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.
43415 Commits
32 Branches
824MB
Tree: 82b7525173
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '82b7525173'
${ noResults }
FFmpeg/libavcodec/motion_est.c

1731 lines
66KB
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 "internal.h"

  36. #include "mathops.h"

  37. #include "motion_est.h"

  38. #include "mpegutils.h"

  39. #include "mpegvideo.h"

  40. 

  41. #undef NDEBUG

  42. #include <assert.h>

  43. 

  44. #define P_LEFT P[1]

  45. #define P_TOP P[2]

  46. #define P_TOPRIGHT P[3]

  47. #define P_MEDIAN P[4]

  48. #define P_MV1 P[9]

  49. 

  50. #define ME_MAP_SHIFT 3

  51. #define ME_MAP_MV_BITS 11

  52. 

  53. static int sad_hpel_motion_search(MpegEncContext * s,

  54.                                   int *mx_ptr, int *my_ptr, int dmin,

  55.                                   int src_index, int ref_index,

  56.                                   int size, int h);

  57. 

  58. static inline unsigned update_map_generation(MotionEstContext *c)

  59. {

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

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

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

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

  64.     }

  65.     return c->map_generation;

  66. }

  67. 

  68. /* shape adaptive search stuff */

  69. typedef struct Minima{

  70.     int height;

  71.     int x, y;

  72.     int checked;

  73. }Minima;

  74. 

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

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

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

  78. 

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

  80. }

  81. 

  82. #define FLAG_QPEL   1 //must be 1

  83. #define FLAG_CHROMA 2

  84. #define FLAG_DIRECT 4

  85. 

  86. 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){

  87.     const int offset[3]= {

  88.           y*c->  stride + x,

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

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

  91.     };

  92.     int i;

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

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

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

  96.     }

  97.     if(ref_index){

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

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

  100.         }

  101.     }

  102. }

  103. 

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

  105.     return   ((c->avctx->flags&AV_CODEC_FLAG_QPEL) ? FLAG_QPEL : 0)

  106.            + (direct ? FLAG_DIRECT : 0)

  107.            + (chroma ? FLAG_CHROMA : 0);

  108. }

  109. 

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

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

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

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

  114.     const int stride= c->stride;

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

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

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

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

  119.     int d;

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

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

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

  123.             const int time_pp= s->pp_time;

  124.             const int time_pb= s->pb_time;

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

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

  127.                 int i;

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

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

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

  131.                     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));

  132.                     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));

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

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

  135. 

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

  137.                     if(qpel){

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

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

  140.                     }else{

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

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

  143.                     }

  144.                 }

  145.             }else{

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

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

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

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

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

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

  152. 

  153.                 if(qpel){

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

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

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

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

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

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

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

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

  162.                 }else{

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

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

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

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

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

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

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

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

  171. 

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

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

  174.                 }

  175.             }

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

  177.         }else

  178.             d= 256*256*256*32;

  179.     return d;

  180. }

  181. 

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

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

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

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

  186.     const int stride= c->stride;

  187.     const int uvstride= c->uvstride;

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

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

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

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

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

  193.     int d;

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

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

  196.         if(dxy){

  197.             if(qpel){

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

  199.                 if(chroma){

  200.                     int cx= hx/2;

  201.                     int cy= hy/2;

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

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

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

  205.                     // FIXME x/y wrong, but MPEG-4 qpel is sick anyway, we should drop as much of it as possible in favor for H.264

  206.                 }

  207.             }else{

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

  209.                 if(chroma)

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

  211.             }

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

  213.         }else{

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

  215.             if(chroma)

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

  217.         }

  218.         if(chroma){

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

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

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

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

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

  224.         }

  225.     return d;

  226. }

  227. 

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

  229.                       int ref_index, int src_index,

  230.                       me_cmp_func cmp_func, me_cmp_func chroma_cmp_func){

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

  232. }

  233. 

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

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

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

  237.     if(flags&FLAG_DIRECT){

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

  239.     }else{

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

  241.     }

  242. }

  243. 

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

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

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

  247.     if(flags&FLAG_DIRECT){

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

  249.     }else{

  250.         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);

  251.     }

  252. }

  253. 

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

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

  256.  */

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

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

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

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

  261.        && av_builtin_constant_p(subx) && av_builtin_constant_p(suby)

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

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

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

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

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

  267.     }else{

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

  269.     }

  270. }

  271. 

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

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

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

  275.     if(flags&FLAG_DIRECT){

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

  277.     }else{

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

  279.     }

  280. }

  281. 

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

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

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

  285.     if(flags&FLAG_DIRECT){

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

  287.     }else{

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

  289.     }

  290. }

  291. 

  292. #include "motion_est_template.c"

  293. 

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

  295.                     ptrdiff_t stride, int h)

  296. {

  297.     return 0;

  298. }

  299. 

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

  301. }

  302. 

  303. int ff_init_me(MpegEncContext *s){

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

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

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

  307. 

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

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

  310.         return -1;

  311.     }

  312. 

  313. #if FF_API_MOTION_EST

  314. FF_DISABLE_DEPRECATION_WARNINGS

  315.     if (s->motion_est == FF_ME_EPZS) {

  316.         if (s->me_method == ME_ZERO)

  317.             s->motion_est = FF_ME_ZERO;

  318.         else if (s->me_method == ME_EPZS)

  319.             s->motion_est = FF_ME_EPZS;

  320.         else if (s->me_method == ME_X1)

  321.             s->motion_est = FF_ME_XONE;

  322.         else {

  323.             av_log(s->avctx, AV_LOG_ERROR,

  324.                    "me_method is only allowed to be set to zero and epzs; "

  325.                    "for hex,umh,full and others see dia_size\n");

  326.             return -1;

  327.         }

  328.     }

  329. FF_ENABLE_DEPRECATION_WARNINGS

  330. #endif

  331. 

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

  333. 

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

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

  336.     }

  337. 

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

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

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

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

  342. 

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

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

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

  346. 

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

  348.     if (s->avctx->flags & AV_CODEC_FLAG_QPEL) {

  349.         c->sub_motion_search= qpel_motion_search;

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

  351.         if (s->no_rounding)

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

  353.         else

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

  355.     }else{

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

  357.             c->sub_motion_search= hpel_motion_search;

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

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

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

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

  362.         else

  363.             c->sub_motion_search= hpel_motion_search;

  364.     }

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

  366.     if (s->no_rounding)

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

  368.     else

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

  370. 

  371.     if(s->linesize){

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

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

  374.     }else{

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

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

  377.     }

  378. 

  379.     /* 8x8 fullpel search would need a 4x4 chroma compare, which we do

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

  381.      * does not expect it. */

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

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

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

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

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

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

  388. 

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

  390.         c->sub_motion_search= no_sub_motion_search;

  391.     }

  392. 

  393.     return 0;

  394. }

  395. 

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

  397. {\

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

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

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

  401. }

  402. 

  403. static int sad_hpel_motion_search(MpegEncContext * s,

  404.                                   int *mx_ptr, int *my_ptr, int dmin,

  405.                                   int src_index, int ref_index,

  406.                                   int size, int h)

  407. {

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

  409.     const int penalty_factor= c->sub_penalty_factor;

  410.     int mx, my, dminh;

  411.     uint8_t *pix, *ptr;

  412.     int stride= c->stride;

  413.     const int flags= c->sub_flags;

  414.     LOAD_COMMON

  415. 

