Prevent AAC frame size overflows.
[ffmpeg.git] / libavcodec / mpegvideo_enc.c
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 libavcodec/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 && avctx->pix_fmt != PIX_FMT_RGB32 &&
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     s->progressive_frame=
653     s->progressive_sequence= !(avctx->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN));
654
655     /* init */
656     if (MPV_common_init(s) < 0)
657         return -1;
658
659     if(!s->dct_quantize)
660         s->dct_quantize = dct_quantize_c;
661     if(!s->denoise_dct)
662         s->denoise_dct = denoise_dct_c;
663     s->fast_dct_quantize = s->dct_quantize;
664     if(avctx->trellis)
665         s->dct_quantize = dct_quantize_trellis_c;
666
667     if((CONFIG_H263P_ENCODER || CONFIG_RV20_ENCODER) && s->modified_quant)
668         s->chroma_qscale_table= ff_h263_chroma_qscale_table;
669
670     s->quant_precision=5;
671
672     ff_set_cmp(&s->dsp, s->dsp.ildct_cmp, s->avctx->ildct_cmp);
673     ff_set_cmp(&s->dsp, s->dsp.frame_skip_cmp, s->avctx->frame_skip_cmp);
674
675     if (CONFIG_H261_ENCODER && s->out_format == FMT_H261)
676         ff_h261_encode_init(s);
677     if (CONFIG_ANY_H263_ENCODER && s->out_format == FMT_H263)
678         h263_encode_init(s);
679     if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version)
680         ff_msmpeg4_encode_init(s);
681     if ((CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
682         && s->out_format == FMT_MPEG1)
683         ff_mpeg1_encode_init(s);
684
685     /* init q matrix */
686     for(i=0;i<64;i++) {
687         int j= s->dsp.idct_permutation[i];
688         if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4 && s->mpeg_quant){
689             s->intra_matrix[j] = ff_mpeg4_default_intra_matrix[i];
690             s->inter_matrix[j] = ff_mpeg4_default_non_intra_matrix[i];
691         }else if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
692             s->intra_matrix[j] =
693             s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
694         }else
695         { /* mpeg1/2 */
696             s->intra_matrix[j] = ff_mpeg1_default_intra_matrix[i];
697             s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
698         }
699         if(s->avctx->intra_matrix)
700             s->intra_matrix[j] = s->avctx->intra_matrix[i];
701         if(s->avctx->inter_matrix)
702             s->inter_matrix[j] = s->avctx->inter_matrix[i];
703     }
704
705     /* precompute matrix */
706     /* for mjpeg, we do include qscale in the matrix */
707     if (s->out_format != FMT_MJPEG) {
708         ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
709                        s->intra_matrix, s->intra_quant_bias, avctx->qmin, 31, 1);
710         ff_convert_matrix(&s->dsp, s->q_inter_matrix, s->q_inter_matrix16,
711                        s->inter_matrix, s->inter_quant_bias, avctx->qmin, 31, 0);
712     }
713
714     if(ff_rate_control_init(s) < 0)
715         return -1;
716
717     return 0;
718 }
719
720 av_cold int MPV_encode_end(AVCodecContext *avctx)
721 {
722     MpegEncContext *s = avctx->priv_data;
723
724     ff_rate_control_uninit(s);
725
726     MPV_common_end(s);
727     if ((CONFIG_MJPEG_ENCODER || CONFIG_LJPEG_ENCODER) && s->out_format == FMT_MJPEG)
728         ff_mjpeg_encode_close(s);
729
730     av_freep(&avctx->extradata);
731
732     return 0;
733 }
734
735 static int get_sae(uint8_t *src, int ref, int stride){
736     int x,y;
737     int acc=0;
738
739     for(y=0; y<16; y++){
740         for(x=0; x<16; x++){
741             acc+= FFABS(src[x+y*stride] - ref);
742         }
743     }
744
745     return acc;
746 }
747
748 static int get_intra_count(MpegEncContext *s, uint8_t *src, uint8_t *ref, int stride){
749     int x, y, w, h;
750     int acc=0;
751
752     w= s->width &~15;
753     h= s->height&~15;
754
755     for(y=0; y<h; y+=16){
756         for(x=0; x<w; x+=16){
757             int offset= x + y*stride;
758             int sad = s->dsp.sad[0](NULL, src + offset, ref + offset, stride, 16);
759             int mean= (s->dsp.pix_sum(src + offset, stride) + 128)>>8;
760             int sae = get_sae(src + offset, mean, stride);
761
762             acc+= sae + 500 < sad;
763         }
764     }
765     return acc;
766 }
767
768
769 static int load_input_picture(MpegEncContext *s, AVFrame *pic_arg){
770     AVFrame *pic=NULL;
771     int64_t pts;
772     int i;
773     const int encoding_delay= s->max_b_frames;
774     int direct=1;
775
776     if(pic_arg){
777         pts= pic_arg->pts;
778         pic_arg->display_picture_number= s->input_picture_number++;
779
780         if(pts != AV_NOPTS_VALUE){
781             if(s->user_specified_pts != AV_NOPTS_VALUE){
782                 int64_t time= pts;
783                 int64_t last= s->user_specified_pts;
784
785                 if(time <= last){
786                     av_log(s->avctx, AV_LOG_ERROR, "Error, Invalid timestamp=%"PRId64", last=%"PRId64"\n", pts, s->user_specified_pts);
787                     return -1;
788                 }
789             }
790             s->user_specified_pts= pts;
791         }else{
792             if(s->user_specified_pts != AV_NOPTS_VALUE){
793                 s->user_specified_pts=
794                 pts= s->user_specified_pts + 1;
795                 av_log(s->avctx, AV_LOG_INFO, "Warning: AVFrame.pts=? trying to guess (%"PRId64")\n", pts);
796             }else{
797                 pts= pic_arg->display_picture_number;
798             }
799         }
800     }
801
802   if(pic_arg){
803     if(encoding_delay && !(s->flags&CODEC_FLAG_INPUT_PRESERVED)) direct=0;
804     if(pic_arg->linesize[0] != s->linesize) direct=0;
805     if(pic_arg->linesize[1] != s->uvlinesize) direct=0;
806     if(pic_arg->linesize[2] != s->uvlinesize) direct=0;
807
808 //    av_log(AV_LOG_DEBUG, "%d %d %d %d\n",pic_arg->linesize[0], pic_arg->linesize[1], s->linesize, s->uvlinesize);
809
810     if(direct){
811         i= ff_find_unused_picture(s, 1);
812
813         pic= (AVFrame*)&s->picture[i];
814         pic->reference= 3;
815
816         for(i=0; i<4; i++){
817             pic->data[i]= pic_arg->data[i];
818             pic->linesize[i]= pic_arg->linesize[i];
819         }
820         ff_alloc_picture(s, (Picture*)pic, 1);
821     }else{
822         i= ff_find_unused_picture(s, 0);
823
824         pic= (AVFrame*)&s->picture[i];
825         pic->reference= 3;
826
827         ff_alloc_picture(s, (Picture*)pic, 0);
828
829         if(   pic->data[0] + INPLACE_OFFSET == pic_arg->data[0]
830            && pic->data[1] + INPLACE_OFFSET == pic_arg->data[1]
831            && pic->data[2] + INPLACE_OFFSET == pic_arg->data[2]){
832        // empty
833         }else{
834             int h_chroma_shift, v_chroma_shift;
835             avcodec_get_chroma_sub_sample(s->avctx->pix_fmt, &h_chroma_shift, &v_chroma_shift);
836
837             for(i=0; i<3; i++){
838                 int src_stride= pic_arg->linesize[i];
839                 int dst_stride= i ? s->uvlinesize : s->linesize;
840                 int h_shift= i ? h_chroma_shift : 0;
841                 int v_shift= i ? v_chroma_shift : 0;
842                 int w= s->width >>h_shift;
843                 int h= s->height>>v_shift;
844                 uint8_t *src= pic_arg->data[i];
845                 uint8_t *dst= pic->data[i];
846
847                 if(!s->avctx->rc_buffer_size)
848                     dst +=INPLACE_OFFSET;
849
850                 if(src_stride==dst_stride)
851                     memcpy(dst, src, src_stride*h);
852                 else{
853                     while(h--){
854                         memcpy(dst, src, w);
855                         dst += dst_stride;
856                         src += src_stride;
857                     }
858                 }
859             }
860         }
861     }
862     copy_picture_attributes(s, pic, pic_arg);
863     pic->pts= pts; //we set this here to avoid modifiying pic_arg
864   }
865
866     /* shift buffer entries */
867     for(i=1; i<MAX_PICTURE_COUNT /*s->encoding_delay+1*/; i++)
868         s->input_picture[i-1]= s->input_picture[i];
869
870     s->input_picture[encoding_delay]= (Picture*)pic;
871
872     return 0;
873 }
874
875 static int skip_check(MpegEncContext *s, Picture *p, Picture *ref){
876     int x, y, plane;
877     int score=0;
878     int64_t score64=0;
879
880     for(plane=0; plane<3; plane++){
881         const int stride= p->linesize[plane];
882         const int bw= plane ? 1 : 2;
883         for(y=0; y<s->mb_height*bw; y++){
884             for(x=0; x<s->mb_width*bw; x++){
885                 int off= p->type == FF_BUFFER_TYPE_SHARED ? 0: 16;
886                 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);
887
888                 switch(s->avctx->frame_skip_exp){
889                     case 0: score= FFMAX(score, v); break;
890                     case 1: score+= FFABS(v);break;
891                     case 2: score+= v*v;break;
892                     case 3: score64+= FFABS(v*v*(int64_t)v);break;
893                     case 4: score64+= v*v*(int64_t)(v*v);break;
894                 }
895             }
896         }
897     }
898
899     if(score) score64= score;
900
901     if(score64 < s->avctx->frame_skip_threshold)
902         return 1;
903     if(score64 < ((s->avctx->frame_skip_factor * (int64_t)s->lambda)>>8))
904         return 1;
905     return 0;
906 }
907
908 static int estimate_best_b_count(MpegEncContext *s){
909     AVCodec *codec= avcodec_find_encoder(s->avctx->codec_id);
910     AVCodecContext *c= avcodec_alloc_context();
911     AVFrame input[FF_MAX_B_FRAMES+2];
912     const int scale= s->avctx->brd_scale;
913     int i, j, out_size, p_lambda, b_lambda, lambda2;
914     int outbuf_size= s->width * s->height; //FIXME
915     uint8_t *outbuf= av_malloc(outbuf_size);
916     int64_t best_rd= INT64_MAX;
917     int best_b_count= -1;
918
919     assert(scale>=0 && scale <=3);
920
921 //    emms_c();
922     p_lambda= s->last_lambda_for[FF_P_TYPE]; //s->next_picture_ptr->quality;
923     b_lambda= s->last_lambda_for[FF_B_TYPE]; //p_lambda *FFABS(s->avctx->b_quant_factor) + s->avctx->b_quant_offset;
924     if(!b_lambda) b_lambda= p_lambda; //FIXME we should do this somewhere else
925     lambda2= (b_lambda*b_lambda + (1<<FF_LAMBDA_SHIFT)/2 ) >> FF_LAMBDA_SHIFT;
926
927     c->width = s->width >> scale;
928     c->height= s->height>> scale;
929     c->flags= CODEC_FLAG_QSCALE | CODEC_FLAG_PSNR | CODEC_FLAG_INPUT_PRESERVED /*| CODEC_FLAG_EMU_EDGE*/;
930     c->flags|= s->avctx->flags & CODEC_FLAG_QPEL;
931     c->mb_decision= s->avctx->mb_decision;
932     c->me_cmp= s->avctx->me_cmp;
933     c->mb_cmp= s->avctx->mb_cmp;
934     c->me_sub_cmp= s->avctx->me_sub_cmp;
935     c->pix_fmt = PIX_FMT_YUV420P;
936     c->time_base= s->avctx->time_base;
937     c->max_b_frames= s->max_b_frames;
938
939     if (avcodec_open(c, codec) < 0)
940         return -1;
941
942     for(i=0; i<s->max_b_frames+2; i++){
943         int ysize= c->width*c->height;
944         int csize= (c->width/2)*(c->height/2);
945         Picture pre_input, *pre_input_ptr= i ? s->input_picture[i-1] : s->next_picture_ptr;
946
947         avcodec_get_frame_defaults(&input[i]);
948         input[i].data[0]= av_malloc(ysize + 2*csize);
949         input[i].data[1]= input[i].data[0] + ysize;
950         input[i].data[2]= input[i].data[1] + csize;
951         input[i].linesize[0]= c->width;
952         input[i].linesize[1]=
953         input[i].linesize[2]= c->width/2;
954
955         if(pre_input_ptr && (!i || s->input_picture[i-1])) {
956             pre_input= *pre_input_ptr;
957
958             if(pre_input.type != FF_BUFFER_TYPE_SHARED && i) {
959                 pre_input.data[0]+=INPLACE_OFFSET;
960                 pre_input.data[1]+=INPLACE_OFFSET;
961                 pre_input.data[2]+=INPLACE_OFFSET;
962             }
963
964             s->dsp.shrink[scale](input[i].data[0], input[i].linesize[0], pre_input.data[0], pre_input.linesize[0], c->width, c->height);
965             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);
966             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);
967         }
968     }
969
970     for(j=0; j<s->max_b_frames+1; j++){
971         int64_t rd=0;
972
973         if(!s->input_picture[j])
974             break;
975
976         c->error[0]= c->error[1]= c->error[2]= 0;
977
978         input[0].pict_type= FF_I_TYPE;
979         input[0].quality= 1 * FF_QP2LAMBDA;
980         out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[0]);
981 //        rd += (out_size * lambda2) >> FF_LAMBDA_SHIFT;
982
983         for(i=0; i<s->max_b_frames+1; i++){
984             int is_p= i % (j+1) == j || i==s->max_b_frames;
985
986             input[i+1].pict_type= is_p ? FF_P_TYPE : FF_B_TYPE;
987             input[i+1].quality= is_p ? p_lambda : b_lambda;
988             out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[i+1]);
989             rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
990         }
991
992         /* get the delayed frames */
993         while(out_size){
994             out_size = avcodec_encode_video(c, outbuf, outbuf_size, NULL);
995             rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
996         }
997
998         rd += c->error[0] + c->error[1] + c->error[2];
999
1000         if(rd < best_rd){
1001             best_rd= rd;
1002             best_b_count= j;
1003         }
1004     }
1005
1006     av_freep(&outbuf);
1007     avcodec_close(c);
1008     av_freep(&c);
1009
1010     for(i=0; i<s->max_b_frames+2; i++){
1011         av_freep(&input[i].data[0]);
1012     }
1013
1014     return best_b_count;
1015 }
1016
1017 static void select_input_picture(MpegEncContext *s){
1018     int i;
1019
1020     for(i=1; i<MAX_PICTURE_COUNT; i++)
1021         s->reordered_input_picture[i-1]= s->reordered_input_picture[i];
