Statistics
| Branch: | Revision:

ffmpeg / libavcodec / mpegvideo_enc.c @ 57be00be

History | View | Annotate | Download (138 KB)

1
/*
2
 * The simplest mpeg encoder (well, it was the simplest!)
3
 * Copyright (c) 2000,2001 Fabrice Bellard
4
 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
5
 *
6
 * 4MV & hq & B-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
7
 *
8
 * This file is part of FFmpeg.
9
 *
10
 * FFmpeg is free software; you can redistribute it and/or
11
 * modify it under the terms of the GNU Lesser General Public
12
 * License as published by the Free Software Foundation; either
13
 * version 2.1 of the License, or (at your option) any later version.
14
 *
15
 * FFmpeg is distributed in the hope that it will be useful,
16
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
18
 * Lesser General Public License for more details.
19
 *
20
 * You should have received a copy of the GNU Lesser General Public
21
 * License along with FFmpeg; if not, write to the Free Software
22
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23
 */
24

    
25
/**
26
 * @file mpegvideo_enc.c
27
 * The simplest mpeg encoder (well, it was the simplest!).
28
 */
29

    
30
#include "avcodec.h"
31
#include "dsputil.h"
32
#include "mpegvideo.h"
33
#include "mpegvideo_common.h"
34
#include "mjpegenc.h"
35
#include "msmpeg4.h"
36
#include "h263.h"
37
#include "faandct.h"
38
#include "aandcttab.h"
39
#include <limits.h>
40

    
41
//#undef NDEBUG
42
//#include <assert.h>
43

    
44
static int encode_picture(MpegEncContext *s, int picture_number);
45
static int dct_quantize_refine(MpegEncContext *s, DCTELEM *block, int16_t *weight, DCTELEM *orig, int n, int qscale);
46
static int sse_mb(MpegEncContext *s);
47

    
48
/* enable all paranoid tests for rounding, overflows, etc... */
49
//#define PARANOID
50

    
51
//#define DEBUG
52

    
53
static uint8_t default_mv_penalty[MAX_FCODE+1][MAX_MV*2+1];
54
static uint8_t default_fcode_tab[MAX_MV*2+1];
55

    
56
void ff_convert_matrix(DSPContext *dsp, int (*qmat)[64], uint16_t (*qmat16)[2][64],
57
                           const uint16_t *quant_matrix, int bias, int qmin, int qmax, int intra)
58
{
59
    int qscale;
60
    int shift=0;
61

    
62
    for(qscale=qmin; qscale<=qmax; qscale++){
63
        int i;
64
        if (dsp->fdct == ff_jpeg_fdct_islow
65
#ifdef FAAN_POSTSCALE
66
            || dsp->fdct == ff_faandct
67
#endif
68
            ) {
69
            for(i=0;i<64;i++) {
70
                const int j= dsp->idct_permutation[i];
71
                /* 16 <= qscale * quant_matrix[i] <= 7905 */
72
                /* 19952             <= ff_aanscales[i] * qscale * quant_matrix[i]               <= 249205026 */
73
                /* (1 << 36) / 19952 >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= (1 << 36) / 249205026 */
74
                /* 3444240           >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= 275 */
75

    
76
                qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) /
77
                                (qscale * quant_matrix[j]));
78
            }
79
        } else if (dsp->fdct == fdct_ifast
80
#ifndef FAAN_POSTSCALE
81
                   || dsp->fdct == ff_faandct
82
#endif
83
                   ) {
84
            for(i=0;i<64;i++) {
85
                const int j= dsp->idct_permutation[i];
86
                /* 16 <= qscale * quant_matrix[i] <= 7905 */
87
                /* 19952             <= ff_aanscales[i] * qscale * quant_matrix[i]               <= 249205026 */
88
                /* (1 << 36) / 19952 >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */
89
                /* 3444240           >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= 275 */
90

    
91
                qmat[qscale][i] = (int)((UINT64_C(1) << (QMAT_SHIFT + 14)) /
92
                                (ff_aanscales[i] * qscale * quant_matrix[j]));
93
            }
94
        } else {
95
            for(i=0;i<64;i++) {
96
                const int j= dsp->idct_permutation[i];
97
                /* We can safely suppose that 16 <= quant_matrix[i] <= 255
98
                   So 16           <= qscale * quant_matrix[i]             <= 7905
99
                   so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905
100
                   so 32768        >= (1<<19) / (qscale * quant_matrix[i]) >= 67
101
                */
102
                qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) / (qscale * quant_matrix[j]));
103
//                qmat  [qscale][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]);
104
                qmat16[qscale][0][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[j]);
105

    
106
                if(qmat16[qscale][0][i]==0 || qmat16[qscale][0][i]==128*256) qmat16[qscale][0][i]=128*256-1;
107
                qmat16[qscale][1][i]= ROUNDED_DIV(bias<<(16-QUANT_BIAS_SHIFT), qmat16[qscale][0][i]);
108
            }
109
        }
110

    
111
        for(i=intra; i<64; i++){
112
            int64_t max= 8191;
113
            if (dsp->fdct == fdct_ifast
114
#ifndef FAAN_POSTSCALE
115
                   || dsp->fdct == ff_faandct
116
#endif
117
                   ) {
118
                max = (8191LL*ff_aanscales[i]) >> 14;
119
            }
120
            while(((max * qmat[qscale][i]) >> shift) > INT_MAX){
121
                shift++;
122
            }
123
        }
124
    }
125
    if(shift){
126
        av_log(NULL, AV_LOG_INFO, "Warning, QMAT_SHIFT is larger than %d, overflows possible\n", QMAT_SHIFT - shift);
127
    }
128
}
129

    
130
static inline void update_qscale(MpegEncContext *s){
131
    s->qscale= (s->lambda*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7);
132
    s->qscale= av_clip(s->qscale, s->avctx->qmin, s->avctx->qmax);
133

    
134
    s->lambda2= (s->lambda*s->lambda + FF_LAMBDA_SCALE/2) >> FF_LAMBDA_SHIFT;
135
}
136

    
137
void ff_write_quant_matrix(PutBitContext *pb, uint16_t *matrix){
138
    int i;
139

    
140
    if(matrix){
141
        put_bits(pb, 1, 1);
142
        for(i=0;i<64;i++) {
143
            put_bits(pb, 8, matrix[ ff_zigzag_direct[i] ]);
144
        }
145
    }else
146
        put_bits(pb, 1, 0);
147
}
148

    
149
static void copy_picture_attributes(MpegEncContext *s, AVFrame *dst, AVFrame *src){
150
    int i;
151

    
152
    dst->pict_type              = src->pict_type;
153
    dst->quality                = src->quality;
154
    dst->coded_picture_number   = src->coded_picture_number;
155
    dst->display_picture_number = src->display_picture_number;
156
//    dst->reference              = src->reference;
157
    dst->pts                    = src->pts;
158
    dst->interlaced_frame       = src->interlaced_frame;
159
    dst->top_field_first        = src->top_field_first;
160

    
161
    if(s->avctx->me_threshold){
162
        if(!src->motion_val[0])
163
            av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_val not set!\n");
164
        if(!src->mb_type)
165
            av_log(s->avctx, AV_LOG_ERROR, "AVFrame.mb_type not set!\n");
166
        if(!src->ref_index[0])
167
            av_log(s->avctx, AV_LOG_ERROR, "AVFrame.ref_index not set!\n");
168
        if(src->motion_subsample_log2 != dst->motion_subsample_log2)
169
            av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_subsample_log2 doesn't match! (%d!=%d)\n",
170
            src->motion_subsample_log2, dst->motion_subsample_log2);
171

    
172
        memcpy(dst->mb_type, src->mb_type, s->mb_stride * s->mb_height * sizeof(dst->mb_type[0]));
173

    
174
        for(i=0; i<2; i++){
175
            int stride= ((16*s->mb_width )>>src->motion_subsample_log2) + 1;
176
            int height= ((16*s->mb_height)>>src->motion_subsample_log2);
177

    
178
            if(src->motion_val[i] && src->motion_val[i] != dst->motion_val[i]){
179
                memcpy(dst->motion_val[i], src->motion_val[i], 2*stride*height*sizeof(int16_t));
180
            }
181
            if(src->ref_index[i] && src->ref_index[i] != dst->ref_index[i]){
182
                memcpy(dst->ref_index[i], src->ref_index[i], s->b8_stride*2*s->mb_height*sizeof(int8_t));
183
            }
184
        }
185
    }
186
}
187

    
188
static void update_duplicate_context_after_me(MpegEncContext *dst, MpegEncContext *src){
189
#define COPY(a) dst->a= src->a
190
    COPY(pict_type);
191
    COPY(current_picture);
192
    COPY(f_code);
193
    COPY(b_code);
194
    COPY(qscale);
195
    COPY(lambda);
196
    COPY(lambda2);
197
    COPY(picture_in_gop_number);
198
    COPY(gop_picture_number);
199
    COPY(frame_pred_frame_dct); //FIXME don't set in encode_header
200
    COPY(progressive_frame); //FIXME don't set in encode_header
201
    COPY(partitioned_frame); //FIXME don't set in encode_header
202
#undef COPY
203
}
204

    
205
/**
206
 * sets the given MpegEncContext to defaults for encoding.
207
 * the changed fields will not depend upon the prior state of the MpegEncContext.
208
 */
209
static void MPV_encode_defaults(MpegEncContext *s){
210
    int i;
211
    MPV_common_defaults(s);
212

    
213
    for(i=-16; i<16; i++){
214
        default_fcode_tab[i + MAX_MV]= 1;
215
    }
216
    s->me.mv_penalty= default_mv_penalty;
217
    s->fcode_tab= default_fcode_tab;
218
}
219

    
220
/* init video encoder */
221
av_cold int MPV_encode_init(AVCodecContext *avctx)
222
{
223
    MpegEncContext *s = avctx->priv_data;
224
    int i;
225
    int chroma_h_shift, chroma_v_shift;
226

    
227
    MPV_encode_defaults(s);
228

    
229
    switch (avctx->codec_id) {
230
    case CODEC_ID_MPEG2VIDEO:
231
        if(avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P){
232
            av_log(avctx, AV_LOG_ERROR, "only YUV420 and YUV422 are supported\n");
233
            return -1;
234
        }
235
        break;
236
    case CODEC_ID_LJPEG:
237
    case CODEC_ID_MJPEG:
238
        if(avctx->pix_fmt != PIX_FMT_YUVJ420P && avctx->pix_fmt != PIX_FMT_YUVJ422P &&
239
           ((avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P) || avctx->strict_std_compliance>FF_COMPLIANCE_INOFFICIAL)){
240
            av_log(avctx, AV_LOG_ERROR, "colorspace not supported in jpeg\n");
241
            return -1;
242
        }
243
        break;
244
    default:
245
        if(avctx->pix_fmt != PIX_FMT_YUV420P){
246
            av_log(avctx, AV_LOG_ERROR, "only YUV420 is supported\n");
247
            return -1;
248
        }
249
    }
250

    
251
    switch (avctx->pix_fmt) {
252
    case PIX_FMT_YUVJ422P:
253
    case PIX_FMT_YUV422P:
254
        s->chroma_format = CHROMA_422;
255
        break;
256
    case PIX_FMT_YUVJ420P:
257
    case PIX_FMT_YUV420P:
258
    default:
259
        s->chroma_format = CHROMA_420;
260
        break;
261
    }
262

    
263
    s->bit_rate = avctx->bit_rate;
264
    s->width = avctx->width;
265
    s->height = avctx->height;
266
    if(avctx->gop_size > 600 && avctx->strict_std_compliance>FF_COMPLIANCE_EXPERIMENTAL){
267
        av_log(avctx, AV_LOG_ERROR, "Warning keyframe interval too large! reducing it ...\n");
268
        avctx->gop_size=600;
269
    }
270
    s->gop_size = avctx->gop_size;
271
    s->avctx = avctx;
272
    s->flags= avctx->flags;
273
    s->flags2= avctx->flags2;
274
    s->max_b_frames= avctx->max_b_frames;
275
    s->codec_id= avctx->codec->id;
276
    s->luma_elim_threshold  = avctx->luma_elim_threshold;
277
    s->chroma_elim_threshold= avctx->chroma_elim_threshold;
278
    s->strict_std_compliance= avctx->strict_std_compliance;
279
    s->data_partitioning= avctx->flags & CODEC_FLAG_PART;
280
    s->quarter_sample= (avctx->flags & CODEC_FLAG_QPEL)!=0;
281
    s->mpeg_quant= avctx->mpeg_quant;
282
    s->rtp_mode= !!avctx->rtp_payload_size;
283
    s->intra_dc_precision= avctx->intra_dc_precision;
284
    s->user_specified_pts = AV_NOPTS_VALUE;
285

    
286
    if (s->gop_size <= 1) {
287
        s->intra_only = 1;
288
        s->gop_size = 12;
289
    } else {
290
        s->intra_only = 0;
291
    }
292

    
293
    s->me_method = avctx->me_method;
294

    
295
    /* Fixed QSCALE */
296
    s->fixed_qscale = !!(avctx->flags & CODEC_FLAG_QSCALE);
297

    
298
    s->adaptive_quant= (   s->avctx->lumi_masking
299
                        || s->avctx->dark_masking
300
                        || s->avctx->temporal_cplx_masking
301
                        || s->avctx->spatial_cplx_masking
302
                        || s->avctx->p_masking
303
                        || s->avctx->border_masking
304
                        || (s->flags&CODEC_FLAG_QP_RD))
305
                       && !s->fixed_qscale;
306

    
307
    s->obmc= !!(s->flags & CODEC_FLAG_OBMC);
308
    s->loop_filter= !!(s->flags & CODEC_FLAG_LOOP_FILTER);
309
    s->alternate_scan= !!(s->flags & CODEC_FLAG_ALT_SCAN);
310
    s->intra_vlc_format= !!(s->flags2 & CODEC_FLAG2_INTRA_VLC);
311
    s->q_scale_type= !!(s->flags2 & CODEC_FLAG2_NON_LINEAR_QUANT);
312

    
313
    if(avctx->rc_max_rate && !avctx->rc_buffer_size){
314
        av_log(avctx, AV_LOG_ERROR, "a vbv buffer size is needed, for encoding with a maximum bitrate\n");
315
        return -1;
316
    }
317

    
318
    if(avctx->rc_min_rate && avctx->rc_max_rate != avctx->rc_min_rate){
319
        av_log(avctx, AV_LOG_INFO, "Warning min_rate > 0 but min_rate != max_rate isn't recommended!\n");
320
    }
321

    
322
    if(avctx->rc_min_rate && avctx->rc_min_rate > avctx->bit_rate){
323
        av_log(avctx, AV_LOG_ERROR, "bitrate below min bitrate\n");
324
        return -1;
325
    }
326

    
327
    if(avctx->rc_max_rate && avctx->rc_max_rate < avctx->bit_rate){
328
        av_log(avctx, AV_LOG_INFO, "bitrate above max bitrate\n");
329
        return -1;
330
    }
331

    
332
    if(avctx->rc_max_rate && avctx->rc_max_rate == avctx->bit_rate && avctx->rc_max_rate != avctx->rc_min_rate){
333
        av_log(avctx, AV_LOG_INFO, "impossible bitrate constraints, this will fail\n");
334
    }
335

    
336
    if(avctx->rc_buffer_size && avctx->bit_rate*av_q2d(avctx->time_base) > avctx->rc_buffer_size){
337
        av_log(avctx, AV_LOG_ERROR, "VBV buffer too small for bitrate\n");
338
        return -1;
339
    }
340

    
341
    if(avctx->bit_rate*av_q2d(avctx->time_base) > avctx->bit_rate_tolerance){
342
        av_log(avctx, AV_LOG_ERROR, "bitrate tolerance too small for bitrate\n");
343
        return -1;
344
    }
345

    
346
    if(   s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate
347
       && (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO)
348
       && 90000LL * (avctx->rc_buffer_size-1) > s->avctx->rc_max_rate*0xFFFFLL){
349

    
350
        av_log(avctx, AV_LOG_INFO, "Warning vbv_delay will be set to 0xFFFF (=VBR) as the specified vbv buffer is too large for the given bitrate!\n");
351
    }
352

    
353
    if((s->flags & CODEC_FLAG_4MV) && s->codec_id != CODEC_ID_MPEG4
354
       && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P && s->codec_id != CODEC_ID_FLV1){
355
        av_log(avctx, AV_LOG_ERROR, "4MV not supported by codec\n");
356
        return -1;
357
    }
358

    
359
    if(s->obmc && s->avctx->mb_decision != FF_MB_DECISION_SIMPLE){
360
        av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with simple mb decision\n");
361
        return -1;
362
    }
363

    
364
    if(s->obmc && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P){
365
        av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with H263(+)\n");
366
        return -1;
367
    }
368

    
369
    if(s->quarter_sample && s->codec_id != CODEC_ID_MPEG4){
370
        av_log(avctx, AV_LOG_ERROR, "qpel not supported by codec\n");
371
        return -1;
372
    }
373

    
374
    if(s->data_partitioning && s->codec_id != CODEC_ID_MPEG4){
375
        av_log(avctx, AV_LOG_ERROR, "data partitioning not supported by codec\n");
376
        return -1;
377
    }
378

    
379
    if(s->max_b_frames && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO){
380
        av_log(avctx, AV_LOG_ERROR, "b frames not supported by codec\n");
381
        return -1;
382
    }
383

    
384
    if((s->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN))
385
       && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG2VIDEO){
386
        av_log(avctx, AV_LOG_ERROR, "interlacing not supported by codec\n");
387
        return -1;
388
    }
389

    
390
    if(s->mpeg_quant && s->codec_id != CODEC_ID_MPEG4){ //FIXME mpeg2 uses that too
391
        av_log(avctx, AV_LOG_ERROR, "mpeg2 style quantization not supported by codec\n");
392
        return -1;
393
    }
394

    
395
    if((s->flags & CODEC_FLAG_CBP_RD) && !avctx->trellis){
396
        av_log(avctx, AV_LOG_ERROR, "CBP RD needs trellis quant\n");
397
        return -1;
398
    }
399

    
400
    if((s->flags & CODEC_FLAG_QP_RD) && s->avctx->mb_decision != FF_MB_DECISION_RD){
401
        av_log(avctx, AV_LOG_ERROR, "QP RD needs mbd=2\n");
402
        return -1;
403
    }
404

    
405
    if(s->avctx->scenechange_threshold < 1000000000 && (s->flags & CODEC_FLAG_CLOSED_GOP)){
406
        av_log(avctx, AV_LOG_ERROR, "closed gop with scene change detection are not supported yet, set threshold to 1000000000\n");
407
        return -1;
408
    }
409

    
410
    if((s->flags2 & CODEC_FLAG2_INTRA_VLC) && s->codec_id != CODEC_ID_MPEG2VIDEO){
411
        av_log(avctx, AV_LOG_ERROR, "intra vlc table not supported by codec\n");
412
        return -1;
413
    }
414

    
415
    if(s->flags & CODEC_FLAG_LOW_DELAY){
416
        if (s->codec_id != CODEC_ID_MPEG2VIDEO){
417
            av_log(avctx, AV_LOG_ERROR, "low delay forcing is only available for mpeg2\n");
418
            return -1;
419
        }
420
        if (s->max_b_frames != 0){
421
            av_log(avctx, AV_LOG_ERROR, "b frames cannot be used with low delay\n");
422
            return -1;
423
        }
424
    }
425

    
426
    if(s->q_scale_type == 1){
427
        if(s->codec_id != CODEC_ID_MPEG2VIDEO){
428
            av_log(avctx, AV_LOG_ERROR, "non linear quant is only available for mpeg2\n");
429
            return -1;
430
        }
431
        if(avctx->qmax > 12){
432
            av_log(avctx, AV_LOG_ERROR, "non linear quant only supports qmax <= 12 currently\n");
433
            return -1;
434
        }
435
    }
436

    
437
    if(s->avctx->thread_count > 1 && s->codec_id != CODEC_ID_MPEG4
438
       && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO
439
       && (s->codec_id != CODEC_ID_H263P || !(s->flags & CODEC_FLAG_H263P_SLICE_STRUCT))){
440
        av_log(avctx, AV_LOG_ERROR, "multi threaded encoding not supported by codec\n");
441
        return -1;
442
    }
443

    
444
    if(s->avctx->thread_count > 1)
445
        s->rtp_mode= 1;
446

    
447
    if(!avctx->time_base.den || !avctx->time_base.num){
448
        av_log(avctx, AV_LOG_ERROR, "framerate not set\n");
449
        return -1;
450
    }
451

    
452
    i= (INT_MAX/2+128)>>8;
453
    if(avctx->me_threshold >= i){
454
        av_log(avctx, AV_LOG_ERROR, "me_threshold too large, max is %d\n", i - 1);
455
        return -1;
456
    }
457
    if(avctx->mb_threshold >= i){
458
        av_log(avctx, AV_LOG_ERROR, "mb_threshold too large, max is %d\n", i - 1);
459
        return -1;
460
    }
461

    
462
    if(avctx->b_frame_strategy && (avctx->flags&CODEC_FLAG_PASS2)){
463
        av_log(avctx, AV_LOG_INFO, "notice: b_frame_strategy only affects the first pass\n");
464
        avctx->b_frame_strategy = 0;
465
    }
466

    
467
    i= av_gcd(avctx->time_base.den, avctx->time_base.num);
468
    if(i > 1){
469
        av_log(avctx, AV_LOG_INFO, "removing common factors from framerate\n");
470
        avctx->time_base.den /= i;
471
        avctx->time_base.num /= i;
472
//        return -1;
473
    }
474

