- ME method compatibility with legacy apps.
[libav.git] / libavcodec / mpegvideo.c
1 /*
2 * The simplest mpeg encoder (well, it was the simplest!)
3 * Copyright (c) 2000,2001 Gerard Lantau.
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
9 *
10 * This program 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
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
18 *
19 * 4MV & hq encoding stuff by Michael Niedermayer <michaelni@gmx.at>
20 */
21 #include <stdlib.h>
22 #include <stdio.h>
23 #include <math.h>
24 #include <string.h>
25 #include "avcodec.h"
26 #include "dsputil.h"
27 #include "mpegvideo.h"
28
29 #ifdef USE_FASTMEMCPY
30 #include "fastmemcpy.h"
31 #endif
32
33 static void encode_picture(MpegEncContext *s, int picture_number);
34 static void rate_control_init(MpegEncContext *s);
35 static int rate_estimate_qscale(MpegEncContext *s);
36 static void dct_unquantize_mpeg1_c(MpegEncContext *s,
37 DCTELEM *block, int n, int qscale);
38 static void dct_unquantize_h263_c(MpegEncContext *s,
39 DCTELEM *block, int n, int qscale);
40 static void draw_edges_c(UINT8 *buf, int wrap, int width, int height, int w);
41 static int dct_quantize_c(MpegEncContext *s, DCTELEM *block, int n, int qscale);
42
43 int (*dct_quantize)(MpegEncContext *s, DCTELEM *block, int n, int qscale)= dct_quantize_c;
44 void (*draw_edges)(UINT8 *buf, int wrap, int width, int height, int w)= draw_edges_c;
45
46 #define EDGE_WIDTH 16
47
48 /* enable all paranoid tests for rounding, overflows, etc... */
49 //#define PARANOID
50
51 //#define DEBUG
52
53
54 /* for jpeg fast DCT */
55 #define CONST_BITS 14
56
57 static const unsigned short aanscales[64] = {
58 /* precomputed values scaled up by 14 bits */
59 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520,
60 22725, 31521, 29692, 26722, 22725, 17855, 12299, 6270,
61 21407, 29692, 27969, 25172, 21407, 16819, 11585, 5906,
62 19266, 26722, 25172, 22654, 19266, 15137, 10426, 5315,
63 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520,
64 12873, 17855, 16819, 15137, 12873, 10114, 6967, 3552,
65 8867, 12299, 11585, 10426, 8867, 6967, 4799, 2446,
66 4520, 6270, 5906, 5315, 4520, 3552, 2446, 1247
67 };
68
69 static UINT8 h263_chroma_roundtab[16] = {
70 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2,
71 };
72
73 static UINT16 default_mv_penalty[MAX_FCODE+1][MAX_MV*2+1];
74 static UINT8 default_fcode_tab[MAX_MV*2+1];
75
76 extern UINT8 zigzag_end[64];
77
78 /* default motion estimation */
79 int motion_estimation_method = ME_EPZS;
80
81 static void convert_matrix(int *qmat, UINT16 *qmat16, const UINT16 *quant_matrix, int qscale)
82 {
83 int i;
84
85 if (av_fdct == jpeg_fdct_ifast) {
86 for(i=0;i<64;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[block_permute_op(i)] = (int)((UINT64_C(1) << (QMAT_SHIFT + 11)) /
93 (aanscales[i] * qscale * quant_matrix[block_permute_op(i)]));
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[i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]);
103 qmat16[i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[block_permute_op(i)]);
104 }
105 }
106 }
107
108 /* init common structure for both encoder and decoder */
109 int MPV_common_init(MpegEncContext *s)
110 {
111 int c_size, i;
112 UINT8 *pict;
113
114 s->dct_unquantize_h263 = dct_unquantize_h263_c;
115 s->dct_unquantize_mpeg = dct_unquantize_mpeg1_c;
116
117 #ifdef HAVE_MMX
118 MPV_common_init_mmx(s);
119 #endif
120 //setup default unquantizers (mpeg4 might change it later)
121 if(s->out_format == FMT_H263)
122 s->dct_unquantize = s->dct_unquantize_h263;
123 else
124 s->dct_unquantize = s->dct_unquantize_mpeg;
125
126 s->mb_width = (s->width + 15) / 16;
127 s->mb_height = (s->height + 15) / 16;
128 s->mb_num = s->mb_width * s->mb_height;
129 s->linesize = s->mb_width * 16 + 2 * EDGE_WIDTH;
130
131 for(i=0;i<3;i++) {
132 int w, h, shift, pict_start;
133
134 w = s->linesize;
135 h = s->mb_height * 16 + 2 * EDGE_WIDTH;
136 shift = (i == 0) ? 0 : 1;
137 c_size = (w >> shift) * (h >> shift);
138 pict_start = (w >> shift) * (EDGE_WIDTH >> shift) + (EDGE_WIDTH >> shift);
139
140 pict = av_mallocz(c_size);
141 if (pict == NULL)
142 goto fail;
143 s->last_picture_base[i] = pict;
144 s->last_picture[i] = pict + pict_start;
145
146 pict = av_mallocz(c_size);
147 if (pict == NULL)
148 goto fail;
149 s->next_picture_base[i] = pict;
150 s->next_picture[i] = pict + pict_start;
151
152 if (s->has_b_frames) {
153 pict = av_mallocz(c_size);
154 if (pict == NULL)
155 goto fail;
156 s->aux_picture_base[i] = pict;
157 s->aux_picture[i] = pict + pict_start;
158 }
159 }
160
161 if (s->encoding) {
162 /* Allocate MB type table */
163 s->mb_type = av_mallocz(s->mb_num * sizeof(char));
164 if (s->mb_type == NULL) {
165 perror("malloc");
166 goto fail;
167 }
168
169 s->mb_var = av_mallocz(s->mb_num * sizeof(INT16));
170 if (s->mb_var == NULL) {
171 perror("malloc");
172 goto fail;
173 }
174 /* Allocate MV table */
175 /* By now we just have one MV per MB */
176 s->mv_table[0] = av_mallocz(s->mb_num * sizeof(INT16));
177 s->mv_table[1] = av_mallocz(s->mb_num * sizeof(INT16));
178 if (s->mv_table[1] == NULL || s->mv_table[0] == NULL) {
179 perror("malloc");
180 goto fail;
181 }
182 }
183
184 if (s->out_format == FMT_H263 || s->encoding) {
185 int size;
186 /* MV prediction */
187 size = (2 * s->mb_width + 2) * (2 * s->mb_height + 2);
188 s->motion_val = malloc(size * 2 * sizeof(INT16));
189 if (s->motion_val == NULL)
190 goto fail;
191 memset(s->motion_val, 0, size * 2 * sizeof(INT16));
192 }
193
194 if (s->h263_pred || s->h263_plus) {
195 int y_size, c_size, i, size;
196
197 /* dc values */
198
199 y_size = (2 * s->mb_width + 2) * (2 * s->mb_height + 2);
200 c_size = (s->mb_width + 2) * (s->mb_height + 2);
201 size = y_size + 2 * c_size;
202 s->dc_val[0] = malloc(size * sizeof(INT16));
203 if (s->dc_val[0] == NULL)
204 goto fail;
205 s->dc_val[1] = s->dc_val[0] + y_size;
206 s->dc_val[2] = s->dc_val[1] + c_size;
207 for(i=0;i<size;i++)
208 s->dc_val[0][i] = 1024;
209
210 /* ac values */
211 s->ac_val[0] = av_mallocz(size * sizeof(INT16) * 16);
212 if (s->ac_val[0] == NULL)
213 goto fail;
214 s->ac_val[1] = s->ac_val[0] + y_size;
215 s->ac_val[2] = s->ac_val[1] + c_size;
216
217 /* cbp values */
218 s->coded_block = av_mallocz(y_size);
219 if (!s->coded_block)
220 goto fail;
221
222 /* which mb is a intra block */
223 s->mbintra_table = av_mallocz(s->mb_num);
224 if (!s->mbintra_table)
225 goto fail;
226 memset(s->mbintra_table, 1, s->mb_num);
227 }
228 /* default structure is frame */
229 s->picture_structure = PICT_FRAME;
230
231 /* init macroblock skip table */
232 if (!s->encoding) {
233 s->mbskip_table = av_mallocz(s->mb_num);
234 if (!s->mbskip_table)
235 goto fail;
236 }
237
238 s->block= s->intra_block;
239
240 s->context_initialized = 1;
241 return 0;
242 fail:
243 MPV_common_end(s);
244 return -1;
245 }
246
247 /* init common structure for both encoder and decoder */
248 void MPV_common_end(MpegEncContext *s)
249 {
250 int i;
251
252 if (s->mb_type)
253 free(s->mb_type);
254 if (s->mb_var)
255 free(s->mb_var);
256 if (s->mv_table[0])
257 free(s->mv_table[0]);
258 if (s->mv_table[1])
259 free(s->mv_table[1]);
260 if (s->motion_val)
261 free(s->motion_val);
262 if (s->dc_val[0])
263 free(s->dc_val[0]);
264 if (s->ac_val[0])
265 free(s->ac_val[0]);
266 if (s->coded_block)
267 free(s->coded_block);
268 if (s->mbintra_table)
269 free(s->mbintra_table);
270
271 if (s->mbskip_table)
272 free(s->mbskip_table);
273 for(i=0;i<3;i++) {
274 if (s->last_picture_base[i])
275 free(s->last_picture_base[i]);
276 if (s->next_picture_base[i])
