falkTX
/
FFmpeg
mirror of https://github.com/falkTX/FFmpeg.git
1
0
Fork 0
Code Issues 0 Releases 338 Wiki Activity
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
51585 Commits
32 Branches
2.0GB
Tree: 39406ea321
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '39406ea321'
${ noResults }
FFmpeg/libavcodec/mpegvideo_motion.c

891 lines
34KB
Raw Blame History 

  1. /*

  2.  * Copyright (c) 2000,2001 Fabrice Bellard

  3.  * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>

  4.  *

  5.  * 4MV & hq & B-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>

  6.  *

  7.  * This file is part of FFmpeg.

  8.  *

  9.  * FFmpeg is free software; you can redistribute it and/or

  10.  * modify it under the terms of the GNU Lesser General Public

  11.  * License as published by the Free Software Foundation; either

  12.  * version 2.1 of the License, or (at your option) any later version.

  13.  *

  14.  * FFmpeg is distributed in the hope that it will be useful,

  15.  * but WITHOUT ANY WARRANTY; without even the implied warranty of

  16.  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU

  17.  * Lesser General Public License for more details.

  18.  *

  19.  * You should have received a copy of the GNU Lesser General Public

  20.  * License along with FFmpeg; if not, write to the Free Software

  21.  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA

  22.  */

  23. 

  24. #include <string.h>

  25. 

  26. #include "libavutil/avassert.h"

  27. #include "libavutil/internal.h"

  28. #include "avcodec.h"

  29. #include "dsputil.h"

  30. #include "mpegvideo.h"

  31. #include "mjpegenc.h"

  32. #include "msmpeg4.h"

  33. #include <limits.h>

  34. 

  35. static void gmc1_motion(MpegEncContext *s,

  36.                         uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,

  37.                         uint8_t **ref_picture)

  38. {

  39.     uint8_t *ptr;

  40.     int offset, src_x, src_y, linesize, uvlinesize;

  41.     int motion_x, motion_y;

  42.     int emu=0;

  43. 

  44.     motion_x= s->sprite_offset[0][0];

  45.     motion_y= s->sprite_offset[0][1];

  46.     src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy+1));

  47.     src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy+1));

  48.     motion_x<<=(3-s->sprite_warping_accuracy);

  49.     motion_y<<=(3-s->sprite_warping_accuracy);

  50.     src_x = av_clip(src_x, -16, s->width);

  51.     if (src_x == s->width)

  52.         motion_x =0;

  53.     src_y = av_clip(src_y, -16, s->height);

  54.     if (src_y == s->height)

  55.         motion_y =0;

  56. 

  57.     linesize = s->linesize;

  58.     uvlinesize = s->uvlinesize;

  59. 

  60.     ptr = ref_picture[0] + (src_y * linesize) + src_x;

  61. 

  62.         if(   (unsigned)src_x >= FFMAX(s->h_edge_pos - 17, 0)

  63.            || (unsigned)src_y >= FFMAX(s->v_edge_pos - 17, 0)){

  64.             s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr, linesize, 17, 17, src_x, src_y, s->h_edge_pos, s->v_edge_pos);

  65.             ptr= s->edge_emu_buffer;

  66.         }

  67. 

  68.     if((motion_x|motion_y)&7){

  69.         s->dsp.gmc1(dest_y  , ptr  , linesize, 16, motion_x&15, motion_y&15, 128 - s->no_rounding);

  70.         s->dsp.gmc1(dest_y+8, ptr+8, linesize, 16, motion_x&15, motion_y&15, 128 - s->no_rounding);

  71.     }else{

  72.         int dxy;

  73. 

  74.         dxy= ((motion_x>>3)&1) | ((motion_y>>2)&2);

  75.         if (s->no_rounding){

  76.             s->hdsp.put_no_rnd_pixels_tab[0][dxy](dest_y, ptr, linesize, 16);

  77.         }else{

  78.             s->hdsp.put_pixels_tab       [0][dxy](dest_y, ptr, linesize, 16);

  79.         }

  80.     }

  81. 

  82.     if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;

  83. 

  84.     motion_x= s->sprite_offset[1][0];

  85.     motion_y= s->sprite_offset[1][1];

  86.     src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy+1));

  87.     src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy+1));

  88.     motion_x<<=(3-s->sprite_warping_accuracy);

  89.     motion_y<<=(3-s->sprite_warping_accuracy);

  90.     src_x = av_clip(src_x, -8, s->width>>1);

  91.     if (src_x == s->width>>1)

  92.         motion_x =0;

  93.     src_y = av_clip(src_y, -8, s->height>>1);

  94.     if (src_y == s->height>>1)

  95.         motion_y =0;

  96. 

  97.     offset = (src_y * uvlinesize) + src_x;

  98.     ptr = ref_picture[1] + offset;

  99.         if(   (unsigned)src_x >= FFMAX((s->h_edge_pos>>1) - 9, 0)

  100.            || (unsigned)src_y >= FFMAX((s->v_edge_pos>>1) - 9, 0)){

  101.             s->vdsp.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);

  102.             ptr= s->edge_emu_buffer;

  103.             emu=1;

  104.         }

  105.     s->dsp.gmc1(dest_cb, ptr, uvlinesize, 8, motion_x&15, motion_y&15, 128 - s->no_rounding);

  106. 

  107.     ptr = ref_picture[2] + offset;

  108.     if(emu){

  109.         s->vdsp.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);

  110.         ptr= s->edge_emu_buffer;

  111.     }

  112.     s->dsp.gmc1(dest_cr, ptr, uvlinesize, 8, motion_x&15, motion_y&15, 128 - s->no_rounding);

  113. 

  114.     return;

  115. }

  116. 

