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.
3330 Commits
32 Branches
593MB
Tree: d890d4aed1
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'd890d4aed1'
${ noResults }
FFmpeg/libavcodec/motion_est.c

1989 lines
69KB
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 src_index, int ref_index,

  50.                                   int size, int h);

  51. 

  52. static inline int update_map_generation(MpegEncContext * s)

  53. {

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

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

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

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

  58.     }

  59.     return s->me.map_generation;

  60. }

  61. 

  62. /* shape adaptive search stuff */

  63. typedef struct Minima{

  64.     int height;

  65.     int x, y;

  66.     int checked;

  67. }Minima;

  68. 

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

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

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

  72.     

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

  74. }

  75. 

  76. #define FLAG_QPEL   1 //must be 1

  77. #define FLAG_CHROMA 2

  78. #define FLAG_DIRECT 4

  79. 

  80. static inline void init_ref(MpegEncContext *s, uint8_t *src[3], uint8_t *ref[3], uint8_t *ref2[3], int x, int y, int ref_index){

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

  82.     const int offset[3]= {

  83.           y*c->  stride + x,

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

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

  86.     };

  87.     int i;

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

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

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

  91.     }

  92.     if(ref_index){

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

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

  95.         }

  96.     }

  97. }

  98. 

  99. static int get_flags(MpegEncContext *s, int direct, int chroma){

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

  101.            + (direct ? FLAG_DIRECT : 0) 

  102.            + (chroma ? FLAG_CHROMA : 0);

  103. }

  104. 

  105. static always_inline int cmp(MpegEncContext *s, const int x, const int y, const int subx, const int suby,

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

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

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

  109.     const int stride= c->stride;

  110.     const int uvstride= c->uvstride;

  111.     const int qpel= flags&FLAG_QPEL;

  112.     const int chroma= flags&FLAG_CHROMA;

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

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

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

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

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

  118.     int d;

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

  120.     if(flags&FLAG_DIRECT){

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

  122.             const int time_pp= s->pp_time;

  123.             const int time_pb= s->pb_time;

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

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

  126.                 int i;

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

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

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

  130.                     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));

  131.                     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));

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

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

  134.         

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

  136.                     if(qpel){

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

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

  139.                     }else{

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

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

  142.                     }

  143.                 }

  144.             }else{

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

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

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

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

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

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

  151.                 

  152.                 if(qpel){

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

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

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

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

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

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

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

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

  161.                 }else{            

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

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

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

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

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

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

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

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

  170. 

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

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

  173.                 }

  174.             }

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

  176.         }else

  177.             d= 256*256*256*32;

  178.     }else{

  179.         int uvdxy;

  180.         if(dxy){

  181.             if(qpel){

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

  183.                 if(chroma){

  184.                     int cx= hx/2;

  185.                     int cy= hy/2;

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

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

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

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

  190.                 }

  191.             }else{

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

  193.                 if(chroma)

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

  195.             }

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

  197.         }else{

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

  199.             if(chroma)

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

  201.         }

  202.         if(chroma){

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

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

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

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

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

  208.         }

  209.     }

  210. #if 0

  211.     if(full_pel){

  212.         const int index= (((y)<<ME_MAP_SHIFT) + (x))&(ME_MAP_SIZE-1);

  213.         score_map[index]= d;

  214.     }

  215. 

  216.     d += (c->mv_penalty[hx - c->pred_x] + c->mv_penalty[hy - c->pred_y])*c->penalty_factor;

  217. #endif

  218.     return d;

  219. }

  220. 

  221. #include "motion_est_template.c"

  222. 

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

  224.     switch(type&0xFF){

  225.     default:

  226.     case FF_CMP_SAD:

  227.         return s->qscale*2;

  228.     case FF_CMP_DCT:

  229.         return s->qscale*3;

  230.     case FF_CMP_SATD:

  231.         return s->qscale*6;

  232.     case FF_CMP_SSE:

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

  234.     case FF_CMP_BIT:

  235.         return 1;

  236.     case FF_CMP_RD:

  237.     case FF_CMP_PSNR:

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

  239.     }

  240. }

  241. 

  242. void ff_init_me(MpegEncContext *s){

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

  244. 

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

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

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

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

  249.     

  250.     s->me.flags    = get_flags(s, 0, s->avctx->me_cmp    &FF_CMP_CHROMA);

  251.     s->me.sub_flags= get_flags(s, 0, s->avctx->me_sub_cmp&FF_CMP_CHROMA);

  252.     s->me.mb_flags = get_flags(s, 0, s->avctx->mb_cmp    &FF_CMP_CHROMA);

  253. 

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

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

  256.         s->me.sub_motion_search= qpel_motion_search;

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

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

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

  260.     }else{

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

  262.             s->me.sub_motion_search= hpel_motion_search;

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

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

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

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

  267.         else

  268.             s->me.sub_motion_search= hpel_motion_search;

  269.         c->hpel_avg= s->dsp.avg_pixels_tab;

  270.         if(s->no_rounding) c->hpel_put= s->dsp.put_no_rnd_pixels_tab;

  271.         else               c->hpel_put= s->dsp.put_pixels_tab;

  272.     }

  273.     if(s->linesize){

  274.         s->me.stride  = s->linesize; 

  275.         s->me.uvstride= s->uvlinesize;

  276.     }else{

  277.         s->me.stride  = 16*s->mb_width + 32;

  278.         s->me.uvstride=  8*s->mb_width + 16;

  279.     }

  280. 

  281.     c->temp= c->scratchpad;

  282. }

  283.       

  284. #if 0

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

  286. {

  287.     int s, i, j;

  288. 

  289.     s = 0;

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

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

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

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

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

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

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

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

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

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

  300. 	    pix += 8;

  301. 	}

  302. 	pix += line_size - 16;

  303.     }

  304.     return s;

  305. }

  306. #endif

  307. 

  308. static inline void no_motion_search(MpegEncContext * s,

  309. 				    int *mx_ptr, int *my_ptr)

  310. {

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

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

  313. }

  314. 

  315. static int full_motion_search(MpegEncContext * s,

  316.                               int *mx_ptr, int *my_ptr, int range,

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

  318. {

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

  320.     int mx, my, dmin, d;

  321.     uint8_t *pix;

  322. 

  323.     xx = 16 * s->mb_x;

  324.     yy = 16 * s->mb_y;

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

  326.     if (x1 < xmin)

  327. 	x1 = xmin;

  328.     x2 = xx + range - 1;

  329.     if (x2 > xmax)

  330. 	x2 = xmax;

  331.     y1 = yy - range + 1;

  332.     if (y1 < ymin)

  333. 	y1 = ymin;

  334.     y2 = yy + range - 1;

  335.     if (y2 > ymax)

  336. 	y2 = ymax;

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

  338.     dmin = 0x7fffffff;

  339.     mx = 0;

  340.     my = 0;

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

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

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

  344. 			     s->linesize, 16);

  345. 	    if (d < dmin ||

  346. 		(d == dmin &&

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

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

  349. 		dmin = d;

  350. 		mx = x;

  351. 		my = y;

  352. 	    }

  353. 	}

  354.     }

  355. 

  356.     *mx_ptr = mx;

  357.     *my_ptr = my;

  358. 

  359. #if 0

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

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

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

  363.     }

  364. #endif

  365.     return dmin;

  366. }

  367. 

  368. 

  369. static int log_motion_search(MpegEncContext * s,

  370.                              int *mx_ptr, int *my_ptr, int range,

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

  372. {

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

  374.     int mx, my, dmin, d;

  375.     uint8_t *pix;

  376. 

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

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

  379. 

  380.     /* Left limit */

  381.     x1 = xx - range;

  382.     if (x1 < xmin)

  383. 	x1 = xmin;

  384. 

  385.     /* Right limit */

  386.     x2 = xx + range;

  387.     if (x2 > xmax)

  388. 	x2 = xmax;

  389. 

  390.     /* Upper limit */

  391.     y1 = yy - range;

  392.     if (y1 < ymin)

  393. 	y1 = ymin;

  394. 

