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 const uint16_t inv_aanscales[64] = {
  65   4096,  2953,  3135,  3483,  4096,  5213,  7568, 14846,
  66   2953,  2129,  2260,  2511,  2953,  3759,  5457, 10703,
  67   3135,  2260,  2399,  2666,  3135,  3990,  5793, 11363,
  68   3483,  2511,  2666,  2962,  3483,  4433,  6436, 12625,
  69   4096,  2953,  3135,  3483,  4096,  5213,  7568, 14846,
  70   5213,  3759,  3990,  4433,  5213,  6635,  9633, 18895,
  71   7568,  5457,  5793,  6436,  7568,  9633, 13985, 27432,
  72  14846, 10703, 11363, 12625, 14846, 18895, 27432, 53809,
  73 };
  74
  75 static uint8_t default_mv_penalty[MAX_FCODE+1][MAX_MV*2+1];
  76 static uint8_t default_fcode_tab[MAX_MV*2+1];
  77
  78 void ff_convert_matrix(DSPContext *dsp, int (*qmat)[64], uint16_t (*qmat16)[2][64],
  79                            const uint16_t *quant_matrix, int bias, int qmin, int qmax, int intra)
  80 {
  81     int qscale;
  82     int shift=0;
  83
  84     for(qscale=qmin; qscale<=qmax; qscale++){
  85         int i;
  86         if (dsp->fdct == ff_jpeg_fdct_islow
  87 #ifdef FAAN_POSTSCALE
  88             || dsp->fdct == ff_faandct
  89 #endif
  90             ) {
  91             for(i=0;i<64;i++) {
  92                 const int j= dsp->idct_permutation[i];
  93                 /* 16 <= qscale * quant_matrix[i] <= 7905 */
  94                 /* 19952         <= aanscales[i] * qscale * quant_matrix[i]           <= 249205026 */
  95                 /* (1<<36)/19952 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */
  96                 /* 3444240       >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= 275 */
  97
  98                 qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) /
  99                                 (qscale * quant_matrix[j]));
 100             }
 101         } else if (dsp->fdct == fdct_ifast
 102 #ifndef FAAN_POSTSCALE
 103                    || dsp->fdct == ff_faandct
 104 #endif
 105                    ) {
 106             for(i=0;i<64;i++) {
 107                 const int j= dsp->idct_permutation[i];
 108                 /* 16 <= qscale * quant_matrix[i] <= 7905 */
 109                 /* 19952         <= aanscales[i] * qscale * quant_matrix[i]           <= 249205026 */
 110                 /* (1<<36)/19952 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */
 111                 /* 3444240       >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= 275 */
 112
 113                 qmat[qscale][i] = (int)((UINT64_C(1) << (QMAT_SHIFT + 14)) /
 114                                 (aanscales[i] * qscale * quant_matrix[j]));
 115             }
 116         } else {
 117             for(i=0;i<64;i++) {
 118                 const int j= dsp->idct_permutation[i];
 119                 /* We can safely suppose that 16 <= quant_matrix[i] <= 255
 120                    So 16           <= qscale * quant_matrix[i]             <= 7905
 121                    so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905
 122                    so 32768        >= (1<<19) / (qscale * quant_matrix[i]) >= 67
 123                 */
 124                 qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) / (qscale * quant_matrix[j]));
 125 //                qmat  [qscale][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]);
 126                 qmat16[qscale][0][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[j]);
 127
 128                 if(qmat16[qscale][0][i]==0 || qmat16[qscale][0][i]==128*256) qmat16[qscale][0][i]=128*256-1;
 129                 qmat16[qscale][1][i]= ROUNDED_DIV(bias<<(16-QUANT_BIAS_SHIFT), qmat16[qscale][0][i]);
 130             }
 131         }
 132
 133         for(i=intra; i<64; i++){
 134             int64_t max= 8191;
 135             if (dsp->fdct == fdct_ifast
 136 #ifndef FAAN_POSTSCALE
 137                    || dsp->fdct == ff_faandct
 138 #endif
 139                    ) {
 140                 max= (8191LL*aanscales[i]) >> 14;
 141             }
 142             while(((max * qmat[qscale][i]) >> shift) > INT_MAX){
 143                 shift++;
 144             }
 145         }
 146     }
 147     if(shift){
 148         av_log(NULL, AV_LOG_INFO, "Warning, QMAT_SHIFT is larger than %d, overflows possible\n", QMAT_SHIFT - shift);
 149     }
 150 }
 151
 152 static inline void update_qscale(MpegEncContext *s){
 153     s->qscale= (s->lambda*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7);
 154     s->qscale= av_clip(s->qscale, s->avctx->qmin, s->avctx->qmax);
 155
 156     s->lambda2= (s->lambda*s->lambda + FF_LAMBDA_SCALE/2) >> FF_LAMBDA_SHIFT;
 157 }
 158
 159 void ff_write_quant_matrix(PutBitContext *pb, uint16_t *matrix){
 160     int i;
 161
 162     if(matrix){
 163         put_bits(pb, 1, 1);
 164         for(i=0;i<64;i++) {
 165             put_bits(pb, 8, matrix[ ff_zigzag_direct[i] ]);
 166         }
 167     }else
 168         put_bits(pb, 1, 0);
 169 }
 170
 171 static void copy_picture_attributes(MpegEncContext *s, AVFrame *dst, AVFrame *src){
 172     int i;
 173
 174     dst->pict_type              = src->pict_type;
 175     dst->quality                = src->quality;
 176     dst->coded_picture_number   = src->coded_picture_number;
 177     dst->display_picture_number = src->display_picture_number;
 178 //    dst->reference              = src->reference;
 179     dst->pts                    = src->pts;
 180     dst->interlaced_frame       = src->interlaced_frame;
 181     dst->top_field_first        = src->top_field_first;
 182
 183     if(s->avctx->me_threshold){
 184         if(!src->motion_val[0])
 185             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_val not set!\n");
 186         if(!src->mb_type)
 187             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.mb_type not set!\n");
 188         if(!src->ref_index[0])
 189             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.ref_index not set!\n");
 190         if(src->motion_subsample_log2 != dst->motion_subsample_log2)
 191             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_subsample_log2 doesn't match! (%d!=%d)\n",
 192             src->motion_subsample_log2, dst->motion_subsample_log2);
 193
 194         memcpy(dst->mb_type, src->mb_type, s->mb_stride * s->mb_height * sizeof(dst->mb_type[0]));
 195
 196         for(i=0; i<2; i++){
 197             int stride= ((16*s->mb_width )>>src->motion_subsample_log2) + 1;
 198             int height= ((16*s->mb_height)>>src->motion_subsample_log2);
 199
 200             if(src->motion_val[i] && src->motion_val[i] != dst->motion_val[i]){
 201                 memcpy(dst->motion_val[i], src->motion_val[i], 2*stride*height*sizeof(int16_t));
 202             }
 203             if(src->ref_index[i] && src->ref_index[i] != dst->ref_index[i]){
 204                 memcpy(dst->ref_index[i], src->ref_index[i], s->b8_stride*2*s->mb_height*sizeof(int8_t));
 205             }
 206         }
 207     }
 208 }
 209
 210 static void update_duplicate_context_after_me(MpegEncContext *dst, MpegEncContext *src){
 211 #define COPY(a) dst->a= src->a
 212     COPY(pict_type);
 213     COPY(current_picture);
 214     COPY(f_code);
 215     COPY(b_code);
 216     COPY(qscale);
 217     COPY(lambda);
 218     COPY(lambda2);
 219     COPY(picture_in_gop_number);
 220     COPY(gop_picture_number);
 221     COPY(frame_pred_frame_dct); //FIXME don't set in encode_header
 222     COPY(progressive_frame); //FIXME don't set in encode_header
 223     COPY(partitioned_frame); //FIXME don't set in encode_header
 224 #undef COPY
 225 }
 226
 227 /**
 228  * sets the given MpegEncContext to defaults for encoding.
 229  * the changed fields will not depend upon the prior state of the MpegEncContext.
 230  */
 231 static void MPV_encode_defaults(MpegEncContext *s){
 232     int i;
 233     MPV_common_defaults(s);
 234
 235     for(i=-16; i<16; i++){
 236         default_fcode_tab[i + MAX_MV]= 1;
 237     }
 238     s->me.mv_penalty= default_mv_penalty;
 239     s->fcode_tab= default_fcode_tab;
 240 }
 241
 242 /* init video encoder */
 243 av_cold int MPV_encode_init(AVCodecContext *avctx)
 244 {
 245     MpegEncContext *s = avctx->priv_data;
 246     int i;
 247     int chroma_h_shift, chroma_v_shift;
 248
 249     MPV_encode_defaults(s);
 250
 251     switch (avctx->codec_id) {
 252     case CODEC_ID_MPEG2VIDEO:
 253         if(avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P){
 254             av_log(avctx, AV_LOG_ERROR, "only YUV420 and YUV422 are supported\n");
 255             return -1;
 256         }
 257         break;
 258     case CODEC_ID_LJPEG:
 259     case CODEC_ID_MJPEG:
 260         if(avctx->pix_fmt != PIX_FMT_YUVJ420P && avctx->pix_fmt != PIX_FMT_YUVJ422P &&
 261            ((avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P) || avctx->strict_std_compliance>FF_COMPLIANCE_INOFFICIAL)){
 262             av_log(avctx, AV_LOG_ERROR, "colorspace not supported in jpeg\n");
 263             return -1;
 264         }
 265         break;
 266     default:
 267         if(avctx->pix_fmt != PIX_FMT_YUV420P){
 268             av_log(avctx, AV_LOG_ERROR, "only YUV420 is supported\n");
 269             return -1;
 270         }
 271     }
 272
 273     switch (avctx->pix_fmt) {
 274     case PIX_FMT_YUVJ422P:
 275     case PIX_FMT_YUV422P:
 276         s->chroma_format = CHROMA_422;
 277         break;
 278     case PIX_FMT_YUVJ420P:
 279     case PIX_FMT_YUV420P:
 280     default:
 281         s->chroma_format = CHROMA_420;
 282         break;
 283     }
 284
 285     s->bit_rate = avctx->bit_rate;
 286     s->width = avctx->width;
 287     s->height = avctx->height;
 288     if(avctx->gop_size > 600 && avctx->strict_std_compliance>FF_COMPLIANCE_EXPERIMENTAL){
 289         av_log(avctx, AV_LOG_ERROR, "Warning keyframe interval too large! reducing it ...\n");
 290         avctx->gop_size=600;
 291     }
 292     s->gop_size = avctx->gop_size;
 293     s->avctx = avctx;
 294     s->flags= avctx->flags;
 295     s->flags2= avctx->flags2;
 296     s->max_b_frames= avctx->max_b_frames;
 297     s->codec_id= avctx->codec->id;
 298     s->luma_elim_threshold  = avctx->luma_elim_threshold;
 299     s->chroma_elim_threshold= avctx->chroma_elim_threshold;
 300     s->strict_std_compliance= avctx->strict_std_compliance;
 301     s->data_partitioning= avctx->flags & CODEC_FLAG_PART;
 302     s->quarter_sample= (avctx->flags & CODEC_FLAG_QPEL)!=0;
 303     s->mpeg_quant= avctx->mpeg_quant;
 304     s->rtp_mode= !!avctx->rtp_payload_size;
 305     s->intra_dc_precision= avctx->intra_dc_precision;
 306     s->user_specified_pts = AV_NOPTS_VALUE;
 307
 308     if (s->gop_size <= 1) {
 309         s->intra_only = 1;
 310         s->gop_size = 12;
 311     } else {
 312         s->intra_only = 0;
 313     }
 314
 315     s->me_method = avctx->me_method;
 316
 317     /* Fixed QSCALE */
 318     s->fixed_qscale = !!(avctx->flags & CODEC_FLAG_QSCALE);
 319
 320     s->adaptive_quant= (   s->avctx->lumi_masking
 321                         || s->avctx->dark_masking
 322                         || s->avctx->temporal_cplx_masking
 323                         || s->avctx->spatial_cplx_masking
 324                         || s->avctx->p_masking
 325                         || s->avctx->border_masking
 326                         || (s->flags&CODEC_FLAG_QP_RD))
 327                        && !s->fixed_qscale;
 328
 329     s->obmc= !!(s->flags & CODEC_FLAG_OBMC);
 330     s->loop_filter= !!(s->flags & CODEC_FLAG_LOOP_FILTER);
 331     s->alternate_scan= !!(s->flags & CODEC_FLAG_ALT_SCAN);
 332     s->intra_vlc_format= !!(s->flags2 & CODEC_FLAG2_INTRA_VLC);
 333     s->q_scale_type= !!(s->flags2 & CODEC_FLAG2_NON_LINEAR_QUANT);
 334
 335 #if LIBAVCODEC_VERSION_INT < ((52<<16)+(0<<8)+0)
 336     if (s->flags & CODEC_FLAG_TRELLIS_QUANT)
 337         avctx->trellis = 1;
 338 #endif
 339
 340     if(avctx->rc_max_rate && !avctx->rc_buffer_size){
 341         av_log(avctx, AV_LOG_ERROR, "a vbv buffer size is needed, for encoding with a maximum bitrate\n");
 342         return -1;
 343     }
 344
 345     if(avctx->rc_min_rate && avctx->rc_max_rate != avctx->rc_min_rate){
 346         av_log(avctx, AV_LOG_INFO, "Warning min_rate > 0 but min_rate != max_rate isn't recommended!\n");
 347     }
 348
 349     if(avctx->rc_min_rate && avctx->rc_min_rate > avctx->bit_rate){
 350         av_log(avctx, AV_LOG_ERROR, "bitrate below min bitrate\n");
 351         return -1;
 352     }
 353
 354     if(avctx->rc_max_rate && avctx->rc_max_rate < avctx->bit_rate){
 355         av_log(avctx, AV_LOG_INFO, "bitrate above max bitrate\n");
 356         return -1;
 357     }
 358
 359     if(avctx->rc_max_rate && avctx->rc_max_rate == avctx->bit_rate && avctx->rc_max_rate != avctx->rc_min_rate){
 360         av_log(avctx, AV_LOG_INFO, "impossible bitrate constraints, this will fail\n");
 361     }
 362
 363     if(avctx->rc_buffer_size && avctx->bit_rate*av_q2d(avctx->time_base) > avctx->rc_buffer_size){
 364         av_log(avctx, AV_LOG_ERROR, "VBV buffer too small for bitrate\n");
 365         return -1;
 366     }
 367
 368     if(avctx->bit_rate*av_q2d(avctx->time_base) > avctx->bit_rate_tolerance){
 369         av_log(avctx, AV_LOG_ERROR, "bitrate tolerance too small for bitrate\n");
 370         return -1;
 371     }
 372
 373     if(   s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate
 374        && (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO)
 375        && 90000LL * (avctx->rc_buffer_size-1) > s->avctx->rc_max_rate*0xFFFFLL){
 376
 377         av_log(avctx, AV_LOG_INFO, "Warning vbv_delay will be set to 0xFFFF (=VBR) as the specified vbv buffer is too large for the given bitrate!\n");
 378     }
 379
 380     if((s->flags & CODEC_FLAG_4MV) && s->codec_id != CODEC_ID_MPEG4
 381        && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P && s->codec_id != CODEC_ID_FLV1){
 382         av_log(avctx, AV_LOG_ERROR, "4MV not supported by codec\n");
 383         return -1;
 384     }
 385
 386     if(s->obmc && s->avctx->mb_decision != FF_MB_DECISION_SIMPLE){
 387         av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with simple mb decision\n");
 388         return -1;
 389     }
 390
 391     if(s->obmc && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P){
 392         av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with H263(+)\n");
 393         return -1;
 394     }
 395
 396     if(s->quarter_sample && s->codec_id != CODEC_ID_MPEG4){
 397         av_log(avctx, AV_LOG_ERROR, "qpel not supported by codec\n");
 398         return -1;
 399     }
 400
 401     if(s->data_partitioning && s->codec_id != CODEC_ID_MPEG4){
 402         av_log(avctx, AV_LOG_ERROR, "data partitioning not supported by codec\n");
 403         return -1;
 404     }
 405
 406     if(s->max_b_frames && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO){
 407         av_log(avctx, AV_LOG_ERROR, "b frames not supported by codec\n");
 408         return -1;
 409     }
 410
 411     if((s->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN))
 412        && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG2VIDEO){
 413         av_log(avctx, AV_LOG_ERROR, "interlacing not supported by codec\n");
 414         return -1;
 415     }
 416
 417     if(s->mpeg_quant && s->codec_id != CODEC_ID_MPEG4){ //FIXME mpeg2 uses that too
 418         av_log(avctx, AV_LOG_ERROR, "mpeg2 style quantization not supported by codec\n");
 419         return -1;
 420     }
 421
 422     if((s->flags & CODEC_FLAG_CBP_RD) && !avctx->trellis){
 423         av_log(avctx, AV_LOG_ERROR, "CBP RD needs trellis quant\n");
 424         return -1;
 425     }
 426
 427     if((s->flags & CODEC_FLAG_QP_RD) && s->avctx->mb_decision != FF_MB_DECISION_RD){
 428         av_log(avctx, AV_LOG_ERROR, "QP RD needs mbd=2\n");
 429         return -1;
 430     }
 431
 432     if(s->avctx->scenechange_threshold < 1000000000 && (s->flags & CODEC_FLAG_CLOSED_GOP)){
 433         av_log(avctx, AV_LOG_ERROR, "closed gop with scene change detection are not supported yet, set threshold to 1000000000\n");
 434         return -1;
 435     }
 436
 437     if((s->flags2 & CODEC_FLAG2_INTRA_VLC) && s->codec_id != CODEC_ID_MPEG2VIDEO){
 438         av_log(avctx, AV_LOG_ERROR, "intra vlc table not supported by codec\n");
 439         return -1;
 440     }
 441
 442     if(s->flags & CODEC_FLAG_LOW_DELAY){
 443         if (s->codec_id != CODEC_ID_MPEG2VIDEO && s->codec_id != CODEC_ID_MPEG1VIDEO){
 444             av_log(avctx, AV_LOG_ERROR, "low delay forcing is only available for mpeg1/2\n");
 445             return -1;
 446         }
 447         if (s->max_b_frames != 0){
 448             av_log(avctx, AV_LOG_ERROR, "b frames cannot be used with low delay\n");
 449             return -1;
 450         }
 451     }
 452
 453     if(s->q_scale_type == 1){
 454         if(s->codec_id != CODEC_ID_MPEG2VIDEO){
 455             av_log(avctx, AV_LOG_ERROR, "non linear quant is only available for mpeg2\n");
 456             return -1;
 457         }
 458         if(avctx->qmax > 12){
 459             av_log(avctx, AV_LOG_ERROR, "non linear quant only supports qmax <= 12 currently\n");
 460             return -1;
 461         }
 462     }
 463
 464     if(s->avctx->thread_count > 1 && s->codec_id != CODEC_ID_MPEG4
 465        && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO
 466        && (s->codec_id != CODEC_ID_H263P || !(s->flags & CODEC_FLAG_H263P_SLICE_STRUCT))){
 467         av_log(avctx, AV_LOG_ERROR, "multi threaded encoding not supported by codec\n");
 468         return -1;
 469     }
 470
 471     if(s->avctx->thread_count > 1)
 472         s->rtp_mode= 1;
 473
 474     if(!avctx->time_base.den || !avctx->time_base.num){
 475         av_log(avctx, AV_LOG_ERROR, "framerate not set\n");
 476         return -1;
 477     }
 478
 479     i= (INT_MAX/2+128)>>8;
 480     if(avctx->me_threshold >= i){
 481         av_log(avctx, AV_LOG_ERROR, "me_threshold too large, max is %d\n", i - 1);
 482         return -1;
 483     }
 484     if(avctx->mb_threshold >= i){
 485         av_log(avctx, AV_LOG_ERROR, "mb_threshold too large, max is %d\n", i - 1);
 486         return -1;
 487     }
 488
 489     if(avctx->b_frame_strategy && (avctx->flags&CODEC_FLAG_PASS2)){
 490         av_log(avctx, AV_LOG_INFO, "notice: b_frame_strategy only affects the first pass\n");
 491         avctx->b_frame_strategy = 0;
 492     }
 493
 494     i= ff_gcd(avctx->time_base.den, avctx->time_base.num);
 495     if(i > 1){
 496         av_log(avctx, AV_LOG_INFO, "removing common factors from framerate\n");
 497         avctx->time_base.den /= i;
 498         avctx->time_base.num /= i;
 499 //        return -1;
 500     }
 501
 502     if(s->codec_id==CODEC_ID_MJPEG){
 503         s->intra_quant_bias= 1<<(QUANT_BIAS_SHIFT-1); //(a + x/2)/x
 504         s->inter_quant_bias= 0;
 505     }else if(s->mpeg_quant || s->codec_id==CODEC_ID_MPEG1VIDEO || s->codec_id==CODEC_ID_MPEG2VIDEO){
