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

   1 /*
   2  * The simplest mpeg encoder (well, it was the simplest!)
   3  * Copyright (c) 2000,2001 Fabrice Bellard.
   4  * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
   5  *
   6  * 4MV & hq & B-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
   7  *
   8  * This file is part of FFmpeg.
   9  *
  10  * FFmpeg is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU Lesser General Public
  12  * License as published by the Free Software Foundation; either
  13  * version 2.1 of the License, or (at your option) any later version.
  14  *
  15  * FFmpeg is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  * Lesser General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU Lesser General Public
  21  * License along with FFmpeg; if not, write to the Free Software
  22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  23  */
  24
  25 /**
  26  * @file mpegvideo_enc.c
  27  * The simplest mpeg encoder (well, it was the simplest!).
  28  */
  29
  30 #include "avcodec.h"
  31 #include "dsputil.h"
  32 #include "mpegvideo.h"
  33 #include "mpegvideo_common.h"
  34 #include "mjpegenc.h"
  35 #include "msmpeg4.h"
  36 #include "h263.h"
  37 #include "faandct.h"
  38 #include <limits.h>
  39
  40 //#undef NDEBUG
  41 //#include <assert.h>
  42
  43 static int encode_picture(MpegEncContext *s, int picture_number);
  44 static int dct_quantize_refine(MpegEncContext *s, DCTELEM *block, int16_t *weight, DCTELEM *orig, int n, int qscale);
  45 static int sse_mb(MpegEncContext *s);
  46
  47 /* enable all paranoid tests for rounding, overflows, etc... */
  48 //#define PARANOID
  49
  50 //#define DEBUG
  51
  52 static const uint16_t aanscales[64] = {
  53     /* precomputed values scaled up by 14 bits */
  54     16384, 22725, 21407, 19266, 16384, 12873,  8867,  4520,
  55     22725, 31521, 29692, 26722, 22725, 17855, 12299,  6270,
  56     21407, 29692, 27969, 25172, 21407, 16819, 11585,  5906,
  57     19266, 26722, 25172, 22654, 19266, 15137, 10426,  5315,
  58     16384, 22725, 21407, 19266, 16384, 12873,  8867,  4520,
  59     12873, 17855, 16819, 15137, 12873, 10114,  6967,  3552,
  60     8867 , 12299, 11585, 10426,  8867,  6967,  4799,  2446,
  61     4520 ,  6270,  5906,  5315,  4520,  3552,  2446,  1247
  62 };
  63
  64 static uint8_t default_mv_penalty[MAX_FCODE+1][MAX_MV*2+1];
  65 static uint8_t default_fcode_tab[MAX_MV*2+1];
  66
  67 enum PixelFormat ff_yuv420p_list[2]= {PIX_FMT_YUV420P, -1};
  68
  69 void ff_convert_matrix(DSPContext *dsp, int (*qmat)[64], uint16_t (*qmat16)[2][64],
  70                            const uint16_t *quant_matrix, int bias, int qmin, int qmax, int intra)
  71 {
  72     int qscale;
  73     int shift=0;
  74
  75     for(qscale=qmin; qscale<=qmax; qscale++){
  76         int i;
  77         if (dsp->fdct == ff_jpeg_fdct_islow
  78 #ifdef FAAN_POSTSCALE
  79             || dsp->fdct == ff_faandct
  80 #endif
  81             ) {
  82             for(i=0;i<64;i++) {
  83                 const int j= dsp->idct_permutation[i];
  84                 /* 16 <= qscale * quant_matrix[i] <= 7905 */
  85                 /* 19952         <= aanscales[i] * qscale * quant_matrix[i]           <= 249205026 */
  86                 /* (1<<36)/19952 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */
  87                 /* 3444240       >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= 275 */
  88
  89                 qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) /
  90                                 (qscale * quant_matrix[j]));
  91             }
  92         } else if (dsp->fdct == fdct_ifast
  93 #ifndef FAAN_POSTSCALE
  94                    || dsp->fdct == ff_faandct
  95 #endif
  96                    ) {
  97             for(i=0;i<64;i++) {
  98                 const int j= dsp->idct_permutation[i];
  99                 /* 16 <= qscale * quant_matrix[i] <= 7905 */
 100                 /* 19952         <= aanscales[i] * qscale * quant_matrix[i]           <= 249205026 */
 101                 /* (1<<36)/19952 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */
 102                 /* 3444240       >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= 275 */
 103
 104                 qmat[qscale][i] = (int)((UINT64_C(1) << (QMAT_SHIFT + 14)) /
 105                                 (aanscales[i] * qscale * quant_matrix[j]));
 106             }
 107         } else {
 108             for(i=0;i<64;i++) {
 109                 const int j= dsp->idct_permutation[i];
 110                 /* We can safely suppose that 16 <= quant_matrix[i] <= 255
 111                    So 16           <= qscale * quant_matrix[i]             <= 7905
 112                    so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905
 113                    so 32768        >= (1<<19) / (qscale * quant_matrix[i]) >= 67
 114                 */
 115                 qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) / (qscale * quant_matrix[j]));
 116 //                qmat  [qscale][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]);
 117                 qmat16[qscale][0][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[j]);
 118
 119                 if(qmat16[qscale][0][i]==0 || qmat16[qscale][0][i]==128*256) qmat16[qscale][0][i]=128*256-1;
 120                 qmat16[qscale][1][i]= ROUNDED_DIV(bias<<(16-QUANT_BIAS_SHIFT), qmat16[qscale][0][i]);
 121             }
 122         }
 123
 124         for(i=intra; i<64; i++){
 125             int64_t max= 8191;
 126             if (dsp->fdct == fdct_ifast
 127 #ifndef FAAN_POSTSCALE
 128                    || dsp->fdct == ff_faandct
 129 #endif
 130                    ) {
 131                 max= (8191LL*aanscales[i]) >> 14;
 132             }
 133             while(((max * qmat[qscale][i]) >> shift) > INT_MAX){
 134                 shift++;
 135             }
 136         }
 137     }
 138     if(shift){
 139         av_log(NULL, AV_LOG_INFO, "Warning, QMAT_SHIFT is larger than %d, overflows possible\n", QMAT_SHIFT - shift);
 140     }
 141 }
 142
 143 static inline void update_qscale(MpegEncContext *s){
 144     s->qscale= (s->lambda*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7);
 145     s->qscale= av_clip(s->qscale, s->avctx->qmin, s->avctx->qmax);
 146
 147     s->lambda2= (s->lambda*s->lambda + FF_LAMBDA_SCALE/2) >> FF_LAMBDA_SHIFT;
 148 }
 149
 150 void ff_write_quant_matrix(PutBitContext *pb, uint16_t *matrix){
 151     int i;
 152
 153     if(matrix){
 154         put_bits(pb, 1, 1);
 155         for(i=0;i<64;i++) {
 156             put_bits(pb, 8, matrix[ ff_zigzag_direct[i] ]);
 157         }
 158     }else
 159         put_bits(pb, 1, 0);
 160 }
 161
 162 static void copy_picture_attributes(MpegEncContext *s, AVFrame *dst, AVFrame *src){
 163     int i;
 164
 165     dst->pict_type              = src->pict_type;
 166     dst->quality                = src->quality;
 167     dst->coded_picture_number   = src->coded_picture_number;
 168     dst->display_picture_number = src->display_picture_number;
 169 //    dst->reference              = src->reference;
 170     dst->pts                    = src->pts;
 171     dst->interlaced_frame       = src->interlaced_frame;
 172     dst->top_field_first        = src->top_field_first;
 173
 174     if(s->avctx->me_threshold){
 175         if(!src->motion_val[0])
 176             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_val not set!\n");
 177         if(!src->mb_type)
 178             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.mb_type not set!\n");
 179         if(!src->ref_index[0])
 180             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.ref_index not set!\n");
 181         if(src->motion_subsample_log2 != dst->motion_subsample_log2)
 182             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_subsample_log2 doesn't match! (%d!=%d)\n",
 183             src->motion_subsample_log2, dst->motion_subsample_log2);
 184
 185         memcpy(dst->mb_type, src->mb_type, s->mb_stride * s->mb_height * sizeof(dst->mb_type[0]));
 186
 187         for(i=0; i<2; i++){
 188             int stride= ((16*s->mb_width )>>src->motion_subsample_log2) + 1;
 189             int height= ((16*s->mb_height)>>src->motion_subsample_log2);
 190
 191             if(src->motion_val[i] && src->motion_val[i] != dst->motion_val[i]){
 192                 memcpy(dst->motion_val[i], src->motion_val[i], 2*stride*height*sizeof(int16_t));
 193             }
 194             if(src->ref_index[i] && src->ref_index[i] != dst->ref_index[i]){
 195                 memcpy(dst->ref_index[i], src->ref_index[i], s->b8_stride*2*s->mb_height*sizeof(int8_t));
 196             }
 197         }
 198     }
 199 }
 200
 201 static void update_duplicate_context_after_me(MpegEncContext *dst, MpegEncContext *src){
 202 #define COPY(a) dst->a= src->a
 203     COPY(pict_type);
 204     COPY(current_picture);
 205     COPY(f_code);
 206     COPY(b_code);
 207     COPY(qscale);
 208     COPY(lambda);
 209     COPY(lambda2);
 210     COPY(picture_in_gop_number);
 211     COPY(gop_picture_number);
 212     COPY(frame_pred_frame_dct); //FIXME don't set in encode_header
 213     COPY(progressive_frame); //FIXME don't set in encode_header
 214     COPY(partitioned_frame); //FIXME don't set in encode_header
 215 #undef COPY
 216 }
 217
 218 /**
 219  * sets the given MpegEncContext to defaults for encoding.
 220  * the changed fields will not depend upon the prior state of the MpegEncContext.
 221  */
 222 static void MPV_encode_defaults(MpegEncContext *s){
 223     int i;
 224     MPV_common_defaults(s);
 225
 226     for(i=-16; i<16; i++){
 227         default_fcode_tab[i + MAX_MV]= 1;
 228     }
 229     s->me.mv_penalty= default_mv_penalty;
 230     s->fcode_tab= default_fcode_tab;
 231 }
 232
 233 /* init video encoder */
 234 int MPV_encode_init(AVCodecContext *avctx)
 235 {
 236     MpegEncContext *s = avctx->priv_data;
 237     int i;
 238     int chroma_h_shift, chroma_v_shift;
 239
 240     MPV_encode_defaults(s);
 241
 242     switch (avctx->codec_id) {
 243     case CODEC_ID_MPEG2VIDEO:
 244         if(avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P){
 245             av_log(avctx, AV_LOG_ERROR, "only YUV420 and YUV422 are supported\n");
 246             return -1;
 247         }
 248         break;
 249     case CODEC_ID_LJPEG:
 250     case CODEC_ID_MJPEG:
 251         if(avctx->pix_fmt != PIX_FMT_YUVJ420P && avctx->pix_fmt != PIX_FMT_YUVJ422P &&
 252            ((avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P) || avctx->strict_std_compliance>FF_COMPLIANCE_INOFFICIAL)){
 253             av_log(avctx, AV_LOG_ERROR, "colorspace not supported in jpeg\n");
 254             return -1;
 255         }
 256         break;
 257     default:
 258         if(avctx->pix_fmt != PIX_FMT_YUV420P){
 259             av_log(avctx, AV_LOG_ERROR, "only YUV420 is supported\n");
 260             return -1;
 261         }
 262     }
 263
 264     switch (avctx->pix_fmt) {
 265     case PIX_FMT_YUVJ422P:
 266     case PIX_FMT_YUV422P:
 267         s->chroma_format = CHROMA_422;
 268         break;
 269     case PIX_FMT_YUVJ420P:
 270     case PIX_FMT_YUV420P:
 271     default:
 272         s->chroma_format = CHROMA_420;
 273         break;
 274     }
 275
 276     s->bit_rate = avctx->bit_rate;
 277     s->width = avctx->width;
 278     s->height = avctx->height;
 279     if(avctx->gop_size > 600 && avctx->strict_std_compliance>FF_COMPLIANCE_EXPERIMENTAL){
 280         av_log(avctx, AV_LOG_ERROR, "Warning keyframe interval too large! reducing it ...\n");
 281         avctx->gop_size=600;
 282     }
 283     s->gop_size = avctx->gop_size;
 284     s->avctx = avctx;
 285     s->flags= avctx->flags;
 286     s->flags2= avctx->flags2;
 287     s->max_b_frames= avctx->max_b_frames;
 288     s->codec_id= avctx->codec->id;
 289     s->luma_elim_threshold  = avctx->luma_elim_threshold;
 290     s->chroma_elim_threshold= avctx->chroma_elim_threshold;
 291     s->strict_std_compliance= avctx->strict_std_compliance;
 292     s->data_partitioning= avctx->flags & CODEC_FLAG_PART;
 293     s->quarter_sample= (avctx->flags & CODEC_FLAG_QPEL)!=0;
 294     s->mpeg_quant= avctx->mpeg_quant;
 295     s->rtp_mode= !!avctx->rtp_payload_size;
 296     s->intra_dc_precision= avctx->intra_dc_precision;
 297     s->user_specified_pts = AV_NOPTS_VALUE;
 298
 299     if (s->gop_size <= 1) {
 300         s->intra_only = 1;
 301         s->gop_size = 12;
 302     } else {
 303         s->intra_only = 0;
 304     }
 305
 306     s->me_method = avctx->me_method;
 307
 308     /* Fixed QSCALE */
 309     s->fixed_qscale = !!(avctx->flags & CODEC_FLAG_QSCALE);
 310
 311     s->adaptive_quant= (   s->avctx->lumi_masking
 312                         || s->avctx->dark_masking
 313                         || s->avctx->temporal_cplx_masking
 314                         || s->avctx->spatial_cplx_masking
 315                         || s->avctx->p_masking
 316                         || s->avctx->border_masking
 317                         || (s->flags&CODEC_FLAG_QP_RD))
 318                        && !s->fixed_qscale;
 319
 320     s->obmc= !!(s->flags & CODEC_FLAG_OBMC);
 321     s->loop_filter= !!(s->flags & CODEC_FLAG_LOOP_FILTER);
 322     s->alternate_scan= !!(s->flags & CODEC_FLAG_ALT_SCAN);
 323     s->intra_vlc_format= !!(s->flags2 & CODEC_FLAG2_INTRA_VLC);
 324     s->q_scale_type= !!(s->flags2 & CODEC_FLAG2_NON_LINEAR_QUANT);
 325
 326     if(avctx->rc_max_rate && !avctx->rc_buffer_size){
 327         av_log(avctx, AV_LOG_ERROR, "a vbv buffer size is needed, for encoding with a maximum bitrate\n");
 328         return -1;
 329     }
 330
 331     if(avctx->rc_min_rate && avctx->rc_max_rate != avctx->rc_min_rate){
 332         av_log(avctx, AV_LOG_INFO, "Warning min_rate > 0 but min_rate != max_rate isn't recommended!\n");
 333     }
 334
 335     if(avctx->rc_min_rate && avctx->rc_min_rate > avctx->bit_rate){
 336         av_log(avctx, AV_LOG_ERROR, "bitrate below min bitrate\n");
 337         return -1;
 338     }
 339
 340     if(avctx->rc_max_rate && avctx->rc_max_rate < avctx->bit_rate){
 341         av_log(avctx, AV_LOG_INFO, "bitrate above max bitrate\n");
 342         return -1;
 343     }
 344
 345     if(avctx->rc_max_rate && avctx->rc_max_rate == avctx->bit_rate && avctx->rc_max_rate != avctx->rc_min_rate){
 346         av_log(avctx, AV_LOG_INFO, "impossible bitrate constraints, this will fail\n");
 347     }
 348
 349     if(avctx->rc_buffer_size && avctx->bit_rate*av_q2d(avctx->time_base) > avctx->rc_buffer_size){
 350         av_log(avctx, AV_LOG_ERROR, "VBV buffer too small for bitrate\n");
 351         return -1;
 352     }
 353
 354     if(avctx->bit_rate*av_q2d(avctx->time_base) > avctx->bit_rate_tolerance){
 355         av_log(avctx, AV_LOG_ERROR, "bitrate tolerance too small for bitrate\n");
 356         return -1;
 357     }
 358
 359     if(   s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate
 360        && (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO)
 361        && 90000LL * (avctx->rc_buffer_size-1) > s->avctx->rc_max_rate*0xFFFFLL){
 362
 363         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");
 364     }
 365
 366     if((s->flags & CODEC_FLAG_4MV) && s->codec_id != CODEC_ID_MPEG4
 367        && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P && s->codec_id != CODEC_ID_FLV1){
 368         av_log(avctx, AV_LOG_ERROR, "4MV not supported by codec\n");
 369         return -1;
 370     }
 371
 372     if(s->obmc && s->avctx->mb_decision != FF_MB_DECISION_SIMPLE){
 373         av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with simple mb decision\n");
 374         return -1;
 375     }
 376
 377     if(s->obmc && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P){
 378         av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with H263(+)\n");
 379         return -1;
 380     }
 381
 382     if(s->quarter_sample && s->codec_id != CODEC_ID_MPEG4){
 383         av_log(avctx, AV_LOG_ERROR, "qpel not supported by codec\n");
 384         return -1;
 385     }
 386
 387     if(s->data_partitioning && s->codec_id != CODEC_ID_MPEG4){
 388         av_log(avctx, AV_LOG_ERROR, "data partitioning not supported by codec\n");
 389         return -1;
 390     }
 391
 392     if(s->max_b_frames && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO){
 393         av_log(avctx, AV_LOG_ERROR, "b frames not supported by codec\n");
 394         return -1;
 395     }
 396
 397     if((s->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN))
 398        && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG2VIDEO){
 399         av_log(avctx, AV_LOG_ERROR, "interlacing not supported by codec\n");
 400         return -1;
 401     }
 402
 403     if(s->mpeg_quant && s->codec_id != CODEC_ID_MPEG4){ //FIXME mpeg2 uses that too
 404         av_log(avctx, AV_LOG_ERROR, "mpeg2 style quantization not supported by codec\n");
 405         return -1;
 406     }
 407
 408     if((s->flags & CODEC_FLAG_CBP_RD) && !(s->flags & CODEC_FLAG_TRELLIS_QUANT)){
 409         av_log(avctx, AV_LOG_ERROR, "CBP RD needs trellis quant\n");
 410         return -1;
 411     }
 412
 413     if((s->flags & CODEC_FLAG_QP_RD) && s->avctx->mb_decision != FF_MB_DECISION_RD){
 414         av_log(avctx, AV_LOG_ERROR, "QP RD needs mbd=2\n");
 415         return -1;
 416     }
 417
 418     if(s->avctx->scenechange_threshold < 1000000000 && (s->flags & CODEC_FLAG_CLOSED_GOP)){
 419         av_log(avctx, AV_LOG_ERROR, "closed gop with scene change detection arent supported yet, set threshold to 1000000000\n");
 420         return -1;
 421     }
 422
 423     if((s->flags2 & CODEC_FLAG2_INTRA_VLC) && s->codec_id != CODEC_ID_MPEG2VIDEO){
 424         av_log(avctx, AV_LOG_ERROR, "intra vlc table not supported by codec\n");
 425         return -1;
 426     }
 427
 428     if(s->flags & CODEC_FLAG_LOW_DELAY){
 429         if (s->codec_id != CODEC_ID_MPEG2VIDEO && s->codec_id != CODEC_ID_MPEG1VIDEO){
 430             av_log(avctx, AV_LOG_ERROR, "low delay forcing is only available for mpeg1/2\n");
 431             return -1;
 432         }
 433         if (s->max_b_frames != 0){
 434             av_log(avctx, AV_LOG_ERROR, "b frames cannot be used with low delay\n");
 435             return -1;
 436         }
 437     }
 438
 439     if(s->q_scale_type == 1){
 440         if(s->codec_id != CODEC_ID_MPEG2VIDEO){
 441             av_log(avctx, AV_LOG_ERROR, "non linear quant is only available for mpeg2\n");
 442             return -1;
 443         }
 444         if(avctx->qmax > 12){
 445             av_log(avctx, AV_LOG_ERROR, "non linear quant only supports qmax <= 12 currently\n");
 446             return -1;
 447         }
 448     }
 449
 450     if(s->avctx->thread_count > 1 && s->codec_id != CODEC_ID_MPEG4
 451        && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO
 452        && (s->codec_id != CODEC_ID_H263P || !(s->flags & CODEC_FLAG_H263P_SLICE_STRUCT))){
 453         av_log(avctx, AV_LOG_ERROR, "multi threaded encoding not supported by codec\n");
 454         return -1;
 455     }
 456
 457     if(s->avctx->thread_count > 1)
