FFmpeg: libavcodec/mpegvideo_motion.c Source File

00001 /*
00002  * Copyright (c) 2000,2001 Fabrice Bellard
00003  * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
00004  *
00005  * 4MV & hq & B-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
00006  *
00007  * This file is part of FFmpeg.
00008  *
00009  * FFmpeg is free software; you can redistribute it and/or
00010  * modify it under the terms of the GNU Lesser General Public
00011  * License as published by the Free Software Foundation; either
00012  * version 2.1 of the License, or (at your option) any later version.
00013  *
00014  * FFmpeg is distributed in the hope that it will be useful,
00015  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00016  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00017  * Lesser General Public License for more details.
00018  *
00019  * You should have received a copy of the GNU Lesser General Public
00020  * License along with FFmpeg; if not, write to the Free Software
00021  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
00022  */
00023 
00024 #include <string.h>
00025 #include "libavutil/avassert.h"
00026 #include "avcodec.h"
00027 #include "dsputil.h"
00028 #include "mpegvideo.h"
00029 #include "mjpegenc.h"
00030 #include "msmpeg4.h"
00031 #include <limits.h>
00032 
00033 static void gmc1_motion(MpegEncContext *s,
00034                         uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
00035                         uint8_t **ref_picture)
00036 {
00037     uint8_t *ptr;
00038     int offset, src_x, src_y, linesize, uvlinesize;
00039     int motion_x, motion_y;
00040     int emu=0;
00041 
00042     motion_x= s->sprite_offset[0][0];
00043     motion_y= s->sprite_offset[0][1];
00044     src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy+1));
00045     src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy+1));
00046     motion_x<<=(3-s->sprite_warping_accuracy);
00047     motion_y<<=(3-s->sprite_warping_accuracy);
00048     src_x = av_clip(src_x, -16, s->width);
00049     if (src_x == s->width)
00050         motion_x =0;
00051     src_y = av_clip(src_y, -16, s->height);
00052     if (src_y == s->height)
00053         motion_y =0;
00054 
00055     linesize = s->linesize;
00056     uvlinesize = s->uvlinesize;
00057 
00058     ptr = ref_picture[0] + (src_y * linesize) + src_x;
00059 
00060     if(s->flags&CODEC_FLAG_EMU_EDGE){
00061         if(   (unsigned)src_x >= FFMAX(s->h_edge_pos - 17, 0)
00062            || (unsigned)src_y >= FFMAX(s->v_edge_pos - 17, 0)){
00063             s->dsp.emulated_edge_mc(s->edge_emu_buffer, ptr, linesize, 17, 17, src_x, src_y, s->h_edge_pos, s->v_edge_pos);
00064             ptr= s->edge_emu_buffer;
00065         }
00066     }
00067 
00068     if((motion_x|motion_y)&7){
00069         s->dsp.gmc1(dest_y  , ptr  , linesize, 16, motion_x&15, motion_y&15, 128 - s->no_rounding);
00070         s->dsp.gmc1(dest_y+8, ptr+8, linesize, 16, motion_x&15, motion_y&15, 128 - s->no_rounding);
00071     }else{
00072         int dxy;
00073 
00074         dxy= ((motion_x>>3)&1) | ((motion_y>>2)&2);
00075         if (s->no_rounding){
00076             s->dsp.put_no_rnd_pixels_tab[0][dxy](dest_y, ptr, linesize, 16);
00077         }else{
00078             s->dsp.put_pixels_tab       [0][dxy](dest_y, ptr, linesize, 16);
00079         }
00080     }
00081 
00082     if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
00083 
00084     motion_x= s->sprite_offset[1][0];
00085     motion_y= s->sprite_offset[1][1];
00086     src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy+1));
00087     src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy+1));
00088     motion_x<<=(3-s->sprite_warping_accuracy);
00089     motion_y<<=(3-s->sprite_warping_accuracy);
00090     src_x = av_clip(src_x, -8, s->width>>1);
00091     if (src_x == s->width>>1)
00092         motion_x =0;
00093     src_y = av_clip(src_y, -8, s->height>>1);
00094     if (src_y == s->height>>1)
00095         motion_y =0;
00096 
00097     offset = (src_y * uvlinesize) + src_x;
00098     ptr = ref_picture[1] + offset;
00099     if(s->flags&CODEC_FLAG_EMU_EDGE){
00100         if(   (unsigned)src_x >= FFMAX((s->h_edge_pos>>1) - 9, 0)
00101            || (unsigned)src_y >= FFMAX((s->v_edge_pos>>1) - 9, 0)){
00102             s->dsp.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);
00103             ptr= s->edge_emu_buffer;
00104             emu=1;
00105         }
00106     }
00107     s->dsp.gmc1(dest_cb, ptr, uvlinesize, 8, motion_x&15, motion_y&15, 128 - s->no_rounding);
00108 
00109     ptr = ref_picture[2] + offset;
00110     if(emu){
00111         s->dsp.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);
00112         ptr= s->edge_emu_buffer;
00113     }
00114     s->dsp.gmc1(dest_cr, ptr, uvlinesize, 8, motion_x&15, motion_y&15, 128 - s->no_rounding);
00115 
