libavcodec/mpegvideo_enc.c

   1 /*
   2  * The simplest mpeg encoder (well, it was the simplest!)
   3  * Copyright (c) 2000,2001 Fabrice Bellard
   4  * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
   5  *
   6  * 4MV & hq & B-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
   7  *
   8  * This file is part of FFmpeg.
   9  *
  10  * FFmpeg is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU Lesser General Public
  12  * License as published by the Free Software Foundation; either
  13  * version 2.1 of the License, or (at your option) any later version.
  14  *
  15  * FFmpeg is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  * Lesser General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU Lesser General Public
  21  * License along with FFmpeg; if not, write to the Free Software
  22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  23  */
  24
  25 /**
  26  * @file libavcodec/mpegvideo_enc.c
  27  * The simplest mpeg encoder (well, it was the simplest!).
  28  */
  29
  30 #include "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 "aandcttab.h"
  39 #include <limits.h>
  40
  41 //#undef NDEBUG
  42 //#include <assert.h>
  43
  44 static int encode_picture(MpegEncContext *s, int picture_number);
  45 static int dct_quantize_refine(MpegEncContext *s, DCTELEM *block, int16_t *weight, DCTELEM *orig, int n, int qscale);
  46 static int sse_mb(MpegEncContext *s);
  47
  48 /* enable all paranoid tests for rounding, overflows, etc... */
  49 //#define PARANOID
  50
  51 //#define DEBUG
  52
  53 static uint8_t default_mv_penalty[MAX_FCODE+1][MAX_MV*2+1];
  54 static uint8_t default_fcode_tab[MAX_MV*2+1];
  55
  56 void ff_convert_matrix(DSPContext *dsp, int (*qmat)[64], uint16_t (*qmat16)[2][64],
  57                            const uint16_t *quant_matrix, int bias, int qmin, int qmax, int intra)
  58 {
  59     int qscale;
  60     int shift=0;
  61
  62     for(qscale=qmin; qscale<=qmax; qscale++){
  63         int i;
  64         if (dsp->fdct == ff_jpeg_fdct_islow
  65 #ifdef FAAN_POSTSCALE
  66             || dsp->fdct == ff_faandct
  67 #endif
  68             ) {
  69             for(i=0;i<64;i++) {
  70                 const int j= dsp->idct_permutation[i];
  71                 /* 16 <= qscale * quant_matrix[i] <= 7905 */
  72                 /* 19952             <= ff_aanscales[i] * qscale * quant_matrix[i]               <= 249205026 */
  73                 /* (1 << 36) / 19952 >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= (1 << 36) / 249205026 */
  74                 /* 3444240           >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= 275 */
  75
  76                 qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) /
  77                                 (qscale * quant_matrix[j]));
  78             }
  79         } else if (dsp->fdct == fdct_ifast
  80 #ifndef FAAN_POSTSCALE
  81                    || dsp->fdct == ff_faandct
  82 #endif
  83                    ) {
  84             for(i=0;i<64;i++) {
  85                 const int j= dsp->idct_permutation[i];
  86                 /* 16 <= qscale * quant_matrix[i] <= 7905 */
  87                 /* 19952             <= ff_aanscales[i] * qscale * quant_matrix[i]               <= 249205026 */
  88                 /* (1 << 36) / 19952 >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */
  89                 /* 3444240           >= (1 << 36) / (ff_aanscales[i] * qscale * quant_matrix[i]) >= 275 */
  90
  91                 qmat[qscale][i] = (int)((UINT64_C(1) << (QMAT_SHIFT + 14)) /
  92                                 (ff_aanscales[i] * qscale * quant_matrix[j]));
  93             }
  94         } else {
  95             for(i=0;i<64;i++) {
  96                 const int j= dsp->idct_permutation[i];
  97                 /* We can safely suppose that 16 <= quant_matrix[i] <= 255
  98                    So 16           <= qscale * quant_matrix[i]             <= 7905
  99                    so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905
 100                    so 32768        >= (1<<19) / (qscale * quant_matrix[i]) >= 67
 101                 */
 102                 qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) / (qscale * quant_matrix[j]));
 103 //                qmat  [qscale][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]);
 104                 qmat16[qscale][0][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[j]);
 105
 106                 if(qmat16[qscale][0][i]==0 || qmat16[qscale][0][i]==128*256) qmat16[qscale][0][i]=128*256-1;
 107                 qmat16[qscale][1][i]= ROUNDED_DIV(bias<<(16-QUANT_BIAS_SHIFT), qmat16[qscale][0][i]);
 108             }
 109         }
 110
 111         for(i=intra; i<64; i++){
 112             int64_t max= 8191;
 113             if (dsp->fdct == fdct_ifast
 114 #ifndef FAAN_POSTSCALE
 115                    || dsp->fdct == ff_faandct
 116 #endif
 117                    ) {
 118                 max = (8191LL*ff_aanscales[i]) >> 14;
 119             }
 120             while(((max * qmat[qscale][i]) >> shift) > INT_MAX){
 121                 shift++;
 122             }
 123         }
 124     }
 125     if(shift){
 126         av_log(NULL, AV_LOG_INFO, "Warning, QMAT_SHIFT is larger than %d, overflows possible\n", QMAT_SHIFT - shift);
 127     }
 128 }
 129
 130 static inline void update_qscale(MpegEncContext *s){
 131     s->qscale= (s->lambda*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7);
 132     s->qscale= av_clip(s->qscale, s->avctx->qmin, s->avctx->qmax);
 133
 134     s->lambda2= (s->lambda*s->lambda + FF_LAMBDA_SCALE/2) >> FF_LAMBDA_SHIFT;
 135 }
 136
 137 void ff_write_quant_matrix(PutBitContext *pb, uint16_t *matrix){
 138     int i;
 139
 140     if(matrix){
 141         put_bits(pb, 1, 1);
 142         for(i=0;i<64;i++) {
 143             put_bits(pb, 8, matrix[ ff_zigzag_direct[i] ]);
 144         }
 145     }else
 146         put_bits(pb, 1, 0);
 147 }
 148
 149 static void copy_picture_attributes(MpegEncContext *s, AVFrame *dst, AVFrame *src){
 150     int i;
 151
 152     dst->pict_type              = src->pict_type;
 153     dst->quality                = src->quality;
 154     dst->coded_picture_number   = src->coded_picture_number;
 155     dst->display_picture_number = src->display_picture_number;
 156 //    dst->reference              = src->reference;
 157     dst->pts                    = src->pts;
 158     dst->interlaced_frame       = src->interlaced_frame;
 159     dst->top_field_first        = src->top_field_first;
 160
 161     if(s->avctx->me_threshold){
 162         if(!src->motion_val[0])
 163             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_val not set!\n");
 164         if(!src->mb_type)
 165             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.mb_type not set!\n");
 166         if(!src->ref_index[0])
 167             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.ref_index not set!\n");
 168         if(src->motion_subsample_log2 != dst->motion_subsample_log2)
 169             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_subsample_log2 doesn't match! (%d!=%d)\n",
 170             src->motion_subsample_log2, dst->motion_subsample_log2);
 171
 172         memcpy(dst->mb_type, src->mb_type, s->mb_stride * s->mb_height * sizeof(dst->mb_type[0]));
 173
 174         for(i=0; i<2; i++){
 175             int stride= ((16*s->mb_width )>>src->motion_subsample_log2) + 1;
 176             int height= ((16*s->mb_height)>>src->motion_subsample_log2);
 177
 178             if(src->motion_val[i] && src->motion_val[i] != dst->motion_val[i]){
 179                 memcpy(dst->motion_val[i], src->motion_val[i], 2*stride*height*sizeof(int16_t));
 180             }
 181             if(src->ref_index[i] && src->ref_index[i] != dst->ref_index[i]){
 182                 memcpy(dst->ref_index[i], src->ref_index[i], s->b8_stride*2*s->mb_height*sizeof(int8_t));
 183             }
 184         }
 185     }
 186 }
 187
 188 static void update_duplicate_context_after_me(MpegEncContext *dst, MpegEncContext *src){
 189 #define COPY(a) dst->a= src->a
 190     COPY(pict_type);
 191     COPY(current_picture);
 192     COPY(f_code);
 193     COPY(b_code);
 194     COPY(qscale);
 195     COPY(lambda);
 196     COPY(lambda2);
 197     COPY(picture_in_gop_number);
 198     COPY(gop_picture_number);
 199     COPY(frame_pred_frame_dct); //FIXME don't set in encode_header
 200     COPY(progressive_frame); //FIXME don't set in encode_header
 201     COPY(partitioned_frame); //FIXME don't set in encode_header
 202 #undef COPY
 203 }
 204
 205 /**
 206  * sets the given MpegEncContext to defaults for encoding.
 207  * the changed fields will not depend upon the prior state of the MpegEncContext.
 208  */
 209 static void MPV_encode_defaults(MpegEncContext *s){
 210     int i;
 211     MPV_common_defaults(s);
 212
 213     for(i=-16; i<16; i++){
 214         default_fcode_tab[i + MAX_MV]= 1;
 215     }
 216     s->me.mv_penalty= default_mv_penalty;
 217     s->fcode_tab= default_fcode_tab;
 218 }
 219
 220 /* init video encoder */
 221 av_cold int MPV_encode_init(AVCodecContext *avctx)
 222 {
 223     MpegEncContext *s = avctx->priv_data;
 224     int i;
 225     int chroma_h_shift, chroma_v_shift;
 226
 227     MPV_encode_defaults(s);
 228
 229     switch (avctx->codec_id) {
 230     case CODEC_ID_MPEG2VIDEO:
 231         if(avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P){
 232             av_log(avctx, AV_LOG_ERROR, "only YUV420 and YUV422 are supported\n");
 233             return -1;
 234         }
 235         break;
 236     case CODEC_ID_LJPEG:
 237     case CODEC_ID_MJPEG:
 238         if(avctx->pix_fmt != PIX_FMT_YUVJ420P && avctx->pix_fmt != PIX_FMT_YUVJ422P && avctx->pix_fmt != PIX_FMT_RGB32 &&
 239            ((avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P) || avctx->strict_std_compliance>FF_COMPLIANCE_INOFFICIAL)){
 240             av_log(avctx, AV_LOG_ERROR, "colorspace not supported in jpeg\n");
 241             return -1;
 242         }
 243         break;
 244     default:
 245         if(avctx->pix_fmt != PIX_FMT_YUV420P){
 246             av_log(avctx, AV_LOG_ERROR, "only YUV420 is supported\n");
 247             return -1;
 248         }
 249     }
 250
 251     switch (avctx->pix_fmt) {
 252     case PIX_FMT_YUVJ422P:
 253     case PIX_FMT_YUV422P:
 254         s->chroma_format = CHROMA_422;
 255         break;
 256     case PIX_FMT_YUVJ420P:
 257     case PIX_FMT_YUV420P:
 258     default:
 259         s->chroma_format = CHROMA_420;
 260         break;
 261     }
 262
 263     s->bit_rate = avctx->bit_rate;
 264     s->width = avctx->width;
 265     s->height = avctx->height;
 266     if(avctx->gop_size > 600 && avctx->strict_std_compliance>FF_COMPLIANCE_EXPERIMENTAL){
 267         av_log(avctx, AV_LOG_ERROR, "Warning keyframe interval too large! reducing it ...\n");
 268         avctx->gop_size=600;
 269     }
 270     s->gop_size = avctx->gop_size;
 271     s->avctx = avctx;
 272     s->flags= avctx->flags;
 273     s->flags2= avctx->flags2;
 274     s->max_b_frames= avctx->max_b_frames;
 275     s->codec_id= avctx->codec->id;
 276     s->luma_elim_threshold  = avctx->luma_elim_threshold;
 277     s->chroma_elim_threshold= avctx->chroma_elim_threshold;
 278     s->strict_std_compliance= avctx->strict_std_compliance;
 279     s->data_partitioning= avctx->flags & CODEC_FLAG_PART;
 280     s->quarter_sample= (avctx->flags & CODEC_FLAG_QPEL)!=0;
 281     s->mpeg_quant= avctx->mpeg_quant;
 282     s->rtp_mode= !!avctx->rtp_payload_size;
 283     s->intra_dc_precision= avctx->intra_dc_precision;
 284     s->user_specified_pts = AV_NOPTS_VALUE;
 285
 286     if (s->gop_size <= 1) {
 287         s->intra_only = 1;
 288         s->gop_size = 12;
 289     } else {
 290         s->intra_only = 0;
 291     }
 292
 293     s->me_method = avctx->me_method;
 294
 295     /* Fixed QSCALE */
 296     s->fixed_qscale = !!(avctx->flags & CODEC_FLAG_QSCALE);
 297
 298     s->adaptive_quant= (   s->avctx->lumi_masking
 299                         || s->avctx->dark_masking
 300                         || s->avctx->temporal_cplx_masking
 301                         || s->avctx->spatial_cplx_masking
 302                         || s->avctx->p_masking
 303                         || s->avctx->border_masking
 304                         || (s->flags&CODEC_FLAG_QP_RD))
 305                        && !s->fixed_qscale;
 306
 307     s->obmc= !!(s->flags & CODEC_FLAG_OBMC);
 308     s->loop_filter= !!(s->flags & CODEC_FLAG_LOOP_FILTER);
 309     s->alternate_scan= !!(s->flags & CODEC_FLAG_ALT_SCAN);
 310     s->intra_vlc_format= !!(s->flags2 & CODEC_FLAG2_INTRA_VLC);
 311     s->q_scale_type= !!(s->flags2 & CODEC_FLAG2_NON_LINEAR_QUANT);
 312
 313     if(avctx->rc_max_rate && !avctx->rc_buffer_size){
 314         av_log(avctx, AV_LOG_ERROR, "a vbv buffer size is needed, for encoding with a maximum bitrate\n");
 315         return -1;
 316     }
 317
 318     if(avctx->rc_min_rate && avctx->rc_max_rate != avctx->rc_min_rate){
 319         av_log(avctx, AV_LOG_INFO, "Warning min_rate > 0 but min_rate != max_rate isn't recommended!\n");
 320     }
 321
 322     if(avctx->rc_min_rate && avctx->rc_min_rate > avctx->bit_rate){
 323         av_log(avctx, AV_LOG_ERROR, "bitrate below min bitrate\n");
 324         return -1;
 325     }
 326
 327     if(avctx->rc_max_rate && avctx->rc_max_rate < avctx->bit_rate){
 328         av_log(avctx, AV_LOG_INFO, "bitrate above max bitrate\n");
 329         return -1;
 330     }
 331
 332     if(avctx->rc_max_rate && avctx->rc_max_rate == avctx->bit_rate && avctx->rc_max_rate != avctx->rc_min_rate){
 333         av_log(avctx, AV_LOG_INFO, "impossible bitrate constraints, this will fail\n");
 334     }
 335
 336     if(avctx->rc_buffer_size && avctx->bit_rate*av_q2d(avctx->time_base) > avctx->rc_buffer_size){
 337         av_log(avctx, AV_LOG_ERROR, "VBV buffer too small for bitrate\n");
 338         return -1;
 339     }
 340
 341     if(avctx->bit_rate*av_q2d(avctx->time_base) > avctx->bit_rate_tolerance){
 342         av_log(avctx, AV_LOG_ERROR, "bitrate tolerance too small for bitrate\n");
 343         return -1;
 344     }
 345
 346     if(   s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate
 347        && (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO)
 348        && 90000LL * (avctx->rc_buffer_size-1) > s->avctx->rc_max_rate*0xFFFFLL){
 349
 350         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");
 351     }
 352
 353     if((s->flags & CODEC_FLAG_4MV) && s->codec_id != CODEC_ID_MPEG4
 354        && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P && s->codec_id != CODEC_ID_FLV1){
 355         av_log(avctx, AV_LOG_ERROR, "4MV not supported by codec\n");
 356         return -1;
 357     }
 358
 359     if(s->obmc && s->avctx->mb_decision != FF_MB_DECISION_SIMPLE){
 360         av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with simple mb decision\n");
 361         return -1;
 362     }
 363
 364     if(s->obmc && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P){
 365         av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with H263(+)\n");
 366         return -1;
 367     }
 368
 369     if(s->quarter_sample && s->codec_id != CODEC_ID_MPEG4){
 370         av_log(avctx, AV_LOG_ERROR, "qpel not supported by codec\n");
 371         return -1;
 372     }
 373
 374     if(s->data_partitioning && s->codec_id != CODEC_ID_MPEG4){
 375         av_log(avctx, AV_LOG_ERROR, "data partitioning not supported by codec\n");
 376         return -1;
 377     }
 378
 379     if(s->max_b_frames && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO){
 380         av_log(avctx, AV_LOG_ERROR, "b frames not supported by codec\n");
 381         return -1;
 382     }
 383
 384     if((s->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN))
 385        && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG2VIDEO){
 386         av_log(avctx, AV_LOG_ERROR, "interlacing not supported by codec\n");
 387         return -1;
 388     }
 389
 390     if(s->mpeg_quant && s->codec_id != CODEC_ID_MPEG4){ //FIXME mpeg2 uses that too
 391         av_log(avctx, AV_LOG_ERROR, "mpeg2 style quantization not supported by codec\n");
 392         return -1;
 393     }
 394
 395     if((s->flags & CODEC_FLAG_CBP_RD) && !avctx->trellis){
 396         av_log(avctx, AV_LOG_ERROR, "CBP RD needs trellis quant\n");
 397         return -1;
 398     }
 399
 400     if((s->flags & CODEC_FLAG_QP_RD) && s->avctx->mb_decision != FF_MB_DECISION_RD){
 401         av_log(avctx, AV_LOG_ERROR, "QP RD needs mbd=2\n");
 402         return -1;
 403     }
 404
 405     if(s->avctx->scenechange_threshold < 1000000000 && (s->flags & CODEC_FLAG_CLOSED_GOP)){
 406         av_log(avctx, AV_LOG_ERROR, "closed gop with scene change detection are not supported yet, set threshold to 1000000000\n");
 407         return -1;
 408     }
 409
 410     if((s->flags2 & CODEC_FLAG2_INTRA_VLC) && s->codec_id != CODEC_ID_MPEG2VIDEO){
 411         av_log(avctx, AV_LOG_ERROR, "intra vlc table not supported by codec\n");
 412         return -1;
 413     }
 414
 415     if(s->flags & CODEC_FLAG_LOW_DELAY){
 416         if (s->codec_id != CODEC_ID_MPEG2VIDEO){
 417             av_log(avctx, AV_LOG_ERROR, "low delay forcing is only available for mpeg2\n");
 418             return -1;
 419         }
 420         if (s->max_b_frames != 0){
 421             av_log(avctx, AV_LOG_ERROR, "b frames cannot be used with low delay\n");
 422             return -1;
 423         }
 424     }
 425
 426     if(s->q_scale_type == 1){
 427         if(s->codec_id != CODEC_ID_MPEG2VIDEO){
 428             av_log(avctx, AV_LOG_ERROR, "non linear quant is only available for mpeg2\n");
 429             return -1;
 430         }
 431         if(avctx->qmax > 12){
 432             av_log(avctx, AV_LOG_ERROR, "non linear quant only supports qmax <= 12 currently\n");
 433             return -1;
 434         }
 435     }
 436
 437     if(s->avctx->thread_count > 1 && s->codec_id != CODEC_ID_MPEG4
 438        && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO
 439        && (s->codec_id != CODEC_ID_H263P || !(s->flags & CODEC_FLAG_H263P_SLICE_STRUCT))){
 440         av_log(avctx, AV_LOG_ERROR, "multi threaded encoding not supported by codec\n");
 441         return -1;
 442     }
 443
 444     if(s->avctx->thread_count > 1)
 445         s->rtp_mode= 1;
 446
 447     if(!avctx->time_base.den || !avctx->time_base.num){
 448         av_log(avctx, AV_LOG_ERROR, "framerate not set\n");
 449         return -1;
 450     }
 451
 452     i= (INT_MAX/2+128)>>8;
 453     if(avctx->me_threshold >= i){
 454         av_log(avctx, AV_LOG_ERROR, "me_threshold too large, max is %d\n", i - 1);
 455         return -1;
 456     }
 457     if(avctx->mb_threshold >= i){
 458         av_log(avctx, AV_LOG_ERROR, "mb_threshold too large, max is %d\n", i - 1);
 459         return -1;
 460     }
 461
 462     if(avctx->b_frame_strategy && (avctx->flags&CODEC_FLAG_PASS2)){
 463         av_log(avctx, AV_LOG_INFO, "notice: b_frame_strategy only affects the first pass\n");
 464         avctx->b_frame_strategy = 0;
 465     }
 466
