Remove casts that are useless since the argument is void *.
[ffmpeg.git] / libavcodec / mpegvideo_enc.c
1 /*
2  * The simplest mpeg encoder (well, it was the simplest!)
3  * Copyright (c) 2000,2001 Fabrice Bellard
4  * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
5  *
6  * 4MV & hq & B-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
7  *
8  * This file is part of FFmpeg.
9  *
10  * FFmpeg is free software; you can redistribute it and/or
11  * modify it under the terms of the GNU Lesser General Public
12  * License as published by the Free Software Foundation; either
13  * version 2.1 of the License, or (at your option) any later version.
14  *
15  * FFmpeg is distributed in the hope that it will be useful,
16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
18  * Lesser General Public License for more details.
19  *
20  * You should have received a copy of the GNU Lesser General Public
21  * License along with FFmpeg; if not, write to the Free Software
22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23  */
24
25 /**
26  * @file libavcodec/mpegvideo_enc.c
27  * The simplest mpeg encoder (well, it was the simplest!).
28  */
29
30 #include "avcodec.h"
31 #include "dsputil.h"
32 #include "mpegvideo.h"
33 #include "mpegvideo_common.h"
34 #include "mjpegenc.h"
35 #include "msmpeg4.h"
36 #include "h263.h"
37 #include "faandct.h"
38 #include "aandcttab.h"
39 #include <limits.h>
40
41 //#undef NDEBUG
42 //#include <assert.h>
43
44 static int encode_picture(MpegEncContext *s, int picture_number);
45 static int dct_quantize_refine(MpegEncContext *s, DCTELEM *block, int16_t *weight, DCTELEM *orig, int n, int qscale);
46 static int sse_mb(MpegEncContext *s);
47
48 /* enable all paranoid tests for rounding, overflows, etc... */
49 //#define PARANOID
50
51 //#define DEBUG
52
53 static uint8_t default_mv_penalty[MAX_FCODE+1][MAX_MV*2+1];
54 static uint8_t default_fcode_tab[MAX_MV*2+1];
55
56 void ff_convert_matrix(DSPContext *dsp, int (*qmat)[64], uint16_t (*qmat16)[2][64],
57                            const uint16_t *quant_matrix, int bias, int qmin, int qmax, int intra)
58 {
59     int qscale;
60     int shift=0;
61
62     for(qscale=qmin; qscale<=qmax; qscale++){
63         int i;
64         if (dsp->fdct == ff_jpeg_fdct_islow
65 #ifdef FAAN_POSTSCALE
66             || dsp->fdct == ff_faandct
67 #endif
68             ) {
69             for(i=0;i<64;i++) {
70                 const int j= dsp->idct_permutation[i];
71                 /* 16 <= qscale * quant_matrix[i] <= 7905 */
72                 /* 19952             <= ff_aanscales[i] * qscale * quant_matrix[i]               <= 249205026 */
73                 /* (1 << 36) / 19952 >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= (1 << 36) / 249205026 */
74                 /* 3444240           >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= 275 */
75
76                 qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) /
77                                 (qscale * quant_matrix[j]));
78             }
79         } else if (dsp->fdct == fdct_ifast
80 #ifndef FAAN_POSTSCALE
81                    || dsp->fdct == ff_faandct
82 #endif
83                    ) {
84             for(i=0;i<64;i++) {
85                 const int j= dsp->idct_permutation[i];
86                 /* 16 <= qscale * quant_matrix[i] <= 7905 */
87                 /* 19952             <= ff_aanscales[i] * qscale * quant_matrix[i]               <= 249205026 */
88                 /* (1 << 36) / 19952 >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */
89                 /* 3444240           >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= 275 */
90
91                 qmat[qscale][i] = (int)((UINT64_C(1) << (QMAT_SHIFT + 14)) /
92                                 (ff_aanscales[i] * qscale * quant_matrix[j]));
93             }
94         } else {
95             for(i=0;i<64;i++) {
96                 const int j= dsp->idct_permutation[i];
97                 /* We can safely suppose that 16 <= quant_matrix[i] <= 255
98                    So 16           <= qscale * quant_matrix[i]             <= 7905
99                    so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905
100                    so 32768        >= (1<<19) / (qscale * quant_matrix[i]) >= 67
101                 */
102                 qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) / (qscale * quant_matrix[j]));
103 //                qmat  [qscale][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]);
104                 qmat16[qscale][0][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[j]);
105
106                 if(qmat16[qscale][0][i]==0 || qmat16[qscale][0][i]==128*256) qmat16[qscale][0][i]=128*256-1;
107                 qmat16[qscale][1][i]= ROUNDED_DIV(bias<<(16-QUANT_BIAS_SHIFT), qmat16[qscale][0][i]);
108             }
109         }
110
111         for(i=intra; i<64; i++){
112             int64_t max= 8191;
113             if (dsp->fdct == fdct_ifast
114 #ifndef FAAN_POSTSCALE
115                    || dsp->fdct == ff_faandct
116 #endif
117                    ) {
118                 max = (8191LL*ff_aanscales[i]) >> 14;
119             }
120             while(((max * qmat[qscale][i]) >> shift) > INT_MAX){
121                 shift++;
122             }
123         }
124     }
125     if(shift){
126         av_log(NULL, AV_LOG_INFO, "Warning, QMAT_SHIFT is larger than %d, overflows possible\n", QMAT_SHIFT - shift);
127     }
128 }
129
130 static inline void update_qscale(MpegEncContext *s){
131     s->qscale= (s->lambda*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7);
132     s->qscale= av_clip(s->qscale, s->avctx->qmin, s->avctx->qmax);
133
134     s->lambda2= (s->lambda*s->lambda + FF_LAMBDA_SCALE/2) >> FF_LAMBDA_SHIFT;
135 }
136
137 void ff_write_quant_matrix(PutBitContext *pb, uint16_t *matrix){
138     int i;
139
140     if(matrix){
141         put_bits(pb, 1, 1);
142         for(i=0;i<64;i++) {
143             put_bits(pb, 8, matrix[ ff_zigzag_direct[i] ]);
144         }
145     }else
146         put_bits(pb, 1, 0);
147 }
148
149 static void copy_picture_attributes(MpegEncContext *s, AVFrame *dst, AVFrame *src){
150     int i;
151
152     dst->pict_type              = src->pict_type;
153     dst->quality                = src->quality;
154     dst->coded_picture_number   = src->coded_picture_number;
155     dst->display_picture_number = src->display_picture_number;
156 //    dst->reference              = src->reference;
157     dst->pts                    = src->pts;
158     dst->interlaced_frame       = src->interlaced_frame;
159     dst->top_field_first        = src->top_field_first;
160
161     if(s->avctx->me_threshold){
162         if(!src->motion_val[0])
163             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_val not set!\n");
164         if(!src->mb_type)
165             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.mb_type not set!\n");
166         if(!src->ref_index[0])
167             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.ref_index not set!\n");
168         if(src->motion_subsample_log2 != dst->motion_subsample_log2)
169             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_subsample_log2 doesn't match! (%d!=%d)\n",
170             src->motion_subsample_log2, dst->motion_subsample_log2);
171
172         memcpy(dst->mb_type, src->mb_type, s->mb_stride * s->mb_height * sizeof(dst->mb_type[0]));
173
174         for(i=0; i<2; i++){
175             int stride= ((16*s->mb_width )>>src->motion_subsample_log2) + 1;
176             int height= ((16*s->mb_height)>>src->motion_subsample_log2);
177
178             if(src->motion_val[i] && src->motion_val[i] != dst->motion_val[i]){
179                 memcpy(dst->motion_val[i], src->motion_val[i], 2*stride*height*sizeof(int16_t));
180             }
181             if(src->ref_index[i] && src->ref_index[i] != dst->ref_index[i]){
182                 memcpy(dst->ref_index[i], src->ref_index[i], s->b8_stride*2*s->mb_height*sizeof(int8_t));
183             }
184         }
185     }
186 }
187
188 static void update_duplicate_context_after_me(MpegEncContext *dst, MpegEncContext *src){
189 #define COPY(a) dst->a= src->a
190     COPY(pict_type);
191     COPY(current_picture);
192     COPY(f_code);
193     COPY(b_code);
194     COPY(qscale);
195     COPY(lambda);
196     COPY(lambda2);
197     COPY(picture_in_gop_number);
198     COPY(gop_picture_number);
199     COPY(frame_pred_frame_dct); //FIXME don't set in encode_header
200     COPY(progressive_frame); //FIXME don't set in encode_header
201     COPY(partitioned_frame); //FIXME don't set in encode_header
202 #undef COPY
203 }
204
205 /**
206  * sets the given MpegEncContext to defaults for encoding.
207  * the changed fields will not depend upon the prior state of the MpegEncContext.
208  */
209 static void MPV_encode_defaults(MpegEncContext *s){
210     int i;
211     MPV_common_defaults(s);
212
213     for(i=-16; i<16; i++){
214         default_fcode_tab[i + MAX_MV]= 1;
215     }
216     s->me.mv_penalty= default_mv_penalty;
217     s->fcode_tab= default_fcode_tab;
218 }
219
220 /* init video encoder */
221 av_cold int MPV_encode_init(AVCodecContext *avctx)
222 {
223     MpegEncContext *s = avctx->priv_data;
224     int i;
225     int chroma_h_shift, chroma_v_shift;
226
227     MPV_encode_defaults(s);
228
229     switch (avctx->codec_id) {
230     case CODEC_ID_MPEG2VIDEO:
231         if(avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P){
232             av_log(avctx, AV_LOG_ERROR, "only YUV420 and YUV422 are supported\n");
233             return -1;
234         }
235         break;
236     case CODEC_ID_LJPEG:
237     case CODEC_ID_MJPEG:
238         if(avctx->pix_fmt != PIX_FMT_YUVJ420P && avctx->pix_fmt != PIX_FMT_YUVJ422P && avctx->pix_fmt != PIX_FMT_RGB32 &&
239            ((avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P) || avctx->strict_std_compliance>FF_COMPLIANCE_INOFFICIAL)){
240             av_log(avctx, AV_LOG_ERROR, "colorspace not supported in jpeg\n");
241             return -1;
242         }
243         break;
244     default:
245         if(avctx->pix_fmt != PIX_FMT_YUV420P){
246             av_log(avctx, AV_LOG_ERROR, "only YUV420 is supported\n");
247             return -1;
248         }
249     }
250
251     switch (avctx->pix_fmt) {
252     case PIX_FMT_YUVJ422P:
253     case PIX_FMT_YUV422P:
254         s->chroma_format = CHROMA_422;
255         break;
256     case PIX_FMT_YUVJ420P:
257     case PIX_FMT_YUV420P:
258     default:
259         s->chroma_format = CHROMA_420;
260         break;
261     }
262
263     s->bit_rate = avctx->bit_rate;
264     s->width = avctx->width;
265     s->height = avctx->height;
266     if(avctx->gop_size > 600 && avctx->strict_std_compliance>FF_COMPLIANCE_EXPERIMENTAL){
267         av_log(avctx, AV_LOG_ERROR, "Warning keyframe interval too large! reducing it ...\n");
268         avctx->gop_size=600;
269     }
270     s->gop_size = avctx->gop_size;
271     s->avctx = avctx;
272     s->flags= avctx->flags;
273     s->flags2= avctx->flags2;
274     s->max_b_frames= avctx->max_b_frames;
275     s->codec_id= avctx->codec->id;
276     s->luma_elim_threshold  = avctx->luma_elim_threshold;
277     s->chroma_elim_threshold= avctx->chroma_elim_threshold;
278     s->strict_std_compliance= avctx->strict_std_compliance;
279     s->data_partitioning= avctx->flags & CODEC_FLAG_PART;
280     s->quarter_sample= (avctx->flags & CODEC_FLAG_QPEL)!=0;
281     s->mpeg_quant= avctx->mpeg_quant;
282     s->rtp_mode= !!avctx->rtp_payload_size;
283     s->intra_dc_precision= avctx->intra_dc_precision;
284     s->user_specified_pts = AV_NOPTS_VALUE;
285
286     if (s->gop_size <= 1) {
287         s->intra_only = 1;
288         s->gop_size = 12;
289     } else {
290         s->intra_only = 0;
291     }
292
293     s->me_method = avctx->me_method;
294
295     /* Fixed QSCALE */
296     s->fixed_qscale = !!(avctx->flags & CODEC_FLAG_QSCALE);
297
298     s->adaptive_quant= (   s->avctx->lumi_masking
299                         || s->avctx->dark_masking
300                         || s->avctx->temporal_cplx_masking
301                         || s->avctx->spatial_cplx_masking
302                         || s->avctx->p_masking
303                         || s->avctx->border_masking
304                         || (s->flags&CODEC_FLAG_QP_RD))
305                        && !s->fixed_qscale;
306
307     s->obmc= !!(s->flags & CODEC_FLAG_OBMC);
308     s->loop_filter= !!(s->flags & CODEC_FLAG_LOOP_FILTER);
309     s->alternate_scan= !!(s->flags & CODEC_FLAG_ALT_SCAN);
310     s->intra_vlc_format= !!(s->flags2 & CODEC_FLAG2_INTRA_VLC);
311     s->q_scale_type= !!(s->flags2 & CODEC_FLAG2_NON_LINEAR_QUANT);
312
313     if(avctx->rc_max_rate && !avctx->rc_buffer_size){
314         av_log(avctx, AV_LOG_ERROR, "a vbv buffer size is needed, for encoding with a maximum bitrate\n");
315         return -1;
316     }
317
318     if(avctx->rc_min_rate && avctx->rc_max_rate != avctx->rc_min_rate){
319         av_log(avctx, AV_LOG_INFO, "Warning min_rate > 0 but min_rate != max_rate isn't recommended!\n");
320     }
321
322     if(avctx->rc_min_rate && avctx->rc_min_rate > avctx->bit_rate){
323         av_log(avctx, AV_LOG_ERROR, "bitrate below min bitrate\n");
324         return -1;
325     }
326
327     if(avctx->rc_max_rate && avctx->rc_max_rate < avctx->bit_rate){
328         av_log(avctx, AV_LOG_INFO, "bitrate above max bitrate\n");
329         return -1;
330     }
331
332     if(avctx->rc_max_rate && avctx->rc_max_rate == avctx->bit_rate && avctx->rc_max_rate != avctx->rc_min_rate){
333         av_log(avctx, AV_LOG_INFO, "impossible bitrate constraints, this will fail\n");
334     }
335
336     if(avctx->rc_buffer_size && avctx->bit_rate*av_q2d(avctx->time_base) > avctx->rc_buffer_size){
337         av_log(avctx, AV_LOG_ERROR, "VBV buffer too small for bitrate\n");
338         return -1;
339     }
340
341     if(avctx->bit_rate*av_q2d(avctx->time_base) > avctx->bit_rate_tolerance){
342         av_log(avctx, AV_LOG_ERROR, "bitrate tolerance too small for bitrate\n");
343         return -1;
344     }
345
346     if(   s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate
347        && (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO)
348        && 90000LL * (avctx->rc_buffer_size-1) > s->avctx->rc_max_rate*0xFFFFLL){
349
350         av_log(avctx, AV_LOG_INFO, "Warning vbv_delay will be set to 0xFFFF (=VBR) as the specified vbv buffer is too large for the given bitrate!\n");
351     }
352
353     if((s->flags & CODEC_FLAG_4MV) && s->codec_id != CODEC_ID_MPEG4
354        && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P && s->codec_id != CODEC_ID_FLV1){
355         av_log(avctx, AV_LOG_ERROR, "4MV not supported by codec\n");
356         return -1;
357     }
358
359     if(s->obmc && s->avctx->mb_decision != FF_MB_DECISION_SIMPLE){
360         av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with simple mb decision\n");
361         return -1;
362     }
363
364     if(s->obmc && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P){
365         av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with H263(+)\n");
366         return -1;
367     }
368
369     if(s->quarter_sample && s->codec_id != CODEC_ID_MPEG4){
370         av_log(avctx, AV_LOG_ERROR, "qpel not supported by codec\n");
371         return -1;
372     }
373
374     if(s->data_partitioning && s->codec_id != CODEC_ID_MPEG4){
375         av_log(avctx, AV_LOG_ERROR, "data partitioning not supported by codec\n");
376         return -1;
377     }
378
379     if(s->max_b_frames && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO){
380         av_log(avctx, AV_LOG_ERROR, "b frames not supported by codec\n");
381         return -1;
382     }
383
384     if((s->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN))
385        && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG2VIDEO){
386         av_log(avctx, AV_LOG_ERROR, "interlacing not supported by codec\n");
387         return -1;
388     }
389
390     if(s->mpeg_quant && s->codec_id != CODEC_ID_MPEG4){ //FIXME mpeg2 uses that too
391         av_log(avctx, AV_LOG_ERROR, "mpeg2 style quantization not supported by codec\n");
392         return -1;
393     }
394
395     if((s->flags & CODEC_FLAG_CBP_RD) && !avctx->trellis){
396         av_log(avctx, AV_LOG_ERROR, "CBP RD needs trellis quant\n");
397         return -1;
398     }
399
400     if((s->flags & CODEC_FLAG_QP_RD) && s->avctx->mb_decision != FF_MB_DECISION_RD){
401         av_log(avctx, AV_LOG_ERROR, "QP RD needs mbd=2\n");
402         return -1;
403     }
404
405     if(s->avctx->scenechange_threshold < 1000000000 && (s->flags & CODEC_FLAG_CLOSED_GOP)){
406         av_log(avctx, AV_LOG_ERROR, "closed gop with scene change detection are not supported yet, set threshold to 1000000000\n");
407         return -1;
408     }
409
410     if((s->flags2 & CODEC_FLAG2_INTRA_VLC) && s->codec_id != CODEC_ID_MPEG2VIDEO){
411         av_log(avctx, AV_LOG_ERROR, "intra vlc table not supported by codec\n");
412         return -1;
413     }
414
415     if(s->flags & CODEC_FLAG_LOW_DELAY){
416         if (s->codec_id != CODEC_ID_MPEG2VIDEO){
417             av_log(avctx, AV_LOG_ERROR, "low delay forcing is only available for mpeg2\n");
418             return -1;
419         }
420         if (s->max_b_frames != 0){
421             av_log(avctx, AV_LOG_ERROR, "b frames cannot be used with low delay\n");
422             return -1;
423         }
424     }
425
426     if(s->q_scale_type == 1){
427         if(s->codec_id != CODEC_ID_MPEG2VIDEO){
428             av_log(avctx, AV_LOG_ERROR, "non linear quant is only available for mpeg2\n");