277 free(s->next_picture_base[i]);
278 if (s->has_b_frames)
279 free(s->aux_picture_base[i]);
280 }
281 s->context_initialized = 0;
282 }
283
284 /* init video encoder */
285 int MPV_encode_init(AVCodecContext *avctx)
286 {
287 MpegEncContext *s = avctx->priv_data;
288 int i;
289
290 avctx->pix_fmt = PIX_FMT_YUV420P;
291
292 s->bit_rate = avctx->bit_rate;
293 s->bit_rate_tolerance = avctx->bit_rate_tolerance;
294 s->frame_rate = avctx->frame_rate;
295 s->width = avctx->width;
296 s->height = avctx->height;
297 s->gop_size = avctx->gop_size;
298 s->rtp_mode = avctx->rtp_mode;
299 s->rtp_payload_size = avctx->rtp_payload_size;
300 if (avctx->rtp_callback)
301 s->rtp_callback = avctx->rtp_callback;
302 s->qmin= avctx->qmin;
303 s->qmax= avctx->qmax;
304 s->max_qdiff= avctx->max_qdiff;
305 s->qcompress= avctx->qcompress;
306 s->qblur= avctx->qblur;
307 s->avctx = avctx;
308 s->aspect_ratio_info= avctx->aspect_ratio_info;
309 s->flags= avctx->flags;
310
311 if (s->gop_size <= 1) {
312 s->intra_only = 1;
313 s->gop_size = 12;
314 } else {
315 s->intra_only = 0;
316 }
317
318 /* ME algorithm */
319 if (avctx->me_method == 0)
320 /* For compatibility */
321 s->me_method = motion_estimation_method;
322 else
323 s->me_method = avctx->me_method;
324
325 /* Fixed QSCALE */
326 s->fixed_qscale = (avctx->flags & CODEC_FLAG_QSCALE);
327
328 switch(avctx->codec->id) {
329 case CODEC_ID_MPEG1VIDEO:
330 s->out_format = FMT_MPEG1;
331 break;
332 case CODEC_ID_MJPEG:
333 s->out_format = FMT_MJPEG;
334 s->intra_only = 1; /* force intra only for jpeg */
335 s->mjpeg_write_tables = 1; /* write all tables */
336 s->mjpeg_vsample[0] = 2; /* set up default sampling factors */
337 s->mjpeg_vsample[1] = 1; /* the only currently supported values */
338 s->mjpeg_vsample[2] = 1;
339 s->mjpeg_hsample[0] = 2;
340 s->mjpeg_hsample[1] = 1;
341 s->mjpeg_hsample[2] = 1;
342 if (mjpeg_init(s) < 0)
343 return -1;
344 break;
345 case CODEC_ID_H263:
346 if (h263_get_picture_format(s->width, s->height) == 7) {
347 printf("Input picture size isn't suitable for h263 codec! try h263+\n");
348 return -1;
349 }
350 s->out_format = FMT_H263;
351 break;
352 case CODEC_ID_H263P:
353 s->out_format = FMT_H263;
354 s->rtp_mode = 1;
355 s->rtp_payload_size = 1200;
356 s->h263_plus = 1;
357 s->unrestricted_mv = 1;
358
359 /* These are just to be sure */
360 s->umvplus = 0;
361 s->umvplus_dec = 0;
362 break;
363 case CODEC_ID_RV10:
364 s->out_format = FMT_H263;
365 s->h263_rv10 = 1;
366 break;
367 case CODEC_ID_MPEG4:
368 s->out_format = FMT_H263;
369 s->h263_pred = 1;
370 s->unrestricted_mv = 1;
371 break;
372 case CODEC_ID_MSMPEG4V1:
373 s->out_format = FMT_H263;
374 s->h263_msmpeg4 = 1;
375 s->h263_pred = 1;
376 s->unrestricted_mv = 1;
377 s->msmpeg4_version= 1;
378 break;
379 case CODEC_ID_MSMPEG4V2:
380 s->out_format = FMT_H263;
381 s->h263_msmpeg4 = 1;
382 s->h263_pred = 1;
383 s->unrestricted_mv = 1;
384 s->msmpeg4_version= 2;
385 break;
386 case CODEC_ID_MSMPEG4V3:
387 s->out_format = FMT_H263;
388 s->h263_msmpeg4 = 1;
389 s->h263_pred = 1;
390 s->unrestricted_mv = 1;
391 s->msmpeg4_version= 3;
392 break;
393 default:
394 return -1;
395 }
396
397 if((s->flags&CODEC_FLAG_4MV) && !(s->flags&CODEC_FLAG_HQ)){
398 printf("4MV is currently only supported in HQ mode\n");
399 return -1;
400 }
401
402 { /* set up some save defaults, some codecs might override them later */
403 static int done=0;
404 if(!done){
405 int i;
406 done=1;
407 memset(default_mv_penalty, 0, sizeof(UINT16)*(MAX_FCODE+1)*(2*MAX_MV+1));
408 memset(default_fcode_tab , 0, sizeof(UINT8)*(2*MAX_MV+1));
409
410 for(i=-16; i<16; i++){
411 default_fcode_tab[i + MAX_MV]= 1;
412 }
413 }
414 }
415 s->mv_penalty= default_mv_penalty;
416 s->fcode_tab= default_fcode_tab;
417
418 if (s->out_format == FMT_H263)
419 h263_encode_init(s);
420 else if (s->out_format == FMT_MPEG1)
421 mpeg1_encode_init(s);
422
423 /* dont use mv_penalty table for crap MV as it would be confused */
424 if (s->me_method < 5) s->mv_penalty = default_mv_penalty;
425
426 s->encoding = 1;
427
428 /* init */
429 if (MPV_common_init(s) < 0)
430 return -1;
431
432 /* init default q matrix */
433 for(i=0;i<64;i++) {
434 s->intra_matrix[i] = default_intra_matrix[i];
435 s->non_intra_matrix[i] = default_non_intra_matrix[i];
436 }
437
438 /* rate control init */
439 rate_control_init(s);
440
441 s->picture_number = 0;
442 s->picture_in_gop_number = 0;
443 s->fake_picture_number = 0;
444 /* motion detector init */
445 s->f_code = 1;
446
447 return 0;
448 }
449
450 int MPV_encode_end(AVCodecContext *avctx)
451 {
452 MpegEncContext *s = avctx->priv_data;
453
454 #ifdef STATS
455 print_stats();
456 #endif
457 MPV_common_end(s);
458 if (s->out_format == FMT_MJPEG)
459 mjpeg_close(s);
460
461 return 0;
462 }
463
464 /* draw the edges of width 'w' of an image of size width, height */
465 static void draw_edges_c(UINT8 *buf, int wrap, int width, int height, int w)
466 {
467 UINT8 *ptr, *last_line;
468 int i;
469
470 last_line = buf + (height - 1) * wrap;
471 for(i=0;i<w;i++) {
472 /* top and bottom */
473 memcpy(buf - (i + 1) * wrap, buf, width);
474 memcpy(last_line + (i + 1) * wrap, last_line, width);
475 }
476 /* left and right */
477 ptr = buf;
478 for(i=0;i<height;i++) {
479 memset(ptr - w, ptr[0], w);
480 memset(ptr + width, ptr[width-1], w);
481 ptr += wrap;
482 }
483 /* corners */
484 for(i=0;i<w;i++) {
485 memset(buf - (i + 1) * wrap - w, buf[0], w); /* top left */
486 memset(buf - (i + 1) * wrap + width, buf[width-1], w); /* top right */
487 memset(last_line + (i + 1) * wrap - w, last_line[0], w); /* top left */
488 memset(last_line + (i + 1) * wrap + width, last_line[width-1], w); /* top right */
489 }
490 }
491
492 /* generic function for encode/decode called before a frame is coded/decoded */
493 void MPV_frame_start(MpegEncContext *s)
494 {
495 int i;
496 UINT8 *tmp;
497
498 s->mb_skiped = 0;
499 if (s->pict_type == B_TYPE) {
500 for(i=0;i<3;i++) {
501 s->current_picture[i] = s->aux_picture[i];
502 }
503 } else {
504 s->last_non_b_pict_type= s->pict_type;
505 for(i=0;i<3;i++) {
506 /* swap next and last */
507 tmp = s->last_picture[i];
508 s->last_picture[i] = s->next_picture[i];
509 s->next_picture[i] = tmp;
510 s->current_picture[i] = tmp;
511 }
512 }
513 }
514
515 /* generic function for encode/decode called after a frame has been coded/decoded */
516 void MPV_frame_end(MpegEncContext *s)
517 {
518 /* draw edge for correct motion prediction if outside */
519 if (s->pict_type != B_TYPE && !s->intra_only) {
520 if(s->avctx==NULL || s->avctx->codec->id!=CODEC_ID_MPEG4 || s->divx_version==500){
521 draw_edges(s->current_picture[0], s->linesize, s->mb_width*16, s->mb_height*16, EDGE_WIDTH);
522 draw_edges(s->current_picture[1], s->linesize/2, s->mb_width*8, s->mb_height*8, EDGE_WIDTH/2);
523 draw_edges(s->current_picture[2], s->linesize/2, s->mb_width*8, s->mb_height*8, EDGE_WIDTH/2);
524 }else{
525 /* mpeg4? / opendivx / xvid */
526 draw_edges(s->current_picture[0], s->linesize, s->width, s->height, EDGE_WIDTH);
527 draw_edges(s->current_picture[1], s->linesize/2, s->width/2, s->height/2, EDGE_WIDTH/2);
528 draw_edges(s->current_picture[2], s->linesize/2, s->width/2, s->height/2, EDGE_WIDTH/2);
529 }
530 }
531 emms_c();
532 }
533
534 int MPV_encode_picture(AVCodecContext *avctx,
535 unsigned char *buf, int buf_size, void *data)
536 {
537 MpegEncContext *s = avctx->priv_data;
538 AVPicture *pict = data;
539 int i, j;
540
541 if (s->fixed_qscale)
542 s->qscale = avctx->quality;
543
544 init_put_bits(&s->pb, buf, buf_size, NULL, NULL);
545
546 s->force_type= (avctx->flags&CODEC_FLAG_TYPE) ?