 467     i= av_gcd(avctx->time_base.den, avctx->time_base.num);
 468     if(i > 1){
 469         av_log(avctx, AV_LOG_INFO, "removing common factors from framerate\n");
 470         avctx->time_base.den /= i;
 471         avctx->time_base.num /= i;
 472 //        return -1;
 473     }
 474
 475     if(s->codec_id==CODEC_ID_MJPEG){
 476         s->intra_quant_bias= 1<<(QUANT_BIAS_SHIFT-1); //(a + x/2)/x
 477         s->inter_quant_bias= 0;
 478     }else if(s->mpeg_quant || s->codec_id==CODEC_ID_MPEG1VIDEO || s->codec_id==CODEC_ID_MPEG2VIDEO){
 479         s->intra_quant_bias= 3<<(QUANT_BIAS_SHIFT-3); //(a + x*3/8)/x
 480         s->inter_quant_bias= 0;
 481     }else{
 482         s->intra_quant_bias=0;
 483         s->inter_quant_bias=-(1<<(QUANT_BIAS_SHIFT-2)); //(a - x/4)/x
 484     }
 485
 486     if(avctx->intra_quant_bias != FF_DEFAULT_QUANT_BIAS)
 487         s->intra_quant_bias= avctx->intra_quant_bias;
 488     if(avctx->inter_quant_bias != FF_DEFAULT_QUANT_BIAS)
 489         s->inter_quant_bias= avctx->inter_quant_bias;
 490
 491     avcodec_get_chroma_sub_sample(avctx->pix_fmt, &chroma_h_shift, &chroma_v_shift);
 492
 493     if(avctx->codec_id == CODEC_ID_MPEG4 && s->avctx->time_base.den > (1<<16)-1){
 494         av_log(avctx, AV_LOG_ERROR, "timebase not supported by mpeg 4 standard\n");
 495         return -1;
 496     }
 497     s->time_increment_bits = av_log2(s->avctx->time_base.den - 1) + 1;
 498
 499     switch(avctx->codec->id) {
 500     case CODEC_ID_MPEG1VIDEO:
 501         s->out_format = FMT_MPEG1;
 502         s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY);
 503         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
 504         break;
 505     case CODEC_ID_MPEG2VIDEO:
 506         s->out_format = FMT_MPEG1;
 507         s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY);
 508         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
 509         s->rtp_mode= 1;
 510         break;
 511     case CODEC_ID_LJPEG:
 512     case CODEC_ID_MJPEG:
 513         s->out_format = FMT_MJPEG;
 514         s->intra_only = 1; /* force intra only for jpeg */
 515         s->mjpeg_vsample[0] = 2;
 516         s->mjpeg_vsample[1] = 2>>chroma_v_shift;
 517         s->mjpeg_vsample[2] = 2>>chroma_v_shift;
 518         s->mjpeg_hsample[0] = 2;
 519         s->mjpeg_hsample[1] = 2>>chroma_h_shift;
 520         s->mjpeg_hsample[2] = 2>>chroma_h_shift;
 521         if (!(CONFIG_MJPEG_ENCODER || CONFIG_LJPEG_ENCODER)
 522             || ff_mjpeg_encode_init(s) < 0)
 523             return -1;
 524         avctx->delay=0;
 525         s->low_delay=1;
 526         break;
 527     case CODEC_ID_H261:
 528         if (!CONFIG_H261_ENCODER)  return -1;
 529         if (ff_h261_get_picture_format(s->width, s->height) < 0) {
 530             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);
 531             return -1;
 532         }
 533         s->out_format = FMT_H261;
 534         avctx->delay=0;
 535         s->low_delay=1;
 536         break;
 537     case CODEC_ID_H263:
 538         if (!CONFIG_H263_ENCODER)  return -1;
 539         if (h263_get_picture_format(s->width, s->height) == 7) {
 540             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);
 541             return -1;
 542         }
 543         s->out_format = FMT_H263;
 544         s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
 545         avctx->delay=0;
 546         s->low_delay=1;
 547         break;
 548     case CODEC_ID_H263P:
 549         s->out_format = FMT_H263;
 550         s->h263_plus = 1;
 551         /* Fx */
 552         s->umvplus = (avctx->flags & CODEC_FLAG_H263P_UMV) ? 1:0;
 553         s->h263_aic= (avctx->flags & CODEC_FLAG_AC_PRED) ? 1:0;
 554         s->modified_quant= s->h263_aic;
 555         s->alt_inter_vlc= (avctx->flags & CODEC_FLAG_H263P_AIV) ? 1:0;
 556         s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
 557         s->loop_filter= (avctx->flags & CODEC_FLAG_LOOP_FILTER) ? 1:0;
 558         s->unrestricted_mv= s->obmc || s->loop_filter || s->umvplus;
 559         s->h263_slice_structured= (s->flags & CODEC_FLAG_H263P_SLICE_STRUCT) ? 1:0;
 560
 561         /* /Fx */
 562         /* These are just to be sure */
 563         avctx->delay=0;
 564         s->low_delay=1;
 565         break;
 566     case CODEC_ID_FLV1:
 567         s->out_format = FMT_H263;
 568         s->h263_flv = 2; /* format = 1; 11-bit codes */
 569         s->unrestricted_mv = 1;
 570         s->rtp_mode=0; /* don't allow GOB */
 571         avctx->delay=0;
 572         s->low_delay=1;
 573         break;
 574     case CODEC_ID_RV10:
 575         s->out_format = FMT_H263;
 576         avctx->delay=0;
 577         s->low_delay=1;
 578         break;
 579     case CODEC_ID_RV20:
 580         s->out_format = FMT_H263;
 581         avctx->delay=0;
 582         s->low_delay=1;
 583         s->modified_quant=1;
 584         s->h263_aic=1;
 585         s->h263_plus=1;
 586         s->loop_filter=1;
 587         s->unrestricted_mv= s->obmc || s->loop_filter || s->umvplus;
 588         break;
 589     case CODEC_ID_MPEG4:
 590         s->out_format = FMT_H263;
 591         s->h263_pred = 1;
 592         s->unrestricted_mv = 1;
 593         s->low_delay= s->max_b_frames ? 0 : 1;
 594         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
 595         break;
 596     case CODEC_ID_MSMPEG4V1:
 597         s->out_format = FMT_H263;
 598         s->h263_msmpeg4 = 1;
 599         s->h263_pred = 1;
 600         s->unrestricted_mv = 1;
 601         s->msmpeg4_version= 1;
 602         avctx->delay=0;
 603         s->low_delay=1;
 604         break;
 605     case CODEC_ID_MSMPEG4V2:
 606         s->out_format = FMT_H263;
 607         s->h263_msmpeg4 = 1;
 608         s->h263_pred = 1;
 609         s->unrestricted_mv = 1;
 610         s->msmpeg4_version= 2;
 611         avctx->delay=0;
 612         s->low_delay=1;
 613         break;
 614     case CODEC_ID_MSMPEG4V3:
 615         s->out_format = FMT_H263;
 616         s->h263_msmpeg4 = 1;
 617         s->h263_pred = 1;
 618         s->unrestricted_mv = 1;
 619         s->msmpeg4_version= 3;
 620         s->flipflop_rounding=1;
 621         avctx->delay=0;
 622         s->low_delay=1;
 623         break;
 624     case CODEC_ID_WMV1:
 625         s->out_format = FMT_H263;
 626         s->h263_msmpeg4 = 1;
 627         s->h263_pred = 1;
 628         s->unrestricted_mv = 1;
 629         s->msmpeg4_version= 4;
 630         s->flipflop_rounding=1;
 631         avctx->delay=0;
 632         s->low_delay=1;
 633         break;
 634     case CODEC_ID_WMV2:
 635         s->out_format = FMT_H263;
 636         s->h263_msmpeg4 = 1;
 637         s->h263_pred = 1;
 638         s->unrestricted_mv = 1;
 639         s->msmpeg4_version= 5;
 640         s->flipflop_rounding=1;
 641         avctx->delay=0;
 642         s->low_delay=1;
 643         break;
 644     default:
 645         return -1;
 646     }
 647
 648     avctx->has_b_frames= !s->low_delay;
 649
 650     s->encoding = 1;
 651
 652     /* init */
 653     if (MPV_common_init(s) < 0)
 654         return -1;
 655
 656     if(!s->dct_quantize)
 657         s->dct_quantize = dct_quantize_c;
 658     if(!s->denoise_dct)
 659         s->denoise_dct = denoise_dct_c;
 660     s->fast_dct_quantize = s->dct_quantize;
 661     if(avctx->trellis)
 662         s->dct_quantize = dct_quantize_trellis_c;
 663
 664     if((CONFIG_H263P_ENCODER || CONFIG_RV20_ENCODER) && s->modified_quant)
 665         s->chroma_qscale_table= ff_h263_chroma_qscale_table;
 666     s->progressive_frame=
 667     s->progressive_sequence= !(avctx->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN));
 668     s->quant_precision=5;
 669
 670     ff_set_cmp(&s->dsp, s->dsp.ildct_cmp, s->avctx->ildct_cmp);
 671     ff_set_cmp(&s->dsp, s->dsp.frame_skip_cmp, s->avctx->frame_skip_cmp);
 672
 673     if (CONFIG_H261_ENCODER && s->out_format == FMT_H261)
 674         ff_h261_encode_init(s);
 675     if (CONFIG_ANY_H263_ENCODER && s->out_format == FMT_H263)
 676         h263_encode_init(s);
 677     if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version)
 678         ff_msmpeg4_encode_init(s);
 679     if ((CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
 680         && s->out_format == FMT_MPEG1)
 681         ff_mpeg1_encode_init(s);
 682
 683     /* init q matrix */
 684     for(i=0;i<64;i++) {
 685         int j= s->dsp.idct_permutation[i];
 686         if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4 && s->mpeg_quant){
 687             s->intra_matrix[j] = ff_mpeg4_default_intra_matrix[i];
 688             s->inter_matrix[j] = ff_mpeg4_default_non_intra_matrix[i];
 689         }else if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
 690             s->intra_matrix[j] =
 691             s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
 692         }else
 693         { /* mpeg1/2 */
 694             s->intra_matrix[j] = ff_mpeg1_default_intra_matrix[i];
 695             s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
 696         }
 697         if(s->avctx->intra_matrix)
 698             s->intra_matrix[j] = s->avctx->intra_matrix[i];
 699         if(s->avctx->inter_matrix)
 700             s->inter_matrix[j] = s->avctx->inter_matrix[i];
 701     }
 702
 703     /* precompute matrix */
 704     /* for mjpeg, we do include qscale in the matrix */
 705     if (s->out_format != FMT_MJPEG) {
 706         ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
 707                        s->intra_matrix, s->intra_quant_bias, avctx->qmin, 31, 1);
 708         ff_convert_matrix(&s->dsp, s->q_inter_matrix, s->q_inter_matrix16,
 709                        s->inter_matrix, s->inter_quant_bias, avctx->qmin, 31, 0);
 710     }
 711
 712     if(ff_rate_control_init(s) < 0)
 713         return -1;
 714
 715     return 0;
 716 }
 717
 718 av_cold int MPV_encode_end(AVCodecContext *avctx)
 719 {
 720     MpegEncContext *s = avctx->priv_data;
 721
 722     ff_rate_control_uninit(s);
 723
 724     MPV_common_end(s);
 725     if ((CONFIG_MJPEG_ENCODER || CONFIG_LJPEG_ENCODER) && s->out_format == FMT_MJPEG)
 726         ff_mjpeg_encode_close(s);
 727
 728     av_freep(&avctx->extradata);
 729
 730     return 0;
 731 }
 732
 733 static int get_sae(uint8_t *src, int ref, int stride){
 734     int x,y;
 735     int acc=0;
 736
 737     for(y=0; y<16; y++){
 738         for(x=0; x<16; x++){
 739             acc+= FFABS(src[x+y*stride] - ref);
 740         }
 741     }
 742
 743     return acc;
 744 }
 745
 746 static int get_intra_count(MpegEncContext *s, uint8_t *src, uint8_t *ref, int stride){
 747     int x, y, w, h;
 748     int acc=0;
 749
 750     w= s->width &~15;
 751     h= s->height&~15;
 752
 753     for(y=0; y<h; y+=16){
 754         for(x=0; x<w; x+=16){
 755             int offset= x + y*stride;
 756             int sad = s->dsp.sad[0](NULL, src + offset, ref + offset, stride, 16);
 757             int mean= (s->dsp.pix_sum(src + offset, stride) + 128)>>8;
 758             int sae = get_sae(src + offset, mean, stride);
 759
 760             acc+= sae + 500 < sad;
 761         }
 762     }
 763     return acc;
 764 }
 765
 766
 767 static int load_input_picture(MpegEncContext *s, AVFrame *pic_arg){
 768     AVFrame *pic=NULL;
 769     int64_t pts;
 770     int i;
 771     const int encoding_delay= s->max_b_frames;
 772     int direct=1;
 773
 774     if(pic_arg){
 775         pts= pic_arg->pts;
 776         pic_arg->display_picture_number= s->input_picture_number++;
 777
 778         if(pts != AV_NOPTS_VALUE){
 779             if(s->user_specified_pts != AV_NOPTS_VALUE){
 780                 int64_t time= pts;
 781                 int64_t last= s->user_specified_pts;
 782
 783                 if(time <= last){
 784                     av_log(s->avctx, AV_LOG_ERROR, "Error, Invalid timestamp=%"PRId64", last=%"PRId64"\n", pts, s->user_specified_pts);
 785                     return -1;
 786                 }
 787             }
 788             s->user_specified_pts= pts;
 789         }else{
 790             if(s->user_specified_pts != AV_NOPTS_VALUE){
 791                 s->user_specified_pts=
 792                 pts= s->user_specified_pts + 1;
 793                 av_log(s->avctx, AV_LOG_INFO, "Warning: AVFrame.pts=? trying to guess (%"PRId64")\n", pts);
 794             }else{
 795                 pts= pic_arg->display_picture_number;
 796             }
 797         }
 798     }
 799
 800   if(pic_arg){
 801     if(encoding_delay && !(s->flags&CODEC_FLAG_INPUT_PRESERVED)) direct=0;
 802     if(pic_arg->linesize[0] != s->linesize) direct=0;
 803     if(pic_arg->linesize[1] != s->uvlinesize) direct=0;
 804     if(pic_arg->linesize[2] != s->uvlinesize) direct=0;
 805
 806 //    av_log(AV_LOG_DEBUG, "%d %d %d %d\n",pic_arg->linesize[0], pic_arg->linesize[1], s->linesize, s->uvlinesize);
 807
 808     if(direct){
 809         i= ff_find_unused_picture(s, 1);
 810
 811         pic= (AVFrame*)&s->picture[i];
 812         pic->reference= 3;
 813
 814         for(i=0; i<4; i++){
 815             pic->data[i]= pic_arg->data[i];
 816             pic->linesize[i]= pic_arg->linesize[i];
 817         }
 818         alloc_picture(s, (Picture*)pic, 1);
 819     }else{
 820         i= ff_find_unused_picture(s, 0);
 821
 822         pic= (AVFrame*)&s->picture[i];
 823         pic->reference= 3;
 824
 825         alloc_picture(s, (Picture*)pic, 0);
 826
 827         if(   pic->data[0] + INPLACE_OFFSET == pic_arg->data[0]
 828            && pic->data[1] + INPLACE_OFFSET == pic_arg->data[1]
 829            && pic->data[2] + INPLACE_OFFSET == pic_arg->data[2]){
 830        // empty
 831         }else{
 832             int h_chroma_shift, v_chroma_shift;
 833             avcodec_get_chroma_sub_sample(s->avctx->pix_fmt, &h_chroma_shift, &v_chroma_shift);
 834
 835             for(i=0; i<3; i++){
 836                 int src_stride= pic_arg->linesize[i];
 837                 int dst_stride= i ? s->uvlinesize : s->linesize;
 838                 int h_shift= i ? h_chroma_shift : 0;
 839                 int v_shift= i ? v_chroma_shift : 0;
 840                 int w= s->width >>h_shift;
 841                 int h= s->height>>v_shift;
 842                 uint8_t *src= pic_arg->data[i];
 843                 uint8_t *dst= pic->data[i];
 844
 845                 if(!s->avctx->rc_buffer_size)
 846                     dst +=INPLACE_OFFSET;
 847
 848                 if(src_stride==dst_stride)
 849                     memcpy(dst, src, src_stride*h);
 850                 else{
 851                     while(h--){
 852                         memcpy(dst, src, w);
 853                         dst += dst_stride;
 854                         src += src_stride;
 855                     }
 856                 }
 857             }
 858         }
 859     }
 860     copy_picture_attributes(s, pic, pic_arg);
 861     pic->pts= pts; //we set this here to avoid modifiying pic_arg
 862   }
 863
 864     /* shift buffer entries */
 865     for(i=1; i<MAX_PICTURE_COUNT /*s->encoding_delay+1*/; i++)
 866         s->input_picture[i-1]= s->input_picture[i];
 867
 868     s->input_picture[encoding_delay]= (Picture*)pic;
 869
 870     return 0;
 871 }
 872
 873 static int skip_check(MpegEncContext *s, Picture *p, Picture *ref){
 874     int x, y, plane;
 875     int score=0;
 876     int64_t score64=0;
 877
 878     for(plane=0; plane<3; plane++){
 879         const int stride= p->linesize[plane];
 880         const int bw= plane ? 1 : 2;
 881         for(y=0; y<s->mb_height*bw; y++){
 882             for(x=0; x<s->mb_width*bw; x++){
 883                 int off= p->type == FF_BUFFER_TYPE_SHARED ? 0: 16;
 884                 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);
 885
 886                 switch(s->avctx->frame_skip_exp){
 887                     case 0: score= FFMAX(score, v); break;
 888                     case 1: score+= FFABS(v);break;
 889                     case 2: score+= v*v;break;
 890                     case 3: score64+= FFABS(v*v*(int64_t)v);break;
 891                     case 4: score64+= v*v*(int64_t)(v*v);break;
 892                 }
 893             }
 894         }
 895     }
 896
 897     if(score) score64= score;
 898
 899     if(score64 < s->avctx->frame_skip_threshold)
 900         return 1;
 901     if(score64 < ((s->avctx->frame_skip_factor * (int64_t)s->lambda)>>8))
 902         return 1;
 903     return 0;
 904 }
 905
 906 static int estimate_best_b_count(MpegEncContext *s){
 907     AVCodec *codec= avcodec_find_encoder(s->avctx->codec_id);
 908     AVCodecContext *c= avcodec_alloc_context();
 909     AVFrame input[FF_MAX_B_FRAMES+2];
 910     const int scale= s->avctx->brd_scale;
 911     int i, j, out_size, p_lambda, b_lambda, lambda2;
 912     int outbuf_size= s->width * s->height; //FIXME
 913     uint8_t *outbuf= av_malloc(outbuf_size);
 914     int64_t best_rd= INT64_MAX;
 915     int best_b_count= -1;
 916
 917     assert(scale>=0 && scale <=3);
 918
 919 //    emms_c();
 920     p_lambda= s->last_lambda_for[FF_P_TYPE]; //s->next_picture_ptr->quality;
 921     b_lambda= s->last_lambda_for[FF_B_TYPE]; //p_lambda *FFABS(s->avctx->b_quant_factor) + s->avctx->b_quant_offset;
 922     if(!b_lambda) b_lambda= p_lambda; //FIXME we should do this somewhere else
 923     lambda2= (b_lambda*b_lambda + (1<<FF_LAMBDA_SHIFT)/2 ) >> FF_LAMBDA_SHIFT;
 924
 925     c->width = s->width >> scale;
 926     c->height= s->height>> scale;
 927     c->flags= CODEC_FLAG_QSCALE | CODEC_FLAG_PSNR | CODEC_FLAG_INPUT_PRESERVED /*| CODEC_FLAG_EMU_EDGE*/;
 928     c->flags|= s->avctx->flags & CODEC_FLAG_QPEL;
 929     c->mb_decision= s->avctx->mb_decision;
 930     c->me_cmp= s->avctx->me_cmp;
 931     c->mb_cmp= s->avctx->mb_cmp;
 932     c->me_sub_cmp= s->avctx->me_sub_cmp;
 933     c->pix_fmt = PIX_FMT_YUV420P;
 934     c->time_base= s->avctx->time_base;
 935     c->max_b_frames= s->max_b_frames;
 936
 937     if (avcodec_open(c, codec) < 0)
 938         return -1;
 939
 940     for(i=0; i<s->max_b_frames+2; i++){
 941         int ysize= c->width*c->height;
 942         int csize= (c->width/2)*(c->height/2);
 943         Picture pre_input, *pre_input_ptr= i ? s->input_picture[i-1] : s->next_picture_ptr;
 944
 945         avcodec_get_frame_defaults(&input[i]);
 946         input[i].data[0]= av_malloc(ysize + 2*csize);
 947         input[i].data[1]= input[i].data[0] + ysize;
 948         input[i].data[2]= input[i].data[1] + csize;
 949         input[i].linesize[0]= c->width;
 950         input[i].linesize[1]=
 951         input[i].linesize[2]= c->width/2;
 952
 953         if(pre_input_ptr && (!i || s->input_picture[i-1])) {
 954             pre_input= *pre_input_ptr;
 955
 956             if(pre_input.type != FF_BUFFER_TYPE_SHARED && i) {
 957                 pre_input.data[0]+=INPLACE_OFFSET;
 958                 pre_input.data[1]+=INPLACE_OFFSET;
 959                 pre_input.data[2]+=INPLACE_OFFSET;
 960             }
 961
 962             s->dsp.shrink[scale](input[i].data[0], input[i].linesize[0], pre_input.data[0], pre_input.linesize[0], c->width, c->height);
 963             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);
 964             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);