00116     return;
00117 }
00118 
00119 static void gmc_motion(MpegEncContext *s,
00120                        uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
00121                        uint8_t **ref_picture)
00122 {
00123     uint8_t *ptr;
00124     int linesize, uvlinesize;
00125     const int a= s->sprite_warping_accuracy;
00126     int ox, oy;
00127 
00128     linesize = s->linesize;
00129     uvlinesize = s->uvlinesize;
00130 
00131     ptr = ref_picture[0];
00132 
00133     ox= s->sprite_offset[0][0] + s->sprite_delta[0][0]*s->mb_x*16 + s->sprite_delta[0][1]*s->mb_y*16;
00134     oy= s->sprite_offset[0][1] + s->sprite_delta[1][0]*s->mb_x*16 + s->sprite_delta[1][1]*s->mb_y*16;
00135 
00136     s->dsp.gmc(dest_y, ptr, linesize, 16,
00137            ox,
00138            oy,
00139            s->sprite_delta[0][0], s->sprite_delta[0][1],
00140            s->sprite_delta[1][0], s->sprite_delta[1][1],
00141            a+1, (1<<(2*a+1)) - s->no_rounding,
00142            s->h_edge_pos, s->v_edge_pos);
00143     s->dsp.gmc(dest_y+8, ptr, linesize, 16,
00144            ox + s->sprite_delta[0][0]*8,
00145            oy + s->sprite_delta[1][0]*8,
00146            s->sprite_delta[0][0], s->sprite_delta[0][1],
00147            s->sprite_delta[1][0], s->sprite_delta[1][1],
00148            a+1, (1<<(2*a+1)) - s->no_rounding,
00149            s->h_edge_pos, s->v_edge_pos);
00150 
00151     if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
00152 
00153     ox= s->sprite_offset[1][0] + s->sprite_delta[0][0]*s->mb_x*8 + s->sprite_delta[0][1]*s->mb_y*8;
00154     oy= s->sprite_offset[1][1] + s->sprite_delta[1][0]*s->mb_x*8 + s->sprite_delta[1][1]*s->mb_y*8;
00155 
00156     ptr = ref_picture[1];
00157     s->dsp.gmc(dest_cb, ptr, uvlinesize, 8,
00158            ox,
00159            oy,
00160            s->sprite_delta[0][0], s->sprite_delta[0][1],
00161            s->sprite_delta[1][0], s->sprite_delta[1][1],
00162            a+1, (1<<(2*a+1)) - s->no_rounding,
00163            s->h_edge_pos>>1, s->v_edge_pos>>1);
00164 
00165     ptr = ref_picture[2];
00166     s->dsp.gmc(dest_cr, ptr, uvlinesize, 8,
00167            ox,
00168            oy,
00169            s->sprite_delta[0][0], s->sprite_delta[0][1],
00170            s->sprite_delta[1][0], s->sprite_delta[1][1],
00171            a+1, (1<<(2*a+1)) - s->no_rounding,
00172            s->h_edge_pos>>1, s->v_edge_pos>>1);
00173 }
00174 
00175 static inline int hpel_motion(MpegEncContext *s,
00176                                   uint8_t *dest, uint8_t *src,
00177                                   int src_x, int src_y,
00178                                   op_pixels_func *pix_op,
00179                                   int motion_x, int motion_y)
00180 {
00181     int dxy = 0;
00182     int emu=0;
00183 
00184     src_x += motion_x >> 1;
00185     src_y += motion_y >> 1;
00186 
00187     /* WARNING: do no forget half pels */
00188     src_x = av_clip(src_x, -16, s->width); //FIXME unneeded for emu?
00189     if (src_x != s->width)
00190         dxy |= motion_x & 1;
00191     src_y = av_clip(src_y, -16, s->height);
00192     if (src_y != s->height)
00193         dxy |= (motion_y & 1) << 1;
00194     src += src_y * s->linesize + src_x;
00195 
00196     if(s->unrestricted_mv && (s->flags&CODEC_FLAG_EMU_EDGE)){
00197         if(   (unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x&1) - 8, 0)
00198            || (unsigned)src_y > FFMAX(s->v_edge_pos - (motion_y&1) - 8, 0)){
00199             s->dsp.emulated_edge_mc(s->edge_emu_buffer, src, s->linesize, 9, 9,
00200                              src_x, src_y, s->h_edge_pos, s->v_edge_pos);
00201             src= s->edge_emu_buffer;
00202             emu=1;
00203         }
00204     }
00205     pix_op[dxy](dest, src, s->linesize, 8);
00206     return emu;
00207 }
00208 
00209 static av_always_inline
00210 void mpeg_motion_internal(MpegEncContext *s,
00211                  uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
00212                  int field_based, int bottom_field, int field_select,
00213                  uint8_t **ref_picture, op_pixels_func (*pix_op)[4],
00214                  int motion_x, int motion_y, int h, int is_mpeg12, int mb_y)
00215 {
00216     uint8_t *ptr_y, *ptr_cb, *ptr_cr;
00217     int dxy, uvdxy, mx, my, src_x, src_y,
00218         uvsrc_x, uvsrc_y, v_edge_pos, uvlinesize, linesize;
00219 
00220 #if 0
00221 if(s->quarter_sample)
00222 {
00223     motion_x>>=1;
00224     motion_y>>=1;
00225 }
00226 #endif
00227 
00228     v_edge_pos = s->v_edge_pos >> field_based;
00229     linesize   = s->current_picture.f.linesize[0] << field_based;
00230     uvlinesize = s->current_picture.f.linesize[1] << field_based;
00231 
00232     dxy = ((motion_y & 1) << 1) | (motion_x & 1);
00233     src_x = s->mb_x* 16               + (motion_x >> 1);
