Merge remote branch 'qatar/master'
[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
27  * The simplest mpeg encoder (well, it was the simplest!).
28  */
29
30 #include "libavutil/intmath.h"
31 #include "avcodec.h"
32 #include "dsputil.h"
33 #include "mpegvideo.h"
34 #include "mpegvideo_common.h"
35 #include "h263.h"
36 #include "mjpegenc.h"
37 #include "msmpeg4.h"
38 #include "faandct.h"
39 #include "thread.h"
40 #include "aandcttab.h"
41 #include "flv.h"
42 #include "mpeg4video.h"
43 #include "internal.h"
44 #include <limits.h>
45
46 //#undef NDEBUG
47 //#include <assert.h>
48
49 static int encode_picture(MpegEncContext *s, int picture_number);
50 static int dct_quantize_refine(MpegEncContext *s, DCTELEM *block, int16_t *weight, DCTELEM *orig, int n, int qscale);
51 static int sse_mb(MpegEncContext *s);
52 static void denoise_dct_c(MpegEncContext *s, DCTELEM *block);
53 static int dct_quantize_trellis_c(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow);
54
55 /* enable all paranoid tests for rounding, overflows, etc... */
56 //#define PARANOID
57
58 //#define DEBUG
59
60 static uint8_t default_mv_penalty[MAX_FCODE+1][MAX_MV*2+1];
61 static uint8_t default_fcode_tab[MAX_MV*2+1];
62
63 void ff_convert_matrix(DSPContext *dsp, int (*qmat)[64], uint16_t (*qmat16)[2][64],
64                            const uint16_t *quant_matrix, int bias, int qmin, int qmax, int intra)
65 {
66     int qscale;
67     int shift=0;
68
69     for(qscale=qmin; qscale<=qmax; qscale++){
70         int i;
71         if (dsp->fdct == ff_jpeg_fdct_islow
72 #ifdef FAAN_POSTSCALE
73             || dsp->fdct == ff_faandct
74 #endif
75             ) {
76             for(i=0;i<64;i++) {
77                 const int j= dsp->idct_permutation[i];
78                 /* 16 <= qscale * quant_matrix[i] <= 7905 */
79                 /* 19952             <= ff_aanscales[i] * qscale * quant_matrix[i]               <= 249205026 */
80                 /* (1 << 36) / 19952 >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= (1 << 36) / 249205026 */
81                 /* 3444240           >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= 275 */
82
83                 qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) /
84                                 (qscale * quant_matrix[j]));
85             }
86         } else if (dsp->fdct == fdct_ifast
87 #ifndef FAAN_POSTSCALE
88                    || dsp->fdct == ff_faandct
89 #endif
90                    ) {
91             for(i=0;i<64;i++) {
92                 const int j= dsp->idct_permutation[i];
93                 /* 16 <= qscale * quant_matrix[i] <= 7905 */
94                 /* 19952             <= ff_aanscales[i] * qscale * quant_matrix[i]               <= 249205026 */
95                 /* (1 << 36) / 19952 >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */
96                 /* 3444240           >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= 275 */
97
98                 qmat[qscale][i] = (int)((UINT64_C(1) << (QMAT_SHIFT + 14)) /
99                                 (ff_aanscales[i] * qscale * quant_matrix[j]));
100             }
101         } else {
102             for(i=0;i<64;i++) {
103                 const int j= dsp->idct_permutation[i];
104                 /* We can safely suppose that 16 <= quant_matrix[i] <= 255
105                    So 16           <= qscale * quant_matrix[i]             <= 7905
106                    so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905
107                    so 32768        >= (1<<19) / (qscale * quant_matrix[i]) >= 67
108                 */
109                 qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) / (qscale * quant_matrix[j]));
110 //                qmat  [qscale][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]);
111                 qmat16[qscale][0][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[j]);
112
113                 if(qmat16[qscale][0][i]==0 || qmat16[qscale][0][i]==128*256) qmat16[qscale][0][i]=128*256-1;
114                 qmat16[qscale][1][i]= ROUNDED_DIV(bias<<(16-QUANT_BIAS_SHIFT), qmat16[qscale][0][i]);
115             }
116         }
117
118         for(i=intra; i<64; i++){
119             int64_t max= 8191;
120             if (dsp->fdct == fdct_ifast
121 #ifndef FAAN_POSTSCALE
122                    || dsp->fdct == ff_faandct
123 #endif
124                    ) {
125                 max = (8191LL*ff_aanscales[i]) >> 14;
126             }
127             while(((max * qmat[qscale][i]) >> shift) > INT_MAX){
128                 shift++;
129             }
130         }
131     }
132     if(shift){
133         av_log(NULL, AV_LOG_INFO, "Warning, QMAT_SHIFT is larger than %d, overflows possible\n", QMAT_SHIFT - shift);
134     }
135 }
136
137 static inline void update_qscale(MpegEncContext *s){
138     s->qscale= (s->lambda*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7);
139     s->qscale= av_clip(s->qscale, s->avctx->qmin, s->avctx->qmax);
140
141     s->lambda2= (s->lambda*s->lambda + FF_LAMBDA_SCALE/2) >> FF_LAMBDA_SHIFT;
142 }
143
144 void ff_write_quant_matrix(PutBitContext *pb, uint16_t *matrix){
145     int i;
146
147     if(matrix){
148         put_bits(pb, 1, 1);
149         for(i=0;i<64;i++) {
150             put_bits(pb, 8, matrix[ ff_zigzag_direct[i] ]);
151         }
152     }else
153         put_bits(pb, 1, 0);
154 }
155
156 /**
157  * init s->current_picture.qscale_table from s->lambda_table
158  */
159 void ff_init_qscale_tab(MpegEncContext *s){
160     int8_t * const qscale_table= s->current_picture.qscale_table;
161     int i;
162
163     for(i=0; i<s->mb_num; i++){
164         unsigned int lam= s->lambda_table[ s->mb_index2xy[i] ];
165         int qp= (lam*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7);
166         qscale_table[ s->mb_index2xy[i] ]= av_clip(qp, s->avctx->qmin, s->avctx->qmax);
167     }
168 }
169
170 static void copy_picture_attributes(MpegEncContext *s, AVFrame *dst, AVFrame *src){
171     int i;
172
173     dst->pict_type              = src->pict_type;
174     dst->quality                = src->quality;
175     dst->coded_picture_number   = src->coded_picture_number;
176     dst->display_picture_number = src->display_picture_number;
177 //    dst->reference              = src->reference;
178     dst->pts                    = src->pts;
179     dst->interlaced_frame       = src->interlaced_frame;
180     dst->top_field_first        = src->top_field_first;
181
182     if(s->avctx->me_threshold){
183         if(!src->motion_val[0])
184             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_val not set!\n");
185         if(!src->mb_type)
186             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.mb_type not set!\n");
187         if(!src->ref_index[0])
188             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.ref_index not set!\n");
189         if(src->motion_subsample_log2 != dst->motion_subsample_log2)
190             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_subsample_log2 doesn't match! (%d!=%d)\n",
191             src->motion_subsample_log2, dst->motion_subsample_log2);
192
193         memcpy(dst->mb_type, src->mb_type, s->mb_stride * s->mb_height * sizeof(dst->mb_type[0]));
194
195         for(i=0; i<2; i++){
196             int stride= ((16*s->mb_width )>>src->motion_subsample_log2) + 1;
197             int height= ((16*s->mb_height)>>src->motion_subsample_log2);
198
199             if(src->motion_val[i] && src->motion_val[i] != dst->motion_val[i]){
200                 memcpy(dst->motion_val[i], src->motion_val[i], 2*stride*height*sizeof(int16_t));
201             }
202             if(src->ref_index[i] && src->ref_index[i] != dst->ref_index[i]){
203                 memcpy(dst->ref_index[i], src->ref_index[i], s->mb_stride*4*s->mb_height*sizeof(int8_t));
204             }
205         }
206     }
207 }
208
209 static void update_duplicate_context_after_me(MpegEncContext *dst, MpegEncContext *src){
210 #define COPY(a) dst->a= src->a
211     COPY(pict_type);
212     COPY(current_picture);
213     COPY(f_code);
214     COPY(b_code);
215     COPY(qscale);
216     COPY(lambda);
217     COPY(lambda2);
218     COPY(picture_in_gop_number);
219     COPY(gop_picture_number);
220     COPY(frame_pred_frame_dct); //FIXME don't set in encode_header
221     COPY(progressive_frame); //FIXME don't set in encode_header
222     COPY(partitioned_frame); //FIXME don't set in encode_header
223 #undef COPY
224 }
225
226 /**
227  * sets the given MpegEncContext to defaults for encoding.
228  * the changed fields will not depend upon the prior state of the MpegEncContext.
229  */
230 static void MPV_encode_defaults(MpegEncContext *s){
231     int i;
232     MPV_common_defaults(s);
233
234     for(i=-16; i<16; i++){
235         default_fcode_tab[i + MAX_MV]= 1;
236     }
237     s->me.mv_penalty= default_mv_penalty;
238     s->fcode_tab= default_fcode_tab;
239 }
240
241 /* init video encoder */
242 av_cold int MPV_encode_init(AVCodecContext *avctx)
243 {
244     MpegEncContext *s = avctx->priv_data;
245     int i;
246     int chroma_h_shift, chroma_v_shift;
247
248     MPV_encode_defaults(s);
249
250     switch (avctx->codec_id) {
251     case CODEC_ID_MPEG2VIDEO:
252         if(avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P){
253             av_log(avctx, AV_LOG_ERROR, "only YUV420 and YUV422 are supported\n");
254             return -1;
255         }
256         break;
257     case CODEC_ID_LJPEG:
258         if(avctx->pix_fmt != PIX_FMT_YUVJ420P && avctx->pix_fmt != PIX_FMT_YUVJ422P && avctx->pix_fmt != PIX_FMT_YUVJ444P && avctx->pix_fmt != PIX_FMT_BGRA &&
259            ((avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P && avctx->pix_fmt != PIX_FMT_YUV444P) || avctx->strict_std_compliance>FF_COMPLIANCE_UNOFFICIAL)){
260             av_log(avctx, AV_LOG_ERROR, "colorspace not supported in LJPEG\n");
261             return -1;
262         }
263         break;
264     case CODEC_ID_MJPEG:
265         if(avctx->pix_fmt != PIX_FMT_YUVJ420P && avctx->pix_fmt != PIX_FMT_YUVJ422P &&
266            ((avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P) || avctx->strict_std_compliance>FF_COMPLIANCE_UNOFFICIAL)){
267             av_log(avctx, AV_LOG_ERROR, "colorspace not supported in jpeg\n");
268             return -1;
269         }
270         break;
271     default:
272         if(avctx->pix_fmt != PIX_FMT_YUV420P){
273             av_log(avctx, AV_LOG_ERROR, "only YUV420 is supported\n");
274             return -1;
275         }
276     }
277
278     switch (avctx->pix_fmt) {
279     case PIX_FMT_YUVJ422P:
280     case PIX_FMT_YUV422P:
281         s->chroma_format = CHROMA_422;
282         break;
283     case PIX_FMT_YUVJ420P:
284     case PIX_FMT_YUV420P:
285     default:
286         s->chroma_format = CHROMA_420;
287         break;
288     }
289
290     s->bit_rate = avctx->bit_rate;
291     s->width = avctx->width;
292     s->height = avctx->height;
293     if(avctx->gop_size > 600 && avctx->strict_std_compliance>FF_COMPLIANCE_EXPERIMENTAL){
294         av_log(avctx, AV_LOG_ERROR, "Warning keyframe interval too large! reducing it ...\n");
295         avctx->gop_size=600;
296     }
297     s->gop_size = avctx->gop_size;
298     s->avctx = avctx;
299     s->flags= avctx->flags;
300     s->flags2= avctx->flags2;
301     s->max_b_frames= avctx->max_b_frames;
302     s->codec_id= avctx->codec->id;
303     s->luma_elim_threshold  = avctx->luma_elim_threshold;
304     s->chroma_elim_threshold= avctx->chroma_elim_threshold;
305     s->strict_std_compliance= avctx->strict_std_compliance;
306     s->data_partitioning= avctx->flags & CODEC_FLAG_PART;
307     s->quarter_sample= (avctx->flags & CODEC_FLAG_QPEL)!=0;
308     s->mpeg_quant= avctx->mpeg_quant;
309     s->rtp_mode= !!avctx->rtp_payload_size;
310     s->intra_dc_precision= avctx->intra_dc_precision;
311     s->user_specified_pts = AV_NOPTS_VALUE;
312
313     if (s->gop_size <= 1) {
314         s->intra_only = 1;
315         s->gop_size = 12;
316     } else {
317         s->intra_only = 0;
318     }
319
320     s->me_method = avctx->me_method;
321
322     /* Fixed QSCALE */
323     s->fixed_qscale = !!(avctx->flags & CODEC_FLAG_QSCALE);
324
325     s->adaptive_quant= (   s->avctx->lumi_masking
326                         || s->avctx->dark_masking
327                         || s->avctx->temporal_cplx_masking
328                         || s->avctx->spatial_cplx_masking
329                         || s->avctx->p_masking
330                         || s->avctx->border_masking
331                         || (s->flags&CODEC_FLAG_QP_RD))
332                        && !s->fixed_qscale;
333
334     s->obmc= !!(s->flags & CODEC_FLAG_OBMC);
335     s->loop_filter= !!(s->flags & CODEC_FLAG_LOOP_FILTER);
336     s->alternate_scan= !!(s->flags & CODEC_FLAG_ALT_SCAN);
337     s->intra_vlc_format= !!(s->flags2 & CODEC_FLAG2_INTRA_VLC);
338     s->q_scale_type= !!(s->flags2 & CODEC_FLAG2_NON_LINEAR_QUANT);
339
340     if(avctx->rc_max_rate && !avctx->rc_buffer_size){
341         av_log(avctx, AV_LOG_ERROR, "a vbv buffer size is needed, for encoding with a maximum bitrate\n");
342         return -1;
343     }
344
345     if(avctx->rc_min_rate && avctx->rc_max_rate != avctx->rc_min_rate){
346         av_log(avctx, AV_LOG_INFO, "Warning min_rate > 0 but min_rate != max_rate isn't recommended!\n");
347     }
348
349     if(avctx->rc_min_rate && avctx->rc_min_rate > avctx->bit_rate){
350         av_log(avctx, AV_LOG_ERROR, "bitrate below min bitrate\n");
351         return -1;
352     }
353
354     if(avctx->rc_max_rate && avctx->rc_max_rate < avctx->bit_rate){
355         av_log(avctx, AV_LOG_INFO, "bitrate above max bitrate\n");
356         return -1;
357     }
358
359     if(avctx->rc_max_rate && avctx->rc_max_rate == avctx->bit_rate && avctx->rc_max_rate != avctx->rc_min_rate){
360         av_log(avctx, AV_LOG_INFO, "impossible bitrate constraints, this will fail\n");
361     }
362
363     if(avctx->rc_buffer_size && avctx->bit_rate*(int64_t)avctx->time_base.num > avctx->rc_buffer_size * (int64_t)avctx->time_base.den){
364         av_log(avctx, AV_LOG_ERROR, "VBV buffer too small for bitrate\n");
365         return -1;
366     }
367
368     if(!s->fixed_qscale && avctx->bit_rate*av_q2d(avctx->time_base) > avctx->bit_rate_tolerance){
369         av_log(avctx, AV_LOG_ERROR, "bitrate tolerance too small for bitrate\n");
370         return -1;
371     }
372
373     if(   s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate
374        && (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO)
375        && 90000LL * (avctx->rc_buffer_size-1) > s->avctx->rc_max_rate*0xFFFFLL){
376
377         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");
378     }
379
380     if((s->flags & CODEC_FLAG_4MV) && s->codec_id != CODEC_ID_MPEG4
381        && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P && s->codec_id != CODEC_ID_FLV1){
382         av_log(avctx, AV_LOG_ERROR, "4MV not supported by codec\n");
383         return -1;
384     }
385
386     if(s->obmc && s->avctx->mb_decision != FF_MB_DECISION_SIMPLE){
387         av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with simple mb decision\n");
388         return -1;
389     }
390
391     if(s->obmc && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P){
392         av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with H263(+)\n");
393         return -1;
394     }
395
396     if(s->quarter_sample && s->codec_id != CODEC_ID_MPEG4){
397         av_log(avctx, AV_LOG_ERROR, "qpel not supported by codec\n");
398         return -1;
399     }
400
401     if(s->data_partitioning && s->codec_id != CODEC_ID_MPEG4){
402         av_log(avctx, AV_LOG_ERROR, "data partitioning not supported by codec\n");
403         return -1;
404     }
405
406     if(s->max_b_frames && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO){
407         av_log(avctx, AV_LOG_ERROR, "b frames not supported by codec\n");
408         return -1;
409     }
410
411     if ((s->codec_id == CODEC_ID_MPEG4 || s->codec_id == CODEC_ID_H263 ||
412          s->codec_id == CODEC_ID_H263P) &&
413         (avctx->sample_aspect_ratio.num > 255 || avctx->sample_aspect_ratio.den > 255)) {
414         av_log(avctx, AV_LOG_ERROR, "Invalid pixel aspect ratio %i/%i, limit is 255/255\n",
415                avctx->sample_aspect_ratio.num, avctx->sample_aspect_ratio.den);
416         return -1;
417     }
418
419     if((s->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN))
420        && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG2VIDEO){
421         av_log(avctx, AV_LOG_ERROR, "interlacing not supported by codec\n");
422         return -1;
423     }
424
425     if(s->mpeg_quant && s->codec_id != CODEC_ID_MPEG4){ //FIXME mpeg2 uses that too
426         av_log(avctx, AV_LOG_ERROR, "mpeg2 style quantization not supported by codec\n");
427         return -1;
428     }
429
430     if((s->flags & CODEC_FLAG_CBP_RD) && !avctx->trellis){
431         av_log(avctx, AV_LOG_ERROR, "CBP RD needs trellis quant\n");
432         return -1;
433     }
434
435     if((s->flags & CODEC_FLAG_QP_RD) && s->avctx->mb_decision != FF_MB_DECISION_RD){
436         av_log(avctx, AV_LOG_ERROR, "QP RD needs mbd=2\n");
437         return -1;
438     }
439
440     if(s->avctx->scenechange_threshold < 1000000000 && (s->flags & CODEC_FLAG_CLOSED_GOP)){
441         av_log(avctx, AV_LOG_ERROR, "closed gop with scene change detection are not supported yet, set threshold to 1000000000\n");
442         return -1;