 965         }
 966     }
 967
 968     for(j=0; j<s->max_b_frames+1; j++){
 969         int64_t rd=0;
 970
 971         if(!s->input_picture[j])
 972             break;
 973
 974         c->error[0]= c->error[1]= c->error[2]= 0;
 975
 976         input[0].pict_type= FF_I_TYPE;
 977         input[0].quality= 1 * FF_QP2LAMBDA;
 978         out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[0]);
 979 //        rd += (out_size * lambda2) >> FF_LAMBDA_SHIFT;
 980
 981         for(i=0; i<s->max_b_frames+1; i++){
 982             int is_p= i % (j+1) == j || i==s->max_b_frames;
 983
 984             input[i+1].pict_type= is_p ? FF_P_TYPE : FF_B_TYPE;
 985             input[i+1].quality= is_p ? p_lambda : b_lambda;
 986             out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[i+1]);
 987             rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
 988         }
 989
 990         /* get the delayed frames */
 991         while(out_size){
 992             out_size = avcodec_encode_video(c, outbuf, outbuf_size, NULL);
 993             rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
 994         }
 995
 996         rd += c->error[0] + c->error[1] + c->error[2];
 997
 998         if(rd < best_rd){
 999             best_rd= rd;
1000             best_b_count= j;
1001         }
1002     }
1003
1004     av_freep(&outbuf);
1005     avcodec_close(c);
1006     av_freep(&c);
1007
1008     for(i=0; i<s->max_b_frames+2; i++){
1009         av_freep(&input[i].data[0]);
1010     }
1011
1012     return best_b_count;
1013 }
1014
1015 static void select_input_picture(MpegEncContext *s){
1016     int i;
1017
1018     for(i=1; i<MAX_PICTURE_COUNT; i++)
1019         s->reordered_input_picture[i-1]= s->reordered_input_picture[i];
1020     s->reordered_input_picture[MAX_PICTURE_COUNT-1]= NULL;
1021
1022     /* set next picture type & ordering */
1023     if(s->reordered_input_picture[0]==NULL && s->input_picture[0]){
1024         if(/*s->picture_in_gop_number >= s->gop_size ||*/ s->next_picture_ptr==NULL || s->intra_only){
1025             s->reordered_input_picture[0]= s->input_picture[0];
1026             s->reordered_input_picture[0]->pict_type= FF_I_TYPE;
1027             s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++;
1028         }else{
1029             int b_frames;
1030
1031             if(s->avctx->frame_skip_threshold || s->avctx->frame_skip_factor){
1032                 if(s->picture_in_gop_number < s->gop_size && skip_check(s, s->input_picture[0], s->next_picture_ptr)){
1033                 //FIXME check that te gop check above is +-1 correct
1034 //av_log(NULL, AV_LOG_DEBUG, "skip %p %"PRId64"\n", s->input_picture[0]->data[0], s->input_picture[0]->pts);
1035
1036                     if(s->input_picture[0]->type == FF_BUFFER_TYPE_SHARED){
1037                         for(i=0; i<4; i++)
1038                             s->input_picture[0]->data[i]= NULL;
1039                         s->input_picture[0]->type= 0;
1040                     }else{
1041                         assert(   s->input_picture[0]->type==FF_BUFFER_TYPE_USER
1042                                || s->input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
1043
1044                         s->avctx->release_buffer(s->avctx, (AVFrame*)s->input_picture[0]);
1045                     }
1046
1047                     emms_c();
1048                     ff_vbv_update(s, 0);
1049
1050                     goto no_output_pic;
1051                 }
1052             }
1053
1054             if(s->flags&CODEC_FLAG_PASS2){
1055                 for(i=0; i<s->max_b_frames+1; i++){
1056                     int pict_num= s->input_picture[0]->display_picture_number + i;
1057
1058                     if(pict_num >= s->rc_context.num_entries)
1059                         break;
1060                     if(!s->input_picture[i]){
1061                         s->rc_context.entry[pict_num-1].new_pict_type = FF_P_TYPE;
1062                         break;
1063                     }
1064
1065                     s->input_picture[i]->pict_type=
1066                         s->rc_context.entry[pict_num].new_pict_type;
1067                 }
1068             }
1069
1070             if(s->avctx->b_frame_strategy==0){
1071                 b_frames= s->max_b_frames;
1072                 while(b_frames && !s->input_picture[b_frames]) b_frames--;
1073             }else if(s->avctx->b_frame_strategy==1){
1074                 for(i=1; i<s->max_b_frames+1; i++){
1075                     if(s->input_picture[i] && s->input_picture[i]->b_frame_score==0){
1076                         s->input_picture[i]->b_frame_score=
1077                             get_intra_count(s, s->input_picture[i  ]->data[0],
1078                                                s->input_picture[i-1]->data[0], s->linesize) + 1;
1079                     }
1080                 }
1081                 for(i=0; i<s->max_b_frames+1; i++){
1082                     if(s->input_picture[i]==NULL || s->input_picture[i]->b_frame_score - 1 > s->mb_num/s->avctx->b_sensitivity) break;
1083                 }
1084
1085                 b_frames= FFMAX(0, i-1);
1086
1087                 /* reset scores */
1088                 for(i=0; i<b_frames+1; i++){
1089                     s->input_picture[i]->b_frame_score=0;
1090                 }
1091             }else if(s->avctx->b_frame_strategy==2){
1092                 b_frames= estimate_best_b_count(s);
1093             }else{
1094                 av_log(s->avctx, AV_LOG_ERROR, "illegal b frame strategy\n");
1095                 b_frames=0;
1096             }
1097
1098             emms_c();
1099 //static int b_count=0;
1100 //b_count+= b_frames;
1101 //av_log(s->avctx, AV_LOG_DEBUG, "b_frames: %d\n", b_count);
1102
1103             for(i= b_frames - 1; i>=0; i--){
1104                 int type= s->input_picture[i]->pict_type;
1105                 if(type && type != FF_B_TYPE)
1106                     b_frames= i;
1107             }
1108             if(s->input_picture[b_frames]->pict_type == FF_B_TYPE && b_frames == s->max_b_frames){
1109                 av_log(s->avctx, AV_LOG_ERROR, "warning, too many b frames in a row\n");
1110             }
1111
1112             if(s->picture_in_gop_number + b_frames >= s->gop_size){
1113               if((s->flags2 & CODEC_FLAG2_STRICT_GOP) && s->gop_size > s->picture_in_gop_number){
1114                     b_frames= s->gop_size - s->picture_in_gop_number - 1;
1115               }else{
1116                 if(s->flags & CODEC_FLAG_CLOSED_GOP)
1117                     b_frames=0;
1118                 s->input_picture[b_frames]->pict_type= FF_I_TYPE;
1119               }
1120             }
1121
1122             if(   (s->flags & CODEC_FLAG_CLOSED_GOP)
1123                && b_frames
1124                && s->input_picture[b_frames]->pict_type== FF_I_TYPE)
1125                 b_frames--;
1126
1127             s->reordered_input_picture[0]= s->input_picture[b_frames];
1128             if(s->reordered_input_picture[0]->pict_type != FF_I_TYPE)
1129                 s->reordered_input_picture[0]->pict_type= FF_P_TYPE;
1130             s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++;
1131             for(i=0; i<b_frames; i++){
1132                 s->reordered_input_picture[i+1]= s->input_picture[i];
1133                 s->reordered_input_picture[i+1]->pict_type= FF_B_TYPE;
1134                 s->reordered_input_picture[i+1]->coded_picture_number= s->coded_picture_number++;
1135             }
1136         }
1137     }
1138 no_output_pic:
1139     if(s->reordered_input_picture[0]){
1140         s->reordered_input_picture[0]->reference= s->reordered_input_picture[0]->pict_type!=FF_B_TYPE ? 3 : 0;
1141
1142         ff_copy_picture(&s->new_picture, s->reordered_input_picture[0]);
1143
1144         if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_SHARED || s->avctx->rc_buffer_size){
1145             // input is a shared pix, so we can't modifiy it -> alloc a new one & ensure that the shared one is reuseable
1146
1147             int i= ff_find_unused_picture(s, 0);
1148             Picture *pic= &s->picture[i];
1149
1150             pic->reference              = s->reordered_input_picture[0]->reference;
1151             alloc_picture(s, pic, 0);
1152
1153             /* mark us unused / free shared pic */
1154             if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_INTERNAL)
1155                 s->avctx->release_buffer(s->avctx, (AVFrame*)s->reordered_input_picture[0]);
1156             for(i=0; i<4; i++)
1157                 s->reordered_input_picture[0]->data[i]= NULL;
1158             s->reordered_input_picture[0]->type= 0;
1159
1160             copy_picture_attributes(s, (AVFrame*)pic, (AVFrame*)s->reordered_input_picture[0]);
1161
1162             s->current_picture_ptr= pic;
1163         }else{
1164             // input is not a shared pix -> reuse buffer for current_pix
1165
1166             assert(   s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_USER
1167                    || s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
1168
1169             s->current_picture_ptr= s->reordered_input_picture[0];
1170             for(i=0; i<4; i++){
1171                 s->new_picture.data[i]+= INPLACE_OFFSET;
1172             }
1173         }
1174         ff_copy_picture(&s->current_picture, s->current_picture_ptr);
1175
1176         s->picture_number= s->new_picture.display_picture_number;
1177 //printf("dpn:%d\n", s->picture_number);
1178     }else{
1179        memset(&s->new_picture, 0, sizeof(Picture));
1180     }
1181 }
1182
1183 int MPV_encode_picture(AVCodecContext *avctx,
1184                        unsigned char *buf, int buf_size, void *data)
1185 {
1186     MpegEncContext *s = avctx->priv_data;
1187     AVFrame *pic_arg = data;
1188     int i, stuffing_count;
1189
1190     for(i=0; i<avctx->thread_count; i++){
1191         int start_y= s->thread_context[i]->start_mb_y;
1192         int   end_y= s->thread_context[i]->  end_mb_y;
1193         int h= s->mb_height;
1194         uint8_t *start= buf + (size_t)(((int64_t) buf_size)*start_y/h);
1195         uint8_t *end  = buf + (size_t)(((int64_t) buf_size)*  end_y/h);
1196
1197         init_put_bits(&s->thread_context[i]->pb, start, end - start);
1198     }
1199
1200     s->picture_in_gop_number++;
1201
1202     if(load_input_picture(s, pic_arg) < 0)
1203         return -1;
1204
1205     select_input_picture(s);
1206
1207     /* output? */
1208     if(s->new_picture.data[0]){
1209         s->pict_type= s->new_picture.pict_type;
1210 //emms_c();
1211 //printf("qs:%f %f %d\n", s->new_picture.quality, s->current_picture.quality, s->qscale);
1212         MPV_frame_start(s, avctx);
1213 vbv_retry:
1214         if (encode_picture(s, s->picture_number) < 0)
1215             return -1;
1216
1217         avctx->real_pict_num  = s->picture_number;
1218         avctx->header_bits = s->header_bits;
1219         avctx->mv_bits     = s->mv_bits;
1220         avctx->misc_bits   = s->misc_bits;
1221         avctx->i_tex_bits  = s->i_tex_bits;
1222         avctx->p_tex_bits  = s->p_tex_bits;
1223         avctx->i_count     = s->i_count;
1224         avctx->p_count     = s->mb_num - s->i_count - s->skip_count; //FIXME f/b_count in avctx
1225         avctx->skip_count  = s->skip_count;
1226
1227         MPV_frame_end(s);
1228
1229         if (CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG)
1230             ff_mjpeg_encode_picture_trailer(s);
1231
1232         if(avctx->rc_buffer_size){
1233             RateControlContext *rcc= &s->rc_context;
1234             int max_size= rcc->buffer_index * avctx->rc_max_available_vbv_use;
1235
1236             if(put_bits_count(&s->pb) > max_size && s->lambda < s->avctx->lmax){
1237                 s->next_lambda= FFMAX(s->lambda+1, s->lambda*(s->qscale+1) / s->qscale);
1238                 if(s->adaptive_quant){
1239                     int i;
1240                     for(i=0; i<s->mb_height*s->mb_stride; i++)
1241                         s->lambda_table[i]= FFMAX(s->lambda_table[i]+1, s->lambda_table[i]*(s->qscale+1) / s->qscale);
1242                 }
1243                 s->mb_skipped = 0;        //done in MPV_frame_start()
1244                 if(s->pict_type==FF_P_TYPE){ //done in encode_picture() so we must undo it
1245                     if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
1246                         s->no_rounding ^= 1;
1247                 }
1248                 if(s->pict_type!=FF_B_TYPE){
1249                     s->time_base= s->last_time_base;
1250                     s->last_non_b_time= s->time - s->pp_time;
1251                 }
1252 //                av_log(NULL, AV_LOG_ERROR, "R:%d ", s->next_lambda);
1253                 for(i=0; i<avctx->thread_count; i++){
1254                     PutBitContext *pb= &s->thread_context[i]->pb;
1255                     init_put_bits(pb, pb->buf, pb->buf_end - pb->buf);
1256                 }
1257                 goto vbv_retry;
1258             }
1259
1260             assert(s->avctx->rc_max_rate);
1261         }
1262
1263         if(s->flags&CODEC_FLAG_PASS1)
1264             ff_write_pass1_stats(s);
1265
1266         for(i=0; i<4; i++){
1267             s->current_picture_ptr->error[i]= s->current_picture.error[i];
1268             avctx->error[i] += s->current_picture_ptr->error[i];
1269         }
1270
1271         if(s->flags&CODEC_FLAG_PASS1)
1272             assert(avctx->header_bits + avctx->mv_bits + avctx->misc_bits + avctx->i_tex_bits + avctx->p_tex_bits == put_bits_count(&s->pb));
1273         flush_put_bits(&s->pb);
1274         s->frame_bits  = put_bits_count(&s->pb);
1275
1276         stuffing_count= ff_vbv_update(s, s->frame_bits);
1277         if(stuffing_count){
1278             if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < stuffing_count + 50){
1279                 av_log(s->avctx, AV_LOG_ERROR, "stuffing too large\n");
1280                 return -1;
1281             }
1282
1283             switch(s->codec_id){
1284             case CODEC_ID_MPEG1VIDEO:
1285             case CODEC_ID_MPEG2VIDEO:
1286                 while(stuffing_count--){
1287                     put_bits(&s->pb, 8, 0);
1288                 }
1289             break;
1290             case CODEC_ID_MPEG4:
1291                 put_bits(&s->pb, 16, 0);
1292                 put_bits(&s->pb, 16, 0x1C3);
1293                 stuffing_count -= 4;
1294                 while(stuffing_count--){
1295                     put_bits(&s->pb, 8, 0xFF);
1296                 }
1297             break;
1298             default:
1299                 av_log(s->avctx, AV_LOG_ERROR, "vbv buffer overflow\n");
1300             }
1301             flush_put_bits(&s->pb);
1302             s->frame_bits  = put_bits_count(&s->pb);
1303         }
1304
1305         /* update mpeg1/2 vbv_delay for CBR */
1306         if(s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate && s->out_format == FMT_MPEG1
1307            && 90000LL * (avctx->rc_buffer_size-1) <= s->avctx->rc_max_rate*0xFFFFLL){
1308             int vbv_delay, min_delay;
1309             double inbits = s->avctx->rc_max_rate*av_q2d(s->avctx->time_base);
1310             int    minbits= s->frame_bits - 8*(s->vbv_delay_ptr - s->pb.buf - 1);
1311             double bits   = s->rc_context.buffer_index + minbits - inbits;
1312
1313             if(bits<0)
1314                 av_log(s->avctx, AV_LOG_ERROR, "Internal error, negative bits\n");
1315
1316             assert(s->repeat_first_field==0);
1317
1318             vbv_delay=     bits * 90000                               / s->avctx->rc_max_rate;
1319             min_delay= (minbits * 90000LL + s->avctx->rc_max_rate - 1)/ s->avctx->rc_max_rate;
1320
1321             vbv_delay= FFMAX(vbv_delay, min_delay);
1322
1323             assert(vbv_delay < 0xFFFF);
1324
1325             s->vbv_delay_ptr[0] &= 0xF8;
1326             s->vbv_delay_ptr[0] |= vbv_delay>>13;
1327             s->vbv_delay_ptr[1]  = vbv_delay>>5;
1328             s->vbv_delay_ptr[2] &= 0x07;
1329             s->vbv_delay_ptr[2] |= vbv_delay<<3;
1330         }
1331         s->total_bits += s->frame_bits;
1332         avctx->frame_bits  = s->frame_bits;
1333     }else{
1334         assert((put_bits_ptr(&s->pb) == s->pb.buf));
1335         s->frame_bits=0;
1336     }
1337     assert((s->frame_bits&7)==0);
1338
1339     return s->frame_bits/8;
1340 }
1341
1342 static inline void dct_single_coeff_elimination(MpegEncContext *s, int n, int threshold)
1343 {
1344     static const char tab[64]=
1345         {3,2,2,1,1,1,1,1,
1346          1,1,1,1,1,1,1,1,
1347          1,1,1,1,1,1,1,1,
1348          0,0,0,0,0,0,0,0,
1349          0,0,0,0,0,0,0,0,
1350          0,0,0,0,0,0,0,0,
1351          0,0,0,0,0,0,0,0,
1352          0,0,0,0,0,0,0,0};
1353     int score=0;
1354     int run=0;
1355     int i;
1356     DCTELEM *block= s->block[n];
1357     const int last_index= s->block_last_index[n];
1358     int skip_dc;
1359
1360     if(threshold<0){
1361         skip_dc=0;
1362         threshold= -threshold;
1363     }else
1364         skip_dc=1;
1365
1366     /* Are all we could set to zero already zero? */
1367     if(last_index<=skip_dc - 1) return;
1368
1369     for(i=0; i<=last_index; i++){
1370         const int j = s->intra_scantable.permutated[i];
1371         const int level = FFABS(block[j]);
1372         if(level==1){
1373             if(skip_dc && i==0) continue;
1374             score+= tab[run];
1375             run=0;
1376         }else if(level>1){
1377             return;
1378         }else{
1379             run++;
1380         }
1381     }
1382     if(score >= threshold) return;
1383     for(i=skip_dc; i<=last_index; i++){
1384         const int j = s->intra_scantable.permutated[i];
1385         block[j]=0;
1386     }
1387     if(block[0]) s->block_last_index[n]= 0;
1388     else         s->block_last_index[n]= -1;
1389 }
1390
1391 static inline void clip_coeffs(MpegEncContext *s, DCTELEM *block, int last_index)
1392 {
1393     int i;
1394     const int maxlevel= s->max_qcoeff;
1395     const int minlevel= s->min_qcoeff;
1396     int overflow=0;
1397
1398     if(s->mb_intra){
1399         i=1; //skip clipping of intra dc
1400     }else
1401         i=0;
1402
1403     for(;i<=last_index; i++){
1404         const int j= s->intra_scantable.permutated[i];
1405         int level = block[j];
1406
1407         if     (level>maxlevel){
1408             level=maxlevel;
1409             overflow++;
1410         }else if(level<minlevel){
1411             level=minlevel;
1412             overflow++;
1413         }
1414
1415         block[j]= level;
1416     }
1417
1418     if(overflow && s->avctx->mb_decision == FF_MB_DECISION_SIMPLE)
1419         av_log(s->avctx, AV_LOG_INFO, "warning, clipping %d dct coefficients to %d..%d\n", overflow, minlevel, maxlevel);
1420 }
1421
1422 static void get_visual_weight(int16_t *weight, uint8_t *ptr, int stride){
1423     int x, y;
1424 //FIXME optimize
1425     for(y=0; y<8; y++){
1426         for(x=0; x<8; x++){
1427             int x2, y2;
1428             int sum=0;
1429             int sqr=0;
1430             int count=0;
1431
1432             for(y2= FFMAX(y-1, 0); y2 < FFMIN(8, y+2); y2++){
