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){
 444             av_log(avctx, AV_LOG_ERROR, "low delay forcing is only available for mpeg2\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 chr_h= 16>>s->chroma_y_shift;
2016     int i, j;
2017     MpegEncContext best_s, backup_s;
2018     uint8_t bit_buf[2][MAX_MB_BYTES];
2019     uint8_t bit_buf2[2][MAX_MB_BYTES];
2020     uint8_t bit_buf_tex[2][MAX_MB_BYTES];
2021     PutBitContext pb[2], pb2[2], tex_pb[2];
2022 //printf("%d->%d\n", s->resync_mb_y, s->end_mb_y);
2023
2024     ff_check_alignment();
2025
2026     for(i=0; i<2; i++){
2027         init_put_bits(&pb    [i], bit_buf    [i], MAX_MB_BYTES);
2028         init_put_bits(&pb2   [i], bit_buf2   [i], MAX_MB_BYTES);
2029         init_put_bits(&tex_pb[i], bit_buf_tex[i], MAX_MB_BYTES);
2030     }
2031
2032     s->last_bits= put_bits_count(&s->pb);
2033     s->mv_bits=0;
2034     s->misc_bits=0;
2035     s->i_tex_bits=0;
2036     s->p_tex_bits=0;
2037     s->i_count=0;
2038     s->f_count=0;
2039     s->b_count=0;
2040     s->skip_count=0;
2041
2042     for(i=0; i<3; i++){
2043         /* init last dc values */
2044         /* note: quant matrix value (8) is implied here */
2045         s->last_dc[i] = 128 << s->intra_dc_precision;
2046
2047         s->current_picture.error[i] = 0;
2048     }
2049     s->mb_skip_run = 0;
2050     memset(s->last_mv, 0, sizeof(s->last_mv));
2051
2052     s->last_mv_dir = 0;
2053
2054     switch(s->codec_id){
2055     case CODEC_ID_H263:
2056     case CODEC_ID_H263P:
2057     case CODEC_ID_FLV1:
2058         if (ENABLE_H263_ENCODER || ENABLE_H263P_ENCODER || ENABLE_FLV_ENCODER)
2059             s->gob_index = ff_h263_get_gob_height(s);
2060         break;
2061     case CODEC_ID_MPEG4:
2062         if(ENABLE_MPEG4_ENCODER && s->partitioned_frame)
2063             ff_mpeg4_init_partitions(s);
2064         break;
2065     }
2066
2067     s->resync_mb_x=0;
2068     s->resync_mb_y=0;
2069     s->first_slice_line = 1;
2070     s->ptr_lastgob = s->pb.buf;
2071     for(mb_y= s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
2072 //    printf("row %d at %X\n", s->mb_y, (int)s);
2073         s->mb_x=0;
2074         s->mb_y= mb_y;
2075
2076         ff_set_qscale(s, s->qscale);
2077         ff_init_block_index(s);
2078
2079         for(mb_x=0; mb_x < s->mb_width; mb_x++) {
2080             int xy= mb_y*s->mb_stride + mb_x; // removed const, H261 needs to adjust this
2081             int mb_type= s->mb_type[xy];
2082 //            int d;
2083             int dmin= INT_MAX;
2084             int dir;
2085
2086             if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < MAX_MB_BYTES){
2087                 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
2088                 return -1;
2089             }
2090             if(s->data_partitioning){
2091                 if(   s->pb2   .buf_end - s->pb2   .buf - (put_bits_count(&s->    pb2)>>3) < MAX_MB_BYTES
2092                    || s->tex_pb.buf_end - s->tex_pb.buf - (put_bits_count(&s->tex_pb )>>3) < MAX_MB_BYTES){
2093                     av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
2094                     return -1;
2095                 }
2096             }
2097
2098             s->mb_x = mb_x;
2099             s->mb_y = mb_y;  // moved into loop, can get changed by H.261
2100             ff_update_block_index(s);
2101
2102             if(ENABLE_H261_ENCODER && s->codec_id == CODEC_ID_H261){
2103                 ff_h261_reorder_mb_index(s);
2104                 xy= s->mb_y*s->mb_stride + s->mb_x;
2105                 mb_type= s->mb_type[xy];
2106             }
2107
2108             /* write gob / video packet header  */
2109             if(s->rtp_mode){
2110                 int current_packet_size, is_gob_start;
2111
2112                 current_packet_size= ((put_bits_count(&s->pb)+7)>>3) - (s->ptr_lastgob - s->pb.buf);
2113
2114                 is_gob_start= s->avctx->rtp_payload_size && current_packet_size >= s->avctx->rtp_payload_size && mb_y + mb_x>0;
2115
2116                 if(s->start_mb_y == mb_y && mb_y > 0 && mb_x==0) is_gob_start=1;
2117
2118                 switch(s->codec_id){
2119                 case CODEC_ID_H263:
2120                 case CODEC_ID_H263P:
2121                     if(!s->h263_slice_structured)
2122                         if(s->mb_x || s->mb_y%s->gob_index) is_gob_start=0;
2123                     break;
2124                 case CODEC_ID_MPEG2VIDEO:
2125                     if(s->mb_x==0 && s->mb_y!=0) is_gob_start=1;
2126                 case CODEC_ID_MPEG1VIDEO:
2127                     if(s->mb_skip_run) is_gob_start=0;
2128                     break;
2129                 }
2130
2131                 if(is_gob_start){
2132                     if(s->start_mb_y != mb_y || mb_x!=0){
2133                         write_slice_end(s);
2134
2135                         if(ENABLE_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4 && s->partitioned_frame){
2136                             ff_mpeg4_init_partitions(s);
2137                         }
2138                     }
2139
2140                     assert((put_bits_count(&s->pb)&7) == 0);
2141                     current_packet_size= pbBufPtr(&s->pb) - s->ptr_lastgob;
2142
2143                     if(s->avctx->error_rate && s->resync_mb_x + s->resync_mb_y > 0){
2144                         int r= put_bits_count(&s->pb)/8 + s->picture_number + 16 + s->mb_x + s->mb_y;
2145                         int d= 100 / s->avctx->error_rate;
2146                         if(r % d == 0){
2147                             current_packet_size=0;
2148 #ifndef ALT_BITSTREAM_WRITER
2149                             s->pb.buf_ptr= s->ptr_lastgob;
2150 #endif
2151                             assert(pbBufPtr(&s->pb) == s->ptr_lastgob);
2152                         }
2153                     }
2154
2155                     if (s->avctx->rtp_callback){
2156                         int number_mb = (mb_y - s->resync_mb_y)*s->mb_width + mb_x - s->resync_mb_x;
2157                         s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, current_packet_size, number_mb);
2158                     }
2159
2160                     switch(s->codec_id){
2161                     case CODEC_ID_MPEG4:
2162                         if (ENABLE_MPEG4_ENCODER) {
2163                             ff_mpeg4_encode_video_packet_header(s);
2164                             ff_mpeg4_clean_buffers(s);
2165                         }
2166                     break;
2167                     case CODEC_ID_MPEG1VIDEO:
2168                     case CODEC_ID_MPEG2VIDEO:
2169                         if (ENABLE_MPEG1VIDEO_ENCODER || ENABLE_MPEG2VIDEO_ENCODER) {
2170                             ff_mpeg1_encode_slice_header(s);
2171                             ff_mpeg1_clean_buffers(s);
2172                         }
2173                     break;
2174                     case CODEC_ID_H263:
2175                     case CODEC_ID_H263P:
2176                         if (ENABLE_H263_ENCODER || ENABLE_H263P_ENCODER)
2177                             h263_encode_gob_header(s, mb_y);
2178                     break;
2179                     }
2180
2181                     if(s->flags&CODEC_FLAG_PASS1){
2182                         int bits= put_bits_count(&s->pb);
2183                         s->misc_bits+= bits - s->last_bits;
2184                         s->last_bits= bits;
2185                     }
2186
2187                     s->ptr_lastgob += current_packet_size;
2188                     s->first_slice_line=1;
2189                     s->resync_mb_x=mb_x;
2190                     s->resync_mb_y=mb_y;
2191                 }
2192             }
2193
2194             if(  (s->resync_mb_x   == s->mb_x)
2195                && s->resync_mb_y+1 == s->mb_y){
2196                 s->first_slice_line=0;
2197             }
2198
2199             s->mb_skipped=0;
2200             s->dquant=0; //only for QP_RD
2201
2202             if(mb_type & (mb_type-1) || (s->flags & CODEC_FLAG_QP_RD)){ // more than 1 MB type possible or CODEC_FLAG_QP_RD
2203                 int next_block=0;
2204                 int pb_bits_count, pb2_bits_count, tex_pb_bits_count;
2205
2206                 copy_context_before_encode(&backup_s, s, -1);
2207                 backup_s.pb= s->pb;
2208                 best_s.data_partitioning= s->data_partitioning;
2209                 best_s.partitioned_frame= s->partitioned_frame;
2210                 if(s->data_partitioning){
2211                     backup_s.pb2= s->pb2;
2212                     backup_s.tex_pb= s->tex_pb;
2213                 }
2214
2215                 if(mb_type&CANDIDATE_MB_TYPE_INTER){
2216                     s->mv_dir = MV_DIR_FORWARD;
2217                     s->mv_type = MV_TYPE_16X16;
2218                     s->mb_intra= 0;
2219                     s->mv[0][0][0] = s->p_mv_table[xy][0];
2220                     s->mv[0][0][1] = s->p_mv_table[xy][1];
2221                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER, pb, pb2, tex_pb,
2222                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2223                 }
2224                 if(mb_type&CANDIDATE_MB_TYPE_INTER_I){
2225                     s->mv_dir = MV_DIR_FORWARD;
2226                     s->mv_type = MV_TYPE_FIELD;
2227                     s->mb_intra= 0;
2228                     for(i=0; i<2; i++){
2229                         j= s->field_select[0][i] = s->p_field_select_table[i][xy];
2230                         s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