00234     src_y =(   mb_y<<(4-field_based)) + (motion_y >> 1);
00235 
00236     if (!is_mpeg12 && s->out_format == FMT_H263) {
00237         if((s->workaround_bugs & FF_BUG_HPEL_CHROMA) && field_based){
00238             mx = (motion_x>>1)|(motion_x&1);
00239             my = motion_y >>1;
00240             uvdxy = ((my & 1) << 1) | (mx & 1);
00241             uvsrc_x = s->mb_x* 8               + (mx >> 1);
00242             uvsrc_y =(   mb_y<<(3-field_based))+ (my >> 1);
00243         }else{
00244             uvdxy = dxy | (motion_y & 2) | ((motion_x & 2) >> 1);
00245             uvsrc_x = src_x>>1;
00246             uvsrc_y = src_y>>1;
00247         }
00248     }else if(!is_mpeg12 && s->out_format == FMT_H261){//even chroma mv's are full pel in H261
00249         mx = motion_x / 4;
00250         my = motion_y / 4;
00251         uvdxy = 0;
00252         uvsrc_x = s->mb_x*8 + mx;
00253         uvsrc_y =    mb_y*8 + my;
00254     } else {
00255         if(s->chroma_y_shift){
00256             mx = motion_x / 2;
00257             my = motion_y / 2;
00258             uvdxy = ((my & 1) << 1) | (mx & 1);
00259             uvsrc_x = s->mb_x* 8               + (mx >> 1);
00260             uvsrc_y =(   mb_y<<(3-field_based))+ (my >> 1);
00261         } else {
00262             if(s->chroma_x_shift){
00263             //Chroma422
00264                 mx = motion_x / 2;
00265                 uvdxy = ((motion_y & 1) << 1) | (mx & 1);
00266                 uvsrc_x = s->mb_x* 8           + (mx >> 1);
00267                 uvsrc_y = src_y;
00268             } else {
00269             //Chroma444
00270                 uvdxy = dxy;
00271                 uvsrc_x = src_x;
00272                 uvsrc_y = src_y;
00273             }
00274         }
00275     }
00276 
00277     ptr_y  = ref_picture[0] + src_y * linesize + src_x;
00278     ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
00279     ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
00280 
00281     if(   (unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x&1) - 16, 0)
00282        || (unsigned)src_y > FFMAX(   v_edge_pos - (motion_y&1) - h , 0)){
00283             if(is_mpeg12 || s->codec_id == AV_CODEC_ID_MPEG2VIDEO ||
00284                s->codec_id == AV_CODEC_ID_MPEG1VIDEO){
00285                 av_log(s->avctx,AV_LOG_DEBUG,
00286                         "MPEG motion vector out of boundary (%d %d)\n", src_x, src_y);
00287                 return;
00288             }
00289             s->dsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y, s->linesize,
00290                                 17, 17+field_based,
00291                                 src_x, src_y<<field_based,
00292                                 s->h_edge_pos, s->v_edge_pos);
00293             ptr_y = s->edge_emu_buffer;
00294             if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
00295                 uint8_t *uvbuf= s->edge_emu_buffer+18*s->linesize;
00296                 s->dsp.emulated_edge_mc(uvbuf ,
00297                                     ptr_cb, s->uvlinesize,
00298                                     9, 9+field_based,
00299                                     uvsrc_x, uvsrc_y<<field_based,
00300                                     s->h_edge_pos>>1, s->v_edge_pos>>1);
00301                 s->dsp.emulated_edge_mc(uvbuf+16,
00302                                     ptr_cr, s->uvlinesize,
00303                                     9, 9+field_based,
00304                                     uvsrc_x, uvsrc_y<<field_based,
00305                                     s->h_edge_pos>>1, s->v_edge_pos>>1);
00306                 ptr_cb= uvbuf;
00307                 ptr_cr= uvbuf+16;
00308             }
00309     }
00310 
00311     if(bottom_field){ //FIXME use this for field pix too instead of the obnoxious hack which changes picture.data
00312         dest_y += s->linesize;
00313         dest_cb+= s->uvlinesize;
00314         dest_cr+= s->uvlinesize;
00315     }
00316 
00317     if(field_select){
00318         ptr_y += s->linesize;
00319         ptr_cb+= s->uvlinesize;
00320         ptr_cr+= s->uvlinesize;
00321     }
00322 
00323     pix_op[0][dxy](dest_y, ptr_y, linesize, h);
00324 
00325     if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
00326         pix_op[s->chroma_x_shift][uvdxy]
00327                 (dest_cb, ptr_cb, uvlinesize, h >> s->chroma_y_shift);
00328         pix_op[s->chroma_x_shift][uvdxy]
00329                 (dest_cr, ptr_cr, uvlinesize, h >> s->chroma_y_shift);
00330     }
00331     if(!is_mpeg12 && (CONFIG_H261_ENCODER || CONFIG_H261_DECODER) &&
00332          s->out_format == FMT_H261){
00333         ff_h261_loop_filter(s);
00334     }
00335 }
00336 /* apply one mpeg motion vector to the three components */
00337 static void mpeg_motion(MpegEncContext *s,
00338                         uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
