Merge remote-tracking branch 'qatar/master'
[ffmpeg.git] / libavcodec / mpegvideo_enc.c
1 /*
2  * The simplest mpeg encoder (well, it was the simplest!)
3  * Copyright (c) 2000,2001 Fabrice Bellard
4  * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
5  *
6  * 4MV & hq & B-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
7  *
8  * This file is part of FFmpeg.
9  *
10  * FFmpeg is free software; you can redistribute it and/or
11  * modify it under the terms of the GNU Lesser General Public
12  * License as published by the Free Software Foundation; either
13  * version 2.1 of the License, or (at your option) any later version.
14  *
15  * FFmpeg is distributed in the hope that it will be useful,
16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
18  * Lesser General Public License for more details.
19  *
20  * You should have received a copy of the GNU Lesser General Public
21  * License along with FFmpeg; if not, write to the Free Software
22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23  */
24
25 /**
26  * @file
27  * The simplest mpeg encoder (well, it was the simplest!).
28  */
29
30 #include "libavutil/intmath.h"
31 #include "avcodec.h"
32 #include "dsputil.h"
33 #include "mpegvideo.h"
34 #include "mpegvideo_common.h"
35 #include "h263.h"
36 #include "mjpegenc.h"
37 #include "msmpeg4.h"
38 #include "faandct.h"
39 #include "thread.h"
40 #include "aandcttab.h"
41 #include "flv.h"
42 #include "mpeg4video.h"
43 #include "internal.h"
44 #include <limits.h>
45
46 //#undef NDEBUG
47 //#include <assert.h>
48
49 static int encode_picture(MpegEncContext *s, int picture_number);
50 static int dct_quantize_refine(MpegEncContext *s, DCTELEM *block, int16_t *weight, DCTELEM *orig, int n, int qscale);
51 static int sse_mb(MpegEncContext *s);
52 static void denoise_dct_c(MpegEncContext *s, DCTELEM *block);
53 static int dct_quantize_trellis_c(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow);
54
55 /* enable all paranoid tests for rounding, overflows, etc... */
56 //#define PARANOID
57
58 //#define DEBUG
59
60 static uint8_t default_mv_penalty[MAX_FCODE+1][MAX_MV*2+1];
61 static uint8_t default_fcode_tab[MAX_MV*2+1];
62
63 void ff_convert_matrix(DSPContext *dsp, int (*qmat)[64], uint16_t (*qmat16)[2][64],
64                            const uint16_t *quant_matrix, int bias, int qmin, int qmax, int intra)
65 {
66     int qscale;
67     int shift=0;
68
69     for(qscale=qmin; qscale<=qmax; qscale++){
70         int i;
71         if (dsp->fdct == ff_jpeg_fdct_islow
72 #ifdef FAAN_POSTSCALE
73             || dsp->fdct == ff_faandct
74 #endif
75             ) {
76             for(i=0;i<64;i++) {
77                 const int j= dsp->idct_permutation[i];
78                 /* 16 <= qscale * quant_matrix[i] <= 7905 */
79                 /* 19952             <= ff_aanscales[i] * qscale * quant_matrix[i]               <= 249205026 */
80                 /* (1 << 36) / 19952 >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= (1 << 36) / 249205026 */
81                 /* 3444240           >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= 275 */
82
83                 qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) /
84                                 (qscale * quant_matrix[j]));
85             }
86         } else if (dsp->fdct == fdct_ifast
87 #ifndef FAAN_POSTSCALE
88                    || dsp->fdct == ff_faandct
89 #endif
90                    ) {
91             for(i=0;i<64;i++) {
92                 const int j= dsp->idct_permutation[i];
93                 /* 16 <= qscale * quant_matrix[i] <= 7905 */
94                 /* 19952             <= ff_aanscales[i] * qscale * quant_matrix[i]               <= 249205026 */
95                 /* (1 << 36) / 19952 >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */
96                 /* 3444240           >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= 275 */
97
98                 qmat[qscale][i] = (int)((UINT64_C(1) << (QMAT_SHIFT + 14)) /
99                                 (ff_aanscales[i] * qscale * quant_matrix[j]));
100             }
101         } else {
102             for(i=0;i<64;i++) {
103                 const int j= dsp->idct_permutation[i];
104                 /* We can safely suppose that 16 <= quant_matrix[i] <= 255
105                    So 16           <= qscale * quant_matrix[i]             <= 7905
106                    so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905
107                    so 32768        >= (1<<19) / (qscale * quant_matrix[i]) >= 67
108                 */
109                 qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) / (qscale * quant_matrix[j]));
110 //                qmat  [qscale][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]);
111                 qmat16[qscale][0][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[j]);
112
113                 if(qmat16[qscale][0][i]==0 || qmat16[qscale][0][i]==128*256) qmat16[qscale][0][i]=128*256-1;
114                 qmat16[qscale][1][i]= ROUNDED_DIV(bias<<(16-QUANT_BIAS_SHIFT), qmat16[qscale][0][i]);
115             }
116         }
117
118         for(i=intra; i<64; i++){
119             int64_t max= 8191;
120             if (dsp->fdct == fdct_ifast
121 #ifndef FAAN_POSTSCALE
122                    || dsp->fdct == ff_faandct
123 #endif
124                    ) {
125                 max = (8191LL*ff_aanscales[i]) >> 14;
126             }
127             while(((max * qmat[qscale][i]) >> shift) > INT_MAX){
128                 shift++;
129             }
130         }
131     }
132     if(shift){
133         av_log(NULL, AV_LOG_INFO, "Warning, QMAT_SHIFT is larger than %d, overflows possible\n", QMAT_SHIFT - shift);
134     }
135 }
136
137 static inline void update_qscale(MpegEncContext *s){
138     s->qscale= (s->lambda*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7);
139     s->qscale= av_clip(s->qscale, s->avctx->qmin, s->avctx->qmax);
140
141     s->lambda2= (s->lambda*s->lambda + FF_LAMBDA_SCALE/2) >> FF_LAMBDA_SHIFT;
142 }
143
144 void ff_write_quant_matrix(PutBitContext *pb, uint16_t *matrix){
145     int i;
146
147     if(matrix){
148         put_bits(pb, 1, 1);
149         for(i=0;i<64;i++) {
150             put_bits(pb, 8, matrix[ ff_zigzag_direct[i] ]);
151         }
152     }else
153         put_bits(pb, 1, 0);
154 }
155
156 /**
157  * init s->current_picture.qscale_table from s->lambda_table
158  */
159 void ff_init_qscale_tab(MpegEncContext *s){
160     int8_t * const qscale_table= s->current_picture.qscale_table;
161     int i;
162
163     for(i=0; i<s->mb_num; i++){
164         unsigned int lam= s->lambda_table[ s->mb_index2xy[i] ];
165         int qp= (lam*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7);
166         qscale_table[ s->mb_index2xy[i] ]= av_clip(qp, s->avctx->qmin, s->avctx->qmax);
167     }
168 }
169
170 static void copy_picture_attributes(MpegEncContext *s, AVFrame *dst, AVFrame *src){
171     int i;
172
173     dst->pict_type              = src->pict_type;
174     dst->quality                = src->quality;
175     dst->coded_picture_number   = src->coded_picture_number;
176     dst->display_picture_number = src->display_picture_number;
177 //    dst->reference              = src->reference;
178     dst->pts                    = src->pts;
179     dst->interlaced_frame       = src->interlaced_frame;
180     dst->top_field_first        = src->top_field_first;
181
182     if(s->avctx->me_threshold){
183         if(!src->motion_val[0])
184             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_val not set!\n");
185         if(!src->mb_type)
186             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.mb_type not set!\n");
187         if(!src->ref_index[0])
188             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.ref_index not set!\n");
189         if(src->motion_subsample_log2 != dst->motion_subsample_log2)
190             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_subsample_log2 doesn't match! (%d!=%d)\n",
191             src->motion_subsample_log2, dst->motion_subsample_log2);
192
193         memcpy(dst->mb_type, src->mb_type, s->mb_stride * s->mb_height * sizeof(dst->mb_type[0]));
194
195         for(i=0; i<2; i++){
196             int stride= ((16*s->mb_width )>>src->motion_subsample_log2) + 1;
197             int height= ((16*s->mb_height)>>src->motion_subsample_log2);
198
199             if(src->motion_val[i] && src->motion_val[i] != dst->motion_val[i]){
200                 memcpy(dst->motion_val[i], src->motion_val[i], 2*stride*height*sizeof(int16_t));
201             }
202             if(src->ref_index[i] && src->ref_index[i] != dst->ref_index[i]){
203                 memcpy(dst->ref_index[i], src->ref_index[i], s->mb_stride*4*s->mb_height*sizeof(int8_t));
204             }
205         }
206     }
207 }
208
209 static void update_duplicate_context_after_me(MpegEncContext *dst, MpegEncContext *src){
210 #define COPY(a) dst->a= src->a
211     COPY(pict_type);
212     COPY(current_picture);
213     COPY(f_code);
214     COPY(b_code);
215     COPY(qscale);
216     COPY(lambda);
217     COPY(lambda2);
218     COPY(picture_in_gop_number);
219     COPY(gop_picture_number);
220     COPY(frame_pred_frame_dct); //FIXME don't set in encode_header
221     COPY(progressive_frame); //FIXME don't set in encode_header
222     COPY(partitioned_frame); //FIXME don't set in encode_header
223 #undef COPY
224 }
225
226 /**
227  * sets the given MpegEncContext to defaults for encoding.
228  * the changed fields will not depend upon the prior state of the MpegEncContext.
229  */
230 static void MPV_encode_defaults(MpegEncContext *s){
231     int i;
232     MPV_common_defaults(s);
233
234     for(i=-16; i<16; i++){
235         default_fcode_tab[i + MAX_MV]= 1;
236     }
237     s->me.mv_penalty= default_mv_penalty;
238     s->fcode_tab= default_fcode_tab;
239 }
240
241 /* init video encoder */
242 av_cold int MPV_encode_init(AVCodecContext *avctx)
243 {
244     MpegEncContext *s = avctx->priv_data;
245     int i;
246     int chroma_h_shift, chroma_v_shift;
247
248     MPV_encode_defaults(s);
249
250     switch (avctx->codec_id) {
251     case CODEC_ID_MPEG2VIDEO:
252         if(avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P){
253             av_log(avctx, AV_LOG_ERROR, "only YUV420 and YUV422 are supported\n");
254             return -1;
255         }
256         break;
257     case CODEC_ID_LJPEG:
258         if(avctx->pix_fmt != PIX_FMT_YUVJ420P && avctx->pix_fmt != PIX_FMT_YUVJ422P && avctx->pix_fmt != PIX_FMT_YUVJ444P && avctx->pix_fmt != PIX_FMT_BGRA &&
259            ((avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P && avctx->pix_fmt != PIX_FMT_YUV444P) || avctx->strict_std_compliance>FF_COMPLIANCE_UNOFFICIAL)){
260             av_log(avctx, AV_LOG_ERROR, "colorspace not supported in LJPEG\n");
261             return -1;
262         }
263         break;
264     case CODEC_ID_MJPEG:
265         if(avctx->pix_fmt != PIX_FMT_YUVJ420P && avctx->pix_fmt != PIX_FMT_YUVJ422P &&
266            ((avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P) || avctx->strict_std_compliance>FF_COMPLIANCE_UNOFFICIAL)){
267             av_log(avctx, AV_LOG_ERROR, "colorspace not supported in jpeg\n");
268             return -1;
269         }
270         break;
271     default:
272         if(avctx->pix_fmt != PIX_FMT_YUV420P){
273             av_log(avctx, AV_LOG_ERROR, "only YUV420 is supported\n");
274             return -1;
275         }
276     }
277
278     switch (avctx->pix_fmt) {
279     case PIX_FMT_YUVJ422P:
280     case PIX_FMT_YUV422P:
281         s->chroma_format = CHROMA_422;
282         break;
283     case PIX_FMT_YUVJ420P:
284     case PIX_FMT_YUV420P:
285     default:
286         s->chroma_format = CHROMA_420;
287         break;
288     }
289
290     s->bit_rate = avctx->bit_rate;
291     s->width = avctx->width;
292     s->height = avctx->height;
293     if(avctx->gop_size > 600 && avctx->strict_std_compliance>FF_COMPLIANCE_EXPERIMENTAL){
294         av_log(avctx, AV_LOG_ERROR, "Warning keyframe interval too large! reducing it ...\n");
295         avctx->gop_size=600;
296     }
297     s->gop_size = avctx->gop_size;
298     s->avctx = avctx;
299     s->flags= avctx->flags;
300     s->flags2= avctx->flags2;
301     s->max_b_frames= avctx->max_b_frames;
302     s->codec_id= avctx->codec->id;
303     s->luma_elim_threshold  = avctx->luma_elim_threshold;
304     s->chroma_elim_threshold= avctx->chroma_elim_threshold;
305     s->strict_std_compliance= avctx->strict_std_compliance;
306     s->data_partitioning= avctx->flags & CODEC_FLAG_PART;
307     s->quarter_sample= (avctx->flags & CODEC_FLAG_QPEL)!=0;
308     s->mpeg_quant= avctx->mpeg_quant;
309     s->rtp_mode= !!avctx->rtp_payload_size;
310     s->intra_dc_precision= avctx->intra_dc_precision;
311     s->user_specified_pts = AV_NOPTS_VALUE;
312
313     if (s->gop_size <= 1) {
314         s->intra_only = 1;
315         s->gop_size = 12;
316     } else {
317         s->intra_only = 0;
318     }
319
320     s->me_method = avctx->me_method;
321
322     /* Fixed QSCALE */
323     s->fixed_qscale = !!(avctx->flags & CODEC_FLAG_QSCALE);
324
325     s->adaptive_quant= (   s->avctx->lumi_masking
326                         || s->avctx->dark_masking
327                         || s->avctx->temporal_cplx_masking
328                         || s->avctx->spatial_cplx_masking
329                         || s->avctx->p_masking
330                         || s->avctx->border_masking
331                         || (s->flags&CODEC_FLAG_QP_RD))
332                        && !s->fixed_qscale;
333
334     s->obmc= !!(s->flags & CODEC_FLAG_OBMC);
335     s->loop_filter= !!(s->flags & CODEC_FLAG_LOOP_FILTER);
336     s->alternate_scan= !!(s->flags & CODEC_FLAG_ALT_SCAN);
337     s->intra_vlc_format= !!(s->flags2 & CODEC_FLAG2_INTRA_VLC);
338     s->q_scale_type= !!(s->flags2 & CODEC_FLAG2_NON_LINEAR_QUANT);
339
340     if(avctx->rc_max_rate && !avctx->rc_buffer_size){
341         av_log(avctx, AV_LOG_ERROR, "a vbv buffer size is needed, for encoding with a maximum bitrate\n");
342         return -1;
343     }
344
345     if(avctx->rc_min_rate && avctx->rc_max_rate != avctx->rc_min_rate){
346         av_log(avctx, AV_LOG_INFO, "Warning min_rate > 0 but min_rate != max_rate isn't recommended!\n");
347     }
348
349     if(avctx->rc_min_rate && avctx->rc_min_rate > avctx->bit_rate){
350         av_log(avctx, AV_LOG_ERROR, "bitrate below min bitrate\n");
351         return -1;
352     }
353
354     if(avctx->rc_max_rate && avctx->rc_max_rate < avctx->bit_rate){
355         av_log(avctx, AV_LOG_ERROR, "bitrate above max bitrate\n");
356         return -1;
357     }
358
359     if(avctx->rc_max_rate && avctx->rc_max_rate == avctx->bit_rate && avctx->rc_max_rate != avctx->rc_min_rate){
360         av_log(avctx, AV_LOG_INFO, "impossible bitrate constraints, this will fail\n");
361     }
362
363     if(avctx->rc_buffer_size && avctx->bit_rate*(int64_t)avctx->time_base.num > avctx->rc_buffer_size * (int64_t)avctx->time_base.den){
364         av_log(avctx, AV_LOG_ERROR, "VBV buffer too small for bitrate\n");
365         return -1;
366     }
367
368     if(!s->fixed_qscale && avctx->bit_rate*av_q2d(avctx->time_base) > avctx->bit_rate_tolerance){
369         av_log(avctx, AV_LOG_ERROR, "bitrate tolerance too small for bitrate\n");
370         return -1;
371     }
372
373     if(   s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate
374        && (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO)
375        && 90000LL * (avctx->rc_buffer_size-1) > s->avctx->rc_max_rate*0xFFFFLL){
376
377         av_log(avctx, AV_LOG_INFO, "Warning vbv_delay will be set to 0xFFFF (=VBR) as the specified vbv buffer is too large for the given bitrate!\n");
378     }
379
380     if((s->flags & CODEC_FLAG_4MV) && s->codec_id != CODEC_ID_MPEG4
381        && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P && s->codec_id != CODEC_ID_FLV1){
382         av_log(avctx, AV_LOG_ERROR, "4MV not supported by codec\n");
383         return -1;
384     }
385
386     if(s->obmc && s->avctx->mb_decision != FF_MB_DECISION_SIMPLE){
387         av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with simple mb decision\n");
388         return -1;
389     }
390
391     if(s->obmc && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P){
392         av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with H263(+)\n");
393         return -1;
394     }
395
396     if(s->quarter_sample && s->codec_id != CODEC_ID_MPEG4){
397         av_log(avctx, AV_LOG_ERROR, "qpel not supported by codec\n");
398         return -1;
399     }
400
401     if(s->data_partitioning && s->codec_id != CODEC_ID_MPEG4){
402         av_log(avctx, AV_LOG_ERROR, "data partitioning not supported by codec\n");
403         return -1;
404     }
405
406     if(s->max_b_frames && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO){
407         av_log(avctx, AV_LOG_ERROR, "b frames not supported by codec\n");
408         return -1;
409     }
410
411     if ((s->codec_id == CODEC_ID_MPEG4 || s->codec_id == CODEC_ID_H263 ||
412          s->codec_id == CODEC_ID_H263P) &&
413         (avctx->sample_aspect_ratio.num > 255 || avctx->sample_aspect_ratio.den > 255)) {
414         av_log(avctx, AV_LOG_ERROR, "Invalid pixel aspect ratio %i/%i, limit is 255/255\n",
415                avctx->sample_aspect_ratio.num, avctx->sample_aspect_ratio.den);
416         return -1;
417     }
418
419     if((s->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN))
420        && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG2VIDEO){
421         av_log(avctx, AV_LOG_ERROR, "interlacing not supported by codec\n");
422         return -1;
423     }
424
425     if(s->mpeg_quant && s->codec_id != CODEC_ID_MPEG4){ //FIXME mpeg2 uses that too
