h263p encoder: add 'umv' private option.
[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 Libav.
9  *
10  * Libav 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  * Libav 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 Libav; 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
27  * The simplest mpeg encoder (well, it was the simplest!).
28  */
29
30 #include "libavutil/intmath.h"
31 #include "libavutil/mathematics.h"
32 #include "libavutil/opt.h"
33 #include "avcodec.h"
34 #include "dsputil.h"
35 #include "mpegvideo.h"
36 #include "mpegvideo_common.h"
37 #include "h263.h"
38 #include "mjpegenc.h"
39 #include "msmpeg4.h"
40 #include "faandct.h"
41 #include "thread.h"
42 #include "aandcttab.h"
43 #include "flv.h"
44 #include "mpeg4video.h"
45 #include "internal.h"
46 #include <limits.h>
47
48 //#undef NDEBUG
49 //#include <assert.h>
50
51 static int encode_picture(MpegEncContext *s, int picture_number);
52 static int dct_quantize_refine(MpegEncContext *s, DCTELEM *block, int16_t *weight, DCTELEM *orig, int n, int qscale);
53 static int sse_mb(MpegEncContext *s);
54 static void denoise_dct_c(MpegEncContext *s, DCTELEM *block);
55 static int dct_quantize_trellis_c(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow);
56
57 /* enable all paranoid tests for rounding, overflows, etc... */
58 //#define PARANOID
59
60 //#define DEBUG
61
62 static uint8_t default_mv_penalty[MAX_FCODE+1][MAX_MV*2+1];
63 static uint8_t default_fcode_tab[MAX_MV*2+1];
64
65 void ff_convert_matrix(DSPContext *dsp, int (*qmat)[64], uint16_t (*qmat16)[2][64],
66                            const uint16_t *quant_matrix, int bias, int qmin, int qmax, int intra)
67 {
68     int qscale;
69     int shift=0;
70
71     for(qscale=qmin; qscale<=qmax; qscale++){
72         int i;
73         if (dsp->fdct == ff_jpeg_fdct_islow_8 ||
74             dsp->fdct == ff_jpeg_fdct_islow_10
75 #ifdef FAAN_POSTSCALE
76             || dsp->fdct == ff_faandct
77 #endif
78             ) {
79             for(i=0;i<64;i++) {
80                 const int j= dsp->idct_permutation[i];
81                 /* 16 <= qscale * quant_matrix[i] <= 7905 */
82                 /* 19952             <= ff_aanscales[i] * qscale * quant_matrix[i]               <= 249205026 */
83                 /* (1 << 36) / 19952 >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= (1 << 36) / 249205026 */
84                 /* 3444240           >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= 275 */
85
86                 qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) /
87                                 (qscale * quant_matrix[j]));
88             }
89         } else if (dsp->fdct == fdct_ifast
90 #ifndef FAAN_POSTSCALE
91                    || dsp->fdct == ff_faandct
92 #endif
93                    ) {
94             for(i=0;i<64;i++) {
95                 const int j= dsp->idct_permutation[i];
96                 /* 16 <= qscale * quant_matrix[i] <= 7905 */
97                 /* 19952             <= ff_aanscales[i] * qscale * quant_matrix[i]               <= 249205026 */
98                 /* (1 << 36) / 19952 >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */
99                 /* 3444240           >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= 275 */
100
101                 qmat[qscale][i] = (int)((UINT64_C(1) << (QMAT_SHIFT + 14)) /
102                                 (ff_aanscales[i] * qscale * quant_matrix[j]));
103             }
104         } else {
105             for(i=0;i<64;i++) {
106                 const int j= dsp->idct_permutation[i];
107                 /* We can safely suppose that 16 <= quant_matrix[i] <= 255
108                    So 16           <= qscale * quant_matrix[i]             <= 7905
109                    so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905
110                    so 32768        >= (1<<19) / (qscale * quant_matrix[i]) >= 67
111                 */
112                 qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) / (qscale * quant_matrix[j]));
113 //                qmat  [qscale][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]);
114                 qmat16[qscale][0][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[j]);
115
116                 if(qmat16[qscale][0][i]==0 || qmat16[qscale][0][i]==128*256) qmat16[qscale][0][i]=128*256-1;
117                 qmat16[qscale][1][i]= ROUNDED_DIV(bias<<(16-QUANT_BIAS_SHIFT), qmat16[qscale][0][i]);
118             }
119         }
120
121         for(i=intra; i<64; i++){
122             int64_t max= 8191;
123             if (dsp->fdct == fdct_ifast
124 #ifndef FAAN_POSTSCALE
125                    || dsp->fdct == ff_faandct
126 #endif
127                    ) {
128                 max = (8191LL*ff_aanscales[i]) >> 14;
129             }
130             while(((max * qmat[qscale][i]) >> shift) > INT_MAX){
131                 shift++;
132             }
133         }
134     }
135     if(shift){
136         av_log(NULL, AV_LOG_INFO, "Warning, QMAT_SHIFT is larger than %d, overflows possible\n", QMAT_SHIFT - shift);
137     }
138 }
139
140 static inline void update_qscale(MpegEncContext *s){
141     s->qscale= (s->lambda*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7);
142     s->qscale= av_clip(s->qscale, s->avctx->qmin, s->avctx->qmax);
143
144     s->lambda2= (s->lambda*s->lambda + FF_LAMBDA_SCALE/2) >> FF_LAMBDA_SHIFT;
145 }
146
147 void ff_write_quant_matrix(PutBitContext *pb, uint16_t *matrix){
148     int i;
149
150     if(matrix){
151         put_bits(pb, 1, 1);
152         for(i=0;i<64;i++) {
153             put_bits(pb, 8, matrix[ ff_zigzag_direct[i] ]);
154         }
155     }else
156         put_bits(pb, 1, 0);
157 }
158
159 /**
160  * init s->current_picture.qscale_table from s->lambda_table
161  */
162 void ff_init_qscale_tab(MpegEncContext *s){
163     int8_t * const qscale_table = s->current_picture.f.qscale_table;
164     int i;
165
166     for(i=0; i<s->mb_num; i++){
167         unsigned int lam= s->lambda_table[ s->mb_index2xy[i] ];
168         int qp= (lam*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7);
169         qscale_table[ s->mb_index2xy[i] ]= av_clip(qp, s->avctx->qmin, s->avctx->qmax);
170     }
171 }
172
173 static void copy_picture_attributes(MpegEncContext *s, AVFrame *dst, AVFrame *src){
174     int i;
175
176     dst->pict_type              = src->pict_type;
177     dst->quality                = src->quality;
178     dst->coded_picture_number   = src->coded_picture_number;
179     dst->display_picture_number = src->display_picture_number;
180 //    dst->reference              = src->reference;
181     dst->pts                    = src->pts;
182     dst->interlaced_frame       = src->interlaced_frame;
183     dst->top_field_first        = src->top_field_first;
184
185     if(s->avctx->me_threshold){
186         if(!src->motion_val[0])
187             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_val not set!\n");
188         if(!src->mb_type)
189             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.mb_type not set!\n");
190         if(!src->ref_index[0])
191             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.ref_index not set!\n");
192         if(src->motion_subsample_log2 != dst->motion_subsample_log2)
193             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_subsample_log2 doesn't match! (%d!=%d)\n",
194             src->motion_subsample_log2, dst->motion_subsample_log2);
195
196         memcpy(dst->mb_type, src->mb_type, s->mb_stride * s->mb_height * sizeof(dst->mb_type[0]));
197
198         for(i=0; i<2; i++){
199             int stride= ((16*s->mb_width )>>src->motion_subsample_log2) + 1;
200             int height= ((16*s->mb_height)>>src->motion_subsample_log2);
201
202             if(src->motion_val[i] && src->motion_val[i] != dst->motion_val[i]){
203                 memcpy(dst->motion_val[i], src->motion_val[i], 2*stride*height*sizeof(int16_t));
204             }
205             if(src->ref_index[i] && src->ref_index[i] != dst->ref_index[i]){
206                 memcpy(dst->ref_index[i], src->ref_index[i], s->mb_stride*4*s->mb_height*sizeof(int8_t));
207             }
208         }
209     }
210 }
211
212 static void update_duplicate_context_after_me(MpegEncContext *dst, MpegEncContext *src){
213 #define COPY(a) dst->a= src->a
214     COPY(pict_type);
215     COPY(current_picture);
216     COPY(f_code);
217     COPY(b_code);
218     COPY(qscale);
219     COPY(lambda);
220     COPY(lambda2);
221     COPY(picture_in_gop_number);
222     COPY(gop_picture_number);
223     COPY(frame_pred_frame_dct); //FIXME don't set in encode_header
224     COPY(progressive_frame); //FIXME don't set in encode_header
225     COPY(partitioned_frame); //FIXME don't set in encode_header
226 #undef COPY
227 }
228
229 /**
230  * sets the given MpegEncContext to defaults for encoding.
231  * the changed fields will not depend upon the prior state of the MpegEncContext.
232  */
233 static void MPV_encode_defaults(MpegEncContext *s){
234     int i;
235     MPV_common_defaults(s);
236
237     for(i=-16; i<16; i++){
238         default_fcode_tab[i + MAX_MV]= 1;
239     }
240     s->me.mv_penalty= default_mv_penalty;
241     s->fcode_tab= default_fcode_tab;
242 }
243
244 /* init video encoder */
245 av_cold int MPV_encode_init(AVCodecContext *avctx)
246 {
247     MpegEncContext *s = avctx->priv_data;
248     int i;
249     int chroma_h_shift, chroma_v_shift;
250
251     MPV_encode_defaults(s);
252
253     switch (avctx->codec_id) {
254     case CODEC_ID_MPEG2VIDEO:
255         if(avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P){
256             av_log(avctx, AV_LOG_ERROR, "only YUV420 and YUV422 are supported\n");
257             return -1;
258         }
259         break;
260     case CODEC_ID_LJPEG:
261         if(avctx->pix_fmt != PIX_FMT_YUVJ420P && avctx->pix_fmt != PIX_FMT_YUVJ422P && avctx->pix_fmt != PIX_FMT_YUVJ444P && avctx->pix_fmt != PIX_FMT_BGRA &&
262            ((avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P && avctx->pix_fmt != PIX_FMT_YUV444P) || avctx->strict_std_compliance>FF_COMPLIANCE_UNOFFICIAL)){
263             av_log(avctx, AV_LOG_ERROR, "colorspace not supported in LJPEG\n");
264             return -1;
265         }
266         break;
267     case CODEC_ID_MJPEG:
268         if(avctx->pix_fmt != PIX_FMT_YUVJ420P && avctx->pix_fmt != PIX_FMT_YUVJ422P &&
269            ((avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P) || avctx->strict_std_compliance>FF_COMPLIANCE_UNOFFICIAL)){
270             av_log(avctx, AV_LOG_ERROR, "colorspace not supported in jpeg\n");
271             return -1;
272         }
273         break;
274     default:
275         if(avctx->pix_fmt != PIX_FMT_YUV420P){
276             av_log(avctx, AV_LOG_ERROR, "only YUV420 is supported\n");
277             return -1;
278         }
279     }
280
281     switch (avctx->pix_fmt) {
282     case PIX_FMT_YUVJ422P:
283     case PIX_FMT_YUV422P:
284         s->chroma_format = CHROMA_422;
285         break;
286     case PIX_FMT_YUVJ420P:
287     case PIX_FMT_YUV420P:
288     default:
289         s->chroma_format = CHROMA_420;
290         break;
291     }
292
293     s->bit_rate = avctx->bit_rate;
294     s->width = avctx->width;
295     s->height = avctx->height;
296     if(avctx->gop_size > 600 && avctx->strict_std_compliance>FF_COMPLIANCE_EXPERIMENTAL){
297         av_log(avctx, AV_LOG_ERROR, "Warning keyframe interval too large! reducing it ...\n");
298         avctx->gop_size=600;
299     }
300     s->gop_size = avctx->gop_size;
301     s->avctx = avctx;
302     s->flags= avctx->flags;
303     s->flags2= avctx->flags2;
304     s->max_b_frames= avctx->max_b_frames;
305     s->codec_id= avctx->codec->id;
306     s->luma_elim_threshold  = avctx->luma_elim_threshold;
307     s->chroma_elim_threshold= avctx->chroma_elim_threshold;
308     s->strict_std_compliance= avctx->strict_std_compliance;
309 #if FF_API_MPEGVIDEO_GLOBAL_OPTS
310     if (avctx->flags & CODEC_FLAG_PART)
311         s->data_partitioning = 1;
312 #endif
313     s->quarter_sample= (avctx->flags & CODEC_FLAG_QPEL)!=0;
314     s->mpeg_quant= avctx->mpeg_quant;
315     s->rtp_mode= !!avctx->rtp_payload_size;
316     s->intra_dc_precision= avctx->intra_dc_precision;
317     s->user_specified_pts = AV_NOPTS_VALUE;
318
319     if (s->gop_size <= 1) {
320         s->intra_only = 1;
321         s->gop_size = 12;
322     } else {
323         s->intra_only = 0;
324     }
325
326     s->me_method = avctx->me_method;
327
328     /* Fixed QSCALE */
329     s->fixed_qscale = !!(avctx->flags & CODEC_FLAG_QSCALE);
330
331     s->adaptive_quant= (   s->avctx->lumi_masking
332                         || s->avctx->dark_masking
333                         || s->avctx->temporal_cplx_masking
334                         || s->avctx->spatial_cplx_masking
335                         || s->avctx->p_masking
336                         || s->avctx->border_masking
337                         || (s->flags&CODEC_FLAG_QP_RD))
338                        && !s->fixed_qscale;
339
340     s->obmc= !!(s->flags & CODEC_FLAG_OBMC);
341     s->loop_filter= !!(s->flags & CODEC_FLAG_LOOP_FILTER);
342 #if FF_API_MPEGVIDEO_GLOBAL_OPTS
343     s->alternate_scan= !!(s->flags & CODEC_FLAG_ALT_SCAN);
344     s->intra_vlc_format= !!(s->flags2 & CODEC_FLAG2_INTRA_VLC);
345     s->q_scale_type= !!(s->flags2 & CODEC_FLAG2_NON_LINEAR_QUANT);
346 #endif
347
348     if(avctx->rc_max_rate && !avctx->rc_buffer_size){
349         av_log(avctx, AV_LOG_ERROR, "a vbv buffer size is needed, for encoding with a maximum bitrate\n");
350         return -1;
351     }
352
353     if(avctx->rc_min_rate && avctx->rc_max_rate != avctx->rc_min_rate){
354         av_log(avctx, AV_LOG_INFO, "Warning min_rate > 0 but min_rate != max_rate isn't recommended!\n");
355     }
356
357     if(avctx->rc_min_rate && avctx->rc_min_rate > avctx->bit_rate){
358         av_log(avctx, AV_LOG_ERROR, "bitrate below min bitrate\n");
359         return -1;
360     }
361
362     if(avctx->rc_max_rate && avctx->rc_max_rate < avctx->bit_rate){
363         av_log(avctx, AV_LOG_INFO, "bitrate above max bitrate\n");
364         return -1;
365     }
366
367     if(avctx->rc_max_rate && avctx->rc_max_rate == avctx->bit_rate && avctx->rc_max_rate != avctx->rc_min_rate){
368         av_log(avctx, AV_LOG_INFO, "impossible bitrate constraints, this will fail\n");
369     }
370
371     if(avctx->rc_buffer_size && avctx->bit_rate*(int64_t)avctx->time_base.num > avctx->rc_buffer_size * (int64_t)avctx->time_base.den){
372         av_log(avctx, AV_LOG_ERROR, "VBV buffer too small for bitrate\n");
373         return -1;
374     }
375
376     if(!s->fixed_qscale && avctx->bit_rate*av_q2d(avctx->time_base) > avctx->bit_rate_tolerance){
377         av_log(avctx, AV_LOG_ERROR, "bitrate tolerance too small for bitrate\n");
378         return -1;
379     }
380
381     if(   s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate
382        && (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO)
383        && 90000LL * (avctx->rc_buffer_size-1) > s->avctx->rc_max_rate*0xFFFFLL){
384
385         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");
386     }
387
388     if((s->flags & CODEC_FLAG_4MV) && s->codec_id != CODEC_ID_MPEG4
389        && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P && s->codec_id != CODEC_ID_FLV1){
390         av_log(avctx, AV_LOG_ERROR, "4MV not supported by codec\n");
391         return -1;
392     }
393
394     if(s->obmc && s->avctx->mb_decision != FF_MB_DECISION_SIMPLE){
395         av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with simple mb decision\n");
396         return -1;
397     }
398
399     if(s->obmc && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P){
400         av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with H263(+)\n");
401         return -1;
402     }
403
404     if(s->quarter_sample && s->codec_id != CODEC_ID_MPEG4){
405         av_log(avctx, AV_LOG_ERROR, "qpel not supported by codec\n");
406         return -1;
407     }
408
409 #if FF_API_MPEGVIDEO_GLOBAL_OPTS
410     if(s->data_partitioning && s->codec_id != CODEC_ID_MPEG4){
411         av_log(avctx, AV_LOG_ERROR, "data partitioning not supported by codec\n");
412         return -1;
413     }
414 #endif
415
416     if(s->max_b_frames && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO){
417         av_log(avctx, AV_LOG_ERROR, "b frames not supported by codec\n");
418         return -1;
419     }
420
421     if ((s->codec_id == CODEC_ID_MPEG4 || s->codec_id == CODEC_ID_H263 ||
422          s->codec_id == CODEC_ID_H263P) &&
423         (avctx->sample_aspect_ratio.num > 255 || avctx->sample_aspect_ratio.den > 255)) {
424         av_log(avctx, AV_LOG_ERROR, "Invalid pixel aspect ratio %i/%i, limit is 255/255\n",
425                avctx->sample_aspect_ratio.num, avctx->sample_aspect_ratio.den);
426         return -1;
427     }
428
429     if((s->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN))
430        && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG2VIDEO){
431         av_log(avctx, AV_LOG_ERROR, "interlacing not supported by codec\n");
432         return -1;
433     }
434
435     if(s->mpeg_quant && s->codec_id != CODEC_ID_MPEG4){ //FIXME mpeg2 uses that too
436         av_log(avctx, AV_LOG_ERROR, "mpeg2 style quantization not supported by codec\n");
437         return -1;
438     }
439
440     if((s->flags & CODEC_FLAG_CBP_RD) && !avctx->trellis){
441         av_log(avctx, AV_LOG_ERROR, "CBP RD needs trellis quant\n");
442         return -1;
443     }
444
445     if((s->flags & CODEC_FLAG_QP_RD) && s->avctx->mb_decision != FF_MB_DECISION_RD){