429             return -1;
430         }
431         if(avctx->qmax > 12){
432             av_log(avctx, AV_LOG_ERROR, "non linear quant only supports qmax <= 12 currently\n");
433             return -1;
434         }
435     }
436
437     if(s->avctx->thread_count > 1 && s->codec_id != CODEC_ID_MPEG4
438        && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO
439        && (s->codec_id != CODEC_ID_H263P || !(s->flags & CODEC_FLAG_H263P_SLICE_STRUCT))){
440         av_log(avctx, AV_LOG_ERROR, "multi threaded encoding not supported by codec\n");
441         return -1;
442     }
443
444     if(s->avctx->thread_count < 1){
445         av_log(avctx, AV_LOG_ERROR, "automatic thread number detection not supported by codec, patch welcome\n");
446         return -1;
447     }
448
449     if(s->avctx->thread_count > 1)
450         s->rtp_mode= 1;
451
452     if(!avctx->time_base.den || !avctx->time_base.num){
453         av_log(avctx, AV_LOG_ERROR, "framerate not set\n");
454         return -1;
455     }
456
457     i= (INT_MAX/2+128)>>8;
458     if(avctx->me_threshold >= i){
459         av_log(avctx, AV_LOG_ERROR, "me_threshold too large, max is %d\n", i - 1);
460         return -1;
461     }
462     if(avctx->mb_threshold >= i){
463         av_log(avctx, AV_LOG_ERROR, "mb_threshold too large, max is %d\n", i - 1);
464         return -1;
465     }
466
467     if(avctx->b_frame_strategy && (avctx->flags&CODEC_FLAG_PASS2)){
468         av_log(avctx, AV_LOG_INFO, "notice: b_frame_strategy only affects the first pass\n");
469         avctx->b_frame_strategy = 0;
470     }
471
472     i= av_gcd(avctx->time_base.den, avctx->time_base.num);
473     if(i > 1){
474         av_log(avctx, AV_LOG_INFO, "removing common factors from framerate\n");
475         avctx->time_base.den /= i;
476         avctx->time_base.num /= i;
477 //        return -1;
478     }
479
480     if(s->codec_id==CODEC_ID_MJPEG){
481         s->intra_quant_bias= 1<<(QUANT_BIAS_SHIFT-1); //(a + x/2)/x
482         s->inter_quant_bias= 0;
483     }else if(s->mpeg_quant || s->codec_id==CODEC_ID_MPEG1VIDEO || s->codec_id==CODEC_ID_MPEG2VIDEO){
484         s->intra_quant_bias= 3<<(QUANT_BIAS_SHIFT-3); //(a + x*3/8)/x
485         s->inter_quant_bias= 0;
486     }else{
487         s->intra_quant_bias=0;
488         s->inter_quant_bias=-(1<<(QUANT_BIAS_SHIFT-2)); //(a - x/4)/x
489     }
490
491     if(avctx->intra_quant_bias != FF_DEFAULT_QUANT_BIAS)
492         s->intra_quant_bias= avctx->intra_quant_bias;
493     if(avctx->inter_quant_bias != FF_DEFAULT_QUANT_BIAS)
494         s->inter_quant_bias= avctx->inter_quant_bias;
495
496     avcodec_get_chroma_sub_sample(avctx->pix_fmt, &chroma_h_shift, &chroma_v_shift);
497
498     if(avctx->codec_id == CODEC_ID_MPEG4 && s->avctx->time_base.den > (1<<16)-1){
499         av_log(avctx, AV_LOG_ERROR, "timebase not supported by mpeg 4 standard\n");
500         return -1;
501     }
502     s->time_increment_bits = av_log2(s->avctx->time_base.den - 1) + 1;
503
504     switch(avctx->codec->id) {
505     case CODEC_ID_MPEG1VIDEO:
506         s->out_format = FMT_MPEG1;
507         s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY);
508         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
509         break;
510     case CODEC_ID_MPEG2VIDEO:
511         s->out_format = FMT_MPEG1;
512         s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY);
513         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
514         s->rtp_mode= 1;
515         break;
516     case CODEC_ID_LJPEG:
517     case CODEC_ID_MJPEG:
518         s->out_format = FMT_MJPEG;
519         s->intra_only = 1; /* force intra only for jpeg */
520         s->mjpeg_vsample[0] = 2;
521         s->mjpeg_vsample[1] = 2>>chroma_v_shift;
522         s->mjpeg_vsample[2] = 2>>chroma_v_shift;
523         s->mjpeg_hsample[0] = 2;
524         s->mjpeg_hsample[1] = 2>>chroma_h_shift;
525         s->mjpeg_hsample[2] = 2>>chroma_h_shift;
526         if (!(CONFIG_MJPEG_ENCODER || CONFIG_LJPEG_ENCODER)
527             || ff_mjpeg_encode_init(s) < 0)
528             return -1;
529         avctx->delay=0;
530         s->low_delay=1;
531         break;
532     case CODEC_ID_H261:
533         if (!CONFIG_H261_ENCODER)  return -1;
534         if (ff_h261_get_picture_format(s->width, s->height) < 0) {
535             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);
536             return -1;
537         }
538         s->out_format = FMT_H261;
539         avctx->delay=0;
540         s->low_delay=1;
541         break;
542     case CODEC_ID_H263:
543         if (!CONFIG_H263_ENCODER)  return -1;
544         if (h263_get_picture_format(s->width, s->height) == 7) {
545             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);
546             return -1;
547         }
548         s->out_format = FMT_H263;
549         s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
550         avctx->delay=0;
551         s->low_delay=1;
552         break;
553     case CODEC_ID_H263P:
554         s->out_format = FMT_H263;
555         s->h263_plus = 1;
556         /* Fx */
557         s->umvplus = (avctx->flags & CODEC_FLAG_H263P_UMV) ? 1:0;
558         s->h263_aic= (avctx->flags & CODEC_FLAG_AC_PRED) ? 1:0;
559         s->modified_quant= s->h263_aic;
560         s->alt_inter_vlc= (avctx->flags & CODEC_FLAG_H263P_AIV) ? 1:0;
561         s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
562         s->loop_filter= (avctx->flags & CODEC_FLAG_LOOP_FILTER) ? 1:0;
563         s->unrestricted_mv= s->obmc || s->loop_filter || s->umvplus;
564         s->h263_slice_structured= (s->flags & CODEC_FLAG_H263P_SLICE_STRUCT) ? 1:0;
565
566         /* /Fx */
567         /* These are just to be sure */
568         avctx->delay=0;
569         s->low_delay=1;
570         break;
571     case CODEC_ID_FLV1:
572         s->out_format = FMT_H263;
573         s->h263_flv = 2; /* format = 1; 11-bit codes */
574         s->unrestricted_mv = 1;
575         s->rtp_mode=0; /* don't allow GOB */
576         avctx->delay=0;
577         s->low_delay=1;
578         break;
579     case CODEC_ID_RV10:
580         s->out_format = FMT_H263;
581         avctx->delay=0;
582         s->low_delay=1;
583         break;
584     case CODEC_ID_RV20:
585         s->out_format = FMT_H263;
586         avctx->delay=0;
587         s->low_delay=1;
588         s->modified_quant=1;
589         s->h263_aic=1;
590         s->h263_plus=1;
591         s->loop_filter=1;
592         s->unrestricted_mv= s->obmc || s->loop_filter || s->umvplus;
593         break;
594     case CODEC_ID_MPEG4:
595         s->out_format = FMT_H263;
596         s->h263_pred = 1;
597         s->unrestricted_mv = 1;
598         s->low_delay= s->max_b_frames ? 0 : 1;
599         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
600         break;
601     case CODEC_ID_MSMPEG4V1:
602         s->out_format = FMT_H263;
603         s->h263_msmpeg4 = 1;
604         s->h263_pred = 1;
605         s->unrestricted_mv = 1;
606         s->msmpeg4_version= 1;
607         avctx->delay=0;
608         s->low_delay=1;
609         break;
610     case CODEC_ID_MSMPEG4V2:
611         s->out_format = FMT_H263;
612         s->h263_msmpeg4 = 1;
613         s->h263_pred = 1;
614         s->unrestricted_mv = 1;
615         s->msmpeg4_version= 2;
616         avctx->delay=0;
617         s->low_delay=1;
618         break;
619     case CODEC_ID_MSMPEG4V3:
620         s->out_format = FMT_H263;
621         s->h263_msmpeg4 = 1;
622         s->h263_pred = 1;
623         s->unrestricted_mv = 1;
624         s->msmpeg4_version= 3;
625         s->flipflop_rounding=1;
626         avctx->delay=0;
627         s->low_delay=1;
628         break;
629     case CODEC_ID_WMV1:
630         s->out_format = FMT_H263;
631         s->h263_msmpeg4 = 1;
632         s->h263_pred = 1;
633         s->unrestricted_mv = 1;
634         s->msmpeg4_version= 4;
635         s->flipflop_rounding=1;
636         avctx->delay=0;
637         s->low_delay=1;
638         break;
639     case CODEC_ID_WMV2:
640         s->out_format = FMT_H263;
641         s->h263_msmpeg4 = 1;
642         s->h263_pred = 1;
643         s->unrestricted_mv = 1;
644         s->msmpeg4_version= 5;
645         s->flipflop_rounding=1;
646         avctx->delay=0;
647         s->low_delay=1;
648         break;
649     default:
650         return -1;
651     }
652
653     avctx->has_b_frames= !s->low_delay;
654
655     s->encoding = 1;
656
657     s->progressive_frame=
658     s->progressive_sequence= !(avctx->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN));
659
660     /* init */
661     if (MPV_common_init(s) < 0)
662         return -1;
663
664     if(!s->dct_quantize)
665         s->dct_quantize = dct_quantize_c;
666     if(!s->denoise_dct)
667         s->denoise_dct = denoise_dct_c;
668     s->fast_dct_quantize = s->dct_quantize;
669     if(avctx->trellis)
670         s->dct_quantize = dct_quantize_trellis_c;
671
672     if((CONFIG_H263P_ENCODER || CONFIG_RV20_ENCODER) && s->modified_quant)
673         s->chroma_qscale_table= ff_h263_chroma_qscale_table;
674
675     s->quant_precision=5;
676
677     ff_set_cmp(&s->dsp, s->dsp.ildct_cmp, s->avctx->ildct_cmp);
678     ff_set_cmp(&s->dsp, s->dsp.frame_skip_cmp, s->avctx->frame_skip_cmp);
679
680     if (CONFIG_H261_ENCODER && s->out_format == FMT_H261)
681         ff_h261_encode_init(s);
682     if (CONFIG_ANY_H263_ENCODER && s->out_format == FMT_H263)
683         h263_encode_init(s);
684     if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version)
685         ff_msmpeg4_encode_init(s);
686     if ((CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
687         && s->out_format == FMT_MPEG1)
688         ff_mpeg1_encode_init(s);
689
690     /* init q matrix */
691     for(i=0;i<64;i++) {
692         int j= s->dsp.idct_permutation[i];
693         if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4 && s->mpeg_quant){
694             s->intra_matrix[j] = ff_mpeg4_default_intra_matrix[i];
695             s->inter_matrix[j] = ff_mpeg4_default_non_intra_matrix[i];
696         }else if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
697             s->intra_matrix[j] =
698             s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
699         }else
700         { /* mpeg1/2 */
701             s->intra_matrix[j] = ff_mpeg1_default_intra_matrix[i];
702             s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
703         }
704         if(s->avctx->intra_matrix)
705             s->intra_matrix[j] = s->avctx->intra_matrix[i];
706         if(s->avctx->inter_matrix)
707             s->inter_matrix[j] = s->avctx->inter_matrix[i];
708     }
709
710     /* precompute matrix */
711     /* for mjpeg, we do include qscale in the matrix */
712     if (s->out_format != FMT_MJPEG) {
713         ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
714                        s->intra_matrix, s->intra_quant_bias, avctx->qmin, 31, 1);
715         ff_convert_matrix(&s->dsp, s->q_inter_matrix, s->q_inter_matrix16,
716                        s->inter_matrix, s->inter_quant_bias, avctx->qmin, 31, 0);
717     }
718
719     if(ff_rate_control_init(s) < 0)
720         return -1;
721
722     return 0;
723 }
724
725 av_cold int MPV_encode_end(AVCodecContext *avctx)
726 {
727     MpegEncContext *s = avctx->priv_data;
728
729     ff_rate_control_uninit(s);
730
731     MPV_common_end(s);
732     if ((CONFIG_MJPEG_ENCODER || CONFIG_LJPEG_ENCODER) && s->out_format == FMT_MJPEG)
733         ff_mjpeg_encode_close(s);
734
735     av_freep(&avctx->extradata);
736
737     return 0;
738 }
739
740 static int get_sae(uint8_t *src, int ref, int stride){
741     int x,y;
742     int acc=0;
743
744     for(y=0; y<16; y++){
745         for(x=0; x<16; x++){
746             acc+= FFABS(src[x+y*stride] - ref);
747         }
748     }
749
750     return acc;
751 }
752
753 static int get_intra_count(MpegEncContext *s, uint8_t *src, uint8_t *ref, int stride){
754     int x, y, w, h;
755     int acc=0;
756
757     w= s->width &~15;
758     h= s->height&~15;
759
760     for(y=0; y<h; y+=16){
761         for(x=0; x<w; x+=16){
762             int offset= x + y*stride;
763             int sad = s->dsp.sad[0](NULL, src + offset, ref + offset, stride, 16);
764             int mean= (s->dsp.pix_sum(src + offset, stride) + 128)>>8;
765             int sae = get_sae(src + offset, mean, stride);
766
767             acc+= sae + 500 < sad;
768         }
769     }
770     return acc;
771 }
772
773
774 static int load_input_picture(MpegEncContext *s, AVFrame *pic_arg){
775     AVFrame *pic=NULL;
776     int64_t pts;
777     int i;
778     const int encoding_delay= s->max_b_frames;
779     int direct=1;
780
781     if(pic_arg){
782         pts= pic_arg->pts;
783         pic_arg->display_picture_number= s->input_picture_number++;
784
785         if(pts != AV_NOPTS_VALUE){
786             if(s->user_specified_pts != AV_NOPTS_VALUE){
787                 int64_t time= pts;
788                 int64_t last= s->user_specified_pts;
789
790                 if(time <= last){
791                     av_log(s->avctx, AV_LOG_ERROR, "Error, Invalid timestamp=%"PRId64", last=%"PRId64"\n", pts, s->user_specified_pts);
792                     return -1;
793                 }
794             }
795             s->user_specified_pts= pts;
796         }else{
797             if(s->user_specified_pts != AV_NOPTS_VALUE){
798                 s->user_specified_pts=
799                 pts= s->user_specified_pts + 1;
800                 av_log(s->avctx, AV_LOG_INFO, "Warning: AVFrame.pts=? trying to guess (%"PRId64")\n", pts);
801             }else{
802                 pts= pic_arg->display_picture_number;
803             }
804         }
805     }
806
807   if(pic_arg){
808     if(encoding_delay && !(s->flags&CODEC_FLAG_INPUT_PRESERVED)) direct=0;
809     if(pic_arg->linesize[0] != s->linesize) direct=0;
810     if(pic_arg->linesize[1] != s->uvlinesize) direct=0;
811     if(pic_arg->linesize[2] != s->uvlinesize) direct=0;
812
813 //    av_log(AV_LOG_DEBUG, "%d %d %d %d\n",pic_arg->linesize[0], pic_arg->linesize[1], s->linesize, s->uvlinesize);
814
815     if(direct){
816         i= ff_find_unused_picture(s, 1);
817
818         pic= (AVFrame*)&s->picture[i];
819         pic->reference= 3;
820
821         for(i=0; i<4; i++){
822             pic->data[i]= pic_arg->data[i];
823             pic->linesize[i]= pic_arg->linesize[i];
824         }
825         ff_alloc_picture(s, (Picture*)pic, 1);
826     }else{
827         i= ff_find_unused_picture(s, 0);
828
829         pic= (AVFrame*)&s->picture[i];
830         pic->reference= 3;
831
832         ff_alloc_picture(s, (Picture*)pic, 0);
833
834         if(   pic->data[0] + INPLACE_OFFSET == pic_arg->data[0]
835            && pic->data[1] + INPLACE_OFFSET == pic_arg->data[1]
836            && pic->data[2] + INPLACE_OFFSET == pic_arg->data[2]){
837        // empty
838         }else{
839             int h_chroma_shift, v_chroma_shift;
840             avcodec_get_chroma_sub_sample(s->avctx->pix_fmt, &h_chroma_shift, &v_chroma_shift);
841
842             for(i=0; i<3; i++){
843                 int src_stride= pic_arg->linesize[i];
844                 int dst_stride= i ? s->uvlinesize : s->linesize;
845                 int h_shift= i ? h_chroma_shift : 0;
846                 int v_shift= i ? v_chroma_shift : 0;
847                 int w= s->width >>h_shift;
848                 int h= s->height>>v_shift;
849                 uint8_t *src= pic_arg->data[i];
850                 uint8_t *dst= pic->data[i];
851
852                 if(!s->avctx->rc_buffer_size)
853                     dst +=INPLACE_OFFSET;
854
855                 if(src_stride==dst_stride)
856                     memcpy(dst, src, src_stride*h);
857                 else{
858                     while(h--){
859                         memcpy(dst, src, w);
860                         dst += dst_stride;
861                         src += src_stride;
862                     }
863                 }
864             }
865         }
866     }
867     copy_picture_attributes(s, pic, pic_arg);
868     pic->pts= pts; //we set this here to avoid modifiying pic_arg
869   }
870
871     /* shift buffer entries */
872     for(i=1; i<MAX_PICTURE_COUNT /*s->encoding_delay+1*/; i++)
873         s->input_picture[i-1]= s->input_picture[i];
874
875     s->input_picture[encoding_delay]= (Picture*)pic;
876
877     return 0;
878 }
879
880 static int skip_check(MpegEncContext *s, Picture *p, Picture *ref){
881     int x, y, plane;
882     int score=0;
883     int64_t score64=0;
884
885     for(plane=0; plane<3; plane++){
886         const int stride= p->linesize[plane];
887         const int bw= plane ? 1 : 2;
888         for(y=0; y<s->mb_height*bw; y++){
889             for(x=0; x<s->mb_width*bw; x++){
890                 int off= p->type == FF_BUFFER_TYPE_SHARED ? 0: 16;
891                 int v= s->dsp.frame_skip_cmp[1](s, p->data[plane] + 8*(x + y*stride)+off, ref->data[plane] + 8*(x + y*stride), stride, 8);
892
893                 switch(s->avctx->frame_skip_exp){
894                     case 0: score= FFMAX(score, v); break;
895                     case 1: score+= FFABS(v);break;
896                     case 2: score+= v*v;break;
897                     case 3: score64+= FFABS(v*v*(int64_t)v);break;
898                     case 4: score64+= v*v*(int64_t)(v*v);break;
899                 }
900             }
901         }
902     }
903
904     if(score) score64= score;
905
906     if(score64 < s->avctx->frame_skip_threshold)
907         return 1;
908     if(score64 < ((s->avctx->frame_skip_factor * (int64_t)s->lambda)>>8))