  395.     /* Lower limit */

  396.     y2 = yy + range;

  397.     if (y2 > ymax)

  398. 	y2 = ymax;

  399. 

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

  401.     dmin = 0x7fffffff;

  402.     mx = 0;

  403.     my = 0;

  404. 

  405.     do {

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

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

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

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

  410. 		    dmin = d;

  411. 		    mx = x;

  412. 		    my = y;

  413. 		}

  414. 	    }

  415. 	}

  416. 

  417. 	range = range >> 1;

  418. 

  419. 	x1 = mx - range;

  420. 	if (x1 < xmin)

  421. 	    x1 = xmin;

  422. 

  423. 	x2 = mx + range;

  424. 	if (x2 > xmax)

  425. 	    x2 = xmax;

  426. 

  427. 	y1 = my - range;

  428. 	if (y1 < ymin)

  429. 	    y1 = ymin;

  430. 

  431. 	y2 = my + range;

  432. 	if (y2 > ymax)

  433. 	    y2 = ymax;

  434. 

  435.     } while (range >= 1);

  436. 

  437. #ifdef DEBUG

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

  439. #endif

  440.     *mx_ptr = mx;

  441.     *my_ptr = my;

  442.     return dmin;

  443. }

  444. 

  445. static int phods_motion_search(MpegEncContext * s,

  446.                                int *mx_ptr, int *my_ptr, int range,

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

  448. {

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

  450.     int mx, my, dminx, dminy;

  451.     uint8_t *pix;

  452. 

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

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

  455. 

  456.     /* Left limit */

  457.     x1 = xx - range;

  458.     if (x1 < xmin)

  459. 	x1 = xmin;

  460. 

  461.     /* Right limit */

  462.     x2 = xx + range;

  463.     if (x2 > xmax)

  464. 	x2 = xmax;

  465. 

  466.     /* Upper limit */

  467.     y1 = yy - range;

  468.     if (y1 < ymin)

  469. 	y1 = ymin;

  470. 

  471.     /* Lower limit */

  472.     y2 = yy + range;

  473.     if (y2 > ymax)

  474. 	y2 = ymax;

  475. 

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

  477.     mx = 0;

  478.     my = 0;

  479. 

  480.     x = xx;

  481.     y = yy;

  482.     do {

  483.         dminx = 0x7fffffff;

  484.         dminy = 0x7fffffff;

  485. 

  486. 	lastx = x;

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

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

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

  490. 		dminx = d;

  491. 		mx = x;

  492. 	    }

  493. 	}

  494. 

  495. 	x = lastx;

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

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

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

  499. 		dminy = d;

  500. 		my = y;

  501. 	    }

  502. 	}

  503. 

  504. 	range = range >> 1;

  505. 

  506. 	x = mx;

  507. 	y = my;

  508. 	x1 = mx - range;

  509. 	if (x1 < xmin)

  510. 	    x1 = xmin;

  511. 

  512. 	x2 = mx + range;

  513. 	if (x2 > xmax)

  514. 	    x2 = xmax;

  515. 

  516. 	y1 = my - range;

  517. 	if (y1 < ymin)

  518. 	    y1 = ymin;

  519. 

  520. 	y2 = my + range;

  521. 	if (y2 > ymax)

  522. 	    y2 = ymax;

  523. 

  524.     } while (range >= 1);

  525. 

  526. #ifdef DEBUG

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

  528. #endif

  529. 

  530.     /* half pixel search */

  531.     *mx_ptr = mx;

  532.     *my_ptr = my;

  533.     return dminy;

  534. }

  535. 

  536. 

  537. #define Z_THRESHOLD 256

  538. 

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

  540. {\

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

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

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

  544. }

  545. 

  546. static inline int sad_hpel_motion_search(MpegEncContext * s,

  547. 				  int *mx_ptr, int *my_ptr, int dmin,

  548.                                   int src_index, int ref_index,

  549.                                   int size, int h)

  550. {

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

  552.     int mx, my, dminh;

  553.     uint8_t *pix, *ptr;

  554.     int stride= s->me.stride;

  555.     const int flags= s->me.sub_flags;

  556.     LOAD_COMMON

  557.     

  558.     assert(flags == 0);

  559. 

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

  561. //    printf("S");

  562.         *mx_ptr = 0;

  563.         *my_ptr = 0;

  564.         return dmin;

  565.     }

  566. //    printf("N");

  567.         

  568.     pix = s->me.src[src_index][0];

  569. 

  570.     mx = *mx_ptr;

  571.     my = *my_ptr;

  572.     ptr = s->me.ref[ref_index][0] + (my * stride) + mx;

  573.     

  574.     dminh = dmin;

  575. 

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

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

  578.         int dx=0, dy=0;

  579.         int d, pen_x, pen_y; 

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

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

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

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

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

  585.         mx<<=1;

  586.         my<<=1;

  587. 

  588.         

  589.         pen_x= pred_x + mx;

  590.         pen_y= pred_y + my;

  591. 

  592.         ptr-= stride;

  593.         if(t<=b){

  594.             CHECK_SAD_HALF_MV(y2 , 0, -1)

  595.             if(l<=r){

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

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

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

  599.                     ptr+= stride;

  600.                 }else{

  601.                     ptr+= stride;

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

  603.                 }

  604.                 CHECK_SAD_HALF_MV(x2 , -1,  0)

  605.             }else{

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

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

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

  609.                     ptr+= stride;

  610.                 }else{

  611.                     ptr+= stride;

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

  613.                 }

  614.                 CHECK_SAD_HALF_MV(x2 , +1,  0)

  615.             }

  616.         }else{

  617.             if(l<=r){

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

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

  620.                     ptr+= stride;

  621.                 }else{

  622.                     ptr+= stride;

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

  624.                 }

  625.                 CHECK_SAD_HALF_MV(x2 , -1,  0)

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

  627.             }else{

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

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

  630.                     ptr+= stride;

  631.                 }else{

  632.                     ptr+= stride;

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

  634.                 }

  635.                 CHECK_SAD_HALF_MV(x2 , +1,  0)

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

  637.             }

  638.             CHECK_SAD_HALF_MV(y2 ,  0, +1)

  639.         }

  640.         mx+=dx;

  641.         my+=dy;

  642. 

  643.     }else{

  644.         mx<<=1;

  645.         my<<=1;

  646.     }

  647. 

  648.     *mx_ptr = mx;

  649.     *my_ptr = my;

  650.     return dminh;

  651. }

  652. 

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

  654. {

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

  656.     

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

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

  659. 

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

  661.     if(mv4){

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

  663. 

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

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

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

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

  668. 

  669.         mot_xy += s->b8_stride;

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

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

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

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

  674.     }

  675. }

  676. 

  677. /**

  678.  * get fullpel ME search limits.

  679.  */

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

  681. {

  682. /*

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

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

  685. */

  686.     if (s->unrestricted_mv) {

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

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

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

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

  691.     } else {

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

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

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

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

  696.     }

  697. }

  698. 

  699. static inline void init_mv4_ref(MpegEncContext *s){

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

  701.     const int stride= s->linesize;

  702. 

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

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

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

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

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

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

  709. }

  710. 

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

  712. {

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

  714.     const int size= 1;

  715.     const int h=8;

  716.     int block;

  717.     int P[10][2];

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

  719.     int same=1;

  720.     const int stride= s->linesize;

  721.     const int uvstride= s->uvlinesize;

  722.     uint8_t *mv_penalty= s->me.current_mv_penalty;

  723. 

  724.     init_mv4_ref(s);

  725.     

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

  727.         int mx4, my4;

  728.         int pred_x4, pred_y4;

  729.         int dmin4;

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

  731.         const int mot_stride = s->b8_stride;

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

  733. 

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

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

  736. 

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

  738. 

  739.         /* special case for first line */

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

  741.             s->me.pred_x= pred_x4= P_LEFT[0];

  742.             s->me.pred_y= pred_y4= P_LEFT[1];

  743.         } else {

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

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

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

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

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

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

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

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

  752.     

