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.
40228 Commits
32 Branches
411MB
Tree: c0de9159a7
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'c0de9159a7'
${ noResults }
FFmpeg/libavcodec/motion_est.c

1753 lines
67KB
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 Libav.

  9.  *

  10.  * Libav 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.  * Libav 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 Libav; 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.         assert(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.                     assert((fx>>1) + 16*s->mb_x >= -16);

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

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

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

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

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

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

  165.                     assert((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) < -ME_MAP_SIZE){

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

  305.         return -1;

  306.     }

  307.     if (s->me_method != ME_ZERO &&

  308.         s->me_method != ME_EPZS &&

  309.         s->me_method != ME_X1) {

  310.         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");

  311.         return -1;

  312.     }

  313. 

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

  315. 

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

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

  318.     }

  319. 

  320.     ff_set_cmp(&s->mecc, s->mecc.me_pre_cmp, c->avctx->me_pre_cmp);

  321.     ff_set_cmp(&s->mecc, s->mecc.me_cmp,     c->avctx->me_cmp);

  322.     ff_set_cmp(&s->mecc, s->mecc.me_sub_cmp, c->avctx->me_sub_cmp);

  323.     ff_set_cmp(&s->mecc, s->mecc.mb_cmp,     c->avctx->mb_cmp);

  324. 

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

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

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

  328. 

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

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

  331.         c->sub_motion_search= qpel_motion_search;

  332.         c->qpel_avg = s->qdsp.avg_qpel_pixels_tab;

  333.         if (s->no_rounding)

  334.             c->qpel_put = s->qdsp.put_no_rnd_qpel_pixels_tab;

  335.         else

  336.             c->qpel_put = s->qdsp.put_qpel_pixels_tab;

  337.     }else{

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

  339.             c->sub_motion_search= hpel_motion_search;

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

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

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

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

  344.         else

  345.             c->sub_motion_search= hpel_motion_search;

  346.     }

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

  348.     if (s->no_rounding)

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

  350.     else

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

  352. 

  353.     if(s->linesize){

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

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

  356.     }else{

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

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

  359.     }

  360. 

  361.     /* 8x8 fullpel search would need a 4x4 chroma compare, which we do

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

  363.      * does not expect it. */

  364.     if ((c->avctx->me_cmp & FF_CMP_CHROMA) /* && !s->mecc.me_cmp[2] */)

  365.         s->mecc.me_cmp[2] = zero_cmp;

  366.     if ((c->avctx->me_sub_cmp & FF_CMP_CHROMA) && !s->mecc.me_sub_cmp[2])

  367.         s->mecc.me_sub_cmp[2] = zero_cmp;

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

  372.         c->sub_motion_search= no_sub_motion_search;

  373.     }

  374. 

  375.     return 0;

  376. }

  377. 

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

  379. {\

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

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

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

  383. }

  384. 

  385. static int sad_hpel_motion_search(MpegEncContext * s,

  386.                                   int *mx_ptr, int *my_ptr, int dmin,

  387.                                   int src_index, int ref_index,

  388.                                   int size, int h)

  389. {

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

  391.     const int penalty_factor= c->sub_penalty_factor;

  392.     int mx, my, dminh;

  393.     uint8_t *pix, *ptr;

  394.     int stride= c->stride;

  395.     const int flags= c->sub_flags;

  396.     LOAD_COMMON

  397. 

  398.     assert(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. /*

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

  524.     else                   c->range= 16;

  525. */

  526.     if (s->unrestricted_mv) {

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

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

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

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

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

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

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

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

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

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

  537.     } else {

  538.         c->xmin = - x;

  539.         c->ymin = - y;

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

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

  542.     }

  543.     if(range){

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

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

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

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

  548.     }

  549. }

  550. 

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

  552.     const int stride= c->stride;

  553. 

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

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

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

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

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

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

  560. }

  561. 

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

  563. {

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

  565.     const int size= 1;

  566.     const int h=8;

  567.     int block;

  568.     int P[10][2];

  569.     int dmin_sum=0, mx4_sum=0, my4_sum=0;

  570.     int same=1;

  571.     const int stride= c->stride;

  572.     uint8_t *mv_penalty= c->current_mv_penalty;

  573. 

  574.     init_mv4_ref(c);

  575. 

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

  577.         int mx4, my4;

  578.         int pred_x4, pred_y4;

  579.         int dmin4;

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

  581.         const int mot_stride = s->b8_stride;

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

  583. 

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

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

  586. 

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

  588. 

  589.         /* special case for first line */

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

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

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

  593.         } else {

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

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

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

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

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

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

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

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

  602. 

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

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

  605. 

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

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

  608.         }

  609.         P_MV1[0]= mx;

  610.         P_MV1[1]= my;

  611. 

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

  613. 

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

  615. 

  616.         if (s->mecc.me_sub_cmp[0] != s->mecc.mb_cmp[0]) {

  617.             int dxy;

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

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

  620.             if(s->quarter_sample){

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

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

  623. 

  624.                 if(s->no_rounding)

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

  626.                 else

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

  628.             }else{

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

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

  631. 

  632.                 if(s->no_rounding)

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

  634.                 else

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

  636.             }

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

  638.         }else

  639.             dmin_sum+= dmin4;

  640. 

  641.         if(s->quarter_sample){

  642.             mx4_sum+= mx4/2;

  643.             my4_sum+= my4/2;

  644.         }else{

  645.             mx4_sum+= mx4;

  646.             my4_sum+= my4;

  647.         }

  648. 

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

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

  651. 

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

  653.     }

  654. 

  655.     if(same)

  656.         return INT_MAX;

  657. 

  658.     if (s->mecc.me_sub_cmp[0] != s->mecc.mb_cmp[0]) {

  659.         dmin_sum += s->mecc.mb_cmp[0](s,

  660.                                       s->new_picture.f->data[0] +

  661.                                       s->mb_x * 16 + s->mb_y * 16 * stride,

  662.                                       c->scratchpad, stride, 16);

  663.     }

  664. 

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

  666.         int dxy;

  667.         int mx, my;

  668.         int offset;

  669. 

  670.         mx= ff_h263_round_chroma(mx4_sum);

  671.         my= ff_h263_round_chroma(my4_sum);

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

  673. 

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

  675. 

  676.         if(s->no_rounding){

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

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

  679.         }else{

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

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

  682.         }

  683. 

  684.         dmin_sum += s->mecc.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);

  685.         dmin_sum += s->mecc.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);

  686.     }

  687. 

  688.     c->pred_x= mx;

  689.     c->pred_y= my;

  690. 

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

  692.     /*case FF_CMP_SSE:

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

  694.     case FF_CMP_RD:

  695.         return dmin_sum;

  696.     default:

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

  698.     }

  699. }

  700. 

