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.
3088 Commits
32 Branches
293MB
Tree: ed263031f6
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'ed263031f6'
${ noResults }
FFmpeg/libavcodec/motion_est.c

1893 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.  *

  7.  * This library is free software; you can redistribute it and/or

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

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

  10.  * version 2 of the License, or (at your option) any later version.

  11.  *

  12.  * This library is distributed in the hope that it will be useful,

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

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

  15.  * Lesser General Public License for more details.

  16.  *

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

  18.  * License along with this library; if not, write to the Free Software

  19.  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA

  20.  *

  21.  * new Motion Estimation (X1/EPZS) by Michael Niedermayer <michaelni@gmx.at>

  22.  */

  23.  

  24. /**

  25.  * @file motion_est.c

  26.  * Motion estimation.

  27.  */

  28.  

  29. #include <stdlib.h>

  30. #include <stdio.h>

  31. #include <limits.h>

  32. #include "avcodec.h"

  33. #include "dsputil.h"

  34. #include "mpegvideo.h"

  35. 

  36. //#undef NDEBUG

  37. //#include <assert.h>

  38. 

  39. #define SQ(a) ((a)*(a))

  40. 

  41. #define P_LEFT P[1]

  42. #define P_TOP P[2]

  43. #define P_TOPRIGHT P[3]

  44. #define P_MEDIAN P[4]

  45. #define P_MV1 P[9]

  46. 

  47. static inline int sad_hpel_motion_search(MpegEncContext * s,

  48. 				  int *mx_ptr, int *my_ptr, int dmin,

  49.                                   int pred_x, int pred_y, uint8_t *src_data[3],

  50.                                   uint8_t *ref_data[6], int stride, int uvstride,

  51.                                   int size, int h, uint8_t * const mv_penalty);

  52. 

  53. static inline int update_map_generation(MpegEncContext * s)

  54. {

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

  56.     if(s->me.map_generation==0){

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

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

  59.     }

  60.     return s->me.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. /* SIMPLE */

  78. #define RENAME(a) simple_ ## a

  79. 

  80. #define CMP(d, x, y, size)\

  81. d = cmp(s, src_y, (ref_y) + (x) + (y)*(stride), stride, h);

  82. 

  83. #define CMP_HPEL(d, dx, dy, x, y, size)\

  84. {\

  85.     const int dxy= (dx) + 2*(dy);\

  86.     hpel_put[0][dxy](s->me.scratchpad, (ref_y) + (x) + (y)*(stride), stride, h);\

  87.     d = cmp_sub(s, s->me.scratchpad, src_y, stride, h);\

  88. }

  89. 

  90. 

  91. #define CMP_QPEL(d, dx, dy, x, y, size)\

  92. {\

  93.     const int dxy= (dx) + 4*(dy);\

  94.     qpel_put[0][dxy](s->me.scratchpad, (ref_y) + (x) + (y)*(stride), stride);\

  95.     d = cmp_sub(s, s->me.scratchpad, src_y, stride, h);\

  96. }

  97. 

  98. #include "motion_est_template.c"

  99. #undef RENAME

  100. #undef CMP

  101. #undef CMP_HPEL

  102. #undef CMP_QPEL

  103. #undef INIT

  104. 

  105. /* SIMPLE CHROMA */

  106. #define RENAME(a) simple_chroma_ ## a

  107. 

  108. #define CMP(d, x, y, size)\

  109. d = cmp(s, src_y, (ref_y) + (x) + (y)*(stride), stride, h);\

  110. if(chroma_cmp){\

  111.     int dxy= ((x)&1) + 2*((y)&1);\

  112.     int c= ((x)>>1) + ((y)>>1)*uvstride;\

  113. \

  114.     chroma_hpel_put[0][dxy](s->me.scratchpad, ref_u + c, uvstride, h>>1);\

  115.     d += chroma_cmp(s, s->me.scratchpad, src_u, uvstride, h>>1);\

  116.     chroma_hpel_put[0][dxy](s->me.scratchpad, ref_v + c, uvstride, h>>1);\

  117.     d += chroma_cmp(s, s->me.scratchpad, src_v, uvstride, h>>1);\

  118. }

  119. 

  120. #define CMP_HPEL(d, dx, dy, x, y, size)\

  121. {\

  122.     const int dxy= (dx) + 2*(dy);\

  123.     hpel_put[0][dxy](s->me.scratchpad, (ref_y) + (x) + (y)*(stride), stride, h);\

  124.     d = cmp_sub(s, s->me.scratchpad, src_y, stride, h);\

  125.     if(chroma_cmp_sub){\

  126.         int cxy= (dxy) | ((x)&1) | (2*((y)&1));\

  127.         int c= ((x)>>1) + ((y)>>1)*uvstride;\

  128.         chroma_hpel_put[0][cxy](s->me.scratchpad, ref_u + c, uvstride, h>>1);\

  129.         d += chroma_cmp_sub(s, s->me.scratchpad, src_u, uvstride, h>>1);\

  130.         chroma_hpel_put[0][cxy](s->me.scratchpad, ref_v + c, uvstride, h>>1);\

  131.         d += chroma_cmp_sub(s, s->me.scratchpad, src_v, uvstride, h>>1);\

  132.     }\

  133. }

  134. 

  135. #define CMP_QPEL(d, dx, dy, x, y, size)\

  136. {\

  137.     const int dxy= (dx) + 4*(dy);\

  138.     qpel_put[0][dxy](s->me.scratchpad, (ref_y) + (x) + (y)*(stride), stride);\

  139.     d = cmp_sub(s, s->me.scratchpad, src_y, stride, h);\

  140.     if(chroma_cmp_sub){\

  141.         int cxy, c;\

  142.         int cx= (4*(x) + (dx))/2;\

  143.         int cy= (4*(y) + (dy))/2;\

  144.         cx= (cx>>1)|(cx&1);\

  145.         cy= (cy>>1)|(cy&1);\

  146.         cxy= (cx&1) + 2*(cy&1);\

  147.         c= ((cx)>>1) + ((cy)>>1)*uvstride;\

  148.         chroma_hpel_put[0][cxy](s->me.scratchpad, ref_u + c, uvstride, h>>1);\

  149.         d += chroma_cmp_sub(s, s->me.scratchpad, src_u, uvstride, h>>1);\

  150.         chroma_hpel_put[0][cxy](s->me.scratchpad, ref_v + c, uvstride, h>>1);\

  151.         d += chroma_cmp_sub(s, s->me.scratchpad, src_v, uvstride, h>>1);\

  152.     }\

  153. }

  154. 

  155. #include "motion_est_template.c"

  156. #undef RENAME

  157. #undef CMP

  158. #undef CMP_HPEL

  159. #undef CMP_QPEL

  160. #undef INIT

  161. 

  162. /* SIMPLE DIRECT HPEL */

  163. #define RENAME(a) simple_direct_hpel_ ## a

  164. //FIXME precalc divisions stuff

  165. 

  166. #define CMP_DIRECT(d, dx, dy, x, y, size, cmp_func)\

  167. if((x) >= xmin && 2*(x) + (dx) <= 2*xmax && (y) >= ymin && 2*(y) + (dy) <= 2*ymax){\

  168.     const int hx= 2*(x) + (dx);\

  169.     const int hy= 2*(y) + (dy);\

  170.     if(s->mv_type==MV_TYPE_8X8){\

  171.         int i;\

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

  173.             int fx = s->me.direct_basis_mv[i][0] + hx;\

  174.             int fy = s->me.direct_basis_mv[i][1] + hy;\

  175.             int bx = hx ? fx - s->me.co_located_mv[i][0] : s->me.co_located_mv[i][0]*(time_pb - time_pp)/time_pp + (i &1)*16;\

  176.             int by = hy ? fy - s->me.co_located_mv[i][1] : s->me.co_located_mv[i][1]*(time_pb - time_pp)/time_pp + (i>>1)*16;\

  177.             int fxy= (fx&1) + 2*(fy&1);\

  178.             int bxy= (bx&1) + 2*(by&1);\

  179. \

  180.             uint8_t *dst= s->me.scratchpad + 8*(i&1) + 8*stride*(i>>1);\

  181.             hpel_put[1][fxy](dst, (ref_y ) + (fx>>1) + (fy>>1)*(stride), stride, 8);\

  182.             hpel_avg[1][bxy](dst, (ref_data[3]) + (bx>>1) + (by>>1)*(stride), stride, 8);\

  183.         }\

  184.     }else{\

  185.         int fx = s->me.direct_basis_mv[0][0] + hx;\

  186.         int fy = s->me.direct_basis_mv[0][1] + hy;\

  187.         int bx = hx ? fx - s->me.co_located_mv[0][0] : (s->me.co_located_mv[0][0]*(time_pb - time_pp)/time_pp);\

  188.         int by = hy ? fy - s->me.co_located_mv[0][1] : (s->me.co_located_mv[0][1]*(time_pb - time_pp)/time_pp);\

  189.         int fxy= (fx&1) + 2*(fy&1);\

  190.         int bxy= (bx&1) + 2*(by&1);\

  191.         \

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

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

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

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

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

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

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

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

  200. \

  201.         hpel_put[0][fxy](s->me.scratchpad, (ref_y ) + (fx>>1) + (fy>>1)*(stride), stride, 16);\

  202.         hpel_avg[0][bxy](s->me.scratchpad, (ref_data[3]) + (bx>>1) + (by>>1)*(stride), stride, 16);\

  203.     }\

  204.     d = cmp_func(s, s->me.scratchpad, src_y, stride, 16);\

  205. }else\

  206.     d= 256*256*256*32;

  207. 

  208. 

  209. #define CMP_HPEL(d, dx, dy, x, y, size)\

  210.     CMP_DIRECT(d, dx, dy, x, y, size, cmp_sub)

  211. 

  212. #define CMP(d, x, y, size)\

  213.     CMP_DIRECT(d, 0, 0, x, y, size, cmp)

  214.     

  215. #include "motion_est_template.c"

  216. #undef RENAME

  217. #undef CMP

  218. #undef CMP_HPEL

  219. #undef CMP_QPEL

  220. #undef INIT

  221. #undef CMP_DIRECT

  222. 

  223. /* SIMPLE DIRECT QPEL */

  224. #define RENAME(a) simple_direct_qpel_ ## a

  225. 