909         return 1;
910     return 0;
911 }
912
913 static int estimate_best_b_count(MpegEncContext *s){
914     AVCodec *codec= avcodec_find_encoder(s->avctx->codec_id);
915     AVCodecContext *c= avcodec_alloc_context();
916     AVFrame input[FF_MAX_B_FRAMES+2];
917     const int scale= s->avctx->brd_scale;
918     int i, j, out_size, p_lambda, b_lambda, lambda2;
919     int outbuf_size= s->width * s->height; //FIXME
920     uint8_t *outbuf= av_malloc(outbuf_size);
921     int64_t best_rd= INT64_MAX;
922     int best_b_count= -1;
923
924     assert(scale>=0 && scale <=3);
925
926 //    emms_c();
927     p_lambda= s->last_lambda_for[FF_P_TYPE]; //s->next_picture_ptr->quality;
928     b_lambda= s->last_lambda_for[FF_B_TYPE]; //p_lambda *FFABS(s->avctx->b_quant_factor) + s->avctx->b_quant_offset;
929     if(!b_lambda) b_lambda= p_lambda; //FIXME we should do this somewhere else
930     lambda2= (b_lambda*b_lambda + (1<<FF_LAMBDA_SHIFT)/2 ) >> FF_LAMBDA_SHIFT;
931
932     c->width = s->width >> scale;
933     c->height= s->height>> scale;
934     c->flags= CODEC_FLAG_QSCALE | CODEC_FLAG_PSNR | CODEC_FLAG_INPUT_PRESERVED /*| CODEC_FLAG_EMU_EDGE*/;
935     c->flags|= s->avctx->flags & CODEC_FLAG_QPEL;
936     c->mb_decision= s->avctx->mb_decision;
937     c->me_cmp= s->avctx->me_cmp;
938     c->mb_cmp= s->avctx->mb_cmp;
939     c->me_sub_cmp= s->avctx->me_sub_cmp;
940     c->pix_fmt = PIX_FMT_YUV420P;
941     c->time_base= s->avctx->time_base;
942     c->max_b_frames= s->max_b_frames;
943
944     if (avcodec_open(c, codec) < 0)
945         return -1;
946
947     for(i=0; i<s->max_b_frames+2; i++){
948         int ysize= c->width*c->height;
949         int csize= (c->width/2)*(c->height/2);
950         Picture pre_input, *pre_input_ptr= i ? s->input_picture[i-1] : s->next_picture_ptr;
951
952         avcodec_get_frame_defaults(&input[i]);
953         input[i].data[0]= av_malloc(ysize + 2*csize);
954         input[i].data[1]= input[i].data[0] + ysize;
955         input[i].data[2]= input[i].data[1] + csize;
956         input[i].linesize[0]= c->width;
957         input[i].linesize[1]=
958         input[i].linesize[2]= c->width/2;
959
960         if(pre_input_ptr && (!i || s->input_picture[i-1])) {
961             pre_input= *pre_input_ptr;
962
963             if(pre_input.type != FF_BUFFER_TYPE_SHARED && i) {
964                 pre_input.data[0]+=INPLACE_OFFSET;
965                 pre_input.data[1]+=INPLACE_OFFSET;
966                 pre_input.data[2]+=INPLACE_OFFSET;
967             }
968
969             s->dsp.shrink[scale](input[i].data[0], input[i].linesize[0], pre_input.data[0], pre_input.linesize[0], c->width, c->height);
970             s->dsp.shrink[scale](input[i].data[1], input[i].linesize[1], pre_input.data[1], pre_input.linesize[1], c->width>>1, c->height>>1);
971             s->dsp.shrink[scale](input[i].data[2], input[i].linesize[2], pre_input.data[2], pre_input.linesize[2], c->width>>1, c->height>>1);
972         }
973     }
974
975     for(j=0; j<s->max_b_frames+1; j++){
976         int64_t rd=0;
977
978         if(!s->input_picture[j])
979             break;
980
981         c->error[0]= c->error[1]= c->error[2]= 0;
982
983         input[0].pict_type= FF_I_TYPE;
984         input[0].quality= 1 * FF_QP2LAMBDA;
985         out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[0]);
986 //        rd += (out_size * lambda2) >> FF_LAMBDA_SHIFT;
987
988         for(i=0; i<s->max_b_frames+1; i++){
989             int is_p= i % (j+1) == j || i==s->max_b_frames;
990
991             input[i+1].pict_type= is_p ? FF_P_TYPE : FF_B_TYPE;
992             input[i+1].quality= is_p ? p_lambda : b_lambda;
993             out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[i+1]);
994             rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
995         }
996
997         /* get the delayed frames */
998         while(out_size){
999             out_size = avcodec_encode_video(c, outbuf, outbuf_size, NULL);
1000             rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
1001         }
1002
1003         rd += c->error[0] + c->error[1] + c->error[2];
1004
1005         if(rd < best_rd){
1006             best_rd= rd;
1007             best_b_count= j;
1008         }
1009     }
1010
1011     av_freep(&outbuf);
1012     avcodec_close(c);
1013     av_freep(&c);
1014
1015     for(i=0; i<s->max_b_frames+2; i++){
1016         av_freep(&input[i].data[0]);
1017     }
1018
1019     return best_b_count;
1020 }
1021
1022 static void select_input_picture(MpegEncContext *s){
1023     int i;
1024
1025     for(i=1; i<MAX_PICTURE_COUNT; i++)
1026         s->reordered_input_picture[i-1]= s->reordered_input_picture[i];
1027     s->reordered_input_picture[MAX_PICTURE_COUNT-1]= NULL;
1028
1029     /* set next picture type & ordering */
1030     if(s->reordered_input_picture[0]==NULL && s->input_picture[0]){
1031         if(/*s->picture_in_gop_number >= s->gop_size ||*/ s->next_picture_ptr==NULL || s->intra_only){
1032             s->reordered_input_picture[0]= s->input_picture[0];
1033             s->reordered_input_picture[0]->pict_type= FF_I_TYPE;
1034             s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++;
1035         }else{
1036             int b_frames;
1037
1038             if(s->avctx->frame_skip_threshold || s->avctx->frame_skip_factor){
1039                 if(s->picture_in_gop_number < s->gop_size && skip_check(s, s->input_picture[0], s->next_picture_ptr)){
1040                 //FIXME check that te gop check above is +-1 correct
1041 //av_log(NULL, AV_LOG_DEBUG, "skip %p %"PRId64"\n", s->input_picture[0]->data[0], s->input_picture[0]->pts);
1042
1043                     if(s->input_picture[0]->type == FF_BUFFER_TYPE_SHARED){
1044                         for(i=0; i<4; i++)
1045                             s->input_picture[0]->data[i]= NULL;
1046                         s->input_picture[0]->type= 0;
1047                     }else{
1048                         assert(   s->input_picture[0]->type==FF_BUFFER_TYPE_USER
1049                                || s->input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
1050
1051                         s->avctx->release_buffer(s->avctx, (AVFrame*)s->input_picture[0]);
1052                     }
1053
1054                     emms_c();
1055                     ff_vbv_update(s, 0);
1056
1057                     goto no_output_pic;
1058                 }
1059             }
1060
1061             if(s->flags&CODEC_FLAG_PASS2){
1062                 for(i=0; i<s->max_b_frames+1; i++){
1063                     int pict_num= s->input_picture[0]->display_picture_number + i;
1064
1065                     if(pict_num >= s->rc_context.num_entries)
1066                         break;
1067                     if(!s->input_picture[i]){
1068                         s->rc_context.entry[pict_num-1].new_pict_type = FF_P_TYPE;
1069                         break;
1070                     }
1071
1072                     s->input_picture[i]->pict_type=
1073                         s->rc_context.entry[pict_num].new_pict_type;
1074                 }
1075             }
1076
1077             if(s->avctx->b_frame_strategy==0){
1078                 b_frames= s->max_b_frames;
1079                 while(b_frames && !s->input_picture[b_frames]) b_frames--;
1080             }else if(s->avctx->b_frame_strategy==1){
1081                 for(i=1; i<s->max_b_frames+1; i++){
1082                     if(s->input_picture[i] && s->input_picture[i]->b_frame_score==0){
1083                         s->input_picture[i]->b_frame_score=
1084                             get_intra_count(s, s->input_picture[i  ]->data[0],
1085                                                s->input_picture[i-1]->data[0], s->linesize) + 1;
1086                     }
1087                 }
1088                 for(i=0; i<s->max_b_frames+1; i++){
1089                     if(s->input_picture[i]==NULL || s->input_picture[i]->b_frame_score - 1 > s->mb_num/s->avctx->b_sensitivity) break;
1090                 }
1091
1092                 b_frames= FFMAX(0, i-1);
1093
1094                 /* reset scores */
1095                 for(i=0; i<b_frames+1; i++){
1096                     s->input_picture[i]->b_frame_score=0;
1097                 }
1098             }else if(s->avctx->b_frame_strategy==2){
1099                 b_frames= estimate_best_b_count(s);
1100             }else{
1101                 av_log(s->avctx, AV_LOG_ERROR, "illegal b frame strategy\n");
1102                 b_frames=0;
1103             }
1104
1105             emms_c();
1106 //static int b_count=0;
1107 //b_count+= b_frames;
1108 //av_log(s->avctx, AV_LOG_DEBUG, "b_frames: %d\n", b_count);
1109
1110             for(i= b_frames - 1; i>=0; i--){
1111                 int type= s->input_picture[i]->pict_type;
1112                 if(type && type != FF_B_TYPE)
1113                     b_frames= i;
1114             }
1115             if(s->input_picture[b_frames]->pict_type == FF_B_TYPE && b_frames == s->max_b_frames){
1116                 av_log(s->avctx, AV_LOG_ERROR, "warning, too many b frames in a row\n");
1117             }
1118
1119             if(s->picture_in_gop_number + b_frames >= s->gop_size){
1120               if((s->flags2 & CODEC_FLAG2_STRICT_GOP) && s->gop_size > s->picture_in_gop_number){
1121                     b_frames= s->gop_size - s->picture_in_gop_number - 1;
1122               }else{
1123                 if(s->flags & CODEC_FLAG_CLOSED_GOP)
1124                     b_frames=0;
1125                 s->input_picture[b_frames]->pict_type= FF_I_TYPE;
1126               }
1127             }
1128
1129             if(   (s->flags & CODEC_FLAG_CLOSED_GOP)
1130                && b_frames
1131                && s->input_picture[b_frames]->pict_type== FF_I_TYPE)
1132                 b_frames--;
1133
1134             s->reordered_input_picture[0]= s->input_picture[b_frames];
1135             if(s->reordered_input_picture[0]->pict_type != FF_I_TYPE)
1136                 s->reordered_input_picture[0]->pict_type= FF_P_TYPE;
1137             s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++;
1138             for(i=0; i<b_frames; i++){
1139                 s->reordered_input_picture[i+1]= s->input_picture[i];
1140                 s->reordered_input_picture[i+1]->pict_type= FF_B_TYPE;
1141                 s->reordered_input_picture[i+1]->coded_picture_number= s->coded_picture_number++;
1142             }
1143         }
1144     }
1145 no_output_pic:
1146     if(s->reordered_input_picture[0]){
1147         s->reordered_input_picture[0]->reference= s->reordered_input_picture[0]->pict_type!=FF_B_TYPE ? 3 : 0;
1148
1149         ff_copy_picture(&s->new_picture, s->reordered_input_picture[0]);
1150
1151         if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_SHARED || s->avctx->rc_buffer_size){
1152             // input is a shared pix, so we can't modifiy it -> alloc a new one & ensure that the shared one is reuseable
1153
1154             int i= ff_find_unused_picture(s, 0);
1155             Picture *pic= &s->picture[i];
1156
1157             pic->reference              = s->reordered_input_picture[0]->reference;
1158             ff_alloc_picture(s, pic, 0);
1159
1160             /* mark us unused / free shared pic */
1161             if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_INTERNAL)
1162                 s->avctx->release_buffer(s->avctx, (AVFrame*)s->reordered_input_picture[0]);
1163             for(i=0; i<4; i++)
1164                 s->reordered_input_picture[0]->data[i]= NULL;
1165             s->reordered_input_picture[0]->type= 0;
1166
1167             copy_picture_attributes(s, (AVFrame*)pic, (AVFrame*)s->reordered_input_picture[0]);
1168
1169             s->current_picture_ptr= pic;
1170         }else{
1171             // input is not a shared pix -> reuse buffer for current_pix
1172
1173             assert(   s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_USER
1174                    || s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
1175
1176             s->current_picture_ptr= s->reordered_input_picture[0];
1177             for(i=0; i<4; i++){
1178                 s->new_picture.data[i]+= INPLACE_OFFSET;
1179             }
1180         }
1181         ff_copy_picture(&s->current_picture, s->current_picture_ptr);
1182
1183         s->picture_number= s->new_picture.display_picture_number;
1184 //printf("dpn:%d\n", s->picture_number);
1185     }else{
1186        memset(&s->new_picture, 0, sizeof(Picture));
1187     }
1188 }
1189
1190 int MPV_encode_picture(AVCodecContext *avctx,
1191                        unsigned char *buf, int buf_size, void *data)
1192 {
1193     MpegEncContext *s = avctx->priv_data;
1194     AVFrame *pic_arg = data;
1195     int i, stuffing_count;
1196
1197     for(i=0; i<avctx->thread_count; i++){
1198         int start_y= s->thread_context[i]->start_mb_y;
1199         int   end_y= s->thread_context[i]->  end_mb_y;
1200         int h= s->mb_height;
1201         uint8_t *start= buf + (size_t)(((int64_t) buf_size)*start_y/h);
1202         uint8_t *end  = buf + (size_t)(((int64_t) buf_size)*  end_y/h);
1203
1204         init_put_bits(&s->thread_context[i]->pb, start, end - start);
1205     }
1206
1207     s->picture_in_gop_number++;
1208
1209     if(load_input_picture(s, pic_arg) < 0)
1210         return -1;
1211
1212     select_input_picture(s);
1213
1214     /* output? */
1215     if(s->new_picture.data[0]){
1216         s->pict_type= s->new_picture.pict_type;
1217 //emms_c();
1218 //printf("qs:%f %f %d\n", s->new_picture.quality, s->current_picture.quality, s->qscale);
1219         MPV_frame_start(s, avctx);
1220 vbv_retry:
1221         if (encode_picture(s, s->picture_number) < 0)
1222             return -1;
1223
1224         avctx->header_bits = s->header_bits;
1225         avctx->mv_bits     = s->mv_bits;
1226         avctx->misc_bits   = s->misc_bits;
1227         avctx->i_tex_bits  = s->i_tex_bits;
1228         avctx->p_tex_bits  = s->p_tex_bits;
1229         avctx->i_count     = s->i_count;
1230         avctx->p_count     = s->mb_num - s->i_count - s->skip_count; //FIXME f/b_count in avctx
1231         avctx->skip_count  = s->skip_count;
1232
1233         MPV_frame_end(s);
1234
1235         if (CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG)
1236             ff_mjpeg_encode_picture_trailer(s);
1237
1238         if(avctx->rc_buffer_size){
1239             RateControlContext *rcc= &s->rc_context;
1240             int max_size= rcc->buffer_index * avctx->rc_max_available_vbv_use;
1241
1242             if(put_bits_count(&s->pb) > max_size && s->lambda < s->avctx->lmax){
1243                 s->next_lambda= FFMAX(s->lambda+1, s->lambda*(s->qscale+1) / s->qscale);
1244                 if(s->adaptive_quant){
1245                     int i;
1246                     for(i=0; i<s->mb_height*s->mb_stride; i++)
1247                         s->lambda_table[i]= FFMAX(s->lambda_table[i]+1, s->lambda_table[i]*(s->qscale+1) / s->qscale);
1248                 }
1249                 s->mb_skipped = 0;        //done in MPV_frame_start()
1250                 if(s->pict_type==FF_P_TYPE){ //done in encode_picture() so we must undo it
1251                     if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
1252                         s->no_rounding ^= 1;
1253                 }
1254                 if(s->pict_type!=FF_B_TYPE){
1255                     s->time_base= s->last_time_base;
1256                     s->last_non_b_time= s->time - s->pp_time;
1257                 }
1258 //                av_log(NULL, AV_LOG_ERROR, "R:%d ", s->next_lambda);
1259                 for(i=0; i<avctx->thread_count; i++){
1260                     PutBitContext *pb= &s->thread_context[i]->pb;
1261                     init_put_bits(pb, pb->buf, pb->buf_end - pb->buf);
1262                 }
1263                 goto vbv_retry;
1264             }
1265
1266             assert(s->avctx->rc_max_rate);
1267         }
1268
1269         if(s->flags&CODEC_FLAG_PASS1)
1270             ff_write_pass1_stats(s);
1271
1272         for(i=0; i<4; i++){
1273             s->current_picture_ptr->error[i]= s->current_picture.error[i];
1274             avctx->error[i] += s->current_picture_ptr->error[i];
1275         }
1276
1277         if(s->flags&CODEC_FLAG_PASS1)
1278             assert(avctx->header_bits + avctx->mv_bits + avctx->misc_bits + avctx->i_tex_bits + avctx->p_tex_bits == put_bits_count(&s->pb));
1279         flush_put_bits(&s->pb);
1280         s->frame_bits  = put_bits_count(&s->pb);
1281
1282         stuffing_count= ff_vbv_update(s, s->frame_bits);
1283         if(stuffing_count){
1284             if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < stuffing_count + 50){
1285                 av_log(s->avctx, AV_LOG_ERROR, "stuffing too large\n");
1286                 return -1;
1287             }
1288
1289             switch(s->codec_id){
1290             case CODEC_ID_MPEG1VIDEO:
1291             case CODEC_ID_MPEG2VIDEO:
1292                 while(stuffing_count--){
1293                     put_bits(&s->pb, 8, 0);
1294                 }
1295             break;
1296             case CODEC_ID_MPEG4:
1297                 put_bits(&s->pb, 16, 0);
1298                 put_bits(&s->pb, 16, 0x1C3);
1299                 stuffing_count -= 4;
1300                 while(stuffing_count--){
1301                     put_bits(&s->pb, 8, 0xFF);
1302                 }
1303             break;
1304             default:
1305                 av_log(s->avctx, AV_LOG_ERROR, "vbv buffer overflow\n");
1306             }
1307             flush_put_bits(&s->pb);
1308             s->frame_bits  = put_bits_count(&s->pb);
