git.sesse.net Git - ffmpeg/blob - libavcodec/mpegvideo_enc.c

   1 /*
   2  * The simplest mpeg encoder (well, it was the simplest!)
   3  * Copyright (c) 2000,2001 Fabrice Bellard
   4  * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
   5  *
   6  * 4MV & hq & B-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
   7  *
   8  * This file is part of FFmpeg.
   9  *
  10  * FFmpeg is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU Lesser General Public
  12  * License as published by the Free Software Foundation; either
  13  * version 2.1 of the License, or (at your option) any later version.
  14  *
  15  * FFmpeg is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  * Lesser General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU Lesser General Public
  21  * License along with FFmpeg; if not, write to the Free Software
  22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  23  */
  24
  25 /**
  26  * @file libavcodec/mpegvideo_enc.c
  27  * The simplest mpeg encoder (well, it was the simplest!).
  28  */
  29
  30 #include "avcodec.h"
  31 #include "dsputil.h"
  32 #include "mpegvideo.h"
  33 #include "mpegvideo_common.h"
  34 #include "h263.h"
  35 #include "mjpegenc.h"
  36 #include "msmpeg4.h"
  37 #include "faandct.h"
  38 #include "aandcttab.h"
  39 #include "flv.h"
  40 #include "mpeg4video.h"
  41 #include "internal.h"
  42 #include <limits.h>
  43
  44 //#undef NDEBUG
  45 //#include <assert.h>
  46
  47 static int encode_picture(MpegEncContext *s, int picture_number);
  48 static int dct_quantize_refine(MpegEncContext *s, DCTELEM *block, int16_t *weight, DCTELEM *orig, int n, int qscale);
  49 static int sse_mb(MpegEncContext *s);
  50
  51 /* enable all paranoid tests for rounding, overflows, etc... */
  52 //#define PARANOID
  53
  54 //#define DEBUG
  55
  56 static uint8_t default_mv_penalty[MAX_FCODE+1][MAX_MV*2+1];
  57 static uint8_t default_fcode_tab[MAX_MV*2+1];
  58
  59 void ff_convert_matrix(DSPContext *dsp, int (*qmat)[64], uint16_t (*qmat16)[2][64],
  60                            const uint16_t *quant_matrix, int bias, int qmin, int qmax, int intra)
  61 {
  62     int qscale;
  63     int shift=0;
  64
  65     for(qscale=qmin; qscale<=qmax; qscale++){
  66         int i;
  67         if (dsp->fdct == ff_jpeg_fdct_islow
  68 #ifdef FAAN_POSTSCALE
  69             || dsp->fdct == ff_faandct
  70 #endif
  71             ) {
  72             for(i=0;i<64;i++) {
  73                 const int j= dsp->idct_permutation[i];
  74                 /* 16 <= qscale * quant_matrix[i] <= 7905 */
  75                 /* 19952             <= ff_aanscales[i] * qscale * quant_matrix[i]               <= 249205026 */
  76                 /* (1 << 36) / 19952 >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= (1 << 36) / 249205026 */
  77                 /* 3444240           >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= 275 */
  78
  79                 qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) /
  80                                 (qscale * quant_matrix[j]));
  81             }
  82         } else if (dsp->fdct == fdct_ifast
  83 #ifndef FAAN_POSTSCALE
  84                    || dsp->fdct == ff_faandct
  85 #endif
  86                    ) {
  87             for(i=0;i<64;i++) {
  88                 const int j= dsp->idct_permutation[i];
  89                 /* 16 <= qscale * quant_matrix[i] <= 7905 */
  90                 /* 19952             <= ff_aanscales[i] * qscale * quant_matrix[i]               <= 249205026 */
  91                 /* (1 << 36) / 19952 >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */
  92                 /* 3444240           >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= 275 */
  93
  94                 qmat[qscale][i] = (int)((UINT64_C(1) << (QMAT_SHIFT + 14)) /
  95                                 (ff_aanscales[i] * qscale * quant_matrix[j]));
  96             }
  97         } else {
  98             for(i=0;i<64;i++) {
  99                 const int j= dsp->idct_permutation[i];
 100                 /* We can safely suppose that 16 <= quant_matrix[i] <= 255
 101                    So 16           <= qscale * quant_matrix[i]             <= 7905
 102                    so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905
 103                    so 32768        >= (1<<19) / (qscale * quant_matrix[i]) >= 67
 104                 */
 105                 qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) / (qscale * quant_matrix[j]));
 106 //                qmat  [qscale][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]);
 107                 qmat16[qscale][0][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[j]);
 108
 109                 if(qmat16[qscale][0][i]==0 || qmat16[qscale][0][i]==128*256) qmat16[qscale][0][i]=128*256-1;
 110                 qmat16[qscale][1][i]= ROUNDED_DIV(bias<<(16-QUANT_BIAS_SHIFT), qmat16[qscale][0][i]);
 111             }
 112         }
 113
 114         for(i=intra; i<64; i++){
 115             int64_t max= 8191;
 116             if (dsp->fdct == fdct_ifast
 117 #ifndef FAAN_POSTSCALE
 118                    || dsp->fdct == ff_faandct
 119 #endif
 120                    ) {
 121                 max = (8191LL*ff_aanscales[i]) >> 14;
 122             }
 123             while(((max * qmat[qscale][i]) >> shift) > INT_MAX){
 124                 shift++;
 125             }
 126         }
 127     }
 128     if(shift){
 129         av_log(NULL, AV_LOG_INFO, "Warning, QMAT_SHIFT is larger than %d, overflows possible\n", QMAT_SHIFT - shift);
 130     }
 131 }
 132
 133 static inline void update_qscale(MpegEncContext *s){
 134     s->qscale= (s->lambda*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7);
 135     s->qscale= av_clip(s->qscale, s->avctx->qmin, s->avctx->qmax);
 136
 137     s->lambda2= (s->lambda*s->lambda + FF_LAMBDA_SCALE/2) >> FF_LAMBDA_SHIFT;
 138 }
 139
 140 void ff_write_quant_matrix(PutBitContext *pb, uint16_t *matrix){
 141     int i;
 142
 143     if(matrix){
 144         put_bits(pb, 1, 1);
 145         for(i=0;i<64;i++) {
 146             put_bits(pb, 8, matrix[ ff_zigzag_direct[i] ]);
 147         }
 148     }else
 149         put_bits(pb, 1, 0);
 150 }
 151
 152 /**
 153  * init s->current_picture.qscale_table from s->lambda_table
 154  */
 155 void ff_init_qscale_tab(MpegEncContext *s){
 156     int8_t * const qscale_table= s->current_picture.qscale_table;
 157     int i;
 158
 159     for(i=0; i<s->mb_num; i++){
 160         unsigned int lam= s->lambda_table[ s->mb_index2xy[i] ];
 161         int qp= (lam*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7);
 162         qscale_table[ s->mb_index2xy[i] ]= av_clip(qp, s->avctx->qmin, s->avctx->qmax);
 163     }
 164 }
 165
 166 static void copy_picture_attributes(MpegEncContext *s, AVFrame *dst, AVFrame *src){
 167     int i;
 168
 169     dst->pict_type              = src->pict_type;
 170     dst->quality                = src->quality;
 171     dst->coded_picture_number   = src->coded_picture_number;
 172     dst->display_picture_number = src->display_picture_number;
 173 //    dst->reference              = src->reference;
 174     dst->pts                    = src->pts;
 175     dst->interlaced_frame       = src->interlaced_frame;
 176     dst->top_field_first        = src->top_field_first;
 177
 178     if(s->avctx->me_threshold){
 179         if(!src->motion_val[0])
 180             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_val not set!\n");
 181         if(!src->mb_type)
 182             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.mb_type not set!\n");
 183         if(!src->ref_index[0])
 184             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.ref_index not set!\n");
 185         if(src->motion_subsample_log2 != dst->motion_subsample_log2)
 186             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_subsample_log2 doesn't match! (%d!=%d)\n",
 187             src->motion_subsample_log2, dst->motion_subsample_log2);
 188
 189         memcpy(dst->mb_type, src->mb_type, s->mb_stride * s->mb_height * sizeof(dst->mb_type[0]));
 190
 191         for(i=0; i<2; i++){
 192             int stride= ((16*s->mb_width )>>src->motion_subsample_log2) + 1;
 193             int height= ((16*s->mb_height)>>src->motion_subsample_log2);
 194
 195             if(src->motion_val[i] && src->motion_val[i] != dst->motion_val[i]){
 196                 memcpy(dst->motion_val[i], src->motion_val[i], 2*stride*height*sizeof(int16_t));
 197             }
 198             if(src->ref_index[i] && src->ref_index[i] != dst->ref_index[i]){
 199                 memcpy(dst->ref_index[i], src->ref_index[i], s->b8_stride*2*s->mb_height*sizeof(int8_t));
 200             }
 201         }
 202     }
 203 }
 204
 205 static void update_duplicate_context_after_me(MpegEncContext *dst, MpegEncContext *src){
 206 #define COPY(a) dst->a= src->a
 207     COPY(pict_type);
 208     COPY(current_picture);
 209     COPY(f_code);
 210     COPY(b_code);
 211     COPY(qscale);
 212     COPY(lambda);
 213     COPY(lambda2);
 214     COPY(picture_in_gop_number);
 215     COPY(gop_picture_number);
 216     COPY(frame_pred_frame_dct); //FIXME don't set in encode_header
 217     COPY(progressive_frame); //FIXME don't set in encode_header
 218     COPY(partitioned_frame); //FIXME don't set in encode_header
 219 #undef COPY
 220 }
 221
 222 /**
 223  * sets the given MpegEncContext to defaults for encoding.
 224  * the changed fields will not depend upon the prior state of the MpegEncContext.
 225  */
 226 static void MPV_encode_defaults(MpegEncContext *s){
 227     int i;
 228     MPV_common_defaults(s);
 229
 230     for(i=-16; i<16; i++){
 231         default_fcode_tab[i + MAX_MV]= 1;
 232     }
 233     s->me.mv_penalty= default_mv_penalty;
 234     s->fcode_tab= default_fcode_tab;
 235 }
 236
 237 /* init video encoder */
 238 av_cold int MPV_encode_init(AVCodecContext *avctx)
 239 {
 240     MpegEncContext *s = avctx->priv_data;
 241     int i;
 242     int chroma_h_shift, chroma_v_shift;
 243
 244     MPV_encode_defaults(s);
 245
 246     switch (avctx->codec_id) {
 247     case CODEC_ID_MPEG2VIDEO:
 248         if(avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P){
 249             av_log(avctx, AV_LOG_ERROR, "only YUV420 and YUV422 are supported\n");
 250             return -1;
 251         }
 252         break;
 253     case CODEC_ID_LJPEG:
 254     case CODEC_ID_MJPEG:
 255         if(avctx->pix_fmt != PIX_FMT_YUVJ420P && avctx->pix_fmt != PIX_FMT_YUVJ422P && avctx->pix_fmt != PIX_FMT_RGB32 &&
 256            ((avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P) || avctx->strict_std_compliance>FF_COMPLIANCE_INOFFICIAL)){
 257             av_log(avctx, AV_LOG_ERROR, "colorspace not supported in jpeg\n");
 258             return -1;
 259         }
 260         break;
 261     default:
 262         if(avctx->pix_fmt != PIX_FMT_YUV420P){
 263             av_log(avctx, AV_LOG_ERROR, "only YUV420 is supported\n");
 264             return -1;
 265         }
 266     }
 267
 268     switch (avctx->pix_fmt) {
 269     case PIX_FMT_YUVJ422P:
 270     case PIX_FMT_YUV422P:
 271         s->chroma_format = CHROMA_422;
 272         break;
 273     case PIX_FMT_YUVJ420P:
 274     case PIX_FMT_YUV420P:
 275     default:
 276         s->chroma_format = CHROMA_420;
 277         break;
 278     }
 279
 280     s->bit_rate = avctx->bit_rate;
 281     s->width = avctx->width;
 282     s->height = avctx->height;
 283     if(avctx->gop_size > 600 && avctx->strict_std_compliance>FF_COMPLIANCE_EXPERIMENTAL){
 284         av_log(avctx, AV_LOG_ERROR, "Warning keyframe interval too large! reducing it ...\n");
 285         avctx->gop_size=600;
 286     }
 287     s->gop_size = avctx->gop_size;
 288     s->avctx = avctx;
 289     s->flags= avctx->flags;
 290     s->flags2= avctx->flags2;
 291     s->max_b_frames= avctx->max_b_frames;
 292     s->codec_id= avctx->codec->id;
 293     s->luma_elim_threshold  = avctx->luma_elim_threshold;
 294     s->chroma_elim_threshold= avctx->chroma_elim_threshold;
 295     s->strict_std_compliance= avctx->strict_std_compliance;
 296     s->data_partitioning= avctx->flags & CODEC_FLAG_PART;
 297     s->quarter_sample= (avctx->flags & CODEC_FLAG_QPEL)!=0;
 298     s->mpeg_quant= avctx->mpeg_quant;
 299     s->rtp_mode= !!avctx->rtp_payload_size;
 300     s->intra_dc_precision= avctx->intra_dc_precision;
 301     s->user_specified_pts = AV_NOPTS_VALUE;
 302
 303     if (s->gop_size <= 1) {
 304         s->intra_only = 1;
 305         s->gop_size = 12;
 306     } else {
 307         s->intra_only = 0;
 308     }
 309
 310     s->me_method = avctx->me_method;
 311
 312     /* Fixed QSCALE */
 313     s->fixed_qscale = !!(avctx->flags & CODEC_FLAG_QSCALE);
 314
 315     s->adaptive_quant= (   s->avctx->lumi_masking
 316                         || s->avctx->dark_masking
 317                         || s->avctx->temporal_cplx_masking
 318                         || s->avctx->spatial_cplx_masking
 319                         || s->avctx->p_masking
 320                         || s->avctx->border_masking
 321                         || (s->flags&CODEC_FLAG_QP_RD))
 322                        && !s->fixed_qscale;
 323
 324     s->obmc= !!(s->flags & CODEC_FLAG_OBMC);
 325     s->loop_filter= !!(s->flags & CODEC_FLAG_LOOP_FILTER);
 326     s->alternate_scan= !!(s->flags & CODEC_FLAG_ALT_SCAN);
 327     s->intra_vlc_format= !!(s->flags2 & CODEC_FLAG2_INTRA_VLC);
 328     s->q_scale_type= !!(s->flags2 & CODEC_FLAG2_NON_LINEAR_QUANT);
 329
 330     if(avctx->rc_max_rate && !avctx->rc_buffer_size){
 331         av_log(avctx, AV_LOG_ERROR, "a vbv buffer size is needed, for encoding with a maximum bitrate\n");
 332         return -1;
 333     }
 334
 335     if(avctx->rc_min_rate && avctx->rc_max_rate != avctx->rc_min_rate){
 336         av_log(avctx, AV_LOG_INFO, "Warning min_rate > 0 but min_rate != max_rate isn't recommended!\n");
 337     }
 338
 339     if(avctx->rc_min_rate && avctx->rc_min_rate > avctx->bit_rate){
 340         av_log(avctx, AV_LOG_ERROR, "bitrate below min bitrate\n");
 341         return -1;
 342     }
 343
 344     if(avctx->rc_max_rate && avctx->rc_max_rate < avctx->bit_rate){
 345         av_log(avctx, AV_LOG_INFO, "bitrate above max bitrate\n");
 346         return -1;
 347     }
 348
 349     if(avctx->rc_max_rate && avctx->rc_max_rate == avctx->bit_rate && avctx->rc_max_rate != avctx->rc_min_rate){
 350         av_log(avctx, AV_LOG_INFO, "impossible bitrate constraints, this will fail\n");
 351     }
 352
 353     if(avctx->rc_buffer_size && avctx->bit_rate*av_q2d(avctx->time_base) > avctx->rc_buffer_size){
 354         av_log(avctx, AV_LOG_ERROR, "VBV buffer too small for bitrate\n");
 355         return -1;
 356     }
 357
 358     if(avctx->bit_rate*av_q2d(avctx->time_base) > avctx->bit_rate_tolerance){
 359         av_log(avctx, AV_LOG_ERROR, "bitrate tolerance too small for bitrate\n");
 360         return -1;
 361     }
 362
 363     if(   s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate
 364        && (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO)
 365        && 90000LL * (avctx->rc_buffer_size-1) > s->avctx->rc_max_rate*0xFFFFLL){
 366
 367         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");
 368     }
 369
 370     if((s->flags & CODEC_FLAG_4MV) && s->codec_id != CODEC_ID_MPEG4
 371        && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P && s->codec_id != CODEC_ID_FLV1){
 372         av_log(avctx, AV_LOG_ERROR, "4MV not supported by codec\n");
 373         return -1;
 374     }
 375
 376     if(s->obmc && s->avctx->mb_decision != FF_MB_DECISION_SIMPLE){
 377         av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with simple mb decision\n");
 378         return -1;
 379     }
 380
 381     if(s->obmc && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P){
 382         av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with H263(+)\n");
 383         return -1;
 384     }
 385
 386     if(s->quarter_sample && s->codec_id != CODEC_ID_MPEG4){
 387         av_log(avctx, AV_LOG_ERROR, "qpel not supported by codec\n");
 388         return -1;
 389     }
 390
 391     if(s->data_partitioning && s->codec_id != CODEC_ID_MPEG4){
 392         av_log(avctx, AV_LOG_ERROR, "data partitioning not supported by codec\n");
 393         return -1;
 394     }
 395
 396     if(s->max_b_frames && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO){
 397         av_log(avctx, AV_LOG_ERROR, "b frames not supported by codec\n");
 398         return -1;
 399     }
 400
 401     if ((s->codec_id == CODEC_ID_MPEG4 || s->codec_id == CODEC_ID_H263 ||
 402          s->codec_id == CODEC_ID_H263P) &&
 403         (avctx->sample_aspect_ratio.num > 255 || avctx->sample_aspect_ratio.den > 255)) {
 404         av_log(avctx, AV_LOG_ERROR, "Invalid pixel aspect ratio %i/%i, limit is 255/255\n",
 405                avctx->sample_aspect_ratio.num, avctx->sample_aspect_ratio.den);
 406         return -1;
 407     }
 408