1022     s->reordered_input_picture[MAX_PICTURE_COUNT-1]= NULL;
1023
1024     /* set next picture type & ordering */
1025     if(s->reordered_input_picture[0]==NULL && s->input_picture[0]){
1026         if(/*s->picture_in_gop_number >= s->gop_size ||*/ s->next_picture_ptr==NULL || s->intra_only){
1027             s->reordered_input_picture[0]= s->input_picture[0];
1028             s->reordered_input_picture[0]->pict_type= FF_I_TYPE;
1029             s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++;
1030         }else{
1031             int b_frames;
1032
1033             if(s->avctx->frame_skip_threshold || s->avctx->frame_skip_factor){
1034                 if(s->picture_in_gop_number < s->gop_size && skip_check(s, s->input_picture[0], s->next_picture_ptr)){
1035                 //FIXME check that te gop check above is +-1 correct
1036 //av_log(NULL, AV_LOG_DEBUG, "skip %p %"PRId64"\n", s->input_picture[0]->data[0], s->input_picture[0]->pts);
1037
1038                     if(s->input_picture[0]->type == FF_BUFFER_TYPE_SHARED){
1039                         for(i=0; i<4; i++)
1040                             s->input_picture[0]->data[i]= NULL;
1041                         s->input_picture[0]->type= 0;
1042                     }else{
1043                         assert(   s->input_picture[0]->type==FF_BUFFER_TYPE_USER
1044                                || s->input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
1045
1046                         s->avctx->release_buffer(s->avctx, (AVFrame*)s->input_picture[0]);
1047                     }
1048
1049                     emms_c();
1050                     ff_vbv_update(s, 0);
1051
1052                     goto no_output_pic;
1053                 }
1054             }
1055
1056             if(s->flags&CODEC_FLAG_PASS2){
1057                 for(i=0; i<s->max_b_frames+1; i++){
1058                     int pict_num= s->input_picture[0]->display_picture_number + i;
1059
1060                     if(pict_num >= s->rc_context.num_entries)
1061                         break;
1062                     if(!s->input_picture[i]){
1063                         s->rc_context.entry[pict_num-1].new_pict_type = FF_P_TYPE;
1064                         break;
1065                     }
1066
1067                     s->input_picture[i]->pict_type=
1068                         s->rc_context.entry[pict_num].new_pict_type;
1069                 }
1070             }
1071
1072             if(s->avctx->b_frame_strategy==0){
1073                 b_frames= s->max_b_frames;
1074                 while(b_frames && !s->input_picture[b_frames]) b_frames--;
1075             }else if(s->avctx->b_frame_strategy==1){
1076                 for(i=1; i<s->max_b_frames+1; i++){
1077                     if(s->input_picture[i] && s->input_picture[i]->b_frame_score==0){
1078                         s->input_picture[i]->b_frame_score=
1079                             get_intra_count(s, s->input_picture[i  ]->data[0],
1080                                                s->input_picture[i-1]->data[0], s->linesize) + 1;
1081                     }
1082                 }
1083                 for(i=0; i<s->max_b_frames+1; i++){
1084                     if(s->input_picture[i]==NULL || s->input_picture[i]->b_frame_score - 1 > s->mb_num/s->avctx->b_sensitivity) break;
1085                 }
1086
1087                 b_frames= FFMAX(0, i-1);
1088
1089                 /* reset scores */
1090                 for(i=0; i<b_frames+1; i++){
1091                     s->input_picture[i]->b_frame_score=0;
1092                 }
1093             }else if(s->avctx->b_frame_strategy==2){
1094                 b_frames= estimate_best_b_count(s);
1095             }else{
1096                 av_log(s->avctx, AV_LOG_ERROR, "illegal b frame strategy\n");
1097                 b_frames=0;
1098             }
1099
1100             emms_c();
1101 //static int b_count=0;
1102 //b_count+= b_frames;
1103 //av_log(s->avctx, AV_LOG_DEBUG, "b_frames: %d\n", b_count);
1104
1105             for(i= b_frames - 1; i>=0; i--){
1106                 int type= s->input_picture[i]->pict_type;
1107                 if(type && type != FF_B_TYPE)
1108                     b_frames= i;
1109             }
1110             if(s->input_picture[b_frames]->pict_type == FF_B_TYPE && b_frames == s->max_b_frames){
1111                 av_log(s->avctx, AV_LOG_ERROR, "warning, too many b frames in a row\n");
1112             }
1113
1114             if(s->picture_in_gop_number + b_frames >= s->gop_size){
1115               if((s->flags2 & CODEC_FLAG2_STRICT_GOP) && s->gop_size > s->picture_in_gop_number){
1116                     b_frames= s->gop_size - s->picture_in_gop_number - 1;
1117               }else{
1118                 if(s->flags & CODEC_FLAG_CLOSED_GOP)
1119                     b_frames=0;
1120                 s->input_picture[b_frames]->pict_type= FF_I_TYPE;
1121               }
1122             }
1123
1124             if(   (s->flags & CODEC_FLAG_CLOSED_GOP)
1125                && b_frames
1126                && s->input_picture[b_frames]->pict_type== FF_I_TYPE)
1127                 b_frames--;
1128
1129             s->reordered_input_picture[0]= s->input_picture[b_frames];
1130             if(s->reordered_input_picture[0]->pict_type != FF_I_TYPE)
1131                 s->reordered_input_picture[0]->pict_type= FF_P_TYPE;
1132             s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++;
1133             for(i=0; i<b_frames; i++){
1134                 s->reordered_input_picture[i+1]= s->input_picture[i];
1135                 s->reordered_input_picture[i+1]->pict_type= FF_B_TYPE;
1136                 s->reordered_input_picture[i+1]->coded_picture_number= s->coded_picture_number++;
1137             }
1138         }
1139     }
1140 no_output_pic:
1141     if(s->reordered_input_picture[0]){
1142         s->reordered_input_picture[0]->reference= s->reordered_input_picture[0]->pict_type!=FF_B_TYPE ? 3 : 0;
1143
1144         ff_copy_picture(&s->new_picture, s->reordered_input_picture[0]);
1145
1146         if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_SHARED || s->avctx->rc_buffer_size){
1147             // input is a shared pix, so we can't modifiy it -> alloc a new one & ensure that the shared one is reuseable
1148
1149             int i= ff_find_unused_picture(s, 0);
1150             Picture *pic= &s->picture[i];
1151
1152             pic->reference              = s->reordered_input_picture[0]->reference;
1153             ff_alloc_picture(s, pic, 0);
1154
1155             /* mark us unused / free shared pic */
1156             if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_INTERNAL)
1157                 s->avctx->release_buffer(s->avctx, (AVFrame*)s->reordered_input_picture[0]);
1158             for(i=0; i<4; i++)
1159                 s->reordered_input_picture[0]->data[i]= NULL;
1160             s->reordered_input_picture[0]->type= 0;
1161
1162             copy_picture_attributes(s, (AVFrame*)pic, (AVFrame*)s->reordered_input_picture[0]);
1163
1164             s->current_picture_ptr= pic;
1165         }else{
1166             // input is not a shared pix -> reuse buffer for current_pix
1167
1168             assert(   s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_USER
1169                    || s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
1170
1171             s->current_picture_ptr= s->reordered_input_picture[0];
1172             for(i=0; i<4; i++){
1173                 s->new_picture.data[i]+= INPLACE_OFFSET;
1174             }
1175         }
1176         ff_copy_picture(&s->current_picture, s->current_picture_ptr);
1177
1178         s->picture_number= s->new_picture.display_picture_number;
1179 //printf("dpn:%d\n", s->picture_number);
1180     }else{
1181        memset(&s->new_picture, 0, sizeof(Picture));
1182     }
1183 }
1184
1185 int MPV_encode_picture(AVCodecContext *avctx,
1186                        unsigned char *buf, int buf_size, void *data)
1187 {
1188     MpegEncContext *s = avctx->priv_data;
1189     AVFrame *pic_arg = data;
1190     int i, stuffing_count;
1191
1192     for(i=0; i<avctx->thread_count; i++){
1193         int start_y= s->thread_context[i]->start_mb_y;
1194         int   end_y= s->thread_context[i]->  end_mb_y;
1195         int h= s->mb_height;
1196         uint8_t *start= buf + (size_t)(((int64_t) buf_size)*start_y/h);
1197         uint8_t *end  = buf + (size_t)(((int64_t) buf_size)*  end_y/h);
1198
1199         init_put_bits(&s->thread_context[i]->pb, start, end - start);
1200     }
1201
1202     s->picture_in_gop_number++;
1203
1204     if(load_input_picture(s, pic_arg) < 0)
1205         return -1;
1206
1207     select_input_picture(s);
1208
1209     /* output? */
1210     if(s->new_picture.data[0]){
1211         s->pict_type= s->new_picture.pict_type;
1212 //emms_c();
1213 //printf("qs:%f %f %d\n", s->new_picture.quality, s->current_picture.quality, s->qscale);
1214         MPV_frame_start(s, avctx);
1215 vbv_retry:
1216         if (encode_picture(s, s->picture_number) < 0)
1217             return -1;
1218
1219         avctx->header_bits = s->header_bits;
1220         avctx->mv_bits     = s->mv_bits;
1221         avctx->misc_bits   = s->misc_bits;
1222         avctx->i_tex_bits  = s->i_tex_bits;
1223         avctx->p_tex_bits  = s->p_tex_bits;
1224         avctx->i_count     = s->i_count;
1225         avctx->p_count     = s->mb_num - s->i_count - s->skip_count; //FIXME f/b_count in avctx
1226         avctx->skip_count  = s->skip_count;
1227
1228         MPV_frame_end(s);
1229
1230         if (CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG)
1231             ff_mjpeg_encode_picture_trailer(s);
1232
1233         if(avctx->rc_buffer_size){
1234             RateControlContext *rcc= &s->rc_context;
1235             int max_size= rcc->buffer_index * avctx->rc_max_available_vbv_use;
1236
1237             if(put_bits_count(&s->pb) > max_size && s->lambda < s->avctx->lmax){
1238                 s->next_lambda= FFMAX(s->lambda+1, s->lambda*(s->qscale+1) / s->qscale);
1239                 if(s->adaptive_quant){
1240                     int i;
1241                     for(i=0; i<s->mb_height*s->mb_stride; i++)
1242                         s->lambda_table[i]= FFMAX(s->lambda_table[i]+1, s->lambda_table[i]*(s->qscale+1) / s->qscale);
1243                 }
1244                 s->mb_skipped = 0;        //done in MPV_frame_start()
1245                 if(s->pict_type==FF_P_TYPE){ //done in encode_picture() so we must undo it
1246                     if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
1247                         s->no_rounding ^= 1;
1248                 }
1249                 if(s->pict_type!=FF_B_TYPE){
1250                     s->time_base= s->last_time_base;
1251                     s->last_non_b_time= s->time - s->pp_time;
1252                 }
1253 //                av_log(NULL, AV_LOG_ERROR, "R:%d ", s->next_lambda);
1254                 for(i=0; i<avctx->thread_count; i++){
1255                     PutBitContext *pb= &s->thread_context[i]->pb;
1256                     init_put_bits(pb, pb->buf, pb->buf_end - pb->buf);
1257                 }
1258                 goto vbv_retry;
1259             }
1260
1261             assert(s->avctx->rc_max_rate);
1262         }
1263
1264         if(s->flags&CODEC_FLAG_PASS1)
1265             ff_write_pass1_stats(s);
1266
1267         for(i=0; i<4; i++){
1268             s->current_picture_ptr->error[i]= s->current_picture.error[i];
1269             avctx->error[i] += s->current_picture_ptr->error[i];
1270         }
1271
1272         if(s->flags&CODEC_FLAG_PASS1)
1273             assert(avctx->header_bits + avctx->mv_bits + avctx->misc_bits + avctx->i_tex_bits + avctx->p_tex_bits == put_bits_count(&s->pb));
1274         flush_put_bits(&s->pb);
1275         s->frame_bits  = put_bits_count(&s->pb);
1276
1277         stuffing_count= ff_vbv_update(s, s->frame_bits);
1278         if(stuffing_count){
1279             if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < stuffing_count + 50){
1280                 av_log(s->avctx, AV_LOG_ERROR, "stuffing too large\n");
1281                 return -1;
1282             }
1283
1284             switch(s->codec_id){
1285             case CODEC_ID_MPEG1VIDEO:
1286             case CODEC_ID_MPEG2VIDEO:
1287                 while(stuffing_count--){
1288                     put_bits(&s->pb, 8, 0);
1289                 }
1290             break;
1291             case CODEC_ID_MPEG4:
1292                 put_bits(&s->pb, 16, 0);
1293                 put_bits(&s->pb, 16, 0x1C3);
1294                 stuffing_count -= 4;
1295                 while(stuffing_count--){
1296                     put_bits(&s->pb, 8, 0xFF);
1297                 }
1298             break;
1299             default:
1300                 av_log(s->avctx, AV_LOG_ERROR, "vbv buffer overflow\n");
1301             }
1302             flush_put_bits(&s->pb);
1303             s->frame_bits  = put_bits_count(&s->pb);
1304         }
1305
1306         /* update mpeg1/2 vbv_delay for CBR */
1307         if(s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate && s->out_format == FMT_MPEG1
1308            && 90000LL * (avctx->rc_buffer_size-1) <= s->avctx->rc_max_rate*0xFFFFLL){
1309             int vbv_delay, min_delay;
1310             double inbits = s->avctx->rc_max_rate*av_q2d(s->avctx->time_base);
1311             int    minbits= s->frame_bits - 8*(s->vbv_delay_ptr - s->pb.buf - 1);
1312             double bits   = s->rc_context.buffer_index + minbits - inbits;
1313
1314             if(bits<0)
1315                 av_log(s->avctx, AV_LOG_ERROR, "Internal error, negative bits\n");
1316
1317             assert(s->repeat_first_field==0);
1318
1319             vbv_delay=     bits * 90000                               / s->avctx->rc_max_rate;
1320             min_delay= (minbits * 90000LL + s->avctx->rc_max_rate - 1)/ s->avctx->rc_max_rate;
1321
1322             vbv_delay= FFMAX(vbv_delay, min_delay);
1323