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

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

  703. 

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

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

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

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

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

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

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

  711.     }

  712. }

  713. 

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

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

  716. {

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

  718.     const int size=0;

  719.     const int h=8;

  720.     int block;

  721.     int P[10][2];

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

  723.     int same=1;

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

  725.     int dmin_sum= 0;

  726.     const int mot_stride= s->mb_stride;

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

  728. 

  729.     c->ymin>>=1;

  730.     c->ymax>>=1;

  731.     c->stride<<=1;

  732.     c->uvstride<<=1;

  733.     init_interlaced_ref(s, ref_index);

  734. 

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

  736.         int field_select;

  737.         int best_dmin= INT_MAX;

  738.         int best_field= -1;

  739. 

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

  741.             int dmin, mx_i, my_i;

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

  743. 

  744.             if(user_field_select){

  745.                 assert(field_select==0 || field_select==1);

  746.                 assert(field_select_tables[block][xy]==0 || field_select_tables[block][xy]==1);

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

  748.                     continue;

  749.             }

  750. 

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

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

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

  754. 

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

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

  757. 

  758.             if(!s->first_slice_line){

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

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

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

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

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

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

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

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

  767. 

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

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

  770.             }

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

  772.             P_MV1[1]= my / 2;

  773. 

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

  775. 

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

  777. 

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

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

  780. 

  781.             if (s->mecc.me_sub_cmp[0] != s->mecc.mb_cmp[0]) {

  782.                 int dxy;

  783. 

  784.                 //FIXME chroma ME

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

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

  787. 

  788.                 if(s->no_rounding){

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

  790.                 }else{

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

  792.                 }

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

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

  795.             }else

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

  797. 

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

  799. 

  800.             if(dmin < best_dmin){

  801.                 best_dmin= dmin;

  802.                 best_field= field_select;

  803.             }

  804.         }

  805.         {

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

  807. 

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

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

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

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

  812.         }

  813. 

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

  815.         dmin_sum += best_dmin;

  816.     }

  817. 

  818.     c->ymin<<=1;

  819.     c->ymax<<=1;

  820.     c->stride>>=1;

  821.     c->uvstride>>=1;

  822. 

  823.     if(same)

  824.         return INT_MAX;

  825. 

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

  827.     /*case FF_CMP_SSE:

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

  829.     case FF_CMP_RD:

  830.         return dmin_sum;

  831.     default:

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

  833.     }

  834. }

  835. 

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

  837.     switch(type&0xFF){

  838.     default:

  839.     case FF_CMP_SAD:

  840.         return lambda>>FF_LAMBDA_SHIFT;

  841.     case FF_CMP_DCT:

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

  843.     case FF_CMP_SATD:

  844.     case FF_CMP_DCT264:

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

  846.     case FF_CMP_RD:

  847.     case FF_CMP_PSNR:

  848.     case FF_CMP_SSE:

  849.     case FF_CMP_NSSE:

  850.         return lambda2>>FF_LAMBDA_SHIFT;

  851.     case FF_CMP_BIT:

  852.         return 1;

  853.     }

  854. }

  855. 

  856. void ff_estimate_p_frame_motion(MpegEncContext * s,

  857.                                 int mb_x, int mb_y)

  858. {

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

  860.     uint8_t *pix, *ppix;

  861.     int sum, mx, my, dmin;

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

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

  864.     int P[10][2];

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

  866.     int mb_type=0;

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

  868. 

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

  870. 

  871.     assert(s->quarter_sample==0 || s->quarter_sample==1);

  872.     assert(s->linesize == c->stride);

  873.     assert(s->uvlinesize == c->uvstride);

  874. 

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

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

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

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

  879. 

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

  881.     c->skip=0;

  882. 

  883.     /* intra / predictive decision */

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

  885.     sum  = s->mpvencdsp.pix_sum(pix, s->linesize);

  886.     varc = s->mpvencdsp.pix_norm1(pix, s->linesize) -

  887.            (((unsigned) sum * sum) >> 8) + 500;

  888. 

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

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

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

  892. 

  893.     switch(s->me_method) {

  894.     case ME_ZERO:

  895.     default:

  896.         mx   = 0;

  897.         my   = 0;

  898.         dmin = 0;

  899.         break;

  900.     case ME_X1:

  901.     case ME_EPZS:

  902.        {

  903.             const int mot_stride = s->b8_stride;

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

  905. 

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

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

  908. 

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

  910. 

  911.             if(!s->first_slice_line) {

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

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

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

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

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

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

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

  919. 

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

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

  922. 

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

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

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

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

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

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

  929.                 }

  930.             }else{

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

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

  933.             }

  934. 

  935.         }

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

  937. 

  938.         break;

  939.     }

  940. 

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

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

  943. 

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

  945. 

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

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

  948. 

  949.     if(mb_type){

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

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

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

  953. 

  954.         if(mb_type == CANDIDATE_MB_TYPE_INTER){

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

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

  957.         }else{

  958.             mx <<=shift;

  959.             my <<=shift;

  960.         }

  961.         if(mb_type == CANDIDATE_MB_TYPE_INTER4V){

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

  963. 

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

  965.         }

  966.         if(mb_type == CANDIDATE_MB_TYPE_INTER_I){

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

  968.         }

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

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

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

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

  973. 

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

  975.             mb_type|= CANDIDATE_MB_TYPE_INTRA;

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

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

  978.             mb_type|= CANDIDATE_MB_TYPE_INTER;

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

  980.             if (s->mpv_flags & FF_MPV_FLAG_MV0)

  981.                 if(mx || my)

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

  983.         }else{

  984.             mx <<=shift;

  985.             my <<=shift;

  986.         }

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

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

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

  990.                 mb_type|=CANDIDATE_MB_TYPE_INTER4V;

  991. 

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

  993.         }else

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

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

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

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

  998.                 mb_type |= CANDIDATE_MB_TYPE_INTER_I;

  999.         }

  1000.     }else{

  1001.         int intra_score, i;

  1002.         mb_type= CANDIDATE_MB_TYPE_INTER;

  1003. 

  1004.         dmin= c->sub_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);

  1005.         if(c->avctx->me_sub_cmp != c->avctx->mb_cmp && !c->skip)

  1006.             dmin= get_mb_score(s, mx, my, 0, 0, 0, 16, 1);

  1007. 

