git.sesse.net Git - ffmpeg/blob - libavcodec/mpegvideo.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  * This file is part of FFmpeg.
   7  *
   8  * FFmpeg is free software; you can redistribute it and/or
   9  * modify it under the terms of the GNU Lesser General Public
  10  * License as published by the Free Software Foundation; either
  11  * version 2.1 of the License, or (at your option) any later version.
  12  *
  13  * FFmpeg is distributed in the hope that it will be useful,
  14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  16  * Lesser General Public License for more details.
  17  *
  18  * You should have received a copy of the GNU Lesser General Public
  19  * License along with FFmpeg; if not, write to the Free Software
  20  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  21  *
  22  * 4MV & hq & b-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
  23  */
  24
  25 /**
  26  * @file mpegvideo.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 "faandct.h"
  34 #include <limits.h>
  35
  36 #ifdef USE_FASTMEMCPY
  37 #include "libvo/fastmemcpy.h"
  38 #endif
  39
  40 //#undef NDEBUG
  41 //#include <assert.h>
  42
  43 #ifdef CONFIG_ENCODERS
  44 static int encode_picture(MpegEncContext *s, int picture_number);
  45 #endif //CONFIG_ENCODERS
  46 static void dct_unquantize_mpeg1_intra_c(MpegEncContext *s,
  47                                    DCTELEM *block, int n, int qscale);
  48 static void dct_unquantize_mpeg1_inter_c(MpegEncContext *s,
  49                                    DCTELEM *block, int n, int qscale);
  50 static void dct_unquantize_mpeg2_intra_c(MpegEncContext *s,
  51                                    DCTELEM *block, int n, int qscale);
  52 static void dct_unquantize_mpeg2_intra_bitexact(MpegEncContext *s,
  53                                    DCTELEM *block, int n, int qscale);
  54 static void dct_unquantize_mpeg2_inter_c(MpegEncContext *s,
  55                                    DCTELEM *block, int n, int qscale);
  56 static void dct_unquantize_h263_intra_c(MpegEncContext *s,
  57                                   DCTELEM *block, int n, int qscale);
  58 static void dct_unquantize_h263_inter_c(MpegEncContext *s,
  59                                   DCTELEM *block, int n, int qscale);
  60 static void draw_edges_c(uint8_t *buf, int wrap, int width, int height, int w);
  61 #ifdef CONFIG_ENCODERS
  62 static int dct_quantize_c(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow);
  63 static int dct_quantize_trellis_c(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow);
  64 static int dct_quantize_refine(MpegEncContext *s, DCTELEM *block, int16_t *weight, DCTELEM *orig, int n, int qscale);
  65 static int sse_mb(MpegEncContext *s);
  66 static void  denoise_dct_c(MpegEncContext *s, DCTELEM *block);
  67 #endif //CONFIG_ENCODERS
  68
  69 #ifdef HAVE_XVMC
  70 extern int  XVMC_field_start(MpegEncContext*s, AVCodecContext *avctx);
  71 extern void XVMC_field_end(MpegEncContext *s);
  72 extern void XVMC_decode_mb(MpegEncContext *s);
  73 #endif
  74
  75 void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w)= draw_edges_c;
  76
  77
  78 /* enable all paranoid tests for rounding, overflows, etc... */
  79 //#define PARANOID
  80
  81 //#define DEBUG
  82
  83
  84 /* for jpeg fast DCT */
  85 #define CONST_BITS 14
  86
  87 static const uint16_t aanscales[64] = {
  88     /* precomputed values scaled up by 14 bits */
  89     16384, 22725, 21407, 19266, 16384, 12873,  8867,  4520,
  90     22725, 31521, 29692, 26722, 22725, 17855, 12299,  6270,
  91     21407, 29692, 27969, 25172, 21407, 16819, 11585,  5906,
  92     19266, 26722, 25172, 22654, 19266, 15137, 10426,  5315,
  93     16384, 22725, 21407, 19266, 16384, 12873,  8867,  4520,
  94     12873, 17855, 16819, 15137, 12873, 10114,  6967,  3552,
  95     8867 , 12299, 11585, 10426,  8867,  6967,  4799,  2446,
  96     4520 ,  6270,  5906,  5315,  4520,  3552,  2446,  1247
  97 };
  98
  99 static const uint8_t h263_chroma_roundtab[16] = {
 100 //  0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15
 101     0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2,
 102 };
 103
 104 static const uint8_t ff_default_chroma_qscale_table[32]={
 105 //  0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
 106     0, 1, 2, 3, 4, 5, 6, 7, 8, 9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31
 107 };
 108
 109 #ifdef CONFIG_ENCODERS
 110 static uint8_t (*default_mv_penalty)[MAX_MV*2+1]=NULL;
 111 static uint8_t default_fcode_tab[MAX_MV*2+1];
 112
 113 enum PixelFormat ff_yuv420p_list[2]= {PIX_FMT_YUV420P, -1};
 114
 115 static void convert_matrix(DSPContext *dsp, int (*qmat)[64], uint16_t (*qmat16)[2][64],
 116                            const uint16_t *quant_matrix, int bias, int qmin, int qmax, int intra)
 117 {
 118     int qscale;
 119     int shift=0;
 120
 121     for(qscale=qmin; qscale<=qmax; qscale++){
 122         int i;
 123         if (dsp->fdct == ff_jpeg_fdct_islow
 124 #ifdef FAAN_POSTSCALE
 125             || dsp->fdct == ff_faandct
 126 #endif
 127             ) {
 128             for(i=0;i<64;i++) {
 129                 const int j= dsp->idct_permutation[i];
 130                 /* 16 <= qscale * quant_matrix[i] <= 7905 */
 131                 /* 19952         <= aanscales[i] * qscale * quant_matrix[i]           <= 249205026 */
 132                 /* (1<<36)/19952 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */
 133                 /* 3444240       >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= 275 */
 134
 135                 qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) /
 136                                 (qscale * quant_matrix[j]));
 137             }
 138         } else if (dsp->fdct == fdct_ifast
 139 #ifndef FAAN_POSTSCALE
 140                    || dsp->fdct == ff_faandct
 141 #endif
 142                    ) {
 143             for(i=0;i<64;i++) {
 144                 const int j= dsp->idct_permutation[i];
 145                 /* 16 <= qscale * quant_matrix[i] <= 7905 */
 146                 /* 19952         <= aanscales[i] * qscale * quant_matrix[i]           <= 249205026 */
 147                 /* (1<<36)/19952 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */
 148                 /* 3444240       >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= 275 */
 149
 150                 qmat[qscale][i] = (int)((UINT64_C(1) << (QMAT_SHIFT + 14)) /
 151                                 (aanscales[i] * qscale * quant_matrix[j]));
 152             }
 153         } else {
 154             for(i=0;i<64;i++) {
 155                 const int j= dsp->idct_permutation[i];
 156                 /* We can safely suppose that 16 <= quant_matrix[i] <= 255
 157                    So 16           <= qscale * quant_matrix[i]             <= 7905
 158                    so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905
 159                    so 32768        >= (1<<19) / (qscale * quant_matrix[i]) >= 67
 160                 */
 161                 qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) / (qscale * quant_matrix[j]));
 162 //                qmat  [qscale][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]);
 163                 qmat16[qscale][0][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[j]);
 164
 165                 if(qmat16[qscale][0][i]==0 || qmat16[qscale][0][i]==128*256) qmat16[qscale][0][i]=128*256-1;
 166                 qmat16[qscale][1][i]= ROUNDED_DIV(bias<<(16-QUANT_BIAS_SHIFT), qmat16[qscale][0][i]);
 167             }
 168         }
 169
 170         for(i=intra; i<64; i++){
 171             int64_t max= 8191;
 172             if (dsp->fdct == fdct_ifast
 173 #ifndef FAAN_POSTSCALE
 174                    || dsp->fdct == ff_faandct
 175 #endif
 176                    ) {
 177                 max= (8191LL*aanscales[i]) >> 14;
 178             }
 179             while(((max * qmat[qscale][i]) >> shift) > INT_MAX){
 180                 shift++;
 181             }
 182         }
 183     }
 184     if(shift){
 185         av_log(NULL, AV_LOG_INFO, "Warning, QMAT_SHIFT is larger then %d, overflows possible\n", QMAT_SHIFT - shift);
 186     }
 187 }
 188
 189 static inline void update_qscale(MpegEncContext *s){
 190     s->qscale= (s->lambda*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7);
 191     s->qscale= clip(s->qscale, s->avctx->qmin, s->avctx->qmax);
 192
 193     s->lambda2= (s->lambda*s->lambda + FF_LAMBDA_SCALE/2) >> FF_LAMBDA_SHIFT;
 194 }
 195 #endif //CONFIG_ENCODERS
 196
 197 void ff_init_scantable(uint8_t *permutation, ScanTable *st, const uint8_t *src_scantable){
 198     int i;
 199     int end;
 200
 201     st->scantable= src_scantable;
 202
 203     for(i=0; i<64; i++){
 204         int j;
 205         j = src_scantable[i];
 206         st->permutated[i] = permutation[j];
 207 #ifdef ARCH_POWERPC
 208         st->inverse[j] = i;
 209 #endif
 210     }
 211
 212     end=-1;
 213     for(i=0; i<64; i++){
 214         int j;
 215         j = st->permutated[i];
 216         if(j>end) end=j;
 217         st->raster_end[i]= end;
 218     }
 219 }
 220
 221 #ifdef CONFIG_ENCODERS
 222 void ff_write_quant_matrix(PutBitContext *pb, uint16_t *matrix){
 223     int i;
 224
 225     if(matrix){
 226         put_bits(pb, 1, 1);
 227         for(i=0;i<64;i++) {
 228             put_bits(pb, 8, matrix[ ff_zigzag_direct[i] ]);
 229         }
 230     }else
 231         put_bits(pb, 1, 0);
 232 }
 233 #endif //CONFIG_ENCODERS
 234
 235 const uint8_t *ff_find_start_code(const uint8_t * restrict p, const uint8_t *end, uint32_t * restrict state){
 236     int i;
 237
 238     assert(p<=end);
 239     if(p>=end)
 240         return end;
 241
 242     for(i=0; i<3; i++){
 243         uint32_t tmp= *state << 8;
 244         *state= tmp + *(p++);
 245         if(tmp == 0x100 || p==end)
 246             return p;
 247     }
 248
 249     while(p<end){
 250         if     (p[-1] > 1      ) p+= 3;
 251         else if(p[-2]          ) p+= 2;
 252         else if(p[-3]|(p[-1]-1)) p++;
 253         else{
 254             p++;
 255             break;
 256         }
 257     }
 258
 259     p= FFMIN(p, end)-4;
 260     *state=  be2me_32(unaligned32(p));
 261
 262     return p+4;
 263 }
 264
 265 /* init common dct for both encoder and decoder */
 266 int DCT_common_init(MpegEncContext *s)
 267 {
 268     s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_c;
 269     s->dct_unquantize_h263_inter = dct_unquantize_h263_inter_c;
 270     s->dct_unquantize_mpeg1_intra = dct_unquantize_mpeg1_intra_c;
 271     s->dct_unquantize_mpeg1_inter = dct_unquantize_mpeg1_inter_c;
 272     s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_c;
 273     if(s->flags & CODEC_FLAG_BITEXACT)
 274         s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_bitexact;
 275     s->dct_unquantize_mpeg2_inter = dct_unquantize_mpeg2_inter_c;
 276
 277 #ifdef CONFIG_ENCODERS
 278     s->dct_quantize= dct_quantize_c;
 279     s->denoise_dct= denoise_dct_c;
 280 #endif //CONFIG_ENCODERS
 281
 282 #ifdef HAVE_MMX
 283     MPV_common_init_mmx(s);
 284 #endif
 285 #ifdef ARCH_ALPHA
 286     MPV_common_init_axp(s);
 287 #endif
 288 #ifdef HAVE_MLIB
 289     MPV_common_init_mlib(s);
 290 #endif
 291 #ifdef HAVE_MMI
 292     MPV_common_init_mmi(s);
 293 #endif
 294 #ifdef ARCH_ARMV4L
 295     MPV_common_init_armv4l(s);
 296 #endif
 297 #ifdef ARCH_POWERPC
 298     MPV_common_init_ppc(s);
 299 #endif
 300
 301 #ifdef CONFIG_ENCODERS
 302     s->fast_dct_quantize= s->dct_quantize;
 303
 304     if(s->flags&CODEC_FLAG_TRELLIS_QUANT){
 305         s->dct_quantize= dct_quantize_trellis_c; //move before MPV_common_init_*
 306     }
 307
 308 #endif //CONFIG_ENCODERS
 309
 310     /* load & permutate scantables
 311        note: only wmv uses different ones
 312     */
 313     if(s->alternate_scan){
 314         ff_init_scantable(s->dsp.idct_permutation, &s->inter_scantable  , ff_alternate_vertical_scan);
 315         ff_init_scantable(s->dsp.idct_permutation, &s->intra_scantable  , ff_alternate_vertical_scan);
 316     }else{
 317         ff_init_scantable(s->dsp.idct_permutation, &s->inter_scantable  , ff_zigzag_direct);
 318         ff_init_scantable(s->dsp.idct_permutation, &s->intra_scantable  , ff_zigzag_direct);
 319     }
 320     ff_init_scantable(s->dsp.idct_permutation, &s->intra_h_scantable, ff_alternate_horizontal_scan);
 321     ff_init_scantable(s->dsp.idct_permutation, &s->intra_v_scantable, ff_alternate_vertical_scan);
 322
 323     return 0;
 324 }
 325
 326 static void copy_picture(Picture *dst, Picture *src){
 327     *dst = *src;
 328     dst->type= FF_BUFFER_TYPE_COPY;
 329 }
 330
 331 #ifdef CONFIG_ENCODERS
 332 static void copy_picture_attributes(MpegEncContext *s, AVFrame *dst, AVFrame *src){
 333     int i;
 334
 335     dst->pict_type              = src->pict_type;
 336     dst->quality                = src->quality;
 337     dst->coded_picture_number   = src->coded_picture_number;
 338     dst->display_picture_number = src->display_picture_number;
 339 //    dst->reference              = src->reference;
 340     dst->pts                    = src->pts;
 341     dst->interlaced_frame       = src->interlaced_frame;
 342     dst->top_field_first        = src->top_field_first;
 343
 344     if(s->avctx->me_threshold){
 345         if(!src->motion_val[0])
 346             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_val not set!\n");
 347         if(!src->mb_type)
 348             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.mb_type not set!\n");
 349         if(!src->ref_index[0])
 350             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.ref_index not set!\n");
 351         if(src->motion_subsample_log2 != dst->motion_subsample_log2)
 352             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_subsample_log2 doesn't match! (%d!=%d)\n",
 353             src->motion_subsample_log2, dst->motion_subsample_log2);
 354
 355         memcpy(dst->mb_type, src->mb_type, s->mb_stride * s->mb_height * sizeof(dst->mb_type[0]));
 356
 357         for(i=0; i<2; i++){
 358             int stride= ((16*s->mb_width )>>src->motion_subsample_log2) + 1;
 359             int height= ((16*s->mb_height)>>src->motion_subsample_log2);
 360
 361             if(src->motion_val[i] && src->motion_val[i] != dst->motion_val[i]){
 362                 memcpy(dst->motion_val[i], src->motion_val[i], 2*stride*height*sizeof(int16_t));
 363             }
 364             if(src->ref_index[i] && src->ref_index[i] != dst->ref_index[i]){
 365                 memcpy(dst->ref_index[i], src->ref_index[i], s->b8_stride*2*s->mb_height*sizeof(int8_t));
 366             }
 367         }
 368     }
 369 }
 370 #endif
 371
 372 /**
 373  * allocates a Picture
 374  * The pixels are allocated/set by calling get_buffer() if shared=0
 375  */
 376 static int alloc_picture(MpegEncContext *s, Picture *pic, int shared){
 377     const int big_mb_num= s->mb_stride*(s->mb_height+1) + 1; //the +1 is needed so memset(,,stride*height) doesnt sig11
 378     const int mb_array_size= s->mb_stride*s->mb_height;
 379     const int b8_array_size= s->b8_stride*s->mb_height*2;
 380     const int b4_array_size= s->b4_stride*s->mb_height*4;
 381     int i;
 382
 383     if(shared){
 384         assert(pic->data[0]);
 385         assert(pic->type == 0 || pic->type == FF_BUFFER_TYPE_SHARED);
 386         pic->type= FF_BUFFER_TYPE_SHARED;
 387     }else{
 388         int r;
 389
 390         assert(!pic->data[0]);
 391
 392         r= s->avctx->get_buffer(s->avctx, (AVFrame*)pic);
 393
 394         if(r<0 || !pic->age || !pic->type || !pic->data[0]){
 395             av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed (%d %d %d %p)\n", r, pic->age, pic->type, pic->data[0]);
 396             return -1;
 397         }
 398
 399         if(s->linesize && (s->linesize != pic->linesize[0] || s->uvlinesize != pic->linesize[1])){
 400             av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed (stride changed)\n");
 401             return -1;
 402         }
 403
 404         if(pic->linesize[1] != pic->linesize[2]){
 405             av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed (uv stride mismatch)\n");
 406             return -1;
 407         }
 408
 409         s->linesize  = pic->linesize[0];
 410         s->uvlinesize= pic->linesize[1];
 411     }
 412
 413     if(pic->qscale_table==NULL){
 414         if (s->encoding) {
 415             CHECKED_ALLOCZ(pic->mb_var   , mb_array_size * sizeof(int16_t))
 416             CHECKED_ALLOCZ(pic->mc_mb_var, mb_array_size * sizeof(int16_t))
 417             CHECKED_ALLOCZ(pic->mb_mean  , mb_array_size * sizeof(int8_t))
 418         }
 419
 420         CHECKED_ALLOCZ(pic->mbskip_table , mb_array_size * sizeof(uint8_t)+2) //the +2 is for the slice end check
 421         CHECKED_ALLOCZ(pic->qscale_table , mb_array_size * sizeof(uint8_t))
 422         CHECKED_ALLOCZ(pic->mb_type_base , big_mb_num    * sizeof(uint32_t))
 423         pic->mb_type= pic->mb_type_base + s->mb_stride+1;
 424         if(s->out_format == FMT_H264){
 425             for(i=0; i<2; i++){
 426                 CHECKED_ALLOCZ(pic->motion_val_base[i], 2 * (b4_array_size+4)  * sizeof(int16_t))
 427                 pic->motion_val[i]= pic->motion_val_base[i]+4;
 428                 CHECKED_ALLOCZ(pic->ref_index[i], b8_array_size * sizeof(uint8_t))
 429             }
 430             pic->motion_subsample_log2= 2;
 431         }else if(s->out_format == FMT_H263 || s->encoding || (s->avctx->debug&FF_DEBUG_MV) || (s->avctx->debug_mv)){
 432             for(i=0; i<2; i++){
 433                 CHECKED_ALLOCZ(pic->motion_val_base[i], 2 * (b8_array_size+4) * sizeof(int16_t))
 434                 pic->motion_val[i]= pic->motion_val_base[i]+4;
 435                 CHECKED_ALLOCZ(pic->ref_index[i], b8_array_size * sizeof(uint8_t))
 436             }
 437             pic->motion_subsample_log2= 3;
 438         }
 439         if(s->avctx->debug&FF_DEBUG_DCT_COEFF) {
 440             CHECKED_ALLOCZ(pic->dct_coeff, 64 * mb_array_size * sizeof(DCTELEM)*6)
 441         }
 442         pic->qstride= s->mb_stride;
 443         CHECKED_ALLOCZ(pic->pan_scan , 1 * sizeof(AVPanScan))
 444     }
 445
 446     //it might be nicer if the application would keep track of these but it would require a API change
 447     memmove(s->prev_pict_types+1, s->prev_pict_types, PREV_PICT_TYPES_BUFFER_SIZE-1);
 448     s->prev_pict_types[0]= s->pict_type;
 449     if(pic->age < PREV_PICT_TYPES_BUFFER_SIZE && s->prev_pict_types[pic->age] == B_TYPE)
 450         pic->age= INT_MAX; // skipped MBs in b frames are quite rare in mpeg1/2 and its a bit tricky to skip them anyway
 451
 452     return 0;
 453 fail: //for the CHECKED_ALLOCZ macro
 454     return -1;
 455 }
 456
 457 /**
 458  * deallocates a picture
 459  */
 460 static void free_picture(MpegEncContext *s, Picture *pic){
 461     int i;
 462
 463     if(pic->data[0] && pic->type!=FF_BUFFER_TYPE_SHARED){
 464         s->avctx->release_buffer(s->avctx, (AVFrame*)pic);
 465     }
 466
 467     av_freep(&pic->mb_var);
 468     av_freep(&pic->mc_mb_var);
 469     av_freep(&pic->mb_mean);
 470     av_freep(&pic->mbskip_table);
 471     av_freep(&pic->qscale_table);
 472     av_freep(&pic->mb_type_base);
 473     av_freep(&pic->dct_coeff);
 474     av_freep(&pic->pan_scan);
 475     pic->mb_type= NULL;
 476     for(i=0; i<2; i++){
 477         av_freep(&pic->motion_val_base[i]);
 478         av_freep(&pic->ref_index[i]);
 479     }
 480
 481     if(pic->type == FF_BUFFER_TYPE_SHARED){
 482         for(i=0; i<4; i++){
 483             pic->base[i]=
 484             pic->data[i]= NULL;
 485         }
 486         pic->type= 0;
 487     }
 488 }
 489
 490 static int init_duplicate_context(MpegEncContext *s, MpegEncContext *base){
 491     int i;
 492
 493     // edge emu needs blocksize + filter length - 1 (=17x17 for halfpel / 21x21 for h264)
 494     CHECKED_ALLOCZ(s->allocated_edge_emu_buffer, (s->width+64)*2*21*2); //(width + edge + align)*interlaced*MBsize*tolerance
 495     s->edge_emu_buffer= s->allocated_edge_emu_buffer + (s->width+64)*2*21;
 496
 497      //FIXME should be linesize instead of s->width*2 but that isnt known before get_buffer()
 498     CHECKED_ALLOCZ(s->me.scratchpad,  (s->width+64)*4*16*2*sizeof(uint8_t))
 499     s->rd_scratchpad=   s->me.scratchpad;
 500     s->b_scratchpad=    s->me.scratchpad;
 501     s->obmc_scratchpad= s->me.scratchpad + 16;
 502     if (s->encoding) {
 503         CHECKED_ALLOCZ(s->me.map      , ME_MAP_SIZE*sizeof(uint32_t))
 504         CHECKED_ALLOCZ(s->me.score_map, ME_MAP_SIZE*sizeof(uint32_t))
 505         if(s->avctx->noise_reduction){
 506             CHECKED_ALLOCZ(s->dct_error_sum, 2 * 64 * sizeof(int))
 507         }
 508     }
 509     CHECKED_ALLOCZ(s->blocks, 64*12*2 * sizeof(DCTELEM))
 510     s->block= s->blocks[0];
 511
 512     for(i=0;i<12;i++){
 513         s->pblocks[i] = (short *)(&s->block[i]);
 514     }
 515     return 0;
 516 fail:
 517     return -1; //free() through MPV_common_end()
 518 }
 519
 520 static void free_duplicate_context(MpegEncContext *s){
 521     if(s==NULL) return;
 522
 523     av_freep(&s->allocated_edge_emu_buffer); s->edge_emu_buffer= NULL;
 524     av_freep(&s->me.scratchpad);
 525     s->rd_scratchpad=
 526     s->b_scratchpad=
 527     s->obmc_scratchpad= NULL;
 528
 529     av_freep(&s->dct_error_sum);
 530     av_freep(&s->me.map);
 531     av_freep(&s->me.score_map);
 532     av_freep(&s->blocks);
 533     s->block= NULL;
 534 }
 535
 536 static void backup_duplicate_context(MpegEncContext *bak, MpegEncContext *src){
 537 #define COPY(a) bak->a= src->a
 538     COPY(allocated_edge_emu_buffer);
 539     COPY(edge_emu_buffer);
 540     COPY(me.scratchpad);
 541     COPY(rd_scratchpad);
 542     COPY(b_scratchpad);
 543     COPY(obmc_scratchpad);
 544     COPY(me.map);
 545     COPY(me.score_map);
 546     COPY(blocks);
 547     COPY(block);
 548     COPY(start_mb_y);
 549     COPY(end_mb_y);
 550     COPY(me.map_generation);
 551     COPY(pb);
 552     COPY(dct_error_sum);
 553     COPY(dct_count[0]);
 554     COPY(dct_count[1]);
 555 #undef COPY
 556 }
 557
 558 void ff_update_duplicate_context(MpegEncContext *dst, MpegEncContext *src){
 559     MpegEncContext bak;
 560     int i;
 561     //FIXME copy only needed parts
 562 //START_TIMER
 563     backup_duplicate_context(&bak, dst);
 564     memcpy(dst, src, sizeof(MpegEncContext));
 565     backup_duplicate_context(dst, &bak);
 566     for(i=0;i<12;i++){
 567         dst->pblocks[i] = (short *)(&dst->block[i]);
 568     }
 569 //STOP_TIMER("update_duplicate_context") //about 10k cycles / 0.01 sec for 1000frames on 1ghz with 2 threads
 570 }
 571
 572 #ifdef CONFIG_ENCODERS
 573 static void update_duplicate_context_after_me(MpegEncContext *dst, MpegEncContext *src){
 574 #define COPY(a) dst->a= src->a
 575     COPY(pict_type);
 576     COPY(current_picture);
 577     COPY(f_code);
 578     COPY(b_code);
 579     COPY(qscale);
 580     COPY(lambda);
 581     COPY(lambda2);
 582     COPY(picture_in_gop_number);
 583     COPY(gop_picture_number);
 584     COPY(frame_pred_frame_dct); //FIXME don't set in encode_header
 585     COPY(progressive_frame); //FIXME don't set in encode_header
 586     COPY(partitioned_frame); //FIXME don't set in encode_header
 587 #undef COPY
 588 }
 589 #endif
 590
 591 /**
 592  * sets the given MpegEncContext to common defaults (same for encoding and decoding).
 593  * the changed fields will not depend upon the prior state of the MpegEncContext.
 594  */
 595 static void MPV_common_defaults(MpegEncContext *s){
 596     s->y_dc_scale_table=
 597     s->c_dc_scale_table= ff_mpeg1_dc_scale_table;
 598     s->chroma_qscale_table= ff_default_chroma_qscale_table;
 599     s->progressive_frame= 1;
 600     s->progressive_sequence= 1;
 601     s->picture_structure= PICT_FRAME;
 602
 603     s->coded_picture_number = 0;
 604     s->picture_number = 0;
 605     s->input_picture_number = 0;
 606
 607     s->picture_in_gop_number = 0;
 608
 609     s->f_code = 1;
 610     s->b_code = 1;
 611 }
 612
 613 /**
 614  * sets the given MpegEncContext to defaults for decoding.
 615  * the changed fields will not depend upon the prior state of the MpegEncContext.
 616  */
 617 void MPV_decode_defaults(MpegEncContext *s){
 618     MPV_common_defaults(s);
 619 }
 620
 621 /**
 622  * sets the given MpegEncContext to defaults for encoding.
 623  * the changed fields will not depend upon the prior state of the MpegEncContext.
 624  */
 625
 626 #ifdef CONFIG_ENCODERS
 627 static void MPV_encode_defaults(MpegEncContext *s){
 628     static int done=0;
 629
 630     MPV_common_defaults(s);
 631
 632     if(!done){
 633         int i;
 634         done=1;
 635
 636         default_mv_penalty= av_mallocz( sizeof(uint8_t)*(MAX_FCODE+1)*(2*MAX_MV+1) );
 637         memset(default_fcode_tab , 0, sizeof(uint8_t)*(2*MAX_MV+1));
 638
 639         for(i=-16; i<16; i++){
 640             default_fcode_tab[i + MAX_MV]= 1;
 641         }
 642     }
 643     s->me.mv_penalty= default_mv_penalty;
 644     s->fcode_tab= default_fcode_tab;
 645 }
 646 #endif //CONFIG_ENCODERS
 647
 648 /**
 649  * init common structure for both encoder and decoder.
 650  * this assumes that some variables like width/height are already set
 651  */
 652 int MPV_common_init(MpegEncContext *s)
 653 {
 654     int y_size, c_size, yc_size, i, mb_array_size, mv_table_size, x, y;
 655
 656     s->mb_height = (s->height + 15) / 16;
 657
 658     if(s->avctx->thread_count > MAX_THREADS || (s->avctx->thread_count > s->mb_height && s->mb_height)){
 659         av_log(s->avctx, AV_LOG_ERROR, "too many threads\n");
 660         return -1;
 661     }
 662
 663     if((s->width || s->height) && avcodec_check_dimensions(s->avctx, s->width, s->height))
 664         return -1;
 665
 666     dsputil_init(&s->dsp, s->avctx);
 667     DCT_common_init(s);
 668
 669     s->flags= s->avctx->flags;
 670     s->flags2= s->avctx->flags2;
 671
 672     s->mb_width  = (s->width  + 15) / 16;
 673     s->mb_stride = s->mb_width + 1;
 674     s->b8_stride = s->mb_width*2 + 1;
 675     s->b4_stride = s->mb_width*4 + 1;
 676     mb_array_size= s->mb_height * s->mb_stride;
 677     mv_table_size= (s->mb_height+2) * s->mb_stride + 1;
 678
 679     /* set chroma shifts */
 680     avcodec_get_chroma_sub_sample(s->avctx->pix_fmt,&(s->chroma_x_shift),
 681                                                     &(s->chroma_y_shift) );
 682
 683     /* set default edge pos, will be overriden in decode_header if needed */
 684     s->h_edge_pos= s->mb_width*16;
 685     s->v_edge_pos= s->mb_height*16;
 686
 687     s->mb_num = s->mb_width * s->mb_height;
 688
 689     s->block_wrap[0]=
 690     s->block_wrap[1]=
 691     s->block_wrap[2]=
 692     s->block_wrap[3]= s->b8_stride;
 693     s->block_wrap[4]=
 694     s->block_wrap[5]= s->mb_stride;
 695
 696     y_size = s->b8_stride * (2 * s->mb_height + 1);
 697     c_size = s->mb_stride * (s->mb_height + 1);
 698     yc_size = y_size + 2 * c_size;
 699
 700     /* convert fourcc to upper case */
 701     s->codec_tag=          toupper( s->avctx->codec_tag     &0xFF)
 702                         + (toupper((s->avctx->codec_tag>>8 )&0xFF)<<8 )
 703                         + (toupper((s->avctx->codec_tag>>16)&0xFF)<<16)
 704                         + (toupper((s->avctx->codec_tag>>24)&0xFF)<<24);
 705
 706     s->stream_codec_tag=          toupper( s->avctx->stream_codec_tag     &0xFF)
 707                                + (toupper((s->avctx->stream_codec_tag>>8 )&0xFF)<<8 )
 708                                + (toupper((s->avctx->stream_codec_tag>>16)&0xFF)<<16)
 709                                + (toupper((s->avctx->stream_codec_tag>>24)&0xFF)<<24);
 710
 711     s->avctx->coded_frame= (AVFrame*)&s->current_picture;
 712
 713     CHECKED_ALLOCZ(s->mb_index2xy, (s->mb_num+1)*sizeof(int)) //error ressilience code looks cleaner with this
 714     for(y=0; y<s->mb_height; y++){
 715         for(x=0; x<s->mb_width; x++){
 716             s->mb_index2xy[ x + y*s->mb_width ] = x + y*s->mb_stride;
 717         }
 718     }
 719     s->mb_index2xy[ s->mb_height*s->mb_width ] = (s->mb_height-1)*s->mb_stride + s->mb_width; //FIXME really needed?