  117. static void gmc_motion(MpegEncContext *s,

  118.                        uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,

  119.                        uint8_t **ref_picture)

  120. {

  121.     uint8_t *ptr;

  122.     int linesize, uvlinesize;

  123.     const int a= s->sprite_warping_accuracy;

  124.     int ox, oy;

  125. 

  126.     linesize = s->linesize;

  127.     uvlinesize = s->uvlinesize;

  128. 

  129.     ptr = ref_picture[0];

  130. 

  131.     ox= s->sprite_offset[0][0] + s->sprite_delta[0][0]*s->mb_x*16 + s->sprite_delta[0][1]*s->mb_y*16;

  132.     oy= s->sprite_offset[0][1] + s->sprite_delta[1][0]*s->mb_x*16 + s->sprite_delta[1][1]*s->mb_y*16;

  133. 

  134.     s->dsp.gmc(dest_y, ptr, linesize, 16,

  135.            ox,

  136.            oy,

  137.            s->sprite_delta[0][0], s->sprite_delta[0][1],

  138.            s->sprite_delta[1][0], s->sprite_delta[1][1],

  139.            a+1, (1<<(2*a+1)) - s->no_rounding,

  140.            s->h_edge_pos, s->v_edge_pos);

  141.     s->dsp.gmc(dest_y+8, ptr, linesize, 16,

  142.            ox + s->sprite_delta[0][0]*8,

  143.            oy + s->sprite_delta[1][0]*8,

  144.            s->sprite_delta[0][0], s->sprite_delta[0][1],

  145.            s->sprite_delta[1][0], s->sprite_delta[1][1],

  146.            a+1, (1<<(2*a+1)) - s->no_rounding,

  147.            s->h_edge_pos, s->v_edge_pos);

  148. 

  149.     if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;

  150. 

  151.     ox= s->sprite_offset[1][0] + s->sprite_delta[0][0]*s->mb_x*8 + s->sprite_delta[0][1]*s->mb_y*8;

  152.     oy= s->sprite_offset[1][1] + s->sprite_delta[1][0]*s->mb_x*8 + s->sprite_delta[1][1]*s->mb_y*8;

  153. 

  154.     ptr = ref_picture[1];

  155.     s->dsp.gmc(dest_cb, ptr, uvlinesize, 8,

  156.            ox,

  157.            oy,

  158.            s->sprite_delta[0][0], s->sprite_delta[0][1],

  159.            s->sprite_delta[1][0], s->sprite_delta[1][1],

  160.            a+1, (1<<(2*a+1)) - s->no_rounding,

  161.            s->h_edge_pos>>1, s->v_edge_pos>>1);

  162. 

  163.     ptr = ref_picture[2];

  164.     s->dsp.gmc(dest_cr, ptr, uvlinesize, 8,

  165.            ox,

  166.            oy,

  167.            s->sprite_delta[0][0], s->sprite_delta[0][1],

  168.            s->sprite_delta[1][0], s->sprite_delta[1][1],

  169.            a+1, (1<<(2*a+1)) - s->no_rounding,

  170.            s->h_edge_pos>>1, s->v_edge_pos>>1);

  171. }

  172. 

  173. static inline int hpel_motion(MpegEncContext *s,

  174.                                   uint8_t *dest, uint8_t *src,

  175.                                   int src_x, int src_y,

  176.                                   op_pixels_func *pix_op,

  177.                                   int motion_x, int motion_y)

  178. {

  179.     int dxy = 0;

  180.     int emu=0;

  181. 

  182.     src_x += motion_x >> 1;

  183.     src_y += motion_y >> 1;

  184. 

  185.     /* WARNING: do no forget half pels */

  186.     src_x = av_clip(src_x, -16, s->width); //FIXME unneeded for emu?

  187.     if (src_x != s->width)

  188.         dxy |= motion_x & 1;

  189.     src_y = av_clip(src_y, -16, s->height);

  190.     if (src_y != s->height)

  191.         dxy |= (motion_y & 1) << 1;

  192.     src += src_y * s->linesize + src_x;

  193. 

  194.     if(s->unrestricted_mv && (s->flags&CODEC_FLAG_EMU_EDGE)){

  195.         if(   (unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x&1) - 8, 0)

  196.            || (unsigned)src_y > FFMAX(s->v_edge_pos - (motion_y&1) - 8, 0)){

  197.             s->vdsp.emulated_edge_mc(s->edge_emu_buffer, src, s->linesize, 9, 9,

  198.                              src_x, src_y, s->h_edge_pos, s->v_edge_pos);

  199.             src= s->edge_emu_buffer;

  200.             emu=1;

  201.         }

  202.     }

  203.     pix_op[dxy](dest, src, s->linesize, 8);

  204.     return emu;

  205. }

  206. 

  207. static av_always_inline

  208. void mpeg_motion_internal(MpegEncContext *s,

  209.                  uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,

  210.                  int field_based, int bottom_field, int field_select,

  211.                  uint8_t **ref_picture, op_pixels_func (*pix_op)[4],

  212.                  int motion_x, int motion_y, int h, int is_mpeg12, int mb_y)

  213. {

  214.     uint8_t *ptr_y, *ptr_cb, *ptr_cr;

  215.     int dxy, uvdxy, mx, my, src_x, src_y,

  216.         uvsrc_x, uvsrc_y, v_edge_pos, uvlinesize, linesize;

  217. 

  218. #if 0

  219. if(s->quarter_sample)

  220. {

  221.     motion_x>>=1;

  222.     motion_y>>=1;

  223. }

  224. #endif

  225. 

  226.     v_edge_pos = s->v_edge_pos >> field_based;

  227.     linesize   = s->current_picture.f.linesize[0] << field_based;

  228.     uvlinesize = s->current_picture.f.linesize[1] << field_based;

  229. 

