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.
61658 Commits
32 Branches
366MB
Tree: e98bac82e5
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'e98bac82e5'
${ noResults }
FFmpeg/libavcodec/motion_est.c

1766 lines
68KB
Raw Blame History 

  1. /*

  2.  * Motion estimation

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

  4.  * Copyright (c) 2002-2004 Michael Niedermayer

  5.  *

  6.  * new motion estimation (X1/EPZS) 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

  27.  * Motion estimation.

  28.  */

  29. 

  30. #include <stdlib.h>

  31. #include <stdio.h>

  32. #include <limits.h>

  33. 

  34. #include "avcodec.h"

  35. #include "mathops.h"

  36. #include "mpegutils.h"

  37. #include "mpegvideo.h"

  38. 

  39. #undef NDEBUG

  40. #include <assert.h>

  41. 

  42. #define P_LEFT P[1]

  43. #define P_TOP P[2]

  44. #define P_TOPRIGHT P[3]

  45. #define P_MEDIAN P[4]

  46. #define P_MV1 P[9]

  47. 

  48. static int sad_hpel_motion_search(MpegEncContext * s,

  49.                                   int *mx_ptr, int *my_ptr, int dmin,

  50.                                   int src_index, int ref_index,

  51.                                   int size, int h);

  52. 

  53. static inline unsigned update_map_generation(MotionEstContext *c)

  54. {

  55.     c->map_generation+= 1<<(ME_MAP_MV_BITS*2);

  56.     if(c->map_generation==0){

  57.         c->map_generation= 1<<(ME_MAP_MV_BITS*2);

  58.         memset(c->map, 0, sizeof(uint32_t)*ME_MAP_SIZE);

  59.     }

  60.     return c->map_generation;

  61. }

  62. 

  63. /* shape adaptive search stuff */

  64. typedef struct Minima{

  65.     int height;

  66.     int x, y;

  67.     int checked;

  68. }Minima;

  69. 

  70. static int minima_cmp(const void *a, const void *b){

  71.     const Minima *da = (const Minima *) a;

  72.     const Minima *db = (const Minima *) b;

  73. 

  74.     return da->height - db->height;

  75. }

  76. 

  77. #define FLAG_QPEL   1 //must be 1

  78. #define FLAG_CHROMA 2

  79. #define FLAG_DIRECT 4

  80. 

  81. static inline void init_ref(MotionEstContext *c, uint8_t *src[3], uint8_t *ref[3], uint8_t *ref2[3], int x, int y, int ref_index){

  82.     const int offset[3]= {

  83.           y*c->  stride + x,

  84.         ((y*c->uvstride + x)>>1),

  85.         ((y*c->uvstride + x)>>1),

  86.     };

  87.     int i;

  88.     for(i=0; i<3; i++){

  89.         c->src[0][i]= src [i] + offset[i];

  90.         c->ref[0][i]= ref [i] + offset[i];

  91.     }

  92.     if(ref_index){

  93.         for(i=0; i<3; i++){

  94.             c->ref[ref_index][i]= ref2[i] + offset[i];

  95.         }

  96.     }

  97. }

  98. 

  99. static int get_flags(MotionEstContext *c, int direct, int chroma){

  100.     return   ((c->avctx->flags&CODEC_FLAG_QPEL) ? FLAG_QPEL : 0)

  101.            + (direct ? FLAG_DIRECT : 0)

  102.            + (chroma ? FLAG_CHROMA : 0);

  103. }

  104. 

  105. static av_always_inline int cmp_direct_inline(MpegEncContext *s, const int x, const int y, const int subx, const int suby,

  106.                       const int size, const int h, int ref_index, int src_index,

  107.                       me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, int qpel){

  108.     MotionEstContext * const c= &s->me;

  109.     const int stride= c->stride;

  110.     const int hx= subx + (x<<(1+qpel));

  111.     const int hy= suby + (y<<(1+qpel));

  112.     uint8_t * const * const ref= c->ref[ref_index];

  113.     uint8_t * const * const src= c->src[src_index];

  114.     int d;

  115.     //FIXME check chroma 4mv, (no crashes ...)

  116.         av_assert2(x >= c->xmin && hx <= c->xmax<<(qpel+1) && y >= c->ymin && hy <= c->ymax<<(qpel+1));

  117.         if(x >= c->xmin && hx <= c->xmax<<(qpel+1) && y >= c->ymin && hy <= c->ymax<<(qpel+1)){

  118.             const int time_pp= s->pp_time;

  119.             const int time_pb= s->pb_time;

  120.             const int mask= 2*qpel+1;

  121.             if(s->mv_type==MV_TYPE_8X8){

  122.                 int i;

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

  124.                     int fx = c->direct_basis_mv[i][0] + hx;

  125.                     int fy = c->direct_basis_mv[i][1] + hy;

  126.                     int bx = hx ? fx - c->co_located_mv[i][0] : c->co_located_mv[i][0]*(time_pb - time_pp)/time_pp + ((i &1)<<(qpel+4));

  127.                     int by = hy ? fy - c->co_located_mv[i][1] : c->co_located_mv[i][1]*(time_pb - time_pp)/time_pp + ((i>>1)<<(qpel+4));

  128.                     int fxy= (fx&mask) + ((fy&mask)<<(qpel+1));

  129.                     int bxy= (bx&mask) + ((by&mask)<<(qpel+1));

  130. 

  131.                     uint8_t *dst= c->temp + 8*(i&1) + 8*stride*(i>>1);

  132.                     if(qpel){

  133.                         c->qpel_put[1][fxy](dst, ref[0] + (fx>>2) + (fy>>2)*stride, stride);

  134.                         c->qpel_avg[1][bxy](dst, ref[8] + (bx>>2) + (by>>2)*stride, stride);

  135.                     }else{

  136.                         c->hpel_put[1][fxy](dst, ref[0] + (fx>>1) + (fy>>1)*stride, stride, 8);

  137.                         c->hpel_avg[1][bxy](dst, ref[8] + (bx>>1) + (by>>1)*stride, stride, 8);

  138.                     }

  139.                 }

  140.             }else{

  141.                 int fx = c->direct_basis_mv[0][0] + hx;

  142.                 int fy = c->direct_basis_mv[0][1] + hy;

  143.                 int bx = hx ? fx - c->co_located_mv[0][0] : (c->co_located_mv[0][0]*(time_pb - time_pp)/time_pp);

  144.                 int by = hy ? fy - c->co_located_mv[0][1] : (c->co_located_mv[0][1]*(time_pb - time_pp)/time_pp);

  145.                 int fxy= (fx&mask) + ((fy&mask)<<(qpel+1));

  146.                 int bxy= (bx&mask) + ((by&mask)<<(qpel+1));

  147. 

  148.                 if(qpel){

  149.                     c->qpel_put[1][fxy](c->temp               , ref[0] + (fx>>2) + (fy>>2)*stride               , stride);

  150.                     c->qpel_put[1][fxy](c->temp + 8           , ref[0] + (fx>>2) + (fy>>2)*stride + 8           , stride);

  151.                     c->qpel_put[1][fxy](c->temp     + 8*stride, ref[0] + (fx>>2) + (fy>>2)*stride     + 8*stride, stride);

  152.                     c->qpel_put[1][fxy](c->temp + 8 + 8*stride, ref[0] + (fx>>2) + (fy>>2)*stride + 8 + 8*stride, stride);

  153.                     c->qpel_avg[1][bxy](c->temp               , ref[8] + (bx>>2) + (by>>2)*stride               , stride);

  154.                     c->qpel_avg[1][bxy](c->temp + 8           , ref[8] + (bx>>2) + (by>>2)*stride + 8           , stride);

  155.                     c->qpel_avg[1][bxy](c->temp     + 8*stride, ref[8] + (bx>>2) + (by>>2)*stride     + 8*stride, stride);

  156.                     c->qpel_avg[1][bxy](c->temp + 8 + 8*stride, ref[8] + (bx>>2) + (by>>2)*stride + 8 + 8*stride, stride);

  157.                 }else{

  158.                     av_assert2((fx>>1) + 16*s->mb_x >= -16);

  159.                     av_assert2((fy>>1) + 16*s->mb_y >= -16);

  160.                     av_assert2((fx>>1) + 16*s->mb_x <= s->width);

  161.                     av_assert2((fy>>1) + 16*s->mb_y <= s->height);

  162.                     av_assert2((bx>>1) + 16*s->mb_x >= -16);

  163.                     av_assert2((by>>1) + 16*s->mb_y >= -16);

  164.                     av_assert2((bx>>1) + 16*s->mb_x <= s->width);

  165.                     av_assert2((by>>1) + 16*s->mb_y <= s->height);

  166. 

  167.                     c->hpel_put[0][fxy](c->temp, ref[0] + (fx>>1) + (fy>>1)*stride, stride, 16);

  168.                     c->hpel_avg[0][bxy](c->temp, ref[8] + (bx>>1) + (by>>1)*stride, stride, 16);

  169.                 }

  170.             }

  171.             d = cmp_func(s, c->temp, src[0], stride, 16);

  172.         }else

  173.             d= 256*256*256*32;

  174.     return d;

  175. }

  176. 

  177. static av_always_inline int cmp_inline(MpegEncContext *s, const int x, const int y, const int subx, const int suby,

  178.                       const int size, const int h, int ref_index, int src_index,

  179.                       me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, int qpel, int chroma){

  180.     MotionEstContext * const c= &s->me;

  181.     const int stride= c->stride;

  182.     const int uvstride= c->uvstride;

  183.     const int dxy= subx + (suby<<(1+qpel)); //FIXME log2_subpel?

  184.     const int hx= subx + (x<<(1+qpel));

  185.     const int hy= suby + (y<<(1+qpel));

  186.     uint8_t * const * const ref= c->ref[ref_index];

  187.     uint8_t * const * const src= c->src[src_index];

  188.     int d;

  189.     //FIXME check chroma 4mv, (no crashes ...)

  190.         int uvdxy;              /* no, it might not be used uninitialized */

  191.         if(dxy){

  192.             if(qpel){

  193.                 c->qpel_put[size][dxy](c->temp, ref[0] + x + y*stride, stride); //FIXME prototype (add h)

  194.                 if(chroma){

  195.                     int cx= hx/2;

  196.                     int cy= hy/2;

  197.                     cx= (cx>>1)|(cx&1);

  198.                     cy= (cy>>1)|(cy&1);

  199.                     uvdxy= (cx&1) + 2*(cy&1);

  200.                     //FIXME x/y wrong, but mpeg4 qpel is sick anyway, we should drop as much of it as possible in favor for h264

  201.                 }

  202.             }else{

  203.                 c->hpel_put[size][dxy](c->temp, ref[0] + x + y*stride, stride, h);

  204.                 if(chroma)

  205.                     uvdxy= dxy | (x&1) | (2*(y&1));

  206.             }

  207.             d = cmp_func(s, c->temp, src[0], stride, h);

  208.         }else{

  209.             d = cmp_func(s, src[0], ref[0] + x + y*stride, stride, h);

  210.             if(chroma)

  211.                 uvdxy= (x&1) + 2*(y&1);

  212.         }

  213.         if(chroma){

  214.             uint8_t * const uvtemp= c->temp + 16*stride;

  215.             c->hpel_put[size+1][uvdxy](uvtemp  , ref[1] + (x>>1) + (y>>1)*uvstride, uvstride, h>>1);

  216.             c->hpel_put[size+1][uvdxy](uvtemp+8, ref[2] + (x>>1) + (y>>1)*uvstride, uvstride, h>>1);

  217.             d += chroma_cmp_func(s, uvtemp  , src[1], uvstride, h>>1);

  218.             d += chroma_cmp_func(s, uvtemp+8, src[2], uvstride, h>>1);

  219.         }

  220.     return d;

  221. }

  222. 

  223. static int cmp_simple(MpegEncContext *s, const int x, const int y,

  224.                       int ref_index, int src_index,

  225.                       me_cmp_func cmp_func, me_cmp_func chroma_cmp_func){

  226.     return cmp_inline(s,x,y,0,0,0,16,ref_index,src_index, cmp_func, chroma_cmp_func, 0, 0);

  227. }

  228. 

