libavcodec/mpegvideo_common.h

   1 /*
   2  * The simplest mpeg encoder (well, it was the simplest!)
   3  * Copyright (c) 2000,2001 Fabrice Bellard.
   4  * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
   5  *
   6  * 4MV & hq & B-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
   7  *
   8  * This file is part of FFmpeg.
   9  *
  10  * FFmpeg is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU Lesser General Public
  12  * License as published by the Free Software Foundation; either
  13  * version 2.1 of the License, or (at your option) any later version.
  14  *
  15  * FFmpeg is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  * Lesser General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU Lesser General Public
  21  * License along with FFmpeg; if not, write to the Free Software
  22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  23  */
  24
  25 /**
  26  * @file mpegvideo_common.h
  27  * The simplest mpeg encoder (well, it was the simplest!).
  28  */
  29
  30 #ifndef FFMPEG_MPEGVIDEO_COMMON_H
  31 #define FFMPEG_MPEGVIDEO_COMMON_H
  32
  33 #include "avcodec.h"
  34 #include "dsputil.h"
  35 #include "mpegvideo.h"
  36 #include "mjpegenc.h"
  37 #include "msmpeg4.h"
  38 #include "faandct.h"
  39 #include <limits.h>
  40
  41 int dct_quantize_c(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow);
  42 int dct_quantize_trellis_c(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow);
  43 void  denoise_dct_c(MpegEncContext *s, DCTELEM *block);
  44 void copy_picture(Picture *dst, Picture *src);
  45
  46 /**
  47  * allocates a Picture
  48  * The pixels are allocated/set by calling get_buffer() if shared=0
  49  */
  50 int alloc_picture(MpegEncContext *s, Picture *pic, int shared);
  51
  52 /**
  53  * sets the given MpegEncContext to common defaults (same for encoding and decoding).
  54  * the changed fields will not depend upon the prior state of the MpegEncContext.
  55  */
  56 void MPV_common_defaults(MpegEncContext *s);
  57
  58 static inline void gmc1_motion(MpegEncContext *s,
  59                                uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
  60                                uint8_t **ref_picture)
  61 {
  62     uint8_t *ptr;
  63     int offset, src_x, src_y, linesize, uvlinesize;
  64     int motion_x, motion_y;
  65     int emu=0;
  66
  67     motion_x= s->sprite_offset[0][0];
  68     motion_y= s->sprite_offset[0][1];
  69     src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy+1));
  70     src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy+1));
  71     motion_x<<=(3-s->sprite_warping_accuracy);
  72     motion_y<<=(3-s->sprite_warping_accuracy);
  73     src_x = av_clip(src_x, -16, s->width);
  74     if (src_x == s->width)
  75         motion_x =0;
  76     src_y = av_clip(src_y, -16, s->height);
  77     if (src_y == s->height)
  78         motion_y =0;
  79
  80     linesize = s->linesize;
  81     uvlinesize = s->uvlinesize;
  82
  83     ptr = ref_picture[0] + (src_y * linesize) + src_x;
  84
  85     if(s->flags&CODEC_FLAG_EMU_EDGE){
  86         if(   (unsigned)src_x >= s->h_edge_pos - 17
  87            || (unsigned)src_y >= s->v_edge_pos - 17){
  88             ff_emulated_edge_mc(s->edge_emu_buffer, ptr, linesize, 17, 17, src_x, src_y, s->h_edge_pos, s->v_edge_pos);
  89             ptr= s->edge_emu_buffer;
  90         }
  91     }
  92
  93     if((motion_x|motion_y)&7){
  94         s->dsp.gmc1(dest_y  , ptr  , linesize, 16, motion_x&15, motion_y&15, 128 - s->no_rounding);
  95         s->dsp.gmc1(dest_y+8, ptr+8, linesize, 16, motion_x&15, motion_y&15, 128 - s->no_rounding);
  96     }else{
  97         int dxy;
  98
  99         dxy= ((motion_x>>3)&1) | ((motion_y>>2)&2);
 100         if (s->no_rounding){
 101             s->dsp.put_no_rnd_pixels_tab[0][dxy](dest_y, ptr, linesize, 16);
 102         }else{
 103             s->dsp.put_pixels_tab       [0][dxy](dest_y, ptr, linesize, 16);
 104         }
 105     }
 106
 107     if(ENABLE_GRAY && s->flags&CODEC_FLAG_GRAY) return;
 108
 109     motion_x= s->sprite_offset[1][0];
 110     motion_y= s->sprite_offset[1][1];
 111     src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy+1));
 112     src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy+1));
