10l
[libav.git] / libavcodec / mpegvideo.c
CommitLineData
de6d9b64
FB
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#include <stdlib.h>
20#include <stdio.h>
21#include <math.h>
22#include <string.h>
23#include "avcodec.h"
24#include "dsputil.h"
25#include "mpegvideo.h"
26
54329dd5
NK
27#ifdef USE_FASTMEMCPY
28#include "fastmemcpy.h"
29#endif
30
21af69f7
FB
31static void encode_picture(MpegEncContext *s, int picture_number);
32static void rate_control_init(MpegEncContext *s);
33static int rate_estimate_qscale(MpegEncContext *s);
34static void dct_unquantize_mpeg1_c(MpegEncContext *s,
35 DCTELEM *block, int n, int qscale);
36static void dct_unquantize_h263_c(MpegEncContext *s,
37 DCTELEM *block, int n, int qscale);
3d9fccbf 38static void draw_edges_c(UINT8 *buf, int wrap, int width, int height, int w);
2f349de2 39static int dct_quantize_c(MpegEncContext *s, DCTELEM *block, int n, int qscale);
3d9fccbf 40
2f349de2 41int (*dct_quantize)(MpegEncContext *s, DCTELEM *block, int n, int qscale)= dct_quantize_c;
3d9fccbf
MN
42void (*draw_edges)(UINT8 *buf, int wrap, int width, int height, int w)= draw_edges_c;
43
de6d9b64
FB
44#define EDGE_WIDTH 16
45
46/* enable all paranoid tests for rounding, overflows, etc... */
47//#define PARANOID
48
49//#define DEBUG
50
51/* for jpeg fast DCT */
52#define CONST_BITS 14
53
54static const unsigned short aanscales[64] = {
55 /* precomputed values scaled up by 14 bits */
56 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520,
57 22725, 31521, 29692, 26722, 22725, 17855, 12299, 6270,
58 21407, 29692, 27969, 25172, 21407, 16819, 11585, 5906,
59 19266, 26722, 25172, 22654, 19266, 15137, 10426, 5315,
60 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520,
61 12873, 17855, 16819, 15137, 12873, 10114, 6967, 3552,
62 8867, 12299, 11585, 10426, 8867, 6967, 4799, 2446,
63 4520, 6270, 5906, 5315, 4520, 3552, 2446, 1247
64};
65
66static UINT8 h263_chroma_roundtab[16] = {
67 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2,
68};
69
9c15096e 70static UINT16 default_mv_penalty[MAX_FCODE+1][MAX_MV*2+1];
45870f57
MN
71static UINT8 default_fcode_tab[MAX_MV*2+1];
72
de6d9b64
FB
73/* default motion estimation */
74int motion_estimation_method = ME_LOG;
75
badaf88e
MN
76extern UINT8 zigzag_end[64];
77
2f349de2 78static void convert_matrix(int *qmat, UINT16 *qmat16, const UINT16 *quant_matrix, int qscale)
de6d9b64
FB
79{
80 int i;
81
82 if (av_fdct == jpeg_fdct_ifast) {
83 for(i=0;i<64;i++) {
84 /* 16 <= qscale * quant_matrix[i] <= 7905 */
2f349de2
MN
85 /* 19952 <= aanscales[i] * qscale * quant_matrix[i] <= 249205026 */
86 /* (1<<36)/19952 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */
87 /* 3444240 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= 275 */
de6d9b64 88
2f349de2
MN
89 qmat[block_permute_op(i)] = (int)((UINT64_C(1) << (QMAT_SHIFT + 11)) /
90 (aanscales[i] * qscale * quant_matrix[block_permute_op(i)]));
de6d9b64
FB
91 }
92 } else {
93 for(i=0;i<64;i++) {
94 /* We can safely suppose that 16 <= quant_matrix[i] <= 255
2f349de2
MN
95 So 16 <= qscale * quant_matrix[i] <= 7905
96 so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905
97 so 32768 >= (1<<19) / (qscale * quant_matrix[i]) >= 67
de6d9b64 98 */
2f349de2
MN
99 qmat[i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]);
100 qmat16[i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[block_permute_op(i)]);
de6d9b64
FB
101 }
102 }
103}
104
105/* init common structure for both encoder and decoder */
106int MPV_common_init(MpegEncContext *s)
107{
108 int c_size, i;
109 UINT8 *pict;
110
21af69f7
FB
111 if (s->out_format == FMT_H263)
112 s->dct_unquantize = dct_unquantize_h263_c;
113 else
114 s->dct_unquantize = dct_unquantize_mpeg1_c;
115
116#ifdef HAVE_MMX
117 MPV_common_init_mmx(s);
a9b3f630 118#endif
de6d9b64
FB
119 s->mb_width = (s->width + 15) / 16;
120 s->mb_height = (s->height + 15) / 16;
e03c341e 121 s->mb_num = s->mb_width * s->mb_height;
de6d9b64
FB
122 s->linesize = s->mb_width * 16 + 2 * EDGE_WIDTH;
123
124 for(i=0;i<3;i++) {
125 int w, h, shift, pict_start;
126
127 w = s->linesize;
128 h = s->mb_height * 16 + 2 * EDGE_WIDTH;
129 shift = (i == 0) ? 0 : 1;
130 c_size = (w >> shift) * (h >> shift);
131 pict_start = (w >> shift) * (EDGE_WIDTH >> shift) + (EDGE_WIDTH >> shift);
132
133 pict = av_mallocz(c_size);
134 if (pict == NULL)
135 goto fail;
136 s->last_picture_base[i] = pict;
137 s->last_picture[i] = pict + pict_start;
138
139 pict = av_mallocz(c_size);
140 if (pict == NULL)
141 goto fail;
142 s->next_picture_base[i] = pict;
143 s->next_picture[i] = pict + pict_start;
144
145 if (s->has_b_frames) {
146 pict = av_mallocz(c_size);
147 if (pict == NULL)
148 goto fail;
149 s->aux_picture_base[i] = pict;
150 s->aux_picture[i] = pict + pict_start;
151 }
152 }
37fbfd0a
J
153
154 if (s->encoding) {
155 /* Allocate MB type table */
b4dbd87c 156 s->mb_type = av_mallocz(s->mb_num * sizeof(char));
37fbfd0a
J
157 if (s->mb_type == NULL) {
158 perror("malloc");
159 goto fail;
160 }
b4dbd87c
J
161
162 s->mb_var = av_mallocz(s->mb_num * sizeof(INT16));
163 if (s->mb_var == NULL) {
164 perror("malloc");
165 goto fail;
166 }
37fbfd0a
J
167 /* Allocate MV table */
168 /* By now we just have one MV per MB */
b4dbd87c
J
169 s->mv_table[0] = av_mallocz(s->mb_num * sizeof(INT16));
170 s->mv_table[1] = av_mallocz(s->mb_num * sizeof(INT16));
37fbfd0a
J
171 if (s->mv_table[1] == NULL || s->mv_table[0] == NULL) {
172 perror("malloc");
173 goto fail;
174 }
175 }
176
174489bd 177 if (s->out_format == FMT_H263 || s->encoding) {
de6d9b64
FB
178 int size;
179 /* MV prediction */
180 size = (2 * s->mb_width + 2) * (2 * s->mb_height + 2);
181 s->motion_val = malloc(size * 2 * sizeof(INT16));
182 if (s->motion_val == NULL)
183 goto fail;
184 memset(s->motion_val, 0, size * 2 * sizeof(INT16));
185 }
186
d140623f 187 if (s->h263_pred || s->h263_plus) {
de6d9b64
FB
188 int y_size, c_size, i, size;
189
190 /* dc values */
191
192 y_size = (2 * s->mb_width + 2) * (2 * s->mb_height + 2);
193 c_size = (s->mb_width + 2) * (s->mb_height + 2);
194 size = y_size + 2 * c_size;
195 s->dc_val[0] = malloc(size * sizeof(INT16));
196 if (s->dc_val[0] == NULL)
197 goto fail;
198 s->dc_val[1] = s->dc_val[0] + y_size;
199 s->dc_val[2] = s->dc_val[1] + c_size;
200 for(i=0;i<size;i++)
201 s->dc_val[0][i] = 1024;
202
203 /* ac values */
204 s->ac_val[0] = av_mallocz(size * sizeof(INT16) * 16);
205 if (s->ac_val[0] == NULL)
206 goto fail;
207 s->ac_val[1] = s->ac_val[0] + y_size;
208 s->ac_val[2] = s->ac_val[1] + c_size;
209
210 /* cbp values */
211 s->coded_block = av_mallocz(y_size);
212 if (!s->coded_block)
213 goto fail;
bff6ecaa
MN
214
215 /* which mb is a intra block */
e03c341e 216 s->mbintra_table = av_mallocz(s->mb_num);
bff6ecaa
MN
217 if (!s->mbintra_table)
218 goto fail;
e03c341e 219 memset(s->mbintra_table, 1, s->mb_num);
5b3438c6 220 }
de6d9b64
FB
221 /* default structure is frame */
222 s->picture_structure = PICT_FRAME;
223
3bb4e23a
FB
224 /* init macroblock skip table */
225 if (!s->encoding) {
e03c341e 226 s->mbskip_table = av_mallocz(s->mb_num);
3bb4e23a
FB
227 if (!s->mbskip_table)
228 goto fail;
229 }
230
de6d9b64
FB
231 s->context_initialized = 1;
232 return 0;
233 fail:
8257bf05 234 MPV_common_end(s);
de6d9b64
FB
235 return -1;
236}
237
238/* init common structure for both encoder and decoder */
239void MPV_common_end(MpegEncContext *s)
240{
241 int i;
242
37fbfd0a
J
243 if (s->mb_type)
244 free(s->mb_type);
b4dbd87c
J
245 if (s->mb_var)
246 free(s->mb_var);
37fbfd0a
J
247 if (s->mv_table[0])
248 free(s->mv_table[0]);