 458         s->rtp_mode= 1;
 459
 460     if(!avctx->time_base.den || !avctx->time_base.num){
 461         av_log(avctx, AV_LOG_ERROR, "framerate not set\n");
 462         return -1;
 463     }
 464
 465     i= (INT_MAX/2+128)>>8;
 466     if(avctx->me_threshold >= i){
 467         av_log(avctx, AV_LOG_ERROR, "me_threshold too large, max is %d\n", i - 1);
 468         return -1;
 469     }
 470     if(avctx->mb_threshold >= i){
 471         av_log(avctx, AV_LOG_ERROR, "mb_threshold too large, max is %d\n", i - 1);
 472         return -1;
 473     }
 474
 475     if(avctx->b_frame_strategy && (avctx->flags&CODEC_FLAG_PASS2)){
 476         av_log(avctx, AV_LOG_INFO, "notice: b_frame_strategy only affects the first pass\n");
 477         avctx->b_frame_strategy = 0;
 478     }
 479
 480     i= ff_gcd(avctx->time_base.den, avctx->time_base.num);
 481     if(i > 1){
 482         av_log(avctx, AV_LOG_INFO, "removing common factors from framerate\n");
 483         avctx->time_base.den /= i;
 484         avctx->time_base.num /= i;
 485 //        return -1;
 486     }
 487
 488     if(s->codec_id==CODEC_ID_MJPEG){
 489         s->intra_quant_bias= 1<<(QUANT_BIAS_SHIFT-1); //(a + x/2)/x
 490         s->inter_quant_bias= 0;
 491     }else if(s->mpeg_quant || s->codec_id==CODEC_ID_MPEG1VIDEO || s->codec_id==CODEC_ID_MPEG2VIDEO){
 492         s->intra_quant_bias= 3<<(QUANT_BIAS_SHIFT-3); //(a + x*3/8)/x
 493         s->inter_quant_bias= 0;
 494     }else{
 495         s->intra_quant_bias=0;
 496         s->inter_quant_bias=-(1<<(QUANT_BIAS_SHIFT-2)); //(a - x/4)/x
 497     }
 498
 499     if(avctx->intra_quant_bias != FF_DEFAULT_QUANT_BIAS)
 500         s->intra_quant_bias= avctx->intra_quant_bias;
 501     if(avctx->inter_quant_bias != FF_DEFAULT_QUANT_BIAS)
 502         s->inter_quant_bias= avctx->inter_quant_bias;
 503
 504     avcodec_get_chroma_sub_sample(avctx->pix_fmt, &chroma_h_shift, &chroma_v_shift);
 505
 506     if(avctx->codec_id == CODEC_ID_MPEG4 && s->avctx->time_base.den > (1<<16)-1){
 507         av_log(avctx, AV_LOG_ERROR, "timebase not supported by mpeg 4 standard\n");
 508         return -1;
 509     }
 510     s->time_increment_bits = av_log2(s->avctx->time_base.den - 1) + 1;
 511
 512     switch(avctx->codec->id) {
 513     case CODEC_ID_MPEG1VIDEO:
 514         s->out_format = FMT_MPEG1;
 515         s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY);
 516         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
 517         break;
 518     case CODEC_ID_MPEG2VIDEO:
 519         s->out_format = FMT_MPEG1;
 520         s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY);
 521         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
 522         s->rtp_mode= 1;
 523         break;
 524     case CODEC_ID_LJPEG:
 525     case CODEC_ID_MJPEG:
 526         s->out_format = FMT_MJPEG;
 527         s->intra_only = 1; /* force intra only for jpeg */
 528         s->mjpeg_vsample[0] = 2;
 529         s->mjpeg_vsample[1] = 2>>chroma_v_shift;
 530         s->mjpeg_vsample[2] = 2>>chroma_v_shift;
 531         s->mjpeg_hsample[0] = 2;
 532         s->mjpeg_hsample[1] = 2>>chroma_h_shift;
 533         s->mjpeg_hsample[2] = 2>>chroma_h_shift;
 534         if (!(ENABLE_MJPEG_ENCODER || ENABLE_LJPEG_ENCODER)
 535             || ff_mjpeg_encode_init(s) < 0)
 536             return -1;
 537         avctx->delay=0;
 538         s->low_delay=1;
 539         break;
 540     case CODEC_ID_H261:
 541         if (!ENABLE_H261_ENCODER)  return -1;
 542         if (ff_h261_get_picture_format(s->width, s->height) < 0) {
 543             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);
 544             return -1;
 545         }
 546         s->out_format = FMT_H261;
 547         avctx->delay=0;
 548         s->low_delay=1;
 549         break;
 550     case CODEC_ID_H263:
 551         if (!ENABLE_H263_ENCODER)  return -1;
 552         if (h263_get_picture_format(s->width, s->height) == 7) {
 553             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);
 554             return -1;
 555         }
 556         s->out_format = FMT_H263;
 557         s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
 558         avctx->delay=0;
 559         s->low_delay=1;
 560         break;
 561     case CODEC_ID_H263P:
 562         s->out_format = FMT_H263;
 563         s->h263_plus = 1;
 564         /* Fx */
 565         s->umvplus = (avctx->flags & CODEC_FLAG_H263P_UMV) ? 1:0;
 566         s->h263_aic= (avctx->flags & CODEC_FLAG_AC_PRED) ? 1:0;
 567         s->modified_quant= s->h263_aic;
 568         s->alt_inter_vlc= (avctx->flags & CODEC_FLAG_H263P_AIV) ? 1:0;
 569         s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
 570         s->loop_filter= (avctx->flags & CODEC_FLAG_LOOP_FILTER) ? 1:0;
 571         s->unrestricted_mv= s->obmc || s->loop_filter || s->umvplus;
 572         s->h263_slice_structured= (s->flags & CODEC_FLAG_H263P_SLICE_STRUCT) ? 1:0;
 573
 574         /* /Fx */
 575         /* These are just to be sure */
 576         avctx->delay=0;
 577         s->low_delay=1;
 578         break;
 579     case CODEC_ID_FLV1:
 580         s->out_format = FMT_H263;
 581         s->h263_flv = 2; /* format = 1; 11-bit codes */
 582         s->unrestricted_mv = 1;
 583         s->rtp_mode=0; /* don't allow GOB */
 584         avctx->delay=0;
 585         s->low_delay=1;
 586         break;
 587     case CODEC_ID_RV10:
 588         s->out_format = FMT_H263;
 589         avctx->delay=0;
 590         s->low_delay=1;
 591         break;
 592     case CODEC_ID_RV20:
 593         s->out_format = FMT_H263;
 594         avctx->delay=0;
 595         s->low_delay=1;
 596         s->modified_quant=1;
 597         s->h263_aic=1;
 598         s->h263_plus=1;
 599         s->loop_filter=1;
 600         s->unrestricted_mv= s->obmc || s->loop_filter || s->umvplus;
 601         break;
 602     case CODEC_ID_MPEG4:
 603         s->out_format = FMT_H263;
 604         s->h263_pred = 1;
 605         s->unrestricted_mv = 1;
 606         s->low_delay= s->max_b_frames ? 0 : 1;
 607         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
 608         break;
 609     case CODEC_ID_MSMPEG4V1:
 610         s->out_format = FMT_H263;
 611         s->h263_msmpeg4 = 1;
 612         s->h263_pred = 1;
 613         s->unrestricted_mv = 1;
 614         s->msmpeg4_version= 1;
 615         avctx->delay=0;
 616         s->low_delay=1;
 617         break;
 618     case CODEC_ID_MSMPEG4V2:
 619         s->out_format = FMT_H263;
 620         s->h263_msmpeg4 = 1;
 621         s->h263_pred = 1;
 622         s->unrestricted_mv = 1;
 623         s->msmpeg4_version= 2;
 624         avctx->delay=0;
 625         s->low_delay=1;
 626         break;
 627     case CODEC_ID_MSMPEG4V3:
 628         s->out_format = FMT_H263;
 629         s->h263_msmpeg4 = 1;
 630         s->h263_pred = 1;
 631         s->unrestricted_mv = 1;
 632         s->msmpeg4_version= 3;
 633         s->flipflop_rounding=1;
 634         avctx->delay=0;
 635         s->low_delay=1;
 636         break;
 637     case CODEC_ID_WMV1:
 638         s->out_format = FMT_H263;
 639         s->h263_msmpeg4 = 1;
 640         s->h263_pred = 1;
 641         s->unrestricted_mv = 1;
 642         s->msmpeg4_version= 4;
 643         s->flipflop_rounding=1;
 644         avctx->delay=0;
 645         s->low_delay=1;
 646         break;
 647     case CODEC_ID_WMV2:
 648         s->out_format = FMT_H263;
 649         s->h263_msmpeg4 = 1;
 650         s->h263_pred = 1;
 651         s->unrestricted_mv = 1;
 652         s->msmpeg4_version= 5;
 653         s->flipflop_rounding=1;
 654         avctx->delay=0;
 655         s->low_delay=1;
 656         break;
 657     default:
 658         return -1;
 659     }
 660
 661     avctx->has_b_frames= !s->low_delay;
 662
 663     s->encoding = 1;
 664
 665     /* init */
 666     if (MPV_common_init(s) < 0)
 667         return -1;
 668
 669     if(!s->dct_quantize)
 670         s->dct_quantize = dct_quantize_c;
 671     if(!s->denoise_dct)
 672         s->denoise_dct = denoise_dct_c;
 673     s->fast_dct_quantize = s->dct_quantize;
 674     if(s->flags & CODEC_FLAG_TRELLIS_QUANT)
 675         s->dct_quantize = dct_quantize_trellis_c;
 676
 677     if((ENABLE_H263P_ENCODER || ENABLE_RV20_ENCODER) && s->modified_quant)
 678         s->chroma_qscale_table= ff_h263_chroma_qscale_table;
 679     s->progressive_frame=
 680     s->progressive_sequence= !(avctx->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN));
 681     s->quant_precision=5;
 682
 683     ff_set_cmp(&s->dsp, s->dsp.ildct_cmp, s->avctx->ildct_cmp);
 684     ff_set_cmp(&s->dsp, s->dsp.frame_skip_cmp, s->avctx->frame_skip_cmp);
 685
 686     if (ENABLE_H261_ENCODER && s->out_format == FMT_H261)
 687         ff_h261_encode_init(s);
 688     if (ENABLE_ANY_H263_ENCODER && s->out_format == FMT_H263)
 689         h263_encode_init(s);
 690     if (ENABLE_MSMPEG4_ENCODER && s->msmpeg4_version)
 691         ff_msmpeg4_encode_init(s);
 692     if ((ENABLE_MPEG1VIDEO_ENCODER || ENABLE_MPEG2VIDEO_ENCODER)
 693         && s->out_format == FMT_MPEG1)
 694         ff_mpeg1_encode_init(s);
 695
 696     /* init q matrix */
 697     for(i=0;i<64;i++) {
 698         int j= s->dsp.idct_permutation[i];
 699         if(ENABLE_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4 && s->mpeg_quant){
 700             s->intra_matrix[j] = ff_mpeg4_default_intra_matrix[i];
 701             s->inter_matrix[j] = ff_mpeg4_default_non_intra_matrix[i];
 702         }else if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
 703             s->intra_matrix[j] =
 704             s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
 705         }else
 706         { /* mpeg1/2 */
 707             s->intra_matrix[j] = ff_mpeg1_default_intra_matrix[i];
 708             s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
 709         }
 710         if(s->avctx->intra_matrix)
 711             s->intra_matrix[j] = s->avctx->intra_matrix[i];
 712         if(s->avctx->inter_matrix)
 713             s->inter_matrix[j] = s->avctx->inter_matrix[i];
 714     }
 715
 716     /* precompute matrix */
 717     /* for mjpeg, we do include qscale in the matrix */
 718     if (s->out_format != FMT_MJPEG) {
 719         ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
 720                        s->intra_matrix, s->intra_quant_bias, avctx->qmin, 31, 1);
 721         ff_convert_matrix(&s->dsp, s->q_inter_matrix, s->q_inter_matrix16,
 722                        s->inter_matrix, s->inter_quant_bias, avctx->qmin, 31, 0);
 723     }
 724
 725     if(ff_rate_control_init(s) < 0)
 726         return -1;
 727
 728     return 0;
 729 }
 730
 731 int MPV_encode_end(AVCodecContext *avctx)
 732 {
 733     MpegEncContext *s = avctx->priv_data;
 734
 735     ff_rate_control_uninit(s);
 736
 737     MPV_common_end(s);
 738     if ((ENABLE_MJPEG_ENCODER || ENABLE_LJPEG_ENCODER) && s->out_format == FMT_MJPEG)
 739         ff_mjpeg_encode_close(s);
 740
 741     av_freep(&avctx->extradata);
 742
 743     return 0;
 744 }
 745
 746 static int get_sae(uint8_t *src, int ref, int stride){
 747     int x,y;
 748     int acc=0;
 749
 750     for(y=0; y<16; y++){
 751         for(x=0; x<16; x++){
 752             acc+= FFABS(src[x+y*stride] - ref);
 753         }
 754     }
 755
 756     return acc;
 757 }
 758
 759 static int get_intra_count(MpegEncContext *s, uint8_t *src, uint8_t *ref, int stride){
 760     int x, y, w, h;
 761     int acc=0;
 762
 763     w= s->width &~15;
 764     h= s->height&~15;
 765
 766     for(y=0; y<h; y+=16){
 767         for(x=0; x<w; x+=16){
 768             int offset= x + y*stride;
 769             int sad = s->dsp.sad[0](NULL, src + offset, ref + offset, stride, 16);
 770             int mean= (s->dsp.pix_sum(src + offset, stride) + 128)>>8;
 771             int sae = get_sae(src + offset, mean, stride);
 772
 773             acc+= sae + 500 < sad;
 774         }
 775     }
 776     return acc;
 777 }
 778
 779
 780 static int load_input_picture(MpegEncContext *s, AVFrame *pic_arg){
 781     AVFrame *pic=NULL;
 782     int64_t pts;
 783     int i;
 784     const int encoding_delay= s->max_b_frames;
 785     int direct=1;
 786
 787     if(pic_arg){
 788         pts= pic_arg->pts;
 789         pic_arg->display_picture_number= s->input_picture_number++;
 790
 791         if(pts != AV_NOPTS_VALUE){
 792             if(s->user_specified_pts != AV_NOPTS_VALUE){
 793                 int64_t time= pts;
 794                 int64_t last= s->user_specified_pts;
 795
 796                 if(time <= last){
 797                     av_log(s->avctx, AV_LOG_ERROR, "Error, Invalid timestamp=%"PRId64", last=%"PRId64"\n", pts, s->user_specified_pts);
 798                     return -1;
 799                 }
 800             }
 801             s->user_specified_pts= pts;
 802         }else{
 803             if(s->user_specified_pts != AV_NOPTS_VALUE){
 804                 s->user_specified_pts=
 805                 pts= s->user_specified_pts + 1;
 806                 av_log(s->avctx, AV_LOG_INFO, "Warning: AVFrame.pts=? trying to guess (%"PRId64")\n", pts);
 807             }else{
 808                 pts= pic_arg->display_picture_number;
 809             }
 810         }
 811     }
 812
 813   if(pic_arg){
 814     if(encoding_delay && !(s->flags&CODEC_FLAG_INPUT_PRESERVED)) direct=0;
 815     if(pic_arg->linesize[0] != s->linesize) direct=0;
 816     if(pic_arg->linesize[1] != s->uvlinesize) direct=0;
 817     if(pic_arg->linesize[2] != s->uvlinesize) direct=0;
 818
 819 //    av_log(AV_LOG_DEBUG, "%d %d %d %d\n",pic_arg->linesize[0], pic_arg->linesize[1], s->linesize, s->uvlinesize);
 820
 821     if(direct){
 822         i= ff_find_unused_picture(s, 1);
 823
 824         pic= (AVFrame*)&s->picture[i];
 825         pic->reference= 3;
 826
 827         for(i=0; i<4; i++){
 828             pic->data[i]= pic_arg->data[i];
 829             pic->linesize[i]= pic_arg->linesize[i];
 830         }
 831         alloc_picture(s, (Picture*)pic, 1);
 832     }else{
 833         i= ff_find_unused_picture(s, 0);
 834
 835         pic= (AVFrame*)&s->picture[i];
 836         pic->reference= 3;
 837
 838         alloc_picture(s, (Picture*)pic, 0);
 839
 840         if(   pic->data[0] + INPLACE_OFFSET == pic_arg->data[0]
 841            && pic->data[1] + INPLACE_OFFSET == pic_arg->data[1]
 842            && pic->data[2] + INPLACE_OFFSET == pic_arg->data[2]){
 843        // empty
 844         }else{
 845             int h_chroma_shift, v_chroma_shift;
 846             avcodec_get_chroma_sub_sample(s->avctx->pix_fmt, &h_chroma_shift, &v_chroma_shift);
 847
 848             for(i=0; i<3; i++){
 849                 int src_stride= pic_arg->linesize[i];
 850                 int dst_stride= i ? s->uvlinesize : s->linesize;
 851                 int h_shift= i ? h_chroma_shift : 0;
 852                 int v_shift= i ? v_chroma_shift : 0;
 853                 int w= s->width >>h_shift;
 854                 int h= s->height>>v_shift;
 855                 uint8_t *src= pic_arg->data[i];
 856                 uint8_t *dst= pic->data[i];
 857
 858                 if(!s->avctx->rc_buffer_size)
 859                     dst +=INPLACE_OFFSET;
 860
 861                 if(src_stride==dst_stride)
 862                     memcpy(dst, src, src_stride*h);
 863                 else{
 864                     while(h--){
 865                         memcpy(dst, src, w);
 866                         dst += dst_stride;
 867                         src += src_stride;
 868                     }
 869                 }
 870             }
 871         }
 872     }
 873     copy_picture_attributes(s, pic, pic_arg);
 874     pic->pts= pts; //we set this here to avoid modifiying pic_arg
 875   }
 876
 877     /* shift buffer entries */
 878     for(i=1; i<MAX_PICTURE_COUNT /*s->encoding_delay+1*/; i++)
 879         s->input_picture[i-1]= s->input_picture[i];
 880
 881     s->input_picture[encoding_delay]= (Picture*)pic;
 882
 883     return 0;
 884 }
 885
 886 static int skip_check(MpegEncContext *s, Picture *p, Picture *ref){
 887     int x, y, plane;
 888     int score=0;
 889     int64_t score64=0;
 890
 891     for(plane=0; plane<3; plane++){
 892         const int stride= p->linesize[plane];
 893         const int bw= plane ? 1 : 2;
 894         for(y=0; y<s->mb_height*bw; y++){
 895             for(x=0; x<s->mb_width*bw; x++){
 896                 int off= p->type == FF_BUFFER_TYPE_SHARED ? 0: 16;
 897                 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);
 898
 899                 switch(s->avctx->frame_skip_exp){
 900                     case 0: score= FFMAX(score, v); break;
 901                     case 1: score+= FFABS(v);break;
 902                     case 2: score+= v*v;break;
 903                     case 3: score64+= FFABS(v*v*(int64_t)v);break;
 904                     case 4: score64+= v*v*(int64_t)(v*v);break;
 905                 }
 906             }
 907         }
 908     }
 909
 910     if(score) score64= score;
 911
 912     if(score64 < s->avctx->frame_skip_threshold)
 913         return 1;
 914     if(score64 < ((s->avctx->frame_skip_factor * (int64_t)s->lambda)>>8))
 915         return 1;
 916     return 0;
 917 }
 918
 919 static int estimate_best_b_count(MpegEncContext *s){
 920     AVCodec *codec= avcodec_find_encoder(s->avctx->codec_id);
 921     AVCodecContext *c= avcodec_alloc_context();
 922     AVFrame input[FF_MAX_B_FRAMES+2];
 923     const int scale= s->avctx->brd_scale;
 924     int i, j, out_size, p_lambda, b_lambda, lambda2;
 925     int outbuf_size= s->width * s->height; //FIXME
 926     uint8_t *outbuf= av_malloc(outbuf_size);
 927     int64_t best_rd= INT64_MAX;
 928     int best_b_count= -1;
 929
 930     assert(scale>=0 && scale <=3);
 931
 932 //    emms_c();
 933     p_lambda= s->last_lambda_for[P_TYPE]; //s->next_picture_ptr->quality;
 934     b_lambda= s->last_lambda_for[B_TYPE]; //p_lambda *FFABS(s->avctx->b_quant_factor) + s->avctx->b_quant_offset;
 935     if(!b_lambda) b_lambda= p_lambda; //FIXME we should do this somewhere else
 936     lambda2= (b_lambda*b_lambda + (1<<FF_LAMBDA_SHIFT)/2 ) >> FF_LAMBDA_SHIFT;
 937
 938     c->width = s->width >> scale;
 939     c->height= s->height>> scale;
 940     c->flags= CODEC_FLAG_QSCALE | CODEC_FLAG_PSNR | CODEC_FLAG_INPUT_PRESERVED /*| CODEC_FLAG_EMU_EDGE*/;
 941     c->flags|= s->avctx->flags & CODEC_FLAG_QPEL;
 942     c->mb_decision= s->avctx->mb_decision;
 943     c->me_cmp= s->avctx->me_cmp;
 944     c->mb_cmp= s->avctx->mb_cmp;
 945     c->me_sub_cmp= s->avctx->me_sub_cmp;
 946     c->pix_fmt = PIX_FMT_YUV420P;
 947     c->time_base= s->avctx->time_base;
 948     c->max_b_frames= s->max_b_frames;
 949
 950     if (avcodec_open(c, codec) < 0)
 951         return -1;
 952
 953     for(i=0; i<s->max_b_frames+2; i++){
 954         int ysize= c->width*c->height;
 955         int csize= (c->width/2)*(c->height/2);
 956         Picture pre_input, *pre_input_ptr= i ? s->input_picture[i-1] : s->next_picture_ptr;
 957
 958         avcodec_get_frame_defaults(&input[i]);
 959         input[i].data[0]= av_malloc(ysize + 2*csize);
 960         input[i].data[1]= input[i].data[0] + ysize;
 961         input[i].data[2]= input[i].data[1] + csize;