547 (avctx->key_frame ? I_TYPE : P_TYPE) : 0;
548 if (!s->intra_only) {
549 /* first picture of GOP is intra */
550 if (s->picture_in_gop_number % s->gop_size==0 || s->force_type==I_TYPE){
551 s->picture_in_gop_number=0;
552 s->pict_type = I_TYPE;
553 }else
554 s->pict_type = P_TYPE;
555 } else {
556 s->pict_type = I_TYPE;
557 }
558
559 MPV_frame_start(s);
560
561 for(i=0;i<3;i++) {
562 UINT8 *src = pict->data[i];
563 UINT8 *dest = s->current_picture[i];
564 int src_wrap = pict->linesize[i];
565 int dest_wrap = s->linesize;
566 int w = s->width;
567 int h = s->height;
568
569 if (i >= 1) {
570 dest_wrap >>= 1;
571 w >>= 1;
572 h >>= 1;
573 }
574
575 if(dest_wrap==src_wrap){
576 s->new_picture[i] = pict->data[i];
577 } else {
578 for(j=0;j<h;j++) {
579 memcpy(dest, src, w);
580 dest += dest_wrap;
581 src += src_wrap;
582 }
583 s->new_picture[i] = s->current_picture[i];
584 }
585 }
586
587 encode_picture(s, s->picture_number);
588 avctx->key_frame = (s->pict_type == I_TYPE);
589 avctx->header_bits = s->header_bits;
590 avctx->mv_bits = s->mv_bits;
591 avctx->misc_bits = s->misc_bits;
592 avctx->i_tex_bits = s->i_tex_bits;
593 avctx->p_tex_bits = s->p_tex_bits;
594 avctx->i_count = s->i_count;
595 avctx->p_count = s->p_count;
596 avctx->skip_count = s->skip_count;
597
598 MPV_frame_end(s);
599 s->picture_number++;
600 s->picture_in_gop_number++;
601
602 if (s->out_format == FMT_MJPEG)
603 mjpeg_picture_trailer(s);
604
605 flush_put_bits(&s->pb);
606 s->last_frame_bits= s->frame_bits;
607 s->frame_bits = (pbBufPtr(&s->pb) - s->pb.buf) * 8;
608 s->total_bits += s->frame_bits;
609 avctx->frame_bits = s->frame_bits;
610 //printf("fcode: %d, type: %d, head: %d, mv: %d, misc: %d, frame: %d, itex: %d, ptex: %d\n",
611 //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);
612
613 avctx->quality = s->qscale;
614 if (avctx->get_psnr) {
615 /* At this point pict->data should have the original frame */
616 /* an s->current_picture should have the coded/decoded frame */
617 get_psnr(pict->data, s->current_picture,
618 pict->linesize, s->linesize, avctx);
619 }
620 return pbBufPtr(&s->pb) - s->pb.buf;
621 }
622
623 static inline int clip(int a, int amin, int amax)
624 {
625 if (a < amin)
626 return amin;
627 else if (a > amax)
628 return amax;
629 else
630 return a;
631 }
632
633 static inline void gmc1_motion(MpegEncContext *s,
634 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr,
635 int dest_offset,
636 UINT8 **ref_picture, int src_offset,
637 int h)
638 {
639 UINT8 *ptr;
640 int dxy, offset, mx, my, src_x, src_y, height, linesize;
641 int motion_x, motion_y;
642
643 if(s->real_sprite_warping_points>1) printf("more than 1 warp point isnt supported\n");
644 motion_x= s->sprite_offset[0][0];
645 motion_y= s->sprite_offset[0][1];
646 src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy+1));
647 src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy+1));
648 motion_x<<=(3-s->sprite_warping_accuracy);
649 motion_y<<=(3-s->sprite_warping_accuracy);
650 src_x = clip(src_x, -16, s->width);
651 if (src_x == s->width)
652 motion_x =0;
653 src_y = clip(src_y, -16, s->height);
654 if (src_y == s->height)
655 motion_y =0;
656
657 linesize = s->linesize;
658 ptr = ref_picture[0] + (src_y * linesize) + src_x + src_offset;
659
660 dest_y+=dest_offset;
661 gmc1(dest_y , ptr , linesize, h, motion_x&15, motion_y&15, s->no_rounding);
662 gmc1(dest_y+8, ptr+8, linesize, h, motion_x&15, motion_y&15, s->no_rounding);
663
664 motion_x= s->sprite_offset[1][0];
665 motion_y= s->sprite_offset[1][1];
666 src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy+1));
667 src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy+1));
668 motion_x<<=(3-s->sprite_warping_accuracy);
669 motion_y<<=(3-s->sprite_warping_accuracy);
670 src_x = clip(src_x, -8, s->width>>1);
671 if (src_x == s->width>>1)
672 motion_x =0;
673 src_y = clip(src_y, -8, s->height>>1);
674 if (src_y == s->height>>1)
675 motion_y =0;
676
677 offset = (src_y * linesize>>1) + src_x + (src_offset>>1);
678 ptr = ref_picture[1] + offset;
679 gmc1(dest_cb + (dest_offset>>1), ptr, linesize>>1, h>>1, motion_x&15, motion_y&15, s->no_rounding);
680 ptr = ref_picture[2] + offset;
681 gmc1(dest_cr + (dest_offset>>1), ptr, linesize>>1, h>>1, motion_x&15, motion_y&15, s->no_rounding);
682
683 return;
684 }
685
686 /* apply one mpeg motion vector to the three components */
687 static inline void mpeg_motion(MpegEncContext *s,
688 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr,
689 int dest_offset,
690 UINT8 **ref_picture, int src_offset,
691 int field_based, op_pixels_func *pix_op,
692 int motion_x, int motion_y, int h)
693 {
694 UINT8 *ptr;
695 int dxy, offset, mx, my, src_x, src_y, height, linesize;
696 if(s->quarter_sample)
697 {
698 motion_x>>=1;
699 motion_y>>=1;
700 }
701 dxy = ((motion_y & 1) << 1) | (motion_x & 1);
702 src_x = s->mb_x * 16 + (motion_x >> 1);
703 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 1);
704
705 /* WARNING: do no forget half pels */
706 height = s->height >> field_based;
707 src_x = clip(src_x, -16, s->width);
708 if (src_x == s->width)
709 dxy &= ~1;
710 src_y = clip(src_y, -16, height);
711 if (src_y == height)
712 dxy &= ~2;
713 linesize = s->linesize << field_based;
714 ptr = ref_picture[0] + (src_y * linesize) + (src_x) + src_offset;
715 dest_y += dest_offset;
716 pix_op[dxy](dest_y, ptr, linesize, h);
717 pix_op[dxy](dest_y + 8, ptr + 8, linesize, h);
718
719 if (s->out_format == FMT_H263) {
720 dxy = 0;
721 if ((motion_x & 3) != 0)
722 dxy |= 1;
723 if ((motion_y & 3) != 0)
724 dxy |= 2;
725 mx = motion_x >> 2;
726 my = motion_y >> 2;
727 } else {
728 mx = motion_x / 2;
729 my = motion_y / 2;
730 dxy = ((my & 1) << 1) | (mx & 1);
731 mx >>= 1;
732 my >>= 1;
733 }
734
735 src_x = s->mb_x * 8 + mx;
736 src_y = s->mb_y * (8 >> field_based) + my;
737 src_x = clip(src_x, -8, s->width >> 1);
738 if (src_x == (s->width >> 1))
739 dxy &= ~1;
740 src_y = clip(src_y, -8, height >> 1);
741 if (src_y == (height >> 1))
742 dxy &= ~2;
743
744 offset = (src_y * (linesize >> 1)) + src_x + (src_offset >> 1);
745 ptr = ref_picture[1] + offset;
746 pix_op[dxy](dest_cb + (dest_offset >> 1), ptr, linesize >> 1, h >> 1);
747 ptr = ref_picture[2] + offset;
748 pix_op[dxy](dest_cr + (dest_offset >> 1), ptr, linesize >> 1, h >> 1);
749 }
750
751 static inline void qpel_motion(MpegEncContext *s,
752 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr,
753 int dest_offset,
754 UINT8 **ref_picture, int src_offset,
755 int field_based, op_pixels_func *pix_op,
756 qpel_mc_func *qpix_op,
757 int motion_x, int motion_y, int h)
758 {
759 UINT8 *ptr;
760 int dxy, offset, mx, my, src_x, src_y, height, linesize;
761
762 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
763 src_x = s->mb_x * 16 + (motion_x >> 2);
764 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2);
765
766 height = s->height >> field_based;
767 src_x = clip(src_x, -16, s->width);
768 if (src_x == s->width)
769 dxy &= ~3;
770 src_y = clip(src_y, -16, height);
771 if (src_y == height)
772 dxy &= ~12;
773 linesize = s->linesize << field_based;
774 ptr = ref_picture[0] + (src_y * linesize) + src_x + src_offset;
775 dest_y += dest_offset;
776 //printf("%d %d %d\n", src_x, src_y, dxy);
777 qpix_op[dxy](dest_y , ptr , linesize, linesize, motion_x&3, motion_y&3);
778 qpix_op[dxy](dest_y + 8, ptr + 8, linesize, linesize, motion_x&3, motion_y&3);
779 qpix_op[dxy](dest_y + linesize*8 , ptr + linesize*8 , linesize, linesize, motion_x&3, motion_y&3);
780 qpix_op[dxy](dest_y + linesize*8 + 8, ptr + linesize*8 + 8, linesize, linesize, motion_x&3, motion_y&3);
781
782 mx= (motion_x>>1) | (motion_x&1);
783 my= (motion_y>>1) | (motion_y&1);
784
785 dxy = 0;
786 if ((mx & 3) != 0)
787 dxy |= 1;
788 if ((my & 3) != 0)
789 dxy |= 2;
790 mx = mx >> 2;
791 my = my >> 2;
792
793 src_x = s->mb_x * 8 + mx;
794 src_y = s->mb_y * (8 >> field_based) + my;
795 src_x = clip(src_x, -8, s->width >> 1);
796 if (src_x == (s->width >> 1))
797 dxy &= ~1;
798 src_y = clip(src_y, -8, height >> 1);
799 if (src_y == (height >> 1))
800 dxy &= ~2;
801
802 offset = (src_y * (linesize >> 1)) + src_x + (src_offset >> 1);
803 ptr = ref_picture[1] + offset;
804 pix_op[dxy](dest_cb + (dest_offset >> 1), ptr, linesize >> 1, h >> 1);
805 ptr = ref_picture[2] + offset;
806 pix_op[dxy](dest_cr + (dest_offset >> 1), ptr, linesize >> 1, h >> 1);
807 }
808
809
810 static inline void MPV_motion(MpegEncContext *s,
811 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr,
812 int dir, UINT8 **ref_picture,
813 op_pixels_func *pix_op, qpel_mc_func *qpix_op)
814 {
815 int dxy, offset, mx, my, src_x, src_y, motion_x, motion_y;
816 int mb_x, mb_y, i;
817 UINT8 *ptr, *dest;
818
819 mb_x = s->mb_x;
820 mb_y = s->mb_y;
821
822 switch(s->mv_type) {
823 case MV_TYPE_16X16:
824 if(s->mcsel){
825 #if 0
826 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
827 ref_picture, 0,
828 0, pix_op,
829 s->sprite_offset[0][0]>>3,
830 s->sprite_offset[0][1]>>3,
831 16);
832 #else
833 gmc1_motion(s, dest_y, dest_cb, dest_cr, 0,
834 ref_picture, 0,
835 16);
836 #endif
837 }else if(s->quarter_sample && dir==0){ //FIXME
838 qpel_motion(s, dest_y, dest_cb, dest_cr, 0,
839 ref_picture, 0,
840 0, pix_op, qpix_op,
841 s->mv[dir][0][0], s->mv[dir][0][1], 16);
842 }else{
843 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
844 ref_picture, 0,
845 0, pix_op,
846 s->mv[dir][0][0], s->mv[dir][0][1], 16);
847 }
848 break;
849 case MV_TYPE_8X8:
850 for(i=0;i<4;i++) {
851 motion_x = s->mv[dir][i][0];
852 motion_y = s->mv[dir][i][1];
853
854 dxy = ((motion_y & 1) << 1) | (motion_x & 1);
855 src_x = mb_x * 16 + (motion_x >> 1) + (i & 1) * 8;
856 src_y = mb_y * 16 + (motion_y >> 1) + (i >>1) * 8;
857
858 /* WARNING: do no forget half pels */
859 src_x = clip(src_x, -16, s->width);
860 if (src_x == s->width)
861 dxy &= ~1;
862 src_y = clip(src_y, -16, s->height);
863 if (src_y == s->height)
864 dxy &= ~2;
865
866 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x);
867 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize;
868 pix_op[dxy](dest, ptr, s->linesize, 8);
869 }
870 /* In case of 8X8, we construct a single chroma motion vector
871 with a special rounding */
872 mx = 0;
873 my = 0;
874 for(i=0;i<4;i++) {
875 mx += s->mv[dir][i][0];
876 my += s->mv[dir][i][1];
877 }
878 if (mx >= 0)
879 mx = (h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1));
880 else {
881 mx = -mx;
882 mx = -(h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1));
883 }
884 if (my >= 0)
885 my = (h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1));
886 else {
887 my = -my;
888 my = -(h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1));
889 }
890 dxy = ((my & 1) << 1) | (mx & 1);
891 mx >>= 1;
892 my >>= 1;
893
894 src_x = mb_x * 8 + mx;
895 src_y = mb_y * 8 + my;
896 src_x = clip(src_x, -8, s->width/2);
897 if (src_x == s->width/2)
898 dxy &= ~1;
899 src_y = clip(src_y, -8, s->height/2);
900 if (src_y == s->height/2)
901 dxy &= ~2;
902
903 offset = (src_y * (s->linesize >> 1)) + src_x;
904 ptr = ref_picture[1] + offset;
905 pix_op[dxy](dest_cb, ptr, s->linesize >> 1, 8);
906 ptr = ref_picture[2] + offset;
907 pix_op[dxy](dest_cr, ptr, s->linesize >> 1, 8);
908 break;
909 case MV_TYPE_FIELD:
910 if (s->picture_structure == PICT_FRAME) {
911 /* top field */
912 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
913 ref_picture, s->field_select[dir][0] ? s->linesize : 0,
914 1, pix_op,
915 s->mv[dir][0][0], s->mv[dir][0][1], 8);
916 /* bottom field */
917 mpeg_motion(s, dest_y, dest_cb, dest_cr, s->linesize,
918 ref_picture, s->field_select[dir][1] ? s->linesize : 0,
919 1, pix_op,
920 s->mv[dir][1][0], s->mv[dir][1][1], 8);
921 } else {
922
923
924 }
925 break;
926 }
927 }
928
929
930 /* put block[] to dest[] */
931 static inline void put_dct(MpegEncContext *s,
932 DCTELEM *block, int i, UINT8 *dest, int line_size)
933 {
934 if (!s->mpeg2)
935 s->dct_unquantize(s, block, i, s->qscale);
936 ff_idct (block);
937 put_pixels_clamped(block, dest, line_size);
938 }
939
940 /* add block[] to dest[] */
941 static inline void add_dct(MpegEncContext *s,
942 DCTELEM *block, int i, UINT8 *dest, int line_size)
943 {
944 if (s->block_last_index[i] >= 0) {
945 if (!s->mpeg2)
946 if(s->encoding || (!s->h263_msmpeg4))
947 s->dct_unquantize(s, block, i, s->qscale);
948 ff_idct (block);
949 add_pixels_clamped(block, dest, line_size);
950 }
951 }
952
953 /* generic function called after a macroblock has been parsed by the
954 decoder or after it has been encoded by the encoder.