1433                 for(x2= FFMAX(x-1, 0); x2 < FFMIN(8, x+2); x2++){
1434                     int v= ptr[x2 + y2*stride];
1435                     sum += v;
1436                     sqr += v*v;
1437                     count++;
1438                 }
1439             }
1440             weight[x + 8*y]= (36*ff_sqrt(count*sqr - sum*sum)) / count;
1441         }
1442     }
1443 }
1444
1445 static av_always_inline void encode_mb_internal(MpegEncContext *s, int motion_x, int motion_y, int mb_block_height, int mb_block_count)
1446 {
1447     int16_t weight[8][64];
1448     DCTELEM orig[8][64];
1449     const int mb_x= s->mb_x;
1450     const int mb_y= s->mb_y;
1451     int i;
1452     int skip_dct[8];
1453     int dct_offset   = s->linesize*8; //default for progressive frames
1454     uint8_t *ptr_y, *ptr_cb, *ptr_cr;
1455     int wrap_y, wrap_c;
1456
1457     for(i=0; i<mb_block_count; i++) skip_dct[i]=s->skipdct;
1458
1459     if(s->adaptive_quant){
1460         const int last_qp= s->qscale;
1461         const int mb_xy= mb_x + mb_y*s->mb_stride;
1462
1463         s->lambda= s->lambda_table[mb_xy];
1464         update_qscale(s);
1465
1466         if(!(s->flags&CODEC_FLAG_QP_RD)){
1467             s->qscale= s->current_picture_ptr->qscale_table[mb_xy];
1468             s->dquant= s->qscale - last_qp;
1469
1470             if(s->out_format==FMT_H263){
1471                 s->dquant= av_clip(s->dquant, -2, 2);
1472
1473                 if(s->codec_id==CODEC_ID_MPEG4){
1474                     if(!s->mb_intra){
1475                         if(s->pict_type == FF_B_TYPE){
1476                             if(s->dquant&1 || s->mv_dir&MV_DIRECT)
1477                                 s->dquant= 0;
1478                         }
1479                         if(s->mv_type==MV_TYPE_8X8)
1480                             s->dquant=0;
1481                     }
1482                 }
1483             }
1484         }
1485         ff_set_qscale(s, last_qp + s->dquant);
1486     }else if(s->flags&CODEC_FLAG_QP_RD)
1487         ff_set_qscale(s, s->qscale + s->dquant);
1488
1489     wrap_y = s->linesize;
1490     wrap_c = s->uvlinesize;
1491     ptr_y = s->new_picture.data[0] + (mb_y * 16 * wrap_y) + mb_x * 16;
1492     ptr_cb = s->new_picture.data[1] + (mb_y * mb_block_height * wrap_c) + mb_x * 8;
1493     ptr_cr = s->new_picture.data[2] + (mb_y * mb_block_height * wrap_c) + mb_x * 8;
1494
1495     if(mb_x*16+16 > s->width || mb_y*16+16 > s->height){
1496         uint8_t *ebuf= s->edge_emu_buffer + 32;
1497         ff_emulated_edge_mc(ebuf            , ptr_y , wrap_y,16,16,mb_x*16,mb_y*16, s->width   , s->height);
1498         ptr_y= ebuf;
1499         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);
1500         ptr_cb= ebuf+18*wrap_y;
1501         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);
1502         ptr_cr= ebuf+18*wrap_y+8;
1503     }
1504
1505     if (s->mb_intra) {
1506         if(s->flags&CODEC_FLAG_INTERLACED_DCT){
1507             int progressive_score, interlaced_score;
1508
1509             s->interlaced_dct=0;
1510             progressive_score= s->dsp.ildct_cmp[4](s, ptr_y           , NULL, wrap_y, 8)
1511                               +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y*8, NULL, wrap_y, 8) - 400;
1512
1513             if(progressive_score > 0){
1514                 interlaced_score = s->dsp.ildct_cmp[4](s, ptr_y           , NULL, wrap_y*2, 8)
1515                                   +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y  , NULL, wrap_y*2, 8);
1516                 if(progressive_score > interlaced_score){
1517                     s->interlaced_dct=1;
1518
1519                     dct_offset= wrap_y;
1520                     wrap_y<<=1;
1521                     if (s->chroma_format == CHROMA_422)
1522                         wrap_c<<=1;
1523                 }
1524             }
1525         }
1526
1527         s->dsp.get_pixels(s->block[0], ptr_y                 , wrap_y);
1528         s->dsp.get_pixels(s->block[1], ptr_y              + 8, wrap_y);
1529         s->dsp.get_pixels(s->block[2], ptr_y + dct_offset    , wrap_y);
1530         s->dsp.get_pixels(s->block[3], ptr_y + dct_offset + 8, wrap_y);
1531
1532         if(s->flags&CODEC_FLAG_GRAY){
1533             skip_dct[4]= 1;
1534             skip_dct[5]= 1;
1535         }else{
1536             s->dsp.get_pixels(s->block[4], ptr_cb, wrap_c);
1537             s->dsp.get_pixels(s->block[5], ptr_cr, wrap_c);
1538             if(!s->chroma_y_shift){ /* 422 */
1539                 s->dsp.get_pixels(s->block[6], ptr_cb + (dct_offset>>1), wrap_c);
1540                 s->dsp.get_pixels(s->block[7], ptr_cr + (dct_offset>>1), wrap_c);
1541             }
1542         }
1543     }else{
1544         op_pixels_func (*op_pix)[4];
1545         qpel_mc_func (*op_qpix)[16];
1546         uint8_t *dest_y, *dest_cb, *dest_cr;
1547
1548         dest_y  = s->dest[0];
1549         dest_cb = s->dest[1];
1550         dest_cr = s->dest[2];
1551
1552         if ((!s->no_rounding) || s->pict_type==FF_B_TYPE){
1553             op_pix = s->dsp.put_pixels_tab;
1554             op_qpix= s->dsp.put_qpel_pixels_tab;
1555         }else{
1556             op_pix = s->dsp.put_no_rnd_pixels_tab;
1557             op_qpix= s->dsp.put_no_rnd_qpel_pixels_tab;
1558         }
1559
1560         if (s->mv_dir & MV_DIR_FORWARD) {
1561             MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.data, op_pix, op_qpix);
1562             op_pix = s->dsp.avg_pixels_tab;
1563             op_qpix= s->dsp.avg_qpel_pixels_tab;
1564         }
1565         if (s->mv_dir & MV_DIR_BACKWARD) {
1566             MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.data, op_pix, op_qpix);
1567         }
1568
1569         if(s->flags&CODEC_FLAG_INTERLACED_DCT){
1570             int progressive_score, interlaced_score;
1571
1572             s->interlaced_dct=0;
1573             progressive_score= s->dsp.ildct_cmp[0](s, dest_y           , ptr_y           , wrap_y, 8)
1574                               +s->dsp.ildct_cmp[0](s, dest_y + wrap_y*8, ptr_y + wrap_y*8, wrap_y, 8) - 400;
1575
1576             if(s->avctx->ildct_cmp == FF_CMP_VSSE) progressive_score -= 400;
1577
1578             if(progressive_score>0){
1579                 interlaced_score = s->dsp.ildct_cmp[0](s, dest_y           , ptr_y           , wrap_y*2, 8)
1580                                   +s->dsp.ildct_cmp[0](s, dest_y + wrap_y  , ptr_y + wrap_y  , wrap_y*2, 8);
1581
1582                 if(progressive_score > interlaced_score){
1583                     s->interlaced_dct=1;
1584
1585                     dct_offset= wrap_y;
1586                     wrap_y<<=1;
1587                     if (s->chroma_format == CHROMA_422)
1588                         wrap_c<<=1;
1589                 }
1590             }
1591         }
1592
1593         s->dsp.diff_pixels(s->block[0], ptr_y                 , dest_y                 , wrap_y);
1594         s->dsp.diff_pixels(s->block[1], ptr_y              + 8, dest_y              + 8, wrap_y);
1595         s->dsp.diff_pixels(s->block[2], ptr_y + dct_offset    , dest_y + dct_offset    , wrap_y);
1596         s->dsp.diff_pixels(s->block[3], ptr_y + dct_offset + 8, dest_y + dct_offset + 8, wrap_y);
1597
1598         if(s->flags&CODEC_FLAG_GRAY){
1599             skip_dct[4]= 1;
1600             skip_dct[5]= 1;
1601         }else{
1602             s->dsp.diff_pixels(s->block[4], ptr_cb, dest_cb, wrap_c);
1603             s->dsp.diff_pixels(s->block[5], ptr_cr, dest_cr, wrap_c);
1604             if(!s->chroma_y_shift){ /* 422 */
1605                 s->dsp.diff_pixels(s->block[6], ptr_cb + (dct_offset>>1), dest_cb + (dct_offset>>1), wrap_c);
1606                 s->dsp.diff_pixels(s->block[7], ptr_cr + (dct_offset>>1), dest_cr + (dct_offset>>1), wrap_c);
1607             }
1608         }
1609         /* pre quantization */
1610         if(s->current_picture.mc_mb_var[s->mb_stride*mb_y+ mb_x]<2*s->qscale*s->qscale){
1611             //FIXME optimize
1612             if(s->dsp.sad[1](NULL, ptr_y               , dest_y               , wrap_y, 8) < 20*s->qscale) skip_dct[0]= 1;
1613             if(s->dsp.sad[1](NULL, ptr_y            + 8, dest_y            + 8, wrap_y, 8) < 20*s->qscale) skip_dct[1]= 1;
1614             if(s->dsp.sad[1](NULL, ptr_y +dct_offset   , dest_y +dct_offset   , wrap_y, 8) < 20*s->qscale) skip_dct[2]= 1;
1615             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;
1616             if(s->dsp.sad[1](NULL, ptr_cb              , dest_cb              , wrap_c, 8) < 20*s->qscale) skip_dct[4]= 1;
1617             if(s->dsp.sad[1](NULL, ptr_cr              , dest_cr              , wrap_c, 8) < 20*s->qscale) skip_dct[5]= 1;
1618             if(!s->chroma_y_shift){ /* 422 */
1619                 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;
1620                 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;
1621             }
1622         }
1623     }
1624
1625     if(s->avctx->quantizer_noise_shaping){
1626         if(!skip_dct[0]) get_visual_weight(weight[0], ptr_y                 , wrap_y);
1627         if(!skip_dct[1]) get_visual_weight(weight[1], ptr_y              + 8, wrap_y);
1628         if(!skip_dct[2]) get_visual_weight(weight[2], ptr_y + dct_offset    , wrap_y);
1629         if(!skip_dct[3]) get_visual_weight(weight[3], ptr_y + dct_offset + 8, wrap_y);
1630         if(!skip_dct[4]) get_visual_weight(weight[4], ptr_cb                , wrap_c);
1631         if(!skip_dct[5]) get_visual_weight(weight[5], ptr_cr                , wrap_c);
1632         if(!s->chroma_y_shift){ /* 422 */
1633             if(!skip_dct[6]) get_visual_weight(weight[6], ptr_cb + (dct_offset>>1), wrap_c);
1634             if(!skip_dct[7]) get_visual_weight(weight[7], ptr_cr + (dct_offset>>1), wrap_c);
1635         }
1636         memcpy(orig[0], s->block[0], sizeof(DCTELEM)*64*mb_block_count);
1637     }
1638
1639     /* DCT & quantize */
1640     assert(s->out_format!=FMT_MJPEG || s->qscale==8);
1641     {
1642         for(i=0;i<mb_block_count;i++) {
1643             if(!skip_dct[i]){
1644                 int overflow;
1645                 s->block_last_index[i] = s->dct_quantize(s, s->block[i], i, s->qscale, &overflow);
1646             // FIXME we could decide to change to quantizer instead of clipping
1647             // JS: I don't think that would be a good idea it could lower quality instead
1648             //     of improve it. Just INTRADC clipping deserves changes in quantizer
1649                 if (overflow) clip_coeffs(s, s->block[i], s->block_last_index[i]);
1650             }else
1651                 s->block_last_index[i]= -1;
1652         }
1653         if(s->avctx->quantizer_noise_shaping){
1654             for(i=0;i<mb_block_count;i++) {
1655                 if(!skip_dct[i]){
1656                     s->block_last_index[i] = dct_quantize_refine(s, s->block[i], weight[i], orig[i], i, s->qscale);
1657                 }
1658             }
1659         }
1660
1661         if(s->luma_elim_threshold && !s->mb_intra)
1662             for(i=0; i<4; i++)
1663                 dct_single_coeff_elimination(s, i, s->luma_elim_threshold);
1664         if(s->chroma_elim_threshold && !s->mb_intra)
1665             for(i=4; i<mb_block_count; i++)
1666                 dct_single_coeff_elimination(s, i, s->chroma_elim_threshold);
1667
1668         if(s->flags & CODEC_FLAG_CBP_RD){
1669             for(i=0;i<mb_block_count;i++) {
1670                 if(s->block_last_index[i] == -1)
1671                     s->coded_score[i]= INT_MAX/256;
1672             }
1673         }
1674     }
1675
1676     if((s->flags&CODEC_FLAG_GRAY) && s->mb_intra){
1677         s->block_last_index[4]=
1678         s->block_last_index[5]= 0;
1679         s->block[4][0]=
1680         s->block[5][0]= (1024 + s->c_dc_scale/2)/ s->c_dc_scale;
1681     }
1682
1683     //non c quantize code returns incorrect block_last_index FIXME
1684     if(s->alternate_scan && s->dct_quantize != dct_quantize_c){
1685         for(i=0; i<mb_block_count; i++){
1686             int j;
1687             if(s->block_last_index[i]>0){
1688                 for(j=63; j>0; j--){
1689                     if(s->block[i][ s->intra_scantable.permutated[j] ]) break;
1690                 }
1691                 s->block_last_index[i]= j;
1692             }
1693         }
1694     }
1695
1696     /* huffman encode */
1697     switch(s->codec_id){ //FIXME funct ptr could be slightly faster
1698     case CODEC_ID_MPEG1VIDEO:
1699     case CODEC_ID_MPEG2VIDEO:
1700         if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
1701             mpeg1_encode_mb(s, s->block, motion_x, motion_y);
1702         break;
1703     case CODEC_ID_MPEG4:
1704         if (CONFIG_MPEG4_ENCODER)
1705             mpeg4_encode_mb(s, s->block, motion_x, motion_y);
1706         break;
1707     case CODEC_ID_MSMPEG4V2:
1708     case CODEC_ID_MSMPEG4V3:
1709     case CODEC_ID_WMV1:
1710         if (CONFIG_MSMPEG4_ENCODER)
1711             msmpeg4_encode_mb(s, s->block, motion_x, motion_y);
1712         break;
1713     case CODEC_ID_WMV2:
1714         if (CONFIG_WMV2_ENCODER)
1715             ff_wmv2_encode_mb(s, s->block, motion_x, motion_y);
1716         break;
1717     case CODEC_ID_H261:
1718         if (CONFIG_H261_ENCODER)
1719             ff_h261_encode_mb(s, s->block, motion_x, motion_y);
1720         break;
1721     case CODEC_ID_H263:
1722     case CODEC_ID_H263P:
1723     case CODEC_ID_FLV1:
1724     case CODEC_ID_RV10:
1725     case CODEC_ID_RV20:
1726         if (CONFIG_H263_ENCODER || CONFIG_H263P_ENCODER ||
1727             CONFIG_FLV_ENCODER  || CONFIG_RV10_ENCODER  || CONFIG_RV20_ENCODER)
1728             h263_encode_mb(s, s->block, motion_x, motion_y);
1729         break;
1730     case CODEC_ID_MJPEG:
1731         if (CONFIG_MJPEG_ENCODER)
1732             ff_mjpeg_encode_mb(s, s->block);
1733         break;
1734     default:
1735         assert(0);
1736     }
1737 }
1738
1739 static av_always_inline void encode_mb(MpegEncContext *s, int motion_x, int motion_y)
1740 {
1741     if (s->chroma_format == CHROMA_420) encode_mb_internal(s, motion_x, motion_y,  8, 6);
1742     else                                encode_mb_internal(s, motion_x, motion_y, 16, 8);
1743 }
1744
1745 static inline void copy_context_before_encode(MpegEncContext *d, MpegEncContext *s, int type){
1746     int i;
1747
1748     memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
1749
1750     /* mpeg1 */
1751     d->mb_skip_run= s->mb_skip_run;
1752     for(i=0; i<3; i++)
1753         d->last_dc[i]= s->last_dc[i];
1754
1755     /* statistics */
1756     d->mv_bits= s->mv_bits;
1757     d->i_tex_bits= s->i_tex_bits;
1758     d->p_tex_bits= s->p_tex_bits;
1759     d->i_count= s->i_count;
1760     d->f_count= s->f_count;
1761     d->b_count= s->b_count;
1762     d->skip_count= s->skip_count;
1763     d->misc_bits= s->misc_bits;
1764     d->last_bits= 0;
1765
1766     d->mb_skipped= 0;
1767     d->qscale= s->qscale;
1768     d->dquant= s->dquant;
1769
1770     d->esc3_level_length= s->esc3_level_length;
1771 }
1772
1773 static inline void copy_context_after_encode(MpegEncContext *d, MpegEncContext *s, int type){
1774     int i;
1775
1776     memcpy(d->mv, s->mv, 2*4*2*sizeof(int));
1777     memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
1778
1779     /* mpeg1 */
1780     d->mb_skip_run= s->mb_skip_run;
1781     for(i=0; i<3; i++)
1782         d->last_dc[i]= s->last_dc[i];
1783
1784     /* statistics */
1785     d->mv_bits= s->mv_bits;
1786     d->i_tex_bits= s->i_tex_bits;
1787     d->p_tex_bits= s->p_tex_bits;
1788     d->i_count= s->i_count;
1789     d->f_count= s->f_count;
1790     d->b_count= s->b_count;
1791     d->skip_count= s->skip_count;
1792     d->misc_bits= s->misc_bits;
1793
1794     d->mb_intra= s->mb_intra;
1795     d->mb_skipped= s->mb_skipped;
1796     d->mv_type= s->mv_type;
1797     d->mv_dir= s->mv_dir;
1798     d->pb= s->pb;
1799     if(s->data_partitioning){
1800         d->pb2= s->pb2;
1801         d->tex_pb= s->tex_pb;
1802     }
1803     d->block= s->block;
1804     for(i=0; i<8; i++)
1805         d->block_last_index[i]= s->block_last_index[i];
1806     d->interlaced_dct= s->interlaced_dct;
1807     d->qscale= s->qscale;
1808
1809     d->esc3_level_length= s->esc3_level_length;
1810 }
1811
1812 static inline void encode_mb_hq(MpegEncContext *s, MpegEncContext *backup, MpegEncContext *best, int type,
1813                            PutBitContext pb[2], PutBitContext pb2[2], PutBitContext tex_pb[2],
1814                            int *dmin, int *next_block, int motion_x, int motion_y)
1815 {
1816     int score;
1817     uint8_t *dest_backup[3];
1818
1819     copy_context_before_encode(s, backup, type);
1820
1821     s->block= s->blocks[*next_block];
1822     s->pb= pb[*next_block];
1823     if(s->data_partitioning){
1824         s->pb2   = pb2   [*next_block];
1825         s->tex_pb= tex_pb[*next_block];
1826     }
1827
1828     if(*next_block){
1829         memcpy(dest_backup, s->dest, sizeof(s->dest));
1830         s->dest[0] = s->rd_scratchpad;
1831         s->dest[1] = s->rd_scratchpad + 16*s->linesize;
1832         s->dest[2] = s->rd_scratchpad + 16*s->linesize + 8;
1833         assert(s->linesize >= 32); //FIXME
1834     }
1835
1836     encode_mb(s, motion_x, motion_y);
1837
1838     score= put_bits_count(&s->pb);
1839     if(s->data_partitioning){
1840         score+= put_bits_count(&s->pb2);
1841         score+= put_bits_count(&s->tex_pb);
1842     }
1843
1844     if(s->avctx->mb_decision == FF_MB_DECISION_RD){
1845         MPV_decode_mb(s, s->block);
1846
1847         score *= s->lambda2;
1848         score += sse_mb(s) << FF_LAMBDA_SHIFT;
1849     }
1850
1851     if(*next_block){
1852         memcpy(s->dest, dest_backup, sizeof(s->dest));
1853     }
1854
1855     if(score<*dmin){
1856         *dmin= score;
1857         *next_block^=1;
1858
1859         copy_context_after_encode(best, s, type);
1860     }
1861 }
1862
1863 static int sse(MpegEncContext *s, uint8_t *src1, uint8_t *src2, int w, int h, int stride){
1864     uint32_t *sq = ff_squareTbl + 256;
1865     int acc=0;
1866     int x,y;
1867
1868     if(w==16 && h==16)
1869         return s->dsp.sse[0](NULL, src1, src2, stride, 16);
1870     else if(w==8 && h==8)
1871         return s->dsp.sse[1](NULL, src1, src2, stride, 8);
1872
1873     for(y=0; y<h; y++){
1874         for(x=0; x<w; x++){
1875             acc+= sq[src1[x + y*stride] - src2[x + y*stride]];
1876         }
1877     }
1878
1879     assert(acc>=0);
1880
1881     return acc;
1882 }
1883
1884 static int sse_mb(MpegEncContext *s){
1885     int w= 16;
1886     int h= 16;
1887
1888     if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
1889     if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
1890
1891     if(w==16 && h==16)
1892       if(s->avctx->mb_cmp == FF_CMP_NSSE){
1893         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)
1894                +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)
1895                +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);
1896       }else{
1897         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)
1898                +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)