 720
 721     if (s->encoding) {
 722         /* Allocate MV tables */
 723         CHECKED_ALLOCZ(s->p_mv_table_base            , mv_table_size * 2 * sizeof(int16_t))
 724         CHECKED_ALLOCZ(s->b_forw_mv_table_base       , mv_table_size * 2 * sizeof(int16_t))
 725         CHECKED_ALLOCZ(s->b_back_mv_table_base       , mv_table_size * 2 * sizeof(int16_t))
 726         CHECKED_ALLOCZ(s->b_bidir_forw_mv_table_base , mv_table_size * 2 * sizeof(int16_t))
 727         CHECKED_ALLOCZ(s->b_bidir_back_mv_table_base , mv_table_size * 2 * sizeof(int16_t))
 728         CHECKED_ALLOCZ(s->b_direct_mv_table_base     , mv_table_size * 2 * sizeof(int16_t))
 729         s->p_mv_table           = s->p_mv_table_base            + s->mb_stride + 1;
 730         s->b_forw_mv_table      = s->b_forw_mv_table_base       + s->mb_stride + 1;
 731         s->b_back_mv_table      = s->b_back_mv_table_base       + s->mb_stride + 1;
 732         s->b_bidir_forw_mv_table= s->b_bidir_forw_mv_table_base + s->mb_stride + 1;
 733         s->b_bidir_back_mv_table= s->b_bidir_back_mv_table_base + s->mb_stride + 1;
 734         s->b_direct_mv_table    = s->b_direct_mv_table_base     + s->mb_stride + 1;
 735
 736         if(s->msmpeg4_version){
 737             CHECKED_ALLOCZ(s->ac_stats, 2*2*(MAX_LEVEL+1)*(MAX_RUN+1)*2*sizeof(int));
 738         }
 739         CHECKED_ALLOCZ(s->avctx->stats_out, 256);
 740
 741         /* Allocate MB type table */
 742         CHECKED_ALLOCZ(s->mb_type  , mb_array_size * sizeof(uint16_t)) //needed for encoding
 743
 744         CHECKED_ALLOCZ(s->lambda_table, mb_array_size * sizeof(int))
 745
 746         CHECKED_ALLOCZ(s->q_intra_matrix, 64*32 * sizeof(int))
 747         CHECKED_ALLOCZ(s->q_inter_matrix, 64*32 * sizeof(int))
 748         CHECKED_ALLOCZ(s->q_intra_matrix16, 64*32*2 * sizeof(uint16_t))
 749         CHECKED_ALLOCZ(s->q_inter_matrix16, 64*32*2 * sizeof(uint16_t))
 750         CHECKED_ALLOCZ(s->input_picture, MAX_PICTURE_COUNT * sizeof(Picture*))
 751         CHECKED_ALLOCZ(s->reordered_input_picture, MAX_PICTURE_COUNT * sizeof(Picture*))
 752
 753         if(s->avctx->noise_reduction){
 754             CHECKED_ALLOCZ(s->dct_offset, 2 * 64 * sizeof(uint16_t))
 755         }
 756     }
 757     CHECKED_ALLOCZ(s->picture, MAX_PICTURE_COUNT * sizeof(Picture))
 758
 759     CHECKED_ALLOCZ(s->error_status_table, mb_array_size*sizeof(uint8_t))
 760
 761     if(s->codec_id==CODEC_ID_MPEG4 || (s->flags & CODEC_FLAG_INTERLACED_ME)){
 762         /* interlaced direct mode decoding tables */
 763             for(i=0; i<2; i++){
 764                 int j, k;
 765                 for(j=0; j<2; j++){
 766                     for(k=0; k<2; k++){
 767                         CHECKED_ALLOCZ(s->b_field_mv_table_base[i][j][k]     , mv_table_size * 2 * sizeof(int16_t))
 768                         s->b_field_mv_table[i][j][k]    = s->b_field_mv_table_base[i][j][k]     + s->mb_stride + 1;
 769                     }
 770                     CHECKED_ALLOCZ(s->b_field_select_table[i][j]     , mb_array_size * 2 * sizeof(uint8_t))
 771                     CHECKED_ALLOCZ(s->p_field_mv_table_base[i][j]     , mv_table_size * 2 * sizeof(int16_t))
 772                     s->p_field_mv_table[i][j]    = s->p_field_mv_table_base[i][j]     + s->mb_stride + 1;
 773                 }
 774                 CHECKED_ALLOCZ(s->p_field_select_table[i]      , mb_array_size * 2 * sizeof(uint8_t))
 775             }
 776     }
 777     if (s->out_format == FMT_H263) {
 778         /* ac values */
 779         CHECKED_ALLOCZ(s->ac_val_base, yc_size * sizeof(int16_t) * 16);
 780         s->ac_val[0] = s->ac_val_base + s->b8_stride + 1;
 781         s->ac_val[1] = s->ac_val_base + y_size + s->mb_stride + 1;
 782         s->ac_val[2] = s->ac_val[1] + c_size;
 783
 784         /* cbp values */
 785         CHECKED_ALLOCZ(s->coded_block_base, y_size);
 786         s->coded_block= s->coded_block_base + s->b8_stride + 1;
 787
 788         /* cbp, ac_pred, pred_dir */
 789         CHECKED_ALLOCZ(s->cbp_table  , mb_array_size * sizeof(uint8_t))
 790         CHECKED_ALLOCZ(s->pred_dir_table, mb_array_size * sizeof(uint8_t))
 791     }
 792
 793     if (s->h263_pred || s->h263_plus || !s->encoding) {
 794         /* dc values */
 795         //MN: we need these for error resilience of intra-frames
 796         CHECKED_ALLOCZ(s->dc_val_base, yc_size * sizeof(int16_t));
 797         s->dc_val[0] = s->dc_val_base + s->b8_stride + 1;
 798         s->dc_val[1] = s->dc_val_base + y_size + s->mb_stride + 1;
 799         s->dc_val[2] = s->dc_val[1] + c_size;
 800         for(i=0;i<yc_size;i++)
 801             s->dc_val_base[i] = 1024;
 802     }
 803
 804     /* which mb is a intra block */
 805     CHECKED_ALLOCZ(s->mbintra_table, mb_array_size);
 806     memset(s->mbintra_table, 1, mb_array_size);
 807
 808     /* init macroblock skip table */
 809     CHECKED_ALLOCZ(s->mbskip_table, mb_array_size+2);
 810     //Note the +1 is for a quicker mpeg4 slice_end detection
 811     CHECKED_ALLOCZ(s->prev_pict_types, PREV_PICT_TYPES_BUFFER_SIZE);
 812
 813     s->parse_context.state= -1;
 814     if((s->avctx->debug&(FF_DEBUG_VIS_QP|FF_DEBUG_VIS_MB_TYPE)) || (s->avctx->debug_mv)){
 815        s->visualization_buffer[0] = av_malloc((s->mb_width*16 + 2*EDGE_WIDTH) * s->mb_height*16 + 2*EDGE_WIDTH);
 816        s->visualization_buffer[1] = av_malloc((s->mb_width*8 + EDGE_WIDTH) * s->mb_height*8 + EDGE_WIDTH);
 817        s->visualization_buffer[2] = av_malloc((s->mb_width*8 + EDGE_WIDTH) * s->mb_height*8 + EDGE_WIDTH);
 818     }
 819
 820     s->context_initialized = 1;
 821
 822     s->thread_context[0]= s;
 823     for(i=1; i<s->avctx->thread_count; i++){
 824         s->thread_context[i]= av_malloc(sizeof(MpegEncContext));
 825         memcpy(s->thread_context[i], s, sizeof(MpegEncContext));
 826     }
 827
 828     for(i=0; i<s->avctx->thread_count; i++){
 829         if(init_duplicate_context(s->thread_context[i], s) < 0)
 830            goto fail;
 831         s->thread_context[i]->start_mb_y= (s->mb_height*(i  ) + s->avctx->thread_count/2) / s->avctx->thread_count;
 832         s->thread_context[i]->end_mb_y  = (s->mb_height*(i+1) + s->avctx->thread_count/2) / s->avctx->thread_count;
 833     }
 834
 835     return 0;
 836  fail:
 837     MPV_common_end(s);
 838     return -1;
 839 }
 840
 841 /* init common structure for both encoder and decoder */
 842 void MPV_common_end(MpegEncContext *s)
 843 {
 844     int i, j, k;
 845
 846     for(i=0; i<s->avctx->thread_count; i++){
 847         free_duplicate_context(s->thread_context[i]);
 848     }
 849     for(i=1; i<s->avctx->thread_count; i++){
 850         av_freep(&s->thread_context[i]);
 851     }
 852
 853     av_freep(&s->parse_context.buffer);
 854     s->parse_context.buffer_size=0;
 855
 856     av_freep(&s->mb_type);
 857     av_freep(&s->p_mv_table_base);
 858     av_freep(&s->b_forw_mv_table_base);
 859     av_freep(&s->b_back_mv_table_base);
 860     av_freep(&s->b_bidir_forw_mv_table_base);
 861     av_freep(&s->b_bidir_back_mv_table_base);
 862     av_freep(&s->b_direct_mv_table_base);
 863     s->p_mv_table= NULL;
 864     s->b_forw_mv_table= NULL;
 865     s->b_back_mv_table= NULL;
 866     s->b_bidir_forw_mv_table= NULL;
 867     s->b_bidir_back_mv_table= NULL;
 868     s->b_direct_mv_table= NULL;
 869     for(i=0; i<2; i++){
 870         for(j=0; j<2; j++){
 871             for(k=0; k<2; k++){
 872                 av_freep(&s->b_field_mv_table_base[i][j][k]);
 873                 s->b_field_mv_table[i][j][k]=NULL;
 874             }
 875             av_freep(&s->b_field_select_table[i][j]);
 876             av_freep(&s->p_field_mv_table_base[i][j]);
 877             s->p_field_mv_table[i][j]=NULL;
 878         }
 879         av_freep(&s->p_field_select_table[i]);
 880     }
 881
 882     av_freep(&s->dc_val_base);
 883     av_freep(&s->ac_val_base);
 884     av_freep(&s->coded_block_base);
 885     av_freep(&s->mbintra_table);
 886     av_freep(&s->cbp_table);
 887     av_freep(&s->pred_dir_table);
 888
 889     av_freep(&s->mbskip_table);
 890     av_freep(&s->prev_pict_types);
 891     av_freep(&s->bitstream_buffer);
 892     s->allocated_bitstream_buffer_size=0;
 893
 894     av_freep(&s->avctx->stats_out);
 895     av_freep(&s->ac_stats);
 896     av_freep(&s->error_status_table);
 897     av_freep(&s->mb_index2xy);
 898     av_freep(&s->lambda_table);
 899     av_freep(&s->q_intra_matrix);
 900     av_freep(&s->q_inter_matrix);
 901     av_freep(&s->q_intra_matrix16);
 902     av_freep(&s->q_inter_matrix16);
 903     av_freep(&s->input_picture);
 904     av_freep(&s->reordered_input_picture);
 905     av_freep(&s->dct_offset);
 906
 907     if(s->picture){
 908         for(i=0; i<MAX_PICTURE_COUNT; i++){
 909             free_picture(s, &s->picture[i]);
 910         }
 911     }
 912     av_freep(&s->picture);
 913     s->context_initialized = 0;
 914     s->last_picture_ptr=
 915     s->next_picture_ptr=
 916     s->current_picture_ptr= NULL;
 917     s->linesize= s->uvlinesize= 0;
 918
 919     for(i=0; i<3; i++)
 920         av_freep(&s->visualization_buffer[i]);
 921
 922     avcodec_default_free_buffers(s->avctx);
 923 }
 924
 925 #ifdef CONFIG_ENCODERS
 926
 927 /* init video encoder */
 928 int MPV_encode_init(AVCodecContext *avctx)
 929 {
 930     MpegEncContext *s = avctx->priv_data;
 931     int i;
 932     int chroma_h_shift, chroma_v_shift;
 933
 934     MPV_encode_defaults(s);
 935
 936     switch (avctx->codec_id) {
 937     case CODEC_ID_MPEG2VIDEO:
 938         if(avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P){
 939             av_log(avctx, AV_LOG_ERROR, "only YUV420 and YUV422 are supported\n");
 940             return -1;
 941         }
 942         break;
 943     case CODEC_ID_LJPEG:
 944     case CODEC_ID_MJPEG:
 945         if(avctx->pix_fmt != PIX_FMT_YUVJ420P && avctx->pix_fmt != PIX_FMT_YUVJ422P &&
 946            ((avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUV422P) || avctx->strict_std_compliance>FF_COMPLIANCE_INOFFICIAL)){
 947             av_log(avctx, AV_LOG_ERROR, "colorspace not supported in jpeg\n");
 948             return -1;
 949         }
 950         break;
 951     default:
 952         if(avctx->pix_fmt != PIX_FMT_YUV420P){
 953             av_log(avctx, AV_LOG_ERROR, "only YUV420 is supported\n");
 954             return -1;
 955         }
 956     }
 957
 958     switch (avctx->pix_fmt) {
 959     case PIX_FMT_YUVJ422P:
 960     case PIX_FMT_YUV422P:
 961         s->chroma_format = CHROMA_422;
 962         break;
 963     case PIX_FMT_YUVJ420P:
 964     case PIX_FMT_YUV420P:
 965     default:
 966         s->chroma_format = CHROMA_420;
 967         break;
 968     }
 969
 970     s->bit_rate = avctx->bit_rate;
 971     s->width = avctx->width;
 972     s->height = avctx->height;
 973     if(avctx->gop_size > 600 && avctx->strict_std_compliance>FF_COMPLIANCE_EXPERIMENTAL){
 974         av_log(avctx, AV_LOG_ERROR, "Warning keyframe interval too large! reducing it ...\n");
 975         avctx->gop_size=600;
 976     }
 977     s->gop_size = avctx->gop_size;
 978     s->avctx = avctx;
 979     s->flags= avctx->flags;
 980     s->flags2= avctx->flags2;
 981     s->max_b_frames= avctx->max_b_frames;
 982     s->codec_id= avctx->codec->id;
 983     s->luma_elim_threshold  = avctx->luma_elim_threshold;
 984     s->chroma_elim_threshold= avctx->chroma_elim_threshold;
 985     s->strict_std_compliance= avctx->strict_std_compliance;
 986     s->data_partitioning= avctx->flags & CODEC_FLAG_PART;
 987     s->quarter_sample= (avctx->flags & CODEC_FLAG_QPEL)!=0;
 988     s->mpeg_quant= avctx->mpeg_quant;
 989     s->rtp_mode= !!avctx->rtp_payload_size;
 990     s->intra_dc_precision= avctx->intra_dc_precision;
 991     s->user_specified_pts = AV_NOPTS_VALUE;
 992
 993     if (s->gop_size <= 1) {
 994         s->intra_only = 1;
 995         s->gop_size = 12;
 996     } else {
 997         s->intra_only = 0;
 998     }
 999
1000     s->me_method = avctx->me_method;
1001
1002     /* Fixed QSCALE */
1003     s->fixed_qscale = !!(avctx->flags & CODEC_FLAG_QSCALE);
1004
1005     s->adaptive_quant= (   s->avctx->lumi_masking
1006                         || s->avctx->dark_masking
1007                         || s->avctx->temporal_cplx_masking
1008                         || s->avctx->spatial_cplx_masking
1009                         || s->avctx->p_masking
1010                         || s->avctx->border_masking
1011                         || (s->flags&CODEC_FLAG_QP_RD))
1012                        && !s->fixed_qscale;
1013
1014     s->obmc= !!(s->flags & CODEC_FLAG_OBMC);
1015     s->loop_filter= !!(s->flags & CODEC_FLAG_LOOP_FILTER);
1016     s->alternate_scan= !!(s->flags & CODEC_FLAG_ALT_SCAN);
1017     s->intra_vlc_format= !!(s->flags2 & CODEC_FLAG2_INTRA_VLC);
1018
1019     if(avctx->rc_max_rate && !avctx->rc_buffer_size){
1020         av_log(avctx, AV_LOG_ERROR, "a vbv buffer size is needed, for encoding with a maximum bitrate\n");
1021         return -1;
1022     }
1023
1024     if(avctx->rc_min_rate && avctx->rc_max_rate != avctx->rc_min_rate){
1025         av_log(avctx, AV_LOG_INFO, "Warning min_rate > 0 but min_rate != max_rate isn't recommended!\n");
1026     }
1027
1028     if(avctx->rc_min_rate && avctx->rc_min_rate > avctx->bit_rate){
1029         av_log(avctx, AV_LOG_INFO, "bitrate below min bitrate\n");
1030         return -1;
1031     }
1032
1033     if(avctx->rc_max_rate && avctx->rc_max_rate < avctx->bit_rate){
1034         av_log(avctx, AV_LOG_INFO, "bitrate above max bitrate\n");
1035         return -1;
1036     }
1037
1038     if(   s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate
1039        && (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO)
1040        && 90000LL * (avctx->rc_buffer_size-1) > s->avctx->rc_max_rate*0xFFFFLL){
1041
1042         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");
1043     }
1044
1045     if((s->flags & CODEC_FLAG_4MV) && s->codec_id != CODEC_ID_MPEG4
1046        && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P && s->codec_id != CODEC_ID_FLV1){
1047         av_log(avctx, AV_LOG_ERROR, "4MV not supported by codec\n");
1048         return -1;
1049     }
1050
1051     if(s->obmc && s->avctx->mb_decision != FF_MB_DECISION_SIMPLE){
1052         av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with simple mb decision\n");
1053         return -1;
1054     }
1055
1056     if(s->obmc && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P){
1057         av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with H263(+)\n");
1058         return -1;
1059     }
1060
1061     if(s->quarter_sample && s->codec_id != CODEC_ID_MPEG4){
1062         av_log(avctx, AV_LOG_ERROR, "qpel not supported by codec\n");
1063         return -1;
1064     }
1065
1066     if(s->data_partitioning && s->codec_id != CODEC_ID_MPEG4){
1067         av_log(avctx, AV_LOG_ERROR, "data partitioning not supported by codec\n");
1068         return -1;
1069     }
1070
1071     if(s->max_b_frames && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO){
1072         av_log(avctx, AV_LOG_ERROR, "b frames not supported by codec\n");
1073         return -1;
1074     }
1075
1076     if((s->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN))
1077        && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG2VIDEO){
1078         av_log(avctx, AV_LOG_ERROR, "interlacing not supported by codec\n");
1079         return -1;
1080     }
1081
1082     if(s->mpeg_quant && s->codec_id != CODEC_ID_MPEG4){ //FIXME mpeg2 uses that too
1083         av_log(avctx, AV_LOG_ERROR, "mpeg2 style quantization not supported by codec\n");
1084         return -1;
1085     }
1086
1087     if((s->flags & CODEC_FLAG_CBP_RD) && !(s->flags & CODEC_FLAG_TRELLIS_QUANT)){
1088         av_log(avctx, AV_LOG_ERROR, "CBP RD needs trellis quant\n");
1089         return -1;
1090     }
1091
1092     if((s->flags & CODEC_FLAG_QP_RD) && s->avctx->mb_decision != FF_MB_DECISION_RD){
1093         av_log(avctx, AV_LOG_ERROR, "QP RD needs mbd=2\n");
1094         return -1;
1095     }
1096
1097     if(s->avctx->scenechange_threshold < 1000000000 && (s->flags & CODEC_FLAG_CLOSED_GOP)){
1098         av_log(avctx, AV_LOG_ERROR, "closed gop with scene change detection arent supported yet\n");
1099         return -1;
1100     }
1101
1102     if((s->flags2 & CODEC_FLAG2_INTRA_VLC) && s->codec_id != CODEC_ID_MPEG2VIDEO){
1103         av_log(avctx, AV_LOG_ERROR, "intra vlc table not supported by codec\n");
1104         return -1;
1105     }
1106
1107     if(s->flags & CODEC_FLAG_LOW_DELAY){
1108         if (s->codec_id != CODEC_ID_MPEG2VIDEO && s->codec_id != CODEC_ID_MPEG1VIDEO){
1109             av_log(avctx, AV_LOG_ERROR, "low delay forcing is only available for mpeg1/2\n");
1110             return -1;
1111         }
1112         if (s->max_b_frames != 0){
1113             av_log(avctx, AV_LOG_ERROR, "b frames cannot be used with low delay\n");
1114             return -1;
1115         }
1116     }
1117
1118     if(s->avctx->thread_count > 1 && s->codec_id != CODEC_ID_MPEG4
1119        && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO
1120        && (s->codec_id != CODEC_ID_H263P || !(s->flags & CODEC_FLAG_H263P_SLICE_STRUCT))){
1121         av_log(avctx, AV_LOG_ERROR, "multi threaded encoding not supported by codec\n");
1122         return -1;
1123     }
1124
1125     if(s->avctx->thread_count > 1)
1126         s->rtp_mode= 1;
1127
1128     if(!avctx->time_base.den || !avctx->time_base.num){
1129         av_log(avctx, AV_LOG_ERROR, "framerate not set\n");
1130         return -1;
1131     }
1132
1133     i= (INT_MAX/2+128)>>8;
1134     if(avctx->me_threshold >= i){
1135         av_log(avctx, AV_LOG_ERROR, "me_threshold too large, max is %d\n", i - 1);
1136         return -1;
1137     }
1138     if(avctx->mb_threshold >= i){
1139         av_log(avctx, AV_LOG_ERROR, "mb_threshold too large, max is %d\n", i - 1);
1140         return -1;
1141     }
1142
1143     if(avctx->b_frame_strategy && (avctx->flags&CODEC_FLAG_PASS2)){
1144         av_log(avctx, AV_LOG_INFO, "notice: b_frame_strategy only affects the first pass\n");
1145         avctx->b_frame_strategy = 0;
1146     }
1147
1148     i= ff_gcd(avctx->time_base.den, avctx->time_base.num);
1149     if(i > 1){
1150         av_log(avctx, AV_LOG_INFO, "removing common factors from framerate\n");
1151         avctx->time_base.den /= i;
1152         avctx->time_base.num /= i;
1153 //        return -1;
1154     }
1155
1156     if(s->codec_id==CODEC_ID_MJPEG){
1157         s->intra_quant_bias= 1<<(QUANT_BIAS_SHIFT-1); //(a + x/2)/x
1158         s->inter_quant_bias= 0;
1159     }else if(s->mpeg_quant || s->codec_id==CODEC_ID_MPEG1VIDEO || s->codec_id==CODEC_ID_MPEG2VIDEO){
1160         s->intra_quant_bias= 3<<(QUANT_BIAS_SHIFT-3); //(a + x*3/8)/x
1161         s->inter_quant_bias= 0;
1162     }else{
1163         s->intra_quant_bias=0;
1164         s->inter_quant_bias=-(1<<(QUANT_BIAS_SHIFT-2)); //(a - x/4)/x
1165     }
1166
1167     if(avctx->intra_quant_bias != FF_DEFAULT_QUANT_BIAS)
1168         s->intra_quant_bias= avctx->intra_quant_bias;
1169     if(avctx->inter_quant_bias != FF_DEFAULT_QUANT_BIAS)
1170         s->inter_quant_bias= avctx->inter_quant_bias;
1171
1172     avcodec_get_chroma_sub_sample(avctx->pix_fmt, &chroma_h_shift, &chroma_v_shift);
1173
1174     if(avctx->codec_id == CODEC_ID_MPEG4 && s->avctx->time_base.den > (1<<16)-1){
1175         av_log(avctx, AV_LOG_ERROR, "timebase not supported by mpeg 4 standard\n");
1176         return -1;
1177     }
1178     s->time_increment_bits = av_log2(s->avctx->time_base.den - 1) + 1;
1179
1180     switch(avctx->codec->id) {
1181     case CODEC_ID_MPEG1VIDEO:
1182         s->out_format = FMT_MPEG1;
1183         s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY);
1184         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
1185         break;
1186     case CODEC_ID_MPEG2VIDEO:
1187         s->out_format = FMT_MPEG1;
1188         s->low_delay= !!(s->flags & CODEC_FLAG_LOW_DELAY);
1189         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
1190         s->rtp_mode= 1;
1191         break;
1192     case CODEC_ID_LJPEG:
1193     case CODEC_ID_JPEGLS:
1194     case CODEC_ID_MJPEG:
1195         s->out_format = FMT_MJPEG;
1196         s->intra_only = 1; /* force intra only for jpeg */
1197         s->mjpeg_write_tables = avctx->codec->id != CODEC_ID_JPEGLS;
1198         s->mjpeg_data_only_frames = 0; /* write all the needed headers */
1199         s->mjpeg_vsample[0] = 2;
1200         s->mjpeg_vsample[1] = 2>>chroma_v_shift;
1201         s->mjpeg_vsample[2] = 2>>chroma_v_shift;
1202         s->mjpeg_hsample[0] = 2;
1203         s->mjpeg_hsample[1] = 2>>chroma_h_shift;
1204         s->mjpeg_hsample[2] = 2>>chroma_h_shift;
1205         if (mjpeg_init(s) < 0)
1206             return -1;
1207         avctx->delay=0;
1208         s->low_delay=1;
1209         break;
1210 #ifdef CONFIG_H261_ENCODER
1211     case CODEC_ID_H261:
1212         if (ff_h261_get_picture_format(s->width, s->height) < 0) {
1213             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);
1214             return -1;
1215         }
1216         s->out_format = FMT_H261;
1217         avctx->delay=0;
1218         s->low_delay=1;
1219         break;
1220 #endif
1221     case CODEC_ID_H263:
1222         if (h263_get_picture_format(s->width, s->height) == 7) {
1223             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);
1224             return -1;
1225         }
1226         s->out_format = FMT_H263;
1227         s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
1228         avctx->delay=0;
1229         s->low_delay=1;
1230         break;
1231     case CODEC_ID_H263P:
1232         s->out_format = FMT_H263;
1233         s->h263_plus = 1;
1234         /* Fx */
1235         s->umvplus = (avctx->flags & CODEC_FLAG_H263P_UMV) ? 1:0;
1236         s->h263_aic= (avctx->flags & CODEC_FLAG_H263P_AIC) ? 1:0;
1237         s->modified_quant= s->h263_aic;
1238         s->alt_inter_vlc= (avctx->flags & CODEC_FLAG_H263P_AIV) ? 1:0;
1239         s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
1240         s->loop_filter= (avctx->flags & CODEC_FLAG_LOOP_FILTER) ? 1:0;
1241         s->unrestricted_mv= s->obmc || s->loop_filter || s->umvplus;
1242         s->h263_slice_structured= (s->flags & CODEC_FLAG_H263P_SLICE_STRUCT) ? 1:0;
1243
1244         /* /Fx */
1245         /* These are just to be sure */
1246         avctx->delay=0;
1247         s->low_delay=1;
1248         break;
1249     case CODEC_ID_FLV1:
1250         s->out_format = FMT_H263;
1251         s->h263_flv = 2; /* format = 1; 11-bit codes */
1252         s->unrestricted_mv = 1;
1253         s->rtp_mode=0; /* don't allow GOB */
1254         avctx->delay=0;
1255         s->low_delay=1;
1256         break;
1257     case CODEC_ID_RV10:
1258         s->out_format = FMT_H263;
1259         avctx->delay=0;
1260         s->low_delay=1;
1261         break;
1262     case CODEC_ID_RV20:
1263         s->out_format = FMT_H263;
1264         avctx->delay=0;
1265         s->low_delay=1;
1266         s->modified_quant=1;
1267         s->h263_aic=1;
1268         s->h263_plus=1;
1269         s->loop_filter=1;
1270         s->unrestricted_mv= s->obmc || s->loop_filter || s->umvplus;
1271         break;
1272     case CODEC_ID_MPEG4:
1273         s->out_format = FMT_H263;
1274         s->h263_pred = 1;
1275         s->unrestricted_mv = 1;
1276         s->low_delay= s->max_b_frames ? 0 : 1;
1277         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
1278         break;
1279     case CODEC_ID_MSMPEG4V1:
1280         s->out_format = FMT_H263;
1281         s->h263_msmpeg4 = 1;
1282         s->h263_pred = 1;
1283         s->unrestricted_mv = 1;
1284         s->msmpeg4_version= 1;
1285         avctx->delay=0;
1286         s->low_delay=1;
1287         break;
1288     case CODEC_ID_MSMPEG4V2:
1289         s->out_format = FMT_H263;
1290         s->h263_msmpeg4 = 1;
1291         s->h263_pred = 1;
1292         s->unrestricted_mv = 1;
1293         s->msmpeg4_version= 2;
1294         avctx->delay=0;
1295         s->low_delay=1;
1296         break;
1297     case CODEC_ID_MSMPEG4V3:
1298         s->out_format = FMT_H263;
1299         s->h263_msmpeg4 = 1;
1300         s->h263_pred = 1;
1301         s->unrestricted_mv = 1;
1302         s->msmpeg4_version= 3;
1303         s->flipflop_rounding=1;
1304         avctx->delay=0;
1305         s->low_delay=1;
1306         break;
1307     case CODEC_ID_WMV1:
1308         s->out_format = FMT_H263;
1309         s->h263_msmpeg4 = 1;
1310         s->h263_pred = 1;
1311         s->unrestricted_mv = 1;
1312         s->msmpeg4_version= 4;
1313         s->flipflop_rounding=1;
1314         avctx->delay=0;
1315         s->low_delay=1;
1316         break;
1317     case CODEC_ID_WMV2:
1318         s->out_format = FMT_H263;
1319         s->h263_msmpeg4 = 1;
1320         s->h263_pred = 1;
1321         s->unrestricted_mv = 1;
1322         s->msmpeg4_version= 5;
1323         s->flipflop_rounding=1;
1324         avctx->delay=0;
1325         s->low_delay=1;
1326         break;
1327     default:
1328         return -1;
1329     }
1330
1331     avctx->has_b_frames= !s->low_delay;
1332
1333     s->encoding = 1;
1334
1335     /* init */
1336     if (MPV_common_init(s) < 0)
1337         return -1;
1338
1339     if(s->modified_quant)
1340         s->chroma_qscale_table= ff_h263_chroma_qscale_table;
1341     s->progressive_frame=
1342     s->progressive_sequence= !(avctx->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN));
1343     s->quant_precision=5;
1344
1345     ff_set_cmp(&s->dsp, s->dsp.ildct_cmp, s->avctx->ildct_cmp);
1346     ff_set_cmp(&s->dsp, s->dsp.frame_skip_cmp, s->avctx->frame_skip_cmp);
1347
1348 #ifdef CONFIG_H261_ENCODER
1349     if (s->out_format == FMT_H261)
1350         ff_h261_encode_init(s);
1351 #endif
1352     if (s->out_format == FMT_H263)
1353         h263_encode_init(s);
1354     if(s->msmpeg4_version)
1355         ff_msmpeg4_encode_init(s);
1356     if (s->out_format == FMT_MPEG1)
1357         ff_mpeg1_encode_init(s);
1358
1359     /* init q matrix */
1360     for(i=0;i<64;i++) {
1361         int j= s->dsp.idct_permutation[i];
1362         if(s->codec_id==CODEC_ID_MPEG4 && s->mpeg_quant){
1363             s->intra_matrix[j] = ff_mpeg4_default_intra_matrix[i];
1364             s->inter_matrix[j] = ff_mpeg4_default_non_intra_matrix[i];
1365         }else if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
1366             s->intra_matrix[j] =
1367             s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
1368         }else
1369         { /* mpeg1/2 */
1370             s->intra_matrix[j] = ff_mpeg1_default_intra_matrix[i];
1371             s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
1372         }
1373         if(s->avctx->intra_matrix)
1374             s->intra_matrix[j] = s->avctx->intra_matrix[i];
1375         if(s->avctx->inter_matrix)
1376             s->inter_matrix[j] = s->avctx->inter_matrix[i];
1377     }
1378
1379     /* precompute matrix */
1380     /* for mjpeg, we do include qscale in the matrix */
1381     if (s->out_format != FMT_MJPEG) {
1382         convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
1383                        s->intra_matrix, s->intra_quant_bias, avctx->qmin, 31, 1);
1384         convert_matrix(&s->dsp, s->q_inter_matrix, s->q_inter_matrix16,
1385                        s->inter_matrix, s->inter_quant_bias, avctx->qmin, 31, 0);
1386     }
1387
1388     if(ff_rate_control_init(s) < 0)
1389         return -1;
1390
1391     return 0;
1392 }
1393
1394 int MPV_encode_end(AVCodecContext *avctx)
1395 {
1396     MpegEncContext *s = avctx->priv_data;
1397
1398     ff_rate_control_uninit(s);
1399
1400     MPV_common_end(s);
1401     if (s->out_format == FMT_MJPEG)
1402         mjpeg_close(s);
1403
1404     av_freep(&avctx->extradata);
1405
1406     return 0;
1407 }
1408
1409 #endif //CONFIG_ENCODERS
1410
1411 void init_rl(RLTable *rl, int use_static)
1412 {
1413     int8_t max_level[MAX_RUN+1], max_run[MAX_LEVEL+1];
1414     uint8_t index_run[MAX_RUN+1];
1415     int last, run, level, start, end, i;
1416
1417     /* If table is static, we can quit if rl->max_level[0] is not NULL */
1418     if(use_static && rl->max_level[0])
1419         return;
1420
1421     /* compute max_level[], max_run[] and index_run[] */
1422     for(last=0;last<2;last++) {
1423         if (last == 0) {
1424             start = 0;
1425             end = rl->last;
1426         } else {
1427             start = rl->last;
1428             end = rl->n;
1429         }
1430
1431         memset(max_level, 0, MAX_RUN + 1);
1432         memset(max_run, 0, MAX_LEVEL + 1);
1433         memset(index_run, rl->n, MAX_RUN + 1);
1434         for(i=start;i<end;i++) {
1435             run = rl->table_run[i];
1436             level = rl->table_level[i];
1437             if (index_run[run] == rl->n)
1438                 index_run[run] = i;
1439             if (level > max_level[run])
1440                 max_level[run] = level;
1441             if (run > max_run[level])
1442                 max_run[level] = run;
1443         }
1444         if(use_static)
1445             rl->max_level[last] = av_mallocz_static(MAX_RUN + 1);
1446         else
1447             rl->max_level[last] = av_malloc(MAX_RUN + 1);
1448         memcpy(rl->max_level[last], max_level, MAX_RUN + 1);
1449         if(use_static)
1450             rl->max_run[last] = av_mallocz_static(MAX_LEVEL + 1);
1451         else
1452             rl->max_run[last] = av_malloc(MAX_LEVEL + 1);
1453         memcpy(rl->max_run[last], max_run, MAX_LEVEL + 1);
1454         if(use_static)
1455             rl->index_run[last] = av_mallocz_static(MAX_RUN + 1);
1456         else
1457             rl->index_run[last] = av_malloc(MAX_RUN + 1);
1458         memcpy(rl->index_run[last], index_run, MAX_RUN + 1);
1459     }
1460 }
1461
1462 /* draw the edges of width 'w' of an image of size width, height */
1463 //FIXME check that this is ok for mpeg4 interlaced
1464 static void draw_edges_c(uint8_t *buf, int wrap, int width, int height, int w)
1465 {
1466     uint8_t *ptr, *last_line;
1467     int i;
1468
1469     last_line = buf + (height - 1) * wrap;
1470     for(i=0;i<w;i++) {
1471         /* top and bottom */
1472         memcpy(buf - (i + 1) * wrap, buf, width);
1473         memcpy(last_line + (i + 1) * wrap, last_line, width);
1474     }
1475     /* left and right */
1476     ptr = buf;
1477     for(i=0;i<height;i++) {
1478         memset(ptr - w, ptr[0], w);
1479         memset(ptr + width, ptr[width-1], w);
1480         ptr += wrap;
1481     }
1482     /* corners */
1483     for(i=0;i<w;i++) {
1484         memset(buf - (i + 1) * wrap - w, buf[0], w); /* top left */
1485         memset(buf - (i + 1) * wrap + width, buf[width-1], w); /* top right */
1486         memset(last_line + (i + 1) * wrap - w, last_line[0], w); /* top left */
1487         memset(last_line + (i + 1) * wrap + width, last_line[width-1], w); /* top right */
1488     }
1489 }
1490
1491 int ff_find_unused_picture(MpegEncContext *s, int shared){
1492     int i;
1493
1494     if(shared){
1495         for(i=0; i<MAX_PICTURE_COUNT; i++){
1496             if(s->picture[i].data[0]==NULL && s->picture[i].type==0) return i;
1497         }
1498     }else{
1499         for(i=0; i<MAX_PICTURE_COUNT; i++){
1500             if(s->picture[i].data[0]==NULL && s->picture[i].type!=0) return i; //FIXME
1501         }
1502         for(i=0; i<MAX_PICTURE_COUNT; i++){
1503             if(s->picture[i].data[0]==NULL) return i;
1504         }
1505     }
1506
1507     assert(0);
1508     return -1;
1509 }
1510
1511 static void update_noise_reduction(MpegEncContext *s){
1512     int intra, i;
1513
1514     for(intra=0; intra<2; intra++){
1515         if(s->dct_count[intra] > (1<<16)){
1516             for(i=0; i<64; i++){
1517                 s->dct_error_sum[intra][i] >>=1;
1518             }
1519             s->dct_count[intra] >>= 1;
1520         }
1521
1522         for(i=0; i<64; i++){
1523             s->dct_offset[intra][i]= (s->avctx->noise_reduction * s->dct_count[intra] + s->dct_error_sum[intra][i]/2) / (s->dct_error_sum[intra][i]+1);
1524         }
1525     }
1526 }
1527
1528 /**
1529  * generic function for encode/decode called after coding/decoding the header and before a frame is coded/decoded
1530  */
1531 int MPV_frame_start(MpegEncContext *s, AVCodecContext *avctx)
1532 {
1533     int i;
1534     AVFrame *pic;
1535     s->mb_skipped = 0;
1536
1537     assert(s->last_picture_ptr==NULL || s->out_format != FMT_H264 || s->codec_id == CODEC_ID_SVQ3);
1538
1539     /* mark&release old frames */
1540     if (s->pict_type != B_TYPE && s->last_picture_ptr && s->last_picture_ptr != s->next_picture_ptr && s->last_picture_ptr->data[0]) {
1541         avctx->release_buffer(avctx, (AVFrame*)s->last_picture_ptr);
1542
1543         /* release forgotten pictures */
1544         /* if(mpeg124/h263) */
1545         if(!s->encoding){
1546             for(i=0; i<MAX_PICTURE_COUNT; i++){
1547                 if(s->picture[i].data[0] && &s->picture[i] != s->next_picture_ptr && s->picture[i].reference){
1548                     av_log(avctx, AV_LOG_ERROR, "releasing zombie picture\n");
1549                     avctx->release_buffer(avctx, (AVFrame*)&s->picture[i]);
1550                 }
1551             }
1552         }
1553     }
1554 alloc:
1555     if(!s->encoding){
1556         /* release non reference frames */
1557         for(i=0; i<MAX_PICTURE_COUNT; i++){
1558             if(s->picture[i].data[0] && !s->picture[i].reference /*&& s->picture[i].type!=FF_BUFFER_TYPE_SHARED*/){
1559                 s->avctx->release_buffer(s->avctx, (AVFrame*)&s->picture[i]);
1560             }
1561         }
1562
1563         if(s->current_picture_ptr && s->current_picture_ptr->data[0]==NULL)
1564             pic= (AVFrame*)s->current_picture_ptr; //we allready have a unused image (maybe it was set before reading the header)
1565         else{
1566             i= ff_find_unused_picture(s, 0);
1567             pic= (AVFrame*)&s->picture[i];
1568         }
1569
1570         pic->reference= (s->pict_type != B_TYPE || s->codec_id == CODEC_ID_H264)
1571                         && !s->dropable ? 3 : 0;
1572
1573         pic->coded_picture_number= s->coded_picture_number++;
1574
1575         if( alloc_picture(s, (Picture*)pic, 0) < 0)
1576             return -1;
1577
1578         s->current_picture_ptr= (Picture*)pic;
1579         s->current_picture_ptr->top_field_first= s->top_field_first; //FIXME use only the vars from current_pic
1580         s->current_picture_ptr->interlaced_frame= !s->progressive_frame && !s->progressive_sequence;
1581     }
1582
1583     s->current_picture_ptr->pict_type= s->pict_type;
1584 //    if(s->flags && CODEC_FLAG_QSCALE)
1585   //      s->current_picture_ptr->quality= s->new_picture_ptr->quality;
1586     s->current_picture_ptr->key_frame= s->pict_type == I_TYPE;
1587
1588     copy_picture(&s->current_picture, s->current_picture_ptr);
1589
1590   if(s->out_format != FMT_H264 || s->codec_id == CODEC_ID_SVQ3){
1591     if (s->pict_type != B_TYPE) {
1592         s->last_picture_ptr= s->next_picture_ptr;
1593         if(!s->dropable)
1594             s->next_picture_ptr= s->current_picture_ptr;
1595     }
1596 /*    av_log(s->avctx, AV_LOG_DEBUG, "L%p N%p C%p L%p N%p C%p type:%d drop:%d\n", s->last_picture_ptr, s->next_picture_ptr,s->current_picture_ptr,
1597         s->last_picture_ptr    ? s->last_picture_ptr->data[0] : NULL,
1598         s->next_picture_ptr    ? s->next_picture_ptr->data[0] : NULL,
1599         s->current_picture_ptr ? s->current_picture_ptr->data[0] : NULL,
1600         s->pict_type, s->dropable);*/
1601
1602     if(s->last_picture_ptr) copy_picture(&s->last_picture, s->last_picture_ptr);
1603     if(s->next_picture_ptr) copy_picture(&s->next_picture, s->next_picture_ptr);
1604
1605     if(s->pict_type != I_TYPE && (s->last_picture_ptr==NULL || s->last_picture_ptr->data[0]==NULL)){
1606         av_log(avctx, AV_LOG_ERROR, "warning: first frame is no keyframe\n");
1607         assert(s->pict_type != B_TYPE); //these should have been dropped if we don't have a reference
1608         goto alloc;
1609     }
1610
1611     assert(s->pict_type == I_TYPE || (s->last_picture_ptr && s->last_picture_ptr->data[0]));
1612
1613     if(s->picture_structure!=PICT_FRAME){
1614         int i;
1615         for(i=0; i<4; i++){
1616             if(s->picture_structure == PICT_BOTTOM_FIELD){
1617                  s->current_picture.data[i] += s->current_picture.linesize[i];
1618             }
1619             s->current_picture.linesize[i] *= 2;
1620             s->last_picture.linesize[i] *=2;
1621             s->next_picture.linesize[i] *=2;
1622         }
1623     }
1624   }
1625
1626     s->hurry_up= s->avctx->hurry_up;
1627     s->error_resilience= avctx->error_resilience;
1628
1629     /* set dequantizer, we can't do it during init as it might change for mpeg4
1630        and we can't do it in the header decode as init isnt called for mpeg4 there yet */
1631     if(s->mpeg_quant || s->codec_id == CODEC_ID_MPEG2VIDEO){
1632         s->dct_unquantize_intra = s->dct_unquantize_mpeg2_intra;
1633         s->dct_unquantize_inter = s->dct_unquantize_mpeg2_inter;
1634     }else if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
1635         s->dct_unquantize_intra = s->dct_unquantize_h263_intra;
1636         s->dct_unquantize_inter = s->dct_unquantize_h263_inter;
1637     }else{
1638         s->dct_unquantize_intra = s->dct_unquantize_mpeg1_intra;
1639         s->dct_unquantize_inter = s->dct_unquantize_mpeg1_inter;
1640     }
1641
1642     if(s->dct_error_sum){
1643         assert(s->avctx->noise_reduction && s->encoding);
1644
1645         update_noise_reduction(s);
1646     }
1647
1648 #ifdef HAVE_XVMC
1649     if(s->avctx->xvmc_acceleration)
1650         return XVMC_field_start(s, avctx);
1651 #endif
1652     return 0;
1653 }
1654
1655 /* generic function for encode/decode called after a frame has been coded/decoded */
1656 void MPV_frame_end(MpegEncContext *s)
1657 {
1658     int i;
1659     /* draw edge for correct motion prediction if outside */
1660 #ifdef HAVE_XVMC
1661 //just to make sure that all data is rendered.