1324             assert(vbv_delay < 0xFFFF);
1325
1326             s->vbv_delay_ptr[0] &= 0xF8;
1327             s->vbv_delay_ptr[0] |= vbv_delay>>13;
1328             s->vbv_delay_ptr[1]  = vbv_delay>>5;
1329             s->vbv_delay_ptr[2] &= 0x07;
1330             s->vbv_delay_ptr[2] |= vbv_delay<<3;
1331         }
1332         s->total_bits += s->frame_bits;
1333         avctx->frame_bits  = s->frame_bits;
1334     }else{
1335         assert((put_bits_ptr(&s->pb) == s->pb.buf));
1336         s->frame_bits=0;
1337     }
1338     assert((s->frame_bits&7)==0);
1339
1340     return s->frame_bits/8;
1341 }
1342
1343 static inline void dct_single_coeff_elimination(MpegEncContext *s, int n, int threshold)
1344 {
1345     static const char tab[64]=
1346         {3,2,2,1,1,1,1,1,
1347          1,1,1,1,1,1,1,1,
1348          1,1,1,1,1,1,1,1,
1349          0,0,0,0,0,0,0,0,
1350          0,0,0,0,0,0,0,0,
1351          0,0,0,0,0,0,0,0,
1352          0,0,0,0,0,0,0,0,
1353          0,0,0,0,0,0,0,0};
1354     int score=0;
1355     int run=0;
1356     int i;
1357     DCTELEM *block= s->block[n];
1358     const int last_index= s->block_last_index[n];
1359     int skip_dc;
1360
1361     if(threshold<0){
1362         skip_dc=0;
1363         threshold= -threshold;
1364     }else
1365         skip_dc=1;
1366
1367     /* Are all we could set to zero already zero? */
1368     if(last_index<=skip_dc - 1) return;
1369
1370     for(i=0; i<=last_index; i++){
1371         const int j = s->intra_scantable.permutated[i];
1372         const int level = FFABS(block[j]);
1373         if(level==1){
1374             if(skip_dc && i==0) continue;
1375             score+= tab[run];
1376             run=0;
1377         }else if(level>1){
1378             return;
1379         }else{
1380             run++;
1381         }
1382     }
1383     if(score >= threshold) return;
1384     for(i=skip_dc; i<=last_index; i++){
1385         const int j = s->intra_scantable.permutated[i];
1386         block[j]=0;
1387     }
1388     if(block[0]) s->block_last_index[n]= 0;
1389     else         s->block_last_index[n]= -1;
1390 }
1391
1392 static inline void clip_coeffs(MpegEncContext *s, DCTELEM *block, int last_index)
1393 {
1394     int i;
1395     const int maxlevel= s->max_qcoeff;
1396     const int minlevel= s->min_qcoeff;
1397     int overflow=0;
1398
1399     if(s->mb_intra){
1400         i=1; //skip clipping of intra dc
1401     }else
1402         i=0;
1403
1404     for(;i<=last_index; i++){
1405         const int j= s->intra_scantable.permutated[i];
1406         int level = block[j];
1407
1408         if     (level>maxlevel){
1409             level=maxlevel;
1410             overflow++;
1411         }else if(level<minlevel){
1412             level=minlevel;
1413             overflow++;
1414         }
1415
1416         block[j]= level;
1417     }
1418
1419     if(overflow && s->avctx->mb_decision == FF_MB_DECISION_SIMPLE)
1420         av_log(s->avctx, AV_LOG_INFO, "warning, clipping %d dct coefficients to %d..%d\n", overflow, minlevel, maxlevel);
1421 }
1422
1423 static void get_visual_weight(int16_t *weight, uint8_t *ptr, int stride){
1424     int x, y;
1425 //FIXME optimize
1426     for(y=0; y<8; y++){
1427         for(x=0; x<8; x++){
1428             int x2, y2;
1429             int sum=0;
1430             int sqr=0;
1431             int count=0;
1432
1433             for(y2= FFMAX(y-1, 0); y2 < FFMIN(8, y+2); y2++){
1434                 for(x2= FFMAX(x-1, 0); x2 < FFMIN(8, x+2); x2++){
1435                     int v= ptr[x2 + y2*stride];
1436                     sum += v;
1437                     sqr += v*v;
1438                     count++;
1439                 }
1440             }
1441             weight[x + 8*y]= (36*ff_sqrt(count*sqr - sum*sum)) / count;
1442         }
1443     }
1444 }
1445
1446 static av_always_inline void encode_mb_internal(MpegEncContext *s, int motion_x, int motion_y, int mb_block_height, int mb_block_count)
1447 {
1448     int16_t weight[8][64];
1449     DCTELEM orig[8][64];
1450     const int mb_x= s->mb_x;
1451     const int mb_y= s->mb_y;
1452     int i;
1453     int skip_dct[8];
1454     int dct_offset   = s->linesize*8; //default for progressive frames
1455     uint8_t *ptr_y, *ptr_cb, *ptr_cr;
1456     int wrap_y, wrap_c;
1457
1458     for(i=0; i<mb_block_count; i++) skip_dct[i]=s->skipdct;
1459
1460     if(s->adaptive_quant){
1461         const int last_qp= s->qscale;
1462         const int mb_xy= mb_x + mb_y*s->mb_stride;
1463
1464         s->lambda= s->lambda_table[mb_xy];
1465         update_qscale(s);
1466
1467         if(!(s->flags&CODEC_FLAG_QP_RD)){
1468             s->qscale= s->current_picture_ptr->qscale_table[mb_xy];
1469             s->dquant= s->qscale - last_qp;
1470
1471             if(s->out_format==FMT_H263){
1472                 s->dquant= av_clip(s->dquant, -2, 2);
1473
1474                 if(s->codec_id==CODEC_ID_MPEG4){
1475                     if(!s->mb_intra){
1476                         if(s->pict_type == FF_B_TYPE){
1477                             if(s->dquant&1 || s->mv_dir&MV_DIRECT)
1478                                 s->dquant= 0;
1479                         }
1480                         if(s->mv_type==MV_TYPE_8X8)
1481                             s->dquant=0;
1482                     }
1483                 }
1484             }
1485         }
1486         ff_set_qscale(s, last_qp + s->dquant);
1487     }else if(s->flags&CODEC_FLAG_QP_RD)
1488         ff_set_qscale(s, s->qscale + s->dquant);
1489
1490     wrap_y = s->linesize;
1491     wrap_c = s->uvlinesize;
1492     ptr_y = s->new_picture.data[0] + (mb_y * 16 * wrap_y) + mb_x * 16;
1493     ptr_cb = s->new_picture.data[1] + (mb_y * mb_block_height * wrap_c) + mb_x * 8;
1494     ptr_cr = s->new_picture.data[2] + (mb_y * mb_block_height * wrap_c) + mb_x * 8;
1495
1496     if(mb_x*16+16 > s->width || mb_y*16+16 > s->height){
1497         uint8_t *ebuf= s->edge_emu_buffer + 32;
1498         ff_emulated_edge_mc(ebuf            , ptr_y , wrap_y,16,16,mb_x*16,mb_y*16, s->width   , s->height);
1499         ptr_y= ebuf;
1500         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);
1501         ptr_cb= ebuf+18*wrap_y;
1502         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);
1503         ptr_cr= ebuf+18*wrap_y+8;
1504     }
1505
1506     if (s->mb_intra) {
1507         if(s->flags&CODEC_FLAG_INTERLACED_DCT){
1508             int progressive_score, interlaced_score;
1509
1510             s->interlaced_dct=0;
1511             progressive_score= s->dsp.ildct_cmp[4](s, ptr_y           , NULL, wrap_y, 8)
1512                               +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y*8, NULL, wrap_y, 8) - 400;
1513
1514             if(progressive_score > 0){
1515                 interlaced_score = s->dsp.ildct_cmp[4](s, ptr_y           , NULL, wrap_y*2, 8)
1516                                   +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y  , NULL, wrap_y*2, 8);
1517                 if(progressive_score > interlaced_score){
1518                     s->interlaced_dct=1;
1519
1520                     dct_offset= wrap_y;
1521                     wrap_y<<=1;
1522                     if (s->chroma_format == CHROMA_422)
1523                         wrap_c<<=1;
1524                 }
1525             }
1526         }
1527
1528         s->dsp.get_pixels(s->block[0], ptr_y                 , wrap_y);
1529         s->dsp.get_pixels(s->block[1], ptr_y              + 8, wrap_y);
1530         s->dsp.get_pixels(s->block[2], ptr_y + dct_offset    , wrap_y);
1531         s->dsp.get_pixels(s->block[3], ptr_y + dct_offset + 8, wrap_y);
1532
1533         if(s->flags&CODEC_FLAG_GRAY){
1534             skip_dct[4]= 1;
1535             skip_dct[5]= 1;
1536         }else{
1537             s->dsp.get_pixels(s->block[4], ptr_cb, wrap_c);
1538             s->dsp.get_pixels(s->block[5], ptr_cr, wrap_c);
1539             if(!s->chroma_y_shift){ /* 422 */
1540                 s->dsp.get_pixels(s->block[6], ptr_cb + (dct_offset>>1), wrap_c);
1541                 s->dsp.get_pixels(s->block[7], ptr_cr + (dct_offset>>1), wrap_c);
1542             }
1543         }
1544     }else{
1545         op_pixels_func (*op_pix)[4];
1546         qpel_mc_func (*op_qpix)[16];
1547         uint8_t *dest_y, *dest_cb, *dest_cr;
1548
1549         dest_y  = s->dest[0];
1550         dest_cb = s->dest[1];
1551         dest_cr = s->dest[2];
1552
1553         if ((!s->no_rounding) || s->pict_type==FF_B_TYPE){
1554             op_pix = s->dsp.put_pixels_tab;
1555             op_qpix= s->dsp.put_qpel_pixels_tab;
1556         }else{
1557             op_pix = s->dsp.put_no_rnd_pixels_tab;
1558             op_qpix= s->dsp.put_no_rnd_qpel_pixels_tab;
1559         }
1560
1561         if (s->mv_dir & MV_DIR_FORWARD) {
1562             MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.data, op_pix, op_qpix);
1563             op_pix = s->dsp.avg_pixels_tab;
1564             op_qpix= s->dsp.avg_qpel_pixels_tab;
1565         }
1566         if (s->mv_dir & MV_DIR_BACKWARD) {
1567             MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.data, op_pix, op_qpix);
1568         }
1569
1570         if(s->flags&CODEC_FLAG_INTERLACED_DCT){
1571             int progressive_score, interlaced_score;
1572
1573             s->interlaced_dct=0;
1574             progressive_score= s->dsp.ildct_cmp[0](s, dest_y           , ptr_y           , wrap_y, 8)
1575                               +s->dsp.ildct_cmp[0](s, dest_y + wrap_y*8, ptr_y + wrap_y*8, wrap_y, 8) - 400;
1576
1577             if(s->avctx->ildct_cmp == FF_CMP_VSSE) progressive_score -= 400;
1578
1579             if(progressive_score>0){
1580                 interlaced_score = s->dsp.ildct_cmp[0](s, dest_y           , ptr_y           , wrap_y*2, 8)
1581                                   +s->dsp.ildct_cmp[0](s, dest_y + wrap_y  , ptr_y + wrap_y  , wrap_y*2, 8);
1582
1583                 if(progressive_score > interlaced_score){
1584                     s->interlaced_dct=1;
1585
1586                     dct_offset= wrap_y;
1587                     wrap_y<<=1;
1588                     if (s->chroma_format == CHROMA_422)
1589                         wrap_c<<=1;
1590                 }
1591             }
1592         }
1593
1594         s->dsp.diff_pixels(s->block[0], ptr_y                 , dest_y                 , wrap_y);
1595         s->dsp.diff_pixels(s->block[1], ptr_y              + 8, dest_y              + 8, wrap_y);
1596         s->dsp.diff_pixels(s->block[2], ptr_y + dct_offset    , dest_y + dct_offset    , wrap_y);
1597         s->dsp.diff_pixels(s->block[3], ptr_y + dct_offset + 8, dest_y + dct_offset + 8, wrap_y);
1598
1599         if(s->flags&CODEC_FLAG_GRAY){
1600             skip_dct[4]= 1;
1601             skip_dct[5]= 1;
1602         }else{
1603             s->dsp.diff_pixels(s->block[4], ptr_cb, dest_cb, wrap_c);
1604             s->dsp.diff_pixels(s->block[5], ptr_cr, dest_cr, wrap_c);
1605             if(!s->chroma_y_shift){ /* 422 */
1606                 s->dsp.diff_pixels(s->block[6], ptr_cb + (dct_offset>>1), dest_cb + (dct_offset>>1), wrap_c);
1607                 s->dsp.diff_pixels(s->block[7], ptr_cr + (dct_offset>>1), dest_cr + (dct_offset>>1), wrap_c);
1608             }
1609         }
1610         /* pre quantization */
1611         if(s->current_picture.mc_mb_var[s->mb_stride*mb_y+ mb_x]<2*s->qscale*s->qscale){
1612             //FIXME optimize
1613             if(s->dsp.sad[1](NULL, ptr_y               , dest_y               , wrap_y, 8) < 20*s->qscale) skip_dct[0]= 1;
1614             if(s->dsp.sad[1](NULL, ptr_y            + 8, dest_y            + 8, wrap_y, 8) < 20*s->qscale) skip_dct[1]= 1;
1615             if(s->dsp.sad[1](NULL, ptr_y +dct_offset   , dest_y +dct_offset   , wrap_y, 8) < 20*s->qscale) skip_dct[2]= 1;
1616             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;
1617             if(s->dsp.sad[1](NULL, ptr_cb              , dest_cb              , wrap_c, 8) < 20*s->qscale) skip_dct[4]= 1;
1618             if(s->dsp.sad[1](NULL, ptr_cr              , dest_cr              , wrap_c, 8) < 20*s->qscale) skip_dct[5]= 1;
1619             if(!s->chroma_y_shift){ /* 422 */
1620                 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;
1621                 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;
1622             }
1623         }
1624     }
1625
1626     if(s->avctx->quantizer_noise_shaping){
1627         if(!skip_dct[0]) get_visual_weight(weight[0], ptr_y                 , wrap_y);
1628         if(!skip_dct[1]) get_visual_weight(weight[1], ptr_y              + 8, wrap_y);
1629         if(!skip_dct[2]) get_visual_weight(weight[2], ptr_y + dct_offset    , wrap_y);
1630         if(!skip_dct[3]) get_visual_weight(weight[3], ptr_y + dct_offset + 8, wrap_y);
1631         if(!skip_dct[4]) get_visual_weight(weight[4], ptr_cb                , wrap_c);
1632         if(!skip_dct[5]) get_visual_weight(weight[5], ptr_cr                , wrap_c);
1633         if(!s->chroma_y_shift){ /* 422 */
1634             if(!skip_dct[6]) get_visual_weight(weight[6], ptr_cb + (dct_offset>>1), wrap_c);
1635             if(!skip_dct[7]) get_visual_weight(weight[7], ptr_cr + (dct_offset>>1), wrap_c);
1636         }
1637         memcpy(orig[0], s->block[0], sizeof(DCTELEM)*64*mb_block_count);
1638     }
1639
1640     /* DCT & quantize */
1641     assert(s->out_format!=FMT_MJPEG || s->qscale==8);
1642     {
1643         for(i=0;i<mb_block_count;i++) {
1644             if(!skip_dct[i]){
1645                 int overflow;
1646                 s->block_last_index[i] = s->dct_quantize(s, s->block[i], i, s->qscale, &overflow);
1647             // FIXME we could decide to change to quantizer instead of clipping
1648             // JS: I don't think that would be a good idea it could lower quality instead
1649             //     of improve it. Just INTRADC clipping deserves changes in quantizer