1309         }
1310
1311         /* update mpeg1/2 vbv_delay for CBR */
1312         if(s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate && s->out_format == FMT_MPEG1
1313            && 90000LL * (avctx->rc_buffer_size-1) <= s->avctx->rc_max_rate*0xFFFFLL){
1314             int vbv_delay, min_delay;
1315             double inbits = s->avctx->rc_max_rate*av_q2d(s->avctx->time_base);
1316             int    minbits= s->frame_bits - 8*(s->vbv_delay_ptr - s->pb.buf - 1);
1317             double bits   = s->rc_context.buffer_index + minbits - inbits;
1318
1319             if(bits<0)
1320                 av_log(s->avctx, AV_LOG_ERROR, "Internal error, negative bits\n");
1321
1322             assert(s->repeat_first_field==0);
1323
1324             vbv_delay=     bits * 90000                               / s->avctx->rc_max_rate;
1325             min_delay= (minbits * 90000LL + s->avctx->rc_max_rate - 1)/ s->avctx->rc_max_rate;
1326
1327             vbv_delay= FFMAX(vbv_delay, min_delay);
1328
1329             assert(vbv_delay < 0xFFFF);
1330
1331             s->vbv_delay_ptr[0] &= 0xF8;
1332             s->vbv_delay_ptr[0] |= vbv_delay>>13;
1333             s->vbv_delay_ptr[1]  = vbv_delay>>5;
1334             s->vbv_delay_ptr[2] &= 0x07;
1335             s->vbv_delay_ptr[2] |= vbv_delay<<3;
1336         }
1337         s->total_bits += s->frame_bits;
1338         avctx->frame_bits  = s->frame_bits;
1339     }else{
1340         assert((put_bits_ptr(&s->pb) == s->pb.buf));
1341         s->frame_bits=0;
1342     }
1343     assert((s->frame_bits&7)==0);
1344
1345     return s->frame_bits/8;
1346 }
1347
1348 static inline void dct_single_coeff_elimination(MpegEncContext *s, int n, int threshold)
1349 {
1350     static const char tab[64]=
1351         {3,2,2,1,1,1,1,1,
1352          1,1,1,1,1,1,1,1,
1353          1,1,1,1,1,1,1,1,
1354          0,0,0,0,0,0,0,0,
1355          0,0,0,0,0,0,0,0,
1356          0,0,0,0,0,0,0,0,
1357          0,0,0,0,0,0,0,0,
1358          0,0,0,0,0,0,0,0};
1359     int score=0;
1360     int run=0;
1361     int i;
1362     DCTELEM *block= s->block[n];
1363     const int last_index= s->block_last_index[n];
1364     int skip_dc;
1365
1366     if(threshold<0){
1367         skip_dc=0;
1368         threshold= -threshold;
1369     }else
1370         skip_dc=1;
1371
1372     /* Are all we could set to zero already zero? */
1373     if(last_index<=skip_dc - 1) return;
1374
1375     for(i=0; i<=last_index; i++){
1376         const int j = s->intra_scantable.permutated[i];
1377         const int level = FFABS(block[j]);
1378         if(level==1){
1379             if(skip_dc && i==0) continue;
1380             score+= tab[run];
1381             run=0;
1382         }else if(level>1){
1383             return;
1384         }else{
1385             run++;
1386         }
1387     }
1388     if(score >= threshold) return;
1389     for(i=skip_dc; i<=last_index; i++){
1390         const int j = s->intra_scantable.permutated[i];
1391         block[j]=0;
1392     }
1393     if(block[0]) s->block_last_index[n]= 0;
1394     else         s->block_last_index[n]= -1;
1395 }
1396
1397 static inline void clip_coeffs(MpegEncContext *s, DCTELEM *block, int last_index)
1398 {
1399     int i;
1400     const int maxlevel= s->max_qcoeff;
1401     const int minlevel= s->min_qcoeff;
1402     int overflow=0;
1403
1404     if(s->mb_intra){
1405         i=1; //skip clipping of intra dc
1406     }else
1407         i=0;
1408
1409     for(;i<=last_index; i++){
1410         const int j= s->intra_scantable.permutated[i];
1411         int level = block[j];
1412
1413         if     (level>maxlevel){
1414             level=maxlevel;
1415             overflow++;
1416         }else if(level<minlevel){
1417             level=minlevel;
1418             overflow++;
1419         }
1420
1421         block[j]= level;
1422     }
1423
1424     if(overflow && s->avctx->mb_decision == FF_MB_DECISION_SIMPLE)
1425         av_log(s->avctx, AV_LOG_INFO, "warning, clipping %d dct coefficients to %d..%d\n", overflow, minlevel, maxlevel);
1426 }
1427
1428 static void get_visual_weight(int16_t *weight, uint8_t *ptr, int stride){
1429     int x, y;
1430 //FIXME optimize
1431     for(y=0; y<8; y++){
1432         for(x=0; x<8; x++){
1433             int x2, y2;
1434             int sum=0;
1435             int sqr=0;
1436             int count=0;
1437
1438             for(y2= FFMAX(y-1, 0); y2 < FFMIN(8, y+2); y2++){
1439                 for(x2= FFMAX(x-1, 0); x2 < FFMIN(8, x+2); x2++){
1440                     int v= ptr[x2 + y2*stride];
1441                     sum += v;
1442                     sqr += v*v;
1443                     count++;
1444                 }
1445             }
1446             weight[x + 8*y]= (36*ff_sqrt(count*sqr - sum*sum)) / count;
1447         }
1448     }
1449 }
1450
1451 static av_always_inline void encode_mb_internal(MpegEncContext *s, int motion_x, int motion_y, int mb_block_height, int mb_block_count)
1452 {
1453     int16_t weight[8][64];
1454     DCTELEM orig[8][64];
1455     const int mb_x= s->mb_x;
1456     const int mb_y= s->mb_y;
1457     int i;
1458     int skip_dct[8];
1459     int dct_offset   = s->linesize*8; //default for progressive frames
1460     uint8_t *ptr_y, *ptr_cb, *ptr_cr;
1461     int wrap_y, wrap_c;
1462
1463     for(i=0; i<mb_block_count; i++) skip_dct[i]=s->skipdct;
1464
1465     if(s->adaptive_quant){
1466         const int last_qp= s->qscale;
1467         const int mb_xy= mb_x + mb_y*s->mb_stride;
1468
1469         s->lambda= s->lambda_table[mb_xy];
1470         update_qscale(s);
1471
1472         if(!(s->flags&CODEC_FLAG_QP_RD)){
1473             s->qscale= s->current_picture_ptr->qscale_table[mb_xy];
1474             s->dquant= s->qscale - last_qp;
1475
1476             if(s->out_format==FMT_H263){
1477                 s->dquant= av_clip(s->dquant, -2, 2);
1478
1479                 if(s->codec_id==CODEC_ID_MPEG4){
1480                     if(!s->mb_intra){
1481                         if(s->pict_type == FF_B_TYPE){
1482                             if(s->dquant&1 || s->mv_dir&MV_DIRECT)
1483                                 s->dquant= 0;
1484                         }
1485                         if(s->mv_type==MV_TYPE_8X8)
1486                             s->dquant=0;
1487                     }
1488                 }
1489             }
1490         }
1491         ff_set_qscale(s, last_qp + s->dquant);
1492     }else if(s->flags&CODEC_FLAG_QP_RD)
1493         ff_set_qscale(s, s->qscale + s->dquant);
1494
1495     wrap_y = s->linesize;
1496     wrap_c = s->uvlinesize;
1497     ptr_y = s->new_picture.data[0] + (mb_y * 16 * wrap_y) + mb_x * 16;
1498     ptr_cb = s->new_picture.data[1] + (mb_y * mb_block_height * wrap_c) + mb_x * 8;
1499     ptr_cr = s->new_picture.data[2] + (mb_y * mb_block_height * wrap_c) + mb_x * 8;
1500
1501     if(mb_x*16+16 > s->width || mb_y*16+16 > s->height){
1502         uint8_t *ebuf= s->edge_emu_buffer + 32;
1503         ff_emulated_edge_mc(ebuf            , ptr_y , wrap_y,16,16,mb_x*16,mb_y*16, s->width   , s->height);
1504         ptr_y= ebuf;
1505         ff_emulated_edge_mc(ebuf+18*wrap_y  , ptr_cb, wrap_c, 8, mb_block_height, mb_x*8, mb_y*8, s->width>>1, s->height>>1);
1506         ptr_cb= ebuf+18*wrap_y;
1507         ff_emulated_edge_mc(ebuf+18*wrap_y+8, ptr_cr, wrap_c, 8, mb_block_height, mb_x*8, mb_y*8, s->width>>1, s->height>>1);
1508         ptr_cr= ebuf+18*wrap_y+8;
1509     }
1510
1511     if (s->mb_intra) {
1512         if(s->flags&CODEC_FLAG_INTERLACED_DCT){
1513             int progressive_score, interlaced_score;
1514
1515             s->interlaced_dct=0;
1516             progressive_score= s->dsp.ildct_cmp[4](s, ptr_y           , NULL, wrap_y, 8)
1517                               +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y*8, NULL, wrap_y, 8) - 400;
1518
1519             if(progressive_score > 0){
1520                 interlaced_score = s->dsp.ildct_cmp[4](s, ptr_y           , NULL, wrap_y*2, 8)
1521                                   +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y  , NULL, wrap_y*2, 8);
1522                 if(progressive_score > interlaced_score){
1523                     s->interlaced_dct=1;
1524
1525                     dct_offset= wrap_y;
1526                     wrap_y<<=1;
1527                     if (s->chroma_format == CHROMA_422)
1528                         wrap_c<<=1;
1529                 }
1530             }
1531         }
1532
1533         s->dsp.get_pixels(s->block[0], ptr_y                 , wrap_y);
1534         s->dsp.get_pixels(s->block[1], ptr_y              + 8, wrap_y);
1535         s->dsp.get_pixels(s->block[2], ptr_y + dct_offset    , wrap_y);
1536         s->dsp.get_pixels(s->block[3], ptr_y + dct_offset + 8, wrap_y);
1537
1538         if(s->flags&CODEC_FLAG_GRAY){
1539             skip_dct[4]= 1;
1540             skip_dct[5]= 1;
1541         }else{
1542             s->dsp.get_pixels(s->block[4], ptr_cb, wrap_c);
1543             s->dsp.get_pixels(s->block[5], ptr_cr, wrap_c);
1544             if(!s->chroma_y_shift){ /* 422 */
1545                 s->dsp.get_pixels(s->block[6], ptr_cb + (dct_offset>>1), wrap_c);
1546                 s->dsp.get_pixels(s->block[7], ptr_cr + (dct_offset>>1), wrap_c);
1547             }
1548         }
1549     }else{
1550         op_pixels_func (*op_pix)[4];
1551         qpel_mc_func (*op_qpix)[16];
1552         uint8_t *dest_y, *dest_cb, *dest_cr;
1553
1554         dest_y  = s->dest[0];
1555         dest_cb = s->dest[1];
1556         dest_cr = s->dest[2];
1557
1558         if ((!s->no_rounding) || s->pict_type==FF_B_TYPE){
1559             op_pix = s->dsp.put_pixels_tab;
1560             op_qpix= s->dsp.put_qpel_pixels_tab;
1561         }else{
1562             op_pix = s->dsp.put_no_rnd_pixels_tab;
1563             op_qpix= s->dsp.put_no_rnd_qpel_pixels_tab;
1564         }
1565
1566         if (s->mv_dir & MV_DIR_FORWARD) {
1567             MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.data, op_pix, op_qpix);
1568             op_pix = s->dsp.avg_pixels_tab;
1569             op_qpix= s->dsp.avg_qpel_pixels_tab;
1570         }
1571         if (s->mv_dir & MV_DIR_BACKWARD) {
1572             MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.data, op_pix, op_qpix);
1573         }
1574
1575         if(s->flags&CODEC_FLAG_INTERLACED_DCT){
1576             int progressive_score, interlaced_score;
1577
1578             s->interlaced_dct=0;
1579             progressive_score= s->dsp.ildct_cmp[0](s, dest_y           , ptr_y           , wrap_y, 8)
1580                               +s->dsp.ildct_cmp[0](s, dest_y + wrap_y*8, ptr_y + wrap_y*8, wrap_y, 8) - 400;
1581
1582             if(s->avctx->ildct_cmp == FF_CMP_VSSE) progressive_score -= 400;
1583
1584             if(progressive_score>0){
1585                 interlaced_score = s->dsp.ildct_cmp[0](s, dest_y           , ptr_y           , wrap_y*2, 8)
1586                                   +s->dsp.ildct_cmp[0](s, dest_y + wrap_y  , ptr_y + wrap_y  , wrap_y*2, 8);
1587
1588                 if(progressive_score > interlaced_score){
1589                     s->interlaced_dct=1;
1590
1591                     dct_offset= wrap_y;
1592                     wrap_y<<=1;
1593                     if (s->chroma_format == CHROMA_422)
1594                         wrap_c<<=1;
1595                 }
1596             }
1597         }
1598
1599         s->dsp.diff_pixels(s->block[0], ptr_y                 , dest_y                 , wrap_y);
1600         s->dsp.diff_pixels(s->block[1], ptr_y              + 8, dest_y              + 8, wrap_y);
1601         s->dsp.diff_pixels(s->block[2], ptr_y + dct_offset    , dest_y + dct_offset    , wrap_y);
1602         s->dsp.diff_pixels(s->block[3], ptr_y + dct_offset + 8, dest_y + dct_offset + 8, wrap_y);
1603
1604         if(s->flags&CODEC_FLAG_GRAY){
1605             skip_dct[4]= 1;
1606             skip_dct[5]= 1;
1607         }else{
1608             s->dsp.diff_pixels(s->block[4], ptr_cb, dest_cb, wrap_c);
1609             s->dsp.diff_pixels(s->block[5], ptr_cr, dest_cr, wrap_c);
1610             if(!s->chroma_y_shift){ /* 422 */
1611                 s->dsp.diff_pixels(s->block[6], ptr_cb + (dct_offset>>1), dest_cb + (dct_offset>>1), wrap_c);
1612                 s->dsp.diff_pixels(s->block[7], ptr_cr + (dct_offset>>1), dest_cr + (dct_offset>>1), wrap_c);
1613             }
1614         }
1615         /* pre quantization */
1616         if(s->current_picture.mc_mb_var[s->mb_stride*mb_y+ mb_x]<2*s->qscale*s->qscale){
1617             //FIXME optimize
1618             if(s->dsp.sad[1](NULL, ptr_y               , dest_y               , wrap_y, 8) < 20*s->qscale) skip_dct[0]= 1;
1619             if(s->dsp.sad[1](NULL, ptr_y            + 8, dest_y            + 8, wrap_y, 8) < 20*s->qscale) skip_dct[1]= 1;
1620             if(s->dsp.sad[1](NULL, ptr_y +dct_offset   , dest_y +dct_offset   , wrap_y, 8) < 20*s->qscale) skip_dct[2]= 1;
1621             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;
1622             if(s->dsp.sad[1](NULL, ptr_cb              , dest_cb              , wrap_c, 8) < 20*s->qscale) skip_dct[4]= 1;
1623             if(s->dsp.sad[1](NULL, ptr_cr              , dest_cr              , wrap_c, 8) < 20*s->qscale) skip_dct[5]= 1;
1624             if(!s->chroma_y_shift){ /* 422 */
1625                 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;
1626                 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;
1627             }
1628         }
1629     }
1630
1631     if(s->avctx->quantizer_noise_shaping){
1632         if(!skip_dct[0]) get_visual_weight(weight[0], ptr_y                 , wrap_y);
1633         if(!skip_dct[1]) get_visual_weight(weight[1], ptr_y              + 8, wrap_y);
1634         if(!skip_dct[2]) get_visual_weight(weight[2], ptr_y + dct_offset    , wrap_y);
1635         if(!skip_dct[3]) get_visual_weight(weight[3], ptr_y + dct_offset + 8, wrap_y);
1636         if(!skip_dct[4]) get_visual_weight(weight[4], ptr_cb                , wrap_c);
1637         if(!skip_dct[5]) get_visual_weight(weight[5], ptr_cr                , wrap_c);
1638         if(!s->chroma_y_shift){ /* 422 */
1639             if(!skip_dct[6]) get_visual_weight(weight[6], ptr_cb + (dct_offset>>1), wrap_c);
1640             if(!skip_dct[7]) get_visual_weight(weight[7], ptr_cr + (dct_offset>>1), wrap_c);
1641         }
1642         memcpy(orig[0], s->block[0], sizeof(DCTELEM)*64*mb_block_count);
1643     }
1644
1645     /* DCT & quantize */
1646     assert(s->out_format!=FMT_MJPEG || s->qscale==8);
1647     {
1648         for(i=0;i<mb_block_count;i++) {
1649             if(!skip_dct[i]){
1650                 int overflow;
1651                 s->block_last_index[i] = s->dct_quantize(s, s->block[i], i, s->qscale, &overflow);
1652             // FIXME we could decide to change to quantizer instead of clipping
1653             // JS: I don't think that would be a good idea it could lower quality instead
1654             //     of improve it. Just INTRADC clipping deserves changes in quantizer
1655                 if (overflow) clip_coeffs(s, s->block[i], s->block_last_index[i]);
1656             }else
1657                 s->block_last_index[i]= -1;
1658         }
1659         if(s->avctx->quantizer_noise_shaping){
1660             for(i=0;i<mb_block_count;i++) {
1661                 if(!skip_dct[i]){
1662                     s->block_last_index[i] = dct_quantize_refine(s, s->block[i], weight[i], orig[i], i, s->qscale);
1663                 }
1664             }
1665         }
1666
1667         if(s->luma_elim_threshold && !s->mb_intra)
1668             for(i=0; i<4; i++)
1669                 dct_single_coeff_elimination(s, i, s->luma_elim_threshold);
1670         if(s->chroma_elim_threshold && !s->mb_intra)
1671             for(i=4; i<mb_block_count; i++)
1672                 dct_single_coeff_elimination(s, i, s->chroma_elim_threshold);
1673
1674         if(s->flags & CODEC_FLAG_CBP_RD){
1675             for(i=0;i<mb_block_count;i++) {
1676                 if(s->block_last_index[i] == -1)
1677                     s->coded_score[i]= INT_MAX/256;
1678             }
1679         }
1680     }
1681
1682     if((s->flags&CODEC_FLAG_GRAY) && s->mb_intra){
1683         s->block_last_index[4]=
1684         s->block_last_index[5]= 0;
1685         s->block[4][0]=
1686         s->block[5][0]= (1024 + s->c_dc_scale/2)/ s->c_dc_scale;
1687     }
1688
1689     //non c quantize code returns incorrect block_last_index FIXME
1690     if(s->alternate_scan && s->dct_quantize != dct_quantize_c){
1691         for(i=0; i<mb_block_count; i++){
1692             int j;
1693             if(s->block_last_index[i]>0){
1694                 for(j=63; j>0; j--){
1695                     if(s->block[i][ s->intra_scantable.permutated[j] ]) break;
1696                 }
1697                 s->block_last_index[i]= j;
1698             }
1699         }
1700     }
1701
1702     /* huffman encode */
1703     switch(s->codec_id){ //FIXME funct ptr could be slightly faster
1704     case CODEC_ID_MPEG1VIDEO:
1705     case CODEC_ID_MPEG2VIDEO:
1706         if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
1707             mpeg1_encode_mb(s, s->block, motion_x, motion_y);
1708         break;
1709     case CODEC_ID_MPEG4:
1710         if (CONFIG_MPEG4_ENCODER)
1711             mpeg4_encode_mb(s, s->block, motion_x, motion_y);
1712         break;
1713     case CODEC_ID_MSMPEG4V2:
1714     case CODEC_ID_MSMPEG4V3:
1715     case CODEC_ID_WMV1:
1716         if (CONFIG_MSMPEG4_ENCODER)
1717             msmpeg4_encode_mb(s, s->block, motion_x, motion_y);
1718         break;