  1008.         if((s->flags&CODEC_FLAG_4MV)

  1009.            && !c->skip && varc>50<<8 && vard>10<<8){

  1010.             int dmin4= h263_mv4_search(s, mx, my, shift);

  1011.             if(dmin4 < dmin){

  1012.                 mb_type= CANDIDATE_MB_TYPE_INTER4V;

  1013.                 dmin=dmin4;

  1014.             }

  1015.         }

  1016.         if((s->flags&CODEC_FLAG_INTERLACED_ME)

  1017.            && !c->skip){ //FIXME varc/d checks

  1018.             int dmin_i= interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my, 0);

  1019.             if(dmin_i < dmin){

  1020.                 mb_type = CANDIDATE_MB_TYPE_INTER_I;

  1021.                 dmin= dmin_i;

  1022.             }

  1023.         }

  1024. 

  1025.         set_p_mv_tables(s, mx, my, mb_type!=CANDIDATE_MB_TYPE_INTER4V);

  1026. 

  1027.         /* get intra luma score */

  1028.         if((c->avctx->mb_cmp&0xFF)==FF_CMP_SSE){

  1029.             intra_score= varc - 500;

  1030.         }else{

  1031.             unsigned mean = (sum+128)>>8;

  1032.             mean*= 0x01010101;

  1033. 

  1034.             for(i=0; i<16; i++){

  1035.                 *(uint32_t*)(&c->scratchpad[i*s->linesize+ 0]) = mean;

  1036.                 *(uint32_t*)(&c->scratchpad[i*s->linesize+ 4]) = mean;

  1037.                 *(uint32_t*)(&c->scratchpad[i*s->linesize+ 8]) = mean;

  1038.                 *(uint32_t*)(&c->scratchpad[i*s->linesize+12]) = mean;

  1039.             }

  1040. 

  1041.             intra_score= s->mecc.mb_cmp[0](s, c->scratchpad, pix, s->linesize, 16);

  1042.         }

  1043.         intra_score += c->mb_penalty_factor*16;

  1044. 

  1045.         if(intra_score < dmin){

  1046.             mb_type= CANDIDATE_MB_TYPE_INTRA;

  1047.             s->current_picture.mb_type[mb_y*s->mb_stride + mb_x] = CANDIDATE_MB_TYPE_INTRA; //FIXME cleanup

  1048.         }else

  1049.             s->current_picture.mb_type[mb_y*s->mb_stride + mb_x] = 0;

  1050. 

  1051.         {

  1052.             int p_score= FFMIN(vard, varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*100);

  1053.             int i_score= varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*20;

  1054.             c->scene_change_score+= ff_sqrt(p_score) - ff_sqrt(i_score);

  1055.         }

  1056.     }

  1057. 

  1058.     s->mb_type[mb_y*s->mb_stride + mb_x]= mb_type;

  1059. }

  1060. 

  1061. int ff_pre_estimate_p_frame_motion(MpegEncContext * s,

  1062.                                     int mb_x, int mb_y)

  1063. {

  1064.     MotionEstContext * const c= &s->me;

  1065.     int mx, my, dmin;

  1066.     int P[10][2];

  1067.     const int shift= 1+s->quarter_sample;

  1068.     const int xy= mb_x + mb_y*s->mb_stride;

  1069.     init_ref(c, s->new_picture.f->data, s->last_picture.f->data, NULL, 16*mb_x, 16*mb_y, 0);

  1070. 

  1071.     assert(s->quarter_sample==0 || s->quarter_sample==1);

  1072. 

  1073.     c->pre_penalty_factor    = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_pre_cmp);

  1074.     c->current_mv_penalty= c->mv_penalty[s->f_code] + MAX_MV;

  1075. 

  1076.     get_limits(s, 16*mb_x, 16*mb_y);

  1077.     c->skip=0;

  1078. 

  1079.     P_LEFT[0]       = s->p_mv_table[xy + 1][0];

  1080.     P_LEFT[1]       = s->p_mv_table[xy + 1][1];

  1081. 

  1082.     if(P_LEFT[0]       < (c->xmin<<shift)) P_LEFT[0]       = (c->xmin<<shift);

  1083. 

  1084.     /* special case for first line */

  1085.     if (s->first_slice_line) {

  1086.         c->pred_x= P_LEFT[0];

  1087.         c->pred_y= P_LEFT[1];

  1088.         P_TOP[0]= P_TOPRIGHT[0]= P_MEDIAN[0]=

  1089.         P_TOP[1]= P_TOPRIGHT[1]= P_MEDIAN[1]= 0; //FIXME

  1090.     } else {

  1091.         P_TOP[0]      = s->p_mv_table[xy + s->mb_stride    ][0];

  1092.         P_TOP[1]      = s->p_mv_table[xy + s->mb_stride    ][1];

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

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

  1095.         if(P_TOP[1]      < (c->ymin<<shift)) P_TOP[1]     = (c->ymin<<shift);

  1096.         if(P_TOPRIGHT[0] > (c->xmax<<shift)) P_TOPRIGHT[0]= (c->xmax<<shift);

  1097.         if(P_TOPRIGHT[1] < (c->ymin<<shift)) P_TOPRIGHT[1]= (c->ymin<<shift);

  1098. 

  1099.         P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);

  1100.         P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);

  1101. 

  1102.         c->pred_x = P_MEDIAN[0];

  1103.         c->pred_y = P_MEDIAN[1];

  1104.     }

  1105. 

  1106.     dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, 0, s->p_mv_table, (1<<16)>>shift, 0, 16);

  1107. 

  1108.     s->p_mv_table[xy][0] = mx<<shift;

  1109.     s->p_mv_table[xy][1] = my<<shift;

  1110. 

  1111.     return dmin;

  1112. }

  1113. 

  1114. static int estimate_motion_b(MpegEncContext *s, int mb_x, int mb_y,

  1115.                              int16_t (*mv_table)[2], int ref_index, int f_code)

  1116. {

  1117.     MotionEstContext * const c= &s->me;

  1118.     int mx, my, dmin;

  1119.     int P[10][2];

  1120.     const int shift= 1+s->quarter_sample;

  1121.     const int mot_stride = s->mb_stride;

  1122.     const int mot_xy = mb_y*mot_stride + mb_x;

  1123.     uint8_t * const mv_penalty= c->mv_penalty[f_code] + MAX_MV;

  1124.     int mv_scale;

  1125. 

  1126.     c->penalty_factor    = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_cmp);

  1127.     c->sub_penalty_factor= get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_sub_cmp);

  1128.     c->mb_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->mb_cmp);

  1129.     c->current_mv_penalty= mv_penalty;

  1130. 

  1131.     get_limits(s, 16*mb_x, 16*mb_y);

  1132. 

  1133.     switch(s->me_method) {

  1134.     case ME_ZERO:

  1135.     default:

  1136.         mx   = 0;

  1137.         my   = 0;

  1138.         dmin = 0;

  1139.         break;

  1140.     case ME_X1:

  1141.     case ME_EPZS:

  1142.         P_LEFT[0] = mv_table[mot_xy - 1][0];

  1143.         P_LEFT[1] = mv_table[mot_xy - 1][1];

  1144. 