955
956 Important variables used:
957 s->mb_intra : true if intra macroblock
958 s->mv_dir : motion vector direction
959 s->mv_type : motion vector type
960 s->mv : motion vector
961 s->interlaced_dct : true if interlaced dct used (mpeg2)
962 */
963 void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64])
964 {
965 int mb_x, mb_y;
966 int dct_linesize, dct_offset;
967 op_pixels_func *op_pix;
968 qpel_mc_func *op_qpix;
969
970 mb_x = s->mb_x;
971 mb_y = s->mb_y;
972
973 #ifdef FF_POSTPROCESS
974 quant_store[mb_y][mb_x]=s->qscale;
975 //printf("[%02d][%02d] %d\n",mb_x,mb_y,s->qscale);
976 #endif
977
978 /* update DC predictors for P macroblocks */
979 if (!s->mb_intra) {
980 if (s->h263_pred || s->h263_aic) {
981 if(s->mbintra_table[mb_x + mb_y*s->mb_width])
982 {
983 int wrap, xy, v;
984 s->mbintra_table[mb_x + mb_y*s->mb_width]=0;
985 wrap = 2 * s->mb_width + 2;
986 xy = 2 * mb_x + 1 + (2 * mb_y + 1) * wrap;
987 v = 1024;
988
989 s->dc_val[0][xy] = v;
990 s->dc_val[0][xy + 1] = v;
991 s->dc_val[0][xy + wrap] = v;
992 s->dc_val[0][xy + 1 + wrap] = v;
993 /* ac pred */
994 memset(s->ac_val[0][xy], 0, 16 * sizeof(INT16));
995 memset(s->ac_val[0][xy + 1], 0, 16 * sizeof(INT16));
996 memset(s->ac_val[0][xy + wrap], 0, 16 * sizeof(INT16));
997 memset(s->ac_val[0][xy + 1 + wrap], 0, 16 * sizeof(INT16));
998 if (s->h263_msmpeg4) {
999 s->coded_block[xy] = 0;
1000 s->coded_block[xy + 1] = 0;
1001 s->coded_block[xy + wrap] = 0;
1002 s->coded_block[xy + 1 + wrap] = 0;
1003 }
1004 /* chroma */
1005 wrap = s->mb_width + 2;
1006 xy = mb_x + 1 + (mb_y + 1) * wrap;
1007 s->dc_val[1][xy] = v;
1008 s->dc_val[2][xy] = v;
1009 /* ac pred */
1010 memset(s->ac_val[1][xy], 0, 16 * sizeof(INT16));
1011 memset(s->ac_val[2][xy], 0, 16 * sizeof(INT16));
1012 }
1013 } else {
1014 s->last_dc[0] = 128 << s->intra_dc_precision;
1015 s->last_dc[1] = 128 << s->intra_dc_precision;
1016 s->last_dc[2] = 128 << s->intra_dc_precision;
1017 }
1018 }
1019 else if (s->h263_pred || s->h263_aic)
1020 s->mbintra_table[mb_x + mb_y*s->mb_width]=1;
1021
1022 /* update motion predictor, not for B-frames as they need the motion_val from the last P/S-Frame */
1023 if (s->out_format == FMT_H263) {
1024 if(s->pict_type!=B_TYPE){
1025 int xy, wrap, motion_x, motion_y;
1026
1027 wrap = 2 * s->mb_width + 2;
1028 xy = 2 * mb_x + 1 + (2 * mb_y + 1) * wrap;
1029 if (s->mb_intra) {
1030 motion_x = 0;
1031 motion_y = 0;
1032 goto motion_init;
1033 } else if (s->mv_type == MV_TYPE_16X16) {
1034 motion_x = s->mv[0][0][0];
1035 motion_y = s->mv[0][0][1];
1036 motion_init:
1037 /* no update if 8X8 because it has been done during parsing */
1038 s->motion_val[xy][0] = motion_x;
1039 s->motion_val[xy][1] = motion_y;
1040 s->motion_val[xy + 1][0] = motion_x;
1041 s->motion_val[xy + 1][1] = motion_y;
1042 s->motion_val[xy + wrap][0] = motion_x;
1043 s->motion_val[xy + wrap][1] = motion_y;
1044 s->motion_val[xy + 1 + wrap][0] = motion_x;
1045 s->motion_val[xy + 1 + wrap][1] = motion_y;
1046 }
1047 }
1048 }
1049
1050 if (!s->intra_only) {
1051 UINT8 *dest_y, *dest_cb, *dest_cr;
1052 UINT8 *mbskip_ptr;
1053
1054 /* avoid copy if macroblock skipped in last frame too */
1055 if (!s->encoding && s->pict_type != B_TYPE) {
1056 mbskip_ptr = &s->mbskip_table[s->mb_y * s->mb_width + s->mb_x];
1057 if (s->mb_skiped) {
1058 s->mb_skiped = 0;
1059 /* if previous was skipped too, then nothing to do ! */
1060 if (*mbskip_ptr != 0)
1061 goto the_end;
1062 *mbskip_ptr = 1; /* indicate that this time we skiped it */
1063 } else {
1064 *mbskip_ptr = 0; /* not skipped */
1065 }
1066 }
1067
1068 dest_y = s->current_picture[0] + (mb_y * 16 * s->linesize) + mb_x * 16;
1069 dest_cb = s->current_picture[1] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8;
1070 dest_cr = s->current_picture[2] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8;
1071
1072 if (s->interlaced_dct) {
1073 dct_linesize = s->linesize * 2;
1074 dct_offset = s->linesize;
1075 } else {
1076 dct_linesize = s->linesize;
1077 dct_offset = s->linesize * 8;
1078 }
1079
1080 if (!s->mb_intra) {
1081 /* motion handling */
1082 if (!s->no_rounding){
1083 op_pix = put_pixels_tab;
1084 op_qpix= qpel_mc_rnd_tab;
1085 }else{
1086 op_pix = put_no_rnd_pixels_tab;
1087 op_qpix= qpel_mc_no_rnd_tab;
1088 }
1089
1090 if (s->mv_dir & MV_DIR_FORWARD) {
1091 MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture, op_pix, op_qpix);
1092 if (!s->no_rounding)
1093 op_pix = avg_pixels_tab;
1094 else
1095 op_pix = avg_no_rnd_pixels_tab;
1096 }
1097 if (s->mv_dir & MV_DIR_BACKWARD) {
1098 MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture, op_pix, op_qpix);
1099 }
1100
1101 /* add dct residue */
1102 add_dct(s, block[0], 0, dest_y, dct_linesize);
1103 add_dct(s, block[1], 1, dest_y + 8, dct_linesize);
1104 add_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize);
1105 add_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize);
1106
1107 add_dct(s, block[4], 4, dest_cb, s->linesize >> 1);
1108 add_dct(s, block[5], 5, dest_cr, s->linesize >> 1);
1109 } else {
1110 /* dct only in intra block */
1111 put_dct(s, block[0], 0, dest_y, dct_linesize);
1112 put_dct(s, block[1], 1, dest_y + 8, dct_linesize);
1113 put_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize);
1114 put_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize);
1115
1116 put_dct(s, block[4], 4, dest_cb, s->linesize >> 1);
1117 put_dct(s, block[5], 5, dest_cr, s->linesize >> 1);
1118 }
1119 }
1120 the_end:
1121 emms_c(); //FIXME remove
1122 }
1123
1124 static void encode_mb(MpegEncContext *s)
1125 {
1126 int wrap;
1127 const int mb_x= s->mb_x;
1128 const int mb_y= s->mb_y;
1129 UINT8 *ptr;
1130 const int motion_x= s->mv[0][0][0];
1131 const int motion_y= s->mv[0][0][1];
1132 int i;
1133
1134 /* get the pixels */
1135 wrap = s->linesize;
1136 ptr = s->new_picture[0] + (mb_y * 16 * wrap) + mb_x * 16;
1137 get_pixels(s->block[0], ptr, wrap);
1138 get_pixels(s->block[1], ptr + 8, wrap);
1139 get_pixels(s->block[2], ptr + 8 * wrap, wrap);
1140 get_pixels(s->block[3], ptr + 8 * wrap + 8, wrap);
1141 wrap = s->linesize >> 1;