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

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

  755. 

  756.             s->me.pred_x= pred_x4 = P_MEDIAN[0];

  757.             s->me.pred_y= pred_y4 = P_MEDIAN[1];

  758.         }

  759.         P_MV1[0]= mx;

  760.         P_MV1[1]= my;

  761. 

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

  763. 

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

  765.         

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

  767.             int dxy;

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

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

  770.             if(s->quarter_sample){

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

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

  773. 

  774.                 if(s->no_rounding)

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

  776.                 else

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

  778.             }else{

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

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

  781. 

  782.                 if(s->no_rounding)

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

  784.                 else

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

  786.             }

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

  788.         }else

  789.             dmin_sum+= dmin4;

  790. 

  791.         if(s->quarter_sample){

  792.             mx4_sum+= mx4/2;

  793.             my4_sum+= my4/2;

  794.         }else{

  795.             mx4_sum+= mx4;

  796.             my4_sum+= my4;

  797.         }

  798.             

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

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

  801. 

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

  803.     }

  804.     

  805.     if(same)

  806.         return INT_MAX;

  807.     

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

  809.         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);

  810.     }

  811.     

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

  813.         int dxy;

  814.         int mx, my;

  815.         int offset;

  816. 

  817.         mx= ff_h263_round_chroma(mx4_sum);

  818.         my= ff_h263_round_chroma(my4_sum);

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

  820.         

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

  822.        

  823.         if(s->no_rounding){

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

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

  826.         }else{

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

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

  829.         }

  830. 

  831.         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);

  832.         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);

  833.     }

  834.     

  835.     s->me.pred_x= mx;

  836.     s->me.pred_y= my;

  837. 

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

  839.     /*case FF_CMP_SSE:

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

  841.     case FF_CMP_RD:

  842.         return dmin_sum;

  843.     default:

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

  845.     }

  846. }

  847. 

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

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

  850. 

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

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

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

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

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

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

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

  858.     }

  859. }

  860. 

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

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

  863. {

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

  865.     const int size=0;

  866.     const int h=8;

  867.     int block;

  868.     int P[10][2];

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

  870.     int same=1;

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

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

  873.     int dmin_sum= 0;

  874.     const int mot_stride= s->mb_stride;

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

  876.     

  877.     c->ymin>>=1;

  878.     c->ymax>>=1;

  879.     c->stride<<=1;

  880.     c->uvstride<<=1;

  881.     init_interlaced_ref(s, ref_index);

  882.     

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

  884.         int field_select;

  885.         int best_dmin= INT_MAX;

  886.         int best_field= -1;

  887. 

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

  889.             int dmin, mx_i, my_i;

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

  891.             

  892.             if(user_field_select){

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

  894.                     continue;

  895.             }

  896.             

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

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

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

  900.             

  901.             s->me.pred_x= P_LEFT[0];

  902.             s->me.pred_y= P_LEFT[1];

  903.             

  904.             if(!s->first_slice_line){

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

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

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

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

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

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

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

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

  913.     

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

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

  916.             }

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

  918.             P_MV1[1]= my / 2;

  919.             

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

  921. 

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

  923.             

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

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

  926.             

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

  928.                 int dxy;

  929. 

  930.                 //FIXME chroma ME

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

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

  933. 

  934.                 if(s->no_rounding){

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

  936.                 }else{

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

  938.                 }

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

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

  941.             }else

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

  943.                 

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

  945.             

  946.             if(dmin < best_dmin){

  947.                 best_dmin= dmin;

  948.                 best_field= field_select;

  949.             }

  950.         }

  951.         {

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

  953. 

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

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

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

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

  958.         }

  959. 

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

  961.         dmin_sum += best_dmin;

  962.     }

  963.     

  964.     c->ymin<<=1;

  965.     c->ymax<<=1;

  966.     c->stride>>=1;

  967.     c->uvstride>>=1;

  968. 

  969.     if(same)

  970.         return INT_MAX;

  971.     

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

  973.     /*case FF_CMP_SSE:

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

  975.     case FF_CMP_RD:

  976.         return dmin_sum;

  977.     default:

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

  979.     }

  980. }

  981. 

  982. static inline int check_input_motion(MpegEncContext * s, int mb_x, int mb_y, int p_type){

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

  984.     Picture *p= s->current_picture_ptr;

  985.     int mb_xy= mb_x + mb_y*s->mb_stride;

  986.     int xy= 2*mb_x + 2*mb_y*s->b8_stride;

  987.     int mb_type= s->current_picture.mb_type[mb_xy];

  988.     int flags= c->flags;

  989.     int shift= (flags&FLAG_QPEL) + 1;

  990.     int mask= (1<<shift)-1;

  991.     int x, y, i;

  992.     int d=0;

  993.     me_cmp_func cmpf= s->dsp.sse[0];

  994.     me_cmp_func chroma_cmpf= s->dsp.sse[1];

  995.     

  996.     assert(p_type==0 || !USES_LIST(mb_type, 1));

  997.     assert(IS_INTRA(mb_type) || USES_LIST(mb_type,0) || USES_LIST(mb_type,1));

  998.     

  999.     if(IS_INTERLACED(mb_type)){

  1000.         int xy2= xy  + s->b8_stride;

  1001.         s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_INTRA;

  1002.         c->stride<<=1;

  1003.         c->uvstride<<=1;

  1004.         

  1005.         assert(s->flags & CODEC_FLAG_INTERLACED_ME);

  1006. 

  1007.         if(USES_LIST(mb_type, 0)){

  1008.             int field_select0= p->ref_index[0][xy ];

  1009.             int field_select1= p->ref_index[0][xy2];

  1010.             assert(field_select0==0 ||field_select0==1);

  1011.             assert(field_select1==0 ||field_select1==1);

  1012.             init_interlaced_ref(s, 0);

  1013. 

  1014.             if(p_type){

  1015.                 s->p_field_select_table[0][mb_xy]= field_select0;

  1016.                 s->p_field_select_table[1][mb_xy]= field_select1;

  1017.                 *(uint32_t*)s->p_field_mv_table[0][field_select0][mb_xy]= *(uint32_t*)p->motion_val[0][xy ];

  1018.                 *(uint32_t*)s->p_field_mv_table[1][field_select1][mb_xy]= *(uint32_t*)p->motion_val[0][xy2];

  1019.                 s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_INTER_I;

  1020.             }else{

  1021.                 s->b_field_select_table[0][0][mb_xy]= field_select0;

  1022.                 s->b_field_select_table[0][1][mb_xy]= field_select1;

  1023.                 *(uint32_t*)s->b_field_mv_table[0][0][field_select0][mb_xy]= *(uint32_t*)p->motion_val[0][xy ];

  1024.                 *(uint32_t*)s->b_field_mv_table[0][1][field_select1][mb_xy]= *(uint32_t*)p->motion_val[0][xy2];

  1025.                 s->mb_type[mb_xy]= CANDIDATE_MB_TYPE_FORWARD_I;

  1026.             }

  1027. 

  1028.             x= p->motion_val[0][xy ][0]; 

  1029.             y= p->motion_val[0][xy ][1];

  1030.             d = cmp(s, x>>shift, y>>shift, x&mask, y&mask, 0, 8, field_select0, 0, cmpf, chroma_cmpf, flags);

  1031.             x= p->motion_val[0][xy2][0]; 

  1032.             y= p->motion_val[0][xy2][1];

  1033.             d+= cmp(s, x>>shift, y>>shift, x&mask, y&mask, 0, 8, field_select1, 1, cmpf, chroma_cmpf, flags);

  1034.         }

  1035.         if(USES_LIST(mb_type, 1)){

  1036.             int field_select0= p->ref_index[1][xy ];

  1037.             int field_select1= p->ref_index[1][xy2];

  1038.             assert(field_select0==0 ||field_select0==1);

  1039.             assert(field_select1==0 ||field_select1==1);

  1040.             init_interlaced_ref(s, 2);

  1041. 