1650                 if (overflow) clip_coeffs(s, s->block[i], s->block_last_index[i]);
1651             }else
1652                 s->block_last_index[i]= -1;
1653         }
1654         if(s->avctx->quantizer_noise_shaping){
1655             for(i=0;i<mb_block_count;i++) {
1656                 if(!skip_dct[i]){
1657                     s->block_last_index[i] = dct_quantize_refine(s, s->block[i], weight[i], orig[i], i, s->qscale);
1658                 }
1659             }
1660         }
1661
1662         if(s->luma_elim_threshold && !s->mb_intra)
1663             for(i=0; i<4; i++)
1664                 dct_single_coeff_elimination(s, i, s->luma_elim_threshold);
1665         if(s->chroma_elim_threshold && !s->mb_intra)
1666             for(i=4; i<mb_block_count; i++)
1667                 dct_single_coeff_elimination(s, i, s->chroma_elim_threshold);
1668
1669         if(s->flags & CODEC_FLAG_CBP_RD){
1670             for(i=0;i<mb_block_count;i++) {
1671                 if(s->block_last_index[i] == -1)
1672                     s->coded_score[i]= INT_MAX/256;
1673             }
1674         }
1675     }
1676
1677     if((s->flags&CODEC_FLAG_GRAY) && s->mb_intra){
1678         s->block_last_index[4]=
1679         s->block_last_index[5]= 0;
1680         s->block[4][0]=
1681         s->block[5][0]= (1024 + s->c_dc_scale/2)/ s->c_dc_scale;
1682     }
1683
1684     //non c quantize code returns incorrect block_last_index FIXME
1685     if(s->alternate_scan && s->dct_quantize != dct_quantize_c){
1686         for(i=0; i<mb_block_count; i++){
1687             int j;
1688             if(s->block_last_index[i]>0){
1689                 for(j=63; j>0; j--){
1690                     if(s->block[i][ s->intra_scantable.permutated[j] ]) break;
1691                 }
1692                 s->block_last_index[i]= j;
1693             }
1694         }
1695     }
1696
1697     /* huffman encode */
1698     switch(s->codec_id){ //FIXME funct ptr could be slightly faster
1699     case CODEC_ID_MPEG1VIDEO:
1700     case CODEC_ID_MPEG2VIDEO:
1701         if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
1702             mpeg1_encode_mb(s, s->block, motion_x, motion_y);
1703         break;
1704     case CODEC_ID_MPEG4:
1705         if (CONFIG_MPEG4_ENCODER)
1706             mpeg4_encode_mb(s, s->block, motion_x, motion_y);
1707         break;
1708     case CODEC_ID_MSMPEG4V2:
1709     case CODEC_ID_MSMPEG4V3:
1710     case CODEC_ID_WMV1:
1711         if (CONFIG_MSMPEG4_ENCODER)
1712             msmpeg4_encode_mb(s, s->block, motion_x, motion_y);
1713         break;
1714     case CODEC_ID_WMV2:
1715         if (CONFIG_WMV2_ENCODER)
1716             ff_wmv2_encode_mb(s, s->block, motion_x, motion_y);
1717         break;
1718     case CODEC_ID_H261:
1719         if (CONFIG_H261_ENCODER)
1720             ff_h261_encode_mb(s, s->block, motion_x, motion_y);
1721         break;
1722     case CODEC_ID_H263:
1723     case CODEC_ID_H263P:
1724     case CODEC_ID_FLV1:
1725     case CODEC_ID_RV10:
1726     case CODEC_ID_RV20:
1727         if (CONFIG_H263_ENCODER || CONFIG_H263P_ENCODER ||
1728             CONFIG_FLV_ENCODER  || CONFIG_RV10_ENCODER  || CONFIG_RV20_ENCODER)
1729             h263_encode_mb(s, s->block, motion_x, motion_y);
1730         break;
1731     case CODEC_ID_MJPEG:
1732         if (CONFIG_MJPEG_ENCODER)
1733             ff_mjpeg_encode_mb(s, s->block);
1734         break;
1735     default:
1736         assert(0);
1737     }
1738 }
1739
1740 static av_always_inline void encode_mb(MpegEncContext *s, int motion_x, int motion_y)
1741 {
1742     if (s->chroma_format == CHROMA_420) encode_mb_internal(s, motion_x, motion_y,  8, 6);
1743     else                                encode_mb_internal(s, motion_x, motion_y, 16, 8);
1744 }
1745
1746 static inline void copy_context_before_encode(MpegEncContext *d, MpegEncContext *s, int type){
1747     int i;
1748
1749     memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
1750
1751     /* mpeg1 */
1752     d->mb_skip_run= s->mb_skip_run;
1753     for(i=0; i<3; i++)
1754         d->last_dc[i]= s->last_dc[i];
1755
1756     /* statistics */
1757     d->mv_bits= s->mv_bits;
1758     d->i_tex_bits= s->i_tex_bits;
1759     d->p_tex_bits= s->p_tex_bits;
1760     d->i_count= s->i_count;
1761     d->f_count= s->f_count;
1762     d->b_count= s->b_count;
1763     d->skip_count= s->skip_count;
1764     d->misc_bits= s->misc_bits;
1765     d->last_bits= 0;
1766
1767     d->mb_skipped= 0;
1768     d->qscale= s->qscale;
1769     d->dquant= s->dquant;
1770
1771     d->esc3_level_length= s->esc3_level_length;
1772 }
1773
1774 static inline void copy_context_after_encode(MpegEncContext *d, MpegEncContext *s, int type){
1775     int i;
1776
1777     memcpy(d->mv, s->mv, 2*4*2*sizeof(int));
1778     memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
1779
1780     /* mpeg1 */
1781     d->mb_skip_run= s->mb_skip_run;
1782     for(i=0; i<3; i++)
1783         d->last_dc[i]= s->last_dc[i];
1784
1785     /* statistics */
1786     d->mv_bits= s->mv_bits;
1787     d->i_tex_bits= s->i_tex_bits;
1788     d->p_tex_bits= s->p_tex_bits;
1789     d->i_count= s->i_count;
1790     d->f_count= s->f_count;
1791     d->b_count= s->b_count;
1792     d->skip_count= s->skip_count;
1793     d->misc_bits= s->misc_bits;
1794
1795     d->mb_intra= s->mb_intra;
1796     d->mb_skipped= s->mb_skipped;
1797     d->mv_type= s->mv_type;
1798     d->mv_dir= s->mv_dir;
1799     d->pb= s->pb;
1800     if(s->data_partitioning){
1801         d->pb2= s->pb2;
1802         d->tex_pb= s->tex_pb;
1803     }
1804     d->block= s->block;
1805     for(i=0; i<8; i++)
1806         d->block_last_index[i]= s->block_last_index[i];
1807     d->interlaced_dct= s->interlaced_dct;
1808     d->qscale= s->qscale;
1809
1810     d->esc3_level_length= s->esc3_level_length;
1811 }
1812
1813 static inline void encode_mb_hq(MpegEncContext *s, MpegEncContext *backup, MpegEncContext *best, int type,
1814                            PutBitContext pb[2], PutBitContext pb2[2], PutBitContext tex_pb[2],
1815                            int *dmin, int *next_block, int motion_x, int motion_y)
1816 {
1817     int score;
1818     uint8_t *dest_backup[3];
1819
1820     copy_context_before_encode(s, backup, type);
1821
1822     s->block= s->blocks[*next_block];
1823     s->pb= pb[*next_block];
1824     if(s->data_partitioning){
1825         s->pb2   = pb2   [*next_block];
1826         s->tex_pb= tex_pb[*next_block];
1827     }
1828
1829     if(*next_block){
1830         memcpy(dest_backup, s->dest, sizeof(s->dest));
1831         s->dest[0] = s->rd_scratchpad;
1832         s->dest[1] = s->rd_scratchpad + 16*s->linesize;
1833         s->dest[2] = s->rd_scratchpad + 16*s->linesize + 8;
1834         assert(s->linesize >= 32); //FIXME
1835     }
1836
1837     encode_mb(s, motion_x, motion_y);
1838
1839     score= put_bits_count(&s->pb);
1840     if(s->data_partitioning){
1841         score+= put_bits_count(&s->pb2);
1842         score+= put_bits_count(&s->tex_pb);
1843     }
1844
1845     if(s->avctx->mb_decision == FF_MB_DECISION_RD){
1846         MPV_decode_mb(s, s->block);
1847
1848         score *= s->lambda2;
1849         score += sse_mb(s) << FF_LAMBDA_SHIFT;
1850     }
1851
1852     if(*next_block){
1853         memcpy(s->dest, dest_backup, sizeof(s->dest));
1854     }
1855
1856     if(score<*dmin){
1857         *dmin= score;
1858         *next_block^=1;
1859
1860         copy_context_after_encode(best, s, type);
1861     }
1862 }
1863
1864 static int sse(MpegEncContext *s, uint8_t *src1, uint8_t *src2, int w, int h, int stride){
1865     uint32_t *sq = ff_squareTbl + 256;
1866     int acc=0;
1867     int x,y;
1868
1869     if(w==16 && h==16)
1870         return s->dsp.sse[0](NULL, src1, src2, stride, 16);
1871     else if(w==8 && h==8)
1872         return s->dsp.sse[1](NULL, src1, src2, stride, 8);
1873
1874     for(y=0; y<h; y++){
1875         for(x=0; x<w; x++){
1876             acc+= sq[src1[x + y*stride] - src2[x + y*stride]];
1877         }
1878     }
1879
1880     assert(acc>=0);
1881
1882     return acc;
1883 }
1884
1885 static int sse_mb(MpegEncContext *s){
1886     int w= 16;
1887     int h= 16;
1888
1889     if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
1890     if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
1891
1892     if(w==16 && h==16)
1893       if(s->avctx->mb_cmp == FF_CMP_NSSE){
1894         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)
1895                +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)
1896                +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);
1897       }else{
1898         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)
1899                +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)
1900                +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);
1901       }
1902     else
1903         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)
1904                +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)
1905                +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);
1906 }
1907
1908 static int pre_estimate_motion_thread(AVCodecContext *c, void *arg){
1909     MpegEncContext *s= *(void**)arg;
1910
1911
1912     s->me.pre_pass=1;
1913     s->me.dia_size= s->avctx->pre_dia_size;
1914     s->first_slice_line=1;
1915     for(s->mb_y= s->end_mb_y-1; s->mb_y >= s->start_mb_y; s->mb_y--) {
1916         for(s->mb_x=s->mb_width-1; s->mb_x >=0 ;s->mb_x--) {
1917             ff_pre_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
1918         }
1919         s->first_slice_line=0;
1920     }
1921
1922     s->me.pre_pass=0;
1923
1924     return 0;
1925 }
1926
1927 static int estimate_motion_thread(AVCodecContext *c, void *arg){
1928     MpegEncContext *s= *(void**)arg;
1929
1930     ff_check_alignment();
1931
1932     s->me.dia_size= s->avctx->dia_size;
1933     s->first_slice_line=1;
1934     for(s->mb_y= s->start_mb_y; s->mb_y < s->end_mb_y; s->mb_y++) {
1935         s->mb_x=0; //for block init below
1936         ff_init_block_index(s);
1937         for(s->mb_x=0; s->mb_x < s->mb_width; s->mb_x++) {
1938             s->block_index[0]+=2;
1939             s->block_index[1]+=2;
1940             s->block_index[2]+=2;
1941             s->block_index[3]+=2;
1942
1943             /* compute motion vector & mb_type and store in context */
1944             if(s->pict_type==FF_B_TYPE)
1945                 ff_estimate_b_frame_motion(s, s->mb_x, s->mb_y);
1946             else
1947                 ff_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
1948         }
1949         s->first_slice_line=0;
1950     }
1951     return 0;
1952 }
1953
1954 static int mb_var_thread(AVCodecContext *c, void *arg){
1955     MpegEncContext *s= *(void**)arg;
1956     int mb_x, mb_y;
1957
1958     ff_check_alignment();
1959
1960     for(mb_y=s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
1961         for(mb_x=0; mb_x < s->mb_width; mb_x++) {
1962             int xx = mb_x * 16;
1963             int yy = mb_y * 16;
1964             uint8_t *pix = s->new_picture.data[0] + (yy * s->linesize) + xx;
1965             int varc;
1966             int sum = s->dsp.pix_sum(pix, s->linesize);
1967
1968             varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500 + 128)>>8;
1969
1970             s->current_picture.mb_var [s->mb_stride * mb_y + mb_x] = varc;
1971             s->current_picture.mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
1972             s->me.mb_var_sum_temp    += varc;
1973         }
1974     }
1975     return 0;
1976 }
1977
1978 static void write_slice_end(MpegEncContext *s){
1979     if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4){
1980         if(s->partitioned_frame){
1981             ff_mpeg4_merge_partitions(s);
1982         }
1983
1984         ff_mpeg4_stuffing(&s->pb);
1985     }else if(CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG){
1986         ff_mjpeg_encode_stuffing(&s->pb);
1987     }
1988
1989     align_put_bits(&s->pb);
1990     flush_put_bits(&s->pb);
1991
1992     if((s->flags&CODEC_FLAG_PASS1) && !s->partitioned_frame)
1993         s->misc_bits+= get_bits_diff(s);
1994 }
1995
1996 static int encode_thread(AVCodecContext *c, void *arg){
1997     MpegEncContext *s= *(void**)arg;
1998     int mb_x, mb_y, pdif = 0;
1999     int chr_h= 16>>s->chroma_y_shift;
2000     int i, j;
2001     MpegEncContext best_s, backup_s;
2002     uint8_t bit_buf[2][MAX_MB_BYTES];
2003     uint8_t bit_buf2[2][MAX_MB_BYTES];
2004     uint8_t bit_buf_tex[2][MAX_MB_BYTES];
2005     PutBitContext pb[2], pb2[2], tex_pb[2];
2006 //printf("%d->%d\n", s->resync_mb_y, s->end_mb_y);
2007
2008     ff_check_alignment();
2009
2010     for(i=0; i<2; i++){
2011         init_put_bits(&pb    [i], bit_buf    [i], MAX_MB_BYTES);
2012         init_put_bits(&pb2   [i], bit_buf2   [i], MAX_MB_BYTES);
2013         init_put_bits(&tex_pb[i], bit_buf_tex[i], MAX_MB_BYTES);
2014     }
2015
2016     s->last_bits= put_bits_count(&s->pb);
2017     s->mv_bits=0;
2018     s->misc_bits=0;
2019     s->i_tex_bits=0;
2020     s->p_tex_bits=0;
2021     s->i_count=0;
2022     s->f_count=0;
2023     s->b_count=0;
2024     s->skip_count=0;
2025
2026     for(i=0; i<3; i++){
2027         /* init last dc values */
2028         /* note: quant matrix value (8) is implied here */
2029         s->last_dc[i] = 128 << s->intra_dc_precision;
2030
2031         s->current_picture.error[i] = 0;
2032     }
2033     s->mb_skip_run = 0;
2034     memset(s->last_mv, 0, sizeof(s->last_mv));
2035
2036     s->last_mv_dir = 0;
2037
2038     switch(s->codec_id){
2039     case CODEC_ID_H263:
2040     case CODEC_ID_H263P:
2041     case CODEC_ID_FLV1:
2042         if (CONFIG_H263_ENCODER || CONFIG_H263P_ENCODER || CONFIG_FLV_ENCODER)
2043             s->gob_index = ff_h263_get_gob_height(s);
2044         break;
2045     case CODEC_ID_MPEG4:
2046         if(CONFIG_MPEG4_ENCODER && s->partitioned_frame)
2047             ff_mpeg4_init_partitions(s);
2048         break;