446         av_log(avctx, AV_LOG_ERROR, "QP RD needs mbd=2\n");
447         return -1;
448     }
449
450     if(s->avctx->scenechange_threshold < 1000000000 && (s->flags & CODEC_FLAG_CLOSED_GOP)){
451         av_log(avctx, AV_LOG_ERROR, "closed gop with scene change detection are not supported yet, set threshold to 1000000000\n");
452         return -1;
453     }
454
455     if((s->flags2 & CODEC_FLAG2_INTRA_VLC) && s->codec_id != CODEC_ID_MPEG2VIDEO){
456         av_log(avctx, AV_LOG_ERROR, "intra vlc table not supported by codec\n");
457         return -1;
458     }
459
460     if(s->flags & CODEC_FLAG_LOW_DELAY){
461         if (s->codec_id != CODEC_ID_MPEG2VIDEO){
462             av_log(avctx, AV_LOG_ERROR, "low delay forcing is only available for mpeg2\n");
463             return -1;
464         }
465         if (s->max_b_frames != 0){
466             av_log(avctx, AV_LOG_ERROR, "b frames cannot be used with low delay\n");
467             return -1;
468         }
469     }
470
471     if(s->q_scale_type == 1){
472 #if FF_API_MPEGVIDEO_GLOBAL_OPTS
473         if(s->codec_id != CODEC_ID_MPEG2VIDEO){
474             av_log(avctx, AV_LOG_ERROR, "non linear quant is only available for mpeg2\n");
475             return -1;
476         }
477 #endif
478         if(avctx->qmax > 12){
479             av_log(avctx, AV_LOG_ERROR, "non linear quant only supports qmax <= 12 currently\n");
480             return -1;
481         }
482     }
483
484     if(s->avctx->thread_count > 1 && s->codec_id != CODEC_ID_MPEG4
485        && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO
486        && (s->codec_id != CODEC_ID_H263P || !(s->flags & CODEC_FLAG_H263P_SLICE_STRUCT))){
487         av_log(avctx, AV_LOG_ERROR, "multi threaded encoding not supported by codec\n");
488         return -1;
489     }
490
491     if(s->avctx->thread_count < 1){
492         av_log(avctx, AV_LOG_ERROR, "automatic thread number detection not supported by codec, patch welcome\n");
493         return -1;
494     }
495
496     if(s->avctx->thread_count > 1)
497         s->rtp_mode= 1;
498
499     if(!avctx->time_base.den || !avctx->time_base.num){
500         av_log(avctx, AV_LOG_ERROR, "framerate not set\n");
501         return -1;
502     }
503
504     i= (INT_MAX/2+128)>>8;
505     if(avctx->me_threshold >= i){
506         av_log(avctx, AV_LOG_ERROR, "me_threshold too large, max is %d\n", i - 1);
507         return -1;
508     }
509     if(avctx->mb_threshold >= i){
510         av_log(avctx, AV_LOG_ERROR, "mb_threshold too large, max is %d\n", i - 1);
511         return -1;
512     }
513
514     if(avctx->b_frame_strategy && (avctx->flags&CODEC_FLAG_PASS2)){
515         av_log(avctx, AV_LOG_INFO, "notice: b_frame_strategy only affects the first pass\n");
516         avctx->b_frame_strategy = 0;
517     }
518
519     i= av_gcd(avctx->time_base.den, avctx->time_base.num);
520     if(i > 1){
521         av_log(avctx, AV_LOG_INFO, "removing common factors from framerate\n");
522         avctx->time_base.den /= i;
523         avctx->time_base.num /= i;
524 //        return -1;
525     }
526
527     if(s->mpeg_quant || s->codec_id==CODEC_ID_MPEG1VIDEO || s->codec_id==CODEC_ID_MPEG2VIDEO || s->codec_id==CODEC_ID_MJPEG){
528         s->intra_quant_bias= 3<<(QUANT_BIAS_SHIFT-3); //(a + x*3/8)/x
529         s->inter_quant_bias= 0;
530     }else{
531         s->intra_quant_bias=0;
532         s->inter_quant_bias=-(1<<(QUANT_BIAS_SHIFT-2)); //(a - x/4)/x
533     }
534
535     if(avctx->intra_quant_bias != FF_DEFAULT_QUANT_BIAS)
536         s->intra_quant_bias= avctx->intra_quant_bias;
537     if(avctx->inter_quant_bias != FF_DEFAULT_QUANT_BIAS)
538         s->inter_quant_bias= avctx->inter_quant_bias;
539
540     avcodec_get_chroma_sub_sample(avctx->pix_fmt, &chroma_h_shift, &chroma_v_shift);
541
542     if(avctx->codec_id == CODEC_ID_MPEG4 && s->avctx->time_base.den > (1<<16)-1){
543         av_log(avctx, AV_LOG_ERROR, "timebase %d/%d not supported by MPEG 4 standard, "
544                "the maximum admitted value for the timebase denominator is %d\n",
545                s->avctx->time_base.num, s->avctx->time_base.den, (1<<16)-1);
546         return -1;
547     }
548     s->time_increment_bits = av_log2(s->avctx->time_base.den - 1) + 1;
549
550     switch(avctx->codec->id) {
551     case CODEC_ID_MPEG1VIDEO:
552         s->out_format = FMT_MPEG1;
553         s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY);
554         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
555         break;
556     case CODEC_ID_MPEG2VIDEO:
557         s->out_format = FMT_MPEG1;
558         s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY);
559         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
560         s->rtp_mode= 1;
561         break;
562     case CODEC_ID_LJPEG:
563     case CODEC_ID_MJPEG:
564         s->out_format = FMT_MJPEG;
565         s->intra_only = 1; /* force intra only for jpeg */
566         if(avctx->codec->id == CODEC_ID_LJPEG && avctx->pix_fmt == PIX_FMT_BGRA){
567             s->mjpeg_vsample[0] = s->mjpeg_hsample[0] =
568             s->mjpeg_vsample[1] = s->mjpeg_hsample[1] =
569             s->mjpeg_vsample[2] = s->mjpeg_hsample[2] = 1;
570         }else{
571             s->mjpeg_vsample[0] = 2;
572             s->mjpeg_vsample[1] = 2>>chroma_v_shift;
573             s->mjpeg_vsample[2] = 2>>chroma_v_shift;
574             s->mjpeg_hsample[0] = 2;
575             s->mjpeg_hsample[1] = 2>>chroma_h_shift;
576             s->mjpeg_hsample[2] = 2>>chroma_h_shift;
577         }
578         if (!(CONFIG_MJPEG_ENCODER || CONFIG_LJPEG_ENCODER)
579             || ff_mjpeg_encode_init(s) < 0)
580             return -1;
581         avctx->delay=0;
582         s->low_delay=1;
583         break;
584     case CODEC_ID_H261:
585         if (!CONFIG_H261_ENCODER)  return -1;
586         if (ff_h261_get_picture_format(s->width, s->height) < 0) {
587             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);
588             return -1;
589         }
590         s->out_format = FMT_H261;
591         avctx->delay=0;
592         s->low_delay=1;
593         break;
594     case CODEC_ID_H263:
595         if (!CONFIG_H263_ENCODER)  return -1;
596         if (ff_match_2uint16(h263_format, FF_ARRAY_ELEMS(h263_format), s->width, s->height) == 8) {
597             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);
598             return -1;
599         }
600         s->out_format = FMT_H263;
601         s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
602         avctx->delay=0;
603         s->low_delay=1;
604         break;
605     case CODEC_ID_H263P:
606         s->out_format = FMT_H263;
607         s->h263_plus = 1;
608         /* Fx */
609 #if FF_API_MPEGVIDEO_GLOBAL_OPTS
610         if (avctx->flags & CODEC_FLAG_H263P_UMV)
611             s->umvplus = 1;
612 #endif
613         s->h263_aic= (avctx->flags & CODEC_FLAG_AC_PRED) ? 1:0;
614         s->modified_quant= s->h263_aic;
615         s->alt_inter_vlc= (avctx->flags & CODEC_FLAG_H263P_AIV) ? 1:0;
616         s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
617         s->loop_filter= (avctx->flags & CODEC_FLAG_LOOP_FILTER) ? 1:0;
618         s->unrestricted_mv= s->obmc || s->loop_filter || s->umvplus;
619         s->h263_slice_structured= (s->flags & CODEC_FLAG_H263P_SLICE_STRUCT) ? 1:0;
620
621         /* /Fx */
622         /* These are just to be sure */
623         avctx->delay=0;
624         s->low_delay=1;
625         break;
626     case CODEC_ID_FLV1:
627         s->out_format = FMT_H263;
628         s->h263_flv = 2; /* format = 1; 11-bit codes */
629         s->unrestricted_mv = 1;
630         s->rtp_mode=0; /* don't allow GOB */
631         avctx->delay=0;
632         s->low_delay=1;
633         break;
634     case CODEC_ID_RV10:
635         s->out_format = FMT_H263;
636         avctx->delay=0;
637         s->low_delay=1;
638         break;
639     case CODEC_ID_RV20:
640         s->out_format = FMT_H263;
641         avctx->delay=0;
642         s->low_delay=1;
643         s->modified_quant=1;
644         s->h263_aic=1;
645         s->h263_plus=1;
646         s->loop_filter=1;
647         s->unrestricted_mv= 0;
648         break;
649     case CODEC_ID_MPEG4:
650         s->out_format = FMT_H263;
651         s->h263_pred = 1;
652         s->unrestricted_mv = 1;
653         s->low_delay= s->max_b_frames ? 0 : 1;
654         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
655         break;
656     case CODEC_ID_MSMPEG4V2:
657         s->out_format = FMT_H263;
658         s->h263_pred = 1;
659         s->unrestricted_mv = 1;
660         s->msmpeg4_version= 2;
661         avctx->delay=0;
662         s->low_delay=1;
663         break;
664     case CODEC_ID_MSMPEG4V3:
665         s->out_format = FMT_H263;
666         s->h263_pred = 1;
667         s->unrestricted_mv = 1;
668         s->msmpeg4_version= 3;
669         s->flipflop_rounding=1;
670         avctx->delay=0;
671         s->low_delay=1;
672         break;
673     case CODEC_ID_WMV1:
674         s->out_format = FMT_H263;
675         s->h263_pred = 1;
676         s->unrestricted_mv = 1;
677         s->msmpeg4_version= 4;
678         s->flipflop_rounding=1;
679         avctx->delay=0;
680         s->low_delay=1;
681         break;
682     case CODEC_ID_WMV2:
683         s->out_format = FMT_H263;
684         s->h263_pred = 1;
685         s->unrestricted_mv = 1;
686         s->msmpeg4_version= 5;
687         s->flipflop_rounding=1;
688         avctx->delay=0;
689         s->low_delay=1;
690         break;
691     default:
692         return -1;
693     }
694
695     avctx->has_b_frames= !s->low_delay;
696
697     s->encoding = 1;
698
699     s->progressive_frame=
700     s->progressive_sequence= !(avctx->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN));
701
702     /* init */
703     if (MPV_common_init(s) < 0)
704         return -1;
705
706     if(!s->dct_quantize)
707         s->dct_quantize = dct_quantize_c;
708     if(!s->denoise_dct)
709         s->denoise_dct = denoise_dct_c;
710     s->fast_dct_quantize = s->dct_quantize;
711     if(avctx->trellis)
712         s->dct_quantize = dct_quantize_trellis_c;
713
714     if((CONFIG_H263P_ENCODER || CONFIG_RV20_ENCODER) && s->modified_quant)
715         s->chroma_qscale_table= ff_h263_chroma_qscale_table;
716
717     s->quant_precision=5;
718
719     ff_set_cmp(&s->dsp, s->dsp.ildct_cmp, s->avctx->ildct_cmp);
720     ff_set_cmp(&s->dsp, s->dsp.frame_skip_cmp, s->avctx->frame_skip_cmp);
721
722     if (CONFIG_H261_ENCODER && s->out_format == FMT_H261)
723         ff_h261_encode_init(s);
724     if (CONFIG_H263_ENCODER && s->out_format == FMT_H263)
725         h263_encode_init(s);
726     if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version)
727         ff_msmpeg4_encode_init(s);
728     if ((CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
729         && s->out_format == FMT_MPEG1)
730         ff_mpeg1_encode_init(s);
731
732     /* init q matrix */
733     for(i=0;i<64;i++) {
734         int j= s->dsp.idct_permutation[i];
735         if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4 && s->mpeg_quant){
736             s->intra_matrix[j] = ff_mpeg4_default_intra_matrix[i];
737             s->inter_matrix[j] = ff_mpeg4_default_non_intra_matrix[i];
738         }else if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
739             s->intra_matrix[j] =
740             s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
741         }else
742         { /* mpeg1/2 */
743             s->intra_matrix[j] = ff_mpeg1_default_intra_matrix[i];
744             s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
745         }
746         if(s->avctx->intra_matrix)
747             s->intra_matrix[j] = s->avctx->intra_matrix[i];
748         if(s->avctx->inter_matrix)
749             s->inter_matrix[j] = s->avctx->inter_matrix[i];
750     }
751
752     /* precompute matrix */
753     /* for mjpeg, we do include qscale in the matrix */
754     if (s->out_format != FMT_MJPEG) {
755         ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
756                        s->intra_matrix, s->intra_quant_bias, avctx->qmin, 31, 1);
757         ff_convert_matrix(&s->dsp, s->q_inter_matrix, s->q_inter_matrix16,
758                        s->inter_matrix, s->inter_quant_bias, avctx->qmin, 31, 0);
759     }
760
761     if(ff_rate_control_init(s) < 0)
762         return -1;
763
764     return 0;
765 }
766
767 av_cold int MPV_encode_end(AVCodecContext *avctx)
768 {
769     MpegEncContext *s = avctx->priv_data;
770
771     ff_rate_control_uninit(s);
772
773     MPV_common_end(s);
774     if ((CONFIG_MJPEG_ENCODER || CONFIG_LJPEG_ENCODER) && s->out_format == FMT_MJPEG)
775         ff_mjpeg_encode_close(s);
776
777     av_freep(&avctx->extradata);
778
779     return 0;
780 }
781
782 static int get_sae(uint8_t *src, int ref, int stride){
783     int x,y;
784     int acc=0;
785
786     for(y=0; y<16; y++){
787         for(x=0; x<16; x++){
788             acc+= FFABS(src[x+y*stride] - ref);
789         }
790     }
791
792     return acc;
793 }
794
795 static int get_intra_count(MpegEncContext *s, uint8_t *src, uint8_t *ref, int stride){
796     int x, y, w, h;
797     int acc=0;
798
799     w= s->width &~15;
800     h= s->height&~15;
801
802     for(y=0; y<h; y+=16){
803         for(x=0; x<w; x+=16){
804             int offset= x + y*stride;
805             int sad = s->dsp.sad[0](NULL, src + offset, ref + offset, stride, 16);
806             int mean= (s->dsp.pix_sum(src + offset, stride) + 128)>>8;
807             int sae = get_sae(src + offset, mean, stride);
808
809             acc+= sae + 500 < sad;
810         }
811     }
812     return acc;
813 }
814
815
816 static int load_input_picture(MpegEncContext *s, AVFrame *pic_arg){
817     AVFrame *pic=NULL;
818     int64_t pts;
819     int i;
820     const int encoding_delay= s->max_b_frames;
821     int direct=1;
822
823     if(pic_arg){
824         pts= pic_arg->pts;
825         pic_arg->display_picture_number= s->input_picture_number++;
826
827         if(pts != AV_NOPTS_VALUE){
828             if(s->user_specified_pts != AV_NOPTS_VALUE){
829                 int64_t time= pts;
830                 int64_t last= s->user_specified_pts;
831
832                 if(time <= last){
833                     av_log(s->avctx, AV_LOG_ERROR, "Error, Invalid timestamp=%"PRId64", last=%"PRId64"\n", pts, s->user_specified_pts);
834                     return -1;
835                 }
836             }
837             s->user_specified_pts= pts;
838         }else{
839             if(s->user_specified_pts != AV_NOPTS_VALUE){
840                 s->user_specified_pts=
841                 pts= s->user_specified_pts + 1;
842                 av_log(s->avctx, AV_LOG_INFO, "Warning: AVFrame.pts=? trying to guess (%"PRId64")\n", pts);
843             }else{
844                 pts= pic_arg->display_picture_number;
845             }
846         }
847     }
848
849   if(pic_arg){
850     if(encoding_delay && !(s->flags&CODEC_FLAG_INPUT_PRESERVED)) direct=0;
851     if(pic_arg->linesize[0] != s->linesize) direct=0;
852     if(pic_arg->linesize[1] != s->uvlinesize) direct=0;
853     if(pic_arg->linesize[2] != s->uvlinesize) direct=0;
854
855 //    av_log(AV_LOG_DEBUG, "%d %d %d %d\n",pic_arg->linesize[0], pic_arg->linesize[1], s->linesize, s->uvlinesize);
856
857     if(direct){
858         i= ff_find_unused_picture(s, 1);
859
860         pic= (AVFrame*)&s->picture[i];
861         pic->reference= 3;
862
863         for(i=0; i<4; i++){
864             pic->data[i]= pic_arg->data[i];
865             pic->linesize[i]= pic_arg->linesize[i];
866         }
867         if(ff_alloc_picture(s, (Picture*)pic, 1) < 0){
868             return -1;
869         }
870     }else{
871         i= ff_find_unused_picture(s, 0);
872
873         pic= (AVFrame*)&s->picture[i];
874         pic->reference= 3;
875
876         if(ff_alloc_picture(s, (Picture*)pic, 0) < 0){
877             return -1;
878         }
879
880         if(   pic->data[0] + INPLACE_OFFSET == pic_arg->data[0]
881            && pic->data[1] + INPLACE_OFFSET == pic_arg->data[1]
882            && pic->data[2] + INPLACE_OFFSET == pic_arg->data[2]){
883        // empty
884         }else{
885             int h_chroma_shift, v_chroma_shift;
886             avcodec_get_chroma_sub_sample(s->avctx->pix_fmt, &h_chroma_shift, &v_chroma_shift);
887
888             for(i=0; i<3; i++){
889                 int src_stride= pic_arg->linesize[i];
890                 int dst_stride= i ? s->uvlinesize : s->linesize;
891                 int h_shift= i ? h_chroma_shift : 0;
892                 int v_shift= i ? v_chroma_shift : 0;
893                 int w= s->width >>h_shift;
894                 int h= s->height>>v_shift;
895                 uint8_t *src= pic_arg->data[i];
896                 uint8_t *dst= pic->data[i];
897
898                 if(!s->avctx->rc_buffer_size)
899                     dst +=INPLACE_OFFSET;
900
901                 if(src_stride==dst_stride)
902                     memcpy(dst, src, src_stride*h);
903                 else{
904                     while(h--){
905                         memcpy(dst, src, w);
906                         dst += dst_stride;
907                         src += src_stride;
908                     }
909                 }
910             }
911         }
912     }
913     copy_picture_attributes(s, pic, pic_arg);
914     pic->pts= pts; //we set this here to avoid modifiying pic_arg
915   }
916
917     /* shift buffer entries */
918     for(i=1; i<MAX_PICTURE_COUNT /*s->encoding_delay+1*/; i++)
919         s->input_picture[i-1]= s->input_picture[i];
920
921     s->input_picture[encoding_delay]= (Picture*)pic;
