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 mpegvideo_enc.c
  27  * The simplest mpeg encoder (well, it was the simplest!).
  28  */
  29
  30 #include "avcodec.h"
  31 #include "dsputil.h"
  32 #include "mpegvideo.h"
  33 #include "mpegvideo_common.h"
  34 #include "mjpegenc.h"
  35 #include "msmpeg4.h"
  36 #include "h263.h"
  37 #include "faandct.h"
  38 #include <limits.h>
  39
  40 //#undef NDEBUG
  41 //#include <assert.h>
  42
  43 static int encode_picture(MpegEncContext *s, int picture_number);
  44 static int dct_quantize_refine(MpegEncContext *s, DCTELEM *block, int16_t *weight, DCTELEM *orig, int n, int qscale);
  45 static int sse_mb(MpegEncContext *s);
  46
  47 /* enable all paranoid tests for rounding, overflows, etc... */
  48 //#define PARANOID
  49
  50 //#define DEBUG
  51
  52 static const uint16_t aanscales[64] = {
  53     /* precomputed values scaled up by 14 bits */
  54     16384, 22725, 21407, 19266, 16384, 12873,  8867,  4520,
  55     22725, 31521, 29692, 26722, 22725, 17855, 12299,  6270,
  56     21407, 29692, 27969, 25172, 21407, 16819, 11585,  5906,
  57     19266, 26722, 25172, 22654, 19266, 15137, 10426,  5315,
  58     16384, 22725, 21407, 19266, 16384, 12873,  8867,  4520,
  59     12873, 17855, 16819, 15137, 12873, 10114,  6967,  3552,
  60     8867 , 12299, 11585, 10426,  8867,  6967,  4799,  2446,
  61     4520 ,  6270,  5906,  5315,  4520,  3552,  2446,  1247
  62 };
  63
  64 static uint8_t default_mv_penalty[MAX_FCODE+1][MAX_MV*2+1];
  65 static uint8_t default_fcode_tab[MAX_MV*2+1];
  66
  67 void ff_convert_matrix(DSPContext *dsp, int (*qmat)[64], uint16_t (*qmat16)[2][64],
  68                            const uint16_t *quant_matrix, int bias, int qmin, int qmax, int intra)
  69 {
  70     int qscale;
  71     int shift=0;
  72
  73     for(qscale=qmin; qscale<=qmax; qscale++){
  74         int i;
  75         if (dsp->fdct == ff_jpeg_fdct_islow
  76 #ifdef FAAN_POSTSCALE
  77             || dsp->fdct == ff_faandct
  78 #endif
  79             ) {
  80             for(i=0;i<64;i++) {
  81                 const int j= dsp->idct_permutation[i];
  82                 /* 16 <= qscale * quant_matrix[i] <= 7905 */
  83                 /* 19952         <= aanscales[i] * qscale * quant_matrix[i]           <= 249205026 */
  84                 /* (1<<36)/19952 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */
  85                 /* 3444240       >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= 275 */
  86
  87                 qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) /
  88                                 (qscale * quant_matrix[j]));
  89             }
  90         } else if (dsp->fdct == fdct_ifast
  91 #ifndef FAAN_POSTSCALE
  92                    || dsp->fdct == ff_faandct
  93 #endif
  94                    ) {
  95             for(i=0;i<64;i++) {
  96                 const int j= dsp->idct_permutation[i];
  97                 /* 16 <= qscale * quant_matrix[i] <= 7905 */
  98                 /* 19952         <= aanscales[i] * qscale * quant_matrix[i]           <= 249205026 */
  99                 /* (1<<36)/19952 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */
 100                 /* 3444240       >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= 275 */
 101
 102                 qmat[qscale][i] = (int)((UINT64_C(1) << (QMAT_SHIFT + 14)) /
 103                                 (aanscales[i] * qscale * quant_matrix[j]));
 104             }
 105         } else {
 106             for(i=0;i<64;i++) {
 107                 const int j= dsp->idct_permutation[i];
 108                 /* We can safely suppose that 16 <= quant_matrix[i] <= 255
 109                    So 16           <= qscale * quant_matrix[i]             <= 7905
 110                    so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905
 111                    so 32768        >= (1<<19) / (qscale * quant_matrix[i]) >= 67
 112                 */
 113                 qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) / (qscale * quant_matrix[j]));
 114 //                qmat  [qscale][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]);
 115                 qmat16[qscale][0][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[j]);
 116
 117                 if(qmat16[qscale][0][i]==0 || qmat16[qscale][0][i]==128*256) qmat16[qscale][0][i]=128*256-1;
 118                 qmat16[qscale][1][i]= ROUNDED_DIV(bias<<(16-QUANT_BIAS_SHIFT), qmat16[qscale][0][i]);
 119             }
 120         }
 121
 122         for(i=intra; i<64; i++){
 123             int64_t max= 8191;
 124             if (dsp->fdct == fdct_ifast
 125 #ifndef FAAN_POSTSCALE
 126                    || dsp->fdct == ff_faandct
 127 #endif
 128                    ) {
 129                 max= (8191LL*aanscales[i]) >> 14;
 130             }
 131             while(((max * qmat[qscale][i]) >> shift) > INT_MAX){
 132                 shift++;
 133             }
 134         }
 135     }
 136     if(shift){
 137         av_log(NULL, AV_LOG_INFO, "Warning, QMAT_SHIFT is larger than %d, overflows possible\n", QMAT_SHIFT - shift);
 138     }
 139 }
 140
 141 static inline void update_qscale(MpegEncContext *s){
 142     s->qscale= (s->lambda*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7);
 143     s->qscale= av_clip(s->qscale, s->avctx->qmin, s->avctx->qmax);
 144
 145     s->lambda2= (s->lambda*s->lambda + FF_LAMBDA_SCALE/2) >> FF_LAMBDA_SHIFT;
 146 }
 147
 148 void ff_write_quant_matrix(PutBitContext *pb, uint16_t *matrix){
 149     int i;
 150
 151     if(matrix){
 152         put_bits(pb, 1, 1);
 153         for(i=0;i<64;i++) {
 154             put_bits(pb, 8, matrix[ ff_zigzag_direct[i] ]);
 155         }
 156     }else
 157         put_bits(pb, 1, 0);
 158 }
 159
 160 static void copy_picture_attributes(MpegEncContext *s, AVFrame *dst, AVFrame *src){
 161     int i;
 162
 163     dst->pict_type              = src->pict_type;
 164     dst->quality                = src->quality;
 165     dst->coded_picture_number   = src->coded_picture_number;
 166     dst->display_picture_number = src->display_picture_number;
 167 //    dst->reference              = src->reference;
 168     dst->pts                    = src->pts;
 169     dst->interlaced_frame       = src->interlaced_frame;
 170     dst->top_field_first        = src->top_field_first;
 171
 172     if(s->avctx->me_threshold){
 173         if(!src->motion_val[0])
 174             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_val not set!\n");
 175         if(!src->mb_type)
 176             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.mb_type not set!\n");
 177         if(!src->ref_index[0])
 178             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.ref_index not set!\n");
 179         if(src->motion_subsample_log2 != dst->motion_subsample_log2)
 180             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_subsample_log2 doesn't match! (%d!=%d)\n",
 181             src->motion_subsample_log2, dst->motion_subsample_log2);
 182
 183         memcpy(dst->mb_type, src->mb_type, s->mb_stride * s->mb_height * sizeof(dst->mb_type[0]));
 184
 185         for(i=0; i<2; i++){
 186             int stride= ((16*s->mb_width )>>src->motion_subsample_log2) + 1;
 187             int height= ((16*s->mb_height)>>src->motion_subsample_log2);
 188
 189             if(src->motion_val[i] && src->motion_val[i] != dst->motion_val[i]){
 190                 memcpy(dst->motion_val[i], src->motion_val[i], 2*stride*height*sizeof(int16_t));
 191             }
 192             if(src->ref_index[i] && src->ref_index[i] != dst->ref_index[i]){
 193                 memcpy(dst->ref_index[i], src->ref_index[i], s->b8_stride*2*s->mb_height*sizeof(int8_t));
 194             }
 195         }
 196     }
 197 }
 198
 199 static void update_duplicate_context_after_me(MpegEncContext *dst, MpegEncContext *src){
 200 #define COPY(a) dst->a= src->a
 201     COPY(pict_type);
 202     COPY(current_picture);
 203     COPY(f_code);
 204     COPY(b_code);
 205     COPY(qscale);
 206     COPY(lambda);
 207     COPY(lambda2);
 208     COPY(picture_in_gop_number);
 209     COPY(gop_picture_number);
 210     COPY(frame_pred_frame_dct); //FIXME don't set in encode_header
 211     COPY(progressive_frame); //FIXME don't set in encode_header
 212     COPY(partitioned_frame); //FIXME don't set in encode_header
 213 #undef COPY
 214 }
 215
 216 /**
 217  * sets the given MpegEncContext to defaults for encoding.
 218  * the changed fields will not depend upon the prior state of the MpegEncContext.
 219  */
 220 static void MPV_encode_defaults(MpegEncContext *s){
 221     int i;
 222     MPV_common_defaults(s);
 223
 224     for(i=-16; i<16; i++){
 225         default_fcode_tab[i + MAX_MV]= 1;
 226     }
 227     s->me.mv_penalty= default_mv_penalty;
 228     s->fcode_tab= default_fcode_tab;
 229 }
 230
 231 /* init video encoder */
 232 int MPV_encode_init(AVCodecContext *avctx)
 233 {
 234     MpegEncContext *s = avctx->priv_data;
 235     int i;
 236     int chroma_h_shift, chroma_v_shift;
 237
 238     MPV_encode_defaults(s);
 239
 240     switch (avctx->codec_id) {
 241     case CODEC_ID_MPEG2VIDEO:
 242         if(avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P){
 243             av_log(avctx, AV_LOG_ERROR, "only YUV420 and YUV422 are supported\n");
 244             return -1;
 245         }
 246         break;
 247     case CODEC_ID_LJPEG:
 248     case CODEC_ID_MJPEG:
 249         if(avctx->pix_fmt != PIX_FMT_YUVJ420P && avctx->pix_fmt != PIX_FMT_YUVJ422P &&
 250            ((avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P) || avctx->strict_std_compliance>FF_COMPLIANCE_INOFFICIAL)){
 251             av_log(avctx, AV_LOG_ERROR, "colorspace not supported in jpeg\n");
 252             return -1;
 253         }
 254         break;
 255     default:
 256         if(avctx->pix_fmt != PIX_FMT_YUV420P){
 257             av_log(avctx, AV_LOG_ERROR, "only YUV420 is supported\n");
 258             return -1;
 259         }
 260     }
 261
 262     switch (avctx->pix_fmt) {
 263     case PIX_FMT_YUVJ422P:
 264     case PIX_FMT_YUV422P:
 265         s->chroma_format = CHROMA_422;
 266         break;
 267     case PIX_FMT_YUVJ420P:
 268     case PIX_FMT_YUV420P:
 269     default:
 270         s->chroma_format = CHROMA_420;
 271         break;
 272     }
 273
 274     s->bit_rate = avctx->bit_rate;
 275     s->width = avctx->width;
 276     s->height = avctx->height;
 277     if(avctx->gop_size > 600 && avctx->strict_std_compliance>FF_COMPLIANCE_EXPERIMENTAL){
 278         av_log(avctx, AV_LOG_ERROR, "Warning keyframe interval too large! reducing it ...\n");
 279         avctx->gop_size=600;
 280     }
 281     s->gop_size = avctx->gop_size;
 282     s->avctx = avctx;
 283     s->flags= avctx->flags;
 284     s->flags2= avctx->flags2;
 285     s->max_b_frames= avctx->max_b_frames;
 286     s->codec_id= avctx->codec->id;
 287     s->luma_elim_threshold  = avctx->luma_elim_threshold;
 288     s->chroma_elim_threshold= avctx->chroma_elim_threshold;
 289     s->strict_std_compliance= avctx->strict_std_compliance;
 290     s->data_partitioning= avctx->flags & CODEC_FLAG_PART;
 291     s->quarter_sample= (avctx->flags & CODEC_FLAG_QPEL)!=0;
 292     s->mpeg_quant= avctx->mpeg_quant;
 293     s->rtp_mode= !!avctx->rtp_payload_size;
 294     s->intra_dc_precision= avctx->intra_dc_precision;
 295     s->user_specified_pts = AV_NOPTS_VALUE;
 296
 297     if (s->gop_size <= 1) {
 298         s->intra_only = 1;
 299         s->gop_size = 12;
 300     } else {
 301         s->intra_only = 0;
 302     }
 303
 304     s->me_method = avctx->me_method;
 305
 306     /* Fixed QSCALE */
 307     s->fixed_qscale = !!(avctx->flags & CODEC_FLAG_QSCALE);
 308
 309     s->adaptive_quant= (   s->avctx->lumi_masking
 310                         || s->avctx->dark_masking
 311                         || s->avctx->temporal_cplx_masking
 312                         || s->avctx->spatial_cplx_masking
 313                         || s->avctx->p_masking
 314                         || s->avctx->border_masking
 315                         || (s->flags&CODEC_FLAG_QP_RD))
 316                        && !s->fixed_qscale;
 317
 318     s->obmc= !!(s->flags & CODEC_FLAG_OBMC);
 319     s->loop_filter= !!(s->flags & CODEC_FLAG_LOOP_FILTER);
 320     s->alternate_scan= !!(s->flags & CODEC_FLAG_ALT_SCAN);
 321     s->intra_vlc_format= !!(s->flags2 & CODEC_FLAG2_INTRA_VLC);
 322     s->q_scale_type= !!(s->flags2 & CODEC_FLAG2_NON_LINEAR_QUANT);
 323
 324     if(avctx->rc_max_rate && !avctx->rc_buffer_size){
 325         av_log(avctx, AV_LOG_ERROR, "a vbv buffer size is needed, for encoding with a maximum bitrate\n");
 326         return -1;
 327     }
 328
 329     if(avctx->rc_min_rate && avctx->rc_max_rate != avctx->rc_min_rate){
 330         av_log(avctx, AV_LOG_INFO, "Warning min_rate > 0 but min_rate != max_rate isn't recommended!\n");
 331     }
 332
 333     if(avctx->rc_min_rate && avctx->rc_min_rate > avctx->bit_rate){
 334         av_log(avctx, AV_LOG_ERROR, "bitrate below min bitrate\n");
 335         return -1;
 336     }
 337
 338     if(avctx->rc_max_rate && avctx->rc_max_rate < avctx->bit_rate){
 339         av_log(avctx, AV_LOG_INFO, "bitrate above max bitrate\n");
 340         return -1;
 341     }
 342
 343     if(avctx->rc_max_rate && avctx->rc_max_rate == avctx->bit_rate && avctx->rc_max_rate != avctx->rc_min_rate){
 344         av_log(avctx, AV_LOG_INFO, "impossible bitrate constraints, this will fail\n");
 345     }
 346
 347     if(avctx->rc_buffer_size && avctx->bit_rate*av_q2d(avctx->time_base) > avctx->rc_buffer_size){
 348         av_log(avctx, AV_LOG_ERROR, "VBV buffer too small for bitrate\n");
 349         return -1;
 350     }
 351
 352     if(avctx->bit_rate*av_q2d(avctx->time_base) > avctx->bit_rate_tolerance){
 353         av_log(avctx, AV_LOG_ERROR, "bitrate tolerance too small for bitrate\n");
 354         return -1;
 355     }
 356
 357     if(   s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate
 358        && (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO)
 359        && 90000LL * (avctx->rc_buffer_size-1) > s->avctx->rc_max_rate*0xFFFFLL){
 360
 361         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");
 362     }
 363
 364     if((s->flags & CODEC_FLAG_4MV) && s->codec_id != CODEC_ID_MPEG4
 365        && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P && s->codec_id != CODEC_ID_FLV1){
 366         av_log(avctx, AV_LOG_ERROR, "4MV not supported by codec\n");
 367         return -1;
 368     }
 369
 370     if(s->obmc && s->avctx->mb_decision != FF_MB_DECISION_SIMPLE){
 371         av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with simple mb decision\n");
 372         return -1;
 373     }
 374
 375     if(s->obmc && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P){
 376         av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with H263(+)\n");
 377         return -1;
 378     }
 379
 380     if(s->quarter_sample && s->codec_id != CODEC_ID_MPEG4){
 381         av_log(avctx, AV_LOG_ERROR, "qpel not supported by codec\n");
 382         return -1;
 383     }
 384
 385     if(s->data_partitioning && s->codec_id != CODEC_ID_MPEG4){
 386         av_log(avctx, AV_LOG_ERROR, "data partitioning not supported by codec\n");
 387         return -1;
 388     }
 389
 390     if(s->max_b_frames && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO){
 391         av_log(avctx, AV_LOG_ERROR, "b frames not supported by codec\n");
 392         return -1;
 393     }
 394
 395     if((s->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN))
 396        && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG2VIDEO){
 397         av_log(avctx, AV_LOG_ERROR, "interlacing not supported by codec\n");
 398         return -1;
 399     }
 400
 401     if(s->mpeg_quant && s->codec_id != CODEC_ID_MPEG4){ //FIXME mpeg2 uses that too
 402         av_log(avctx, AV_LOG_ERROR, "mpeg2 style quantization not supported by codec\n");
 403         return -1;
 404     }
 405
 406     if((s->flags & CODEC_FLAG_CBP_RD) && !(s->flags & CODEC_FLAG_TRELLIS_QUANT)){
 407         av_log(avctx, AV_LOG_ERROR, "CBP RD needs trellis quant\n");
 408         return -1;
 409     }
 410
 411     if((s->flags & CODEC_FLAG_QP_RD) && s->avctx->mb_decision != FF_MB_DECISION_RD){
 412         av_log(avctx, AV_LOG_ERROR, "QP RD needs mbd=2\n");
 413         return -1;
 414     }
 415
 416     if(s->avctx->scenechange_threshold < 1000000000 && (s->flags & CODEC_FLAG_CLOSED_GOP)){
 417         av_log(avctx, AV_LOG_ERROR, "closed gop with scene change detection arent supported yet, set threshold to 1000000000\n");
 418         return -1;
 419     }
 420
 421     if((s->flags2 & CODEC_FLAG2_INTRA_VLC) && s->codec_id != CODEC_ID_MPEG2VIDEO){
 422         av_log(avctx, AV_LOG_ERROR, "intra vlc table not supported by codec\n");
 423         return -1;
 424     }
 425
 426     if(s->flags & CODEC_FLAG_LOW_DELAY){
 427         if (s->codec_id != CODEC_ID_MPEG2VIDEO && s->codec_id != CODEC_ID_MPEG1VIDEO){
 428             av_log(avctx, AV_LOG_ERROR, "low delay forcing is only available for mpeg1/2\n");
 429             return -1;
 430         }
 431         if (s->max_b_frames != 0){
 432             av_log(avctx, AV_LOG_ERROR, "b frames cannot be used with low delay\n");
 433             return -1;
 434         }
 435     }
 436
 437     if(s->q_scale_type == 1){
 438         if(s->codec_id != CODEC_ID_MPEG2VIDEO){
 439             av_log(avctx, AV_LOG_ERROR, "non linear quant is only available for mpeg2\n");
 440             return -1;
 441         }
 442         if(avctx->qmax > 12){
 443             av_log(avctx, AV_LOG_ERROR, "non linear quant only supports qmax <= 12 currently\n");
 444             return -1;
 445         }
 446     }
 447
 448     if(s->avctx->thread_count > 1 && s->codec_id != CODEC_ID_MPEG4
 449        && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO
 450        && (s->codec_id != CODEC_ID_H263P || !(s->flags & CODEC_FLAG_H263P_SLICE_STRUCT))){
 451         av_log(avctx, AV_LOG_ERROR, "multi threaded encoding not supported by codec\n");
 452         return -1;
 453     }
 454
 455     if(s->avctx->thread_count > 1)
 456         s->rtp_mode= 1;
 457
 458     if(!avctx->time_base.den || !avctx->time_base.num){
 459         av_log(avctx, AV_LOG_ERROR, "framerate not set\n");
 460         return -1;
 461     }
 462
 463     i= (INT_MAX/2+128)>>8;
 464     if(avctx->me_threshold >= i){
 465         av_log(avctx, AV_LOG_ERROR, "me_threshold too large, max is %d\n", i - 1);
 466         return -1;
 467     }
 468     if(avctx->mb_threshold >= i){
 469         av_log(avctx, AV_LOG_ERROR, "mb_threshold too large, max is %d\n", i - 1);
 470         return -1;
 471     }
 472
 473     if(avctx->b_frame_strategy && (avctx->flags&CODEC_FLAG_PASS2)){
 474         av_log(avctx, AV_LOG_INFO, "notice: b_frame_strategy only affects the first pass\n");
 475         avctx->b_frame_strategy = 0;
 476     }
 477