443     }
444
445     if((s->flags2 & CODEC_FLAG2_INTRA_VLC) && s->codec_id != CODEC_ID_MPEG2VIDEO){
446         av_log(avctx, AV_LOG_ERROR, "intra vlc table not supported by codec\n");
447         return -1;
448     }
449
450     if(s->flags & CODEC_FLAG_LOW_DELAY){
451         if (s->codec_id != CODEC_ID_MPEG2VIDEO){
452             av_log(avctx, AV_LOG_ERROR, "low delay forcing is only available for mpeg2\n");
453             return -1;
454         }
455         if (s->max_b_frames != 0){
456             av_log(avctx, AV_LOG_ERROR, "b frames cannot be used with low delay\n");
457             return -1;
458         }
459     }
460
461     if(s->q_scale_type == 1){
462         if(s->codec_id != CODEC_ID_MPEG2VIDEO){
463             av_log(avctx, AV_LOG_ERROR, "non linear quant is only available for mpeg2\n");
464             return -1;
465         }
466         if(avctx->qmax > 12){
467             av_log(avctx, AV_LOG_ERROR, "non linear quant only supports qmax <= 12 currently\n");
468             return -1;
469         }
470     }
471
472     if(s->avctx->thread_count > 1 && s->codec_id != CODEC_ID_MPEG4
473        && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO
474        && (s->codec_id != CODEC_ID_H263P || !(s->flags & CODEC_FLAG_H263P_SLICE_STRUCT))){
475         av_log(avctx, AV_LOG_ERROR, "multi threaded encoding not supported by codec\n");
476         return -1;
477     }
478
479     if(s->avctx->thread_count < 1){
480         av_log(avctx, AV_LOG_ERROR, "automatic thread number detection not supported by codec, patch welcome\n");
481         return -1;
482     }
483
484     if(s->avctx->thread_count > 1)
485         s->rtp_mode= 1;
486
487     if(!avctx->time_base.den || !avctx->time_base.num){
488         av_log(avctx, AV_LOG_ERROR, "framerate not set\n");
489         return -1;
490     }
491
492     i= (INT_MAX/2+128)>>8;
493     if(avctx->me_threshold >= i){
494         av_log(avctx, AV_LOG_ERROR, "me_threshold too large, max is %d\n", i - 1);
495         return -1;
496     }
497     if(avctx->mb_threshold >= i){
498         av_log(avctx, AV_LOG_ERROR, "mb_threshold too large, max is %d\n", i - 1);
499         return -1;
500     }
501
502     if(avctx->b_frame_strategy && (avctx->flags&CODEC_FLAG_PASS2)){
503         av_log(avctx, AV_LOG_INFO, "notice: b_frame_strategy only affects the first pass\n");
504         avctx->b_frame_strategy = 0;
505     }
506
507     i= av_gcd(avctx->time_base.den, avctx->time_base.num);
508     if(i > 1){
509         av_log(avctx, AV_LOG_INFO, "removing common factors from framerate\n");
510         avctx->time_base.den /= i;
511         avctx->time_base.num /= i;
512 //        return -1;
513     }
514
515     if(s->mpeg_quant || s->codec_id==CODEC_ID_MPEG1VIDEO || s->codec_id==CODEC_ID_MPEG2VIDEO || s->codec_id==CODEC_ID_MJPEG){
516         s->intra_quant_bias= 3<<(QUANT_BIAS_SHIFT-3); //(a + x*3/8)/x
517         s->inter_quant_bias= 0;
518     }else{
519         s->intra_quant_bias=0;
520         s->inter_quant_bias=-(1<<(QUANT_BIAS_SHIFT-2)); //(a - x/4)/x
521     }
522
523     if(avctx->intra_quant_bias != FF_DEFAULT_QUANT_BIAS)
524         s->intra_quant_bias= avctx->intra_quant_bias;
525     if(avctx->inter_quant_bias != FF_DEFAULT_QUANT_BIAS)
526         s->inter_quant_bias= avctx->inter_quant_bias;
527
528     avcodec_get_chroma_sub_sample(avctx->pix_fmt, &chroma_h_shift, &chroma_v_shift);
529
530     if(avctx->codec_id == CODEC_ID_MPEG4 && s->avctx->time_base.den > (1<<16)-1){
531         av_log(avctx, AV_LOG_ERROR, "timebase %d/%d not supported by MPEG 4 standard, "
532                "the maximum admitted value for the timebase denominator is %d\n",
533                s->avctx->time_base.num, s->avctx->time_base.den, (1<<16)-1);
534         return -1;
535     }
536     s->time_increment_bits = av_log2(s->avctx->time_base.den - 1) + 1;
537
538     switch(avctx->codec->id) {
539     case CODEC_ID_MPEG1VIDEO:
540         s->out_format = FMT_MPEG1;
541         s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY);
542         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
543         break;
544     case CODEC_ID_MPEG2VIDEO:
545         s->out_format = FMT_MPEG1;
546         s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY);
547         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
548         s->rtp_mode= 1;
549         break;
550     case CODEC_ID_LJPEG:
551     case CODEC_ID_MJPEG:
552         s->out_format = FMT_MJPEG;
553         s->intra_only = 1; /* force intra only for jpeg */
554         if(avctx->codec->id == CODEC_ID_LJPEG && avctx->pix_fmt == PIX_FMT_BGRA){
555             s->mjpeg_vsample[0] = s->mjpeg_hsample[0] =
556             s->mjpeg_vsample[1] = s->mjpeg_hsample[1] =
557             s->mjpeg_vsample[2] = s->mjpeg_hsample[2] = 1;
558         }else{
559             s->mjpeg_vsample[0] = 2;
560             s->mjpeg_vsample[1] = 2>>chroma_v_shift;
561             s->mjpeg_vsample[2] = 2>>chroma_v_shift;
562             s->mjpeg_hsample[0] = 2;
563             s->mjpeg_hsample[1] = 2>>chroma_h_shift;
564             s->mjpeg_hsample[2] = 2>>chroma_h_shift;
565         }
566         if (!(CONFIG_MJPEG_ENCODER || CONFIG_LJPEG_ENCODER)
567             || ff_mjpeg_encode_init(s) < 0)
568             return -1;
569         avctx->delay=0;
570         s->low_delay=1;
571         break;
572     case CODEC_ID_H261:
573         if (!CONFIG_H261_ENCODER)  return -1;
574         if (ff_h261_get_picture_format(s->width, s->height) < 0) {
575             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);
576             return -1;
577         }
578         s->out_format = FMT_H261;
579         avctx->delay=0;
580         s->low_delay=1;
581         break;
582     case CODEC_ID_H263:
583         if (!CONFIG_H263_ENCODER)  return -1;
584         if (ff_match_2uint16(h263_format, FF_ARRAY_ELEMS(h263_format), s->width, s->height) == 8) {
585             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);
586             return -1;
587         }
588         s->out_format = FMT_H263;
589         s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
590         avctx->delay=0;
591         s->low_delay=1;
592         break;
593     case CODEC_ID_H263P:
594         s->out_format = FMT_H263;
595         s->h263_plus = 1;
596         /* Fx */
597         s->umvplus = (avctx->flags & CODEC_FLAG_H263P_UMV) ? 1:0;
598         s->h263_aic= (avctx->flags & CODEC_FLAG_AC_PRED) ? 1:0;
599         s->modified_quant= s->h263_aic;
600         s->alt_inter_vlc= (avctx->flags & CODEC_FLAG_H263P_AIV) ? 1:0;
601         s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
602         s->loop_filter= (avctx->flags & CODEC_FLAG_LOOP_FILTER) ? 1:0;
603         s->unrestricted_mv= s->obmc || s->loop_filter || s->umvplus;
604         s->h263_slice_structured= (s->flags & CODEC_FLAG_H263P_SLICE_STRUCT) ? 1:0;
605
606         /* /Fx */
607         /* These are just to be sure */
608         avctx->delay=0;
609         s->low_delay=1;
610         break;
611     case CODEC_ID_FLV1:
612         s->out_format = FMT_H263;
613         s->h263_flv = 2; /* format = 1; 11-bit codes */
614         s->unrestricted_mv = 1;
615         s->rtp_mode=0; /* don't allow GOB */
616         avctx->delay=0;
617         s->low_delay=1;
618         break;
619     case CODEC_ID_RV10:
620         s->out_format = FMT_H263;
621         avctx->delay=0;
622         s->low_delay=1;
623         break;
624     case CODEC_ID_RV20:
625         s->out_format = FMT_H263;
626         avctx->delay=0;
627         s->low_delay=1;
628         s->modified_quant=1;
629         s->h263_aic=1;
630         s->h263_plus=1;
631         s->loop_filter=1;
632         s->unrestricted_mv= 0;
633         break;
634     case CODEC_ID_MPEG4:
635         s->out_format = FMT_H263;
636         s->h263_pred = 1;
637         s->unrestricted_mv = 1;
638         s->low_delay= s->max_b_frames ? 0 : 1;
639         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
640         break;
641     case CODEC_ID_MSMPEG4V1:
642         s->out_format = FMT_H263;
643         s->h263_msmpeg4 = 1;
644         s->h263_pred = 1;
645         s->unrestricted_mv = 1;
646         s->msmpeg4_version= 1;
647         avctx->delay=0;
648         s->low_delay=1;
649         break;
650     case CODEC_ID_MSMPEG4V2:
651         s->out_format = FMT_H263;
652         s->h263_msmpeg4 = 1;
653         s->h263_pred = 1;
654         s->unrestricted_mv = 1;
655         s->msmpeg4_version= 2;
656         avctx->delay=0;
657         s->low_delay=1;
658         break;
659     case CODEC_ID_MSMPEG4V3:
660         s->out_format = FMT_H263;
661         s->h263_msmpeg4 = 1;
662         s->h263_pred = 1;
663         s->unrestricted_mv = 1;
664         s->msmpeg4_version= 3;
665         s->flipflop_rounding=1;
666         avctx->delay=0;
667         s->low_delay=1;
668         break;
669     case CODEC_ID_WMV1:
670         s->out_format = FMT_H263;
671         s->h263_msmpeg4 = 1;
672         s->h263_pred = 1;
673         s->unrestricted_mv = 1;
674         s->msmpeg4_version= 4;
675         s->flipflop_rounding=1;
676         avctx->delay=0;
677         s->low_delay=1;
678         break;
679     case CODEC_ID_WMV2:
680         s->out_format = FMT_H263;
681         s->h263_msmpeg4 = 1;
682         s->h263_pred = 1;
683         s->unrestricted_mv = 1;
684         s->msmpeg4_version= 5;
685         s->flipflop_rounding=1;
686         avctx->delay=0;
687         s->low_delay=1;
688         break;
689     default:
690         return -1;
691     }
692
693     avctx->has_b_frames= !s->low_delay;
694
695     s->encoding = 1;
696
697     s->progressive_frame=
698     s->progressive_sequence= !(avctx->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN));
699
700     /* init */
701     if (MPV_common_init(s) < 0)
702         return -1;
703
704     if(!s->dct_quantize)
705         s->dct_quantize = dct_quantize_c;
706     if(!s->denoise_dct)
707         s->denoise_dct = denoise_dct_c;
708     s->fast_dct_quantize = s->dct_quantize;
709     if(avctx->trellis)
710         s->dct_quantize = dct_quantize_trellis_c;
711
712     if((CONFIG_H263P_ENCODER || CONFIG_RV20_ENCODER) && s->modified_quant)
713         s->chroma_qscale_table= ff_h263_chroma_qscale_table;
714
715     s->quant_precision=5;
716
717     ff_set_cmp(&s->dsp, s->dsp.ildct_cmp, s->avctx->ildct_cmp);
718     ff_set_cmp(&s->dsp, s->dsp.frame_skip_cmp, s->avctx->frame_skip_cmp);
719
720     if (CONFIG_H261_ENCODER && s->out_format == FMT_H261)
721         ff_h261_encode_init(s);
722     if (CONFIG_H263_ENCODER && s->out_format == FMT_H263)
723         h263_encode_init(s);
724     if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version)
725         ff_msmpeg4_encode_init(s);
726     if ((CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
727         && s->out_format == FMT_MPEG1)
728         ff_mpeg1_encode_init(s);
729
730     /* init q matrix */
731     for(i=0;i<64;i++) {
732         int j= s->dsp.idct_permutation[i];
733         if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4 && s->mpeg_quant){
734             s->intra_matrix[j] = ff_mpeg4_default_intra_matrix[i];
735             s->inter_matrix[j] = ff_mpeg4_default_non_intra_matrix[i];
736         }else if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
737             s->intra_matrix[j] =
738             s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
739         }else
740         { /* mpeg1/2 */
741             s->intra_matrix[j] = ff_mpeg1_default_intra_matrix[i];
742             s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
743         }
744         if(s->avctx->intra_matrix)
745             s->intra_matrix[j] = s->avctx->intra_matrix[i];
746         if(s->avctx->inter_matrix)
747             s->inter_matrix[j] = s->avctx->inter_matrix[i];
748     }
749
750     /* precompute matrix */
751     /* for mjpeg, we do include qscale in the matrix */
752     if (s->out_format != FMT_MJPEG) {
753         ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
754                        s->intra_matrix, s->intra_quant_bias, avctx->qmin, 31, 1);
755         ff_convert_matrix(&s->dsp, s->q_inter_matrix, s->q_inter_matrix16,
756                        s->inter_matrix, s->inter_quant_bias, avctx->qmin, 31, 0);
757     }
758
759     if(ff_rate_control_init(s) < 0)
760         return -1;
761
762     return 0;
763 }
764
765 av_cold int MPV_encode_end(AVCodecContext *avctx)
766 {
767     MpegEncContext *s = avctx->priv_data;
768
769     ff_rate_control_uninit(s);
770
771     MPV_common_end(s);
772     if ((CONFIG_MJPEG_ENCODER || CONFIG_LJPEG_ENCODER) && s->out_format == FMT_MJPEG)
773         ff_mjpeg_encode_close(s);
774
775     av_freep(&avctx->extradata);
776
777     return 0;
778 }
779
780 static int get_sae(uint8_t *src, int ref, int stride){
781     int x,y;
782     int acc=0;
783
784     for(y=0; y<16; y++){
785         for(x=0; x<16; x++){
786             acc+= FFABS(src[x+y*stride] - ref);
787         }
788     }
789
790     return acc;
791 }
792
793 static int get_intra_count(MpegEncContext *s, uint8_t *src, uint8_t *ref, int stride){
794     int x, y, w, h;
795     int acc=0;
796
797     w= s->width &~15;
798     h= s->height&~15;
799
800     for(y=0; y<h; y+=16){
801         for(x=0; x<w; x+=16){
802             int offset= x + y*stride;
803             int sad = s->dsp.sad[0](NULL, src + offset, ref + offset, stride, 16);
804             int mean= (s->dsp.pix_sum(src + offset, stride) + 128)>>8;
805             int sae = get_sae(src + offset, mean, stride);
806
807             acc+= sae + 500 < sad;
808         }
809     }
810     return acc;
811 }
812
813
814 static int load_input_picture(MpegEncContext *s, AVFrame *pic_arg){
815     AVFrame *pic=NULL;
816     int64_t pts;
817     int i;
818     const int encoding_delay= s->max_b_frames;
819     int direct=1;
820
821     if(pic_arg){
822         pts= pic_arg->pts;
823         pic_arg->display_picture_number= s->input_picture_number++;
824
825         if(pts != AV_NOPTS_VALUE){
826             if(s->user_specified_pts != AV_NOPTS_VALUE){
827                 int64_t time= pts;
828                 int64_t last= s->user_specified_pts;
829
830                 if(time <= last){
831                     av_log(s->avctx, AV_LOG_ERROR, "Error, Invalid timestamp=%"PRId64", last=%"PRId64"\n", pts, s->user_specified_pts);
832                     return -1;
833                 }
834             }
835             s->user_specified_pts= pts;
836         }else{
837             if(s->user_specified_pts != AV_NOPTS_VALUE){
838                 s->user_specified_pts=
839                 pts= s->user_specified_pts + 1;
840                 av_log(s->avctx, AV_LOG_INFO, "Warning: AVFrame.pts=? trying to guess (%"PRId64")\n", pts);
841             }else{
842                 pts= pic_arg->display_picture_number;
843             }
844         }
845     }
846
847   if(pic_arg){
848     if(encoding_delay && !(s->flags&CODEC_FLAG_INPUT_PRESERVED)) direct=0;
849     if(pic_arg->linesize[0] != s->linesize) direct=0;
850     if(pic_arg->linesize[1] != s->uvlinesize) direct=0;
851     if(pic_arg->linesize[2] != s->uvlinesize) direct=0;
852
853 //    av_log(AV_LOG_DEBUG, "%d %d %d %d\n",pic_arg->linesize[0], pic_arg->linesize[1], s->linesize, s->uvlinesize);
854
855     if(direct){
856         i= ff_find_unused_picture(s, 1);
857
858         pic= (AVFrame*)&s->picture[i];
859         pic->reference= 3;
860
861         for(i=0; i<4; i++){
862             pic->data[i]= pic_arg->data[i];
863             pic->linesize[i]= pic_arg->linesize[i];
864         }
865         if(ff_alloc_picture(s, (Picture*)pic, 1) < 0){
866             return -1;
867         }
868     }else{
869         i= ff_find_unused_picture(s, 0);
870
871         pic= (AVFrame*)&s->picture[i];
872         pic->reference= 3;
873
874         if(ff_alloc_picture(s, (Picture*)pic, 0) < 0){
875             return -1;
876         }
877
878         if(   pic->data[0] + INPLACE_OFFSET == pic_arg->data[0]
879            && pic->data[1] + INPLACE_OFFSET == pic_arg->data[1]
880            && pic->data[2] + INPLACE_OFFSET == pic_arg->data[2]){
881        // empty
882         }else{
883             int h_chroma_shift, v_chroma_shift;
884             avcodec_get_chroma_sub_sample(s->avctx->pix_fmt, &h_chroma_shift, &v_chroma_shift);
885
886             for(i=0; i<3; i++){
887                 int src_stride= pic_arg->linesize[i];
888                 int dst_stride= i ? s->uvlinesize : s->linesize;
889                 int h_shift= i ? h_chroma_shift : 0;
890                 int v_shift= i ? v_chroma_shift : 0;
891                 int w= s->width >>h_shift;
892                 int h= s->height>>v_shift;
893                 uint8_t *src= pic_arg->data[i];
894                 uint8_t *dst= pic->data[i];
895
896                 if(!s->avctx->rc_buffer_size)
897                     dst +=INPLACE_OFFSET;
898
899                 if(src_stride==dst_stride)
900                     memcpy(dst, src, src_stride*h);
901                 else{
902                     while(h--){
903                         memcpy(dst, src, w);
904                         dst += dst_stride;
905                         src += src_stride;
906                     }
907                 }
908             }
909         }
910     }
911     copy_picture_attributes(s, pic, pic_arg);
912     pic->pts= pts; //we set this here to avoid modifiying pic_arg
913   }
914
915     /* shift buffer entries */
916     for(i=1; i<MAX_PICTURE_COUNT /*s->encoding_delay+1*/; i++)