 962         input[i].linesize[0]= c->width;
 963         input[i].linesize[1]=
 964         input[i].linesize[2]= c->width/2;
 965
 966         if(pre_input_ptr && (!i || s->input_picture[i-1])) {
 967             pre_input= *pre_input_ptr;
 968
 969             if(pre_input.type != FF_BUFFER_TYPE_SHARED && i) {
 970                 pre_input.data[0]+=INPLACE_OFFSET;
 971                 pre_input.data[1]+=INPLACE_OFFSET;
 972                 pre_input.data[2]+=INPLACE_OFFSET;
 973             }
 974
 975             s->dsp.shrink[scale](input[i].data[0], input[i].linesize[0], pre_input.data[0], pre_input.linesize[0], c->width, c->height);
 976             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);
 977             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);
 978         }
 979     }
 980
 981     for(j=0; j<s->max_b_frames+1; j++){
 982         int64_t rd=0;
 983
 984         if(!s->input_picture[j])
 985             break;
 986
 987         c->error[0]= c->error[1]= c->error[2]= 0;
 988
 989         input[0].pict_type= I_TYPE;
 990         input[0].quality= 1 * FF_QP2LAMBDA;
 991         out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[0]);
 992 //        rd += (out_size * lambda2) >> FF_LAMBDA_SHIFT;
 993
 994         for(i=0; i<s->max_b_frames+1; i++){
 995             int is_p= i % (j+1) == j || i==s->max_b_frames;
 996
 997             input[i+1].pict_type= is_p ? P_TYPE : B_TYPE;
 998             input[i+1].quality= is_p ? p_lambda : b_lambda;
 999             out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[i+1]);
1000             rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
1001         }
1002
1003         /* get the delayed frames */
1004         while(out_size){
1005             out_size = avcodec_encode_video(c, outbuf, outbuf_size, NULL);
1006             rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
1007         }
1008
1009         rd += c->error[0] + c->error[1] + c->error[2];
1010
1011         if(rd < best_rd){
1012             best_rd= rd;
1013             best_b_count= j;
1014         }
1015     }
1016
1017     av_freep(&outbuf);
1018     avcodec_close(c);
1019     av_freep(&c);
1020
1021     for(i=0; i<s->max_b_frames+2; i++){
1022         av_freep(&input[i].data[0]);
1023     }
1024
1025     return best_b_count;
1026 }
1027
1028 static void select_input_picture(MpegEncContext *s){
1029     int i;
1030
1031     for(i=1; i<MAX_PICTURE_COUNT; i++)
1032         s->reordered_input_picture[i-1]= s->reordered_input_picture[i];
1033     s->reordered_input_picture[MAX_PICTURE_COUNT-1]= NULL;
1034
1035     /* set next picture type & ordering */
1036     if(s->reordered_input_picture[0]==NULL && s->input_picture[0]){
1037         if(/*s->picture_in_gop_number >= s->gop_size ||*/ s->next_picture_ptr==NULL || s->intra_only){
1038             s->reordered_input_picture[0]= s->input_picture[0];
1039             s->reordered_input_picture[0]->pict_type= I_TYPE;
1040             s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++;
1041         }else{
1042             int b_frames;
1043
1044             if(s->avctx->frame_skip_threshold || s->avctx->frame_skip_factor){
1045                 if(s->picture_in_gop_number < s->gop_size && skip_check(s, s->input_picture[0], s->next_picture_ptr)){
1046                 //FIXME check that te gop check above is +-1 correct
1047 //av_log(NULL, AV_LOG_DEBUG, "skip %p %"PRId64"\n", s->input_picture[0]->data[0], s->input_picture[0]->pts);
1048
1049                     if(s->input_picture[0]->type == FF_BUFFER_TYPE_SHARED){
1050                         for(i=0; i<4; i++)
1051                             s->input_picture[0]->data[i]= NULL;
1052                         s->input_picture[0]->type= 0;
1053                     }else{
1054                         assert(   s->input_picture[0]->type==FF_BUFFER_TYPE_USER
1055                                || s->input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
1056
1057                         s->avctx->release_buffer(s->avctx, (AVFrame*)s->input_picture[0]);
1058                     }
1059
1060                     emms_c();
1061                     ff_vbv_update(s, 0);
1062
1063                     goto no_output_pic;
1064                 }
1065             }
1066
1067             if(s->flags&CODEC_FLAG_PASS2){
1068                 for(i=0; i<s->max_b_frames+1; i++){
1069                     int pict_num= s->input_picture[0]->display_picture_number + i;
1070
1071                     if(pict_num >= s->rc_context.num_entries)
1072                         break;
1073                     if(!s->input_picture[i]){
1074                         s->rc_context.entry[pict_num-1].new_pict_type = P_TYPE;
1075                         break;
1076                     }
1077
1078                     s->input_picture[i]->pict_type=
1079                         s->rc_context.entry[pict_num].new_pict_type;
1080                 }
1081             }
1082
1083             if(s->avctx->b_frame_strategy==0){
1084                 b_frames= s->max_b_frames;
1085                 while(b_frames && !s->input_picture[b_frames]) b_frames--;
1086             }else if(s->avctx->b_frame_strategy==1){
1087                 for(i=1; i<s->max_b_frames+1; i++){
1088                     if(s->input_picture[i] && s->input_picture[i]->b_frame_score==0){
1089                         s->input_picture[i]->b_frame_score=
1090                             get_intra_count(s, s->input_picture[i  ]->data[0],
1091                                                s->input_picture[i-1]->data[0], s->linesize) + 1;
1092                     }
1093                 }
1094                 for(i=0; i<s->max_b_frames+1; i++){
1095                     if(s->input_picture[i]==NULL || s->input_picture[i]->b_frame_score - 1 > s->mb_num/s->avctx->b_sensitivity) break;
1096                 }
1097
1098                 b_frames= FFMAX(0, i-1);
1099
1100                 /* reset scores */
1101                 for(i=0; i<b_frames+1; i++){
1102                     s->input_picture[i]->b_frame_score=0;
1103                 }
1104             }else if(s->avctx->b_frame_strategy==2){
1105                 b_frames= estimate_best_b_count(s);
1106             }else{
1107                 av_log(s->avctx, AV_LOG_ERROR, "illegal b frame strategy\n");
1108                 b_frames=0;
1109             }
1110
1111             emms_c();
1112 //static int b_count=0;
1113 //b_count+= b_frames;
1114 //av_log(s->avctx, AV_LOG_DEBUG, "b_frames: %d\n", b_count);
1115
1116             for(i= b_frames - 1; i>=0; i--){
1117                 int type= s->input_picture[i]->pict_type;
1118                 if(type && type != B_TYPE)
1119                     b_frames= i;
1120             }
1121             if(s->input_picture[b_frames]->pict_type == B_TYPE && b_frames == s->max_b_frames){
1122                 av_log(s->avctx, AV_LOG_ERROR, "warning, too many b frames in a row\n");
1123             }
1124
1125             if(s->picture_in_gop_number + b_frames >= s->gop_size){
1126               if((s->flags2 & CODEC_FLAG2_STRICT_GOP) && s->gop_size > s->picture_in_gop_number){
1127                     b_frames= s->gop_size - s->picture_in_gop_number - 1;
1128               }else{
1129                 if(s->flags & CODEC_FLAG_CLOSED_GOP)
1130                     b_frames=0;
1131                 s->input_picture[b_frames]->pict_type= I_TYPE;
1132               }
1133             }
1134
1135             if(   (s->flags & CODEC_FLAG_CLOSED_GOP)
1136                && b_frames
1137                && s->input_picture[b_frames]->pict_type== I_TYPE)
1138                 b_frames--;
1139
1140             s->reordered_input_picture[0]= s->input_picture[b_frames];
1141             if(s->reordered_input_picture[0]->pict_type != I_TYPE)
1142                 s->reordered_input_picture[0]->pict_type= P_TYPE;
1143             s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++;
1144             for(i=0; i<b_frames; i++){
1145                 s->reordered_input_picture[i+1]= s->input_picture[i];
1146                 s->reordered_input_picture[i+1]->pict_type= B_TYPE;
1147                 s->reordered_input_picture[i+1]->coded_picture_number= s->coded_picture_number++;
1148             }
1149         }
1150     }
1151 no_output_pic:
1152     if(s->reordered_input_picture[0]){
1153         s->reordered_input_picture[0]->reference= s->reordered_input_picture[0]->pict_type!=B_TYPE ? 3 : 0;
1154
1155         copy_picture(&s->new_picture, s->reordered_input_picture[0]);
1156
1157         if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_SHARED || s->avctx->rc_buffer_size){
1158             // input is a shared pix, so we can't modifiy it -> alloc a new one & ensure that the shared one is reuseable
1159
1160             int i= ff_find_unused_picture(s, 0);
1161             Picture *pic= &s->picture[i];
1162
1163             pic->reference              = s->reordered_input_picture[0]->reference;
1164             alloc_picture(s, pic, 0);
1165
1166             /* mark us unused / free shared pic */
1167             if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_INTERNAL)
1168                 s->avctx->release_buffer(s->avctx, (AVFrame*)s->reordered_input_picture[0]);
1169             for(i=0; i<4; i++)
1170                 s->reordered_input_picture[0]->data[i]= NULL;
1171             s->reordered_input_picture[0]->type= 0;
1172
1173             copy_picture_attributes(s, (AVFrame*)pic, (AVFrame*)s->reordered_input_picture[0]);
1174
1175             s->current_picture_ptr= pic;
1176         }else{
1177             // input is not a shared pix -> reuse buffer for current_pix
1178
1179             assert(   s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_USER
1180                    || s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
1181
1182             s->current_picture_ptr= s->reordered_input_picture[0];
1183             for(i=0; i<4; i++){
1184                 s->new_picture.data[i]+= INPLACE_OFFSET;
1185             }
1186         }
1187         copy_picture(&s->current_picture, s->current_picture_ptr);
1188
1189         s->picture_number= s->new_picture.display_picture_number;
1190 //printf("dpn:%d\n", s->picture_number);
1191     }else{
1192        memset(&s->new_picture, 0, sizeof(Picture));
1193     }
1194 }
1195
1196 int MPV_encode_picture(AVCodecContext *avctx,
1197                        unsigned char *buf, int buf_size, void *data)
1198 {
1199     MpegEncContext *s = avctx->priv_data;
1200     AVFrame *pic_arg = data;
1201     int i, stuffing_count;
1202
1203     for(i=0; i<avctx->thread_count; i++){
1204         int start_y= s->thread_context[i]->start_mb_y;
1205         int   end_y= s->thread_context[i]->  end_mb_y;
1206         int h= s->mb_height;
1207         uint8_t *start= buf + (size_t)(((int64_t) buf_size)*start_y/h);
1208         uint8_t *end  = buf + (size_t)(((int64_t) buf_size)*  end_y/h);
1209
1210         init_put_bits(&s->thread_context[i]->pb, start, end - start);
1211     }
1212
1213     s->picture_in_gop_number++;
1214
1215     if(load_input_picture(s, pic_arg) < 0)
1216         return -1;
1217
1218     select_input_picture(s);
1219
1220     /* output? */
1221     if(s->new_picture.data[0]){
1222         s->pict_type= s->new_picture.pict_type;
1223 //emms_c();
1224 //printf("qs:%f %f %d\n", s->new_picture.quality, s->current_picture.quality, s->qscale);
1225         MPV_frame_start(s, avctx);
1226 vbv_retry:
1227         if (encode_picture(s, s->picture_number) < 0)
1228             return -1;
1229
1230         avctx->real_pict_num  = s->picture_number;
1231         avctx->header_bits = s->header_bits;
1232         avctx->mv_bits     = s->mv_bits;
1233         avctx->misc_bits   = s->misc_bits;
1234         avctx->i_tex_bits  = s->i_tex_bits;
1235         avctx->p_tex_bits  = s->p_tex_bits;
1236         avctx->i_count     = s->i_count;
1237         avctx->p_count     = s->mb_num - s->i_count - s->skip_count; //FIXME f/b_count in avctx
1238         avctx->skip_count  = s->skip_count;
1239
1240         MPV_frame_end(s);
1241
1242         if (ENABLE_MJPEG_ENCODER && s->out_format == FMT_MJPEG)
1243             ff_mjpeg_encode_picture_trailer(s);
1244
1245         if(avctx->rc_buffer_size){
1246             RateControlContext *rcc= &s->rc_context;
1247             int max_size= rcc->buffer_index/3;
1248
1249             if(put_bits_count(&s->pb) > max_size && s->lambda < s->avctx->lmax){
1250                 s->next_lambda= FFMAX(s->lambda+1, s->lambda*(s->qscale+1) / s->qscale);
1251                 if(s->adaptive_quant){
1252                     int i;
1253                     for(i=0; i<s->mb_height*s->mb_stride; i++)
1254                         s->lambda_table[i]= FFMAX(s->lambda_table[i]+1, s->lambda_table[i]*(s->qscale+1) / s->qscale);
1255                 }
1256                 s->mb_skipped = 0;        //done in MPV_frame_start()
1257                 if(s->pict_type==P_TYPE){ //done in encode_picture() so we must undo it
1258                     if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
1259                         s->no_rounding ^= 1;
1260                 }
1261                 if(s->pict_type!=B_TYPE){
1262                     s->time_base= s->last_time_base;
1263                     s->last_non_b_time= s->time - s->pp_time;
1264                 }
1265 //                av_log(NULL, AV_LOG_ERROR, "R:%d ", s->next_lambda);
1266                 for(i=0; i<avctx->thread_count; i++){
1267                     PutBitContext *pb= &s->thread_context[i]->pb;
1268                     init_put_bits(pb, pb->buf, pb->buf_end - pb->buf);
1269                 }
1270                 goto vbv_retry;
1271             }
1272
1273             assert(s->avctx->rc_max_rate);
1274         }
1275
1276         if(s->flags&CODEC_FLAG_PASS1)
1277             ff_write_pass1_stats(s);
1278
1279         for(i=0; i<4; i++){
1280             s->current_picture_ptr->error[i]= s->current_picture.error[i];
1281             avctx->error[i] += s->current_picture_ptr->error[i];
1282         }
1283
1284         if(s->flags&CODEC_FLAG_PASS1)
1285             assert(avctx->header_bits + avctx->mv_bits + avctx->misc_bits + avctx->i_tex_bits + avctx->p_tex_bits == put_bits_count(&s->pb));
1286         flush_put_bits(&s->pb);
1287         s->frame_bits  = put_bits_count(&s->pb);
1288
1289         stuffing_count= ff_vbv_update(s, s->frame_bits);
1290         if(stuffing_count){
1291             if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < stuffing_count + 50){
1292                 av_log(s->avctx, AV_LOG_ERROR, "stuffing too large\n");
1293                 return -1;
1294             }
1295
1296             switch(s->codec_id){
1297             case CODEC_ID_MPEG1VIDEO:
1298             case CODEC_ID_MPEG2VIDEO:
1299                 while(stuffing_count--){
1300                     put_bits(&s->pb, 8, 0);
1301                 }
1302             break;
1303             case CODEC_ID_MPEG4:
1304                 put_bits(&s->pb, 16, 0);
1305                 put_bits(&s->pb, 16, 0x1C3);
1306                 stuffing_count -= 4;
1307                 while(stuffing_count--){
1308                     put_bits(&s->pb, 8, 0xFF);
1309                 }
1310             break;
1311             default:
1312                 av_log(s->avctx, AV_LOG_ERROR, "vbv buffer overflow\n");
1313             }
1314             flush_put_bits(&s->pb);
1315             s->frame_bits  = put_bits_count(&s->pb);
1316         }
1317
1318         /* update mpeg1/2 vbv_delay for CBR */
1319         if(s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate && s->out_format == FMT_MPEG1
1320            && 90000LL * (avctx->rc_buffer_size-1) <= s->avctx->rc_max_rate*0xFFFFLL){
1321             int vbv_delay;
1322
1323             assert(s->repeat_first_field==0);
1324
1325             vbv_delay= lrintf(90000 * s->rc_context.buffer_index / s->avctx->rc_max_rate);
1326             assert(vbv_delay < 0xFFFF);
1327
1328             s->vbv_delay_ptr[0] &= 0xF8;
1329             s->vbv_delay_ptr[0] |= vbv_delay>>13;
1330             s->vbv_delay_ptr[1]  = vbv_delay>>5;
1331             s->vbv_delay_ptr[2] &= 0x07;
1332             s->vbv_delay_ptr[2] |= vbv_delay<<3;
1333         }
1334         s->total_bits += s->frame_bits;
1335         avctx->frame_bits  = s->frame_bits;
1336     }else{
1337         assert((pbBufPtr(&s->pb) == s->pb.buf));
1338         s->frame_bits=0;
1339     }
1340     assert((s->frame_bits&7)==0);
1341
1342     return s->frame_bits/8;
1343 }
1344
1345 static inline void dct_single_coeff_elimination(MpegEncContext *s, int n, int threshold)
1346 {
1347     static const char tab[64]=
1348         {3,2,2,1,1,1,1,1,
1349          1,1,1,1,1,1,1,1,
1350          1,1,1,1,1,1,1,1,
1351          0,0,0,0,0,0,0,0,
1352          0,0,0,0,0,0,0,0,
1353          0,0,0,0,0,0,0,0,
1354          0,0,0,0,0,0,0,0,
1355          0,0,0,0,0,0,0,0};
1356     int score=0;
1357     int run=0;
1358     int i;
1359     DCTELEM *block= s->block[n];
1360     const int last_index= s->block_last_index[n];
1361     int skip_dc;
1362
1363     if(threshold<0){
1364         skip_dc=0;
1365         threshold= -threshold;
1366     }else
1367         skip_dc=1;
1368
1369     /* are all which we could set to zero are allready zero? */
1370     if(last_index<=skip_dc - 1) return;
1371
1372     for(i=0; i<=last_index; i++){
1373         const int j = s->intra_scantable.permutated[i];
1374         const int level = FFABS(block[j]);
1375         if(level==1){
1376             if(skip_dc && i==0) continue;
1377             score+= tab[run];
1378             run=0;
1379         }else if(level>1){
1380             return;
1381         }else{
1382             run++;
1383         }
1384     }
1385     if(score >= threshold) return;
1386     for(i=skip_dc; i<=last_index; i++){
1387         const int j = s->intra_scantable.permutated[i];
1388         block[j]=0;
1389     }
1390     if(block[0]) s->block_last_index[n]= 0;
1391     else         s->block_last_index[n]= -1;
1392 }
1393
1394 static inline void clip_coeffs(MpegEncContext *s, DCTELEM *block, int last_index)
1395 {
1396     int i;
1397     const int maxlevel= s->max_qcoeff;
1398     const int minlevel= s->min_qcoeff;
1399     int overflow=0;
1400
1401     if(s->mb_intra){
1402         i=1; //skip clipping of intra dc
1403     }else
1404         i=0;
1405
1406     for(;i<=last_index; i++){
1407         const int j= s->intra_scantable.permutated[i];
1408         int level = block[j];
1409
1410         if     (level>maxlevel){
1411             level=maxlevel;
1412             overflow++;
1413         }else if(level<minlevel){
1414             level=minlevel;
1415             overflow++;
1416         }
1417
1418         block[j]= level;
1419     }
1420
1421     if(overflow && s->avctx->mb_decision == FF_MB_DECISION_SIMPLE)
1422         av_log(s->avctx, AV_LOG_INFO, "warning, clipping %d dct coefficients to %d..%d\n", overflow, minlevel, maxlevel);
1423 }
1424
1425 static void get_visual_weight(int16_t *weight, uint8_t *ptr, int stride){
1426     int x, y;
1427 //FIXME optimize
1428     for(y=0; y<8; y++){
1429         for(x=0; x<8; x++){