2231                         s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
2232                     }
2233                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER_I, pb, pb2, tex_pb,
2234                                  &dmin, &next_block, 0, 0);
2235                 }
2236                 if(mb_type&CANDIDATE_MB_TYPE_SKIPPED){
2237                     s->mv_dir = MV_DIR_FORWARD;
2238                     s->mv_type = MV_TYPE_16X16;
2239                     s->mb_intra= 0;
2240                     s->mv[0][0][0] = 0;
2241                     s->mv[0][0][1] = 0;
2242                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_SKIPPED, pb, pb2, tex_pb,
2243                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2244                 }
2245                 if(mb_type&CANDIDATE_MB_TYPE_INTER4V){
2246                     s->mv_dir = MV_DIR_FORWARD;
2247                     s->mv_type = MV_TYPE_8X8;
2248                     s->mb_intra= 0;
2249                     for(i=0; i<4; i++){
2250                         s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
2251                         s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
2252                     }
2253                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER4V, pb, pb2, tex_pb,
2254                                  &dmin, &next_block, 0, 0);
2255                 }
2256                 if(mb_type&CANDIDATE_MB_TYPE_FORWARD){
2257                     s->mv_dir = MV_DIR_FORWARD;
2258                     s->mv_type = MV_TYPE_16X16;
2259                     s->mb_intra= 0;
2260                     s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
2261                     s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2262                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD, pb, pb2, tex_pb,
2263                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2264                 }
2265                 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD){
2266                     s->mv_dir = MV_DIR_BACKWARD;
2267                     s->mv_type = MV_TYPE_16X16;
2268                     s->mb_intra= 0;
2269                     s->mv[1][0][0] = s->b_back_mv_table[xy][0];
2270                     s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2271                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD, pb, pb2, tex_pb,
2272                                  &dmin, &next_block, s->mv[1][0][0], s->mv[1][0][1]);
2273                 }
2274                 if(mb_type&CANDIDATE_MB_TYPE_BIDIR){
2275                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2276                     s->mv_type = MV_TYPE_16X16;
2277                     s->mb_intra= 0;
2278                     s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
2279                     s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
2280                     s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
2281                     s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2282                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR, pb, pb2, tex_pb,
2283                                  &dmin, &next_block, 0, 0);
2284                 }
2285                 if(mb_type&CANDIDATE_MB_TYPE_FORWARD_I){
2286                     s->mv_dir = MV_DIR_FORWARD;
2287                     s->mv_type = MV_TYPE_FIELD;
2288                     s->mb_intra= 0;
2289                     for(i=0; i<2; i++){
2290                         j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
2291                         s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
2292                         s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
2293                     }
2294                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD_I, pb, pb2, tex_pb,
2295                                  &dmin, &next_block, 0, 0);
2296                 }
2297                 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD_I){
2298                     s->mv_dir = MV_DIR_BACKWARD;
2299                     s->mv_type = MV_TYPE_FIELD;
2300                     s->mb_intra= 0;
2301                     for(i=0; i<2; i++){
2302                         j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
2303                         s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
2304                         s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
2305                     }
2306                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD_I, pb, pb2, tex_pb,
2307                                  &dmin, &next_block, 0, 0);
2308                 }
2309                 if(mb_type&CANDIDATE_MB_TYPE_BIDIR_I){
2310                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2311                     s->mv_type = MV_TYPE_FIELD;
2312                     s->mb_intra= 0;
2313                     for(dir=0; dir<2; dir++){
2314                         for(i=0; i<2; i++){
2315                             j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
2316                             s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
2317                             s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
2318                         }
2319                     }
2320                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR_I, pb, pb2, tex_pb,
2321                                  &dmin, &next_block, 0, 0);
2322                 }
2323                 if(mb_type&CANDIDATE_MB_TYPE_INTRA){
2324                     s->mv_dir = 0;
2325                     s->mv_type = MV_TYPE_16X16;
2326                     s->mb_intra= 1;
2327                     s->mv[0][0][0] = 0;
2328                     s->mv[0][0][1] = 0;
2329                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTRA, pb, pb2, tex_pb,
2330                                  &dmin, &next_block, 0, 0);
2331                     if(s->h263_pred || s->h263_aic){
2332                         if(best_s.mb_intra)
2333                             s->mbintra_table[mb_x + mb_y*s->mb_stride]=1;
2334                         else
2335                             ff_clean_intra_table_entries(s); //old mode?
2336                     }
2337                 }
2338
2339                 if((s->flags & CODEC_FLAG_QP_RD) && dmin < INT_MAX){
2340                     if(best_s.mv_type==MV_TYPE_16X16){ //FIXME move 4mv after QPRD
2341                         const int last_qp= backup_s.qscale;
2342                         int qpi, qp, dc[6];
2343                         DCTELEM ac[6][16];
2344                         const int mvdir= (best_s.mv_dir&MV_DIR_BACKWARD) ? 1 : 0;
2345                         static const int dquant_tab[4]={-1,1,-2,2};
2346
2347                         assert(backup_s.dquant == 0);
2348
2349                         //FIXME intra
2350                         s->mv_dir= best_s.mv_dir;
2351                         s->mv_type = MV_TYPE_16X16;
2352                         s->mb_intra= best_s.mb_intra;
2353                         s->mv[0][0][0] = best_s.mv[0][0][0];
2354                         s->mv[0][0][1] = best_s.mv[0][0][1];
2355                         s->mv[1][0][0] = best_s.mv[1][0][0];
2356                         s->mv[1][0][1] = best_s.mv[1][0][1];
2357
2358                         qpi = s->pict_type == FF_B_TYPE ? 2 : 0;
2359                         for(; qpi<4; qpi++){
2360                             int dquant= dquant_tab[qpi];
2361                             qp= last_qp + dquant;
2362                             if(qp < s->avctx->qmin || qp > s->avctx->qmax)
2363                                 continue;
2364                             backup_s.dquant= dquant;
2365                             if(s->mb_intra && s->dc_val[0]){
2366                                 for(i=0; i<6; i++){
2367                                     dc[i]= s->dc_val[0][ s->block_index[i] ];
2368                                     memcpy(ac[i], s->ac_val[0][s->block_index[i]], sizeof(DCTELEM)*16);
2369                                 }
2370                             }
2371
2372                             encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
2373                                          &dmin, &next_block, s->mv[mvdir][0][0], s->mv[mvdir][0][1]);
2374                             if(best_s.qscale != qp){
2375                                 if(s->mb_intra && s->dc_val[0]){
2376                                     for(i=0; i<6; i++){
2377                                         s->dc_val[0][ s->block_index[i] ]= dc[i];
2378                                         memcpy(s->ac_val[0][s->block_index[i]], ac[i], sizeof(DCTELEM)*16);
2379                                     }
2380                                 }
2381                             }
2382                         }
2383                     }
2384                 }
2385                 if(ENABLE_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT){
2386                     int mx= s->b_direct_mv_table[xy][0];
2387                     int my= s->b_direct_mv_table[xy][1];
2388
2389                     backup_s.dquant = 0;
2390                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2391                     s->mb_intra= 0;
2392                     ff_mpeg4_set_direct_mv(s, mx, my);
2393                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
2394                                  &dmin, &next_block, mx, my);
2395                 }