917         s->input_picture[i-1]= s->input_picture[i];
918
919     s->input_picture[encoding_delay]= (Picture*)pic;
920
921     return 0;
922 }
923
924 static int skip_check(MpegEncContext *s, Picture *p, Picture *ref){
925     int x, y, plane;
926     int score=0;
927     int64_t score64=0;
928
929     for(plane=0; plane<3; plane++){
930         const int stride= p->linesize[plane];
931         const int bw= plane ? 1 : 2;
932         for(y=0; y<s->mb_height*bw; y++){
933             for(x=0; x<s->mb_width*bw; x++){
934                 int off= p->type == FF_BUFFER_TYPE_SHARED ? 0: 16;
935                 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);
936
937                 switch(s->avctx->frame_skip_exp){
938                     case 0: score= FFMAX(score, v); break;
939                     case 1: score+= FFABS(v);break;
940                     case 2: score+= v*v;break;
941                     case 3: score64+= FFABS(v*v*(int64_t)v);break;
942                     case 4: score64+= v*v*(int64_t)(v*v);break;
943                 }
944             }
945         }
946     }
947
948     if(score) score64= score;
949
950     if(score64 < s->avctx->frame_skip_threshold)
951         return 1;
952     if(score64 < ((s->avctx->frame_skip_factor * (int64_t)s->lambda)>>8))
953         return 1;
954     return 0;
955 }
956
957 static int estimate_best_b_count(MpegEncContext *s){
958     AVCodec *codec= avcodec_find_encoder(s->avctx->codec_id);
959     AVCodecContext *c= avcodec_alloc_context();
960     AVFrame input[FF_MAX_B_FRAMES+2];
961     const int scale= s->avctx->brd_scale;
962     int i, j, out_size, p_lambda, b_lambda, lambda2;
963     int outbuf_size= s->width * s->height; //FIXME
964     uint8_t *outbuf= av_malloc(outbuf_size);
965     int64_t best_rd= INT64_MAX;
966     int best_b_count= -1;
967
968     assert(scale>=0 && scale <=3);
969
970 //    emms_c();
971     p_lambda= s->last_lambda_for[FF_P_TYPE]; //s->next_picture_ptr->quality;
972     b_lambda= s->last_lambda_for[FF_B_TYPE]; //p_lambda *FFABS(s->avctx->b_quant_factor) + s->avctx->b_quant_offset;
973     if(!b_lambda) b_lambda= p_lambda; //FIXME we should do this somewhere else
974     lambda2= (b_lambda*b_lambda + (1<<FF_LAMBDA_SHIFT)/2 ) >> FF_LAMBDA_SHIFT;
975
976     c->width = s->width >> scale;
977     c->height= s->height>> scale;
978     c->flags= CODEC_FLAG_QSCALE | CODEC_FLAG_PSNR | CODEC_FLAG_INPUT_PRESERVED /*| CODEC_FLAG_EMU_EDGE*/;
979     c->flags|= s->avctx->flags & CODEC_FLAG_QPEL;
980     c->mb_decision= s->avctx->mb_decision;
981     c->me_cmp= s->avctx->me_cmp;
982     c->mb_cmp= s->avctx->mb_cmp;
983     c->me_sub_cmp= s->avctx->me_sub_cmp;
984     c->pix_fmt = PIX_FMT_YUV420P;
985     c->time_base= s->avctx->time_base;
986     c->max_b_frames= s->max_b_frames;
987
988     if (avcodec_open(c, codec) < 0)
989         return -1;
990
991     for(i=0; i<s->max_b_frames+2; i++){
992         int ysize= c->width*c->height;
993         int csize= (c->width/2)*(c->height/2);
994         Picture pre_input, *pre_input_ptr= i ? s->input_picture[i-1] : s->next_picture_ptr;
995
996         avcodec_get_frame_defaults(&input[i]);
997         input[i].data[0]= av_malloc(ysize + 2*csize);
998         input[i].data[1]= input[i].data[0] + ysize;
999         input[i].data[2]= input[i].data[1] + csize;
1000         input[i].linesize[0]= c->width;
1001         input[i].linesize[1]=
1002         input[i].linesize[2]= c->width/2;
1003
1004         if(pre_input_ptr && (!i || s->input_picture[i-1])) {
1005             pre_input= *pre_input_ptr;
1006
1007             if(pre_input.type != FF_BUFFER_TYPE_SHARED && i) {
1008                 pre_input.data[0]+=INPLACE_OFFSET;
1009                 pre_input.data[1]+=INPLACE_OFFSET;
1010                 pre_input.data[2]+=INPLACE_OFFSET;
1011             }
1012
1013             s->dsp.shrink[scale](input[i].data[0], input[i].linesize[0], pre_input.data[0], pre_input.linesize[0], c->width, c->height);
1014             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);
1015             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);
1016         }
1017     }
1018
1019     for(j=0; j<s->max_b_frames+1; j++){
1020         int64_t rd=0;
1021
1022         if(!s->input_picture[j])
1023             break;
1024
1025         c->error[0]= c->error[1]= c->error[2]= 0;
1026
1027         input[0].pict_type= FF_I_TYPE;
1028         input[0].quality= 1 * FF_QP2LAMBDA;
1029         out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[0]);
1030 //        rd += (out_size * lambda2) >> FF_LAMBDA_SHIFT;
1031
1032         for(i=0; i<s->max_b_frames+1; i++){
1033             int is_p= i % (j+1) == j || i==s->max_b_frames;
1034
1035             input[i+1].pict_type= is_p ? FF_P_TYPE : FF_B_TYPE;
1036             input[i+1].quality= is_p ? p_lambda : b_lambda;
1037             out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[i+1]);
1038             rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
1039         }
1040
1041         /* get the delayed frames */
1042         while(out_size){
1043             out_size = avcodec_encode_video(c, outbuf, outbuf_size, NULL);
1044             rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
1045         }
1046
1047         rd += c->error[0] + c->error[1] + c->error[2];
1048
1049         if(rd < best_rd){
1050             best_rd= rd;
1051             best_b_count= j;
1052         }
1053     }
1054
1055     av_freep(&outbuf);
1056     avcodec_close(c);
1057     av_freep(&c);
1058
1059     for(i=0; i<s->max_b_frames+2; i++){
1060         av_freep(&input[i].data[0]);
1061     }
1062
1063     return best_b_count;
1064 }
1065
1066 static int select_input_picture(MpegEncContext *s){
1067     int i;
1068
1069     for(i=1; i<MAX_PICTURE_COUNT; i++)
1070         s->reordered_input_picture[i-1]= s->reordered_input_picture[i];
1071     s->reordered_input_picture[MAX_PICTURE_COUNT-1]= NULL;
1072
1073     /* set next picture type & ordering */
1074     if(s->reordered_input_picture[0]==NULL && s->input_picture[0]){
1075         if(/*s->picture_in_gop_number >= s->gop_size ||*/ s->next_picture_ptr==NULL || s->intra_only){
1076             s->reordered_input_picture[0]= s->input_picture[0];
1077             s->reordered_input_picture[0]->pict_type= FF_I_TYPE;
1078             s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++;
1079         }else{
1080             int b_frames;
1081
1082             if(s->avctx->frame_skip_threshold || s->avctx->frame_skip_factor){
1083                 if(s->picture_in_gop_number < s->gop_size && skip_check(s, s->input_picture[0], s->next_picture_ptr)){
1084                 //FIXME check that te gop check above is +-1 correct
1085 //av_log(NULL, AV_LOG_DEBUG, "skip %p %"PRId64"\n", s->input_picture[0]->data[0], s->input_picture[0]->pts);
1086
1087                     if(s->input_picture[0]->type == FF_BUFFER_TYPE_SHARED){
1088                         for(i=0; i<4; i++)
1089                             s->input_picture[0]->data[i]= NULL;
1090                         s->input_picture[0]->type= 0;
1091                     }else{
1092                         assert(   s->input_picture[0]->type==FF_BUFFER_TYPE_USER
1093                                || s->input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
1094
1095                         s->avctx->release_buffer(s->avctx, (AVFrame*)s->input_picture[0]);
1096                     }
1097
1098                     emms_c();
1099                     ff_vbv_update(s, 0);
1100
1101                     goto no_output_pic;
1102                 }
1103             }
1104
1105             if(s->flags&CODEC_FLAG_PASS2){
1106                 for(i=0; i<s->max_b_frames+1; i++){
1107                     int pict_num= s->input_picture[0]->display_picture_number + i;
1108
1109                     if(pict_num >= s->rc_context.num_entries)
1110                         break;
1111                     if(!s->input_picture[i]){
1112                         s->rc_context.entry[pict_num-1].new_pict_type = FF_P_TYPE;
1113                         break;
1114                     }
1115
1116                     s->input_picture[i]->pict_type=
1117                         s->rc_context.entry[pict_num].new_pict_type;
1118                 }
1119             }
1120
1121             if(s->avctx->b_frame_strategy==0){
1122                 b_frames= s->max_b_frames;
1123                 while(b_frames && !s->input_picture[b_frames]) b_frames--;
1124             }else if(s->avctx->b_frame_strategy==1){
1125                 for(i=1; i<s->max_b_frames+1; i++){
1126                     if(s->input_picture[i] && s->input_picture[i]->b_frame_score==0){
1127                         s->input_picture[i]->b_frame_score=
1128                             get_intra_count(s, s->input_picture[i  ]->data[0],
1129                                                s->input_picture[i-1]->data[0], s->linesize) + 1;
1130                     }
1131                 }
1132                 for(i=0; i<s->max_b_frames+1; i++){
1133                     if(s->input_picture[i]==NULL || s->input_picture[i]->b_frame_score - 1 > s->mb_num/s->avctx->b_sensitivity) break;
1134                 }
1135
1136                 b_frames= FFMAX(0, i-1);
1137
1138                 /* reset scores */
1139                 for(i=0; i<b_frames+1; i++){
1140                     s->input_picture[i]->b_frame_score=0;
1141                 }
1142             }else if(s->avctx->b_frame_strategy==2){
1143                 b_frames= estimate_best_b_count(s);
1144             }else{
1145                 av_log(s->avctx, AV_LOG_ERROR, "illegal b frame strategy\n");
1146                 b_frames=0;
1147             }
1148
1149             emms_c();
1150 //static int b_count=0;
1151 //b_count+= b_frames;
1152 //av_log(s->avctx, AV_LOG_DEBUG, "b_frames: %d\n", b_count);
1153
1154             for(i= b_frames - 1; i>=0; i--){
1155                 int type= s->input_picture[i]->pict_type;
1156                 if(type && type != FF_B_TYPE)
1157                     b_frames= i;
1158             }
1159             if(s->input_picture[b_frames]->pict_type == FF_B_TYPE && b_frames == s->max_b_frames){
1160                 av_log(s->avctx, AV_LOG_ERROR, "warning, too many b frames in a row\n");
1161             }
1162
1163             if(s->picture_in_gop_number + b_frames >= s->gop_size){
1164               if((s->flags2 & CODEC_FLAG2_STRICT_GOP) && s->gop_size > s->picture_in_gop_number){
1165                     b_frames= s->gop_size - s->picture_in_gop_number - 1;
1166               }else{
1167                 if(s->flags & CODEC_FLAG_CLOSED_GOP)
1168                     b_frames=0;
1169                 s->input_picture[b_frames]->pict_type= FF_I_TYPE;
1170               }
1171             }
1172
1173             if(   (s->flags & CODEC_FLAG_CLOSED_GOP)
1174                && b_frames
1175                && s->input_picture[b_frames]->pict_type== FF_I_TYPE)
1176                 b_frames--;
1177
1178             s->reordered_input_picture[0]= s->input_picture[b_frames];
1179             if(s->reordered_input_picture[0]->pict_type != FF_I_TYPE)
1180                 s->reordered_input_picture[0]->pict_type= FF_P_TYPE;
1181             s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++;
1182             for(i=0; i<b_frames; i++){
1183                 s->reordered_input_picture[i+1]= s->input_picture[i];
1184                 s->reordered_input_picture[i+1]->pict_type= FF_B_TYPE;
1185                 s->reordered_input_picture[i+1]->coded_picture_number= s->coded_picture_number++;
1186             }
1187         }
1188     }
1189 no_output_pic:
1190     if(s->reordered_input_picture[0]){
1191         s->reordered_input_picture[0]->reference= s->reordered_input_picture[0]->pict_type!=FF_B_TYPE ? 3 : 0;
1192
1193         ff_copy_picture(&s->new_picture, s->reordered_input_picture[0]);
1194
1195         if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_SHARED || s->avctx->rc_buffer_size){
1196             // input is a shared pix, so we can't modifiy it -> alloc a new one & ensure that the shared one is reuseable
1197
1198             int i= ff_find_unused_picture(s, 0);
1199             Picture *pic= &s->picture[i];
1200
1201             pic->reference              = s->reordered_input_picture[0]->reference;
1202             if(ff_alloc_picture(s, pic, 0) < 0){
1203                 return -1;
1204             }
1205
1206             /* mark us unused / free shared pic */
1207             if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_INTERNAL)
1208                 s->avctx->release_buffer(s->avctx, (AVFrame*)s->reordered_input_picture[0]);
1209             for(i=0; i<4; i++)
1210                 s->reordered_input_picture[0]->data[i]= NULL;
1211             s->reordered_input_picture[0]->type= 0;
1212
1213             copy_picture_attributes(s, (AVFrame*)pic, (AVFrame*)s->reordered_input_picture[0]);
1214
1215             s->current_picture_ptr= pic;
1216         }else{
1217             // input is not a shared pix -> reuse buffer for current_pix
1218
1219             assert(   s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_USER
1220                    || s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
1221
1222             s->current_picture_ptr= s->reordered_input_picture[0];
1223             for(i=0; i<4; i++){
1224                 s->new_picture.data[i]+= INPLACE_OFFSET;
1225             }
1226         }
1227         ff_copy_picture(&s->current_picture, s->current_picture_ptr);
1228
1229         s->picture_number= s->new_picture.display_picture_number;
1230 //printf("dpn:%d\n", s->picture_number);
1231     }else{
1232        memset(&s->new_picture, 0, sizeof(Picture));
1233     }
1234     return 0;
1235 }
1236
1237 int MPV_encode_picture(AVCodecContext *avctx,
1238                        unsigned char *buf, int buf_size, void *data)
1239 {
1240     MpegEncContext *s = avctx->priv_data;
1241     AVFrame *pic_arg = data;
1242     int i, stuffing_count, context_count = avctx->active_thread_type&FF_THREAD_SLICE ? avctx->thread_count : 1;
1243
1244     for(i=0; i<context_count; i++){
1245         int start_y= s->thread_context[i]->start_mb_y;
1246         int   end_y= s->thread_context[i]->  end_mb_y;
1247         int h= s->mb_height;
1248         uint8_t *start= buf + (size_t)(((int64_t) buf_size)*start_y/h);
1249         uint8_t *end  = buf + (size_t)(((int64_t) buf_size)*  end_y/h);
1250
1251         init_put_bits(&s->thread_context[i]->pb, start, end - start);
1252     }
1253
1254     s->picture_in_gop_number++;
1255
1256     if(load_input_picture(s, pic_arg) < 0)
1257         return -1;
1258
1259     if(select_input_picture(s) < 0){
1260         return -1;
1261     }
1262
1263     /* output? */
1264     if(s->new_picture.data[0]){
1265         s->pict_type= s->new_picture.pict_type;
1266 //emms_c();
1267 //printf("qs:%f %f %d\n", s->new_picture.quality, s->current_picture.quality, s->qscale);
1268         MPV_frame_start(s, avctx);
1269 vbv_retry:
1270         if (encode_picture(s, s->picture_number) < 0)
1271             return -1;
1272
1273         avctx->header_bits = s->header_bits;
1274         avctx->mv_bits     = s->mv_bits;
1275         avctx->misc_bits   = s->misc_bits;
1276         avctx->i_tex_bits  = s->i_tex_bits;
1277         avctx->p_tex_bits  = s->p_tex_bits;
1278         avctx->i_count     = s->i_count;
1279         avctx->p_count     = s->mb_num - s->i_count - s->skip_count; //FIXME f/b_count in avctx
1280         avctx->skip_count  = s->skip_count;
1281
1282         MPV_frame_end(s);
1283
1284         if (CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG)
1285             ff_mjpeg_encode_picture_trailer(s);
1286
1287         if(avctx->rc_buffer_size){
1288             RateControlContext *rcc= &s->rc_context;
1289             int max_size= rcc->buffer_index * avctx->rc_max_available_vbv_use;
1290
1291             if(put_bits_count(&s->pb) > max_size && s->lambda < s->avctx->lmax){
1292                 s->next_lambda= FFMAX(s->lambda+1, s->lambda*(s->qscale+1) / s->qscale);
1293                 if(s->adaptive_quant){
1294                     int i;
1295                     for(i=0; i<s->mb_height*s->mb_stride; i++)
1296                         s->lambda_table[i]= FFMAX(s->lambda_table[i]+1, s->lambda_table[i]*(s->qscale+1) / s->qscale);
1297                 }
1298                 s->mb_skipped = 0;        //done in MPV_frame_start()
1299                 if(s->pict_type==FF_P_TYPE){ //done in encode_picture() so we must undo it
1300                     if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
1301                         s->no_rounding ^= 1;
1302                 }
1303                 if(s->pict_type!=FF_B_TYPE){
1304                     s->time_base= s->last_time_base;
1305                     s->last_non_b_time= s->time - s->pp_time;
1306                 }
1307 //                av_log(NULL, AV_LOG_ERROR, "R:%d ", s->next_lambda);
1308                 for(i=0; i<context_count; i++){
1309                     PutBitContext *pb= &s->thread_context[i]->pb;
1310                     init_put_bits(pb, pb->buf, pb->buf_end - pb->buf);
1311                 }
1312                 goto vbv_retry;
1313             }
1314
1315             assert(s->avctx->rc_max_rate);
1316         }
1317
1318         if(s->flags&CODEC_FLAG_PASS1)
1319             ff_write_pass1_stats(s);
1320
1321         for(i=0; i<4; i++){
1322             s->current_picture_ptr->error[i]= s->current_picture.error[i];