1430             int x2, y2;
1431             int sum=0;
1432             int sqr=0;
1433             int count=0;
1434
1435             for(y2= FFMAX(y-1, 0); y2 < FFMIN(8, y+2); y2++){
1436                 for(x2= FFMAX(x-1, 0); x2 < FFMIN(8, x+2); x2++){
1437                     int v= ptr[x2 + y2*stride];
1438                     sum += v;
1439                     sqr += v*v;
1440                     count++;
1441                 }
1442             }
1443             weight[x + 8*y]= (36*ff_sqrt(count*sqr - sum*sum)) / count;
1444         }
1445     }
1446 }
1447
1448 static av_always_inline void encode_mb_internal(MpegEncContext *s, int motion_x, int motion_y, int mb_block_height, int mb_block_count)
1449 {
1450     int16_t weight[8][64];
1451     DCTELEM orig[8][64];
1452     const int mb_x= s->mb_x;
1453     const int mb_y= s->mb_y;
1454     int i;
1455     int skip_dct[8];
1456     int dct_offset   = s->linesize*8; //default for progressive frames
1457     uint8_t *ptr_y, *ptr_cb, *ptr_cr;
1458     int wrap_y, wrap_c;
1459
1460     for(i=0; i<mb_block_count; i++) skip_dct[i]=s->skipdct;
1461
1462     if(s->adaptive_quant){
1463         const int last_qp= s->qscale;
1464         const int mb_xy= mb_x + mb_y*s->mb_stride;
1465
1466         s->lambda= s->lambda_table[mb_xy];
1467         update_qscale(s);
1468
1469         if(!(s->flags&CODEC_FLAG_QP_RD)){
1470             s->qscale= s->current_picture_ptr->qscale_table[mb_xy];
1471             s->dquant= s->qscale - last_qp;
1472
1473             if(s->out_format==FMT_H263){
1474                 s->dquant= av_clip(s->dquant, -2, 2);
1475
1476                 if(s->codec_id==CODEC_ID_MPEG4){
1477                     if(!s->mb_intra){
1478                         if(s->pict_type == B_TYPE){
1479                             if(s->dquant&1 || s->mv_dir&MV_DIRECT)
1480                                 s->dquant= 0;
1481                         }
1482                         if(s->mv_type==MV_TYPE_8X8)
1483                             s->dquant=0;
1484                     }
1485                 }
1486             }
1487         }
1488         ff_set_qscale(s, last_qp + s->dquant);
1489     }else if(s->flags&CODEC_FLAG_QP_RD)
1490         ff_set_qscale(s, s->qscale + s->dquant);
1491
1492     wrap_y = s->linesize;
1493     wrap_c = s->uvlinesize;
1494     ptr_y = s->new_picture.data[0] + (mb_y * 16 * wrap_y) + mb_x * 16;
1495     ptr_cb = s->new_picture.data[1] + (mb_y * mb_block_height * wrap_c) + mb_x * 8;
1496     ptr_cr = s->new_picture.data[2] + (mb_y * mb_block_height * wrap_c) + mb_x * 8;
1497
1498     if(mb_x*16+16 > s->width || mb_y*16+16 > s->height){
1499         uint8_t *ebuf= s->edge_emu_buffer + 32;
1500         ff_emulated_edge_mc(ebuf            , ptr_y , wrap_y,16,16,mb_x*16,mb_y*16, s->width   , s->height);
1501         ptr_y= ebuf;
1502         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);
1503         ptr_cb= ebuf+18*wrap_y;
1504         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);
1505         ptr_cr= ebuf+18*wrap_y+8;
1506     }
1507
1508     if (s->mb_intra) {
1509         if(s->flags&CODEC_FLAG_INTERLACED_DCT){
1510             int progressive_score, interlaced_score;
1511
1512             s->interlaced_dct=0;
1513             progressive_score= s->dsp.ildct_cmp[4](s, ptr_y           , NULL, wrap_y, 8)
1514                               +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y*8, NULL, wrap_y, 8) - 400;
1515
1516             if(progressive_score > 0){
1517                 interlaced_score = s->dsp.ildct_cmp[4](s, ptr_y           , NULL, wrap_y*2, 8)
1518                                   +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y  , NULL, wrap_y*2, 8);
1519                 if(progressive_score > interlaced_score){
1520                     s->interlaced_dct=1;
1521
1522                     dct_offset= wrap_y;
1523                     wrap_y<<=1;
1524                     if (s->chroma_format == CHROMA_422)
1525                         wrap_c<<=1;
1526                 }
1527             }
1528         }
1529
1530         s->dsp.get_pixels(s->block[0], ptr_y                 , wrap_y);
1531         s->dsp.get_pixels(s->block[1], ptr_y              + 8, wrap_y);
1532         s->dsp.get_pixels(s->block[2], ptr_y + dct_offset    , wrap_y);
1533         s->dsp.get_pixels(s->block[3], ptr_y + dct_offset + 8, wrap_y);
1534
1535         if(s->flags&CODEC_FLAG_GRAY){
1536             skip_dct[4]= 1;
1537             skip_dct[5]= 1;
1538         }else{
1539             s->dsp.get_pixels(s->block[4], ptr_cb, wrap_c);
1540             s->dsp.get_pixels(s->block[5], ptr_cr, wrap_c);
1541             if(!s->chroma_y_shift){ /* 422 */
1542                 s->dsp.get_pixels(s->block[6], ptr_cb + (dct_offset>>1), wrap_c);
1543                 s->dsp.get_pixels(s->block[7], ptr_cr + (dct_offset>>1), wrap_c);
1544             }
1545         }
1546     }else{
1547         op_pixels_func (*op_pix)[4];
1548         qpel_mc_func (*op_qpix)[16];
1549         uint8_t *dest_y, *dest_cb, *dest_cr;
1550
1551         dest_y  = s->dest[0];
1552         dest_cb = s->dest[1];
1553         dest_cr = s->dest[2];
1554
1555         if ((!s->no_rounding) || s->pict_type==B_TYPE){
1556             op_pix = s->dsp.put_pixels_tab;
1557             op_qpix= s->dsp.put_qpel_pixels_tab;
1558         }else{
1559             op_pix = s->dsp.put_no_rnd_pixels_tab;
1560             op_qpix= s->dsp.put_no_rnd_qpel_pixels_tab;
1561         }
1562
1563         if (s->mv_dir & MV_DIR_FORWARD) {
1564             MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.data, op_pix, op_qpix);
1565             op_pix = s->dsp.avg_pixels_tab;
1566             op_qpix= s->dsp.avg_qpel_pixels_tab;
1567         }
1568         if (s->mv_dir & MV_DIR_BACKWARD) {
1569             MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.data, op_pix, op_qpix);
1570         }
1571
1572         if(s->flags&CODEC_FLAG_INTERLACED_DCT){
1573             int progressive_score, interlaced_score;
1574
1575             s->interlaced_dct=0;
1576             progressive_score= s->dsp.ildct_cmp[0](s, dest_y           , ptr_y           , wrap_y, 8)
1577                               +s->dsp.ildct_cmp[0](s, dest_y + wrap_y*8, ptr_y + wrap_y*8, wrap_y, 8) - 400;
1578
1579             if(s->avctx->ildct_cmp == FF_CMP_VSSE) progressive_score -= 400;
1580
1581             if(progressive_score>0){
1582                 interlaced_score = s->dsp.ildct_cmp[0](s, dest_y           , ptr_y           , wrap_y*2, 8)
1583                                   +s->dsp.ildct_cmp[0](s, dest_y + wrap_y  , ptr_y + wrap_y  , wrap_y*2, 8);
1584
1585                 if(progressive_score > interlaced_score){
1586                     s->interlaced_dct=1;
1587
1588                     dct_offset= wrap_y;
1589                     wrap_y<<=1;
1590                     if (s->chroma_format == CHROMA_422)
1591                         wrap_c<<=1;
1592                 }
1593             }
1594         }
1595
1596         s->dsp.diff_pixels(s->block[0], ptr_y                 , dest_y                 , wrap_y);
1597         s->dsp.diff_pixels(s->block[1], ptr_y              + 8, dest_y              + 8, wrap_y);
1598         s->dsp.diff_pixels(s->block[2], ptr_y + dct_offset    , dest_y + dct_offset    , wrap_y);
1599         s->dsp.diff_pixels(s->block[3], ptr_y + dct_offset + 8, dest_y + dct_offset + 8, wrap_y);
1600
1601         if(s->flags&CODEC_FLAG_GRAY){
1602             skip_dct[4]= 1;
1603             skip_dct[5]= 1;
1604         }else{
1605             s->dsp.diff_pixels(s->block[4], ptr_cb, dest_cb, wrap_c);
1606             s->dsp.diff_pixels(s->block[5], ptr_cr, dest_cr, wrap_c);
1607             if(!s->chroma_y_shift){ /* 422 */
1608                 s->dsp.diff_pixels(s->block[6], ptr_cb + (dct_offset>>1), dest_cb + (dct_offset>>1), wrap_c);
1609                 s->dsp.diff_pixels(s->block[7], ptr_cr + (dct_offset>>1), dest_cr + (dct_offset>>1), wrap_c);
1610             }
1611         }
1612         /* pre quantization */
1613         if(s->current_picture.mc_mb_var[s->mb_stride*mb_y+ mb_x]<2*s->qscale*s->qscale){
1614             //FIXME optimize
1615             if(s->dsp.sad[1](NULL, ptr_y               , dest_y               , wrap_y, 8) < 20*s->qscale) skip_dct[0]= 1;
1616             if(s->dsp.sad[1](NULL, ptr_y            + 8, dest_y            + 8, wrap_y, 8) < 20*s->qscale) skip_dct[1]= 1;
1617             if(s->dsp.sad[1](NULL, ptr_y +dct_offset   , dest_y +dct_offset   , wrap_y, 8) < 20*s->qscale) skip_dct[2]= 1;
1618             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;
1619             if(s->dsp.sad[1](NULL, ptr_cb              , dest_cb              , wrap_c, 8) < 20*s->qscale) skip_dct[4]= 1;
1620             if(s->dsp.sad[1](NULL, ptr_cr              , dest_cr              , wrap_c, 8) < 20*s->qscale) skip_dct[5]= 1;
1621             if(!s->chroma_y_shift){ /* 422 */
1622                 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;
1623                 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;
1624             }
1625         }
1626     }
1627
1628     if(s->avctx->quantizer_noise_shaping){
1629         if(!skip_dct[0]) get_visual_weight(weight[0], ptr_y                 , wrap_y);
1630         if(!skip_dct[1]) get_visual_weight(weight[1], ptr_y              + 8, wrap_y);
1631         if(!skip_dct[2]) get_visual_weight(weight[2], ptr_y + dct_offset    , wrap_y);
1632         if(!skip_dct[3]) get_visual_weight(weight[3], ptr_y + dct_offset + 8, wrap_y);
1633         if(!skip_dct[4]) get_visual_weight(weight[4], ptr_cb                , wrap_c);
1634         if(!skip_dct[5]) get_visual_weight(weight[5], ptr_cr                , wrap_c);
1635         if(!s->chroma_y_shift){ /* 422 */
1636             if(!skip_dct[6]) get_visual_weight(weight[6], ptr_cb + (dct_offset>>1), wrap_c);
1637             if(!skip_dct[7]) get_visual_weight(weight[7], ptr_cr + (dct_offset>>1), wrap_c);
1638         }
1639         memcpy(orig[0], s->block[0], sizeof(DCTELEM)*64*mb_block_count);
1640     }
1641
1642     /* DCT & quantize */
1643     assert(s->out_format!=FMT_MJPEG || s->qscale==8);
1644     {
1645         for(i=0;i<mb_block_count;i++) {
1646             if(!skip_dct[i]){
1647                 int overflow;
1648                 s->block_last_index[i] = s->dct_quantize(s, s->block[i], i, s->qscale, &overflow);
1649             // FIXME we could decide to change to quantizer instead of clipping
1650             // JS: I don't think that would be a good idea it could lower quality instead
1651             //     of improve it. Just INTRADC clipping deserves changes in quantizer
1652                 if (overflow) clip_coeffs(s, s->block[i], s->block_last_index[i]);
1653             }else
1654                 s->block_last_index[i]= -1;
1655         }
1656         if(s->avctx->quantizer_noise_shaping){
1657             for(i=0;i<mb_block_count;i++) {
1658                 if(!skip_dct[i]){
1659                     s->block_last_index[i] = dct_quantize_refine(s, s->block[i], weight[i], orig[i], i, s->qscale);
1660                 }
1661             }
1662         }
1663
1664         if(s->luma_elim_threshold && !s->mb_intra)
1665             for(i=0; i<4; i++)
1666                 dct_single_coeff_elimination(s, i, s->luma_elim_threshold);
1667         if(s->chroma_elim_threshold && !s->mb_intra)
1668             for(i=4; i<mb_block_count; i++)
1669                 dct_single_coeff_elimination(s, i, s->chroma_elim_threshold);
1670
1671         if(s->flags & CODEC_FLAG_CBP_RD){
1672             for(i=0;i<mb_block_count;i++) {
1673                 if(s->block_last_index[i] == -1)
1674                     s->coded_score[i]= INT_MAX/256;
1675             }
1676         }
1677     }
1678
1679     if((s->flags&CODEC_FLAG_GRAY) && s->mb_intra){
1680         s->block_last_index[4]=
1681         s->block_last_index[5]= 0;
1682         s->block[4][0]=
1683         s->block[5][0]= (1024 + s->c_dc_scale/2)/ s->c_dc_scale;
1684     }
1685
1686     //non c quantize code returns incorrect block_last_index FIXME
1687     if(s->alternate_scan && s->dct_quantize != dct_quantize_c){
1688         for(i=0; i<mb_block_count; i++){
1689             int j;
1690             if(s->block_last_index[i]>0){
1691                 for(j=63; j>0; j--){
1692                     if(s->block[i][ s->intra_scantable.permutated[j] ]) break;
1693                 }
1694                 s->block_last_index[i]= j;
1695             }
1696         }
1697     }
1698
1699     /* huffman encode */
1700     switch(s->codec_id){ //FIXME funct ptr could be slightly faster
1701     case CODEC_ID_MPEG1VIDEO:
1702     case CODEC_ID_MPEG2VIDEO:
1703         if (ENABLE_MPEG1VIDEO_ENCODER || ENABLE_MPEG2VIDEO_ENCODER)
1704             mpeg1_encode_mb(s, s->block, motion_x, motion_y);
1705         break;
1706     case CODEC_ID_MPEG4:
1707         if (ENABLE_MPEG4_ENCODER)
1708             mpeg4_encode_mb(s, s->block, motion_x, motion_y);
1709         break;
1710     case CODEC_ID_MSMPEG4V2:
1711     case CODEC_ID_MSMPEG4V3:
1712     case CODEC_ID_WMV1:
1713         if (ENABLE_MSMPEG4_ENCODER)
1714             msmpeg4_encode_mb(s, s->block, motion_x, motion_y);
1715         break;
1716     case CODEC_ID_WMV2:
1717         if (ENABLE_WMV2_ENCODER)
1718             ff_wmv2_encode_mb(s, s->block, motion_x, motion_y);
1719         break;
1720     case CODEC_ID_H261:
1721         if (ENABLE_H261_ENCODER)
1722             ff_h261_encode_mb(s, s->block, motion_x, motion_y);
1723         break;
1724     case CODEC_ID_H263:
1725     case CODEC_ID_H263P:
1726     case CODEC_ID_FLV1:
1727     case CODEC_ID_RV10:
1728     case CODEC_ID_RV20:
1729         if (ENABLE_H263_ENCODER || ENABLE_H263P_ENCODER ||
1730             ENABLE_FLV_ENCODER  || ENABLE_RV10_ENCODER  || ENABLE_RV20_ENCODER)
1731             h263_encode_mb(s, s->block, motion_x, motion_y);
1732         break;
1733     case CODEC_ID_MJPEG:
1734         if (ENABLE_MJPEG_ENCODER)
1735             ff_mjpeg_encode_mb(s, s->block);
1736         break;
1737     default:
1738         assert(0);
1739     }
1740 }
1741
1742 static av_always_inline void encode_mb(MpegEncContext *s, int motion_x, int motion_y)
1743 {
1744     if (s->chroma_format == CHROMA_420) encode_mb_internal(s, motion_x, motion_y,  8, 6);
1745     else                                encode_mb_internal(s, motion_x, motion_y, 16, 8);
1746 }
1747
1748 static inline void copy_context_before_encode(MpegEncContext *d, MpegEncContext *s, int type){
1749     int i;
1750
1751     memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
1752
1753     /* mpeg1 */
1754     d->mb_skip_run= s->mb_skip_run;
1755     for(i=0; i<3; i++)
1756         d->last_dc[i]= s->last_dc[i];
1757
1758     /* statistics */
1759     d->mv_bits= s->mv_bits;
1760     d->i_tex_bits= s->i_tex_bits;
1761     d->p_tex_bits= s->p_tex_bits;
1762     d->i_count= s->i_count;
1763     d->f_count= s->f_count;
1764     d->b_count= s->b_count;
1765     d->skip_count= s->skip_count;
1766     d->misc_bits= s->misc_bits;
1767     d->last_bits= 0;
1768
1769     d->mb_skipped= 0;
1770     d->qscale= s->qscale;
1771     d->dquant= s->dquant;
1772 }
1773
1774 static inline void copy_context_after_encode(MpegEncContext *d, MpegEncContext *s, int type){
1775     int i;
1776
1777     memcpy(d->mv, s->mv, 2*4*2*sizeof(int));
1778     memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
1779
1780     /* mpeg1 */
1781     d->mb_skip_run= s->mb_skip_run;
1782     for(i=0; i<3; i++)
1783         d->last_dc[i]= s->last_dc[i];
1784
1785     /* statistics */
1786     d->mv_bits= s->mv_bits;
1787     d->i_tex_bits= s->i_tex_bits;
1788     d->p_tex_bits= s->p_tex_bits;
1789     d->i_count= s->i_count;
1790     d->f_count= s->f_count;
1791     d->b_count= s->b_count;
1792     d->skip_count= s->skip_count;
1793     d->misc_bits= s->misc_bits;
1794
1795     d->mb_intra= s->mb_intra;
1796     d->mb_skipped= s->mb_skipped;
1797     d->mv_type= s->mv_type;
1798     d->mv_dir= s->mv_dir;
1799     d->pb= s->pb;
1800     if(s->data_partitioning){
1801         d->pb2= s->pb2;
1802         d->tex_pb= s->tex_pb;
1803     }
1804     d->block= s->block;
1805     for(i=0; i<8; i++)
1806         d->block_last_index[i]= s->block_last_index[i];
1807     d->interlaced_dct= s->interlaced_dct;
1808     d->qscale= s->qscale;
1809 }
1810
1811 static inline void encode_mb_hq(MpegEncContext *s, MpegEncContext *backup, MpegEncContext *best, int type,
1812                            PutBitContext pb[2], PutBitContext pb2[2], PutBitContext tex_pb[2],
1813                            int *dmin, int *next_block, int motion_x, int motion_y)
1814 {
1815     int score;
1816     uint8_t *dest_backup[3];
1817
1818     copy_context_before_encode(s, backup, type);
1819
1820     s->block= s->blocks[*next_block];
1821     s->pb= pb[*next_block];
1822     if(s->data_partitioning){
1823         s->pb2   = pb2   [*next_block];
1824         s->tex_pb= tex_pb[*next_block];
1825     }
1826
1827     if(*next_block){
1828         memcpy(dest_backup, s->dest, sizeof(s->dest));
1829         s->dest[0] = s->rd_scratchpad;
1830         s->dest[1] = s->rd_scratchpad + 16*s->linesize;
1831         s->dest[2] = s->rd_scratchpad + 16*s->linesize + 8;
1832         assert(s->linesize >= 32); //FIXME
1833     }
1834
1835     encode_mb(s, motion_x, motion_y);
1836
1837     score= put_bits_count(&s->pb);
1838     if(s->data_partitioning){
1839         score+= put_bits_count(&s->pb2);
1840         score+= put_bits_count(&s->tex_pb);
1841     }
1842
1843     if(s->avctx->mb_decision == FF_MB_DECISION_RD){
1844         MPV_decode_mb(s, s->block);
1845
1846         score *= s->lambda2;
1847         score += sse_mb(s) << FF_LAMBDA_SHIFT;
1848     }
1849
1850     if(*next_block){
1851         memcpy(s->dest, dest_backup, sizeof(s->dest));
1852     }
1853
1854     if(score<*dmin){
1855         *dmin= score;
1856         *next_block^=1;
1857
1858         copy_context_after_encode(best, s, type);
1859     }
1860 }
1861
1862 static int sse(MpegEncContext *s, uint8_t *src1, uint8_t *src2, int w, int h, int stride){
1863     uint32_t *sq = ff_squareTbl + 256;
1864     int acc=0;
1865     int x,y;
1866
1867     if(w==16 && h==16)
1868         return s->dsp.sse[0](NULL, src1, src2, stride, 16);
1869     else if(w==8 && h==8)
1870         return s->dsp.sse[1](NULL, src1, src2, stride, 8);
1871
1872     for(y=0; y<h; y++){
1873         for(x=0; x<w; x++){
1874             acc+= sq[src1[x + y*stride] - src2[x + y*stride]];
1875         }
1876     }
1877
1878     assert(acc>=0);
1879
1880     return acc;
1881 }
1882
1883 static int sse_mb(MpegEncContext *s){
1884     int w= 16;
1885     int h= 16;
1886
1887     if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
1888     if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
1889
1890     if(w==16 && h==16)
1891       if(s->avctx->mb_cmp == FF_CMP_NSSE){
1892         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)
1893                +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)
1894                +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);
1895       }else{
1896         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)