 113     motion_x<<=(3-s->sprite_warping_accuracy);
 114     motion_y<<=(3-s->sprite_warping_accuracy);
 115     src_x = av_clip(src_x, -8, s->width>>1);
 116     if (src_x == s->width>>1)
 117         motion_x =0;
 118     src_y = av_clip(src_y, -8, s->height>>1);
 119     if (src_y == s->height>>1)
 120         motion_y =0;
 121
 122     offset = (src_y * uvlinesize) + src_x;
 123     ptr = ref_picture[1] + offset;
 124     if(s->flags&CODEC_FLAG_EMU_EDGE){
 125         if(   (unsigned)src_x >= (s->h_edge_pos>>1) - 9
 126            || (unsigned)src_y >= (s->v_edge_pos>>1) - 9){
 127             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);
 128             ptr= s->edge_emu_buffer;
 129             emu=1;
 130         }
 131     }
 132     s->dsp.gmc1(dest_cb, ptr, uvlinesize, 8, motion_x&15, motion_y&15, 128 - s->no_rounding);
 133
 134     ptr = ref_picture[2] + offset;
 135     if(emu){
 136         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);
 137         ptr= s->edge_emu_buffer;
 138     }
 139     s->dsp.gmc1(dest_cr, ptr, uvlinesize, 8, motion_x&15, motion_y&15, 128 - s->no_rounding);
 140
 141     return;
 142 }
 143
 144 static inline void gmc_motion(MpegEncContext *s,
 145                                uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 146                                uint8_t **ref_picture)
 147 {
 148     uint8_t *ptr;
 149     int linesize, uvlinesize;
 150     const int a= s->sprite_warping_accuracy;
 151     int ox, oy;
 152
 153     linesize = s->linesize;
 154     uvlinesize = s->uvlinesize;
 155
 156     ptr = ref_picture[0];
 157
 158     ox= s->sprite_offset[0][0] + s->sprite_delta[0][0]*s->mb_x*16 + s->sprite_delta[0][1]*s->mb_y*16;
 159     oy= s->sprite_offset[0][1] + s->sprite_delta[1][0]*s->mb_x*16 + s->sprite_delta[1][1]*s->mb_y*16;
 160
 161     s->dsp.gmc(dest_y, ptr, linesize, 16,
 162            ox,
 163            oy,
 164            s->sprite_delta[0][0], s->sprite_delta[0][1],
 165            s->sprite_delta[1][0], s->sprite_delta[1][1],
 166            a+1, (1<<(2*a+1)) - s->no_rounding,
 167            s->h_edge_pos, s->v_edge_pos);
 168     s->dsp.gmc(dest_y+8, ptr, linesize, 16,
 169            ox + s->sprite_delta[0][0]*8,
 170            oy + s->sprite_delta[1][0]*8,
 171            s->sprite_delta[0][0], s->sprite_delta[0][1],
 172            s->sprite_delta[1][0], s->sprite_delta[1][1],
 173            a+1, (1<<(2*a+1)) - s->no_rounding,
 174            s->h_edge_pos, s->v_edge_pos);
 175
 176     if(ENABLE_GRAY && s->flags&CODEC_FLAG_GRAY) return;
 177
 178     ox= s->sprite_offset[1][0] + s->sprite_delta[0][0]*s->mb_x*8 + s->sprite_delta[0][1]*s->mb_y*8;
 179     oy= s->sprite_offset[1][1] + s->sprite_delta[1][0]*s->mb_x*8 + s->sprite_delta[1][1]*s->mb_y*8;
 180
 181     ptr = ref_picture[1];
 182     s->dsp.gmc(dest_cb, ptr, uvlinesize, 8,
 183            ox,
 184            oy,
 185            s->sprite_delta[0][0], s->sprite_delta[0][1],
 186            s->sprite_delta[1][0], s->sprite_delta[1][1],
 187            a+1, (1<<(2*a+1)) - s->no_rounding,
 188            s->h_edge_pos>>1, s->v_edge_pos>>1);
 189
 190     ptr = ref_picture[2];
 191     s->dsp.gmc(dest_cr, ptr, uvlinesize, 8,
 192            ox,
 193            oy,
 194            s->sprite_delta[0][0], s->sprite_delta[0][1],
 195            s->sprite_delta[1][0], s->sprite_delta[1][1],
 196            a+1, (1<<(2*a+1)) - s->no_rounding,
 197            s->h_edge_pos>>1, s->v_edge_pos>>1);
 198 }
 199
 200 static inline int hpel_motion(MpegEncContext *s,
 201                                   uint8_t *dest, uint8_t *src,
 202                                   int field_based, int field_select,
 203                                   int src_x, int src_y,
 204                                   int width, int height, int stride,
 205                                   int h_edge_pos, int v_edge_pos,
 206                                   int w, int h, op_pixels_func *pix_op,
 207                                   int motion_x, int motion_y)
 208 {
 209     int dxy;
 210     int emu=0;
 211
 212     dxy = ((motion_y & 1) << 1) | (motion_x & 1);
 213     src_x += motion_x >> 1;
 214     src_y += motion_y >> 1;
 215
 216     /* WARNING: do no forget half pels */
 217     src_x = av_clip(src_x, -16, width); //FIXME unneeded for emu?