1662     if(s->avctx->xvmc_acceleration){
1663         XVMC_field_end(s);
1664     }else
1665 #endif
1666     if(s->unrestricted_mv && s->current_picture.reference && !s->intra_only && !(s->flags&CODEC_FLAG_EMU_EDGE)) {
1667             draw_edges(s->current_picture.data[0], s->linesize  , s->h_edge_pos   , s->v_edge_pos   , EDGE_WIDTH  );
1668             draw_edges(s->current_picture.data[1], s->uvlinesize, s->h_edge_pos>>1, s->v_edge_pos>>1, EDGE_WIDTH/2);
1669             draw_edges(s->current_picture.data[2], s->uvlinesize, s->h_edge_pos>>1, s->v_edge_pos>>1, EDGE_WIDTH/2);
1670     }
1671     emms_c();
1672
1673     s->last_pict_type    = s->pict_type;
1674     s->last_lambda_for[s->pict_type]= s->current_picture_ptr->quality;
1675     if(s->pict_type!=B_TYPE){
1676         s->last_non_b_pict_type= s->pict_type;
1677     }
1678 #if 0
1679         /* copy back current_picture variables */
1680     for(i=0; i<MAX_PICTURE_COUNT; i++){
1681         if(s->picture[i].data[0] == s->current_picture.data[0]){
1682             s->picture[i]= s->current_picture;
1683             break;
1684         }
1685     }
1686     assert(i<MAX_PICTURE_COUNT);
1687 #endif
1688
1689     if(s->encoding){
1690         /* release non-reference frames */
1691         for(i=0; i<MAX_PICTURE_COUNT; i++){
1692             if(s->picture[i].data[0] && !s->picture[i].reference /*&& s->picture[i].type!=FF_BUFFER_TYPE_SHARED*/){
1693                 s->avctx->release_buffer(s->avctx, (AVFrame*)&s->picture[i]);
1694             }
1695         }
1696     }
1697     // clear copies, to avoid confusion
1698 #if 0
1699     memset(&s->last_picture, 0, sizeof(Picture));
1700     memset(&s->next_picture, 0, sizeof(Picture));
1701     memset(&s->current_picture, 0, sizeof(Picture));
1702 #endif
1703     s->avctx->coded_frame= (AVFrame*)s->current_picture_ptr;
1704 }
1705
1706 /**
1707  * draws an line from (ex, ey) -> (sx, sy).
1708  * @param w width of the image
1709  * @param h height of the image
1710  * @param stride stride/linesize of the image
1711  * @param color color of the arrow
1712  */
1713 static void draw_line(uint8_t *buf, int sx, int sy, int ex, int ey, int w, int h, int stride, int color){
1714     int x, y, fr, f;
1715
1716     sx= clip(sx, 0, w-1);
1717     sy= clip(sy, 0, h-1);
1718     ex= clip(ex, 0, w-1);
1719     ey= clip(ey, 0, h-1);
1720
1721     buf[sy*stride + sx]+= color;
1722
1723     if(FFABS(ex - sx) > FFABS(ey - sy)){
1724         if(sx > ex){
1725             FFSWAP(int, sx, ex);
1726             FFSWAP(int, sy, ey);
1727         }
1728         buf+= sx + sy*stride;
1729         ex-= sx;
1730         f= ((ey-sy)<<16)/ex;
1731         for(x= 0; x <= ex; x++){
1732             y = (x*f)>>16;
1733             fr= (x*f)&0xFFFF;
1734             buf[ y   *stride + x]+= (color*(0x10000-fr))>>16;
1735             buf[(y+1)*stride + x]+= (color*         fr )>>16;
1736         }
1737     }else{
1738         if(sy > ey){
1739             FFSWAP(int, sx, ex);
1740             FFSWAP(int, sy, ey);
1741         }
1742         buf+= sx + sy*stride;
1743         ey-= sy;
1744         if(ey) f= ((ex-sx)<<16)/ey;
1745         else   f= 0;
1746         for(y= 0; y <= ey; y++){
1747             x = (y*f)>>16;
1748             fr= (y*f)&0xFFFF;
1749             buf[y*stride + x  ]+= (color*(0x10000-fr))>>16;;
1750             buf[y*stride + x+1]+= (color*         fr )>>16;;
1751         }
1752     }
1753 }
1754
1755 /**
1756  * draws an arrow from (ex, ey) -> (sx, sy).
1757  * @param w width of the image
1758  * @param h height of the image
1759  * @param stride stride/linesize of the image
1760  * @param color color of the arrow
1761  */
1762 static void draw_arrow(uint8_t *buf, int sx, int sy, int ex, int ey, int w, int h, int stride, int color){
1763     int dx,dy;
1764
1765     sx= clip(sx, -100, w+100);
1766     sy= clip(sy, -100, h+100);
1767     ex= clip(ex, -100, w+100);
1768     ey= clip(ey, -100, h+100);
1769
1770     dx= ex - sx;
1771     dy= ey - sy;
1772
1773     if(dx*dx + dy*dy > 3*3){
1774         int rx=  dx + dy;
1775         int ry= -dx + dy;
1776         int length= ff_sqrt((rx*rx + ry*ry)<<8);
1777
1778         //FIXME subpixel accuracy
1779         rx= ROUNDED_DIV(rx*3<<4, length);
1780         ry= ROUNDED_DIV(ry*3<<4, length);
1781
1782         draw_line(buf, sx, sy, sx + rx, sy + ry, w, h, stride, color);
1783         draw_line(buf, sx, sy, sx - ry, sy + rx, w, h, stride, color);
1784     }
1785     draw_line(buf, sx, sy, ex, ey, w, h, stride, color);
1786 }
1787
1788 /**
1789  * prints debuging info for the given picture.
1790  */
1791 void ff_print_debug_info(MpegEncContext *s, AVFrame *pict){
1792
1793     if(!pict || !pict->mb_type) return;
1794
1795     if(s->avctx->debug&(FF_DEBUG_SKIP | FF_DEBUG_QP | FF_DEBUG_MB_TYPE)){
1796         int x,y;
1797
1798         av_log(s->avctx,AV_LOG_DEBUG,"New frame, type: ");
1799         switch (pict->pict_type) {
1800             case FF_I_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"I\n"); break;
1801             case FF_P_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"P\n"); break;
1802             case FF_B_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"B\n"); break;
1803             case FF_S_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"S\n"); break;
1804             case FF_SI_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"SI\n"); break;
1805             case FF_SP_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"SP\n"); break;
1806         }
1807         for(y=0; y<s->mb_height; y++){
1808             for(x=0; x<s->mb_width; x++){
1809                 if(s->avctx->debug&FF_DEBUG_SKIP){
1810                     int count= s->mbskip_table[x + y*s->mb_stride];
1811                     if(count>9) count=9;
1812                     av_log(s->avctx, AV_LOG_DEBUG, "%1d", count);
1813                 }
1814                 if(s->avctx->debug&FF_DEBUG_QP){
1815                     av_log(s->avctx, AV_LOG_DEBUG, "%2d", pict->qscale_table[x + y*s->mb_stride]);
1816                 }
1817                 if(s->avctx->debug&FF_DEBUG_MB_TYPE){
1818                     int mb_type= pict->mb_type[x + y*s->mb_stride];
1819                     //Type & MV direction
1820                     if(IS_PCM(mb_type))
1821                         av_log(s->avctx, AV_LOG_DEBUG, "P");
1822                     else if(IS_INTRA(mb_type) && IS_ACPRED(mb_type))
1823                         av_log(s->avctx, AV_LOG_DEBUG, "A");
1824                     else if(IS_INTRA4x4(mb_type))
1825                         av_log(s->avctx, AV_LOG_DEBUG, "i");
1826                     else if(IS_INTRA16x16(mb_type))
1827                         av_log(s->avctx, AV_LOG_DEBUG, "I");
1828                     else if(IS_DIRECT(mb_type) && IS_SKIP(mb_type))
1829                         av_log(s->avctx, AV_LOG_DEBUG, "d");
1830                     else if(IS_DIRECT(mb_type))
1831                         av_log(s->avctx, AV_LOG_DEBUG, "D");
1832                     else if(IS_GMC(mb_type) && IS_SKIP(mb_type))
1833                         av_log(s->avctx, AV_LOG_DEBUG, "g");
1834                     else if(IS_GMC(mb_type))
1835                         av_log(s->avctx, AV_LOG_DEBUG, "G");
1836                     else if(IS_SKIP(mb_type))
1837                         av_log(s->avctx, AV_LOG_DEBUG, "S");
1838                     else if(!USES_LIST(mb_type, 1))
1839                         av_log(s->avctx, AV_LOG_DEBUG, ">");
1840                     else if(!USES_LIST(mb_type, 0))
1841                         av_log(s->avctx, AV_LOG_DEBUG, "<");
1842                     else{
1843                         assert(USES_LIST(mb_type, 0) && USES_LIST(mb_type, 1));
1844                         av_log(s->avctx, AV_LOG_DEBUG, "X");
1845                     }
1846
1847                     //segmentation
1848                     if(IS_8X8(mb_type))
1849                         av_log(s->avctx, AV_LOG_DEBUG, "+");
1850                     else if(IS_16X8(mb_type))
1851                         av_log(s->avctx, AV_LOG_DEBUG, "-");
1852                     else if(IS_8X16(mb_type))
1853                         av_log(s->avctx, AV_LOG_DEBUG, "|");
1854                     else if(IS_INTRA(mb_type) || IS_16X16(mb_type))
1855                         av_log(s->avctx, AV_LOG_DEBUG, " ");
1856                     else
1857                         av_log(s->avctx, AV_LOG_DEBUG, "?");
1858
1859
1860                     if(IS_INTERLACED(mb_type) && s->codec_id == CODEC_ID_H264)
1861                         av_log(s->avctx, AV_LOG_DEBUG, "=");
1862                     else
1863                         av_log(s->avctx, AV_LOG_DEBUG, " ");
1864                 }
1865 //                av_log(s->avctx, AV_LOG_DEBUG, " ");
1866             }
1867             av_log(s->avctx, AV_LOG_DEBUG, "\n");
1868         }
1869     }
1870
1871     if((s->avctx->debug&(FF_DEBUG_VIS_QP|FF_DEBUG_VIS_MB_TYPE)) || (s->avctx->debug_mv)){
1872         const int shift= 1 + s->quarter_sample;
1873         int mb_y;
1874         uint8_t *ptr;
1875         int i;
1876         int h_chroma_shift, v_chroma_shift;
1877         const int width = s->avctx->width;
1878         const int height= s->avctx->height;
1879         const int mv_sample_log2= 4 - pict->motion_subsample_log2;
1880         const int mv_stride= (s->mb_width << mv_sample_log2) + (s->codec_id == CODEC_ID_H264 ? 0 : 1);
1881         s->low_delay=0; //needed to see the vectors without trashing the buffers
1882
1883         avcodec_get_chroma_sub_sample(s->avctx->pix_fmt, &h_chroma_shift, &v_chroma_shift);
1884         for(i=0; i<3; i++){
1885             memcpy(s->visualization_buffer[i], pict->data[i], (i==0) ? pict->linesize[i]*height:pict->linesize[i]*height >> v_chroma_shift);
1886             pict->data[i]= s->visualization_buffer[i];
1887         }
1888         pict->type= FF_BUFFER_TYPE_COPY;
1889         ptr= pict->data[0];
1890
1891         for(mb_y=0; mb_y<s->mb_height; mb_y++){
1892             int mb_x;
1893             for(mb_x=0; mb_x<s->mb_width; mb_x++){
1894                 const int mb_index= mb_x + mb_y*s->mb_stride;
1895                 if((s->avctx->debug_mv) && pict->motion_val){
1896                   int type;
1897                   for(type=0; type<3; type++){
1898                     int direction = 0;
1899                     switch (type) {
1900                       case 0: if ((!(s->avctx->debug_mv&FF_DEBUG_VIS_MV_P_FOR)) || (pict->pict_type!=FF_P_TYPE))
1901                                 continue;
1902                               direction = 0;
1903                               break;
1904                       case 1: if ((!(s->avctx->debug_mv&FF_DEBUG_VIS_MV_B_FOR)) || (pict->pict_type!=FF_B_TYPE))
1905                                 continue;
1906                               direction = 0;
1907                               break;
1908                       case 2: if ((!(s->avctx->debug_mv&FF_DEBUG_VIS_MV_B_BACK)) || (pict->pict_type!=FF_B_TYPE))
1909                                 continue;
1910                               direction = 1;
1911                               break;
1912                     }
1913                     if(!USES_LIST(pict->mb_type[mb_index], direction))
1914                         continue;
1915
1916                     if(IS_8X8(pict->mb_type[mb_index])){
1917                       int i;
1918                       for(i=0; i<4; i++){
1919                         int sx= mb_x*16 + 4 + 8*(i&1);
1920                         int sy= mb_y*16 + 4 + 8*(i>>1);
1921                         int xy= (mb_x*2 + (i&1) + (mb_y*2 + (i>>1))*mv_stride) << (mv_sample_log2-1);
1922                         int mx= (pict->motion_val[direction][xy][0]>>shift) + sx;
1923                         int my= (pict->motion_val[direction][xy][1]>>shift) + sy;
1924                         draw_arrow(ptr, sx, sy, mx, my, width, height, s->linesize, 100);
1925                       }
1926                     }else if(IS_16X8(pict->mb_type[mb_index])){
1927                       int i;
1928                       for(i=0; i<2; i++){
1929                         int sx=mb_x*16 + 8;
1930                         int sy=mb_y*16 + 4 + 8*i;
1931                         int xy= (mb_x*2 + (mb_y*2 + i)*mv_stride) << (mv_sample_log2-1);
1932                         int mx=(pict->motion_val[direction][xy][0]>>shift);
1933                         int my=(pict->motion_val[direction][xy][1]>>shift);
1934
1935                         if(IS_INTERLACED(pict->mb_type[mb_index]))
1936                             my*=2;
1937
1938                         draw_arrow(ptr, sx, sy, mx+sx, my+sy, width, height, s->linesize, 100);
1939                       }
1940                     }else if(IS_8X16(pict->mb_type[mb_index])){
1941                       int i;
1942                       for(i=0; i<2; i++){
1943                         int sx=mb_x*16 + 4 + 8*i;
1944                         int sy=mb_y*16 + 8;
1945                         int xy= (mb_x*2 + i + mb_y*2*mv_stride) << (mv_sample_log2-1);
1946                         int mx=(pict->motion_val[direction][xy][0]>>shift);
1947                         int my=(pict->motion_val[direction][xy][1]>>shift);
1948
1949                         if(IS_INTERLACED(pict->mb_type[mb_index]))
1950                             my*=2;
1951
1952                         draw_arrow(ptr, sx, sy, mx+sx, my+sy, width, height, s->linesize, 100);
1953                       }
1954                     }else{
1955                       int sx= mb_x*16 + 8;
1956                       int sy= mb_y*16 + 8;
1957                       int xy= (mb_x + mb_y*mv_stride) << mv_sample_log2;
1958                       int mx= (pict->motion_val[direction][xy][0]>>shift) + sx;
1959                       int my= (pict->motion_val[direction][xy][1]>>shift) + sy;
1960                       draw_arrow(ptr, sx, sy, mx, my, width, height, s->linesize, 100);
1961                     }
1962                   }
1963                 }
1964                 if((s->avctx->debug&FF_DEBUG_VIS_QP) && pict->motion_val){
1965                     uint64_t c= (pict->qscale_table[mb_index]*128/31) * 0x0101010101010101ULL;
1966                     int y;
1967                     for(y=0; y<8; y++){
1968                         *(uint64_t*)(pict->data[1] + 8*mb_x + (8*mb_y + y)*pict->linesize[1])= c;
1969                         *(uint64_t*)(pict->data[2] + 8*mb_x + (8*mb_y + y)*pict->linesize[2])= c;
1970                     }
1971                 }
1972                 if((s->avctx->debug&FF_DEBUG_VIS_MB_TYPE) && pict->motion_val){
1973                     int mb_type= pict->mb_type[mb_index];
1974                     uint64_t u,v;
1975                     int y;
1976 #define COLOR(theta, r)\
1977 u= (int)(128 + r*cos(theta*3.141592/180));\
1978 v= (int)(128 + r*sin(theta*3.141592/180));
1979
1980
1981                     u=v=128;
1982                     if(IS_PCM(mb_type)){
1983                         COLOR(120,48)
1984                     }else if((IS_INTRA(mb_type) && IS_ACPRED(mb_type)) || IS_INTRA16x16(mb_type)){
1985                         COLOR(30,48)
1986                     }else if(IS_INTRA4x4(mb_type)){
1987                         COLOR(90,48)
1988                     }else if(IS_DIRECT(mb_type) && IS_SKIP(mb_type)){
1989 //                        COLOR(120,48)
1990                     }else if(IS_DIRECT(mb_type)){
1991                         COLOR(150,48)
1992                     }else if(IS_GMC(mb_type) && IS_SKIP(mb_type)){
1993                         COLOR(170,48)
1994                     }else if(IS_GMC(mb_type)){
1995                         COLOR(190,48)
1996                     }else if(IS_SKIP(mb_type)){
1997 //                        COLOR(180,48)
1998                     }else if(!USES_LIST(mb_type, 1)){
1999                         COLOR(240,48)
2000                     }else if(!USES_LIST(mb_type, 0)){
2001                         COLOR(0,48)
2002                     }else{
2003                         assert(USES_LIST(mb_type, 0) && USES_LIST(mb_type, 1));
2004                         COLOR(300,48)
2005                     }
2006
2007                     u*= 0x0101010101010101ULL;
2008                     v*= 0x0101010101010101ULL;
2009                     for(y=0; y<8; y++){
2010                         *(uint64_t*)(pict->data[1] + 8*mb_x + (8*mb_y + y)*pict->linesize[1])= u;
2011                         *(uint64_t*)(pict->data[2] + 8*mb_x + (8*mb_y + y)*pict->linesize[2])= v;
2012                     }
2013
2014                     //segmentation
2015                     if(IS_8X8(mb_type) || IS_16X8(mb_type)){
2016                         *(uint64_t*)(pict->data[0] + 16*mb_x + 0 + (16*mb_y + 8)*pict->linesize[0])^= 0x8080808080808080ULL;
2017                         *(uint64_t*)(pict->data[0] + 16*mb_x + 8 + (16*mb_y + 8)*pict->linesize[0])^= 0x8080808080808080ULL;
2018                     }
2019                     if(IS_8X8(mb_type) || IS_8X16(mb_type)){
2020                         for(y=0; y<16; y++)
2021                             pict->data[0][16*mb_x + 8 + (16*mb_y + y)*pict->linesize[0]]^= 0x80;
2022                     }
2023                     if(IS_8X8(mb_type) && mv_sample_log2 >= 2){
2024                         int dm= 1 << (mv_sample_log2-2);
2025                         for(i=0; i<4; i++){
2026                             int sx= mb_x*16 + 8*(i&1);
2027                             int sy= mb_y*16 + 8*(i>>1);
2028                             int xy= (mb_x*2 + (i&1) + (mb_y*2 + (i>>1))*mv_stride) << (mv_sample_log2-1);
2029                             //FIXME bidir
2030                             int32_t *mv = (int32_t*)&pict->motion_val[0][xy];
2031                             if(mv[0] != mv[dm] || mv[dm*mv_stride] != mv[dm*(mv_stride+1)])
2032                                 for(y=0; y<8; y++)
2033                                     pict->data[0][sx + 4 + (sy + y)*pict->linesize[0]]^= 0x80;
2034                             if(mv[0] != mv[dm*mv_stride] || mv[dm] != mv[dm*(mv_stride+1)])
2035                                 *(uint64_t*)(pict->data[0] + sx + (sy + 4)*pict->linesize[0])^= 0x8080808080808080ULL;
2036                         }
2037                     }
2038
2039                     if(IS_INTERLACED(mb_type) && s->codec_id == CODEC_ID_H264){
2040                         // hmm
2041                     }
2042                 }
2043                 s->mbskip_table[mb_index]=0;
2044             }
2045         }
2046     }
2047 }
2048
2049 #ifdef CONFIG_ENCODERS
2050
2051 static int get_sae(uint8_t *src, int ref, int stride){
2052     int x,y;
2053     int acc=0;
2054
2055     for(y=0; y<16; y++){
2056         for(x=0; x<16; x++){
2057             acc+= FFABS(src[x+y*stride] - ref);
2058         }
2059     }
2060
2061     return acc;
2062 }
2063
2064 static int get_intra_count(MpegEncContext *s, uint8_t *src, uint8_t *ref, int stride){
2065     int x, y, w, h;
2066     int acc=0;
2067
2068     w= s->width &~15;
2069     h= s->height&~15;
2070
2071     for(y=0; y<h; y+=16){
2072         for(x=0; x<w; x+=16){
2073             int offset= x + y*stride;
2074             int sad = s->dsp.sad[0](NULL, src + offset, ref + offset, stride, 16);
2075             int mean= (s->dsp.pix_sum(src + offset, stride) + 128)>>8;
2076             int sae = get_sae(src + offset, mean, stride);
2077
2078             acc+= sae + 500 < sad;
2079         }
2080     }
2081     return acc;
2082 }
2083
2084
2085 static int load_input_picture(MpegEncContext *s, AVFrame *pic_arg){
2086     AVFrame *pic=NULL;
2087     int64_t pts;
2088     int i;
2089     const int encoding_delay= s->max_b_frames;
2090     int direct=1;
2091
2092     if(pic_arg){
2093         pts= pic_arg->pts;
2094         pic_arg->display_picture_number= s->input_picture_number++;
2095
2096         if(pts != AV_NOPTS_VALUE){
2097             if(s->user_specified_pts != AV_NOPTS_VALUE){
2098                 int64_t time= pts;
2099                 int64_t last= s->user_specified_pts;
2100
2101                 if(time <= last){
2102                     av_log(s->avctx, AV_LOG_ERROR, "Error, Invalid timestamp=%"PRId64", last=%"PRId64"\n", pts, s->user_specified_pts);
2103                     return -1;
2104                 }
2105             }
2106             s->user_specified_pts= pts;
2107         }else{
2108             if(s->user_specified_pts != AV_NOPTS_VALUE){
2109                 s->user_specified_pts=
2110                 pts= s->user_specified_pts + 1;
2111                 av_log(s->avctx, AV_LOG_INFO, "Warning: AVFrame.pts=? trying to guess (%"PRId64")\n", pts);
2112             }else{
2113                 pts= pic_arg->display_picture_number;
2114             }
2115         }
2116     }
2117
2118   if(pic_arg){
2119     if(encoding_delay && !(s->flags&CODEC_FLAG_INPUT_PRESERVED)) direct=0;
2120     if(pic_arg->linesize[0] != s->linesize) direct=0;
2121     if(pic_arg->linesize[1] != s->uvlinesize) direct=0;
2122     if(pic_arg->linesize[2] != s->uvlinesize) direct=0;
2123
2124 //    av_log(AV_LOG_DEBUG, "%d %d %d %d\n",pic_arg->linesize[0], pic_arg->linesize[1], s->linesize, s->uvlinesize);
2125
2126     if(direct){
2127         i= ff_find_unused_picture(s, 1);
2128
2129         pic= (AVFrame*)&s->picture[i];
2130         pic->reference= 3;
2131
2132         for(i=0; i<4; i++){
2133             pic->data[i]= pic_arg->data[i];
2134             pic->linesize[i]= pic_arg->linesize[i];
2135         }
2136         alloc_picture(s, (Picture*)pic, 1);
2137     }else{
2138         i= ff_find_unused_picture(s, 0);
2139
2140         pic= (AVFrame*)&s->picture[i];
2141         pic->reference= 3;
2142
2143         alloc_picture(s, (Picture*)pic, 0);
2144
2145         if(   pic->data[0] + INPLACE_OFFSET == pic_arg->data[0]
2146            && pic->data[1] + INPLACE_OFFSET == pic_arg->data[1]
2147            && pic->data[2] + INPLACE_OFFSET == pic_arg->data[2]){
2148        // empty
2149         }else{
2150             int h_chroma_shift, v_chroma_shift;
2151             avcodec_get_chroma_sub_sample(s->avctx->pix_fmt, &h_chroma_shift, &v_chroma_shift);
2152
2153             for(i=0; i<3; i++){
2154                 int src_stride= pic_arg->linesize[i];
2155                 int dst_stride= i ? s->uvlinesize : s->linesize;
2156                 int h_shift= i ? h_chroma_shift : 0;
2157                 int v_shift= i ? v_chroma_shift : 0;
2158                 int w= s->width >>h_shift;
2159                 int h= s->height>>v_shift;
2160                 uint8_t *src= pic_arg->data[i];
2161                 uint8_t *dst= pic->data[i];
2162
2163                 if(!s->avctx->rc_buffer_size)
2164                     dst +=INPLACE_OFFSET;
2165
2166                 if(src_stride==dst_stride)
2167                     memcpy(dst, src, src_stride*h);
2168                 else{
2169                     while(h--){
2170                         memcpy(dst, src, w);
2171                         dst += dst_stride;
2172                         src += src_stride;
2173                     }
2174                 }
2175             }
2176         }
2177     }
2178     copy_picture_attributes(s, pic, pic_arg);
2179     pic->pts= pts; //we set this here to avoid modifiying pic_arg
2180   }
2181
2182     /* shift buffer entries */
2183     for(i=1; i<MAX_PICTURE_COUNT /*s->encoding_delay+1*/; i++)
2184         s->input_picture[i-1]= s->input_picture[i];
2185
2186     s->input_picture[encoding_delay]= (Picture*)pic;
2187
2188     return 0;
2189 }
2190
2191 static int skip_check(MpegEncContext *s, Picture *p, Picture *ref){
2192     int x, y, plane;
2193     int score=0;
2194     int64_t score64=0;
2195
2196     for(plane=0; plane<3; plane++){
2197         const int stride= p->linesize[plane];
2198         const int bw= plane ? 1 : 2;
2199         for(y=0; y<s->mb_height*bw; y++){
2200             for(x=0; x<s->mb_width*bw; x++){
2201                 int off= p->type == FF_BUFFER_TYPE_SHARED ? 0: 16;
2202                 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);
2203
2204                 switch(s->avctx->frame_skip_exp){
2205                     case 0: score= FFMAX(score, v); break;
2206                     case 1: score+= FFABS(v);break;
2207                     case 2: score+= v*v;break;
2208                     case 3: score64+= FFABS(v*v*(int64_t)v);break;
2209                     case 4: score64+= v*v*(int64_t)(v*v);break;
2210                 }
2211             }
2212         }
2213     }
2214
2215     if(score) score64= score;
2216
2217     if(score64 < s->avctx->frame_skip_threshold)
2218         return 1;
2219     if(score64 < ((s->avctx->frame_skip_factor * (int64_t)s->lambda)>>8))
2220         return 1;
2221     return 0;
2222 }
2223
2224 static int estimate_best_b_count(MpegEncContext *s){
2225     AVCodec *codec= avcodec_find_encoder(s->avctx->codec_id);
2226     AVCodecContext *c= avcodec_alloc_context();
2227     AVFrame input[FF_MAX_B_FRAMES+2];
2228     const int scale= s->avctx->brd_scale;
2229     int i, j, out_size, p_lambda, b_lambda, lambda2;
2230     int outbuf_size= s->width * s->height; //FIXME
2231     uint8_t *outbuf= av_malloc(outbuf_size);
2232     int64_t best_rd= INT64_MAX;
2233     int best_b_count= -1;
2234
2235     assert(scale>=0 && scale <=3);
2236
2237 //    emms_c();
2238     p_lambda= s->last_lambda_for[P_TYPE]; //s->next_picture_ptr->quality;
2239     b_lambda= s->last_lambda_for[B_TYPE]; //p_lambda *FFABS(s->avctx->b_quant_factor) + s->avctx->b_quant_offset;
2240     if(!b_lambda) b_lambda= p_lambda; //FIXME we should do this somewhere else
2241     lambda2= (b_lambda*b_lambda + (1<<FF_LAMBDA_SHIFT)/2 ) >> FF_LAMBDA_SHIFT;
2242
2243     c->width = s->width >> scale;
2244     c->height= s->height>> scale;
2245     c->flags= CODEC_FLAG_QSCALE | CODEC_FLAG_PSNR | CODEC_FLAG_INPUT_PRESERVED /*| CODEC_FLAG_EMU_EDGE*/;
2246     c->flags|= s->avctx->flags & CODEC_FLAG_QPEL;
2247     c->mb_decision= s->avctx->mb_decision;
2248     c->me_cmp= s->avctx->me_cmp;
2249     c->mb_cmp= s->avctx->mb_cmp;
2250     c->me_sub_cmp= s->avctx->me_sub_cmp;
2251     c->pix_fmt = PIX_FMT_YUV420P;
2252     c->time_base= s->avctx->time_base;
2253     c->max_b_frames= s->max_b_frames;
2254
2255     if (avcodec_open(c, codec) < 0)
2256         return -1;
2257
2258     for(i=0; i<s->max_b_frames+2; i++){
2259         int ysize= c->width*c->height;
2260         int csize= (c->width/2)*(c->height/2);
2261         Picture pre_input, *pre_input_ptr= i ? s->input_picture[i-1] : s->next_picture_ptr;
2262
2263         if(pre_input_ptr)
2264             pre_input= *pre_input_ptr;
2265
2266         if(pre_input.type != FF_BUFFER_TYPE_SHARED && i){
2267             pre_input.data[0]+=INPLACE_OFFSET;
2268             pre_input.data[1]+=INPLACE_OFFSET;
2269             pre_input.data[2]+=INPLACE_OFFSET;
2270         }
2271
2272         avcodec_get_frame_defaults(&input[i]);
2273         input[i].data[0]= av_malloc(ysize + 2*csize);
2274         input[i].data[1]= input[i].data[0] + ysize;
2275         input[i].data[2]= input[i].data[1] + csize;
2276         input[i].linesize[0]= c->width;
2277         input[i].linesize[1]=
2278         input[i].linesize[2]= c->width/2;
2279
2280         if(!i || s->input_picture[i-1]){
2281             s->dsp.shrink[scale](input[i].data[0], input[i].linesize[0], pre_input.data[0], pre_input.linesize[0], c->width, c->height);
2282             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);
2283             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);
2284         }
2285     }
2286
2287     for(j=0; j<s->max_b_frames+1; j++){
2288         int64_t rd=0;
2289
2290         if(!s->input_picture[j])
2291             break;
2292
2293         c->error[0]= c->error[1]= c->error[2]= 0;
2294
2295         input[0].pict_type= I_TYPE;
2296         input[0].quality= 1 * FF_QP2LAMBDA;
2297         out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[0]);
2298 //        rd += (out_size * lambda2) >> FF_LAMBDA_SHIFT;
2299
2300         for(i=0; i<s->max_b_frames+1; i++){
2301             int is_p= i % (j+1) == j || i==s->max_b_frames;
2302
2303             input[i+1].pict_type= is_p ? P_TYPE : B_TYPE;
2304             input[i+1].quality= is_p ? p_lambda : b_lambda;
2305             out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[i+1]);
2306             rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
2307         }
2308
2309         /* get the delayed frames */
2310         while(out_size){
2311             out_size = avcodec_encode_video(c, outbuf, outbuf_size, NULL);
2312             rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
2313         }
2314
2315         rd += c->error[0] + c->error[1] + c->error[2];
2316
2317         if(rd < best_rd){
2318             best_rd= rd;
2319             best_b_count= j;
2320         }
2321     }
2322
2323     av_freep(&outbuf);
2324     avcodec_close(c);
2325     av_freep(&c);
2326
2327     for(i=0; i<s->max_b_frames+2; i++){
2328         av_freep(&input[i].data[0]);
2329     }
2330
2331     return best_b_count;
2332 }
2333
2334 static void select_input_picture(MpegEncContext *s){
2335     int i;
2336
2337     for(i=1; i<MAX_PICTURE_COUNT; i++)
2338         s->reordered_input_picture[i-1]= s->reordered_input_picture[i];
2339     s->reordered_input_picture[MAX_PICTURE_COUNT-1]= NULL;
2340
2341     /* set next picture type & ordering */
2342     if(s->reordered_input_picture[0]==NULL && s->input_picture[0]){
2343         if(/*s->picture_in_gop_number >= s->gop_size ||*/ s->next_picture_ptr==NULL || s->intra_only){
2344             s->reordered_input_picture[0]= s->input_picture[0];
2345             s->reordered_input_picture[0]->pict_type= I_TYPE;
2346             s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++;
2347         }else{
2348             int b_frames;
2349
2350             if(s->avctx->frame_skip_threshold || s->avctx->frame_skip_factor){
2351                 if(s->picture_in_gop_number < s->gop_size && skip_check(s, s->input_picture[0], s->next_picture_ptr)){
2352                 //FIXME check that te gop check above is +-1 correct
2353 //av_log(NULL, AV_LOG_DEBUG, "skip %p %"PRId64"\n", s->input_picture[0]->data[0], s->input_picture[0]->pts);
2354
2355                     if(s->input_picture[0]->type == FF_BUFFER_TYPE_SHARED){
2356                         for(i=0; i<4; i++)
2357                             s->input_picture[0]->data[i]= NULL;
2358                         s->input_picture[0]->type= 0;
2359                     }else{
2360                         assert(   s->input_picture[0]->type==FF_BUFFER_TYPE_USER
2361                                || s->input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
2362
2363                         s->avctx->release_buffer(s->avctx, (AVFrame*)s->input_picture[0]);
2364                     }
2365
2366                     emms_c();
2367                     ff_vbv_update(s, 0);
2368
2369                     goto no_output_pic;
2370                 }
2371             }
2372
2373             if(s->flags&CODEC_FLAG_PASS2){
2374                 for(i=0; i<s->max_b_frames+1; i++){
2375                     int pict_num= s->input_picture[0]->display_picture_number + i;
2376
2377                     if(pict_num >= s->rc_context.num_entries)
2378                         break;
2379                     if(!s->input_picture[i]){
2380                         s->rc_context.entry[pict_num-1].new_pict_type = P_TYPE;
2381                         break;
2382                     }
2383
2384                     s->input_picture[i]->pict_type=
2385                         s->rc_context.entry[pict_num].new_pict_type;
2386                 }
2387             }
2388
2389             if(s->avctx->b_frame_strategy==0){
2390                 b_frames= s->max_b_frames;
2391                 while(b_frames && !s->input_picture[b_frames]) b_frames--;
2392             }else if(s->avctx->b_frame_strategy==1){
2393                 for(i=1; i<s->max_b_frames+1; i++){
2394                     if(s->input_picture[i] && s->input_picture[i]->b_frame_score==0){
2395                         s->input_picture[i]->b_frame_score=
2396                             get_intra_count(s, s->input_picture[i  ]->data[0],
2397                                                s->input_picture[i-1]->data[0], s->linesize) + 1;
2398                     }
2399                 }
2400                 for(i=0; i<s->max_b_frames+1; i++){
2401                     if(s->input_picture[i]==NULL || s->input_picture[i]->b_frame_score - 1 > s->mb_num/s->avctx->b_sensitivity) break;
2402                 }
2403
2404                 b_frames= FFMAX(0, i-1);
2405
2406                 /* reset scores */
2407                 for(i=0; i<b_frames+1; i++){
2408                     s->input_picture[i]->b_frame_score=0;
2409                 }
2410             }else if(s->avctx->b_frame_strategy==2){
2411                 b_frames= estimate_best_b_count(s);
2412             }else{
2413                 av_log(s->avctx, AV_LOG_ERROR, "illegal b frame strategy\n");
2414                 b_frames=0;
2415             }
2416
2417             emms_c();
2418 //static int b_count=0;
2419 //b_count+= b_frames;
2420 //av_log(s->avctx, AV_LOG_DEBUG, "b_frames: %d\n", b_count);
2421
2422             for(i= b_frames - 1; i>=0; i--){
2423                 int type= s->input_picture[i]->pict_type;
2424                 if(type && type != B_TYPE)
2425                     b_frames= i;
2426             }
2427             if(s->input_picture[b_frames]->pict_type == B_TYPE && b_frames == s->max_b_frames){
2428                 av_log(s->avctx, AV_LOG_ERROR, "warning, too many b frames in a row\n");
2429             }
2430
2431             if(s->picture_in_gop_number + b_frames >= s->gop_size){
2432               if((s->flags2 & CODEC_FLAG2_STRICT_GOP) && s->gop_size > s->picture_in_gop_number){
2433                     b_frames= s->gop_size - s->picture_in_gop_number - 1;
2434               }else{
2435                 if(s->flags & CODEC_FLAG_CLOSED_GOP)
2436                     b_frames=0;
2437                 s->input_picture[b_frames]->pict_type= I_TYPE;
2438               }
2439             }
2440
2441             if(   (s->flags & CODEC_FLAG_CLOSED_GOP)
2442                && b_frames
2443                && s->input_picture[b_frames]->pict_type== I_TYPE)
2444                 b_frames--;
2445
2446             s->reordered_input_picture[0]= s->input_picture[b_frames];
2447             if(s->reordered_input_picture[0]->pict_type != I_TYPE)
2448                 s->reordered_input_picture[0]->pict_type= P_TYPE;
2449             s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++;
2450             for(i=0; i<b_frames; i++){
2451                 s->reordered_input_picture[i+1]= s->input_picture[i];
2452                 s->reordered_input_picture[i+1]->pict_type= B_TYPE;
2453                 s->reordered_input_picture[i+1]->coded_picture_number= s->coded_picture_number++;
2454             }
2455         }
2456     }
2457 no_output_pic:
2458     if(s->reordered_input_picture[0]){
2459         s->reordered_input_picture[0]->reference= s->reordered_input_picture[0]->pict_type!=B_TYPE ? 3 : 0;
2460
2461         copy_picture(&s->new_picture, s->reordered_input_picture[0]);
2462
2463         if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_SHARED || s->avctx->rc_buffer_size){
2464             // input is a shared pix, so we can't modifiy it -> alloc a new one & ensure that the shared one is reuseable
2465
2466             int i= ff_find_unused_picture(s, 0);
2467             Picture *pic= &s->picture[i];
2468
2469             pic->reference              = s->reordered_input_picture[0]->reference;
2470             alloc_picture(s, pic, 0);
2471
2472             /* mark us unused / free shared pic */
