042b91d6bcd735a81f7440fb53dfdf1ac9edc0e9
[ffmpeg.git] / libavcodec / mpegvideo_enc.c
1 /*
2  * The simplest mpeg encoder (well, it was the simplest!)
3  * Copyright (c) 2000,2001 Fabrice Bellard
4  * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
5  *
6  * 4MV & hq & B-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
7  *
8  * This file is part of Libav.
9  *
10  * Libav is free software; you can redistribute it and/or
11  * modify it under the terms of the GNU Lesser General Public
12  * License as published by the Free Software Foundation; either
13  * version 2.1 of the License, or (at your option) any later version.
14  *
15  * Libav is distributed in the hope that it will be useful,
16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
18  * Lesser General Public License for more details.
19  *
20  * You should have received a copy of the GNU Lesser General Public
21  * License along with Libav; if not, write to the Free Software
22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23  */
24
25 /**
26  * @file
27  * The simplest mpeg encoder (well, it was the simplest!).
28  */
29
30 #include "libavutil/intmath.h"
31 #include "libavutil/mathematics.h"
32 #include "avcodec.h"
33 #include "dsputil.h"
34 #include "mpegvideo.h"
35 #include "mpegvideo_common.h"
36 #include "h263.h"
37 #include "mjpegenc.h"
38 #include "msmpeg4.h"
39 #include "faandct.h"
40 #include "thread.h"
41 #include "aandcttab.h"
42 #include "flv.h"
43 #include "mpeg4video.h"
44 #include "internal.h"
45 #include <limits.h>
46
47 //#undef NDEBUG
48 //#include <assert.h>
49
50 static int encode_picture(MpegEncContext *s, int picture_number);
51 static int dct_quantize_refine(MpegEncContext *s, DCTELEM *block, int16_t *weight, DCTELEM *orig, int n, int qscale);
52 static int sse_mb(MpegEncContext *s);
53 static void denoise_dct_c(MpegEncContext *s, DCTELEM *block);
54 static int dct_quantize_trellis_c(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow);
55
56 /* enable all paranoid tests for rounding, overflows, etc... */
57 //#define PARANOID
58
59 //#define DEBUG
60
61 static uint8_t default_mv_penalty[MAX_FCODE+1][MAX_MV*2+1];
62 static uint8_t default_fcode_tab[MAX_MV*2+1];
63
64 void ff_convert_matrix(DSPContext *dsp, int (*qmat)[64], uint16_t (*qmat16)[2][64],
65                            const uint16_t *quant_matrix, int bias, int qmin, int qmax, int intra)
66 {
67     int qscale;
68     int shift=0;
69
70     for(qscale=qmin; qscale<=qmax; qscale++){
71         int i;
72         if (dsp->fdct == ff_jpeg_fdct_islow_8 ||
73             dsp->fdct == ff_jpeg_fdct_islow_10
74 #ifdef FAAN_POSTSCALE
75             || dsp->fdct == ff_faandct
76 #endif
77             ) {
78             for(i=0;i<64;i++) {
79                 const int j= dsp->idct_permutation[i];
80                 /* 16 <= qscale * quant_matrix[i] <= 7905 */
81                 /* 19952             <= ff_aanscales[i] * qscale * quant_matrix[i]               <= 249205026 */
82                 /* (1 << 36) / 19952 >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= (1 << 36) / 249205026 */
83                 /* 3444240           >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= 275 */
84
85                 qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) /
86                                 (qscale * quant_matrix[j]));
87             }
88         } else if (dsp->fdct == fdct_ifast
89 #ifndef FAAN_POSTSCALE
90                    || dsp->fdct == ff_faandct
91 #endif
92                    ) {
93             for(i=0;i<64;i++) {
94                 const int j= dsp->idct_permutation[i];
95                 /* 16 <= qscale * quant_matrix[i] <= 7905 */
96                 /* 19952             <= ff_aanscales[i] * qscale * quant_matrix[i]               <= 249205026 */
97                 /* (1 << 36) / 19952 >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */
98                 /* 3444240           >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= 275 */
99
100                 qmat[qscale][i] = (int)((UINT64_C(1) << (QMAT_SHIFT + 14)) /
101                                 (ff_aanscales[i] * qscale * quant_matrix[j]));
102             }
103         } else {
104             for(i=0;i<64;i++) {
105                 const int j= dsp->idct_permutation[i];
106                 /* We can safely suppose that 16 <= quant_matrix[i] <= 255
107                    So 16           <= qscale * quant_matrix[i]             <= 7905
108                    so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905
109                    so 32768        >= (1<<19) / (qscale * quant_matrix[i]) >= 67
110                 */
111                 qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) / (qscale * quant_matrix[j]));
112 //                qmat  [qscale][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]);
113                 qmat16[qscale][0][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[j]);
114
115                 if(qmat16[qscale][0][i]==0 || qmat16[qscale][0][i]==128*256) qmat16[qscale][0][i]=128*256-1;
116                 qmat16[qscale][1][i]= ROUNDED_DIV(bias<<(16-QUANT_BIAS_SHIFT), qmat16[qscale][0][i]);
117             }
118         }
119
120         for(i=intra; i<64; i++){
121             int64_t max= 8191;
122             if (dsp->fdct == fdct_ifast
123 #ifndef FAAN_POSTSCALE
124                    || dsp->fdct == ff_faandct
125 #endif
126                    ) {
127                 max = (8191LL*ff_aanscales[i]) >> 14;
128             }
129             while(((max * qmat[qscale][i]) >> shift) > INT_MAX){
130                 shift++;
131             }
132         }
133     }
134     if(shift){
135         av_log(NULL, AV_LOG_INFO, "Warning, QMAT_SHIFT is larger than %d, overflows possible\n", QMAT_SHIFT - shift);
136     }
137 }
138
139 static inline void update_qscale(MpegEncContext *s){
140     s->qscale= (s->lambda*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7);
141     s->qscale= av_clip(s->qscale, s->avctx->qmin, s->avctx->qmax);
142
143     s->lambda2= (s->lambda*s->lambda + FF_LAMBDA_SCALE/2) >> FF_LAMBDA_SHIFT;
144 }
145
146 void ff_write_quant_matrix(PutBitContext *pb, uint16_t *matrix){
147     int i;
148
149     if(matrix){
150         put_bits(pb, 1, 1);
151         for(i=0;i<64;i++) {
152             put_bits(pb, 8, matrix[ ff_zigzag_direct[i] ]);
153         }
154     }else
155         put_bits(pb, 1, 0);
156 }
157
158 /**
159  * init s->current_picture.qscale_table from s->lambda_table
160  */
161 void ff_init_qscale_tab(MpegEncContext *s){
162     int8_t * const qscale_table = s->current_picture.f.qscale_table;
163     int i;
164
165     for(i=0; i<s->mb_num; i++){
166         unsigned int lam= s->lambda_table[ s->mb_index2xy[i] ];
167         int qp= (lam*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7);
168         qscale_table[ s->mb_index2xy[i] ]= av_clip(qp, s->avctx->qmin, s->avctx->qmax);
169     }
170 }
171
172 static void copy_picture_attributes(MpegEncContext *s, AVFrame *dst, AVFrame *src){
173     int i;
174
175     dst->pict_type              = src->pict_type;
176     dst->quality                = src->quality;
177     dst->coded_picture_number   = src->coded_picture_number;
178     dst->display_picture_number = src->display_picture_number;
179 //    dst->reference              = src->reference;
180     dst->pts                    = src->pts;
181     dst->interlaced_frame       = src->interlaced_frame;
182     dst->top_field_first        = src->top_field_first;
183
184     if(s->avctx->me_threshold){
185         if(!src->motion_val[0])
186             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_val not set!\n");
187         if(!src->mb_type)
188             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.mb_type not set!\n");
189         if(!src->ref_index[0])
190             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.ref_index not set!\n");
191         if(src->motion_subsample_log2 != dst->motion_subsample_log2)
192             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_subsample_log2 doesn't match! (%d!=%d)\n",
193             src->motion_subsample_log2, dst->motion_subsample_log2);
194
195         memcpy(dst->mb_type, src->mb_type, s->mb_stride * s->mb_height * sizeof(dst->mb_type[0]));
196
197         for(i=0; i<2; i++){
198             int stride= ((16*s->mb_width )>>src->motion_subsample_log2) + 1;
199             int height= ((16*s->mb_height)>>src->motion_subsample_log2);
200
201             if(src->motion_val[i] && src->motion_val[i] != dst->motion_val[i]){
202                 memcpy(dst->motion_val[i], src->motion_val[i], 2*stride*height*sizeof(int16_t));
203             }
204             if(src->ref_index[i] && src->ref_index[i] != dst->ref_index[i]){
205                 memcpy(dst->ref_index[i], src->ref_index[i], s->mb_stride*4*s->mb_height*sizeof(int8_t));
206             }
207         }
208     }
209 }
210
211 static void update_duplicate_context_after_me(MpegEncContext *dst, MpegEncContext *src){
212 #define COPY(a) dst->a= src->a
213     COPY(pict_type);
214     COPY(current_picture);
215     COPY(f_code);
216     COPY(b_code);
217     COPY(qscale);
218     COPY(lambda);
219     COPY(lambda2);
220     COPY(picture_in_gop_number);
221     COPY(gop_picture_number);
222     COPY(frame_pred_frame_dct); //FIXME don't set in encode_header
223     COPY(progressive_frame); //FIXME don't set in encode_header
224     COPY(partitioned_frame); //FIXME don't set in encode_header
225 #undef COPY
226 }
227
228 /**
229  * sets the given MpegEncContext to defaults for encoding.
230  * the changed fields will not depend upon the prior state of the MpegEncContext.
231  */
232 static void MPV_encode_defaults(MpegEncContext *s){
233     int i;
234     MPV_common_defaults(s);
235
236     for(i=-16; i<16; i++){
237         default_fcode_tab[i + MAX_MV]= 1;
238     }
239     s->me.mv_penalty= default_mv_penalty;
240     s->fcode_tab= default_fcode_tab;
241 }
242
243 /* init video encoder */
244 av_cold int MPV_encode_init(AVCodecContext *avctx)
245 {
246     MpegEncContext *s = avctx->priv_data;
247     int i;
248     int chroma_h_shift, chroma_v_shift;
249
250     MPV_encode_defaults(s);
251
252     switch (avctx->codec_id) {
253     case CODEC_ID_MPEG2VIDEO:
254         if(avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P){
255             av_log(avctx, AV_LOG_ERROR, "only YUV420 and YUV422 are supported\n");
256             return -1;
257         }
258         break;
259     case CODEC_ID_LJPEG:
260         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 &&
261            ((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)){
262             av_log(avctx, AV_LOG_ERROR, "colorspace not supported in LJPEG\n");
263             return -1;
264         }
265         break;
266     case CODEC_ID_MJPEG:
267         if(avctx->pix_fmt != PIX_FMT_YUVJ420P && avctx->pix_fmt != PIX_FMT_YUVJ422P &&
268            ((avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P) || avctx->strict_std_compliance>FF_COMPLIANCE_UNOFFICIAL)){
269             av_log(avctx, AV_LOG_ERROR, "colorspace not supported in jpeg\n");
270             return -1;
271         }
272         break;
273     default:
274         if(avctx->pix_fmt != PIX_FMT_YUV420P){
275             av_log(avctx, AV_LOG_ERROR, "only YUV420 is supported\n");
276             return -1;
277         }
278     }
279
280     switch (avctx->pix_fmt) {
281     case PIX_FMT_YUVJ422P:
282     case PIX_FMT_YUV422P:
283         s->chroma_format = CHROMA_422;
284         break;
285     case PIX_FMT_YUVJ420P:
286     case PIX_FMT_YUV420P:
287     default:
288         s->chroma_format = CHROMA_420;
289         break;
290     }
291
292     s->bit_rate = avctx->bit_rate;
293     s->width = avctx->width;
294     s->height = avctx->height;
295     if(avctx->gop_size > 600 && avctx->strict_std_compliance>FF_COMPLIANCE_EXPERIMENTAL){
296         av_log(avctx, AV_LOG_ERROR, "Warning keyframe interval too large! reducing it ...\n");
297         avctx->gop_size=600;
298     }
299     s->gop_size = avctx->gop_size;
300     s->avctx = avctx;
301     s->flags= avctx->flags;
302     s->flags2= avctx->flags2;
303     s->max_b_frames= avctx->max_b_frames;
304     s->codec_id= avctx->codec->id;
305     s->luma_elim_threshold  = avctx->luma_elim_threshold;
306     s->chroma_elim_threshold= avctx->chroma_elim_threshold;
307     s->strict_std_compliance= avctx->strict_std_compliance;
308 #if FF_API_MPEGVIDEO_GLOBAL_OPTS
309     if (avctx->flags & CODEC_FLAG_PART)
310         s->data_partitioning = 1;
311 #endif
312     s->quarter_sample= (avctx->flags & CODEC_FLAG_QPEL)!=0;
313     s->mpeg_quant= avctx->mpeg_quant;
314     s->rtp_mode= !!avctx->rtp_payload_size;
315     s->intra_dc_precision= avctx->intra_dc_precision;
316     s->user_specified_pts = AV_NOPTS_VALUE;
317
318     if (s->gop_size <= 1) {
319         s->intra_only = 1;
320         s->gop_size = 12;
321     } else {
322         s->intra_only = 0;
323     }
324
325     s->me_method = avctx->me_method;
326
327     /* Fixed QSCALE */
328     s->fixed_qscale = !!(avctx->flags & CODEC_FLAG_QSCALE);
329
330     s->adaptive_quant= (   s->avctx->lumi_masking
331                         || s->avctx->dark_masking
332                         || s->avctx->temporal_cplx_masking
333                         || s->avctx->spatial_cplx_masking
334                         || s->avctx->p_masking
335                         || s->avctx->border_masking
336                         || (s->flags&CODEC_FLAG_QP_RD))
337                        && !s->fixed_qscale;
338
339     s->obmc= !!(s->flags & CODEC_FLAG_OBMC);
340     s->loop_filter= !!(s->flags & CODEC_FLAG_LOOP_FILTER);
341 #if FF_API_MPEGVIDEO_GLOBAL_OPTS
342     s->alternate_scan= !!(s->flags & CODEC_FLAG_ALT_SCAN);
343     s->intra_vlc_format= !!(s->flags2 & CODEC_FLAG2_INTRA_VLC);
344     s->q_scale_type= !!(s->flags2 & CODEC_FLAG2_NON_LINEAR_QUANT);
345 #endif
346
347     if(avctx->rc_max_rate && !avctx->rc_buffer_size){
348         av_log(avctx, AV_LOG_ERROR, "a vbv buffer size is needed, for encoding with a maximum bitrate\n");
349         return -1;
350     }
351
352     if(avctx->rc_min_rate && avctx->rc_max_rate != avctx->rc_min_rate){
353         av_log(avctx, AV_LOG_INFO, "Warning min_rate > 0 but min_rate != max_rate isn't recommended!\n");
354     }
355
356     if(avctx->rc_min_rate && avctx->rc_min_rate > avctx->bit_rate){
357         av_log(avctx, AV_LOG_ERROR, "bitrate below min bitrate\n");
358         return -1;
359     }
360
361     if(avctx->rc_max_rate && avctx->rc_max_rate < avctx->bit_rate){
362         av_log(avctx, AV_LOG_INFO, "bitrate above max bitrate\n");
363         return -1;
364     }
365
366     if(avctx->rc_max_rate && avctx->rc_max_rate == avctx->bit_rate && avctx->rc_max_rate != avctx->rc_min_rate){
367         av_log(avctx, AV_LOG_INFO, "impossible bitrate constraints, this will fail\n");
368     }
369
370     if(avctx->rc_buffer_size && avctx->bit_rate*(int64_t)avctx->time_base.num > avctx->rc_buffer_size * (int64_t)avctx->time_base.den){
371         av_log(avctx, AV_LOG_ERROR, "VBV buffer too small for bitrate\n");
372         return -1;
373     }
374
375     if(!s->fixed_qscale && avctx->bit_rate*av_q2d(avctx->time_base) > avctx->bit_rate_tolerance){
376         av_log(avctx, AV_LOG_ERROR, "bitrate tolerance too small for bitrate\n");
377         return -1;
378     }
379
380     if(   s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate
381        && (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO)
382        && 90000LL * (avctx->rc_buffer_size-1) > s->avctx->rc_max_rate*0xFFFFLL){
383
384         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");
385     }
386
387     if((s->flags & CODEC_FLAG_4MV) && s->codec_id != CODEC_ID_MPEG4
388        && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P && s->codec_id != CODEC_ID_FLV1){
389         av_log(avctx, AV_LOG_ERROR, "4MV not supported by codec\n");
390         return -1;
391     }
392
393     if(s->obmc && s->avctx->mb_decision != FF_MB_DECISION_SIMPLE){
394         av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with simple mb decision\n");
395         return -1;
396     }
397
398     if(s->obmc && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P){
399         av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with H263(+)\n");
400         return -1;
401     }
402
403     if(s->quarter_sample && s->codec_id != CODEC_ID_MPEG4){
404         av_log(avctx, AV_LOG_ERROR, "qpel not supported by codec\n");
405         return -1;
406     }
407
408 #if FF_API_MPEGVIDEO_GLOBAL_OPTS
409     if(s->data_partitioning && s->codec_id != CODEC_ID_MPEG4){
410         av_log(avctx, AV_LOG_ERROR, "data partitioning not supported by codec\n");
411         return -1;
412     }
413 #endif
414
415     if(s->max_b_frames && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO){
416         av_log(avctx, AV_LOG_ERROR, "b frames not supported by codec\n");
417         return -1;
418     }
419
420     if ((s->codec_id == CODEC_ID_MPEG4 || s->codec_id == CODEC_ID_H263 ||
421          s->codec_id == CODEC_ID_H263P) &&
422         (avctx->sample_aspect_ratio.num > 255 || avctx->sample_aspect_ratio.den > 255)) {
423         av_log(avctx, AV_LOG_ERROR, "Invalid pixel aspect ratio %i/%i, limit is 255/255\n",
424                avctx->sample_aspect_ratio.num, avctx->sample_aspect_ratio.den);
425         return -1;
426     }
427
428     if((s->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN))
429        && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG2VIDEO){
430         av_log(avctx, AV_LOG_ERROR, "interlacing not supported by codec\n");
431         return -1;
432     }
433
434     if(s->mpeg_quant && s->codec_id != CODEC_ID_MPEG4){ //FIXME mpeg2 uses that too
435         av_log(avctx, AV_LOG_ERROR, "mpeg2 style quantization not supported by codec\n");
436         return -1;
437     }
438
439     if((s->flags & CODEC_FLAG_CBP_RD) && !avctx->trellis){
440         av_log(avctx, AV_LOG_ERROR, "CBP RD needs trellis quant\n");
441         return -1;
442     }
443
444     if((s->flags & CODEC_FLAG_QP_RD) && s->avctx->mb_decision != FF_MB_DECISION_RD){
445         av_log(avctx, AV_LOG_ERROR, "QP RD needs mbd=2\n");
446         return -1;
447     }
448
449     if(s->avctx->scenechange_threshold < 1000000000 && (s->flags & CODEC_FLAG_CLOSED_GOP)){
450         av_log(avctx, AV_LOG_ERROR, "closed gop with scene change detection are not supported yet, set threshold to 1000000000\n");
451         return -1;
452     }
453
454     if((s->flags2 & CODEC_FLAG2_INTRA_VLC) && s->codec_id != CODEC_ID_MPEG2VIDEO){
455         av_log(avctx, AV_LOG_ERROR, "intra vlc table not supported by codec\n");
