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