2473             if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_INTERNAL)
2474                 s->avctx->release_buffer(s->avctx, (AVFrame*)s->reordered_input_picture[0]);
2475             for(i=0; i<4; i++)
2476                 s->reordered_input_picture[0]->data[i]= NULL;
2477             s->reordered_input_picture[0]->type= 0;
2478
2479             copy_picture_attributes(s, (AVFrame*)pic, (AVFrame*)s->reordered_input_picture[0]);
2480
2481             s->current_picture_ptr= pic;
2482         }else{
2483             // input is not a shared pix -> reuse buffer for current_pix
2484
2485             assert(   s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_USER
2486                    || s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
2487
2488             s->current_picture_ptr= s->reordered_input_picture[0];
2489             for(i=0; i<4; i++){
2490                 s->new_picture.data[i]+= INPLACE_OFFSET;
2491             }
2492         }
2493         copy_picture(&s->current_picture, s->current_picture_ptr);
2494
2495         s->picture_number= s->new_picture.display_picture_number;
2496 //printf("dpn:%d\n", s->picture_number);
2497     }else{
2498        memset(&s->new_picture, 0, sizeof(Picture));
2499     }
2500 }
2501
2502 int MPV_encode_picture(AVCodecContext *avctx,
2503                        unsigned char *buf, int buf_size, void *data)
2504 {
2505     MpegEncContext *s = avctx->priv_data;
2506     AVFrame *pic_arg = data;
2507     int i, stuffing_count;
2508
2509     for(i=0; i<avctx->thread_count; i++){
2510         int start_y= s->thread_context[i]->start_mb_y;
2511         int   end_y= s->thread_context[i]->  end_mb_y;
2512         int h= s->mb_height;
2513         uint8_t *start= buf + (size_t)(((int64_t) buf_size)*start_y/h);
2514         uint8_t *end  = buf + (size_t)(((int64_t) buf_size)*  end_y/h);
2515
2516         init_put_bits(&s->thread_context[i]->pb, start, end - start);
2517     }
2518
2519     s->picture_in_gop_number++;
2520
2521     if(load_input_picture(s, pic_arg) < 0)
2522         return -1;
2523
2524     select_input_picture(s);
2525
2526     /* output? */
2527     if(s->new_picture.data[0]){
2528         s->pict_type= s->new_picture.pict_type;
2529 //emms_c();
2530 //printf("qs:%f %f %d\n", s->new_picture.quality, s->current_picture.quality, s->qscale);
2531         MPV_frame_start(s, avctx);
2532 vbv_retry:
2533         if (encode_picture(s, s->picture_number) < 0)
2534             return -1;
2535
2536         avctx->real_pict_num  = s->picture_number;
2537         avctx->header_bits = s->header_bits;
2538         avctx->mv_bits     = s->mv_bits;
2539         avctx->misc_bits   = s->misc_bits;
2540         avctx->i_tex_bits  = s->i_tex_bits;
2541         avctx->p_tex_bits  = s->p_tex_bits;
2542         avctx->i_count     = s->i_count;
2543         avctx->p_count     = s->mb_num - s->i_count - s->skip_count; //FIXME f/b_count in avctx
2544         avctx->skip_count  = s->skip_count;
2545
2546         MPV_frame_end(s);
2547
2548         if (s->out_format == FMT_MJPEG)
2549             mjpeg_picture_trailer(s);
2550
2551         if(avctx->rc_buffer_size){
2552             RateControlContext *rcc= &s->rc_context;
2553             int max_size= rcc->buffer_index/3;
2554
2555             if(put_bits_count(&s->pb) > max_size && s->lambda < s->avctx->lmax){
2556                 s->next_lambda= FFMAX(s->lambda+1, s->lambda*(s->qscale+1) / s->qscale);
2557                 if(s->adaptive_quant){
2558                     int i;
2559                     for(i=0; i<s->mb_height*s->mb_stride; i++)
2560                         s->lambda_table[i]= FFMAX(s->lambda_table[i]+1, s->lambda_table[i]*(s->qscale+1) / s->qscale);
2561                 }
2562                 s->mb_skipped = 0;        //done in MPV_frame_start()
2563                 if(s->pict_type==P_TYPE){ //done in encode_picture() so we must undo it
2564                     if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
2565                         s->no_rounding ^= 1;
2566                 }
2567                 if(s->pict_type!=B_TYPE){
2568                     s->time_base= s->last_time_base;
2569                     s->last_non_b_time= s->time - s->pp_time;
2570                 }
2571 //                av_log(NULL, AV_LOG_ERROR, "R:%d ", s->next_lambda);
2572                 for(i=0; i<avctx->thread_count; i++){
2573                     PutBitContext *pb= &s->thread_context[i]->pb;
2574                     init_put_bits(pb, pb->buf, pb->buf_end - pb->buf);
2575                 }
2576                 goto vbv_retry;
2577             }
2578
2579             assert(s->avctx->rc_max_rate);
2580         }
2581
2582         if(s->flags&CODEC_FLAG_PASS1)
2583             ff_write_pass1_stats(s);
2584
2585         for(i=0; i<4; i++){
2586             s->current_picture_ptr->error[i]= s->current_picture.error[i];
2587             avctx->error[i] += s->current_picture_ptr->error[i];
2588         }
2589
2590         if(s->flags&CODEC_FLAG_PASS1)
2591             assert(avctx->header_bits + avctx->mv_bits + avctx->misc_bits + avctx->i_tex_bits + avctx->p_tex_bits == put_bits_count(&s->pb));
2592         flush_put_bits(&s->pb);
2593         s->frame_bits  = put_bits_count(&s->pb);
2594
2595         stuffing_count= ff_vbv_update(s, s->frame_bits);
2596         if(stuffing_count){
2597             if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < stuffing_count + 50){
2598                 av_log(s->avctx, AV_LOG_ERROR, "stuffing too large\n");
2599                 return -1;
2600             }
2601
2602             switch(s->codec_id){
2603             case CODEC_ID_MPEG1VIDEO:
2604             case CODEC_ID_MPEG2VIDEO:
2605                 while(stuffing_count--){
2606                     put_bits(&s->pb, 8, 0);
2607                 }
2608             break;
2609             case CODEC_ID_MPEG4:
2610                 put_bits(&s->pb, 16, 0);
2611                 put_bits(&s->pb, 16, 0x1C3);
2612                 stuffing_count -= 4;
2613                 while(stuffing_count--){
2614                     put_bits(&s->pb, 8, 0xFF);
2615                 }
2616             break;
2617             default:
2618                 av_log(s->avctx, AV_LOG_ERROR, "vbv buffer overflow\n");
2619             }
2620             flush_put_bits(&s->pb);
2621             s->frame_bits  = put_bits_count(&s->pb);
2622         }
2623
2624         /* update mpeg1/2 vbv_delay for CBR */
2625         if(s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate && s->out_format == FMT_MPEG1
2626            && 90000LL * (avctx->rc_buffer_size-1) <= s->avctx->rc_max_rate*0xFFFFLL){
2627             int vbv_delay;
2628
2629             assert(s->repeat_first_field==0);
2630
2631             vbv_delay= lrintf(90000 * s->rc_context.buffer_index / s->avctx->rc_max_rate);
2632             assert(vbv_delay < 0xFFFF);
2633
2634             s->vbv_delay_ptr[0] &= 0xF8;
2635             s->vbv_delay_ptr[0] |= vbv_delay>>13;
2636             s->vbv_delay_ptr[1]  = vbv_delay>>5;
2637             s->vbv_delay_ptr[2] &= 0x07;
2638             s->vbv_delay_ptr[2] |= vbv_delay<<3;
2639         }
2640         s->total_bits += s->frame_bits;
2641         avctx->frame_bits  = s->frame_bits;
2642     }else{
2643         assert((pbBufPtr(&s->pb) == s->pb.buf));
2644         s->frame_bits=0;
2645     }
2646     assert((s->frame_bits&7)==0);
2647
2648     return s->frame_bits/8;
2649 }
2650
2651 #endif //CONFIG_ENCODERS
2652
2653 static inline void gmc1_motion(MpegEncContext *s,
2654                                uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2655                                uint8_t **ref_picture)
2656 {
2657     uint8_t *ptr;
2658     int offset, src_x, src_y, linesize, uvlinesize;
2659     int motion_x, motion_y;
2660     int emu=0;
2661
2662     motion_x= s->sprite_offset[0][0];
2663     motion_y= s->sprite_offset[0][1];
2664     src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy+1));
2665     src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy+1));
2666     motion_x<<=(3-s->sprite_warping_accuracy);
2667     motion_y<<=(3-s->sprite_warping_accuracy);
2668     src_x = clip(src_x, -16, s->width);
2669     if (src_x == s->width)
2670         motion_x =0;
2671     src_y = clip(src_y, -16, s->height);
2672     if (src_y == s->height)
2673         motion_y =0;
2674
2675     linesize = s->linesize;
2676     uvlinesize = s->uvlinesize;
2677
2678     ptr = ref_picture[0] + (src_y * linesize) + src_x;
2679
2680     if(s->flags&CODEC_FLAG_EMU_EDGE){
2681         if(   (unsigned)src_x >= s->h_edge_pos - 17
2682            || (unsigned)src_y >= s->v_edge_pos - 17){
2683             ff_emulated_edge_mc(s->edge_emu_buffer, ptr, linesize, 17, 17, src_x, src_y, s->h_edge_pos, s->v_edge_pos);
2684             ptr= s->edge_emu_buffer;
2685         }
2686     }
2687
2688     if((motion_x|motion_y)&7){
2689         s->dsp.gmc1(dest_y  , ptr  , linesize, 16, motion_x&15, motion_y&15, 128 - s->no_rounding);
2690         s->dsp.gmc1(dest_y+8, ptr+8, linesize, 16, motion_x&15, motion_y&15, 128 - s->no_rounding);
2691     }else{
2692         int dxy;
2693
2694         dxy= ((motion_x>>3)&1) | ((motion_y>>2)&2);
2695         if (s->no_rounding){
2696             s->dsp.put_no_rnd_pixels_tab[0][dxy](dest_y, ptr, linesize, 16);
2697         }else{
2698             s->dsp.put_pixels_tab       [0][dxy](dest_y, ptr, linesize, 16);
2699         }
2700     }
2701
2702     if(s->flags&CODEC_FLAG_GRAY) return;
2703
2704     motion_x= s->sprite_offset[1][0];
2705     motion_y= s->sprite_offset[1][1];
2706     src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy+1));
2707     src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy+1));
2708     motion_x<<=(3-s->sprite_warping_accuracy);
2709     motion_y<<=(3-s->sprite_warping_accuracy);
2710     src_x = clip(src_x, -8, s->width>>1);
2711     if (src_x == s->width>>1)
2712         motion_x =0;
2713     src_y = clip(src_y, -8, s->height>>1);
2714     if (src_y == s->height>>1)
2715         motion_y =0;
2716
2717     offset = (src_y * uvlinesize) + src_x;
2718     ptr = ref_picture[1] + offset;
2719     if(s->flags&CODEC_FLAG_EMU_EDGE){
2720         if(   (unsigned)src_x >= (s->h_edge_pos>>1) - 9
2721            || (unsigned)src_y >= (s->v_edge_pos>>1) - 9){
2722             ff_emulated_edge_mc(s->edge_emu_buffer, ptr, uvlinesize, 9, 9, src_x, src_y, s->h_edge_pos>>1, s->v_edge_pos>>1);
2723             ptr= s->edge_emu_buffer;
2724             emu=1;
2725         }
2726     }
2727     s->dsp.gmc1(dest_cb, ptr, uvlinesize, 8, motion_x&15, motion_y&15, 128 - s->no_rounding);
2728
2729     ptr = ref_picture[2] + offset;
2730     if(emu){
2731         ff_emulated_edge_mc(s->edge_emu_buffer, ptr, uvlinesize, 9, 9, src_x, src_y, s->h_edge_pos>>1, s->v_edge_pos>>1);
2732         ptr= s->edge_emu_buffer;
2733     }
2734     s->dsp.gmc1(dest_cr, ptr, uvlinesize, 8, motion_x&15, motion_y&15, 128 - s->no_rounding);
2735
2736     return;
2737 }
2738
2739 static inline void gmc_motion(MpegEncContext *s,
2740                                uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2741                                uint8_t **ref_picture)
2742 {
2743     uint8_t *ptr;
2744     int linesize, uvlinesize;
2745     const int a= s->sprite_warping_accuracy;
2746     int ox, oy;
2747
2748     linesize = s->linesize;
2749     uvlinesize = s->uvlinesize;
2750
2751     ptr = ref_picture[0];
2752
2753     ox= s->sprite_offset[0][0] + s->sprite_delta[0][0]*s->mb_x*16 + s->sprite_delta[0][1]*s->mb_y*16;
2754     oy= s->sprite_offset[0][1] + s->sprite_delta[1][0]*s->mb_x*16 + s->sprite_delta[1][1]*s->mb_y*16;
2755
2756     s->dsp.gmc(dest_y, ptr, linesize, 16,
2757            ox,
2758            oy,
2759            s->sprite_delta[0][0], s->sprite_delta[0][1],
2760            s->sprite_delta[1][0], s->sprite_delta[1][1],
2761            a+1, (1<<(2*a+1)) - s->no_rounding,
2762            s->h_edge_pos, s->v_edge_pos);
2763     s->dsp.gmc(dest_y+8, ptr, linesize, 16,
2764            ox + s->sprite_delta[0][0]*8,
2765            oy + s->sprite_delta[1][0]*8,
2766            s->sprite_delta[0][0], s->sprite_delta[0][1],
2767            s->sprite_delta[1][0], s->sprite_delta[1][1],
2768            a+1, (1<<(2*a+1)) - s->no_rounding,
2769            s->h_edge_pos, s->v_edge_pos);
2770
2771     if(s->flags&CODEC_FLAG_GRAY) return;
2772
2773     ox= s->sprite_offset[1][0] + s->sprite_delta[0][0]*s->mb_x*8 + s->sprite_delta[0][1]*s->mb_y*8;
2774     oy= s->sprite_offset[1][1] + s->sprite_delta[1][0]*s->mb_x*8 + s->sprite_delta[1][1]*s->mb_y*8;
2775
2776     ptr = ref_picture[1];
2777     s->dsp.gmc(dest_cb, ptr, uvlinesize, 8,
2778            ox,
2779            oy,
2780            s->sprite_delta[0][0], s->sprite_delta[0][1],
2781            s->sprite_delta[1][0], s->sprite_delta[1][1],
2782            a+1, (1<<(2*a+1)) - s->no_rounding,
2783            s->h_edge_pos>>1, s->v_edge_pos>>1);
2784
2785     ptr = ref_picture[2];
2786     s->dsp.gmc(dest_cr, ptr, uvlinesize, 8,
2787            ox,
2788            oy,
2789            s->sprite_delta[0][0], s->sprite_delta[0][1],
2790            s->sprite_delta[1][0], s->sprite_delta[1][1],
2791            a+1, (1<<(2*a+1)) - s->no_rounding,
2792            s->h_edge_pos>>1, s->v_edge_pos>>1);
2793 }
2794
2795 /**
2796  * Copies a rectangular area of samples to a temporary buffer and replicates the boarder samples.
2797  * @param buf destination buffer
2798  * @param src source buffer
2799  * @param linesize number of bytes between 2 vertically adjacent samples in both the source and destination buffers
2800  * @param block_w width of block
2801  * @param block_h height of block
2802  * @param src_x x coordinate of the top left sample of the block in the source buffer
2803  * @param src_y y coordinate of the top left sample of the block in the source buffer
2804  * @param w width of the source buffer
2805  * @param h height of the source buffer
2806  */
2807 void ff_emulated_edge_mc(uint8_t *buf, uint8_t *src, int linesize, int block_w, int block_h,
2808                                     int src_x, int src_y, int w, int h){
2809     int x, y;
2810     int start_y, start_x, end_y, end_x;
2811
2812     if(src_y>= h){
2813         src+= (h-1-src_y)*linesize;
2814         src_y=h-1;
2815     }else if(src_y<=-block_h){
2816         src+= (1-block_h-src_y)*linesize;
2817         src_y=1-block_h;
2818     }
2819     if(src_x>= w){
2820         src+= (w-1-src_x);
2821         src_x=w-1;
2822     }else if(src_x<=-block_w){
2823         src+= (1-block_w-src_x);
2824         src_x=1-block_w;
2825     }
2826
2827     start_y= FFMAX(0, -src_y);
2828     start_x= FFMAX(0, -src_x);
2829     end_y= FFMIN(block_h, h-src_y);
2830     end_x= FFMIN(block_w, w-src_x);
2831
2832     // copy existing part
2833     for(y=start_y; y<end_y; y++){
2834         for(x=start_x; x<end_x; x++){
2835             buf[x + y*linesize]= src[x + y*linesize];
2836         }
2837     }
2838
2839     //top
2840     for(y=0; y<start_y; y++){
2841         for(x=start_x; x<end_x; x++){
2842             buf[x + y*linesize]= buf[x + start_y*linesize];
2843         }
2844     }
2845
2846     //bottom
2847     for(y=end_y; y<block_h; y++){
2848         for(x=start_x; x<end_x; x++){
2849             buf[x + y*linesize]= buf[x + (end_y-1)*linesize];
2850         }
2851     }
2852
2853     for(y=0; y<block_h; y++){
2854        //left
2855         for(x=0; x<start_x; x++){
2856             buf[x + y*linesize]= buf[start_x + y*linesize];
2857         }
2858
2859        //right
2860         for(x=end_x; x<block_w; x++){
2861             buf[x + y*linesize]= buf[end_x - 1 + y*linesize];
2862         }
2863     }
2864 }
2865
2866 static inline int hpel_motion(MpegEncContext *s,
2867                                   uint8_t *dest, uint8_t *src,
2868                                   int field_based, int field_select,
2869                                   int src_x, int src_y,
2870                                   int width, int height, int stride,
2871                                   int h_edge_pos, int v_edge_pos,
2872                                   int w, int h, op_pixels_func *pix_op,
2873                                   int motion_x, int motion_y)
2874 {
2875     int dxy;
2876     int emu=0;
2877
2878     dxy = ((motion_y & 1) << 1) | (motion_x & 1);
2879     src_x += motion_x >> 1;
2880     src_y += motion_y >> 1;
2881
2882     /* WARNING: do no forget half pels */
2883     src_x = clip(src_x, -16, width); //FIXME unneeded for emu?
2884     if (src_x == width)
2885         dxy &= ~1;
2886     src_y = clip(src_y, -16, height);
2887     if (src_y == height)
2888         dxy &= ~2;
2889     src += src_y * stride + src_x;
2890
2891     if(s->unrestricted_mv && (s->flags&CODEC_FLAG_EMU_EDGE)){
2892         if(   (unsigned)src_x > h_edge_pos - (motion_x&1) - w
2893            || (unsigned)src_y > v_edge_pos - (motion_y&1) - h){
2894             ff_emulated_edge_mc(s->edge_emu_buffer, src, s->linesize, w+1, (h+1)<<field_based,
2895                              src_x, src_y<<field_based, h_edge_pos, s->v_edge_pos);
2896             src= s->edge_emu_buffer;
2897             emu=1;
2898         }
2899     }
2900     if(field_select)
2901         src += s->linesize;
2902     pix_op[dxy](dest, src, stride, h);
2903     return emu;
2904 }
2905
2906 static inline int hpel_motion_lowres(MpegEncContext *s,
2907                                   uint8_t *dest, uint8_t *src,
2908                                   int field_based, int field_select,
2909                                   int src_x, int src_y,
2910                                   int width, int height, int stride,
2911                                   int h_edge_pos, int v_edge_pos,
2912                                   int w, int h, h264_chroma_mc_func *pix_op,
2913                                   int motion_x, int motion_y)
2914 {
2915     const int lowres= s->avctx->lowres;
2916     const int s_mask= (2<<lowres)-1;
2917     int emu=0;
2918     int sx, sy;
2919
2920     if(s->quarter_sample){
2921         motion_x/=2;
2922         motion_y/=2;
2923     }
2924
2925     sx= motion_x & s_mask;
2926     sy= motion_y & s_mask;
2927     src_x += motion_x >> (lowres+1);
2928     src_y += motion_y >> (lowres+1);
2929
2930     src += src_y * stride + src_x;
2931
2932     if(   (unsigned)src_x > h_edge_pos                 - (!!sx) - w
2933        || (unsigned)src_y >(v_edge_pos >> field_based) - (!!sy) - h){
2934         ff_emulated_edge_mc(s->edge_emu_buffer, src, s->linesize, w+1, (h+1)<<field_based,
2935                             src_x, src_y<<field_based, h_edge_pos, v_edge_pos);
2936         src= s->edge_emu_buffer;
2937         emu=1;
2938     }
2939
2940     sx <<= 2 - lowres;
2941     sy <<= 2 - lowres;
2942     if(field_select)
2943         src += s->linesize;
2944     pix_op[lowres](dest, src, stride, h, sx, sy);
2945     return emu;
2946 }
2947
2948 /* apply one mpeg motion vector to the three components */
2949 static av_always_inline void mpeg_motion(MpegEncContext *s,
2950                                uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2951                                int field_based, int bottom_field, int field_select,
2952                                uint8_t **ref_picture, op_pixels_func (*pix_op)[4],
2953                                int motion_x, int motion_y, int h)
2954 {
2955     uint8_t *ptr_y, *ptr_cb, *ptr_cr;
2956     int dxy, uvdxy, mx, my, src_x, src_y, uvsrc_x, uvsrc_y, v_edge_pos, uvlinesize, linesize;
2957
2958 #if 0
2959 if(s->quarter_sample)
2960 {
2961     motion_x>>=1;
2962     motion_y>>=1;
2963 }
2964 #endif
2965
2966     v_edge_pos = s->v_edge_pos >> field_based;
2967     linesize   = s->current_picture.linesize[0] << field_based;
2968     uvlinesize = s->current_picture.linesize[1] << field_based;
2969
2970     dxy = ((motion_y & 1) << 1) | (motion_x & 1);
2971     src_x = s->mb_x* 16               + (motion_x >> 1);
2972     src_y =(s->mb_y<<(4-field_based)) + (motion_y >> 1);
2973
2974     if (s->out_format == FMT_H263) {
2975         if((s->workaround_bugs & FF_BUG_HPEL_CHROMA) && field_based){
2976             mx = (motion_x>>1)|(motion_x&1);
2977             my = motion_y >>1;
2978             uvdxy = ((my & 1) << 1) | (mx & 1);
2979             uvsrc_x = s->mb_x* 8               + (mx >> 1);
2980             uvsrc_y = (s->mb_y<<(3-field_based)) + (my >> 1);
2981         }else{
2982             uvdxy = dxy | (motion_y & 2) | ((motion_x & 2) >> 1);
2983             uvsrc_x = src_x>>1;
2984             uvsrc_y = src_y>>1;
2985         }
2986     }else if(s->out_format == FMT_H261){//even chroma mv's are full pel in H261
2987         mx = motion_x / 4;
2988         my = motion_y / 4;
2989         uvdxy = 0;
2990         uvsrc_x = s->mb_x*8 + mx;
2991         uvsrc_y = s->mb_y*8 + my;
2992     } else {
2993         if(s->chroma_y_shift){
2994             mx = motion_x / 2;
2995             my = motion_y / 2;
2996             uvdxy = ((my & 1) << 1) | (mx & 1);
2997             uvsrc_x = s->mb_x* 8               + (mx >> 1);
2998             uvsrc_y = (s->mb_y<<(3-field_based)) + (my >> 1);
2999         } else {
3000             if(s->chroma_x_shift){
3001             //Chroma422
3002                 mx = motion_x / 2;
3003                 uvdxy = ((motion_y & 1) << 1) | (mx & 1);
3004                 uvsrc_x = s->mb_x* 8           + (mx >> 1);
3005                 uvsrc_y = src_y;
3006             } else {
3007             //Chroma444
3008                 uvdxy = dxy;
3009                 uvsrc_x = src_x;
3010                 uvsrc_y = src_y;
3011             }
3012         }
3013     }
3014
3015     ptr_y  = ref_picture[0] + src_y * linesize + src_x;
3016     ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
3017     ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
3018
3019     if(   (unsigned)src_x > s->h_edge_pos - (motion_x&1) - 16
3020        || (unsigned)src_y >    v_edge_pos - (motion_y&1) - h){
3021             if(s->codec_id == CODEC_ID_MPEG2VIDEO ||
3022                s->codec_id == CODEC_ID_MPEG1VIDEO){
3023                 av_log(s->avctx,AV_LOG_DEBUG,"MPEG motion vector out of boundary\n");
3024                 return ;
3025             }
3026             ff_emulated_edge_mc(s->edge_emu_buffer, ptr_y, s->linesize, 17, 17+field_based,
3027                              src_x, src_y<<field_based, s->h_edge_pos, s->v_edge_pos);
3028             ptr_y = s->edge_emu_buffer;
3029             if(!(s->flags&CODEC_FLAG_GRAY)){
3030                 uint8_t *uvbuf= s->edge_emu_buffer+18*s->linesize;
3031                 ff_emulated_edge_mc(uvbuf  , ptr_cb, s->uvlinesize, 9, 9+field_based,
3032                                  uvsrc_x, uvsrc_y<<field_based, s->h_edge_pos>>1, s->v_edge_pos>>1);
3033                 ff_emulated_edge_mc(uvbuf+16, ptr_cr, s->uvlinesize, 9, 9+field_based,
3034                                  uvsrc_x, uvsrc_y<<field_based, s->h_edge_pos>>1, s->v_edge_pos>>1);
3035                 ptr_cb= uvbuf;
3036                 ptr_cr= uvbuf+16;
3037             }
3038     }
3039
3040     if(bottom_field){ //FIXME use this for field pix too instead of the obnoxious hack which changes picture.data
3041         dest_y += s->linesize;
3042         dest_cb+= s->uvlinesize;
3043         dest_cr+= s->uvlinesize;
3044     }
3045
3046     if(field_select){
3047         ptr_y += s->linesize;
3048         ptr_cb+= s->uvlinesize;
3049         ptr_cr+= s->uvlinesize;
3050     }
3051
3052     pix_op[0][dxy](dest_y, ptr_y, linesize, h);
3053
3054     if(!(s->flags&CODEC_FLAG_GRAY)){
3055         pix_op[s->chroma_x_shift][uvdxy](dest_cb, ptr_cb, uvlinesize, h >> s->chroma_y_shift);
3056         pix_op[s->chroma_x_shift][uvdxy](dest_cr, ptr_cr, uvlinesize, h >> s->chroma_y_shift);
3057     }
3058 #if defined(CONFIG_H261_ENCODER) || defined(CONFIG_H261_DECODER)
3059     if(s->out_format == FMT_H261){
3060         ff_h261_loop_filter(s);
3061     }
3062 #endif
3063 }
3064
3065 /* apply one mpeg motion vector to the three components */
3066 static av_always_inline void mpeg_motion_lowres(MpegEncContext *s,
3067                                uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
3068                                int field_based, int bottom_field, int field_select,
3069                                uint8_t **ref_picture, h264_chroma_mc_func *pix_op,
3070                                int motion_x, int motion_y, int h)
3071 {
3072     uint8_t *ptr_y, *ptr_cb, *ptr_cr;
3073     int mx, my, src_x, src_y, uvsrc_x, uvsrc_y, uvlinesize, linesize, sx, sy, uvsx, uvsy;
3074     const int lowres= s->avctx->lowres;
3075     const int block_s= 8>>lowres;
3076     const int s_mask= (2<<lowres)-1;
3077     const int h_edge_pos = s->h_edge_pos >> lowres;
3078     const int v_edge_pos = s->v_edge_pos >> lowres;
3079     linesize   = s->current_picture.linesize[0] << field_based;
3080     uvlinesize = s->current_picture.linesize[1] << field_based;
3081
3082     if(s->quarter_sample){ //FIXME obviously not perfect but qpel wont work in lowres anyway
3083         motion_x/=2;
3084         motion_y/=2;
3085     }
3086
3087     if(field_based){
3088         motion_y += (bottom_field - field_select)*((1<<lowres)-1);
3089     }
3090
3091     sx= motion_x & s_mask;
3092     sy= motion_y & s_mask;
3093     src_x = s->mb_x*2*block_s               + (motion_x >> (lowres+1));
3094     src_y =(s->mb_y*2*block_s>>field_based) + (motion_y >> (lowres+1));
3095
3096     if (s->out_format == FMT_H263) {
3097         uvsx = ((motion_x>>1) & s_mask) | (sx&1);
3098         uvsy = ((motion_y>>1) & s_mask) | (sy&1);
3099         uvsrc_x = src_x>>1;
3100         uvsrc_y = src_y>>1;
3101     }else if(s->out_format == FMT_H261){//even chroma mv's are full pel in H261
3102         mx = motion_x / 4;
3103         my = motion_y / 4;
3104         uvsx = (2*mx) & s_mask;
3105         uvsy = (2*my) & s_mask;
3106         uvsrc_x = s->mb_x*block_s               + (mx >> lowres);
3107         uvsrc_y = s->mb_y*block_s               + (my >> lowres);
3108     } else {
3109         mx = motion_x / 2;
3110         my = motion_y / 2;
3111         uvsx = mx & s_mask;
3112         uvsy = my & s_mask;
3113         uvsrc_x = s->mb_x*block_s               + (mx >> (lowres+1));
3114         uvsrc_y =(s->mb_y*block_s>>field_based) + (my >> (lowres+1));
3115     }
3116
3117     ptr_y  = ref_picture[0] + src_y * linesize + src_x;
3118     ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
3119     ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
3120
3121     if(   (unsigned)src_x > h_edge_pos                 - (!!sx) - 2*block_s
3122        || (unsigned)src_y >(v_edge_pos >> field_based) - (!!sy) - h){
3123             ff_emulated_edge_mc(s->edge_emu_buffer, ptr_y, s->linesize, 17, 17+field_based,
3124                              src_x, src_y<<field_based, h_edge_pos, v_edge_pos);
3125             ptr_y = s->edge_emu_buffer;
3126             if(!(s->flags&CODEC_FLAG_GRAY)){
3127                 uint8_t *uvbuf= s->edge_emu_buffer+18*s->linesize;
3128                 ff_emulated_edge_mc(uvbuf  , ptr_cb, s->uvlinesize, 9, 9+field_based,
3129                                  uvsrc_x, uvsrc_y<<field_based, h_edge_pos>>1, v_edge_pos>>1);
3130                 ff_emulated_edge_mc(uvbuf+16, ptr_cr, s->uvlinesize, 9, 9+field_based,
3131                                  uvsrc_x, uvsrc_y<<field_based, h_edge_pos>>1, v_edge_pos>>1);
3132                 ptr_cb= uvbuf;
3133                 ptr_cr= uvbuf+16;
3134             }
3135     }
3136
3137     if(bottom_field){ //FIXME use this for field pix too instead of the obnoxious hack which changes picture.data
3138         dest_y += s->linesize;
3139         dest_cb+= s->uvlinesize;
3140         dest_cr+= s->uvlinesize;
3141     }
3142
3143     if(field_select){
3144         ptr_y += s->linesize;
3145         ptr_cb+= s->uvlinesize;
3146         ptr_cr+= s->uvlinesize;
3147     }
3148
3149     sx <<= 2 - lowres;
3150     sy <<= 2 - lowres;
3151     pix_op[lowres-1](dest_y, ptr_y, linesize, h, sx, sy);
3152
3153     if(!(s->flags&CODEC_FLAG_GRAY)){
3154         uvsx <<= 2 - lowres;
3155         uvsy <<= 2 - lowres;
3156         pix_op[lowres](dest_cb, ptr_cb, uvlinesize, h >> s->chroma_y_shift, uvsx, uvsy);
3157         pix_op[lowres](dest_cr, ptr_cr, uvlinesize, h >> s->chroma_y_shift, uvsx, uvsy);
3158     }
3159     //FIXME h261 lowres loop filter
3160 }
3161
3162 //FIXME move to dsputil, avg variant, 16x16 version
3163 static inline void put_obmc(uint8_t *dst, uint8_t *src[5], int stride){
3164     int x;
3165     uint8_t * const top   = src[1];
3166     uint8_t * const left  = src[2];
3167     uint8_t * const mid   = src[0];
3168     uint8_t * const right = src[3];
3169     uint8_t * const bottom= src[4];
3170 #define OBMC_FILTER(x, t, l, m, r, b)\
3171     dst[x]= (t*top[x] + l*left[x] + m*mid[x] + r*right[x] + b*bottom[x] + 4)>>3
3172 #define OBMC_FILTER4(x, t, l, m, r, b)\
3173     OBMC_FILTER(x         , t, l, m, r, b);\
3174     OBMC_FILTER(x+1       , t, l, m, r, b);\
3175     OBMC_FILTER(x  +stride, t, l, m, r, b);\
3176     OBMC_FILTER(x+1+stride, t, l, m, r, b);
3177
3178     x=0;
3179     OBMC_FILTER (x  , 2, 2, 4, 0, 0);
3180     OBMC_FILTER (x+1, 2, 1, 5, 0, 0);
3181     OBMC_FILTER4(x+2, 2, 1, 5, 0, 0);
3182     OBMC_FILTER4(x+4, 2, 0, 5, 1, 0);
3183     OBMC_FILTER (x+6, 2, 0, 5, 1, 0);
3184     OBMC_FILTER (x+7, 2, 0, 4, 2, 0);
3185     x+= stride;
3186     OBMC_FILTER (x  , 1, 2, 5, 0, 0);
3187     OBMC_FILTER (x+1, 1, 2, 5, 0, 0);
3188     OBMC_FILTER (x+6, 1, 0, 5, 2, 0);
3189     OBMC_FILTER (x+7, 1, 0, 5, 2, 0);
3190     x+= stride;
3191     OBMC_FILTER4(x  , 1, 2, 5, 0, 0);
3192     OBMC_FILTER4(x+2, 1, 1, 6, 0, 0);
3193     OBMC_FILTER4(x+4, 1, 0, 6, 1, 0);
3194     OBMC_FILTER4(x+6, 1, 0, 5, 2, 0);
3195     x+= 2*stride;
3196     OBMC_FILTER4(x  , 0, 2, 5, 0, 1);
3197     OBMC_FILTER4(x+2, 0, 1, 6, 0, 1);
3198     OBMC_FILTER4(x+4, 0, 0, 6, 1, 1);
3199     OBMC_FILTER4(x+6, 0, 0, 5, 2, 1);
3200     x+= 2*stride;
3201     OBMC_FILTER (x  , 0, 2, 5, 0, 1);
3202     OBMC_FILTER (x+1, 0, 2, 5, 0, 1);
3203     OBMC_FILTER4(x+2, 0, 1, 5, 0, 2);
3204     OBMC_FILTER4(x+4, 0, 0, 5, 1, 2);
3205     OBMC_FILTER (x+6, 0, 0, 5, 2, 1);
3206     OBMC_FILTER (x+7, 0, 0, 5, 2, 1);
3207     x+= stride;
3208     OBMC_FILTER (x  , 0, 2, 4, 0, 2);
3209     OBMC_FILTER (x+1, 0, 1, 5, 0, 2);
3210     OBMC_FILTER (x+6, 0, 0, 5, 1, 2);
3211     OBMC_FILTER (x+7, 0, 0, 4, 2, 2);
3212 }
3213
3214 /* obmc for 1 8x8 luma block */
3215 static inline void obmc_motion(MpegEncContext *s,
3216                                uint8_t *dest, uint8_t *src,
3217                                int src_x, int src_y,
3218                                op_pixels_func *pix_op,
3219                                int16_t mv[5][2]/* mid top left right bottom*/)
3220 #define MID    0
3221 {
3222     int i;
3223     uint8_t *ptr[5];
3224
3225     assert(s->quarter_sample==0);
3226
3227     for(i=0; i<5; i++){
3228         if(i && mv[i][0]==mv[MID][0] && mv[i][1]==mv[MID][1]){
3229             ptr[i]= ptr[MID];
3230         }else{
3231             ptr[i]= s->obmc_scratchpad + 8*(i&1) + s->linesize*8*(i>>1);
3232             hpel_motion(s, ptr[i], src, 0, 0,
3233                         src_x, src_y,
3234                         s->width, s->height, s->linesize,
3235                         s->h_edge_pos, s->v_edge_pos,
3236                         8, 8, pix_op,
3237                         mv[i][0], mv[i][1]);
3238         }
3239     }
3240
3241     put_obmc(dest, ptr, s->linesize);
3242 }
3243
3244 static inline void qpel_motion(MpegEncContext *s,
3245                                uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
3246                                int field_based, int bottom_field, int field_select,
3247                                uint8_t **ref_picture, op_pixels_func (*pix_op)[4],
3248                                qpel_mc_func (*qpix_op)[16],
3249                                int motion_x, int motion_y, int h)
3250 {
3251     uint8_t *ptr_y, *ptr_cb, *ptr_cr;
3252     int dxy, uvdxy, mx, my, src_x, src_y, uvsrc_x, uvsrc_y, v_edge_pos, linesize, uvlinesize;
3253
3254     dxy = ((motion_y & 3) << 2) | (motion_x & 3);
3255     src_x = s->mb_x *  16                 + (motion_x >> 2);
3256     src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2);
3257
3258     v_edge_pos = s->v_edge_pos >> field_based;
3259     linesize = s->linesize << field_based;
3260     uvlinesize = s->uvlinesize << field_based;
3261
3262     if(field_based){
3263         mx= motion_x/2;
3264         my= motion_y>>1;
3265     }else if(s->workaround_bugs&FF_BUG_QPEL_CHROMA2){
3266         static const int rtab[8]= {0,0,1,1,0,0,0,1};
3267         mx= (motion_x>>1) + rtab[motion_x&7];
3268         my= (motion_y>>1) + rtab[motion_y&7];
3269     }else if(s->workaround_bugs&FF_BUG_QPEL_CHROMA){
3270         mx= (motion_x>>1)|(motion_x&1);
3271         my= (motion_y>>1)|(motion_y&1);
3272     }else{
3273         mx= motion_x/2;
3274         my= motion_y/2;
3275     }
3276     mx= (mx>>1)|(mx&1);
3277     my= (my>>1)|(my&1);
3278
3279     uvdxy= (mx&1) | ((my&1)<<1);
3280     mx>>=1;
3281     my>>=1;
3282
3283     uvsrc_x = s->mb_x *  8                 + mx;
3284     uvsrc_y = s->mb_y * (8 >> field_based) + my;
3285
3286     ptr_y  = ref_picture[0] +   src_y *   linesize +   src_x;
3287     ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
3288     ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
3289
3290     if(   (unsigned)src_x > s->h_edge_pos - (motion_x&3) - 16
3291        || (unsigned)src_y >    v_edge_pos - (motion_y&3) - h  ){
3292         ff_emulated_edge_mc(s->edge_emu_buffer, ptr_y, s->linesize, 17, 17+field_based,
3293                          src_x, src_y<<field_based, s->h_edge_pos, s->v_edge_pos);
3294         ptr_y= s->edge_emu_buffer;
3295         if(!(s->flags&CODEC_FLAG_GRAY)){
3296             uint8_t *uvbuf= s->edge_emu_buffer + 18*s->linesize;
3297             ff_emulated_edge_mc(uvbuf, ptr_cb, s->uvlinesize, 9, 9 + field_based,
3298                              uvsrc_x, uvsrc_y<<field_based, s->h_edge_pos>>1, s->v_edge_pos>>1);
3299             ff_emulated_edge_mc(uvbuf + 16, ptr_cr, s->uvlinesize, 9, 9 + field_based,
3300                              uvsrc_x, uvsrc_y<<field_based, s->h_edge_pos>>1, s->v_edge_pos>>1);
3301             ptr_cb= uvbuf;
3302             ptr_cr= uvbuf + 16;
3303         }
3304     }
3305
3306     if(!field_based)
3307         qpix_op[0][dxy](dest_y, ptr_y, linesize);
3308     else{
3309         if(bottom_field){
3310             dest_y += s->linesize;
3311             dest_cb+= s->uvlinesize;
3312             dest_cr+= s->uvlinesize;
3313         }
3314
3315         if(field_select){
3316             ptr_y  += s->linesize;
3317             ptr_cb += s->uvlinesize;
3318             ptr_cr += s->uvlinesize;
3319         }
3320         //damn interlaced mode
3321         //FIXME boundary mirroring is not exactly correct here
3322         qpix_op[1][dxy](dest_y  , ptr_y  , linesize);
3323         qpix_op[1][dxy](dest_y+8, ptr_y+8, linesize);
3324     }
3325     if(!(s->flags&CODEC_FLAG_GRAY)){
3326         pix_op[1][uvdxy](dest_cr, ptr_cr, uvlinesize, h >> 1);
3327         pix_op[1][uvdxy](dest_cb, ptr_cb, uvlinesize, h >> 1);
3328     }
3329 }
3330
3331 inline int ff_h263_round_chroma(int x){
3332     if (x >= 0)
3333         return  (h263_chroma_roundtab[x & 0xf] + ((x >> 3) & ~1));
3334     else {
3335         x = -x;
3336         return -(h263_chroma_roundtab[x & 0xf] + ((x >> 3) & ~1));
3337     }
3338 }
3339
3340 /**
3341  * h263 chorma 4mv motion compensation.