 478     i= ff_gcd(avctx->time_base.den, avctx->time_base.num);
 479     if(i > 1){
 480         av_log(avctx, AV_LOG_INFO, "removing common factors from framerate\n");
 481         avctx->time_base.den /= i;
 482         avctx->time_base.num /= i;
 483 //        return -1;
 484     }
 485
 486     if(s->codec_id==CODEC_ID_MJPEG){
 487         s->intra_quant_bias= 1<<(QUANT_BIAS_SHIFT-1); //(a + x/2)/x
 488         s->inter_quant_bias= 0;
 489     }else if(s->mpeg_quant || s->codec_id==CODEC_ID_MPEG1VIDEO || s->codec_id==CODEC_ID_MPEG2VIDEO){
 490         s->intra_quant_bias= 3<<(QUANT_BIAS_SHIFT-3); //(a + x*3/8)/x
 491         s->inter_quant_bias= 0;
 492     }else{
 493         s->intra_quant_bias=0;
 494         s->inter_quant_bias=-(1<<(QUANT_BIAS_SHIFT-2)); //(a - x/4)/x
 495     }
 496
 497     if(avctx->intra_quant_bias != FF_DEFAULT_QUANT_BIAS)
 498         s->intra_quant_bias= avctx->intra_quant_bias;
 499     if(avctx->inter_quant_bias != FF_DEFAULT_QUANT_BIAS)
 500         s->inter_quant_bias= avctx->inter_quant_bias;
 501
 502     avcodec_get_chroma_sub_sample(avctx->pix_fmt, &chroma_h_shift, &chroma_v_shift);
 503
 504     if(avctx->codec_id == CODEC_ID_MPEG4 && s->avctx->time_base.den > (1<<16)-1){
 505         av_log(avctx, AV_LOG_ERROR, "timebase not supported by mpeg 4 standard\n");
 506         return -1;
 507     }
 508     s->time_increment_bits = av_log2(s->avctx->time_base.den - 1) + 1;
 509
 510     switch(avctx->codec->id) {
 511     case CODEC_ID_MPEG1VIDEO:
 512         s->out_format = FMT_MPEG1;
 513         s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY);
 514         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
 515         break;
 516     case CODEC_ID_MPEG2VIDEO:
 517         s->out_format = FMT_MPEG1;
 518         s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY);
 519         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
 520         s->rtp_mode= 1;
 521         break;
 522     case CODEC_ID_LJPEG:
 523     case CODEC_ID_MJPEG:
 524         s->out_format = FMT_MJPEG;
 525         s->intra_only = 1; /* force intra only for jpeg */
 526         s->mjpeg_vsample[0] = 2;
 527         s->mjpeg_vsample[1] = 2>>chroma_v_shift;
 528         s->mjpeg_vsample[2] = 2>>chroma_v_shift;
 529         s->mjpeg_hsample[0] = 2;
 530         s->mjpeg_hsample[1] = 2>>chroma_h_shift;
 531         s->mjpeg_hsample[2] = 2>>chroma_h_shift;
 532         if (!(ENABLE_MJPEG_ENCODER || ENABLE_LJPEG_ENCODER)
 533             || ff_mjpeg_encode_init(s) < 0)
 534             return -1;
 535         avctx->delay=0;
 536         s->low_delay=1;
 537         break;
 538     case CODEC_ID_H261:
 539         if (!ENABLE_H261_ENCODER)  return -1;
 540         if (ff_h261_get_picture_format(s->width, s->height) < 0) {
 541             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);
 542             return -1;
 543         }
 544         s->out_format = FMT_H261;
 545         avctx->delay=0;
 546         s->low_delay=1;
 547         break;
 548     case CODEC_ID_H263:
 549         if (!ENABLE_H263_ENCODER)  return -1;
 550         if (h263_get_picture_format(s->width, s->height) == 7) {
 551             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);
 552             return -1;
 553         }
 554         s->out_format = FMT_H263;
 555         s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
 556         avctx->delay=0;
 557         s->low_delay=1;
 558         break;
 559     case CODEC_ID_H263P:
 560         s->out_format = FMT_H263;
 561         s->h263_plus = 1;
 562         /* Fx */
 563         s->umvplus = (avctx->flags & CODEC_FLAG_H263P_UMV) ? 1:0;
 564         s->h263_aic= (avctx->flags & CODEC_FLAG_AC_PRED) ? 1:0;
 565         s->modified_quant= s->h263_aic;
 566         s->alt_inter_vlc= (avctx->flags & CODEC_FLAG_H263P_AIV) ? 1:0;
 567         s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
 568         s->loop_filter= (avctx->flags & CODEC_FLAG_LOOP_FILTER) ? 1:0;
 569         s->unrestricted_mv= s->obmc || s->loop_filter || s->umvplus;
 570         s->h263_slice_structured= (s->flags & CODEC_FLAG_H263P_SLICE_STRUCT) ? 1:0;
 571
 572         /* /Fx */
 573         /* These are just to be sure */
 574         avctx->delay=0;
 575         s->low_delay=1;
 576         break;
 577     case CODEC_ID_FLV1:
 578         s->out_format = FMT_H263;
 579         s->h263_flv = 2; /* format = 1; 11-bit codes */
 580         s->unrestricted_mv = 1;
 581         s->rtp_mode=0; /* don't allow GOB */
 582         avctx->delay=0;
 583         s->low_delay=1;
 584         break;
 585     case CODEC_ID_RV10:
 586         s->out_format = FMT_H263;
 587         avctx->delay=0;
 588         s->low_delay=1;
 589         break;
 590     case CODEC_ID_RV20:
 591         s->out_format = FMT_H263;
 592         avctx->delay=0;
 593         s->low_delay=1;
 594         s->modified_quant=1;
 595         s->h263_aic=1;
 596         s->h263_plus=1;
 597         s->loop_filter=1;
 598         s->unrestricted_mv= s->obmc || s->loop_filter || s->umvplus;
 599         break;
 600     case CODEC_ID_MPEG4:
 601         s->out_format = FMT_H263;
 602         s->h263_pred = 1;
 603         s->unrestricted_mv = 1;
 604         s->low_delay= s->max_b_frames ? 0 : 1;
 605         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
 606         break;
 607     case CODEC_ID_MSMPEG4V1:
 608         s->out_format = FMT_H263;
 609         s->h263_msmpeg4 = 1;
 610         s->h263_pred = 1;
 611         s->unrestricted_mv = 1;
 612         s->msmpeg4_version= 1;
 613         avctx->delay=0;
 614         s->low_delay=1;
 615         break;
 616     case CODEC_ID_MSMPEG4V2:
 617         s->out_format = FMT_H263;
 618         s->h263_msmpeg4 = 1;
 619         s->h263_pred = 1;
 620         s->unrestricted_mv = 1;
 621         s->msmpeg4_version= 2;
 622         avctx->delay=0;
 623         s->low_delay=1;
 624         break;
 625     case CODEC_ID_MSMPEG4V3:
 626         s->out_format = FMT_H263;
 627         s->h263_msmpeg4 = 1;
 628         s->h263_pred = 1;
 629         s->unrestricted_mv = 1;
 630         s->msmpeg4_version= 3;
 631         s->flipflop_rounding=1;
 632         avctx->delay=0;
 633         s->low_delay=1;
 634         break;
 635     case CODEC_ID_WMV1:
 636         s->out_format = FMT_H263;
 637         s->h263_msmpeg4 = 1;
 638         s->h263_pred = 1;
 639         s->unrestricted_mv = 1;
 640         s->msmpeg4_version= 4;
 641         s->flipflop_rounding=1;
 642         avctx->delay=0;
 643         s->low_delay=1;
 644         break;
 645     case CODEC_ID_WMV2:
 646         s->out_format = FMT_H263;
 647         s->h263_msmpeg4 = 1;
 648         s->h263_pred = 1;
 649         s->unrestricted_mv = 1;
 650         s->msmpeg4_version= 5;
 651         s->flipflop_rounding=1;
 652         avctx->delay=0;
 653         s->low_delay=1;
 654         break;
 655     default:
 656         return -1;
 657     }
 658
 659     avctx->has_b_frames= !s->low_delay;
 660
 661     s->encoding = 1;
 662
 663     /* init */
 664     if (MPV_common_init(s) < 0)
 665         return -1;
 666
 667     if(!s->dct_quantize)
 668         s->dct_quantize = dct_quantize_c;
 669     if(!s->denoise_dct)
 670         s->denoise_dct = denoise_dct_c;
 671     s->fast_dct_quantize = s->dct_quantize;
 672     if(s->flags & CODEC_FLAG_TRELLIS_QUANT)
 673         s->dct_quantize = dct_quantize_trellis_c;
 674
 675     if((ENABLE_H263P_ENCODER || ENABLE_RV20_ENCODER) && s->modified_quant)
 676         s->chroma_qscale_table= ff_h263_chroma_qscale_table;
 677     s->progressive_frame=
 678     s->progressive_sequence= !(avctx->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN));
 679     s->quant_precision=5;
 680
 681     ff_set_cmp(&s->dsp, s->dsp.ildct_cmp, s->avctx->ildct_cmp);
 682     ff_set_cmp(&s->dsp, s->dsp.frame_skip_cmp, s->avctx->frame_skip_cmp);
 683
 684     if (ENABLE_H261_ENCODER && s->out_format == FMT_H261)
 685         ff_h261_encode_init(s);
 686     if (ENABLE_ANY_H263_ENCODER && s->out_format == FMT_H263)
 687         h263_encode_init(s);
 688     if (ENABLE_MSMPEG4_ENCODER && s->msmpeg4_version)
 689         ff_msmpeg4_encode_init(s);
 690     if ((ENABLE_MPEG1VIDEO_ENCODER || ENABLE_MPEG2VIDEO_ENCODER)
 691         && s->out_format == FMT_MPEG1)
 692         ff_mpeg1_encode_init(s);
 693
 694     /* init q matrix */
 695     for(i=0;i<64;i++) {
 696         int j= s->dsp.idct_permutation[i];
 697         if(ENABLE_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4 && s->mpeg_quant){
 698             s->intra_matrix[j] = ff_mpeg4_default_intra_matrix[i];
 699             s->inter_matrix[j] = ff_mpeg4_default_non_intra_matrix[i];
 700         }else if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
 701             s->intra_matrix[j] =
 702             s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
 703         }else
 704         { /* mpeg1/2 */
 705             s->intra_matrix[j] = ff_mpeg1_default_intra_matrix[i];
 706             s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
 707         }
 708         if(s->avctx->intra_matrix)
 709             s->intra_matrix[j] = s->avctx->intra_matrix[i];
 710         if(s->avctx->inter_matrix)
 711             s->inter_matrix[j] = s->avctx->inter_matrix[i];
 712     }
 713
 714     /* precompute matrix */
 715     /* for mjpeg, we do include qscale in the matrix */
 716     if (s->out_format != FMT_MJPEG) {
 717         ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
 718                        s->intra_matrix, s->intra_quant_bias, avctx->qmin, 31, 1);
 719         ff_convert_matrix(&s->dsp, s->q_inter_matrix, s->q_inter_matrix16,
 720                        s->inter_matrix, s->inter_quant_bias, avctx->qmin, 31, 0);
 721     }
 722
 723     if(ff_rate_control_init(s) < 0)
 724         return -1;
 725
 726     return 0;
 727 }
 728
 729 int MPV_encode_end(AVCodecContext *avctx)
 730 {
 731     MpegEncContext *s = avctx->priv_data;
 732
 733     ff_rate_control_uninit(s);
 734
 735     MPV_common_end(s);
 736     if ((ENABLE_MJPEG_ENCODER || ENABLE_LJPEG_ENCODER) && s->out_format == FMT_MJPEG)
 737         ff_mjpeg_encode_close(s);
 738
 739     av_freep(&avctx->extradata);
 740
 741     return 0;
 742 }
 743
 744 static int get_sae(uint8_t *src, int ref, int stride){
 745     int x,y;
 746     int acc=0;
 747
 748     for(y=0; y<16; y++){
 749         for(x=0; x<16; x++){
 750             acc+= FFABS(src[x+y*stride] - ref);
 751         }
 752     }
 753
 754     return acc;
 755 }
 756
 757 static int get_intra_count(MpegEncContext *s, uint8_t *src, uint8_t *ref, int stride){
 758     int x, y, w, h;
 759     int acc=0;
 760
 761     w= s->width &~15;
 762     h= s->height&~15;
 763
 764     for(y=0; y<h; y+=16){
 765         for(x=0; x<w; x+=16){
 766             int offset= x + y*stride;
 767             int sad = s->dsp.sad[0](NULL, src + offset, ref + offset, stride, 16);
 768             int mean= (s->dsp.pix_sum(src + offset, stride) + 128)>>8;
 769             int sae = get_sae(src + offset, mean, stride);
 770
 771             acc+= sae + 500 < sad;
 772         }
 773     }
 774     return acc;
 775 }
 776
 777
 778 static int load_input_picture(MpegEncContext *s, AVFrame *pic_arg){
 779     AVFrame *pic=NULL;
 780     int64_t pts;
 781     int i;
 782     const int encoding_delay= s->max_b_frames;
 783     int direct=1;
 784
 785     if(pic_arg){
 786         pts= pic_arg->pts;
 787         pic_arg->display_picture_number= s->input_picture_number++;
 788
 789         if(pts != AV_NOPTS_VALUE){
 790             if(s->user_specified_pts != AV_NOPTS_VALUE){
 791                 int64_t time= pts;
 792                 int64_t last= s->user_specified_pts;
 793
 794                 if(time <= last){
 795                     av_log(s->avctx, AV_LOG_ERROR, "Error, Invalid timestamp=%"PRId64", last=%"PRId64"\n", pts, s->user_specified_pts);
 796                     return -1;
 797                 }
 798             }
 799             s->user_specified_pts= pts;
 800         }else{
 801             if(s->user_specified_pts != AV_NOPTS_VALUE){
 802                 s->user_specified_pts=
 803                 pts= s->user_specified_pts + 1;
 804                 av_log(s->avctx, AV_LOG_INFO, "Warning: AVFrame.pts=? trying to guess (%"PRId64")\n", pts);
 805             }else{
 806                 pts= pic_arg->display_picture_number;
 807             }
 808         }
 809     }
 810
 811   if(pic_arg){
 812     if(encoding_delay && !(s->flags&CODEC_FLAG_INPUT_PRESERVED)) direct=0;
 813     if(pic_arg->linesize[0] != s->linesize) direct=0;
 814     if(pic_arg->linesize[1] != s->uvlinesize) direct=0;
 815     if(pic_arg->linesize[2] != s->uvlinesize) direct=0;
 816
 817 //    av_log(AV_LOG_DEBUG, "%d %d %d %d\n",pic_arg->linesize[0], pic_arg->linesize[1], s->linesize, s->uvlinesize);
 818
 819     if(direct){
 820         i= ff_find_unused_picture(s, 1);
 821
 822         pic= (AVFrame*)&s->picture[i];
 823         pic->reference= 3;
 824
 825         for(i=0; i<4; i++){
 826             pic->data[i]= pic_arg->data[i];
 827             pic->linesize[i]= pic_arg->linesize[i];
 828         }
 829         alloc_picture(s, (Picture*)pic, 1);
 830     }else{
 831         i= ff_find_unused_picture(s, 0);
 832
 833         pic= (AVFrame*)&s->picture[i];
 834         pic->reference= 3;
 835
 836         alloc_picture(s, (Picture*)pic, 0);
 837
 838         if(   pic->data[0] + INPLACE_OFFSET == pic_arg->data[0]
 839            && pic->data[1] + INPLACE_OFFSET == pic_arg->data[1]
 840            && pic->data[2] + INPLACE_OFFSET == pic_arg->data[2]){
 841        // empty
 842         }else{
 843             int h_chroma_shift, v_chroma_shift;
 844             avcodec_get_chroma_sub_sample(s->avctx->pix_fmt, &h_chroma_shift, &v_chroma_shift);
 845
 846             for(i=0; i<3; i++){
 847                 int src_stride= pic_arg->linesize[i];
 848                 int dst_stride= i ? s->uvlinesize : s->linesize;
 849                 int h_shift= i ? h_chroma_shift : 0;
 850                 int v_shift= i ? v_chroma_shift : 0;
 851                 int w= s->width >>h_shift;
 852                 int h= s->height>>v_shift;
 853                 uint8_t *src= pic_arg->data[i];
 854                 uint8_t *dst= pic->data[i];
 855
 856                 if(!s->avctx->rc_buffer_size)
 857                     dst +=INPLACE_OFFSET;
 858
 859                 if(src_stride==dst_stride)
 860                     memcpy(dst, src, src_stride*h);
 861                 else{
 862                     while(h--){
 863                         memcpy(dst, src, w);
 864                         dst += dst_stride;
 865                         src += src_stride;
 866                     }
 867                 }
 868             }
 869         }
 870     }
 871     copy_picture_attributes(s, pic, pic_arg);
 872     pic->pts= pts; //we set this here to avoid modifiying pic_arg
 873   }
 874
 875     /* shift buffer entries */
 876     for(i=1; i<MAX_PICTURE_COUNT /*s->encoding_delay+1*/; i++)
 877         s->input_picture[i-1]= s->input_picture[i];
 878
 879     s->input_picture[encoding_delay]= (Picture*)pic;
 880
 881     return 0;
 882 }
 883
 884 static int skip_check(MpegEncContext *s, Picture *p, Picture *ref){
 885     int x, y, plane;
 886     int score=0;
 887     int64_t score64=0;
 888
 889     for(plane=0; plane<3; plane++){
 890         const int stride= p->linesize[plane];
 891         const int bw= plane ? 1 : 2;
 892         for(y=0; y<s->mb_height*bw; y++){
 893             for(x=0; x<s->mb_width*bw; x++){
 894                 int off= p->type == FF_BUFFER_TYPE_SHARED ? 0: 16;
 895                 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);
 896
 897                 switch(s->avctx->frame_skip_exp){
 898                     case 0: score= FFMAX(score, v); break;
 899                     case 1: score+= FFABS(v);break;
 900                     case 2: score+= v*v;break;
 901                     case 3: score64+= FFABS(v*v*(int64_t)v);break;
 902                     case 4: score64+= v*v*(int64_t)(v*v);break;
 903                 }
 904             }
 905         }
 906     }
 907
 908     if(score) score64= score;
 909
 910     if(score64 < s->avctx->frame_skip_threshold)
 911         return 1;
 912     if(score64 < ((s->avctx->frame_skip_factor * (int64_t)s->lambda)>>8))
 913         return 1;
 914     return 0;
 915 }
 916
 917 static int estimate_best_b_count(MpegEncContext *s){
 918     AVCodec *codec= avcodec_find_encoder(s->avctx->codec_id);
 919     AVCodecContext *c= avcodec_alloc_context();
 920     AVFrame input[FF_MAX_B_FRAMES+2];
 921     const int scale= s->avctx->brd_scale;
 922     int i, j, out_size, p_lambda, b_lambda, lambda2;
 923     int outbuf_size= s->width * s->height; //FIXME
 924     uint8_t *outbuf= av_malloc(outbuf_size);
 925     int64_t best_rd= INT64_MAX;
 926     int best_b_count= -1;
 927
 928     assert(scale>=0 && scale <=3);
 929
 930 //    emms_c();
 931     p_lambda= s->last_lambda_for[P_TYPE]; //s->next_picture_ptr->quality;
 932     b_lambda= s->last_lambda_for[B_TYPE]; //p_lambda *FFABS(s->avctx->b_quant_factor) + s->avctx->b_quant_offset;
 933     if(!b_lambda) b_lambda= p_lambda; //FIXME we should do this somewhere else
 934     lambda2= (b_lambda*b_lambda + (1<<FF_LAMBDA_SHIFT)/2 ) >> FF_LAMBDA_SHIFT;
 935
 936     c->width = s->width >> scale;
 937     c->height= s->height>> scale;
 938     c->flags= CODEC_FLAG_QSCALE | CODEC_FLAG_PSNR | CODEC_FLAG_INPUT_PRESERVED /*| CODEC_FLAG_EMU_EDGE*/;
 939     c->flags|= s->avctx->flags & CODEC_FLAG_QPEL;
 940     c->mb_decision= s->avctx->mb_decision;
 941     c->me_cmp= s->avctx->me_cmp;
 942     c->mb_cmp= s->avctx->mb_cmp;
 943     c->me_sub_cmp= s->avctx->me_sub_cmp;
 944     c->pix_fmt = PIX_FMT_YUV420P;
 945     c->time_base= s->avctx->time_base;
 946     c->max_b_frames= s->max_b_frames;
 947
 948     if (avcodec_open(c, codec) < 0)
 949         return -1;
 950
 951     for(i=0; i<s->max_b_frames+2; i++){
 952         int ysize= c->width*c->height;
 953         int csize= (c->width/2)*(c->height/2);
 954         Picture pre_input, *pre_input_ptr= i ? s->input_picture[i-1] : s->next_picture_ptr;
 955
 956         avcodec_get_frame_defaults(&input[i]);
 957         input[i].data[0]= av_malloc(ysize + 2*csize);
 958         input[i].data[1]= input[i].data[0] + ysize;
 959         input[i].data[2]= input[i].data[1] + csize;
 960         input[i].linesize[0]= c->width;
 961         input[i].linesize[1]=
 962         input[i].linesize[2]= c->width/2;
 963
 964         if(pre_input_ptr && (!i || s->input_picture[i-1])) {
 965             pre_input= *pre_input_ptr;
 966
 967             if(pre_input.type != FF_BUFFER_TYPE_SHARED && i) {
 968                 pre_input.data[0]+=INPLACE_OFFSET;
 969                 pre_input.data[1]+=INPLACE_OFFSET;
 970                 pre_input.data[2]+=INPLACE_OFFSET;
 971             }
 972
 973             s->dsp.shrink[scale](input[i].data[0], input[i].linesize[0], pre_input.data[0], pre_input.linesize[0], c->width, c->height);