456         return -1;
457     }
458
459     if(s->flags & CODEC_FLAG_LOW_DELAY){
460         if (s->codec_id != CODEC_ID_MPEG2VIDEO){
461             av_log(avctx, AV_LOG_ERROR, "low delay forcing is only available for mpeg2\n");
462             return -1;
463         }
464         if (s->max_b_frames != 0){
465             av_log(avctx, AV_LOG_ERROR, "b frames cannot be used with low delay\n");
466             return -1;
467         }
468     }
469
470     if(s->q_scale_type == 1){
471 #if FF_API_MPEGVIDEO_GLOBAL_OPTS
472         if(s->codec_id != CODEC_ID_MPEG2VIDEO){
473             av_log(avctx, AV_LOG_ERROR, "non linear quant is only available for mpeg2\n");
474             return -1;
475         }
476 #endif
477         if(avctx->qmax > 12){
478             av_log(avctx, AV_LOG_ERROR, "non linear quant only supports qmax <= 12 currently\n");
479             return -1;
480         }
481     }
482
483     if(s->avctx->thread_count > 1 && s->codec_id != CODEC_ID_MPEG4
484        && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO
485        && (s->codec_id != CODEC_ID_H263P || !(s->flags & CODEC_FLAG_H263P_SLICE_STRUCT))){
486         av_log(avctx, AV_LOG_ERROR, "multi threaded encoding not supported by codec\n");
487         return -1;
488     }
489
490     if(s->avctx->thread_count < 1){
491         av_log(avctx, AV_LOG_ERROR, "automatic thread number detection not supported by codec, patch welcome\n");
492         return -1;
493     }
494
495     if(s->avctx->thread_count > 1)
496         s->rtp_mode= 1;
497
498     if(!avctx->time_base.den || !avctx->time_base.num){
499         av_log(avctx, AV_LOG_ERROR, "framerate not set\n");
500         return -1;
501     }
502
503     i= (INT_MAX/2+128)>>8;
504     if(avctx->me_threshold >= i){
505         av_log(avctx, AV_LOG_ERROR, "me_threshold too large, max is %d\n", i - 1);
506         return -1;
507     }
508     if(avctx->mb_threshold >= i){
509         av_log(avctx, AV_LOG_ERROR, "mb_threshold too large, max is %d\n", i - 1);
510         return -1;
511     }
512
513     if(avctx->b_frame_strategy && (avctx->flags&CODEC_FLAG_PASS2)){
514         av_log(avctx, AV_LOG_INFO, "notice: b_frame_strategy only affects the first pass\n");
515         avctx->b_frame_strategy = 0;
516     }
517
518     i= av_gcd(avctx->time_base.den, avctx->time_base.num);
519     if(i > 1){
520         av_log(avctx, AV_LOG_INFO, "removing common factors from framerate\n");
521         avctx->time_base.den /= i;
522         avctx->time_base.num /= i;
523 //        return -1;
524     }
525
526     if(s->mpeg_quant || s->codec_id==CODEC_ID_MPEG1VIDEO || s->codec_id==CODEC_ID_MPEG2VIDEO || s->codec_id==CODEC_ID_MJPEG){
527         s->intra_quant_bias= 3<<(QUANT_BIAS_SHIFT-3); //(a + x*3/8)/x
528         s->inter_quant_bias= 0;
529     }else{
530         s->intra_quant_bias=0;
531         s->inter_quant_bias=-(1<<(QUANT_BIAS_SHIFT-2)); //(a - x/4)/x
532     }
533
534     if(avctx->intra_quant_bias != FF_DEFAULT_QUANT_BIAS)
535         s->intra_quant_bias= avctx->intra_quant_bias;
536     if(avctx->inter_quant_bias != FF_DEFAULT_QUANT_BIAS)
537         s->inter_quant_bias= avctx->inter_quant_bias;
538
539     avcodec_get_chroma_sub_sample(avctx->pix_fmt, &chroma_h_shift, &chroma_v_shift);
540
541     if(avctx->codec_id == CODEC_ID_MPEG4 && s->avctx->time_base.den > (1<<16)-1){
542         av_log(avctx, AV_LOG_ERROR, "timebase %d/%d not supported by MPEG 4 standard, "
543                "the maximum admitted value for the timebase denominator is %d\n",
544                s->avctx->time_base.num, s->avctx->time_base.den, (1<<16)-1);
545         return -1;
546     }
547     s->time_increment_bits = av_log2(s->avctx->time_base.den - 1) + 1;
548
549     switch(avctx->codec->id) {
550     case CODEC_ID_MPEG1VIDEO:
551         s->out_format = FMT_MPEG1;
552         s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY);
553         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
554         break;
555     case CODEC_ID_MPEG2VIDEO:
556         s->out_format = FMT_MPEG1;
557         s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY);
558         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
559         s->rtp_mode= 1;
560         break;
561     case CODEC_ID_LJPEG:
562     case CODEC_ID_MJPEG:
563         s->out_format = FMT_MJPEG;
564         s->intra_only = 1; /* force intra only for jpeg */
565         if(avctx->codec->id == CODEC_ID_LJPEG && avctx->pix_fmt == PIX_FMT_BGRA){
566             s->mjpeg_vsample[0] = s->mjpeg_hsample[0] =
567             s->mjpeg_vsample[1] = s->mjpeg_hsample[1] =
568             s->mjpeg_vsample[2] = s->mjpeg_hsample[2] = 1;
569         }else{
570             s->mjpeg_vsample[0] = 2;
571             s->mjpeg_vsample[1] = 2>>chroma_v_shift;
572             s->mjpeg_vsample[2] = 2>>chroma_v_shift;
573             s->mjpeg_hsample[0] = 2;
574             s->mjpeg_hsample[1] = 2>>chroma_h_shift;
575             s->mjpeg_hsample[2] = 2>>chroma_h_shift;
576         }
577         if (!(CONFIG_MJPEG_ENCODER || CONFIG_LJPEG_ENCODER)
578             || ff_mjpeg_encode_init(s) < 0)
579             return -1;
580         avctx->delay=0;
581         s->low_delay=1;
582         break;
583     case CODEC_ID_H261:
584         if (!CONFIG_H261_ENCODER)  return -1;
585         if (ff_h261_get_picture_format(s->width, s->height) < 0) {
586             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);
587             return -1;
588         }
589         s->out_format = FMT_H261;
590         avctx->delay=0;
591         s->low_delay=1;
592         break;
593     case CODEC_ID_H263:
594         if (!CONFIG_H263_ENCODER)  return -1;
595         if (ff_match_2uint16(h263_format, FF_ARRAY_ELEMS(h263_format), s->width, s->height) == 8) {
596             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);
597             return -1;
598         }
599         s->out_format = FMT_H263;
600         s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
601         avctx->delay=0;
602         s->low_delay=1;
603         break;
604     case CODEC_ID_H263P:
605         s->out_format = FMT_H263;
606         s->h263_plus = 1;
607         /* Fx */
608         s->umvplus = (avctx->flags & CODEC_FLAG_H263P_UMV) ? 1:0;
609         s->h263_aic= (avctx->flags & CODEC_FLAG_AC_PRED) ? 1:0;
610         s->modified_quant= s->h263_aic;
611         s->alt_inter_vlc= (avctx->flags & CODEC_FLAG_H263P_AIV) ? 1:0;
612         s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
613         s->loop_filter= (avctx->flags & CODEC_FLAG_LOOP_FILTER) ? 1:0;
614         s->unrestricted_mv= s->obmc || s->loop_filter || s->umvplus;
615         s->h263_slice_structured= (s->flags & CODEC_FLAG_H263P_SLICE_STRUCT) ? 1:0;
616
617         /* /Fx */
618         /* These are just to be sure */
619         avctx->delay=0;
620         s->low_delay=1;
621         break;
622     case CODEC_ID_FLV1:
623         s->out_format = FMT_H263;
624         s->h263_flv = 2; /* format = 1; 11-bit codes */
625         s->unrestricted_mv = 1;
626         s->rtp_mode=0; /* don't allow GOB */
627         avctx->delay=0;
628         s->low_delay=1;
629         break;
630     case CODEC_ID_RV10:
631         s->out_format = FMT_H263;
632         avctx->delay=0;
633         s->low_delay=1;
634         break;
635     case CODEC_ID_RV20:
636         s->out_format = FMT_H263;
637         avctx->delay=0;
638         s->low_delay=1;
639         s->modified_quant=1;
640         s->h263_aic=1;
641         s->h263_plus=1;
642         s->loop_filter=1;
643         s->unrestricted_mv= 0;
644         break;
645     case CODEC_ID_MPEG4:
646         s->out_format = FMT_H263;
647         s->h263_pred = 1;
648         s->unrestricted_mv = 1;
649         s->low_delay= s->max_b_frames ? 0 : 1;
650         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
651         break;
652     case CODEC_ID_MSMPEG4V2:
653         s->out_format = FMT_H263;
654         s->h263_pred = 1;
655         s->unrestricted_mv = 1;
656         s->msmpeg4_version= 2;
657         avctx->delay=0;
658         s->low_delay=1;
659         break;
660     case CODEC_ID_MSMPEG4V3:
661         s->out_format = FMT_H263;
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_pred = 1;
672         s->unrestricted_mv = 1;
673         s->msmpeg4_version= 4;
674         s->flipflop_rounding=1;
675         avctx->delay=0;
676         s->low_delay=1;
677         break;
678     case CODEC_ID_WMV2:
679         s->out_format = FMT_H263;
680         s->h263_pred = 1;
681         s->unrestricted_mv = 1;
682         s->msmpeg4_version= 5;
683         s->flipflop_rounding=1;
684         avctx->delay=0;
685         s->low_delay=1;
686         break;
687     default:
688         return -1;
689     }
690
691     avctx->has_b_frames= !s->low_delay;
692
693     s->encoding = 1;
694
695     s->progressive_frame=
696     s->progressive_sequence= !(avctx->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN));
697
698     /* init */
699     if (MPV_common_init(s) < 0)
700         return -1;
701
702     if(!s->dct_quantize)
703         s->dct_quantize = dct_quantize_c;
704     if(!s->denoise_dct)
705         s->denoise_dct = denoise_dct_c;
706     s->fast_dct_quantize = s->dct_quantize;
707     if(avctx->trellis)
708         s->dct_quantize = dct_quantize_trellis_c;
709
710     if((CONFIG_H263P_ENCODER || CONFIG_RV20_ENCODER) && s->modified_quant)
711         s->chroma_qscale_table= ff_h263_chroma_qscale_table;
712
713     s->quant_precision=5;
714
715     ff_set_cmp(&s->dsp, s->dsp.ildct_cmp, s->avctx->ildct_cmp);
716     ff_set_cmp(&s->dsp, s->dsp.frame_skip_cmp, s->avctx->frame_skip_cmp);
717
718     if (CONFIG_H261_ENCODER && s->out_format == FMT_H261)
719         ff_h261_encode_init(s);
720     if (CONFIG_H263_ENCODER && s->out_format == FMT_H263)
721         h263_encode_init(s);
722     if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version)
723         ff_msmpeg4_encode_init(s);
724     if ((CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
725         && s->out_format == FMT_MPEG1)
726         ff_mpeg1_encode_init(s);
727
728     /* init q matrix */
729     for(i=0;i<64;i++) {
730         int j= s->dsp.idct_permutation[i];
731         if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4 && s->mpeg_quant){
732             s->intra_matrix[j] = ff_mpeg4_default_intra_matrix[i];
733             s->inter_matrix[j] = ff_mpeg4_default_non_intra_matrix[i];
734         }else if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
735             s->intra_matrix[j] =
736             s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
737         }else
738         { /* mpeg1/2 */
739             s->intra_matrix[j] = ff_mpeg1_default_intra_matrix[i];
740             s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
741         }
742         if(s->avctx->intra_matrix)
743             s->intra_matrix[j] = s->avctx->intra_matrix[i];
744         if(s->avctx->inter_matrix)
745             s->inter_matrix[j] = s->avctx->inter_matrix[i];
746     }
747
748     /* precompute matrix */
749     /* for mjpeg, we do include qscale in the matrix */
750     if (s->out_format != FMT_MJPEG) {
751         ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
752                        s->intra_matrix, s->intra_quant_bias, avctx->qmin, 31, 1);
753         ff_convert_matrix(&s->dsp, s->q_inter_matrix, s->q_inter_matrix16,
754                        s->inter_matrix, s->inter_quant_bias, avctx->qmin, 31, 0);
755     }
756
757     if(ff_rate_control_init(s) < 0)
758         return -1;
759
760     return 0;
761 }
762
763 av_cold int MPV_encode_end(AVCodecContext *avctx)
764 {
765     MpegEncContext *s = avctx->priv_data;
766
767     ff_rate_control_uninit(s);
768
769     MPV_common_end(s);
770     if ((CONFIG_MJPEG_ENCODER || CONFIG_LJPEG_ENCODER) && s->out_format == FMT_MJPEG)
771         ff_mjpeg_encode_close(s);
772
773     av_freep(&avctx->extradata);
774
775     return 0;
776 }
777
778 static int get_sae(uint8_t *src, int ref, int stride){
779     int x,y;
780     int acc=0;
781
782     for(y=0; y<16; y++){
783         for(x=0; x<16; x++){
784             acc+= FFABS(src[x+y*stride] - ref);
785         }
786     }
787
788     return acc;
789 }
790
791 static int get_intra_count(MpegEncContext *s, uint8_t *src, uint8_t *ref, int stride){
792     int x, y, w, h;
793     int acc=0;
794
795     w= s->width &~15;
796     h= s->height&~15;
797
798     for(y=0; y<h; y+=16){
799         for(x=0; x<w; x+=16){
800             int offset= x + y*stride;
801             int sad = s->dsp.sad[0](NULL, src + offset, ref + offset, stride, 16);
802             int mean= (s->dsp.pix_sum(src + offset, stride) + 128)>>8;
803             int sae = get_sae(src + offset, mean, stride);
804
805             acc+= sae + 500 < sad;
806         }
807     }
808     return acc;
809 }
810
811
812 static int load_input_picture(MpegEncContext *s, AVFrame *pic_arg){
813     AVFrame *pic=NULL;
814     int64_t pts;
815     int i;
816     const int encoding_delay= s->max_b_frames;
817     int direct=1;
818
819     if(pic_arg){
820         pts= pic_arg->pts;
821         pic_arg->display_picture_number= s->input_picture_number++;
822
823         if(pts != AV_NOPTS_VALUE){
824             if(s->user_specified_pts != AV_NOPTS_VALUE){
825                 int64_t time= pts;
826                 int64_t last= s->user_specified_pts;
827
828                 if(time <= last){
829                     av_log(s->avctx, AV_LOG_ERROR, "Error, Invalid timestamp=%"PRId64", last=%"PRId64"\n", pts, s->user_specified_pts);
830                     return -1;
831                 }
832             }
833             s->user_specified_pts= pts;
834         }else{
835             if(s->user_specified_pts != AV_NOPTS_VALUE){
836                 s->user_specified_pts=
837                 pts= s->user_specified_pts + 1;
838                 av_log(s->avctx, AV_LOG_INFO, "Warning: AVFrame.pts=? trying to guess (%"PRId64")\n", pts);
839             }else{
840                 pts= pic_arg->display_picture_number;
841             }
842         }
843     }
844
845   if(pic_arg){
846     if(encoding_delay && !(s->flags&CODEC_FLAG_INPUT_PRESERVED)) direct=0;
847     if(pic_arg->linesize[0] != s->linesize) direct=0;
848     if(pic_arg->linesize[1] != s->uvlinesize) direct=0;
849     if(pic_arg->linesize[2] != s->uvlinesize) direct=0;
850
851 //    av_log(AV_LOG_DEBUG, "%d %d %d %d\n",pic_arg->linesize[0], pic_arg->linesize[1], s->linesize, s->uvlinesize);
852
853     if(direct){
854         i= ff_find_unused_picture(s, 1);
855
856         pic= (AVFrame*)&s->picture[i];
857         pic->reference= 3;
858
859         for(i=0; i<4; i++){
860             pic->data[i]= pic_arg->data[i];
861             pic->linesize[i]= pic_arg->linesize[i];
862         }
863         if(ff_alloc_picture(s, (Picture*)pic, 1) < 0){
864             return -1;
865         }
866     }else{
867         i= ff_find_unused_picture(s, 0);
868
869         pic= (AVFrame*)&s->picture[i];
870         pic->reference= 3;
871
872         if(ff_alloc_picture(s, (Picture*)pic, 0) < 0){
873             return -1;
874         }
875
876         if(   pic->data[0] + INPLACE_OFFSET == pic_arg->data[0]
877            && pic->data[1] + INPLACE_OFFSET == pic_arg->data[1]
878            && pic->data[2] + INPLACE_OFFSET == pic_arg->data[2]){
879        // empty
880         }else{
881             int h_chroma_shift, v_chroma_shift;
882             avcodec_get_chroma_sub_sample(s->avctx->pix_fmt, &h_chroma_shift, &v_chroma_shift);
883
884             for(i=0; i<3; i++){
885                 int src_stride= pic_arg->linesize[i];
886                 int dst_stride= i ? s->uvlinesize : s->linesize;
887                 int h_shift= i ? h_chroma_shift : 0;
888                 int v_shift= i ? v_chroma_shift : 0;
889                 int w= s->width >>h_shift;
890                 int h= s->height>>v_shift;
891                 uint8_t *src= pic_arg->data[i];
892                 uint8_t *dst= pic->data[i];
893
894                 if(!s->avctx->rc_buffer_size)
895                     dst +=INPLACE_OFFSET;
896
897                 if(src_stride==dst_stride)
898                     memcpy(dst, src, src_stride*h);
899                 else{
900                     while(h--){
901                         memcpy(dst, src, w);
902                         dst += dst_stride;
903                         src += src_stride;
904                     }
905                 }
906             }
907         }
908     }
909     copy_picture_attributes(s, pic, pic_arg);
910     pic->pts= pts; //we set this here to avoid modifiying pic_arg
911   }
912
913     /* shift buffer entries */
914     for(i=1; i<MAX_PICTURE_COUNT /*s->encoding_delay+1*/; i++)
915         s->input_picture[i-1]= s->input_picture[i];
916
917     s->input_picture[encoding_delay]= (Picture*)pic;
918
919     return 0;
920 }
921
922 static int skip_check(MpegEncContext *s, Picture *p, Picture *ref){
923     int x, y, plane;
924     int score=0;
925     int64_t score64=0;
926
927     for(plane=0; plane<3; plane++){
928         const int stride = p->f.linesize[plane];
929         const int bw= plane ? 1 : 2;
930         for(y=0; y<s->mb_height*bw; y++){
931             for(x=0; x<s->mb_width*bw; x++){
932                 int off = p->f.type == FF_BUFFER_TYPE_SHARED ? 0: 16;
933                 int v   = s->dsp.frame_skip_cmp[1](s, p->f.data[plane] + 8*(x + y*stride)+off, ref->f.data[plane] + 8*(x + y*stride), stride, 8);
934
935                 switch(s->avctx->frame_skip_exp){
936                     case 0: score= FFMAX(score, v); break;
937                     case 1: score+= FFABS(v);break;
938                     case 2: score+= v*v;break;
939                     case 3: score64+= FFABS(v*v*(int64_t)v);break;
940                     case 4: score64+= v*v*(int64_t)(v*v);break;
941                 }
942             }
943         }
944     }
945
946     if(score) score64= score;
947
948     if(score64 < s->avctx->frame_skip_threshold)
949         return 1;
950     if(score64 < ((s->avctx->frame_skip_factor * (int64_t)s->lambda)>>8))
951         return 1;
952     return 0;
953 }
954
955 static int estimate_best_b_count(MpegEncContext *s){
956     AVCodec *codec= avcodec_find_encoder(s->avctx->codec_id);
957     AVCodecContext *c = avcodec_alloc_context3(NULL);
958     AVFrame input[FF_MAX_B_FRAMES+2];
959     const int scale= s->avctx->brd_scale;
960     int i, j, out_size, p_lambda, b_lambda, lambda2;
961     int outbuf_size= s->width * s->height; //FIXME
962     uint8_t *outbuf= av_malloc(outbuf_size);
963     int64_t best_rd= INT64_MAX;