1323             avctx->error[i] += s->current_picture_ptr->error[i];
1324         }
1325
1326         if(s->flags&CODEC_FLAG_PASS1)
1327             assert(avctx->header_bits + avctx->mv_bits + avctx->misc_bits + avctx->i_tex_bits + avctx->p_tex_bits == put_bits_count(&s->pb));
1328         flush_put_bits(&s->pb);
1329         s->frame_bits  = put_bits_count(&s->pb);
1330
1331         stuffing_count= ff_vbv_update(s, s->frame_bits);
1332         if(stuffing_count){
1333             if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < stuffing_count + 50){
1334                 av_log(s->avctx, AV_LOG_ERROR, "stuffing too large\n");
1335                 return -1;
1336             }
1337
1338             switch(s->codec_id){
1339             case CODEC_ID_MPEG1VIDEO:
1340             case CODEC_ID_MPEG2VIDEO:
1341                 while(stuffing_count--){
1342                     put_bits(&s->pb, 8, 0);
1343                 }
1344             break;
1345             case CODEC_ID_MPEG4:
1346                 put_bits(&s->pb, 16, 0);
1347                 put_bits(&s->pb, 16, 0x1C3);
1348                 stuffing_count -= 4;
1349                 while(stuffing_count--){
1350                     put_bits(&s->pb, 8, 0xFF);
1351                 }
1352             break;
1353             default:
1354                 av_log(s->avctx, AV_LOG_ERROR, "vbv buffer overflow\n");
1355             }
1356             flush_put_bits(&s->pb);
1357             s->frame_bits  = put_bits_count(&s->pb);
1358         }
1359
1360         /* update mpeg1/2 vbv_delay for CBR */
1361         if(s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate && s->out_format == FMT_MPEG1
1362            && 90000LL * (avctx->rc_buffer_size-1) <= s->avctx->rc_max_rate*0xFFFFLL){
1363             int vbv_delay, min_delay;
1364             double inbits = s->avctx->rc_max_rate*av_q2d(s->avctx->time_base);
1365             int    minbits= s->frame_bits - 8*(s->vbv_delay_ptr - s->pb.buf - 1);
1366             double bits   = s->rc_context.buffer_index + minbits - inbits;
1367
1368             if(bits<0)
1369                 av_log(s->avctx, AV_LOG_ERROR, "Internal error, negative bits\n");
1370
1371             assert(s->repeat_first_field==0);
1372
1373             vbv_delay=     bits * 90000                               / s->avctx->rc_max_rate;
1374             min_delay= (minbits * 90000LL + s->avctx->rc_max_rate - 1)/ s->avctx->rc_max_rate;
1375
1376             vbv_delay= FFMAX(vbv_delay, min_delay);
1377
1378             assert(vbv_delay < 0xFFFF);
1379
1380             s->vbv_delay_ptr[0] &= 0xF8;
1381             s->vbv_delay_ptr[0] |= vbv_delay>>13;
1382             s->vbv_delay_ptr[1]  = vbv_delay>>5;
1383             s->vbv_delay_ptr[2] &= 0x07;
1384             s->vbv_delay_ptr[2] |= vbv_delay<<3;
1385             avctx->vbv_delay = vbv_delay*300;
1386         }
1387         s->total_bits += s->frame_bits;
1388         avctx->frame_bits  = s->frame_bits;
1389     }else{
1390         assert((put_bits_ptr(&s->pb) == s->pb.buf));
1391         s->frame_bits=0;
1392     }
1393     assert((s->frame_bits&7)==0);
1394
1395     return s->frame_bits/8;
1396 }
1397
1398 static inline void dct_single_coeff_elimination(MpegEncContext *s, int n, int threshold)
1399 {
1400     static const char tab[64]=
1401         {3,2,2,1,1,1,1,1,
1402          1,1,1,1,1,1,1,1,
1403          1,1,1,1,1,1,1,1,
1404          0,0,0,0,0,0,0,0,
1405          0,0,0,0,0,0,0,0,
1406          0,0,0,0,0,0,0,0,
1407          0,0,0,0,0,0,0,0,
1408          0,0,0,0,0,0,0,0};
1409     int score=0;
1410     int run=0;
1411     int i;
1412     DCTELEM *block= s->block[n];
1413     const int last_index= s->block_last_index[n];
1414     int skip_dc;
1415
1416     if(threshold<0){
1417         skip_dc=0;
1418         threshold= -threshold;
1419     }else
1420         skip_dc=1;
1421
1422     /* Are all we could set to zero already zero? */
1423     if(last_index<=skip_dc - 1) return;
1424
1425     for(i=0; i<=last_index; i++){
1426         const int j = s->intra_scantable.permutated[i];
1427         const int level = FFABS(block[j]);
1428         if(level==1){
1429             if(skip_dc && i==0) continue;
1430             score+= tab[run];
1431             run=0;
1432         }else if(level>1){
1433             return;
1434         }else{
1435             run++;
1436         }
1437     }
1438     if(score >= threshold) return;
1439     for(i=skip_dc; i<=last_index; i++){
1440         const int j = s->intra_scantable.permutated[i];
1441         block[j]=0;
1442     }
1443     if(block[0]) s->block_last_index[n]= 0;
1444     else         s->block_last_index[n]= -1;
1445 }
1446
1447 static inline void clip_coeffs(MpegEncContext *s, DCTELEM *block, int last_index)
1448 {
1449     int i;
1450     const int maxlevel= s->max_qcoeff;
1451     const int minlevel= s->min_qcoeff;
1452     int overflow=0;
1453
1454     if(s->mb_intra){
1455         i=1; //skip clipping of intra dc
1456     }else
1457         i=0;
1458
1459     for(;i<=last_index; i++){
1460         const int j= s->intra_scantable.permutated[i];
1461         int level = block[j];
1462
1463         if     (level>maxlevel){
1464             level=maxlevel;
1465             overflow++;
1466         }else if(level<minlevel){
1467             level=minlevel;
1468             overflow++;
1469         }
1470
1471         block[j]= level;
1472     }
1473
1474     if(overflow && s->avctx->mb_decision == FF_MB_DECISION_SIMPLE)
1475         av_log(s->avctx, AV_LOG_INFO, "warning, clipping %d dct coefficients to %d..%d\n", overflow, minlevel, maxlevel);
1476 }
1477
1478 static void get_visual_weight(int16_t *weight, uint8_t *ptr, int stride){
1479     int x, y;
1480 //FIXME optimize
1481     for(y=0; y<8; y++){
1482         for(x=0; x<8; x++){
1483             int x2, y2;
1484             int sum=0;
1485             int sqr=0;
1486             int count=0;
1487
1488             for(y2= FFMAX(y-1, 0); y2 < FFMIN(8, y+2); y2++){
1489                 for(x2= FFMAX(x-1, 0); x2 < FFMIN(8, x+2); x2++){
1490                     int v= ptr[x2 + y2*stride];
1491                     sum += v;
1492                     sqr += v*v;
1493                     count++;
1494                 }
1495             }
1496             weight[x + 8*y]= (36*ff_sqrt(count*sqr - sum*sum)) / count;
1497         }
1498     }
1499 }
1500
1501 static av_always_inline void encode_mb_internal(MpegEncContext *s, int motion_x, int motion_y, int mb_block_height, int mb_block_count)
1502 {
1503     int16_t weight[8][64];
1504     DCTELEM orig[8][64];
1505     const int mb_x= s->mb_x;
1506     const int mb_y= s->mb_y;
1507     int i;
1508     int skip_dct[8];
1509     int dct_offset   = s->linesize*8; //default for progressive frames
1510     uint8_t *ptr_y, *ptr_cb, *ptr_cr;
1511     int wrap_y, wrap_c;
1512
1513     for(i=0; i<mb_block_count; i++) skip_dct[i]=s->skipdct;
1514
1515     if(s->adaptive_quant){
1516         const int last_qp= s->qscale;
1517         const int mb_xy= mb_x + mb_y*s->mb_stride;
1518
1519         s->lambda= s->lambda_table[mb_xy];
1520         update_qscale(s);
1521
1522         if(!(s->flags&CODEC_FLAG_QP_RD)){
1523             s->qscale= s->current_picture_ptr->qscale_table[mb_xy];
1524             s->dquant= s->qscale - last_qp;
1525
1526             if(s->out_format==FMT_H263){
1527                 s->dquant= av_clip(s->dquant, -2, 2);
1528
1529                 if(s->codec_id==CODEC_ID_MPEG4){
1530                     if(!s->mb_intra){
1531                         if(s->pict_type == FF_B_TYPE){
1532                             if(s->dquant&1 || s->mv_dir&MV_DIRECT)
1533                                 s->dquant= 0;
1534                         }
1535                         if(s->mv_type==MV_TYPE_8X8)
1536                             s->dquant=0;
1537                     }
1538                 }
1539             }
1540         }
1541         ff_set_qscale(s, last_qp + s->dquant);
1542     }else if(s->flags&CODEC_FLAG_QP_RD)
1543         ff_set_qscale(s, s->qscale + s->dquant);
1544
1545     wrap_y = s->linesize;
1546     wrap_c = s->uvlinesize;
1547     ptr_y = s->new_picture.data[0] + (mb_y * 16 * wrap_y) + mb_x * 16;
1548     ptr_cb = s->new_picture.data[1] + (mb_y * mb_block_height * wrap_c) + mb_x * 8;
1549     ptr_cr = s->new_picture.data[2] + (mb_y * mb_block_height * wrap_c) + mb_x * 8;
1550
1551     if(mb_x*16+16 > s->width || mb_y*16+16 > s->height){
1552         uint8_t *ebuf= s->edge_emu_buffer + 32;
1553         s->dsp.emulated_edge_mc(ebuf            , ptr_y , wrap_y,16,16,mb_x*16,mb_y*16, s->width   , s->height);
1554         ptr_y= ebuf;
1555         s->dsp.emulated_edge_mc(ebuf+18*wrap_y  , ptr_cb, wrap_c, 8, mb_block_height, mb_x*8, mb_y*8, s->width>>1, s->height>>1);
1556         ptr_cb= ebuf+18*wrap_y;
1557         s->dsp.emulated_edge_mc(ebuf+18*wrap_y+8, ptr_cr, wrap_c, 8, mb_block_height, mb_x*8, mb_y*8, s->width>>1, s->height>>1);
1558         ptr_cr= ebuf+18*wrap_y+8;
1559     }
1560
1561     if (s->mb_intra) {
1562         if(s->flags&CODEC_FLAG_INTERLACED_DCT){
1563             int progressive_score, interlaced_score;
1564
1565             s->interlaced_dct=0;
1566             progressive_score= s->dsp.ildct_cmp[4](s, ptr_y           , NULL, wrap_y, 8)
1567                               +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y*8, NULL, wrap_y, 8) - 400;
1568
1569             if(progressive_score > 0){
1570                 interlaced_score = s->dsp.ildct_cmp[4](s, ptr_y           , NULL, wrap_y*2, 8)
1571                                   +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y  , NULL, wrap_y*2, 8);
1572                 if(progressive_score > interlaced_score){
1573                     s->interlaced_dct=1;
1574
1575                     dct_offset= wrap_y;
1576                     wrap_y<<=1;
1577                     if (s->chroma_format == CHROMA_422)
1578                         wrap_c<<=1;
1579                 }
1580             }
1581         }
1582
1583         s->dsp.get_pixels(s->block[0], ptr_y                 , wrap_y);
1584         s->dsp.get_pixels(s->block[1], ptr_y              + 8, wrap_y);
1585         s->dsp.get_pixels(s->block[2], ptr_y + dct_offset    , wrap_y);
1586         s->dsp.get_pixels(s->block[3], ptr_y + dct_offset + 8, wrap_y);
1587
1588         if(s->flags&CODEC_FLAG_GRAY){
1589             skip_dct[4]= 1;
1590             skip_dct[5]= 1;
1591         }else{
1592             s->dsp.get_pixels(s->block[4], ptr_cb, wrap_c);
1593             s->dsp.get_pixels(s->block[5], ptr_cr, wrap_c);
1594             if(!s->chroma_y_shift){ /* 422 */
1595                 s->dsp.get_pixels(s->block[6], ptr_cb + (dct_offset>>1), wrap_c);
1596                 s->dsp.get_pixels(s->block[7], ptr_cr + (dct_offset>>1), wrap_c);
1597             }
1598         }
1599     }else{
1600         op_pixels_func (*op_pix)[4];
1601         qpel_mc_func (*op_qpix)[16];
1602         uint8_t *dest_y, *dest_cb, *dest_cr;
1603
1604         dest_y  = s->dest[0];
1605         dest_cb = s->dest[1];
1606         dest_cr = s->dest[2];
1607
1608         if ((!s->no_rounding) || s->pict_type==FF_B_TYPE){
1609             op_pix = s->dsp.put_pixels_tab;
1610             op_qpix= s->dsp.put_qpel_pixels_tab;
1611         }else{
1612             op_pix = s->dsp.put_no_rnd_pixels_tab;
1613             op_qpix= s->dsp.put_no_rnd_qpel_pixels_tab;
1614         }
1615
1616         if (s->mv_dir & MV_DIR_FORWARD) {
1617             MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.data, op_pix, op_qpix);
1618             op_pix = s->dsp.avg_pixels_tab;
1619             op_qpix= s->dsp.avg_qpel_pixels_tab;
1620         }
1621         if (s->mv_dir & MV_DIR_BACKWARD) {
1622             MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.data, op_pix, op_qpix);
1623         }
1624
1625         if(s->flags&CODEC_FLAG_INTERLACED_DCT){
1626             int progressive_score, interlaced_score;
1627
1628             s->interlaced_dct=0;
1629             progressive_score= s->dsp.ildct_cmp[0](s, dest_y           , ptr_y           , wrap_y, 8)
1630                               +s->dsp.ildct_cmp[0](s, dest_y + wrap_y*8, ptr_y + wrap_y*8, wrap_y, 8) - 400;
1631
1632             if(s->avctx->ildct_cmp == FF_CMP_VSSE) progressive_score -= 400;
1633
1634             if(progressive_score>0){
1635                 interlaced_score = s->dsp.ildct_cmp[0](s, dest_y           , ptr_y           , wrap_y*2, 8)
1636                                   +s->dsp.ildct_cmp[0](s, dest_y + wrap_y  , ptr_y + wrap_y  , wrap_y*2, 8);
1637
1638                 if(progressive_score > interlaced_score){
1639                     s->interlaced_dct=1;
1640
1641                     dct_offset= wrap_y;
1642                     wrap_y<<=1;
1643                     if (s->chroma_format == CHROMA_422)
1644                         wrap_c<<=1;
1645                 }
1646             }
1647         }
1648
1649         s->dsp.diff_pixels(s->block[0], ptr_y                 , dest_y                 , wrap_y);
1650         s->dsp.diff_pixels(s->block[1], ptr_y              + 8, dest_y              + 8, wrap_y);
1651         s->dsp.diff_pixels(s->block[2], ptr_y + dct_offset    , dest_y + dct_offset    , wrap_y);
1652         s->dsp.diff_pixels(s->block[3], ptr_y + dct_offset + 8, dest_y + dct_offset + 8, wrap_y);
1653
1654         if(s->flags&CODEC_FLAG_GRAY){
1655             skip_dct[4]= 1;
1656             skip_dct[5]= 1;
1657         }else{
1658             s->dsp.diff_pixels(s->block[4], ptr_cb, dest_cb, wrap_c);
1659             s->dsp.diff_pixels(s->block[5], ptr_cr, dest_cr, wrap_c);
1660             if(!s->chroma_y_shift){ /* 422 */
1661                 s->dsp.diff_pixels(s->block[6], ptr_cb + (dct_offset>>1), dest_cb + (dct_offset>>1), wrap_c);
1662                 s->dsp.diff_pixels(s->block[7], ptr_cr + (dct_offset>>1), dest_cr + (dct_offset>>1), wrap_c);
1663             }
1664         }
1665         /* pre quantization */
1666         if(s->current_picture.mc_mb_var[s->mb_stride*mb_y+ mb_x]<2*s->qscale*s->qscale){
1667             //FIXME optimize
1668             if(s->dsp.sad[1](NULL, ptr_y               , dest_y               , wrap_y, 8) < 20*s->qscale) skip_dct[0]= 1;
1669             if(s->dsp.sad[1](NULL, ptr_y            + 8, dest_y            + 8, wrap_y, 8) < 20*s->qscale) skip_dct[1]= 1;
1670             if(s->dsp.sad[1](NULL, ptr_y +dct_offset   , dest_y +dct_offset   , wrap_y, 8) < 20*s->qscale) skip_dct[2]= 1;
1671             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;
1672             if(s->dsp.sad[1](NULL, ptr_cb              , dest_cb              , wrap_c, 8) < 20*s->qscale) skip_dct[4]= 1;
1673             if(s->dsp.sad[1](NULL, ptr_cr              , dest_cr              , wrap_c, 8) < 20*s->qscale) skip_dct[5]= 1;
1674             if(!s->chroma_y_shift){ /* 422 */
1675                 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;
1676                 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;
1677             }
1678         }
1679     }
1680
1681     if(s->avctx->quantizer_noise_shaping){
1682         if(!skip_dct[0]) get_visual_weight(weight[0], ptr_y                 , wrap_y);
1683         if(!skip_dct[1]) get_visual_weight(weight[1], ptr_y              + 8, wrap_y);
1684         if(!skip_dct[2]) get_visual_weight(weight[2], ptr_y + dct_offset    , wrap_y);
1685         if(!skip_dct[3]) get_visual_weight(weight[3], ptr_y + dct_offset + 8, wrap_y);
1686         if(!skip_dct[4]) get_visual_weight(weight[4], ptr_cb                , wrap_c);
1687         if(!skip_dct[5]) get_visual_weight(weight[5], ptr_cr                , wrap_c);
1688         if(!s->chroma_y_shift){ /* 422 */
1689             if(!skip_dct[6]) get_visual_weight(weight[6], ptr_cb + (dct_offset>>1), wrap_c);
1690             if(!skip_dct[7]) get_visual_weight(weight[7], ptr_cr + (dct_offset>>1), wrap_c);
1691         }
1692         memcpy(orig[0], s->block[0], sizeof(DCTELEM)*64*mb_block_count);
1693     }
1694
1695     /* DCT & quantize */
1696     assert(s->out_format!=FMT_MJPEG || s->qscale==8);
1697     {
1698         for(i=0;i<mb_block_count;i++) {
1699             if(!skip_dct[i]){
1700                 int overflow;
1701                 s->block_last_index[i] = s->dct_quantize(s, s->block[i], i, s->qscale, &overflow);
1702             // FIXME we could decide to change to quantizer instead of clipping
1703             // JS: I don't think that would be a good idea it could lower quality instead
1704             //     of improve it. Just INTRADC clipping deserves changes in quantizer
1705                 if (overflow) clip_coeffs(s, s->block[i], s->block_last_index[i]);
1706             }else
1707                 s->block_last_index[i]= -1;
1708         }
1709         if(s->avctx->quantizer_noise_shaping){
1710             for(i=0;i<mb_block_count;i++) {
1711                 if(!skip_dct[i]){
1712                     s->block_last_index[i] = dct_quantize_refine(s, s->block[i], weight[i], orig[i], i, s->qscale);
1713                 }
1714             }
1715         }
1716
1717         if(s->luma_elim_threshold && !s->mb_intra)
1718             for(i=0; i<4; i++)
1719                 dct_single_coeff_elimination(s, i, s->luma_elim_threshold);
1720         if(s->chroma_elim_threshold && !s->mb_intra)
1721             for(i=4; i<mb_block_count; i++)
1722                 dct_single_coeff_elimination(s, i, s->chroma_elim_threshold);
1723
1724         if(s->flags & CODEC_FLAG_CBP_RD){