922
923     return 0;
924 }
925
926 static int skip_check(MpegEncContext *s, Picture *p, Picture *ref){
927     int x, y, plane;
928     int score=0;
929     int64_t score64=0;
930
931     for(plane=0; plane<3; plane++){
932         const int stride = p->f.linesize[plane];
933         const int bw= plane ? 1 : 2;
934         for(y=0; y<s->mb_height*bw; y++){
935             for(x=0; x<s->mb_width*bw; x++){
936                 int off = p->f.type == FF_BUFFER_TYPE_SHARED ? 0: 16;
937                 int v   = s->dsp.frame_skip_cmp[1](s, p->f.data[plane] + 8*(x + y*stride)+off, ref->f.data[plane] + 8*(x + y*stride), stride, 8);
938
939                 switch(s->avctx->frame_skip_exp){
940                     case 0: score= FFMAX(score, v); break;
941                     case 1: score+= FFABS(v);break;
942                     case 2: score+= v*v;break;
943                     case 3: score64+= FFABS(v*v*(int64_t)v);break;
944                     case 4: score64+= v*v*(int64_t)(v*v);break;
945                 }
946             }
947         }
948     }
949
950     if(score) score64= score;
951
952     if(score64 < s->avctx->frame_skip_threshold)
953         return 1;
954     if(score64 < ((s->avctx->frame_skip_factor * (int64_t)s->lambda)>>8))
955         return 1;
956     return 0;
957 }
958
959 static int estimate_best_b_count(MpegEncContext *s){
960     AVCodec *codec= avcodec_find_encoder(s->avctx->codec_id);
961     AVCodecContext *c = avcodec_alloc_context3(NULL);
962     AVFrame input[FF_MAX_B_FRAMES+2];
963     const int scale= s->avctx->brd_scale;
964     int i, j, out_size, p_lambda, b_lambda, lambda2;
965     int outbuf_size= s->width * s->height; //FIXME
966     uint8_t *outbuf= av_malloc(outbuf_size);
967     int64_t best_rd= INT64_MAX;
968     int best_b_count= -1;
969
970     assert(scale>=0 && scale <=3);
971
972 //    emms_c();
973     p_lambda= s->last_lambda_for[AV_PICTURE_TYPE_P]; //s->next_picture_ptr->quality;
974     b_lambda= s->last_lambda_for[AV_PICTURE_TYPE_B]; //p_lambda *FFABS(s->avctx->b_quant_factor) + s->avctx->b_quant_offset;
975     if(!b_lambda) b_lambda= p_lambda; //FIXME we should do this somewhere else
976     lambda2= (b_lambda*b_lambda + (1<<FF_LAMBDA_SHIFT)/2 ) >> FF_LAMBDA_SHIFT;
977
978     c->width = s->width >> scale;
979     c->height= s->height>> scale;
980     c->flags= CODEC_FLAG_QSCALE | CODEC_FLAG_PSNR | CODEC_FLAG_INPUT_PRESERVED /*| CODEC_FLAG_EMU_EDGE*/;
981     c->flags|= s->avctx->flags & CODEC_FLAG_QPEL;
982     c->mb_decision= s->avctx->mb_decision;
983     c->me_cmp= s->avctx->me_cmp;
984     c->mb_cmp= s->avctx->mb_cmp;
985     c->me_sub_cmp= s->avctx->me_sub_cmp;
986     c->pix_fmt = PIX_FMT_YUV420P;
987     c->time_base= s->avctx->time_base;
988     c->max_b_frames= s->max_b_frames;
989
990     if (avcodec_open2(c, codec, NULL) < 0)
991         return -1;
992
993     for(i=0; i<s->max_b_frames+2; i++){
994         int ysize= c->width*c->height;
995         int csize= (c->width/2)*(c->height/2);
996         Picture pre_input, *pre_input_ptr= i ? s->input_picture[i-1] : s->next_picture_ptr;
997
998         avcodec_get_frame_defaults(&input[i]);
999         input[i].data[0]= av_malloc(ysize + 2*csize);
1000         input[i].data[1]= input[i].data[0] + ysize;
1001         input[i].data[2]= input[i].data[1] + csize;
1002         input[i].linesize[0]= c->width;
1003         input[i].linesize[1]=
1004         input[i].linesize[2]= c->width/2;
1005
1006         if(pre_input_ptr && (!i || s->input_picture[i-1])) {
1007             pre_input= *pre_input_ptr;
1008
1009             if (pre_input.f.type != FF_BUFFER_TYPE_SHARED && i) {
1010                 pre_input.f.data[0] += INPLACE_OFFSET;
1011                 pre_input.f.data[1] += INPLACE_OFFSET;
1012                 pre_input.f.data[2] += INPLACE_OFFSET;
1013             }
1014
1015             s->dsp.shrink[scale](input[i].data[0], input[i].linesize[0], pre_input.f.data[0], pre_input.f.linesize[0], c->width,      c->height);
1016             s->dsp.shrink[scale](input[i].data[1], input[i].linesize[1], pre_input.f.data[1], pre_input.f.linesize[1], c->width >> 1, c->height >> 1);
1017             s->dsp.shrink[scale](input[i].data[2], input[i].linesize[2], pre_input.f.data[2], pre_input.f.linesize[2], c->width >> 1, c->height >> 1);
1018         }
1019     }
1020
1021     for(j=0; j<s->max_b_frames+1; j++){
1022         int64_t rd=0;
1023
1024         if(!s->input_picture[j])
1025             break;
1026
1027         c->error[0]= c->error[1]= c->error[2]= 0;
1028
1029         input[0].pict_type= AV_PICTURE_TYPE_I;
1030         input[0].quality= 1 * FF_QP2LAMBDA;
1031         out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[0]);
1032 //        rd += (out_size * lambda2) >> FF_LAMBDA_SHIFT;
1033
1034         for(i=0; i<s->max_b_frames+1; i++){
1035             int is_p= i % (j+1) == j || i==s->max_b_frames;
1036
1037             input[i+1].pict_type= is_p ? AV_PICTURE_TYPE_P : AV_PICTURE_TYPE_B;
1038             input[i+1].quality= is_p ? p_lambda : b_lambda;
1039             out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[i+1]);
1040             rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
1041         }
1042
1043         /* get the delayed frames */
1044         while(out_size){
1045             out_size = avcodec_encode_video(c, outbuf, outbuf_size, NULL);
1046             rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
1047         }
1048
1049         rd += c->error[0] + c->error[1] + c->error[2];
1050
1051         if(rd < best_rd){
1052             best_rd= rd;
1053             best_b_count= j;
1054         }
1055     }
1056
1057     av_freep(&outbuf);
1058     avcodec_close(c);
1059     av_freep(&c);
1060
1061     for(i=0; i<s->max_b_frames+2; i++){
1062         av_freep(&input[i].data[0]);
1063     }
1064
1065     return best_b_count;
1066 }
1067
1068 static int select_input_picture(MpegEncContext *s){
1069     int i;
1070
1071     for(i=1; i<MAX_PICTURE_COUNT; i++)
1072         s->reordered_input_picture[i-1]= s->reordered_input_picture[i];
1073     s->reordered_input_picture[MAX_PICTURE_COUNT-1]= NULL;
1074
1075     /* set next picture type & ordering */
1076     if(s->reordered_input_picture[0]==NULL && s->input_picture[0]){
1077         if(/*s->picture_in_gop_number >= s->gop_size ||*/ s->next_picture_ptr==NULL || s->intra_only){
1078             s->reordered_input_picture[0]= s->input_picture[0];
1079             s->reordered_input_picture[0]->f.pict_type = AV_PICTURE_TYPE_I;
1080             s->reordered_input_picture[0]->f.coded_picture_number = s->coded_picture_number++;
1081         }else{
1082             int b_frames;
1083
1084             if(s->avctx->frame_skip_threshold || s->avctx->frame_skip_factor){
1085                 if(s->picture_in_gop_number < s->gop_size && skip_check(s, s->input_picture[0], s->next_picture_ptr)){
1086                 //FIXME check that te gop check above is +-1 correct
1087 //av_log(NULL, AV_LOG_DEBUG, "skip %p %"PRId64"\n", s->input_picture[0]->f.data[0], s->input_picture[0]->pts);
1088
1089                     if (s->input_picture[0]->f.type == FF_BUFFER_TYPE_SHARED) {
1090                         for(i=0; i<4; i++)
1091                             s->input_picture[0]->f.data[i] = NULL;
1092                         s->input_picture[0]->f.type = 0;
1093                     }else{
1094                         assert(   s->input_picture[0]->type==FF_BUFFER_TYPE_USER
1095                                || s->input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
1096
1097                         s->avctx->release_buffer(s->avctx, (AVFrame*)s->input_picture[0]);
1098                     }
1099
1100                     emms_c();
1101                     ff_vbv_update(s, 0);
1102
1103                     goto no_output_pic;
1104                 }
1105             }
1106
1107             if(s->flags&CODEC_FLAG_PASS2){
1108                 for(i=0; i<s->max_b_frames+1; i++){
1109                     int pict_num = s->input_picture[0]->f.display_picture_number + i;
1110
1111                     if(pict_num >= s->rc_context.num_entries)
1112                         break;
1113                     if(!s->input_picture[i]){
1114                         s->rc_context.entry[pict_num-1].new_pict_type = AV_PICTURE_TYPE_P;
1115                         break;
1116                     }
1117
1118                     s->input_picture[i]->f.pict_type =
1119                         s->rc_context.entry[pict_num].new_pict_type;
1120                 }
1121             }
1122
1123             if(s->avctx->b_frame_strategy==0){
1124                 b_frames= s->max_b_frames;
1125                 while(b_frames && !s->input_picture[b_frames]) b_frames--;
1126             }else if(s->avctx->b_frame_strategy==1){
1127                 for(i=1; i<s->max_b_frames+1; i++){
1128                     if(s->input_picture[i] && s->input_picture[i]->b_frame_score==0){
1129                         s->input_picture[i]->b_frame_score=
1130                             get_intra_count(s, s->input_picture[i  ]->f.data[0],
1131                                                s->input_picture[i-1]->f.data[0], s->linesize) + 1;
1132                     }
1133                 }
1134                 for(i=0; i<s->max_b_frames+1; i++){
1135                     if(s->input_picture[i]==NULL || s->input_picture[i]->b_frame_score - 1 > s->mb_num/s->avctx->b_sensitivity) break;
1136                 }
1137
1138                 b_frames= FFMAX(0, i-1);
1139
1140                 /* reset scores */
1141                 for(i=0; i<b_frames+1; i++){
1142                     s->input_picture[i]->b_frame_score=0;
1143                 }
1144             }else if(s->avctx->b_frame_strategy==2){
1145                 b_frames= estimate_best_b_count(s);
1146             }else{
1147                 av_log(s->avctx, AV_LOG_ERROR, "illegal b frame strategy\n");
1148                 b_frames=0;
1149             }
1150
1151             emms_c();
1152 //static int b_count=0;
1153 //b_count+= b_frames;
1154 //av_log(s->avctx, AV_LOG_DEBUG, "b_frames: %d\n", b_count);
1155
1156             for(i= b_frames - 1; i>=0; i--){
1157                 int type = s->input_picture[i]->f.pict_type;
1158                 if(type && type != AV_PICTURE_TYPE_B)
1159                     b_frames= i;
1160             }
1161             if (s->input_picture[b_frames]->f.pict_type == AV_PICTURE_TYPE_B && b_frames == s->max_b_frames){
1162                 av_log(s->avctx, AV_LOG_ERROR, "warning, too many b frames in a row\n");
1163             }
1164
1165             if(s->picture_in_gop_number + b_frames >= s->gop_size){
1166               if((s->flags2 & CODEC_FLAG2_STRICT_GOP) && s->gop_size > s->picture_in_gop_number){
1167                     b_frames= s->gop_size - s->picture_in_gop_number - 1;
1168               }else{
1169                 if(s->flags & CODEC_FLAG_CLOSED_GOP)
1170                     b_frames=0;
1171                 s->input_picture[b_frames]->f.pict_type = AV_PICTURE_TYPE_I;
1172               }
1173             }
1174
1175             if(   (s->flags & CODEC_FLAG_CLOSED_GOP)
1176                && b_frames
1177                && s->input_picture[b_frames]->f.pict_type== AV_PICTURE_TYPE_I)
1178                 b_frames--;
1179
1180             s->reordered_input_picture[0]= s->input_picture[b_frames];
1181             if (s->reordered_input_picture[0]->f.pict_type != AV_PICTURE_TYPE_I)
1182                 s->reordered_input_picture[0]->f.pict_type = AV_PICTURE_TYPE_P;
1183             s->reordered_input_picture[0]->f.coded_picture_number = s->coded_picture_number++;
1184             for(i=0; i<b_frames; i++){
1185                 s->reordered_input_picture[i + 1] = s->input_picture[i];
1186                 s->reordered_input_picture[i + 1]->f.pict_type = AV_PICTURE_TYPE_B;
1187                 s->reordered_input_picture[i + 1]->f.coded_picture_number = s->coded_picture_number++;
1188             }
1189         }
1190     }
1191 no_output_pic:
1192     if(s->reordered_input_picture[0]){
1193         s->reordered_input_picture[0]->f.reference = s->reordered_input_picture[0]->f.pict_type!=AV_PICTURE_TYPE_B ? 3 : 0;
1194
1195         ff_copy_picture(&s->new_picture, s->reordered_input_picture[0]);
1196
1197         if (s->reordered_input_picture[0]->f.type == FF_BUFFER_TYPE_SHARED || s->avctx->rc_buffer_size) {
1198             // input is a shared pix, so we can't modifiy it -> alloc a new one & ensure that the shared one is reuseable
1199
1200             int i= ff_find_unused_picture(s, 0);
1201             Picture *pic= &s->picture[i];
1202
1203             pic->f.reference = s->reordered_input_picture[0]->f.reference;
1204             if(ff_alloc_picture(s, pic, 0) < 0){
1205                 return -1;
1206             }
1207
1208             /* mark us unused / free shared pic */
1209             if (s->reordered_input_picture[0]->f.type == FF_BUFFER_TYPE_INTERNAL)
1210                 s->avctx->release_buffer(s->avctx, (AVFrame*)s->reordered_input_picture[0]);
1211             for(i=0; i<4; i++)
1212                 s->reordered_input_picture[0]->f.data[i] = NULL;
1213             s->reordered_input_picture[0]->f.type = 0;
1214
1215             copy_picture_attributes(s, (AVFrame*)pic, (AVFrame*)s->reordered_input_picture[0]);
1216
1217             s->current_picture_ptr= pic;
1218         }else{
1219             // input is not a shared pix -> reuse buffer for current_pix
1220
1221             assert(   s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_USER
1222                    || s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
1223
1224             s->current_picture_ptr= s->reordered_input_picture[0];
1225             for(i=0; i<4; i++){
1226                 s->new_picture.f.data[i] += INPLACE_OFFSET;
1227             }
1228         }
1229         ff_copy_picture(&s->current_picture, s->current_picture_ptr);
1230
1231         s->picture_number = s->new_picture.f.display_picture_number;
1232 //printf("dpn:%d\n", s->picture_number);
1233     }else{
1234        memset(&s->new_picture, 0, sizeof(Picture));
1235     }
1236     return 0;
1237 }
1238
1239 int MPV_encode_picture(AVCodecContext *avctx,
1240                        unsigned char *buf, int buf_size, void *data)
1241 {
1242     MpegEncContext *s = avctx->priv_data;
1243     AVFrame *pic_arg = data;
1244     int i, stuffing_count, context_count = avctx->thread_count;
1245
1246     for(i=0; i<context_count; i++){
1247         int start_y= s->thread_context[i]->start_mb_y;
1248         int   end_y= s->thread_context[i]->  end_mb_y;
1249         int h= s->mb_height;
1250         uint8_t *start= buf + (size_t)(((int64_t) buf_size)*start_y/h);
1251         uint8_t *end  = buf + (size_t)(((int64_t) buf_size)*  end_y/h);
1252
1253         init_put_bits(&s->thread_context[i]->pb, start, end - start);
1254     }
1255
1256     s->picture_in_gop_number++;
1257
1258     if(load_input_picture(s, pic_arg) < 0)
1259         return -1;
1260
1261     if(select_input_picture(s) < 0){
1262         return -1;
1263     }
1264
1265     /* output? */
1266     if (s->new_picture.f.data[0]) {
1267         s->pict_type = s->new_picture.f.pict_type;
1268 //emms_c();
1269 //printf("qs:%f %f %d\n", s->new_picture.quality, s->current_picture.quality, s->qscale);
1270         MPV_frame_start(s, avctx);
1271 vbv_retry:
1272         if (encode_picture(s, s->picture_number) < 0)
1273             return -1;
1274
1275         avctx->header_bits = s->header_bits;
1276         avctx->mv_bits     = s->mv_bits;
1277         avctx->misc_bits   = s->misc_bits;
1278         avctx->i_tex_bits  = s->i_tex_bits;
1279         avctx->p_tex_bits  = s->p_tex_bits;
1280         avctx->i_count     = s->i_count;
1281         avctx->p_count     = s->mb_num - s->i_count - s->skip_count; //FIXME f/b_count in avctx
1282         avctx->skip_count  = s->skip_count;
1283
1284         MPV_frame_end(s);
1285
1286         if (CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG)
1287             ff_mjpeg_encode_picture_trailer(s);
1288
1289         if(avctx->rc_buffer_size){
1290             RateControlContext *rcc= &s->rc_context;
1291             int max_size= rcc->buffer_index * avctx->rc_max_available_vbv_use;
1292
1293             if(put_bits_count(&s->pb) > max_size && s->lambda < s->avctx->lmax){
1294                 s->next_lambda= FFMAX(s->lambda+1, s->lambda*(s->qscale+1) / s->qscale);
1295                 if(s->adaptive_quant){
1296                     int i;
1297                     for(i=0; i<s->mb_height*s->mb_stride; i++)
1298                         s->lambda_table[i]= FFMAX(s->lambda_table[i]+1, s->lambda_table[i]*(s->qscale+1) / s->qscale);
1299                 }
1300                 s->mb_skipped = 0;        //done in MPV_frame_start()
1301                 if(s->pict_type==AV_PICTURE_TYPE_P){ //done in encode_picture() so we must undo it
1302                     if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
1303                         s->no_rounding ^= 1;
1304                 }
1305                 if(s->pict_type!=AV_PICTURE_TYPE_B){
1306                     s->time_base= s->last_time_base;
1307                     s->last_non_b_time= s->time - s->pp_time;
1308                 }
1309 //                av_log(NULL, AV_LOG_ERROR, "R:%d ", s->next_lambda);
1310                 for(i=0; i<context_count; i++){
1311                     PutBitContext *pb= &s->thread_context[i]->pb;
1312                     init_put_bits(pb, pb->buf, pb->buf_end - pb->buf);
1313                 }
1314                 goto vbv_retry;
1315             }
1316
1317             assert(s->avctx->rc_max_rate);