00339                         int field_select, uint8_t **ref_picture,
00340                         op_pixels_func (*pix_op)[4],
00341                         int motion_x, int motion_y, int h, int mb_y)
00342 {
00343 #if !CONFIG_SMALL
00344     if(s->out_format == FMT_MPEG1)
00345         mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 0, 0,
00346                     field_select, ref_picture, pix_op,
00347                     motion_x, motion_y, h, 1, mb_y);
00348     else
00349 #endif
00350         mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 0, 0,
00351                     field_select, ref_picture, pix_op,
00352                     motion_x, motion_y, h, 0, mb_y);
00353 }
00354 
00355 static void mpeg_motion_field(MpegEncContext *s, uint8_t *dest_y,
00356                               uint8_t *dest_cb, uint8_t *dest_cr,
00357                               int bottom_field, int field_select,
00358                               uint8_t **ref_picture,
00359                               op_pixels_func (*pix_op)[4],
00360                               int motion_x, int motion_y, int h, int mb_y)
00361 {
00362 #if !CONFIG_SMALL
00363     if(s->out_format == FMT_MPEG1)
00364         mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 1,
00365                     bottom_field, field_select, ref_picture, pix_op,
00366                     motion_x, motion_y, h, 1, mb_y);
00367     else
00368 #endif
00369         mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 1,
00370                     bottom_field, field_select, ref_picture, pix_op,
00371                     motion_x, motion_y, h, 0, mb_y);
00372 }
00373 
00374 //FIXME move to dsputil, avg variant, 16x16 version
00375 static inline void put_obmc(uint8_t *dst, uint8_t *src[5], int stride){
00376     int x;
00377     uint8_t * const top   = src[1];
00378     uint8_t * const left  = src[2];
00379     uint8_t * const mid   = src[0];
00380     uint8_t * const right = src[3];
00381     uint8_t * const bottom= src[4];
00382 #define OBMC_FILTER(x, t, l, m, r, b)\
00383     dst[x]= (t*top[x] + l*left[x] + m*mid[x] + r*right[x] + b*bottom[x] + 4)>>3
00384 #define OBMC_FILTER4(x, t, l, m, r, b)\
00385     OBMC_FILTER(x         , t, l, m, r, b);\
00386     OBMC_FILTER(x+1       , t, l, m, r, b);\
00387     OBMC_FILTER(x  +stride, t, l, m, r, b);\
00388     OBMC_FILTER(x+1+stride, t, l, m, r, b);
00389 
00390     x=0;
00391     OBMC_FILTER (x  , 2, 2, 4, 0, 0);
00392     OBMC_FILTER (x+1, 2, 1, 5, 0, 0);
00393     OBMC_FILTER4(x+2, 2, 1, 5, 0, 0);
00394     OBMC_FILTER4(x+4, 2, 0, 5, 1, 0);
00395     OBMC_FILTER (x+6, 2, 0, 5, 1, 0);
00396     OBMC_FILTER (x+7, 2, 0, 4, 2, 0);
00397     x+= stride;
00398     OBMC_FILTER (x  , 1, 2, 5, 0, 0);
00399     OBMC_FILTER (x+1, 1, 2, 5, 0, 0);
00400     OBMC_FILTER (x+6, 1, 0, 5, 2, 0);
00401     OBMC_FILTER (x+7, 1, 0, 5, 2, 0);
00402     x+= stride;
00403     OBMC_FILTER4(x  , 1, 2, 5, 0, 0);
00404     OBMC_FILTER4(x+2, 1, 1, 6, 0, 0);
00405     OBMC_FILTER4(x+4, 1, 0, 6, 1, 0);
00406     OBMC_FILTER4(x+6, 1, 0, 5, 2, 0);
00407     x+= 2*stride;
00408     OBMC_FILTER4(x  , 0, 2, 5, 0, 1);
00409     OBMC_FILTER4(x+2, 0, 1, 6, 0, 1);
00410     OBMC_FILTER4(x+4, 0, 0, 6, 1, 1);
00411     OBMC_FILTER4(x+6, 0, 0, 5, 2, 1);
00412     x+= 2*stride;
00413     OBMC_FILTER (x  , 0, 2, 5, 0, 1);
00414     OBMC_FILTER (x+1, 0, 2, 5, 0, 1);
00415     OBMC_FILTER4(x+2, 0, 1, 5, 0, 2);
00416     OBMC_FILTER4(x+4, 0, 0, 5, 1, 2);
00417     OBMC_FILTER (x+6, 0, 0, 5, 2, 1);
00418     OBMC_FILTER (x+7, 0, 0, 5, 2, 1);
00419     x+= stride;
00420     OBMC_FILTER (x  , 0, 2, 4, 0, 2);
00421     OBMC_FILTER (x+1, 0, 1, 5, 0, 2);
00422     OBMC_FILTER (x+6, 0, 0, 5, 1, 2);
00423     OBMC_FILTER (x+7, 0, 0, 4, 2, 2);
00424 }
00425 
00426 /* obmc for 1 8x8 luma block */
00427 static inline void obmc_motion(MpegEncContext *s,
00428                                uint8_t *dest, uint8_t *src,
00429                                int src_x, int src_y,
00430                                op_pixels_func *pix_op,
00431                                int16_t mv[5][2]/* mid top left right bottom*/)
00432 #define MID    0
00433 {
00434     int i;
00435     uint8_t *ptr[5];
00436 
00437     av_assert2(s->quarter_sample==0);
00438 
00439     for(i=0; i<5; i++){
00440         if(i && mv[i][0]==mv[MID][0] && mv[i][1]==mv[MID][1]){
00441             ptr[i]= ptr[MID];
00442         }else{
00443             ptr[i]= s->obmc_scratchpad + 8*(i&1) + s->linesize*8*(i>>1);
00444             hpel_motion(s, ptr[i], src,
00445                         src_x, src_y,
00446                         pix_op,
00447                         mv[i][0], mv[i][1]);
00448         }
00449     }
00450 
00451     put_obmc(dest, ptr, s->linesize);
00452 }
00453 
00454 static inline void qpel_motion(MpegEncContext *s,