  229. static int cmp_fpel_internal(MpegEncContext *s, const int x, const int y,

  230.                       const int size, const int h, int ref_index, int src_index,

  231.                       me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){

  232.     if(flags&FLAG_DIRECT){

  233.         return cmp_direct_inline(s,x,y,0,0,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, flags&FLAG_QPEL);

  234.     }else{

  235.         return cmp_inline(s,x,y,0,0,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, 0, flags&FLAG_CHROMA);

  236.     }

  237. }

  238. 

  239. static int cmp_internal(MpegEncContext *s, const int x, const int y, const int subx, const int suby,

  240.                       const int size, const int h, int ref_index, int src_index,

  241.                       me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){

  242.     if(flags&FLAG_DIRECT){

  243.         return cmp_direct_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, flags&FLAG_QPEL);

  244.     }else{

  245.         return cmp_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, flags&FLAG_QPEL, flags&FLAG_CHROMA);

  246.     }

  247. }

  248. 

  249. /** @brief compares a block (either a full macroblock or a partition thereof)

  250.     against a proposed motion-compensated prediction of that block

  251.  */

  252. static av_always_inline int cmp(MpegEncContext *s, const int x, const int y, const int subx, const int suby,

  253.                       const int size, const int h, int ref_index, int src_index,

  254.                       me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){

  255.     if(av_builtin_constant_p(flags) && av_builtin_constant_p(h) && av_builtin_constant_p(size)

  256.        && av_builtin_constant_p(subx) && av_builtin_constant_p(suby)

  257.        && flags==0 && h==16 && size==0 && subx==0 && suby==0){

  258.         return cmp_simple(s,x,y,ref_index,src_index, cmp_func, chroma_cmp_func);

  259.     }else if(av_builtin_constant_p(subx) && av_builtin_constant_p(suby)

  260.        && subx==0 && suby==0){

  261.         return cmp_fpel_internal(s,x,y,size,h,ref_index,src_index, cmp_func, chroma_cmp_func,flags);

  262.     }else{

  263.         return cmp_internal(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, flags);

  264.     }

  265. }

  266. 

  267. static int cmp_hpel(MpegEncContext *s, const int x, const int y, const int subx, const int suby,

  268.                       const int size, const int h, int ref_index, int src_index,

  269.                       me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){

  270.     if(flags&FLAG_DIRECT){

  271.         return cmp_direct_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, 0);

  272.     }else{

  273.         return cmp_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, 0, flags&FLAG_CHROMA);

  274.     }

  275. }

  276. 

  277. static int cmp_qpel(MpegEncContext *s, const int x, const int y, const int subx, const int suby,

  278.                       const int size, const int h, int ref_index, int src_index,

  279.                       me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){

  280.     if(flags&FLAG_DIRECT){

  281.         return cmp_direct_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, 1);

  282.     }else{

  283.         return cmp_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, 1, flags&FLAG_CHROMA);

  284.     }

  285. }

  286. 

  287. #include "motion_est_template.c"

  288. 

  289. static int zero_cmp(MpegEncContext *s, uint8_t *a, uint8_t *b,

  290.                     int stride, int h)

  291. {

  292.     return 0;

  293. }

  294. 

  295. static void zero_hpel(uint8_t *a, const uint8_t *b, ptrdiff_t stride, int h){

  296. }

  297. 

  298. int ff_init_me(MpegEncContext *s){

  299.     MotionEstContext * const c= &s->me;

  300.     int cache_size= FFMIN(ME_MAP_SIZE>>ME_MAP_SHIFT, 1<<ME_MAP_SHIFT);

  301.     int dia_size= FFMAX(FFABS(s->avctx->dia_size)&255, FFABS(s->avctx->pre_dia_size)&255);

  302. 

  303.     if(FFMIN(s->avctx->dia_size, s->avctx->pre_dia_size) < -FFMIN(ME_MAP_SIZE, MAX_SAB_SIZE)){

  304.         av_log(s->avctx, AV_LOG_ERROR, "ME_MAP size is too small for SAB diamond\n");

  305.         return -1;

  306.     }

  307.     //special case of snow is needed because snow uses its own iterative ME code

  308.     if(s->me_method!=ME_ZERO && s->me_method!=ME_EPZS && s->me_method!=ME_X1 && s->avctx->codec_id != AV_CODEC_ID_SNOW){

  309.         av_log(s->avctx, AV_LOG_ERROR, "me_method is only allowed to be set to zero and epzs; for hex,umh,full and others see dia_size\n");

  310.         return -1;

  311.     }

  312. 

  313.     c->avctx= s->avctx;

  314. 

  315.     if(cache_size < 2*dia_size && !c->stride){

  316.         av_log(s->avctx, AV_LOG_INFO, "ME_MAP size may be a little small for the selected diamond size\n");

  317.     }

  318. 

  319.     ff_set_cmp(&s->dsp, s->dsp.me_pre_cmp, c->avctx->me_pre_cmp);

  320.     ff_set_cmp(&s->dsp, s->dsp.me_cmp, c->avctx->me_cmp);

  321.     ff_set_cmp(&s->dsp, s->dsp.me_sub_cmp, c->avctx->me_sub_cmp);

  322.     ff_set_cmp(&s->dsp, s->dsp.mb_cmp, c->avctx->mb_cmp);

  323. 

  324.     c->flags    = get_flags(c, 0, c->avctx->me_cmp    &FF_CMP_CHROMA);

  325.     c->sub_flags= get_flags(c, 0, c->avctx->me_sub_cmp&FF_CMP_CHROMA);

  326.     c->mb_flags = get_flags(c, 0, c->avctx->mb_cmp    &FF_CMP_CHROMA);

  327. 

  328. /*FIXME s->no_rounding b_type*/

  329.     if(s->flags&CODEC_FLAG_QPEL){

  330.         c->sub_motion_search= qpel_motion_search;

  331.         c->qpel_avg= s->dsp.avg_qpel_pixels_tab;

  332.         if(s->no_rounding) c->qpel_put= s->dsp.put_no_rnd_qpel_pixels_tab;

  333.         else               c->qpel_put= s->dsp.put_qpel_pixels_tab;

  334.     }else{

  335.         if(c->avctx->me_sub_cmp&FF_CMP_CHROMA)

  336.             c->sub_motion_search= hpel_motion_search;

  337.         else if(   c->avctx->me_sub_cmp == FF_CMP_SAD

  338.                 && c->avctx->    me_cmp == FF_CMP_SAD

  339.                 && c->avctx->    mb_cmp == FF_CMP_SAD)

  340.             c->sub_motion_search= sad_hpel_motion_search; // 2050 vs. 2450 cycles

  341.         else

  342.             c->sub_motion_search= hpel_motion_search;

  343.     }

  344.     c->hpel_avg = s->hdsp.avg_pixels_tab;

  345.     if (s->no_rounding)

  346.         c->hpel_put = s->hdsp.put_no_rnd_pixels_tab;

  347.     else

  348.         c->hpel_put = s->hdsp.put_pixels_tab;

  349. 

  350.     if(s->linesize){

  351.         c->stride  = s->linesize;

  352.         c->uvstride= s->uvlinesize;

  353.     }else{

  354.         c->stride  = 16*s->mb_width + 32;

  355.         c->uvstride=  8*s->mb_width + 16;

  356.     }

  357. 

  358.     /* 8x8 fullpel search would need a 4x4 chroma compare, which we do

  359.      * not have yet, and even if we had, the motion estimation code

  360.      * does not expect it. */

  361.     if(s->codec_id != AV_CODEC_ID_SNOW){

  362.         if((c->avctx->me_cmp&FF_CMP_CHROMA)/* && !s->dsp.me_cmp[2]*/){

  363.             s->dsp.me_cmp[2]= zero_cmp;

  364.         }

  365.         if((c->avctx->me_sub_cmp&FF_CMP_CHROMA) && !s->dsp.me_sub_cmp[2]){

  366.             s->dsp.me_sub_cmp[2]= zero_cmp;

  367.         }

  368.         c->hpel_put[2][0]= c->hpel_put[2][1]=

  369.         c->hpel_put[2][2]= c->hpel_put[2][3]= zero_hpel;

  370.     }

  371. 

  372.     if(s->codec_id == AV_CODEC_ID_H261){

  373.         c->sub_motion_search= no_sub_motion_search;

  374.     }

  375. 

  376.     return 0;

  377. }

  378. 

  379. #define CHECK_SAD_HALF_MV(suffix, x, y) \

  380. {\

  381.     d= s->dsp.pix_abs[size][(x?1:0)+(y?2:0)](NULL, pix, ptr+((x)>>1), stride, h);\

  382.     d += (mv_penalty[pen_x + x] + mv_penalty[pen_y + y])*penalty_factor;\

  383.     COPY3_IF_LT(dminh, d, dx, x, dy, y)\

  384. }

  385. 

  386. static int sad_hpel_motion_search(MpegEncContext * s,

  387.                                   int *mx_ptr, int *my_ptr, int dmin,

  388.                                   int src_index, int ref_index,

  389.                                   int size, int h)

  390. {

  391.     MotionEstContext * const c= &s->me;

  392.     const int penalty_factor= c->sub_penalty_factor;

  393.     int mx, my, dminh;

  394.     uint8_t *pix, *ptr;

  395.     int stride= c->stride;

  396.     LOAD_COMMON

  397. 

  398.     av_assert2(c->sub_flags == 0);

  399. 

  400.     if(c->skip){

  401.         *mx_ptr = 0;

  402.         *my_ptr = 0;

  403.         return dmin;

  404.     }

  405. 

  406.     pix = c->src[src_index][0];

  407. 

  408.     mx = *mx_ptr;

  409.     my = *my_ptr;

  410.     ptr = c->ref[ref_index][0] + (my * stride) + mx;

  411. 

  412.     dminh = dmin;

  413. 

  414.     if (mx > xmin && mx < xmax &&

  415.         my > ymin && my < ymax) {

  416.         int dx=0, dy=0;

  417.         int d, pen_x, pen_y;

  418.         const int index= (my<<ME_MAP_SHIFT) + mx;

  419.         const int t= score_map[(index-(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)];

  420.         const int l= score_map[(index- 1               )&(ME_MAP_SIZE-1)];

  421.         const int r= score_map[(index+ 1               )&(ME_MAP_SIZE-1)];

  422.         const int b= score_map[(index+(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)];

  423.         mx<<=1;

  424.         my<<=1;

  425. 

  426. 

  427.         pen_x= pred_x + mx;

  428.         pen_y= pred_y + my;

  429. 

  430.         ptr-= stride;

  431.         if(t<=b){

  432.             CHECK_SAD_HALF_MV(y2 , 0, -1)

  433.             if(l<=r){

  434.                 CHECK_SAD_HALF_MV(xy2, -1, -1)

  435.                 if(t+r<=b+l){

  436.                     CHECK_SAD_HALF_MV(xy2, +1, -1)

  437.                     ptr+= stride;

  438.                 }else{

  439.                     ptr+= stride;

  440.                     CHECK_SAD_HALF_MV(xy2, -1, +1)

  441.                 }

  442.                 CHECK_SAD_HALF_MV(x2 , -1,  0)

  443.             }else{

  444.                 CHECK_SAD_HALF_MV(xy2, +1, -1)

  445.                 if(t+l<=b+r){

  446.                     CHECK_SAD_HALF_MV(xy2, -1, -1)

  447.                     ptr+= stride;

  448.                 }else{

  449.                     ptr+= stride;

  450.                     CHECK_SAD_HALF_MV(xy2, +1, +1)

  451.                 }

  452.                 CHECK_SAD_HALF_MV(x2 , +1,  0)

  453.             }

  454.         }else{

  455.             if(l<=r){

  456.                 if(t+l<=b+r){

  457.                     CHECK_SAD_HALF_MV(xy2, -1, -1)

  458.                     ptr+= stride;

  459.                 }else{

  460.                     ptr+= stride;

  461.                     CHECK_SAD_HALF_MV(xy2, +1, +1)

  462.                 }

  463.                 CHECK_SAD_HALF_MV(x2 , -1,  0)

  464.                 CHECK_SAD_HALF_MV(xy2, -1, +1)

  465.             }else{

  466.                 if(t+r<=b+l){

  467.                     CHECK_SAD_HALF_MV(xy2, +1, -1)

  468.                     ptr+= stride;

  469.                 }else{

  470.                     ptr+= stride;

  471.                     CHECK_SAD_HALF_MV(xy2, -1, +1)

  472.                 }

  473.                 CHECK_SAD_HALF_MV(x2 , +1,  0)

  474.                 CHECK_SAD_HALF_MV(xy2, +1, +1)

  475.             }

  476.             CHECK_SAD_HALF_MV(y2 ,  0, +1)

  477.         }

  478.         mx+=dx;

  479.         my+=dy;

  480. 