1318         }
1319
1320         if(s->flags&CODEC_FLAG_PASS1)
1321             ff_write_pass1_stats(s);
1322
1323         for(i=0; i<4; i++){
1324             s->current_picture_ptr->f.error[i]  = s->current_picture.f.error[i];
1325             avctx->error[i]                        += s->current_picture_ptr->f.error[i];
1326         }
1327
1328         if(s->flags&CODEC_FLAG_PASS1)
1329             assert(avctx->header_bits + avctx->mv_bits + avctx->misc_bits + avctx->i_tex_bits + avctx->p_tex_bits == put_bits_count(&s->pb));
1330         flush_put_bits(&s->pb);
1331         s->frame_bits  = put_bits_count(&s->pb);
1332
1333         stuffing_count= ff_vbv_update(s, s->frame_bits);
1334         if(stuffing_count){
1335             if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < stuffing_count + 50){
1336                 av_log(s->avctx, AV_LOG_ERROR, "stuffing too large\n");
1337                 return -1;
1338             }
1339
1340             switch(s->codec_id){
1341             case CODEC_ID_MPEG1VIDEO:
1342             case CODEC_ID_MPEG2VIDEO:
1343                 while(stuffing_count--){
1344                     put_bits(&s->pb, 8, 0);
1345                 }
1346             break;
1347             case CODEC_ID_MPEG4:
1348                 put_bits(&s->pb, 16, 0);
1349                 put_bits(&s->pb, 16, 0x1C3);
1350                 stuffing_count -= 4;
1351                 while(stuffing_count--){
1352                     put_bits(&s->pb, 8, 0xFF);
1353                 }
1354             break;
1355             default:
1356                 av_log(s->avctx, AV_LOG_ERROR, "vbv buffer overflow\n");
1357             }
1358             flush_put_bits(&s->pb);
1359             s->frame_bits  = put_bits_count(&s->pb);
1360         }
1361
1362         /* update mpeg1/2 vbv_delay for CBR */
1363         if(s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate && s->out_format == FMT_MPEG1
1364            && 90000LL * (avctx->rc_buffer_size-1) <= s->avctx->rc_max_rate*0xFFFFLL){
1365             int vbv_delay, min_delay;
1366             double inbits = s->avctx->rc_max_rate*av_q2d(s->avctx->time_base);
1367             int    minbits= s->frame_bits - 8*(s->vbv_delay_ptr - s->pb.buf - 1);
1368             double bits   = s->rc_context.buffer_index + minbits - inbits;
1369
1370             if(bits<0)
1371                 av_log(s->avctx, AV_LOG_ERROR, "Internal error, negative bits\n");
1372
1373             assert(s->repeat_first_field==0);
1374
1375             vbv_delay=     bits * 90000                               / s->avctx->rc_max_rate;
1376             min_delay= (minbits * 90000LL + s->avctx->rc_max_rate - 1)/ s->avctx->rc_max_rate;
1377
1378             vbv_delay= FFMAX(vbv_delay, min_delay);
1379
1380             assert(vbv_delay < 0xFFFF);
1381
1382             s->vbv_delay_ptr[0] &= 0xF8;
1383             s->vbv_delay_ptr[0] |= vbv_delay>>13;
1384             s->vbv_delay_ptr[1]  = vbv_delay>>5;
1385             s->vbv_delay_ptr[2] &= 0x07;
1386             s->vbv_delay_ptr[2] |= vbv_delay<<3;
1387             avctx->vbv_delay = vbv_delay*300;
1388         }
1389         s->total_bits += s->frame_bits;
1390         avctx->frame_bits  = s->frame_bits;
1391     }else{
1392         assert((put_bits_ptr(&s->pb) == s->pb.buf));
1393         s->frame_bits=0;
1394     }
1395     assert((s->frame_bits&7)==0);
1396
1397     return s->frame_bits/8;
1398 }
1399
1400 static inline void dct_single_coeff_elimination(MpegEncContext *s, int n, int threshold)
1401 {
1402     static const char tab[64]=
1403         {3,2,2,1,1,1,1,1,
1404          1,1,1,1,1,1,1,1,
1405          1,1,1,1,1,1,1,1,
1406          0,0,0,0,0,0,0,0,
1407          0,0,0,0,0,0,0,0,
1408          0,0,0,0,0,0,0,0,
1409          0,0,0,0,0,0,0,0,
1410          0,0,0,0,0,0,0,0};
1411     int score=0;
1412     int run=0;
1413     int i;
1414     DCTELEM *block= s->block[n];
1415     const int last_index= s->block_last_index[n];
1416     int skip_dc;
1417
1418     if(threshold<0){
1419         skip_dc=0;
1420         threshold= -threshold;
1421     }else
1422         skip_dc=1;
1423
1424     /* Are all we could set to zero already zero? */
1425     if(last_index<=skip_dc - 1) return;
1426
1427     for(i=0; i<=last_index; i++){
1428         const int j = s->intra_scantable.permutated[i];
1429         const int level = FFABS(block[j]);
1430         if(level==1){
1431             if(skip_dc && i==0) continue;
1432             score+= tab[run];
1433             run=0;
1434         }else if(level>1){
1435             return;
1436         }else{
1437             run++;
1438         }
1439     }
1440     if(score >= threshold) return;
1441     for(i=skip_dc; i<=last_index; i++){
1442         const int j = s->intra_scantable.permutated[i];
1443         block[j]=0;
1444     }
1445     if(block[0]) s->block_last_index[n]= 0;
1446     else         s->block_last_index[n]= -1;
1447 }
1448
1449 static inline void clip_coeffs(MpegEncContext *s, DCTELEM *block, int last_index)
1450 {
1451     int i;
1452     const int maxlevel= s->max_qcoeff;
1453     const int minlevel= s->min_qcoeff;
1454     int overflow=0;
1455
1456     if(s->mb_intra){
1457         i=1; //skip clipping of intra dc
1458     }else
1459         i=0;
1460
1461     for(;i<=last_index; i++){
1462         const int j= s->intra_scantable.permutated[i];
1463         int level = block[j];
1464
1465         if     (level>maxlevel){
1466             level=maxlevel;
1467             overflow++;
1468         }else if(level<minlevel){
1469             level=minlevel;
1470             overflow++;
1471         }
1472
1473         block[j]= level;
1474     }
1475
1476     if(overflow && s->avctx->mb_decision == FF_MB_DECISION_SIMPLE)
1477         av_log(s->avctx, AV_LOG_INFO, "warning, clipping %d dct coefficients to %d..%d\n", overflow, minlevel, maxlevel);
1478 }
1479
1480 static void get_visual_weight(int16_t *weight, uint8_t *ptr, int stride){
1481     int x, y;
1482 //FIXME optimize
1483     for(y=0; y<8; y++){
1484         for(x=0; x<8; x++){
1485             int x2, y2;
1486             int sum=0;
1487             int sqr=0;
1488             int count=0;
1489
1490             for(y2= FFMAX(y-1, 0); y2 < FFMIN(8, y+2); y2++){
1491                 for(x2= FFMAX(x-1, 0); x2 < FFMIN(8, x+2); x2++){
1492                     int v= ptr[x2 + y2*stride];
1493                     sum += v;
1494                     sqr += v*v;
1495                     count++;
1496                 }
1497             }
1498             weight[x + 8*y]= (36*ff_sqrt(count*sqr - sum*sum)) / count;
1499         }
1500     }
1501 }
1502
1503 static av_always_inline void encode_mb_internal(MpegEncContext *s, int motion_x, int motion_y, int mb_block_height, int mb_block_count)
1504 {
1505     int16_t weight[8][64];
1506     DCTELEM orig[8][64];
1507     const int mb_x= s->mb_x;
1508     const int mb_y= s->mb_y;
1509     int i;
1510     int skip_dct[8];
1511     int dct_offset   = s->linesize*8; //default for progressive frames
1512     uint8_t *ptr_y, *ptr_cb, *ptr_cr;
1513     int wrap_y, wrap_c;
1514
1515     for(i=0; i<mb_block_count; i++) skip_dct[i]=s->skipdct;
1516
1517     if(s->adaptive_quant){
1518         const int last_qp= s->qscale;
1519         const int mb_xy= mb_x + mb_y*s->mb_stride;
1520
1521         s->lambda= s->lambda_table[mb_xy];
1522         update_qscale(s);
1523
1524         if(!(s->flags&CODEC_FLAG_QP_RD)){
1525             s->qscale = s->current_picture_ptr->f.qscale_table[mb_xy];
1526             s->dquant= s->qscale - last_qp;
1527
1528             if(s->out_format==FMT_H263){
1529                 s->dquant= av_clip(s->dquant, -2, 2);
1530
1531                 if(s->codec_id==CODEC_ID_MPEG4){
1532                     if(!s->mb_intra){
1533                         if(s->pict_type == AV_PICTURE_TYPE_B){
1534                             if(s->dquant&1 || s->mv_dir&MV_DIRECT)
1535                                 s->dquant= 0;
1536                         }
1537                         if(s->mv_type==MV_TYPE_8X8)
1538                             s->dquant=0;
1539                     }
1540                 }
1541             }
1542         }
1543         ff_set_qscale(s, last_qp + s->dquant);
1544     }else if(s->flags&CODEC_FLAG_QP_RD)
1545         ff_set_qscale(s, s->qscale + s->dquant);
1546
1547     wrap_y = s->linesize;
1548     wrap_c = s->uvlinesize;
1549     ptr_y  = s->new_picture.f.data[0] + (mb_y * 16 * wrap_y) + mb_x * 16;
1550     ptr_cb = s->new_picture.f.data[1] + (mb_y * mb_block_height * wrap_c) + mb_x * 8;
1551     ptr_cr = s->new_picture.f.data[2] + (mb_y * mb_block_height * wrap_c) + mb_x * 8;
1552
1553     if(mb_x*16+16 > s->width || mb_y*16+16 > s->height){
1554         uint8_t *ebuf= s->edge_emu_buffer + 32;
1555         s->dsp.emulated_edge_mc(ebuf            , ptr_y , wrap_y,16,16,mb_x*16,mb_y*16, s->width   , s->height);
1556         ptr_y= ebuf;
1557         s->dsp.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);
1558         ptr_cb= ebuf+18*wrap_y;
1559         s->dsp.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);
1560         ptr_cr= ebuf+18*wrap_y+8;
1561     }
1562
1563     if (s->mb_intra) {
1564         if(s->flags&CODEC_FLAG_INTERLACED_DCT){
1565             int progressive_score, interlaced_score;
1566
1567             s->interlaced_dct=0;
1568             progressive_score= s->dsp.ildct_cmp[4](s, ptr_y           , NULL, wrap_y, 8)
1569                               +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y*8, NULL, wrap_y, 8) - 400;
1570
1571             if(progressive_score > 0){
1572                 interlaced_score = s->dsp.ildct_cmp[4](s, ptr_y           , NULL, wrap_y*2, 8)
1573                                   +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y  , NULL, wrap_y*2, 8);
1574                 if(progressive_score > interlaced_score){
1575                     s->interlaced_dct=1;
1576
1577                     dct_offset= wrap_y;
1578                     wrap_y<<=1;
1579                     if (s->chroma_format == CHROMA_422)
1580                         wrap_c<<=1;
1581                 }
1582             }
1583         }
1584
1585         s->dsp.get_pixels(s->block[0], ptr_y                 , wrap_y);
1586         s->dsp.get_pixels(s->block[1], ptr_y              + 8, wrap_y);
1587         s->dsp.get_pixels(s->block[2], ptr_y + dct_offset    , wrap_y);
1588         s->dsp.get_pixels(s->block[3], ptr_y + dct_offset + 8, wrap_y);
1589
1590         if(s->flags&CODEC_FLAG_GRAY){
1591             skip_dct[4]= 1;
1592             skip_dct[5]= 1;
1593         }else{
1594             s->dsp.get_pixels(s->block[4], ptr_cb, wrap_c);
1595             s->dsp.get_pixels(s->block[5], ptr_cr, wrap_c);
1596             if(!s->chroma_y_shift){ /* 422 */
1597                 s->dsp.get_pixels(s->block[6], ptr_cb + (dct_offset>>1), wrap_c);
1598                 s->dsp.get_pixels(s->block[7], ptr_cr + (dct_offset>>1), wrap_c);
1599             }
1600         }
1601     }else{
1602         op_pixels_func (*op_pix)[4];
1603         qpel_mc_func (*op_qpix)[16];
1604         uint8_t *dest_y, *dest_cb, *dest_cr;
1605
1606         dest_y  = s->dest[0];
1607         dest_cb = s->dest[1];
1608         dest_cr = s->dest[2];
1609
1610         if ((!s->no_rounding) || s->pict_type==AV_PICTURE_TYPE_B){
1611             op_pix = s->dsp.put_pixels_tab;
1612             op_qpix= s->dsp.put_qpel_pixels_tab;
1613         }else{
1614             op_pix = s->dsp.put_no_rnd_pixels_tab;
1615             op_qpix= s->dsp.put_no_rnd_qpel_pixels_tab;
1616         }
1617
1618         if (s->mv_dir & MV_DIR_FORWARD) {
1619             MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.f.data, op_pix, op_qpix);
1620             op_pix = s->dsp.avg_pixels_tab;
1621             op_qpix= s->dsp.avg_qpel_pixels_tab;
1622         }
1623         if (s->mv_dir & MV_DIR_BACKWARD) {
1624             MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.f.data, op_pix, op_qpix);
1625         }
1626
1627         if(s->flags&CODEC_FLAG_INTERLACED_DCT){
1628             int progressive_score, interlaced_score;
1629
1630             s->interlaced_dct=0;
1631             progressive_score= s->dsp.ildct_cmp[0](s, dest_y           , ptr_y           , wrap_y, 8)
1632                               +s->dsp.ildct_cmp[0](s, dest_y + wrap_y*8, ptr_y + wrap_y*8, wrap_y, 8) - 400;
1633
1634             if(s->avctx->ildct_cmp == FF_CMP_VSSE) progressive_score -= 400;
1635
1636             if(progressive_score>0){
1637                 interlaced_score = s->dsp.ildct_cmp[0](s, dest_y           , ptr_y           , wrap_y*2, 8)
1638                                   +s->dsp.ildct_cmp[0](s, dest_y + wrap_y  , ptr_y + wrap_y  , wrap_y*2, 8);
1639
1640                 if(progressive_score > interlaced_score){
1641                     s->interlaced_dct=1;
1642
1643                     dct_offset= wrap_y;
1644                     wrap_y<<=1;
1645                     if (s->chroma_format == CHROMA_422)
1646                         wrap_c<<=1;
1647                 }
1648             }
1649         }
1650
1651         s->dsp.diff_pixels(s->block[0], ptr_y                 , dest_y                 , wrap_y);
1652         s->dsp.diff_pixels(s->block[1], ptr_y              + 8, dest_y              + 8, wrap_y);
1653         s->dsp.diff_pixels(s->block[2], ptr_y + dct_offset    , dest_y + dct_offset    , wrap_y);
1654         s->dsp.diff_pixels(s->block[3], ptr_y + dct_offset + 8, dest_y + dct_offset + 8, wrap_y);
1655
1656         if(s->flags&CODEC_FLAG_GRAY){
1657             skip_dct[4]= 1;
1658             skip_dct[5]= 1;
1659         }else{
1660             s->dsp.diff_pixels(s->block[4], ptr_cb, dest_cb, wrap_c);
1661             s->dsp.diff_pixels(s->block[5], ptr_cr, dest_cr, wrap_c);
1662             if(!s->chroma_y_shift){ /* 422 */
1663                 s->dsp.diff_pixels(s->block[6], ptr_cb + (dct_offset>>1), dest_cb + (dct_offset>>1), wrap_c);
1664                 s->dsp.diff_pixels(s->block[7], ptr_cr + (dct_offset>>1), dest_cr + (dct_offset>>1), wrap_c);
1665             }
1666         }
1667         /* pre quantization */
1668         if(s->current_picture.mc_mb_var[s->mb_stride*mb_y+ mb_x]<2*s->qscale*s->qscale){
1669             //FIXME optimize
1670             if(s->dsp.sad[1](NULL, ptr_y               , dest_y               , wrap_y, 8) < 20*s->qscale) skip_dct[0]= 1;
1671             if(s->dsp.sad[1](NULL, ptr_y            + 8, dest_y            + 8, wrap_y, 8) < 20*s->qscale) skip_dct[1]= 1;
1672             if(s->dsp.sad[1](NULL, ptr_y +dct_offset   , dest_y +dct_offset   , wrap_y, 8) < 20*s->qscale) skip_dct[2]= 1;
1673             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;
1674             if(s->dsp.sad[1](NULL, ptr_cb              , dest_cb              , wrap_c, 8) < 20*s->qscale) skip_dct[4]= 1;
1675             if(s->dsp.sad[1](NULL, ptr_cr              , dest_cr              , wrap_c, 8) < 20*s->qscale) skip_dct[5]= 1;
1676             if(!s->chroma_y_shift){ /* 422 */
1677                 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;
1678                 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;
1679             }
1680         }
1681     }
1682
1683     if(s->avctx->quantizer_noise_shaping){
1684         if(!skip_dct[0]) get_visual_weight(weight[0], ptr_y                 , wrap_y);
1685         if(!skip_dct[1]) get_visual_weight(weight[1], ptr_y              + 8, wrap_y);
1686         if(!skip_dct[2]) get_visual_weight(weight[2], ptr_y + dct_offset    , wrap_y);
1687         if(!skip_dct[3]) get_visual_weight(weight[3], ptr_y + dct_offset + 8, wrap_y);
1688         if(!skip_dct[4]) get_visual_weight(weight[4], ptr_cb                , wrap_c);
1689         if(!skip_dct[5]) get_visual_weight(weight[5], ptr_cr                , wrap_c);
1690         if(!s->chroma_y_shift){ /* 422 */
1691             if(!skip_dct[6]) get_visual_weight(weight[6], ptr_cb + (dct_offset>>1), wrap_c);
1692             if(!skip_dct[7]) get_visual_weight(weight[7], ptr_cr + (dct_offset>>1), wrap_c);
1693         }
1694         memcpy(orig[0], s->block[0], sizeof(DCTELEM)*64*mb_block_count);
1695     }
1696
1697     /* DCT & quantize */
1698     assert(s->out_format!=FMT_MJPEG || s->qscale==8);
1699     {
1700         for(i=0;i<mb_block_count;i++) {
1701             if(!skip_dct[i]){
1702                 int overflow;
1703                 s->block_last_index[i] = s->dct_quantize(s, s->block[i], i, s->qscale, &overflow);
1704             // FIXME we could decide to change to quantizer instead of clipping
1705             // JS: I don't think that would be a good idea it could lower quality instead
1706             //     of improve it. Just INTRADC clipping deserves changes in quantizer
1707                 if (overflow) clip_coeffs(s, s->block[i], s->block_last_index[i]);
1708             }else
1709                 s->block_last_index[i]= -1;
1710         }
1711         if(s->avctx->quantizer_noise_shaping){
1712             for(i=0;i<mb_block_count;i++) {
1713                 if(!skip_dct[i]){
1714                     s->block_last_index[i] = dct_quantize_refine(s, s->block[i], weight[i], orig[i], i, s->qscale);
1715                 }
1716             }
1717         }
1718
1719         if(s->luma_elim_threshold && !s->mb_intra)
1720             for(i=0; i<4; i++)
1721                 dct_single_coeff_elimination(s, i, s->luma_elim_threshold);
1722         if(s->chroma_elim_threshold && !s->mb_intra)
1723             for(i=4; i<mb_block_count; i++)
1724                 dct_single_coeff_elimination(s, i, s->chroma_elim_threshold);
1725
1726         if(s->flags & CODEC_FLAG_CBP_RD){