3342  */
3343 static inline void chroma_4mv_motion(MpegEncContext *s,
3344                                      uint8_t *dest_cb, uint8_t *dest_cr,
3345                                      uint8_t **ref_picture,
3346                                      op_pixels_func *pix_op,
3347                                      int mx, int my){
3348     int dxy, emu=0, src_x, src_y, offset;
3349     uint8_t *ptr;
3350
3351     /* In case of 8X8, we construct a single chroma motion vector
3352        with a special rounding */
3353     mx= ff_h263_round_chroma(mx);
3354     my= ff_h263_round_chroma(my);
3355
3356     dxy = ((my & 1) << 1) | (mx & 1);
3357     mx >>= 1;
3358     my >>= 1;
3359
3360     src_x = s->mb_x * 8 + mx;
3361     src_y = s->mb_y * 8 + my;
3362     src_x = clip(src_x, -8, s->width/2);
3363     if (src_x == s->width/2)
3364         dxy &= ~1;
3365     src_y = clip(src_y, -8, s->height/2);
3366     if (src_y == s->height/2)
3367         dxy &= ~2;
3368
3369     offset = (src_y * (s->uvlinesize)) + src_x;
3370     ptr = ref_picture[1] + offset;
3371     if(s->flags&CODEC_FLAG_EMU_EDGE){
3372         if(   (unsigned)src_x > (s->h_edge_pos>>1) - (dxy &1) - 8
3373            || (unsigned)src_y > (s->v_edge_pos>>1) - (dxy>>1) - 8){
3374             ff_emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, 9, 9, src_x, src_y, s->h_edge_pos>>1, s->v_edge_pos>>1);
3375             ptr= s->edge_emu_buffer;
3376             emu=1;
3377         }
3378     }
3379     pix_op[dxy](dest_cb, ptr, s->uvlinesize, 8);
3380
3381     ptr = ref_picture[2] + offset;
3382     if(emu){
3383         ff_emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, 9, 9, src_x, src_y, s->h_edge_pos>>1, s->v_edge_pos>>1);
3384         ptr= s->edge_emu_buffer;
3385     }
3386     pix_op[dxy](dest_cr, ptr, s->uvlinesize, 8);
3387 }
3388
3389 static inline void chroma_4mv_motion_lowres(MpegEncContext *s,
3390                                      uint8_t *dest_cb, uint8_t *dest_cr,
3391                                      uint8_t **ref_picture,
3392                                      h264_chroma_mc_func *pix_op,
3393                                      int mx, int my){
3394     const int lowres= s->avctx->lowres;
3395     const int block_s= 8>>lowres;
3396     const int s_mask= (2<<lowres)-1;
3397     const int h_edge_pos = s->h_edge_pos >> (lowres+1);
3398     const int v_edge_pos = s->v_edge_pos >> (lowres+1);
3399     int emu=0, src_x, src_y, offset, sx, sy;
3400     uint8_t *ptr;
3401
3402     if(s->quarter_sample){
3403         mx/=2;
3404         my/=2;
3405     }
3406
3407     /* In case of 8X8, we construct a single chroma motion vector
3408        with a special rounding */
3409     mx= ff_h263_round_chroma(mx);
3410     my= ff_h263_round_chroma(my);
3411
3412     sx= mx & s_mask;
3413     sy= my & s_mask;
3414     src_x = s->mb_x*block_s + (mx >> (lowres+1));
3415     src_y = s->mb_y*block_s + (my >> (lowres+1));
3416
3417     offset = src_y * s->uvlinesize + src_x;
3418     ptr = ref_picture[1] + offset;
3419     if(s->flags&CODEC_FLAG_EMU_EDGE){
3420         if(   (unsigned)src_x > h_edge_pos - (!!sx) - block_s
3421            || (unsigned)src_y > v_edge_pos - (!!sy) - block_s){
3422             ff_emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, 9, 9, src_x, src_y, h_edge_pos, v_edge_pos);
3423             ptr= s->edge_emu_buffer;
3424             emu=1;
3425         }
3426     }
3427     sx <<= 2 - lowres;
3428     sy <<= 2 - lowres;
3429     pix_op[lowres](dest_cb, ptr, s->uvlinesize, block_s, sx, sy);
3430
3431     ptr = ref_picture[2] + offset;
3432     if(emu){
3433         ff_emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, 9, 9, src_x, src_y, h_edge_pos, v_edge_pos);
3434         ptr= s->edge_emu_buffer;
3435     }
3436     pix_op[lowres](dest_cr, ptr, s->uvlinesize, block_s, sx, sy);
3437 }
3438
3439 static inline void prefetch_motion(MpegEncContext *s, uint8_t **pix, int dir){
3440     /* fetch pixels for estimated mv 4 macroblocks ahead
3441      * optimized for 64byte cache lines */
3442     const int shift = s->quarter_sample ? 2 : 1;
3443     const int mx= (s->mv[dir][0][0]>>shift) + 16*s->mb_x + 8;
3444     const int my= (s->mv[dir][0][1]>>shift) + 16*s->mb_y;
3445     int off= mx + (my + (s->mb_x&3)*4)*s->linesize + 64;
3446     s->dsp.prefetch(pix[0]+off, s->linesize, 4);
3447     off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
3448     s->dsp.prefetch(pix[1]+off, pix[2]-pix[1], 2);
3449 }
3450
3451 /**
3452  * motion compensation of a single macroblock
3453  * @param s context
3454  * @param dest_y luma destination pointer
3455  * @param dest_cb chroma cb/u destination pointer
3456  * @param dest_cr chroma cr/v destination pointer
3457  * @param dir direction (0->forward, 1->backward)
3458  * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
3459  * @param pic_op halfpel motion compensation function (average or put normally)
3460  * @param pic_op qpel motion compensation function (average or put normally)
3461  * the motion vectors are taken from s->mv and the MV type from s->mv_type
3462  */
3463 static inline void MPV_motion(MpegEncContext *s,
3464                               uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
3465                               int dir, uint8_t **ref_picture,
3466                               op_pixels_func (*pix_op)[4], qpel_mc_func (*qpix_op)[16])
3467 {
3468     int dxy, mx, my, src_x, src_y, motion_x, motion_y;
3469     int mb_x, mb_y, i;
3470     uint8_t *ptr, *dest;
3471
3472     mb_x = s->mb_x;
3473     mb_y = s->mb_y;
3474
3475     prefetch_motion(s, ref_picture, dir);
3476
3477     if(s->obmc && s->pict_type != B_TYPE){
3478         int16_t mv_cache[4][4][2];
3479         const int xy= s->mb_x + s->mb_y*s->mb_stride;
3480         const int mot_stride= s->b8_stride;
3481         const int mot_xy= mb_x*2 + mb_y*2*mot_stride;
3482
3483         assert(!s->mb_skipped);
3484
3485         memcpy(mv_cache[1][1], s->current_picture.motion_val[0][mot_xy           ], sizeof(int16_t)*4);
3486         memcpy(mv_cache[2][1], s->current_picture.motion_val[0][mot_xy+mot_stride], sizeof(int16_t)*4);
3487         memcpy(mv_cache[3][1], s->current_picture.motion_val[0][mot_xy+mot_stride], sizeof(int16_t)*4);
3488
3489         if(mb_y==0 || IS_INTRA(s->current_picture.mb_type[xy-s->mb_stride])){
3490             memcpy(mv_cache[0][1], mv_cache[1][1], sizeof(int16_t)*4);
3491         }else{
3492             memcpy(mv_cache[0][1], s->current_picture.motion_val[0][mot_xy-mot_stride], sizeof(int16_t)*4);
3493         }
3494
3495         if(mb_x==0 || IS_INTRA(s->current_picture.mb_type[xy-1])){
3496             *(int32_t*)mv_cache[1][0]= *(int32_t*)mv_cache[1][1];
3497             *(int32_t*)mv_cache[2][0]= *(int32_t*)mv_cache[2][1];
3498         }else{
3499             *(int32_t*)mv_cache[1][0]= *(int32_t*)s->current_picture.motion_val[0][mot_xy-1];
3500             *(int32_t*)mv_cache[2][0]= *(int32_t*)s->current_picture.motion_val[0][mot_xy-1+mot_stride];
3501         }
3502
3503         if(mb_x+1>=s->mb_width || IS_INTRA(s->current_picture.mb_type[xy+1])){
3504             *(int32_t*)mv_cache[1][3]= *(int32_t*)mv_cache[1][2];
3505             *(int32_t*)mv_cache[2][3]= *(int32_t*)mv_cache[2][2];
3506         }else{
3507             *(int32_t*)mv_cache[1][3]= *(int32_t*)s->current_picture.motion_val[0][mot_xy+2];
3508             *(int32_t*)mv_cache[2][3]= *(int32_t*)s->current_picture.motion_val[0][mot_xy+2+mot_stride];
3509         }
3510
3511         mx = 0;
3512         my = 0;
3513         for(i=0;i<4;i++) {
3514             const int x= (i&1)+1;
3515             const int y= (i>>1)+1;
3516             int16_t mv[5][2]= {
3517                 {mv_cache[y][x  ][0], mv_cache[y][x  ][1]},
3518                 {mv_cache[y-1][x][0], mv_cache[y-1][x][1]},
3519                 {mv_cache[y][x-1][0], mv_cache[y][x-1][1]},
3520                 {mv_cache[y][x+1][0], mv_cache[y][x+1][1]},
3521                 {mv_cache[y+1][x][0], mv_cache[y+1][x][1]}};
3522             //FIXME cleanup
3523             obmc_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
3524                         ref_picture[0],
3525                         mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >>1) * 8,
3526                         pix_op[1],
3527                         mv);
3528
3529             mx += mv[0][0];
3530             my += mv[0][1];
3531         }
3532         if(!(s->flags&CODEC_FLAG_GRAY))
3533             chroma_4mv_motion(s, dest_cb, dest_cr, ref_picture, pix_op[1], mx, my);
3534
3535         return;
3536     }
3537
3538     switch(s->mv_type) {
3539     case MV_TYPE_16X16:
3540         if(s->mcsel){
3541             if(s->real_sprite_warping_points==1){
3542                 gmc1_motion(s, dest_y, dest_cb, dest_cr,
3543                             ref_picture);
3544             }else{
3545                 gmc_motion(s, dest_y, dest_cb, dest_cr,
3546                             ref_picture);
3547             }
3548         }else if(s->quarter_sample){
3549             qpel_motion(s, dest_y, dest_cb, dest_cr,
3550                         0, 0, 0,
3551                         ref_picture, pix_op, qpix_op,
3552                         s->mv[dir][0][0], s->mv[dir][0][1], 16);
3553         }else if(s->mspel){
3554             ff_mspel_motion(s, dest_y, dest_cb, dest_cr,
3555                         ref_picture, pix_op,
3556                         s->mv[dir][0][0], s->mv[dir][0][1], 16);
3557         }else
3558         {
3559             mpeg_motion(s, dest_y, dest_cb, dest_cr,
3560                         0, 0, 0,
3561                         ref_picture, pix_op,
3562                         s->mv[dir][0][0], s->mv[dir][0][1], 16);
3563         }
3564         break;
3565     case MV_TYPE_8X8:
3566         mx = 0;
3567         my = 0;
3568         if(s->quarter_sample){
3569             for(i=0;i<4;i++) {
3570                 motion_x = s->mv[dir][i][0];
3571                 motion_y = s->mv[dir][i][1];
3572
3573                 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
3574                 src_x = mb_x * 16 + (motion_x >> 2) + (i & 1) * 8;
3575                 src_y = mb_y * 16 + (motion_y >> 2) + (i >>1) * 8;
3576
3577                 /* WARNING: do no forget half pels */
3578                 src_x = clip(src_x, -16, s->width);
3579                 if (src_x == s->width)
3580                     dxy &= ~3;
3581                 src_y = clip(src_y, -16, s->height);
3582                 if (src_y == s->height)
3583                     dxy &= ~12;
3584
3585                 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x);
3586                 if(s->flags&CODEC_FLAG_EMU_EDGE){
3587                     if(   (unsigned)src_x > s->h_edge_pos - (motion_x&3) - 8
3588                        || (unsigned)src_y > s->v_edge_pos - (motion_y&3) - 8 ){
3589                         ff_emulated_edge_mc(s->edge_emu_buffer, ptr, s->linesize, 9, 9, src_x, src_y, s->h_edge_pos, s->v_edge_pos);
3590                         ptr= s->edge_emu_buffer;
3591                     }
3592                 }
3593                 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize;
3594                 qpix_op[1][dxy](dest, ptr, s->linesize);
3595
3596                 mx += s->mv[dir][i][0]/2;
3597                 my += s->mv[dir][i][1]/2;
3598             }
3599         }else{
3600             for(i=0;i<4;i++) {
3601                 hpel_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
3602                             ref_picture[0], 0, 0,
3603                             mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >>1) * 8,
3604                             s->width, s->height, s->linesize,
3605                             s->h_edge_pos, s->v_edge_pos,
3606                             8, 8, pix_op[1],
3607                             s->mv[dir][i][0], s->mv[dir][i][1]);
3608
3609                 mx += s->mv[dir][i][0];
3610                 my += s->mv[dir][i][1];
3611             }
3612         }
3613
3614         if(!(s->flags&CODEC_FLAG_GRAY))
3615             chroma_4mv_motion(s, dest_cb, dest_cr, ref_picture, pix_op[1], mx, my);
3616         break;
3617     case MV_TYPE_FIELD:
3618         if (s->picture_structure == PICT_FRAME) {
3619             if(s->quarter_sample){
3620                 for(i=0; i<2; i++){
3621                     qpel_motion(s, dest_y, dest_cb, dest_cr,
3622                                 1, i, s->field_select[dir][i],
3623                                 ref_picture, pix_op, qpix_op,
3624                                 s->mv[dir][i][0], s->mv[dir][i][1], 8);
3625                 }
3626             }else{
3627                 /* top field */
3628                 mpeg_motion(s, dest_y, dest_cb, dest_cr,
3629                             1, 0, s->field_select[dir][0],
3630                             ref_picture, pix_op,
3631                             s->mv[dir][0][0], s->mv[dir][0][1], 8);
3632                 /* bottom field */
3633                 mpeg_motion(s, dest_y, dest_cb, dest_cr,
3634                             1, 1, s->field_select[dir][1],
3635                             ref_picture, pix_op,
3636                             s->mv[dir][1][0], s->mv[dir][1][1], 8);
3637             }
3638         } else {
3639             if(s->picture_structure != s->field_select[dir][0] + 1 && s->pict_type != B_TYPE && !s->first_field){
3640                 ref_picture= s->current_picture_ptr->data;
3641             }
3642
3643             mpeg_motion(s, dest_y, dest_cb, dest_cr,
3644                         0, 0, s->field_select[dir][0],
3645                         ref_picture, pix_op,
3646                         s->mv[dir][0][0], s->mv[dir][0][1], 16);
3647         }
3648         break;
3649     case MV_TYPE_16X8:
3650         for(i=0; i<2; i++){
3651             uint8_t ** ref2picture;
3652
3653             if(s->picture_structure == s->field_select[dir][i] + 1 || s->pict_type == B_TYPE || s->first_field){
3654                 ref2picture= ref_picture;
3655             }else{
3656                 ref2picture= s->current_picture_ptr->data;
3657             }
3658
3659             mpeg_motion(s, dest_y, dest_cb, dest_cr,
3660                         0, 0, s->field_select[dir][i],
3661                         ref2picture, pix_op,
3662                         s->mv[dir][i][0], s->mv[dir][i][1] + 16*i, 8);
3663
3664             dest_y += 16*s->linesize;
3665             dest_cb+= (16>>s->chroma_y_shift)*s->uvlinesize;
3666             dest_cr+= (16>>s->chroma_y_shift)*s->uvlinesize;
3667         }
3668         break;
3669     case MV_TYPE_DMV:
3670         if(s->picture_structure == PICT_FRAME){
3671             for(i=0; i<2; i++){
3672                 int j;
3673                 for(j=0; j<2; j++){
3674                     mpeg_motion(s, dest_y, dest_cb, dest_cr,
3675                                 1, j, j^i,
3676                                 ref_picture, pix_op,
3677                                 s->mv[dir][2*i + j][0], s->mv[dir][2*i + j][1], 8);
3678                 }
3679                 pix_op = s->dsp.avg_pixels_tab;
3680             }
3681         }else{
3682             for(i=0; i<2; i++){
3683                 mpeg_motion(s, dest_y, dest_cb, dest_cr,
3684                             0, 0, s->picture_structure != i+1,
3685                             ref_picture, pix_op,
3686                             s->mv[dir][2*i][0],s->mv[dir][2*i][1],16);
3687
3688                 // after put we make avg of the same block
3689                 pix_op=s->dsp.avg_pixels_tab;
3690
3691                 //opposite parity is always in the same frame if this is second field
3692                 if(!s->first_field){
3693                     ref_picture = s->current_picture_ptr->data;
3694                 }
3695             }
3696         }
3697     break;
3698     default: assert(0);
3699     }
3700 }
3701
3702 /**
3703  * motion compensation of a single macroblock
3704  * @param s context
3705  * @param dest_y luma destination pointer
3706  * @param dest_cb chroma cb/u destination pointer
3707  * @param dest_cr chroma cr/v destination pointer
3708  * @param dir direction (0->forward, 1->backward)
3709  * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
3710  * @param pic_op halfpel motion compensation function (average or put normally)
3711  * the motion vectors are taken from s->mv and the MV type from s->mv_type
3712  */
3713 static inline void MPV_motion_lowres(MpegEncContext *s,
3714                               uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
3715                               int dir, uint8_t **ref_picture,
3716                               h264_chroma_mc_func *pix_op)
3717 {
3718     int mx, my;
3719     int mb_x, mb_y, i;
3720     const int lowres= s->avctx->lowres;
3721     const int block_s= 8>>lowres;
3722
3723     mb_x = s->mb_x;
3724     mb_y = s->mb_y;
3725
3726     switch(s->mv_type) {
3727     case MV_TYPE_16X16:
3728         mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
3729                     0, 0, 0,
3730                     ref_picture, pix_op,
3731                     s->mv[dir][0][0], s->mv[dir][0][1], 2*block_s);
3732         break;
3733     case MV_TYPE_8X8:
3734         mx = 0;
3735         my = 0;
3736             for(i=0;i<4;i++) {
3737                 hpel_motion_lowres(s, dest_y + ((i & 1) + (i >> 1) * s->linesize)*block_s,
3738                             ref_picture[0], 0, 0,
3739                             (2*mb_x + (i & 1))*block_s, (2*mb_y + (i >>1))*block_s,
3740                             s->width, s->height, s->linesize,
3741                             s->h_edge_pos >> lowres, s->v_edge_pos >> lowres,
3742                             block_s, block_s, pix_op,
3743                             s->mv[dir][i][0], s->mv[dir][i][1]);
3744
3745                 mx += s->mv[dir][i][0];
3746                 my += s->mv[dir][i][1];
3747             }
3748
3749         if(!(s->flags&CODEC_FLAG_GRAY))
3750             chroma_4mv_motion_lowres(s, dest_cb, dest_cr, ref_picture, pix_op, mx, my);
3751         break;
3752     case MV_TYPE_FIELD:
3753         if (s->picture_structure == PICT_FRAME) {
3754             /* top field */
3755             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
3756                         1, 0, s->field_select[dir][0],
3757                         ref_picture, pix_op,
3758                         s->mv[dir][0][0], s->mv[dir][0][1], block_s);
3759             /* bottom field */
3760             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
3761                         1, 1, s->field_select[dir][1],
3762                         ref_picture, pix_op,
3763                         s->mv[dir][1][0], s->mv[dir][1][1], block_s);
3764         } else {
3765             if(s->picture_structure != s->field_select[dir][0] + 1 && s->pict_type != B_TYPE && !s->first_field){
3766                 ref_picture= s->current_picture_ptr->data;
3767             }
3768
3769             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
3770                         0, 0, s->field_select[dir][0],
3771                         ref_picture, pix_op,
3772                         s->mv[dir][0][0], s->mv[dir][0][1], 2*block_s);
3773         }
3774         break;
3775     case MV_TYPE_16X8:
3776         for(i=0; i<2; i++){
3777             uint8_t ** ref2picture;
3778
3779             if(s->picture_structure == s->field_select[dir][i] + 1 || s->pict_type == B_TYPE || s->first_field){
3780                 ref2picture= ref_picture;
3781             }else{
3782                 ref2picture= s->current_picture_ptr->data;
3783             }
3784
3785             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
3786                         0, 0, s->field_select[dir][i],
3787                         ref2picture, pix_op,
3788                         s->mv[dir][i][0], s->mv[dir][i][1] + 2*block_s*i, block_s);
3789
3790             dest_y += 2*block_s*s->linesize;
3791             dest_cb+= (2*block_s>>s->chroma_y_shift)*s->uvlinesize;
3792             dest_cr+= (2*block_s>>s->chroma_y_shift)*s->uvlinesize;
3793         }
3794         break;
3795     case MV_TYPE_DMV:
3796         if(s->picture_structure == PICT_FRAME){
3797             for(i=0; i<2; i++){
3798                 int j;
3799                 for(j=0; j<2; j++){
3800                     mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
3801                                 1, j, j^i,
3802                                 ref_picture, pix_op,
3803                                 s->mv[dir][2*i + j][0], s->mv[dir][2*i + j][1], block_s);
3804                 }
3805                 pix_op = s->dsp.avg_h264_chroma_pixels_tab;
3806             }
3807         }else{
3808             for(i=0; i<2; i++){
3809                 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
3810                             0, 0, s->picture_structure != i+1,
3811                             ref_picture, pix_op,
3812                             s->mv[dir][2*i][0],s->mv[dir][2*i][1],2*block_s);
3813
3814                 // after put we make avg of the same block
3815                 pix_op = s->dsp.avg_h264_chroma_pixels_tab;
3816
3817                 //opposite parity is always in the same frame if this is second field
3818                 if(!s->first_field){
3819                     ref_picture = s->current_picture_ptr->data;
3820                 }
3821             }
3822         }
3823     break;
3824     default: assert(0);
3825     }
3826 }
3827
3828 /* put block[] to dest[] */
3829 static inline void put_dct(MpegEncContext *s,
3830                            DCTELEM *block, int i, uint8_t *dest, int line_size, int qscale)
3831 {
3832     s->dct_unquantize_intra(s, block, i, qscale);
3833     s->dsp.idct_put (dest, line_size, block);
3834 }
3835
3836 /* add block[] to dest[] */
3837 static inline void add_dct(MpegEncContext *s,
3838                            DCTELEM *block, int i, uint8_t *dest, int line_size)
3839 {
3840     if (s->block_last_index[i] >= 0) {
3841         s->dsp.idct_add (dest, line_size, block);
3842     }
3843 }
3844
3845 static inline void add_dequant_dct(MpegEncContext *s,
3846                            DCTELEM *block, int i, uint8_t *dest, int line_size, int qscale)
3847 {
3848     if (s->block_last_index[i] >= 0) {
3849         s->dct_unquantize_inter(s, block, i, qscale);
3850
3851         s->dsp.idct_add (dest, line_size, block);
3852     }
3853 }
3854
3855 /**
3856  * cleans dc, ac, coded_block for the current non intra MB
3857  */
3858 void ff_clean_intra_table_entries(MpegEncContext *s)
3859 {
3860     int wrap = s->b8_stride;
3861     int xy = s->block_index[0];
3862
3863     s->dc_val[0][xy           ] =
3864     s->dc_val[0][xy + 1       ] =
3865     s->dc_val[0][xy     + wrap] =
3866     s->dc_val[0][xy + 1 + wrap] = 1024;
3867     /* ac pred */
3868     memset(s->ac_val[0][xy       ], 0, 32 * sizeof(int16_t));
3869     memset(s->ac_val[0][xy + wrap], 0, 32 * sizeof(int16_t));
3870     if (s->msmpeg4_version>=3) {
3871         s->coded_block[xy           ] =
3872         s->coded_block[xy + 1       ] =
3873         s->coded_block[xy     + wrap] =
3874         s->coded_block[xy + 1 + wrap] = 0;
3875     }
3876     /* chroma */
3877     wrap = s->mb_stride;
3878     xy = s->mb_x + s->mb_y * wrap;
3879     s->dc_val[1][xy] =
3880     s->dc_val[2][xy] = 1024;
3881     /* ac pred */
3882     memset(s->ac_val[1][xy], 0, 16 * sizeof(int16_t));
3883     memset(s->ac_val[2][xy], 0, 16 * sizeof(int16_t));
3884
3885     s->mbintra_table[xy]= 0;
3886 }
3887
3888 /* generic function called after a macroblock has been parsed by the
3889    decoder or after it has been encoded by the encoder.