  481.     }else{

  482.         mx<<=1;

  483.         my<<=1;

  484.     }

  485. 

  486.     *mx_ptr = mx;

  487.     *my_ptr = my;

  488.     return dminh;

  489. }

  490. 

  491. static inline void set_p_mv_tables(MpegEncContext * s, int mx, int my, int mv4)

  492. {

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

  494. 

  495.     s->p_mv_table[xy][0] = mx;

  496.     s->p_mv_table[xy][1] = my;

  497. 

  498.     /* has already been set to the 4 MV if 4MV is done */

  499.     if(mv4){

  500.         int mot_xy= s->block_index[0];

  501. 

  502.         s->current_picture.motion_val[0][mot_xy    ][0] = mx;

  503.         s->current_picture.motion_val[0][mot_xy    ][1] = my;

  504.         s->current_picture.motion_val[0][mot_xy + 1][0] = mx;

  505.         s->current_picture.motion_val[0][mot_xy + 1][1] = my;

  506. 

  507.         mot_xy += s->b8_stride;

  508.         s->current_picture.motion_val[0][mot_xy    ][0] = mx;

  509.         s->current_picture.motion_val[0][mot_xy    ][1] = my;

  510.         s->current_picture.motion_val[0][mot_xy + 1][0] = mx;

  511.         s->current_picture.motion_val[0][mot_xy + 1][1] = my;

  512.     }

  513. }

  514. 

  515. /**

  516.  * get fullpel ME search limits.

  517.  */

  518. static inline void get_limits(MpegEncContext *s, int x, int y)

  519. {

  520.     MotionEstContext * const c= &s->me;

  521.     int range= c->avctx->me_range >> (1 + !!(c->flags&FLAG_QPEL));

  522.     int max_range = MAX_MV >> (1 + !!(c->flags&FLAG_QPEL));

  523. /*

  524.     if(c->avctx->me_range) c->range= c->avctx->me_range >> 1;

  525.     else                   c->range= 16;

  526. */

  527.     if (s->unrestricted_mv) {

  528.         c->xmin = - x - 16;

  529.         c->ymin = - y - 16;

  530.         c->xmax = - x + s->width;

  531.         c->ymax = - y + s->height;

  532.     } else if (s->out_format == FMT_H261){

  533.         // Search range of H261 is different from other codec standards

  534.         c->xmin = (x > 15) ? - 15 : 0;

  535.         c->ymin = (y > 15) ? - 15 : 0;

  536.         c->xmax = (x < s->mb_width * 16 - 16) ? 15 : 0;

  537.         c->ymax = (y < s->mb_height * 16 - 16) ? 15 : 0;

  538.     } else {

  539.         c->xmin = - x;

  540.         c->ymin = - y;

  541.         c->xmax = - x + s->mb_width *16 - 16;

  542.         c->ymax = - y + s->mb_height*16 - 16;

  543.     }

  544.     if(!range || range > max_range)

  545.         range = max_range;

  546.     if(range){

  547.         c->xmin = FFMAX(c->xmin,-range);

  548.         c->xmax = FFMIN(c->xmax, range);

  549.         c->ymin = FFMAX(c->ymin,-range);

  550.         c->ymax = FFMIN(c->ymax, range);

  551.     }

  552. }

  553. 

  554. static inline void init_mv4_ref(MotionEstContext *c){

  555.     const int stride= c->stride;

  556. 

  557.     c->ref[1][0] = c->ref[0][0] + 8;

  558.     c->ref[2][0] = c->ref[0][0] + 8*stride;

  559.     c->ref[3][0] = c->ref[2][0] + 8;

  560.     c->src[1][0] = c->src[0][0] + 8;

  561.     c->src[2][0] = c->src[0][0] + 8*stride;

  562.     c->src[3][0] = c->src[2][0] + 8;

  563. }

  564. 

  565. static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift)

  566. {

  567.     MotionEstContext * const c= &s->me;

  568.     const int size= 1;

  569.     const int h=8;

  570.     int block;

  571.     int P[10][2];

  572.     int dmin_sum=0, mx4_sum=0, my4_sum=0, i;

  573.     int same=1;

  574.     const int stride= c->stride;

  575.     uint8_t *mv_penalty= c->current_mv_penalty;

  576.     int saftey_cliping= s->unrestricted_mv && (s->width&15) && (s->height&15);

  577. 

  578.     init_mv4_ref(c);

  579. 

  580.     for(block=0; block<4; block++){

  581.         int mx4, my4;

  582.         int pred_x4, pred_y4;

  583.         int dmin4;

  584.         static const int off[4]= {2, 1, 1, -1};

  585.         const int mot_stride = s->b8_stride;

  586.         const int mot_xy = s->block_index[block];

  587. 

  588.         if(saftey_cliping){

  589.             c->xmax = - 16*s->mb_x + s->width  - 8*(block &1);

  590.             c->ymax = - 16*s->mb_y + s->height - 8*(block>>1);

  591.         }

  592. 

  593.         P_LEFT[0] = s->current_picture.motion_val[0][mot_xy - 1][0];

  594.         P_LEFT[1] = s->current_picture.motion_val[0][mot_xy - 1][1];

  595. 

  596.         if(P_LEFT[0]       > (c->xmax<<shift)) P_LEFT[0]       = (c->xmax<<shift);

  597. 

  598.         /* special case for first line */

  599.         if (s->first_slice_line && block<2) {

  600.             c->pred_x= pred_x4= P_LEFT[0];

  601.             c->pred_y= pred_y4= P_LEFT[1];

  602.         } else {

  603.             P_TOP[0]      = s->current_picture.motion_val[0][mot_xy - mot_stride             ][0];

  604.             P_TOP[1]      = s->current_picture.motion_val[0][mot_xy - mot_stride             ][1];

  605.             P_TOPRIGHT[0] = s->current_picture.motion_val[0][mot_xy - mot_stride + off[block]][0];

  606.             P_TOPRIGHT[1] = s->current_picture.motion_val[0][mot_xy - mot_stride + off[block]][1];

  607.             if(P_TOP[1]      > (c->ymax<<shift)) P_TOP[1]     = (c->ymax<<shift);

  608.             if(P_TOPRIGHT[0] < (c->xmin<<shift)) P_TOPRIGHT[0]= (c->xmin<<shift);

  609.             if(P_TOPRIGHT[0] > (c->xmax<<shift)) P_TOPRIGHT[0]= (c->xmax<<shift);

  610.             if(P_TOPRIGHT[1] > (c->ymax<<shift)) P_TOPRIGHT[1]= (c->ymax<<shift);

  611. 

  612.             P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);

  613.             P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);

  614. 

  615.             c->pred_x= pred_x4 = P_MEDIAN[0];

  616.             c->pred_y= pred_y4 = P_MEDIAN[1];

  617.         }

  618.         P_MV1[0]= mx;

  619.         P_MV1[1]= my;

  620.         if(saftey_cliping)

  621.             for(i=0; i<10; i++){

  622.                 if(P[i][0] > (c->xmax<<shift)) P[i][0]= (c->xmax<<shift);

  623.                 if(P[i][1] > (c->ymax<<shift)) P[i][1]= (c->ymax<<shift);

  624.             }

  625. 

  626.         dmin4 = epzs_motion_search4(s, &mx4, &my4, P, block, block, s->p_mv_table, (1<<16)>>shift);

  627. 

  628.         dmin4= c->sub_motion_search(s, &mx4, &my4, dmin4, block, block, size, h);

  629. 

  630.         if(s->dsp.me_sub_cmp[0] != s->dsp.mb_cmp[0]){

  631.             int dxy;

  632.             const int offset= ((block&1) + (block>>1)*stride)*8;

  633.             uint8_t *dest_y = c->scratchpad + offset;

  634.             if(s->quarter_sample){

  635.                 uint8_t *ref= c->ref[block][0] + (mx4>>2) + (my4>>2)*stride;

  636.                 dxy = ((my4 & 3) << 2) | (mx4 & 3);

  637. 

  638.                 if(s->no_rounding)

  639.                     s->dsp.put_no_rnd_qpel_pixels_tab[1][dxy](dest_y   , ref    , stride);

  640.                 else

  641.                     s->dsp.put_qpel_pixels_tab       [1][dxy](dest_y   , ref    , stride);

  642.             }else{

  643.                 uint8_t *ref= c->ref[block][0] + (mx4>>1) + (my4>>1)*stride;

  644.                 dxy = ((my4 & 1) << 1) | (mx4 & 1);

  645. 

  646.                 if(s->no_rounding)

  647.                     s->hdsp.put_no_rnd_pixels_tab[1][dxy](dest_y    , ref    , stride, h);

  648.                 else

  649.                     s->hdsp.put_pixels_tab       [1][dxy](dest_y    , ref    , stride, h);

  650.             }

  651.             dmin_sum+= (mv_penalty[mx4-pred_x4] + mv_penalty[my4-pred_y4])*c->mb_penalty_factor;

  652.         }else

  653.             dmin_sum+= dmin4;

  654. 

  655.         if(s->quarter_sample){

  656.             mx4_sum+= mx4/2;

  657.             my4_sum+= my4/2;

  658.         }else{

  659.             mx4_sum+= mx4;

  660.             my4_sum+= my4;

  661.         }

  662. 

  663.         s->current_picture.motion_val[0][s->block_index[block]][0] = mx4;

  664.         s->current_picture.motion_val[0][s->block_index[block]][1] = my4;

  665. 

  666.         if(mx4 != mx || my4 != my) same=0;

  667.     }

  668. 

  669.     if(same)

  670.         return INT_MAX;

  671. 

  672.     if(s->dsp.me_sub_cmp[0] != s->dsp.mb_cmp[0]){

  673.         dmin_sum += s->dsp.mb_cmp[0](s, s->new_picture.f.data[0] + s->mb_x*16 + s->mb_y*16*stride, c->scratchpad, stride, 16);

  674.     }

  675. 

  676.     if(c->avctx->mb_cmp&FF_CMP_CHROMA){

  677.         int dxy;

  678.         int mx, my;

  679.         int offset;

  680. 

  681.         mx= ff_h263_round_chroma(mx4_sum);

  682.         my= ff_h263_round_chroma(my4_sum);

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

  684. 

  685.         offset= (s->mb_x*8 + (mx>>1)) + (s->mb_y*8 + (my>>1))*s->uvlinesize;

  686. 

  687.         if(s->no_rounding){

  688.             s->hdsp.put_no_rnd_pixels_tab[1][dxy](c->scratchpad    , s->last_picture.f.data[1] + offset, s->uvlinesize, 8);

  689.             s->hdsp.put_no_rnd_pixels_tab[1][dxy](c->scratchpad + 8, s->last_picture.f.data[2] + offset, s->uvlinesize, 8);

  690.         }else{

  691.             s->hdsp.put_pixels_tab       [1][dxy](c->scratchpad    , s->last_picture.f.data[1] + offset, s->uvlinesize, 8);

  692.             s->hdsp.put_pixels_tab       [1][dxy](c->scratchpad + 8, s->last_picture.f.data[2] + offset, s->uvlinesize, 8);

  693.         }

  694. 

  695.         dmin_sum += s->dsp.mb_cmp[1](s, s->new_picture.f.data[1] + s->mb_x*8 + s->mb_y*8*s->uvlinesize, c->scratchpad  , s->uvlinesize, 8);

  696.         dmin_sum += s->dsp.mb_cmp[1](s, s->new_picture.f.data[2] + s->mb_x*8 + s->mb_y*8*s->uvlinesize, c->scratchpad+8, s->uvlinesize, 8);

  697.     }

  698. 

  699.     c->pred_x= mx;

  700.     c->pred_y= my;

  701. 

  702.     switch(c->avctx->mb_cmp&0xFF){

  703.     /*case FF_CMP_SSE:

  704.         return dmin_sum+ 32*s->qscale*s->qscale;*/

  705.     case FF_CMP_RD:

  706.         return dmin_sum;

  707.     default:

  708.         return dmin_sum+ 11*c->mb_penalty_factor;

  709.     }

  710. }

  711. 