1725             for(i=0;i<mb_block_count;i++) {
1726                 if(s->block_last_index[i] == -1)
1727                     s->coded_score[i]= INT_MAX/256;
1728             }
1729         }
1730     }
1731
1732     if((s->flags&CODEC_FLAG_GRAY) && s->mb_intra){
1733         s->block_last_index[4]=
1734         s->block_last_index[5]= 0;
1735         s->block[4][0]=
1736         s->block[5][0]= (1024 + s->c_dc_scale/2)/ s->c_dc_scale;
1737     }
1738
1739     //non c quantize code returns incorrect block_last_index FIXME
1740     if(s->alternate_scan && s->dct_quantize != dct_quantize_c){
1741         for(i=0; i<mb_block_count; i++){
1742             int j;
1743             if(s->block_last_index[i]>0){
1744                 for(j=63; j>0; j--){
1745                     if(s->block[i][ s->intra_scantable.permutated[j] ]) break;
1746                 }
1747                 s->block_last_index[i]= j;
1748             }
1749         }
1750     }
1751
1752     /* huffman encode */
1753     switch(s->codec_id){ //FIXME funct ptr could be slightly faster
1754     case CODEC_ID_MPEG1VIDEO:
1755     case CODEC_ID_MPEG2VIDEO:
1756         if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
1757             mpeg1_encode_mb(s, s->block, motion_x, motion_y);
1758         break;
1759     case CODEC_ID_MPEG4:
1760         if (CONFIG_MPEG4_ENCODER)
1761             mpeg4_encode_mb(s, s->block, motion_x, motion_y);
1762         break;
1763     case CODEC_ID_MSMPEG4V2:
1764     case CODEC_ID_MSMPEG4V3:
1765     case CODEC_ID_WMV1:
1766         if (CONFIG_MSMPEG4_ENCODER)
1767             msmpeg4_encode_mb(s, s->block, motion_x, motion_y);
1768         break;
1769     case CODEC_ID_WMV2:
1770         if (CONFIG_WMV2_ENCODER)
1771             ff_wmv2_encode_mb(s, s->block, motion_x, motion_y);
1772         break;
1773     case CODEC_ID_H261:
1774         if (CONFIG_H261_ENCODER)
1775             ff_h261_encode_mb(s, s->block, motion_x, motion_y);
1776         break;
1777     case CODEC_ID_H263:
1778     case CODEC_ID_H263P:
1779     case CODEC_ID_FLV1:
1780     case CODEC_ID_RV10:
1781     case CODEC_ID_RV20:
1782         if (CONFIG_H263_ENCODER)
1783             h263_encode_mb(s, s->block, motion_x, motion_y);
1784         break;
1785     case CODEC_ID_MJPEG:
1786         if (CONFIG_MJPEG_ENCODER)
1787             ff_mjpeg_encode_mb(s, s->block);
1788         break;
1789     default:
1790         assert(0);
1791     }
1792 }
1793
1794 static av_always_inline void encode_mb(MpegEncContext *s, int motion_x, int motion_y)
1795 {
1796     if (s->chroma_format == CHROMA_420) encode_mb_internal(s, motion_x, motion_y,  8, 6);
1797     else                                encode_mb_internal(s, motion_x, motion_y, 16, 8);
1798 }
1799
1800 static inline void copy_context_before_encode(MpegEncContext *d, MpegEncContext *s, int type){
1801     int i;
1802
1803     memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
1804
1805     /* mpeg1 */
1806     d->mb_skip_run= s->mb_skip_run;
1807     for(i=0; i<3; i++)
1808         d->last_dc[i]= s->last_dc[i];
1809
1810     /* statistics */
1811     d->mv_bits= s->mv_bits;
1812     d->i_tex_bits= s->i_tex_bits;
1813     d->p_tex_bits= s->p_tex_bits;
1814     d->i_count= s->i_count;
1815     d->f_count= s->f_count;
1816     d->b_count= s->b_count;
1817     d->skip_count= s->skip_count;
1818     d->misc_bits= s->misc_bits;
1819     d->last_bits= 0;
1820
1821     d->mb_skipped= 0;
1822     d->qscale= s->qscale;
1823     d->dquant= s->dquant;
1824
1825     d->esc3_level_length= s->esc3_level_length;
1826 }
1827
1828 static inline void copy_context_after_encode(MpegEncContext *d, MpegEncContext *s, int type){
1829     int i;
1830
1831     memcpy(d->mv, s->mv, 2*4*2*sizeof(int));
1832     memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
1833
1834     /* mpeg1 */
1835     d->mb_skip_run= s->mb_skip_run;
1836     for(i=0; i<3; i++)
1837         d->last_dc[i]= s->last_dc[i];
1838
1839     /* statistics */
1840     d->mv_bits= s->mv_bits;
1841     d->i_tex_bits= s->i_tex_bits;
1842     d->p_tex_bits= s->p_tex_bits;
1843     d->i_count= s->i_count;
1844     d->f_count= s->f_count;
1845     d->b_count= s->b_count;
1846     d->skip_count= s->skip_count;
1847     d->misc_bits= s->misc_bits;
1848
1849     d->mb_intra= s->mb_intra;
1850     d->mb_skipped= s->mb_skipped;
1851     d->mv_type= s->mv_type;
1852     d->mv_dir= s->mv_dir;
1853     d->pb= s->pb;
1854     if(s->data_partitioning){
1855         d->pb2= s->pb2;
1856         d->tex_pb= s->tex_pb;
1857     }
1858     d->block= s->block;
1859     for(i=0; i<8; i++)
1860         d->block_last_index[i]= s->block_last_index[i];
1861     d->interlaced_dct= s->interlaced_dct;
1862     d->qscale= s->qscale;
1863
1864     d->esc3_level_length= s->esc3_level_length;
1865 }
1866
1867 static inline void encode_mb_hq(MpegEncContext *s, MpegEncContext *backup, MpegEncContext *best, int type,
1868                            PutBitContext pb[2], PutBitContext pb2[2], PutBitContext tex_pb[2],
1869                            int *dmin, int *next_block, int motion_x, int motion_y)
1870 {
1871     int score;
1872     uint8_t *dest_backup[3];
1873
1874     copy_context_before_encode(s, backup, type);
1875
1876     s->block= s->blocks[*next_block];
1877     s->pb= pb[*next_block];
1878     if(s->data_partitioning){
1879         s->pb2   = pb2   [*next_block];
1880         s->tex_pb= tex_pb[*next_block];
1881     }
1882
1883     if(*next_block){
1884         memcpy(dest_backup, s->dest, sizeof(s->dest));
1885         s->dest[0] = s->rd_scratchpad;
1886         s->dest[1] = s->rd_scratchpad + 16*s->linesize;
1887         s->dest[2] = s->rd_scratchpad + 16*s->linesize + 8;
1888         assert(s->linesize >= 32); //FIXME
1889     }
1890
1891     encode_mb(s, motion_x, motion_y);
1892
1893     score= put_bits_count(&s->pb);
1894     if(s->data_partitioning){
1895         score+= put_bits_count(&s->pb2);
1896         score+= put_bits_count(&s->tex_pb);
1897     }
1898
1899     if(s->avctx->mb_decision == FF_MB_DECISION_RD){
1900         MPV_decode_mb(s, s->block);
1901
1902         score *= s->lambda2;
1903         score += sse_mb(s) << FF_LAMBDA_SHIFT;
1904     }
1905
1906     if(*next_block){
1907         memcpy(s->dest, dest_backup, sizeof(s->dest));
1908     }
1909
1910     if(score<*dmin){
1911         *dmin= score;
1912         *next_block^=1;
1913
1914         copy_context_after_encode(best, s, type);
1915     }
1916 }
1917
1918 static int sse(MpegEncContext *s, uint8_t *src1, uint8_t *src2, int w, int h, int stride){
1919     uint32_t *sq = ff_squareTbl + 256;
1920     int acc=0;
1921     int x,y;
1922
1923     if(w==16 && h==16)
1924         return s->dsp.sse[0](NULL, src1, src2, stride, 16);
1925     else if(w==8 && h==8)
1926         return s->dsp.sse[1](NULL, src1, src2, stride, 8);
1927
1928     for(y=0; y<h; y++){
1929         for(x=0; x<w; x++){
1930             acc+= sq[src1[x + y*stride] - src2[x + y*stride]];
1931         }
1932     }
1933
1934     assert(acc>=0);
1935
1936     return acc;
1937 }
1938
1939 static int sse_mb(MpegEncContext *s){
1940     int w= 16;
1941     int h= 16;
1942
1943     if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
1944     if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
1945
1946     if(w==16 && h==16)
1947       if(s->avctx->mb_cmp == FF_CMP_NSSE){
1948         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)
1949                +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)
1950                +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);
1951       }else{
1952         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)
1953                +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)
1954                +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);
1955       }
1956     else
1957         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)
1958                +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)
1959                +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);
1960 }
1961
1962 static int pre_estimate_motion_thread(AVCodecContext *c, void *arg){
1963     MpegEncContext *s= *(void**)arg;
1964
1965
1966     s->me.pre_pass=1;
1967     s->me.dia_size= s->avctx->pre_dia_size;
1968     s->first_slice_line=1;
1969     for(s->mb_y= s->end_mb_y-1; s->mb_y >= s->start_mb_y; s->mb_y--) {
1970         for(s->mb_x=s->mb_width-1; s->mb_x >=0 ;s->mb_x--) {
1971             ff_pre_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
1972         }
1973         s->first_slice_line=0;
1974     }
1975
1976     s->me.pre_pass=0;
1977
1978     return 0;
1979 }
1980
1981 static int estimate_motion_thread(AVCodecContext *c, void *arg){
1982     MpegEncContext *s= *(void**)arg;
1983
1984     ff_check_alignment();
1985
1986     s->me.dia_size= s->avctx->dia_size;
1987     s->first_slice_line=1;
1988     for(s->mb_y= s->start_mb_y; s->mb_y < s->end_mb_y; s->mb_y++) {
1989         s->mb_x=0; //for block init below
1990         ff_init_block_index(s);
1991         for(s->mb_x=0; s->mb_x < s->mb_width; s->mb_x++) {
1992             s->block_index[0]+=2;
1993             s->block_index[1]+=2;
1994             s->block_index[2]+=2;
1995             s->block_index[3]+=2;
1996
1997             /* compute motion vector & mb_type and store in context */
1998             if(s->pict_type==FF_B_TYPE)
1999                 ff_estimate_b_frame_motion(s, s->mb_x, s->mb_y);
2000             else
2001                 ff_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
2002         }
2003         s->first_slice_line=0;
2004     }
2005     return 0;
2006 }
2007
2008 static int mb_var_thread(AVCodecContext *c, void *arg){
2009     MpegEncContext *s= *(void**)arg;
2010     int mb_x, mb_y;
2011
2012     ff_check_alignment();
2013
2014     for(mb_y=s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
2015         for(mb_x=0; mb_x < s->mb_width; mb_x++) {
2016             int xx = mb_x * 16;
2017             int yy = mb_y * 16;
2018             uint8_t *pix = s->new_picture.data[0] + (yy * s->linesize) + xx;
2019             int varc;
2020             int sum = s->dsp.pix_sum(pix, s->linesize);
2021
2022             varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500 + 128)>>8;
2023
2024             s->current_picture.mb_var [s->mb_stride * mb_y + mb_x] = varc;
2025             s->current_picture.mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
2026             s->me.mb_var_sum_temp    += varc;
2027         }
2028     }
2029     return 0;
2030 }
2031
2032 static void write_slice_end(MpegEncContext *s){
2033     if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4){
2034         if(s->partitioned_frame){
2035             ff_mpeg4_merge_partitions(s);
2036         }
2037
2038         ff_mpeg4_stuffing(&s->pb);
2039     }else if(CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG){
2040         ff_mjpeg_encode_stuffing(&s->pb);
2041     }
2042
2043     align_put_bits(&s->pb);
2044     flush_put_bits(&s->pb);
2045
2046     if((s->flags&CODEC_FLAG_PASS1) && !s->partitioned_frame)
2047         s->misc_bits+= get_bits_diff(s);
2048 }
2049
2050 static int encode_thread(AVCodecContext *c, void *arg){
2051     MpegEncContext *s= *(void**)arg;
2052     int mb_x, mb_y, pdif = 0;
2053     int chr_h= 16>>s->chroma_y_shift;
2054     int i, j;
2055     MpegEncContext best_s, backup_s;
2056     uint8_t bit_buf[2][MAX_MB_BYTES];
2057     uint8_t bit_buf2[2][MAX_MB_BYTES];
2058     uint8_t bit_buf_tex[2][MAX_MB_BYTES];
2059     PutBitContext pb[2], pb2[2], tex_pb[2];
2060 //printf("%d->%d\n", s->resync_mb_y, s->end_mb_y);
2061
2062     ff_check_alignment();
2063
2064     for(i=0; i<2; i++){
2065         init_put_bits(&pb    [i], bit_buf    [i], MAX_MB_BYTES);
2066         init_put_bits(&pb2   [i], bit_buf2   [i], MAX_MB_BYTES);
2067         init_put_bits(&tex_pb[i], bit_buf_tex[i], MAX_MB_BYTES);
2068     }
2069
2070     s->last_bits= put_bits_count(&s->pb);
2071     s->mv_bits=0;
2072     s->misc_bits=0;
2073     s->i_tex_bits=0;
2074     s->p_tex_bits=0;
2075     s->i_count=0;
2076     s->f_count=0;
2077     s->b_count=0;
2078     s->skip_count=0;
2079
2080     for(i=0; i<3; i++){
2081         /* init last dc values */
2082         /* note: quant matrix value (8) is implied here */
2083         s->last_dc[i] = 128 << s->intra_dc_precision;
2084
2085         s->current_picture.error[i] = 0;
2086     }
2087     s->mb_skip_run = 0;
2088     memset(s->last_mv, 0, sizeof(s->last_mv));
2089
2090     s->last_mv_dir = 0;
2091
2092     switch(s->codec_id){
2093     case CODEC_ID_H263:
2094     case CODEC_ID_H263P:
2095     case CODEC_ID_FLV1:
2096         if (CONFIG_H263_ENCODER)
2097             s->gob_index = ff_h263_get_gob_height(s);
2098         break;
2099     case CODEC_ID_MPEG4:
2100         if(CONFIG_MPEG4_ENCODER && s->partitioned_frame)
2101             ff_mpeg4_init_partitions(s);
2102         break;
2103     }
2104
2105     s->resync_mb_x=0;
2106     s->resync_mb_y=0;
2107     s->first_slice_line = 1;
2108     s->ptr_lastgob = s->pb.buf;
2109     for(mb_y= s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
2110 //    printf("row %d at %X\n", s->mb_y, (int)s);
2111         s->mb_x=0;
2112         s->mb_y= mb_y;
2113
2114         ff_set_qscale(s, s->qscale);
2115         ff_init_block_index(s);
2116
2117         for(mb_x=0; mb_x < s->mb_width; mb_x++) {
2118             int xy= mb_y*s->mb_stride + mb_x; // removed const, H261 needs to adjust this
2119             int mb_type= s->mb_type[xy];
2120 //            int d;
2121             int dmin= INT_MAX;
2122             int dir;
2123
2124             if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < MAX_MB_BYTES){
2125                 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
2126                 return -1;
2127             }
2128             if(s->data_partitioning){
2129                 if(   s->pb2   .buf_end - s->pb2   .buf - (put_bits_count(&s->    pb2)>>3) < MAX_MB_BYTES
2130                    || s->tex_pb.buf_end - s->tex_pb.buf - (put_bits_count(&s->tex_pb )>>3) < MAX_MB_BYTES){
2131                     av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
2132                     return -1;
2133                 }
2134             }
2135
2136             s->mb_x = mb_x;
2137             s->mb_y = mb_y;  // moved into loop, can get changed by H.261
2138             ff_update_block_index(s);
2139
2140             if(CONFIG_H261_ENCODER && s->codec_id == CODEC_ID_H261){
2141                 ff_h261_reorder_mb_index(s);
2142                 xy= s->mb_y*s->mb_stride + s->mb_x;
2143                 mb_type= s->mb_type[xy];
2144             }
2145
2146             /* write gob / video packet header  */
2147             if(s->rtp_mode){
2148                 int current_packet_size, is_gob_start;
2149
2150                 current_packet_size= ((put_bits_count(&s->pb)+7)>>3) - (s->ptr_lastgob - s->pb.buf);
2151
2152                 is_gob_start= s->avctx->rtp_payload_size && current_packet_size >= s->avctx->rtp_payload_size && mb_y + mb_x>0;
2153
2154                 if(s->start_mb_y == mb_y && mb_y > 0 && mb_x==0) is_gob_start=1;
2155
2156                 switch(s->codec_id){
2157                 case CODEC_ID_H263:
2158                 case CODEC_ID_H263P:
2159                     if(!s->h263_slice_structured)
2160                         if(s->mb_x || s->mb_y%s->gob_index) is_gob_start=0;
2161                     break;
2162                 case CODEC_ID_MPEG2VIDEO:
2163                     if(s->mb_x==0 && s->mb_y!=0) is_gob_start=1;
2164                 case CODEC_ID_MPEG1VIDEO:
2165                     if(s->mb_skip_run) is_gob_start=0;
2166                     break;
2167                 }
2168
2169                 if(is_gob_start){
2170                     if(s->start_mb_y != mb_y || mb_x!=0){
2171                         write_slice_end(s);
2172
2173                         if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4 && s->partitioned_frame){
2174                             ff_mpeg4_init_partitions(s);
2175                         }
2176                     }
2177
2178                     assert((put_bits_count(&s->pb)&7) == 0);
2179                     current_packet_size= put_bits_ptr(&s->pb) - s->ptr_lastgob;
2180
2181                     if(s->avctx->error_rate && s->resync_mb_x + s->resync_mb_y > 0){
2182                         int r= put_bits_count(&s->pb)/8 + s->picture_number + 16 + s->mb_x + s->mb_y;
2183                         int d= 100 / s->avctx->error_rate;
2184                         if(r % d == 0){
2185                             current_packet_size=0;
2186 #ifndef ALT_BITSTREAM_WRITER
2187                             s->pb.buf_ptr= s->ptr_lastgob;
2188 #endif
2189                             assert(put_bits_ptr(&s->pb) == s->ptr_lastgob);
2190                         }
2191                     }
2192
2193                     if (s->avctx->rtp_callback){
2194                         int number_mb = (mb_y - s->resync_mb_y)*s->mb_width + mb_x - s->resync_mb_x;
2195                         s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, current_packet_size, number_mb);
2196                     }
2197
2198                     switch(s->codec_id){
2199                     case CODEC_ID_MPEG4:
2200                         if (CONFIG_MPEG4_ENCODER) {
2201                             ff_mpeg4_encode_video_packet_header(s);
2202                             ff_mpeg4_clean_buffers(s);
2203                         }