426         av_log(avctx, AV_LOG_ERROR, "mpeg2 style quantization not supported by codec\n");
427         return -1;
428     }
429
430     if((s->flags & CODEC_FLAG_CBP_RD) && !avctx->trellis){
431         av_log(avctx, AV_LOG_ERROR, "CBP RD needs trellis quant\n");
432         return -1;
433     }
434
435     if((s->flags & CODEC_FLAG_QP_RD) && s->avctx->mb_decision != FF_MB_DECISION_RD){
436         av_log(avctx, AV_LOG_ERROR, "QP RD needs mbd=2\n");
437         return -1;
438     }
439
440     if(s->avctx->scenechange_threshold < 1000000000 && (s->flags & CODEC_FLAG_CLOSED_GOP)){
441         av_log(avctx, AV_LOG_ERROR, "closed gop with scene change detection are not supported yet, set threshold to 1000000000\n");
442         return -1;
443     }
444
445     if((s->flags2 & CODEC_FLAG2_INTRA_VLC) && s->codec_id != CODEC_ID_MPEG2VIDEO){
446         av_log(avctx, AV_LOG_ERROR, "intra vlc table not supported by codec\n");
447         return -1;
448     }
449
450     if(s->flags & CODEC_FLAG_LOW_DELAY){
451         if (s->codec_id != CODEC_ID_MPEG2VIDEO){
452             av_log(avctx, AV_LOG_ERROR, "low delay forcing is only available for mpeg2\n");
453             return -1;
454         }
455         if (s->max_b_frames != 0){
456             av_log(avctx, AV_LOG_ERROR, "b frames cannot be used with low delay\n");
457             return -1;
458         }
459     }
460
461     if(s->q_scale_type == 1){
462         if(s->codec_id != CODEC_ID_MPEG2VIDEO){
463             av_log(avctx, AV_LOG_ERROR, "non linear quant is only available for mpeg2\n");
464             return -1;
465         }
466         if(avctx->qmax > 12){
467             av_log(avctx, AV_LOG_ERROR, "non linear quant only supports qmax <= 12 currently\n");
468             return -1;
469         }
470     }
471
472     if(s->avctx->thread_count > 1 && s->codec_id != CODEC_ID_MPEG4
473        && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO
474        && (s->codec_id != CODEC_ID_H263P || !(s->flags & CODEC_FLAG_H263P_SLICE_STRUCT))){
475         av_log(avctx, AV_LOG_ERROR, "multi threaded encoding not supported by codec\n");
476         return -1;
477     }
478
479     if(s->avctx->thread_count < 1){
480         av_log(avctx, AV_LOG_ERROR, "automatic thread number detection not supported by codec, patch welcome\n");
481         return -1;
482     }
483
484     if(s->avctx->thread_count > 1)
485         s->rtp_mode= 1;
486
487     if(!avctx->time_base.den || !avctx->time_base.num){
488         av_log(avctx, AV_LOG_ERROR, "framerate not set\n");
489         return -1;
490     }
491
492     i= (INT_MAX/2+128)>>8;
493     if(avctx->me_threshold >= i){
494         av_log(avctx, AV_LOG_ERROR, "me_threshold too large, max is %d\n", i - 1);
495         return -1;
496     }
497     if(avctx->mb_threshold >= i){
498         av_log(avctx, AV_LOG_ERROR, "mb_threshold too large, max is %d\n", i - 1);
499         return -1;
500     }
501
502     if(avctx->b_frame_strategy && (avctx->flags&CODEC_FLAG_PASS2)){
503         av_log(avctx, AV_LOG_INFO, "notice: b_frame_strategy only affects the first pass\n");
504         avctx->b_frame_strategy = 0;
505     }
506
507     i= av_gcd(avctx->time_base.den, avctx->time_base.num);
508     if(i > 1){
509         av_log(avctx, AV_LOG_INFO, "removing common factors from framerate\n");
510         avctx->time_base.den /= i;
511         avctx->time_base.num /= i;
512 //        return -1;
513     }
514
515     if(s->mpeg_quant || s->codec_id==CODEC_ID_MPEG1VIDEO || s->codec_id==CODEC_ID_MPEG2VIDEO || s->codec_id==CODEC_ID_MJPEG){
516         s->intra_quant_bias= 3<<(QUANT_BIAS_SHIFT-3); //(a + x*3/8)/x
517         s->inter_quant_bias= 0;
518     }else{
519         s->intra_quant_bias=0;
520         s->inter_quant_bias=-(1<<(QUANT_BIAS_SHIFT-2)); //(a - x/4)/x
521     }
522
523     if(avctx->intra_quant_bias != FF_DEFAULT_QUANT_BIAS)
524         s->intra_quant_bias= avctx->intra_quant_bias;
525     if(avctx->inter_quant_bias != FF_DEFAULT_QUANT_BIAS)
526         s->inter_quant_bias= avctx->inter_quant_bias;
527
528     avcodec_get_chroma_sub_sample(avctx->pix_fmt, &chroma_h_shift, &chroma_v_shift);
529
530     if(avctx->codec_id == CODEC_ID_MPEG4 && s->avctx->time_base.den > (1<<16)-1){
531         av_log(avctx, AV_LOG_ERROR, "timebase %d/%d not supported by MPEG 4 standard, "
532                "the maximum admitted value for the timebase denominator is %d\n",
533                s->avctx->time_base.num, s->avctx->time_base.den, (1<<16)-1);
534         return -1;
535     }
536     s->time_increment_bits = av_log2(s->avctx->time_base.den - 1) + 1;
537
538     switch(avctx->codec->id) {
539     case CODEC_ID_MPEG1VIDEO:
540         s->out_format = FMT_MPEG1;
541         s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY);
542         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
543         break;
544     case CODEC_ID_MPEG2VIDEO:
545         s->out_format = FMT_MPEG1;
546         s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY);
547         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
548         s->rtp_mode= 1;
549         break;
550     case CODEC_ID_LJPEG:
551     case CODEC_ID_MJPEG:
552         s->out_format = FMT_MJPEG;
553         s->intra_only = 1; /* force intra only for jpeg */
554         if(avctx->codec->id == CODEC_ID_LJPEG && avctx->pix_fmt == PIX_FMT_BGRA){
555             s->mjpeg_vsample[0] = s->mjpeg_hsample[0] =
556             s->mjpeg_vsample[1] = s->mjpeg_hsample[1] =
557             s->mjpeg_vsample[2] = s->mjpeg_hsample[2] = 1;
558         }else{
559             s->mjpeg_vsample[0] = 2;
560             s->mjpeg_vsample[1] = 2>>chroma_v_shift;
561             s->mjpeg_vsample[2] = 2>>chroma_v_shift;
562             s->mjpeg_hsample[0] = 2;
563             s->mjpeg_hsample[1] = 2>>chroma_h_shift;
564             s->mjpeg_hsample[2] = 2>>chroma_h_shift;
565         }
566         if (!(CONFIG_MJPEG_ENCODER || CONFIG_LJPEG_ENCODER)
567             || ff_mjpeg_encode_init(s) < 0)
568             return -1;
569         avctx->delay=0;
570         s->low_delay=1;
571         break;
572     case CODEC_ID_H261:
573         if (!CONFIG_H261_ENCODER)  return -1;
574         if (ff_h261_get_picture_format(s->width, s->height) < 0) {
575             av_log(avctx, AV_LOG_ERROR, "The specified picture size of %dx%d is not valid for the H.261 codec.\nValid sizes are 176x144, 352x288\n", s->width, s->height);
576             return -1;
577         }
578         s->out_format = FMT_H261;
579         avctx->delay=0;
580         s->low_delay=1;
581         break;
582     case CODEC_ID_H263:
583         if (!CONFIG_H263_ENCODER)  return -1;
584         if (ff_match_2uint16(h263_format, FF_ARRAY_ELEMS(h263_format), s->width, s->height) == 8) {
585             av_log(avctx, AV_LOG_ERROR, "The specified picture size of %dx%d is not valid for the H.263 codec.\nValid sizes are 128x96, 176x144, 352x288, 704x576, and 1408x1152. Try H.263+.\n", s->width, s->height);
586             return -1;
587         }
588         s->out_format = FMT_H263;
589         s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
590         avctx->delay=0;
591         s->low_delay=1;
592         break;
593     case CODEC_ID_H263P:
594         s->out_format = FMT_H263;
595         s->h263_plus = 1;
596         /* Fx */
597         s->umvplus = (avctx->flags & CODEC_FLAG_H263P_UMV) ? 1:0;
598         s->h263_aic= (avctx->flags & CODEC_FLAG_AC_PRED) ? 1:0;
599         s->modified_quant= s->h263_aic;
600         s->alt_inter_vlc= (avctx->flags & CODEC_FLAG_H263P_AIV) ? 1:0;
601         s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
602         s->loop_filter= (avctx->flags & CODEC_FLAG_LOOP_FILTER) ? 1:0;
603         s->unrestricted_mv= s->obmc || s->loop_filter || s->umvplus;
604         s->h263_slice_structured= (s->flags & CODEC_FLAG_H263P_SLICE_STRUCT) ? 1:0;
605
606         /* /Fx */
607         /* These are just to be sure */
608         avctx->delay=0;
609         s->low_delay=1;
610         break;
611     case CODEC_ID_FLV1:
612         s->out_format = FMT_H263;
613         s->h263_flv = 2; /* format = 1; 11-bit codes */
614         s->unrestricted_mv = 1;
615         s->rtp_mode=0; /* don't allow GOB */
616         avctx->delay=0;
617         s->low_delay=1;
618         break;
619     case CODEC_ID_RV10:
620         s->out_format = FMT_H263;
621         avctx->delay=0;
622         s->low_delay=1;
623         break;
624     case CODEC_ID_RV20:
625         s->out_format = FMT_H263;
626         avctx->delay=0;
627         s->low_delay=1;
628         s->modified_quant=1;
629         s->h263_aic=1;
630         s->h263_plus=1;
631         s->loop_filter=1;
632         s->unrestricted_mv= 0;
633         break;
634     case CODEC_ID_MPEG4:
635         s->out_format = FMT_H263;
636         s->h263_pred = 1;
637         s->unrestricted_mv = 1;
638         s->low_delay= s->max_b_frames ? 0 : 1;
639         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
640         break;
641     case CODEC_ID_MSMPEG4V2:
642         s->out_format = FMT_H263;
643         s->h263_pred = 1;
644         s->unrestricted_mv = 1;
645         s->msmpeg4_version= 2;
646         avctx->delay=0;
647         s->low_delay=1;
648         break;
649     case CODEC_ID_MSMPEG4V3:
650         s->out_format = FMT_H263;
651         s->h263_pred = 1;
652         s->unrestricted_mv = 1;
653         s->msmpeg4_version= 3;
654         s->flipflop_rounding=1;
655         avctx->delay=0;
656         s->low_delay=1;
657         break;
658     case CODEC_ID_WMV1:
659         s->out_format = FMT_H263;
660         s->h263_pred = 1;
661         s->unrestricted_mv = 1;
662         s->msmpeg4_version= 4;
663         s->flipflop_rounding=1;
664         avctx->delay=0;
665         s->low_delay=1;
666         break;
667     case CODEC_ID_WMV2:
668         s->out_format = FMT_H263;
669         s->h263_pred = 1;
670         s->unrestricted_mv = 1;
671         s->msmpeg4_version= 5;
672         s->flipflop_rounding=1;
673         avctx->delay=0;
674         s->low_delay=1;
675         break;
676     default:
677         return -1;
678     }
679
680     avctx->has_b_frames= !s->low_delay;
681
682     s->encoding = 1;
683
684     s->progressive_frame=
685     s->progressive_sequence= !(avctx->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN));
686
687     /* init */
688     if (MPV_common_init(s) < 0)
689         return -1;
690
691     if(!s->dct_quantize)
692         s->dct_quantize = dct_quantize_c;
693     if(!s->denoise_dct)
694         s->denoise_dct = denoise_dct_c;
695     s->fast_dct_quantize = s->dct_quantize;
696     if(avctx->trellis)
697         s->dct_quantize = dct_quantize_trellis_c;
698
699     if((CONFIG_H263P_ENCODER || CONFIG_RV20_ENCODER) && s->modified_quant)
700         s->chroma_qscale_table= ff_h263_chroma_qscale_table;
701
702     s->quant_precision=5;
703
704     ff_set_cmp(&s->dsp, s->dsp.ildct_cmp, s->avctx->ildct_cmp);
705     ff_set_cmp(&s->dsp, s->dsp.frame_skip_cmp, s->avctx->frame_skip_cmp);
706
707     if (CONFIG_H261_ENCODER && s->out_format == FMT_H261)
708         ff_h261_encode_init(s);
709     if (CONFIG_H263_ENCODER && s->out_format == FMT_H263)
710         h263_encode_init(s);
711     if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version)
712         ff_msmpeg4_encode_init(s);
713     if ((CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
714         && s->out_format == FMT_MPEG1)
715         ff_mpeg1_encode_init(s);
716
717     /* init q matrix */
718     for(i=0;i<64;i++) {
719         int j= s->dsp.idct_permutation[i];
720         if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4 && s->mpeg_quant){
721             s->intra_matrix[j] = ff_mpeg4_default_intra_matrix[i];
722             s->inter_matrix[j] = ff_mpeg4_default_non_intra_matrix[i];
723         }else if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
724             s->intra_matrix[j] =
725             s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
726         }else
727         { /* mpeg1/2 */
728             s->intra_matrix[j] = ff_mpeg1_default_intra_matrix[i];
729             s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
730         }
731         if(s->avctx->intra_matrix)
732             s->intra_matrix[j] = s->avctx->intra_matrix[i];
733         if(s->avctx->inter_matrix)
734             s->inter_matrix[j] = s->avctx->inter_matrix[i];
735     }
736
737     /* precompute matrix */
738     /* for mjpeg, we do include qscale in the matrix */
739     if (s->out_format != FMT_MJPEG) {
740         ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
741                        s->intra_matrix, s->intra_quant_bias, avctx->qmin, 31, 1);
742         ff_convert_matrix(&s->dsp, s->q_inter_matrix, s->q_inter_matrix16,
743                        s->inter_matrix, s->inter_quant_bias, avctx->qmin, 31, 0);
744     }
745
746     if(ff_rate_control_init(s) < 0)
747         return -1;
748
749     return 0;
750 }
751
752 av_cold int MPV_encode_end(AVCodecContext *avctx)
753 {
754     MpegEncContext *s = avctx->priv_data;
755
756     ff_rate_control_uninit(s);
757
758     MPV_common_end(s);
759     if ((CONFIG_MJPEG_ENCODER || CONFIG_LJPEG_ENCODER) && s->out_format == FMT_MJPEG)
760         ff_mjpeg_encode_close(s);
761
762     av_freep(&avctx->extradata);
763
764     return 0;
765 }
766
767 static int get_sae(uint8_t *src, int ref, int stride){
768     int x,y;
769     int acc=0;
770
771     for(y=0; y<16; y++){
772         for(x=0; x<16; x++){
773             acc+= FFABS(src[x+y*stride] - ref);
774         }
775     }
776
777     return acc;
778 }
779
780 static int get_intra_count(MpegEncContext *s, uint8_t *src, uint8_t *ref, int stride){
781     int x, y, w, h;
782     int acc=0;
783
784     w= s->width &~15;
785     h= s->height&~15;
786
787     for(y=0; y<h; y+=16){
788         for(x=0; x<w; x+=16){
789             int offset= x + y*stride;
790             int sad = s->dsp.sad[0](NULL, src + offset, ref + offset, stride, 16);
791             int mean= (s->dsp.pix_sum(src + offset, stride) + 128)>>8;
792             int sae = get_sae(src + offset, mean, stride);
793
794             acc+= sae + 500 < sad;
795         }
796     }
797     return acc;
798 }
799
800
801 static int load_input_picture(MpegEncContext *s, AVFrame *pic_arg){
802     AVFrame *pic=NULL;
803     int64_t pts;
804     int i;
805     const int encoding_delay= s->max_b_frames;
806     int direct=1;
807
808     if(pic_arg){
809         pts= pic_arg->pts;
810         pic_arg->display_picture_number= s->input_picture_number++;
811
812         if(pts != AV_NOPTS_VALUE){
813             if(s->user_specified_pts != AV_NOPTS_VALUE){
814                 int64_t time= pts;
815                 int64_t last= s->user_specified_pts;
816
817                 if(time <= last){
818                     av_log(s->avctx, AV_LOG_ERROR, "Error, Invalid timestamp=%"PRId64", last=%"PRId64"\n", pts, s->user_specified_pts);
819                     return -1;
820                 }
821             }
822             s->user_specified_pts= pts;
823         }else{
824             if(s->user_specified_pts != AV_NOPTS_VALUE){
825                 s->user_specified_pts=
826                 pts= s->user_specified_pts + 1;
827                 av_log(s->avctx, AV_LOG_INFO, "Warning: AVFrame.pts=? trying to guess (%"PRId64")\n", pts);
828             }else{
829                 pts= pic_arg->display_picture_number;
830             }
831         }
832     }
833
834   if(pic_arg){
835     if(encoding_delay && !(s->flags&CODEC_FLAG_INPUT_PRESERVED)) direct=0;
836     if(pic_arg->linesize[0] != s->linesize) direct=0;
837     if(pic_arg->linesize[1] != s->uvlinesize) direct=0;
838     if(pic_arg->linesize[2] != s->uvlinesize) direct=0;
839
840 //    av_log(AV_LOG_DEBUG, "%d %d %d %d\n",pic_arg->linesize[0], pic_arg->linesize[1], s->linesize, s->uvlinesize);
841
842     if(direct){
843         i= ff_find_unused_picture(s, 1);
844
845         pic= (AVFrame*)&s->picture[i];
846         pic->reference= 3;
847
848         for(i=0; i<4; i++){
849             pic->data[i]= pic_arg->data[i];
850             pic->linesize[i]= pic_arg->linesize[i];
851         }
852         if(ff_alloc_picture(s, (Picture*)pic, 1) < 0){
853             return -1;
854         }
855     }else{
856         i= ff_find_unused_picture(s, 0);
857
858         pic= (AVFrame*)&s->picture[i];
859         pic->reference= 3;
860
861         if(ff_alloc_picture(s, (Picture*)pic, 0) < 0){
862             return -1;
863         }
864
865         if(   pic->data[0] + INPLACE_OFFSET == pic_arg->data[0]
866            && pic->data[1] + INPLACE_OFFSET == pic_arg->data[1]
867            && pic->data[2] + INPLACE_OFFSET == pic_arg->data[2]){
868        // empty
869         }else{
870             int h_chroma_shift, v_chroma_shift;
871             avcodec_get_chroma_sub_sample(s->avctx->pix_fmt, &h_chroma_shift, &v_chroma_shift);
872
873             for(i=0; i<3; i++){
874                 int src_stride= pic_arg->linesize[i];
875                 int dst_stride= i ? s->uvlinesize : s->linesize;
876                 int h_shift= i ? h_chroma_shift : 0;
877                 int v_shift= i ? v_chroma_shift : 0;
878                 int w= s->width >>h_shift;
879                 int h= s->height>>v_shift;
880                 uint8_t *src= pic_arg->data[i];
881                 uint8_t *dst= pic->data[i];
882
883                 if(!s->avctx->rc_buffer_size)
884                     dst +=INPLACE_OFFSET;
885
886                 if(src_stride==dst_stride)
887                     memcpy(dst, src, src_stride*h);
888                 else{
889                     while(h--){
890                         memcpy(dst, src, w);
891                         dst += dst_stride;
