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 "libavutil/intmath.h"
  31 #include "avcodec.h"
  32 #include "dsputil.h"
  33 #include "mpegvideo.h"
  34 #include "mpegvideo_common.h"
  35 #include "h263.h"
  36 #include "mjpegenc.h"
  37 #include "msmpeg4.h"
  38 #include "faandct.h"
  39 #include "aandcttab.h"
  40 #include "flv.h"
  41 #include "mpeg4video.h"
  42 #include "internal.h"
  43 #include <limits.h>
  44
  45 //#undef NDEBUG
  46 //#include <assert.h>
  47
  48 static int encode_picture(MpegEncContext *s, int picture_number);
  49 static int dct_quantize_refine(MpegEncContext *s, DCTELEM *block, int16_t *weight, DCTELEM *orig, int n, int qscale);
  50 static int sse_mb(MpegEncContext *s);
  51
  52 /* enable all paranoid tests for rounding, overflows, etc... */
  53 //#define PARANOID
  54
  55 //#define DEBUG
  56
  57 static uint8_t default_mv_penalty[MAX_FCODE+1][MAX_MV*2+1];
  58 static uint8_t default_fcode_tab[MAX_MV*2+1];
  59
  60 void ff_convert_matrix(DSPContext *dsp, int (*qmat)[64], uint16_t (*qmat16)[2][64],
  61                            const uint16_t *quant_matrix, int bias, int qmin, int qmax, int intra)
  62 {
  63     int qscale;
  64     int shift=0;
  65
  66     for(qscale=qmin; qscale<=qmax; qscale++){
  67         int i;
  68         if (dsp->fdct == ff_jpeg_fdct_islow
  69 #ifdef FAAN_POSTSCALE
  70             || dsp->fdct == ff_faandct
  71 #endif
  72             ) {
  73             for(i=0;i<64;i++) {
  74                 const int j= dsp->idct_permutation[i];
  75                 /* 16 <= qscale * quant_matrix[i] <= 7905 */
  76                 /* 19952             <= ff_aanscales[i] * qscale * quant_matrix[i]               <= 249205026 */
  77                 /* (1 << 36) / 19952 >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= (1 << 36) / 249205026 */
  78                 /* 3444240           >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= 275 */
  79
  80                 qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) /
  81                                 (qscale * quant_matrix[j]));
  82             }
  83         } else if (dsp->fdct == fdct_ifast
  84 #ifndef FAAN_POSTSCALE
  85                    || dsp->fdct == ff_faandct
  86 #endif
  87                    ) {
  88             for(i=0;i<64;i++) {
  89                 const int j= dsp->idct_permutation[i];
  90                 /* 16 <= qscale * quant_matrix[i] <= 7905 */
  91                 /* 19952             <= ff_aanscales[i] * qscale * quant_matrix[i]               <= 249205026 */
  92                 /* (1 << 36) / 19952 >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */
  93                 /* 3444240           >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= 275 */
  94
  95                 qmat[qscale][i] = (int)((UINT64_C(1) << (QMAT_SHIFT + 14)) /
  96                                 (ff_aanscales[i] * qscale * quant_matrix[j]));
  97             }
  98         } else {
  99             for(i=0;i<64;i++) {
 100                 const int j= dsp->idct_permutation[i];
 101                 /* We can safely suppose that 16 <= quant_matrix[i] <= 255
 102                    So 16           <= qscale * quant_matrix[i]             <= 7905
 103                    so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905
 104                    so 32768        >= (1<<19) / (qscale * quant_matrix[i]) >= 67
 105                 */
 106                 qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) / (qscale * quant_matrix[j]));
 107 //                qmat  [qscale][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]);
 108                 qmat16[qscale][0][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[j]);
 109
 110                 if(qmat16[qscale][0][i]==0 || qmat16[qscale][0][i]==128*256) qmat16[qscale][0][i]=128*256-1;
 111                 qmat16[qscale][1][i]= ROUNDED_DIV(bias<<(16-QUANT_BIAS_SHIFT), qmat16[qscale][0][i]);
 112             }
 113         }
 114
 115         for(i=intra; i<64; i++){
 116             int64_t max= 8191;
 117             if (dsp->fdct == fdct_ifast
 118 #ifndef FAAN_POSTSCALE
 119                    || dsp->fdct == ff_faandct
 120 #endif
 121                    ) {
 122                 max = (8191LL*ff_aanscales[i]) >> 14;
 123             }
 124             while(((max * qmat[qscale][i]) >> shift) > INT_MAX){
 125                 shift++;
 126             }
 127         }
 128     }
 129     if(shift){
 130         av_log(NULL, AV_LOG_INFO, "Warning, QMAT_SHIFT is larger than %d, overflows possible\n", QMAT_SHIFT - shift);
 131     }
 132 }
 133
 134 static inline void update_qscale(MpegEncContext *s){
 135     s->qscale= (s->lambda*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7);
 136     s->qscale= av_clip(s->qscale, s->avctx->qmin, s->avctx->qmax);
 137
 138     s->lambda2= (s->lambda*s->lambda + FF_LAMBDA_SCALE/2) >> FF_LAMBDA_SHIFT;
 139 }
 140
 141 void ff_write_quant_matrix(PutBitContext *pb, uint16_t *matrix){
 142     int i;
 143
 144     if(matrix){
 145         put_bits(pb, 1, 1);
 146         for(i=0;i<64;i++) {
 147             put_bits(pb, 8, matrix[ ff_zigzag_direct[i] ]);
 148         }
 149     }else
 150         put_bits(pb, 1, 0);
 151 }
 152
 153 /**
 154  * init s->current_picture.qscale_table from s->lambda_table
 155  */
 156 void ff_init_qscale_tab(MpegEncContext *s){
 157     int8_t * const qscale_table= s->current_picture.qscale_table;
 158     int i;
 159
 160     for(i=0; i<s->mb_num; i++){
 161         unsigned int lam= s->lambda_table[ s->mb_index2xy[i] ];
 162         int qp= (lam*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7);
 163         qscale_table[ s->mb_index2xy[i] ]= av_clip(qp, s->avctx->qmin, s->avctx->qmax);
 164     }
 165 }
 166
 167 static void copy_picture_attributes(MpegEncContext *s, AVFrame *dst, AVFrame *src){
 168     int i;
 169
 170     dst->pict_type              = src->pict_type;
 171     dst->quality                = src->quality;
 172     dst->coded_picture_number   = src->coded_picture_number;
 173     dst->display_picture_number = src->display_picture_number;
 174 //    dst->reference              = src->reference;
 175     dst->pts                    = src->pts;
 176     dst->interlaced_frame       = src->interlaced_frame;
 177     dst->top_field_first        = src->top_field_first;
 178
 179     if(s->avctx->me_threshold){
 180         if(!src->motion_val[0])
 181             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_val not set!\n");
 182         if(!src->mb_type)
 183             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.mb_type not set!\n");
 184         if(!src->ref_index[0])
 185             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.ref_index not set!\n");
 186         if(src->motion_subsample_log2 != dst->motion_subsample_log2)
 187             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_subsample_log2 doesn't match! (%d!=%d)\n",
 188             src->motion_subsample_log2, dst->motion_subsample_log2);
 189
 190         memcpy(dst->mb_type, src->mb_type, s->mb_stride * s->mb_height * sizeof(dst->mb_type[0]));
 191
 192         for(i=0; i<2; i++){
 193             int stride= ((16*s->mb_width )>>src->motion_subsample_log2) + 1;
 194             int height= ((16*s->mb_height)>>src->motion_subsample_log2);
 195
 196             if(src->motion_val[i] && src->motion_val[i] != dst->motion_val[i]){
 197                 memcpy(dst->motion_val[i], src->motion_val[i], 2*stride*height*sizeof(int16_t));
 198             }
 199             if(src->ref_index[i] && src->ref_index[i] != dst->ref_index[i]){
 200                 memcpy(dst->ref_index[i], src->ref_index[i], s->b8_stride*2*s->mb_height*sizeof(int8_t));
 201             }
 202         }
 203     }
 204 }
 205
 206 static void update_duplicate_context_after_me(MpegEncContext *dst, MpegEncContext *src){
 207 #define COPY(a) dst->a= src->a
 208     COPY(pict_type);
 209     COPY(current_picture);
 210     COPY(f_code);
 211     COPY(b_code);
 212     COPY(qscale);
 213     COPY(lambda);
 214     COPY(lambda2);
 215     COPY(picture_in_gop_number);
 216     COPY(gop_picture_number);
 217     COPY(frame_pred_frame_dct); //FIXME don't set in encode_header
 218     COPY(progressive_frame); //FIXME don't set in encode_header
 219     COPY(partitioned_frame); //FIXME don't set in encode_header
 220 #undef COPY
 221 }
 222
 223 /**
 224  * sets the given MpegEncContext to defaults for encoding.
 225  * the changed fields will not depend upon the prior state of the MpegEncContext.
 226  */
 227 static void MPV_encode_defaults(MpegEncContext *s){
 228     int i;
 229     MPV_common_defaults(s);
 230
 231     for(i=-16; i<16; i++){
 232         default_fcode_tab[i + MAX_MV]= 1;
 233     }
 234     s->me.mv_penalty= default_mv_penalty;
 235     s->fcode_tab= default_fcode_tab;
 236 }
 237
 238 /* init video encoder */
 239 av_cold int MPV_encode_init(AVCodecContext *avctx)
 240 {
 241     MpegEncContext *s = avctx->priv_data;
 242     int i;
 243     int chroma_h_shift, chroma_v_shift;
 244
 245     MPV_encode_defaults(s);
 246
 247     switch (avctx->codec_id) {
 248     case CODEC_ID_MPEG2VIDEO:
 249         if(avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P){
 250             av_log(avctx, AV_LOG_ERROR, "only YUV420 and YUV422 are supported\n");
 251             return -1;
 252         }
 253         break;
 254     case CODEC_ID_LJPEG:
 255     case CODEC_ID_MJPEG:
 256         if(avctx->pix_fmt != PIX_FMT_YUVJ420P && avctx->pix_fmt != PIX_FMT_YUVJ422P && avctx->pix_fmt != PIX_FMT_RGB32 &&
 257            ((avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P) || avctx->strict_std_compliance>FF_COMPLIANCE_INOFFICIAL)){
 258             av_log(avctx, AV_LOG_ERROR, "colorspace not supported in jpeg\n");
 259             return -1;
 260         }
 261         break;
 262     default:
 263         if(avctx->pix_fmt != PIX_FMT_YUV420P){
 264             av_log(avctx, AV_LOG_ERROR, "only YUV420 is supported\n");
 265             return -1;
 266         }
 267     }
 268
 269     switch (avctx->pix_fmt) {
 270     case PIX_FMT_YUVJ422P:
 271     case PIX_FMT_YUV422P:
 272         s->chroma_format = CHROMA_422;
 273         break;
 274     case PIX_FMT_YUVJ420P:
 275     case PIX_FMT_YUV420P:
 276     default:
 277         s->chroma_format = CHROMA_420;
 278         break;
 279     }
 280
 281     s->bit_rate = avctx->bit_rate;
 282     s->width = avctx->width;
 283     s->height = avctx->height;
 284     if(avctx->gop_size > 600 && avctx->strict_std_compliance>FF_COMPLIANCE_EXPERIMENTAL){
 285         av_log(avctx, AV_LOG_ERROR, "Warning keyframe interval too large! reducing it ...\n");
 286         avctx->gop_size=600;
 287     }
 288     s->gop_size = avctx->gop_size;
 289     s->avctx = avctx;
 290     s->flags= avctx->flags;
 291     s->flags2= avctx->flags2;
 292     s->max_b_frames= avctx->max_b_frames;
 293     s->codec_id= avctx->codec->id;
 294     s->luma_elim_threshold  = avctx->luma_elim_threshold;
 295     s->chroma_elim_threshold= avctx->chroma_elim_threshold;
 296     s->strict_std_compliance= avctx->strict_std_compliance;
 297     s->data_partitioning= avctx->flags & CODEC_FLAG_PART;
 298     s->quarter_sample= (avctx->flags & CODEC_FLAG_QPEL)!=0;
 299     s->mpeg_quant= avctx->mpeg_quant;
 300     s->rtp_mode= !!avctx->rtp_payload_size;
 301     s->intra_dc_precision= avctx->intra_dc_precision;
 302     s->user_specified_pts = AV_NOPTS_VALUE;
 303
 304     if (s->gop_size <= 1) {
 305         s->intra_only = 1;
 306         s->gop_size = 12;
 307     } else {
 308         s->intra_only = 0;
 309     }
 310
 311     s->me_method = avctx->me_method;
 312
 313     /* Fixed QSCALE */
 314     s->fixed_qscale = !!(avctx->flags & CODEC_FLAG_QSCALE);
 315
 316     s->adaptive_quant= (   s->avctx->lumi_masking
 317                         || s->avctx->dark_masking
 318                         || s->avctx->temporal_cplx_masking
 319                         || s->avctx->spatial_cplx_masking
 320                         || s->avctx->p_masking
 321                         || s->avctx->border_masking
 322                         || (s->flags&CODEC_FLAG_QP_RD))
 323                        && !s->fixed_qscale;
 324
 325     s->obmc= !!(s->flags & CODEC_FLAG_OBMC);
 326     s->loop_filter= !!(s->flags & CODEC_FLAG_LOOP_FILTER);
 327     s->alternate_scan= !!(s->flags & CODEC_FLAG_ALT_SCAN);
 328     s->intra_vlc_format= !!(s->flags2 & CODEC_FLAG2_INTRA_VLC);
 329     s->q_scale_type= !!(s->flags2 & CODEC_FLAG2_NON_LINEAR_QUANT);
 330
 331     if(avctx->rc_max_rate && !avctx->rc_buffer_size){
 332         av_log(avctx, AV_LOG_ERROR, "a vbv buffer size is needed, for encoding with a maximum bitrate\n");
 333         return -1;
 334     }
 335
 336     if(avctx->rc_min_rate && avctx->rc_max_rate != avctx->rc_min_rate){
 337         av_log(avctx, AV_LOG_INFO, "Warning min_rate > 0 but min_rate != max_rate isn't recommended!\n");
 338     }
 339
 340     if(avctx->rc_min_rate && avctx->rc_min_rate > avctx->bit_rate){
 341         av_log(avctx, AV_LOG_ERROR, "bitrate below min bitrate\n");
 342         return -1;
 343     }
 344
 345     if(avctx->rc_max_rate && avctx->rc_max_rate < avctx->bit_rate){
 346         av_log(avctx, AV_LOG_INFO, "bitrate above max bitrate\n");
 347         return -1;
 348     }
 349
 350     if(avctx->rc_max_rate && avctx->rc_max_rate == avctx->bit_rate && avctx->rc_max_rate != avctx->rc_min_rate){
 351         av_log(avctx, AV_LOG_INFO, "impossible bitrate constraints, this will fail\n");
 352     }
 353
 354     if(avctx->rc_buffer_size && avctx->bit_rate*av_q2d(avctx->time_base) > avctx->rc_buffer_size){
 355         av_log(avctx, AV_LOG_ERROR, "VBV buffer too small for bitrate\n");
 356         return -1;
 357     }
 358
 359     if(avctx->bit_rate*av_q2d(avctx->time_base) > avctx->bit_rate_tolerance){
 360         av_log(avctx, AV_LOG_ERROR, "bitrate tolerance too small for bitrate\n");
 361         return -1;
 362     }
 363
 364     if(   s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate
 365        && (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO)
 366        && 90000LL * (avctx->rc_buffer_size-1) > s->avctx->rc_max_rate*0xFFFFLL){
 367
 368         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");
 369     }
 370
 371     if((s->flags & CODEC_FLAG_4MV) && s->codec_id != CODEC_ID_MPEG4
 372        && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P && s->codec_id != CODEC_ID_FLV1){
 373         av_log(avctx, AV_LOG_ERROR, "4MV not supported by codec\n");
 374         return -1;
 375     }
 376
 377     if(s->obmc && s->avctx->mb_decision != FF_MB_DECISION_SIMPLE){
 378         av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with simple mb decision\n");
 379         return -1;
 380     }
 381
 382     if(s->obmc && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P){
 383         av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with H263(+)\n");
 384         return -1;
 385     }
 386
 387     if(s->quarter_sample && s->codec_id != CODEC_ID_MPEG4){
 388         av_log(avctx, AV_LOG_ERROR, "qpel not supported by codec\n");
 389         return -1;
 390     }
 391
 392     if(s->data_partitioning && s->codec_id != CODEC_ID_MPEG4){
 393         av_log(avctx, AV_LOG_ERROR, "data partitioning not supported by codec\n");
 394         return -1;
 395     }