  1145.         if (P_LEFT[0] > (c->xmax << shift)) P_LEFT[0] = (c->xmax << shift);

  1146. 

  1147.         /* special case for first line */

  1148.         if (!s->first_slice_line) {

  1149.             P_TOP[0]      = mv_table[mot_xy - mot_stride    ][0];

  1150.             P_TOP[1]      = mv_table[mot_xy - mot_stride    ][1];

  1151.             P_TOPRIGHT[0] = mv_table[mot_xy - mot_stride + 1][0];

  1152.             P_TOPRIGHT[1] = mv_table[mot_xy - mot_stride + 1][1];

  1153.             if (P_TOP[1] > (c->ymax << shift)) P_TOP[1] = (c->ymax << shift);

  1154.             if (P_TOPRIGHT[0] < (c->xmin << shift)) P_TOPRIGHT[0] = (c->xmin << shift);

  1155.             if (P_TOPRIGHT[1] > (c->ymax << shift)) P_TOPRIGHT[1] = (c->ymax << shift);

  1156. 

  1157.             P_MEDIAN[0] = mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);

  1158.             P_MEDIAN[1] = mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);

  1159.         }

  1160.         c->pred_x = P_LEFT[0];

  1161.         c->pred_y = P_LEFT[1];

  1162. 

  1163.         if(mv_table == s->b_forw_mv_table){

  1164.             mv_scale= (s->pb_time<<16) / (s->pp_time<<shift);

  1165.         }else{

  1166.             mv_scale= ((s->pb_time - s->pp_time)<<16) / (s->pp_time<<shift);

  1167.         }

  1168. 

  1169.         dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, ref_index, s->p_mv_table, mv_scale, 0, 16);

  1170. 

  1171.         break;

  1172.     }

  1173. 

  1174.     dmin= c->sub_motion_search(s, &mx, &my, dmin, 0, ref_index, 0, 16);

  1175. 

  1176.     if(c->avctx->me_sub_cmp != c->avctx->mb_cmp && !c->skip)

  1177.         dmin= get_mb_score(s, mx, my, 0, ref_index, 0, 16, 1);

  1178. 

  1179. //    s->mb_type[mb_y*s->mb_width + mb_x]= mb_type;

  1180.     mv_table[mot_xy][0]= mx;

  1181.     mv_table[mot_xy][1]= my;

  1182. 

  1183.     return dmin;

  1184. }

  1185. 

  1186. static inline int check_bidir_mv(MpegEncContext * s,

  1187.                    int motion_fx, int motion_fy,

  1188.                    int motion_bx, int motion_by,

  1189.                    int pred_fx, int pred_fy,

  1190.                    int pred_bx, int pred_by,

  1191.                    int size, int h)

  1192. {

  1193.     //FIXME optimize?

  1194.     //FIXME better f_code prediction (max mv & distance)

  1195.     //FIXME pointers

  1196.     MotionEstContext * const c= &s->me;

  1197.     uint8_t * const mv_penalty_f= c->mv_penalty[s->f_code] + MAX_MV; // f_code of the prev frame

  1198.     uint8_t * const mv_penalty_b= c->mv_penalty[s->b_code] + MAX_MV; // f_code of the prev frame

  1199.     int stride= c->stride;

  1200.     uint8_t *dest_y = c->scratchpad;

  1201.     uint8_t *ptr;

  1202.     int dxy;

  1203.     int src_x, src_y;

  1204.     int fbmin;

  1205.     uint8_t **src_data= c->src[0];

  1206.     uint8_t **ref_data= c->ref[0];

  1207.     uint8_t **ref2_data= c->ref[2];

  1208. 

  1209.     if(s->quarter_sample){

  1210.         dxy = ((motion_fy & 3) << 2) | (motion_fx & 3);

  1211.         src_x = motion_fx >> 2;

  1212.         src_y = motion_fy >> 2;

  1213. 

  1214.         ptr = ref_data[0] + (src_y * stride) + src_x;

  1215.         s->qdsp.put_qpel_pixels_tab[0][dxy](dest_y, ptr, stride);

  1216. 

  1217.         dxy = ((motion_by & 3) << 2) | (motion_bx & 3);

  1218.         src_x = motion_bx >> 2;

  1219.         src_y = motion_by >> 2;

  1220. 

  1221.         ptr = ref2_data[0] + (src_y * stride) + src_x;

  1222.         s->qdsp.avg_qpel_pixels_tab[size][dxy](dest_y, ptr, stride);

  1223.     }else{

  1224.         dxy = ((motion_fy & 1) << 1) | (motion_fx & 1);

  1225.         src_x = motion_fx >> 1;

  1226.         src_y = motion_fy >> 1;

  1227. 

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

  1229.         s->hdsp.put_pixels_tab[size][dxy](dest_y    , ptr    , stride, h);

  1230. 

  1231.         dxy = ((motion_by & 1) << 1) | (motion_bx & 1);

  1232.         src_x = motion_bx >> 1;

  1233.         src_y = motion_by >> 1;

  1234. 

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

  1236.         s->hdsp.avg_pixels_tab[size][dxy](dest_y    , ptr    , stride, h);

  1237.     }

  1238. 

  1239.     fbmin = (mv_penalty_f[motion_fx-pred_fx] + mv_penalty_f[motion_fy-pred_fy])*c->mb_penalty_factor

  1240.            +(mv_penalty_b[motion_bx-pred_bx] + mv_penalty_b[motion_by-pred_by])*c->mb_penalty_factor

  1241.            + s->mecc.mb_cmp[size](s, src_data[0], dest_y, stride, h); // FIXME new_pic

  1242. 

  1243.     if(c->avctx->mb_cmp&FF_CMP_CHROMA){

  1244.     }

  1245.     //FIXME CHROMA !!!

  1246. 

  1247.     return fbmin;

  1248. }

  1249. 