 409     if((s->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN))
 410        && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG2VIDEO){
 411         av_log(avctx, AV_LOG_ERROR, "interlacing not supported by codec\n");
 412         return -1;
 413     }
 414
 415     if(s->mpeg_quant && s->codec_id != CODEC_ID_MPEG4){ //FIXME mpeg2 uses that too
 416         av_log(avctx, AV_LOG_ERROR, "mpeg2 style quantization not supported by codec\n");
 417         return -1;
 418     }
 419
 420     if((s->flags & CODEC_FLAG_CBP_RD) && !avctx->trellis){
 421         av_log(avctx, AV_LOG_ERROR, "CBP RD needs trellis quant\n");
 422         return -1;
 423     }
 424
 425     if((s->flags & CODEC_FLAG_QP_RD) && s->avctx->mb_decision != FF_MB_DECISION_RD){
 426         av_log(avctx, AV_LOG_ERROR, "QP RD needs mbd=2\n");
 427         return -1;
 428     }
 429
 430     if(s->avctx->scenechange_threshold < 1000000000 && (s->flags & CODEC_FLAG_CLOSED_GOP)){
 431         av_log(avctx, AV_LOG_ERROR, "closed gop with scene change detection are not supported yet, set threshold to 1000000000\n");
 432         return -1;
 433     }
 434
 435     if((s->flags2 & CODEC_FLAG2_INTRA_VLC) && s->codec_id != CODEC_ID_MPEG2VIDEO){
 436         av_log(avctx, AV_LOG_ERROR, "intra vlc table not supported by codec\n");
 437         return -1;
 438     }
 439
 440     if(s->flags & CODEC_FLAG_LOW_DELAY){
 441         if (s->codec_id != CODEC_ID_MPEG2VIDEO){
 442             av_log(avctx, AV_LOG_ERROR, "low delay forcing is only available for mpeg2\n");
 443             return -1;
 444         }
 445         if (s->max_b_frames != 0){
 446             av_log(avctx, AV_LOG_ERROR, "b frames cannot be used with low delay\n");
 447             return -1;
 448         }
 449     }
 450
 451     if(s->q_scale_type == 1){
 452         if(s->codec_id != CODEC_ID_MPEG2VIDEO){
 453             av_log(avctx, AV_LOG_ERROR, "non linear quant is only available for mpeg2\n");
 454             return -1;
 455         }
 456         if(avctx->qmax > 12){
 457             av_log(avctx, AV_LOG_ERROR, "non linear quant only supports qmax <= 12 currently\n");
 458             return -1;
 459         }
 460     }
 461
 462     if(s->avctx->thread_count > 1 && s->codec_id != CODEC_ID_MPEG4
 463        && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO
 464        && (s->codec_id != CODEC_ID_H263P || !(s->flags & CODEC_FLAG_H263P_SLICE_STRUCT))){
 465         av_log(avctx, AV_LOG_ERROR, "multi threaded encoding not supported by codec\n");
 466         return -1;
 467     }
 468
 469     if(s->avctx->thread_count < 1){
 470         av_log(avctx, AV_LOG_ERROR, "automatic thread number detection not supported by codec, patch welcome\n");
 471         return -1;
 472     }
 473
 474     if(s->avctx->thread_count > 1)
 475         s->rtp_mode= 1;
 476
 477     if(!avctx->time_base.den || !avctx->time_base.num){
 478         av_log(avctx, AV_LOG_ERROR, "framerate not set\n");
 479         return -1;
 480     }
 481
 482     i= (INT_MAX/2+128)>>8;
 483     if(avctx->me_threshold >= i){
 484         av_log(avctx, AV_LOG_ERROR, "me_threshold too large, max is %d\n", i - 1);
 485         return -1;
 486     }
 487     if(avctx->mb_threshold >= i){
 488         av_log(avctx, AV_LOG_ERROR, "mb_threshold too large, max is %d\n", i - 1);
 489         return -1;
 490     }
 491
 492     if(avctx->b_frame_strategy && (avctx->flags&CODEC_FLAG_PASS2)){
 493         av_log(avctx, AV_LOG_INFO, "notice: b_frame_strategy only affects the first pass\n");
 494         avctx->b_frame_strategy = 0;
 495     }
 496
 497     i= av_gcd(avctx->time_base.den, avctx->time_base.num);
 498     if(i > 1){
 499         av_log(avctx, AV_LOG_INFO, "removing common factors from framerate\n");
 500         avctx->time_base.den /= i;
 501         avctx->time_base.num /= i;
 502 //        return -1;
 503     }
 504
 505     if(s->codec_id==CODEC_ID_MJPEG){
 506         s->intra_quant_bias= 1<<(QUANT_BIAS_SHIFT-1); //(a + x/2)/x
 507         s->inter_quant_bias= 0;
 508     }else if(s->mpeg_quant || s->codec_id==CODEC_ID_MPEG1VIDEO || s->codec_id==CODEC_ID_MPEG2VIDEO){
 509         s->intra_quant_bias= 3<<(QUANT_BIAS_SHIFT-3); //(a + x*3/8)/x
 510         s->inter_quant_bias= 0;
 511     }else{
 512         s->intra_quant_bias=0;
 513         s->inter_quant_bias=-(1<<(QUANT_BIAS_SHIFT-2)); //(a - x/4)/x
 514     }
 515
 516     if(avctx->intra_quant_bias != FF_DEFAULT_QUANT_BIAS)
 517         s->intra_quant_bias= avctx->intra_quant_bias;
 518     if(avctx->inter_quant_bias != FF_DEFAULT_QUANT_BIAS)
 519         s->inter_quant_bias= avctx->inter_quant_bias;
 520
 521     avcodec_get_chroma_sub_sample(avctx->pix_fmt, &chroma_h_shift, &chroma_v_shift);
 522
 523     if(avctx->codec_id == CODEC_ID_MPEG4 && s->avctx->time_base.den > (1<<16)-1){
 524         av_log(avctx, AV_LOG_ERROR, "timebase not supported by mpeg 4 standard\n");
 525         return -1;
 526     }
 527     s->time_increment_bits = av_log2(s->avctx->time_base.den - 1) + 1;
 528
 529     switch(avctx->codec->id) {
 530     case CODEC_ID_MPEG1VIDEO:
 531         s->out_format = FMT_MPEG1;
 532         s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY);
 533         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
 534         break;
 535     case CODEC_ID_MPEG2VIDEO:
 536         s->out_format = FMT_MPEG1;
 537         s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY);
 538         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
 539         s->rtp_mode= 1;
 540         break;
 541     case CODEC_ID_LJPEG:
 542     case CODEC_ID_MJPEG:
 543         s->out_format = FMT_MJPEG;
 544         s->intra_only = 1; /* force intra only for jpeg */
 545         if(avctx->codec->id == CODEC_ID_LJPEG && avctx->pix_fmt == PIX_FMT_BGRA){
 546             s->mjpeg_vsample[0] = s->mjpeg_hsample[0] =
 547             s->mjpeg_vsample[1] = s->mjpeg_hsample[1] =
 548             s->mjpeg_vsample[2] = s->mjpeg_hsample[2] = 1;
 549         }else{
 550             s->mjpeg_vsample[0] = 2;
 551             s->mjpeg_vsample[1] = 2>>chroma_v_shift;
 552             s->mjpeg_vsample[2] = 2>>chroma_v_shift;
 553             s->mjpeg_hsample[0] = 2;
 554             s->mjpeg_hsample[1] = 2>>chroma_h_shift;
 555             s->mjpeg_hsample[2] = 2>>chroma_h_shift;
 556         }
 557         if (!(CONFIG_MJPEG_ENCODER || CONFIG_LJPEG_ENCODER)
 558             || ff_mjpeg_encode_init(s) < 0)
 559             return -1;
 560         avctx->delay=0;
 561         s->low_delay=1;
 562         break;
 563     case CODEC_ID_H261:
 564         if (!CONFIG_H261_ENCODER)  return -1;
 565         if (ff_h261_get_picture_format(s->width, s->height) < 0) {
 566             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);
 567             return -1;
 568         }
 569         s->out_format = FMT_H261;
 570         avctx->delay=0;
 571         s->low_delay=1;
 572         break;
 573     case CODEC_ID_H263:
 574         if (!CONFIG_H263_ENCODER)  return -1;
 575         if (ff_match_2uint16(h263_format, FF_ARRAY_ELEMS(h263_format), s->width, s->height) == 7) {
 576             av_log(avctx, AV_LOG_INFO, "The specified picture size of %dx%d is not valid for the H.263 codec.\nValid sizes are 128x96, 176x144, 352x288, 704x576, and 1408x1152. Try H.263+.\n", s->width, s->height);
 577             return -1;
 578         }
 579         s->out_format = FMT_H263;
 580         s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
 581         avctx->delay=0;
 582         s->low_delay=1;
 583         break;
 584     case CODEC_ID_H263P:
 585         s->out_format = FMT_H263;
 586         s->h263_plus = 1;
 587         /* Fx */
 588         s->umvplus = (avctx->flags & CODEC_FLAG_H263P_UMV) ? 1:0;
 589         s->h263_aic= (avctx->flags & CODEC_FLAG_AC_PRED) ? 1:0;
 590         s->modified_quant= s->h263_aic;
 591         s->alt_inter_vlc= (avctx->flags & CODEC_FLAG_H263P_AIV) ? 1:0;
 592         s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
 593         s->loop_filter= (avctx->flags & CODEC_FLAG_LOOP_FILTER) ? 1:0;
 594         s->unrestricted_mv= s->obmc || s->loop_filter || s->umvplus;
 595         s->h263_slice_structured= (s->flags & CODEC_FLAG_H263P_SLICE_STRUCT) ? 1:0;
 596
 597         /* /Fx */
 598         /* These are just to be sure */
 599         avctx->delay=0;
 600         s->low_delay=1;
 601         break;
 602     case CODEC_ID_FLV1:
 603         s->out_format = FMT_H263;
 604         s->h263_flv = 2; /* format = 1; 11-bit codes */
 605         s->unrestricted_mv = 1;
 606         s->rtp_mode=0; /* don't allow GOB */
 607         avctx->delay=0;
 608         s->low_delay=1;
 609         break;
 610     case CODEC_ID_RV10:
 611         s->out_format = FMT_H263;
 612         avctx->delay=0;
 613         s->low_delay=1;
 614         break;
 615     case CODEC_ID_RV20:
 616         s->out_format = FMT_H263;
 617         avctx->delay=0;
 618         s->low_delay=1;
 619         s->modified_quant=1;
 620         s->h263_aic=1;
 621         s->h263_plus=1;
 622         s->loop_filter=1;
 623         s->unrestricted_mv= s->obmc || s->loop_filter || s->umvplus;
 624         break;
 625     case CODEC_ID_MPEG4:
 626         s->out_format = FMT_H263;
 627         s->h263_pred = 1;
 628         s->unrestricted_mv = 1;
 629         s->low_delay= s->max_b_frames ? 0 : 1;
 630         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
 631         break;
 632     case CODEC_ID_MSMPEG4V1:
 633         s->out_format = FMT_H263;
 634         s->h263_msmpeg4 = 1;
 635         s->h263_pred = 1;
 636         s->unrestricted_mv = 1;
 637         s->msmpeg4_version= 1;
 638         avctx->delay=0;
 639         s->low_delay=1;
 640         break;
 641     case CODEC_ID_MSMPEG4V2:
 642         s->out_format = FMT_H263;
 643         s->h263_msmpeg4 = 1;
 644         s->h263_pred = 1;
 645         s->unrestricted_mv = 1;
 646         s->msmpeg4_version= 2;
 647         avctx->delay=0;
 648         s->low_delay=1;
 649         break;
 650     case CODEC_ID_MSMPEG4V3:
 651         s->out_format = FMT_H263;
 652         s->h263_msmpeg4 = 1;
 653         s->h263_pred = 1;
 654         s->unrestricted_mv = 1;
 655         s->msmpeg4_version= 3;
 656         s->flipflop_rounding=1;
 657         avctx->delay=0;
 658         s->low_delay=1;
 659         break;
 660     case CODEC_ID_WMV1:
 661         s->out_format = FMT_H263;
 662         s->h263_msmpeg4 = 1;
 663         s->h263_pred = 1;
 664         s->unrestricted_mv = 1;
 665         s->msmpeg4_version= 4;
 666         s->flipflop_rounding=1;
 667         avctx->delay=0;
 668         s->low_delay=1;
 669         break;
 670     case CODEC_ID_WMV2:
 671         s->out_format = FMT_H263;
 672         s->h263_msmpeg4 = 1;
 673         s->h263_pred = 1;
 674         s->unrestricted_mv = 1;
 675         s->msmpeg4_version= 5;
 676         s->flipflop_rounding=1;
 677         avctx->delay=0;
 678         s->low_delay=1;
 679         break;
 680     default:
 681         return -1;
 682     }
 683
 684     avctx->has_b_frames= !s->low_delay;
 685
 686     s->encoding = 1;
 687
 688     s->progressive_frame=
 689     s->progressive_sequence= !(avctx->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN));
 690
 691     /* init */
 692     if (MPV_common_init(s) < 0)
 693         return -1;
 694
 695     if(!s->dct_quantize)
 696         s->dct_quantize = dct_quantize_c;
 697     if(!s->denoise_dct)
 698         s->denoise_dct = denoise_dct_c;
 699     s->fast_dct_quantize = s->dct_quantize;
 700     if(avctx->trellis)
 701         s->dct_quantize = dct_quantize_trellis_c;
 702
 703     if((CONFIG_H263P_ENCODER || CONFIG_RV20_ENCODER) && s->modified_quant)
 704         s->chroma_qscale_table= ff_h263_chroma_qscale_table;
 705
 706     s->quant_precision=5;
 707
 708     ff_set_cmp(&s->dsp, s->dsp.ildct_cmp, s->avctx->ildct_cmp);
 709     ff_set_cmp(&s->dsp, s->dsp.frame_skip_cmp, s->avctx->frame_skip_cmp);
 710
 711     if (CONFIG_H261_ENCODER && s->out_format == FMT_H261)
 712         ff_h261_encode_init(s);
 713     if (CONFIG_H263_ENCODER && s->out_format == FMT_H263)
 714         h263_encode_init(s);
 715     if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version)
 716         ff_msmpeg4_encode_init(s);
 717     if ((CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
 718         && s->out_format == FMT_MPEG1)
 719         ff_mpeg1_encode_init(s);
 720
 721     /* init q matrix */
 722     for(i=0;i<64;i++) {
 723         int j= s->dsp.idct_permutation[i];
 724         if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4 && s->mpeg_quant){
 725             s->intra_matrix[j] = ff_mpeg4_default_intra_matrix[i];
 726             s->inter_matrix[j] = ff_mpeg4_default_non_intra_matrix[i];
 727         }else if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
 728             s->intra_matrix[j] =
 729             s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
 730         }else
 731         { /* mpeg1/2 */
 732             s->intra_matrix[j] = ff_mpeg1_default_intra_matrix[i];
 733             s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
 734         }
 735         if(s->avctx->intra_matrix)
 736             s->intra_matrix[j] = s->avctx->intra_matrix[i];
 737         if(s->avctx->inter_matrix)
 738             s->inter_matrix[j] = s->avctx->inter_matrix[i];
 739     }
 740
 741     /* precompute matrix */
 742     /* for mjpeg, we do include qscale in the matrix */
 743     if (s->out_format != FMT_MJPEG) {
 744         ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
 745                        s->intra_matrix, s->intra_quant_bias, avctx->qmin, 31, 1);
 746         ff_convert_matrix(&s->dsp, s->q_inter_matrix, s->q_inter_matrix16,
 747                        s->inter_matrix, s->inter_quant_bias, avctx->qmin, 31, 0);
 748     }
 749
 750     if(ff_rate_control_init(s) < 0)
 751         return -1;
 752
 753     return 0;
 754 }
 755
 756 av_cold int MPV_encode_end(AVCodecContext *avctx)
 757 {
 758     MpegEncContext *s = avctx->priv_data;
 759
 760     ff_rate_control_uninit(s);
 761
 762     MPV_common_end(s);
 763     if ((CONFIG_MJPEG_ENCODER || CONFIG_LJPEG_ENCODER) && s->out_format == FMT_MJPEG)
 764         ff_mjpeg_encode_close(s);
 765
 766     av_freep(&avctx->extradata);
 767
 768     return 0;
 769 }
 770
 771 static int get_sae(uint8_t *src, int ref, int stride){
 772     int x,y;
 773     int acc=0;
 774
 775     for(y=0; y<16; y++){
 776         for(x=0; x<16; x++){
 777             acc+= FFABS(src[x+y*stride] - ref);
 778         }
 779     }
 780
 781     return acc;
 782 }
 783
 784 static int get_intra_count(MpegEncContext *s, uint8_t *src, uint8_t *ref, int stride){
 785     int x, y, w, h;
 786     int acc=0;
 787
 788     w= s->width &~15;
 789     h= s->height&~15;
 790
 791     for(y=0; y<h; y+=16){
 792         for(x=0; x<w; x+=16){
 793             int offset= x + y*stride;
 794             int sad = s->dsp.sad[0](NULL, src + offset, ref + offset, stride, 16);
 795             int mean= (s->dsp.pix_sum(src + offset, stride) + 128)>>8;
 796             int sae = get_sae(src + offset, mean, stride);
 797
 798             acc+= sae + 500 < sad;
 799         }
 800     }
 801     return acc;
 802 }
 803
 804
 805 static int load_input_picture(MpegEncContext *s, AVFrame *pic_arg){
 806     AVFrame *pic=NULL;
 807     int64_t pts;
 808     int i;
 809     const int encoding_delay= s->max_b_frames;
 810     int direct=1;
 811
 812     if(pic_arg){
 813         pts= pic_arg->pts;
 814         pic_arg->display_picture_number= s->input_picture_number++;
 815
 816         if(pts != AV_NOPTS_VALUE){
 817             if(s->user_specified_pts != AV_NOPTS_VALUE){
 818                 int64_t time= pts;
 819                 int64_t last= s->user_specified_pts;
 820
 821                 if(time <= last){
 822                     av_log(s->avctx, AV_LOG_ERROR, "Error, Invalid timestamp=%"PRId64", last=%"PRId64"\n", pts, s->user_specified_pts);
 823                     return -1;
 824                 }
 825             }
 826             s->user_specified_pts= pts;
 827         }else{
 828             if(s->user_specified_pts != AV_NOPTS_VALUE){
 829                 s->user_specified_pts=
 830                 pts= s->user_specified_pts + 1;
 831                 av_log(s->avctx, AV_LOG_INFO, "Warning: AVFrame.pts=? trying to guess (%"PRId64")\n", pts);
 832             }else{
 833                 pts= pic_arg->display_picture_number;
 834             }
 835         }
 836     }
 837
 838   if(pic_arg){
 839     if(encoding_delay && !(s->flags&CODEC_FLAG_INPUT_PRESERVED)) direct=0;
 840     if(pic_arg->linesize[0] != s->linesize) direct=0;
 841     if(pic_arg->linesize[1] != s->uvlinesize) direct=0;
 842     if(pic_arg->linesize[2] != s->uvlinesize) direct=0;
 843
 844 //    av_log(AV_LOG_DEBUG, "%d %d %d %d\n",pic_arg->linesize[0], pic_arg->linesize[1], s->linesize, s->uvlinesize);
 845
 846     if(direct){
 847         i= ff_find_unused_picture(s, 1);
 848
 849         pic= (AVFrame*)&s->picture[i];