964     int best_b_count= -1;
965
966     assert(scale>=0 && scale <=3);
967
968 //    emms_c();
969     p_lambda= s->last_lambda_for[AV_PICTURE_TYPE_P]; //s->next_picture_ptr->quality;
970     b_lambda= s->last_lambda_for[AV_PICTURE_TYPE_B]; //p_lambda *FFABS(s->avctx->b_quant_factor) + s->avctx->b_quant_offset;
971     if(!b_lambda) b_lambda= p_lambda; //FIXME we should do this somewhere else
972     lambda2= (b_lambda*b_lambda + (1<<FF_LAMBDA_SHIFT)/2 ) >> FF_LAMBDA_SHIFT;
973
974     c->width = s->width >> scale;
975     c->height= s->height>> scale;
976     c->flags= CODEC_FLAG_QSCALE | CODEC_FLAG_PSNR | CODEC_FLAG_INPUT_PRESERVED /*| CODEC_FLAG_EMU_EDGE*/;
977     c->flags|= s->avctx->flags & CODEC_FLAG_QPEL;
978     c->mb_decision= s->avctx->mb_decision;
979     c->me_cmp= s->avctx->me_cmp;
980     c->mb_cmp= s->avctx->mb_cmp;
981     c->me_sub_cmp= s->avctx->me_sub_cmp;
982     c->pix_fmt = PIX_FMT_YUV420P;
983     c->time_base= s->avctx->time_base;
984     c->max_b_frames= s->max_b_frames;
985
986     if (avcodec_open2(c, codec, NULL) < 0)
987         return -1;
988
989     for(i=0; i<s->max_b_frames+2; i++){
990         int ysize= c->width*c->height;
991         int csize= (c->width/2)*(c->height/2);
992         Picture pre_input, *pre_input_ptr= i ? s->input_picture[i-1] : s->next_picture_ptr;
993
994         avcodec_get_frame_defaults(&input[i]);
995         input[i].data[0]= av_malloc(ysize + 2*csize);
996         input[i].data[1]= input[i].data[0] + ysize;
997         input[i].data[2]= input[i].data[1] + csize;
998         input[i].linesize[0]= c->width;
999         input[i].linesize[1]=
1000         input[i].linesize[2]= c->width/2;
1001
1002         if(pre_input_ptr && (!i || s->input_picture[i-1])) {
1003             pre_input= *pre_input_ptr;
1004
1005             if (pre_input.f.type != FF_BUFFER_TYPE_SHARED && i) {
1006                 pre_input.f.data[0] += INPLACE_OFFSET;
1007                 pre_input.f.data[1] += INPLACE_OFFSET;
1008                 pre_input.f.data[2] += INPLACE_OFFSET;
1009             }
1010
1011             s->dsp.shrink[scale](input[i].data[0], input[i].linesize[0], pre_input.f.data[0], pre_input.f.linesize[0], c->width,      c->height);
1012             s->dsp.shrink[scale](input[i].data[1], input[i].linesize[1], pre_input.f.data[1], pre_input.f.linesize[1], c->width >> 1, c->height >> 1);
1013             s->dsp.shrink[scale](input[i].data[2], input[i].linesize[2], pre_input.f.data[2], pre_input.f.linesize[2], c->width >> 1, c->height >> 1);
1014         }
1015     }
1016
1017     for(j=0; j<s->max_b_frames+1; j++){
1018         int64_t rd=0;
1019
1020         if(!s->input_picture[j])
1021             break;
1022
1023         c->error[0]= c->error[1]= c->error[2]= 0;
1024
1025         input[0].pict_type= AV_PICTURE_TYPE_I;
1026         input[0].quality= 1 * FF_QP2LAMBDA;
1027         out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[0]);
1028 //        rd += (out_size * lambda2) >> FF_LAMBDA_SHIFT;
1029
1030         for(i=0; i<s->max_b_frames+1; i++){
1031             int is_p= i % (j+1) == j || i==s->max_b_frames;
1032
1033             input[i+1].pict_type= is_p ? AV_PICTURE_TYPE_P : AV_PICTURE_TYPE_B;
1034             input[i+1].quality= is_p ? p_lambda : b_lambda;
1035             out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[i+1]);
1036             rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
1037         }
1038
1039         /* get the delayed frames */
1040         while(out_size){
1041             out_size = avcodec_encode_video(c, outbuf, outbuf_size, NULL);
1042             rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
1043         }
1044
1045         rd += c->error[0] + c->error[1] + c->error[2];
1046
1047         if(rd < best_rd){
1048             best_rd= rd;
1049             best_b_count= j;
1050         }
1051     }
1052
1053     av_freep(&outbuf);
1054     avcodec_close(c);
1055     av_freep(&c);
1056
1057     for(i=0; i<s->max_b_frames+2; i++){
1058         av_freep(&input[i].data[0]);
1059     }
1060
1061     return best_b_count;
1062 }
1063
1064 static int select_input_picture(MpegEncContext *s){
1065     int i;
1066
1067     for(i=1; i<MAX_PICTURE_COUNT; i++)
1068         s->reordered_input_picture[i-1]= s->reordered_input_picture[i];
1069     s->reordered_input_picture[MAX_PICTURE_COUNT-1]= NULL;
1070
1071     /* set next picture type & ordering */
1072     if(s->reordered_input_picture[0]==NULL && s->input_picture[0]){
1073         if(/*s->picture_in_gop_number >= s->gop_size ||*/ s->next_picture_ptr==NULL || s->intra_only){
1074             s->reordered_input_picture[0]= s->input_picture[0];
1075             s->reordered_input_picture[0]->f.pict_type = AV_PICTURE_TYPE_I;
1076             s->reordered_input_picture[0]->f.coded_picture_number = s->coded_picture_number++;
1077         }else{
1078             int b_frames;
1079
1080             if(s->avctx->frame_skip_threshold || s->avctx->frame_skip_factor){
1081                 if(s->picture_in_gop_number < s->gop_size && skip_check(s, s->input_picture[0], s->next_picture_ptr)){
1082                 //FIXME check that te gop check above is +-1 correct
1083 //av_log(NULL, AV_LOG_DEBUG, "skip %p %"PRId64"\n", s->input_picture[0]->f.data[0], s->input_picture[0]->pts);
1084
1085                     if (s->input_picture[0]->f.type == FF_BUFFER_TYPE_SHARED) {
1086                         for(i=0; i<4; i++)
1087                             s->input_picture[0]->f.data[i] = NULL;
1088                         s->input_picture[0]->f.type = 0;
1089                     }else{
1090                         assert(   s->input_picture[0]->type==FF_BUFFER_TYPE_USER
1091                                || s->input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
1092
1093                         s->avctx->release_buffer(s->avctx, (AVFrame*)s->input_picture[0]);
1094                     }
1095
1096                     emms_c();
1097                     ff_vbv_update(s, 0);
1098
1099                     goto no_output_pic;
1100                 }
1101             }
1102
1103             if(s->flags&CODEC_FLAG_PASS2){
1104                 for(i=0; i<s->max_b_frames+1; i++){
1105                     int pict_num = s->input_picture[0]->f.display_picture_number + i;
1106
1107                     if(pict_num >= s->rc_context.num_entries)
1108                         break;
1109                     if(!s->input_picture[i]){
1110                         s->rc_context.entry[pict_num-1].new_pict_type = AV_PICTURE_TYPE_P;
1111                         break;
1112                     }
1113
1114                     s->input_picture[i]->f.pict_type =
1115                         s->rc_context.entry[pict_num].new_pict_type;
1116                 }
1117             }
1118
1119             if(s->avctx->b_frame_strategy==0){
1120                 b_frames= s->max_b_frames;
1121                 while(b_frames && !s->input_picture[b_frames]) b_frames--;
1122             }else if(s->avctx->b_frame_strategy==1){
1123                 for(i=1; i<s->max_b_frames+1; i++){
1124                     if(s->input_picture[i] && s->input_picture[i]->b_frame_score==0){
1125                         s->input_picture[i]->b_frame_score=
1126                             get_intra_count(s, s->input_picture[i  ]->f.data[0],
1127                                                s->input_picture[i-1]->f.data[0], s->linesize) + 1;
1128                     }
1129                 }
1130                 for(i=0; i<s->max_b_frames+1; i++){
1131                     if(s->input_picture[i]==NULL || s->input_picture[i]->b_frame_score - 1 > s->mb_num/s->avctx->b_sensitivity) break;
1132                 }
1133
1134                 b_frames= FFMAX(0, i-1);
1135
1136                 /* reset scores */
1137                 for(i=0; i<b_frames+1; i++){
1138                     s->input_picture[i]->b_frame_score=0;
1139                 }
1140             }else if(s->avctx->b_frame_strategy==2){
1141                 b_frames= estimate_best_b_count(s);
1142             }else{
1143                 av_log(s->avctx, AV_LOG_ERROR, "illegal b frame strategy\n");
1144                 b_frames=0;
1145             }
1146
1147             emms_c();
1148 //static int b_count=0;
1149 //b_count+= b_frames;
1150 //av_log(s->avctx, AV_LOG_DEBUG, "b_frames: %d\n", b_count);
1151
1152             for(i= b_frames - 1; i>=0; i--){
1153                 int type = s->input_picture[i]->f.pict_type;
1154                 if(type && type != AV_PICTURE_TYPE_B)
1155                     b_frames= i;
1156             }
1157             if (s->input_picture[b_frames]->f.pict_type == AV_PICTURE_TYPE_B && b_frames == s->max_b_frames){
1158                 av_log(s->avctx, AV_LOG_ERROR, "warning, too many b frames in a row\n");
1159             }
1160
1161             if(s->picture_in_gop_number + b_frames >= s->gop_size){
1162               if((s->flags2 & CODEC_FLAG2_STRICT_GOP) && s->gop_size > s->picture_in_gop_number){
1163                     b_frames= s->gop_size - s->picture_in_gop_number - 1;
1164               }else{
1165                 if(s->flags & CODEC_FLAG_CLOSED_GOP)
1166                     b_frames=0;
1167                 s->input_picture[b_frames]->f.pict_type = AV_PICTURE_TYPE_I;
1168               }
1169             }
1170
1171             if(   (s->flags & CODEC_FLAG_CLOSED_GOP)
1172                && b_frames
1173                && s->input_picture[b_frames]->f.pict_type== AV_PICTURE_TYPE_I)
1174                 b_frames--;
1175
1176             s->reordered_input_picture[0]= s->input_picture[b_frames];
1177             if (s->reordered_input_picture[0]->f.pict_type != AV_PICTURE_TYPE_I)
1178                 s->reordered_input_picture[0]->f.pict_type = AV_PICTURE_TYPE_P;
1179             s->reordered_input_picture[0]->f.coded_picture_number = s->coded_picture_number++;
1180             for(i=0; i<b_frames; i++){
1181                 s->reordered_input_picture[i + 1] = s->input_picture[i];
1182                 s->reordered_input_picture[i + 1]->f.pict_type = AV_PICTURE_TYPE_B;
1183                 s->reordered_input_picture[i + 1]->f.coded_picture_number = s->coded_picture_number++;
1184             }
1185         }
1186     }
1187 no_output_pic:
1188     if(s->reordered_input_picture[0]){
1189         s->reordered_input_picture[0]->f.reference = s->reordered_input_picture[0]->f.pict_type!=AV_PICTURE_TYPE_B ? 3 : 0;
1190
1191         ff_copy_picture(&s->new_picture, s->reordered_input_picture[0]);
1192
1193         if (s->reordered_input_picture[0]->f.type == FF_BUFFER_TYPE_SHARED || s->avctx->rc_buffer_size) {
1194             // input is a shared pix, so we can't modifiy it -> alloc a new one & ensure that the shared one is reuseable
1195
1196             int i= ff_find_unused_picture(s, 0);
1197             Picture *pic= &s->picture[i];
1198
1199             pic->f.reference = s->reordered_input_picture[0]->f.reference;
1200             if(ff_alloc_picture(s, pic, 0) < 0){
1201                 return -1;
1202             }
1203
1204             /* mark us unused / free shared pic */
1205             if (s->reordered_input_picture[0]->f.type == FF_BUFFER_TYPE_INTERNAL)
1206                 s->avctx->release_buffer(s->avctx, (AVFrame*)s->reordered_input_picture[0]);
1207             for(i=0; i<4; i++)
1208                 s->reordered_input_picture[0]->f.data[i] = NULL;
1209             s->reordered_input_picture[0]->f.type = 0;
1210
1211             copy_picture_attributes(s, (AVFrame*)pic, (AVFrame*)s->reordered_input_picture[0]);
1212
1213             s->current_picture_ptr= pic;
1214         }else{
1215             // input is not a shared pix -> reuse buffer for current_pix
1216
1217             assert(   s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_USER
1218                    || s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
1219
1220             s->current_picture_ptr= s->reordered_input_picture[0];
1221             for(i=0; i<4; i++){
1222                 s->new_picture.f.data[i] += INPLACE_OFFSET;
1223             }
1224         }
1225         ff_copy_picture(&s->current_picture, s->current_picture_ptr);
1226
1227         s->picture_number = s->new_picture.f.display_picture_number;
1228 //printf("dpn:%d\n", s->picture_number);
1229     }else{
1230        memset(&s->new_picture, 0, sizeof(Picture));
1231     }
1232     return 0;
1233 }
1234
1235 int MPV_encode_picture(AVCodecContext *avctx,
1236                        unsigned char *buf, int buf_size, void *data)
1237 {
1238     MpegEncContext *s = avctx->priv_data;
1239     AVFrame *pic_arg = data;
1240     int i, stuffing_count, context_count = avctx->thread_count;
1241
1242     for(i=0; i<context_count; i++){
1243         int start_y= s->thread_context[i]->start_mb_y;
1244         int   end_y= s->thread_context[i]->  end_mb_y;
1245         int h= s->mb_height;
1246         uint8_t *start= buf + (size_t)(((int64_t) buf_size)*start_y/h);
1247         uint8_t *end  = buf + (size_t)(((int64_t) buf_size)*  end_y/h);
1248
1249         init_put_bits(&s->thread_context[i]->pb, start, end - start);
1250     }
1251
1252     s->picture_in_gop_number++;
1253
1254     if(load_input_picture(s, pic_arg) < 0)
1255         return -1;
1256
1257     if(select_input_picture(s) < 0){
1258         return -1;
1259     }
1260
1261     /* output? */
1262     if (s->new_picture.f.data[0]) {
1263         s->pict_type = s->new_picture.f.pict_type;
1264 //emms_c();
1265 //printf("qs:%f %f %d\n", s->new_picture.quality, s->current_picture.quality, s->qscale);
1266         MPV_frame_start(s, avctx);
1267 vbv_retry:
1268         if (encode_picture(s, s->picture_number) < 0)
1269             return -1;
1270
1271         avctx->header_bits = s->header_bits;
1272         avctx->mv_bits     = s->mv_bits;
1273         avctx->misc_bits   = s->misc_bits;
1274         avctx->i_tex_bits  = s->i_tex_bits;
1275         avctx->p_tex_bits  = s->p_tex_bits;
1276         avctx->i_count     = s->i_count;
1277         avctx->p_count     = s->mb_num - s->i_count - s->skip_count; //FIXME f/b_count in avctx
1278         avctx->skip_count  = s->skip_count;
1279
1280         MPV_frame_end(s);
1281
1282         if (CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG)
1283             ff_mjpeg_encode_picture_trailer(s);
1284
1285         if(avctx->rc_buffer_size){
1286             RateControlContext *rcc= &s->rc_context;
1287             int max_size= rcc->buffer_index * avctx->rc_max_available_vbv_use;
1288
1289             if(put_bits_count(&s->pb) > max_size && s->lambda < s->avctx->lmax){
1290                 s->next_lambda= FFMAX(s->lambda+1, s->lambda*(s->qscale+1) / s->qscale);
1291                 if(s->adaptive_quant){
1292                     int i;
1293                     for(i=0; i<s->mb_height*s->mb_stride; i++)
1294                         s->lambda_table[i]= FFMAX(s->lambda_table[i]+1, s->lambda_table[i]*(s->qscale+1) / s->qscale);
1295                 }
1296                 s->mb_skipped = 0;        //done in MPV_frame_start()
1297                 if(s->pict_type==AV_PICTURE_TYPE_P){ //done in encode_picture() so we must undo it
1298                     if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
1299                         s->no_rounding ^= 1;
1300                 }
1301                 if(s->pict_type!=AV_PICTURE_TYPE_B){
1302                     s->time_base= s->last_time_base;
1303                     s->last_non_b_time= s->time - s->pp_time;
1304                 }
1305 //                av_log(NULL, AV_LOG_ERROR, "R:%d ", s->next_lambda);
1306                 for(i=0; i<context_count; i++){
1307                     PutBitContext *pb= &s->thread_context[i]->pb;
1308                     init_put_bits(pb, pb->buf, pb->buf_end - pb->buf);
1309                 }
1310                 goto vbv_retry;
1311             }
1312
1313             assert(s->avctx->rc_max_rate);
1314         }
1315
1316         if(s->flags&CODEC_FLAG_PASS1)
1317             ff_write_pass1_stats(s);
1318
1319         for(i=0; i<4; i++){
1320             s->current_picture_ptr->f.error[i]  = s->current_picture.f.error[i];
1321             avctx->error[i]                        += s->current_picture_ptr->f.error[i];
1322         }
1323
1324         if(s->flags&CODEC_FLAG_PASS1)
1325             assert(avctx->header_bits + avctx->mv_bits + avctx->misc_bits + avctx->i_tex_bits + avctx->p_tex_bits == put_bits_count(&s->pb));
1326         flush_put_bits(&s->pb);
1327         s->frame_bits  = put_bits_count(&s->pb);
1328
1329         stuffing_count= ff_vbv_update(s, s->frame_bits);
1330         if(stuffing_count){
1331             if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < stuffing_count + 50){
1332                 av_log(s->avctx, AV_LOG_ERROR, "stuffing too large\n");
1333                 return -1;
1334             }
1335
1336             switch(s->codec_id){
1337             case CODEC_ID_MPEG1VIDEO:
1338             case CODEC_ID_MPEG2VIDEO:
1339                 while(stuffing_count--){
1340                     put_bits(&s->pb, 8, 0);
1341                 }
1342             break;
1343             case CODEC_ID_MPEG4:
1344                 put_bits(&s->pb, 16, 0);
1345                 put_bits(&s->pb, 16, 0x1C3);
1346                 stuffing_count -= 4;
1347                 while(stuffing_count--){
1348                     put_bits(&s->pb, 8, 0xFF);
1349                 }
1350             break;
1351             default:
1352                 av_log(s->avctx, AV_LOG_ERROR, "vbv buffer overflow\n");
1353             }
1354             flush_put_bits(&s->pb);
1355             s->frame_bits  = put_bits_count(&s->pb);
1356         }
1357
1358         /* update mpeg1/2 vbv_delay for CBR */
1359         if(s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate && s->out_format == FMT_MPEG1
1360            && 90000LL * (avctx->rc_buffer_size-1) <= s->avctx->rc_max_rate*0xFFFFLL){
1361             int vbv_delay, min_delay;
1362             double inbits = s->avctx->rc_max_rate*av_q2d(s->avctx->time_base);
1363             int    minbits= s->frame_bits - 8*(s->vbv_delay_ptr - s->pb.buf - 1);
1364             double bits   = s->rc_context.buffer_index + minbits - inbits;
1365
1366             if(bits<0)
1367                 av_log(s->avctx, AV_LOG_ERROR, "Internal error, negative bits\n");
1368
1369             assert(s->repeat_first_field==0);
1370
1371             vbv_delay=     bits * 90000                               / s->avctx->rc_max_rate;
1372             min_delay= (minbits * 90000LL + s->avctx->rc_max_rate - 1)/ s->avctx->rc_max_rate;
1373
1374             vbv_delay= FFMAX(vbv_delay, min_delay);
1375