00455                                uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
00456                                int field_based, int bottom_field, int field_select,
00457                                uint8_t **ref_picture, op_pixels_func (*pix_op)[4],
00458                                qpel_mc_func (*qpix_op)[16],
00459                                int motion_x, int motion_y, int h)
00460 {
00461     uint8_t *ptr_y, *ptr_cb, *ptr_cr;
00462     int dxy, uvdxy, mx, my, src_x, src_y, uvsrc_x, uvsrc_y, v_edge_pos, linesize, uvlinesize;
00463 
00464     dxy = ((motion_y & 3) << 2) | (motion_x & 3);
00465     src_x = s->mb_x *  16                 + (motion_x >> 2);
00466     src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2);
00467 
00468     v_edge_pos = s->v_edge_pos >> field_based;
00469     linesize = s->linesize << field_based;
00470     uvlinesize = s->uvlinesize << field_based;
00471 
00472     if(field_based){
00473         mx= motion_x/2;
00474         my= motion_y>>1;
00475     }else if(s->workaround_bugs&FF_BUG_QPEL_CHROMA2){
00476         static const int rtab[8]= {0,0,1,1,0,0,0,1};
00477         mx= (motion_x>>1) + rtab[motion_x&7];
00478         my= (motion_y>>1) + rtab[motion_y&7];
00479     }else if(s->workaround_bugs&FF_BUG_QPEL_CHROMA){
00480         mx= (motion_x>>1)|(motion_x&1);
00481         my= (motion_y>>1)|(motion_y&1);
00482     }else{
00483         mx= motion_x/2;
00484         my= motion_y/2;
00485     }
00486     mx= (mx>>1)|(mx&1);
00487     my= (my>>1)|(my&1);
00488 
00489     uvdxy= (mx&1) | ((my&1)<<1);
00490     mx>>=1;
00491     my>>=1;
00492 
00493     uvsrc_x = s->mb_x *  8                 + mx;
00494     uvsrc_y = s->mb_y * (8 >> field_based) + my;
00495 
00496     ptr_y  = ref_picture[0] +   src_y *   linesize +   src_x;
00497     ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
00498     ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
00499 
00500     if(   (unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x&3) - 16, 0)
00501        || (unsigned)src_y > FFMAX(   v_edge_pos - (motion_y&3) - h , 0)){
00502         s->dsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y, s->linesize,
00503                             17, 17+field_based, src_x, src_y<<field_based,
00504                             s->h_edge_pos, s->v_edge_pos);
00505         ptr_y= s->edge_emu_buffer;
00506         if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
00507             uint8_t *uvbuf= s->edge_emu_buffer + 18*s->linesize;
00508             s->dsp.emulated_edge_mc(uvbuf, ptr_cb, s->uvlinesize,
00509                                 9, 9 + field_based,
00510                                 uvsrc_x, uvsrc_y<<field_based,
00511                                 s->h_edge_pos>>1, s->v_edge_pos>>1);
00512             s->dsp.emulated_edge_mc(uvbuf + 16, ptr_cr, s->uvlinesize,
00513                                 9, 9 + field_based,
00514                                 uvsrc_x, uvsrc_y<<field_based,
00515                                 s->h_edge_pos>>1, s->v_edge_pos>>1);
00516             ptr_cb= uvbuf;
00517             ptr_cr= uvbuf + 16;
00518         }
00519     }
00520 
00521     if(!field_based)
00522         qpix_op[0][dxy](dest_y, ptr_y, linesize);
00523     else{
00524         if(bottom_field){
00525             dest_y += s->linesize;
00526             dest_cb+= s->uvlinesize;
00527             dest_cr+= s->uvlinesize;
00528         }
00529 
00530         if(field_select){
00531             ptr_y  += s->linesize;
00532             ptr_cb += s->uvlinesize;
00533             ptr_cr += s->uvlinesize;
00534         }
00535         //damn interlaced mode
00536         //FIXME boundary mirroring is not exactly correct here
00537         qpix_op[1][dxy](dest_y  , ptr_y  , linesize);
00538         qpix_op[1][dxy](dest_y+8, ptr_y+8, linesize);
00539     }
00540     if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
00541         pix_op[1][uvdxy](dest_cr, ptr_cr, uvlinesize, h >> 1);
00542         pix_op[1][uvdxy](dest_cb, ptr_cb, uvlinesize, h >> 1);
00543     }
00544 }
00545 
00549 static void chroma_4mv_motion(MpegEncContext *s,
00550                               uint8_t *dest_cb, uint8_t *dest_cr,
00551                               uint8_t **ref_picture,
00552                               op_pixels_func *pix_op,
00553                               int mx, int my)
00554 {
00555     int dxy, emu=0, src_x, src_y, offset;
00556     uint8_t *ptr;
00557 
00558     /* In case of 8X8, we construct a single chroma motion vector
00559        with a special rounding */
00560     mx= ff_h263_round_chroma(mx);
00561     my= ff_h263_round_chroma(my);
00562 
00563     dxy = ((my & 1) << 1) | (mx & 1);
00564     mx >>= 1;
00565     my >>= 1;
00566 
00567     src_x = s->mb_x * 8 + mx;
00568     src_y = s->mb_y * 8 + my;
00569     src_x = av_clip(src_x, -8, (s->width >> 1));