    
475
    if(s->codec_id==CODEC_ID_MJPEG){
476
        s->intra_quant_bias= 1<<(QUANT_BIAS_SHIFT-1); //(a + x/2)/x
477
        s->inter_quant_bias= 0;
478
    }else if(s->mpeg_quant || s->codec_id==CODEC_ID_MPEG1VIDEO || s->codec_id==CODEC_ID_MPEG2VIDEO){
479
        s->intra_quant_bias= 3<<(QUANT_BIAS_SHIFT-3); //(a + x*3/8)/x
480
        s->inter_quant_bias= 0;
481
    }else{
482
        s->intra_quant_bias=0;
483
        s->inter_quant_bias=-(1<<(QUANT_BIAS_SHIFT-2)); //(a - x/4)/x
484
    }
485

    
486
    if(avctx->intra_quant_bias != FF_DEFAULT_QUANT_BIAS)
487
        s->intra_quant_bias= avctx->intra_quant_bias;
488
    if(avctx->inter_quant_bias != FF_DEFAULT_QUANT_BIAS)
489
        s->inter_quant_bias= avctx->inter_quant_bias;
490

    
491
    avcodec_get_chroma_sub_sample(avctx->pix_fmt, &chroma_h_shift, &chroma_v_shift);
492

    
493
    if(avctx->codec_id == CODEC_ID_MPEG4 && s->avctx->time_base.den > (1<<16)-1){
494
        av_log(avctx, AV_LOG_ERROR, "timebase not supported by mpeg 4 standard\n");
495
        return -1;
496
    }
497
    s->time_increment_bits = av_log2(s->avctx->time_base.den - 1) + 1;
498

    
499
    switch(avctx->codec->id) {
500
    case CODEC_ID_MPEG1VIDEO:
501
        s->out_format = FMT_MPEG1;
502
        s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY);
503
        avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
504
        break;
505
    case CODEC_ID_MPEG2VIDEO:
506
        s->out_format = FMT_MPEG1;
507
        s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY);
508
        avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
509
        s->rtp_mode= 1;
510
        break;
511
    case CODEC_ID_LJPEG:
512
    case CODEC_ID_MJPEG:
513
        s->out_format = FMT_MJPEG;
514
        s->intra_only = 1; /* force intra only for jpeg */
515
        s->mjpeg_vsample[0] = 2;
516
        s->mjpeg_vsample[1] = 2>>chroma_v_shift;
517
        s->mjpeg_vsample[2] = 2>>chroma_v_shift;
518
        s->mjpeg_hsample[0] = 2;
519
        s->mjpeg_hsample[1] = 2>>chroma_h_shift;
520
        s->mjpeg_hsample[2] = 2>>chroma_h_shift;
521
        if (!(CONFIG_MJPEG_ENCODER || CONFIG_LJPEG_ENCODER)
522
            || ff_mjpeg_encode_init(s) < 0)
523
            return -1;
524
        avctx->delay=0;
525
        s->low_delay=1;
526
        break;
527
    case CODEC_ID_H261:
528
        if (!CONFIG_H261_ENCODER)  return -1;
529
        if (ff_h261_get_picture_format(s->width, s->height) < 0) {
530
            av_log(avctx, AV_LOG_ERROR, "The specified picture size of %dx%d is not valid for the H.261 codec.\nValid sizes are 176x144, 352x288\n", s->width, s->height);
531
            return -1;
532
        }
533
        s->out_format = FMT_H261;
534
        avctx->delay=0;
535
        s->low_delay=1;
536
        break;
537
    case CODEC_ID_H263:
538
        if (!CONFIG_H263_ENCODER)  return -1;
539
        if (h263_get_picture_format(s->width, s->height) == 7) {
540
            av_log(avctx, AV_LOG_INFO, "The specified picture size of %dx%d is not valid for the H.263 codec.\nValid sizes are 128x96, 176x144, 352x288, 704x576, and 1408x1152. Try H.263+.\n", s->width, s->height);
541
            return -1;
542
        }
543
        s->out_format = FMT_H263;
544
        s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
545
        avctx->delay=0;
546
        s->low_delay=1;
547
        break;
548
    case CODEC_ID_H263P:
549
        s->out_format = FMT_H263;
550
        s->h263_plus = 1;
551
        /* Fx */
552
        s->umvplus = (avctx->flags & CODEC_FLAG_H263P_UMV) ? 1:0;
553
        s->h263_aic= (avctx->flags & CODEC_FLAG_AC_PRED) ? 1:0;
554
        s->modified_quant= s->h263_aic;
555
        s->alt_inter_vlc= (avctx->flags & CODEC_FLAG_H263P_AIV) ? 1:0;
556
        s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
557
        s->loop_filter= (avctx->flags & CODEC_FLAG_LOOP_FILTER) ? 1:0;
558
        s->unrestricted_mv= s->obmc || s->loop_filter || s->umvplus;
559
        s->h263_slice_structured= (s->flags & CODEC_FLAG_H263P_SLICE_STRUCT) ? 1:0;
560

    
561
        /* /Fx */
562
        /* These are just to be sure */
563
        avctx->delay=0;
564
        s->low_delay=1;
565
        break;
566
    case CODEC_ID_FLV1:
567
        s->out_format = FMT_H263;
568
        s->h263_flv = 2; /* format = 1; 11-bit codes */
569
        s->unrestricted_mv = 1;
570
        s->rtp_mode=0; /* don't allow GOB */
571
        avctx->delay=0;
572
        s->low_delay=1;
573
        break;
574
    case CODEC_ID_RV10:
575
        s->out_format = FMT_H263;
576
        avctx->delay=0;
577
        s->low_delay=1;
578
        break;
579
    case CODEC_ID_RV20:
580
        s->out_format = FMT_H263;
581
        avctx->delay=0;
582
        s->low_delay=1;
583
        s->modified_quant=1;
584
        s->h263_aic=1;
585
        s->h263_plus=1;
586
        s->loop_filter=1;
587
        s->unrestricted_mv= s->obmc || s->loop_filter || s->umvplus;
588
        break;
589
    case CODEC_ID_MPEG4:
590
        s->out_format = FMT_H263;
591
        s->h263_pred = 1;
592
        s->unrestricted_mv = 1;
593
        s->low_delay= s->max_b_frames ? 0 : 1;
594
        avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
595
        break;
596
    case CODEC_ID_MSMPEG4V1:
597
        s->out_format = FMT_H263;
598
        s->h263_msmpeg4 = 1;
599
        s->h263_pred = 1;
600
        s->unrestricted_mv = 1;
601
        s->msmpeg4_version= 1;
602
        avctx->delay=0;
603
        s->low_delay=1;
604
        break;
605
    case CODEC_ID_MSMPEG4V2:
606
        s->out_format = FMT_H263;
607
        s->h263_msmpeg4 = 1;
608
        s->h263_pred = 1;
609
        s->unrestricted_mv = 1;
610
        s->msmpeg4_version= 2;
611
        avctx->delay=0;
612
        s->low_delay=1;
613
        break;
614
    case CODEC_ID_MSMPEG4V3:
615
        s->out_format = FMT_H263;
616
        s->h263_msmpeg4 = 1;
617
        s->h263_pred = 1;
618
        s->unrestricted_mv = 1;
619
        s->msmpeg4_version= 3;
620
        s->flipflop_rounding=1;
621
        avctx->delay=0;
622
        s->low_delay=1;
623
        break;
624
    case CODEC_ID_WMV1:
625
        s->out_format = FMT_H263;
626
        s->h263_msmpeg4 = 1;
627
        s->h263_pred = 1;
628
        s->unrestricted_mv = 1;
629
        s->msmpeg4_version= 4;
630
        s->flipflop_rounding=1;
631
        avctx->delay=0;
632
        s->low_delay=1;
633
        break;
634
    case CODEC_ID_WMV2:
635
        s->out_format = FMT_H263;
636
        s->h263_msmpeg4 = 1;
637
        s->h263_pred = 1;
638
        s->unrestricted_mv = 1;
639
        s->msmpeg4_version= 5;
640
        s->flipflop_rounding=1;
641
        avctx->delay=0;
642
        s->low_delay=1;
643
        break;
644
    default:
645
        return -1;
646
    }
647

    
648
    avctx->has_b_frames= !s->low_delay;
649

    
650
    s->encoding = 1;
651

    
652
    /* init */
653
    if (MPV_common_init(s) < 0)
654
        return -1;
655

    
656
    if(!s->dct_quantize)
657
        s->dct_quantize = dct_quantize_c;
658
    if(!s->denoise_dct)
659
        s->denoise_dct = denoise_dct_c;
660
    s->fast_dct_quantize = s->dct_quantize;
661
    if(avctx->trellis)
662
        s->dct_quantize = dct_quantize_trellis_c;
663

    
664
    if((CONFIG_H263P_ENCODER || CONFIG_RV20_ENCODER) && s->modified_quant)
665
        s->chroma_qscale_table= ff_h263_chroma_qscale_table;
666
    s->progressive_frame=
667
    s->progressive_sequence= !(avctx->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN));
668
    s->quant_precision=5;
669

    
670
    ff_set_cmp(&s->dsp, s->dsp.ildct_cmp, s->avctx->ildct_cmp);
671
    ff_set_cmp(&s->dsp, s->dsp.frame_skip_cmp, s->avctx->frame_skip_cmp);
672

    
673
    if (CONFIG_H261_ENCODER && s->out_format == FMT_H261)
674
        ff_h261_encode_init(s);
675
    if (CONFIG_ANY_H263_ENCODER && s->out_format == FMT_H263)
676
        h263_encode_init(s);
677
    if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version)
678
        ff_msmpeg4_encode_init(s);
679
    if ((CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
680
        && s->out_format == FMT_MPEG1)
681
        ff_mpeg1_encode_init(s);
682

    
683
    /* init q matrix */
684
    for(i=0;i<64;i++) {
685
        int j= s->dsp.idct_permutation[i];
686
        if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4 && s->mpeg_quant){
687
            s->intra_matrix[j] = ff_mpeg4_default_intra_matrix[i];
688
            s->inter_matrix[j] = ff_mpeg4_default_non_intra_matrix[i];
689
        }else if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
690
            s->intra_matrix[j] =
691
            s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
692
        }else
693
        { /* mpeg1/2 */
694
            s->intra_matrix[j] = ff_mpeg1_default_intra_matrix[i];
695
            s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
696
        }
697
        if(s->avctx->intra_matrix)
698
            s->intra_matrix[j] = s->avctx->intra_matrix[i];
699
        if(s->avctx->inter_matrix)
700
            s->inter_matrix[j] = s->avctx->inter_matrix[i];
701
    }
702

    
703
    /* precompute matrix */
704
    /* for mjpeg, we do include qscale in the matrix */
705
    if (s->out_format != FMT_MJPEG) {
706
        ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
707
                       s->intra_matrix, s->intra_quant_bias, avctx->qmin, 31, 1);
708
        ff_convert_matrix(&s->dsp, s->q_inter_matrix, s->q_inter_matrix16,
709
                       s->inter_matrix, s->inter_quant_bias, avctx->qmin, 31, 0);
710
    }
711

    
712
    if(ff_rate_control_init(s) < 0)
713
        return -1;
714

    
715
    return 0;
716
}
717

    
718
av_cold int MPV_encode_end(AVCodecContext *avctx)
719
{
720
    MpegEncContext *s = avctx->priv_data;
721

    
722
    ff_rate_control_uninit(s);
723

    
724
    MPV_common_end(s);
725
    if ((CONFIG_MJPEG_ENCODER || CONFIG_LJPEG_ENCODER) && s->out_format == FMT_MJPEG)
726
        ff_mjpeg_encode_close(s);
727

    
728
    av_freep(&avctx->extradata);
729

    
730
    return 0;
731
}
732

    
733
static int get_sae(uint8_t *src, int ref, int stride){
734
    int x,y;
735
    int acc=0;
736

    
737
    for(y=0; y<16; y++){
738
        for(x=0; x<16; x++){
739
            acc+= FFABS(src[x+y*stride] - ref);
740
        }
741
    }
742

    
743
    return acc;
744
}
745

    
746
static int get_intra_count(MpegEncContext *s, uint8_t *src, uint8_t *ref, int stride){
747
    int x, y, w, h;
748
    int acc=0;
749

    
750
    w= s->width &~15;
751
    h= s->height&~15;
752

    
753
    for(y=0; y<h; y+=16){
754
        for(x=0; x<w; x+=16){
755
            int offset= x + y*stride;
756
            int sad = s->dsp.sad[0](NULL, src + offset, ref + offset, stride, 16);
757
            int mean= (s->dsp.pix_sum(src + offset, stride) + 128)>>8;
758
            int sae = get_sae(src + offset, mean, stride);
759

    
760
            acc+= sae + 500 < sad;
761
        }
762
    }
763
    return acc;
764
}
765

    
766

    
767
static int load_input_picture(MpegEncContext *s, AVFrame *pic_arg){
768
    AVFrame *pic=NULL;
769
    int64_t pts;
770
    int i;
771
    const int encoding_delay= s->max_b_frames;
772
    int direct=1;
773

    
774
    if(pic_arg){
775
        pts= pic_arg->pts;
776
        pic_arg->display_picture_number= s->input_picture_number++;
777

    
778
        if(pts != AV_NOPTS_VALUE){
779
            if(s->user_specified_pts != AV_NOPTS_VALUE){
780
                int64_t time= pts;
781
                int64_t last= s->user_specified_pts;
782

    
783
                if(time <= last){
784
                    av_log(s->avctx, AV_LOG_ERROR, "Error, Invalid timestamp=%"PRId64", last=%"PRId64"\n", pts, s->user_specified_pts);
785
                    return -1;
786
                }
787
            }
788
            s->user_specified_pts= pts;
789
        }else{
790
            if(s->user_specified_pts != AV_NOPTS_VALUE){
791
                s->user_specified_pts=
792
                pts= s->user_specified_pts + 1;
793
                av_log(s->avctx, AV_LOG_INFO, "Warning: AVFrame.pts=? trying to guess (%"PRId64")\n", pts);
794
            }else{
795
                pts= pic_arg->display_picture_number;
796
            }
797
        }
798
    }
799

    
800
  if(pic_arg){
801
    if(encoding_delay && !(s->flags&CODEC_FLAG_INPUT_PRESERVED)) direct=0;
802
    if(pic_arg->linesize[0] != s->linesize) direct=0;
803
    if(pic_arg->linesize[1] != s->uvlinesize) direct=0;
804
    if(pic_arg->linesize[2] != s->uvlinesize) direct=0;
805

    
806
//    av_log(AV_LOG_DEBUG, "%d %d %d %d\n",pic_arg->linesize[0], pic_arg->linesize[1], s->linesize, s->uvlinesize);
807

    
808
    if(direct){
809
        i= ff_find_unused_picture(s, 1);
810

    
811
        pic= (AVFrame*)&s->picture[i];
812
        pic->reference= 3;
813

    
814
        for(i=0; i<4; i++){
815
            pic->data[i]= pic_arg->data[i];
816
            pic->linesize[i]= pic_arg->linesize[i];
817
        }
818
        alloc_picture(s, (Picture*)pic, 1);
819
    }else{
820
        i= ff_find_unused_picture(s, 0);
821

    
822
        pic= (AVFrame*)&s->picture[i];
823
        pic->reference= 3;
824

    
825
        alloc_picture(s, (Picture*)pic, 0);
826

    
827
        if(   pic->data[0] + INPLACE_OFFSET == pic_arg->data[0]
828
           && pic->data[1] + INPLACE_OFFSET == pic_arg->data[1]
829
           && pic->data[2] + INPLACE_OFFSET == pic_arg->data[2]){
830
       // empty
831
        }else{
832
            int h_chroma_shift, v_chroma_shift;
833
            avcodec_get_chroma_sub_sample(s->avctx->pix_fmt, &h_chroma_shift, &v_chroma_shift);
834

    
835
            for(i=0; i<3; i++){
836
                int src_stride= pic_arg->linesize[i];
837
                int dst_stride= i ? s->uvlinesize : s->linesize;
838
                int h_shift= i ? h_chroma_shift : 0;
839
                int v_shift= i ? v_chroma_shift : 0;
840
                int w= s->width >>h_shift;
841
                int h= s->height>>v_shift;
842
                uint8_t *src= pic_arg->data[i];
843
                uint8_t *dst= pic->data[i];
844

    
845
                if(!s->avctx->rc_buffer_size)
846
                    dst +=INPLACE_OFFSET;
847

    
848
                if(src_stride==dst_stride)
849
                    memcpy(dst, src, src_stride*h);
850
                else{
851
                    while(h--){
852
                        memcpy(dst, src, w);
853
                        dst += dst_stride;
854
                        src += src_stride;
855
                    }
856
                }
857
            }
858
        }
859
    }
860
    copy_picture_attributes(s, pic, pic_arg);
861
    pic->pts= pts; //we set this here to avoid modifiying pic_arg
862
  }
863

    
864
    /* shift buffer entries */
865
    for(i=1; i<MAX_PICTURE_COUNT /*s->encoding_delay+1*/; i++)
866
        s->input_picture[i-1]= s->input_picture[i];
867

    
868
    s->input_picture[encoding_delay]= (Picture*)pic;
869

    
870
    return 0;
871
}
872

    
873
static int skip_check(MpegEncContext *s, Picture *p, Picture *ref){
874
    int x, y, plane;
875
    int score=0;
876
    int64_t score64=0;
877

    
878
    for(plane=0; plane<3; plane++){
879
        const int stride= p->linesize[plane];
880
        const int bw= plane ? 1 : 2;
881
        for(y=0; y<s->mb_height*bw; y++){
882
            for(x=0; x<s->mb_width*bw; x++){
883
                int off= p->type == FF_BUFFER_TYPE_SHARED ? 0: 16;
884
                int v= s->dsp.frame_skip_cmp[1](s, p->data[plane] + 8*(x + y*stride)+off, ref->data[plane] + 8*(x + y*stride), stride, 8);
885

    
886
                switch(s->avctx->frame_skip_exp){
887
                    case 0: score= FFMAX(score, v); break;
888
                    case 1: score+= FFABS(v);break;
889
                    case 2: score+= v*v;break;
890
                    case 3: score64+= FFABS(v*v*(int64_t)v);break;
891
                    case 4: score64+= v*v*(int64_t)(v*v);break;
892
                }
893
            }
894
        }
895
    }
896

    
897
    if(score) score64= score;
898

    
899
    if(score64 < s->avctx->frame_skip_threshold)
900
        return 1;
901
    if(score64 < ((s->avctx->frame_skip_factor * (int64_t)s->lambda)>>8))
902
        return 1;
903
    return 0;
904
}
905

    
906
static int estimate_best_b_count(MpegEncContext *s){
907
    AVCodec *codec= avcodec_find_encoder(s->avctx->codec_id);
908
    AVCodecContext *c= avcodec_alloc_context();
909
    AVFrame input[FF_MAX_B_FRAMES+2];
910
    const int scale= s->avctx->brd_scale;
911
    int i, j, out_size, p_lambda, b_lambda, lambda2;
912
    int outbuf_size= s->width * s->height; //FIXME
913
    uint8_t *outbuf= av_malloc(outbuf_size);
914
    int64_t best_rd= INT64_MAX;
915
    int best_b_count= -1;
916

    
917
    assert(scale>=0 && scale <=3);
918

    
919
//    emms_c();
920
    p_lambda= s->last_lambda_for[FF_P_TYPE]; //s->next_picture_ptr->quality;
921
    b_lambda= s->last_lambda_for[FF_B_TYPE]; //p_lambda *FFABS(s->avctx->b_quant_factor) + s->avctx->b_quant_offset;
922
    if(!b_lambda) b_lambda= p_lambda; //FIXME we should do this somewhere else
923
    lambda2= (b_lambda*b_lambda + (1<<FF_LAMBDA_SHIFT)/2 ) >> FF_LAMBDA_SHIFT;
924

    
925
    c->width = s->width >> scale;
926
    c->height= s->height>> scale;
927
    c->flags= CODEC_FLAG_QSCALE | CODEC_FLAG_PSNR | CODEC_FLAG_INPUT_PRESERVED /*| CODEC_FLAG_EMU_EDGE*/;
928
    c->flags|= s->avctx->flags & CODEC_FLAG_QPEL;
929
    c->mb_decision= s->avctx->mb_decision;
930
    c->me_cmp= s->avctx->me_cmp;
931
    c->mb_cmp= s->avctx->mb_cmp;
932
    c->me_sub_cmp= s->avctx->me_sub_cmp;
933
    c->pix_fmt = PIX_FMT_YUV420P;
934
    c->time_base= s->avctx->time_base;
935
    c->max_b_frames= s->max_b_frames;
936

    
937
    if (avcodec_open(c, codec) < 0)
938
        return -1;
939

    
940
    for(i=0; i<s->max_b_frames+2; i++){
941
        int ysize= c->width*c->height;
942
        int csize= (c->width/2)*(c->height/2);
943
        Picture pre_input, *pre_input_ptr= i ? s->input_picture[i-1] : s->next_picture_ptr;
944

    
945
        avcodec_get_frame_defaults(&input[i]);
946
        input[i].data[0]= av_malloc(ysize + 2*csize);
947
        input[i].data[1]= input[i].data[0] + ysize;
948
        input[i].data[2]= input[i].data[1] + csize;
949
        input[i].linesize[0]= c->width;
950
        input[i].linesize[1]=
951
        input[i].linesize[2]= c->width/2;
952