 850         pic->reference= 3;
 851
 852         for(i=0; i<4; i++){
 853             pic->data[i]= pic_arg->data[i];
 854             pic->linesize[i]= pic_arg->linesize[i];
 855         }
 856         ff_alloc_picture(s, (Picture*)pic, 1);
 857     }else{
 858         i= ff_find_unused_picture(s, 0);
 859
 860         pic= (AVFrame*)&s->picture[i];
 861         pic->reference= 3;
 862
 863         ff_alloc_picture(s, (Picture*)pic, 0);
 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[FF_P_TYPE]; //s->next_picture_ptr->quality;
 959     b_lambda= s->last_lambda_for[FF_B_TYPE]; //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= FF_I_TYPE;
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 ? FF_P_TYPE : FF_B_TYPE;
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 void 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= FF_I_TYPE;
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 = FF_P_TYPE;
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 != FF_B_TYPE)
1144                     b_frames= i;
1145             }
1146             if(s->input_picture[b_frames]->pict_type == FF_B_TYPE && 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= FF_I_TYPE;
1157               }
1158             }
1159
1160             if(   (s->flags & CODEC_FLAG_CLOSED_GOP)
1161                && b_frames
1162                && s->input_picture[b_frames]->pict_type== FF_I_TYPE)
1163                 b_frames--;
1164
1165             s->reordered_input_picture[0]= s->input_picture[b_frames];
1166             if(s->reordered_input_picture[0]->pict_type != FF_I_TYPE)
1167                 s->reordered_input_picture[0]->pict_type= FF_P_TYPE;
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= FF_B_TYPE;
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!=FF_B_TYPE ? 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             ff_alloc_picture(s, pic, 0);
1190
1191             /* mark us unused / free shared pic */
1192             if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_INTERNAL)
1193                 s->avctx->release_buffer(s->avctx, (AVFrame*)s->reordered_input_picture[0]);
1194             for(i=0; i<4; i++)
1195                 s->reordered_input_picture[0]->data[i]= NULL;
1196             s->reordered_input_picture[0]->type= 0;
1197
1198             copy_picture_attributes(s, (AVFrame*)pic, (AVFrame*)s->reordered_input_picture[0]);
1199
1200             s->current_picture_ptr= pic;
1201         }else{
1202             // input is not a shared pix -> reuse buffer for current_pix
1203
1204             assert(   s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_USER
1205                    || s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
1206
1207             s->current_picture_ptr= s->reordered_input_picture[0];
1208             for(i=0; i<4; i++){
1209                 s->new_picture.data[i]+= INPLACE_OFFSET;
1210             }
1211         }
1212         ff_copy_picture(&s->current_picture, s->current_picture_ptr);
1213
1214         s->picture_number= s->new_picture.display_picture_number;
1215 //printf("dpn:%d\n", s->picture_number);
1216     }else{
1217        memset(&s->new_picture, 0, sizeof(Picture));
1218     }
1219 }
1220
1221 int MPV_encode_picture(AVCodecContext *avctx,
1222                        unsigned char *buf, int buf_size, void *data)
1223 {
1224     MpegEncContext *s = avctx->priv_data;
1225     AVFrame *pic_arg = data;
1226     int i, stuffing_count;
1227
1228     for(i=0; i<avctx->thread_count; i++){
1229         int start_y= s->thread_context[i]->start_mb_y;
1230         int   end_y= s->thread_context[i]->  end_mb_y;
1231         int h= s->mb_height;
1232         uint8_t *start= buf + (size_t)(((int64_t) buf_size)*start_y/h);
1233         uint8_t *end  = buf + (size_t)(((int64_t) buf_size)*  end_y/h);
1234
1235         init_put_bits(&s->thread_context[i]->pb, start, end - start);
1236     }
1237
1238     s->picture_in_gop_number++;
1239
1240     if(load_input_picture(s, pic_arg) < 0)
1241         return -1;
1242
1243     select_input_picture(s);
1244
1245     /* output? */
1246     if(s->new_picture.data[0]){
1247         s->pict_type= s->new_picture.pict_type;
1248 //emms_c();
1249 //printf("qs:%f %f %d\n", s->new_picture.quality, s->current_picture.quality, s->qscale);
1250         MPV_frame_start(s, avctx);
1251 vbv_retry:
1252         if (encode_picture(s, s->picture_number) < 0)
1253             return -1;
1254
1255         avctx->header_bits = s->header_bits;
1256         avctx->mv_bits     = s->mv_bits;
1257         avctx->misc_bits   = s->misc_bits;
1258         avctx->i_tex_bits  = s->i_tex_bits;
1259         avctx->p_tex_bits  = s->p_tex_bits;
1260         avctx->i_count     = s->i_count;
1261         avctx->p_count     = s->mb_num - s->i_count - s->skip_count; //FIXME f/b_count in avctx
1262         avctx->skip_count  = s->skip_count;
1263
1264         MPV_frame_end(s);
1265
1266         if (CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG)
1267             ff_mjpeg_encode_picture_trailer(s);
1268
1269         if(avctx->rc_buffer_size){
1270             RateControlContext *rcc= &s->rc_context;
1271             int max_size= rcc->buffer_index * avctx->rc_max_available_vbv_use;
1272
1273             if(put_bits_count(&s->pb) > max_size && s->lambda < s->avctx->lmax){
1274                 s->next_lambda= FFMAX(s->lambda+1, s->lambda*(s->qscale+1) / s->qscale);
1275                 if(s->adaptive_quant){
1276                     int i;
1277                     for(i=0; i<s->mb_height*s->mb_stride; i++)
1278                         s->lambda_table[i]= FFMAX(s->lambda_table[i]+1, s->lambda_table[i]*(s->qscale+1) / s->qscale);
1279                 }
1280                 s->mb_skipped = 0;        //done in MPV_frame_start()
1281                 if(s->pict_type==FF_P_TYPE){ //done in encode_picture() so we must undo it
1282                     if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
1283                         s->no_rounding ^= 1;
1284                 }
1285                 if(s->pict_type!=FF_B_TYPE){
1286                     s->time_base= s->last_time_base;
1287                     s->last_non_b_time= s->time - s->pp_time;
1288                 }
1289 //                av_log(NULL, AV_LOG_ERROR, "R:%d ", s->next_lambda);
1290                 for(i=0; i<avctx->thread_count; i++){
1291                     PutBitContext *pb= &s->thread_context[i]->pb;
1292                     init_put_bits(pb, pb->buf, pb->buf_end - pb->buf);
1293                 }
1294                 goto vbv_retry;
1295             }
1296
1297             assert(s->avctx->rc_max_rate);
1298         }
1299
1300         if(s->flags&CODEC_FLAG_PASS1)
1301             ff_write_pass1_stats(s);
1302
1303         for(i=0; i<4; i++){
1304             s->current_picture_ptr->error[i]= s->current_picture.error[i];
1305             avctx->error[i] += s->current_picture_ptr->error[i];
1306         }
1307
1308         if(s->flags&CODEC_FLAG_PASS1)
1309             assert(avctx->header_bits + avctx->mv_bits + avctx->misc_bits + avctx->i_tex_bits + avctx->p_tex_bits == put_bits_count(&s->pb));
1310         flush_put_bits(&s->pb);
1311         s->frame_bits  = put_bits_count(&s->pb);
1312
1313         stuffing_count= ff_vbv_update(s, s->frame_bits);
1314         if(stuffing_count){
1315             if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < stuffing_count + 50){
1316                 av_log(s->avctx, AV_LOG_ERROR, "stuffing too large\n");
1317                 return -1;
1318             }
1319
1320             switch(s->codec_id){
1321             case CODEC_ID_MPEG1VIDEO:
1322             case CODEC_ID_MPEG2VIDEO:
1323                 while(stuffing_count--){
1324                     put_bits(&s->pb, 8, 0);
1325                 }
1326             break;
1327             case CODEC_ID_MPEG4:
1328                 put_bits(&s->pb, 16, 0);
1329                 put_bits(&s->pb, 16, 0x1C3);
1330                 stuffing_count -= 4;
1331                 while(stuffing_count--){
1332                     put_bits(&s->pb, 8, 0xFF);
1333                 }
1334             break;
1335             default:
1336                 av_log(s->avctx, AV_LOG_ERROR, "vbv buffer overflow\n");
1337             }
1338             flush_put_bits(&s->pb);
1339             s->frame_bits  = put_bits_count(&s->pb);
1340         }
1341
1342         /* update mpeg1/2 vbv_delay for CBR */
1343         if(s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate && s->out_format == FMT_MPEG1
1344            && 90000LL * (avctx->rc_buffer_size-1) <= s->avctx->rc_max_rate*0xFFFFLL){
1345             int vbv_delay, min_delay;
1346             double inbits = s->avctx->rc_max_rate*av_q2d(s->avctx->time_base);
1347             int    minbits= s->frame_bits - 8*(s->vbv_delay_ptr - s->pb.buf - 1);
1348             double bits   = s->rc_context.buffer_index + minbits - inbits;
1349
1350             if(bits<0)
1351                 av_log(s->avctx, AV_LOG_ERROR, "Internal error, negative bits\n");
1352
1353             assert(s->repeat_first_field==0);
1354
1355             vbv_delay=     bits * 90000                               / s->avctx->rc_max_rate;
1356             min_delay= (minbits * 90000LL + s->avctx->rc_max_rate - 1)/ s->avctx->rc_max_rate;
1357
1358             vbv_delay= FFMAX(vbv_delay, min_delay);
1359
1360             assert(vbv_delay < 0xFFFF);
1361
1362             s->vbv_delay_ptr[0] &= 0xF8;
1363             s->vbv_delay_ptr[0] |= vbv_delay>>13;
1364             s->vbv_delay_ptr[1]  = vbv_delay>>5;
1365             s->vbv_delay_ptr[2] &= 0x07;
1366             s->vbv_delay_ptr[2] |= vbv_delay<<3;
1367         }
1368         s->total_bits += s->frame_bits;
1369         avctx->frame_bits  = s->frame_bits;
1370     }else{
1371         assert((put_bits_ptr(&s->pb) == s->pb.buf));
1372         s->frame_bits=0;
1373     }
1374     assert((s->frame_bits&7)==0);
1375
1376     return s->frame_bits/8;
1377 }
1378
1379 static inline void dct_single_coeff_elimination(MpegEncContext *s, int n, int threshold)
1380 {
1381     static const char tab[64]=
1382         {3,2,2,1,1,1,1,1,
1383          1,1,1,1,1,1,1,1,
1384          1,1,1,1,1,1,1,1,
1385          0,0,0,0,0,0,0,0,
1386          0,0,0,0,0,0,0,0,
1387          0,0,0,0,0,0,0,0,
1388          0,0,0,0,0,0,0,0,
1389          0,0,0,0,0,0,0,0};
1390     int score=0;
1391     int run=0;
1392     int i;
1393     DCTELEM *block= s->block[n];
1394     const int last_index= s->block_last_index[n];
1395     int skip_dc;
1396
1397     if(threshold<0){
1398         skip_dc=0;
1399         threshold= -threshold;
1400     }else
1401         skip_dc=1;
1402
1403     /* Are all we could set to zero already zero? */
1404     if(last_index<=skip_dc - 1) return;
1405
1406     for(i=0; i<=last_index; i++){
1407         const int j = s->intra_scantable.permutated[i];
1408         const int level = FFABS(block[j]);
1409         if(level==1){
1410             if(skip_dc && i==0) continue;
1411             score+= tab[run];
1412             run=0;
1413         }else if(level>1){
1414             return;
1415         }else{
1416             run++;
1417         }
1418     }
1419     if(score >= threshold) return;
1420     for(i=skip_dc; i<=last_index; i++){
1421         const int j = s->intra_scantable.permutated[i];
1422         block[j]=0;
1423     }
1424     if(block[0]) s->block_last_index[n]= 0;
1425     else         s->block_last_index[n]= -1;
1426 }
1427
1428 static inline void clip_coeffs(MpegEncContext *s, DCTELEM *block, int last_index)
1429 {
1430     int i;
1431     const int maxlevel= s->max_qcoeff;
1432     const int minlevel= s->min_qcoeff;
1433     int overflow=0;
1434
1435     if(s->mb_intra){
1436         i=1; //skip clipping of intra dc
1437     }else
1438         i=0;
1439
1440     for(;i<=last_index; i++){
1441         const int j= s->intra_scantable.permutated[i];
1442         int level = block[j];
1443
1444         if     (level>maxlevel){
1445             level=maxlevel;
1446             overflow++;
1447         }else if(level<minlevel){
1448             level=minlevel;
1449             overflow++;
1450         }
1451
1452         block[j]= level;
1453     }
1454
1455     if(overflow && s->avctx->mb_decision == FF_MB_DECISION_SIMPLE)
1456         av_log(s->avctx, AV_LOG_INFO, "warning, clipping %d dct coefficients to %d..%d\n", overflow, minlevel, maxlevel);
1457 }
1458
1459 static void get_visual_weight(int16_t *weight, uint8_t *ptr, int stride){
1460     int x, y;
1461 //FIXME optimize
1462     for(y=0; y<8; y++){
1463         for(x=0; x<8; x++){
1464             int x2, y2;
1465             int sum=0;
1466             int sqr=0;
1467             int count=0;
1468
1469             for(y2= FFMAX(y-1, 0); y2 < FFMIN(8, y+2); y2++){
1470                 for(x2= FFMAX(x-1, 0); x2 < FFMIN(8, x+2); x2++){
1471                     int v= ptr[x2 + y2*stride];
1472                     sum += v;
1473                     sqr += v*v;
1474                     count++;
1475                 }
1476             }
1477             weight[x + 8*y]= (36*ff_sqrt(count*sqr - sum*sum)) / count;
1478         }
1479     }
1480 }
1481
1482 static av_always_inline void encode_mb_internal(MpegEncContext *s, int motion_x, int motion_y, int mb_block_height, int mb_block_count)
1483 {
1484     int16_t weight[8][64];
1485     DCTELEM orig[8][64];
1486     const int mb_x= s->mb_x;
1487     const int mb_y= s->mb_y;
1488     int i;
1489     int skip_dct[8];
1490     int dct_offset   = s->linesize*8; //default for progressive frames
1491     uint8_t *ptr_y, *ptr_cb, *ptr_cr;
1492     int wrap_y, wrap_c;
1493
1494     for(i=0; i<mb_block_count; i++) skip_dct[i]=s->skipdct;
1495
1496     if(s->adaptive_quant){
1497         const int last_qp= s->qscale;
1498         const int mb_xy= mb_x + mb_y*s->mb_stride;
1499
1500         s->lambda= s->lambda_table[mb_xy];
1501         update_qscale(s);
1502
1503         if(!(s->flags&CODEC_FLAG_QP_RD)){
1504             s->qscale= s->current_picture_ptr->qscale_table[mb_xy];
1505             s->dquant= s->qscale - last_qp;
1506
1507             if(s->out_format==FMT_H263){
1508                 s->dquant= av_clip(s->dquant, -2, 2);
1509
1510                 if(s->codec_id==CODEC_ID_MPEG4){
1511                     if(!s->mb_intra){
1512                         if(s->pict_type == FF_B_TYPE){
1513                             if(s->dquant&1 || s->mv_dir&MV_DIRECT)
1514                                 s->dquant= 0;
1515                         }
1516                         if(s->mv_type==MV_TYPE_8X8)
1517                             s->dquant=0;
1518                     }
1519                 }
1520             }
1521         }
1522         ff_set_qscale(s, last_qp + s->dquant);
1523     }else if(s->flags&CODEC_FLAG_QP_RD)
1524         ff_set_qscale(s, s->qscale + s->dquant);
1525
1526     wrap_y = s->linesize;
1527     wrap_c = s->uvlinesize;
1528     ptr_y = s->new_picture.data[0] + (mb_y * 16 * wrap_y) + mb_x * 16;
1529     ptr_cb = s->new_picture.data[1] + (mb_y * mb_block_height * wrap_c) + mb_x * 8;
1530     ptr_cr = s->new_picture.data[2] + (mb_y * mb_block_height * wrap_c) + mb_x * 8;
1531
1532     if(mb_x*16+16 > s->width || mb_y*16+16 > s->height){
1533         uint8_t *ebuf= s->edge_emu_buffer + 32;
1534         ff_emulated_edge_mc(ebuf            , ptr_y , wrap_y,16,16,mb_x*16,mb_y*16, s->width   , s->height);
1535         ptr_y= ebuf;
1536         ff_emulated_edge_mc(ebuf+18*wrap_y  , ptr_cb, wrap_c, 8, mb_block_height, mb_x*8, mb_y*8, s->width>>1, s->height>>1);
1537         ptr_cb= ebuf+18*wrap_y;
1538         ff_emulated_edge_mc(ebuf+18*wrap_y+8, ptr_cr, wrap_c, 8, mb_block_height, mb_x*8, mb_y*8, s->width>>1, s->height>>1);
1539         ptr_cr= ebuf+18*wrap_y+8;
1540     }
1541
1542     if (s->mb_intra) {
1543         if(s->flags&CODEC_FLAG_INTERLACED_DCT){
1544             int progressive_score, interlaced_score;
1545
1546             s->interlaced_dct=0;
1547             progressive_score= s->dsp.ildct_cmp[4](s, ptr_y           , NULL, wrap_y, 8)
1548                               +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y*8, NULL, wrap_y, 8) - 400;
1549
1550             if(progressive_score > 0){
1551                 interlaced_score = s->dsp.ildct_cmp[4](s, ptr_y           , NULL, wrap_y*2, 8)
1552                                   +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y  , NULL, wrap_y*2, 8);
1553                 if(progressive_score > interlaced_score){
1554                     s->interlaced_dct=1;
1555
1556                     dct_offset= wrap_y;
1557                     wrap_y<<=1;
1558                     if (s->chroma_format == CHROMA_422)
1559                         wrap_c<<=1;
1560                 }
1561             }
1562         }
1563
1564         s->dsp.get_pixels(s->block[0], ptr_y                 , wrap_y);
1565         s->dsp.get_pixels(s->block[1], ptr_y              + 8, wrap_y);
1566         s->dsp.get_pixels(s->block[2], ptr_y + dct_offset    , wrap_y);