  226. #define CMP_DIRECT(d, dx, dy, x, y, size, cmp_func)\

  227. if((x) >= xmin && 4*(x) + (dx) <= 4*xmax && (y) >= ymin && 4*(y) + (dy) <= 4*ymax){\

  228.     const int qx= 4*(x) + (dx);\

  229.     const int qy= 4*(y) + (dy);\

  230.     if(s->mv_type==MV_TYPE_8X8){\

  231.         int i;\

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

  233.             int fx = s->me.direct_basis_mv[i][0] + qx;\

  234.             int fy = s->me.direct_basis_mv[i][1] + qy;\

  235.             int bx = qx ? fx - s->me.co_located_mv[i][0] : s->me.co_located_mv[i][0]*(time_pb - time_pp)/time_pp + (i &1)*16;\

  236.             int by = qy ? fy - s->me.co_located_mv[i][1] : s->me.co_located_mv[i][1]*(time_pb - time_pp)/time_pp + (i>>1)*16;\

  237.             int fxy= (fx&3) + 4*(fy&3);\

  238.             int bxy= (bx&3) + 4*(by&3);\

  239. \

  240.             uint8_t *dst= s->me.scratchpad + 8*(i&1) + 8*stride*(i>>1);\

  241.             qpel_put[1][fxy](dst, (ref_y ) + (fx>>2) + (fy>>2)*(stride), stride);\

  242.             qpel_avg[1][bxy](dst, (ref_data[3]) + (bx>>2) + (by>>2)*(stride), stride);\

  243.         }\

  244.     }else{\

  245.         int fx = s->me.direct_basis_mv[0][0] + qx;\

  246.         int fy = s->me.direct_basis_mv[0][1] + qy;\

  247.         int bx = qx ? fx - s->me.co_located_mv[0][0] : s->me.co_located_mv[0][0]*(time_pb - time_pp)/time_pp;\

  248.         int by = qy ? fy - s->me.co_located_mv[0][1] : s->me.co_located_mv[0][1]*(time_pb - time_pp)/time_pp;\

  249.         int fxy= (fx&3) + 4*(fy&3);\

  250.         int bxy= (bx&3) + 4*(by&3);\

  251. \

  252.         qpel_put[1][fxy](s->me.scratchpad               , (ref_y ) + (fx>>2) + (fy>>2)*(stride)               , stride);\

  253.         qpel_put[1][fxy](s->me.scratchpad + 8           , (ref_y ) + (fx>>2) + (fy>>2)*(stride) + 8           , stride);\

  254.         qpel_put[1][fxy](s->me.scratchpad     + 8*stride, (ref_y ) + (fx>>2) + (fy>>2)*(stride)     + 8*stride, stride);\

  255.         qpel_put[1][fxy](s->me.scratchpad + 8 + 8*stride, (ref_y ) + (fx>>2) + (fy>>2)*(stride) + 8 + 8*stride, stride);\

  256.         qpel_avg[1][bxy](s->me.scratchpad               , (ref_data[3]) + (bx>>2) + (by>>2)*(stride)               , stride);\

  257.         qpel_avg[1][bxy](s->me.scratchpad + 8           , (ref_data[3]) + (bx>>2) + (by>>2)*(stride) + 8           , stride);\

  258.         qpel_avg[1][bxy](s->me.scratchpad     + 8*stride, (ref_data[3]) + (bx>>2) + (by>>2)*(stride)     + 8*stride, stride);\

  259.         qpel_avg[1][bxy](s->me.scratchpad + 8 + 8*stride, (ref_data[3]) + (bx>>2) + (by>>2)*(stride) + 8 + 8*stride, stride);\

  260.     }\

  261.     d = cmp_func(s, s->me.scratchpad, src_y, stride, 16);\

  262. }else\

  263.     d= 256*256*256*32;

  264. 

  265. 

  266. #define CMP_QPEL(d, dx, dy, x, y, size)\

  267.     CMP_DIRECT(d, dx, dy, x, y, size, cmp_sub)

  268. 

  269. #define CMP(d, x, y, size)\

  270.     CMP_DIRECT(d, 0, 0, x, y, size, cmp)

  271. 

  272. #include "motion_est_template.c"

  273. #undef RENAME

  274. #undef CMP

  275. #undef CMP_HPEL

  276. #undef CMP_QPEL

  277. #undef INIT

  278. #undef CMP__DIRECT

  279. 

  280. static inline int get_penalty_factor(MpegEncContext *s, int type){

  281.     switch(type&0xFF){

  282.     default:

  283.     case FF_CMP_SAD:

  284.         return s->qscale*2;

  285.     case FF_CMP_DCT:

  286.         return s->qscale*3;

  287.     case FF_CMP_SATD:

  288.         return s->qscale*6;

  289.     case FF_CMP_SSE:

  290.         return s->qscale*s->qscale*2;

  291.     case FF_CMP_BIT:

  292.         return 1;

  293.     case FF_CMP_RD:

  294.     case FF_CMP_PSNR:

  295.         return (s->qscale*s->qscale*185 + 64)>>7;

  296.     }

  297. }

  298. 

  299. void ff_init_me(MpegEncContext *s){

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

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

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

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

  304. 

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

  306.         if(s->avctx->me_sub_cmp&FF_CMP_CHROMA)

  307.             s->me.sub_motion_search= simple_chroma_qpel_motion_search;

  308.         else

  309.             s->me.sub_motion_search= simple_qpel_motion_search;

  310.     }else{

  311.         if(s->avctx->me_sub_cmp&FF_CMP_CHROMA)

  312.             s->me.sub_motion_search= simple_chroma_hpel_motion_search;

  313.         else if(   s->avctx->me_sub_cmp == FF_CMP_SAD 

  314.                 && s->avctx->    me_cmp == FF_CMP_SAD 

  315.                 && s->avctx->    mb_cmp == FF_CMP_SAD)

  316.             s->me.sub_motion_search= sad_hpel_motion_search; // 2050 vs. 2450 cycles

  317.         else

  318.             s->me.sub_motion_search= simple_hpel_motion_search;

  319.     }

  320. 

  321.     if(s->avctx->me_cmp&FF_CMP_CHROMA){

  322.         s->me.motion_search[0]= simple_chroma_epzs_motion_search;

  323.         s->me.motion_search[1]= simple_chroma_epzs_motion_search4;

  324.         s->me.motion_search[4]= simple_chroma_epzs_motion_search2;

  325.     }else{

  326.         s->me.motion_search[0]= simple_epzs_motion_search;

  327.         s->me.motion_search[1]= simple_epzs_motion_search4;

  328.         s->me.motion_search[4]= simple_epzs_motion_search2;

  329.     }

  330.     

  331.     if(s->avctx->me_pre_cmp&FF_CMP_CHROMA){

  332.         s->me.pre_motion_search= simple_chroma_epzs_motion_search;

  333.     }else{

  334.         s->me.pre_motion_search= simple_epzs_motion_search;

  335.     }

  336.     

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

  338.         if(s->avctx->mb_cmp&FF_CMP_CHROMA)

  339.             s->me.get_mb_score= simple_chroma_qpel_get_mb_score;

  340.         else

  341.             s->me.get_mb_score= simple_qpel_get_mb_score;

  342.     }else{

  343.         if(s->avctx->mb_cmp&FF_CMP_CHROMA)

  344.             s->me.get_mb_score= simple_chroma_hpel_get_mb_score;

  345.         else

  346.             s->me.get_mb_score= simple_hpel_get_mb_score;

  347.     }

  348. }

  349.       

  350. #if 0

  351. static int pix_dev(uint8_t * pix, int line_size, int mean)

  352. {

  353.     int s, i, j;

  354. 

  355.     s = 0;

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

  357. 	for (j = 0; j < 16; j += 8) {

  358. 	    s += ABS(pix[0]-mean);

  359. 	    s += ABS(pix[1]-mean);

  360. 	    s += ABS(pix[2]-mean);

  361. 	    s += ABS(pix[3]-mean);

  362. 	    s += ABS(pix[4]-mean);

  363. 	    s += ABS(pix[5]-mean);

  364. 	    s += ABS(pix[6]-mean);

  365. 	    s += ABS(pix[7]-mean);

  366. 	    pix += 8;

  367. 	}

  368. 	pix += line_size - 16;

  369.     }

  370.     return s;

  371. }

  372. #endif

  373. 

  374. static inline void no_motion_search(MpegEncContext * s,

  375. 				    int *mx_ptr, int *my_ptr)

  376. {

  377.     *mx_ptr = 16 * s->mb_x;

  378.     *my_ptr = 16 * s->mb_y;

  379. }

  380. 

  381. static int full_motion_search(MpegEncContext * s,

  382.                               int *mx_ptr, int *my_ptr, int range,

  383.                               int xmin, int ymin, int xmax, int ymax, uint8_t *ref_picture)

  384. {

  385.     int x1, y1, x2, y2, xx, yy, x, y;

  386.     int mx, my, dmin, d;

  387.     uint8_t *pix;

  388. 

  389.     xx = 16 * s->mb_x;

  390.     yy = 16 * s->mb_y;

  391.     x1 = xx - range + 1;	/* we loose one pixel to avoid boundary pb with half pixel pred */

  392.     if (x1 < xmin)

  393. 	x1 = xmin;

  394.     x2 = xx + range - 1;

  395.     if (x2 > xmax)

  396. 	x2 = xmax;

  397.     y1 = yy - range + 1;

  398.     if (y1 < ymin)

  399. 	y1 = ymin;

  400.     y2 = yy + range - 1;

  401.     if (y2 > ymax)

  402. 	y2 = ymax;

  403.     pix = s->new_picture.data[0] + (yy * s->linesize) + xx;

  404.     dmin = 0x7fffffff;

  405.     mx = 0;

  406.     my = 0;

  407.     for (y = y1; y <= y2; y++) {

  408. 	for (x = x1; x <= x2; x++) {

  409. 	    d = s->dsp.pix_abs[0][0](NULL, pix, ref_picture + (y * s->linesize) + x,

  410. 			     s->linesize, 16);

  411. 	    if (d < dmin ||

  412. 		(d == dmin &&

  413. 		 (abs(x - xx) + abs(y - yy)) <

  414. 		 (abs(mx - xx) + abs(my - yy)))) {

  415. 		dmin = d;

  416. 		mx = x;

  417. 		my = y;

  418. 	    }

  419. 	}

  420.     }

  421. 

  422.     *mx_ptr = mx;

  423.     *my_ptr = my;

  424. 

  425. #if 0

  426.     if (*mx_ptr < -(2 * range) || *mx_ptr >= (2 * range) ||

  427. 	*my_ptr < -(2 * range) || *my_ptr >= (2 * range)) {

  428. 	fprintf(stderr, "error %d %d\n", *mx_ptr, *my_ptr);

  429.     }

  430. #endif

  431.     return dmin;

  432. }

  433. 

  434. 

  435. static int log_motion_search(MpegEncContext * s,

  436.                              int *mx_ptr, int *my_ptr, int range,

  437.                              int xmin, int ymin, int xmax, int ymax, uint8_t *ref_picture)

  438. {

  439.     int x1, y1, x2, y2, xx, yy, x, y;

  440.     int mx, my, dmin, d;

  441.     uint8_t *pix;

  442. 

  443.     xx = s->mb_x << 4;

  444.     yy = s->mb_y << 4;

  445. 

  446.     /* Left limit */

  447.     x1 = xx - range;

  448.     if (x1 < xmin)

  449. 	x1 = xmin;

  450. 

  451.     /* Right limit */

  452.     x2 = xx + range;

  453.     if (x2 > xmax)

  454. 	x2 = xmax;

  455. 

  456.     /* Upper limit */

  457.     y1 = yy - range;

  458.     if (y1 < ymin)

  459. 	y1 = ymin;

  460. 

  461.     /* Lower limit */

  462.     y2 = yy + range;

  463.     if (y2 > ymax)

  464. 	y2 = ymax;

  465. 

  466.     pix = s->new_picture.data[0] + (yy * s->linesize) + xx;

  467.     dmin = 0x7fffffff;

  468.     mx = 0;

  469.     my = 0;

  470. 

  471.     do {

  472. 	for (y = y1; y <= y2; y += range) {

  473. 	    for (x = x1; x <= x2; x += range) {

  474. 		d = s->dsp.pix_abs[0][0](NULL, pix, ref_picture + (y * s->linesize) + x, s->linesize, 16);

  475. 		if (d < dmin || (d == dmin && (abs(x - xx) + abs(y - yy)) < (abs(mx - xx) + abs(my - yy)))) {

  476. 		    dmin = d;

  477. 		    mx = x;

  478. 		    my = y;

  479. 		}

  480. 	    }

  481. 	}

  482. 