1899                +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);
1900       }
1901     else
1902         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)
1903                +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)
1904                +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);
1905 }
1906
1907 static int pre_estimate_motion_thread(AVCodecContext *c, void *arg){
1908     MpegEncContext *s= *(void**)arg;
1909
1910
1911     s->me.pre_pass=1;
1912     s->me.dia_size= s->avctx->pre_dia_size;
1913     s->first_slice_line=1;
1914     for(s->mb_y= s->end_mb_y-1; s->mb_y >= s->start_mb_y; s->mb_y--) {
1915         for(s->mb_x=s->mb_width-1; s->mb_x >=0 ;s->mb_x--) {
1916             ff_pre_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
1917         }
1918         s->first_slice_line=0;
1919     }
1920
1921     s->me.pre_pass=0;
1922
1923     return 0;
1924 }
1925
1926 static int estimate_motion_thread(AVCodecContext *c, void *arg){
1927     MpegEncContext *s= *(void**)arg;
1928
1929     ff_check_alignment();
1930
1931     s->me.dia_size= s->avctx->dia_size;
1932     s->first_slice_line=1;
1933     for(s->mb_y= s->start_mb_y; s->mb_y < s->end_mb_y; s->mb_y++) {
1934         s->mb_x=0; //for block init below
1935         ff_init_block_index(s);
1936         for(s->mb_x=0; s->mb_x < s->mb_width; s->mb_x++) {
1937             s->block_index[0]+=2;
1938             s->block_index[1]+=2;
1939             s->block_index[2]+=2;
1940             s->block_index[3]+=2;
1941
1942             /* compute motion vector & mb_type and store in context */
1943             if(s->pict_type==FF_B_TYPE)
1944                 ff_estimate_b_frame_motion(s, s->mb_x, s->mb_y);
1945             else
1946                 ff_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
1947         }
1948         s->first_slice_line=0;
1949     }
1950     return 0;
1951 }
1952
1953 static int mb_var_thread(AVCodecContext *c, void *arg){
1954     MpegEncContext *s= *(void**)arg;
1955     int mb_x, mb_y;
1956
1957     ff_check_alignment();
1958
1959     for(mb_y=s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
1960         for(mb_x=0; mb_x < s->mb_width; mb_x++) {
1961             int xx = mb_x * 16;
1962             int yy = mb_y * 16;
1963             uint8_t *pix = s->new_picture.data[0] + (yy * s->linesize) + xx;
1964             int varc;
1965             int sum = s->dsp.pix_sum(pix, s->linesize);
1966
1967             varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500 + 128)>>8;
1968
1969             s->current_picture.mb_var [s->mb_stride * mb_y + mb_x] = varc;
1970             s->current_picture.mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
1971             s->me.mb_var_sum_temp    += varc;
1972         }
1973     }
1974     return 0;
1975 }
1976
1977 static void write_slice_end(MpegEncContext *s){
1978     if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4){
1979         if(s->partitioned_frame){
1980             ff_mpeg4_merge_partitions(s);
1981         }
1982
1983         ff_mpeg4_stuffing(&s->pb);
1984     }else if(CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG){
1985         ff_mjpeg_encode_stuffing(&s->pb);
1986     }
1987
1988     align_put_bits(&s->pb);
1989     flush_put_bits(&s->pb);
1990
1991     if((s->flags&CODEC_FLAG_PASS1) && !s->partitioned_frame)
1992         s->misc_bits+= get_bits_diff(s);
1993 }
1994
1995 static int encode_thread(AVCodecContext *c, void *arg){
1996     MpegEncContext *s= *(void**)arg;
1997     int mb_x, mb_y, pdif = 0;
1998     int chr_h= 16>>s->chroma_y_shift;
1999     int i, j;
2000     MpegEncContext best_s, backup_s;
2001     uint8_t bit_buf[2][MAX_MB_BYTES];
2002     uint8_t bit_buf2[2][MAX_MB_BYTES];
2003     uint8_t bit_buf_tex[2][MAX_MB_BYTES];
2004     PutBitContext pb[2], pb2[2], tex_pb[2];
2005 //printf("%d->%d\n", s->resync_mb_y, s->end_mb_y);
2006
2007     ff_check_alignment();
2008
2009     for(i=0; i<2; i++){
2010         init_put_bits(&pb    [i], bit_buf    [i], MAX_MB_BYTES);
2011         init_put_bits(&pb2   [i], bit_buf2   [i], MAX_MB_BYTES);
2012         init_put_bits(&tex_pb[i], bit_buf_tex[i], MAX_MB_BYTES);
2013     }
2014
2015     s->last_bits= put_bits_count(&s->pb);
2016     s->mv_bits=0;
2017     s->misc_bits=0;
2018     s->i_tex_bits=0;
2019     s->p_tex_bits=0;
2020     s->i_count=0;
2021     s->f_count=0;
2022     s->b_count=0;
2023     s->skip_count=0;
2024
2025     for(i=0; i<3; i++){
2026         /* init last dc values */
2027         /* note: quant matrix value (8) is implied here */
2028         s->last_dc[i] = 128 << s->intra_dc_precision;
2029
2030         s->current_picture.error[i] = 0;
2031     }
2032     s->mb_skip_run = 0;
2033     memset(s->last_mv, 0, sizeof(s->last_mv));
2034
2035     s->last_mv_dir = 0;
2036
2037     switch(s->codec_id){
2038     case CODEC_ID_H263:
2039     case CODEC_ID_H263P:
2040     case CODEC_ID_FLV1:
2041         if (CONFIG_H263_ENCODER || CONFIG_H263P_ENCODER || CONFIG_FLV_ENCODER)
2042             s->gob_index = ff_h263_get_gob_height(s);
2043         break;
2044     case CODEC_ID_MPEG4:
2045         if(CONFIG_MPEG4_ENCODER && s->partitioned_frame)
2046             ff_mpeg4_init_partitions(s);
2047         break;
2048     }
2049
2050     s->resync_mb_x=0;
2051     s->resync_mb_y=0;
2052     s->first_slice_line = 1;
2053     s->ptr_lastgob = s->pb.buf;
2054     for(mb_y= s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
2055 //    printf("row %d at %X\n", s->mb_y, (int)s);
2056         s->mb_x=0;
2057         s->mb_y= mb_y;
2058
2059         ff_set_qscale(s, s->qscale);
2060         ff_init_block_index(s);
2061
2062         for(mb_x=0; mb_x < s->mb_width; mb_x++) {
2063             int xy= mb_y*s->mb_stride + mb_x; // removed const, H261 needs to adjust this
2064             int mb_type= s->mb_type[xy];
2065 //            int d;
2066             int dmin= INT_MAX;
2067             int dir;
2068
2069             if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < MAX_MB_BYTES){
2070                 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
2071                 return -1;
2072             }
2073             if(s->data_partitioning){
2074                 if(   s->pb2   .buf_end - s->pb2   .buf - (put_bits_count(&s->    pb2)>>3) < MAX_MB_BYTES
2075                    || s->tex_pb.buf_end - s->tex_pb.buf - (put_bits_count(&s->tex_pb )>>3) < MAX_MB_BYTES){
2076                     av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
2077                     return -1;
2078                 }
2079             }
2080
2081             s->mb_x = mb_x;
2082             s->mb_y = mb_y;  // moved into loop, can get changed by H.261
2083             ff_update_block_index(s);
2084
2085             if(CONFIG_H261_ENCODER && s->codec_id == CODEC_ID_H261){
2086                 ff_h261_reorder_mb_index(s);
2087                 xy= s->mb_y*s->mb_stride + s->mb_x;
2088                 mb_type= s->mb_type[xy];
2089             }
2090
2091             /* write gob / video packet header  */
2092             if(s->rtp_mode){
2093                 int current_packet_size, is_gob_start;
2094
2095                 current_packet_size= ((put_bits_count(&s->pb)+7)>>3) - (s->ptr_lastgob - s->pb.buf);
2096
2097                 is_gob_start= s->avctx->rtp_payload_size && current_packet_size >= s->avctx->rtp_payload_size && mb_y + mb_x>0;
2098
2099                 if(s->start_mb_y == mb_y && mb_y > 0 && mb_x==0) is_gob_start=1;
2100
2101                 switch(s->codec_id){
2102                 case CODEC_ID_H263:
2103                 case CODEC_ID_H263P:
2104                     if(!s->h263_slice_structured)
2105                         if(s->mb_x || s->mb_y%s->gob_index) is_gob_start=0;
2106                     break;
2107                 case CODEC_ID_MPEG2VIDEO:
2108                     if(s->mb_x==0 && s->mb_y!=0) is_gob_start=1;
2109                 case CODEC_ID_MPEG1VIDEO:
2110                     if(s->mb_skip_run) is_gob_start=0;
2111                     break;
2112                 }
2113
2114                 if(is_gob_start){
2115                     if(s->start_mb_y != mb_y || mb_x!=0){
2116                         write_slice_end(s);
2117
2118                         if(CONFIG_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4 && s->partitioned_frame){
2119                             ff_mpeg4_init_partitions(s);
2120                         }
2121                     }
2122
2123                     assert((put_bits_count(&s->pb)&7) == 0);
2124                     current_packet_size= put_bits_ptr(&s->pb) - s->ptr_lastgob;
2125
2126                     if(s->avctx->error_rate && s->resync_mb_x + s->resync_mb_y > 0){
2127                         int r= put_bits_count(&s->pb)/8 + s->picture_number + 16 + s->mb_x + s->mb_y;
2128                         int d= 100 / s->avctx->error_rate;
2129                         if(r % d == 0){
2130                             current_packet_size=0;
2131 #ifndef ALT_BITSTREAM_WRITER
2132                             s->pb.buf_ptr= s->ptr_lastgob;
2133 #endif
2134                             assert(put_bits_ptr(&s->pb) == s->ptr_lastgob);
2135                         }
2136                     }
2137
2138                     if (s->avctx->rtp_callback){
2139                         int number_mb = (mb_y - s->resync_mb_y)*s->mb_width + mb_x - s->resync_mb_x;
2140                         s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, current_packet_size, number_mb);
2141                     }
2142
2143                     switch(s->codec_id){
2144                     case CODEC_ID_MPEG4:
2145                         if (CONFIG_MPEG4_ENCODER) {
2146                             ff_mpeg4_encode_video_packet_header(s);
2147                             ff_mpeg4_clean_buffers(s);
2148                         }
2149                     break;
2150                     case CODEC_ID_MPEG1VIDEO:
2151                     case CODEC_ID_MPEG2VIDEO:
2152                         if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER) {
2153                             ff_mpeg1_encode_slice_header(s);
2154                             ff_mpeg1_clean_buffers(s);
2155                         }
2156                     break;
2157                     case CODEC_ID_H263:
2158                     case CODEC_ID_H263P:
2159                         if (CONFIG_H263_ENCODER || CONFIG_H263P_ENCODER)
2160                             h263_encode_gob_header(s, mb_y);
2161                     break;
2162                     }
2163
2164                     if(s->flags&CODEC_FLAG_PASS1){
2165                         int bits= put_bits_count(&s->pb);
2166                         s->misc_bits+= bits - s->last_bits;
2167                         s->last_bits= bits;
2168                     }
2169
2170                     s->ptr_lastgob += current_packet_size;
2171                     s->first_slice_line=1;
2172                     s->resync_mb_x=mb_x;
2173                     s->resync_mb_y=mb_y;
2174                 }
2175             }
2176
2177             if(  (s->resync_mb_x   == s->mb_x)
2178                && s->resync_mb_y+1 == s->mb_y){
2179                 s->first_slice_line=0;
2180             }
2181
2182             s->mb_skipped=0;
2183             s->dquant=0; //only for QP_RD
2184
2185             if(mb_type & (mb_type-1) || (s->flags & CODEC_FLAG_QP_RD)){ // more than 1 MB type possible or CODEC_FLAG_QP_RD
2186                 int next_block=0;
2187                 int pb_bits_count, pb2_bits_count, tex_pb_bits_count;
2188
2189                 copy_context_before_encode(&backup_s, s, -1);
2190                 backup_s.pb= s->pb;
2191                 best_s.data_partitioning= s->data_partitioning;
2192                 best_s.partitioned_frame= s->partitioned_frame;
2193                 if(s->data_partitioning){
2194                     backup_s.pb2= s->pb2;
2195                     backup_s.tex_pb= s->tex_pb;
2196                 }
2197
2198                 if(mb_type&CANDIDATE_MB_TYPE_INTER){
2199                     s->mv_dir = MV_DIR_FORWARD;
2200                     s->mv_type = MV_TYPE_16X16;
2201                     s->mb_intra= 0;
2202                     s->mv[0][0][0] = s->p_mv_table[xy][0];
2203                     s->mv[0][0][1] = s->p_mv_table[xy][1];
2204                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER, pb, pb2, tex_pb,
2205                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2206                 }
2207                 if(mb_type&CANDIDATE_MB_TYPE_INTER_I){
2208                     s->mv_dir = MV_DIR_FORWARD;
2209                     s->mv_type = MV_TYPE_FIELD;
2210                     s->mb_intra= 0;
2211                     for(i=0; i<2; i++){
2212                         j= s->field_select[0][i] = s->p_field_select_table[i][xy];
2213                         s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
2214                         s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
2215                     }
2216                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER_I, pb, pb2, tex_pb,
2217                                  &dmin, &next_block, 0, 0);
2218                 }
2219                 if(mb_type&CANDIDATE_MB_TYPE_SKIPPED){
2220                     s->mv_dir = MV_DIR_FORWARD;
2221                     s->mv_type = MV_TYPE_16X16;
2222                     s->mb_intra= 0;
2223                     s->mv[0][0][0] = 0;
2224                     s->mv[0][0][1] = 0;
2225                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_SKIPPED, pb, pb2, tex_pb,
2226                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2227                 }
2228                 if(mb_type&CANDIDATE_MB_TYPE_INTER4V){
2229                     s->mv_dir = MV_DIR_FORWARD;
2230                     s->mv_type = MV_TYPE_8X8;
2231                     s->mb_intra= 0;
2232                     for(i=0; i<4; i++){
2233                         s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
2234                         s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
2235                     }
2236                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER4V, pb, pb2, tex_pb,
2237                                  &dmin, &next_block, 0, 0);
2238                 }
2239                 if(mb_type&CANDIDATE_MB_TYPE_FORWARD){
2240                     s->mv_dir = MV_DIR_FORWARD;
2241                     s->mv_type = MV_TYPE_16X16;
2242                     s->mb_intra= 0;
2243                     s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
2244                     s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2245                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD, pb, pb2, tex_pb,
2246                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
2247                 }
2248                 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD){
2249                     s->mv_dir = MV_DIR_BACKWARD;
2250                     s->mv_type = MV_TYPE_16X16;
2251                     s->mb_intra= 0;
2252                     s->mv[1][0][0] = s->b_back_mv_table[xy][0];
2253                     s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2254                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD, pb, pb2, tex_pb,
2255                                  &dmin, &next_block, s->mv[1][0][0], s->mv[1][0][1]);
2256                 }
2257                 if(mb_type&CANDIDATE_MB_TYPE_BIDIR){
2258                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2259                     s->mv_type = MV_TYPE_16X16;
2260                     s->mb_intra= 0;
2261                     s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
2262                     s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
2263                     s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
2264                     s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2265                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR, pb, pb2, tex_pb,
2266                                  &dmin, &next_block, 0, 0);
2267                 }
2268                 if(mb_type&CANDIDATE_MB_TYPE_FORWARD_I){
2269                     s->mv_dir = MV_DIR_FORWARD;
2270                     s->mv_type = MV_TYPE_FIELD;
2271                     s->mb_intra= 0;
2272                     for(i=0; i<2; i++){
2273                         j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
2274                         s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
2275                         s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
2276                     }
2277                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD_I, pb, pb2, tex_pb,
2278                                  &dmin, &next_block, 0, 0);
2279                 }
2280                 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD_I){
2281                     s->mv_dir = MV_DIR_BACKWARD;
2282                     s->mv_type = MV_TYPE_FIELD;
2283                     s->mb_intra= 0;
2284                     for(i=0; i<2; i++){
2285                         j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
2286                         s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
2287                         s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
2288                     }
2289                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD_I, pb, pb2, tex_pb,
2290                                  &dmin, &next_block, 0, 0);
2291                 }
2292                 if(mb_type&CANDIDATE_MB_TYPE_BIDIR_I){
2293                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2294                     s->mv_type = MV_TYPE_FIELD;
2295                     s->mb_intra= 0;
2296                     for(dir=0; dir<2; dir++){
2297                         for(i=0; i<2; i++){
2298                             j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
2299                             s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
2300                             s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
2301                         }
2302                     }
2303                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR_I, pb, pb2, tex_pb,
2304                                  &dmin, &next_block, 0, 0);
2305                 }
2306                 if(mb_type&CANDIDATE_MB_TYPE_INTRA){
2307                     s->mv_dir = 0;
2308                     s->mv_type = MV_TYPE_16X16;
2309                     s->mb_intra= 1;
2310                     s->mv[0][0][0] = 0;
2311                     s->mv[0][0][1] = 0;
2312                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTRA, pb, pb2, tex_pb,
2313                                  &dmin, &next_block, 0, 0);
2314                     if(s->h263_pred || s->h263_aic){
2315                         if(best_s.mb_intra)
2316                             s->mbintra_table[mb_x + mb_y*s->mb_stride]=1;
2317                         else
2318                             ff_clean_intra_table_entries(s); //old mode?