 506         s->intra_quant_bias= 3<<(QUANT_BIAS_SHIFT-3); //(a + x*3/8)/x
 507         s->inter_quant_bias= 0;
 508     }else{
 509         s->intra_quant_bias=0;
 510         s->inter_quant_bias=-(1<<(QUANT_BIAS_SHIFT-2)); //(a - x/4)/x
 511     }
 512
 513     if(avctx->intra_quant_bias != FF_DEFAULT_QUANT_BIAS)
 514         s->intra_quant_bias= avctx->intra_quant_bias;
 515     if(avctx->inter_quant_bias != FF_DEFAULT_QUANT_BIAS)
 516         s->inter_quant_bias= avctx->inter_quant_bias;
 517
 518     avcodec_get_chroma_sub_sample(avctx->pix_fmt, &chroma_h_shift, &chroma_v_shift);
 519
 520     if(avctx->codec_id == CODEC_ID_MPEG4 && s->avctx->time_base.den > (1<<16)-1){
 521         av_log(avctx, AV_LOG_ERROR, "timebase not supported by mpeg 4 standard\n");
 522         return -1;
 523     }
 524     s->time_increment_bits = av_log2(s->avctx->time_base.den - 1) + 1;
 525
 526     switch(avctx->codec->id) {
 527     case CODEC_ID_MPEG1VIDEO:
 528         s->out_format = FMT_MPEG1;
 529         s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY);
 530         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
 531         break;
 532     case CODEC_ID_MPEG2VIDEO:
 533         s->out_format = FMT_MPEG1;
 534         s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY);
 535         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
 536         s->rtp_mode= 1;
 537         break;
 538     case CODEC_ID_LJPEG:
 539     case CODEC_ID_MJPEG:
 540         s->out_format = FMT_MJPEG;
 541         s->intra_only = 1; /* force intra only for jpeg */
 542         s->mjpeg_vsample[0] = 2;
 543         s->mjpeg_vsample[1] = 2>>chroma_v_shift;
 544         s->mjpeg_vsample[2] = 2>>chroma_v_shift;
 545         s->mjpeg_hsample[0] = 2;
 546         s->mjpeg_hsample[1] = 2>>chroma_h_shift;
 547         s->mjpeg_hsample[2] = 2>>chroma_h_shift;
 548         if (!(ENABLE_MJPEG_ENCODER || ENABLE_LJPEG_ENCODER)
 549             || ff_mjpeg_encode_init(s) < 0)
 550             return -1;
 551         avctx->delay=0;
 552         s->low_delay=1;
 553         break;
 554     case CODEC_ID_H261:
 555         if (!ENABLE_H261_ENCODER)  return -1;
 556         if (ff_h261_get_picture_format(s->width, s->height) < 0) {
 557             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);
 558             return -1;
 559         }
 560         s->out_format = FMT_H261;
 561         avctx->delay=0;
 562         s->low_delay=1;
 563         break;
 564     case CODEC_ID_H263:
 565         if (!ENABLE_H263_ENCODER)  return -1;
 566         if (h263_get_picture_format(s->width, s->height) == 7) {
 567             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);
 568             return -1;
 569         }
 570         s->out_format = FMT_H263;
 571         s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
 572         avctx->delay=0;
 573         s->low_delay=1;
 574         break;
 575     case CODEC_ID_H263P:
 576         s->out_format = FMT_H263;
 577         s->h263_plus = 1;
 578         /* Fx */
 579         s->umvplus = (avctx->flags & CODEC_FLAG_H263P_UMV) ? 1:0;
 580         s->h263_aic= (avctx->flags & CODEC_FLAG_AC_PRED) ? 1:0;
 581         s->modified_quant= s->h263_aic;
 582         s->alt_inter_vlc= (avctx->flags & CODEC_FLAG_H263P_AIV) ? 1:0;
 583         s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
 584         s->loop_filter= (avctx->flags & CODEC_FLAG_LOOP_FILTER) ? 1:0;
 585         s->unrestricted_mv= s->obmc || s->loop_filter || s->umvplus;
 586         s->h263_slice_structured= (s->flags & CODEC_FLAG_H263P_SLICE_STRUCT) ? 1:0;
 587
 588         /* /Fx */
 589         /* These are just to be sure */
 590         avctx->delay=0;
 591         s->low_delay=1;
 592         break;
 593     case CODEC_ID_FLV1:
 594         s->out_format = FMT_H263;
 595         s->h263_flv = 2; /* format = 1; 11-bit codes */
 596         s->unrestricted_mv = 1;
 597         s->rtp_mode=0; /* don't allow GOB */
 598         avctx->delay=0;
 599         s->low_delay=1;
 600         break;
 601     case CODEC_ID_RV10:
 602         s->out_format = FMT_H263;
 603         avctx->delay=0;
 604         s->low_delay=1;
 605         break;
 606     case CODEC_ID_RV20:
 607         s->out_format = FMT_H263;
 608         avctx->delay=0;
 609         s->low_delay=1;
 610         s->modified_quant=1;
 611         s->h263_aic=1;
 612         s->h263_plus=1;
 613         s->loop_filter=1;
 614         s->unrestricted_mv= s->obmc || s->loop_filter || s->umvplus;
 615         break;
 616     case CODEC_ID_MPEG4:
 617         s->out_format = FMT_H263;
 618         s->h263_pred = 1;
 619         s->unrestricted_mv = 1;
 620         s->low_delay= s->max_b_frames ? 0 : 1;
 621         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
 622         break;
 623     case CODEC_ID_MSMPEG4V1:
 624         s->out_format = FMT_H263;
 625         s->h263_msmpeg4 = 1;
 626         s->h263_pred = 1;
 627         s->unrestricted_mv = 1;
 628         s->msmpeg4_version= 1;
 629         avctx->delay=0;
 630         s->low_delay=1;
 631         break;
 632     case CODEC_ID_MSMPEG4V2:
 633         s->out_format = FMT_H263;
 634         s->h263_msmpeg4 = 1;
 635         s->h263_pred = 1;
 636         s->unrestricted_mv = 1;
 637         s->msmpeg4_version= 2;
 638         avctx->delay=0;
 639         s->low_delay=1;
 640         break;
 641     case CODEC_ID_MSMPEG4V3:
 642         s->out_format = FMT_H263;
 643         s->h263_msmpeg4 = 1;
 644         s->h263_pred = 1;
 645         s->unrestricted_mv = 1;
 646         s->msmpeg4_version= 3;
 647         s->flipflop_rounding=1;
 648         avctx->delay=0;
 649         s->low_delay=1;
 650         break;
 651     case CODEC_ID_WMV1:
 652         s->out_format = FMT_H263;
 653         s->h263_msmpeg4 = 1;
 654         s->h263_pred = 1;
 655         s->unrestricted_mv = 1;
 656         s->msmpeg4_version= 4;
 657         s->flipflop_rounding=1;
 658         avctx->delay=0;
 659         s->low_delay=1;
 660         break;
 661     case CODEC_ID_WMV2:
 662         s->out_format = FMT_H263;
 663         s->h263_msmpeg4 = 1;
 664         s->h263_pred = 1;
 665         s->unrestricted_mv = 1;
 666         s->msmpeg4_version= 5;
 667         s->flipflop_rounding=1;
 668         avctx->delay=0;
 669         s->low_delay=1;
 670         break;
 671     default:
 672         return -1;
 673     }
 674
 675     avctx->has_b_frames= !s->low_delay;
 676
 677     s->encoding = 1;
 678
 679     /* init */
 680     if (MPV_common_init(s) < 0)
 681         return -1;
 682
 683     if(!s->dct_quantize)
 684         s->dct_quantize = dct_quantize_c;
 685     if(!s->denoise_dct)
 686         s->denoise_dct = denoise_dct_c;
 687     s->fast_dct_quantize = s->dct_quantize;
 688     if(avctx->trellis)
 689         s->dct_quantize = dct_quantize_trellis_c;
 690
 691     if((ENABLE_H263P_ENCODER || ENABLE_RV20_ENCODER) && s->modified_quant)
 692         s->chroma_qscale_table= ff_h263_chroma_qscale_table;
 693     s->progressive_frame=
 694     s->progressive_sequence= !(avctx->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN));
 695     s->quant_precision=5;
 696
 697     ff_set_cmp(&s->dsp, s->dsp.ildct_cmp, s->avctx->ildct_cmp);
 698     ff_set_cmp(&s->dsp, s->dsp.frame_skip_cmp, s->avctx->frame_skip_cmp);
 699
 700     if (ENABLE_H261_ENCODER && s->out_format == FMT_H261)
 701         ff_h261_encode_init(s);
 702     if (ENABLE_ANY_H263_ENCODER && s->out_format == FMT_H263)
 703         h263_encode_init(s);
 704     if (ENABLE_MSMPEG4_ENCODER && s->msmpeg4_version)
 705         ff_msmpeg4_encode_init(s);
 706     if ((ENABLE_MPEG1VIDEO_ENCODER || ENABLE_MPEG2VIDEO_ENCODER)
 707         && s->out_format == FMT_MPEG1)
 708         ff_mpeg1_encode_init(s);
 709
 710     /* init q matrix */
 711     for(i=0;i<64;i++) {
 712         int j= s->dsp.idct_permutation[i];
 713         if(ENABLE_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4 && s->mpeg_quant){
 714             s->intra_matrix[j] = ff_mpeg4_default_intra_matrix[i];
 715             s->inter_matrix[j] = ff_mpeg4_default_non_intra_matrix[i];
 716         }else if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
 717             s->intra_matrix[j] =
 718             s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
 719         }else
 720         { /* mpeg1/2 */
 721             s->intra_matrix[j] = ff_mpeg1_default_intra_matrix[i];
 722             s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
 723         }
 724         if(s->avctx->intra_matrix)
 725             s->intra_matrix[j] = s->avctx->intra_matrix[i];
 726         if(s->avctx->inter_matrix)
 727             s->inter_matrix[j] = s->avctx->inter_matrix[i];
 728     }
 729
 730     /* precompute matrix */
 731     /* for mjpeg, we do include qscale in the matrix */
 732     if (s->out_format != FMT_MJPEG) {
 733         ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
 734                        s->intra_matrix, s->intra_quant_bias, avctx->qmin, 31, 1);
 735         ff_convert_matrix(&s->dsp, s->q_inter_matrix, s->q_inter_matrix16,
 736                        s->inter_matrix, s->inter_quant_bias, avctx->qmin, 31, 0);
 737     }
 738
 739     if(ff_rate_control_init(s) < 0)
 740         return -1;
 741
 742     return 0;
 743 }
 744
 745 av_cold int MPV_encode_end(AVCodecContext *avctx)
 746 {
 747     MpegEncContext *s = avctx->priv_data;
 748
 749     ff_rate_control_uninit(s);
 750
 751     MPV_common_end(s);
 752     if ((ENABLE_MJPEG_ENCODER || ENABLE_LJPEG_ENCODER) && s->out_format == FMT_MJPEG)
 753         ff_mjpeg_encode_close(s);
 754
 755     av_freep(&avctx->extradata);
 756
 757     return 0;
 758 }
 759
 760 static int get_sae(uint8_t *src, int ref, int stride){
 761     int x,y;
 762     int acc=0;
 763
 764     for(y=0; y<16; y++){
 765         for(x=0; x<16; x++){
 766             acc+= FFABS(src[x+y*stride] - ref);
 767         }
 768     }
 769
 770     return acc;
 771 }
 772
 773 static int get_intra_count(MpegEncContext *s, uint8_t *src, uint8_t *ref, int stride){
 774     int x, y, w, h;
 775     int acc=0;
 776
 777     w= s->width &~15;
 778     h= s->height&~15;
 779
 780     for(y=0; y<h; y+=16){
 781         for(x=0; x<w; x+=16){
 782             int offset= x + y*stride;
 783             int sad = s->dsp.sad[0](NULL, src + offset, ref + offset, stride, 16);
 784             int mean= (s->dsp.pix_sum(src + offset, stride) + 128)>>8;
 785             int sae = get_sae(src + offset, mean, stride);
 786
 787             acc+= sae + 500 < sad;
 788         }
 789     }
 790     return acc;
 791 }
 792
 793
 794 static int load_input_picture(MpegEncContext *s, AVFrame *pic_arg){
 795     AVFrame *pic=NULL;
 796     int64_t pts;
 797     int i;
 798     const int encoding_delay= s->max_b_frames;
 799     int direct=1;
 800
 801     if(pic_arg){
 802         pts= pic_arg->pts;
 803         pic_arg->display_picture_number= s->input_picture_number++;
 804
 805         if(pts != AV_NOPTS_VALUE){
 806             if(s->user_specified_pts != AV_NOPTS_VALUE){
 807                 int64_t time= pts;
 808                 int64_t last= s->user_specified_pts;
 809
 810                 if(time <= last){
 811                     av_log(s->avctx, AV_LOG_ERROR, "Error, Invalid timestamp=%"PRId64", last=%"PRId64"\n", pts, s->user_specified_pts);
 812                     return -1;
 813                 }
 814             }
 815             s->user_specified_pts= pts;
 816         }else{
 817             if(s->user_specified_pts != AV_NOPTS_VALUE){
 818                 s->user_specified_pts=
 819                 pts= s->user_specified_pts + 1;
 820                 av_log(s->avctx, AV_LOG_INFO, "Warning: AVFrame.pts=? trying to guess (%"PRId64")\n", pts);
 821             }else{
 822                 pts= pic_arg->display_picture_number;
 823             }
 824         }
 825     }
 826
 827   if(pic_arg){
 828     if(encoding_delay && !(s->flags&CODEC_FLAG_INPUT_PRESERVED)) direct=0;
 829     if(pic_arg->linesize[0] != s->linesize) direct=0;
 830     if(pic_arg->linesize[1] != s->uvlinesize) direct=0;
 831     if(pic_arg->linesize[2] != s->uvlinesize) direct=0;
 832
 833 //    av_log(AV_LOG_DEBUG, "%d %d %d %d\n",pic_arg->linesize[0], pic_arg->linesize[1], s->linesize, s->uvlinesize);
 834
 835     if(direct){
 836         i= ff_find_unused_picture(s, 1);
 837
 838         pic= (AVFrame*)&s->picture[i];
 839         pic->reference= 3;
 840
 841         for(i=0; i<4; i++){
 842             pic->data[i]= pic_arg->data[i];
 843             pic->linesize[i]= pic_arg->linesize[i];
 844         }
 845         alloc_picture(s, (Picture*)pic, 1);
 846     }else{
 847         i= ff_find_unused_picture(s, 0);
 848
 849         pic= (AVFrame*)&s->picture[i];
 850         pic->reference= 3;
 851
 852         alloc_picture(s, (Picture*)pic, 0);
 853
 854         if(   pic->data[0] + INPLACE_OFFSET == pic_arg->data[0]
 855            && pic->data[1] + INPLACE_OFFSET == pic_arg->data[1]
 856            && pic->data[2] + INPLACE_OFFSET == pic_arg->data[2]){
 857        // empty
 858         }else{
 859             int h_chroma_shift, v_chroma_shift;
 860             avcodec_get_chroma_sub_sample(s->avctx->pix_fmt, &h_chroma_shift, &v_chroma_shift);
 861
 862             for(i=0; i<3; i++){
 863                 int src_stride= pic_arg->linesize[i];
 864                 int dst_stride= i ? s->uvlinesize : s->linesize;
 865                 int h_shift= i ? h_chroma_shift : 0;
 866                 int v_shift= i ? v_chroma_shift : 0;
 867                 int w= s->width >>h_shift;
 868                 int h= s->height>>v_shift;
 869                 uint8_t *src= pic_arg->data[i];
 870                 uint8_t *dst= pic->data[i];
 871
 872                 if(!s->avctx->rc_buffer_size)
 873                     dst +=INPLACE_OFFSET;
 874
 875                 if(src_stride==dst_stride)
 876                     memcpy(dst, src, src_stride*h);
 877                 else{
 878                     while(h--){
 879                         memcpy(dst, src, w);
 880                         dst += dst_stride;
 881                         src += src_stride;
 882                     }
 883                 }
 884             }
 885         }
 886     }
 887     copy_picture_attributes(s, pic, pic_arg);
 888     pic->pts= pts; //we set this here to avoid modifiying pic_arg
 889   }
 890
 891     /* shift buffer entries */
 892     for(i=1; i<MAX_PICTURE_COUNT /*s->encoding_delay+1*/; i++)
 893         s->input_picture[i-1]= s->input_picture[i];
 894
 895     s->input_picture[encoding_delay]= (Picture*)pic;
 896
 897     return 0;
 898 }
 899
 900 static int skip_check(MpegEncContext *s, Picture *p, Picture *ref){
 901     int x, y, plane;
 902     int score=0;
 903     int64_t score64=0;
 904
 905     for(plane=0; plane<3; plane++){
 906         const int stride= p->linesize[plane];
 907         const int bw= plane ? 1 : 2;
 908         for(y=0; y<s->mb_height*bw; y++){
 909             for(x=0; x<s->mb_width*bw; x++){
 910                 int off= p->type == FF_BUFFER_TYPE_SHARED ? 0: 16;
 911                 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);
 912
 913                 switch(s->avctx->frame_skip_exp){
 914                     case 0: score= FFMAX(score, v); break;
 915                     case 1: score+= FFABS(v);break;
 916                     case 2: score+= v*v;break;
 917                     case 3: score64+= FFABS(v*v*(int64_t)v);break;
 918                     case 4: score64+= v*v*(int64_t)(v*v);break;
 919                 }
 920             }
 921         }
 922     }
 923
 924     if(score) score64= score;
 925
 926     if(score64 < s->avctx->frame_skip_threshold)
 927         return 1;
 928     if(score64 < ((s->avctx->frame_skip_factor * (int64_t)s->lambda)>>8))
 929         return 1;
 930     return 0;
 931 }
 932
 933 static int estimate_best_b_count(MpegEncContext *s){
 934     AVCodec *codec= avcodec_find_encoder(s->avctx->codec_id);
 935     AVCodecContext *c= avcodec_alloc_context();
 936     AVFrame input[FF_MAX_B_FRAMES+2];
 937     const int scale= s->avctx->brd_scale;
 938     int i, j, out_size, p_lambda, b_lambda, lambda2;
 939     int outbuf_size= s->width * s->height; //FIXME
 940     uint8_t *outbuf= av_malloc(outbuf_size);
 941     int64_t best_rd= INT64_MAX;
 942     int best_b_count= -1;
 943
 944     assert(scale>=0 && scale <=3);
 945
 946 //    emms_c();
 947     p_lambda= s->last_lambda_for[FF_P_TYPE]; //s->next_picture_ptr->quality;
 948     b_lambda= s->last_lambda_for[FF_B_TYPE]; //p_lambda *FFABS(s->avctx->b_quant_factor) + s->avctx->b_quant_offset;
 949     if(!b_lambda) b_lambda= p_lambda; //FIXME we should do this somewhere else
 950     lambda2= (b_lambda*b_lambda + (1<<FF_LAMBDA_SHIFT)/2 ) >> FF_LAMBDA_SHIFT;
 951
 952     c->width = s->width >> scale;
 953     c->height= s->height>> scale;
 954     c->flags= CODEC_FLAG_QSCALE | CODEC_FLAG_PSNR | CODEC_FLAG_INPUT_PRESERVED /*| CODEC_FLAG_EMU_EDGE*/;
 955     c->flags|= s->avctx->flags & CODEC_FLAG_QPEL;
 956     c->mb_decision= s->avctx->mb_decision;
 957     c->me_cmp= s->avctx->me_cmp;
 958     c->mb_cmp= s->avctx->mb_cmp;
 959     c->me_sub_cmp= s->avctx->me_sub_cmp;
 960     c->pix_fmt = PIX_FMT_YUV420P;
 961     c->time_base= s->avctx->time_base;
 962     c->max_b_frames= s->max_b_frames;
 963
 964     if (avcodec_open(c, codec) < 0)
 965         return -1;
 966
 967     for(i=0; i<s->max_b_frames+2; i++){
 968         int ysize= c->width*c->height;
 969         int csize= (c->width/2)*(c->height/2);
 970         Picture pre_input, *pre_input_ptr= i ? s->input_picture[i-1] : s->next_picture_ptr;
 971
 972         avcodec_get_frame_defaults(&input[i]);
 973         input[i].data[0]= av_malloc(ysize + 2*csize);
 974         input[i].data[1]= input[i].data[0] + ysize;
 975         input[i].data[2]= input[i].data[1] + csize;
 976         input[i].linesize[0]= c->width;
 977         input[i].linesize[1]=
 978         input[i].linesize[2]= c->width/2;
 979
 980         if(pre_input_ptr && (!i || s->input_picture[i-1])) {
 981             pre_input= *pre_input_ptr;
 982
 983             if(pre_input.type != FF_BUFFER_TYPE_SHARED && i) {
 984                 pre_input.data[0]+=INPLACE_OFFSET;
 985                 pre_input.data[1]+=INPLACE_OFFSET;
 986                 pre_input.data[2]+=INPLACE_OFFSET;
 987             }
 988
 989             s->dsp.shrink[scale](input[i].data[0], input[i].linesize[0], pre_input.data[0], pre_input.linesize[0], c->width, c->height);
 990             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);
 991             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);
 992         }
 993     }
 994
 995     for(j=0; j<s->max_b_frames+1; j++){
 996         int64_t rd=0;
 997
 998         if(!s->input_picture[j])
 999             break;
1000