  1042.             s->b_field_select_table[1][0][mb_xy]= field_select0;

  1043.             s->b_field_select_table[1][1][mb_xy]= field_select1;

  1044.             *(uint32_t*)s->b_field_mv_table[1][0][field_select0][mb_xy]= *(uint32_t*)p->motion_val[1][xy ];

  1045.             *(uint32_t*)s->b_field_mv_table[1][1][field_select1][mb_xy]= *(uint32_t*)p->motion_val[1][xy2];

  1046.             if(USES_LIST(mb_type, 0)){

  1047.                 s->mb_type[mb_xy]= CANDIDATE_MB_TYPE_BIDIR_I;

  1048.             }else{

  1049.                 s->mb_type[mb_xy]= CANDIDATE_MB_TYPE_BACKWARD_I;

  1050.             }

  1051. 

  1052.             x= p->motion_val[1][xy ][0]; 

  1053.             y= p->motion_val[1][xy ][1];

  1054.             d = cmp(s, x>>shift, y>>shift, x&mask, y&mask, 0, 8, field_select0+2, 0, cmpf, chroma_cmpf, flags);

  1055.             x= p->motion_val[1][xy2][0]; 

  1056.             y= p->motion_val[1][xy2][1];

  1057.             d+= cmp(s, x>>shift, y>>shift, x&mask, y&mask, 0, 8, field_select1+2, 1, cmpf, chroma_cmpf, flags);

  1058.             //FIXME bidir scores

  1059.         }

  1060.         c->stride>>=1;

  1061.         c->uvstride>>=1;

  1062.     }else if(IS_8X8(mb_type)){

  1063.         assert(s->flags & CODEC_FLAG_4MV);

  1064.         cmpf= s->dsp.sse[1];

  1065.         chroma_cmpf= s->dsp.sse[1];

  1066.         init_mv4_ref(s);

  1067.         for(i=0; i<4; i++){

  1068.             xy= s->block_index[i];

  1069.             x= p->motion_val[0][xy][0]; 

  1070.             y= p->motion_val[0][xy][1];

  1071.             d+= cmp(s, x>>shift, y>>shift, x&mask, y&mask, 1, 8, i, i, cmpf, chroma_cmpf, flags);

  1072.         }

  1073.         s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_INTER4V;

  1074.     }else{

  1075.         if(USES_LIST(mb_type, 0)){

  1076.             if(p_type){

  1077.                 *(uint32_t*)s->p_mv_table[mb_xy]= *(uint32_t*)p->motion_val[0][xy];

  1078.                 s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_INTER;

  1079.             }else if(USES_LIST(mb_type, 1)){

  1080.                 *(uint32_t*)s->b_bidir_forw_mv_table[mb_xy]= *(uint32_t*)p->motion_val[0][xy];

  1081.                 *(uint32_t*)s->b_bidir_back_mv_table[mb_xy]= *(uint32_t*)p->motion_val[1][xy];

  1082.                 s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_BIDIR;

  1083.             }else{

  1084.                 *(uint32_t*)s->b_forw_mv_table[mb_xy]= *(uint32_t*)p->motion_val[0][xy];

  1085.                 s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_FORWARD;

  1086.             }

  1087.             x= p->motion_val[0][xy][0]; 

  1088.             y= p->motion_val[0][xy][1];

  1089.             d = cmp(s, x>>shift, y>>shift, x&mask, y&mask, 0, 16, 0, 0, cmpf, chroma_cmpf, flags);

  1090.         }else if(USES_LIST(mb_type, 1)){

  1091.             *(uint32_t*)s->b_back_mv_table[mb_xy]= *(uint32_t*)p->motion_val[1][xy];

  1092.             s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_BACKWARD;

  1093.            

  1094.             x= p->motion_val[1][xy][0]; 

  1095.             y= p->motion_val[1][xy][1];

  1096.             d = cmp(s, x>>shift, y>>shift, x&mask, y&mask, 0, 16, 2, 0, cmpf, chroma_cmpf, flags);

  1097.         }else

  1098.             s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_INTRA;

  1099.     }

  1100.     return d;

  1101. }

  1102. 

  1103. void ff_estimate_p_frame_motion(MpegEncContext * s,

  1104.                                 int mb_x, int mb_y)

  1105. {

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

  1107.     uint8_t *pix, *ppix;

  1108.     int sum, varc, vard, mx, my, dmin;

  1109.     int P[10][2];

  1110.     const int shift= 1+s->quarter_sample;

  1111.     int mb_type=0;

  1112.     Picture * const pic= &s->current_picture;

  1113.     

  1114.     init_ref(s, s->new_picture.data, s->last_picture.data, NULL, 16*mb_x, 16*mb_y, 0);

  1115. 

  1116.     assert(s->quarter_sample==0 || s->quarter_sample==1);

  1117.     assert(s->linesize == s->me.stride);

  1118.     assert(s->uvlinesize == s->me.uvstride);

  1119. 

  1120.     s->me.penalty_factor    = get_penalty_factor(s, s->avctx->me_cmp);

  1121.     s->me.sub_penalty_factor= get_penalty_factor(s, s->avctx->me_sub_cmp);

  1122.     s->me.mb_penalty_factor = get_penalty_factor(s, s->avctx->mb_cmp);

  1123.     s->me.current_mv_penalty= s->me.mv_penalty[s->f_code] + MAX_MV;

  1124. 

  1125.     get_limits(s, 16*mb_x, 16*mb_y);

  1126.     s->me.skip=0;

  1127. 

  1128.     /* intra / predictive decision */

  1129.     pix = c->src[0][0];

  1130.     sum = s->dsp.pix_sum(pix, s->linesize);

  1131.     varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500 + 128)>>8;

  1132. 

  1133.     pic->mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;

  1134.     pic->mb_var [s->mb_stride * mb_y + mb_x] = varc;

  1135.     s->mb_var_sum_temp += varc;

  1136. 

  1137.     if(s->avctx->me_threshold){

  1138.         vard= (check_input_motion(s, mb_x, mb_y, 1)+128)>>8;

  1139.         

  1140.         if(vard<s->avctx->me_threshold){

  1141.             pic->mc_mb_var[s->mb_stride * mb_y + mb_x] = vard;

  1142.             s->mc_mb_var_sum_temp += vard;

  1143.             if (vard <= 64 || vard < varc) { //FIXME

  1144.                 s->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);

  1145.             }else{

  1146.                 s->scene_change_score+= s->qscale;

  1147.             }

  1148.             return;

  1149.         }

  1150.         if(vard<s->avctx->mb_threshold)

  1151.             mb_type= s->mb_type[mb_x + mb_y*s->mb_stride];

  1152.     }

  1153. 

  1154.     switch(s->me_method) {

  1155.     case ME_ZERO:

  1156.     default:

  1157. 	no_motion_search(s, &mx, &my);

  1158.         mx-= mb_x*16;

  1159.         my-= mb_y*16;

  1160.         dmin = 0;

  1161.         break;

  1162. #if 0

  1163.     case ME_FULL:

  1164. 	dmin = full_motion_search(s, &mx, &my, range, ref_picture);

  1165.         mx-= mb_x*16;

  1166.         my-= mb_y*16;

  1167.         break;

  1168.     case ME_LOG:

  1169. 	dmin = log_motion_search(s, &mx, &my, range / 2, ref_picture);

  1170.         mx-= mb_x*16;

  1171.         my-= mb_y*16;

  1172.         break;

  1173.     case ME_PHODS:

  1174. 	dmin = phods_motion_search(s, &mx, &my, range / 2, ref_picture);

  1175.         mx-= mb_x*16;

  1176.         my-= mb_y*16;

  1177.         break;

  1178. #endif

  1179.     case ME_X1:

  1180.     case ME_EPZS:

  1181.        {

  1182.             const int mot_stride = s->b8_stride;

  1183.             const int mot_xy = s->block_index[0];

  1184. 

  1185.             P_LEFT[0]       = s->current_picture.motion_val[0][mot_xy - 1][0];

  1186.             P_LEFT[1]       = s->current_picture.motion_val[0][mot_xy - 1][1];

  1187. 