1376             assert(vbv_delay < 0xFFFF);
1377
1378             s->vbv_delay_ptr[0] &= 0xF8;
1379             s->vbv_delay_ptr[0] |= vbv_delay>>13;
1380             s->vbv_delay_ptr[1]  = vbv_delay>>5;
1381             s->vbv_delay_ptr[2] &= 0x07;
1382             s->vbv_delay_ptr[2] |= vbv_delay<<3;
1383             avctx->vbv_delay = vbv_delay*300;
1384         }
1385         s->total_bits += s->frame_bits;
1386         avctx->frame_bits  = s->frame_bits;
1387     }else{
1388         assert((put_bits_ptr(&s->pb) == s->pb.buf));
1389         s->frame_bits=0;
1390     }
1391     assert((s->frame_bits&7)==0);
1392
1393     return s->frame_bits/8;
1394 }
1395
1396 static inline void dct_single_coeff_elimination(MpegEncContext *s, int n, int threshold)
1397 {
1398     static const char tab[64]=
1399         {3,2,2,1,1,1,1,1,
1400          1,1,1,1,1,1,1,1,
1401          1,1,1,1,1,1,1,1,
1402          0,0,0,0,0,0,0,0,
1403          0,0,0,0,0,0,0,0,
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     int score=0;
1408     int run=0;
1409     int i;
1410     DCTELEM *block= s->block[n];
1411     const int last_index= s->block_last_index[n];
1412     int skip_dc;
1413
1414     if(threshold<0){
1415         skip_dc=0;
1416         threshold= -threshold;
1417     }else
1418         skip_dc=1;
1419
1420     /* Are all we could set to zero already zero? */
1421     if(last_index<=skip_dc - 1) return;
1422
1423     for(i=0; i<=last_index; i++){
1424         const int j = s->intra_scantable.permutated[i];
1425         const int level = FFABS(block[j]);
1426         if(level==1){
1427             if(skip_dc && i==0) continue;
1428             score+= tab[run];
1429             run=0;
1430         }else if(level>1){
1431             return;
1432         }else{
1433             run++;
1434         }
1435     }
1436     if(score >= threshold) return;
1437     for(i=skip_dc; i<=last_index; i++){
1438         const int j = s->intra_scantable.permutated[i];
1439         block[j]=0;
1440     }
1441     if(block[0]) s->block_last_index[n]= 0;
1442     else         s->block_last_index[n]= -1;
1443 }
1444
1445 static inline void clip_coeffs(MpegEncContext *s, DCTELEM *block, int last_index)
1446 {
1447     int i;
1448     const int maxlevel= s->max_qcoeff;
1449     const int minlevel= s->min_qcoeff;
1450     int overflow=0;
1451
1452     if(s->mb_intra){
1453         i=1; //skip clipping of intra dc
1454     }else
1455         i=0;
1456
1457     for(;i<=last_index; i++){
1458         const int j= s->intra_scantable.permutated[i];
1459         int level = block[j];
1460
1461         if     (level>maxlevel){
1462             level=maxlevel;
1463             overflow++;
1464         }else if(level<minlevel){
1465             level=minlevel;
1466             overflow++;
1467         }
1468
1469         block[j]= level;
1470     }
1471
1472     if(overflow && s->avctx->mb_decision == FF_MB_DECISION_SIMPLE)
1473         av_log(s->avctx, AV_LOG_INFO, "warning, clipping %d dct coefficients to %d..%d\n", overflow, minlevel, maxlevel);
1474 }
1475
1476 static void get_visual_weight(int16_t *weight, uint8_t *ptr, int stride){
1477     int x, y;
1478 //FIXME optimize
1479     for(y=0; y<8; y++){
1480         for(x=0; x<8; x++){
1481             int x2, y2;
1482             int sum=0;
1483             int sqr=0;
1484             int count=0;
1485
1486             for(y2= FFMAX(y-1, 0); y2 < FFMIN(8, y+2); y2++){
1487                 for(x2= FFMAX(x-1, 0); x2 < FFMIN(8, x+2); x2++){
1488                     int v= ptr[x2 + y2*stride];
1489                     sum += v;
1490                     sqr += v*v;
1491                     count++;
1492                 }
1493             }
1494             weight[x + 8*y]= (36*ff_sqrt(count*sqr - sum*sum)) / count;
1495         }
1496     }
1497 }
1498
1499 static av_always_inline void encode_mb_internal(MpegEncContext *s, int motion_x, int motion_y, int mb_block_height, int mb_block_count)
1500 {
1501     int16_t weight[8][64];
1502     DCTELEM orig[8][64];
1503     const int mb_x= s->mb_x;
1504     const int mb_y= s->mb_y;
1505     int i;
1506     int skip_dct[8];
1507     int dct_offset   = s->linesize*8; //default for progressive frames
1508     uint8_t *ptr_y, *ptr_cb, *ptr_cr;
1509     int wrap_y, wrap_c;
1510
1511     for(i=0; i<mb_block_count; i++) skip_dct[i]=s->skipdct;
1512
1513     if(s->adaptive_quant){
1514         const int last_qp= s->qscale;
1515         const int mb_xy= mb_x + mb_y*s->mb_stride;
1516
1517         s->lambda= s->lambda_table[mb_xy];
1518         update_qscale(s);
1519
1520         if(!(s->flags&CODEC_FLAG_QP_RD)){
1521             s->qscale = s->current_picture_ptr->f.qscale_table[mb_xy];
1522             s->dquant= s->qscale - last_qp;
1523
1524             if(s->out_format==FMT_H263){
1525                 s->dquant= av_clip(s->dquant, -2, 2);
1526
1527                 if(s->codec_id==CODEC_ID_MPEG4){
1528                     if(!s->mb_intra){
1529                         if(s->pict_type == AV_PICTURE_TYPE_B){
1530                             if(s->dquant&1 || s->mv_dir&MV_DIRECT)
1531                                 s->dquant= 0;
1532                         }
1533                         if(s->mv_type==MV_TYPE_8X8)
1534                             s->dquant=0;
1535                     }
1536                 }
1537             }
1538         }
1539         ff_set_qscale(s, last_qp + s->dquant);
1540     }else if(s->flags&CODEC_FLAG_QP_RD)
1541         ff_set_qscale(s, s->qscale + s->dquant);
1542
1543     wrap_y = s->linesize;
1544     wrap_c = s->uvlinesize;
1545     ptr_y  = s->new_picture.f.data[0] + (mb_y * 16 * wrap_y) + mb_x * 16;
1546     ptr_cb = s->new_picture.f.data[1] + (mb_y * mb_block_height * wrap_c) + mb_x * 8;
1547     ptr_cr = s->new_picture.f.data[2] + (mb_y * mb_block_height * wrap_c) + mb_x * 8;
1548
1549     if(mb_x*16+16 > s->width || mb_y*16+16 > s->height){
1550         uint8_t *ebuf= s->edge_emu_buffer + 32;
1551         s->dsp.emulated_edge_mc(ebuf            , ptr_y , wrap_y,16,16,mb_x*16,mb_y*16, s->width   , s->height);
1552         ptr_y= ebuf;
1553         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);
1554         ptr_cb= ebuf+18*wrap_y;
1555         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);
1556         ptr_cr= ebuf+18*wrap_y+8;
1557     }
1558
1559     if (s->mb_intra) {
1560         if(s->flags&CODEC_FLAG_INTERLACED_DCT){
1561             int progressive_score, interlaced_score;
1562
1563             s->interlaced_dct=0;
1564             progressive_score= s->dsp.ildct_cmp[4](s, ptr_y           , NULL, wrap_y, 8)
1565                               +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y*8, NULL, wrap_y, 8) - 400;
1566
1567             if(progressive_score > 0){
1568                 interlaced_score = s->dsp.ildct_cmp[4](s, ptr_y           , NULL, wrap_y*2, 8)
1569                                   +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y  , NULL, wrap_y*2, 8);
1570                 if(progressive_score > interlaced_score){
1571                     s->interlaced_dct=1;
1572
1573                     dct_offset= wrap_y;
1574                     wrap_y<<=1;
1575                     if (s->chroma_format == CHROMA_422)
1576                         wrap_c<<=1;
1577                 }
1578             }
1579         }
1580
1581         s->dsp.get_pixels(s->block[0], ptr_y                 , wrap_y);
1582         s->dsp.get_pixels(s->block[1], ptr_y              + 8, wrap_y);
1583         s->dsp.get_pixels(s->block[2], ptr_y + dct_offset    , wrap_y);
1584         s->dsp.get_pixels(s->block[3], ptr_y + dct_offset + 8, wrap_y);
1585
1586         if(s->flags&CODEC_FLAG_GRAY){
1587             skip_dct[4]= 1;
1588             skip_dct[5]= 1;
1589         }else{
1590             s->dsp.get_pixels(s->block[4], ptr_cb, wrap_c);
1591             s->dsp.get_pixels(s->block[5], ptr_cr, wrap_c);
1592             if(!s->chroma_y_shift){ /* 422 */
1593                 s->dsp.get_pixels(s->block[6], ptr_cb + (dct_offset>>1), wrap_c);
1594                 s->dsp.get_pixels(s->block[7], ptr_cr + (dct_offset>>1), wrap_c);
1595             }
1596         }
1597     }else{
1598         op_pixels_func (*op_pix)[4];
1599         qpel_mc_func (*op_qpix)[16];
1600         uint8_t *dest_y, *dest_cb, *dest_cr;
1601
1602         dest_y  = s->dest[0];
1603         dest_cb = s->dest[1];
1604         dest_cr = s->dest[2];
1605
1606         if ((!s->no_rounding) || s->pict_type==AV_PICTURE_TYPE_B){
1607             op_pix = s->dsp.put_pixels_tab;
1608             op_qpix= s->dsp.put_qpel_pixels_tab;
1609         }else{
1610             op_pix = s->dsp.put_no_rnd_pixels_tab;
1611             op_qpix= s->dsp.put_no_rnd_qpel_pixels_tab;
1612         }
1613
1614         if (s->mv_dir & MV_DIR_FORWARD) {
1615             MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.f.data, op_pix, op_qpix);
1616             op_pix = s->dsp.avg_pixels_tab;
1617             op_qpix= s->dsp.avg_qpel_pixels_tab;
1618         }
1619         if (s->mv_dir & MV_DIR_BACKWARD) {
1620             MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.f.data, op_pix, op_qpix);
1621         }
1622
1623         if(s->flags&CODEC_FLAG_INTERLACED_DCT){
1624             int progressive_score, interlaced_score;
1625
1626             s->interlaced_dct=0;
1627             progressive_score= s->dsp.ildct_cmp[0](s, dest_y           , ptr_y           , wrap_y, 8)
1628                               +s->dsp.ildct_cmp[0](s, dest_y + wrap_y*8, ptr_y + wrap_y*8, wrap_y, 8) - 400;
1629
1630             if(s->avctx->ildct_cmp == FF_CMP_VSSE) progressive_score -= 400;
1631
1632             if(progressive_score>0){
1633                 interlaced_score = s->dsp.ildct_cmp[0](s, dest_y           , ptr_y           , wrap_y*2, 8)
1634                                   +s->dsp.ildct_cmp[0](s, dest_y + wrap_y  , ptr_y + wrap_y  , wrap_y*2, 8);
1635
1636                 if(progressive_score > interlaced_score){
1637                     s->interlaced_dct=1;
1638
1639                     dct_offset= wrap_y;
1640                     wrap_y<<=1;
1641                     if (s->chroma_format == CHROMA_422)
1642                         wrap_c<<=1;
1643                 }
1644             }
1645         }
1646
1647         s->dsp.diff_pixels(s->block[0], ptr_y                 , dest_y                 , wrap_y);
1648         s->dsp.diff_pixels(s->block[1], ptr_y              + 8, dest_y              + 8, wrap_y);
1649         s->dsp.diff_pixels(s->block[2], ptr_y + dct_offset    , dest_y + dct_offset    , wrap_y);
1650         s->dsp.diff_pixels(s->block[3], ptr_y + dct_offset + 8, dest_y + dct_offset + 8, wrap_y);
1651
1652         if(s->flags&CODEC_FLAG_GRAY){
1653             skip_dct[4]= 1;
1654             skip_dct[5]= 1;
1655         }else{
1656             s->dsp.diff_pixels(s->block[4], ptr_cb, dest_cb, wrap_c);
1657             s->dsp.diff_pixels(s->block[5], ptr_cr, dest_cr, wrap_c);
1658             if(!s->chroma_y_shift){ /* 422 */
1659                 s->dsp.diff_pixels(s->block[6], ptr_cb + (dct_offset>>1), dest_cb + (dct_offset>>1), wrap_c);
1660                 s->dsp.diff_pixels(s->block[7], ptr_cr + (dct_offset>>1), dest_cr + (dct_offset>>1), wrap_c);
1661             }
1662         }
1663         /* pre quantization */
1664         if(s->current_picture.mc_mb_var[s->mb_stride*mb_y+ mb_x]<2*s->qscale*s->qscale){
1665             //FIXME optimize
1666             if(s->dsp.sad[1](NULL, ptr_y               , dest_y               , wrap_y, 8) < 20*s->qscale) skip_dct[0]= 1;
1667             if(s->dsp.sad[1](NULL, ptr_y            + 8, dest_y            + 8, wrap_y, 8) < 20*s->qscale) skip_dct[1]= 1;
1668             if(s->dsp.sad[1](NULL, ptr_y +dct_offset   , dest_y +dct_offset   , wrap_y, 8) < 20*s->qscale) skip_dct[2]= 1;
1669             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;
1670             if(s->dsp.sad[1](NULL, ptr_cb              , dest_cb              , wrap_c, 8) < 20*s->qscale) skip_dct[4]= 1;
1671             if(s->dsp.sad[1](NULL, ptr_cr              , dest_cr              , wrap_c, 8) < 20*s->qscale) skip_dct[5]= 1;
1672             if(!s->chroma_y_shift){ /* 422 */
1673                 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;
1674                 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;
1675             }
1676         }
1677     }
1678
1679     if(s->avctx->quantizer_noise_shaping){
1680         if(!skip_dct[0]) get_visual_weight(weight[0], ptr_y                 , wrap_y);
1681         if(!skip_dct[1]) get_visual_weight(weight[1], ptr_y              + 8, wrap_y);
1682         if(!skip_dct[2]) get_visual_weight(weight[2], ptr_y + dct_offset    , wrap_y);
1683         if(!skip_dct[3]) get_visual_weight(weight[3], ptr_y + dct_offset + 8, wrap_y);
1684         if(!skip_dct[4]) get_visual_weight(weight[4], ptr_cb                , wrap_c);
1685         if(!skip_dct[5]) get_visual_weight(weight[5], ptr_cr                , wrap_c);
1686         if(!s->chroma_y_shift){ /* 422 */
1687             if(!skip_dct[6]) get_visual_weight(weight[6], ptr_cb + (dct_offset>>1), wrap_c);
1688             if(!skip_dct[7]) get_visual_weight(weight[7], ptr_cr + (dct_offset>>1), wrap_c);
1689         }
1690         memcpy(orig[0], s->block[0], sizeof(DCTELEM)*64*mb_block_count);
1691     }
1692
1693     /* DCT & quantize */
1694     assert(s->out_format!=FMT_MJPEG || s->qscale==8);
1695     {
1696         for(i=0;i<mb_block_count;i++) {
1697             if(!skip_dct[i]){
1698                 int overflow;
1699                 s->block_last_index[i] = s->dct_quantize(s, s->block[i], i, s->qscale, &overflow);
1700             // FIXME we could decide to change to quantizer instead of clipping
1701             // JS: I don't think that would be a good idea it could lower quality instead
1702             //     of improve it. Just INTRADC clipping deserves changes in quantizer
1703                 if (overflow) clip_coeffs(s, s->block[i], s->block_last_index[i]);
1704             }else
1705                 s->block_last_index[i]= -1;
1706         }
1707         if(s->avctx->quantizer_noise_shaping){
1708             for(i=0;i<mb_block_count;i++) {
1709                 if(!skip_dct[i]){
1710                     s->block_last_index[i] = dct_quantize_refine(s, s->block[i], weight[i], orig[i], i, s->qscale);
1711                 }
1712             }
1713         }
1714
1715         if(s->luma_elim_threshold && !s->mb_intra)
1716             for(i=0; i<4; i++)
1717                 dct_single_coeff_elimination(s, i, s->luma_elim_threshold);
1718         if(s->chroma_elim_threshold && !s->mb_intra)
1719             for(i=4; i<mb_block_count; i++)
1720                 dct_single_coeff_elimination(s, i, s->chroma_elim_threshold);
1721
1722         if(s->flags & CODEC_FLAG_CBP_RD){
1723             for(i=0;i<mb_block_count;i++) {
1724                 if(s->block_last_index[i] == -1)
1725                     s->coded_score[i]= INT_MAX/256;
1726             }
1727         }
1728     }
1729
1730     if((s->flags&CODEC_FLAG_GRAY) && s->mb_intra){
1731         s->block_last_index[4]=
1732         s->block_last_index[5]= 0;
1733         s->block[4][0]=
1734         s->block[5][0]= (1024 + s->c_dc_scale/2)/ s->c_dc_scale;
1735     }
1736
1737     //non c quantize code returns incorrect block_last_index FIXME
1738     if(s->alternate_scan && s->dct_quantize != dct_quantize_c){
1739         for(i=0; i<mb_block_count; i++){
1740             int j;
1741             if(s->block_last_index[i]>0){
1742                 for(j=63; j>0; j--){
1743                     if(s->block[i][ s->intra_scantable.permutated[j] ]) break;
1744                 }
1745                 s->block_last_index[i]= j;
1746             }
1747         }
1748     }
1749
1750     /* huffman encode */
1751     switch(s->codec_id){ //FIXME funct ptr could be slightly faster
1752     case CODEC_ID_MPEG1VIDEO:
1753     case CODEC_ID_MPEG2VIDEO:
1754         if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
1755             mpeg1_encode_mb(s, s->block, motion_x, motion_y);
1756         break;
1757     case CODEC_ID_MPEG4:
1758         if (CONFIG_MPEG4_ENCODER)
1759             mpeg4_encode_mb(s, s->block, motion_x, motion_y);
1760         break;
1761     case CODEC_ID_MSMPEG4V2:
1762     case CODEC_ID_MSMPEG4V3:
1763     case CODEC_ID_WMV1:
1764         if (CONFIG_MSMPEG4_ENCODER)
1765             msmpeg4_encode_mb(s, s->block, motion_x, motion_y);
1766         break;
1767     case CODEC_ID_WMV2:
1768         if (CONFIG_WMV2_ENCODER)
1769             ff_wmv2_encode_mb(s, s->block, motion_x, motion_y);
1770         break;
1771     case CODEC_ID_H261:
1772         if (CONFIG_H261_ENCODER)
1773             ff_h261_encode_mb(s, s->block, motion_x, motion_y);
1774         break;
1775     case CODEC_ID_H263:
1776     case CODEC_ID_H263P:
1777     case CODEC_ID_FLV1:
1778     case CODEC_ID_RV10:
1779     case CODEC_ID_RV20:
1780         if (CONFIG_H263_ENCODER)
1781             h263_encode_mb(s, s->block, motion_x, motion_y);
1782         break;
1783     case CODEC_ID_MJPEG:
1784         if (CONFIG_MJPEG_ENCODER)
1785             ff_mjpeg_encode_mb(s, s->block);
1786         break;
1787     default:
1788         assert(0);
1789     }
1790 }
1791
1792 static av_always_inline void encode_mb(MpegEncContext *s, int motion_x, int motion_y)
1793 {
1794     if (s->chroma_format == CHROMA_420) encode_mb_internal(s, motion_x, motion_y,  8, 6);
1795     else                                encode_mb_internal(s, motion_x, motion_y, 16, 8);
1796 }
1797
1798 static inline void copy_context_before_encode(MpegEncContext *d, MpegEncContext *s, int type){
1799     int i;
1800
1801     memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster than a loop?
1802
1803     /* mpeg1 */
1804     d->mb_skip_run= s->mb_skip_run;
1805     for(i=0; i<3; i++)
1806         d->last_dc[i]= s->last_dc[i];
1807
1808     /* statistics */
1809     d->mv_bits= s->mv_bits;
1810     d->i_tex_bits= s->i_tex_bits;
1811     d->p_tex_bits= s->p_tex_bits;
1812     d->i_count= s->i_count;
1813     d->f_count= s->f_count;
1814     d->b_count= s->b_count;
1815     d->skip_count= s->skip_count;
1816     d->misc_bits= s->misc_bits;
1817     d->last_bits= 0;
1818
1819     d->mb_skipped= 0;
1820     d->qscale= s->qscale;
1821     d->dquant= s->dquant;
1822
1823     d->esc3_level_length= s->esc3_level_length;
1824 }
1825
1826 static inline void copy_context_after_encode(MpegEncContext *d, MpegEncContext *s, int type){
1827     int i;
1828
1829     memcpy(d->mv, s->mv, 2*4*2*sizeof(int));
1830     memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster than a loop?