  230.     dxy = ((motion_y & 1) << 1) | (motion_x & 1);

  231.     src_x = s->mb_x* 16               + (motion_x >> 1);

  232.     src_y =(   mb_y<<(4-field_based)) + (motion_y >> 1);

  233. 

  234.     if (!is_mpeg12 && s->out_format == FMT_H263) {

  235.         if((s->workaround_bugs & FF_BUG_HPEL_CHROMA) && field_based){

  236.             mx = (motion_x>>1)|(motion_x&1);

  237.             my = motion_y >>1;

  238.             uvdxy = ((my & 1) << 1) | (mx & 1);

  239.             uvsrc_x = s->mb_x* 8               + (mx >> 1);

  240.             uvsrc_y =(   mb_y<<(3-field_based))+ (my >> 1);

  241.         }else{

  242.             uvdxy = dxy | (motion_y & 2) | ((motion_x & 2) >> 1);

  243.             uvsrc_x = src_x>>1;

  244.             uvsrc_y = src_y>>1;

  245.         }

  246.     }else if(!is_mpeg12 && s->out_format == FMT_H261){//even chroma mv's are full pel in H261

  247.         mx = motion_x / 4;

  248.         my = motion_y / 4;

  249.         uvdxy = 0;

  250.         uvsrc_x = s->mb_x*8 + mx;

  251.         uvsrc_y =    mb_y*8 + my;

  252.     } else {

  253.         if(s->chroma_y_shift){

  254.             mx = motion_x / 2;

  255.             my = motion_y / 2;

  256.             uvdxy = ((my & 1) << 1) | (mx & 1);

  257.             uvsrc_x = s->mb_x* 8               + (mx >> 1);

  258.             uvsrc_y =(   mb_y<<(3-field_based))+ (my >> 1);

  259.         } else {

  260.             if(s->chroma_x_shift){

  261.             //Chroma422

  262.                 mx = motion_x / 2;

  263.                 uvdxy = ((motion_y & 1) << 1) | (mx & 1);

  264.                 uvsrc_x = s->mb_x* 8           + (mx >> 1);

  265.                 uvsrc_y = src_y;

  266.             } else {

  267.             //Chroma444

  268.                 uvdxy = dxy;

  269.                 uvsrc_x = src_x;

  270.                 uvsrc_y = src_y;

  271.             }

  272.         }

  273.     }

  274. 

  275.     ptr_y  = ref_picture[0] + src_y * linesize + src_x;

  276.     ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;

  277.     ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;

  278. 

  279.     if(   (unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x&1) - 16, 0)

  280.        || (unsigned)src_y > FFMAX(   v_edge_pos - (motion_y&1) - h , 0)){

  281.             if(is_mpeg12 || s->codec_id == AV_CODEC_ID_MPEG2VIDEO ||

  282.                s->codec_id == AV_CODEC_ID_MPEG1VIDEO){

  283.                 av_log(s->avctx,AV_LOG_DEBUG,

  284.                         "MPEG motion vector out of boundary (%d %d)\n", src_x, src_y);

  285.                 return;

  286.             }

  287.             s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y, s->linesize,

  288.                                 17, 17+field_based,

  289.                                 src_x, src_y<<field_based,

  290.                                 s->h_edge_pos, s->v_edge_pos);

  291.             ptr_y = s->edge_emu_buffer;

  292.             if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){

  293.                 uint8_t *uvbuf= s->edge_emu_buffer+18*s->linesize;

  294.                 s->vdsp.emulated_edge_mc(uvbuf ,

  295.                                     ptr_cb, s->uvlinesize,

  296.                                     9, 9+field_based,

  297.                                     uvsrc_x, uvsrc_y<<field_based,

  298.                                     s->h_edge_pos>>1, s->v_edge_pos>>1);

  299.                 s->vdsp.emulated_edge_mc(uvbuf+16,

  300.                                     ptr_cr, s->uvlinesize,

  301.                                     9, 9+field_based,

  302.                                     uvsrc_x, uvsrc_y<<field_based,

  303.                                     s->h_edge_pos>>1, s->v_edge_pos>>1);

  304.                 ptr_cb= uvbuf;

  305.                 ptr_cr= uvbuf+16;

  306.             }

  307.     }

  308. 

  309.     if(bottom_field){ //FIXME use this for field pix too instead of the obnoxious hack which changes picture.data

  310.         dest_y += s->linesize;

  311.         dest_cb+= s->uvlinesize;

  312.         dest_cr+= s->uvlinesize;

  313.     }

  314. 

  315.     if(field_select){

  316.         ptr_y += s->linesize;

  317.         ptr_cb+= s->uvlinesize;

  318.         ptr_cr+= s->uvlinesize;

  319.     }

  320. 

  321.     pix_op[0][dxy](dest_y, ptr_y, linesize, h);

  322. 

  323.     if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){

  324.         pix_op[s->chroma_x_shift][uvdxy]

  325.                 (dest_cb, ptr_cb, uvlinesize, h >> s->chroma_y_shift);

  326.         pix_op[s->chroma_x_shift][uvdxy]

  327.                 (dest_cr, ptr_cr, uvlinesize, h >> s->chroma_y_shift);

  328.     }

  329.     if(!is_mpeg12 && (CONFIG_H261_ENCODER || CONFIG_H261_DECODER) &&

  330.          s->out_format == FMT_H261){

  331.         ff_h261_loop_filter(s);

  332.     }

  333. }

  334. /* apply one mpeg motion vector to the three components */

  335. static void mpeg_motion(MpegEncContext *s,

  336.                         uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,

  337.                         int field_select, uint8_t **ref_picture,

  338.                         op_pixels_func (*pix_op)[4],

  339.                         int motion_x, int motion_y, int h, int mb_y)

  340. {

  341. #if !CONFIG_SMALL

  342.     if(s->out_format == FMT_MPEG1)

  343.         mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 0, 0,

  344.                     field_select, ref_picture, pix_op,

  345.                     motion_x, motion_y, h, 1, mb_y);

  346.     else

  347. #endif

  348.         mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 0, 0,

  349.                     field_select, ref_picture, pix_op,

  350.                     motion_x, motion_y, h, 0, mb_y);

  351. }

  352. 