249 if (s->mv_table[1])
250 free(s->mv_table[1]);
de6d9b64
FB
251 if (s->motion_val)
252 free(s->motion_val);
8257bf05 253 if (s->dc_val[0])
de6d9b64 254 free(s->dc_val[0]);
8257bf05 255 if (s->ac_val[0])
de6d9b64 256 free(s->ac_val[0]);
8257bf05 257 if (s->coded_block)
de6d9b64 258 free(s->coded_block);
8257bf05 259 if (s->mbintra_table)
5b3438c6 260 free(s->mbintra_table);
8257bf05 261
3bb4e23a
FB
262 if (s->mbskip_table)
263 free(s->mbskip_table);
de6d9b64 264 for(i=0;i<3;i++) {
8257bf05
ZK
265 if (s->last_picture_base[i])
266 free(s->last_picture_base[i]);
267 if (s->next_picture_base[i])
268 free(s->next_picture_base[i]);
de6d9b64
FB
269 if (s->has_b_frames)
270 free(s->aux_picture_base[i]);
271 }
272 s->context_initialized = 0;
273}
274
275/* init video encoder */
276int MPV_encode_init(AVCodecContext *avctx)
277{
278 MpegEncContext *s = avctx->priv_data;
519c2b6d 279 int i;
de6d9b64
FB
280
281 s->bit_rate = avctx->bit_rate;
9cdd6a24 282 s->bit_rate_tolerance = avctx->bit_rate_tolerance;
de6d9b64
FB
283 s->frame_rate = avctx->frame_rate;
284 s->width = avctx->width;
285 s->height = avctx->height;
286 s->gop_size = avctx->gop_size;
644d98a4
J
287 s->rtp_mode = avctx->rtp_mode;
288 s->rtp_payload_size = avctx->rtp_payload_size;
81401c1f
J
289 if (avctx->rtp_callback)
290 s->rtp_callback = avctx->rtp_callback;
9cdd6a24
MN
291 s->qmin= avctx->qmin;
292 s->qmax= avctx->qmax;
293 s->max_qdiff= avctx->max_qdiff;
294 s->qcompress= avctx->qcompress;
295 s->qblur= avctx->qblur;
477c35a9 296 s->avctx = avctx;
11ce8834 297 s->aspect_ratio_info= avctx->aspect_ratio_info;
174489bd 298 s->hq= (avctx->flags & CODEC_FLAG_HQ);
644d98a4 299
de6d9b64
FB
300 if (s->gop_size <= 1) {
301 s->intra_only = 1;
302 s->gop_size = 12;
303 } else {
304 s->intra_only = 0;
305 }
306 s->full_search = motion_estimation_method;
307
308 s->fixed_qscale = (avctx->flags & CODEC_FLAG_QSCALE);
37fbfd0a 309
de6d9b64
FB
310 switch(avctx->codec->id) {
311 case CODEC_ID_MPEG1VIDEO:
312 s->out_format = FMT_MPEG1;
313 break;
314 case CODEC_ID_MJPEG:
315 s->out_format = FMT_MJPEG;
316 s->intra_only = 1; /* force intra only for jpeg */
37fbfd0a
J
317 s->mjpeg_write_tables = 1; /* write all tables */
318 s->mjpeg_vsample[0] = 2; /* set up default sampling factors */
319 s->mjpeg_vsample[1] = 1; /* the only currently supported values */
320 s->mjpeg_vsample[2] = 1;
321 s->mjpeg_hsample[0] = 2;
322 s->mjpeg_hsample[1] = 1;
323 s->mjpeg_hsample[2] = 1;
de6d9b64
FB
324 if (mjpeg_init(s) < 0)
325 return -1;
326 break;
327 case CODEC_ID_H263:
37fbfd0a
J
328 if (h263_get_picture_format(s->width, s->height) == 7) {
329 printf("Input picture size isn't suitable for h263 codec! try h263+\n");
de6d9b64 330 return -1;
37fbfd0a 331 }
de6d9b64
FB
332 s->out_format = FMT_H263;
333 break;
334 case CODEC_ID_H263P:
335 s->out_format = FMT_H263;
644d98a4
J
336 s->rtp_mode = 1;
337 s->rtp_payload_size = 1200;
de6d9b64 338 s->h263_plus = 1;
6dbd39fe 339 s->unrestricted_mv = 1;
544286b3
J
340
341 /* These are just to be sure */
342 s->umvplus = 0;
343 s->umvplus_dec = 0;
de6d9b64
FB
344 break;
345 case CODEC_ID_RV10:
346 s->out_format = FMT_H263;
347 s->h263_rv10 = 1;
348 break;
58f26ba9 349 case CODEC_ID_MPEG4:
de6d9b64
FB
350 s->out_format = FMT_H263;
351 s->h263_pred = 1;
352 s->unrestricted_mv = 1;
353 break;
354 case CODEC_ID_MSMPEG4:
355 s->out_format = FMT_H263;
356 s->h263_msmpeg4 = 1;
357 s->h263_pred = 1;
358 s->unrestricted_mv = 1;
359 break;
360 default:
361 return -1;
362 }
363
45870f57
MN
364 { /* set up some save defaults, some codecs might override them later */
365 static int done=0;
366 if(!done){
367 int i;
368 done=1;
9c15096e 369 memset(default_mv_penalty, 0, sizeof(UINT16)*(MAX_FCODE+1)*(2*MAX_MV+1));
45870f57
MN
370 memset(default_fcode_tab , 0, sizeof(UINT8)*(2*MAX_MV+1));
371
372 for(i=-16; i<16; i++){
373 default_fcode_tab[i + MAX_MV]= 1;
374 }
375 }
376 }
377 s->mv_penalty= default_mv_penalty;
378 s->fcode_tab= default_fcode_tab;
379
de6d9b64 380 if (s->out_format == FMT_H263)
45870f57 381 h263_encode_init(s);
11ce8834
MN
382 else if (s->out_format == FMT_MPEG1)
383 mpeg1_encode_init(s);
de6d9b64 384
9d2a0355
MN
385 /* dont use mv_penalty table for crap MV as it would be confused */
386 if(s->full_search<4) s->mv_penalty= default_mv_penalty;
387
3bb4e23a
FB
388 s->encoding = 1;
389
de6d9b64
FB
390 /* init */
391 if (MPV_common_init(s) < 0)
392 return -1;
393
519c2b6d
FB
394 /* init default q matrix */
395 for(i=0;i<64;i++) {
396 s->intra_matrix[i] = default_intra_matrix[i];
397 s->non_intra_matrix[i] = default_non_intra_matrix[i];
398 }
399
de6d9b64
FB
400 /* rate control init */
401 rate_control_init(s);
402
403 s->picture_number = 0;
45870f57 404 s->picture_in_gop_number = 0;
de6d9b64
FB
405 s->fake_picture_number = 0;
406 /* motion detector init */
407 s->f_code = 1;
408
409 return 0;
410}
411
412int MPV_encode_end(AVCodecContext *avctx)
413{
414 MpegEncContext *s = avctx->priv_data;
415
416#ifdef STATS
417 print_stats();
418#endif
419 MPV_common_end(s);
420 if (s->out_format == FMT_MJPEG)
421 mjpeg_close(s);
37fbfd0a 422
de6d9b64
FB
423 return 0;
424}
425
426/* draw the edges of width 'w' of an image of size width, height */
3d9fccbf 427static void draw_edges_c(UINT8 *buf, int wrap, int width, int height, int w)
de6d9b64
FB
428{
429 UINT8 *ptr, *last_line;
430 int i;
431
432 last_line = buf + (height - 1) * wrap;
433 for(i=0;i<w;i++) {
434 /* top and bottom */
435 memcpy(buf - (i + 1) * wrap, buf, width);
436 memcpy(last_line + (i + 1) * wrap, last_line, width);
437 }
438 /* left and right */
439 ptr = buf;
440 for(i=0;i<height;i++) {
441 memset(ptr - w, ptr[0], w);
442 memset(ptr + width, ptr[width-1], w);
443 ptr += wrap;
444 }
445 /* corners */
446 for(i=0;i<w;i++) {
447 memset(buf - (i + 1) * wrap - w, buf[0], w); /* top left */
448 memset(buf - (i + 1) * wrap + width, buf[width-1], w); /* top right */
449 memset(last_line + (i + 1) * wrap - w, last_line[0], w); /* top left */
450 memset(last_line + (i + 1) * wrap + width, last_line[width-1], w); /* top right */
451 }
452}
453
de6d9b64
FB
454/* generic function for encode/decode called before a frame is coded/decoded */
455void MPV_frame_start(MpegEncContext *s)
456{
457 int i;
458 UINT8 *tmp;
459
425dddb7 460 s->mb_skiped = 0;
de6d9b64
FB
461 if (s->pict_type == B_TYPE) {
462 for(i=0;i<3;i++) {
463 s->current_picture[i] = s->aux_picture[i];
464 }
465 } else {
6f91bcd1 466 s->last_non_b_pict_type= s->pict_type;
de6d9b64
FB
467 for(i=0;i<3;i++) {
468 /* swap next and last */
469 tmp = s->last_picture[i];
470 s->last_picture[i] = s->next_picture[i];
471 s->next_picture[i] = tmp;
472 s->current_picture[i] = tmp;
473 }
474 }
475}
21af69f7 476
de6d9b64
FB
477/* generic function for encode/decode called after a frame has been coded/decoded */
478void MPV_frame_end(MpegEncContext *s)
479{
480 /* draw edge for correct motion prediction if outside */
2f349de2 481 if (s->pict_type != B_TYPE && !s->intra_only) {
1699d376 482 if(s->avctx==NULL || s->avctx->codec->id!=CODEC_ID_MPEG4 || s->divx_version==500){
393fe8dd
A
483 draw_edges(s->current_picture[0], s->linesize, s->mb_width*16, s->mb_height*16, EDGE_WIDTH);
484 draw_edges(s->current_picture[1], s->linesize/2, s->mb_width*8, s->mb_height*8, EDGE_WIDTH/2);
485 draw_edges(s->current_picture[2], s->linesize/2, s->mb_width*8, s->mb_height*8, EDGE_WIDTH/2);
ef093645 486 }else{
1699d376 487 /* mpeg4? / opendivx / xvid */
de6d9b64
FB
488 draw_edges(s->current_picture[0], s->linesize, s->width, s->height, EDGE_WIDTH);