1001         c->error[0]= c->error[1]= c->error[2]= 0;
1002
1003         input[0].pict_type= FF_I_TYPE;
1004         input[0].quality= 1 * FF_QP2LAMBDA;
1005         out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[0]);
1006 //        rd += (out_size * lambda2) >> FF_LAMBDA_SHIFT;
1007
1008         for(i=0; i<s->max_b_frames+1; i++){
1009             int is_p= i % (j+1) == j || i==s->max_b_frames;
1010
1011             input[i+1].pict_type= is_p ? FF_P_TYPE : FF_B_TYPE;
1012             input[i+1].quality= is_p ? p_lambda : b_lambda;
1013             out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[i+1]);
1014             rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
1015         }
1016
1017         /* get the delayed frames */
1018         while(out_size){
1019             out_size = avcodec_encode_video(c, outbuf, outbuf_size, NULL);
1020             rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
1021         }
1022
1023         rd += c->error[0] + c->error[1] + c->error[2];
1024
1025         if(rd < best_rd){
1026             best_rd= rd;
1027             best_b_count= j;
1028         }
1029     }
1030
1031     av_freep(&outbuf);
1032     avcodec_close(c);
1033     av_freep(&c);
1034
1035     for(i=0; i<s->max_b_frames+2; i++){
1036         av_freep(&input[i].data[0]);
1037     }
1038
1039     return best_b_count;
1040 }
1041
1042 static void select_input_picture(MpegEncContext *s){
1043     int i;
1044
1045     for(i=1; i<MAX_PICTURE_COUNT; i++)
1046         s->reordered_input_picture[i-1]= s->reordered_input_picture[i];
1047     s->reordered_input_picture[MAX_PICTURE_COUNT-1]= NULL;
1048
1049     /* set next picture type & ordering */
1050     if(s->reordered_input_picture[0]==NULL && s->input_picture[0]){
1051         if(/*s->picture_in_gop_number >= s->gop_size ||*/ s->next_picture_ptr==NULL || s->intra_only){
1052             s->reordered_input_picture[0]= s->input_picture[0];
1053             s->reordered_input_picture[0]->pict_type= FF_I_TYPE;
1054             s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++;
1055         }else{
1056             int b_frames;
1057
1058             if(s->avctx->frame_skip_threshold || s->avctx->frame_skip_factor){
1059                 if(s->picture_in_gop_number < s->gop_size && skip_check(s, s->input_picture[0], s->next_picture_ptr)){
1060                 //FIXME check that te gop check above is +-1 correct
1061 //av_log(NULL, AV_LOG_DEBUG, "skip %p %"PRId64"\n", s->input_picture[0]->data[0], s->input_picture[0]->pts);
1062
1063                     if(s->input_picture[0]->type == FF_BUFFER_TYPE_SHARED){
1064                         for(i=0; i<4; i++)
1065                             s->input_picture[0]->data[i]= NULL;
1066                         s->input_picture[0]->type= 0;
1067                     }else{
1068                         assert(   s->input_picture[0]->type==FF_BUFFER_TYPE_USER
1069                                || s->input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
1070
1071                         s->avctx->release_buffer(s->avctx, (AVFrame*)s->input_picture[0]);
1072                     }
1073
1074                     emms_c();
1075                     ff_vbv_update(s, 0);
1076
1077                     goto no_output_pic;
1078                 }
1079             }
1080
1081             if(s->flags&CODEC_FLAG_PASS2){
1082                 for(i=0; i<s->max_b_frames+1; i++){
1083                     int pict_num= s->input_picture[0]->display_picture_number + i;
1084
1085                     if(pict_num >= s->rc_context.num_entries)
1086                         break;
1087                     if(!s->input_picture[i]){
1088                         s->rc_context.entry[pict_num-1].new_pict_type = FF_P_TYPE;
1089                         break;
1090                     }
1091
1092                     s->input_picture[i]->pict_type=
1093                         s->rc_context.entry[pict_num].new_pict_type;
1094                 }
1095             }
1096
1097             if(s->avctx->b_frame_strategy==0){
1098                 b_frames= s->max_b_frames;
1099                 while(b_frames && !s->input_picture[b_frames]) b_frames--;
1100             }else if(s->avctx->b_frame_strategy==1){
1101                 for(i=1; i<s->max_b_frames+1; i++){
1102                     if(s->input_picture[i] && s->input_picture[i]->b_frame_score==0){
1103                         s->input_picture[i]->b_frame_score=
1104                             get_intra_count(s, s->input_picture[i  ]->data[0],
1105                                                s->input_picture[i-1]->data[0], s->linesize) + 1;
1106                     }
1107                 }
1108                 for(i=0; i<s->max_b_frames+1; i++){
1109                     if(s->input_picture[i]==NULL || s->input_picture[i]->b_frame_score - 1 > s->mb_num/s->avctx->b_sensitivity) break;
1110                 }
1111
1112                 b_frames= FFMAX(0, i-1);
1113
1114                 /* reset scores */
1115                 for(i=0; i<b_frames+1; i++){
1116                     s->input_picture[i]->b_frame_score=0;
1117                 }
1118             }else if(s->avctx->b_frame_strategy==2){
1119                 b_frames= estimate_best_b_count(s);
1120             }else{
1121                 av_log(s->avctx, AV_LOG_ERROR, "illegal b frame strategy\n");
1122                 b_frames=0;
1123             }
1124
1125             emms_c();
1126 //static int b_count=0;
1127 //b_count+= b_frames;
1128 //av_log(s->avctx, AV_LOG_DEBUG, "b_frames: %d\n", b_count);
1129
1130             for(i= b_frames - 1; i>=0; i--){
1131                 int type= s->input_picture[i]->pict_type;
1132                 if(type && type != FF_B_TYPE)
1133                     b_frames= i;
1134             }
1135             if(s->input_picture[b_frames]->pict_type == FF_B_TYPE && b_frames == s->max_b_frames){
1136                 av_log(s->avctx, AV_LOG_ERROR, "warning, too many b frames in a row\n");
1137             }
1138
1139             if(s->picture_in_gop_number + b_frames >= s->gop_size){
1140               if((s->flags2 & CODEC_FLAG2_STRICT_GOP) && s->gop_size > s->picture_in_gop_number){
1141                     b_frames= s->gop_size - s->picture_in_gop_number - 1;
1142               }else{
1143                 if(s->flags & CODEC_FLAG_CLOSED_GOP)
1144                     b_frames=0;
1145                 s->input_picture[b_frames]->pict_type= FF_I_TYPE;
1146               }
1147             }
1148
1149             if(   (s->flags & CODEC_FLAG_CLOSED_GOP)
1150                && b_frames
1151                && s->input_picture[b_frames]->pict_type== FF_I_TYPE)
1152                 b_frames--;
1153
1154             s->reordered_input_picture[0]= s->input_picture[b_frames];
1155             if(s->reordered_input_picture[0]->pict_type != FF_I_TYPE)
1156                 s->reordered_input_picture[0]->pict_type= FF_P_TYPE;
1157             s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++;
1158             for(i=0; i<b_frames; i++){
1159                 s->reordered_input_picture[i+1]= s->input_picture[i];
1160                 s->reordered_input_picture[i+1]->pict_type= FF_B_TYPE;
1161                 s->reordered_input_picture[i+1]->coded_picture_number= s->coded_picture_number++;
1162             }
1163         }
1164     }
1165 no_output_pic:
1166     if(s->reordered_input_picture[0]){
1167         s->reordered_input_picture[0]->reference= s->reordered_input_picture[0]->pict_type!=FF_B_TYPE ? 3 : 0;
1168
1169         copy_picture(&s->new_picture, s->reordered_input_picture[0]);
1170
1171         if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_SHARED || s->avctx->rc_buffer_size){
1172             // input is a shared pix, so we can't modifiy it -> alloc a new one & ensure that the shared one is reuseable
1173
1174             int i= ff_find_unused_picture(s, 0);
1175             Picture *pic= &s->picture[i];
1176
1177             pic->reference              = s->reordered_input_picture[0]->reference;
1178             alloc_picture(s, pic, 0);
1179
1180             /* mark us unused / free shared pic */
1181             if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_INTERNAL)
1182                 s->avctx->release_buffer(s->avctx, (AVFrame*)s->reordered_input_picture[0]);
1183             for(i=0; i<4; i++)
1184                 s->reordered_input_picture[0]->data[i]= NULL;
1185             s->reordered_input_picture[0]->type= 0;
1186
1187             copy_picture_attributes(s, (AVFrame*)pic, (AVFrame*)s->reordered_input_picture[0]);
1188
1189             s->current_picture_ptr= pic;
1190         }else{
1191             // input is not a shared pix -> reuse buffer for current_pix
1192
1193             assert(   s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_USER
1194                    || s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
1195
1196             s->current_picture_ptr= s->reordered_input_picture[0];
1197             for(i=0; i<4; i++){
1198                 s->new_picture.data[i]+= INPLACE_OFFSET;
1199             }
1200         }
1201         copy_picture(&s->current_picture, s->current_picture_ptr);
1202
1203         s->picture_number= s->new_picture.display_picture_number;
1204 //printf("dpn:%d\n", s->picture_number);
1205     }else{
1206        memset(&s->new_picture, 0, sizeof(Picture));
1207     }
1208 }
1209
1210 int MPV_encode_picture(AVCodecContext *avctx,
1211                        unsigned char *buf, int buf_size, void *data)
1212 {
1213     MpegEncContext *s = avctx->priv_data;
1214     AVFrame *pic_arg = data;
1215     int i, stuffing_count;
1216
1217     for(i=0; i<avctx->thread_count; i++){
1218         int start_y= s->thread_context[i]->start_mb_y;
1219         int   end_y= s->thread_context[i]->  end_mb_y;
1220         int h= s->mb_height;
1221         uint8_t *start= buf + (size_t)(((int64_t) buf_size)*start_y/h);
1222         uint8_t *end  = buf + (size_t)(((int64_t) buf_size)*  end_y/h);
1223
1224         init_put_bits(&s->thread_context[i]->pb, start, end - start);
1225     }
1226
1227     s->picture_in_gop_number++;
1228
1229     if(load_input_picture(s, pic_arg) < 0)
1230         return -1;
1231
1232     select_input_picture(s);
1233
1234     /* output? */
1235     if(s->new_picture.data[0]){
1236         s->pict_type= s->new_picture.pict_type;
1237 //emms_c();
1238 //printf("qs:%f %f %d\n", s->new_picture.quality, s->current_picture.quality, s->qscale);
1239         MPV_frame_start(s, avctx);
1240 vbv_retry:
1241         if (encode_picture(s, s->picture_number) < 0)
1242             return -1;
1243
1244         avctx->real_pict_num  = s->picture_number;
1245         avctx->header_bits = s->header_bits;
1246         avctx->mv_bits     = s->mv_bits;
1247         avctx->misc_bits   = s->misc_bits;
1248         avctx->i_tex_bits  = s->i_tex_bits;
1249         avctx->p_tex_bits  = s->p_tex_bits;
1250         avctx->i_count     = s->i_count;
1251         avctx->p_count     = s->mb_num - s->i_count - s->skip_count; //FIXME f/b_count in avctx
1252         avctx->skip_count  = s->skip_count;
1253
1254         MPV_frame_end(s);
1255
1256         if (ENABLE_MJPEG_ENCODER && s->out_format == FMT_MJPEG)
1257             ff_mjpeg_encode_picture_trailer(s);
1258
1259         if(avctx->rc_buffer_size){
1260             RateControlContext *rcc= &s->rc_context;
1261             int max_size= rcc->buffer_index/3;
1262
1263             if(put_bits_count(&s->pb) > max_size && s->lambda < s->avctx->lmax){
1264                 s->next_lambda= FFMAX(s->lambda+1, s->lambda*(s->qscale+1) / s->qscale);
1265                 if(s->adaptive_quant){
1266                     int i;
1267                     for(i=0; i<s->mb_height*s->mb_stride; i++)
1268                         s->lambda_table[i]= FFMAX(s->lambda_table[i]+1, s->lambda_table[i]*(s->qscale+1) / s->qscale);
1269                 }
1270                 s->mb_skipped = 0;        //done in MPV_frame_start()
1271                 if(s->pict_type==FF_P_TYPE){ //done in encode_picture() so we must undo it
1272                     if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
1273                         s->no_rounding ^= 1;
1274                 }
1275                 if(s->pict_type!=FF_B_TYPE){
1276                     s->time_base= s->last_time_base;
1277                     s->last_non_b_time= s->time - s->pp_time;
1278                 }
1279 //                av_log(NULL, AV_LOG_ERROR, "R:%d ", s->next_lambda);
1280                 for(i=0; i<avctx->thread_count; i++){
1281                     PutBitContext *pb= &s->thread_context[i]->pb;
1282                     init_put_bits(pb, pb->buf, pb->buf_end - pb->buf);
1283                 }
1284                 goto vbv_retry;
1285             }
1286
1287             assert(s->avctx->rc_max_rate);
1288         }
1289
1290         if(s->flags&CODEC_FLAG_PASS1)
1291             ff_write_pass1_stats(s);
1292
1293         for(i=0; i<4; i++){
1294             s->current_picture_ptr->error[i]= s->current_picture.error[i];
1295             avctx->error[i] += s->current_picture_ptr->error[i];
1296         }
1297
1298         if(s->flags&CODEC_FLAG_PASS1)
1299             assert(avctx->header_bits + avctx->mv_bits + avctx->misc_bits + avctx->i_tex_bits + avctx->p_tex_bits == put_bits_count(&s->pb));
1300         flush_put_bits(&s->pb);
1301         s->frame_bits  = put_bits_count(&s->pb);
1302
1303         stuffing_count= ff_vbv_update(s, s->frame_bits);
1304         if(stuffing_count){
1305             if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < stuffing_count + 50){
1306                 av_log(s->avctx, AV_LOG_ERROR, "stuffing too large\n");
1307                 return -1;
1308             }
1309
1310             switch(s->codec_id){
1311             case CODEC_ID_MPEG1VIDEO:
1312             case CODEC_ID_MPEG2VIDEO:
1313                 while(stuffing_count--){
1314                     put_bits(&s->pb, 8, 0);
1315                 }
1316             break;
1317             case CODEC_ID_MPEG4:
1318                 put_bits(&s->pb, 16, 0);
1319                 put_bits(&s->pb, 16, 0x1C3);
1320                 stuffing_count -= 4;
1321                 while(stuffing_count--){
1322                     put_bits(&s->pb, 8, 0xFF);
1323                 }
1324             break;
1325             default:
1326                 av_log(s->avctx, AV_LOG_ERROR, "vbv buffer overflow\n");
1327             }
1328             flush_put_bits(&s->pb);
1329             s->frame_bits  = put_bits_count(&s->pb);
1330         }
1331
1332         /* update mpeg1/2 vbv_delay for CBR */
1333         if(s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate && s->out_format == FMT_MPEG1
1334            && 90000LL * (avctx->rc_buffer_size-1) <= s->avctx->rc_max_rate*0xFFFFLL){
1335             int vbv_delay;
1336
1337             assert(s->repeat_first_field==0);
1338
1339             vbv_delay= lrintf(90000 * s->rc_context.buffer_index / s->avctx->rc_max_rate);
1340             assert(vbv_delay < 0xFFFF);
1341
1342             s->vbv_delay_ptr[0] &= 0xF8;
1343             s->vbv_delay_ptr[0] |= vbv_delay>>13;
1344             s->vbv_delay_ptr[1]  = vbv_delay>>5;
1345             s->vbv_delay_ptr[2] &= 0x07;
1346             s->vbv_delay_ptr[2] |= vbv_delay<<3;
1347         }
1348         s->total_bits += s->frame_bits;
1349         avctx->frame_bits  = s->frame_bits;
1350     }else{
1351         assert((pbBufPtr(&s->pb) == s->pb.buf));
1352         s->frame_bits=0;
1353     }
1354     assert((s->frame_bits&7)==0);
1355
1356     return s->frame_bits/8;
1357 }
1358
1359 static inline void dct_single_coeff_elimination(MpegEncContext *s, int n, int threshold)
1360 {
1361     static const char tab[64]=
1362         {3,2,2,1,1,1,1,1,
1363          1,1,1,1,1,1,1,1,
1364          1,1,1,1,1,1,1,1,
1365          0,0,0,0,0,0,0,0,
1366          0,0,0,0,0,0,0,0,
1367          0,0,0,0,0,0,0,0,
1368          0,0,0,0,0,0,0,0,
1369          0,0,0,0,0,0,0,0};
1370     int score=0;
1371     int run=0;
1372     int i;
1373     DCTELEM *block= s->block[n];
1374     const int last_index= s->block_last_index[n];
1375     int skip_dc;
1376
1377     if(threshold<0){
1378         skip_dc=0;
1379         threshold= -threshold;
1380     }else
1381         skip_dc=1;
1382
1383     /* Are all we could set to zero already zero? */
1384     if(last_index<=skip_dc - 1) return;
1385
1386     for(i=0; i<=last_index; i++){
1387         const int j = s->intra_scantable.permutated[i];
1388         const int level = FFABS(block[j]);
1389         if(level==1){
1390             if(skip_dc && i==0) continue;
1391             score+= tab[run];
1392             run=0;
1393         }else if(level>1){
1394             return;
1395         }else{
1396             run++;
1397         }
1398     }
1399     if(score >= threshold) return;
1400     for(i=skip_dc; i<=last_index; i++){
1401         const int j = s->intra_scantable.permutated[i];
1402         block[j]=0;
1403     }
1404     if(block[0]) s->block_last_index[n]= 0;
1405     else         s->block_last_index[n]= -1;
1406 }
1407
1408 static inline void clip_coeffs(MpegEncContext *s, DCTELEM *block, int last_index)
1409 {
1410     int i;
1411     const int maxlevel= s->max_qcoeff;
1412     const int minlevel= s->min_qcoeff;
1413     int overflow=0;
1414
1415     if(s->mb_intra){
1416         i=1; //skip clipping of intra dc
1417     }else
1418         i=0;
1419
1420     for(;i<=last_index; i++){
1421         const int j= s->intra_scantable.permutated[i];
1422         int level = block[j];
1423
1424         if     (level>maxlevel){
1425             level=maxlevel;
1426             overflow++;
1427         }else if(level<minlevel){
1428             level=minlevel;
1429             overflow++;
1430         }
1431
1432         block[j]= level;
1433     }
1434
1435     if(overflow && s->avctx->mb_decision == FF_MB_DECISION_SIMPLE)
1436         av_log(s->avctx, AV_LOG_INFO, "warning, clipping %d dct coefficients to %d..%d\n", overflow, minlevel, maxlevel);
1437 }
1438
1439 static void get_visual_weight(int16_t *weight, uint8_t *ptr, int stride){
1440     int x, y;
1441 //FIXME optimize
1442     for(y=0; y<8; y++){
1443         for(x=0; x<8; x++){
1444             int x2, y2;
1445             int sum=0;
1446             int sqr=0;
1447             int count=0;
1448
1449             for(y2= FFMAX(y-1, 0); y2 < FFMIN(8, y+2); y2++){
1450                 for(x2= FFMAX(x-1, 0); x2 < FFMIN(8, x+2); x2++){
1451                     int v= ptr[x2 + y2*stride];
1452                     sum += v;
1453                     sqr += v*v;
1454                     count++;
1455                 }
1456             }
1457             weight[x + 8*y]= (36*ff_sqrt(count*sqr - sum*sum)) / count;
1458         }
1459     }
1460 }
1461
1462 static av_always_inline void encode_mb_internal(MpegEncContext *s, int motion_x, int motion_y, int mb_block_height, int mb_block_count)