00570     if (src_x == (s->width >> 1))
00571         dxy &= ~1;
00572     src_y = av_clip(src_y, -8, (s->height >> 1));
00573     if (src_y == (s->height >> 1))
00574         dxy &= ~2;
00575 
00576     offset = src_y * s->uvlinesize + src_x;
00577     ptr = ref_picture[1] + offset;
00578     if(s->flags&CODEC_FLAG_EMU_EDGE){
00579         if(   (unsigned)src_x > FFMAX((s->h_edge_pos>>1) - (dxy &1) - 8, 0)
00580            || (unsigned)src_y > FFMAX((s->v_edge_pos>>1) - (dxy>>1) - 8, 0)){
00581             s->dsp.emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize,
00582                                 9, 9, src_x, src_y,
00583                                 s->h_edge_pos>>1, s->v_edge_pos>>1);
00584             ptr= s->edge_emu_buffer;
00585             emu=1;
00586         }
00587     }
00588     pix_op[dxy](dest_cb, ptr, s->uvlinesize, 8);
00589 
00590     ptr = ref_picture[2] + offset;
00591     if(emu){
00592         s->dsp.emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize,
00593                             9, 9, src_x, src_y,
00594                             s->h_edge_pos>>1, s->v_edge_pos>>1);
00595         ptr= s->edge_emu_buffer;
00596     }
00597     pix_op[dxy](dest_cr, ptr, s->uvlinesize, 8);
00598 }
00599 
00600 static inline void prefetch_motion(MpegEncContext *s, uint8_t **pix, int dir){
00601     /* fetch pixels for estimated mv 4 macroblocks ahead
00602      * optimized for 64byte cache lines */
00603     const int shift = s->quarter_sample ? 2 : 1;
00604     const int mx= (s->mv[dir][0][0]>>shift) + 16*s->mb_x + 8;
00605     const int my= (s->mv[dir][0][1]>>shift) + 16*s->mb_y;
00606     int off= mx + (my + (s->mb_x&3)*4)*s->linesize + 64;
00607     s->dsp.prefetch(pix[0]+off, s->linesize, 4);
00608     off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
00609     s->dsp.prefetch(pix[1]+off, pix[2]-pix[1], 2);
00610 }
00611 
00624 static av_always_inline void MPV_motion_internal(MpegEncContext *s,
00625                               uint8_t *dest_y, uint8_t *dest_cb,
00626                               uint8_t *dest_cr, int dir,
00627                               uint8_t **ref_picture,
00628                               op_pixels_func (*pix_op)[4],
00629                               qpel_mc_func (*qpix_op)[16], int is_mpeg12)
00630 {
00631     int dxy, mx, my, src_x, src_y, motion_x, motion_y;
00632     int mb_x, mb_y, i;
00633     uint8_t *ptr, *dest;
00634 
00635     mb_x = s->mb_x;
00636     mb_y = s->mb_y;
00637 
00638     prefetch_motion(s, ref_picture, dir);
00639 
00640     if(!is_mpeg12 && s->obmc && s->pict_type != AV_PICTURE_TYPE_B){
00641         LOCAL_ALIGNED_8(int16_t, mv_cache, [4], [4][2]);
00642         AVFrame *cur_frame = &s->current_picture.f;
00643         const int xy= s->mb_x + s->mb_y*s->mb_stride;
00644         const int mot_stride= s->b8_stride;
00645         const int mot_xy= mb_x*2 + mb_y*2*mot_stride;
00646 
00647         av_assert2(!s->mb_skipped);
00648 
00649         AV_COPY32(mv_cache[1][1], cur_frame->motion_val[0][mot_xy    ]);
00650         AV_COPY32(mv_cache[1][2], cur_frame->motion_val[0][mot_xy + 1]);
00651 
00652         AV_COPY32(mv_cache[2][1], cur_frame->motion_val[0][mot_xy + mot_stride    ]);
00653         AV_COPY32(mv_cache[2][2], cur_frame->motion_val[0][mot_xy + mot_stride + 1]);
00654 
00655         AV_COPY32(mv_cache[3][1], cur_frame->motion_val[0][mot_xy + mot_stride    ]);
00656         AV_COPY32(mv_cache[3][2], cur_frame->motion_val[0][mot_xy + mot_stride + 1]);
00657 
00658         if (mb_y == 0 || IS_INTRA(cur_frame->mb_type[xy - s->mb_stride])) {
00659             AV_COPY32(mv_cache[0][1], mv_cache[1][1]);
00660             AV_COPY32(mv_cache[0][2], mv_cache[1][2]);
00661         }else{
00662             AV_COPY32(mv_cache[0][1], cur_frame->motion_val[0][mot_xy - mot_stride    ]);
00663             AV_COPY32(mv_cache[0][2], cur_frame->motion_val[0][mot_xy - mot_stride + 1]);
00664         }
00665 
00666         if (mb_x == 0 || IS_INTRA(cur_frame->mb_type[xy - 1])) {
00667             AV_COPY32(mv_cache[1][0], mv_cache[1][1]);
00668             AV_COPY32(mv_cache[2][0], mv_cache[2][1]);
00669         }else{
00670             AV_COPY32(mv_cache[1][0], cur_frame->motion_val[0][mot_xy - 1]);
00671             AV_COPY32(mv_cache[2][0], cur_frame->motion_val[0][mot_xy - 1 + mot_stride]);
00672         }
00673 
00674         if (mb_x + 1 >= s->mb_width || IS_INTRA(cur_frame->mb_type[xy + 1])) {
00675             AV_COPY32(mv_cache[1][3], mv_cache[1][2]);
00676             AV_COPY32(mv_cache[2][3], mv_cache[2][2]);
00677         }else{
00678             AV_COPY32(mv_cache[1][3], cur_frame->motion_val[0][mot_xy + 2]);
00679             AV_COPY32(mv_cache[2][3], cur_frame->motion_val[0][mot_xy + 2 + mot_stride]);
00680         }
00681 
00682         mx = 0;
00683         my = 0;