 218     if (src_x == width)
 219         dxy &= ~1;
 220     src_y = av_clip(src_y, -16, height);
 221     if (src_y == height)
 222         dxy &= ~2;
 223     src += src_y * stride + src_x;
 224
 225     if(s->unrestricted_mv && (s->flags&CODEC_FLAG_EMU_EDGE)){
 226         if(   (unsigned)src_x > h_edge_pos - (motion_x&1) - w
 227            || (unsigned)src_y > v_edge_pos - (motion_y&1) - h){
 228             ff_emulated_edge_mc(s->edge_emu_buffer, src, s->linesize, w+1, (h+1)<<field_based,
 229                              src_x, src_y<<field_based, h_edge_pos, s->v_edge_pos);
 230             src= s->edge_emu_buffer;
 231             emu=1;
 232         }
 233     }
 234     if(field_select)
 235         src += s->linesize;
 236     pix_op[dxy](dest, src, stride, h);
 237     return emu;
 238 }
 239
 240 static av_always_inline
 241 void mpeg_motion_internal(MpegEncContext *s,
 242                  uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 243                  int field_based, int bottom_field, int field_select,
 244                  uint8_t **ref_picture, op_pixels_func (*pix_op)[4],
 245                  int motion_x, int motion_y, int h, int is_mpeg12)
 246 {
 247     uint8_t *ptr_y, *ptr_cb, *ptr_cr;
 248     int dxy, uvdxy, mx, my, src_x, src_y,
 249         uvsrc_x, uvsrc_y, v_edge_pos, uvlinesize, linesize;
 250
 251 #if 0
 252 if(s->quarter_sample)
 253 {
 254     motion_x>>=1;
 255     motion_y>>=1;
 256 }
 257 #endif
 258
 259     v_edge_pos = s->v_edge_pos >> field_based;
 260     linesize   = s->current_picture.linesize[0] << field_based;
 261     uvlinesize = s->current_picture.linesize[1] << field_based;
 262
 263     dxy = ((motion_y & 1) << 1) | (motion_x & 1);
 264     src_x = s->mb_x* 16               + (motion_x >> 1);
 265     src_y =(s->mb_y<<(4-field_based)) + (motion_y >> 1);
 266
 267     if (!is_mpeg12 && s->out_format == FMT_H263) {
 268         if((s->workaround_bugs & FF_BUG_HPEL_CHROMA) && field_based){
 269             mx = (motion_x>>1)|(motion_x&1);
 270             my = motion_y >>1;
 271             uvdxy = ((my & 1) << 1) | (mx & 1);
 272             uvsrc_x = s->mb_x* 8               + (mx >> 1);
 273             uvsrc_y = (s->mb_y<<(3-field_based)) + (my >> 1);
 274         }else{
 275             uvdxy = dxy | (motion_y & 2) | ((motion_x & 2) >> 1);
 276             uvsrc_x = src_x>>1;
 277             uvsrc_y = src_y>>1;
 278         }
 279     }else if(!is_mpeg12 && s->out_format == FMT_H261){//even chroma mv's are full pel in H261
 280         mx = motion_x / 4;
 281         my = motion_y / 4;
 282         uvdxy = 0;
 283         uvsrc_x = s->mb_x*8 + mx;
 284         uvsrc_y = s->mb_y*8 + my;
 285     } else {
 286         if(s->chroma_y_shift){
 287             mx = motion_x / 2;
 288             my = motion_y / 2;
 289             uvdxy = ((my & 1) << 1) | (mx & 1);
 290             uvsrc_x = s->mb_x* 8               + (mx >> 1);
 291             uvsrc_y = (s->mb_y<<(3-field_based)) + (my >> 1);
 292         } else {
 293             if(s->chroma_x_shift){
 294             //Chroma422
 295                 mx = motion_x / 2;
 296                 uvdxy = ((motion_y & 1) << 1) | (mx & 1);
 297                 uvsrc_x = s->mb_x* 8           + (mx >> 1);
 298                 uvsrc_y = src_y;
 299             } else {
 300             //Chroma444
 301                 uvdxy = dxy;
 302                 uvsrc_x = src_x;
 303                 uvsrc_y = src_y;
 304             }
 305         }
 306     }
 307
 308     ptr_y  = ref_picture[0] + src_y * linesize + src_x;
 309     ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
 310     ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
 311
 312     if(   (unsigned)src_x > s->h_edge_pos - (motion_x&1) - 16
 313        || (unsigned)src_y >    v_edge_pos - (motion_y&1) - h){
 314             if(is_mpeg12 || s->codec_id == CODEC_ID_MPEG2VIDEO ||
 315                s->codec_id == CODEC_ID_MPEG1VIDEO){
 316                 av_log(s->avctx,AV_LOG_DEBUG,
 317                         "MPEG motion vector out of boundary\n");
 318                 return ;
 319             }
 320             ff_emulated_edge_mc(s->edge_emu_buffer, ptr_y, s->linesize,
 321                                 17, 17+field_based,
 322                                 src_x, src_y<<field_based,
 323                                 s->h_edge_pos, s->v_edge_pos);
 324             ptr_y = s->edge_emu_buffer;
 325             if(!ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
 326                 uint8_t *uvbuf= s->edge_emu_buffer+18*s->linesize;
 327                 ff_emulated_edge_mc(uvbuf ,
 328                                     ptr_cb, s->uvlinesize,
 329                                     9, 9+field_based,
 330                                     uvsrc_x, uvsrc_y<<field_based,