1463 {
1464     int16_t weight[8][64];
1465     DCTELEM orig[8][64];
1466     const int mb_x= s->mb_x;
1467     const int mb_y= s->mb_y;
1468     int i;
1469     int skip_dct[8];
1470     int dct_offset   = s->linesize*8; //default for progressive frames
1471     uint8_t *ptr_y, *ptr_cb, *ptr_cr;
1472     int wrap_y, wrap_c;
1473
1474     for(i=0; i<mb_block_count; i++) skip_dct[i]=s->skipdct;
1475
1476     if(s->adaptive_quant){
1477         const int last_qp= s->qscale;
1478         const int mb_xy= mb_x + mb_y*s->mb_stride;
1479
1480         s->lambda= s->lambda_table[mb_xy];
1481         update_qscale(s);
1482
1483         if(!(s->flags&CODEC_FLAG_QP_RD)){
1484             s->qscale= s->current_picture_ptr->qscale_table[mb_xy];
1485             s->dquant= s->qscale - last_qp;
1486
1487             if(s->out_format==FMT_H263){
1488                 s->dquant= av_clip(s->dquant, -2, 2);
1489
1490                 if(s->codec_id==CODEC_ID_MPEG4){
1491                     if(!s->mb_intra){
1492                         if(s->pict_type == FF_B_TYPE){
1493                             if(s->dquant&1 || s->mv_dir&MV_DIRECT)
1494                                 s->dquant= 0;
1495                         }
1496                         if(s->mv_type==MV_TYPE_8X8)
1497                             s->dquant=0;
1498                     }
1499                 }
1500             }
1501         }
1502         ff_set_qscale(s, last_qp + s->dquant);
1503     }else if(s->flags&CODEC_FLAG_QP_RD)
1504         ff_set_qscale(s, s->qscale + s->dquant);
1505
1506     wrap_y = s->linesize;
1507     wrap_c = s->uvlinesize;
1508     ptr_y = s->new_picture.data[0] + (mb_y * 16 * wrap_y) + mb_x * 16;
1509     ptr_cb = s->new_picture.data[1] + (mb_y * mb_block_height * wrap_c) + mb_x * 8;
1510     ptr_cr = s->new_picture.data[2] + (mb_y * mb_block_height * wrap_c) + mb_x * 8;
1511
1512     if(mb_x*16+16 > s->width || mb_y*16+16 > s->height){
1513         uint8_t *ebuf= s->edge_emu_buffer + 32;
1514         ff_emulated_edge_mc(ebuf            , ptr_y , wrap_y,16,16,mb_x*16,mb_y*16, s->width   , s->height);
1515         ptr_y= ebuf;
1516         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);
1517         ptr_cb= ebuf+18*wrap_y;
1518         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);
1519         ptr_cr= ebuf+18*wrap_y+8;
1520     }
1521
1522     if (s->mb_intra) {
1523         if(s->flags&CODEC_FLAG_INTERLACED_DCT){
1524             int progressive_score, interlaced_score;
1525
1526             s->interlaced_dct=0;
1527             progressive_score= s->dsp.ildct_cmp[4](s, ptr_y           , NULL, wrap_y, 8)
1528                               +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y*8, NULL, wrap_y, 8) - 400;
1529
1530             if(progressive_score > 0){
1531                 interlaced_score = s->dsp.ildct_cmp[4](s, ptr_y           , NULL, wrap_y*2, 8)
1532                                   +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y  , NULL, wrap_y*2, 8);
1533                 if(progressive_score > interlaced_score){
1534                     s->interlaced_dct=1;
1535
1536                     dct_offset= wrap_y;
1537                     wrap_y<<=1;
1538                     if (s->chroma_format == CHROMA_422)
1539                         wrap_c<<=1;
1540                 }
1541             }
1542         }
1543
1544         s->dsp.get_pixels(s->block[0], ptr_y                 , wrap_y);
1545         s->dsp.get_pixels(s->block[1], ptr_y              + 8, wrap_y);
1546         s->dsp.get_pixels(s->block[2], ptr_y + dct_offset    , wrap_y);
1547         s->dsp.get_pixels(s->block[3], ptr_y + dct_offset + 8, wrap_y);
1548
1549         if(s->flags&CODEC_FLAG_GRAY){
1550             skip_dct[4]= 1;
1551             skip_dct[5]= 1;
1552         }else{
1553             s->dsp.get_pixels(s->block[4], ptr_cb, wrap_c);
1554             s->dsp.get_pixels(s->block[5], ptr_cr, wrap_c);
1555             if(!s->chroma_y_shift){ /* 422 */
1556                 s->dsp.get_pixels(s->block[6], ptr_cb + (dct_offset>>1), wrap_c);
1557                 s->dsp.get_pixels(s->block[7], ptr_cr + (dct_offset>>1), wrap_c);
1558             }
1559         }
1560     }else{
1561         op_pixels_func (*op_pix)[4];
1562         qpel_mc_func (*op_qpix)[16];
1563         uint8_t *dest_y, *dest_cb, *dest_cr;
1564
1565         dest_y  = s->dest[0];
1566         dest_cb = s->dest[1];
1567         dest_cr = s->dest[2];
1568
1569         if ((!s->no_rounding) || s->pict_type==FF_B_TYPE){
1570             op_pix = s->dsp.put_pixels_tab;
1571             op_qpix= s->dsp.put_qpel_pixels_tab;
1572         }else{
1573             op_pix = s->dsp.put_no_rnd_pixels_tab;
1574             op_qpix= s->dsp.put_no_rnd_qpel_pixels_tab;
1575         }
1576
1577         if (s->mv_dir & MV_DIR_FORWARD) {
1578             MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.data, op_pix, op_qpix);
1579             op_pix = s->dsp.avg_pixels_tab;
1580             op_qpix= s->dsp.avg_qpel_pixels_tab;
1581         }
1582         if (s->mv_dir & MV_DIR_BACKWARD) {
1583             MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.data, op_pix, op_qpix);
1584         }
1585
1586         if(s->flags&CODEC_FLAG_INTERLACED_DCT){
1587             int progressive_score, interlaced_score;
1588
1589             s->interlaced_dct=0;
1590             progressive_score= s->dsp.ildct_cmp[0](s, dest_y           , ptr_y           , wrap_y, 8)
1591                               +s->dsp.ildct_cmp[0](s, dest_y + wrap_y*8, ptr_y + wrap_y*8, wrap_y, 8) - 400;
1592
1593             if(s->avctx->ildct_cmp == FF_CMP_VSSE) progressive_score -= 400;
1594
1595             if(progressive_score>0){
1596                 interlaced_score = s->dsp.ildct_cmp[0](s, dest_y           , ptr_y           , wrap_y*2, 8)
1597                                   +s->dsp.ildct_cmp[0](s, dest_y + wrap_y  , ptr_y + wrap_y  , wrap_y*2, 8);
1598
1599                 if(progressive_score > interlaced_score){
1600                     s->interlaced_dct=1;
1601
1602                     dct_offset= wrap_y;
1603                     wrap_y<<=1;
1604                     if (s->chroma_format == CHROMA_422)
1605                         wrap_c<<=1;
1606                 }
1607             }
1608         }
1609
1610         s->dsp.diff_pixels(s->block[0], ptr_y                 , dest_y                 , wrap_y);
1611         s->dsp.diff_pixels(s->block[1], ptr_y              + 8, dest_y              + 8, wrap_y);
1612         s->dsp.diff_pixels(s->block[2], ptr_y + dct_offset    , dest_y + dct_offset    , wrap_y);
1613         s->dsp.diff_pixels(s->block[3], ptr_y + dct_offset + 8, dest_y + dct_offset + 8, wrap_y);
1614
1615         if(s->flags&CODEC_FLAG_GRAY){
1616             skip_dct[4]= 1;
1617             skip_dct[5]= 1;
1618         }else{
1619             s->dsp.diff_pixels(s->block[4], ptr_cb, dest_cb, wrap_c);
1620             s->dsp.diff_pixels(s->block[5], ptr_cr, dest_cr, wrap_c);
1621             if(!s->chroma_y_shift){ /* 422 */
1622                 s->dsp.diff_pixels(s->block[6], ptr_cb + (dct_offset>>1), dest_cb + (dct_offset>>1), wrap_c);
1623                 s->dsp.diff_pixels(s->block[7], ptr_cr + (dct_offset>>1), dest_cr + (dct_offset>>1), wrap_c);
1624             }
1625         }
1626         /* pre quantization */
1627         if(s->current_picture.mc_mb_var[s->mb_stride*mb_y+ mb_x]<2*s->qscale*s->qscale){
1628             //FIXME optimize
1629             if(s->dsp.sad[1](NULL, ptr_y               , dest_y               , wrap_y, 8) < 20*s->qscale) skip_dct[0]= 1;
1630             if(s->dsp.sad[1](NULL, ptr_y            + 8, dest_y            + 8, wrap_y, 8) < 20*s->qscale) skip_dct[1]= 1;
1631             if(s->dsp.sad[1](NULL, ptr_y +dct_offset   , dest_y +dct_offset   , wrap_y, 8) < 20*s->qscale) skip_dct[2]= 1;
1632             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;
1633             if(s->dsp.sad[1](NULL, ptr_cb              , dest_cb              , wrap_c, 8) < 20*s->qscale) skip_dct[4]= 1;
1634             if(s->dsp.sad[1](NULL, ptr_cr              , dest_cr              , wrap_c, 8) < 20*s->qscale) skip_dct[5]= 1;
1635             if(!s->chroma_y_shift){ /* 422 */
1636                 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;
1637                 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;
1638             }
1639         }
1640     }
1641
1642     if(s->avctx->quantizer_noise_shaping){
1643         if(!skip_dct[0]) get_visual_weight(weight[0], ptr_y                 , wrap_y);
1644         if(!skip_dct[1]) get_visual_weight(weight[1], ptr_y              + 8, wrap_y);
1645         if(!skip_dct[2]) get_visual_weight(weight[2], ptr_y + dct_offset    , wrap_y);
1646         if(!skip_dct[3]) get_visual_weight(weight[3], ptr_y + dct_offset + 8, wrap_y);
1647         if(!skip_dct[4]) get_visual_weight(weight[4], ptr_cb                , wrap_c);
1648         if(!skip_dct[5]) get_visual_weight(weight[5], ptr_cr                , wrap_c);
1649         if(!s->chroma_y_shift){ /* 422 */
1650             if(!skip_dct[6]) get_visual_weight(weight[6], ptr_cb + (dct_offset>>1), wrap_c);
1651             if(!skip_dct[7]) get_visual_weight(weight[7], ptr_cr + (dct_offset>>1), wrap_c);
1652         }
1653         memcpy(orig[0], s->block[0], sizeof(DCTELEM)*64*mb_block_count);
1654     }
1655
1656     /* DCT & quantize */
1657     assert(s->out_format!=FMT_MJPEG || s->qscale==8);
1658     {
1659         for(i=0;i<mb_block_count;i++) {
1660             if(!skip_dct[i]){
1661                 int overflow;
1662                 s->block_last_index[i] = s->dct_quantize(s, s->block[i], i, s->qscale, &overflow);
1663             // FIXME we could decide to change to quantizer instead of clipping
1664             // JS: I don't think that would be a good idea it could lower quality instead
1665             //     of improve it. Just INTRADC clipping deserves changes in quantizer
1666                 if (overflow) clip_coeffs(s, s->block[i], s->block_last_index[i]);
1667             }else
1668                 s->block_last_index[i]= -1;
1669         }
1670         if(s->avctx->quantizer_noise_shaping){
1671             for(i=0;i<mb_block_count;i++) {
1672                 if(!skip_dct[i]){
1673                     s->block_last_index[i] = dct_quantize_refine(s, s->block[i], weight[i], orig[i], i, s->qscale);
1674                 }
1675             }
1676         }
1677
1678         if(s->luma_elim_threshold && !s->mb_intra)
1679             for(i=0; i<4; i++)
1680                 dct_single_coeff_elimination(s, i, s->luma_elim_threshold);
1681         if(s->chroma_elim_threshold && !s->mb_intra)
1682             for(i=4; i<mb_block_count; i++)
1683                 dct_single_coeff_elimination(s, i, s->chroma_elim_threshold);
1684
1685         if(s->flags & CODEC_FLAG_CBP_RD){
1686             for(i=0;i<mb_block_count;i++) {
1687                 if(s->block_last_index[i] == -1)
1688                     s->coded_score[i]= INT_MAX/256;
1689             }
1690         }
1691     }
1692
1693     if((s->flags&CODEC_FLAG_GRAY) && s->mb_intra){
1694         s->block_last_index[4]=
1695         s->block_last_index[5]= 0;
1696         s->block[4][0]=
1697         s->block[5][0]= (1024 + s->c_dc_scale/2)/ s->c_dc_scale;
1698     }
1699
1700     //non c quantize code returns incorrect block_last_index FIXME
1701     if(s->alternate_scan && s->dct_quantize != dct_quantize_c){
1702         for(i=0; i<mb_block_count; i++){
1703             int j;
1704             if(s->block_last_index[i]>0){
1705                 for(j=63; j>0; j--){
1706                     if(s->block[i][ s->intra_scantable.permutated[j] ]) break;
1707                 }
1708                 s->block_last_index[i]= j;
1709             }
1710         }
1711     }
1712
1713     /* huffman encode */
1714     switch(s->codec_id){ //FIXME funct ptr could be slightly faster
1715     case CODEC_ID_MPEG1VIDEO:
1716     case CODEC_ID_MPEG2VIDEO:
1717         if (ENABLE_MPEG1VIDEO_ENCODER || ENABLE_MPEG2VIDEO_ENCODER)
1718             mpeg1_encode_mb(s, s->block, motion_x, motion_y);
1719         break;
1720     case CODEC_ID_MPEG4:
1721         if (ENABLE_MPEG4_ENCODER)
1722             mpeg4_encode_mb(s, s->block, motion_x, motion_y);
1723         break;
1724     case CODEC_ID_MSMPEG4V2:
1725     case CODEC_ID_MSMPEG4V3:
1726     case CODEC_ID_WMV1:
1727         if (ENABLE_MSMPEG4_ENCODER)
1728             msmpeg4_encode_mb(s, s->block, motion_x, motion_y);
1729         break;
1730     case CODEC_ID_WMV2:
1731         if (ENABLE_WMV2_ENCODER)
1732             ff_wmv2_encode_mb(s, s->block, motion_x, motion_y);
1733         break;
1734     case CODEC_ID_H261:
1735         if (ENABLE_H261_ENCODER)
1736             ff_h261_encode_mb(s, s->block, motion_x, motion_y);
1737         break;
1738     case CODEC_ID_H263:
1739     case CODEC_ID_H263P:
1740     case CODEC_ID_FLV1:
1741     case CODEC_ID_RV10:
1742     case CODEC_ID_RV20:
1743         if (ENABLE_H263_ENCODER || ENABLE_H263P_ENCODER ||
1744             ENABLE_FLV_ENCODER  || ENABLE_RV10_ENCODER  || ENABLE_RV20_ENCODER)
1745             h263_encode_mb(s, s->block, motion_x, motion_y);
1746         break;
1747     case CODEC_ID_MJPEG:
1748         if (ENABLE_MJPEG_ENCODER)
1749             ff_mjpeg_encode_mb(s, s->block);
1750         break;
1751     default:
1752         assert(0);
1753     }
1754 }
1755
1756 static av_always_inline void encode_mb(MpegEncContext *s, int motion_x, int motion_y)
1757 {
1758     if (s->chroma_format == CHROMA_420) encode_mb_internal(s, motion_x, motion_y,  8, 6);
1759     else                                encode_mb_internal(s, motion_x, motion_y, 16, 8);
1760 }
1761
1762 static inline void copy_context_before_encode(MpegEncContext *d, MpegEncContext *s, int type){
1763     int i;
1764
1765     memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
1766
1767     /* mpeg1 */
1768     d->mb_skip_run= s->mb_skip_run;
1769     for(i=0; i<3; i++)
1770         d->last_dc[i]= s->last_dc[i];
1771
1772     /* statistics */
1773     d->mv_bits= s->mv_bits;
1774     d->i_tex_bits= s->i_tex_bits;
1775     d->p_tex_bits= s->p_tex_bits;
1776     d->i_count= s->i_count;
1777     d->f_count= s->f_count;
1778     d->b_count= s->b_count;
1779     d->skip_count= s->skip_count;
1780     d->misc_bits= s->misc_bits;
1781     d->last_bits= 0;
1782
1783     d->mb_skipped= 0;
1784     d->qscale= s->qscale;
1785     d->dquant= s->dquant;
1786
1787     d->esc3_level_length= s->esc3_level_length;
1788 }
1789
1790 static inline void copy_context_after_encode(MpegEncContext *d, MpegEncContext *s, int type){
1791     int i;
1792
1793     memcpy(d->mv, s->mv, 2*4*2*sizeof(int));
1794     memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
1795
1796     /* mpeg1 */
1797     d->mb_skip_run= s->mb_skip_run;
1798     for(i=0; i<3; i++)
1799         d->last_dc[i]= s->last_dc[i];
1800
1801     /* statistics */
1802     d->mv_bits= s->mv_bits;
1803     d->i_tex_bits= s->i_tex_bits;
1804     d->p_tex_bits= s->p_tex_bits;
1805     d->i_count= s->i_count;
1806     d->f_count= s->f_count;
1807     d->b_count= s->b_count;
1808     d->skip_count= s->skip_count;
1809     d->misc_bits= s->misc_bits;
1810
1811     d->mb_intra= s->mb_intra;
1812     d->mb_skipped= s->mb_skipped;
1813     d->mv_type= s->mv_type;
1814     d->mv_dir= s->mv_dir;
1815     d->pb= s->pb;
1816     if(s->data_partitioning){
1817         d->pb2= s->pb2;
1818         d->tex_pb= s->tex_pb;
1819     }
1820     d->block= s->block;
1821     for(i=0; i<8; i++)
1822         d->block_last_index[i]= s->block_last_index[i];
1823     d->interlaced_dct= s->interlaced_dct;
1824     d->qscale= s->qscale;
1825
1826     d->esc3_level_length= s->esc3_level_length;
1827 }
1828
1829 static inline void encode_mb_hq(MpegEncContext *s, MpegEncContext *backup, MpegEncContext *best, int type,
1830                            PutBitContext pb[2], PutBitContext pb2[2], PutBitContext tex_pb[2],
1831                            int *dmin, int *next_block, int motion_x, int motion_y)
1832 {
1833     int score;
1834     uint8_t *dest_backup[3];
1835
1836     copy_context_before_encode(s, backup, type);
1837
1838     s->block= s->blocks[*next_block];
1839     s->pb= pb[*next_block];
1840     if(s->data_partitioning){
1841         s->pb2   = pb2   [*next_block];
1842         s->tex_pb= tex_pb[*next_block];
1843     }
1844
1845     if(*next_block){
1846         memcpy(dest_backup, s->dest, sizeof(s->dest));
1847         s->dest[0] = s->rd_scratchpad;
1848         s->dest[1] = s->rd_scratchpad + 16*s->linesize;
1849         s->dest[2] = s->rd_scratchpad + 16*s->linesize + 8;
1850         assert(s->linesize >= 32); //FIXME
1851     }
1852
1853     encode_mb(s, motion_x, motion_y);
1854
1855     score= put_bits_count(&s->pb);
1856     if(s->data_partitioning){
1857         score+= put_bits_count(&s->pb2);
1858         score+= put_bits_count(&s->tex_pb);
1859     }
1860
1861     if(s->avctx->mb_decision == FF_MB_DECISION_RD){
1862         MPV_decode_mb(s, s->block);
1863
1864         score *= s->lambda2;
1865         score += sse_mb(s) << FF_LAMBDA_SHIFT;
1866     }
1867
1868     if(*next_block){
1869         memcpy(s->dest, dest_backup, sizeof(s->dest));
1870     }
1871
1872     if(score<*dmin){
1873         *dmin= score;
1874         *next_block^=1;
1875
1876         copy_context_after_encode(best, s, type);
1877     }
1878 }
1879
1880 static int sse(MpegEncContext *s, uint8_t *src1, uint8_t *src2, int w, int h, int stride){
1881     uint32_t *sq = ff_squareTbl + 256;
1882     int acc=0;
1883     int x,y;
1884
1885     if(w==16 && h==16)
1886         return s->dsp.sse[0](NULL, src1, src2, stride, 16);
1887     else if(w==8 && h==8)
1888         return s->dsp.sse[1](NULL, src1, src2, stride, 8);
1889
1890     for(y=0; y<h; y++){
1891         for(x=0; x<w; x++){
1892             acc+= sq[src1[x + y*stride] - src2[x + y*stride]];
1893         }
1894     }
1895
1896     assert(acc>=0);
1897
1898     return acc;
1899 }
1900
1901 static int sse_mb(MpegEncContext *s){
1902     int w= 16;
1903     int h= 16;
1904
1905     if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
1906     if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
1907
1908     if(w==16 && h==16)
1909       if(s->avctx->mb_cmp == FF_CMP_NSSE){
1910         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)
1911                +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)
1912                +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);
1913       }else{
1914         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)
1915                +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)
1916                +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);
1917       }
1918     else
1919         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)