  416.     assert(flags == 0);

  417. 

  418.     if(c->skip){

  419.         *mx_ptr = 0;

  420.         *my_ptr = 0;

  421.         return dmin;

  422.     }

  423. 

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

  425. 

  426.     mx = *mx_ptr;

  427.     my = *my_ptr;

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

  429. 

  430.     dminh = dmin;

  431. 

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

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

  434.         int dx=0, dy=0;

  435.         int d, pen_x, pen_y;

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

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

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

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

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

  441.         mx<<=1;

  442.         my<<=1;

  443. 

  444. 

  445.         pen_x= pred_x + mx;

  446.         pen_y= pred_y + my;

  447. 

  448.         ptr-= stride;

  449.         if(t<=b){

  450.             CHECK_SAD_HALF_MV(y2 , 0, -1)

  451.             if(l<=r){

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

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

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

  455.                     ptr+= stride;

  456.                 }else{

  457.                     ptr+= stride;

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

  459.                 }

  460.                 CHECK_SAD_HALF_MV(x2 , -1,  0)

  461.             }else{

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

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

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

  465.                     ptr+= stride;

  466.                 }else{

  467.                     ptr+= stride;

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

  469.                 }

  470.                 CHECK_SAD_HALF_MV(x2 , +1,  0)

  471.             }

  472.         }else{

  473.             if(l<=r){

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

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

  476.                     ptr+= stride;

  477.                 }else{

  478.                     ptr+= stride;

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

  480.                 }

  481.                 CHECK_SAD_HALF_MV(x2 , -1,  0)

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

  483.             }else{

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

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

  486.                     ptr+= stride;

  487.                 }else{

  488.                     ptr+= stride;

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

  490.                 }

  491.                 CHECK_SAD_HALF_MV(x2 , +1,  0)

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

  493.             }

  494.             CHECK_SAD_HALF_MV(y2 ,  0, +1)

  495.         }

  496.         mx+=dx;

  497.         my+=dy;

  498. 

  499.     }else{

  500.         mx<<=1;

  501.         my<<=1;

  502.     }

  503. 

  504.     *mx_ptr = mx;

  505.     *my_ptr = my;

  506.     return dminh;

  507. }

  508. 

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

  510. {

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

  512. 

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

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

  515. 

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

  517.     if(mv4){

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

  519. 

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

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

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

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

  524. 

  525.         mot_xy += s->b8_stride;

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

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

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

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

  530.     }

  531. }

  532. 

  533. /**

  534.  * get fullpel ME search limits.

  535.  */

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

  537. {

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

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

  540. /*

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

  542.     else                   c->range= 16;

  543. */

  544.     if (s->unrestricted_mv) {

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

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

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

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

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

  550.         // Search range of H.261 is different from other codec standards

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

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

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

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

  555.     } else {

  556.         c->xmin = - x;

  557.         c->ymin = - y;

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

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

  560.     }

  561.     if(range){

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

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

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

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

  566.     }

  567. }

  568. 

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

  570.     const int stride= c->stride;

  571. 

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

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

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

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

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

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

  578. }

  579. 

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

  581. {

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

  583.     const int size= 1;

  584.     const int h=8;

  585.     int block;

  586.     int P[10][2];

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

  588.     int same=1;

  589.     const int stride= c->stride;

  590.     uint8_t *mv_penalty= c->current_mv_penalty;

  591. 

  592.     init_mv4_ref(c);

  593. 

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

  595.         int mx4, my4;

  596.         int pred_x4, pred_y4;

  597.         int dmin4;

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

  599.         const int mot_stride = s->b8_stride;

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

  601. 

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

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

  604. 

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

  606. 

  607.         /* special case for first line */

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

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

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

  611.         } else {

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

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

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

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

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

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

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

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

  620. 

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

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

  623. 

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

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

  626.         }

  627.         P_MV1[0]= mx;

  628.         P_MV1[1]= my;

  629. 

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

  631. 

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

  633. 

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

  635.             int dxy;

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

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

  638.             if(s->quarter_sample){

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

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

  641. 

  642.                 if(s->no_rounding)

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

  644.                 else

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

  646.             }else{

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

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

  649. 

  650.                 if(s->no_rounding)

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

  652.                 else

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

  654.             }

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

  656.         }else

  657.             dmin_sum+= dmin4;

  658. 

  659.         if(s->quarter_sample){

  660.             mx4_sum+= mx4/2;

  661.             my4_sum+= my4/2;

  662.         }else{

  663.             mx4_sum+= mx4;

  664.             my4_sum+= my4;

  665.         }

  666. 

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

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

  669. 

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

  671.     }

  672. 

  673.     if(same)

  674.         return INT_MAX;

  675. 

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

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

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

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

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

  681.     }

  682. 

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

  684.         int dxy;

  685.         int mx, my;

  686.         int offset;

  687. 

  688.         mx= ff_h263_round_chroma(mx4_sum);

  689.         my= ff_h263_round_chroma(my4_sum);

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

  691. 

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

  693. 

  694.         if(s->no_rounding){

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

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

  697.         }else{

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

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

  700.         }

  701. 

  702.         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);

  703.         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);

  704.     }

  705. 

  706.     c->pred_x= mx;

  707.     c->pred_y= my;

  708. 

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

  710.     /*case FF_CMP_SSE:

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

  712.     case FF_CMP_RD:

  713.         return dmin_sum;

  714.     default:

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

  716.     }

  717. }

  718. 

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

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

  721. 

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

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

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

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

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

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

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

  729.     }

  730. }

  731. 

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

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

  734. {

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

  736.     const int size=0;

  737.     const int h=8;

  738.     int block;

  739.     int P[10][2];

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

  741.     int same=1;

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

  743.     int dmin_sum= 0;

  744.     const int mot_stride= s->mb_stride;

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

  746. 

  747.     c->ymin>>=1;

  748.     c->ymax>>=1;

  749.     c->stride<<=1;

  750.     c->uvstride<<=1;

  751.     init_interlaced_ref(s, ref_index);

  752. 

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

  754.         int field_select;

  755.         int best_dmin= INT_MAX;

  756.         int best_field= -1;

  757. 

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

  759.             int dmin, mx_i, my_i;

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

  761. 

  762.             if(user_field_select){

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

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

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

  766.                     continue;

  767.             }

  768. 

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

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

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

  772. 

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

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

  775. 

  776.             if(!s->first_slice_line){

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

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

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

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

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

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

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

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

  785. 

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

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

  788.             }

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

  790.             P_MV1[1]= my / 2;

  791. 

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

  793. 

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

  795. 

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

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

  798. 

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

  800.                 int dxy;

  801. 

  802.                 //FIXME chroma ME

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

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

  805. 

  806.                 if(s->no_rounding){

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

  808.                 }else{

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

  810.                 }

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

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

  813.             }else

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

  815. 

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

  817. 

  818.             if(dmin < best_dmin){

  819.                 best_dmin= dmin;

  820.                 best_field= field_select;

  821.             }

  822.         }

  823.         {

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

  825. 

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

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

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

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

  830.         }

  831. 

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

  833.         dmin_sum += best_dmin;

  834.     }

  835. 

  836.     c->ymin<<=1;

  837.     c->ymax<<=1;

  838.     c->stride>>=1;

  839.     c->uvstride>>=1;

  840. 

  841.     if(same)

  842.         return INT_MAX;

  843. 