1727             for(i=0;i<mb_block_count;i++) {
1728                 if(s->block_last_index[i] == -1)
1729                     s->coded_score[i]= INT_MAX/256;
1730             }
1731         }
1732     }
1733
1734     if((s->flags&CODEC_FLAG_GRAY) && s->mb_intra){
1735         s->block_last_index[4]=
1736         s->block_last_index[5]= 0;
1737         s->block[4][0]=
1738         s->block[5][0]= (1024 + s->c_dc_scale/2)/ s->c_dc_scale;
1739     }
1740
1741     //non c quantize code returns incorrect block_last_index FIXME
1742     if(s->alternate_scan && s->dct_quantize != dct_quantize_c){
1743         for(i=0; i<mb_block_count; i++){
1744             int j;
1745             if(s->block_last_index[i]>0){
1746                 for(j=63; j>0; j--){
1747                     if(s->block[i][ s->intra_scantable.permutated[j] ]) break;
1748                 }
1749                 s->block_last_index[i]= j;
1750             }
1751         }
1752     }
1753
1754     /* huffman encode */
1755     switch(s->codec_id){ //FIXME funct ptr could be slightly faster
1756     case CODEC_ID_MPEG1VIDEO:
1757     case CODEC_ID_MPEG2VIDEO:
1758         if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
1759             mpeg1_encode_mb(s, s->block, motion_x, motion_y);
1760         break;
1761     case CODEC_ID_MPEG4:
1762         if (CONFIG_MPEG4_ENCODER)
1763             mpeg4_encode_mb(s, s->block, motion_x, motion_y);
1764         break;
1765     case CODEC_ID_MSMPEG4V2:
1766     case CODEC_ID_MSMPEG4V3:
1767     case CODEC_ID_WMV1:
1768         if (CONFIG_MSMPEG4_ENCODER)
1769             msmpeg4_encode_mb(s, s->block, motion_x, motion_y);
1770         break;
1771     case CODEC_ID_WMV2:
1772         if (CONFIG_WMV2_ENCODER)
1773             ff_wmv2_encode_mb(s, s->block, motion_x, motion_y);
1774         break;
1775     case CODEC_ID_H261:
1776         if (CONFIG_H261_ENCODER)
1777             ff_h261_encode_mb(s, s->block, motion_x, motion_y);
1778         break;
1779     case CODEC_ID_H263:
1780     case CODEC_ID_H263P:
1781     case CODEC_ID_FLV1:
1782     case CODEC_ID_RV10:
1783     case CODEC_ID_RV20:
1784         if (CONFIG_H263_ENCODER)
1785             h263_encode_mb(s, s->block, motion_x, motion_y);
1786         break;
1787     case CODEC_ID_MJPEG:
1788         if (CONFIG_MJPEG_ENCODER)
1789             ff_mjpeg_encode_mb(s, s->block);
1790         break;
1791     default:
1792         assert(0);
1793     }
1794 }
1795
1796 static av_always_inline void encode_mb(MpegEncContext *s, int motion_x, int motion_y)
1797 {
1798     if (s->chroma_format == CHROMA_420) encode_mb_internal(s, motion_x, motion_y,  8, 6);
1799     else                                encode_mb_internal(s, motion_x, motion_y, 16, 8);
1800 }
1801
1802 static inline void copy_context_before_encode(MpegEncContext *d, MpegEncContext *s, int type){
1803     int i;
1804
1805     memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster than a loop?
1806
1807     /* mpeg1 */
1808     d->mb_skip_run= s->mb_skip_run;
1809     for(i=0; i<3; i++)
1810         d->last_dc[i]= s->last_dc[i];
1811
1812     /* statistics */
1813     d->mv_bits= s->mv_bits;
1814     d->i_tex_bits= s->i_tex_bits;
1815     d->p_tex_bits= s->p_tex_bits;
1816     d->i_count= s->i_count;
1817     d->f_count= s->f_count;
1818     d->b_count= s->b_count;
1819     d->skip_count= s->skip_count;
1820     d->misc_bits= s->misc_bits;
1821     d->last_bits= 0;
1822
1823     d->mb_skipped= 0;
1824     d->qscale= s->qscale;
1825     d->dquant= s->dquant;
1826
1827     d->esc3_level_length= s->esc3_level_length;
1828 }
1829
1830 static inline void copy_context_after_encode(MpegEncContext *d, MpegEncContext *s, int type){
1831     int i;
1832
1833     memcpy(d->mv, s->mv, 2*4*2*sizeof(int));
1834     memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster than a loop?
1835
1836     /* mpeg1 */
1837     d->mb_skip_run= s->mb_skip_run;
1838     for(i=0; i<3; i++)
1839         d->last_dc[i]= s->last_dc[i];
1840
1841     /* statistics */
1842     d->mv_bits= s->mv_bits;
1843     d->i_tex_bits= s->i_tex_bits;
1844     d->p_tex_bits= s->p_tex_bits;
1845     d->i_count= s->i_count;
1846     d->f_count= s->f_count;
1847     d->b_count= s->b_count;
1848     d->skip_count= s->skip_count;
1849     d->misc_bits= s->misc_bits;
1850
1851     d->mb_intra= s->mb_intra;
1852     d->mb_skipped= s->mb_skipped;
1853     d->mv_type= s->mv_type;
1854     d->mv_dir= s->mv_dir;
1855     d->pb= s->pb;
1856     if(s->data_partitioning){
1857         d->pb2= s->pb2;
1858         d->tex_pb= s->tex_pb;
1859     }
1860     d->block= s->block;
1861     for(i=0; i<8; i++)
1862         d->block_last_index[i]= s->block_last_index[i];
1863     d->interlaced_dct= s->interlaced_dct;
1864     d->qscale= s->qscale;
1865
1866     d->esc3_level_length= s->esc3_level_length;
1867 }
1868
1869 static inline void encode_mb_hq(MpegEncContext *s, MpegEncContext *backup, MpegEncContext *best, int type,
1870                            PutBitContext pb[2], PutBitContext pb2[2], PutBitContext tex_pb[2],
1871                            int *dmin, int *next_block, int motion_x, int motion_y)
1872 {
1873     int score;
1874     uint8_t *dest_backup[3];
1875
1876     copy_context_before_encode(s, backup, type);
1877
1878     s->block= s->blocks[*next_block];
1879     s->pb= pb[*next_block];
1880     if(s->data_partitioning){
1881         s->pb2   = pb2   [*next_block];
1882         s->tex_pb= tex_pb[*next_block];
1883     }
1884
1885     if(*next_block){
1886         memcpy(dest_backup, s->dest, sizeof(s->dest));
1887         s->dest[0] = s->rd_scratchpad;
1888         s->dest[1] = s->rd_scratchpad + 16*s->linesize;
1889         s->dest[2] = s->rd_scratchpad + 16*s->linesize + 8;
1890         assert(s->linesize >= 32); //FIXME
1891     }
1892
1893     encode_mb(s, motion_x, motion_y);
1894
1895     score= put_bits_count(&s->pb);
1896     if(s->data_partitioning){
1897         score+= put_bits_count(&s->pb2);
1898         score+= put_bits_count(&s->tex_pb);
1899     }
1900
1901     if(s->avctx->mb_decision == FF_MB_DECISION_RD){
1902         MPV_decode_mb(s, s->block);
1903
1904         score *= s->lambda2;
1905         score += sse_mb(s) << FF_LAMBDA_SHIFT;
1906     }
1907
1908     if(*next_block){
1909         memcpy(s->dest, dest_backup, sizeof(s->dest));
1910     }
1911
1912     if(score<*dmin){
1913         *dmin= score;
1914         *next_block^=1;
1915
1916         copy_context_after_encode(best, s, type);
1917     }
1918 }
1919
1920 static int sse(MpegEncContext *s, uint8_t *src1, uint8_t *src2, int w, int h, int stride){
1921     uint32_t *sq = ff_squareTbl + 256;
1922     int acc=0;
1923     int x,y;
1924
1925     if(w==16 && h==16)
1926         return s->dsp.sse[0](NULL, src1, src2, stride, 16);
1927     else if(w==8 && h==8)
1928         return s->dsp.sse[1](NULL, src1, src2, stride, 8);
1929
1930     for(y=0; y<h; y++){
1931         for(x=0; x<w; x++){
1932             acc+= sq[src1[x + y*stride] - src2[x + y*stride]];
1933         }
1934     }
1935
1936     assert(acc>=0);
1937
1938     return acc;
1939 }
1940
1941 static int sse_mb(MpegEncContext *s){
1942     int w= 16;
1943     int h= 16;
1944
1945     if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
1946     if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
1947
1948     if(w==16 && h==16)
1949       if(s->avctx->mb_cmp == FF_CMP_NSSE){
1950         return  s->dsp.nsse[0](s, s->new_picture.f.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, s->dest[0], s->linesize, 16)
1951                +s->dsp.nsse[1](s, s->new_picture.f.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[1], s->uvlinesize, 8)
1952                +s->dsp.nsse[1](s, s->new_picture.f.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[2], s->uvlinesize, 8);
1953       }else{
1954         return  s->dsp.sse[0](NULL, s->new_picture.f.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, s->dest[0], s->linesize, 16)
1955                +s->dsp.sse[1](NULL, s->new_picture.f.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[1], s->uvlinesize, 8)
1956                +s->dsp.sse[1](NULL, s->new_picture.f.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[2], s->uvlinesize, 8);
1957       }
1958     else
1959         return  sse(s, s->new_picture.f.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, s->dest[0], w, h, s->linesize)
1960                +sse(s, s->new_picture.f.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[1], w>>1, h>>1, s->uvlinesize)
1961                +sse(s, s->new_picture.f.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[2], w>>1, h>>1, s->uvlinesize);
1962 }
1963
1964 static int pre_estimate_motion_thread(AVCodecContext *c, void *arg){
1965     MpegEncContext *s= *(void**)arg;
1966
1967
1968     s->me.pre_pass=1;
1969     s->me.dia_size= s->avctx->pre_dia_size;
1970     s->first_slice_line=1;
1971     for(s->mb_y= s->end_mb_y-1; s->mb_y >= s->start_mb_y; s->mb_y--) {
1972         for(s->mb_x=s->mb_width-1; s->mb_x >=0 ;s->mb_x--) {
1973             ff_pre_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
1974         }
1975         s->first_slice_line=0;
1976     }
1977
1978     s->me.pre_pass=0;
1979
1980     return 0;
1981 }
1982
1983 static int estimate_motion_thread(AVCodecContext *c, void *arg){
1984     MpegEncContext *s= *(void**)arg;
1985
1986     ff_check_alignment();
1987
1988     s->me.dia_size= s->avctx->dia_size;
1989     s->first_slice_line=1;
1990     for(s->mb_y= s->start_mb_y; s->mb_y < s->end_mb_y; s->mb_y++) {
1991         s->mb_x=0; //for block init below
1992         ff_init_block_index(s);
1993         for(s->mb_x=0; s->mb_x < s->mb_width; s->mb_x++) {
1994             s->block_index[0]+=2;
1995             s->block_index[1]+=2;
1996             s->block_index[2]+=2;
1997             s->block_index[3]+=2;
1998
1999             /* compute motion vector & mb_type and store in context */
2000             if(s->pict_type==AV_PICTURE_TYPE_B)
2001                 ff_estimate_b_frame_motion(s, s->mb_x, s->mb_y);
2002             else
2003                 ff_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
2004         }
2005         s->first_slice_line=0;
2006     }
2007     return 0;
2008 }
2009
2010 static int mb_var_thread(AVCodecContext *c, void *arg){
2011     MpegEncContext *s= *(void**)arg;
2012     int mb_x, mb_y;
2013
2014     ff_check_alignment();
2015
2016     for(mb_y=s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
2017         for(mb_x=0; mb_x < s->mb_width; mb_x++) {
2018             int xx = mb_x * 16;
2019             int yy = mb_y * 16;
2020             uint8_t *pix = s->new_picture.f.data[0] + (yy * s->linesize) + xx;
2021             int varc;
2022             int sum = s->dsp.pix_sum(pix, s->linesize);
2023
2024             varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500 + 128)>>8;
2025
2026             s->current_picture.mb_var [s->mb_stride * mb_y + mb_x] = varc;
2027             s->current_picture.mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
2028             s->me.mb_var_sum_temp    += varc;
2029         }
2030     }
2031     return 0;
2032 }
2033
2034 static void write_slice_end(MpegEncContext *s){
2035     if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4){
2036         if(s->partitioned_frame){
2037             ff_mpeg4_merge_partitions(s);
2038         }
2039
2040         ff_mpeg4_stuffing(&s->pb);
2041     }else if(CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG){
2042         ff_mjpeg_encode_stuffing(&s->pb);
2043     }
2044
2045     align_put_bits(&s->pb);
2046     flush_put_bits(&s->pb);
2047
2048     if((s->flags&CODEC_FLAG_PASS1) && !s->partitioned_frame)
2049         s->misc_bits+= get_bits_diff(s);
2050 }
2051
2052 static int encode_thread(AVCodecContext *c, void *arg){
2053     MpegEncContext *s= *(void**)arg;
2054     int mb_x, mb_y, pdif = 0;
2055     int chr_h= 16>>s->chroma_y_shift;
2056     int i, j;
2057     MpegEncContext best_s, backup_s;
2058     uint8_t bit_buf[2][MAX_MB_BYTES];
2059     uint8_t bit_buf2[2][MAX_MB_BYTES];
2060     uint8_t bit_buf_tex[2][MAX_MB_BYTES];
2061     PutBitContext pb[2], pb2[2], tex_pb[2];
2062 //printf("%d->%d\n", s->resync_mb_y, s->end_mb_y);
2063
2064     ff_check_alignment();
2065
2066     for(i=0; i<2; i++){
2067         init_put_bits(&pb    [i], bit_buf    [i], MAX_MB_BYTES);
2068         init_put_bits(&pb2   [i], bit_buf2   [i], MAX_MB_BYTES);
2069         init_put_bits(&tex_pb[i], bit_buf_tex[i], MAX_MB_BYTES);
2070     }
2071
2072     s->last_bits= put_bits_count(&s->pb);
2073     s->mv_bits=0;
2074     s->misc_bits=0;
2075     s->i_tex_bits=0;
2076     s->p_tex_bits=0;
2077     s->i_count=0;
2078     s->f_count=0;
2079     s->b_count=0;
2080     s->skip_count=0;
2081
2082     for(i=0; i<3; i++){
2083         /* init last dc values */
2084         /* note: quant matrix value (8) is implied here */
2085         s->last_dc[i] = 128 << s->intra_dc_precision;
2086
2087         s->current_picture.f.error[i] = 0;
2088     }
2089     s->mb_skip_run = 0;
2090     memset(s->last_mv, 0, sizeof(s->last_mv));
2091
2092     s->last_mv_dir = 0;
2093
2094     switch(s->codec_id){
2095     case CODEC_ID_H263:
2096     case CODEC_ID_H263P:
2097     case CODEC_ID_FLV1:
2098         if (CONFIG_H263_ENCODER)
2099             s->gob_index = ff_h263_get_gob_height(s);
2100         break;
2101     case CODEC_ID_MPEG4:
2102         if(CONFIG_MPEG4_ENCODER && s->partitioned_frame)
2103             ff_mpeg4_init_partitions(s);
2104         break;
2105     }
2106
2107     s->resync_mb_x=0;
2108     s->resync_mb_y=0;
2109     s->first_slice_line = 1;
2110     s->ptr_lastgob = s->pb.buf;
2111     for(mb_y= s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
2112 //    printf("row %d at %X\n", s->mb_y, (int)s);
2113         s->mb_x=0;
2114         s->mb_y= mb_y;
2115
2116         ff_set_qscale(s, s->qscale);
2117         ff_init_block_index(s);
2118
2119         for(mb_x=0; mb_x < s->mb_width; mb_x++) {
2120             int xy= mb_y*s->mb_stride + mb_x; // removed const, H261 needs to adjust this
2121             int mb_type= s->mb_type[xy];
2122 //            int d;
2123             int dmin= INT_MAX;
2124             int dir;
2125
2126             if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < MAX_MB_BYTES){
2127                 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
2128                 return -1;
2129             }
2130             if(s->data_partitioning){
2131                 if(   s->pb2   .buf_end - s->pb2   .buf - (put_bits_count(&s->    pb2)>>3) < MAX_MB_BYTES
2132                    || s->tex_pb.buf_end - s->tex_pb.buf - (put_bits_count(&s->tex_pb )>>3) < MAX_MB_BYTES){
2133                     av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
2134                     return -1;
2135                 }
2136             }
2137
2138             s->mb_x = mb_x;
2139             s->mb_y = mb_y;  // moved into loop, can get changed by H.261
2140             ff_update_block_index(s);
2141
2142             if(CONFIG_H261_ENCODER && s->codec_id == CODEC_ID_H261){
2143                 ff_h261_reorder_mb_index(s);
2144                 xy= s->mb_y*s->mb_stride + s->mb_x;
2145                 mb_type= s->mb_type[xy];
2146             }
2147
2148             /* write gob / video packet header  */
2149             if(s->rtp_mode){
2150                 int current_packet_size, is_gob_start;
2151
2152                 current_packet_size= ((put_bits_count(&s->pb)+7)>>3) - (s->ptr_lastgob - s->pb.buf);
2153
2154                 is_gob_start= s->avctx->rtp_payload_size && current_packet_size >= s->avctx->rtp_payload_size && mb_y + mb_x>0;
2155
2156                 if(s->start_mb_y == mb_y && mb_y > 0 && mb_x==0) is_gob_start=1;
2157
2158                 switch(s->codec_id){
2159                 case CODEC_ID_H263:
2160                 case CODEC_ID_H263P:
2161                     if(!s->h263_slice_structured)
2162                         if(s->mb_x || s->mb_y%s->gob_index) is_gob_start=0;
2163                     break;
2164                 case CODEC_ID_MPEG2VIDEO:
2165                     if(s->mb_x==0 && s->mb_y!=0) is_gob_start=1;
2166                 case CODEC_ID_MPEG1VIDEO:
2167                     if(s->mb_skip_run) is_gob_start=0;
2168                     break;
2169                 }
2170
2171                 if(is_gob_start){
2172                     if(s->start_mb_y != mb_y || mb_x!=0){
2173                         write_slice_end(s);
2174
2175                         if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4 && s->partitioned_frame){
2176                             ff_mpeg4_init_partitions(s);
2177                         }
2178                     }
2179
2180                     assert((put_bits_count(&s->pb)&7) == 0);
2181                     current_packet_size= put_bits_ptr(&s->pb) - s->ptr_lastgob;
2182
2183                     if(s->avctx->error_rate && s->resync_mb_x + s->resync_mb_y > 0){
2184                         int r= put_bits_count(&s->pb)/8 + s->picture_number + 16 + s->mb_x + s->mb_y;
2185                         int d= 100 / s->avctx->error_rate;
2186                         if(r % d == 0){
2187                             current_packet_size=0;
2188                             s->pb.buf_ptr= s->ptr_lastgob;
2189                             assert(put_bits_ptr(&s->pb) == s->ptr_lastgob);
2190                         }
2191                     }
2192
2193                     if (s->avctx->rtp_callback){
2194                         int number_mb = (mb_y - s->resync_mb_y)*s->mb_width + mb_x - s->resync_mb_x;
2195                         s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, current_packet_size, number_mb);