1920                +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)
1921                +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);
1922 }
1923
1924 static int pre_estimate_motion_thread(AVCodecContext *c, void *arg){
1925     MpegEncContext *s= arg;
1926
1927
1928     s->me.pre_pass=1;
1929     s->me.dia_size= s->avctx->pre_dia_size;
1930     s->first_slice_line=1;
1931     for(s->mb_y= s->end_mb_y-1; s->mb_y >= s->start_mb_y; s->mb_y--) {
1932         for(s->mb_x=s->mb_width-1; s->mb_x >=0 ;s->mb_x--) {
1933             ff_pre_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
1934         }
1935         s->first_slice_line=0;
1936     }
1937
1938     s->me.pre_pass=0;
1939
1940     return 0;
1941 }
1942
1943 static int estimate_motion_thread(AVCodecContext *c, void *arg){
1944     MpegEncContext *s= arg;
1945
1946     ff_check_alignment();
1947
1948     s->me.dia_size= s->avctx->dia_size;
1949     s->first_slice_line=1;
1950     for(s->mb_y= s->start_mb_y; s->mb_y < s->end_mb_y; s->mb_y++) {
1951         s->mb_x=0; //for block init below
1952         ff_init_block_index(s);
1953         for(s->mb_x=0; s->mb_x < s->mb_width; s->mb_x++) {
1954             s->block_index[0]+=2;
1955             s->block_index[1]+=2;
1956             s->block_index[2]+=2;
1957             s->block_index[3]+=2;
1958
1959             /* compute motion vector & mb_type and store in context */
1960             if(s->pict_type==FF_B_TYPE)
1961                 ff_estimate_b_frame_motion(s, s->mb_x, s->mb_y);
1962             else
1963                 ff_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
1964         }
1965         s->first_slice_line=0;
1966     }
1967     return 0;
1968 }
1969
1970 static int mb_var_thread(AVCodecContext *c, void *arg){
1971     MpegEncContext *s= arg;
1972     int mb_x, mb_y;
1973
1974     ff_check_alignment();
1975
1976     for(mb_y=s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
1977         for(mb_x=0; mb_x < s->mb_width; mb_x++) {
1978             int xx = mb_x * 16;
1979             int yy = mb_y * 16;
1980             uint8_t *pix = s->new_picture.data[0] + (yy * s->linesize) + xx;
1981             int varc;
1982             int sum = s->dsp.pix_sum(pix, s->linesize);
1983
1984             varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500 + 128)>>8;
1985
1986             s->current_picture.mb_var [s->mb_stride * mb_y + mb_x] = varc;
1987             s->current_picture.mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
1988             s->me.mb_var_sum_temp    += varc;
1989         }
1990     }
1991     return 0;
1992 }
1993
1994 static void write_slice_end(MpegEncContext *s){
1995     if(ENABLE_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4){
1996         if(s->partitioned_frame){
1997             ff_mpeg4_merge_partitions(s);
1998         }
1999
2000         ff_mpeg4_stuffing(&s->pb);
2001     }else if(ENABLE_MJPEG_ENCODER && s->out_format == FMT_MJPEG){
2002         ff_mjpeg_encode_stuffing(&s->pb);
2003     }
2004
2005     align_put_bits(&s->pb);
2006     flush_put_bits(&s->pb);
2007
2008     if((s->flags&CODEC_FLAG_PASS1) && !s->partitioned_frame)
2009         s->misc_bits+= get_bits_diff(s);
2010 }
2011
2012 static int encode_thread(AVCodecContext *c, void *arg){
2013     MpegEncContext *s= arg;
2014     int mb_x, mb_y, pdif = 0;
2015     int i, j;
2016     MpegEncContext best_s, backup_s;
2017     uint8_t bit_buf[2][MAX_MB_BYTES];
2018     uint8_t bit_buf2[2][MAX_MB_BYTES];
2019     uint8_t bit_buf_tex[2][MAX_MB_BYTES];
2020     PutBitContext pb[2], pb2[2], tex_pb[2];
2021 //printf("%d->%d\n", s->resync_mb_y, s->end_mb_y);
2022
2023     ff_check_alignment();
2024
2025     for(i=0; i<2; i++){
2026         init_put_bits(&pb    [i], bit_buf    [i], MAX_MB_BYTES);
2027         init_put_bits(&pb2   [i], bit_buf2   [i], MAX_MB_BYTES);
2028         init_put_bits(&tex_pb[i], bit_buf_tex[i], MAX_MB_BYTES);
2029     }
2030
2031     s->last_bits= put_bits_count(&s->pb);
2032     s->mv_bits=0;
2033     s->misc_bits=0;
2034     s->i_tex_bits=0;
2035     s->p_tex_bits=0;
2036     s->i_count=0;
2037     s->f_count=0;
2038     s->b_count=0;
2039     s->skip_count=0;
2040
2041     for(i=0; i<3; i++){
2042         /* init last dc values */
2043         /* note: quant matrix value (8) is implied here */
2044         s->last_dc[i] = 128 << s->intra_dc_precision;
2045
2046         s->current_picture.error[i] = 0;
2047     }
2048     s->mb_skip_run = 0;
2049     memset(s->last_mv, 0, sizeof(s->last_mv));
2050
2051     s->last_mv_dir = 0;
2052
2053     switch(s->codec_id){
2054     case CODEC_ID_H263:
2055     case CODEC_ID_H263P:
2056     case CODEC_ID_FLV1:
2057         if (ENABLE_H263_ENCODER || ENABLE_H263P_ENCODER || ENABLE_FLV_ENCODER)
2058             s->gob_index = ff_h263_get_gob_height(s);
2059         break;
2060     case CODEC_ID_MPEG4:
2061         if(ENABLE_MPEG4_ENCODER && s->partitioned_frame)
2062             ff_mpeg4_init_partitions(s);
2063         break;
2064     }
2065
2066     s->resync_mb_x=0;
2067     s->resync_mb_y=0;
2068     s->first_slice_line = 1;
2069     s->ptr_lastgob = s->pb.buf;
2070     for(mb_y= s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
2071 //    printf("row %d at %X\n", s->mb_y, (int)s);
2072         s->mb_x=0;
2073         s->mb_y= mb_y;
2074
2075         ff_set_qscale(s, s->qscale);
2076         ff_init_block_index(s);
2077
2078         for(mb_x=0; mb_x < s->mb_width; mb_x++) {
2079             int xy= mb_y*s->mb_stride + mb_x; // removed const, H261 needs to adjust this
2080             int mb_type= s->mb_type[xy];
2081 //            int d;
2082             int dmin= INT_MAX;
2083             int dir;
2084
2085             if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < MAX_MB_BYTES){
2086                 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
2087                 return -1;
2088             }
2089             if(s->data_partitioning){
2090                 if(   s->pb2   .buf_end - s->pb2   .buf - (put_bits_count(&s->    pb2)>>3) < MAX_MB_BYTES
2091                    || s->tex_pb.buf_end - s->tex_pb.buf - (put_bits_count(&s->tex_pb )>>3) < MAX_MB_BYTES){
2092                     av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
2093                     return -1;
2094                 }
2095             }
2096
2097             s->mb_x = mb_x;
2098             s->mb_y = mb_y;  // moved into loop, can get changed by H.261
2099             ff_update_block_index(s);
2100
2101             if(ENABLE_H261_ENCODER && s->codec_id == CODEC_ID_H261){
2102                 ff_h261_reorder_mb_index(s);
2103                 xy= s->mb_y*s->mb_stride + s->mb_x;
2104                 mb_type= s->mb_type[xy];
2105             }
2106
2107             /* write gob / video packet header  */
2108             if(s->rtp_mode){
2109                 int current_packet_size, is_gob_start;
2110
2111                 current_packet_size= ((put_bits_count(&s->pb)+7)>>3) - (s->ptr_lastgob - s->pb.buf);
2112
2113                 is_gob_start= s->avctx->rtp_payload_size && current_packet_size >= s->avctx->rtp_payload_size && mb_y + mb_x>0;
2114
2115                 if(s->start_mb_y == mb_y && mb_y > 0 && mb_x==0) is_gob_start=1;
2116
2117                 switch(s->codec_id){
2118                 case CODEC_ID_H263:
2119                 case CODEC_ID_H263P:
2120                     if(!s->h263_slice_structured)
2121                         if(s->mb_x || s->mb_y%s->gob_index) is_gob_start=0;
2122                     break;
2123                 case CODEC_ID_MPEG2VIDEO:
2124                     if(s->mb_x==0 && s->mb_y!=0) is_gob_start=1;
2125                 case CODEC_ID_MPEG1VIDEO:
2126                     if(s->mb_skip_run) is_gob_start=0;
2127                     break;
2128                 }
2129
2130                 if(is_gob_start){
2131                     if(s->start_mb_y != mb_y || mb_x!=0){
2132                         write_slice_end(s);
2133
2134                         if(ENABLE_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4 && s->partitioned_frame){
2135                             ff_mpeg4_init_partitions(s);
2136                         }
2137                     }
2138
2139                     assert((put_bits_count(&s->pb)&7) == 0);
2140                     current_packet_size= pbBufPtr(&s->pb) - s->ptr_lastgob;
2141
2142                     if(s->avctx->error_rate && s->resync_mb_x + s->resync_mb_y > 0){
2143                         int r= put_bits_count(&s->pb)/8 + s->picture_number + 16 + s->mb_x + s->mb_y;
2144                         int d= 100 / s->avctx->error_rate;
2145                         if(r % d == 0){
2146                             current_packet_size=0;
2147 #ifndef ALT_BITSTREAM_WRITER
2148                             s->pb.buf_ptr= s->ptr_lastgob;
2149 #endif
2150                             assert(pbBufPtr(&s->pb) == s->ptr_lastgob);
2151                         }
2152                     }
2153
2154                     if (s->avctx->rtp_callback){
2155                         int number_mb = (mb_y - s->resync_mb_y)*s->mb_width + mb_x - s->resync_mb_x;
2156                         s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, current_packet_size, number_mb);
2157                     }
2158
2159                     switch(s->codec_id){
2160                     case CODEC_ID_MPEG4:
2161                         if (ENABLE_MPEG4_ENCODER) {
2162                             ff_mpeg4_encode_video_packet_header(s);
2163                             ff_mpeg4_clean_buffers(s);
2164                         }
2165                     break;
2166                     case CODEC_ID_MPEG1VIDEO:
2167                     case CODEC_ID_MPEG2VIDEO:
2168                         if (ENABLE_MPEG1VIDEO_ENCODER || ENABLE_MPEG2VIDEO_ENCODER) {
2169                             ff_mpeg1_encode_slice_header(s);
2170                             ff_mpeg1_clean_buffers(s);
2171                         }
2172                     break;
2173                     case CODEC_ID_H263:
2174                     case CODEC_ID_H263P:
2175                         if (ENABLE_H263_ENCODER || ENABLE_H263P_ENCODER)
2176                             h263_encode_gob_header(s, mb_y);
2177                     break;
2178                     }
2179
2180                     if(s->flags&CODEC_FLAG_PASS1){
2181                         int bits= put_bits_count(&s->pb);
2182                         s->misc_bits+= bits - s->last_bits;
2183                         s->last_bits= bits;
2184                     }
2185
2186                     s->ptr_lastgob += current_packet_size;
2187                     s->first_slice_line=1;
2188                     s->resync_mb_x=mb_x;
2189                     s->resync_mb_y=mb_y;
2190                 }
2191             }
2192
2193             if(  (s->resync_mb_x   == s->mb_x)
2194                && s->resync_mb_y+1 == s->mb_y){
2195                 s->first_slice_line=0;
2196             }
2197
2198             s->mb_skipped=0;
2199             s->dquant=0; //only for QP_RD
2200
2201             if(mb_type & (mb_type-1) || (s->flags & CODEC_FLAG_QP_RD)){ // more than 1 MB type possible or CODEC_FLAG_QP_RD
2202                 int next_block=0;
2203                 int pb_bits_count, pb2_bits_count, tex_pb_bits_count;
2204
2205                 copy_context_before_encode(&backup_s, s, -1);
2206                 backup_s.pb= s->pb;
2207                 best_s.data_partitioning= s->data_partitioning;
2208                 best_s.partitioned_frame= s->partitioned_frame;
2209                 if(s->data_partitioning){
2210                     backup_s.pb2= s->pb2;
2211                     backup_s.tex_pb= s->tex_pb;
2212                 }
2213
2214                 if(mb_type&CANDIDATE_MB_TYPE_INTER){
2215                     s->mv_dir = MV_DIR_FORWARD;
2216                     s->mv_type = MV_TYPE_16X16;
2217                     s->mb_intra= 0;
2218                     s->mv[0][0][0] = s->p_mv_table[xy][0];
2219                     s->mv[0][0][1] = s->p_mv_table[xy][1];
2220                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER, pb, pb2, tex_pb,
2221                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2222                 }
2223                 if(mb_type&CANDIDATE_MB_TYPE_INTER_I){
2224                     s->mv_dir = MV_DIR_FORWARD;
2225                     s->mv_type = MV_TYPE_FIELD;
2226                     s->mb_intra= 0;
2227                     for(i=0; i<2; i++){
2228                         j= s->field_select[0][i] = s->p_field_select_table[i][xy];
2229                         s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
2230                         s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
2231                     }
2232                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER_I, pb, pb2, tex_pb,
2233                                  &dmin, &next_block, 0, 0);
2234                 }
2235                 if(mb_type&CANDIDATE_MB_TYPE_SKIPPED){
2236                     s->mv_dir = MV_DIR_FORWARD;
2237                     s->mv_type = MV_TYPE_16X16;
2238                     s->mb_intra= 0;
2239                     s->mv[0][0][0] = 0;
2240                     s->mv[0][0][1] = 0;
2241                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_SKIPPED, pb, pb2, tex_pb,
2242                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2243                 }
2244                 if(mb_type&CANDIDATE_MB_TYPE_INTER4V){
2245                     s->mv_dir = MV_DIR_FORWARD;
2246                     s->mv_type = MV_TYPE_8X8;
2247                     s->mb_intra= 0;
2248                     for(i=0; i<4; i++){
2249                         s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
2250                         s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
2251                     }
2252                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER4V, pb, pb2, tex_pb,
2253                                  &dmin, &next_block, 0, 0);
2254                 }
2255                 if(mb_type&CANDIDATE_MB_TYPE_FORWARD){
2256                     s->mv_dir = MV_DIR_FORWARD;
2257                     s->mv_type = MV_TYPE_16X16;
2258                     s->mb_intra= 0;
2259                     s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
2260                     s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2261                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD, pb, pb2, tex_pb,
2262                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2263                 }
2264                 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD){
2265                     s->mv_dir = MV_DIR_BACKWARD;
2266                     s->mv_type = MV_TYPE_16X16;
2267                     s->mb_intra= 0;
2268                     s->mv[1][0][0] = s->b_back_mv_table[xy][0];
2269                     s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2270                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD, pb, pb2, tex_pb,
2271                                  &dmin, &next_block, s->mv[1][0][0], s->mv[1][0][1]);
2272                 }
2273                 if(mb_type&CANDIDATE_MB_TYPE_BIDIR){
2274                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2275                     s->mv_type = MV_TYPE_16X16;
2276                     s->mb_intra= 0;
2277                     s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
2278                     s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
2279                     s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
2280                     s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2281                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR, pb, pb2, tex_pb,
2282                                  &dmin, &next_block, 0, 0);
2283                 }
2284                 if(mb_type&CANDIDATE_MB_TYPE_FORWARD_I){
2285                     s->mv_dir = MV_DIR_FORWARD;
2286                     s->mv_type = MV_TYPE_FIELD;
2287                     s->mb_intra= 0;
2288                     for(i=0; i<2; i++){
2289                         j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
2290                         s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
2291                         s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
2292                     }
2293                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD_I, pb, pb2, tex_pb,
2294                                  &dmin, &next_block, 0, 0);
2295                 }
2296                 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD_I){
2297                     s->mv_dir = MV_DIR_BACKWARD;
2298                     s->mv_type = MV_TYPE_FIELD;
2299                     s->mb_intra= 0;
2300                     for(i=0; i<2; i++){
2301                         j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
2302                         s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
2303                         s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
2304                     }
2305                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD_I, pb, pb2, tex_pb,
2306                                  &dmin, &next_block, 0, 0);
2307                 }
2308                 if(mb_type&CANDIDATE_MB_TYPE_BIDIR_I){
2309                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2310                     s->mv_type = MV_TYPE_FIELD;
2311                     s->mb_intra= 0;
2312                     for(dir=0; dir<2; dir++){
2313                         for(i=0; i<2; i++){
2314                             j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
2315                             s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
2316                             s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
2317                         }
2318                     }
2319                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR_I, pb, pb2, tex_pb,
2320                                  &dmin, &next_block, 0, 0);
2321                 }
2322                 if(mb_type&CANDIDATE_MB_TYPE_INTRA){
2323                     s->mv_dir = 0;
2324                     s->mv_type = MV_TYPE_16X16;
2325                     s->mb_intra= 1;
2326                     s->mv[0][0][0] = 0;
2327                     s->mv[0][0][1] = 0;
2328                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTRA, pb, pb2, tex_pb,
2329                                  &dmin, &next_block, 0, 0);
2330                     if(s->h263_pred || s->h263_aic){
2331                         if(best_s.mb_intra)
2332                             s->mbintra_table[mb_x + mb_y*s->mb_stride]=1;
2333                         else
2334                             ff_clean_intra_table_entries(s); //old mode?