  353. static void mpeg_motion_field(MpegEncContext *s, uint8_t *dest_y,

  354.                               uint8_t *dest_cb, uint8_t *dest_cr,

  355.                               int bottom_field, int field_select,

  356.                               uint8_t **ref_picture,

  357.                               op_pixels_func (*pix_op)[4],

  358.                               int motion_x, int motion_y, int h, int mb_y)

  359. {

  360. #if !CONFIG_SMALL

  361.     if(s->out_format == FMT_MPEG1)

  362.         mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 1,

  363.                     bottom_field, field_select, ref_picture, pix_op,

  364.                     motion_x, motion_y, h, 1, mb_y);

  365.     else

  366. #endif

  367.         mpeg_motion_internal(s, dest_y, dest_cb, dest_cr, 1,

  368.                     bottom_field, field_select, ref_picture, pix_op,

  369.                     motion_x, motion_y, h, 0, mb_y);

  370. }

  371. 

  372. //FIXME move to dsputil, avg variant, 16x16 version

  373. static inline void put_obmc(uint8_t *dst, uint8_t *src[5], int stride){

  374.     int x;

  375.     uint8_t * const top   = src[1];

  376.     uint8_t * const left  = src[2];

  377.     uint8_t * const mid   = src[0];

  378.     uint8_t * const right = src[3];

  379.     uint8_t * const bottom= src[4];

  380. #define OBMC_FILTER(x, t, l, m, r, b)\

  381.     dst[x]= (t*top[x] + l*left[x] + m*mid[x] + r*right[x] + b*bottom[x] + 4)>>3

  382. #define OBMC_FILTER4(x, t, l, m, r, b)\

  383.     OBMC_FILTER(x         , t, l, m, r, b);\

  384.     OBMC_FILTER(x+1       , t, l, m, r, b);\

  385.     OBMC_FILTER(x  +stride, t, l, m, r, b);\

  386.     OBMC_FILTER(x+1+stride, t, l, m, r, b);

  387. 

  388.     x=0;

  389.     OBMC_FILTER (x  , 2, 2, 4, 0, 0);

  390.     OBMC_FILTER (x+1, 2, 1, 5, 0, 0);

  391.     OBMC_FILTER4(x+2, 2, 1, 5, 0, 0);

  392.     OBMC_FILTER4(x+4, 2, 0, 5, 1, 0);

  393.     OBMC_FILTER (x+6, 2, 0, 5, 1, 0);

  394.     OBMC_FILTER (x+7, 2, 0, 4, 2, 0);

  395.     x+= stride;

  396.     OBMC_FILTER (x  , 1, 2, 5, 0, 0);

  397.     OBMC_FILTER (x+1, 1, 2, 5, 0, 0);

  398.     OBMC_FILTER (x+6, 1, 0, 5, 2, 0);

  399.     OBMC_FILTER (x+7, 1, 0, 5, 2, 0);

  400.     x+= stride;

  401.     OBMC_FILTER4(x  , 1, 2, 5, 0, 0);

  402.     OBMC_FILTER4(x+2, 1, 1, 6, 0, 0);

  403.     OBMC_FILTER4(x+4, 1, 0, 6, 1, 0);

  404.     OBMC_FILTER4(x+6, 1, 0, 5, 2, 0);

  405.     x+= 2*stride;

  406.     OBMC_FILTER4(x  , 0, 2, 5, 0, 1);

  407.     OBMC_FILTER4(x+2, 0, 1, 6, 0, 1);

  408.     OBMC_FILTER4(x+4, 0, 0, 6, 1, 1);

  409.     OBMC_FILTER4(x+6, 0, 0, 5, 2, 1);

  410.     x+= 2*stride;

  411.     OBMC_FILTER (x  , 0, 2, 5, 0, 1);

  412.     OBMC_FILTER (x+1, 0, 2, 5, 0, 1);

  413.     OBMC_FILTER4(x+2, 0, 1, 5, 0, 2);

  414.     OBMC_FILTER4(x+4, 0, 0, 5, 1, 2);

  415.     OBMC_FILTER (x+6, 0, 0, 5, 2, 1);

  416.     OBMC_FILTER (x+7, 0, 0, 5, 2, 1);

  417.     x+= stride;

  418.     OBMC_FILTER (x  , 0, 2, 4, 0, 2);

  419.     OBMC_FILTER (x+1, 0, 1, 5, 0, 2);

  420.     OBMC_FILTER (x+6, 0, 0, 5, 1, 2);

  421.     OBMC_FILTER (x+7, 0, 0, 4, 2, 2);

  422. }

  423. 

  424. /* obmc for 1 8x8 luma block */

  425. static inline void obmc_motion(MpegEncContext *s,

  426.                                uint8_t *dest, uint8_t *src,

  427.                                int src_x, int src_y,

  428.                                op_pixels_func *pix_op,

  429.                                int16_t mv[5][2]/* mid top left right bottom*/)

  430. #define MID    0

  431. {

  432.     int i;

  433.     uint8_t *ptr[5];

  434. 

  435.     av_assert2(s->quarter_sample==0);

  436. 