2196                     }
2197
2198                     switch(s->codec_id){
2199                     case CODEC_ID_MPEG4:
2200                         if (CONFIG_MPEG4_ENCODER) {
2201                             ff_mpeg4_encode_video_packet_header(s);
2202                             ff_mpeg4_clean_buffers(s);
2203                         }
2204                     break;
2205                     case CODEC_ID_MPEG1VIDEO:
2206                     case CODEC_ID_MPEG2VIDEO:
2207                         if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER) {
2208                             ff_mpeg1_encode_slice_header(s);
2209                             ff_mpeg1_clean_buffers(s);
2210                         }
2211                     break;
2212                     case CODEC_ID_H263:
2213                     case CODEC_ID_H263P:
2214                         if (CONFIG_H263_ENCODER)
2215                             h263_encode_gob_header(s, mb_y);
2216                     break;
2217                     }
2218
2219                     if(s->flags&CODEC_FLAG_PASS1){
2220                         int bits= put_bits_count(&s->pb);
2221                         s->misc_bits+= bits - s->last_bits;
2222                         s->last_bits= bits;
2223                     }
2224
2225                     s->ptr_lastgob += current_packet_size;
2226                     s->first_slice_line=1;
2227                     s->resync_mb_x=mb_x;
2228                     s->resync_mb_y=mb_y;
2229                 }
2230             }
2231
2232             if(  (s->resync_mb_x   == s->mb_x)
2233                && s->resync_mb_y+1 == s->mb_y){
2234                 s->first_slice_line=0;
2235             }
2236
2237             s->mb_skipped=0;
2238             s->dquant=0; //only for QP_RD
2239
2240             if(mb_type & (mb_type-1) || (s->flags & CODEC_FLAG_QP_RD)){ // more than 1 MB type possible or CODEC_FLAG_QP_RD
2241                 int next_block=0;
2242                 int pb_bits_count, pb2_bits_count, tex_pb_bits_count;
2243
2244                 copy_context_before_encode(&backup_s, s, -1);
2245                 backup_s.pb= s->pb;
2246                 best_s.data_partitioning= s->data_partitioning;
2247                 best_s.partitioned_frame= s->partitioned_frame;
2248                 if(s->data_partitioning){
2249                     backup_s.pb2= s->pb2;
2250                     backup_s.tex_pb= s->tex_pb;
2251                 }
2252
2253                 if(mb_type&CANDIDATE_MB_TYPE_INTER){
2254                     s->mv_dir = MV_DIR_FORWARD;
2255                     s->mv_type = MV_TYPE_16X16;
2256                     s->mb_intra= 0;
2257                     s->mv[0][0][0] = s->p_mv_table[xy][0];
2258                     s->mv[0][0][1] = s->p_mv_table[xy][1];
2259                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER, pb, pb2, tex_pb,
2260                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2261                 }
2262                 if(mb_type&CANDIDATE_MB_TYPE_INTER_I){
2263                     s->mv_dir = MV_DIR_FORWARD;
2264                     s->mv_type = MV_TYPE_FIELD;
2265                     s->mb_intra= 0;
2266                     for(i=0; i<2; i++){
2267                         j= s->field_select[0][i] = s->p_field_select_table[i][xy];
2268                         s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
2269                         s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
2270                     }
2271                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER_I, pb, pb2, tex_pb,
2272                                  &dmin, &next_block, 0, 0);
2273                 }
2274                 if(mb_type&CANDIDATE_MB_TYPE_SKIPPED){
2275                     s->mv_dir = MV_DIR_FORWARD;
2276                     s->mv_type = MV_TYPE_16X16;
2277                     s->mb_intra= 0;
2278                     s->mv[0][0][0] = 0;
2279                     s->mv[0][0][1] = 0;
2280                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_SKIPPED, pb, pb2, tex_pb,
2281                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2282                 }
2283                 if(mb_type&CANDIDATE_MB_TYPE_INTER4V){
2284                     s->mv_dir = MV_DIR_FORWARD;
2285                     s->mv_type = MV_TYPE_8X8;
2286                     s->mb_intra= 0;
2287                     for(i=0; i<4; i++){
2288                         s->mv[0][i][0] = s->current_picture.f.motion_val[0][s->block_index[i]][0];
2289                         s->mv[0][i][1] = s->current_picture.f.motion_val[0][s->block_index[i]][1];
2290                     }
2291                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER4V, pb, pb2, tex_pb,
2292                                  &dmin, &next_block, 0, 0);
2293                 }
2294                 if(mb_type&CANDIDATE_MB_TYPE_FORWARD){
2295                     s->mv_dir = MV_DIR_FORWARD;
2296                     s->mv_type = MV_TYPE_16X16;
2297                     s->mb_intra= 0;
2298                     s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
2299                     s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2300                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD, pb, pb2, tex_pb,
2301                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2302                 }
2303                 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD){
2304                     s->mv_dir = MV_DIR_BACKWARD;
2305                     s->mv_type = MV_TYPE_16X16;
2306                     s->mb_intra= 0;
2307                     s->mv[1][0][0] = s->b_back_mv_table[xy][0];
2308                     s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2309                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD, pb, pb2, tex_pb,
2310                                  &dmin, &next_block, s->mv[1][0][0], s->mv[1][0][1]);
2311                 }
2312                 if(mb_type&CANDIDATE_MB_TYPE_BIDIR){
2313                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2314                     s->mv_type = MV_TYPE_16X16;
2315                     s->mb_intra= 0;
2316                     s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
2317                     s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
2318                     s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
2319                     s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2320                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR, pb, pb2, tex_pb,
2321                                  &dmin, &next_block, 0, 0);
2322                 }
2323                 if(mb_type&CANDIDATE_MB_TYPE_FORWARD_I){
2324                     s->mv_dir = MV_DIR_FORWARD;
2325                     s->mv_type = MV_TYPE_FIELD;
2326                     s->mb_intra= 0;
2327                     for(i=0; i<2; i++){
2328                         j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
2329                         s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
2330                         s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
2331                     }
2332                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD_I, pb, pb2, tex_pb,
2333                                  &dmin, &next_block, 0, 0);
2334                 }
2335                 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD_I){
2336                     s->mv_dir = MV_DIR_BACKWARD;
2337                     s->mv_type = MV_TYPE_FIELD;
2338                     s->mb_intra= 0;
2339                     for(i=0; i<2; i++){
2340                         j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
2341                         s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
2342                         s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
2343                     }
2344                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD_I, pb, pb2, tex_pb,
2345                                  &dmin, &next_block, 0, 0);
2346                 }
2347                 if(mb_type&CANDIDATE_MB_TYPE_BIDIR_I){
2348                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2349                     s->mv_type = MV_TYPE_FIELD;
2350                     s->mb_intra= 0;
2351                     for(dir=0; dir<2; dir++){
2352                         for(i=0; i<2; i++){
2353                             j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
2354                             s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
2355                             s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
2356                         }
2357                     }
2358                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR_I, pb, pb2, tex_pb,
2359                                  &dmin, &next_block, 0, 0);
2360                 }
2361                 if(mb_type&CANDIDATE_MB_TYPE_INTRA){
2362                     s->mv_dir = 0;
2363                     s->mv_type = MV_TYPE_16X16;
2364                     s->mb_intra= 1;
2365                     s->mv[0][0][0] = 0;
2366                     s->mv[0][0][1] = 0;
2367                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTRA, pb, pb2, tex_pb,
2368                                  &dmin, &next_block, 0, 0);
2369                     if(s->h263_pred || s->h263_aic){
2370                         if(best_s.mb_intra)
2371                             s->mbintra_table[mb_x + mb_y*s->mb_stride]=1;
2372                         else
2373                             ff_clean_intra_table_entries(s); //old mode?
2374                     }
2375                 }
2376
2377                 if((s->flags & CODEC_FLAG_QP_RD) && dmin < INT_MAX){
2378                     if(best_s.mv_type==MV_TYPE_16X16){ //FIXME move 4mv after QPRD
2379                         const int last_qp= backup_s.qscale;
2380                         int qpi, qp, dc[6];
2381                         DCTELEM ac[6][16];
2382                         const int mvdir= (best_s.mv_dir&MV_DIR_BACKWARD) ? 1 : 0;
2383                         static const int dquant_tab[4]={-1,1,-2,2};
2384
2385                         assert(backup_s.dquant == 0);
2386
2387                         //FIXME intra
2388                         s->mv_dir= best_s.mv_dir;
2389                         s->mv_type = MV_TYPE_16X16;
2390                         s->mb_intra= best_s.mb_intra;
2391                         s->mv[0][0][0] = best_s.mv[0][0][0];
2392                         s->mv[0][0][1] = best_s.mv[0][0][1];
2393                         s->mv[1][0][0] = best_s.mv[1][0][0];
2394                         s->mv[1][0][1] = best_s.mv[1][0][1];
2395
2396                         qpi = s->pict_type == AV_PICTURE_TYPE_B ? 2 : 0;
2397                         for(; qpi<4; qpi++){
2398                             int dquant= dquant_tab[qpi];
2399                             qp= last_qp + dquant;
2400                             if(qp < s->avctx->qmin || qp > s->avctx->qmax)
2401                                 continue;
2402                             backup_s.dquant= dquant;
2403                             if(s->mb_intra && s->dc_val[0]){
2404                                 for(i=0; i<6; i++){
2405                                     dc[i]= s->dc_val[0][ s->block_index[i] ];
2406                                     memcpy(ac[i], s->ac_val[0][s->block_index[i]], sizeof(DCTELEM)*16);
2407                                 }
2408                             }
2409
2410                             encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
2411                                          &dmin, &next_block, s->mv[mvdir][0][0], s->mv[mvdir][0][1]);
2412                             if(best_s.qscale != qp){
2413                                 if(s->mb_intra && s->dc_val[0]){
2414                                     for(i=0; i<6; i++){
2415                                         s->dc_val[0][ s->block_index[i] ]= dc[i];
2416                                         memcpy(s->ac_val[0][s->block_index[i]], ac[i], sizeof(DCTELEM)*16);
2417                                     }
2418                                 }
2419                             }
2420                         }
2421                     }
2422                 }
2423                 if(CONFIG_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT){
2424                     int mx= s->b_direct_mv_table[xy][0];
2425                     int my= s->b_direct_mv_table[xy][1];
2426
2427                     backup_s.dquant = 0;
2428                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2429                     s->mb_intra= 0;
2430                     ff_mpeg4_set_direct_mv(s, mx, my);
2431                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
2432                                  &dmin, &next_block, mx, my);
2433                 }
2434                 if(CONFIG_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT0){
2435                     backup_s.dquant = 0;
2436                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2437                     s->mb_intra= 0;
2438                     ff_mpeg4_set_direct_mv(s, 0, 0);
2439                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
2440                                  &dmin, &next_block, 0, 0);
2441                 }
2442                 if(!best_s.mb_intra && s->flags2&CODEC_FLAG2_SKIP_RD){
2443                     int coded=0;
2444                     for(i=0; i<6; i++)
2445                         coded |= s->block_last_index[i];
2446                     if(coded){
2447                         int mx,my;
2448                         memcpy(s->mv, best_s.mv, sizeof(s->mv));
2449                         if(CONFIG_MPEG4_ENCODER && best_s.mv_dir & MV_DIRECT){
2450                             mx=my=0; //FIXME find the one we actually used
2451                             ff_mpeg4_set_direct_mv(s, mx, my);
2452                         }else if(best_s.mv_dir&MV_DIR_BACKWARD){
2453                             mx= s->mv[1][0][0];
2454                             my= s->mv[1][0][1];
2455                         }else{
2456                             mx= s->mv[0][0][0];
2457                             my= s->mv[0][0][1];
2458                         }
2459
2460                         s->mv_dir= best_s.mv_dir;
2461                         s->mv_type = best_s.mv_type;
2462                         s->mb_intra= 0;
2463 /*                        s->mv[0][0][0] = best_s.mv[0][0][0];
2464                         s->mv[0][0][1] = best_s.mv[0][0][1];
2465                         s->mv[1][0][0] = best_s.mv[1][0][0];
2466                         s->mv[1][0][1] = best_s.mv[1][0][1];*/
2467                         backup_s.dquant= 0;
2468                         s->skipdct=1;
2469                         encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
2470                                         &dmin, &next_block, mx, my);
2471                         s->skipdct=0;
2472                     }
2473                 }
2474
2475                 s->current_picture.f.qscale_table[xy] = best_s.qscale;
2476
2477                 copy_context_after_encode(s, &best_s, -1);
2478
2479                 pb_bits_count= put_bits_count(&s->pb);
2480                 flush_put_bits(&s->pb);
2481                 ff_copy_bits(&backup_s.pb, bit_buf[next_block^1], pb_bits_count);
2482                 s->pb= backup_s.pb;
2483
2484                 if(s->data_partitioning){
2485                     pb2_bits_count= put_bits_count(&s->pb2);
2486                     flush_put_bits(&s->pb2);
2487                     ff_copy_bits(&backup_s.pb2, bit_buf2[next_block^1], pb2_bits_count);
2488                     s->pb2= backup_s.pb2;
2489
2490                     tex_pb_bits_count= put_bits_count(&s->tex_pb);
2491                     flush_put_bits(&s->tex_pb);
2492                     ff_copy_bits(&backup_s.tex_pb, bit_buf_tex[next_block^1], tex_pb_bits_count);
2493                     s->tex_pb= backup_s.tex_pb;
2494                 }
2495                 s->last_bits= put_bits_count(&s->pb);
2496
2497                 if (CONFIG_H263_ENCODER &&
2498                     s->out_format == FMT_H263 && s->pict_type!=AV_PICTURE_TYPE_B)
2499                     ff_h263_update_motion_val(s);
2500
2501                 if(next_block==0){ //FIXME 16 vs linesize16
2502                     s->dsp.put_pixels_tab[0][0](s->dest[0], s->rd_scratchpad                     , s->linesize  ,16);
2503                     s->dsp.put_pixels_tab[1][0](s->dest[1], s->rd_scratchpad + 16*s->linesize    , s->uvlinesize, 8);
2504                     s->dsp.put_pixels_tab[1][0](s->dest[2], s->rd_scratchpad + 16*s->linesize + 8, s->uvlinesize, 8);
2505                 }
2506
2507                 if(s->avctx->mb_decision == FF_MB_DECISION_BITS)
2508                     MPV_decode_mb(s, s->block);
2509             } else {
2510                 int motion_x = 0, motion_y = 0;
2511                 s->mv_type=MV_TYPE_16X16;
2512                 // only one MB-Type possible
2513
2514                 switch(mb_type){
2515                 case CANDIDATE_MB_TYPE_INTRA:
2516                     s->mv_dir = 0;
2517                     s->mb_intra= 1;
2518                     motion_x= s->mv[0][0][0] = 0;
2519                     motion_y= s->mv[0][0][1] = 0;
2520                     break;
2521                 case CANDIDATE_MB_TYPE_INTER:
2522                     s->mv_dir = MV_DIR_FORWARD;
2523                     s->mb_intra= 0;