2396                 if(ENABLE_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT0){
2397                     backup_s.dquant = 0;
2398                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2399                     s->mb_intra= 0;
2400                     ff_mpeg4_set_direct_mv(s, 0, 0);
2401                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
2402                                  &dmin, &next_block, 0, 0);
2403                 }
2404                 if(!best_s.mb_intra && s->flags2&CODEC_FLAG2_SKIP_RD){
2405                     int coded=0;
2406                     for(i=0; i<6; i++)
2407                         coded |= s->block_last_index[i];
2408                     if(coded){
2409                         int mx,my;
2410                         memcpy(s->mv, best_s.mv, sizeof(s->mv));
2411                         if(ENABLE_MPEG4_ENCODER && best_s.mv_dir & MV_DIRECT){
2412                             mx=my=0; //FIXME find the one we actually used
2413                             ff_mpeg4_set_direct_mv(s, mx, my);
2414                         }else if(best_s.mv_dir&MV_DIR_BACKWARD){
2415                             mx= s->mv[1][0][0];
2416                             my= s->mv[1][0][1];
2417                         }else{
2418                             mx= s->mv[0][0][0];
2419                             my= s->mv[0][0][1];
2420                         }
2421
2422                         s->mv_dir= best_s.mv_dir;
2423                         s->mv_type = best_s.mv_type;
2424                         s->mb_intra= 0;
2425 /*                        s->mv[0][0][0] = best_s.mv[0][0][0];
2426                         s->mv[0][0][1] = best_s.mv[0][0][1];
2427                         s->mv[1][0][0] = best_s.mv[1][0][0];
2428                         s->mv[1][0][1] = best_s.mv[1][0][1];*/
2429                         backup_s.dquant= 0;
2430                         s->skipdct=1;
2431                         encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
2432                                         &dmin, &next_block, mx, my);
2433                         s->skipdct=0;
2434                     }
2435                 }
2436
2437                 s->current_picture.qscale_table[xy]= best_s.qscale;
2438
2439                 copy_context_after_encode(s, &best_s, -1);
2440
2441                 pb_bits_count= put_bits_count(&s->pb);
2442                 flush_put_bits(&s->pb);
2443                 ff_copy_bits(&backup_s.pb, bit_buf[next_block^1], pb_bits_count);
2444                 s->pb= backup_s.pb;
2445
2446                 if(s->data_partitioning){
2447                     pb2_bits_count= put_bits_count(&s->pb2);
2448                     flush_put_bits(&s->pb2);
2449                     ff_copy_bits(&backup_s.pb2, bit_buf2[next_block^1], pb2_bits_count);
2450                     s->pb2= backup_s.pb2;
2451
2452                     tex_pb_bits_count= put_bits_count(&s->tex_pb);
2453                     flush_put_bits(&s->tex_pb);
2454                     ff_copy_bits(&backup_s.tex_pb, bit_buf_tex[next_block^1], tex_pb_bits_count);
2455                     s->tex_pb= backup_s.tex_pb;
2456                 }
2457                 s->last_bits= put_bits_count(&s->pb);
2458
2459                 if (ENABLE_ANY_H263_ENCODER &&
2460                     s->out_format == FMT_H263 && s->pict_type!=FF_B_TYPE)
2461                     ff_h263_update_motion_val(s);
2462
2463                 if(next_block==0){ //FIXME 16 vs linesize16
2464                     s->dsp.put_pixels_tab[0][0](s->dest[0], s->rd_scratchpad                     , s->linesize  ,16);
2465                     s->dsp.put_pixels_tab[1][0](s->dest[1], s->rd_scratchpad + 16*s->linesize    , s->uvlinesize, 8);
2466                     s->dsp.put_pixels_tab[1][0](s->dest[2], s->rd_scratchpad + 16*s->linesize + 8, s->uvlinesize, 8);
2467                 }
2468
2469                 if(s->avctx->mb_decision == FF_MB_DECISION_BITS)
2470                     MPV_decode_mb(s, s->block);
2471             } else {
2472                 int motion_x = 0, motion_y = 0;
2473                 s->mv_type=MV_TYPE_16X16;
2474                 // only one MB-Type possible
2475
2476                 switch(mb_type){
2477                 case CANDIDATE_MB_TYPE_INTRA:
2478                     s->mv_dir = 0;
2479                     s->mb_intra= 1;
2480                     motion_x= s->mv[0][0][0] = 0;
2481                     motion_y= s->mv[0][0][1] = 0;
2482                     break;
2483                 case CANDIDATE_MB_TYPE_INTER:
2484                     s->mv_dir = MV_DIR_FORWARD;
2485                     s->mb_intra= 0;
2486                     motion_x= s->mv[0][0][0] = s->p_mv_table[xy][0];
2487                     motion_y= s->mv[0][0][1] = s->p_mv_table[xy][1];
2488                     break;
2489                 case CANDIDATE_MB_TYPE_INTER_I:
2490                     s->mv_dir = MV_DIR_FORWARD;
2491                     s->mv_type = MV_TYPE_FIELD;
2492                     s->mb_intra= 0;
2493                     for(i=0; i<2; i++){
2494                         j= s->field_select[0][i] = s->p_field_select_table[i][xy];
2495                         s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
2496                         s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
2497                     }
2498                     break;
2499                 case CANDIDATE_MB_TYPE_INTER4V:
2500                     s->mv_dir = MV_DIR_FORWARD;
2501                     s->mv_type = MV_TYPE_8X8;
2502                     s->mb_intra= 0;
2503                     for(i=0; i<4; i++){
2504                         s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
2505                         s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
2506                     }
2507                     break;
2508                 case CANDIDATE_MB_TYPE_DIRECT:
2509                     if (ENABLE_MPEG4_ENCODER) {
2510                         s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
2511                         s->mb_intra= 0;
2512                         motion_x=s->b_direct_mv_table[xy][0];
2513                         motion_y=s->b_direct_mv_table[xy][1];
2514                         ff_mpeg4_set_direct_mv(s, motion_x, motion_y);
2515                     }
2516                     break;
2517                 case CANDIDATE_MB_TYPE_DIRECT0:
2518                     if (ENABLE_MPEG4_ENCODER) {
2519                         s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
2520                         s->mb_intra= 0;
2521                         ff_mpeg4_set_direct_mv(s, 0, 0);
2522                     }
2523                     break;
2524                 case CANDIDATE_MB_TYPE_BIDIR:
2525                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2526                     s->mb_intra= 0;
2527                     s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
2528                     s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
2529                     s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
2530                     s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2531                     break;
2532                 case CANDIDATE_MB_TYPE_BACKWARD:
2533                     s->mv_dir = MV_DIR_BACKWARD;
2534                     s->mb_intra= 0;
2535                     motion_x= s->mv[1][0][0] = s->b_back_mv_table[xy][0];
2536                     motion_y= s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2537                     break;
2538                 case CANDIDATE_MB_TYPE_FORWARD:
2539                     s->mv_dir = MV_DIR_FORWARD;
2540                     s->mb_intra= 0;
2541                     motion_x= s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
2542                     motion_y= s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2543 //                    printf(" %d %d ", motion_x, motion_y);
2544                     break;
2545                 case CANDIDATE_MB_TYPE_FORWARD_I:
2546                     s->mv_dir = MV_DIR_FORWARD;
2547                     s->mv_type = MV_TYPE_FIELD;
2548                     s->mb_intra= 0;
2549                     for(i=0; i<2; i++){
2550                         j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
2551                         s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
2552                         s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
2553                     }
2554                     break;
2555                 case CANDIDATE_MB_TYPE_BACKWARD_I:
2556                     s->mv_dir = MV_DIR_BACKWARD;
2557                     s->mv_type = MV_TYPE_FIELD;
2558                     s->mb_intra= 0;
2559                     for(i=0; i<2; i++){
2560                         j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
2561                         s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
2562                         s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
2563                     }
2564                     break;
2565                 case CANDIDATE_MB_TYPE_BIDIR_I:
2566                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2567                     s->mv_type = MV_TYPE_FIELD;
2568                     s->mb_intra= 0;
2569                     for(dir=0; dir<2; dir++){
2570                         for(i=0; i<2; i++){
2571                             j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
2572                             s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