1719     case CODEC_ID_WMV2:
1720         if (CONFIG_WMV2_ENCODER)
1721             ff_wmv2_encode_mb(s, s->block, motion_x, motion_y);
1722         break;
1723     case CODEC_ID_H261:
1724         if (CONFIG_H261_ENCODER)
1725             ff_h261_encode_mb(s, s->block, motion_x, motion_y);
1726         break;
1727     case CODEC_ID_H263:
1728     case CODEC_ID_H263P:
1729     case CODEC_ID_FLV1:
1730     case CODEC_ID_RV10:
1731     case CODEC_ID_RV20:
1732         if (CONFIG_H263_ENCODER)
1733             h263_encode_mb(s, s->block, motion_x, motion_y);
1734         break;
1735     case CODEC_ID_MJPEG:
1736         if (CONFIG_MJPEG_ENCODER)
1737             ff_mjpeg_encode_mb(s, s->block);
1738         break;
1739     default:
1740         assert(0);
1741     }
1742 }
1743
1744 static av_always_inline void encode_mb(MpegEncContext *s, int motion_x, int motion_y)
1745 {
1746     if (s->chroma_format == CHROMA_420) encode_mb_internal(s, motion_x, motion_y,  8, 6);
1747     else                                encode_mb_internal(s, motion_x, motion_y, 16, 8);
1748 }
1749
1750 static inline void copy_context_before_encode(MpegEncContext *d, MpegEncContext *s, int type){
1751     int i;
1752
1753     memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
1754
1755     /* mpeg1 */
1756     d->mb_skip_run= s->mb_skip_run;
1757     for(i=0; i<3; i++)
1758         d->last_dc[i]= s->last_dc[i];
1759
1760     /* statistics */
1761     d->mv_bits= s->mv_bits;
1762     d->i_tex_bits= s->i_tex_bits;
1763     d->p_tex_bits= s->p_tex_bits;
1764     d->i_count= s->i_count;
1765     d->f_count= s->f_count;
1766     d->b_count= s->b_count;
1767     d->skip_count= s->skip_count;
1768     d->misc_bits= s->misc_bits;
1769     d->last_bits= 0;
1770
1771     d->mb_skipped= 0;
1772     d->qscale= s->qscale;
1773     d->dquant= s->dquant;
1774
1775     d->esc3_level_length= s->esc3_level_length;
1776 }
1777
1778 static inline void copy_context_after_encode(MpegEncContext *d, MpegEncContext *s, int type){
1779     int i;
1780
1781     memcpy(d->mv, s->mv, 2*4*2*sizeof(int));
1782     memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
1783
1784     /* mpeg1 */
1785     d->mb_skip_run= s->mb_skip_run;
1786     for(i=0; i<3; i++)
1787         d->last_dc[i]= s->last_dc[i];
1788
1789     /* statistics */
1790     d->mv_bits= s->mv_bits;
1791     d->i_tex_bits= s->i_tex_bits;
1792     d->p_tex_bits= s->p_tex_bits;
1793     d->i_count= s->i_count;
1794     d->f_count= s->f_count;
1795     d->b_count= s->b_count;
1796     d->skip_count= s->skip_count;
1797     d->misc_bits= s->misc_bits;
1798
1799     d->mb_intra= s->mb_intra;
1800     d->mb_skipped= s->mb_skipped;
1801     d->mv_type= s->mv_type;
1802     d->mv_dir= s->mv_dir;
1803     d->pb= s->pb;
1804     if(s->data_partitioning){
1805         d->pb2= s->pb2;
1806         d->tex_pb= s->tex_pb;
1807     }
1808     d->block= s->block;
1809     for(i=0; i<8; i++)
1810         d->block_last_index[i]= s->block_last_index[i];
1811     d->interlaced_dct= s->interlaced_dct;
1812     d->qscale= s->qscale;
1813
1814     d->esc3_level_length= s->esc3_level_length;
1815 }
1816
1817 static inline void encode_mb_hq(MpegEncContext *s, MpegEncContext *backup, MpegEncContext *best, int type,
1818                            PutBitContext pb[2], PutBitContext pb2[2], PutBitContext tex_pb[2],
1819                            int *dmin, int *next_block, int motion_x, int motion_y)
1820 {
1821     int score;
1822     uint8_t *dest_backup[3];
1823
1824     copy_context_before_encode(s, backup, type);
1825
1826     s->block= s->blocks[*next_block];
1827     s->pb= pb[*next_block];
1828     if(s->data_partitioning){
1829         s->pb2   = pb2   [*next_block];
1830         s->tex_pb= tex_pb[*next_block];
1831     }
1832
1833     if(*next_block){
1834         memcpy(dest_backup, s->dest, sizeof(s->dest));
1835         s->dest[0] = s->rd_scratchpad;
1836         s->dest[1] = s->rd_scratchpad + 16*s->linesize;
1837         s->dest[2] = s->rd_scratchpad + 16*s->linesize + 8;
1838         assert(s->linesize >= 32); //FIXME
1839     }
1840
1841     encode_mb(s, motion_x, motion_y);
1842
1843     score= put_bits_count(&s->pb);
1844     if(s->data_partitioning){
1845         score+= put_bits_count(&s->pb2);
1846         score+= put_bits_count(&s->tex_pb);
1847     }
1848
1849     if(s->avctx->mb_decision == FF_MB_DECISION_RD){
1850         MPV_decode_mb(s, s->block);
1851
1852         score *= s->lambda2;
1853         score += sse_mb(s) << FF_LAMBDA_SHIFT;
1854     }
1855
1856     if(*next_block){
1857         memcpy(s->dest, dest_backup, sizeof(s->dest));
1858     }
1859
1860     if(score<*dmin){
1861         *dmin= score;
1862         *next_block^=1;
1863
1864         copy_context_after_encode(best, s, type);
1865     }
1866 }
1867
1868 static int sse(MpegEncContext *s, uint8_t *src1, uint8_t *src2, int w, int h, int stride){
1869     uint32_t *sq = ff_squareTbl + 256;
1870     int acc=0;
1871     int x,y;
1872
1873     if(w==16 && h==16)
1874         return s->dsp.sse[0](NULL, src1, src2, stride, 16);
1875     else if(w==8 && h==8)
1876         return s->dsp.sse[1](NULL, src1, src2, stride, 8);
1877
1878     for(y=0; y<h; y++){
1879         for(x=0; x<w; x++){
1880             acc+= sq[src1[x + y*stride] - src2[x + y*stride]];
1881         }
1882     }
1883
1884     assert(acc>=0);
1885
1886     return acc;
1887 }
1888
1889 static int sse_mb(MpegEncContext *s){
1890     int w= 16;
1891     int h= 16;
1892
1893     if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
1894     if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
1895
1896     if(w==16 && h==16)
1897       if(s->avctx->mb_cmp == FF_CMP_NSSE){
1898         return  s->dsp.nsse[0](s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, s->dest[0], s->linesize, 16)
1899                +s->dsp.nsse[1](s, s->new_picture.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[1], s->uvlinesize, 8)
1900                +s->dsp.nsse[1](s, s->new_picture.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[2], s->uvlinesize, 8);
1901       }else{
1902         return  s->dsp.sse[0](NULL, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, s->dest[0], s->linesize, 16)
1903                +s->dsp.sse[1](NULL, s->new_picture.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[1], s->uvlinesize, 8)
1904                +s->dsp.sse[1](NULL, s->new_picture.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[2], s->uvlinesize, 8);
1905       }
1906     else
1907         return  sse(s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, s->dest[0], w, h, s->linesize)
1908                +sse(s, s->new_picture.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[1], w>>1, h>>1, s->uvlinesize)
1909                +sse(s, s->new_picture.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[2], w>>1, h>>1, s->uvlinesize);
1910 }
1911
1912 static int pre_estimate_motion_thread(AVCodecContext *c, void *arg){
1913     MpegEncContext *s= *(void**)arg;
1914
1915
1916     s->me.pre_pass=1;
1917     s->me.dia_size= s->avctx->pre_dia_size;
1918     s->first_slice_line=1;
1919     for(s->mb_y= s->end_mb_y-1; s->mb_y >= s->start_mb_y; s->mb_y--) {
1920         for(s->mb_x=s->mb_width-1; s->mb_x >=0 ;s->mb_x--) {
1921             ff_pre_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
1922         }
1923         s->first_slice_line=0;
1924     }
1925
1926     s->me.pre_pass=0;
1927
1928     return 0;
1929 }
1930
1931 static int estimate_motion_thread(AVCodecContext *c, void *arg){
1932     MpegEncContext *s= *(void**)arg;
1933
1934     ff_check_alignment();
1935
1936     s->me.dia_size= s->avctx->dia_size;
1937     s->first_slice_line=1;
1938     for(s->mb_y= s->start_mb_y; s->mb_y < s->end_mb_y; s->mb_y++) {
1939         s->mb_x=0; //for block init below
1940         ff_init_block_index(s);
1941         for(s->mb_x=0; s->mb_x < s->mb_width; s->mb_x++) {
1942             s->block_index[0]+=2;
1943             s->block_index[1]+=2;
1944             s->block_index[2]+=2;
1945             s->block_index[3]+=2;
1946
1947             /* compute motion vector & mb_type and store in context */
1948             if(s->pict_type==FF_B_TYPE)
1949                 ff_estimate_b_frame_motion(s, s->mb_x, s->mb_y);
1950             else
1951                 ff_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
1952         }
1953         s->first_slice_line=0;
1954     }
1955     return 0;
1956 }
1957
1958 static int mb_var_thread(AVCodecContext *c, void *arg){
1959     MpegEncContext *s= *(void**)arg;
1960     int mb_x, mb_y;
1961
1962     ff_check_alignment();
1963
1964     for(mb_y=s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
1965         for(mb_x=0; mb_x < s->mb_width; mb_x++) {
1966             int xx = mb_x * 16;
1967             int yy = mb_y * 16;
1968             uint8_t *pix = s->new_picture.data[0] + (yy * s->linesize) + xx;
1969             int varc;
1970             int sum = s->dsp.pix_sum(pix, s->linesize);
1971
1972             varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500 + 128)>>8;
1973
1974             s->current_picture.mb_var [s->mb_stride * mb_y + mb_x] = varc;
1975             s->current_picture.mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
1976             s->me.mb_var_sum_temp    += varc;
1977         }
1978     }
1979     return 0;
1980 }
1981
1982 static void write_slice_end(MpegEncContext *s){
1983     if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4){
1984         if(s->partitioned_frame){
1985             ff_mpeg4_merge_partitions(s);
1986         }
1987
1988         ff_mpeg4_stuffing(&s->pb);
1989     }else if(CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG){
1990         ff_mjpeg_encode_stuffing(&s->pb);
1991     }
1992
1993     align_put_bits(&s->pb);
1994     flush_put_bits(&s->pb);
1995
1996     if((s->flags&CODEC_FLAG_PASS1) && !s->partitioned_frame)
1997         s->misc_bits+= get_bits_diff(s);
1998 }
1999
2000 static int encode_thread(AVCodecContext *c, void *arg){
2001     MpegEncContext *s= *(void**)arg;
2002     int mb_x, mb_y, pdif = 0;
2003     int chr_h= 16>>s->chroma_y_shift;
2004     int i, j;
2005     MpegEncContext best_s, backup_s;
2006     uint8_t bit_buf[2][MAX_MB_BYTES];
2007     uint8_t bit_buf2[2][MAX_MB_BYTES];
2008     uint8_t bit_buf_tex[2][MAX_MB_BYTES];
2009     PutBitContext pb[2], pb2[2], tex_pb[2];
2010 //printf("%d->%d\n", s->resync_mb_y, s->end_mb_y);
2011
2012     ff_check_alignment();
2013
2014     for(i=0; i<2; i++){
2015         init_put_bits(&pb    [i], bit_buf    [i], MAX_MB_BYTES);
2016         init_put_bits(&pb2   [i], bit_buf2   [i], MAX_MB_BYTES);
2017         init_put_bits(&tex_pb[i], bit_buf_tex[i], MAX_MB_BYTES);
2018     }
2019
2020     s->last_bits= put_bits_count(&s->pb);
2021     s->mv_bits=0;
2022     s->misc_bits=0;
2023     s->i_tex_bits=0;
2024     s->p_tex_bits=0;
2025     s->i_count=0;
2026     s->f_count=0;
2027     s->b_count=0;
2028     s->skip_count=0;
2029
2030     for(i=0; i<3; i++){
2031         /* init last dc values */
2032         /* note: quant matrix value (8) is implied here */
2033         s->last_dc[i] = 128 << s->intra_dc_precision;
2034
2035         s->current_picture.error[i] = 0;
2036     }
2037     s->mb_skip_run = 0;
2038     memset(s->last_mv, 0, sizeof(s->last_mv));
2039
2040     s->last_mv_dir = 0;
2041
2042     switch(s->codec_id){
2043     case CODEC_ID_H263:
2044     case CODEC_ID_H263P:
2045     case CODEC_ID_FLV1:
2046         if (CONFIG_H263_ENCODER)
2047             s->gob_index = ff_h263_get_gob_height(s);
2048         break;
2049     case CODEC_ID_MPEG4:
2050         if(CONFIG_MPEG4_ENCODER && s->partitioned_frame)
2051             ff_mpeg4_init_partitions(s);
2052         break;
2053     }
2054
2055     s->resync_mb_x=0;
2056     s->resync_mb_y=0;
2057     s->first_slice_line = 1;
2058     s->ptr_lastgob = s->pb.buf;
2059     for(mb_y= s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
2060 //    printf("row %d at %X\n", s->mb_y, (int)s);
2061         s->mb_x=0;
2062         s->mb_y= mb_y;
2063
2064         ff_set_qscale(s, s->qscale);
2065         ff_init_block_index(s);
2066
2067         for(mb_x=0; mb_x < s->mb_width; mb_x++) {
2068             int xy= mb_y*s->mb_stride + mb_x; // removed const, H261 needs to adjust this
2069             int mb_type= s->mb_type[xy];
2070 //            int d;
2071             int dmin= INT_MAX;
2072             int dir;
2073
2074             if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < MAX_MB_BYTES){
2075                 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
2076                 return -1;
2077             }
2078             if(s->data_partitioning){
2079                 if(   s->pb2   .buf_end - s->pb2   .buf - (put_bits_count(&s->    pb2)>>3) < MAX_MB_BYTES
2080                    || s->tex_pb.buf_end - s->tex_pb.buf - (put_bits_count(&s->tex_pb )>>3) < MAX_MB_BYTES){
2081                     av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
2082                     return -1;
2083                 }
2084             }
2085
2086             s->mb_x = mb_x;
2087             s->mb_y = mb_y;  // moved into loop, can get changed by H.261
2088             ff_update_block_index(s);
2089
2090             if(CONFIG_H261_ENCODER && s->codec_id == CODEC_ID_H261){
2091                 ff_h261_reorder_mb_index(s);
2092                 xy= s->mb_y*s->mb_stride + s->mb_x;
2093                 mb_type= s->mb_type[xy];
2094             }
2095
2096             /* write gob / video packet header  */
2097             if(s->rtp_mode){
2098                 int current_packet_size, is_gob_start;
2099
2100                 current_packet_size= ((put_bits_count(&s->pb)+7)>>3) - (s->ptr_lastgob - s->pb.buf);
2101
2102                 is_gob_start= s->avctx->rtp_payload_size && current_packet_size >= s->avctx->rtp_payload_size && mb_y + mb_x>0;
2103
2104                 if(s->start_mb_y == mb_y && mb_y > 0 && mb_x==0) is_gob_start=1;
2105
2106                 switch(s->codec_id){
2107                 case CODEC_ID_H263:
2108                 case CODEC_ID_H263P:
2109                     if(!s->h263_slice_structured)
2110                         if(s->mb_x || s->mb_y%s->gob_index) is_gob_start=0;
2111                     break;
2112                 case CODEC_ID_MPEG2VIDEO:
2113                     if(s->mb_x==0 && s->mb_y!=0) is_gob_start=1;
2114                 case CODEC_ID_MPEG1VIDEO:
2115                     if(s->mb_skip_run) is_gob_start=0;
2116                     break;
2117                 }
2118
2119                 if(is_gob_start){
2120                     if(s->start_mb_y != mb_y || mb_x!=0){
2121                         write_slice_end(s);
2122
2123                         if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4 && s->partitioned_frame){
2124                             ff_mpeg4_init_partitions(s);
2125                         }
2126                     }
2127
2128                     assert((put_bits_count(&s->pb)&7) == 0);
2129                     current_packet_size= put_bits_ptr(&s->pb) - s->ptr_lastgob;
2130
2131                     if(s->avctx->error_rate && s->resync_mb_x + s->resync_mb_y > 0){
2132                         int r= put_bits_count(&s->pb)/8 + s->picture_number + 16 + s->mb_x + s->mb_y;
2133                         int d= 100 / s->avctx->error_rate;
2134                         if(r % d == 0){
2135                             current_packet_size=0;
2136 #ifndef ALT_BITSTREAM_WRITER
2137                             s->pb.buf_ptr= s->ptr_lastgob;
2138 #endif
2139                             assert(put_bits_ptr(&s->pb) == s->ptr_lastgob);
2140                         }
2141                     }
2142
2143                     if (s->avctx->rtp_callback){
2144                         int number_mb = (mb_y - s->resync_mb_y)*s->mb_width + mb_x - s->resync_mb_x;
2145                         s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, current_packet_size, number_mb);
2146                     }
2147
2148                     switch(s->codec_id){
2149                     case CODEC_ID_MPEG4:
2150                         if (CONFIG_MPEG4_ENCODER) {
2151                             ff_mpeg4_encode_video_packet_header(s);
2152                             ff_mpeg4_clean_buffers(s);
2153                         }
2154                     break;
2155                     case CODEC_ID_MPEG1VIDEO:
2156                     case CODEC_ID_MPEG2VIDEO:
2157                         if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER) {
2158                             ff_mpeg1_encode_slice_header(s);
2159                             ff_mpeg1_clean_buffers(s);
2160                         }
2161                     break;
2162                     case CODEC_ID_H263:
2163                     case CODEC_ID_H263P:
2164                         if (CONFIG_H263_ENCODER)
2165                             h263_encode_gob_header(s, mb_y);
2166                     break;
2167                     }
2168
2169                     if(s->flags&CODEC_FLAG_PASS1){
2170                         int bits= put_bits_count(&s->pb);
2171                         s->misc_bits+= bits - s->last_bits;
2172                         s->last_bits= bits;
2173                     }
2174
2175                     s->ptr_lastgob += current_packet_size;
2176                     s->first_slice_line=1;
2177                     s->resync_mb_x=mb_x;
2178                     s->resync_mb_y=mb_y;
2179                 }
2180             }
2181
2182             if(  (s->resync_mb_x   == s->mb_x)
2183                && s->resync_mb_y+1 == s->mb_y){