1831
1832     /* mpeg1 */
1833     d->mb_skip_run= s->mb_skip_run;
1834     for(i=0; i<3; i++)
1835         d->last_dc[i]= s->last_dc[i];
1836
1837     /* statistics */
1838     d->mv_bits= s->mv_bits;
1839     d->i_tex_bits= s->i_tex_bits;
1840     d->p_tex_bits= s->p_tex_bits;
1841     d->i_count= s->i_count;
1842     d->f_count= s->f_count;
1843     d->b_count= s->b_count;
1844     d->skip_count= s->skip_count;
1845     d->misc_bits= s->misc_bits;
1846
1847     d->mb_intra= s->mb_intra;
1848     d->mb_skipped= s->mb_skipped;
1849     d->mv_type= s->mv_type;
1850     d->mv_dir= s->mv_dir;
1851     d->pb= s->pb;
1852     if(s->data_partitioning){
1853         d->pb2= s->pb2;
1854         d->tex_pb= s->tex_pb;
1855     }
1856     d->block= s->block;
1857     for(i=0; i<8; i++)
1858         d->block_last_index[i]= s->block_last_index[i];
1859     d->interlaced_dct= s->interlaced_dct;
1860     d->qscale= s->qscale;
1861
1862     d->esc3_level_length= s->esc3_level_length;
1863 }
1864
1865 static inline void encode_mb_hq(MpegEncContext *s, MpegEncContext *backup, MpegEncContext *best, int type,
1866                            PutBitContext pb[2], PutBitContext pb2[2], PutBitContext tex_pb[2],
1867                            int *dmin, int *next_block, int motion_x, int motion_y)
1868 {
1869     int score;
1870     uint8_t *dest_backup[3];
1871
1872     copy_context_before_encode(s, backup, type);
1873
1874     s->block= s->blocks[*next_block];
1875     s->pb= pb[*next_block];
1876     if(s->data_partitioning){
1877         s->pb2   = pb2   [*next_block];
1878         s->tex_pb= tex_pb[*next_block];
1879     }
1880
1881     if(*next_block){
1882         memcpy(dest_backup, s->dest, sizeof(s->dest));
1883         s->dest[0] = s->rd_scratchpad;
1884         s->dest[1] = s->rd_scratchpad + 16*s->linesize;
1885         s->dest[2] = s->rd_scratchpad + 16*s->linesize + 8;
1886         assert(s->linesize >= 32); //FIXME
1887     }
1888
1889     encode_mb(s, motion_x, motion_y);
1890
1891     score= put_bits_count(&s->pb);
1892     if(s->data_partitioning){
1893         score+= put_bits_count(&s->pb2);
1894         score+= put_bits_count(&s->tex_pb);
1895     }
1896
1897     if(s->avctx->mb_decision == FF_MB_DECISION_RD){
1898         MPV_decode_mb(s, s->block);
1899
1900         score *= s->lambda2;
1901         score += sse_mb(s) << FF_LAMBDA_SHIFT;
1902     }
1903
1904     if(*next_block){
1905         memcpy(s->dest, dest_backup, sizeof(s->dest));
1906     }
1907
1908     if(score<*dmin){
1909         *dmin= score;
1910         *next_block^=1;
1911
1912         copy_context_after_encode(best, s, type);
1913     }
1914 }
1915
1916 static int sse(MpegEncContext *s, uint8_t *src1, uint8_t *src2, int w, int h, int stride){
1917     uint32_t *sq = ff_squareTbl + 256;
1918     int acc=0;
1919     int x,y;
1920
1921     if(w==16 && h==16)
1922         return s->dsp.sse[0](NULL, src1, src2, stride, 16);
1923     else if(w==8 && h==8)
1924         return s->dsp.sse[1](NULL, src1, src2, stride, 8);
1925
1926     for(y=0; y<h; y++){
1927         for(x=0; x<w; x++){
1928             acc+= sq[src1[x + y*stride] - src2[x + y*stride]];
1929         }
1930     }
1931
1932     assert(acc>=0);
1933
1934     return acc;
1935 }
1936
1937 static int sse_mb(MpegEncContext *s){
1938     int w= 16;
1939     int h= 16;
1940
1941     if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
1942     if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
1943
1944     if(w==16 && h==16)
1945       if(s->avctx->mb_cmp == FF_CMP_NSSE){
1946         return  s->dsp.nsse[0](s, s->new_picture.f.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, s->dest[0], s->linesize, 16)
1947                +s->dsp.nsse[1](s, s->new_picture.f.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[1], s->uvlinesize, 8)
1948                +s->dsp.nsse[1](s, s->new_picture.f.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[2], s->uvlinesize, 8);
1949       }else{
1950         return  s->dsp.sse[0](NULL, s->new_picture.f.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, s->dest[0], s->linesize, 16)
1951                +s->dsp.sse[1](NULL, s->new_picture.f.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[1], s->uvlinesize, 8)
1952                +s->dsp.sse[1](NULL, s->new_picture.f.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[2], s->uvlinesize, 8);
1953       }
1954     else
1955         return  sse(s, s->new_picture.f.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, s->dest[0], w, h, s->linesize)
1956                +sse(s, s->new_picture.f.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[1], w>>1, h>>1, s->uvlinesize)
1957                +sse(s, s->new_picture.f.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[2], w>>1, h>>1, s->uvlinesize);
1958 }
1959
1960 static int pre_estimate_motion_thread(AVCodecContext *c, void *arg){
1961     MpegEncContext *s= *(void**)arg;
1962
1963
1964     s->me.pre_pass=1;
1965     s->me.dia_size= s->avctx->pre_dia_size;
1966     s->first_slice_line=1;
1967     for(s->mb_y= s->end_mb_y-1; s->mb_y >= s->start_mb_y; s->mb_y--) {
1968         for(s->mb_x=s->mb_width-1; s->mb_x >=0 ;s->mb_x--) {
1969             ff_pre_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
1970         }
1971         s->first_slice_line=0;
1972     }
1973
1974     s->me.pre_pass=0;
1975
1976     return 0;
1977 }
1978
1979 static int estimate_motion_thread(AVCodecContext *c, void *arg){
1980     MpegEncContext *s= *(void**)arg;
1981
1982     ff_check_alignment();
1983
1984     s->me.dia_size= s->avctx->dia_size;
1985     s->first_slice_line=1;
1986     for(s->mb_y= s->start_mb_y; s->mb_y < s->end_mb_y; s->mb_y++) {
1987         s->mb_x=0; //for block init below
1988         ff_init_block_index(s);
1989         for(s->mb_x=0; s->mb_x < s->mb_width; s->mb_x++) {
1990             s->block_index[0]+=2;
1991             s->block_index[1]+=2;
1992             s->block_index[2]+=2;
1993             s->block_index[3]+=2;
1994
1995             /* compute motion vector & mb_type and store in context */
1996             if(s->pict_type==AV_PICTURE_TYPE_B)
1997                 ff_estimate_b_frame_motion(s, s->mb_x, s->mb_y);
1998             else
1999                 ff_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
2000         }
2001         s->first_slice_line=0;
2002     }
2003     return 0;
2004 }
2005
2006 static int mb_var_thread(AVCodecContext *c, void *arg){
2007     MpegEncContext *s= *(void**)arg;
2008     int mb_x, mb_y;
2009
2010     ff_check_alignment();
2011
2012     for(mb_y=s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
2013         for(mb_x=0; mb_x < s->mb_width; mb_x++) {
2014             int xx = mb_x * 16;
2015             int yy = mb_y * 16;
2016             uint8_t *pix = s->new_picture.f.data[0] + (yy * s->linesize) + xx;
2017             int varc;
2018             int sum = s->dsp.pix_sum(pix, s->linesize);
2019
2020             varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500 + 128)>>8;
2021
2022             s->current_picture.mb_var [s->mb_stride * mb_y + mb_x] = varc;
2023             s->current_picture.mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
2024             s->me.mb_var_sum_temp    += varc;
2025         }
2026     }
2027     return 0;
2028 }
2029
2030 static void write_slice_end(MpegEncContext *s){
2031     if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4){
2032         if(s->partitioned_frame){
2033             ff_mpeg4_merge_partitions(s);
2034         }
2035
2036         ff_mpeg4_stuffing(&s->pb);
2037     }else if(CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG){
2038         ff_mjpeg_encode_stuffing(&s->pb);
2039     }
2040
2041     align_put_bits(&s->pb);
2042     flush_put_bits(&s->pb);
2043
2044     if((s->flags&CODEC_FLAG_PASS1) && !s->partitioned_frame)
2045         s->misc_bits+= get_bits_diff(s);
2046 }
2047
2048 static int encode_thread(AVCodecContext *c, void *arg){
2049     MpegEncContext *s= *(void**)arg;
2050     int mb_x, mb_y, pdif = 0;
2051     int chr_h= 16>>s->chroma_y_shift;
2052     int i, j;
2053     MpegEncContext best_s, backup_s;
2054     uint8_t bit_buf[2][MAX_MB_BYTES];
2055     uint8_t bit_buf2[2][MAX_MB_BYTES];
2056     uint8_t bit_buf_tex[2][MAX_MB_BYTES];
2057     PutBitContext pb[2], pb2[2], tex_pb[2];
2058 //printf("%d->%d\n", s->resync_mb_y, s->end_mb_y);
2059
2060     ff_check_alignment();
2061
2062     for(i=0; i<2; i++){
2063         init_put_bits(&pb    [i], bit_buf    [i], MAX_MB_BYTES);
2064         init_put_bits(&pb2   [i], bit_buf2   [i], MAX_MB_BYTES);
2065         init_put_bits(&tex_pb[i], bit_buf_tex[i], MAX_MB_BYTES);
2066     }
2067
2068     s->last_bits= put_bits_count(&s->pb);
2069     s->mv_bits=0;
2070     s->misc_bits=0;
2071     s->i_tex_bits=0;
2072     s->p_tex_bits=0;
2073     s->i_count=0;
2074     s->f_count=0;
2075     s->b_count=0;
2076     s->skip_count=0;
2077
2078     for(i=0; i<3; i++){
2079         /* init last dc values */
2080         /* note: quant matrix value (8) is implied here */
2081         s->last_dc[i] = 128 << s->intra_dc_precision;
2082
2083         s->current_picture.f.error[i] = 0;
2084     }
2085     s->mb_skip_run = 0;
2086     memset(s->last_mv, 0, sizeof(s->last_mv));
2087
2088     s->last_mv_dir = 0;
2089
2090     switch(s->codec_id){
2091     case CODEC_ID_H263:
2092     case CODEC_ID_H263P:
2093     case CODEC_ID_FLV1:
2094         if (CONFIG_H263_ENCODER)
2095             s->gob_index = ff_h263_get_gob_height(s);
2096         break;
2097     case CODEC_ID_MPEG4:
2098         if(CONFIG_MPEG4_ENCODER && s->partitioned_frame)
2099             ff_mpeg4_init_partitions(s);
2100         break;
2101     }
2102
2103     s->resync_mb_x=0;
2104     s->resync_mb_y=0;
2105     s->first_slice_line = 1;
2106     s->ptr_lastgob = s->pb.buf;
2107     for(mb_y= s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
2108 //    printf("row %d at %X\n", s->mb_y, (int)s);
2109         s->mb_x=0;
2110         s->mb_y= mb_y;
2111
2112         ff_set_qscale(s, s->qscale);
2113         ff_init_block_index(s);
2114
2115         for(mb_x=0; mb_x < s->mb_width; mb_x++) {
2116             int xy= mb_y*s->mb_stride + mb_x; // removed const, H261 needs to adjust this
2117             int mb_type= s->mb_type[xy];
2118 //            int d;
2119             int dmin= INT_MAX;
2120             int dir;
2121
2122             if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < MAX_MB_BYTES){
2123                 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
2124                 return -1;
2125             }
2126             if(s->data_partitioning){
2127                 if(   s->pb2   .buf_end - s->pb2   .buf - (put_bits_count(&s->    pb2)>>3) < MAX_MB_BYTES
2128                    || s->tex_pb.buf_end - s->tex_pb.buf - (put_bits_count(&s->tex_pb )>>3) < MAX_MB_BYTES){
2129                     av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
2130                     return -1;
2131                 }
2132             }
2133
2134             s->mb_x = mb_x;
2135             s->mb_y = mb_y;  // moved into loop, can get changed by H.261
2136             ff_update_block_index(s);
2137
2138             if(CONFIG_H261_ENCODER && s->codec_id == CODEC_ID_H261){
2139                 ff_h261_reorder_mb_index(s);
2140                 xy= s->mb_y*s->mb_stride + s->mb_x;
2141                 mb_type= s->mb_type[xy];
2142             }
2143
2144             /* write gob / video packet header  */
2145             if(s->rtp_mode){
2146                 int current_packet_size, is_gob_start;
2147
2148                 current_packet_size= ((put_bits_count(&s->pb)+7)>>3) - (s->ptr_lastgob - s->pb.buf);
2149
2150                 is_gob_start= s->avctx->rtp_payload_size && current_packet_size >= s->avctx->rtp_payload_size && mb_y + mb_x>0;
2151
2152                 if(s->start_mb_y == mb_y && mb_y > 0 && mb_x==0) is_gob_start=1;
2153
2154                 switch(s->codec_id){
2155                 case CODEC_ID_H263:
2156                 case CODEC_ID_H263P:
2157                     if(!s->h263_slice_structured)
2158                         if(s->mb_x || s->mb_y%s->gob_index) is_gob_start=0;
2159                     break;
2160                 case CODEC_ID_MPEG2VIDEO:
2161                     if(s->mb_x==0 && s->mb_y!=0) is_gob_start=1;
2162                 case CODEC_ID_MPEG1VIDEO:
2163                     if(s->mb_skip_run) is_gob_start=0;
2164                     break;
2165                 }
2166
2167                 if(is_gob_start){
2168                     if(s->start_mb_y != mb_y || mb_x!=0){
2169                         write_slice_end(s);
2170
2171                         if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4 && s->partitioned_frame){
2172                             ff_mpeg4_init_partitions(s);
2173                         }
2174                     }
2175
2176                     assert((put_bits_count(&s->pb)&7) == 0);
2177                     current_packet_size= put_bits_ptr(&s->pb) - s->ptr_lastgob;
2178
2179                     if(s->avctx->error_rate && s->resync_mb_x + s->resync_mb_y > 0){
2180                         int r= put_bits_count(&s->pb)/8 + s->picture_number + 16 + s->mb_x + s->mb_y;
2181                         int d= 100 / s->avctx->error_rate;
2182                         if(r % d == 0){
2183                             current_packet_size=0;
2184                             s->pb.buf_ptr= s->ptr_lastgob;
2185                             assert(put_bits_ptr(&s->pb) == s->ptr_lastgob);
2186                         }
2187                     }
2188
2189                     if (s->avctx->rtp_callback){
2190                         int number_mb = (mb_y - s->resync_mb_y)*s->mb_width + mb_x - s->resync_mb_x;
2191                         s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, current_packet_size, number_mb);
2192                     }
2193
2194                     switch(s->codec_id){
2195                     case CODEC_ID_MPEG4:
2196                         if (CONFIG_MPEG4_ENCODER) {
2197                             ff_mpeg4_encode_video_packet_header(s);
2198                             ff_mpeg4_clean_buffers(s);
2199                         }
2200                     break;
2201                     case CODEC_ID_MPEG1VIDEO:
2202                     case CODEC_ID_MPEG2VIDEO:
2203                         if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER) {
2204                             ff_mpeg1_encode_slice_header(s);
2205                             ff_mpeg1_clean_buffers(s);
2206                         }
2207                     break;
2208                     case CODEC_ID_H263:
2209                     case CODEC_ID_H263P:
2210                         if (CONFIG_H263_ENCODER)
2211                             h263_encode_gob_header(s, mb_y);
2212                     break;
2213                     }
2214
2215                     if(s->flags&CODEC_FLAG_PASS1){
2216                         int bits= put_bits_count(&s->pb);
2217                         s->misc_bits+= bits - s->last_bits;
2218                         s->last_bits= bits;
2219                     }
2220
2221                     s->ptr_lastgob += current_packet_size;
2222                     s->first_slice_line=1;
2223                     s->resync_mb_x=mb_x;
2224                     s->resync_mb_y=mb_y;
2225                 }
2226             }
2227
2228             if(  (s->resync_mb_x   == s->mb_x)
2229                && s->resync_mb_y+1 == s->mb_y){
2230                 s->first_slice_line=0;
2231             }
2232
2233             s->mb_skipped=0;
2234             s->dquant=0; //only for QP_RD
2235
2236             if(mb_type & (mb_type-1) || (s->flags & CODEC_FLAG_QP_RD)){ // more than 1 MB type possible or CODEC_FLAG_QP_RD
2237                 int next_block=0;
2238                 int pb_bits_count, pb2_bits_count, tex_pb_bits_count;
2239
2240                 copy_context_before_encode(&backup_s, s, -1);
2241                 backup_s.pb= s->pb;
2242                 best_s.data_partitioning= s->data_partitioning;
2243                 best_s.partitioned_frame= s->partitioned_frame;
2244                 if(s->data_partitioning){
2245                     backup_s.pb2= s->pb2;
2246                     backup_s.tex_pb= s->tex_pb;
2247                 }
2248
2249                 if(mb_type&CANDIDATE_MB_TYPE_INTER){
2250                     s->mv_dir = MV_DIR_FORWARD;
2251                     s->mv_type = MV_TYPE_16X16;
2252                     s->mb_intra= 0;
2253                     s->mv[0][0][0] = s->p_mv_table[xy][0];
2254                     s->mv[0][0][1] = s->p_mv_table[xy][1];
2255                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER, pb, pb2, tex_pb,
2256                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2257                 }
2258                 if(mb_type&CANDIDATE_MB_TYPE_INTER_I){
2259                     s->mv_dir = MV_DIR_FORWARD;
2260                     s->mv_type = MV_TYPE_FIELD;
2261                     s->mb_intra= 0;
2262                     for(i=0; i<2; i++){
2263                         j= s->field_select[0][i] = s->p_field_select_table[i][xy];
2264                         s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
2265                         s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
2266                     }
2267                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER_I, pb, pb2, tex_pb,
2268                                  &dmin, &next_block, 0, 0);
2269                 }
2270                 if(mb_type&CANDIDATE_MB_TYPE_SKIPPED){
2271                     s->mv_dir = MV_DIR_FORWARD;
2272                     s->mv_type = MV_TYPE_16X16;
2273                     s->mb_intra= 0;
2274                     s->mv[0][0][0] = 0;
2275                     s->mv[0][0][1] = 0;
2276                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_SKIPPED, pb, pb2, tex_pb,
2277                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2278                 }
2279                 if(mb_type&CANDIDATE_MB_TYPE_INTER4V){
2280                     s->mv_dir = MV_DIR_FORWARD;
2281                     s->mv_type = MV_TYPE_8X8;
2282                     s->mb_intra= 0;
2283                     for(i=0; i<4; i++){
2284                         s->mv[0][i][0] = s->current_picture.f.motion_val[0][s->block_index[i]][0];
2285                         s->mv[0][i][1] = s->current_picture.f.motion_val[0][s->block_index[i]][1];
2286                     }
2287                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER4V, pb, pb2, tex_pb,
2288                                  &dmin, &next_block, 0, 0);
2289                 }
2290                 if(mb_type&CANDIDATE_MB_TYPE_FORWARD){
2291                     s->mv_dir = MV_DIR_FORWARD;
2292                     s->mv_type = MV_TYPE_16X16;
2293                     s->mb_intra= 0;
2294                     s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
2295                     s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2296                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD, pb, pb2, tex_pb,
2297                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2298                 }
2299                 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD){
2300                     s->mv_dir = MV_DIR_BACKWARD;
2301                     s->mv_type = MV_TYPE_16X16;
2302                     s->mb_intra= 0;
2303                     s->mv[1][0][0] = s->b_back_mv_table[xy][0];
2304                     s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2305                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD, pb, pb2, tex_pb,
2306                                  &dmin, &next_block, s->mv[1][0][0], s->mv[1][0][1]);
2307                 }
2308                 if(mb_type&CANDIDATE_MB_TYPE_BIDIR){
2309                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2310                     s->mv_type = MV_TYPE_16X16;
2311                     s->mb_intra= 0;
2312                     s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
2313                     s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
2314                     s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
2315                     s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2316                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR, pb, pb2, tex_pb,
2317                                  &dmin, &next_block, 0, 0);
2318                 }
2319                 if(mb_type&CANDIDATE_MB_TYPE_FORWARD_I){
2320                     s->mv_dir = MV_DIR_FORWARD;
2321                     s->mv_type = MV_TYPE_FIELD;
2322                     s->mb_intra= 0;
2323                     for(i=0; i<2; i++){
2324                         j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
2325                         s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
2326                         s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
2327                     }
2328                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD_I, pb, pb2, tex_pb,
2329                                  &dmin, &next_block, 0, 0);
2330                 }
2331                 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD_I){
2332                     s->mv_dir = MV_DIR_BACKWARD;
2333                     s->mv_type = MV_TYPE_FIELD;
2334                     s->mb_intra= 0;
2335                     for(i=0; i<2; i++){
2336                         j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
2337                         s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
2338                         s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
2339                     }
2340                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD_I, pb, pb2, tex_pb,
2341                                  &dmin, &next_block, 0, 0);
2342                 }
2343                 if(mb_type&CANDIDATE_MB_TYPE_BIDIR_I){
2344                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2345                     s->mv_type = MV_TYPE_FIELD;
2346                     s->mb_intra= 0;
2347                     for(dir=0; dir<2; dir++){
2348                         for(i=0; i<2; i++){
2349                             j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
2350                             s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
2351                             s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
2352                         }
2353                     }
2354                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR_I, pb, pb2, tex_pb,
2355                                  &dmin, &next_block, 0, 0);
2356                 }
2357                 if(mb_type&CANDIDATE_MB_TYPE_INTRA){
2358                     s->mv_dir = 0;
2359                     s->mv_type = MV_TYPE_16X16;
2360                     s->mb_intra= 1;
2361                     s->mv[0][0][0] = 0;
2362                     s->mv[0][0][1] = 0;
2363                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTRA, pb, pb2, tex_pb,
2364                                  &dmin, &next_block, 0, 0);
2365                     if(s->h263_pred || s->h263_aic){
2366                         if(best_s.mb_intra)
2367                             s->mbintra_table[mb_x + mb_y*s->mb_stride]=1;
2368                         else
2369                             ff_clean_intra_table_entries(s); //old mode?