 331                                     s->h_edge_pos>>1, s->v_edge_pos>>1);
 332                 ff_emulated_edge_mc(uvbuf+16,
 333                                     ptr_cr, s->uvlinesize,
 334                                     9, 9+field_based,
 335                                     uvsrc_x, uvsrc_y<<field_based,
 336                                     s->h_edge_pos>>1, s->v_edge_pos>>1);
 337                 ptr_cb= uvbuf;
 338                 ptr_cr= uvbuf+16;
 339             }
 340     }
 341
 342     if(bottom_field){ //FIXME use this for field pix too instead of the obnoxious hack which changes picture.data
 343         dest_y += s->linesize;
 344         dest_cb+= s->uvlinesize;
 345         dest_cr+= s->uvlinesize;
 346     }
 347
 348     if(field_select){
 349         ptr_y += s->linesize;
 350         ptr_cb+= s->uvlinesize;
 351         ptr_cr+= s->uvlinesize;
 352     }
 353
 354     pix_op[0][dxy](dest_y, ptr_y, linesize, h);
 355
 356     if(!ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
 357         pix_op[s->chroma_x_shift][uvdxy]
 358                 (dest_cb, ptr_cb, uvlinesize, h >> s->chroma_y_shift);
 359         pix_op[s->chroma_x_shift][uvdxy]
 360                 (dest_cr, ptr_cr, uvlinesize, h >> s->chroma_y_shift);
 361     }
 362     if(!is_mpeg12 && (ENABLE_H261_ENCODER || ENABLE_H261_DECODER) &&
 363          s->out_format == FMT_H261){
 364         ff_h261_loop_filter(s);
 365     }
 366 }
 367 /* apply one mpeg motion vector to the three components */
 368 static av_always_inline
 369 void mpeg_motion(MpegEncContext *s,
 370                  uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 371                  int field_based, int bottom_field, int field_select,
 372                  uint8_t **ref_picture, op_pixels_func (*pix_op)[4],
 373                  int motion_x, int motion_y, int h)
 374 {
 375 #ifndef CONFIG_SMALL
 376     if(s->out_format == FMT_MPEG1)
 377         mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, field_based,
 378                     bottom_field, field_select, ref_picture, pix_op,
 379                     motion_x, motion_y, h, 1);
 380     else
 381 #endif
 382         mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, field_based,
 383                     bottom_field, field_select, ref_picture, pix_op,
 384                     motion_x, motion_y, h, 0);
 385 }
 386
 387 //FIXME move to dsputil, avg variant, 16x16 version
 388 static inline void put_obmc(uint8_t *dst, uint8_t *src[5], int stride){
 389     int x;
 390     uint8_t * const top   = src[1];
 391     uint8_t * const left  = src[2];
 392     uint8_t * const mid   = src[0];
 393     uint8_t * const right = src[3];
 394     uint8_t * const bottom= src[4];
 395 #define OBMC_FILTER(x, t, l, m, r, b)\
 396     dst[x]= (t*top[x] + l*left[x] + m*mid[x] + r*right[x] + b*bottom[x] + 4)>>3
 397 #define OBMC_FILTER4(x, t, l, m, r, b)\
 398     OBMC_FILTER(x         , t, l, m, r, b);\
 399     OBMC_FILTER(x+1       , t, l, m, r, b);\
 400     OBMC_FILTER(x  +stride, t, l, m, r, b);\
 401     OBMC_FILTER(x+1+stride, t, l, m, r, b);
 402
 403     x=0;
 404     OBMC_FILTER (x  , 2, 2, 4, 0, 0);
 405     OBMC_FILTER (x+1, 2, 1, 5, 0, 0);
 406     OBMC_FILTER4(x+2, 2, 1, 5, 0, 0);
 407     OBMC_FILTER4(x+4, 2, 0, 5, 1, 0);
 408     OBMC_FILTER (x+6, 2, 0, 5, 1, 0);
 409     OBMC_FILTER (x+7, 2, 0, 4, 2, 0);
 410     x+= stride;
 411     OBMC_FILTER (x  , 1, 2, 5, 0, 0);
 412     OBMC_FILTER (x+1, 1, 2, 5, 0, 0);
 413     OBMC_FILTER (x+6, 1, 0, 5, 2, 0);
 414     OBMC_FILTER (x+7, 1, 0, 5, 2, 0);
 415     x+= stride;
 416     OBMC_FILTER4(x  , 1, 2, 5, 0, 0);
 417     OBMC_FILTER4(x+2, 1, 1, 6, 0, 0);
 418     OBMC_FILTER4(x+4, 1, 0, 6, 1, 0);
 419     OBMC_FILTER4(x+6, 1, 0, 5, 2, 0);
 420     x+= 2*stride;
 421     OBMC_FILTER4(x  , 0, 2, 5, 0, 1);
 422     OBMC_FILTER4(x+2, 0, 1, 6, 0, 1);
 423     OBMC_FILTER4(x+4, 0, 0, 6, 1, 1);
 424     OBMC_FILTER4(x+6, 0, 0, 5, 2, 1);
 425     x+= 2*stride;
 426     OBMC_FILTER (x  , 0, 2, 5, 0, 1);
 427     OBMC_FILTER (x+1, 0, 2, 5, 0, 1);
 428     OBMC_FILTER4(x+2, 0, 1, 5, 0, 2);
 429     OBMC_FILTER4(x+4, 0, 0, 5, 1, 2);
 430     OBMC_FILTER (x+6, 0, 0, 5, 2, 1);
 431     OBMC_FILTER (x+7, 0, 0, 5, 2, 1);
 432     x+= stride;
 433     OBMC_FILTER (x  , 0, 2, 4, 0, 2);
 434     OBMC_FILTER (x+1, 0, 1, 5, 0, 2);
 435     OBMC_FILTER (x+6, 0, 0, 5, 1, 2);
 436     OBMC_FILTER (x+7, 0, 0, 4, 2, 2);
 437 }
 438
 439 /* obmc for 1 8x8 luma block */
 440 static inline void obmc_motion(MpegEncContext *s,
 441                                uint8_t *dest, uint8_t *src,
 442                                int src_x, int src_y,
 443                                op_pixels_func *pix_op,
 444                                int16_t mv[5][2]/* mid top left right bottom*/)