2184                 s->first_slice_line=0;
2185             }
2186
2187             s->mb_skipped=0;
2188             s->dquant=0; //only for QP_RD
2189
2190             if(mb_type & (mb_type-1) || (s->flags & CODEC_FLAG_QP_RD)){ // more than 1 MB type possible or CODEC_FLAG_QP_RD
2191                 int next_block=0;
2192                 int pb_bits_count, pb2_bits_count, tex_pb_bits_count;
2193
2194                 copy_context_before_encode(&backup_s, s, -1);
2195                 backup_s.pb= s->pb;
2196                 best_s.data_partitioning= s->data_partitioning;
2197                 best_s.partitioned_frame= s->partitioned_frame;
2198                 if(s->data_partitioning){
2199                     backup_s.pb2= s->pb2;
2200                     backup_s.tex_pb= s->tex_pb;
2201                 }
2202
2203                 if(mb_type&CANDIDATE_MB_TYPE_INTER){
2204                     s->mv_dir = MV_DIR_FORWARD;
2205                     s->mv_type = MV_TYPE_16X16;
2206                     s->mb_intra= 0;
2207                     s->mv[0][0][0] = s->p_mv_table[xy][0];
2208                     s->mv[0][0][1] = s->p_mv_table[xy][1];
2209                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER, pb, pb2, tex_pb,
2210                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2211                 }
2212                 if(mb_type&CANDIDATE_MB_TYPE_INTER_I){
2213                     s->mv_dir = MV_DIR_FORWARD;
2214                     s->mv_type = MV_TYPE_FIELD;
2215                     s->mb_intra= 0;
2216                     for(i=0; i<2; i++){
2217                         j= s->field_select[0][i] = s->p_field_select_table[i][xy];
2218                         s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
2219                         s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
2220                     }
2221                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER_I, pb, pb2, tex_pb,
2222                                  &dmin, &next_block, 0, 0);
2223                 }
2224                 if(mb_type&CANDIDATE_MB_TYPE_SKIPPED){
2225                     s->mv_dir = MV_DIR_FORWARD;
2226                     s->mv_type = MV_TYPE_16X16;
2227                     s->mb_intra= 0;
2228                     s->mv[0][0][0] = 0;
2229                     s->mv[0][0][1] = 0;
2230                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_SKIPPED, pb, pb2, tex_pb,
2231                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2232                 }
2233                 if(mb_type&CANDIDATE_MB_TYPE_INTER4V){
2234                     s->mv_dir = MV_DIR_FORWARD;
2235                     s->mv_type = MV_TYPE_8X8;
2236                     s->mb_intra= 0;
2237                     for(i=0; i<4; i++){
2238                         s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
2239                         s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
2240                     }
2241                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER4V, pb, pb2, tex_pb,
2242                                  &dmin, &next_block, 0, 0);
2243                 }
2244                 if(mb_type&CANDIDATE_MB_TYPE_FORWARD){
2245                     s->mv_dir = MV_DIR_FORWARD;
2246                     s->mv_type = MV_TYPE_16X16;
2247                     s->mb_intra= 0;
2248                     s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
2249                     s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2250                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD, pb, pb2, tex_pb,
2251                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2252                 }
2253                 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD){
2254                     s->mv_dir = MV_DIR_BACKWARD;
2255                     s->mv_type = MV_TYPE_16X16;
2256                     s->mb_intra= 0;
2257                     s->mv[1][0][0] = s->b_back_mv_table[xy][0];
2258                     s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2259                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD, pb, pb2, tex_pb,
2260                                  &dmin, &next_block, s->mv[1][0][0], s->mv[1][0][1]);
2261                 }
2262                 if(mb_type&CANDIDATE_MB_TYPE_BIDIR){
2263                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2264                     s->mv_type = MV_TYPE_16X16;
2265                     s->mb_intra= 0;
2266                     s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
2267                     s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
2268                     s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
2269                     s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2270                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR, pb, pb2, tex_pb,
2271                                  &dmin, &next_block, 0, 0);
2272                 }
2273                 if(mb_type&CANDIDATE_MB_TYPE_FORWARD_I){
2274                     s->mv_dir = MV_DIR_FORWARD;
2275                     s->mv_type = MV_TYPE_FIELD;
2276                     s->mb_intra= 0;
2277                     for(i=0; i<2; i++){
2278                         j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
2279                         s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
2280                         s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
2281                     }
2282                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD_I, pb, pb2, tex_pb,
2283                                  &dmin, &next_block, 0, 0);
2284                 }
2285                 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD_I){
2286                     s->mv_dir = MV_DIR_BACKWARD;
2287                     s->mv_type = MV_TYPE_FIELD;
2288                     s->mb_intra= 0;
2289                     for(i=0; i<2; i++){
2290                         j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
2291                         s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
2292                         s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
2293                     }
2294                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD_I, pb, pb2, tex_pb,
2295                                  &dmin, &next_block, 0, 0);
2296                 }
2297                 if(mb_type&CANDIDATE_MB_TYPE_BIDIR_I){
2298                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2299                     s->mv_type = MV_TYPE_FIELD;
2300                     s->mb_intra= 0;
2301                     for(dir=0; dir<2; dir++){
2302                         for(i=0; i<2; i++){
2303                             j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
2304                             s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
2305                             s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
2306                         }
2307                     }
2308                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR_I, pb, pb2, tex_pb,
2309                                  &dmin, &next_block, 0, 0);
2310                 }
2311                 if(mb_type&CANDIDATE_MB_TYPE_INTRA){
2312                     s->mv_dir = 0;
2313                     s->mv_type = MV_TYPE_16X16;
2314                     s->mb_intra= 1;
2315                     s->mv[0][0][0] = 0;
2316                     s->mv[0][0][1] = 0;
2317                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTRA, pb, pb2, tex_pb,
2318                                  &dmin, &next_block, 0, 0);
2319                     if(s->h263_pred || s->h263_aic){
2320                         if(best_s.mb_intra)
2321                             s->mbintra_table[mb_x + mb_y*s->mb_stride]=1;
2322                         else
2323                             ff_clean_intra_table_entries(s); //old mode?
2324                     }
2325                 }
2326
2327                 if((s->flags & CODEC_FLAG_QP_RD) && dmin < INT_MAX){
2328                     if(best_s.mv_type==MV_TYPE_16X16){ //FIXME move 4mv after QPRD
2329                         const int last_qp= backup_s.qscale;
2330                         int qpi, qp, dc[6];
2331                         DCTELEM ac[6][16];
2332                         const int mvdir= (best_s.mv_dir&MV_DIR_BACKWARD) ? 1 : 0;
2333                         static const int dquant_tab[4]={-1,1,-2,2};
2334
2335                         assert(backup_s.dquant == 0);
2336
2337                         //FIXME intra
2338                         s->mv_dir= best_s.mv_dir;
2339                         s->mv_type = MV_TYPE_16X16;
2340                         s->mb_intra= best_s.mb_intra;
2341                         s->mv[0][0][0] = best_s.mv[0][0][0];
2342                         s->mv[0][0][1] = best_s.mv[0][0][1];
2343                         s->mv[1][0][0] = best_s.mv[1][0][0];
2344                         s->mv[1][0][1] = best_s.mv[1][0][1];
2345
2346                         qpi = s->pict_type == FF_B_TYPE ? 2 : 0;
2347                         for(; qpi<4; qpi++){
2348                             int dquant= dquant_tab[qpi];
2349                             qp= last_qp + dquant;
2350                             if(qp < s->avctx->qmin || qp > s->avctx->qmax)
2351                                 continue;
2352                             backup_s.dquant= dquant;
2353                             if(s->mb_intra && s->dc_val[0]){
2354                                 for(i=0; i<6; i++){
2355                                     dc[i]= s->dc_val[0][ s->block_index[i] ];
2356                                     memcpy(ac[i], s->ac_val[0][s->block_index[i]], sizeof(DCTELEM)*16);
2357                                 }
2358                             }
2359
2360                             encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
2361                                          &dmin, &next_block, s->mv[mvdir][0][0], s->mv[mvdir][0][1]);
2362                             if(best_s.qscale != qp){
2363                                 if(s->mb_intra && s->dc_val[0]){
2364                                     for(i=0; i<6; i++){
2365                                         s->dc_val[0][ s->block_index[i] ]= dc[i];
2366                                         memcpy(s->ac_val[0][s->block_index[i]], ac[i], sizeof(DCTELEM)*16);
2367                                     }
2368                                 }
2369                             }
2370                         }
2371                     }
2372                 }
2373                 if(CONFIG_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT){
2374                     int mx= s->b_direct_mv_table[xy][0];
2375                     int my= s->b_direct_mv_table[xy][1];
2376
2377                     backup_s.dquant = 0;
2378                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2379                     s->mb_intra= 0;
2380                     ff_mpeg4_set_direct_mv(s, mx, my);
2381                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
2382                                  &dmin, &next_block, mx, my);
2383                 }
2384                 if(CONFIG_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT0){
2385                     backup_s.dquant = 0;
2386                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2387                     s->mb_intra= 0;
2388                     ff_mpeg4_set_direct_mv(s, 0, 0);
2389                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
2390                                  &dmin, &next_block, 0, 0);
2391                 }
2392                 if(!best_s.mb_intra && s->flags2&CODEC_FLAG2_SKIP_RD){
2393                     int coded=0;
2394                     for(i=0; i<6; i++)
2395                         coded |= s->block_last_index[i];
2396                     if(coded){
2397                         int mx,my;
2398                         memcpy(s->mv, best_s.mv, sizeof(s->mv));
2399                         if(CONFIG_MPEG4_ENCODER && best_s.mv_dir & MV_DIRECT){
2400                             mx=my=0; //FIXME find the one we actually used
2401                             ff_mpeg4_set_direct_mv(s, mx, my);
2402                         }else if(best_s.mv_dir&MV_DIR_BACKWARD){
2403                             mx= s->mv[1][0][0];
2404                             my= s->mv[1][0][1];
2405                         }else{
2406                             mx= s->mv[0][0][0];
2407                             my= s->mv[0][0][1];
2408                         }
2409
2410                         s->mv_dir= best_s.mv_dir;
2411                         s->mv_type = best_s.mv_type;
2412                         s->mb_intra= 0;
2413 /*                        s->mv[0][0][0] = best_s.mv[0][0][0];
2414                         s->mv[0][0][1] = best_s.mv[0][0][1];
2415                         s->mv[1][0][0] = best_s.mv[1][0][0];
2416                         s->mv[1][0][1] = best_s.mv[1][0][1];*/
2417                         backup_s.dquant= 0;
2418                         s->skipdct=1;
2419                         encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
2420                                         &dmin, &next_block, mx, my);
2421                         s->skipdct=0;
2422                     }
2423                 }
2424
2425                 s->current_picture.qscale_table[xy]= best_s.qscale;
2426
2427                 copy_context_after_encode(s, &best_s, -1);
2428
2429                 pb_bits_count= put_bits_count(&s->pb);
2430                 flush_put_bits(&s->pb);
2431                 ff_copy_bits(&backup_s.pb, bit_buf[next_block^1], pb_bits_count);
2432                 s->pb= backup_s.pb;
2433
2434                 if(s->data_partitioning){
2435                     pb2_bits_count= put_bits_count(&s->pb2);
2436                     flush_put_bits(&s->pb2);
2437                     ff_copy_bits(&backup_s.pb2, bit_buf2[next_block^1], pb2_bits_count);
2438                     s->pb2= backup_s.pb2;
2439
2440                     tex_pb_bits_count= put_bits_count(&s->tex_pb);
2441                     flush_put_bits(&s->tex_pb);
2442                     ff_copy_bits(&backup_s.tex_pb, bit_buf_tex[next_block^1], tex_pb_bits_count);
2443                     s->tex_pb= backup_s.tex_pb;
2444                 }
2445                 s->last_bits= put_bits_count(&s->pb);
2446
2447                 if (CONFIG_ANY_H263_ENCODER &&
2448                     s->out_format == FMT_H263 && s->pict_type!=FF_B_TYPE)
2449                     ff_h263_update_motion_val(s);
2450
2451                 if(next_block==0){ //FIXME 16 vs linesize16
2452                     s->dsp.put_pixels_tab[0][0](s->dest[0], s->rd_scratchpad                     , s->linesize  ,16);
2453                     s->dsp.put_pixels_tab[1][0](s->dest[1], s->rd_scratchpad + 16*s->linesize    , s->uvlinesize, 8);
2454                     s->dsp.put_pixels_tab[1][0](s->dest[2], s->rd_scratchpad + 16*s->linesize + 8, s->uvlinesize, 8);
2455                 }
2456
2457                 if(s->avctx->mb_decision == FF_MB_DECISION_BITS)
2458                     MPV_decode_mb(s, s->block);
2459             } else {
2460                 int motion_x = 0, motion_y = 0;
2461                 s->mv_type=MV_TYPE_16X16;
2462                 // only one MB-Type possible
2463
2464                 switch(mb_type){
2465                 case CANDIDATE_MB_TYPE_INTRA:
2466                     s->mv_dir = 0;
2467                     s->mb_intra= 1;
2468                     motion_x= s->mv[0][0][0] = 0;
2469                     motion_y= s->mv[0][0][1] = 0;
2470                     break;
2471                 case CANDIDATE_MB_TYPE_INTER:
2472                     s->mv_dir = MV_DIR_FORWARD;
2473                     s->mb_intra= 0;
2474                     motion_x= s->mv[0][0][0] = s->p_mv_table[xy][0];
2475                     motion_y= s->mv[0][0][1] = s->p_mv_table[xy][1];
2476                     break;
2477                 case CANDIDATE_MB_TYPE_INTER_I:
2478                     s->mv_dir = MV_DIR_FORWARD;
2479                     s->mv_type = MV_TYPE_FIELD;
2480                     s->mb_intra= 0;
2481                     for(i=0; i<2; i++){
2482                         j= s->field_select[0][i] = s->p_field_select_table[i][xy];
2483                         s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
2484                         s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
2485                     }
2486                     break;
2487                 case CANDIDATE_MB_TYPE_INTER4V:
2488                     s->mv_dir = MV_DIR_FORWARD;
2489                     s->mv_type = MV_TYPE_8X8;
2490                     s->mb_intra= 0;
2491                     for(i=0; i<4; i++){
2492                         s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
2493                         s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
2494                     }
2495                     break;
2496                 case CANDIDATE_MB_TYPE_DIRECT:
2497                     if (CONFIG_MPEG4_ENCODER) {
2498                         s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
2499                         s->mb_intra= 0;
2500                         motion_x=s->b_direct_mv_table[xy][0];
2501                         motion_y=s->b_direct_mv_table[xy][1];
2502                         ff_mpeg4_set_direct_mv(s, motion_x, motion_y);
2503                     }
2504                     break;