  1188.             if(P_LEFT[0]       > (s->me.xmax<<shift)) P_LEFT[0]       = (s->me.xmax<<shift);

  1189. 

  1190.             if(!s->first_slice_line) {

  1191.                 P_TOP[0]      = s->current_picture.motion_val[0][mot_xy - mot_stride    ][0];

  1192.                 P_TOP[1]      = s->current_picture.motion_val[0][mot_xy - mot_stride    ][1];

  1193.                 P_TOPRIGHT[0] = s->current_picture.motion_val[0][mot_xy - mot_stride + 2][0];

  1194.                 P_TOPRIGHT[1] = s->current_picture.motion_val[0][mot_xy - mot_stride + 2][1];

  1195.                 if(P_TOP[1]      > (s->me.ymax<<shift)) P_TOP[1]     = (s->me.ymax<<shift);

  1196.                 if(P_TOPRIGHT[0] < (s->me.xmin<<shift)) P_TOPRIGHT[0]= (s->me.xmin<<shift);

  1197.                 if(P_TOPRIGHT[1] > (s->me.ymax<<shift)) P_TOPRIGHT[1]= (s->me.ymax<<shift);

  1198.         

  1199.                 P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);

  1200.                 P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);

  1201. 

  1202.                 if(s->out_format == FMT_H263){

  1203.                     c->pred_x = P_MEDIAN[0];

  1204.                     c->pred_y = P_MEDIAN[1];

  1205.                 }else { /* mpeg1 at least */

  1206.                     c->pred_x= P_LEFT[0];

  1207.                     c->pred_y= P_LEFT[1];

  1208.                 }

  1209.             }else{

  1210.                 c->pred_x= P_LEFT[0];

  1211.                 c->pred_y= P_LEFT[1];

  1212.             }

  1213. 

  1214.         }

  1215.         dmin = epzs_motion_search(s, &mx, &my, P, 0, 0, s->p_mv_table, (1<<16)>>shift);       

  1216. 

  1217.         break;

  1218.     }

  1219. 

  1220.     /* At this point (mx,my) are full-pell and the relative displacement */

  1221.     ppix = c->ref[0][0] + (my * s->linesize) + mx;

  1222.         

  1223.     vard = (s->dsp.sse[0](NULL, pix, ppix, s->linesize, 16)+128)>>8;

  1224. 

  1225.     pic->mc_mb_var[s->mb_stride * mb_y + mb_x] = vard;

  1226. //    pic->mb_cmp_score[s->mb_stride * mb_y + mb_x] = dmin; 

  1227.     s->mc_mb_var_sum_temp += vard;

  1228.     

  1229. #if 0

  1230.     printf("varc=%4d avg_var=%4d (sum=%4d) vard=%4d mx=%2d my=%2d\n",

  1231. 	   varc, s->avg_mb_var, sum, vard, mx - xx, my - yy);

  1232. #endif

  1233.     if(mb_type){

  1234.         if (vard <= 64 || vard < varc)

  1235.             s->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);

  1236.         else

  1237.             s->scene_change_score+= s->qscale;

  1238. 

  1239.         if(mb_type == CANDIDATE_MB_TYPE_INTER){

  1240.             s->me.sub_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);

  1241.             set_p_mv_tables(s, mx, my, 1);

  1242.         }else{

  1243.             mx <<=shift;

  1244.             my <<=shift;

  1245.         }

  1246.         if(mb_type == CANDIDATE_MB_TYPE_INTER4V){

  1247.             h263_mv4_search(s, mx, my, shift);

  1248. 

  1249.             set_p_mv_tables(s, mx, my, 0);

  1250.         }

  1251.         if(mb_type == CANDIDATE_MB_TYPE_INTER_I){

  1252.             interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my, 1);

  1253.         }

  1254.     }else if(s->avctx->mb_decision > FF_MB_DECISION_SIMPLE){

  1255.         if (vard <= 64 || vard < varc)

  1256.             s->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);

  1257.         else

  1258.             s->scene_change_score+= s->qscale;

  1259. 

  1260.         if (vard*2 + 200 > varc)

  1261.             mb_type|= CANDIDATE_MB_TYPE_INTRA;

  1262.         if (varc*2 + 200 > vard){

  1263.             mb_type|= CANDIDATE_MB_TYPE_INTER;

  1264.             s->me.sub_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);

  1265.             if(s->flags&CODEC_FLAG_MV0)

  1266.                 if(mx || my)

  1267.                     mb_type |= CANDIDATE_MB_TYPE_SKIPED; //FIXME check difference

  1268.         }else{

  1269.             mx <<=shift;

  1270.             my <<=shift;

  1271.         }

  1272.         if((s->flags&CODEC_FLAG_4MV)

  1273.            && !s->me.skip && varc>50 && vard>10){

  1274.             if(h263_mv4_search(s, mx, my, shift) < INT_MAX)

  1275.                 mb_type|=CANDIDATE_MB_TYPE_INTER4V;

  1276. 

  1277.             set_p_mv_tables(s, mx, my, 0);

  1278.         }else

  1279.             set_p_mv_tables(s, mx, my, 1);

  1280.         if((s->flags&CODEC_FLAG_INTERLACED_ME)

  1281.            && !s->me.skip){ //FIXME varc/d checks

  1282.             if(interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my, 0) < INT_MAX)

  1283.                 mb_type |= CANDIDATE_MB_TYPE_INTER_I;

  1284.         }

  1285.     }else{

  1286.         int intra_score, i;

  1287.         mb_type= CANDIDATE_MB_TYPE_INTER;

  1288. 

  1289.         dmin= s->me.sub_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);

  1290.         if(s->avctx->me_sub_cmp != s->avctx->mb_cmp && !s->me.skip)

  1291.             dmin= get_mb_score(s, mx, my, 0, 0);

  1292. 

  1293.         if((s->flags&CODEC_FLAG_4MV)

  1294.            && !s->me.skip && varc>50 && vard>10){

  1295.             int dmin4= h263_mv4_search(s, mx, my, shift);

  1296.             if(dmin4 < dmin){

  1297.                 mb_type= CANDIDATE_MB_TYPE_INTER4V;

  1298.                 dmin=dmin4;

  1299.             }

  1300.         }

  1301.         if((s->flags&CODEC_FLAG_INTERLACED_ME)

  1302.            && !s->me.skip){ //FIXME varc/d checks

  1303.             int dmin_i= interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my, 0);

  1304.             if(dmin_i < dmin){

  1305.                 mb_type = CANDIDATE_MB_TYPE_INTER_I;

  1306.                 dmin= dmin_i;

  1307.             }

  1308.         }

  1309.                 

  1310. //        pic->mb_cmp_score[s->mb_stride * mb_y + mb_x] = dmin; 

  1311.         set_p_mv_tables(s, mx, my, mb_type!=CANDIDATE_MB_TYPE_INTER4V);

  1312. 

  1313.         /* get intra luma score */

  1314.         if((s->avctx->mb_cmp&0xFF)==FF_CMP_SSE){

  1315.             intra_score= (varc<<8) - 500; //FIXME dont scale it down so we dont have to fix it

  1316.         }else{

  1317.             int mean= (sum+128)>>8;

  1318.             mean*= 0x01010101;

  1319.             

  1320.             for(i=0; i<16; i++){

  1321.                 *(uint32_t*)(&s->me.scratchpad[i*s->linesize+ 0]) = mean;

  1322.                 *(uint32_t*)(&s->me.scratchpad[i*s->linesize+ 4]) = mean;

  1323.                 *(uint32_t*)(&s->me.scratchpad[i*s->linesize+ 8]) = mean;

  1324.                 *(uint32_t*)(&s->me.scratchpad[i*s->linesize+12]) = mean;

  1325.             }

  1326. 

  1327.             intra_score= s->dsp.mb_cmp[0](s, s->me.scratchpad, pix, s->linesize, 16);

  1328.         }

  1329. #if 0 //FIXME

  1330.         /* get chroma score */

  1331.         if(s->avctx->mb_cmp&FF_CMP_CHROMA){

  1332.             for(i=1; i<3; i++){

  1333.                 uint8_t *dest_c;

  1334.                 int mean;

  1335.                 