 445 #define MID    0
 446 {
 447     int i;
 448     uint8_t *ptr[5];
 449
 450     assert(s->quarter_sample==0);
 451
 452     for(i=0; i<5; i++){
 453         if(i && mv[i][0]==mv[MID][0] && mv[i][1]==mv[MID][1]){
 454             ptr[i]= ptr[MID];
 455         }else{
 456             ptr[i]= s->obmc_scratchpad + 8*(i&1) + s->linesize*8*(i>>1);
 457             hpel_motion(s, ptr[i], src, 0, 0,
 458                         src_x, src_y,
 459                         s->width, s->height, s->linesize,
 460                         s->h_edge_pos, s->v_edge_pos,
 461                         8, 8, pix_op,
 462                         mv[i][0], mv[i][1]);
 463         }
 464     }
 465
 466     put_obmc(dest, ptr, s->linesize);
 467 }
 468
 469 static inline void qpel_motion(MpegEncContext *s,
 470                                uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
 471                                int field_based, int bottom_field, int field_select,
 472                                uint8_t **ref_picture, op_pixels_func (*pix_op)[4],
 473                                qpel_mc_func (*qpix_op)[16],
 474                                int motion_x, int motion_y, int h)
 475 {
 476     uint8_t *ptr_y, *ptr_cb, *ptr_cr;
 477     int dxy, uvdxy, mx, my, src_x, src_y, uvsrc_x, uvsrc_y, v_edge_pos, linesize, uvlinesize;
 478
 479     dxy = ((motion_y & 3) << 2) | (motion_x & 3);
 480     src_x = s->mb_x *  16                 + (motion_x >> 2);
 481     src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2);
 482
 483     v_edge_pos = s->v_edge_pos >> field_based;
 484     linesize = s->linesize << field_based;
 485     uvlinesize = s->uvlinesize << field_based;
 486
 487     if(field_based){
 488         mx= motion_x/2;
 489         my= motion_y>>1;
 490     }else if(s->workaround_bugs&FF_BUG_QPEL_CHROMA2){
 491         static const int rtab[8]= {0,0,1,1,0,0,0,1};
 492         mx= (motion_x>>1) + rtab[motion_x&7];
 493         my= (motion_y>>1) + rtab[motion_y&7];
 494     }else if(s->workaround_bugs&FF_BUG_QPEL_CHROMA){
 495         mx= (motion_x>>1)|(motion_x&1);
 496         my= (motion_y>>1)|(motion_y&1);
 497     }else{
 498         mx= motion_x/2;
 499         my= motion_y/2;
 500     }
 501     mx= (mx>>1)|(mx&1);
 502     my= (my>>1)|(my&1);
 503
 504     uvdxy= (mx&1) | ((my&1)<<1);
 505     mx>>=1;
 506     my>>=1;
 507
 508     uvsrc_x = s->mb_x *  8                 + mx;
 509     uvsrc_y = s->mb_y * (8 >> field_based) + my;
 510
 511     ptr_y  = ref_picture[0] +   src_y *   linesize +   src_x;
 512     ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
 513     ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
 514
 515     if(   (unsigned)src_x > s->h_edge_pos - (motion_x&3) - 16
 516        || (unsigned)src_y >    v_edge_pos - (motion_y&3) - h  ){
 517         ff_emulated_edge_mc(s->edge_emu_buffer, ptr_y, s->linesize,
 518                             17, 17+field_based, src_x, src_y<<field_based,
 519                             s->h_edge_pos, s->v_edge_pos);
 520         ptr_y= s->edge_emu_buffer;
 521         if(!ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
 522             uint8_t *uvbuf= s->edge_emu_buffer + 18*s->linesize;
 523             ff_emulated_edge_mc(uvbuf, ptr_cb, s->uvlinesize,
 524                                 9, 9 + field_based,
 525                                 uvsrc_x, uvsrc_y<<field_based,
 526                                 s->h_edge_pos>>1, s->v_edge_pos>>1);
 527             ff_emulated_edge_mc(uvbuf + 16, ptr_cr, s->uvlinesize,
 528                                 9, 9 + field_based,
 529                                 uvsrc_x, uvsrc_y<<field_based,
 530                                 s->h_edge_pos>>1, s->v_edge_pos>>1);
 531             ptr_cb= uvbuf;
 532             ptr_cr= uvbuf + 16;
 533         }
 534     }
 535
 536     if(!field_based)
 537         qpix_op[0][dxy](dest_y, ptr_y, linesize);
 538     else{
 539         if(bottom_field){
 540             dest_y += s->linesize;
 541             dest_cb+= s->uvlinesize;
 542             dest_cr+= s->uvlinesize;
 543         }
 544
 545         if(field_select){
 546             ptr_y  += s->linesize;
 547             ptr_cb += s->uvlinesize;
 548             ptr_cr += s->uvlinesize;
 549         }
 550         //damn interlaced mode
 551         //FIXME boundary mirroring is not exactly correct here
 552         qpix_op[1][dxy](dest_y  , ptr_y  , linesize);
 553         qpix_op[1][dxy](dest_y+8, ptr_y+8, linesize);
 554     }
 555     if(!ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
 556         pix_op[1][uvdxy](dest_cr, ptr_cr, uvlinesize, h >> 1);
 557         pix_op[1][uvdxy](dest_cb, ptr_cb, uvlinesize, h >> 1);
 558     }
 559 }
 560
 561 /**
 562  * h263 chroma 4mv motion compensation.