1897                +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)
1898                +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);
1899       }
1900     else
1901         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)
1902                +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)
1903                +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);
1904 }
1905
1906 static int pre_estimate_motion_thread(AVCodecContext *c, void *arg){
1907     MpegEncContext *s= arg;
1908
1909
1910     s->me.pre_pass=1;
1911     s->me.dia_size= s->avctx->pre_dia_size;
1912     s->first_slice_line=1;
1913     for(s->mb_y= s->end_mb_y-1; s->mb_y >= s->start_mb_y; s->mb_y--) {
1914         for(s->mb_x=s->mb_width-1; s->mb_x >=0 ;s->mb_x--) {
1915             ff_pre_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
1916         }
1917         s->first_slice_line=0;
1918     }
1919
1920     s->me.pre_pass=0;
1921
1922     return 0;
1923 }
1924
1925 static int estimate_motion_thread(AVCodecContext *c, void *arg){
1926     MpegEncContext *s= arg;
1927
1928     ff_check_alignment();
1929
1930     s->me.dia_size= s->avctx->dia_size;
1931     s->first_slice_line=1;
1932     for(s->mb_y= s->start_mb_y; s->mb_y < s->end_mb_y; s->mb_y++) {
1933         s->mb_x=0; //for block init below
1934         ff_init_block_index(s);
1935         for(s->mb_x=0; s->mb_x < s->mb_width; s->mb_x++) {
1936             s->block_index[0]+=2;
1937             s->block_index[1]+=2;
1938             s->block_index[2]+=2;
1939             s->block_index[3]+=2;
1940
1941             /* compute motion vector & mb_type and store in context */
1942             if(s->pict_type==B_TYPE)
1943                 ff_estimate_b_frame_motion(s, s->mb_x, s->mb_y);
1944             else
1945                 ff_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
1946         }
1947         s->first_slice_line=0;
1948     }
1949     return 0;
1950 }
1951
1952 static int mb_var_thread(AVCodecContext *c, void *arg){
1953     MpegEncContext *s= arg;
1954     int mb_x, mb_y;
1955
1956     ff_check_alignment();
1957
1958     for(mb_y=s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
1959         for(mb_x=0; mb_x < s->mb_width; mb_x++) {
1960             int xx = mb_x * 16;
1961             int yy = mb_y * 16;
1962             uint8_t *pix = s->new_picture.data[0] + (yy * s->linesize) + xx;
1963             int varc;
1964             int sum = s->dsp.pix_sum(pix, s->linesize);
1965
1966             varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500 + 128)>>8;
1967
1968             s->current_picture.mb_var [s->mb_stride * mb_y + mb_x] = varc;
1969             s->current_picture.mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
1970             s->me.mb_var_sum_temp    += varc;
1971         }
1972     }
1973     return 0;
1974 }
1975
1976 static void write_slice_end(MpegEncContext *s){
1977     if(ENABLE_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4){
1978         if(s->partitioned_frame){
1979             ff_mpeg4_merge_partitions(s);
1980         }
1981
1982         ff_mpeg4_stuffing(&s->pb);
1983     }else if(ENABLE_MJPEG_ENCODER && s->out_format == FMT_MJPEG){
1984         ff_mjpeg_encode_stuffing(&s->pb);
1985     }
1986
1987     align_put_bits(&s->pb);
1988     flush_put_bits(&s->pb);
1989
1990     if((s->flags&CODEC_FLAG_PASS1) && !s->partitioned_frame)
1991         s->misc_bits+= get_bits_diff(s);
1992 }
1993
1994 static int encode_thread(AVCodecContext *c, void *arg){
1995     MpegEncContext *s= arg;
1996     int mb_x, mb_y, pdif = 0;
1997     int i, j;
1998     MpegEncContext best_s, backup_s;
1999     uint8_t bit_buf[2][MAX_MB_BYTES];
2000     uint8_t bit_buf2[2][MAX_MB_BYTES];
2001     uint8_t bit_buf_tex[2][MAX_MB_BYTES];
2002     PutBitContext pb[2], pb2[2], tex_pb[2];
2003 //printf("%d->%d\n", s->resync_mb_y, s->end_mb_y);
2004
2005     ff_check_alignment();
2006
2007     for(i=0; i<2; i++){
2008         init_put_bits(&pb    [i], bit_buf    [i], MAX_MB_BYTES);
2009         init_put_bits(&pb2   [i], bit_buf2   [i], MAX_MB_BYTES);
2010         init_put_bits(&tex_pb[i], bit_buf_tex[i], MAX_MB_BYTES);
2011     }
2012
2013     s->last_bits= put_bits_count(&s->pb);
2014     s->mv_bits=0;
2015     s->misc_bits=0;
2016     s->i_tex_bits=0;
2017     s->p_tex_bits=0;
2018     s->i_count=0;
2019     s->f_count=0;
2020     s->b_count=0;
2021     s->skip_count=0;
2022
2023     for(i=0; i<3; i++){
2024         /* init last dc values */
2025         /* note: quant matrix value (8) is implied here */
2026         s->last_dc[i] = 128 << s->intra_dc_precision;
2027
2028         s->current_picture.error[i] = 0;
2029     }
2030     s->mb_skip_run = 0;
2031     memset(s->last_mv, 0, sizeof(s->last_mv));
2032
2033     s->last_mv_dir = 0;
2034
2035     switch(s->codec_id){
2036     case CODEC_ID_H263:
2037     case CODEC_ID_H263P:
2038     case CODEC_ID_FLV1:
2039         if (ENABLE_H263_ENCODER || ENABLE_H263P_ENCODER || ENABLE_FLV_ENCODER)
2040             s->gob_index = ff_h263_get_gob_height(s);
2041         break;
2042     case CODEC_ID_MPEG4:
2043         if(ENABLE_MPEG4_ENCODER && s->partitioned_frame)
2044             ff_mpeg4_init_partitions(s);
2045         break;
2046     }
2047
2048     s->resync_mb_x=0;
2049     s->resync_mb_y=0;
2050     s->first_slice_line = 1;
2051     s->ptr_lastgob = s->pb.buf;
2052     for(mb_y= s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
2053 //    printf("row %d at %X\n", s->mb_y, (int)s);
2054         s->mb_x=0;
2055         s->mb_y= mb_y;
2056
2057         ff_set_qscale(s, s->qscale);
2058         ff_init_block_index(s);
2059
2060         for(mb_x=0; mb_x < s->mb_width; mb_x++) {
2061             int xy= mb_y*s->mb_stride + mb_x; // removed const, H261 needs to adjust this
2062             int mb_type= s->mb_type[xy];
2063 //            int d;
2064             int dmin= INT_MAX;
2065             int dir;
2066
2067             if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < MAX_MB_BYTES){
2068                 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
2069                 return -1;
2070             }
2071             if(s->data_partitioning){
2072                 if(   s->pb2   .buf_end - s->pb2   .buf - (put_bits_count(&s->    pb2)>>3) < MAX_MB_BYTES
2073                    || s->tex_pb.buf_end - s->tex_pb.buf - (put_bits_count(&s->tex_pb )>>3) < MAX_MB_BYTES){
2074                     av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
2075                     return -1;
2076                 }
2077             }
2078
2079             s->mb_x = mb_x;
2080             s->mb_y = mb_y;  // moved into loop, can get changed by H.261
2081             ff_update_block_index(s);
2082
2083             if(ENABLE_H261_ENCODER && s->codec_id == CODEC_ID_H261){
2084                 ff_h261_reorder_mb_index(s);
2085                 xy= s->mb_y*s->mb_stride + s->mb_x;
2086                 mb_type= s->mb_type[xy];
2087             }
2088
2089             /* write gob / video packet header  */
2090             if(s->rtp_mode){
2091                 int current_packet_size, is_gob_start;
2092
2093                 current_packet_size= ((put_bits_count(&s->pb)+7)>>3) - (s->ptr_lastgob - s->pb.buf);
2094
2095                 is_gob_start= s->avctx->rtp_payload_size && current_packet_size >= s->avctx->rtp_payload_size && mb_y + mb_x>0;
2096
2097                 if(s->start_mb_y == mb_y && mb_y > 0 && mb_x==0) is_gob_start=1;
2098
2099                 switch(s->codec_id){
2100                 case CODEC_ID_H263:
2101                 case CODEC_ID_H263P:
2102                     if(!s->h263_slice_structured)
2103                         if(s->mb_x || s->mb_y%s->gob_index) is_gob_start=0;
2104                     break;
2105                 case CODEC_ID_MPEG2VIDEO:
2106                     if(s->mb_x==0 && s->mb_y!=0) is_gob_start=1;
2107                 case CODEC_ID_MPEG1VIDEO:
2108                     if(s->mb_skip_run) is_gob_start=0;
2109                     break;
2110                 }
2111
2112                 if(is_gob_start){
2113                     if(s->start_mb_y != mb_y || mb_x!=0){
2114                         write_slice_end(s);
2115
2116                         if(ENABLE_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4 && s->partitioned_frame){
2117                             ff_mpeg4_init_partitions(s);
2118                         }
2119                     }
2120
2121                     assert((put_bits_count(&s->pb)&7) == 0);
2122                     current_packet_size= pbBufPtr(&s->pb) - s->ptr_lastgob;
2123
2124                     if(s->avctx->error_rate && s->resync_mb_x + s->resync_mb_y > 0){
2125                         int r= put_bits_count(&s->pb)/8 + s->picture_number + 16 + s->mb_x + s->mb_y;
2126                         int d= 100 / s->avctx->error_rate;
2127                         if(r % d == 0){
2128                             current_packet_size=0;
2129 #ifndef ALT_BITSTREAM_WRITER
2130                             s->pb.buf_ptr= s->ptr_lastgob;
2131 #endif
2132                             assert(pbBufPtr(&s->pb) == s->ptr_lastgob);
2133                         }
2134                     }
2135
2136                     if (s->avctx->rtp_callback){
2137                         int number_mb = (mb_y - s->resync_mb_y)*s->mb_width + mb_x - s->resync_mb_x;
2138                         s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, current_packet_size, number_mb);
2139                     }
2140
2141                     switch(s->codec_id){
2142                     case CODEC_ID_MPEG4:
2143                         if (ENABLE_MPEG4_ENCODER) {
2144                             ff_mpeg4_encode_video_packet_header(s);
2145                             ff_mpeg4_clean_buffers(s);
2146                         }
2147                     break;
2148                     case CODEC_ID_MPEG1VIDEO:
2149                     case CODEC_ID_MPEG2VIDEO:
2150                         if (ENABLE_MPEG1VIDEO_ENCODER || ENABLE_MPEG2VIDEO_ENCODER) {
2151                             ff_mpeg1_encode_slice_header(s);
2152                             ff_mpeg1_clean_buffers(s);
2153                         }
2154                     break;
2155                     case CODEC_ID_H263:
2156                     case CODEC_ID_H263P:
2157                         if (ENABLE_H263_ENCODER || ENABLE_H263P_ENCODER)
2158                             h263_encode_gob_header(s, mb_y);
2159                     break;
2160                     }
2161
2162                     if(s->flags&CODEC_FLAG_PASS1){
2163                         int bits= put_bits_count(&s->pb);
2164                         s->misc_bits+= bits - s->last_bits;
2165                         s->last_bits= bits;
2166                     }
2167
2168                     s->ptr_lastgob += current_packet_size;
2169                     s->first_slice_line=1;
2170                     s->resync_mb_x=mb_x;
2171                     s->resync_mb_y=mb_y;
2172                 }
2173             }
2174
2175             if(  (s->resync_mb_x   == s->mb_x)
2176                && s->resync_mb_y+1 == s->mb_y){
2177                 s->first_slice_line=0;
2178             }
2179
2180             s->mb_skipped=0;
2181             s->dquant=0; //only for QP_RD
2182
2183             if(mb_type & (mb_type-1) || (s->flags & CODEC_FLAG_QP_RD)){ // more than 1 MB type possible or CODEC_FLAG_QP_RD
2184                 int next_block=0;
2185                 int pb_bits_count, pb2_bits_count, tex_pb_bits_count;
2186
2187                 copy_context_before_encode(&backup_s, s, -1);
2188                 backup_s.pb= s->pb;
2189                 best_s.data_partitioning= s->data_partitioning;
2190                 best_s.partitioned_frame= s->partitioned_frame;
2191                 if(s->data_partitioning){
2192                     backup_s.pb2= s->pb2;
2193                     backup_s.tex_pb= s->tex_pb;
2194                 }
2195
2196                 if(mb_type&CANDIDATE_MB_TYPE_INTER){
2197                     s->mv_dir = MV_DIR_FORWARD;
2198                     s->mv_type = MV_TYPE_16X16;
2199                     s->mb_intra= 0;
2200                     s->mv[0][0][0] = s->p_mv_table[xy][0];
2201                     s->mv[0][0][1] = s->p_mv_table[xy][1];
2202                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER, pb, pb2, tex_pb,
2203                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2204                 }
2205                 if(mb_type&CANDIDATE_MB_TYPE_INTER_I){
2206                     s->mv_dir = MV_DIR_FORWARD;
2207                     s->mv_type = MV_TYPE_FIELD;
2208                     s->mb_intra= 0;
2209                     for(i=0; i<2; i++){
2210                         j= s->field_select[0][i] = s->p_field_select_table[i][xy];
2211                         s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
2212                         s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
2213                     }
2214                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER_I, pb, pb2, tex_pb,
2215                                  &dmin, &next_block, 0, 0);
2216                 }
2217                 if(mb_type&CANDIDATE_MB_TYPE_SKIPPED){
2218                     s->mv_dir = MV_DIR_FORWARD;
2219                     s->mv_type = MV_TYPE_16X16;
2220                     s->mb_intra= 0;
2221                     s->mv[0][0][0] = 0;
2222                     s->mv[0][0][1] = 0;
2223                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_SKIPPED, pb, pb2, tex_pb,
2224                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2225                 }
2226                 if(mb_type&CANDIDATE_MB_TYPE_INTER4V){
2227                     s->mv_dir = MV_DIR_FORWARD;
2228                     s->mv_type = MV_TYPE_8X8;
2229                     s->mb_intra= 0;
2230                     for(i=0; i<4; i++){
2231                         s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
2232                         s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
2233                     }
2234                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER4V, pb, pb2, tex_pb,
2235                                  &dmin, &next_block, 0, 0);
2236                 }
2237                 if(mb_type&CANDIDATE_MB_TYPE_FORWARD){
2238                     s->mv_dir = MV_DIR_FORWARD;
2239                     s->mv_type = MV_TYPE_16X16;
2240                     s->mb_intra= 0;
2241                     s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
2242                     s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2243                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD, pb, pb2, tex_pb,
2244                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2245                 }
2246                 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD){
2247                     s->mv_dir = MV_DIR_BACKWARD;
2248                     s->mv_type = MV_TYPE_16X16;
2249                     s->mb_intra= 0;
2250                     s->mv[1][0][0] = s->b_back_mv_table[xy][0];
2251                     s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2252                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD, pb, pb2, tex_pb,
2253                                  &dmin, &next_block, s->mv[1][0][0], s->mv[1][0][1]);
2254                 }
2255                 if(mb_type&CANDIDATE_MB_TYPE_BIDIR){
2256                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2257                     s->mv_type = MV_TYPE_16X16;
2258                     s->mb_intra= 0;
2259                     s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
2260                     s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
2261                     s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
2262                     s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2263                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR, pb, pb2, tex_pb,
2264                                  &dmin, &next_block, 0, 0);
2265                 }
2266                 if(mb_type&CANDIDATE_MB_TYPE_FORWARD_I){
2267                     s->mv_dir = MV_DIR_FORWARD;
2268                     s->mv_type = MV_TYPE_FIELD;
2269                     s->mb_intra= 0;
2270                     for(i=0; i<2; i++){
2271                         j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
2272                         s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
2273                         s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
2274                     }
2275                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD_I, pb, pb2, tex_pb,
2276                                  &dmin, &next_block, 0, 0);
2277                 }
2278                 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD_I){
2279                     s->mv_dir = MV_DIR_BACKWARD;
2280                     s->mv_type = MV_TYPE_FIELD;
2281                     s->mb_intra= 0;
2282                     for(i=0; i<2; i++){
2283                         j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
2284                         s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
2285                         s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
2286                     }
2287                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD_I, pb, pb2, tex_pb,
2288                                  &dmin, &next_block, 0, 0);
2289                 }
2290                 if(mb_type&CANDIDATE_MB_TYPE_BIDIR_I){
2291                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2292                     s->mv_type = MV_TYPE_FIELD;
2293                     s->mb_intra= 0;
2294                     for(dir=0; dir<2; dir++){
2295                         for(i=0; i<2; i++){
2296                             j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
2297                             s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
2298                             s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
2299                         }
2300                     }
2301                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR_I, pb, pb2, tex_pb,
2302                                  &dmin, &next_block, 0, 0);
2303                 }
2304                 if(mb_type&CANDIDATE_MB_TYPE_INTRA){
2305                     s->mv_dir = 0;
2306                     s->mv_type = MV_TYPE_16X16;
2307                     s->mb_intra= 1;
2308                     s->mv[0][0][0] = 0;
2309                     s->mv[0][0][1] = 0;
2310                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTRA, pb, pb2, tex_pb,
2311                                  &dmin, &next_block, 0, 0);
2312                     if(s->h263_pred || s->h263_aic){
2313                         if(best_s.mb_intra)
2314                             s->mbintra_table[mb_x + mb_y*s->mb_stride]=1;
2315                         else
2316                             ff_clean_intra_table_entries(s); //old mode?