2573                             s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
2574                         }
2575                     }
2576                     break;
2577                 default:
2578                     av_log(s->avctx, AV_LOG_ERROR, "illegal MB type\n");
2579                 }
2580
2581                 encode_mb(s, motion_x, motion_y);
2582
2583                 // RAL: Update last macroblock type
2584                 s->last_mv_dir = s->mv_dir;
2585
2586                 if (ENABLE_ANY_H263_ENCODER &&
2587                     s->out_format == FMT_H263 && s->pict_type!=FF_B_TYPE)
2588                     ff_h263_update_motion_val(s);
2589
2590                 MPV_decode_mb(s, s->block);
2591             }
2592
2593             /* clean the MV table in IPS frames for direct mode in B frames */
2594             if(s->mb_intra /* && I,P,S_TYPE */){
2595                 s->p_mv_table[xy][0]=0;
2596                 s->p_mv_table[xy][1]=0;
2597             }
2598
2599             if(s->flags&CODEC_FLAG_PSNR){
2600                 int w= 16;
2601                 int h= 16;
2602
2603                 if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
2604                 if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
2605
2606                 s->current_picture.error[0] += sse(
2607                     s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16,
2608                     s->dest[0], w, h, s->linesize);
2609                 s->current_picture.error[1] += sse(
2610                     s, s->new_picture.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*chr_h,
2611                     s->dest[1], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
2612                 s->current_picture.error[2] += sse(
2613                     s, s->new_picture.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*chr_h,
2614                     s->dest[2], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
2615             }
2616             if(s->loop_filter){
2617                 if(ENABLE_ANY_H263_ENCODER && s->out_format == FMT_H263)
2618                     ff_h263_loop_filter(s);
2619             }
2620 //printf("MB %d %d bits\n", s->mb_x+s->mb_y*s->mb_stride, put_bits_count(&s->pb));
2621         }
2622     }
2623
2624     //not beautiful here but we must write it before flushing so it has to be here
2625     if (ENABLE_MSMPEG4_ENCODER && s->msmpeg4_version && s->msmpeg4_version<4 && s->pict_type == FF_I_TYPE)
2626         msmpeg4_encode_ext_header(s);
2627
2628     write_slice_end(s);
2629
2630     /* Send the last GOB if RTP */
2631     if (s->avctx->rtp_callback) {
2632         int number_mb = (mb_y - s->resync_mb_y)*s->mb_width - s->resync_mb_x;
2633         pdif = pbBufPtr(&s->pb) - s->ptr_lastgob;
2634         /* Call the RTP callback to send the last GOB */
2635         emms_c();
2636         s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, pdif, number_mb);
2637     }
2638
2639     return 0;
2640 }
2641
2642 #define MERGE(field) dst->field += src->field; src->field=0
2643 static void merge_context_after_me(MpegEncContext *dst, MpegEncContext *src){
2644     MERGE(me.scene_change_score);
2645     MERGE(me.mc_mb_var_sum_temp);
2646     MERGE(me.mb_var_sum_temp);
2647 }
2648
2649 static void merge_context_after_encode(MpegEncContext *dst, MpegEncContext *src){
2650     int i;
2651
2652     MERGE(dct_count[0]); //note, the other dct vars are not part of the context
2653     MERGE(dct_count[1]);
2654     MERGE(mv_bits);
2655     MERGE(i_tex_bits);
2656     MERGE(p_tex_bits);
2657     MERGE(i_count);
2658     MERGE(f_count);
2659     MERGE(b_count);
2660     MERGE(skip_count);
2661     MERGE(misc_bits);
2662     MERGE(error_count);
2663     MERGE(padding_bug_score);
2664     MERGE(current_picture.error[0]);
2665     MERGE(current_picture.error[1]);
2666     MERGE(current_picture.error[2]);
2667
2668     if(dst->avctx->noise_reduction){
2669         for(i=0; i<64; i++){
2670             MERGE(dct_error_sum[0][i]);
2671             MERGE(dct_error_sum[1][i]);
2672         }
2673     }
2674
2675     assert(put_bits_count(&src->pb) % 8 ==0);
2676     assert(put_bits_count(&dst->pb) % 8 ==0);
2677     ff_copy_bits(&dst->pb, src->pb.buf, put_bits_count(&src->pb));
2678     flush_put_bits(&dst->pb);
2679 }
2680
2681 static int estimate_qp(MpegEncContext *s, int dry_run){
2682     if (s->next_lambda){
2683         s->current_picture_ptr->quality=
2684         s->current_picture.quality = s->next_lambda;
2685         if(!dry_run) s->next_lambda= 0;
2686     } else if (!s->fixed_qscale) {
2687         s->current_picture_ptr->quality=
2688         s->current_picture.quality = ff_rate_estimate_qscale(s, dry_run);
2689         if (s->current_picture.quality < 0)
2690             return -1;
2691     }
2692
2693     if(s->adaptive_quant){
2694         switch(s->codec_id){
2695         case CODEC_ID_MPEG4:
2696             if (ENABLE_MPEG4_ENCODER)
2697                 ff_clean_mpeg4_qscales(s);
2698             break;
2699         case CODEC_ID_H263:
2700         case CODEC_ID_H263P:
2701         case CODEC_ID_FLV1:
2702             if (ENABLE_H263_ENCODER||ENABLE_H263P_ENCODER||ENABLE_FLV_ENCODER)
2703                 ff_clean_h263_qscales(s);
2704             break;
2705         }
2706
2707         s->lambda= s->lambda_table[0];
2708         //FIXME broken
2709     }else
2710         s->lambda= s->current_picture.quality;
2711 //printf("%d %d\n", s->avctx->global_quality, s->current_picture.quality);
2712     update_qscale(s);
2713     return 0;
2714 }
2715
2716 /* must be called before writing the header */
2717 static void set_frame_distances(MpegEncContext * s){
2718     assert(s->current_picture_ptr->pts != AV_NOPTS_VALUE);
2719     s->time= s->current_picture_ptr->pts*s->avctx->time_base.num;
2720
2721     if(s->pict_type==FF_B_TYPE){
2722         s->pb_time= s->pp_time - (s->last_non_b_time - s->time);
2723         assert(s->pb_time > 0 && s->pb_time < s->pp_time);
2724     }else{
2725         s->pp_time= s->time - s->last_non_b_time;
2726         s->last_non_b_time= s->time;
2727         assert(s->picture_number==0 || s->pp_time > 0);
2728     }
2729 }
2730
2731 static int encode_picture(MpegEncContext *s, int picture_number)
2732 {
2733     int i;
2734     int bits;
2735
2736     s->picture_number = picture_number;
2737
2738     /* Reset the average MB variance */
2739     s->me.mb_var_sum_temp    =
2740     s->me.mc_mb_var_sum_temp = 0;
2741
2742     /* we need to initialize some time vars before we can encode b-frames */
2743     // RAL: Condition added for MPEG1VIDEO
2744     if (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO || (s->h263_pred && !s->h263_msmpeg4))
2745         set_frame_distances(s);
2746     if(ENABLE_MPEG4_ENCODER && s->codec_id == CODEC_ID_MPEG4)
2747         ff_set_mpeg4_time(s);
2748
2749     s->me.scene_change_score=0;
2750
2751 //    s->lambda= s->current_picture_ptr->quality; //FIXME qscale / ... stuff for ME rate distortion
2752
2753     if(s->pict_type==FF_I_TYPE){
2754         if(s->msmpeg4_version >= 3) s->no_rounding=1;
2755         else                        s->no_rounding=0;
2756     }else if(s->pict_type!=FF_B_TYPE){
2757         if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
2758             s->no_rounding ^= 1;
2759     }
2760
2761     if(s->flags & CODEC_FLAG_PASS2){
2762         if (estimate_qp(s,1) < 0)
2763             return -1;
2764         ff_get_2pass_fcode(s);
2765     }else if(!(s->flags & CODEC_FLAG_QSCALE)){
2766         if(s->pict_type==FF_B_TYPE)
2767             s->lambda= s->last_lambda_for[s->pict_type];
2768         else
2769             s->lambda= s->last_lambda_for[s->last_non_b_pict_type];
2770         update_qscale(s);
2771     }
2772
2773     s->mb_intra=0; //for the rate distortion & bit compare functions
2774     for(i=1; i<s->avctx->thread_count; i++){
2775         ff_update_duplicate_context(s->thread_context[i], s);
2776     }
2777
2778     ff_init_me(s);
2779
2780     /* Estimate motion for every MB */
2781     if(s->pict_type != FF_I_TYPE){
2782         s->lambda = (s->lambda * s->avctx->me_penalty_compensation + 128)>>8;
2783         s->lambda2= (s->lambda2* (int64_t)s->avctx->me_penalty_compensation + 128)>>8;
2784         if(s->pict_type != FF_B_TYPE && s->avctx->me_threshold==0){
2785             if((s->avctx->pre_me && s->last_non_b_pict_type==FF_I_TYPE) || s->avctx->pre_me==2){
2786                 s->avctx->execute(s->avctx, pre_estimate_motion_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count);
2787             }
2788         }
2789
2790         s->avctx->execute(s->avctx, estimate_motion_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count);
2791     }else /* if(s->pict_type == FF_I_TYPE) */{
2792         /* I-Frame */
2793         for(i=0; i<s->mb_stride*s->mb_height; i++)
2794             s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
2795
2796         if(!s->fixed_qscale){
2797             /* finding spatial complexity for I-frame rate control */
2798             s->avctx->execute(s->avctx, mb_var_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count);
2799         }
2800     }
2801     for(i=1; i<s->avctx->thread_count; i++){
2802         merge_context_after_me(s, s->thread_context[i]);
2803     }
2804     s->current_picture.mc_mb_var_sum= s->current_picture_ptr->mc_mb_var_sum= s->me.mc_mb_var_sum_temp;
2805     s->current_picture.   mb_var_sum= s->current_picture_ptr->   mb_var_sum= s->me.   mb_var_sum_temp;
2806     emms_c();
2807