1142 ptr = s->new_picture[1] + (mb_y * 8 * wrap) + mb_x * 8;
1143 get_pixels(s->block[4], ptr, wrap);
1144
1145 wrap = s->linesize >> 1;
1146 ptr = s->new_picture[2] + (mb_y * 8 * wrap) + mb_x * 8;
1147 get_pixels(s->block[5], ptr, wrap);
1148
1149 /* subtract previous frame if non intra */
1150 if (!s->mb_intra) {
1151 int dxy, offset, mx, my;
1152
1153 if(s->mv_type==MV_TYPE_16X16){
1154 dxy = ((motion_y & 1) << 1) | (motion_x & 1);
1155 ptr = s->last_picture[0] +
1156 ((mb_y * 16 + (motion_y >> 1)) * s->linesize) +
1157 (mb_x * 16 + (motion_x >> 1));
1158
1159 sub_pixels_2(s->block[0], ptr, s->linesize, dxy);
1160 sub_pixels_2(s->block[1], ptr + 8, s->linesize, dxy);
1161 sub_pixels_2(s->block[2], ptr + s->linesize * 8, s->linesize, dxy);
1162 sub_pixels_2(s->block[3], ptr + 8 + s->linesize * 8, s->linesize ,dxy);
1163
1164 if (s->out_format == FMT_H263) {
1165 /* special rounding for h263 */
1166 dxy = 0;
1167 if ((motion_x & 3) != 0)
1168 dxy |= 1;
1169 if ((motion_y & 3) != 0)
1170 dxy |= 2;
1171 mx = motion_x >> 2;
1172 my = motion_y >> 2;
1173 } else {
1174 mx = motion_x / 2;
1175 my = motion_y / 2;
1176 dxy = ((my & 1) << 1) | (mx & 1);
1177 mx >>= 1;
1178 my >>= 1;
1179 }
1180 offset = ((mb_y * 8 + my) * (s->linesize >> 1)) + (mb_x * 8 + mx);
1181 ptr = s->last_picture[1] + offset;
1182 sub_pixels_2(s->block[4], ptr, s->linesize >> 1, dxy);
1183 ptr = s->last_picture[2] + offset;
1184 sub_pixels_2(s->block[5], ptr, s->linesize >> 1, dxy);
1185 }else{
1186 int src_x, src_y;
1187
1188 for(i=0;i<4;i++) {
1189 int motion_x = s->mv[0][i][0];
1190 int motion_y = s->mv[0][i][1];
1191
1192 dxy = ((motion_y & 1) << 1) | (motion_x & 1);
1193 src_x = mb_x * 16 + (motion_x >> 1) + (i & 1) * 8;
1194 src_y = mb_y * 16 + (motion_y >> 1) + (i >>1) * 8;
1195
1196 ptr = s->last_picture[0] + (src_y * s->linesize) + (src_x);
1197 sub_pixels_2(s->block[i], ptr, s->linesize, dxy);
1198 }
1199 /* In case of 8X8, we construct a single chroma motion vector
1200 with a special rounding */
1201 mx = 0;
1202 my = 0;
1203 for(i=0;i<4;i++) {
1204 mx += s->mv[0][i][0];
1205 my += s->mv[0][i][1];
1206 }
1207 if (mx >= 0)
1208 mx = (h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1));
1209 else {
1210 mx = -mx;
1211 mx = -(h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1));
1212 }
1213 if (my >= 0)
1214 my = (h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1));
1215 else {
1216 my = -my;
1217 my = -(h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1));
1218 }
1219 dxy = ((my & 1) << 1) | (mx & 1);
1220 mx >>= 1;
1221 my >>= 1;
1222
1223 src_x = mb_x * 8 + mx;
1224 src_y = mb_y * 8 + my;
1225 src_x = clip(src_x, -8, s->width/2);
1226 if (src_x == s->width/2)
1227 dxy &= ~1;
1228 src_y = clip(src_y, -8, s->height/2);
1229 if (src_y == s->height/2)
1230 dxy &= ~2;
1231
1232 offset = (src_y * (s->linesize >> 1)) + src_x;
1233 ptr = s->last_picture[1] + offset;
1234 sub_pixels_2(s->block[4], ptr, s->linesize >> 1, dxy);
1235 ptr = s->last_picture[2] + offset;
1236 sub_pixels_2(s->block[5], ptr, s->linesize >> 1, dxy);
1237 }
1238 }
1239
1240 #if 0
1241 {
1242 float adap_parm;
1243
1244 adap_parm = ((s->avg_mb_var << 1) + s->mb_var[s->mb_width*mb_y+mb_x] + 1.0) /
1245 ((s->mb_var[s->mb_width*mb_y+mb_x] << 1) + s->avg_mb_var + 1.0);
1246
1247 printf("\ntype=%c qscale=%2d adap=%0.2f dquant=%4.2f var=%4d avgvar=%4d",
1248 (s->mb_type[s->mb_width*mb_y+mb_x] > 0) ? 'I' : 'P',
1249 s->qscale, adap_parm, s->qscale*adap_parm,
1250 s->mb_var[s->mb_width*mb_y+mb_x], s->avg_mb_var);
1251 }
1252 #endif
1253 /* DCT & quantize */
1254 if (s->h263_msmpeg4) {
1255 msmpeg4_dc_scale(s);
1256 } else if (s->h263_pred) {
1257 h263_dc_scale(s);
1258 } else {
1259 /* default quantization values */
1260 s->y_dc_scale = 8;
1261 s->c_dc_scale = 8;
1262 }
1263 for(i=0;i<6;i++) {
1264 s->block_last_index[i] = dct_quantize(s, s->block[i], i, s->qscale);
1265 }
1266
1267 /* huffman encode */
1268 switch(s->out_format) {
1269 case FMT_MPEG1:
1270 mpeg1_encode_mb(s, s->block, motion_x, motion_y);
1271 break;
1272 case FMT_H263:
1273 if (s->h263_msmpeg4)
1274 msmpeg4_encode_mb(s, s->block, motion_x, motion_y);
1275 else if(s->h263_pred)
1276 mpeg4_encode_mb(s, s->block, motion_x, motion_y);
1277 else
1278 h263_encode_mb(s, s->block, motion_x, motion_y);
1279 break;
1280 case FMT_MJPEG:
1281 mjpeg_encode_mb(s, s->block);
1282 break;
1283 }
1284 }
1285
1286 static void copy_bits(PutBitContext *pb, UINT8 *src, int length)
1287 {
1288 int bytes= length>>3;
1289 int bits= length&7;
1290 int i;
1291
1292 for(i=0; i<bytes; i++) put_bits(pb, 8, src[i]);
1293 put_bits(pb, bits, src[i]>>(8-bits));
1294 }
1295
1296 static void encode_picture(MpegEncContext *s, int picture_number)
1297 {
1298 int mb_x, mb_y, last_gob, pdif = 0;
1299 int i;
1300 int bits;
1301 MpegEncContext best_s;
1302 UINT8 bit_buf[4][3000]; //FIXME check that this is ALLWAYS large enogh for a MB
1303
1304 s->picture_number = picture_number;
1305
1306 s->block_wrap[0]=
1307 s->block_wrap[1]=
1308 s->block_wrap[2]=
1309 s->block_wrap[3]= s->mb_width*2 + 2;
1310 s->block_wrap[4]=
1311 s->block_wrap[5]= s->mb_width + 2;
1312
1313 s->last_mc_mb_var = s->mc_mb_var;
1314 /* Reset the average MB variance */
1315 s->avg_mb_var = 0;
1316 s->mc_mb_var = 0;
1317 /* Estimate motion for every MB */
1318 if(s->pict_type == P_TYPE){
1319 for(mb_y=0; mb_y < s->mb_height; mb_y++) {
1320 s->block_index[0]= s->block_wrap[0]*(mb_y*2 + 1) - 1;
1321 s->block_index[1]= s->block_wrap[0]*(mb_y*2 + 1);
1322 s->block_index[2]= s->block_wrap[0]*(mb_y*2 + 2) - 1;
1323 s->block_index[3]= s->block_wrap[0]*(mb_y*2 + 2);
1324 for(mb_x=0; mb_x < s->mb_width; mb_x++) {
1325 s->mb_x = mb_x;
1326 s->mb_y = mb_y;
1327 s->block_index[0]+=2;
1328 s->block_index[1]+=2;
1329 s->block_index[2]+=2;
1330 s->block_index[3]+=2;
1331
1332 /* compute motion vector & mb_type and store in context */
1333 estimate_motion(s, mb_x, mb_y);
1334 // s->mb_type[mb_y*s->mb_width + mb_x]=MB_TYPE_INTER;
1335 }
1336 }
1337 emms_c();
1338 }else{
1339 /* I-Frame */
1340 //FIXME do we need to zero them?