2319                     }
2320                 }
2321
2322                 if((s->flags & CODEC_FLAG_QP_RD) && dmin < INT_MAX){
2323                     if(best_s.mv_type==MV_TYPE_16X16){ //FIXME move 4mv after QPRD
2324                         const int last_qp= backup_s.qscale;
2325                         int qpi, qp, dc[6];
2326                         DCTELEM ac[6][16];
2327                         const int mvdir= (best_s.mv_dir&MV_DIR_BACKWARD) ? 1 : 0;
2328                         static const int dquant_tab[4]={-1,1,-2,2};
2329
2330                         assert(backup_s.dquant == 0);
2331
2332                         //FIXME intra
2333                         s->mv_dir= best_s.mv_dir;
2334                         s->mv_type = MV_TYPE_16X16;
2335                         s->mb_intra= best_s.mb_intra;
2336                         s->mv[0][0][0] = best_s.mv[0][0][0];
2337                         s->mv[0][0][1] = best_s.mv[0][0][1];
2338                         s->mv[1][0][0] = best_s.mv[1][0][0];
2339                         s->mv[1][0][1] = best_s.mv[1][0][1];
2340
2341                         qpi = s->pict_type == FF_B_TYPE ? 2 : 0;
2342                         for(; qpi<4; qpi++){
2343                             int dquant= dquant_tab[qpi];
2344                             qp= last_qp + dquant;
2345                             if(qp < s->avctx->qmin || qp > s->avctx->qmax)
2346                                 continue;
2347                             backup_s.dquant= dquant;
2348                             if(s->mb_intra && s->dc_val[0]){
2349                                 for(i=0; i<6; i++){
2350                                     dc[i]= s->dc_val[0][ s->block_index[i] ];
2351                                     memcpy(ac[i], s->ac_val[0][s->block_index[i]], sizeof(DCTELEM)*16);
2352                                 }
2353                             }
2354
2355                             encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
2356                                          &dmin, &next_block, s->mv[mvdir][0][0], s->mv[mvdir][0][1]);
2357                             if(best_s.qscale != qp){
2358                                 if(s->mb_intra && s->dc_val[0]){
2359                                     for(i=0; i<6; i++){
2360                                         s->dc_val[0][ s->block_index[i] ]= dc[i];
2361                                         memcpy(s->ac_val[0][s->block_index[i]], ac[i], sizeof(DCTELEM)*16);
2362                                     }
2363                                 }
2364                             }
2365                         }
2366                     }
2367                 }
2368                 if(CONFIG_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT){
2369                     int mx= s->b_direct_mv_table[xy][0];
2370                     int my= s->b_direct_mv_table[xy][1];
2371
2372                     backup_s.dquant = 0;
2373                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2374                     s->mb_intra= 0;
2375                     ff_mpeg4_set_direct_mv(s, mx, my);
2376                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
2377                                  &dmin, &next_block, mx, my);
2378                 }
2379                 if(CONFIG_MPEG4_ENCODER && mb_type&CANDIDATE_MB_TYPE_DIRECT0){
2380                     backup_s.dquant = 0;
2381                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
2382                     s->mb_intra= 0;
2383                     ff_mpeg4_set_direct_mv(s, 0, 0);
2384                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
2385                                  &dmin, &next_block, 0, 0);
2386                 }
2387                 if(!best_s.mb_intra && s->flags2&CODEC_FLAG2_SKIP_RD){
2388                     int coded=0;
2389                     for(i=0; i<6; i++)
2390                         coded |= s->block_last_index[i];
2391                     if(coded){
2392                         int mx,my;
2393                         memcpy(s->mv, best_s.mv, sizeof(s->mv));
2394                         if(CONFIG_MPEG4_ENCODER && best_s.mv_dir & MV_DIRECT){
2395                             mx=my=0; //FIXME find the one we actually used
2396                             ff_mpeg4_set_direct_mv(s, mx, my);
2397                         }else if(best_s.mv_dir&MV_DIR_BACKWARD){
2398                             mx= s->mv[1][0][0];
2399                             my= s->mv[1][0][1];
2400                         }else{
2401                             mx= s->mv[0][0][0];
2402                             my= s->mv[0][0][1];
2403                         }
2404
2405                         s->mv_dir= best_s.mv_dir;
2406                         s->mv_type = best_s.mv_type;
2407                         s->mb_intra= 0;
2408 /*                        s->mv[0][0][0] = best_s.mv[0][0][0];
2409                         s->mv[0][0][1] = best_s.mv[0][0][1];
2410                         s->mv[1][0][0] = best_s.mv[1][0][0];
2411                         s->mv[1][0][1] = best_s.mv[1][0][1];*/
2412                         backup_s.dquant= 0;
2413                         s->skipdct=1;
2414                         encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
2415                                         &dmin, &next_block, mx, my);
2416                         s->skipdct=0;
2417                     }
2418                 }
2419
2420                 s->current_picture.qscale_table[xy]= best_s.qscale;
2421
2422                 copy_context_after_encode(s, &best_s, -1);
2423
2424                 pb_bits_count= put_bits_count(&s->pb);
2425                 flush_put_bits(&s->pb);
2426                 ff_copy_bits(&backup_s.pb, bit_buf[next_block^1], pb_bits_count);
2427                 s->pb= backup_s.pb;
2428
2429                 if(s->data_partitioning){
2430                     pb2_bits_count= put_bits_count(&s->pb2);
2431                     flush_put_bits(&s->pb2);
2432                     ff_copy_bits(&backup_s.pb2, bit_buf2[next_block^1], pb2_bits_count);
2433                     s->pb2= backup_s.pb2;
2434
2435                     tex_pb_bits_count= put_bits_count(&s->tex_pb);
2436                     flush_put_bits(&s->tex_pb);
2437                     ff_copy_bits(&backup_s.tex_pb, bit_buf_tex[next_block^1], tex_pb_bits_count);
2438                     s->tex_pb= backup_s.tex_pb;
2439                 }
2440                 s->last_bits= put_bits_count(&s->pb);
2441
2442                 if (CONFIG_ANY_H263_ENCODER &&
2443                     s->out_format == FMT_H263 && s->pict_type!=FF_B_TYPE)
2444                     ff_h263_update_motion_val(s);
2445
2446                 if(next_block==0){ //FIXME 16 vs linesize16
2447                     s->dsp.put_pixels_tab[0][0](s->dest[0], s->rd_scratchpad                     , s->linesize  ,16);
2448                     s->dsp.put_pixels_tab[1][0](s->dest[1], s->rd_scratchpad + 16*s->linesize    , s->uvlinesize, 8);
2449                     s->dsp.put_pixels_tab[1][0](s->dest[2], s->rd_scratchpad + 16*s->linesize + 8, s->uvlinesize, 8);
2450                 }
2451
2452                 if(s->avctx->mb_decision == FF_MB_DECISION_BITS)
2453                     MPV_decode_mb(s, s->block);
2454             } else {
2455                 int motion_x = 0, motion_y = 0;
2456                 s->mv_type=MV_TYPE_16X16;
2457                 // only one MB-Type possible
2458
2459                 switch(mb_type){
2460                 case CANDIDATE_MB_TYPE_INTRA:
2461                     s->mv_dir = 0;
2462                     s->mb_intra= 1;
2463                     motion_x= s->mv[0][0][0] = 0;
2464                     motion_y= s->mv[0][0][1] = 0;
2465                     break;
2466                 case CANDIDATE_MB_TYPE_INTER:
2467                     s->mv_dir = MV_DIR_FORWARD;
2468                     s->mb_intra= 0;
2469                     motion_x= s->mv[0][0][0] = s->p_mv_table[xy][0];
2470                     motion_y= s->mv[0][0][1] = s->p_mv_table[xy][1];
2471                     break;
2472                 case CANDIDATE_MB_TYPE_INTER_I:
2473                     s->mv_dir = MV_DIR_FORWARD;
2474                     s->mv_type = MV_TYPE_FIELD;
2475                     s->mb_intra= 0;
2476                     for(i=0; i<2; i++){
2477                         j= s->field_select[0][i] = s->p_field_select_table[i][xy];
2478                         s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
2479                         s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
2480                     }
2481                     break;
2482                 case CANDIDATE_MB_TYPE_INTER4V:
2483                     s->mv_dir = MV_DIR_FORWARD;
2484                     s->mv_type = MV_TYPE_8X8;
2485                     s->mb_intra= 0;
2486                     for(i=0; i<4; i++){
2487                         s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
2488                         s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
2489                     }
2490                     break;
2491                 case CANDIDATE_MB_TYPE_DIRECT:
2492                     if (CONFIG_MPEG4_ENCODER) {
2493                         s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
2494                         s->mb_intra= 0;
2495                         motion_x=s->b_direct_mv_table[xy][0];
2496                         motion_y=s->b_direct_mv_table[xy][1];
2497                         ff_mpeg4_set_direct_mv(s, motion_x, motion_y);
2498                     }
2499                     break;
2500                 case CANDIDATE_MB_TYPE_DIRECT0:
2501                     if (CONFIG_MPEG4_ENCODER) {
2502                         s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD|MV_DIRECT;
2503                         s->mb_intra= 0;
2504                         ff_mpeg4_set_direct_mv(s, 0, 0);
2505                     }
2506                     break;
2507                 case CANDIDATE_MB_TYPE_BIDIR:
2508                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2509                     s->mb_intra= 0;
2510                     s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
2511                     s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
2512                     s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
2513                     s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
2514                     break;
2515                 case CANDIDATE_MB_TYPE_BACKWARD:
2516                     s->mv_dir = MV_DIR_BACKWARD;
2517                     s->mb_intra= 0;
2518                     motion_x= s->mv[1][0][0] = s->b_back_mv_table[xy][0];
2519                     motion_y= s->mv[1][0][1] = s->b_back_mv_table[xy][1];
2520                     break;
2521                 case CANDIDATE_MB_TYPE_FORWARD:
2522                     s->mv_dir = MV_DIR_FORWARD;
2523                     s->mb_intra= 0;
2524                     motion_x= s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
2525                     motion_y= s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
2526 //                    printf(" %d %d ", motion_x, motion_y);
2527                     break;
2528                 case CANDIDATE_MB_TYPE_FORWARD_I:
2529                     s->mv_dir = MV_DIR_FORWARD;
2530                     s->mv_type = MV_TYPE_FIELD;
2531                     s->mb_intra= 0;
2532                     for(i=0; i<2; i++){
2533                         j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
2534                         s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
2535                         s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
2536                     }
2537                     break;
2538                 case CANDIDATE_MB_TYPE_BACKWARD_I:
2539                     s->mv_dir = MV_DIR_BACKWARD;
2540                     s->mv_type = MV_TYPE_FIELD;
2541                     s->mb_intra= 0;
2542                     for(i=0; i<2; i++){
2543                         j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
2544                         s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
2545                         s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
2546                     }
2547                     break;
2548                 case CANDIDATE_MB_TYPE_BIDIR_I:
2549                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
2550                     s->mv_type = MV_TYPE_FIELD;
2551                     s->mb_intra= 0;
2552                     for(dir=0; dir<2; dir++){
2553                         for(i=0; i<2; i++){
2554                             j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
2555                             s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
2556                             s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
2557                         }
2558                     }
2559                     break;
2560                 default:
2561                     av_log(s->avctx, AV_LOG_ERROR, "illegal MB type\n");
2562                 }
2563
2564                 encode_mb(s, motion_x, motion_y);
2565
2566                 // RAL: Update last macroblock type
2567                 s->last_mv_dir = s->mv_dir;
2568
2569                 if (CONFIG_ANY_H263_ENCODER &&
2570                     s->out_format == FMT_H263 && s->pict_type!=FF_B_TYPE)
2571                     ff_h263_update_motion_val(s);
2572
2573                 MPV_decode_mb(s, s->block);
2574             }
2575
2576             /* clean the MV table in IPS frames for direct mode in B frames */
2577             if(s->mb_intra /* && I,P,S_TYPE */){
2578                 s->p_mv_table[xy][0]=0;
2579                 s->p_mv_table[xy][1]=0;
2580             }
2581
2582             if(s->flags&CODEC_FLAG_PSNR){
2583                 int w= 16;
2584                 int h= 16;
2585
2586                 if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
2587                 if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
2588
2589                 s->current_picture.error[0] += sse(
2590                     s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16,
2591                     s->dest[0], w, h, s->linesize);
2592                 s->current_picture.error[1] += sse(
2593                     s, s->new_picture.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*chr_h,
2594                     s->dest[1], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
2595                 s->current_picture.error[2] += sse(
2596                     s, s->new_picture.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*chr_h,
2597                     s->dest[2], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
2598             }
2599             if(s->loop_filter){
2600                 if(CONFIG_ANY_H263_ENCODER && s->out_format == FMT_H263)
2601                     ff_h263_loop_filter(s);
2602             }
2603 //printf("MB %d %d bits\n", s->mb_x+s->mb_y*s->mb_stride, put_bits_count(&s->pb));
2604         }
2605     }
2606
2607     //not beautiful here but we must write it before flushing so it has to be here
2608     if (CONFIG_MSMPEG4_ENCODER && s->msmpeg4_version && s->msmpeg4_version<4 && s->pict_type == FF_I_TYPE)
2609         msmpeg4_encode_ext_header(s);
2610
2611     write_slice_end(s);
2612
2613     /* Send the last GOB if RTP */
2614     if (s->avctx->rtp_callback) {
2615         int number_mb = (mb_y - s->resync_mb_y)*s->mb_width - s->resync_mb_x;
2616         pdif = put_bits_ptr(&s->pb) - s->ptr_lastgob;
2617         /* Call the RTP callback to send the last GOB */
2618         emms_c();
2619         s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, pdif, number_mb);
2620     }
2621
2622     return 0;
2623 }
2624
2625 #define MERGE(field) dst->field += src->field; src->field=0
2626 static void merge_context_after_me(MpegEncContext *dst, MpegEncContext *src){
2627     MERGE(me.scene_change_score);
2628     MERGE(me.mc_mb_var_sum_temp);
2629     MERGE(me.mb_var_sum_temp);
2630 }
2631
2632 static void merge_context_after_encode(MpegEncContext *dst, MpegEncContext *src){
2633     int i;
2634
2635     MERGE(dct_count[0]); //note, the other dct vars are not part of the context
2636     MERGE(dct_count[1]);
2637     MERGE(mv_bits);
2638     MERGE(i_tex_bits);
2639     MERGE(p_tex_bits);
2640     MERGE(i_count);
2641     MERGE(f_count);
2642     MERGE(b_count);
2643     MERGE(skip_count);
2644     MERGE(misc_bits);
2645     MERGE(error_count);
2646     MERGE(padding_bug_score);
2647     MERGE(current_picture.error[0]);
2648     MERGE(current_picture.error[1]);
2649     MERGE(current_picture.error[2]);
2650
2651     if(dst->avctx->noise_reduction){
2652         for(i=0; i<64; i++){
2653             MERGE(dct_error_sum[0][i]);
2654             MERGE(dct_error_sum[1][i]);
2655         }
2656     }
2657
2658     assert(put_bits_count(&src->pb) % 8 ==0);
2659     assert(put_bits_count(&dst->pb) % 8 ==0);
2660     ff_copy_bits(&dst->pb, src->pb.buf, put_bits_count(&src->pb));
2661     flush_put_bits(&dst->pb);
2662 }
2663
2664 static int estimate_qp(MpegEncContext *s, int dry_run){
2665     if (s->next_lambda){
2666         s->current_picture_ptr->quality=
2667         s->current_picture.quality = s->next_lambda;
2668         if(!dry_run) s->next_lambda= 0;
2669     } else if (!s->fixed_qscale) {
2670         s->current_picture_ptr->quality=
2671         s->current_picture.quality = ff_rate_estimate_qscale(s, dry_run);
2672         if (s->current_picture.quality < 0)
2673             return -1;
2674     }
2675
2676     if(s->adaptive_quant){
2677         switch(s->codec_id){
2678         case CODEC_ID_MPEG4:
2679             if (CONFIG_MPEG4_ENCODER)
2680                 ff_clean_mpeg4_qscales(s);
2681             break;
2682         case CODEC_ID_H263:
2683         case CODEC_ID_H263P:
2684         case CODEC_ID_FLV1:
2685             if (CONFIG_H263_ENCODER||CONFIG_H263P_ENCODER||CONFIG_FLV_ENCODER)
2686                 ff_clean_h263_qscales(s);
2687             break;
2688         }
2689
2690         s->lambda= s->lambda_table[0];
2691         //FIXME broken
2692     }else
2693         s->lambda= s->current_picture.quality;
2694 //printf("%d %d\n", s->avctx->global_quality, s->current_picture.quality);
2695     update_qscale(s);
2696     return 0;
2697 }
2698
2699 /* must be called before writing the header */
2700 static void set_frame_distances(MpegEncContext * s){
2701     assert(s->current_picture_ptr->pts != AV_NOPTS_VALUE);
2702     s->time= s->current_picture_ptr->pts*s->avctx->time_base.num;
2703
2704     if(s->pict_type==FF_B_TYPE){
2705         s->pb_time= s->pp_time - (s->last_non_b_time - s->time);
2706         assert(s->pb_time > 0 && s->pb_time < s->pp_time);
2707     }else{
2708         s->pp_time= s->time - s->last_non_b_time;
2709         s->last_non_b_time= s->time;
2710         assert(s->picture_number==0 || s->pp_time > 0);
2711     }
2712 }
2713
2714 static int encode_picture(MpegEncContext *s, int picture_number)
2715 {
2716     int i;
2717     int bits;
2718
2719     s->picture_number = picture_number;
2720
2721     /* Reset the average MB variance */
2722     s->me.mb_var_sum_temp    =
2723     s->me.mc_mb_var_sum_temp = 0;
2724
2725     /* we need to initialize some time vars before we can encode b-frames */
2726     // RAL: Condition added for MPEG1VIDEO
2727     if (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO || (s->h263_pred && !s->h263_msmpeg4))
2728         set_frame_distances(s);
2729     if(CONFIG_MPEG4_ENCODER && s->codec_id == CODEC_ID_MPEG4)
2730         ff_set_mpeg4_time(s);
2731
2732     s->me.scene_change_score=0;
2733
2734 //    s->lambda= s->current_picture_ptr->quality; //FIXME qscale / ... stuff for ME rate distortion
2735
2736     if(s->pict_type==FF_I_TYPE){
2737         if(s->msmpeg4_version >= 3) s->no_rounding=1;
2738         else                        s->no_rounding=0;
2739     }else if(s->pict_type!=FF_B_TYPE){
2740         if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
2741             s->no_rounding ^= 1;
2742     }
2743
2744     if(s->flags & CODEC_FLAG_PASS2){
2745         if (estimate_qp(s,1) < 0)
2746             return -1;
2747         ff_get_2pass_fcode(s);
2748     }else if(!(s->flags & CODEC_FLAG_QSCALE)){
2749         if(s->pict_type==FF_B_TYPE)
2750             s->lambda= s->last_lambda_for[s->pict_type];
2751         else
2752             s->lambda= s->last_lambda_for[s->last_non_b_pict_type];
2753         update_qscale(s);
2754     }
2755
2756     s->mb_intra=0; //for the rate distortion & bit compare functions
2757     for(i=1; i<s->avctx->thread_count; i++){
2758         ff_update_duplicate_context(s->thread_context[i], s);