2808     if(s->me.scene_change_score > s->avctx->scenechange_threshold && s->pict_type == FF_P_TYPE){
2809         s->pict_type= FF_I_TYPE;
2810         for(i=0; i<s->mb_stride*s->mb_height; i++)
2811             s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
2812 //printf("Scene change detected, encoding as I Frame %d %d\n", s->current_picture.mb_var_sum, s->current_picture.mc_mb_var_sum);
2813     }
2814
2815     if(!s->umvplus){
2816         if(s->pict_type==FF_P_TYPE || s->pict_type==FF_S_TYPE) {
2817             s->f_code= ff_get_best_fcode(s, s->p_mv_table, CANDIDATE_MB_TYPE_INTER);
2818
2819             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2820                 int a,b;
2821                 a= ff_get_best_fcode(s, s->p_field_mv_table[0][0], CANDIDATE_MB_TYPE_INTER_I); //FIXME field_select
2822                 b= ff_get_best_fcode(s, s->p_field_mv_table[1][1], CANDIDATE_MB_TYPE_INTER_I);
2823                 s->f_code= FFMAX3(s->f_code, a, b);
2824             }
2825
2826             ff_fix_long_p_mvs(s);
2827             ff_fix_long_mvs(s, NULL, 0, s->p_mv_table, s->f_code, CANDIDATE_MB_TYPE_INTER, 0);
2828             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2829                 int j;
2830                 for(i=0; i<2; i++){
2831                     for(j=0; j<2; j++)
2832                         ff_fix_long_mvs(s, s->p_field_select_table[i], j,
2833                                         s->p_field_mv_table[i][j], s->f_code, CANDIDATE_MB_TYPE_INTER_I, 0);
2834                 }
2835             }
2836         }
2837
2838         if(s->pict_type==FF_B_TYPE){
2839             int a, b;
2840
2841             a = ff_get_best_fcode(s, s->b_forw_mv_table, CANDIDATE_MB_TYPE_FORWARD);
2842             b = ff_get_best_fcode(s, s->b_bidir_forw_mv_table, CANDIDATE_MB_TYPE_BIDIR);
2843             s->f_code = FFMAX(a, b);
2844
2845             a = ff_get_best_fcode(s, s->b_back_mv_table, CANDIDATE_MB_TYPE_BACKWARD);
2846             b = ff_get_best_fcode(s, s->b_bidir_back_mv_table, CANDIDATE_MB_TYPE_BIDIR);
2847             s->b_code = FFMAX(a, b);
2848
2849             ff_fix_long_mvs(s, NULL, 0, s->b_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_FORWARD, 1);
2850             ff_fix_long_mvs(s, NULL, 0, s->b_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BACKWARD, 1);
2851             ff_fix_long_mvs(s, NULL, 0, s->b_bidir_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_BIDIR, 1);
2852             ff_fix_long_mvs(s, NULL, 0, s->b_bidir_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BIDIR, 1);
2853             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2854                 int dir, j;
2855                 for(dir=0; dir<2; dir++){
2856                     for(i=0; i<2; i++){
2857                         for(j=0; j<2; j++){
2858                             int type= dir ? (CANDIDATE_MB_TYPE_BACKWARD_I|CANDIDATE_MB_TYPE_BIDIR_I)
2859                                           : (CANDIDATE_MB_TYPE_FORWARD_I |CANDIDATE_MB_TYPE_BIDIR_I);
2860                             ff_fix_long_mvs(s, s->b_field_select_table[dir][i], j,
2861                                             s->b_field_mv_table[dir][i][j], dir ? s->b_code : s->f_code, type, 1);
2862                         }
2863                     }
2864                 }
2865             }
2866         }
2867     }
2868
2869     if (estimate_qp(s, 0) < 0)
2870         return -1;
2871
2872     if(s->qscale < 3 && s->max_qcoeff<=128 && s->pict_type==FF_I_TYPE && !(s->flags & CODEC_FLAG_QSCALE))
2873         s->qscale= 3; //reduce clipping problems
2874
2875     if (s->out_format == FMT_MJPEG) {
2876         /* for mjpeg, we do include qscale in the matrix */
2877         s->intra_matrix[0] = ff_mpeg1_default_intra_matrix[0];
2878         for(i=1;i<64;i++){
2879             int j= s->dsp.idct_permutation[i];
2880
2881             s->intra_matrix[j] = av_clip_uint8((ff_mpeg1_default_intra_matrix[i] * s->qscale) >> 3);
2882         }
2883         ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
2884                        s->intra_matrix, s->intra_quant_bias, 8, 8, 1);
2885         s->qscale= 8;
2886     }
2887
2888     //FIXME var duplication
2889     s->current_picture_ptr->key_frame=
2890     s->current_picture.key_frame= s->pict_type == FF_I_TYPE; //FIXME pic_ptr
2891     s->current_picture_ptr->pict_type=
2892     s->current_picture.pict_type= s->pict_type;
2893
2894     if(s->current_picture.key_frame)
2895         s->picture_in_gop_number=0;
2896
2897     s->last_bits= put_bits_count(&s->pb);
2898     switch(s->out_format) {
2899     case FMT_MJPEG:
2900         if (ENABLE_MJPEG_ENCODER)
2901             ff_mjpeg_encode_picture_header(s);
2902         break;
2903     case FMT_H261:
2904         if (ENABLE_H261_ENCODER)
2905             ff_h261_encode_picture_header(s, picture_number);
2906         break;
2907     case FMT_H263:
2908         if (ENABLE_WMV2_ENCODER && s->codec_id == CODEC_ID_WMV2)
2909             ff_wmv2_encode_picture_header(s, picture_number);
2910         else if (ENABLE_MSMPEG4_ENCODER && s->h263_msmpeg4)
2911             msmpeg4_encode_picture_header(s, picture_number);
2912         else if (ENABLE_MPEG4_ENCODER && s->h263_pred)
2913             mpeg4_encode_picture_header(s, picture_number);
2914         else if (ENABLE_RV10_ENCODER && s->codec_id == CODEC_ID_RV10)
2915             rv10_encode_picture_header(s, picture_number);
2916         else if (ENABLE_RV20_ENCODER && s->codec_id == CODEC_ID_RV20)
2917             rv20_encode_picture_header(s, picture_number);
2918         else if (ENABLE_FLV_ENCODER && s->codec_id == CODEC_ID_FLV1)
2919             ff_flv_encode_picture_header(s, picture_number);
2920         else if (ENABLE_ANY_H263_ENCODER)
2921             h263_encode_picture_header(s, picture_number);
2922         break;
2923     case FMT_MPEG1:
2924         if (ENABLE_MPEG1VIDEO_ENCODER || ENABLE_MPEG2VIDEO_ENCODER)
2925             mpeg1_encode_picture_header(s, picture_number);
2926         break;
2927     case FMT_H264:
2928         break;
2929     default:
2930         assert(0);
2931     }
2932     bits= put_bits_count(&s->pb);
2933     s->header_bits= bits - s->last_bits;
2934
2935     for(i=1; i<s->avctx->thread_count; i++){
2936         update_duplicate_context_after_me(s->thread_context[i], s);
2937     }
2938     s->avctx->execute(s->avctx, encode_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count);
2939     for(i=1; i<s->avctx->thread_count; i++){
2940         merge_context_after_encode(s, s->thread_context[i]);
2941     }
2942     emms_c();
2943     return 0;
2944 }
2945
2946 void  denoise_dct_c(MpegEncContext *s, DCTELEM *block){
2947     const int intra= s->mb_intra;
2948     int i;
2949
2950     s->dct_count[intra]++;
2951
2952     for(i=0; i<64; i++){
2953         int level= block[i];
2954
2955         if(level){
2956             if(level>0){
2957                 s->dct_error_sum[intra][i] += level;
2958                 level -= s->dct_offset[intra][i];
2959                 if(level<0) level=0;
2960             }else{
2961                 s->dct_error_sum[intra][i] -= level;
2962                 level += s->dct_offset[intra][i];
2963                 if(level>0) level=0;
2964             }
2965             block[i]= level;
2966         }
2967     }
2968 }
2969
2970 int dct_quantize_trellis_c(MpegEncContext *s,
2971                         DCTELEM *block, int n,
2972                         int qscale, int *overflow){
2973     const int *qmat;
2974     const uint8_t *scantable= s->intra_scantable.scantable;
2975     const uint8_t *perm_scantable= s->intra_scantable.permutated;
2976     int max=0;
2977     unsigned int threshold1, threshold2;
2978     int bias=0;
2979     int run_tab[65];
2980     int level_tab[65];
2981     int score_tab[65];
2982     int survivor[65];
2983     int survivor_count;
2984     int last_run=0;
2985     int last_level=0;
2986     int last_score= 0;
2987     int last_i;
2988     int coeff[2][64];
2989     int coeff_count[64];
2990     int qmul, qadd, start_i, last_non_zero, i, dc;
2991     const int esc_length= s->ac_esc_length;
2992     uint8_t * length;
2993     uint8_t * last_length;
2994     const int lambda= s->lambda2 >> (FF_LAMBDA_SHIFT - 6);
2995
2996     s->dsp.fdct (block);
2997
2998     if(s->dct_error_sum)
2999         s->denoise_dct(s, block);
3000     qmul= qscale*16;
3001     qadd= ((qscale-1)|1)*8;
3002
3003     if (s->mb_intra) {
3004         int q;
3005         if (!s->h263_aic) {
3006             if (n < 4)
3007                 q = s->y_dc_scale;
3008             else
3009                 q = s->c_dc_scale;
3010             q = q << 3;
3011         } else{
3012             /* For AIC we skip quant/dequant of INTRADC */
3013             q = 1 << 3;
3014             qadd=0;
3015         }
3016
3017         /* note: block[0] is assumed to be positive */
3018         block[0] = (block[0] + (q >> 1)) / q;
3019         start_i = 1;
3020         last_non_zero = 0;
3021         qmat = s->q_intra_matrix[qscale];
3022         if(s->mpeg_quant || s->out_format == FMT_MPEG1)
3023             bias= 1<<(QMAT_SHIFT-1);
3024         length     = s->intra_ac_vlc_length;
3025         last_length= s->intra_ac_vlc_last_length;
3026     } else {
3027         start_i = 0;
3028         last_non_zero = -1;
3029         qmat = s->q_inter_matrix[qscale];
3030         length     = s->inter_ac_vlc_length;
3031         last_length= s->inter_ac_vlc_last_length;
3032     }
3033     last_i= start_i;
3034
3035     threshold1= (1<<QMAT_SHIFT) - bias - 1;
3036     threshold2= (threshold1<<1);