  483. 	range = range >> 1;

  484. 

  485. 	x1 = mx - range;

  486. 	if (x1 < xmin)

  487. 	    x1 = xmin;

  488. 

  489. 	x2 = mx + range;

  490. 	if (x2 > xmax)

  491. 	    x2 = xmax;

  492. 

  493. 	y1 = my - range;

  494. 	if (y1 < ymin)

  495. 	    y1 = ymin;

  496. 

  497. 	y2 = my + range;

  498. 	if (y2 > ymax)

  499. 	    y2 = ymax;

  500. 

  501.     } while (range >= 1);

  502. 

  503. #ifdef DEBUG

  504.     fprintf(stderr, "log       - MX: %d\tMY: %d\n", mx, my);

  505. #endif

  506.     *mx_ptr = mx;

  507.     *my_ptr = my;

  508.     return dmin;

  509. }

  510. 

  511. static int phods_motion_search(MpegEncContext * s,

  512.                                int *mx_ptr, int *my_ptr, int range,

  513.                                int xmin, int ymin, int xmax, int ymax, uint8_t *ref_picture)

  514. {

  515.     int x1, y1, x2, y2, xx, yy, x, y, lastx, d;

  516.     int mx, my, dminx, dminy;

  517.     uint8_t *pix;

  518. 

  519.     xx = s->mb_x << 4;

  520.     yy = s->mb_y << 4;

  521. 

  522.     /* Left limit */

  523.     x1 = xx - range;

  524.     if (x1 < xmin)

  525. 	x1 = xmin;

  526. 

  527.     /* Right limit */

  528.     x2 = xx + range;

  529.     if (x2 > xmax)

  530. 	x2 = xmax;

  531. 

  532.     /* Upper limit */

  533.     y1 = yy - range;

  534.     if (y1 < ymin)

  535. 	y1 = ymin;

  536. 

  537.     /* Lower limit */

  538.     y2 = yy + range;

  539.     if (y2 > ymax)

  540. 	y2 = ymax;

  541. 

  542.     pix = s->new_picture.data[0] + (yy * s->linesize) + xx;

  543.     mx = 0;

  544.     my = 0;

  545. 

  546.     x = xx;

  547.     y = yy;

  548.     do {

  549.         dminx = 0x7fffffff;

  550.         dminy = 0x7fffffff;

  551. 

  552. 	lastx = x;

  553. 	for (x = x1; x <= x2; x += range) {

  554. 	    d = s->dsp.pix_abs[0][0](NULL, pix, ref_picture + (y * s->linesize) + x, s->linesize, 16);

  555. 	    if (d < dminx || (d == dminx && (abs(x - xx) + abs(y - yy)) < (abs(mx - xx) + abs(my - yy)))) {

  556. 		dminx = d;

  557. 		mx = x;

  558. 	    }

  559. 	}

  560. 

  561. 	x = lastx;

  562. 	for (y = y1; y <= y2; y += range) {

  563. 	    d = s->dsp.pix_abs[0][0](NULL, pix, ref_picture + (y * s->linesize) + x, s->linesize, 16);

  564. 	    if (d < dminy || (d == dminy && (abs(x - xx) + abs(y - yy)) < (abs(mx - xx) + abs(my - yy)))) {

  565. 		dminy = d;

  566. 		my = y;

  567. 	    }

  568. 	}

  569. 

  570. 	range = range >> 1;

  571. 

  572. 	x = mx;

  573. 	y = my;

  574. 	x1 = mx - range;

  575. 	if (x1 < xmin)

  576. 	    x1 = xmin;

  577. 

  578. 	x2 = mx + range;

  579. 	if (x2 > xmax)

  580. 	    x2 = xmax;

  581. 

  582. 	y1 = my - range;

  583. 	if (y1 < ymin)

  584. 	    y1 = ymin;

  585. 

  586. 	y2 = my + range;

  587. 	if (y2 > ymax)

  588. 	    y2 = ymax;

  589. 

  590.     } while (range >= 1);

  591. 

  592. #ifdef DEBUG

  593.     fprintf(stderr, "phods     - MX: %d\tMY: %d\n", mx, my);

  594. #endif

  595. 

  596.     /* half pixel search */

  597.     *mx_ptr = mx;

  598.     *my_ptr = my;

  599.     return dminy;

  600. }

  601. 

  602. 

  603. #define Z_THRESHOLD 256

  604. 

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

  606. {\

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

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

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

  610. }

  611. 

  612. static inline int sad_hpel_motion_search(MpegEncContext * s,

  613. 				  int *mx_ptr, int *my_ptr, int dmin,

  614.                                   int pred_x, int pred_y, uint8_t *src_data[3],

  615.                                   uint8_t *ref_data[6], int stride, int uvstride,                                  

  616.                                   int size, int h, uint8_t * const mv_penalty)

  617. {

  618.     uint32_t *score_map= s->me.score_map;

  619.     const int penalty_factor= s->me.sub_penalty_factor;

  620.     int mx, my, dminh;

  621.     uint8_t *pix, *ptr;

  622.     const int xmin= s->me.xmin;

  623.     const int ymin= s->me.ymin;

  624.     const int xmax= s->me.xmax;

  625.     const int ymax= s->me.ymax;

  626. 

  627.     if(s->me.skip){

  628. //    printf("S");

  629.         *mx_ptr = 0;

  630.         *my_ptr = 0;

  631.         return dmin;

  632.     }

  633. //    printf("N");

  634.         

  635.     pix = src_data[0];

  636. 

  637.     mx = *mx_ptr;

  638.     my = *my_ptr;

  639.     ptr = ref_data[0] + (my * stride) + mx;

  640.     

  641.     dminh = dmin;

  642. 

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

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

  645.         int dx=0, dy=0;

  646.         int d, pen_x, pen_y; 

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

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

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

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

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

  652.         mx<<=1;

  653.         my<<=1;

  654. 

  655.         

  656.         pen_x= pred_x + mx;

  657.         pen_y= pred_y + my;

  658. 

  659.         ptr-= stride;

  660.         if(t<=b){

  661.             CHECK_SAD_HALF_MV(y2 , 0, -1)

  662.             if(l<=r){

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

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

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

  666.                     ptr+= stride;

  667.                 }else{

  668.                     ptr+= stride;

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

  670.                 }

  671.                 CHECK_SAD_HALF_MV(x2 , -1,  0)

  672.             }else{

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

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

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

  676.                     ptr+= stride;

  677.                 }else{

  678.                     ptr+= stride;

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

  680.                 }

  681.                 CHECK_SAD_HALF_MV(x2 , +1,  0)

  682.             }

  683.         }else{

  684.             if(l<=r){

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

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

  687.                     ptr+= stride;

  688.                 }else{

  689.                     ptr+= stride;

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

  691.                 }

  692.                 CHECK_SAD_HALF_MV(x2 , -1,  0)

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

  694.             }else{

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

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

  697.                     ptr+= stride;

  698.                 }else{

  699.                     ptr+= stride;

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

  701.                 }

  702.                 CHECK_SAD_HALF_MV(x2 , +1,  0)

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

  704.             }

  705.             CHECK_SAD_HALF_MV(y2 ,  0, +1)

  706.         }

  707.         mx+=dx;

  708.         my+=dy;

  709. 

  710.     }else{

  711.         mx<<=1;

  712.         my<<=1;

  713.     }

  714. 

  715.     *mx_ptr = mx;

  716.     *my_ptr = my;

  717.     return dminh;

  718. }

  719. 

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

  721. {

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

  723.     

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

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

  726. 

  727.     /* has allready been set to the 4 MV if 4MV is done */

  728.     if(mv4){

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

  730. 

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

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

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

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

  735. 

  736.         mot_xy += s->block_wrap[0];

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

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

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

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

  741.     }

  742. }

  743. 

  744. /**

  745.  * get fullpel ME search limits.

  746.  */

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

  748. {

  749. /*

  750.     if(s->avctx->me_range) s->me.range= s->avctx->me_range >> 1;

  751.     else                   s->me.range= 16;

  752. */

  753.     if (s->unrestricted_mv) {

  754.         s->me.xmin = - x - 16;

  755.         s->me.ymin = - y - 16;

  756.         s->me.xmax = - x + s->mb_width *16;

  757.         s->me.ymax = - y + s->mb_height*16;

  758.     } else {

  759.         s->me.xmin = - x;

  760.         s->me.ymin = - y;

  761.         s->me.xmax = - x + s->mb_width *16 - 16;

  762.         s->me.ymax = - y + s->mb_height*16 - 16;

  763.     }

  764. }

  765. 

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

  767. {

  768.     const int size= 1;

  769.     const int h=8;

  770.     int block;

  771.     int P[10][2];

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

  773.     uint8_t * const mv_penalty= s->me.mv_penalty[s->f_code] + MAX_MV;

  774.     int same=1;

  775.     const int stride= s->linesize;

  776.     const int uvstride= s->uvlinesize;

  777.     const int xmin= s->me.xmin;

  778.     const int ymin= s->me.ymin;

  779.     const int xmax= s->me.xmax;

  780.     const int ymax= s->me.ymax;

  781. 

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

  783.         int mx4, my4;

  784.         int pred_x4, pred_y4;

  785.         int dmin4;

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

  787.         const int mot_stride = s->block_wrap[0];

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

  789.         const int block_x= (block&1);

  790.         const int block_y= (block>>1);

  791.         uint8_t *src_data[3]= {

  792.             s->new_picture.data[0] + 8*(2*s->mb_x + block_x) + stride  *8*(2*s->mb_y + block_y), //FIXME chroma?

  793.             s->new_picture.data[1] + 4*(2*s->mb_x + block_x) + uvstride*4*(2*s->mb_y + block_y),

  794.             s->new_picture.data[2] + 4*(2*s->mb_x + block_x) + uvstride*4*(2*s->mb_y + block_y)

  795.         };

  796.         uint8_t *ref_data[3]= {

  797.             s->last_picture.data[0] + 8*(2*s->mb_x + block_x) + stride  *8*(2*s->mb_y + block_y), //FIXME chroma?

  798.             s->last_picture.data[1] + 4*(2*s->mb_x + block_x) + uvstride*4*(2*s->mb_y + block_y),

  799.             s->last_picture.data[2] + 4*(2*s->mb_x + block_x) + uvstride*4*(2*s->mb_y + block_y)

  800.         };

  801. 

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

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

  804. 

  805.         if(P_LEFT[0]       > (s->me.xmax<<shift)) P_LEFT[0]       = (s->me.xmax<<shift);

  806. 

  807.         /* special case for first line */

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

  809.             pred_x4= P_LEFT[0];

  810.             pred_y4= P_LEFT[1];

  811.         } else {

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

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

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

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

  816.             if(P_TOP[1]      > (s->me.ymax<<shift)) P_TOP[1]     = (s->me.ymax<<shift);

  817.             if(P_TOPRIGHT[0] < (s->me.xmin<<shift)) P_TOPRIGHT[0]= (s->me.xmin<<shift);

  818.             if(P_TOPRIGHT[0] > (s->me.xmax<<shift)) P_TOPRIGHT[0]= (s->me.xmax<<shift);

  819.             if(P_TOPRIGHT[1] > (s->me.ymax<<shift)) P_TOPRIGHT[1]= (s->me.ymax<<shift);

  820.     

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

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

  823. 