    
953
        if(pre_input_ptr && (!i || s->input_picture[i-1])) {
954
            pre_input= *pre_input_ptr;
955

    
956
            if(pre_input.type != FF_BUFFER_TYPE_SHARED && i) {
957
                pre_input.data[0]+=INPLACE_OFFSET;
958
                pre_input.data[1]+=INPLACE_OFFSET;
959
                pre_input.data[2]+=INPLACE_OFFSET;
960
            }
961

    
962
            s->dsp.shrink[scale](input[i].data[0], input[i].linesize[0], pre_input.data[0], pre_input.linesize[0], c->width, c->height);
963
            s->dsp.shrink[scale](input[i].data[1], input[i].linesize[1], pre_input.data[1], pre_input.linesize[1], c->width>>1, c->height>>1);
964
            s->dsp.shrink[scale](input[i].data[2], input[i].linesize[2], pre_input.data[2], pre_input.linesize[2], c->width>>1, c->height>>1);
965
        }
966
    }
967

    
968
    for(j=0; j<s->max_b_frames+1; j++){
969
        int64_t rd=0;
970

    
971
        if(!s->input_picture[j])
972
            break;
973

    
974
        c->error[0]= c->error[1]= c->error[2]= 0;
975

    
976
        input[0].pict_type= FF_I_TYPE;
977
        input[0].quality= 1 * FF_QP2LAMBDA;
978
        out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[0]);
979
//        rd += (out_size * lambda2) >> FF_LAMBDA_SHIFT;
980

    
981
        for(i=0; i<s->max_b_frames+1; i++){
982
            int is_p= i % (j+1) == j || i==s->max_b_frames;
983

    
984
            input[i+1].pict_type= is_p ? FF_P_TYPE : FF_B_TYPE;
985
            input[i+1].quality= is_p ? p_lambda : b_lambda;
986
            out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[i+1]);
987
            rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
988
        }
989

    
990
        /* get the delayed frames */
991
        while(out_size){
992
            out_size = avcodec_encode_video(c, outbuf, outbuf_size, NULL);
993
            rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
994
        }
995

    
996
        rd += c->error[0] + c->error[1] + c->error[2];
997

    
998
        if(rd < best_rd){
999
            best_rd= rd;
1000
            best_b_count= j;
1001
        }
1002
    }
1003

    
1004
    av_freep(&outbuf);
1005
    avcodec_close(c);
1006
    av_freep(&c);
1007

    
1008
    for(i=0; i<s->max_b_frames+2; i++){
1009
        av_freep(&input[i].data[0]);
1010
    }
1011

    
1012
    return best_b_count;
1013
}
1014

    
1015
static void select_input_picture(MpegEncContext *s){
1016
    int i;
1017

    
1018
    for(i=1; i<MAX_PICTURE_COUNT; i++)
1019
        s->reordered_input_picture[i-1]= s->reordered_input_picture[i];
1020
    s->reordered_input_picture[MAX_PICTURE_COUNT-1]= NULL;
1021

    
1022
    /* set next picture type & ordering */
1023
    if(s->reordered_input_picture[0]==NULL && s->input_picture[0]){
1024
        if(/*s->picture_in_gop_number >= s->gop_size ||*/ s->next_picture_ptr==NULL || s->intra_only){
1025
            s->reordered_input_picture[0]= s->input_picture[0];
1026
            s->reordered_input_picture[0]->pict_type= FF_I_TYPE;
1027
            s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++;
1028
        }else{
1029
            int b_frames;
1030

    
1031
            if(s->avctx->frame_skip_threshold || s->avctx->frame_skip_factor){
1032
                if(s->picture_in_gop_number < s->gop_size && skip_check(s, s->input_picture[0], s->next_picture_ptr)){
1033
                //FIXME check that te gop check above is +-1 correct
1034
//av_log(NULL, AV_LOG_DEBUG, "skip %p %"PRId64"\n", s->input_picture[0]->data[0], s->input_picture[0]->pts);
1035

    
1036
                    if(s->input_picture[0]->type == FF_BUFFER_TYPE_SHARED){
1037
                        for(i=0; i<4; i++)
1038
                            s->input_picture[0]->data[i]= NULL;
1039
                        s->input_picture[0]->type= 0;
1040
                    }else{
1041
                        assert(   s->input_picture[0]->type==FF_BUFFER_TYPE_USER
1042
                               || s->input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
1043

    
1044
                        s->avctx->release_buffer(s->avctx, (AVFrame*)s->input_picture[0]);
1045
                    }
1046

    
1047
                    emms_c();
1048
                    ff_vbv_update(s, 0);
1049

    
1050
                    goto no_output_pic;
1051
                }
1052
            }
1053

    
1054
            if(s->flags&CODEC_FLAG_PASS2){
1055
                for(i=0; i<s->max_b_frames+1; i++){
1056
                    int pict_num= s->input_picture[0]->display_picture_number + i;
1057

    
1058
                    if(pict_num >= s->rc_context.num_entries)
1059
                        break;
1060
                    if(!s->input_picture[i]){
1061
                        s->rc_context.entry[pict_num-1].new_pict_type = FF_P_TYPE;
1062
                        break;
1063
                    }
1064

    
1065
                    s->input_picture[i]->pict_type=
1066
                        s->rc_context.entry[pict_num].new_pict_type;
1067
                }
1068
            }
1069

    
1070
            if(s->avctx->b_frame_strategy==0){
1071
                b_frames= s->max_b_frames;
1072
                while(b_frames && !s->input_picture[b_frames]) b_frames--;
1073
            }else if(s->avctx->b_frame_strategy==1){
1074
                for(i=1; i<s->max_b_frames+1; i++){
1075
                    if(s->input_picture[i] && s->input_picture[i]->b_frame_score==0){
1076
                        s->input_picture[i]->b_frame_score=
1077
                            get_intra_count(s, s->input_picture[i  ]->data[0],
1078
                                               s->input_picture[i-1]->data[0], s->linesize) + 1;
1079
                    }
1080
                }
1081
                for(i=0; i<s->max_b_frames+1; i++){
1082
                    if(s->input_picture[i]==NULL || s->input_picture[i]->b_frame_score - 1 > s->mb_num/s->avctx->b_sensitivity) break;
1083
                }
1084

    
1085
                b_frames= FFMAX(0, i-1);
1086

    
1087
                /* reset scores */
1088
                for(i=0; i<b_frames+1; i++){
1089
                    s->input_picture[i]->b_frame_score=0;
1090
                }
1091
            }else if(s->avctx->b_frame_strategy==2){
1092
                b_frames= estimate_best_b_count(s);
1093
            }else{
1094
                av_log(s->avctx, AV_LOG_ERROR, "illegal b frame strategy\n");
1095
                b_frames=0;
1096
            }
1097

    
1098
            emms_c();
1099
//static int b_count=0;
1100
//b_count+= b_frames;
1101
//av_log(s->avctx, AV_LOG_DEBUG, "b_frames: %d\n", b_count);
1102

    
1103
            for(i= b_frames - 1; i>=0; i--){
1104
                int type= s->input_picture[i]->pict_type;
1105
                if(type && type != FF_B_TYPE)
1106
                    b_frames= i;
1107
            }
1108
            if(s->input_picture[b_frames]->pict_type == FF_B_TYPE && b_frames == s->max_b_frames){
1109
                av_log(s->avctx, AV_LOG_ERROR, "warning, too many b frames in a row\n");
1110
            }
1111

    
1112
            if(s->picture_in_gop_number + b_frames >= s->gop_size){
1113
              if((s->flags2 & CODEC_FLAG2_STRICT_GOP) && s->gop_size > s->picture_in_gop_number){
1114
                    b_frames= s->gop_size - s->picture_in_gop_number - 1;
1115
              }else{
1116
                if(s->flags & CODEC_FLAG_CLOSED_GOP)
1117
                    b_frames=0;
1118
                s->input_picture[b_frames]->pict_type= FF_I_TYPE;
1119
              }
1120
            }
1121

    
1122
            if(   (s->flags & CODEC_FLAG_CLOSED_GOP)
1123
               && b_frames
1124
               && s->input_picture[b_frames]->pict_type== FF_I_TYPE)
1125
                b_frames--;
1126

    
1127
            s->reordered_input_picture[0]= s->input_picture[b_frames];
1128
            if(s->reordered_input_picture[0]->pict_type != FF_I_TYPE)
1129
                s->reordered_input_picture[0]->pict_type= FF_P_TYPE;
1130
            s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++;
1131
            for(i=0; i<b_frames; i++){
1132
                s->reordered_input_picture[i+1]= s->input_picture[i];
1133
                s->reordered_input_picture[i+1]->pict_type= FF_B_TYPE;
1134
                s->reordered_input_picture[i+1]->coded_picture_number= s->coded_picture_number++;
1135
            }
1136
        }
1137
    }
1138
no_output_pic:
1139
    if(s->reordered_input_picture[0]){
1140
        s->reordered_input_picture[0]->reference= s->reordered_input_picture[0]->pict_type!=FF_B_TYPE ? 3 : 0;
1141

    
1142
        ff_copy_picture(&s->new_picture, s->reordered_input_picture[0]);
1143

    
1144
        if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_SHARED || s->avctx->rc_buffer_size){
1145
            // input is a shared pix, so we can't modifiy it -> alloc a new one & ensure that the shared one is reuseable
1146

    
1147
            int i= ff_find_unused_picture(s, 0);
1148
            Picture *pic= &s->picture[i];
1149

    
1150
            pic->reference              = s->reordered_input_picture[0]->reference;
1151
            alloc_picture(s, pic, 0);
1152

    
1153
            /* mark us unused / free shared pic */
1154
            if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_INTERNAL)
1155
                s->avctx->release_buffer(s->avctx, (AVFrame*)s->reordered_input_picture[0]);
1156
            for(i=0; i<4; i++)
1157
                s->reordered_input_picture[0]->data[i]= NULL;
1158
            s->reordered_input_picture[0]->type= 0;
1159

    
1160
            copy_picture_attributes(s, (AVFrame*)pic, (AVFrame*)s->reordered_input_picture[0]);
1161

    
1162
            s->current_picture_ptr= pic;
1163
        }else{
1164
            // input is not a shared pix -> reuse buffer for current_pix
1165

    
1166
            assert(   s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_USER
1167
                   || s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
1168

    
1169
            s->current_picture_ptr= s->reordered_input_picture[0];
1170
            for(i=0; i<4; i++){
1171
                s->new_picture.data[i]+= INPLACE_OFFSET;
1172
            }
1173
        }
1174
        ff_copy_picture(&s->current_picture, s->current_picture_ptr);
1175

    
1176
        s->picture_number= s->new_picture.display_picture_number;
1177
//printf("dpn:%d\n", s->picture_number);
1178
    }else{
1179
       memset(&s->new_picture, 0, sizeof(Picture));
1180
    }
1181
}
1182

    
1183
int MPV_encode_picture(AVCodecContext *avctx,
1184
                       unsigned char *buf, int buf_size, void *data)
1185
{
1186
    MpegEncContext *s = avctx->priv_data;
1187
    AVFrame *pic_arg = data;
1188
    int i, stuffing_count;
1189

    
1190
    for(i=0; i<avctx->thread_count; i++){
1191
        int start_y= s->thread_context[i]->start_mb_y;
1192
        int   end_y= s->thread_context[i]->  end_mb_y;
1193
        int h= s->mb_height;
1194
        uint8_t *start= buf + (size_t)(((int64_t) buf_size)*start_y/h);
1195
        uint8_t *end  = buf + (size_t)(((int64_t) buf_size)*  end_y/h);
1196

    
1197
        init_put_bits(&s->thread_context[i]->pb, start, end - start);
1198
    }
1199

    
1200
    s->picture_in_gop_number++;
1201

    
1202
    if(load_input_picture(s, pic_arg) < 0)
1203
        return -1;
1204

    
1205
    select_input_picture(s);
1206

    
1207
    /* output? */
1208
    if(s->new_picture.data[0]){
1209
        s->pict_type= s->new_picture.pict_type;
1210
//emms_c();
1211
//printf("qs:%f %f %d\n", s->new_picture.quality, s->current_picture.quality, s->qscale);
1212
        MPV_frame_start(s, avctx);
1213
vbv_retry:
1214
        if (encode_picture(s, s->picture_number) < 0)
1215
            return -1;
1216

    
1217
        avctx->real_pict_num  = s->picture_number;
1218
        avctx->header_bits = s->header_bits;
1219
        avctx->mv_bits     = s->mv_bits;
1220
        avctx->misc_bits   = s->misc_bits;
1221
        avctx->i_tex_bits  = s->i_tex_bits;
1222
        avctx->p_tex_bits  = s->p_tex_bits;
1223
        avctx->i_count     = s->i_count;
1224
        avctx->p_count     = s->mb_num - s->i_count - s->skip_count; //FIXME f/b_count in avctx
1225
        avctx->skip_count  = s->skip_count;
1226

    
1227
        MPV_frame_end(s);
1228

    
1229
        if (CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG)
1230
            ff_mjpeg_encode_picture_trailer(s);
1231

    
1232
        if(avctx->rc_buffer_size){
1233
            RateControlContext *rcc= &s->rc_context;
1234
            int max_size= rcc->buffer_index * avctx->rc_max_available_vbv_use;
1235

    
1236
            if(put_bits_count(&s->pb) > max_size && s->lambda < s->avctx->lmax){
1237
                s->next_lambda= FFMAX(s->lambda+1, s->lambda*(s->qscale+1) / s->qscale);
1238
                if(s->adaptive_quant){
1239
                    int i;
1240
                    for(i=0; i<s->mb_height*s->mb_stride; i++)
1241
                        s->lambda_table[i]= FFMAX(s->lambda_table[i]+1, s->lambda_table[i]*(s->qscale+1) / s->qscale);
1242
                }
1243
                s->mb_skipped = 0;        //done in MPV_frame_start()
1244
                if(s->pict_type==FF_P_TYPE){ //done in encode_picture() so we must undo it
1245
                    if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
1246
                        s->no_rounding ^= 1;
1247
                }
1248
                if(s->pict_type!=FF_B_TYPE){
1249
                    s->time_base= s->last_time_base;
1250
                    s->last_non_b_time= s->time - s->pp_time;
1251
                }
1252
//                av_log(NULL, AV_LOG_ERROR, "R:%d ", s->next_lambda);
1253
                for(i=0; i<avctx->thread_count; i++){
1254
                    PutBitContext *pb= &s->thread_context[i]->pb;
1255
                    init_put_bits(pb, pb->buf, pb->buf_end - pb->buf);
1256
                }
1257
                goto vbv_retry;
1258
            }
1259

    
1260
            assert(s->avctx->rc_max_rate);
1261
        }
1262

    
1263
        if(s->flags&CODEC_FLAG_PASS1)
1264
            ff_write_pass1_stats(s);
1265

    
1266
        for(i=0; i<4; i++){
1267
            s->current_picture_ptr->error[i]= s->current_picture.error[i];
1268
            avctx->error[i] += s->current_picture_ptr->error[i];
1269
        }
1270

    
1271
        if(s->flags&CODEC_FLAG_PASS1)
1272
            assert(avctx->header_bits + avctx->mv_bits + avctx->misc_bits + avctx->i_tex_bits + avctx->p_tex_bits == put_bits_count(&s->pb));
1273
        flush_put_bits(&s->pb);
1274
        s->frame_bits  = put_bits_count(&s->pb);
1275

    
1276
        stuffing_count= ff_vbv_update(s, s->frame_bits);
1277
        if(stuffing_count){
1278
            if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < stuffing_count + 50){
1279
                av_log(s->avctx, AV_LOG_ERROR, "stuffing too large\n");
1280
                return -1;
1281
            }
1282

    
1283
            switch(s->codec_id){
1284
            case CODEC_ID_MPEG1VIDEO:
1285
            case CODEC_ID_MPEG2VIDEO:
1286
                while(stuffing_count--){
1287
                    put_bits(&s->pb, 8, 0);
1288
                }
1289
            break;
1290
            case CODEC_ID_MPEG4:
1291
                put_bits(&s->pb, 16, 0);
1292
                put_bits(&s->pb, 16, 0x1C3);
1293
                stuffing_count -= 4;
1294
                while(stuffing_count--){
1295
                    put_bits(&s->pb, 8, 0xFF);
1296
                }
1297
            break;
1298
            default:
1299
                av_log(s->avctx, AV_LOG_ERROR, "vbv buffer overflow\n");
1300
            }
1301
            flush_put_bits(&s->pb);
1302
            s->frame_bits  = put_bits_count(&s->pb);
1303
        }
1304

    
1305
        /* update mpeg1/2 vbv_delay for CBR */
1306
        if(s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate && s->out_format == FMT_MPEG1
1307
           && 90000LL * (avctx->rc_buffer_size-1) <= s->avctx->rc_max_rate*0xFFFFLL){
1308
            int vbv_delay;
1309

    
1310
            assert(s->repeat_first_field==0);
1311

    
1312
            vbv_delay= lrintf(90000 * s->rc_context.buffer_index / s->avctx->rc_max_rate);
1313
            assert(vbv_delay < 0xFFFF);
1314

    
1315
            s->vbv_delay_ptr[0] &= 0xF8;
1316
            s->vbv_delay_ptr[0] |= vbv_delay>>13;
1317
            s->vbv_delay_ptr[1]  = vbv_delay>>5;
1318
            s->vbv_delay_ptr[2] &= 0x07;
1319
            s->vbv_delay_ptr[2] |= vbv_delay<<3;
1320
        }
1321
        s->total_bits += s->frame_bits;
1322
        avctx->frame_bits  = s->frame_bits;
1323
    }else{
1324
        assert((pbBufPtr(&s->pb) == s->pb.buf));
1325
        s->frame_bits=0;
1326
    }
1327
    assert((s->frame_bits&7)==0);
1328

    
1329
    return s->frame_bits/8;
1330
}
1331

    
1332
static inline void dct_single_coeff_elimination(MpegEncContext *s, int n, int threshold)
1333
{
1334
    static const char tab[64]=
1335
        {3,2,2,1,1,1,1,1,
1336
         1,1,1,1,1,1,1,1,
1337
         1,1,1,1,1,1,1,1,
1338
         0,0,0,0,0,0,0,0,
1339
         0,0,0,0,0,0,0,0,
1340
         0,0,0,0,0,0,0,0,
1341
         0,0,0,0,0,0,0,0,
1342
         0,0,0,0,0,0,0,0};
1343
    int score=0;
1344
    int run=0;
1345
    int i;
1346
    DCTELEM *block= s->block[n];
1347
    const int last_index= s->block_last_index[n];
1348
    int skip_dc;
1349

    
1350
    if(threshold<0){
1351
        skip_dc=0;
1352
        threshold= -threshold;
1353
    }else
1354
        skip_dc=1;
1355

    
1356
    /* Are all we could set to zero already zero? */
1357
    if(last_index<=skip_dc - 1) return;
1358

    
1359
    for(i=0; i<=last_index; i++){
1360
        const int j = s->intra_scantable.permutated[i];
1361
        const int level = FFABS(block[j]);
1362
        if(level==1){
1363
            if(skip_dc && i==0) continue;
1364
            score+= tab[run];
1365
            run=0;
1366
        }else if(level>1){
1367
            return;
1368
        }else{
1369
            run++;
1370
        }
1371
    }
1372
    if(score >= threshold) return;
1373
    for(i=skip_dc; i<=last_index; i++){
1374
        const int j = s->intra_scantable.permutated[i];
1375
        block[j]=0;
1376
    }
1377
    if(block[0]) s->block_last_index[n]= 0;
1378
    else         s->block_last_index[n]= -1;
1379
}
1380

    
1381
static inline void clip_coeffs(MpegEncContext *s, DCTELEM *block, int last_index)
1382
{
1383
    int i;
1384
    const int maxlevel= s->max_qcoeff;
1385
    const int minlevel= s->min_qcoeff;
1386
    int overflow=0;
1387

    
1388
    if(s->mb_intra){
1389
        i=1; //skip clipping of intra dc
1390
    }else
1391
        i=0;
1392

    
1393
    for(;i<=last_index; i++){
1394
        const int j= s->intra_scantable.permutated[i];
1395
        int level = block[j];
1396

    
1397
        if     (level>maxlevel){
1398
            level=maxlevel;
1399
            overflow++;
1400
        }else if(level<minlevel){
1401
            level=minlevel;
1402
            overflow++;
1403
        }
1404

    
1405
        block[j]= level;
1406
    }
1407

    
1408
    if(overflow && s->avctx->mb_decision == FF_MB_DECISION_SIMPLE)
1409
        av_log(s->avctx, AV_LOG_INFO, "warning, clipping %d dct coefficients to %d..%d\n", overflow, minlevel, maxlevel);
1410
}
1411

    
1412
static void get_visual_weight(int16_t *weight, uint8_t *ptr, int stride){
1413
    int x, y;
1414
//FIXME optimize
1415
    for(y=0; y<8; y++){
1416
        for(x=0; x<8; x++){
1417
            int x2, y2;
1418
            int sum=0;
1419
            int sqr=0;
1420
            int count=0;
1421

    
1422
            for(y2= FFMAX(y-1, 0); y2 < FFMIN(8, y+2); y2++){
1423
                for(x2= FFMAX(x-1, 0); x2 < FFMIN(8, x+2); x2++){
1424
                    int v= ptr[x2 + y2*stride];
1425
                    sum += v;
1426
                    sqr += v*v;
1427
                    count++;
1428
                }
1429
            }
1430
            weight[x + 8*y]= (36*ff_sqrt(count*sqr - sum*sum)) / count;
1431
        }
1432
    }
1433
}
1434