2049     }
2050
2051     s->resync_mb_x=0;
2052     s->resync_mb_y=0;
2053     s->first_slice_line = 1;
2054     s->ptr_lastgob = s->pb.buf;
2055     for(mb_y= s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
2056 //    printf("row %d at %X\n", s->mb_y, (int)s);
2057         s->mb_x=0;
2058         s->mb_y= mb_y;
2059
2060         ff_set_qscale(s, s->qscale);
2061         ff_init_block_index(s);
2062
2063         for(mb_x=0; mb_x < s->mb_width; mb_x++) {
2064             int xy= mb_y*s->mb_stride + mb_x; // removed const, H261 needs to adjust this
2065             int mb_type= s->mb_type[xy];
2066 //            int d;
2067             int dmin= INT_MAX;
2068             int dir;
2069
2070             if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < MAX_MB_BYTES){
2071                 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
2072                 return -1;
2073             }
2074             if(s->data_partitioning){
2075                 if(   s->pb2   .buf_end - s->pb2   .buf - (put_bits_count(&s->    pb2)>>3) < MAX_MB_BYTES
2076                    || s->tex_pb.buf_end - s->tex_pb.buf - (put_bits_count(&s->tex_pb )>>3) < MAX_MB_BYTES){
2077                     av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
2078                     return -1;
2079                 }
2080             }
2081
2082             s->mb_x = mb_x;
2083             s->mb_y = mb_y;  // moved into loop, can get changed by H.261
2084             ff_update_block_index(s);
2085
2086             if(CONFIG_H261_ENCODER && s->codec_id == CODEC_ID_H261){
2087                 ff_h261_reorder_mb_index(s);
2088                 xy= s->mb_y*s->mb_stride + s->mb_x;
2089                 mb_type= s->mb_type[xy];
2090             }
2091
2092             /* write gob / video packet header  */
2093             if(s->rtp_mode){
2094                 int current_packet_size, is_gob_start;
2095
2096                 current_packet_size= ((put_bits_count(&s->pb)+7)>>3) - (s->ptr_lastgob - s->pb.buf);
2097
2098                 is_gob_start= s->avctx->rtp_payload_size && current_packet_size >= s->avctx->rtp_payload_size && mb_y + mb_x>0;
2099
2100                 if(s->start_mb_y == mb_y && mb_y > 0 && mb_x==0) is_gob_start=1;
2101
2102                 switch(s->codec_id){
2103                 case CODEC_ID_H263:
2104                 case CODEC_ID_H263P:
2105                     if(!s->h263_slice_structured)
2106                         if(s->mb_x || s->mb_y%s->gob_index) is_gob_start=0;
2107                     break;
2108                 case CODEC_ID_MPEG2VIDEO:
2109                     if(s->mb_x==0 && s->mb_y!=0) is_gob_start=1;
2110                 case CODEC_ID_MPEG1VIDEO:
2111                     if(s->mb_skip_run) is_gob_start=0;
2112                     break;
2113                 }
2114
2115                 if(is_gob_start){
2116                     if(s->start_mb_y != mb_y || mb_x!=0){
2117                         write_slice_end(s);
2118
2119                         if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4 && s->partitioned_frame){
2120                             ff_mpeg4_init_partitions(s);
2121                         }
2122                     }
2123
2124                     assert((put_bits_count(&s->pb)&7) == 0);
2125                     current_packet_size= put_bits_ptr(&s->pb) - s->ptr_lastgob;
2126
2127                     if(s->avctx->error_rate && s->resync_mb_x + s->resync_mb_y > 0){
2128                         int r= put_bits_count(&s->pb)/8 + s->picture_number + 16 + s->mb_x + s->mb_y;
2129                         int d= 100 / s->avctx->error_rate;
2130                         if(r % d == 0){
2131                             current_packet_size=0;
2132 #ifndef ALT_BITSTREAM_WRITER
2133                             s->pb.buf_ptr= s->ptr_lastgob;
2134 #endif
2135                             assert(put_bits_ptr(&s->pb) == s->ptr_lastgob);
2136                         }
2137                     }
2138
2139                     if (s->avctx->rtp_callback){
2140                         int number_mb = (mb_y - s->resync_mb_y)*s->mb_width + mb_x - s->resync_mb_x;
2141                         s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, current_packet_size, number_mb);
2142                     }
2143
2144                     switch(s->codec_id){
2145                     case CODEC_ID_MPEG4:
2146                         if (CONFIG_MPEG4_ENCODER) {
2147                             ff_mpeg4_encode_video_packet_header(s);
2148                             ff_mpeg4_clean_buffers(s);
2149                         }
2150                     break;
2151                     case CODEC_ID_MPEG1VIDEO:
2152                     case CODEC_ID_MPEG2VIDEO:
2153                         if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER) {
2154                             ff_mpeg1_encode_slice_header(s);
2155                             ff_mpeg1_clean_buffers(s);
2156                         }
2157                     break;
2158                     case CODEC_ID_H263:
2159                     case CODEC_ID_H263P:
2160                         if (CONFIG_H263_ENCODER || CONFIG_H263P_ENCODER)
2161                             h263_encode_gob_header(s, mb_y);
2162                     break;
2163                     }
2164
2165                     if(s->flags&CODEC_FLAG_PASS1){
2166                         int bits= put_bits_count(&s->pb);
2167                         s->misc_bits+= bits - s->last_bits;
2168                         s->last_bits= bits;
2169                     }
2170
2171                     s->ptr_lastgob += current_packet_size;
2172                     s->first_slice_line=1;
2173                     s->resync_mb_x=mb_x;
2174                     s->resync_mb_y=mb_y;
2175                 }
2176             }
2177
2178             if(  (s->resync_mb_x   == s->mb_x)
2179                && s->resync_mb_y+1 == s->mb_y){
2180                 s->first_slice_line=0;
2181             }
2182
2183             s->mb_skipped=0;
2184             s->dquant=0; //only for QP_RD
2185
2186             if(mb_type & (mb_type-1) || (s->flags & CODEC_FLAG_QP_RD)){ // more than 1 MB type possible or CODEC_FLAG_QP_RD
2187                 int next_block=0;
2188                 int pb_bits_count, pb2_bits_count, tex_pb_bits_count;
2189
2190                 copy_context_before_encode(&backup_s, s, -1);
2191                 backup_s.pb= s->pb;
2192                 best_s.data_partitioning= s->data_partitioning;
2193                 best_s.partitioned_frame= s->partitioned_frame;
2194                 if(s->data_partitioning){
2195                     backup_s.pb2= s->pb2;
2196                     backup_s.tex_pb= s->tex_pb;
2197                 }
2198
2199                 if(mb_type&CANDIDATE_MB_TYPE_INTER){
2200                     s->mv_dir = MV_DIR_FORWARD;
2201                     s->mv_type = MV_TYPE_16X16;
2202                     s->mb_intra= 0;
2203                     s->mv[0][0][0] = s->p_mv_table[xy][0];
2204                     s->mv[0][0][1] = s->p_mv_table[xy][1];
2205                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER, pb, pb2, tex_pb,
2206                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2207                 }
2208                 if(mb_type&CANDIDATE_MB_TYPE_INTER_I){
2209                     s->mv_dir = MV_DIR_FORWARD;
2210                     s->mv_type = MV_TYPE_FIELD;
2211                     s->mb_intra= 0;
2212                     for(i=0; i<2; i++){
2213                         j= s->field_select[0][i] = s->p_field_select_table[i][xy];
2214                         s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
2215                         s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
2216                     }
2217                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER_I, pb, pb2, tex_pb,
2218                                  &dmin, &next_block, 0, 0);
2219                 }
2220                 if(mb_type&CANDIDATE_MB_TYPE_SKIPPED){
2221                     s->mv_dir = MV_DIR_FORWARD;
2222                     s->mv_type = MV_TYPE_16X16;
2223                     s->mb_intra= 0;
2224                     s->mv[0][0][0] = 0;
2225                     s->mv[0][0][1] = 0;
2226                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_SKIPPED, pb, pb2, tex_pb,
2227                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2228                 }
2229                 if(mb_type&CANDIDATE_MB_TYPE_INTER4V){
2230                     s->mv_dir = MV_DIR_FORWARD;
2231                     s->mv_type = MV_TYPE_8X8;
2232                     s->mb_intra= 0;
2233                     for(i=0; i<4; i++){
2234                         s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
2235                         s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
2236                     }
2237                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER4V, pb, pb2, tex_pb,
2238                                  &dmin, &next_block, 0, 0);
2239                 }
2240                 if(mb_type&CANDIDATE_MB_TYPE_FORWARD){
2241                     s->mv_dir = MV_DIR_FORWARD;
2242                     s->mv_type = MV_TYPE_16X16;
2243                     s->mb_intra= 0;
2244                     s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
2245                     s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2246                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD, pb, pb2, tex_pb,
2247                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2248                 }
2249                 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD){
2250                     s->mv_dir = MV_DIR_BACKWARD;
2251                     s->mv_type = MV_TYPE_16X16;
2252                     s->mb_intra= 0;
2253                     s->mv[1][0][0] = s->b_back_mv_table[xy][0];
2254                     s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2255                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD, pb, pb2, tex_pb,
2256                                  &dmin, &next_block, s->mv[1][0][0], s->mv[1][0][1]);
2257                 }
2258                 if(mb_type&CANDIDATE_MB_TYPE_BIDIR){
2259                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2260                     s->mv_type = MV_TYPE_16X16;
2261                     s->mb_intra= 0;
2262                     s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
2263                     s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
2264                     s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
2265                     s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2266                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR, pb, pb2, tex_pb,
2267                                  &dmin, &next_block, 0, 0);
2268                 }
2269                 if(mb_type&CANDIDATE_MB_TYPE_FORWARD_I){
2270                     s->mv_dir = MV_DIR_FORWARD;
2271                     s->mv_type = MV_TYPE_FIELD;
2272                     s->mb_intra= 0;
2273                     for(i=0; i<2; i++){
2274                         j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
2275                         s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
2276                         s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
2277                     }
2278                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD_I, pb, pb2, tex_pb,
2279                                  &dmin, &next_block, 0, 0);
2280                 }
2281                 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD_I){
2282                     s->mv_dir = MV_DIR_BACKWARD;
2283                     s->mv_type = MV_TYPE_FIELD;
2284                     s->mb_intra= 0;
2285                     for(i=0; i<2; i++){
2286                         j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
2287                         s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
2288                         s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
2289                     }
2290                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD_I, pb, pb2, tex_pb,
2291                                  &dmin, &next_block, 0, 0);
2292                 }
2293                 if(mb_type&CANDIDATE_MB_TYPE_BIDIR_I){
2294                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2295                     s->mv_type = MV_TYPE_FIELD;
2296                     s->mb_intra= 0;
2297                     for(dir=0; dir<2; dir++){
2298                         for(i=0; i<2; i++){
2299                             j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
2300                             s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
2301                             s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
2302                         }
2303                     }
2304                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR_I, pb, pb2, tex_pb,
2305                                  &dmin, &next_block, 0, 0);
2306                 }
2307                 if(mb_type&CANDIDATE_MB_TYPE_INTRA){
2308                     s->mv_dir = 0;
2309                     s->mv_type = MV_TYPE_16X16;
2310                     s->mb_intra= 1;
2311                     s->mv[0][0][0] = 0;
2312                     s->mv[0][0][1] = 0;
2313                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTRA, pb, pb2, tex_pb,
2314                                  &dmin, &next_block, 0, 0);
2315                     if(s->h263_pred || s->h263_aic){
2316                         if(best_s.mb_intra)
2317                             s->mbintra_table[mb_x + mb_y*s->mb_stride]=1;
2318                         else
2319                             ff_clean_intra_table_entries(s); //old mode?