00684         for(i=0;i<4;i++) {
00685             const int x= (i&1)+1;
00686             const int y= (i>>1)+1;
00687             int16_t mv[5][2]= {
00688                 {mv_cache[y][x  ][0], mv_cache[y][x  ][1]},
00689                 {mv_cache[y-1][x][0], mv_cache[y-1][x][1]},
00690                 {mv_cache[y][x-1][0], mv_cache[y][x-1][1]},
00691                 {mv_cache[y][x+1][0], mv_cache[y][x+1][1]},
00692                 {mv_cache[y+1][x][0], mv_cache[y+1][x][1]}};
00693             //FIXME cleanup
00694             obmc_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
00695                         ref_picture[0],
00696                         mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >>1) * 8,
00697                         pix_op[1],
00698                         mv);
00699 
00700             mx += mv[0][0];
00701             my += mv[0][1];
00702         }
00703         if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY))
00704             chroma_4mv_motion(s, dest_cb, dest_cr, ref_picture, pix_op[1], mx, my);
00705 
00706         return;
00707     }
00708 
00709     switch(s->mv_type) {
00710     case MV_TYPE_16X16:
00711         if(s->mcsel){
00712             if(s->real_sprite_warping_points==1){
00713                 gmc1_motion(s, dest_y, dest_cb, dest_cr,
00714                             ref_picture);
00715             }else{
00716                 gmc_motion(s, dest_y, dest_cb, dest_cr,
00717                             ref_picture);
00718             }
00719         }else if(!is_mpeg12 && s->quarter_sample){
00720             qpel_motion(s, dest_y, dest_cb, dest_cr,
00721                         0, 0, 0,
00722                         ref_picture, pix_op, qpix_op,
00723                         s->mv[dir][0][0], s->mv[dir][0][1], 16);
00724         } else if (!is_mpeg12 && (CONFIG_WMV2_DECODER || CONFIG_WMV2_ENCODER) &&
00725                    s->mspel && s->codec_id == AV_CODEC_ID_WMV2) {
00726             ff_mspel_motion(s, dest_y, dest_cb, dest_cr,
00727                         ref_picture, pix_op,
00728                         s->mv[dir][0][0], s->mv[dir][0][1], 16);
00729         }else
00730         {
00731             mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
00732                         ref_picture, pix_op,
00733                         s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y);
00734         }
00735         break;
00736     case MV_TYPE_8X8:
00737     if (!is_mpeg12) {
00738         mx = 0;
00739         my = 0;
00740         if(s->quarter_sample){
00741             for(i=0;i<4;i++) {
00742                 motion_x = s->mv[dir][i][0];
00743                 motion_y = s->mv[dir][i][1];
00744 
00745                 dxy = ((motion_y & 3) << 2) | (motion_x & 3);
00746                 src_x = mb_x * 16 + (motion_x >> 2) + (i & 1) * 8;
00747                 src_y = mb_y * 16 + (motion_y >> 2) + (i >>1) * 8;
00748 
00749                 /* WARNING: do no forget half pels */
00750                 src_x = av_clip(src_x, -16, s->width);
00751                 if (src_x == s->width)
00752                     dxy &= ~3;
00753                 src_y = av_clip(src_y, -16, s->height);
00754                 if (src_y == s->height)
00755                     dxy &= ~12;
00756 
00757                 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x);
00758                 if(s->flags&CODEC_FLAG_EMU_EDGE){
00759                     if(   (unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x&3) - 8, 0)
00760                        || (unsigned)src_y > FFMAX(s->v_edge_pos - (motion_y&3) - 8, 0)){
00761                         s->dsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
00762                                             s->linesize, 9, 9,
00763                                             src_x, src_y,
00764                                             s->h_edge_pos, s->v_edge_pos);
00765                         ptr= s->edge_emu_buffer;
00766                     }
00767                 }
00768                 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize;
00769                 qpix_op[1][dxy](dest, ptr, s->linesize);
00770 
00771                 mx += s->mv[dir][i][0]/2;
00772                 my += s->mv[dir][i][1]/2;
00773             }
00774         }else{
00775             for(i=0;i<4;i++) {
00776                 hpel_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
00777                             ref_picture[0],
00778                             mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >>1) * 8,
00779                             pix_op[1],
00780                             s->mv[dir][i][0], s->mv[dir][i][1]);
00781 
00782                 mx += s->mv[dir][i][0];
00783                 my += s->mv[dir][i][1];
00784             }
00785         }
00786 
00787         if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY))