  1250. /* refine the bidir vectors in hq mode and return the score in both lq & hq mode*/

  1251. static inline int bidir_refine(MpegEncContext * s, int mb_x, int mb_y)

  1252. {

  1253.     MotionEstContext * const c= &s->me;

  1254.     const int mot_stride = s->mb_stride;

  1255.     const int xy = mb_y *mot_stride + mb_x;

  1256.     int fbmin;

  1257.     int pred_fx= s->b_bidir_forw_mv_table[xy-1][0];

  1258.     int pred_fy= s->b_bidir_forw_mv_table[xy-1][1];

  1259.     int pred_bx= s->b_bidir_back_mv_table[xy-1][0];

  1260.     int pred_by= s->b_bidir_back_mv_table[xy-1][1];

  1261.     int motion_fx= s->b_bidir_forw_mv_table[xy][0]= s->b_forw_mv_table[xy][0];

  1262.     int motion_fy= s->b_bidir_forw_mv_table[xy][1]= s->b_forw_mv_table[xy][1];

  1263.     int motion_bx= s->b_bidir_back_mv_table[xy][0]= s->b_back_mv_table[xy][0];

  1264.     int motion_by= s->b_bidir_back_mv_table[xy][1]= s->b_back_mv_table[xy][1];

  1265.     const int flags= c->sub_flags;

  1266.     const int qpel= flags&FLAG_QPEL;

  1267.     const int shift= 1+qpel;

  1268.     const int xmin= c->xmin<<shift;

  1269.     const int ymin= c->ymin<<shift;

  1270.     const int xmax= c->xmax<<shift;

  1271.     const int ymax= c->ymax<<shift;

  1272. #define HASH(fx,fy,bx,by) ((fx)+17*(fy)+63*(bx)+117*(by))

  1273. #define HASH8(fx,fy,bx,by) ((uint8_t)HASH(fx,fy,bx,by))

  1274.     int hashidx= HASH(motion_fx,motion_fy, motion_bx, motion_by);

  1275.     uint8_t map[256] = { 0 };

  1276. 

  1277.     map[hashidx&255] = 1;

  1278. 

  1279.     fbmin= check_bidir_mv(s, motion_fx, motion_fy,

  1280.                           motion_bx, motion_by,

  1281.                           pred_fx, pred_fy,

  1282.                           pred_bx, pred_by,

  1283.                           0, 16);

  1284. 

  1285.     if(s->avctx->bidir_refine){

  1286.         int end;

  1287.         static const uint8_t limittab[5]={0,8,32,64,80};

  1288.         const int limit= limittab[s->avctx->bidir_refine];

  1289.         static const int8_t vect[][4]={

  1290. { 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},

  1291. 

  1292. { 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},

  1293. { 0, 1, 0, 1}, { 0,-1, 0,-1}, { 1, 0, 1, 0}, {-1, 0,-1, 0},

  1294. { 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},

  1295. { 0,-1, 0, 1}, { 0, 1, 0,-1}, {-1, 0, 1, 0}, { 1, 0,-1, 0},

  1296. 

  1297. { 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},

  1298. { 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},

  1299. { 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},

  1300. { 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},

  1301. 

  1302. { 1, 1, 1, 1}, {-1,-1,-1,-1},

  1303. { 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},

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

  1305.         };

  1306.         static const uint8_t hash[]={

  1307. 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),

  1308. 

  1309. 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),

  1310. HASH8( 0, 1, 0, 1), HASH8( 0,-1, 0,-1), HASH8( 1, 0, 1, 0), HASH8(-1, 0,-1, 0),

  1311. 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),

  1312. HASH8( 0,-1, 0, 1), HASH8( 0, 1, 0,-1), HASH8(-1, 0, 1, 0), HASH8( 1, 0,-1, 0),

  1313. 

  1314. 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),

  1315. 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),

  1316. 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),

  1317. 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),

  1318. 

  1319. HASH8( 1, 1, 1, 1), HASH8(-1,-1,-1,-1),

  1320. 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),

  1321. 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),

  1322. };

  1323. 

  1324. #define CHECK_BIDIR(fx,fy,bx,by)\

  1325.     if( !map[(hashidx+HASH(fx,fy,bx,by))&255]\

  1326.        &&(fx<=0 || motion_fx+fx<=xmax) && (fy<=0 || motion_fy+fy<=ymax) && (bx<=0 || motion_bx+bx<=xmax) && (by<=0 || motion_by+by<=ymax)\

  1327.        &&(fx>=0 || motion_fx+fx>=xmin) && (fy>=0 || motion_fy+fy>=ymin) && (bx>=0 || motion_bx+bx>=xmin) && (by>=0 || motion_by+by>=ymin)){\

  1328.         int score;\

  1329.         map[(hashidx+HASH(fx,fy,bx,by))&255] = 1;\

  1330.         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);\

  1331.         if(score < fbmin){\

  1332.             hashidx += HASH(fx,fy,bx,by);\

  1333.             fbmin= score;\

  1334.             motion_fx+=fx;\

  1335.             motion_fy+=fy;\

  1336.             motion_bx+=bx;\

  1337.             motion_by+=by;\

  1338.             end=0;\

  1339.         }\

  1340.     }

  1341. #define CHECK_BIDIR2(a,b,c,d)\

  1342. CHECK_BIDIR(a,b,c,d)\

  1343. CHECK_BIDIR(-(a),-(b),-(c),-(d))

  1344. 

  1345.         do{

  1346.             int i;

  1347.             int borderdist=0;

  1348.             end=1;

  1349. 

  1350.             CHECK_BIDIR2(0,0,0,1)

  1351.             CHECK_BIDIR2(0,0,1,0)

  1352.             CHECK_BIDIR2(0,1,0,0)

  1353.             CHECK_BIDIR2(1,0,0,0)

  1354. 

  1355.             for(i=8; i<limit; i++){

  1356.                 int fx= motion_fx+vect[i][0];

  1357.                 int fy= motion_fy+vect[i][1];

  1358.                 int bx= motion_bx+vect[i][2];

  1359.                 int by= motion_by+vect[i][3];

  1360.                 if(borderdist<=0){

  1361.                     int a= (xmax - FFMAX(fx,bx))|(FFMIN(fx,bx) - xmin);

  1362.                     int b= (ymax - FFMAX(fy,by))|(FFMIN(fy,by) - ymin);

  1363.                     if((a|b) < 0)

  1364.                         map[(hashidx+hash[i])&255] = 1;

  1365.                 }

  1366.                 if(!map[(hashidx+hash[i])&255]){

  1367.                     int score;

  1368.                     map[(hashidx+hash[i])&255] = 1;

  1369.                     score= check_bidir_mv(s, fx, fy, bx, by, pred_fx, pred_fy, pred_bx, pred_by, 0, 16);

  1370.                     if(score < fbmin){

  1371.                         hashidx += hash[i];

  1372.                         fbmin= score;

  1373.                         motion_fx=fx;

  1374.                         motion_fy=fy;

  1375.                         motion_bx=bx;

  1376.                         motion_by=by;

  1377.                         end=0;

  1378.                         borderdist--;

  1379.                         if(borderdist<=0){

  1380.                             int a= FFMIN(xmax - FFMAX(fx,bx), FFMIN(fx,bx) - xmin);

  1381.                             int b= FFMIN(ymax - FFMAX(fy,by), FFMIN(fy,by) - ymin);

  1382.                             borderdist= FFMIN(a,b);

  1383.                         }

  1384.                     }

  1385.                 }

  1386.             }

  1387.         }while(!end);

  1388.     }

  1389. 