2320                     }
2321                 }
2322
2323                 if((s->flags & CODEC_FLAG_QP_RD) && dmin < INT_MAX){
2324                     if(best_s.mv_type==MV_TYPE_16X16){ //FIXME move 4mv after QPRD
2325                         const int last_qp= backup_s.qscale;
2326                         int qpi, qp, dc[6];
2327                         DCTELEM ac[6][16];
2328                         const int mvdir= (best_s.mv_dir&MV_DIR_BACKWARD) ? 1 : 0;
2329                         static const int dquant_tab[4]={-1,1,-2,2};
2330
2331                         assert(backup_s.dquant == 0);
2332
2333                         //FIXME intra
2334                         s->mv_dir= best_s.mv_dir;
2335                         s->mv_type = MV_TYPE_16X16;
2336                         s->mb_intra= best_s.mb_intra;
2337                         s->mv[0][0][0] = best_s.mv[0][0][0];
2338                         s->mv[0][0][1] = best_s.mv[0][0][1];
2339                         s->mv[1][0][0] = best_s.mv[1][0][0];
2340                         s->mv[1][0][1] = best_s.mv[1][0][1];
2341
2342                         qpi = s->pict_type == FF_B_TYPE ? 2 : 0;
2343                         for(; qpi<4; qpi++){
2344                             int dquant= dquant_tab[qpi];
2345                             qp= last_qp + dquant;
2346                             if(qp < s->avctx->qmin || qp > s->avctx->qmax)
2347                                 continue;
2348                             backup_s.dquant= dquant;
2349                             if(s->mb_intra && s->dc_val[0]){
2350                                 for(i=0; i<6; i++){
2351                                     dc[i]= s->dc_val[0][ s->block_index[i] ];
2352                                     memcpy(ac[i], s->ac_val[0][s->block_index[i]], sizeof(DCTELEM)*16);
2353                                 }
2354                             }
2355
2356                             encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
2357                                          &dmin, &next_block, s->mv[mvdir][0][0], s->mv[mvdir][0][1]);
2358                             if(best_s.qscale != qp){
2359                                 if(s->mb_intra && s->dc_val[0]){
2360                                     for(i=0; i<6; i++){
2361                                         s->dc_val[0][ s->block_index[i] ]= dc[i];
2362                                         memcpy(s->ac_val[0][s->block_index[i]], ac[i], sizeof(DCTELEM)*16);
2363                                     }
2364                                 }
2365                             }
2366                         }
2367                     }
2368                 }
2369                 if(CONFIG_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT){
2370                     int mx= s->b_direct_mv_table[xy][0];
2371                     int my= s->b_direct_mv_table[xy][1];
2372
2373                     backup_s.dquant = 0;
2374                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2375                     s->mb_intra= 0;
2376                     ff_mpeg4_set_direct_mv(s, mx, my);
2377                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
2378                                  &dmin, &next_block, mx, my);
2379                 }
2380                 if(CONFIG_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT0){
2381                     backup_s.dquant = 0;
2382                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2383                     s->mb_intra= 0;
2384                     ff_mpeg4_set_direct_mv(s, 0, 0);
2385                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
2386                                  &dmin, &next_block, 0, 0);
2387                 }
2388                 if(!best_s.mb_intra && s->flags2&CODEC_FLAG2_SKIP_RD){
2389                     int coded=0;
2390                     for(i=0; i<6; i++)
2391                         coded |= s->block_last_index[i];
2392                     if(coded){
2393                         int mx,my;
2394                         memcpy(s->mv, best_s.mv, sizeof(s->mv));
2395                         if(CONFIG_MPEG4_ENCODER && best_s.mv_dir & MV_DIRECT){
2396                             mx=my=0; //FIXME find the one we actually used
2397                             ff_mpeg4_set_direct_mv(s, mx, my);
2398                         }else if(best_s.mv_dir&MV_DIR_BACKWARD){
2399                             mx= s->mv[1][0][0];
2400                             my= s->mv[1][0][1];
2401                         }else{
2402                             mx= s->mv[0][0][0];
2403                             my= s->mv[0][0][1];
2404                         }
2405
2406                         s->mv_dir= best_s.mv_dir;
2407                         s->mv_type = best_s.mv_type;
2408                         s->mb_intra= 0;
2409 /*                        s->mv[0][0][0] = best_s.mv[0][0][0];
2410                         s->mv[0][0][1] = best_s.mv[0][0][1];
2411                         s->mv[1][0][0] = best_s.mv[1][0][0];
2412                         s->mv[1][0][1] = best_s.mv[1][0][1];*/
2413                         backup_s.dquant= 0;
2414                         s->skipdct=1;
2415                         encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
2416                                         &dmin, &next_block, mx, my);
2417                         s->skipdct=0;
2418                     }
2419                 }
2420
2421                 s->current_picture.qscale_table[xy]= best_s.qscale;
2422
2423                 copy_context_after_encode(s, &best_s, -1);
2424
2425                 pb_bits_count= put_bits_count(&s->pb);
2426                 flush_put_bits(&s->pb);
2427                 ff_copy_bits(&backup_s.pb, bit_buf[next_block^1], pb_bits_count);
2428                 s->pb= backup_s.pb;
2429
2430                 if(s->data_partitioning){
2431                     pb2_bits_count= put_bits_count(&s->pb2);
2432                     flush_put_bits(&s->pb2);
2433                     ff_copy_bits(&backup_s.pb2, bit_buf2[next_block^1], pb2_bits_count);
2434                     s->pb2= backup_s.pb2;
2435
2436                     tex_pb_bits_count= put_bits_count(&s->tex_pb);
2437                     flush_put_bits(&s->tex_pb);
2438                     ff_copy_bits(&backup_s.tex_pb, bit_buf_tex[next_block^1], tex_pb_bits_count);
2439                     s->tex_pb= backup_s.tex_pb;
2440                 }
2441                 s->last_bits= put_bits_count(&s->pb);
2442
2443                 if (CONFIG_ANY_H263_ENCODER &&
2444                     s->out_format == FMT_H263 && s->pict_type!=FF_B_TYPE)
2445                     ff_h263_update_motion_val(s);
2446
2447                 if(next_block==0){ //FIXME 16 vs linesize16
2448                     s->dsp.put_pixels_tab[0][0](s->dest[0], s->rd_scratchpad                     , s->linesize  ,16);
2449                     s->dsp.put_pixels_tab[1][0](s->dest[1], s->rd_scratchpad + 16*s->linesize    , s->uvlinesize, 8);
2450                     s->dsp.put_pixels_tab[1][0](s->dest[2], s->rd_scratchpad + 16*s->linesize + 8, s->uvlinesize, 8);
2451                 }
2452
2453                 if(s->avctx->mb_decision == FF_MB_DECISION_BITS)
2454                     MPV_decode_mb(s, s->block);
2455             } else {
2456                 int motion_x = 0, motion_y = 0;
2457                 s->mv_type=MV_TYPE_16X16;
2458                 // only one MB-Type possible
2459
2460                 switch(mb_type){
2461                 case CANDIDATE_MB_TYPE_INTRA:
2462                     s->mv_dir = 0;
2463                     s->mb_intra= 1;
2464                     motion_x= s->mv[0][0][0] = 0;
2465                     motion_y= s->mv[0][0][1] = 0;
2466                     break;
2467                 case CANDIDATE_MB_TYPE_INTER:
2468                     s->mv_dir = MV_DIR_FORWARD;
2469                     s->mb_intra= 0;
2470                     motion_x= s->mv[0][0][0] = s->p_mv_table[xy][0];
2471                     motion_y= s->mv[0][0][1] = s->p_mv_table[xy][1];
2472                     break;
2473                 case CANDIDATE_MB_TYPE_INTER_I:
2474                     s->mv_dir = MV_DIR_FORWARD;
2475                     s->mv_type = MV_TYPE_FIELD;
2476                     s->mb_intra= 0;
2477                     for(i=0; i<2; i++){
2478                         j= s->field_select[0][i] = s->p_field_select_table[i][xy];
2479                         s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
2480                         s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
2481                     }
2482                     break;
2483                 case CANDIDATE_MB_TYPE_INTER4V:
2484                     s->mv_dir = MV_DIR_FORWARD;
2485                     s->mv_type = MV_TYPE_8X8;
2486                     s->mb_intra= 0;
2487                     for(i=0; i<4; i++){
2488                         s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
2489                         s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
2490                     }
2491                     break;
2492                 case CANDIDATE_MB_TYPE_DIRECT:
2493                     if (CONFIG_MPEG4_ENCODER) {
2494                         s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
2495                         s->mb_intra= 0;
2496                         motion_x=s->b_direct_mv_table[xy][0];
2497                         motion_y=s->b_direct_mv_table[xy][1];
2498                         ff_mpeg4_set_direct_mv(s, motion_x, motion_y);
2499                     }
2500                     break;
2501                 case CANDIDATE_MB_TYPE_DIRECT0:
2502                     if (CONFIG_MPEG4_ENCODER) {
2503                         s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
2504                         s->mb_intra= 0;
2505                         ff_mpeg4_set_direct_mv(s, 0, 0);
2506                     }
2507                     break;
2508                 case CANDIDATE_MB_TYPE_BIDIR:
2509                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2510                     s->mb_intra= 0;
2511                     s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
2512                     s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
2513                     s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
2514                     s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2515                     break;
2516                 case CANDIDATE_MB_TYPE_BACKWARD:
2517                     s->mv_dir = MV_DIR_BACKWARD;
2518                     s->mb_intra= 0;
2519                     motion_x= s->mv[1][0][0] = s->b_back_mv_table[xy][0];
2520                     motion_y= s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2521                     break;
2522                 case CANDIDATE_MB_TYPE_FORWARD:
2523                     s->mv_dir = MV_DIR_FORWARD;
2524                     s->mb_intra= 0;
2525                     motion_x= s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
2526                     motion_y= s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2527 //                    printf(" %d %d ", motion_x, motion_y);
2528                     break;
2529                 case CANDIDATE_MB_TYPE_FORWARD_I:
2530                     s->mv_dir = MV_DIR_FORWARD;
2531                     s->mv_type = MV_TYPE_FIELD;
2532                     s->mb_intra= 0;
2533                     for(i=0; i<2; i++){
2534                         j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
2535                         s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
2536                         s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
2537                     }
2538                     break;
2539                 case CANDIDATE_MB_TYPE_BACKWARD_I:
2540                     s->mv_dir = MV_DIR_BACKWARD;
2541                     s->mv_type = MV_TYPE_FIELD;
2542                     s->mb_intra= 0;
2543                     for(i=0; i<2; i++){
2544                         j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
2545                         s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
2546                         s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
2547                     }
2548                     break;
2549                 case CANDIDATE_MB_TYPE_BIDIR_I:
2550                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2551                     s->mv_type = MV_TYPE_FIELD;
2552                     s->mb_intra= 0;
2553                     for(dir=0; dir<2; dir++){
2554                         for(i=0; i<2; i++){
2555                             j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
2556                             s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
2557                             s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
2558                         }
2559                     }
2560                     break;
2561                 default:
2562                     av_log(s->avctx, AV_LOG_ERROR, "illegal MB type\n");
2563                 }
2564
2565                 encode_mb(s, motion_x, motion_y);
2566
2567                 // RAL: Update last macroblock type
2568                 s->last_mv_dir = s->mv_dir;
2569
2570                 if (CONFIG_ANY_H263_ENCODER &&
2571                     s->out_format == FMT_H263 && s->pict_type!=FF_B_TYPE)
2572                     ff_h263_update_motion_val(s);
2573
2574                 MPV_decode_mb(s, s->block);
2575             }
2576
2577             /* clean the MV table in IPS frames for direct mode in B frames */
2578             if(s->mb_intra /* && I,P,S_TYPE */){
2579                 s->p_mv_table[xy][0]=0;
2580                 s->p_mv_table[xy][1]=0;
2581             }
2582
2583             if(s->flags&CODEC_FLAG_PSNR){
2584                 int w= 16;
2585                 int h= 16;
2586
2587                 if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
2588                 if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
2589
2590                 s->current_picture.error[0] += sse(
2591                     s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16,
2592                     s->dest[0], w, h, s->linesize);
2593                 s->current_picture.error[1] += sse(
2594                     s, s->new_picture.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*chr_h,
2595                     s->dest[1], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
2596                 s->current_picture.error[2] += sse(
2597                     s, s->new_picture.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*chr_h,
2598                     s->dest[2], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
2599             }
2600             if(s->loop_filter){
2601                 if(CONFIG_ANY_H263_ENCODER && s->out_format == FMT_H263)
2602                     ff_h263_loop_filter(s);