  824. //            if(s->out_format == FMT_H263){

  825.                 pred_x4 = P_MEDIAN[0];

  826.                 pred_y4 = P_MEDIAN[1];

  827. #if 0

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

  829.                 pred_x4= P_LEFT[0];

  830.                 pred_y4= P_LEFT[1];

  831.             }

  832. #endif

  833.         }

  834.         P_MV1[0]= mx;

  835.         P_MV1[1]= my;

  836. 

  837.         dmin4 = s->me.motion_search[1](s, &mx4, &my4, P, pred_x4, pred_y4, 

  838.                                        src_data, ref_data, stride, uvstride, s->p_mv_table, (1<<16)>>shift, mv_penalty);

  839. 

  840.         dmin4= s->me.sub_motion_search(s, &mx4, &my4, dmin4, 

  841. 					  pred_x4, pred_y4, src_data, ref_data, stride, uvstride, size, h, mv_penalty);

  842.         

  843.         if(s->dsp.me_sub_cmp[0] != s->dsp.mb_cmp[0]

  844.            && s->avctx->mb_decision == FF_MB_DECISION_SIMPLE){

  845.             int dxy;

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

  847.             uint8_t *dest_y = s->me.scratchpad + offset;

  848.             if(s->quarter_sample){

  849.                 uint8_t *ref= ref_data[0] + (mx4>>2) + (my4>>2)*stride;

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

  851. 

  852.                 if(s->no_rounding)

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

  854.                 else

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

  856.             }else{

  857.                 uint8_t *ref= ref_data[0] + (mx4>>1) + (my4>>1)*stride;

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

  859. 

  860.                 if(s->no_rounding)

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

  862.                 else

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

  864.             }

  865.             dmin_sum+= (mv_penalty[mx4-pred_x4] + mv_penalty[my4-pred_y4])*s->me.mb_penalty_factor;

  866.         }else

  867.             dmin_sum+= dmin4;

  868. 

  869.         if(s->quarter_sample){

  870.             mx4_sum+= mx4/2;

  871.             my4_sum+= my4/2;

  872.         }else{

  873.             mx4_sum+= mx4;

  874.             my4_sum+= my4;

  875.         }

  876.             

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

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

  879. 

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

  881.     }

  882.     

  883.     if(same)

  884.         return INT_MAX;

  885.     

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

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

  888.     }

  889.     

  890.     if(s->avctx->mb_cmp&FF_CMP_CHROMA){

  891.         int dxy;

  892.         int mx, my;

  893.         int offset;

  894. 

  895.         mx= ff_h263_round_chroma(mx4_sum);

  896.         my= ff_h263_round_chroma(my4_sum);

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

  898.         

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

  900.        

  901.         if(s->no_rounding){

  902.             s->dsp.put_no_rnd_pixels_tab[1][dxy](s->me.scratchpad    , s->last_picture.data[1] + offset, s->uvlinesize, 8);

  903.             s->dsp.put_no_rnd_pixels_tab[1][dxy](s->me.scratchpad+8  , s->last_picture.data[2] + offset, s->uvlinesize, 8);

  904.         }else{

  905.             s->dsp.put_pixels_tab       [1][dxy](s->me.scratchpad    , s->last_picture.data[1] + offset, s->uvlinesize, 8);

  906.             s->dsp.put_pixels_tab       [1][dxy](s->me.scratchpad+8  , s->last_picture.data[2] + offset, s->uvlinesize, 8);

  907.         }

  908. 

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

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

  911.     }

  912. 

  913.     switch(s->avctx->mb_cmp&0xFF){

  914.     /*case FF_CMP_SSE:

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

  916.     case FF_CMP_RD:

  917.         return dmin_sum;

  918.     default:

  919.         return dmin_sum+ 11*s->me.mb_penalty_factor;

  920.     }

  921. }

  922. 

  923. static int interlaced_search(MpegEncContext *s, uint8_t *frame_src_data[3], uint8_t *frame_ref_data[3], 

  924.                              int16_t (*mv_tables[2][2])[2], uint8_t *field_select_tables[2], int f_code, int mx, int my)

  925. {

  926.     const int size=0;

  927.     const int h=8;

  928.     int block;

  929.     int P[10][2];

  930.     uint8_t * const mv_penalty= s->me.mv_penalty[f_code] + MAX_MV;

  931.     int same=1;

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

  933.     const int uvstride= 2*s->uvlinesize;

  934.     int dmin_sum= 0;

  935.     const int mot_stride= s->mb_stride;

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

  937.     

  938.     s->me.ymin>>=1;

  939.     s->me.ymax>>=1;

  940.     

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

  942.         int field_select;

  943.         int best_dmin= INT_MAX;

  944.         int best_field= -1;

  945. 

  946.         uint8_t *src_data[3]= {

  947.             frame_src_data[0] + s->  linesize*block,

  948.             frame_src_data[1] + s->uvlinesize*block,

  949.             frame_src_data[2] + s->uvlinesize*block

  950.         };

  951. 

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

  953.             int dmin, mx_i, my_i, pred_x, pred_y;

  954.             uint8_t *ref_data[3]= {

  955.                 frame_ref_data[0] + s->  linesize*field_select,

  956.                 frame_ref_data[1] + s->uvlinesize*field_select,

  957.                 frame_ref_data[2] + s->uvlinesize*field_select

  958.             };

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

  960.             

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

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

  963.             if(P_LEFT[0]       > (s->me.xmax<<1)) P_LEFT[0]       = (s->me.xmax<<1);

  964.             

  965.             pred_x= P_LEFT[0];

  966.             pred_y= P_LEFT[1];

  967.             

  968.             if(!s->first_slice_line){

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

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

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

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

  973.                 if(P_TOP[1]      > (s->me.ymax<<1)) P_TOP[1]     = (s->me.ymax<<1);

  974.                 if(P_TOPRIGHT[0] < (s->me.xmin<<1)) P_TOPRIGHT[0]= (s->me.xmin<<1);

  975.                 if(P_TOPRIGHT[0] > (s->me.xmax<<1)) P_TOPRIGHT[0]= (s->me.xmax<<1);

  976.                 if(P_TOPRIGHT[1] > (s->me.ymax<<1)) P_TOPRIGHT[1]= (s->me.ymax<<1);

  977.     

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

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

  980.             }

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

  982.             P_MV1[1]= my / 2;

  983.             

  984.             dmin = s->me.motion_search[4](s, &mx_i, &my_i, P, pred_x, pred_y, 

  985.                                            src_data, ref_data, stride, uvstride, mv_table, (1<<16)>>1, mv_penalty);

  986. 

  987.             dmin= s->me.sub_motion_search(s, &mx_i, &my_i, dmin, 

  988.                                            pred_x, pred_y, src_data, ref_data, stride, uvstride, size, h, mv_penalty);

  989.             

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

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

  992.             

  993.             if(s->dsp.me_sub_cmp[0] != s->dsp.mb_cmp[0]

  994.                && s->avctx->mb_decision == FF_MB_DECISION_SIMPLE){

  995.                 int dxy;

  996. 

  997.                 //FIXME chroma ME

  998.                 uint8_t *ref= ref_data[0] + (mx_i>>1) + (my_i>>1)*stride;

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

  1000. 

  1001.                 if(s->no_rounding){

  1002.                     s->dsp.put_no_rnd_pixels_tab[size][dxy](s->me.scratchpad, ref    , stride, h);

  1003.                 }else{

  1004.                     s->dsp.put_pixels_tab       [size][dxy](s->me.scratchpad, ref    , stride, h);

  1005.                 }

  1006.                 dmin= s->dsp.mb_cmp[size](s, src_data[0], s->me.scratchpad, stride, h);

  1007.                 dmin+= (mv_penalty[mx_i-pred_x] + mv_penalty[my_i-pred_y] + 1)*s->me.mb_penalty_factor;

  1008.             }else

  1009.                 dmin+= s->me.mb_penalty_factor; //field_select bits

  1010.                 

  1011.             dmin += field_select != block; //slightly prefer same field

  1012.             

  1013.             if(dmin < best_dmin){

  1014.                 best_dmin= dmin;

  1015.                 best_field= field_select;

  1016.             }

  1017.         }

  1018.         {

  1019.             int16_t (*mv_table)[2]= mv_tables[block][best_field];

  1020. 

  1021.             if(mv_table[xy][0] != mx) same=0; //FIXME check if these checks work and are any good at all

  1022.             if(mv_table[xy][1]&1) same=0;

  1023.             if(mv_table[xy][1]*2 != my) same=0; 

  1024.             if(best_field != block) same=0;

  1025.         }

  1026. 

  1027.         field_select_tables[block][xy]= best_field;

  1028.         dmin_sum += best_dmin;

  1029.     }

  1030.     

  1031.     s->me.ymin<<=1;

  1032.     s->me.ymax<<=1;

  1033. 

  1034.     if(same)

  1035.         return INT_MAX;

  1036.     

  1037.     switch(s->avctx->mb_cmp&0xFF){

  1038.     /*case FF_CMP_SSE:

  1039.         return dmin_sum+ 32*s->qscale*s->qscale;*/

  1040.     case FF_CMP_RD:

  1041.         return dmin_sum;

  1042.     default:

  1043.         return dmin_sum+ 11*s->me.mb_penalty_factor;

  1044.     }

  1045. }

  1046. 

  1047. void ff_estimate_p_frame_motion(MpegEncContext * s,

  1048.                                 int mb_x, int mb_y)

  1049. {

  1050.     uint8_t *pix, *ppix;

  1051.     int sum, varc, vard, mx, my, dmin, xx, yy;

  1052.     int pred_x=0, pred_y=0;

  1053.     int P[10][2];

  1054.     const int shift= 1+s->quarter_sample;

  1055.     int mb_type=0;

  1056.     uint8_t *ref_picture= s->last_picture.data[0];

  1057.     Picture * const pic= &s->current_picture;

  1058.     uint8_t * const mv_penalty= s->me.mv_penalty[s->f_code] + MAX_MV;

  1059.     const int stride= s->linesize;

  1060.     const int uvstride= s->uvlinesize;

  1061.     uint8_t *src_data[3]= {

  1062.         s->new_picture.data[0] + 16*(mb_x + stride*mb_y),

  1063.         s->new_picture.data[1] + 8*(mb_x + uvstride*mb_y),

  1064.         s->new_picture.data[2] + 8*(mb_x + uvstride*mb_y)

  1065.     };

  1066.     uint8_t *ref_data[3]= {

  1067.         s->last_picture.data[0] + 16*(mb_x + stride*mb_y),

  1068.         s->last_picture.data[1] + 8*(mb_x + uvstride*mb_y),

  1069.         s->last_picture.data[2] + 8*(mb_x + uvstride*mb_y)

  1070.     };

  1071. 

  1072.     assert(s->quarter_sample==0 || s->quarter_sample==1);

  1073. 

  1074.     s->me.penalty_factor    = get_penalty_factor(s, s->avctx->me_cmp);

  1075.     s->me.sub_penalty_factor= get_penalty_factor(s, s->avctx->me_sub_cmp);

  1076.     s->me.mb_penalty_factor = get_penalty_factor(s, s->avctx->mb_cmp);

  1077. 

  1078.     get_limits(s, 16*mb_x, 16*mb_y);

  1079.     s->me.skip=0;

  1080. 