    
1435
static av_always_inline void encode_mb_internal(MpegEncContext *s, int motion_x, int motion_y, int mb_block_height, int mb_block_count)
1436
{
1437
    int16_t weight[8][64];
1438
    DCTELEM orig[8][64];
1439
    const int mb_x= s->mb_x;
1440
    const int mb_y= s->mb_y;
1441
    int i;
1442
    int skip_dct[8];
1443
    int dct_offset   = s->linesize*8; //default for progressive frames
1444
    uint8_t *ptr_y, *ptr_cb, *ptr_cr;
1445
    int wrap_y, wrap_c;
1446

    
1447
    for(i=0; i<mb_block_count; i++) skip_dct[i]=s->skipdct;
1448

    
1449
    if(s->adaptive_quant){
1450
        const int last_qp= s->qscale;
1451
        const int mb_xy= mb_x + mb_y*s->mb_stride;
1452

    
1453
        s->lambda= s->lambda_table[mb_xy];
1454
        update_qscale(s);
1455

    
1456
        if(!(s->flags&CODEC_FLAG_QP_RD)){
1457
            s->qscale= s->current_picture_ptr->qscale_table[mb_xy];
1458
            s->dquant= s->qscale - last_qp;
1459

    
1460
            if(s->out_format==FMT_H263){
1461
                s->dquant= av_clip(s->dquant, -2, 2);
1462

    
1463
                if(s->codec_id==CODEC_ID_MPEG4){
1464
                    if(!s->mb_intra){
1465
                        if(s->pict_type == FF_B_TYPE){
1466
                            if(s->dquant&1 || s->mv_dir&MV_DIRECT)
1467
                                s->dquant= 0;
1468
                        }
1469
                        if(s->mv_type==MV_TYPE_8X8)
1470
                            s->dquant=0;
1471
                    }
1472
                }
1473
            }
1474
        }
1475
        ff_set_qscale(s, last_qp + s->dquant);
1476
    }else if(s->flags&CODEC_FLAG_QP_RD)
1477
        ff_set_qscale(s, s->qscale + s->dquant);
1478

    
1479
    wrap_y = s->linesize;
1480
    wrap_c = s->uvlinesize;
1481
    ptr_y = s->new_picture.data[0] + (mb_y * 16 * wrap_y) + mb_x * 16;
1482
    ptr_cb = s->new_picture.data[1] + (mb_y * mb_block_height * wrap_c) + mb_x * 8;
1483
    ptr_cr = s->new_picture.data[2] + (mb_y * mb_block_height * wrap_c) + mb_x * 8;
1484

    
1485
    if(mb_x*16+16 > s->width || mb_y*16+16 > s->height){
1486
        uint8_t *ebuf= s->edge_emu_buffer + 32;
1487
        ff_emulated_edge_mc(ebuf            , ptr_y , wrap_y,16,16,mb_x*16,mb_y*16, s->width   , s->height);
1488
        ptr_y= ebuf;
1489
        ff_emulated_edge_mc(ebuf+18*wrap_y  , ptr_cb, wrap_c, 8, mb_block_height, mb_x*8, mb_y*8, s->width>>1, s->height>>1);
1490
        ptr_cb= ebuf+18*wrap_y;
1491
        ff_emulated_edge_mc(ebuf+18*wrap_y+8, ptr_cr, wrap_c, 8, mb_block_height, mb_x*8, mb_y*8, s->width>>1, s->height>>1);
1492
        ptr_cr= ebuf+18*wrap_y+8;
1493
    }
1494

    
1495
    if (s->mb_intra) {
1496
        if(s->flags&CODEC_FLAG_INTERLACED_DCT){
1497
            int progressive_score, interlaced_score;
1498

    
1499
            s->interlaced_dct=0;
1500
            progressive_score= s->dsp.ildct_cmp[4](s, ptr_y           , NULL, wrap_y, 8)
1501
                              +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y*8, NULL, wrap_y, 8) - 400;
1502

    
1503
            if(progressive_score > 0){
1504
                interlaced_score = s->dsp.ildct_cmp[4](s, ptr_y           , NULL, wrap_y*2, 8)
1505
                                  +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y  , NULL, wrap_y*2, 8);
1506
                if(progressive_score > interlaced_score){
1507
                    s->interlaced_dct=1;
1508

    
1509
                    dct_offset= wrap_y;
1510
                    wrap_y<<=1;
1511
                    if (s->chroma_format == CHROMA_422)
1512
                        wrap_c<<=1;
1513
                }
1514
            }
1515
        }
1516

    
1517
        s->dsp.get_pixels(s->block[0], ptr_y                 , wrap_y);
1518
        s->dsp.get_pixels(s->block[1], ptr_y              + 8, wrap_y);
1519
        s->dsp.get_pixels(s->block[2], ptr_y + dct_offset    , wrap_y);
1520
        s->dsp.get_pixels(s->block[3], ptr_y + dct_offset + 8, wrap_y);
1521

    
1522
        if(s->flags&CODEC_FLAG_GRAY){
1523
            skip_dct[4]= 1;
1524
            skip_dct[5]= 1;
1525
        }else{
1526
            s->dsp.get_pixels(s->block[4], ptr_cb, wrap_c);
1527
            s->dsp.get_pixels(s->block[5], ptr_cr, wrap_c);
1528
            if(!s->chroma_y_shift){ /* 422 */
1529
                s->dsp.get_pixels(s->block[6], ptr_cb + (dct_offset>>1), wrap_c);
1530
                s->dsp.get_pixels(s->block[7], ptr_cr + (dct_offset>>1), wrap_c);
1531
            }
1532
        }
1533
    }else{
1534
        op_pixels_func (*op_pix)[4];
1535
        qpel_mc_func (*op_qpix)[16];
1536
        uint8_t *dest_y, *dest_cb, *dest_cr;
1537

    
1538
        dest_y  = s->dest[0];
1539
        dest_cb = s->dest[1];
1540
        dest_cr = s->dest[2];
1541

    
1542
        if ((!s->no_rounding) || s->pict_type==FF_B_TYPE){
1543
            op_pix = s->dsp.put_pixels_tab;
1544
            op_qpix= s->dsp.put_qpel_pixels_tab;
1545
        }else{
1546
            op_pix = s->dsp.put_no_rnd_pixels_tab;
1547
            op_qpix= s->dsp.put_no_rnd_qpel_pixels_tab;
1548
        }
1549

    
1550
        if (s->mv_dir & MV_DIR_FORWARD) {
1551
            MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.data, op_pix, op_qpix);
1552
            op_pix = s->dsp.avg_pixels_tab;
1553
            op_qpix= s->dsp.avg_qpel_pixels_tab;
1554
        }
1555
        if (s->mv_dir & MV_DIR_BACKWARD) {
1556
            MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.data, op_pix, op_qpix);
1557
        }
1558

    
1559
        if(s->flags&CODEC_FLAG_INTERLACED_DCT){
1560
            int progressive_score, interlaced_score;
1561

    
1562
            s->interlaced_dct=0;
1563
            progressive_score= s->dsp.ildct_cmp[0](s, dest_y           , ptr_y           , wrap_y, 8)
1564
                              +s->dsp.ildct_cmp[0](s, dest_y + wrap_y*8, ptr_y + wrap_y*8, wrap_y, 8) - 400;
1565

    
1566
            if(s->avctx->ildct_cmp == FF_CMP_VSSE) progressive_score -= 400;
1567

    
1568
            if(progressive_score>0){
1569
                interlaced_score = s->dsp.ildct_cmp[0](s, dest_y           , ptr_y           , wrap_y*2, 8)
1570
                                  +s->dsp.ildct_cmp[0](s, dest_y + wrap_y  , ptr_y + wrap_y  , wrap_y*2, 8);
1571

    
1572
                if(progressive_score > interlaced_score){
1573
                    s->interlaced_dct=1;
1574

    
1575
                    dct_offset= wrap_y;
1576
                    wrap_y<<=1;
1577
                    if (s->chroma_format == CHROMA_422)
1578
                        wrap_c<<=1;
1579
                }
1580
            }
1581
        }
1582

    
1583
        s->dsp.diff_pixels(s->block[0], ptr_y                 , dest_y                 , wrap_y);
1584
        s->dsp.diff_pixels(s->block[1], ptr_y              + 8, dest_y              + 8, wrap_y);
1585
        s->dsp.diff_pixels(s->block[2], ptr_y + dct_offset    , dest_y + dct_offset    , wrap_y);
1586
        s->dsp.diff_pixels(s->block[3], ptr_y + dct_offset + 8, dest_y + dct_offset + 8, wrap_y);
1587

    
1588
        if(s->flags&CODEC_FLAG_GRAY){
1589
            skip_dct[4]= 1;
1590
            skip_dct[5]= 1;
1591
        }else{
1592
            s->dsp.diff_pixels(s->block[4], ptr_cb, dest_cb, wrap_c);
1593
            s->dsp.diff_pixels(s->block[5], ptr_cr, dest_cr, wrap_c);
1594
            if(!s->chroma_y_shift){ /* 422 */
1595
                s->dsp.diff_pixels(s->block[6], ptr_cb + (dct_offset>>1), dest_cb + (dct_offset>>1), wrap_c);
1596
                s->dsp.diff_pixels(s->block[7], ptr_cr + (dct_offset>>1), dest_cr + (dct_offset>>1), wrap_c);
1597
            }
1598
        }
1599
        /* pre quantization */
1600
        if(s->current_picture.mc_mb_var[s->mb_stride*mb_y+ mb_x]<2*s->qscale*s->qscale){
1601
            //FIXME optimize
1602
            if(s->dsp.sad[1](NULL, ptr_y               , dest_y               , wrap_y, 8) < 20*s->qscale) skip_dct[0]= 1;
1603
            if(s->dsp.sad[1](NULL, ptr_y            + 8, dest_y            + 8, wrap_y, 8) < 20*s->qscale) skip_dct[1]= 1;
1604
            if(s->dsp.sad[1](NULL, ptr_y +dct_offset   , dest_y +dct_offset   , wrap_y, 8) < 20*s->qscale) skip_dct[2]= 1;
1605
            if(s->dsp.sad[1](NULL, ptr_y +dct_offset+ 8, dest_y +dct_offset+ 8, wrap_y, 8) < 20*s->qscale) skip_dct[3]= 1;
1606
            if(s->dsp.sad[1](NULL, ptr_cb              , dest_cb              , wrap_c, 8) < 20*s->qscale) skip_dct[4]= 1;
1607
            if(s->dsp.sad[1](NULL, ptr_cr              , dest_cr              , wrap_c, 8) < 20*s->qscale) skip_dct[5]= 1;
1608
            if(!s->chroma_y_shift){ /* 422 */
1609
                if(s->dsp.sad[1](NULL, ptr_cb +(dct_offset>>1), dest_cb +(dct_offset>>1), wrap_c, 8) < 20*s->qscale) skip_dct[6]= 1;
1610
                if(s->dsp.sad[1](NULL, ptr_cr +(dct_offset>>1), dest_cr +(dct_offset>>1), wrap_c, 8) < 20*s->qscale) skip_dct[7]= 1;
1611
            }
1612
        }
1613
    }
1614

    
1615
    if(s->avctx->quantizer_noise_shaping){
1616
        if(!skip_dct[0]) get_visual_weight(weight[0], ptr_y                 , wrap_y);
1617
        if(!skip_dct[1]) get_visual_weight(weight[1], ptr_y              + 8, wrap_y);
1618
        if(!skip_dct[2]) get_visual_weight(weight[2], ptr_y + dct_offset    , wrap_y);
1619
        if(!skip_dct[3]) get_visual_weight(weight[3], ptr_y + dct_offset + 8, wrap_y);
1620
        if(!skip_dct[4]) get_visual_weight(weight[4], ptr_cb                , wrap_c);
1621
        if(!skip_dct[5]) get_visual_weight(weight[5], ptr_cr                , wrap_c);
1622
        if(!s->chroma_y_shift){ /* 422 */
1623
            if(!skip_dct[6]) get_visual_weight(weight[6], ptr_cb + (dct_offset>>1), wrap_c);
1624
            if(!skip_dct[7]) get_visual_weight(weight[7], ptr_cr + (dct_offset>>1), wrap_c);
1625
        }
1626
        memcpy(orig[0], s->block[0], sizeof(DCTELEM)*64*mb_block_count);
1627
    }
1628

    
1629
    /* DCT & quantize */
1630
    assert(s->out_format!=FMT_MJPEG || s->qscale==8);
1631
    {
1632
        for(i=0;i<mb_block_count;i++) {
1633
            if(!skip_dct[i]){
1634
                int overflow;
1635
                s->block_last_index[i] = s->dct_quantize(s, s->block[i], i, s->qscale, &overflow);
1636
            // FIXME we could decide to change to quantizer instead of clipping
1637
            // JS: I don't think that would be a good idea it could lower quality instead
1638
            //     of improve it. Just INTRADC clipping deserves changes in quantizer
1639
                if (overflow) clip_coeffs(s, s->block[i], s->block_last_index[i]);
1640
            }else
1641
                s->block_last_index[i]= -1;
1642
        }
1643
        if(s->avctx->quantizer_noise_shaping){
1644
            for(i=0;i<mb_block_count;i++) {
1645
                if(!skip_dct[i]){
1646
                    s->block_last_index[i] = dct_quantize_refine(s, s->block[i], weight[i], orig[i], i, s->qscale);
1647
                }
1648
            }
1649
        }
1650

    
1651
        if(s->luma_elim_threshold && !s->mb_intra)
1652
            for(i=0; i<4; i++)
1653
                dct_single_coeff_elimination(s, i, s->luma_elim_threshold);
1654
        if(s->chroma_elim_threshold && !s->mb_intra)
1655
            for(i=4; i<mb_block_count; i++)
1656
                dct_single_coeff_elimination(s, i, s->chroma_elim_threshold);
1657

    
1658
        if(s->flags & CODEC_FLAG_CBP_RD){
1659
            for(i=0;i<mb_block_count;i++) {
1660
                if(s->block_last_index[i] == -1)
1661
                    s->coded_score[i]= INT_MAX/256;
1662
            }
1663
        }
1664
    }
1665

    
1666
    if((s->flags&CODEC_FLAG_GRAY) && s->mb_intra){
1667
        s->block_last_index[4]=
1668
        s->block_last_index[5]= 0;
1669
        s->block[4][0]=
1670
        s->block[5][0]= (1024 + s->c_dc_scale/2)/ s->c_dc_scale;
1671
    }
1672

    
1673
    //non c quantize code returns incorrect block_last_index FIXME
1674
    if(s->alternate_scan && s->dct_quantize != dct_quantize_c){
1675
        for(i=0; i<mb_block_count; i++){
1676
            int j;
1677
            if(s->block_last_index[i]>0){
1678
                for(j=63; j>0; j--){
1679
                    if(s->block[i][ s->intra_scantable.permutated[j] ]) break;
1680
                }
1681
                s->block_last_index[i]= j;
1682
            }
1683
        }
1684
    }
1685

    
1686
    /* huffman encode */
1687
    switch(s->codec_id){ //FIXME funct ptr could be slightly faster
1688
    case CODEC_ID_MPEG1VIDEO:
1689
    case CODEC_ID_MPEG2VIDEO:
1690
        if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
1691
            mpeg1_encode_mb(s, s->block, motion_x, motion_y);
1692
        break;
1693
    case CODEC_ID_MPEG4:
1694
        if (CONFIG_MPEG4_ENCODER)
1695
            mpeg4_encode_mb(s, s->block, motion_x, motion_y);
1696
        break;
1697
    case CODEC_ID_MSMPEG4V2:
1698
    case CODEC_ID_MSMPEG4V3:
1699
    case CODEC_ID_WMV1:
1700
        if (CONFIG_MSMPEG4_ENCODER)
1701
            msmpeg4_encode_mb(s, s->block, motion_x, motion_y);
1702
        break;
1703
    case CODEC_ID_WMV2:
1704
        if (CONFIG_WMV2_ENCODER)
1705
            ff_wmv2_encode_mb(s, s->block, motion_x, motion_y);
1706
        break;
1707
    case CODEC_ID_H261:
1708
        if (CONFIG_H261_ENCODER)
1709
            ff_h261_encode_mb(s, s->block, motion_x, motion_y);
1710
        break;
1711
    case CODEC_ID_H263:
1712
    case CODEC_ID_H263P:
1713
    case CODEC_ID_FLV1:
1714
    case CODEC_ID_RV10:
1715
    case CODEC_ID_RV20:
1716
        if (CONFIG_H263_ENCODER || CONFIG_H263P_ENCODER ||
1717
            CONFIG_FLV_ENCODER  || CONFIG_RV10_ENCODER  || CONFIG_RV20_ENCODER)
1718
            h263_encode_mb(s, s->block, motion_x, motion_y);
1719
        break;
1720
    case CODEC_ID_MJPEG:
1721
        if (CONFIG_MJPEG_ENCODER)
1722
            ff_mjpeg_encode_mb(s, s->block);
1723
        break;
1724
    default:
1725
        assert(0);
1726
    }
1727
}
1728

    
1729
static av_always_inline void encode_mb(MpegEncContext *s, int motion_x, int motion_y)
1730
{
1731
    if (s->chroma_format == CHROMA_420) encode_mb_internal(s, motion_x, motion_y,  8, 6);
1732
    else                                encode_mb_internal(s, motion_x, motion_y, 16, 8);
1733
}
1734

    
1735
static inline void copy_context_before_encode(MpegEncContext *d, MpegEncContext *s, int type){
1736
    int i;
1737

    
1738
    memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
1739

    
1740
    /* mpeg1 */
1741
    d->mb_skip_run= s->mb_skip_run;
1742
    for(i=0; i<3; i++)
1743
        d->last_dc[i]= s->last_dc[i];
1744

    
1745
    /* statistics */
1746
    d->mv_bits= s->mv_bits;
1747
    d->i_tex_bits= s->i_tex_bits;
1748
    d->p_tex_bits= s->p_tex_bits;
1749
    d->i_count= s->i_count;
1750
    d->f_count= s->f_count;
1751
    d->b_count= s->b_count;
1752
    d->skip_count= s->skip_count;
1753
    d->misc_bits= s->misc_bits;
1754
    d->last_bits= 0;
1755

    
1756
    d->mb_skipped= 0;
1757
    d->qscale= s->qscale;
1758
    d->dquant= s->dquant;
1759

    
1760
    d->esc3_level_length= s->esc3_level_length;
1761
}
1762

    
1763
static inline void copy_context_after_encode(MpegEncContext *d, MpegEncContext *s, int type){
1764
    int i;
1765

    
1766
    memcpy(d->mv, s->mv, 2*4*2*sizeof(int));
1767
    memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
1768

    
1769
    /* mpeg1 */
1770
    d->mb_skip_run= s->mb_skip_run;
1771
    for(i=0; i<3; i++)
1772
        d->last_dc[i]= s->last_dc[i];
1773

    
1774
    /* statistics */
1775
    d->mv_bits= s->mv_bits;
1776
    d->i_tex_bits= s->i_tex_bits;
1777
    d->p_tex_bits= s->p_tex_bits;
1778
    d->i_count= s->i_count;
1779
    d->f_count= s->f_count;
1780
    d->b_count= s->b_count;
1781
    d->skip_count= s->skip_count;
1782
    d->misc_bits= s->misc_bits;
1783

    
1784
    d->mb_intra= s->mb_intra;
1785
    d->mb_skipped= s->mb_skipped;
1786
    d->mv_type= s->mv_type;
1787
    d->mv_dir= s->mv_dir;
1788
    d->pb= s->pb;
1789
    if(s->data_partitioning){
1790
        d->pb2= s->pb2;
1791
        d->tex_pb= s->tex_pb;
1792
    }
1793
    d->block= s->block;
1794
    for(i=0; i<8; i++)
1795
        d->block_last_index[i]= s->block_last_index[i];
1796
    d->interlaced_dct= s->interlaced_dct;
1797
    d->qscale= s->qscale;
1798

    
1799
    d->esc3_level_length= s->esc3_level_length;
1800
}
1801

    
1802
static inline void encode_mb_hq(MpegEncContext *s, MpegEncContext *backup, MpegEncContext *best, int type,
1803
                           PutBitContext pb[2], PutBitContext pb2[2], PutBitContext tex_pb[2],
1804
                           int *dmin, int *next_block, int motion_x, int motion_y)
1805
{
1806
    int score;
1807
    uint8_t *dest_backup[3];
1808

    
1809
    copy_context_before_encode(s, backup, type);
1810

    
1811
    s->block= s->blocks[*next_block];
1812
    s->pb= pb[*next_block];
1813
    if(s->data_partitioning){
1814
        s->pb2   = pb2   [*next_block];
1815
        s->tex_pb= tex_pb[*next_block];
1816
    }
1817

    
1818
    if(*next_block){
1819
        memcpy(dest_backup, s->dest, sizeof(s->dest));
1820
        s->dest[0] = s->rd_scratchpad;
1821
        s->dest[1] = s->rd_scratchpad + 16*s->linesize;
1822
        s->dest[2] = s->rd_scratchpad + 16*s->linesize + 8;
1823
        assert(s->linesize >= 32); //FIXME
1824
    }
1825