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

  845.     /*case FF_CMP_SSE:

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

  847.     case FF_CMP_RD:

  848.         return dmin_sum;

  849.     default:

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

  851.     }

  852. }

  853. 

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

  855.     switch(type&0xFF){

  856.     default:

  857.     case FF_CMP_SAD:

  858.         return lambda>>FF_LAMBDA_SHIFT;

  859.     case FF_CMP_DCT:

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

  861.     case FF_CMP_SATD:

  862.     case FF_CMP_DCT264:

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

  864.     case FF_CMP_RD:

  865.     case FF_CMP_PSNR:

  866.     case FF_CMP_SSE:

  867.     case FF_CMP_NSSE:

  868.         return lambda2>>FF_LAMBDA_SHIFT;

  869.     case FF_CMP_BIT:

  870.         return 1;

  871.     }

  872. }

  873. 

  874. void ff_estimate_p_frame_motion(MpegEncContext * s,

  875.                                 int mb_x, int mb_y)

  876. {

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

  878.     uint8_t *pix, *ppix;

  879.     int sum, mx = 0, my = 0, dmin = 0;

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

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

  882.     int P[10][2];

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

  884.     int mb_type=0;

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

  886. 

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

  888. 

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

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

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

  892. 

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

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

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

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

  897. 

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

  899.     c->skip=0;

  900. 

  901.     /* intra / predictive decision */

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

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

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

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

  906. 

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

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

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

  910. 

  911.     if (s->motion_est != FF_ME_ZERO) {

  912.         const int mot_stride = s->b8_stride;

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

  914. 

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

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

  917. 

  918.         if (P_LEFT[0] > (c->xmax << shift))

  919.             P_LEFT[0] =  c->xmax << shift;

  920. 

  921.         if (!s->first_slice_line) {

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

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

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

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

  926.             if (P_TOP[1] > (c->ymax << shift))

  927.                 P_TOP[1] =  c->ymax << shift;

  928.             if (P_TOPRIGHT[0] < (c->xmin << shift))

  929.                 P_TOPRIGHT[0] =  c->xmin << shift;

  930.             if (P_TOPRIGHT[1] > (c->ymax << shift))

  931.                 P_TOPRIGHT[1] =  c->ymax << shift;

  932. 

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

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

  935. 

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

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

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

  939.             } else { /* MPEG-1 at least */

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

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

  942.             }

  943.         } else {

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

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

  946.         }

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

  948.     }

  949. 

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

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

  952. 

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

  954. 

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

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

  957. 

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

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

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

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

  962. 

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

  964.             mb_type|= CANDIDATE_MB_TYPE_INTRA;

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

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

  967.             mb_type|= CANDIDATE_MB_TYPE_INTER;

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

  969.             if (s->mpv_flags & FF_MPV_FLAG_MV0)

  970.                 if(mx || my)

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

  972.         }else{

  973.             mx <<=shift;

  974.             my <<=shift;

  975.         }

  976.         if ((s->avctx->flags & AV_CODEC_FLAG_4MV)

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

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

  979.                 mb_type|=CANDIDATE_MB_TYPE_INTER4V;

  980. 

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

  982.         }else

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

  984.         if ((s->avctx->flags & AV_CODEC_FLAG_INTERLACED_ME)

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

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

  987.                 mb_type |= CANDIDATE_MB_TYPE_INTER_I;

  988.         }

  989.     }else{

  990.         int intra_score, i;

  991.         mb_type= CANDIDATE_MB_TYPE_INTER;

  992. 

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

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

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

  996. 

  997.         if ((s->avctx->flags & AV_CODEC_FLAG_4MV)

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

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

  1000.             if(dmin4 < dmin){

  1001.                 mb_type= CANDIDATE_MB_TYPE_INTER4V;

  1002.                 dmin=dmin4;

  1003.             }

  1004.         }

  1005.         if ((s->avctx->flags & AV_CODEC_FLAG_INTERLACED_ME)

  1006.            && !c->skip){ //FIXME varc/d checks

  1007.             int dmin_i= interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my, 0);

  1008.             if(dmin_i < dmin){

  1009.                 mb_type = CANDIDATE_MB_TYPE_INTER_I;

  1010.                 dmin= dmin_i;

  1011.             }

  1012.         }

  1013. 

  1014.         set_p_mv_tables(s, mx, my, mb_type!=CANDIDATE_MB_TYPE_INTER4V);

  1015. 

  1016.         /* get intra luma score */

  1017.         if((c->avctx->mb_cmp&0xFF)==FF_CMP_SSE){

  1018.             intra_score= varc - 500;

  1019.         }else{

  1020.             unsigned mean = (sum+128)>>8;

  1021.             mean*= 0x01010101;

  1022. 

  1023.             for(i=0; i<16; i++){

  1024.                 *(uint32_t*)(&c->scratchpad[i*s->linesize+ 0]) = mean;

  1025.                 *(uint32_t*)(&c->scratchpad[i*s->linesize+ 4]) = mean;

  1026.                 *(uint32_t*)(&c->scratchpad[i*s->linesize+ 8]) = mean;

  1027.                 *(uint32_t*)(&c->scratchpad[i*s->linesize+12]) = mean;

  1028.             }

  1029. 

  1030.             intra_score= s->mecc.mb_cmp[0](s, c->scratchpad, pix, s->linesize, 16);

  1031.         }

  1032.         intra_score += c->mb_penalty_factor*16;

  1033. 

  1034.         if(intra_score < dmin){

  1035.             mb_type= CANDIDATE_MB_TYPE_INTRA;

  1036.             s->current_picture.mb_type[mb_y*s->mb_stride + mb_x] = CANDIDATE_MB_TYPE_INTRA; //FIXME cleanup

  1037.         }else

  1038.             s->current_picture.mb_type[mb_y*s->mb_stride + mb_x] = 0;

  1039. 

  1040.         {

  1041.             int p_score= FFMIN(vard, varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*100);

  1042.             int i_score= varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*20;

  1043.             c->scene_change_score+= ff_sqrt(p_score) - ff_sqrt(i_score);

  1044.         }

  1045.     }

  1046. 

  1047.     s->mb_type[mb_y*s->mb_stride + mb_x]= mb_type;

  1048. }

  1049. 

  1050. int ff_pre_estimate_p_frame_motion(MpegEncContext * s,

  1051.                                     int mb_x, int mb_y)

  1052. {

  1053.     MotionEstContext * const c= &s->me;

  1054.     int mx, my, dmin;

  1055.     int P[10][2];

  1056.     const int shift= 1+s->quarter_sample;

  1057.     const int xy= mb_x + mb_y*s->mb_stride;

  1058.     init_ref(c, s->new_picture.f->data, s->last_picture.f->data, NULL, 16*mb_x, 16*mb_y, 0);

  1059. 

  1060.     assert(s->quarter_sample==0 || s->quarter_sample==1);

  1061. 

  1062.     c->pre_penalty_factor    = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_pre_cmp);

  1063.     c->current_mv_penalty= c->mv_penalty[s->f_code] + MAX_MV;

  1064. 

  1065.     get_limits(s, 16*mb_x, 16*mb_y);

  1066.     c->skip=0;

  1067. 

  1068.     P_LEFT[0]       = s->p_mv_table[xy + 1][0];

  1069.     P_LEFT[1]       = s->p_mv_table[xy + 1][1];

  1070. 