00788             chroma_4mv_motion(s, dest_cb, dest_cr, ref_picture, pix_op[1], mx, my);
00789     }
00790         break;
00791     case MV_TYPE_FIELD:
00792         if (s->picture_structure == PICT_FRAME) {
00793             if(!is_mpeg12 && s->quarter_sample){
00794                 for(i=0; i<2; i++){
00795                     qpel_motion(s, dest_y, dest_cb, dest_cr,
00796                                 1, i, s->field_select[dir][i],
00797                                 ref_picture, pix_op, qpix_op,
00798                                 s->mv[dir][i][0], s->mv[dir][i][1], 8);
00799                 }
00800             }else{
00801                 /* top field */
00802                 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
00803                                   0, s->field_select[dir][0],
00804                                   ref_picture, pix_op,
00805                                   s->mv[dir][0][0], s->mv[dir][0][1], 8, mb_y);
00806                 /* bottom field */
00807                 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
00808                                   1, s->field_select[dir][1],
00809                                   ref_picture, pix_op,
00810                                   s->mv[dir][1][0], s->mv[dir][1][1], 8, mb_y);
00811             }
00812         } else {
00813             if(s->picture_structure != s->field_select[dir][0] + 1 && s->pict_type != AV_PICTURE_TYPE_B && !s->first_field){
00814                 ref_picture = s->current_picture_ptr->f.data;
00815             }
00816 
00817             mpeg_motion(s, dest_y, dest_cb, dest_cr,
00818                         s->field_select[dir][0],
00819                         ref_picture, pix_op,
00820                         s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y>>1);
00821         }
00822         break;
00823     case MV_TYPE_16X8:
00824         for(i=0; i<2; i++){
00825             uint8_t ** ref2picture;
00826 
00827             if(s->picture_structure == s->field_select[dir][i] + 1
00828                || s->pict_type == AV_PICTURE_TYPE_B || s->first_field){
00829                 ref2picture= ref_picture;
00830             }else{
00831                 ref2picture = s->current_picture_ptr->f.data;
00832             }
00833 
00834             mpeg_motion(s, dest_y, dest_cb, dest_cr,
00835                         s->field_select[dir][i],
00836                         ref2picture, pix_op,
00837                         s->mv[dir][i][0], s->mv[dir][i][1] + 16*i, 8, mb_y>>1);
00838 
00839             dest_y += 16*s->linesize;
00840             dest_cb+= (16>>s->chroma_y_shift)*s->uvlinesize;
00841             dest_cr+= (16>>s->chroma_y_shift)*s->uvlinesize;
00842         }
00843         break;
00844     case MV_TYPE_DMV:
00845         if(s->picture_structure == PICT_FRAME){
00846             for(i=0; i<2; i++){
00847                 int j;
00848                 for(j=0; j<2; j++){
00849                     mpeg_motion_field(s, dest_y, dest_cb, dest_cr,
00850                                       j, j^i, ref_picture, pix_op,
00851                                       s->mv[dir][2*i + j][0],
00852                                       s->mv[dir][2*i + j][1], 8, mb_y);
00853                 }
00854                 pix_op = s->dsp.avg_pixels_tab;
00855             }
00856         }else{
00857             for(i=0; i<2; i++){
00858                 mpeg_motion(s, dest_y, dest_cb, dest_cr,
00859                             s->picture_structure != i+1,
00860                             ref_picture, pix_op,
00861                             s->mv[dir][2*i][0],s->mv[dir][2*i][1],16, mb_y>>1);
00862 
00863                 // after put we make avg of the same block
00864                 pix_op=s->dsp.avg_pixels_tab;
00865 
00866                 //opposite parity is always in the same frame if this is second field
00867                 if(!s->first_field){
00868                     ref_picture = s->current_picture_ptr->f.data;
00869                 }
00870             }
00871         }
00872     break;
00873     default: av_assert2(0);
00874     }
00875 }
00876 
00877 void ff_MPV_motion(MpegEncContext *s,
00878                    uint8_t *dest_y, uint8_t *dest_cb,
00879                    uint8_t *dest_cr, int dir,
00880                    uint8_t **ref_picture,
00881                    op_pixels_func (*pix_op)[4],
00882                    qpel_mc_func (*qpix_op)[16])
00883 {
00884 #if !CONFIG_SMALL
00885     if(s->out_format == FMT_MPEG1)
00886         MPV_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
00887                             ref_picture, pix_op, qpix_op, 1);
00888     else
00889 #endif
00890         MPV_motion_internal(s, dest_y, dest_cb, dest_cr, dir,
00891                             ref_picture, pix_op, qpix_op, 0);
00892 }