    
1826
    encode_mb(s, motion_x, motion_y);
1827

    
1828
    score= put_bits_count(&s->pb);
1829
    if(s->data_partitioning){
1830
        score+= put_bits_count(&s->pb2);
1831
        score+= put_bits_count(&s->tex_pb);
1832
    }
1833

    
1834
    if(s->avctx->mb_decision == FF_MB_DECISION_RD){
1835
        MPV_decode_mb(s, s->block);
1836

    
1837
        score *= s->lambda2;
1838
        score += sse_mb(s) << FF_LAMBDA_SHIFT;
1839
    }
1840

    
1841
    if(*next_block){
1842
        memcpy(s->dest, dest_backup, sizeof(s->dest));
1843
    }
1844

    
1845
    if(score<*dmin){
1846
        *dmin= score;
1847
        *next_block^=1;
1848

    
1849
        copy_context_after_encode(best, s, type);
1850
    }
1851
}
1852

    
1853
static int sse(MpegEncContext *s, uint8_t *src1, uint8_t *src2, int w, int h, int stride){
1854
    uint32_t *sq = ff_squareTbl + 256;
1855
    int acc=0;
1856
    int x,y;
1857

    
1858
    if(w==16 && h==16)
1859
        return s->dsp.sse[0](NULL, src1, src2, stride, 16);
1860
    else if(w==8 && h==8)
1861
        return s->dsp.sse[1](NULL, src1, src2, stride, 8);
1862

    
1863
    for(y=0; y<h; y++){
1864
        for(x=0; x<w; x++){
1865
            acc+= sq[src1[x + y*stride] - src2[x + y*stride]];
1866
        }
1867
    }
1868

    
1869
    assert(acc>=0);
1870

    
1871
    return acc;
1872
}
1873

    
1874
static int sse_mb(MpegEncContext *s){
1875
    int w= 16;
1876
    int h= 16;
1877

    
1878
    if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
1879
    if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
1880

    
1881
    if(w==16 && h==16)
1882
      if(s->avctx->mb_cmp == FF_CMP_NSSE){
1883
        return  s->dsp.nsse[0](s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, s->dest[0], s->linesize, 16)
1884
               +s->dsp.nsse[1](s, s->new_picture.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[1], s->uvlinesize, 8)
1885
               +s->dsp.nsse[1](s, s->new_picture.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[2], s->uvlinesize, 8);
1886
      }else{
1887
        return  s->dsp.sse[0](NULL, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, s->dest[0], s->linesize, 16)
1888
               +s->dsp.sse[1](NULL, s->new_picture.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[1], s->uvlinesize, 8)
1889
               +s->dsp.sse[1](NULL, s->new_picture.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[2], s->uvlinesize, 8);
1890
      }
1891
    else
1892
        return  sse(s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, s->dest[0], w, h, s->linesize)
1893
               +sse(s, s->new_picture.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[1], w>>1, h>>1, s->uvlinesize)
1894
               +sse(s, s->new_picture.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[2], w>>1, h>>1, s->uvlinesize);
1895
}
1896

    
1897
static int pre_estimate_motion_thread(AVCodecContext *c, void *arg){
1898
    MpegEncContext *s= *(void**)arg;
1899

    
1900

    
1901
    s->me.pre_pass=1;
1902
    s->me.dia_size= s->avctx->pre_dia_size;
1903
    s->first_slice_line=1;
1904
    for(s->mb_y= s->end_mb_y-1; s->mb_y >= s->start_mb_y; s->mb_y--) {
1905
        for(s->mb_x=s->mb_width-1; s->mb_x >=0 ;s->mb_x--) {
1906
            ff_pre_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
1907
        }
1908
        s->first_slice_line=0;
1909
    }
1910

    
1911
    s->me.pre_pass=0;
1912

    
1913
    return 0;
1914
}
1915

    
1916
static int estimate_motion_thread(AVCodecContext *c, void *arg){
1917
    MpegEncContext *s= *(void**)arg;
1918

    
1919
    ff_check_alignment();
1920

    
1921
    s->me.dia_size= s->avctx->dia_size;
1922
    s->first_slice_line=1;
1923
    for(s->mb_y= s->start_mb_y; s->mb_y < s->end_mb_y; s->mb_y++) {
1924
        s->mb_x=0; //for block init below
1925
        ff_init_block_index(s);
1926
        for(s->mb_x=0; s->mb_x < s->mb_width; s->mb_x++) {
1927
            s->block_index[0]+=2;
1928
            s->block_index[1]+=2;
1929
            s->block_index[2]+=2;
1930
            s->block_index[3]+=2;
1931

    
1932
            /* compute motion vector & mb_type and store in context */
1933
            if(s->pict_type==FF_B_TYPE)
1934
                ff_estimate_b_frame_motion(s, s->mb_x, s->mb_y);
1935
            else
1936
                ff_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
1937
        }
1938
        s->first_slice_line=0;
1939
    }
1940
    return 0;
1941
}
1942

    
1943
static int mb_var_thread(AVCodecContext *c, void *arg){
1944
    MpegEncContext *s= *(void**)arg;
1945
    int mb_x, mb_y;
1946

    
1947
    ff_check_alignment();
1948

    
1949
    for(mb_y=s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
1950
        for(mb_x=0; mb_x < s->mb_width; mb_x++) {
1951
            int xx = mb_x * 16;
1952
            int yy = mb_y * 16;
1953
            uint8_t *pix = s->new_picture.data[0] + (yy * s->linesize) + xx;
1954
            int varc;
1955
            int sum = s->dsp.pix_sum(pix, s->linesize);
1956

    
1957
            varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500 + 128)>>8;
1958

    
1959
            s->current_picture.mb_var [s->mb_stride * mb_y + mb_x] = varc;
1960
            s->current_picture.mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
1961
            s->me.mb_var_sum_temp    += varc;
1962
        }
1963
    }
1964
    return 0;
1965
}
1966

    
1967
static void write_slice_end(MpegEncContext *s){
1968
    if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4){
1969
        if(s->partitioned_frame){
1970
            ff_mpeg4_merge_partitions(s);
1971
        }
1972

    
1973
        ff_mpeg4_stuffing(&s->pb);
1974
    }else if(CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG){
1975
        ff_mjpeg_encode_stuffing(&s->pb);
1976
    }
1977

    
1978
    align_put_bits(&s->pb);
1979
    flush_put_bits(&s->pb);
1980

    
1981
    if((s->flags&CODEC_FLAG_PASS1) && !s->partitioned_frame)
1982
        s->misc_bits+= get_bits_diff(s);
1983
}
1984

    
1985
static int encode_thread(AVCodecContext *c, void *arg){
1986
    MpegEncContext *s= *(void**)arg;
1987
    int mb_x, mb_y, pdif = 0;
1988
    int chr_h= 16>>s->chroma_y_shift;
1989
    int i, j;
1990
    MpegEncContext best_s, backup_s;
1991
    uint8_t bit_buf[2][MAX_MB_BYTES];
1992
    uint8_t bit_buf2[2][MAX_MB_BYTES];
1993
    uint8_t bit_buf_tex[2][MAX_MB_BYTES];
1994
    PutBitContext pb[2], pb2[2], tex_pb[2];
1995
//printf("%d->%d\n", s->resync_mb_y, s->end_mb_y);
1996

    
1997
    ff_check_alignment();
1998

    
1999
    for(i=0; i<2; i++){
2000
        init_put_bits(&pb    [i], bit_buf    [i], MAX_MB_BYTES);
2001
        init_put_bits(&pb2   [i], bit_buf2   [i], MAX_MB_BYTES);
2002
        init_put_bits(&tex_pb[i], bit_buf_tex[i], MAX_MB_BYTES);
2003
    }
2004

    
2005
    s->last_bits= put_bits_count(&s->pb);
2006
    s->mv_bits=0;
2007
    s->misc_bits=0;
2008
    s->i_tex_bits=0;
2009
    s->p_tex_bits=0;
2010
    s->i_count=0;
2011
    s->f_count=0;
2012
    s->b_count=0;
2013
    s->skip_count=0;
2014

    
2015
    for(i=0; i<3; i++){
2016
        /* init last dc values */
2017
        /* note: quant matrix value (8) is implied here */
2018
        s->last_dc[i] = 128 << s->intra_dc_precision;
2019

    
2020
        s->current_picture.error[i] = 0;
2021
    }
2022
    s->mb_skip_run = 0;
2023
    memset(s->last_mv, 0, sizeof(s->last_mv));
2024

    
2025
    s->last_mv_dir = 0;
2026

    
2027
    switch(s->codec_id){
2028
    case CODEC_ID_H263:
2029
    case CODEC_ID_H263P:
2030
    case CODEC_ID_FLV1:
2031
        if (CONFIG_H263_ENCODER || CONFIG_H263P_ENCODER || CONFIG_FLV_ENCODER)
2032
            s->gob_index = ff_h263_get_gob_height(s);
2033
        break;
2034
    case CODEC_ID_MPEG4:
2035
        if(CONFIG_MPEG4_ENCODER && s->partitioned_frame)
2036
            ff_mpeg4_init_partitions(s);
2037
        break;
2038
    }
2039

    
2040
    s->resync_mb_x=0;
2041
    s->resync_mb_y=0;
2042
    s->first_slice_line = 1;
2043
    s->ptr_lastgob = s->pb.buf;
2044
    for(mb_y= s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
2045
//    printf("row %d at %X\n", s->mb_y, (int)s);
2046
        s->mb_x=0;
2047
        s->mb_y= mb_y;
2048

    
2049
        ff_set_qscale(s, s->qscale);
2050
        ff_init_block_index(s);
2051

    
2052
        for(mb_x=0; mb_x < s->mb_width; mb_x++) {
2053
            int xy= mb_y*s->mb_stride + mb_x; // removed const, H261 needs to adjust this
2054
            int mb_type= s->mb_type[xy];
2055
//            int d;
2056
            int dmin= INT_MAX;
2057
            int dir;
2058

    
2059
            if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < MAX_MB_BYTES){
2060
                av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
2061
                return -1;
2062
            }
2063
            if(s->data_partitioning){
2064
                if(   s->pb2   .buf_end - s->pb2   .buf - (put_bits_count(&s->    pb2)>>3) < MAX_MB_BYTES
2065
                   || s->tex_pb.buf_end - s->tex_pb.buf - (put_bits_count(&s->tex_pb )>>3) < MAX_MB_BYTES){
2066
                    av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
2067
                    return -1;
2068
                }
2069
            }
2070

    
2071
            s->mb_x = mb_x;
2072
            s->mb_y = mb_y;  // moved into loop, can get changed by H.261
2073
            ff_update_block_index(s);
2074

    
2075
            if(CONFIG_H261_ENCODER && s->codec_id == CODEC_ID_H261){
2076
                ff_h261_reorder_mb_index(s);
2077
                xy= s->mb_y*s->mb_stride + s->mb_x;
2078
                mb_type= s->mb_type[xy];
2079
            }
2080

    
2081
            /* write gob / video packet header  */
2082
            if(s->rtp_mode){
2083
                int current_packet_size, is_gob_start;
2084

    
2085
                current_packet_size= ((put_bits_count(&s->pb)+7)>>3) - (s->ptr_lastgob - s->pb.buf);
2086

    
2087
                is_gob_start= s->avctx->rtp_payload_size && current_packet_size >= s->avctx->rtp_payload_size && mb_y + mb_x>0;
2088

    
2089
                if(s->start_mb_y == mb_y && mb_y > 0 && mb_x==0) is_gob_start=1;
2090

    
2091
                switch(s->codec_id){
2092
                case CODEC_ID_H263:
2093
                case CODEC_ID_H263P:
2094
                    if(!s->h263_slice_structured)
2095
                        if(s->mb_x || s->mb_y%s->gob_index) is_gob_start=0;
2096
                    break;
2097
                case CODEC_ID_MPEG2VIDEO:
2098
                    if(s->mb_x==0 && s->mb_y!=0) is_gob_start=1;
2099
                case CODEC_ID_MPEG1VIDEO:
2100
                    if(s->mb_skip_run) is_gob_start=0;
2101
                    break;
2102
                }
2103

    
2104
                if(is_gob_start){
2105
                    if(s->start_mb_y != mb_y || mb_x!=0){
2106
                        write_slice_end(s);
2107

    
2108
                        if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4 && s->partitioned_frame){
2109
                            ff_mpeg4_init_partitions(s);
2110
                        }
2111
                    }
2112

    
2113
                    assert((put_bits_count(&s->pb)&7) == 0);
2114
                    current_packet_size= pbBufPtr(&s->pb) - s->ptr_lastgob;
2115

    
2116
                    if(s->avctx->error_rate && s->resync_mb_x + s->resync_mb_y > 0){
2117
                        int r= put_bits_count(&s->pb)/8 + s->picture_number + 16 + s->mb_x + s->mb_y;
2118
                        int d= 100 / s->avctx->error_rate;
2119
                        if(r % d == 0){
2120
                            current_packet_size=0;
2121
#ifndef ALT_BITSTREAM_WRITER
2122
                            s->pb.buf_ptr= s->ptr_lastgob;
2123
#endif
2124
                            assert(pbBufPtr(&s->pb) == s->ptr_lastgob);
2125
                        }
2126
                    }
2127

    
2128
                    if (s->avctx->rtp_callback){
2129
                        int number_mb = (mb_y - s->resync_mb_y)*s->mb_width + mb_x - s->resync_mb_x;
2130
                        s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, current_packet_size, number_mb);
2131
                    }
2132

    
2133
                    switch(s->codec_id){
2134
                    case CODEC_ID_MPEG4:
2135
                        if (CONFIG_MPEG4_ENCODER) {
2136
                            ff_mpeg4_encode_video_packet_header(s);
2137
                            ff_mpeg4_clean_buffers(s);
2138
                        }
2139
                    break;
2140
                    case CODEC_ID_MPEG1VIDEO:
2141
                    case CODEC_ID_MPEG2VIDEO:
2142
                        if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER) {
2143
                            ff_mpeg1_encode_slice_header(s);
2144
                            ff_mpeg1_clean_buffers(s);
2145
                        }
2146
                    break;
2147
                    case CODEC_ID_H263:
2148
                    case CODEC_ID_H263P:
2149
                        if (CONFIG_H263_ENCODER || CONFIG_H263P_ENCODER)
2150
                            h263_encode_gob_header(s, mb_y);
2151
                    break;
2152
                    }
2153

    
2154
                    if(s->flags&CODEC_FLAG_PASS1){
2155
                        int bits= put_bits_count(&s->pb);
2156
                        s->misc_bits+= bits - s->last_bits;
2157
                        s->last_bits= bits;
2158
                    }
2159

    
2160
                    s->ptr_lastgob += current_packet_size;
2161
                    s->first_slice_line=1;
2162
                    s->resync_mb_x=mb_x;
2163
                    s->resync_mb_y=mb_y;
2164
                }
2165
            }
2166

    
2167
            if(  (s->resync_mb_x   == s->mb_x)
2168
               && s->resync_mb_y+1 == s->mb_y){
2169
                s->first_slice_line=0;
2170
            }
2171

    
2172
            s->mb_skipped=0;
2173
            s->dquant=0; //only for QP_RD
2174

    
2175
            if(mb_type & (mb_type-1) || (s->flags & CODEC_FLAG_QP_RD)){ // more than 1 MB type possible or CODEC_FLAG_QP_RD
2176
                int next_block=0;
2177
                int pb_bits_count, pb2_bits_count, tex_pb_bits_count;
2178

    
2179
                copy_context_before_encode(&backup_s, s, -1);
2180
                backup_s.pb= s->pb;
2181
                best_s.data_partitioning= s->data_partitioning;
2182
                best_s.partitioned_frame= s->partitioned_frame;
2183
                if(s->data_partitioning){
2184
                    backup_s.pb2= s->pb2;
2185
                    backup_s.tex_pb= s->tex_pb;
2186
                }
2187

    
2188
                if(mb_type&CANDIDATE_MB_TYPE_INTER){
2189
                    s->mv_dir = MV_DIR_FORWARD;
2190
                    s->mv_type = MV_TYPE_16X16;
2191
                    s->mb_intra= 0;
2192
                    s->mv[0][0][0] = s->p_mv_table[xy][0];
2193
                    s->mv[0][0][1] = s->p_mv_table[xy][1];
2194
                    encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER, pb, pb2, tex_pb,
2195
                                 &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2196
                }
2197
                if(mb_type&CANDIDATE_MB_TYPE_INTER_I){
2198
                    s->mv_dir = MV_DIR_FORWARD;
2199
                    s->mv_type = MV_TYPE_FIELD;
2200
                    s->mb_intra= 0;
2201
                    for(i=0; i<2; i++){
2202
                        j= s->field_select[0][i] = s->p_field_select_table[i][xy];
2203
                        s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
2204
                        s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
2205
                    }
2206
                    encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER_I, pb, pb2, tex_pb,
2207
                                 &dmin, &next_block, 0, 0);
2208
                }
2209
                if(mb_type&CANDIDATE_MB_TYPE_SKIPPED){
2210
                    s->mv_dir = MV_DIR_FORWARD;
2211
                    s->mv_type = MV_TYPE_16X16;
2212
                    s->mb_intra= 0;
2213
                    s->mv[0][0][0] = 0;
2214
                    s->mv[0][0][1] = 0;
2215
                    encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_SKIPPED, pb, pb2, tex_pb,
2216
                                 &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2217
                }
2218
                if(mb_type&CANDIDATE_MB_TYPE_INTER4V){
2219
                    s->mv_dir = MV_DIR_FORWARD;
2220
                    s->mv_type = MV_TYPE_8X8;
2221
                    s->mb_intra= 0;
2222
                    for(i=0; i<4; i++){
2223
                        s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
2224
                        s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
2225
                    }
2226
                    encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER4V, pb, pb2, tex_pb,
2227
                                 &dmin, &next_block, 0, 0);
2228
                }
2229
                if(mb_type&CANDIDATE_MB_TYPE_FORWARD){
2230
                    s->mv_dir = MV_DIR_FORWARD;
2231
                    s->mv_type = MV_TYPE_16X16;
2232
                    s->mb_intra= 0;
2233
                    s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
2234
                    s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2235
                    encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD, pb, pb2, tex_pb,
2236
                                 &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2237
                }
2238
                if(mb_type&CANDIDATE_MB_TYPE_BACKWARD){
2239
                    s->mv_dir = MV_DIR_BACKWARD;
2240
                    s->mv_type = MV_TYPE_16X16;
2241
                    s->mb_intra= 0;
2242
                    s->mv[1][0][0] = s->b_back_mv_table[xy][0];
2243
                    s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2244
                    encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD, pb, pb2, tex_pb,
2245
                                 &dmin, &next_block, s->mv[1][0][0], s->mv[1][0][1]);
2246
                }
2247
                if(mb_type&CANDIDATE_MB_TYPE_BIDIR){
2248
                    s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2249
                    s->mv_type = MV_TYPE_16X16;
2250
                    s->mb_intra= 0;
2251
                    s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
2252
                    s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
2253
                    s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
2254
                    s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2255
                    encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR, pb, pb2, tex_pb,
2256
                                 &dmin, &next_block, 0, 0);
2257
                }
2258
                if(mb_type&CANDIDATE_MB_TYPE_FORWARD_I){
2259
                    s->mv_dir = MV_DIR_FORWARD;
2260
                    s->mv_type = MV_TYPE_FIELD;
2261
                    s->mb_intra= 0;
2262
                    for(i=0; i<2; i++){
2263
                        j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
2264
                        s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
2265
                        s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
2266
                    }
2267
                    encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD_I, pb, pb2, tex_pb,
2268
                                 &dmin, &next_block, 0, 0);
2269
                }
2270
                if(mb_type&CANDIDATE_MB_TYPE_BACKWARD_I){
2271
                    s->mv_dir = MV_DIR_BACKWARD;
2272
                    s->mv_type = MV_TYPE_FIELD;
2273
                    s->mb_intra= 0;
2274
                    for(i=0; i<2; i++){
2275
                        j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
2276
                        s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
2277
                        s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
2278
                    }
2279
                    encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD_I, pb, pb2, tex_pb,
2280
                                 &dmin, &next_block, 0, 0);
2281
                }
2282
                if(mb_type&CANDIDATE_MB_TYPE_BIDIR_I){
2283
                    s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2284
                    s->mv_type = MV_TYPE_FIELD;
2285
                    s->mb_intra= 0;
2286
                    for(dir=0; dir<2; dir++){
2287
                        for(i=0; i<2; i++){
2288
                            j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
2289
                            s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
2290
                            s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
2291
                        }
2292
                    }
2293
                    encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR_I, pb, pb2, tex_pb,
2294
                                 &dmin, &next_block, 0, 0);
2295
                }
2296
                if(mb_type&CANDIDATE_MB_TYPE_INTRA){
2297
                    s->mv_dir = 0;
2298
                    s->mv_type = MV_TYPE_16X16;
2299
                    s->mb_intra= 1;
2300
                    s->mv[0][0][0] = 0;
2301
                    s->mv[0][0][1] = 0;
2302
                    encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTRA, pb, pb2, tex_pb,
2303
                                 &dmin, &next_block, 0, 0);
2304
                    if(s->h263_pred || s->h263_aic){
2305
                        if(best_s.mb_intra)
2306
                            s->mbintra_table[mb_x + mb_y*s->mb_stride]=1;
2307
                        else
2308
                            ff_clean_intra_table_entries(s); //old mode?
2309
                    }
2310
                }
2311