 563  */
 564 static inline void chroma_4mv_motion(MpegEncContext *s,
 565                                      uint8_t *dest_cb, uint8_t *dest_cr,
 566                                      uint8_t **ref_picture,
 567                                      op_pixels_func *pix_op,
 568                                      int mx, int my){
 569     int dxy, emu=0, src_x, src_y, offset;
 570     uint8_t *ptr;
 571
 572     /* In case of 8X8, we construct a single chroma motion vector
 573        with a special rounding */
 574     mx= ff_h263_round_chroma(mx);
 575     my= ff_h263_round_chroma(my);
 576
 577     dxy = ((my & 1) << 1) | (mx & 1);
 578     mx >>= 1;
 579     my >>= 1;
 580
 581     src_x = s->mb_x * 8 + mx;
 582     src_y = s->mb_y * 8 + my;
 583     src_x = av_clip(src_x, -8, s->width/2);
 584     if (src_x == s->width/2)
 585         dxy &= ~1;
 586     src_y = av_clip(src_y, -8, s->height/2);
 587     if (src_y == s->height/2)
 588         dxy &= ~2;
 589
 590     offset = (src_y * (s->uvlinesize)) + src_x;
 591     ptr = ref_picture[1] + offset;
 592     if(s->flags&CODEC_FLAG_EMU_EDGE){
 593         if(   (unsigned)src_x > (s->h_edge_pos>>1) - (dxy &1) - 8
 594            || (unsigned)src_y > (s->v_edge_pos>>1) - (dxy>>1) - 8){
 595             ff_emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize,
 596                                 9, 9, src_x, src_y,
 597                                 s->h_edge_pos>>1, s->v_edge_pos>>1);
 598             ptr= s->edge_emu_buffer;
 599             emu=1;
 600         }
 601     }
 602     pix_op[dxy](dest_cb, ptr, s->uvlinesize, 8);
 603
 604     ptr = ref_picture[2] + offset;
 605     if(emu){
 606         ff_emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize,
 607                             9, 9, src_x, src_y,
 608                             s->h_edge_pos>>1, s->v_edge_pos>>1);
 609         ptr= s->edge_emu_buffer;
 610     }
 611     pix_op[dxy](dest_cr, ptr, s->uvlinesize, 8);
 612 }
 613
 614 static inline void prefetch_motion(MpegEncContext *s, uint8_t **pix, int dir){
 615     /* fetch pixels for estimated mv 4 macroblocks ahead
 616      * optimized for 64byte cache lines */
 617     const int shift = s->quarter_sample ? 2 : 1;
 618     const int mx= (s->mv[dir][0][0]>>shift) + 16*s->mb_x + 8;
 619     const int my= (s->mv[dir][0][1]>>shift) + 16*s->mb_y;
 620     int off= mx + (my + (s->mb_x&3)*4)*s->linesize + 64;
 621     s->dsp.prefetch(pix[0]+off, s->linesize, 4);
 622     off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
 623     s->dsp.prefetch(pix[1]+off, pix[2]-pix[1], 2);
 624 }
 625
 626 /**
 627  * motion compensation of a single macroblock
 628  * @param s context
 629  * @param dest_y luma destination pointer
 630  * @param dest_cb chroma cb/u destination pointer
 631  * @param dest_cr chroma cr/v destination pointer
 632  * @param dir direction (0->forward, 1->backward)
 633  * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
 634  * @param pic_op halfpel motion compensation function (average or put normally)
 635  * @param pic_op qpel motion compensation function (average or put normally)
 636  * the motion vectors are taken from s->mv and the MV type from s->mv_type
 637  */
 638 static av_always_inline void MPV_motion_internal(MpegEncContext *s,
 639                               uint8_t *dest_y, uint8_t *dest_cb,
 640                               uint8_t *dest_cr, int dir,
 641                               uint8_t **ref_picture,
 642                               op_pixels_func (*pix_op)[4],
 643                               qpel_mc_func (*qpix_op)[16], int is_mpeg12)
 644 {
 645     int dxy, mx, my, src_x, src_y, motion_x, motion_y;
 646     int mb_x, mb_y, i;
 647     uint8_t *ptr, *dest;
 648
 649     mb_x = s->mb_x;
 650     mb_y = s->mb_y;
 651
 652     prefetch_motion(s, ref_picture, dir);
 653
 654     if(!is_mpeg12 && s->obmc && s->pict_type != FF_B_TYPE){
 655         int16_t mv_cache[4][4][2];
 656         const int xy= s->mb_x + s->mb_y*s->mb_stride;
 657         const int mot_stride= s->b8_stride;
 658         const int mot_xy= mb_x*2 + mb_y*2*mot_stride;
 659
 660         assert(!s->mb_skipped);
 661
 662         memcpy(mv_cache[1][1], s->current_picture.motion_val[0][mot_xy           ], sizeof(int16_t)*4);
 663         memcpy(mv_cache[2][1], s->current_picture.motion_val[0][mot_xy+mot_stride], sizeof(int16_t)*4);
 664         memcpy(mv_cache[3][1], s->current_picture.motion_val[0][mot_xy+mot_stride], sizeof(int16_t)*4);
 665
 666         if(mb_y==0 || IS_INTRA(s->current_picture.mb_type[xy-s->mb_stride])){
 667             memcpy(mv_cache[0][1], mv_cache[1][1], sizeof(int16_t)*4);
 668         }else{
 669             memcpy(mv_cache[0][1], s->current_picture.motion_val[0][mot_xy-mot_stride], sizeof(int16_t)*4);
 670         }
 671
 672         if(mb_x==0 || IS_INTRA(s->current_picture.mb_type[xy-1])){
 673             *(int32_t*)mv_cache[1][0]= *(int32_t*)mv_cache[1][1];
 674             *(int32_t*)mv_cache[2][0]= *(int32_t*)mv_cache[2][1];
 675         }else{