892                         src += src_stride;
893                     }
894                 }
895             }
896         }
897     }
898     copy_picture_attributes(s, pic, pic_arg);
899     pic->pts= pts; //we set this here to avoid modifiying pic_arg
900   }
901
902     /* shift buffer entries */
903     for(i=1; i<MAX_PICTURE_COUNT /*s->encoding_delay+1*/; i++)
904         s->input_picture[i-1]= s->input_picture[i];
905
906     s->input_picture[encoding_delay]= (Picture*)pic;
907
908     return 0;
909 }
910
911 static int skip_check(MpegEncContext *s, Picture *p, Picture *ref){
912     int x, y, plane;
913     int score=0;
914     int64_t score64=0;
915
916     for(plane=0; plane<3; plane++){
917         const int stride= p->linesize[plane];
918         const int bw= plane ? 1 : 2;
919         for(y=0; y<s->mb_height*bw; y++){
920             for(x=0; x<s->mb_width*bw; x++){
921                 int off= p->type == FF_BUFFER_TYPE_SHARED ? 0: 16;
922                 int v= s->dsp.frame_skip_cmp[1](s, p->data[plane] + 8*(x + y*stride)+off, ref->data[plane] + 8*(x + y*stride), stride, 8);
923
924                 switch(s->avctx->frame_skip_exp){
925                     case 0: score= FFMAX(score, v); break;
926                     case 1: score+= FFABS(v);break;
927                     case 2: score+= v*v;break;
928                     case 3: score64+= FFABS(v*v*(int64_t)v);break;
929                     case 4: score64+= v*v*(int64_t)(v*v);break;
930                 }
931             }
932         }
933     }
934
935     if(score) score64= score;
936
937     if(score64 < s->avctx->frame_skip_threshold)
938         return 1;
939     if(score64 < ((s->avctx->frame_skip_factor * (int64_t)s->lambda)>>8))
940         return 1;
941     return 0;
942 }
943
944 static int estimate_best_b_count(MpegEncContext *s){
945     AVCodec *codec= avcodec_find_encoder(s->avctx->codec_id);
946     AVCodecContext *c= avcodec_alloc_context();
947     AVFrame input[FF_MAX_B_FRAMES+2];
948     const int scale= s->avctx->brd_scale;
949     int i, j, out_size, p_lambda, b_lambda, lambda2;
950     int outbuf_size= s->width * s->height; //FIXME
951     uint8_t *outbuf= av_malloc(outbuf_size);
952     int64_t best_rd= INT64_MAX;
953     int best_b_count= -1;
954
955     assert(scale>=0 && scale <=3);
956
957 //    emms_c();
958     p_lambda= s->last_lambda_for[AV_PICTURE_TYPE_P]; //s->next_picture_ptr->quality;
959     b_lambda= s->last_lambda_for[AV_PICTURE_TYPE_B]; //p_lambda *FFABS(s->avctx->b_quant_factor) + s->avctx->b_quant_offset;
960     if(!b_lambda) b_lambda= p_lambda; //FIXME we should do this somewhere else
961     lambda2= (b_lambda*b_lambda + (1<<FF_LAMBDA_SHIFT)/2 ) >> FF_LAMBDA_SHIFT;
962
963     c->width = s->width >> scale;
964     c->height= s->height>> scale;
965     c->flags= CODEC_FLAG_QSCALE | CODEC_FLAG_PSNR | CODEC_FLAG_INPUT_PRESERVED /*| CODEC_FLAG_EMU_EDGE*/;
966     c->flags|= s->avctx->flags & CODEC_FLAG_QPEL;
967     c->mb_decision= s->avctx->mb_decision;
968     c->me_cmp= s->avctx->me_cmp;
969     c->mb_cmp= s->avctx->mb_cmp;
970     c->me_sub_cmp= s->avctx->me_sub_cmp;
971     c->pix_fmt = PIX_FMT_YUV420P;
972     c->time_base= s->avctx->time_base;
973     c->max_b_frames= s->max_b_frames;
974
975     if (avcodec_open(c, codec) < 0)
976         return -1;
977
978     for(i=0; i<s->max_b_frames+2; i++){
979         int ysize= c->width*c->height;
980         int csize= (c->width/2)*(c->height/2);
981         Picture pre_input, *pre_input_ptr= i ? s->input_picture[i-1] : s->next_picture_ptr;
982
983         avcodec_get_frame_defaults(&input[i]);
984         input[i].data[0]= av_malloc(ysize + 2*csize);
985         input[i].data[1]= input[i].data[0] + ysize;
986         input[i].data[2]= input[i].data[1] + csize;
987         input[i].linesize[0]= c->width;
988         input[i].linesize[1]=
989         input[i].linesize[2]= c->width/2;
990
991         if(pre_input_ptr && (!i || s->input_picture[i-1])) {
992             pre_input= *pre_input_ptr;
993
994             if(pre_input.type != FF_BUFFER_TYPE_SHARED && i) {
995                 pre_input.data[0]+=INPLACE_OFFSET;
996                 pre_input.data[1]+=INPLACE_OFFSET;
997                 pre_input.data[2]+=INPLACE_OFFSET;
998             }
999
1000             s->dsp.shrink[scale](input[i].data[0], input[i].linesize[0], pre_input.data[0], pre_input.linesize[0], c->width, c->height);
1001             s->dsp.shrink[scale](input[i].data[1], input[i].linesize[1], pre_input.data[1], pre_input.linesize[1], c->width>>1, c->height>>1);
1002             s->dsp.shrink[scale](input[i].data[2], input[i].linesize[2], pre_input.data[2], pre_input.linesize[2], c->width>>1, c->height>>1);
1003         }
1004     }
1005
1006     for(j=0; j<s->max_b_frames+1; j++){
1007         int64_t rd=0;
1008
1009         if(!s->input_picture[j])
1010             break;
1011
1012         c->error[0]= c->error[1]= c->error[2]= 0;
1013
1014         input[0].pict_type= AV_PICTURE_TYPE_I;
1015         input[0].quality= 1 * FF_QP2LAMBDA;
1016         out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[0]);
1017 //        rd += (out_size * lambda2) >> FF_LAMBDA_SHIFT;
1018
1019         for(i=0; i<s->max_b_frames+1; i++){
1020             int is_p= i % (j+1) == j || i==s->max_b_frames;
1021
1022             input[i+1].pict_type= is_p ? AV_PICTURE_TYPE_P : AV_PICTURE_TYPE_B;
1023             input[i+1].quality= is_p ? p_lambda : b_lambda;
1024             out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[i+1]);
1025             rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
1026         }
1027
1028         /* get the delayed frames */
1029         while(out_size){
1030             out_size = avcodec_encode_video(c, outbuf, outbuf_size, NULL);
1031             rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
1032         }
1033
1034         rd += c->error[0] + c->error[1] + c->error[2];
1035
1036         if(rd < best_rd){
1037             best_rd= rd;
1038             best_b_count= j;
1039         }
1040     }
1041
1042     av_freep(&outbuf);
1043     avcodec_close(c);
1044     av_freep(&c);
1045
1046     for(i=0; i<s->max_b_frames+2; i++){
1047         av_freep(&input[i].data[0]);
1048     }
1049
1050     return best_b_count;
1051 }
1052
1053 static int select_input_picture(MpegEncContext *s){
1054     int i;
1055
1056     for(i=1; i<MAX_PICTURE_COUNT; i++)
1057         s->reordered_input_picture[i-1]= s->reordered_input_picture[i];
1058     s->reordered_input_picture[MAX_PICTURE_COUNT-1]= NULL;
1059
1060     /* set next picture type & ordering */
1061     if(s->reordered_input_picture[0]==NULL && s->input_picture[0]){
1062         if(/*s->picture_in_gop_number >= s->gop_size ||*/ s->next_picture_ptr==NULL || s->intra_only){
1063             s->reordered_input_picture[0]= s->input_picture[0];
1064             s->reordered_input_picture[0]->pict_type= AV_PICTURE_TYPE_I;
1065             s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++;
1066         }else{
1067             int b_frames;
1068
1069             if(s->avctx->frame_skip_threshold || s->avctx->frame_skip_factor){
1070                 if(s->picture_in_gop_number < s->gop_size && skip_check(s, s->input_picture[0], s->next_picture_ptr)){
1071                 //FIXME check that te gop check above is +-1 correct
1072 //av_log(NULL, AV_LOG_DEBUG, "skip %p %"PRId64"\n", s->input_picture[0]->data[0], s->input_picture[0]->pts);
1073
1074                     if(s->input_picture[0]->type == FF_BUFFER_TYPE_SHARED){
1075                         for(i=0; i<4; i++)
1076                             s->input_picture[0]->data[i]= NULL;
1077                         s->input_picture[0]->type= 0;
1078                     }else{
1079                         assert(   s->input_picture[0]->type==FF_BUFFER_TYPE_USER
1080                                || s->input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
1081
1082                         s->avctx->release_buffer(s->avctx, (AVFrame*)s->input_picture[0]);
1083                     }
1084
1085                     emms_c();
1086                     ff_vbv_update(s, 0);
1087
1088                     goto no_output_pic;
1089                 }
1090             }
1091
1092             if(s->flags&CODEC_FLAG_PASS2){
1093                 for(i=0; i<s->max_b_frames+1; i++){
1094                     int pict_num= s->input_picture[0]->display_picture_number + i;
1095
1096                     if(pict_num >= s->rc_context.num_entries)
1097                         break;
1098                     if(!s->input_picture[i]){
1099                         s->rc_context.entry[pict_num-1].new_pict_type = AV_PICTURE_TYPE_P;
1100                         break;
1101                     }
1102
1103                     s->input_picture[i]->pict_type=
1104                         s->rc_context.entry[pict_num].new_pict_type;
1105                 }
1106             }
1107
1108             if(s->avctx->b_frame_strategy==0){
1109                 b_frames= s->max_b_frames;
1110                 while(b_frames && !s->input_picture[b_frames]) b_frames--;
1111             }else if(s->avctx->b_frame_strategy==1){
1112                 for(i=1; i<s->max_b_frames+1; i++){
1113                     if(s->input_picture[i] && s->input_picture[i]->b_frame_score==0){
1114                         s->input_picture[i]->b_frame_score=
1115                             get_intra_count(s, s->input_picture[i  ]->data[0],
1116                                                s->input_picture[i-1]->data[0], s->linesize) + 1;
1117                     }
1118                 }
1119                 for(i=0; i<s->max_b_frames+1; i++){
1120                     if(s->input_picture[i]==NULL || s->input_picture[i]->b_frame_score - 1 > s->mb_num/s->avctx->b_sensitivity) break;
1121                 }
1122
1123                 b_frames= FFMAX(0, i-1);
1124
1125                 /* reset scores */
1126                 for(i=0; i<b_frames+1; i++){
1127                     s->input_picture[i]->b_frame_score=0;
1128                 }
1129             }else if(s->avctx->b_frame_strategy==2){
1130                 b_frames= estimate_best_b_count(s);
1131             }else{
1132                 av_log(s->avctx, AV_LOG_ERROR, "illegal b frame strategy\n");
1133                 b_frames=0;
1134             }
1135
1136             emms_c();
1137 //static int b_count=0;
1138 //b_count+= b_frames;
1139 //av_log(s->avctx, AV_LOG_DEBUG, "b_frames: %d\n", b_count);
1140
1141             for(i= b_frames - 1; i>=0; i--){
1142                 int type= s->input_picture[i]->pict_type;
1143                 if(type && type != AV_PICTURE_TYPE_B)
1144                     b_frames= i;
1145             }
1146             if(s->input_picture[b_frames]->pict_type == AV_PICTURE_TYPE_B && b_frames == s->max_b_frames){
1147                 av_log(s->avctx, AV_LOG_ERROR, "warning, too many b frames in a row\n");
1148             }
1149
1150             if(s->picture_in_gop_number + b_frames >= s->gop_size){
1151               if((s->flags2 & CODEC_FLAG2_STRICT_GOP) && s->gop_size > s->picture_in_gop_number){
1152                     b_frames= s->gop_size - s->picture_in_gop_number - 1;
1153               }else{
1154                 if(s->flags & CODEC_FLAG_CLOSED_GOP)
1155                     b_frames=0;
1156                 s->input_picture[b_frames]->pict_type= AV_PICTURE_TYPE_I;
1157               }
1158             }
1159
1160             if(   (s->flags & CODEC_FLAG_CLOSED_GOP)
1161                && b_frames
1162                && s->input_picture[b_frames]->pict_type== AV_PICTURE_TYPE_I)
1163                 b_frames--;
1164
1165             s->reordered_input_picture[0]= s->input_picture[b_frames];
1166             if(s->reordered_input_picture[0]->pict_type != AV_PICTURE_TYPE_I)
1167                 s->reordered_input_picture[0]->pict_type= AV_PICTURE_TYPE_P;
1168             s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++;
1169             for(i=0; i<b_frames; i++){
1170                 s->reordered_input_picture[i+1]= s->input_picture[i];
1171                 s->reordered_input_picture[i+1]->pict_type= AV_PICTURE_TYPE_B;
1172                 s->reordered_input_picture[i+1]->coded_picture_number= s->coded_picture_number++;
1173             }
1174         }
1175     }
1176 no_output_pic:
1177     if(s->reordered_input_picture[0]){
1178         s->reordered_input_picture[0]->reference= s->reordered_input_picture[0]->pict_type!=AV_PICTURE_TYPE_B ? 3 : 0;
1179
1180         ff_copy_picture(&s->new_picture, s->reordered_input_picture[0]);
1181
1182         if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_SHARED || s->avctx->rc_buffer_size){
1183             // input is a shared pix, so we can't modifiy it -> alloc a new one & ensure that the shared one is reuseable
1184
1185             int i= ff_find_unused_picture(s, 0);
1186             Picture *pic= &s->picture[i];
1187
1188             pic->reference              = s->reordered_input_picture[0]->reference;
1189             if(ff_alloc_picture(s, pic, 0) < 0){
1190                 return -1;
1191             }
1192
1193             /* mark us unused / free shared pic */
1194             if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_INTERNAL)
1195                 s->avctx->release_buffer(s->avctx, (AVFrame*)s->reordered_input_picture[0]);
1196             for(i=0; i<4; i++)
1197                 s->reordered_input_picture[0]->data[i]= NULL;
1198             s->reordered_input_picture[0]->type= 0;
1199
1200             copy_picture_attributes(s, (AVFrame*)pic, (AVFrame*)s->reordered_input_picture[0]);
1201
1202             s->current_picture_ptr= pic;
1203         }else{
1204             // input is not a shared pix -> reuse buffer for current_pix
1205
1206             assert(   s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_USER
1207                    || s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
1208
1209             s->current_picture_ptr= s->reordered_input_picture[0];
1210             for(i=0; i<4; i++){
1211                 s->new_picture.data[i]+= INPLACE_OFFSET;
1212             }
1213         }
1214         ff_copy_picture(&s->current_picture, s->current_picture_ptr);
1215
1216         s->picture_number= s->new_picture.display_picture_number;
1217 //printf("dpn:%d\n", s->picture_number);
1218     }else{
1219        memset(&s->new_picture, 0, sizeof(Picture));
1220     }
1221     return 0;
1222 }
1223
1224 int MPV_encode_picture(AVCodecContext *avctx,
1225                        unsigned char *buf, int buf_size, void *data)
1226 {
1227     MpegEncContext *s = avctx->priv_data;
1228     AVFrame *pic_arg = data;
1229     int i, stuffing_count, context_count = avctx->thread_count;
1230
1231     for(i=0; i<context_count; i++){
1232         int start_y= s->thread_context[i]->start_mb_y;
1233         int   end_y= s->thread_context[i]->  end_mb_y;
1234         int h= s->mb_height;
1235         uint8_t *start= buf + (size_t)(((int64_t) buf_size)*start_y/h);
1236         uint8_t *end  = buf + (size_t)(((int64_t) buf_size)*  end_y/h);
1237
1238         init_put_bits(&s->thread_context[i]->pb, start, end - start);
1239     }
1240
1241     s->picture_in_gop_number++;
1242
1243     if(load_input_picture(s, pic_arg) < 0)
1244         return -1;
1245
1246     if(select_input_picture(s) < 0){
1247         return -1;
1248     }
1249
1250     /* output? */
1251     if(s->new_picture.data[0]){
1252         s->pict_type= s->new_picture.pict_type;
1253 //emms_c();
1254 //printf("qs:%f %f %d\n", s->new_picture.quality, s->current_picture.quality, s->qscale);
1255         MPV_frame_start(s, avctx);
1256 vbv_retry:
1257         if (encode_picture(s, s->picture_number) < 0)
1258             return -1;
1259
1260         avctx->header_bits = s->header_bits;
1261         avctx->mv_bits     = s->mv_bits;
1262         avctx->misc_bits   = s->misc_bits;
1263         avctx->i_tex_bits  = s->i_tex_bits;
1264         avctx->p_tex_bits  = s->p_tex_bits;
1265         avctx->i_count     = s->i_count;
1266         avctx->p_count     = s->mb_num - s->i_count - s->skip_count; //FIXME f/b_count in avctx
1267         avctx->skip_count  = s->skip_count;
1268
1269         MPV_frame_end(s);
1270
1271         if (CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG)
1272             ff_mjpeg_encode_picture_trailer(s);
1273
1274         if(avctx->rc_buffer_size){
1275             RateControlContext *rcc= &s->rc_context;
1276             int max_size= rcc->buffer_index * avctx->rc_max_available_vbv_use;
1277
1278             if(put_bits_count(&s->pb) > max_size && s->lambda < s->avctx->lmax){
1279                 s->next_lambda= FFMAX(s->lambda+1, s->lambda*(s->qscale+1) / s->qscale);
1280                 if(s->adaptive_quant){
1281                     int i;
1282                     for(i=0; i<s->mb_height*s->mb_stride; i++)
1283                         s->lambda_table[i]= FFMAX(s->lambda_table[i]+1, s->lambda_table[i]*(s->qscale+1) / s->qscale);
1284                 }
1285                 s->mb_skipped = 0;        //done in MPV_frame_start()
1286                 if(s->pict_type==AV_PICTURE_TYPE_P){ //done in encode_picture() so we must undo it
1287                     if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
1288                         s->no_rounding ^= 1;
1289                 }
1290                 if(s->pict_type!=AV_PICTURE_TYPE_B){
1291                     s->time_base= s->last_time_base;
1292                     s->last_non_b_time= s->time - s->pp_time;
1293                 }
1294 //                av_log(NULL, AV_LOG_ERROR, "R:%d ", s->next_lambda);
1295                 for(i=0; i<context_count; i++){
1296                     PutBitContext *pb= &s->thread_context[i]->pb;