2317                     }
2318                 }
2319
2320                 if((s->flags & CODEC_FLAG_QP_RD) && dmin < INT_MAX){
2321                     if(best_s.mv_type==MV_TYPE_16X16){ //FIXME move 4mv after QPRD
2322                         const int last_qp= backup_s.qscale;
2323                         int qpi, qp, dc[6];
2324                         DCTELEM ac[6][16];
2325                         const int mvdir= (best_s.mv_dir&MV_DIR_BACKWARD) ? 1 : 0;
2326                         static const int dquant_tab[4]={-1,1,-2,2};
2327
2328                         assert(backup_s.dquant == 0);
2329
2330                         //FIXME intra
2331                         s->mv_dir= best_s.mv_dir;
2332                         s->mv_type = MV_TYPE_16X16;
2333                         s->mb_intra= best_s.mb_intra;
2334                         s->mv[0][0][0] = best_s.mv[0][0][0];
2335                         s->mv[0][0][1] = best_s.mv[0][0][1];
2336                         s->mv[1][0][0] = best_s.mv[1][0][0];
2337                         s->mv[1][0][1] = best_s.mv[1][0][1];
2338
2339                         qpi = s->pict_type == B_TYPE ? 2 : 0;
2340                         for(; qpi<4; qpi++){
2341                             int dquant= dquant_tab[qpi];
2342                             qp= last_qp + dquant;
2343                             if(qp < s->avctx->qmin || qp > s->avctx->qmax)
2344                                 continue;
2345                             backup_s.dquant= dquant;
2346                             if(s->mb_intra && s->dc_val[0]){
2347                                 for(i=0; i<6; i++){
2348                                     dc[i]= s->dc_val[0][ s->block_index[i] ];
2349                                     memcpy(ac[i], s->ac_val[0][s->block_index[i]], sizeof(DCTELEM)*16);
2350                                 }
2351                             }
2352
2353                             encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
2354                                          &dmin, &next_block, s->mv[mvdir][0][0], s->mv[mvdir][0][1]);
2355                             if(best_s.qscale != qp){
2356                                 if(s->mb_intra && s->dc_val[0]){
2357                                     for(i=0; i<6; i++){
2358                                         s->dc_val[0][ s->block_index[i] ]= dc[i];
2359                                         memcpy(s->ac_val[0][s->block_index[i]], ac[i], sizeof(DCTELEM)*16);
2360                                     }
2361                                 }
2362                             }
2363                         }
2364                     }
2365                 }
2366                 if(ENABLE_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT){
2367                     int mx= s->b_direct_mv_table[xy][0];
2368                     int my= s->b_direct_mv_table[xy][1];
2369
2370                     backup_s.dquant = 0;
2371                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2372                     s->mb_intra= 0;
2373                     ff_mpeg4_set_direct_mv(s, mx, my);
2374                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
2375                                  &dmin, &next_block, mx, my);
2376                 }
2377                 if(ENABLE_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT0){
2378                     backup_s.dquant = 0;
2379                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2380                     s->mb_intra= 0;
2381                     ff_mpeg4_set_direct_mv(s, 0, 0);
2382                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
2383                                  &dmin, &next_block, 0, 0);
2384                 }
2385                 if(!best_s.mb_intra && s->flags2&CODEC_FLAG2_SKIP_RD){
2386                     int coded=0;
2387                     for(i=0; i<6; i++)
2388                         coded |= s->block_last_index[i];
2389                     if(coded){
2390                         int mx,my;
2391                         memcpy(s->mv, best_s.mv, sizeof(s->mv));
2392                         if(ENABLE_MPEG4_ENCODER && best_s.mv_dir & MV_DIRECT){
2393                             mx=my=0; //FIXME find the one we actually used
2394                             ff_mpeg4_set_direct_mv(s, mx, my);
2395                         }else if(best_s.mv_dir&MV_DIR_BACKWARD){
2396                             mx= s->mv[1][0][0];
2397                             my= s->mv[1][0][1];
2398                         }else{
2399                             mx= s->mv[0][0][0];
2400                             my= s->mv[0][0][1];
2401                         }
2402
2403                         s->mv_dir= best_s.mv_dir;
2404                         s->mv_type = best_s.mv_type;
2405                         s->mb_intra= 0;
2406 /*                        s->mv[0][0][0] = best_s.mv[0][0][0];
2407                         s->mv[0][0][1] = best_s.mv[0][0][1];
2408                         s->mv[1][0][0] = best_s.mv[1][0][0];
2409                         s->mv[1][0][1] = best_s.mv[1][0][1];*/
2410                         backup_s.dquant= 0;
2411                         s->skipdct=1;
2412                         encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
2413                                         &dmin, &next_block, mx, my);
2414                         s->skipdct=0;
2415                     }
2416                 }
2417
2418                 s->current_picture.qscale_table[xy]= best_s.qscale;
2419
2420                 copy_context_after_encode(s, &best_s, -1);
2421
2422                 pb_bits_count= put_bits_count(&s->pb);
2423                 flush_put_bits(&s->pb);
2424                 ff_copy_bits(&backup_s.pb, bit_buf[next_block^1], pb_bits_count);
2425                 s->pb= backup_s.pb;
2426
2427                 if(s->data_partitioning){
2428                     pb2_bits_count= put_bits_count(&s->pb2);
2429                     flush_put_bits(&s->pb2);
2430                     ff_copy_bits(&backup_s.pb2, bit_buf2[next_block^1], pb2_bits_count);
2431                     s->pb2= backup_s.pb2;
2432
2433                     tex_pb_bits_count= put_bits_count(&s->tex_pb);
2434                     flush_put_bits(&s->tex_pb);
2435                     ff_copy_bits(&backup_s.tex_pb, bit_buf_tex[next_block^1], tex_pb_bits_count);
2436                     s->tex_pb= backup_s.tex_pb;
2437                 }
2438                 s->last_bits= put_bits_count(&s->pb);
2439
2440                 if (ENABLE_ANY_H263_ENCODER &&
2441                     s->out_format == FMT_H263 && s->pict_type!=B_TYPE)
2442                     ff_h263_update_motion_val(s);
2443
2444                 if(next_block==0){ //FIXME 16 vs linesize16
2445                     s->dsp.put_pixels_tab[0][0](s->dest[0], s->rd_scratchpad                     , s->linesize  ,16);
2446                     s->dsp.put_pixels_tab[1][0](s->dest[1], s->rd_scratchpad + 16*s->linesize    , s->uvlinesize, 8);
2447                     s->dsp.put_pixels_tab[1][0](s->dest[2], s->rd_scratchpad + 16*s->linesize + 8, s->uvlinesize, 8);
2448                 }
2449
2450                 if(s->avctx->mb_decision == FF_MB_DECISION_BITS)
2451                     MPV_decode_mb(s, s->block);
2452             } else {
2453                 int motion_x = 0, motion_y = 0;
2454                 s->mv_type=MV_TYPE_16X16;
2455                 // only one MB-Type possible
2456
2457                 switch(mb_type){
2458                 case CANDIDATE_MB_TYPE_INTRA:
2459                     s->mv_dir = 0;
2460                     s->mb_intra= 1;
2461                     motion_x= s->mv[0][0][0] = 0;
2462                     motion_y= s->mv[0][0][1] = 0;
2463                     break;
2464                 case CANDIDATE_MB_TYPE_INTER:
2465                     s->mv_dir = MV_DIR_FORWARD;
2466                     s->mb_intra= 0;
2467                     motion_x= s->mv[0][0][0] = s->p_mv_table[xy][0];
2468                     motion_y= s->mv[0][0][1] = s->p_mv_table[xy][1];
2469                     break;
2470                 case CANDIDATE_MB_TYPE_INTER_I:
2471                     s->mv_dir = MV_DIR_FORWARD;
2472                     s->mv_type = MV_TYPE_FIELD;
2473                     s->mb_intra= 0;
2474                     for(i=0; i<2; i++){
2475                         j= s->field_select[0][i] = s->p_field_select_table[i][xy];
2476                         s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
2477                         s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
2478                     }
2479                     break;
2480                 case CANDIDATE_MB_TYPE_INTER4V:
2481                     s->mv_dir = MV_DIR_FORWARD;
2482                     s->mv_type = MV_TYPE_8X8;
2483                     s->mb_intra= 0;
2484                     for(i=0; i<4; i++){
2485                         s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
2486                         s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
2487                     }
2488                     break;
2489                 case CANDIDATE_MB_TYPE_DIRECT:
2490                     if (ENABLE_MPEG4_ENCODER) {
2491                         s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
2492                         s->mb_intra= 0;
2493                         motion_x=s->b_direct_mv_table[xy][0];
2494                         motion_y=s->b_direct_mv_table[xy][1];
2495                         ff_mpeg4_set_direct_mv(s, motion_x, motion_y);
2496                     }
2497                     break;
2498                 case CANDIDATE_MB_TYPE_DIRECT0:
2499                     if (ENABLE_MPEG4_ENCODER) {
2500                         s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
2501                         s->mb_intra= 0;
2502                         ff_mpeg4_set_direct_mv(s, 0, 0);
2503                     }
2504                     break;
2505                 case CANDIDATE_MB_TYPE_BIDIR:
2506                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2507                     s->mb_intra= 0;
2508                     s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
2509                     s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
2510                     s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
2511                     s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2512                     break;
2513                 case CANDIDATE_MB_TYPE_BACKWARD:
2514                     s->mv_dir = MV_DIR_BACKWARD;
2515                     s->mb_intra= 0;
2516                     motion_x= s->mv[1][0][0] = s->b_back_mv_table[xy][0];
2517                     motion_y= s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2518                     break;
2519                 case CANDIDATE_MB_TYPE_FORWARD:
2520                     s->mv_dir = MV_DIR_FORWARD;
2521                     s->mb_intra= 0;
2522                     motion_x= s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
2523                     motion_y= s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2524 //                    printf(" %d %d ", motion_x, motion_y);
2525                     break;
2526                 case CANDIDATE_MB_TYPE_FORWARD_I:
2527                     s->mv_dir = MV_DIR_FORWARD;
2528                     s->mv_type = MV_TYPE_FIELD;
2529                     s->mb_intra= 0;
2530                     for(i=0; i<2; i++){
2531                         j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
2532                         s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
2533                         s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
2534                     }
2535                     break;
2536                 case CANDIDATE_MB_TYPE_BACKWARD_I:
2537                     s->mv_dir = MV_DIR_BACKWARD;
2538                     s->mv_type = MV_TYPE_FIELD;
2539                     s->mb_intra= 0;
2540                     for(i=0; i<2; i++){
2541                         j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
2542                         s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
2543                         s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
2544                     }
2545                     break;
2546                 case CANDIDATE_MB_TYPE_BIDIR_I:
2547                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2548                     s->mv_type = MV_TYPE_FIELD;
2549                     s->mb_intra= 0;
2550                     for(dir=0; dir<2; dir++){
2551                         for(i=0; i<2; i++){
2552                             j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
2553                             s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
2554                             s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
2555                         }
2556                     }
2557                     break;
2558                 default:
2559                     av_log(s->avctx, AV_LOG_ERROR, "illegal MB type\n");
2560                 }
2561
2562                 encode_mb(s, motion_x, motion_y);
2563
2564                 // RAL: Update last macroblock type
2565                 s->last_mv_dir = s->mv_dir;
2566
2567                 if (ENABLE_ANY_H263_ENCODER &&
2568                     s->out_format == FMT_H263 && s->pict_type!=B_TYPE)
2569                     ff_h263_update_motion_val(s);
2570
2571                 MPV_decode_mb(s, s->block);
2572             }
2573
2574             /* clean the MV table in IPS frames for direct mode in B frames */
2575             if(s->mb_intra /* && I,P,S_TYPE */){
2576                 s->p_mv_table[xy][0]=0;
2577                 s->p_mv_table[xy][1]=0;
2578             }
2579
2580             if(s->flags&CODEC_FLAG_PSNR){
2581                 int w= 16;
2582                 int h= 16;
2583
2584                 if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
2585                 if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
2586
2587                 s->current_picture.error[0] += sse(
2588                     s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16,
2589                     s->dest[0], w, h, s->linesize);
2590                 s->current_picture.error[1] += sse(
2591                     s, s->new_picture.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,
2592                     s->dest[1], w>>1, h>>1, s->uvlinesize);
2593                 s->current_picture.error[2] += sse(
2594                     s, s->new_picture    .data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,
2595                     s->dest[2], w>>1, h>>1, s->uvlinesize);
2596             }
2597             if(s->loop_filter){
2598                 if(ENABLE_ANY_H263_ENCODER && s->out_format == FMT_H263)
2599                     ff_h263_loop_filter(s);
2600             }
2601 //printf("MB %d %d bits\n", s->mb_x+s->mb_y*s->mb_stride, put_bits_count(&s->pb));
2602         }
2603     }
2604
2605     //not beautiful here but we must write it before flushing so it has to be here
2606     if (ENABLE_MSMPEG4_ENCODER && s->msmpeg4_version && s->msmpeg4_version<4 && s->pict_type == I_TYPE)
2607         msmpeg4_encode_ext_header(s);
2608
2609     write_slice_end(s);
2610
2611     /* Send the last GOB if RTP */
2612     if (s->avctx->rtp_callback) {
2613         int number_mb = (mb_y - s->resync_mb_y)*s->mb_width - s->resync_mb_x;
2614         pdif = pbBufPtr(&s->pb) - s->ptr_lastgob;
2615         /* Call the RTP callback to send the last GOB */
2616         emms_c();
2617         s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, pdif, number_mb);
2618     }
2619
2620     return 0;
2621 }
2622
2623 #define MERGE(field) dst->field += src->field; src->field=0
2624 static void merge_context_after_me(MpegEncContext *dst, MpegEncContext *src){
2625     MERGE(me.scene_change_score);
2626     MERGE(me.mc_mb_var_sum_temp);
2627     MERGE(me.mb_var_sum_temp);
2628 }
2629
2630 static void merge_context_after_encode(MpegEncContext *dst, MpegEncContext *src){
2631     int i;
2632
2633     MERGE(dct_count[0]); //note, the other dct vars are not part of the context
2634     MERGE(dct_count[1]);
2635     MERGE(mv_bits);
2636     MERGE(i_tex_bits);
2637     MERGE(p_tex_bits);
2638     MERGE(i_count);
2639     MERGE(f_count);
2640     MERGE(b_count);
2641     MERGE(skip_count);
2642     MERGE(misc_bits);
2643     MERGE(error_count);
2644     MERGE(padding_bug_score);
2645     MERGE(current_picture.error[0]);
2646     MERGE(current_picture.error[1]);
2647     MERGE(current_picture.error[2]);
2648
2649     if(dst->avctx->noise_reduction){
2650         for(i=0; i<64; i++){
2651             MERGE(dct_error_sum[0][i]);
2652             MERGE(dct_error_sum[1][i]);
2653         }
2654     }
2655
2656     assert(put_bits_count(&src->pb) % 8 ==0);
2657     assert(put_bits_count(&dst->pb) % 8 ==0);
2658     ff_copy_bits(&dst->pb, src->pb.buf, put_bits_count(&src->pb));
2659     flush_put_bits(&dst->pb);
2660 }
2661
2662 static int estimate_qp(MpegEncContext *s, int dry_run){
2663     if (s->next_lambda){
2664         s->current_picture_ptr->quality=
2665         s->current_picture.quality = s->next_lambda;
2666         if(!dry_run) s->next_lambda= 0;
2667     } else if (!s->fixed_qscale) {
2668         s->current_picture_ptr->quality=
2669         s->current_picture.quality = ff_rate_estimate_qscale(s, dry_run);
2670         if (s->current_picture.quality < 0)
2671             return -1;
2672     }
2673
2674     if(s->adaptive_quant){
2675         switch(s->codec_id){
2676         case CODEC_ID_MPEG4:
2677             if (ENABLE_MPEG4_ENCODER)
2678                 ff_clean_mpeg4_qscales(s);
2679             break;
2680         case CODEC_ID_H263:
2681         case CODEC_ID_H263P:
2682         case CODEC_ID_FLV1:
2683             if (ENABLE_H263_ENCODER||ENABLE_H263P_ENCODER||ENABLE_FLV_ENCODER)
2684                 ff_clean_h263_qscales(s);
2685             break;
2686         }
2687
2688         s->lambda= s->lambda_table[0];
2689         //FIXME broken
2690     }else
2691         s->lambda= s->current_picture.quality;
2692 //printf("%d %d\n", s->avctx->global_quality, s->current_picture.quality);
2693     update_qscale(s);
2694     return 0;
2695 }
2696
2697 /* must be called before writing the header */
2698 static void set_frame_distances(MpegEncContext * s){
2699     assert(s->current_picture_ptr->pts != AV_NOPTS_VALUE);
2700     s->time= s->current_picture_ptr->pts*s->avctx->time_base.num;
2701
2702     if(s->pict_type==B_TYPE){
2703         s->pb_time= s->pp_time - (s->last_non_b_time - s->time);
2704         assert(s->pb_time > 0 && s->pb_time < s->pp_time);
2705     }else{
2706         s->pp_time= s->time - s->last_non_b_time;
2707         s->last_non_b_time= s->time;
2708         assert(s->picture_number==0 || s->pp_time > 0);
2709     }
2710 }
2711
2712 static int encode_picture(MpegEncContext *s, int picture_number)
2713 {
2714     int i;
2715     int bits;
2716
2717     s->picture_number = picture_number;
2718
2719     /* Reset the average MB variance */
2720     s->me.mb_var_sum_temp    =
2721     s->me.mc_mb_var_sum_temp = 0;
2722
2723     /* we need to initialize some time vars before we can encode b-frames */
2724     // RAL: Condition added for MPEG1VIDEO
2725     if (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO || (s->h263_pred && !s->h263_msmpeg4))
2726         set_frame_distances(s);
2727     if(ENABLE_MPEG4_ENCODER && s->codec_id == CODEC_ID_MPEG4)
2728         ff_set_mpeg4_time(s);
2729
2730     s->me.scene_change_score=0;
2731
2732 //    s->lambda= s->current_picture_ptr->quality; //FIXME qscale / ... stuff for ME ratedistoration
2733
2734     if(s->pict_type==I_TYPE){
2735         if(s->msmpeg4_version >= 3) s->no_rounding=1;
2736         else                        s->no_rounding=0;
2737     }else if(s->pict_type!=B_TYPE){
2738         if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
2739             s->no_rounding ^= 1;
2740     }
2741
2742     if(s->flags & CODEC_FLAG_PASS2){
2743         if (estimate_qp(s,1) < 0)
2744             return -1;
2745         ff_get_2pass_fcode(s);
2746     }else if(!(s->flags & CODEC_FLAG_QSCALE)){
2747         if(s->pict_type==B_TYPE)
2748             s->lambda= s->last_lambda_for[s->pict_type];
2749         else
2750             s->lambda= s->last_lambda_for[s->last_non_b_pict_type];
2751         update_qscale(s);
2752     }
2753
2754     s->mb_intra=0; //for the rate distortion & bit compare functions