2759     }
2760
2761     if(ff_init_me(s)<0)
2762         return -1;
2763
2764     /* Estimate motion for every MB */
2765     if(s->pict_type != FF_I_TYPE){
2766         s->lambda = (s->lambda * s->avctx->me_penalty_compensation + 128)>>8;
2767         s->lambda2= (s->lambda2* (int64_t)s->avctx->me_penalty_compensation + 128)>>8;
2768         if(s->pict_type != FF_B_TYPE && s->avctx->me_threshold==0){
2769             if((s->avctx->pre_me && s->last_non_b_pict_type==FF_I_TYPE) || s->avctx->pre_me==2){
2770                 s->avctx->execute(s->avctx, pre_estimate_motion_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count, sizeof(void*));
2771             }
2772         }
2773
2774         s->avctx->execute(s->avctx, estimate_motion_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count, sizeof(void*));
2775     }else /* if(s->pict_type == FF_I_TYPE) */{
2776         /* I-Frame */
2777         for(i=0; i<s->mb_stride*s->mb_height; i++)
2778             s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
2779
2780         if(!s->fixed_qscale){
2781             /* finding spatial complexity for I-frame rate control */
2782             s->avctx->execute(s->avctx, mb_var_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count, sizeof(void*));
2783         }
2784     }
2785     for(i=1; i<s->avctx->thread_count; i++){
2786         merge_context_after_me(s, s->thread_context[i]);
2787     }
2788     s->current_picture.mc_mb_var_sum= s->current_picture_ptr->mc_mb_var_sum= s->me.mc_mb_var_sum_temp;
2789     s->current_picture.   mb_var_sum= s->current_picture_ptr->   mb_var_sum= s->me.   mb_var_sum_temp;
2790     emms_c();
2791
2792     if(s->me.scene_change_score > s->avctx->scenechange_threshold && s->pict_type == FF_P_TYPE){
2793         s->pict_type= FF_I_TYPE;
2794         for(i=0; i<s->mb_stride*s->mb_height; i++)
2795             s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
2796 //printf("Scene change detected, encoding as I Frame %d %d\n", s->current_picture.mb_var_sum, s->current_picture.mc_mb_var_sum);
2797     }
2798
2799     if(!s->umvplus){
2800         if(s->pict_type==FF_P_TYPE || s->pict_type==FF_S_TYPE) {
2801             s->f_code= ff_get_best_fcode(s, s->p_mv_table, CANDIDATE_MB_TYPE_INTER);
2802
2803             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2804                 int a,b;
2805                 a= ff_get_best_fcode(s, s->p_field_mv_table[0][0], CANDIDATE_MB_TYPE_INTER_I); //FIXME field_select
2806                 b= ff_get_best_fcode(s, s->p_field_mv_table[1][1], CANDIDATE_MB_TYPE_INTER_I);
2807                 s->f_code= FFMAX3(s->f_code, a, b);
2808             }
2809
2810             ff_fix_long_p_mvs(s);
2811             ff_fix_long_mvs(s, NULL, 0, s->p_mv_table, s->f_code, CANDIDATE_MB_TYPE_INTER, 0);
2812             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2813                 int j;
2814                 for(i=0; i<2; i++){
2815                     for(j=0; j<2; j++)
2816                         ff_fix_long_mvs(s, s->p_field_select_table[i], j,
2817                                         s->p_field_mv_table[i][j], s->f_code, CANDIDATE_MB_TYPE_INTER_I, 0);
2818                 }
2819             }
2820         }
2821
2822         if(s->pict_type==FF_B_TYPE){
2823             int a, b;
2824
2825             a = ff_get_best_fcode(s, s->b_forw_mv_table, CANDIDATE_MB_TYPE_FORWARD);
2826             b = ff_get_best_fcode(s, s->b_bidir_forw_mv_table, CANDIDATE_MB_TYPE_BIDIR);
2827             s->f_code = FFMAX(a, b);
2828
2829             a = ff_get_best_fcode(s, s->b_back_mv_table, CANDIDATE_MB_TYPE_BACKWARD);
2830             b = ff_get_best_fcode(s, s->b_bidir_back_mv_table, CANDIDATE_MB_TYPE_BIDIR);
2831             s->b_code = FFMAX(a, b);
2832
2833             ff_fix_long_mvs(s, NULL, 0, s->b_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_FORWARD, 1);
2834             ff_fix_long_mvs(s, NULL, 0, s->b_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BACKWARD, 1);
2835             ff_fix_long_mvs(s, NULL, 0, s->b_bidir_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_BIDIR, 1);
2836             ff_fix_long_mvs(s, NULL, 0, s->b_bidir_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BIDIR, 1);
2837             if(s->flags & CODEC_FLAG_INTERLACED_ME){
2838                 int dir, j;
2839                 for(dir=0; dir<2; dir++){
2840                     for(i=0; i<2; i++){
2841                         for(j=0; j<2; j++){
2842                             int type= dir ? (CANDIDATE_MB_TYPE_BACKWARD_I|CANDIDATE_MB_TYPE_BIDIR_I)
2843                                           : (CANDIDATE_MB_TYPE_FORWARD_I |CANDIDATE_MB_TYPE_BIDIR_I);
2844                             ff_fix_long_mvs(s, s->b_field_select_table[dir][i], j,
2845                                             s->b_field_mv_table[dir][i][j], dir ? s->b_code : s->f_code, type, 1);
2846                         }
2847                     }
2848                 }
2849             }
2850         }
2851     }
2852
2853     if (estimate_qp(s, 0) < 0)
2854         return -1;
2855
2856     if(s->qscale < 3 && s->max_qcoeff<=128 && s->pict_type==FF_I_TYPE && !(s->flags & CODEC_FLAG_QSCALE))
2857         s->qscale= 3; //reduce clipping problems
2858
2859     if (s->out_format == FMT_MJPEG) {
2860         /* for mjpeg, we do include qscale in the matrix */
2861         s->intra_matrix[0] = ff_mpeg1_default_intra_matrix[0];
2862         for(i=1;i<64;i++){
2863             int j= s->dsp.idct_permutation[i];
2864
2865             s->intra_matrix[j] = av_clip_uint8((ff_mpeg1_default_intra_matrix[i] * s->qscale) >> 3);
2866         }
2867         ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
2868                        s->intra_matrix, s->intra_quant_bias, 8, 8, 1);
2869         s->qscale= 8;
2870     }
2871
2872     //FIXME var duplication
2873     s->current_picture_ptr->key_frame=
2874     s->current_picture.key_frame= s->pict_type == FF_I_TYPE; //FIXME pic_ptr
2875     s->current_picture_ptr->pict_type=
2876     s->current_picture.pict_type= s->pict_type;
2877
2878     if(s->current_picture.key_frame)
2879         s->picture_in_gop_number=0;
2880
2881     s->last_bits= put_bits_count(&s->pb);
2882     switch(s->out_format) {
2883     case FMT_MJPEG:
2884         if (CONFIG_MJPEG_ENCODER)
2885             ff_mjpeg_encode_picture_header(s);
2886         break;
2887     case FMT_H261:
2888         if (CONFIG_H261_ENCODER)
2889             ff_h261_encode_picture_header(s, picture_number);
2890         break;
2891     case FMT_H263:
2892         if (CONFIG_WMV2_ENCODER && s->codec_id == CODEC_ID_WMV2)
2893             ff_wmv2_encode_picture_header(s, picture_number);
2894         else if (CONFIG_MSMPEG4_ENCODER && s->h263_msmpeg4)
2895             msmpeg4_encode_picture_header(s, picture_number);
2896         else if (CONFIG_MPEG4_ENCODER && s->h263_pred)
2897             mpeg4_encode_picture_header(s, picture_number);
2898         else if (CONFIG_RV10_ENCODER && s->codec_id == CODEC_ID_RV10)
2899             rv10_encode_picture_header(s, picture_number);
2900         else if (CONFIG_RV20_ENCODER && s->codec_id == CODEC_ID_RV20)
2901             rv20_encode_picture_header(s, picture_number);
2902         else if (CONFIG_FLV_ENCODER && s->codec_id == CODEC_ID_FLV1)
2903             ff_flv_encode_picture_header(s, picture_number);
2904         else if (CONFIG_ANY_H263_ENCODER)
2905             h263_encode_picture_header(s, picture_number);
2906         break;
2907     case FMT_MPEG1:
2908         if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
2909             mpeg1_encode_picture_header(s, picture_number);
2910         break;
2911     case FMT_H264:
2912         break;
2913     default:
2914         assert(0);
2915     }
2916     bits= put_bits_count(&s->pb);
2917     s->header_bits= bits - s->last_bits;
2918
2919     for(i=1; i<s->avctx->thread_count; i++){
2920         update_duplicate_context_after_me(s->thread_context[i], s);
2921     }
2922     s->avctx->execute(s->avctx, encode_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count, sizeof(void*));
2923     for(i=1; i<s->avctx->thread_count; i++){
2924         merge_context_after_encode(s, s->thread_context[i]);
2925     }
2926     emms_c();
2927     return 0;
2928 }
2929
2930 void  denoise_dct_c(MpegEncContext *s, DCTELEM *block){
2931     const int intra= s->mb_intra;
2932     int i;
2933
2934     s->dct_count[intra]++;
2935
2936     for(i=0; i<64; i++){
2937         int level= block[i];
2938
2939         if(level){
2940             if(level>0){
2941                 s->dct_error_sum[intra][i] += level;
2942                 level -= s->dct_offset[intra][i];
2943                 if(level<0) level=0;
2944             }else{
2945                 s->dct_error_sum[intra][i] -= level;
2946                 level += s->dct_offset[intra][i];
2947                 if(level>0) level=0;
2948             }
2949             block[i]= level;
2950         }
2951     }
2952 }
2953
2954 int dct_quantize_trellis_c(MpegEncContext *s,
2955                         DCTELEM *block, int n,
2956                         int qscale, int *overflow){
2957     const int *qmat;
2958     const uint8_t *scantable= s->intra_scantable.scantable;
2959     const uint8_t *perm_scantable= s->intra_scantable.permutated;
2960     int max=0;
2961     unsigned int threshold1, threshold2;
2962     int bias=0;
2963     int run_tab[65];
2964     int level_tab[65];
2965     int score_tab[65];
2966     int survivor[65];
2967     int survivor_count;
2968     int last_run=0;
2969     int last_level=0;
2970     int last_score= 0;
2971     int last_i;
2972     int coeff[2][64];
2973     int coeff_count[64];
2974     int qmul, qadd, start_i, last_non_zero, i, dc;
2975     const int esc_length= s->ac_esc_length;
2976     uint8_t * length;
2977     uint8_t * last_length;
2978     const int lambda= s->lambda2 >> (FF_LAMBDA_SHIFT - 6);
2979
2980     s->dsp.fdct (block);
2981
2982     if(s->dct_error_sum)
2983         s->denoise_dct(s, block);
2984     qmul= qscale*16;
2985     qadd= ((qscale-1)|1)*8;
2986
2987     if (s->mb_intra) {
2988         int q;
2989         if (!s->h263_aic) {
2990             if (n < 4)
2991                 q = s->y_dc_scale;
2992             else
2993                 q = s->c_dc_scale;
2994             q = q << 3;
2995         } else{
2996             /* For AIC we skip quant/dequant of INTRADC */
2997             q = 1 << 3;
2998             qadd=0;
2999         }
3000
3001         /* note: block[0] is assumed to be positive */
3002         block[0] = (block[0] + (q >> 1)) / q;
3003         start_i = 1;
3004         last_non_zero = 0;
3005         qmat = s->q_intra_matrix[qscale];
3006         if(s->mpeg_quant || s->out_format == FMT_MPEG1)
3007             bias= 1<<(QMAT_SHIFT-1);
3008         length     = s->intra_ac_vlc_length;
3009         last_length= s->intra_ac_vlc_last_length;
3010     } else {
3011         start_i = 0;
3012         last_non_zero = -1;
3013         qmat = s->q_inter_matrix[qscale];
3014         length     = s->inter_ac_vlc_length;
3015         last_length= s->inter_ac_vlc_last_length;
3016     }
3017     last_i= start_i;
3018
3019     threshold1= (1<<QMAT_SHIFT) - bias - 1;
3020     threshold2= (threshold1<<1);
3021
3022     for(i=63; i>=start_i; i--) {
3023         const int j = scantable[i];
3024         int level = block[j] * qmat[j];
3025
3026         if(((unsigned)(level+threshold1))>threshold2){
3027             last_non_zero = i;
3028             break;
3029         }
3030     }
3031
3032     for(i=start_i; i<=last_non_zero; i++) {
3033         const int j = scantable[i];
3034         int level = block[j] * qmat[j];
3035
3036 //        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
3037 //           || bias-level >= (1<<(QMAT_SHIFT - 3))){
3038         if(((unsigned)(level+threshold1))>threshold2){
3039             if(level>0){
3040                 level= (bias + level)>>QMAT_SHIFT;
3041                 coeff[0][i]= level;
3042                 coeff[1][i]= level-1;
3043 //                coeff[2][k]= level-2;
3044             }else{
3045                 level= (bias - level)>>QMAT_SHIFT;
3046                 coeff[0][i]= -level;
3047                 coeff[1][i]= -level+1;
3048 //                coeff[2][k]= -level+2;
3049             }
3050             coeff_count[i]= FFMIN(level, 2);
3051             assert(coeff_count[i]);
3052             max |=level;
3053         }else{
3054             coeff[0][i]= (level>>31)|1;
3055             coeff_count[i]= 1;
3056         }
3057     }
3058
3059     *overflow= s->max_qcoeff < max; //overflow might have happened
3060
3061     if(last_non_zero < start_i){
3062         memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
3063         return last_non_zero;
3064     }
3065
3066     score_tab[start_i]= 0;
3067     survivor[0]= start_i;
3068     survivor_count= 1;
3069
3070     for(i=start_i; i<=last_non_zero; i++){
3071         int level_index, j, zero_distortion;
3072         int dct_coeff= FFABS(block[ scantable[i] ]);
3073         int best_score=256*256*256*120;
3074
3075         if (   s->dsp.fdct == fdct_ifast
3076 #ifndef FAAN_POSTSCALE
3077             || s->dsp.fdct == ff_faandct
3078 #endif
3079            )
3080             dct_coeff= (dct_coeff*ff_inv_aanscales[ scantable[i] ]) >> 12;
3081         zero_distortion= dct_coeff*dct_coeff;
3082
3083         for(level_index=0; level_index < coeff_count[i]; level_index++){
3084             int distortion;
3085             int level= coeff[level_index][i];
3086             const int alevel= FFABS(level);
3087             int unquant_coeff;
3088
3089             assert(level);
3090
3091             if(s->out_format == FMT_H263){
3092                 unquant_coeff= alevel*qmul + qadd;
3093             }else{ //MPEG1
3094                 j= s->dsp.idct_permutation[ scantable[i] ]; //FIXME optimize
3095                 if(s->mb_intra){
3096                         unquant_coeff = (int)(  alevel  * qscale * s->intra_matrix[j]) >> 3;
3097                         unquant_coeff =   (unquant_coeff - 1) | 1;
3098                 }else{
3099                         unquant_coeff = (((  alevel  << 1) + 1) * qscale * ((int) s->inter_matrix[j])) >> 4;
3100                         unquant_coeff =   (unquant_coeff - 1) | 1;
3101                 }
3102                 unquant_coeff<<= 3;
3103             }
3104
3105             distortion= (unquant_coeff - dct_coeff) * (unquant_coeff - dct_coeff) - zero_distortion;
3106             level+=64;
3107             if((level&(~127)) == 0){
3108                 for(j=survivor_count-1; j>=0; j--){
3109                     int run= i - survivor[j];
3110                     int score= distortion + length[UNI_AC_ENC_INDEX(run, level)]*lambda;
3111                     score += score_tab[i-run];
3112
3113                     if(score < best_score){
3114                         best_score= score;
3115                         run_tab[i+1]= run;
3116                         level_tab[i+1]= level-64;
3117                     }
3118                 }
3119
3120                 if(s->out_format == FMT_H263){
3121                     for(j=survivor_count-1; j>=0; j--){
3122                         int run= i - survivor[j];
3123                         int score= distortion + last_length[UNI_AC_ENC_INDEX(run, level)]*lambda;
3124                         score += score_tab[i-run];
3125                         if(score < last_score){
3126                             last_score= score;
3127                             last_run= run;
3128                             last_level= level-64;
3129                             last_i= i+1;
3130                         }
3131                     }
3132                 }
3133             }else{
3134                 distortion += esc_length*lambda;
3135                 for(j=survivor_count-1; j>=0; j--){
3136                     int run= i - survivor[j];
3137                     int score= distortion + score_tab[i-run];
3138
3139                     if(score < best_score){
3140                         best_score= score;
3141                         run_tab[i+1]= run;
3142                         level_tab[i+1]= level-64;
3143                     }
3144                 }
3145
3146                 if(s->out_format == FMT_H263){
3147                   for(j=survivor_count-1; j>=0; j--){
3148                         int run= i - survivor[j];
3149                         int score= distortion + score_tab[i-run];
3150                         if(score < last_score){
3151                             last_score= score;
3152                             last_run= run;
3153                             last_level= level-64;
3154                             last_i= i+1;
3155                         }
3156                     }
3157                 }
3158             }
3159         }
3160
3161         score_tab[i+1]= best_score;
3162
3163         //Note: there is a vlc code in mpeg4 which is 1 bit shorter then another one with a shorter run and the same level
3164         if(last_non_zero <= 27){
3165             for(; survivor_count; survivor_count--){
3166                 if(score_tab[ survivor[survivor_count-1] ] <= best_score)
3167                     break;
3168             }
3169         }else{
3170             for(; survivor_count; survivor_count--){
3171                 if(score_tab[ survivor[survivor_count-1] ] <= best_score + lambda)
3172                     break;
3173             }
3174         }
3175
3176         survivor[ survivor_count++ ]= i+1;
3177     }
3178
3179     if(s->out_format != FMT_H263){
3180         last_score= 256*256*256*120;
3181         for(i= survivor[0]; i<=last_non_zero + 1; i++){
3182             int score= score_tab[i];
3183             if(i) score += lambda*2; //FIXME exacter?
3184
3185             if(score < last_score){
3186                 last_score= score;
3187                 last_i= i;
3188                 last_level= level_tab[i];
3189                 last_run= run_tab[i];
3190             }
3191         }
3192     }
3193
3194     s->coded_score[n] = last_score;
3195
3196     dc= FFABS(block[0]);
3197     last_non_zero= last_i - 1;
3198     memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
3199
3200     if(last_non_zero < start_i)
3201         return last_non_zero;
3202
3203     if(last_non_zero == 0 && start_i == 0){
3204         int best_level= 0;
3205         int best_score= dc * dc;
3206
3207         for(i=0; i<coeff_count[0]; i++){
3208             int level= coeff[i][0];
3209             int alevel= FFABS(level);
3210             int unquant_coeff, score, distortion;
3211
3212             if(s->out_format == FMT_H263){
3213                     unquant_coeff= (alevel*qmul + qadd)>>3;
3214             }else{ //MPEG1
3215                     unquant_coeff = (((  alevel  << 1) + 1) * qscale * ((int) s->inter_matrix[0])) >> 4;
3216                     unquant_coeff =   (unquant_coeff - 1) | 1;
3217             }
3218             unquant_coeff = (unquant_coeff + 4) >> 3;
3219             unquant_coeff<<= 3 + 3;
3220
3221             distortion= (unquant_coeff - dc) * (unquant_coeff - dc);
3222             level+=64;
3223             if((level&(~127)) == 0) score= distortion + last_length[UNI_AC_ENC_INDEX(0, level)]*lambda;
3224             else                    score= distortion + esc_length*lambda;
3225
3226             if(score < best_score){
3227                 best_score= score;
3228                 best_level= level - 64;
3229             }
3230         }
3231         block[0]= best_level;
3232         s->coded_score[n] = best_score - dc*dc;
3233         if(best_level == 0) return -1;
3234         else                return last_non_zero;
3235     }
3236
3237     i= last_i;
3238     assert(last_level);
3239
3240     block[ perm_scantable[last_non_zero] ]= last_level;
3241     i -= last_run + 1;
3242
3243     for(; i>start_i; i -= run_tab[i] + 1){
3244         block[ perm_scantable[i-1] ]= level_tab[i];
3245     }
3246
3247     return last_non_zero;
3248 }
3249
3250 //#define REFINE_STATS 1
3251 static int16_t basis[64][64];
3252
3253 static void build_basis(uint8_t *perm){
3254     int i, j, x, y;
3255     emms_c();
3256     for(i=0; i<8; i++){
3257         for(j=0; j<8; j++){
3258             for(y=0; y<8; y++){
3259                 for(x=0; x<8; x++){
3260                     double s= 0.25*(1<<BASIS_SHIFT);
3261                     int index= 8*i + j;
3262                     int perm_index= perm[index];
3263                     if(i==0) s*= sqrt(0.5);
3264                     if(j==0) s*= sqrt(0.5);
3265                     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)));
3266                 }
3267             }
3268         }
3269     }
3270 }
3271
3272 static int dct_quantize_refine(MpegEncContext *s, //FIXME breaks denoise?