 396
 397     if(s->max_b_frames && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO){
 398         av_log(avctx, AV_LOG_ERROR, "b frames not supported by codec\n");
 399         return -1;
 400     }
 401
 402     if ((s->codec_id == CODEC_ID_MPEG4 || s->codec_id == CODEC_ID_H263 ||
 403          s->codec_id == CODEC_ID_H263P) &&
 404         (avctx->sample_aspect_ratio.num > 255 || avctx->sample_aspect_ratio.den > 255)) {
 405         av_log(avctx, AV_LOG_ERROR, "Invalid pixel aspect ratio %i/%i, limit is 255/255\n",
 406                avctx->sample_aspect_ratio.num, avctx->sample_aspect_ratio.den);
 407         return -1;
 408     }
 409
 410     if((s->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN))
 411        && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG2VIDEO){
 412         av_log(avctx, AV_LOG_ERROR, "interlacing not supported by codec\n");
 413         return -1;
 414     }
 415
 416     if(s->mpeg_quant && s->codec_id != CODEC_ID_MPEG4){ //FIXME mpeg2 uses that too
 417         av_log(avctx, AV_LOG_ERROR, "mpeg2 style quantization not supported by codec\n");
 418         return -1;
 419     }
 420
 421     if((s->flags & CODEC_FLAG_CBP_RD) && !avctx->trellis){
 422         av_log(avctx, AV_LOG_ERROR, "CBP RD needs trellis quant\n");
 423         return -1;
 424     }
 425
 426     if((s->flags & CODEC_FLAG_QP_RD) && s->avctx->mb_decision != FF_MB_DECISION_RD){
 427         av_log(avctx, AV_LOG_ERROR, "QP RD needs mbd=2\n");
 428         return -1;
 429     }
 430
 431     if(s->avctx->scenechange_threshold < 1000000000 && (s->flags & CODEC_FLAG_CLOSED_GOP)){
 432         av_log(avctx, AV_LOG_ERROR, "closed gop with scene change detection are not supported yet, set threshold to 1000000000\n");
 433         return -1;
 434     }
 435
 436     if((s->flags2 & CODEC_FLAG2_INTRA_VLC) && s->codec_id != CODEC_ID_MPEG2VIDEO){
 437         av_log(avctx, AV_LOG_ERROR, "intra vlc table not supported by codec\n");
 438         return -1;
 439     }
 440
 441     if(s->flags & CODEC_FLAG_LOW_DELAY){
 442         if (s->codec_id != CODEC_ID_MPEG2VIDEO){
 443             av_log(avctx, AV_LOG_ERROR, "low delay forcing is only available for mpeg2\n");
 444             return -1;
 445         }
 446         if (s->max_b_frames != 0){
 447             av_log(avctx, AV_LOG_ERROR, "b frames cannot be used with low delay\n");
 448             return -1;
 449         }
 450     }
 451
 452     if(s->q_scale_type == 1){
 453         if(s->codec_id != CODEC_ID_MPEG2VIDEO){
 454             av_log(avctx, AV_LOG_ERROR, "non linear quant is only available for mpeg2\n");
 455             return -1;
 456         }
 457         if(avctx->qmax > 12){
 458             av_log(avctx, AV_LOG_ERROR, "non linear quant only supports qmax <= 12 currently\n");
 459             return -1;
 460         }
 461     }
 462
 463     if(s->avctx->thread_count > 1 && s->codec_id != CODEC_ID_MPEG4
 464        && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO
 465        && (s->codec_id != CODEC_ID_H263P || !(s->flags & CODEC_FLAG_H263P_SLICE_STRUCT))){
 466         av_log(avctx, AV_LOG_ERROR, "multi threaded encoding not supported by codec\n");
 467         return -1;
 468     }
 469
 470     if(s->avctx->thread_count < 1){
 471         av_log(avctx, AV_LOG_ERROR, "automatic thread number detection not supported by codec, patch welcome\n");
 472         return -1;
 473     }
 474
 475     if(s->avctx->thread_count > 1)
 476         s->rtp_mode= 1;
 477
 478     if(!avctx->time_base.den || !avctx->time_base.num){
 479         av_log(avctx, AV_LOG_ERROR, "framerate not set\n");
 480         return -1;
 481     }
 482
 483     i= (INT_MAX/2+128)>>8;
 484     if(avctx->me_threshold >= i){
 485         av_log(avctx, AV_LOG_ERROR, "me_threshold too large, max is %d\n", i - 1);
 486         return -1;
 487     }
 488     if(avctx->mb_threshold >= i){
 489         av_log(avctx, AV_LOG_ERROR, "mb_threshold too large, max is %d\n", i - 1);
 490         return -1;
 491     }
 492
 493     if(avctx->b_frame_strategy && (avctx->flags&CODEC_FLAG_PASS2)){
 494         av_log(avctx, AV_LOG_INFO, "notice: b_frame_strategy only affects the first pass\n");
 495         avctx->b_frame_strategy = 0;
 496     }
 497
 498     i= av_gcd(avctx->time_base.den, avctx->time_base.num);
 499     if(i > 1){
 500         av_log(avctx, AV_LOG_INFO, "removing common factors from framerate\n");
 501         avctx->time_base.den /= i;
 502         avctx->time_base.num /= i;
 503 //        return -1;
 504     }
 505
 506     if(s->mpeg_quant || s->codec_id==CODEC_ID_MPEG1VIDEO || s->codec_id==CODEC_ID_MPEG2VIDEO || s->codec_id==CODEC_ID_MJPEG){
 507         s->intra_quant_bias= 3<<(QUANT_BIAS_SHIFT-3); //(a + x*3/8)/x
 508         s->inter_quant_bias= 0;
 509     }else{
 510         s->intra_quant_bias=0;
 511         s->inter_quant_bias=-(1<<(QUANT_BIAS_SHIFT-2)); //(a - x/4)/x
 512     }
 513
 514     if(avctx->intra_quant_bias != FF_DEFAULT_QUANT_BIAS)
 515         s->intra_quant_bias= avctx->intra_quant_bias;
 516     if(avctx->inter_quant_bias != FF_DEFAULT_QUANT_BIAS)
 517         s->inter_quant_bias= avctx->inter_quant_bias;
 518
 519     avcodec_get_chroma_sub_sample(avctx->pix_fmt, &chroma_h_shift, &chroma_v_shift);
 520
 521     if(avctx->codec_id == CODEC_ID_MPEG4 && s->avctx->time_base.den > (1<<16)-1){
 522         av_log(avctx, AV_LOG_ERROR, "timebase not supported by mpeg 4 standard\n");
 523         return -1;
 524     }
 525     s->time_increment_bits = av_log2(s->avctx->time_base.den - 1) + 1;
 526
 527     switch(avctx->codec->id) {
 528     case CODEC_ID_MPEG1VIDEO:
 529         s->out_format = FMT_MPEG1;
 530         s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY);
 531         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
 532         break;
 533     case CODEC_ID_MPEG2VIDEO:
 534         s->out_format = FMT_MPEG1;
 535         s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY);
 536         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
 537         s->rtp_mode= 1;
 538         break;
 539     case CODEC_ID_LJPEG:
 540     case CODEC_ID_MJPEG:
 541         s->out_format = FMT_MJPEG;
 542         s->intra_only = 1; /* force intra only for jpeg */
 543         if(avctx->codec->id == CODEC_ID_LJPEG && avctx->pix_fmt == PIX_FMT_BGRA){
 544             s->mjpeg_vsample[0] = s->mjpeg_hsample[0] =
 545             s->mjpeg_vsample[1] = s->mjpeg_hsample[1] =
 546             s->mjpeg_vsample[2] = s->mjpeg_hsample[2] = 1;
 547         }else{
 548             s->mjpeg_vsample[0] = 2;
 549             s->mjpeg_vsample[1] = 2>>chroma_v_shift;
 550             s->mjpeg_vsample[2] = 2>>chroma_v_shift;
 551             s->mjpeg_hsample[0] = 2;
 552             s->mjpeg_hsample[1] = 2>>chroma_h_shift;
 553             s->mjpeg_hsample[2] = 2>>chroma_h_shift;
 554         }
 555         if (!(CONFIG_MJPEG_ENCODER || CONFIG_LJPEG_ENCODER)
 556             || ff_mjpeg_encode_init(s) < 0)
 557             return -1;
 558         avctx->delay=0;
 559         s->low_delay=1;
 560         break;
 561     case CODEC_ID_H261:
 562         if (!CONFIG_H261_ENCODER)  return -1;
 563         if (ff_h261_get_picture_format(s->width, s->height) < 0) {
 564             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);
 565             return -1;
 566         }
 567         s->out_format = FMT_H261;
 568         avctx->delay=0;
 569         s->low_delay=1;
 570         break;
 571     case CODEC_ID_H263:
 572         if (!CONFIG_H263_ENCODER)  return -1;
 573         if (ff_match_2uint16(h263_format, FF_ARRAY_ELEMS(h263_format), s->width, s->height) == 7) {
 574             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);
 575             return -1;
 576         }
 577         s->out_format = FMT_H263;
 578         s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
 579         avctx->delay=0;
 580         s->low_delay=1;
 581         break;
 582     case CODEC_ID_H263P:
 583         s->out_format = FMT_H263;
 584         s->h263_plus = 1;
 585         /* Fx */
 586         s->umvplus = (avctx->flags & CODEC_FLAG_H263P_UMV) ? 1:0;
 587         s->h263_aic= (avctx->flags & CODEC_FLAG_AC_PRED) ? 1:0;
 588         s->modified_quant= s->h263_aic;
 589         s->alt_inter_vlc= (avctx->flags & CODEC_FLAG_H263P_AIV) ? 1:0;
 590         s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
 591         s->loop_filter= (avctx->flags & CODEC_FLAG_LOOP_FILTER) ? 1:0;
 592         s->unrestricted_mv= s->obmc || s->loop_filter || s->umvplus;
 593         s->h263_slice_structured= (s->flags & CODEC_FLAG_H263P_SLICE_STRUCT) ? 1:0;
 594
 595         /* /Fx */
 596         /* These are just to be sure */
 597         avctx->delay=0;
 598         s->low_delay=1;
 599         break;
 600     case CODEC_ID_FLV1:
 601         s->out_format = FMT_H263;
 602         s->h263_flv = 2; /* format = 1; 11-bit codes */
 603         s->unrestricted_mv = 1;
 604         s->rtp_mode=0; /* don't allow GOB */
 605         avctx->delay=0;
 606         s->low_delay=1;
 607         break;
 608     case CODEC_ID_RV10:
 609         s->out_format = FMT_H263;
 610         avctx->delay=0;
 611         s->low_delay=1;
 612         break;
 613     case CODEC_ID_RV20:
 614         s->out_format = FMT_H263;
 615         avctx->delay=0;
 616         s->low_delay=1;
 617         s->modified_quant=1;
 618         s->h263_aic=1;
 619         s->h263_plus=1;
 620         s->loop_filter=1;
 621         s->unrestricted_mv= s->obmc || s->loop_filter || s->umvplus;
 622         break;
 623     case CODEC_ID_MPEG4:
 624         s->out_format = FMT_H263;
 625         s->h263_pred = 1;
 626         s->unrestricted_mv = 1;
 627         s->low_delay= s->max_b_frames ? 0 : 1;
 628         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
 629         break;
 630     case CODEC_ID_MSMPEG4V1:
 631         s->out_format = FMT_H263;
 632         s->h263_msmpeg4 = 1;
 633         s->h263_pred = 1;
 634         s->unrestricted_mv = 1;
 635         s->msmpeg4_version= 1;
 636         avctx->delay=0;
 637         s->low_delay=1;
 638         break;
 639     case CODEC_ID_MSMPEG4V2:
 640         s->out_format = FMT_H263;
 641         s->h263_msmpeg4 = 1;
 642         s->h263_pred = 1;
 643         s->unrestricted_mv = 1;
 644         s->msmpeg4_version= 2;
 645         avctx->delay=0;
 646         s->low_delay=1;
 647         break;
 648     case CODEC_ID_MSMPEG4V3:
 649         s->out_format = FMT_H263;
 650         s->h263_msmpeg4 = 1;
 651         s->h263_pred = 1;
 652         s->unrestricted_mv = 1;
 653         s->msmpeg4_version= 3;
 654         s->flipflop_rounding=1;
 655         avctx->delay=0;
 656         s->low_delay=1;
 657         break;
 658     case CODEC_ID_WMV1:
 659         s->out_format = FMT_H263;
 660         s->h263_msmpeg4 = 1;
 661         s->h263_pred = 1;
 662         s->unrestricted_mv = 1;
 663         s->msmpeg4_version= 4;
 664         s->flipflop_rounding=1;
 665         avctx->delay=0;
 666         s->low_delay=1;
 667         break;
 668     case CODEC_ID_WMV2:
 669         s->out_format = FMT_H263;
 670         s->h263_msmpeg4 = 1;
 671         s->h263_pred = 1;
 672         s->unrestricted_mv = 1;
 673         s->msmpeg4_version= 5;
 674         s->flipflop_rounding=1;
 675         avctx->delay=0;
 676         s->low_delay=1;
 677         break;
 678     default:
 679         return -1;
 680     }
 681
 682     avctx->has_b_frames= !s->low_delay;
 683
 684     s->encoding = 1;
 685
 686     s->progressive_frame=
 687     s->progressive_sequence= !(avctx->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN));
 688
 689     /* init */
 690     if (MPV_common_init(s) < 0)
 691         return -1;
 692
 693     if(!s->dct_quantize)
 694         s->dct_quantize = dct_quantize_c;
 695     if(!s->denoise_dct)
 696         s->denoise_dct = denoise_dct_c;
 697     s->fast_dct_quantize = s->dct_quantize;
 698     if(avctx->trellis)
 699         s->dct_quantize = dct_quantize_trellis_c;
 700
 701     if((CONFIG_H263P_ENCODER || CONFIG_RV20_ENCODER) && s->modified_quant)
 702         s->chroma_qscale_table= ff_h263_chroma_qscale_table;
 703
 704     s->quant_precision=5;
 705
 706     ff_set_cmp(&s->dsp, s->dsp.ildct_cmp, s->avctx->ildct_cmp);
 707     ff_set_cmp(&s->dsp, s->dsp.frame_skip_cmp, s->avctx->frame_skip_cmp);
 708
 709     if (CONFIG_H261_ENCODER && s->out_format == FMT_H261)
 710         ff_h261_encode_init(s);
 711     if (CONFIG_H263_ENCODER && s->out_format == FMT_H263)
 712         h263_encode_init(s);
 713     if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version)
 714         ff_msmpeg4_encode_init(s);
 715     if ((CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
 716         && s->out_format == FMT_MPEG1)
 717         ff_mpeg1_encode_init(s);
 718
 719     /* init q matrix */
 720     for(i=0;i<64;i++) {
 721         int j= s->dsp.idct_permutation[i];
 722         if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4 && s->mpeg_quant){
 723             s->intra_matrix[j] = ff_mpeg4_default_intra_matrix[i];
 724             s->inter_matrix[j] = ff_mpeg4_default_non_intra_matrix[i];
 725         }else if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
 726             s->intra_matrix[j] =
 727             s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
 728         }else
 729         { /* mpeg1/2 */
 730             s->intra_matrix[j] = ff_mpeg1_default_intra_matrix[i];
 731             s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
 732         }
 733         if(s->avctx->intra_matrix)
 734             s->intra_matrix[j] = s->avctx->intra_matrix[i];
 735         if(s->avctx->inter_matrix)
 736             s->inter_matrix[j] = s->avctx->inter_matrix[i];
 737     }
 738
 739     /* precompute matrix */
 740     /* for mjpeg, we do include qscale in the matrix */
 741     if (s->out_format != FMT_MJPEG) {
 742         ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
 743                        s->intra_matrix, s->intra_quant_bias, avctx->qmin, 31, 1);
 744         ff_convert_matrix(&s->dsp, s->q_inter_matrix, s->q_inter_matrix16,
 745                        s->inter_matrix, s->inter_quant_bias, avctx->qmin, 31, 0);
 746     }
 747
 748     if(ff_rate_control_init(s) < 0)
 749         return -1;
 750
 751     return 0;
 752 }
 753
 754 av_cold int MPV_encode_end(AVCodecContext *avctx)
 755 {
 756     MpegEncContext *s = avctx->priv_data;
 757
 758     ff_rate_control_uninit(s);
 759
 760     MPV_common_end(s);
 761     if ((CONFIG_MJPEG_ENCODER || CONFIG_LJPEG_ENCODER) && s->out_format == FMT_MJPEG)
 762         ff_mjpeg_encode_close(s);
 763
 764     av_freep(&avctx->extradata);
 765
 766     return 0;
 767 }
 768
 769 static int get_sae(uint8_t *src, int ref, int stride){
 770     int x,y;
 771     int acc=0;
 772
 773     for(y=0; y<16; y++){
 774         for(x=0; x<16; x++){
 775             acc+= FFABS(src[x+y*stride] - ref);
 776         }
 777     }
 778
 779     return acc;
 780 }
 781
 782 static int get_intra_count(MpegEncContext *s, uint8_t *src, uint8_t *ref, int stride){
 783     int x, y, w, h;
 784     int acc=0;
 785
 786     w= s->width &~15;
 787     h= s->height&~15;
 788
 789     for(y=0; y<h; y+=16){
 790         for(x=0; x<w; x+=16){
 791             int offset= x + y*stride;
 792             int sad = s->dsp.sad[0](NULL, src + offset, ref + offset, stride, 16);