  437.     for(i=0; i<5; i++){

  438.         if(i && mv[i][0]==mv[MID][0] && mv[i][1]==mv[MID][1]){

  439.             ptr[i]= ptr[MID];

  440.         }else{

  441.             ptr[i]= s->obmc_scratchpad + 8*(i&1) + s->linesize*8*(i>>1);

  442.             hpel_motion(s, ptr[i], src,

  443.                         src_x, src_y,

  444.                         pix_op,

  445.                         mv[i][0], mv[i][1]);

  446.         }

  447.     }

  448. 

  449.     put_obmc(dest, ptr, s->linesize);

  450. }

  451. 

  452. static inline void qpel_motion(MpegEncContext *s,

  453.                                uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,

  454.                                int field_based, int bottom_field, int field_select,

  455.                                uint8_t **ref_picture, op_pixels_func (*pix_op)[4],

  456.                                qpel_mc_func (*qpix_op)[16],

  457.                                int motion_x, int motion_y, int h)

  458. {

  459.     uint8_t *ptr_y, *ptr_cb, *ptr_cr;

  460.     int dxy, uvdxy, mx, my, src_x, src_y, uvsrc_x, uvsrc_y, v_edge_pos, linesize, uvlinesize;

  461. 

  462.     dxy = ((motion_y & 3) << 2) | (motion_x & 3);

  463.     src_x = s->mb_x *  16                 + (motion_x >> 2);

  464.     src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2);

  465. 

  466.     v_edge_pos = s->v_edge_pos >> field_based;

  467.     linesize = s->linesize << field_based;

  468.     uvlinesize = s->uvlinesize << field_based;

  469. 

  470.     if(field_based){

  471.         mx= motion_x/2;

  472.         my= motion_y>>1;

  473.     }else if(s->workaround_bugs&FF_BUG_QPEL_CHROMA2){

  474.         static const int rtab[8]= {0,0,1,1,0,0,0,1};

  475.         mx= (motion_x>>1) + rtab[motion_x&7];

  476.         my= (motion_y>>1) + rtab[motion_y&7];

  477.     }else if(s->workaround_bugs&FF_BUG_QPEL_CHROMA){

  478.         mx= (motion_x>>1)|(motion_x&1);

  479.         my= (motion_y>>1)|(motion_y&1);

  480.     }else{

  481.         mx= motion_x/2;

  482.         my= motion_y/2;

  483.     }

  484.     mx= (mx>>1)|(mx&1);

  485.     my= (my>>1)|(my&1);

  486. 

  487.     uvdxy= (mx&1) | ((my&1)<<1);

  488.     mx>>=1;

  489.     my>>=1;

  490. 

  491.     uvsrc_x = s->mb_x *  8                 + mx;

  492.     uvsrc_y = s->mb_y * (8 >> field_based) + my;

  493. 

  494.     ptr_y  = ref_picture[0] +   src_y *   linesize +   src_x;

  495.     ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;

  496.     ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;

  497. 

  498.     if(   (unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x&3) - 16, 0)

  499.        || (unsigned)src_y > FFMAX(   v_edge_pos - (motion_y&3) - h , 0)){

  500.         s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y, s->linesize,

  501.                             17, 17+field_based, src_x, src_y<<field_based,

  502.                             s->h_edge_pos, s->v_edge_pos);

  503.         ptr_y= s->edge_emu_buffer;

  504.         if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){

  505.             uint8_t *uvbuf= s->edge_emu_buffer + 18*s->linesize;

  506.             s->vdsp.emulated_edge_mc(uvbuf, ptr_cb, s->uvlinesize,

  507.                                 9, 9 + field_based,

  508.                                 uvsrc_x, uvsrc_y<<field_based,

  509.                                 s->h_edge_pos>>1, s->v_edge_pos>>1);

  510.             s->vdsp.emulated_edge_mc(uvbuf + 16, ptr_cr, s->uvlinesize,

  511.                                 9, 9 + field_based,

  512.                                 uvsrc_x, uvsrc_y<<field_based,

  513.                                 s->h_edge_pos>>1, s->v_edge_pos>>1);

  514.             ptr_cb= uvbuf;

  515.             ptr_cr= uvbuf + 16;

  516.         }

  517.     }

  518. 

  519.     if(!field_based)

  520.         qpix_op[0][dxy](dest_y, ptr_y, linesize);

  521.     else{

  522.         if(bottom_field){

  523.             dest_y += s->linesize;

  524.             dest_cb+= s->uvlinesize;

  525.             dest_cr+= s->uvlinesize;

  526.         }

  527. 

  528.         if(field_select){

  529.             ptr_y  += s->linesize;

  530.             ptr_cb += s->uvlinesize;

  531.             ptr_cr += s->uvlinesize;

  532.         }

  533.         //damn interlaced mode

  534.         //FIXME boundary mirroring is not exactly correct here

  535.         qpix_op[1][dxy](dest_y  , ptr_y  , linesize);

  536.         qpix_op[1][dxy](dest_y+8, ptr_y+8, linesize);

  537.     }

  538.     if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){

  539.         pix_op[1][uvdxy](dest_cr, ptr_cr, uvlinesize, h >> 1);

  540.         pix_op[1][uvdxy](dest_cb, ptr_cb, uvlinesize, h >> 1);

  541.     }

  542. }

  543. 

  544. /**

  545.  * h263 chroma 4mv motion compensation.

  546.  */

  547. static void chroma_4mv_motion(MpegEncContext *s,

  548.                               uint8_t *dest_cb, uint8_t *dest_cr,

  549.                               uint8_t **ref_picture,

  550.                               op_pixels_func *pix_op,

  551.                               int mx, int my)

  552. {

  553.     int dxy, emu=0, src_x, src_y, offset;

  554.     uint8_t *ptr;

  555. 