2370                     }
2371                 }
2372
2373                 if((s->flags & CODEC_FLAG_QP_RD) && dmin < INT_MAX){
2374                     if(best_s.mv_type==MV_TYPE_16X16){ //FIXME move 4mv after QPRD
2375                         const int last_qp= backup_s.qscale;
2376                         int qpi, qp, dc[6];
2377                         DCTELEM ac[6][16];
2378                         const int mvdir= (best_s.mv_dir&MV_DIR_BACKWARD) ? 1 : 0;
2379                         static const int dquant_tab[4]={-1,1,-2,2};
2380
2381                         assert(backup_s.dquant == 0);
2382
2383                         //FIXME intra
2384                         s->mv_dir= best_s.mv_dir;
2385                         s->mv_type = MV_TYPE_16X16;
2386                         s->mb_intra= best_s.mb_intra;
2387                         s->mv[0][0][0] = best_s.mv[0][0][0];
2388                         s->mv[0][0][1] = best_s.mv[0][0][1];
2389                         s->mv[1][0][0] = best_s.mv[1][0][0];
2390                         s->mv[1][0][1] = best_s.mv[1][0][1];
2391
2392                         qpi = s->pict_type == AV_PICTURE_TYPE_B ? 2 : 0;
2393                         for(; qpi<4; qpi++){
2394                             int dquant= dquant_tab[qpi];
2395                             qp= last_qp + dquant;
2396                             if(qp < s->avctx->qmin || qp > s->avctx->qmax)
2397                                 continue;
2398                             backup_s.dquant= dquant;
2399                             if(s->mb_intra && s->dc_val[0]){
2400                                 for(i=0; i<6; i++){
2401                                     dc[i]= s->dc_val[0][ s->block_index[i] ];
2402                                     memcpy(ac[i], s->ac_val[0][s->block_index[i]], sizeof(DCTELEM)*16);
2403                                 }
2404                             }
2405
2406                             encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
2407                                          &dmin, &next_block, s->mv[mvdir][0][0], s->mv[mvdir][0][1]);
2408                             if(best_s.qscale != qp){
2409                                 if(s->mb_intra && s->dc_val[0]){
2410                                     for(i=0; i<6; i++){
2411                                         s->dc_val[0][ s->block_index[i] ]= dc[i];
2412                                         memcpy(s->ac_val[0][s->block_index[i]], ac[i], sizeof(DCTELEM)*16);
2413                                     }
2414                                 }
2415                             }
2416                         }
2417                     }
2418                 }
2419                 if(CONFIG_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT){
2420                     int mx= s->b_direct_mv_table[xy][0];
2421                     int my= s->b_direct_mv_table[xy][1];
2422
2423                     backup_s.dquant = 0;
2424                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2425                     s->mb_intra= 0;
2426                     ff_mpeg4_set_direct_mv(s, mx, my);
2427                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
2428                                  &dmin, &next_block, mx, my);
2429                 }
2430                 if(CONFIG_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT0){
2431                     backup_s.dquant = 0;
2432                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2433                     s->mb_intra= 0;
2434                     ff_mpeg4_set_direct_mv(s, 0, 0);
2435                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
2436                                  &dmin, &next_block, 0, 0);
2437                 }
2438                 if(!best_s.mb_intra && s->flags2&CODEC_FLAG2_SKIP_RD){
2439                     int coded=0;
2440                     for(i=0; i<6; i++)
2441                         coded |= s->block_last_index[i];
2442                     if(coded){
2443                         int mx,my;
2444                         memcpy(s->mv, best_s.mv, sizeof(s->mv));
2445                         if(CONFIG_MPEG4_ENCODER && best_s.mv_dir & MV_DIRECT){
2446                             mx=my=0; //FIXME find the one we actually used
2447                             ff_mpeg4_set_direct_mv(s, mx, my);
2448                         }else if(best_s.mv_dir&MV_DIR_BACKWARD){
2449                             mx= s->mv[1][0][0];
2450                             my= s->mv[1][0][1];
2451                         }else{
2452                             mx= s->mv[0][0][0];
2453                             my= s->mv[0][0][1];
2454                         }
2455
2456                         s->mv_dir= best_s.mv_dir;
2457                         s->mv_type = best_s.mv_type;
2458                         s->mb_intra= 0;
2459 /*                        s->mv[0][0][0] = best_s.mv[0][0][0];
2460                         s->mv[0][0][1] = best_s.mv[0][0][1];
2461                         s->mv[1][0][0] = best_s.mv[1][0][0];
2462                         s->mv[1][0][1] = best_s.mv[1][0][1];*/
2463                         backup_s.dquant= 0;
2464                         s->skipdct=1;
2465                         encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
2466                                         &dmin, &next_block, mx, my);
2467                         s->skipdct=0;
2468                     }
2469                 }
2470
2471                 s->current_picture.f.qscale_table[xy] = best_s.qscale;
2472
2473                 copy_context_after_encode(s, &best_s, -1);
2474
2475                 pb_bits_count= put_bits_count(&s->pb);
2476                 flush_put_bits(&s->pb);
2477                 ff_copy_bits(&backup_s.pb, bit_buf[next_block^1], pb_bits_count);
2478                 s->pb= backup_s.pb;
2479
2480                 if(s->data_partitioning){
2481                     pb2_bits_count= put_bits_count(&s->pb2);
2482                     flush_put_bits(&s->pb2);
2483                     ff_copy_bits(&backup_s.pb2, bit_buf2[next_block^1], pb2_bits_count);
2484                     s->pb2= backup_s.pb2;
2485
2486                     tex_pb_bits_count= put_bits_count(&s->tex_pb);
2487                     flush_put_bits(&s->tex_pb);
2488                     ff_copy_bits(&backup_s.tex_pb, bit_buf_tex[next_block^1], tex_pb_bits_count);
2489                     s->tex_pb= backup_s.tex_pb;
2490                 }
2491                 s->last_bits= put_bits_count(&s->pb);
2492
2493                 if (CONFIG_H263_ENCODER &&
2494                     s->out_format == FMT_H263 && s->pict_type!=AV_PICTURE_TYPE_B)
2495                     ff_h263_update_motion_val(s);
2496
2497                 if(next_block==0){ //FIXME 16 vs linesize16
2498                     s->dsp.put_pixels_tab[0][0](s->dest[0], s->rd_scratchpad                     , s->linesize  ,16);
2499                     s->dsp.put_pixels_tab[1][0](s->dest[1], s->rd_scratchpad + 16*s->linesize    , s->uvlinesize, 8);
2500                     s->dsp.put_pixels_tab[1][0](s->dest[2], s->rd_scratchpad + 16*s->linesize + 8, s->uvlinesize, 8);
2501                 }
2502
2503                 if(s->avctx->mb_decision == FF_MB_DECISION_BITS)
2504                     MPV_decode_mb(s, s->block);
2505             } else {
2506                 int motion_x = 0, motion_y = 0;
2507                 s->mv_type=MV_TYPE_16X16;
2508                 // only one MB-Type possible
2509
2510                 switch(mb_type){
2511                 case CANDIDATE_MB_TYPE_INTRA:
2512                     s->mv_dir = 0;
2513                     s->mb_intra= 1;
2514                     motion_x= s->mv[0][0][0] = 0;
2515                     motion_y= s->mv[0][0][1] = 0;
2516                     break;
2517                 case CANDIDATE_MB_TYPE_INTER:
2518                     s->mv_dir = MV_DIR_FORWARD;
2519                     s->mb_intra= 0;
2520                     motion_x= s->mv[0][0][0] = s->p_mv_table[xy][0];
2521                     motion_y= s->mv[0][0][1] = s->p_mv_table[xy][1];
2522                     break;
2523                 case CANDIDATE_MB_TYPE_INTER_I:
2524                     s->mv_dir = MV_DIR_FORWARD;
2525                     s->mv_type = MV_TYPE_FIELD;
2526                     s->mb_intra= 0;
2527                     for(i=0; i<2; i++){
2528                         j= s->field_select[0][i] = s->p_field_select_table[i][xy];
2529                         s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
2530                         s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
2531                     }
2532                     break;
2533                 case CANDIDATE_MB_TYPE_INTER4V:
2534                     s->mv_dir = MV_DIR_FORWARD;
2535                     s->mv_type = MV_TYPE_8X8;
2536                     s->mb_intra= 0;
2537                     for(i=0; i<4; i++){
2538                         s->mv[0][i][0] = s->current_picture.f.motion_val[0][s->block_index[i]][0];
2539                         s->mv[0][i][1] = s->current_picture.f.motion_val[0][s->block_index[i]][1];
2540                     }
2541                     break;
2542                 case CANDIDATE_MB_TYPE_DIRECT:
2543                     if (CONFIG_MPEG4_ENCODER) {
2544                         s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
2545                         s->mb_intra= 0;
2546                         motion_x=s->b_direct_mv_table[xy][0];
2547                         motion_y=s->b_direct_mv_table[xy][1];
2548                         ff_mpeg4_set_direct_mv(s, motion_x, motion_y);
2549                     }
2550                     break;
2551                 case CANDIDATE_MB_TYPE_DIRECT0:
2552                     if (CONFIG_MPEG4_ENCODER) {
2553                         s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
2554                         s->mb_intra= 0;
2555                         ff_mpeg4_set_direct_mv(s, 0, 0);
2556                     }
2557                     break;
2558                 case CANDIDATE_MB_TYPE_BIDIR:
2559                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2560                     s->mb_intra= 0;
2561                     s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
2562                     s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
2563                     s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
2564                     s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2565                     break;
2566                 case CANDIDATE_MB_TYPE_BACKWARD:
2567                     s->mv_dir = MV_DIR_BACKWARD;
2568                     s->mb_intra= 0;
2569                     motion_x= s->mv[1][0][0] = s->b_back_mv_table[xy][0];
2570                     motion_y= s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2571                     break;
2572                 case CANDIDATE_MB_TYPE_FORWARD:
2573                     s->mv_dir = MV_DIR_FORWARD;
2574                     s->mb_intra= 0;
2575                     motion_x= s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
2576                     motion_y= s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2577 //                    printf(" %d %d ", motion_x, motion_y);
2578                     break;
2579                 case CANDIDATE_MB_TYPE_FORWARD_I:
2580                     s->mv_dir = MV_DIR_FORWARD;
2581                     s->mv_type = MV_TYPE_FIELD;
2582                     s->mb_intra= 0;
2583                     for(i=0; i<2; i++){
2584                         j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
2585                         s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
2586                         s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
2587                     }
2588                     break;
2589                 case CANDIDATE_MB_TYPE_BACKWARD_I:
2590                     s->mv_dir = MV_DIR_BACKWARD;
2591                     s->mv_type = MV_TYPE_FIELD;
2592                     s->mb_intra= 0;
2593                     for(i=0; i<2; i++){
2594                         j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
2595                         s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
2596                         s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
2597                     }
2598                     break;
2599                 case CANDIDATE_MB_TYPE_BIDIR_I:
2600                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2601                     s->mv_type = MV_TYPE_FIELD;
2602                     s->mb_intra= 0;
2603                     for(dir=0; dir<2; dir++){
2604                         for(i=0; i<2; i++){
2605                             j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
2606                             s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
2607                             s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
2608                         }
2609                     }
2610                     break;
2611                 default:
2612                     av_log(s->avctx, AV_LOG_ERROR, "illegal MB type\n");
2613                 }
2614
2615                 encode_mb(s, motion_x, motion_y);
2616
2617                 // RAL: Update last macroblock type
2618                 s->last_mv_dir = s->mv_dir;
2619
2620                 if (CONFIG_H263_ENCODER &&
2621                     s->out_format == FMT_H263 && s->pict_type!=AV_PICTURE_TYPE_B)
2622                     ff_h263_update_motion_val(s);
2623
2624                 MPV_decode_mb(s, s->block);
2625             }
2626
2627             /* clean the MV table in IPS frames for direct mode in B frames */
2628             if(s->mb_intra /* && I,P,S_TYPE */){
2629                 s->p_mv_table[xy][0]=0;
2630                 s->p_mv_table[xy][1]=0;
2631             }
2632
2633             if(s->flags&CODEC_FLAG_PSNR){
2634                 int w= 16;
2635                 int h= 16;
2636
2637                 if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
2638                 if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
2639
2640                 s->current_picture.f.error[0] += sse(
2641                     s, s->new_picture.f.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16,
2642                     s->dest[0], w, h, s->linesize);
2643                 s->current_picture.f.error[1] += sse(
2644                     s, s->new_picture.f.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*chr_h,
2645                     s->dest[1], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
2646                 s->current_picture.f.error[2] += sse(
2647                     s, s->new_picture.f.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*chr_h,
2648                     s->dest[2], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
2649             }
2650             if(s->loop_filter){
2651                 if(CONFIG_H263_ENCODER && s->out_format == FMT_H263)
2652                     ff_h263_loop_filter(s);
2653             }
2654 //printf("MB %d %d bits\n", s->mb_x+s->mb_y*s->mb_stride, put_bits_count(&s->pb));
2655         }
2656     }
2657
2658     //not beautiful here but we must write it before flushing so it has to be here
2659     if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version && s->msmpeg4_version<4 && s->pict_type == AV_PICTURE_TYPE_I)
2660         msmpeg4_encode_ext_header(s);
2661
2662     write_slice_end(s);
2663
2664     /* Send the last GOB if RTP */
2665     if (s->avctx->rtp_callback) {
2666         int number_mb = (mb_y - s->resync_mb_y)*s->mb_width - s->resync_mb_x;
2667         pdif = put_bits_ptr(&s->pb) - s->ptr_lastgob;
2668         /* Call the RTP callback to send the last GOB */
2669         emms_c();
2670         s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, pdif, number_mb);
2671     }
2672
2673     return 0;
2674 }
2675
2676 #define MERGE(field) dst->field += src->field; src->field=0
2677 static void merge_context_after_me(MpegEncContext *dst, MpegEncContext *src){
2678     MERGE(me.scene_change_score);
2679     MERGE(me.mc_mb_var_sum_temp);
2680     MERGE(me.mb_var_sum_temp);
2681 }
2682
2683 static void merge_context_after_encode(MpegEncContext *dst, MpegEncContext *src){
2684     int i;
2685
2686     MERGE(dct_count[0]); //note, the other dct vars are not part of the context
2687     MERGE(dct_count[1]);
2688     MERGE(mv_bits);
2689     MERGE(i_tex_bits);
2690     MERGE(p_tex_bits);
2691     MERGE(i_count);
2692     MERGE(f_count);
2693     MERGE(b_count);
2694     MERGE(skip_count);
2695     MERGE(misc_bits);
2696     MERGE(error_count);
2697     MERGE(padding_bug_score);
2698     MERGE(current_picture.f.error[0]);
2699     MERGE(current_picture.f.error[1]);
2700     MERGE(current_picture.f.error[2]);
2701
2702     if(dst->avctx->noise_reduction){
2703         for(i=0; i<64; i++){
2704             MERGE(dct_error_sum[0][i]);
2705             MERGE(dct_error_sum[1][i]);
2706         }
2707     }
2708
2709     assert(put_bits_count(&src->pb) % 8 ==0);
2710     assert(put_bits_count(&dst->pb) % 8 ==0);
2711     ff_copy_bits(&dst->pb, src->pb.buf, put_bits_count(&src->pb));
2712     flush_put_bits(&dst->pb);
2713 }
2714
2715 static int estimate_qp(MpegEncContext *s, int dry_run){
2716     if (s->next_lambda){
2717         s->current_picture_ptr->f.quality =
2718         s->current_picture.f.quality = s->next_lambda;
2719         if(!dry_run) s->next_lambda= 0;
2720     } else if (!s->fixed_qscale) {
2721         s->current_picture_ptr->f.quality =
2722         s->current_picture.f.quality = ff_rate_estimate_qscale(s, dry_run);
2723         if (s->current_picture.f.quality < 0)
2724             return -1;
2725     }
2726
2727     if(s->adaptive_quant){
2728         switch(s->codec_id){
2729         case CODEC_ID_MPEG4:
2730             if (CONFIG_MPEG4_ENCODER)
2731                 ff_clean_mpeg4_qscales(s);
2732             break;
2733         case CODEC_ID_H263:
2734         case CODEC_ID_H263P:
2735         case CODEC_ID_FLV1:
2736             if (CONFIG_H263_ENCODER)
2737                 ff_clean_h263_qscales(s);
2738             break;
2739         default:
2740             ff_init_qscale_tab(s);
2741         }
2742
2743         s->lambda= s->lambda_table[0];
2744         //FIXME broken
2745     }else
2746         s->lambda = s->current_picture.f.quality;
2747 //printf("%d %d\n", s->avctx->global_quality, s->current_picture.quality);
2748     update_qscale(s);
2749     return 0;
2750 }
2751
2752 /* must be called before writing the header */
2753 static void set_frame_distances(MpegEncContext * s){
2754     assert(s->current_picture_ptr->pts != AV_NOPTS_VALUE);
2755     s->time = s->current_picture_ptr->f.pts * s->avctx->time_base.num;
2756
2757     if(s->pict_type==AV_PICTURE_TYPE_B){
2758         s->pb_time= s->pp_time - (s->last_non_b_time - s->time);
2759         assert(s->pb_time > 0 && s->pb_time < s->pp_time);
2760     }else{
2761         s->pp_time= s->time - s->last_non_b_time;
2762         s->last_non_b_time= s->time;
2763         assert(s->picture_number==0 || s->pp_time > 0);
2764     }
2765 }
2766
2767 static int encode_picture(MpegEncContext *s, int picture_number)
2768 {
2769     int i;
2770     int bits;
2771     int context_count = s->avctx->thread_count;
2772
2773     s->picture_number = picture_number;
2774
2775     /* Reset the average MB variance */
2776     s->me.mb_var_sum_temp    =