1567         s->dsp.get_pixels(s->block[3], ptr_y + dct_offset + 8, wrap_y);
1568
1569         if(s->flags&CODEC_FLAG_GRAY){
1570             skip_dct[4]= 1;
1571             skip_dct[5]= 1;
1572         }else{
1573             s->dsp.get_pixels(s->block[4], ptr_cb, wrap_c);
1574             s->dsp.get_pixels(s->block[5], ptr_cr, wrap_c);
1575             if(!s->chroma_y_shift){ /* 422 */
1576                 s->dsp.get_pixels(s->block[6], ptr_cb + (dct_offset>>1), wrap_c);
1577                 s->dsp.get_pixels(s->block[7], ptr_cr + (dct_offset>>1), wrap_c);
1578             }
1579         }
1580     }else{
1581         op_pixels_func (*op_pix)[4];
1582         qpel_mc_func (*op_qpix)[16];
1583         uint8_t *dest_y, *dest_cb, *dest_cr;
1584
1585         dest_y  = s->dest[0];
1586         dest_cb = s->dest[1];
1587         dest_cr = s->dest[2];
1588
1589         if ((!s->no_rounding) || s->pict_type==FF_B_TYPE){
1590             op_pix = s->dsp.put_pixels_tab;
1591             op_qpix= s->dsp.put_qpel_pixels_tab;
1592         }else{
1593             op_pix = s->dsp.put_no_rnd_pixels_tab;
1594             op_qpix= s->dsp.put_no_rnd_qpel_pixels_tab;
1595         }
1596
1597         if (s->mv_dir & MV_DIR_FORWARD) {
1598             MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.data, op_pix, op_qpix);
1599             op_pix = s->dsp.avg_pixels_tab;
1600             op_qpix= s->dsp.avg_qpel_pixels_tab;
1601         }
1602         if (s->mv_dir & MV_DIR_BACKWARD) {
1603             MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.data, op_pix, op_qpix);
1604         }
1605
1606         if(s->flags&CODEC_FLAG_INTERLACED_DCT){
1607             int progressive_score, interlaced_score;
1608
1609             s->interlaced_dct=0;
1610             progressive_score= s->dsp.ildct_cmp[0](s, dest_y           , ptr_y           , wrap_y, 8)
1611                               +s->dsp.ildct_cmp[0](s, dest_y + wrap_y*8, ptr_y + wrap_y*8, wrap_y, 8) - 400;
1612
1613             if(s->avctx->ildct_cmp == FF_CMP_VSSE) progressive_score -= 400;
1614
1615             if(progressive_score>0){
1616                 interlaced_score = s->dsp.ildct_cmp[0](s, dest_y           , ptr_y           , wrap_y*2, 8)
1617                                   +s->dsp.ildct_cmp[0](s, dest_y + wrap_y  , ptr_y + wrap_y  , wrap_y*2, 8);
1618
1619                 if(progressive_score > interlaced_score){
1620                     s->interlaced_dct=1;
1621
1622                     dct_offset= wrap_y;
1623                     wrap_y<<=1;
1624                     if (s->chroma_format == CHROMA_422)
1625                         wrap_c<<=1;
1626                 }
1627             }
1628         }
1629
1630         s->dsp.diff_pixels(s->block[0], ptr_y                 , dest_y                 , wrap_y);
1631         s->dsp.diff_pixels(s->block[1], ptr_y              + 8, dest_y              + 8, wrap_y);
1632         s->dsp.diff_pixels(s->block[2], ptr_y + dct_offset    , dest_y + dct_offset    , wrap_y);
1633         s->dsp.diff_pixels(s->block[3], ptr_y + dct_offset + 8, dest_y + dct_offset + 8, wrap_y);
1634
1635         if(s->flags&CODEC_FLAG_GRAY){
1636             skip_dct[4]= 1;
1637             skip_dct[5]= 1;
1638         }else{
1639             s->dsp.diff_pixels(s->block[4], ptr_cb, dest_cb, wrap_c);
1640             s->dsp.diff_pixels(s->block[5], ptr_cr, dest_cr, wrap_c);
1641             if(!s->chroma_y_shift){ /* 422 */
1642                 s->dsp.diff_pixels(s->block[6], ptr_cb + (dct_offset>>1), dest_cb + (dct_offset>>1), wrap_c);
1643                 s->dsp.diff_pixels(s->block[7], ptr_cr + (dct_offset>>1), dest_cr + (dct_offset>>1), wrap_c);
1644             }
1645         }
1646         /* pre quantization */
1647         if(s->current_picture.mc_mb_var[s->mb_stride*mb_y+ mb_x]<2*s->qscale*s->qscale){
1648             //FIXME optimize
1649             if(s->dsp.sad[1](NULL, ptr_y               , dest_y               , wrap_y, 8) < 20*s->qscale) skip_dct[0]= 1;
1650             if(s->dsp.sad[1](NULL, ptr_y            + 8, dest_y            + 8, wrap_y, 8) < 20*s->qscale) skip_dct[1]= 1;
1651             if(s->dsp.sad[1](NULL, ptr_y +dct_offset   , dest_y +dct_offset   , wrap_y, 8) < 20*s->qscale) skip_dct[2]= 1;
1652             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;
1653             if(s->dsp.sad[1](NULL, ptr_cb              , dest_cb              , wrap_c, 8) < 20*s->qscale) skip_dct[4]= 1;
1654             if(s->dsp.sad[1](NULL, ptr_cr              , dest_cr              , wrap_c, 8) < 20*s->qscale) skip_dct[5]= 1;
1655             if(!s->chroma_y_shift){ /* 422 */
1656                 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;
1657                 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;
1658             }
1659         }
1660     }
1661
1662     if(s->avctx->quantizer_noise_shaping){
1663         if(!skip_dct[0]) get_visual_weight(weight[0], ptr_y                 , wrap_y);
1664         if(!skip_dct[1]) get_visual_weight(weight[1], ptr_y              + 8, wrap_y);
1665         if(!skip_dct[2]) get_visual_weight(weight[2], ptr_y + dct_offset    , wrap_y);
1666         if(!skip_dct[3]) get_visual_weight(weight[3], ptr_y + dct_offset + 8, wrap_y);
1667         if(!skip_dct[4]) get_visual_weight(weight[4], ptr_cb                , wrap_c);
1668         if(!skip_dct[5]) get_visual_weight(weight[5], ptr_cr                , wrap_c);
1669         if(!s->chroma_y_shift){ /* 422 */
1670             if(!skip_dct[6]) get_visual_weight(weight[6], ptr_cb + (dct_offset>>1), wrap_c);
1671             if(!skip_dct[7]) get_visual_weight(weight[7], ptr_cr + (dct_offset>>1), wrap_c);
1672         }
1673         memcpy(orig[0], s->block[0], sizeof(DCTELEM)*64*mb_block_count);
1674     }
1675
1676     /* DCT & quantize */
1677     assert(s->out_format!=FMT_MJPEG || s->qscale==8);
1678     {
1679         for(i=0;i<mb_block_count;i++) {
1680             if(!skip_dct[i]){
1681                 int overflow;
1682                 s->block_last_index[i] = s->dct_quantize(s, s->block[i], i, s->qscale, &overflow);
1683             // FIXME we could decide to change to quantizer instead of clipping
1684             // JS: I don't think that would be a good idea it could lower quality instead
1685             //     of improve it. Just INTRADC clipping deserves changes in quantizer
1686                 if (overflow) clip_coeffs(s, s->block[i], s->block_last_index[i]);
1687             }else
1688                 s->block_last_index[i]= -1;
1689         }
1690         if(s->avctx->quantizer_noise_shaping){
1691             for(i=0;i<mb_block_count;i++) {
1692                 if(!skip_dct[i]){
1693                     s->block_last_index[i] = dct_quantize_refine(s, s->block[i], weight[i], orig[i], i, s->qscale);
1694                 }
1695             }
1696         }
1697
1698         if(s->luma_elim_threshold && !s->mb_intra)
1699             for(i=0; i<4; i++)
1700                 dct_single_coeff_elimination(s, i, s->luma_elim_threshold);
1701         if(s->chroma_elim_threshold && !s->mb_intra)
1702             for(i=4; i<mb_block_count; i++)
1703                 dct_single_coeff_elimination(s, i, s->chroma_elim_threshold);
1704
1705         if(s->flags & CODEC_FLAG_CBP_RD){
1706             for(i=0;i<mb_block_count;i++) {
1707                 if(s->block_last_index[i] == -1)
1708                     s->coded_score[i]= INT_MAX/256;
1709             }
1710         }
1711     }
1712
1713     if((s->flags&CODEC_FLAG_GRAY) && s->mb_intra){
1714         s->block_last_index[4]=
1715         s->block_last_index[5]= 0;
1716         s->block[4][0]=
1717         s->block[5][0]= (1024 + s->c_dc_scale/2)/ s->c_dc_scale;
1718     }
1719
1720     //non c quantize code returns incorrect block_last_index FIXME
1721     if(s->alternate_scan && s->dct_quantize != dct_quantize_c){
1722         for(i=0; i<mb_block_count; i++){
1723             int j;
1724             if(s->block_last_index[i]>0){
1725                 for(j=63; j>0; j--){
1726                     if(s->block[i][ s->intra_scantable.permutated[j] ]) break;
1727                 }
1728                 s->block_last_index[i]= j;
1729             }
1730         }
1731     }
1732
1733     /* huffman encode */
1734     switch(s->codec_id){ //FIXME funct ptr could be slightly faster
1735     case CODEC_ID_MPEG1VIDEO:
1736     case CODEC_ID_MPEG2VIDEO:
1737         if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
1738             mpeg1_encode_mb(s, s->block, motion_x, motion_y);
1739         break;
1740     case CODEC_ID_MPEG4:
1741         if (CONFIG_MPEG4_ENCODER)
1742             mpeg4_encode_mb(s, s->block, motion_x, motion_y);
1743         break;
1744     case CODEC_ID_MSMPEG4V2:
1745     case CODEC_ID_MSMPEG4V3:
1746     case CODEC_ID_WMV1:
1747         if (CONFIG_MSMPEG4_ENCODER)
1748             msmpeg4_encode_mb(s, s->block, motion_x, motion_y);
1749         break;
1750     case CODEC_ID_WMV2:
1751         if (CONFIG_WMV2_ENCODER)
1752             ff_wmv2_encode_mb(s, s->block, motion_x, motion_y);
1753         break;
1754     case CODEC_ID_H261:
1755         if (CONFIG_H261_ENCODER)
1756             ff_h261_encode_mb(s, s->block, motion_x, motion_y);
1757         break;
1758     case CODEC_ID_H263:
1759     case CODEC_ID_H263P:
1760     case CODEC_ID_FLV1:
1761     case CODEC_ID_RV10:
1762     case CODEC_ID_RV20:
1763         if (CONFIG_H263_ENCODER)
1764             h263_encode_mb(s, s->block, motion_x, motion_y);
1765         break;
1766     case CODEC_ID_MJPEG:
1767         if (CONFIG_MJPEG_ENCODER)
1768             ff_mjpeg_encode_mb(s, s->block);
1769         break;
1770     default:
1771         assert(0);
1772     }
1773 }
1774
1775 static av_always_inline void encode_mb(MpegEncContext *s, int motion_x, int motion_y)
1776 {
1777     if (s->chroma_format == CHROMA_420) encode_mb_internal(s, motion_x, motion_y,  8, 6);
1778     else                                encode_mb_internal(s, motion_x, motion_y, 16, 8);
1779 }
1780
1781 static inline void copy_context_before_encode(MpegEncContext *d, MpegEncContext *s, int type){
1782     int i;
1783
1784     memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
1785
1786     /* mpeg1 */
1787     d->mb_skip_run= s->mb_skip_run;
1788     for(i=0; i<3; i++)
1789         d->last_dc[i]= s->last_dc[i];
1790
1791     /* statistics */
1792     d->mv_bits= s->mv_bits;
1793     d->i_tex_bits= s->i_tex_bits;
1794     d->p_tex_bits= s->p_tex_bits;
1795     d->i_count= s->i_count;
1796     d->f_count= s->f_count;
1797     d->b_count= s->b_count;
1798     d->skip_count= s->skip_count;
1799     d->misc_bits= s->misc_bits;
1800     d->last_bits= 0;
1801
1802     d->mb_skipped= 0;
1803     d->qscale= s->qscale;
1804     d->dquant= s->dquant;
1805
1806     d->esc3_level_length= s->esc3_level_length;
1807 }
1808
1809 static inline void copy_context_after_encode(MpegEncContext *d, MpegEncContext *s, int type){
1810     int i;
1811
1812     memcpy(d->mv, s->mv, 2*4*2*sizeof(int));
1813     memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
1814
1815     /* mpeg1 */
1816     d->mb_skip_run= s->mb_skip_run;
1817     for(i=0; i<3; i++)
1818         d->last_dc[i]= s->last_dc[i];
1819
1820     /* statistics */
1821     d->mv_bits= s->mv_bits;
1822     d->i_tex_bits= s->i_tex_bits;
1823     d->p_tex_bits= s->p_tex_bits;
1824     d->i_count= s->i_count;
1825     d->f_count= s->f_count;
1826     d->b_count= s->b_count;
1827     d->skip_count= s->skip_count;
1828     d->misc_bits= s->misc_bits;
1829
1830     d->mb_intra= s->mb_intra;
1831     d->mb_skipped= s->mb_skipped;
1832     d->mv_type= s->mv_type;
1833     d->mv_dir= s->mv_dir;
1834     d->pb= s->pb;
1835     if(s->data_partitioning){
1836         d->pb2= s->pb2;
1837         d->tex_pb= s->tex_pb;
1838     }
1839     d->block= s->block;
1840     for(i=0; i<8; i++)
1841         d->block_last_index[i]= s->block_last_index[i];
1842     d->interlaced_dct= s->interlaced_dct;
1843     d->qscale= s->qscale;
1844
1845     d->esc3_level_length= s->esc3_level_length;
1846 }
1847
1848 static inline void encode_mb_hq(MpegEncContext *s, MpegEncContext *backup, MpegEncContext *best, int type,
1849                            PutBitContext pb[2], PutBitContext pb2[2], PutBitContext tex_pb[2],
1850                            int *dmin, int *next_block, int motion_x, int motion_y)
1851 {
1852     int score;
1853     uint8_t *dest_backup[3];
1854
1855     copy_context_before_encode(s, backup, type);
1856
1857     s->block= s->blocks[*next_block];
1858     s->pb= pb[*next_block];
1859     if(s->data_partitioning){
1860         s->pb2   = pb2   [*next_block];
1861         s->tex_pb= tex_pb[*next_block];
1862     }
1863
1864     if(*next_block){
1865         memcpy(dest_backup, s->dest, sizeof(s->dest));
1866         s->dest[0] = s->rd_scratchpad;
1867         s->dest[1] = s->rd_scratchpad + 16*s->linesize;
1868         s->dest[2] = s->rd_scratchpad + 16*s->linesize + 8;
1869         assert(s->linesize >= 32); //FIXME
1870     }
1871
1872     encode_mb(s, motion_x, motion_y);
1873
1874     score= put_bits_count(&s->pb);
1875     if(s->data_partitioning){
1876         score+= put_bits_count(&s->pb2);
1877         score+= put_bits_count(&s->tex_pb);
1878     }
1879
1880     if(s->avctx->mb_decision == FF_MB_DECISION_RD){
1881         MPV_decode_mb(s, s->block);
1882
1883         score *= s->lambda2;
1884         score += sse_mb(s) << FF_LAMBDA_SHIFT;
1885     }
1886
1887     if(*next_block){
1888         memcpy(s->dest, dest_backup, sizeof(s->dest));
1889     }
1890
1891     if(score<*dmin){
1892         *dmin= score;
1893         *next_block^=1;
1894
1895         copy_context_after_encode(best, s, type);
1896     }
1897 }
1898
1899 static int sse(MpegEncContext *s, uint8_t *src1, uint8_t *src2, int w, int h, int stride){
1900     uint32_t *sq = ff_squareTbl + 256;
1901     int acc=0;
1902     int x,y;
1903
1904     if(w==16 && h==16)
1905         return s->dsp.sse[0](NULL, src1, src2, stride, 16);
1906     else if(w==8 && h==8)
1907         return s->dsp.sse[1](NULL, src1, src2, stride, 8);
1908
1909     for(y=0; y<h; y++){
1910         for(x=0; x<w; x++){
1911             acc+= sq[src1[x + y*stride] - src2[x + y*stride]];
1912         }
1913     }
1914
1915     assert(acc>=0);
1916
1917     return acc;
1918 }
1919
1920 static int sse_mb(MpegEncContext *s){
1921     int w= 16;
1922     int h= 16;
1923
1924     if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
1925     if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
1926
1927     if(w==16 && h==16)
1928       if(s->avctx->mb_cmp == FF_CMP_NSSE){
1929         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)
1930                +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)
1931                +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);
1932       }else{
1933         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)
1934                +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)
1935                +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);
1936       }
1937     else
1938         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)
1939                +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)
1940                +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);
1941 }
1942
1943 static int pre_estimate_motion_thread(AVCodecContext *c, void *arg){
1944     MpegEncContext *s= *(void**)arg;
1945
1946
1947     s->me.pre_pass=1;
1948     s->me.dia_size= s->avctx->pre_dia_size;
1949     s->first_slice_line=1;
1950     for(s->mb_y= s->end_mb_y-1; s->mb_y >= s->start_mb_y; s->mb_y--) {
1951         for(s->mb_x=s->mb_width-1; s->mb_x >=0 ;s->mb_x--) {
1952             ff_pre_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
1953         }
1954         s->first_slice_line=0;
1955     }
1956
1957     s->me.pre_pass=0;
1958
1959     return 0;
1960 }
1961
1962 static int estimate_motion_thread(AVCodecContext *c, void *arg){
1963     MpegEncContext *s= *(void**)arg;
1964
1965     ff_check_alignment();
1966
1967     s->me.dia_size= s->avctx->dia_size;
1968     s->first_slice_line=1;
1969     for(s->mb_y= s->start_mb_y; s->mb_y < s->end_mb_y; s->mb_y++) {
1970         s->mb_x=0; //for block init below
1971         ff_init_block_index(s);
1972         for(s->mb_x=0; s->mb_x < s->mb_width; s->mb_x++) {
1973             s->block_index[0]+=2;
1974             s->block_index[1]+=2;
1975             s->block_index[2]+=2;
1976             s->block_index[3]+=2;
1977
1978             /* compute motion vector & mb_type and store in context */
1979             if(s->pict_type==FF_B_TYPE)
1980                 ff_estimate_b_frame_motion(s, s->mb_x, s->mb_y);
1981             else
1982                 ff_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
1983         }
1984         s->first_slice_line=0;
1985     }
1986     return 0;
1987 }
1988
1989 static int mb_var_thread(AVCodecContext *c, void *arg){
1990     MpegEncContext *s= *(void**)arg;
1991     int mb_x, mb_y;
1992
1993     ff_check_alignment();