    
2312
                if((s->flags & CODEC_FLAG_QP_RD) && dmin < INT_MAX){
2313
                    if(best_s.mv_type==MV_TYPE_16X16){ //FIXME move 4mv after QPRD
2314
                        const int last_qp= backup_s.qscale;
2315
                        int qpi, qp, dc[6];
2316
                        DCTELEM ac[6][16];
2317
                        const int mvdir= (best_s.mv_dir&MV_DIR_BACKWARD) ? 1 : 0;
2318
                        static const int dquant_tab[4]={-1,1,-2,2};
2319

    
2320
                        assert(backup_s.dquant == 0);
2321

    
2322
                        //FIXME intra
2323
                        s->mv_dir= best_s.mv_dir;
2324
                        s->mv_type = MV_TYPE_16X16;
2325
                        s->mb_intra= best_s.mb_intra;
2326
                        s->mv[0][0][0] = best_s.mv[0][0][0];
2327
                        s->mv[0][0][1] = best_s.mv[0][0][1];
2328
                        s->mv[1][0][0] = best_s.mv[1][0][0];
2329
                        s->mv[1][0][1] = best_s.mv[1][0][1];
2330

    
2331
                        qpi = s->pict_type == FF_B_TYPE ? 2 : 0;
2332
                        for(; qpi<4; qpi++){
2333
                            int dquant= dquant_tab[qpi];
2334
                            qp= last_qp + dquant;
2335
                            if(qp < s->avctx->qmin || qp > s->avctx->qmax)
2336
                                continue;
2337
                            backup_s.dquant= dquant;
2338
                            if(s->mb_intra && s->dc_val[0]){
2339
                                for(i=0; i<6; i++){
2340
                                    dc[i]= s->dc_val[0][ s->block_index[i] ];
2341
                                    memcpy(ac[i], s->ac_val[0][s->block_index[i]], sizeof(DCTELEM)*16);
2342
                                }
2343
                            }
2344

    
2345
                            encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
2346
                                         &dmin, &next_block, s->mv[mvdir][0][0], s->mv[mvdir][0][1]);
2347
                            if(best_s.qscale != qp){
2348
                                if(s->mb_intra && s->dc_val[0]){
2349
                                    for(i=0; i<6; i++){
2350
                                        s->dc_val[0][ s->block_index[i] ]= dc[i];
2351
                                        memcpy(s->ac_val[0][s->block_index[i]], ac[i], sizeof(DCTELEM)*16);
2352
                                    }
2353
                                }
2354
                            }
2355
                        }
2356
                    }
2357
                }
2358
                if(CONFIG_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT){
2359
                    int mx= s->b_direct_mv_table[xy][0];
2360
                    int my= s->b_direct_mv_table[xy][1];
2361

    
2362
                    backup_s.dquant = 0;
2363
                    s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2364
                    s->mb_intra= 0;
2365
                    ff_mpeg4_set_direct_mv(s, mx, my);
2366
                    encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
2367
                                 &dmin, &next_block, mx, my);
2368
                }
2369
                if(CONFIG_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT0){
2370
                    backup_s.dquant = 0;
2371
                    s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2372
                    s->mb_intra= 0;
2373
                    ff_mpeg4_set_direct_mv(s, 0, 0);
2374
                    encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
2375
                                 &dmin, &next_block, 0, 0);
2376
                }
2377
                if(!best_s.mb_intra && s->flags2&CODEC_FLAG2_SKIP_RD){
2378
                    int coded=0;
2379
                    for(i=0; i<6; i++)
2380
                        coded |= s->block_last_index[i];
2381
                    if(coded){
2382
                        int mx,my;
2383
                        memcpy(s->mv, best_s.mv, sizeof(s->mv));
2384
                        if(CONFIG_MPEG4_ENCODER && best_s.mv_dir & MV_DIRECT){
2385
                            mx=my=0; //FIXME find the one we actually used
2386
                            ff_mpeg4_set_direct_mv(s, mx, my);
2387
                        }else if(best_s.mv_dir&MV_DIR_BACKWARD){
2388
                            mx= s->mv[1][0][0];
2389
                            my= s->mv[1][0][1];
2390
                        }else{
2391
                            mx= s->mv[0][0][0];
2392
                            my= s->mv[0][0][1];
2393
                        }
2394

    
2395
                        s->mv_dir= best_s.mv_dir;
2396
                        s->mv_type = best_s.mv_type;
2397
                        s->mb_intra= 0;
2398
/*                        s->mv[0][0][0] = best_s.mv[0][0][0];
2399
                        s->mv[0][0][1] = best_s.mv[0][0][1];
2400
                        s->mv[1][0][0] = best_s.mv[1][0][0];
2401
                        s->mv[1][0][1] = best_s.mv[1][0][1];*/
2402
                        backup_s.dquant= 0;
2403
                        s->skipdct=1;
2404
                        encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
2405
                                        &dmin, &next_block, mx, my);
2406
                        s->skipdct=0;
2407
                    }
2408
                }
2409

    
2410
                s->current_picture.qscale_table[xy]= best_s.qscale;
2411

    
2412
                copy_context_after_encode(s, &best_s, -1);
2413

    
2414
                pb_bits_count= put_bits_count(&s->pb);
2415
                flush_put_bits(&s->pb);
2416
                ff_copy_bits(&backup_s.pb, bit_buf[next_block^1], pb_bits_count);
2417
                s->pb= backup_s.pb;
2418

    
2419
                if(s->data_partitioning){
2420
                    pb2_bits_count= put_bits_count(&s->pb2);
2421
                    flush_put_bits(&s->pb2);
2422
                    ff_copy_bits(&backup_s.pb2, bit_buf2[next_block^1], pb2_bits_count);
2423
                    s->pb2= backup_s.pb2;
2424

    
2425
                    tex_pb_bits_count= put_bits_count(&s->tex_pb);
2426
                    flush_put_bits(&s->tex_pb);
2427
                    ff_copy_bits(&backup_s.tex_pb, bit_buf_tex[next_block^1], tex_pb_bits_count);
2428
                    s->tex_pb= backup_s.tex_pb;
2429
                }
2430
                s->last_bits= put_bits_count(&s->pb);
2431

    
2432
                if (CONFIG_ANY_H263_ENCODER &&
2433
                    s->out_format == FMT_H263 && s->pict_type!=FF_B_TYPE)
2434
                    ff_h263_update_motion_val(s);
2435

    
2436
                if(next_block==0){ //FIXME 16 vs linesize16
2437
                    s->dsp.put_pixels_tab[0][0](s->dest[0], s->rd_scratchpad                     , s->linesize  ,16);
2438
                    s->dsp.put_pixels_tab[1][0](s->dest[1], s->rd_scratchpad + 16*s->linesize    , s->uvlinesize, 8);
2439
                    s->dsp.put_pixels_tab[1][0](s->dest[2], s->rd_scratchpad + 16*s->linesize + 8, s->uvlinesize, 8);
2440
                }
2441

    
2442
                if(s->avctx->mb_decision == FF_MB_DECISION_BITS)
2443
                    MPV_decode_mb(s, s->block);
2444
            } else {
2445
                int motion_x = 0, motion_y = 0;
2446
                s->mv_type=MV_TYPE_16X16;
2447
                // only one MB-Type possible
2448

    
2449
                switch(mb_type){
2450
                case CANDIDATE_MB_TYPE_INTRA:
2451
                    s->mv_dir = 0;
2452
                    s->mb_intra= 1;
2453
                    motion_x= s->mv[0][0][0] = 0;
2454
                    motion_y= s->mv[0][0][1] = 0;
2455
                    break;
2456
                case CANDIDATE_MB_TYPE_INTER:
2457
                    s->mv_dir = MV_DIR_FORWARD;
2458
                    s->mb_intra= 0;
2459
                    motion_x= s->mv[0][0][0] = s->p_mv_table[xy][0];
2460
                    motion_y= s->mv[0][0][1] = s->p_mv_table[xy][1];
2461
                    break;
2462
                case CANDIDATE_MB_TYPE_INTER_I:
2463
                    s->mv_dir = MV_DIR_FORWARD;
2464
                    s->mv_type = MV_TYPE_FIELD;
2465
                    s->mb_intra= 0;
2466
                    for(i=0; i<2; i++){
2467
                        j= s->field_select[0][i] = s->p_field_select_table[i][xy];
2468
                        s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
2469
                        s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
2470
                    }
2471
                    break;
2472
                case CANDIDATE_MB_TYPE_INTER4V:
2473
                    s->mv_dir = MV_DIR_FORWARD;
2474
                    s->mv_type = MV_TYPE_8X8;
2475
                    s->mb_intra= 0;
2476
                    for(i=0; i<4; i++){
2477
                        s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
2478
                        s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
2479
                    }
2480
                    break;
2481
                case CANDIDATE_MB_TYPE_DIRECT:
2482
                    if (CONFIG_MPEG4_ENCODER) {
2483
                        s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
2484
                        s->mb_intra= 0;
2485
                        motion_x=s->b_direct_mv_table[xy][0];
2486
                        motion_y=s->b_direct_mv_table[xy][1];
2487
                        ff_mpeg4_set_direct_mv(s, motion_x, motion_y);
2488
                    }
2489
                    break;
2490
                case CANDIDATE_MB_TYPE_DIRECT0:
2491
                    if (CONFIG_MPEG4_ENCODER) {
2492
                        s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
2493
                        s->mb_intra= 0;
2494
                        ff_mpeg4_set_direct_mv(s, 0, 0);
2495
                    }
2496
                    break;
2497
                case CANDIDATE_MB_TYPE_BIDIR:
2498
                    s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2499
                    s->mb_intra= 0;
2500
                    s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
2501
                    s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
2502
                    s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
2503
                    s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2504
                    break;
2505
                case CANDIDATE_MB_TYPE_BACKWARD:
2506
                    s->mv_dir = MV_DIR_BACKWARD;
2507
                    s->mb_intra= 0;
2508
                    motion_x= s->mv[1][0][0] = s->b_back_mv_table[xy][0];
2509
                    motion_y= s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2510
                    break;
2511
                case CANDIDATE_MB_TYPE_FORWARD:
2512
                    s->mv_dir = MV_DIR_FORWARD;
2513
                    s->mb_intra= 0;
2514
                    motion_x= s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
2515
                    motion_y= s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2516
//                    printf(" %d %d ", motion_x, motion_y);
2517
                    break;
2518
                case CANDIDATE_MB_TYPE_FORWARD_I:
2519
                    s->mv_dir = MV_DIR_FORWARD;
2520
                    s->mv_type = MV_TYPE_FIELD;
2521
                    s->mb_intra= 0;
2522
                    for(i=0; i<2; i++){
2523
                        j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
2524
                        s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
2525
                        s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
2526
                    }
2527
                    break;
2528
                case CANDIDATE_MB_TYPE_BACKWARD_I:
2529
                    s->mv_dir = MV_DIR_BACKWARD;
2530
                    s->mv_type = MV_TYPE_FIELD;
2531
                    s->mb_intra= 0;
2532
                    for(i=0; i<2; i++){
2533
                        j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
2534
                        s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
2535
                        s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
2536
                    }
2537
                    break;
2538
                case CANDIDATE_MB_TYPE_BIDIR_I:
2539
                    s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2540
                    s->mv_type = MV_TYPE_FIELD;
2541
                    s->mb_intra= 0;
2542
                    for(dir=0; dir<2; dir++){
2543
                        for(i=0; i<2; i++){
2544
                            j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
2545
                            s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
2546
                            s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
2547
                        }
2548
                    }
2549
                    break;
2550
                default:
2551
                    av_log(s->avctx, AV_LOG_ERROR, "illegal MB type\n");
2552
                }
2553

    
2554
                encode_mb(s, motion_x, motion_y);
2555

    
2556
                // RAL: Update last macroblock type
2557
                s->last_mv_dir = s->mv_dir;
2558

    
2559
                if (CONFIG_ANY_H263_ENCODER &&
2560
                    s->out_format == FMT_H263 && s->pict_type!=FF_B_TYPE)
2561
                    ff_h263_update_motion_val(s);
2562

    
2563
                MPV_decode_mb(s, s->block);
2564
            }
2565

    
2566
            /* clean the MV table in IPS frames for direct mode in B frames */
2567
            if(s->mb_intra /* && I,P,S_TYPE */){
2568
                s->p_mv_table[xy][0]=0;
2569
                s->p_mv_table[xy][1]=0;
2570
            }
2571

    
2572
            if(s->flags&CODEC_FLAG_PSNR){
2573
                int w= 16;
2574
                int h= 16;
2575

    
2576
                if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
2577
                if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
2578

    
2579
                s->current_picture.error[0] += sse(
2580
                    s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16,
2581
                    s->dest[0], w, h, s->linesize);
2582
                s->current_picture.error[1] += sse(
2583
                    s, s->new_picture.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*chr_h,
2584
                    s->dest[1], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
2585
                s->current_picture.error[2] += sse(
2586
                    s, s->new_picture.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*chr_h,
2587
                    s->dest[2], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
2588
            }
2589
            if(s->loop_filter){
2590
                if(CONFIG_ANY_H263_ENCODER && s->out_format == FMT_H263)
2591
                    ff_h263_loop_filter(s);
2592
            }
2593
//printf("MB %d %d bits\n", s->mb_x+s->mb_y*s->mb_stride, put_bits_count(&s->pb));
2594
        }
2595
    }
2596

    
2597
    //not beautiful here but we must write it before flushing so it has to be here
2598
    if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version && s->msmpeg4_version<4 && s->pict_type == FF_I_TYPE)
2599
        msmpeg4_encode_ext_header(s);
2600

    
2601
    write_slice_end(s);
2602

    
2603
    /* Send the last GOB if RTP */
2604
    if (s->avctx->rtp_callback) {
2605
        int number_mb = (mb_y - s->resync_mb_y)*s->mb_width - s->resync_mb_x;
2606
        pdif = pbBufPtr(&s->pb) - s->ptr_lastgob;
2607
        /* Call the RTP callback to send the last GOB */
2608
        emms_c();
2609
        s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, pdif, number_mb);
2610
    }
2611

    
2612
    return 0;
2613
}
2614

    
2615
#define MERGE(field) dst->field += src->field; src->field=0
2616
static void merge_context_after_me(MpegEncContext *dst, MpegEncContext *src){
2617
    MERGE(me.scene_change_score);
2618
    MERGE(me.mc_mb_var_sum_temp);
2619
    MERGE(me.mb_var_sum_temp);
2620
}
2621

    
2622
static void merge_context_after_encode(MpegEncContext *dst, MpegEncContext *src){
2623
    int i;
2624

    
2625
    MERGE(dct_count[0]); //note, the other dct vars are not part of the context
2626
    MERGE(dct_count[1]);
2627
    MERGE(mv_bits);
2628
    MERGE(i_tex_bits);
2629
    MERGE(p_tex_bits);
2630
    MERGE(i_count);
2631
    MERGE(f_count);
2632
    MERGE(b_count);
2633
    MERGE(skip_count);
2634
    MERGE(misc_bits);
2635
    MERGE(error_count);
2636
    MERGE(padding_bug_score);
2637
    MERGE(current_picture.error[0]);
2638
    MERGE(current_picture.error[1]);
2639
    MERGE(current_picture.error[2]);
2640

    
2641
    if(dst->avctx->noise_reduction){
2642
        for(i=0; i<64; i++){
2643
            MERGE(dct_error_sum[0][i]);
2644
            MERGE(dct_error_sum[1][i]);
2645
        }
2646
    }
2647

    
2648
    assert(put_bits_count(&src->pb) % 8 ==0);
2649
    assert(put_bits_count(&dst->pb) % 8 ==0);
2650
    ff_copy_bits(&dst->pb, src->pb.buf, put_bits_count(&src->pb));
2651
    flush_put_bits(&dst->pb);
2652
}
2653

    
2654
static int estimate_qp(MpegEncContext *s, int dry_run){
2655
    if (s->next_lambda){
2656
        s->current_picture_ptr->quality=
2657
        s->current_picture.quality = s->next_lambda;
2658
        if(!dry_run) s->next_lambda= 0;
2659
    } else if (!s->fixed_qscale) {
2660
        s->current_picture_ptr->quality=
2661
        s->current_picture.quality = ff_rate_estimate_qscale(s, dry_run);
2662
        if (s->current_picture.quality < 0)
2663
            return -1;
2664
    }
2665

    
2666
    if(s->adaptive_quant){
2667
        switch(s->codec_id){
2668
        case CODEC_ID_MPEG4:
2669
            if (CONFIG_MPEG4_ENCODER)
2670
                ff_clean_mpeg4_qscales(s);
2671
            break;
2672
        case CODEC_ID_H263:
2673
        case CODEC_ID_H263P:
2674
        case CODEC_ID_FLV1:
2675
            if (CONFIG_H263_ENCODER||CONFIG_H263P_ENCODER||CONFIG_FLV_ENCODER)
2676
                ff_clean_h263_qscales(s);
2677
            break;
2678
        }
2679

    
2680
        s->lambda= s->lambda_table[0];
2681
        //FIXME broken
2682
    }else
2683
        s->lambda= s->current_picture.quality;
2684
//printf("%d %d\n", s->avctx->global_quality, s->current_picture.quality);
2685
    update_qscale(s);
2686
    return 0;
2687
}
2688

    
2689
/* must be called before writing the header */
2690
static void set_frame_distances(MpegEncContext * s){
2691
    assert(s->current_picture_ptr->pts != AV_NOPTS_VALUE);
2692
    s->time= s->current_picture_ptr->pts*s->avctx->time_base.num;
2693

    
2694
    if(s->pict_type==FF_B_TYPE){
2695
        s->pb_time= s->pp_time - (s->last_non_b_time - s->time);
2696
        assert(s->pb_time > 0 && s->pb_time < s->pp_time);
2697
    }else{
2698
        s->pp_time= s->time - s->last_non_b_time;
2699
        s->last_non_b_time= s->time;
2700
        assert(s->picture_number==0 || s->pp_time > 0);
2701
    }
2702
}
2703

    
2704
static int encode_picture(MpegEncContext *s, int picture_number)
2705
{
2706
    int i;
2707
    int bits;
2708

    
2709
    s->picture_number = picture_number;
2710

    
2711
    /* Reset the average MB variance */
2712
    s->me.mb_var_sum_temp    =
2713
    s->me.mc_mb_var_sum_temp = 0;
2714

    
2715
    /* we need to initialize some time vars before we can encode b-frames */
2716
    // RAL: Condition added for MPEG1VIDEO
2717
    if (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO || (s->h263_pred && !s->h263_msmpeg4))
2718
        set_frame_distances(s);
2719
    if(CONFIG_MPEG4_ENCODER && s->codec_id == CODEC_ID_MPEG4)
2720
        ff_set_mpeg4_time(s);
2721

    
2722
    s->me.scene_change_score=0;
2723

    
2724
//    s->lambda= s->current_picture_ptr->quality; //FIXME qscale / ... stuff for ME rate distortion
2725

    
2726
    if(s->pict_type==FF_I_TYPE){
2727
        if(s->msmpeg4_version >= 3) s->no_rounding=1;
2728
        else                        s->no_rounding=0;
2729
    }else if(s->pict_type!=FF_B_TYPE){
2730
        if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
2731
            s->no_rounding ^= 1;
2732
    }
2733

    
2734
    if(s->flags & CODEC_FLAG_PASS2){
2735
        if (estimate_qp(s,1) < 0)
2736
            return -1;
2737
        ff_get_2pass_fcode(s);
2738
    }else if(!(s->flags & CODEC_FLAG_QSCALE)){
2739
        if(s->pict_type==FF_B_TYPE)
2740
            s->lambda= s->last_lambda_for[s->pict_type];
2741
        else
2742
            s->lambda= s->last_lambda_for[s->last_non_b_pict_type];
2743
        update_qscale(s);
2744
    }
2745

    
2746
    s->mb_intra=0; //for the rate distortion & bit compare functions
2747
    for(i=1; i<s->avctx->thread_count; i++){
2748
        ff_update_duplicate_context(s->thread_context[i], s);
2749
    }
2750

    
2751
    if(ff_init_me(s)<0)
2752
        return -1;
2753

    
2754
    /* Estimate motion for every MB */
2755
    if(s->pict_type != FF_I_TYPE){
2756
        s->lambda = (s->lambda * s->avctx->me_penalty_compensation + 128)>>8;
2757
        s->lambda2= (s->lambda2* (int64_t)s->avctx->me_penalty_compensation + 128)>>8;
2758
        if(s->pict_type != FF_B_TYPE && s->avctx->me_threshold==0){
2759
            if((s->avctx->pre_me && s->last_non_b_pict_type==FF_I_TYPE) || s->avctx->pre_me==2){
2760
                s->avctx->execute(s->avctx, pre_estimate_motion_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count, sizeof(void*));
2761
            }
2762
        }
2763

    
2764
        s->avctx->execute(s->avctx, estimate_motion_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count, sizeof(void*));
2765
    }else /* if(s->pict_type == FF_I_TYPE) */{
2766
        /* I-Frame */
2767
        for(i=0; i<s->mb_stride*s->mb_height; i++)
2768
            s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
2769

    
2770
        if(!s->fixed_qscale){
2771
            /* finding spatial complexity for I-frame rate control */
2772
            s->avctx->execute(s->avctx, mb_var_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count, sizeof(void*));
2773
        }
2774
    }
2775
    for(i=1; i<s->avctx->thread_count; i++){
2776
        merge_context_after_me(s, s->thread_context[i]);
2777
    }
2778
    s->current_picture.mc_mb_var_sum= s->current_picture_ptr->mc_mb_var_sum= s->me.mc_mb_var_sum_temp;
2779
    s->current_picture.   mb_var_sum= s->current_picture_ptr->   mb_var_sum= s->me.   mb_var_sum_temp;
2780
    emms_c();
2781