  712. static inline void init_interlaced_ref(MpegEncContext *s, int ref_index){

  713.     MotionEstContext * const c= &s->me;

  714. 

  715.     c->ref[1+ref_index][0] = c->ref[0+ref_index][0] + s->linesize;

  716.     c->src[1][0] = c->src[0][0] + s->linesize;

  717.     if(c->flags & FLAG_CHROMA){

  718.         c->ref[1+ref_index][1] = c->ref[0+ref_index][1] + s->uvlinesize;

  719.         c->ref[1+ref_index][2] = c->ref[0+ref_index][2] + s->uvlinesize;

  720.         c->src[1][1] = c->src[0][1] + s->uvlinesize;

  721.         c->src[1][2] = c->src[0][2] + s->uvlinesize;

  722.     }

  723. }

  724. 

  725. static int interlaced_search(MpegEncContext *s, int ref_index,

  726.                              int16_t (*mv_tables[2][2])[2], uint8_t *field_select_tables[2], int mx, int my, int user_field_select)

  727. {

  728.     MotionEstContext * const c= &s->me;

  729.     const int size=0;

  730.     const int h=8;

  731.     int block;

  732.     int P[10][2];

  733.     uint8_t * const mv_penalty= c->current_mv_penalty;

  734.     int same=1;

  735.     const int stride= 2*s->linesize;

  736.     int dmin_sum= 0;

  737.     const int mot_stride= s->mb_stride;

  738.     const int xy= s->mb_x + s->mb_y*mot_stride;

  739. 

  740.     c->ymin>>=1;

  741.     c->ymax>>=1;

  742.     c->stride<<=1;

  743.     c->uvstride<<=1;

  744.     init_interlaced_ref(s, ref_index);

  745. 

  746.     for(block=0; block<2; block++){

  747.         int field_select;

  748.         int best_dmin= INT_MAX;

  749.         int best_field= -1;

  750. 

  751.         for(field_select=0; field_select<2; field_select++){

  752.             int dmin, mx_i, my_i;

  753.             int16_t (*mv_table)[2]= mv_tables[block][field_select];

  754. 

  755.             if(user_field_select){

  756.                 av_assert1(field_select==0 || field_select==1);

  757.                 av_assert1(field_select_tables[block][xy]==0 || field_select_tables[block][xy]==1);

  758.                 if(field_select_tables[block][xy] != field_select)

  759.                     continue;

  760.             }

  761. 

  762.             P_LEFT[0] = mv_table[xy - 1][0];

  763.             P_LEFT[1] = mv_table[xy - 1][1];

  764.             if(P_LEFT[0]       > (c->xmax<<1)) P_LEFT[0]       = (c->xmax<<1);

  765. 

  766.             c->pred_x= P_LEFT[0];

  767.             c->pred_y= P_LEFT[1];

  768. 

  769.             if(!s->first_slice_line){

  770.                 P_TOP[0]      = mv_table[xy - mot_stride][0];

  771.                 P_TOP[1]      = mv_table[xy - mot_stride][1];

  772.                 P_TOPRIGHT[0] = mv_table[xy - mot_stride + 1][0];

  773.                 P_TOPRIGHT[1] = mv_table[xy - mot_stride + 1][1];

  774.                 if(P_TOP[1]      > (c->ymax<<1)) P_TOP[1]     = (c->ymax<<1);

  775.                 if(P_TOPRIGHT[0] < (c->xmin<<1)) P_TOPRIGHT[0]= (c->xmin<<1);

  776.                 if(P_TOPRIGHT[0] > (c->xmax<<1)) P_TOPRIGHT[0]= (c->xmax<<1);

  777.                 if(P_TOPRIGHT[1] > (c->ymax<<1)) P_TOPRIGHT[1]= (c->ymax<<1);

  778. 

  779.                 P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);

  780.                 P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);

  781.             }

  782.             P_MV1[0]= mx; //FIXME not correct if block != field_select

  783.             P_MV1[1]= my / 2;

  784. 

  785.             dmin = epzs_motion_search2(s, &mx_i, &my_i, P, block, field_select+ref_index, mv_table, (1<<16)>>1);

  786. 

  787.             dmin= c->sub_motion_search(s, &mx_i, &my_i, dmin, block, field_select+ref_index, size, h);

  788. 

  789.             mv_table[xy][0]= mx_i;

  790.             mv_table[xy][1]= my_i;

  791. 

  792.             if(s->dsp.me_sub_cmp[0] != s->dsp.mb_cmp[0]){

  793.                 int dxy;

  794. 

  795.                 //FIXME chroma ME

  796.                 uint8_t *ref= c->ref[field_select+ref_index][0] + (mx_i>>1) + (my_i>>1)*stride;

  797.                 dxy = ((my_i & 1) << 1) | (mx_i & 1);

  798. 

  799.                 if(s->no_rounding){

  800.                     s->hdsp.put_no_rnd_pixels_tab[size][dxy](c->scratchpad, ref    , stride, h);

  801.                 }else{

  802.                     s->hdsp.put_pixels_tab       [size][dxy](c->scratchpad, ref    , stride, h);

  803.                 }

  804.                 dmin= s->dsp.mb_cmp[size](s, c->src[block][0], c->scratchpad, stride, h);

  805.                 dmin+= (mv_penalty[mx_i-c->pred_x] + mv_penalty[my_i-c->pred_y] + 1)*c->mb_penalty_factor;

  806.             }else

  807.                 dmin+= c->mb_penalty_factor; //field_select bits

  808. 

  809.             dmin += field_select != block; //slightly prefer same field

  810. 

  811.             if(dmin < best_dmin){

  812.                 best_dmin= dmin;

  813.                 best_field= field_select;

  814.             }

  815.         }

  816.         {

  817.             int16_t (*mv_table)[2]= mv_tables[block][best_field];

  818. 

  819.             if(mv_table[xy][0] != mx) same=0; //FIXME check if these checks work and are any good at all

  820.             if(mv_table[xy][1]&1) same=0;

  821.             if(mv_table[xy][1]*2 != my) same=0;

  822.             if(best_field != block) same=0;

  823.         }

  824. 

  825.         field_select_tables[block][xy]= best_field;

  826.         dmin_sum += best_dmin;

  827.     }

  828. 

  829.     c->ymin<<=1;

  830.     c->ymax<<=1;

  831.     c->stride>>=1;

  832.     c->uvstride>>=1;

  833. 

  834.     if(same)

  835.         return INT_MAX;

  836. 

  837.     switch(c->avctx->mb_cmp&0xFF){

  838.     /*case FF_CMP_SSE:

  839.         return dmin_sum+ 32*s->qscale*s->qscale;*/

  840.     case FF_CMP_RD:

  841.         return dmin_sum;

  842.     default:

  843.         return dmin_sum+ 11*c->mb_penalty_factor;

  844.     }

  845. }

  846. 

  847. static inline int get_penalty_factor(int lambda, int lambda2, int type){

  848.     switch(type&0xFF){

  849.     default:

  850.     case FF_CMP_SAD:

  851.         return lambda>>FF_LAMBDA_SHIFT;

  852.     case FF_CMP_DCT:

  853.         return (3*lambda)>>(FF_LAMBDA_SHIFT+1);

  854.     case FF_CMP_W53:

  855.         return (4*lambda)>>(FF_LAMBDA_SHIFT);

  856.     case FF_CMP_W97:

  857.         return (2*lambda)>>(FF_LAMBDA_SHIFT);

  858.     case FF_CMP_SATD:

  859.     case FF_CMP_DCT264:

  860.         return (2*lambda)>>FF_LAMBDA_SHIFT;

  861.     case FF_CMP_RD:

  862.     case FF_CMP_PSNR:

  863.     case FF_CMP_SSE:

  864.     case FF_CMP_NSSE:

  865.         return lambda2>>FF_LAMBDA_SHIFT;

  866.     case FF_CMP_BIT:

  867.         return 1;

  868.     }

  869. }

  870. 

  871. void ff_estimate_p_frame_motion(MpegEncContext * s,

  872.                                 int mb_x, int mb_y)

  873. {

  874.     MotionEstContext * const c= &s->me;

  875.     uint8_t *pix, *ppix;

  876.     int sum, mx, my, dmin;

  877.     int varc;            ///< the variance of the block (sum of squared (p[y][x]-average))

  878.     int vard;            ///< sum of squared differences with the estimated motion vector

  879.     int P[10][2];

  880.     const int shift= 1+s->quarter_sample;

  881.     int mb_type=0;

  882.     Picture * const pic= &s->current_picture;

  883. 

  884.     init_ref(c, s->new_picture.f.data, s->last_picture.f.data, NULL, 16*mb_x, 16*mb_y, 0);

  885. 

  886.     av_assert0(s->quarter_sample==0 || s->quarter_sample==1);

  887.     av_assert0(s->linesize == c->stride);

  888.     av_assert0(s->uvlinesize == c->uvstride);

  889. 

  890.     c->penalty_factor    = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_cmp);

  891.     c->sub_penalty_factor= get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_sub_cmp);

  892.     c->mb_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->mb_cmp);

  893.     c->current_mv_penalty= c->mv_penalty[s->f_code] + MAX_MV;

  894. 

  895.     get_limits(s, 16*mb_x, 16*mb_y);

  896.     c->skip=0;

  897. 

  898.     /* intra / predictive decision */

  899.     pix = c->src[0][0];

  900.     sum = s->dsp.pix_sum(pix, s->linesize);

  901.     varc = s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)sum*sum)>>8) + 500;

  902. 

  903.     pic->mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;

  904.     pic->mb_var [s->mb_stride * mb_y + mb_x] = (varc+128)>>8;

  905.     c->mb_var_sum_temp += (varc+128)>>8;

  906. 

  907.     switch(s->me_method) {

  908.     case ME_ZERO:

  909.     default:

  910.         mx   = 0;

  911.         my   = 0;

  912.         dmin = 0;

  913.         break;

  914.     case ME_X1:

  915.     case ME_EPZS:

  916.        {

  917.             const int mot_stride = s->b8_stride;

  918.             const int mot_xy = s->block_index[0];

  919. 

  920.             P_LEFT[0] = s->current_picture.motion_val[0][mot_xy - 1][0];

  921.             P_LEFT[1] = s->current_picture.motion_val[0][mot_xy - 1][1];

  922. 

  923.             if(P_LEFT[0]       > (c->xmax<<shift)) P_LEFT[0]       = (c->xmax<<shift);

  924. 

  925.             if(!s->first_slice_line) {

  926.                 P_TOP[0]      = s->current_picture.motion_val[0][mot_xy - mot_stride    ][0];

  927.                 P_TOP[1]      = s->current_picture.motion_val[0][mot_xy - mot_stride    ][1];

  928.                 P_TOPRIGHT[0] = s->current_picture.motion_val[0][mot_xy - mot_stride + 2][0];

  929.                 P_TOPRIGHT[1] = s->current_picture.motion_val[0][mot_xy - mot_stride + 2][1];

  930.                 if(P_TOP[1]      > (c->ymax<<shift)) P_TOP[1]     = (c->ymax<<shift);

  931.                 if(P_TOPRIGHT[0] < (c->xmin<<shift)) P_TOPRIGHT[0]= (c->xmin<<shift);

  932.                 if(P_TOPRIGHT[1] > (c->ymax<<shift)) P_TOPRIGHT[1]= (c->ymax<<shift);

  933. 

  934.                 P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);

  935.                 P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);

  936. 

  937.                 if(s->out_format == FMT_H263){

  938.                     c->pred_x = P_MEDIAN[0];

  939.                     c->pred_y = P_MEDIAN[1];

  940.                 }else { /* mpeg1 at least */

  941.                     c->pred_x= P_LEFT[0];

  942.                     c->pred_y= P_LEFT[1];

  943.                 }

  944.             }else{

  945.                 c->pred_x= P_LEFT[0];

  946.                 c->pred_y= P_LEFT[1];

  947.             }

  948. 

  949.         }

  950.         dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, 0, s->p_mv_table, (1<<16)>>shift, 0, 16);

  951. 

  952.         break;

  953.     }

  954. 

  955.     /* At this point (mx,my) are full-pell and the relative displacement */

  956.     ppix = c->ref[0][0] + (my * s->linesize) + mx;

  957. 