1994
1995     for(mb_y=s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
1996         for(mb_x=0; mb_x < s->mb_width; mb_x++) {
1997             int xx = mb_x * 16;
1998             int yy = mb_y * 16;
1999             uint8_t *pix = s->new_picture.data[0] + (yy * s->linesize) + xx;
2000             int varc;
2001             int sum = s->dsp.pix_sum(pix, s->linesize);
2002
2003             varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500 + 128)>>8;
2004
2005             s->current_picture.mb_var [s->mb_stride * mb_y + mb_x] = varc;
2006             s->current_picture.mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
2007             s->me.mb_var_sum_temp    += varc;
2008         }
2009     }
2010     return 0;
2011 }
2012
2013 static void write_slice_end(MpegEncContext *s){
2014     if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4){
2015         if(s->partitioned_frame){
2016             ff_mpeg4_merge_partitions(s);
2017         }
2018
2019         ff_mpeg4_stuffing(&s->pb);
2020     }else if(CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG){
2021         ff_mjpeg_encode_stuffing(&s->pb);
2022     }
2023
2024     align_put_bits(&s->pb);
2025     flush_put_bits(&s->pb);
2026
2027     if((s->flags&CODEC_FLAG_PASS1) && !s->partitioned_frame)
2028         s->misc_bits+= get_bits_diff(s);
2029 }
2030
2031 static int encode_thread(AVCodecContext *c, void *arg){
2032     MpegEncContext *s= *(void**)arg;
2033     int mb_x, mb_y, pdif = 0;
2034     int chr_h= 16>>s->chroma_y_shift;
2035     int i, j;
2036     MpegEncContext best_s, backup_s;
2037     uint8_t bit_buf[2][MAX_MB_BYTES];
2038     uint8_t bit_buf2[2][MAX_MB_BYTES];
2039     uint8_t bit_buf_tex[2][MAX_MB_BYTES];
2040     PutBitContext pb[2], pb2[2], tex_pb[2];
2041 //printf("%d->%d\n", s->resync_mb_y, s->end_mb_y);
2042
2043     ff_check_alignment();
2044
2045     for(i=0; i<2; i++){
2046         init_put_bits(&pb    [i], bit_buf    [i], MAX_MB_BYTES);
2047         init_put_bits(&pb2   [i], bit_buf2   [i], MAX_MB_BYTES);
2048         init_put_bits(&tex_pb[i], bit_buf_tex[i], MAX_MB_BYTES);
2049     }
2050
2051     s->last_bits= put_bits_count(&s->pb);
2052     s->mv_bits=0;
2053     s->misc_bits=0;
2054     s->i_tex_bits=0;
2055     s->p_tex_bits=0;
2056     s->i_count=0;
2057     s->f_count=0;
2058     s->b_count=0;
2059     s->skip_count=0;
2060
2061     for(i=0; i<3; i++){
2062         /* init last dc values */
2063         /* note: quant matrix value (8) is implied here */
2064         s->last_dc[i] = 128 << s->intra_dc_precision;
2065
2066         s->current_picture.error[i] = 0;
2067     }
2068     s->mb_skip_run = 0;
2069     memset(s->last_mv, 0, sizeof(s->last_mv));
2070
2071     s->last_mv_dir = 0;
2072
2073     switch(s->codec_id){
2074     case CODEC_ID_H263:
2075     case CODEC_ID_H263P:
2076     case CODEC_ID_FLV1:
2077         if (CONFIG_H263_ENCODER)
2078             s->gob_index = ff_h263_get_gob_height(s);
2079         break;
2080     case CODEC_ID_MPEG4:
2081         if(CONFIG_MPEG4_ENCODER && s->partitioned_frame)
2082             ff_mpeg4_init_partitions(s);
2083         break;
2084     }
2085
2086     s->resync_mb_x=0;
2087     s->resync_mb_y=0;
2088     s->first_slice_line = 1;
2089     s->ptr_lastgob = s->pb.buf;
2090     for(mb_y= s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
2091 //    printf("row %d at %X\n", s->mb_y, (int)s);
2092         s->mb_x=0;
2093         s->mb_y= mb_y;
2094
2095         ff_set_qscale(s, s->qscale);
2096         ff_init_block_index(s);
2097
2098         for(mb_x=0; mb_x < s->mb_width; mb_x++) {
2099             int xy= mb_y*s->mb_stride + mb_x; // removed const, H261 needs to adjust this
2100             int mb_type= s->mb_type[xy];
2101 //            int d;
2102             int dmin= INT_MAX;
2103             int dir;
2104
2105             if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < MAX_MB_BYTES){
2106                 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
2107                 return -1;
2108             }
2109             if(s->data_partitioning){
2110                 if(   s->pb2   .buf_end - s->pb2   .buf - (put_bits_count(&s->    pb2)>>3) < MAX_MB_BYTES
2111                    || s->tex_pb.buf_end - s->tex_pb.buf - (put_bits_count(&s->tex_pb )>>3) < MAX_MB_BYTES){
2112                     av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
2113                     return -1;
2114                 }
2115             }
2116
2117             s->mb_x = mb_x;
2118             s->mb_y = mb_y;  // moved into loop, can get changed by H.261
2119             ff_update_block_index(s);
2120
2121             if(CONFIG_H261_ENCODER && s->codec_id == CODEC_ID_H261){
2122                 ff_h261_reorder_mb_index(s);
2123                 xy= s->mb_y*s->mb_stride + s->mb_x;
2124                 mb_type= s->mb_type[xy];
2125             }
2126
2127             /* write gob / video packet header  */
2128             if(s->rtp_mode){
2129                 int current_packet_size, is_gob_start;
2130
2131                 current_packet_size= ((put_bits_count(&s->pb)+7)>>3) - (s->ptr_lastgob - s->pb.buf);
2132
2133                 is_gob_start= s->avctx->rtp_payload_size && current_packet_size >= s->avctx->rtp_payload_size && mb_y + mb_x>0;
2134
2135                 if(s->start_mb_y == mb_y && mb_y > 0 && mb_x==0) is_gob_start=1;
2136
2137                 switch(s->codec_id){
2138                 case CODEC_ID_H263:
2139                 case CODEC_ID_H263P:
2140                     if(!s->h263_slice_structured)
2141                         if(s->mb_x || s->mb_y%s->gob_index) is_gob_start=0;
2142                     break;
2143                 case CODEC_ID_MPEG2VIDEO:
2144                     if(s->mb_x==0 && s->mb_y!=0) is_gob_start=1;
2145                 case CODEC_ID_MPEG1VIDEO:
2146                     if(s->mb_skip_run) is_gob_start=0;
2147                     break;
2148                 }
2149
2150                 if(is_gob_start){
2151                     if(s->start_mb_y != mb_y || mb_x!=0){
2152                         write_slice_end(s);
2153
2154                         if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4 && s->partitioned_frame){
2155                             ff_mpeg4_init_partitions(s);
2156                         }
2157                     }
2158
2159                     assert((put_bits_count(&s->pb)&7) == 0);
2160                     current_packet_size= put_bits_ptr(&s->pb) - s->ptr_lastgob;
2161
2162                     if(s->avctx->error_rate && s->resync_mb_x + s->resync_mb_y > 0){
2163                         int r= put_bits_count(&s->pb)/8 + s->picture_number + 16 + s->mb_x + s->mb_y;
2164                         int d= 100 / s->avctx->error_rate;
2165                         if(r % d == 0){
2166                             current_packet_size=0;
2167 #ifndef ALT_BITSTREAM_WRITER
2168                             s->pb.buf_ptr= s->ptr_lastgob;
2169 #endif
2170                             assert(put_bits_ptr(&s->pb) == s->ptr_lastgob);
2171                         }
2172                     }
2173
2174                     if (s->avctx->rtp_callback){
2175                         int number_mb = (mb_y - s->resync_mb_y)*s->mb_width + mb_x - s->resync_mb_x;
2176                         s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, current_packet_size, number_mb);
2177                     }
2178
2179                     switch(s->codec_id){
2180                     case CODEC_ID_MPEG4:
2181                         if (CONFIG_MPEG4_ENCODER) {
2182                             ff_mpeg4_encode_video_packet_header(s);
2183                             ff_mpeg4_clean_buffers(s);
2184                         }
2185                     break;
2186                     case CODEC_ID_MPEG1VIDEO:
2187                     case CODEC_ID_MPEG2VIDEO:
2188                         if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER) {
2189                             ff_mpeg1_encode_slice_header(s);
2190                             ff_mpeg1_clean_buffers(s);
2191                         }
2192                     break;
2193                     case CODEC_ID_H263:
2194                     case CODEC_ID_H263P:
2195                         if (CONFIG_H263_ENCODER)
2196                             h263_encode_gob_header(s, mb_y);
2197                     break;
2198                     }
2199
2200                     if(s->flags&CODEC_FLAG_PASS1){
2201                         int bits= put_bits_count(&s->pb);
2202                         s->misc_bits+= bits - s->last_bits;
2203                         s->last_bits= bits;
2204                     }
2205
2206                     s->ptr_lastgob += current_packet_size;
2207                     s->first_slice_line=1;
2208                     s->resync_mb_x=mb_x;
2209                     s->resync_mb_y=mb_y;
2210                 }
2211             }
2212
2213             if(  (s->resync_mb_x   == s->mb_x)
2214                && s->resync_mb_y+1 == s->mb_y){
2215                 s->first_slice_line=0;
2216             }
2217
2218             s->mb_skipped=0;
2219             s->dquant=0; //only for QP_RD
2220
2221             if(mb_type & (mb_type-1) || (s->flags & CODEC_FLAG_QP_RD)){ // more than 1 MB type possible or CODEC_FLAG_QP_RD
2222                 int next_block=0;
2223                 int pb_bits_count, pb2_bits_count, tex_pb_bits_count;
2224
2225                 copy_context_before_encode(&backup_s, s, -1);
2226                 backup_s.pb= s->pb;
2227                 best_s.data_partitioning= s->data_partitioning;
2228                 best_s.partitioned_frame= s->partitioned_frame;
2229                 if(s->data_partitioning){
2230                     backup_s.pb2= s->pb2;
2231                     backup_s.tex_pb= s->tex_pb;
2232                 }
2233
2234                 if(mb_type&CANDIDATE_MB_TYPE_INTER){
2235                     s->mv_dir = MV_DIR_FORWARD;
2236                     s->mv_type = MV_TYPE_16X16;
2237                     s->mb_intra= 0;
2238                     s->mv[0][0][0] = s->p_mv_table[xy][0];
2239                     s->mv[0][0][1] = s->p_mv_table[xy][1];
2240                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER, pb, pb2, tex_pb,
2241                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2242                 }
2243                 if(mb_type&CANDIDATE_MB_TYPE_INTER_I){
2244                     s->mv_dir = MV_DIR_FORWARD;
2245                     s->mv_type = MV_TYPE_FIELD;
2246                     s->mb_intra= 0;
2247                     for(i=0; i<2; i++){
2248                         j= s->field_select[0][i] = s->p_field_select_table[i][xy];
2249                         s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
2250                         s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
2251                     }
2252                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER_I, pb, pb2, tex_pb,
2253                                  &dmin, &next_block, 0, 0);
2254                 }
2255                 if(mb_type&CANDIDATE_MB_TYPE_SKIPPED){
2256                     s->mv_dir = MV_DIR_FORWARD;
2257                     s->mv_type = MV_TYPE_16X16;
2258                     s->mb_intra= 0;
2259                     s->mv[0][0][0] = 0;
2260                     s->mv[0][0][1] = 0;
2261                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_SKIPPED, pb, pb2, tex_pb,
2262                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2263                 }
2264                 if(mb_type&CANDIDATE_MB_TYPE_INTER4V){
2265                     s->mv_dir = MV_DIR_FORWARD;
2266                     s->mv_type = MV_TYPE_8X8;
2267                     s->mb_intra= 0;
2268                     for(i=0; i<4; i++){
2269                         s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
2270                         s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
2271                     }
2272                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER4V, pb, pb2, tex_pb,
2273                                  &dmin, &next_block, 0, 0);
2274                 }
2275                 if(mb_type&CANDIDATE_MB_TYPE_FORWARD){
2276                     s->mv_dir = MV_DIR_FORWARD;
2277                     s->mv_type = MV_TYPE_16X16;
2278                     s->mb_intra= 0;
2279                     s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
2280                     s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2281                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD, pb, pb2, tex_pb,
2282                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2283                 }
2284                 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD){
2285                     s->mv_dir = MV_DIR_BACKWARD;
2286                     s->mv_type = MV_TYPE_16X16;
2287                     s->mb_intra= 0;
2288                     s->mv[1][0][0] = s->b_back_mv_table[xy][0];
2289                     s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2290                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD, pb, pb2, tex_pb,
2291                                  &dmin, &next_block, s->mv[1][0][0], s->mv[1][0][1]);
2292                 }
2293                 if(mb_type&CANDIDATE_MB_TYPE_BIDIR){
2294                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2295                     s->mv_type = MV_TYPE_16X16;
2296                     s->mb_intra= 0;
2297                     s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
2298                     s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
2299                     s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
2300                     s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2301                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR, pb, pb2, tex_pb,
2302                                  &dmin, &next_block, 0, 0);
2303                 }
2304                 if(mb_type&CANDIDATE_MB_TYPE_FORWARD_I){
2305                     s->mv_dir = MV_DIR_FORWARD;
2306                     s->mv_type = MV_TYPE_FIELD;
2307                     s->mb_intra= 0;
2308                     for(i=0; i<2; i++){
2309                         j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
2310                         s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
2311                         s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
2312                     }
2313                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD_I, pb, pb2, tex_pb,
2314                                  &dmin, &next_block, 0, 0);
2315                 }
2316                 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD_I){
2317                     s->mv_dir = MV_DIR_BACKWARD;
2318                     s->mv_type = MV_TYPE_FIELD;
2319                     s->mb_intra= 0;
2320                     for(i=0; i<2; i++){
2321                         j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
2322                         s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
2323                         s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
2324                     }
2325                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD_I, pb, pb2, tex_pb,
2326                                  &dmin, &next_block, 0, 0);
2327                 }
2328                 if(mb_type&CANDIDATE_MB_TYPE_BIDIR_I){
2329                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2330                     s->mv_type = MV_TYPE_FIELD;
2331                     s->mb_intra= 0;
2332                     for(dir=0; dir<2; dir++){
2333                         for(i=0; i<2; i++){
2334                             j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
2335                             s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
2336                             s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
2337                         }
2338                     }
2339                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR_I, pb, pb2, tex_pb,
2340                                  &dmin, &next_block, 0, 0);
2341                 }
2342                 if(mb_type&CANDIDATE_MB_TYPE_INTRA){
2343                     s->mv_dir = 0;
2344                     s->mv_type = MV_TYPE_16X16;
2345                     s->mb_intra= 1;
2346                     s->mv[0][0][0] = 0;
2347                     s->mv[0][0][1] = 0;
2348                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTRA, pb, pb2, tex_pb,
2349                                  &dmin, &next_block, 0, 0);
2350                     if(s->h263_pred || s->h263_aic){
2351                         if(best_s.mb_intra)
2352                             s->mbintra_table[mb_x + mb_y*s->mb_stride]=1;
2353                         else
2354                             ff_clean_intra_table_entries(s); //old mode?