2755     for(i=1; i<s->avctx->thread_count; i++){
2756         ff_update_duplicate_context(s->thread_context[i], s);
2757     }
2758
2759     ff_init_me(s);
2760
2761     /* Estimate motion for every MB */
2762     if(s->pict_type != I_TYPE){
2763         s->lambda = (s->lambda * s->avctx->me_penalty_compensation + 128)>>8;
2764         s->lambda2= (s->lambda2* (int64_t)s->avctx->me_penalty_compensation + 128)>>8;
2765         if(s->pict_type != B_TYPE && s->avctx->me_threshold==0){
2766             if((s->avctx->pre_me && s->last_non_b_pict_type==I_TYPE) || s->avctx->pre_me==2){
2767                 s->avctx->execute(s->avctx, pre_estimate_motion_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count);
2768             }
2769         }
2770
2771         s->avctx->execute(s->avctx, estimate_motion_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count);
2772     }else /* if(s->pict_type == I_TYPE) */{
2773         /* I-Frame */
2774         for(i=0; i<s->mb_stride*s->mb_height; i++)
2775             s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
2776
2777         if(!s->fixed_qscale){
2778             /* finding spatial complexity for I-frame rate control */
2779             s->avctx->execute(s->avctx, mb_var_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count);
2780         }
2781     }
2782     for(i=1; i<s->avctx->thread_count; i++){
2783         merge_context_after_me(s, s->thread_context[i]);
2784     }
2785     s->current_picture.mc_mb_var_sum= s->current_picture_ptr->mc_mb_var_sum= s->me.mc_mb_var_sum_temp;
2786     s->current_picture.   mb_var_sum= s->current_picture_ptr->   mb_var_sum= s->me.   mb_var_sum_temp;
2787     emms_c();
2788
2789     if(s->me.scene_change_score > s->avctx->scenechange_threshold && s->pict_type == P_TYPE){
2790         s->pict_type= I_TYPE;
2791         for(i=0; i<s->mb_stride*s->mb_height; i++)
2792             s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
2793 //printf("Scene change detected, encoding as I Frame %d %d\n", s->current_picture.mb_var_sum, s->current_picture.mc_mb_var_sum);
2794     }
2795
2796     if(!s->umvplus){
2797         if(s->pict_type==P_TYPE || s->pict_type==S_TYPE) {
2798             s->f_code= ff_get_best_fcode(s, s->p_mv_table, CANDIDATE_MB_TYPE_INTER);
2799
2800             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2801                 int a,b;
2802                 a= ff_get_best_fcode(s, s->p_field_mv_table[0][0], CANDIDATE_MB_TYPE_INTER_I); //FIXME field_select
2803                 b= ff_get_best_fcode(s, s->p_field_mv_table[1][1], CANDIDATE_MB_TYPE_INTER_I);
2804                 s->f_code= FFMAX(s->f_code, FFMAX(a,b));
2805             }
2806
2807             ff_fix_long_p_mvs(s);
2808             ff_fix_long_mvs(s, NULL, 0, s->p_mv_table, s->f_code, CANDIDATE_MB_TYPE_INTER, 0);
2809             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2810                 int j;
2811                 for(i=0; i<2; i++){
2812                     for(j=0; j<2; j++)
2813                         ff_fix_long_mvs(s, s->p_field_select_table[i], j,
2814                                         s->p_field_mv_table[i][j], s->f_code, CANDIDATE_MB_TYPE_INTER_I, 0);
2815                 }
2816             }
2817         }
2818
2819         if(s->pict_type==B_TYPE){
2820             int a, b;
2821
2822             a = ff_get_best_fcode(s, s->b_forw_mv_table, CANDIDATE_MB_TYPE_FORWARD);
2823             b = ff_get_best_fcode(s, s->b_bidir_forw_mv_table, CANDIDATE_MB_TYPE_BIDIR);
2824             s->f_code = FFMAX(a, b);
2825
2826             a = ff_get_best_fcode(s, s->b_back_mv_table, CANDIDATE_MB_TYPE_BACKWARD);
2827             b = ff_get_best_fcode(s, s->b_bidir_back_mv_table, CANDIDATE_MB_TYPE_BIDIR);
2828             s->b_code = FFMAX(a, b);
2829
2830             ff_fix_long_mvs(s, NULL, 0, s->b_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_FORWARD, 1);
2831             ff_fix_long_mvs(s, NULL, 0, s->b_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BACKWARD, 1);
2832             ff_fix_long_mvs(s, NULL, 0, s->b_bidir_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_BIDIR, 1);
2833             ff_fix_long_mvs(s, NULL, 0, s->b_bidir_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BIDIR, 1);
2834             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2835                 int dir, j;
2836                 for(dir=0; dir<2; dir++){
2837                     for(i=0; i<2; i++){
2838                         for(j=0; j<2; j++){
2839                             int type= dir ? (CANDIDATE_MB_TYPE_BACKWARD_I|CANDIDATE_MB_TYPE_BIDIR_I)
2840                                           : (CANDIDATE_MB_TYPE_FORWARD_I |CANDIDATE_MB_TYPE_BIDIR_I);
2841                             ff_fix_long_mvs(s, s->b_field_select_table[dir][i], j,
2842                                             s->b_field_mv_table[dir][i][j], dir ? s->b_code : s->f_code, type, 1);
2843                         }
2844                     }
2845                 }
2846             }
2847         }
2848     }
2849
2850     if (estimate_qp(s, 0) < 0)
2851         return -1;
2852
2853     if(s->qscale < 3 && s->max_qcoeff<=128 && s->pict_type==I_TYPE && !(s->flags & CODEC_FLAG_QSCALE))
2854         s->qscale= 3; //reduce clipping problems
2855
2856     if (s->out_format == FMT_MJPEG) {
2857         /* for mjpeg, we do include qscale in the matrix */
2858         s->intra_matrix[0] = ff_mpeg1_default_intra_matrix[0];
2859         for(i=1;i<64;i++){
2860             int j= s->dsp.idct_permutation[i];
2861
2862             s->intra_matrix[j] = av_clip_uint8((ff_mpeg1_default_intra_matrix[i] * s->qscale) >> 3);
2863         }
2864         ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
2865                        s->intra_matrix, s->intra_quant_bias, 8, 8, 1);
2866         s->qscale= 8;
2867     }
2868
2869     //FIXME var duplication
2870     s->current_picture_ptr->key_frame=
2871     s->current_picture.key_frame= s->pict_type == I_TYPE; //FIXME pic_ptr
2872     s->current_picture_ptr->pict_type=
2873     s->current_picture.pict_type= s->pict_type;
2874
2875     if(s->current_picture.key_frame)
2876         s->picture_in_gop_number=0;
2877
2878     s->last_bits= put_bits_count(&s->pb);
2879     switch(s->out_format) {
2880     case FMT_MJPEG:
2881         if (ENABLE_MJPEG_ENCODER)
2882             ff_mjpeg_encode_picture_header(s);
2883         break;
2884     case FMT_H261:
2885         if (ENABLE_H261_ENCODER)
2886             ff_h261_encode_picture_header(s, picture_number);
2887         break;
2888     case FMT_H263:
2889         if (ENABLE_WMV2_ENCODER && s->codec_id == CODEC_ID_WMV2)
2890             ff_wmv2_encode_picture_header(s, picture_number);
2891         else if (ENABLE_MSMPEG4_ENCODER && s->h263_msmpeg4)
2892             msmpeg4_encode_picture_header(s, picture_number);
2893         else if (ENABLE_MPEG4_ENCODER && s->h263_pred)
2894             mpeg4_encode_picture_header(s, picture_number);
2895         else if (ENABLE_RV10_ENCODER && s->codec_id == CODEC_ID_RV10)
2896             rv10_encode_picture_header(s, picture_number);
2897         else if (ENABLE_RV20_ENCODER && s->codec_id == CODEC_ID_RV20)
2898             rv20_encode_picture_header(s, picture_number);
2899         else if (ENABLE_FLV_ENCODER && s->codec_id == CODEC_ID_FLV1)
2900             ff_flv_encode_picture_header(s, picture_number);
2901         else if (ENABLE_ANY_H263_ENCODER)
2902             h263_encode_picture_header(s, picture_number);
2903         break;
2904     case FMT_MPEG1:
2905         if (ENABLE_MPEG1VIDEO_ENCODER || ENABLE_MPEG2VIDEO_ENCODER)
2906             mpeg1_encode_picture_header(s, picture_number);
2907         break;
2908     case FMT_H264:
2909         break;
2910     default:
2911         assert(0);
2912     }
2913     bits= put_bits_count(&s->pb);
2914     s->header_bits= bits - s->last_bits;
2915
2916     for(i=1; i<s->avctx->thread_count; i++){
2917         update_duplicate_context_after_me(s->thread_context[i], s);
2918     }
2919     s->avctx->execute(s->avctx, encode_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count);
2920     for(i=1; i<s->avctx->thread_count; i++){
2921         merge_context_after_encode(s, s->thread_context[i]);
2922     }
2923     emms_c();
2924     return 0;
2925 }
2926
2927 void  denoise_dct_c(MpegEncContext *s, DCTELEM *block){
2928     const int intra= s->mb_intra;
2929     int i;
2930
2931     s->dct_count[intra]++;
2932
2933     for(i=0; i<64; i++){
2934         int level= block[i];
2935
2936         if(level){
2937             if(level>0){
2938                 s->dct_error_sum[intra][i] += level;
2939                 level -= s->dct_offset[intra][i];
2940                 if(level<0) level=0;
2941             }else{
2942                 s->dct_error_sum[intra][i] -= level;
2943                 level += s->dct_offset[intra][i];
2944                 if(level>0) level=0;
2945             }
2946             block[i]= level;
2947         }
2948     }
2949 }
2950
2951 int dct_quantize_trellis_c(MpegEncContext *s,
2952                         DCTELEM *block, int n,
2953                         int qscale, int *overflow){
2954     const int *qmat;
2955     const uint8_t *scantable= s->intra_scantable.scantable;
2956     const uint8_t *perm_scantable= s->intra_scantable.permutated;
2957     int max=0;
2958     unsigned int threshold1, threshold2;
2959     int bias=0;
2960     int run_tab[65];
2961     int level_tab[65];
2962     int score_tab[65];
2963     int survivor[65];
2964     int survivor_count;
2965     int last_run=0;
2966     int last_level=0;
2967     int last_score= 0;
2968     int last_i;
2969     int coeff[2][64];
2970     int coeff_count[64];
2971     int qmul, qadd, start_i, last_non_zero, i, dc;
2972     const int esc_length= s->ac_esc_length;
2973     uint8_t * length;
2974     uint8_t * last_length;
2975     const int lambda= s->lambda2 >> (FF_LAMBDA_SHIFT - 6);
2976
2977     s->dsp.fdct (block);
2978
2979     if(s->dct_error_sum)
2980         s->denoise_dct(s, block);
2981     qmul= qscale*16;
2982     qadd= ((qscale-1)|1)*8;
2983
2984     if (s->mb_intra) {
2985         int q;
2986         if (!s->h263_aic) {
2987             if (n < 4)
2988                 q = s->y_dc_scale;
2989             else
2990                 q = s->c_dc_scale;
2991             q = q << 3;
2992         } else{
2993             /* For AIC we skip quant/dequant of INTRADC */
2994             q = 1 << 3;
2995             qadd=0;
2996         }
2997
2998         /* note: block[0] is assumed to be positive */
2999         block[0] = (block[0] + (q >> 1)) / q;
3000         start_i = 1;
3001         last_non_zero = 0;
3002         qmat = s->q_intra_matrix[qscale];
3003         if(s->mpeg_quant || s->out_format == FMT_MPEG1)
3004             bias= 1<<(QMAT_SHIFT-1);
3005         length     = s->intra_ac_vlc_length;
3006         last_length= s->intra_ac_vlc_last_length;
3007     } else {
3008         start_i = 0;
3009         last_non_zero = -1;
3010         qmat = s->q_inter_matrix[qscale];
3011         length     = s->inter_ac_vlc_length;
3012         last_length= s->inter_ac_vlc_last_length;
3013     }
3014     last_i= start_i;
3015
3016     threshold1= (1<<QMAT_SHIFT) - bias - 1;
3017     threshold2= (threshold1<<1);
3018
3019     for(i=63; i>=start_i; i--) {
3020         const int j = scantable[i];
3021         int level = block[j] * qmat[j];
3022
3023         if(((unsigned)(level+threshold1))>threshold2){
3024             last_non_zero = i;
3025             break;
3026         }
3027     }
3028
3029     for(i=start_i; i<=last_non_zero; i++) {
3030         const int j = scantable[i];
3031         int level = block[j] * qmat[j];
3032
3033 //        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
3034 //           || bias-level >= (1<<(QMAT_SHIFT - 3))){
3035         if(((unsigned)(level+threshold1))>threshold2){
3036             if(level>0){
3037                 level= (bias + level)>>QMAT_SHIFT;
3038                 coeff[0][i]= level;
3039                 coeff[1][i]= level-1;
3040 //                coeff[2][k]= level-2;
3041             }else{
3042                 level= (bias - level)>>QMAT_SHIFT;
3043                 coeff[0][i]= -level;
3044                 coeff[1][i]= -level+1;
3045 //                coeff[2][k]= -level+2;
3046             }
3047             coeff_count[i]= FFMIN(level, 2);
3048             assert(coeff_count[i]);
3049             max |=level;
3050         }else{
3051             coeff[0][i]= (level>>31)|1;
3052             coeff_count[i]= 1;
3053         }
3054     }
3055
3056     *overflow= s->max_qcoeff < max; //overflow might have happened
3057
3058     if(last_non_zero < start_i){
3059         memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
3060         return last_non_zero;
3061     }
3062
3063     score_tab[start_i]= 0;
3064     survivor[0]= start_i;
3065     survivor_count= 1;
3066
3067     for(i=start_i; i<=last_non_zero; i++){
3068         int level_index, j;
3069         const int dct_coeff= FFABS(block[ scantable[i] ]);
3070         const int zero_distoration= dct_coeff*dct_coeff;
3071         int best_score=256*256*256*120;
3072         for(level_index=0; level_index < coeff_count[i]; level_index++){
3073             int distoration;
3074             int level= coeff[level_index][i];
3075             const int alevel= FFABS(level);
3076             int unquant_coeff;
3077
3078             assert(level);
3079
3080             if(s->out_format == FMT_H263){
3081                 unquant_coeff= alevel*qmul + qadd;
3082             }else{ //MPEG1
3083                 j= s->dsp.idct_permutation[ scantable[i] ]; //FIXME optimize
3084                 if(s->mb_intra){
3085                         unquant_coeff = (int)(  alevel  * qscale * s->intra_matrix[j]) >> 3;
3086                         unquant_coeff =   (unquant_coeff - 1) | 1;
3087                 }else{
3088                         unquant_coeff = (((  alevel  << 1) + 1) * qscale * ((int) s->inter_matrix[j])) >> 4;
3089                         unquant_coeff =   (unquant_coeff - 1) | 1;
3090                 }
3091                 unquant_coeff<<= 3;
3092             }
3093
3094             distoration= (unquant_coeff - dct_coeff) * (unquant_coeff - dct_coeff) - zero_distoration;
3095             level+=64;
3096             if((level&(~127)) == 0){
3097                 for(j=survivor_count-1; j>=0; j--){
3098                     int run= i - survivor[j];
3099                     int score= distoration + length[UNI_AC_ENC_INDEX(run, level)]*lambda;
3100                     score += score_tab[i-run];
3101
3102                     if(score < best_score){
3103                         best_score= score;
3104                         run_tab[i+1]= run;
3105                         level_tab[i+1]= level-64;
3106                     }
3107                 }
3108
3109                 if(s->out_format == FMT_H263){
3110                     for(j=survivor_count-1; j>=0; j--){
3111                         int run= i - survivor[j];
3112                         int score= distoration + last_length[UNI_AC_ENC_INDEX(run, level)]*lambda;
3113                         score += score_tab[i-run];
3114                         if(score < last_score){
3115                             last_score= score;
3116                             last_run= run;
3117                             last_level= level-64;
3118                             last_i= i+1;
3119                         }
3120                     }
3121                 }
3122             }else{
3123                 distoration += esc_length*lambda;
3124                 for(j=survivor_count-1; j>=0; j--){
3125                     int run= i - survivor[j];
3126                     int score= distoration + score_tab[i-run];
3127
3128                     if(score < best_score){
3129                         best_score= score;
3130                         run_tab[i+1]= run;
3131                         level_tab[i+1]= level-64;
3132                     }
3133                 }
3134
3135                 if(s->out_format == FMT_H263){
3136                   for(j=survivor_count-1; j>=0; j--){
3137                         int run= i - survivor[j];
3138                         int score= distoration + score_tab[i-run];
3139                         if(score < last_score){
3140                             last_score= score;
3141                             last_run= run;
3142                             last_level= level-64;
3143                             last_i= i+1;
3144                         }
3145                     }
3146                 }
3147             }
3148         }
3149
3150         score_tab[i+1]= best_score;
3151
3152         //Note: there is a vlc code in mpeg4 which is 1 bit shorter then another one with a shorter run and the same level
3153         if(last_non_zero <= 27){
3154             for(; survivor_count; survivor_count--){
3155                 if(score_tab[ survivor[survivor_count-1] ] <= best_score)
3156                     break;
3157             }
3158         }else{
3159             for(; survivor_count; survivor_count--){
3160                 if(score_tab[ survivor[survivor_count-1] ] <= best_score + lambda)
3161                     break;
3162             }
3163         }
3164
3165         survivor[ survivor_count++ ]= i+1;
3166     }
3167
3168     if(s->out_format != FMT_H263){
3169         last_score= 256*256*256*120;
3170         for(i= survivor[0]; i<=last_non_zero + 1; i++){
3171             int score= score_tab[i];
3172             if(i) score += lambda*2; //FIXME exacter?
3173
3174             if(score < last_score){
3175                 last_score= score;
3176                 last_i= i;
3177                 last_level= level_tab[i];
3178                 last_run= run_tab[i];
3179             }
3180         }
3181     }
3182
3183     s->coded_score[n] = last_score;
3184
3185     dc= FFABS(block[0]);
3186     last_non_zero= last_i - 1;
3187     memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
3188
3189     if(last_non_zero < start_i)
3190         return last_non_zero;
3191
3192     if(last_non_zero == 0 && start_i == 0){
3193         int best_level= 0;
3194         int best_score= dc * dc;
3195
3196         for(i=0; i<coeff_count[0]; i++){
3197             int level= coeff[i][0];
3198             int alevel= FFABS(level);
3199             int unquant_coeff, score, distortion;
3200
3201             if(s->out_format == FMT_H263){
3202                     unquant_coeff= (alevel*qmul + qadd)>>3;
3203             }else{ //MPEG1
3204                     unquant_coeff = (((  alevel  << 1) + 1) * qscale * ((int) s->inter_matrix[0])) >> 4;
3205                     unquant_coeff =   (unquant_coeff - 1) | 1;
3206             }
3207             unquant_coeff = (unquant_coeff + 4) >> 3;
3208             unquant_coeff<<= 3 + 3;
3209
3210             distortion= (unquant_coeff - dc) * (unquant_coeff - dc);
3211             level+=64;
3212             if((level&(~127)) == 0) score= distortion + last_length[UNI_AC_ENC_INDEX(0, level)]*lambda;
3213             else                    score= distortion + esc_length*lambda;
3214
3215             if(score < best_score){
3216                 best_score= score;
3217                 best_level= level - 64;
3218             }
3219         }
3220         block[0]= best_level;
3221         s->coded_score[n] = best_score - dc*dc;
3222         if(best_level == 0) return -1;
3223         else                return last_non_zero;
3224     }
3225
3226     i= last_i;
3227     assert(last_level);
3228
3229     block[ perm_scantable[last_non_zero] ]= last_level;
3230     i -= last_run + 1;
3231
3232     for(; i>start_i; i -= run_tab[i] + 1){
3233         block[ perm_scantable[i-1] ]= level_tab[i];
3234     }
3235
3236     return last_non_zero;
3237 }
3238
3239 //#define REFINE_STATS 1
3240 static int16_t basis[64][64];
3241
3242 static void build_basis(uint8_t *perm){
3243     int i, j, x, y;
3244     emms_c();
3245     for(i=0; i<8; i++){
3246         for(j=0; j<8; j++){
3247             for(y=0; y<8; y++){
3248                 for(x=0; x<8; x++){
3249                     double s= 0.25*(1<<BASIS_SHIFT);
3250                     int index= 8*i + j;
3251                     int perm_index= perm[index];
3252                     if(i==0) s*= sqrt(0.5);
3253                     if(j==0) s*= sqrt(0.5);
3254                     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)));
3255                 }
3256             }
3257         }
3258     }
3259 }
3260
3261 static int dct_quantize_refine(MpegEncContext *s, //FIXME breaks denoise?