  1071.     if(P_LEFT[0]       < (c->xmin<<shift)) P_LEFT[0]       = (c->xmin<<shift);

  1072. 

  1073.     /* special case for first line */

  1074.     if (s->first_slice_line) {

  1075.         c->pred_x= P_LEFT[0];

  1076.         c->pred_y= P_LEFT[1];

  1077.         P_TOP[0]= P_TOPRIGHT[0]= P_MEDIAN[0]=

  1078.         P_TOP[1]= P_TOPRIGHT[1]= P_MEDIAN[1]= 0; //FIXME

  1079.     } else {

  1080.         P_TOP[0]      = s->p_mv_table[xy + s->mb_stride    ][0];

  1081.         P_TOP[1]      = s->p_mv_table[xy + s->mb_stride    ][1];

  1082.         P_TOPRIGHT[0] = s->p_mv_table[xy + s->mb_stride - 1][0];

  1083.         P_TOPRIGHT[1] = s->p_mv_table[xy + s->mb_stride - 1][1];

  1084.         if(P_TOP[1]      < (c->ymin<<shift)) P_TOP[1]     = (c->ymin<<shift);

  1085.         if(P_TOPRIGHT[0] > (c->xmax<<shift)) P_TOPRIGHT[0]= (c->xmax<<shift);

  1086.         if(P_TOPRIGHT[1] < (c->ymin<<shift)) P_TOPRIGHT[1]= (c->ymin<<shift);

  1087. 

  1088.         P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);

  1089.         P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);

  1090. 

  1091.         c->pred_x = P_MEDIAN[0];

  1092.         c->pred_y = P_MEDIAN[1];

  1093.     }

  1094. 

  1095.     dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, 0, s->p_mv_table, (1<<16)>>shift, 0, 16);

  1096. 

  1097.     s->p_mv_table[xy][0] = mx<<shift;

  1098.     s->p_mv_table[xy][1] = my<<shift;

  1099. 

  1100.     return dmin;

  1101. }

  1102. 

  1103. static int estimate_motion_b(MpegEncContext *s, int mb_x, int mb_y,

  1104.                              int16_t (*mv_table)[2], int ref_index, int f_code)

  1105. {

  1106.     MotionEstContext * const c= &s->me;

  1107.     int mx = 0, my = 0, dmin = 0;

  1108.     int P[10][2];

  1109.     const int shift= 1+s->quarter_sample;

  1110.     const int mot_stride = s->mb_stride;

  1111.     const int mot_xy = mb_y*mot_stride + mb_x;

  1112.     uint8_t * const mv_penalty= c->mv_penalty[f_code] + MAX_MV;

  1113.     int mv_scale;

  1114. 

  1115.     c->penalty_factor    = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_cmp);

  1116.     c->sub_penalty_factor= get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_sub_cmp);

  1117.     c->mb_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->mb_cmp);

  1118.     c->current_mv_penalty= mv_penalty;

  1119. 

  1120.     get_limits(s, 16*mb_x, 16*mb_y);

  1121. 

  1122.     if (s->motion_est != FF_ME_ZERO) {

  1123.         P_LEFT[0] = mv_table[mot_xy - 1][0];

  1124.         P_LEFT[1] = mv_table[mot_xy - 1][1];

  1125. 

  1126.         if (P_LEFT[0] > (c->xmax << shift)) P_LEFT[0] = (c->xmax << shift);

  1127. 

  1128.         /* special case for first line */

  1129.         if (!s->first_slice_line) {

  1130.             P_TOP[0]      = mv_table[mot_xy - mot_stride    ][0];

  1131.             P_TOP[1]      = mv_table[mot_xy - mot_stride    ][1];

  1132.             P_TOPRIGHT[0] = mv_table[mot_xy - mot_stride + 1][0];

  1133.             P_TOPRIGHT[1] = mv_table[mot_xy - mot_stride + 1][1];

  1134.             if (P_TOP[1] > (c->ymax << shift)) P_TOP[1] = (c->ymax << shift);

  1135.             if (P_TOPRIGHT[0] < (c->xmin << shift)) P_TOPRIGHT[0] = (c->xmin << shift);

  1136.             if (P_TOPRIGHT[1] > (c->ymax << shift)) P_TOPRIGHT[1] = (c->ymax << shift);

  1137. 

  1138.             P_MEDIAN[0] = mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);

  1139.             P_MEDIAN[1] = mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);

  1140.         }

  1141.         c->pred_x = P_LEFT[0];

  1142.         c->pred_y = P_LEFT[1];

  1143. 

  1144.         if(mv_table == s->b_forw_mv_table){

  1145.             mv_scale= (s->pb_time<<16) / (s->pp_time<<shift);

  1146.         }else{

  1147.             mv_scale= ((s->pb_time - s->pp_time)<<16) / (s->pp_time<<shift);

  1148.         }

  1149. 

  1150.         dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, ref_index, s->p_mv_table, mv_scale, 0, 16);

  1151.     }

  1152. 

  1153.     dmin= c->sub_motion_search(s, &mx, &my, dmin, 0, ref_index, 0, 16);

  1154. 

  1155.     if(c->avctx->me_sub_cmp != c->avctx->mb_cmp && !c->skip)

  1156.         dmin= get_mb_score(s, mx, my, 0, ref_index, 0, 16, 1);

  1157. 

  1158. //    s->mb_type[mb_y*s->mb_width + mb_x]= mb_type;

  1159.     mv_table[mot_xy][0]= mx;

  1160.     mv_table[mot_xy][1]= my;

  1161. 

  1162.     return dmin;

  1163. }

  1164. 

  1165. static inline int check_bidir_mv(MpegEncContext * s,

  1166.                    int motion_fx, int motion_fy,

  1167.                    int motion_bx, int motion_by,

  1168.                    int pred_fx, int pred_fy,

  1169.                    int pred_bx, int pred_by,

  1170.                    int size, int h)

  1171. {

  1172.     //FIXME optimize?

  1173.     //FIXME better f_code prediction (max mv & distance)

  1174.     //FIXME pointers

  1175.     MotionEstContext * const c= &s->me;

  1176.     uint8_t * const mv_penalty_f= c->mv_penalty[s->f_code] + MAX_MV; // f_code of the prev frame

  1177.     uint8_t * const mv_penalty_b= c->mv_penalty[s->b_code] + MAX_MV; // f_code of the prev frame

  1178.     int stride= c->stride;

  1179.     uint8_t *dest_y = c->scratchpad;

  1180.     uint8_t *ptr;

  1181.     int dxy;

  1182.     int src_x, src_y;

  1183.     int fbmin;

  1184.     uint8_t **src_data= c->src[0];

  1185.     uint8_t **ref_data= c->ref[0];

  1186.     uint8_t **ref2_data= c->ref[2];

  1187. 

  1188.     if(s->quarter_sample){

  1189.         dxy = ((motion_fy & 3) << 2) | (motion_fx & 3);

  1190.         src_x = motion_fx >> 2;

  1191.         src_y = motion_fy >> 2;

  1192. 

  1193.         ptr = ref_data[0] + (src_y * stride) + src_x;

  1194.         s->qdsp.put_qpel_pixels_tab[0][dxy](dest_y, ptr, stride);

  1195. 

  1196.         dxy = ((motion_by & 3) << 2) | (motion_bx & 3);

  1197.         src_x = motion_bx >> 2;

  1198.         src_y = motion_by >> 2;

  1199. 