489 draw_edges(s->current_picture[1], s->linesize/2, s->width/2, s->height/2, EDGE_WIDTH/2);
490 draw_edges(s->current_picture[2], s->linesize/2, s->width/2, s->height/2, EDGE_WIDTH/2);
ef093645 491 }
de6d9b64 492 }
5975626d 493 emms_c();
de6d9b64
FB
494}
495
496int MPV_encode_picture(AVCodecContext *avctx,
497 unsigned char *buf, int buf_size, void *data)
498{
499 MpegEncContext *s = avctx->priv_data;
500 AVPicture *pict = data;
501 int i, j;
502
503 if (s->fixed_qscale)
504 s->qscale = avctx->quality;
505
506 init_put_bits(&s->pb, buf, buf_size, NULL, NULL);
507
508 if (!s->intra_only) {
509 /* first picture of GOP is intra */
174489bd 510 if (s->picture_in_gop_number % s->gop_size==0){
45870f57 511 s->picture_in_gop_number=0;
de6d9b64 512 s->pict_type = I_TYPE;
45870f57 513 }else
de6d9b64
FB
514 s->pict_type = P_TYPE;
515 } else {
516 s->pict_type = I_TYPE;
517 }
de6d9b64
FB
518
519 MPV_frame_start(s);
2f349de2 520
de6d9b64
FB
521 for(i=0;i<3;i++) {
522 UINT8 *src = pict->data[i];
523 UINT8 *dest = s->current_picture[i];
524 int src_wrap = pict->linesize[i];
525 int dest_wrap = s->linesize;
526 int w = s->width;
527 int h = s->height;
528
529 if (i >= 1) {
530 dest_wrap >>= 1;
531 w >>= 1;
532 h >>= 1;
533 }
534
37fbfd0a
J
535 if(dest_wrap==src_wrap){
536 s->new_picture[i] = pict->data[i];
537 } else {
2f349de2
MN
538 for(j=0;j<h;j++) {
539 memcpy(dest, src, w);
540 dest += dest_wrap;
541 src += src_wrap;
542 }
0b2540a0 543 s->new_picture[i] = s->current_picture[i];
37fbfd0a 544 }
de6d9b64
FB
545 }
546
547 encode_picture(s, s->picture_number);
bd31a775 548 avctx->key_frame = (s->pict_type == I_TYPE);
098eefe1
MN
549 avctx->header_bits = s->header_bits;
550 avctx->mv_bits = s->mv_bits;
551 avctx->misc_bits = s->misc_bits;
552 avctx->i_tex_bits = s->i_tex_bits;
553 avctx->p_tex_bits = s->p_tex_bits;
554 avctx->i_count = s->i_count;
555 avctx->p_count = s->p_count;
556 avctx->skip_count = s->skip_count;
557
de6d9b64
FB
558 MPV_frame_end(s);
559 s->picture_number++;
45870f57 560 s->picture_in_gop_number++;
de6d9b64
FB
561
562 if (s->out_format == FMT_MJPEG)
563 mjpeg_picture_trailer(s);
564
565 flush_put_bits(&s->pb);
9cdd6a24
MN
566 s->last_frame_bits= s->frame_bits;
567 s->frame_bits = (pbBufPtr(&s->pb) - s->pb.buf) * 8;
568 s->total_bits += s->frame_bits;
098eefe1
MN
569 avctx->frame_bits = s->frame_bits;
570//printf("fcode: %d, type: %d, head: %d, mv: %d, misc: %d, frame: %d, itex: %d, ptex: %d\n",
571//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);
17592475 572
de6d9b64 573 avctx->quality = s->qscale;
43f1708f
J
574 if (avctx->get_psnr) {
575 /* At this point pict->data should have the original frame */
576 /* an s->current_picture should have the coded/decoded frame */
577 get_psnr(pict->data, s->current_picture,
578 pict->linesize, s->linesize, avctx);
579 }
17592475 580 return pbBufPtr(&s->pb) - s->pb.buf;
de6d9b64
FB
581}
582
583static inline int clip(int a, int amin, int amax)
584{
585 if (a < amin)
586 return amin;
587 else if (a > amax)
588 return amax;
589 else
590 return a;
591}
592
44eb4951
MN
593static inline void gmc1_motion(MpegEncContext *s,
594 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr,
595 int dest_offset,
596 UINT8 **ref_picture, int src_offset,
597 int h)
598{
599 UINT8 *ptr;
600 int dxy, offset, mx, my, src_x, src_y, height, linesize;
601 int motion_x, motion_y;
602
d6231b9e 603 if(s->real_sprite_warping_points>1) printf("more than 1 warp point isnt supported\n");
44eb4951
MN
604 motion_x= s->sprite_offset[0][0];
605 motion_y= s->sprite_offset[0][1];
606 src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy+1));
607 src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy+1));
608 motion_x<<=(3-s->sprite_warping_accuracy);
609 motion_y<<=(3-s->sprite_warping_accuracy);
610 src_x = clip(src_x, -16, s->width);
611 if (src_x == s->width)
612 motion_x =0;
613 src_y = clip(src_y, -16, s->height);
614 if (src_y == s->height)
615 motion_y =0;
616
617 linesize = s->linesize;
618 ptr = ref_picture[0] + (src_y * linesize) + src_x + src_offset;
619
620 dest_y+=dest_offset;
621 gmc1(dest_y , ptr , linesize, h, motion_x&15, motion_y&15, s->no_rounding);
622 gmc1(dest_y+8, ptr+8, linesize, h, motion_x&15, motion_y&15, s->no_rounding);
623
624 motion_x= s->sprite_offset[1][0];
625 motion_y= s->sprite_offset[1][1];
626 src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy+1));
627 src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy+1));
628 motion_x<<=(3-s->sprite_warping_accuracy);
629 motion_y<<=(3-s->sprite_warping_accuracy);
630 src_x = clip(src_x, -8, s->width>>1);
631 if (src_x == s->width>>1)
632 motion_x =0;
633 src_y = clip(src_y, -8, s->height>>1);
634 if (src_y == s->height>>1)
635 motion_y =0;
636
637 offset = (src_y * linesize>>1) + src_x + (src_offset>>1);
638 ptr = ref_picture[1] + offset;
639 gmc1(dest_cb + (dest_offset>>1), ptr, linesize>>1, h>>1, motion_x&15, motion_y&15, s->no_rounding);
640 ptr = ref_picture[2] + offset;
641 gmc1(dest_cr + (dest_offset>>1), ptr, linesize>>1, h>>1, motion_x&15, motion_y&15, s->no_rounding);
642
643 return;
644}
645
de6d9b64
FB
646/* apply one mpeg motion vector to the three components */
647static inline void mpeg_motion(MpegEncContext *s,
648 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr,
649 int dest_offset,
650 UINT8 **ref_picture, int src_offset,
651 int field_based, op_pixels_func *pix_op,
652 int motion_x, int motion_y, int h)
653{
654 UINT8 *ptr;
655 int dxy, offset, mx, my, src_x, src_y, height, linesize;
44eb4951
MN
656if(s->quarter_sample)
657{
658 motion_x>>=1;
659 motion_y>>=1;
660}
de6d9b64
FB
661 dxy = ((motion_y & 1) << 1) | (motion_x & 1);
662 src_x = s->mb_x * 16 + (motion_x >> 1);
663 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 1);
664
665 /* WARNING: do no forget half pels */
666 height = s->height >> field_based;
667 src_x = clip(src_x, -16, s->width);
668 if (src_x == s->width)
669 dxy &= ~1;
670 src_y = clip(src_y, -16, height);
671 if (src_y == height)
672 dxy &= ~2;
673 linesize = s->linesize << field_based;
674 ptr = ref_picture[0] + (src_y * linesize) + (src_x) + src_offset;
675 dest_y += dest_offset;
676 pix_op[dxy](dest_y, ptr, linesize, h);
677 pix_op[dxy](dest_y + 8, ptr + 8, linesize, h);
678
679 if (s->out_format == FMT_H263) {
680 dxy = 0;
681 if ((motion_x & 3) != 0)
682 dxy |= 1;
683 if ((motion_y & 3) != 0)
684 dxy |= 2;
685 mx = motion_x >> 2;
686 my = motion_y >> 2;
687 } else {
688 mx = motion_x / 2;
689 my = motion_y / 2;
690 dxy = ((my & 1) << 1) | (mx & 1);
691 mx >>= 1;
692 my >>= 1;
693 }
694
695 src_x = s->mb_x * 8 + mx;
696 src_y = s->mb_y * (8 >> field_based) + my;
697 src_x = clip(src_x, -8, s->width >> 1);
698 if (src_x == (s->width >> 1))
699 dxy &= ~1;
700 src_y = clip(src_y, -8, height >> 1);
701 if (src_y == (height >> 1))
702 dxy &= ~2;
703
704 offset = (src_y * (linesize >> 1)) + src_x + (src_offset >> 1);
705 ptr = ref_picture[1] + offset;
706 pix_op[dxy](dest_cb + (dest_offset >> 1), ptr, linesize >> 1, h >> 1);
707 ptr = ref_picture[2] + offset;
708 pix_op[dxy](dest_cr + (dest_offset >> 1), ptr, linesize >> 1, h >> 1);
709}
710
44eb4951
MN
711static inline void qpel_motion(MpegEncContext *s,
712 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr,
713 int dest_offset,
714 UINT8 **ref_picture, int src_offset,
715 int field_based, op_pixels_func *pix_op,
716 qpel_mc_func *qpix_op,
717 int motion_x, int motion_y, int h)
718{
719 UINT8 *ptr;
720 int dxy, offset, mx, my, src_x, src_y, height, linesize;
721
722 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