2777     s->me.mc_mb_var_sum_temp = 0;
2778
2779     /* we need to initialize some time vars before we can encode b-frames */
2780     // RAL: Condition added for MPEG1VIDEO
2781     if (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO || (s->h263_pred && !s->msmpeg4_version))
2782         set_frame_distances(s);
2783     if(CONFIG_MPEG4_ENCODER && s->codec_id == CODEC_ID_MPEG4)
2784         ff_set_mpeg4_time(s);
2785
2786     s->me.scene_change_score=0;
2787
2788 //    s->lambda= s->current_picture_ptr->quality; //FIXME qscale / ... stuff for ME rate distortion
2789
2790     if(s->pict_type==AV_PICTURE_TYPE_I){
2791         if(s->msmpeg4_version >= 3) s->no_rounding=1;
2792         else                        s->no_rounding=0;
2793     }else if(s->pict_type!=AV_PICTURE_TYPE_B){
2794         if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
2795             s->no_rounding ^= 1;
2796     }
2797
2798     if(s->flags & CODEC_FLAG_PASS2){
2799         if (estimate_qp(s,1) < 0)
2800             return -1;
2801         ff_get_2pass_fcode(s);
2802     }else if(!(s->flags & CODEC_FLAG_QSCALE)){
2803         if(s->pict_type==AV_PICTURE_TYPE_B)
2804             s->lambda= s->last_lambda_for[s->pict_type];
2805         else
2806             s->lambda= s->last_lambda_for[s->last_non_b_pict_type];
2807         update_qscale(s);
2808     }
2809
2810     s->mb_intra=0; //for the rate distortion & bit compare functions
2811     for(i=1; i<context_count; i++){
2812         ff_update_duplicate_context(s->thread_context[i], s);
2813     }
2814
2815     if(ff_init_me(s)<0)
2816         return -1;
2817
2818     /* Estimate motion for every MB */
2819     if(s->pict_type != AV_PICTURE_TYPE_I){
2820         s->lambda = (s->lambda * s->avctx->me_penalty_compensation + 128)>>8;
2821         s->lambda2= (s->lambda2* (int64_t)s->avctx->me_penalty_compensation + 128)>>8;
2822         if(s->pict_type != AV_PICTURE_TYPE_B && s->avctx->me_threshold==0){
2823             if((s->avctx->pre_me && s->last_non_b_pict_type==AV_PICTURE_TYPE_I) || s->avctx->pre_me==2){
2824                 s->avctx->execute(s->avctx, pre_estimate_motion_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
2825             }
2826         }
2827
2828         s->avctx->execute(s->avctx, estimate_motion_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
2829     }else /* if(s->pict_type == AV_PICTURE_TYPE_I) */{
2830         /* I-Frame */
2831         for(i=0; i<s->mb_stride*s->mb_height; i++)
2832             s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
2833
2834         if(!s->fixed_qscale){
2835             /* finding spatial complexity for I-frame rate control */
2836             s->avctx->execute(s->avctx, mb_var_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
2837         }
2838     }
2839     for(i=1; i<context_count; i++){
2840         merge_context_after_me(s, s->thread_context[i]);
2841     }
2842     s->current_picture.mc_mb_var_sum= s->current_picture_ptr->mc_mb_var_sum= s->me.mc_mb_var_sum_temp;
2843     s->current_picture.   mb_var_sum= s->current_picture_ptr->   mb_var_sum= s->me.   mb_var_sum_temp;
2844     emms_c();
2845
2846     if(s->me.scene_change_score > s->avctx->scenechange_threshold && s->pict_type == AV_PICTURE_TYPE_P){
2847         s->pict_type= AV_PICTURE_TYPE_I;
2848         for(i=0; i<s->mb_stride*s->mb_height; i++)
2849             s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
2850 //printf("Scene change detected, encoding as I Frame %d %d\n", s->current_picture.mb_var_sum, s->current_picture.mc_mb_var_sum);
2851     }
2852
2853     if(!s->umvplus){
2854         if(s->pict_type==AV_PICTURE_TYPE_P || s->pict_type==AV_PICTURE_TYPE_S) {
2855             s->f_code= ff_get_best_fcode(s, s->p_mv_table, CANDIDATE_MB_TYPE_INTER);
2856
2857             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2858                 int a,b;
2859                 a= ff_get_best_fcode(s, s->p_field_mv_table[0][0], CANDIDATE_MB_TYPE_INTER_I); //FIXME field_select
2860                 b= ff_get_best_fcode(s, s->p_field_mv_table[1][1], CANDIDATE_MB_TYPE_INTER_I);
2861                 s->f_code= FFMAX3(s->f_code, a, b);
2862             }
2863
2864             ff_fix_long_p_mvs(s);
2865             ff_fix_long_mvs(s, NULL, 0, s->p_mv_table, s->f_code, CANDIDATE_MB_TYPE_INTER, 0);
2866             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2867                 int j;
2868                 for(i=0; i<2; i++){
2869                     for(j=0; j<2; j++)
2870                         ff_fix_long_mvs(s, s->p_field_select_table[i], j,
2871                                         s->p_field_mv_table[i][j], s->f_code, CANDIDATE_MB_TYPE_INTER_I, 0);
2872                 }
2873             }
2874         }
2875
2876         if(s->pict_type==AV_PICTURE_TYPE_B){
2877             int a, b;
2878
2879             a = ff_get_best_fcode(s, s->b_forw_mv_table, CANDIDATE_MB_TYPE_FORWARD);
2880             b = ff_get_best_fcode(s, s->b_bidir_forw_mv_table, CANDIDATE_MB_TYPE_BIDIR);
2881             s->f_code = FFMAX(a, b);
2882
2883             a = ff_get_best_fcode(s, s->b_back_mv_table, CANDIDATE_MB_TYPE_BACKWARD);
2884             b = ff_get_best_fcode(s, s->b_bidir_back_mv_table, CANDIDATE_MB_TYPE_BIDIR);
2885             s->b_code = FFMAX(a, b);
2886
2887             ff_fix_long_mvs(s, NULL, 0, s->b_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_FORWARD, 1);
2888             ff_fix_long_mvs(s, NULL, 0, s->b_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BACKWARD, 1);
2889             ff_fix_long_mvs(s, NULL, 0, s->b_bidir_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_BIDIR, 1);
2890             ff_fix_long_mvs(s, NULL, 0, s->b_bidir_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BIDIR, 1);
2891             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2892                 int dir, j;
2893                 for(dir=0; dir<2; dir++){
2894                     for(i=0; i<2; i++){
2895                         for(j=0; j<2; j++){
2896                             int type= dir ? (CANDIDATE_MB_TYPE_BACKWARD_I|CANDIDATE_MB_TYPE_BIDIR_I)
2897                                           : (CANDIDATE_MB_TYPE_FORWARD_I |CANDIDATE_MB_TYPE_BIDIR_I);
2898                             ff_fix_long_mvs(s, s->b_field_select_table[dir][i], j,
2899                                             s->b_field_mv_table[dir][i][j], dir ? s->b_code : s->f_code, type, 1);
2900                         }
2901                     }
2902                 }
2903             }
2904         }
2905     }
2906
2907     if (estimate_qp(s, 0) < 0)
2908         return -1;
2909
2910     if(s->qscale < 3 && s->max_qcoeff<=128 && s->pict_type==AV_PICTURE_TYPE_I && !(s->flags & CODEC_FLAG_QSCALE))
2911         s->qscale= 3; //reduce clipping problems
2912
2913     if (s->out_format == FMT_MJPEG) {
2914         /* for mjpeg, we do include qscale in the matrix */
2915         for(i=1;i<64;i++){
2916             int j= s->dsp.idct_permutation[i];
2917
2918             s->intra_matrix[j] = av_clip_uint8((ff_mpeg1_default_intra_matrix[i] * s->qscale) >> 3);
2919         }
2920         s->y_dc_scale_table=
2921         s->c_dc_scale_table= ff_mpeg2_dc_scale_table[s->intra_dc_precision];
2922         s->intra_matrix[0] = ff_mpeg2_dc_scale_table[s->intra_dc_precision][8];
2923         ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
2924                        s->intra_matrix, s->intra_quant_bias, 8, 8, 1);
2925         s->qscale= 8;
2926     }
2927
2928     //FIXME var duplication
2929     s->current_picture_ptr->f.key_frame =
2930     s->current_picture.f.key_frame = s->pict_type == AV_PICTURE_TYPE_I; //FIXME pic_ptr
2931     s->current_picture_ptr->f.pict_type =
2932     s->current_picture.f.pict_type = s->pict_type;
2933
2934     if (s->current_picture.f.key_frame)
2935         s->picture_in_gop_number=0;
2936
2937     s->last_bits= put_bits_count(&s->pb);
2938     switch(s->out_format) {
2939     case FMT_MJPEG:
2940         if (CONFIG_MJPEG_ENCODER)
2941             ff_mjpeg_encode_picture_header(s);
2942         break;
2943     case FMT_H261:
2944         if (CONFIG_H261_ENCODER)
2945             ff_h261_encode_picture_header(s, picture_number);
2946         break;
2947     case FMT_H263:
2948         if (CONFIG_WMV2_ENCODER && s->codec_id == CODEC_ID_WMV2)
2949             ff_wmv2_encode_picture_header(s, picture_number);
2950         else if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version)
2951             msmpeg4_encode_picture_header(s, picture_number);
2952         else if (CONFIG_MPEG4_ENCODER && s->h263_pred)
2953             mpeg4_encode_picture_header(s, picture_number);
2954         else if (CONFIG_RV10_ENCODER && s->codec_id == CODEC_ID_RV10)
2955             rv10_encode_picture_header(s, picture_number);
2956         else if (CONFIG_RV20_ENCODER && s->codec_id == CODEC_ID_RV20)
2957             rv20_encode_picture_header(s, picture_number);
2958         else if (CONFIG_FLV_ENCODER && s->codec_id == CODEC_ID_FLV1)
2959             ff_flv_encode_picture_header(s, picture_number);
2960         else if (CONFIG_H263_ENCODER)
2961             h263_encode_picture_header(s, picture_number);
2962         break;
2963     case FMT_MPEG1:
2964         if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
2965             mpeg1_encode_picture_header(s, picture_number);
2966         break;
2967     case FMT_H264:
2968         break;
2969     default:
2970         assert(0);
2971     }
2972     bits= put_bits_count(&s->pb);
2973     s->header_bits= bits - s->last_bits;
2974
2975     for(i=1; i<context_count; i++){
2976         update_duplicate_context_after_me(s->thread_context[i], s);
2977     }
2978     s->avctx->execute(s->avctx, encode_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
2979     for(i=1; i<context_count; i++){
2980         merge_context_after_encode(s, s->thread_context[i]);
2981     }
2982     emms_c();
2983     return 0;
2984 }
2985
2986 static void denoise_dct_c(MpegEncContext *s, DCTELEM *block){
2987     const int intra= s->mb_intra;
2988     int i;
2989
2990     s->dct_count[intra]++;
2991
2992     for(i=0; i<64; i++){
2993         int level= block[i];
2994
2995         if(level){
2996             if(level>0){
2997                 s->dct_error_sum[intra][i] += level;
2998                 level -= s->dct_offset[intra][i];
2999                 if(level<0) level=0;
3000             }else{
3001                 s->dct_error_sum[intra][i] -= level;
3002                 level += s->dct_offset[intra][i];
3003                 if(level>0) level=0;
3004             }
3005             block[i]= level;
3006         }
3007     }
3008 }
3009
3010 static int dct_quantize_trellis_c(MpegEncContext *s,
3011                                   DCTELEM *block, int n,
3012                                   int qscale, int *overflow){
3013     const int *qmat;
3014     const uint8_t *scantable= s->intra_scantable.scantable;
3015     const uint8_t *perm_scantable= s->intra_scantable.permutated;
3016     int max=0;
3017     unsigned int threshold1, threshold2;
3018     int bias=0;
3019     int run_tab[65];
3020     int level_tab[65];
3021     int score_tab[65];
3022     int survivor[65];
3023     int survivor_count;
3024     int last_run=0;
3025     int last_level=0;
3026     int last_score= 0;
3027     int last_i;
3028     int coeff[2][64];
3029     int coeff_count[64];
3030     int qmul, qadd, start_i, last_non_zero, i, dc;
3031     const int esc_length= s->ac_esc_length;
3032     uint8_t * length;
3033     uint8_t * last_length;
3034     const int lambda= s->lambda2 >> (FF_LAMBDA_SHIFT - 6);
3035
3036     s->dsp.fdct (block);
3037
3038     if(s->dct_error_sum)
3039         s->denoise_dct(s, block);
3040     qmul= qscale*16;
3041     qadd= ((qscale-1)|1)*8;
3042
3043     if (s->mb_intra) {
3044         int q;
3045         if (!s->h263_aic) {
3046             if (n < 4)
3047                 q = s->y_dc_scale;
3048             else
3049                 q = s->c_dc_scale;
3050             q = q << 3;
3051         } else{
3052             /* For AIC we skip quant/dequant of INTRADC */
3053             q = 1 << 3;
3054             qadd=0;
3055         }
3056
3057         /* note: block[0] is assumed to be positive */
3058         block[0] = (block[0] + (q >> 1)) / q;
3059         start_i = 1;
3060         last_non_zero = 0;
3061         qmat = s->q_intra_matrix[qscale];
3062         if(s->mpeg_quant || s->out_format == FMT_MPEG1)
3063             bias= 1<<(QMAT_SHIFT-1);
3064         length     = s->intra_ac_vlc_length;
3065         last_length= s->intra_ac_vlc_last_length;
3066     } else {
3067         start_i = 0;
3068         last_non_zero = -1;
3069         qmat = s->q_inter_matrix[qscale];
3070         length     = s->inter_ac_vlc_length;
3071         last_length= s->inter_ac_vlc_last_length;
3072     }
3073     last_i= start_i;
3074
3075     threshold1= (1<<QMAT_SHIFT) - bias - 1;
3076     threshold2= (threshold1<<1);
3077
3078     for(i=63; i>=start_i; i--) {
3079         const int j = scantable[i];
3080         int level = block[j] * qmat[j];
3081
3082         if(((unsigned)(level+threshold1))>threshold2){
3083             last_non_zero = i;
3084             break;
3085         }
3086     }
3087
3088     for(i=start_i; i<=last_non_zero; i++) {
3089         const int j = scantable[i];
3090         int level = block[j] * qmat[j];
3091
3092 //        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
3093 //           || bias-level >= (1<<(QMAT_SHIFT - 3))){
3094         if(((unsigned)(level+threshold1))>threshold2){
3095             if(level>0){
3096                 level= (bias + level)>>QMAT_SHIFT;
3097                 coeff[0][i]= level;
3098                 coeff[1][i]= level-1;
3099 //                coeff[2][k]= level-2;
3100             }else{
3101                 level= (bias - level)>>QMAT_SHIFT;
3102                 coeff[0][i]= -level;
3103                 coeff[1][i]= -level+1;
3104 //                coeff[2][k]= -level+2;
3105             }
3106             coeff_count[i]= FFMIN(level, 2);
3107             assert(coeff_count[i]);
3108             max |=level;
3109         }else{
3110             coeff[0][i]= (level>>31)|1;
3111             coeff_count[i]= 1;
3112         }
3113     }
3114
3115     *overflow= s->max_qcoeff < max; //overflow might have happened
3116
3117     if(last_non_zero < start_i){
3118         memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
3119         return last_non_zero;
3120     }
3121
3122     score_tab[start_i]= 0;
3123     survivor[0]= start_i;
3124     survivor_count= 1;
3125
3126     for(i=start_i; i<=last_non_zero; i++){
3127         int level_index, j, zero_distortion;
3128         int dct_coeff= FFABS(block[ scantable[i] ]);
3129         int best_score=256*256*256*120;
3130
3131         if (   s->dsp.fdct == fdct_ifast
3132 #ifndef FAAN_POSTSCALE
3133             || s->dsp.fdct == ff_faandct
3134 #endif
3135            )
3136             dct_coeff= (dct_coeff*ff_inv_aanscales[ scantable[i] ]) >> 12;
3137         zero_distortion= dct_coeff*dct_coeff;
3138
3139         for(level_index=0; level_index < coeff_count[i]; level_index++){
3140             int distortion;
3141             int level= coeff[level_index][i];
3142             const int alevel= FFABS(level);
3143             int unquant_coeff;
3144
3145             assert(level);
3146
3147             if(s->out_format == FMT_H263){
3148                 unquant_coeff= alevel*qmul + qadd;
3149             }else{ //MPEG1
3150                 j= s->dsp.idct_permutation[ scantable[i] ]; //FIXME optimize
3151                 if(s->mb_intra){
3152                         unquant_coeff = (int)(  alevel  * qscale * s->intra_matrix[j]) >> 3;
3153                         unquant_coeff =   (unquant_coeff - 1) | 1;
3154                 }else{
3155                         unquant_coeff = (((  alevel  << 1) + 1) * qscale * ((int) s->inter_matrix[j])) >> 4;
3156                         unquant_coeff =   (unquant_coeff - 1) | 1;
3157                 }
3158                 unquant_coeff<<= 3;
3159             }
3160
3161             distortion= (unquant_coeff - dct_coeff) * (unquant_coeff - dct_coeff) - zero_distortion;
3162             level+=64;
3163             if((level&(~127)) == 0){
3164                 for(j=survivor_count-1; j>=0; j--){
3165                     int run= i - survivor[j];
3166                     int score= distortion + length[UNI_AC_ENC_INDEX(run, level)]*lambda;
3167                     score += score_tab[i-run];
3168
3169                     if(score < best_score){
3170                         best_score= score;
3171                         run_tab[i+1]= run;
3172                         level_tab[i+1]= level-64;
3173                     }
3174                 }
3175
3176                 if(s->out_format == FMT_H263){
3177                     for(j=survivor_count-1; j>=0; j--){
3178                         int run= i - survivor[j];
3179                         int score= distortion + last_length[UNI_AC_ENC_INDEX(run, level)]*lambda;
3180                         score += score_tab[i-run];
3181                         if(score < last_score){
3182                             last_score= score;
3183                             last_run= run;
3184                             last_level= level-64;
3185                             last_i= i+1;
3186                         }
3187                     }
3188                 }
3189             }else{
3190                 distortion += esc_length*lambda;
3191                 for(j=survivor_count-1; j>=0; j--){
3192                     int run= i - survivor[j];
3193                     int score= distortion + score_tab[i-run];
3194
3195                     if(score < best_score){
3196                         best_score= score;
3197                         run_tab[i+1]= run;
3198                         level_tab[i+1]= level-64;
3199                     }
3200                 }
3201
3202                 if(s->out_format == FMT_H263){
3203                   for(j=survivor_count-1; j>=0; j--){
3204                         int run= i - survivor[j];
3205                         int score= distortion + score_tab[i-run];
3206                         if(score < last_score){
3207                             last_score= score;
3208                             last_run= run;
3209                             last_level= level-64;
3210                             last_i= i+1;
3211                         }
3212                     }
3213                 }
3214             }
3215         }
3216
3217         score_tab[i+1]= best_score;
3218
3219         //Note: there is a vlc code in mpeg4 which is 1 bit shorter then another one with a shorter run and the same level
3220         if(last_non_zero <= 27){
3221             for(; survivor_count; survivor_count--){
3222                 if(score_tab[ survivor[survivor_count-1] ] <= best_score)
3223                     break;
3224             }
3225         }else{
3226             for(; survivor_count; survivor_count--){
3227                 if(score_tab[ survivor[survivor_count-1] ] <= best_score + lambda)
3228                     break;
3229             }
3230         }
3231
3232         survivor[ survivor_count++ ]= i+1;
3233     }
3234
3235     if(s->out_format != FMT_H263){
3236         last_score= 256*256*256*120;
3237         for(i= survivor[0]; i<=last_non_zero + 1; i++){
3238             int score= score_tab[i];
3239             if(i) score += lambda*2; //FIXME exacter?