2603             }
2604 //printf("MB %d %d bits\n", s->mb_x+s->mb_y*s->mb_stride, put_bits_count(&s->pb));
2605         }
2606     }
2607
2608     //not beautiful here but we must write it before flushing so it has to be here
2609     if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version && s->msmpeg4_version<4 && s->pict_type == FF_I_TYPE)
2610         msmpeg4_encode_ext_header(s);
2611
2612     write_slice_end(s);
2613
2614     /* Send the last GOB if RTP */
2615     if (s->avctx->rtp_callback) {
2616         int number_mb = (mb_y - s->resync_mb_y)*s->mb_width - s->resync_mb_x;
2617         pdif = put_bits_ptr(&s->pb) - s->ptr_lastgob;
2618         /* Call the RTP callback to send the last GOB */
2619         emms_c();
2620         s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, pdif, number_mb);
2621     }
2622
2623     return 0;
2624 }
2625
2626 #define MERGE(field) dst->field += src->field; src->field=0
2627 static void merge_context_after_me(MpegEncContext *dst, MpegEncContext *src){
2628     MERGE(me.scene_change_score);
2629     MERGE(me.mc_mb_var_sum_temp);
2630     MERGE(me.mb_var_sum_temp);
2631 }
2632
2633 static void merge_context_after_encode(MpegEncContext *dst, MpegEncContext *src){
2634     int i;
2635
2636     MERGE(dct_count[0]); //note, the other dct vars are not part of the context
2637     MERGE(dct_count[1]);
2638     MERGE(mv_bits);
2639     MERGE(i_tex_bits);
2640     MERGE(p_tex_bits);
2641     MERGE(i_count);
2642     MERGE(f_count);
2643     MERGE(b_count);
2644     MERGE(skip_count);
2645     MERGE(misc_bits);
2646     MERGE(error_count);
2647     MERGE(padding_bug_score);
2648     MERGE(current_picture.error[0]);
2649     MERGE(current_picture.error[1]);
2650     MERGE(current_picture.error[2]);
2651
2652     if(dst->avctx->noise_reduction){
2653         for(i=0; i<64; i++){
2654             MERGE(dct_error_sum[0][i]);
2655             MERGE(dct_error_sum[1][i]);
2656         }
2657     }
2658
2659     assert(put_bits_count(&src->pb) % 8 ==0);
2660     assert(put_bits_count(&dst->pb) % 8 ==0);
2661     ff_copy_bits(&dst->pb, src->pb.buf, put_bits_count(&src->pb));
2662     flush_put_bits(&dst->pb);
2663 }
2664
2665 static int estimate_qp(MpegEncContext *s, int dry_run){
2666     if (s->next_lambda){
2667         s->current_picture_ptr->quality=
2668         s->current_picture.quality = s->next_lambda;
2669         if(!dry_run) s->next_lambda= 0;
2670     } else if (!s->fixed_qscale) {
2671         s->current_picture_ptr->quality=
2672         s->current_picture.quality = ff_rate_estimate_qscale(s, dry_run);
2673         if (s->current_picture.quality < 0)
2674             return -1;
2675     }
2676
2677     if(s->adaptive_quant){
2678         switch(s->codec_id){
2679         case CODEC_ID_MPEG4:
2680             if (CONFIG_MPEG4_ENCODER)
2681                 ff_clean_mpeg4_qscales(s);
2682             break;
2683         case CODEC_ID_H263:
2684         case CODEC_ID_H263P:
2685         case CODEC_ID_FLV1:
2686             if (CONFIG_H263_ENCODER||CONFIG_H263P_ENCODER||CONFIG_FLV_ENCODER)
2687                 ff_clean_h263_qscales(s);
2688             break;
2689         }
2690
2691         s->lambda= s->lambda_table[0];
2692         //FIXME broken
2693     }else
2694         s->lambda= s->current_picture.quality;
2695 //printf("%d %d\n", s->avctx->global_quality, s->current_picture.quality);
2696     update_qscale(s);
2697     return 0;
2698 }
2699
2700 /* must be called before writing the header */
2701 static void set_frame_distances(MpegEncContext * s){
2702     assert(s->current_picture_ptr->pts != AV_NOPTS_VALUE);
2703     s->time= s->current_picture_ptr->pts*s->avctx->time_base.num;
2704
2705     if(s->pict_type==FF_B_TYPE){
2706         s->pb_time= s->pp_time - (s->last_non_b_time - s->time);
2707         assert(s->pb_time > 0 && s->pb_time < s->pp_time);
2708     }else{
2709         s->pp_time= s->time - s->last_non_b_time;
2710         s->last_non_b_time= s->time;
2711         assert(s->picture_number==0 || s->pp_time > 0);
2712     }
2713 }
2714
2715 static int encode_picture(MpegEncContext *s, int picture_number)
2716 {
2717     int i;
2718     int bits;
2719
2720     s->picture_number = picture_number;
2721
2722     /* Reset the average MB variance */
2723     s->me.mb_var_sum_temp    =
2724     s->me.mc_mb_var_sum_temp = 0;
2725
2726     /* we need to initialize some time vars before we can encode b-frames */
2727     // RAL: Condition added for MPEG1VIDEO
2728     if (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO || (s->h263_pred && !s->h263_msmpeg4))
2729         set_frame_distances(s);
2730     if(CONFIG_MPEG4_ENCODER && s->codec_id == CODEC_ID_MPEG4)
2731         ff_set_mpeg4_time(s);
2732
2733     s->me.scene_change_score=0;
2734
2735 //    s->lambda= s->current_picture_ptr->quality; //FIXME qscale / ... stuff for ME rate distortion
2736
2737     if(s->pict_type==FF_I_TYPE){
2738         if(s->msmpeg4_version >= 3) s->no_rounding=1;
2739         else                        s->no_rounding=0;
2740     }else if(s->pict_type!=FF_B_TYPE){
2741         if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
2742             s->no_rounding ^= 1;
2743     }
2744
2745     if(s->flags & CODEC_FLAG_PASS2){
2746         if (estimate_qp(s,1) < 0)
2747             return -1;
2748         ff_get_2pass_fcode(s);
2749     }else if(!(s->flags & CODEC_FLAG_QSCALE)){
2750         if(s->pict_type==FF_B_TYPE)
2751             s->lambda= s->last_lambda_for[s->pict_type];
2752         else
2753             s->lambda= s->last_lambda_for[s->last_non_b_pict_type];
2754         update_qscale(s);
2755     }
2756
2757     s->mb_intra=0; //for the rate distortion & bit compare functions
2758     for(i=1; i<s->avctx->thread_count; i++){
2759         ff_update_duplicate_context(s->thread_context[i], s);
2760     }
2761
2762     if(ff_init_me(s)<0)
2763         return -1;
2764
2765     /* Estimate motion for every MB */
2766     if(s->pict_type != FF_I_TYPE){
2767         s->lambda = (s->lambda * s->avctx->me_penalty_compensation + 128)>>8;
2768         s->lambda2= (s->lambda2* (int64_t)s->avctx->me_penalty_compensation + 128)>>8;
2769         if(s->pict_type != FF_B_TYPE && s->avctx->me_threshold==0){
2770             if((s->avctx->pre_me && s->last_non_b_pict_type==FF_I_TYPE) || s->avctx->pre_me==2){
2771                 s->avctx->execute(s->avctx, pre_estimate_motion_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count, sizeof(void*));
2772             }
2773         }
2774
2775         s->avctx->execute(s->avctx, estimate_motion_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count, sizeof(void*));
2776     }else /* if(s->pict_type == FF_I_TYPE) */{
2777         /* I-Frame */
2778         for(i=0; i<s->mb_stride*s->mb_height; i++)
2779             s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
2780
2781         if(!s->fixed_qscale){
2782             /* finding spatial complexity for I-frame rate control */
2783             s->avctx->execute(s->avctx, mb_var_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count, sizeof(void*));
2784         }
2785     }
2786     for(i=1; i<s->avctx->thread_count; i++){
2787         merge_context_after_me(s, s->thread_context[i]);
2788     }
2789     s->current_picture.mc_mb_var_sum= s->current_picture_ptr->mc_mb_var_sum= s->me.mc_mb_var_sum_temp;
2790     s->current_picture.   mb_var_sum= s->current_picture_ptr->   mb_var_sum= s->me.   mb_var_sum_temp;
2791     emms_c();
2792
2793     if(s->me.scene_change_score > s->avctx->scenechange_threshold && s->pict_type == FF_P_TYPE){
2794         s->pict_type= FF_I_TYPE;
2795         for(i=0; i<s->mb_stride*s->mb_height; i++)
2796             s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
2797 //printf("Scene change detected, encoding as I Frame %d %d\n", s->current_picture.mb_var_sum, s->current_picture.mc_mb_var_sum);
2798     }
2799
2800     if(!s->umvplus){
2801         if(s->pict_type==FF_P_TYPE || s->pict_type==FF_S_TYPE) {
2802             s->f_code= ff_get_best_fcode(s, s->p_mv_table, CANDIDATE_MB_TYPE_INTER);
2803
2804             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2805                 int a,b;
2806                 a= ff_get_best_fcode(s, s->p_field_mv_table[0][0], CANDIDATE_MB_TYPE_INTER_I); //FIXME field_select
2807                 b= ff_get_best_fcode(s, s->p_field_mv_table[1][1], CANDIDATE_MB_TYPE_INTER_I);
2808                 s->f_code= FFMAX3(s->f_code, a, b);
2809             }
2810
2811             ff_fix_long_p_mvs(s);
2812             ff_fix_long_mvs(s, NULL, 0, s->p_mv_table, s->f_code, CANDIDATE_MB_TYPE_INTER, 0);
2813             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2814                 int j;
2815                 for(i=0; i<2; i++){
2816                     for(j=0; j<2; j++)
2817                         ff_fix_long_mvs(s, s->p_field_select_table[i], j,
2818                                         s->p_field_mv_table[i][j], s->f_code, CANDIDATE_MB_TYPE_INTER_I, 0);
2819                 }
2820             }
2821         }
2822
2823         if(s->pict_type==FF_B_TYPE){
2824             int a, b;
2825
2826             a = ff_get_best_fcode(s, s->b_forw_mv_table, CANDIDATE_MB_TYPE_FORWARD);
2827             b = ff_get_best_fcode(s, s->b_bidir_forw_mv_table, CANDIDATE_MB_TYPE_BIDIR);
2828             s->f_code = FFMAX(a, b);
2829
2830             a = ff_get_best_fcode(s, s->b_back_mv_table, CANDIDATE_MB_TYPE_BACKWARD);
2831             b = ff_get_best_fcode(s, s->b_bidir_back_mv_table, CANDIDATE_MB_TYPE_BIDIR);
2832             s->b_code = FFMAX(a, b);
2833
2834             ff_fix_long_mvs(s, NULL, 0, s->b_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_FORWARD, 1);
2835             ff_fix_long_mvs(s, NULL, 0, s->b_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BACKWARD, 1);
2836             ff_fix_long_mvs(s, NULL, 0, s->b_bidir_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_BIDIR, 1);
2837             ff_fix_long_mvs(s, NULL, 0, s->b_bidir_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BIDIR, 1);
2838             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2839                 int dir, j;
2840                 for(dir=0; dir<2; dir++){
2841                     for(i=0; i<2; i++){
2842                         for(j=0; j<2; j++){
2843                             int type= dir ? (CANDIDATE_MB_TYPE_BACKWARD_I|CANDIDATE_MB_TYPE_BIDIR_I)
2844                                           : (CANDIDATE_MB_TYPE_FORWARD_I |CANDIDATE_MB_TYPE_BIDIR_I);
2845                             ff_fix_long_mvs(s, s->b_field_select_table[dir][i], j,
2846                                             s->b_field_mv_table[dir][i][j], dir ? s->b_code : s->f_code, type, 1);
2847                         }
2848                     }
2849                 }
2850             }
2851         }
2852     }
2853
2854     if (estimate_qp(s, 0) < 0)
2855         return -1;
2856
2857     if(s->qscale < 3 && s->max_qcoeff<=128 && s->pict_type==FF_I_TYPE && !(s->flags & CODEC_FLAG_QSCALE))
2858         s->qscale= 3; //reduce clipping problems
2859
2860     if (s->out_format == FMT_MJPEG) {
2861         /* for mjpeg, we do include qscale in the matrix */
2862         s->intra_matrix[0] = ff_mpeg1_default_intra_matrix[0];
2863         for(i=1;i<64;i++){
2864             int j= s->dsp.idct_permutation[i];
2865
2866             s->intra_matrix[j] = av_clip_uint8((ff_mpeg1_default_intra_matrix[i] * s->qscale) >> 3);
2867         }
2868         ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
2869                        s->intra_matrix, s->intra_quant_bias, 8, 8, 1);
2870         s->qscale= 8;
2871     }
2872
2873     //FIXME var duplication
2874     s->current_picture_ptr->key_frame=
2875     s->current_picture.key_frame= s->pict_type == FF_I_TYPE; //FIXME pic_ptr
2876     s->current_picture_ptr->pict_type=
2877     s->current_picture.pict_type= s->pict_type;
2878
2879     if(s->current_picture.key_frame)
2880         s->picture_in_gop_number=0;
2881
2882     s->last_bits= put_bits_count(&s->pb);
2883     switch(s->out_format) {
2884     case FMT_MJPEG:
2885         if (CONFIG_MJPEG_ENCODER)
2886             ff_mjpeg_encode_picture_header(s);
2887         break;
2888     case FMT_H261:
2889         if (CONFIG_H261_ENCODER)
2890             ff_h261_encode_picture_header(s, picture_number);
2891         break;
2892     case FMT_H263:
2893         if (CONFIG_WMV2_ENCODER && s->codec_id == CODEC_ID_WMV2)
2894             ff_wmv2_encode_picture_header(s, picture_number);
2895         else if (CONFIG_MSMPEG4_ENCODER && s->h263_msmpeg4)
2896             msmpeg4_encode_picture_header(s, picture_number);
2897         else if (CONFIG_MPEG4_ENCODER && s->h263_pred)
2898             mpeg4_encode_picture_header(s, picture_number);
2899         else if (CONFIG_RV10_ENCODER && s->codec_id == CODEC_ID_RV10)
2900             rv10_encode_picture_header(s, picture_number);
2901         else if (CONFIG_RV20_ENCODER && s->codec_id == CODEC_ID_RV20)
2902             rv20_encode_picture_header(s, picture_number);
2903         else if (CONFIG_FLV_ENCODER && s->codec_id == CODEC_ID_FLV1)
2904             ff_flv_encode_picture_header(s, picture_number);
2905         else if (CONFIG_ANY_H263_ENCODER)
2906             h263_encode_picture_header(s, picture_number);
2907         break;
2908     case FMT_MPEG1:
2909         if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
2910             mpeg1_encode_picture_header(s, picture_number);
2911         break;
2912     case FMT_H264:
2913         break;
2914     default:
2915         assert(0);
2916     }
2917     bits= put_bits_count(&s->pb);
2918     s->header_bits= bits - s->last_bits;
2919
2920     for(i=1; i<s->avctx->thread_count; i++){
2921         update_duplicate_context_after_me(s->thread_context[i], s);
2922     }
2923     s->avctx->execute(s->avctx, encode_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count, sizeof(void*));
2924     for(i=1; i<s->avctx->thread_count; i++){
2925         merge_context_after_encode(s, s->thread_context[i]);
2926     }
2927     emms_c();
2928     return 0;
2929 }
2930
2931 void  denoise_dct_c(MpegEncContext *s, DCTELEM *block){
2932     const int intra= s->mb_intra;
2933     int i;
2934
2935     s->dct_count[intra]++;
2936
2937     for(i=0; i<64; i++){
2938         int level= block[i];
2939
2940         if(level){
2941             if(level>0){
2942                 s->dct_error_sum[intra][i] += level;
2943                 level -= s->dct_offset[intra][i];
2944                 if(level<0) level=0;
2945             }else{
2946                 s->dct_error_sum[intra][i] -= level;
2947                 level += s->dct_offset[intra][i];
2948                 if(level>0) level=0;
2949             }
2950             block[i]= level;
2951         }
2952     }
2953 }
2954
2955 int dct_quantize_trellis_c(MpegEncContext *s,
2956                         DCTELEM *block, int n,
2957                         int qscale, int *overflow){
2958     const int *qmat;