1341 memset(s->motion_val[0], 0, sizeof(INT16)*(s->mb_width*2 + 2)*(s->mb_height*2 + 2)*2);
1342 memset(s->mv_table[0] , 0, sizeof(INT16)*s->mb_width*s->mb_height);
1343 memset(s->mv_table[1] , 0, sizeof(INT16)*s->mb_width*s->mb_height);
1344 memset(s->mb_type , MB_TYPE_INTRA, sizeof(UINT8)*s->mb_width*s->mb_height);
1345 }
1346
1347 if(s->avg_mb_var < s->mc_mb_var && s->pict_type != B_TYPE && (!s->force_type)){ //FIXME subtract MV bits
1348 s->pict_type= I_TYPE;
1349 s->picture_in_gop_number=0;
1350 memset(s->mb_type , MB_TYPE_INTRA, sizeof(UINT8)*s->mb_width*s->mb_height);
1351 //printf("Scene change detected, encoding as I Frame\n");
1352 }
1353
1354 /* find best f_code for ME which do unlimited searches */
1355 if(s->pict_type == P_TYPE && s->me_method >= 5){
1356 int mv_num[8];
1357 int i;
1358 int loose=0;
1359 UINT8 * fcode_tab= s->fcode_tab;
1360
1361 for(i=0; i<8; i++) mv_num[i]=0;
1362
1363 for(i=0; i<s->mb_num; i++){
1364 if(s->mb_type[i] & MB_TYPE_INTER){
1365 mv_num[ fcode_tab[s->mv_table[0][i] + MAX_MV] ]++;
1366 mv_num[ fcode_tab[s->mv_table[1][i] + MAX_MV] ]++;
1367 //printf("%d %d %d\n", s->mv_table[0][i], fcode_tab[s->mv_table[0][i] + MAX_MV], i);
1368 }
1369 //else printf("I");
1370 }
1371
1372 for(i=MAX_FCODE; i>1; i--){
1373 loose+= mv_num[i];
1374 if(loose > 10) break; //FIXME this is pretty ineffective
1375 }
1376 s->f_code= i;
1377 /* for(i=0; i<=MAX_FCODE; i++){
1378 printf("%d ", mv_num[i]);
1379 }
1380 printf("\n");*/
1381 }else{
1382 s->f_code= 1;
1383 }
1384
1385 //printf("f_code %d ///\n", s->f_code);
1386 /* convert MBs with too long MVs to I-Blocks */
1387 if(s->pict_type==P_TYPE){
1388 int i, x, y;
1389 const int f_code= s->f_code;
1390 UINT8 * fcode_tab= s->fcode_tab;
1391 //FIXME try to clip instead of intra izing ;)
1392 /* clip / convert to intra 16x16 type MVs */
1393 for(i=0; i<s->mb_num; i++){
1394 if(s->mb_type[i]&MB_TYPE_INTER){
1395 if( fcode_tab[s->mv_table[0][i] + MAX_MV] > f_code
1396 || fcode_tab[s->mv_table[0][i] + MAX_MV] == 0
1397 || fcode_tab[s->mv_table[1][i] + MAX_MV] > f_code
1398 || fcode_tab[s->mv_table[1][i] + MAX_MV] == 0 ){
1399 s->mb_type[i] &= ~MB_TYPE_INTER;
1400 s->mb_type[i] |= MB_TYPE_INTRA;
1401 s->mv_table[0][i] = 0;
1402 s->mv_table[1][i] = 0;
1403 }
1404 }
1405 }
1406
1407 if(s->flags&CODEC_FLAG_4MV){
1408 int wrap= 2+ s->mb_width*2;
1409
1410 /* clip / convert to intra 8x8 type MVs */
1411 for(y=0; y<s->mb_height; y++){
1412 int xy= (y*2 + 1)*wrap + 1;
1413 i= y*s->mb_width;
1414
1415 for(x=0; x<s->mb_width; x++){
1416 if(s->mb_type[i]&MB_TYPE_INTER4V){
1417 int block;
1418 for(block=0; block<4; block++){
1419 int off= (block& 1) + (block>>1)*wrap;
1420 int mx= s->motion_val[ xy + off ][0];
1421 int my= s->motion_val[ xy + off ][1];
1422
1423 if( fcode_tab[mx + MAX_MV] > f_code
1424 || fcode_tab[mx + MAX_MV] == 0
1425 || fcode_tab[my + MAX_MV] > f_code
1426 || fcode_tab[my + MAX_MV] == 0 ){
1427 s->mb_type[i] &= ~MB_TYPE_INTER4V;
1428 s->mb_type[i] |= MB_TYPE_INTRA;
1429 }
1430 }
1431 xy+=2;
1432 i++;
1433 }
1434 }
1435 }
1436 }
1437 }
1438
1439 // printf("%d %d\n", s->avg_mb_var, s->mc_mb_var);
1440
1441 if (!s->fixed_qscale)
1442 s->qscale = rate_estimate_qscale(s);
1443
1444 /* precompute matrix */
1445 if (s->out_format == FMT_MJPEG) {
1446 /* for mjpeg, we do include qscale in the matrix */
1447 s->intra_matrix[0] = default_intra_matrix[0];
1448 for(i=1;i<64;i++)
1449 s->intra_matrix[i] = (default_intra_matrix[i] * s->qscale) >> 3;
1450 convert_matrix(s->q_intra_matrix, s->q_intra_matrix16, s->intra_matrix, 8);
1451 } else {
1452 convert_matrix(s->q_intra_matrix, s->q_intra_matrix16, s->intra_matrix, s->qscale);
1453 convert_matrix(s->q_non_intra_matrix, s->q_non_intra_matrix16, s->non_intra_matrix, s->qscale);
1454 }
1455
1456 s->last_bits= get_bit_count(&s->pb);
1457 switch(s->out_format) {
1458 case FMT_MJPEG:
1459 mjpeg_picture_header(s);
1460 break;
1461 case FMT_H263:
1462 if (s->h263_msmpeg4)
1463 msmpeg4_encode_picture_header(s, picture_number);
1464 else if (s->h263_pred)
1465 mpeg4_encode_picture_header(s, picture_number);
1466 else if (s->h263_rv10)
1467 rv10_encode_picture_header(s, picture_number);
1468 else
1469 h263_encode_picture_header(s, picture_number);
1470 break;
1471 case FMT_MPEG1:
1472 mpeg1_encode_picture_header(s, picture_number);
1473 break;
1474 }
1475 bits= get_bit_count(&s->pb);
1476 s->header_bits= bits - s->last_bits;
1477 s->last_bits= bits;
1478 s->mv_bits=0;
1479 s->misc_bits=0;
1480 s->i_tex_bits=0;
1481 s->p_tex_bits=0;
1482 s->i_count=0;
1483 s->p_count=0;
1484 s->skip_count=0;
1485
1486 /* init last dc values */
1487 /* note: quant matrix value (8) is implied here */
1488 s->last_dc[0] = 128;
1489 s->last_dc[1] = 128;
1490 s->last_dc[2] = 128;
1491 s->mb_incr = 1;
1492 s->last_mv[0][0][0] = 0;
1493 s->last_mv[0][0][1] = 0;
1494
1495 /* Get the GOB height based on picture height */
1496 if (s->out_format == FMT_H263 && !s->h263_pred && !s->h263_msmpeg4) {
1497 if (s->height <= 400)
1498 s->gob_index = 1;
1499 else if (s->height <= 800)
1500 s->gob_index = 2;
1501 else
1502 s->gob_index = 4;
1503 }
1504
1505 s->avg_mb_var = s->avg_mb_var / s->mb_num;
1506
1507 for(mb_y=0; mb_y < s->mb_height; mb_y++) {
1508 /* Put GOB header based on RTP MTU */
1509 /* TODO: Put all this stuff in a separate generic function */
1510 if (s->rtp_mode) {
1511 if (!mb_y) {
1512 s->ptr_lastgob = s->pb.buf;
1513 s->ptr_last_mb_line = s->pb.buf;
1514 } else if (s->out_format == FMT_H263 && !s->h263_pred && !s->h263_msmpeg4 && !(mb_y % s->gob_index)) {
1515 last_gob = h263_encode_gob_header(s, mb_y);
1516 if (last_gob) {
1517 s->first_gob_line = 1;
1518 }
1519 }
1520 }
1521
1522 s->block_index[0]= s->block_wrap[0]*(mb_y*2 + 1) - 1;
1523 s->block_index[1]= s->block_wrap[0]*(mb_y*2 + 1);
1524 s->block_index[2]= s->block_wrap[0]*(mb_y*2 + 2) - 1;
1525 s->block_index[3]= s->block_wrap[0]*(mb_y*2 + 2);
1526 s->block_index[4]= s->block_wrap[4]*(mb_y + 1) + s->block_wrap[0]*(s->mb_height*2 + 2);
1527 s->block_index[5]= s->block_wrap[4]*(mb_y + 1 + s->mb_height + 2) + s->block_wrap[0]*(s->mb_height*2 + 2);
1528 for(mb_x=0; mb_x < s->mb_width; mb_x++) {
1529 const int mb_type= s->mb_type[mb_y * s->mb_width + mb_x];
1530 PutBitContext pb;
1531 int d;
1532 int dmin=10000000;
1533 int best=0;
1534
1535 s->mb_x = mb_x;
1536 s->mb_y = mb_y;
1537 s->block_index[0]+=2;
1538 s->block_index[1]+=2;
1539 s->block_index[2]+=2;
1540 s->block_index[3]+=2;
1541 s->block_index[4]++;
1542 s->block_index[5]++;
1543
1544 s->mv_dir = MV_DIR_FORWARD;
1545 if(mb_type & (mb_type-1)){ // more than 1 MB type possible
1546 pb= s->pb;
1547 if(mb_type&MB_TYPE_INTER){
1548 s->mv_type = MV_TYPE_16X16;
1549 s->mb_intra= 0;
1550 s->mv[0][0][0] = s->mv_table[0][mb_y * s->mb_width + mb_x];
1551 s->mv[0][0][1] = s->mv_table[1][mb_y * s->mb_width + mb_x];
1552 init_put_bits(&s->pb, bit_buf[1], 3000, NULL, NULL);
1553 s->block= s->inter_block;
1554
1555 encode_mb(s);
1556 d= get_bit_count(&s->pb);
1557 if(d<dmin){
1558 flush_put_bits(&s->pb);
1559 dmin=d;
1560 best_s.mv[0][0][0]= s->mv[0][0][0];
1561 best_s.mv[0][0][1]= s->mv[0][0][1];
1562 best_s.mb_intra= 0;
1563 best_s.mv_type = MV_TYPE_16X16;
1564 best_s.pb=s->pb;
1565 best_s.block= s->block;
1566 best=1;
1567 for(i=0; i<6; i++)
1568 best_s.block_last_index[i]= s->block_last_index[i];
1569 }
1570 }
1571 if(mb_type&MB_TYPE_INTER4V){
1572 s->mv_type = MV_TYPE_8X8;
1573 s->mb_intra= 0;
1574 for(i=0; i<4; i++){
1575 s->mv[0][i][0] = s->motion_val[s->block_index[i]][0];
1576 s->mv[0][i][1] = s->motion_val[s->block_index[i]][1];
1577 }
1578 init_put_bits(&s->pb, bit_buf[2], 3000, NULL, NULL);
1579 s->block= s->inter4v_block;
1580
1581 encode_mb(s);
1582 d= get_bit_count(&s->pb);
1583 if(d<dmin){
1584 flush_put_bits(&s->pb);
1585 dmin=d;
1586 for(i=0; i<4; i++){