  1336.                 if(s->out_format == FMT_H263){

  1337.                     mean= (s->dc_val[i][mb_x + mb_y*s->b8_stride] + 4)>>3; //FIXME not exact but simple ;)

  1338.                 }else{

  1339.                     mean= (s->last_dc[i] + 4)>>3;

  1340.                 }

  1341.                 dest_c = s->new_picture.data[i] + (mb_y * 8  * (s->uvlinesize)) + mb_x * 8;

  1342.                 

  1343.                 mean*= 0x01010101;

  1344.                 for(i=0; i<8; i++){

  1345.                     *(uint32_t*)(&s->me.scratchpad[i*s->uvlinesize+ 0]) = mean;

  1346.                     *(uint32_t*)(&s->me.scratchpad[i*s->uvlinesize+ 4]) = mean;

  1347.                 }

  1348.                 

  1349.                 intra_score+= s->dsp.mb_cmp[1](s, s->me.scratchpad, dest_c, s->uvlinesize);

  1350.             }                

  1351.         }

  1352. #endif

  1353.         intra_score += s->me.mb_penalty_factor*16;

  1354.         

  1355.         if(intra_score < dmin){

  1356.             mb_type= CANDIDATE_MB_TYPE_INTRA;

  1357.             s->current_picture.mb_type[mb_y*s->mb_stride + mb_x]= CANDIDATE_MB_TYPE_INTRA; //FIXME cleanup

  1358.         }else

  1359.             s->current_picture.mb_type[mb_y*s->mb_stride + mb_x]= 0;

  1360.         

  1361.         if (vard <= 64 || vard < varc) { //FIXME

  1362.             s->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);

  1363.         }else{

  1364.             s->scene_change_score+= s->qscale;

  1365.         }

  1366.     }

  1367. 

  1368.     s->mb_type[mb_y*s->mb_stride + mb_x]= mb_type;

  1369. }

  1370. 

  1371. int ff_pre_estimate_p_frame_motion(MpegEncContext * s,

  1372.                                     int mb_x, int mb_y)

  1373. {

  1374.     MotionEstContext * const c= &s->me;

  1375.     int mx, my, dmin;

  1376.     int P[10][2];

  1377.     const int shift= 1+s->quarter_sample;

  1378.     const int xy= mb_x + mb_y*s->mb_stride;

  1379.     init_ref(s, s->new_picture.data, s->last_picture.data, NULL, 16*mb_x, 16*mb_y, 0);

  1380.     

  1381.     assert(s->quarter_sample==0 || s->quarter_sample==1);

  1382. 

  1383.     s->me.pre_penalty_factor    = get_penalty_factor(s, s->avctx->me_pre_cmp);

  1384.     s->me.current_mv_penalty= s->me.mv_penalty[s->f_code] + MAX_MV;

  1385. 

  1386.     get_limits(s, 16*mb_x, 16*mb_y);

  1387.     s->me.skip=0;

  1388. 

  1389.     P_LEFT[0]       = s->p_mv_table[xy + 1][0];

  1390.     P_LEFT[1]       = s->p_mv_table[xy + 1][1];

  1391. 

  1392.     if(P_LEFT[0]       < (s->me.xmin<<shift)) P_LEFT[0]       = (s->me.xmin<<shift);

  1393. 

  1394.     /* special case for first line */

  1395.     if (s->first_slice_line) {

  1396.         c->pred_x= P_LEFT[0];

  1397.         c->pred_y= P_LEFT[1];

  1398.         P_TOP[0]= P_TOPRIGHT[0]= P_MEDIAN[0]=

  1399.         P_TOP[1]= P_TOPRIGHT[1]= P_MEDIAN[1]= 0; //FIXME 

  1400.     } else {

  1401.         P_TOP[0]      = s->p_mv_table[xy + s->mb_stride    ][0];

  1402.         P_TOP[1]      = s->p_mv_table[xy + s->mb_stride    ][1];

  1403.         P_TOPRIGHT[0] = s->p_mv_table[xy + s->mb_stride - 1][0];

  1404.         P_TOPRIGHT[1] = s->p_mv_table[xy + s->mb_stride - 1][1];

  1405.         if(P_TOP[1]      < (s->me.ymin<<shift)) P_TOP[1]     = (s->me.ymin<<shift);

  1406.         if(P_TOPRIGHT[0] > (s->me.xmax<<shift)) P_TOPRIGHT[0]= (s->me.xmax<<shift);

  1407.         if(P_TOPRIGHT[1] < (s->me.ymin<<shift)) P_TOPRIGHT[1]= (s->me.ymin<<shift);

  1408.     

  1409.         P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);

  1410.         P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);

  1411. 

  1412.         c->pred_x = P_MEDIAN[0];

  1413.         c->pred_y = P_MEDIAN[1];

  1414.     }

  1415. 

  1416.     dmin = epzs_motion_search(s, &mx, &my, P, 0, 0, s->p_mv_table, (1<<16)>>shift);       

  1417. 

  1418.     s->p_mv_table[xy][0] = mx<<shift;

  1419.     s->p_mv_table[xy][1] = my<<shift;

  1420.     

  1421.     return dmin;

  1422. }

  1423. 

  1424. static int ff_estimate_motion_b(MpegEncContext * s,

  1425.                        int mb_x, int mb_y, int16_t (*mv_table)[2], int ref_index, int f_code)

  1426. {

  1427.     int mx, my, dmin;

  1428.     int P[10][2];

  1429.     const int shift= 1+s->quarter_sample;

  1430.     const int mot_stride = s->mb_stride;

  1431.     const int mot_xy = mb_y*mot_stride + mb_x;

  1432.     uint8_t * const mv_penalty= s->me.mv_penalty[f_code] + MAX_MV;

  1433.     int mv_scale;

  1434.         

  1435.     s->me.penalty_factor    = get_penalty_factor(s, s->avctx->me_cmp);

  1436.     s->me.sub_penalty_factor= get_penalty_factor(s, s->avctx->me_sub_cmp);

  1437.     s->me.mb_penalty_factor = get_penalty_factor(s, s->avctx->mb_cmp);

  1438.     s->me.current_mv_penalty= mv_penalty;

  1439. 

  1440.     get_limits(s, 16*mb_x, 16*mb_y);

  1441. 

  1442.     switch(s->me_method) {

  1443.     case ME_ZERO:

  1444.     default:

  1445. 	no_motion_search(s, &mx, &my);

  1446.         dmin = 0;

  1447.         mx-= mb_x*16;

  1448.         my-= mb_y*16;

  1449.         break;

  1450. #if 0

  1451.     case ME_FULL:

  1452. 	dmin = full_motion_search(s, &mx, &my, range, ref_picture);

  1453.         mx-= mb_x*16;

  1454.         my-= mb_y*16;

  1455.         break;

  1456.     case ME_LOG:

  1457. 	dmin = log_motion_search(s, &mx, &my, range / 2, ref_picture);

  1458.         mx-= mb_x*16;

  1459.         my-= mb_y*16;

  1460.         break;

  1461.     case ME_PHODS:

  1462. 	dmin = phods_motion_search(s, &mx, &my, range / 2, ref_picture);

  1463.         mx-= mb_x*16;

  1464.         my-= mb_y*16;

  1465.         break;

  1466. #endif

  1467.     case ME_X1:

  1468.     case ME_EPZS:

  1469.        {

  1470.             P_LEFT[0]        = mv_table[mot_xy - 1][0];

  1471.             P_LEFT[1]        = mv_table[mot_xy - 1][1];

  1472. 

  1473.             if(P_LEFT[0]       > (s->me.xmax<<shift)) P_LEFT[0]       = (s->me.xmax<<shift);

  1474. 