  556.     /* In case of 8X8, we construct a single chroma motion vector

  557.        with a special rounding */

  558.     mx= ff_h263_round_chroma(mx);

  559.     my= ff_h263_round_chroma(my);

  560. 

  561.     dxy = ((my & 1) << 1) | (mx & 1);

  562.     mx >>= 1;

  563.     my >>= 1;

  564. 

  565.     src_x = s->mb_x * 8 + mx;

  566.     src_y = s->mb_y * 8 + my;

  567.     src_x = av_clip(src_x, -8, (s->width >> 1));

  568.     if (src_x == (s->width >> 1))

  569.         dxy &= ~1;

  570.     src_y = av_clip(src_y, -8, (s->height >> 1));

  571.     if (src_y == (s->height >> 1))

  572.         dxy &= ~2;

  573. 

  574.     offset = src_y * s->uvlinesize + src_x;

  575.     ptr = ref_picture[1] + offset;

  576.     if(s->flags&CODEC_FLAG_EMU_EDGE){

  577.         if(   (unsigned)src_x > FFMAX((s->h_edge_pos>>1) - (dxy &1) - 8, 0)

  578.            || (unsigned)src_y > FFMAX((s->v_edge_pos>>1) - (dxy>>1) - 8, 0)){

  579.             s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize,

  580.                                 9, 9, src_x, src_y,

  581.                                 s->h_edge_pos>>1, s->v_edge_pos>>1);

  582.             ptr= s->edge_emu_buffer;

  583.             emu=1;

  584.         }

  585.     }

  586.     pix_op[dxy](dest_cb, ptr, s->uvlinesize, 8);

  587. 

  588.     ptr = ref_picture[2] + offset;

  589.     if(emu){

  590.         s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize,

  591.                             9, 9, src_x, src_y,

  592.                             s->h_edge_pos>>1, s->v_edge_pos>>1);

  593.         ptr= s->edge_emu_buffer;

  594.     }

  595.     pix_op[dxy](dest_cr, ptr, s->uvlinesize, 8);

  596. }

  597. 

  598. static inline void prefetch_motion(MpegEncContext *s, uint8_t **pix, int dir){

  599.     /* fetch pixels for estimated mv 4 macroblocks ahead

  600.      * optimized for 64byte cache lines */

  601.     const int shift = s->quarter_sample ? 2 : 1;

  602.     const int mx= (s->mv[dir][0][0]>>shift) + 16*s->mb_x + 8;

  603.     const int my= (s->mv[dir][0][1]>>shift) + 16*s->mb_y;

  604.     int off= mx + (my + (s->mb_x&3)*4)*s->linesize + 64;

  605.     s->vdsp.prefetch(pix[0]+off, s->linesize, 4);

  606.     off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;

  607.     s->vdsp.prefetch(pix[1]+off, pix[2]-pix[1], 2);

  608. }

  609. 

  610. /**

  611.  * motion compensation of a single macroblock

  612.  * @param s context

  613.  * @param dest_y luma destination pointer

  614.  * @param dest_cb chroma cb/u destination pointer

  615.  * @param dest_cr chroma cr/v destination pointer

  616.  * @param dir direction (0->forward, 1->backward)

  617.  * @param ref_picture array[3] of pointers to the 3 planes of the reference picture

  618.  * @param pix_op halfpel motion compensation function (average or put normally)

  619.  * @param qpix_op qpel motion compensation function (average or put normally)

  620.  * the motion vectors are taken from s->mv and the MV type from s->mv_type

  621.  */

  622. static av_always_inline void MPV_motion_internal(MpegEncContext *s,

  623.                               uint8_t *dest_y, uint8_t *dest_cb,

  624.                               uint8_t *dest_cr, int dir,

  625.                               uint8_t **ref_picture,

  626.                               op_pixels_func (*pix_op)[4],

  627.                               qpel_mc_func (*qpix_op)[16], int is_mpeg12)

  628. {

  629.     int dxy, mx, my, src_x, src_y, motion_x, motion_y;

  630.     int mb_x, mb_y, i;

  631.     uint8_t *ptr, *dest;

  632. 

  633.     mb_x = s->mb_x;

  634.     mb_y = s->mb_y;

  635. 

  636.     prefetch_motion(s, ref_picture, dir);

  637. 

  638.     if(!is_mpeg12 && s->obmc && s->pict_type != AV_PICTURE_TYPE_B){

  639.         LOCAL_ALIGNED_8(int16_t, mv_cache, [4], [4][2]);

  640.         Picture *cur_frame = &s->current_picture;

  641.         const int xy= s->mb_x + s->mb_y*s->mb_stride;

  642.         const int mot_stride= s->b8_stride;

  643.         const int mot_xy= mb_x*2 + mb_y*2*mot_stride;

  644. 

  645.         av_assert2(!s->mb_skipped);

  646. 

  647.         AV_COPY32(mv_cache[1][1], cur_frame->motion_val[0][mot_xy    ]);

  648.         AV_COPY32(mv_cache[1][2], cur_frame->motion_val[0][mot_xy + 1]);

  649. 

  650.         AV_COPY32(mv_cache[2][1], cur_frame->motion_val[0][mot_xy + mot_stride    ]);

  651.         AV_COPY32(mv_cache[2][2], cur_frame->motion_val[0][mot_xy + mot_stride + 1]);

  652. 

  653.         AV_COPY32(mv_cache[3][1], cur_frame->motion_val[0][mot_xy + mot_stride    ]);

  654.         AV_COPY32(mv_cache[3][2], cur_frame->motion_val[0][mot_xy + mot_stride + 1]);

  655. 