  1200.         ptr = ref2_data[0] + (src_y * stride) + src_x;

  1201.         s->qdsp.avg_qpel_pixels_tab[size][dxy](dest_y, ptr, stride);

  1202.     }else{

  1203.         dxy = ((motion_fy & 1) << 1) | (motion_fx & 1);

  1204.         src_x = motion_fx >> 1;

  1205.         src_y = motion_fy >> 1;

  1206. 

  1207.         ptr = ref_data[0] + (src_y * stride) + src_x;

  1208.         s->hdsp.put_pixels_tab[size][dxy](dest_y    , ptr    , stride, h);

  1209. 

  1210.         dxy = ((motion_by & 1) << 1) | (motion_bx & 1);

  1211.         src_x = motion_bx >> 1;

  1212.         src_y = motion_by >> 1;

  1213. 

  1214.         ptr = ref2_data[0] + (src_y * stride) + src_x;

  1215.         s->hdsp.avg_pixels_tab[size][dxy](dest_y    , ptr    , stride, h);

  1216.     }

  1217. 

  1218.     fbmin = (mv_penalty_f[motion_fx-pred_fx] + mv_penalty_f[motion_fy-pred_fy])*c->mb_penalty_factor

  1219.            +(mv_penalty_b[motion_bx-pred_bx] + mv_penalty_b[motion_by-pred_by])*c->mb_penalty_factor

  1220.            + s->mecc.mb_cmp[size](s, src_data[0], dest_y, stride, h); // FIXME new_pic

  1221. 

  1222.     if(c->avctx->mb_cmp&FF_CMP_CHROMA){

  1223.     }

  1224.     //FIXME CHROMA !!!

  1225. 

  1226.     return fbmin;

  1227. }

  1228. 

  1229. /* refine the bidir vectors in hq mode and return the score in both lq & hq mode*/

  1230. static inline int bidir_refine(MpegEncContext * s, int mb_x, int mb_y)

  1231. {

  1232.     MotionEstContext * const c= &s->me;

  1233.     const int mot_stride = s->mb_stride;

  1234.     const int xy = mb_y *mot_stride + mb_x;

  1235.     int fbmin;

  1236.     int pred_fx= s->b_bidir_forw_mv_table[xy-1][0];

  1237.     int pred_fy= s->b_bidir_forw_mv_table[xy-1][1];

  1238.     int pred_bx= s->b_bidir_back_mv_table[xy-1][0];

  1239.     int pred_by= s->b_bidir_back_mv_table[xy-1][1];

  1240.     int motion_fx= s->b_bidir_forw_mv_table[xy][0]= s->b_forw_mv_table[xy][0];

  1241.     int motion_fy= s->b_bidir_forw_mv_table[xy][1]= s->b_forw_mv_table[xy][1];

  1242.     int motion_bx= s->b_bidir_back_mv_table[xy][0]= s->b_back_mv_table[xy][0];

  1243.     int motion_by= s->b_bidir_back_mv_table[xy][1]= s->b_back_mv_table[xy][1];

  1244.     const int flags= c->sub_flags;

  1245.     const int qpel= flags&FLAG_QPEL;

  1246.     const int shift= 1+qpel;

  1247.     const int xmin= c->xmin<<shift;

  1248.     const int ymin= c->ymin<<shift;

  1249.     const int xmax= c->xmax<<shift;

  1250.     const int ymax= c->ymax<<shift;

  1251. #define HASH(fx,fy,bx,by) ((fx)+17*(fy)+63*(bx)+117*(by))

  1252. #define HASH8(fx,fy,bx,by) ((uint8_t)HASH(fx,fy,bx,by))

  1253.     int hashidx= HASH(motion_fx,motion_fy, motion_bx, motion_by);

  1254.     uint8_t map[256] = { 0 };

  1255. 

  1256.     map[hashidx&255] = 1;

  1257. 

  1258.     fbmin= check_bidir_mv(s, motion_fx, motion_fy,

  1259.                           motion_bx, motion_by,

  1260.                           pred_fx, pred_fy,

  1261.                           pred_bx, pred_by,

  1262.                           0, 16);

  1263. 

  1264.     if(s->avctx->bidir_refine){

  1265.         int end;

  1266.         static const uint8_t limittab[5]={0,8,32,64,80};

  1267.         const int limit= limittab[s->avctx->bidir_refine];

  1268.         static const int8_t vect[][4]={

  1269. { 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},

  1270. 

  1271. { 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},

  1272. { 0, 1, 0, 1}, { 0,-1, 0,-1}, { 1, 0, 1, 0}, {-1, 0,-1, 0},

  1273. { 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},

  1274. { 0,-1, 0, 1}, { 0, 1, 0,-1}, {-1, 0, 1, 0}, { 1, 0,-1, 0},

  1275. 

  1276. { 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},

  1277. { 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},

  1278. { 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},

  1279. { 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},

  1280. 

  1281. { 1, 1, 1, 1}, {-1,-1,-1,-1},

  1282. { 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},

  1283. { 1, 1,-1,-1}, {-1,-1, 1, 1}, { 1,-1,-1, 1}, {-1, 1, 1,-1}, { 1,-1, 1,-1}, {-1, 1,-1, 1},

  1284.         };

  1285.         static const uint8_t hash[]={

  1286. 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),

  1287. 

  1288. 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),

  1289. HASH8( 0, 1, 0, 1), HASH8( 0,-1, 0,-1), HASH8( 1, 0, 1, 0), HASH8(-1, 0,-1, 0),

  1290. 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),

  1291. HASH8( 0,-1, 0, 1), HASH8( 0, 1, 0,-1), HASH8(-1, 0, 1, 0), HASH8( 1, 0,-1, 0),

  1292. 

  1293. 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),

  1294. 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),

  1295. 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),

  1296. 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),

  1297. 

  1298. HASH8( 1, 1, 1, 1), HASH8(-1,-1,-1,-1),

  1299. 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),

  1300. 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),

  1301. };

  1302. 

  1303. #define CHECK_BIDIR(fx,fy,bx,by)\

  1304.     if( !map[(hashidx+HASH(fx,fy,bx,by))&255]\

  1305.        &&(fx<=0 || motion_fx+fx<=xmax) && (fy<=0 || motion_fy+fy<=ymax) && (bx<=0 || motion_bx+bx<=xmax) && (by<=0 || motion_by+by<=ymax)\

  1306.        &&(fx>=0 || motion_fx+fx>=xmin) && (fy>=0 || motion_fy+fy>=ymin) && (bx>=0 || motion_bx+bx>=xmin) && (by>=0 || motion_by+by>=ymin)){\

  1307.         int score;\

  1308.         map[(hashidx+HASH(fx,fy,bx,by))&255] = 1;\

  1309.         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);\

  1310.         if(score < fbmin){\

  1311.             hashidx += HASH(fx,fy,bx,by);\

  1312.             fbmin= score;\

  1313.             motion_fx+=fx;\

  1314.             motion_fy+=fy;\

  1315.             motion_bx+=bx;\

  1316.             motion_by+=by;\

  1317.             end=0;\

  1318.         }\

  1319.     }

  1320. #define CHECK_BIDIR2(a,b,c,d)\

  1321. CHECK_BIDIR(a,b,c,d)\

  1322. CHECK_BIDIR(-(a),-(b),-(c),-(d))

  1323. 