723 src_x = s->mb_x * 16 + (motion_x >> 2);
724 src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2);
725
726 height = s->height >> field_based;
727 src_x = clip(src_x, -16, s->width);
728 if (src_x == s->width)
729 dxy &= ~3;
730 src_y = clip(src_y, -16, height);
731 if (src_y == height)
732 dxy &= ~12;
733 linesize = s->linesize << field_based;
734 ptr = ref_picture[0] + (src_y * linesize) + src_x + src_offset;
735 dest_y += dest_offset;
736//printf("%d %d %d\n", src_x, src_y, dxy);
737 qpix_op[dxy](dest_y , ptr , linesize, linesize, motion_x&3, motion_y&3);
738 qpix_op[dxy](dest_y + 8, ptr + 8, linesize, linesize, motion_x&3, motion_y&3);
739 qpix_op[dxy](dest_y + linesize*8 , ptr + linesize*8 , linesize, linesize, motion_x&3, motion_y&3);
740 qpix_op[dxy](dest_y + linesize*8 + 8, ptr + linesize*8 + 8, linesize, linesize, motion_x&3, motion_y&3);
741
742 mx= (motion_x>>1) | (motion_x&1);
743 my= (motion_y>>1) | (motion_y&1);
744
745 dxy = 0;
746 if ((mx & 3) != 0)
747 dxy |= 1;
748 if ((my & 3) != 0)
749 dxy |= 2;
750 mx = mx >> 2;
751 my = my >> 2;
752
753 src_x = s->mb_x * 8 + mx;
754 src_y = s->mb_y * (8 >> field_based) + my;
755 src_x = clip(src_x, -8, s->width >> 1);
756 if (src_x == (s->width >> 1))
757 dxy &= ~1;
758 src_y = clip(src_y, -8, height >> 1);
759 if (src_y == (height >> 1))
760 dxy &= ~2;
761
762 offset = (src_y * (linesize >> 1)) + src_x + (src_offset >> 1);
763 ptr = ref_picture[1] + offset;
764 pix_op[dxy](dest_cb + (dest_offset >> 1), ptr, linesize >> 1, h >> 1);
765 ptr = ref_picture[2] + offset;
766 pix_op[dxy](dest_cr + (dest_offset >> 1), ptr, linesize >> 1, h >> 1);
767}
768
769
de6d9b64
FB
770static inline void MPV_motion(MpegEncContext *s,
771 UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr,
772 int dir, UINT8 **ref_picture,
44eb4951 773 op_pixels_func *pix_op, qpel_mc_func *qpix_op)
de6d9b64
FB
774{
775 int dxy, offset, mx, my, src_x, src_y, motion_x, motion_y;
776 int mb_x, mb_y, i;
777 UINT8 *ptr, *dest;
778
779 mb_x = s->mb_x;
780 mb_y = s->mb_y;
781
782 switch(s->mv_type) {
783 case MV_TYPE_16X16:
44eb4951
MN
784 if(s->mcsel){
785#if 0
786 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
787 ref_picture, 0,
788 0, pix_op,
789 s->sprite_offset[0][0]>>3,
790 s->sprite_offset[0][1]>>3,
791 16);
792#else
793 gmc1_motion(s, dest_y, dest_cb, dest_cr, 0,
794 ref_picture, 0,
795 16);
796#endif
6f91bcd1 797 }else if(s->quarter_sample && dir==0){ //FIXME
44eb4951
MN
798 qpel_motion(s, dest_y, dest_cb, dest_cr, 0,
799 ref_picture, 0,
800 0, pix_op, qpix_op,
801 s->mv[dir][0][0], s->mv[dir][0][1], 16);
802 }else{
803 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
804 ref_picture, 0,
805 0, pix_op,
806 s->mv[dir][0][0], s->mv[dir][0][1], 16);
807 }
de6d9b64
FB
808 break;
809 case MV_TYPE_8X8:
810 for(i=0;i<4;i++) {
811 motion_x = s->mv[dir][i][0];
812 motion_y = s->mv[dir][i][1];
813
814 dxy = ((motion_y & 1) << 1) | (motion_x & 1);
815 src_x = mb_x * 16 + (motion_x >> 1) + (i & 1) * 8;
816 src_y = mb_y * 16 + (motion_y >> 1) + ((i >> 1) & 1) * 8;
817
818 /* WARNING: do no forget half pels */
819 src_x = clip(src_x, -16, s->width);
820 if (src_x == s->width)
821 dxy &= ~1;
822 src_y = clip(src_y, -16, s->height);
823 if (src_y == s->height)
824 dxy &= ~2;
825
826 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x);
827 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize;
828 pix_op[dxy](dest, ptr, s->linesize, 8);
829 }
830 /* In case of 8X8, we construct a single chroma motion vector
831 with a special rounding */
832 mx = 0;
833 my = 0;
834 for(i=0;i<4;i++) {
835 mx += s->mv[dir][i][0];
836 my += s->mv[dir][i][1];
837 }
838 if (mx >= 0)
839 mx = (h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1));
840 else {
841 mx = -mx;
842 mx = -(h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1));
843 }
844 if (my >= 0)
845 my = (h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1));
846 else {
847 my = -my;
848 my = -(h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1));
849 }
850 dxy = ((my & 1) << 1) | (mx & 1);
851 mx >>= 1;
852 my >>= 1;
853
854 src_x = mb_x * 8 + mx;
855 src_y = mb_y * 8 + my;
856 src_x = clip(src_x, -8, s->width/2);
857 if (src_x == s->width/2)
858 dxy &= ~1;
859 src_y = clip(src_y, -8, s->height/2);
860 if (src_y == s->height/2)
861 dxy &= ~2;
862
863 offset = (src_y * (s->linesize >> 1)) + src_x;
864 ptr = ref_picture[1] + offset;
865 pix_op[dxy](dest_cb, ptr, s->linesize >> 1, 8);
866 ptr = ref_picture[2] + offset;
867 pix_op[dxy](dest_cr, ptr, s->linesize >> 1, 8);
868 break;
869 case MV_TYPE_FIELD:
870 if (s->picture_structure == PICT_FRAME) {
871 /* top field */
872 mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
873 ref_picture, s->field_select[dir][0] ? s->linesize : 0,
874 1, pix_op,
875 s->mv[dir][0][0], s->mv[dir][0][1], 8);
876 /* bottom field */
877 mpeg_motion(s, dest_y, dest_cb, dest_cr, s->linesize,
878 ref_picture, s->field_select[dir][1] ? s->linesize : 0,
879 1, pix_op,
880 s->mv[dir][1][0], s->mv[dir][1][1], 8);
881 } else {
882
883
884 }
885 break;
886 }
887}
888
889
890/* put block[] to dest[] */
891static inline void put_dct(MpegEncContext *s,
892 DCTELEM *block, int i, UINT8 *dest, int line_size)
893{
894 if (!s->mpeg2)
21af69f7 895 s->dct_unquantize(s, block, i, s->qscale);
4af7bcc1 896 ff_idct (block);
de6d9b64
FB
897 put_pixels_clamped(block, dest, line_size);
898}
899
900/* add block[] to dest[] */
901static inline void add_dct(MpegEncContext *s,
902 DCTELEM *block, int i, UINT8 *dest, int line_size)
903{
904 if (s->block_last_index[i] >= 0) {
905 if (!s->mpeg2)
3d9fccbf 906 if(s->encoding || (!s->h263_msmpeg4))
badaf88e 907 s->dct_unquantize(s, block, i, s->qscale);
4af7bcc1 908 ff_idct (block);
de6d9b64
FB
909 add_pixels_clamped(block, dest, line_size);
910 }
911}
912
913/* generic function called after a macroblock has been parsed by the
914 decoder or after it has been encoded by the encoder.
915
916 Important variables used:
917 s->mb_intra : true if intra macroblock
918 s->mv_dir : motion vector direction
919 s->mv_type : motion vector type
920 s->mv : motion vector
921 s->interlaced_dct : true if interlaced dct used (mpeg2)
922 */
923void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64])
924{
8257bf05 925 int mb_x, mb_y;
de6d9b64
FB
926 int dct_linesize, dct_offset;
927 op_pixels_func *op_pix;
44eb4951 928 qpel_mc_func *op_qpix;
de6d9b64
FB
929
930 mb_x = s->mb_x;
931 mb_y = s->mb_y;
932
79e7b305
A
933#ifdef FF_POSTPROCESS
934 quant_store[mb_y][mb_x]=s->qscale;
935 //printf("[%02d][%02d] %d\n",mb_x,mb_y,s->qscale);
936#endif
937
de6d9b64
FB
938 /* update DC predictors for P macroblocks */
939 if (!s->mb_intra) {
9e15ad28 940 if (s->h263_pred || s->h263_aic) {
bff6ecaa
MN
941 if(s->mbintra_table[mb_x + mb_y*s->mb_width])
942 {
8257bf05 943 int wrap, xy, v;
bff6ecaa 944 s->mbintra_table[mb_x + mb_y*s->mb_width]=0;
de6d9b64 945 wrap = 2 * s->mb_width + 2;
8257bf05 946 xy = 2 * mb_x + 1 + (2 * mb_y + 1) * wrap;
de6d9b64 947 v = 1024;
bff6ecaa 948
9e15ad28 949 s->dc_val[0][xy] = v;
8257bf05
ZK
950 s->dc_val[0][xy + 1] = v;
951 s->dc_val[0][xy + wrap] = v;
952 s->dc_val[0][xy + 1 + wrap] = v;
de6d9b64 953 /* ac pred */
8257bf05
ZK
954 memset(s->ac_val[0][xy], 0, 16 * sizeof(INT16));
955 memset(s->ac_val[0][xy + 1], 0, 16 * sizeof(INT16));
956 memset(s->ac_val[0][xy + wrap], 0, 16 * sizeof(INT16));
957 memset(s->ac_val[0][xy + 1 + wrap], 0, 16 * sizeof(INT16));
de6d9b64 958 if (s->h263_msmpeg4) {
8257bf05
ZK
959 s->coded_block[xy] = 0;
960 s->coded_block[xy + 1] = 0;