  958.     vard = s->dsp.sse[0](NULL, pix, ppix, s->linesize, 16);

  959. 

  960.     pic->mc_mb_var[s->mb_stride * mb_y + mb_x] = (vard+128)>>8;

  961.     c->mc_mb_var_sum_temp += (vard+128)>>8;

  962. 

  963.     if(mb_type){

  964.         int p_score= FFMIN(vard, varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*100);

  965.         int i_score= varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*20;

  966.         c->scene_change_score+= ff_sqrt(p_score) - ff_sqrt(i_score);

  967. 

  968.         if(mb_type == CANDIDATE_MB_TYPE_INTER){

  969.             c->sub_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);

  970.             set_p_mv_tables(s, mx, my, 1);

  971.         }else{

  972.             mx <<=shift;

  973.             my <<=shift;

  974.         }

  975.         if(mb_type == CANDIDATE_MB_TYPE_INTER4V){

  976.             h263_mv4_search(s, mx, my, shift);

  977. 

  978.             set_p_mv_tables(s, mx, my, 0);

  979.         }

  980.         if(mb_type == CANDIDATE_MB_TYPE_INTER_I){

  981.             interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my, 1);

  982.         }

  983.     }else if(c->avctx->mb_decision > FF_MB_DECISION_SIMPLE){

  984.         int p_score= FFMIN(vard, varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*100);

  985.         int i_score= varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*20;

  986.         c->scene_change_score+= ff_sqrt(p_score) - ff_sqrt(i_score);

  987. 

  988.         if (vard*2 + 200*256 > varc)

  989.             mb_type|= CANDIDATE_MB_TYPE_INTRA;

  990.         if (varc*2 + 200*256 > vard || s->qscale > 24){

  991. //        if (varc*2 + 200*256 + 50*(s->lambda2>>FF_LAMBDA_SHIFT) > vard){

  992.             mb_type|= CANDIDATE_MB_TYPE_INTER;

  993.             c->sub_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);

  994.             if(s->flags&CODEC_FLAG_MV0)

  995.                 if(mx || my)

  996.                     mb_type |= CANDIDATE_MB_TYPE_SKIPPED; //FIXME check difference

  997.         }else{

  998.             mx <<=shift;

  999.             my <<=shift;

  1000.         }

  1001.         if((s->flags&CODEC_FLAG_4MV)

  1002.            && !c->skip && varc>50<<8 && vard>10<<8){

  1003.             if(h263_mv4_search(s, mx, my, shift) < INT_MAX)

  1004.                 mb_type|=CANDIDATE_MB_TYPE_INTER4V;

  1005. 

  1006.             set_p_mv_tables(s, mx, my, 0);

  1007.         }else

  1008.             set_p_mv_tables(s, mx, my, 1);

  1009.         if((s->flags&CODEC_FLAG_INTERLACED_ME)

  1010.            && !c->skip){ //FIXME varc/d checks

  1011.             if(interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my, 0) < INT_MAX)

  1012.                 mb_type |= CANDIDATE_MB_TYPE_INTER_I;

  1013.         }

  1014.     }else{

  1015.         int intra_score, i;

  1016.         mb_type= CANDIDATE_MB_TYPE_INTER;

  1017. 

  1018.         dmin= c->sub_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);

  1019.         if(c->avctx->me_sub_cmp != c->avctx->mb_cmp && !c->skip)

  1020.             dmin= get_mb_score(s, mx, my, 0, 0, 0, 16, 1);

  1021. 

  1022.         if((s->flags&CODEC_FLAG_4MV)

  1023.            && !c->skip && varc>50<<8 && vard>10<<8){

  1024.             int dmin4= h263_mv4_search(s, mx, my, shift);

  1025.             if(dmin4 < dmin){

  1026.                 mb_type= CANDIDATE_MB_TYPE_INTER4V;

  1027.                 dmin=dmin4;

  1028.             }

  1029.         }

  1030.         if((s->flags&CODEC_FLAG_INTERLACED_ME)

  1031.            && !c->skip){ //FIXME varc/d checks

  1032.             int dmin_i= interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my, 0);

  1033.             if(dmin_i < dmin){

  1034.                 mb_type = CANDIDATE_MB_TYPE_INTER_I;

  1035.                 dmin= dmin_i;

  1036.             }

  1037.         }

  1038. 

  1039.         set_p_mv_tables(s, mx, my, mb_type!=CANDIDATE_MB_TYPE_INTER4V);

  1040. 

  1041.         /* get intra luma score */

  1042.         if((c->avctx->mb_cmp&0xFF)==FF_CMP_SSE){

  1043.             intra_score= varc - 500;

  1044.         }else{

  1045.             unsigned mean = (sum+128)>>8;

  1046.             mean*= 0x01010101;

  1047. 

  1048.             for(i=0; i<16; i++){

  1049.                 *(uint32_t*)(&c->scratchpad[i*s->linesize+ 0]) = mean;

  1050.                 *(uint32_t*)(&c->scratchpad[i*s->linesize+ 4]) = mean;

  1051.                 *(uint32_t*)(&c->scratchpad[i*s->linesize+ 8]) = mean;

  1052.                 *(uint32_t*)(&c->scratchpad[i*s->linesize+12]) = mean;

  1053.             }

  1054. 

  1055.             intra_score= s->dsp.mb_cmp[0](s, c->scratchpad, pix, s->linesize, 16);

  1056.         }

  1057.         intra_score += c->mb_penalty_factor*16;

  1058. 

  1059.         if(intra_score < dmin){

  1060.             mb_type= CANDIDATE_MB_TYPE_INTRA;

  1061.             s->current_picture.mb_type[mb_y*s->mb_stride + mb_x] = CANDIDATE_MB_TYPE_INTRA; //FIXME cleanup

  1062.         }else

  1063.             s->current_picture.mb_type[mb_y*s->mb_stride + mb_x] = 0;

  1064. 

  1065.         {

  1066.             int p_score= FFMIN(vard, varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*100);

  1067.             int i_score= varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*20;

  1068.             c->scene_change_score+= ff_sqrt(p_score) - ff_sqrt(i_score);

  1069.         }

  1070.     }

  1071. 

  1072.     s->mb_type[mb_y*s->mb_stride + mb_x]= mb_type;

  1073. }

  1074. 

  1075. int ff_pre_estimate_p_frame_motion(MpegEncContext * s,

  1076.                                     int mb_x, int mb_y)

  1077. {

  1078.     MotionEstContext * const c= &s->me;

  1079.     int mx, my, dmin;

  1080.     int P[10][2];

  1081.     const int shift= 1+s->quarter_sample;

  1082.     const int xy= mb_x + mb_y*s->mb_stride;

  1083.     init_ref(c, s->new_picture.f.data, s->last_picture.f.data, NULL, 16*mb_x, 16*mb_y, 0);

  1084. 

  1085.     av_assert0(s->quarter_sample==0 || s->quarter_sample==1);

  1086. 

  1087.     c->pre_penalty_factor    = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_pre_cmp);

  1088.     c->current_mv_penalty= c->mv_penalty[s->f_code] + MAX_MV;

  1089. 

  1090.     get_limits(s, 16*mb_x, 16*mb_y);

  1091.     c->skip=0;

  1092. 

  1093.     P_LEFT[0]       = s->p_mv_table[xy + 1][0];

  1094.     P_LEFT[1]       = s->p_mv_table[xy + 1][1];

  1095. 

  1096.     if(P_LEFT[0]       < (c->xmin<<shift)) P_LEFT[0]       = (c->xmin<<shift);

  1097. 

  1098.     /* special case for first line */

  1099.     if (s->first_slice_line) {

  1100.         c->pred_x= P_LEFT[0];

  1101.         c->pred_y= P_LEFT[1];

  1102.         P_TOP[0]= P_TOPRIGHT[0]= P_MEDIAN[0]=

  1103.         P_TOP[1]= P_TOPRIGHT[1]= P_MEDIAN[1]= 0; //FIXME

  1104.     } else {

  1105.         P_TOP[0]      = s->p_mv_table[xy + s->mb_stride    ][0];

  1106.         P_TOP[1]      = s->p_mv_table[xy + s->mb_stride    ][1];

  1107.         P_TOPRIGHT[0] = s->p_mv_table[xy + s->mb_stride - 1][0];

  1108.         P_TOPRIGHT[1] = s->p_mv_table[xy + s->mb_stride - 1][1];

  1109.         if(P_TOP[1]      < (c->ymin<<shift)) P_TOP[1]     = (c->ymin<<shift);

  1110.         if(P_TOPRIGHT[0] > (c->xmax<<shift)) P_TOPRIGHT[0]= (c->xmax<<shift);

  1111.         if(P_TOPRIGHT[1] < (c->ymin<<shift)) P_TOPRIGHT[1]= (c->ymin<<shift);

  1112. 

  1113.         P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);

  1114.         P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);

  1115. 

  1116.         c->pred_x = P_MEDIAN[0];

  1117.         c->pred_y = P_MEDIAN[1];

  1118.     }

  1119. 

  1120.     dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, 0, s->p_mv_table, (1<<16)>>shift, 0, 16);

  1121. 

  1122.     s->p_mv_table[xy][0] = mx<<shift;

  1123.     s->p_mv_table[xy][1] = my<<shift;

  1124. 

  1125.     return dmin;

  1126. }

  1127. 

  1128. static int estimate_motion_b(MpegEncContext *s, int mb_x, int mb_y,

  1129.                              int16_t (*mv_table)[2], int ref_index, int f_code)

  1130. {

  1131.     MotionEstContext * const c= &s->me;

  1132.     int mx, my, dmin;

  1133.     int P[10][2];

  1134.     const int shift= 1+s->quarter_sample;

  1135.     const int mot_stride = s->mb_stride;

  1136.     const int mot_xy = mb_y*mot_stride + mb_x;

  1137.     uint8_t * const mv_penalty= c->mv_penalty[f_code] + MAX_MV;

  1138.     int mv_scale;

  1139. 

  1140.     c->penalty_factor    = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_cmp);

  1141.     c->sub_penalty_factor= get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_sub_cmp);

  1142.     c->mb_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->mb_cmp);

  1143.     c->current_mv_penalty= mv_penalty;

  1144. 

  1145.     get_limits(s, 16*mb_x, 16*mb_y);

  1146. 

  1147.     switch(s->me_method) {

  1148.     case ME_ZERO:

  1149.     default:

  1150.         mx   = 0;

  1151.         my   = 0;

  1152.         dmin = 0;

  1153.         break;

  1154.     case ME_X1:

  1155.     case ME_EPZS:

  1156.         P_LEFT[0] = mv_table[mot_xy - 1][0];

  1157.         P_LEFT[1] = mv_table[mot_xy - 1][1];

  1158. 

  1159.         if (P_LEFT[0] > (c->xmax << shift)) P_LEFT[0] = (c->xmax << shift);

  1160. 

  1161.         /* special case for first line */

  1162.         if (!s->first_slice_line) {

  1163.             P_TOP[0]      = mv_table[mot_xy - mot_stride    ][0];

  1164.             P_TOP[1]      = mv_table[mot_xy - mot_stride    ][1];

  1165.             P_TOPRIGHT[0] = mv_table[mot_xy - mot_stride + 1][0];

  1166.             P_TOPRIGHT[1] = mv_table[mot_xy - mot_stride + 1][1];

  1167.             if (P_TOP[1] > (c->ymax << shift)) P_TOP[1] = (c->ymax << shift);

  1168.             if (P_TOPRIGHT[0] < (c->xmin << shift)) P_TOPRIGHT[0] = (c->xmin << shift);

  1169.             if (P_TOPRIGHT[1] > (c->ymax << shift)) P_TOPRIGHT[1] = (c->ymax << shift);

  1170. 

  1171.             P_MEDIAN[0] = mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);

  1172.             P_MEDIAN[1] = mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);

  1173.         }

  1174.         c->pred_x = P_LEFT[0];

  1175.         c->pred_y = P_LEFT[1];

  1176. 

  1177.         if(mv_table == s->b_forw_mv_table){

  1178.             mv_scale= (s->pb_time<<16) / (s->pp_time<<shift);

  1179.         }else{

  1180.             mv_scale= ((s->pb_time - s->pp_time)<<16) / (s->pp_time<<shift);

  1181.         }

  1182. 