3240
3241             if(score < last_score){
3242                 last_score= score;
3243                 last_i= i;
3244                 last_level= level_tab[i];
3245                 last_run= run_tab[i];
3246             }
3247         }
3248     }
3249
3250     s->coded_score[n] = last_score;
3251
3252     dc= FFABS(block[0]);
3253     last_non_zero= last_i - 1;
3254     memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
3255
3256     if(last_non_zero < start_i)
3257         return last_non_zero;
3258
3259     if(last_non_zero == 0 && start_i == 0){
3260         int best_level= 0;
3261         int best_score= dc * dc;
3262
3263         for(i=0; i<coeff_count[0]; i++){
3264             int level= coeff[i][0];
3265             int alevel= FFABS(level);
3266             int unquant_coeff, score, distortion;
3267
3268             if(s->out_format == FMT_H263){
3269                     unquant_coeff= (alevel*qmul + qadd)>>3;
3270             }else{ //MPEG1
3271                     unquant_coeff = (((  alevel  << 1) + 1) * qscale * ((int) s->inter_matrix[0])) >> 4;
3272                     unquant_coeff =   (unquant_coeff - 1) | 1;
3273             }
3274             unquant_coeff = (unquant_coeff + 4) >> 3;
3275             unquant_coeff<<= 3 + 3;
3276
3277             distortion= (unquant_coeff - dc) * (unquant_coeff - dc);
3278             level+=64;
3279             if((level&(~127)) == 0) score= distortion + last_length[UNI_AC_ENC_INDEX(0, level)]*lambda;
3280             else                    score= distortion + esc_length*lambda;
3281
3282             if(score < best_score){
3283                 best_score= score;
3284                 best_level= level - 64;
3285             }
3286         }
3287         block[0]= best_level;
3288         s->coded_score[n] = best_score - dc*dc;
3289         if(best_level == 0) return -1;
3290         else                return last_non_zero;
3291     }
3292
3293     i= last_i;
3294     assert(last_level);
3295
3296     block[ perm_scantable[last_non_zero] ]= last_level;
3297     i -= last_run + 1;
3298
3299     for(; i>start_i; i -= run_tab[i] + 1){
3300         block[ perm_scantable[i-1] ]= level_tab[i];
3301     }
3302
3303     return last_non_zero;
3304 }
3305
3306 //#define REFINE_STATS 1
3307 static int16_t basis[64][64];
3308
3309 static void build_basis(uint8_t *perm){
3310     int i, j, x, y;
3311     emms_c();
3312     for(i=0; i<8; i++){
3313         for(j=0; j<8; j++){
3314             for(y=0; y<8; y++){
3315                 for(x=0; x<8; x++){
3316                     double s= 0.25*(1<<BASIS_SHIFT);
3317                     int index= 8*i + j;
3318                     int perm_index= perm[index];
3319                     if(i==0) s*= sqrt(0.5);
3320                     if(j==0) s*= sqrt(0.5);
3321                     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)));
3322                 }
3323             }
3324         }
3325     }
3326 }
3327
3328 static int dct_quantize_refine(MpegEncContext *s, //FIXME breaks denoise?
3329                         DCTELEM *block, int16_t *weight, DCTELEM *orig,
3330                         int n, int qscale){
3331     int16_t rem[64];
3332     LOCAL_ALIGNED_16(DCTELEM, d1, [64]);
3333     const uint8_t *scantable= s->intra_scantable.scantable;
3334     const uint8_t *perm_scantable= s->intra_scantable.permutated;
3335 //    unsigned int threshold1, threshold2;
3336 //    int bias=0;
3337     int run_tab[65];
3338     int prev_run=0;
3339     int prev_level=0;
3340     int qmul, qadd, start_i, last_non_zero, i, dc;
3341     uint8_t * length;
3342     uint8_t * last_length;
3343     int lambda;
3344     int rle_index, run, q = 1, sum; //q is only used when s->mb_intra is true
3345 #ifdef REFINE_STATS
3346 static int count=0;
3347 static int after_last=0;
3348 static int to_zero=0;
3349 static int from_zero=0;
3350 static int raise=0;
3351 static int lower=0;
3352 static int messed_sign=0;
3353 #endif
3354
3355     if(basis[0][0] == 0)
3356         build_basis(s->dsp.idct_permutation);
3357
3358     qmul= qscale*2;
3359     qadd= (qscale-1)|1;
3360     if (s->mb_intra) {
3361         if (!s->h263_aic) {
3362             if (n < 4)
3363                 q = s->y_dc_scale;
3364             else
3365                 q = s->c_dc_scale;
3366         } else{
3367             /* For AIC we skip quant/dequant of INTRADC */
3368             q = 1;
3369             qadd=0;
3370         }
3371         q <<= RECON_SHIFT-3;
3372         /* note: block[0] is assumed to be positive */
3373         dc= block[0]*q;
3374 //        block[0] = (block[0] + (q >> 1)) / q;
3375         start_i = 1;
3376 //        if(s->mpeg_quant || s->out_format == FMT_MPEG1)
3377 //            bias= 1<<(QMAT_SHIFT-1);
3378         length     = s->intra_ac_vlc_length;
3379         last_length= s->intra_ac_vlc_last_length;
3380     } else {
3381         dc= 0;
3382         start_i = 0;
3383         length     = s->inter_ac_vlc_length;
3384         last_length= s->inter_ac_vlc_last_length;
3385     }
3386     last_non_zero = s->block_last_index[n];
3387
3388 #ifdef REFINE_STATS
3389 {START_TIMER
3390 #endif
3391     dc += (1<<(RECON_SHIFT-1));
3392     for(i=0; i<64; i++){
3393         rem[i]= dc - (orig[i]<<RECON_SHIFT); //FIXME  use orig dirrectly instead of copying to rem[]
3394     }
3395 #ifdef REFINE_STATS
3396 STOP_TIMER("memset rem[]")}
3397 #endif
3398     sum=0;
3399     for(i=0; i<64; i++){
3400         int one= 36;
3401         int qns=4;
3402         int w;
3403
3404         w= FFABS(weight[i]) + qns*one;
3405         w= 15 + (48*qns*one + w/2)/w; // 16 .. 63
3406
3407         weight[i] = w;
3408 //        w=weight[i] = (63*qns + (w/2)) / w;
3409
3410         assert(w>0);
3411         assert(w<(1<<6));
3412         sum += w*w;
3413     }
3414     lambda= sum*(uint64_t)s->lambda2 >> (FF_LAMBDA_SHIFT - 6 + 6 + 6 + 6);
3415 #ifdef REFINE_STATS
3416 {START_TIMER
3417 #endif
3418     run=0;
3419     rle_index=0;
3420     for(i=start_i; i<=last_non_zero; i++){
3421         int j= perm_scantable[i];
3422         const int level= block[j];
3423         int coeff;
3424
3425         if(level){
3426             if(level<0) coeff= qmul*level - qadd;
3427             else        coeff= qmul*level + qadd;
3428             run_tab[rle_index++]=run;
3429             run=0;
3430
3431             s->dsp.add_8x8basis(rem, basis[j], coeff);
3432         }else{
3433             run++;
3434         }
3435     }
3436 #ifdef REFINE_STATS
3437 if(last_non_zero>0){
3438 STOP_TIMER("init rem[]")
3439 }
3440 }
3441
3442 {START_TIMER
3443 #endif
3444     for(;;){
3445         int best_score=s->dsp.try_8x8basis(rem, weight, basis[0], 0);
3446         int best_coeff=0;
3447         int best_change=0;
3448         int run2, best_unquant_change=0, analyze_gradient;
3449 #ifdef REFINE_STATS
3450 {START_TIMER
3451 #endif
3452         analyze_gradient = last_non_zero > 2 || s->avctx->quantizer_noise_shaping >= 3;
3453
3454         if(analyze_gradient){
3455 #ifdef REFINE_STATS
3456 {START_TIMER
3457 #endif
3458             for(i=0; i<64; i++){
3459                 int w= weight[i];
3460
3461                 d1[i] = (rem[i]*w*w + (1<<(RECON_SHIFT+12-1)))>>(RECON_SHIFT+12);
3462             }
3463 #ifdef REFINE_STATS
3464 STOP_TIMER("rem*w*w")}
3465 {START_TIMER
3466 #endif
3467             s->dsp.fdct(d1);
3468 #ifdef REFINE_STATS
3469 STOP_TIMER("dct")}
3470 #endif
3471         }
3472
3473         if(start_i){
3474             const int level= block[0];
3475             int change, old_coeff;
3476
3477             assert(s->mb_intra);
3478
3479             old_coeff= q*level;
3480
3481             for(change=-1; change<=1; change+=2){
3482                 int new_level= level + change;
3483                 int score, new_coeff;
3484
3485                 new_coeff= q*new_level;
3486                 if(new_coeff >= 2048 || new_coeff < 0)
3487                     continue;
3488
3489                 score= s->dsp.try_8x8basis(rem, weight, basis[0], new_coeff - old_coeff);
3490                 if(score<best_score){
3491                     best_score= score;
3492                     best_coeff= 0;
3493                     best_change= change;
3494                     best_unquant_change= new_coeff - old_coeff;
3495                 }
3496             }
3497         }
3498
3499         run=0;
3500         rle_index=0;
3501         run2= run_tab[rle_index++];
3502         prev_level=0;
3503         prev_run=0;
3504
3505         for(i=start_i; i<64; i++){
3506             int j= perm_scantable[i];
3507             const int level= block[j];
3508             int change, old_coeff;
3509
3510             if(s->avctx->quantizer_noise_shaping < 3 && i > last_non_zero + 1)
3511                 break;
3512
3513             if(level){
3514                 if(level<0) old_coeff= qmul*level - qadd;
3515                 else        old_coeff= qmul*level + qadd;
3516                 run2= run_tab[rle_index++]; //FIXME ! maybe after last
3517             }else{
3518                 old_coeff=0;
3519                 run2--;
3520                 assert(run2>=0 || i >= last_non_zero );
3521             }
3522
3523             for(change=-1; change<=1; change+=2){
3524                 int new_level= level + change;
3525                 int score, new_coeff, unquant_change;
3526
3527                 score=0;
3528                 if(s->avctx->quantizer_noise_shaping < 2 && FFABS(new_level) > FFABS(level))
3529                    continue;
3530
3531                 if(new_level){
3532                     if(new_level<0) new_coeff= qmul*new_level - qadd;
3533                     else            new_coeff= qmul*new_level + qadd;
3534                     if(new_coeff >= 2048 || new_coeff <= -2048)
3535                         continue;
3536                     //FIXME check for overflow
3537
3538                     if(level){
3539                         if(level < 63 && level > -63){
3540                             if(i < last_non_zero)
3541                                 score +=   length[UNI_AC_ENC_INDEX(run, new_level+64)]
3542                                          - length[UNI_AC_ENC_INDEX(run, level+64)];
3543                             else
3544                                 score +=   last_length[UNI_AC_ENC_INDEX(run, new_level+64)]
3545                                          - last_length[UNI_AC_ENC_INDEX(run, level+64)];
3546                         }
3547                     }else{
3548                         assert(FFABS(new_level)==1);
3549
3550                         if(analyze_gradient){
3551                             int g= d1[ scantable[i] ];
3552                             if(g && (g^new_level) >= 0)
3553                                 continue;
3554                         }
3555
3556                         if(i < last_non_zero){
3557                             int next_i= i + run2 + 1;
3558                             int next_level= block[ perm_scantable[next_i] ] + 64;
3559
3560                             if(next_level&(~127))
3561                                 next_level= 0;
3562
3563                             if(next_i < last_non_zero)
3564                                 score +=   length[UNI_AC_ENC_INDEX(run, 65)]
3565                                          + length[UNI_AC_ENC_INDEX(run2, next_level)]
3566                                          - length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)];
3567                             else
3568                                 score +=  length[UNI_AC_ENC_INDEX(run, 65)]
3569                                         + last_length[UNI_AC_ENC_INDEX(run2, next_level)]
3570                                         - last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)];
3571                         }else{
3572                             score += last_length[UNI_AC_ENC_INDEX(run, 65)];
3573                             if(prev_level){
3574                                 score +=  length[UNI_AC_ENC_INDEX(prev_run, prev_level)]
3575                                         - last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)];
3576                             }
3577                         }
3578                     }
3579                 }else{
3580                     new_coeff=0;
3581                     assert(FFABS(level)==1);
3582
3583                     if(i < last_non_zero){
3584                         int next_i= i + run2 + 1;
3585                         int next_level= block[ perm_scantable[next_i] ] + 64;
3586
3587                         if(next_level&(~127))
3588                             next_level= 0;
3589
3590                         if(next_i < last_non_zero)
3591                             score +=   length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)]
3592                                      - length[UNI_AC_ENC_INDEX(run2, next_level)]
3593                                      - length[UNI_AC_ENC_INDEX(run, 65)];
3594                         else
3595                             score +=   last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)]
3596                                      - last_length[UNI_AC_ENC_INDEX(run2, next_level)]
3597                                      - length[UNI_AC_ENC_INDEX(run, 65)];
3598                     }else{
3599                         score += -last_length[UNI_AC_ENC_INDEX(run, 65)];
3600                         if(prev_level){
3601                             score +=  last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)]
3602                                     - length[UNI_AC_ENC_INDEX(prev_run, prev_level)];
3603                         }
3604                     }
3605                 }
3606
3607                 score *= lambda;
3608
3609                 unquant_change= new_coeff - old_coeff;
3610                 assert((score < 100*lambda && score > -100*lambda) || lambda==0);
3611
3612                 score+= s->dsp.try_8x8basis(rem, weight, basis[j], unquant_change);
3613                 if(score<best_score){
3614                     best_score= score;
3615                     best_coeff= i;
3616                     best_change= change;
3617                     best_unquant_change= unquant_change;
3618                 }
3619             }
3620             if(level){
3621                 prev_level= level + 64;
3622                 if(prev_level&(~127))
3623                     prev_level= 0;
3624                 prev_run= run;
3625                 run=0;
3626             }else{
3627                 run++;
3628             }
3629         }
3630 #ifdef REFINE_STATS
3631 STOP_TIMER("iterative step")}
3632 #endif
3633
3634         if(best_change){
3635             int j= perm_scantable[ best_coeff ];
3636
3637             block[j] += best_change;
3638
3639             if(best_coeff > last_non_zero){
3640                 last_non_zero= best_coeff;
3641                 assert(block[j]);
3642 #ifdef REFINE_STATS
3643 after_last++;
3644 #endif
3645             }else{
3646 #ifdef REFINE_STATS
3647 if(block[j]){
3648     if(block[j] - best_change){
3649         if(FFABS(block[j]) > FFABS(block[j] - best_change)){
3650             raise++;
3651         }else{
3652             lower++;
3653         }
3654     }else{
3655         from_zero++;
3656     }
3657 }else{
3658     to_zero++;
3659 }
3660 #endif
3661                 for(; last_non_zero>=start_i; last_non_zero--){
3662                     if(block[perm_scantable[last_non_zero]])
3663                         break;
3664                 }
3665             }
3666 #ifdef REFINE_STATS
3667 count++;
3668 if(256*256*256*64 % count == 0){
3669     printf("after_last:%d to_zero:%d from_zero:%d raise:%d lower:%d sign:%d xyp:%d/%d/%d\n", after_last, to_zero, from_zero, raise, lower, messed_sign, s->mb_x, s->mb_y, s->picture_number);
3670 }
3671 #endif
3672             run=0;
3673             rle_index=0;
3674             for(i=start_i; i<=last_non_zero; i++){
3675                 int j= perm_scantable[i];
3676                 const int level= block[j];
3677
3678                  if(level){
3679                      run_tab[rle_index++]=run;
3680                      run=0;
3681                  }else{
3682                      run++;
3683                  }
3684             }
3685
3686             s->dsp.add_8x8basis(rem, basis[j], best_unquant_change);
3687         }else{
3688             break;
3689         }
3690     }
3691 #ifdef REFINE_STATS
3692 if(last_non_zero>0){
3693 STOP_TIMER("iterative search")
3694 }
3695 }
3696 #endif
3697
3698     return last_non_zero;
3699 }
3700
3701 int dct_quantize_c(MpegEncContext *s,
3702                         DCTELEM *block, int n,
3703                         int qscale, int *overflow)
3704 {
3705     int i, j, level, last_non_zero, q, start_i;
3706     const int *qmat;
3707     const uint8_t *scantable= s->intra_scantable.scantable;
3708     int bias;
3709     int max=0;
3710     unsigned int threshold1, threshold2;
3711
3712     s->dsp.fdct (block);
3713
3714     if(s->dct_error_sum)
3715         s->denoise_dct(s, block);
3716
3717     if (s->mb_intra) {
3718         if (!s->h263_aic) {
3719             if (n < 4)
3720                 q = s->y_dc_scale;
3721             else
3722                 q = s->c_dc_scale;
3723             q = q << 3;
3724         } else
3725             /* For AIC we skip quant/dequant of INTRADC */
3726             q = 1 << 3;
3727
3728         /* note: block[0] is assumed to be positive */
3729         block[0] = (block[0] + (q >> 1)) / q;
3730         start_i = 1;
3731         last_non_zero = 0;
3732         qmat = s->q_intra_matrix[qscale];
3733         bias= s->intra_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT);
3734     } else {
3735         start_i = 0;
3736         last_non_zero = -1;
3737         qmat = s->q_inter_matrix[qscale];
3738         bias= s->inter_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT);
3739     }
3740     threshold1= (1<<QMAT_SHIFT) - bias - 1;
3741     threshold2= (threshold1<<1);
3742     for(i=63;i>=start_i;i--) {
3743         j = scantable[i];
3744         level = block[j] * qmat[j];
3745
3746         if(((unsigned)(level+threshold1))>threshold2){
3747             last_non_zero = i;
3748             break;
3749         }else{
3750             block[j]=0;
3751         }
3752     }
3753     for(i=start_i; i<=last_non_zero; i++) {
3754         j = scantable[i];
3755         level = block[j] * qmat[j];
3756
3757 //        if(   bias+level >= (1<<QMAT_SHIFT)
3758 //           || bias-level >= (1<<QMAT_SHIFT)){
3759         if(((unsigned)(level+threshold1))>threshold2){
3760             if(level>0){
3761                 level= (bias + level)>>QMAT_SHIFT;
3762                 block[j]= level;
3763             }else{
3764                 level= (bias - level)>>QMAT_SHIFT;
3765                 block[j]= -level;
3766             }
3767             max |=level;
3768         }else{
3769             block[j]=0;
3770         }
3771     }
3772     *overflow= s->max_qcoeff < max; //overflow might have happened
3773
3774     /* we need this permutation so that we correct the IDCT, we only permute the !=0 elements */
3775     if (s->dsp.idct_permutation_type != FF_NO_IDCT_PERM)
3776         ff_block_permute(block, s->dsp.idct_permutation, scantable, last_non_zero);
3777
3778     return last_non_zero;
3779 }
3780
3781 AVCodec ff_h263_encoder = {
3782     .name           = "h263",
3783     .type           = AVMEDIA_TYPE_VIDEO,
3784     .id             = CODEC_ID_H263,
3785     .priv_data_size = sizeof(MpegEncContext),
3786     .init           = MPV_encode_init,
3787     .encode         = MPV_encode_picture,
3788     .close          = MPV_encode_end,
3789     .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3790     .long_name= NULL_IF_CONFIG_SMALL("H.263 / H.263-1996"),
3791 };
3792
3793 AVCodec ff_h263p_encoder = {
3794     .name           = "h263p",
3795     .type           = AVMEDIA_TYPE_VIDEO,
3796     .id             = CODEC_ID_H263P,
3797     .priv_data_size = sizeof(MpegEncContext),
3798     .init           = MPV_encode_init,
3799     .encode         = MPV_encode_picture,
3800     .close          = MPV_encode_end,
3801     .capabilities = CODEC_CAP_SLICE_THREADS,
3802     .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3803     .long_name= NULL_IF_CONFIG_SMALL("H.263+ / H.263-1998 / H.263 version 2"),
3804 };
3805
3806 AVCodec ff_msmpeg4v2_encoder = {
3807     .name           = "msmpeg4v2",
3808     .type           = AVMEDIA_TYPE_VIDEO,
3809     .id             = CODEC_ID_MSMPEG4V2,
3810     .priv_data_size = sizeof(MpegEncContext),
3811     .init           = MPV_encode_init,
3812     .encode         = MPV_encode_picture,
3813     .close          = MPV_encode_end,
3814     .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3815     .long_name= NULL_IF_CONFIG_SMALL("MPEG-4 part 2 Microsoft variant version 2"),
3816 };
3817
3818 AVCodec ff_msmpeg4v3_encoder = {
3819     .name           = "msmpeg4",
3820     .type           = AVMEDIA_TYPE_VIDEO,
3821     .id             = CODEC_ID_MSMPEG4V3,
3822     .priv_data_size = sizeof(MpegEncContext),
3823     .init           = MPV_encode_init,
3824     .encode         = MPV_encode_picture,
3825     .close          = MPV_encode_end,
3826     .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3827     .long_name= NULL_IF_CONFIG_SMALL("MPEG-4 part 2 Microsoft variant version 3"),
3828 };
3829
3830 AVCodec ff_wmv1_encoder = {
3831     .name           = "wmv1",
3832     .type           = AVMEDIA_TYPE_VIDEO,
3833     .id             = CODEC_ID_WMV1,
3834     .priv_data_size = sizeof(MpegEncContext),
3835     .init           = MPV_encode_init,
3836     .encode         = MPV_encode_picture,
3837     .close          = MPV_encode_end,
3838     .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3839     .long_name= NULL_IF_CONFIG_SMALL("Windows Media Video 7"),
3840 };