  656.         if (mb_y == 0 || IS_INTRA(cur_frame->mb_type[xy - s->mb_stride])) {

  657.             AV_COPY32(mv_cache[0][1], mv_cache[1][1]);

  658.             AV_COPY32(mv_cache[0][2], mv_cache[1][2]);

  659.         }else{

  660.             AV_COPY32(mv_cache[0][1], cur_frame->motion_val[0][mot_xy - mot_stride    ]);

  661.             AV_COPY32(mv_cache[0][2], cur_frame->motion_val[0][mot_xy - mot_stride + 1]);

  662.         }

  663. 

  664.         if (mb_x == 0 || IS_INTRA(cur_frame->mb_type[xy - 1])) {

  665.             AV_COPY32(mv_cache[1][0], mv_cache[1][1]);

  666.             AV_COPY32(mv_cache[2][0], mv_cache[2][1]);

  667.         }else{

  668.             AV_COPY32(mv_cache[1][0], cur_frame->motion_val[0][mot_xy - 1]);

  669.             AV_COPY32(mv_cache[2][0], cur_frame->motion_val[0][mot_xy - 1 + mot_stride]);

  670.         }

  671. 

  672.         if (mb_x + 1 >= s->mb_width || IS_INTRA(cur_frame->mb_type[xy + 1])) {

  673.             AV_COPY32(mv_cache[1][3], mv_cache[1][2]);

  674.             AV_COPY32(mv_cache[2][3], mv_cache[2][2]);

  675.         }else{

  676.             AV_COPY32(mv_cache[1][3], cur_frame->motion_val[0][mot_xy + 2]);

  677.             AV_COPY32(mv_cache[2][3], cur_frame->motion_val[0][mot_xy + 2 + mot_stride]);

  678.         }

  679. 

  680.         mx = 0;

  681.         my = 0;

  682.         for(i=0;i<4;i++) {

  683.             const int x= (i&1)+1;

  684.             const int y= (i>>1)+1;

  685.             int16_t mv[5][2]= {

  686.                 {mv_cache[y][x  ][0], mv_cache[y][x  ][1]},

  687.                 {mv_cache[y-1][x][0], mv_cache[y-1][x][1]},

  688.                 {mv_cache[y][x-1][0], mv_cache[y][x-1][1]},

  689.                 {mv_cache[y][x+1][0], mv_cache[y][x+1][1]},

  690.                 {mv_cache[y+1][x][0], mv_cache[y+1][x][1]}};

  691.             //FIXME cleanup

  692.             obmc_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,

  693.                         ref_picture[0],

  694.                         mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >>1) * 8,

  695.                         pix_op[1],

  696.                         mv);

  697. 

  698.             mx += mv[0][0];

  699.             my += mv[0][1];

  700.         }

  701.         if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY))

  702.             chroma_4mv_motion(s, dest_cb, dest_cr, ref_picture, pix_op[1], mx, my);

  703. 

  704.         return;

  705.     }

  706. 

  707.     switch(s->mv_type) {

  708.     case MV_TYPE_16X16:

  709.         if(s->mcsel){

  710.             if(s->real_sprite_warping_points==1){

  711.                 gmc1_motion(s, dest_y, dest_cb, dest_cr,

  712.                             ref_picture);

  713.             }else{

  714.                 gmc_motion(s, dest_y, dest_cb, dest_cr,

  715.                             ref_picture);

  716.             }

  717.         }else if(!is_mpeg12 && s->quarter_sample){

  718.             qpel_motion(s, dest_y, dest_cb, dest_cr,

  719.                         0, 0, 0,

  720.                         ref_picture, pix_op, qpix_op,

  721.                         s->mv[dir][0][0], s->mv[dir][0][1], 16);

  722.         } else if (!is_mpeg12 && (CONFIG_WMV2_DECODER || CONFIG_WMV2_ENCODER) &&

  723.                    s->mspel && s->codec_id == AV_CODEC_ID_WMV2) {

  724.             ff_mspel_motion(s, dest_y, dest_cb, dest_cr,

  725.                         ref_picture, pix_op,

  726.                         s->mv[dir][0][0], s->mv[dir][0][1], 16);

  727.         }else

  728.         {

  729.             mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,

  730.                         ref_picture, pix_op,

  731.                         s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y);

  732.         }

  733.         break;

  734.     case MV_TYPE_8X8:

  735.     if (!is_mpeg12) {

  736.         mx = 0;

  737.         my = 0;

  738.         if(s->quarter_sample){

  739.             for(i=0;i<4;i++) {

  740.                 motion_x = s->mv[dir][i][0];

  741.                 motion_y = s->mv[dir][i][1];

  742. 

  743.                 dxy = ((motion_y & 3) << 2) | (motion_x & 3);

  744.                 src_x = mb_x * 16 + (motion_x >> 2) + (i & 1) * 8;

  745.                 src_y = mb_y * 16 + (motion_y >> 2) + (i >>1) * 8;

  746. 

  747.                 /* WARNING: do no forget half pels */

  748.                 src_x = av_clip(src_x, -16, s->width);

  749.                 if (src_x == s->width)

  750.                     dxy &= ~3;

  751.                 src_y = av_clip(src_y, -16, s->height);

  752.                 if (src_y == s->height)

  753.                     dxy &= ~12;

  754. 

  755.                 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x);

  756.                 if(s->flags&CODEC_FLAG_EMU_EDGE){

  757.                     if(   (unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x&3) - 8, 0)

  758.                        || (unsigned)src_y > FFMAX(s->v_edge_pos - (motion_y&3) - 8, 0)){

  759.                         s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,

  760.                                             s->linesize, 9, 9,

  761.                                             src_x, src_y,

  762.                                             s->h_edge_pos, s->v_edge_pos);

  763.                         ptr= s->edge_emu_buffer;

  764.                     }

  765.                 }

  766.                 dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize;

  767.                 qpix_op[1][dxy](dest, ptr, s->linesize);

  768. 