3273                         DCTELEM *block, int16_t *weight, DCTELEM *orig,
3274                         int n, int qscale){
3275     int16_t rem[64];
3276     DECLARE_ALIGNED_16(DCTELEM, d1[64]);
3277     const uint8_t *scantable= s->intra_scantable.scantable;
3278     const uint8_t *perm_scantable= s->intra_scantable.permutated;
3279 //    unsigned int threshold1, threshold2;
3280 //    int bias=0;
3281     int run_tab[65];
3282     int prev_run=0;
3283     int prev_level=0;
3284     int qmul, qadd, start_i, last_non_zero, i, dc;
3285     uint8_t * length;
3286     uint8_t * last_length;
3287     int lambda;
3288     int rle_index, run, q = 1, sum; //q is only used when s->mb_intra is true
3289 #ifdef REFINE_STATS
3290 static int count=0;
3291 static int after_last=0;
3292 static int to_zero=0;
3293 static int from_zero=0;
3294 static int raise=0;
3295 static int lower=0;
3296 static int messed_sign=0;
3297 #endif
3298
3299     if(basis[0][0] == 0)
3300         build_basis(s->dsp.idct_permutation);
3301
3302     qmul= qscale*2;
3303     qadd= (qscale-1)|1;
3304     if (s->mb_intra) {
3305         if (!s->h263_aic) {
3306             if (n < 4)
3307                 q = s->y_dc_scale;
3308             else
3309                 q = s->c_dc_scale;
3310         } else{
3311             /* For AIC we skip quant/dequant of INTRADC */
3312             q = 1;
3313             qadd=0;
3314         }
3315         q <<= RECON_SHIFT-3;
3316         /* note: block[0] is assumed to be positive */
3317         dc= block[0]*q;
3318 //        block[0] = (block[0] + (q >> 1)) / q;
3319         start_i = 1;
3320 //        if(s->mpeg_quant || s->out_format == FMT_MPEG1)
3321 //            bias= 1<<(QMAT_SHIFT-1);
3322         length     = s->intra_ac_vlc_length;
3323         last_length= s->intra_ac_vlc_last_length;
3324     } else {
3325         dc= 0;
3326         start_i = 0;
3327         length     = s->inter_ac_vlc_length;
3328         last_length= s->inter_ac_vlc_last_length;
3329     }
3330     last_non_zero = s->block_last_index[n];
3331
3332 #ifdef REFINE_STATS
3333 {START_TIMER
3334 #endif
3335     dc += (1<<(RECON_SHIFT-1));
3336     for(i=0; i<64; i++){
3337         rem[i]= dc - (orig[i]<<RECON_SHIFT); //FIXME  use orig dirrectly instead of copying to rem[]
3338     }
3339 #ifdef REFINE_STATS
3340 STOP_TIMER("memset rem[]")}
3341 #endif
3342     sum=0;
3343     for(i=0; i<64; i++){
3344         int one= 36;
3345         int qns=4;
3346         int w;
3347
3348         w= FFABS(weight[i]) + qns*one;
3349         w= 15 + (48*qns*one + w/2)/w; // 16 .. 63
3350
3351         weight[i] = w;
3352 //        w=weight[i] = (63*qns + (w/2)) / w;
3353
3354         assert(w>0);
3355         assert(w<(1<<6));
3356         sum += w*w;
3357     }
3358     lambda= sum*(uint64_t)s->lambda2 >> (FF_LAMBDA_SHIFT - 6 + 6 + 6 + 6);
3359 #ifdef REFINE_STATS
3360 {START_TIMER
3361 #endif
3362     run=0;
3363     rle_index=0;
3364     for(i=start_i; i<=last_non_zero; i++){
3365         int j= perm_scantable[i];
3366         const int level= block[j];
3367         int coeff;
3368
3369         if(level){
3370             if(level<0) coeff= qmul*level - qadd;
3371             else        coeff= qmul*level + qadd;
3372             run_tab[rle_index++]=run;
3373             run=0;
3374
3375             s->dsp.add_8x8basis(rem, basis[j], coeff);
3376         }else{
3377             run++;
3378         }
3379     }
3380 #ifdef REFINE_STATS
3381 if(last_non_zero>0){
3382 STOP_TIMER("init rem[]")
3383 }
3384 }
3385
3386 {START_TIMER
3387 #endif
3388     for(;;){
3389         int best_score=s->dsp.try_8x8basis(rem, weight, basis[0], 0);
3390         int best_coeff=0;
3391         int best_change=0;
3392         int run2, best_unquant_change=0, analyze_gradient;
3393 #ifdef REFINE_STATS
3394 {START_TIMER
3395 #endif
3396         analyze_gradient = last_non_zero > 2 || s->avctx->quantizer_noise_shaping >= 3;
3397
3398         if(analyze_gradient){
3399 #ifdef REFINE_STATS
3400 {START_TIMER
3401 #endif
3402             for(i=0; i<64; i++){
3403                 int w= weight[i];
3404
3405                 d1[i] = (rem[i]*w*w + (1<<(RECON_SHIFT+12-1)))>>(RECON_SHIFT+12);
3406             }
3407 #ifdef REFINE_STATS
3408 STOP_TIMER("rem*w*w")}
3409 {START_TIMER
3410 #endif
3411             s->dsp.fdct(d1);
3412 #ifdef REFINE_STATS
3413 STOP_TIMER("dct")}
3414 #endif
3415         }
3416
3417         if(start_i){
3418             const int level= block[0];
3419             int change, old_coeff;
3420
3421             assert(s->mb_intra);
3422
3423             old_coeff= q*level;
3424
3425             for(change=-1; change<=1; change+=2){
3426                 int new_level= level + change;
3427                 int score, new_coeff;
3428
3429                 new_coeff= q*new_level;
3430                 if(new_coeff >= 2048 || new_coeff < 0)
3431                     continue;
3432
3433                 score= s->dsp.try_8x8basis(rem, weight, basis[0], new_coeff - old_coeff);
3434                 if(score<best_score){
3435                     best_score= score;
3436                     best_coeff= 0;
3437                     best_change= change;
3438                     best_unquant_change= new_coeff - old_coeff;
3439                 }
3440             }
3441         }
3442
3443         run=0;
3444         rle_index=0;
3445         run2= run_tab[rle_index++];
3446         prev_level=0;
3447         prev_run=0;
3448
3449         for(i=start_i; i<64; i++){
3450             int j= perm_scantable[i];
3451             const int level= block[j];
3452             int change, old_coeff;
3453
3454             if(s->avctx->quantizer_noise_shaping < 3 && i > last_non_zero + 1)
3455                 break;
3456
3457             if(level){
3458                 if(level<0) old_coeff= qmul*level - qadd;
3459                 else        old_coeff= qmul*level + qadd;
3460                 run2= run_tab[rle_index++]; //FIXME ! maybe after last
3461             }else{
3462                 old_coeff=0;
3463                 run2--;
3464                 assert(run2>=0 || i >= last_non_zero );
3465             }
3466
3467             for(change=-1; change<=1; change+=2){
3468                 int new_level= level + change;
3469                 int score, new_coeff, unquant_change;
3470
3471                 score=0;
3472                 if(s->avctx->quantizer_noise_shaping < 2 && FFABS(new_level) > FFABS(level))
3473                    continue;
3474
3475                 if(new_level){
3476                     if(new_level<0) new_coeff= qmul*new_level - qadd;
3477                     else            new_coeff= qmul*new_level + qadd;
3478                     if(new_coeff >= 2048 || new_coeff <= -2048)
3479                         continue;
3480                     //FIXME check for overflow
3481
3482                     if(level){
3483                         if(level < 63 && level > -63){
3484                             if(i < last_non_zero)
3485                                 score +=   length[UNI_AC_ENC_INDEX(run, new_level+64)]
3486                                          - length[UNI_AC_ENC_INDEX(run, level+64)];
3487                             else
3488                                 score +=   last_length[UNI_AC_ENC_INDEX(run, new_level+64)]
3489                                          - last_length[UNI_AC_ENC_INDEX(run, level+64)];
3490                         }
3491                     }else{
3492                         assert(FFABS(new_level)==1);
3493
3494                         if(analyze_gradient){
3495                             int g= d1[ scantable[i] ];
3496                             if(g && (g^new_level) >= 0)
3497                                 continue;
3498                         }
3499
3500                         if(i < last_non_zero){
3501                             int next_i= i + run2 + 1;
3502                             int next_level= block[ perm_scantable[next_i] ] + 64;
3503
3504                             if(next_level&(~127))
3505                                 next_level= 0;
3506
3507                             if(next_i < last_non_zero)
3508                                 score +=   length[UNI_AC_ENC_INDEX(run, 65)]
3509                                          + length[UNI_AC_ENC_INDEX(run2, next_level)]
3510                                          - length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)];
3511                             else
3512                                 score +=  length[UNI_AC_ENC_INDEX(run, 65)]
3513                                         + last_length[UNI_AC_ENC_INDEX(run2, next_level)]
3514                                         - last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)];
3515                         }else{
3516                             score += last_length[UNI_AC_ENC_INDEX(run, 65)];
3517                             if(prev_level){
3518                                 score +=  length[UNI_AC_ENC_INDEX(prev_run, prev_level)]
3519                                         - last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)];
3520                             }
3521                         }
3522                     }
3523                 }else{
3524                     new_coeff=0;
3525                     assert(FFABS(level)==1);
3526
3527                     if(i < last_non_zero){
3528                         int next_i= i + run2 + 1;
3529                         int next_level= block[ perm_scantable[next_i] ] + 64;
3530
3531                         if(next_level&(~127))
3532                             next_level= 0;
3533
3534                         if(next_i < last_non_zero)
3535                             score +=   length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)]
3536                                      - length[UNI_AC_ENC_INDEX(run2, next_level)]
3537                                      - length[UNI_AC_ENC_INDEX(run, 65)];
3538                         else
3539                             score +=   last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)]
3540                                      - last_length[UNI_AC_ENC_INDEX(run2, next_level)]
3541                                      - length[UNI_AC_ENC_INDEX(run, 65)];
3542                     }else{
3543                         score += -last_length[UNI_AC_ENC_INDEX(run, 65)];
3544                         if(prev_level){
3545                             score +=  last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)]
3546                                     - length[UNI_AC_ENC_INDEX(prev_run, prev_level)];
3547                         }
3548                     }
3549                 }
3550
3551                 score *= lambda;
3552
3553                 unquant_change= new_coeff - old_coeff;
3554                 assert((score < 100*lambda && score > -100*lambda) || lambda==0);
3555
3556                 score+= s->dsp.try_8x8basis(rem, weight, basis[j], unquant_change);
3557                 if(score<best_score){
3558                     best_score= score;
3559                     best_coeff= i;
3560                     best_change= change;
3561                     best_unquant_change= unquant_change;
3562                 }
3563             }
3564             if(level){
3565                 prev_level= level + 64;
3566                 if(prev_level&(~127))
3567                     prev_level= 0;
3568                 prev_run= run;
3569                 run=0;
3570             }else{
3571                 run++;
3572             }
3573         }
3574 #ifdef REFINE_STATS
3575 STOP_TIMER("iterative step")}
3576 #endif
3577
3578         if(best_change){
3579             int j= perm_scantable[ best_coeff ];
3580
3581             block[j] += best_change;
3582
3583             if(best_coeff > last_non_zero){
3584                 last_non_zero= best_coeff;
3585                 assert(block[j]);
3586 #ifdef REFINE_STATS
3587 after_last++;
3588 #endif
3589             }else{
3590 #ifdef REFINE_STATS
3591 if(block[j]){
3592     if(block[j] - best_change){
3593         if(FFABS(block[j]) > FFABS(block[j] - best_change)){
3594             raise++;
3595         }else{
3596             lower++;
3597         }
3598     }else{
3599         from_zero++;
3600     }
3601 }else{
3602     to_zero++;
3603 }
3604 #endif
3605                 for(; last_non_zero>=start_i; last_non_zero--){
3606                     if(block[perm_scantable[last_non_zero]])
3607                         break;
3608                 }
3609             }
3610 #ifdef REFINE_STATS
3611 count++;
3612 if(256*256*256*64 % count == 0){
3613     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);
3614 }
3615 #endif
3616             run=0;
3617             rle_index=0;
3618             for(i=start_i; i<=last_non_zero; i++){
3619                 int j= perm_scantable[i];
3620                 const int level= block[j];
3621
3622                  if(level){
3623                      run_tab[rle_index++]=run;
3624                      run=0;
3625                  }else{
3626                      run++;
3627                  }
3628             }
3629
3630             s->dsp.add_8x8basis(rem, basis[j], best_unquant_change);
3631         }else{
3632             break;
3633         }
3634     }
3635 #ifdef REFINE_STATS
3636 if(last_non_zero>0){
3637 STOP_TIMER("iterative search")
3638 }
3639 }
3640 #endif
3641
3642     return last_non_zero;
3643 }
3644
3645 int dct_quantize_c(MpegEncContext *s,
3646                         DCTELEM *block, int n,
3647                         int qscale, int *overflow)
3648 {
3649     int i, j, level, last_non_zero, q, start_i;
3650     const int *qmat;
3651     const uint8_t *scantable= s->intra_scantable.scantable;
3652     int bias;
3653     int max=0;
3654     unsigned int threshold1, threshold2;
3655
3656     s->dsp.fdct (block);
3657
3658     if(s->dct_error_sum)
3659         s->denoise_dct(s, block);
3660
3661     if (s->mb_intra) {
3662         if (!s->h263_aic) {
3663             if (n < 4)
3664                 q = s->y_dc_scale;
3665             else
3666                 q = s->c_dc_scale;
3667             q = q << 3;
3668         } else
3669             /* For AIC we skip quant/dequant of INTRADC */
3670             q = 1 << 3;
3671
3672         /* note: block[0] is assumed to be positive */
3673         block[0] = (block[0] + (q >> 1)) / q;
3674         start_i = 1;
3675         last_non_zero = 0;
3676         qmat = s->q_intra_matrix[qscale];
3677         bias= s->intra_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT);
3678     } else {
3679         start_i = 0;
3680         last_non_zero = -1;
3681         qmat = s->q_inter_matrix[qscale];
3682         bias= s->inter_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT);
3683     }
3684     threshold1= (1<<QMAT_SHIFT) - bias - 1;
3685     threshold2= (threshold1<<1);
3686     for(i=63;i>=start_i;i--) {
3687         j = scantable[i];
3688         level = block[j] * qmat[j];
3689
3690         if(((unsigned)(level+threshold1))>threshold2){
3691             last_non_zero = i;
3692             break;
3693         }else{
3694             block[j]=0;
3695         }
3696     }
3697     for(i=start_i; i<=last_non_zero; i++) {
3698         j = scantable[i];
3699         level = block[j] * qmat[j];
3700
3701 //        if(   bias+level >= (1<<QMAT_SHIFT)
3702 //           || bias-level >= (1<<QMAT_SHIFT)){
3703         if(((unsigned)(level+threshold1))>threshold2){
3704             if(level>0){
3705                 level= (bias + level)>>QMAT_SHIFT;
3706                 block[j]= level;
3707             }else{
3708                 level= (bias - level)>>QMAT_SHIFT;
3709                 block[j]= -level;
3710             }
3711             max |=level;
3712         }else{
3713             block[j]=0;
3714         }
3715     }
3716     *overflow= s->max_qcoeff < max; //overflow might have happened
3717
3718     /* we need this permutation so that we correct the IDCT, we only permute the !=0 elements */
3719     if (s->dsp.idct_permutation_type != FF_NO_IDCT_PERM)
3720         ff_block_permute(block, s->dsp.idct_permutation, scantable, last_non_zero);
3721
3722     return last_non_zero;
3723 }
3724
3725 AVCodec h263_encoder = {
3726     "h263",
3727     CODEC_TYPE_VIDEO,
3728     CODEC_ID_H263,
3729     sizeof(MpegEncContext),
3730     MPV_encode_init,
3731     MPV_encode_picture,
3732     MPV_encode_end,
3733     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3734     .long_name= NULL_IF_CONFIG_SMALL("H.263 / H.263-1996"),
3735 };
3736
3737 AVCodec h263p_encoder = {
3738     "h263p",
3739     CODEC_TYPE_VIDEO,
3740     CODEC_ID_H263P,
3741     sizeof(MpegEncContext),
3742     MPV_encode_init,
3743     MPV_encode_picture,
3744     MPV_encode_end,
3745     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3746     .long_name= NULL_IF_CONFIG_SMALL("H.263+ / H.263-1998 / H.263 version 2"),
3747 };
3748
3749 AVCodec flv_encoder = {
3750     "flv",
3751     CODEC_TYPE_VIDEO,
3752     CODEC_ID_FLV1,
3753     sizeof(MpegEncContext),
3754     MPV_encode_init,
3755     MPV_encode_picture,
3756     MPV_encode_end,
3757     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3758     .long_name= NULL_IF_CONFIG_SMALL("Flash Video (FLV)"),
3759 };
3760
3761 AVCodec rv10_encoder = {
3762     "rv10",
3763     CODEC_TYPE_VIDEO,
3764     CODEC_ID_RV10,
3765     sizeof(MpegEncContext),
3766     MPV_encode_init,
3767     MPV_encode_picture,
3768     MPV_encode_end,
3769     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3770     .long_name= NULL_IF_CONFIG_SMALL("RealVideo 1.0"),
3771 };
3772
3773 AVCodec rv20_encoder = {
3774     "rv20",
3775     CODEC_TYPE_VIDEO,
3776     CODEC_ID_RV20,
3777     sizeof(MpegEncContext),
3778     MPV_encode_init,
3779     MPV_encode_picture,
3780     MPV_encode_end,
3781     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3782     .long_name= NULL_IF_CONFIG_SMALL("RealVideo 2.0"),
3783 };
3784
3785 AVCodec mpeg4_encoder = {
3786     "mpeg4",
3787     CODEC_TYPE_VIDEO,
3788     CODEC_ID_MPEG4,
3789     sizeof(MpegEncContext),
3790     MPV_encode_init,
3791     MPV_encode_picture,
3792     MPV_encode_end,
3793     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3794     .capabilities= CODEC_CAP_DELAY,
3795     .long_name= NULL_IF_CONFIG_SMALL("MPEG-4 part 2"),
3796 };
3797
3798 AVCodec msmpeg4v1_encoder = {
3799     "msmpeg4v1",
3800     CODEC_TYPE_VIDEO,
3801     CODEC_ID_MSMPEG4V1,
3802     sizeof(MpegEncContext),
3803     MPV_encode_init,
3804     MPV_encode_picture,
3805     MPV_encode_end,
3806     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3807     .long_name= NULL_IF_CONFIG_SMALL("MPEG-4 part 2 Microsoft variant version 1"),
3808 };
3809
3810 AVCodec msmpeg4v2_encoder = {
3811     "msmpeg4v2",
3812     CODEC_TYPE_VIDEO,
3813     CODEC_ID_MSMPEG4V2,
3814     sizeof(MpegEncContext),
3815     MPV_encode_init,
3816     MPV_encode_picture,
3817     MPV_encode_end,
3818     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3819     .long_name= NULL_IF_CONFIG_SMALL("MPEG-4 part 2 Microsoft variant version 2"),
3820 };
3821
3822 AVCodec msmpeg4v3_encoder = {
3823     "msmpeg4",
3824     CODEC_TYPE_VIDEO,
3825     CODEC_ID_MSMPEG4V3,
3826     sizeof(MpegEncContext),
3827     MPV_encode_init,
3828     MPV_encode_picture,
3829     MPV_encode_end,
3830     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3831     .long_name= NULL_IF_CONFIG_SMALL("MPEG-4 part 2 Microsoft variant version 3"),
3832 };
3833
3834 AVCodec wmv1_encoder = {
3835     "wmv1",
3836     CODEC_TYPE_VIDEO,
3837     CODEC_ID_WMV1,
3838     sizeof(MpegEncContext),
3839     MPV_encode_init,
3840     MPV_encode_picture,
3841     MPV_encode_end,
3842     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
3843     .long_name= NULL_IF_CONFIG_SMALL("Windows Media Video 7"),
3844 };