2355                     }
2356                 }
2357
2358                 if((s->flags & CODEC_FLAG_QP_RD) && dmin < INT_MAX){
2359                     if(best_s.mv_type==MV_TYPE_16X16){ //FIXME move 4mv after QPRD
2360                         const int last_qp= backup_s.qscale;
2361                         int qpi, qp, dc[6];
2362                         DCTELEM ac[6][16];
2363                         const int mvdir= (best_s.mv_dir&MV_DIR_BACKWARD) ? 1 : 0;
2364                         static const int dquant_tab[4]={-1,1,-2,2};
2365
2366                         assert(backup_s.dquant == 0);
2367
2368                         //FIXME intra
2369                         s->mv_dir= best_s.mv_dir;
2370                         s->mv_type = MV_TYPE_16X16;
2371                         s->mb_intra= best_s.mb_intra;
2372                         s->mv[0][0][0] = best_s.mv[0][0][0];
2373                         s->mv[0][0][1] = best_s.mv[0][0][1];
2374                         s->mv[1][0][0] = best_s.mv[1][0][0];
2375                         s->mv[1][0][1] = best_s.mv[1][0][1];
2376
2377                         qpi = s->pict_type == FF_B_TYPE ? 2 : 0;
2378                         for(; qpi<4; qpi++){
2379                             int dquant= dquant_tab[qpi];
2380                             qp= last_qp + dquant;
2381                             if(qp < s->avctx->qmin || qp > s->avctx->qmax)
2382                                 continue;
2383                             backup_s.dquant= dquant;
2384                             if(s->mb_intra && s->dc_val[0]){
2385                                 for(i=0; i<6; i++){
2386                                     dc[i]= s->dc_val[0][ s->block_index[i] ];
2387                                     memcpy(ac[i], s->ac_val[0][s->block_index[i]], sizeof(DCTELEM)*16);
2388                                 }
2389                             }
2390
2391                             encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
2392                                          &dmin, &next_block, s->mv[mvdir][0][0], s->mv[mvdir][0][1]);
2393                             if(best_s.qscale != qp){
2394                                 if(s->mb_intra && s->dc_val[0]){
2395                                     for(i=0; i<6; i++){
2396                                         s->dc_val[0][ s->block_index[i] ]= dc[i];
2397                                         memcpy(s->ac_val[0][s->block_index[i]], ac[i], sizeof(DCTELEM)*16);
2398                                     }
2399                                 }
2400                             }
2401                         }
2402                     }
2403                 }
2404                 if(CONFIG_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT){
2405                     int mx= s->b_direct_mv_table[xy][0];
2406                     int my= s->b_direct_mv_table[xy][1];
2407
2408                     backup_s.dquant = 0;
2409                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2410                     s->mb_intra= 0;
2411                     ff_mpeg4_set_direct_mv(s, mx, my);
2412                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
2413                                  &dmin, &next_block, mx, my);
2414                 }
2415                 if(CONFIG_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT0){
2416                     backup_s.dquant = 0;
2417                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2418                     s->mb_intra= 0;
2419                     ff_mpeg4_set_direct_mv(s, 0, 0);
2420                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
2421                                  &dmin, &next_block, 0, 0);
2422                 }
2423                 if(!best_s.mb_intra && s->flags2&CODEC_FLAG2_SKIP_RD){
2424                     int coded=0;
2425                     for(i=0; i<6; i++)
2426                         coded |= s->block_last_index[i];
2427                     if(coded){
2428                         int mx,my;
2429                         memcpy(s->mv, best_s.mv, sizeof(s->mv));
2430                         if(CONFIG_MPEG4_ENCODER && best_s.mv_dir & MV_DIRECT){
2431                             mx=my=0; //FIXME find the one we actually used
2432                             ff_mpeg4_set_direct_mv(s, mx, my);
2433                         }else if(best_s.mv_dir&MV_DIR_BACKWARD){
2434                             mx= s->mv[1][0][0];
2435                             my= s->mv[1][0][1];
2436                         }else{
2437                             mx= s->mv[0][0][0];
2438                             my= s->mv[0][0][1];
2439                         }
2440
2441                         s->mv_dir= best_s.mv_dir;
2442                         s->mv_type = best_s.mv_type;
2443                         s->mb_intra= 0;
2444 /*                        s->mv[0][0][0] = best_s.mv[0][0][0];
2445                         s->mv[0][0][1] = best_s.mv[0][0][1];
2446                         s->mv[1][0][0] = best_s.mv[1][0][0];
2447                         s->mv[1][0][1] = best_s.mv[1][0][1];*/
2448                         backup_s.dquant= 0;
2449                         s->skipdct=1;
2450                         encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
2451                                         &dmin, &next_block, mx, my);
2452                         s->skipdct=0;
2453                     }
2454                 }
2455
2456                 s->current_picture.qscale_table[xy]= best_s.qscale;
2457
2458                 copy_context_after_encode(s, &best_s, -1);
2459
2460                 pb_bits_count= put_bits_count(&s->pb);
2461                 flush_put_bits(&s->pb);
2462                 ff_copy_bits(&backup_s.pb, bit_buf[next_block^1], pb_bits_count);
2463                 s->pb= backup_s.pb;
2464
2465                 if(s->data_partitioning){
2466                     pb2_bits_count= put_bits_count(&s->pb2);
2467                     flush_put_bits(&s->pb2);
2468                     ff_copy_bits(&backup_s.pb2, bit_buf2[next_block^1], pb2_bits_count);
2469                     s->pb2= backup_s.pb2;
2470
2471                     tex_pb_bits_count= put_bits_count(&s->tex_pb);
2472                     flush_put_bits(&s->tex_pb);
2473                     ff_copy_bits(&backup_s.tex_pb, bit_buf_tex[next_block^1], tex_pb_bits_count);
2474                     s->tex_pb= backup_s.tex_pb;
2475                 }
2476                 s->last_bits= put_bits_count(&s->pb);
2477
2478                 if (CONFIG_H263_ENCODER &&
2479                     s->out_format == FMT_H263 && s->pict_type!=FF_B_TYPE)
2480                     ff_h263_update_motion_val(s);
2481
2482                 if(next_block==0){ //FIXME 16 vs linesize16
2483                     s->dsp.put_pixels_tab[0][0](s->dest[0], s->rd_scratchpad                     , s->linesize  ,16);
2484                     s->dsp.put_pixels_tab[1][0](s->dest[1], s->rd_scratchpad + 16*s->linesize    , s->uvlinesize, 8);
2485                     s->dsp.put_pixels_tab[1][0](s->dest[2], s->rd_scratchpad + 16*s->linesize + 8, s->uvlinesize, 8);
2486                 }
2487
2488                 if(s->avctx->mb_decision == FF_MB_DECISION_BITS)
2489                     MPV_decode_mb(s, s->block);
2490             } else {
2491                 int motion_x = 0, motion_y = 0;
2492                 s->mv_type=MV_TYPE_16X16;
2493                 // only one MB-Type possible
2494
2495                 switch(mb_type){
2496                 case CANDIDATE_MB_TYPE_INTRA:
2497                     s->mv_dir = 0;
2498                     s->mb_intra= 1;
2499                     motion_x= s->mv[0][0][0] = 0;
2500                     motion_y= s->mv[0][0][1] = 0;
2501                     break;
2502                 case CANDIDATE_MB_TYPE_INTER:
2503                     s->mv_dir = MV_DIR_FORWARD;
2504                     s->mb_intra= 0;
2505                     motion_x= s->mv[0][0][0] = s->p_mv_table[xy][0];
2506                     motion_y= s->mv[0][0][1] = s->p_mv_table[xy][1];
2507                     break;
2508                 case CANDIDATE_MB_TYPE_INTER_I:
2509                     s->mv_dir = MV_DIR_FORWARD;
2510                     s->mv_type = MV_TYPE_FIELD;
2511                     s->mb_intra= 0;
2512                     for(i=0; i<2; i++){
2513                         j= s->field_select[0][i] = s->p_field_select_table[i][xy];
2514                         s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
2515                         s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
2516                     }
2517                     break;
2518                 case CANDIDATE_MB_TYPE_INTER4V:
2519                     s->mv_dir = MV_DIR_FORWARD;
2520                     s->mv_type = MV_TYPE_8X8;
2521                     s->mb_intra= 0;
2522                     for(i=0; i<4; i++){
2523                         s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
2524                         s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
2525                     }
2526                     break;
2527                 case CANDIDATE_MB_TYPE_DIRECT:
2528                     if (CONFIG_MPEG4_ENCODER) {
2529                         s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
2530                         s->mb_intra= 0;
2531                         motion_x=s->b_direct_mv_table[xy][0];
2532                         motion_y=s->b_direct_mv_table[xy][1];
2533                         ff_mpeg4_set_direct_mv(s, motion_x, motion_y);
2534                     }
2535                     break;
2536                 case CANDIDATE_MB_TYPE_DIRECT0:
2537                     if (CONFIG_MPEG4_ENCODER) {
2538                         s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
2539                         s->mb_intra= 0;
2540                         ff_mpeg4_set_direct_mv(s, 0, 0);
2541                     }
2542                     break;
2543                 case CANDIDATE_MB_TYPE_BIDIR:
2544                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2545                     s->mb_intra= 0;
2546                     s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
2547                     s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
2548                     s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
2549                     s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2550                     break;
2551                 case CANDIDATE_MB_TYPE_BACKWARD:
2552                     s->mv_dir = MV_DIR_BACKWARD;
2553                     s->mb_intra= 0;
2554                     motion_x= s->mv[1][0][0] = s->b_back_mv_table[xy][0];
2555                     motion_y= s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2556                     break;
2557                 case CANDIDATE_MB_TYPE_FORWARD:
2558                     s->mv_dir = MV_DIR_FORWARD;
2559                     s->mb_intra= 0;
2560                     motion_x= s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
2561                     motion_y= s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2562 //                    printf(" %d %d ", motion_x, motion_y);
2563                     break;
2564                 case CANDIDATE_MB_TYPE_FORWARD_I:
2565                     s->mv_dir = MV_DIR_FORWARD;
2566                     s->mv_type = MV_TYPE_FIELD;
2567                     s->mb_intra= 0;
2568                     for(i=0; i<2; i++){
2569                         j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
2570                         s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
2571                         s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
2572                     }
2573                     break;
2574                 case CANDIDATE_MB_TYPE_BACKWARD_I:
2575                     s->mv_dir = MV_DIR_BACKWARD;
2576                     s->mv_type = MV_TYPE_FIELD;
2577                     s->mb_intra= 0;
2578                     for(i=0; i<2; i++){
2579                         j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
2580                         s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
2581                         s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
2582                     }
2583                     break;
2584                 case CANDIDATE_MB_TYPE_BIDIR_I:
2585                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2586                     s->mv_type = MV_TYPE_FIELD;
2587                     s->mb_intra= 0;
2588                     for(dir=0; dir<2; dir++){
2589                         for(i=0; i<2; i++){
2590                             j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
2591                             s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
2592                             s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
2593                         }
2594                     }
2595                     break;
2596                 default:
2597                     av_log(s->avctx, AV_LOG_ERROR, "illegal MB type\n");
2598                 }
2599
2600                 encode_mb(s, motion_x, motion_y);
2601
2602                 // RAL: Update last macroblock type
2603                 s->last_mv_dir = s->mv_dir;
2604
2605                 if (CONFIG_H263_ENCODER &&
2606                     s->out_format == FMT_H263 && s->pict_type!=FF_B_TYPE)
2607                     ff_h263_update_motion_val(s);
2608
2609                 MPV_decode_mb(s, s->block);
2610             }
2611
2612             /* clean the MV table in IPS frames for direct mode in B frames */
2613             if(s->mb_intra /* && I,P,S_TYPE */){
2614                 s->p_mv_table[xy][0]=0;
2615                 s->p_mv_table[xy][1]=0;
2616             }
2617
2618             if(s->flags&CODEC_FLAG_PSNR){
2619                 int w= 16;
2620                 int h= 16;
2621
2622                 if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
2623                 if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
2624
2625                 s->current_picture.error[0] += sse(
2626                     s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16,
2627                     s->dest[0], w, h, s->linesize);
2628                 s->current_picture.error[1] += sse(
2629                     s, s->new_picture.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*chr_h,
2630                     s->dest[1], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
2631                 s->current_picture.error[2] += sse(
2632                     s, s->new_picture.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*chr_h,
2633                     s->dest[2], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
2634             }
2635             if(s->loop_filter){
2636                 if(CONFIG_H263_ENCODER && s->out_format == FMT_H263)
2637                     ff_h263_loop_filter(s);
2638             }
2639 //printf("MB %d %d bits\n", s->mb_x+s->mb_y*s->mb_stride, put_bits_count(&s->pb));
2640         }
2641     }
2642
2643     //not beautiful here but we must write it before flushing so it has to be here
2644     if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version && s->msmpeg4_version<4 && s->pict_type == FF_I_TYPE)
2645         msmpeg4_encode_ext_header(s);
2646
2647     write_slice_end(s);
2648
2649     /* Send the last GOB if RTP */
2650     if (s->avctx->rtp_callback) {
2651         int number_mb = (mb_y - s->resync_mb_y)*s->mb_width - s->resync_mb_x;
2652         pdif = put_bits_ptr(&s->pb) - s->ptr_lastgob;
2653         /* Call the RTP callback to send the last GOB */
2654         emms_c();
2655         s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, pdif, number_mb);
2656     }
2657
2658     return 0;
2659 }
2660
2661 #define MERGE(field) dst->field += src->field; src->field=0
2662 static void merge_context_after_me(MpegEncContext *dst, MpegEncContext *src){
2663     MERGE(me.scene_change_score);
2664     MERGE(me.mc_mb_var_sum_temp);
2665     MERGE(me.mb_var_sum_temp);
2666 }
2667
2668 static void merge_context_after_encode(MpegEncContext *dst, MpegEncContext *src){
2669     int i;
2670
2671     MERGE(dct_count[0]); //note, the other dct vars are not part of the context
2672     MERGE(dct_count[1]);
2673     MERGE(mv_bits);
2674     MERGE(i_tex_bits);
2675     MERGE(p_tex_bits);
2676     MERGE(i_count);
2677     MERGE(f_count);
2678     MERGE(b_count);
2679     MERGE(skip_count);
2680     MERGE(misc_bits);
2681     MERGE(error_count);
2682     MERGE(padding_bug_score);
2683     MERGE(current_picture.error[0]);
2684     MERGE(current_picture.error[1]);
2685     MERGE(current_picture.error[2]);
2686
2687     if(dst->avctx->noise_reduction){
2688         for(i=0; i<64; i++){
2689             MERGE(dct_error_sum[0][i]);
2690             MERGE(dct_error_sum[1][i]);
2691         }
2692     }
2693
2694     assert(put_bits_count(&src->pb) % 8 ==0);
2695     assert(put_bits_count(&dst->pb) % 8 ==0);
2696     ff_copy_bits(&dst->pb, src->pb.buf, put_bits_count(&src->pb));
2697     flush_put_bits(&dst->pb);
2698 }
2699
2700 static int estimate_qp(MpegEncContext *s, int dry_run){
2701     if (s->next_lambda){
2702         s->current_picture_ptr->quality=
2703         s->current_picture.quality = s->next_lambda;
2704         if(!dry_run) s->next_lambda= 0;
2705     } else if (!s->fixed_qscale) {
2706         s->current_picture_ptr->quality=
2707         s->current_picture.quality = ff_rate_estimate_qscale(s, dry_run);
2708         if (s->current_picture.quality < 0)
2709             return -1;
2710     }
2711
2712     if(s->adaptive_quant){
2713         switch(s->codec_id){
2714         case CODEC_ID_MPEG4:
2715             if (CONFIG_MPEG4_ENCODER)
2716                 ff_clean_mpeg4_qscales(s);
2717             break;
2718         case CODEC_ID_H263:
2719         case CODEC_ID_H263P:
2720         case CODEC_ID_FLV1:
2721             if (CONFIG_H263_ENCODER)
2722                 ff_clean_h263_qscales(s);
2723             break;
2724         default:
2725             ff_init_qscale_tab(s);
2726         }
2727
2728         s->lambda= s->lambda_table[0];
2729         //FIXME broken
2730     }else
2731         s->lambda= s->current_picture.quality;
2732 //printf("%d %d\n", s->avctx->global_quality, s->current_picture.quality);
2733     update_qscale(s);
2734     return 0;
2735 }
2736
2737 /* must be called before writing the header */
2738 static void set_frame_distances(MpegEncContext * s){
2739     assert(s->current_picture_ptr->pts != AV_NOPTS_VALUE);
2740     s->time= s->current_picture_ptr->pts*s->avctx->time_base.num;
2741
2742     if(s->pict_type==FF_B_TYPE){
2743         s->pb_time= s->pp_time - (s->last_non_b_time - s->time);
2744         assert(s->pb_time > 0 && s->pb_time < s->pp_time);
2745     }else{
2746         s->pp_time= s->time - s->last_non_b_time;
2747         s->last_non_b_time= s->time;
2748         assert(s->picture_number==0 || s->pp_time > 0);
2749     }
2750 }
2751
2752 static int encode_picture(MpegEncContext *s, int picture_number)
2753 {
2754     int i;
2755     int bits;
2756
2757     s->picture_number = picture_number;
2758
2759     /* Reset the average MB variance */
2760     s->me.mb_var_sum_temp    =
2761     s->me.mc_mb_var_sum_temp = 0;
2762
2763     /* we need to initialize some time vars before we can encode b-frames */
2764     // RAL: Condition added for MPEG1VIDEO
2765     if (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO || (s->h263_pred && !s->h263_msmpeg4))
2766         set_frame_distances(s);
2767     if(CONFIG_MPEG4_ENCODER && s->codec_id == CODEC_ID_MPEG4)
2768         ff_set_mpeg4_time(s);
2769
2770     s->me.scene_change_score=0;
2771
2772 //    s->lambda= s->current_picture_ptr->quality; //FIXME qscale / ... stuff for ME rate distortion
2773
2774     if(s->pict_type==FF_I_TYPE){
2775         if(s->msmpeg4_version >= 3) s->no_rounding=1;
2776         else                        s->no_rounding=0;
2777     }else if(s->pict_type!=FF_B_TYPE){
2778         if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
2779             s->no_rounding ^= 1;
2780     }
2781
2782     if(s->flags & CODEC_FLAG_PASS2){
2783         if (estimate_qp(s,1) < 0)
2784             return -1;
2785         ff_get_2pass_fcode(s);
2786     }else if(!(s->flags & CODEC_FLAG_QSCALE)){
2787         if(s->pict_type==FF_B_TYPE)
2788             s->lambda= s->last_lambda_for[s->pict_type];
2789         else
2790             s->lambda= s->last_lambda_for[s->last_non_b_pict_type];
2791         update_qscale(s);
2792     }
2793
2794     s->mb_intra=0; //for the rate distortion & bit compare functions
2795     for(i=1; i<s->avctx->thread_count; i++){
2796         ff_update_duplicate_context(s->thread_context[i], s);