2524                     motion_x= s->mv[0][0][0] = s->p_mv_table[xy][0];
2525                     motion_y= s->mv[0][0][1] = s->p_mv_table[xy][1];
2526                     break;
2527                 case CANDIDATE_MB_TYPE_INTER_I:
2528                     s->mv_dir = MV_DIR_FORWARD;
2529                     s->mv_type = MV_TYPE_FIELD;
2530                     s->mb_intra= 0;
2531                     for(i=0; i<2; i++){
2532                         j= s->field_select[0][i] = s->p_field_select_table[i][xy];
2533                         s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
2534                         s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
2535                     }
2536                     break;
2537                 case CANDIDATE_MB_TYPE_INTER4V:
2538                     s->mv_dir = MV_DIR_FORWARD;
2539                     s->mv_type = MV_TYPE_8X8;
2540                     s->mb_intra= 0;
2541                     for(i=0; i<4; i++){
2542                         s->mv[0][i][0] = s->current_picture.f.motion_val[0][s->block_index[i]][0];
2543                         s->mv[0][i][1] = s->current_picture.f.motion_val[0][s->block_index[i]][1];
2544                     }
2545                     break;
2546                 case CANDIDATE_MB_TYPE_DIRECT:
2547                     if (CONFIG_MPEG4_ENCODER) {
2548                         s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
2549                         s->mb_intra= 0;
2550                         motion_x=s->b_direct_mv_table[xy][0];
2551                         motion_y=s->b_direct_mv_table[xy][1];
2552                         ff_mpeg4_set_direct_mv(s, motion_x, motion_y);
2553                     }
2554                     break;
2555                 case CANDIDATE_MB_TYPE_DIRECT0:
2556                     if (CONFIG_MPEG4_ENCODER) {
2557                         s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
2558                         s->mb_intra= 0;
2559                         ff_mpeg4_set_direct_mv(s, 0, 0);
2560                     }
2561                     break;
2562                 case CANDIDATE_MB_TYPE_BIDIR:
2563                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2564                     s->mb_intra= 0;
2565                     s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
2566                     s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
2567                     s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
2568                     s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2569                     break;
2570                 case CANDIDATE_MB_TYPE_BACKWARD:
2571                     s->mv_dir = MV_DIR_BACKWARD;
2572                     s->mb_intra= 0;
2573                     motion_x= s->mv[1][0][0] = s->b_back_mv_table[xy][0];
2574                     motion_y= s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2575                     break;
2576                 case CANDIDATE_MB_TYPE_FORWARD:
2577                     s->mv_dir = MV_DIR_FORWARD;
2578                     s->mb_intra= 0;
2579                     motion_x= s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
2580                     motion_y= s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2581 //                    printf(" %d %d ", motion_x, motion_y);
2582                     break;
2583                 case CANDIDATE_MB_TYPE_FORWARD_I:
2584                     s->mv_dir = MV_DIR_FORWARD;
2585                     s->mv_type = MV_TYPE_FIELD;
2586                     s->mb_intra= 0;
2587                     for(i=0; i<2; i++){
2588                         j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
2589                         s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
2590                         s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
2591                     }
2592                     break;
2593                 case CANDIDATE_MB_TYPE_BACKWARD_I:
2594                     s->mv_dir = MV_DIR_BACKWARD;
2595                     s->mv_type = MV_TYPE_FIELD;
2596                     s->mb_intra= 0;
2597                     for(i=0; i<2; i++){
2598                         j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
2599                         s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
2600                         s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
2601                     }
2602                     break;
2603                 case CANDIDATE_MB_TYPE_BIDIR_I:
2604                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2605                     s->mv_type = MV_TYPE_FIELD;
2606                     s->mb_intra= 0;
2607                     for(dir=0; dir<2; dir++){
2608                         for(i=0; i<2; i++){
2609                             j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
2610                             s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
2611                             s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
2612                         }
2613                     }
2614                     break;
2615                 default:
2616                     av_log(s->avctx, AV_LOG_ERROR, "illegal MB type\n");
2617                 }
2618
2619                 encode_mb(s, motion_x, motion_y);
2620
2621                 // RAL: Update last macroblock type
2622                 s->last_mv_dir = s->mv_dir;
2623
2624                 if (CONFIG_H263_ENCODER &&
2625                     s->out_format == FMT_H263 && s->pict_type!=AV_PICTURE_TYPE_B)
2626                     ff_h263_update_motion_val(s);
2627
2628                 MPV_decode_mb(s, s->block);
2629             }
2630
2631             /* clean the MV table in IPS frames for direct mode in B frames */
2632             if(s->mb_intra /* && I,P,S_TYPE */){
2633                 s->p_mv_table[xy][0]=0;
2634                 s->p_mv_table[xy][1]=0;
2635             }
2636
2637             if(s->flags&CODEC_FLAG_PSNR){
2638                 int w= 16;
2639                 int h= 16;
2640
2641                 if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
2642                 if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
2643
2644                 s->current_picture.f.error[0] += sse(
2645                     s, s->new_picture.f.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16,
2646                     s->dest[0], w, h, s->linesize);
2647                 s->current_picture.f.error[1] += sse(
2648                     s, s->new_picture.f.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*chr_h,
2649                     s->dest[1], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
2650                 s->current_picture.f.error[2] += sse(
2651                     s, s->new_picture.f.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*chr_h,
2652                     s->dest[2], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
2653             }
2654             if(s->loop_filter){
2655                 if(CONFIG_H263_ENCODER && s->out_format == FMT_H263)
2656                     ff_h263_loop_filter(s);
2657             }
2658 //printf("MB %d %d bits\n", s->mb_x+s->mb_y*s->mb_stride, put_bits_count(&s->pb));
2659         }
2660     }
2661
2662     //not beautiful here but we must write it before flushing so it has to be here
2663     if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version && s->msmpeg4_version<4 && s->pict_type == AV_PICTURE_TYPE_I)
2664         msmpeg4_encode_ext_header(s);
2665
2666     write_slice_end(s);
2667
2668     /* Send the last GOB if RTP */
2669     if (s->avctx->rtp_callback) {
2670         int number_mb = (mb_y - s->resync_mb_y)*s->mb_width - s->resync_mb_x;
2671         pdif = put_bits_ptr(&s->pb) - s->ptr_lastgob;
2672         /* Call the RTP callback to send the last GOB */
2673         emms_c();
2674         s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, pdif, number_mb);
2675     }
2676
2677     return 0;
2678 }
2679
2680 #define MERGE(field) dst->field += src->field; src->field=0
2681 static void merge_context_after_me(MpegEncContext *dst, MpegEncContext *src){
2682     MERGE(me.scene_change_score);
2683     MERGE(me.mc_mb_var_sum_temp);
2684     MERGE(me.mb_var_sum_temp);
2685 }
2686
2687 static void merge_context_after_encode(MpegEncContext *dst, MpegEncContext *src){
2688     int i;
2689
2690     MERGE(dct_count[0]); //note, the other dct vars are not part of the context
2691     MERGE(dct_count[1]);
2692     MERGE(mv_bits);
2693     MERGE(i_tex_bits);
2694     MERGE(p_tex_bits);
2695     MERGE(i_count);
2696     MERGE(f_count);
2697     MERGE(b_count);
2698     MERGE(skip_count);
2699     MERGE(misc_bits);
2700     MERGE(error_count);
2701     MERGE(padding_bug_score);
2702     MERGE(current_picture.f.error[0]);
2703     MERGE(current_picture.f.error[1]);
2704     MERGE(current_picture.f.error[2]);
2705
2706     if(dst->avctx->noise_reduction){
2707         for(i=0; i<64; i++){
2708             MERGE(dct_error_sum[0][i]);
2709             MERGE(dct_error_sum[1][i]);
2710         }
2711     }
2712
2713     assert(put_bits_count(&src->pb) % 8 ==0);
2714     assert(put_bits_count(&dst->pb) % 8 ==0);
2715     ff_copy_bits(&dst->pb, src->pb.buf, put_bits_count(&src->pb));
2716     flush_put_bits(&dst->pb);
2717 }
2718
2719 static int estimate_qp(MpegEncContext *s, int dry_run){
2720     if (s->next_lambda){
2721         s->current_picture_ptr->f.quality =
2722         s->current_picture.f.quality = s->next_lambda;
2723         if(!dry_run) s->next_lambda= 0;
2724     } else if (!s->fixed_qscale) {
2725         s->current_picture_ptr->f.quality =
2726         s->current_picture.f.quality = ff_rate_estimate_qscale(s, dry_run);
2727         if (s->current_picture.f.quality < 0)
2728             return -1;
2729     }
2730
2731     if(s->adaptive_quant){
2732         switch(s->codec_id){
2733         case CODEC_ID_MPEG4:
2734             if (CONFIG_MPEG4_ENCODER)
2735                 ff_clean_mpeg4_qscales(s);
2736             break;
2737         case CODEC_ID_H263:
2738         case CODEC_ID_H263P:
2739         case CODEC_ID_FLV1:
2740             if (CONFIG_H263_ENCODER)
2741                 ff_clean_h263_qscales(s);
2742             break;
2743         default:
2744             ff_init_qscale_tab(s);
2745         }
2746
2747         s->lambda= s->lambda_table[0];
2748         //FIXME broken
2749     }else
2750         s->lambda = s->current_picture.f.quality;
2751 //printf("%d %d\n", s->avctx->global_quality, s->current_picture.quality);
2752     update_qscale(s);
2753     return 0;
2754 }
2755
2756 /* must be called before writing the header */
2757 static void set_frame_distances(MpegEncContext * s){
2758     assert(s->current_picture_ptr->pts != AV_NOPTS_VALUE);
2759     s->time = s->current_picture_ptr->f.pts * s->avctx->time_base.num;
2760
2761     if(s->pict_type==AV_PICTURE_TYPE_B){
2762         s->pb_time= s->pp_time - (s->last_non_b_time - s->time);
2763         assert(s->pb_time > 0 && s->pb_time < s->pp_time);
2764     }else{
2765         s->pp_time= s->time - s->last_non_b_time;
2766         s->last_non_b_time= s->time;
2767         assert(s->picture_number==0 || s->pp_time > 0);
2768     }
2769 }
2770
2771 static int encode_picture(MpegEncContext *s, int picture_number)
2772 {
2773     int i;
2774     int bits;
2775     int context_count = s->avctx->thread_count;
2776
2777     s->picture_number = picture_number;
2778
2779     /* Reset the average MB variance */
2780     s->me.mb_var_sum_temp    =
2781     s->me.mc_mb_var_sum_temp = 0;
2782
2783     /* we need to initialize some time vars before we can encode b-frames */
2784     // RAL: Condition added for MPEG1VIDEO
2785     if (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO || (s->h263_pred && !s->msmpeg4_version))
2786         set_frame_distances(s);
2787     if(CONFIG_MPEG4_ENCODER && s->codec_id == CODEC_ID_MPEG4)
2788         ff_set_mpeg4_time(s);
2789
2790     s->me.scene_change_score=0;
2791
2792 //    s->lambda= s->current_picture_ptr->quality; //FIXME qscale / ... stuff for ME rate distortion
2793
2794     if(s->pict_type==AV_PICTURE_TYPE_I){
2795         if(s->msmpeg4_version >= 3) s->no_rounding=1;
2796         else                        s->no_rounding=0;
2797     }else if(s->pict_type!=AV_PICTURE_TYPE_B){
2798         if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
2799             s->no_rounding ^= 1;
2800     }
2801
2802     if(s->flags & CODEC_FLAG_PASS2){
2803         if (estimate_qp(s,1) < 0)
2804             return -1;
2805         ff_get_2pass_fcode(s);
2806     }else if(!(s->flags & CODEC_FLAG_QSCALE)){
2807         if(s->pict_type==AV_PICTURE_TYPE_B)
2808             s->lambda= s->last_lambda_for[s->pict_type];
2809         else
2810             s->lambda= s->last_lambda_for[s->last_non_b_pict_type];
2811         update_qscale(s);
2812     }
2813
2814     s->mb_intra=0; //for the rate distortion & bit compare functions
2815     for(i=1; i<context_count; i++){
2816         ff_update_duplicate_context(s->thread_context[i], s);
2817     }
2818
2819     if(ff_init_me(s)<0)
2820         return -1;
2821
2822     /* Estimate motion for every MB */
2823     if(s->pict_type != AV_PICTURE_TYPE_I){
2824         s->lambda = (s->lambda * s->avctx->me_penalty_compensation + 128)>>8;
2825         s->lambda2= (s->lambda2* (int64_t)s->avctx->me_penalty_compensation + 128)>>8;
2826         if(s->pict_type != AV_PICTURE_TYPE_B && s->avctx->me_threshold==0){
2827             if((s->avctx->pre_me && s->last_non_b_pict_type==AV_PICTURE_TYPE_I) || s->avctx->pre_me==2){
2828                 s->avctx->execute(s->avctx, pre_estimate_motion_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
2829             }
2830         }
2831
2832         s->avctx->execute(s->avctx, estimate_motion_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
2833     }else /* if(s->pict_type == AV_PICTURE_TYPE_I) */{
2834         /* I-Frame */
2835         for(i=0; i<s->mb_stride*s->mb_height; i++)
2836             s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
2837
2838         if(!s->fixed_qscale){
2839             /* finding spatial complexity for I-frame rate control */
2840             s->avctx->execute(s->avctx, mb_var_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
2841         }
2842     }
2843     for(i=1; i<context_count; i++){
2844         merge_context_after_me(s, s->thread_context[i]);
2845     }
2846     s->current_picture.mc_mb_var_sum= s->current_picture_ptr->mc_mb_var_sum= s->me.mc_mb_var_sum_temp;
2847     s->current_picture.   mb_var_sum= s->current_picture_ptr->   mb_var_sum= s->me.   mb_var_sum_temp;
2848     emms_c();
2849
2850     if(s->me.scene_change_score > s->avctx->scenechange_threshold && s->pict_type == AV_PICTURE_TYPE_P){
2851         s->pict_type= AV_PICTURE_TYPE_I;
2852         for(i=0; i<s->mb_stride*s->mb_height; i++)
2853             s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
2854 //printf("Scene change detected, encoding as I Frame %d %d\n", s->current_picture.mb_var_sum, s->current_picture.mc_mb_var_sum);
2855     }
2856
2857     if(!s->umvplus){
2858         if(s->pict_type==AV_PICTURE_TYPE_P || s->pict_type==AV_PICTURE_TYPE_S) {
2859             s->f_code= ff_get_best_fcode(s, s->p_mv_table, CANDIDATE_MB_TYPE_INTER);
2860
2861             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2862                 int a,b;
2863                 a= ff_get_best_fcode(s, s->p_field_mv_table[0][0], CANDIDATE_MB_TYPE_INTER_I); //FIXME field_select
2864                 b= ff_get_best_fcode(s, s->p_field_mv_table[1][1], CANDIDATE_MB_TYPE_INTER_I);
2865                 s->f_code= FFMAX3(s->f_code, a, b);
2866             }
2867
2868             ff_fix_long_p_mvs(s);
2869             ff_fix_long_mvs(s, NULL, 0, s->p_mv_table, s->f_code, CANDIDATE_MB_TYPE_INTER, 0);
2870             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2871                 int j;
2872                 for(i=0; i<2; i++){
2873                     for(j=0; j<2; j++)
2874                         ff_fix_long_mvs(s, s->p_field_select_table[i], j,
2875                                         s->p_field_mv_table[i][j], s->f_code, CANDIDATE_MB_TYPE_INTER_I, 0);
2876                 }
2877             }
2878         }
2879
2880         if(s->pict_type==AV_PICTURE_TYPE_B){
2881             int a, b;
2882
2883             a = ff_get_best_fcode(s, s->b_forw_mv_table, CANDIDATE_MB_TYPE_FORWARD);
2884             b = ff_get_best_fcode(s, s->b_bidir_forw_mv_table, CANDIDATE_MB_TYPE_BIDIR);
2885             s->f_code = FFMAX(a, b);
2886
2887             a = ff_get_best_fcode(s, s->b_back_mv_table, CANDIDATE_MB_TYPE_BACKWARD);
2888             b = ff_get_best_fcode(s, s->b_bidir_back_mv_table, CANDIDATE_MB_TYPE_BIDIR);
2889             s->b_code = FFMAX(a, b);
2890
2891             ff_fix_long_mvs(s, NULL, 0, s->b_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_FORWARD, 1);
2892             ff_fix_long_mvs(s, NULL, 0, s->b_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BACKWARD, 1);
2893             ff_fix_long_mvs(s, NULL, 0, s->b_bidir_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_BIDIR, 1);
2894             ff_fix_long_mvs(s, NULL, 0, s->b_bidir_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BIDIR, 1);
2895             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2896                 int dir, j;
2897                 for(dir=0; dir<2; dir++){
2898                     for(i=0; i<2; i++){
2899                         for(j=0; j<2; j++){
2900                             int type= dir ? (CANDIDATE_MB_TYPE_BACKWARD_I|CANDIDATE_MB_TYPE_BIDIR_I)