2505                 case CANDIDATE_MB_TYPE_DIRECT0:
2506                     if (CONFIG_MPEG4_ENCODER) {
2507                         s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
2508                         s->mb_intra= 0;
2509                         ff_mpeg4_set_direct_mv(s, 0, 0);
2510                     }
2511                     break;
2512                 case CANDIDATE_MB_TYPE_BIDIR:
2513                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2514                     s->mb_intra= 0;
2515                     s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
2516                     s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
2517                     s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
2518                     s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2519                     break;
2520                 case CANDIDATE_MB_TYPE_BACKWARD:
2521                     s->mv_dir = MV_DIR_BACKWARD;
2522                     s->mb_intra= 0;
2523                     motion_x= s->mv[1][0][0] = s->b_back_mv_table[xy][0];
2524                     motion_y= s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2525                     break;
2526                 case CANDIDATE_MB_TYPE_FORWARD:
2527                     s->mv_dir = MV_DIR_FORWARD;
2528                     s->mb_intra= 0;
2529                     motion_x= s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
2530                     motion_y= s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2531 //                    printf(" %d %d ", motion_x, motion_y);
2532                     break;
2533                 case CANDIDATE_MB_TYPE_FORWARD_I:
2534                     s->mv_dir = MV_DIR_FORWARD;
2535                     s->mv_type = MV_TYPE_FIELD;
2536                     s->mb_intra= 0;
2537                     for(i=0; i<2; i++){
2538                         j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
2539                         s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
2540                         s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
2541                     }
2542                     break;
2543                 case CANDIDATE_MB_TYPE_BACKWARD_I:
2544                     s->mv_dir = MV_DIR_BACKWARD;
2545                     s->mv_type = MV_TYPE_FIELD;
2546                     s->mb_intra= 0;
2547                     for(i=0; i<2; i++){
2548                         j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
2549                         s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
2550                         s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
2551                     }
2552                     break;
2553                 case CANDIDATE_MB_TYPE_BIDIR_I:
2554                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2555                     s->mv_type = MV_TYPE_FIELD;
2556                     s->mb_intra= 0;
2557                     for(dir=0; dir<2; dir++){
2558                         for(i=0; i<2; i++){
2559                             j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
2560                             s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
2561                             s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
2562                         }
2563                     }
2564                     break;
2565                 default:
2566                     av_log(s->avctx, AV_LOG_ERROR, "illegal MB type\n");
2567                 }
2568
2569                 encode_mb(s, motion_x, motion_y);
2570
2571                 // RAL: Update last macroblock type
2572                 s->last_mv_dir = s->mv_dir;
2573
2574                 if (CONFIG_ANY_H263_ENCODER &&
2575                     s->out_format == FMT_H263 && s->pict_type!=FF_B_TYPE)
2576                     ff_h263_update_motion_val(s);
2577
2578                 MPV_decode_mb(s, s->block);
2579             }
2580
2581             /* clean the MV table in IPS frames for direct mode in B frames */
2582             if(s->mb_intra /* && I,P,S_TYPE */){
2583                 s->p_mv_table[xy][0]=0;
2584                 s->p_mv_table[xy][1]=0;
2585             }
2586
2587             if(s->flags&CODEC_FLAG_PSNR){
2588                 int w= 16;
2589                 int h= 16;
2590
2591                 if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
2592                 if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
2593
2594                 s->current_picture.error[0] += sse(
2595                     s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16,
2596                     s->dest[0], w, h, s->linesize);
2597                 s->current_picture.error[1] += sse(
2598                     s, s->new_picture.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*chr_h,
2599                     s->dest[1], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
2600                 s->current_picture.error[2] += sse(
2601                     s, s->new_picture.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*chr_h,
2602                     s->dest[2], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
2603             }
2604             if(s->loop_filter){
2605                 if(CONFIG_ANY_H263_ENCODER && s->out_format == FMT_H263)
2606                     ff_h263_loop_filter(s);
2607             }
2608 //printf("MB %d %d bits\n", s->mb_x+s->mb_y*s->mb_stride, put_bits_count(&s->pb));
2609         }
2610     }
2611
2612     //not beautiful here but we must write it before flushing so it has to be here
2613     if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version && s->msmpeg4_version<4 && s->pict_type == FF_I_TYPE)
2614         msmpeg4_encode_ext_header(s);
2615
2616     write_slice_end(s);
2617
2618     /* Send the last GOB if RTP */
2619     if (s->avctx->rtp_callback) {
2620         int number_mb = (mb_y - s->resync_mb_y)*s->mb_width - s->resync_mb_x;
2621         pdif = put_bits_ptr(&s->pb) - s->ptr_lastgob;
2622         /* Call the RTP callback to send the last GOB */
2623         emms_c();
2624         s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, pdif, number_mb);
2625     }
2626
2627     return 0;
2628 }
2629
2630 #define MERGE(field) dst->field += src->field; src->field=0
2631 static void merge_context_after_me(MpegEncContext *dst, MpegEncContext *src){
2632     MERGE(me.scene_change_score);
2633     MERGE(me.mc_mb_var_sum_temp);
2634     MERGE(me.mb_var_sum_temp);
2635 }
2636
2637 static void merge_context_after_encode(MpegEncContext *dst, MpegEncContext *src){
2638     int i;
2639
2640     MERGE(dct_count[0]); //note, the other dct vars are not part of the context
2641     MERGE(dct_count[1]);
2642     MERGE(mv_bits);
2643     MERGE(i_tex_bits);
2644     MERGE(p_tex_bits);
2645     MERGE(i_count);
2646     MERGE(f_count);
2647     MERGE(b_count);
2648     MERGE(skip_count);
2649     MERGE(misc_bits);
2650     MERGE(error_count);
2651     MERGE(padding_bug_score);
2652     MERGE(current_picture.error[0]);
2653     MERGE(current_picture.error[1]);
2654     MERGE(current_picture.error[2]);
2655
2656     if(dst->avctx->noise_reduction){
2657         for(i=0; i<64; i++){
2658             MERGE(dct_error_sum[0][i]);
2659             MERGE(dct_error_sum[1][i]);
2660         }
2661     }
2662
2663     assert(put_bits_count(&src->pb) % 8 ==0);
2664     assert(put_bits_count(&dst->pb) % 8 ==0);
2665     ff_copy_bits(&dst->pb, src->pb.buf, put_bits_count(&src->pb));
2666     flush_put_bits(&dst->pb);
2667 }
2668
2669 static int estimate_qp(MpegEncContext *s, int dry_run){
2670     if (s->next_lambda){
2671         s->current_picture_ptr->quality=
2672         s->current_picture.quality = s->next_lambda;
2673         if(!dry_run) s->next_lambda= 0;
2674     } else if (!s->fixed_qscale) {
2675         s->current_picture_ptr->quality=
2676         s->current_picture.quality = ff_rate_estimate_qscale(s, dry_run);
2677         if (s->current_picture.quality < 0)
2678             return -1;
2679     }
2680
2681     if(s->adaptive_quant){
2682         switch(s->codec_id){
2683         case CODEC_ID_MPEG4:
2684             if (CONFIG_MPEG4_ENCODER)
2685                 ff_clean_mpeg4_qscales(s);
2686             break;
2687         case CODEC_ID_H263:
2688         case CODEC_ID_H263P:
2689         case CODEC_ID_FLV1:
2690             if (CONFIG_H263_ENCODER)
2691                 ff_clean_h263_qscales(s);
2692             break;
2693         }
2694
2695         s->lambda= s->lambda_table[0];
2696         //FIXME broken
2697     }else
2698         s->lambda= s->current_picture.quality;
2699 //printf("%d %d\n", s->avctx->global_quality, s->current_picture.quality);
2700     update_qscale(s);
2701     return 0;
2702 }
2703
2704 /* must be called before writing the header */
2705 static void set_frame_distances(MpegEncContext * s){
2706     assert(s->current_picture_ptr->pts != AV_NOPTS_VALUE);
2707     s->time= s->current_picture_ptr->pts*s->avctx->time_base.num;
2708
2709     if(s->pict_type==FF_B_TYPE){
2710         s->pb_time= s->pp_time - (s->last_non_b_time - s->time);
2711         assert(s->pb_time > 0 && s->pb_time < s->pp_time);
2712     }else{
2713         s->pp_time= s->time - s->last_non_b_time;
2714         s->last_non_b_time= s->time;
2715         assert(s->picture_number==0 || s->pp_time > 0);
2716     }
2717 }
2718
2719 static int encode_picture(MpegEncContext *s, int picture_number)
2720 {
2721     int i;
2722     int bits;
2723
2724     s->picture_number = picture_number;
2725
2726     /* Reset the average MB variance */
2727     s->me.mb_var_sum_temp    =
2728     s->me.mc_mb_var_sum_temp = 0;
2729
2730     /* we need to initialize some time vars before we can encode b-frames */
2731     // RAL: Condition added for MPEG1VIDEO
2732     if (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO || (s->h263_pred && !s->h263_msmpeg4))
2733         set_frame_distances(s);
2734     if(CONFIG_MPEG4_ENCODER && s->codec_id == CODEC_ID_MPEG4)
2735         ff_set_mpeg4_time(s);
2736
2737     s->me.scene_change_score=0;
2738
2739 //    s->lambda= s->current_picture_ptr->quality; //FIXME qscale / ... stuff for ME rate distortion
2740
2741     if(s->pict_type==FF_I_TYPE){
2742         if(s->msmpeg4_version >= 3) s->no_rounding=1;
2743         else                        s->no_rounding=0;
2744     }else if(s->pict_type!=FF_B_TYPE){
2745         if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
2746             s->no_rounding ^= 1;
2747     }
2748
2749     if(s->flags & CODEC_FLAG_PASS2){
2750         if (estimate_qp(s,1) < 0)
2751             return -1;
2752         ff_get_2pass_fcode(s);
2753     }else if(!(s->flags & CODEC_FLAG_QSCALE)){
2754         if(s->pict_type==FF_B_TYPE)
2755             s->lambda= s->last_lambda_for[s->pict_type];
2756         else
2757             s->lambda= s->last_lambda_for[s->last_non_b_pict_type];
2758         update_qscale(s);
2759     }
2760
2761     s->mb_intra=0; //for the rate distortion & bit compare functions
2762     for(i=1; i<s->avctx->thread_count; i++){
2763         ff_update_duplicate_context(s->thread_context[i], s);
2764     }
2765
2766     if(ff_init_me(s)<0)
2767         return -1;
2768
2769     /* Estimate motion for every MB */
2770     if(s->pict_type != FF_I_TYPE){
2771         s->lambda = (s->lambda * s->avctx->me_penalty_compensation + 128)>>8;
2772         s->lambda2= (s->lambda2* (int64_t)s->avctx->me_penalty_compensation + 128)>>8;
2773         if(s->pict_type != FF_B_TYPE && s->avctx->me_threshold==0){
2774             if((s->avctx->pre_me && s->last_non_b_pict_type==FF_I_TYPE) || s->avctx->pre_me==2){
2775                 s->avctx->execute(s->avctx, pre_estimate_motion_thread, &s->thread_context[0], NULL, s->avctx->thread_count, sizeof(void*));
2776             }
2777         }
2778
2779         s->avctx->execute(s->avctx, estimate_motion_thread, &s->thread_context[0], NULL, s->avctx->thread_count, sizeof(void*));
2780     }else /* if(s->pict_type == FF_I_TYPE) */{
2781         /* I-Frame */
2782         for(i=0; i<s->mb_stride*s->mb_height; i++)
2783             s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
2784
2785         if(!s->fixed_qscale){
2786             /* finding spatial complexity for I-frame rate control */
2787             s->avctx->execute(s->avctx, mb_var_thread, &s->thread_context[0], NULL, s->avctx->thread_count, sizeof(void*));
2788         }
2789     }
2790     for(i=1; i<s->avctx->thread_count; i++){
2791         merge_context_after_me(s, s->thread_context[i]);
2792     }
2793     s->current_picture.mc_mb_var_sum= s->current_picture_ptr->mc_mb_var_sum= s->me.mc_mb_var_sum_temp;
2794     s->current_picture.   mb_var_sum= s->current_picture_ptr->   mb_var_sum= s->me.   mb_var_sum_temp;
2795     emms_c();
2796
2797     if(s->me.scene_change_score > s->avctx->scenechange_threshold && s->pict_type == FF_P_TYPE){
2798         s->pict_type= FF_I_TYPE;
2799         for(i=0; i<s->mb_stride*s->mb_height; i++)
2800             s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
2801 //printf("Scene change detected, encoding as I Frame %d %d\n", s->current_picture.mb_var_sum, s->current_picture.mc_mb_var_sum);
2802     }
2803
2804     if(!s->umvplus){
2805         if(s->pict_type==FF_P_TYPE || s->pict_type==FF_S_TYPE) {
2806             s->f_code= ff_get_best_fcode(s, s->p_mv_table, CANDIDATE_MB_TYPE_INTER);
2807
2808             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2809                 int a,b;
2810                 a= ff_get_best_fcode(s, s->p_field_mv_table[0][0], CANDIDATE_MB_TYPE_INTER_I); //FIXME field_select
2811                 b= ff_get_best_fcode(s, s->p_field_mv_table[1][1], CANDIDATE_MB_TYPE_INTER_I);
2812                 s->f_code= FFMAX3(s->f_code, a, b);
2813             }
2814
2815             ff_fix_long_p_mvs(s);
2816             ff_fix_long_mvs(s, NULL, 0, s->p_mv_table, s->f_code, CANDIDATE_MB_TYPE_INTER, 0);
2817             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2818                 int j;
2819                 for(i=0; i<2; i++){
2820                     for(j=0; j<2; j++)
2821                         ff_fix_long_mvs(s, s->p_field_select_table[i], j,
2822                                         s->p_field_mv_table[i][j], s->f_code, CANDIDATE_MB_TYPE_INTER_I, 0);
2823                 }
2824             }
2825         }
2826
2827         if(s->pict_type==FF_B_TYPE){
2828             int a, b;
2829
2830             a = ff_get_best_fcode(s, s->b_forw_mv_table, CANDIDATE_MB_TYPE_FORWARD);
2831             b = ff_get_best_fcode(s, s->b_bidir_forw_mv_table, CANDIDATE_MB_TYPE_BIDIR);
2832             s->f_code = FFMAX(a, b);
2833
2834             a = ff_get_best_fcode(s, s->b_back_mv_table, CANDIDATE_MB_TYPE_BACKWARD);
2835             b = ff_get_best_fcode(s, s->b_bidir_back_mv_table, CANDIDATE_MB_TYPE_BIDIR);
2836             s->b_code = FFMAX(a, b);
2837
2838             ff_fix_long_mvs(s, NULL, 0, s->b_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_FORWARD, 1);
2839             ff_fix_long_mvs(s, NULL, 0, s->b_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BACKWARD, 1);
2840             ff_fix_long_mvs(s, NULL, 0, s->b_bidir_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_BIDIR, 1);
2841             ff_fix_long_mvs(s, NULL, 0, s->b_bidir_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BIDIR, 1);
2842             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2843                 int dir, j;
2844                 for(dir=0; dir<2; dir++){
2845                     for(i=0; i<2; i++){
2846                         for(j=0; j<2; j++){
2847                             int type= dir ? (CANDIDATE_MB_TYPE_BACKWARD_I|CANDIDATE_MB_TYPE_BIDIR_I)
2848                                           : (CANDIDATE_MB_TYPE_FORWARD_I |CANDIDATE_MB_TYPE_BIDIR_I);
2849                             ff_fix_long_mvs(s, s->b_field_select_table[dir][i], j,
2850                                             s->b_field_mv_table[dir][i][j], dir ? s->b_code : s->f_code, type, 1);
2851                         }
2852                     }
2853                 }
2854             }
2855         }
2856     }
2857
2858     if (estimate_qp(s, 0) < 0)
2859         return -1;
2860
2861     if(s->qscale < 3 && s->max_qcoeff<=128 && s->pict_type==FF_I_TYPE && !(s->flags & CODEC_FLAG_QSCALE))
2862         s->qscale= 3; //reduce clipping problems
2863
2864     if (s->out_format == FMT_MJPEG) {
2865         /* for mjpeg, we do include qscale in the matrix */
2866         s->intra_matrix[0] = ff_mpeg1_default_intra_matrix[0];
2867         for(i=1;i<64;i++){
2868             int j= s->dsp.idct_permutation[i];
2869
2870             s->intra_matrix[j] = av_clip_uint8((ff_mpeg1_default_intra_matrix[i] * s->qscale) >> 3);
2871         }
2872         ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
2873                        s->intra_matrix, s->intra_quant_bias, 8, 8, 1);
2874         s->qscale= 8;
2875     }
2876
2877     //FIXME var duplication
2878     s->current_picture_ptr->key_frame=
2879     s->current_picture.key_frame= s->pict_type == FF_I_TYPE; //FIXME pic_ptr
2880     s->current_picture_ptr->pict_type=
2881     s->current_picture.pict_type= s->pict_type;
2882
2883     if(s->current_picture.key_frame)
2884         s->picture_in_gop_number=0;
2885
2886     s->last_bits= put_bits_count(&s->pb);
2887     switch(s->out_format) {
2888     case FMT_MJPEG:
2889         if (CONFIG_MJPEG_ENCODER)
2890             ff_mjpeg_encode_picture_header(s);
2891         break;
2892     case FMT_H261:
2893         if (CONFIG_H261_ENCODER)