2797     }
2798
2799     if(ff_init_me(s)<0)
2800         return -1;
2801
2802     /* Estimate motion for every MB */
2803     if(s->pict_type != FF_I_TYPE){
2804         s->lambda = (s->lambda * s->avctx->me_penalty_compensation + 128)>>8;
2805         s->lambda2= (s->lambda2* (int64_t)s->avctx->me_penalty_compensation + 128)>>8;
2806         if(s->pict_type != FF_B_TYPE && s->avctx->me_threshold==0){
2807             if((s->avctx->pre_me && s->last_non_b_pict_type==FF_I_TYPE) || s->avctx->pre_me==2){
2808                 s->avctx->execute(s->avctx, pre_estimate_motion_thread, &s->thread_context[0], NULL, s->avctx->thread_count, sizeof(void*));
2809             }
2810         }
2811
2812         s->avctx->execute(s->avctx, estimate_motion_thread, &s->thread_context[0], NULL, s->avctx->thread_count, sizeof(void*));
2813     }else /* if(s->pict_type == FF_I_TYPE) */{
2814         /* I-Frame */
2815         for(i=0; i<s->mb_stride*s->mb_height; i++)
2816             s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
2817
2818         if(!s->fixed_qscale){
2819             /* finding spatial complexity for I-frame rate control */
2820             s->avctx->execute(s->avctx, mb_var_thread, &s->thread_context[0], NULL, s->avctx->thread_count, sizeof(void*));
2821         }
2822     }
2823     for(i=1; i<s->avctx->thread_count; i++){
2824         merge_context_after_me(s, s->thread_context[i]);
2825     }
2826     s->current_picture.mc_mb_var_sum= s->current_picture_ptr->mc_mb_var_sum= s->me.mc_mb_var_sum_temp;
2827     s->current_picture.   mb_var_sum= s->current_picture_ptr->   mb_var_sum= s->me.   mb_var_sum_temp;
2828     emms_c();
2829
2830     if(s->me.scene_change_score > s->avctx->scenechange_threshold && s->pict_type == FF_P_TYPE){
2831         s->pict_type= FF_I_TYPE;
2832         for(i=0; i<s->mb_stride*s->mb_height; i++)
2833             s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
2834 //printf("Scene change detected, encoding as I Frame %d %d\n", s->current_picture.mb_var_sum, s->current_picture.mc_mb_var_sum);
2835     }
2836
2837     if(!s->umvplus){
2838         if(s->pict_type==FF_P_TYPE || s->pict_type==FF_S_TYPE) {
2839             s->f_code= ff_get_best_fcode(s, s->p_mv_table, CANDIDATE_MB_TYPE_INTER);
2840
2841             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2842                 int a,b;
2843                 a= ff_get_best_fcode(s, s->p_field_mv_table[0][0], CANDIDATE_MB_TYPE_INTER_I); //FIXME field_select
2844                 b= ff_get_best_fcode(s, s->p_field_mv_table[1][1], CANDIDATE_MB_TYPE_INTER_I);
2845                 s->f_code= FFMAX3(s->f_code, a, b);
2846             }
2847
2848             ff_fix_long_p_mvs(s);
2849             ff_fix_long_mvs(s, NULL, 0, s->p_mv_table, s->f_code, CANDIDATE_MB_TYPE_INTER, 0);
2850             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2851                 int j;
2852                 for(i=0; i<2; i++){
2853                     for(j=0; j<2; j++)
2854                         ff_fix_long_mvs(s, s->p_field_select_table[i], j,
2855                                         s->p_field_mv_table[i][j], s->f_code, CANDIDATE_MB_TYPE_INTER_I, 0);
2856                 }
2857             }
2858         }
2859
2860         if(s->pict_type==FF_B_TYPE){
2861             int a, b;
2862
2863             a = ff_get_best_fcode(s, s->b_forw_mv_table, CANDIDATE_MB_TYPE_FORWARD);
2864             b = ff_get_best_fcode(s, s->b_bidir_forw_mv_table, CANDIDATE_MB_TYPE_BIDIR);
2865             s->f_code = FFMAX(a, b);
2866
2867             a = ff_get_best_fcode(s, s->b_back_mv_table, CANDIDATE_MB_TYPE_BACKWARD);
2868             b = ff_get_best_fcode(s, s->b_bidir_back_mv_table, CANDIDATE_MB_TYPE_BIDIR);
2869             s->b_code = FFMAX(a, b);
2870
2871             ff_fix_long_mvs(s, NULL, 0, s->b_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_FORWARD, 1);
2872             ff_fix_long_mvs(s, NULL, 0, s->b_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BACKWARD, 1);
2873             ff_fix_long_mvs(s, NULL, 0, s->b_bidir_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_BIDIR, 1);
2874             ff_fix_long_mvs(s, NULL, 0, s->b_bidir_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BIDIR, 1);
2875             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2876                 int dir, j;
2877                 for(dir=0; dir<2; dir++){
2878                     for(i=0; i<2; i++){
2879                         for(j=0; j<2; j++){
2880                             int type= dir ? (CANDIDATE_MB_TYPE_BACKWARD_I|CANDIDATE_MB_TYPE_BIDIR_I)
2881                                           : (CANDIDATE_MB_TYPE_FORWARD_I |CANDIDATE_MB_TYPE_BIDIR_I);
2882                             ff_fix_long_mvs(s, s->b_field_select_table[dir][i], j,
2883                                             s->b_field_mv_table[dir][i][j], dir ? s->b_code : s->f_code, type, 1);
2884                         }
2885                     }
2886                 }
2887             }
2888         }
2889     }
2890
2891     if (estimate_qp(s, 0) < 0)
2892         return -1;
2893
2894     if(s->qscale < 3 && s->max_qcoeff<=128 && s->pict_type==FF_I_TYPE && !(s->flags & CODEC_FLAG_QSCALE))
2895         s->qscale= 3; //reduce clipping problems
2896
2897     if (s->out_format == FMT_MJPEG) {
2898         /* for mjpeg, we do include qscale in the matrix */
2899         s->intra_matrix[0] = ff_mpeg1_default_intra_matrix[0];
2900         for(i=1;i<64;i++){
2901             int j= s->dsp.idct_permutation[i];
2902
2903             s->intra_matrix[j] = av_clip_uint8((ff_mpeg1_default_intra_matrix[i] * s->qscale) >> 3);
2904         }
2905         ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
2906                        s->intra_matrix, s->intra_quant_bias, 8, 8, 1);
2907         s->qscale= 8;
2908     }
2909
2910     //FIXME var duplication
2911     s->current_picture_ptr->key_frame=
2912     s->current_picture.key_frame= s->pict_type == FF_I_TYPE; //FIXME pic_ptr
2913     s->current_picture_ptr->pict_type=
2914     s->current_picture.pict_type= s->pict_type;
2915
2916     if(s->current_picture.key_frame)
2917         s->picture_in_gop_number=0;
2918
2919     s->last_bits= put_bits_count(&s->pb);
2920     switch(s->out_format) {
2921     case FMT_MJPEG:
2922         if (CONFIG_MJPEG_ENCODER)
2923             ff_mjpeg_encode_picture_header(s);
2924         break;
2925     case FMT_H261:
2926         if (CONFIG_H261_ENCODER)
2927             ff_h261_encode_picture_header(s, picture_number);
2928         break;
2929     case FMT_H263:
2930         if (CONFIG_WMV2_ENCODER && s->codec_id == CODEC_ID_WMV2)
2931             ff_wmv2_encode_picture_header(s, picture_number);
2932         else if (CONFIG_MSMPEG4_ENCODER && s->h263_msmpeg4)
2933             msmpeg4_encode_picture_header(s, picture_number);
2934         else if (CONFIG_MPEG4_ENCODER && s->h263_pred)
2935             mpeg4_encode_picture_header(s, picture_number);
2936         else if (CONFIG_RV10_ENCODER && s->codec_id == CODEC_ID_RV10)
2937             rv10_encode_picture_header(s, picture_number);
2938         else if (CONFIG_RV20_ENCODER && s->codec_id == CODEC_ID_RV20)
2939             rv20_encode_picture_header(s, picture_number);
2940         else if (CONFIG_FLV_ENCODER && s->codec_id == CODEC_ID_FLV1)
2941             ff_flv_encode_picture_header(s, picture_number);
2942         else if (CONFIG_H263_ENCODER)
2943             h263_encode_picture_header(s, picture_number);
2944         break;
2945     case FMT_MPEG1:
2946         if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
2947             mpeg1_encode_picture_header(s, picture_number);
2948         break;
2949     case FMT_H264:
2950         break;
2951     default:
2952         assert(0);
2953     }
2954     bits= put_bits_count(&s->pb);
2955     s->header_bits= bits - s->last_bits;
2956
2957     for(i=1; i<s->avctx->thread_count; i++){
2958         update_duplicate_context_after_me(s->thread_context[i], s);
2959     }
2960     s->avctx->execute(s->avctx, encode_thread, &s->thread_context[0], NULL, s->avctx->thread_count, sizeof(void*));
2961     for(i=1; i<s->avctx->thread_count; i++){
2962         merge_context_after_encode(s, s->thread_context[i]);
2963     }
2964     emms_c();
2965     return 0;
2966 }
2967
2968 void  denoise_dct_c(MpegEncContext *s, DCTELEM *block){
2969     const int intra= s->mb_intra;
2970     int i;
2971
2972     s->dct_count[intra]++;
2973
2974     for(i=0; i<64; i++){
2975         int level= block[i];
2976
2977         if(level){
2978             if(level>0){
2979                 s->dct_error_sum[intra][i] += level;
2980                 level -= s->dct_offset[intra][i];
2981                 if(level<0) level=0;
2982             }else{
2983                 s->dct_error_sum[intra][i] -= level;
2984                 level += s->dct_offset[intra][i];
2985                 if(level>0) level=0;
2986             }
2987             block[i]= level;
2988         }
2989     }
2990 }
2991
2992 int dct_quantize_trellis_c(MpegEncContext *s,
2993                         DCTELEM *block, int n,
2994                         int qscale, int *overflow){
2995     const int *qmat;
2996     const uint8_t *scantable= s->intra_scantable.scantable;
2997     const uint8_t *perm_scantable= s->intra_scantable.permutated;
2998     int max=0;
2999     unsigned int threshold1, threshold2;
3000     int bias=0;
3001     int run_tab[65];
3002     int level_tab[65];
3003     int score_tab[65];
3004     int survivor[65];
3005     int survivor_count;
3006     int last_run=0;
3007     int last_level=0;
3008     int last_score= 0;
3009     int last_i;
3010     int coeff[2][64];
3011     int coeff_count[64];
3012     int qmul, qadd, start_i, last_non_zero, i, dc;
3013     const int esc_length= s->ac_esc_length;
3014     uint8_t * length;
3015     uint8_t * last_length;
3016     const int lambda= s->lambda2 >> (FF_LAMBDA_SHIFT - 6);
3017
3018     s->dsp.fdct (block);
3019
3020     if(s->dct_error_sum)
3021         s->denoise_dct(s, block);
3022     qmul= qscale*16;
3023     qadd= ((qscale-1)|1)*8;
3024
3025     if (s->mb_intra) {
3026         int q;
3027         if (!s->h263_aic) {
3028             if (n < 4)
3029                 q = s->y_dc_scale;
3030             else
3031                 q = s->c_dc_scale;
3032             q = q << 3;
3033         } else{
3034             /* For AIC we skip quant/dequant of INTRADC */
3035             q = 1 << 3;
3036             qadd=0;
3037         }
3038
3039         /* note: block[0] is assumed to be positive */
3040         block[0] = (block[0] + (q >> 1)) / q;
3041         start_i = 1;
3042         last_non_zero = 0;
3043         qmat = s->q_intra_matrix[qscale];
3044         if(s->mpeg_quant || s->out_format == FMT_MPEG1)
3045             bias= 1<<(QMAT_SHIFT-1);
3046         length     = s->intra_ac_vlc_length;
3047         last_length= s->intra_ac_vlc_last_length;
3048     } else {
3049         start_i = 0;
3050         last_non_zero = -1;
3051         qmat = s->q_inter_matrix[qscale];
3052         length     = s->inter_ac_vlc_length;
3053         last_length= s->inter_ac_vlc_last_length;
3054     }
3055     last_i= start_i;
3056
3057     threshold1= (1<<QMAT_SHIFT) - bias - 1;
3058     threshold2= (threshold1<<1);
3059
3060     for(i=63; i>=start_i; i--) {
3061         const int j = scantable[i];
3062         int level = block[j] * qmat[j];
3063
3064         if(((unsigned)(level+threshold1))>threshold2){
3065             last_non_zero = i;
3066             break;
3067         }
3068     }
3069
3070     for(i=start_i; i<=last_non_zero; i++) {
3071         const int j = scantable[i];
3072         int level = block[j] * qmat[j];
3073
3074 //        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
3075 //           || bias-level >= (1<<(QMAT_SHIFT - 3))){
3076         if(((unsigned)(level+threshold1))>threshold2){
3077             if(level>0){
3078                 level= (bias + level)>>QMAT_SHIFT;
3079                 coeff[0][i]= level;
3080                 coeff[1][i]= level-1;
3081 //                coeff[2][k]= level-2;
3082             }else{
3083                 level= (bias - level)>>QMAT_SHIFT;
3084                 coeff[0][i]= -level;
3085                 coeff[1][i]= -level+1;
3086 //                coeff[2][k]= -level+2;
3087             }
3088             coeff_count[i]= FFMIN(level, 2);
3089             assert(coeff_count[i]);
3090             max |=level;
3091         }else{
3092             coeff[0][i]= (level>>31)|1;
3093             coeff_count[i]= 1;
3094         }
3095     }
3096
3097     *overflow= s->max_qcoeff < max; //overflow might have happened
3098
3099     if(last_non_zero < start_i){
3100         memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
3101         return last_non_zero;
3102     }
3103
3104     score_tab[start_i]= 0;
3105     survivor[0]= start_i;
3106     survivor_count= 1;
3107
3108     for(i=start_i; i<=last_non_zero; i++){
3109         int level_index, j, zero_distortion;
3110         int dct_coeff= FFABS(block[ scantable[i] ]);
3111         int best_score=256*256*256*120;
3112
3113         if (   s->dsp.fdct == fdct_ifast
3114 #ifndef FAAN_POSTSCALE
3115             || s->dsp.fdct == ff_faandct
3116 #endif
3117            )
3118             dct_coeff= (dct_coeff*ff_inv_aanscales[ scantable[i] ]) >> 12;
3119         zero_distortion= dct_coeff*dct_coeff;
3120
3121         for(level_index=0; level_index < coeff_count[i]; level_index++){
3122             int distortion;
3123             int level= coeff[level_index][i];
3124             const int alevel= FFABS(level);
3125             int unquant_coeff;
3126
3127             assert(level);
3128
3129             if(s->out_format == FMT_H263){
3130                 unquant_coeff= alevel*qmul + qadd;
3131             }else{ //MPEG1
3132                 j= s->dsp.idct_permutation[ scantable[i] ]; //FIXME optimize
3133                 if(s->mb_intra){
3134                         unquant_coeff = (int)(  alevel  * qscale * s->intra_matrix[j]) >> 3;
3135                         unquant_coeff =   (unquant_coeff - 1) | 1;
3136                 }else{
3137                         unquant_coeff = (((  alevel  << 1) + 1) * qscale * ((int) s->inter_matrix[j])) >> 4;
3138                         unquant_coeff =   (unquant_coeff - 1) | 1;
3139                 }
3140                 unquant_coeff<<= 3;
3141             }
3142
3143             distortion= (unquant_coeff - dct_coeff) * (unquant_coeff - dct_coeff) - zero_distortion;
3144             level+=64;
3145             if((level&(~127)) == 0){
3146                 for(j=survivor_count-1; j>=0; j--){
3147                     int run= i - survivor[j];
3148                     int score= distortion + length[UNI_AC_ENC_INDEX(run, level)]*lambda;
3149                     score += score_tab[i-run];
3150
3151                     if(score < best_score){
3152                         best_score= score;
3153                         run_tab[i+1]= run;
3154                         level_tab[i+1]= level-64;
3155                     }
3156                 }
3157
3158                 if(s->out_format == FMT_H263){
3159                     for(j=survivor_count-1; j>=0; j--){
3160                         int run= i - survivor[j];
3161                         int score= distortion + last_length[UNI_AC_ENC_INDEX(run, level)]*lambda;
3162                         score += score_tab[i-run];
3163                         if(score < last_score){
3164                             last_score= score;
3165                             last_run= run;
3166                             last_level= level-64;
3167                             last_i= i+1;
3168                         }
3169                     }
3170                 }
3171             }else{
3172                 distortion += esc_length*lambda;
3173                 for(j=survivor_count-1; j>=0; j--){
3174                     int run= i - survivor[j];
3175                     int score= distortion + score_tab[i-run];
3176
3177                     if(score < best_score){
3178                         best_score= score;
3179                         run_tab[i+1]= run;
3180                         level_tab[i+1]= level-64;
3181                     }
3182                 }
3183
3184                 if(s->out_format == FMT_H263){
3185                   for(j=survivor_count-1; j>=0; j--){
3186                         int run= i - survivor[j];
3187                         int score= distortion + score_tab[i-run];
3188                         if(score < last_score){
3189                             last_score= score;
3190                             last_run= run;
3191                             last_level= level-64;
3192                             last_i= i+1;
3193                         }
3194                     }
3195                 }
3196             }
3197         }
3198
3199         score_tab[i+1]= best_score;
3200
3201         //Note: there is a vlc code in mpeg4 which is 1 bit shorter then another one with a shorter run and the same level
3202         if(last_non_zero <= 27){
3203             for(; survivor_count; survivor_count--){
3204                 if(score_tab[ survivor[survivor_count-1] ] <= best_score)
3205                     break;
3206             }
3207         }else{
3208             for(; survivor_count; survivor_count--){
3209                 if(score_tab[ survivor[survivor_count-1] ] <= best_score + lambda)
3210                     break;
3211             }
3212         }
3213
3214         survivor[ survivor_count++ ]= i+1;
3215     }
3216
3217     if(s->out_format != FMT_H263){
3218         last_score= 256*256*256*120;
3219         for(i= survivor[0]; i<=last_non_zero + 1; i++){
3220             int score= score_tab[i];
3221             if(i) score += lambda*2; //FIXME exacter?
3222
3223             if(score < last_score){
3224                 last_score= score;
3225                 last_i= i;
3226                 last_level= level_tab[i];
3227                 last_run= run_tab[i];
3228             }
3229         }
3230     }
3231
3232     s->coded_score[n] = last_score;
3233
3234     dc= FFABS(block[0]);
3235     last_non_zero= last_i - 1;
3236     memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
3237
3238     if(last_non_zero < start_i)
3239         return last_non_zero;
3240
3241     if(last_non_zero == 0 && start_i == 0){
3242         int best_level= 0;
3243         int best_score= dc * dc;
3244
3245         for(i=0; i<coeff_count[0]; i++){
3246             int level= coeff[i][0];
3247             int alevel= FFABS(level);
3248             int unquant_coeff, score, distortion;
3249
3250             if(s->out_format == FMT_H263){
3251                     unquant_coeff= (alevel*qmul + qadd)>>3;
3252             }else{ //MPEG1
3253                     unquant_coeff = (((  alevel  << 1) + 1) * qscale * ((int) s->inter_matrix[0])) >> 4;
3254                     unquant_coeff =   (unquant_coeff - 1) | 1;
3255             }
3256             unquant_coeff = (unquant_coeff + 4) >> 3;
3257             unquant_coeff<<= 3 + 3;
3258
3259             distortion= (unquant_coeff - dc) * (unquant_coeff - dc);
3260             level+=64;
3261             if((level&(~127)) == 0) score= distortion + last_length[UNI_AC_ENC_INDEX(0, level)]*lambda;
3262             else                    score= distortion + esc_length*lambda;
3263
3264             if(score < best_score){
3265                 best_score= score;
3266                 best_level= level - 64;
3267             }
3268         }
3269         block[0]= best_level;
3270         s->coded_score[n] = best_score - dc*dc;
3271         if(best_level == 0) return -1;
3272         else                return last_non_zero;
3273     }
3274
3275     i= last_i;
3276     assert(last_level);
3277
3278     block[ perm_scantable[last_non_zero] ]= last_level;
3279     i -= last_run + 1;
3280
3281     for(; i>start_i; i -= run_tab[i] + 1){
3282         block[ perm_scantable[i-1] ]= level_tab[i];
3283     }
3284
3285     return last_non_zero;
3286 }
3287
3288 //#define REFINE_STATS 1
3289 static int16_t basis[64][64];
3290
3291 static void build_basis(uint8_t *perm){
3292     int i, j, x, y;
3293     emms_c();
3294     for(i=0; i<8; i++){
3295         for(j=0; j<8; j++){
3296             for(y=0; y<8; y++){
3297                 for(x=0; x<8; x++){
3298                     double s= 0.25*(1<<BASIS_SHIFT);
3299                     int index= 8*i + j;
3300                     int perm_index= perm[index];
3301                     if(i==0) s*= sqrt(0.5);
3302                     if(j==0) s*= sqrt(0.5);
3303                     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)));
3304                 }
3305             }
3306         }
3307     }
3308 }
3309
3310 static int dct_quantize_refine(MpegEncContext *s, //FIXME breaks denoise?