  1475.             /* special case for first line */

  1476.             if (!s->first_slice_line) {

  1477.                 P_TOP[0] = mv_table[mot_xy - mot_stride             ][0];

  1478.                 P_TOP[1] = mv_table[mot_xy - mot_stride             ][1];

  1479.                 P_TOPRIGHT[0] = mv_table[mot_xy - mot_stride + 1         ][0];

  1480.                 P_TOPRIGHT[1] = mv_table[mot_xy - mot_stride + 1         ][1];

  1481.                 if(P_TOP[1] > (s->me.ymax<<shift)) P_TOP[1]= (s->me.ymax<<shift);

  1482.                 if(P_TOPRIGHT[0] < (s->me.xmin<<shift)) P_TOPRIGHT[0]= (s->me.xmin<<shift);

  1483.                 if(P_TOPRIGHT[1] > (s->me.ymax<<shift)) P_TOPRIGHT[1]= (s->me.ymax<<shift);

  1484.         

  1485.                 P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);

  1486.                 P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);

  1487.             }

  1488.             s->me.pred_x= P_LEFT[0];

  1489.             s->me.pred_y= P_LEFT[1];

  1490.         }

  1491.         

  1492.         if(mv_table == s->b_forw_mv_table){

  1493.             mv_scale= (s->pb_time<<16) / (s->pp_time<<shift);

  1494.         }else{

  1495.             mv_scale= ((s->pb_time - s->pp_time)<<16) / (s->pp_time<<shift);

  1496.         }

  1497.         

  1498.         dmin = epzs_motion_search(s, &mx, &my, P, 0, ref_index, s->p_mv_table, mv_scale);

  1499.  

  1500.         break;

  1501.     }

  1502.     

  1503.     dmin= s->me.sub_motion_search(s, &mx, &my, dmin, 0, ref_index, 0, 16);

  1504.                                    

  1505.     if(s->avctx->me_sub_cmp != s->avctx->mb_cmp && !s->me.skip)

  1506.         dmin= get_mb_score(s, mx, my, 0, ref_index);

  1507. 

  1508. //printf("%d %d %d %d//", s->mb_x, s->mb_y, mx, my);

  1509. //    s->mb_type[mb_y*s->mb_width + mb_x]= mb_type;

  1510.     mv_table[mot_xy][0]= mx;

  1511.     mv_table[mot_xy][1]= my;

  1512. 

  1513.     return dmin;

  1514. }

  1515. 

  1516. static inline int check_bidir_mv(MpegEncContext * s,

  1517.                    int motion_fx, int motion_fy,

  1518.                    int motion_bx, int motion_by,

  1519.                    int pred_fx, int pred_fy,

  1520.                    int pred_bx, int pred_by,

  1521.                    int size, int h)

  1522. {

  1523.     //FIXME optimize?

  1524.     //FIXME better f_code prediction (max mv & distance)

  1525.     //FIXME pointers

  1526.     MotionEstContext * const c= &s->me;

  1527.     uint8_t * const mv_penalty= s->me.mv_penalty[s->f_code] + MAX_MV; // f_code of the prev frame

  1528.     int stride= s->me.stride;

  1529.     int uvstride= s->me.uvstride;

  1530.     uint8_t *dest_y = s->me.scratchpad;

  1531.     uint8_t *ptr;

  1532.     int dxy;

  1533.     int src_x, src_y;

  1534.     int fbmin;

  1535.     uint8_t **src_data= c->src[0];

  1536.     uint8_t **ref_data= c->ref[0];

  1537.     uint8_t **ref2_data= c->ref[2];

  1538. 

  1539.     if(s->quarter_sample){

  1540.         dxy = ((motion_fy & 3) << 2) | (motion_fx & 3);

  1541.         src_x = motion_fx >> 2;

  1542.         src_y = motion_fy >> 2;

  1543. 

  1544.         ptr = ref_data[0] + (src_y * stride) + src_x;

  1545.         s->dsp.put_qpel_pixels_tab[0][dxy](dest_y    , ptr    , stride);

  1546. 

  1547.         dxy = ((motion_by & 3) << 2) | (motion_bx & 3);

  1548.         src_x = motion_bx >> 2;

  1549.         src_y = motion_by >> 2;

  1550.     

  1551.         ptr = ref2_data[0] + (src_y * stride) + src_x;

  1552.         s->dsp.avg_qpel_pixels_tab[size][dxy](dest_y    , ptr    , stride);

  1553.     }else{

  1554.         dxy = ((motion_fy & 1) << 1) | (motion_fx & 1);

  1555.         src_x = motion_fx >> 1;

  1556.         src_y = motion_fy >> 1;

  1557. 

  1558.         ptr = ref_data[0] + (src_y * stride) + src_x;

  1559.         s->dsp.put_pixels_tab[size][dxy](dest_y    , ptr    , stride, h);

  1560. 

  1561.         dxy = ((motion_by & 1) << 1) | (motion_bx & 1);

  1562.         src_x = motion_bx >> 1;

  1563.         src_y = motion_by >> 1;

  1564.     

  1565.         ptr = ref2_data[0] + (src_y * stride) + src_x;

  1566.         s->dsp.avg_pixels_tab[size][dxy](dest_y    , ptr    , stride, h);

  1567.     }

  1568. 

  1569.     fbmin = (mv_penalty[motion_fx-pred_fx] + mv_penalty[motion_fy-pred_fy])*s->me.mb_penalty_factor

  1570.            +(mv_penalty[motion_bx-pred_bx] + mv_penalty[motion_by-pred_by])*s->me.mb_penalty_factor

  1571.            + s->dsp.mb_cmp[size](s, src_data[0], dest_y, stride, h); //FIXME new_pic

  1572.            

  1573.     if(s->avctx->mb_cmp&FF_CMP_CHROMA){

  1574.     }

  1575.     //FIXME CHROMA !!!

  1576.            

  1577.     return fbmin;

  1578. }

  1579. 

  1580. /* refine the bidir vectors in hq mode and return the score in both lq & hq mode*/

  1581. static inline int bidir_refine(MpegEncContext * s, int mb_x, int mb_y)

  1582. {

  1583.     const int mot_stride = s->mb_stride;

  1584.     const int xy = mb_y *mot_stride + mb_x;

  1585.     int fbmin;

  1586.     int pred_fx= s->b_bidir_forw_mv_table[xy-1][0];

  1587.     int pred_fy= s->b_bidir_forw_mv_table[xy-1][1];

  1588.     int pred_bx= s->b_bidir_back_mv_table[xy-1][0];

  1589.     int pred_by= s->b_bidir_back_mv_table[xy-1][1];

  1590.     int motion_fx= s->b_bidir_forw_mv_table[xy][0]= s->b_forw_mv_table[xy][0];

  1591.     int motion_fy= s->b_bidir_forw_mv_table[xy][1]= s->b_forw_mv_table[xy][1];

  1592.     int motion_bx= s->b_bidir_back_mv_table[xy][0]= s->b_back_mv_table[xy][0];

  1593.     int motion_by= s->b_bidir_back_mv_table[xy][1]= s->b_back_mv_table[xy][1];

  1594. 

  1595.     //FIXME do refinement and add flag

  1596.     

  1597.     fbmin= check_bidir_mv(s, motion_fx, motion_fy,

  1598.                           motion_bx, motion_by,

  1599.                           pred_fx, pred_fy,

  1600.                           pred_bx, pred_by,

  1601.                           0, 16);

  1602. 

  1603.    return fbmin;

  1604. }

  1605. 

  1606. static inline int direct_search(MpegEncContext * s, int mb_x, int mb_y)

  1607. {

  1608.     int P[10][2];

  1609.     const int mot_stride = s->mb_stride;

  1610.     const int mot_xy = mb_y*mot_stride + mb_x;

  1611.     const int shift= 1+s->quarter_sample;

  1612.     int dmin, i;

  1613.     const int time_pp= s->pp_time;

  1614.     const int time_pb= s->pb_time;

  1615.     int mx, my, xmin, xmax, ymin, ymax;

  1616.     int16_t (*mv_table)[2]= s->b_direct_mv_table;

  1617.     