  769.                 mx += s->mv[dir][i][0]/2;

  770.                 my += s->mv[dir][i][1]/2;

  771.             }

  772.         }else{

  773.             for(i=0;i<4;i++) {

  774.                 hpel_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,

  775.                             ref_picture[0],

  776.                             mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >>1) * 8,

  777.                             pix_op[1],

  778.                             s->mv[dir][i][0], s->mv[dir][i][1]);

  779. 

  780.                 mx += s->mv[dir][i][0];

  781.                 my += s->mv[dir][i][1];

  782.             }

  783.         }

  784. 

  785.         if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY))

  786.             chroma_4mv_motion(s, dest_cb, dest_cr, ref_picture, pix_op[1], mx, my);

  787.     }

  788.         break;

  789.     case MV_TYPE_FIELD:

  790.         if (s->picture_structure == PICT_FRAME) {

  791.             if(!is_mpeg12 && s->quarter_sample){

  792.                 for(i=0; i<2; i++){

  793.                     qpel_motion(s, dest_y, dest_cb, dest_cr,

  794.                                 1, i, s->field_select[dir][i],

  795.                                 ref_picture, pix_op, qpix_op,

  796.                                 s->mv[dir][i][0], s->mv[dir][i][1], 8);

  797.                 }

  798.             }else{

  799.                 /* top field */

  800.                 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,

  801.                                   0, s->field_select[dir][0],

  802.                                   ref_picture, pix_op,

  803.                                   s->mv[dir][0][0], s->mv[dir][0][1], 8, mb_y);

  804.                 /* bottom field */

  805.                 mpeg_motion_field(s, dest_y, dest_cb, dest_cr,

  806.                                   1, s->field_select[dir][1],

  807.                                   ref_picture, pix_op,

  808.                                   s->mv[dir][1][0], s->mv[dir][1][1], 8, mb_y);

  809.             }

  810.         } else {

  811.             if(s->picture_structure != s->field_select[dir][0] + 1 && s->pict_type != AV_PICTURE_TYPE_B && !s->first_field){

  812.                 ref_picture = s->current_picture_ptr->f.data;

  813.             }

  814. 

  815.             mpeg_motion(s, dest_y, dest_cb, dest_cr,

  816.                         s->field_select[dir][0],

  817.                         ref_picture, pix_op,

  818.                         s->mv[dir][0][0], s->mv[dir][0][1], 16, mb_y>>1);

  819.         }

  820.         break;

  821.     case MV_TYPE_16X8:

  822.         for(i=0; i<2; i++){

  823.             uint8_t ** ref2picture;

  824. 

  825.             if(s->picture_structure == s->field_select[dir][i] + 1

  826.                || s->pict_type == AV_PICTURE_TYPE_B || s->first_field){

  827.                 ref2picture= ref_picture;

  828.             }else{

  829.                 ref2picture = s->current_picture_ptr->f.data;

  830.             }

  831. 

  832.             mpeg_motion(s, dest_y, dest_cb, dest_cr,

  833.                         s->field_select[dir][i],

  834.                         ref2picture, pix_op,

  835.                         s->mv[dir][i][0], s->mv[dir][i][1] + 16*i, 8, mb_y>>1);

  836. 

  837.             dest_y += 16*s->linesize;

  838.             dest_cb+= (16>>s->chroma_y_shift)*s->uvlinesize;

  839.             dest_cr+= (16>>s->chroma_y_shift)*s->uvlinesize;

  840.         }

  841.         break;

  842.     case MV_TYPE_DMV:

  843.         if(s->picture_structure == PICT_FRAME){

  844.             for(i=0; i<2; i++){

  845.                 int j;

  846.                 for(j=0; j<2; j++){

  847.                     mpeg_motion_field(s, dest_y, dest_cb, dest_cr,

  848.                                       j, j^i, ref_picture, pix_op,

  849.                                       s->mv[dir][2*i + j][0],

  850.                                       s->mv[dir][2*i + j][1], 8, mb_y);

  851.                 }

  852.                 pix_op = s->hdsp.avg_pixels_tab;

  853.             }

  854.         }else{

  855.             for(i=0; i<2; i++){

  856.                 mpeg_motion(s, dest_y, dest_cb, dest_cr,

  857.                             s->picture_structure != i+1,

  858.                             ref_picture, pix_op,

  859.                             s->mv[dir][2*i][0],s->mv[dir][2*i][1],16, mb_y>>1);

  860. 

  861.                 // after put we make avg of the same block

  862.                 pix_op=s->hdsp.avg_pixels_tab;

  863. 

  864.                 //opposite parity is always in the same frame if this is second field

  865.                 if(!s->first_field){

  866.                     ref_picture = s->current_picture_ptr->f.data;

  867.                 }

  868.             }

  869.         }

  870.     break;

  871.     default: av_assert2(0);

  872.     }

  873. }

  874. 

  875. void ff_MPV_motion(MpegEncContext *s,

  876.                    uint8_t *dest_y, uint8_t *dest_cb,

  877.                    uint8_t *dest_cr, int dir,

  878.                    uint8_t **ref_picture,

  879.                    op_pixels_func (*pix_op)[4],

  880.                    qpel_mc_func (*qpix_op)[16])

  881. {

  882. #if !CONFIG_SMALL

  883.     if(s->out_format == FMT_MPEG1)

  884.         MPV_motion_internal(s, dest_y, dest_cb, dest_cr, dir,

  885.                             ref_picture, pix_op, qpix_op, 1);

  886.     else

  887. #endif

  888.         MPV_motion_internal(s, dest_y, dest_cb, dest_cr, dir,

  889.                             ref_picture, pix_op, qpix_op, 0);

  890. }