1297                     init_put_bits(pb, pb->buf, pb->buf_end - pb->buf);
1298                 }
1299                 goto vbv_retry;
1300             }
1301
1302             assert(s->avctx->rc_max_rate);
1303         }
1304
1305         if(s->flags&CODEC_FLAG_PASS1)
1306             ff_write_pass1_stats(s);
1307
1308         for(i=0; i<4; i++){
1309             s->current_picture_ptr->error[i]= s->current_picture.error[i];
1310             avctx->error[i] += s->current_picture_ptr->error[i];
1311         }
1312
1313         if(s->flags&CODEC_FLAG_PASS1)
1314             assert(avctx->header_bits + avctx->mv_bits + avctx->misc_bits + avctx->i_tex_bits + avctx->p_tex_bits == put_bits_count(&s->pb));
1315         flush_put_bits(&s->pb);
1316         s->frame_bits  = put_bits_count(&s->pb);
1317
1318         stuffing_count= ff_vbv_update(s, s->frame_bits);
1319         if(stuffing_count){
1320             if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < stuffing_count + 50){
1321                 av_log(s->avctx, AV_LOG_ERROR, "stuffing too large\n");
1322                 return -1;
1323             }
1324
1325             switch(s->codec_id){
1326             case CODEC_ID_MPEG1VIDEO:
1327             case CODEC_ID_MPEG2VIDEO:
1328                 while(stuffing_count--){
1329                     put_bits(&s->pb, 8, 0);
1330                 }
1331             break;
1332             case CODEC_ID_MPEG4:
1333                 put_bits(&s->pb, 16, 0);
1334                 put_bits(&s->pb, 16, 0x1C3);
1335                 stuffing_count -= 4;
1336                 while(stuffing_count--){
1337                     put_bits(&s->pb, 8, 0xFF);
1338                 }
1339             break;
1340             default:
1341                 av_log(s->avctx, AV_LOG_ERROR, "vbv buffer overflow\n");
1342             }
1343             flush_put_bits(&s->pb);
1344             s->frame_bits  = put_bits_count(&s->pb);
1345         }
1346
1347         /* update mpeg1/2 vbv_delay for CBR */
1348         if(s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate && s->out_format == FMT_MPEG1
1349            && 90000LL * (avctx->rc_buffer_size-1) <= s->avctx->rc_max_rate*0xFFFFLL){
1350             int vbv_delay, min_delay;
1351             double inbits = s->avctx->rc_max_rate*av_q2d(s->avctx->time_base);
1352             int    minbits= s->frame_bits - 8*(s->vbv_delay_ptr - s->pb.buf - 1);
1353             double bits   = s->rc_context.buffer_index + minbits - inbits;
1354
1355             if(bits<0)
1356                 av_log(s->avctx, AV_LOG_ERROR, "Internal error, negative bits\n");
1357
1358             assert(s->repeat_first_field==0);
1359
1360             vbv_delay=     bits * 90000                               / s->avctx->rc_max_rate;
1361             min_delay= (minbits * 90000LL + s->avctx->rc_max_rate - 1)/ s->avctx->rc_max_rate;
1362
1363             vbv_delay= FFMAX(vbv_delay, min_delay);
1364
1365             assert(vbv_delay < 0xFFFF);
1366
1367             s->vbv_delay_ptr[0] &= 0xF8;
1368             s->vbv_delay_ptr[0] |= vbv_delay>>13;
1369             s->vbv_delay_ptr[1]  = vbv_delay>>5;
1370             s->vbv_delay_ptr[2] &= 0x07;
1371             s->vbv_delay_ptr[2] |= vbv_delay<<3;
1372             avctx->vbv_delay = vbv_delay*300;
1373         }
1374         s->total_bits += s->frame_bits;
1375         avctx->frame_bits  = s->frame_bits;
1376     }else{
1377         assert((put_bits_ptr(&s->pb) == s->pb.buf));
1378         s->frame_bits=0;
1379     }
1380     assert((s->frame_bits&7)==0);
1381
1382     return s->frame_bits/8;
1383 }
1384
1385 static inline void dct_single_coeff_elimination(MpegEncContext *s, int n, int threshold)
1386 {
1387     static const char tab[64]=
1388         {3,2,2,1,1,1,1,1,
1389          1,1,1,1,1,1,1,1,
1390          1,1,1,1,1,1,1,1,
1391          0,0,0,0,0,0,0,0,
1392          0,0,0,0,0,0,0,0,
1393          0,0,0,0,0,0,0,0,
1394          0,0,0,0,0,0,0,0,
1395          0,0,0,0,0,0,0,0};
1396     int score=0;
1397     int run=0;
1398     int i;
1399     DCTELEM *block= s->block[n];
1400     const int last_index= s->block_last_index[n];
1401     int skip_dc;
1402
1403     if(threshold<0){
1404         skip_dc=0;
1405         threshold= -threshold;
1406     }else
1407         skip_dc=1;
1408
1409     /* Are all we could set to zero already zero? */
1410     if(last_index<=skip_dc - 1) return;
1411
1412     for(i=0; i<=last_index; i++){
1413         const int j = s->intra_scantable.permutated[i];
1414         const int level = FFABS(block[j]);
1415         if(level==1){
1416             if(skip_dc && i==0) continue;
1417             score+= tab[run];
1418             run=0;
1419         }else if(level>1){
1420             return;
1421         }else{
1422             run++;
1423         }
1424     }
1425     if(score >= threshold) return;
1426     for(i=skip_dc; i<=last_index; i++){
1427         const int j = s->intra_scantable.permutated[i];
1428         block[j]=0;
1429     }
1430     if(block[0]) s->block_last_index[n]= 0;
1431     else         s->block_last_index[n]= -1;
1432 }
1433
1434 static inline void clip_coeffs(MpegEncContext *s, DCTELEM *block, int last_index)
1435 {
1436     int i;
1437     const int maxlevel= s->max_qcoeff;
1438     const int minlevel= s->min_qcoeff;
1439     int overflow=0;
1440
1441     if(s->mb_intra){
1442         i=1; //skip clipping of intra dc
1443     }else
1444         i=0;
1445
1446     for(;i<=last_index; i++){
1447         const int j= s->intra_scantable.permutated[i];
1448         int level = block[j];
1449
1450         if     (level>maxlevel){
1451             level=maxlevel;
1452             overflow++;
1453         }else if(level<minlevel){
1454             level=minlevel;
1455             overflow++;
1456         }
1457
1458         block[j]= level;
1459     }
1460
1461     if(overflow && s->avctx->mb_decision == FF_MB_DECISION_SIMPLE)
1462         av_log(s->avctx, AV_LOG_INFO, "warning, clipping %d dct coefficients to %d..%d\n", overflow, minlevel, maxlevel);
1463 }
1464
1465 static void get_visual_weight(int16_t *weight, uint8_t *ptr, int stride){
1466     int x, y;
1467 //FIXME optimize
1468     for(y=0; y<8; y++){
1469         for(x=0; x<8; x++){
1470             int x2, y2;
1471             int sum=0;
1472             int sqr=0;
1473             int count=0;
1474
1475             for(y2= FFMAX(y-1, 0); y2 < FFMIN(8, y+2); y2++){
1476                 for(x2= FFMAX(x-1, 0); x2 < FFMIN(8, x+2); x2++){
1477                     int v= ptr[x2 + y2*stride];
1478                     sum += v;
1479                     sqr += v*v;
1480                     count++;
1481                 }
1482             }
1483             weight[x + 8*y]= (36*ff_sqrt(count*sqr - sum*sum)) / count;
1484         }
1485     }
1486 }
1487
1488 static av_always_inline void encode_mb_internal(MpegEncContext *s, int motion_x, int motion_y, int mb_block_height, int mb_block_count)
1489 {
1490     int16_t weight[8][64];
1491     DCTELEM orig[8][64];
1492     const int mb_x= s->mb_x;
1493     const int mb_y= s->mb_y;
1494     int i;
1495     int skip_dct[8];
1496     int dct_offset   = s->linesize*8; //default for progressive frames
1497     uint8_t *ptr_y, *ptr_cb, *ptr_cr;
1498     int wrap_y, wrap_c;
1499
1500     for(i=0; i<mb_block_count; i++) skip_dct[i]=s->skipdct;
1501
1502     if(s->adaptive_quant){
1503         const int last_qp= s->qscale;
1504         const int mb_xy= mb_x + mb_y*s->mb_stride;
1505
1506         s->lambda= s->lambda_table[mb_xy];
1507         update_qscale(s);
1508
1509         if(!(s->flags&CODEC_FLAG_QP_RD)){
1510             s->qscale= s->current_picture_ptr->qscale_table[mb_xy];
1511             s->dquant= s->qscale - last_qp;
1512
1513             if(s->out_format==FMT_H263){
1514                 s->dquant= av_clip(s->dquant, -2, 2);
1515
1516                 if(s->codec_id==CODEC_ID_MPEG4){
1517                     if(!s->mb_intra){
1518                         if(s->pict_type == AV_PICTURE_TYPE_B){
1519                             if(s->dquant&1 || s->mv_dir&MV_DIRECT)
1520                                 s->dquant= 0;
1521                         }
1522                         if(s->mv_type==MV_TYPE_8X8)
1523                             s->dquant=0;
1524                     }
1525                 }
1526             }
1527         }
1528         ff_set_qscale(s, last_qp + s->dquant);
1529     }else if(s->flags&CODEC_FLAG_QP_RD)
1530         ff_set_qscale(s, s->qscale + s->dquant);
1531
1532     wrap_y = s->linesize;
1533     wrap_c = s->uvlinesize;
1534     ptr_y = s->new_picture.data[0] + (mb_y * 16 * wrap_y) + mb_x * 16;
1535     ptr_cb = s->new_picture.data[1] + (mb_y * mb_block_height * wrap_c) + mb_x * 8;
1536     ptr_cr = s->new_picture.data[2] + (mb_y * mb_block_height * wrap_c) + mb_x * 8;
1537
1538     if(mb_x*16+16 > s->width || mb_y*16+16 > s->height){
1539         uint8_t *ebuf= s->edge_emu_buffer + 32;
1540         s->dsp.emulated_edge_mc(ebuf            , ptr_y , wrap_y,16,16,mb_x*16,mb_y*16, s->width   , s->height);
1541         ptr_y= ebuf;
1542         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);
1543         ptr_cb= ebuf+18*wrap_y;
1544         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);
1545         ptr_cr= ebuf+18*wrap_y+8;
1546     }
1547
1548     if (s->mb_intra) {
1549         if(s->flags&CODEC_FLAG_INTERLACED_DCT){
1550             int progressive_score, interlaced_score;
1551
1552             s->interlaced_dct=0;
1553             progressive_score= s->dsp.ildct_cmp[4](s, ptr_y           , NULL, wrap_y, 8)
1554                               +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y*8, NULL, wrap_y, 8) - 400;
1555
1556             if(progressive_score > 0){
1557                 interlaced_score = s->dsp.ildct_cmp[4](s, ptr_y           , NULL, wrap_y*2, 8)
1558                                   +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y  , NULL, wrap_y*2, 8);
1559                 if(progressive_score > interlaced_score){
1560                     s->interlaced_dct=1;
1561
1562                     dct_offset= wrap_y;
1563                     wrap_y<<=1;
1564                     if (s->chroma_format == CHROMA_422)
1565                         wrap_c<<=1;
1566                 }
1567             }
1568         }
1569
1570         s->dsp.get_pixels(s->block[0], ptr_y                 , wrap_y);
1571         s->dsp.get_pixels(s->block[1], ptr_y              + 8, wrap_y);
1572         s->dsp.get_pixels(s->block[2], ptr_y + dct_offset    , wrap_y);
1573         s->dsp.get_pixels(s->block[3], ptr_y + dct_offset + 8, wrap_y);
1574
1575         if(s->flags&CODEC_FLAG_GRAY){
1576             skip_dct[4]= 1;
1577             skip_dct[5]= 1;
1578         }else{
1579             s->dsp.get_pixels(s->block[4], ptr_cb, wrap_c);
1580             s->dsp.get_pixels(s->block[5], ptr_cr, wrap_c);
1581             if(!s->chroma_y_shift){ /* 422 */
1582                 s->dsp.get_pixels(s->block[6], ptr_cb + (dct_offset>>1), wrap_c);
1583                 s->dsp.get_pixels(s->block[7], ptr_cr + (dct_offset>>1), wrap_c);
1584             }
1585         }
1586     }else{
1587         op_pixels_func (*op_pix)[4];
1588         qpel_mc_func (*op_qpix)[16];
1589         uint8_t *dest_y, *dest_cb, *dest_cr;
1590
1591         dest_y  = s->dest[0];
1592         dest_cb = s->dest[1];
1593         dest_cr = s->dest[2];
1594
1595         if ((!s->no_rounding) || s->pict_type==AV_PICTURE_TYPE_B){
1596             op_pix = s->dsp.put_pixels_tab;
1597             op_qpix= s->dsp.put_qpel_pixels_tab;
1598         }else{
1599             op_pix = s->dsp.put_no_rnd_pixels_tab;
1600             op_qpix= s->dsp.put_no_rnd_qpel_pixels_tab;
1601         }
1602
1603         if (s->mv_dir & MV_DIR_FORWARD) {
1604             MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.data, op_pix, op_qpix);
1605             op_pix = s->dsp.avg_pixels_tab;
1606             op_qpix= s->dsp.avg_qpel_pixels_tab;
1607         }
1608         if (s->mv_dir & MV_DIR_BACKWARD) {
1609             MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.data, op_pix, op_qpix);
1610         }
1611
1612         if(s->flags&CODEC_FLAG_INTERLACED_DCT){
1613             int progressive_score, interlaced_score;
1614
1615             s->interlaced_dct=0;
1616             progressive_score= s->dsp.ildct_cmp[0](s, dest_y           , ptr_y           , wrap_y, 8)
1617                               +s->dsp.ildct_cmp[0](s, dest_y + wrap_y*8, ptr_y + wrap_y*8, wrap_y, 8) - 400;
1618
1619             if(s->avctx->ildct_cmp == FF_CMP_VSSE) progressive_score -= 400;
1620
1621             if(progressive_score>0){
1622                 interlaced_score = s->dsp.ildct_cmp[0](s, dest_y           , ptr_y           , wrap_y*2, 8)
1623                                   +s->dsp.ildct_cmp[0](s, dest_y + wrap_y  , ptr_y + wrap_y  , wrap_y*2, 8);
1624
1625                 if(progressive_score > interlaced_score){
1626                     s->interlaced_dct=1;
1627
1628                     dct_offset= wrap_y;
1629                     wrap_y<<=1;
1630                     if (s->chroma_format == CHROMA_422)
1631                         wrap_c<<=1;
1632                 }
1633             }
1634         }
1635
1636         s->dsp.diff_pixels(s->block[0], ptr_y                 , dest_y                 , wrap_y);
1637         s->dsp.diff_pixels(s->block[1], ptr_y              + 8, dest_y              + 8, wrap_y);
1638         s->dsp.diff_pixels(s->block[2], ptr_y + dct_offset    , dest_y + dct_offset    , wrap_y);
1639         s->dsp.diff_pixels(s->block[3], ptr_y + dct_offset + 8, dest_y + dct_offset + 8, wrap_y);
1640
1641         if(s->flags&CODEC_FLAG_GRAY){
1642             skip_dct[4]= 1;
1643             skip_dct[5]= 1;
1644         }else{
1645             s->dsp.diff_pixels(s->block[4], ptr_cb, dest_cb, wrap_c);
1646             s->dsp.diff_pixels(s->block[5], ptr_cr, dest_cr, wrap_c);
1647             if(!s->chroma_y_shift){ /* 422 */
1648                 s->dsp.diff_pixels(s->block[6], ptr_cb + (dct_offset>>1), dest_cb + (dct_offset>>1), wrap_c);
1649                 s->dsp.diff_pixels(s->block[7], ptr_cr + (dct_offset>>1), dest_cr + (dct_offset>>1), wrap_c);
1650             }
1651         }
1652         /* pre quantization */
1653         if(s->current_picture.mc_mb_var[s->mb_stride*mb_y+ mb_x]<2*s->qscale*s->qscale){
1654             //FIXME optimize
1655             if(s->dsp.sad[1](NULL, ptr_y               , dest_y               , wrap_y, 8) < 20*s->qscale) skip_dct[0]= 1;
1656             if(s->dsp.sad[1](NULL, ptr_y            + 8, dest_y            + 8, wrap_y, 8) < 20*s->qscale) skip_dct[1]= 1;
1657             if(s->dsp.sad[1](NULL, ptr_y +dct_offset   , dest_y +dct_offset   , wrap_y, 8) < 20*s->qscale) skip_dct[2]= 1;
1658             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;
1659             if(s->dsp.sad[1](NULL, ptr_cb              , dest_cb              , wrap_c, 8) < 20*s->qscale) skip_dct[4]= 1;
1660             if(s->dsp.sad[1](NULL, ptr_cr              , dest_cr              , wrap_c, 8) < 20*s->qscale) skip_dct[5]= 1;
1661             if(!s->chroma_y_shift){ /* 422 */
1662                 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;
1663                 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;
1664             }
1665         }
1666     }
1667
1668     if(s->avctx->quantizer_noise_shaping){
1669         if(!skip_dct[0]) get_visual_weight(weight[0], ptr_y                 , wrap_y);
1670         if(!skip_dct[1]) get_visual_weight(weight[1], ptr_y              + 8, wrap_y);
1671         if(!skip_dct[2]) get_visual_weight(weight[2], ptr_y + dct_offset    , wrap_y);
1672         if(!skip_dct[3]) get_visual_weight(weight[3], ptr_y + dct_offset + 8, wrap_y);
1673         if(!skip_dct[4]) get_visual_weight(weight[4], ptr_cb                , wrap_c);
1674         if(!skip_dct[5]) get_visual_weight(weight[5], ptr_cr                , wrap_c);
1675         if(!s->chroma_y_shift){ /* 422 */
1676             if(!skip_dct[6]) get_visual_weight(weight[6], ptr_cb + (dct_offset>>1), wrap_c);
1677             if(!skip_dct[7]) get_visual_weight(weight[7], ptr_cr + (dct_offset>>1), wrap_c);
1678         }
1679         memcpy(orig[0], s->block[0], sizeof(DCTELEM)*64*mb_block_count);
1680     }
1681
1682     /* DCT & quantize */
1683     assert(s->out_format!=FMT_MJPEG || s->qscale==8);
1684     {
1685         for(i=0;i<mb_block_count;i++) {
1686             if(!skip_dct[i]){
1687                 int overflow;
1688                 s->block_last_index[i] = s->dct_quantize(s, s->block[i], i, s->qscale, &overflow);
1689             // FIXME we could decide to change to quantizer instead of clipping
1690             // JS: I don't think that would be a good idea it could lower quality instead
1691             //     of improve it. Just INTRADC clipping deserves changes in quantizer
1692                 if (overflow) clip_coeffs(s, s->block[i], s->block_last_index[i]);
1693             }else
1694                 s->block_last_index[i]= -1;
1695         }
1696         if(s->avctx->quantizer_noise_shaping){
1697             for(i=0;i<mb_block_count;i++) {
1698                 if(!skip_dct[i]){
1699                     s->block_last_index[i] = dct_quantize_refine(s, s->block[i], weight[i], orig[i], i, s->qscale);
1700                 }
1701             }
1702         }
1703
1704         if(s->luma_elim_threshold && !s->mb_intra)
1705             for(i=0; i<4; i++)
1706                 dct_single_coeff_elimination(s, i, s->luma_elim_threshold);
1707         if(s->chroma_elim_threshold && !s->mb_intra)
1708             for(i=4; i<mb_block_count; i++)
1709                 dct_single_coeff_elimination(s, i, s->chroma_elim_threshold);
1710
1711         if(s->flags & CODEC_FLAG_CBP_RD){