3890
3891    Important variables used:
3892    s->mb_intra : true if intra macroblock
3893    s->mv_dir   : motion vector direction
3894    s->mv_type  : motion vector type
3895    s->mv       : motion vector
3896    s->interlaced_dct : true if interlaced dct used (mpeg2)
3897  */
3898 static av_always_inline void MPV_decode_mb_internal(MpegEncContext *s, DCTELEM block[12][64], int lowres_flag)
3899 {
3900     int mb_x, mb_y;
3901     const int mb_xy = s->mb_y * s->mb_stride + s->mb_x;
3902 #ifdef HAVE_XVMC
3903     if(s->avctx->xvmc_acceleration){
3904         XVMC_decode_mb(s);//xvmc uses pblocks
3905         return;
3906     }
3907 #endif
3908
3909     mb_x = s->mb_x;
3910     mb_y = s->mb_y;
3911
3912     if(s->avctx->debug&FF_DEBUG_DCT_COEFF) {
3913        /* save DCT coefficients */
3914        int i,j;
3915        DCTELEM *dct = &s->current_picture.dct_coeff[mb_xy*64*6];
3916        for(i=0; i<6; i++)
3917            for(j=0; j<64; j++)
3918                *dct++ = block[i][s->dsp.idct_permutation[j]];
3919     }
3920
3921     s->current_picture.qscale_table[mb_xy]= s->qscale;
3922
3923     /* update DC predictors for P macroblocks */
3924     if (!s->mb_intra) {
3925         if (s->h263_pred || s->h263_aic) {
3926             if(s->mbintra_table[mb_xy])
3927                 ff_clean_intra_table_entries(s);
3928         } else {
3929             s->last_dc[0] =
3930             s->last_dc[1] =
3931             s->last_dc[2] = 128 << s->intra_dc_precision;
3932         }
3933     }
3934     else if (s->h263_pred || s->h263_aic)
3935         s->mbintra_table[mb_xy]=1;
3936
3937     if ((s->flags&CODEC_FLAG_PSNR) || !(s->encoding && (s->intra_only || s->pict_type==B_TYPE) && s->avctx->mb_decision != FF_MB_DECISION_RD)) { //FIXME precalc
3938         uint8_t *dest_y, *dest_cb, *dest_cr;
3939         int dct_linesize, dct_offset;
3940         op_pixels_func (*op_pix)[4];
3941         qpel_mc_func (*op_qpix)[16];
3942         const int linesize= s->current_picture.linesize[0]; //not s->linesize as this would be wrong for field pics
3943         const int uvlinesize= s->current_picture.linesize[1];
3944         const int readable= s->pict_type != B_TYPE || s->encoding || s->avctx->draw_horiz_band || lowres_flag;
3945         const int block_size= lowres_flag ? 8>>s->avctx->lowres : 8;
3946
3947         /* avoid copy if macroblock skipped in last frame too */
3948         /* skip only during decoding as we might trash the buffers during encoding a bit */
3949         if(!s->encoding){
3950             uint8_t *mbskip_ptr = &s->mbskip_table[mb_xy];
3951             const int age= s->current_picture.age;
3952
3953             assert(age);
3954
3955             if (s->mb_skipped) {
3956                 s->mb_skipped= 0;
3957                 assert(s->pict_type!=I_TYPE);
3958
3959                 (*mbskip_ptr) ++; /* indicate that this time we skipped it */
3960                 if(*mbskip_ptr >99) *mbskip_ptr= 99;
3961
3962                 /* if previous was skipped too, then nothing to do !  */
3963                 if (*mbskip_ptr >= age && s->current_picture.reference){
3964                     return;
3965                 }
3966             } else if(!s->current_picture.reference){
3967                 (*mbskip_ptr) ++; /* increase counter so the age can be compared cleanly */
3968                 if(*mbskip_ptr >99) *mbskip_ptr= 99;
3969             } else{
3970                 *mbskip_ptr = 0; /* not skipped */
3971             }
3972         }
3973
3974         dct_linesize = linesize << s->interlaced_dct;
3975         dct_offset =(s->interlaced_dct)? linesize : linesize*block_size;
3976
3977         if(readable){
3978             dest_y=  s->dest[0];
3979             dest_cb= s->dest[1];
3980             dest_cr= s->dest[2];
3981         }else{
3982             dest_y = s->b_scratchpad;
3983             dest_cb= s->b_scratchpad+16*linesize;
3984             dest_cr= s->b_scratchpad+32*linesize;
3985         }
3986
3987         if (!s->mb_intra) {
3988             /* motion handling */
3989             /* decoding or more than one mb_type (MC was already done otherwise) */
3990             if(!s->encoding){
3991                 if(lowres_flag){
3992                     h264_chroma_mc_func *op_pix = s->dsp.put_h264_chroma_pixels_tab;
3993
3994                     if (s->mv_dir & MV_DIR_FORWARD) {
3995                         MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.data, op_pix);
3996                         op_pix = s->dsp.avg_h264_chroma_pixels_tab;
3997                     }
3998                     if (s->mv_dir & MV_DIR_BACKWARD) {
3999                         MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.data, op_pix);
4000                     }
4001                 }else{
4002                     op_qpix= s->me.qpel_put;
4003                     if ((!s->no_rounding) || s->pict_type==B_TYPE){
4004                         op_pix = s->dsp.put_pixels_tab;
4005                     }else{
4006                         op_pix = s->dsp.put_no_rnd_pixels_tab;
4007                     }
4008                     if (s->mv_dir & MV_DIR_FORWARD) {
4009                         MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.data, op_pix, op_qpix);
4010                         op_pix = s->dsp.avg_pixels_tab;
4011                         op_qpix= s->me.qpel_avg;
4012                     }
4013                     if (s->mv_dir & MV_DIR_BACKWARD) {
4014                         MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.data, op_pix, op_qpix);
4015                     }
4016                 }
4017             }
4018
4019             /* skip dequant / idct if we are really late ;) */
4020             if(s->hurry_up>1) goto skip_idct;
4021             if(s->avctx->skip_idct){
4022                 if(  (s->avctx->skip_idct >= AVDISCARD_NONREF && s->pict_type == B_TYPE)
4023                    ||(s->avctx->skip_idct >= AVDISCARD_NONKEY && s->pict_type != I_TYPE)
4024                    || s->avctx->skip_idct >= AVDISCARD_ALL)
4025                     goto skip_idct;
4026             }
4027
4028             /* add dct residue */
4029             if(s->encoding || !(   s->h263_msmpeg4 || s->codec_id==CODEC_ID_MPEG1VIDEO || s->codec_id==CODEC_ID_MPEG2VIDEO
4030                                 || (s->codec_id==CODEC_ID_MPEG4 && !s->mpeg_quant))){
4031                 add_dequant_dct(s, block[0], 0, dest_y                          , dct_linesize, s->qscale);
4032                 add_dequant_dct(s, block[1], 1, dest_y              + block_size, dct_linesize, s->qscale);
4033                 add_dequant_dct(s, block[2], 2, dest_y + dct_offset             , dct_linesize, s->qscale);
4034                 add_dequant_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize, s->qscale);
4035
4036                 if(!(s->flags&CODEC_FLAG_GRAY)){
4037                     if (s->chroma_y_shift){
4038                         add_dequant_dct(s, block[4], 4, dest_cb, uvlinesize, s->chroma_qscale);
4039                         add_dequant_dct(s, block[5], 5, dest_cr, uvlinesize, s->chroma_qscale);
4040                     }else{
4041                         dct_linesize >>= 1;
4042                         dct_offset >>=1;
4043                         add_dequant_dct(s, block[4], 4, dest_cb,              dct_linesize, s->chroma_qscale);
4044                         add_dequant_dct(s, block[5], 5, dest_cr,              dct_linesize, s->chroma_qscale);
4045                         add_dequant_dct(s, block[6], 6, dest_cb + dct_offset, dct_linesize, s->chroma_qscale);
4046                         add_dequant_dct(s, block[7], 7, dest_cr + dct_offset, dct_linesize, s->chroma_qscale);
4047                     }
4048                 }
4049             } else if(s->codec_id != CODEC_ID_WMV2){
4050                 add_dct(s, block[0], 0, dest_y                          , dct_linesize);
4051                 add_dct(s, block[1], 1, dest_y              + block_size, dct_linesize);
4052                 add_dct(s, block[2], 2, dest_y + dct_offset             , dct_linesize);
4053                 add_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize);
4054
4055                 if(!(s->flags&CODEC_FLAG_GRAY)){
4056                     if(s->chroma_y_shift){//Chroma420
4057                         add_dct(s, block[4], 4, dest_cb, uvlinesize);
4058                         add_dct(s, block[5], 5, dest_cr, uvlinesize);
4059                     }else{
4060                         //chroma422
4061                         dct_linesize = uvlinesize << s->interlaced_dct;
4062                         dct_offset =(s->interlaced_dct)? uvlinesize : uvlinesize*8;
4063
4064                         add_dct(s, block[4], 4, dest_cb, dct_linesize);
4065                         add_dct(s, block[5], 5, dest_cr, dct_linesize);
4066                         add_dct(s, block[6], 6, dest_cb+dct_offset, dct_linesize);
4067                         add_dct(s, block[7], 7, dest_cr+dct_offset, dct_linesize);
4068                         if(!s->chroma_x_shift){//Chroma444
4069                             add_dct(s, block[8], 8, dest_cb+8, dct_linesize);
4070                             add_dct(s, block[9], 9, dest_cr+8, dct_linesize);
4071                             add_dct(s, block[10], 10, dest_cb+8+dct_offset, dct_linesize);
4072                             add_dct(s, block[11], 11, dest_cr+8+dct_offset, dct_linesize);
4073                         }
4074                     }
4075                 }//fi gray
4076             }
4077             else{
4078                 ff_wmv2_add_mb(s, block, dest_y, dest_cb, dest_cr);
4079             }
4080         } else {
4081             /* dct only in intra block */
4082             if(s->encoding || !(s->codec_id==CODEC_ID_MPEG1VIDEO || s->codec_id==CODEC_ID_MPEG2VIDEO)){
4083                 put_dct(s, block[0], 0, dest_y                          , dct_linesize, s->qscale);
4084                 put_dct(s, block[1], 1, dest_y              + block_size, dct_linesize, s->qscale);
4085                 put_dct(s, block[2], 2, dest_y + dct_offset             , dct_linesize, s->qscale);
4086                 put_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize, s->qscale);
4087
4088                 if(!(s->flags&CODEC_FLAG_GRAY)){
4089                     if(s->chroma_y_shift){
4090                         put_dct(s, block[4], 4, dest_cb, uvlinesize, s->chroma_qscale);
4091                         put_dct(s, block[5], 5, dest_cr, uvlinesize, s->chroma_qscale);
4092                     }else{
4093                         dct_offset >>=1;
4094                         dct_linesize >>=1;
4095                         put_dct(s, block[4], 4, dest_cb,              dct_linesize, s->chroma_qscale);
4096                         put_dct(s, block[5], 5, dest_cr,              dct_linesize, s->chroma_qscale);
4097                         put_dct(s, block[6], 6, dest_cb + dct_offset, dct_linesize, s->chroma_qscale);
4098                         put_dct(s, block[7], 7, dest_cr + dct_offset, dct_linesize, s->chroma_qscale);
4099                     }
4100                 }
4101             }else{
4102                 s->dsp.idct_put(dest_y                          , dct_linesize, block[0]);
4103                 s->dsp.idct_put(dest_y              + block_size, dct_linesize, block[1]);
4104                 s->dsp.idct_put(dest_y + dct_offset             , dct_linesize, block[2]);
4105                 s->dsp.idct_put(dest_y + dct_offset + block_size, dct_linesize, block[3]);
4106
4107                 if(!(s->flags&CODEC_FLAG_GRAY)){
4108                     if(s->chroma_y_shift){
4109                         s->dsp.idct_put(dest_cb, uvlinesize, block[4]);
4110                         s->dsp.idct_put(dest_cr, uvlinesize, block[5]);
4111                     }else{
4112
4113                         dct_linesize = uvlinesize << s->interlaced_dct;
4114                         dct_offset =(s->interlaced_dct)? uvlinesize : uvlinesize*8;
4115
4116                         s->dsp.idct_put(dest_cb,              dct_linesize, block[4]);
4117                         s->dsp.idct_put(dest_cr,              dct_linesize, block[5]);
4118                         s->dsp.idct_put(dest_cb + dct_offset, dct_linesize, block[6]);
4119                         s->dsp.idct_put(dest_cr + dct_offset, dct_linesize, block[7]);
4120                         if(!s->chroma_x_shift){//Chroma444
4121                             s->dsp.idct_put(dest_cb + 8,              dct_linesize, block[8]);
4122                             s->dsp.idct_put(dest_cr + 8,              dct_linesize, block[9]);
4123                             s->dsp.idct_put(dest_cb + 8 + dct_offset, dct_linesize, block[10]);
4124                             s->dsp.idct_put(dest_cr + 8 + dct_offset, dct_linesize, block[11]);
4125                         }
4126                     }
4127                 }//gray
4128             }
4129         }
4130 skip_idct:
4131         if(!readable){
4132             s->dsp.put_pixels_tab[0][0](s->dest[0], dest_y ,   linesize,16);
4133             s->dsp.put_pixels_tab[s->chroma_x_shift][0](s->dest[1], dest_cb, uvlinesize,16 >> s->chroma_y_shift);
4134             s->dsp.put_pixels_tab[s->chroma_x_shift][0](s->dest[2], dest_cr, uvlinesize,16 >> s->chroma_y_shift);
4135         }
4136     }
4137 }
4138
4139 void MPV_decode_mb(MpegEncContext *s, DCTELEM block[12][64]){
4140     if(s->avctx->lowres) MPV_decode_mb_internal(s, block, 1);
4141     else                  MPV_decode_mb_internal(s, block, 0);
4142 }
4143
4144 #ifdef CONFIG_ENCODERS
4145
4146 static inline void dct_single_coeff_elimination(MpegEncContext *s, int n, int threshold)
4147 {
4148     static const char tab[64]=
4149         {3,2,2,1,1,1,1,1,
4150          1,1,1,1,1,1,1,1,
4151          1,1,1,1,1,1,1,1,
4152          0,0,0,0,0,0,0,0,
4153          0,0,0,0,0,0,0,0,
4154          0,0,0,0,0,0,0,0,
4155          0,0,0,0,0,0,0,0,
4156          0,0,0,0,0,0,0,0};
4157     int score=0;
4158     int run=0;
4159     int i;
4160     DCTELEM *block= s->block[n];
4161     const int last_index= s->block_last_index[n];
4162     int skip_dc;
4163
4164     if(threshold<0){
4165         skip_dc=0;
4166         threshold= -threshold;
4167     }else
4168         skip_dc=1;
4169
4170     /* are all which we could set to zero are allready zero? */
4171     if(last_index<=skip_dc - 1) return;
4172
4173     for(i=0; i<=last_index; i++){
4174         const int j = s->intra_scantable.permutated[i];
4175         const int level = FFABS(block[j]);
4176         if(level==1){
4177             if(skip_dc && i==0) continue;
4178             score+= tab[run];
4179             run=0;
4180         }else if(level>1){
4181             return;
4182         }else{
4183             run++;
4184         }
4185     }
4186     if(score >= threshold) return;
4187     for(i=skip_dc; i<=last_index; i++){
4188         const int j = s->intra_scantable.permutated[i];
4189         block[j]=0;
4190     }
4191     if(block[0]) s->block_last_index[n]= 0;
4192     else         s->block_last_index[n]= -1;
4193 }
4194
4195 static inline void clip_coeffs(MpegEncContext *s, DCTELEM *block, int last_index)
4196 {
4197     int i;
4198     const int maxlevel= s->max_qcoeff;
4199     const int minlevel= s->min_qcoeff;
4200     int overflow=0;
4201
4202     if(s->mb_intra){
4203         i=1; //skip clipping of intra dc
4204     }else
4205         i=0;
4206
4207     for(;i<=last_index; i++){
4208         const int j= s->intra_scantable.permutated[i];
4209         int level = block[j];
4210
4211         if     (level>maxlevel){
4212             level=maxlevel;
4213             overflow++;
4214         }else if(level<minlevel){
4215             level=minlevel;
4216             overflow++;
4217         }
4218
4219         block[j]= level;
4220     }
4221
4222     if(overflow && s->avctx->mb_decision == FF_MB_DECISION_SIMPLE)
4223         av_log(s->avctx, AV_LOG_INFO, "warning, clipping %d dct coefficients to %d..%d\n", overflow, minlevel, maxlevel);
4224 }
4225
4226 #endif //CONFIG_ENCODERS
4227
4228 /**
4229  *
4230  * @param h is the normal height, this will be reduced automatically if needed for the last row
4231  */
4232 void ff_draw_horiz_band(MpegEncContext *s, int y, int h){
4233     if (s->avctx->draw_horiz_band) {
4234         AVFrame *src;
4235         int offset[4];
4236
4237         if(s->picture_structure != PICT_FRAME){
4238             h <<= 1;
4239             y <<= 1;
4240             if(s->first_field  && !(s->avctx->slice_flags&SLICE_FLAG_ALLOW_FIELD)) return;
4241         }
4242
4243         h= FFMIN(h, s->avctx->height - y);
4244
4245         if(s->pict_type==B_TYPE || s->low_delay || (s->avctx->slice_flags&SLICE_FLAG_CODED_ORDER))
4246             src= (AVFrame*)s->current_picture_ptr;
4247         else if(s->last_picture_ptr)
4248             src= (AVFrame*)s->last_picture_ptr;
4249         else
4250             return;
4251
4252         if(s->pict_type==B_TYPE && s->picture_structure == PICT_FRAME && s->out_format != FMT_H264){
4253             offset[0]=
4254             offset[1]=
4255             offset[2]=
4256             offset[3]= 0;
4257         }else{
4258             offset[0]= y * s->linesize;;
4259             offset[1]=
4260             offset[2]= (y >> s->chroma_y_shift) * s->uvlinesize;
4261             offset[3]= 0;
4262         }
4263
4264         emms_c();
4265
4266         s->avctx->draw_horiz_band(s->avctx, src, offset,
4267                                   y, s->picture_structure, h);
4268     }
4269 }
4270
4271 void ff_init_block_index(MpegEncContext *s){ //FIXME maybe rename
4272     const int linesize= s->current_picture.linesize[0]; //not s->linesize as this would be wrong for field pics
4273     const int uvlinesize= s->current_picture.linesize[1];
4274     const int mb_size= 4 - s->avctx->lowres;
4275
4276     s->block_index[0]= s->b8_stride*(s->mb_y*2    ) - 2 + s->mb_x*2;
4277     s->block_index[1]= s->b8_stride*(s->mb_y*2    ) - 1 + s->mb_x*2;
4278     s->block_index[2]= s->b8_stride*(s->mb_y*2 + 1) - 2 + s->mb_x*2;
4279     s->block_index[3]= s->b8_stride*(s->mb_y*2 + 1) - 1 + s->mb_x*2;
4280     s->block_index[4]= s->mb_stride*(s->mb_y + 1)                + s->b8_stride*s->mb_height*2 + s->mb_x - 1;
4281     s->block_index[5]= s->mb_stride*(s->mb_y + s->mb_height + 2) + s->b8_stride*s->mb_height*2 + s->mb_x - 1;
4282     //block_index is not used by mpeg2, so it is not affected by chroma_format
4283
4284     s->dest[0] = s->current_picture.data[0] + ((s->mb_x - 1) << mb_size);
4285     s->dest[1] = s->current_picture.data[1] + ((s->mb_x - 1) << (mb_size - s->chroma_x_shift));
4286     s->dest[2] = s->current_picture.data[2] + ((s->mb_x - 1) << (mb_size - s->chroma_x_shift));
4287
4288     if(!(s->pict_type==B_TYPE && s->avctx->draw_horiz_band && s->picture_structure==PICT_FRAME))
4289     {
4290         s->dest[0] += s->mb_y *   linesize << mb_size;
4291         s->dest[1] += s->mb_y * uvlinesize << (mb_size - s->chroma_y_shift);
4292         s->dest[2] += s->mb_y * uvlinesize << (mb_size - s->chroma_y_shift);
4293     }
4294 }
4295
4296 #ifdef CONFIG_ENCODERS
4297
4298 static void get_vissual_weight(int16_t *weight, uint8_t *ptr, int stride){
4299     int x, y;
4300 //FIXME optimize
4301     for(y=0; y<8; y++){
4302         for(x=0; x<8; x++){
4303             int x2, y2;
4304             int sum=0;
4305             int sqr=0;
4306             int count=0;
4307
4308             for(y2= FFMAX(y-1, 0); y2 < FFMIN(8, y+2); y2++){
4309                 for(x2= FFMAX(x-1, 0); x2 < FFMIN(8, x+2); x2++){
4310                     int v= ptr[x2 + y2*stride];
4311                     sum += v;
4312                     sqr += v*v;
4313                     count++;
4314                 }
4315             }
4316             weight[x + 8*y]= (36*ff_sqrt(count*sqr - sum*sum)) / count;
4317         }
4318     }
4319 }
4320
4321 static av_always_inline void encode_mb_internal(MpegEncContext *s, int motion_x, int motion_y, int mb_block_height, int mb_block_count)
4322 {
4323     int16_t weight[8][64];
4324     DCTELEM orig[8][64];
4325     const int mb_x= s->mb_x;
4326     const int mb_y= s->mb_y;
4327     int i;
4328     int skip_dct[8];
4329     int dct_offset   = s->linesize*8; //default for progressive frames
4330     uint8_t *ptr_y, *ptr_cb, *ptr_cr;
4331     int wrap_y, wrap_c;
4332
4333     for(i=0; i<mb_block_count; i++) skip_dct[i]=0;
4334
4335     if(s->adaptive_quant){
4336         const int last_qp= s->qscale;
4337         const int mb_xy= mb_x + mb_y*s->mb_stride;
4338
4339         s->lambda= s->lambda_table[mb_xy];
4340         update_qscale(s);
4341
4342         if(!(s->flags&CODEC_FLAG_QP_RD)){
4343             s->qscale= s->current_picture_ptr->qscale_table[mb_xy];
4344             s->dquant= s->qscale - last_qp;
4345
4346             if(s->out_format==FMT_H263){
4347                 s->dquant= clip(s->dquant, -2, 2); //FIXME RD
4348
4349                 if(s->codec_id==CODEC_ID_MPEG4){
4350                     if(!s->mb_intra){
4351                         if(s->pict_type == B_TYPE){
4352                             if(s->dquant&1)
4353                                 s->dquant= 0;
4354                             if(s->mv_dir&MV_DIRECT)
4355                                 s->dquant= 0;
4356                         }
4357                         if(s->mv_type==MV_TYPE_8X8)
4358                             s->dquant=0;
4359                     }
4360                 }
4361             }
4362         }
4363         ff_set_qscale(s, last_qp + s->dquant);
4364     }else if(s->flags&CODEC_FLAG_QP_RD)
4365         ff_set_qscale(s, s->qscale + s->dquant);
4366
4367     wrap_y = s->linesize;
4368     wrap_c = s->uvlinesize;
4369     ptr_y = s->new_picture.data[0] + (mb_y * 16 * wrap_y) + mb_x * 16;
4370     ptr_cb = s->new_picture.data[1] + (mb_y * mb_block_height * wrap_c) + mb_x * 8;
4371     ptr_cr = s->new_picture.data[2] + (mb_y * mb_block_height * wrap_c) + mb_x * 8;
4372
4373     if(mb_x*16+16 > s->width || mb_y*16+16 > s->height){
4374         uint8_t *ebuf= s->edge_emu_buffer + 32;
4375         ff_emulated_edge_mc(ebuf            , ptr_y , wrap_y,16,16,mb_x*16,mb_y*16, s->width   , s->height);
4376         ptr_y= ebuf;
4377         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);
4378         ptr_cb= ebuf+18*wrap_y;
4379         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);
4380         ptr_cr= ebuf+18*wrap_y+8;
4381     }
4382
4383     if (s->mb_intra) {
4384         if(s->flags&CODEC_FLAG_INTERLACED_DCT){
4385             int progressive_score, interlaced_score;
4386
4387             s->interlaced_dct=0;
4388             progressive_score= s->dsp.ildct_cmp[4](s, ptr_y           , NULL, wrap_y, 8)
4389                               +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y*8, NULL, wrap_y, 8) - 400;
4390
4391             if(progressive_score > 0){
4392                 interlaced_score = s->dsp.ildct_cmp[4](s, ptr_y           , NULL, wrap_y*2, 8)
4393                                   +s->dsp.ildct_cmp[4](s, ptr_y + wrap_y  , NULL, wrap_y*2, 8);
4394                 if(progressive_score > interlaced_score){
4395                     s->interlaced_dct=1;
4396
4397                     dct_offset= wrap_y;
4398                     wrap_y<<=1;
4399                     if (s->chroma_format == CHROMA_422)
4400                         wrap_c<<=1;
4401                 }
4402             }
4403         }
4404
4405         s->dsp.get_pixels(s->block[0], ptr_y                 , wrap_y);
4406         s->dsp.get_pixels(s->block[1], ptr_y              + 8, wrap_y);
4407         s->dsp.get_pixels(s->block[2], ptr_y + dct_offset    , wrap_y);
4408         s->dsp.get_pixels(s->block[3], ptr_y + dct_offset + 8, wrap_y);
4409
4410         if(s->flags&CODEC_FLAG_GRAY){
4411             skip_dct[4]= 1;
4412             skip_dct[5]= 1;
4413         }else{
4414             s->dsp.get_pixels(s->block[4], ptr_cb, wrap_c);
4415             s->dsp.get_pixels(s->block[5], ptr_cr, wrap_c);
4416             if(!s->chroma_y_shift){ /* 422 */
4417                 s->dsp.get_pixels(s->block[6], ptr_cb + (dct_offset>>1), wrap_c);
4418                 s->dsp.get_pixels(s->block[7], ptr_cr + (dct_offset>>1), wrap_c);
4419             }
4420         }
4421     }else{
4422         op_pixels_func (*op_pix)[4];
4423         qpel_mc_func (*op_qpix)[16];
4424         uint8_t *dest_y, *dest_cb, *dest_cr;
4425
4426         dest_y  = s->dest[0];
4427         dest_cb = s->dest[1];
4428         dest_cr = s->dest[2];
4429
4430         if ((!s->no_rounding) || s->pict_type==B_TYPE){
4431             op_pix = s->dsp.put_pixels_tab;
4432             op_qpix= s->dsp.put_qpel_pixels_tab;
4433         }else{
4434             op_pix = s->dsp.put_no_rnd_pixels_tab;
4435             op_qpix= s->dsp.put_no_rnd_qpel_pixels_tab;
4436         }
4437
4438         if (s->mv_dir & MV_DIR_FORWARD) {
4439             MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.data, op_pix, op_qpix);
4440             op_pix = s->dsp.avg_pixels_tab;
4441             op_qpix= s->dsp.avg_qpel_pixels_tab;
4442         }
4443         if (s->mv_dir & MV_DIR_BACKWARD) {
4444             MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.data, op_pix, op_qpix);
4445         }
4446
4447         if(s->flags&CODEC_FLAG_INTERLACED_DCT){
4448             int progressive_score, interlaced_score;
4449
4450             s->interlaced_dct=0;
4451             progressive_score= s->dsp.ildct_cmp[0](s, dest_y           , ptr_y           , wrap_y, 8)
4452                               +s->dsp.ildct_cmp[0](s, dest_y + wrap_y*8, ptr_y + wrap_y*8, wrap_y, 8) - 400;
4453
4454             if(s->avctx->ildct_cmp == FF_CMP_VSSE) progressive_score -= 400;
4455
4456             if(progressive_score>0){
4457                 interlaced_score = s->dsp.ildct_cmp[0](s, dest_y           , ptr_y           , wrap_y*2, 8)
4458                                   +s->dsp.ildct_cmp[0](s, dest_y + wrap_y  , ptr_y + wrap_y  , wrap_y*2, 8);
4459
4460                 if(progressive_score > interlaced_score){
4461                     s->interlaced_dct=1;
4462
4463                     dct_offset= wrap_y;
4464                     wrap_y<<=1;
4465                     if (s->chroma_format == CHROMA_422)
4466                         wrap_c<<=1;
4467                 }
4468             }
4469         }
4470
4471         s->dsp.diff_pixels(s->block[0], ptr_y                 , dest_y                 , wrap_y);
4472         s->dsp.diff_pixels(s->block[1], ptr_y              + 8, dest_y              + 8, wrap_y);
4473         s->dsp.diff_pixels(s->block[2], ptr_y + dct_offset    , dest_y + dct_offset    , wrap_y);
4474         s->dsp.diff_pixels(s->block[3], ptr_y + dct_offset + 8, dest_y + dct_offset + 8, wrap_y);
4475
4476         if(s->flags&CODEC_FLAG_GRAY){
4477             skip_dct[4]= 1;
4478             skip_dct[5]= 1;
4479         }else{
4480             s->dsp.diff_pixels(s->block[4], ptr_cb, dest_cb, wrap_c);
4481             s->dsp.diff_pixels(s->block[5], ptr_cr, dest_cr, wrap_c);
4482             if(!s->chroma_y_shift){ /* 422 */
4483                 s->dsp.diff_pixels(s->block[6], ptr_cb + (dct_offset>>1), dest_cb + (dct_offset>>1), wrap_c);
4484                 s->dsp.diff_pixels(s->block[7], ptr_cr + (dct_offset>>1), dest_cr + (dct_offset>>1), wrap_c);
4485             }
4486         }
4487         /* pre quantization */
4488         if(s->current_picture.mc_mb_var[s->mb_stride*mb_y+ mb_x]<2*s->qscale*s->qscale){
4489             //FIXME optimize
4490             if(s->dsp.sad[1](NULL, ptr_y               , dest_y               , wrap_y, 8) < 20*s->qscale) skip_dct[0]= 1;
4491             if(s->dsp.sad[1](NULL, ptr_y            + 8, dest_y            + 8, wrap_y, 8) < 20*s->qscale) skip_dct[1]= 1;
4492             if(s->dsp.sad[1](NULL, ptr_y +dct_offset   , dest_y +dct_offset   , wrap_y, 8) < 20*s->qscale) skip_dct[2]= 1;
4493             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;
4494             if(s->dsp.sad[1](NULL, ptr_cb              , dest_cb              , wrap_c, 8) < 20*s->qscale) skip_dct[4]= 1;
4495             if(s->dsp.sad[1](NULL, ptr_cr              , dest_cr              , wrap_c, 8) < 20*s->qscale) skip_dct[5]= 1;
4496             if(!s->chroma_y_shift){ /* 422 */
4497                 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;
4498                 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;
4499             }
4500         }
4501     }
4502
4503     if(s->avctx->quantizer_noise_shaping){
4504         if(!skip_dct[0]) get_vissual_weight(weight[0], ptr_y                 , wrap_y);
4505         if(!skip_dct[1]) get_vissual_weight(weight[1], ptr_y              + 8, wrap_y);
4506         if(!skip_dct[2]) get_vissual_weight(weight[2], ptr_y + dct_offset    , wrap_y);
4507         if(!skip_dct[3]) get_vissual_weight(weight[3], ptr_y + dct_offset + 8, wrap_y);
4508         if(!skip_dct[4]) get_vissual_weight(weight[4], ptr_cb                , wrap_c);
4509         if(!skip_dct[5]) get_vissual_weight(weight[5], ptr_cr                , wrap_c);
4510         if(!s->chroma_y_shift){ /* 422 */
4511             if(!skip_dct[6]) get_vissual_weight(weight[6], ptr_cb + (dct_offset>>1), wrap_c);
4512             if(!skip_dct[7]) get_vissual_weight(weight[7], ptr_cr + (dct_offset>>1), wrap_c);
4513         }
4514         memcpy(orig[0], s->block[0], sizeof(DCTELEM)*64*mb_block_count);
4515     }
4516
4517     /* DCT & quantize */
4518     assert(s->out_format!=FMT_MJPEG || s->qscale==8);
4519     {
4520         for(i=0;i<mb_block_count;i++) {
4521             if(!skip_dct[i]){
4522                 int overflow;
4523                 s->block_last_index[i] = s->dct_quantize(s, s->block[i], i, s->qscale, &overflow);
4524             // FIXME we could decide to change to quantizer instead of clipping
4525             // JS: I don't think that would be a good idea it could lower quality instead
4526             //     of improve it. Just INTRADC clipping deserves changes in quantizer
4527                 if (overflow) clip_coeffs(s, s->block[i], s->block_last_index[i]);
4528             }else
4529                 s->block_last_index[i]= -1;
4530         }
4531         if(s->avctx->quantizer_noise_shaping){
4532             for(i=0;i<mb_block_count;i++) {
4533                 if(!skip_dct[i]){
4534                     s->block_last_index[i] = dct_quantize_refine(s, s->block[i], weight[i], orig[i], i, s->qscale);
4535                 }
4536             }
4537         }
4538
4539         if(s->luma_elim_threshold && !s->mb_intra)
4540             for(i=0; i<4; i++)
4541                 dct_single_coeff_elimination(s, i, s->luma_elim_threshold);
4542         if(s->chroma_elim_threshold && !s->mb_intra)
4543             for(i=4; i<mb_block_count; i++)
4544                 dct_single_coeff_elimination(s, i, s->chroma_elim_threshold);
4545
4546         if(s->flags & CODEC_FLAG_CBP_RD){
4547             for(i=0;i<mb_block_count;i++) {
4548                 if(s->block_last_index[i] == -1)
4549                     s->coded_score[i]= INT_MAX/256;
4550             }
4551         }
4552     }
4553
4554     if((s->flags&CODEC_FLAG_GRAY) && s->mb_intra){
4555         s->block_last_index[4]=
4556         s->block_last_index[5]= 0;
4557         s->block[4][0]=
4558         s->block[5][0]= (1024 + s->c_dc_scale/2)/ s->c_dc_scale;
4559     }
4560
4561     //non c quantize code returns incorrect block_last_index FIXME
4562     if(s->alternate_scan && s->dct_quantize != dct_quantize_c){
4563         for(i=0; i<mb_block_count; i++){
4564             int j;
4565             if(s->block_last_index[i]>0){
4566                 for(j=63; j>0; j--){
4567                     if(s->block[i][ s->intra_scantable.permutated[j] ]) break;
4568                 }
4569                 s->block_last_index[i]= j;
4570             }
4571         }
4572     }
4573
4574     /* huffman encode */
4575     switch(s->codec_id){ //FIXME funct ptr could be slightly faster
4576     case CODEC_ID_MPEG1VIDEO:
4577     case CODEC_ID_MPEG2VIDEO:
4578         mpeg1_encode_mb(s, s->block, motion_x, motion_y); break;
4579     case CODEC_ID_MPEG4:
4580         mpeg4_encode_mb(s, s->block, motion_x, motion_y); break;
4581     case CODEC_ID_MSMPEG4V2:
4582     case CODEC_ID_MSMPEG4V3:
4583     case CODEC_ID_WMV1:
4584         msmpeg4_encode_mb(s, s->block, motion_x, motion_y); break;
4585     case CODEC_ID_WMV2:
4586          ff_wmv2_encode_mb(s, s->block, motion_x, motion_y); break;
4587 #ifdef CONFIG_H261_ENCODER
4588     case CODEC_ID_H261:
4589         ff_h261_encode_mb(s, s->block, motion_x, motion_y); break;
4590 #endif
4591     case CODEC_ID_H263:
4592     case CODEC_ID_H263P:
4593     case CODEC_ID_FLV1:
4594     case CODEC_ID_RV10:
4595     case CODEC_ID_RV20:
4596         h263_encode_mb(s, s->block, motion_x, motion_y); break;
4597     case CODEC_ID_MJPEG:
4598         mjpeg_encode_mb(s, s->block); break;
4599     default:
4600         assert(0);
4601     }
4602 }
4603
4604 static av_always_inline void encode_mb(MpegEncContext *s, int motion_x, int motion_y)
4605 {
4606     if (s->chroma_format == CHROMA_420) encode_mb_internal(s, motion_x, motion_y,  8, 6);
4607     else                                encode_mb_internal(s, motion_x, motion_y, 16, 8);
4608 }
4609
4610 #endif //CONFIG_ENCODERS
4611
4612 void ff_mpeg_flush(AVCodecContext *avctx){
4613     int i;
4614     MpegEncContext *s = avctx->priv_data;
4615
4616     if(s==NULL || s->picture==NULL)
4617         return;
4618
4619     for(i=0; i<MAX_PICTURE_COUNT; i++){
4620        if(s->picture[i].data[0] && (   s->picture[i].type == FF_BUFFER_TYPE_INTERNAL
4621                                     || s->picture[i].type == FF_BUFFER_TYPE_USER))
4622         avctx->release_buffer(avctx, (AVFrame*)&s->picture[i]);
4623     }
4624     s->current_picture_ptr = s->last_picture_ptr = s->next_picture_ptr = NULL;
4625
4626     s->mb_x= s->mb_y= 0;
4627
4628     s->parse_context.state= -1;
4629     s->parse_context.frame_start_found= 0;
4630     s->parse_context.overread= 0;
4631     s->parse_context.overread_index= 0;
4632     s->parse_context.index= 0;
4633     s->parse_context.last_index= 0;
4634     s->bitstream_buffer_size=0;
4635 }
4636
4637 #ifdef CONFIG_ENCODERS
4638 void ff_copy_bits(PutBitContext *pb, uint8_t *src, int length)
4639 {
4640     const uint16_t *srcw= (uint16_t*)src;
4641     int words= length>>4;
4642     int bits= length&15;
4643     int i;
4644
4645     if(length==0) return;
4646
4647     if(words < 16){
4648         for(i=0; i<words; i++) put_bits(pb, 16, be2me_16(srcw[i]));
4649     }else if(put_bits_count(pb)&7){
4650         for(i=0; i<words; i++) put_bits(pb, 16, be2me_16(srcw[i]));
4651     }else{
4652         for(i=0; put_bits_count(pb)&31; i++)
4653             put_bits(pb, 8, src[i]);
4654         flush_put_bits(pb);
4655         memcpy(pbBufPtr(pb), src+i, 2*words-i);
4656         skip_put_bytes(pb, 2*words-i);
4657     }
4658
4659     put_bits(pb, bits, be2me_16(srcw[words])>>(16-bits));
4660 }
4661
4662 static inline void copy_context_before_encode(MpegEncContext *d, MpegEncContext *s, int type){
4663     int i;
4664
4665     memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
4666
4667     /* mpeg1 */
4668     d->mb_skip_run= s->mb_skip_run;
4669     for(i=0; i<3; i++)
4670         d->last_dc[i]= s->last_dc[i];
4671
4672     /* statistics */
4673     d->mv_bits= s->mv_bits;
4674     d->i_tex_bits= s->i_tex_bits;
4675     d->p_tex_bits= s->p_tex_bits;
4676     d->i_count= s->i_count;
4677     d->f_count= s->f_count;
4678     d->b_count= s->b_count;
4679     d->skip_count= s->skip_count;
4680     d->misc_bits= s->misc_bits;
4681     d->last_bits= 0;
4682
4683     d->mb_skipped= 0;
4684     d->qscale= s->qscale;
4685     d->dquant= s->dquant;
4686 }
4687
4688 static inline void copy_context_after_encode(MpegEncContext *d, MpegEncContext *s, int type){
4689     int i;
4690
4691     memcpy(d->mv, s->mv, 2*4*2*sizeof(int));
4692     memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
4693
4694     /* mpeg1 */
4695     d->mb_skip_run= s->mb_skip_run;
4696     for(i=0; i<3; i++)
4697         d->last_dc[i]= s->last_dc[i];
4698
4699     /* statistics */
4700     d->mv_bits= s->mv_bits;
4701     d->i_tex_bits= s->i_tex_bits;
4702     d->p_tex_bits= s->p_tex_bits;
4703     d->i_count= s->i_count;
4704     d->f_count= s->f_count;
4705     d->b_count= s->b_count;
4706     d->skip_count= s->skip_count;
4707     d->misc_bits= s->misc_bits;
4708
4709     d->mb_intra= s->mb_intra;
4710     d->mb_skipped= s->mb_skipped;
4711     d->mv_type= s->mv_type;
4712     d->mv_dir= s->mv_dir;
4713     d->pb= s->pb;
4714     if(s->data_partitioning){
4715         d->pb2= s->pb2;
4716         d->tex_pb= s->tex_pb;
4717     }
4718     d->block= s->block;
4719     for(i=0; i<8; i++)
4720         d->block_last_index[i]= s->block_last_index[i];
4721     d->interlaced_dct= s->interlaced_dct;
4722     d->qscale= s->qscale;
4723 }
4724
4725 static inline void encode_mb_hq(MpegEncContext *s, MpegEncContext *backup, MpegEncContext *best, int type,
4726                            PutBitContext pb[2], PutBitContext pb2[2], PutBitContext tex_pb[2],
4727                            int *dmin, int *next_block, int motion_x, int motion_y)
4728 {
4729     int score;
4730     uint8_t *dest_backup[3];
4731
4732     copy_context_before_encode(s, backup, type);
4733
4734     s->block= s->blocks[*next_block];
4735     s->pb= pb[*next_block];
4736     if(s->data_partitioning){
4737         s->pb2   = pb2   [*next_block];
4738         s->tex_pb= tex_pb[*next_block];
4739     }
4740
4741     if(*next_block){
4742         memcpy(dest_backup, s->dest, sizeof(s->dest));
4743         s->dest[0] = s->rd_scratchpad;
4744         s->dest[1] = s->rd_scratchpad + 16*s->linesize;
4745         s->dest[2] = s->rd_scratchpad + 16*s->linesize + 8;
4746         assert(s->linesize >= 32); //FIXME
4747     }
4748
4749     encode_mb(s, motion_x, motion_y);
4750
4751     score= put_bits_count(&s->pb);
4752     if(s->data_partitioning){
4753         score+= put_bits_count(&s->pb2);
4754         score+= put_bits_count(&s->tex_pb);
4755     }
4756
4757     if(s->avctx->mb_decision == FF_MB_DECISION_RD){
4758         MPV_decode_mb(s, s->block);
4759
4760         score *= s->lambda2;
4761         score += sse_mb(s) << FF_LAMBDA_SHIFT;
4762     }
4763
4764     if(*next_block){
4765         memcpy(s->dest, dest_backup, sizeof(s->dest));
4766     }
4767
4768     if(score<*dmin){
4769         *dmin= score;
4770         *next_block^=1;
4771
4772         copy_context_after_encode(best, s, type);
4773     }
4774 }
4775
4776 static int sse(MpegEncContext *s, uint8_t *src1, uint8_t *src2, int w, int h, int stride){
4777     uint32_t *sq = ff_squareTbl + 256;
4778     int acc=0;
4779     int x,y;
4780
4781     if(w==16 && h==16)
4782         return s->dsp.sse[0](NULL, src1, src2, stride, 16);
4783     else if(w==8 && h==8)
4784         return s->dsp.sse[1](NULL, src1, src2, stride, 8);
4785
4786     for(y=0; y<h; y++){
4787         for(x=0; x<w; x++){
4788             acc+= sq[src1[x + y*stride] - src2[x + y*stride]];
4789         }
4790     }
4791
4792     assert(acc>=0);
4793
4794     return acc;
4795 }
4796
4797 static int sse_mb(MpegEncContext *s){
4798     int w= 16;
4799     int h= 16;
4800
4801     if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
4802     if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
4803
4804     if(w==16 && h==16)
4805       if(s->avctx->mb_cmp == FF_CMP_NSSE){
4806         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)
4807                +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)
4808                +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);
4809       }else{
4810         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)
4811                +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)
4812                +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);
4813       }
4814     else
4815         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)
4816                +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)
4817                +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);
4818 }
4819
4820 static int pre_estimate_motion_thread(AVCodecContext *c, void *arg){
4821     MpegEncContext *s= arg;
4822
4823
4824     s->me.pre_pass=1;
4825     s->me.dia_size= s->avctx->pre_dia_size;
4826     s->first_slice_line=1;
4827     for(s->mb_y= s->end_mb_y-1; s->mb_y >= s->start_mb_y; s->mb_y--) {
4828         for(s->mb_x=s->mb_width-1; s->mb_x >=0 ;s->mb_x--) {
4829             ff_pre_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
4830         }
4831         s->first_slice_line=0;
4832     }
4833
4834     s->me.pre_pass=0;
4835
4836     return 0;
4837 }
4838
4839 static int estimate_motion_thread(AVCodecContext *c, void *arg){
4840     MpegEncContext *s= arg;
4841
4842     ff_check_alignment();
4843
4844     s->me.dia_size= s->avctx->dia_size;
4845     s->first_slice_line=1;
4846     for(s->mb_y= s->start_mb_y; s->mb_y < s->end_mb_y; s->mb_y++) {
4847         s->mb_x=0; //for block init below
4848         ff_init_block_index(s);
4849         for(s->mb_x=0; s->mb_x < s->mb_width; s->mb_x++) {
4850             s->block_index[0]+=2;
4851             s->block_index[1]+=2;
4852             s->block_index[2]+=2;
4853             s->block_index[3]+=2;
4854
4855             /* compute motion vector & mb_type and store in context */
4856             if(s->pict_type==B_TYPE)
4857                 ff_estimate_b_frame_motion(s, s->mb_x, s->mb_y);
4858             else
4859                 ff_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
4860         }
4861         s->first_slice_line=0;
4862     }
4863     return 0;
4864 }
4865
4866 static int mb_var_thread(AVCodecContext *c, void *arg){
4867     MpegEncContext *s= arg;
4868     int mb_x, mb_y;
4869
4870     ff_check_alignment();
4871
4872     for(mb_y=s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
4873         for(mb_x=0; mb_x < s->mb_width; mb_x++) {
4874             int xx = mb_x * 16;
4875             int yy = mb_y * 16;
4876             uint8_t *pix = s->new_picture.data[0] + (yy * s->linesize) + xx;
4877             int varc;
4878             int sum = s->dsp.pix_sum(pix, s->linesize);
4879
4880             varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500 + 128)>>8;
4881
4882             s->current_picture.mb_var [s->mb_stride * mb_y + mb_x] = varc;
4883             s->current_picture.mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
4884             s->me.mb_var_sum_temp    += varc;
4885         }
4886     }
4887     return 0;
4888 }
4889
4890 static void write_slice_end(MpegEncContext *s){
4891     if(s->codec_id==CODEC_ID_MPEG4){
4892         if(s->partitioned_frame){
4893             ff_mpeg4_merge_partitions(s);
4894         }
4895
4896         ff_mpeg4_stuffing(&s->pb);
4897     }else if(s->out_format == FMT_MJPEG){
4898         ff_mjpeg_stuffing(&s->pb);
4899     }
4900
4901     align_put_bits(&s->pb);
4902     flush_put_bits(&s->pb);
4903
4904     if((s->flags&CODEC_FLAG_PASS1) && !s->partitioned_frame)
4905         s->misc_bits+= get_bits_diff(s);
4906 }
4907
4908 static int encode_thread(AVCodecContext *c, void *arg){
4909     MpegEncContext *s= arg;
4910     int mb_x, mb_y, pdif = 0;
4911     int i, j;
4912     MpegEncContext best_s, backup_s;
4913     uint8_t bit_buf[2][MAX_MB_BYTES];
4914     uint8_t bit_buf2[2][MAX_MB_BYTES];
4915     uint8_t bit_buf_tex[2][MAX_MB_BYTES];
4916     PutBitContext pb[2], pb2[2], tex_pb[2];
4917 //printf("%d->%d\n", s->resync_mb_y, s->end_mb_y);
4918
4919     ff_check_alignment();
4920
4921     for(i=0; i<2; i++){
4922         init_put_bits(&pb    [i], bit_buf    [i], MAX_MB_BYTES);
4923         init_put_bits(&pb2   [i], bit_buf2   [i], MAX_MB_BYTES);
4924         init_put_bits(&tex_pb[i], bit_buf_tex[i], MAX_MB_BYTES);
4925     }
4926
4927     s->last_bits= put_bits_count(&s->pb);
4928     s->mv_bits=0;
4929     s->misc_bits=0;
4930     s->i_tex_bits=0;
4931     s->p_tex_bits=0;
4932     s->i_count=0;
4933     s->f_count=0;
4934     s->b_count=0;
4935     s->skip_count=0;
4936
4937     for(i=0; i<3; i++){
4938         /* init last dc values */
4939         /* note: quant matrix value (8) is implied here */
4940         s->last_dc[i] = 128 << s->intra_dc_precision;
4941
4942         s->current_picture.error[i] = 0;
4943     }
4944     s->mb_skip_run = 0;
4945     memset(s->last_mv, 0, sizeof(s->last_mv));
4946
4947     s->last_mv_dir = 0;
4948
4949     switch(s->codec_id){
4950     case CODEC_ID_H263:
4951     case CODEC_ID_H263P:
4952     case CODEC_ID_FLV1:
4953         s->gob_index = ff_h263_get_gob_height(s);
4954         break;
4955     case CODEC_ID_MPEG4:
4956         if(s->partitioned_frame)
4957             ff_mpeg4_init_partitions(s);
4958         break;
4959     }
4960
4961     s->resync_mb_x=0;
4962     s->resync_mb_y=0;
4963     s->first_slice_line = 1;
4964     s->ptr_lastgob = s->pb.buf;
4965     for(mb_y= s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
4966 //    printf("row %d at %X\n", s->mb_y, (int)s);
4967         s->mb_x=0;
4968         s->mb_y= mb_y;
4969
4970         ff_set_qscale(s, s->qscale);
4971         ff_init_block_index(s);
4972
4973         for(mb_x=0; mb_x < s->mb_width; mb_x++) {
4974             int xy= mb_y*s->mb_stride + mb_x; // removed const, H261 needs to adjust this
4975             int mb_type= s->mb_type[xy];
4976 //            int d;
4977             int dmin= INT_MAX;
4978             int dir;
4979
4980             if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < MAX_MB_BYTES){
4981                 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
4982                 return -1;
4983             }
4984             if(s->data_partitioning){
4985                 if(   s->pb2   .buf_end - s->pb2   .buf - (put_bits_count(&s->    pb2)>>3) < MAX_MB_BYTES
4986                    || s->tex_pb.buf_end - s->tex_pb.buf - (put_bits_count(&s->tex_pb )>>3) < MAX_MB_BYTES){
4987                     av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
4988                     return -1;
4989                 }
4990             }
4991
4992             s->mb_x = mb_x;
4993             s->mb_y = mb_y;  // moved into loop, can get changed by H.261
4994             ff_update_block_index(s);
4995
4996 #ifdef CONFIG_H261_ENCODER
4997             if(s->codec_id == CODEC_ID_H261){
4998                 ff_h261_reorder_mb_index(s);
4999                 xy= s->mb_y*s->mb_stride + s->mb_x;
5000                 mb_type= s->mb_type[xy];
5001             }
5002 #endif
5003
5004             /* write gob / video packet header  */
5005             if(s->rtp_mode){
5006                 int current_packet_size, is_gob_start;