 974             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);
 975             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);
 976         }
 977     }
 978
 979     for(j=0; j<s->max_b_frames+1; j++){
 980         int64_t rd=0;
 981
 982         if(!s->input_picture[j])
 983             break;
 984
 985         c->error[0]= c->error[1]= c->error[2]= 0;
 986
 987         input[0].pict_type= I_TYPE;
 988         input[0].quality= 1 * FF_QP2LAMBDA;
 989         out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[0]);
 990 //        rd += (out_size * lambda2) >> FF_LAMBDA_SHIFT;
 991
 992         for(i=0; i<s->max_b_frames+1; i++){
 993             int is_p= i % (j+1) == j || i==s->max_b_frames;
 994
 995             input[i+1].pict_type= is_p ? P_TYPE : B_TYPE;
 996             input[i+1].quality= is_p ? p_lambda : b_lambda;
 997             out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[i+1]);
 998             rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
 999         }
1000
1001         /* get the delayed frames */
1002         while(out_size){
1003             out_size = avcodec_encode_video(c, outbuf, outbuf_size, NULL);
1004             rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
1005         }
1006
1007         rd += c->error[0] + c->error[1] + c->error[2];
1008
1009         if(rd < best_rd){
1010             best_rd= rd;
1011             best_b_count= j;
1012         }
1013     }
1014
1015     av_freep(&outbuf);
1016     avcodec_close(c);
1017     av_freep(&c);
1018
1019     for(i=0; i<s->max_b_frames+2; i++){
1020         av_freep(&input[i].data[0]);
1021     }
1022
1023     return best_b_count;
1024 }
1025
1026 static void select_input_picture(MpegEncContext *s){
1027     int i;
1028
1029     for(i=1; i<MAX_PICTURE_COUNT; i++)
1030         s->reordered_input_picture[i-1]= s->reordered_input_picture[i];
1031     s->reordered_input_picture[MAX_PICTURE_COUNT-1]= NULL;
1032
1033     /* set next picture type & ordering */
1034     if(s->reordered_input_picture[0]==NULL && s->input_picture[0]){
1035         if(/*s->picture_in_gop_number >= s->gop_size ||*/ s->next_picture_ptr==NULL || s->intra_only){
1036             s->reordered_input_picture[0]= s->input_picture[0];
1037             s->reordered_input_picture[0]->pict_type= I_TYPE;
1038             s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++;
1039         }else{
1040             int b_frames;
1041
1042             if(s->avctx->frame_skip_threshold || s->avctx->frame_skip_factor){
1043                 if(s->picture_in_gop_number < s->gop_size && skip_check(s, s->input_picture[0], s->next_picture_ptr)){
1044                 //FIXME check that te gop check above is +-1 correct
1045 //av_log(NULL, AV_LOG_DEBUG, "skip %p %"PRId64"\n", s->input_picture[0]->data[0], s->input_picture[0]->pts);
1046
1047                     if(s->input_picture[0]->type == FF_BUFFER_TYPE_SHARED){
1048                         for(i=0; i<4; i++)
1049                             s->input_picture[0]->data[i]= NULL;
1050                         s->input_picture[0]->type= 0;
1051                     }else{
1052                         assert(   s->input_picture[0]->type==FF_BUFFER_TYPE_USER
1053                                || s->input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
1054
1055                         s->avctx->release_buffer(s->avctx, (AVFrame*)s->input_picture[0]);
1056                     }
1057
1058                     emms_c();
1059                     ff_vbv_update(s, 0);
1060
1061                     goto no_output_pic;
1062                 }
1063             }
1064
1065             if(s->flags&CODEC_FLAG_PASS2){
1066                 for(i=0; i<s->max_b_frames+1; i++){
1067                     int pict_num= s->input_picture[0]->display_picture_number + i;
1068
1069                     if(pict_num >= s->rc_context.num_entries)
1070                         break;
1071                     if(!s->input_picture[i]){
1072                         s->rc_context.entry[pict_num-1].new_pict_type = P_TYPE;
1073                         break;
1074                     }
1075
1076                     s->input_picture[i]->pict_type=
1077                         s->rc_context.entry[pict_num].new_pict_type;
1078                 }
1079             }
1080
1081             if(s->avctx->b_frame_strategy==0){
1082                 b_frames= s->max_b_frames;
1083                 while(b_frames && !s->input_picture[b_frames]) b_frames--;
1084             }else if(s->avctx->b_frame_strategy==1){
1085                 for(i=1; i<s->max_b_frames+1; i++){
1086                     if(s->input_picture[i] && s->input_picture[i]->b_frame_score==0){
1087                         s->input_picture[i]->b_frame_score=
1088                             get_intra_count(s, s->input_picture[i  ]->data[0],
1089                                                s->input_picture[i-1]->data[0], s->linesize) + 1;
1090                     }
1091                 }
1092                 for(i=0; i<s->max_b_frames+1; i++){
1093                     if(s->input_picture[i]==NULL || s->input_picture[i]->b_frame_score - 1 > s->mb_num/s->avctx->b_sensitivity) break;
1094                 }
1095
1096                 b_frames= FFMAX(0, i-1);
1097
1098                 /* reset scores */
1099                 for(i=0; i<b_frames+1; i++){
1100                     s->input_picture[i]->b_frame_score=0;
1101                 }
1102             }else if(s->avctx->b_frame_strategy==2){
1103                 b_frames= estimate_best_b_count(s);
1104             }else{
1105                 av_log(s->avctx, AV_LOG_ERROR, "illegal b frame strategy\n");
1106                 b_frames=0;
1107             }
1108
1109             emms_c();
1110 //static int b_count=0;
1111 //b_count+= b_frames;
1112 //av_log(s->avctx, AV_LOG_DEBUG, "b_frames: %d\n", b_count);
1113
1114             for(i= b_frames - 1; i>=0; i--){
1115                 int type= s->input_picture[i]->pict_type;
1116                 if(type && type != B_TYPE)
1117                     b_frames= i;
1118             }
1119             if(s->input_picture[b_frames]->pict_type == B_TYPE && b_frames == s->max_b_frames){
1120                 av_log(s->avctx, AV_LOG_ERROR, "warning, too many b frames in a row\n");
1121             }
1122
1123             if(s->picture_in_gop_number + b_frames >= s->gop_size){
1124               if((s->flags2 & CODEC_FLAG2_STRICT_GOP) && s->gop_size > s->picture_in_gop_number){
1125                     b_frames= s->gop_size - s->picture_in_gop_number - 1;
1126               }else{
1127                 if(s->flags & CODEC_FLAG_CLOSED_GOP)
1128                     b_frames=0;
1129                 s->input_picture[b_frames]->pict_type= I_TYPE;
1130               }
1131             }
1132
1133             if(   (s->flags & CODEC_FLAG_CLOSED_GOP)
1134                && b_frames
1135                && s->input_picture[b_frames]->pict_type== I_TYPE)
1136                 b_frames--;
1137
1138             s->reordered_input_picture[0]= s->input_picture[b_frames];
1139             if(s->reordered_input_picture[0]->pict_type != I_TYPE)
1140                 s->reordered_input_picture[0]->pict_type= P_TYPE;
1141             s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++;
1142             for(i=0; i<b_frames; i++){
1143                 s->reordered_input_picture[i+1]= s->input_picture[i];
1144                 s->reordered_input_picture[i+1]->pict_type= B_TYPE;
1145                 s->reordered_input_picture[i+1]->coded_picture_number= s->coded_picture_number++;
1146             }
1147         }
1148     }
1149 no_output_pic:
1150     if(s->reordered_input_picture[0]){
1151         s->reordered_input_picture[0]->reference= s->reordered_input_picture[0]->pict_type!=B_TYPE ? 3 : 0;
1152
1153         copy_picture(&s->new_picture, s->reordered_input_picture[0]);
1154
1155         if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_SHARED || s->avctx->rc_buffer_size){
1156             // input is a shared pix, so we can't modifiy it -> alloc a new one & ensure that the shared one is reuseable
1157
1158             int i= ff_find_unused_picture(s, 0);
1159             Picture *pic= &s->picture[i];
1160
1161             pic->reference              = s->reordered_input_picture[0]->reference;
1162             alloc_picture(s, pic, 0);
1163
1164             /* mark us unused / free shared pic */
1165             if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_INTERNAL)
1166                 s->avctx->release_buffer(s->avctx, (AVFrame*)s->reordered_input_picture[0]);
1167             for(i=0; i<4; i++)
1168                 s->reordered_input_picture[0]->data[i]= NULL;
1169             s->reordered_input_picture[0]->type= 0;
1170
1171             copy_picture_attributes(s, (AVFrame*)pic, (AVFrame*)s->reordered_input_picture[0]);
1172
1173             s->current_picture_ptr= pic;
1174         }else{
1175             // input is not a shared pix -> reuse buffer for current_pix
1176
1177             assert(   s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_USER
1178                    || s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
1179
1180             s->current_picture_ptr= s->reordered_input_picture[0];
1181             for(i=0; i<4; i++){
1182                 s->new_picture.data[i]+= INPLACE_OFFSET;
1183             }
1184         }
1185         copy_picture(&s->current_picture, s->current_picture_ptr);
1186
1187         s->picture_number= s->new_picture.display_picture_number;
1188 //printf("dpn:%d\n", s->picture_number);
1189     }else{
1190        memset(&s->new_picture, 0, sizeof(Picture));
1191     }
1192 }
1193
1194 int MPV_encode_picture(AVCodecContext *avctx,
1195                        unsigned char *buf, int buf_size, void *data)
1196 {
1197     MpegEncContext *s = avctx->priv_data;
1198     AVFrame *pic_arg = data;
1199     int i, stuffing_count;
1200
1201     for(i=0; i<avctx->thread_count; i++){
1202         int start_y= s->thread_context[i]->start_mb_y;
1203         int   end_y= s->thread_context[i]->  end_mb_y;
1204         int h= s->mb_height;
1205         uint8_t *start= buf + (size_t)(((int64_t) buf_size)*start_y/h);
1206         uint8_t *end  = buf + (size_t)(((int64_t) buf_size)*  end_y/h);
1207
1208         init_put_bits(&s->thread_context[i]->pb, start, end - start);
1209     }
1210
1211     s->picture_in_gop_number++;
1212
1213     if(load_input_picture(s, pic_arg) < 0)
1214         return -1;
1215
1216     select_input_picture(s);
1217
1218     /* output? */
1219     if(s->new_picture.data[0]){
1220         s->pict_type= s->new_picture.pict_type;
1221 //emms_c();
1222 //printf("qs:%f %f %d\n", s->new_picture.quality, s->current_picture.quality, s->qscale);
1223         MPV_frame_start(s, avctx);
1224 vbv_retry:
1225         if (encode_picture(s, s->picture_number) < 0)
1226             return -1;
1227
1228         avctx->real_pict_num  = s->picture_number;
1229         avctx->header_bits = s->header_bits;
1230         avctx->mv_bits     = s->mv_bits;
1231         avctx->misc_bits   = s->misc_bits;
1232         avctx->i_tex_bits  = s->i_tex_bits;
1233         avctx->p_tex_bits  = s->p_tex_bits;
1234         avctx->i_count     = s->i_count;
1235         avctx->p_count     = s->mb_num - s->i_count - s->skip_count; //FIXME f/b_count in avctx
1236         avctx->skip_count  = s->skip_count;
1237
1238         MPV_frame_end(s);
1239
1240         if (ENABLE_MJPEG_ENCODER && s->out_format == FMT_MJPEG)
1241             ff_mjpeg_encode_picture_trailer(s);
1242
1243         if(avctx->rc_buffer_size){
1244             RateControlContext *rcc= &s->rc_context;
1245             int max_size= rcc->buffer_index/3;
1246
1247             if(put_bits_count(&s->pb) > max_size && s->lambda < s->avctx->lmax){
1248                 s->next_lambda= FFMAX(s->lambda+1, s->lambda*(s->qscale+1) / s->qscale);
1249                 if(s->adaptive_quant){
1250                     int i;
1251                     for(i=0; i<s->mb_height*s->mb_stride; i++)
1252                         s->lambda_table[i]= FFMAX(s->lambda_table[i]+1, s->lambda_table[i]*(s->qscale+1) / s->qscale);
1253                 }
1254                 s->mb_skipped = 0;        //done in MPV_frame_start()
1255                 if(s->pict_type==P_TYPE){ //done in encode_picture() so we must undo it
1256                     if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
1257                         s->no_rounding ^= 1;
1258                 }
1259                 if(s->pict_type!=B_TYPE){
1260                     s->time_base= s->last_time_base;
1261                     s->last_non_b_time= s->time - s->pp_time;
1262                 }
1263 //                av_log(NULL, AV_LOG_ERROR, "R:%d ", s->next_lambda);
1264                 for(i=0; i<avctx->thread_count; i++){
1265                     PutBitContext *pb= &s->thread_context[i]->pb;
1266                     init_put_bits(pb, pb->buf, pb->buf_end - pb->buf);
1267                 }
1268                 goto vbv_retry;
1269             }
1270
1271             assert(s->avctx->rc_max_rate);
1272         }
1273
1274         if(s->flags&CODEC_FLAG_PASS1)
1275             ff_write_pass1_stats(s);
1276
1277         for(i=0; i<4; i++){
1278             s->current_picture_ptr->error[i]= s->current_picture.error[i];
1279             avctx->error[i] += s->current_picture_ptr->error[i];
1280         }
1281
1282         if(s->flags&CODEC_FLAG_PASS1)
1283             assert(avctx->header_bits + avctx->mv_bits + avctx->misc_bits + avctx->i_tex_bits + avctx->p_tex_bits == put_bits_count(&s->pb));
1284         flush_put_bits(&s->pb);
1285         s->frame_bits  = put_bits_count(&s->pb);
1286
1287         stuffing_count= ff_vbv_update(s, s->frame_bits);
1288         if(stuffing_count){
1289             if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < stuffing_count + 50){
1290                 av_log(s->avctx, AV_LOG_ERROR, "stuffing too large\n");
1291                 return -1;
1292             }
1293
1294             switch(s->codec_id){
1295             case CODEC_ID_MPEG1VIDEO:
1296             case CODEC_ID_MPEG2VIDEO:
1297                 while(stuffing_count--){
1298                     put_bits(&s->pb, 8, 0);
1299                 }
1300             break;
1301             case CODEC_ID_MPEG4:
1302                 put_bits(&s->pb, 16, 0);
1303                 put_bits(&s->pb, 16, 0x1C3);
1304                 stuffing_count -= 4;
1305                 while(stuffing_count--){
1306                     put_bits(&s->pb, 8, 0xFF);
1307                 }
1308             break;
1309             default:
1310                 av_log(s->avctx, AV_LOG_ERROR, "vbv buffer overflow\n");
1311             }
1312             flush_put_bits(&s->pb);
1313             s->frame_bits  = put_bits_count(&s->pb);
1314         }
1315
1316         /* update mpeg1/2 vbv_delay for CBR */
1317         if(s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate && s->out_format == FMT_MPEG1
1318            && 90000LL * (avctx->rc_buffer_size-1) <= s->avctx->rc_max_rate*0xFFFFLL){
1319             int vbv_delay;
1320
1321             assert(s->repeat_first_field==0);
1322
1323             vbv_delay= lrintf(90000 * s->rc_context.buffer_index / s->avctx->rc_max_rate);
1324             assert(vbv_delay < 0xFFFF);
1325
1326             s->vbv_delay_ptr[0] &= 0xF8;
1327             s->vbv_delay_ptr[0] |= vbv_delay>>13;
1328             s->vbv_delay_ptr[1]  = vbv_delay>>5;
1329             s->vbv_delay_ptr[2] &= 0x07;
1330             s->vbv_delay_ptr[2] |= vbv_delay<<3;
1331         }
1332         s->total_bits += s->frame_bits;
1333         avctx->frame_bits  = s->frame_bits;
1334     }else{
1335         assert((pbBufPtr(&s->pb) == s->pb.buf));
1336         s->frame_bits=0;
1337     }
1338     assert((s->frame_bits&7)==0);
1339
1340     return s->frame_bits/8;
1341 }
1342
1343 static inline void dct_single_coeff_elimination(MpegEncContext *s, int n, int threshold)
1344 {
1345     static const char tab[64]=
1346         {3,2,2,1,1,1,1,1,
1347          1,1,1,1,1,1,1,1,
1348          1,1,1,1,1,1,1,1,
1349          0,0,0,0,0,0,0,0,
1350          0,0,0,0,0,0,0,0,
1351          0,0,0,0,0,0,0,0,
1352          0,0,0,0,0,0,0,0,
1353          0,0,0,0,0,0,0,0};
1354     int score=0;
1355     int run=0;
1356     int i;
1357     DCTELEM *block= s->block[n];
1358     const int last_index= s->block_last_index[n];
1359     int skip_dc;
1360
1361     if(threshold<0){
1362         skip_dc=0;
1363         threshold= -threshold;
1364     }else
1365         skip_dc=1;
1366
1367     /* are all which we could set to zero are allready zero? */
1368     if(last_index<=skip_dc - 1) return;
1369
1370     for(i=0; i<=last_index; i++){
1371         const int j = s->intra_scantable.permutated[i];
1372         const int level = FFABS(block[j]);
1373         if(level==1){
1374             if(skip_dc && i==0) continue;
1375             score+= tab[run];
1376             run=0;
1377         }else if(level>1){
1378             return;
1379         }else{
1380             run++;
1381         }
1382     }
1383     if(score >= threshold) return;
1384     for(i=skip_dc; i<=last_index; i++){
1385         const int j = s->intra_scantable.permutated[i];
1386         block[j]=0;
1387     }
1388     if(block[0]) s->block_last_index[n]= 0;
1389     else         s->block_last_index[n]= -1;
1390 }
1391
1392 static inline void clip_coeffs(MpegEncContext *s, DCTELEM *block, int last_index)
1393 {
1394     int i;
1395     const int maxlevel= s->max_qcoeff;
1396     const int minlevel= s->min_qcoeff;
1397     int overflow=0;
1398
1399     if(s->mb_intra){
1400         i=1; //skip clipping of intra dc
1401     }else
1402         i=0;
1403
1404     for(;i<=last_index; i++){
1405         const int j= s->intra_scantable.permutated[i];
1406         int level = block[j];
1407
1408         if     (level>maxlevel){
1409             level=maxlevel;
1410             overflow++;
1411         }else if(level<minlevel){
1412             level=minlevel;
1413             overflow++;
1414         }
1415
1416         block[j]= level;
1417     }
1418
1419     if(overflow && s->avctx->mb_decision == FF_MB_DECISION_SIMPLE)
1420         av_log(s->avctx, AV_LOG_INFO, "warning, clipping %d dct coefficients to %d..%d\n", overflow, minlevel, maxlevel);
1421 }
1422
1423 static void get_visual_weight(int16_t *weight, uint8_t *ptr, int stride){
1424     int x, y;
1425 //FIXME optimize
1426     for(y=0; y<8; y++){
1427         for(x=0; x<8; x++){
1428             int x2, y2;
1429             int sum=0;
1430             int sqr=0;
1431             int count=0;
1432
1433             for(y2= FFMAX(y-1, 0); y2 < FFMIN(8, y+2); y2++){
1434                 for(x2= FFMAX(x-1, 0); x2 < FFMIN(8, x+2); x2++){
1435                     int v= ptr[x2 + y2*stride];
1436                     sum += v;
1437                     sqr += v*v;
1438                     count++;
1439                 }
1440             }