  1324.         do{

  1325.             int i;

  1326.             int borderdist=0;

  1327.             end=1;

  1328. 

  1329.             CHECK_BIDIR2(0,0,0,1)

  1330.             CHECK_BIDIR2(0,0,1,0)

  1331.             CHECK_BIDIR2(0,1,0,0)

  1332.             CHECK_BIDIR2(1,0,0,0)

  1333. 

  1334.             for(i=8; i<limit; i++){

  1335.                 int fx= motion_fx+vect[i][0];

  1336.                 int fy= motion_fy+vect[i][1];

  1337.                 int bx= motion_bx+vect[i][2];

  1338.                 int by= motion_by+vect[i][3];

  1339.                 if(borderdist<=0){

  1340.                     int a= (xmax - FFMAX(fx,bx))|(FFMIN(fx,bx) - xmin);

  1341.                     int b= (ymax - FFMAX(fy,by))|(FFMIN(fy,by) - ymin);

  1342.                     if((a|b) < 0)

  1343.                         map[(hashidx+hash[i])&255] = 1;

  1344.                 }

  1345.                 if(!map[(hashidx+hash[i])&255]){

  1346.                     int score;

  1347.                     map[(hashidx+hash[i])&255] = 1;

  1348.                     score= check_bidir_mv(s, fx, fy, bx, by, pred_fx, pred_fy, pred_bx, pred_by, 0, 16);

  1349.                     if(score < fbmin){

  1350.                         hashidx += hash[i];

  1351.                         fbmin= score;

  1352.                         motion_fx=fx;

  1353.                         motion_fy=fy;

  1354.                         motion_bx=bx;

  1355.                         motion_by=by;

  1356.                         end=0;

  1357.                         borderdist--;

  1358.                         if(borderdist<=0){

  1359.                             int a= FFMIN(xmax - FFMAX(fx,bx), FFMIN(fx,bx) - xmin);

  1360.                             int b= FFMIN(ymax - FFMAX(fy,by), FFMIN(fy,by) - ymin);

  1361.                             borderdist= FFMIN(a,b);

  1362.                         }

  1363.                     }

  1364.                 }

  1365.             }

  1366.         }while(!end);

  1367.     }

  1368. 

  1369.     s->b_bidir_forw_mv_table[xy][0]= motion_fx;

  1370.     s->b_bidir_forw_mv_table[xy][1]= motion_fy;

  1371.     s->b_bidir_back_mv_table[xy][0]= motion_bx;

  1372.     s->b_bidir_back_mv_table[xy][1]= motion_by;

  1373. 

  1374.     return fbmin;

  1375. }

  1376. 

  1377. static inline int direct_search(MpegEncContext * s, int mb_x, int mb_y)

  1378. {

  1379.     MotionEstContext * const c= &s->me;

  1380.     int P[10][2];

  1381.     const int mot_stride = s->mb_stride;

  1382.     const int mot_xy = mb_y*mot_stride + mb_x;

  1383.     const int shift= 1+s->quarter_sample;

  1384.     int dmin, i;

  1385.     const int time_pp= s->pp_time;

  1386.     const int time_pb= s->pb_time;

  1387.     int mx, my, xmin, xmax, ymin, ymax;

  1388.     int16_t (*mv_table)[2]= s->b_direct_mv_table;

  1389. 

  1390.     c->current_mv_penalty= c->mv_penalty[1] + MAX_MV;

  1391.     ymin= xmin=(-32)>>shift;

  1392.     ymax= xmax=   31>>shift;

  1393. 

  1394.     if (IS_8X8(s->next_picture.mb_type[mot_xy])) {

  1395.         s->mv_type= MV_TYPE_8X8;

  1396.     }else{

  1397.         s->mv_type= MV_TYPE_16X16;

  1398.     }

  1399. 

  1400.     for(i=0; i<4; i++){

  1401.         int index= s->block_index[i];

  1402.         int min, max;

  1403. 

  1404.         c->co_located_mv[i][0] = s->next_picture.motion_val[0][index][0];

  1405.         c->co_located_mv[i][1] = s->next_picture.motion_val[0][index][1];

  1406.         c->direct_basis_mv[i][0]= c->co_located_mv[i][0]*time_pb/time_pp + ((i& 1)<<(shift+3));

  1407.         c->direct_basis_mv[i][1]= c->co_located_mv[i][1]*time_pb/time_pp + ((i>>1)<<(shift+3));

  1408. //        c->direct_basis_mv[1][i][0]= c->co_located_mv[i][0]*(time_pb - time_pp)/time_pp + ((i &1)<<(shift+3);

  1409. //        c->direct_basis_mv[1][i][1]= c->co_located_mv[i][1]*(time_pb - time_pp)/time_pp + ((i>>1)<<(shift+3);

  1410. 

  1411.         max= FFMAX(c->direct_basis_mv[i][0], c->direct_basis_mv[i][0] - c->co_located_mv[i][0])>>shift;

  1412.         min= FFMIN(c->direct_basis_mv[i][0], c->direct_basis_mv[i][0] - c->co_located_mv[i][0])>>shift;

  1413.         max+= 16*mb_x + 1; // +-1 is for the simpler rounding

  1414.         min+= 16*mb_x - 1;

  1415.         xmax= FFMIN(xmax, s->width - max);

  1416.         xmin= FFMAX(xmin, - 16     - min);

  1417. 

  1418.         max= FFMAX(c->direct_basis_mv[i][1], c->direct_basis_mv[i][1] - c->co_located_mv[i][1])>>shift;

  1419.         min= FFMIN(c->direct_basis_mv[i][1], c->direct_basis_mv[i][1] - c->co_located_mv[i][1])>>shift;

  1420.         max+= 16*mb_y + 1; // +-1 is for the simpler rounding

  1421.         min+= 16*mb_y - 1;

  1422.         ymax= FFMIN(ymax, s->height - max);

  1423.         ymin= FFMAX(ymin, - 16      - min);

  1424. 

  1425.         if(s->mv_type == MV_TYPE_16X16) break;

  1426.     }

  1427. 

  1428.     assert(xmax <= 15 && ymax <= 15 && xmin >= -16 && ymin >= -16);

  1429. 

  1430.     if(xmax < 0 || xmin >0 || ymax < 0 || ymin > 0){

  1431.         s->b_direct_mv_table[mot_xy][0]= 0;

  1432.         s->b_direct_mv_table[mot_xy][1]= 0;

  1433. 

  1434.         return 256*256*256*64;

  1435.     }

  1436. 

  1437.     c->xmin= xmin;

  1438.     c->ymin= ymin;

  1439.     c->xmax= xmax;

  1440.     c->ymax= ymax;

  1441.     c->flags     |= FLAG_DIRECT;

  1442.     c->sub_flags |= FLAG_DIRECT;

  1443.     c->pred_x=0;

  1444.     c->pred_y=0;

  1445. 

  1446.     P_LEFT[0]        = av_clip(mv_table[mot_xy - 1][0], xmin<<shift, xmax<<shift);

  1447.     P_LEFT[1]        = av_clip(mv_table[mot_xy - 1][1], ymin<<shift, ymax<<shift);

  1448. 