3037
3038     for(i=63; i>=start_i; i--) {
3039         const int j = scantable[i];
3040         int level = block[j] * qmat[j];
3041
3042         if(((unsigned)(level+threshold1))>threshold2){
3043             last_non_zero = i;
3044             break;
3045         }
3046     }
3047
3048     for(i=start_i; i<=last_non_zero; i++) {
3049         const int j = scantable[i];
3050         int level = block[j] * qmat[j];
3051
3052 //        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
3053 //           || bias-level >= (1<<(QMAT_SHIFT - 3))){
3054         if(((unsigned)(level+threshold1))>threshold2){
3055             if(level>0){
3056                 level= (bias + level)>>QMAT_SHIFT;
3057                 coeff[0][i]= level;
3058                 coeff[1][i]= level-1;
3059 //                coeff[2][k]= level-2;
3060             }else{
3061                 level= (bias - level)>>QMAT_SHIFT;
3062                 coeff[0][i]= -level;
3063                 coeff[1][i]= -level+1;
3064 //                coeff[2][k]= -level+2;
3065             }
3066             coeff_count[i]= FFMIN(level, 2);
3067             assert(coeff_count[i]);
3068             max |=level;
3069         }else{
3070             coeff[0][i]= (level>>31)|1;
3071             coeff_count[i]= 1;
3072         }
3073     }
3074
3075     *overflow= s->max_qcoeff < max; //overflow might have happened
3076
3077     if(last_non_zero < start_i){
3078         memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
3079         return last_non_zero;
3080     }
3081
3082     score_tab[start_i]= 0;
3083     survivor[0]= start_i;
3084     survivor_count= 1;
3085
3086     for(i=start_i; i<=last_non_zero; i++){
3087         int level_index, j, zero_distortion;
3088         int dct_coeff= FFABS(block[ scantable[i] ]);
3089         int best_score=256*256*256*120;
3090
3091         if (   s->dsp.fdct == fdct_ifast
3092 #ifndef FAAN_POSTSCALE
3093             || s->dsp.fdct == ff_faandct
3094 #endif
3095            )
3096             dct_coeff= (dct_coeff*inv_aanscales[ scantable[i] ]) >> 12;
3097         zero_distortion= dct_coeff*dct_coeff;
3098
3099         for(level_index=0; level_index < coeff_count[i]; level_index++){
3100             int distortion;
3101             int level= coeff[level_index][i];
3102             const int alevel= FFABS(level);
3103             int unquant_coeff;
3104
3105             assert(level);
3106
3107             if(s->out_format == FMT_H263){
3108                 unquant_coeff= alevel*qmul + qadd;
3109             }else{ //MPEG1
3110                 j= s->dsp.idct_permutation[ scantable[i] ]; //FIXME optimize
3111                 if(s->mb_intra){
3112                         unquant_coeff = (int)(  alevel  * qscale * s->intra_matrix[j]) >> 3;
3113                         unquant_coeff =   (unquant_coeff - 1) | 1;
3114                 }else{
3115                         unquant_coeff = (((  alevel  << 1) + 1) * qscale * ((int) s->inter_matrix[j])) >> 4;
3116                         unquant_coeff =   (unquant_coeff - 1) | 1;
3117                 }
3118                 unquant_coeff<<= 3;
3119             }
3120
3121             distortion= (unquant_coeff - dct_coeff) * (unquant_coeff - dct_coeff) - zero_distortion;
3122             level+=64;
3123             if((level&(~127)) == 0){
3124                 for(j=survivor_count-1; j>=0; j--){
3125                     int run= i - survivor[j];
3126                     int score= distortion + length[UNI_AC_ENC_INDEX(run, level)]*lambda;
3127                     score += score_tab[i-run];
3128
3129                     if(score < best_score){
3130                         best_score= score;
3131                         run_tab[i+1]= run;
3132                         level_tab[i+1]= level-64;
3133                     }
3134                 }
3135
3136                 if(s->out_format == FMT_H263){
3137                     for(j=survivor_count-1; j>=0; j--){
3138                         int run= i - survivor[j];
3139                         int score= distortion + last_length[UNI_AC_ENC_INDEX(run, level)]*lambda;
3140                         score += score_tab[i-run];
3141                         if(score < last_score){
3142                             last_score= score;
3143                             last_run= run;
3144                             last_level= level-64;
3145                             last_i= i+1;
3146                         }
3147                     }
3148                 }
3149             }else{
3150                 distortion += esc_length*lambda;
3151                 for(j=survivor_count-1; j>=0; j--){
3152                     int run= i - survivor[j];
3153                     int score= distortion + score_tab[i-run];
3154
3155                     if(score < best_score){
3156                         best_score= score;
3157                         run_tab[i+1]= run;
3158                         level_tab[i+1]= level-64;
3159                     }
3160                 }
3161
3162                 if(s->out_format == FMT_H263){
3163                   for(j=survivor_count-1; j>=0; j--){
3164                         int run= i - survivor[j];
3165                         int score= distortion + score_tab[i-run];
3166                         if(score < last_score){
3167                             last_score= score;
3168                             last_run= run;
3169                             last_level= level-64;
3170                             last_i= i+1;
3171                         }
3172                     }
3173                 }
3174             }
3175         }
3176
3177         score_tab[i+1]= best_score;
3178
3179         //Note: there is a vlc code in mpeg4 which is 1 bit shorter then another one with a shorter run and the same level
3180         if(last_non_zero <= 27){
3181             for(; survivor_count; survivor_count--){
3182                 if(score_tab[ survivor[survivor_count-1] ] <= best_score)
3183                     break;
3184             }
3185         }else{
3186             for(; survivor_count; survivor_count--){
3187                 if(score_tab[ survivor[survivor_count-1] ] <= best_score + lambda)
3188                     break;
3189             }
3190         }
3191
3192         survivor[ survivor_count++ ]= i+1;
3193     }
3194
3195     if(s->out_format != FMT_H263){
3196         last_score= 256*256*256*120;
3197         for(i= survivor[0]; i<=last_non_zero + 1; i++){
3198             int score= score_tab[i];
3199             if(i) score += lambda*2; //FIXME exacter?
3200
3201             if(score < last_score){
3202                 last_score= score;
3203                 last_i= i;
3204                 last_level= level_tab[i];
3205                 last_run= run_tab[i];
3206             }
3207         }
3208     }
3209
3210     s->coded_score[n] = last_score;
3211
3212     dc= FFABS(block[0]);
3213     last_non_zero= last_i - 1;
3214     memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
3215
3216     if(last_non_zero < start_i)
3217         return last_non_zero;
3218
3219     if(last_non_zero == 0 && start_i == 0){
3220         int best_level= 0;
3221         int best_score= dc * dc;
3222
3223         for(i=0; i<coeff_count[0]; i++){
3224             int level= coeff[i][0];
3225             int alevel= FFABS(level);
3226             int unquant_coeff, score, distortion;
3227
3228             if(s->out_format == FMT_H263){
3229                     unquant_coeff= (alevel*qmul + qadd)>>3;
3230             }else{ //MPEG1
3231                     unquant_coeff = (((  alevel  << 1) + 1) * qscale * ((int) s->inter_matrix[0])) >> 4;
3232                     unquant_coeff =   (unquant_coeff - 1) | 1;
3233             }
3234             unquant_coeff = (unquant_coeff + 4) >> 3;
3235             unquant_coeff<<= 3 + 3;
3236
3237             distortion= (unquant_coeff - dc) * (unquant_coeff - dc);
3238             level+=64;
3239             if((level&(~127)) == 0) score= distortion + last_length[UNI_AC_ENC_INDEX(0, level)]*lambda;
3240             else                    score= distortion + esc_length*lambda;
3241
3242             if(score < best_score){
3243                 best_score= score;
3244                 best_level= level - 64;
3245             }
3246         }
3247         block[0]= best_level;
3248         s->coded_score[n] = best_score - dc*dc;
3249         if(best_level == 0) return -1;
3250         else                return last_non_zero;
3251     }
3252
3253     i= last_i;
3254     assert(last_level);
3255
3256     block[ perm_scantable[last_non_zero] ]= last_level;
3257     i -= last_run + 1;
3258
3259     for(; i>start_i; i -= run_tab[i] + 1){
3260         block[ perm_scantable[i-1] ]= level_tab[i];
3261     }
3262
3263     return last_non_zero;
3264 }
3265
3266 //#define REFINE_STATS 1
3267 static int16_t basis[64][64];
3268
3269 static void build_basis(uint8_t *perm){
3270     int i, j, x, y;
3271     emms_c();
3272     for(i=0; i<8; i++){
3273         for(j=0; j<8; j++){
3274             for(y=0; y<8; y++){
3275                 for(x=0; x<8; x++){
3276                     double s= 0.25*(1<<BASIS_SHIFT);
3277                     int index= 8*i + j;
3278                     int perm_index= perm[index];
3279                     if(i==0) s*= sqrt(0.5);
3280                     if(j==0) s*= sqrt(0.5);
3281                     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)));
3282                 }
3283             }
3284         }
3285     }
3286 }
3287
3288 static int dct_quantize_refine(MpegEncContext *s, //FIXME breaks denoise?