1441             weight[x + 8*y]= (36*ff_sqrt(count*sqr - sum*sum)) / count;
1442         }
1443     }
1444 }
1445
1446 static av_always_inline void encode_mb_internal(MpegEncContext *s, int motion_x, int motion_y, int mb_block_height, int mb_block_count)
1447 {
1448     int16_t weight[8][64];
1449     DCTELEM orig[8][64];
1450     const int mb_x= s->mb_x;
1451     const int mb_y= s->mb_y;
1452     int i;
1453     int skip_dct[8];
1454     int dct_offset   = s->linesize*8; //default for progressive frames
1455     uint8_t *ptr_y, *ptr_cb, *ptr_cr;
1456     int wrap_y, wrap_c;
1457
1458     for(i=0; i<mb_block_count; i++) skip_dct[i]=s->skipdct;
1459
1460     if(s->adaptive_quant){
1461         const int last_qp= s->qscale;
1462         const int mb_xy= mb_x + mb_y*s->mb_stride;
1463
1464         s->lambda= s->lambda_table[mb_xy];
1465         update_qscale(s);
1466
1467         if(!(s->flags&CODEC_FLAG_QP_RD)){
1468             s->qscale= s->current_picture_ptr->qscale_table[mb_xy];
1469             s->dquant= s->qscale - last_qp;
1470
1471             if(s->out_format==FMT_H263){
1472                 s->dquant= av_clip(s->dquant, -2, 2);
1473
1474                 if(s->codec_id==CODEC_ID_MPEG4){
1475                     if(!s->mb_intra){
1476                         if(s->pict_type == B_TYPE){
1477                             if(s->dquant&1 || s->mv_dir&MV_DIRECT)
1478                                 s->dquant= 0;
1479                         }
1480                         if(s->mv_type==MV_TYPE_8X8)
1481                             s->dquant=0;
1482                     }
1483                 }
1484             }
1485         }
1486         ff_set_qscale(s, last_qp + s->dquant);
1487     }else if(s->flags&CODEC_FLAG_QP_RD)
1488         ff_set_qscale(s, s->qscale + s->dquant);
1489
1490     wrap_y = s->linesize;
1491     wrap_c = s->uvlinesize;
1492     ptr_y = s->new_picture.data[0] + (mb_y * 16 * wrap_y) + mb_x * 16;
1493     ptr_cb = s->new_picture.data[1] + (mb_y * mb_block_height * wrap_c) + mb_x * 8;
1494     ptr_cr = s->new_picture.data[2] + (mb_y * mb_block_height * wrap_c) + mb_x * 8;
1495
1496     if(mb_x*16+16 > s->width || mb_y*16+16 > s->height){
1497         uint8_t *ebuf= s->edge_emu_buffer + 32;
1498         ff_emulated_edge_mc(ebuf            , ptr_y , wrap_y,16,16,mb_x*16,mb_y*16, s->width   , s->height);
1499         ptr_y= ebuf;
1500         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);
1501         ptr_cb= ebuf+18*wrap_y;
1502         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);
1503         ptr_cr= ebuf+18*wrap_y+8;
1504     }
1505
1506     if (s->mb_intra) {
1507         if(s->flags&CODEC_FLAG_INTERLACED_DCT){
1508             int progressive_score, interlaced_score;
1509
1510             s->interlaced_dct=0;
1511             progressive_score= s->dsp.ildct_cmp[4](s, ptr_y           , NULL, wrap_y, 8)
1512                               +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y*8, NULL, wrap_y, 8) - 400;
1513
1514             if(progressive_score > 0){
1515                 interlaced_score = s->dsp.ildct_cmp[4](s, ptr_y           , NULL, wrap_y*2, 8)
1516                                   +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y  , NULL, wrap_y*2, 8);
1517                 if(progressive_score > interlaced_score){
1518                     s->interlaced_dct=1;
1519
1520                     dct_offset= wrap_y;
1521                     wrap_y<<=1;
1522                     if (s->chroma_format == CHROMA_422)
1523                         wrap_c<<=1;
1524                 }
1525             }
1526         }
1527
1528         s->dsp.get_pixels(s->block[0], ptr_y                 , wrap_y);
1529         s->dsp.get_pixels(s->block[1], ptr_y              + 8, wrap_y);
1530         s->dsp.get_pixels(s->block[2], ptr_y + dct_offset    , wrap_y);
1531         s->dsp.get_pixels(s->block[3], ptr_y + dct_offset + 8, wrap_y);
1532
1533         if(s->flags&CODEC_FLAG_GRAY){
1534             skip_dct[4]= 1;
1535             skip_dct[5]= 1;
1536         }else{
1537             s->dsp.get_pixels(s->block[4], ptr_cb, wrap_c);
1538             s->dsp.get_pixels(s->block[5], ptr_cr, wrap_c);
1539             if(!s->chroma_y_shift){ /* 422 */
1540                 s->dsp.get_pixels(s->block[6], ptr_cb + (dct_offset>>1), wrap_c);
1541                 s->dsp.get_pixels(s->block[7], ptr_cr + (dct_offset>>1), wrap_c);
1542             }
1543         }
1544     }else{
1545         op_pixels_func (*op_pix)[4];
1546         qpel_mc_func (*op_qpix)[16];
1547         uint8_t *dest_y, *dest_cb, *dest_cr;
1548
1549         dest_y  = s->dest[0];
1550         dest_cb = s->dest[1];
1551         dest_cr = s->dest[2];
1552
1553         if ((!s->no_rounding) || s->pict_type==B_TYPE){
1554             op_pix = s->dsp.put_pixels_tab;
1555             op_qpix= s->dsp.put_qpel_pixels_tab;
1556         }else{
1557             op_pix = s->dsp.put_no_rnd_pixels_tab;
1558             op_qpix= s->dsp.put_no_rnd_qpel_pixels_tab;
1559         }
1560
1561         if (s->mv_dir & MV_DIR_FORWARD) {
1562             MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.data, op_pix, op_qpix);
1563             op_pix = s->dsp.avg_pixels_tab;
1564             op_qpix= s->dsp.avg_qpel_pixels_tab;
1565         }
1566         if (s->mv_dir & MV_DIR_BACKWARD) {
1567             MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.data, op_pix, op_qpix);
1568         }
1569
1570         if(s->flags&CODEC_FLAG_INTERLACED_DCT){
1571             int progressive_score, interlaced_score;
1572
1573             s->interlaced_dct=0;
1574             progressive_score= s->dsp.ildct_cmp[0](s, dest_y           , ptr_y           , wrap_y, 8)
1575                               +s->dsp.ildct_cmp[0](s, dest_y + wrap_y*8, ptr_y + wrap_y*8, wrap_y, 8) - 400;
1576
1577             if(s->avctx->ildct_cmp == FF_CMP_VSSE) progressive_score -= 400;
1578
1579             if(progressive_score>0){
1580                 interlaced_score = s->dsp.ildct_cmp[0](s, dest_y           , ptr_y           , wrap_y*2, 8)
1581                                   +s->dsp.ildct_cmp[0](s, dest_y + wrap_y  , ptr_y + wrap_y  , wrap_y*2, 8);
1582
1583                 if(progressive_score > interlaced_score){
1584                     s->interlaced_dct=1;
1585
1586                     dct_offset= wrap_y;
1587                     wrap_y<<=1;
1588                     if (s->chroma_format == CHROMA_422)
1589                         wrap_c<<=1;
1590                 }
1591             }
1592         }
1593
1594         s->dsp.diff_pixels(s->block[0], ptr_y                 , dest_y                 , wrap_y);
1595         s->dsp.diff_pixels(s->block[1], ptr_y              + 8, dest_y              + 8, wrap_y);
1596         s->dsp.diff_pixels(s->block[2], ptr_y + dct_offset    , dest_y + dct_offset    , wrap_y);
1597         s->dsp.diff_pixels(s->block[3], ptr_y + dct_offset + 8, dest_y + dct_offset + 8, wrap_y);
1598
1599         if(s->flags&CODEC_FLAG_GRAY){
1600             skip_dct[4]= 1;
1601             skip_dct[5]= 1;
1602         }else{
1603             s->dsp.diff_pixels(s->block[4], ptr_cb, dest_cb, wrap_c);
1604             s->dsp.diff_pixels(s->block[5], ptr_cr, dest_cr, wrap_c);
1605             if(!s->chroma_y_shift){ /* 422 */
1606                 s->dsp.diff_pixels(s->block[6], ptr_cb + (dct_offset>>1), dest_cb + (dct_offset>>1), wrap_c);
1607                 s->dsp.diff_pixels(s->block[7], ptr_cr + (dct_offset>>1), dest_cr + (dct_offset>>1), wrap_c);
1608             }
1609         }
1610         /* pre quantization */
1611         if(s->current_picture.mc_mb_var[s->mb_stride*mb_y+ mb_x]<2*s->qscale*s->qscale){
1612             //FIXME optimize
1613             if(s->dsp.sad[1](NULL, ptr_y               , dest_y               , wrap_y, 8) < 20*s->qscale) skip_dct[0]= 1;
1614             if(s->dsp.sad[1](NULL, ptr_y            + 8, dest_y            + 8, wrap_y, 8) < 20*s->qscale) skip_dct[1]= 1;
1615             if(s->dsp.sad[1](NULL, ptr_y +dct_offset   , dest_y +dct_offset   , wrap_y, 8) < 20*s->qscale) skip_dct[2]= 1;
1616             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;
1617             if(s->dsp.sad[1](NULL, ptr_cb              , dest_cb              , wrap_c, 8) < 20*s->qscale) skip_dct[4]= 1;
1618             if(s->dsp.sad[1](NULL, ptr_cr              , dest_cr              , wrap_c, 8) < 20*s->qscale) skip_dct[5]= 1;
1619             if(!s->chroma_y_shift){ /* 422 */
1620                 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;
1621                 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;
1622             }
1623         }
1624     }
1625
1626     if(s->avctx->quantizer_noise_shaping){
1627         if(!skip_dct[0]) get_visual_weight(weight[0], ptr_y                 , wrap_y);
1628         if(!skip_dct[1]) get_visual_weight(weight[1], ptr_y              + 8, wrap_y);
1629         if(!skip_dct[2]) get_visual_weight(weight[2], ptr_y + dct_offset    , wrap_y);
1630         if(!skip_dct[3]) get_visual_weight(weight[3], ptr_y + dct_offset + 8, wrap_y);
1631         if(!skip_dct[4]) get_visual_weight(weight[4], ptr_cb                , wrap_c);
1632         if(!skip_dct[5]) get_visual_weight(weight[5], ptr_cr                , wrap_c);
1633         if(!s->chroma_y_shift){ /* 422 */
1634             if(!skip_dct[6]) get_visual_weight(weight[6], ptr_cb + (dct_offset>>1), wrap_c);
1635             if(!skip_dct[7]) get_visual_weight(weight[7], ptr_cr + (dct_offset>>1), wrap_c);
1636         }
1637         memcpy(orig[0], s->block[0], sizeof(DCTELEM)*64*mb_block_count);
1638     }
1639
1640     /* DCT & quantize */
1641     assert(s->out_format!=FMT_MJPEG || s->qscale==8);
1642     {
1643         for(i=0;i<mb_block_count;i++) {
1644             if(!skip_dct[i]){
1645                 int overflow;
1646                 s->block_last_index[i] = s->dct_quantize(s, s->block[i], i, s->qscale, &overflow);
1647             // FIXME we could decide to change to quantizer instead of clipping
1648             // JS: I don't think that would be a good idea it could lower quality instead
1649             //     of improve it. Just INTRADC clipping deserves changes in quantizer
1650                 if (overflow) clip_coeffs(s, s->block[i], s->block_last_index[i]);
1651             }else
1652                 s->block_last_index[i]= -1;
1653         }
1654         if(s->avctx->quantizer_noise_shaping){
1655             for(i=0;i<mb_block_count;i++) {
1656                 if(!skip_dct[i]){
1657                     s->block_last_index[i] = dct_quantize_refine(s, s->block[i], weight[i], orig[i], i, s->qscale);
1658                 }
1659             }
1660         }
1661
1662         if(s->luma_elim_threshold && !s->mb_intra)
1663             for(i=0; i<4; i++)
1664                 dct_single_coeff_elimination(s, i, s->luma_elim_threshold);
1665         if(s->chroma_elim_threshold && !s->mb_intra)
1666             for(i=4; i<mb_block_count; i++)
1667                 dct_single_coeff_elimination(s, i, s->chroma_elim_threshold);
1668
1669         if(s->flags & CODEC_FLAG_CBP_RD){
1670             for(i=0;i<mb_block_count;i++) {
1671                 if(s->block_last_index[i] == -1)
1672                     s->coded_score[i]= INT_MAX/256;
1673             }
1674         }
1675     }
1676
1677     if((s->flags&CODEC_FLAG_GRAY) && s->mb_intra){
1678         s->block_last_index[4]=
1679         s->block_last_index[5]= 0;
1680         s->block[4][0]=
1681         s->block[5][0]= (1024 + s->c_dc_scale/2)/ s->c_dc_scale;
1682     }
1683
1684     //non c quantize code returns incorrect block_last_index FIXME
1685     if(s->alternate_scan && s->dct_quantize != dct_quantize_c){
1686         for(i=0; i<mb_block_count; i++){
1687             int j;
1688             if(s->block_last_index[i]>0){
1689                 for(j=63; j>0; j--){
1690                     if(s->block[i][ s->intra_scantable.permutated[j] ]) break;
1691                 }
1692                 s->block_last_index[i]= j;
1693             }
1694         }
1695     }
1696
1697     /* huffman encode */
1698     switch(s->codec_id){ //FIXME funct ptr could be slightly faster
1699     case CODEC_ID_MPEG1VIDEO:
1700     case CODEC_ID_MPEG2VIDEO:
1701         if (ENABLE_MPEG1VIDEO_ENCODER || ENABLE_MPEG2VIDEO_ENCODER)
1702             mpeg1_encode_mb(s, s->block, motion_x, motion_y);
1703         break;
1704     case CODEC_ID_MPEG4:
1705         if (ENABLE_MPEG4_ENCODER)
1706             mpeg4_encode_mb(s, s->block, motion_x, motion_y);
1707         break;
1708     case CODEC_ID_MSMPEG4V2:
1709     case CODEC_ID_MSMPEG4V3:
1710     case CODEC_ID_WMV1:
1711         if (ENABLE_MSMPEG4_ENCODER)
1712             msmpeg4_encode_mb(s, s->block, motion_x, motion_y);
1713         break;
1714     case CODEC_ID_WMV2:
1715         if (ENABLE_WMV2_ENCODER)
1716             ff_wmv2_encode_mb(s, s->block, motion_x, motion_y);
1717         break;
1718     case CODEC_ID_H261:
1719         if (ENABLE_H261_ENCODER)
1720             ff_h261_encode_mb(s, s->block, motion_x, motion_y);
1721         break;
1722     case CODEC_ID_H263:
1723     case CODEC_ID_H263P:
1724     case CODEC_ID_FLV1:
1725     case CODEC_ID_RV10:
1726     case CODEC_ID_RV20:
1727         if (ENABLE_H263_ENCODER || ENABLE_H263P_ENCODER ||
1728             ENABLE_FLV_ENCODER  || ENABLE_RV10_ENCODER  || ENABLE_RV20_ENCODER)
1729             h263_encode_mb(s, s->block, motion_x, motion_y);
1730         break;
1731     case CODEC_ID_MJPEG:
1732         if (ENABLE_MJPEG_ENCODER)
1733             ff_mjpeg_encode_mb(s, s->block);
1734         break;
1735     default:
1736         assert(0);
1737     }
1738 }
1739
1740 static av_always_inline void encode_mb(MpegEncContext *s, int motion_x, int motion_y)
1741 {
1742     if (s->chroma_format == CHROMA_420) encode_mb_internal(s, motion_x, motion_y,  8, 6);
1743     else                                encode_mb_internal(s, motion_x, motion_y, 16, 8);
1744 }
1745
1746 static inline void copy_context_before_encode(MpegEncContext *d, MpegEncContext *s, int type){
1747     int i;
1748
1749     memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
1750
1751     /* mpeg1 */
1752     d->mb_skip_run= s->mb_skip_run;
1753     for(i=0; i<3; i++)
1754         d->last_dc[i]= s->last_dc[i];
1755
1756     /* statistics */
1757     d->mv_bits= s->mv_bits;
1758     d->i_tex_bits= s->i_tex_bits;
1759     d->p_tex_bits= s->p_tex_bits;
1760     d->i_count= s->i_count;
1761     d->f_count= s->f_count;
1762     d->b_count= s->b_count;
1763     d->skip_count= s->skip_count;
1764     d->misc_bits= s->misc_bits;
1765     d->last_bits= 0;
1766
1767     d->mb_skipped= 0;
1768     d->qscale= s->qscale;
1769     d->dquant= s->dquant;
1770
1771     d->esc3_level_length= s->esc3_level_length;
1772 }
1773
1774 static inline void copy_context_after_encode(MpegEncContext *d, MpegEncContext *s, int type){
1775     int i;
1776
1777     memcpy(d->mv, s->mv, 2*4*2*sizeof(int));
1778     memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
1779
1780     /* mpeg1 */
1781     d->mb_skip_run= s->mb_skip_run;
1782     for(i=0; i<3; i++)
1783         d->last_dc[i]= s->last_dc[i];
1784
1785     /* statistics */
1786     d->mv_bits= s->mv_bits;
1787     d->i_tex_bits= s->i_tex_bits;
1788     d->p_tex_bits= s->p_tex_bits;
1789     d->i_count= s->i_count;
1790     d->f_count= s->f_count;
1791     d->b_count= s->b_count;
1792     d->skip_count= s->skip_count;
1793     d->misc_bits= s->misc_bits;
1794
1795     d->mb_intra= s->mb_intra;
1796     d->mb_skipped= s->mb_skipped;
1797     d->mv_type= s->mv_type;
1798     d->mv_dir= s->mv_dir;
1799     d->pb= s->pb;
1800     if(s->data_partitioning){
1801         d->pb2= s->pb2;
1802         d->tex_pb= s->tex_pb;
1803     }
1804     d->block= s->block;
1805     for(i=0; i<8; i++)
1806         d->block_last_index[i]= s->block_last_index[i];
1807     d->interlaced_dct= s->interlaced_dct;
1808     d->qscale= s->qscale;
1809
1810     d->esc3_level_length= s->esc3_level_length;
1811 }
1812
1813 static inline void encode_mb_hq(MpegEncContext *s, MpegEncContext *backup, MpegEncContext *best, int type,
1814                            PutBitContext pb[2], PutBitContext pb2[2], PutBitContext tex_pb[2],
1815                            int *dmin, int *next_block, int motion_x, int motion_y)
1816 {
1817     int score;
1818     uint8_t *dest_backup[3];
1819
1820     copy_context_before_encode(s, backup, type);
1821
1822     s->block= s->blocks[*next_block];
1823     s->pb= pb[*next_block];
1824     if(s->data_partitioning){
1825         s->pb2   = pb2   [*next_block];
1826         s->tex_pb= tex_pb[*next_block];
1827     }
1828
1829     if(*next_block){
1830         memcpy(dest_backup, s->dest, sizeof(s->dest));
1831         s->dest[0] = s->rd_scratchpad;
1832         s->dest[1] = s->rd_scratchpad + 16*s->linesize;
1833         s->dest[2] = s->rd_scratchpad + 16*s->linesize + 8;
1834         assert(s->linesize >= 32); //FIXME
1835     }
1836
1837     encode_mb(s, motion_x, motion_y);
1838
1839     score= put_bits_count(&s->pb);
1840     if(s->data_partitioning){
1841         score+= put_bits_count(&s->pb2);
1842         score+= put_bits_count(&s->tex_pb);
1843     }
1844
1845     if(s->avctx->mb_decision == FF_MB_DECISION_RD){
1846         MPV_decode_mb(s, s->block);
1847
1848         score *= s->lambda2;
1849         score += sse_mb(s) << FF_LAMBDA_SHIFT;
1850     }
1851
1852     if(*next_block){
1853         memcpy(s->dest, dest_backup, sizeof(s->dest));
1854     }
1855
1856     if(score<*dmin){
1857         *dmin= score;
1858         *next_block^=1;
1859
1860         copy_context_after_encode(best, s, type);
1861     }
1862 }
1863
1864 static int sse(MpegEncContext *s, uint8_t *src1, uint8_t *src2, int w, int h, int stride){
1865     uint32_t *sq = ff_squareTbl + 256;
1866     int acc=0;
1867     int x,y;
1868
1869     if(w==16 && h==16)
1870         return s->dsp.sse[0](NULL, src1, src2, stride, 16);
1871     else if(w==8 && h==8)
1872         return s->dsp.sse[1](NULL, src1, src2, stride, 8);
1873
1874     for(y=0; y<h; y++){
1875         for(x=0; x<w; x++){
1876             acc+= sq[src1[x + y*stride] - src2[x + y*stride]];
1877         }
1878     }
1879
1880     assert(acc>=0);
1881
1882     return acc;
1883 }
1884
1885 static int sse_mb(MpegEncContext *s){
1886     int w= 16;
1887     int h= 16;
1888
1889     if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
1890     if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
1891
1892     if(w==16 && h==16)
1893       if(s->avctx->mb_cmp == FF_CMP_NSSE){
1894         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)
1895                +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)
1896                +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);
1897       }else{
1898         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)
1899                +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)
1900                +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);