  1390.     s->b_bidir_forw_mv_table[xy][0]= motion_fx;

  1391.     s->b_bidir_forw_mv_table[xy][1]= motion_fy;

  1392.     s->b_bidir_back_mv_table[xy][0]= motion_bx;

  1393.     s->b_bidir_back_mv_table[xy][1]= motion_by;

  1394. 

  1395.     return fbmin;

  1396. }

  1397. 

  1398. static inline int direct_search(MpegEncContext * s, int mb_x, int mb_y)

  1399. {

  1400.     MotionEstContext * const c= &s->me;

  1401.     int P[10][2];

  1402.     const int mot_stride = s->mb_stride;

  1403.     const int mot_xy = mb_y*mot_stride + mb_x;

  1404.     const int shift= 1+s->quarter_sample;

  1405.     int dmin, i;

  1406.     const int time_pp= s->pp_time;

  1407.     const int time_pb= s->pb_time;

  1408.     int mx, my, xmin, xmax, ymin, ymax;

  1409.     int16_t (*mv_table)[2]= s->b_direct_mv_table;

  1410. 

  1411.     c->current_mv_penalty= c->mv_penalty[1] + MAX_MV;

  1412.     ymin= xmin=(-32)>>shift;

  1413.     ymax= xmax=   31>>shift;

  1414. 

  1415.     if (IS_8X8(s->next_picture.mb_type[mot_xy])) {

  1416.         s->mv_type= MV_TYPE_8X8;

  1417.     }else{

  1418.         s->mv_type= MV_TYPE_16X16;

  1419.     }

  1420. 

  1421.     for(i=0; i<4; i++){

  1422.         int index= s->block_index[i];

  1423.         int min, max;

  1424. 

  1425.         c->co_located_mv[i][0] = s->next_picture.motion_val[0][index][0];

  1426.         c->co_located_mv[i][1] = s->next_picture.motion_val[0][index][1];

  1427.         c->direct_basis_mv[i][0]= c->co_located_mv[i][0]*time_pb/time_pp + ((i& 1)<<(shift+3));

  1428.         c->direct_basis_mv[i][1]= c->co_located_mv[i][1]*time_pb/time_pp + ((i>>1)<<(shift+3));

  1429. //        c->direct_basis_mv[1][i][0]= c->co_located_mv[i][0]*(time_pb - time_pp)/time_pp + ((i &1)<<(shift+3);

  1430. //        c->direct_basis_mv[1][i][1]= c->co_located_mv[i][1]*(time_pb - time_pp)/time_pp + ((i>>1)<<(shift+3);

  1431. 

  1432.         max= FFMAX(c->direct_basis_mv[i][0], c->direct_basis_mv[i][0] - c->co_located_mv[i][0])>>shift;

  1433.         min= FFMIN(c->direct_basis_mv[i][0], c->direct_basis_mv[i][0] - c->co_located_mv[i][0])>>shift;

  1434.         max+= 16*mb_x + 1; // +-1 is for the simpler rounding

  1435.         min+= 16*mb_x - 1;

  1436.         xmax= FFMIN(xmax, s->width - max);

  1437.         xmin= FFMAX(xmin, - 16     - min);

  1438. 

  1439.         max= FFMAX(c->direct_basis_mv[i][1], c->direct_basis_mv[i][1] - c->co_located_mv[i][1])>>shift;

  1440.         min= FFMIN(c->direct_basis_mv[i][1], c->direct_basis_mv[i][1] - c->co_located_mv[i][1])>>shift;

  1441.         max+= 16*mb_y + 1; // +-1 is for the simpler rounding

  1442.         min+= 16*mb_y - 1;

  1443.         ymax= FFMIN(ymax, s->height - max);

  1444.         ymin= FFMAX(ymin, - 16      - min);

  1445. 

  1446.         if(s->mv_type == MV_TYPE_16X16) break;

  1447.     }

  1448. 

  1449.     assert(xmax <= 15 && ymax <= 15 && xmin >= -16 && ymin >= -16);

  1450. 

  1451.     if(xmax < 0 || xmin >0 || ymax < 0 || ymin > 0){

  1452.         s->b_direct_mv_table[mot_xy][0]= 0;

  1453.         s->b_direct_mv_table[mot_xy][1]= 0;

  1454. 

  1455.         return 256*256*256*64;

  1456.     }

  1457. 

  1458.     c->xmin= xmin;

  1459.     c->ymin= ymin;

  1460.     c->xmax= xmax;

  1461.     c->ymax= ymax;

  1462.     c->flags     |= FLAG_DIRECT;

  1463.     c->sub_flags |= FLAG_DIRECT;

  1464.     c->pred_x=0;

  1465.     c->pred_y=0;

  1466. 

  1467.     P_LEFT[0]        = av_clip(mv_table[mot_xy - 1][0], xmin<<shift, xmax<<shift);

  1468.     P_LEFT[1]        = av_clip(mv_table[mot_xy - 1][1], ymin<<shift, ymax<<shift);

  1469. 

  1470.     /* special case for first line */

  1471.     if (!s->first_slice_line) { //FIXME maybe allow this over thread boundary as it is clipped

  1472.         P_TOP[0]      = av_clip(mv_table[mot_xy - mot_stride             ][0], xmin<<shift, xmax<<shift);

  1473.         P_TOP[1]      = av_clip(mv_table[mot_xy - mot_stride             ][1], ymin<<shift, ymax<<shift);

  1474.         P_TOPRIGHT[0] = av_clip(mv_table[mot_xy - mot_stride + 1         ][0], xmin<<shift, xmax<<shift);

  1475.         P_TOPRIGHT[1] = av_clip(mv_table[mot_xy - mot_stride + 1         ][1], ymin<<shift, ymax<<shift);

  1476. 

  1477.         P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);

  1478.         P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);

  1479.     }

  1480. 

  1481.     dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, 0, mv_table, 1<<(16-shift), 0, 16);

  1482.     if(c->sub_flags&FLAG_QPEL)

  1483.         dmin = qpel_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);

  1484.     else

  1485.         dmin = hpel_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);

  1486. 

  1487.     if(c->avctx->me_sub_cmp != c->avctx->mb_cmp && !c->skip)

  1488.         dmin= get_mb_score(s, mx, my, 0, 0, 0, 16, 1);

  1489. 

  1490.     get_limits(s, 16*mb_x, 16*mb_y); //restore c->?min/max, maybe not needed

  1491. 

  1492.     mv_table[mot_xy][0]= mx;

  1493.     mv_table[mot_xy][1]= my;

  1494.     c->flags     &= ~FLAG_DIRECT;

  1495.     c->sub_flags &= ~FLAG_DIRECT;

  1496. 