 793             int mean= (s->dsp.pix_sum(src + offset, stride) + 128)>>8;
 794             int sae = get_sae(src + offset, mean, stride);
 795
 796             acc+= sae + 500 < sad;
 797         }
 798     }
 799     return acc;
 800 }
 801
 802
 803 static int load_input_picture(MpegEncContext *s, AVFrame *pic_arg){
 804     AVFrame *pic=NULL;
 805     int64_t pts;
 806     int i;
 807     const int encoding_delay= s->max_b_frames;
 808     int direct=1;
 809
 810     if(pic_arg){
 811         pts= pic_arg->pts;
 812         pic_arg->display_picture_number= s->input_picture_number++;
 813
 814         if(pts != AV_NOPTS_VALUE){
 815             if(s->user_specified_pts != AV_NOPTS_VALUE){
 816                 int64_t time= pts;
 817                 int64_t last= s->user_specified_pts;
 818
 819                 if(time <= last){
 820                     av_log(s->avctx, AV_LOG_ERROR, "Error, Invalid timestamp=%"PRId64", last=%"PRId64"\n", pts, s->user_specified_pts);
 821                     return -1;
 822                 }
 823             }
 824             s->user_specified_pts= pts;
 825         }else{
 826             if(s->user_specified_pts != AV_NOPTS_VALUE){
 827                 s->user_specified_pts=
 828                 pts= s->user_specified_pts + 1;
 829                 av_log(s->avctx, AV_LOG_INFO, "Warning: AVFrame.pts=? trying to guess (%"PRId64")\n", pts);
 830             }else{
 831                 pts= pic_arg->display_picture_number;
 832             }
 833         }
 834     }
 835
 836   if(pic_arg){
 837     if(encoding_delay && !(s->flags&CODEC_FLAG_INPUT_PRESERVED)) direct=0;
 838     if(pic_arg->linesize[0] != s->linesize) direct=0;
 839     if(pic_arg->linesize[1] != s->uvlinesize) direct=0;
 840     if(pic_arg->linesize[2] != s->uvlinesize) direct=0;
 841
 842 //    av_log(AV_LOG_DEBUG, "%d %d %d %d\n",pic_arg->linesize[0], pic_arg->linesize[1], s->linesize, s->uvlinesize);
 843
 844     if(direct){
 845         i= ff_find_unused_picture(s, 1);
 846
 847         pic= (AVFrame*)&s->picture[i];
 848         pic->reference= 3;
 849
 850         for(i=0; i<4; i++){
 851             pic->data[i]= pic_arg->data[i];
 852             pic->linesize[i]= pic_arg->linesize[i];
 853         }
 854         ff_alloc_picture(s, (Picture*)pic, 1);
 855     }else{
 856         i= ff_find_unused_picture(s, 0);
 857
 858         pic= (AVFrame*)&s->picture[i];
 859         pic->reference= 3;
 860
 861         ff_alloc_picture(s, (Picture*)pic, 0);
 862
 863         if(   pic->data[0] + INPLACE_OFFSET == pic_arg->data[0]
 864            && pic->data[1] + INPLACE_OFFSET == pic_arg->data[1]
 865            && pic->data[2] + INPLACE_OFFSET == pic_arg->data[2]){
 866        // empty
 867         }else{
 868             int h_chroma_shift, v_chroma_shift;
 869             avcodec_get_chroma_sub_sample(s->avctx->pix_fmt, &h_chroma_shift, &v_chroma_shift);
 870
 871             for(i=0; i<3; i++){
 872                 int src_stride= pic_arg->linesize[i];
 873                 int dst_stride= i ? s->uvlinesize : s->linesize;
 874                 int h_shift= i ? h_chroma_shift : 0;
 875                 int v_shift= i ? v_chroma_shift : 0;
 876                 int w= s->width >>h_shift;
 877                 int h= s->height>>v_shift;
 878                 uint8_t *src= pic_arg->data[i];
 879                 uint8_t *dst= pic->data[i];
 880
 881                 if(!s->avctx->rc_buffer_size)
 882                     dst +=INPLACE_OFFSET;
 883
 884                 if(src_stride==dst_stride)
 885                     memcpy(dst, src, src_stride*h);
 886                 else{
 887                     while(h--){
 888                         memcpy(dst, src, w);
 889                         dst += dst_stride;
 890                         src += src_stride;
 891                     }
 892                 }
 893             }
 894         }
 895     }
 896     copy_picture_attributes(s, pic, pic_arg);
 897     pic->pts= pts; //we set this here to avoid modifiying pic_arg
 898   }
 899
 900     /* shift buffer entries */
 901     for(i=1; i<MAX_PICTURE_COUNT /*s->encoding_delay+1*/; i++)
 902         s->input_picture[i-1]= s->input_picture[i];
 903
 904     s->input_picture[encoding_delay]= (Picture*)pic;
 905
 906     return 0;
 907 }
 908
 909 static int skip_check(MpegEncContext *s, Picture *p, Picture *ref){
 910     int x, y, plane;
 911     int score=0;
 912     int64_t score64=0;
 913
 914     for(plane=0; plane<3; plane++){
 915         const int stride= p->linesize[plane];
 916         const int bw= plane ? 1 : 2;
 917         for(y=0; y<s->mb_height*bw; y++){
 918             for(x=0; x<s->mb_width*bw; x++){
 919                 int off= p->type == FF_BUFFER_TYPE_SHARED ? 0: 16;
 920                 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);
 921
 922                 switch(s->avctx->frame_skip_exp){
 923                     case 0: score= FFMAX(score, v); break;
 924                     case 1: score+= FFABS(v);break;
 925                     case 2: score+= v*v;break;
 926                     case 3: score64+= FFABS(v*v*(int64_t)v);break;
 927                     case 4: score64+= v*v*(int64_t)(v*v);break;
 928                 }
 929             }
 930         }
 931     }
 932
 933     if(score) score64= score;
 934
 935     if(score64 < s->avctx->frame_skip_threshold)
 936         return 1;
 937     if(score64 < ((s->avctx->frame_skip_factor * (int64_t)s->lambda)>>8))
 938         return 1;
 939     return 0;
 940 }
 941
 942 static int estimate_best_b_count(MpegEncContext *s){
 943     AVCodec *codec= avcodec_find_encoder(s->avctx->codec_id);
 944     AVCodecContext *c= avcodec_alloc_context();
 945     AVFrame input[FF_MAX_B_FRAMES+2];
 946     const int scale= s->avctx->brd_scale;
 947     int i, j, out_size, p_lambda, b_lambda, lambda2;
 948     int outbuf_size= s->width * s->height; //FIXME
 949     uint8_t *outbuf= av_malloc(outbuf_size);
 950     int64_t best_rd= INT64_MAX;
 951     int best_b_count= -1;
 952
 953     assert(scale>=0 && scale <=3);
 954
 955 //    emms_c();
 956     p_lambda= s->last_lambda_for[FF_P_TYPE]; //s->next_picture_ptr->quality;
 957     b_lambda= s->last_lambda_for[FF_B_TYPE]; //p_lambda *FFABS(s->avctx->b_quant_factor) + s->avctx->b_quant_offset;
 958     if(!b_lambda) b_lambda= p_lambda; //FIXME we should do this somewhere else
 959     lambda2= (b_lambda*b_lambda + (1<<FF_LAMBDA_SHIFT)/2 ) >> FF_LAMBDA_SHIFT;
 960
 961     c->width = s->width >> scale;
 962     c->height= s->height>> scale;
 963     c->flags= CODEC_FLAG_QSCALE | CODEC_FLAG_PSNR | CODEC_FLAG_INPUT_PRESERVED /*| CODEC_FLAG_EMU_EDGE*/;
 964     c->flags|= s->avctx->flags & CODEC_FLAG_QPEL;
 965     c->mb_decision= s->avctx->mb_decision;
 966     c->me_cmp= s->avctx->me_cmp;
 967     c->mb_cmp= s->avctx->mb_cmp;
 968     c->me_sub_cmp= s->avctx->me_sub_cmp;
 969     c->pix_fmt = PIX_FMT_YUV420P;
 970     c->time_base= s->avctx->time_base;
 971     c->max_b_frames= s->max_b_frames;
 972
 973     if (avcodec_open(c, codec) < 0)
 974         return -1;
 975
 976     for(i=0; i<s->max_b_frames+2; i++){
 977         int ysize= c->width*c->height;
 978         int csize= (c->width/2)*(c->height/2);
 979         Picture pre_input, *pre_input_ptr= i ? s->input_picture[i-1] : s->next_picture_ptr;
 980
 981         avcodec_get_frame_defaults(&input[i]);
 982         input[i].data[0]= av_malloc(ysize + 2*csize);
 983         input[i].data[1]= input[i].data[0] + ysize;
 984         input[i].data[2]= input[i].data[1] + csize;
 985         input[i].linesize[0]= c->width;
 986         input[i].linesize[1]=
 987         input[i].linesize[2]= c->width/2;
 988
 989         if(pre_input_ptr && (!i || s->input_picture[i-1])) {
 990             pre_input= *pre_input_ptr;
 991
 992             if(pre_input.type != FF_BUFFER_TYPE_SHARED && i) {
 993                 pre_input.data[0]+=INPLACE_OFFSET;
 994                 pre_input.data[1]+=INPLACE_OFFSET;
 995                 pre_input.data[2]+=INPLACE_OFFSET;
 996             }
 997
 998             s->dsp.shrink[scale](input[i].data[0], input[i].linesize[0], pre_input.data[0], pre_input.linesize[0], c->width, c->height);
 999             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);
1000             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);
1001         }
1002     }
1003
1004     for(j=0; j<s->max_b_frames+1; j++){
1005         int64_t rd=0;
1006
1007         if(!s->input_picture[j])
1008             break;
1009
1010         c->error[0]= c->error[1]= c->error[2]= 0;
1011
1012         input[0].pict_type= FF_I_TYPE;
1013         input[0].quality= 1 * FF_QP2LAMBDA;
1014         out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[0]);
1015 //        rd += (out_size * lambda2) >> FF_LAMBDA_SHIFT;
1016
1017         for(i=0; i<s->max_b_frames+1; i++){
1018             int is_p= i % (j+1) == j || i==s->max_b_frames;
1019
1020             input[i+1].pict_type= is_p ? FF_P_TYPE : FF_B_TYPE;
1021             input[i+1].quality= is_p ? p_lambda : b_lambda;
1022             out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[i+1]);
1023             rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
1024         }
1025
1026         /* get the delayed frames */
1027         while(out_size){
1028             out_size = avcodec_encode_video(c, outbuf, outbuf_size, NULL);
1029             rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
1030         }
1031
1032         rd += c->error[0] + c->error[1] + c->error[2];
1033
1034         if(rd < best_rd){
1035             best_rd= rd;
1036             best_b_count= j;
1037         }
1038     }
1039
1040     av_freep(&outbuf);
1041     avcodec_close(c);
1042     av_freep(&c);
1043
1044     for(i=0; i<s->max_b_frames+2; i++){
1045         av_freep(&input[i].data[0]);
1046     }
1047
1048     return best_b_count;
1049 }
1050
1051 static void select_input_picture(MpegEncContext *s){
1052     int i;
1053
1054     for(i=1; i<MAX_PICTURE_COUNT; i++)
1055         s->reordered_input_picture[i-1]= s->reordered_input_picture[i];
1056     s->reordered_input_picture[MAX_PICTURE_COUNT-1]= NULL;
1057
1058     /* set next picture type & ordering */
1059     if(s->reordered_input_picture[0]==NULL && s->input_picture[0]){
1060         if(/*s->picture_in_gop_number >= s->gop_size ||*/ s->next_picture_ptr==NULL || s->intra_only){
1061             s->reordered_input_picture[0]= s->input_picture[0];
1062             s->reordered_input_picture[0]->pict_type= FF_I_TYPE;
1063             s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++;
1064         }else{
1065             int b_frames;
1066
1067             if(s->avctx->frame_skip_threshold || s->avctx->frame_skip_factor){
1068                 if(s->picture_in_gop_number < s->gop_size && skip_check(s, s->input_picture[0], s->next_picture_ptr)){
1069                 //FIXME check that te gop check above is +-1 correct
1070 //av_log(NULL, AV_LOG_DEBUG, "skip %p %"PRId64"\n", s->input_picture[0]->data[0], s->input_picture[0]->pts);
1071
1072                     if(s->input_picture[0]->type == FF_BUFFER_TYPE_SHARED){
1073                         for(i=0; i<4; i++)
1074                             s->input_picture[0]->data[i]= NULL;
1075                         s->input_picture[0]->type= 0;
1076                     }else{
1077                         assert(   s->input_picture[0]->type==FF_BUFFER_TYPE_USER
1078                                || s->input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
1079
1080                         s->avctx->release_buffer(s->avctx, (AVFrame*)s->input_picture[0]);
1081                     }
1082
1083                     emms_c();
1084                     ff_vbv_update(s, 0);
1085
1086                     goto no_output_pic;
1087                 }
1088             }
1089
1090             if(s->flags&CODEC_FLAG_PASS2){
1091                 for(i=0; i<s->max_b_frames+1; i++){
1092                     int pict_num= s->input_picture[0]->display_picture_number + i;
1093
1094                     if(pict_num >= s->rc_context.num_entries)
1095                         break;
1096                     if(!s->input_picture[i]){
1097                         s->rc_context.entry[pict_num-1].new_pict_type = FF_P_TYPE;
1098                         break;
1099                     }
1100
1101                     s->input_picture[i]->pict_type=
1102                         s->rc_context.entry[pict_num].new_pict_type;
1103                 }
1104             }
1105
1106             if(s->avctx->b_frame_strategy==0){
1107                 b_frames= s->max_b_frames;
1108                 while(b_frames && !s->input_picture[b_frames]) b_frames--;
1109             }else if(s->avctx->b_frame_strategy==1){
1110                 for(i=1; i<s->max_b_frames+1; i++){
1111                     if(s->input_picture[i] && s->input_picture[i]->b_frame_score==0){
1112                         s->input_picture[i]->b_frame_score=
1113                             get_intra_count(s, s->input_picture[i  ]->data[0],
1114                                                s->input_picture[i-1]->data[0], s->linesize) + 1;
1115                     }
1116                 }
1117                 for(i=0; i<s->max_b_frames+1; i++){
1118                     if(s->input_picture[i]==NULL || s->input_picture[i]->b_frame_score - 1 > s->mb_num/s->avctx->b_sensitivity) break;
1119                 }
1120
1121                 b_frames= FFMAX(0, i-1);
1122
1123                 /* reset scores */
1124                 for(i=0; i<b_frames+1; i++){
1125                     s->input_picture[i]->b_frame_score=0;
1126                 }
1127             }else if(s->avctx->b_frame_strategy==2){
1128                 b_frames= estimate_best_b_count(s);
1129             }else{
1130                 av_log(s->avctx, AV_LOG_ERROR, "illegal b frame strategy\n");
1131                 b_frames=0;
1132             }
1133
1134             emms_c();
1135 //static int b_count=0;
1136 //b_count+= b_frames;
1137 //av_log(s->avctx, AV_LOG_DEBUG, "b_frames: %d\n", b_count);
1138
1139             for(i= b_frames - 1; i>=0; i--){
1140                 int type= s->input_picture[i]->pict_type;
1141                 if(type && type != FF_B_TYPE)
1142                     b_frames= i;
1143             }
1144             if(s->input_picture[b_frames]->pict_type == FF_B_TYPE && b_frames == s->max_b_frames){
1145                 av_log(s->avctx, AV_LOG_ERROR, "warning, too many b frames in a row\n");
1146             }
1147
1148             if(s->picture_in_gop_number + b_frames >= s->gop_size){
1149               if((s->flags2 & CODEC_FLAG2_STRICT_GOP) && s->gop_size > s->picture_in_gop_number){
1150                     b_frames= s->gop_size - s->picture_in_gop_number - 1;
1151               }else{
1152                 if(s->flags & CODEC_FLAG_CLOSED_GOP)
1153                     b_frames=0;
1154                 s->input_picture[b_frames]->pict_type= FF_I_TYPE;
1155               }
1156             }
1157
1158             if(   (s->flags & CODEC_FLAG_CLOSED_GOP)
1159                && b_frames
1160                && s->input_picture[b_frames]->pict_type== FF_I_TYPE)
1161                 b_frames--;
1162
1163             s->reordered_input_picture[0]= s->input_picture[b_frames];
1164             if(s->reordered_input_picture[0]->pict_type != FF_I_TYPE)
1165                 s->reordered_input_picture[0]->pict_type= FF_P_TYPE;