1587 best_s.mv[0][i][0] = s->mv[0][i][0];
1588 best_s.mv[0][i][1] = s->mv[0][i][1];
1589 }
1590 best_s.mb_intra= 0;
1591 best_s.mv_type = MV_TYPE_8X8;
1592 best_s.pb=s->pb;
1593 best_s.block= s->block;
1594 best=2;
1595 for(i=0; i<6; i++)
1596 best_s.block_last_index[i]= s->block_last_index[i];
1597 }
1598 }
1599 if(mb_type&MB_TYPE_INTRA){
1600 s->mv_type = MV_TYPE_16X16;
1601 s->mb_intra= 1;
1602 s->mv[0][0][0] = 0;
1603 s->mv[0][0][1] = 0;
1604 init_put_bits(&s->pb, bit_buf[0], 3000, NULL, NULL);
1605 s->block= s->intra_block;
1606
1607 encode_mb(s);
1608 d= get_bit_count(&s->pb);
1609 if(d<dmin){
1610 flush_put_bits(&s->pb);
1611 dmin=d;
1612 best_s.mv[0][0][0]= 0;
1613 best_s.mv[0][0][1]= 0;
1614 best_s.mb_intra= 1;
1615 best_s.mv_type = MV_TYPE_16X16;
1616 best_s.pb=s->pb;
1617 best_s.block= s->block;
1618 for(i=0; i<6; i++)
1619 best_s.block_last_index[i]= s->block_last_index[i];
1620 best=0;
1621 }
1622 /* force cleaning of ac/dc if needed ... */
1623 s->mbintra_table[mb_x + mb_y*s->mb_width]=1;
1624 }
1625 for(i=0; i<4; i++){
1626 s->mv[0][i][0] = best_s.mv[0][i][0];
1627 s->mv[0][i][1] = best_s.mv[0][i][1];
1628 }
1629 s->mb_intra= best_s.mb_intra;
1630 s->mv_type= best_s.mv_type;
1631 for(i=0; i<6; i++)
1632 s->block_last_index[i]= best_s.block_last_index[i];
1633 copy_bits(&pb, bit_buf[best], dmin);
1634 s->block= best_s.block;
1635 s->pb= pb;
1636 } else {
1637 // only one MB-Type possible
1638 if(mb_type&MB_TYPE_INTRA){
1639 s->mb_intra= 1;
1640 s->mv[0][0][0] = 0;
1641 s->mv[0][0][1] = 0;
1642 }else{
1643 s->mb_intra= 0;
1644 s->mv[0][0][0] = s->mv_table[0][mb_y * s->mb_width + mb_x];
1645 s->mv[0][0][1] = s->mv_table[1][mb_y * s->mb_width + mb_x];
1646 }
1647 encode_mb(s);
1648 }
1649
1650 MPV_decode_mb(s, s->block);
1651 }
1652
1653
1654 /* Obtain average GOB size for RTP */
1655 if (s->rtp_mode) {
1656 if (!mb_y)
1657 s->mb_line_avgsize = pbBufPtr(&s->pb) - s->ptr_last_mb_line;
1658 else if (!(mb_y % s->gob_index)) {
1659 s->mb_line_avgsize = (s->mb_line_avgsize + pbBufPtr(&s->pb) - s->ptr_last_mb_line) >> 1;
1660 s->ptr_last_mb_line = pbBufPtr(&s->pb);
1661 }
1662 //fprintf(stderr, "\nMB line: %d\tSize: %u\tAvg. Size: %u", s->mb_y,
1663 // (s->pb.buf_ptr - s->ptr_last_mb_line), s->mb_line_avgsize);
1664 s->first_gob_line = 0;
1665 }
1666 }
1667 emms_c();
1668
1669 if (s->h263_msmpeg4 && s->msmpeg4_version<4 && s->pict_type == I_TYPE)
1670 msmpeg4_encode_ext_header(s);
1671
1672 //if (s->gob_number)
1673 // fprintf(stderr,"\nNumber of GOB: %d", s->gob_number);
1674
1675 /* Send the last GOB if RTP */
1676 if (s->rtp_mode) {
1677 flush_put_bits(&s->pb);
1678 pdif = pbBufPtr(&s->pb) - s->ptr_lastgob;
1679 /* Call the RTP callback to send the last GOB */
1680 if (s->rtp_callback)
1681 s->rtp_callback(s->ptr_lastgob, pdif, s->gob_number);
1682 s->ptr_lastgob = pbBufPtr(&s->pb);
1683 //fprintf(stderr,"\nGOB: %2d size: %d (last)", s->gob_number, pdif);
1684 }
1685 }
1686
1687 static int dct_quantize_c(MpegEncContext *s,
1688 DCTELEM *block, int n,
1689 int qscale)
1690 {
1691 int i, j, level, last_non_zero, q;
1692 const int *qmat;
1693 int minLevel, maxLevel;
1694
1695 if(s->avctx!=NULL && s->avctx->codec->id==CODEC_ID_MPEG4){
1696 /* mpeg4 */
1697 minLevel= -2048;
1698 maxLevel= 2047;
1699 }else if(s->out_format==FMT_MPEG1){
1700 /* mpeg1 */
1701 minLevel= -255;
1702 maxLevel= 255;
1703 }else if(s->out_format==FMT_MJPEG){
1704 /* (m)jpeg */
1705 minLevel= -1023;
1706 maxLevel= 1023;
1707 }else{
1708 /* h263 / msmpeg4 */
1709 minLevel= -128;
1710 maxLevel= 127;
1711 }
1712
1713 av_fdct (block);
1714
1715 /* we need this permutation so that we correct the IDCT
1716 permutation. will be moved into DCT code */
1717 block_permute(block);
1718
1719 if (s->mb_intra) {
1720 if (n < 4)
1721 q = s->y_dc_scale;
1722 else
1723 q = s->c_dc_scale;
1724 q = q << 3;
1725
1726 /* note: block[0] is assumed to be positive */
1727 block[0] = (block[0] + (q >> 1)) / q;
1728 i = 1;
1729 last_non_zero = 0;
1730 if (s->out_format == FMT_H263) {
1731 qmat = s->q_non_intra_matrix;
1732 } else {
1733 qmat = s->q_intra_matrix;
1734 }
1735 } else {
1736 i = 0;
1737 last_non_zero = -1;
1738 qmat = s->q_non_intra_matrix;
1739 }
1740
1741 for(;i<64;i++) {
1742 j = zigzag_direct[i];
1743 level = block[j];
1744 level = level * qmat[j];
1745 #ifdef PARANOID
1746 {
1747 static int count = 0;
1748 int level1, level2, qmat1;
1749 double val;
1750 if (qmat == s->q_non_intra_matrix) {
1751 qmat1 = default_non_intra_matrix[j] * s->qscale;
1752 } else {
1753 qmat1 = default_intra_matrix[j] * s->qscale;
1754 }
1755 if (av_fdct != jpeg_fdct_ifast)
1756 val = ((double)block[j] * 8.0) / (double)qmat1;
1757 else
1758 val = ((double)block[j] * 8.0 * 2048.0) /
1759 ((double)qmat1 * aanscales[j]);
1760 level1 = (int)val;
1761 level2 = level / (1 << (QMAT_SHIFT - 3));
1762 if (level1 != level2) {
1763 fprintf(stderr, "%d: quant error qlevel=%d wanted=%d level=%d qmat1=%d qmat=%d wantedf=%0.6f\n",
1764 count, level2, level1, block[j], qmat1, qmat[j],
1765 val);
1766 count++;
1767 }
1768
1769 }
1770 #endif
1771 /* XXX: slight error for the low range. Test should be equivalent to
1772 (level <= -(1 << (QMAT_SHIFT - 3)) || level >= (1 <<
1773 (QMAT_SHIFT - 3)))
1774 */
1775 if (((level << (31 - (QMAT_SHIFT - 3))) >> (31 - (QMAT_SHIFT - 3))) !=
1776 level) {
1777 level = level / (1 << (QMAT_SHIFT - 3));
1778 /* XXX: currently, this code is not optimal. the range should be:
1779 mpeg1: -255..255
1780 mpeg2: -2048..2047
1781 h263: -128..127
1782 mpeg4: -2048..2047
1783 */
1784 if (level > maxLevel)
1785 level = maxLevel;
1786 else if (level < minLevel)
1787 level = minLevel;
1788
1789 block[j] = level;
1790 last_non_zero = i;
1791 } else {
1792 block[j] = 0;
1793 }
1794 }
1795 return last_non_zero;
1796 }
1797
1798 static void dct_unquantize_mpeg1_c(MpegEncContext *s,
1799 DCTELEM *block, int n, int qscale)
1800 {
1801 int i, level, nCoeffs;
1802 const UINT16 *quant_matrix;
1803
1804 if(s->alternate_scan) nCoeffs= 64;
1805 else nCoeffs= s->block_last_index[n]+1;
1806
1807 if (s->mb_intra) {
1808 if (n < 4)
1809 block[0] = block[0] * s->y_dc_scale;
1810 else
1811 block[0] = block[0] * s->c_dc_scale;
1812 /* XXX: only mpeg1 */
1813 quant_matrix = s->intra_matrix;
1814 for(i=1;i<nCoeffs;i++) {
1815 int j= zigzag_direct[i];
1816 level = block[j];
1817 if (level) {
1818 if (level < 0) {
1819 level = -level;
1820 level = (int)(level * qscale * quant_matrix[j]) >> 3;
1821 level = (level - 1) | 1;
1822 level = -level;
1823 } else {
1824 level = (int)(level * qscale * quant_matrix[j]) >> 3;
1825 level = (level - 1) | 1;
1826 }
1827 #ifdef PARANOID
1828 if (level < -2048 || level > 2047)
1829 fprintf(stderr, "unquant error %d %d\n", i, level);
1830 #endif
1831 block[j] = level;
1832 }
1833 }
1834 } else {
1835 i = 0;
1836 quant_matrix = s->non_intra_matrix;
1837 for(;i<nCoeffs;i++) {
1838 int j= zigzag_direct[i];
1839 level = block[j];
1840 if (level) {
1841 if (level < 0) {
1842 level = -level;
1843 level = (((level << 1) + 1) * qscale *
1844 ((int) (quant_matrix[j]))) >> 4;
1845 level = (level - 1) | 1;
1846 level = -level;
1847 } else {
1848 level = (((level << 1) + 1) * qscale *
1849 ((int) (quant_matrix[j]))) >> 4;
1850 level = (level - 1) | 1;
1851 }
1852 #ifdef PARANOID
1853 if (level < -2048 || level > 2047)
1854 fprintf(stderr, "unquant error %d %d\n", i, level);
1855 #endif
1856 block[j] = level;
1857 }
1858 }
1859 }
1860 }
1861
1862 static void dct_unquantize_h263_c(MpegEncContext *s,
1863 DCTELEM *block, int n, int qscale)
1864 {
1865 int i, level, qmul, qadd;
1866 int nCoeffs;
1867
1868 if (s->mb_intra) {
1869 if (!s->h263_aic) {
1870 if (n < 4)
1871 block[0] = block[0] * s->y_dc_scale;
1872 else
1873 block[0] = block[0] * s->c_dc_scale;
1874 }
1875 i = 1;
1876 nCoeffs= 64; //does not allways use zigzag table
1877 } else {