961 s->coded_block[xy + wrap] = 0;
962 s->coded_block[xy + 1 + wrap] = 0;
de6d9b64
FB
963 }
964 /* chroma */
965 wrap = s->mb_width + 2;
8257bf05
ZK
966 xy = mb_x + 1 + (mb_y + 1) * wrap;
967 s->dc_val[1][xy] = v;
968 s->dc_val[2][xy] = v;
de6d9b64 969 /* ac pred */
8257bf05
ZK
970 memset(s->ac_val[1][xy], 0, 16 * sizeof(INT16));
971 memset(s->ac_val[2][xy], 0, 16 * sizeof(INT16));
bff6ecaa 972 }
de6d9b64
FB
973 } else {
974 s->last_dc[0] = 128 << s->intra_dc_precision;
975 s->last_dc[1] = 128 << s->intra_dc_precision;
976 s->last_dc[2] = 128 << s->intra_dc_precision;
977 }
978 }
9e15ad28 979 else if (s->h263_pred || s->h263_aic)
bff6ecaa
MN
980 s->mbintra_table[mb_x + mb_y*s->mb_width]=1;
981
c5b1c10a
MN
982 /* update motion predictor, not for B-frames as they need the motion_val from the last P/S-Frame */
983 if (s->out_format == FMT_H263) {
6f91bcd1 984 if(s->pict_type!=B_TYPE){
8257bf05 985 int xy, wrap, motion_x, motion_y;
de6d9b64 986
de6d9b64 987 wrap = 2 * s->mb_width + 2;
8257bf05 988 xy = 2 * mb_x + 1 + (2 * mb_y + 1) * wrap;
de6d9b64
FB
989 if (s->mb_intra) {
990 motion_x = 0;
991 motion_y = 0;
992 goto motion_init;
993 } else if (s->mv_type == MV_TYPE_16X16) {
994 motion_x = s->mv[0][0][0];
995 motion_y = s->mv[0][0][1];
996 motion_init:
997 /* no update if 8X8 because it has been done during parsing */
8257bf05
ZK
998 s->motion_val[xy][0] = motion_x;
999 s->motion_val[xy][1] = motion_y;
1000 s->motion_val[xy + 1][0] = motion_x;
1001 s->motion_val[xy + 1][1] = motion_y;
1002 s->motion_val[xy + wrap][0] = motion_x;
1003 s->motion_val[xy + wrap][1] = motion_y;
1004 s->motion_val[xy + 1 + wrap][0] = motion_x;
1005 s->motion_val[xy + 1 + wrap][1] = motion_y;
de6d9b64 1006 }
6f91bcd1 1007 }
de6d9b64
FB
1008 }
1009
1010 if (!s->intra_only) {
1011 UINT8 *dest_y, *dest_cb, *dest_cr;
3bb4e23a
FB
1012 UINT8 *mbskip_ptr;
1013
1014 /* avoid copy if macroblock skipped in last frame too */
21af69f7 1015 if (!s->encoding && s->pict_type != B_TYPE) {
3bb4e23a
FB
1016 mbskip_ptr = &s->mbskip_table[s->mb_y * s->mb_width + s->mb_x];
1017 if (s->mb_skiped) {
1018 s->mb_skiped = 0;
1019 /* if previous was skipped too, then nothing to do ! */
1020 if (*mbskip_ptr != 0)
1021 goto the_end;
1022 *mbskip_ptr = 1; /* indicate that this time we skiped it */
1023 } else {
1024 *mbskip_ptr = 0; /* not skipped */
1025 }
1026 }
de6d9b64
FB
1027
1028 dest_y = s->current_picture[0] + (mb_y * 16 * s->linesize) + mb_x * 16;
1029 dest_cb = s->current_picture[1] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8;
1030 dest_cr = s->current_picture[2] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8;
1031
1032 if (s->interlaced_dct) {
1033 dct_linesize = s->linesize * 2;
1034 dct_offset = s->linesize;
1035 } else {
1036 dct_linesize = s->linesize;
1037 dct_offset = s->linesize * 8;
1038 }
1039
1040 if (!s->mb_intra) {
1041 /* motion handling */
44eb4951 1042 if (!s->no_rounding){
de6d9b64 1043 op_pix = put_pixels_tab;
44eb4951
MN
1044 op_qpix= qpel_mc_rnd_tab;
1045 }else{
de6d9b64 1046 op_pix = put_no_rnd_pixels_tab;
44eb4951
MN
1047 op_qpix= qpel_mc_no_rnd_tab;
1048 }
de6d9b64
FB
1049
1050 if (s->mv_dir & MV_DIR_FORWARD) {
44eb4951 1051 MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture, op_pix, op_qpix);
de6d9b64
FB
1052 if (!s->no_rounding)
1053 op_pix = avg_pixels_tab;
1054 else
1055 op_pix = avg_no_rnd_pixels_tab;
1056 }
1057 if (s->mv_dir & MV_DIR_BACKWARD) {
44eb4951 1058 MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture, op_pix, op_qpix);
de6d9b64
FB
1059 }
1060
1061 /* add dct residue */
1062 add_dct(s, block[0], 0, dest_y, dct_linesize);
1063 add_dct(s, block[1], 1, dest_y + 8, dct_linesize);
1064 add_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize);
1065 add_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize);
1066
51454751
FB
1067 add_dct(s, block[4], 4, dest_cb, s->linesize >> 1);
1068 add_dct(s, block[5], 5, dest_cr, s->linesize >> 1);
de6d9b64
FB
1069 } else {
1070 /* dct only in intra block */
1071 put_dct(s, block[0], 0, dest_y, dct_linesize);
1072 put_dct(s, block[1], 1, dest_y + 8, dct_linesize);
1073 put_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize);
1074 put_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize);
1075
51454751
FB
1076 put_dct(s, block[4], 4, dest_cb, s->linesize >> 1);
1077 put_dct(s, block[5], 5, dest_cr, s->linesize >> 1);
de6d9b64
FB
1078 }
1079 }
3bb4e23a
FB
1080 the_end:
1081 emms_c();
de6d9b64
FB
1082}
1083
1084static void encode_picture(MpegEncContext *s, int picture_number)
1085{
81401c1f 1086 int mb_x, mb_y, wrap, last_gob, pdif = 0;
de6d9b64 1087 UINT8 *ptr;
de6d9b64 1088 int i, motion_x, motion_y;
098eefe1 1089 int bits;
de6d9b64
FB
1090
1091 s->picture_number = picture_number;
9cdd6a24
MN
1092
1093 s->last_mc_mb_var = s->mc_mb_var;
1094 /* Reset the average MB variance */
1095 s->avg_mb_var = 0;
1096 s->mc_mb_var = 0;
9cdd6a24
MN
1097 /* Estimate motion for every MB */
1098 for(mb_y=0; mb_y < s->mb_height; mb_y++) {
1099 for(mb_x=0; mb_x < s->mb_width; mb_x++) {
1100 int xy= mb_y * s->mb_width + mb_x;
174489bd
MN
1101 const int mot_stride = s->mb_width*2 + 2;
1102 int mot_xy = (mb_y*2 + 1)*mot_stride + mb_x*2 + 1;
9cdd6a24
MN
1103 s->mb_x = mb_x;
1104 s->mb_y = mb_y;
1105
1106 /* compute motion vector and macro block type (intra or non intra) */
1107 motion_x = 0;
1108 motion_y = 0;
1109 if (s->pict_type == P_TYPE) {
1110 s->mb_intra = estimate_motion(s, mb_x, mb_y,
1111 &motion_x,
1112 &motion_y);
1113 } else {
1114 s->mb_intra = 1;
1115 }
1116 /* Store MB type and MV */
1117 s->mb_type[xy] = s->mb_intra;
1118 s->mv_table[0][xy] = motion_x;
1119 s->mv_table[1][xy] = motion_y;
174489bd
MN
1120
1121 s->motion_val[mot_xy ][0]= motion_x;
1122 s->motion_val[mot_xy ][1]= motion_y;
1123 s->motion_val[mot_xy+1][0]= motion_x;
1124 s->motion_val[mot_xy+1][1]= motion_y;
1125 mot_xy += mot_stride;
1126 s->motion_val[mot_xy ][0]= motion_x;
1127 s->motion_val[mot_xy ][1]= motion_y;
1128 s->motion_val[mot_xy+1][0]= motion_x;
1129 s->motion_val[mot_xy+1][1]= motion_y;
9cdd6a24
MN
1130 }
1131 }
45870f57 1132 emms_c();
9cdd6a24 1133
bd31a775
MN
1134 if(s->avg_mb_var < s->mc_mb_var && s->pict_type != B_TYPE){ //FIXME subtract MV bits
1135 int i;
1136 s->pict_type= I_TYPE;
45870f57
MN
1137 s->picture_in_gop_number=0;
1138 for(i=0; i<s->mb_num; i++){
1139 s->mb_type[i] = 1;
bd31a775
MN
1140 s->mv_table[0][i] = 0;
1141 s->mv_table[1][i] = 0;
1142 }
1143 }
45870f57 1144
9d2a0355
MN
1145 /* find best f_code for ME which do unlimited searches */
1146 if(s->pict_type==P_TYPE && s->full_search>3){
45870f57
MN
1147 int mv_num[8];
1148 int i;
1149 int loose=0;
1150 UINT8 * fcode_tab= s->fcode_tab;
1151
1152 for(i=0; i<8; i++) mv_num[i]=0;
1153
1154 for(i=0; i<s->mb_num; i++){
1155 if(s->mb_type[i] == 0){
1156 mv_num[ fcode_tab[s->mv_table[0][i] + MAX_MV] ]++;
1157 mv_num[ fcode_tab[s->mv_table[1][i] + MAX_MV] ]++;
1158//printf("%d %d %d\n", s->mv_table[0][i], fcode_tab[s->mv_table[0][i] + MAX_MV], i);
1159 }
1160//else printf("I");
1161 }
1162
1163 for(i=MAX_FCODE; i>1; i--){
1164 loose+= mv_num[i];
098eefe1 1165 if(loose > 10) break; //FIXME this is pretty ineffective
45870f57
MN
1166 }
1167 s->f_code= i;
9d2a0355
MN
1168/* for(i=0; i<=MAX_FCODE; i++){
1169 printf("%d ", mv_num[i]);
1170 }
1171 printf("\n");*/
45870f57
MN
1172 }else{
1173 s->f_code= 1;
1174 }
11ce8834 1175
45870f57
MN
1176//printf("f_code %d ///\n", s->f_code);
1177 /* convert MBs with too long MVs to I-Blocks */
1178 if(s->pict_type==P_TYPE){
1179 int i;
1180 const int f_code= s->f_code;