1166             s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++;
1167             for(i=0; i<b_frames; i++){
1168                 s->reordered_input_picture[i+1]= s->input_picture[i];
1169                 s->reordered_input_picture[i+1]->pict_type= FF_B_TYPE;
1170                 s->reordered_input_picture[i+1]->coded_picture_number= s->coded_picture_number++;
1171             }
1172         }
1173     }
1174 no_output_pic:
1175     if(s->reordered_input_picture[0]){
1176         s->reordered_input_picture[0]->reference= s->reordered_input_picture[0]->pict_type!=FF_B_TYPE ? 3 : 0;
1177
1178         ff_copy_picture(&s->new_picture, s->reordered_input_picture[0]);
1179
1180         if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_SHARED || s->avctx->rc_buffer_size){
1181             // input is a shared pix, so we can't modifiy it -> alloc a new one & ensure that the shared one is reuseable
1182
1183             int i= ff_find_unused_picture(s, 0);
1184             Picture *pic= &s->picture[i];
1185
1186             pic->reference              = s->reordered_input_picture[0]->reference;
1187             ff_alloc_picture(s, pic, 0);
1188
1189             /* mark us unused / free shared pic */
1190             if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_INTERNAL)
1191                 s->avctx->release_buffer(s->avctx, (AVFrame*)s->reordered_input_picture[0]);
1192             for(i=0; i<4; i++)
1193                 s->reordered_input_picture[0]->data[i]= NULL;
1194             s->reordered_input_picture[0]->type= 0;
1195
1196             copy_picture_attributes(s, (AVFrame*)pic, (AVFrame*)s->reordered_input_picture[0]);
1197
1198             s->current_picture_ptr= pic;
1199         }else{
1200             // input is not a shared pix -> reuse buffer for current_pix
1201
1202             assert(   s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_USER
1203                    || s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
1204
1205             s->current_picture_ptr= s->reordered_input_picture[0];
1206             for(i=0; i<4; i++){
1207                 s->new_picture.data[i]+= INPLACE_OFFSET;
1208             }
1209         }
1210         ff_copy_picture(&s->current_picture, s->current_picture_ptr);
1211
1212         s->picture_number= s->new_picture.display_picture_number;
1213 //printf("dpn:%d\n", s->picture_number);
1214     }else{
1215        memset(&s->new_picture, 0, sizeof(Picture));
1216     }
1217 }
1218
1219 int MPV_encode_picture(AVCodecContext *avctx,
1220                        unsigned char *buf, int buf_size, void *data)
1221 {
1222     MpegEncContext *s = avctx->priv_data;
1223     AVFrame *pic_arg = data;
1224     int i, stuffing_count;
1225
1226     for(i=0; i<avctx->thread_count; i++){
1227         int start_y= s->thread_context[i]->start_mb_y;
1228         int   end_y= s->thread_context[i]->  end_mb_y;
1229         int h= s->mb_height;
1230         uint8_t *start= buf + (size_t)(((int64_t) buf_size)*start_y/h);
1231         uint8_t *end  = buf + (size_t)(((int64_t) buf_size)*  end_y/h);
1232
1233         init_put_bits(&s->thread_context[i]->pb, start, end - start);
1234     }
1235
1236     s->picture_in_gop_number++;
1237
1238     if(load_input_picture(s, pic_arg) < 0)
1239         return -1;
1240
1241     select_input_picture(s);
1242
1243     /* output? */
1244     if(s->new_picture.data[0]){
1245         s->pict_type= s->new_picture.pict_type;
1246 //emms_c();
1247 //printf("qs:%f %f %d\n", s->new_picture.quality, s->current_picture.quality, s->qscale);
1248         MPV_frame_start(s, avctx);
1249 vbv_retry:
1250         if (encode_picture(s, s->picture_number) < 0)
1251             return -1;
1252
1253         avctx->header_bits = s->header_bits;
1254         avctx->mv_bits     = s->mv_bits;
1255         avctx->misc_bits   = s->misc_bits;
1256         avctx->i_tex_bits  = s->i_tex_bits;
1257         avctx->p_tex_bits  = s->p_tex_bits;
1258         avctx->i_count     = s->i_count;
1259         avctx->p_count     = s->mb_num - s->i_count - s->skip_count; //FIXME f/b_count in avctx
1260         avctx->skip_count  = s->skip_count;
1261
1262         MPV_frame_end(s);
1263
1264         if (CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG)
1265             ff_mjpeg_encode_picture_trailer(s);
1266
1267         if(avctx->rc_buffer_size){
1268             RateControlContext *rcc= &s->rc_context;
1269             int max_size= rcc->buffer_index * avctx->rc_max_available_vbv_use;
1270
1271             if(put_bits_count(&s->pb) > max_size && s->lambda < s->avctx->lmax){
1272                 s->next_lambda= FFMAX(s->lambda+1, s->lambda*(s->qscale+1) / s->qscale);
1273                 if(s->adaptive_quant){
1274                     int i;
1275                     for(i=0; i<s->mb_height*s->mb_stride; i++)
1276                         s->lambda_table[i]= FFMAX(s->lambda_table[i]+1, s->lambda_table[i]*(s->qscale+1) / s->qscale);
1277                 }
1278                 s->mb_skipped = 0;        //done in MPV_frame_start()
1279                 if(s->pict_type==FF_P_TYPE){ //done in encode_picture() so we must undo it
1280                     if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
1281                         s->no_rounding ^= 1;
1282                 }
1283                 if(s->pict_type!=FF_B_TYPE){
1284                     s->time_base= s->last_time_base;
1285                     s->last_non_b_time= s->time - s->pp_time;
1286                 }
1287 //                av_log(NULL, AV_LOG_ERROR, "R:%d ", s->next_lambda);
1288                 for(i=0; i<avctx->thread_count; i++){
1289                     PutBitContext *pb= &s->thread_context[i]->pb;
1290                     init_put_bits(pb, pb->buf, pb->buf_end - pb->buf);
1291                 }
1292                 goto vbv_retry;
1293             }
1294
1295             assert(s->avctx->rc_max_rate);
1296         }
1297
1298         if(s->flags&CODEC_FLAG_PASS1)
1299             ff_write_pass1_stats(s);
1300
1301         for(i=0; i<4; i++){
1302             s->current_picture_ptr->error[i]= s->current_picture.error[i];
1303             avctx->error[i] += s->current_picture_ptr->error[i];
1304         }
1305
1306         if(s->flags&CODEC_FLAG_PASS1)
1307             assert(avctx->header_bits + avctx->mv_bits + avctx->misc_bits + avctx->i_tex_bits + avctx->p_tex_bits == put_bits_count(&s->pb));
1308         flush_put_bits(&s->pb);
1309         s->frame_bits  = put_bits_count(&s->pb);
1310
1311         stuffing_count= ff_vbv_update(s, s->frame_bits);
1312         if(stuffing_count){
1313             if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < stuffing_count + 50){
1314                 av_log(s->avctx, AV_LOG_ERROR, "stuffing too large\n");
1315                 return -1;
1316             }
1317
1318             switch(s->codec_id){
1319             case CODEC_ID_MPEG1VIDEO:
1320             case CODEC_ID_MPEG2VIDEO:
1321                 while(stuffing_count--){
1322                     put_bits(&s->pb, 8, 0);
1323                 }
1324             break;
1325             case CODEC_ID_MPEG4:
1326                 put_bits(&s->pb, 16, 0);
1327                 put_bits(&s->pb, 16, 0x1C3);
1328                 stuffing_count -= 4;
1329                 while(stuffing_count--){
1330                     put_bits(&s->pb, 8, 0xFF);
1331                 }
1332             break;
1333             default:
1334                 av_log(s->avctx, AV_LOG_ERROR, "vbv buffer overflow\n");
1335             }
1336             flush_put_bits(&s->pb);
1337             s->frame_bits  = put_bits_count(&s->pb);
1338         }
1339
1340         /* update mpeg1/2 vbv_delay for CBR */
1341         if(s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate && s->out_format == FMT_MPEG1
1342            && 90000LL * (avctx->rc_buffer_size-1) <= s->avctx->rc_max_rate*0xFFFFLL){
1343             int vbv_delay, min_delay;
1344             double inbits = s->avctx->rc_max_rate*av_q2d(s->avctx->time_base);
1345             int    minbits= s->frame_bits - 8*(s->vbv_delay_ptr - s->pb.buf - 1);
1346             double bits   = s->rc_context.buffer_index + minbits - inbits;
1347
1348             if(bits<0)
1349                 av_log(s->avctx, AV_LOG_ERROR, "Internal error, negative bits\n");
1350
1351             assert(s->repeat_first_field==0);
1352
1353             vbv_delay=     bits * 90000                               / s->avctx->rc_max_rate;
1354             min_delay= (minbits * 90000LL + s->avctx->rc_max_rate - 1)/ s->avctx->rc_max_rate;
1355
1356             vbv_delay= FFMAX(vbv_delay, min_delay);
1357
1358             assert(vbv_delay < 0xFFFF);
1359
1360             s->vbv_delay_ptr[0] &= 0xF8;
1361             s->vbv_delay_ptr[0] |= vbv_delay>>13;
1362             s->vbv_delay_ptr[1]  = vbv_delay>>5;
1363             s->vbv_delay_ptr[2] &= 0x07;
1364             s->vbv_delay_ptr[2] |= vbv_delay<<3;
1365         }
1366         s->total_bits += s->frame_bits;
1367         avctx->frame_bits  = s->frame_bits;
1368     }else{
1369         assert((put_bits_ptr(&s->pb) == s->pb.buf));
1370         s->frame_bits=0;
1371     }
1372     assert((s->frame_bits&7)==0);
1373
1374     return s->frame_bits/8;
1375 }
1376
1377 static inline void dct_single_coeff_elimination(MpegEncContext *s, int n, int threshold)
1378 {
1379     static const char tab[64]=
1380         {3,2,2,1,1,1,1,1,
1381          1,1,1,1,1,1,1,1,
1382          1,1,1,1,1,1,1,1,
1383          0,0,0,0,0,0,0,0,
1384          0,0,0,0,0,0,0,0,
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     int score=0;
1389     int run=0;
1390     int i;
1391     DCTELEM *block= s->block[n];
1392     const int last_index= s->block_last_index[n];
1393     int skip_dc;
1394
1395     if(threshold<0){
1396         skip_dc=0;
1397         threshold= -threshold;
1398     }else
1399         skip_dc=1;
1400
1401     /* Are all we could set to zero already zero? */
1402     if(last_index<=skip_dc - 1) return;
1403
1404     for(i=0; i<=last_index; i++){
1405         const int j = s->intra_scantable.permutated[i];
1406         const int level = FFABS(block[j]);
1407         if(level==1){
1408             if(skip_dc && i==0) continue;
1409             score+= tab[run];
1410             run=0;
1411         }else if(level>1){
1412             return;
1413         }else{
1414             run++;
1415         }
1416     }
1417     if(score >= threshold) return;
1418     for(i=skip_dc; i<=last_index; i++){
1419         const int j = s->intra_scantable.permutated[i];
1420         block[j]=0;
1421     }
1422     if(block[0]) s->block_last_index[n]= 0;
1423     else         s->block_last_index[n]= -1;
1424 }
1425
1426 static inline void clip_coeffs(MpegEncContext *s, DCTELEM *block, int last_index)
1427 {
1428     int i;
1429     const int maxlevel= s->max_qcoeff;
1430     const int minlevel= s->min_qcoeff;
1431     int overflow=0;
1432
1433     if(s->mb_intra){
1434         i=1; //skip clipping of intra dc
1435     }else
1436         i=0;
1437
1438     for(;i<=last_index; i++){
1439         const int j= s->intra_scantable.permutated[i];
1440         int level = block[j];
1441
1442         if     (level>maxlevel){
1443             level=maxlevel;
1444             overflow++;
1445         }else if(level<minlevel){
1446             level=minlevel;
1447             overflow++;
1448         }
1449
1450         block[j]= level;
1451     }
1452
1453     if(overflow && s->avctx->mb_decision == FF_MB_DECISION_SIMPLE)
1454         av_log(s->avctx, AV_LOG_INFO, "warning, clipping %d dct coefficients to %d..%d\n", overflow, minlevel, maxlevel);
1455 }
1456
1457 static void get_visual_weight(int16_t *weight, uint8_t *ptr, int stride){
1458     int x, y;
1459 //FIXME optimize
1460     for(y=0; y<8; y++){
1461         for(x=0; x<8; x++){
1462             int x2, y2;
1463             int sum=0;
1464             int sqr=0;
1465             int count=0;
1466
1467             for(y2= FFMAX(y-1, 0); y2 < FFMIN(8, y+2); y2++){
1468                 for(x2= FFMAX(x-1, 0); x2 < FFMIN(8, x+2); x2++){
1469                     int v= ptr[x2 + y2*stride];
1470                     sum += v;
1471                     sqr += v*v;
1472                     count++;
1473                 }
1474             }
1475             weight[x + 8*y]= (36*ff_sqrt(count*sqr - sum*sum)) / count;
1476         }
1477     }
1478 }
1479
1480 static av_always_inline void encode_mb_internal(MpegEncContext *s, int motion_x, int motion_y, int mb_block_height, int mb_block_count)
1481 {
1482     int16_t weight[8][64];
1483     DCTELEM orig[8][64];
1484     const int mb_x= s->mb_x;
1485     const int mb_y= s->mb_y;
1486     int i;
1487     int skip_dct[8];
1488     int dct_offset   = s->linesize*8; //default for progressive frames
1489     uint8_t *ptr_y, *ptr_cb, *ptr_cr;
1490     int wrap_y, wrap_c;
1491
1492     for(i=0; i<mb_block_count; i++) skip_dct[i]=s->skipdct;
1493
1494     if(s->adaptive_quant){
1495         const int last_qp= s->qscale;
1496         const int mb_xy= mb_x + mb_y*s->mb_stride;
1497
1498         s->lambda= s->lambda_table[mb_xy];
1499         update_qscale(s);
1500
1501         if(!(s->flags&CODEC_FLAG_QP_RD)){
1502             s->qscale= s->current_picture_ptr->qscale_table[mb_xy];
1503             s->dquant= s->qscale - last_qp;
1504
1505             if(s->out_format==FMT_H263){
1506                 s->dquant= av_clip(s->dquant, -2, 2);
1507
1508                 if(s->codec_id==CODEC_ID_MPEG4){
1509                     if(!s->mb_intra){
1510                         if(s->pict_type == FF_B_TYPE){
1511                             if(s->dquant&1 || s->mv_dir&MV_DIRECT)
1512                                 s->dquant= 0;
1513                         }
1514                         if(s->mv_type==MV_TYPE_8X8)
1515                             s->dquant=0;
1516                     }
1517                 }
1518             }
1519         }
1520         ff_set_qscale(s, last_qp + s->dquant);
1521     }else if(s->flags&CODEC_FLAG_QP_RD)
1522         ff_set_qscale(s, s->qscale + s->dquant);
1523
1524     wrap_y = s->linesize;
1525     wrap_c = s->uvlinesize;
1526     ptr_y = s->new_picture.data[0] + (mb_y * 16 * wrap_y) + mb_x * 16;
1527     ptr_cb = s->new_picture.data[1] + (mb_y * mb_block_height * wrap_c) + mb_x * 8;
1528     ptr_cr = s->new_picture.data[2] + (mb_y * mb_block_height * wrap_c) + mb_x * 8;
1529
1530     if(mb_x*16+16 > s->width || mb_y*16+16 > s->height){
1531         uint8_t *ebuf= s->edge_emu_buffer + 32;
1532         ff_emulated_edge_mc(ebuf            , ptr_y , wrap_y,16,16,mb_x*16,mb_y*16, s->width   , s->height);
1533         ptr_y= ebuf;
1534         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);
1535         ptr_cb= ebuf+18*wrap_y;
1536         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);
1537         ptr_cr= ebuf+18*wrap_y+8;
1538     }
1539
1540     if (s->mb_intra) {
1541         if(s->flags&CODEC_FLAG_INTERLACED_DCT){
1542             int progressive_score, interlaced_score;
1543
1544             s->interlaced_dct=0;
1545             progressive_score= s->dsp.ildct_cmp[4](s, ptr_y           , NULL, wrap_y, 8)
1546                               +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y*8, NULL, wrap_y, 8) - 400;
1547
1548             if(progressive_score > 0){