  1497.     return dmin;

  1498. }

  1499. 

  1500. void ff_estimate_b_frame_motion(MpegEncContext * s,

  1501.                              int mb_x, int mb_y)

  1502. {

  1503.     MotionEstContext * const c= &s->me;

  1504.     const int penalty_factor= c->mb_penalty_factor;

  1505.     int fmin, bmin, dmin, fbmin, bimin, fimin;

  1506.     int type=0;

  1507.     const int xy = mb_y*s->mb_stride + mb_x;

  1508.     init_ref(c, s->new_picture.f->data, s->last_picture.f->data,

  1509.              s->next_picture.f->data, 16 * mb_x, 16 * mb_y, 2);

  1510. 

  1511.     get_limits(s, 16*mb_x, 16*mb_y);

  1512. 

  1513.     c->skip=0;

  1514. 

  1515.     if (s->codec_id == AV_CODEC_ID_MPEG4 && s->next_picture.mbskip_table[xy]) {

  1516.         int score= direct_search(s, mb_x, mb_y); //FIXME just check 0,0

  1517. 

  1518.         score= ((unsigned)(score*score + 128*256))>>16;

  1519.         c->mc_mb_var_sum_temp += score;

  1520.         s->current_picture.mc_mb_var[mb_y*s->mb_stride + mb_x] = score; //FIXME use SSE

  1521.         s->mb_type[mb_y*s->mb_stride + mb_x]= CANDIDATE_MB_TYPE_DIRECT0;

  1522. 

  1523.         return;

  1524.     }

  1525. 

  1526.     if (s->codec_id == AV_CODEC_ID_MPEG4)

  1527.         dmin= direct_search(s, mb_x, mb_y);

  1528.     else

  1529.         dmin= INT_MAX;

  1530. //FIXME penalty stuff for non mpeg4

  1531.     c->skip=0;

  1532.     fmin = estimate_motion_b(s, mb_x, mb_y, s->b_forw_mv_table, 0, s->f_code) +

  1533.            3 * penalty_factor;

  1534. 

  1535.     c->skip=0;

  1536.     bmin = estimate_motion_b(s, mb_x, mb_y, s->b_back_mv_table, 2, s->b_code) +

  1537.            2 * penalty_factor;

  1538.     av_dlog(s, " %d %d ", s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1]);

  1539. 

  1540.     c->skip=0;

  1541.     fbmin= bidir_refine(s, mb_x, mb_y) + penalty_factor;

  1542.     av_dlog(s, "%d %d %d %d\n", dmin, fmin, bmin, fbmin);

  1543. 

  1544.     if(s->flags & CODEC_FLAG_INTERLACED_ME){

  1545. //FIXME mb type penalty

  1546.         c->skip=0;

  1547.         c->current_mv_penalty= c->mv_penalty[s->f_code] + MAX_MV;

  1548.         fimin= interlaced_search(s, 0,

  1549.                                  s->b_field_mv_table[0], s->b_field_select_table[0],

  1550.                                  s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1], 0);

  1551.         c->current_mv_penalty= c->mv_penalty[s->b_code] + MAX_MV;

  1552.         bimin= interlaced_search(s, 2,

  1553.                                  s->b_field_mv_table[1], s->b_field_select_table[1],

  1554.                                  s->b_back_mv_table[xy][0], s->b_back_mv_table[xy][1], 0);

  1555.     }else

  1556.         fimin= bimin= INT_MAX;

  1557. 

  1558.     {

  1559.         int score= fmin;

  1560.         type = CANDIDATE_MB_TYPE_FORWARD;

  1561. 

  1562.         if (dmin <= score){

  1563.             score = dmin;

  1564.             type = CANDIDATE_MB_TYPE_DIRECT;

  1565.         }

  1566.         if(bmin<score){

  1567.             score=bmin;

  1568.             type= CANDIDATE_MB_TYPE_BACKWARD;

  1569.         }

  1570.         if(fbmin<score){

  1571.             score=fbmin;

  1572.             type= CANDIDATE_MB_TYPE_BIDIR;

  1573.         }

  1574.         if(fimin<score){

  1575.             score=fimin;

  1576.             type= CANDIDATE_MB_TYPE_FORWARD_I;

  1577.         }

  1578.         if(bimin<score){

  1579.             score=bimin;

  1580.             type= CANDIDATE_MB_TYPE_BACKWARD_I;

  1581.         }

  1582. 

  1583.         score= ((unsigned)(score*score + 128*256))>>16;

  1584.         c->mc_mb_var_sum_temp += score;

  1585.         s->current_picture.mc_mb_var[mb_y*s->mb_stride + mb_x] = score; //FIXME use SSE

  1586.     }

  1587. 

  1588.     if(c->avctx->mb_decision > FF_MB_DECISION_SIMPLE){

  1589.         type= CANDIDATE_MB_TYPE_FORWARD | CANDIDATE_MB_TYPE_BACKWARD | CANDIDATE_MB_TYPE_BIDIR | CANDIDATE_MB_TYPE_DIRECT;

  1590.         if(fimin < INT_MAX)

  1591.             type |= CANDIDATE_MB_TYPE_FORWARD_I;

  1592.         if(bimin < INT_MAX)

  1593.             type |= CANDIDATE_MB_TYPE_BACKWARD_I;

  1594.         if(fimin < INT_MAX && bimin < INT_MAX){

  1595.             type |= CANDIDATE_MB_TYPE_BIDIR_I;

  1596.         }

  1597.          //FIXME something smarter

  1598.         if(dmin>256*256*16) type&= ~CANDIDATE_MB_TYPE_DIRECT; //do not try direct mode if it is invalid for this MB

  1599.         if (s->codec_id == AV_CODEC_ID_MPEG4 && type&CANDIDATE_MB_TYPE_DIRECT &&

  1600.             s->mpv_flags & FF_MPV_FLAG_MV0 && *(uint32_t*)s->b_direct_mv_table[xy])

  1601.             type |= CANDIDATE_MB_TYPE_DIRECT0;

  1602.     }

  1603. 

  1604.     s->mb_type[mb_y*s->mb_stride + mb_x]= type;

  1605. }

  1606. 

  1607. /* find best f_code for ME which do unlimited searches */

  1608. int ff_get_best_fcode(MpegEncContext * s, int16_t (*mv_table)[2], int type)

  1609. {

  1610.     if(s->me_method>=ME_EPZS){

  1611.         int score[8];

  1612.         int i, y, range= s->avctx->me_range ? s->avctx->me_range : (INT_MAX/2);

  1613.         uint8_t * fcode_tab= s->fcode_tab;

  1614.         int best_fcode=-1;

  1615.         int best_score=-10000000;

  1616. 