3311                         DCTELEM *block, int16_t *weight, DCTELEM *orig,
3312                         int n, int qscale){
3313     int16_t rem[64];
3314     DECLARE_ALIGNED_16(DCTELEM, d1)[64];
3315     const uint8_t *scantable= s->intra_scantable.scantable;
3316     const uint8_t *perm_scantable= s->intra_scantable.permutated;
3317 //    unsigned int threshold1, threshold2;
3318 //    int bias=0;
3319     int run_tab[65];
3320     int prev_run=0;
3321     int prev_level=0;
3322     int qmul, qadd, start_i, last_non_zero, i, dc;
3323     uint8_t * length;
3324     uint8_t * last_length;
3325     int lambda;
3326     int rle_index, run, q = 1, sum; //q is only used when s->mb_intra is true
3327 #ifdef REFINE_STATS
3328 static int count=0;
3329 static int after_last=0;
3330 static int to_zero=0;
3331 static int from_zero=0;
3332 static int raise=0;
3333 static int lower=0;
3334 static int messed_sign=0;
3335 #endif
3336
3337     if(basis[0][0] == 0)
3338         build_basis(s->dsp.idct_permutation);
3339
3340     qmul= qscale*2;
3341     qadd= (qscale-1)|1;
3342     if (s->mb_intra) {
3343         if (!s->h263_aic) {
3344             if (n < 4)
3345                 q = s->y_dc_scale;
3346             else
3347                 q = s->c_dc_scale;
3348         } else{
3349             /* For AIC we skip quant/dequant of INTRADC */
3350             q = 1;
3351             qadd=0;
3352         }
3353         q <<= RECON_SHIFT-3;
3354         /* note: block[0] is assumed to be positive */
3355         dc= block[0]*q;
3356 //        block[0] = (block[0] + (q >> 1)) / q;
3357         start_i = 1;
3358 //        if(s->mpeg_quant || s->out_format == FMT_MPEG1)
3359 //            bias= 1<<(QMAT_SHIFT-1);
3360         length     = s->intra_ac_vlc_length;
3361         last_length= s->intra_ac_vlc_last_length;
3362     } else {
3363         dc= 0;
3364         start_i = 0;
3365         length     = s->inter_ac_vlc_length;
3366         last_length= s->inter_ac_vlc_last_length;
3367     }
3368     last_non_zero = s->block_last_index[n];
3369
3370 #ifdef REFINE_STATS
3371 {START_TIMER
3372 #endif
3373     dc += (1<<(RECON_SHIFT-1));
3374     for(i=0; i<64; i++){
3375         rem[i]= dc - (orig[i]<<RECON_SHIFT); //FIXME  use orig dirrectly instead of copying to rem[]
3376     }
3377 #ifdef REFINE_STATS
3378 STOP_TIMER("memset rem[]")}
3379 #endif
3380     sum=0;
3381     for(i=0; i<64; i++){
3382         int one= 36;
3383         int qns=4;
3384         int w;
3385
3386         w= FFABS(weight[i]) + qns*one;
3387         w= 15 + (48*qns*one + w/2)/w; // 16 .. 63
3388
3389         weight[i] = w;
3390 //        w=weight[i] = (63*qns + (w/2)) / w;
3391
3392         assert(w>0);
3393         assert(w<(1<<6));
3394         sum += w*w;
3395     }
3396     lambda= sum*(uint64_t)s->lambda2 >> (FF_LAMBDA_SHIFT - 6 + 6 + 6 + 6);
3397 #ifdef REFINE_STATS
3398 {START_TIMER
3399 #endif
3400     run=0;
3401     rle_index=0;
3402     for(i=start_i; i<=last_non_zero; i++){
3403         int j= perm_scantable[i];
3404         const int level= block[j];
3405         int coeff;
3406
3407         if(level){
3408             if(level<0) coeff= qmul*level - qadd;
3409             else        coeff= qmul*level + qadd;
3410             run_tab[rle_index++]=run;
3411             run=0;
3412
3413             s->dsp.add_8x8basis(rem, basis[j], coeff);
3414         }else{
3415             run++;
3416         }
3417     }
3418 #ifdef REFINE_STATS
3419 if(last_non_zero>0){
3420 STOP_TIMER("init rem[]")
3421 }
3422 }
3423
3424 {START_TIMER
3425 #endif
3426     for(;;){
3427         int best_score=s->dsp.try_8x8basis(rem, weight, basis[0], 0);
3428         int best_coeff=0;
3429         int best_change=0;
3430         int run2, best_unquant_change=0, analyze_gradient;
3431 #ifdef REFINE_STATS
3432 {START_TIMER
3433 #endif
3434         analyze_gradient = last_non_zero > 2 || s->avctx->quantizer_noise_shaping >= 3;
3435
3436         if(analyze_gradient){
3437 #ifdef REFINE_STATS
3438 {START_TIMER
3439 #endif
3440             for(i=0; i<64; i++){
3441                 int w= weight[i];
3442
3443                 d1[i] = (rem[i]*w*w + (1<<(RECON_SHIFT+12-1)))>>(RECON_SHIFT+12);
3444             }
3445 #ifdef REFINE_STATS
3446 STOP_TIMER("rem*w*w")}
3447 {START_TIMER
3448 #endif
3449             s->dsp.fdct(d1);
3450 #ifdef REFINE_STATS
3451 STOP_TIMER("dct")}
3452 #endif
3453         }
3454
3455         if(start_i){
3456             const int level= block[0];
3457             int change, old_coeff;
3458
3459             assert(s->mb_intra);
3460
3461             old_coeff= q*level;
3462
3463             for(change=-1; change<=1; change+=2){
3464                 int new_level= level + change;
3465                 int score, new_coeff;
3466
3467                 new_coeff= q*new_level;
3468                 if(new_coeff >= 2048 || new_coeff < 0)
3469                     continue;
3470
3471                 score= s->dsp.try_8x8basis(rem, weight, basis[0], new_coeff - old_coeff);
3472                 if(score<best_score){
3473                     best_score= score;
3474                     best_coeff= 0;
3475                     best_change= change;
3476                     best_unquant_change= new_coeff - old_coeff;
3477                 }
3478             }
3479         }
3480
3481         run=0;
3482         rle_index=0;
3483         run2= run_tab[rle_index++];
3484         prev_level=0;
3485         prev_run=0;
3486
3487         for(i=start_i; i<64; i++){
3488             int j= perm_scantable[i];
3489             const int level= block[j];
3490             int change, old_coeff;
3491
3492             if(s->avctx->quantizer_noise_shaping < 3 && i > last_non_zero + 1)
3493                 break;
3494
3495             if(level){
3496                 if(level<0) old_coeff= qmul*level - qadd;
3497                 else        old_coeff= qmul*level + qadd;
3498                 run2= run_tab[rle_index++]; //FIXME ! maybe after last
3499             }else{
3500                 old_coeff=0;
3501                 run2--;
3502                 assert(run2>=0 || i >= last_non_zero );
3503             }
3504
3505             for(change=-1; change<=1; change+=2){
3506                 int new_level= level + change;
3507                 int score, new_coeff, unquant_change;
3508
3509                 score=0;
3510                 if(s->avctx->quantizer_noise_shaping < 2 && FFABS(new_level) > FFABS(level))
3511                    continue;
3512
3513                 if(new_level){
3514                     if(new_level<0) new_coeff= qmul*new_level - qadd;
3515                     else            new_coeff= qmul*new_level + qadd;
3516                     if(new_coeff >= 2048 || new_coeff <= -2048)
3517                         continue;
3518                     //FIXME check for overflow
3519
3520                     if(level){
3521                         if(level < 63 && level > -63){
3522                             if(i < last_non_zero)
3523                                 score +=   length[UNI_AC_ENC_INDEX(run, new_level+64)]
3524                                          - length[UNI_AC_ENC_INDEX(run, level+64)];
3525                             else
3526                                 score +=   last_length[UNI_AC_ENC_INDEX(run, new_level+64)]
3527                                          - last_length[UNI_AC_ENC_INDEX(run, level+64)];
3528                         }
3529                     }else{
3530                         assert(FFABS(new_level)==1);
3531
3532                         if(analyze_gradient){
3533                             int g= d1[ scantable[i] ];
3534                             if(g && (g^new_level) >= 0)
3535                                 continue;
3536                         }
3537
3538                         if(i < last_non_zero){
3539                             int next_i= i + run2 + 1;
3540                             int next_level= block[ perm_scantable[next_i] ] + 64;
3541
3542                             if(next_level&(~127))
3543                                 next_level= 0;
3544
3545                             if(next_i < last_non_zero)
3546                                 score +=   length[UNI_AC_ENC_INDEX(run, 65)]
3547                                          + length[UNI_AC_ENC_INDEX(run2, next_level)]
3548                                          - length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)];
3549                             else
3550                                 score +=  length[UNI_AC_ENC_INDEX(run, 65)]
3551                                         + last_length[UNI_AC_ENC_INDEX(run2, next_level)]
3552                                         - last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)];
3553                         }else{
3554                             score += last_length[UNI_AC_ENC_INDEX(run, 65)];
3555                             if(prev_level){
3556                                 score +=  length[UNI_AC_ENC_INDEX(prev_run, prev_level)]
3557                                         - last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)];
3558                             }
3559                         }
3560                     }
3561                 }else{
3562                     new_coeff=0;
3563                     assert(FFABS(level)==1);
3564
3565                     if(i < last_non_zero){
3566                         int next_i= i + run2 + 1;
3567                         int next_level= block[ perm_scantable[next_i] ] + 64;
3568
3569                         if(next_level&(~127))
3570                             next_level= 0;
3571
3572                         if(next_i < last_non_zero)
3573                             score +=   length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)]
3574                                      - length[UNI_AC_ENC_INDEX(run2, next_level)]
3575                                      - length[UNI_AC_ENC_INDEX(run, 65)];
3576                         else
3577                             score +=   last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)]
3578                                      - last_length[UNI_AC_ENC_INDEX(run2, next_level)]
3579                                      - length[UNI_AC_ENC_INDEX(run, 65)];
3580                     }else{
3581                         score += -last_length[UNI_AC_ENC_INDEX(run, 65)];
3582                         if(prev_level){
3583                             score +=  last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)]
3584                                     - length[UNI_AC_ENC_INDEX(prev_run, prev_level)];
3585                         }
3586                     }
3587                 }
3588
3589                 score *= lambda;
3590
3591                 unquant_change= new_coeff - old_coeff;
3592                 assert((score < 100*lambda && score > -100*lambda) || lambda==0);
3593
3594                 score+= s->dsp.try_8x8basis(rem, weight, basis[j], unquant_change);
3595                 if(score<best_score){
3596                     best_score= score;
3597                     best_coeff= i;
3598                     best_change= change;
3599                     best_unquant_change= unquant_change;
3600                 }
3601             }
3602             if(level){
3603                 prev_level= level + 64;
3604                 if(prev_level&(~127))
3605                     prev_level= 0;
3606                 prev_run= run;
3607                 run=0;
3608             }else{
3609                 run++;
3610             }
3611         }
3612 #ifdef REFINE_STATS
3613 STOP_TIMER("iterative step")}
3614 #endif
3615
3616         if(best_change){
3617             int j= perm_scantable[ best_coeff ];
3618
3619             block[j] += best_change;
3620
3621             if(best_coeff > last_non_zero){
3622                 last_non_zero= best_coeff;
3623                 assert(block[j]);
3624 #ifdef REFINE_STATS
3625 after_last++;
3626 #endif
3627             }else{
3628 #ifdef REFINE_STATS
3629 if(block[j]){
3630     if(block[j] - best_change){
3631         if(FFABS(block[j]) > FFABS(block[j] - best_change)){
3632             raise++;
3633         }else{
3634             lower++;
3635         }
3636     }else{
3637         from_zero++;
3638     }
3639 }else{
3640     to_zero++;
3641 }
3642 #endif
3643                 for(; last_non_zero>=start_i; last_non_zero--){
3644                     if(block[perm_scantable[last_non_zero]])
3645                         break;
3646                 }
3647             }
3648 #ifdef REFINE_STATS
3649 count++;
3650 if(256*256*256*64 % count == 0){
3651     printf("after_last:%d to_zero:%d from_zero:%d raise:%d lower:%d sign:%d xyp:%d/%d/%d\n", after_last, to_zero, from_zero, raise, lower, messed_sign, s->mb_x, s->mb_y, s->picture_number);
3652 }
3653 #endif
3654             run=0;
3655             rle_index=0;
3656             for(i=start_i; i<=last_non_zero; i++){
3657                 int j= perm_scantable[i];
3658                 const int level= block[j];
3659
3660                  if(level){
3661                      run_tab[rle_index++]=run;
3662                      run=0;
3663                  }else{
3664                      run++;
3665                  }
3666             }
3667
3668             s->dsp.add_8x8basis(rem, basis[j], best_unquant_change);
3669         }else{
3670             break;
3671         }
3672     }
3673 #ifdef REFINE_STATS
3674 if(last_non_zero>0){
3675 STOP_TIMER("iterative search")
3676 }
3677 }
3678 #endif
3679
3680     return last_non_zero;
3681 }
3682
3683 int dct_quantize_c(MpegEncContext *s,
3684                         DCTELEM *block, int n,
3685                         int qscale, int *overflow)
3686 {
3687     int i, j, level, last_non_zero, q, start_i;
3688     const int *qmat;
3689     const uint8_t *scantable= s->intra_scantable.scantable;
3690     int bias;
3691     int max=0;
3692     unsigned int threshold1, threshold2;
3693
3694     s->dsp.fdct (block);
3695
3696     if(s->dct_error_sum)
3697         s->denoise_dct(s, block);
3698
3699     if (s->mb_intra) {
3700         if (!s->h263_aic) {
3701             if (n < 4)
3702                 q = s->y_dc_scale;
3703             else
3704                 q = s->c_dc_scale;
3705             q = q << 3;
3706         } else
3707             /* For AIC we skip quant/dequant of INTRADC */
3708             q = 1 << 3;
3709
3710         /* note: block[0] is assumed to be positive */
3711         block[0] = (block[0] + (q >> 1)) / q;
3712         start_i = 1;
3713         last_non_zero = 0;
3714         qmat = s->q_intra_matrix[qscale];
3715         bias= s->intra_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT);
3716     } else {
3717         start_i = 0;
3718         last_non_zero = -1;
3719         qmat = s->q_inter_matrix[qscale];
3720         bias= s->inter_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT);
3721     }
3722     threshold1= (1<<QMAT_SHIFT) - bias - 1;
3723     threshold2= (threshold1<<1);
3724     for(i=63;i>=start_i;i--) {
3725         j = scantable[i];
3726         level = block[j] * qmat[j];
3727
3728         if(((unsigned)(level+threshold1))>threshold2){
3729             last_non_zero = i;
3730             break;
3731         }else{
3732             block[j]=0;
3733         }
3734     }
3735     for(i=start_i; i<=last_non_zero; i++) {
3736         j = scantable[i];
3737         level = block[j] * qmat[j];
3738
3739 //        if(   bias+level >= (1<<QMAT_SHIFT)
3740 //           || bias-level >= (1<<QMAT_SHIFT)){
3741         if(((unsigned)(level+threshold1))>threshold2){
3742             if(level>0){
3743                 level= (bias + level)>>QMAT_SHIFT;
3744                 block[j]= level;
3745             }else{
3746                 level= (bias - level)>>QMAT_SHIFT;
3747                 block[j]= -level;
3748             }
3749             max |=level;
3750         }else{
3751             block[j]=0;
3752         }
3753     }
3754     *overflow= s->max_qcoeff < max; //overflow might have happened
3755
3756     /* we need this permutation so that we correct the IDCT, we only permute the !=0 elements */
3757     if (s->dsp.idct_permutation_type != FF_NO_IDCT_PERM)
3758         ff_block_permute(block, s->dsp.idct_permutation, scantable, last_non_zero);
3759
3760     return last_non_zero;
3761 }
3762
3763 AVCodec h263_encoder = {
3764     "h263",
3765     CODEC_TYPE_VIDEO,
3766     CODEC_ID_H263,
3767     sizeof(MpegEncContext),
3768     MPV_encode_init,
3769     MPV_encode_picture,
3770     MPV_encode_end,
3771     .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3772     .long_name= NULL_IF_CONFIG_SMALL("H.263 / H.263-1996"),
3773 };
3774
3775 AVCodec h263p_encoder = {
3776     "h263p",
3777     CODEC_TYPE_VIDEO,
3778     CODEC_ID_H263P,
3779     sizeof(MpegEncContext),
3780     MPV_encode_init,
3781     MPV_encode_picture,
3782     MPV_encode_end,
3783     .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3784     .long_name= NULL_IF_CONFIG_SMALL("H.263+ / H.263-1998 / H.263 version 2"),
3785 };
3786
3787 AVCodec msmpeg4v1_encoder = {
3788     "msmpeg4v1",
3789     CODEC_TYPE_VIDEO,
3790     CODEC_ID_MSMPEG4V1,
3791     sizeof(MpegEncContext),
3792     MPV_encode_init,
3793     MPV_encode_picture,
3794     MPV_encode_end,
3795     .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3796     .long_name= NULL_IF_CONFIG_SMALL("MPEG-4 part 2 Microsoft variant version 1"),
3797 };
3798
3799 AVCodec msmpeg4v2_encoder = {
3800     "msmpeg4v2",
3801     CODEC_TYPE_VIDEO,
3802     CODEC_ID_MSMPEG4V2,
3803     sizeof(MpegEncContext),
3804     MPV_encode_init,
3805     MPV_encode_picture,
3806     MPV_encode_end,
3807     .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3808     .long_name= NULL_IF_CONFIG_SMALL("MPEG-4 part 2 Microsoft variant version 2"),
3809 };
3810
3811 AVCodec msmpeg4v3_encoder = {
3812     "msmpeg4",
3813     CODEC_TYPE_VIDEO,
3814     CODEC_ID_MSMPEG4V3,
3815     sizeof(MpegEncContext),
3816     MPV_encode_init,
3817     MPV_encode_picture,
3818     MPV_encode_end,
3819     .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3820     .long_name= NULL_IF_CONFIG_SMALL("MPEG-4 part 2 Microsoft variant version 3"),
3821 };
3822
3823 AVCodec wmv1_encoder = {
3824     "wmv1",
3825     CODEC_TYPE_VIDEO,
3826     CODEC_ID_WMV1,
3827     sizeof(MpegEncContext),
3828     MPV_encode_init,
3829     MPV_encode_picture,
3830     MPV_encode_end,
3831     .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3832     .long_name= NULL_IF_CONFIG_SMALL("Windows Media Video 7"),
3833 };