5007
5008                 current_packet_size= ((put_bits_count(&s->pb)+7)>>3) - (s->ptr_lastgob - s->pb.buf);
5009
5010                 is_gob_start= s->avctx->rtp_payload_size && current_packet_size >= s->avctx->rtp_payload_size && mb_y + mb_x>0;
5011
5012                 if(s->start_mb_y == mb_y && mb_y > 0 && mb_x==0) is_gob_start=1;
5013
5014                 switch(s->codec_id){
5015                 case CODEC_ID_H263:
5016                 case CODEC_ID_H263P:
5017                     if(!s->h263_slice_structured)
5018                         if(s->mb_x || s->mb_y%s->gob_index) is_gob_start=0;
5019                     break;
5020                 case CODEC_ID_MPEG2VIDEO:
5021                     if(s->mb_x==0 && s->mb_y!=0) is_gob_start=1;
5022                 case CODEC_ID_MPEG1VIDEO:
5023                     if(s->mb_skip_run) is_gob_start=0;
5024                     break;
5025                 }
5026
5027                 if(is_gob_start){
5028                     if(s->start_mb_y != mb_y || mb_x!=0){
5029                         write_slice_end(s);
5030
5031                         if(s->codec_id==CODEC_ID_MPEG4 && s->partitioned_frame){
5032                             ff_mpeg4_init_partitions(s);
5033                         }
5034                     }
5035
5036                     assert((put_bits_count(&s->pb)&7) == 0);
5037                     current_packet_size= pbBufPtr(&s->pb) - s->ptr_lastgob;
5038
5039                     if(s->avctx->error_rate && s->resync_mb_x + s->resync_mb_y > 0){
5040                         int r= put_bits_count(&s->pb)/8 + s->picture_number + 16 + s->mb_x + s->mb_y;
5041                         int d= 100 / s->avctx->error_rate;
5042                         if(r % d == 0){
5043                             current_packet_size=0;
5044 #ifndef ALT_BITSTREAM_WRITER
5045                             s->pb.buf_ptr= s->ptr_lastgob;
5046 #endif
5047                             assert(pbBufPtr(&s->pb) == s->ptr_lastgob);
5048                         }
5049                     }
5050
5051                     if (s->avctx->rtp_callback){
5052                         int number_mb = (mb_y - s->resync_mb_y)*s->mb_width + mb_x - s->resync_mb_x;
5053                         s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, current_packet_size, number_mb);
5054                     }
5055
5056                     switch(s->codec_id){
5057                     case CODEC_ID_MPEG4:
5058                         ff_mpeg4_encode_video_packet_header(s);
5059                         ff_mpeg4_clean_buffers(s);
5060                     break;
5061                     case CODEC_ID_MPEG1VIDEO:
5062                     case CODEC_ID_MPEG2VIDEO:
5063                         ff_mpeg1_encode_slice_header(s);
5064                         ff_mpeg1_clean_buffers(s);
5065                     break;
5066                     case CODEC_ID_H263:
5067                     case CODEC_ID_H263P:
5068                         h263_encode_gob_header(s, mb_y);
5069                     break;
5070                     }
5071
5072                     if(s->flags&CODEC_FLAG_PASS1){
5073                         int bits= put_bits_count(&s->pb);
5074                         s->misc_bits+= bits - s->last_bits;
5075                         s->last_bits= bits;
5076                     }
5077
5078                     s->ptr_lastgob += current_packet_size;
5079                     s->first_slice_line=1;
5080                     s->resync_mb_x=mb_x;
5081                     s->resync_mb_y=mb_y;
5082                 }
5083             }
5084
5085             if(  (s->resync_mb_x   == s->mb_x)
5086                && s->resync_mb_y+1 == s->mb_y){
5087                 s->first_slice_line=0;
5088             }
5089
5090             s->mb_skipped=0;
5091             s->dquant=0; //only for QP_RD
5092
5093             if(mb_type & (mb_type-1) || (s->flags & CODEC_FLAG_QP_RD)){ // more than 1 MB type possible or CODEC_FLAG_QP_RD
5094                 int next_block=0;
5095                 int pb_bits_count, pb2_bits_count, tex_pb_bits_count;
5096
5097                 copy_context_before_encode(&backup_s, s, -1);
5098                 backup_s.pb= s->pb;
5099                 best_s.data_partitioning= s->data_partitioning;
5100                 best_s.partitioned_frame= s->partitioned_frame;
5101                 if(s->data_partitioning){
5102                     backup_s.pb2= s->pb2;
5103                     backup_s.tex_pb= s->tex_pb;
5104                 }
5105
5106                 if(mb_type&CANDIDATE_MB_TYPE_INTER){
5107                     s->mv_dir = MV_DIR_FORWARD;
5108                     s->mv_type = MV_TYPE_16X16;
5109                     s->mb_intra= 0;
5110                     s->mv[0][0][0] = s->p_mv_table[xy][0];
5111                     s->mv[0][0][1] = s->p_mv_table[xy][1];
5112                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER, pb, pb2, tex_pb,
5113                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
5114                 }
5115                 if(mb_type&CANDIDATE_MB_TYPE_INTER_I){
5116                     s->mv_dir = MV_DIR_FORWARD;
5117                     s->mv_type = MV_TYPE_FIELD;
5118                     s->mb_intra= 0;
5119                     for(i=0; i<2; i++){
5120                         j= s->field_select[0][i] = s->p_field_select_table[i][xy];
5121                         s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
5122                         s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
5123                     }
5124                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER_I, pb, pb2, tex_pb,
5125                                  &dmin, &next_block, 0, 0);
5126                 }
5127                 if(mb_type&CANDIDATE_MB_TYPE_SKIPPED){
5128                     s->mv_dir = MV_DIR_FORWARD;
5129                     s->mv_type = MV_TYPE_16X16;
5130                     s->mb_intra= 0;
5131                     s->mv[0][0][0] = 0;
5132                     s->mv[0][0][1] = 0;
5133                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_SKIPPED, pb, pb2, tex_pb,
5134                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
5135                 }
5136                 if(mb_type&CANDIDATE_MB_TYPE_INTER4V){
5137                     s->mv_dir = MV_DIR_FORWARD;
5138                     s->mv_type = MV_TYPE_8X8;
5139                     s->mb_intra= 0;
5140                     for(i=0; i<4; i++){
5141                         s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
5142                         s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
5143                     }
5144                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER4V, pb, pb2, tex_pb,
5145                                  &dmin, &next_block, 0, 0);
5146                 }
5147                 if(mb_type&CANDIDATE_MB_TYPE_FORWARD){
5148                     s->mv_dir = MV_DIR_FORWARD;
5149                     s->mv_type = MV_TYPE_16X16;
5150                     s->mb_intra= 0;
5151                     s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
5152                     s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
5153                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD, pb, pb2, tex_pb,
5154                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
5155                 }
5156                 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD){
5157                     s->mv_dir = MV_DIR_BACKWARD;
5158                     s->mv_type = MV_TYPE_16X16;
5159                     s->mb_intra= 0;
5160                     s->mv[1][0][0] = s->b_back_mv_table[xy][0];
5161                     s->mv[1][0][1] = s->b_back_mv_table[xy][1];
5162                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD, pb, pb2, tex_pb,
5163                                  &dmin, &next_block, s->mv[1][0][0], s->mv[1][0][1]);
5164                 }
5165                 if(mb_type&CANDIDATE_MB_TYPE_BIDIR){
5166                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
5167                     s->mv_type = MV_TYPE_16X16;
5168                     s->mb_intra= 0;
5169                     s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
5170                     s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
5171                     s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
5172                     s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
5173                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR, pb, pb2, tex_pb,
5174                                  &dmin, &next_block, 0, 0);
5175                 }
5176                 if(mb_type&CANDIDATE_MB_TYPE_FORWARD_I){
5177                     s->mv_dir = MV_DIR_FORWARD;
5178                     s->mv_type = MV_TYPE_FIELD;
5179                     s->mb_intra= 0;
5180                     for(i=0; i<2; i++){
5181                         j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
5182                         s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
5183                         s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
5184                     }
5185                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD_I, pb, pb2, tex_pb,
5186                                  &dmin, &next_block, 0, 0);
5187                 }
5188                 if(mb_type&CANDIDATE_MB_TYPE_BACKWARD_I){
5189                     s->mv_dir = MV_DIR_BACKWARD;
5190                     s->mv_type = MV_TYPE_FIELD;
5191                     s->mb_intra= 0;
5192                     for(i=0; i<2; i++){
5193                         j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
5194                         s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
5195                         s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
5196                     }
5197                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD_I, pb, pb2, tex_pb,
5198                                  &dmin, &next_block, 0, 0);
5199                 }
5200                 if(mb_type&CANDIDATE_MB_TYPE_BIDIR_I){
5201                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
5202                     s->mv_type = MV_TYPE_FIELD;
5203                     s->mb_intra= 0;
5204                     for(dir=0; dir<2; dir++){
5205                         for(i=0; i<2; i++){
5206                             j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
5207                             s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
5208                             s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
5209                         }
5210                     }
5211                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR_I, pb, pb2, tex_pb,
5212                                  &dmin, &next_block, 0, 0);
5213                 }
5214                 if(mb_type&CANDIDATE_MB_TYPE_INTRA){
5215                     s->mv_dir = 0;
5216                     s->mv_type = MV_TYPE_16X16;
5217                     s->mb_intra= 1;
5218                     s->mv[0][0][0] = 0;
5219                     s->mv[0][0][1] = 0;
5220                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTRA, pb, pb2, tex_pb,
5221                                  &dmin, &next_block, 0, 0);
5222                     if(s->h263_pred || s->h263_aic){
5223                         if(best_s.mb_intra)
5224                             s->mbintra_table[mb_x + mb_y*s->mb_stride]=1;
5225                         else
5226                             ff_clean_intra_table_entries(s); //old mode?
5227                     }
5228                 }
5229
5230                 if((s->flags & CODEC_FLAG_QP_RD) && dmin < INT_MAX){
5231                     if(best_s.mv_type==MV_TYPE_16X16){ //FIXME move 4mv after QPRD
5232                         const int last_qp= backup_s.qscale;
5233                         int qpi, qp, dc[6];
5234                         DCTELEM ac[6][16];
5235                         const int mvdir= (best_s.mv_dir&MV_DIR_BACKWARD) ? 1 : 0;
5236                         static const int dquant_tab[4]={-1,1,-2,2};
5237
5238                         assert(backup_s.dquant == 0);
5239
5240                         //FIXME intra
5241                         s->mv_dir= best_s.mv_dir;
5242                         s->mv_type = MV_TYPE_16X16;
5243                         s->mb_intra= best_s.mb_intra;
5244                         s->mv[0][0][0] = best_s.mv[0][0][0];
5245                         s->mv[0][0][1] = best_s.mv[0][0][1];
5246                         s->mv[1][0][0] = best_s.mv[1][0][0];
5247                         s->mv[1][0][1] = best_s.mv[1][0][1];
5248
5249                         qpi = s->pict_type == B_TYPE ? 2 : 0;
5250                         for(; qpi<4; qpi++){
5251                             int dquant= dquant_tab[qpi];
5252                             qp= last_qp + dquant;
5253                             if(qp < s->avctx->qmin || qp > s->avctx->qmax)
5254                                 continue;
5255                             backup_s.dquant= dquant;
5256                             if(s->mb_intra && s->dc_val[0]){
5257                                 for(i=0; i<6; i++){
5258                                     dc[i]= s->dc_val[0][ s->block_index[i] ];
5259                                     memcpy(ac[i], s->ac_val[0][s->block_index[i]], sizeof(DCTELEM)*16);
5260                                 }
5261                             }
5262
5263                             encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
5264                                          &dmin, &next_block, s->mv[mvdir][0][0], s->mv[mvdir][0][1]);
5265                             if(best_s.qscale != qp){
5266                                 if(s->mb_intra && s->dc_val[0]){
5267                                     for(i=0; i<6; i++){
5268                                         s->dc_val[0][ s->block_index[i] ]= dc[i];
5269                                         memcpy(s->ac_val[0][s->block_index[i]], ac[i], sizeof(DCTELEM)*16);
5270                                     }
5271                                 }
5272                             }
5273                         }
5274                     }
5275                 }
5276                 if(mb_type&CANDIDATE_MB_TYPE_DIRECT){
5277                     int mx= s->b_direct_mv_table[xy][0];
5278                     int my= s->b_direct_mv_table[xy][1];
5279
5280                     backup_s.dquant = 0;
5281                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
5282                     s->mb_intra= 0;
5283                     ff_mpeg4_set_direct_mv(s, mx, my);
5284                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
5285                                  &dmin, &next_block, mx, my);
5286                 }
5287                 if(mb_type&CANDIDATE_MB_TYPE_DIRECT0){
5288                     backup_s.dquant = 0;
5289                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
5290                     s->mb_intra= 0;
5291                     ff_mpeg4_set_direct_mv(s, 0, 0);
5292                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
5293                                  &dmin, &next_block, 0, 0);
5294                 }
5295                 s->current_picture.qscale_table[xy]= best_s.qscale;
5296
5297                 copy_context_after_encode(s, &best_s, -1);
5298
5299                 pb_bits_count= put_bits_count(&s->pb);
5300                 flush_put_bits(&s->pb);
5301                 ff_copy_bits(&backup_s.pb, bit_buf[next_block^1], pb_bits_count);
5302                 s->pb= backup_s.pb;
5303
5304                 if(s->data_partitioning){
5305                     pb2_bits_count= put_bits_count(&s->pb2);
5306                     flush_put_bits(&s->pb2);
5307                     ff_copy_bits(&backup_s.pb2, bit_buf2[next_block^1], pb2_bits_count);
5308                     s->pb2= backup_s.pb2;
5309
5310                     tex_pb_bits_count= put_bits_count(&s->tex_pb);
5311                     flush_put_bits(&s->tex_pb);
5312                     ff_copy_bits(&backup_s.tex_pb, bit_buf_tex[next_block^1], tex_pb_bits_count);
5313                     s->tex_pb= backup_s.tex_pb;
5314                 }
5315                 s->last_bits= put_bits_count(&s->pb);
5316
5317                 if (s->out_format == FMT_H263 && s->pict_type!=B_TYPE)
5318                     ff_h263_update_motion_val(s);
5319
5320                 if(next_block==0){ //FIXME 16 vs linesize16
5321                     s->dsp.put_pixels_tab[0][0](s->dest[0], s->rd_scratchpad                     , s->linesize  ,16);
5322                     s->dsp.put_pixels_tab[1][0](s->dest[1], s->rd_scratchpad + 16*s->linesize    , s->uvlinesize, 8);
5323                     s->dsp.put_pixels_tab[1][0](s->dest[2], s->rd_scratchpad + 16*s->linesize + 8, s->uvlinesize, 8);
5324                 }
5325
5326                 if(s->avctx->mb_decision == FF_MB_DECISION_BITS)
5327                     MPV_decode_mb(s, s->block);
5328             } else {
5329                 int motion_x, motion_y;
5330                 s->mv_type=MV_TYPE_16X16;
5331                 // only one MB-Type possible
5332
5333                 switch(mb_type){
5334                 case CANDIDATE_MB_TYPE_INTRA:
5335                     s->mv_dir = 0;
5336                     s->mb_intra= 1;
5337                     motion_x= s->mv[0][0][0] = 0;
5338                     motion_y= s->mv[0][0][1] = 0;
5339                     break;
5340                 case CANDIDATE_MB_TYPE_INTER:
5341                     s->mv_dir = MV_DIR_FORWARD;
5342                     s->mb_intra= 0;
5343                     motion_x= s->mv[0][0][0] = s->p_mv_table[xy][0];
5344                     motion_y= s->mv[0][0][1] = s->p_mv_table[xy][1];
5345                     break;
5346                 case CANDIDATE_MB_TYPE_INTER_I:
5347                     s->mv_dir = MV_DIR_FORWARD;
5348                     s->mv_type = MV_TYPE_FIELD;
5349                     s->mb_intra= 0;
5350                     for(i=0; i<2; i++){
5351                         j= s->field_select[0][i] = s->p_field_select_table[i][xy];
5352                         s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
5353                         s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
5354                     }
5355                     motion_x = motion_y = 0;
5356                     break;
5357                 case CANDIDATE_MB_TYPE_INTER4V:
5358                     s->mv_dir = MV_DIR_FORWARD;
5359                     s->mv_type = MV_TYPE_8X8;
5360                     s->mb_intra= 0;
5361                     for(i=0; i<4; i++){
5362                         s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
5363                         s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
5364                     }
5365                     motion_x= motion_y= 0;
5366                     break;
5367                 case CANDIDATE_MB_TYPE_DIRECT:
5368                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
5369                     s->mb_intra= 0;
5370                     motion_x=s->b_direct_mv_table[xy][0];
5371                     motion_y=s->b_direct_mv_table[xy][1];
5372                     ff_mpeg4_set_direct_mv(s, motion_x, motion_y);
5373                     break;
5374                 case CANDIDATE_MB_TYPE_BIDIR:
5375                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
5376                     s->mb_intra= 0;
5377                     motion_x=0;
5378                     motion_y=0;
5379                     s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
5380                     s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
5381                     s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
5382                     s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
5383                     break;
5384                 case CANDIDATE_MB_TYPE_BACKWARD:
5385                     s->mv_dir = MV_DIR_BACKWARD;
5386                     s->mb_intra= 0;
5387                     motion_x= s->mv[1][0][0] = s->b_back_mv_table[xy][0];
5388                     motion_y= s->mv[1][0][1] = s->b_back_mv_table[xy][1];
5389                     break;
5390                 case CANDIDATE_MB_TYPE_FORWARD:
5391                     s->mv_dir = MV_DIR_FORWARD;
5392                     s->mb_intra= 0;
5393                     motion_x= s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
5394                     motion_y= s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
5395 //                    printf(" %d %d ", motion_x, motion_y);
5396                     break;
5397                 case CANDIDATE_MB_TYPE_FORWARD_I:
5398                     s->mv_dir = MV_DIR_FORWARD;
5399                     s->mv_type = MV_TYPE_FIELD;
5400                     s->mb_intra= 0;
5401                     for(i=0; i<2; i++){
5402                         j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
5403                         s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
5404                         s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
5405                     }
5406                     motion_x=motion_y=0;
5407                     break;
5408                 case CANDIDATE_MB_TYPE_BACKWARD_I:
5409                     s->mv_dir = MV_DIR_BACKWARD;
5410                     s->mv_type = MV_TYPE_FIELD;
5411                     s->mb_intra= 0;
5412                     for(i=0; i<2; i++){
5413                         j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
5414                         s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
5415                         s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
5416                     }
5417                     motion_x=motion_y=0;
5418                     break;
5419                 case CANDIDATE_MB_TYPE_BIDIR_I:
5420                     s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
5421                     s->mv_type = MV_TYPE_FIELD;
5422                     s->mb_intra= 0;
5423                     for(dir=0; dir<2; dir++){
5424                         for(i=0; i<2; i++){
5425                             j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
5426                             s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
5427                             s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
5428                         }
5429                     }
5430                     motion_x=motion_y=0;
5431                     break;
5432                 default:
5433                     motion_x=motion_y=0; //gcc warning fix
5434                     av_log(s->avctx, AV_LOG_ERROR, "illegal MB type\n");
5435                 }
5436
5437                 encode_mb(s, motion_x, motion_y);
5438
5439                 // RAL: Update last macroblock type
5440                 s->last_mv_dir = s->mv_dir;
5441
5442                 if (s->out_format == FMT_H263 && s->pict_type!=B_TYPE)
5443                     ff_h263_update_motion_val(s);
5444
5445                 MPV_decode_mb(s, s->block);
5446             }
5447
5448             /* clean the MV table in IPS frames for direct mode in B frames */
5449             if(s->mb_intra /* && I,P,S_TYPE */){
5450                 s->p_mv_table[xy][0]=0;
5451                 s->p_mv_table[xy][1]=0;
5452             }
5453
5454             if(s->flags&CODEC_FLAG_PSNR){
5455                 int w= 16;
5456                 int h= 16;
5457
5458                 if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
5459                 if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
5460
5461                 s->current_picture.error[0] += sse(
5462                     s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16,
5463                     s->dest[0], w, h, s->linesize);
5464                 s->current_picture.error[1] += sse(
5465                     s, s->new_picture.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,
5466                     s->dest[1], w>>1, h>>1, s->uvlinesize);
5467                 s->current_picture.error[2] += sse(
5468                     s, s->new_picture    .data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,
5469                     s->dest[2], w>>1, h>>1, s->uvlinesize);
5470             }
5471             if(s->loop_filter){
5472                 if(s->out_format == FMT_H263)
5473                     ff_h263_loop_filter(s);
5474             }
5475 //printf("MB %d %d bits\n", s->mb_x+s->mb_y*s->mb_stride, put_bits_count(&s->pb));
5476         }
5477     }
5478
5479     //not beautiful here but we must write it before flushing so it has to be here
5480     if (s->msmpeg4_version && s->msmpeg4_version<4 && s->pict_type == I_TYPE)
5481         msmpeg4_encode_ext_header(s);
5482
5483     write_slice_end(s);
5484
5485     /* Send the last GOB if RTP */
5486     if (s->avctx->rtp_callback) {
5487         int number_mb = (mb_y - s->resync_mb_y)*s->mb_width - s->resync_mb_x;
5488         pdif = pbBufPtr(&s->pb) - s->ptr_lastgob;
5489         /* Call the RTP callback to send the last GOB */
5490         emms_c();
5491         s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, pdif, number_mb);
5492     }
5493
5494     return 0;
5495 }
5496
5497 #define MERGE(field) dst->field += src->field; src->field=0
5498 static void merge_context_after_me(MpegEncContext *dst, MpegEncContext *src){
5499     MERGE(me.scene_change_score);
5500     MERGE(me.mc_mb_var_sum_temp);
5501     MERGE(me.mb_var_sum_temp);
5502 }
5503
5504 static void merge_context_after_encode(MpegEncContext *dst, MpegEncContext *src){
5505     int i;
5506
5507     MERGE(dct_count[0]); //note, the other dct vars are not part of the context
5508     MERGE(dct_count[1]);
5509     MERGE(mv_bits);
5510     MERGE(i_tex_bits);
5511     MERGE(p_tex_bits);
5512     MERGE(i_count);
5513     MERGE(f_count);
5514     MERGE(b_count);
5515     MERGE(skip_count);
5516     MERGE(misc_bits);
5517     MERGE(error_count);
5518     MERGE(padding_bug_score);
5519     MERGE(current_picture.error[0]);
5520     MERGE(current_picture.error[1]);
5521     MERGE(current_picture.error[2]);
5522
5523     if(dst->avctx->noise_reduction){
5524         for(i=0; i<64; i++){
5525             MERGE(dct_error_sum[0][i]);
5526             MERGE(dct_error_sum[1][i]);
5527         }
5528     }
5529
5530     assert(put_bits_count(&src->pb) % 8 ==0);
5531     assert(put_bits_count(&dst->pb) % 8 ==0);
5532     ff_copy_bits(&dst->pb, src->pb.buf, put_bits_count(&src->pb));
5533     flush_put_bits(&dst->pb);
5534 }
5535
5536 static int estimate_qp(MpegEncContext *s, int dry_run){
5537     if (s->next_lambda){
5538         s->current_picture_ptr->quality=
5539         s->current_picture.quality = s->next_lambda;
5540         if(!dry_run) s->next_lambda= 0;
5541     } else if (!s->fixed_qscale) {
5542         s->current_picture_ptr->quality=
5543         s->current_picture.quality = ff_rate_estimate_qscale(s, dry_run);
5544         if (s->current_picture.quality < 0)
5545             return -1;
5546     }
5547
5548     if(s->adaptive_quant){
5549         switch(s->codec_id){
5550         case CODEC_ID_MPEG4:
5551             ff_clean_mpeg4_qscales(s);
5552             break;
5553         case CODEC_ID_H263:
5554         case CODEC_ID_H263P:
5555         case CODEC_ID_FLV1:
5556             ff_clean_h263_qscales(s);
5557             break;
5558         }
5559
5560         s->lambda= s->lambda_table[0];
5561         //FIXME broken
5562     }else
5563         s->lambda= s->current_picture.quality;
5564 //printf("%d %d\n", s->avctx->global_quality, s->current_picture.quality);
5565     update_qscale(s);
5566     return 0;
5567 }
5568
5569 static int encode_picture(MpegEncContext *s, int picture_number)
5570 {
5571     int i;
5572     int bits;
5573
5574     s->picture_number = picture_number;
5575
5576     /* Reset the average MB variance */
5577     s->me.mb_var_sum_temp    =
5578     s->me.mc_mb_var_sum_temp = 0;
5579
5580     /* we need to initialize some time vars before we can encode b-frames */
5581     // RAL: Condition added for MPEG1VIDEO
5582     if (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO || (s->h263_pred && !s->h263_msmpeg4))
5583         ff_set_mpeg4_time(s, s->picture_number);  //FIXME rename and use has_b_frames or similar
5584
5585     s->me.scene_change_score=0;
5586
5587 //    s->lambda= s->current_picture_ptr->quality; //FIXME qscale / ... stuff for ME ratedistoration
5588
5589     if(s->pict_type==I_TYPE){
5590         if(s->msmpeg4_version >= 3) s->no_rounding=1;
5591         else                        s->no_rounding=0;
5592     }else if(s->pict_type!=B_TYPE){
5593         if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
5594             s->no_rounding ^= 1;
5595     }
5596
5597     if(s->flags & CODEC_FLAG_PASS2){
5598         if (estimate_qp(s,1) < 0)
5599             return -1;
5600         ff_get_2pass_fcode(s);
5601     }else if(!(s->flags & CODEC_FLAG_QSCALE)){
5602         if(s->pict_type==B_TYPE)
5603             s->lambda= s->last_lambda_for[s->pict_type];
5604         else
5605             s->lambda= s->last_lambda_for[s->last_non_b_pict_type];
5606         update_qscale(s);
5607     }
5608
5609     s->mb_intra=0; //for the rate distortion & bit compare functions
5610     for(i=1; i<s->avctx->thread_count; i++){
5611         ff_update_duplicate_context(s->thread_context[i], s);
5612     }
5613
5614     ff_init_me(s);
5615
5616     /* Estimate motion for every MB */
5617     if(s->pict_type != I_TYPE){
5618         s->lambda = (s->lambda * s->avctx->me_penalty_compensation + 128)>>8;
5619         s->lambda2= (s->lambda2* (int64_t)s->avctx->me_penalty_compensation + 128)>>8;
5620         if(s->pict_type != B_TYPE && s->avctx->me_threshold==0){
5621             if((s->avctx->pre_me && s->last_non_b_pict_type==I_TYPE) || s->avctx->pre_me==2){
5622                 s->avctx->execute(s->avctx, pre_estimate_motion_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count);
5623             }
5624         }
5625
5626         s->avctx->execute(s->avctx, estimate_motion_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count);
5627     }else /* if(s->pict_type == I_TYPE) */{
5628         /* I-Frame */
5629         for(i=0; i<s->mb_stride*s->mb_height; i++)
5630             s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
5631
5632         if(!s->fixed_qscale){
5633             /* finding spatial complexity for I-frame rate control */
5634             s->avctx->execute(s->avctx, mb_var_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count);
5635         }
5636     }
5637     for(i=1; i<s->avctx->thread_count; i++){
5638         merge_context_after_me(s, s->thread_context[i]);
5639     }
5640     s->current_picture.mc_mb_var_sum= s->current_picture_ptr->mc_mb_var_sum= s->me.mc_mb_var_sum_temp;
5641     s->current_picture.   mb_var_sum= s->current_picture_ptr->   mb_var_sum= s->me.   mb_var_sum_temp;
5642     emms_c();
5643
5644     if(s->me.scene_change_score > s->avctx->scenechange_threshold && s->pict_type == P_TYPE){
5645         s->pict_type= I_TYPE;
5646         for(i=0; i<s->mb_stride*s->mb_height; i++)
5647             s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
5648 //printf("Scene change detected, encoding as I Frame %d %d\n", s->current_picture.mb_var_sum, s->current_picture.mc_mb_var_sum);
5649     }
5650
5651     if(!s->umvplus){
5652         if(s->pict_type==P_TYPE || s->pict_type==S_TYPE) {
5653             s->f_code= ff_get_best_fcode(s, s->p_mv_table, CANDIDATE_MB_TYPE_INTER);
5654
5655             if(s->flags & CODEC_FLAG_INTERLACED_ME){
5656                 int a,b;
5657                 a= ff_get_best_fcode(s, s->p_field_mv_table[0][0], CANDIDATE_MB_TYPE_INTER_I); //FIXME field_select
5658                 b= ff_get_best_fcode(s, s->p_field_mv_table[1][1], CANDIDATE_MB_TYPE_INTER_I);
5659                 s->f_code= FFMAX(s->f_code, FFMAX(a,b));
5660             }
5661
5662             ff_fix_long_p_mvs(s);
5663             ff_fix_long_mvs(s, NULL, 0, s->p_mv_table, s->f_code, CANDIDATE_MB_TYPE_INTER, 0);
5664             if(s->flags & CODEC_FLAG_INTERLACED_ME){
5665                 int j;
5666                 for(i=0; i<2; i++){
5667                     for(j=0; j<2; j++)
5668                         ff_fix_long_mvs(s, s->p_field_select_table[i], j,
5669                                         s->p_field_mv_table[i][j], s->f_code, CANDIDATE_MB_TYPE_INTER_I, 0);
5670                 }
5671             }
5672         }
5673
5674         if(s->pict_type==B_TYPE){
5675             int a, b;
5676
5677             a = ff_get_best_fcode(s, s->b_forw_mv_table, CANDIDATE_MB_TYPE_FORWARD);
5678             b = ff_get_best_fcode(s, s->b_bidir_forw_mv_table, CANDIDATE_MB_TYPE_BIDIR);
5679             s->f_code = FFMAX(a, b);
5680
5681             a = ff_get_best_fcode(s, s->b_back_mv_table, CANDIDATE_MB_TYPE_BACKWARD);
5682             b = ff_get_best_fcode(s, s->b_bidir_back_mv_table, CANDIDATE_MB_TYPE_BIDIR);
5683             s->b_code = FFMAX(a, b);
5684
5685             ff_fix_long_mvs(s, NULL, 0, s->b_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_FORWARD, 1);
5686             ff_fix_long_mvs(s, NULL, 0, s->b_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BACKWARD, 1);
5687             ff_fix_long_mvs(s, NULL, 0, s->b_bidir_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_BIDIR, 1);
5688             ff_fix_long_mvs(s, NULL, 0, s->b_bidir_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BIDIR, 1);
5689             if(s->flags & CODEC_FLAG_INTERLACED_ME){
5690                 int dir, j;
5691                 for(dir=0; dir<2; dir++){
5692                     for(i=0; i<2; i++){
5693                         for(j=0; j<2; j++){
5694                             int type= dir ? (CANDIDATE_MB_TYPE_BACKWARD_I|CANDIDATE_MB_TYPE_BIDIR_I)
5695                                           : (CANDIDATE_MB_TYPE_FORWARD_I |CANDIDATE_MB_TYPE_BIDIR_I);
5696                             ff_fix_long_mvs(s, s->b_field_select_table[dir][i], j,
5697                                             s->b_field_mv_table[dir][i][j], dir ? s->b_code : s->f_code, type, 1);
5698                         }
5699                     }
5700                 }
5701             }
5702         }
5703     }
5704
5705     if (estimate_qp(s, 0) < 0)
5706         return -1;
5707
5708     if(s->qscale < 3 && s->max_qcoeff<=128 && s->pict_type==I_TYPE && !(s->flags & CODEC_FLAG_QSCALE))
5709         s->qscale= 3; //reduce clipping problems
5710
5711     if (s->out_format == FMT_MJPEG) {
5712         /* for mjpeg, we do include qscale in the matrix */
5713         s->intra_matrix[0] = ff_mpeg1_default_intra_matrix[0];
5714         for(i=1;i<64;i++){
5715             int j= s->dsp.idct_permutation[i];
5716
5717             s->intra_matrix[j] = clip_uint8((ff_mpeg1_default_intra_matrix[i] * s->qscale) >> 3);
5718         }
5719         convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
5720                        s->intra_matrix, s->intra_quant_bias, 8, 8, 1);
5721         s->qscale= 8;
5722     }
5723
5724     //FIXME var duplication
5725     s->current_picture_ptr->key_frame=
5726     s->current_picture.key_frame= s->pict_type == I_TYPE; //FIXME pic_ptr
5727     s->current_picture_ptr->pict_type=
5728     s->current_picture.pict_type= s->pict_type;
5729
5730     if(s->current_picture.key_frame)
5731         s->picture_in_gop_number=0;
5732
5733     s->last_bits= put_bits_count(&s->pb);
5734     switch(s->out_format) {
5735     case FMT_MJPEG:
5736         mjpeg_picture_header(s);
5737         break;
5738 #ifdef CONFIG_H261_ENCODER
5739     case FMT_H261:
5740         ff_h261_encode_picture_header(s, picture_number);
5741         break;
5742 #endif
5743     case FMT_H263:
5744         if (s->codec_id == CODEC_ID_WMV2)
5745             ff_wmv2_encode_picture_header(s, picture_number);
5746         else if (s->h263_msmpeg4)
5747             msmpeg4_encode_picture_header(s, picture_number);
5748         else if (s->h263_pred)
5749             mpeg4_encode_picture_header(s, picture_number);
5750 #ifdef CONFIG_RV10_ENCODER
5751         else if (s->codec_id == CODEC_ID_RV10)
5752             rv10_encode_picture_header(s, picture_number);
5753 #endif
5754 #ifdef CONFIG_RV20_ENCODER
5755         else if (s->codec_id == CODEC_ID_RV20)
5756             rv20_encode_picture_header(s, picture_number);
5757 #endif
5758         else if (s->codec_id == CODEC_ID_FLV1)
5759             ff_flv_encode_picture_header(s, picture_number);
5760         else
5761             h263_encode_picture_header(s, picture_number);
5762         break;
5763     case FMT_MPEG1:
5764         mpeg1_encode_picture_header(s, picture_number);
5765         break;
5766     case FMT_H264:
5767         break;
5768     default:
5769         assert(0);
5770     }
5771     bits= put_bits_count(&s->pb);
5772     s->header_bits= bits - s->last_bits;
5773
5774     for(i=1; i<s->avctx->thread_count; i++){
5775         update_duplicate_context_after_me(s->thread_context[i], s);
5776     }
5777     s->avctx->execute(s->avctx, encode_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count);
5778     for(i=1; i<s->avctx->thread_count; i++){
5779         merge_context_after_encode(s, s->thread_context[i]);
5780     }
5781     emms_c();
5782     return 0;
5783 }
5784
5785 static void  denoise_dct_c(MpegEncContext *s, DCTELEM *block){
5786     const int intra= s->mb_intra;
5787     int i;
5788
5789     s->dct_count[intra]++;
5790
5791     for(i=0; i<64; i++){
5792         int level= block[i];
5793
5794         if(level){
5795             if(level>0){
5796                 s->dct_error_sum[intra][i] += level;
5797                 level -= s->dct_offset[intra][i];
5798                 if(level<0) level=0;
5799             }else{
5800                 s->dct_error_sum[intra][i] -= level;
5801                 level += s->dct_offset[intra][i];
5802                 if(level>0) level=0;
5803             }
5804             block[i]= level;
5805         }
5806     }
5807 }
5808
5809 static int dct_quantize_trellis_c(MpegEncContext *s,
5810                         DCTELEM *block, int n,
5811                         int qscale, int *overflow){
5812     const int *qmat;
5813     const uint8_t *scantable= s->intra_scantable.scantable;
5814     const uint8_t *perm_scantable= s->intra_scantable.permutated;
5815     int max=0;
5816     unsigned int threshold1, threshold2;
5817     int bias=0;
5818     int run_tab[65];
5819     int level_tab[65];
5820     int score_tab[65];
5821     int survivor[65];
5822     int survivor_count;
5823     int last_run=0;
5824     int last_level=0;
5825     int last_score= 0;
5826     int last_i;
5827     int coeff[2][64];
5828     int coeff_count[64];
5829     int qmul, qadd, start_i, last_non_zero, i, dc;
5830     const int esc_length= s->ac_esc_length;
5831     uint8_t * length;
5832     uint8_t * last_length;
5833     const int lambda= s->lambda2 >> (FF_LAMBDA_SHIFT - 6);
5834
5835     s->dsp.fdct (block);
5836
5837     if(s->dct_error_sum)
5838         s->denoise_dct(s, block);
5839     qmul= qscale*16;
5840     qadd= ((qscale-1)|1)*8;
5841
5842     if (s->mb_intra) {
5843         int q;
5844         if (!s->h263_aic) {
5845             if (n < 4)
5846                 q = s->y_dc_scale;
5847             else
5848                 q = s->c_dc_scale;
5849             q = q << 3;
5850         } else{
5851             /* For AIC we skip quant/dequant of INTRADC */
5852             q = 1 << 3;
5853             qadd=0;
5854         }
5855
5856         /* note: block[0] is assumed to be positive */
5857         block[0] = (block[0] + (q >> 1)) / q;
5858         start_i = 1;
5859         last_non_zero = 0;
5860         qmat = s->q_intra_matrix[qscale];
5861         if(s->mpeg_quant || s->out_format == FMT_MPEG1)
5862             bias= 1<<(QMAT_SHIFT-1);
5863         length     = s->intra_ac_vlc_length;
5864         last_length= s->intra_ac_vlc_last_length;
5865     } else {
5866         start_i = 0;
5867         last_non_zero = -1;
5868         qmat = s->q_inter_matrix[qscale];
5869         length     = s->inter_ac_vlc_length;
5870         last_length= s->inter_ac_vlc_last_length;
5871     }
5872     last_i= start_i;
5873
5874     threshold1= (1<<QMAT_SHIFT) - bias - 1;
5875     threshold2= (threshold1<<1);
5876
5877     for(i=63; i>=start_i; i--) {
5878         const int j = scantable[i];
5879         int level = block[j] * qmat[j];
5880
5881         if(((unsigned)(level+threshold1))>threshold2){
5882             last_non_zero = i;
5883             break;
5884         }
5885     }
5886
5887     for(i=start_i; i<=last_non_zero; i++) {
5888         const int j = scantable[i];
5889         int level = block[j] * qmat[j];
5890
5891 //        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
5892 //           || bias-level >= (1<<(QMAT_SHIFT - 3))){
5893         if(((unsigned)(level+threshold1))>threshold2){
5894             if(level>0){
5895                 level= (bias + level)>>QMAT_SHIFT;
5896                 coeff[0][i]= level;
5897                 coeff[1][i]= level-1;
5898 //                coeff[2][k]= level-2;
5899             }else{
5900                 level= (bias - level)>>QMAT_SHIFT;
5901                 coeff[0][i]= -level;
5902                 coeff[1][i]= -level+1;
5903 //                coeff[2][k]= -level+2;
5904             }
5905             coeff_count[i]= FFMIN(level, 2);
5906             assert(coeff_count[i]);
5907             max |=level;
5908         }else{
5909             coeff[0][i]= (level>>31)|1;
5910             coeff_count[i]= 1;
5911         }
5912     }
5913
5914     *overflow= s->max_qcoeff < max; //overflow might have happened
5915
5916     if(last_non_zero < start_i){
5917         memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
5918         return last_non_zero;
5919     }
5920
5921     score_tab[start_i]= 0;
5922     survivor[0]= start_i;
5923     survivor_count= 1;
5924
5925     for(i=start_i; i<=last_non_zero; i++){
5926         int level_index, j;
5927         const int dct_coeff= FFABS(block[ scantable[i] ]);
5928         const int zero_distoration= dct_coeff*dct_coeff;
5929         int best_score=256*256*256*120;
5930         for(level_index=0; level_index < coeff_count[i]; level_index++){
5931             int distoration;
5932             int level= coeff[level_index][i];
5933             const int alevel= FFABS(level);
5934             int unquant_coeff;
5935
5936             assert(level);
5937
5938             if(s->out_format == FMT_H263){
5939                 unquant_coeff= alevel*qmul + qadd;
5940             }else{ //MPEG1
5941                 j= s->dsp.idct_permutation[ scantable[i] ]; //FIXME optimize
5942                 if(s->mb_intra){
5943                         unquant_coeff = (int)(  alevel  * qscale * s->intra_matrix[j]) >> 3;
5944                         unquant_coeff =   (unquant_coeff - 1) | 1;
5945                 }else{
5946                         unquant_coeff = (((  alevel  << 1) + 1) * qscale * ((int) s->inter_matrix[j])) >> 4;
5947                         unquant_coeff =   (unquant_coeff - 1) | 1;
5948                 }
5949                 unquant_coeff<<= 3;
5950             }
5951
5952             distoration= (unquant_coeff - dct_coeff) * (unquant_coeff - dct_coeff) - zero_distoration;
5953             level+=64;
5954             if((level&(~127)) == 0){
5955                 for(j=survivor_count-1; j>=0; j--){
5956                     int run= i - survivor[j];
5957                     int score= distoration + length[UNI_AC_ENC_INDEX(run, level)]*lambda;
5958                     score += score_tab[i-run];
5959
5960                     if(score < best_score){
5961                         best_score= score;
5962                         run_tab[i+1]= run;
5963                         level_tab[i+1]= level-64;
5964                     }
5965                 }
5966
5967                 if(s->out_format == FMT_H263){
5968                     for(j=survivor_count-1; j>=0; j--){
5969                         int run= i - survivor[j];
5970                         int score= distoration + last_length[UNI_AC_ENC_INDEX(run, level)]*lambda;
5971                         score += score_tab[i-run];
5972                         if(score < last_score){
5973                             last_score= score;
5974                             last_run= run;
5975                             last_level= level-64;
5976                             last_i= i+1;
5977                         }
5978                     }
5979                 }
5980             }else{
5981                 distoration += esc_length*lambda;
5982                 for(j=survivor_count-1; j>=0; j--){
5983                     int run= i - survivor[j];
5984                     int score= distoration + score_tab[i-run];
5985
5986                     if(score < best_score){
5987                         best_score= score;
5988                         run_tab[i+1]= run;
5989                         level_tab[i+1]= level-64;
5990                     }
5991                 }
5992
5993                 if(s->out_format == FMT_H263){
5994                   for(j=survivor_count-1; j>=0; j--){
5995                         int run= i - survivor[j];
5996                         int score= distoration + score_tab[i-run];
5997                         if(score < last_score){
5998                             last_score= score;
5999                             last_run= run;
6000                             last_level= level-64;
6001                             last_i= i+1;
6002                         }
6003                     }
6004                 }
6005             }
6006         }
6007
6008         score_tab[i+1]= best_score;
6009
6010         //Note: there is a vlc code in mpeg4 which is 1 bit shorter then another one with a shorter run and the same level
6011         if(last_non_zero <= 27){
6012             for(; survivor_count; survivor_count--){
6013                 if(score_tab[ survivor[survivor_count-1] ] <= best_score)
6014                     break;
6015             }
6016         }else{
6017             for(; survivor_count; survivor_count--){
6018                 if(score_tab[ survivor[survivor_count-1] ] <= best_score + lambda)
6019                     break;
6020             }
6021         }
6022
6023         survivor[ survivor_count++ ]= i+1;
6024     }
6025
6026     if(s->out_format != FMT_H263){
6027         last_score= 256*256*256*120;
6028         for(i= survivor[0]; i<=last_non_zero + 1; i++){
6029             int score= score_tab[i];
6030             if(i) score += lambda*2; //FIXME exacter?