  1081.     switch(s->me_method) {

  1082.     case ME_ZERO:

  1083.     default:

  1084. 	no_motion_search(s, &mx, &my);

  1085.         mx-= mb_x*16;

  1086.         my-= mb_y*16;

  1087.         dmin = 0;

  1088.         break;

  1089. #if 0

  1090.     case ME_FULL:

  1091. 	dmin = full_motion_search(s, &mx, &my, range, ref_picture);

  1092.         mx-= mb_x*16;

  1093.         my-= mb_y*16;

  1094.         break;

  1095.     case ME_LOG:

  1096. 	dmin = log_motion_search(s, &mx, &my, range / 2, ref_picture);

  1097.         mx-= mb_x*16;

  1098.         my-= mb_y*16;

  1099.         break;

  1100.     case ME_PHODS:

  1101. 	dmin = phods_motion_search(s, &mx, &my, range / 2, ref_picture);

  1102.         mx-= mb_x*16;

  1103.         my-= mb_y*16;

  1104.         break;

  1105. #endif

  1106.     case ME_X1:

  1107.     case ME_EPZS:

  1108.        {

  1109.             const int mot_stride = s->block_wrap[0];

  1110.             const int mot_xy = s->block_index[0];

  1111. 

  1112.             P_LEFT[0]       = s->current_picture.motion_val[0][mot_xy - 1][0];

  1113.             P_LEFT[1]       = s->current_picture.motion_val[0][mot_xy - 1][1];

  1114. 

  1115.             if(P_LEFT[0]       > (s->me.xmax<<shift)) P_LEFT[0]       = (s->me.xmax<<shift);

  1116. 

  1117.             if(!s->first_slice_line) {

  1118.                 P_TOP[0]      = s->current_picture.motion_val[0][mot_xy - mot_stride    ][0];

  1119.                 P_TOP[1]      = s->current_picture.motion_val[0][mot_xy - mot_stride    ][1];

  1120.                 P_TOPRIGHT[0] = s->current_picture.motion_val[0][mot_xy - mot_stride + 2][0];

  1121.                 P_TOPRIGHT[1] = s->current_picture.motion_val[0][mot_xy - mot_stride + 2][1];

  1122.                 if(P_TOP[1]      > (s->me.ymax<<shift)) P_TOP[1]     = (s->me.ymax<<shift);

  1123.                 if(P_TOPRIGHT[0] < (s->me.xmin<<shift)) P_TOPRIGHT[0]= (s->me.xmin<<shift);

  1124.                 if(P_TOPRIGHT[1] > (s->me.ymax<<shift)) P_TOPRIGHT[1]= (s->me.ymax<<shift);

  1125.         

  1126.                 P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);

  1127.                 P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);

  1128. 

  1129.                 if(s->out_format == FMT_H263){

  1130.                     pred_x = P_MEDIAN[0];

  1131.                     pred_y = P_MEDIAN[1];

  1132.                 }else { /* mpeg1 at least */

  1133.                     pred_x= P_LEFT[0];

  1134.                     pred_y= P_LEFT[1];

  1135.                 }

  1136.             }else{

  1137.                 pred_x= P_LEFT[0];

  1138.                 pred_y= P_LEFT[1];

  1139.             }

  1140. 

  1141.         }

  1142.         dmin = s->me.motion_search[0](s, &mx, &my, P, pred_x, pred_y, 

  1143.                                       src_data, ref_data, stride, uvstride, s->p_mv_table, (1<<16)>>shift, mv_penalty);

  1144.  

  1145.         break;

  1146.     }

  1147. 

  1148.     /* intra / predictive decision */

  1149.     xx = mb_x * 16;

  1150.     yy = mb_y * 16;

  1151. 

  1152.     pix = src_data[0];

  1153.     /* At this point (mx,my) are full-pell and the relative displacement */

  1154.     ppix = ref_data[0] + (my * s->linesize) + mx;

  1155.     

  1156.     sum = s->dsp.pix_sum(pix, s->linesize);

  1157.     

  1158.     varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500 + 128)>>8;

  1159.     vard = (s->dsp.sse[0](NULL, pix, ppix, s->linesize, 16)+128)>>8;

  1160. 

  1161. //printf("%d %d %d %X %X %X\n", s->mb_width, mb_x, mb_y,(int)s, (int)s->mb_var, (int)s->mc_mb_var); fflush(stdout);

  1162.     pic->mb_var   [s->mb_stride * mb_y + mb_x] = varc;

  1163.     pic->mc_mb_var[s->mb_stride * mb_y + mb_x] = vard;

  1164.     pic->mb_mean  [s->mb_stride * mb_y + mb_x] = (sum+128)>>8;

  1165. //    pic->mb_cmp_score[s->mb_stride * mb_y + mb_x] = dmin; 

  1166.     s->mb_var_sum_temp    += varc;

  1167.     s->mc_mb_var_sum_temp += vard;

  1168. //printf("E%d %d %d %X %X %X\n", s->mb_width, mb_x, mb_y,(int)s, (int)s->mb_var, (int)s->mc_mb_var); fflush(stdout);

  1169.     

  1170. #if 0

  1171.     printf("varc=%4d avg_var=%4d (sum=%4d) vard=%4d mx=%2d my=%2d\n",

  1172. 	   varc, s->avg_mb_var, sum, vard, mx - xx, my - yy);

  1173. #endif

  1174.     if(s->avctx->mb_decision > FF_MB_DECISION_SIMPLE){

  1175.         if (vard <= 64 || vard < varc)

  1176.             s->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);

  1177.         else

  1178.             s->scene_change_score+= s->qscale;

  1179. 

  1180.         if (vard*2 + 200 > varc)

  1181.             mb_type|= CANDIDATE_MB_TYPE_INTRA;

  1182.         if (varc*2 + 200 > vard){

  1183.             mb_type|= CANDIDATE_MB_TYPE_INTER;

  1184.             s->me.sub_motion_search(s, &mx, &my, dmin,

  1185. 				   pred_x, pred_y, src_data, ref_data, stride, uvstride, 0, 16, mv_penalty);

  1186.             if(s->flags&CODEC_FLAG_MV0)

  1187.                 if(mx || my)

  1188.                     mb_type |= CANDIDATE_MB_TYPE_SKIPED; //FIXME check difference

  1189.         }else{

  1190.             mx <<=shift;

  1191.             my <<=shift;

  1192.         }

  1193.         if((s->flags&CODEC_FLAG_4MV)

  1194.            && !s->me.skip && varc>50 && vard>10){

  1195.             if(h263_mv4_search(s, mx, my, shift) < INT_MAX)

  1196.                 mb_type|=CANDIDATE_MB_TYPE_INTER4V;

  1197. 

  1198.             set_p_mv_tables(s, mx, my, 0);

  1199.         }else

  1200.             set_p_mv_tables(s, mx, my, 1);

  1201.         if((s->flags&CODEC_FLAG_INTERLACED_ME)

  1202.            && !s->me.skip){ //FIXME varc/d checks

  1203.             if(interlaced_search(s, src_data, ref_data, s->p_field_mv_table, s->p_field_select_table, s->f_code, mx, my) < INT_MAX)

  1204.                 mb_type |= CANDIDATE_MB_TYPE_INTER_I;

  1205.         }

  1206.     }else{

  1207.         int intra_score, i;

  1208.         mb_type= CANDIDATE_MB_TYPE_INTER;

  1209. 

  1210.         dmin= s->me.sub_motion_search(s, &mx, &my, dmin,

  1211.                                     pred_x, pred_y, src_data, ref_data, stride, uvstride, 0, 16, mv_penalty);

  1212.         if(s->avctx->me_sub_cmp != s->avctx->mb_cmp && !s->me.skip)

  1213.             dmin= s->me.get_mb_score(s, mx, my, pred_x, pred_y, src_data, ref_data, stride, uvstride, mv_penalty);

  1214. 

  1215.         if((s->flags&CODEC_FLAG_4MV)

  1216.            && !s->me.skip && varc>50 && vard>10){

  1217.             int dmin4= h263_mv4_search(s, mx, my, shift);

  1218.             if(dmin4 < dmin){

  1219.                 mb_type= CANDIDATE_MB_TYPE_INTER4V;

  1220.                 dmin=dmin4;

  1221.             }

  1222.         }

  1223.         if((s->flags&CODEC_FLAG_INTERLACED_ME)

  1224.            && !s->me.skip){ //FIXME varc/d checks

  1225.             int dmin_i= interlaced_search(s, src_data, ref_data, s->p_field_mv_table, s->p_field_select_table, s->f_code, mx, my);

  1226.             if(dmin_i < dmin){

  1227.                 mb_type = CANDIDATE_MB_TYPE_INTER_I;

  1228.                 dmin= dmin_i;

  1229.             }

  1230.         }

  1231.                 

  1232. //        pic->mb_cmp_score[s->mb_stride * mb_y + mb_x] = dmin; 

  1233.         set_p_mv_tables(s, mx, my, mb_type!=CANDIDATE_MB_TYPE_INTER4V);

  1234. 

  1235.         /* get intra luma score */

  1236.         if((s->avctx->mb_cmp&0xFF)==FF_CMP_SSE){

  1237.             intra_score= (varc<<8) - 500; //FIXME dont scale it down so we dont have to fix it

  1238.         }else{

  1239.             int mean= (sum+128)>>8;

  1240.             mean*= 0x01010101;

  1241.             

  1242.             for(i=0; i<16; i++){

  1243.                 *(uint32_t*)(&s->me.scratchpad[i*s->linesize+ 0]) = mean;

  1244.                 *(uint32_t*)(&s->me.scratchpad[i*s->linesize+ 4]) = mean;

  1245.                 *(uint32_t*)(&s->me.scratchpad[i*s->linesize+ 8]) = mean;

  1246.                 *(uint32_t*)(&s->me.scratchpad[i*s->linesize+12]) = mean;

  1247.             }

  1248. 

  1249.             intra_score= s->dsp.mb_cmp[0](s, s->me.scratchpad, pix, s->linesize, 16);

  1250.         }

  1251. #if 0 //FIXME

  1252.         /* get chroma score */

  1253.         if(s->avctx->mb_cmp&FF_CMP_CHROMA){

  1254.             for(i=1; i<3; i++){

  1255.                 uint8_t *dest_c;

  1256.                 int mean;

  1257.                 

  1258.                 if(s->out_format == FMT_H263){

  1259.                     mean= (s->dc_val[i][mb_x + (mb_y+1)*(s->mb_width+2)] + 4)>>3; //FIXME not exact but simple ;)

  1260.                 }else{

  1261.                     mean= (s->last_dc[i] + 4)>>3;

  1262.                 }

  1263.                 dest_c = s->new_picture.data[i] + (mb_y * 8  * (s->uvlinesize)) + mb_x * 8;

  1264.                 

  1265.                 mean*= 0x01010101;

  1266.                 for(i=0; i<8; i++){

  1267.                     *(uint32_t*)(&s->me.scratchpad[i*s->uvlinesize+ 0]) = mean;

  1268.                     *(uint32_t*)(&s->me.scratchpad[i*s->uvlinesize+ 4]) = mean;

  1269.                 }

  1270.                 

  1271.                 intra_score+= s->dsp.mb_cmp[1](s, s->me.scratchpad, dest_c, s->uvlinesize);

  1272.             }                

  1273.         }

  1274. #endif

  1275.         intra_score += s->me.mb_penalty_factor*16;

  1276.         

  1277.         if(intra_score < dmin){

  1278.             mb_type= CANDIDATE_MB_TYPE_INTRA;

  1279.             s->current_picture.mb_type[mb_y*s->mb_stride + mb_x]= CANDIDATE_MB_TYPE_INTRA; //FIXME cleanup

  1280.         }else

  1281.             s->current_picture.mb_type[mb_y*s->mb_stride + mb_x]= 0;

  1282.         