 676             *(int32_t*)mv_cache[1][0]= *(int32_t*)s->current_picture.motion_val[0][mot_xy-1];
 677             *(int32_t*)mv_cache[2][0]= *(int32_t*)s->current_picture.motion_val[0][mot_xy-1+mot_stride];
 678         }
 679
 680         if(mb_x+1>=s->mb_width || IS_INTRA(s->current_picture.mb_type[xy+1])){
 681             *(int32_t*)mv_cache[1][3]= *(int32_t*)mv_cache[1][2];
 682             *(int32_t*)mv_cache[2][3]= *(int32_t*)mv_cache[2][2];
 683         }else{
 684             *(int32_t*)mv_cache[1][3]= *(int32_t*)s->current_picture.motion_val[0][mot_xy+2];
 685             *(int32_t*)mv_cache[2][3]= *(int32_t*)s->current_picture.motion_val[0][mot_xy+2+mot_stride];
 686         }
 687
 688         mx = 0;
 689         my = 0;
 690         for(i=0;i<4;i++) {
 691             const int x= (i&1)+1;
 692             const int y= (i>>1)+1;
 693             int16_t mv[5][2]= {
 694                 {mv_cache[y][x  ][0], mv_cache[y][x  ][1]},
 695                 {mv_cache[y-1][x][0], mv_cache[y-1][x][1]},
 696                 {mv_cache[y][x-1][0], mv_cache[y][x-1][1]},
 697                 {mv_cache[y][x+1][0], mv_cache[y][x+1][1]},
 698                 {mv_cache[y+1][x][0], mv_cache[y+1][x][1]}};
 699             //FIXME cleanup
 700             obmc_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
 701                         ref_picture[0],
 702                         mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >>1) * 8,
 703                         pix_op[1],
 704                         mv);
 705
 706             mx += mv[0][0];
 707             my += mv[0][1];
 708         }
 709         if(!ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY))
 710             chroma_4mv_motion(s, dest_cb, dest_cr, ref_picture, pix_op[1], mx, my);
 711
 712         return;
 713     }
 714
 715     switch(s->mv_type) {
 716     case MV_TYPE_16X16:
 717         if(s->mcsel){
 718             if(s->real_sprite_warping_points==1){
 719                 gmc1_motion(s, dest_y, dest_cb, dest_cr,
 720                             ref_picture);
 721             }else{
 722                 gmc_motion(s, dest_y, dest_cb, dest_cr,
 723                             ref_picture);
 724             }
 725         }else if(!is_mpeg12 && s->quarter_sample){
 726             qpel_motion(s, dest_y, dest_cb, dest_cr,
 727                         0, 0, 0,
 728                         ref_picture, pix_op, qpix_op,
 729                         s->mv[dir][0][0], s->mv[dir][0][1], 16);
 730         }else if(!is_mpeg12 && ENABLE_WMV2 && s->mspel){
 731             ff_mspel_motion(s, dest_y, dest_cb, dest_cr,
 732                         ref_picture, pix_op,
 733                         s->mv[dir][0][0], s->mv[dir][0][1], 16);
 734         }else
 735         {
 736             mpeg_motion(s, dest_y, dest_cb, dest_cr,
 737                         0, 0, 0,
 738                         ref_picture, pix_op,
 739                         s->mv[dir][0][0], s->mv[dir][0][1], 16);
 740         }
 741         break;
 742     case MV_TYPE_8X8:
 743     if (!is_mpeg12) {
 744         mx = 0;
 745         my = 0;
 746         if(s->quarter_sample){
 747             for(i=0;i<4;i++) {
 748                 motion_x = s->mv[dir][i][0];
 749                 motion_y = s->mv[dir][i][1];
 750
 751                 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
 752                 src_x = mb_x * 16 + (motion_x >> 2) + (i & 1) * 8;
 753                 src_y = mb_y * 16 + (motion_y >> 2) + (i >>1) * 8;
 754
 755                 /* WARNING: do no forget half pels */
 756                 src_x = av_clip(src_x, -16, s->width);
 757                 if (src_x == s->width)
 758                     dxy &= ~3;
 759                 src_y = av_clip(src_y, -16, s->height);
 760                 if (src_y == s->height)
 761                     dxy &= ~12;
 762
 763                 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x);
 764                 if(s->flags&CODEC_FLAG_EMU_EDGE){
 765                     if(   (unsigned)src_x > s->h_edge_pos - (motion_x&3) - 8
 766                        || (unsigned)src_y > s->v_edge_pos - (motion_y&3) - 8 ){
 767                         ff_emulated_edge_mc(s->edge_emu_buffer, ptr,
 768                                             s->linesize, 9, 9,
 769                                             src_x, src_y,
 770                                             s->h_edge_pos, s->v_edge_pos);
 771                         ptr= s->edge_emu_buffer;
 772                     }
 773                 }
 774                 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize;
 775                 qpix_op[1][dxy](dest, ptr, s->linesize);
 776
 777                 mx += s->mv[dir][i][0]/2;
 778                 my += s->mv[dir][i][1]/2;
 779             }
 780         }else{
 781             for(i=0;i<4;i++) {
 782                 hpel_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
 783                             ref_picture[0], 0, 0,
 784                             mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >>1) * 8,
 785                             s->width, s->height, s->linesize,