1181 UINT8 * fcode_tab= s->fcode_tab;
1182
1183 for(i=0; i<s->mb_num; i++){
1184 if(s->mb_type[i] == 0){
1185 if( fcode_tab[s->mv_table[0][i] + MAX_MV] > f_code
1186 || fcode_tab[s->mv_table[0][i] + MAX_MV] == 0
1187 || fcode_tab[s->mv_table[1][i] + MAX_MV] > f_code
1188 || fcode_tab[s->mv_table[1][i] + MAX_MV] == 0 ){
1189 s->mb_type[i] = 1;
1190 s->mv_table[0][i] = 0;
1191 s->mv_table[1][i] = 0;
1192 }
1193 }
1194 }
1195 }
1196
bd31a775
MN
1197// printf("%d %d\n", s->avg_mb_var, s->mc_mb_var);
1198
de6d9b64
FB
1199 if (!s->fixed_qscale)
1200 s->qscale = rate_estimate_qscale(s);
1201
1202 /* precompute matrix */
1203 if (s->out_format == FMT_MJPEG) {
1204 /* for mjpeg, we do include qscale in the matrix */
1205 s->intra_matrix[0] = default_intra_matrix[0];
1206 for(i=1;i<64;i++)
1207 s->intra_matrix[i] = (default_intra_matrix[i] * s->qscale) >> 3;
2f349de2 1208 convert_matrix(s->q_intra_matrix, s->q_intra_matrix16, s->intra_matrix, 8);
de6d9b64 1209 } else {
2f349de2
MN
1210 convert_matrix(s->q_intra_matrix, s->q_intra_matrix16, s->intra_matrix, s->qscale);
1211 convert_matrix(s->q_non_intra_matrix, s->q_non_intra_matrix16, s->non_intra_matrix, s->qscale);
de6d9b64
FB
1212 }
1213
098eefe1 1214 s->last_bits= get_bit_count(&s->pb);
de6d9b64
FB
1215 switch(s->out_format) {
1216 case FMT_MJPEG:
1217 mjpeg_picture_header(s);
1218 break;
1219 case FMT_H263:
1220 if (s->h263_msmpeg4)
1221 msmpeg4_encode_picture_header(s, picture_number);
1222 else if (s->h263_pred)
1223 mpeg4_encode_picture_header(s, picture_number);
1224 else if (s->h263_rv10)
1225 rv10_encode_picture_header(s, picture_number);
1226 else
1227 h263_encode_picture_header(s, picture_number);
1228 break;
1229 case FMT_MPEG1:
1230 mpeg1_encode_picture_header(s, picture_number);
1231 break;
1232 }
098eefe1
MN
1233 bits= get_bit_count(&s->pb);
1234 s->header_bits= bits - s->last_bits;
1235 s->last_bits= bits;
1236 s->mv_bits=0;
1237 s->misc_bits=0;
1238 s->i_tex_bits=0;
1239 s->p_tex_bits=0;
1240 s->i_count=0;
1241 s->p_count=0;
1242 s->skip_count=0;
1243
de6d9b64
FB
1244 /* init last dc values */
1245 /* note: quant matrix value (8) is implied here */
1246 s->last_dc[0] = 128;
1247 s->last_dc[1] = 128;
1248 s->last_dc[2] = 128;
1249 s->mb_incr = 1;
1250 s->last_mv[0][0][0] = 0;
1251 s->last_mv[0][0][1] = 0;
1252 s->mv_type = MV_TYPE_16X16;
1253 s->mv_dir = MV_DIR_FORWARD;
1254
644d98a4 1255 /* Get the GOB height based on picture height */
81401c1f 1256 if (s->out_format == FMT_H263 && !s->h263_pred && !s->h263_msmpeg4) {
644d98a4
J
1257 if (s->height <= 400)
1258 s->gob_index = 1;
1259 else if (s->height <= 800)
1260 s->gob_index = 2;
1261 else
1262 s->gob_index = 4;
1263 }
9cdd6a24 1264
e03c341e
J
1265 s->avg_mb_var = s->avg_mb_var / s->mb_num;
1266
4278e7a6
MN
1267 s->block_wrap[0]=
1268 s->block_wrap[1]=
1269 s->block_wrap[2]=
1270 s->block_wrap[3]= s->mb_width*2 + 2;
1271 s->block_wrap[4]=
1272 s->block_wrap[5]= s->mb_width + 2;
e03c341e
J
1273 for(mb_y=0; mb_y < s->mb_height; mb_y++) {
1274 /* Put GOB header based on RTP MTU */
1275 /* TODO: Put all this stuff in a separate generic function */
1276 if (s->rtp_mode) {
1277 if (!mb_y) {
1278 s->ptr_lastgob = s->pb.buf;
1279 s->ptr_last_mb_line = s->pb.buf;
1280 } else if (s->out_format == FMT_H263 && !s->h263_pred && !s->h263_msmpeg4 && !(mb_y % s->gob_index)) {
1281 last_gob = h263_encode_gob_header(s, mb_y);
1282 if (last_gob) {
1283 s->first_gob_line = 1;
1284 }
1285 }
1286 }
1287
4278e7a6
MN
1288 s->block_index[0]= s->block_wrap[0]*(mb_y*2 + 1) - 1;
1289 s->block_index[1]= s->block_wrap[0]*(mb_y*2 + 1);
1290 s->block_index[2]= s->block_wrap[0]*(mb_y*2 + 2) - 1;
1291 s->block_index[3]= s->block_wrap[0]*(mb_y*2 + 2);
1292 s->block_index[4]= s->block_wrap[4]*(mb_y + 1) + s->block_wrap[0]*(s->mb_height*2 + 2);
1293 s->block_index[5]= s->block_wrap[4]*(mb_y + 1 + s->mb_height + 2) + s->block_wrap[0]*(s->mb_height*2 + 2);
37fbfd0a
J
1294 for(mb_x=0; mb_x < s->mb_width; mb_x++) {
1295
1296 s->mb_x = mb_x;
1297 s->mb_y = mb_y;
4278e7a6
MN
1298 s->block_index[0]+=2;
1299 s->block_index[1]+=2;
1300 s->block_index[2]+=2;
1301 s->block_index[3]+=2;
1302 s->block_index[4]++;
1303 s->block_index[5]++;
37fbfd0a
J
1304#if 0
1305 /* compute motion vector and macro block type (intra or non intra) */
1306 motion_x = 0;
1307 motion_y = 0;
1308 if (s->pict_type == P_TYPE) {
1309 s->mb_intra = estimate_motion(s, mb_x, mb_y,
1310 &motion_x,
1311 &motion_y);
1312 } else {
1313 s->mb_intra = 1;
1314 }
1315#endif
de6d9b64 1316
37fbfd0a
J
1317 s->mb_intra = s->mb_type[mb_y * s->mb_width + mb_x];
1318 motion_x = s->mv_table[0][mb_y * s->mb_width + mb_x];
1319 motion_y = s->mv_table[1][mb_y * s->mb_width + mb_x];
1320
de6d9b64
FB
1321 /* get the pixels */
1322 wrap = s->linesize;
1323 ptr = s->new_picture[0] + (mb_y * 16 * wrap) + mb_x * 16;
21af69f7
FB
1324 get_pixels(s->block[0], ptr, wrap);
1325 get_pixels(s->block[1], ptr + 8, wrap);
1326 get_pixels(s->block[2], ptr + 8 * wrap, wrap);
1327 get_pixels(s->block[3], ptr + 8 * wrap + 8, wrap);
de6d9b64
FB
1328 wrap = s->linesize >> 1;
1329 ptr = s->new_picture[1] + (mb_y * 8 * wrap) + mb_x * 8;
21af69f7 1330 get_pixels(s->block[4], ptr, wrap);
de6d9b64
FB
1331
1332 wrap = s->linesize >> 1;
1333 ptr = s->new_picture[2] + (mb_y * 8 * wrap) + mb_x * 8;
21af69f7 1334 get_pixels(s->block[5], ptr, wrap);
de6d9b64
FB
1335
1336 /* subtract previous frame if non intra */
1337 if (!s->mb_intra) {
1338 int dxy, offset, mx, my;
1339
1340 dxy = ((motion_y & 1) << 1) | (motion_x & 1);
1341 ptr = s->last_picture[0] +
1342 ((mb_y * 16 + (motion_y >> 1)) * s->linesize) +
1343 (mb_x * 16 + (motion_x >> 1));
1344
21af69f7
FB
1345 sub_pixels_2(s->block[0], ptr, s->linesize, dxy);
1346 sub_pixels_2(s->block[1], ptr + 8, s->linesize, dxy);
1347 sub_pixels_2(s->block[2], ptr + s->linesize * 8, s->linesize, dxy);
1348 sub_pixels_2(s->block[3], ptr + 8 + s->linesize * 8, s->linesize ,dxy);
de6d9b64
FB
1349
1350 if (s->out_format == FMT_H263) {
1351 /* special rounding for h263 */
1352 dxy = 0;
1353 if ((motion_x & 3) != 0)
1354 dxy |= 1;
1355 if ((motion_y & 3) != 0)
1356 dxy |= 2;
1357 mx = motion_x >> 2;
1358 my = motion_y >> 2;
1359 } else {
1360 mx = motion_x / 2;
1361 my = motion_y / 2;
1362 dxy = ((my & 1) << 1) | (mx & 1);
1363 mx >>= 1;
1364 my >>= 1;
1365 }
1366 offset = ((mb_y * 8 + my) * (s->linesize >> 1)) + (mb_x * 8 + mx);
1367 ptr = s->last_picture[1] + offset;
21af69f7 1368 sub_pixels_2(s->block[4], ptr, s->linesize >> 1, dxy);
de6d9b64 1369 ptr = s->last_picture[2] + offset;
21af69f7 1370 sub_pixels_2(s->block[5], ptr, s->linesize >> 1, dxy);
de6d9b64 1371 }
3bb4e23a 1372 emms_c();
d140623f
J
1373
1374#if 0
1375 {
1376 float adap_parm;
1377
1378 adap_parm = ((s->avg_mb_var << 1) + s->mb_var[s->mb_width*mb_y+mb_x] + 1.0) /
1379 ((s->mb_var[s->mb_width*mb_y+mb_x] << 1) + s->avg_mb_var + 1.0);
1380
1381 printf("\ntype=%c qscale=%2d adap=%0.2f dquant=%4.2f var=%4d avgvar=%4d",
1382 (s->mb_type[s->mb_width*mb_y+mb_x] > 0) ? 'I' : 'P',
1383 s->qscale, adap_parm, s->qscale*adap_parm,
1384 s->mb_var[s->mb_width*mb_y+mb_x], s->avg_mb_var);
1385 }
1386#endif
de6d9b64
FB
1387 /* DCT & quantize */
1388 if (s->h263_msmpeg4) {
1389 msmpeg4_dc_scale(s);
1390 } else if (s->h263_pred) {
1391 h263_dc_scale(s);
1392 } else {
1393 /* default quantization values */
1394 s->y_dc_scale = 8;
1395 s->c_dc_scale = 8;
1396 }
de6d9b64 1397 for(i=0;i<6;i++) {
2f349de2 1398 s->block_last_index[i] = dct_quantize(s, s->block[i], i, s->qscale);
de6d9b64
FB
1399 }
1400
1401 /* huffman encode */
1402 switch(s->out_format) {