2335                     }
2336                 }
2337
2338                 if((s->flags & CODEC_FLAG_QP_RD) && dmin < INT_MAX){
2339                     if(best_s.mv_type==MV_TYPE_16X16){ //FIXME move 4mv after QPRD
2340                         const int last_qp= backup_s.qscale;
2341                         int qpi, qp, dc[6];
2342                         DCTELEM ac[6][16];
2343                         const int mvdir= (best_s.mv_dir&MV_DIR_BACKWARD) ? 1 : 0;
2344                         static const int dquant_tab[4]={-1,1,-2,2};
2345
2346                         assert(backup_s.dquant == 0);
2347
2348                         //FIXME intra
2349                         s->mv_dir= best_s.mv_dir;
2350                         s->mv_type = MV_TYPE_16X16;
2351                         s->mb_intra= best_s.mb_intra;
2352                         s->mv[0][0][0] = best_s.mv[0][0][0];
2353                         s->mv[0][0][1] = best_s.mv[0][0][1];
2354                         s->mv[1][0][0] = best_s.mv[1][0][0];
2355                         s->mv[1][0][1] = best_s.mv[1][0][1];
2356
2357                         qpi = s->pict_type == FF_B_TYPE ? 2 : 0;
2358                         for(; qpi<4; qpi++){
2359                             int dquant= dquant_tab[qpi];
2360                             qp= last_qp + dquant;
2361                             if(qp < s->avctx->qmin || qp > s->avctx->qmax)
2362                                 continue;
2363                             backup_s.dquant= dquant;
2364                             if(s->mb_intra && s->dc_val[0]){
2365                                 for(i=0; i<6; i++){
2366                                     dc[i]= s->dc_val[0][ s->block_index[i] ];
2367                                     memcpy(ac[i], s->ac_val[0][s->block_index[i]], sizeof(DCTELEM)*16);
2368                                 }
2369                             }
2370
2371                             encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
2372                                          &dmin, &next_block, s->mv[mvdir][0][0], s->mv[mvdir][0][1]);
2373                             if(best_s.qscale != qp){
2374                                 if(s->mb_intra && s->dc_val[0]){
2375                                     for(i=0; i<6; i++){
2376                                         s->dc_val[0][ s->block_index[i] ]= dc[i];
2377                                         memcpy(s->ac_val[0][s->block_index[i]], ac[i], sizeof(DCTELEM)*16);
2378                                     }
2379                                 }
2380                             }
2381                         }
2382                     }
2383                 }
2384                 if(ENABLE_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT){
2385                     int mx= s->b_direct_mv_table[xy][0];
2386                     int my= s->b_direct_mv_table[xy][1];
2387
2388                     backup_s.dquant = 0;
2389                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2390                     s->mb_intra= 0;
2391                     ff_mpeg4_set_direct_mv(s, mx, my);
2392                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
2393                                  &dmin, &next_block, mx, my);
2394                 }
2395                 if(ENABLE_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT0){
2396                     backup_s.dquant = 0;
2397                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2398                     s->mb_intra= 0;
2399                     ff_mpeg4_set_direct_mv(s, 0, 0);
2400                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
2401                                  &dmin, &next_block, 0, 0);
2402                 }
2403                 if(!best_s.mb_intra && s->flags2&CODEC_FLAG2_SKIP_RD){
2404                     int coded=0;
2405                     for(i=0; i<6; i++)
2406                         coded |= s->block_last_index[i];
2407                     if(coded){
2408                         int mx,my;
2409                         memcpy(s->mv, best_s.mv, sizeof(s->mv));
2410                         if(ENABLE_MPEG4_ENCODER && best_s.mv_dir & MV_DIRECT){
2411                             mx=my=0; //FIXME find the one we actually used
2412                             ff_mpeg4_set_direct_mv(s, mx, my);
2413                         }else if(best_s.mv_dir&MV_DIR_BACKWARD){
2414                             mx= s->mv[1][0][0];
2415                             my= s->mv[1][0][1];
2416                         }else{
2417                             mx= s->mv[0][0][0];
2418                             my= s->mv[0][0][1];
2419                         }
2420
2421                         s->mv_dir= best_s.mv_dir;
2422                         s->mv_type = best_s.mv_type;
2423                         s->mb_intra= 0;
2424 /*                        s->mv[0][0][0] = best_s.mv[0][0][0];
2425                         s->mv[0][0][1] = best_s.mv[0][0][1];
2426                         s->mv[1][0][0] = best_s.mv[1][0][0];
2427                         s->mv[1][0][1] = best_s.mv[1][0][1];*/
2428                         backup_s.dquant= 0;
2429                         s->skipdct=1;
2430                         encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
2431                                         &dmin, &next_block, mx, my);
2432                         s->skipdct=0;
2433                     }
2434                 }
2435
2436                 s->current_picture.qscale_table[xy]= best_s.qscale;
2437
2438                 copy_context_after_encode(s, &best_s, -1);
2439
2440                 pb_bits_count= put_bits_count(&s->pb);
2441                 flush_put_bits(&s->pb);
2442                 ff_copy_bits(&backup_s.pb, bit_buf[next_block^1], pb_bits_count);
2443                 s->pb= backup_s.pb;
2444
2445                 if(s->data_partitioning){
2446                     pb2_bits_count= put_bits_count(&s->pb2);
2447                     flush_put_bits(&s->pb2);
2448                     ff_copy_bits(&backup_s.pb2, bit_buf2[next_block^1], pb2_bits_count);
2449                     s->pb2= backup_s.pb2;
2450
2451                     tex_pb_bits_count= put_bits_count(&s->tex_pb);
2452                     flush_put_bits(&s->tex_pb);
2453                     ff_copy_bits(&backup_s.tex_pb, bit_buf_tex[next_block^1], tex_pb_bits_count);
2454                     s->tex_pb= backup_s.tex_pb;
2455                 }
2456                 s->last_bits= put_bits_count(&s->pb);
2457
2458                 if (ENABLE_ANY_H263_ENCODER &&
2459                     s->out_format == FMT_H263 && s->pict_type!=FF_B_TYPE)
2460                     ff_h263_update_motion_val(s);
2461
2462                 if(next_block==0){ //FIXME 16 vs linesize16
2463                     s->dsp.put_pixels_tab[0][0](s->dest[0], s->rd_scratchpad                     , s->linesize  ,16);
2464                     s->dsp.put_pixels_tab[1][0](s->dest[1], s->rd_scratchpad + 16*s->linesize    , s->uvlinesize, 8);
2465                     s->dsp.put_pixels_tab[1][0](s->dest[2], s->rd_scratchpad + 16*s->linesize + 8, s->uvlinesize, 8);
2466                 }
2467
2468                 if(s->avctx->mb_decision == FF_MB_DECISION_BITS)
2469                     MPV_decode_mb(s, s->block);
2470             } else {
2471                 int motion_x = 0, motion_y = 0;
2472                 s->mv_type=MV_TYPE_16X16;
2473                 // only one MB-Type possible
2474
2475                 switch(mb_type){
2476                 case CANDIDATE_MB_TYPE_INTRA:
2477                     s->mv_dir = 0;
2478                     s->mb_intra= 1;
2479                     motion_x= s->mv[0][0][0] = 0;
2480                     motion_y= s->mv[0][0][1] = 0;
2481                     break;
2482                 case CANDIDATE_MB_TYPE_INTER:
2483                     s->mv_dir = MV_DIR_FORWARD;
2484                     s->mb_intra= 0;
2485                     motion_x= s->mv[0][0][0] = s->p_mv_table[xy][0];
2486                     motion_y= s->mv[0][0][1] = s->p_mv_table[xy][1];
2487                     break;
2488                 case CANDIDATE_MB_TYPE_INTER_I:
2489                     s->mv_dir = MV_DIR_FORWARD;
2490                     s->mv_type = MV_TYPE_FIELD;
2491                     s->mb_intra= 0;
2492                     for(i=0; i<2; i++){
2493                         j= s->field_select[0][i] = s->p_field_select_table[i][xy];
2494                         s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
2495                         s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
2496                     }
2497                     break;
2498                 case CANDIDATE_MB_TYPE_INTER4V:
2499                     s->mv_dir = MV_DIR_FORWARD;
2500                     s->mv_type = MV_TYPE_8X8;
2501                     s->mb_intra= 0;
2502                     for(i=0; i<4; i++){
2503                         s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
2504                         s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
2505                     }
2506                     break;
2507                 case CANDIDATE_MB_TYPE_DIRECT:
2508                     if (ENABLE_MPEG4_ENCODER) {
2509                         s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
2510                         s->mb_intra= 0;
2511                         motion_x=s->b_direct_mv_table[xy][0];
2512                         motion_y=s->b_direct_mv_table[xy][1];
2513                         ff_mpeg4_set_direct_mv(s, motion_x, motion_y);
2514                     }
2515                     break;
2516                 case CANDIDATE_MB_TYPE_DIRECT0:
2517                     if (ENABLE_MPEG4_ENCODER) {
2518                         s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
2519                         s->mb_intra= 0;
2520                         ff_mpeg4_set_direct_mv(s, 0, 0);
2521                     }
2522                     break;
2523                 case CANDIDATE_MB_TYPE_BIDIR:
2524                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2525                     s->mb_intra= 0;
2526                     s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
2527                     s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
2528                     s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
2529                     s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2530                     break;
2531                 case CANDIDATE_MB_TYPE_BACKWARD:
2532                     s->mv_dir = MV_DIR_BACKWARD;
2533                     s->mb_intra= 0;
2534                     motion_x= s->mv[1][0][0] = s->b_back_mv_table[xy][0];
2535                     motion_y= s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2536                     break;
2537                 case CANDIDATE_MB_TYPE_FORWARD:
2538                     s->mv_dir = MV_DIR_FORWARD;
2539                     s->mb_intra= 0;
2540                     motion_x= s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
2541                     motion_y= s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2542 //                    printf(" %d %d ", motion_x, motion_y);
2543                     break;
2544                 case CANDIDATE_MB_TYPE_FORWARD_I:
2545                     s->mv_dir = MV_DIR_FORWARD;
2546                     s->mv_type = MV_TYPE_FIELD;
2547                     s->mb_intra= 0;
2548                     for(i=0; i<2; i++){
2549                         j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
2550                         s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
2551                         s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
2552                     }
2553                     break;
2554                 case CANDIDATE_MB_TYPE_BACKWARD_I:
2555                     s->mv_dir = MV_DIR_BACKWARD;
2556                     s->mv_type = MV_TYPE_FIELD;
2557                     s->mb_intra= 0;
2558                     for(i=0; i<2; i++){
2559                         j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
2560                         s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
2561                         s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
2562                     }
2563                     break;
2564                 case CANDIDATE_MB_TYPE_BIDIR_I:
2565                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2566                     s->mv_type = MV_TYPE_FIELD;
2567                     s->mb_intra= 0;
2568                     for(dir=0; dir<2; dir++){
2569                         for(i=0; i<2; i++){
2570                             j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
2571                             s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
2572                             s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
2573                         }
2574                     }
2575                     break;
2576                 default:
2577                     av_log(s->avctx, AV_LOG_ERROR, "illegal MB type\n");
2578                 }
2579
2580                 encode_mb(s, motion_x, motion_y);
2581
2582                 // RAL: Update last macroblock type
2583                 s->last_mv_dir = s->mv_dir;
2584
2585                 if (ENABLE_ANY_H263_ENCODER &&
2586                     s->out_format == FMT_H263 && s->pict_type!=FF_B_TYPE)
2587                     ff_h263_update_motion_val(s);
2588
2589                 MPV_decode_mb(s, s->block);
2590             }
2591
2592             /* clean the MV table in IPS frames for direct mode in B frames */
2593             if(s->mb_intra /* && I,P,S_TYPE */){
2594                 s->p_mv_table[xy][0]=0;
2595                 s->p_mv_table[xy][1]=0;
2596             }
2597
2598             if(s->flags&CODEC_FLAG_PSNR){
2599                 int w= 16;
2600                 int h= 16;
2601
2602                 if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
2603                 if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
2604
2605                 s->current_picture.error[0] += sse(
2606                     s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16,
2607                     s->dest[0], w, h, s->linesize);
2608                 s->current_picture.error[1] += sse(
2609                     s, s->new_picture.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,
2610                     s->dest[1], w>>1, h>>1, s->uvlinesize);
2611                 s->current_picture.error[2] += sse(
2612                     s, s->new_picture    .data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,
2613                     s->dest[2], w>>1, h>>1, s->uvlinesize);
2614             }
2615             if(s->loop_filter){
2616                 if(ENABLE_ANY_H263_ENCODER && s->out_format == FMT_H263)
2617                     ff_h263_loop_filter(s);
2618             }
2619 //printf("MB %d %d bits\n", s->mb_x+s->mb_y*s->mb_stride, put_bits_count(&s->pb));
2620         }
2621     }
2622
2623     //not beautiful here but we must write it before flushing so it has to be here
2624     if (ENABLE_MSMPEG4_ENCODER && s->msmpeg4_version && s->msmpeg4_version<4 && s->pict_type == FF_I_TYPE)
2625         msmpeg4_encode_ext_header(s);
2626
2627     write_slice_end(s);
2628
2629     /* Send the last GOB if RTP */
2630     if (s->avctx->rtp_callback) {
2631         int number_mb = (mb_y - s->resync_mb_y)*s->mb_width - s->resync_mb_x;
2632         pdif = pbBufPtr(&s->pb) - s->ptr_lastgob;
2633         /* Call the RTP callback to send the last GOB */
2634         emms_c();
2635         s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, pdif, number_mb);
2636     }
2637
2638     return 0;
2639 }
2640
2641 #define MERGE(field) dst->field += src->field; src->field=0
2642 static void merge_context_after_me(MpegEncContext *dst, MpegEncContext *src){
2643     MERGE(me.scene_change_score);
2644     MERGE(me.mc_mb_var_sum_temp);
2645     MERGE(me.mb_var_sum_temp);
2646 }
2647
2648 static void merge_context_after_encode(MpegEncContext *dst, MpegEncContext *src){
2649     int i;
2650
2651     MERGE(dct_count[0]); //note, the other dct vars are not part of the context
2652     MERGE(dct_count[1]);
2653     MERGE(mv_bits);
2654     MERGE(i_tex_bits);
2655     MERGE(p_tex_bits);
2656     MERGE(i_count);
2657     MERGE(f_count);
2658     MERGE(b_count);
2659     MERGE(skip_count);
2660     MERGE(misc_bits);
2661     MERGE(error_count);
2662     MERGE(padding_bug_score);
2663     MERGE(current_picture.error[0]);
2664     MERGE(current_picture.error[1]);
2665     MERGE(current_picture.error[2]);
2666
2667     if(dst->avctx->noise_reduction){
2668         for(i=0; i<64; i++){
2669             MERGE(dct_error_sum[0][i]);
2670             MERGE(dct_error_sum[1][i]);
2671         }
2672     }
2673
2674     assert(put_bits_count(&src->pb) % 8 ==0);
2675     assert(put_bits_count(&dst->pb) % 8 ==0);
2676     ff_copy_bits(&dst->pb, src->pb.buf, put_bits_count(&src->pb));
2677     flush_put_bits(&dst->pb);
2678 }
2679
2680 static int estimate_qp(MpegEncContext *s, int dry_run){
2681     if (s->next_lambda){
2682         s->current_picture_ptr->quality=
2683         s->current_picture.quality = s->next_lambda;
2684         if(!dry_run) s->next_lambda= 0;
2685     } else if (!s->fixed_qscale) {
2686         s->current_picture_ptr->quality=
2687         s->current_picture.quality = ff_rate_estimate_qscale(s, dry_run);
2688         if (s->current_picture.quality < 0)
2689             return -1;
2690     }
2691
2692     if(s->adaptive_quant){
2693         switch(s->codec_id){
2694         case CODEC_ID_MPEG4:
2695             if (ENABLE_MPEG4_ENCODER)
2696                 ff_clean_mpeg4_qscales(s);
2697             break;
2698         case CODEC_ID_H263:
2699         case CODEC_ID_H263P:
2700         case CODEC_ID_FLV1:
2701             if (ENABLE_H263_ENCODER||ENABLE_H263P_ENCODER||ENABLE_FLV_ENCODER)
2702                 ff_clean_h263_qscales(s);
2703             break;
2704         }
2705
2706         s->lambda= s->lambda_table[0];
2707         //FIXME broken
2708     }else
2709         s->lambda= s->current_picture.quality;
2710 //printf("%d %d\n", s->avctx->global_quality, s->current_picture.quality);
2711     update_qscale(s);
2712     return 0;
2713 }
2714
2715 /* must be called before writing the header */
2716 static void set_frame_distances(MpegEncContext * s){
2717     assert(s->current_picture_ptr->pts != AV_NOPTS_VALUE);
2718     s->time= s->current_picture_ptr->pts*s->avctx->time_base.num;
2719
2720     if(s->pict_type==FF_B_TYPE){
2721         s->pb_time= s->pp_time - (s->last_non_b_time - s->time);
2722         assert(s->pb_time > 0 && s->pb_time < s->pp_time);
2723     }else{
2724         s->pp_time= s->time - s->last_non_b_time;
2725         s->last_non_b_time= s->time;
2726         assert(s->picture_number==0 || s->pp_time > 0);
2727     }
2728 }
2729
2730 static int encode_picture(MpegEncContext *s, int picture_number)
2731 {
2732     int i;
2733     int bits;
2734
2735     s->picture_number = picture_number;
2736
2737     /* Reset the average MB variance */
2738     s->me.mb_var_sum_temp    =
2739     s->me.mc_mb_var_sum_temp = 0;
2740
2741     /* we need to initialize some time vars before we can encode b-frames */
2742     // RAL: Condition added for MPEG1VIDEO
2743     if (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO || (s->h263_pred && !s->h263_msmpeg4))
2744         set_frame_distances(s);
2745     if(ENABLE_MPEG4_ENCODER && s->codec_id == CODEC_ID_MPEG4)
2746         ff_set_mpeg4_time(s);
2747
2748     s->me.scene_change_score=0;
2749
2750 //    s->lambda= s->current_picture_ptr->quality; //FIXME qscale / ... stuff for ME rate distortion
2751
2752     if(s->pict_type==FF_I_TYPE){
2753         if(s->msmpeg4_version >= 3) s->no_rounding=1;
2754         else                        s->no_rounding=0;
2755     }else if(s->pict_type!=FF_B_TYPE){
2756         if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
2757             s->no_rounding ^= 1;
2758     }
2759
2760     if(s->flags & CODEC_FLAG_PASS2){
2761         if (estimate_qp(s,1) < 0)
2762             return -1;
2763         ff_get_2pass_fcode(s);
2764     }else if(!(s->flags & CODEC_FLAG_QSCALE)){
2765         if(s->pict_type==FF_B_TYPE)
2766             s->lambda= s->last_lambda_for[s->pict_type];
2767         else
2768             s->lambda= s->last_lambda_for[s->last_non_b_pict_type];
2769         update_qscale(s);
2770     }
2771
2772     s->mb_intra=0; //for the rate distortion & bit compare functions
2773     for(i=1; i<s->avctx->thread_count; i++){
2774         ff_update_duplicate_context(s->thread_context[i], s);