2901                                           : (CANDIDATE_MB_TYPE_FORWARD_I |CANDIDATE_MB_TYPE_BIDIR_I);
2902                             ff_fix_long_mvs(s, s->b_field_select_table[dir][i], j,
2903                                             s->b_field_mv_table[dir][i][j], dir ? s->b_code : s->f_code, type, 1);
2904                         }
2905                     }
2906                 }
2907             }
2908         }
2909     }
2910
2911     if (estimate_qp(s, 0) < 0)
2912         return -1;
2913
2914     if(s->qscale < 3 && s->max_qcoeff<=128 && s->pict_type==AV_PICTURE_TYPE_I && !(s->flags & CODEC_FLAG_QSCALE))
2915         s->qscale= 3; //reduce clipping problems
2916
2917     if (s->out_format == FMT_MJPEG) {
2918         /* for mjpeg, we do include qscale in the matrix */
2919         for(i=1;i<64;i++){
2920             int j= s->dsp.idct_permutation[i];
2921
2922             s->intra_matrix[j] = av_clip_uint8((ff_mpeg1_default_intra_matrix[i] * s->qscale) >> 3);
2923         }
2924         s->y_dc_scale_table=
2925         s->c_dc_scale_table= ff_mpeg2_dc_scale_table[s->intra_dc_precision];
2926         s->intra_matrix[0] = ff_mpeg2_dc_scale_table[s->intra_dc_precision][8];
2927         ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
2928                        s->intra_matrix, s->intra_quant_bias, 8, 8, 1);
2929         s->qscale= 8;
2930     }
2931
2932     //FIXME var duplication
2933     s->current_picture_ptr->f.key_frame =
2934     s->current_picture.f.key_frame = s->pict_type == AV_PICTURE_TYPE_I; //FIXME pic_ptr
2935     s->current_picture_ptr->f.pict_type =
2936     s->current_picture.f.pict_type = s->pict_type;
2937
2938     if (s->current_picture.f.key_frame)
2939         s->picture_in_gop_number=0;
2940
2941     s->last_bits= put_bits_count(&s->pb);
2942     switch(s->out_format) {
2943     case FMT_MJPEG:
2944         if (CONFIG_MJPEG_ENCODER)
2945             ff_mjpeg_encode_picture_header(s);
2946         break;
2947     case FMT_H261:
2948         if (CONFIG_H261_ENCODER)
2949             ff_h261_encode_picture_header(s, picture_number);
2950         break;
2951     case FMT_H263:
2952         if (CONFIG_WMV2_ENCODER && s->codec_id == CODEC_ID_WMV2)
2953             ff_wmv2_encode_picture_header(s, picture_number);
2954         else if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version)
2955             msmpeg4_encode_picture_header(s, picture_number);
2956         else if (CONFIG_MPEG4_ENCODER && s->h263_pred)
2957             mpeg4_encode_picture_header(s, picture_number);
2958         else if (CONFIG_RV10_ENCODER && s->codec_id == CODEC_ID_RV10)
2959             rv10_encode_picture_header(s, picture_number);
2960         else if (CONFIG_RV20_ENCODER && s->codec_id == CODEC_ID_RV20)
2961             rv20_encode_picture_header(s, picture_number);
2962         else if (CONFIG_FLV_ENCODER && s->codec_id == CODEC_ID_FLV1)
2963             ff_flv_encode_picture_header(s, picture_number);
2964         else if (CONFIG_H263_ENCODER)
2965             h263_encode_picture_header(s, picture_number);
2966         break;
2967     case FMT_MPEG1:
2968         if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
2969             mpeg1_encode_picture_header(s, picture_number);
2970         break;
2971     case FMT_H264:
2972         break;
2973     default:
2974         assert(0);
2975     }
2976     bits= put_bits_count(&s->pb);
2977     s->header_bits= bits - s->last_bits;
2978
2979     for(i=1; i<context_count; i++){
2980         update_duplicate_context_after_me(s->thread_context[i], s);
2981     }
2982     s->avctx->execute(s->avctx, encode_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
2983     for(i=1; i<context_count; i++){
2984         merge_context_after_encode(s, s->thread_context[i]);
2985     }
2986     emms_c();
2987     return 0;
2988 }
2989
2990 static void denoise_dct_c(MpegEncContext *s, DCTELEM *block){
2991     const int intra= s->mb_intra;
2992     int i;
2993
2994     s->dct_count[intra]++;
2995
2996     for(i=0; i<64; i++){
2997         int level= block[i];
2998
2999         if(level){
3000             if(level>0){
3001                 s->dct_error_sum[intra][i] += level;
3002                 level -= s->dct_offset[intra][i];
3003                 if(level<0) level=0;
3004             }else{
3005                 s->dct_error_sum[intra][i] -= level;
3006                 level += s->dct_offset[intra][i];
3007                 if(level>0) level=0;
3008             }
3009             block[i]= level;
3010         }
3011     }
3012 }
3013
3014 static int dct_quantize_trellis_c(MpegEncContext *s,
3015                                   DCTELEM *block, int n,
3016                                   int qscale, int *overflow){
3017     const int *qmat;
3018     const uint8_t *scantable= s->intra_scantable.scantable;
3019     const uint8_t *perm_scantable= s->intra_scantable.permutated;
3020     int max=0;
3021     unsigned int threshold1, threshold2;
3022     int bias=0;
3023     int run_tab[65];
3024     int level_tab[65];
3025     int score_tab[65];
3026     int survivor[65];
3027     int survivor_count;
3028     int last_run=0;
3029     int last_level=0;
3030     int last_score= 0;
3031     int last_i;
3032     int coeff[2][64];
3033     int coeff_count[64];
3034     int qmul, qadd, start_i, last_non_zero, i, dc;
3035     const int esc_length= s->ac_esc_length;
3036     uint8_t * length;
3037     uint8_t * last_length;
3038     const int lambda= s->lambda2 >> (FF_LAMBDA_SHIFT - 6);
3039
3040     s->dsp.fdct (block);
3041
3042     if(s->dct_error_sum)
3043         s->denoise_dct(s, block);
3044     qmul= qscale*16;
3045     qadd= ((qscale-1)|1)*8;
3046
3047     if (s->mb_intra) {
3048         int q;
3049         if (!s->h263_aic) {
3050             if (n < 4)
3051                 q = s->y_dc_scale;
3052             else
3053                 q = s->c_dc_scale;
3054             q = q << 3;
3055         } else{
3056             /* For AIC we skip quant/dequant of INTRADC */
3057             q = 1 << 3;
3058             qadd=0;
3059         }
3060
3061         /* note: block[0] is assumed to be positive */
3062         block[0] = (block[0] + (q >> 1)) / q;
3063         start_i = 1;
3064         last_non_zero = 0;
3065         qmat = s->q_intra_matrix[qscale];
3066         if(s->mpeg_quant || s->out_format == FMT_MPEG1)
3067             bias= 1<<(QMAT_SHIFT-1);
3068         length     = s->intra_ac_vlc_length;
3069         last_length= s->intra_ac_vlc_last_length;
3070     } else {
3071         start_i = 0;
3072         last_non_zero = -1;
3073         qmat = s->q_inter_matrix[qscale];
3074         length     = s->inter_ac_vlc_length;
3075         last_length= s->inter_ac_vlc_last_length;
3076     }
3077     last_i= start_i;
3078
3079     threshold1= (1<<QMAT_SHIFT) - bias - 1;
3080     threshold2= (threshold1<<1);
3081
3082     for(i=63; i>=start_i; i--) {
3083         const int j = scantable[i];
3084         int level = block[j] * qmat[j];
3085
3086         if(((unsigned)(level+threshold1))>threshold2){
3087             last_non_zero = i;
3088             break;
3089         }
3090     }
3091
3092     for(i=start_i; i<=last_non_zero; i++) {
3093         const int j = scantable[i];
3094         int level = block[j] * qmat[j];
3095
3096 //        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
3097 //           || bias-level >= (1<<(QMAT_SHIFT - 3))){
3098         if(((unsigned)(level+threshold1))>threshold2){
3099             if(level>0){
3100                 level= (bias + level)>>QMAT_SHIFT;
3101                 coeff[0][i]= level;
3102                 coeff[1][i]= level-1;
3103 //                coeff[2][k]= level-2;
3104             }else{
3105                 level= (bias - level)>>QMAT_SHIFT;
3106                 coeff[0][i]= -level;
3107                 coeff[1][i]= -level+1;
3108 //                coeff[2][k]= -level+2;
3109             }
3110             coeff_count[i]= FFMIN(level, 2);
3111             assert(coeff_count[i]);
3112             max |=level;
3113         }else{
3114             coeff[0][i]= (level>>31)|1;
3115             coeff_count[i]= 1;
3116         }
3117     }
3118
3119     *overflow= s->max_qcoeff < max; //overflow might have happened
3120
3121     if(last_non_zero < start_i){
3122         memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
3123         return last_non_zero;
3124     }
3125
3126     score_tab[start_i]= 0;
3127     survivor[0]= start_i;
3128     survivor_count= 1;
3129
3130     for(i=start_i; i<=last_non_zero; i++){
3131         int level_index, j, zero_distortion;
3132         int dct_coeff= FFABS(block[ scantable[i] ]);
3133         int best_score=256*256*256*120;
3134
3135         if (   s->dsp.fdct == fdct_ifast
3136 #ifndef FAAN_POSTSCALE
3137             || s->dsp.fdct == ff_faandct
3138 #endif
3139            )
3140             dct_coeff= (dct_coeff*ff_inv_aanscales[ scantable[i] ]) >> 12;
3141         zero_distortion= dct_coeff*dct_coeff;
3142
3143         for(level_index=0; level_index < coeff_count[i]; level_index++){
3144             int distortion;
3145             int level= coeff[level_index][i];
3146             const int alevel= FFABS(level);
3147             int unquant_coeff;
3148
3149             assert(level);
3150
3151             if(s->out_format == FMT_H263){
3152                 unquant_coeff= alevel*qmul + qadd;
3153             }else{ //MPEG1
3154                 j= s->dsp.idct_permutation[ scantable[i] ]; //FIXME optimize
3155                 if(s->mb_intra){
3156                         unquant_coeff = (int)(  alevel  * qscale * s->intra_matrix[j]) >> 3;
3157                         unquant_coeff =   (unquant_coeff - 1) | 1;
3158                 }else{
3159                         unquant_coeff = (((  alevel  << 1) + 1) * qscale * ((int) s->inter_matrix[j])) >> 4;
3160                         unquant_coeff =   (unquant_coeff - 1) | 1;
3161                 }
3162                 unquant_coeff<<= 3;
3163             }
3164
3165             distortion= (unquant_coeff - dct_coeff) * (unquant_coeff - dct_coeff) - zero_distortion;
3166             level+=64;
3167             if((level&(~127)) == 0){
3168                 for(j=survivor_count-1; j>=0; j--){
3169                     int run= i - survivor[j];
3170                     int score= distortion + length[UNI_AC_ENC_INDEX(run, level)]*lambda;
3171                     score += score_tab[i-run];
3172
3173                     if(score < best_score){
3174                         best_score= score;
3175                         run_tab[i+1]= run;
3176                         level_tab[i+1]= level-64;
3177                     }
3178                 }
3179
3180                 if(s->out_format == FMT_H263){
3181                     for(j=survivor_count-1; j>=0; j--){
3182                         int run= i - survivor[j];
3183                         int score= distortion + last_length[UNI_AC_ENC_INDEX(run, level)]*lambda;
3184                         score += score_tab[i-run];
3185                         if(score < last_score){
3186                             last_score= score;
3187                             last_run= run;
3188                             last_level= level-64;
3189                             last_i= i+1;
3190                         }
3191                     }
3192                 }
3193             }else{
3194                 distortion += esc_length*lambda;
3195                 for(j=survivor_count-1; j>=0; j--){
3196                     int run= i - survivor[j];
3197                     int score= distortion + score_tab[i-run];
3198
3199                     if(score < best_score){
3200                         best_score= score;
3201                         run_tab[i+1]= run;
3202                         level_tab[i+1]= level-64;
3203                     }
3204                 }
3205
3206                 if(s->out_format == FMT_H263){
3207                   for(j=survivor_count-1; j>=0; j--){
3208                         int run= i - survivor[j];
3209                         int score= distortion + score_tab[i-run];
3210                         if(score < last_score){
3211                             last_score= score;
3212                             last_run= run;
3213                             last_level= level-64;
3214                             last_i= i+1;
3215                         }
3216                     }
3217                 }
3218             }
3219         }
3220
3221         score_tab[i+1]= best_score;
3222
3223         //Note: there is a vlc code in mpeg4 which is 1 bit shorter then another one with a shorter run and the same level
3224         if(last_non_zero <= 27){
3225             for(; survivor_count; survivor_count--){
3226                 if(score_tab[ survivor[survivor_count-1] ] <= best_score)
3227                     break;
3228             }
3229         }else{
3230             for(; survivor_count; survivor_count--){
3231                 if(score_tab[ survivor[survivor_count-1] ] <= best_score + lambda)
3232                     break;
3233             }
3234         }
3235
3236         survivor[ survivor_count++ ]= i+1;
3237     }
3238
3239     if(s->out_format != FMT_H263){
3240         last_score= 256*256*256*120;
3241         for(i= survivor[0]; i<=last_non_zero + 1; i++){
3242             int score= score_tab[i];
3243             if(i) score += lambda*2; //FIXME exacter?
3244
3245             if(score < last_score){
3246                 last_score= score;
3247                 last_i= i;
3248                 last_level= level_tab[i];
3249                 last_run= run_tab[i];
3250             }
3251         }
3252     }
3253
3254     s->coded_score[n] = last_score;
3255
3256     dc= FFABS(block[0]);
3257     last_non_zero= last_i - 1;
3258     memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
3259
3260     if(last_non_zero < start_i)
3261         return last_non_zero;
3262
3263     if(last_non_zero == 0 && start_i == 0){
3264         int best_level= 0;
3265         int best_score= dc * dc;
3266
3267         for(i=0; i<coeff_count[0]; i++){
3268             int level= coeff[i][0];
3269             int alevel= FFABS(level);
3270             int unquant_coeff, score, distortion;
3271
3272             if(s->out_format == FMT_H263){
3273                     unquant_coeff= (alevel*qmul + qadd)>>3;
3274             }else{ //MPEG1
3275                     unquant_coeff = (((  alevel  << 1) + 1) * qscale * ((int) s->inter_matrix[0])) >> 4;
3276                     unquant_coeff =   (unquant_coeff - 1) | 1;
3277             }
3278             unquant_coeff = (unquant_coeff + 4) >> 3;
3279             unquant_coeff<<= 3 + 3;
3280
3281             distortion= (unquant_coeff - dc) * (unquant_coeff - dc);
3282             level+=64;
3283             if((level&(~127)) == 0) score= distortion + last_length[UNI_AC_ENC_INDEX(0, level)]*lambda;
3284             else                    score= distortion + esc_length*lambda;
3285
3286             if(score < best_score){
3287                 best_score= score;
3288                 best_level= level - 64;
3289             }
3290         }
3291         block[0]= best_level;
3292         s->coded_score[n] = best_score - dc*dc;
3293         if(best_level == 0) return -1;
3294         else                return last_non_zero;
3295     }
3296
3297     i= last_i;
3298     assert(last_level);
3299
3300     block[ perm_scantable[last_non_zero] ]= last_level;
3301     i -= last_run + 1;
3302
3303     for(; i>start_i; i -= run_tab[i] + 1){
3304         block[ perm_scantable[i-1] ]= level_tab[i];
3305     }
3306
3307     return last_non_zero;
3308 }
3309
3310 //#define REFINE_STATS 1
3311 static int16_t basis[64][64];
3312
3313 static void build_basis(uint8_t *perm){
3314     int i, j, x, y;
3315     emms_c();
3316     for(i=0; i<8; i++){
3317         for(j=0; j<8; j++){
3318             for(y=0; y<8; y++){
3319                 for(x=0; x<8; x++){
3320                     double s= 0.25*(1<<BASIS_SHIFT);
3321                     int index= 8*i + j;
3322                     int perm_index= perm[index];
3323                     if(i==0) s*= sqrt(0.5);
3324                     if(j==0) s*= sqrt(0.5);
3325                     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)));
3326                 }
3327             }
3328         }
3329     }
3330 }
3331
3332 static int dct_quantize_refine(MpegEncContext *s, //FIXME breaks denoise?
3333                         DCTELEM *block, int16_t *weight, DCTELEM *orig,
3334                         int n, int qscale){
3335     int16_t rem[64];
3336     LOCAL_ALIGNED_16(DCTELEM, d1, [64]);
3337     const uint8_t *scantable= s->intra_scantable.scantable;
3338     const uint8_t *perm_scantable= s->intra_scantable.permutated;
3339 //    unsigned int threshold1, threshold2;
3340 //    int bias=0;
3341     int run_tab[65];
3342     int prev_run=0;
3343     int prev_level=0;
3344     int qmul, qadd, start_i, last_non_zero, i, dc;
3345     uint8_t * length;
3346     uint8_t * last_length;
3347     int lambda;
3348     int rle_index, run, q = 1, sum; //q is only used when s->mb_intra is true
3349 #ifdef REFINE_STATS
3350 static int count=0;
3351 static int after_last=0;
3352 static int to_zero=0;
3353 static int from_zero=0;
3354 static int raise=0;
3355 static int lower=0;
3356 static int messed_sign=0;
3357 #endif
3358
3359     if(basis[0][0] == 0)
3360         build_basis(s->dsp.idct_permutation);
3361
3362     qmul= qscale*2;
3363     qadd= (qscale-1)|1;
3364     if (s->mb_intra) {
3365         if (!s->h263_aic) {
3366             if (n < 4)
3367                 q = s->y_dc_scale;
3368             else
3369                 q = s->c_dc_scale;