  1183.         dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, ref_index, s->p_mv_table, mv_scale, 0, 16);

  1184. 

  1185.         break;

  1186.     }

  1187. 

  1188.     dmin= c->sub_motion_search(s, &mx, &my, dmin, 0, ref_index, 0, 16);

  1189. 

  1190.     if(c->avctx->me_sub_cmp != c->avctx->mb_cmp && !c->skip)

  1191.         dmin= get_mb_score(s, mx, my, 0, ref_index, 0, 16, 1);

  1192. 

  1193. //    s->mb_type[mb_y*s->mb_width + mb_x]= mb_type;

  1194.     mv_table[mot_xy][0]= mx;

  1195.     mv_table[mot_xy][1]= my;

  1196. 

  1197.     return dmin;

  1198. }

  1199. 

  1200. static inline int check_bidir_mv(MpegEncContext * s,

  1201.                    int motion_fx, int motion_fy,

  1202.                    int motion_bx, int motion_by,

  1203.                    int pred_fx, int pred_fy,

  1204.                    int pred_bx, int pred_by,

  1205.                    int size, int h)

  1206. {

  1207.     //FIXME optimize?

  1208.     //FIXME better f_code prediction (max mv & distance)

  1209.     //FIXME pointers

  1210.     MotionEstContext * const c= &s->me;

  1211.     uint8_t * const mv_penalty_f= c->mv_penalty[s->f_code] + MAX_MV; // f_code of the prev frame

  1212.     uint8_t * const mv_penalty_b= c->mv_penalty[s->b_code] + MAX_MV; // f_code of the prev frame

  1213.     int stride= c->stride;

  1214.     uint8_t *dest_y = c->scratchpad;

  1215.     uint8_t *ptr;

  1216.     int dxy;

  1217.     int src_x, src_y;

  1218.     int fbmin;

  1219.     uint8_t **src_data= c->src[0];

  1220.     uint8_t **ref_data= c->ref[0];

  1221.     uint8_t **ref2_data= c->ref[2];

  1222. 

  1223.     if(s->quarter_sample){

  1224.         dxy = ((motion_fy & 3) << 2) | (motion_fx & 3);

  1225.         src_x = motion_fx >> 2;

  1226.         src_y = motion_fy >> 2;

  1227. 

  1228.         ptr = ref_data[0] + (src_y * stride) + src_x;

  1229.         s->dsp.put_qpel_pixels_tab[0][dxy](dest_y    , ptr    , stride);

  1230. 

  1231.         dxy = ((motion_by & 3) << 2) | (motion_bx & 3);

  1232.         src_x = motion_bx >> 2;

  1233.         src_y = motion_by >> 2;

  1234. 

  1235.         ptr = ref2_data[0] + (src_y * stride) + src_x;

  1236.         s->dsp.avg_qpel_pixels_tab[size][dxy](dest_y    , ptr    , stride);

  1237.     }else{

  1238.         dxy = ((motion_fy & 1) << 1) | (motion_fx & 1);

  1239.         src_x = motion_fx >> 1;

  1240.         src_y = motion_fy >> 1;

  1241. 

  1242.         ptr = ref_data[0] + (src_y * stride) + src_x;

  1243.         s->hdsp.put_pixels_tab[size][dxy](dest_y    , ptr    , stride, h);

  1244. 

  1245.         dxy = ((motion_by & 1) << 1) | (motion_bx & 1);

  1246.         src_x = motion_bx >> 1;

  1247.         src_y = motion_by >> 1;

  1248. 

  1249.         ptr = ref2_data[0] + (src_y * stride) + src_x;

  1250.         s->hdsp.avg_pixels_tab[size][dxy](dest_y    , ptr    , stride, h);

  1251.     }

  1252. 

  1253.     fbmin = (mv_penalty_f[motion_fx-pred_fx] + mv_penalty_f[motion_fy-pred_fy])*c->mb_penalty_factor

  1254.            +(mv_penalty_b[motion_bx-pred_bx] + mv_penalty_b[motion_by-pred_by])*c->mb_penalty_factor

  1255.            + s->dsp.mb_cmp[size](s, src_data[0], dest_y, stride, h); //FIXME new_pic

  1256. 

  1257.     if(c->avctx->mb_cmp&FF_CMP_CHROMA){

  1258.     }

  1259.     //FIXME CHROMA !!!

  1260. 

  1261.     return fbmin;

  1262. }

  1263. 

  1264. /* refine the bidir vectors in hq mode and return the score in both lq & hq mode*/

  1265. static inline int bidir_refine(MpegEncContext * s, int mb_x, int mb_y)

  1266. {

  1267.     MotionEstContext * const c= &s->me;

  1268.     const int mot_stride = s->mb_stride;

  1269.     const int xy = mb_y *mot_stride + mb_x;

  1270.     int fbmin;

  1271.     int pred_fx= s->b_bidir_forw_mv_table[xy-1][0];

  1272.     int pred_fy= s->b_bidir_forw_mv_table[xy-1][1];

  1273.     int pred_bx= s->b_bidir_back_mv_table[xy-1][0];

  1274.     int pred_by= s->b_bidir_back_mv_table[xy-1][1];

  1275.     int motion_fx= s->b_bidir_forw_mv_table[xy][0]= s->b_forw_mv_table[xy][0];

  1276.     int motion_fy= s->b_bidir_forw_mv_table[xy][1]= s->b_forw_mv_table[xy][1];

  1277.     int motion_bx= s->b_bidir_back_mv_table[xy][0]= s->b_back_mv_table[xy][0];

  1278.     int motion_by= s->b_bidir_back_mv_table[xy][1]= s->b_back_mv_table[xy][1];

  1279.     const int flags= c->sub_flags;

  1280.     const int qpel= flags&FLAG_QPEL;

  1281.     const int shift= 1+qpel;

  1282.     const int xmin= c->xmin<<shift;

  1283.     const int ymin= c->ymin<<shift;

  1284.     const int xmax= c->xmax<<shift;

  1285.     const int ymax= c->ymax<<shift;

  1286. #define HASH(fx,fy,bx,by) ((fx)+17*(fy)+63*(bx)+117*(by))

  1287. #define HASH8(fx,fy,bx,by) ((uint8_t)HASH(fx,fy,bx,by))

  1288.     int hashidx= HASH(motion_fx,motion_fy, motion_bx, motion_by);

  1289.     uint8_t map[256] = { 0 };

  1290. 

  1291.     map[hashidx&255] = 1;

  1292. 

  1293.     fbmin= check_bidir_mv(s, motion_fx, motion_fy,

  1294.                           motion_bx, motion_by,

  1295.                           pred_fx, pred_fy,

  1296.                           pred_bx, pred_by,

  1297.                           0, 16);

  1298. 

  1299.     if(s->avctx->bidir_refine){

  1300.         int end;

  1301.         static const uint8_t limittab[5]={0,8,32,64,80};

  1302.         const int limit= limittab[s->avctx->bidir_refine];

  1303.         static const int8_t vect[][4]={

  1304. { 0, 0, 0, 1}, { 0, 0, 0,-1}, { 0, 0, 1, 0}, { 0, 0,-1, 0}, { 0, 1, 0, 0}, { 0,-1, 0, 0}, { 1, 0, 0, 0}, {-1, 0, 0, 0},

  1305. 

  1306. { 0, 0, 1, 1}, { 0, 0,-1,-1}, { 0, 1, 1, 0}, { 0,-1,-1, 0}, { 1, 1, 0, 0}, {-1,-1, 0, 0}, { 1, 0, 0, 1}, {-1, 0, 0,-1},

  1307. { 0, 1, 0, 1}, { 0,-1, 0,-1}, { 1, 0, 1, 0}, {-1, 0,-1, 0},

  1308. { 0, 0,-1, 1}, { 0, 0, 1,-1}, { 0,-1, 1, 0}, { 0, 1,-1, 0}, {-1, 1, 0, 0}, { 1,-1, 0, 0}, { 1, 0, 0,-1}, {-1, 0, 0, 1},

  1309. { 0,-1, 0, 1}, { 0, 1, 0,-1}, {-1, 0, 1, 0}, { 1, 0,-1, 0},

  1310. 

  1311. { 0, 1, 1, 1}, { 0,-1,-1,-1}, { 1, 1, 1, 0}, {-1,-1,-1, 0}, { 1, 1, 0, 1}, {-1,-1, 0,-1}, { 1, 0, 1, 1}, {-1, 0,-1,-1},

  1312. { 0,-1, 1, 1}, { 0, 1,-1,-1}, {-1, 1, 1, 0}, { 1,-1,-1, 0}, { 1, 1, 0,-1}, {-1,-1, 0, 1}, { 1, 0,-1, 1}, {-1, 0, 1,-1},

  1313. { 0, 1,-1, 1}, { 0,-1, 1,-1}, { 1,-1, 1, 0}, {-1, 1,-1, 0}, {-1, 1, 0, 1}, { 1,-1, 0,-1}, { 1, 0, 1,-1}, {-1, 0,-1, 1},

  1314. { 0, 1, 1,-1}, { 0,-1,-1, 1}, { 1, 1,-1, 0}, {-1,-1, 1, 0}, { 1,-1, 0, 1}, {-1, 1, 0,-1}, {-1, 0, 1, 1}, { 1, 0,-1,-1},

  1315. 

  1316. { 1, 1, 1, 1}, {-1,-1,-1,-1},

  1317. { 1, 1, 1,-1}, {-1,-1,-1, 1}, { 1, 1,-1, 1}, {-1,-1, 1,-1}, { 1,-1, 1, 1}, {-1, 1,-1,-1}, {-1, 1, 1, 1}, { 1,-1,-1,-1},

  1318. { 1, 1,-1,-1}, {-1,-1, 1, 1}, { 1,-1,-1, 1}, {-1, 1, 1,-1}, { 1,-1, 1,-1}, {-1, 1,-1, 1},

  1319.         };

  1320.         static const uint8_t hash[]={

  1321. HASH8( 0, 0, 0, 1), HASH8( 0, 0, 0,-1), HASH8( 0, 0, 1, 0), HASH8( 0, 0,-1, 0), HASH8( 0, 1, 0, 0), HASH8( 0,-1, 0, 0), HASH8( 1, 0, 0, 0), HASH8(-1, 0, 0, 0),

  1322. 

  1323. HASH8( 0, 0, 1, 1), HASH8( 0, 0,-1,-1), HASH8( 0, 1, 1, 0), HASH8( 0,-1,-1, 0), HASH8( 1, 1, 0, 0), HASH8(-1,-1, 0, 0), HASH8( 1, 0, 0, 1), HASH8(-1, 0, 0,-1),

  1324. HASH8( 0, 1, 0, 1), HASH8( 0,-1, 0,-1), HASH8( 1, 0, 1, 0), HASH8(-1, 0,-1, 0),

  1325. HASH8( 0, 0,-1, 1), HASH8( 0, 0, 1,-1), HASH8( 0,-1, 1, 0), HASH8( 0, 1,-1, 0), HASH8(-1, 1, 0, 0), HASH8( 1,-1, 0, 0), HASH8( 1, 0, 0,-1), HASH8(-1, 0, 0, 1),

  1326. HASH8( 0,-1, 0, 1), HASH8( 0, 1, 0,-1), HASH8(-1, 0, 1, 0), HASH8( 1, 0,-1, 0),

  1327. 

  1328. HASH8( 0, 1, 1, 1), HASH8( 0,-1,-1,-1), HASH8( 1, 1, 1, 0), HASH8(-1,-1,-1, 0), HASH8( 1, 1, 0, 1), HASH8(-1,-1, 0,-1), HASH8( 1, 0, 1, 1), HASH8(-1, 0,-1,-1),

  1329. HASH8( 0,-1, 1, 1), HASH8( 0, 1,-1,-1), HASH8(-1, 1, 1, 0), HASH8( 1,-1,-1, 0), HASH8( 1, 1, 0,-1), HASH8(-1,-1, 0, 1), HASH8( 1, 0,-1, 1), HASH8(-1, 0, 1,-1),

  1330. HASH8( 0, 1,-1, 1), HASH8( 0,-1, 1,-1), HASH8( 1,-1, 1, 0), HASH8(-1, 1,-1, 0), HASH8(-1, 1, 0, 1), HASH8( 1,-1, 0,-1), HASH8( 1, 0, 1,-1), HASH8(-1, 0,-1, 1),

  1331. HASH8( 0, 1, 1,-1), HASH8( 0,-1,-1, 1), HASH8( 1, 1,-1, 0), HASH8(-1,-1, 1, 0), HASH8( 1,-1, 0, 1), HASH8(-1, 1, 0,-1), HASH8(-1, 0, 1, 1), HASH8( 1, 0,-1,-1),

  1332. 