1549                 interlaced_score = s->dsp.ildct_cmp[4](s, ptr_y           , NULL, wrap_y*2, 8)
1550                                   +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y  , NULL, wrap_y*2, 8);
1551                 if(progressive_score > interlaced_score){
1552                     s->interlaced_dct=1;
1553
1554                     dct_offset= wrap_y;
1555                     wrap_y<<=1;
1556                     if (s->chroma_format == CHROMA_422)
1557                         wrap_c<<=1;
1558                 }
1559             }
1560         }
1561
1562         s->dsp.get_pixels(s->block[0], ptr_y                 , wrap_y);
1563         s->dsp.get_pixels(s->block[1], ptr_y              + 8, wrap_y);
1564         s->dsp.get_pixels(s->block[2], ptr_y + dct_offset    , wrap_y);
1565         s->dsp.get_pixels(s->block[3], ptr_y + dct_offset + 8, wrap_y);
1566
1567         if(s->flags&CODEC_FLAG_GRAY){
1568             skip_dct[4]= 1;
1569             skip_dct[5]= 1;
1570         }else{
1571             s->dsp.get_pixels(s->block[4], ptr_cb, wrap_c);
1572             s->dsp.get_pixels(s->block[5], ptr_cr, wrap_c);
1573             if(!s->chroma_y_shift){ /* 422 */
1574                 s->dsp.get_pixels(s->block[6], ptr_cb + (dct_offset>>1), wrap_c);
1575                 s->dsp.get_pixels(s->block[7], ptr_cr + (dct_offset>>1), wrap_c);
1576             }
1577         }
1578     }else{
1579         op_pixels_func (*op_pix)[4];
1580         qpel_mc_func (*op_qpix)[16];
1581         uint8_t *dest_y, *dest_cb, *dest_cr;
1582
1583         dest_y  = s->dest[0];
1584         dest_cb = s->dest[1];
1585         dest_cr = s->dest[2];
1586
1587         if ((!s->no_rounding) || s->pict_type==FF_B_TYPE){
1588             op_pix = s->dsp.put_pixels_tab;
1589             op_qpix= s->dsp.put_qpel_pixels_tab;
1590         }else{
1591             op_pix = s->dsp.put_no_rnd_pixels_tab;
1592             op_qpix= s->dsp.put_no_rnd_qpel_pixels_tab;
1593         }
1594
1595         if (s->mv_dir & MV_DIR_FORWARD) {
1596             MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.data, op_pix, op_qpix);
1597             op_pix = s->dsp.avg_pixels_tab;
1598             op_qpix= s->dsp.avg_qpel_pixels_tab;
1599         }
1600         if (s->mv_dir & MV_DIR_BACKWARD) {
1601             MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.data, op_pix, op_qpix);
1602         }
1603
1604         if(s->flags&CODEC_FLAG_INTERLACED_DCT){
1605             int progressive_score, interlaced_score;
1606
1607             s->interlaced_dct=0;
1608             progressive_score= s->dsp.ildct_cmp[0](s, dest_y           , ptr_y           , wrap_y, 8)
1609                               +s->dsp.ildct_cmp[0](s, dest_y + wrap_y*8, ptr_y + wrap_y*8, wrap_y, 8) - 400;
1610
1611             if(s->avctx->ildct_cmp == FF_CMP_VSSE) progressive_score -= 400;
1612
1613             if(progressive_score>0){
1614                 interlaced_score = s->dsp.ildct_cmp[0](s, dest_y           , ptr_y           , wrap_y*2, 8)
1615                                   +s->dsp.ildct_cmp[0](s, dest_y + wrap_y  , ptr_y + wrap_y  , wrap_y*2, 8);
1616
1617                 if(progressive_score > interlaced_score){
1618                     s->interlaced_dct=1;
1619
1620                     dct_offset= wrap_y;
1621                     wrap_y<<=1;
1622                     if (s->chroma_format == CHROMA_422)
1623                         wrap_c<<=1;
1624                 }
1625             }
1626         }
1627
1628         s->dsp.diff_pixels(s->block[0], ptr_y                 , dest_y                 , wrap_y);
1629         s->dsp.diff_pixels(s->block[1], ptr_y              + 8, dest_y              + 8, wrap_y);
1630         s->dsp.diff_pixels(s->block[2], ptr_y + dct_offset    , dest_y + dct_offset    , wrap_y);
1631         s->dsp.diff_pixels(s->block[3], ptr_y + dct_offset + 8, dest_y + dct_offset + 8, wrap_y);
1632
1633         if(s->flags&CODEC_FLAG_GRAY){
1634             skip_dct[4]= 1;
1635             skip_dct[5]= 1;
1636         }else{
1637             s->dsp.diff_pixels(s->block[4], ptr_cb, dest_cb, wrap_c);
1638             s->dsp.diff_pixels(s->block[5], ptr_cr, dest_cr, wrap_c);
1639             if(!s->chroma_y_shift){ /* 422 */
1640                 s->dsp.diff_pixels(s->block[6], ptr_cb + (dct_offset>>1), dest_cb + (dct_offset>>1), wrap_c);
1641                 s->dsp.diff_pixels(s->block[7], ptr_cr + (dct_offset>>1), dest_cr + (dct_offset>>1), wrap_c);
1642             }
1643         }
1644         /* pre quantization */
1645         if(s->current_picture.mc_mb_var[s->mb_stride*mb_y+ mb_x]<2*s->qscale*s->qscale){
1646             //FIXME optimize
1647             if(s->dsp.sad[1](NULL, ptr_y               , dest_y               , wrap_y, 8) < 20*s->qscale) skip_dct[0]= 1;
1648             if(s->dsp.sad[1](NULL, ptr_y            + 8, dest_y            + 8, wrap_y, 8) < 20*s->qscale) skip_dct[1]= 1;
1649             if(s->dsp.sad[1](NULL, ptr_y +dct_offset   , dest_y +dct_offset   , wrap_y, 8) < 20*s->qscale) skip_dct[2]= 1;
1650             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;
1651             if(s->dsp.sad[1](NULL, ptr_cb              , dest_cb              , wrap_c, 8) < 20*s->qscale) skip_dct[4]= 1;
1652             if(s->dsp.sad[1](NULL, ptr_cr              , dest_cr              , wrap_c, 8) < 20*s->qscale) skip_dct[5]= 1;
1653             if(!s->chroma_y_shift){ /* 422 */
1654                 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;
1655                 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;
1656             }
1657         }
1658     }
1659
1660     if(s->avctx->quantizer_noise_shaping){
1661         if(!skip_dct[0]) get_visual_weight(weight[0], ptr_y                 , wrap_y);
1662         if(!skip_dct[1]) get_visual_weight(weight[1], ptr_y              + 8, wrap_y);
1663         if(!skip_dct[2]) get_visual_weight(weight[2], ptr_y + dct_offset    , wrap_y);
1664         if(!skip_dct[3]) get_visual_weight(weight[3], ptr_y + dct_offset + 8, wrap_y);
1665         if(!skip_dct[4]) get_visual_weight(weight[4], ptr_cb                , wrap_c);
1666         if(!skip_dct[5]) get_visual_weight(weight[5], ptr_cr                , wrap_c);
1667         if(!s->chroma_y_shift){ /* 422 */
1668             if(!skip_dct[6]) get_visual_weight(weight[6], ptr_cb + (dct_offset>>1), wrap_c);
1669             if(!skip_dct[7]) get_visual_weight(weight[7], ptr_cr + (dct_offset>>1), wrap_c);
1670         }
1671         memcpy(orig[0], s->block[0], sizeof(DCTELEM)*64*mb_block_count);
1672     }
1673
1674     /* DCT & quantize */
1675     assert(s->out_format!=FMT_MJPEG || s->qscale==8);
1676     {
1677         for(i=0;i<mb_block_count;i++) {
1678             if(!skip_dct[i]){
1679                 int overflow;
1680                 s->block_last_index[i] = s->dct_quantize(s, s->block[i], i, s->qscale, &overflow);
1681             // FIXME we could decide to change to quantizer instead of clipping
1682             // JS: I don't think that would be a good idea it could lower quality instead
1683             //     of improve it. Just INTRADC clipping deserves changes in quantizer
1684                 if (overflow) clip_coeffs(s, s->block[i], s->block_last_index[i]);
1685             }else
1686                 s->block_last_index[i]= -1;
1687         }
1688         if(s->avctx->quantizer_noise_shaping){
1689             for(i=0;i<mb_block_count;i++) {
1690                 if(!skip_dct[i]){
1691                     s->block_last_index[i] = dct_quantize_refine(s, s->block[i], weight[i], orig[i], i, s->qscale);
1692                 }
1693             }
1694         }
1695
1696         if(s->luma_elim_threshold && !s->mb_intra)
1697             for(i=0; i<4; i++)
1698                 dct_single_coeff_elimination(s, i, s->luma_elim_threshold);
1699         if(s->chroma_elim_threshold && !s->mb_intra)
1700             for(i=4; i<mb_block_count; i++)
1701                 dct_single_coeff_elimination(s, i, s->chroma_elim_threshold);
1702
1703         if(s->flags & CODEC_FLAG_CBP_RD){
1704             for(i=0;i<mb_block_count;i++) {
1705                 if(s->block_last_index[i] == -1)
1706                     s->coded_score[i]= INT_MAX/256;
1707             }
1708         }
1709     }
1710
1711     if((s->flags&CODEC_FLAG_GRAY) && s->mb_intra){
1712         s->block_last_index[4]=
1713         s->block_last_index[5]= 0;
1714         s->block[4][0]=
1715         s->block[5][0]= (1024 + s->c_dc_scale/2)/ s->c_dc_scale;
1716     }
1717
1718     //non c quantize code returns incorrect block_last_index FIXME
1719     if(s->alternate_scan && s->dct_quantize != dct_quantize_c){
1720         for(i=0; i<mb_block_count; i++){
1721             int j;
1722             if(s->block_last_index[i]>0){
1723                 for(j=63; j>0; j--){
1724                     if(s->block[i][ s->intra_scantable.permutated[j] ]) break;
1725                 }
1726                 s->block_last_index[i]= j;
1727             }
1728         }
1729     }
1730
1731     /* huffman encode */
1732     switch(s->codec_id){ //FIXME funct ptr could be slightly faster
1733     case CODEC_ID_MPEG1VIDEO:
1734     case CODEC_ID_MPEG2VIDEO:
1735         if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
1736             mpeg1_encode_mb(s, s->block, motion_x, motion_y);
1737         break;
1738     case CODEC_ID_MPEG4:
1739         if (CONFIG_MPEG4_ENCODER)
1740             mpeg4_encode_mb(s, s->block, motion_x, motion_y);
1741         break;
1742     case CODEC_ID_MSMPEG4V2:
1743     case CODEC_ID_MSMPEG4V3:
1744     case CODEC_ID_WMV1:
1745         if (CONFIG_MSMPEG4_ENCODER)
1746             msmpeg4_encode_mb(s, s->block, motion_x, motion_y);
1747         break;
1748     case CODEC_ID_WMV2:
1749         if (CONFIG_WMV2_ENCODER)
1750             ff_wmv2_encode_mb(s, s->block, motion_x, motion_y);
1751         break;
1752     case CODEC_ID_H261:
1753         if (CONFIG_H261_ENCODER)
1754             ff_h261_encode_mb(s, s->block, motion_x, motion_y);
1755         break;
1756     case CODEC_ID_H263:
1757     case CODEC_ID_H263P:
1758     case CODEC_ID_FLV1:
1759     case CODEC_ID_RV10:
1760     case CODEC_ID_RV20:
1761         if (CONFIG_H263_ENCODER)
1762             h263_encode_mb(s, s->block, motion_x, motion_y);
1763         break;
1764     case CODEC_ID_MJPEG:
1765         if (CONFIG_MJPEG_ENCODER)
1766             ff_mjpeg_encode_mb(s, s->block);
1767         break;
1768     default:
1769         assert(0);
1770     }
1771 }
1772
1773 static av_always_inline void encode_mb(MpegEncContext *s, int motion_x, int motion_y)
1774 {
1775     if (s->chroma_format == CHROMA_420) encode_mb_internal(s, motion_x, motion_y,  8, 6);
1776     else                                encode_mb_internal(s, motion_x, motion_y, 16, 8);
1777 }
1778
1779 static inline void copy_context_before_encode(MpegEncContext *d, MpegEncContext *s, int type){
1780     int i;
1781
1782     memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
1783
1784     /* mpeg1 */
1785     d->mb_skip_run= s->mb_skip_run;
1786     for(i=0; i<3; i++)
1787         d->last_dc[i]= s->last_dc[i];
1788
1789     /* statistics */
1790     d->mv_bits= s->mv_bits;
1791     d->i_tex_bits= s->i_tex_bits;
1792     d->p_tex_bits= s->p_tex_bits;
1793     d->i_count= s->i_count;
1794     d->f_count= s->f_count;
1795     d->b_count= s->b_count;
1796     d->skip_count= s->skip_count;
1797     d->misc_bits= s->misc_bits;
1798     d->last_bits= 0;
1799
1800     d->mb_skipped= 0;
1801     d->qscale= s->qscale;
1802     d->dquant= s->dquant;
1803
1804     d->esc3_level_length= s->esc3_level_length;
1805 }
1806
1807 static inline void copy_context_after_encode(MpegEncContext *d, MpegEncContext *s, int type){
1808     int i;
1809
1810     memcpy(d->mv, s->mv, 2*4*2*sizeof(int));
1811     memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
1812
1813     /* mpeg1 */
1814     d->mb_skip_run= s->mb_skip_run;
1815     for(i=0; i<3; i++)
1816         d->last_dc[i]= s->last_dc[i];
1817
1818     /* statistics */
1819     d->mv_bits= s->mv_bits;
1820     d->i_tex_bits= s->i_tex_bits;
1821     d->p_tex_bits= s->p_tex_bits;
1822     d->i_count= s->i_count;
1823     d->f_count= s->f_count;
1824     d->b_count= s->b_count;
1825     d->skip_count= s->skip_count;
1826     d->misc_bits= s->misc_bits;
1827
1828     d->mb_intra= s->mb_intra;
1829     d->mb_skipped= s->mb_skipped;
1830     d->mv_type= s->mv_type;
1831     d->mv_dir= s->mv_dir;
1832     d->pb= s->pb;
1833     if(s->data_partitioning){
1834         d->pb2= s->pb2;
1835         d->tex_pb= s->tex_pb;
1836     }
1837     d->block= s->block;
1838     for(i=0; i<8; i++)
1839         d->block_last_index[i]= s->block_last_index[i];
1840     d->interlaced_dct= s->interlaced_dct;
1841     d->qscale= s->qscale;
1842
1843     d->esc3_level_length= s->esc3_level_length;
1844 }
1845
1846 static inline void encode_mb_hq(MpegEncContext *s, MpegEncContext *backup, MpegEncContext *best, int type,
1847                            PutBitContext pb[2], PutBitContext pb2[2], PutBitContext tex_pb[2],
1848                            int *dmin, int *next_block, int motion_x, int motion_y)
1849 {
1850     int score;
1851     uint8_t *dest_backup[3];
1852
1853     copy_context_before_encode(s, backup, type);
1854
1855     s->block= s->blocks[*next_block];
1856     s->pb= pb[*next_block];
1857     if(s->data_partitioning){
1858         s->pb2   = pb2   [*next_block];
1859         s->tex_pb= tex_pb[*next_block];
1860     }
1861
1862     if(*next_block){
1863         memcpy(dest_backup, s->dest, sizeof(s->dest));
1864         s->dest[0] = s->rd_scratchpad;
1865         s->dest[1] = s->rd_scratchpad + 16*s->linesize;
1866         s->dest[2] = s->rd_scratchpad + 16*s->linesize + 8;
1867         assert(s->linesize >= 32); //FIXME
1868     }
1869
1870     encode_mb(s, motion_x, motion_y);
1871
1872     score= put_bits_count(&s->pb);
1873     if(s->data_partitioning){
1874         score+= put_bits_count(&s->pb2);
1875         score+= put_bits_count(&s->tex_pb);
1876     }
1877
1878     if(s->avctx->mb_decision == FF_MB_DECISION_RD){
1879         MPV_decode_mb(s, s->block);
1880
1881         score *= s->lambda2;
1882         score += sse_mb(s) << FF_LAMBDA_SHIFT;
1883     }
1884
1885     if(*next_block){
1886         memcpy(s->dest, dest_backup, sizeof(s->dest));
1887     }
1888
1889     if(score<*dmin){
1890         *dmin= score;
1891         *next_block^=1;
1892
1893         copy_context_after_encode(best, s, type);
1894     }
1895 }
1896
1897 static int sse(MpegEncContext *s, uint8_t *src1, uint8_t *src2, int w, int h, int stride){
1898     uint32_t *sq = ff_squareTbl + 256;
1899     int acc=0;
1900     int x,y;
1901
1902     if(w==16 && h==16)
1903         return s->dsp.sse[0](NULL, src1, src2, stride, 16);
1904     else if(w==8 && h==8)
1905         return s->dsp.sse[1](NULL, src1, src2, stride, 8);
1906
1907     for(y=0; y<h; y++){
1908         for(x=0; x<w; x++){
1909             acc+= sq[src1[x + y*stride] - src2[x + y*stride]];
1910         }
1911     }
1912
1913     assert(acc>=0);
1914
1915     return acc;
1916 }
1917
1918 static int sse_mb(MpegEncContext *s){
1919     int w= 16;
1920     int h= 16;
1921
1922     if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
1923     if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
1924
1925     if(w==16 && h==16)
1926       if(s->avctx->mb_cmp == FF_CMP_NSSE){