2894             ff_h261_encode_picture_header(s, picture_number);
2895         break;
2896     case FMT_H263:
2897         if (CONFIG_WMV2_ENCODER && s->codec_id == CODEC_ID_WMV2)
2898             ff_wmv2_encode_picture_header(s, picture_number);
2899         else if (CONFIG_MSMPEG4_ENCODER && s->h263_msmpeg4)
2900             msmpeg4_encode_picture_header(s, picture_number);
2901         else if (CONFIG_MPEG4_ENCODER && s->h263_pred)
2902             mpeg4_encode_picture_header(s, picture_number);
2903         else if (CONFIG_RV10_ENCODER && s->codec_id == CODEC_ID_RV10)
2904             rv10_encode_picture_header(s, picture_number);
2905         else if (CONFIG_RV20_ENCODER && s->codec_id == CODEC_ID_RV20)
2906             rv20_encode_picture_header(s, picture_number);
2907         else if (CONFIG_FLV_ENCODER && s->codec_id == CODEC_ID_FLV1)
2908             ff_flv_encode_picture_header(s, picture_number);
2909         else if (CONFIG_ANY_H263_ENCODER)
2910             h263_encode_picture_header(s, picture_number);
2911         break;
2912     case FMT_MPEG1:
2913         if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
2914             mpeg1_encode_picture_header(s, picture_number);
2915         break;
2916     case FMT_H264:
2917         break;
2918     default:
2919         assert(0);
2920     }
2921     bits= put_bits_count(&s->pb);
2922     s->header_bits= bits - s->last_bits;
2923
2924     for(i=1; i<s->avctx->thread_count; i++){
2925         update_duplicate_context_after_me(s->thread_context[i], s);
2926     }
2927     s->avctx->execute(s->avctx, encode_thread, &s->thread_context[0], NULL, s->avctx->thread_count, sizeof(void*));
2928     for(i=1; i<s->avctx->thread_count; i++){
2929         merge_context_after_encode(s, s->thread_context[i]);
2930     }
2931     emms_c();
2932     return 0;
2933 }
2934
2935 void  denoise_dct_c(MpegEncContext *s, DCTELEM *block){
2936     const int intra= s->mb_intra;
2937     int i;
2938
2939     s->dct_count[intra]++;
2940
2941     for(i=0; i<64; i++){
2942         int level= block[i];
2943
2944         if(level){
2945             if(level>0){
2946                 s->dct_error_sum[intra][i] += level;
2947                 level -= s->dct_offset[intra][i];
2948                 if(level<0) level=0;
2949             }else{
2950                 s->dct_error_sum[intra][i] -= level;
2951                 level += s->dct_offset[intra][i];
2952                 if(level>0) level=0;
2953             }
2954             block[i]= level;
2955         }
2956     }
2957 }
2958
2959 int dct_quantize_trellis_c(MpegEncContext *s,
2960                         DCTELEM *block, int n,
2961                         int qscale, int *overflow){
2962     const int *qmat;
2963     const uint8_t *scantable= s->intra_scantable.scantable;
2964     const uint8_t *perm_scantable= s->intra_scantable.permutated;
2965     int max=0;
2966     unsigned int threshold1, threshold2;
2967     int bias=0;
2968     int run_tab[65];
2969     int level_tab[65];
2970     int score_tab[65];
2971     int survivor[65];
2972     int survivor_count;
2973     int last_run=0;
2974     int last_level=0;
2975     int last_score= 0;
2976     int last_i;
2977     int coeff[2][64];
2978     int coeff_count[64];
2979     int qmul, qadd, start_i, last_non_zero, i, dc;
2980     const int esc_length= s->ac_esc_length;
2981     uint8_t * length;
2982     uint8_t * last_length;
2983     const int lambda= s->lambda2 >> (FF_LAMBDA_SHIFT - 6);
2984
2985     s->dsp.fdct (block);
2986
2987     if(s->dct_error_sum)
2988         s->denoise_dct(s, block);
2989     qmul= qscale*16;
2990     qadd= ((qscale-1)|1)*8;
2991
2992     if (s->mb_intra) {
2993         int q;
2994         if (!s->h263_aic) {
2995             if (n < 4)
2996                 q = s->y_dc_scale;
2997             else
2998                 q = s->c_dc_scale;
2999             q = q << 3;
3000         } else{
3001             /* For AIC we skip quant/dequant of INTRADC */
3002             q = 1 << 3;
3003             qadd=0;
3004         }
3005
3006         /* note: block[0] is assumed to be positive */
3007         block[0] = (block[0] + (q >> 1)) / q;
3008         start_i = 1;
3009         last_non_zero = 0;
3010         qmat = s->q_intra_matrix[qscale];
3011         if(s->mpeg_quant || s->out_format == FMT_MPEG1)
3012             bias= 1<<(QMAT_SHIFT-1);
3013         length     = s->intra_ac_vlc_length;
3014         last_length= s->intra_ac_vlc_last_length;
3015     } else {
3016         start_i = 0;
3017         last_non_zero = -1;
3018         qmat = s->q_inter_matrix[qscale];
3019         length     = s->inter_ac_vlc_length;
3020         last_length= s->inter_ac_vlc_last_length;
3021     }
3022     last_i= start_i;
3023
3024     threshold1= (1<<QMAT_SHIFT) - bias - 1;
3025     threshold2= (threshold1<<1);
3026
3027     for(i=63; i>=start_i; i--) {
3028         const int j = scantable[i];
3029         int level = block[j] * qmat[j];
3030
3031         if(((unsigned)(level+threshold1))>threshold2){
3032             last_non_zero = i;
3033             break;
3034         }
3035     }
3036
3037     for(i=start_i; i<=last_non_zero; i++) {
3038         const int j = scantable[i];
3039         int level = block[j] * qmat[j];
3040
3041 //        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
3042 //           || bias-level >= (1<<(QMAT_SHIFT - 3))){
3043         if(((unsigned)(level+threshold1))>threshold2){
3044             if(level>0){
3045                 level= (bias + level)>>QMAT_SHIFT;
3046                 coeff[0][i]= level;
3047                 coeff[1][i]= level-1;
3048 //                coeff[2][k]= level-2;
3049             }else{
3050                 level= (bias - level)>>QMAT_SHIFT;
3051                 coeff[0][i]= -level;
3052                 coeff[1][i]= -level+1;
3053 //                coeff[2][k]= -level+2;
3054             }
3055             coeff_count[i]= FFMIN(level, 2);
3056             assert(coeff_count[i]);
3057             max |=level;
3058         }else{
3059             coeff[0][i]= (level>>31)|1;
3060             coeff_count[i]= 1;
3061         }
3062     }
3063
3064     *overflow= s->max_qcoeff < max; //overflow might have happened
3065
3066     if(last_non_zero < start_i){
3067         memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
3068         return last_non_zero;
3069     }
3070
3071     score_tab[start_i]= 0;
3072     survivor[0]= start_i;
3073     survivor_count= 1;
3074
3075     for(i=start_i; i<=last_non_zero; i++){
3076         int level_index, j, zero_distortion;
3077         int dct_coeff= FFABS(block[ scantable[i] ]);
3078         int best_score=256*256*256*120;
3079
3080         if (   s->dsp.fdct == fdct_ifast
3081 #ifndef FAAN_POSTSCALE
3082             || s->dsp.fdct == ff_faandct
3083 #endif
3084            )
3085             dct_coeff= (dct_coeff*ff_inv_aanscales[ scantable[i] ]) >> 12;
3086         zero_distortion= dct_coeff*dct_coeff;
3087
3088         for(level_index=0; level_index < coeff_count[i]; level_index++){
3089             int distortion;
3090             int level= coeff[level_index][i];
3091             const int alevel= FFABS(level);
3092             int unquant_coeff;
3093
3094             assert(level);
3095
3096             if(s->out_format == FMT_H263){
3097                 unquant_coeff= alevel*qmul + qadd;
3098             }else{ //MPEG1
3099                 j= s->dsp.idct_permutation[ scantable[i] ]; //FIXME optimize
3100                 if(s->mb_intra){
3101                         unquant_coeff = (int)(  alevel  * qscale * s->intra_matrix[j]) >> 3;
3102                         unquant_coeff =   (unquant_coeff - 1) | 1;
3103                 }else{
3104                         unquant_coeff = (((  alevel  << 1) + 1) * qscale * ((int) s->inter_matrix[j])) >> 4;
3105                         unquant_coeff =   (unquant_coeff - 1) | 1;
3106                 }
3107                 unquant_coeff<<= 3;
3108             }
3109
3110             distortion= (unquant_coeff - dct_coeff) * (unquant_coeff - dct_coeff) - zero_distortion;
3111             level+=64;
3112             if((level&(~127)) == 0){
3113                 for(j=survivor_count-1; j>=0; j--){
3114                     int run= i - survivor[j];
3115                     int score= distortion + length[UNI_AC_ENC_INDEX(run, level)]*lambda;
3116                     score += score_tab[i-run];
3117
3118                     if(score < best_score){
3119                         best_score= score;
3120                         run_tab[i+1]= run;
3121                         level_tab[i+1]= level-64;
3122                     }
3123                 }
3124
3125                 if(s->out_format == FMT_H263){
3126                     for(j=survivor_count-1; j>=0; j--){
3127                         int run= i - survivor[j];
3128                         int score= distortion + last_length[UNI_AC_ENC_INDEX(run, level)]*lambda;
3129                         score += score_tab[i-run];
3130                         if(score < last_score){
3131                             last_score= score;
3132                             last_run= run;
3133                             last_level= level-64;
3134                             last_i= i+1;
3135                         }
3136                     }
3137                 }
3138             }else{
3139                 distortion += esc_length*lambda;
3140                 for(j=survivor_count-1; j>=0; j--){
3141                     int run= i - survivor[j];
3142                     int score= distortion + score_tab[i-run];
3143
3144                     if(score < best_score){
3145                         best_score= score;
3146                         run_tab[i+1]= run;
3147                         level_tab[i+1]= level-64;
3148                     }
3149                 }
3150
3151                 if(s->out_format == FMT_H263){
3152                   for(j=survivor_count-1; j>=0; j--){
3153                         int run= i - survivor[j];
3154                         int score= distortion + score_tab[i-run];
3155                         if(score < last_score){
3156                             last_score= score;
3157                             last_run= run;
3158                             last_level= level-64;
3159                             last_i= i+1;
3160                         }
3161                     }
3162                 }
3163             }
3164         }
3165
3166         score_tab[i+1]= best_score;
3167
3168         //Note: there is a vlc code in mpeg4 which is 1 bit shorter then another one with a shorter run and the same level
3169         if(last_non_zero <= 27){
3170             for(; survivor_count; survivor_count--){
3171                 if(score_tab[ survivor[survivor_count-1] ] <= best_score)
3172                     break;
3173             }
3174         }else{
3175             for(; survivor_count; survivor_count--){
3176                 if(score_tab[ survivor[survivor_count-1] ] <= best_score + lambda)
3177                     break;
3178             }
3179         }
3180
3181         survivor[ survivor_count++ ]= i+1;
3182     }
3183
3184     if(s->out_format != FMT_H263){
3185         last_score= 256*256*256*120;
3186         for(i= survivor[0]; i<=last_non_zero + 1; i++){
3187             int score= score_tab[i];
3188             if(i) score += lambda*2; //FIXME exacter?
3189
3190             if(score < last_score){
3191                 last_score= score;
3192                 last_i= i;
3193                 last_level= level_tab[i];
3194                 last_run= run_tab[i];
3195             }
3196         }
3197     }
3198
3199     s->coded_score[n] = last_score;
3200
3201     dc= FFABS(block[0]);
3202     last_non_zero= last_i - 1;
3203     memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
3204
3205     if(last_non_zero < start_i)
3206         return last_non_zero;
3207
3208     if(last_non_zero == 0 && start_i == 0){
3209         int best_level= 0;
3210         int best_score= dc * dc;
3211
3212         for(i=0; i<coeff_count[0]; i++){
3213             int level= coeff[i][0];
3214             int alevel= FFABS(level);
3215             int unquant_coeff, score, distortion;
3216
3217             if(s->out_format == FMT_H263){
3218                     unquant_coeff= (alevel*qmul + qadd)>>3;
3219             }else{ //MPEG1
3220                     unquant_coeff = (((  alevel  << 1) + 1) * qscale * ((int) s->inter_matrix[0])) >> 4;
3221                     unquant_coeff =   (unquant_coeff - 1) | 1;
3222             }
3223             unquant_coeff = (unquant_coeff + 4) >> 3;
3224             unquant_coeff<<= 3 + 3;
3225
3226             distortion= (unquant_coeff - dc) * (unquant_coeff - dc);
3227             level+=64;
3228             if((level&(~127)) == 0) score= distortion + last_length[UNI_AC_ENC_INDEX(0, level)]*lambda;
3229             else                    score= distortion + esc_length*lambda;
3230
3231             if(score < best_score){
3232                 best_score= score;
3233                 best_level= level - 64;
3234             }
3235         }
3236         block[0]= best_level;
3237         s->coded_score[n] = best_score - dc*dc;
3238         if(best_level == 0) return -1;
3239         else                return last_non_zero;
3240     }
3241
3242     i= last_i;
3243     assert(last_level);
3244
3245     block[ perm_scantable[last_non_zero] ]= last_level;
3246     i -= last_run + 1;
3247
3248     for(; i>start_i; i -= run_tab[i] + 1){
3249         block[ perm_scantable[i-1] ]= level_tab[i];
3250     }
3251
3252     return last_non_zero;
3253 }
3254
3255 //#define REFINE_STATS 1
3256 static int16_t basis[64][64];
3257
3258 static void build_basis(uint8_t *perm){
3259     int i, j, x, y;
3260     emms_c();
3261     for(i=0; i<8; i++){
3262         for(j=0; j<8; j++){
3263             for(y=0; y<8; y++){
3264                 for(x=0; x<8; x++){
3265                     double s= 0.25*(1<<BASIS_SHIFT);
3266                     int index= 8*i + j;
3267                     int perm_index= perm[index];
3268                     if(i==0) s*= sqrt(0.5);
3269                     if(j==0) s*= sqrt(0.5);
3270                     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)));
3271                 }
3272             }
3273         }
3274     }
3275 }
3276
3277 static int dct_quantize_refine(MpegEncContext *s, //FIXME breaks denoise?
3278                         DCTELEM *block, int16_t *weight, DCTELEM *orig,
3279                         int n, int qscale){
3280     int16_t rem[64];
3281     DECLARE_ALIGNED_16(DCTELEM, d1[64]);
3282     const uint8_t *scantable= s->intra_scantable.scantable;
3283     const uint8_t *perm_scantable= s->intra_scantable.permutated;
3284 //    unsigned int threshold1, threshold2;
3285 //    int bias=0;
3286     int run_tab[65];
3287     int prev_run=0;
3288     int prev_level=0;
3289     int qmul, qadd, start_i, last_non_zero, i, dc;
3290     uint8_t * length;
3291     uint8_t * last_length;
3292     int lambda;
3293     int rle_index, run, q = 1, sum; //q is only used when s->mb_intra is true
3294 #ifdef REFINE_STATS
3295 static int count=0;
3296 static int after_last=0;
3297 static int to_zero=0;
3298 static int from_zero=0;
3299 static int raise=0;
3300 static int lower=0;
3301 static int messed_sign=0;
3302 #endif
3303
3304     if(basis[0][0] == 0)
3305         build_basis(s->dsp.idct_permutation);
3306
3307     qmul= qscale*2;
3308     qadd= (qscale-1)|1;
3309     if (s->mb_intra) {
3310         if (!s->h263_aic) {
3311             if (n < 4)
3312                 q = s->y_dc_scale;
3313             else
3314                 q = s->c_dc_scale;
3315         } else{
3316             /* For AIC we skip quant/dequant of INTRADC */
3317             q = 1;
3318             qadd=0;
3319         }
3320         q <<= RECON_SHIFT-3;
3321         /* note: block[0] is assumed to be positive */
3322         dc= block[0]*q;
3323 //        block[0] = (block[0] + (q >> 1)) / q;
3324         start_i = 1;
3325 //        if(s->mpeg_quant || s->out_format == FMT_MPEG1)
3326 //            bias= 1<<(QMAT_SHIFT-1);
3327         length     = s->intra_ac_vlc_length;
3328         last_length= s->intra_ac_vlc_last_length;
3329     } else {
3330         dc= 0;
3331         start_i = 0;
3332         length     = s->inter_ac_vlc_length;
3333         last_length= s->inter_ac_vlc_last_length;
3334     }
3335     last_non_zero = s->block_last_index[n];
3336
3337 #ifdef REFINE_STATS
3338 {START_TIMER
3339 #endif
3340     dc += (1<<(RECON_SHIFT-1));
3341     for(i=0; i<64; i++){
3342         rem[i]= dc - (orig[i]<<RECON_SHIFT); //FIXME  use orig dirrectly instead of copying to rem[]
3343     }
3344 #ifdef REFINE_STATS
3345 STOP_TIMER("memset rem[]")}
3346 #endif
3347     sum=0;
3348     for(i=0; i<64; i++){
3349         int one= 36;
3350         int qns=4;
3351         int w;
3352
3353         w= FFABS(weight[i]) + qns*one;
3354         w= 15 + (48*qns*one + w/2)/w; // 16 .. 63
3355
3356         weight[i] = w;
3357 //        w=weight[i] = (63*qns + (w/2)) / w;
3358
3359         assert(w>0);
3360         assert(w<(1<<6));
3361         sum += w*w;
3362     }
3363     lambda= sum*(uint64_t)s->lambda2 >> (FF_LAMBDA_SHIFT - 6 + 6 + 6 + 6);
3364 #ifdef REFINE_STATS
3365 {START_TIMER
3366 #endif
3367     run=0;
3368     rle_index=0;
3369     for(i=start_i; i<=last_non_zero; i++){
3370         int j= perm_scantable[i];
3371         const int level= block[j];
3372         int coeff;
3373
3374         if(level){
3375             if(level<0) coeff= qmul*level - qadd;
3376             else        coeff= qmul*level + qadd;
3377             run_tab[rle_index++]=run;
3378