  1449.     /* special case for first line */

  1450.     if (!s->first_slice_line) { //FIXME maybe allow this over thread boundary as it is clipped

  1451.         P_TOP[0]      = av_clip(mv_table[mot_xy - mot_stride             ][0], xmin<<shift, xmax<<shift);

  1452.         P_TOP[1]      = av_clip(mv_table[mot_xy - mot_stride             ][1], ymin<<shift, ymax<<shift);

  1453.         P_TOPRIGHT[0] = av_clip(mv_table[mot_xy - mot_stride + 1         ][0], xmin<<shift, xmax<<shift);

  1454.         P_TOPRIGHT[1] = av_clip(mv_table[mot_xy - mot_stride + 1         ][1], ymin<<shift, ymax<<shift);

  1455. 

  1456.         P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);

  1457.         P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);

  1458.     }

  1459. 

  1460.     dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, 0, mv_table, 1<<(16-shift), 0, 16);

  1461.     if(c->sub_flags&FLAG_QPEL)

  1462.         dmin = qpel_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);

  1463.     else

  1464.         dmin = hpel_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);

  1465. 

  1466.     if(c->avctx->me_sub_cmp != c->avctx->mb_cmp && !c->skip)

  1467.         dmin= get_mb_score(s, mx, my, 0, 0, 0, 16, 1);

  1468. 

  1469.     get_limits(s, 16*mb_x, 16*mb_y); //restore c->?min/max, maybe not needed

  1470. 

  1471.     mv_table[mot_xy][0]= mx;

  1472.     mv_table[mot_xy][1]= my;

  1473.     c->flags     &= ~FLAG_DIRECT;

  1474.     c->sub_flags &= ~FLAG_DIRECT;

  1475. 

  1476.     return dmin;

  1477. }

  1478. 

  1479. void ff_estimate_b_frame_motion(MpegEncContext * s,

  1480.                              int mb_x, int mb_y)

  1481. {

  1482.     MotionEstContext * const c= &s->me;

  1483.     const int penalty_factor= c->mb_penalty_factor;

  1484.     int fmin, bmin, dmin, fbmin, bimin, fimin;

  1485.     int type=0;

  1486.     const int xy = mb_y*s->mb_stride + mb_x;

  1487.     init_ref(c, s->new_picture.f->data, s->last_picture.f->data,

  1488.              s->next_picture.f->data, 16 * mb_x, 16 * mb_y, 2);

  1489. 

  1490.     get_limits(s, 16*mb_x, 16*mb_y);

  1491. 

  1492.     c->skip=0;

  1493. 

  1494.     if (s->codec_id == AV_CODEC_ID_MPEG4 && s->next_picture.mbskip_table[xy]) {

  1495.         int score= direct_search(s, mb_x, mb_y); //FIXME just check 0,0

  1496. 

  1497.         score= ((unsigned)(score*score + 128*256))>>16;

  1498.         c->mc_mb_var_sum_temp += score;

  1499.         s->current_picture.mc_mb_var[mb_y*s->mb_stride + mb_x] = score; //FIXME use SSE

  1500.         s->mb_type[mb_y*s->mb_stride + mb_x]= CANDIDATE_MB_TYPE_DIRECT0;

  1501. 

  1502.         return;

  1503.     }

  1504. 

  1505.     if (s->codec_id == AV_CODEC_ID_MPEG4)

  1506.         dmin= direct_search(s, mb_x, mb_y);

  1507.     else

  1508.         dmin= INT_MAX;

  1509. // FIXME penalty stuff for non-MPEG-4

  1510.     c->skip=0;

  1511.     fmin = estimate_motion_b(s, mb_x, mb_y, s->b_forw_mv_table, 0, s->f_code) +

  1512.            3 * penalty_factor;

  1513. 

  1514.     c->skip=0;

  1515.     bmin = estimate_motion_b(s, mb_x, mb_y, s->b_back_mv_table, 2, s->b_code) +

  1516.            2 * penalty_factor;

  1517.     ff_dlog(s, " %d %d ", s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1]);

  1518. 

  1519.     c->skip=0;

  1520.     fbmin= bidir_refine(s, mb_x, mb_y) + penalty_factor;

  1521.     ff_dlog(s, "%d %d %d %d\n", dmin, fmin, bmin, fbmin);

  1522. 

  1523.     if (s->avctx->flags & AV_CODEC_FLAG_INTERLACED_ME) {

  1524. //FIXME mb type penalty

  1525.         c->skip=0;

  1526.         c->current_mv_penalty= c->mv_penalty[s->f_code] + MAX_MV;

  1527.         fimin= interlaced_search(s, 0,

  1528.                                  s->b_field_mv_table[0], s->b_field_select_table[0],

  1529.                                  s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1], 0);

  1530.         c->current_mv_penalty= c->mv_penalty[s->b_code] + MAX_MV;

  1531.         bimin= interlaced_search(s, 2,

  1532.                                  s->b_field_mv_table[1], s->b_field_select_table[1],

  1533.                                  s->b_back_mv_table[xy][0], s->b_back_mv_table[xy][1], 0);

  1534.     }else

  1535.         fimin= bimin= INT_MAX;

  1536. 

  1537.     {

  1538.         int score= fmin;

  1539.         type = CANDIDATE_MB_TYPE_FORWARD;

  1540. 

  1541.         if (dmin <= score){

  1542.             score = dmin;

  1543.             type = CANDIDATE_MB_TYPE_DIRECT;

  1544.         }

  1545.         if(bmin<score){

  1546.             score=bmin;

  1547.             type= CANDIDATE_MB_TYPE_BACKWARD;

  1548.         }

  1549.         if(fbmin<score){

  1550.             score=fbmin;

  1551.             type= CANDIDATE_MB_TYPE_BIDIR;

  1552.         }

  1553.         if(fimin<score){

  1554.             score=fimin;

  1555.             type= CANDIDATE_MB_TYPE_FORWARD_I;

  1556.         }

  1557.         if(bimin<score){

  1558.             score=bimin;

  1559.             type= CANDIDATE_MB_TYPE_BACKWARD_I;

  1560.         }

  1561. 

  1562.         score= ((unsigned)(score*score + 128*256))>>16;

  1563.         c->mc_mb_var_sum_temp += score;

  1564.         s->current_picture.mc_mb_var[mb_y*s->mb_stride + mb_x] = score; //FIXME use SSE

  1565.     }

  1566. 

  1567.     if(c->avctx->mb_decision > FF_MB_DECISION_SIMPLE){

  1568.         type= CANDIDATE_MB_TYPE_FORWARD | CANDIDATE_MB_TYPE_BACKWARD | CANDIDATE_MB_TYPE_BIDIR | CANDIDATE_MB_TYPE_DIRECT;

  1569.         if(fimin < INT_MAX)

  1570.             type |= CANDIDATE_MB_TYPE_FORWARD_I;

  1571.         if(bimin < INT_MAX)

  1572.             type |= CANDIDATE_MB_TYPE_BACKWARD_I;

  1573.         if(fimin < INT_MAX && bimin < INT_MAX){

  1574.             type |= CANDIDATE_MB_TYPE_BIDIR_I;

  1575.         }

  1576.          //FIXME something smarter

  1577.         if(dmin>256*256*16) type&= ~CANDIDATE_MB_TYPE_DIRECT; //do not try direct mode if it is invalid for this MB

  1578.         if (s->codec_id == AV_CODEC_ID_MPEG4 && type&CANDIDATE_MB_TYPE_DIRECT &&

  1579.             s->mpv_flags & FF_MPV_FLAG_MV0 && *(uint32_t*)s->b_direct_mv_table[xy])

  1580.             type |= CANDIDATE_MB_TYPE_DIRECT0;

  1581.     }

  1582. 