2775     }
2776
2777     ff_init_me(s);
2778
2779     /* Estimate motion for every MB */
2780     if(s->pict_type != FF_I_TYPE){
2781         s->lambda = (s->lambda * s->avctx->me_penalty_compensation + 128)>>8;
2782         s->lambda2= (s->lambda2* (int64_t)s->avctx->me_penalty_compensation + 128)>>8;
2783         if(s->pict_type != FF_B_TYPE && s->avctx->me_threshold==0){
2784             if((s->avctx->pre_me && s->last_non_b_pict_type==FF_I_TYPE) || s->avctx->pre_me==2){
2785                 s->avctx->execute(s->avctx, pre_estimate_motion_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count);
2786             }
2787         }
2788
2789         s->avctx->execute(s->avctx, estimate_motion_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count);
2790     }else /* if(s->pict_type == FF_I_TYPE) */{
2791         /* I-Frame */
2792         for(i=0; i<s->mb_stride*s->mb_height; i++)
2793             s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
2794
2795         if(!s->fixed_qscale){
2796             /* finding spatial complexity for I-frame rate control */
2797             s->avctx->execute(s->avctx, mb_var_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count);
2798         }
2799     }
2800     for(i=1; i<s->avctx->thread_count; i++){
2801         merge_context_after_me(s, s->thread_context[i]);
2802     }
2803     s->current_picture.mc_mb_var_sum= s->current_picture_ptr->mc_mb_var_sum= s->me.mc_mb_var_sum_temp;
2804     s->current_picture.   mb_var_sum= s->current_picture_ptr->   mb_var_sum= s->me.   mb_var_sum_temp;
2805     emms_c();
2806
2807     if(s->me.scene_change_score > s->avctx->scenechange_threshold && s->pict_type == FF_P_TYPE){
2808         s->pict_type= FF_I_TYPE;
2809         for(i=0; i<s->mb_stride*s->mb_height; i++)
2810             s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
2811 //printf("Scene change detected, encoding as I Frame %d %d\n", s->current_picture.mb_var_sum, s->current_picture.mc_mb_var_sum);
2812     }
2813
2814     if(!s->umvplus){
2815         if(s->pict_type==FF_P_TYPE || s->pict_type==FF_S_TYPE) {
2816             s->f_code= ff_get_best_fcode(s, s->p_mv_table, CANDIDATE_MB_TYPE_INTER);
2817
2818             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2819                 int a,b;
2820                 a= ff_get_best_fcode(s, s->p_field_mv_table[0][0], CANDIDATE_MB_TYPE_INTER_I); //FIXME field_select
2821                 b= ff_get_best_fcode(s, s->p_field_mv_table[1][1], CANDIDATE_MB_TYPE_INTER_I);
2822                 s->f_code= FFMAX3(s->f_code, a, b);
2823             }
2824
2825             ff_fix_long_p_mvs(s);
2826             ff_fix_long_mvs(s, NULL, 0, s->p_mv_table, s->f_code, CANDIDATE_MB_TYPE_INTER, 0);
2827             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2828                 int j;
2829                 for(i=0; i<2; i++){
2830                     for(j=0; j<2; j++)
2831                         ff_fix_long_mvs(s, s->p_field_select_table[i], j,
2832                                         s->p_field_mv_table[i][j], s->f_code, CANDIDATE_MB_TYPE_INTER_I, 0);
2833                 }
2834             }
2835         }
2836
2837         if(s->pict_type==FF_B_TYPE){
2838             int a, b;
2839
2840             a = ff_get_best_fcode(s, s->b_forw_mv_table, CANDIDATE_MB_TYPE_FORWARD);
2841             b = ff_get_best_fcode(s, s->b_bidir_forw_mv_table, CANDIDATE_MB_TYPE_BIDIR);
2842             s->f_code = FFMAX(a, b);
2843
2844             a = ff_get_best_fcode(s, s->b_back_mv_table, CANDIDATE_MB_TYPE_BACKWARD);
2845             b = ff_get_best_fcode(s, s->b_bidir_back_mv_table, CANDIDATE_MB_TYPE_BIDIR);
2846             s->b_code = FFMAX(a, b);
2847
2848             ff_fix_long_mvs(s, NULL, 0, s->b_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_FORWARD, 1);
2849             ff_fix_long_mvs(s, NULL, 0, s->b_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BACKWARD, 1);
2850             ff_fix_long_mvs(s, NULL, 0, s->b_bidir_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_BIDIR, 1);
2851             ff_fix_long_mvs(s, NULL, 0, s->b_bidir_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BIDIR, 1);
2852             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2853                 int dir, j;
2854                 for(dir=0; dir<2; dir++){
2855                     for(i=0; i<2; i++){
2856                         for(j=0; j<2; j++){
2857                             int type= dir ? (CANDIDATE_MB_TYPE_BACKWARD_I|CANDIDATE_MB_TYPE_BIDIR_I)
2858                                           : (CANDIDATE_MB_TYPE_FORWARD_I |CANDIDATE_MB_TYPE_BIDIR_I);
2859                             ff_fix_long_mvs(s, s->b_field_select_table[dir][i], j,
2860                                             s->b_field_mv_table[dir][i][j], dir ? s->b_code : s->f_code, type, 1);
2861                         }
2862                     }
2863                 }
2864             }
2865         }
2866     }
2867
2868     if (estimate_qp(s, 0) < 0)
2869         return -1;
2870
2871     if(s->qscale < 3 && s->max_qcoeff<=128 && s->pict_type==FF_I_TYPE && !(s->flags & CODEC_FLAG_QSCALE))
2872         s->qscale= 3; //reduce clipping problems
2873
2874     if (s->out_format == FMT_MJPEG) {
2875         /* for mjpeg, we do include qscale in the matrix */
2876         s->intra_matrix[0] = ff_mpeg1_default_intra_matrix[0];
2877         for(i=1;i<64;i++){
2878             int j= s->dsp.idct_permutation[i];
2879
2880             s->intra_matrix[j] = av_clip_uint8((ff_mpeg1_default_intra_matrix[i] * s->qscale) >> 3);
2881         }
2882         ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
2883                        s->intra_matrix, s->intra_quant_bias, 8, 8, 1);
2884         s->qscale= 8;
2885     }
2886
2887     //FIXME var duplication
2888     s->current_picture_ptr->key_frame=
2889     s->current_picture.key_frame= s->pict_type == FF_I_TYPE; //FIXME pic_ptr
2890     s->current_picture_ptr->pict_type=
2891     s->current_picture.pict_type= s->pict_type;
2892
2893     if(s->current_picture.key_frame)
2894         s->picture_in_gop_number=0;
2895
2896     s->last_bits= put_bits_count(&s->pb);
2897     switch(s->out_format) {
2898     case FMT_MJPEG:
2899         if (ENABLE_MJPEG_ENCODER)
2900             ff_mjpeg_encode_picture_header(s);
2901         break;
2902     case FMT_H261:
2903         if (ENABLE_H261_ENCODER)
2904             ff_h261_encode_picture_header(s, picture_number);
2905         break;
2906     case FMT_H263:
2907         if (ENABLE_WMV2_ENCODER && s->codec_id == CODEC_ID_WMV2)
2908             ff_wmv2_encode_picture_header(s, picture_number);
2909         else if (ENABLE_MSMPEG4_ENCODER && s->h263_msmpeg4)
2910             msmpeg4_encode_picture_header(s, picture_number);
2911         else if (ENABLE_MPEG4_ENCODER && s->h263_pred)
2912             mpeg4_encode_picture_header(s, picture_number);
2913         else if (ENABLE_RV10_ENCODER && s->codec_id == CODEC_ID_RV10)
2914             rv10_encode_picture_header(s, picture_number);
2915         else if (ENABLE_RV20_ENCODER && s->codec_id == CODEC_ID_RV20)
2916             rv20_encode_picture_header(s, picture_number);
2917         else if (ENABLE_FLV_ENCODER && s->codec_id == CODEC_ID_FLV1)
2918             ff_flv_encode_picture_header(s, picture_number);
2919         else if (ENABLE_ANY_H263_ENCODER)
2920             h263_encode_picture_header(s, picture_number);
2921         break;
2922     case FMT_MPEG1:
2923         if (ENABLE_MPEG1VIDEO_ENCODER || ENABLE_MPEG2VIDEO_ENCODER)
2924             mpeg1_encode_picture_header(s, picture_number);
2925         break;
2926     case FMT_H264:
2927         break;
2928     default:
2929         assert(0);
2930     }
2931     bits= put_bits_count(&s->pb);
2932     s->header_bits= bits - s->last_bits;
2933
2934     for(i=1; i<s->avctx->thread_count; i++){
2935         update_duplicate_context_after_me(s->thread_context[i], s);
2936     }
2937     s->avctx->execute(s->avctx, encode_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count);
2938     for(i=1; i<s->avctx->thread_count; i++){
2939         merge_context_after_encode(s, s->thread_context[i]);
2940     }
2941     emms_c();
2942     return 0;
2943 }
2944
2945 void  denoise_dct_c(MpegEncContext *s, DCTELEM *block){
2946     const int intra= s->mb_intra;
2947     int i;
2948
2949     s->dct_count[intra]++;
2950
2951     for(i=0; i<64; i++){
2952         int level= block[i];
2953
2954         if(level){
2955             if(level>0){
2956                 s->dct_error_sum[intra][i] += level;
2957                 level -= s->dct_offset[intra][i];
2958                 if(level<0) level=0;
2959             }else{
2960                 s->dct_error_sum[intra][i] -= level;
2961                 level += s->dct_offset[intra][i];
2962                 if(level>0) level=0;
2963             }
2964             block[i]= level;
2965         }
2966     }
2967 }
2968
2969 int dct_quantize_trellis_c(MpegEncContext *s,
2970                         DCTELEM *block, int n,
2971                         int qscale, int *overflow){
2972     const int *qmat;
2973     const uint8_t *scantable= s->intra_scantable.scantable;
2974     const uint8_t *perm_scantable= s->intra_scantable.permutated;
2975     int max=0;
2976     unsigned int threshold1, threshold2;
2977     int bias=0;
2978     int run_tab[65];
2979     int level_tab[65];
2980     int score_tab[65];
2981     int survivor[65];
2982     int survivor_count;
2983     int last_run=0;
2984     int last_level=0;
2985     int last_score= 0;
2986     int last_i;
2987     int coeff[2][64];
2988     int coeff_count[64];
2989     int qmul, qadd, start_i, last_non_zero, i, dc;
2990     const int esc_length= s->ac_esc_length;
2991     uint8_t * length;
2992     uint8_t * last_length;
2993     const int lambda= s->lambda2 >> (FF_LAMBDA_SHIFT - 6);
2994
2995     s->dsp.fdct (block);
2996
2997     if(s->dct_error_sum)
2998         s->denoise_dct(s, block);
2999     qmul= qscale*16;
3000     qadd= ((qscale-1)|1)*8;
3001
3002     if (s->mb_intra) {
3003         int q;
3004         if (!s->h263_aic) {
3005             if (n < 4)
3006                 q = s->y_dc_scale;
3007             else
3008                 q = s->c_dc_scale;
3009             q = q << 3;
3010         } else{
3011             /* For AIC we skip quant/dequant of INTRADC */
3012             q = 1 << 3;
3013             qadd=0;
3014         }
3015
3016         /* note: block[0] is assumed to be positive */
3017         block[0] = (block[0] + (q >> 1)) / q;
3018         start_i = 1;
3019         last_non_zero = 0;
3020         qmat = s->q_intra_matrix[qscale];
3021         if(s->mpeg_quant || s->out_format == FMT_MPEG1)
3022             bias= 1<<(QMAT_SHIFT-1);
3023         length     = s->intra_ac_vlc_length;
3024         last_length= s->intra_ac_vlc_last_length;
3025     } else {
3026         start_i = 0;
3027         last_non_zero = -1;
3028         qmat = s->q_inter_matrix[qscale];
3029         length     = s->inter_ac_vlc_length;
3030         last_length= s->inter_ac_vlc_last_length;
3031     }
3032     last_i= start_i;
3033
3034     threshold1= (1<<QMAT_SHIFT) - bias - 1;
3035     threshold2= (threshold1<<1);
3036
3037     for(i=63; i>=start_i; i--) {
3038         const int j = scantable[i];
3039         int level = block[j] * qmat[j];
3040
3041         if(((unsigned)(level+threshold1))>threshold2){
3042             last_non_zero = i;
3043             break;
3044         }
3045     }
3046
3047     for(i=start_i; i<=last_non_zero; i++) {
3048         const int j = scantable[i];
3049         int level = block[j] * qmat[j];
3050
3051 //        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
3052 //           || bias-level >= (1<<(QMAT_SHIFT - 3))){
3053         if(((unsigned)(level+threshold1))>threshold2){
3054             if(level>0){
3055                 level= (bias + level)>>QMAT_SHIFT;
3056                 coeff[0][i]= level;
3057                 coeff[1][i]= level-1;
3058 //                coeff[2][k]= level-2;
3059             }else{
3060                 level= (bias - level)>>QMAT_SHIFT;
3061                 coeff[0][i]= -level;
3062                 coeff[1][i]= -level+1;
3063 //                coeff[2][k]= -level+2;
3064             }
3065             coeff_count[i]= FFMIN(level, 2);
3066             assert(coeff_count[i]);
3067             max |=level;
3068         }else{
3069             coeff[0][i]= (level>>31)|1;
3070             coeff_count[i]= 1;
3071         }
3072     }
3073
3074     *overflow= s->max_qcoeff < max; //overflow might have happened
3075
3076     if(last_non_zero < start_i){
3077         memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
3078         return last_non_zero;
3079     }
3080
3081     score_tab[start_i]= 0;
3082     survivor[0]= start_i;
3083     survivor_count= 1;
3084
3085     for(i=start_i; i<=last_non_zero; i++){
3086         int level_index, j, zero_distortion;
3087         int dct_coeff= FFABS(block[ scantable[i] ]);
3088         int best_score=256*256*256*120;
3089
3090         if (   s->dsp.fdct == fdct_ifast
3091 #ifndef FAAN_POSTSCALE
3092             || s->dsp.fdct == ff_faandct
3093 #endif
3094            )
3095             dct_coeff= (dct_coeff*inv_aanscales[ scantable[i] ]) >> 12;
3096         zero_distortion= dct_coeff*dct_coeff;
3097
3098         for(level_index=0; level_index < coeff_count[i]; level_index++){
3099             int distortion;
3100             int level= coeff[level_index][i];
3101             const int alevel= FFABS(level);
3102             int unquant_coeff;
3103
3104             assert(level);
3105
3106             if(s->out_format == FMT_H263){
3107                 unquant_coeff= alevel*qmul + qadd;
3108             }else{ //MPEG1
3109                 j= s->dsp.idct_permutation[ scantable[i] ]; //FIXME optimize
3110                 if(s->mb_intra){
3111                         unquant_coeff = (int)(  alevel  * qscale * s->intra_matrix[j]) >> 3;
3112                         unquant_coeff =   (unquant_coeff - 1) | 1;
3113                 }else{
3114                         unquant_coeff = (((  alevel  << 1) + 1) * qscale * ((int) s->inter_matrix[j])) >> 4;
3115                         unquant_coeff =   (unquant_coeff - 1) | 1;
3116                 }
3117                 unquant_coeff<<= 3;
3118             }
3119
3120             distortion= (unquant_coeff - dct_coeff) * (unquant_coeff - dct_coeff) - zero_distortion;
3121             level+=64;
3122             if((level&(~127)) == 0){
3123                 for(j=survivor_count-1; j>=0; j--){
3124                     int run= i - survivor[j];
3125                     int score= distortion + length[UNI_AC_ENC_INDEX(run, level)]*lambda;
3126                     score += score_tab[i-run];
3127
3128                     if(score < best_score){
3129                         best_score= score;
3130                         run_tab[i+1]= run;
3131                         level_tab[i+1]= level-64;
3132                     }
3133                 }
3134
3135                 if(s->out_format == FMT_H263){
3136                     for(j=survivor_count-1; j>=0; j--){
3137                         int run= i - survivor[j];
3138                         int score= distortion + last_length[UNI_AC_ENC_INDEX(run, level)]*lambda;
3139                         score += score_tab[i-run];
3140                         if(score < last_score){
3141                             last_score= score;
3142                             last_run= run;
3143                             last_level= level-64;
3144                             last_i= i+1;
3145                         }
3146                     }
3147                 }
3148             }else{
3149                 distortion += esc_length*lambda;
3150                 for(j=survivor_count-1; j>=0; j--){
3151                     int run= i - survivor[j];
3152                     int score= distortion + score_tab[i-run];
3153
3154                     if(score < best_score){
3155                         best_score= score;
3156                         run_tab[i+1]= run;
3157                         level_tab[i+1]= level-64;
3158                     }
3159                 }
3160
3161                 if(s->out_format == FMT_H263){
3162                   for(j=survivor_count-1; j>=0; j--){
3163                         int run= i - survivor[j];
3164                         int score= distortion + score_tab[i-run];
3165                         if(score < last_score){
3166                             last_score= score;
3167                             last_run= run;
3168                             last_level= level-64;
3169                             last_i= i+1;
3170                         }
3171                     }
3172                 }
3173             }
3174         }
3175
3176         score_tab[i+1]= best_score;
3177
3178         //Note: there is a vlc code in mpeg4 which is 1 bit shorter then another one with a shorter run and the same level
3179         if(last_non_zero <= 27){
3180             for(; survivor_count; survivor_count--){
3181                 if(score_tab[ survivor[survivor_count-1] ] <= best_score)
3182                     break;
3183             }
3184         }else{
3185             for(; survivor_count; survivor_count--){
3186                 if(score_tab[ survivor[survivor_count-1] ] <= best_score + lambda)
3187                     break;
3188             }
3189         }
3190
3191         survivor[ survivor_count++ ]= i+1;
3192     }
3193
3194     if(s->out_format != FMT_H263){
3195         last_score= 256*256*256*120;
3196         for(i= survivor[0]; i<=last_non_zero + 1; i++){
3197             int score= score_tab[i];
3198             if(i) score += lambda*2; //FIXME exacter?
3199
3200             if(score < last_score){
3201                 last_score= score;
3202                 last_i= i;
3203                 last_level= level_tab[i];
3204                 last_run= run_tab[i];
3205             }
3206         }
3207     }
3208
3209     s->coded_score[n] = last_score;
3210
3211     dc= FFABS(block[0]);
3212     last_non_zero= last_i - 1;
3213     memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
3214
3215     if(last_non_zero < start_i)
3216         return last_non_zero;
3217
3218     if(last_non_zero == 0 && start_i == 0){
3219         int best_level= 0;
3220         int best_score= dc * dc;
3221
3222         for(i=0; i<coeff_count[0]; i++){
3223             int level= coeff[i][0];
3224             int alevel= FFABS(level);
3225             int unquant_coeff, score, distortion;
3226
3227             if(s->out_format == FMT_H263){
3228                     unquant_coeff= (alevel*qmul + qadd)>>3;
3229             }else{ //MPEG1
3230                     unquant_coeff = (((  alevel  << 1) + 1) * qscale * ((int) s->inter_matrix[0])) >> 4;
3231                     unquant_coeff =   (unquant_coeff - 1) | 1;
3232             }
3233             unquant_coeff = (unquant_coeff + 4) >> 3;
3234             unquant_coeff<<= 3 + 3;
3235
3236             distortion= (unquant_coeff - dc) * (unquant_coeff - dc);
3237             level+=64;
3238             if((level&(~127)) == 0) score= distortion + last_length[UNI_AC_ENC_INDEX(0, level)]*lambda;
3239             else                    score= distortion + esc_length*lambda;
3240
3241             if(score < best_score){
3242                 best_score= score;
3243                 best_level= level - 64;
3244             }
3245         }
3246         block[0]= best_level;
3247         s->coded_score[n] = best_score - dc*dc;
3248         if(best_level == 0) return -1;
3249         else                return last_non_zero;
3250     }
3251
3252     i= last_i;
3253     assert(last_level);
3254
3255     block[ perm_scantable[last_non_zero] ]= last_level;
3256     i -= last_run + 1;
3257
3258     for(; i>start_i; i -= run_tab[i] + 1){
3259         block[ perm_scantable[i-1] ]= level_tab[i];
3260     }
3261
3262     return last_non_zero;
3263 }
3264
3265 //#define REFINE_STATS 1
3266 static int16_t basis[64][64];
3267
3268 static void build_basis(uint8_t *perm){
3269     int i, j, x, y;
3270     emms_c();
3271     for(i=0; i<8; i++){
3272         for(j=0; j<8; j++){
3273             for(y=0; y<8; y++){
3274                 for(x=0; x<8; x++){
3275                     double s= 0.25*(1<<BASIS_SHIFT);
3276                     int index= 8*i + j;
3277                     int perm_index= perm[index];
3278                     if(i==0) s*= sqrt(0.5);
3279                     if(j==0) s*= sqrt(0.5);
3280                     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)));
3281                 }
3282             }
3283         }
3284     }
3285 }
3286
3287 static int dct_quantize_refine(MpegEncContext *s, //FIXME breaks denoise?