1712             for(i=0;i<mb_block_count;i++) {
1713                 if(s->block_last_index[i] == -1)
1714                     s->coded_score[i]= INT_MAX/256;
1715             }
1716         }
1717     }
1718
1719     if((s->flags&CODEC_FLAG_GRAY) && s->mb_intra){
1720         s->block_last_index[4]=
1721         s->block_last_index[5]= 0;
1722         s->block[4][0]=
1723         s->block[5][0]= (1024 + s->c_dc_scale/2)/ s->c_dc_scale;
1724     }
1725
1726     //non c quantize code returns incorrect block_last_index FIXME
1727     if(s->alternate_scan && s->dct_quantize != dct_quantize_c){
1728         for(i=0; i<mb_block_count; i++){
1729             int j;
1730             if(s->block_last_index[i]>0){
1731                 for(j=63; j>0; j--){
1732                     if(s->block[i][ s->intra_scantable.permutated[j] ]) break;
1733                 }
1734                 s->block_last_index[i]= j;
1735             }
1736         }
1737     }
1738
1739     /* huffman encode */
1740     switch(s->codec_id){ //FIXME funct ptr could be slightly faster
1741     case CODEC_ID_MPEG1VIDEO:
1742     case CODEC_ID_MPEG2VIDEO:
1743         if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
1744             mpeg1_encode_mb(s, s->block, motion_x, motion_y);
1745         break;
1746     case CODEC_ID_MPEG4:
1747         if (CONFIG_MPEG4_ENCODER)
1748             mpeg4_encode_mb(s, s->block, motion_x, motion_y);
1749         break;
1750     case CODEC_ID_MSMPEG4V2:
1751     case CODEC_ID_MSMPEG4V3:
1752     case CODEC_ID_WMV1:
1753         if (CONFIG_MSMPEG4_ENCODER)
1754             msmpeg4_encode_mb(s, s->block, motion_x, motion_y);
1755         break;
1756     case CODEC_ID_WMV2:
1757         if (CONFIG_WMV2_ENCODER)
1758             ff_wmv2_encode_mb(s, s->block, motion_x, motion_y);
1759         break;
1760     case CODEC_ID_H261:
1761         if (CONFIG_H261_ENCODER)
1762             ff_h261_encode_mb(s, s->block, motion_x, motion_y);
1763         break;
1764     case CODEC_ID_H263:
1765     case CODEC_ID_H263P:
1766     case CODEC_ID_FLV1:
1767     case CODEC_ID_RV10:
1768     case CODEC_ID_RV20:
1769         if (CONFIG_H263_ENCODER)
1770             h263_encode_mb(s, s->block, motion_x, motion_y);
1771         break;
1772     case CODEC_ID_MJPEG:
1773         if (CONFIG_MJPEG_ENCODER)
1774             ff_mjpeg_encode_mb(s, s->block);
1775         break;
1776     default:
1777         assert(0);
1778     }
1779 }
1780
1781 static av_always_inline void encode_mb(MpegEncContext *s, int motion_x, int motion_y)
1782 {
1783     if (s->chroma_format == CHROMA_420) encode_mb_internal(s, motion_x, motion_y,  8, 6);
1784     else                                encode_mb_internal(s, motion_x, motion_y, 16, 8);
1785 }
1786
1787 static inline void copy_context_before_encode(MpegEncContext *d, MpegEncContext *s, int type){
1788     int i;
1789
1790     memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster than a loop?
1791
1792     /* mpeg1 */
1793     d->mb_skip_run= s->mb_skip_run;
1794     for(i=0; i<3; i++)
1795         d->last_dc[i]= s->last_dc[i];
1796
1797     /* statistics */
1798     d->mv_bits= s->mv_bits;
1799     d->i_tex_bits= s->i_tex_bits;
1800     d->p_tex_bits= s->p_tex_bits;
1801     d->i_count= s->i_count;
1802     d->f_count= s->f_count;
1803     d->b_count= s->b_count;
1804     d->skip_count= s->skip_count;
1805     d->misc_bits= s->misc_bits;
1806     d->last_bits= 0;
1807
1808     d->mb_skipped= 0;
1809     d->qscale= s->qscale;
1810     d->dquant= s->dquant;
1811
1812     d->esc3_level_length= s->esc3_level_length;
1813 }
1814
1815 static inline void copy_context_after_encode(MpegEncContext *d, MpegEncContext *s, int type){
1816     int i;
1817
1818     memcpy(d->mv, s->mv, 2*4*2*sizeof(int));
1819     memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster than a loop?
1820
1821     /* mpeg1 */
1822     d->mb_skip_run= s->mb_skip_run;
1823     for(i=0; i<3; i++)
1824         d->last_dc[i]= s->last_dc[i];
1825
1826     /* statistics */
1827     d->mv_bits= s->mv_bits;
1828     d->i_tex_bits= s->i_tex_bits;
1829     d->p_tex_bits= s->p_tex_bits;
1830     d->i_count= s->i_count;
1831     d->f_count= s->f_count;
1832     d->b_count= s->b_count;
1833     d->skip_count= s->skip_count;
1834     d->misc_bits= s->misc_bits;
1835
1836     d->mb_intra= s->mb_intra;
1837     d->mb_skipped= s->mb_skipped;
1838     d->mv_type= s->mv_type;
1839     d->mv_dir= s->mv_dir;
1840     d->pb= s->pb;
1841     if(s->data_partitioning){
1842         d->pb2= s->pb2;
1843         d->tex_pb= s->tex_pb;
1844     }
1845     d->block= s->block;
1846     for(i=0; i<8; i++)
1847         d->block_last_index[i]= s->block_last_index[i];
1848     d->interlaced_dct= s->interlaced_dct;
1849     d->qscale= s->qscale;
1850
1851     d->esc3_level_length= s->esc3_level_length;
1852 }
1853
1854 static inline void encode_mb_hq(MpegEncContext *s, MpegEncContext *backup, MpegEncContext *best, int type,
1855                            PutBitContext pb[2], PutBitContext pb2[2], PutBitContext tex_pb[2],
1856                            int *dmin, int *next_block, int motion_x, int motion_y)
1857 {
1858     int score;
1859     uint8_t *dest_backup[3];
1860
1861     copy_context_before_encode(s, backup, type);
1862
1863     s->block= s->blocks[*next_block];
1864     s->pb= pb[*next_block];
1865     if(s->data_partitioning){
1866         s->pb2   = pb2   [*next_block];
1867         s->tex_pb= tex_pb[*next_block];
1868     }
1869
1870     if(*next_block){
1871         memcpy(dest_backup, s->dest, sizeof(s->dest));
1872         s->dest[0] = s->rd_scratchpad;
1873         s->dest[1] = s->rd_scratchpad + 16*s->linesize;
1874         s->dest[2] = s->rd_scratchpad + 16*s->linesize + 8;
1875         assert(s->linesize >= 32); //FIXME
1876     }
1877
1878     encode_mb(s, motion_x, motion_y);
1879
1880     score= put_bits_count(&s->pb);
1881     if(s->data_partitioning){
1882         score+= put_bits_count(&s->pb2);
1883         score+= put_bits_count(&s->tex_pb);
1884     }
1885
1886     if(s->avctx->mb_decision == FF_MB_DECISION_RD){
1887         MPV_decode_mb(s, s->block);
1888
1889         score *= s->lambda2;
1890         score += sse_mb(s) << FF_LAMBDA_SHIFT;
1891     }
1892
1893     if(*next_block){
1894         memcpy(s->dest, dest_backup, sizeof(s->dest));
1895     }
1896
1897     if(score<*dmin){
1898         *dmin= score;
1899         *next_block^=1;
1900
1901         copy_context_after_encode(best, s, type);
1902     }
1903 }
1904
1905 static int sse(MpegEncContext *s, uint8_t *src1, uint8_t *src2, int w, int h, int stride){
1906     uint32_t *sq = ff_squareTbl + 256;
1907     int acc=0;
1908     int x,y;
1909
1910     if(w==16 && h==16)
1911         return s->dsp.sse[0](NULL, src1, src2, stride, 16);
1912     else if(w==8 && h==8)
1913         return s->dsp.sse[1](NULL, src1, src2, stride, 8);
1914
1915     for(y=0; y<h; y++){
1916         for(x=0; x<w; x++){
1917             acc+= sq[src1[x + y*stride] - src2[x + y*stride]];
1918         }
1919     }
1920
1921     assert(acc>=0);
1922
1923     return acc;
1924 }
1925
1926 static int sse_mb(MpegEncContext *s){
1927     int w= 16;
1928     int h= 16;
1929
1930     if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
1931     if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
1932
1933     if(w==16 && h==16)
1934       if(s->avctx->mb_cmp == FF_CMP_NSSE){
1935         return  s->dsp.nsse[0](s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, s->dest[0], s->linesize, 16)
1936                +s->dsp.nsse[1](s, s->new_picture.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[1], s->uvlinesize, 8)
1937                +s->dsp.nsse[1](s, s->new_picture.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[2], s->uvlinesize, 8);
1938       }else{
1939         return  s->dsp.sse[0](NULL, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, s->dest[0], s->linesize, 16)
1940                +s->dsp.sse[1](NULL, s->new_picture.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[1], s->uvlinesize, 8)
1941                +s->dsp.sse[1](NULL, s->new_picture.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[2], s->uvlinesize, 8);
1942       }
1943     else
1944         return  sse(s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, s->dest[0], w, h, s->linesize)
1945                +sse(s, s->new_picture.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[1], w>>1, h>>1, s->uvlinesize)
1946                +sse(s, s->new_picture.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[2], w>>1, h>>1, s->uvlinesize);
1947 }
1948
1949 static int pre_estimate_motion_thread(AVCodecContext *c, void *arg){
1950     MpegEncContext *s= *(void**)arg;
1951
1952
1953     s->me.pre_pass=1;
1954     s->me.dia_size= s->avctx->pre_dia_size;
1955     s->first_slice_line=1;
1956     for(s->mb_y= s->end_mb_y-1; s->mb_y >= s->start_mb_y; s->mb_y--) {
1957         for(s->mb_x=s->mb_width-1; s->mb_x >=0 ;s->mb_x--) {
1958             ff_pre_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
1959         }
1960         s->first_slice_line=0;
1961     }
1962
1963     s->me.pre_pass=0;
1964
1965     return 0;
1966 }
1967
1968 static int estimate_motion_thread(AVCodecContext *c, void *arg){
1969     MpegEncContext *s= *(void**)arg;
1970
1971     ff_check_alignment();
1972
1973     s->me.dia_size= s->avctx->dia_size;
1974     s->first_slice_line=1;
1975     for(s->mb_y= s->start_mb_y; s->mb_y < s->end_mb_y; s->mb_y++) {
1976         s->mb_x=0; //for block init below
1977         ff_init_block_index(s);
1978         for(s->mb_x=0; s->mb_x < s->mb_width; s->mb_x++) {
1979             s->block_index[0]+=2;
1980             s->block_index[1]+=2;
1981             s->block_index[2]+=2;
1982             s->block_index[3]+=2;
1983
1984             /* compute motion vector & mb_type and store in context */
1985             if(s->pict_type==AV_PICTURE_TYPE_B)
1986                 ff_estimate_b_frame_motion(s, s->mb_x, s->mb_y);
1987             else
1988                 ff_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
1989         }
1990         s->first_slice_line=0;
1991     }
1992     return 0;
1993 }
1994
1995 static int mb_var_thread(AVCodecContext *c, void *arg){
1996     MpegEncContext *s= *(void**)arg;
1997     int mb_x, mb_y;
1998
1999     ff_check_alignment();
2000
2001     for(mb_y=s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
2002         for(mb_x=0; mb_x < s->mb_width; mb_x++) {
2003             int xx = mb_x * 16;
2004             int yy = mb_y * 16;
2005             uint8_t *pix = s->new_picture.data[0] + (yy * s->linesize) + xx;
2006             int varc;
2007             int sum = s->dsp.pix_sum(pix, s->linesize);
2008
2009             varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500 + 128)>>8;
2010
2011             s->current_picture.mb_var [s->mb_stride * mb_y + mb_x] = varc;
2012             s->current_picture.mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
2013             s->me.mb_var_sum_temp    += varc;
2014         }
2015     }
2016     return 0;
2017 }
2018
2019 static void write_slice_end(MpegEncContext *s){
2020     if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4){
2021         if(s->partitioned_frame){
2022             ff_mpeg4_merge_partitions(s);
2023         }
2024
2025         ff_mpeg4_stuffing(&s->pb);
2026     }else if(CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG){
2027         ff_mjpeg_encode_stuffing(&s->pb);
2028     }
2029
2030     align_put_bits(&s->pb);
2031     flush_put_bits(&s->pb);
2032
2033     if((s->flags&CODEC_FLAG_PASS1) && !s->partitioned_frame)
2034         s->misc_bits+= get_bits_diff(s);
2035 }
2036
2037 static int encode_thread(AVCodecContext *c, void *arg){
2038     MpegEncContext *s= *(void**)arg;
2039     int mb_x, mb_y, pdif = 0;
2040     int chr_h= 16>>s->chroma_y_shift;
2041     int i, j;
2042     MpegEncContext best_s, backup_s;
2043     uint8_t bit_buf[2][MAX_MB_BYTES];
2044     uint8_t bit_buf2[2][MAX_MB_BYTES];
2045     uint8_t bit_buf_tex[2][MAX_MB_BYTES];
2046     PutBitContext pb[2], pb2[2], tex_pb[2];
2047 //printf("%d->%d\n", s->resync_mb_y, s->end_mb_y);
2048
2049     ff_check_alignment();
2050
2051     for(i=0; i<2; i++){
2052         init_put_bits(&pb    [i], bit_buf    [i], MAX_MB_BYTES);
2053         init_put_bits(&pb2   [i], bit_buf2   [i], MAX_MB_BYTES);
2054         init_put_bits(&tex_pb[i], bit_buf_tex[i], MAX_MB_BYTES);
2055     }
2056
2057     s->last_bits= put_bits_count(&s->pb);
2058     s->mv_bits=0;
2059     s->misc_bits=0;
2060     s->i_tex_bits=0;
2061     s->p_tex_bits=0;
2062     s->i_count=0;
2063     s->f_count=0;
2064     s->b_count=0;
2065     s->skip_count=0;
2066
2067     for(i=0; i<3; i++){
2068         /* init last dc values */
2069         /* note: quant matrix value (8) is implied here */
2070         s->last_dc[i] = 128 << s->intra_dc_precision;
2071
2072         s->current_picture.error[i] = 0;
2073     }
2074     s->mb_skip_run = 0;
2075     memset(s->last_mv, 0, sizeof(s->last_mv));
2076
2077     s->last_mv_dir = 0;
2078
2079     switch(s->codec_id){
2080     case CODEC_ID_H263:
2081     case CODEC_ID_H263P:
2082     case CODEC_ID_FLV1:
2083         if (CONFIG_H263_ENCODER)
2084             s->gob_index = ff_h263_get_gob_height(s);
2085         break;
2086     case CODEC_ID_MPEG4:
2087         if(CONFIG_MPEG4_ENCODER && s->partitioned_frame)
2088             ff_mpeg4_init_partitions(s);
2089         break;
2090     }
2091
2092     s->resync_mb_x=0;
2093     s->resync_mb_y=0;
2094     s->first_slice_line = 1;
2095     s->ptr_lastgob = s->pb.buf;
2096     for(mb_y= s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
2097 //    printf("row %d at %X\n", s->mb_y, (int)s);
2098         s->mb_x=0;
2099         s->mb_y= mb_y;
2100
2101         ff_set_qscale(s, s->qscale);
2102         ff_init_block_index(s);
2103
2104         for(mb_x=0; mb_x < s->mb_width; mb_x++) {
2105             int xy= mb_y*s->mb_stride + mb_x; // removed const, H261 needs to adjust this
2106             int mb_type= s->mb_type[xy];
2107 //            int d;
2108             int dmin= INT_MAX;
2109             int dir;
2110
2111             if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < MAX_MB_BYTES){
2112                 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
2113                 return -1;
2114             }
2115             if(s->data_partitioning){
2116                 if(   s->pb2   .buf_end - s->pb2   .buf - (put_bits_count(&s->    pb2)>>3) < MAX_MB_BYTES
2117                    || s->tex_pb.buf_end - s->tex_pb.buf - (put_bits_count(&s->tex_pb )>>3) < MAX_MB_BYTES){
2118                     av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
2119                     return -1;
2120                 }
2121             }
2122
2123             s->mb_x = mb_x;
2124             s->mb_y = mb_y;  // moved into loop, can get changed by H.261
2125             ff_update_block_index(s);
2126
2127             if(CONFIG_H261_ENCODER && s->codec_id == CODEC_ID_H261){
2128                 ff_h261_reorder_mb_index(s);
2129                 xy= s->mb_y*s->mb_stride + s->mb_x;
2130                 mb_type= s->mb_type[xy];
2131             }
2132
2133             /* write gob / video packet header  */
2134             if(s->rtp_mode){
2135                 int current_packet_size, is_gob_start;
2136
2137                 current_packet_size= ((put_bits_count(&s->pb)+7)>>3) - (s->ptr_lastgob - s->pb.buf);
2138
2139                 is_gob_start= s->avctx->rtp_payload_size && current_packet_size >= s->avctx->rtp_payload_size && mb_y + mb_x>0;
2140
2141                 if(s->start_mb_y == mb_y && mb_y > 0 && mb_x==0) is_gob_start=1;
2142
2143                 switch(s->codec_id){
2144                 case CODEC_ID_H263:
2145                 case CODEC_ID_H263P:
2146                     if(!s->h263_slice_structured)
2147                         if(s->mb_x || s->mb_y%s->gob_index) is_gob_start=0;
2148                     break;
2149                 case CODEC_ID_MPEG2VIDEO:
2150                     if(s->mb_x==0 && s->mb_y!=0) is_gob_start=1;
2151                 case CODEC_ID_MPEG1VIDEO:
2152                     if(s->mb_skip_run) is_gob_start=0;
2153                     break;
2154                 }
2155
2156                 if(is_gob_start){
2157                     if(s->start_mb_y != mb_y || mb_x!=0){
2158                         write_slice_end(s);
2159
2160                         if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4 && s->partitioned_frame){
2161                             ff_mpeg4_init_partitions(s);
2162                         }
2163                     }
2164
2165                     assert((put_bits_count(&s->pb)&7) == 0);
2166                     current_packet_size= put_bits_ptr(&s->pb) - s->ptr_lastgob;
2167
2168                     if(s->avctx->error_rate && s->resync_mb_x + s->resync_mb_y > 0){
2169                         int r= put_bits_count(&s->pb)/8 + s->picture_number + 16 + s->mb_x + s->mb_y;
2170                         int d= 100 / s->avctx->error_rate;
2171                         if(r % d == 0){
2172                             current_packet_size=0;
2173                             s->pb.buf_ptr= s->ptr_lastgob;
2174                             assert(put_bits_ptr(&s->pb) == s->ptr_lastgob);
2175                         }
2176                     }
2177
2178                     if (s->avctx->rtp_callback){
2179                         int number_mb = (mb_y - s->resync_mb_y)*s->mb_width + mb_x - s->resync_mb_x;
2180                         s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, current_packet_size, number_mb);
2181                     }
2182
2183                     switch(s->codec_id){
2184                     case CODEC_ID_MPEG4:
2185                         if (CONFIG_MPEG4_ENCODER) {
2186                             ff_mpeg4_encode_video_packet_header(s);
2187                             ff_mpeg4_clean_buffers(s);
2188                         }
2189                     break;
2190                     case CODEC_ID_MPEG1VIDEO:
2191                     case CODEC_ID_MPEG2VIDEO:
2192                         if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER) {
2193                             ff_mpeg1_encode_slice_header(s);
2194                             ff_mpeg1_clean_buffers(s);
2195                         }
2196                     break;
2197                     case CODEC_ID_H263:
2198                     case CODEC_ID_H263P:
2199                         if (CONFIG_H263_ENCODER)
2200                             h263_encode_gob_header(s, mb_y);
2201                     break;
2202                     }
2203
2204                     if(s->flags&CODEC_FLAG_PASS1){
2205                         int bits= put_bits_count(&s->pb);
2206                         s->misc_bits+= bits - s->last_bits;
2207                         s->last_bits= bits;
2208                     }
2209
2210                     s->ptr_lastgob += current_packet_size;
2211                     s->first_slice_line=1;
2212                     s->resync_mb_x=mb_x;
2213                     s->resync_mb_y=mb_y;
2214                 }
2215             }
2216
2217             if(  (s->resync_mb_x   == s->mb_x)
2218                && s->resync_mb_y+1 == s->mb_y){
2219                 s->first_slice_line=0;
2220             }
2221
2222             s->mb_skipped=0;
2223             s->dquant=0; //only for QP_RD
2224
2225             if(mb_type & (mb_type-1) || (s->flags & CODEC_FLAG_QP_RD)){ // more than 1 MB type possible or CODEC_FLAG_QP_RD
2226                 int next_block=0;
2227                 int pb_bits_count, pb2_bits_count, tex_pb_bits_count;
2228
2229                 copy_context_before_encode(&backup_s, s, -1);
2230                 backup_s.pb= s->pb;
2231                 best_s.data_partitioning= s->data_partitioning;
2232                 best_s.partitioned_frame= s->partitioned_frame;
2233                 if(s->data_partitioning){
2234                     backup_s.pb2= s->pb2;
2235                     backup_s.tex_pb= s->tex_pb;
2236                 }
2237
2238                 if(mb_type&CANDIDATE_MB_TYPE_INTER){
2239                     s->mv_dir = MV_DIR_FORWARD;
2240                     s->mv_type = MV_TYPE_16X16;
2241                     s->mb_intra= 0;
2242                     s->mv[0][0][0] = s->p_mv_table[xy][0];
2243                     s->mv[0][0][1] = s->p_mv_table[xy][1];
2244                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER, pb, pb2, tex_pb,
2245                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2246                 }
2247                 if(mb_type&CANDIDATE_MB_TYPE_INTER_I){
2248                     s->mv_dir = MV_DIR_FORWARD;
2249                     s->mv_type = MV_TYPE_FIELD;
2250                     s->mb_intra= 0;
2251                     for(i=0; i<2; i++){
2252                         j= s->field_select[0][i] = s->p_field_select_table[i][xy];
2253                         s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
2254                         s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
2255                     }
2256                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER_I, pb, pb2, tex_pb,
2257                                  &dmin, &next_block, 0, 0);
2258                 }
2259                 if(mb_type&CANDIDATE_MB_TYPE_SKIPPED){
2260                     s->mv_dir = MV_DIR_FORWARD;
2261                     s->mv_type = MV_TYPE_16X16;
2262                     s->mb_intra= 0;
2263                     s->mv[0][0][0] = 0;
2264                     s->mv[0][0][1] = 0;
2265                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_SKIPPED, pb, pb2, tex_pb,
2266                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2267                 }
2268                 if(mb_type&CANDIDATE_MB_TYPE_INTER4V){
2269                     s->mv_dir = MV_DIR_FORWARD;
2270                     s->mv_type = MV_TYPE_8X8;
2271                     s->mb_intra= 0;
2272                     for(i=0; i<4; i++){
2273                         s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
2274                         s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
2275                     }
2276                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER4V, pb, pb2, tex_pb,
2277                                  &dmin, &next_block, 0, 0);
2278                 }
2279                 if(mb_type&CANDIDATE_MB_TYPE_FORWARD){
2280                     s->mv_dir = MV_DIR_FORWARD;
2281                     s->mv_type = MV_TYPE_16X16;
2282                     s->mb_intra= 0;
2283                     s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
2284                     s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2285                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD, pb, pb2, tex_pb,
2286                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2287                 }
2288                 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD){
2289                     s->mv_dir = MV_DIR_BACKWARD;
2290                     s->mv_type = MV_TYPE_16X16;
2291                     s->mb_intra= 0;
2292                     s->mv[1][0][0] = s->b_back_mv_table[xy][0];
2293                     s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2294                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD, pb, pb2, tex_pb,
2295                                  &dmin, &next_block, s->mv[1][0][0], s->mv[1][0][1]);
2296                 }
2297                 if(mb_type&CANDIDATE_MB_TYPE_BIDIR){
2298                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2299                     s->mv_type = MV_TYPE_16X16;
2300                     s->mb_intra= 0;
2301                     s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
2302                     s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
2303                     s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
2304                     s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2305                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR, pb, pb2, tex_pb,
2306                                  &dmin, &next_block, 0, 0);
2307                 }
2308                 if(mb_type&CANDIDATE_MB_TYPE_FORWARD_I){
2309                     s->mv_dir = MV_DIR_FORWARD;
2310                     s->mv_type = MV_TYPE_FIELD;
2311                     s->mb_intra= 0;
2312                     for(i=0; i<2; i++){
2313                         j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
2314                         s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
2315                         s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
2316                     }
2317                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD_I, pb, pb2, tex_pb,
2318                                  &dmin, &next_block, 0, 0);
2319                 }
2320                 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD_I){
2321                     s->mv_dir = MV_DIR_BACKWARD;
2322                     s->mv_type = MV_TYPE_FIELD;
2323                     s->mb_intra= 0;
2324                     for(i=0; i<2; i++){
2325                         j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
2326                         s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
2327                         s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
2328                     }
2329                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD_I, pb, pb2, tex_pb,
2330                                  &dmin, &next_block, 0, 0);
2331                 }
2332                 if(mb_type&CANDIDATE_MB_TYPE_BIDIR_I){
2333                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2334                     s->mv_type = MV_TYPE_FIELD;
2335                     s->mb_intra= 0;
2336                     for(dir=0; dir<2; dir++){
2337                         for(i=0; i<2; i++){
2338                             j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
2339                             s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
2340                             s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
2341                         }
2342                     }
2343                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR_I, pb, pb2, tex_pb,
2344                                  &dmin, &next_block, 0, 0);
2345                 }
2346                 if(mb_type&CANDIDATE_MB_TYPE_INTRA){
2347                     s->mv_dir = 0;
2348                     s->mv_type = MV_TYPE_16X16;
2349                     s->mb_intra= 1;
2350                     s->mv[0][0][0] = 0;
2351                     s->mv[0][0][1] = 0;
2352                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTRA, pb, pb2, tex_pb,
2353                                  &dmin, &next_block, 0, 0);
2354                     if(s->h263_pred || s->h263_aic){
2355                         if(best_s.mb_intra)
2356                             s->mbintra_table[mb_x + mb_y*s->mb_stride]=1;
2357                         else
2358                             ff_clean_intra_table_entries(s); //old mode?
2359                     }
2360                 }
2361
2362                 if((s->flags & CODEC_FLAG_QP_RD) && dmin < INT_MAX){
2363                     if(best_s.mv_type==MV_TYPE_16X16){ //FIXME move 4mv after QPRD
2364                         const int last_qp= backup_s.qscale;
2365                         int qpi, qp, dc[6];
2366                         DCTELEM ac[6][16];
2367                         const int mvdir= (best_s.mv_dir&MV_DIR_BACKWARD) ? 1 : 0;
2368                         static const int dquant_tab[4]={-1,1,-2,2};
2369
2370                         assert(backup_s.dquant == 0);
2371
2372                         //FIXME intra
2373                         s->mv_dir= best_s.mv_dir;
2374                         s->mv_type = MV_TYPE_16X16;
2375                         s->mb_intra= best_s.mb_intra;
2376                         s->mv[0][0][0] = best_s.mv[0][0][0];
2377                         s->mv[0][0][1] = best_s.mv[0][0][1];
2378                         s->mv[1][0][0] = best_s.mv[1][0][0];
2379                         s->mv[1][0][1] = best_s.mv[1][0][1];
2380
2381                         qpi = s->pict_type == AV_PICTURE_TYPE_B ? 2 : 0;
2382                         for(; qpi<4; qpi++){
2383                             int dquant= dquant_tab[qpi];
2384                             qp= last_qp + dquant;
2385                             if(qp < s->avctx->qmin || qp > s->avctx->qmax)
2386                                 continue;
2387                             backup_s.dquant= dquant;
2388                             if(s->mb_intra && s->dc_val[0]){
2389                                 for(i=0; i<6; i++){
2390                                     dc[i]= s->dc_val[0][ s->block_index[i] ];
2391                                     memcpy(ac[i], s->ac_val[0][s->block_index[i]], sizeof(DCTELEM)*16);
2392                                 }
2393                             }
2394
2395                             encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
2396                                          &dmin, &next_block, s->mv[mvdir][0][0], s->mv[mvdir][0][1]);
2397                             if(best_s.qscale != qp){
2398                                 if(s->mb_intra && s->dc_val[0]){
2399                                     for(i=0; i<6; i++){
2400                                         s->dc_val[0][ s->block_index[i] ]= dc[i];
2401                                         memcpy(s->ac_val[0][s->block_index[i]], ac[i], sizeof(DCTELEM)*16);
2402                                     }
2403                                 }
2404                             }
2405                         }
2406                     }
2407                 }
2408                 if(CONFIG_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT){
2409                     int mx= s->b_direct_mv_table[xy][0];
2410                     int my= s->b_direct_mv_table[xy][1];
2411
2412                     backup_s.dquant = 0;
2413                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2414                     s->mb_intra= 0;
2415                     ff_mpeg4_set_direct_mv(s, mx, my);
2416                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
2417                                  &dmin, &next_block, mx, my);
2418                 }
2419                 if(CONFIG_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT0){
2420                     backup_s.dquant = 0;
2421                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2422                     s->mb_intra= 0;
2423                     ff_mpeg4_set_direct_mv(s, 0, 0);
2424                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
2425                                  &dmin, &next_block, 0, 0);
2426                 }
2427                 if(!best_s.mb_intra && s->flags2&CODEC_FLAG2_SKIP_RD){
2428                     int coded=0;
2429                     for(i=0; i<6; i++)
2430                         coded |= s->block_last_index[i];
2431                     if(coded){
2432                         int mx,my;
2433                         memcpy(s->mv, best_s.mv, sizeof(s->mv));
2434                         if(CONFIG_MPEG4_ENCODER && best_s.mv_dir & MV_DIRECT){
2435                             mx=my=0; //FIXME find the one we actually used
2436                             ff_mpeg4_set_direct_mv(s, mx, my);
2437                         }else if(best_s.mv_dir&MV_DIR_BACKWARD){
2438                             mx= s->mv[1][0][0];
2439                             my= s->mv[1][0][1];
2440                         }else{
2441                             mx= s->mv[0][0][0];
2442                             my= s->mv[0][0][1];
2443                         }
2444
2445                         s->mv_dir= best_s.mv_dir;
2446                         s->mv_type = best_s.mv_type;
2447                         s->mb_intra= 0;
2448 /*                        s->mv[0][0][0] = best_s.mv[0][0][0];
2449                         s->mv[0][0][1] = best_s.mv[0][0][1];
2450                         s->mv[1][0][0] = best_s.mv[1][0][0];
2451                         s->mv[1][0][1] = best_s.mv[1][0][1];*/
2452                         backup_s.dquant= 0;
2453                         s->skipdct=1;
2454                         encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
2455                                         &dmin, &next_block, mx, my);
2456                         s->skipdct=0;
2457                     }
2458                 }
2459
2460                 s->current_picture.qscale_table[xy]= best_s.qscale;
2461
2462                 copy_context_after_encode(s, &best_s, -1);
2463
2464                 pb_bits_count= put_bits_count(&s->pb);
2465                 flush_put_bits(&s->pb);
2466                 ff_copy_bits(&backup_s.pb, bit_buf[next_block^1], pb_bits_count);
2467                 s->pb= backup_s.pb;
2468
2469                 if(s->data_partitioning){
2470                     pb2_bits_count= put_bits_count(&s->pb2);
2471                     flush_put_bits(&s->pb2);
2472                     ff_copy_bits(&backup_s.pb2, bit_buf2[next_block^1], pb2_bits_count);
2473                     s->pb2= backup_s.pb2;
2474
2475                     tex_pb_bits_count= put_bits_count(&s->tex_pb);
2476                     flush_put_bits(&s->tex_pb);
2477                     ff_copy_bits(&backup_s.tex_pb, bit_buf_tex[next_block^1], tex_pb_bits_count);
2478                     s->tex_pb= backup_s.tex_pb;
2479                 }
2480                 s->last_bits= put_bits_count(&s->pb);
2481
2482                 if (CONFIG_H263_ENCODER &&
2483                     s->out_format == FMT_H263 && s->pict_type!=AV_PICTURE_TYPE_B)
2484                     ff_h263_update_motion_val(s);
2485
2486                 if(next_block==0){ //FIXME 16 vs linesize16
2487                     s->dsp.put_pixels_tab[0][0](s->dest[0], s->rd_scratchpad                     , s->linesize  ,16);
2488                     s->dsp.put_pixels_tab[1][0](s->dest[1], s->rd_scratchpad + 16*s->linesize    , s->uvlinesize, 8);
2489                     s->dsp.put_pixels_tab[1][0](s->dest[2], s->rd_scratchpad + 16*s->linesize + 8, s->uvlinesize, 8);
2490                 }
2491
2492                 if(s->avctx->mb_decision == FF_MB_DECISION_BITS)
2493                     MPV_decode_mb(s, s->block);
2494             } else {
2495                 int motion_x = 0, motion_y = 0;
2496                 s->mv_type=MV_TYPE_16X16;
2497                 // only one MB-Type possible
2498
2499                 switch(mb_type){
2500                 case CANDIDATE_MB_TYPE_INTRA:
2501                     s->mv_dir = 0;
2502                     s->mb_intra= 1;
2503                     motion_x= s->mv[0][0][0] = 0;
2504                     motion_y= s->mv[0][0][1] = 0;
2505                     break;
2506                 case CANDIDATE_MB_TYPE_INTER:
2507                     s->mv_dir = MV_DIR_FORWARD;
2508                     s->mb_intra= 0;
2509                     motion_x= s->mv[0][0][0] = s->p_mv_table[xy][0];
2510                     motion_y= s->mv[0][0][1] = s->p_mv_table[xy][1];
2511                     break;
2512                 case CANDIDATE_MB_TYPE_INTER_I:
2513                     s->mv_dir = MV_DIR_FORWARD;