  1583.     s->mb_type[mb_y*s->mb_stride + mb_x]= type;

  1584. }

  1585. 

  1586. /* find best f_code for ME which do unlimited searches */

  1587. int ff_get_best_fcode(MpegEncContext * s, int16_t (*mv_table)[2], int type)

  1588. {

  1589.     if (s->motion_est != FF_ME_ZERO) {

  1590.         int score[8];

  1591.         int i, y, range= s->avctx->me_range ? s->avctx->me_range : (INT_MAX/2);

  1592.         uint8_t * fcode_tab= s->fcode_tab;

  1593.         int best_fcode=-1;

  1594.         int best_score=-10000000;

  1595. 

  1596.         if(s->msmpeg4_version)

  1597.             range= FFMIN(range, 16);

  1598.         else if(s->codec_id == AV_CODEC_ID_MPEG2VIDEO && s->avctx->strict_std_compliance >= FF_COMPLIANCE_NORMAL)

  1599.             range= FFMIN(range, 256);

  1600. 

  1601.         for(i=0; i<8; i++) score[i]= s->mb_num*(8-i);

  1602. 

  1603.         for(y=0; y<s->mb_height; y++){

  1604.             int x;

  1605.             int xy= y*s->mb_stride;

  1606.             for(x=0; x<s->mb_width; x++){

  1607.                 if(s->mb_type[xy] & type){

  1608.                     int mx= mv_table[xy][0];

  1609.                     int my= mv_table[xy][1];

  1610.                     int fcode= FFMAX(fcode_tab[mx + MAX_MV],

  1611.                                      fcode_tab[my + MAX_MV]);

  1612.                     int j;

  1613. 

  1614.                         if(mx >= range || mx < -range ||

  1615.                            my >= range || my < -range)

  1616.                             continue;

  1617. 

  1618.                     for(j=0; j<fcode && j<8; j++){

  1619.                         if(s->pict_type==AV_PICTURE_TYPE_B || s->current_picture.mc_mb_var[xy] < s->current_picture.mb_var[xy])

  1620.                             score[j]-= 170;

  1621.                     }

  1622.                 }

  1623.                 xy++;

  1624.             }

  1625.         }

  1626. 

  1627.         for(i=1; i<8; i++){

  1628.             if(score[i] > best_score){

  1629.                 best_score= score[i];

  1630.                 best_fcode= i;

  1631.             }

  1632.         }

  1633. 

  1634.         return best_fcode;

  1635.     }else{

  1636.         return 1;

  1637.     }

  1638. }

  1639. 

  1640. void ff_fix_long_p_mvs(MpegEncContext * s)

  1641. {

  1642.     MotionEstContext * const c= &s->me;

  1643.     const int f_code= s->f_code;

  1644.     int y, range;

  1645.     assert(s->pict_type==AV_PICTURE_TYPE_P);

  1646. 

  1647.     range = (((s->out_format == FMT_MPEG1 || s->msmpeg4_version) ? 8 : 16) << f_code);

  1648. 

  1649.     assert(range <= 16 || !s->msmpeg4_version);

  1650.     assert(range <=256 || !(s->codec_id == AV_CODEC_ID_MPEG2VIDEO && s->avctx->strict_std_compliance >= FF_COMPLIANCE_NORMAL));

  1651. 

  1652.     if(c->avctx->me_range && range > c->avctx->me_range) range= c->avctx->me_range;

  1653. 

  1654.     if (s->avctx->flags & AV_CODEC_FLAG_4MV) {

  1655.         const int wrap= s->b8_stride;

  1656. 

  1657.         /* clip / convert to intra 8x8 type MVs */

  1658.         for(y=0; y<s->mb_height; y++){

  1659.             int xy= y*2*wrap;

  1660.             int i= y*s->mb_stride;

  1661.             int x;

  1662. 

  1663.             for(x=0; x<s->mb_width; x++){

  1664.                 if(s->mb_type[i]&CANDIDATE_MB_TYPE_INTER4V){

  1665.                     int block;

  1666.                     for(block=0; block<4; block++){

  1667.                         int off= (block& 1) + (block>>1)*wrap;

  1668.                         int mx = s->current_picture.motion_val[0][ xy + off ][0];

  1669.                         int my = s->current_picture.motion_val[0][ xy + off ][1];

  1670. 

  1671.                         if(   mx >=range || mx <-range

  1672.                            || my >=range || my <-range){

  1673.                             s->mb_type[i] &= ~CANDIDATE_MB_TYPE_INTER4V;

  1674.                             s->mb_type[i] |= CANDIDATE_MB_TYPE_INTRA;

  1675.                             s->current_picture.mb_type[i] = CANDIDATE_MB_TYPE_INTRA;

  1676.                         }

  1677.                     }

  1678.                 }

  1679.                 xy+=2;

  1680.                 i++;

  1681.             }

  1682.         }

  1683.     }

  1684. }

  1685. 

  1686. /**

  1687.  * @param truncate 1 for truncation, 0 for using intra

  1688.  */

  1689. void ff_fix_long_mvs(MpegEncContext * s, uint8_t *field_select_table, int field_select,

  1690.                      int16_t (*mv_table)[2], int f_code, int type, int truncate)

  1691. {

  1692.     MotionEstContext * const c= &s->me;

  1693.     int y, h_range, v_range;

  1694. 

  1695.     // RAL: 8 in MPEG-1, 16 in MPEG-4

  1696.     int range = (((s->out_format == FMT_MPEG1 || s->msmpeg4_version) ? 8 : 16) << f_code);

  1697. 

  1698.     if(c->avctx->me_range && range > c->avctx->me_range) range= c->avctx->me_range;

  1699. 

  1700.     h_range= range;

  1701.     v_range= field_select_table ? range>>1 : range;

  1702. 

  1703.     /* clip / convert to intra 16x16 type MVs */

  1704.     for(y=0; y<s->mb_height; y++){

  1705.         int x;

  1706.         int xy= y*s->mb_stride;

  1707.         for(x=0; x<s->mb_width; x++){

  1708.             if (s->mb_type[xy] & type){    // RAL: "type" test added...

  1709.                 if (!field_select_table || field_select_table[xy] == field_select) {

  1710.                     if(   mv_table[xy][0] >=h_range || mv_table[xy][0] <-h_range

  1711.                        || mv_table[xy][1] >=v_range || mv_table[xy][1] <-v_range){

  1712. 

  1713.                         if(truncate){

  1714.                             if     (mv_table[xy][0] > h_range-1) mv_table[xy][0]=  h_range-1;

  1715.                             else if(mv_table[xy][0] < -h_range ) mv_table[xy][0]= -h_range;

  1716.                             if     (mv_table[xy][1] > v_range-1) mv_table[xy][1]=  v_range-1;

  1717.                             else if(mv_table[xy][1] < -v_range ) mv_table[xy][1]= -v_range;

  1718.                         }else{

  1719.                             s->mb_type[xy] &= ~type;

  1720.                             s->mb_type[xy] |= CANDIDATE_MB_TYPE_INTRA;

  1721.                             mv_table[xy][0]=

  1722.                             mv_table[xy][1]= 0;

  1723.                         }

  1724.                     }

  1725.                 }

  1726.             }

  1727.             xy++;

  1728.         }

  1729.     }

  1730. }