1901       }
1902     else
1903         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)
1904                +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)
1905                +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);
1906 }
1907
1908 static int pre_estimate_motion_thread(AVCodecContext *c, void *arg){
1909     MpegEncContext *s= arg;
1910
1911
1912     s->me.pre_pass=1;
1913     s->me.dia_size= s->avctx->pre_dia_size;
1914     s->first_slice_line=1;
1915     for(s->mb_y= s->end_mb_y-1; s->mb_y >= s->start_mb_y; s->mb_y--) {
1916         for(s->mb_x=s->mb_width-1; s->mb_x >=0 ;s->mb_x--) {
1917             ff_pre_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
1918         }
1919         s->first_slice_line=0;
1920     }
1921
1922     s->me.pre_pass=0;
1923
1924     return 0;
1925 }
1926
1927 static int estimate_motion_thread(AVCodecContext *c, void *arg){
1928     MpegEncContext *s= arg;
1929
1930     ff_check_alignment();
1931
1932     s->me.dia_size= s->avctx->dia_size;
1933     s->first_slice_line=1;
1934     for(s->mb_y= s->start_mb_y; s->mb_y < s->end_mb_y; s->mb_y++) {
1935         s->mb_x=0; //for block init below
1936         ff_init_block_index(s);
1937         for(s->mb_x=0; s->mb_x < s->mb_width; s->mb_x++) {
1938             s->block_index[0]+=2;
1939             s->block_index[1]+=2;
1940             s->block_index[2]+=2;
1941             s->block_index[3]+=2;
1942
1943             /* compute motion vector & mb_type and store in context */
1944             if(s->pict_type==B_TYPE)
1945                 ff_estimate_b_frame_motion(s, s->mb_x, s->mb_y);
1946             else
1947                 ff_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
1948         }
1949         s->first_slice_line=0;
1950     }
1951     return 0;
1952 }
1953
1954 static int mb_var_thread(AVCodecContext *c, void *arg){
1955     MpegEncContext *s= arg;
1956     int mb_x, mb_y;
1957
1958     ff_check_alignment();
1959
1960     for(mb_y=s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
1961         for(mb_x=0; mb_x < s->mb_width; mb_x++) {
1962             int xx = mb_x * 16;
1963             int yy = mb_y * 16;
1964             uint8_t *pix = s->new_picture.data[0] + (yy * s->linesize) + xx;
1965             int varc;
1966             int sum = s->dsp.pix_sum(pix, s->linesize);
1967
1968             varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500 + 128)>>8;
1969
1970             s->current_picture.mb_var [s->mb_stride * mb_y + mb_x] = varc;
1971             s->current_picture.mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
1972             s->me.mb_var_sum_temp    += varc;
1973         }
1974     }
1975     return 0;
1976 }
1977
1978 static void write_slice_end(MpegEncContext *s){
1979     if(ENABLE_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4){
1980         if(s->partitioned_frame){
1981             ff_mpeg4_merge_partitions(s);
1982         }
1983
1984         ff_mpeg4_stuffing(&s->pb);
1985     }else if(ENABLE_MJPEG_ENCODER && s->out_format == FMT_MJPEG){
1986         ff_mjpeg_encode_stuffing(&s->pb);
1987     }
1988
1989     align_put_bits(&s->pb);
1990     flush_put_bits(&s->pb);
1991
1992     if((s->flags&CODEC_FLAG_PASS1) && !s->partitioned_frame)
1993         s->misc_bits+= get_bits_diff(s);
1994 }
1995
1996 static int encode_thread(AVCodecContext *c, void *arg){
1997     MpegEncContext *s= arg;
1998     int mb_x, mb_y, pdif = 0;
1999     int i, j;
2000     MpegEncContext best_s, backup_s;
2001     uint8_t bit_buf[2][MAX_MB_BYTES];
2002     uint8_t bit_buf2[2][MAX_MB_BYTES];
2003     uint8_t bit_buf_tex[2][MAX_MB_BYTES];
2004     PutBitContext pb[2], pb2[2], tex_pb[2];
2005 //printf("%d->%d\n", s->resync_mb_y, s->end_mb_y);
2006
2007     ff_check_alignment();
2008
2009     for(i=0; i<2; i++){
2010         init_put_bits(&pb    [i], bit_buf    [i], MAX_MB_BYTES);
2011         init_put_bits(&pb2   [i], bit_buf2   [i], MAX_MB_BYTES);
2012         init_put_bits(&tex_pb[i], bit_buf_tex[i], MAX_MB_BYTES);
2013     }
2014
2015     s->last_bits= put_bits_count(&s->pb);
2016     s->mv_bits=0;
2017     s->misc_bits=0;
2018     s->i_tex_bits=0;
2019     s->p_tex_bits=0;
2020     s->i_count=0;
2021     s->f_count=0;
2022     s->b_count=0;
2023     s->skip_count=0;
2024
2025     for(i=0; i<3; i++){
2026         /* init last dc values */
2027         /* note: quant matrix value (8) is implied here */
2028         s->last_dc[i] = 128 << s->intra_dc_precision;
2029
2030         s->current_picture.error[i] = 0;
2031     }
2032     s->mb_skip_run = 0;
2033     memset(s->last_mv, 0, sizeof(s->last_mv));
2034
2035     s->last_mv_dir = 0;
2036
2037     switch(s->codec_id){
2038     case CODEC_ID_H263:
2039     case CODEC_ID_H263P:
2040     case CODEC_ID_FLV1:
2041         if (ENABLE_H263_ENCODER || ENABLE_H263P_ENCODER || ENABLE_FLV_ENCODER)
2042             s->gob_index = ff_h263_get_gob_height(s);
2043         break;
2044     case CODEC_ID_MPEG4:
2045         if(ENABLE_MPEG4_ENCODER && s->partitioned_frame)
2046             ff_mpeg4_init_partitions(s);
2047         break;
2048     }
2049
2050     s->resync_mb_x=0;
2051     s->resync_mb_y=0;
2052     s->first_slice_line = 1;
2053     s->ptr_lastgob = s->pb.buf;
2054     for(mb_y= s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
2055 //    printf("row %d at %X\n", s->mb_y, (int)s);
2056         s->mb_x=0;
2057         s->mb_y= mb_y;
2058
2059         ff_set_qscale(s, s->qscale);
2060         ff_init_block_index(s);
2061
2062         for(mb_x=0; mb_x < s->mb_width; mb_x++) {
2063             int xy= mb_y*s->mb_stride + mb_x; // removed const, H261 needs to adjust this
2064             int mb_type= s->mb_type[xy];
2065 //            int d;
2066             int dmin= INT_MAX;
2067             int dir;
2068
2069             if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < MAX_MB_BYTES){
2070                 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
2071                 return -1;
2072             }
2073             if(s->data_partitioning){
2074                 if(   s->pb2   .buf_end - s->pb2   .buf - (put_bits_count(&s->    pb2)>>3) < MAX_MB_BYTES
2075                    || s->tex_pb.buf_end - s->tex_pb.buf - (put_bits_count(&s->tex_pb )>>3) < MAX_MB_BYTES){
2076                     av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
2077                     return -1;
2078                 }
2079             }
2080
2081             s->mb_x = mb_x;
2082             s->mb_y = mb_y;  // moved into loop, can get changed by H.261
2083             ff_update_block_index(s);
2084
2085             if(ENABLE_H261_ENCODER && s->codec_id == CODEC_ID_H261){
2086                 ff_h261_reorder_mb_index(s);
2087                 xy= s->mb_y*s->mb_stride + s->mb_x;
2088                 mb_type= s->mb_type[xy];
2089             }
2090
2091             /* write gob / video packet header  */
2092             if(s->rtp_mode){
2093                 int current_packet_size, is_gob_start;
2094
2095                 current_packet_size= ((put_bits_count(&s->pb)+7)>>3) - (s->ptr_lastgob - s->pb.buf);
2096
2097                 is_gob_start= s->avctx->rtp_payload_size && current_packet_size >= s->avctx->rtp_payload_size && mb_y + mb_x>0;
2098
2099                 if(s->start_mb_y == mb_y && mb_y > 0 && mb_x==0) is_gob_start=1;
2100
2101                 switch(s->codec_id){
2102                 case CODEC_ID_H263:
2103                 case CODEC_ID_H263P:
2104                     if(!s->h263_slice_structured)
2105                         if(s->mb_x || s->mb_y%s->gob_index) is_gob_start=0;
2106                     break;
2107                 case CODEC_ID_MPEG2VIDEO:
2108                     if(s->mb_x==0 && s->mb_y!=0) is_gob_start=1;
2109                 case CODEC_ID_MPEG1VIDEO:
2110                     if(s->mb_skip_run) is_gob_start=0;
2111                     break;
2112                 }
2113
2114                 if(is_gob_start){
2115                     if(s->start_mb_y != mb_y || mb_x!=0){
2116                         write_slice_end(s);
2117
2118                         if(ENABLE_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4 && s->partitioned_frame){
2119                             ff_mpeg4_init_partitions(s);
2120                         }
2121                     }
2122
2123                     assert((put_bits_count(&s->pb)&7) == 0);
2124                     current_packet_size= pbBufPtr(&s->pb) - s->ptr_lastgob;
2125
2126                     if(s->avctx->error_rate && s->resync_mb_x + s->resync_mb_y > 0){
2127                         int r= put_bits_count(&s->pb)/8 + s->picture_number + 16 + s->mb_x + s->mb_y;
2128                         int d= 100 / s->avctx->error_rate;
2129                         if(r % d == 0){
2130                             current_packet_size=0;
2131 #ifndef ALT_BITSTREAM_WRITER
2132                             s->pb.buf_ptr= s->ptr_lastgob;
2133 #endif
2134                             assert(pbBufPtr(&s->pb) == s->ptr_lastgob);
2135                         }
2136                     }
2137
2138                     if (s->avctx->rtp_callback){
2139                         int number_mb = (mb_y - s->resync_mb_y)*s->mb_width + mb_x - s->resync_mb_x;
2140                         s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, current_packet_size, number_mb);
2141                     }
2142
2143                     switch(s->codec_id){
2144                     case CODEC_ID_MPEG4:
2145                         if (ENABLE_MPEG4_ENCODER) {
2146                             ff_mpeg4_encode_video_packet_header(s);
2147                             ff_mpeg4_clean_buffers(s);
2148                         }
2149                     break;
2150                     case CODEC_ID_MPEG1VIDEO:
2151                     case CODEC_ID_MPEG2VIDEO:
2152                         if (ENABLE_MPEG1VIDEO_ENCODER || ENABLE_MPEG2VIDEO_ENCODER) {
2153                             ff_mpeg1_encode_slice_header(s);
2154                             ff_mpeg1_clean_buffers(s);
2155                         }
2156                     break;
2157                     case CODEC_ID_H263:
2158                     case CODEC_ID_H263P:
2159                         if (ENABLE_H263_ENCODER || ENABLE_H263P_ENCODER)
2160                             h263_encode_gob_header(s, mb_y);
2161                     break;
2162                     }
2163
2164                     if(s->flags&CODEC_FLAG_PASS1){
2165                         int bits= put_bits_count(&s->pb);
2166                         s->misc_bits+= bits - s->last_bits;
2167                         s->last_bits= bits;
2168                     }
2169
2170                     s->ptr_lastgob += current_packet_size;
2171                     s->first_slice_line=1;
2172                     s->resync_mb_x=mb_x;
2173                     s->resync_mb_y=mb_y;
2174                 }
2175             }
2176
2177             if(  (s->resync_mb_x   == s->mb_x)
2178                && s->resync_mb_y+1 == s->mb_y){
2179                 s->first_slice_line=0;
2180             }
2181
2182             s->mb_skipped=0;
2183             s->dquant=0; //only for QP_RD
2184
2185             if(mb_type & (mb_type-1) || (s->flags & CODEC_FLAG_QP_RD)){ // more than 1 MB type possible or CODEC_FLAG_QP_RD
2186                 int next_block=0;
2187                 int pb_bits_count, pb2_bits_count, tex_pb_bits_count;
2188
2189                 copy_context_before_encode(&backup_s, s, -1);
2190                 backup_s.pb= s->pb;
2191                 best_s.data_partitioning= s->data_partitioning;
2192                 best_s.partitioned_frame= s->partitioned_frame;
2193                 if(s->data_partitioning){
2194                     backup_s.pb2= s->pb2;
2195                     backup_s.tex_pb= s->tex_pb;
2196                 }
2197
2198                 if(mb_type&CANDIDATE_MB_TYPE_INTER){
2199                     s->mv_dir = MV_DIR_FORWARD;
2200                     s->mv_type = MV_TYPE_16X16;
2201                     s->mb_intra= 0;
2202                     s->mv[0][0][0] = s->p_mv_table[xy][0];
2203                     s->mv[0][0][1] = s->p_mv_table[xy][1];
2204                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER, pb, pb2, tex_pb,
2205                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2206                 }
2207                 if(mb_type&CANDIDATE_MB_TYPE_INTER_I){
2208                     s->mv_dir = MV_DIR_FORWARD;
2209                     s->mv_type = MV_TYPE_FIELD;
2210                     s->mb_intra= 0;
2211                     for(i=0; i<2; i++){
2212                         j= s->field_select[0][i] = s->p_field_select_table[i][xy];
2213                         s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
2214                         s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
2215                     }
2216                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER_I, pb, pb2, tex_pb,
2217                                  &dmin, &next_block, 0, 0);
2218                 }
2219                 if(mb_type&CANDIDATE_MB_TYPE_SKIPPED){
2220                     s->mv_dir = MV_DIR_FORWARD;
2221                     s->mv_type = MV_TYPE_16X16;
2222                     s->mb_intra= 0;
2223                     s->mv[0][0][0] = 0;
2224                     s->mv[0][0][1] = 0;
2225                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_SKIPPED, pb, pb2, tex_pb,
2226                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2227                 }
2228                 if(mb_type&CANDIDATE_MB_TYPE_INTER4V){
2229                     s->mv_dir = MV_DIR_FORWARD;
2230                     s->mv_type = MV_TYPE_8X8;
2231                     s->mb_intra= 0;
2232                     for(i=0; i<4; i++){
2233                         s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
2234                         s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
2235                     }
2236                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER4V, pb, pb2, tex_pb,
2237                                  &dmin, &next_block, 0, 0);
2238                 }
2239                 if(mb_type&CANDIDATE_MB_TYPE_FORWARD){
2240                     s->mv_dir = MV_DIR_FORWARD;
2241                     s->mv_type = MV_TYPE_16X16;
2242                     s->mb_intra= 0;
2243                     s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
2244                     s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2245                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD, pb, pb2, tex_pb,
2246                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2247                 }
2248                 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD){
2249                     s->mv_dir = MV_DIR_BACKWARD;
2250                     s->mv_type = MV_TYPE_16X16;
2251                     s->mb_intra= 0;
2252                     s->mv[1][0][0] = s->b_back_mv_table[xy][0];
2253                     s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2254                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD, pb, pb2, tex_pb,
2255                                  &dmin, &next_block, s->mv[1][0][0], s->mv[1][0][1]);
2256                 }
2257                 if(mb_type&CANDIDATE_MB_TYPE_BIDIR){
2258                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2259                     s->mv_type = MV_TYPE_16X16;
2260                     s->mb_intra= 0;
2261                     s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
2262                     s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
2263                     s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
2264                     s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2265                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR, pb, pb2, tex_pb,
2266                                  &dmin, &next_block, 0, 0);
2267                 }
2268                 if(mb_type&CANDIDATE_MB_TYPE_FORWARD_I){
2269                     s->mv_dir = MV_DIR_FORWARD;
2270                     s->mv_type = MV_TYPE_FIELD;
2271                     s->mb_intra= 0;
2272                     for(i=0; i<2; i++){
2273                         j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
2274                         s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
2275                         s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
2276                     }
2277                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD_I, pb, pb2, tex_pb,
2278                                  &dmin, &next_block, 0, 0);
2279                 }
2280                 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD_I){
2281                     s->mv_dir = MV_DIR_BACKWARD;
2282                     s->mv_type = MV_TYPE_FIELD;
2283                     s->mb_intra= 0;
2284                     for(i=0; i<2; i++){
2285                         j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
2286                         s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
2287                         s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
2288                     }
2289                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD_I, pb, pb2, tex_pb,
2290                                  &dmin, &next_block, 0, 0);
2291                 }
2292                 if(mb_type&CANDIDATE_MB_TYPE_BIDIR_I){
2293                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2294                     s->mv_type = MV_TYPE_FIELD;
2295                     s->mb_intra= 0;
2296                     for(dir=0; dir<2; dir++){
2297                         for(i=0; i<2; i++){
2298                             j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
2299                             s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
2300                             s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
2301                         }
2302                     }
2303                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR_I, pb, pb2, tex_pb,
2304                                  &dmin, &next_block, 0, 0);
2305                 }
2306                 if(mb_type&CANDIDATE_MB_TYPE_INTRA){
2307                     s->mv_dir = 0;
2308                     s->mv_type = MV_TYPE_16X16;
2309                     s->mb_intra= 1;
2310                     s->mv[0][0][0] = 0;
2311                     s->mv[0][0][1] = 0;
2312                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTRA, pb, pb2, tex_pb,
2313                                  &dmin, &next_block, 0, 0);
2314                     if(s->h263_pred || s->h263_aic){
2315                         if(best_s.mb_intra)
2316                             s->mbintra_table[mb_x + mb_y*s->mb_stride]=1;
2317                         else
2318                             ff_clean_intra_table_entries(s); //old mode?