3289                         DCTELEM *block, int16_t *weight, DCTELEM *orig,
3290                         int n, int qscale){
3291     int16_t rem[64];
3292     DECLARE_ALIGNED_16(DCTELEM, d1[64]);
3293     const int *qmat;
3294     const uint8_t *scantable= s->intra_scantable.scantable;
3295     const uint8_t *perm_scantable= s->intra_scantable.permutated;
3296 //    unsigned int threshold1, threshold2;
3297 //    int bias=0;
3298     int run_tab[65];
3299     int prev_run=0;
3300     int prev_level=0;
3301     int qmul, qadd, start_i, last_non_zero, i, dc;
3302     uint8_t * length;
3303     uint8_t * last_length;
3304     int lambda;
3305     int rle_index, run, q = 1, sum; //q is only used when s->mb_intra is true
3306 #ifdef REFINE_STATS
3307 static int count=0;
3308 static int after_last=0;
3309 static int to_zero=0;
3310 static int from_zero=0;
3311 static int raise=0;
3312 static int lower=0;
3313 static int messed_sign=0;
3314 #endif
3315
3316     if(basis[0][0] == 0)
3317         build_basis(s->dsp.idct_permutation);
3318
3319     qmul= qscale*2;
3320     qadd= (qscale-1)|1;
3321     if (s->mb_intra) {
3322         if (!s->h263_aic) {
3323             if (n < 4)
3324                 q = s->y_dc_scale;
3325             else
3326                 q = s->c_dc_scale;
3327         } else{
3328             /* For AIC we skip quant/dequant of INTRADC */
3329             q = 1;
3330             qadd=0;
3331         }
3332         q <<= RECON_SHIFT-3;
3333         /* note: block[0] is assumed to be positive */
3334         dc= block[0]*q;
3335 //        block[0] = (block[0] + (q >> 1)) / q;
3336         start_i = 1;
3337         qmat = s->q_intra_matrix[qscale];
3338 //        if(s->mpeg_quant || s->out_format == FMT_MPEG1)
3339 //            bias= 1<<(QMAT_SHIFT-1);
3340         length     = s->intra_ac_vlc_length;
3341         last_length= s->intra_ac_vlc_last_length;
3342     } else {
3343         dc= 0;
3344         start_i = 0;
3345         qmat = s->q_inter_matrix[qscale];
3346         length     = s->inter_ac_vlc_length;
3347         last_length= s->inter_ac_vlc_last_length;
3348     }
3349     last_non_zero = s->block_last_index[n];
3350
3351 #ifdef REFINE_STATS
3352 {START_TIMER
3353 #endif
3354     dc += (1<<(RECON_SHIFT-1));
3355     for(i=0; i<64; i++){
3356         rem[i]= dc - (orig[i]<<RECON_SHIFT); //FIXME  use orig dirrectly instead of copying to rem[]
3357     }
3358 #ifdef REFINE_STATS
3359 STOP_TIMER("memset rem[]")}
3360 #endif
3361     sum=0;
3362     for(i=0; i<64; i++){
3363         int one= 36;
3364         int qns=4;
3365         int w;
3366
3367         w= FFABS(weight[i]) + qns*one;
3368         w= 15 + (48*qns*one + w/2)/w; // 16 .. 63
3369
3370         weight[i] = w;
3371 //        w=weight[i] = (63*qns + (w/2)) / w;
3372
3373         assert(w>0);
3374         assert(w<(1<<6));
3375         sum += w*w;
3376     }
3377     lambda= sum*(uint64_t)s->lambda2 >> (FF_LAMBDA_SHIFT - 6 + 6 + 6 + 6);
3378 #ifdef REFINE_STATS
3379 {START_TIMER
3380 #endif
3381     run=0;
3382     rle_index=0;
3383     for(i=start_i; i<=last_non_zero; i++){
3384         int j= perm_scantable[i];
3385         const int level= block[j];
3386         int coeff;
3387
3388         if(level){
3389             if(level<0) coeff= qmul*level - qadd;
3390             else        coeff= qmul*level + qadd;
3391             run_tab[rle_index++]=run;
3392             run=0;
3393
3394             s->dsp.add_8x8basis(rem, basis[j], coeff);
3395         }else{
3396             run++;
3397         }
3398     }
3399 #ifdef REFINE_STATS
3400 if(last_non_zero>0){
3401 STOP_TIMER("init rem[]")
3402 }
3403 }
3404
3405 {START_TIMER
3406 #endif
3407     for(;;){
3408         int best_score=s->dsp.try_8x8basis(rem, weight, basis[0], 0);
3409         int best_coeff=0;
3410         int best_change=0;
3411         int run2, best_unquant_change=0, analyze_gradient;
3412 #ifdef REFINE_STATS
3413 {START_TIMER
3414 #endif
3415         analyze_gradient = last_non_zero > 2 || s->avctx->quantizer_noise_shaping >= 3;
3416
3417         if(analyze_gradient){
3418 #ifdef REFINE_STATS
3419 {START_TIMER
3420 #endif
3421             for(i=0; i<64; i++){
3422                 int w= weight[i];
3423
3424                 d1[i] = (rem[i]*w*w + (1<<(RECON_SHIFT+12-1)))>>(RECON_SHIFT+12);
3425             }
3426 #ifdef REFINE_STATS
3427 STOP_TIMER("rem*w*w")}
3428 {START_TIMER
3429 #endif
3430             s->dsp.fdct(d1);
3431 #ifdef REFINE_STATS
3432 STOP_TIMER("dct")}
3433 #endif
3434         }
3435
3436         if(start_i){
3437             const int level= block[0];
3438             int change, old_coeff;
3439
3440             assert(s->mb_intra);
3441
3442             old_coeff= q*level;
3443
3444             for(change=-1; change<=1; change+=2){
3445                 int new_level= level + change;
3446                 int score, new_coeff;
3447
3448                 new_coeff= q*new_level;
3449                 if(new_coeff >= 2048 || new_coeff < 0)
3450                     continue;
3451
3452                 score= s->dsp.try_8x8basis(rem, weight, basis[0], new_coeff - old_coeff);
3453                 if(score<best_score){
3454                     best_score= score;
3455                     best_coeff= 0;
3456                     best_change= change;
3457                     best_unquant_change= new_coeff - old_coeff;
3458                 }
3459             }
3460         }
3461
3462         run=0;
3463         rle_index=0;
3464         run2= run_tab[rle_index++];
3465         prev_level=0;
3466         prev_run=0;
3467
3468         for(i=start_i; i<64; i++){
3469             int j= perm_scantable[i];
3470             const int level= block[j];
3471             int change, old_coeff;
3472
3473             if(s->avctx->quantizer_noise_shaping < 3 && i > last_non_zero + 1)
3474                 break;
3475
3476             if(level){
3477                 if(level<0) old_coeff= qmul*level - qadd;
3478                 else        old_coeff= qmul*level + qadd;
3479                 run2= run_tab[rle_index++]; //FIXME ! maybe after last
3480             }else{
3481                 old_coeff=0;
3482                 run2--;
3483                 assert(run2>=0 || i >= last_non_zero );
3484             }
3485
3486             for(change=-1; change<=1; change+=2){
3487                 int new_level= level + change;
3488                 int score, new_coeff, unquant_change;
3489
3490                 score=0;
3491                 if(s->avctx->quantizer_noise_shaping < 2 && FFABS(new_level) > FFABS(level))
3492                    continue;
3493
3494                 if(new_level){
3495                     if(new_level<0) new_coeff= qmul*new_level - qadd;
3496                     else            new_coeff= qmul*new_level + qadd;
3497                     if(new_coeff >= 2048 || new_coeff <= -2048)
3498                         continue;
3499                     //FIXME check for overflow
3500
3501                     if(level){
3502                         if(level < 63 && level > -63){
3503                             if(i < last_non_zero)
3504                                 score +=   length[UNI_AC_ENC_INDEX(run, new_level+64)]
3505                                          - length[UNI_AC_ENC_INDEX(run, level+64)];
3506                             else
3507                                 score +=   last_length[UNI_AC_ENC_INDEX(run, new_level+64)]
3508                                          - last_length[UNI_AC_ENC_INDEX(run, level+64)];
3509                         }
3510                     }else{
3511                         assert(FFABS(new_level)==1);
3512
3513                         if(analyze_gradient){
3514                             int g= d1[ scantable[i] ];
3515                             if(g && (g^new_level) >= 0)
3516                                 continue;
3517                         }
3518
3519                         if(i < last_non_zero){
3520                             int next_i= i + run2 + 1;
3521                             int next_level= block[ perm_scantable[next_i] ] + 64;
3522
3523                             if(next_level&(~127))
3524                                 next_level= 0;
3525
3526                             if(next_i < last_non_zero)
3527                                 score +=   length[UNI_AC_ENC_INDEX(run, 65)]
3528                                          + length[UNI_AC_ENC_INDEX(run2, next_level)]
3529                                          - length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)];
3530                             else
3531                                 score +=  length[UNI_AC_ENC_INDEX(run, 65)]
3532                                         + last_length[UNI_AC_ENC_INDEX(run2, next_level)]
3533                                         - last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)];
3534                         }else{
3535                             score += last_length[UNI_AC_ENC_INDEX(run, 65)];
3536                             if(prev_level){
3537                                 score +=  length[UNI_AC_ENC_INDEX(prev_run, prev_level)]
3538                                         - last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)];
3539                             }
3540                         }
3541                     }
3542                 }else{
3543                     new_coeff=0;
3544                     assert(FFABS(level)==1);
3545
3546                     if(i < last_non_zero){
3547                         int next_i= i + run2 + 1;
3548                         int next_level= block[ perm_scantable[next_i] ] + 64;
3549
3550                         if(next_level&(~127))
3551                             next_level= 0;
3552
3553                         if(next_i < last_non_zero)
3554                             score +=   length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)]
3555                                      - length[UNI_AC_ENC_INDEX(run2, next_level)]