  1283.         if (vard <= 64 || vard < varc) { //FIXME

  1284.             s->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);

  1285.         }else{

  1286.             s->scene_change_score+= s->qscale;

  1287.         }

  1288.     }

  1289. 

  1290.     s->mb_type[mb_y*s->mb_stride + mb_x]= mb_type;

  1291. }

  1292. 

  1293. int ff_pre_estimate_p_frame_motion(MpegEncContext * s,

  1294.                                     int mb_x, int mb_y)

  1295. {

  1296.     int mx, my, dmin;

  1297.     int pred_x=0, pred_y=0;

  1298.     int P[10][2];

  1299.     const int shift= 1+s->quarter_sample;

  1300.     uint8_t * const mv_penalty= s->me.mv_penalty[s->f_code] + MAX_MV;

  1301.     const int xy= mb_x + mb_y*s->mb_stride;

  1302.     const int stride= s->linesize;

  1303.     const int uvstride= s->uvlinesize;

  1304.     uint8_t *src_data[3]= {

  1305.         s->new_picture.data[0] + 16*(mb_x + stride*mb_y),

  1306.         s->new_picture.data[1] + 8*(mb_x + uvstride*mb_y),

  1307.         s->new_picture.data[2] + 8*(mb_x + uvstride*mb_y)

  1308.     };

  1309.     uint8_t *ref_data[3]= {

  1310.         s->last_picture.data[0] + 16*(mb_x + stride*mb_y),

  1311.         s->last_picture.data[1] + 8*(mb_x + uvstride*mb_y),

  1312.         s->last_picture.data[2] + 8*(mb_x + uvstride*mb_y)

  1313.     };

  1314.     

  1315.     assert(s->quarter_sample==0 || s->quarter_sample==1);

  1316. 

  1317.     s->me.pre_penalty_factor    = get_penalty_factor(s, s->avctx->me_pre_cmp);

  1318. 

  1319.     get_limits(s, 16*mb_x, 16*mb_y);

  1320.     s->me.skip=0;

  1321. 

  1322.     P_LEFT[0]       = s->p_mv_table[xy + 1][0];

  1323.     P_LEFT[1]       = s->p_mv_table[xy + 1][1];

  1324. 

  1325.     if(P_LEFT[0]       < (s->me.xmin<<shift)) P_LEFT[0]       = (s->me.xmin<<shift);

  1326. 

  1327.     /* special case for first line */

  1328.     if (s->first_slice_line) {

  1329.         pred_x= P_LEFT[0];

  1330.         pred_y= P_LEFT[1];

  1331.         P_TOP[0]= P_TOPRIGHT[0]= P_MEDIAN[0]=

  1332.         P_TOP[1]= P_TOPRIGHT[1]= P_MEDIAN[1]= 0; //FIXME 

  1333.     } else {

  1334.         P_TOP[0]      = s->p_mv_table[xy + s->mb_stride    ][0];

  1335.         P_TOP[1]      = s->p_mv_table[xy + s->mb_stride    ][1];

  1336.         P_TOPRIGHT[0] = s->p_mv_table[xy + s->mb_stride - 1][0];

  1337.         P_TOPRIGHT[1] = s->p_mv_table[xy + s->mb_stride - 1][1];

  1338.         if(P_TOP[1]      < (s->me.ymin<<shift)) P_TOP[1]     = (s->me.ymin<<shift);

  1339.         if(P_TOPRIGHT[0] > (s->me.xmax<<shift)) P_TOPRIGHT[0]= (s->me.xmax<<shift);

  1340.         if(P_TOPRIGHT[1] < (s->me.ymin<<shift)) P_TOPRIGHT[1]= (s->me.ymin<<shift);

  1341.     

  1342.         P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);

  1343.         P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);

  1344. 

  1345.         pred_x = P_MEDIAN[0];

  1346.         pred_y = P_MEDIAN[1];

  1347.     }

  1348.     dmin = s->me.pre_motion_search(s, &mx, &my, P, pred_x, pred_y, 

  1349.                                    src_data, ref_data, stride, uvstride, s->p_mv_table, (1<<16)>>shift, mv_penalty);

  1350. 

  1351.     s->p_mv_table[xy][0] = mx<<shift;

  1352.     s->p_mv_table[xy][1] = my<<shift;

  1353.     

  1354.     return dmin;

  1355. }

  1356. 

  1357. static int ff_estimate_motion_b(MpegEncContext * s,

  1358.                        int mb_x, int mb_y, int16_t (*mv_table)[2], uint8_t *src_data[3],

  1359.                        uint8_t *ref_data[3], int stride, int uvstride, int f_code)

  1360. {

  1361.     int mx, my, dmin;

  1362.     int pred_x=0, pred_y=0;

  1363.     int P[10][2];

  1364.     const int shift= 1+s->quarter_sample;

  1365.     const int mot_stride = s->mb_stride;

  1366.     const int mot_xy = mb_y*mot_stride + mb_x;

  1367.     uint8_t * const ref_picture= ref_data[0] - 16*s->mb_x - 16*s->mb_y*s->linesize; //FIXME ugly

  1368.     uint8_t * const mv_penalty= s->me.mv_penalty[f_code] + MAX_MV;

  1369.     int mv_scale;

  1370.         

  1371.     s->me.penalty_factor    = get_penalty_factor(s, s->avctx->me_cmp);

  1372.     s->me.sub_penalty_factor= get_penalty_factor(s, s->avctx->me_sub_cmp);

  1373.     s->me.mb_penalty_factor = get_penalty_factor(s, s->avctx->mb_cmp);

  1374. 

  1375.     get_limits(s, 16*mb_x, 16*mb_y);

  1376. 

  1377.     switch(s->me_method) {

  1378.     case ME_ZERO:

  1379.     default:

  1380. 	no_motion_search(s, &mx, &my);

  1381.         dmin = 0;

  1382.         mx-= mb_x*16;

  1383.         my-= mb_y*16;

  1384.         break;

  1385. #if 0

  1386.     case ME_FULL:

  1387. 	dmin = full_motion_search(s, &mx, &my, range, ref_picture);

  1388.         mx-= mb_x*16;

  1389.         my-= mb_y*16;

  1390.         break;

  1391.     case ME_LOG:

  1392. 	dmin = log_motion_search(s, &mx, &my, range / 2, ref_picture);

  1393.         mx-= mb_x*16;

  1394.         my-= mb_y*16;

  1395.         break;

  1396.     case ME_PHODS:

  1397. 	dmin = phods_motion_search(s, &mx, &my, range / 2, ref_picture);

  1398.         mx-= mb_x*16;

  1399.         my-= mb_y*16;

  1400.         break;

  1401. #endif

  1402.     case ME_X1:

  1403.     case ME_EPZS:

  1404.        {

  1405.             P_LEFT[0]        = mv_table[mot_xy - 1][0];

  1406.             P_LEFT[1]        = mv_table[mot_xy - 1][1];

  1407. 

  1408.             if(P_LEFT[0]       > (s->me.xmax<<shift)) P_LEFT[0]       = (s->me.xmax<<shift);

  1409. 

  1410.             /* special case for first line */

  1411.             if (!s->first_slice_line) {

  1412.                 P_TOP[0] = mv_table[mot_xy - mot_stride             ][0];

  1413.                 P_TOP[1] = mv_table[mot_xy - mot_stride             ][1];

  1414.                 P_TOPRIGHT[0] = mv_table[mot_xy - mot_stride + 1         ][0];

  1415.                 P_TOPRIGHT[1] = mv_table[mot_xy - mot_stride + 1         ][1];

  1416.                 if(P_TOP[1] > (s->me.ymax<<shift)) P_TOP[1]= (s->me.ymax<<shift);

  1417.                 if(P_TOPRIGHT[0] < (s->me.xmin<<shift)) P_TOPRIGHT[0]= (s->me.xmin<<shift);

  1418.                 if(P_TOPRIGHT[1] > (s->me.ymax<<shift)) P_TOPRIGHT[1]= (s->me.ymax<<shift);

  1419.         

  1420.                 P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);

  1421.                 P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);

  1422.             }

  1423.             pred_x= P_LEFT[0];

  1424.             pred_y= P_LEFT[1];

  1425.         }

  1426.         

  1427.         if(mv_table == s->b_forw_mv_table){

  1428.             mv_scale= (s->pb_time<<16) / (s->pp_time<<shift);

  1429.         }else{

  1430.             mv_scale= ((s->pb_time - s->pp_time)<<16) / (s->pp_time<<shift);

  1431.         }

  1432.         

  1433.         dmin = s->me.motion_search[0](s, &mx, &my, P, pred_x, pred_y, 

  1434.                                       src_data, ref_data, stride, uvstride, s->p_mv_table, mv_scale, mv_penalty);

  1435.  

  1436.         break;

  1437.     }

  1438.     

  1439.     dmin= s->me.sub_motion_search(s, &mx, &my, dmin,

  1440. 				   pred_x, pred_y, src_data, ref_data, stride, uvstride, 0, 16, mv_penalty);

  1441.                                    

  1442.     if(s->avctx->me_sub_cmp != s->avctx->mb_cmp && !s->me.skip)

  1443.         dmin= s->me.get_mb_score(s, mx, my, pred_x, pred_y, src_data, ref_data, stride, uvstride, mv_penalty);

  1444. 

  1445. //printf("%d %d %d %d//", s->mb_x, s->mb_y, mx, my);

  1446. //    s->mb_type[mb_y*s->mb_width + mb_x]= mb_type;

  1447.     mv_table[mot_xy][0]= mx;

  1448.     mv_table[mot_xy][1]= my;

  1449. 

  1450.     return dmin;

  1451. }

  1452. 

  1453. static inline int check_bidir_mv(MpegEncContext * s, uint8_t *src_data[3], uint8_t *ref_data[6],

  1454.                    int stride, int uvstride,

  1455.                    int motion_fx, int motion_fy,

  1456.                    int motion_bx, int motion_by,

  1457.                    int pred_fx, int pred_fy,

  1458.                    int pred_bx, int pred_by,

  1459.                    int size, int h)

  1460. {

  1461.     //FIXME optimize?

  1462.     //FIXME move into template?

  1463.     //FIXME better f_code prediction (max mv & distance)

  1464.     //FIXME pointers

  1465.     uint8_t * const mv_penalty= s->me.mv_penalty[s->f_code] + MAX_MV; // f_code of the prev frame

  1466.     uint8_t *dest_y = s->me.scratchpad;

  1467.     uint8_t *ptr;

  1468.     int dxy;

  1469.     int src_x, src_y;

  1470.     int fbmin;

  1471. 

  1472.     if(s->quarter_sample){

  1473.         dxy = ((motion_fy & 3) << 2) | (motion_fx & 3);

  1474.         src_x = motion_fx >> 2;

  1475.         src_y = motion_fy >> 2;

  1476. 

  1477.         ptr = ref_data[0] + (src_y * stride) + src_x;

  1478.         s->dsp.put_qpel_pixels_tab[0][dxy](dest_y    , ptr    , stride);

  1479. 

  1480.         dxy = ((motion_by & 3) << 2) | (motion_bx & 3);

  1481.         src_x = motion_bx >> 2;

  1482.         src_y = motion_by >> 2;

  1483.     

  1484.         ptr = ref_data[3] + (src_y * stride) + src_x;

  1485.         s->dsp.avg_qpel_pixels_tab[size][dxy](dest_y    , ptr    , stride);

  1486.     }else{

  1487.         dxy = ((motion_fy & 1) << 1) | (motion_fx & 1);

  1488.         src_x = motion_fx >> 1;

  1489.         src_y = motion_fy >> 1;

  1490. 