6031
6032             if(score < last_score){
6033                 last_score= score;
6034                 last_i= i;
6035                 last_level= level_tab[i];
6036                 last_run= run_tab[i];
6037             }
6038         }
6039     }
6040
6041     s->coded_score[n] = last_score;
6042
6043     dc= FFABS(block[0]);
6044     last_non_zero= last_i - 1;
6045     memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
6046
6047     if(last_non_zero < start_i)
6048         return last_non_zero;
6049
6050     if(last_non_zero == 0 && start_i == 0){
6051         int best_level= 0;
6052         int best_score= dc * dc;
6053
6054         for(i=0; i<coeff_count[0]; i++){
6055             int level= coeff[i][0];
6056             int alevel= FFABS(level);
6057             int unquant_coeff, score, distortion;
6058
6059             if(s->out_format == FMT_H263){
6060                     unquant_coeff= (alevel*qmul + qadd)>>3;
6061             }else{ //MPEG1
6062                     unquant_coeff = (((  alevel  << 1) + 1) * qscale * ((int) s->inter_matrix[0])) >> 4;
6063                     unquant_coeff =   (unquant_coeff - 1) | 1;
6064             }
6065             unquant_coeff = (unquant_coeff + 4) >> 3;
6066             unquant_coeff<<= 3 + 3;
6067
6068             distortion= (unquant_coeff - dc) * (unquant_coeff - dc);
6069             level+=64;
6070             if((level&(~127)) == 0) score= distortion + last_length[UNI_AC_ENC_INDEX(0, level)]*lambda;
6071             else                    score= distortion + esc_length*lambda;
6072
6073             if(score < best_score){
6074                 best_score= score;
6075                 best_level= level - 64;
6076             }
6077         }
6078         block[0]= best_level;
6079         s->coded_score[n] = best_score - dc*dc;
6080         if(best_level == 0) return -1;
6081         else                return last_non_zero;
6082     }
6083
6084     i= last_i;
6085     assert(last_level);
6086
6087     block[ perm_scantable[last_non_zero] ]= last_level;
6088     i -= last_run + 1;
6089
6090     for(; i>start_i; i -= run_tab[i] + 1){
6091         block[ perm_scantable[i-1] ]= level_tab[i];
6092     }
6093
6094     return last_non_zero;
6095 }
6096
6097 //#define REFINE_STATS 1
6098 static int16_t basis[64][64];
6099
6100 static void build_basis(uint8_t *perm){
6101     int i, j, x, y;
6102     emms_c();
6103     for(i=0; i<8; i++){
6104         for(j=0; j<8; j++){
6105             for(y=0; y<8; y++){
6106                 for(x=0; x<8; x++){
6107                     double s= 0.25*(1<<BASIS_SHIFT);
6108                     int index= 8*i + j;
6109                     int perm_index= perm[index];
6110                     if(i==0) s*= sqrt(0.5);
6111                     if(j==0) s*= sqrt(0.5);
6112                     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)));
6113                 }
6114             }
6115         }
6116     }
6117 }
6118
6119 static int dct_quantize_refine(MpegEncContext *s, //FIXME breaks denoise?
6120                         DCTELEM *block, int16_t *weight, DCTELEM *orig,
6121                         int n, int qscale){
6122     int16_t rem[64];
6123     DECLARE_ALIGNED_16(DCTELEM, d1[64]);
6124     const int *qmat;
6125     const uint8_t *scantable= s->intra_scantable.scantable;
6126     const uint8_t *perm_scantable= s->intra_scantable.permutated;
6127 //    unsigned int threshold1, threshold2;
6128 //    int bias=0;
6129     int run_tab[65];
6130     int prev_run=0;
6131     int prev_level=0;
6132     int qmul, qadd, start_i, last_non_zero, i, dc;
6133     uint8_t * length;
6134     uint8_t * last_length;
6135     int lambda;
6136     int rle_index, run, q, sum;
6137 #ifdef REFINE_STATS
6138 static int count=0;
6139 static int after_last=0;
6140 static int to_zero=0;
6141 static int from_zero=0;
6142 static int raise=0;
6143 static int lower=0;
6144 static int messed_sign=0;
6145 #endif
6146
6147     if(basis[0][0] == 0)
6148         build_basis(s->dsp.idct_permutation);
6149
6150     qmul= qscale*2;
6151     qadd= (qscale-1)|1;
6152     if (s->mb_intra) {
6153         if (!s->h263_aic) {
6154             if (n < 4)
6155                 q = s->y_dc_scale;
6156             else
6157                 q = s->c_dc_scale;
6158         } else{
6159             /* For AIC we skip quant/dequant of INTRADC */
6160             q = 1;
6161             qadd=0;
6162         }
6163         q <<= RECON_SHIFT-3;
6164         /* note: block[0] is assumed to be positive */
6165         dc= block[0]*q;
6166 //        block[0] = (block[0] + (q >> 1)) / q;
6167         start_i = 1;
6168         qmat = s->q_intra_matrix[qscale];
6169 //        if(s->mpeg_quant || s->out_format == FMT_MPEG1)
6170 //            bias= 1<<(QMAT_SHIFT-1);
6171         length     = s->intra_ac_vlc_length;
6172         last_length= s->intra_ac_vlc_last_length;
6173     } else {
6174         dc= 0;
6175         start_i = 0;
6176         qmat = s->q_inter_matrix[qscale];
6177         length     = s->inter_ac_vlc_length;
6178         last_length= s->inter_ac_vlc_last_length;
6179     }
6180     last_non_zero = s->block_last_index[n];
6181
6182 #ifdef REFINE_STATS
6183 {START_TIMER
6184 #endif
6185     dc += (1<<(RECON_SHIFT-1));
6186     for(i=0; i<64; i++){
6187         rem[i]= dc - (orig[i]<<RECON_SHIFT); //FIXME  use orig dirrectly instead of copying to rem[]
6188     }
6189 #ifdef REFINE_STATS
6190 STOP_TIMER("memset rem[]")}
6191 #endif
6192     sum=0;
6193     for(i=0; i<64; i++){
6194         int one= 36;
6195         int qns=4;
6196         int w;
6197
6198         w= FFABS(weight[i]) + qns*one;
6199         w= 15 + (48*qns*one + w/2)/w; // 16 .. 63
6200
6201         weight[i] = w;
6202 //        w=weight[i] = (63*qns + (w/2)) / w;
6203
6204         assert(w>0);
6205         assert(w<(1<<6));
6206         sum += w*w;
6207     }
6208     lambda= sum*(uint64_t)s->lambda2 >> (FF_LAMBDA_SHIFT - 6 + 6 + 6 + 6);
6209 #ifdef REFINE_STATS
6210 {START_TIMER
6211 #endif
6212     run=0;
6213     rle_index=0;
6214     for(i=start_i; i<=last_non_zero; i++){
6215         int j= perm_scantable[i];
6216         const int level= block[j];
6217         int coeff;
6218
6219         if(level){
6220             if(level<0) coeff= qmul*level - qadd;
6221             else        coeff= qmul*level + qadd;
6222             run_tab[rle_index++]=run;
6223             run=0;
6224
6225             s->dsp.add_8x8basis(rem, basis[j], coeff);
6226         }else{
6227             run++;
6228         }
6229     }
6230 #ifdef REFINE_STATS
6231 if(last_non_zero>0){
6232 STOP_TIMER("init rem[]")
6233 }
6234 }
6235
6236 {START_TIMER
6237 #endif
6238     for(;;){
6239         int best_score=s->dsp.try_8x8basis(rem, weight, basis[0], 0);
6240         int best_coeff=0;
6241         int best_change=0;
6242         int run2, best_unquant_change=0, analyze_gradient;
6243 #ifdef REFINE_STATS
6244 {START_TIMER
6245 #endif
6246         analyze_gradient = last_non_zero > 2 || s->avctx->quantizer_noise_shaping >= 3;
6247
6248         if(analyze_gradient){
6249 #ifdef REFINE_STATS
6250 {START_TIMER
6251 #endif
6252             for(i=0; i<64; i++){
6253                 int w= weight[i];
6254
6255                 d1[i] = (rem[i]*w*w + (1<<(RECON_SHIFT+12-1)))>>(RECON_SHIFT+12);
6256             }
6257 #ifdef REFINE_STATS
6258 STOP_TIMER("rem*w*w")}
6259 {START_TIMER
6260 #endif
6261             s->dsp.fdct(d1);
6262 #ifdef REFINE_STATS
6263 STOP_TIMER("dct")}
6264 #endif
6265         }
6266
6267         if(start_i){
6268             const int level= block[0];
6269             int change, old_coeff;
6270
6271             assert(s->mb_intra);
6272
6273             old_coeff= q*level;
6274
6275             for(change=-1; change<=1; change+=2){
6276                 int new_level= level + change;
6277                 int score, new_coeff;
6278
6279                 new_coeff= q*new_level;
6280                 if(new_coeff >= 2048 || new_coeff < 0)
6281                     continue;
6282
6283                 score= s->dsp.try_8x8basis(rem, weight, basis[0], new_coeff - old_coeff);
6284                 if(score<best_score){
6285                     best_score= score;
6286                     best_coeff= 0;
6287                     best_change= change;
6288                     best_unquant_change= new_coeff - old_coeff;
6289                 }
6290             }
6291         }
6292
6293         run=0;
6294         rle_index=0;
6295         run2= run_tab[rle_index++];
6296         prev_level=0;
6297         prev_run=0;
6298
6299         for(i=start_i; i<64; i++){
6300             int j= perm_scantable[i];
6301             const int level= block[j];
6302             int change, old_coeff;
6303
6304             if(s->avctx->quantizer_noise_shaping < 3 && i > last_non_zero + 1)
6305                 break;
6306
6307             if(level){
6308                 if(level<0) old_coeff= qmul*level - qadd;
6309                 else        old_coeff= qmul*level + qadd;
6310                 run2= run_tab[rle_index++]; //FIXME ! maybe after last
6311             }else{
6312                 old_coeff=0;
6313                 run2--;
6314                 assert(run2>=0 || i >= last_non_zero );
6315             }
6316
6317             for(change=-1; change<=1; change+=2){
6318                 int new_level= level + change;
6319                 int score, new_coeff, unquant_change;
6320
6321                 score=0;
6322                 if(s->avctx->quantizer_noise_shaping < 2 && FFABS(new_level) > FFABS(level))
6323                    continue;
6324
6325                 if(new_level){
6326                     if(new_level<0) new_coeff= qmul*new_level - qadd;
6327                     else            new_coeff= qmul*new_level + qadd;
6328                     if(new_coeff >= 2048 || new_coeff <= -2048)
6329                         continue;
6330                     //FIXME check for overflow
6331
6332                     if(level){
6333                         if(level < 63 && level > -63){
6334                             if(i < last_non_zero)
6335                                 score +=   length[UNI_AC_ENC_INDEX(run, new_level+64)]
6336                                          - length[UNI_AC_ENC_INDEX(run, level+64)];
6337                             else
6338                                 score +=   last_length[UNI_AC_ENC_INDEX(run, new_level+64)]
6339                                          - last_length[UNI_AC_ENC_INDEX(run, level+64)];
6340                         }
6341                     }else{
6342                         assert(FFABS(new_level)==1);
6343
6344                         if(analyze_gradient){
6345                             int g= d1[ scantable[i] ];
6346                             if(g && (g^new_level) >= 0)
6347                                 continue;
6348                         }
6349
6350                         if(i < last_non_zero){
6351                             int next_i= i + run2 + 1;
6352                             int next_level= block[ perm_scantable[next_i] ] + 64;
6353
6354                             if(next_level&(~127))
6355                                 next_level= 0;
6356
6357                             if(next_i < last_non_zero)
6358                                 score +=   length[UNI_AC_ENC_INDEX(run, 65)]
6359                                          + length[UNI_AC_ENC_INDEX(run2, next_level)]
6360                                          - length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)];
6361                             else
6362                                 score +=  length[UNI_AC_ENC_INDEX(run, 65)]
6363                                         + last_length[UNI_AC_ENC_INDEX(run2, next_level)]
6364                                         - last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)];
6365                         }else{
6366                             score += last_length[UNI_AC_ENC_INDEX(run, 65)];
6367                             if(prev_level){
6368                                 score +=  length[UNI_AC_ENC_INDEX(prev_run, prev_level)]
6369                                         - last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)];
6370                             }
6371                         }
6372                     }
6373                 }else{
6374                     new_coeff=0;
6375                     assert(FFABS(level)==1);
6376
6377                     if(i < last_non_zero){
6378                         int next_i= i + run2 + 1;
6379                         int next_level= block[ perm_scantable[next_i] ] + 64;
6380
6381                         if(next_level&(~127))
6382                             next_level= 0;
6383
6384                         if(next_i < last_non_zero)
6385                             score +=   length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)]
6386                                      - length[UNI_AC_ENC_INDEX(run2, next_level)]
6387                                      - length[UNI_AC_ENC_INDEX(run, 65)];
6388                         else
6389                             score +=   last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)]
6390                                      - last_length[UNI_AC_ENC_INDEX(run2, next_level)]
6391                                      - length[UNI_AC_ENC_INDEX(run, 65)];
6392                     }else{
6393                         score += -last_length[UNI_AC_ENC_INDEX(run, 65)];
6394                         if(prev_level){
6395                             score +=  last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)]
6396                                     - length[UNI_AC_ENC_INDEX(prev_run, prev_level)];
6397                         }
6398                     }
6399                 }
6400
6401                 score *= lambda;
6402
6403                 unquant_change= new_coeff - old_coeff;
6404                 assert((score < 100*lambda && score > -100*lambda) || lambda==0);
6405
6406                 score+= s->dsp.try_8x8basis(rem, weight, basis[j], unquant_change);
6407                 if(score<best_score){
6408                     best_score= score;
6409                     best_coeff= i;
6410                     best_change= change;
6411                     best_unquant_change= unquant_change;
6412                 }
6413             }
6414             if(level){
6415                 prev_level= level + 64;
6416                 if(prev_level&(~127))
6417                     prev_level= 0;
6418                 prev_run= run;
6419                 run=0;
6420             }else{
6421                 run++;
6422             }
6423         }
6424 #ifdef REFINE_STATS
6425 STOP_TIMER("iterative step")}
6426 #endif
6427
6428         if(best_change){
6429             int j= perm_scantable[ best_coeff ];
6430
6431             block[j] += best_change;
6432
6433             if(best_coeff > last_non_zero){
6434                 last_non_zero= best_coeff;
6435                 assert(block[j]);
6436 #ifdef REFINE_STATS
6437 after_last++;
6438 #endif
6439             }else{
6440 #ifdef REFINE_STATS
6441 if(block[j]){
6442     if(block[j] - best_change){
6443         if(FFABS(block[j]) > FFABS(block[j] - best_change)){
6444             raise++;
6445         }else{
6446             lower++;
6447         }
6448     }else{
6449         from_zero++;
6450     }
6451 }else{
6452     to_zero++;
6453 }
6454 #endif
6455                 for(; last_non_zero>=start_i; last_non_zero--){
6456                     if(block[perm_scantable[last_non_zero]])
6457                         break;
6458                 }
6459             }
6460 #ifdef REFINE_STATS
6461 count++;
6462 if(256*256*256*64 % count == 0){
6463     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);
6464 }
6465 #endif
6466             run=0;
6467             rle_index=0;
6468             for(i=start_i; i<=last_non_zero; i++){
6469                 int j= perm_scantable[i];
6470                 const int level= block[j];
6471
6472                  if(level){
6473                      run_tab[rle_index++]=run;
6474                      run=0;
6475                  }else{
6476                      run++;
6477                  }
6478             }
6479
6480             s->dsp.add_8x8basis(rem, basis[j], best_unquant_change);
6481         }else{
6482             break;
6483         }
6484     }
6485 #ifdef REFINE_STATS
6486 if(last_non_zero>0){
6487 STOP_TIMER("iterative search")
6488 }
6489 }
6490 #endif
6491
6492     return last_non_zero;
6493 }
6494
6495 static int dct_quantize_c(MpegEncContext *s,
6496                         DCTELEM *block, int n,
6497                         int qscale, int *overflow)
6498 {
6499     int i, j, level, last_non_zero, q, start_i;
6500     const int *qmat;
6501     const uint8_t *scantable= s->intra_scantable.scantable;
6502     int bias;
6503     int max=0;
6504     unsigned int threshold1, threshold2;
6505
6506     s->dsp.fdct (block);
6507
6508     if(s->dct_error_sum)
6509         s->denoise_dct(s, block);
6510
6511     if (s->mb_intra) {
6512         if (!s->h263_aic) {
6513             if (n < 4)
6514                 q = s->y_dc_scale;
6515             else
6516                 q = s->c_dc_scale;
6517             q = q << 3;
6518         } else
6519             /* For AIC we skip quant/dequant of INTRADC */
6520             q = 1 << 3;
6521
6522         /* note: block[0] is assumed to be positive */
6523         block[0] = (block[0] + (q >> 1)) / q;
6524         start_i = 1;
6525         last_non_zero = 0;
6526         qmat = s->q_intra_matrix[qscale];
6527         bias= s->intra_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT);
6528     } else {
6529         start_i = 0;
6530         last_non_zero = -1;
6531         qmat = s->q_inter_matrix[qscale];
6532         bias= s->inter_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT);
6533     }
6534     threshold1= (1<<QMAT_SHIFT) - bias - 1;
6535     threshold2= (threshold1<<1);
6536     for(i=63;i>=start_i;i--) {
6537         j = scantable[i];
6538         level = block[j] * qmat[j];
6539
6540         if(((unsigned)(level+threshold1))>threshold2){
6541             last_non_zero = i;
6542             break;
6543         }else{
6544             block[j]=0;
6545         }
6546     }
6547     for(i=start_i; i<=last_non_zero; i++) {
6548         j = scantable[i];
6549         level = block[j] * qmat[j];
6550
6551 //        if(   bias+level >= (1<<QMAT_SHIFT)
6552 //           || bias-level >= (1<<QMAT_SHIFT)){
6553         if(((unsigned)(level+threshold1))>threshold2){
6554             if(level>0){
6555                 level= (bias + level)>>QMAT_SHIFT;
6556                 block[j]= level;
6557             }else{
6558                 level= (bias - level)>>QMAT_SHIFT;
6559                 block[j]= -level;
6560             }
6561             max |=level;
6562         }else{
6563             block[j]=0;
6564         }
6565     }
6566     *overflow= s->max_qcoeff < max; //overflow might have happened
6567
6568     /* we need this permutation so that we correct the IDCT, we only permute the !=0 elements */
6569     if (s->dsp.idct_permutation_type != FF_NO_IDCT_PERM)
6570         ff_block_permute(block, s->dsp.idct_permutation, scantable, last_non_zero);
6571
6572     return last_non_zero;
6573 }
6574
6575 #endif //CONFIG_ENCODERS
6576
6577 static void dct_unquantize_mpeg1_intra_c(MpegEncContext *s,
6578                                    DCTELEM *block, int n, int qscale)
6579 {
6580     int i, level, nCoeffs;
6581     const uint16_t *quant_matrix;
6582
6583     nCoeffs= s->block_last_index[n];
6584
6585     if (n < 4)
6586         block[0] = block[0] * s->y_dc_scale;
6587     else
6588         block[0] = block[0] * s->c_dc_scale;
6589     /* XXX: only mpeg1 */
6590     quant_matrix = s->intra_matrix;
6591     for(i=1;i<=nCoeffs;i++) {
6592         int j= s->intra_scantable.permutated[i];
6593         level = block[j];
6594         if (level) {
6595             if (level < 0) {
6596                 level = -level;
6597                 level = (int)(level * qscale * quant_matrix[j]) >> 3;
6598                 level = (level - 1) | 1;
6599                 level = -level;
6600             } else {
6601                 level = (int)(level * qscale * quant_matrix[j]) >> 3;
6602                 level = (level - 1) | 1;
6603             }
6604             block[j] = level;
6605         }
6606     }
6607 }
6608
6609 static void dct_unquantize_mpeg1_inter_c(MpegEncContext *s,
6610                                    DCTELEM *block, int n, int qscale)
6611 {
6612     int i, level, nCoeffs;
6613     const uint16_t *quant_matrix;
6614
6615     nCoeffs= s->block_last_index[n];
6616
6617     quant_matrix = s->inter_matrix;
6618     for(i=0; i<=nCoeffs; i++) {
6619         int j= s->intra_scantable.permutated[i];
6620         level = block[j];
6621         if (level) {
6622             if (level < 0) {
6623                 level = -level;
6624                 level = (((level << 1) + 1) * qscale *
6625                          ((int) (quant_matrix[j]))) >> 4;
6626                 level = (level - 1) | 1;
6627                 level = -level;
6628             } else {
6629                 level = (((level << 1) + 1) * qscale *
6630                          ((int) (quant_matrix[j]))) >> 4;
6631                 level = (level - 1) | 1;
6632             }
6633             block[j] = level;
6634         }
6635     }
6636 }
6637
6638 static void dct_unquantize_mpeg2_intra_c(MpegEncContext *s,
6639                                    DCTELEM *block, int n, int qscale)
6640 {
6641     int i, level, nCoeffs;
6642     const uint16_t *quant_matrix;
6643
6644     if(s->alternate_scan) nCoeffs= 63;
6645     else nCoeffs= s->block_last_index[n];
6646
6647     if (n < 4)
6648         block[0] = block[0] * s->y_dc_scale;
6649     else
6650         block[0] = block[0] * s->c_dc_scale;
6651     quant_matrix = s->intra_matrix;
6652     for(i=1;i<=nCoeffs;i++) {
6653         int j= s->intra_scantable.permutated[i];
6654         level = block[j];
6655         if (level) {
6656             if (level < 0) {
6657                 level = -level;
6658                 level = (int)(level * qscale * quant_matrix[j]) >> 3;
6659                 level = -level;
6660             } else {
6661                 level = (int)(level * qscale * quant_matrix[j]) >> 3;
6662             }
6663             block[j] = level;
6664         }
6665     }
6666 }
6667
6668 static void dct_unquantize_mpeg2_intra_bitexact(MpegEncContext *s,
6669                                    DCTELEM *block, int n, int qscale)
6670 {
6671     int i, level, nCoeffs;
6672     const uint16_t *quant_matrix;
6673     int sum=-1;
6674
6675     if(s->alternate_scan) nCoeffs= 63;
6676     else nCoeffs= s->block_last_index[n];
6677
6678     if (n < 4)
6679         block[0] = block[0] * s->y_dc_scale;
6680     else
6681         block[0] = block[0] * s->c_dc_scale;
6682     quant_matrix = s->intra_matrix;
6683     for(i=1;i<=nCoeffs;i++) {
6684         int j= s->intra_scantable.permutated[i];
6685         level = block[j];
6686         if (level) {
6687             if (level < 0) {
6688                 level = -level;
6689                 level = (int)(level * qscale * quant_matrix[j]) >> 3;
6690                 level = -level;
6691             } else {
6692                 level = (int)(level * qscale * quant_matrix[j]) >> 3;
6693             }
6694             block[j] = level;
6695             sum+=level;
6696         }
6697     }
6698     block[63]^=sum&1;
6699 }
6700
6701 static void dct_unquantize_mpeg2_inter_c(MpegEncContext *s,
6702                                    DCTELEM *block, int n, int qscale)
6703 {
6704     int i, level, nCoeffs;
6705     const uint16_t *quant_matrix;
6706     int sum=-1;
6707
6708     if(s->alternate_scan) nCoeffs= 63;
6709     else nCoeffs= s->block_last_index[n];
6710
6711     quant_matrix = s->inter_matrix;
6712     for(i=0; i<=nCoeffs; i++) {
6713         int j= s->intra_scantable.permutated[i];
6714         level = block[j];
6715         if (level) {
6716             if (level < 0) {
6717                 level = -level;
6718                 level = (((level << 1) + 1) * qscale *
6719                          ((int) (quant_matrix[j]))) >> 4;
6720                 level = -level;
6721             } else {
6722                 level = (((level << 1) + 1) * qscale *
6723                          ((int) (quant_matrix[j]))) >> 4;
6724             }
6725             block[j] = level;
6726             sum+=level;
6727         }
6728     }
6729     block[63]^=sum&1;
6730 }
6731
6732 static void dct_unquantize_h263_intra_c(MpegEncContext *s,
6733                                   DCTELEM *block, int n, int qscale)
6734 {
6735     int i, level, qmul, qadd;
6736     int nCoeffs;
6737
6738     assert(s->block_last_index[n]>=0);
6739
6740     qmul = qscale << 1;
6741
6742     if (!s->h263_aic) {
6743         if (n < 4)
6744             block[0] = block[0] * s->y_dc_scale;
6745         else
6746             block[0] = block[0] * s->c_dc_scale;
6747         qadd = (qscale - 1) | 1;
6748     }else{
6749         qadd = 0;
6750     }
6751     if(s->ac_pred)
6752         nCoeffs=63;
6753     else
6754         nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
6755
6756     for(i=1; i<=nCoeffs; i++) {
6757         level = block[i];
6758         if (level) {
6759             if (level < 0) {
6760                 level = level * qmul - qadd;
6761             } else {
6762                 level = level * qmul + qadd;
6763             }
6764             block[i] = level;
6765         }
6766     }
6767 }
6768
6769 static void dct_unquantize_h263_inter_c(MpegEncContext *s,
6770                                   DCTELEM *block, int n, int qscale)
6771 {
6772     int i, level, qmul, qadd;
6773     int nCoeffs;
6774
6775     assert(s->block_last_index[n]>=0);
6776
6777     qadd = (qscale - 1) | 1;
6778     qmul = qscale << 1;
6779
6780     nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
6781
6782     for(i=0; i<=nCoeffs; i++) {
6783         level = block[i];
6784         if (level) {
6785             if (level < 0) {
6786                 level = level * qmul - qadd;
6787             } else {
6788                 level = level * qmul + qadd;
6789             }
6790             block[i] = level;
6791         }
6792     }
6793 }
6794
6795 #ifdef CONFIG_ENCODERS
6796 AVCodec h263_encoder = {
6797     "h263",
6798     CODEC_TYPE_VIDEO,
6799     CODEC_ID_H263,
6800     sizeof(MpegEncContext),
6801     MPV_encode_init,
6802     MPV_encode_picture,
6803     MPV_encode_end,
6804     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
6805 };
6806
6807 AVCodec h263p_encoder = {
6808     "h263p",
6809     CODEC_TYPE_VIDEO,
6810     CODEC_ID_H263P,
6811     sizeof(MpegEncContext),
6812     MPV_encode_init,
6813     MPV_encode_picture,
6814     MPV_encode_end,
6815     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
6816 };
6817
6818 AVCodec flv_encoder = {
6819     "flv",
6820     CODEC_TYPE_VIDEO,
6821     CODEC_ID_FLV1,
6822     sizeof(MpegEncContext),
6823     MPV_encode_init,
6824     MPV_encode_picture,
6825     MPV_encode_end,
6826     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
6827 };
6828
6829 AVCodec rv10_encoder = {
6830     "rv10",
6831     CODEC_TYPE_VIDEO,
6832     CODEC_ID_RV10,
6833     sizeof(MpegEncContext),
6834     MPV_encode_init,
6835     MPV_encode_picture,
6836     MPV_encode_end,
6837     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
6838 };
6839
6840 AVCodec rv20_encoder = {
6841     "rv20",
6842     CODEC_TYPE_VIDEO,
6843     CODEC_ID_RV20,
6844     sizeof(MpegEncContext),
6845     MPV_encode_init,
6846     MPV_encode_picture,
6847     MPV_encode_end,
6848     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
6849 };
6850
6851 AVCodec mpeg4_encoder = {
6852     "mpeg4",
6853     CODEC_TYPE_VIDEO,
6854     CODEC_ID_MPEG4,
6855     sizeof(MpegEncContext),
6856     MPV_encode_init,
6857     MPV_encode_picture,
6858     MPV_encode_end,
6859     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
6860     .capabilities= CODEC_CAP_DELAY,
6861 };
6862
6863 AVCodec msmpeg4v1_encoder = {
6864     "msmpeg4v1",
6865     CODEC_TYPE_VIDEO,
6866     CODEC_ID_MSMPEG4V1,
6867     sizeof(MpegEncContext),
6868     MPV_encode_init,
6869     MPV_encode_picture,
6870     MPV_encode_end,
6871     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
6872 };
6873
6874 AVCodec msmpeg4v2_encoder = {
6875     "msmpeg4v2",
6876     CODEC_TYPE_VIDEO,
6877     CODEC_ID_MSMPEG4V2,
6878     sizeof(MpegEncContext),
6879     MPV_encode_init,
6880     MPV_encode_picture,
6881     MPV_encode_end,
6882     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
6883 };
6884
6885 AVCodec msmpeg4v3_encoder = {
6886     "msmpeg4",
6887     CODEC_TYPE_VIDEO,
6888     CODEC_ID_MSMPEG4V3,
6889     sizeof(MpegEncContext),
6890     MPV_encode_init,
6891     MPV_encode_picture,
6892     MPV_encode_end,
6893     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
6894 };
6895
6896 AVCodec wmv1_encoder = {
6897     "wmv1",
6898     CODEC_TYPE_VIDEO,
6899     CODEC_ID_WMV1,
6900     sizeof(MpegEncContext),
6901     MPV_encode_init,
6902     MPV_encode_picture,
6903     MPV_encode_end,
6904     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV420P, -1},
6905 };
6906
6907 AVCodec mjpeg_encoder = {
6908     "mjpeg",
6909     CODEC_TYPE_VIDEO,
6910     CODEC_ID_MJPEG,
6911     sizeof(MpegEncContext),
6912     MPV_encode_init,
6913     MPV_encode_picture,
6914     MPV_encode_end,
6915     .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUVJ420P, PIX_FMT_YUVJ422P, -1},
6916 };
6917
6918 #endif //CONFIG_ENCODERS