2959     const uint8_t *scantable= s->intra_scantable.scantable;
2960     const uint8_t *perm_scantable= s->intra_scantable.permutated;
2961     int max=0;
2962     unsigned int threshold1, threshold2;
2963     int bias=0;
2964     int run_tab[65];
2965     int level_tab[65];
2966     int score_tab[65];
2967     int survivor[65];
2968     int survivor_count;
2969     int last_run=0;
2970     int last_level=0;
2971     int last_score= 0;
2972     int last_i;
2973     int coeff[2][64];
2974     int coeff_count[64];
2975     int qmul, qadd, start_i, last_non_zero, i, dc;
2976     const int esc_length= s->ac_esc_length;
2977     uint8_t * length;
2978     uint8_t * last_length;
2979     const int lambda= s->lambda2 >> (FF_LAMBDA_SHIFT - 6);
2980
2981     s->dsp.fdct (block);
2982
2983     if(s->dct_error_sum)
2984         s->denoise_dct(s, block);
2985     qmul= qscale*16;
2986     qadd= ((qscale-1)|1)*8;
2987
2988     if (s->mb_intra) {
2989         int q;
2990         if (!s->h263_aic) {
2991             if (n < 4)
2992                 q = s->y_dc_scale;
2993             else
2994                 q = s->c_dc_scale;
2995             q = q << 3;
2996         } else{
2997             /* For AIC we skip quant/dequant of INTRADC */
2998             q = 1 << 3;
2999             qadd=0;
3000         }
3001
3002         /* note: block[0] is assumed to be positive */
3003         block[0] = (block[0] + (q >> 1)) / q;
3004         start_i = 1;
3005         last_non_zero = 0;
3006         qmat = s->q_intra_matrix[qscale];
3007         if(s->mpeg_quant || s->out_format == FMT_MPEG1)
3008             bias= 1<<(QMAT_SHIFT-1);
3009         length     = s->intra_ac_vlc_length;
3010         last_length= s->intra_ac_vlc_last_length;
3011     } else {
3012         start_i = 0;
3013         last_non_zero = -1;
3014         qmat = s->q_inter_matrix[qscale];
3015         length     = s->inter_ac_vlc_length;
3016         last_length= s->inter_ac_vlc_last_length;
3017     }
3018     last_i= start_i;
3019
3020     threshold1= (1<<QMAT_SHIFT) - bias - 1;
3021     threshold2= (threshold1<<1);
3022
3023     for(i=63; i>=start_i; i--) {
3024         const int j = scantable[i];
3025         int level = block[j] * qmat[j];
3026
3027         if(((unsigned)(level+threshold1))>threshold2){
3028             last_non_zero = i;
3029             break;
3030         }
3031     }
3032
3033     for(i=start_i; i<=last_non_zero; i++) {
3034         const int j = scantable[i];
3035         int level = block[j] * qmat[j];
3036
3037 //        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
3038 //           || bias-level >= (1<<(QMAT_SHIFT - 3))){
3039         if(((unsigned)(level+threshold1))>threshold2){
3040             if(level>0){
3041                 level= (bias + level)>>QMAT_SHIFT;
3042                 coeff[0][i]= level;
3043                 coeff[1][i]= level-1;
3044 //                coeff[2][k]= level-2;
3045             }else{
3046                 level= (bias - level)>>QMAT_SHIFT;
3047                 coeff[0][i]= -level;
3048                 coeff[1][i]= -level+1;
3049 //                coeff[2][k]= -level+2;
3050             }
3051             coeff_count[i]= FFMIN(level, 2);
3052             assert(coeff_count[i]);
3053             max |=level;
3054         }else{
3055             coeff[0][i]= (level>>31)|1;
3056             coeff_count[i]= 1;
3057         }
3058     }
3059
3060     *overflow= s->max_qcoeff < max; //overflow might have happened
3061
3062     if(last_non_zero < start_i){
3063         memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
3064         return last_non_zero;
3065     }
3066
3067     score_tab[start_i]= 0;
3068     survivor[0]= start_i;
3069     survivor_count= 1;
3070
3071     for(i=start_i; i<=last_non_zero; i++){
3072         int level_index, j, zero_distortion;
3073         int dct_coeff= FFABS(block[ scantable[i] ]);
3074         int best_score=256*256*256*120;
3075
3076         if (   s->dsp.fdct == fdct_ifast
3077 #ifndef FAAN_POSTSCALE
3078             || s->dsp.fdct == ff_faandct
3079 #endif
3080            )
3081             dct_coeff= (dct_coeff*ff_inv_aanscales[ scantable[i] ]) >> 12;
3082         zero_distortion= dct_coeff*dct_coeff;
3083
3084         for(level_index=0; level_index < coeff_count[i]; level_index++){
3085             int distortion;
3086             int level= coeff[level_index][i];
3087             const int alevel= FFABS(level);
3088             int unquant_coeff;
3089
3090             assert(level);
3091
3092             if(s->out_format == FMT_H263){
3093                 unquant_coeff= alevel*qmul + qadd;
3094             }else{ //MPEG1
3095                 j= s->dsp.idct_permutation[ scantable[i] ]; //FIXME optimize
3096                 if(s->mb_intra){
3097                         unquant_coeff = (int)(  alevel  * qscale * s->intra_matrix[j]) >> 3;
3098                         unquant_coeff =   (unquant_coeff - 1) | 1;
3099                 }else{
3100                         unquant_coeff = (((  alevel  << 1) + 1) * qscale * ((int) s->inter_matrix[j])) >> 4;
3101                         unquant_coeff =   (unquant_coeff - 1) | 1;
3102                 }
3103                 unquant_coeff<<= 3;
3104             }
3105
3106             distortion= (unquant_coeff - dct_coeff) * (unquant_coeff - dct_coeff) - zero_distortion;
3107             level+=64;
3108             if((level&(~127)) == 0){
3109                 for(j=survivor_count-1; j>=0; j--){
3110                     int run= i - survivor[j];
3111                     int score= distortion + length[UNI_AC_ENC_INDEX(run, level)]*lambda;
3112                     score += score_tab[i-run];
3113
3114                     if(score < best_score){
3115                         best_score= score;
3116                         run_tab[i+1]= run;
3117                         level_tab[i+1]= level-64;
3118                     }
3119                 }
3120
3121                 if(s->out_format == FMT_H263){
3122                     for(j=survivor_count-1; j>=0; j--){
3123                         int run= i - survivor[j];
3124                         int score= distortion + last_length[UNI_AC_ENC_INDEX(run, level)]*lambda;
3125                         score += score_tab[i-run];
3126                         if(score < last_score){
3127                             last_score= score;
3128                             last_run= run;
3129                             last_level= level-64;
3130                             last_i= i+1;
3131                         }
3132                     }
3133                 }
3134             }else{
3135                 distortion += esc_length*lambda;
3136                 for(j=survivor_count-1; j>=0; j--){
3137                     int run= i - survivor[j];
3138                     int score= distortion + score_tab[i-run];
3139
3140                     if(score < best_score){
3141                         best_score= score;
3142                         run_tab[i+1]= run;
3143                         level_tab[i+1]= level-64;
3144                     }
3145                 }
3146
3147                 if(s->out_format == FMT_H263){
3148                   for(j=survivor_count-1; j>=0; j--){
3149                         int run= i - survivor[j];
3150                         int score= distortion + score_tab[i-run];
3151                         if(score < last_score){
3152                             last_score= score;
3153                             last_run= run;
3154                             last_level= level-64;
3155                             last_i= i+1;
3156                         }
3157                     }
3158                 }
3159             }
3160         }
3161
3162         score_tab[i+1]= best_score;
3163
3164         //Note: there is a vlc code in mpeg4 which is 1 bit shorter then another one with a shorter run and the same level
3165         if(last_non_zero <= 27){
3166             for(; survivor_count; survivor_count--){
3167                 if(score_tab[ survivor[survivor_count-1] ] <= best_score)
3168                     break;
3169             }
3170         }else{
3171             for(; survivor_count; survivor_count--){
3172                 if(score_tab[ survivor[survivor_count-1] ] <= best_score + lambda)
3173                     break;
3174             }
3175         }
3176
3177         survivor[ survivor_count++ ]= i+1;
3178     }
3179
3180     if(s->out_format != FMT_H263){
3181         last_score= 256*256*256*120;
3182         for(i= survivor[0]; i<=last_non_zero + 1; i++){
3183             int score= score_tab[i];
3184             if(i) score += lambda*2; //FIXME exacter?
3185
3186             if(score < last_score){
3187                 last_score= score;
3188                 last_i= i;
3189                 last_level= level_tab[i];
3190                 last_run= run_tab[i];
3191             }
3192         }
3193     }
3194
3195     s->coded_score[n] = last_score;
3196
3197     dc= FFABS(block[0]);
3198     last_non_zero= last_i - 1;
3199     memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
3200
3201     if(last_non_zero < start_i)
3202         return last_non_zero;
3203
3204     if(last_non_zero == 0 && start_i == 0){
3205         int best_level= 0;
3206         int best_score= dc * dc;
3207
3208         for(i=0; i<coeff_count[0]; i++){
3209             int level= coeff[i][0];
3210             int alevel= FFABS(level);
3211             int unquant_coeff, score, distortion;
3212
3213             if(s->out_format == FMT_H263){
3214                     unquant_coeff= (alevel*qmul + qadd)>>3;
3215             }else{ //MPEG1
3216                     unquant_coeff = (((  alevel  << 1) + 1) * qscale * ((int) s->inter_matrix[0])) >> 4;
3217                     unquant_coeff =   (unquant_coeff - 1) | 1;
3218             }
3219             unquant_coeff = (unquant_coeff + 4) >> 3;
3220             unquant_coeff<<= 3 + 3;
3221
3222             distortion= (unquant_coeff - dc) * (unquant_coeff - dc);
3223             level+=64;
3224             if((level&(~127)) == 0) score= distortion + last_length[UNI_AC_ENC_INDEX(0, level)]*lambda;
3225             else                    score= distortion + esc_length*lambda;
3226
3227             if(score < best_score){
3228                 best_score= score;
3229                 best_level= level - 64;
3230             }
3231         }
3232         block[0]= best_level;
3233         s->coded_score[n] = best_score - dc*dc;
3234         if(best_level == 0) return -1;
3235         else                return last_non_zero;
3236     }
3237
3238     i= last_i;
3239     assert(last_level);
3240
3241     block[ perm_scantable[last_non_zero] ]= last_level;
3242     i -= last_run + 1;
3243
3244     for(; i>start_i; i -= run_tab[i] + 1){
3245         block[ perm_scantable[i-1] ]= level_tab[i];
3246     }
3247
3248     return last_non_zero;
3249 }
3250
3251 //#define REFINE_STATS 1
3252 static int16_t basis[64][64];
3253
3254 static void build_basis(uint8_t *perm){
3255     int i, j, x, y;
3256     emms_c();
3257     for(i=0; i<8; i++){
3258         for(j=0; j<8; j++){
3259             for(y=0; y<8; y++){
3260                 for(x=0; x<8; x++){
3261                     double s= 0.25*(1<<BASIS_SHIFT);
3262                     int index= 8*i + j;
3263                     int perm_index= perm[index];
3264                     if(i==0) s*= sqrt(0.5);
3265                     if(j==0) s*= sqrt(0.5);
3266                     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)));
3267                 }
3268             }
3269         }
3270     }
3271 }
3272
3273 static int dct_quantize_refine(MpegEncContext *s, //FIXME breaks denoise?
3274                         DCTELEM *block, int16_t *weight, DCTELEM *orig,
3275                         int n, int qscale){
3276     int16_t rem[64];
3277     DECLARE_ALIGNED_16(DCTELEM, d1[64]);
3278     const uint8_t *scantable= s->intra_scantable.scantable;
3279     const uint8_t *perm_scantable= s->intra_scantable.permutated;
3280 //    unsigned int threshold1, threshold2;
3281 //    int bias=0;
3282     int run_tab[65];
3283     int prev_run=0;
3284     int prev_level=0;
3285     int qmul, qadd, start_i, last_non_zero, i, dc;
3286     uint8_t * length;
3287     uint8_t * last_length;
3288     int lambda;
3289     int rle_index, run, q = 1, sum; //q is only used when s->mb_intra is true
3290 #ifdef REFINE_STATS
3291 static int count=0;
3292 static int after_last=0;
3293 static int to_zero=0;
3294 static int from_zero=0;
3295 static int raise=0;
3296 static int lower=0;
3297 static int messed_sign=0;
3298 #endif
3299
3300     if(basis[0][0] == 0)
3301         build_basis(s->dsp.idct_permutation);
3302
3303     qmul= qscale*2;
3304     qadd= (qscale-1)|1;
3305     if (s->mb_intra) {
3306         if (!s->h263_aic) {
3307             if (n < 4)
3308                 q = s->y_dc_scale;
3309             else
3310                 q = s->c_dc_scale;
3311         } else{
3312             /* For AIC we skip quant/dequant of INTRADC */
3313             q = 1;
3314             qadd=0;
3315         }
3316         q <<= RECON_SHIFT-3;
3317         /* note: block[0] is assumed to be positive */
3318         dc= block[0]*q;
3319 //        block[0] = (block[0] + (q >> 1)) / q;
3320         start_i = 1;
3321 //        if(s->mpeg_quant || s->out_format == FMT_MPEG1)
3322 //            bias= 1<<(QMAT_SHIFT-1);
3323         length     = s->intra_ac_vlc_length;
3324         last_length= s->intra_ac_vlc_last_length;
3325     } else {
3326         dc= 0;
3327         start_i = 0;
3328         length     = s->inter_ac_vlc_length;
3329         last_length= s->inter_ac_vlc_last_length;
3330     }
3331     last_non_zero = s->block_last_index[n];
3332
3333 #ifdef REFINE_STATS
3334 {START_TIMER
3335 #endif
3336     dc += (1<<(RECON_SHIFT-1));
3337     for(i=0; i<64; i++){
3338         rem[i]= dc - (orig[i]<<RECON_SHIFT); //FIXME  use orig dirrectly instead of copying to rem[]
3339     }
3340 #ifdef REFINE_STATS
3341 STOP_TIMER("memset rem[]")}
3342 #endif
3343     sum=0;
3344     for(i=0; i<64; i++){
3345         int one= 36;
3346         int qns=4;
3347         int w;
3348
3349         w= FFABS(weight[i]) + qns*one;
3350         w= 15 + (48*qns*one + w/2)/w; // 16 .. 63
3351
3352         weight[i] = w;
3353 //        w=weight[i] = (63*qns + (w/2)) / w;
3354
3355         assert(w>0);
3356         assert(w<(1<<6));
3357         sum += w*w;
3358     }
3359     lambda= sum*(uint64_t)s->lambda2 >> (FF_LAMBDA_SHIFT - 6 + 6 + 6 + 6);
3360 #ifdef REFINE_STATS
3361 {START_TIMER
3362 #endif
3363     run=0;
3364     rle_index=0;
3365     for(i=start_i; i<=last_non_zero; i++){
3366         int j= perm_scantable[i];
3367         const int level= block[j];
3368         int coeff;
3369
3370         if(level){
3371             if(level<0) coeff= qmul*level - qadd;
3372             else        coeff= qmul*level + qadd;
3373             run_tab[rle_index++]=run;
3374             run=0;
3375
3376             s->dsp.add_8x8basis(rem, basis[j], coeff);
3377      &nb