2514                     s->mv_type = MV_TYPE_FIELD;
2515                     s->mb_intra= 0;
2516                     for(i=0; i<2; i++){
2517                         j= s->field_select[0][i] = s->p_field_select_table[i][xy];
2518                         s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
2519                         s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
2520                     }
2521                     break;
2522                 case CANDIDATE_MB_TYPE_INTER4V:
2523                     s->mv_dir = MV_DIR_FORWARD;
2524                     s->mv_type = MV_TYPE_8X8;
2525                     s->mb_intra= 0;
2526                     for(i=0; i<4; i++){
2527                         s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
2528                         s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
2529                     }
2530                     break;
2531                 case CANDIDATE_MB_TYPE_DIRECT:
2532                     if (CONFIG_MPEG4_ENCODER) {
2533                         s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
2534                         s->mb_intra= 0;
2535                         motion_x=s->b_direct_mv_table[xy][0];
2536                         motion_y=s->b_direct_mv_table[xy][1];
2537                         ff_mpeg4_set_direct_mv(s, motion_x, motion_y);
2538                     }
2539                     break;
2540                 case CANDIDATE_MB_TYPE_DIRECT0:
2541                     if (CONFIG_MPEG4_ENCODER) {
2542                         s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
2543                         s->mb_intra= 0;
2544                         ff_mpeg4_set_direct_mv(s, 0, 0);
2545                     }
2546                     break;
2547                 case CANDIDATE_MB_TYPE_BIDIR:
2548                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2549                     s->mb_intra= 0;
2550                     s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
2551                     s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
2552                     s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
2553                     s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2554                     break;
2555                 case CANDIDATE_MB_TYPE_BACKWARD:
2556                     s->mv_dir = MV_DIR_BACKWARD;
2557                     s->mb_intra= 0;
2558                     motion_x= s->mv[1][0][0] = s->b_back_mv_table[xy][0];
2559                     motion_y= s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2560                     break;
2561                 case CANDIDATE_MB_TYPE_FORWARD:
2562                     s->mv_dir = MV_DIR_FORWARD;
2563                     s->mb_intra= 0;
2564                     motion_x= s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
2565                     motion_y= s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2566 //                    printf(" %d %d ", motion_x, motion_y);
2567                     break;
2568                 case CANDIDATE_MB_TYPE_FORWARD_I:
2569                     s->mv_dir = MV_DIR_FORWARD;
2570                     s->mv_type = MV_TYPE_FIELD;
2571                     s->mb_intra= 0;
2572                     for(i=0; i<2; i++){
2573                         j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
2574                         s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
2575                         s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
2576                     }
2577                     break;
2578                 case CANDIDATE_MB_TYPE_BACKWARD_I:
2579                     s->mv_dir = MV_DIR_BACKWARD;
2580                     s->mv_type = MV_TYPE_FIELD;
2581                     s->mb_intra= 0;
2582                     for(i=0; i<2; i++){
2583                         j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
2584                         s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
2585                         s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
2586                     }
2587                     break;
2588                 case CANDIDATE_MB_TYPE_BIDIR_I:
2589                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2590                     s->mv_type = MV_TYPE_FIELD;
2591                     s->mb_intra= 0;
2592                     for(dir=0; dir<2; dir++){
2593                         for(i=0; i<2; i++){
2594                             j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
2595                             s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
2596                             s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
2597                         }
2598                     }
2599                     break;
2600                 default:
2601                     av_log(s->avctx, AV_LOG_ERROR, "illegal MB type\n");
2602                 }
2603
2604                 encode_mb(s, motion_x, motion_y);
2605
2606                 // RAL: Update last macroblock type
2607                 s->last_mv_dir = s->mv_dir;
2608
2609                 if (CONFIG_H263_ENCODER &&
2610                     s->out_format == FMT_H263 && s->pict_type!=AV_PICTURE_TYPE_B)
2611                     ff_h263_update_motion_val(s);
2612
2613                 MPV_decode_mb(s, s->block);
2614             }
2615
2616             /* clean the MV table in IPS frames for direct mode in B frames */
2617             if(s->mb_intra /* && I,P,S_TYPE */){
2618                 s->p_mv_table[xy][0]=0;
2619                 s->p_mv_table[xy][1]=0;
2620             }
2621
2622             if(s->flags&CODEC_FLAG_PSNR){
2623                 int w= 16;
2624                 int h= 16;
2625
2626                 if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
2627                 if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
2628
2629                 s->current_picture.error[0] += sse(
2630                     s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16,
2631                     s->dest[0], w, h, s->linesize);
2632                 s->current_picture.error[1] += sse(
2633                     s, s->new_picture.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*chr_h,
2634                     s->dest[1], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
2635                 s->current_picture.error[2] += sse(
2636                     s, s->new_picture.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*chr_h,
2637                     s->dest[2], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
2638             }
2639             if(s->loop_filter){
2640                 if(CONFIG_H263_ENCODER && s->out_format == FMT_H263)
2641                     ff_h263_loop_filter(s);
2642             }
2643 //printf("MB %d %d bits\n", s->mb_x+s->mb_y*s->mb_stride, put_bits_count(&s->pb));
2644         }
2645     }
2646
2647     //not beautiful here but we must write it before flushing so it has to be here
2648     if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version && s->msmpeg4_version<4 && s->pict_type == AV_PICTURE_TYPE_I)
2649         msmpeg4_encode_ext_header(s);
2650
2651     write_slice_end(s);
2652
2653     /* Send the last GOB if RTP */
2654     if (s->avctx->rtp_callback) {
2655         int number_mb = (mb_y - s->resync_mb_y)*s->mb_width - s->resync_mb_x;
2656         pdif = put_bits_ptr(&s->pb) - s->ptr_lastgob;
2657         /* Call the RTP callback to send the last GOB */
2658         emms_c();
2659         s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, pdif, number_mb);
2660     }
2661
2662     return 0;
2663 }
2664
2665 #define MERGE(field) dst->field += src->field; src->field=0
2666 static void merge_context_after_me(MpegEncContext *dst, MpegEncContext *src){
2667     MERGE(me.scene_change_score);
2668     MERGE(me.mc_mb_var_sum_temp);
2669     MERGE(me.mb_var_sum_temp);
2670 }
2671
2672 static void merge_context_after_encode(MpegEncContext *dst, MpegEncContext *src){
2673     int i;
2674
2675     MERGE(dct_count[0]); //note, the other dct vars are not part of the context
2676     MERGE(dct_count[1]);
2677     MERGE(mv_bits);
2678     MERGE(i_tex_bits);
2679     MERGE(p_tex_bits);
2680     MERGE(i_count);
2681     MERGE(f_count);
2682     MERGE(b_count);
2683     MERGE(skip_count);
2684     MERGE(misc_bits);
2685     MERGE(error_count);
2686     MERGE(padding_bug_score);
2687     MERGE(current_picture.error[0]);
2688     MERGE(current_picture.error[1]);
2689     MERGE(current_picture.error[2]);
2690
2691     if(dst->avctx->noise_reduction){
2692         for(i=0; i<64; i++){
2693             MERGE(dct_error_sum[0][i]);
2694             MERGE(dct_error_sum[1][i]);
2695         }
2696     }
2697
2698     assert(put_bits_count(&src->pb) % 8 ==0);
2699     assert(put_bits_count(&dst->pb) % 8 ==0);
2700     ff_copy_bits(&dst->pb, src->pb.buf, put_bits_count(&src->pb));
2701     flush_put_bits(&dst->pb);
2702 }
2703
2704 static int estimate_qp(MpegEncContext *s, int dry_run){
2705     if (s->next_lambda){
2706         s->current_picture_ptr->quality=
2707         s->current_picture.quality = s->next_lambda;
2708         if(!dry_run) s->next_lambda= 0;
2709     } else if (!s->fixed_qscale) {
2710         s->current_picture_ptr->quality=
2711         s->current_picture.quality = ff_rate_estimate_qscale(s, dry_run);
2712         if (s->current_picture.quality < 0)
2713             return -1;
2714     }
2715
2716     if(s->adaptive_quant){
2717         switch(s->codec_id){
2718         case CODEC_ID_MPEG4:
2719             if (CONFIG_MPEG4_ENCODER)
2720                 ff_clean_mpeg4_qscales(s);
2721             break;
2722         case CODEC_ID_H263:
2723         case CODEC_ID_H263P:
2724         case CODEC_ID_FLV1:
2725             if (CONFIG_H263_ENCODER)
2726                 ff_clean_h263_qscales(s);
2727             break;
2728         default:
2729             ff_init_qscale_tab(s);
2730         }
2731
2732         s->lambda= s->lambda_table[0];
2733         //FIXME broken
2734     }else
2735         s->lambda= s->current_picture.quality;
2736 //printf("%d %d\n", s->avctx->global_quality, s->current_picture.quality);
2737     update_qscale(s);
2738     return 0;
2739 }
2740
2741 /* must be called before writing the header */
2742 static void set_frame_distances(MpegEncContext * s){
2743     assert(s->current_picture_ptr->pts != AV_NOPTS_VALUE);
2744     s->time= s->current_picture_ptr->pts*s->avctx->time_base.num;
2745
2746     if(s->pict_type==AV_PICTURE_TYPE_B){
2747         s->pb_time= s->pp_time - (s->last_non_b_time - s->time);
2748         assert(s->pb_time > 0 && s->pb_time < s->pp_time);
2749     }else{
2750         s->pp_time= s->time - s->last_non_b_time;
2751         s->last_non_b_time= s->time;
2752         assert(s->picture_number==0 || s->pp_time > 0);
2753     }
2754 }
2755
2756 static int encode_picture(MpegEncContext *s, int picture_number)
2757 {
2758     int i;
2759     int bits;
2760     int context_count = s->avctx->thread_count;
2761
2762     s->picture_number = picture_number;
2763
2764     /* Reset the average MB variance */
2765     s->me.mb_var_sum_temp    =
2766     s->me.mc_mb_var_sum_temp = 0;
2767
2768     /* we need to initialize some time vars before we can encode b-frames */
2769     // RAL: Condition added for MPEG1VIDEO
2770     if (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO || (s->h263_pred && !s->msmpeg4_version))
2771         set_frame_distances(s);
2772     if(CONFIG_MPEG4_ENCODER && s->codec_id == CODEC_ID_MPEG4)
2773         ff_set_mpeg4_time(s);
2774
2775     s->me.scene_change_score=0;
2776
2777 //    s->lambda= s->current_picture_ptr->quality; //FIXME qscale / ... stuff for ME rate distortion
2778
2779     if(s->pict_type==AV_PICTURE_TYPE_I){
2780         if(s->msmpeg4_version >= 3) s->no_rounding=1;
2781         else                        s->no_rounding=0;
2782     }else if(s->pict_type!=AV_PICTURE_TYPE_B){
2783         if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
2784             s->no_rounding ^= 1;
2785     }
2786
2787     if(s->flags & CODEC_FLAG_PASS2){
2788         if (estimate_qp(s,1) < 0)
2789             return -1;
2790         ff_get_2pass_fcode(s);
2791     }else if(!(s->flags & CODEC_FLAG_QSCALE)){
2792         if(s->pict_type==AV_PICTURE_TYPE_B)
2793             s->lambda= s->last_lambda_for[s->pict_type];
2794         else
2795             s->lambda= s->last_lambda_for[s->last_non_b_pict_type];
2796         update_qscale(s);
2797     }
2798
2799     s->mb_intra=0; //for the rate distortion & bit compare functions
2800     for(i=1; i<context_count; i++){
2801         ff_update_duplicate_context(s->thread_context[i], s);
2802     }
2803
2804     if(ff_init_me(s)<0)
2805         return -1;
2806
2807     /* Estimate motion for every MB */
2808     if(s->pict_type != AV_PICTURE_TYPE_I){
2809         s->lambda = (s->lambda * s->avctx->me_penalty_compensation + 128)>>8;
2810         s->lambda2= (s->lambda2* (int64_t)s->avctx->me_penalty_compensation + 128)>>8;
2811         if(s->pict_type != AV_PICTURE_TYPE_B && s->avctx->me_threshold==0){
2812             if((s->avctx->pre_me && s->last_non_b_pict_type==AV_PICTURE_TYPE_I) || s->avctx->pre_me==2){
2813                 s->avctx->execute(s->avctx, pre_estimate_motion_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
2814             }
2815         }
2816
2817         s->avctx->execute(s->avctx, estimate_motion_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
2818     }else /* if(s->pict_type == AV_PICTURE_TYPE_I) */{
2819         /* I-Frame */
2820         for(i=0; i<s->mb_stride*s->mb_height; i++)
2821             s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
2822
2823         if(!s->fixed_qscale){
2824             /* finding spatial complexity for I-frame rate control */
2825             s->avctx->execute(s->avctx, mb_var_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
2826         }
2827     }
2828     for(i=1; i<context_count; i++){
2829         merge_context_after_me(s, s->thread_context[i]);
2830     }
2831     s->current_picture.mc_mb_var_sum= s->current_picture_ptr->mc_mb_var_sum= s->me.mc_mb_var_sum_temp;
2832     s->current_picture.   mb_var_sum= s->current_picture_ptr->   mb_var_sum= s->me.   mb_var_sum_temp;
2833     emms_c();
2834
2835     if(s->me.scene_change_score > s->avctx->scenechange_threshold && s->pict_type == AV_PICTURE_TYPE_P){
2836         s->pict_type= AV_PICTURE_TYPE_I;
2837         for(i=0; i<s->mb_stride*s->mb_height; i++)
2838             s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
2839 //printf("Scene change detected, encoding as I Frame %d %d\n", s->current_picture.mb_var_sum, s->current_picture.mc_mb_var_sum);
2840     }
2841
2842     if(!s->umvplus){
2843         if(s->pict_type==AV_PICTURE_TYPE_P || s->pict_type==AV_PICTURE_TYPE_S) {
2844             s->f_code= ff_get_best_fcode(s, s->p_mv_table, CANDIDATE_MB_TYPE_INTER);
2845
2846             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2847                 int a,b;
2848                 a= ff_get_best_fcode(s, s->p_field_mv_table[0][0], CANDIDATE_MB_TYPE_INTER_I); //FIXME field_select
2849                 b= ff_get_best_fcode(s, s->p_field_mv_table[1][1], CANDIDATE_MB_TYPE_INTER_I);
2850                 s->f_code= FFMAX3(s->f_code, a, b);
2851             }
2852
2853             ff_fix_long_p_mvs(s);
2854             ff_fix_long_mvs(s, NULL, 0, s->p_mv_table, s->f_code, CANDIDATE_MB_TYPE_INTER, 0);
2855             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2856                 int j;
2857                 for(i=0; i<2; i++){
2858                     for(j=0; j<2; j++)
2859                         ff_fix_long_mvs(s, s->p_field_select_table[i], j,
2860                                         s->p_field_mv_table[i][j], s->f_code, CANDIDATE_MB_TYPE_INTER_I, 0);
2861                 }
2862             }
2863         }
2864
2865         if(s->pict_type==AV_PICTURE_TYPE_B){
2866             int a, b;
2867
2868             a = ff_get_best_fcode(s, s->b_forw_mv_table, CANDIDATE_MB_TYPE_FORWARD);
2869             b = ff_get_best_fcode(s, s->b_bidir_forw_mv_table, CANDIDATE_MB_TYPE_BIDIR);
2870             s->f_code = FFMAX(a, b);
2871
2872             a = ff_get_best_fcode(s, s->b_back_mv_table, CANDIDATE_MB_TYPE_BACKWARD);
2873             b = ff_get_best_fcode(s, s->b_bidir_back_mv_table, CANDIDATE_MB_TYPE_BIDIR);
2874             s->b_code = FFMAX(a, b);
2875
2876             ff_fix_long_mvs(s, NULL, 0, s->b_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_FORWARD, 1);
2877             ff_fix_long_mvs(s, NULL, 0, s->b_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BACKWARD, 1);
2878             ff_fix_long_mvs(s, NULL, 0, s->b_bidir_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_BIDIR, 1);
2879             ff_fix_long_mvs(s, NULL, 0, s->b_bidir_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BIDIR, 1);
2880             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2881                 int dir, j;
2882                 for(dir=0; dir<2; dir++){
2883                     for(i=0; i<2; i++){
2884                         for(j=0; j<2; j++){
2885                             int type= dir ? (CANDIDATE_MB_TYPE_BACKWARD_I|CANDIDATE_MB_TYPE_BIDIR_I)
2886                                           : (CANDIDATE_MB_TYPE_FORWARD_I |CANDIDATE_MB_TYPE_BIDIR_I);
2887                             ff_fix_long_mvs(s, s->b_field_select_table[dir][i], j,
2888                                             s->b_field_mv_table[dir][i][j], dir ? s->b_code : s->f_code, type, 1);
2889                         }
2890                     }
2891                 }
2892             }
2893         }
2894     }
2895
2896     if (estimate_qp(s, 0) < 0)