2319                     }
2320                 }
2321
2322                 if((s->flags & CODEC_FLAG_QP_RD) && dmin < INT_MAX){
2323                     if(best_s.mv_type==MV_TYPE_16X16){ //FIXME move 4mv after QPRD
2324                         const int last_qp= backup_s.qscale;
2325                         int qpi, qp, dc[6];
2326                         DCTELEM ac[6][16];
2327                         const int mvdir= (best_s.mv_dir&MV_DIR_BACKWARD) ? 1 : 0;
2328                         static const int dquant_tab[4]={-1,1,-2,2};
2329
2330                         assert(backup_s.dquant == 0);
2331
2332                         //FIXME intra
2333                         s->mv_dir= best_s.mv_dir;
2334                         s->mv_type = MV_TYPE_16X16;
2335                         s->mb_intra= best_s.mb_intra;
2336                         s->mv[0][0][0] = best_s.mv[0][0][0];
2337                         s->mv[0][0][1] = best_s.mv[0][0][1];
2338                         s->mv[1][0][0] = best_s.mv[1][0][0];
2339                         s->mv[1][0][1] = best_s.mv[1][0][1];
2340
2341                         qpi = s->pict_type == B_TYPE ? 2 : 0;
2342                         for(; qpi<4; qpi++){
2343                             int dquant= dquant_tab[qpi];
2344                             qp= last_qp + dquant;
2345                             if(qp < s->avctx->qmin || qp > s->avctx->qmax)
2346                                 continue;
2347                             backup_s.dquant= dquant;
2348                             if(s->mb_intra && s->dc_val[0]){
2349                                 for(i=0; i<6; i++){
2350                                     dc[i]= s->dc_val[0][ s->block_index[i] ];
2351                                     memcpy(ac[i], s->ac_val[0][s->block_index[i]], sizeof(DCTELEM)*16);
2352                                 }
2353                             }
2354
2355                             encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
2356                                          &dmin, &next_block, s->mv[mvdir][0][0], s->mv[mvdir][0][1]);
2357                             if(best_s.qscale != qp){
2358                                 if(s->mb_intra && s->dc_val[0]){
2359                                     for(i=0; i<6; i++){
2360                                         s->dc_val[0][ s->block_index[i] ]= dc[i];
2361                                         memcpy(s->ac_val[0][s->block_index[i]], ac[i], sizeof(DCTELEM)*16);
2362                                     }
2363                                 }
2364                             }
2365                         }
2366                     }
2367                 }
2368                 if(ENABLE_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT){
2369                     int mx= s->b_direct_mv_table[xy][0];
2370                     int my= s->b_direct_mv_table[xy][1];
2371
2372                     backup_s.dquant = 0;
2373                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2374                     s->mb_intra= 0;
2375                     ff_mpeg4_set_direct_mv(s, mx, my);
2376                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
2377                                  &dmin, &next_block, mx, my);
2378                 }
2379                 if(ENABLE_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT0){
2380                     backup_s.dquant = 0;
2381                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2382                     s->mb_intra= 0;
2383                     ff_mpeg4_set_direct_mv(s, 0, 0);
2384                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
2385                                  &dmin, &next_block, 0, 0);
2386                 }
2387                 if(!best_s.mb_intra && s->flags2&CODEC_FLAG2_SKIP_RD){
2388                     int coded=0;
2389                     for(i=0; i<6; i++)
2390                         coded |= s->block_last_index[i];
2391                     if(coded){
2392                         int mx,my;
2393                         memcpy(s->mv, best_s.mv, sizeof(s->mv));
2394                         if(ENABLE_MPEG4_ENCODER && best_s.mv_dir & MV_DIRECT){
2395                             mx=my=0; //FIXME find the one we actually used
2396                             ff_mpeg4_set_direct_mv(s, mx, my);
2397                         }else if(best_s.mv_dir&MV_DIR_BACKWARD){
2398                             mx= s->mv[1][0][0];
2399                             my= s->mv[1][0][1];
2400                         }else{
2401                             mx= s->mv[0][0][0];
2402                             my= s->mv[0][0][1];
2403                         }
2404
2405                         s->mv_dir= best_s.mv_dir;
2406                         s->mv_type = best_s.mv_type;
2407                         s->mb_intra= 0;
2408 /*                        s->mv[0][0][0] = best_s.mv[0][0][0];
2409                         s->mv[0][0][1] = best_s.mv[0][0][1];
2410                         s->mv[1][0][0] = best_s.mv[1][0][0];
2411                         s->mv[1][0][1] = best_s.mv[1][0][1];*/
2412                         backup_s.dquant= 0;
2413                         s->skipdct=1;
2414                         encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
2415                                         &dmin, &next_block, mx, my);
2416                         s->skipdct=0;
2417                     }
2418                 }
2419
2420                 s->current_picture.qscale_table[xy]= best_s.qscale;
2421
2422                 copy_context_after_encode(s, &best_s, -1);
2423
2424                 pb_bits_count= put_bits_count(&s->pb);
2425                 flush_put_bits(&s->pb);
2426                 ff_copy_bits(&backup_s.pb, bit_buf[next_block^1], pb_bits_count);
2427                 s->pb= backup_s.pb;
2428
2429                 if(s->data_partitioning){
2430                     pb2_bits_count= put_bits_count(&s->pb2);
2431                     flush_put_bits(&s->pb2);
2432                     ff_copy_bits(&backup_s.pb2, bit_buf2[next_block^1], pb2_bits_count);
2433                     s->pb2= backup_s.pb2;
2434
2435                     tex_pb_bits_count= put_bits_count(&s->tex_pb);
2436                     flush_put_bits(&s->tex_pb);
2437                     ff_copy_bits(&backup_s.tex_pb, bit_buf_tex[next_block^1], tex_pb_bits_count);
2438                     s->tex_pb= backup_s.tex_pb;
2439                 }
2440                 s->last_bits= put_bits_count(&s->pb);
2441
2442                 if (ENABLE_ANY_H263_ENCODER &&
2443                     s->out_format == FMT_H263 && s->pict_type!=B_TYPE)
2444                     ff_h263_update_motion_val(s);
2445
2446                 if(next_block==0){ //FIXME 16 vs linesize16
2447                     s->dsp.put_pixels_tab[0][0](s->dest[0], s->rd_scratchpad                     , s->linesize  ,16);
2448                     s->dsp.put_pixels_tab[1][0](s->dest[1], s->rd_scratchpad + 16*s->linesize    , s->uvlinesize, 8);
2449                     s->dsp.put_pixels_tab[1][0](s->dest[2], s->rd_scratchpad + 16*s->linesize + 8, s->uvlinesize, 8);
2450                 }
2451
2452                 if(s->avctx->mb_decision == FF_MB_DECISION_BITS)
2453                     MPV_decode_mb(s, s->block);
2454             } else {
2455                 int motion_x = 0, motion_y = 0;
2456                 s->mv_type=MV_TYPE_16X16;
2457                 // only one MB-Type possible
2458
2459                 switch(mb_type){
2460                 case CANDIDATE_MB_TYPE_INTRA:
2461                     s->mv_dir = 0;
2462                     s->mb_intra= 1;
2463                     motion_x= s->mv[0][0][0] = 0;
2464                     motion_y= s->mv[0][0][1] = 0;
2465                     break;
2466                 case CANDIDATE_MB_TYPE_INTER:
2467                     s->mv_dir = MV_DIR_FORWARD;
2468                     s->mb_intra= 0;
2469                     motion_x= s->mv[0][0][0] = s->p_mv_table[xy][0];
2470                     motion_y= s->mv[0][0][1] = s->p_mv_table[xy][1];
2471                     break;
2472                 case CANDIDATE_MB_TYPE_INTER_I:
2473                     s->mv_dir = MV_DIR_FORWARD;
2474                     s->mv_type = MV_TYPE_FIELD;
2475                     s->mb_intra= 0;
2476                     for(i=0; i<2; i++){
2477                         j= s->field_select[0][i] = s->p_field_select_table[i][xy];
2478                         s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
2479                         s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
2480                     }
2481                     break;
2482                 case CANDIDATE_MB_TYPE_INTER4V:
2483                     s->mv_dir = MV_DIR_FORWARD;
2484                     s->mv_type = MV_TYPE_8X8;
2485                     s->mb_intra= 0;
2486                     for(i=0; i<4; i++){
2487                         s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
2488                         s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
2489                     }
2490                     break;
2491                 case CANDIDATE_MB_TYPE_DIRECT:
2492                     if (ENABLE_MPEG4_ENCODER) {
2493                         s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
2494                         s->mb_intra= 0;
2495                         motion_x=s->b_direct_mv_table[xy][0];
2496                         motion_y=s->b_direct_mv_table[xy][1];
2497                         ff_mpeg4_set_direct_mv(s, motion_x, motion_y);
2498                     }
2499                     break;
2500                 case CANDIDATE_MB_TYPE_DIRECT0:
2501                     if (ENABLE_MPEG4_ENCODER) {
2502                         s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
2503                         s->mb_intra= 0;
2504                         ff_mpeg4_set_direct_mv(s, 0, 0);
2505                     }
2506                     break;
2507                 case CANDIDATE_MB_TYPE_BIDIR:
2508                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2509                     s->mb_intra= 0;
2510                     s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
2511                     s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
2512                     s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
2513                     s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2514                     break;
2515                 case CANDIDATE_MB_TYPE_BACKWARD:
2516                     s->mv_dir = MV_DIR_BACKWARD;
2517                     s->mb_intra= 0;
2518                     motion_x= s->mv[1][0][0] = s->b_back_mv_table[xy][0];
2519                     motion_y= s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2520                     break;
2521                 case CANDIDATE_MB_TYPE_FORWARD:
2522                     s->mv_dir = MV_DIR_FORWARD;
2523                     s->mb_intra= 0;
2524                     motion_x= s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
2525                     motion_y= s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2526 //                    printf(" %d %d ", motion_x, motion_y);
2527                     break;
2528                 case CANDIDATE_MB_TYPE_FORWARD_I:
2529                     s->mv_dir = MV_DIR_FORWARD;
2530                     s->mv_type = MV_TYPE_FIELD;
2531                     s->mb_intra= 0;
2532                     for(i=0; i<2; i++){
2533                         j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
2534                         s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
2535                         s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
2536                     }
2537                     break;
2538                 case CANDIDATE_MB_TYPE_BACKWARD_I:
2539                     s->mv_dir = MV_DIR_BACKWARD;
2540                     s->mv_type = MV_TYPE_FIELD;
2541                     s->mb_intra= 0;
2542                     for(i=0; i<2; i++){
2543                         j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
2544                         s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
2545                         s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
2546                     }
2547                     break;
2548                 case CANDIDATE_MB_TYPE_BIDIR_I:
2549                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2550                     s->mv_type = MV_TYPE_FIELD;
2551                     s->mb_intra= 0;
2552                     for(dir=0; dir<2; dir++){
2553                         for(i=0; i<2; i++){
2554                             j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
2555                             s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
2556                             s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
2557                         }
2558                     }
2559                     break;
2560                 default:
2561                     av_log(s->avctx, AV_LOG_ERROR, "illegal MB type\n");
2562                 }
2563
2564                 encode_mb(s, motion_x, motion_y);
2565
2566                 // RAL: Update last macroblock type
2567                 s->last_mv_dir = s->mv_dir;
2568
2569                 if (ENABLE_ANY_H263_ENCODER &&
2570                     s->out_format == FMT_H263 && s->pict_type!=B_TYPE)
2571                     ff_h263_update_motion_val(s);
2572
2573                 MPV_decode_mb(s, s->block);
2574             }
2575
2576             /* clean the MV table in IPS frames for direct mode in B frames */
2577             if(s->mb_intra /* && I,P,S_TYPE */){
2578                 s->p_mv_table[xy][0]=0;
2579                 s->p_mv_table[xy][1]=0;
2580             }
2581
2582             if(s->flags&CODEC_FLAG_PSNR){
2583                 int w= 16;
2584                 int h= 16;
2585
2586                 if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
2587                 if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
2588
2589                 s->current_picture.error[0] += sse(
2590                     s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16,
2591                     s->dest[0], w, h, s->linesize);
2592                 s->current_picture.error[1] += sse(
2593                     s, s->new_picture.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,
2594                     s->dest[1], w>>1, h>>1, s->uvlinesize);
2595                 s->current_picture.error[2] += sse(
2596                     s, s->new_picture    .data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,
2597                     s->dest[2], w>>1, h>>1, s->uvlinesize);
2598             }
2599             if(s->loop_filter){
2600                 if(ENABLE_ANY_H263_ENCODER && s->out_format == FMT_H263)
2601                     ff_h263_loop_filter(s);
2602             }
2603 //printf("MB %d %d bits\n", s->mb_x+s->mb_y*s->mb_stride, put_bits_count(&s->pb));
2604         }
2605     }
2606
2607     //not beautiful here but we must write it before flushing so it has to be here
2608     if (ENABLE_MSMPEG4_ENCODER && s->msmpeg4_version && s->msmpeg4_version<4 && s->pict_type == I_TYPE)
2609         msmpeg4_encode_ext_header(s);
2610
2611     write_slice_end(s);
2612
2613     /* Send the last GOB if RTP */
2614     if (s->avctx->rtp_callback) {
2615         int number_mb = (mb_y - s->resync_mb_y)*s->mb_width - s->resync_mb_x;
2616         pdif = pbBufPtr(&s->pb) - s->ptr_lastgob;
2617         /* Call the RTP callback to send the last GOB */
2618         emms_c();
2619         s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, pdif, number_mb);
2620     }
2621
2622     return 0;
2623 }
2624
2625 #define MERGE(field) dst->field += src->field; src->field=0
2626 static void merge_context_after_me(MpegEncContext *dst, MpegEncContext *src){
2627     MERGE(me.scene_change_score);
2628     MERGE(me.mc_mb_var_sum_temp);
2629     MERGE(me.mb_var_sum_temp);
2630 }
2631
2632 static void merge_context_after_encode(MpegEncContext *dst, MpegEncContext *src){
2633     int i;
2634
2635     MERGE(dct_count[0]); //note, the other dct vars are not part of the context
2636     MERGE(dct_count[1]);
2637     MERGE(mv_bits);
2638     MERGE(i_tex_bits);
2639     MERGE(p_tex_bits);
2640     MERGE(i_count);
2641     MERGE(f_count);
2642     MERGE(b_count);
2643     MERGE(skip_count);
2644     MERGE(misc_bits);
2645     MERGE(error_count);
2646     MERGE(padding_bug_score);
2647     MERGE(current_picture.error[0]);
2648     MERGE(current_picture.error[1]);
2649     MERGE(current_picture.error[2]);
2650
2651     if(dst->avctx->noise_reduction){
2652         for(i=0; i<64; i++){
2653             MERGE(dct_error_sum[0][i]);
2654             MERGE(dct_error_sum[1][i]);
2655         }
2656     }
2657
2658     assert(put_bits_count(&src->pb) % 8 ==0);
2659     assert(put_bits_count(&dst->pb) % 8 ==0);
2660     ff_copy_bits(&dst->pb, src->pb.buf, put_bits_count(&src->pb));
2661     flush_put_bits(&dst->pb);
2662 }
2663
2664 static int estimate_qp(MpegEncContext *s, int dry_run){
2665     if (s->next_lambda){
2666         s->current_picture_ptr->quality=
2667         s->current_picture.quality = s->next_lambda;
2668         if(!dry_run) s->next_lambda= 0;
2669     } else if (!s->fixed_qscale) {
2670         s->current_picture_ptr->quality=
2671         s->current_picture.quality = ff_rate_estimate_qscale(s, dry_run);
2672         if (s->current_picture.quality < 0)
2673             return -1;
2674     }
2675
2676     if(s->adaptive_quant){
2677         switch(s->codec_id){
2678         case CODEC_ID_MPEG4:
2679             if (ENABLE_MPEG4_ENCODER)
2680                 ff_clean_mpeg4_qscales(s);
2681             break;
2682         case CODEC_ID_H263:
2683         case CODEC_ID_H263P:
2684         case CODEC_ID_FLV1:
2685             if (ENABLE_H263_ENCODER||ENABLE_H263P_ENCODER||ENABLE_FLV_ENCODER)
2686                 ff_clean_h263_qscales(s);
2687             break;
2688         }
2689
2690         s->lambda= s->lambda_table[0];
2691         //FIXME broken
2692     }else
2693         s->lambda= s->current_picture.quality;
2694 //printf("%d %d\n", s->avctx->global_quality, s->current_picture.quality);
2695     update_qscale(s);
2696     return 0;
2697 }
2698
2699 /* must be called before writing the header */
2700 static void set_frame_distances(MpegEncContext * s){
2701     assert(s->current_picture_ptr->pts != AV_NOPTS_VALUE);
2702     s->time= s->current_picture_ptr->pts*s->avctx->time_base.num;
2703
2704     if(s->pict_type==B_TYPE){
2705         s->pb_time= s->pp_time - (s->last_non_b_time - s->time);
2706         assert(s->pb_time > 0 && s->pb_time < s->pp_time);
2707     }else{
2708         s->pp_time= s->time - s->last_non_b_time;
2709         s->last_non_b_time= s->time;
2710         assert(s->picture_number==0 || s->pp_time > 0);
2711     }
2712 }
2713
2714 static int encode_picture(MpegEncContext *s, int picture_number)
2715 {
2716     int i;
2717     int bits;
2718
2719     s->picture_number = picture_number;
2720
2721     /* Reset the average MB variance */
2722     s->me.mb_var_sum_temp    =
2723     s->me.mc_mb_var_sum_temp = 0;
2724
2725     /* we need to initialize some time vars before we can encode b-frames */
2726     // RAL: Condition added for MPEG1VIDEO
2727     if (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO || (s->h263_pred && !s->h263_msmpeg4))
2728         set_frame_distances(s);
2729     if(ENABLE_MPEG4_ENCODER && s->codec_id == CODEC_ID_MPEG4)
2730         ff_set_mpeg4_time(s);
2731
2732     s->me.scene_change_score=0;
2733
2734 //    s->lambda= s->current_picture_ptr->quality; //FIXME qscale / ... stuff for ME ratedistoration
2735
2736     if(s->pict_type==I_TYPE){
2737         if(s->msmpeg4_version >= 3) s->no_rounding=1;
2738         else                        s->no_rounding=0;
2739     }else if(s->pict_type!=B_TYPE){
2740         if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
2741             s->no_rounding ^= 1;
2742     }
2743
2744     if(s->flags & CODEC_FLAG_PASS2){
2745         if (estimate_qp(s,1) < 0)
2746             return -1;
2747         ff_get_2pass_fcode(s);
2748     }else if(!(s->flags & CODEC_FLAG_QSCALE)){
2749         if(s->pict_type==B_TYPE)
2750             s->lambda= s->last_lambda_for[s->pict_type];
2751         else
2752             s->lambda= s->last_lambda_for[s->last_non_b_pict_type];
2753         update_qscale(s);
2754     }
2755
2756     s->mb_intra=0; //for the rate distortion & bit compare functions