3288                         DCTELEM *block, int16_t *weight, DCTELEM *orig,
3289                         int n, int qscale){
3290     int16_t rem[64];
3291     DECLARE_ALIGNED_16(DCTELEM, d1[64]);
3292     const int *qmat;
3293     const uint8_t *scantable= s->intra_scantable.scantable;
3294     const uint8_t *perm_scantable= s->intra_scantable.permutated;
3295 //    unsigned int threshold1, threshold2;
3296 //    int bias=0;
3297     int run_tab[65];
3298     int prev_run=0;
3299     int prev_level=0;
3300     int qmul, qadd, start_i, last_non_zero, i, dc;
3301     uint8_t * length;
3302     uint8_t * last_length;
3303     int lambda;
3304     int rle_index, run, q = 1, sum; //q is only used when s->mb_intra is true
3305 #ifdef REFINE_STATS
3306 static int count=0;
3307 static int after_last=0;
3308 static int to_zero=0;
3309 static int from_zero=0;
3310 static int raise=0;
3311 static int lower=0;
3312 static int messed_sign=0;
3313 #endif
3314
3315     if(basis[0][0] == 0)
3316         build_basis(s->dsp.idct_permutation);
3317
3318     qmul= qscale*2;
3319     qadd= (qscale-1)|1;
3320     if (s->mb_intra) {
3321         if (!s->h263_aic) {
3322             if (n < 4)
3323                 q = s->y_dc_scale;
3324             else
3325                 q = s->c_dc_scale;
3326         } else{
3327             /* For AIC we skip quant/dequant of INTRADC */
3328             q = 1;
3329             qadd=0;
3330         }
3331         q <<= RECON_SHIFT-3;
3332         /* note: block[0] is assumed to be positive */
3333         dc= block[0]*q;
3334 //        block[0] = (block[0] + (q >> 1)) / q;
3335         start_i = 1;
3336         qmat = s->q_intra_matrix[qscale];
3337 //        if(s->mpeg_quant || s->out_format == FMT_MPEG1)
3338 //            bias= 1<<(QMAT_SHIFT-1);
3339         length     = s->intra_ac_vlc_length;
3340         last_length= s->intra_ac_vlc_last_length;
3341     } else {
3342         dc= 0;
3343         start_i = 0;
3344         qmat = s->q_inter_matrix[qscale];
3345         length     = s->inter_ac_vlc_length;
3346         last_length= s->inter_ac_vlc_last_length;
3347     }
3348     last_non_zero = s->block_last_index[n];
3349
3350 #ifdef REFINE_STATS
3351 {START_TIMER
3352 #endif
3353     dc += (1<<(RECON_SHIFT-1));
3354     for(i=0; i<64; i++){
3355         rem[i]= dc - (orig[i]<<RECON_SHIFT); //FIXME  use orig dirrectly instead of copying to rem[]
3356     }
3357 #ifdef REFINE_STATS
3358 STOP_TIMER("memset rem[]")}
3359 #endif
3360     sum=0;
3361     for(i=0; i<64; i++){
3362         int one= 36;
3363         int qns=4;
3364         int w;
3365
3366         w= FFABS(weight[i]) + qns*one;
3367         w= 15 + (48*qns*one + w/2)/w; // 16 .. 63
3368
3369         weight[i] = w;
3370 //        w=weight[i] = (63*qns + (w/2)) / w;
3371
3372         assert(w>0);
3373         assert(w<(1<<6));
3374         sum += w*w;
3375     }
3376     lambda= sum*(uint64_t)s->lambda2 >> (FF_LAMBDA_SHIFT - 6 + 6 + 6 + 6);
3377 #ifdef REFINE_STATS
3378 {START_TIMER
3379 #endif
3380     run=0;
3381     rle_index=0;
3382     for(i=start_i; i<=last_non_zero; i++){
3383         int j= perm_scantable[i];
3384         const int level= block[j];
3385         int coeff;
3386
3387         if(level){
3388             if(level<0) coeff= qmul*level - qadd;
3389             else        coeff= qmul*level + qadd;
3390             run_tab[rle_index++]=run;
3391             run=0;
3392
3393             s->dsp.add_8x8basis(rem, basis[j], coeff);
3394         }else{
3395             run++;
3396         }
3397     }
3398 #ifdef REFINE_STATS
3399 if(last_non_zero>0){
3400 STOP_TIMER("init rem[]")
3401 }
3402 }
3403
3404 {START_TIMER
3405 #endif
3406     for(;;){
3407         int best_score=s->dsp.try_8x8basis(rem, weight, basis[0], 0);
3408         int best_coeff=0;
3409         int best_change=0;
3410         int run2, best_unquant_change=0, analyze_gradient;
3411 #ifdef REFINE_STATS
3412 {START_TIMER
3413 #endif
3414         analyze_gradient = last_non_zero > 2 || s->avctx->quantizer_noise_shaping >= 3;
3415
3416         if(analyze_gradient){
3417 #ifdef REFINE_STATS
3418 {START_TIMER
3419 #endif
3420             for(i=0; i<64; i++){
3421                 int w= weight[i];
3422
3423                 d1[i] = (rem[i]*w*w + (1<<(RECON_SHIFT+12-1)))>>(RECON_SHIFT+12);
3424             }
3425 #ifdef REFINE_STATS
3426 STOP_TIMER("rem*w*w")}
3427 {START_TIMER
3428 #endif
3429             s->dsp.fdct(d1);
3430 #ifdef REFINE_STATS
3431 STOP_TIMER("dct")}
3432 #endif
3433         }
3434
3435         if(start_i){
3436             const int level= block[0];
3437             int change, old_coeff;
3438
3439             assert(s->mb_intra);
3440
3441             old_coeff= q*level;
3442
3443             for(change=-1; change<=1; change+=2){
3444                 int new_level= level + change;
3445                 int score, new_coeff;
3446
3447                 new_coeff= q*new_level;
3448                 if(new_coeff >= 2048 || new_coeff < 0)
3449                     continue;
3450
3451                 score= s->dsp.try_8x8basis(rem, weight, basis[0], new_coeff - old_coeff);
3452                 if(score<best_score){
3453                     best_score= score;
3454                     best_coeff= 0;
3455                     best_change= change;
3456                     best_unquant_change= new_coeff - old_coeff;
3457                 }
3458             }
3459         }
3460
3461         run=0;
3462         rle_index=0;
3463         run2= run_tab[rle_index++];
3464         prev_level=0;
3465         prev_run=0;
3466
3467         for(i=start_i; i<64; i++){
3468             int j= perm_scantable[i];
3469             const int level= block[j];
3470             int change, old_coeff;
3471
3472             if(s->avctx->quantizer_noise_shaping < 3 && i > last_non_zero + 1)
3473                 break;
3474
3475             if(level){
3476                 if(level<0) old_coeff= qmul*level - qadd;
3477                 else        old_coeff= qmul*level + qadd;
3478                 run2= run_tab[rle_index++]; //FIXME ! maybe after last
3479             }else{
3480                 old_coeff=0;
3481                 run2--;
3482                 assert(run2>=0 || i >= last_non_zero );
3483             }
3484
3485             for(change=-1; change<=1; change+=2){
3486                 int new_level= level + change;
3487                 int score, new_coeff, unquant_change;
3488
3489                 score=0;
3490                 if(s->avctx->quantizer_noise_shaping < 2 && FFABS(new_level) > FFABS(level))
3491                    continue;
3492
3493                 if(new_level){
3494                     if(new_level<0) new_coeff= qmul*new_level - qadd;
3495                     else            new_coeff= qmul*new_level + qadd;
3496                     if(new_coeff >= 2048 || new_coeff <= -2048)
3497                         continue;
3498                     //FIXME check for overflow
3499
3500                     if(level){
3501                         if(level < 63 && level > -63){
3502                             if(i < last_non_zero)
3503                                 score +=   length[UNI_AC_ENC_INDEX(run, new_level+64)]
3504                                          - length[UNI_AC_ENC_INDEX(run, level+64)];
3505                             else
3506                                 score +=   last_length[UNI_AC_ENC_INDEX(run, new_level+64)]
3507                                          - last_length[UNI_AC_ENC_INDEX(run, level+64)];
3508                         }
3509                     }else{
3510                         assert(FFABS(new_level)==1);
3511
3512                         if(analyze_gradient){
3513                             int g= d1[ scantable[i] ];
3514                             if(g && (g^new_level) >= 0)
3515                                 continue;
3516                         }
3517
3518                         if(i < last_non_zero){
3519                             int next_i= i + run2 + 1;
3520                             int next_level= block[ perm_scantable[next_i] ] + 64;
3521
3522                             if(next_level&(~127))
3523                                 next_level= 0;
3524
3525                             if(next_i < last_non_zero)
3526                                 score +=   length[UNI_AC_ENC_INDEX(run, 65)]
3527                                          + length[UNI_AC_ENC_INDEX(run2, next_level)]
3528                                          - length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)];
3529                             else
3530                                 score +=  length[UNI_AC_ENC_INDEX(run, 65)]
3531                                         + last_length[UNI_AC_ENC_INDEX(run2, next_level)]
3532                                         - last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)];
3533                         }else{
3534                             score += last_length[UNI_AC_ENC_INDEX(run, 65)];
3535                             if(prev_level){
3536                                 score +=  length[UNI_AC_ENC_INDEX(prev_run, prev_level)]
3537                                         - last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)];
3538                             }
3539                         }
3540                     }
3541                 }else{
3542                     new_coeff=0;
3543                     assert(FFABS(level)==1);
3544
3545                     if(i < last_non_zero){
3546                         int next_i= i + run2 + 1;
3547                         int next_level= block[ perm_scantable[next_i] ] + 64;
3548
3549                         if(next_level&(~127))
3550                             next_level= 0;
3551
3552                         if(next_i < last_non_zero)
3553                             score +=   length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)]
3554                                      - length[UNI_AC_ENC_INDEX(run2, next_level)]
3555                                      - length[UNI_AC_ENC_INDEX(run, 65)];
3556                         else
3557                             score +=   last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)]
3558                                      - last_length[UNI_AC_ENC_INDEX(run2, next_level)]
3559                                      - length[UNI_AC_ENC_INDEX(run, 65)];
3560                     }else{
3561                         score += -last_length[UNI_AC_ENC_INDEX(run, 65)];
3562                         if(prev_level){
3563                             score +=  last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)]
3564                                     - length[UNI_AC_ENC_INDEX(prev_run, prev_level)];
3565                         }
3566                     }
3567                 }
3568
3569                 score *= lambda;
3570
3571                 unquant_change= new_coeff - old_coeff;
3572                 assert((score < 100*lambda && score > -100*lambda) || lambda==0);
3573
3574                 score+= s->dsp.try_8x8basis(rem, weight, basis[j], unquant_change);
3575                 if(score<best_score){
3576                     best_score= score;
3577                     best_coeff= i;
3578                     best_change= change;
3579                     best_unquant_change= unquant_change;
3580                 }
3581             }
3582             if(level){
3583                 prev_level= level + 64;
3584                 if(prev_level&(~127))
3585                     prev_level= 0;
3586                 prev_run= run;
3587                 run=0;
3588             }else{
3589                 run++;
3590             }
3591         }
3592 #ifdef REFINE_STATS
3593 STOP_TIMER("iterative step")}
3594 #endif
3595
3596         if(best_change){
3597             int j= perm_scantable[ best_coeff ];
3598
3599             block[j] += best_change;
3600
3601             if(best_coeff > last_non_zero){
3602                 last_non_zero= best_coeff;
3603                 assert(block[j]);
3604 #ifdef REFINE_STATS
3605 after_last++;
3606 #endif
3607             }else{
3608 #ifdef REFINE_STATS
3609 if(block[j]){
3610     if(block[j] - best_change){
3611         if(FFABS(block[j]) > FFABS(block[j] - best_change)){
3612             raise++;
3613         }else{
3614             lower++;
3615         }
3616     }else{
3617         from_zero++;
3618     }
3619 }else{
3620     to_zero++;
3621 }
3622 #endif
3623                 for(; last_non_zero>=start_i; last_non_zero--){
3624                     if(block[perm_scantable[last_non_zero]])
3625                         break;
3626                 }
3627             }
3628 #ifdef REFINE_STATS
3629 count++;
3630 if(256*256*256*64 % count == 0){
3631     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);
3632 }
3633 #endif
3634             run=0;
3635             rle_index=0;
3636             for(i=start_i; i<=last_non_zero; i++){
3637                 int j= perm_scantable[i];
3638                 const int level= block[j];
3639
3640                  if(level){
3641                      run_tab[rle_index++]=run;
3642                      run=0;
3643                  }else{
3644                      run++;
3645                  }
3646             }
3647
3648             s->dsp.add_8x8basis(rem, basis[j], best_unquant_change);
3649         }else{
3650             break;
3651         }
3652     }
3653 #ifdef REFINE_STATS
3654 if(last_non_zero>0){
3655 STOP_TIMER("iterative search")
3656 }
3657 }
3658 #endif
3659
3660     return last_non_zero;
3661 }
3662
3663 int dct_quantize_c(MpegEncContext *s,
3664                         DCTELEM *block, int n,
3665                         int qscale, int *overflow)
3666 {
3667     int i, j, level, last_non_zero, q, start_i;
3668     const int *qmat;
3669     const uint8_t *scantable= s->intra_scantable.scantable;
3670     int bias;
3671     int max=0;
3672     unsigned int threshold1, threshold2;
3673
3674     s->dsp.fdct (block);
3675
3676     if(s->dct_error_sum)
3677         s->denoise_dct(s, block);
3678
3679     if (s->mb_intra) {
3680         if (!s->h263_aic) {
3681             if (n < 4)
3682                 q = s->y_dc_scale;
3683             else
3684                 q = s->c_dc_scale;
3685             q = q << 3;
3686         } else
3687             /* For AIC we skip quant/dequant of INTRADC */
3688             q = 1 << 3;
3689
3690         /* note: block[0] is assumed to be positive */
3691         block[0] = (block[0] + (q >> 1)) / q;
3692         start_i = 1;
3693         last_non_zero = 0;
3694         qmat = s->q_intra_matrix[qscale];
3695         bias= s->intra_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT);
3696     } else {
3697         start_i = 0;
3698         last_non_zero = -1;
3699         qmat = s->q_inter_matrix[qscale];
3700         bias= s->inter_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT);
3701     }
3702     threshold1= (1<<QMAT_SHIFT) - bias - 1;
3703     threshold2= (threshold1<<1);
3704     for(i=63;i>=start_i;i--) {
3705         j = scantable[i];
3706         level = block[j] * qmat[j];
3707
3708         if(((unsigned)(level+threshold1))>threshold2){
3709             last_non_zero = i;
3710             break;
3711         }else{
3712             block[j]=0;
3713         }
3714     }
3715     for(i=start_i; i<=last_non_zero; i++) {
3716         j = scantable[i];
3717         level = block[j] * qmat[j];
3718
3719 //        if(   bias+level >= (1<<QMAT_SHIFT)
3720 //           || bias-level >= (1<<QMAT_SHIFT)){
3721         if(((unsigned)(level+threshold1))>threshold2){
3722             if(level>0){
3723                 level= (bias + level)>>QMAT_SHIFT;
3724                 block[j]= level;
3725             }else{
3726                 level= (bias - level)>>QMAT_SHIFT;
3727                 block[j]= -level;
3728             }
3729             max |=level;
3730         }else{
3731             block[j]=0;
3732         }
3733     }
3734     *overflow= s->max_qcoeff < max; //overflow might have happened
3735
3736     /* we need this permutation so that we correct the IDCT, we only permute the !=0 elements */
3737     if (s->dsp.idct_permutation_type != FF_NO_IDCT_PERM)
3738         ff_block_permute(block, s->dsp.idct_permutation, scantable, last_non_zero);
3739
3740     return last_non_zero;
3741 }
3742
3743 AVCodec h263_encoder = {
3744     "h263",
3745     CODEC_TYPE_VIDEO,
3746     CODEC_ID_H263,
3747     sizeof(MpegEncContext),
3748     MPV_encode_init,
3749     MPV_encode_picture,
3750     MPV_encode_end,
3751     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3752     .long_name= NULL_IF_CONFIG_SMALL("H.263"),
3753 };
3754
3755 AVCodec h263p_encoder = {
3756     "h263p",
3757     CODEC_TYPE_VIDEO,
3758     CODEC_ID_H263P,
3759     sizeof(MpegEncContext),
3760     MPV_encode_init,
3761     MPV_encode_picture,
3762     MPV_encode_end,
3763     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3764     .long_name= NULL_IF_CONFIG_SMALL("H.263+ / H.263 version 2"),
3765 };
3766
3767 AVCodec flv_encoder = {
3768     "flv",
3769     CODEC_TYPE_VIDEO,
3770     CODEC_ID_FLV1,
3771     sizeof(MpegEncContext),
3772     MPV_encode_init,
3773     MPV_encode_picture,
3774     MPV_encode_end,
3775     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3776     .long_name= NULL_IF_CONFIG_SMALL("Flash Video"),
3777 };
3778
3779 AVCodec rv10_encoder = {
3780     "rv10",
3781     CODEC_TYPE_VIDEO,
3782     CODEC_ID_RV10,
3783     sizeof(MpegEncContext),
3784     MPV_encode_init,
3785     MPV_encode_picture,
3786     MPV_encode_end,
3787     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3788     .long_name= NULL_IF_CONFIG_SMALL("RealVideo 1.0"),
3789 };
3790
3791 AVCodec rv20_encoder = {
3792     "rv20",
3793     CODEC_TYPE_VIDEO,
3794     CODEC_ID_RV20,
3795     sizeof(MpegEncContext),
3796     MPV_encode_init,
3797     MPV_encode_picture,
3798     MPV_encode_end,
3799     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3800     .long_name= NULL_IF_CONFIG_SMALL("RealVideo 2.0"),
3801 };
3802
3803 AVCodec mpeg4_encoder = {
3804     "mpeg4",
3805     CODEC_TYPE_VIDEO,
3806     CODEC_ID_MPEG4,
3807     sizeof(MpegEncContext),
3808     MPV_encode_init,
3809     MPV_encode_picture,
3810     MPV_encode_end,
3811     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3812     .capabilities= CODEC_CAP_DELAY,
3813     .long_name= NULL_IF_CONFIG_SMALL("MPEG-4 part 2"),
3814 };
3815
3816 AVCodec msmpeg4v1_encoder = {
3817     "msmpeg4v1",
3818     CODEC_TYPE_VIDEO,
3819     CODEC_ID_MSMPEG4V1,
3820     sizeof(MpegEncContext),
3821     MPV_encode_init,
3822     MPV_encode_picture,
3823     MPV_encode_end,
3824     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3825     .long_name= NULL_IF_CONFIG_SMALL("MPEG-4 part 2 Microsoft variant version 1"),
3826 };
3827
3828 AVCodec msmpeg4v2_encoder = {
3829     "msmpeg4v2",
3830     CODEC_TYPE_VIDEO,
3831     CODEC_ID_MSMPEG4V2,
3832     sizeof(MpegEncContext),
3833     MPV_encode_init,
3834     MPV_encode_picture,
3835     MPV_encode_end,
3836     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3837     .long_name= NULL_IF_CONFIG_SMALL("MPEG-4 part 2 Microsoft variant version 2"),
3838 };
3839
3840 AVCodec msmpeg4v3_encoder = {
3841     "msmpeg4",
3842     CODEC_TYPE_VIDEO,
3843     CODEC_ID_MSMPEG4V3,
3844     sizeof(MpegEncContext),
3845     MPV_encode_init,
3846     MPV_encode_picture,
3847     MPV_encode_end,
3848     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3849     .long_name= NULL_IF_CONFIG_SMALL("MPEG-4 part 2 Microsoft variant version 3"),
3850 };
3851
3852 AVCodec wmv1_encoder = {
3853     "wmv1",
3854     CODEC_TYPE_VIDEO,
3855     CODEC_ID_WMV1,
3856     sizeof(MpegEncContext),
3857     MPV_encode_init,
3858     MPV_encode_picture,
3859     MPV_encode_end,
3860     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3861     .long_name= NULL_IF_CONFIG_SMALL("Windows Media Video 7"),
3862 };