  1491.         ptr = ref_data[0] + (src_y * stride) + src_x;

  1492.         s->dsp.put_pixels_tab[size][dxy](dest_y    , ptr    , stride, h);

  1493. 

  1494.         dxy = ((motion_by & 1) << 1) | (motion_bx & 1);

  1495.         src_x = motion_bx >> 1;

  1496.         src_y = motion_by >> 1;

  1497.     

  1498.         ptr = ref_data[3] + (src_y * stride) + src_x;

  1499.         s->dsp.avg_pixels_tab[size][dxy](dest_y    , ptr    , stride, h);

  1500.     }

  1501. 

  1502.     fbmin = (mv_penalty[motion_fx-pred_fx] + mv_penalty[motion_fy-pred_fy])*s->me.mb_penalty_factor

  1503.            +(mv_penalty[motion_bx-pred_bx] + mv_penalty[motion_by-pred_by])*s->me.mb_penalty_factor

  1504.            + s->dsp.mb_cmp[size](s, src_data[0], dest_y, stride, h); //FIXME new_pic

  1505.            

  1506.     if(s->avctx->mb_cmp&FF_CMP_CHROMA){

  1507.     }

  1508.     //FIXME CHROMA !!!

  1509.            

  1510.     return fbmin;

  1511. }

  1512. 

  1513. /* refine the bidir vectors in hq mode and return the score in both lq & hq mode*/

  1514. static inline int bidir_refine(MpegEncContext * s, uint8_t *src_data[3], uint8_t *ref_data[6],

  1515.                                   int stride, int uvstride,

  1516.                                   int mb_x, int mb_y)

  1517. {

  1518.     const int mot_stride = s->mb_stride;

  1519.     const int xy = mb_y *mot_stride + mb_x;

  1520.     int fbmin;

  1521.     int pred_fx= s->b_bidir_forw_mv_table[xy-1][0];

  1522.     int pred_fy= s->b_bidir_forw_mv_table[xy-1][1];

  1523.     int pred_bx= s->b_bidir_back_mv_table[xy-1][0];

  1524.     int pred_by= s->b_bidir_back_mv_table[xy-1][1];

  1525.     int motion_fx= s->b_bidir_forw_mv_table[xy][0]= s->b_forw_mv_table[xy][0];

  1526.     int motion_fy= s->b_bidir_forw_mv_table[xy][1]= s->b_forw_mv_table[xy][1];

  1527.     int motion_bx= s->b_bidir_back_mv_table[xy][0]= s->b_back_mv_table[xy][0];

  1528.     int motion_by= s->b_bidir_back_mv_table[xy][1]= s->b_back_mv_table[xy][1];

  1529. 

  1530.     //FIXME do refinement and add flag

  1531.     

  1532.     fbmin= check_bidir_mv(s, src_data, ref_data, stride, uvstride,

  1533.                           motion_fx, motion_fy,

  1534.                           motion_bx, motion_by,

  1535.                           pred_fx, pred_fy,

  1536.                           pred_bx, pred_by,

  1537.                           0, 16);

  1538. 

  1539.    return fbmin;

  1540. }

  1541. 

  1542. static inline int direct_search(MpegEncContext * s, uint8_t *src_data[3], uint8_t *ref_data[6],

  1543.                                 int stride, int uvstride,

  1544.                                 int mb_x, int mb_y)

  1545. {

  1546.     int P[10][2];

  1547.     const int mot_stride = s->mb_stride;

  1548.     const int mot_xy = mb_y*mot_stride + mb_x;

  1549.     const int shift= 1+s->quarter_sample;

  1550.     int dmin, i;

  1551.     const int time_pp= s->pp_time;

  1552.     const int time_pb= s->pb_time;

  1553.     int mx, my, xmin, xmax, ymin, ymax;

  1554.     int16_t (*mv_table)[2]= s->b_direct_mv_table;

  1555.     uint8_t * const mv_penalty= s->me.mv_penalty[1] + MAX_MV;

  1556.     

  1557.     ymin= xmin=(-32)>>shift;

  1558.     ymax= xmax=   31>>shift;

  1559. 

  1560.     if(IS_8X8(s->next_picture.mb_type[mot_xy])){

  1561.         s->mv_type= MV_TYPE_8X8;

  1562.     }else{

  1563.         s->mv_type= MV_TYPE_16X16;

  1564.     }

  1565. 

  1566.     for(i=0; i<4; i++){

  1567.         int index= s->block_index[i];

  1568.         int min, max;

  1569.     

  1570.         s->me.co_located_mv[i][0]= s->next_picture.motion_val[0][index][0];

  1571.         s->me.co_located_mv[i][1]= s->next_picture.motion_val[0][index][1];

  1572.         s->me.direct_basis_mv[i][0]= s->me.co_located_mv[i][0]*time_pb/time_pp + ((i& 1)<<(shift+3));

  1573.         s->me.direct_basis_mv[i][1]= s->me.co_located_mv[i][1]*time_pb/time_pp + ((i>>1)<<(shift+3));

  1574. //        s->me.direct_basis_mv[1][i][0]= s->me.co_located_mv[i][0]*(time_pb - time_pp)/time_pp + ((i &1)<<(shift+3);

  1575. //        s->me.direct_basis_mv[1][i][1]= s->me.co_located_mv[i][1]*(time_pb - time_pp)/time_pp + ((i>>1)<<(shift+3);

  1576. 

  1577.         max= FFMAX(s->me.direct_basis_mv[i][0], s->me.direct_basis_mv[i][0] - s->me.co_located_mv[i][0])>>shift;

  1578.         min= FFMIN(s->me.direct_basis_mv[i][0], s->me.direct_basis_mv[i][0] - s->me.co_located_mv[i][0])>>shift;

  1579.         max+= 16*mb_x + 1; // +-1 is for the simpler rounding

  1580.         min+= 16*mb_x - 1;

  1581.         xmax= FFMIN(xmax, s->width - max);

  1582.         xmin= FFMAX(xmin, - 16     - min);

  1583. 

  1584.         max= FFMAX(s->me.direct_basis_mv[i][1], s->me.direct_basis_mv[i][1] - s->me.co_located_mv[i][1])>>shift;

  1585.         min= FFMIN(s->me.direct_basis_mv[i][1], s->me.direct_basis_mv[i][1] - s->me.co_located_mv[i][1])>>shift;

  1586.         max+= 16*mb_y + 1; // +-1 is for the simpler rounding

  1587.         min+= 16*mb_y - 1;

  1588.         ymax= FFMIN(ymax, s->height - max);

  1589.         ymin= FFMAX(ymin, - 16      - min);

  1590.         

  1591.         if(s->mv_type == MV_TYPE_16X16) break;

  1592.     }

  1593.     

  1594.     assert(xmax <= 15 && ymax <= 15 && xmin >= -16 && ymin >= -16);

  1595.     

  1596.     if(xmax < 0 || xmin >0 || ymax < 0 || ymin > 0){

  1597.         s->b_direct_mv_table[mot_xy][0]= 0;

  1598.         s->b_direct_mv_table[mot_xy][1]= 0;

  1599. 

  1600.         return 256*256*256*64;

  1601.     }

  1602.     

  1603.     s->me.xmin= xmin;

  1604.     s->me.ymin= ymin;

  1605.     s->me.xmax= xmax;

  1606.     s->me.ymax= ymax;

  1607. 

  1608.     P_LEFT[0]        = clip(mv_table[mot_xy - 1][0], xmin<<shift, xmax<<shift);

  1609.     P_LEFT[1]        = clip(mv_table[mot_xy - 1][1], ymin<<shift, ymax<<shift);

  1610. 

  1611.     /* special case for first line */

  1612.     if (!s->first_slice_line) { //FIXME maybe allow this over thread boundary as its cliped

  1613.         P_TOP[0]      = clip(mv_table[mot_xy - mot_stride             ][0], xmin<<shift, xmax<<shift);

  1614.         P_TOP[1]      = clip(mv_table[mot_xy - mot_stride             ][1], ymin<<shift, ymax<<shift);

  1615.         P_TOPRIGHT[0] = clip(mv_table[mot_xy - mot_stride + 1         ][0], xmin<<shift, xmax<<shift);

  1616.         P_TOPRIGHT[1] = clip(mv_table[mot_xy - mot_stride + 1         ][1], ymin<<shift, ymax<<shift);

  1617.     

  1618.         P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);

  1619.         P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);

  1620.     }

  1621.  

  1622.     //FIXME direct_search  ptr in context!!! (needed for chroma anyway or this will get messy)   

  1623.     if(s->flags&CODEC_FLAG_QPEL){

  1624.         dmin = simple_direct_qpel_epzs_motion_search(s, &mx, &my, P, 0, 0, 

  1625.                                                      src_data, ref_data, stride, uvstride, mv_table, 1<<14, mv_penalty);

  1626.         dmin = simple_direct_qpel_qpel_motion_search(s, &mx, &my, dmin,

  1627.                                                 0, 0, src_data, ref_data, stride, uvstride, 0, 16, mv_penalty);

  1628.         

  1629.         if(s->avctx->me_sub_cmp != s->avctx->mb_cmp && !s->me.skip)

  1630.             dmin= simple_direct_qpel_qpel_get_mb_score(s, mx, my, 0, 0, src_data, ref_data, stride, uvstride, mv_penalty);

  1631.     }else{

  1632.         dmin = simple_direct_hpel_epzs_motion_search(s, &mx, &my, P, 0, 0, 

  1633.                                                      src_data, ref_data, stride, uvstride, mv_table, 1<<15, mv_penalty);

  1634.         dmin = simple_direct_hpel_hpel_motion_search(s, &mx, &my, dmin,

  1635.                                                 0, 0, src_data, ref_data, stride, uvstride, 0, 16, mv_penalty);

  1636. 

  1637.         if(s->avctx->me_sub_cmp != s->avctx->mb_cmp && !s->me.skip)

  1638.             dmin= simple_direct_hpel_hpel_get_mb_score(s, mx, my, 0, 0, src_data, ref_data, stride, uvstride, mv_penalty);

  1639.     }

  1640.     

  1641.     get_limits(s, 16*mb_x, 16*mb_y); //restore s->me.?min/max, maybe not needed

  1642. 

  1643.     s->b_direct_mv_table[mot_xy][0]= mx;

  1644.     s->b_direct_mv_table[mot_xy][1]= my;

  1645.     return dmin;

  1646. }

  1647. 

  1648. void ff_estimate_b_frame_motion(MpegEncContext * s,

  1649.                              int mb_x, int mb_y)

  1650. {

  1651.     const int penalty_factor= s->me.mb_penalty_factor;

  1652.     int fmin, bmin, dmin, fbmin, bimin, fimin;

  1653.     int type=0;

  1654.     const int stride= s->linesize;

  1655.     const int uvstride= s->uvlinesize;

  1656.     uint8_t *src_data[3]= {

  1657.         s->new_picture.data[0] + 16*(s->mb_x + stride*s->mb_y),

  1658.         s->new_picture.data[1] + 8*(s->mb_x + uvstride*s->mb_y),

  1659.         s->new_picture.data[2] + 8*(s->mb_x + uvstride*s->mb_y)

  1660.     };

  1661.     uint8_t *ref_data[6]= {

  1662.         s->last_picture.data[0] + 16*(s->mb_x + stride*s->mb_y),

  1663.         s->last_picture.data[1] + 8*(s->mb_x + uvstride*s->mb_y),

  1664.         s->last_picture.data[2] + 8*(s->mb_x + uvstride*s->mb_y),

  1665.         s->next_picture.data[0] + 16*(s->mb_x + stride*s->mb_y),

  1666.         s->next_picture.data[1] + 8*(s->mb_x + uvstride*s->mb_y),

  1667.         s->next_picture.data[2] + 8*(s->mb_x + uvstride*s->mb_y)

  1668.     };

  1669.     