2757     for(i=1; i<s->avctx->thread_count; i++){
2758         ff_update_duplicate_context(s->thread_context[i], s);
2759     }
2760
2761     ff_init_me(s);
2762
2763     /* Estimate motion for every MB */
2764     if(s->pict_type != I_TYPE){
2765         s->lambda = (s->lambda * s->avctx->me_penalty_compensation + 128)>>8;
2766         s->lambda2= (s->lambda2* (int64_t)s->avctx->me_penalty_compensation + 128)>>8;
2767         if(s->pict_type != B_TYPE && s->avctx->me_threshold==0){
2768             if((s->avctx->pre_me && s->last_non_b_pict_type==I_TYPE) || s->avctx->pre_me==2){
2769                 s->avctx->execute(s->avctx, pre_estimate_motion_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count);
2770             }
2771         }
2772
2773         s->avctx->execute(s->avctx, estimate_motion_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count);
2774     }else /* if(s->pict_type == I_TYPE) */{
2775         /* I-Frame */
2776         for(i=0; i<s->mb_stride*s->mb_height; i++)
2777             s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
2778
2779         if(!s->fixed_qscale){
2780             /* finding spatial complexity for I-frame rate control */
2781             s->avctx->execute(s->avctx, mb_var_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count);
2782         }
2783     }
2784     for(i=1; i<s->avctx->thread_count; i++){
2785         merge_context_after_me(s, s->thread_context[i]);
2786     }
2787     s->current_picture.mc_mb_var_sum= s->current_picture_ptr->mc_mb_var_sum= s->me.mc_mb_var_sum_temp;
2788     s->current_picture.   mb_var_sum= s->current_picture_ptr->   mb_var_sum= s->me.   mb_var_sum_temp;
2789     emms_c();
2790
2791     if(s->me.scene_change_score > s->avctx->scenechange_threshold && s->pict_type == P_TYPE){
2792         s->pict_type= I_TYPE;
2793         for(i=0; i<s->mb_stride*s->mb_height; i++)
2794             s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
2795 //printf("Scene change detected, encoding as I Frame %d %d\n", s->current_picture.mb_var_sum, s->current_picture.mc_mb_var_sum);
2796     }
2797
2798     if(!s->umvplus){
2799         if(s->pict_type==P_TYPE || s->pict_type==S_TYPE) {
2800             s->f_code= ff_get_best_fcode(s, s->p_mv_table, CANDIDATE_MB_TYPE_INTER);
2801
2802             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2803                 int a,b;
2804                 a= ff_get_best_fcode(s, s->p_field_mv_table[0][0], CANDIDATE_MB_TYPE_INTER_I); //FIXME field_select
2805                 b= ff_get_best_fcode(s, s->p_field_mv_table[1][1], CANDIDATE_MB_TYPE_INTER_I);
2806                 s->f_code= FFMAX(s->f_code, FFMAX(a,b));
2807             }
2808
2809             ff_fix_long_p_mvs(s);
2810             ff_fix_long_mvs(s, NULL, 0, s->p_mv_table, s->f_code, CANDIDATE_MB_TYPE_INTER, 0);
2811             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2812                 int j;
2813                 for(i=0; i<2; i++){
2814                     for(j=0; j<2; j++)
2815                         ff_fix_long_mvs(s, s->p_field_select_table[i], j,
2816                                         s->p_field_mv_table[i][j], s->f_code, CANDIDATE_MB_TYPE_INTER_I, 0);
2817                 }
2818             }
2819         }
2820
2821         if(s->pict_type==B_TYPE){
2822             int a, b;
2823
2824             a = ff_get_best_fcode(s, s->b_forw_mv_table, CANDIDATE_MB_TYPE_FORWARD);
2825             b = ff_get_best_fcode(s, s->b_bidir_forw_mv_table, CANDIDATE_MB_TYPE_BIDIR);
2826             s->f_code = FFMAX(a, b);
2827
2828             a = ff_get_best_fcode(s, s->b_back_mv_table, CANDIDATE_MB_TYPE_BACKWARD);
2829             b = ff_get_best_fcode(s, s->b_bidir_back_mv_table, CANDIDATE_MB_TYPE_BIDIR);
2830             s->b_code = FFMAX(a, b);
2831
2832             ff_fix_long_mvs(s, NULL, 0, s->b_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_FORWARD, 1);
2833             ff_fix_long_mvs(s, NULL, 0, s->b_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BACKWARD, 1);
2834             ff_fix_long_mvs(s, NULL, 0, s->b_bidir_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_BIDIR, 1);
2835             ff_fix_long_mvs(s, NULL, 0, s->b_bidir_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BIDIR, 1);
2836             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2837                 int dir, j;
2838                 for(dir=0; dir<2; dir++){
2839                     for(i=0; i<2; i++){
2840                         for(j=0; j<2; j++){
2841                             int type= dir ? (CANDIDATE_MB_TYPE_BACKWARD_I|CANDIDATE_MB_TYPE_BIDIR_I)
2842                                           : (CANDIDATE_MB_TYPE_FORWARD_I |CANDIDATE_MB_TYPE_BIDIR_I);
2843                             ff_fix_long_mvs(s, s->b_field_select_table[dir][i], j,
2844                                             s->b_field_mv_table[dir][i][j], dir ? s->b_code : s->f_code, type, 1);
2845                         }
2846                     }
2847                 }
2848             }
2849         }
2850     }
2851
2852     if (estimate_qp(s, 0) < 0)
2853         return -1;
2854
2855     if(s->qscale < 3 && s->max_qcoeff<=128 && s->pict_type==I_TYPE && !(s->flags & CODEC_FLAG_QSCALE))
2856         s->qscale= 3; //reduce clipping problems
2857
2858     if (s->out_format == FMT_MJPEG) {
2859         /* for mjpeg, we do include qscale in the matrix */
2860         s->intra_matrix[0] = ff_mpeg1_default_intra_matrix[0];
2861         for(i=1;i<64;i++){
2862             int j= s->dsp.idct_permutation[i];
2863
2864             s->intra_matrix[j] = av_clip_uint8((ff_mpeg1_default_intra_matrix[i] * s->qscale) >> 3);
2865         }
2866         ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
2867                        s->intra_matrix, s->intra_quant_bias, 8, 8, 1);
2868         s->qscale= 8;
2869     }
2870
2871     //FIXME var duplication
2872     s->current_picture_ptr->key_frame=
2873     s->current_picture.key_frame= s->pict_type == I_TYPE; //FIXME pic_ptr
2874     s->current_picture_ptr->pict_type=
2875     s->current_picture.pict_type= s->pict_type;
2876
2877     if(s->current_picture.key_frame)
2878         s->picture_in_gop_number=0;
2879
2880     s->last_bits= put_bits_count(&s->pb);
2881     switch(s->out_format) {
2882     case FMT_MJPEG:
2883         if (ENABLE_MJPEG_ENCODER)
2884             ff_mjpeg_encode_picture_header(s);
2885         break;
2886     case FMT_H261:
2887         if (ENABLE_H261_ENCODER)
2888             ff_h261_encode_picture_header(s, picture_number);
2889         break;
2890     case FMT_H263:
2891         if (ENABLE_WMV2_ENCODER && s->codec_id == CODEC_ID_WMV2)
2892             ff_wmv2_encode_picture_header(s, picture_number);
2893         else if (ENABLE_MSMPEG4_ENCODER && s->h263_msmpeg4)
2894             msmpeg4_encode_picture_header(s, picture_number);
2895         else if (ENABLE_MPEG4_ENCODER && s->h263_pred)
2896             mpeg4_encode_picture_header(s, picture_number);
2897         else if (ENABLE_RV10_ENCODER && s->codec_id == CODEC_ID_RV10)
2898             rv10_encode_picture_header(s, picture_number);
2899         else if (ENABLE_RV20_ENCODER && s->codec_id == CODEC_ID_RV20)
2900             rv20_encode_picture_header(s, picture_number);
2901         else if (ENABLE_FLV_ENCODER && s->codec_id == CODEC_ID_FLV1)
2902             ff_flv_encode_picture_header(s, picture_number);
2903         else if (ENABLE_ANY_H263_ENCODER)
2904             h263_encode_picture_header(s, picture_number);
2905         break;
2906     case FMT_MPEG1:
2907         if (ENABLE_MPEG1VIDEO_ENCODER || ENABLE_MPEG2VIDEO_ENCODER)
2908             mpeg1_encode_picture_header(s, picture_number);
2909         break;
2910     case FMT_H264:
2911         break;
2912     default:
2913         assert(0);
2914     }
2915     bits= put_bits_count(&s->pb);
2916     s->header_bits= bits - s->last_bits;
2917
2918     for(i=1; i<s->avctx->thread_count; i++){
2919         update_duplicate_context_after_me(s->thread_context[i], s);
2920     }
2921     s->avctx->execute(s->avctx, encode_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count);
2922     for(i=1; i<s->avctx->thread_count; i++){
2923         merge_context_after_encode(s, s->thread_context[i]);
2924     }
2925     emms_c();
2926     return 0;
2927 }
2928
2929 void  denoise_dct_c(MpegEncContext *s, DCTELEM *block){
2930     const int intra= s->mb_intra;
2931     int i;
2932
2933     s->dct_count[intra]++;
2934
2935     for(i=0; i<64; i++){
2936         int level= block[i];
2937
2938         if(level){
2939             if(level>0){
2940                 s->dct_error_sum[intra][i] += level;
2941                 level -= s->dct_offset[intra][i];
2942                 if(level<0) level=0;
2943             }else{
2944                 s->dct_error_sum[intra][i] -= level;
2945                 level += s->dct_offset[intra][i];
2946                 if(level>0) level=0;
2947             }
2948             block[i]= level;
2949         }
2950     }
2951 }
2952
2953 int dct_quantize_trellis_c(MpegEncContext *s,
2954                         DCTELEM *block, int n,
2955                         int qscale, int *overflow){
2956     const int *qmat;
2957     const uint8_t *scantable= s->intra_scantable.scantable;
2958     const uint8_t *perm_scantable= s->intra_scantable.permutated;
2959     int max=0;
2960     unsigned int threshold1, threshold2;
2961     int bias=0;
2962     int run_tab[65];
2963     int level_tab[65];
2964     int score_tab[65];
2965     int survivor[65];
2966     int survivor_count;
2967     int last_run=0;
2968     int last_level=0;
2969     int last_score= 0;
2970     int last_i;
2971     int coeff[2][64];
2972     int coeff_count[64];
2973     int qmul, qadd, start_i, last_non_zero, i, dc;
2974     const int esc_length= s->ac_esc_length;
2975     uint8_t * length;
2976     uint8_t * last_length;
2977     const int lambda= s->lambda2 >> (FF_LAMBDA_SHIFT - 6);
2978
2979     s->dsp.fdct (block);
2980
2981     if(s->dct_error_sum)
2982         s->denoise_dct(s, block);
2983     qmul= qscale*16;
2984     qadd= ((qscale-1)|1)*8;
2985
2986     if (s->mb_intra) {
2987         int q;
2988         if (!s->h263_aic) {
2989             if (n < 4)
2990                 q = s->y_dc_scale;
2991             else
2992                 q = s->c_dc_scale;
2993             q = q << 3;
2994         } else{
2995             /* For AIC we skip quant/dequant of INTRADC */
2996             q = 1 << 3;
2997             qadd=0;
2998         }
2999
3000         /* note: block[0] is assumed to be positive */
3001         block[0] = (block[0] + (q >> 1)) / q;
3002         start_i = 1;
3003         last_non_zero = 0;
3004         qmat = s->q_intra_matrix[qscale];
3005         if(s->mpeg_quant || s->out_format == FMT_MPEG1)
3006             bias= 1<<(QMAT_SHIFT-1);
3007         length     = s->intra_ac_vlc_length;
3008         last_length= s->intra_ac_vlc_last_length;
3009     } else {
3010         start_i = 0;
3011         last_non_zero = -1;
3012         qmat = s->q_inter_matrix[qscale];
3013         length     = s->inter_ac_vlc_length;
3014         last_length= s->inter_ac_vlc_last_length;
3015     }
3016     last_i= start_i;
3017
3018     threshold1= (1<<QMAT_SHIFT) - bias - 1;
3019     threshold2= (threshold1<<1);
3020
3021     for(i=63; i>=start_i; i--) {
3022         const int j = scantable[i];
3023         int level = block[j] * qmat[j];
3024
3025         if(((unsigned)(level+threshold1))>threshold2){
3026             last_non_zero = i;
3027             break;
3028         }
3029     }
3030
3031     for(i=start_i; i<=last_non_zero; i++) {
3032         const int j = scantable[i];
3033         int level = block[j] * qmat[j];
3034
3035 //        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
3036 //           || bias-level >= (1<<(QMAT_SHIFT - 3))){
3037         if(((unsigned)(level+threshold1))>threshold2){
3038             if(level>0){
3039                 level= (bias + level)>>QMAT_SHIFT;
3040                 coeff[0][i]= level;
3041                 coeff[1][i]= level-1;
3042 //                coeff[2][k]= level-2;
3043             }else{
3044                 level= (bias - level)>>QMAT_SHIFT;
3045                 coeff[0][i]= -level;
3046                 coeff[1][i]= -level+1;
3047 //                coeff[2][k]= -level+2;
3048             }
3049             coeff_count[i]= FFMIN(level, 2);
3050             assert(coeff_count[i]);
3051             max |=level;
3052         }else{
3053             coeff[0][i]= (level>>31)|1;
3054             coeff_count[i]= 1;
3055         }
3056     }
3057
3058     *overflow= s->max_qcoeff < max; //overflow might have happened
3059
3060     if(last_non_zero < start_i){
3061         memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
3062         return last_non_zero;
3063     }
3064
3065     score_tab[start_i]= 0;
3066     survivor[0]= start_i;
3067     survivor_count= 1;
3068
3069     for(i=start_i; i<=last_non_zero; i++){
3070         int level_index, j;
3071         const int dct_coeff= FFABS(block[ scantable[i] ]);
3072         const int zero_distoration= dct_coeff*dct_coeff;
3073         int best_score=256*256*256*120;
3074         for(level_index=0; level_index < coeff_count[i]; level_index++){
3075             int distoration;
3076             int level= coeff[level_index][i];
3077             const int alevel= FFABS(level);
3078             int unquant_coeff;
3079
3080             assert(level);
3081
3082             if(s->out_format == FMT_H263){
3083                 unquant_coeff= alevel*qmul + qadd;
3084             }else{ //MPEG1
3085                 j= s->dsp.idct_permutation[ scantable[i] ]; //FIXME optimize
3086                 if(s->mb_intra){
3087                         unquant_coeff = (int)(  alevel  * qscale * s->intra_matrix[j]) >> 3;
3088                         unquant_coeff =   (unquant_coeff - 1) | 1;
3089                 }else{
3090                         unquant_coeff = (((  alevel  << 1) + 1) * qscale * ((int) s->inter_matrix[j])) >> 4;
3091                         unquant_coeff =   (unquant_coeff - 1) | 1;
3092                 }
3093                 unquant_coeff<<= 3;
3094             }
3095
3096             distoration= (unquant_coeff - dct_coeff) * (unquant_coeff - dct_coeff) - zero_distoration;
3097             level+=64;
3098             if((level&(~127)) == 0){
3099                 for(j=survivor_count-1; j>=0; j--){
3100                     int run= i - survivor[j];
3101                     int score= distoration + length[UNI_AC_ENC_INDEX(run, level)]*lambda;
3102                     score += score_tab[i-run];
3103
3104                     if(score < best_score){
3105                         best_score= score;
3106                         run_tab[i+1]= run;
3107                         level_tab[i+1]= level-64;
3108                     }
3109                 }
3110
3111                 if(s->out_format == FMT_H263){
3112                     for(j=survivor_count-1; j>=0; j--){
3113                         int run= i - survivor[j];
3114                         int score= distoration + last_length[UNI_AC_ENC_INDEX(run, level)]*lambda;
3115                         score += score_tab[i-run];
3116                         if(score < last_score){
3117                             last_score= score;
3118                             last_run= run;
3119                             last_level= level-64;
3120                             last_i= i+1;
3121                         }
3122                     }
3123                 }
3124             }else{
3125                 distoration += esc_length*lambda;
3126                 for(j=survivor_count-1; j>=0; j--){
3127                     int run= i - survivor[j];
3128                     int score= distoration + score_tab[i-run];
3129
3130                     if(score < best_score){
3131                         best_score= score;
3132                         run_tab[i+1]= run;
3133                         level_tab[i+1]= level-64;
3134                     }
3135                 }
3136
3137                 if(s->out_format == FMT_H263){
3138                   for(j=survivor_count-1; j>=0; j--){
3139                         int run= i - survivor[j];
3140                         int score= distoration + score_tab[i-run];
3141                         if(score < last_score){
3142                             last_score= score;
3143                             last_run= run;
3144                             last_level= level-64;
3145                             last_i= i+1;
3146                         }
3147                     }
3148                 }
3149             }
3150         }
3151
3152         score_tab[i+1]= best_score;
3153
3154         //Note: there is a vlc code in mpeg4 which is 1 bit shorter then another one with a shorter run and the same level
3155         if(last_non_zero <= 27){
3156             for(; survivor_count; survivor_count--){
3157                 if(score_tab[ survivor[survivor_count-1] ] <= best_score)
3158                     break;
3159             }
3160         }else{
3161             for(; survivor_count; survivor_count--){
3162                 if(score_tab[ survivor[survivor_count-1] ] <= best_score + lambda)
3163                     break;
3164             }
3165         }
3166
3167         survivor[ survivor_count++ ]= i+1;
3168     }
3169
3170     if(s->out_format != FMT_H263){
3171         last_score= 256*256*256*120;
3172         for(i= survivor[0]; i<=last_non_zero + 1; i++){
3173             int score= score_tab[i];
3174             if(i) score += lambda*2; //FIXME exacter?
3175
3176             if(score < last_score){
3177                 last_score= score;
3178                 last_i= i;
3179                 last_level= level_tab[i];
3180                 last_run= run_tab[i];
3181             }
3182         }
3183     }
3184
3185     s->coded_score[n] = last_score;
3186
3187     dc= FFABS(block[0]);
3188     last_non_zero= last_i - 1;
3189     memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
3190
3191     if(last_non_zero < start_i)
3192         return last_non_zero;
3193
3194     if(last_non_zero == 0 && start_i == 0){
3195         int best_level= 0;
3196         int best_score= dc * dc;
3197
3198         for(i=0; i<coeff_count[0]; i++){
3199             int level= coeff[i][0];
3200             int alevel= FFABS(level);
3201             int unquant_coeff, score, distortion;
3202
3203             if(s->out_format == FMT_H263){
3204                     unquant_coeff= (alevel*qmul + qadd)>>3;
3205             }else{ //MPEG1
3206                     unquant_coeff = (((  alevel  << 1) + 1) * qscale * ((int) s->inter_matrix[0])) >> 4;
3207                     unquant_coeff =   (unquant_coeff - 1) | 1;
3208             }
3209             unquant_coeff = (unquant_coeff + 4) >> 3;
3210             unquant_coeff<<= 3 + 3;
3211
3212             distortion= (unquant_coeff - dc) * (unquant_coeff - dc);
3213             level+=64;
3214             if((level&(~127)) == 0) score= distortion + last_length[UNI_AC_ENC_INDEX(0, level)]*lambda;
3215             else                    score= distortion + esc_length*lambda;
3216
3217             if(score < best_score){
3218                 best_score= score;
3219                 best_level= level - 64;
3220             }
3221         }
3222         block[0]= best_level;
3223         s->coded_score[n] = best_score - dc*dc;
3224         if(best_level == 0) return -1;
3225         else                return last_non_zero;
3226     }
3227
3228     i= last_i;
3229     assert(last_level);
3230
3231     block[ perm_scantable[last_non_zero] ]= last_level;
3232     i -= last_run + 1;
3233
3234     for(; i>start_i; i -= run_tab[i] + 1){
3235         block[ perm_scantable[i-1] ]= level_tab[i];
3236     }
3237
3238     return last_non_zero;
3239 }
3240
3241 //#define REFINE_STATS 1
3242 static int16_t basis[64][64];
3243
3244 static void build_basis(uint8_t *perm){
3245     int i, j, x, y;
3246     emms_c();
3247     for(i=0; i<8; i++){
3248         for(j=0; j<8; j++){
3249             for(y=0; y<8; y++){
3250                 for(x=0; x<8; x++){
3251                     double s= 0.25*(1<<BASIS_SHIFT);
3252                     int index= 8*i + j;
3253                     int perm_index= perm[index];
3254                     if(i==0) s*= sqrt(0.5);
3255                     if(j==0) s*= sqrt(0.5);
3256                     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)));
3257                 }
3258             }
3259         }
3260     }
3261 }
3262
3263 static int dct_quantize_refine(MpegEncContext *s, //FIXME breaks denoise?