1403 case FMT_MPEG1:
21af69f7 1404 mpeg1_encode_mb(s, s->block, motion_x, motion_y);
de6d9b64
FB
1405 break;
1406 case FMT_H263:
1407 if (s->h263_msmpeg4)
21af69f7 1408 msmpeg4_encode_mb(s, s->block, motion_x, motion_y);
d6231b9e
MN
1409 else if(s->h263_pred)
1410 mpeg4_encode_mb(s, s->block, motion_x, motion_y);
de6d9b64 1411 else
21af69f7 1412 h263_encode_mb(s, s->block, motion_x, motion_y);
de6d9b64
FB
1413 break;
1414 case FMT_MJPEG:
21af69f7 1415 mjpeg_encode_mb(s, s->block);
de6d9b64
FB
1416 break;
1417 }
098eefe1 1418
de6d9b64
FB
1419 /* decompress blocks so that we keep the state of the decoder */
1420 s->mv[0][0][0] = motion_x;
1421 s->mv[0][0][1] = motion_y;
1422
21af69f7 1423 MPV_decode_mb(s, s->block);
de6d9b64 1424 }
17592475
MN
1425
1426
81401c1f
J
1427 /* Obtain average GOB size for RTP */
1428 if (s->rtp_mode) {
1429 if (!mb_y)
17592475 1430 s->mb_line_avgsize = pbBufPtr(&s->pb) - s->ptr_last_mb_line;
81401c1f 1431 else if (!(mb_y % s->gob_index)) {
17592475
MN
1432 s->mb_line_avgsize = (s->mb_line_avgsize + pbBufPtr(&s->pb) - s->ptr_last_mb_line) >> 1;
1433 s->ptr_last_mb_line = pbBufPtr(&s->pb);
81401c1f
J
1434 }
1435 //fprintf(stderr, "\nMB line: %d\tSize: %u\tAvg. Size: %u", s->mb_y,
1436 // (s->pb.buf_ptr - s->ptr_last_mb_line), s->mb_line_avgsize);
1437 s->first_gob_line = 0;
1438 }
de6d9b64 1439 }
098eefe1 1440
2b9ab1d5 1441 if (s->h263_msmpeg4 && s->pict_type == I_TYPE)
ae40484c
MN
1442 msmpeg4_encode_ext_header(s);
1443
644d98a4
J
1444 //if (s->gob_number)
1445 // fprintf(stderr,"\nNumber of GOB: %d", s->gob_number);
81401c1f
J
1446
1447 /* Send the last GOB if RTP */
1448 if (s->rtp_mode) {
1449 flush_put_bits(&s->pb);
17592475 1450 pdif = pbBufPtr(&s->pb) - s->ptr_lastgob;
81401c1f
J
1451 /* Call the RTP callback to send the last GOB */
1452 if (s->rtp_callback)
1453 s->rtp_callback(s->ptr_lastgob, pdif, s->gob_number);
17592475 1454 s->ptr_lastgob = pbBufPtr(&s->pb);
81401c1f
J
1455 //fprintf(stderr,"\nGOB: %2d size: %d (last)", s->gob_number, pdif);
1456 }
1457
de6d9b64
FB
1458}
1459
2f349de2 1460static int dct_quantize_c(MpegEncContext *s,
de6d9b64
FB
1461 DCTELEM *block, int n,
1462 int qscale)
1463{
1464 int i, j, level, last_non_zero, q;
1465 const int *qmat;
4e3269e8
MN
1466 int minLevel, maxLevel;
1467
1468 if(s->avctx!=NULL && s->avctx->codec->id==CODEC_ID_MPEG4){
1469 /* mpeg4 */
1470 minLevel= -2048;
1471 maxLevel= 2047;
1472 }else if(s->out_format==FMT_MPEG1){
1473 /* mpeg1 */
1474 minLevel= -255;
1475 maxLevel= 255;
3cb32e3d
A
1476 }else if(s->out_format==FMT_MJPEG){
1477 /* (m)jpeg */
1478 minLevel= -1023;
1479 maxLevel= 1023;
4e3269e8
MN
1480 }else{
1481 /* h263 / msmpeg4 */
1482 minLevel= -128;
1483 maxLevel= 127;
1484 }
de6d9b64
FB
1485
1486 av_fdct (block);
1487
1a565432
FB
1488 /* we need this permutation so that we correct the IDCT
1489 permutation. will be moved into DCT code */
1490 block_permute(block);
1491
de6d9b64
FB
1492 if (s->mb_intra) {
1493 if (n < 4)
1494 q = s->y_dc_scale;
1495 else
1496 q = s->c_dc_scale;
1497 q = q << 3;
1498
1499 /* note: block[0] is assumed to be positive */
1500 block[0] = (block[0] + (q >> 1)) / q;
1501 i = 1;
1502 last_non_zero = 0;
1503 if (s->out_format == FMT_H263) {
1504 qmat = s->q_non_intra_matrix;
1505 } else {
1506 qmat = s->q_intra_matrix;
1507 }
1508 } else {
1509 i = 0;
1510 last_non_zero = -1;
1511 qmat = s->q_non_intra_matrix;
1512 }
1513
1514 for(;i<64;i++) {
1515 j = zigzag_direct[i];
1516 level = block[j];
1517 level = level * qmat[j];
1518#ifdef PARANOID
1519 {
1520 static int count = 0;
1521 int level1, level2, qmat1;
1522 double val;
1523 if (qmat == s->q_non_intra_matrix) {
1524 qmat1 = default_non_intra_matrix[j] * s->qscale;
1525 } else {
1526 qmat1 = default_intra_matrix[j] * s->qscale;
1527 }
1528 if (av_fdct != jpeg_fdct_ifast)
1529 val = ((double)block[j] * 8.0) / (double)qmat1;
1530 else
1531 val = ((double)block[j] * 8.0 * 2048.0) /
1532 ((double)qmat1 * aanscales[j]);
1533 level1 = (int)val;
1534 level2 = level / (1 << (QMAT_SHIFT - 3));
1535 if (level1 != level2) {
1536 fprintf(stderr, "%d: quant error qlevel=%d wanted=%d level=%d qmat1=%d qmat=%d wantedf=%0.6f\n",
1537 count, level2, level1, block[j], qmat1, qmat[j],
1538 val);
1539 count++;
1540 }
1541
1542 }
1543#endif
1544 /* XXX: slight error for the low range. Test should be equivalent to
1545 (level <= -(1 << (QMAT_SHIFT - 3)) || level >= (1 <<
1546 (QMAT_SHIFT - 3)))
1547 */
1548 if (((level << (31 - (QMAT_SHIFT - 3))) >> (31 - (QMAT_SHIFT - 3))) !=
1549 level) {
1550 level = level / (1 << (QMAT_SHIFT - 3));
1551 /* XXX: currently, this code is not optimal. the range should be:
1552 mpeg1: -255..255
1553 mpeg2: -2048..2047
1554 h263: -128..127
1555 mpeg4: -2048..2047
1556 */
4e3269e8
MN
1557 if (level > maxLevel)
1558 level = maxLevel;
1559 else if (level < minLevel)
1560 level = minLevel;
de6d9b64 1561
de6d9b64
FB
1562 block[j] = level;
1563 last_non_zero = i;
1564 } else {
1565 block[j] = 0;
1566 }
1567 }
1568 return last_non_zero;
1569}
1570
21af69f7
FB
1571static void dct_unquantize_mpeg1_c(MpegEncContext *s,
1572 DCTELEM *block, int n, int qscale)
de6d9b64 1573{
badaf88e 1574 int i, level, nCoeffs;
de6d9b64
FB
1575 const UINT16 *quant_matrix;
1576
badaf88e
MN
1577 if(s->alternate_scan) nCoeffs= 64;
1578 else nCoeffs= s->block_last_index[n]+1;
1579
de6d9b64
FB
1580 if (s->mb_intra) {
1581 if (n < 4)
1582 block[0] = block[0] * s->y_dc_scale;
1583 else
1584 block[0] = block[0] * s->c_dc_scale;
de6d9b64
FB
1585 /* XXX: only mpeg1 */
1586 quant_matrix = s->intra_matrix;
badaf88e
MN
1587 for(i=1;i<nCoeffs;i++) {
1588 int j= zigzag_direct[i];
1589 level = block[j];
de6d9b64
FB
1590 if (level) {
1591 if (level < 0) {
1592 level = -level;
badaf88e 1593 level = (int)(level * qscale * quant_matrix[j]) >> 3;
de6d9b64
FB
1594 level = (level - 1) | 1;
1595 level = -level;
1596 } else {
badaf88e 1597 level = (int)(level * qscale * quant_matrix[j]) >> 3;
de6d9b64
FB
1598 level = (level - 1) | 1;
1599 }
1600#ifdef PARANOID
1601 if (level < -2048 || level > 2047)
1602 fprintf(stderr, "unquant error %d %d\n", i, level);
1603#endif
badaf88e 1604 block[j] = level;
de6d9b64
FB
1605 }
1606 }
1607 } else {
1608 i = 0;
de6d9b64 1609 quant_matrix = s->non_intra_matrix;
d2b3c3d7 1610 for(;i<nCoeffs;i++) {
badaf88e
MN
1611 int j= zigzag_direct[i];
1612 level = block[j];
de6d9b64
FB
1613 if (level) {
1614 if (level < 0) {
1615 level = -level;
1616 level = (((level << 1) + 1) * qscale *
badaf88e 1617 ((int) (quant_matrix[j]))) >> 4;
de6d9b64
FB
1618 level = (level - 1) | 1;
1619 level = -level;
1620 } else {
1621 level = (((level << 1) + 1) * qscale *
badaf88e 1622 ((int) (quant_matrix[j]))) >> 4;
de6d9b64
FB
1623 level = (level - 1) | 1;
1624 }
1625#ifdef PARANOID
1626 if (level < -2048 || level > 2047)
1627 fprintf(stderr, "unquant error %d %d\n", i, level);
1628#endif
badaf88e 1629 block[j] = level;
de6d9b64
FB
1630 }
1631 }
1632 }
1633}
21af69f7
FB
1634
1635static void dct_unquantize_h263_c(MpegEncContext *s,
1636 DCTELEM *block, int n, int qscale)
1637{
1638 int i, level, qmul, qadd;
badaf88e 1639 int nCoeffs;
9e15ad28 1640
21af69f7 1641 if (s->mb_intra) {
9e15ad28
J
1642 if (!s->h263_aic) {
1643 if (n < 4)
1644 block[0] = block[0] * s->y_dc_scale;
1645 else
1646 block[0] = block[0] * s->c_dc_scale;
1647 }
21af69f7 1648 i = 1;
badaf88e 1649 nCoeffs= 64; //does not allways use zigzag table
21af69f7
FB
1650 } else {
1651 i = 0;
badaf88e 1652 nCoeffs= zigzag_end[ s->block_last_index[n] ];
21af69f7
FB
1653 }
1654
1655 qmul = s->qscale << 1;
d140623f
J