  1333. HASH8( 1, 1, 1, 1), HASH8(-1,-1,-1,-1),

  1334. HASH8( 1, 1, 1,-1), HASH8(-1,-1,-1, 1), HASH8( 1, 1,-1, 1), HASH8(-1,-1, 1,-1), HASH8( 1,-1, 1, 1), HASH8(-1, 1,-1,-1), HASH8(-1, 1, 1, 1), HASH8( 1,-1,-1,-1),

  1335. HASH8( 1, 1,-1,-1), HASH8(-1,-1, 1, 1), HASH8( 1,-1,-1, 1), HASH8(-1, 1, 1,-1), HASH8( 1,-1, 1,-1), HASH8(-1, 1,-1, 1),

  1336. };

  1337. 

  1338. #define CHECK_BIDIR(fx,fy,bx,by)\

  1339.     if( !map[(hashidx+HASH(fx,fy,bx,by))&255]\

  1340.        &&(fx<=0 || motion_fx+fx<=xmax) && (fy<=0 || motion_fy+fy<=ymax) && (bx<=0 || motion_bx+bx<=xmax) && (by<=0 || motion_by+by<=ymax)\

  1341.        &&(fx>=0 || motion_fx+fx>=xmin) && (fy>=0 || motion_fy+fy>=ymin) && (bx>=0 || motion_bx+bx>=xmin) && (by>=0 || motion_by+by>=ymin)){\

  1342.         int score;\

  1343.         map[(hashidx+HASH(fx,fy,bx,by))&255] = 1;\

  1344.         score= check_bidir_mv(s, motion_fx+fx, motion_fy+fy, motion_bx+bx, motion_by+by, pred_fx, pred_fy, pred_bx, pred_by, 0, 16);\

  1345.         if(score < fbmin){\

  1346.             hashidx += HASH(fx,fy,bx,by);\

  1347.             fbmin= score;\

  1348.             motion_fx+=fx;\

  1349.             motion_fy+=fy;\

  1350.             motion_bx+=bx;\

  1351.             motion_by+=by;\

  1352.             end=0;\

  1353.         }\

  1354.     }

  1355. #define CHECK_BIDIR2(a,b,c,d)\

  1356. CHECK_BIDIR(a,b,c,d)\

  1357. CHECK_BIDIR(-(a),-(b),-(c),-(d))

  1358. 

  1359.         do{

  1360.             int i;

  1361.             int borderdist=0;

  1362.             end=1;

  1363. 

  1364.             CHECK_BIDIR2(0,0,0,1)

  1365.             CHECK_BIDIR2(0,0,1,0)

  1366.             CHECK_BIDIR2(0,1,0,0)

  1367.             CHECK_BIDIR2(1,0,0,0)

  1368. 

  1369.             for(i=8; i<limit; i++){

  1370.                 int fx= motion_fx+vect[i][0];

  1371.                 int fy= motion_fy+vect[i][1];

  1372.                 int bx= motion_bx+vect[i][2];

  1373.                 int by= motion_by+vect[i][3];

  1374.                 if(borderdist<=0){

  1375.                     int a= (xmax - FFMAX(fx,bx))|(FFMIN(fx,bx) - xmin);

  1376.                     int b= (ymax - FFMAX(fy,by))|(FFMIN(fy,by) - ymin);

  1377.                     if((a|b) < 0)

  1378.                         map[(hashidx+hash[i])&255] = 1;

  1379.                 }

  1380.                 if(!map[(hashidx+hash[i])&255]){

  1381.                     int score;

  1382.                     map[(hashidx+hash[i])&255] = 1;

  1383.                     score= check_bidir_mv(s, fx, fy, bx, by, pred_fx, pred_fy, pred_bx, pred_by, 0, 16);

  1384.                     if(score < fbmin){

  1385.                         hashidx += hash[i];

  1386.                         fbmin= score;

  1387.                         motion_fx=fx;

  1388.                         motion_fy=fy;

  1389.                         motion_bx=bx;

  1390.                         motion_by=by;

  1391.                         end=0;

  1392.                         borderdist--;

  1393.                         if(borderdist<=0){

  1394.                             int a= FFMIN(xmax - FFMAX(fx,bx), FFMIN(fx,bx) - xmin);

  1395.                             int b= FFMIN(ymax - FFMAX(fy,by), FFMIN(fy,by) - ymin);

  1396.                             borderdist= FFMIN(a,b);

  1397.                         }

  1398.                     }

  1399.                 }

  1400.             }

  1401.         }while(!end);

  1402.     }

  1403. 

  1404.     s->b_bidir_forw_mv_table[xy][0]= motion_fx;

  1405.     s->b_bidir_forw_mv_table[xy][1]= motion_fy;

  1406.     s->b_bidir_back_mv_table[xy][0]= motion_bx;

  1407.     s->b_bidir_back_mv_table[xy][1]= motion_by;

  1408. 

  1409.     return fbmin;

  1410. }

  1411. 

  1412. static inline int direct_search(MpegEncContext * s, int mb_x, int mb_y)

  1413. {

  1414.     MotionEstContext * const c= &s->me;

  1415.     int P[10][2];

  1416.     const int mot_stride = s->mb_stride;

  1417.     const int mot_xy = mb_y*mot_stride + mb_x;

  1418.     const int shift= 1+s->quarter_sample;

  1419.     int dmin, i;

  1420.     const int time_pp= s->pp_time;

  1421.     const int time_pb= s->pb_time;

  1422.     int mx, my, xmin, xmax, ymin, ymax;

  1423.     int16_t (*mv_table)[2]= s->b_direct_mv_table;

  1424. 

  1425.     c->current_mv_penalty= c->mv_penalty[1] + MAX_MV;

  1426.     ymin= xmin=(-32)>>shift;

  1427.     ymax= xmax=   31>>shift;

  1428. 

  1429.     if (IS_8X8(s->next_picture.mb_type[mot_xy])) {

  1430.         s->mv_type= MV_TYPE_8X8;

  1431.     }else{

  1432.         s->mv_type= MV_TYPE_16X16;

  1433.     }

  1434. 

  1435.     for(i=0; i<4; i++){

  1436.         int index= s->block_index[i];

  1437.         int min, max;

  1438. 

  1439.         c->co_located_mv[i][0] = s->next_picture.motion_val[0][index][0];

  1440.         c->co_located_mv[i][1] = s->next_picture.motion_val[0][index][1];

  1441.         c->direct_basis_mv[i][0]= c->co_located_mv[i][0]*time_pb/time_pp + ((i& 1)<<(shift+3));

  1442.         c->direct_basis_mv[i][1]= c->co_located_mv[i][1]*time_pb/time_pp + ((i>>1)<<(shift+3));

  1443. //        c->direct_basis_mv[1][i][0]= c->co_located_mv[i][0]*(time_pb - time_pp)/time_pp + ((i &1)<<(shift+3);

  1444. //        c->direct_basis_mv[1][i][1]= c->co_located_mv[i][1]*(time_pb - time_pp)/time_pp + ((i>>1)<<(shift+3);

  1445. 

  1446.         max= FFMAX(c->direct_basis_mv[i][0], c->direct_basis_mv[i][0] - c->co_located_mv[i][0])>>shift;

  1447.         min= FFMIN(c->direct_basis_mv[i][0], c->direct_basis_mv[i][0] - c->co_located_mv[i][0])>>shift;

  1448.         max+= 16*mb_x + 1; // +-1 is for the simpler rounding

  1449.         min+= 16*mb_x - 1;

  1450.         xmax= FFMIN(xmax, s->width - max);

  1451.         xmin= FFMAX(xmin, - 16     - min);

  1452. 

  1453.         max= FFMAX(c->direct_basis_mv[i][1], c->direct_basis_mv[i][1] - c->co_located_mv[i][1])>>shift;

  1454.         min= FFMIN(c->direct_basis_mv[i][1], c->direct_basis_mv[i][1] - c->co_located_mv[i][1])>>shift;

  1455.         max+= 16*mb_y + 1; // +-1 is for the simpler rounding

  1456.         min+= 16*mb_y - 1;

  1457.         ymax= FFMIN(ymax, s->height - max);

  1458.         ymin= FFMAX(ymin, - 16      - min);

  1459. 

  1460.         if(s->mv_type == MV_TYPE_16X16) break;

  1461.     }

  1462. 

  1463.     av_assert2(xmax <= 15 && ymax <= 15 && xmin >= -16 && ymin >= -16);

  1464. 

  1465.     if(xmax < 0 || xmin >0 || ymax < 0 || ymin > 0){

  1466.         s->b_direct_mv_table[mot_xy][0]= 0;

  1467.         s->b_direct_mv_table[mot_xy][1]= 0;

  1468. 

  1469.         return 256*256*256*64;

  1470.     }

  1471. 

  1472.     c->xmin= xmin;

  1473.     c->ymin= ymin;

  1474.     c->xmax= xmax;

  1475.     c->ymax= ymax;

  1476.     c->flags     |= FLAG_DIRECT;

  1477.     c->sub_flags |= FLAG_DIRECT;

  1478.     c->pred_x=0;

  1479.     c->pred_y=0;

  1480. 

  1481.     P_LEFT[0]        = av_clip(mv_table[mot_xy - 1][0], xmin<<shift, xmax<<shift);

  1482.     P_LEFT[1]        = av_clip(mv_table[mot_xy - 1][1], ymin<<shift, ymax<<shift);

  1483. 

  1484.     /* special case for first line */

  1485.     if (!s->first_slice_line) { //FIXME maybe allow this over thread boundary as it is clipped

  1486.         P_TOP[0]      = av_clip(mv_table[mot_xy - mot_stride             ][0], xmin<<shift, xmax<<shift);

  1487.         P_TOP[1]      = av_clip(mv_table[mot_xy - mot_stride             ][1], ymin<<shift, ymax<<shift);

  1488.         P_TOPRIGHT[0] = av_clip(mv_table[mot_xy - mot_stride + 1         ][0], xmin<<shift, xmax<<shift);

  1489.         P_TOPRIGHT[1] = av_clip(mv_table[mot_xy - mot_stride + 1         ][1], ymin<<shift, ymax<<shift);

  1490. 

  1491.         P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);

  1492.         P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);

  1493.     }

  1494. 

  1495.     dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, 0, mv_table, 1<<(16-shift), 0, 16);

  1496.     if(c->sub_flags&FLAG_QPEL)

  1497.         dmin = qpel_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);

  1498.     else

  1499.         dmin = hpel_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);

  1500. 

  1501.     if(c->avctx->me_sub_cmp != c->avctx->mb_cmp && !c->skip)

  1502.         dmin= get_mb_score(s, mx, my, 0, 0, 0, 16, 1);

  1503. 

  1504.     get_limits(s, 16*mb_x, 16*mb_y); //restore c->?min/max, maybe not needed

  1505. 

  1506.     mv_table[mot_xy][0]= mx;

  1507.     mv_table[mot_xy][1]= my;

  1508.     c->flags     &= ~FLAG_DIRECT;

  1509.     c->sub_flags &= ~FLAG_DIRECT;

  1510. 

  1511.     return dmin;

  1512. }

  1513. 

  1514. void ff_estimate_b_frame_motion(MpegEncContext * s,

  1515.                              int mb_x, int mb_y)

  1516. {

  1517.     MotionEstContext * const c= &s->me;

  1518.     const int penalty_factor= c->mb_penalty_factor;

  1519.     int fmin, bmin, dmin, fbmin, bimin, fimin;

  1520.     int type=0;

  1521.     const int xy = mb_y*s->mb_stride + mb_x;

  1522.     init_ref(c, s->new_picture.f.data, s->last_picture.f.data,

  1523.              s->next_picture.f.data, 16 * mb_x, 16 * mb_y, 2);

  1524. 

  1525.     get_limits(s, 16*mb_x, 16*mb_y);

  1526. 

  1527.     c->skip=0;

  1528. 