3264                         DCTELEM *block, int16_t *weight, DCTELEM *orig,
3265                         int n, int qscale){
3266     int16_t rem[64];
3267     DECLARE_ALIGNED_16(DCTELEM, d1[64]);
3268     const int *qmat;
3269     const uint8_t *scantable= s->intra_scantable.scantable;
3270     const uint8_t *perm_scantable= s->intra_scantable.permutated;
3271 //    unsigned int threshold1, threshold2;
3272 //    int bias=0;
3273     int run_tab[65];
3274     int prev_run=0;
3275     int prev_level=0;
3276     int qmul, qadd, start_i, last_non_zero, i, dc;
3277     uint8_t * length;
3278     uint8_t * last_length;
3279     int lambda;
3280     int rle_index, run, q = 1, sum; //q is only used when s->mb_intra is true
3281 #ifdef REFINE_STATS
3282 static int count=0;
3283 static int after_last=0;
3284 static int to_zero=0;
3285 static int from_zero=0;
3286 static int raise=0;
3287 static int lower=0;
3288 static int messed_sign=0;
3289 #endif
3290
3291     if(basis[0][0] == 0)
3292         build_basis(s->dsp.idct_permutation);
3293
3294     qmul= qscale*2;
3295     qadd= (qscale-1)|1;
3296     if (s->mb_intra) {
3297         if (!s->h263_aic) {
3298             if (n < 4)
3299                 q = s->y_dc_scale;
3300             else
3301                 q = s->c_dc_scale;
3302         } else{
3303             /* For AIC we skip quant/dequant of INTRADC */
3304             q = 1;
3305             qadd=0;
3306         }
3307         q <<= RECON_SHIFT-3;
3308         /* note: block[0] is assumed to be positive */
3309         dc= block[0]*q;
3310 //        block[0] = (block[0] + (q >> 1)) / q;
3311         start_i = 1;
3312         qmat = s->q_intra_matrix[qscale];
3313 //        if(s->mpeg_quant || s->out_format == FMT_MPEG1)
3314 //            bias= 1<<(QMAT_SHIFT-1);
3315         length     = s->intra_ac_vlc_length;
3316         last_length= s->intra_ac_vlc_last_length;
3317     } else {
3318         dc= 0;
3319         start_i = 0;
3320         qmat = s->q_inter_matrix[qscale];
3321         length     = s->inter_ac_vlc_length;
3322         last_length= s->inter_ac_vlc_last_length;
3323     }
3324     last_non_zero = s->block_last_index[n];
3325
3326 #ifdef REFINE_STATS
3327 {START_TIMER
3328 #endif
3329     dc += (1<<(RECON_SHIFT-1));
3330     for(i=0; i<64; i++){
3331         rem[i]= dc - (orig[i]<<RECON_SHIFT); //FIXME  use orig dirrectly instead of copying to rem[]
3332     }
3333 #ifdef REFINE_STATS
3334 STOP_TIMER("memset rem[]")}
3335 #endif
3336     sum=0;
3337     for(i=0; i<64; i++){
3338         int one= 36;
3339         int qns=4;
3340         int w;
3341
3342         w= FFABS(weight[i]) + qns*one;
3343         w= 15 + (48*qns*one + w/2)/w; // 16 .. 63
3344
3345         weight[i] = w;
3346 //        w=weight[i] = (63*qns + (w/2)) / w;
3347
3348         assert(w>0);
3349         assert(w<(1<<6));
3350         sum += w*w;
3351     }
3352     lambda= sum*(uint64_t)s->lambda2 >> (FF_LAMBDA_SHIFT - 6 + 6 + 6 + 6);
3353 #ifdef REFINE_STATS
3354 {START_TIMER
3355 #endif
3356     run=0;
3357     rle_index=0;
3358     for(i=start_i; i<=last_non_zero; i++){
3359         int j= perm_scantable[i];
3360         const int level= block[j];
3361         int coeff;
3362
3363         if(level){
3364             if(level<0) coeff= qmul*level - qadd;
3365             else        coeff= qmul*level + qadd;
3366             run_tab[rle_index++]=run;
3367             run=0;
3368
3369             s->dsp.add_8x8basis(rem, basis[j], coeff);
3370         }else{
3371             run++;
3372         }
3373     }
3374 #ifdef REFINE_STATS
3375 if(last_non_zero>0){
3376 STOP_TIMER("init rem[]")
3377 }
3378 }
3379
3380 {START_TIMER
3381 #endif
3382     for(;;){
3383         int best_score=s->dsp.try_8x8basis(rem, weight, basis[0], 0);
3384         int best_coeff=0;
3385         int best_change=0;
3386         int run2, best_unquant_change=0, analyze_gradient;
3387 #ifdef REFINE_STATS
3388 {START_TIMER
3389 #endif
3390         analyze_gradient = last_non_zero > 2 || s->avctx->quantizer_noise_shaping >= 3;
3391
3392         if(analyze_gradient){
3393 #ifdef REFINE_STATS
3394 {START_TIMER
3395 #endif
3396             for(i=0; i<64; i++){
3397                 int w= weight[i];
3398
3399                 d1[i] = (rem[i]*w*w + (1<<(RECON_SHIFT+12-1)))>>(RECON_SHIFT+12);
3400             }
3401 #ifdef REFINE_STATS
3402 STOP_TIMER("rem*w*w")}
3403 {START_TIMER
3404 #endif
3405             s->dsp.fdct(d1);
3406 #ifdef REFINE_STATS
3407 STOP_TIMER("dct")}
3408 #endif
3409         }
3410
3411         if(start_i){
3412             const int level= block[0];
3413             int change, old_coeff;
3414
3415             assert(s->mb_intra);
3416
3417             old_coeff= q*level;
3418
3419             for(change=-1; change<=1; change+=2){
3420                 int new_level= level + change;
3421                 int score, new_coeff;
3422
3423                 new_coeff= q*new_level;
3424                 if(new_coeff >= 2048 || new_coeff < 0)
3425                     continue;
3426
3427                 score= s->dsp.try_8x8basis(rem, weight, basis[0], new_coeff - old_coeff);
3428                 if(score<best_score){
3429                     best_score= score;
3430                     best_coeff= 0;
3431                     best_change= change;
3432                     best_unquant_change= new_coeff - old_coeff;
3433                 }
3434             }
3435         }
3436
3437         run=0;
3438         rle_index=0;
3439         run2= run_tab[rle_index++];
3440         prev_level=0;
3441         prev_run=0;
3442
3443         for(i=start_i; i<64; i++){
3444             int j= perm_scantable[i];
3445             const int level= block[j];
3446             int change, old_coeff;
3447
3448             if(s->avctx->quantizer_noise_shaping < 3 && i > last_non_zero + 1)
3449                 break;
3450
3451             if(level){
3452                 if(level<0) old_coeff= qmul*level - qadd;
3453                 else        old_coeff= qmul*level + qadd;
3454                 run2= run_tab[rle_index++]; //FIXME ! maybe after last
3455             }else{
3456                 old_coeff=0;
3457                 run2--;
3458                 assert(run2>=0 || i >= last_non_zero );
3459             }
3460
3461             for(change=-1; change<=1; change+=2){
3462                 int new_level= level + change;
3463                 int score, new_coeff, unquant_change;
3464
3465                 score=0;
3466                 if(s->avctx->quantizer_noise_shaping < 2 && FFABS(new_level) > FFABS(level))
3467                    continue;
3468
3469                 if(new_level){
3470                     if(new_level<0) new_coeff= qmul*new_level - qadd;
3471                     else            new_coeff= qmul*new_level + qadd;
3472                     if(new_coeff >= 2048 || new_coeff <= -2048)
3473                         continue;
3474                     //FIXME check for overflow
3475
3476                     if(level){
3477                         if(level < 63 && level > -63){
3478                             if(i < last_non_zero)
3479                                 score +=   length[UNI_AC_ENC_INDEX(run, new_level+64)]
3480                                          - length[UNI_AC_ENC_INDEX(run, level+64)];
3481                             else
3482                                 score +=   last_length[UNI_AC_ENC_INDEX(run, new_level+64)]
3483                                          - last_length[UNI_AC_ENC_INDEX(run, level+64)];
3484                         }
3485                     }else{
3486                         assert(FFABS(new_level)==1);
3487
3488                         if(analyze_gradient){
3489                             int g= d1[ scantable[i] ];
3490                             if(g && (g^new_level) >= 0)
3491                                 continue;
3492                         }
3493
3494                         if(i < last_non_zero){
3495                             int next_i= i + run2 + 1;
3496                             int next_level= block[ perm_scantable[next_i] ] + 64;
3497
3498                             if(next_level&(~127))
3499                                 next_level= 0;
3500
3501                             if(next_i < last_non_zero)
3502                                 score +=   length[UNI_AC_ENC_INDEX(run, 65)]
3503                                          + length[UNI_AC_ENC_INDEX(run2, next_level)]
3504                                          - length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)];
3505                             else
3506                                 score +=  length[UNI_AC_ENC_INDEX(run, 65)]
3507                                         + last_length[UNI_AC_ENC_INDEX(run2, next_level)]
3508                                         - last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)];
3509                         }else{
3510                             score += last_length[UNI_AC_ENC_INDEX(run, 65)];
3511                             if(prev_level){
3512                                 score +=  length[UNI_AC_ENC_INDEX(prev_run, prev_level)]
3513                                         - last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)];
3514                             }
3515                         }
3516                     }
3517                 }else{
3518                     new_coeff=0;
3519                     assert(FFABS(level)==1);
3520
3521                     if(i < last_non_zero){
3522                         int next_i= i + run2 + 1;
3523                         int next_level= block[ perm_scantable[next_i] ] + 64;
3524
3525                         if(next_level&(~127))
3526                             next_level= 0;
3527
3528                         if(next_i < last_non_zero)
3529                             score +=   length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)]
3530                                      - length[UNI_AC_ENC_INDEX(run2, next_level)]
3531                                      - length[UNI_AC_ENC_INDEX(run, 65)];
3532                         else
3533                             score +=   last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)]
3534                                      - last_length[UNI_AC_ENC_INDEX(run2, next_level)]
3535                                      - length[UNI_AC_ENC_INDEX(run, 65)];
3536                     }else{
3537                         score += -last_length[UNI_AC_ENC_INDEX(run, 65)];
3538                         if(prev_level){
3539                             score +=  last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)]
3540                                     - length[UNI_AC_ENC_INDEX(prev_run, prev_level)];
3541                         }
3542                     }
3543                 }
3544
3545                 score *= lambda;
3546
3547                 unquant_change= new_coeff - old_coeff;
3548                 assert((score < 100*lambda && score > -100*lambda) || lambda==0);
3549
3550                 score+= s->dsp.try_8x8basis(rem, weight, basis[j], unquant_change);
3551                 if(score<best_score){
3552                     best_score= score;
3553                     best_coeff= i;
3554                     best_change= change;
3555                     best_unquant_change= unquant_change;
3556                 }
3557             }
3558             if(level){
3559                 prev_level= level + 64;
3560                 if(prev_level&(~127))
3561                     prev_level= 0;
3562                 prev_run= run;
3563                 run=0;
3564             }else{
3565                 run++;
3566             }
3567         }
3568 #ifdef REFINE_STATS
3569 STOP_TIMER("iterative step")}
3570 #endif
3571
3572         if(best_change){
3573             int j= perm_scantable[ best_coeff ];
3574
3575             block[j] += best_change;
3576
3577             if(best_coeff > last_non_zero){
3578                 last_non_zero= best_coeff;
3579                 assert(block[j]);
3580 #ifdef REFINE_STATS
3581 after_last++;
3582 #endif
3583             }else{
3584 #ifdef REFINE_STATS
3585 if(block[j]){
3586     if(block[j] - best_change){
3587         if(FFABS(block[j]) > FFABS(block[j] - best_change)){
3588             raise++;
3589         }else{
3590             lower++;
3591         }
3592     }else{
3593         from_zero++;
3594     }
3595 }else{
3596     to_zero++;
3597 }
3598 #endif
3599                 for(; last_non_zero>=start_i; last_non_zero--){
3600                     if(block[perm_scantable[last_non_zero]])
3601                         break;
3602                 }
3603             }
3604 #ifdef REFINE_STATS
3605 count++;
3606 if(256*256*256*64 % count == 0){
3607     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);
3608 }
3609 #endif
3610             run=0;
3611             rle_index=0;
3612             for(i=start_i; i<=last_non_zero; i++){
3613                 int j= perm_scantable[i];
3614                 const int level= block[j];
3615
3616                  if(level){
3617                      run_tab[rle_index++]=run;
3618                      run=0;
3619                  }else{
3620                      run++;
3621                  }
3622             }
3623
3624             s->dsp.add_8x8basis(rem, basis[j], best_unquant_change);
3625         }else{
3626             break;
3627         }
3628     }
3629 #ifdef REFINE_STATS
3630 if(last_non_zero>0){
3631 STOP_TIMER("iterative search")
3632 }
3633 }
3634 #endif
3635
3636     return last_non_zero;
3637 }
3638
3639 int dct_quantize_c(MpegEncContext *s,
3640                         DCTELEM *block, int n,
3641                         int qscale, int *overflow)
3642 {
3643     int i, j, level, last_non_zero, q, start_i;
3644     const int *qmat;
3645     const uint8_t *scantable= s->intra_scantable.scantable;
3646     int bias;
3647     int max=0;
3648     unsigned int threshold1, threshold2;
3649
3650     s->dsp.fdct (block);
3651
3652     if(s->dct_error_sum)
3653         s->denoise_dct(s, block);
3654
3655     if (s->mb_intra) {
3656         if (!s->h263_aic) {
3657             if (n < 4)
3658                 q = s->y_dc_scale;
3659             else
3660                 q = s->c_dc_scale;
3661             q = q << 3;
3662         } else
3663             /* For AIC we skip quant/dequant of INTRADC */
3664             q = 1 << 3;
3665
3666         /* note: block[0] is assumed to be positive */
3667         block[0] = (block[0] + (q >> 1)) / q;
3668         start_i = 1;
3669         last_non_zero = 0;
3670         qmat = s->q_intra_matrix[qscale];
3671         bias= s->intra_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT);
3672     } else {
3673         start_i = 0;
3674         last_non_zero = -1;
3675         qmat = s->q_inter_matrix[qscale];
3676         bias= s->inter_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT);
3677     }
3678     threshold1= (1<<QMAT_SHIFT) - bias - 1;
3679     threshold2= (threshold1<<1);
3680     for(i=63;i>=start_i;i--) {
3681         j = scantable[i];
3682         level = block[j] * qmat[j];
3683
3684         if(((unsigned)(level+threshold1))>threshold2){
3685             last_non_zero = i;
3686             break;
3687         }else{
3688             block[j]=0;
3689         }
3690     }
3691     for(i=start_i; i<=last_non_zero; i++) {
3692         j = scantable[i];
3693         level = block[j] * qmat[j];
3694
3695 //        if(   bias+level >= (1<<QMAT_SHIFT)
3696 //           || bias-level >= (1<<QMAT_SHIFT)){
3697         if(((unsigned)(level+threshold1))>threshold2){
3698             if(level>0){
3699                 level= (bias + level)>>QMAT_SHIFT;
3700                 block[j]= level;
3701             }else{
3702                 level= (bias - level)>>QMAT_SHIFT;
3703                 block[j]= -level;
3704             }
3705             max |=level;
3706         }else{
3707             block[j]=0;
3708         }
3709     }
3710     *overflow= s->max_qcoeff < max; //overflow might have happened
3711
3712     /* we need this permutation so that we correct the IDCT, we only permute the !=0 elements */
3713     if (s->dsp.idct_permutation_type != FF_NO_IDCT_PERM)
3714         ff_block_permute(block, s->dsp.idct_permutation, scantable, last_non_zero);
3715
3716     return last_non_zero;
3717 }
3718
3719 AVCodec h263_encoder = {
3720     "h263",
3721     CODEC_TYPE_VIDEO,
3722     CODEC_ID_H263,
3723     sizeof(MpegEncContext),
3724     MPV_encode_init,
3725     MPV_encode_picture,
3726     MPV_encode_end,
3727     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
3728 };
3729
3730 AVCodec h263p_encoder = {
3731     "h263p",
3732     CODEC_TYPE_VIDEO,
3733     CODEC_ID_H263P,
3734     sizeof(MpegEncContext),
3735     MPV_encode_init,
3736     MPV_encode_picture,
3737     MPV_encode_end,
3738     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
3739 };
3740
3741 AVCodec flv_encoder = {
3742     "flv",
3743     CODEC_TYPE_VIDEO,
3744     CODEC_ID_FLV1,
3745     sizeof(MpegEncContext),
3746     MPV_encode_init,
3747     MPV_encode_picture,
3748     MPV_encode_end,
3749     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
3750 };
3751
3752 AVCodec rv10_encoder = {
3753     "rv10",
3754     CODEC_TYPE_VIDEO,
3755     CODEC_ID_RV10,
3756     sizeof(MpegEncContext),
3757     MPV_encode_init,
3758     MPV_encode_picture,
3759     MPV_encode_end,
3760     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
3761 };
3762
3763 AVCodec rv20_encoder = {
3764     "rv20",
3765     CODEC_TYPE_VIDEO,
3766     CODEC_ID_RV20,
3767     sizeof(MpegEncContext),
3768     MPV_encode_init,
3769     MPV_encode_picture,
3770     MPV_encode_end,
3771     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
3772 };
3773
3774 AVCodec mpeg4_encoder = {
3775     "mpeg4",
3776     CODEC_TYPE_VIDEO,
3777     CODEC_ID_MPEG4,
3778     sizeof(MpegEncContext),
3779     MPV_encode_init,
3780     MPV_encode_picture,
3781     MPV_encode_end,
3782     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
3783     .capabilities= CODEC_CAP_DELAY,
3784 };
3785
3786 AVCodec msmpeg4v1_encoder = {
3787     "msmpeg4v1",
3788     CODEC_TYPE_VIDEO,
3789     CODEC_ID_MSMPEG4V1,
3790     sizeof(MpegEncContext),
3791     MPV_encode_init,
3792     MPV_encode_picture,
3793     MPV_encode_end,
3794     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
3795 };
3796
3797 AVCodec msmpeg4v2_encoder = {
3798     "msmpeg4v2",
3799     CODEC_TYPE_VIDEO,
3800     CODEC_ID_MSMPEG4V2,
3801     sizeof(MpegEncContext),
3802     MPV_encode_init,
3803     MPV_encode_picture,
3804     MPV_encode_end,
3805     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
3806 };
3807
3808 AVCodec msmpeg4v3_encoder = {
3809     "msmpeg4",
3810     CODEC_TYPE_VIDEO,
3811     CODEC_ID_MSMPEG4V3,
3812     sizeof(MpegEncContext),
3813     MPV_encode_init,
3814     MPV_encode_picture,
3815     MPV_encode_end,
3816     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
3817 };
3818
3819 AVCodec wmv1_encoder = {
3820     "wmv1",
3821     CODEC_TYPE_VIDEO,
3822     CODEC_ID_WMV1,
3823     sizeof(MpegEncContext),
3824     MPV_encode_init,
3825     MPV_encode_picture,
3826     MPV_encode_end,
3827     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
3828 };