    
2782
    if(s->me.scene_change_score > s->avctx->scenechange_threshold && s->pict_type == FF_P_TYPE){
2783
        s->pict_type= FF_I_TYPE;
2784
        for(i=0; i<s->mb_stride*s->mb_height; i++)
2785
            s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
2786
//printf("Scene change detected, encoding as I Frame %d %d\n", s->current_picture.mb_var_sum, s->current_picture.mc_mb_var_sum);
2787
    }
2788

    
2789
    if(!s->umvplus){
2790
        if(s->pict_type==FF_P_TYPE || s->pict_type==FF_S_TYPE) {
2791
            s->f_code= ff_get_best_fcode(s, s->p_mv_table, CANDIDATE_MB_TYPE_INTER);
2792

    
2793
            if(s->flags & CODEC_FLAG_INTERLACED_ME){
2794
                int a,b;
2795
                a= ff_get_best_fcode(s, s->p_field_mv_table[0][0], CANDIDATE_MB_TYPE_INTER_I); //FIXME field_select
2796
                b= ff_get_best_fcode(s, s->p_field_mv_table[1][1], CANDIDATE_MB_TYPE_INTER_I);
2797
                s->f_code= FFMAX3(s->f_code, a, b);
2798
            }
2799

    
2800
            ff_fix_long_p_mvs(s);
2801
            ff_fix_long_mvs(s, NULL, 0, s->p_mv_table, s->f_code, CANDIDATE_MB_TYPE_INTER, 0);
2802
            if(s->flags & CODEC_FLAG_INTERLACED_ME){
2803
                int j;
2804
                for(i=0; i<2; i++){
2805
                    for(j=0; j<2; j++)
2806
                        ff_fix_long_mvs(s, s->p_field_select_table[i], j,
2807
                                        s->p_field_mv_table[i][j], s->f_code, CANDIDATE_MB_TYPE_INTER_I, 0);
2808
                }
2809
            }
2810
        }
2811

    
2812
        if(s->pict_type==FF_B_TYPE){
2813
            int a, b;
2814

    
2815
            a = ff_get_best_fcode(s, s->b_forw_mv_table, CANDIDATE_MB_TYPE_FORWARD);
2816
            b = ff_get_best_fcode(s, s->b_bidir_forw_mv_table, CANDIDATE_MB_TYPE_BIDIR);
2817
            s->f_code = FFMAX(a, b);
2818

    
2819
            a = ff_get_best_fcode(s, s->b_back_mv_table, CANDIDATE_MB_TYPE_BACKWARD);
2820
            b = ff_get_best_fcode(s, s->b_bidir_back_mv_table, CANDIDATE_MB_TYPE_BIDIR);
2821
            s->b_code = FFMAX(a, b);
2822

    
2823
            ff_fix_long_mvs(s, NULL, 0, s->b_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_FORWARD, 1);
2824
            ff_fix_long_mvs(s, NULL, 0, s->b_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BACKWARD, 1);
2825
            ff_fix_long_mvs(s, NULL, 0, s->b_bidir_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_BIDIR, 1);
2826
            ff_fix_long_mvs(s, NULL, 0, s->b_bidir_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BIDIR, 1);
2827
            if(s->flags & CODEC_FLAG_INTERLACED_ME){
2828
                int dir, j;
2829
                for(dir=0; dir<2; dir++){
2830
                    for(i=0; i<2; i++){
2831
                        for(j=0; j<2; j++){
2832
                            int type= dir ? (CANDIDATE_MB_TYPE_BACKWARD_I|CANDIDATE_MB_TYPE_BIDIR_I)
2833
                                          : (CANDIDATE_MB_TYPE_FORWARD_I |CANDIDATE_MB_TYPE_BIDIR_I);
2834
                            ff_fix_long_mvs(s, s->b_field_select_table[dir][i], j,
2835
                                            s->b_field_mv_table[dir][i][j], dir ? s->b_code : s->f_code, type, 1);
2836
                        }
2837
                    }
2838
                }
2839
            }
2840
        }
2841
    }
2842

    
2843
    if (estimate_qp(s, 0) < 0)
2844
        return -1;
2845

    
2846
    if(s->qscale < 3 && s->max_qcoeff<=128 && s->pict_type==FF_I_TYPE && !(s->flags & CODEC_FLAG_QSCALE))
2847
        s->qscale= 3; //reduce clipping problems
2848

    
2849
    if (s->out_format == FMT_MJPEG) {
2850
        /* for mjpeg, we do include qscale in the matrix */
2851
        s->intra_matrix[0] = ff_mpeg1_default_intra_matrix[0];
2852
        for(i=1;i<64;i++){
2853
            int j= s->dsp.idct_permutation[i];
2854

    
2855
            s->intra_matrix[j] = av_clip_uint8((ff_mpeg1_default_intra_matrix[i] * s->qscale) >> 3);
2856
        }
2857
        ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
2858
                       s->intra_matrix, s->intra_quant_bias, 8, 8, 1);
2859
        s->qscale= 8;
2860
    }
2861

    
2862
    //FIXME var duplication
2863
    s->current_picture_ptr->key_frame=
2864
    s->current_picture.key_frame= s->pict_type == FF_I_TYPE; //FIXME pic_ptr
2865
    s->current_picture_ptr->pict_type=
2866
    s->current_picture.pict_type= s->pict_type;
2867

    
2868
    if(s->current_picture.key_frame)
2869
        s->picture_in_gop_number=0;
2870

    
2871
    s->last_bits= put_bits_count(&s->pb);
2872
    switch(s->out_format) {
2873
    case FMT_MJPEG:
2874
        if (CONFIG_MJPEG_ENCODER)
2875
            ff_mjpeg_encode_picture_header(s);
2876
        break;
2877
    case FMT_H261:
2878
        if (CONFIG_H261_ENCODER)
2879
            ff_h261_encode_picture_header(s, picture_number);
2880
        break;
2881
    case FMT_H263:
2882
        if (CONFIG_WMV2_ENCODER && s->codec_id == CODEC_ID_WMV2)
2883
            ff_wmv2_encode_picture_header(s, picture_number);
2884
        else if (CONFIG_MSMPEG4_ENCODER && s->h263_msmpeg4)
2885
            msmpeg4_encode_picture_header(s, picture_number);
2886
        else if (CONFIG_MPEG4_ENCODER && s->h263_pred)
2887
            mpeg4_encode_picture_header(s, picture_number);
2888
        else if (CONFIG_RV10_ENCODER && s->codec_id == CODEC_ID_RV10)
2889
            rv10_encode_picture_header(s, picture_number);
2890
        else if (CONFIG_RV20_ENCODER && s->codec_id == CODEC_ID_RV20)
2891
            rv20_encode_picture_header(s, picture_number);
2892
        else if (CONFIG_FLV_ENCODER && s->codec_id == CODEC_ID_FLV1)
2893
            ff_flv_encode_picture_header(s, picture_number);
2894
        else if (CONFIG_ANY_H263_ENCODER)
2895
            h263_encode_picture_header(s, picture_number);
2896
        break;
2897
    case FMT_MPEG1:
2898
        if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
2899
            mpeg1_encode_picture_header(s, picture_number);
2900
        break;
2901
    case FMT_H264:
2902
        break;
2903
    default:
2904
        assert(0);
2905
    }
2906
    bits= put_bits_count(&s->pb);
2907
    s->header_bits= bits - s->last_bits;
2908

    
2909
    for(i=1; i<s->avctx->thread_count; i++){
2910
        update_duplicate_context_after_me(s->thread_context[i], s);
2911
    }
2912
    s->avctx->execute(s->avctx, encode_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count, sizeof(void*));
2913
    for(i=1; i<s->avctx->thread_count; i++){
2914
        merge_context_after_encode(s, s->thread_context[i]);
2915
    }
2916
    emms_c();
2917
    return 0;
2918
}
2919

    
2920
void  denoise_dct_c(MpegEncContext *s, DCTELEM *block){
2921
    const int intra= s->mb_intra;
2922
    int i;
2923

    
2924
    s->dct_count[intra]++;
2925

    
2926
    for(i=0; i<64; i++){
2927
        int level= block[i];
2928

    
2929
        if(level){
2930
            if(level>0){
2931
                s->dct_error_sum[intra][i] += level;
2932
                level -= s->dct_offset[intra][i];
2933
                if(level<0) level=0;
2934
            }else{
2935
                s->dct_error_sum[intra][i] -= level;
2936
                level += s->dct_offset[intra][i];
2937
                if(level>0) level=0;
2938
            }
2939
            block[i]= level;
2940
        }
2941
    }
2942
}
2943

    
2944
int dct_quantize_trellis_c(MpegEncContext *s,
2945
                        DCTELEM *block, int n,
2946
                        int qscale, int *overflow){
2947
    const int *qmat;
2948
    const uint8_t *scantable= s->intra_scantable.scantable;
2949
    const uint8_t *perm_scantable= s->intra_scantable.permutated;
2950
    int max=0;
2951
    unsigned int threshold1, threshold2;
2952
    int bias=0;
2953
    int run_tab[65];
2954
    int level_tab[65];
2955
    int score_tab[65];
2956
    int survivor[65];
2957
    int survivor_count;
2958
    int last_run=0;
2959
    int last_level=0;
2960
    int last_score= 0;
2961
    int last_i;
2962
    int coeff[2][64];
2963
    int coeff_count[64];
2964
    int qmul, qadd, start_i, last_non_zero, i, dc;
2965
    const int esc_length= s->ac_esc_length;
2966
    uint8_t * length;
2967
    uint8_t * last_length;
2968
    const int lambda= s->lambda2 >> (FF_LAMBDA_SHIFT - 6);
2969

    
2970
    s->dsp.fdct (block);
2971

    
2972
    if(s->dct_error_sum)
2973
        s->denoise_dct(s, block);
2974
    qmul= qscale*16;
2975
    qadd= ((qscale-1)|1)*8;
2976

    
2977
    if (s->mb_intra) {
2978
        int q;
2979
        if (!s->h263_aic) {
2980
            if (n < 4)
2981
                q = s->y_dc_scale;
2982
            else
2983
                q = s->c_dc_scale;
2984
            q = q << 3;
2985
        } else{
2986
            /* For AIC we skip quant/dequant of INTRADC */
2987
            q = 1 << 3;
2988
            qadd=0;
2989
        }
2990

    
2991
        /* note: block[0] is assumed to be positive */
2992
        block[0] = (block[0] + (q >> 1)) / q;
2993
        start_i = 1;
2994
        last_non_zero = 0;
2995
        qmat = s->q_intra_matrix[qscale];
2996
        if(s->mpeg_quant || s->out_format == FMT_MPEG1)
2997
            bias= 1<<(QMAT_SHIFT-1);
2998
        length     = s->intra_ac_vlc_length;
2999
        last_length= s->intra_ac_vlc_last_length;
3000
    } else {
3001
        start_i = 0;
3002
        last_non_zero = -1;
3003
        qmat = s->q_inter_matrix[qscale];
3004
        length     = s->inter_ac_vlc_length;
3005
        last_length= s->inter_ac_vlc_last_length;
3006
    }
3007
    last_i= start_i;
3008

    
3009
    threshold1= (1<<QMAT_SHIFT) - bias - 1;
3010
    threshold2= (threshold1<<1);
3011

    
3012
    for(i=63; i>=start_i; i--) {
3013
        const int j = scantable[i];
3014
        int level = block[j] * qmat[j];
3015

    
3016
        if(((unsigned)(level+threshold1))>threshold2){
3017
            last_non_zero = i;
3018
            break;
3019
        }
3020
    }
3021

    
3022
    for(i=start_i; i<=last_non_zero; i++) {
3023
        const int j = scantable[i];
3024
        int level = block[j] * qmat[j];
3025

    
3026
//        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
3027
//           || bias-level >= (1<<(QMAT_SHIFT - 3))){
3028
        if(((unsigned)(level+threshold1))>threshold2){
3029
            if(level>0){
3030
                level= (bias + level)>>QMAT_SHIFT;
3031
                coeff[0][i]= level;
3032
                coeff[1][i]= level-1;
3033
//                coeff[2][k]= level-2;
3034
            }else{
3035
                level= (bias - level)>>QMAT_SHIFT;
3036
                coeff[0][i]= -level;
3037
                coeff[1][i]= -level+1;
3038
//                coeff[2][k]= -level+2;
3039
            }
3040
            coeff_count[i]= FFMIN(level, 2);
3041
            assert(coeff_count[i]);
3042
            max |=level;
3043
        }else{
3044
            coeff[0][i]= (level>>31)|1;
3045
            coeff_count[i]= 1;
3046
        }
3047
    }
3048

    
3049
    *overflow= s->max_qcoeff < max; //overflow might have happened
3050

    
3051
    if(last_non_zero < start_i){
3052
        memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
3053
        return last_non_zero;
3054
    }
3055

    
3056
    score_tab[start_i]= 0;
3057
    survivor[0]= start_i;
3058
    survivor_count= 1;
3059

    
3060
    for(i=start_i; i<=last_non_zero; i++){
3061
        int level_index, j, zero_distortion;
3062
        int dct_coeff= FFABS(block[ scantable[i] ]);
3063
        int best_score=256*256*256*120;
3064

    
3065
        if (   s->dsp.fdct == fdct_ifast
3066
#ifndef FAAN_POSTSCALE
3067
            || s->dsp.fdct == ff_faandct
3068
#endif
3069
           )
3070
            dct_coeff= (dct_coeff*ff_inv_aanscales[ scantable[i] ]) >> 12;
3071
        zero_distortion= dct_coeff*dct_coeff;
3072

    
3073
        for(level_index=0; level_index < coeff_count[i]; level_index++){
3074
            int distortion;
3075
            int level= coeff[level_index][i];
3076
            const int alevel= FFABS(level);
3077
            int unquant_coeff;
3078

    
3079
            assert(level);
3080

    
3081
            if(s->out_format == FMT_H263){
3082
                unquant_coeff= alevel*qmul + qadd;
3083
            }else{ //MPEG1
3084
                j= s->dsp.idct_permutation[ scantable[i] ]; //FIXME optimize
3085
                if(s->mb_intra){
3086
                        unquant_coeff = (int)(  alevel  * qscale * s->intra_matrix[j]) >> 3;
3087
                        unquant_coeff =   (unquant_coeff - 1) | 1;
3088
                }else{
3089
                        unquant_coeff = (((  alevel  << 1) + 1) * qscale * ((int) s->inter_matrix[j])) >> 4;
3090
                        unquant_coeff =   (unquant_coeff - 1) | 1;
3091
                }
3092
                unquant_coeff<<= 3;
3093
            }
3094

    
3095
            distortion= (unquant_coeff - dct_coeff) * (unquant_coeff - dct_coeff) - zero_distortion;
3096
            level+=64;
3097
            if((level&(~127)) == 0){
3098
                for(j=survivor_count-1; j>=0; j--){
3099
                    int run= i - survivor[j];
3100
                    int score= distortion + length[UNI_AC_ENC_INDEX(run, level)]*lambda;
3101
                    score += score_tab[i-run];
3102

    
3103
                    if(score < best_score){
3104
                        best_score= score;
3105
                        run_tab[i+1]= run;
3106
                        level_tab[i+1]= level-64;
3107
                    }
3108
                }
3109

    
3110
                if(s->out_format == FMT_H263){
3111
                    for(j=survivor_count-1; j>=0; j--){
3112
                        int run= i - survivor[j];
3113
                        int score= distortion + last_length[UNI_AC_ENC_INDEX(run, level)]*lambda;
3114
                        score += score_tab[i-run];
3115
                        if(score < last_score){
3116
                            last_score= score;
3117
                            last_run= run;
3118
                            last_level= level-64;
3119
                            last_i= i+1;
3120
                        }
3121
                    }
3122
                }
3123
            }else{
3124
                distortion += esc_length*lambda;
3125
                for(j=survivor_count-1; j>=0; j--){
3126
                    int run= i - survivor[j];
3127
                    int score= distortion + score_tab[i-run];
3128

    
3129
                    if(score < best_score){
3130
                        best_score= score;
3131
                        run_tab[i+1]= run;
3132
                        level_tab[i+1]= level-64;
3133
                    }
3134
                }
3135

    
3136
                if(s->out_format == FMT_H263){
3137
                  for(j=survivor_count-1; j>=0; j--){
3138
                        int run= i - survivor[j];
3139
                        int score= distortion + score_tab[i-run];
3140
                        if(score < last_score){
3141
                            last_score= score;
3142
                            last_run= run;
3143
                            last_level= level-64;
3144
                            last_i= i+1;
3145
                        }
3146
                    }
3147
                }
3148
            }
3149
        }
3150

    
3151
        score_tab[i+1]= best_score;
3152

    
3153
        //Note: there is a vlc code in mpeg4 which is 1 bit shorter then another one with a shorter run and the same level
3154
        if(last_non_zero <= 27){
3155
            for(; survivor_count; survivor_count--){
3156
                if(score_tab[ survivor[survivor_count-1] ] <= best_score)
3157
                    break;
3158
            }
3159
        }else{
3160
            for(; survivor_count; survivor_count--){
3161
                if(score_tab[ survivor[survivor_count-1] ] <= best_score + lambda)
3162
                    break;
3163
            }
3164
        }
3165

    
3166
        survivor[ survivor_count++ ]= i+1;
3167
    }
3168

    
3169
    if(s->out_format != FMT_H263){
3170
        last_score= 256*256*256*120;
3171
        for(i= survivor[0]; i<=last_non_zero + 1; i++){
3172
            int score= score_tab[i];
3173
            if(i) score += lambda*2; //FIXME exacter?
3174

    
3175
            if(score < last_score){
3176
                last_score= score;
3177
                last_i= i;
3178
                last_level= level_tab[i];
3179
                last_run= run_tab[i];
3180
            }
3181
        }
3182
    }
3183

    
3184
    s->coded_score[n] = last_score;
3185

    
3186
    dc= FFABS(block[0]);
3187
    last_non_zero= last_i - 1;
3188
    memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
3189

    
3190
    if(last_non_zero < start_i)
3191
        return last_non_zero;
3192

    
3193
    if(last_non_zero == 0 && start_i == 0){
3194
        int best_level= 0;
3195
        int best_score= dc * dc;
3196

    
3197
        for(i=0; i<coeff_count[0]; i++){
3198
            int level= coeff[i][0];
3199
            int alevel= FFABS(level);
3200
            int unquant_coeff, score, distortion;
3201

    
3202
            if(s->out_format == FMT_H263){
3203
                    unquant_coeff= (alevel*qmul + qadd)>>3;
3204
            }else{ //MPEG1
3205
                    unquant_coeff = (((  alevel  << 1) + 1) * qscale * ((int) s->inter_matrix[0])) >> 4;
3206
                    unquant_coeff =   (unquant_coeff - 1) | 1;
3207
            }
3208
            unquant_coeff = (unquant_coeff + 4) >> 3;
3209
            unquant_coeff<<= 3 + 3;
3210

    
3211
            distortion= (unquant_coeff - dc) * (unquant_coeff - dc);
3212
            level+=64;
3213
            if((level&(~127)) == 0) score= distortion + last_length[UNI_AC_ENC_INDEX(0, level)]*lambda;
3214
            else                    score= distortion + esc_length*lambda;
3215

    
3216
            if(score < best_score){
3217
                best_score= score;
3218
                best_level= level - 64;
3219
            }
3220
        }
3221
        block[0]= best_level;
3222
        s->coded_score[n] = best_score - dc*dc;
3223
        if(best_level == 0) return -1;
3224
        else                return last_non_zero;
3225
    }
3226

    
3227
    i= last_i;
3228
    assert(last_level);
3229

    
3230
    block[ perm_scantable[last_non_zero] ]= last_level;
3231
    i -= last_run + 1;
3232

    
3233
    for(; i>start_i; i -= run_tab[i] + 1){
3234
        block[ perm_scantable[i-1] ]= level_tab[i];
3235
    }
3236

    
3237
    return last_non_zero;
3238
}
3239

    
3240
//#define REFINE_STATS 1
3241
static int16_t basis[64][64];
3242

    
3243
static void build_basis(uint8_t *perm){
3244
    int i, j, x, y;
3245
    emms_c();
3246
    for(i=0; i<8; i++){
3247
        for(j=0; j<8; j++){
3248
            for(y=0; y<8; y++){
3249
                for(x=0; x<8; x++){
3250
                    double s= 0.25*(1<<BASIS_SHIFT);
3251
                    int index= 8*i + j;
3252
                    int perm_index= perm[index];
3253
                    if(i==0) s*= sqrt(0.5);
3254
                    if(j==0) s*= sqrt(0.5);
3255
                    basis[perm_index][8*x + y]= lrintf(s * cos((M_PI/8.0)*i*(x+0.5)) * cos((M_PI/8.0)*j*(y+0.5)));
3256
                }
3257
            }
3258
        }
3259
    }
3260
}
3261