  1670.     s->me.skip=0;

  1671.     if (s->codec_id == CODEC_ID_MPEG4)

  1672.         dmin= direct_search(s, src_data, ref_data, stride, uvstride, mb_x, mb_y);

  1673.     else

  1674.         dmin= INT_MAX;

  1675. //FIXME penalty stuff for non mpeg4

  1676.     s->me.skip=0;

  1677.     fmin= ff_estimate_motion_b(s, mb_x, mb_y, s->b_forw_mv_table, src_data, 

  1678.                                ref_data, stride, uvstride, s->f_code) + 3*penalty_factor;

  1679.     

  1680.     s->me.skip=0;

  1681.     bmin= ff_estimate_motion_b(s, mb_x, mb_y, s->b_back_mv_table, src_data, 

  1682.                                ref_data+3, stride, uvstride, s->b_code) + 2*penalty_factor;

  1683. //printf(" %d %d ", s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1]);

  1684. 

  1685.     s->me.skip=0;

  1686.     fbmin= bidir_refine(s, src_data, ref_data, stride, uvstride, mb_x, mb_y) + penalty_factor;

  1687. //printf("%d %d %d %d\n", dmin, fmin, bmin, fbmin);

  1688.     

  1689.     if(s->flags & CODEC_FLAG_INTERLACED_ME){

  1690.         const int xy = mb_y*s->mb_stride + mb_x;

  1691. 

  1692. //FIXME mb type penalty

  1693.         s->me.skip=0;

  1694.         fimin= interlaced_search(s, src_data, ref_data  , 

  1695.                                  s->b_field_mv_table[0], s->b_field_select_table[0], s->f_code,

  1696.                                  s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1]);

  1697.         bimin= interlaced_search(s, src_data, ref_data+3, 

  1698.                                  s->b_field_mv_table[1], s->b_field_select_table[1], s->b_code,

  1699.                                  s->b_back_mv_table[xy][0], s->b_back_mv_table[xy][1]);

  1700.     }else

  1701.         fimin= bimin= INT_MAX;

  1702. 

  1703.     {

  1704.         int score= fmin;

  1705.         type = CANDIDATE_MB_TYPE_FORWARD;

  1706.         

  1707.         if (dmin <= score){

  1708.             score = dmin;

  1709.             type = CANDIDATE_MB_TYPE_DIRECT;

  1710.         }

  1711.         if(bmin<score){

  1712.             score=bmin;

  1713.             type= CANDIDATE_MB_TYPE_BACKWARD; 

  1714.         }

  1715.         if(fbmin<score){

  1716.             score=fbmin;

  1717.             type= CANDIDATE_MB_TYPE_BIDIR;

  1718.         }

  1719.         if(fimin<score){

  1720.             score=fimin;

  1721.             type= CANDIDATE_MB_TYPE_FORWARD_I;

  1722.         }

  1723.         if(bimin<score){

  1724.             score=bimin;

  1725.             type= CANDIDATE_MB_TYPE_BACKWARD_I;

  1726.         }

  1727.         

  1728.         score= ((unsigned)(score*score + 128*256))>>16;

  1729.         s->mc_mb_var_sum_temp += score;

  1730.         s->current_picture.mc_mb_var[mb_y*s->mb_stride + mb_x] = score; //FIXME use SSE

  1731.     }

  1732. 

  1733.     if(s->avctx->mb_decision > FF_MB_DECISION_SIMPLE){

  1734.         type= CANDIDATE_MB_TYPE_FORWARD | CANDIDATE_MB_TYPE_BACKWARD | CANDIDATE_MB_TYPE_BIDIR | CANDIDATE_MB_TYPE_DIRECT;

  1735.         if(fimin < INT_MAX)

  1736.             type |= CANDIDATE_MB_TYPE_FORWARD_I;

  1737.         if(bimin < INT_MAX)

  1738.             type |= CANDIDATE_MB_TYPE_BACKWARD_I;

  1739.         if(fimin < INT_MAX && bimin < INT_MAX){

  1740.             type |= CANDIDATE_MB_TYPE_BIDIR_I;

  1741.         }

  1742.          //FIXME something smarter

  1743.         if(dmin>256*256*16) type&= ~CANDIDATE_MB_TYPE_DIRECT; //dont try direct mode if its invalid for this MB

  1744. #if 0        

  1745.         if(s->out_format == FMT_MPEG1)

  1746.             type |= CANDIDATE_MB_TYPE_INTRA;

  1747. #endif

  1748.     }

  1749. 

  1750.     s->mb_type[mb_y*s->mb_stride + mb_x]= type;

  1751. }

  1752. 

  1753. /* find best f_code for ME which do unlimited searches */

  1754. int ff_get_best_fcode(MpegEncContext * s, int16_t (*mv_table)[2], int type)

  1755. {

  1756.     if(s->me_method>=ME_EPZS){

  1757.         int score[8];

  1758.         int i, y;

  1759.         uint8_t * fcode_tab= s->fcode_tab;

  1760.         int best_fcode=-1;

  1761.         int best_score=-10000000;

  1762. 

  1763.         for(i=0; i<8; i++) score[i]= s->mb_num*(8-i);

  1764. 

  1765.         for(y=0; y<s->mb_height; y++){

  1766.             int x;

  1767.             int xy= y*s->mb_stride;

  1768.             for(x=0; x<s->mb_width; x++){

  1769.                 if(s->mb_type[xy] & type){

  1770.                     int fcode= FFMAX(fcode_tab[mv_table[xy][0] + MAX_MV],

  1771.                                      fcode_tab[mv_table[xy][1] + MAX_MV]);

  1772.                     int j;

  1773.                     

  1774.                     for(j=0; j<fcode && j<8; j++){

  1775.                         if(s->pict_type==B_TYPE || s->current_picture.mc_mb_var[xy] < s->current_picture.mb_var[xy])

  1776.                             score[j]-= 170;

  1777.                     }

  1778.                 }

  1779.                 xy++;

  1780.             }

  1781.         }

  1782.         

  1783.         for(i=1; i<8; i++){

  1784.             if(score[i] > best_score){

  1785.                 best_score= score[i];

  1786.                 best_fcode= i;

  1787.             }

  1788. //            printf("%d %d\n", i, score[i]);

  1789.         }

  1790. 

  1791. //    printf("fcode: %d type: %d\n", i, s->pict_type);

  1792.         return best_fcode;

  1793. /*        for(i=0; i<=MAX_FCODE; i++){

  1794.             printf("%d ", mv_num[i]);

  1795.         }

  1796.         printf("\n");*/

  1797.     }else{

  1798.         return 1;

  1799.     }

  1800. }

  1801. 

  1802. void ff_fix_long_p_mvs(MpegEncContext * s)

  1803. {

  1804.     const int f_code= s->f_code;

  1805.     int y, range;

  1806.     assert(s->pict_type==P_TYPE);

  1807. 

  1808.     range = (((s->out_format == FMT_MPEG1) ? 8 : 16) << f_code);

  1809.     

  1810.     if(s->msmpeg4_version) range= 16;

  1811.     

  1812.     if(s->avctx->me_range && range > s->avctx->me_range) range= s->avctx->me_range;

  1813.     

  1814. //printf("%d no:%d %d//\n", clip, noclip, f_code);

  1815.     if(s->flags&CODEC_FLAG_4MV){

  1816.         const int wrap= 2+ s->mb_width*2;

  1817. 

  1818.         /* clip / convert to intra 8x8 type MVs */

  1819.         for(y=0; y<s->mb_height; y++){

  1820.             int xy= (y*2 + 1)*wrap + 1;

  1821.             int i= y*s->mb_stride;

  1822.             int x;

  1823. 

  1824.             for(x=0; x<s->mb_width; x++){

  1825.                 if(s->mb_type[i]&CANDIDATE_MB_TYPE_INTER4V){

  1826.                     int block;

  1827.                     for(block=0; block<4; block++){

  1828.                         int off= (block& 1) + (block>>1)*wrap;

  1829.                         int mx= s->current_picture.motion_val[0][ xy + off ][0];

  1830.                         int my= s->current_picture.motion_val[0][ xy + off ][1];

  1831. 

  1832.                         if(   mx >=range || mx <-range

  1833.                            || my >=range || my <-range){

  1834.                             s->mb_type[i] &= ~CANDIDATE_MB_TYPE_INTER4V;

  1835.                             s->mb_type[i] |= CANDIDATE_MB_TYPE_INTRA;

  1836.                             s->current_picture.mb_type[i]= CANDIDATE_MB_TYPE_INTRA;

  1837.                         }

  1838.                     }

  1839.                 }

  1840.                 xy+=2;

  1841.                 i++;

  1842.             }

  1843.         }

  1844.     }

  1845. }

  1846. 

  1847. /**

  1848.  *

  1849.  * @param truncate 1 for truncation, 0 for using intra

  1850.  */

  1851. void ff_fix_long_mvs(MpegEncContext * s, uint8_t *field_select_table, int field_select, 

  1852.                      int16_t (*mv_table)[2], int f_code, int type, int truncate)

  1853. {

  1854.     int y, h_range, v_range;

  1855. 

  1856.     // RAL: 8 in MPEG-1, 16 in MPEG-4

  1857.     int range = (((s->out_format == FMT_MPEG1) ? 8 : 16) << f_code);

  1858. 

  1859.     if(s->msmpeg4_version) range= 16;

  1860.     if(s->avctx->me_range && range > s->avctx->me_range) range= s->avctx->me_range;

  1861. 

  1862.     h_range= range;

  1863.     v_range= field_select_table ? range>>1 : range;

  1864. 

  1865.     /* clip / convert to intra 16x16 type MVs */

  1866.     for(y=0; y<s->mb_height; y++){

  1867.         int x;

  1868.         int xy= y*s->mb_stride;

  1869.         for(x=0; x<s->mb_width; x++){

  1870.             if (s->mb_type[xy] & type){    // RAL: "type" test added...

  1871.                 if(field_select_table==NULL || field_select_table[xy] == field_select){

  1872.                     if(   mv_table[xy][0] >=h_range || mv_table[xy][0] <-h_range

  1873.                        || mv_table[xy][1] >=v_range || mv_table[xy][1] <-v_range){

  1874. 

  1875.                         if(truncate){

  1876.                             if     (mv_table[xy][0] > h_range-1) mv_table[xy][0]=  h_range-1;

  1877.                             else if(mv_table[xy][0] < -h_range ) mv_table[xy][0]= -h_range;

  1878.                             if     (mv_table[xy][1] > v_range-1) mv_table[xy][1]=  v_range-1;

  1879.                             else if(mv_table[xy][1] < -v_range ) mv_table[xy][1]= -v_range;

  1880.                         }else{

  1881.                             s->mb_type[xy] &= ~type;

  1882.                             s->mb_type[xy] |= CANDIDATE_MB_TYPE_INTRA;

  1883.                             mv_table[xy][0]=

  1884.                             mv_table[xy][1]= 0;

  1885.                         }

  1886.                     }

  1887.                 }

  1888.             }

  1889.             xy++;

  1890.         }

  1891.     }

  1892. }