3262                         DCTELEM *block, int16_t *weight, DCTELEM *orig,
3263                         int n, int qscale){
3264     int16_t rem[64];
3265     DECLARE_ALIGNED_16(DCTELEM, d1[64]);
3266     const int *qmat;
3267     const uint8_t *scantable= s->intra_scantable.scantable;
3268     const uint8_t *perm_scantable= s->intra_scantable.permutated;
3269 //    unsigned int threshold1, threshold2;
3270 //    int bias=0;
3271     int run_tab[65];
3272     int prev_run=0;
3273     int prev_level=0;
3274     int qmul, qadd, start_i, last_non_zero, i, dc;
3275     uint8_t * length;
3276     uint8_t * last_length;
3277     int lambda;
3278     int rle_index, run, q = 1, sum; //q is only used when s->mb_intra is true
3279 #ifdef REFINE_STATS
3280 static int count=0;
3281 static int after_last=0;
3282 static int to_zero=0;
3283 static int from_zero=0;
3284 static int raise=0;
3285 static int lower=0;
3286 static int messed_sign=0;
3287 #endif
3288
3289     if(basis[0][0] == 0)
3290         build_basis(s->dsp.idct_permutation);
3291
3292     qmul= qscale*2;
3293     qadd= (qscale-1)|1;
3294     if (s->mb_intra) {
3295         if (!s->h263_aic) {
3296             if (n < 4)
3297                 q = s->y_dc_scale;
3298             else
3299                 q = s->c_dc_scale;
3300         } else{
3301             /* For AIC we skip quant/dequant of INTRADC */
3302             q = 1;
3303             qadd=0;
3304         }
3305         q <<= RECON_SHIFT-3;
3306         /* note: block[0] is assumed to be positive */
3307         dc= block[0]*q;
3308 //        block[0] = (block[0] + (q >> 1)) / q;
3309         start_i = 1;
3310         qmat = s->q_intra_matrix[qscale];
3311 //        if(s->mpeg_quant || s->out_format == FMT_MPEG1)
3312 //            bias= 1<<(QMAT_SHIFT-1);
3313         length     = s->intra_ac_vlc_length;
3314         last_length= s->intra_ac_vlc_last_length;
3315     } else {
3316         dc= 0;
3317         start_i = 0;
3318         qmat = s->q_inter_matrix[qscale];
3319         length     = s->inter_ac_vlc_length;
3320         last_length= s->inter_ac_vlc_last_length;
3321     }
3322     last_non_zero = s->block_last_index[n];
3323
3324 #ifdef REFINE_STATS
3325 {START_TIMER
3326 #endif
3327     dc += (1<<(RECON_SHIFT-1));
3328     for(i=0; i<64; i++){
3329         rem[i]= dc - (orig[i]<<RECON_SHIFT); //FIXME  use orig dirrectly instead of copying to rem[]
3330     }
3331 #ifdef REFINE_STATS
3332 STOP_TIMER("memset rem[]")}
3333 #endif
3334     sum=0;
3335     for(i=0; i<64; i++){
3336         int one= 36;
3337         int qns=4;
3338         int w;
3339
3340         w= FFABS(weight[i]) + qns*one;
3341         w= 15 + (48*qns*one + w/2)/w; // 16 .. 63
3342
3343         weight[i] = w;
3344 //        w=weight[i] = (63*qns + (w/2)) / w;
3345
3346         assert(w>0);
3347         assert(w<(1<<6));
3348         sum += w*w;
3349     }
3350     lambda= sum*(uint64_t)s->lambda2 >> (FF_LAMBDA_SHIFT - 6 + 6 + 6 + 6);
3351 #ifdef REFINE_STATS
3352 {START_TIMER
3353 #endif
3354     run=0;
3355     rle_index=0;
3356     for(i=start_i; i<=last_non_zero; i++){
3357         int j= perm_scantable[i];
3358         const int level= block[j];
3359         int coeff;
3360
3361         if(level){
3362             if(level<0) coeff= qmul*level - qadd;
3363             else        coeff= qmul*level + qadd;
3364             run_tab[rle_index++]=run;
3365             run=0;
3366
3367             s->dsp.add_8x8basis(rem, basis[j], coeff);
3368         }else{
3369             run++;
3370         }
3371     }
3372 #ifdef REFINE_STATS
3373 if(last_non_zero>0){
3374 STOP_TIMER("init rem[]")
3375 }
3376 }
3377
3378 {START_TIMER
3379 #endif
3380     for(;;){
3381         int best_score=s->dsp.try_8x8basis(rem, weight, basis[0], 0);
3382         int best_coeff=0;
3383         int best_change=0;
3384         int run2, best_unquant_change=0, analyze_gradient;
3385 #ifdef REFINE_STATS
3386 {START_TIMER
3387 #endif
3388         analyze_gradient = last_non_zero > 2 || s->avctx->quantizer_noise_shaping >= 3;
3389
3390         if(analyze_gradient){
3391 #ifdef REFINE_STATS
3392 {START_TIMER
3393 #endif
3394             for(i=0; i<64; i++){
3395                 int w= weight[i];
3396
3397                 d1[i] = (rem[i]*w*w + (1<<(RECON_SHIFT+12-1)))>>(RECON_SHIFT+12);
3398             }
3399 #ifdef REFINE_STATS
3400 STOP_TIMER("rem*w*w")}
3401 {START_TIMER
3402 #endif
3403             s->dsp.fdct(d1);
3404 #ifdef REFINE_STATS
3405 STOP_TIMER("dct")}
3406 #endif
3407         }
3408
3409         if(start_i){
3410             const int level= block[0];
3411             int change, old_coeff;
3412
3413             assert(s->mb_intra);
3414
3415             old_coeff= q*level;
3416
3417             for(change=-1; change<=1; change+=2){
3418                 int new_level= level + change;
3419                 int score, new_coeff;
3420
3421                 new_coeff= q*new_level;
3422                 if(new_coeff >= 2048 || new_coeff < 0)
3423                     continue;
3424
3425                 score= s->dsp.try_8x8basis(rem, weight, basis[0], new_coeff - old_coeff);
3426                 if(score<best_score){
3427                     best_score= score;
3428                     best_coeff= 0;
3429                     best_change= change;
3430                     best_unquant_change= new_coeff - old_coeff;
3431                 }
3432             }
3433         }
3434
3435         run=0;
3436         rle_index=0;
3437         run2= run_tab[rle_index++];
3438         prev_level=0;
3439         prev_run=0;
3440
3441         for(i=start_i; i<64; i++){
3442             int j= perm_scantable[i];
3443             const int level= block[j];
3444             int change, old_coeff;
3445
3446             if(s->avctx->quantizer_noise_shaping < 3 && i > last_non_zero + 1)
3447                 break;
3448
3449             if(level){
3450                 if(level<0) old_coeff= qmul*level - qadd;
3451                 else        old_coeff= qmul*level + qadd;
3452                 run2= run_tab[rle_index++]; //FIXME ! maybe after last
3453             }else{
3454                 old_coeff=0;
3455                 run2--;
3456                 assert(run2>=0 || i >= last_non_zero );
3457             }
3458
3459             for(change=-1; change<=1; change+=2){
3460                 int new_level= level + change;
3461                 int score, new_coeff, unquant_change;
3462
3463                 score=0;
3464                 if(s->avctx->quantizer_noise_shaping < 2 && FFABS(new_level) > FFABS(level))
3465                    continue;
3466
3467                 if(new_level){
3468                     if(new_level<0) new_coeff= qmul*new_level - qadd;
3469                     else            new_coeff= qmul*new_level + qadd;
3470                     if(new_coeff >= 2048 || new_coeff <= -2048)
3471                         continue;
3472                     //FIXME check for overflow
3473
3474                     if(level){
3475                         if(level < 63 && level > -63){
3476                             if(i < last_non_zero)
3477                                 score +=   length[UNI_AC_ENC_INDEX(run, new_level+64)]
3478                                          - length[UNI_AC_ENC_INDEX(run, level+64)];
3479                             else
3480                                 score +=   last_length[UNI_AC_ENC_INDEX(run, new_level+64)]
3481                                          - last_length[UNI_AC_ENC_INDEX(run, level+64)];
3482                         }
3483                     }else{
3484                         assert(FFABS(new_level)==1);
3485
3486                         if(analyze_gradient){
3487                             int g= d1[ scantable[i] ];
3488                             if(g && (g^new_level) >= 0)
3489                                 continue;
3490                         }
3491
3492                         if(i < last_non_zero){
3493                             int next_i= i + run2 + 1;
3494                             int next_level= block[ perm_scantable[next_i] ] + 64;
3495
3496                             if(next_level&(~127))
3497                                 next_level= 0;
3498
3499                             if(next_i < last_non_zero)
3500                                 score +=   length[UNI_AC_ENC_INDEX(run, 65)]
3501                                          + length[UNI_AC_ENC_INDEX(run2, next_level)]
3502                                          - length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)];
3503                             else
3504                                 score +=  length[UNI_AC_ENC_INDEX(run, 65)]
3505                                         + last_length[UNI_AC_ENC_INDEX(run2, next_level)]
3506                                         - last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)];
3507                         }else{
3508                             score += last_length[UNI_AC_ENC_INDEX(run, 65)];
3509                             if(prev_level){
3510                                 score +=  length[UNI_AC_ENC_INDEX(prev_run, prev_level)]
3511                                         - last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)];
3512                             }
3513                         }
3514                     }
3515                 }else{
3516                     new_coeff=0;
3517                     assert(FFABS(level)==1);
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 + run2 + 1, next_level)]
3528                                      - length[UNI_AC_ENC_INDEX(run2, next_level)]
3529                                      - length[UNI_AC_ENC_INDEX(run, 65)];
3530                         else
3531                             score +=   last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)]
3532                                      - last_length[UNI_AC_ENC_INDEX(run2, next_level)]
3533                                      - length[UNI_AC_ENC_INDEX(run, 65)];
3534                     }else{
3535                         score += -last_length[UNI_AC_ENC_INDEX(run, 65)];
3536                         if(prev_level){
3537                             score +=  last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)]
3538                                     - length[UNI_AC_ENC_INDEX(prev_run, prev_level)];
3539                         }
3540                     }
3541                 }
3542
3543                 score *= lambda;
3544
3545                 unquant_change= new_coeff - old_coeff;
3546                 assert((score < 100*lambda && score > -100*lambda) || lambda==0);
3547
3548                 score+= s->dsp.try_8x8basis(rem, weight, basis[j], unquant_change);
3549                 if(score<best_score){
3550                     best_score= score;
3551                     best_coeff= i;
3552                     best_change= change;
3553                     best_unquant_change= unquant_change;
3554                 }
3555             }
3556             if(level){
3557                 prev_level= level + 64;
3558                 if(prev_level&(~127))
3559                     prev_level= 0;
3560                 prev_run= run;
3561                 run=0;
3562             }else{
3563                 run++;
3564             }
3565         }
3566 #ifdef REFINE_STATS
3567 STOP_TIMER("iterative step")}
3568 #endif
3569
3570         if(best_change){
3571             int j= perm_scantable[ best_coeff ];
3572
3573             block[j] += best_change;
3574
3575             if(best_coeff > last_non_zero){
3576                 last_non_zero= best_coeff;
3577                 assert(block[j]);
3578 #ifdef REFINE_STATS
3579 after_last++;
3580 #endif
3581             }else{
3582 #ifdef REFINE_STATS
3583 if(block[j]){
3584     if(block[j] - best_change){
3585         if(FFABS(block[j]) > FFABS(block[j] - best_change)){
3586             raise++;
3587         }else{
3588             lower++;
3589         }
3590     }else{
3591         from_zero++;
3592     }
3593 }else{
3594     to_zero++;
3595 }
3596 #endif
3597                 for(; last_non_zero>=start_i; last_non_zero--){
3598                     if(block[perm_scantable[last_non_zero]])
3599                         break;
3600                 }
3601             }
3602 #ifdef REFINE_STATS
3603 count++;
3604 if(256*256*256*64 % count == 0){
3605     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);
3606 }
3607 #endif
3608             run=0;
3609             rle_index=0;
3610             for(i=start_i; i<=last_non_zero; i++){
3611                 int j= perm_scantable[i];
3612                 const int level= block[j];
3613
3614                  if(level){
3615                      run_tab[rle_index++]=run;
3616                      run=0;
3617                  }else{
3618                      run++;
3619                  }
3620             }
3621
3622             s->dsp.add_8x8basis(rem, basis[j], best_unquant_change);
3623         }else{
3624             break;
3625         }
3626     }
3627 #ifdef REFINE_STATS
3628 if(last_non_zero>0){
3629 STOP_TIMER("iterative search")
3630 }
3631 }
3632 #endif
3633
3634     return last_non_zero;
3635 }
3636
3637 int dct_quantize_c(MpegEncContext *s,
3638                         DCTELEM *block, int n,
3639                         int qscale, int *overflow)
3640 {
3641     int i, j, level, last_non_zero, q, start_i;
3642     const int *qmat;
3643     const uint8_t *scantable= s->intra_scantable.scantable;
3644     int bias;
3645     int max=0;
3646     unsigned int threshold1, threshold2;
3647
3648     s->dsp.fdct (block);
3649
3650     if(s->dct_error_sum)
3651         s->denoise_dct(s, block);
3652
3653     if (s->mb_intra) {
3654         if (!s->h263_aic) {
3655             if (n < 4)
3656                 q = s->y_dc_scale;
3657             else
3658                 q = s->c_dc_scale;
3659             q = q << 3;
3660         } else
3661             /* For AIC we skip quant/dequant of INTRADC */
3662             q = 1 << 3;
3663
3664         /* note: block[0] is assumed to be positive */
3665         block[0] = (block[0] + (q >> 1)) / q;
3666         start_i = 1;
3667         last_non_zero = 0;
3668         qmat = s->q_intra_matrix[qscale];
3669         bias= s->intra_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT);
3670     } else {
3671         start_i = 0;
3672         last_non_zero = -1;
3673         qmat = s->q_inter_matrix[qscale];
3674         bias= s->inter_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT);
3675     }
3676     threshold1= (1<<QMAT_SHIFT) - bias - 1;
3677     threshold2= (threshold1<<1);
3678     for(i=63;i>=start_i;i--) {
3679         j = scantable[i];
3680         level = block[j] * qmat[j];
3681
3682         if(((unsigned)(level+threshold1))>threshold2){
3683             last_non_zero = i;
3684             break;
3685         }else{
3686             block[j]=0;
3687         }
3688     }
3689     for(i=start_i; i<=last_non_zero; i++) {
3690         j = scantable[i];
3691         level = block[j] * qmat[j];
3692
3693 //        if(   bias+level >= (1<<QMAT_SHIFT)
3694 //           || bias-level >= (1<<QMAT_SHIFT)){
3695         if(((unsigned)(level+threshold1))>threshold2){
3696             if(level>0){
3697                 level= (bias + level)>>QMAT_SHIFT;
3698                 block[j]= level;
3699             }else{
3700                 level= (bias - level)>>QMAT_SHIFT;
3701                 block[j]= -level;
3702             }
3703             max |=level;
3704         }else{
3705             block[j]=0;
3706         }
3707     }
3708     *overflow= s->max_qcoeff < max; //overflow might have happened
3709
3710     /* we need this permutation so that we correct the IDCT, we only permute the !=0 elements */
3711     if (s->dsp.idct_permutation_type != FF_NO_IDCT_PERM)
3712         ff_block_permute(block, s->dsp.idct_permutation, scantable, last_non_zero);
3713
3714     return last_non_zero;
3715 }
3716
3717 AVCodec h263_encoder = {
3718     "h263",
3719     CODEC_TYPE_VIDEO,
3720     CODEC_ID_H263,
3721     sizeof(MpegEncContext),
3722     MPV_encode_init,
3723     MPV_encode_picture,
3724     MPV_encode_end,
3725     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
3726 };
3727
3728 AVCodec h263p_encoder = {
3729     "h263p",
3730     CODEC_TYPE_VIDEO,
3731     CODEC_ID_H263P,
3732     sizeof(MpegEncContext),
3733     MPV_encode_init,
3734     MPV_encode_picture,
3735     MPV_encode_end,
3736     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
3737 };
3738
3739 AVCodec flv_encoder = {
3740     "flv",
3741     CODEC_TYPE_VIDEO,
3742     CODEC_ID_FLV1,
3743     sizeof(MpegEncContext),
3744     MPV_encode_init,
3745     MPV_encode_picture,
3746     MPV_encode_end,
3747     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
3748 };
3749
3750 AVCodec rv10_encoder = {
3751     "rv10",
3752     CODEC_TYPE_VIDEO,
3753     CODEC_ID_RV10,
3754     sizeof(MpegEncContext),
3755     MPV_encode_init,
3756     MPV_encode_picture,
3757     MPV_encode_end,
3758     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
3759 };
3760
3761 AVCodec rv20_encoder = {
3762     "rv20",
3763     CODEC_TYPE_VIDEO,
3764     CODEC_ID_RV20,
3765     sizeof(MpegEncContext),
3766     MPV_encode_init,
3767     MPV_encode_picture,
3768     MPV_encode_end,
3769     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
3770 };
3771
3772 AVCodec mpeg4_encoder = {
3773     "mpeg4",
3774     CODEC_TYPE_VIDEO,
3775     CODEC_ID_MPEG4,
3776     sizeof(MpegEncContext),
3777     MPV_encode_init,
3778     MPV_encode_picture,
3779     MPV_encode_end,
3780     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
3781     .capabilities= CODEC_CAP_DELAY,
3782 };
3783
3784 AVCodec msmpeg4v1_encoder = {
3785     "msmpeg4v1",
3786     CODEC_TYPE_VIDEO,
3787     CODEC_ID_MSMPEG4V1,
3788     sizeof(MpegEncContext),
3789     MPV_encode_init,
3790     MPV_encode_picture,
3791     MPV_encode_end,
3792     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
3793 };
3794
3795 AVCodec msmpeg4v2_encoder = {
3796     "msmpeg4v2",
3797     CODEC_TYPE_VIDEO,
3798     CODEC_ID_MSMPEG4V2,
3799     sizeof(MpegEncContext),
3800     MPV_encode_init,
3801     MPV_encode_picture,
3802     MPV_encode_end,
3803     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
3804 };
3805
3806 AVCodec msmpeg4v3_encoder = {
3807     "msmpeg4",
3808     CODEC_TYPE_VIDEO,
3809     CODEC_ID_MSMPEG4V3,
3810     sizeof(MpegEncContext),
3811     MPV_encode_init,
3812     MPV_encode_picture,
3813     MPV_encode_end,
3814     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
3815 };
3816
3817 AVCodec wmv1_encoder = {
3818     "wmv1",
3819     CODEC_TYPE_VIDEO,
3820     CODEC_ID_WMV1,
3821     sizeof(MpegEncContext),
3822     MPV_encode_init,
3823     MPV_encode_picture,
3824     MPV_encode_end,
3825     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
3826 };