3556                                      - length[UNI_AC_ENC_INDEX(run, 65)];
3557                         else
3558                             score +=   last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)]
3559                                      - last_length[UNI_AC_ENC_INDEX(run2, next_level)]
3560                                      - length[UNI_AC_ENC_INDEX(run, 65)];
3561                     }else{
3562                         score += -last_length[UNI_AC_ENC_INDEX(run, 65)];
3563                         if(prev_level){
3564                             score +=  last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)]
3565                                     - length[UNI_AC_ENC_INDEX(prev_run, prev_level)];
3566                         }
3567                     }
3568                 }
3569
3570                 score *= lambda;
3571
3572                 unquant_change= new_coeff - old_coeff;
3573                 assert((score < 100*lambda && score > -100*lambda) || lambda==0);
3574
3575                 score+= s->dsp.try_8x8basis(rem, weight, basis[j], unquant_change);
3576                 if(score<best_score){
3577                     best_score= score;
3578                     best_coeff= i;
3579                     best_change= change;
3580                     best_unquant_change= unquant_change;
3581                 }
3582             }
3583             if(level){
3584                 prev_level= level + 64;
3585                 if(prev_level&(~127))
3586                     prev_level= 0;
3587                 prev_run= run;
3588                 run=0;
3589             }else{
3590                 run++;
3591             }
3592         }
3593 #ifdef REFINE_STATS
3594 STOP_TIMER("iterative step")}
3595 #endif
3596
3597         if(best_change){
3598             int j= perm_scantable[ best_coeff ];
3599
3600             block[j] += best_change;
3601
3602             if(best_coeff > last_non_zero){
3603                 last_non_zero= best_coeff;
3604                 assert(block[j]);
3605 #ifdef REFINE_STATS
3606 after_last++;
3607 #endif
3608             }else{
3609 #ifdef REFINE_STATS
3610 if(block[j]){
3611     if(block[j] - best_change){
3612         if(FFABS(block[j]) > FFABS(block[j] - best_change)){
3613             raise++;
3614         }else{
3615             lower++;
3616         }
3617     }else{
3618         from_zero++;
3619     }
3620 }else{
3621     to_zero++;
3622 }
3623 #endif
3624                 for(; last_non_zero>=start_i; last_non_zero--){
3625                     if(block[perm_scantable[last_non_zero]])
3626                         break;
3627                 }
3628             }
3629 #ifdef REFINE_STATS
3630 count++;
3631 if(256*256*256*64 % count == 0){
3632     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);
3633 }
3634 #endif
3635             run=0;
3636             rle_index=0;
3637             for(i=start_i; i<=last_non_zero; i++){
3638                 int j= perm_scantable[i];
3639                 const int level= block[j];
3640
3641                  if(level){
3642                      run_tab[rle_index++]=run;
3643                      run=0;
3644                  }else{
3645                      run++;
3646                  }
3647             }
3648
3649             s->dsp.add_8x8basis(rem, basis[j], best_unquant_change);
3650         }else{
3651             break;
3652         }
3653     }
3654 #ifdef REFINE_STATS
3655 if(last_non_zero>0){
3656 STOP_TIMER("iterative search")
3657 }
3658 }
3659 #endif
3660
3661     return last_non_zero;
3662 }
3663
3664 int dct_quantize_c(MpegEncContext *s,
3665                         DCTELEM *block, int n,
3666                         int qscale, int *overflow)
3667 {
3668     int i, j, level, last_non_zero, q, start_i;
3669     const int *qmat;
3670     const uint8_t *scantable= s->intra_scantable.scantable;
3671     int bias;
3672     int max=0;
3673     unsigned int threshold1, threshold2;
3674
3675     s->dsp.fdct (block);
3676
3677     if(s->dct_error_sum)
3678         s->denoise_dct(s, block);
3679
3680     if (s->mb_intra) {
3681         if (!s->h263_aic) {
3682             if (n < 4)
3683                 q = s->y_dc_scale;
3684             else
3685                 q = s->c_dc_scale;
3686             q = q << 3;
3687         } else
3688             /* For AIC we skip quant/dequant of INTRADC */
3689             q = 1 << 3;
3690
3691         /* note: block[0] is assumed to be positive */
3692         block[0] = (block[0] + (q >> 1)) / q;
3693         start_i = 1;
3694         last_non_zero = 0;
3695         qmat = s->q_intra_matrix[qscale];
3696         bias= s->intra_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT);
3697     } else {
3698         start_i = 0;
3699         last_non_zero = -1;
3700         qmat = s->q_inter_matrix[qscale];
3701         bias= s->inter_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT);
3702     }
3703     threshold1= (1<<QMAT_SHIFT) - bias - 1;
3704     threshold2= (threshold1<<1);
3705     for(i=63;i>=start_i;i--) {
3706         j = scantable[i];
3707         level = block[j] * qmat[j];
3708
3709         if(((unsigned)(level+threshold1))>threshold2){
3710             last_non_zero = i;
3711             break;
3712         }else{
3713             block[j]=0;
3714         }
3715     }
3716     for(i=start_i; i<=last_non_zero; i++) {
3717         j = scantable[i];
3718         level = block[j] * qmat[j];
3719
3720 //        if(   bias+level >= (1<<QMAT_SHIFT)
3721 //           || bias-level >= (1<<QMAT_SHIFT)){
3722         if(((unsigned)(level+threshold1))>threshold2){
3723             if(level>0){
3724                 level= (bias + level)>>QMAT_SHIFT;
3725                 block[j]= level;
3726             }else{
3727                 level= (bias - level)>>QMAT_SHIFT;
3728                 block[j]= -level;
3729             }
3730             max |=level;
3731         }else{
3732             block[j]=0;
3733         }
3734     }
3735     *overflow= s->max_qcoeff < max; //overflow might have happened
3736
3737     /* we need this permutation so that we correct the IDCT, we only permute the !=0 elements */
3738     if (s->dsp.idct_permutation_type != FF_NO_IDCT_PERM)
3739         ff_block_permute(block, s->dsp.idct_permutation, scantable, last_non_zero);
3740
3741     return last_non_zero;
3742 }
3743
3744 AVCodec h263_encoder = {
3745     "h263",
3746     CODEC_TYPE_VIDEO,
3747     CODEC_ID_H263,
3748     sizeof(MpegEncContext),
3749     MPV_encode_init,
3750     MPV_encode_picture,
3751     MPV_encode_end,
3752     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3753     .long_name= NULL_IF_CONFIG_SMALL("H.263"),
3754 };
3755
3756 AVCodec h263p_encoder = {
3757     "h263p",
3758     CODEC_TYPE_VIDEO,
3759     CODEC_ID_H263P,
3760     sizeof(MpegEncContext),
3761     MPV_encode_init,
3762     MPV_encode_picture,
3763     MPV_encode_end,
3764     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3765     .long_name= NULL_IF_CONFIG_SMALL("H.263+ / H.263 version 2"),
3766 };
3767
3768 AVCodec flv_encoder = {
3769     "flv",
3770     CODEC_TYPE_VIDEO,
3771     CODEC_ID_FLV1,
3772     sizeof(MpegEncContext),
3773     MPV_encode_init,
3774     MPV_encode_picture,
3775     MPV_encode_end,
3776     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3777     .long_name= NULL_IF_CONFIG_SMALL("Flash Video"),
3778 };
3779
3780 AVCodec rv10_encoder = {
3781     "rv10",
3782     CODEC_TYPE_VIDEO,
3783     CODEC_ID_RV10,
3784     sizeof(MpegEncContext),
3785     MPV_encode_init,
3786     MPV_encode_picture,
3787     MPV_encode_end,
3788     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3789     .long_name= NULL_IF_CONFIG_SMALL("RealVideo 1.0"),
3790 };
3791
3792 AVCodec rv20_encoder = {
3793     "rv20",
3794     CODEC_TYPE_VIDEO,
3795     CODEC_ID_RV20,
3796     sizeof(MpegEncContext),
3797     MPV_encode_init,
3798     MPV_encode_picture,
3799     MPV_encode_end,
3800     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3801     .long_name= NULL_IF_CONFIG_SMALL("RealVideo 2.0"),
3802 };
3803
3804 AVCodec mpeg4_encoder = {
3805     "mpeg4",
3806     CODEC_TYPE_VIDEO,
3807     CODEC_ID_MPEG4,
3808     sizeof(MpegEncContext),
3809     MPV_encode_init,
3810     MPV_encode_picture,
3811     MPV_encode_end,
3812     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3813     .capabilities= CODEC_CAP_DELAY,
3814     .long_name= NULL_IF_CONFIG_SMALL("MPEG-4 part 2"),
3815 };
3816
3817 AVCodec msmpeg4v1_encoder = {
3818     "msmpeg4v1",
3819     CODEC_TYPE_VIDEO,
3820     CODEC_ID_MSMPEG4V1,
3821     sizeof(MpegEncContext),
3822     MPV_encode_init,
3823     MPV_encode_picture,
3824     MPV_encode_end,
3825     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3826     .long_name= NULL_IF_CONFIG_SMALL("MPEG-4 part 2 Microsoft variant version 1"),
3827 };
3828
3829 AVCodec msmpeg4v2_encoder = {
3830     "msmpeg4v2",
3831     CODEC_TYPE_VIDEO,
3832     CODEC_ID_MSMPEG4V2,
3833     sizeof(MpegEncContext),
3834     MPV_encode_init,
3835     MPV_encode_picture,
3836     MPV_encode_end,
3837     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3838     .long_name= NULL_IF_CONFIG_SMALL("MPEG-4 part 2 Microsoft variant version 2"),
3839 };
3840
3841 AVCodec msmpeg4v3_encoder = {
3842     "msmpeg4",
3843     CODEC_TYPE_VIDEO,
3844     CODEC_ID_MSMPEG4V3,
3845     sizeof(MpegEncContext),
3846     MPV_encode_init,
3847     MPV_encode_picture,
3848     MPV_encode_end,
3849     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3850     .long_name= NULL_IF_CONFIG_SMALL("MPEG-4 part 2 Microsoft variant version 3"),
3851 };
3852
3853 AVCodec wmv1_encoder = {
3854     "wmv1",
3855     CODEC_TYPE_VIDEO,
3856     CODEC_ID_WMV1,
3857     sizeof(MpegEncContext),
3858     MPV_encode_init,
3859     MPV_encode_picture,
3860     MPV_encode_end,
3861     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3862     .long_name= NULL_IF_CONFIG_SMALL("Windows Media Video 7"),
3863 };