    
3262
static int dct_quantize_refine(MpegEncContext *s, //FIXME breaks denoise?
3263
                        DCTELEM *block, int16_t *weight, DCTELEM *orig,
3264
                        int n, int qscale){
3265
    int16_t rem[64];
3266
    DECLARE_ALIGNED_16(DCTELEM, d1[64]);
3267
    const int *qmat;
3268
    const uint8_t *scantable= s->intra_scantable.scantable;
3269
    const uint8_t *perm_scantable= s->intra_scantable.permutated;
3270
//    unsigned int threshold1, threshold2;
3271
//    int bias=0;
3272
    int run_tab[65];
3273
    int prev_run=0;
3274
    int prev_level=0;
3275
    int qmul, qadd, start_i, last_non_zero, i, dc;
3276
    uint8_t * length;
3277
    uint8_t * last_length;
3278
    int lambda;
3279
    int rle_index, run, q = 1, sum; //q is only used when s->mb_intra is true
3280
#ifdef REFINE_STATS
3281
static int count=0;
3282
static int after_last=0;
3283
static int to_zero=0;
3284
static int from_zero=0;
3285
static int raise=0;
3286
static int lower=0;
3287
static int messed_sign=0;
3288
#endif
3289

    
3290
    if(basis[0][0] == 0)
3291
        build_basis(s->dsp.idct_permutation);
3292

    
3293
    qmul= qscale*2;
3294
    qadd= (qscale-1)|1;
3295
    if (s->mb_intra) {
3296
        if (!s->h263_aic) {
3297
            if (n < 4)
3298
                q = s->y_dc_scale;
3299
            else
3300
                q = s->c_dc_scale;
3301
        } else{
3302
            /* For AIC we skip quant/dequant of INTRADC */
3303
            q = 1;
3304
            qadd=0;
3305
        }
3306
        q <<= RECON_SHIFT-3;
3307
        /* note: block[0] is assumed to be positive */
3308
        dc= block[0]*q;
3309
//        block[0] = (block[0] + (q >> 1)) / q;
3310
        start_i = 1;
3311
        qmat = s->q_intra_matrix[qscale];
3312
//        if(s->mpeg_quant || s->out_format == FMT_MPEG1)
3313
//            bias= 1<<(QMAT_SHIFT-1);
3314
        length     = s->intra_ac_vlc_length;
3315
        last_length= s->intra_ac_vlc_last_length;
3316
    } else {
3317
        dc= 0;
3318
        start_i = 0;
3319
        qmat = s->q_inter_matrix[qscale];
3320
        length     = s->inter_ac_vlc_length;
3321
        last_length= s->inter_ac_vlc_last_length;
3322
    }
3323
    last_non_zero = s->block_last_index[n];
3324

    
3325
#ifdef REFINE_STATS
3326
{START_TIMER
3327
#endif
3328
    dc += (1<<(RECON_SHIFT-1));
3329
    for(i=0; i<64; i++){
3330
        rem[i]= dc - (orig[i]<<RECON_SHIFT); //FIXME  use orig dirrectly instead of copying to rem[]
3331
    }
3332
#ifdef REFINE_STATS
3333
STOP_TIMER("memset rem[]")}
3334
#endif
3335
    sum=0;
3336
    for(i=0; i<64; i++){
3337
        int one= 36;
3338
        int qns=4;
3339
        int w;
3340

    
3341
        w= FFABS(weight[i]) + qns*one;
3342
        w= 15 + (48*qns*one + w/2)/w; // 16 .. 63
3343

    
3344
        weight[i] = w;
3345
//        w=weight[i] = (63*qns + (w/2)) / w;
3346

    
3347
        assert(w>0);
3348
        assert(w<(1<<6));
3349
        sum += w*w;
3350
    }
3351
    lambda= sum*(uint64_t)s->lambda2 >> (FF_LAMBDA_SHIFT - 6 + 6 + 6 + 6);
3352
#ifdef REFINE_STATS
3353
{START_TIMER
3354
#endif
3355
    run=0;
3356
    rle_index=0;
3357
    for(i=start_i; i<=last_non_zero; i++){
3358
        int j= perm_scantable[i];
3359
        const int level= block[j];
3360
        int coeff;
3361

    
3362
        if(level){
3363
            if(level<0) coeff= qmul*level - qadd;
3364
            else        coeff= qmul*level + qadd;
3365
            run_tab[rle_index++]=run;
3366
            run=0;
3367

    
3368
            s->dsp.add_8x8basis(rem, basis[j], coeff);
3369
        }else{
3370
            run++;
3371
        }
3372
    }
3373
#ifdef REFINE_STATS
3374
if(last_non_zero>0){
3375
STOP_TIMER("init rem[]")
3376
}
3377
}
3378

    
3379
{START_TIMER
3380
#endif
3381
    for(;;){
3382
        int best_score=s->dsp.try_8x8basis(rem, weight, basis[0], 0);
3383
        int best_coeff=0;
3384
        int best_change=0;
3385
        int run2, best_unquant_change=0, analyze_gradient;
3386
#ifdef REFINE_STATS
3387
{START_TIMER
3388
#endif
3389
        analyze_gradient = last_non_zero > 2 || s->avctx->quantizer_noise_shaping >= 3;
3390

    
3391
        if(analyze_gradient){
3392
#ifdef REFINE_STATS
3393
{START_TIMER
3394
#endif
3395
            for(i=0; i<64; i++){
3396
                int w= weight[i];
3397

    
3398
                d1[i] = (rem[i]*w*w + (1<<(RECON_SHIFT+12-1)))>>(RECON_SHIFT+12);
3399
            }
3400
#ifdef REFINE_STATS
3401
STOP_TIMER("rem*w*w")}
3402
{START_TIMER
3403
#endif
3404
            s->dsp.fdct(d1);
3405
#ifdef REFINE_STATS
3406
STOP_TIMER("dct")}
3407
#endif
3408
        }
3409

    
3410
        if(start_i){
3411
            const int level= block[0];
3412
            int change, old_coeff;
3413

    
3414
            assert(s->mb_intra);
3415

    
3416
            old_coeff= q*level;
3417

    
3418
            for(change=-1; change<=1; change+=2){
3419
                int new_level= level + change;
3420
                int score, new_coeff;
3421

    
3422
                new_coeff= q*new_level;
3423
                if(new_coeff >= 2048 || new_coeff < 0)
3424
                    continue;
3425

    
3426
                score= s->dsp.try_8x8basis(rem, weight, basis[0], new_coeff - old_coeff);
3427
                if(score<best_score){
3428
                    best_score= score;
3429
                    best_coeff= 0;
3430
                    best_change= change;
3431
                    best_unquant_change= new_coeff - old_coeff;
3432
                }
3433
            }
3434
        }
3435

    
3436
        run=0;
3437
        rle_index=0;
3438
        run2= run_tab[rle_index++];
3439
        prev_level=0;
3440
        prev_run=0;
3441

    
3442
        for(i=start_i; i<64; i++){
3443
            int j= perm_scantable[i];
3444
            const int level= block[j];
3445
            int change, old_coeff;
3446

    
3447
            if(s->avctx->quantizer_noise_shaping < 3 && i > last_non_zero + 1)
3448
                break;
3449

    
3450
            if(level){
3451
                if(level<0) old_coeff= qmul*level - qadd;
3452
                else        old_coeff= qmul*level + qadd;
3453
                run2= run_tab[rle_index++]; //FIXME ! maybe after last
3454
            }else{
3455
                old_coeff=0;
3456
                run2--;
3457
                assert(run2>=0 || i >= last_non_zero );
3458
            }
3459

    
3460
            for(change=-1; change<=1; change+=2){
3461
                int new_level= level + change;
3462
                int score, new_coeff, unquant_change;
3463

    
3464
                score=0;
3465
                if(s->avctx->quantizer_noise_shaping < 2 && FFABS(new_level) > FFABS(level))
3466
                   continue;
3467

    
3468
                if(new_level){
3469
                    if(new_level<0) new_coeff= qmul*new_level - qadd;
3470
                    else            new_coeff= qmul*new_level + qadd;
3471
                    if(new_coeff >= 2048 || new_coeff <= -2048)
3472
                        continue;
3473
                    //FIXME check for overflow
3474

    
3475
                    if(level){
3476
                        if(level < 63 && level > -63){
3477
                            if(i < last_non_zero)
3478
                                score +=   length[UNI_AC_ENC_INDEX(run, new_level+64)]
3479
                                         - length[UNI_AC_ENC_INDEX(run, level+64)];
3480
                            else
3481
                                score +=   last_length[UNI_AC_ENC_INDEX(run, new_level+64)]
3482
                                         - last_length[UNI_AC_ENC_INDEX(run, level+64)];
3483
                        }
3484
                    }else{
3485
                        assert(FFABS(new_level)==1);
3486

    
3487
                        if(analyze_gradient){
3488
                            int g= d1[ scantable[i] ];
3489
                            if(g && (g^new_level) >= 0)
3490
                                continue;
3491
                        }
3492

    
3493
                        if(i < last_non_zero){
3494
                            int next_i= i + run2 + 1;
3495
                            int next_level= block[ perm_scantable[next_i] ] + 64;
3496

    
3497
                            if(next_level&(~127))
3498
                                next_level= 0;
3499

    
3500
                            if(next_i < last_non_zero)
3501
                                score +=   length[UNI_AC_ENC_INDEX(run, 65)]
3502
                                         + length[UNI_AC_ENC_INDEX(run2, next_level)]
3503
                                         - length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)];
3504
                            else
3505
                                score +=  length[UNI_AC_ENC_INDEX(run, 65)]
3506
                                        + last_length[UNI_AC_ENC_INDEX(run2, next_level)]
3507
                                        - last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)];
3508
                        }else{
3509
                            score += last_length[UNI_AC_ENC_INDEX(run, 65)];
3510
                            if(prev_level){
3511
                                score +=  length[UNI_AC_ENC_INDEX(prev_run, prev_level)]
3512
                                        - last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)];
3513
                            }
3514
                        }
3515
                    }
3516
                }else{
3517
                    new_coeff=0;
3518
                    assert(FFABS(level)==1);
3519

    
3520
                    if(i < last_non_zero){
3521
                        int next_i= i + run2 + 1;
3522
                        int next_level= block[ perm_scantable[next_i] ] + 64;
3523

    
3524
                        if(next_level&(~127))
3525
                            next_level= 0;
3526

    
3527
                        if(next_i < last_non_zero)
3528
                            score +=   length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)]
3529
                                     - length[UNI_AC_ENC_INDEX(run2, next_level)]
3530
                                     - length[UNI_AC_ENC_INDEX(run, 65)];
3531
                        else
3532
                            score +=   last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)]
3533
                                     - last_length[UNI_AC_ENC_INDEX(run2, next_level)]
3534
                                     - length[UNI_AC_ENC_INDEX(run, 65)];
3535
                    }else{
3536
                        score += -last_length[UNI_AC_ENC_INDEX(run, 65)];
3537
                        if(prev_level){
3538
                            score +=  last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)]
3539
                                    - length[UNI_AC_ENC_INDEX(prev_run, prev_level)];
3540
                        }
3541
                    }
3542
                }
3543

    
3544
                score *= lambda;
3545

    
3546
                unquant_change= new_coeff - old_coeff;
3547
                assert((score < 100*lambda && score > -100*lambda) || lambda==0);
3548

    
3549
                score+= s->dsp.try_8x8basis(rem, weight, basis[j], unquant_change);
3550
                if(score<best_score){
3551
                    best_score= score;
3552
                    best_coeff= i;
3553
                    best_change= change;
3554
                    best_unquant_change= unquant_change;
3555
                }
3556
            }
3557
            if(level){
3558
                prev_level= level + 64;
3559
                if(prev_level&(~127))
3560
                    prev_level= 0;
3561
                prev_run= run;
3562
                run=0;
3563
            }else{
3564
                run++;
3565
            }
3566
        }
3567
#ifdef REFINE_STATS
3568
STOP_TIMER("iterative step")}
3569
#endif
3570

    
3571
        if(best_change){
3572
            int j= perm_scantable[ best_coeff ];
3573

    
3574
            block[j] += best_change;
3575

    
3576
            if(best_coeff > last_non_zero){
3577
                last_non_zero= best_coeff;
3578
                assert(block[j]);
3579
#ifdef REFINE_STATS
3580
after_last++;
3581
#endif
3582
            }else{
3583
#ifdef REFINE_STATS
3584
if(block[j]){
3585
    if(block[j] - best_change){
3586
        if(FFABS(block[j]) > FFABS(block[j] - best_change)){
3587
            raise++;
3588
        }else{
3589
            lower++;
3590
        }
3591
    }else{
3592
        from_zero++;
3593
    }
3594
}else{
3595
    to_zero++;
3596
}
3597
#endif
3598
                for(; last_non_zero>=start_i; last_non_zero--){
3599
                    if(block[perm_scantable[last_non_zero]])
3600
                        break;
3601
                }
3602
            }
3603
#ifdef REFINE_STATS
3604
count++;
3605
if(256*256*256*64 % count == 0){
3606
    printf("after_last:%d to_zero:%d from_zero:%d raise:%d lower:%d sign:%d xyp:%d/%d/%d\n", after_last, to_zero, from_zero, raise, lower, messed_sign, s->mb_x, s->mb_y, s->picture_number);
3607
}
3608
#endif
3609
            run=0;
3610
            rle_index=0;
3611
            for(i=start_i; i<=last_non_zero; i++){
3612
                int j= perm_scantable[i];
3613
                const int level= block[j];
3614

    
3615
                 if(level){
3616
                     run_tab[rle_index++]=run;
3617
                     run=0;
3618
                 }else{
3619
                     run++;
3620
                 }
3621
            }
3622

    
3623
            s->dsp.add_8x8basis(rem, basis[j], best_unquant_change);
3624
        }else{
3625
            break;
3626
        }
3627
    }
3628
#ifdef REFINE_STATS
3629
if(last_non_zero>0){
3630
STOP_TIMER("iterative search")
3631
}
3632
}
3633
#endif
3634

    
3635
    return last_non_zero;
3636
}
3637

    
3638
int dct_quantize_c(MpegEncContext *s,
3639
                        DCTELEM *block, int n,
3640
                        int qscale, int *overflow)
3641
{
3642
    int i, j, level, last_non_zero, q, start_i;
3643
    const int *qmat;
3644
    const uint8_t *scantable= s->intra_scantable.scantable;
3645
    int bias;
3646
    int max=0;
3647
    unsigned int threshold1, threshold2;
3648

    
3649
    s->dsp.fdct (block);
3650

    
3651
    if(s->dct_error_sum)
3652
        s->denoise_dct(s, block);
3653

    
3654
    if (s->mb_intra) {
3655
        if (!s->h263_aic) {
3656
            if (n < 4)
3657
                q = s->y_dc_scale;
3658
            else
3659
                q = s->c_dc_scale;
3660
            q = q << 3;
3661
        } else
3662
            /* For AIC we skip quant/dequant of INTRADC */
3663
            q = 1 << 3;
3664

    
3665
        /* note: block[0] is assumed to be positive */
3666
        block[0] = (block[0] + (q >> 1)) / q;
3667
        start_i = 1;
3668
        last_non_zero = 0;
3669
        qmat = s->q_intra_matrix[qscale];
3670
        bias= s->intra_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT);
3671
    } else {
3672
        start_i = 0;
3673
        last_non_zero = -1;
3674
        qmat = s->q_inter_matrix[qscale];
3675
        bias= s->inter_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT);
3676
    }
3677
    threshold1= (1<<QMAT_SHIFT) - bias - 1;
3678
    threshold2= (threshold1<<1);
3679
    for(i=63;i>=start_i;i--) {
3680
        j = scantable[i];
3681
        level = block[j] * qmat[j];
3682

    
3683
        if(((unsigned)(level+threshold1))>threshold2){
3684
            last_non_zero = i;
3685
            break;
3686
        }else{
3687
            block[j]=0;
3688
        }
3689
    }
3690
    for(i=start_i; i<=last_non_zero; i++) {
3691
        j = scantable[i];
3692
        level = block[j] * qmat[j];
3693

    
3694
//        if(   bias+level >= (1<<QMAT_SHIFT)
3695
//           || bias-level >= (1<<QMAT_SHIFT)){
3696
        if(((unsigned)(level+threshold1))>threshold2){
3697
            if(level>0){
3698
                level= (bias + level)>>QMAT_SHIFT;
3699
                block[j]= level;
3700
            }else{
3701
                level= (bias - level)>>QMAT_SHIFT;
3702
                block[j]= -level;
3703
            }
3704
            max |=level;
3705
        }else{
3706
            block[j]=0;
3707
        }
3708
    }
3709
    *overflow= s->max_qcoeff < max; //overflow might have happened
3710

    
3711
    /* we need this permutation so that we correct the IDCT, we only permute the !=0 elements */
3712
    if (s->dsp.idct_permutation_type != FF_NO_IDCT_PERM)
3713
        ff_block_permute(block, s->dsp.idct_permutation, scantable, last_non_zero);
3714

    
3715
    return last_non_zero;
3716
}
3717

    
3718
AVCodec h263_encoder = {
3719
    "h263",
3720
    CODEC_TYPE_VIDEO,
3721
    CODEC_ID_H263,
3722
    sizeof(MpegEncContext),
3723
    MPV_encode_init,
3724
    MPV_encode_picture,
3725
    MPV_encode_end,
3726
    .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3727
    .long_name= NULL_IF_CONFIG_SMALL("H.263 / H.263-1996"),
3728
};
3729

    
3730
AVCodec h263p_encoder = {
3731
    "h263p",
3732
    CODEC_TYPE_VIDEO,
3733
    CODEC_ID_H263P,
3734
    sizeof(MpegEncContext),
3735
    MPV_encode_init,
3736
    MPV_encode_picture,
3737
    MPV_encode_end,
3738
    .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3739
    .long_name= NULL_IF_CONFIG_SMALL("H.263+ / H.263-1998 / H.263 version 2"),
3740
};
3741

    
3742
AVCodec flv_encoder = {
3743
    "flv",
3744
    CODEC_TYPE_VIDEO,
3745
    CODEC_ID_FLV1,
3746
    sizeof(MpegEncContext),
3747
    MPV_encode_init,
3748
    MPV_encode_picture,
3749
    MPV_encode_end,
3750
    .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3751
    .long_name= NULL_IF_CONFIG_SMALL("Flash Video"),
3752
};
3753

    
3754
AVCodec rv10_encoder = {
3755
    "rv10",
3756
    CODEC_TYPE_VIDEO,
3757
    CODEC_ID_RV10,
3758
    sizeof(MpegEncContext),
3759
    MPV_encode_init,
3760
    MPV_encode_picture,
3761
    MPV_encode_end,
3762
    .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3763
    .long_name= NULL_IF_CONFIG_SMALL("RealVideo 1.0"),
3764
};
3765

    
3766
AVCodec rv20_encoder = {
3767
    "rv20",
3768
    CODEC_TYPE_VIDEO,
3769
    CODEC_ID_RV20,
3770
    sizeof(MpegEncContext),
3771
    MPV_encode_init,
3772
    MPV_encode_picture,
3773
    MPV_encode_end,
3774
    .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3775
    .long_name= NULL_IF_CONFIG_SMALL("RealVideo 2.0"),
3776
};
3777

    
3778
AVCodec mpeg4_encoder = {
3779
    "mpeg4",
3780
    CODEC_TYPE_VIDEO,
3781
    CODEC_ID_MPEG4,
3782
    sizeof(MpegEncContext),
3783
    MPV_encode_init,
3784
    MPV_encode_picture,
3785
    MPV_encode_end,
3786
    .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3787
    .capabilities= CODEC_CAP_DELAY,
3788
    .long_name= NULL_IF_CONFIG_SMALL("MPEG-4 part 2"),
3789
};
3790

    
3791
AVCodec msmpeg4v1_encoder = {
3792
    "msmpeg4v1",
3793
    CODEC_TYPE_VIDEO,
3794
    CODEC_ID_MSMPEG4V1,
3795
    sizeof(MpegEncContext),
3796
    MPV_encode_init,
3797
    MPV_encode_picture,
3798
    MPV_encode_end,
3799
    .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3800
    .long_name= NULL_IF_CONFIG_SMALL("MPEG-4 part 2 Microsoft variant version 1"),
3801
};
3802

    
3803
AVCodec msmpeg4v2_encoder = {
3804
    "msmpeg4v2",
3805
    CODEC_TYPE_VIDEO,
3806
    CODEC_ID_MSMPEG4V2,
3807
    sizeof(MpegEncContext),
3808
    MPV_encode_init,
3809
    MPV_encode_picture,
3810
    MPV_encode_end,
3811
    .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3812
    .long_name= NULL_IF_CONFIG_SMALL("MPEG-4 part 2 Microsoft variant version 2"),
3813
};
3814

    
3815
AVCodec msmpeg4v3_encoder = {
3816
    "msmpeg4",
3817
    CODEC_TYPE_VIDEO,
3818
    CODEC_ID_MSMPEG4V3,
3819
    sizeof(MpegEncContext),
3820
    MPV_encode_init,
3821
    MPV_encode_picture,
3822
    MPV_encode_end,
3823
    .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3824
    .long_name= NULL_IF_CONFIG_SMALL("MPEG-4 part 2 Microsoft variant version 3"),
3825
};
3826

    
3827
AVCodec wmv1_encoder = {
3828
    "wmv1",
3829
    CODEC_TYPE_VIDEO,
3830
    CODEC_ID_WMV1,
3831
    sizeof(MpegEncContext),
3832
    MPV_encode_init,
3833
    MPV_encode_picture,
3834
    MPV_encode_end,
3835
    .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3836
    .long_name= NULL_IF_CONFIG_SMALL("Windows Media Video 7"),
3837
};