1656 if (s->h263_aic && s->mb_intra)
1657 qadd = 0;
1658 else
1659 qadd = (s->qscale - 1) | 1;
21af69f7 1660
badaf88e 1661 for(;i<nCoeffs;i++) {
21af69f7
FB
1662 level = block[i];
1663 if (level) {
1664 if (level < 0) {
1665 level = level * qmul - qadd;
1666 } else {
1667 level = level * qmul + qadd;
1668 }
1669#ifdef PARANOID
1670 if (level < -2048 || level > 2047)
1671 fprintf(stderr, "unquant error %d %d\n", i, level);
1672#endif
1673 block[i] = level;
1674 }
1675 }
1676}
de6d9b64
FB
1677
1678/* rate control */
1679
1680/* an I frame is I_FRAME_SIZE_RATIO bigger than a P frame */
1681#define I_FRAME_SIZE_RATIO 3.0
1682#define QSCALE_K 20
1683
1684static void rate_control_init(MpegEncContext *s)
1685{
9cdd6a24
MN
1686#if 1
1687 emms_c();
1688
1689 //initial values, they dont really matter as they will be totally different within a few frames
1690 s->i_pred.coeff= s->p_pred.coeff= 7.0;
1691 s->i_pred.count= s->p_pred.count= 1.0;
1692
1693 s->i_pred.decay= s->p_pred.decay= 0.4;
1694
1695 // use more bits at the beginning, otherwise high motion at the begin will look like shit
1696 s->qsum=100;
1697 s->qcount=100;
1698
1699 s->short_term_qsum=0.001;
1700 s->short_term_qcount=0.001;
1701#else
de6d9b64
FB
1702 s->wanted_bits = 0;
1703
1704 if (s->intra_only) {
1705 s->I_frame_bits = ((INT64)s->bit_rate * FRAME_RATE_BASE) / s->frame_rate;
1706 s->P_frame_bits = s->I_frame_bits;
1707 } else {
1708 s->P_frame_bits = (int) ((float)(s->gop_size * s->bit_rate) /
1709 (float)((float)s->frame_rate / FRAME_RATE_BASE * (I_FRAME_SIZE_RATIO + s->gop_size - 1)));
1710 s->I_frame_bits = (int)(s->P_frame_bits * I_FRAME_SIZE_RATIO);
1711 }
9cdd6a24 1712
de6d9b64
FB
1713#if defined(DEBUG)
1714 printf("I_frame_size=%d P_frame_size=%d\n",
1715 s->I_frame_bits, s->P_frame_bits);
1716#endif
9cdd6a24 1717#endif
de6d9b64
FB
1718}
1719
9cdd6a24
MN
1720static double predict(Predictor *p, double q, double var)
1721{
1722 return p->coeff*var / (q*p->count);
1723}
1724
1725static void update_predictor(Predictor *p, double q, double var, double size)
1726{
1727 double new_coeff= size*q / (var + 1);
1728 if(var<1000) return;
1729/*{
1730int pred= predict(p, q, var);
1731int error= abs(pred-size);
1732static double sum=0;
1733static int count=0;
1734if(count>5) sum+=error;
1735count++;
1736if(256*256*256*64%count==0){
1737 printf("%d %f %f\n", count, sum/count, p->coeff);
1738}
1739}*/
1740 p->count*= p->decay;
1741 p->coeff*= p->decay;
1742 p->count++;
1743 p->coeff+= new_coeff;
1744}
de6d9b64 1745
de6d9b64
FB
1746static int rate_estimate_qscale(MpegEncContext *s)
1747{
9cdd6a24
MN
1748#if 1
1749 int qmin= s->qmin;
1750 int qmax= s->qmax;
1751 int rate_q=5;
de6d9b64 1752 float q;
9cdd6a24
MN
1753 int qscale;
1754 float br_compensation;
1755 double diff;
1756 double short_term_q;
1757 double long_term_q;
1758 int last_qscale= s->qscale;
1759 double fps;
1760 INT64 wanted_bits;
1761 emms_c();
1762
1763 fps= (double)s->frame_rate / FRAME_RATE_BASE;
1764 wanted_bits= s->bit_rate*(double)s->picture_number/fps;
1765
1766
1767 if(s->picture_number>2){
1768 /* update predictors */
1769 if(s->last_pict_type == I_TYPE){
1770 //FIXME
1771 }else{ //P Frame
1772//printf("%d %d %d %f\n", s->qscale, s->last_mc_mb_var, s->frame_bits, s->p_pred.coeff);
1773 update_predictor(&s->p_pred, s->qscale, s->last_mc_mb_var, s->frame_bits);
1774 }
1775 }
1776
1777 if(s->pict_type == I_TYPE){
1778 //FIXME
1779 rate_q= s->qsum/s->qcount;
1780 }else{ //P Frame
1781 int i;
1782 int diff, best_diff=1000000000;
1783 for(i=1; i<=31; i++){
1784 diff= predict(&s->p_pred, i, s->mc_mb_var) - (double)s->bit_rate/fps;
1785 if(diff<0) diff= -diff;
1786 if(diff<best_diff){
1787 best_diff= diff;
1788 rate_q= i;
1789 }
1790 }
1791 }
1792
1793 s->short_term_qsum*=s->qblur;
1794 s->short_term_qcount*=s->qblur;
1795
1796 s->short_term_qsum+= rate_q;
1797 s->short_term_qcount++;
1798 short_term_q= s->short_term_qsum/s->short_term_qcount;
1799
1800 long_term_q= s->qsum/s->qcount*s->total_bits/wanted_bits;
1801
1802// q= (long_term_q - short_term_q)*s->qcompress + short_term_q;
1803 q= 1/((1/long_term_q - 1/short_term_q)*s->qcompress + 1/short_term_q);
1804
1805 diff= s->total_bits - wanted_bits;
1806 br_compensation= (s->bit_rate_tolerance - diff)/s->bit_rate_tolerance;
c6741159 1807 if(br_compensation<=0.0) br_compensation=0.001;
9cdd6a24
MN
1808 q/=br_compensation;
1809
1810 qscale= (int)(q + 0.5);
1811 if (qscale<qmin) qscale=qmin;
1812 else if(qscale>qmax) qscale=qmax;
1813
1814 if (qscale<last_qscale-s->max_qdiff) qscale=last_qscale-s->max_qdiff;
1815 else if(qscale>last_qscale+s->max_qdiff) qscale=last_qscale+s->max_qdiff;
1816
1817 s->qsum+= qscale;
1818 s->qcount++;
de6d9b64 1819
9cdd6a24
MN
1820 s->last_pict_type= s->pict_type;
1821//printf("q:%d diff:%d comp:%f rate_q:%d st_q:%d fvar:%d last_size:%d\n", qscale, (int)diff, br_compensation,
1822// rate_q, (int)short_term_q, s->mc_mb_var, s->frame_bits);
1823//printf("%d %d\n", s->bit_rate, (int)fps);
1824 return qscale;
1825#else
1826 INT64 diff, total_bits = s->total_bits;
1827 float q;
1828 int qscale;
de6d9b64
FB
1829 if (s->pict_type == I_TYPE) {
1830 s->wanted_bits += s->I_frame_bits;
1831 } else {
1832 s->wanted_bits += s->P_frame_bits;
1833 }
1834 diff = s->wanted_bits - total_bits;
1835 q = 31.0 - (float)diff / (QSCALE_K * s->mb_height * s->mb_width);
1836 /* adjust for I frame */
1837 if (s->pict_type == I_TYPE && !s->intra_only) {
1838 q /= I_FRAME_SIZE_RATIO;
1839 }
1840
1841 /* using a too small Q scale leeds to problems in mpeg1 and h263
1842 because AC coefficients are clamped to 255 or 127 */
1843 qmin = 3;
1844 if (q < qmin)
1845 q = qmin;
1846 else if (q > 31)
1847 q = 31;
1848 qscale = (int)(q + 0.5);
1849#if defined(DEBUG)
d140623f 1850 printf("\n%d: total=%0.0f wanted=%0.0f br=%0.1f diff=%d qest=%2.1f\n",
de6d9b64 1851 s->picture_number,
1a565432 1852 (double)total_bits,
d140623f 1853 (double)s->wanted_bits,
de6d9b64
FB
1854 (float)s->frame_rate / FRAME_RATE_BASE *
1855 total_bits / s->picture_number,
d140623f 1856 (int)diff, q);
de6d9b64
FB
1857#endif
1858 return qscale;
9cdd6a24 1859#endif
de6d9b64
FB
1860}
1861
1862AVCodec mpeg1video_encoder = {
1863 "mpeg1video",
1864 CODEC_TYPE_VIDEO,
1865 CODEC_ID_MPEG1VIDEO,
1866 sizeof(MpegEncContext),
1867 MPV_encode_init,
1868 MPV_encode_picture,
1869 MPV_encode_end,
1870};
1871
1872AVCodec h263_encoder = {
1873 "h263",
1874 CODEC_TYPE_VIDEO,
1875 CODEC_ID_H263,
1876 sizeof(MpegEncContext),
1877 MPV_encode_init,
1878 MPV_encode_picture,
1879 MPV_encode_end,
1880};
1881
1882AVCodec h263p_encoder = {
1883 "h263p",
1884 CODEC_TYPE_VIDEO,
1885 CODEC_ID_H263P,
1886 sizeof(MpegEncContext),
1887 MPV_encode_init,
1888 MPV_encode_picture,
1889 MPV_encode_end,
1890};
1891
1892AVCodec rv10_encoder = {
1893 "rv10",
1894 CODEC_TYPE_VIDEO,
1895 CODEC_ID_RV10,
1896 sizeof(MpegEncContext),
1897 MPV_encode_init,
1898 MPV_encode_picture,
1899 MPV_encode_end,
1900};
1901
1902AVCodec mjpeg_encoder = {
1903 "mjpeg",
1904 CODEC_TYPE_VIDEO,
1905 CODEC_ID_MJPEG,
1906 sizeof(MpegEncContext),
1907 MPV_encode_init,
1908 MPV_encode_picture,
1909 MPV_encode_end,
1910};
1911
58f26ba9
FB
1912AVCodec mpeg4_encoder = {
1913 "mpeg4",
de6d9b64 1914 CODEC_TYPE_VIDEO,
58f26ba9 1915 CODEC_ID_MPEG4,
de6d9b64
FB
1916 sizeof(MpegEncContext),
1917 MPV_encode_init,
1918 MPV_encode_picture,
1919 MPV_encode_end,
1920};
1921
1922AVCodec msmpeg4_encoder = {
1923 "msmpeg4",
1924 CODEC_TYPE_VIDEO,
1925 CODEC_ID_MSMPEG4,
1926 sizeof(MpegEncContext),
1927 MPV_encode_init,
1928 MPV_encode_picture,
1929 MPV_encode_end,
1930};