1927         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)
1928                +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)
1929                +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);
1930       }else{
1931         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)
1932                +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)
1933                +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);
1934       }
1935     else
1936         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)
1937                +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)
1938                +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);
1939 }
1940
1941 static int pre_estimate_motion_thread(AVCodecContext *c, void *arg){
1942     MpegEncContext *s= *(void**)arg;
1943
1944
1945     s->me.pre_pass=1;
1946     s->me.dia_size= s->avctx->pre_dia_size;
1947     s->first_slice_line=1;
1948     for(s->mb_y= s->end_mb_y-1; s->mb_y >= s->start_mb_y; s->mb_y--) {
1949         for(s->mb_x=s->mb_width-1; s->mb_x >=0 ;s->mb_x--) {
1950             ff_pre_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
1951         }
1952         s->first_slice_line=0;
1953     }
1954
1955     s->me.pre_pass=0;
1956
1957     return 0;
1958 }
1959
1960 static int estimate_motion_thread(AVCodecContext *c, void *arg){
1961     MpegEncContext *s= *(void**)arg;
1962
1963     ff_check_alignment();
1964
1965     s->me.dia_size= s->avctx->dia_size;
1966     s->first_slice_line=1;
1967     for(s->mb_y= s->start_mb_y; s->mb_y < s->end_mb_y; s->mb_y++) {
1968         s->mb_x=0; //for block init below
1969         ff_init_block_index(s);
1970         for(s->mb_x=0; s->mb_x < s->mb_width; s->mb_x++) {
1971             s->block_index[0]+=2;
1972             s->block_index[1]+=2;
1973             s->block_index[2]+=2;
1974             s->block_index[3]+=2;
1975
1976             /* compute motion vector & mb_type and store in context */
1977             if(s->pict_type==FF_B_TYPE)
1978                 ff_estimate_b_frame_motion(s, s->mb_x, s->mb_y);
1979             else
1980                 ff_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
1981         }
1982         s->first_slice_line=0;
1983     }
1984     return 0;
1985 }
1986
1987 static int mb_var_thread(AVCodecContext *c, void *arg){
1988     MpegEncContext *s= *(void**)arg;
1989     int mb_x, mb_y;
1990
1991     ff_check_alignment();
1992
1993     for(mb_y=s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
1994         for(mb_x=0; mb_x < s->mb_width; mb_x++) {
1995             int xx = mb_x * 16;
1996             int yy = mb_y * 16;
1997             uint8_t *pix = s->new_picture.data[0] + (yy * s->linesize) + xx;
1998             int varc;
1999             int sum = s->dsp.pix_sum(pix, s->linesize);
2000
2001             varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500 + 128)>>8;
2002
2003             s->current_picture.mb_var [s->mb_stride * mb_y + mb_x] = varc;
2004             s->current_picture.mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
2005             s->me.mb_var_sum_temp    += varc;
2006         }
2007     }
2008     return 0;
2009 }
2010
2011 static void write_slice_end(MpegEncContext *s){
2012     if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4){
2013         if(s->partitioned_frame){
2014             ff_mpeg4_merge_partitions(s);
2015         }
2016
2017         ff_mpeg4_stuffing(&s->pb);
2018     }else if(CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG){
2019         ff_mjpeg_encode_stuffing(&s->pb);
2020     }
2021
2022     align_put_bits(&s->pb);
2023     flush_put_bits(&s->pb);
2024
2025     if((s->flags&CODEC_FLAG_PASS1) && !s->partitioned_frame)
2026         s->misc_bits+= get_bits_diff(s);
2027 }
2028
2029 static int encode_thread(AVCodecContext *c, void *arg){
2030     MpegEncContext *s= *(void**)arg;
2031     int mb_x, mb_y, pdif = 0;
2032     int chr_h= 16>>s->chroma_y_shift;
2033     int i, j;
2034     MpegEncContext best_s, backup_s;
2035     uint8_t bit_buf[2][MAX_MB_BYTES];
2036     uint8_t bit_buf2[2][MAX_MB_BYTES];
2037     uint8_t bit_buf_tex[2][MAX_MB_BYTES];
2038     PutBitContext pb[2], pb2[2], tex_pb[2];
2039 //printf("%d->%d\n", s->resync_mb_y, s->end_mb_y);
2040
2041     ff_check_alignment();
2042
2043     for(i=0; i<2; i++){
2044         init_put_bits(&pb    [i], bit_buf    [i], MAX_MB_BYTES);
2045         init_put_bits(&pb2   [i], bit_buf2   [i], MAX_MB_BYTES);
2046         init_put_bits(&tex_pb[i], bit_buf_tex[i], MAX_MB_BYTES);
2047     }
2048
2049     s->last_bits= put_bits_count(&s->pb);
2050     s->mv_bits=0;
2051     s->misc_bits=0;
2052     s->i_tex_bits=0;
2053     s->p_tex_bits=0;
2054     s->i_count=0;
2055     s->f_count=0;
2056     s->b_count=0;
2057     s->skip_count=0;
2058
2059     for(i=0; i<3; i++){
2060         /* init last dc values */
2061         /* note: quant matrix value (8) is implied here */
2062         s->last_dc[i] = 128 << s->intra_dc_precision;
2063
2064         s->current_picture.error[i] = 0;
2065     }
2066     s->mb_skip_run = 0;
2067     memset(s->last_mv, 0, sizeof(s->last_mv));
2068
2069     s->last_mv_dir = 0;
2070
2071     switch(s->codec_id){
2072     case CODEC_ID_H263:
2073     case CODEC_ID_H263P:
2074     case CODEC_ID_FLV1:
2075         if (CONFIG_H263_ENCODER)
2076             s->gob_index = ff_h263_get_gob_height(s);
2077         break;
2078     case CODEC_ID_MPEG4:
2079         if(CONFIG_MPEG4_ENCODER && s->partitioned_frame)
2080             ff_mpeg4_init_partitions(s);
2081         break;
2082     }
2083
2084     s->resync_mb_x=0;
2085     s->resync_mb_y=0;
2086     s->first_slice_line = 1;
2087     s->ptr_lastgob = s->pb.buf;
2088     for(mb_y= s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
2089 //    printf("row %d at %X\n", s->mb_y, (int)s);
2090         s->mb_x=0;
2091         s->mb_y= mb_y;
2092
2093         ff_set_qscale(s, s->qscale);
2094         ff_init_block_index(s);
2095
2096         for(mb_x=0; mb_x < s->mb_width; mb_x++) {
2097             int xy= mb_y*s->mb_stride + mb_x; // removed const, H261 needs to adjust this
2098             int mb_type= s->mb_type[xy];
2099 //            int d;
2100             int dmin= INT_MAX;
2101             int dir;
2102
2103             if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < MAX_MB_BYTES){
2104                 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
2105                 return -1;
2106             }
2107             if(s->data_partitioning){
2108                 if(   s->pb2   .buf_end - s->pb2   .buf - (put_bits_count(&s->    pb2)>>3) < MAX_MB_BYTES
2109                    || s->tex_pb.buf_end - s->tex_pb.buf - (put_bits_count(&s->tex_pb )>>3) < MAX_MB_BYTES){
2110                     av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
2111                     return -1;
2112                 }
2113             }
2114
2115             s->mb_x = mb_x;
2116             s->mb_y = mb_y;  // moved into loop, can get changed by H.261
2117             ff_update_block_index(s);
2118
2119             if(CONFIG_H261_ENCODER && s->codec_id == CODEC_ID_H261){
2120                 ff_h261_reorder_mb_index(s);
2121                 xy= s->mb_y*s->mb_stride + s->mb_x;
2122                 mb_type= s->mb_type[xy];
2123             }
2124
2125             /* write gob / video packet header  */
2126             if(s->rtp_mode){
2127                 int current_packet_size, is_gob_start;
2128
2129                 current_packet_size= ((put_bits_count(&s->pb)+7)>>3) - (s->ptr_lastgob - s->pb.buf);
2130
2131                 is_gob_start= s->avctx->rtp_payload_size && current_packet_size >= s->avctx->rtp_payload_size && mb_y + mb_x>0;
2132
2133                 if(s->start_mb_y == mb_y && mb_y > 0 && mb_x==0) is_gob_start=1;
2134
2135                 switch(s->codec_id){
2136                 case CODEC_ID_H263:
2137                 case CODEC_ID_H263P:
2138                     if(!s->h263_slice_structured)
2139                         if(s->mb_x || s->mb_y%s->gob_index) is_gob_start=0;
2140                     break;
2141                 case CODEC_ID_MPEG2VIDEO:
2142                     if(s->mb_x==0 && s->mb_y!=0) is_gob_start=1;
2143                 case CODEC_ID_MPEG1VIDEO:
2144                     if(s->mb_skip_run) is_gob_start=0;
2145                     break;
2146                 }
2147
2148                 if(is_gob_start){
2149                     if(s->start_mb_y != mb_y || mb_x!=0){
2150                         write_slice_end(s);
2151
2152                         if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4 && s->partitioned_frame){
2153                             ff_mpeg4_init_partitions(s);
2154                         }
2155                     }
2156
2157                     assert((put_bits_count(&s->pb)&7) == 0);
2158                     current_packet_size= put_bits_ptr(&s->pb) - s->ptr_lastgob;
2159
2160                     if(s->avctx->error_rate && s->resync_mb_x + s->resync_mb_y > 0){
2161                         int r= put_bits_count(&s->pb)/8 + s->picture_number + 16 + s->mb_x + s->mb_y;
2162                         int d= 100 / s->avctx->error_rate;
2163                         if(r % d == 0){
2164                             current_packet_size=0;
2165 #ifndef ALT_BITSTREAM_WRITER
2166                             s->pb.buf_ptr= s->ptr_lastgob;
2167 #endif
2168                             assert(put_bits_ptr(&s->pb) == s->ptr_lastgob);
2169                         }
2170                     }
2171
2172                     if (s->avctx->rtp_callback){
2173                         int number_mb = (mb_y - s->resync_mb_y)*s->mb_width + mb_x - s->resync_mb_x;
2174                         s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, current_packet_size, number_mb);
2175                     }
2176
2177                     switch(s->codec_id){
2178                     case CODEC_ID_MPEG4:
2179                         if (CONFIG_MPEG4_ENCODER) {
2180                             ff_mpeg4_encode_video_packet_header(s);
2181                             ff_mpeg4_clean_buffers(s);
2182                         }
2183                     break;
2184                     case CODEC_ID_MPEG1VIDEO:
2185                     case CODEC_ID_MPEG2VIDEO:
2186                         if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER) {
2187                             ff_mpeg1_encode_slice_header(s);
2188                             ff_mpeg1_clean_buffers(s);
2189                         }
2190                     break;
2191                     case CODEC_ID_H263:
2192                     case CODEC_ID_H263P:
2193                         if (CONFIG_H263_ENCODER)
2194                             h263_encode_gob_header(s, mb_y);
2195                     break;
2196                     }
2197
2198                     if(s->flags&CODEC_FLAG_PASS1){
2199                         int bits= put_bits_count(&s->pb);
2200                         s->misc_bits+= bits - s->last_bits;
2201                         s->last_bits= bits;
2202                     }
2203
2204                     s->ptr_lastgob += current_packet_size;
2205                     s->first_slice_line=1;
2206                     s->resync_mb_x=mb_x;
2207                     s->resync_mb_y=mb_y;
2208                 }
2209             }
2210
2211             if(  (s->resync_mb_x   == s->mb_x)
2212                && s->resync_mb_y+1 == s->mb_y){
2213                 s->first_slice_line=0;
2214             }
2215
2216             s->mb_skipped=0;
2217             s->dquant=0; //only for QP_RD
2218
2219             if(mb_type & (mb_type-1) || (s->flags & CODEC_FLAG_QP_RD)){ // more than 1 MB type possible or CODEC_FLAG_QP_RD
2220                 int next_block=0;
2221                 int pb_bits_count, pb2_bits_count, tex_pb_bits_count;
2222
2223                 copy_context_before_encode(&backup_s, s, -1);
2224                 backup_s.pb= s->pb;
2225                 best_s.data_partitioning= s->data_partitioning;
2226                 best_s.partitioned_frame= s->partitioned_frame;
2227                 if(s->data_partitioning){
2228                     backup_s.pb2= s->pb2;
2229                     backup_s.tex_pb= s->tex_pb;
2230                 }
2231
2232                 if(mb_type&CANDIDATE_MB_TYPE_INTER){
2233                     s->mv_dir = MV_DIR_FORWARD;
2234                     s->mv_type = MV_TYPE_16X16;
2235                     s->mb_intra= 0;
2236                     s->mv[0][0][0] = s->p_mv_table[xy][0];
2237                     s->mv[0][0][1] = s->p_mv_table[xy][1];
2238                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER, pb, pb2, tex_pb,
2239                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2240                 }
2241                 if(mb_type&CANDIDATE_MB_TYPE_INTER_I){
2242                     s->mv_dir = MV_DIR_FORWARD;
2243                     s->mv_type = MV_TYPE_FIELD;
2244                     s->mb_intra= 0;
2245                     for(i=0; i<2; i++){
2246                         j= s->field_select[0][i] = s->p_field_select_table[i][xy];
2247                         s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
2248                         s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
2249                     }
2250                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER_I, pb, pb2, tex_pb,
2251                                  &dmin, &next_block, 0, 0);
2252                 }
2253                 if(mb_type&CANDIDATE_MB_TYPE_SKIPPED){
2254                     s->mv_dir = MV_DIR_FORWARD;
2255                     s->mv_type = MV_TYPE_16X16;
2256                     s->mb_intra= 0;
2257                     s->mv[0][0][0] = 0;
2258                     s->mv[0][0][1] = 0;
2259                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_SKIPPED, pb, pb2, tex_pb,
2260                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2261                 }
2262                 if(mb_type&CANDIDATE_MB_TYPE_INTER4V){
2263                     s->mv_dir = MV_DIR_FORWARD;
2264                     s->mv_type = MV_TYPE_8X8;
2265                     s->mb_intra= 0;
2266                     for(i=0; i<4; i++){
2267                         s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
2268                         s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
2269                     }
2270                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER4V, pb, pb2, tex_pb,
2271                                  &dmin, &next_block, 0, 0);
2272                 }
2273                 if(mb_type&CANDIDATE_MB_TYPE_FORWARD){
2274                     s->mv_dir = MV_DIR_FORWARD;
2275                     s->mv_type = MV_TYPE_16X16;
2276                     s->mb_intra= 0;
2277                     s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
2278                     s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2279                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD, pb, pb2, tex_pb,
2280                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2281                 }
2282                 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD){
2283                     s->mv_dir = MV_DIR_BACKWARD;
2284                     s->mv_type = MV_TYPE_16X16;
2285                     s->mb_intra= 0;