  1617.         if(s->msmpeg4_version)

  1618.             range= FFMIN(range, 16);

  1619.         else if(s->codec_id == AV_CODEC_ID_MPEG2VIDEO && s->avctx->strict_std_compliance >= FF_COMPLIANCE_NORMAL)

  1620.             range= FFMIN(range, 256);

  1621. 

  1622.         for(i=0; i<8; i++) score[i]= s->mb_num*(8-i);

  1623. 

  1624.         for(y=0; y<s->mb_height; y++){

  1625.             int x;

  1626.             int xy= y*s->mb_stride;

  1627.             for(x=0; x<s->mb_width; x++){

  1628.                 if(s->mb_type[xy] & type){

  1629.                     int mx= mv_table[xy][0];

  1630.                     int my= mv_table[xy][1];

  1631.                     int fcode= FFMAX(fcode_tab[mx + MAX_MV],

  1632.                                      fcode_tab[my + MAX_MV]);

  1633.                     int j;

  1634. 

  1635.                         if(mx >= range || mx < -range ||

  1636.                            my >= range || my < -range)

  1637.                             continue;

  1638. 

  1639.                     for(j=0; j<fcode && j<8; j++){

  1640.                         if(s->pict_type==AV_PICTURE_TYPE_B || s->current_picture.mc_mb_var[xy] < s->current_picture.mb_var[xy])

  1641.                             score[j]-= 170;

  1642.                     }

  1643.                 }

  1644.                 xy++;

  1645.             }

  1646.         }

  1647. 

  1648.         for(i=1; i<8; i++){

  1649.             if(score[i] > best_score){

  1650.                 best_score= score[i];

  1651.                 best_fcode= i;

  1652.             }

  1653.         }

  1654. 

  1655.         return best_fcode;

  1656.     }else{

  1657.         return 1;

  1658.     }

  1659. }

  1660. 

  1661. void ff_fix_long_p_mvs(MpegEncContext * s)

  1662. {

  1663.     MotionEstContext * const c= &s->me;

  1664.     const int f_code= s->f_code;

  1665.     int y, range;

  1666.     assert(s->pict_type==AV_PICTURE_TYPE_P);

  1667. 

  1668.     range = (((s->out_format == FMT_MPEG1 || s->msmpeg4_version) ? 8 : 16) << f_code);

  1669. 

  1670.     assert(range <= 16 || !s->msmpeg4_version);

  1671.     assert(range <=256 || !(s->codec_id == AV_CODEC_ID_MPEG2VIDEO && s->avctx->strict_std_compliance >= FF_COMPLIANCE_NORMAL));

  1672. 

  1673.     if(c->avctx->me_range && range > c->avctx->me_range) range= c->avctx->me_range;

  1674. 

  1675.     if(s->flags&CODEC_FLAG_4MV){

  1676.         const int wrap= s->b8_stride;

  1677. 

  1678.         /* clip / convert to intra 8x8 type MVs */

  1679.         for(y=0; y<s->mb_height; y++){

  1680.             int xy= y*2*wrap;

  1681.             int i= y*s->mb_stride;

  1682.             int x;

  1683. 

  1684.             for(x=0; x<s->mb_width; x++){

  1685.                 if(s->mb_type[i]&CANDIDATE_MB_TYPE_INTER4V){

  1686.                     int block;

  1687.                     for(block=0; block<4; block++){

  1688.                         int off= (block& 1) + (block>>1)*wrap;

  1689.                         int mx = s->current_picture.motion_val[0][ xy + off ][0];

  1690.                         int my = s->current_picture.motion_val[0][ xy + off ][1];

  1691. 

  1692.                         if(   mx >=range || mx <-range

  1693.                            || my >=range || my <-range){

  1694.                             s->mb_type[i] &= ~CANDIDATE_MB_TYPE_INTER4V;

  1695.                             s->mb_type[i] |= CANDIDATE_MB_TYPE_INTRA;

  1696.                             s->current_picture.mb_type[i] = CANDIDATE_MB_TYPE_INTRA;

  1697.                         }

  1698.                     }

  1699.                 }

  1700.                 xy+=2;

  1701.                 i++;

  1702.             }

  1703.         }

  1704.     }

  1705. }

  1706. 

  1707. /**

  1708.  *

  1709.  * @param truncate 1 for truncation, 0 for using intra

  1710.  */

  1711. void ff_fix_long_mvs(MpegEncContext * s, uint8_t *field_select_table, int field_select,

  1712.                      int16_t (*mv_table)[2], int f_code, int type, int truncate)

  1713. {

  1714.     MotionEstContext * const c= &s->me;

  1715.     int y, h_range, v_range;

  1716. 

  1717.     // RAL: 8 in MPEG-1, 16 in MPEG-4

  1718.     int range = (((s->out_format == FMT_MPEG1 || s->msmpeg4_version) ? 8 : 16) << f_code);

  1719. 

  1720.     if(c->avctx->me_range && range > c->avctx->me_range) range= c->avctx->me_range;

  1721. 

  1722.     h_range= range;

  1723.     v_range= field_select_table ? range>>1 : range;

  1724. 

  1725.     /* clip / convert to intra 16x16 type MVs */

  1726.     for(y=0; y<s->mb_height; y++){

  1727.         int x;

  1728.         int xy= y*s->mb_stride;

  1729.         for(x=0; x<s->mb_width; x++){

  1730.             if (s->mb_type[xy] & type){    // RAL: "type" test added...

  1731.                 if(field_select_table==NULL || field_select_table[xy] == field_select){

  1732.                     if(   mv_table[xy][0] >=h_range || mv_table[xy][0] <-h_range

  1733.                        || mv_table[xy][1] >=v_range || mv_table[xy][1] <-v_range){

  1734. 

  1735.                         if(truncate){

  1736.                             if     (mv_table[xy][0] > h_range-1) mv_table[xy][0]=  h_range-1;

  1737.                             else if(mv_table[xy][0] < -h_range ) mv_table[xy][0]= -h_range;

  1738.                             if     (mv_table[xy][1] > v_range-1) mv_table[xy][1]=  v_range-1;

  1739.                             else if(mv_table[xy][1] < -v_range ) mv_table[xy][1]= -v_range;

  1740.                         }else{

  1741.                             s->mb_type[xy] &= ~type;

  1742.                             s->mb_type[xy] |= CANDIDATE_MB_TYPE_INTRA;

  1743.                             mv_table[xy][0]=

  1744.                             mv_table[xy][1]= 0;

  1745.                         }

  1746.                     }

  1747.                 }

  1748.             }

  1749.             xy++;

  1750.         }

  1751.     }

  1752. }