1878 i = 0;
1879 nCoeffs= zigzag_end[ s->block_last_index[n] ];
1880 }
1881
1882 qmul = s->qscale << 1;
1883 if (s->h263_aic && s->mb_intra)
1884 qadd = 0;
1885 else
1886 qadd = (s->qscale - 1) | 1;
1887
1888 for(;i<nCoeffs;i++) {
1889 level = block[i];
1890 if (level) {
1891 if (level < 0) {
1892 level = level * qmul - qadd;
1893 } else {
1894 level = level * qmul + qadd;
1895 }
1896 #ifdef PARANOID
1897 if (level < -2048 || level > 2047)
1898 fprintf(stderr, "unquant error %d %d\n", i, level);
1899 #endif
1900 block[i] = level;
1901 }
1902 }
1903 }
1904
1905 /* rate control */
1906
1907 /* an I frame is I_FRAME_SIZE_RATIO bigger than a P frame */
1908 #define I_FRAME_SIZE_RATIO 3.0
1909 #define QSCALE_K 20
1910
1911 static void rate_control_init(MpegEncContext *s)
1912 {
1913 #if 1
1914 emms_c();
1915
1916 //initial values, they dont really matter as they will be totally different within a few frames
1917 s->i_pred.coeff= s->p_pred.coeff= 7.0;
1918 s->i_pred.count= s->p_pred.count= 1.0;
1919
1920 s->i_pred.decay= s->p_pred.decay= 0.4;
1921
1922 // use more bits at the beginning, otherwise high motion at the begin will look like shit
1923 s->qsum=100;
1924 s->qcount=100;
1925
1926 s->short_term_qsum=0.001;
1927 s->short_term_qcount=0.001;
1928 #else
1929 s->wanted_bits = 0;
1930
1931 if (s->intra_only) {
1932 s->I_frame_bits = ((INT64)s->bit_rate * FRAME_RATE_BASE) / s->frame_rate;
1933 s->P_frame_bits = s->I_frame_bits;
1934 } else {
1935 s->P_frame_bits = (int) ((float)(s->gop_size * s->bit_rate) /
1936 (float)((float)s->frame_rate / FRAME_RATE_BASE * (I_FRAME_SIZE_RATIO + s->gop_size - 1)));
1937 s->I_frame_bits = (int)(s->P_frame_bits * I_FRAME_SIZE_RATIO);
1938 }
1939
1940 #if defined(DEBUG)
1941 printf("I_frame_size=%d P_frame_size=%d\n",
1942 s->I_frame_bits, s->P_frame_bits);
1943 #endif
1944 #endif
1945 }
1946
1947 static double predict(Predictor *p, double q, double var)
1948 {
1949 return p->coeff*var / (q*p->count);
1950 }
1951
1952 static void update_predictor(Predictor *p, double q, double var, double size)
1953 {
1954 double new_coeff= size*q / (var + 1);
1955 if(var<1000) return;
1956 /*{
1957 int pred= predict(p, q, var);
1958 int error= abs(pred-size);
1959 static double sum=0;
1960 static int count=0;
1961 if(count>5) sum+=error;
1962 count++;
1963 if(256*256*256*64%count==0){
1964 printf("%d %f %f\n", count, sum/count, p->coeff);
1965 }
1966 }*/
1967 p->count*= p->decay;
1968 p->coeff*= p->decay;
1969 p->count++;
1970 p->coeff+= new_coeff;
1971 }
1972
1973 static int rate_estimate_qscale(MpegEncContext *s)
1974 {
1975 #if 1
1976 int qmin= s->qmin;
1977 int qmax= s->qmax;
1978 int rate_q=5;
1979 float q;
1980 int qscale;
1981 float br_compensation;
1982 double diff;
1983 double short_term_q;
1984 double long_term_q;
1985 int last_qscale= s->qscale;
1986 double fps;
1987 INT64 wanted_bits;
1988 emms_c();
1989
1990 fps= (double)s->frame_rate / FRAME_RATE_BASE;
1991 wanted_bits= s->bit_rate*(double)s->picture_number/fps;
1992
1993
1994 if(s->picture_number>2){
1995 /* update predictors */
1996 if(s->last_pict_type == I_TYPE){
1997 //FIXME
1998 }else{ //P Frame
1999 //printf("%d %d %d %f\n", s->qscale, s->last_mc_mb_var, s->frame_bits, s->p_pred.coeff);
2000 update_predictor(&s->p_pred, s->qscale, s->last_mc_mb_var, s->frame_bits);
2001 }
2002 }
2003
2004 if(s->pict_type == I_TYPE){
2005 //FIXME
2006 rate_q= s->qsum/s->qcount;
2007 }else{ //P Frame
2008 int i;
2009 int diff, best_diff=1000000000;
2010 for(i=1; i<=31; i++){
2011 diff= predict(&s->p_pred, i, s->mc_mb_var) - (double)s->bit_rate/fps;
2012 if(diff<0) diff= -diff;
2013 if(diff<best_diff){
2014 best_diff= diff;
2015 rate_q= i;
2016 }
2017 }
2018 }
2019
2020 s->short_term_qsum*=s->qblur;
2021 s->short_term_qcount*=s->qblur;
2022
2023 s->short_term_qsum+= rate_q;
2024 s->short_term_qcount++;
2025 short_term_q= s->short_term_qsum/s->short_term_qcount;
2026
2027 long_term_q= s->qsum/s->qcount*(s->total_bits+1)/(wanted_bits+1); //+1 to avoid nan & 0
2028
2029 // q= (long_term_q - short_term_q)*s->qcompress + short_term_q;
2030 q= 1/((1/long_term_q - 1/short_term_q)*s->qcompress + 1/short_term_q);
2031
2032 diff= s->total_bits - wanted_bits;
2033 br_compensation= (s->bit_rate_tolerance - diff)/s->bit_rate_tolerance;
2034 if(br_compensation<=0.0) br_compensation=0.001;
2035 q/=br_compensation;
2036
2037 qscale= (int)(q + 0.5);
2038 if (qscale<qmin) qscale=qmin;
2039 else if(qscale>qmax) qscale=qmax;
2040
2041 if (qscale<last_qscale-s->max_qdiff) qscale=last_qscale-s->max_qdiff;
2042 else if(qscale>last_qscale+s->max_qdiff) qscale=last_qscale+s->max_qdiff;
2043
2044 s->qsum+= qscale;
2045 s->qcount++;
2046
2047 s->last_pict_type= s->pict_type;
2048 //printf("q:%d diff:%d comp:%f rate_q:%d st_q:%f fvar:%d last_size:%d\n", qscale, (int)diff, br_compensation,
2049 // rate_q, short_term_q, s->mc_mb_var, s->frame_bits);
2050 //printf("%d %d\n", s->bit_rate, (int)fps);
2051 return qscale;
2052 #else
2053 INT64 diff, total_bits = s->total_bits;
2054 float q;
2055 int qscale;
2056 if (s->pict_type == I_TYPE) {
2057 s->wanted_bits += s->I_frame_bits;
2058 } else {
2059 s->wanted_bits += s->P_frame_bits;
2060 }
2061 diff = s->wanted_bits - total_bits;
2062 q = 31.0 - (float)diff / (QSCALE_K * s->mb_height * s->mb_width);
2063 /* adjust for I frame */
2064 if (s->pict_type == I_TYPE && !s->intra_only) {
2065 q /= I_FRAME_SIZE_RATIO;
2066 }
2067
2068 /* using a too small Q scale leeds to problems in mpeg1 and h263
2069 because AC coefficients are clamped to 255 or 127 */
2070 qmin = 3;
2071 if (q < qmin)
2072 q = qmin;
2073 else if (q > 31)
2074 q = 31;
2075 qscale = (int)(q + 0.5);
2076 #if defined(DEBUG)
2077 printf("\n%d: total=%0.0f wanted=%0.0f br=%0.1f diff=%d qest=%2.1f\n",
2078 s->picture_number,
2079 (double)total_bits,
2080 (double)s->wanted_bits,
2081 (float)s->frame_rate / FRAME_RATE_BASE *
2082 total_bits / s->picture_number,
2083 (int)diff, q);
2084 #endif
2085 return qscale;
2086 #endif
2087 }
2088
2089 AVCodec mpeg1video_encoder = {
2090 "mpeg1video",
2091 CODEC_TYPE_VIDEO,
2092 CODEC_ID_MPEG1VIDEO,
2093 sizeof(MpegEncContext),
2094 MPV_encode_init,
2095 MPV_encode_picture,
2096 MPV_encode_end,
2097 };
2098
2099 AVCodec h263_encoder = {
2100 "h263",
2101 CODEC_TYPE_VIDEO,
2102 CODEC_ID_H263,
2103 sizeof(MpegEncContext),
2104 MPV_encode_init,
2105 MPV_encode_picture,
2106 MPV_encode_end,
2107 };
2108
2109 AVCodec h263p_encoder = {
2110 "h263p",
2111 CODEC_TYPE_VIDEO,
2112 CODEC_ID_H263P,
2113 sizeof(MpegEncContext),
2114 MPV_encode_init,
2115 MPV_encode_picture,
2116 MPV_encode_end,
2117 };
2118
2119 AVCodec rv10_encoder = {
2120 "rv10",
2121 CODEC_TYPE_VIDEO,
2122 CODEC_ID_RV10,
2123 sizeof(MpegEncContext),
2124 MPV_encode_init,
2125 MPV_encode_picture,
2126 MPV_encode_end,
2127 };
2128
2129 AVCodec mjpeg_encoder = {
2130 "mjpeg",
2131 CODEC_TYPE_VIDEO,
2132 CODEC_ID_MJPEG,
2133 sizeof(MpegEncContext),
2134 MPV_encode_init,
2135 MPV_encode_picture,
2136 MPV_encode_end,
2137 };
2138
2139 AVCodec mpeg4_encoder = {
2140 "mpeg4",
2141 CODEC_TYPE_VIDEO,
2142 CODEC_ID_MPEG4,
2143 sizeof(MpegEncContext),
2144 MPV_encode_init,
2145 MPV_encode_picture,
2146 MPV_encode_end,
2147 };
2148
2149 AVCodec msmpeg4v1_encoder = {
2150 "msmpeg4v1",
2151 CODEC_TYPE_VIDEO,
2152 CODEC_ID_MSMPEG4V1,
2153 sizeof(MpegEncContext),
2154 MPV_encode_init,
2155 MPV_encode_picture,
2156 MPV_encode_end,
2157 };
2158
2159 AVCodec msmpeg4v2_encoder = {
2160 "msmpeg4v2",
2161 CODEC_TYPE_VIDEO,
2162 CODEC_ID_MSMPEG4V2,
2163 sizeof(MpegEncContext),
2164 MPV_encode_init,
2165 MPV_encode_picture,
2166 MPV_encode_end,
2167 };
2168
2169 AVCodec msmpeg4v3_encoder = {
2170 "msmpeg4",
2171 CODEC_TYPE_VIDEO,
2172 CODEC_ID_MSMPEG4V3,
2173 sizeof(MpegEncContext),
2174 MPV_encode_init,
2175 MPV_encode_picture,
2176 MPV_encode_end,
2177 };