2286                     s->mv[1][0][0] = s->b_back_mv_table[xy][0];
2287                     s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2288                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD, pb, pb2, tex_pb,
2289                                  &dmin, &next_block, s->mv[1][0][0], s->mv[1][0][1]);
2290                 }
2291                 if(mb_type&CANDIDATE_MB_TYPE_BIDIR){
2292                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2293                     s->mv_type = MV_TYPE_16X16;
2294                     s->mb_intra= 0;
2295                     s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
2296                     s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
2297                     s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
2298                     s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2299                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR, pb, pb2, tex_pb,
2300                                  &dmin, &next_block, 0, 0);
2301                 }
2302                 if(mb_type&CANDIDATE_MB_TYPE_FORWARD_I){
2303                     s->mv_dir = MV_DIR_FORWARD;
2304                     s->mv_type = MV_TYPE_FIELD;
2305                     s->mb_intra= 0;
2306                     for(i=0; i<2; i++){
2307                         j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
2308                         s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
2309                         s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
2310                     }
2311                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD_I, pb, pb2, tex_pb,
2312                                  &dmin, &next_block, 0, 0);
2313                 }
2314                 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD_I){
2315                     s->mv_dir = MV_DIR_BACKWARD;
2316                     s->mv_type = MV_TYPE_FIELD;
2317                     s->mb_intra= 0;
2318                     for(i=0; i<2; i++){
2319                         j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
2320                         s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
2321                         s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
2322                     }
2323                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD_I, pb, pb2, tex_pb,
2324                                  &dmin, &next_block, 0, 0);
2325                 }
2326                 if(mb_type&CANDIDATE_MB_TYPE_BIDIR_I){
2327                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2328                     s->mv_type = MV_TYPE_FIELD;
2329                     s->mb_intra= 0;
2330                     for(dir=0; dir<2; dir++){
2331                         for(i=0; i<2; i++){
2332                             j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
2333                             s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
2334                             s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
2335                         }
2336                     }
2337                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR_I, pb, pb2, tex_pb,
2338                                  &dmin, &next_block, 0, 0);
2339                 }
2340                 if(mb_type&CANDIDATE_MB_TYPE_INTRA){
2341                     s->mv_dir = 0;
2342                     s->mv_type = MV_TYPE_16X16;
2343                     s->mb_intra= 1;
2344                     s->mv[0][0][0] = 0;
2345                     s->mv[0][0][1] = 0;
2346                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTRA, pb, pb2, tex_pb,
2347                                  &dmin, &next_block, 0, 0);
2348                     if(s->h263_pred || s->h263_aic){
2349                         if(best_s.mb_intra)
2350                             s->mbintra_table[mb_x + mb_y*s->mb_stride]=1;
2351                         else
2352                             ff_clean_intra_table_entries(s); //old mode?
2353                     }
2354                 }
2355
2356                 if((s->flags & CODEC_FLAG_QP_RD) && dmin < INT_MAX){
2357                     if(best_s.mv_type==MV_TYPE_16X16){ //FIXME move 4mv after QPRD
2358                         const int last_qp= backup_s.qscale;
2359                         int qpi, qp, dc[6];
2360                         DCTELEM ac[6][16];
2361                         const int mvdir= (best_s.mv_dir&MV_DIR_BACKWARD) ? 1 : 0;
2362                         static const int dquant_tab[4]={-1,1,-2,2};
2363
2364                         assert(backup_s.dquant == 0);
2365
2366                         //FIXME intra
2367                         s->mv_dir= best_s.mv_dir;
2368                         s->mv_type = MV_TYPE_16X16;
2369                         s->mb_intra= best_s.mb_intra;
2370                         s->mv[0][0][0] = best_s.mv[0][0][0];
2371                         s->mv[0][0][1] = best_s.mv[0][0][1];
2372                         s->mv[1][0][0] = best_s.mv[1][0][0];
2373                         s->mv[1][0][1] = best_s.mv[1][0][1];
2374
2375                         qpi = s->pict_type == FF_B_TYPE ? 2 : 0;
2376                         for(; qpi<4; qpi++){
2377                             int dquant= dquant_tab[qpi];
2378                             qp= last_qp + dquant;
2379                             if(qp < s->avctx->qmin || qp > s->avctx->qmax)
2380                                 continue;
2381                             backup_s.dquant= dquant;
2382                             if(s->mb_intra && s->dc_val[0]){
2383                                 for(i=0; i<6; i++){
2384                                     dc[i]= s->dc_val[0][ s->block_index[i] ];
2385                                     memcpy(ac[i], s->ac_val[0][s->block_index[i]], sizeof(DCTELEM)*16);
2386                                 }
2387                             }
2388
2389                             encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
2390                                          &dmin, &next_block, s->mv[mvdir][0][0], s->mv[mvdir][0][1]);
2391                             if(best_s.qscale != qp){
2392                                 if(s->mb_intra && s->dc_val[0]){
2393                                     for(i=0; i<6; i++){
2394                                         s->dc_val[0][ s->block_index[i] ]= dc[i];
2395                                         memcpy(s->ac_val[0][s->block_index[i]], ac[i], sizeof(DCTELEM)*16);
2396                                     }
2397                                 }
2398                             }
2399                         }
2400                     }
2401                 }
2402                 if(CONFIG_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT){
2403                     int mx= s->b_direct_mv_table[xy][0];
2404                     int my= s->b_direct_mv_table[xy][1];
2405
2406                     backup_s.dquant = 0;
2407                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2408                     s->mb_intra= 0;
2409                     ff_mpeg4_set_direct_mv(s, mx, my);
2410                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
2411                                  &dmin, &next_block, mx, my);
2412                 }
2413                 if(CONFIG_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT0){
2414                     backup_s.dquant = 0;
2415                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2416                     s->mb_intra= 0;
2417                     ff_mpeg4_set_direct_mv(s, 0, 0);
2418                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
2419                                  &dmin, &next_block, 0, 0);
2420                 }
2421                 if(!best_s.mb_intra && s->flags2&CODEC_FLAG2_SKIP_RD){
2422                     int coded=0;
2423                     for(i=0; i<6; i++)
2424                         coded |= s->block_last_index[i];
2425                     if(coded){
2426                         int mx,my;
2427                         memcpy(s->mv, best_s.mv, sizeof(s->mv));
2428                         if(CONFIG_MPEG4_ENCODER && best_s.mv_dir & MV_DIRECT){
2429                             mx=my=0; //FIXME find the one we actually used
2430                             ff_mpeg4_set_direct_mv(s, mx, my);
2431                         }else if(best_s.mv_dir&MV_DIR_BACKWARD){
2432                             mx= s->mv[1][0][0];
2433                             my= s->mv[1][0][1];
2434                         }else{
2435                             mx= s->mv[0][0][0];
2436                             my= s->mv[0][0][1];
2437                         }
2438
2439                         s->mv_dir= best_s.mv_dir;
2440                         s->mv_type = best_s.mv_type;
2441                         s->mb_intra= 0;
2442 /*                        s->mv[0][0][0] = best_s.mv[0][0][0];
2443                         s->mv[0][0][1] = best_s.mv[0][0][1];
2444                         s->mv[1][0][0] = best_s.mv[1][0][0];
2445                         s->mv[1][0][1] = best_s.mv[1][0][1];*/
2446                         backup_s.dquant= 0;
2447                         s->skipdct=1;
2448                         encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
2449                                         &dmin, &next_block, mx, my);
2450                         s->skipdct=0;
2451                     }
2452                 }
2453
2454                 s->current_picture.qscale_table[xy]= best_s.qscale;
2455
2456                 copy_context_after_encode(s, &best_s, -1);
2457
2458                 pb_bits_count= put_bits_count(&s->pb);
2459                 flush_put_bits(&s->pb);
2460                 ff_copy_bits(&backup_s.pb, bit_buf[next_block^1], pb_bits_count);
2461                 s->pb= backup_s.pb;
2462
2463                 if(s->data_partitioning){
2464                     pb2_bits_count= put_bits_count(&s->pb2);
2465                     flush_put_bits(&s->pb2);
2466                     ff_copy_bits(&backup_s.pb2, bit_buf2[next_block^1], pb2_bits_count);
2467                     s->pb2= backup_s.pb2;
2468
2469                     tex_pb_bits_count= put_bits_count(&s->tex_pb);
2470                     flush_put_bits(&s->tex_pb);
2471                     ff_copy_bits(&backup_s.tex_pb, bit_buf_tex[next_block^1], tex_pb_bits_count);
2472                     s->tex_pb= backup_s.tex_pb;
2473                 }
2474                 s->last_bits= put_bits_count(&s->pb);
2475
2476                 if (CONFIG_H263_ENCODER &&
2477                     s->out_format == FMT_H263 && s->pict_type!=FF_B_TYPE)
2478                     ff_h263_update_motion_val(s);
2479
2480                 if(next_block==0){ //FIXME 16 vs linesize16
2481                     s->dsp.put_pixels_tab[0][0](s->dest[0], s->rd_scratchpad                     , s->linesize  ,16);
2482                     s->dsp.put_pixels_tab[1][0](s->dest[1], s->rd_scratchpad + 16*s->linesize    , s->uvlinesize, 8);
2483                     s->dsp.put_pixels_tab[1][0](s->dest[2], s->rd_scratchpad + 16*s->linesize + 8, s->uvlinesize, 8);
2484                 }
2485
2486                 if(s->avctx->mb_decision == FF_MB_DECISION_BITS)
2487                     MPV_decode_mb(s, s->block);
2488             } else {
2489                 int motion_x = 0, motion_y = 0;
2490                 s->mv_type=MV_TYPE_16X16;
2491                 // only one MB-Type possible
2492
2493                 switch(mb_type){
2494                 case CANDIDATE_MB_TYPE_INTRA:
2495                     s->mv_dir = 0;
2496                     s->mb_intra= 1;
2497                     motion_x= s->mv[0][0][0] = 0;
2498                     motion_y= s->mv[0][0][1] = 0;
2499                     break;
2500                 case CANDIDATE_MB_TYPE_INTER:
2501                     s->mv_dir = MV_DIR_FORWARD;
2502                     s->mb_intra= 0;
2503                     motion_x= s->mv[0][0][0] = s->p_mv_table[xy][0];
2504                     motion_y= s->mv[0][0][1] = s->p_mv_table[xy][1];
2505                     break;
2506                 case CANDIDATE_MB_TYPE_INTER_I:
2507                     s->mv_dir = MV_DIR_FORWARD;
2508                     s->mv_type = MV_TYPE_FIELD;
2509                     s->mb_intra= 0;
2510                     for(i=0; i<2; i++){
2511                         j= s->field_select[0][i] = s->p_field_select_table[i][xy];
2512                         s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
2513                         s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
2514                     }
2515                     break;
2516                 case CANDIDATE_MB_TYPE_INTER4V:
2517                     s->mv_dir = MV_DIR_FORWARD;
2518                     s->mv_type = MV_TYPE_8X8;
2519                     s->mb_intra= 0;
2520                     for(i=0; i<4; i++){
2521                         s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
2522                         s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
2523                     }
2524                     break;
2525                 case CANDIDATE_MB_TYPE_DIRECT:
2526                     if (CONFIG_MPEG4_ENCODER) {
2527                         s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
2528                         s->mb_intra= 0;
2529                         motion_x=s->b_direct_mv_table[xy][0];
2530                         motion_y=s->b_direct_mv_table[xy][1];
2531                         ff_mpeg4_set_direct_mv(s, motion_x, motion_y);
2532                     }
2533                     break;
2534                 case CANDIDATE_MB_TYPE_DIRECT0:
2535                     if (CONFIG_MPEG4_ENCODER) {
2536                         s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
2537                         s->mb_intra= 0;
2538                         ff_mpeg4_set_direct_mv(s, 0, 0);
2539                     }
2540                     break;
2541                 case CANDIDATE_MB_TYPE_BIDIR:
2542                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2543                     s->mb_intra= 0;
2544                     s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
2545                     s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
2546                     s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
2547                     s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2548                     break;
2549                 case CANDIDATE_MB_TYPE_BACKWARD:
2550                     s->mv_dir = MV_DIR_BACKWARD;
2551                     s->mb_intra= 0;
2552                     motion_x= s->mv[1][0][0] = s->b_back_mv_table[xy][0];
2553                     motion_y= s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2554                     break;
2555                 case CANDIDATE_MB_TYPE_FORWARD:
2556                     s->mv_dir = MV_DIR_FORWARD;
2557                     s->mb_intra= 0;
2558                     motion_x= s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
2559                     motion_y= s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2560 //                    printf(" %d %d ", motion_x, motion_y);
2561                     break;
2562                 case CANDIDATE_MB_TYPE_FORWARD_I:
2563                     s->mv_dir = MV_DIR_FORWARD;
2564                     s->mv_type = MV_TYPE_FIELD;
2565                     s->mb_intra= 0;
2566                     for(i=0; i<2; i++){
2567                         j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
2568                         s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
2569                         s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
2570                     }
2571                     break;
2572                 case CANDIDATE_MB_TYPE_BACKWARD_I:
2573                     s->mv_dir = MV_DIR_BACKWARD;
2574                     s->mv_type = MV_TYPE_FIELD;
2575                     s->mb_intra= 0;
2576                     for(i=0; i<2; i++){
2577                         j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
2578                         s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
2579                         s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
2580                     }
2581                     break;
2582                 case CANDIDATE_MB_TYPE_BIDIR_I:
2583                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2584                     s->mv_type = MV_TYPE_FIELD;
2585                     s->mb_intra= 0;
2586                     for(dir=0; dir<2; dir++){
2587                         for(i=0; i<2; i++){
2588                             j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
2589                             s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
2590                             s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
2591                         }
2592                     }
2593                     break;
2594                 default:
2595                     av_log(s->avctx, AV_LOG_ERROR, "illegal MB type\n");
2596                 }
2597
2598                 encode_mb(s, motion_x, motion_y);
2599
2600                 // RAL: Update last macroblock type
2601                 s->last_mv_dir = s->mv_dir;
2602
2603                 if (CONFIG_H263_ENCODER &&
2604                     s->out_format == FMT_H263 && s->pict_type!=FF_B_TYPE)
2605                     ff_h263_update_motion_val(s);
2606
2607                 MPV_decode_mb(s, s->block);
2608             }
2609
2610             /* clean the MV table in IPS frames for direct mode in B frames */
2611             if(s->mb_intra /* && I,P,S_TYPE */){
2612                 s->p_mv_table[xy][0]=0;
2613                 s->p_mv_table[xy][1]=0;
2614             }
2615
2616             if(s->flags&CODEC_FLAG_PSNR){
2617                 int w= 16;
2618                 int h= 16;
2619
2620                 if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
2621                 if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
2622
2623                 s->current_picture.error[0] += sse(
2624                     s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16,
2625                     s->dest[0], w, h, s->linesize);
2626                 s->current_picture.error[1] += sse(
2627                     s, s->new_picture.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*chr_h,
2628                     s->dest[1], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
2629                 s->current_picture.error[2] += sse(
2630                     s, s->new_picture.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*chr_h,
2631                     s->dest[2], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
2632             }
2633             if(s->loop_filter){
2634                 if(CONFIG_H263_ENCODER && s->out_format == FMT_H263)
2635                     ff_h263_loop_filter(s);
2636             }
2637 //printf("MB %d %d bits\n", s->mb_x+s->mb_y*s->mb_stride, put_bits_count(&s->pb));
2638         }
2639     }
2640
2641     //not beautiful here but we must write it before flushing so it has to be here
2642     if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version && s->msmpeg4_version<4 && s->pict_type == FF_I_TYPE)
2643         msmpeg4_encode_ext_header(s);
2644
2645     write_slice_end(s);
2646
2647     /* Send the last GOB if RTP */
2648     if (s->avctx->rtp_callback) {
2649         int number_mb = (mb_y - s->resync_mb_y)*s->mb_width - s->resync_mb_x;
2650         pdif = put_bits_ptr(&s->pb) - s->ptr_lastgob;
2651         /* Call the RTP callback to send the last GOB */
2652         emms_c();
2653         s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, pdif, number_mb);
2654     }
2655
2656     return 0;
2657 }
2658
2659 #define MERGE(field) dst->field += src->field; src->field=0
2660 static void merge_context_after_me(MpegEncContext *dst, MpegEncContext *src){
2661     MERGE(me.scene_change_score);
2662     MERGE(me.mc_mb_var_sum_temp);
2663     MERGE(me.mb_var_sum_temp);
2664 }
2665
2666 static void merge_context_after_encode(MpegEncContext *dst, MpegEncContext *src){
2667     int i;
2668
2669     MERGE(dct_count[0]); //note, the other dct vars are not part of the context
2670     MERGE(dct_count[1]);
2671     MERGE(mv_bits);
2672     MERGE(i_tex_bits);
2673     MERGE(p_tex_bits);
2674     MERGE(i_count);
2675     MERGE(f_count);
2676     MERGE(b_count);
2677     MERGE(skip_count);
2678     MERGE(misc_bits);
2679     MERGE(error_count);
2680     MERGE(padding_bug_score);
2681     MERGE(current_picture.error[0]);
2682     MERGE(current_picture.error[1]);
2683     MERGE(current_picture.error[2]);
2684
2685     if(dst->avctx->noise_reduction){
2686         for(i=0; i<64; i++){
2687             MERGE(dct_error_sum[0][i]);
2688             MERGE(dct_error_sum[1][i]);
2689         }
2690     }
2691
2692     assert(put_bits_count(&src->pb) % 8 ==0);
2693     assert(put_bits_count(&dst->pb) % 8 ==0);
2694     ff_copy_bits(&dst->pb, src->pb.buf, put_bits_count(&src->pb));
2695     flush_put_bits(&dst->pb);
2696 }
2697
2698 static int estimate_qp(MpegEncContext *s, int dry_run){
2699     if (s->next_lambda){
2700         s->current_picture_ptr->quality=
2701         s->current_picture.quality = s->next_lambda;
2702         if(!dry_run) s->next_lambda= 0;
2703     } else if (!s->fixed_qscale) {
2704         s->current_picture_ptr->quality=
2705         s->current_picture.quality = ff_rate_estimate_qscale(s, dry_run);
2706         if (s->current_picture.quality < 0)
2707             return -1;
2708     }
2709
2710     if(s->adaptive_quant){
2711         switch(s->codec_id){
2712         case CODEC_ID_MPEG4:
2713             if (CONFIG_MPEG4_ENCODER)
2714                 ff_clean_mpeg4_qscales(s);
2715             break;
2716         case CODEC_ID_H263:
2717         case CODEC_ID_H263P:
2718         case CODEC_ID_FLV1:
2719             if (CONFIG_H263_ENCODER)
2720                 ff_clean_h263_qscales(s);
2721             break;
2722         default:
2723             ff_init_qscale_tab(s);
2724         }
2725
2726         s->lambda= s->lambda_table[0];
2727         //FIXME broken
2728     }else
2729         s->lambda= s->current_picture.quality;
2730 //printf("%d %d\n", s->avctx->global_quality, s->current_picture.quality);
2731     update_qscale(s);
2732     return 0;
2733 }
2734
2735 /* must be called before writing the header */
2736 static void set_frame_distances(MpegEncContext * s){
2737     assert(s->current_picture_ptr->pts != AV_NOPTS_VALUE);
2738     s->time= s->current_picture_ptr->pts*s->avctx->time_base.num;
2739
2740     if(s->pict_type==FF_B_TYPE){
2741         s->pb_time= s->pp_time - (s->last_non_b_time - s->time);
2742         assert(s->pb_time > 0 && s->pb_time < s->pp_time);
2743     }else{
2744         s->pp_time= s->time - s->last_non_b_time;
2745         s->last_non_b_time= s->time;
2746         assert(s->picture_number==0 || s->pp_time > 0);
2747     }
2748 }
2749
2750 static int encode_picture(MpegEncContext *s, int picture_number)
2751 {
2752     int i;
2753     int bits;
2754
2755     s->picture_number = picture_number;
2756
2757     /* Reset the average MB variance */
2758     s->me.mb_var_sum_temp    =
2759     s->me.mc_mb_var_sum_temp = 0;
2760
2761     /* we need to initialize some time vars before we can encode b-frames */
2762     // RAL: Condition added for MPEG1VIDEO
2763     if (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO || (s->h263_pred && !s->h263_msmpeg4))
2764         set_frame_distances(s);
2765     if(CONFIG_MPEG4_ENCODER && s->codec_id == CODEC_ID_MPEG4)
2766         ff_set_mpeg4_time(s);
2767
2768     s->me.scene_change_score=0;
2769
2770 //    s->lambda= s->current_picture_ptr->quality; //FIXME qscale / ... stuff for ME rate distortion
2771
2772     if(s->pict_type==FF_I_TYPE){
2773         if(s->msmpeg4_version >= 3) s->no_rounding=1;
2774         else                        s->no_rounding=0;
2775     }else if(s->pict_type!=FF_B_TYPE){
2776         if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
2777             s->no_rounding ^= 1;
2778     }
2779
2780     if(s->flags & CODEC_FLAG_PASS2){
2781         if (estimate_qp(s,1) < 0)
2782             return -1;
2783         ff_get_2pass_fcode(s);
2784     }else if(!(s->flags & CODEC_FLAG_QSCALE)){
2785         if(s->pict_type==FF_B_TYPE)
2786             s->lambda= s->last_lambda_for[s->pict_type];
2787         else
2788             s->lambda= s->last_lambda_for[s->last_non_b_pict_type];
2789         update_qscale(s);
2790     }
2791
2792     s->mb_intra=0; //for the rate distortion & bit compare functions
2793     for(i=1; i<s->avctx->thread_count; i++){
2794         ff_update_duplicate_context(s->thread_context[i], s);
2795     }
2796
2797     if(ff_init_me(s)<0)
2798         return -1;
2799
2800     /* Estimate motion for every MB */
2801     if(s->pict_type != FF_I_TYPE){
2802         s->lambda = (s->lambda * s->avctx->me_penalty_compensation + 128)>>8;
2803         s->lambda2= (s->lambda2* (int64_t)s->avctx->me_penalty_compensation + 128)>>8;
2804         if(s->pict_type != FF_B_TYPE && s->avctx->me_threshold==0){
2805             if((s->avctx->pre_me && s->last_non_b_pict_type==FF_I_TYPE) || s->avctx->pre_me==2){
2806                 s->avctx->execute(s->avctx, pre_estimate_motion_thread, &s->thread_context[0], NULL, s->avctx->thread_count, sizeof(void*));
2807             }
2808         }
2809
2810         s->avctx->execute(s->avctx, estimate_motion_thread, &s->thread_context[0], NULL, s->avctx->thread_count, sizeof(void*));
2811     }else /* if(s->pict_type == FF_I_TYPE) */{
2812         /* I-Frame */
2813         for(i=0; i<s->mb_stride*s->mb_height; i++)
2814             s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
2815
2816         if(!s->fixed_qscale){
2817             /* finding spatial complexity for I-frame rate control */
2818             s->avctx->execute(s->avctx, mb_var_thread, &s->thread_context[0], NULL, s->avctx->thread_count, sizeof(void*));
2819         }
2820     }
2821     for(i=1; i<s->avctx->thread_count; i++){
2822         merge_context_after_me(s, s->thread_context[i]);
2823     }
2824     s->current_picture.mc_mb_var_sum= s->current_picture_ptr->mc_mb_var_sum= s->me.mc_mb_var_sum_temp;
2825     s->current_picture.   mb_var_sum= s->current_picture_ptr->   mb_var_sum= s->me.   mb_var_sum_temp;
2826     emms_c();
2827
2828     if(s->me.scene_change_score > s->avctx->scenechange_threshold && s->pict_type == FF_P_TYPE){
2829         s->pict_type= FF_I_TYPE;
2830         for(i=0; i<s->mb_stride*s->mb_height; i++)
2831             s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
2832 //printf("Scene change detected, encoding as I Frame %d %d\n", s->current_picture.mb_var_sum, s->current_picture.mc_mb_var_sum);
2833     }
2834
2835     if(!s->umvplus){
2836         if(s->pict_type==FF_P_TYPE || s->pict_type==FF_S_TYPE) {
2837             s->f_code= ff_get_best_fcode(s, s->p_mv_table, CANDIDATE_MB_TYPE_INTER);
2838
2839             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2840                 int a,b;
2841                 a= ff_get_best_fcode(s, s->p_field_mv_table[0][0], CANDIDATE_MB_TYPE_INTER_I); //FIXME field_select
2842                 b= ff_get_best_fcode(s, s->p_field_mv_table[1][1], CANDIDATE_MB_TYPE_INTER_I);
2843                 s->f_code= FFMAX3(s->f_code, a, b);
2844             }
2845
2846             ff_fix_long_p_mvs(s);
2847             ff_fix_long_mvs(s, NULL, 0, s->p_mv_table, s->f_code, CANDIDATE_MB_TYPE_INTER, 0);
2848             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2849                 int j;
2850                 for(i=0; i<2; i++){
2851                     for(j=0; j<2; j++)
2852                         ff_fix_long_mvs(s, s->p_field_select_table[i], j,
2853                                         s->p_field_mv_table[i][j], s->f_code, CANDIDATE_MB_TYPE_INTER_I, 0);
2854                 }
2855             }
2856         }
2857
2858         if(s->pict_type==FF_B_TYPE){
2859             int a, b;
2860
2861             a = ff_get_best_fcode(s, s->b_forw_mv_table, CANDIDATE_MB_TYPE_FORWARD);
2862             b = ff_get_best_fcode(s, s->b_bidir_forw_mv_table, CANDIDATE_MB_TYPE_BIDIR);
2863             s->f_code = FFMAX(a, b);
2864
2865             a = ff_get_best_fcode(s, s->b_back_mv_table, CANDIDATE_MB_TYPE_BACKWARD);
2866             b = ff_get_best_fcode(s, s->b_bidir_back_mv_table, CANDIDATE_MB_TYPE_BIDIR);
2867             s->b_code = FFMAX(a, b);
2868
2869             ff_fix_long_mvs(s, NULL, 0, s->b_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_FORWARD, 1);
2870             ff_fix_long_mvs(s, NULL, 0, s->b_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BACKWARD, 1);
2871             ff_fix_long_mvs(s, NULL, 0, s->b_bidir_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_BIDIR, 1);
2872             ff_fix_long_mvs(s, NULL, 0, s->b_bidir_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BIDIR, 1);
2873             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2874                 int dir, j;
2875                 for(dir=0; dir<2; dir++){
2876                     for(i=0; i<2; i++){
2877                         for(j=0; j<2; j++){
2878                             int type= dir ? (CANDIDATE_MB_TYPE_BACKWARD_I|CANDIDATE_MB_TYPE_BIDIR_I)
2879                                           : (CANDIDATE_MB_TYPE_FORWARD_I |CANDIDATE_MB_TYPE_BIDIR_I);
2880                             ff_fix_long_mvs(s, s->b_field_select_table[dir][i], j,
2881                                             s->b_field_mv_table[dir][i][j], dir ? s->b_code : s->f_code, type, 1);
2882                         }
2883                     }
2884                 }
2885             }
2886         }
2887     }
2888
2889     if (estimate_qp(s, 0) < 0)
2890         return -1;
2891
2892     if(s->qscale < 3 && s->max_qcoeff<=128 && s->pict_type==FF_I_TYPE && !(s->flags & CODEC_FLAG_QSCALE))
2893         s->qscale= 3; //reduce clipping problems
2894
2895     if (s->out_format == FMT_MJPEG) {
2896         /* for mjpeg, we do include qscale in the matrix */
2897         for(i=1;i<64;i++){
2898             int j= s->dsp.idct_permutation[i];
2899
2900             s->intra_matrix[j] = av_clip_uint8((ff_mpeg1_default_intra_matrix[i] * s->qscale) >> 3);
2901         }
2902         s->y_dc_scale_table=
2903         s->c_dc_scale_table= mpeg2_dc_scale_table[s->intra_dc_precision];
2904         s->intra_matrix[0] = mpeg2_dc_scale_table[s->intra_dc_precision][8];
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     LOCAL_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 };