<inter_count)){
2458 intra_conceal= inter_conceal= s->mb_num;
2459 // printf("%d %d\n",intra_count, inter_count);
2460 }
2461
2462 fprintf(stderr, "concealing errors\n");
2463
2464 /* for all MBs from the current one back until the last resync marker */
2465 for(; mb_y>=0 && mb_y>=s->resync_mb_y; mb_y--){
2466 for(; mb_x>=0; mb_x--){
2467 uint8_t *dest_y = s->current_picture[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
2468 uint8_t *dest_cb = s->current_picture[1] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8;
2469 uint8_t *dest_cr = s->current_picture[2] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8;
2470 int mb_x_backup= s->mb_x; //FIXME pass xy to mpeg_motion
2471 int mb_y_backup= s->mb_y;
2472 s->mb_x=mb_x;
2473 s->mb_y=mb_y;
2474 if(s->mbintra_table[mb_y*s->mb_width + mb_x] && mb_dist<intra_conceal){
2475 if(s->decoding_error==DECODING_AC_LOST){
2476 remove_ac(s, dest_y, dest_cb, dest_cr, mb_x, mb_y);
2477 // printf("remove ac to %d %d\n", mb_x, mb_y);
2478 }else{
2479 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
2480 s->last_picture, 0, 0, put_pixels_tab,
2481 0/*mx*/, 0/*my*/, 16);
2482 }
2483 }
2484 else if(!s->mbintra_table[mb_y*s->mb_width + mb_x] && mb_dist<inter_conceal){
2485 int mx=0;
2486 int my=0;
2487
2488 if(s->decoding_error!=DECODING_DESYNC){
2489 int xy= mb_x*2+1 + (mb_y*2+1)*(s->mb_width*2+2);
2490 mx= s->motion_val[ xy ][0];
2491 my= s->motion_val[ xy ][1];
2492 }
2493
2494 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
2495 s->last_picture, 0, 0, put_pixels_tab,
2496 mx, my, 16);
2497 }
2498 s->mb_x= mb_x_backup;
2499 s->mb_y= mb_y_backup;
2500
2501 if(mb_x== s->resync_mb_x && mb_y== s->resync_mb_y) return;
2502 if(!s->mbskip_table[mb_x + mb_y*s->mb_width]) mb_dist++;
2503 }
2504 mb_x=s->mb_width-1;
2505 }
2506 }
2507
2508 AVCodec mpeg1video_encoder = {
2509 "mpeg1video",
2510 CODEC_TYPE_VIDEO,
2511 CODEC_ID_MPEG1VIDEO,
2512 sizeof(MpegEncContext),
2513 MPV_encode_init,
2514 MPV_encode_picture,
2515 MPV_encode_end,
2516 };
2517
2518 AVCodec h263_encoder = {
2519 "h263",
2520 CODEC_TYPE_VIDEO,
2521 CODEC_ID_H263,
2522 sizeof(MpegEncContext),
2523 MPV_encode_init,
2524 MPV_encode_picture,
2525 MPV_encode_end,
2526 };
2527
2528 AVCodec h263p_encoder = {
2529 "h263p",
2530 CODEC_TYPE_VIDEO,
2531 CODEC_ID_H263P,
2532 sizeof(MpegEncContext),
2533 MPV_encode_init,
2534 MPV_encode_picture,
2535 MPV_encode_end,
2536 };
2537
2538 AVCodec rv10_encoder = {
2539 "rv10",
2540 CODEC_TYPE_VIDEO,
2541 CODEC_ID_RV10,
2542 sizeof(MpegEncContext),
2543 MPV_encode_init,
2544 MPV_encode_picture,
2545 MPV_encode_end,
2546 };
2547
2548 AVCodec mjpeg_encoder = {
2549 "mjpeg",
2550 CODEC_TYPE_VIDEO,
2551 CODEC_ID_MJPEG,
2552 sizeof(MpegEncContext),
2553 MPV_encode_init,
2554 MPV_encode_picture,
2555 MPV_encode_end,
2556 };
2557
2558 AVCodec mpeg4_encoder = {
2559 "mpeg4",
2560 CODEC_TYPE_VIDEO,
2561 CODEC_ID_MPEG4,
2562 sizeof(MpegEncContext),
2563 MPV_encode_init,
2564 MPV_encode_picture,
2565 MPV_encode_end,
2566 };
2567
2568 AVCodec msmpeg4v1_encoder = {
2569 "msmpeg4v1",
2570 CODEC_TYPE_VIDEO,
2571 CODEC_ID_MSMPEG4V1,
2572 sizeof(MpegEncContext),
2573 MPV_encode_init,
2574 MPV_encode_picture,
2575 MPV_encode_end,
2576 };
2577
2578 AVCodec msmpeg4v2_encoder = {
2579 "msmpeg4v2",
2580 CODEC_TYPE_VIDEO,
2581 CODEC_ID_MSMPEG4V2,
2582 sizeof(MpegEncContext),
2583 MPV_encode_init,
2584 MPV_encode_picture,
2585 MPV_encode_end,
2586 };
2587
2588 AVCodec msmpeg4v3_encoder = {
2589 "msmpeg4",
2590 CODEC_TYPE_VIDEO,
2591 CODEC_ID_MSMPEG4V3,
2592 sizeof(MpegEncContext),
2593 MPV_encode_init,
2594 MPV_encode_picture,
2595 MPV_encode_end,
2596 };
2597
2598 AVCodec wmv1_encoder = {
2599 "wmv1",
2600 CODEC_TYPE_VIDEO,
2601 CODEC_ID_WMV1,
2602 sizeof(MpegEncContext),
2603 MPV_encode_init,
2604 MPV_encode_picture,
2605 MPV_encode_end,
2606 };
2607
2608 AVCodec wmv2_encoder = {
2609 "wmv2",
2610 CODEC_TYPE_VIDEO,
2611 CODEC_ID_WMV2,
2612 sizeof(MpegEncContext),
2613 MPV_encode_init,
2614 MPV_encode_picture,
2615 MPV_encode_end,
2616 };