 786                             s->h_edge_pos, s->v_edge_pos,
 787                             8, 8, pix_op[1],
 788                             s->mv[dir][i][0], s->mv[dir][i][1]);
 789
 790                 mx += s->mv[dir][i][0];
 791                 my += s->mv[dir][i][1];
 792             }
 793         }
 794
 795         if(!ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY))
 796             chroma_4mv_motion(s, dest_cb, dest_cr, ref_picture, pix_op[1], mx, my);
 797     }
 798         break;
 799     case MV_TYPE_FIELD:
 800         if (s->picture_structure == PICT_FRAME) {
 801             if(!is_mpeg12 && s->quarter_sample){
 802                 for(i=0; i<2; i++){
 803                     qpel_motion(s, dest_y, dest_cb, dest_cr,
 804                                 1, i, s->field_select[dir][i],
 805                                 ref_picture, pix_op, qpix_op,
 806                                 s->mv[dir][i][0], s->mv[dir][i][1], 8);
 807                 }
 808             }else{
 809                 /* top field */
 810                 mpeg_motion(s, dest_y, dest_cb, dest_cr,
 811                             1, 0, s->field_select[dir][0],
 812                             ref_picture, pix_op,
 813                             s->mv[dir][0][0], s->mv[dir][0][1], 8);
 814                 /* bottom field */
 815                 mpeg_motion(s, dest_y, dest_cb, dest_cr,
 816                             1, 1, s->field_select[dir][1],
 817                             ref_picture, pix_op,
 818                             s->mv[dir][1][0], s->mv[dir][1][1], 8);
 819             }
 820         } else {
 821             if(s->picture_structure != s->field_select[dir][0] + 1 && s->pict_type != FF_B_TYPE && !s->first_field){
 822                 ref_picture= s->current_picture_ptr->data;
 823             }
 824
 825             mpeg_motion(s, dest_y, dest_cb, dest_cr,
 826                         0, 0, s->field_select[dir][0],
 827                         ref_picture, pix_op,
 828                         s->mv[dir][0][0], s->mv[dir][0][1], 16);
 829         }
 830         break;
 831     case MV_TYPE_16X8:
 832         for(i=0; i<2; i++){
 833             uint8_t ** ref2picture;
 834
 835             if(s->picture_structure == s->field_select[dir][i] + 1
 836                || s->pict_type == FF_B_TYPE || s->first_field){
 837                 ref2picture= ref_picture;
 838             }else{
 839                 ref2picture= s->current_picture_ptr->data;
 840             }
 841
 842             mpeg_motion(s, dest_y, dest_cb, dest_cr,
 843                         0, 0, s->field_select[dir][i],
 844                         ref2picture, pix_op,
 845                         s->mv[dir][i][0], s->mv[dir][i][1] + 16*i, 8);
 846
 847             dest_y += 16*s->linesize;
 848             dest_cb+= (16>>s->chroma_y_shift)*s->uvlinesize;
 849             dest_cr+= (16>>s->chroma_y_shift)*s->uvlinesize;
 850         }
 851         break;
 852     case MV_TYPE_DMV:
 853         if(s->picture_structure == PICT_FRAME){
 854             for(i=0; i<2; i++){
 855                 int j;
 856                 for(j=0; j<2; j++){
 857                     mpeg_motion(s, dest_y, dest_cb, dest_cr,
 858                                 1, j, j^i,
 859                                 ref_picture, pix_op,
 860                                 s->mv[dir][2*i + j][0], s->mv[dir][2*i + j][1], 8);
 861                 }
 862                 pix_op = s->dsp.avg_pixels_tab;
 863             }
 864         }else{
 865             for(i=0; i<2; i++){
 866                 mpeg_motion(s, dest_y, dest_cb, dest_cr,
 867                             0, 0, s->picture_structure != i+1,
 868                             ref_picture, pix_op,
 869                             s->mv[dir][2*i][0],s->mv[dir][2*i][1],16);
 870
 871                 // after put we make avg of the same block
 872                 pix_op=s->dsp.avg_pixels_tab;
 873
 874                 //opposite parity is always in the same frame if this is second field
 875                 if(!s->first_field){
 876                     ref_picture = s->current_picture_ptr->data;
 877                 }
 878             }
 879         }
 880     break;
 881     default: assert(0);
 882     }
 883 }
 884
 885 static inline void MPV_motion(MpegEncContext *s,
 886                               uint8_t *dest_y, uint8_t *dest_cb,
 887                               uint8_t *dest_cr, int dir,
 888                               uint8_t **ref_picture,
 889                               op_pixels_func (*pix_op)[4],
 890                               qpel_mc_func (*qpix_op)[16])
 891 {
 892 #ifndef CONFIG_SMALL
 893     if(s->out_format == FMT_MPEG1)
 894         MPV_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
 895                             ref_picture, pix_op, qpix_op, 1);
 896     else
 897 #endif
 898         MPV_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
 899                             ref_picture, pix_op, qpix_op, 0);
 900 }
 901 #endif /* FFMPEG_MPEGVIDEO_COMMON_H */