2204                     break;
2205                     case CODEC_ID_MPEG1VIDEO:
2206                     case CODEC_ID_MPEG2VIDEO:
2207                         if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER) {
2208                             ff_mpeg1_encode_slice_header(s);
2209                             ff_mpeg1_clean_buffers(s);
2210                         }
2211                     break;
2212                     case CODEC_ID_H263:
2213                     case CODEC_ID_H263P:
2214                         if (CONFIG_H263_ENCODER)
2215                             h263_encode_gob_header(s, mb_y);
2216                     break;
2217                     }
2218
2219                     if(s->flags&CODEC_FLAG_PASS1){
2220                         int bits= put_bits_count(&s->pb);
2221                         s->misc_bits+= bits - s->last_bits;
2222                         s->last_bits= bits;
2223                     }
2224
2225                     s->ptr_lastgob += current_packet_size;
2226                     s->first_slice_line=1;
2227                     s->resync_mb_x=mb_x;
2228                     s->resync_mb_y=mb_y;
2229                 }
2230             }
2231
2232             if(  (s->resync_mb_x   == s->mb_x)
2233                && s->resync_mb_y+1 == s->mb_y){
2234                 s->first_slice_line=0;
2235             }
2236
2237             s->mb_skipped=0;
2238             s->dquant=0; //only for QP_RD
2239
2240             if(mb_type & (mb_type-1) || (s->flags & CODEC_FLAG_QP_RD)){ // more than 1 MB type possible or CODEC_FLAG_QP_RD
2241                 int next_block=0;
2242                 int pb_bits_count, pb2_bits_count, tex_pb_bits_count;
2243
2244                 copy_context_before_encode(&backup_s, s, -1);
2245                 backup_s.pb= s->pb;
2246                 best_s.data_partitioning= s->data_partitioning;
2247                 best_s.partitioned_frame= s->partitioned_frame;
2248                 if(s->data_partitioning){
2249                     backup_s.pb2= s->pb2;
2250                     backup_s.tex_pb= s->tex_pb;
2251                 }
2252
2253                 if(mb_type&CANDIDATE_MB_TYPE_INTER){
2254                     s->mv_dir = MV_DIR_FORWARD;
2255                     s->mv_type = MV_TYPE_16X16;
2256                     s->mb_intra= 0;
2257                     s->mv[0][0][0] = s->p_mv_table[xy][0];
2258                     s->mv[0][0][1] = s->p_mv_table[xy][1];
2259                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER, pb, pb2, tex_pb,
2260                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2261                 }
2262                 if(mb_type&CANDIDATE_MB_TYPE_INTER_I){
2263                     s->mv_dir = MV_DIR_FORWARD;
2264                     s->mv_type = MV_TYPE_FIELD;
2265                     s->mb_intra= 0;
2266                     for(i=0; i<2; i++){
2267                         j= s->field_select[0][i] = s->p_field_select_table[i][xy];
2268                         s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
2269                         s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
2270                     }
2271                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER_I, pb, pb2, tex_pb,
2272                                  &dmin, &next_block, 0, 0);
2273                 }
2274                 if(mb_type&CANDIDATE_MB_TYPE_SKIPPED){
2275                     s->mv_dir = MV_DIR_FORWARD;
2276                     s->mv_type = MV_TYPE_16X16;
2277                     s->mb_intra= 0;
2278                     s->mv[0][0][0] = 0;
2279                     s->mv[0][0][1] = 0;
2280                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_SKIPPED, pb, pb2, tex_pb,
2281                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2282                 }
2283                 if(mb_type&CANDIDATE_MB_TYPE_INTER4V){
2284                     s->mv_dir = MV_DIR_FORWARD;
2285                     s->mv_type = MV_TYPE_8X8;
2286                     s->mb_intra= 0;
2287                     for(i=0; i<4; i++){
2288                         s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
2289                         s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
2290                     }
2291                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER4V, pb, pb2, tex_pb,
2292                                  &dmin, &next_block, 0, 0);
2293                 }
2294                 if(mb_type&CANDIDATE_MB_TYPE_FORWARD){
2295                     s->mv_dir = MV_DIR_FORWARD;
2296                     s->mv_type = MV_TYPE_16X16;
2297                     s->mb_intra= 0;
2298                     s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
2299                     s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2300                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD, pb, pb2, tex_pb,
2301                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2302                 }
2303                 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD){
2304                     s->mv_dir = MV_DIR_BACKWARD;
2305                     s->mv_type = MV_TYPE_16X16;
2306                     s->mb_intra= 0;
2307                     s->mv[1][0][0] = s->b_back_mv_table[xy][0];
2308                     s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2309                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD, pb, pb2, tex_pb,
2310                                  &dmin, &next_block, s->mv[1][0][0], s->mv[1][0][1]);
2311                 }
2312                 if(mb_type&CANDIDATE_MB_TYPE_BIDIR){
2313                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2314                     s->mv_type = MV_TYPE_16X16;
2315                     s->mb_intra= 0;
2316                     s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
2317                     s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
2318                     s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
2319                     s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2320                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR, pb, pb2, tex_pb,
2321                                  &dmin, &next_block, 0, 0);
2322                 }
2323                 if(mb_type&CANDIDATE_MB_TYPE_FORWARD_I){
2324                     s->mv_dir = MV_DIR_FORWARD;
2325                     s->mv_type = MV_TYPE_FIELD;
2326                     s->mb_intra= 0;
2327                     for(i=0; i<2; i++){
2328                         j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
2329                         s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
2330                         s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
2331                     }
2332                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD_I, pb, pb2, tex_pb,
2333                                  &dmin, &next_block, 0, 0);
2334                 }
2335                 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD_I){
2336                     s->mv_dir = MV_DIR_BACKWARD;
2337                     s->mv_type = MV_TYPE_FIELD;
2338                     s->mb_intra= 0;
2339                     for(i=0; i<2; i++){
2340                         j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
2341                         s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
2342                         s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
2343                     }
2344                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD_I, pb, pb2, tex_pb,
2345                                  &dmin, &next_block, 0, 0);
2346                 }
2347                 if(mb_type&CANDIDATE_MB_TYPE_BIDIR_I){
2348                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2349                     s->mv_type = MV_TYPE_FIELD;
2350                     s->mb_intra= 0;
2351                     for(dir=0; dir<2; dir++){
2352                         for(i=0; i<2; i++){
2353                             j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
2354                             s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
2355                             s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
2356                         }
2357                     }
2358                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR_I, pb, pb2, tex_pb,
2359                                  &dmin, &next_block, 0, 0);
2360                 }
2361                 if(mb_type&CANDIDATE_MB_TYPE_INTRA){
2362                     s->mv_dir = 0;
2363                     s->mv_type = MV_TYPE_16X16;
2364                     s->mb_intra= 1;
2365                     s->mv[0][0][0] = 0;
2366                     s->mv[0][0][1] = 0;
2367                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTRA, pb, pb2, tex_pb,
2368                                  &dmin, &next_block, 0, 0);
2369                     if(s->h263_pred || s->h263_aic){
2370                         if(best_s.mb_intra)
2371                             s->mbintra_table[mb_x + mb_y*s->mb_stride]=1;
2372                         else
2373                             ff_clean_intra_table_entries(s); //old mode?
2374                     }
2375                 }
2376
2377                 if((s->flags & CODEC_FLAG_QP_RD) && dmin < INT_MAX){
2378                     if(best_s.mv_type==MV_TYPE_16X16){ //FIXME move 4mv after QPRD
2379                         const int last_qp= backup_s.qscale;
2380                         int qpi, qp, dc[6];
2381                         DCTELEM ac[6][16];
2382                         const int mvdir= (best_s.mv_dir&MV_DIR_BACKWARD) ? 1 : 0;
2383                         static const int dquant_tab[4]={-1,1,-2,2};
2384
2385                         assert(backup_s.dquant == 0);
2386
2387                         //FIXME intra
2388                         s->mv_dir= best_s.mv_dir;
2389                         s->mv_type = MV_TYPE_16X16;
2390                         s->mb_intra= best_s.mb_intra;
2391                         s->mv[0][0][0] = best_s.mv[0][0][0];
2392                         s->mv[0][0][1] = best_s.mv[0][0][1];
2393                         s->mv[1][0][0] = best_s.mv[1][0][0];
2394                         s->mv[1][0][1] = best_s.mv[1][0][1];
2395
2396                         qpi = s->pict_type == FF_B_TYPE ? 2 : 0;
2397                         for(; qpi<4; qpi++){
2398                             int dquant= dquant_tab[qpi];
2399                             qp= last_qp + dquant;
2400                             if(qp < s->avctx->qmin || qp > s->avctx->qmax)
2401                                 continue;
2402                             backup_s.dquant= dquant;
2403                             if(s->mb_intra && s->dc_val[0]){
2404                                 for(i=0; i<6; i++){
2405                                     dc[i]= s->dc_val[0][ s->block_index[i] ];
2406                                     memcpy(ac[i], s->ac_val[0][s->block_index[i]], sizeof(DCTELEM)*16);
2407                                 }
2408                             }
2409
2410                             encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
2411                                          &dmin, &next_block, s->mv[mvdir][0][0], s->mv[mvdir][0][1]);
2412                             if(best_s.qscale != qp){
2413                                 if(s->mb_intra && s->dc_val[0]){
2414                                     for(i=0; i<6; i++){
2415                                         s->dc_val[0][ s->block_index[i] ]= dc[i];
2416                                         memcpy(s->ac_val[0][s->block_index[i]], ac[i], sizeof(DCTELEM)*16);
2417                                     }
2418                                 }
2419                             }
2420                         }
2421                     }
2422                 }
2423                 if(CONFIG_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT){
2424                     int mx= s->b_direct_mv_table[xy][0];
2425                     int my= s->b_direct_mv_table[xy][1];
2426
2427                     backup_s.dquant = 0;
2428                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2429                     s->mb_intra= 0;
2430                     ff_mpeg4_set_direct_mv(s, mx, my);
2431                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
2432                                  &dmin, &next_block, mx, my);
2433                 }
2434                 if(CONFIG_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT0){
2435                     backup_s.dquant = 0;
2436                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2437                     s->mb_intra= 0;
2438                     ff_mpeg4_set_direct_mv(s, 0, 0);
2439                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
2440                                  &dmin, &next_block, 0, 0);
2441                 }
2442                 if(!best_s.mb_intra && s->flags2&CODEC_FLAG2_SKIP_RD){
2443                     int coded=0;
2444                     for(i=0; i<6; i++)
2445                         coded |= s->block_last_index[i];
2446                     if(coded){
2447                         int mx,my;
2448                         memcpy(s->mv, best_s.mv, sizeof(s->mv));
2449                         if(CONFIG_MPEG4_ENCODER && best_s.mv_dir & MV_DIRECT){
2450                             mx=my=0; //FIXME find the one we actually used
2451                             ff_mpeg4_set_direct_mv(s, mx, my);
2452                         }else if(best_s.mv_dir&MV_DIR_BACKWARD){
2453                             mx= s->mv[1][0][0];
2454                             my= s->mv[1][0][1];
2455                         }else{
2456                             mx= s->mv[0][0][0];
2457                             my= s->mv[0][0][1];
2458                         }
2459
2460                         s->mv_dir= best_s.mv_dir;
2461                         s->mv_type = best_s.mv_type;
2462                         s->mb_intra= 0;
2463 /*                        s->mv[0][0][0] = best_s.mv[0][0][0];
2464                         s->mv[0][0][1] = best_s.mv[0][0][1];
2465                         s->mv[1][0][0] = best_s.mv[1][0][0];
2466                         s->mv[1][0][1] = best_s.mv[1][0][1];*/
2467                         backup_s.dquant= 0;
2468                         s->skipdct=1;
2469                         encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
2470                                         &dmin, &next_block, mx, my);
2471                         s->skipdct=0;
2472                     }
2473                 }
2474
2475                 s->current_picture.qscale_table[xy]= best_s.qscale;
2476
2477                 copy_context_after_encode(s, &best_s, -1);
2478
2479                 pb_bits_count= put_bits_count(&s->pb);
2480                 flush_put_bits(&s->pb);
2481                 ff_copy_bits(&backup_s.pb, bit_buf[next_block^1], pb_bits_count);
2482                 s->pb= backup_s.pb;
2483
2484                 if(s->data_partitioning){
2485                     pb2_bits_count= put_bits_count(&s->pb2);
2486                     flush_put_bits(&s->pb2);
2487                     ff_copy_bits(&backup_s.pb2, bit_buf2[next_block^1], pb2_bits_count);
2488                     s->pb2= backup_s.pb2;
2489
2490                     tex_pb_bits_count= put_bits_count(&s->tex_pb);
2491                     flush_put_bits(&s->tex_pb);
2492                     ff_copy_bits(&backup_s.tex_pb, bit_buf_tex[next_block^1], tex_pb_bits_count);
2493                     s->tex_pb= backup_s.tex_pb;
2494                 }
2495                 s->last_bits= put_bits_count(&s->pb);
2496
2497                 if (CONFIG_H263_ENCODER &&
2498                     s->out_format == FMT_H263 && s->pict_type!=FF_B_TYPE)
2499                     ff_h263_update_motion_val(s);
2500
2501                 if(next_block==0){ //FIXME 16 vs linesize16
2502                     s->dsp.put_pixels_tab[0][0](s->dest[0], s->rd_scratchpad                     , s->linesize  ,16);
2503                     s->dsp.put_pixels_tab[1][0](s->dest[1], s->rd_scratchpad + 16*s->linesize    , s->uvlinesize, 8);
2504                     s->dsp.put_pixels_tab[1][0](s->dest[2], s->rd_scratchpad + 16*s->linesize + 8, s->uvlinesize, 8);
2505                 }
2506
2507                 if(s->avctx->mb_decision == FF_MB_DECISION_BITS)
2508                     MPV_decode_mb(s, s->block);
2509             } else {
2510                 int motion_x = 0, motion_y = 0;
2511                 s->mv_type=MV_TYPE_16X16;
2512                 // only one MB-Type possible
2513
2514                 switch(mb_type){
2515                 case CANDIDATE_MB_TYPE_INTRA:
2516                     s->mv_dir = 0;
2517                     s->mb_intra= 1;
2518                     motion_x= s->mv[0][0][0] = 0;
2519                     motion_y= s->mv[0][0][1] = 0;
2520                     break;
2521                 case CANDIDATE_MB_TYPE_INTER:
2522                     s->mv_dir = MV_DIR_FORWARD;
2523                     s->mb_intra= 0;
2524                     motion_x= s->mv[0][0][0] = s->p_mv_table[xy][0];
2525                     motion_y= s->mv[0][0][1] = s->p_mv_table[xy][1];
2526                     break;
2527                 case CANDIDATE_MB_TYPE_INTER_I:
2528                     s->mv_dir = MV_DIR_FORWARD;
2529                     s->mv_type = MV_TYPE_FIELD;
2530                     s->mb_intra= 0;
2531                     for(i=0; i<2; i++){
2532                         j= s->field_select[0][i] = s->p_field_select_table[i][xy];
2533                         s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
2534                         s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
2535                     }
2536                     break;
2537                 case CANDIDATE_MB_TYPE_INTER4V:
2538                     s->mv_dir = MV_DIR_FORWARD;
2539                     s->mv_type = MV_TYPE_8X8;
2540                     s->mb_intra= 0;
2541                     for(i=0; i<4; i++){
2542                         s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
2543                         s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
2544                     }
2545                     break;
2546                 case CANDIDATE_MB_TYPE_DIRECT:
2547                     if (CONFIG_MPEG4_ENCODER) {
2548                         s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
2549                         s->mb_intra= 0;
2550                         motion_x=s->b_direct_mv_table[xy][0];
2551                         motion_y=s->b_direct_mv_table[xy][1];
2552                         ff_mpeg4_set_direct_mv(s, motion_x, motion_y);
2553                     }
2554                     break;
2555                 case CANDIDATE_MB_TYPE_DIRECT0:
2556                     if (CONFIG_MPEG4_ENCODER) {
2557                         s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
2558                         s->mb_intra= 0;
2559                         ff_mpeg4_set_direct_mv(s, 0, 0);
2560                     }
2561                     break;
2562                 case CANDIDATE_MB_TYPE_BIDIR:
2563                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2564                     s->mb_intra= 0;
2565                     s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
2566                     s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
2567                     s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
2568                     s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2569                     break;
2570                 case CANDIDATE_MB_TYPE_BACKWARD:
2571                     s->mv_dir = MV_DIR_BACKWARD;
2572                     s->mb_intra= 0;
2573                     motion_x= s->mv[1][0][0] = s->b_back_mv_table[xy][0];
2574                     motion_y= s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2575                     break;
2576                 case CANDIDATE_MB_TYPE_FORWARD:
2577                     s->mv_dir = MV_DIR_FORWARD;
2578                     s->mb_intra= 0;
2579                     motion_x= s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
2580                     motion_y= s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2581 //                    printf(" %d %d ", motion_x, motion_y);
2582                     break;
2583                 case CANDIDATE_MB_TYPE_FORWARD_I:
2584                     s->mv_dir = MV_DIR_FORWARD;
2585                     s->mv_type = MV_TYPE_FIELD;
2586                     s->mb_intra= 0;
2587                     for(i=0; i<2; i++){
2588                         j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
2589                         s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
2590                         s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
2591                     }
2592                     break;
2593                 case CANDIDATE_MB_TYPE_BACKWARD_I:
2594                     s->mv_dir = MV_DIR_BACKWARD;
2595                     s->mv_type = MV_TYPE_FIELD;
2596                     s->mb_intra= 0;
2597                     for(i=0; i<2; i++){
2598                         j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
2599                         s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
2600                         s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
2601                     }
2602                     break;
2603                 case CANDIDATE_MB_TYPE_BIDIR_I:
2604                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2605                     s->mv_type = MV_TYPE_FIELD;
2606                     s->mb_intra= 0;
2607                     for(dir=0; dir<2; dir++){
2608                         for(i=0; i<2; i++){
2609                             j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
2610                             s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
2611                             s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
2612                         }
2613                     }
2614                     break;
2615                 default:
2616                     av_log(s->avctx, AV_LOG_ERROR, "illegal MB type\n");
2617                 }
2618
2619                 encode_mb(s, motion_x, motion_y);
2620
2621                 // RAL: Update last macroblock type
2622                 s->last_mv_dir = s->mv_dir;
2623
2624                 if (CONFIG_H263_ENCODER &&
2625                     s->out_format == FMT_H263 && s->pict_type!=FF_B_TYPE)
2626                     ff_h263_update_motion_val(s);
2627
2628                 MPV_decode_mb(s, s->block);
2629             }
2630
2631             /* clean the MV table in IPS frames for direct mode in B frames */
2632             if(s->mb_intra /* && I,P,S_TYPE */){
2633                 s->p_mv_table[xy][0]=0;
2634                 s->p_mv_table[xy][1]=0;
2635             }
2636
2637             if(s->flags&CODEC_FLAG_PSNR){
2638                 int w= 16;
2639                 int h= 16;
2640
2641                 if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
2642                 if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
2643
2644                 s->current_picture.error[0] += sse(
2645                     s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16,
2646                     s->dest[0], w, h, s->linesize);
2647                 s->current_picture.error[1] += sse(
2648                     s, s->new_picture.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*chr_h,
2649                     s->dest[1], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
2650                 s->current_picture.error[2] += sse(
2651                     s, s->new_picture.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*chr_h,
2652                     s->dest[2], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
2653             }
2654             if(s->loop_filter){
2655                 if(CONFIG_H263_ENCODER && s->out_format == FMT_H263)
2656                     ff_h263_loop_filter(s);
2657             }
2658 //printf("MB %d %d bits\n", s->mb_x+s->mb_y*s->mb_stride, put_bits_count(&s->pb));
2659         }
2660     }
2661
2662     //not beautiful here but we must write it before flushing so it has to be here
2663     if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version && s->msmpeg4_version<4 && s->pict_type == FF_I_TYPE)
2664         msmpeg4_encode_ext_header(s);
2665
2666     write_slice_end(s);
2667
2668     /* Send the last GOB if RTP */
2669     if (s->avctx->rtp_callback) {
2670         int number_mb = (mb_y - s->resync_mb_y)*s->mb_width - s->resync_mb_x;
2671         pdif = put_bits_ptr(&s->pb) - s->ptr_lastgob;
2672         /* Call the RTP callback to send the last GOB */
2673         emms_c();
2674         s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, pdif, number_mb);
2675     }
2676
2677     return 0;
2678 }
2679
2680 #define MERGE(field) dst->field += src->field; src->field=0
2681 static void merge_context_after_me(MpegEncContext *dst, MpegEncContext *src){
2682     MERGE(me.scene_change_score);
2683     MERGE(me.mc_mb_var_sum_temp);
2684     MERGE(me.mb_var_sum_temp);
2685 }
2686
2687 static void merge_context_after_encode(MpegEncContext *dst, MpegEncContext *src){
2688     int i;
2689
2690     MERGE(dct_count[0]); //note, the other dct vars are not part of the context
2691     MERGE(dct_count[1]);
2692     MERGE(mv_bits);
2693     MERGE(i_tex_bits);
2694     MERGE(p_tex_bits);
2695     MERGE(i_count);
2696     MERGE(f_count);
2697     MERGE(b_count);
2698     MERGE(skip_count);
2699     MERGE(misc_bits);
2700     MERGE(error_count);
2701     MERGE(padding_bug_score);
2702     MERGE(current_picture.error[0]);
2703     MERGE(current_picture.error[1]);
2704     MERGE(current_picture.error[2]);
2705
2706     if(dst->avctx->noise_reduction){
2707         for(i=0; i<64; i++){
2708             MERGE(dct_error_sum[0][i]);
2709             MERGE(dct_error_sum[1][i]);
2710         }
2711     }
2712
2713     assert(put_bits_count(&src->pb) % 8 ==0);
2714     assert(put_bits_count(&dst->pb) % 8 ==0);
2715     ff_copy_bits(&dst->pb, src->pb.buf, put_bits_count(&src->pb));
2716     flush_put_bits(&dst->pb);
2717 }
2718
2719 static int estimate_qp(MpegEncContext *s, int dry_run){
2720     if (s->next_lambda){
2721         s->current_picture_ptr->quality=
2722         s->current_picture.quality = s->next_lambda;
2723         if(!dry_run) s->next_lambda= 0;
2724     } else if (!s->fixed_qscale) {
2725         s->current_picture_ptr->quality=
2726         s->current_picture.quality = ff_rate_estimate_qscale(s, dry_run);
2727         if (s->current_picture.quality < 0)
2728             return -1;
2729     }
2730
2731     if(s->adaptive_quant){
2732         switch(s->codec_id){
2733         case CODEC_ID_MPEG4:
2734             if (CONFIG_MPEG4_ENCODER)
2735                 ff_clean_mpeg4_qscales(s);
2736             break;
2737         case CODEC_ID_H263:
2738         case CODEC_ID_H263P:
2739         case CODEC_ID_FLV1:
2740             if (CONFIG_H263_ENCODER)
2741                 ff_clean_h263_qscales(s);
2742             break;
2743         default:
2744             ff_init_qscale_tab(s);
2745         }
2746
2747         s->lambda= s->lambda_table[0];
2748         //FIXME broken
2749     }else
2750         s->lambda= s->current_picture.quality;
2751 //printf("%d %d\n", s->avctx->global_quality, s->current_picture.quality);
2752     update_qscale(s);
2753     return 0;
2754 }
2755
2756 /* must be called before writing the header */
2757 static void set_frame_distances(MpegEncContext * s){
2758     assert(s->current_picture_ptr->pts != AV_NOPTS_VALUE);
2759     s->time= s->current_picture_ptr->pts*s->avctx->time_base.num;
2760
2761     if(s->pict_type==FF_B_TYPE){
2762         s->pb_time= s->pp_time - (s->last_non_b_time - s->time);
2763         assert(s->pb_time > 0 && s->pb_time < s->pp_time);
2764     }else{
2765         s->pp_time= s->time - s->last_non_b_time;
2766         s->last_non_b_time= s->time;
2767         assert(s->picture_number==0 || s->pp_time > 0);
2768     }
2769 }
2770
2771 static int encode_picture(MpegEncContext *s, int picture_number)
2772 {
2773     int i;
2774     int bits;
2775     int context_count = s->avctx->active_thread_type&FF_THREAD_SLICE ? s->avctx->thread_count : 1;
2776
2777     s->picture_number = picture_number;
2778
2779     /* Reset the average MB variance */
2780     s->me.mb_var_sum_temp    =
2781     s->me.mc_mb_var_sum_temp = 0;
2782
2783     /* we need to initialize some time vars before we can encode b-frames */
2784     // RAL: Condition added for MPEG1VIDEO
2785     if (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO || (s->h263_pred && !s->h263_msmpeg4))
2786         set_frame_distances(s);
2787     if(CONFIG_MPEG4_ENCODER && s->codec_id == CODEC_ID_MPEG4)
2788         ff_set_mpeg4_time(s);
2789
2790     s->me.scene_change_score=0;
2791
2792 //    s->lambda= s->current_picture_ptr->quality; //FIXME qscale / ... stuff for ME rate distortion
2793
2794     if(s->pict_type==FF_I_TYPE){
2795         if(s->msmpeg4_version >= 3) s->no_rounding=1;
2796         else                        s->no_rounding=0;
2797     }else if(s->pict_type!=FF_B_TYPE){
2798         if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
2799             s->no_rounding ^= 1;
2800     }
2801
2802     if(s->flags & CODEC_FLAG_PASS2){
2803         if (estimate_qp(s,1) < 0)
2804             return -1;
2805         ff_get_2pass_fcode(s);
2806     }else if(!(s->flags & CODEC_FLAG_QSCALE)){
2807         if(s->pict_type==FF_B_TYPE)
2808             s->lambda= s->last_lambda_for[s->pict_type];
2809         else
2810             s->lambda= s->last_lambda_for[s->last_non_b_pict_type];
2811         update_qscale(s);
2812     }
2813
2814     s->mb_intra=0; //for the rate distortion & bit compare functions
2815     for(i=1; i<context_count; i++){
2816         ff_update_duplicate_context(s->thread_context[i], s);
2817     }
2818
2819     if(ff_init_me(s)<0)
2820         return -1;
2821
2822     /* Estimate motion for every MB */
2823     if(s->pict_type != FF_I_TYPE){
2824         s->lambda = (s->lambda * s->avctx->me_penalty_compensation + 128)>>8;
2825         s->lambda2= (s->lambda2* (int64_t)s->avctx->me_penalty_compensation + 128)>>8;
2826         if(s->pict_type != FF_B_TYPE && s->avctx->me_threshold==0){
2827             if((s->avctx->pre_me && s->last_non_b_pict_type==FF_I_TYPE) || s->avctx->pre_me==2){
2828                 s->avctx->execute(s->avctx, pre_estimate_motion_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
2829             }
2830         }
2831
2832         s->avctx->execute(s->avctx, estimate_motion_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
2833     }else /* if(s->pict_type == FF_I_TYPE) */{
2834         /* I-Frame */
2835         for(i=0; i<s->mb_stride*s->mb_height; i++)
2836             s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
2837
2838         if(!s->fixed_qscale){
2839             /* finding spatial complexity for I-frame rate control */
2840             s->avctx->execute(s->avctx, mb_var_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
2841         }
2842     }
2843     for(i=1; i<context_count; i++){
2844         merge_context_after_me(s, s->thread_context[i]);
2845     }
2846     s->current_picture.mc_mb_var_sum= s->current_picture_ptr->mc_mb_var_sum= s->me.mc_mb_var_sum_temp;
2847     s->current_picture.   mb_var_sum= s->current_picture_ptr->   mb_var_sum= s->me.   mb_var_sum_temp;
2848     emms_c();
2849
2850     if(s->me.scene_change_score > s->avctx->scenechange_threshold && s->pict_type == FF_P_TYPE){
2851         s->pict_type= FF_I_TYPE;
2852         for(i=0; i<s->mb_stride*s->mb_height; i++)
2853             s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
2854 //printf("Scene change detected, encoding as I Frame %d %d\n", s->current_picture.mb_var_sum, s->current_picture.mc_mb_var_sum);
2855     }
2856
2857     if(!s->umvplus){
2858         if(s->pict_type==FF_P_TYPE || s->pict_type==FF_S_TYPE) {
2859             s->f_code= ff_get_best_fcode(s, s->p_mv_table, CANDIDATE_MB_TYPE_INTER);
2860
2861             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2862                 int a,b;
2863                 a= ff_get_best_fcode(s, s->p_field_mv_table[0][0], CANDIDATE_MB_TYPE_INTER_I); //FIXME field_select
2864                 b= ff_get_best_fcode(s, s->p_field_mv_table[1][1], CANDIDATE_MB_TYPE_INTER_I);
2865                 s->f_code= FFMAX3(s->f_code, a, b);
2866             }
2867
2868             ff_fix_long_p_mvs(s);
2869             ff_fix_long_mvs(s, NULL, 0, s->p_mv_table, s->f_code, CANDIDATE_MB_TYPE_INTER, 0);
2870             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2871                 int j;
2872                 for(i=0; i<2; i++){
2873                     for(j=0; j<2; j++)
2874                         ff_fix_long_mvs(s, s->p_field_select_table[i], j,
2875                                         s->p_field_mv_table[i][j], s->f_code, CANDIDATE_MB_TYPE_INTER_I, 0);
2876                 }
2877             }
2878         }
2879
2880         if(s->pict_type==FF_B_TYPE){
2881             int a, b;
2882
2883             a = ff_get_best_fcode(s, s->b_forw_mv_table, CANDIDATE_MB_TYPE_FORWARD);
2884             b = ff_get_best_fcode(s, s->b_bidir_forw_mv_table, CANDIDATE_MB_TYPE_BIDIR);
2885             s->f_code = FFMAX(a, b);
2886
2887             a = ff_get_best_fcode(s, s->b_back_mv_table, CANDIDATE_MB_TYPE_BACKWARD);
2888             b = ff_get_best_fcode(s, s->b_bidir_back_mv_table, CANDIDATE_MB_TYPE_BIDIR);
2889             s->b_code = FFMAX(a, b);
2890
2891             ff_fix_long_mvs(s, NULL, 0, s->b_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_FORWARD, 1);
2892             ff_fix_long_mvs(s, NULL, 0, s->b_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BACKWARD, 1);
2893             ff_fix_long_mvs(s, NULL, 0, s->b_bidir_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_BIDIR, 1);
2894             ff_fix_long_mvs(s, NULL, 0, s->b_bidir_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BIDIR, 1);
2895             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2896                 int dir, j;
2897                 for(dir=0; dir<2; dir++){
2898                     for(i=0; i<2; i++){
2899                         for(j=0; j<2; j++){
2900                             int type= dir ? (CANDIDATE_MB_TYPE_BACKWARD_I|CANDIDATE_MB_TYPE_BIDIR_I)
2901                                           : (CANDIDATE_MB_TYPE_FORWARD_I |CANDIDATE_MB_TYPE_BIDIR_I);
2902                             ff_fix_long_mvs(s, s->b_field_select_table[dir][i], j,