2897         return -1;
2898
2899     if(s->qscale < 3 && s->max_qcoeff<=128 && s->pict_type==AV_PICTURE_TYPE_I && !(s->flags & CODEC_FLAG_QSCALE))
2900         s->qscale= 3; //reduce clipping problems
2901
2902     if (s->out_format == FMT_MJPEG) {
2903         /* for mjpeg, we do include qscale in the matrix */
2904         for(i=1;i<64;i++){
2905             int j= s->dsp.idct_permutation[i];
2906
2907             s->intra_matrix[j] = av_clip_uint8((ff_mpeg1_default_intra_matrix[i] * s->qscale) >> 3);
2908         }
2909         s->y_dc_scale_table=
2910         s->c_dc_scale_table= ff_mpeg2_dc_scale_table[s->intra_dc_precision];
2911         s->intra_matrix[0] = ff_mpeg2_dc_scale_table[s->intra_dc_precision][8];
2912         ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
2913                        s->intra_matrix, s->intra_quant_bias, 8, 8, 1);
2914         s->qscale= 8;
2915     }
2916
2917     //FIXME var duplication
2918     s->current_picture_ptr->key_frame=
2919     s->current_picture.key_frame= s->pict_type == AV_PICTURE_TYPE_I; //FIXME pic_ptr
2920     s->current_picture_ptr->pict_type=
2921     s->current_picture.pict_type= s->pict_type;
2922
2923     if(s->current_picture.key_frame)
2924         s->picture_in_gop_number=0;
2925
2926     s->last_bits= put_bits_count(&s->pb);
2927     switch(s->out_format) {
2928     case FMT_MJPEG:
2929         if (CONFIG_MJPEG_ENCODER)
2930             ff_mjpeg_encode_picture_header(s);
2931         break;
2932     case FMT_H261:
2933         if (CONFIG_H261_ENCODER)
2934             ff_h261_encode_picture_header(s, picture_number);
2935         break;
2936     case FMT_H263:
2937         if (CONFIG_WMV2_ENCODER && s->codec_id == CODEC_ID_WMV2)
2938             ff_wmv2_encode_picture_header(s, picture_number);
2939         else if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version)
2940             msmpeg4_encode_picture_header(s, picture_number);
2941         else if (CONFIG_MPEG4_ENCODER && s->h263_pred)
2942             mpeg4_encode_picture_header(s, picture_number);
2943         else if (CONFIG_RV10_ENCODER && s->codec_id == CODEC_ID_RV10)
2944             rv10_encode_picture_header(s, picture_number);
2945         else if (CONFIG_RV20_ENCODER && s->codec_id == CODEC_ID_RV20)
2946             rv20_encode_picture_header(s, picture_number);
2947         else if (CONFIG_FLV_ENCODER && s->codec_id == CODEC_ID_FLV1)
2948             ff_flv_encode_picture_header(s, picture_number);
2949         else if (CONFIG_H263_ENCODER)
2950             h263_encode_picture_header(s, picture_number);
2951         break;
2952     case FMT_MPEG1:
2953         if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
2954             mpeg1_encode_picture_header(s, picture_number);
2955         break;
2956     case FMT_H264:
2957         break;
2958     default:
2959         assert(0);
2960     }
2961     bits= put_bits_count(&s->pb);
2962     s->header_bits= bits - s->last_bits;
2963
2964     for(i=1; i<context_count; i++){
2965         update_duplicate_context_after_me(s->thread_context[i], s);
2966     }
2967     s->avctx->execute(s->avctx, encode_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
2968     for(i=1; i<context_count; i++){
2969         merge_context_after_encode(s, s->thread_context[i]);
2970     }
2971     emms_c();
2972     return 0;
2973 }
2974
2975 static void denoise_dct_c(MpegEncContext *s, DCTELEM *block){
2976     const int intra= s->mb_intra;
2977     int i;
2978
2979     s->dct_count[intra]++;
2980
2981     for(i=0; i<64; i++){
2982         int level= block[i];
2983
2984         if(level){
2985             if(level>0){
2986                 s->dct_error_sum[intra][i] += level;
2987                 level -= s->dct_offset[intra][i];
2988                 if(level<0) level=0;
2989             }else{
2990                 s->dct_error_sum[intra][i] -= level;
2991                 level += s->dct_offset[intra][i];
2992                 if(level>0) level=0;
2993             }
2994             block[i]= level;
2995         }
2996     }
2997 }
2998
2999 static int dct_quantize_trellis_c(MpegEncContext *s,
3000                                   DCTELEM *block, int n,
3001                                   int qscale, int *overflow){
3002     const int *qmat;
3003     const uint8_t *scantable= s->intra_scantable.scantable;
3004     const uint8_t *perm_scantable= s->intra_scantable.permutated;
3005     int max=0;
3006     unsigned int threshold1, threshold2;
3007     int bias=0;
3008     int run_tab[65];
3009     int level_tab[65];
3010     int score_tab[65];
3011     int survivor[65];
3012     int survivor_count;
3013     int last_run=0;
3014     int last_level=0;
3015     int last_score= 0;
3016     int last_i;
3017     int coeff[2][64];
3018     int coeff_count[64];
3019     int qmul, qadd, start_i, last_non_zero, i, dc;
3020     const int esc_length= s->ac_esc_length;
3021     uint8_t * length;
3022     uint8_t * last_length;
3023     const int lambda= s->lambda2 >> (FF_LAMBDA_SHIFT - 6);
3024
3025     s->dsp.fdct (block);
3026
3027     if(s->dct_error_sum)
3028         s->denoise_dct(s, block);
3029     qmul= qscale*16;
3030     qadd= ((qscale-1)|1)*8;
3031
3032     if (s->mb_intra) {
3033         int q;
3034         if (!s->h263_aic) {
3035             if (n < 4)
3036                 q = s->y_dc_scale;
3037             else
3038                 q = s->c_dc_scale;
3039             q = q << 3;
3040         } else{
3041             /* For AIC we skip quant/dequant of INTRADC */
3042             q = 1 << 3;
3043             qadd=0;
3044         }
3045
3046         /* note: block[0] is assumed to be positive */
3047         block[0] = (block[0] + (q >> 1)) / q;
3048         start_i = 1;
3049         last_non_zero = 0;
3050         qmat = s->q_intra_matrix[qscale];
3051         if(s->mpeg_quant || s->out_format == FMT_MPEG1)
3052             bias= 1<<(QMAT_SHIFT-1);
3053         length     = s->intra_ac_vlc_length;
3054         last_length= s->intra_ac_vlc_last_length;
3055     } else {
3056         start_i = 0;
3057         last_non_zero = -1;
3058         qmat = s->q_inter_matrix[qscale];
3059         length     = s->inter_ac_vlc_length;
3060         last_length= s->inter_ac_vlc_last_length;
3061     }
3062     last_i= start_i;
3063
3064     threshold1= (1<<QMAT_SHIFT) - bias - 1;
3065     threshold2= (threshold1<<1);
3066
3067     for(i=63; i>=start_i; i--) {
3068         const int j = scantable[i];
3069         int level = block[j] * qmat[j];
3070
3071         if(((unsigned)(level+threshold1))>threshold2){
3072             last_non_zero = i;
3073             break;
3074         }
3075     }
3076
3077     for(i=start_i; i<=last_non_zero; i++) {
3078         const int j = scantable[i];
3079         int level = block[j] * qmat[j];
3080
3081 //        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
3082 //           || bias-level >= (1<<(QMAT_SHIFT - 3))){
3083         if(((unsigned)(level+threshold1))>threshold2){
3084             if(level>0){
3085                 level= (bias + level)>>QMAT_SHIFT;
3086                 coeff[0][i]= level;
3087                 coeff[1][i]= level-1;
3088 //                coeff[2][k]= level-2;
3089             }else{
3090                 level= (bias - level)>>QMAT_SHIFT;
3091                 coeff[0][i]= -level;
3092                 coeff[1][i]= -level+1;
3093 //                coeff[2][k]= -level+2;
3094             }
3095             coeff_count[i]= FFMIN(level, 2);
3096             assert(coeff_count[i]);
3097             max |=level;
3098         }else{
3099             coeff[0][i]= (level>>31)|1;
3100             coeff_count[i]= 1;
3101         }
3102     }
3103
3104     *overflow= s->max_qcoeff < max; //overflow might have happened
3105
3106     if(last_non_zero < start_i){
3107         memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
3108         return last_non_zero;
3109     }
3110
3111     score_tab[start_i]= 0;
3112     survivor[0]= start_i;
3113     survivor_count= 1;
3114
3115     for(i=start_i; i<=last_non_zero; i++){
3116         int level_index, j, zero_distortion;
3117         int dct_coeff= FFABS(block[ scantable[i] ]);
3118         int best_score=256*256*256*120;
3119
3120         if (   s->dsp.fdct == fdct_ifast
3121 #ifndef FAAN_POSTSCALE
3122             || s->dsp.fdct == ff_faandct
3123 #endif
3124            )
3125             dct_coeff= (dct_coeff*ff_inv_aanscales[ scantable[i] ]) >> 12;
3126         zero_distortion= dct_coeff*dct_coeff;
3127
3128         for(level_index=0; level_index < coeff_count[i]; level_index++){
3129             int distortion;
3130             int level= coeff[level_index][i];
3131             const int alevel= FFABS(level);
3132             int unquant_coeff;
3133
3134             assert(level);
3135
3136             if(s->out_format == FMT_H263){
3137                 unquant_coeff= alevel*qmul + qadd;
3138             }else{ //MPEG1
3139                 j= s->dsp.idct_permutation[ scantable[i] ]; //FIXME optimize
3140                 if(s->mb_intra){
3141                         unquant_coeff = (int)(  alevel  * qscale * s->intra_matrix[j]) >> 3;
3142                         unquant_coeff =   (unquant_coeff - 1) | 1;
3143                 }else{
3144                         unquant_coeff = (((  alevel  << 1) + 1) * qscale * ((int) s->inter_matrix[j])) >> 4;
3145                         unquant_coeff =   (unquant_coeff - 1) | 1;
3146                 }
3147                 unquant_coeff<<= 3;
3148             }
3149
3150             distortion= (unquant_coeff - dct_coeff) * (unquant_coeff - dct_coeff) - zero_distortion;
3151             level+=64;
3152             if((level&(~127)) == 0){
3153                 for(j=survivor_count-1; j>=0; j--){
3154                     int run= i - survivor[j];
3155                     int score= distortion + length[UNI_AC_ENC_INDEX(run, level)]*lambda;
3156                     score += score_tab[i-run];
3157
3158                     if(score < best_score){
3159                         best_score= score;
3160                         run_tab[i+1]= run;
3161                         level_tab[i+1]= level-64;
3162                     }
3163                 }
3164
3165                 if(s->out_format == FMT_H263){
3166                     for(j=survivor_count-1; j>=0; j--){
3167                         int run= i - survivor[j];
3168                         int score= distortion + last_length[UNI_AC_ENC_INDEX(run, level)]*lambda;
3169                         score += score_tab[i-run];
3170                         if(score < last_score){
3171                             last_score= score;
3172                             last_run= run;
3173                             last_level= level-64;
3174                             last_i= i+1;
3175                         }
3176                     }
3177                 }
3178             }else{
3179                 distortion += esc_length*lambda;
3180                 for(j=survivor_count-1; j>=0; j--){
3181                     int run= i - survivor[j];
3182                     int score= distortion + score_tab[i-run];
3183
3184                     if(score < best_score){
3185                         best_score= score;
3186                         run_tab[i+1]= run;
3187                         level_tab[i+1]= level-64;
3188                     }
3189                 }
3190
3191                 if(s->out_format == FMT_H263){
3192                   for(j=survivor_count-1; j>=0; j--){
3193                         int run= i - survivor[j];
3194                         int score= distortion + score_tab[i-run];
3195                         if(score < last_score){
3196                             last_score= score;
3197                             last_run= run;
3198                             last_level= level-64;
3199                             last_i= i+1;
3200                         }
3201                     }
3202                 }
3203             }
3204         }
3205
3206         score_tab[i+1]= best_score;
3207
3208         //Note: there is a vlc code in mpeg4 which is 1 bit shorter then another one with a shorter run and the same level
3209         if(last_non_zero <= 27){
3210             for(; survivor_count; survivor_count--){
3211                 if(score_tab[ survivor[survivor_count-1] ] <= best_score)
3212                     break;
3213             }
3214         }else{
3215             for(; survivor_count; survivor_count--){
3216                 if(score_tab[ survivor[survivor_count-1] ] <= best_score + lambda)
3217                     break;
3218             }
3219         }
3220
3221         survivor[ survivor_count++ ]= i+1;
3222     }
3223
3224     if(s->out_format != FMT_H263){
3225         last_score= 256*256*256*120;
3226         for(i= survivor[0]; i<=last_non_zero + 1; i++){
3227             int score= score_tab[i];
3228             if(i) score += lambda*2; //FIXME exacter?
3229
3230             if(score < last_score){
3231                 last_score= score;
3232                 last_i= i;
3233                 last_level= level_tab[i];
3234                 last_run= run_tab[i];
3235             }
3236         }
3237     }
3238
3239     s->coded_score[n] = last_score;
3240
3241     dc= FFABS(block[0]);
3242     last_non_zero= last_i - 1;
3243     memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
3244
3245     if(last_non_zero < start_i)
3246         return last_non_zero;
3247
3248     if(last_non_zero == 0 && start_i == 0){
3249         int best_level= 0;
3250         int best_score= dc * dc;
3251
3252         for(i=0; i<coeff_count[0]; i++){
3253             int level= coeff[i][0];
3254             int alevel= FFABS(level);
3255             int unquant_coeff, score, distortion;
3256
3257             if(s->out_format == FMT_H263){
3258                     unquant_coeff= (alevel*qmul + qadd)>>3;
3259             }else{ //MPEG1
3260                     unquant_coeff = (((  alevel  << 1) + 1) * qscale * ((int) s->inter_matrix[0])) >> 4;
3261                     unquant_coeff =   (unquant_coeff - 1) | 1;
3262             }
3263             unquant_coeff = (unquant_coeff + 4) >> 3;
3264             unquant_coeff<<= 3 + 3;
3265
3266             distortion= (unquant_coeff - dc) * (unquant_coeff - dc);
3267             level+=64;
3268             if((level&(~127)) == 0) score= distortion + last_length[UNI_AC_ENC_INDEX(0, level)]*lambda;
3269             else                    score= distortion + esc_length*lambda;
3270
3271             if(score < best_score){
3272                 best_score= score;
3273                 best_level= level - 64;
3274             }
3275         }
3276         block[0]= best_level;
3277         s->coded_score[n] = best_score - dc*dc;
3278         if(best_level == 0) return -1;
3279         else                return last_non_zero;
3280     }
3281
3282     i= last_i;
3283     assert(last_level);
3284
3285     block[ perm_scantable[last_non_zero] ]= last_level;
3286     i -= last_run + 1;
3287
3288     for(; i>start_i; i -= run_tab[i] + 1){
3289         block[ perm_scantable[i-1] ]= level_tab[i];
3290     }
3291
3292     return last_non_zero;
3293 }
3294
3295 //#define REFINE_STATS 1
3296 static int16_t basis[64][64];
3297
3298 static void build_basis(uint8_t *perm){
3299     int i, j, x, y;
3300     emms_c();
3301     for(i=0; i<8; i++){
3302         for(j=0; j<8; j++){
3303             for(y=0; y<8; y++){
3304                 for(x=0; x<8; x++){
3305                     double s= 0.25*(1<<BASIS_SHIFT);
3306                     int index= 8*i + j;
3307                     int perm_index= perm[index];
3308                     if(i==0) s*= sqrt(0.5);
3309                     if(j==0) s*= sqrt(0.5);
3310                     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)));
3311                 }
3312             }
3313         }
3314     }
3315 }
3316
3317 static int dct_quantize_refine(MpegEncContext *s, //FIXME breaks denoise?
3318                         DCTELEM *block, int16_t *weight, DCTELEM *orig,
3319                         int n, int qscale){
3320     int16_t rem[64];
3321     LOCAL_ALIGNED_16(DCTELEM, d1, [64]);
3322     const uint8_t *scantable= s->intra_scantable.scantable;
3323     const uint8_t *perm_scantable= s->intra_scantable.permutated;
3324 //    unsigned int threshold1, threshold2;
3325 //    int bias=0;
3326     int run_tab[65];
3327     int prev_run=0;
3328     int prev_level=0;
3329     int qmul, qadd, start_i, last_non_zero, i, dc;
3330     uint8_t * length;
3331     uint8_t * last_length;
3332     int lambda;
3333     int rle_index, run, q = 1, sum; //q is only used when s->mb_intra is true
3334 #ifdef REFINE_STATS
3335 static int count=0;
3336 static int after_last=0;
3337 static int to_zero=0;
3338 static int from_zero=0;
3339 static int raise=0;
3340 static int lower=0;
3341 static int messed_sign=0;
3342 #endif
3343
3344     if(basis[0][0] == 0)
3345         build_basis(s->dsp.idct_permutation);
3346
3347     qmul= qscale*2;
3348     qadd= (qscale-1)|1;
3349     if (s->mb_intra) {
3350         if (!s->h263_aic) {
3351             if (n < 4)
3352                 q = s->y_dc_scale;
3353             else
3354                 q = s->c_dc_scale;
3355         } else{
3356             /* For AIC we skip quant/dequant of INTRADC */
3357             q = 1;
3358             qadd=0;
3359         }
3360         q <<= RECON_SHIFT-3;
3361         /* note: block[0] is assumed to be positive */
3362         dc= block[0]*q;
3363 //        block[0] = (block[0] + (q >> 1)) / q;
3364         start_i = 1;
3365 //        if(s->mpeg_quant || s->out_format == FMT_MPEG1)
3366 //            bias= 1<<(QMAT_SHIFT-1);
3367         length     = s->intra_ac_vlc_length;
3368         last_length= s->intra_ac_vlc_last_length;
3369     } else {