  1529.     if (s->codec_id == AV_CODEC_ID_MPEG4 && s->next_picture.mbskip_table[xy]) {

  1530.         int score= direct_search(s, mb_x, mb_y); //FIXME just check 0,0

  1531. 

  1532.         score= ((unsigned)(score*score + 128*256))>>16;

  1533.         c->mc_mb_var_sum_temp += score;

  1534.         s->current_picture.mc_mb_var[mb_y*s->mb_stride + mb_x] = score; //FIXME use SSE

  1535.         s->mb_type[mb_y*s->mb_stride + mb_x]= CANDIDATE_MB_TYPE_DIRECT0;

  1536. 

  1537.         return;

  1538.     }

  1539. 

  1540.     if (s->codec_id == AV_CODEC_ID_MPEG4)

  1541.         dmin= direct_search(s, mb_x, mb_y);

  1542.     else

  1543.         dmin= INT_MAX;

  1544. //FIXME penalty stuff for non mpeg4

  1545.     c->skip=0;

  1546.     fmin = estimate_motion_b(s, mb_x, mb_y, s->b_forw_mv_table, 0, s->f_code) +

  1547.            3 * penalty_factor;

  1548. 

  1549.     c->skip=0;

  1550.     bmin = estimate_motion_b(s, mb_x, mb_y, s->b_back_mv_table, 2, s->b_code) +

  1551.            2 * penalty_factor;

  1552.     av_dlog(s, " %d %d ", s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1]);

  1553. 

  1554.     c->skip=0;

  1555.     fbmin= bidir_refine(s, mb_x, mb_y) + penalty_factor;

  1556.     av_dlog(s, "%d %d %d %d\n", dmin, fmin, bmin, fbmin);

  1557. 

  1558.     if(s->flags & CODEC_FLAG_INTERLACED_ME){

  1559. //FIXME mb type penalty

  1560.         c->skip=0;

  1561.         c->current_mv_penalty= c->mv_penalty[s->f_code] + MAX_MV;

  1562.         fimin= interlaced_search(s, 0,

  1563.                                  s->b_field_mv_table[0], s->b_field_select_table[0],

  1564.                                  s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1], 0);

  1565.         c->current_mv_penalty= c->mv_penalty[s->b_code] + MAX_MV;

  1566.         bimin= interlaced_search(s, 2,

  1567.                                  s->b_field_mv_table[1], s->b_field_select_table[1],

  1568.                                  s->b_back_mv_table[xy][0], s->b_back_mv_table[xy][1], 0);

  1569.     }else

  1570.         fimin= bimin= INT_MAX;

  1571. 

  1572.     {

  1573.         int score= fmin;

  1574.         type = CANDIDATE_MB_TYPE_FORWARD;

  1575. 

  1576.         if (dmin <= score){

  1577.             score = dmin;

  1578.             type = CANDIDATE_MB_TYPE_DIRECT;

  1579.         }

  1580.         if(bmin<score){

  1581.             score=bmin;

  1582.             type= CANDIDATE_MB_TYPE_BACKWARD;

  1583.         }

  1584.         if(fbmin<score){

  1585.             score=fbmin;

  1586.             type= CANDIDATE_MB_TYPE_BIDIR;

  1587.         }

  1588.         if(fimin<score){

  1589.             score=fimin;

  1590.             type= CANDIDATE_MB_TYPE_FORWARD_I;

  1591.         }

  1592.         if(bimin<score){

  1593.             score=bimin;

  1594.             type= CANDIDATE_MB_TYPE_BACKWARD_I;

  1595.         }

  1596. 

  1597.         score= ((unsigned)(score*score + 128*256))>>16;

  1598.         c->mc_mb_var_sum_temp += score;

  1599.         s->current_picture.mc_mb_var[mb_y*s->mb_stride + mb_x] = score; //FIXME use SSE

  1600.     }

  1601. 

  1602.     if(c->avctx->mb_decision > FF_MB_DECISION_SIMPLE){

  1603.         type= CANDIDATE_MB_TYPE_FORWARD | CANDIDATE_MB_TYPE_BACKWARD | CANDIDATE_MB_TYPE_BIDIR | CANDIDATE_MB_TYPE_DIRECT;

  1604.         if(fimin < INT_MAX)

  1605.             type |= CANDIDATE_MB_TYPE_FORWARD_I;

  1606.         if(bimin < INT_MAX)

  1607.             type |= CANDIDATE_MB_TYPE_BACKWARD_I;

  1608.         if(fimin < INT_MAX && bimin < INT_MAX){

  1609.             type |= CANDIDATE_MB_TYPE_BIDIR_I;

  1610.         }

  1611.          //FIXME something smarter

  1612.         if(dmin>256*256*16) type&= ~CANDIDATE_MB_TYPE_DIRECT; //do not try direct mode if it is invalid for this MB

  1613.         if(s->codec_id == AV_CODEC_ID_MPEG4 && type&CANDIDATE_MB_TYPE_DIRECT && s->flags&CODEC_FLAG_MV0 && *(uint32_t*)s->b_direct_mv_table[xy])

  1614.             type |= CANDIDATE_MB_TYPE_DIRECT0;

  1615.     }

  1616. 

  1617.     s->mb_type[mb_y*s->mb_stride + mb_x]= type;

  1618. }

  1619. 

  1620. /* find best f_code for ME which do unlimited searches */

  1621. int ff_get_best_fcode(MpegEncContext * s, int16_t (*mv_table)[2], int type)

  1622. {

  1623.     if(s->me_method>=ME_EPZS){

  1624.         int score[8];

  1625.         int i, y, range= s->avctx->me_range ? s->avctx->me_range : (INT_MAX/2);

  1626.         uint8_t * fcode_tab= s->fcode_tab;

  1627.         int best_fcode=-1;

  1628.         int best_score=-10000000;

  1629. 

  1630.         if(s->msmpeg4_version)

  1631.             range= FFMIN(range, 16);

  1632.         else if(s->codec_id == AV_CODEC_ID_MPEG2VIDEO && s->avctx->strict_std_compliance >= FF_COMPLIANCE_NORMAL)

  1633.             range= FFMIN(range, 256);

  1634. 

  1635.         for(i=0; i<8; i++) score[i]= s->mb_num*(8-i);

  1636. 

  1637.         for(y=0; y<s->mb_height; y++){

  1638.             int x;

  1639.             int xy= y*s->mb_stride;

  1640.             for(x=0; x<s->mb_width; x++){

  1641.                 if(s->mb_type[xy] & type){

  1642.                     int mx= mv_table[xy][0];

  1643.                     int my= mv_table[xy][1];

  1644.                     int fcode= FFMAX(fcode_tab[mx + MAX_MV],

  1645.                                      fcode_tab[my + MAX_MV]);

  1646.                     int j;

  1647. 

  1648.                         if(mx >= range || mx < -range ||

  1649.                            my >= range || my < -range)

  1650.                             continue;

  1651. 

  1652.                     for(j=0; j<fcode && j<8; j++){

  1653.                         if(s->pict_type==AV_PICTURE_TYPE_B || s->current_picture.mc_mb_var[xy] < s->current_picture.mb_var[xy])

  1654.                             score[j]-= 170;

  1655.                     }

  1656.                 }

  1657.                 xy++;

  1658.             }

  1659.         }

  1660. 

  1661.         for(i=1; i<8; i++){

  1662.             if(score[i] > best_score){

  1663.                 best_score= score[i];

  1664.                 best_fcode= i;

  1665.             }

  1666.         }

  1667. 

  1668.         return best_fcode;

  1669.     }else{

  1670.         return 1;

  1671.     }

  1672. }

  1673. 

  1674. void ff_fix_long_p_mvs(MpegEncContext * s)

  1675. {

  1676.     MotionEstContext * const c= &s->me;

  1677.     const int f_code= s->f_code;

  1678.     int y, range;

  1679.     av_assert0(s->pict_type==AV_PICTURE_TYPE_P);

  1680. 

  1681.     range = (((s->out_format == FMT_MPEG1 || s->msmpeg4_version) ? 8 : 16) << f_code);

  1682. 

  1683.     av_assert0(range <= 16 || !s->msmpeg4_version);

  1684.     av_assert0(range <=256 || !(s->codec_id == AV_CODEC_ID_MPEG2VIDEO && s->avctx->strict_std_compliance >= FF_COMPLIANCE_NORMAL));

  1685. 

  1686.     if(c->avctx->me_range && range > c->avctx->me_range) range= c->avctx->me_range;

  1687. 

  1688.     if(s->flags&CODEC_FLAG_4MV){

  1689.         const int wrap= s->b8_stride;

  1690. 

  1691.         /* clip / convert to intra 8x8 type MVs */

  1692.         for(y=0; y<s->mb_height; y++){

  1693.             int xy= y*2*wrap;

  1694.             int i= y*s->mb_stride;

  1695.             int x;

  1696. 

  1697.             for(x=0; x<s->mb_width; x++){

  1698.                 if(s->mb_type[i]&CANDIDATE_MB_TYPE_INTER4V){

  1699.                     int block;

  1700.                     for(block=0; block<4; block++){

  1701.                         int off= (block& 1) + (block>>1)*wrap;

  1702.                         int mx = s->current_picture.motion_val[0][ xy + off ][0];

  1703.                         int my = s->current_picture.motion_val[0][ xy + off ][1];

  1704. 

  1705.                         if(   mx >=range || mx <-range

  1706.                            || my >=range || my <-range){

  1707.                             s->mb_type[i] &= ~CANDIDATE_MB_TYPE_INTER4V;

  1708.                             s->mb_type[i] |= CANDIDATE_MB_TYPE_INTRA;

  1709.                             s->current_picture.mb_type[i] = CANDIDATE_MB_TYPE_INTRA;

  1710.                         }

  1711.                     }

  1712.                 }

  1713.                 xy+=2;

  1714.                 i++;

  1715.             }

  1716.         }

  1717.     }

  1718. }

  1719. 

  1720. /**

  1721.  *

  1722.  * @param truncate 1 for truncation, 0 for using intra

  1723.  */

  1724. void ff_fix_long_mvs(MpegEncContext * s, uint8_t *field_select_table, int field_select,

  1725.                      int16_t (*mv_table)[2], int f_code, int type, int truncate)

  1726. {

  1727.     MotionEstContext * const c= &s->me;

  1728.     int y, h_range, v_range;

  1729. 

  1730.     // RAL: 8 in MPEG-1, 16 in MPEG-4

  1731.     int range = (((s->out_format == FMT_MPEG1 || s->msmpeg4_version) ? 8 : 16) << f_code);

  1732. 

  1733.     if(c->avctx->me_range && range > c->avctx->me_range) range= c->avctx->me_range;

  1734. 

  1735.     h_range= range;

  1736.     v_range= field_select_table ? range>>1 : range;

  1737. 

  1738.     /* clip / convert to intra 16x16 type MVs */

  1739.     for(y=0; y<s->mb_height; y++){

  1740.         int x;

  1741.         int xy= y*s->mb_stride;

  1742.         for(x=0; x<s->mb_width; x++){

  1743.             if (s->mb_type[xy] & type){    // RAL: "type" test added...

  1744.                 if(field_select_table==NULL || field_select_table[xy] == field_select){

  1745.                     if(   mv_table[xy][0] >=h_range || mv_table[xy][0] <-h_range

  1746.                        || mv_table[xy][1] >=v_range || mv_table[xy][1] <-v_range){

  1747. 

  1748.                         if(truncate){

  1749.                             if     (mv_table[xy][0] > h_range-1) mv_table[xy][0]=  h_range-1;

  1750.                             else if(mv_table[xy][0] < -h_range ) mv_table[xy][0]= -h_range;

  1751.                             if     (mv_table[xy][1] > v_range-1) mv_table[xy][1]=  v_range-1;

  1752.                             else if(mv_table[xy][1] < -v_range ) mv_table[xy][1]= -v_range;

  1753.                         }else{

  1754.                             s->mb_type[xy] &= ~type;

  1755.                             s->mb_type[xy] |= CANDIDATE_MB_TYPE_INTRA;

  1756.                             mv_table[xy][0]=

  1757.                             mv_table[xy][1]= 0;

  1758.                         }

  1759.                     }

  1760.                 }

  1761.             }

  1762.             xy++;

  1763.         }

  1764.     }

  1765. }