2903                                             s->b_field_mv_table[dir][i][j], dir ? s->b_code : s->f_code, type, 1);
2904                         }
2905                     }
2906                 }
2907             }
2908         }
2909     }
2910
2911     if (estimate_qp(s, 0) < 0)
2912         return -1;
2913
2914     if(s->qscale < 3 && s->max_qcoeff<=128 && s->pict_type==FF_I_TYPE && !(s->flags & CODEC_FLAG_QSCALE))
2915         s->qscale= 3; //reduce clipping problems
2916
2917     if (s->out_format == FMT_MJPEG) {
2918         /* for mjpeg, we do include qscale in the matrix */
2919         for(i=1;i<64;i++){
2920             int j= s->dsp.idct_permutation[i];
2921
2922             s->intra_matrix[j] = av_clip_uint8((ff_mpeg1_default_intra_matrix[i] * s->qscale) >> 3);
2923         }
2924         s->y_dc_scale_table=
2925         s->c_dc_scale_table= ff_mpeg2_dc_scale_table[s->intra_dc_precision];
2926         s->intra_matrix[0] = ff_mpeg2_dc_scale_table[s->intra_dc_precision][8];
2927         ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
2928                        s->intra_matrix, s->intra_quant_bias, 8, 8, 1);
2929         s->qscale= 8;
2930     }
2931
2932     //FIXME var duplication
2933     s->current_picture_ptr->key_frame=
2934     s->current_picture.key_frame= s->pict_type == FF_I_TYPE; //FIXME pic_ptr
2935     s->current_picture_ptr->pict_type=
2936     s->current_picture.pict_type= s->pict_type;
2937
2938     if(s->current_picture.key_frame)
2939         s->picture_in_gop_number=0;
2940
2941     s->last_bits= put_bits_count(&s->pb);
2942     switch(s->out_format) {
2943     case FMT_MJPEG:
2944         if (CONFIG_MJPEG_ENCODER)
2945             ff_mjpeg_encode_picture_header(s);
2946         break;
2947     case FMT_H261:
2948         if (CONFIG_H261_ENCODER)
2949             ff_h261_encode_picture_header(s, picture_number);
2950         break;
2951     case FMT_H263:
2952         if (CONFIG_WMV2_ENCODER && s->codec_id == CODEC_ID_WMV2)
2953             ff_wmv2_encode_picture_header(s, picture_number);
2954         else if (CONFIG_MSMPEG4_ENCODER && s->h263_msmpeg4)
2955             msmpeg4_encode_picture_header(s, picture_number);
2956         else if (CONFIG_MPEG4_ENCODER && s->h263_pred)
2957             mpeg4_encode_picture_header(s, picture_number);
2958         else if (CONFIG_RV10_ENCODER && s->codec_id == CODEC_ID_RV10)
2959             rv10_encode_picture_header(s, picture_number);
2960         else if (CONFIG_RV20_ENCODER && s->codec_id == CODEC_ID_RV20)
2961             rv20_encode_picture_header(s, picture_number);
2962         else if (CONFIG_FLV_ENCODER && s->codec_id == CODEC_ID_FLV1)
2963             ff_flv_encode_picture_header(s, picture_number);
2964         else if (CONFIG_H263_ENCODER)
2965             h263_encode_picture_header(s, picture_number);
2966         break;
2967     case FMT_MPEG1:
2968         if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
2969             mpeg1_encode_picture_header(s, picture_number);
2970         break;
2971     case FMT_H264:
2972         break;
2973     default:
2974         assert(0);
2975     }
2976     bits= put_bits_count(&s->pb);
2977     s->header_bits= bits - s->last_bits;
2978
2979     for(i=1; i<context_count; i++){
2980         update_duplicate_context_after_me(s->thread_context[i], s);
2981     }
2982     s->avctx->execute(s->avctx, encode_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
2983     for(i=1; i<context_count; i++){
2984         merge_context_after_encode(s, s->thread_context[i]);
2985     }
2986     emms_c();
2987     return 0;
2988 }
2989
2990 static void denoise_dct_c(MpegEncContext *s, DCTELEM *block){
2991     const int intra= s->mb_intra;
2992     int i;
2993
2994     s->dct_count[intra]++;
2995
2996     for(i=0; i<64; i++){
2997         int level= block[i];
2998
2999         if(level){
3000             if(level>0){
3001                 s->dct_error_sum[intra][i] += level;
3002                 level -= s->dct_offset[intra][i];
3003                 if(level<0) level=0;
3004             }else{
3005                 s->dct_error_sum[intra][i] -= level;
3006                 level += s->dct_offset[intra][i];
3007                 if(level>0) level=0;
3008             }
3009             block[i]= level;
3010         }
3011     }
3012 }
3013
3014 static int dct_quantize_trellis_c(MpegEncContext *s,
3015                                   DCTELEM *block, int n,
3016                                   int qscale, int *overflow){
3017     const int *qmat;
3018     const uint8_t *scantable= s->intra_scantable.scantable;
3019     const uint8_t *perm_scantable= s->intra_scantable.permutated;
3020     int max=0;
3021     unsigned int threshold1, threshold2;
3022     int bias=0;
3023     int run_tab[65];
3024     int level_tab[65];
3025     int score_tab[65];
3026     int survivor[65];
3027     int survivor_count;
3028     int last_run=0;
3029     int last_level=0;
3030     int last_score= 0;
3031     int last_i;
3032     int coeff[2][64];
3033     int coeff_count[64];
3034     int qmul, qadd, start_i, last_non_zero, i, dc;
3035     const int esc_length= s->ac_esc_length;
3036     uint8_t * length;
3037     uint8_t * last_length;
3038     const int lambda= s->lambda2 >> (FF_LAMBDA_SHIFT - 6);
3039
3040     s->dsp.fdct (block);
3041
3042     if(s->dct_error_sum)
3043         s->denoise_dct(s, block);
3044     qmul= qscale*16;
3045     qadd= ((qscale-1)|1)*8;
3046
3047     if (s->mb_intra) {
3048         int q;
3049         if (!s->h263_aic) {
3050             if (n < 4)
3051                 q = s->y_dc_scale;
3052             else
3053                 q = s->c_dc_scale;
3054             q = q << 3;
3055         } else{
3056             /* For AIC we skip quant/dequant of INTRADC */
3057             q = 1 << 3;
3058             qadd=0;
3059         }
3060
3061         /* note: block[0] is assumed to be positive */
3062         block[0] = (block[0] + (q >> 1)) / q;
3063         start_i = 1;
3064         last_non_zero = 0;
3065         qmat = s->q_intra_matrix[qscale];
3066         if(s->mpeg_quant || s->out_format == FMT_MPEG1)
3067             bias= 1<<(QMAT_SHIFT-1);
3068         length     = s->intra_ac_vlc_length;
3069         last_length= s->intra_ac_vlc_last_length;
3070     } else {
3071         start_i = 0;
3072         last_non_zero = -1;
3073         qmat = s->q_inter_matrix[qscale];
3074         length     = s->inter_ac_vlc_length;
3075         last_length= s->inter_ac_vlc_last_length;
3076     }
3077     last_i= start_i;
3078
3079     threshold1= (1<<QMAT_SHIFT) - bias - 1;
3080     threshold2= (threshold1<<1);
3081
3082     for(i=63; i>=start_i; i--) {
3083         const int j = scantable[i];
3084         int level = block[j] * qmat[j];
3085
3086         if(((unsigned)(level+threshold1))>threshold2){
3087             last_non_zero = i;
3088             break;
3089         }
3090     }
3091
3092     for(i=start_i; i<=last_non_zero; i++) {
3093         const int j = scantable[i];
3094         int level = block[j] * qmat[j];
3095
3096 //        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
3097 //           || bias-level >= (1<<(QMAT_SHIFT - 3))){
3098         if(((unsigned)(level+threshold1))>threshold2){
3099             if(level>0){
3100                 level= (bias + level)>>QMAT_SHIFT;
3101                 coeff[0][i]= level;
3102                 coeff[1][i]= level-1;
3103 //                coeff[2][k]= level-2;
3104             }else{
3105                 level= (bias - level)>>QMAT_SHIFT;
3106                 coeff[0][i]= -level;
3107                 coeff[1][i]= -level+1;
3108 //                coeff[2][k]= -level+2;
3109             }
3110             coeff_count[i]= FFMIN(level, 2);
3111             assert(coeff_count[i]);
3112             max |=level;
3113         }else{
3114             coeff[0][i]= (level>>31)|1;
3115             coeff_count[i]= 1;
3116         }
3117     }
3118
3119     *overflow= s->max_qcoeff < max; //overflow might have happened
3120
3121     if(last_non_zero < start_i){
3122         memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
3123         return last_non_zero;
3124     }
3125
3126     score_tab[start_i]= 0;
3127     survivor[0]= start_i;
3128     survivor_count= 1;
3129
3130     for(i=start_i; i<=last_non_zero; i++){
3131         int level_index, j, zero_distortion;
3132         int dct_coeff= FFABS(block[ scantable[i] ]);
3133         int best_score=256*256*256*120;
3134
3135         if (   s->dsp.fdct == fdct_ifast
3136 #ifndef FAAN_POSTSCALE
3137             || s->dsp.fdct == ff_faandct
3138 #endif
3139            )
3140             dct_coeff= (dct_coeff*ff_inv_aanscales[ scantable[i] ]) >> 12;
3141         zero_distortion= dct_coeff*dct_coeff;
3142
3143         for(level_index=0; level_index < coeff_count[i]; level_index++){
3144             int distortion;
3145             int level= coeff[level_index][i];
3146             const int alevel= FFABS(level);
3147             int unquant_coeff;
3148
3149             assert(level);
3150
3151             if(s->out_format == FMT_H263){
3152                 unquant_coeff= alevel*qmul + qadd;
3153             }else{ //MPEG1
3154                 j= s->dsp.idct_permutation[ scantable[i] ]; //FIXME optimize
3155                 if(s->mb_intra){
3156                         unquant_coeff = (int)(  alevel  * qscale * s->intra_matrix[j]) >> 3;
3157                         unquant_coeff =   (unquant_coeff - 1) | 1;
3158                 }else{
3159                         unquant_coeff = (((  alevel  << 1) + 1) * qscale * ((int) s->inter_matrix[j])) >> 4;
3160                         unquant_coeff =   (unquant_coeff - 1) | 1;
3161                 }
3162                 unquant_coeff<<= 3;
3163             }
3164
3165             distortion= (unquant_coeff - dct_coeff) * (unquant_coeff - dct_coeff) - zero_distortion;
3166             level+=64;
3167             if((level&(~127)) == 0){
3168                 for(j=survivor_count-1; j>=0; j--){
3169                     int run= i - survivor[j];
3170                     int score= distortion + length[UNI_AC_ENC_INDEX(run, level)]*lambda;
3171                     score += score_tab[i-run];
3172
3173                     if(score < best_score){
3174                         best_score= score;
3175                         run_tab[i+1]= run;
3176                         level_tab[i+1]= level-64;
3177                     }
3178                 }
3179
3180                 if(s->out_format == FMT_H263){
3181                     for(j=survivor_count-1; j>=0; j--){
3182                         int run= i - survivor[j];
3183                         int score= distortion + last_length[UNI_AC_ENC_INDEX(run, level)]*lambda;
3184                         score += score_tab[i-run];
3185                         if(score < last_score){
3186                             last_score= score;
3187                             last_run= run;
3188                             last_level= level-64;
3189                             last_i= i+1;
3190                         }
3191                     }
3192                 }
3193             }else{
3194                 distortion += esc_length*lambda;
3195                 for(j=survivor_count-1; j>=0; j--){
3196                     int run= i - survivor[j];
3197                     int score= distortion + score_tab[i-run];
3198
3199                     if(score < best_score){
3200                         best_score= score;
3201                         run_tab[i+1]= run;
3202                         level_tab[i+1]= level-64;
3203                     }
3204                 }
3205
3206                 if(s->out_format == FMT_H263){
3207                   for(j=survivor_count-1; j>=0; j--){
3208                         int run= i - survivor[j];
3209                         int score= distortion + score_tab[i-run];
3210                         if(score < last_score){
3211                             last_score= score;
3212                             last_run= run;
3213                             last_level= level-64;
3214                             last_i= i+1;
3215                         }
3216                     }
3217                 }
3218             }
3219         }
3220
3221         score_tab[i+1]= best_score;
3222
3223         //Note: there is a vlc code in mpeg4 which is 1 bit shorter then another one with a shorter run and the same level
3224         if(last_non_zero <= 27){
3225             for(; survivor_count; survivor_count--){
3226                 if(score_tab[ survivor[survivor_count-1] ] <= best_score)
3227                     break;
3228             }
3229         }else{
3230             for(; survivor_count; survivor_count--){
3231                 if(score_tab[ survivor[survivor_count-1] ] <= best_score + lambda)
3232                     break;
3233             }
3234         }
3235
3236         survivor[ survivor_count++ ]= i+1;
3237     }
3238
3239     if(s->out_format != FMT_H263){
3240         last_score= 256*256*256*120;
3241         for(i= survivor[0]; i<=last_non_zero + 1; i++){
3242             int score= score_tab[i];
3243             if(i) score += lambda*2; //FIXME exacter?
3244
3245             if(score < last_score){
3246                 last_score= score;
3247                 last_i= i;
3248                 last_level= level_tab[i];
3249                 last_run= run_tab[i];
3250             }
3251         }
3252     }
3253
3254     s->coded_score[n] = last_score;
3255
3256     dc= FFABS(block[0]);
3257     last_non_zero= last_i - 1;
3258     memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
3259
3260     if(last_non_zero < start_i)
3261         return last_non_zero;
3262
3263     if(last_non_zero == 0 && start_i == 0){
3264         int best_level= 0;
3265         int best_score= dc * dc;
3266
3267         for(i=0; i<coeff_count[0]; i++){
3268             int level= coeff[i][0];
3269             int alevel= FFABS(level);
3270             int unquant_coeff, score, distortion;
3271
3272             if(s->out_format == FMT_H263){
3273                     unquant_coeff= (alevel*qmul + qadd)>>3;
3274             }else{ //MPEG1
3275                     unquant_coeff = (((  alevel  << 1) + 1) * qscale * ((int) s->inter_matrix[0])) >> 4;
3276                     unquant_coeff =   (unquant_coeff - 1) | 1;
3277             }
3278             unquant_coeff = (unquant_coeff + 4) >> 3;
3279             unquant_coeff<<= 3 + 3;
3280
3281             distortion= (unquant_coeff - dc) * (unquant_coeff - dc);
3282             level+=64;
3283             if((level&(~127)) == 0) score= distortion + last_length[UNI_AC_ENC_INDEX(0, level)]*lambda;
3284             else                    score= distortion + esc_length*lambda;
3285
3286             if(score < best_score){
3287                 best_score= score;
3288                 best_level= level - 64;
3289             }
3290         }
3291         block[0]= best_level;
3292         s->coded_score[n] = best_score - dc*dc;
3293         if(best_level == 0) return -1;
3294         else                return last_non_zero;
3295     }
3296
3297     i= last_i;
3298     assert(last_level);
3299
3300     block[ perm_scantable[last_non_zero] ]= last_level;
3301     i -= last_run + 1;
3302
3303     for(; i>start_i; i -= run_tab[i] + 1){
3304         block[ perm_scantable[i-1] ]= level_tab[i];
3305     }
3306
3307     return last_non_zero;
3308 }
3309
3310 //#define REFINE_STATS 1
3311 static int16_t basis[64][64];
3312
3313 static void build_basis(uint8_t *perm){
3314     int i, j, x, y;
3315     emms_c();
3316     for(i=0; i<8; i++){
3317         for(j=0; j<8; j++){
3318             for(y=0; y<8; y++){
3319                 for(x=0; x<8; x++){
3320                     double s= 0.25*(1<<BASIS_SHIFT);
3321                     int index= 8*i + j;
3322                     int perm_index= perm[index];
3323                     if(i==0) s*= sqrt(0.5);
3324                     if(j==0) s*= sqrt(0.5);
3325                     basis[perm_index][8*x + y]= lrintf(s * cos((M_PI/8.0)*i*(x+0.5)) * cos((M_PI/8.0)*j*(y+0.5)));
3326                 }
3327             }
3328         }
3329     }
3330 }
3331
3332 static int dct_quantize_refine(MpegEncContext *s, //FIXME breaks denoise?
3333                         DCTELEM *block, int16_t *weight, DCTELEM *orig,
3334                         int n, int qscale){
3335     int16_t rem[64];
3336     LOCAL_ALIGNED_16(DCTELEM, d1, [64]);
3337     const uint8_t *scantable= s->intra_scantable.scantable;
3338     const uint8_t *perm_scantable= s->intra_scantable.permutated;
3339 //    unsigned int threshold1, threshold2;
3340 //    int bias=0;
3341     int run_tab[65];
3342     int prev_run=0;
3343     int prev_level=0;
3344     int qmul, qadd, start_i, last_non_zero, i, dc;
3345     uint8_t * length;
3346     uint8_t * last_length;
3347     int lambda;
3348     int rle_index, run, q = 1, sum; //q is only used when s->mb_intra is true
3349 #ifdef REFINE_STATS
3350 static int count=0;
3351 static int after_last=0;
3352 static int to_zero=0;
3353 static int from_zero=0;
3354 static int raise=0;
3355 static int lower=0;
3356 static int messed_sign=0;
3357 #endif
3358
3359     if(basis[0][0] == 0)
3360         build_basis(s->dsp.idct_permutation);
3361
3362     qmul= qscale*2;
3363     qadd= (qscale-1)|1;
3364     if (s->mb_intra) {
3365         if (!s->h263_aic) {
3366             if (n < 4)
3367                 q = s->y_dc_scale;
3368             else
3369                 q = s->c_dc_scale;
3370         } else{
3371             /* For AIC we skip quant/dequant of INTRADC */</