  1618.     s->me.current_mv_penalty= s->me.mv_penalty[1] + MAX_MV;

  1619.     ymin= xmin=(-32)>>shift;

  1620.     ymax= xmax=   31>>shift;

  1621. 

  1622.     if(IS_8X8(s->next_picture.mb_type[mot_xy])){

  1623.         s->mv_type= MV_TYPE_8X8;

  1624.     }else{

  1625.         s->mv_type= MV_TYPE_16X16;

  1626.     }

  1627. 

  1628.     for(i=0; i<4; i++){

  1629.         int index= s->block_index[i];

  1630.         int min, max;

  1631.     

  1632.         s->me.co_located_mv[i][0]= s->next_picture.motion_val[0][index][0];

  1633.         s->me.co_located_mv[i][1]= s->next_picture.motion_val[0][index][1];

  1634.         s->me.direct_basis_mv[i][0]= s->me.co_located_mv[i][0]*time_pb/time_pp + ((i& 1)<<(shift+3));

  1635.         s->me.direct_basis_mv[i][1]= s->me.co_located_mv[i][1]*time_pb/time_pp + ((i>>1)<<(shift+3));

  1636. //        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);

  1637. //        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);

  1638. 

  1639.         max= FFMAX(s->me.direct_basis_mv[i][0], s->me.direct_basis_mv[i][0] - s->me.co_located_mv[i][0])>>shift;

  1640.         min= FFMIN(s->me.direct_basis_mv[i][0], s->me.direct_basis_mv[i][0] - s->me.co_located_mv[i][0])>>shift;

  1641.         max+= 16*mb_x + 1; // +-1 is for the simpler rounding

  1642.         min+= 16*mb_x - 1;

  1643.         xmax= FFMIN(xmax, s->width - max);

  1644.         xmin= FFMAX(xmin, - 16     - min);

  1645. 

  1646.         max= FFMAX(s->me.direct_basis_mv[i][1], s->me.direct_basis_mv[i][1] - s->me.co_located_mv[i][1])>>shift;

  1647.         min= FFMIN(s->me.direct_basis_mv[i][1], s->me.direct_basis_mv[i][1] - s->me.co_located_mv[i][1])>>shift;

  1648.         max+= 16*mb_y + 1; // +-1 is for the simpler rounding

  1649.         min+= 16*mb_y - 1;

  1650.         ymax= FFMIN(ymax, s->height - max);

  1651.         ymin= FFMAX(ymin, - 16      - min);

  1652.         

  1653.         if(s->mv_type == MV_TYPE_16X16) break;

  1654.     }

  1655.     

  1656.     assert(xmax <= 15 && ymax <= 15 && xmin >= -16 && ymin >= -16);

  1657.     

  1658.     if(xmax < 0 || xmin >0 || ymax < 0 || ymin > 0){

  1659.         s->b_direct_mv_table[mot_xy][0]= 0;

  1660.         s->b_direct_mv_table[mot_xy][1]= 0;

  1661. 

  1662.         return 256*256*256*64;

  1663.     }

  1664.     

  1665.     s->me.xmin= xmin;

  1666.     s->me.ymin= ymin;

  1667.     s->me.xmax= xmax;

  1668.     s->me.ymax= ymax;

  1669.     s->me.flags     |= FLAG_DIRECT;

  1670.     s->me.sub_flags |= FLAG_DIRECT;

  1671.     s->me.pred_x=0;

  1672.     s->me.pred_y=0;

  1673. 

  1674.     P_LEFT[0]        = clip(mv_table[mot_xy - 1][0], xmin<<shift, xmax<<shift);

  1675.     P_LEFT[1]        = clip(mv_table[mot_xy - 1][1], ymin<<shift, ymax<<shift);

  1676. 

  1677.     /* special case for first line */

  1678.     if (!s->first_slice_line) { //FIXME maybe allow this over thread boundary as its cliped

  1679.         P_TOP[0]      = clip(mv_table[mot_xy - mot_stride             ][0], xmin<<shift, xmax<<shift);

  1680.         P_TOP[1]      = clip(mv_table[mot_xy - mot_stride             ][1], ymin<<shift, ymax<<shift);

  1681.         P_TOPRIGHT[0] = clip(mv_table[mot_xy - mot_stride + 1         ][0], xmin<<shift, xmax<<shift);

  1682.         P_TOPRIGHT[1] = clip(mv_table[mot_xy - mot_stride + 1         ][1], ymin<<shift, ymax<<shift);

  1683.     

  1684.         P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);

  1685.         P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);

  1686.     }

  1687.  

  1688.     dmin = epzs_motion_search(s, &mx, &my, P, 0, 0, mv_table, 1<<(16-shift));

  1689.     if(s->me.sub_flags&FLAG_QPEL) 

  1690.         dmin = qpel_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);

  1691.     else

  1692.         dmin = hpel_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);

  1693.     

  1694.     if(s->avctx->me_sub_cmp != s->avctx->mb_cmp && !s->me.skip)

  1695.         dmin= get_mb_score(s, mx, my, 0, 0);

  1696.     

  1697.     get_limits(s, 16*mb_x, 16*mb_y); //restore s->me.?min/max, maybe not needed

  1698. 

  1699.     s->b_direct_mv_table[mot_xy][0]= mx;

  1700.     s->b_direct_mv_table[mot_xy][1]= my;

  1701.     s->me.flags     &= ~FLAG_DIRECT;

  1702.     s->me.sub_flags &= ~FLAG_DIRECT;

  1703. 

  1704.     return dmin;

  1705. }

  1706. 

  1707. void ff_estimate_b_frame_motion(MpegEncContext * s,

  1708.                              int mb_x, int mb_y)

  1709. {

  1710.     const int penalty_factor= s->me.mb_penalty_factor;

  1711.     int fmin, bmin, dmin, fbmin, bimin, fimin;

  1712.     int type=0;

  1713.     const int xy = mb_y*s->mb_stride + mb_x;

  1714.     init_ref(s, s->new_picture.data, s->last_picture.data, s->next_picture.data, 16*mb_x, 16*mb_y, 2);

  1715. 

  1716.     

  1717.     s->me.skip=0;

  1718.     if(s->avctx->me_threshold){

  1719.         int vard= (check_input_motion(s, mb_x, mb_y, 0)+128)>>8;

  1720.         

  1721.         if(vard<s->avctx->me_threshold){

  1722. //            pix = c->src[0][0];

  1723. //            sum = s->dsp.pix_sum(pix, s->linesize);

  1724. //            varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500 + 128)>>8;

  1725.         

  1726. //            pic->mb_var   [s->mb_stride * mb_y + mb_x] = varc;

  1727.              s->current_picture.mc_mb_var[s->mb_stride * mb_y + mb_x] = vard;

  1728. /*            pic->mb_mean  [s->mb_stride * mb_y + mb_x] = (sum+128)>>8;

  1729.             s->mb_var_sum_temp    += varc;*/

  1730.             s->mc_mb_var_sum_temp += vard;

  1731. /*            if (vard <= 64 || vard < varc) {

  1732.                 s->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);

  1733.             }else{

  1734.                 s->scene_change_score+= s->qscale;

  1735.             }*/

  1736.             return;

  1737.         }

  1738.         if(vard<s->avctx->mb_threshold){

  1739.             type= s->mb_type[mb_y*s->mb_stride + mb_x];

  1740.             if(type == CANDIDATE_MB_TYPE_DIRECT){

  1741.                 direct_search(s, mb_x, mb_y);

  1742.             }

  1743.             if(type == CANDIDATE_MB_TYPE_FORWARD || type == CANDIDATE_MB_TYPE_BIDIR){

  1744.                 s->me.skip=0;

  1745.                 ff_estimate_motion_b(s, mb_x, mb_y, s->b_forw_mv_table, 0, s->f_code);

  1746.             }

  1747.             if(type == CANDIDATE_MB_TYPE_BACKWARD || type == CANDIDATE_MB_TYPE_BIDIR){

  1748.                 s->me.skip=0;

  1749.                 ff_estimate_motion_b(s, mb_x, mb_y, s->b_back_mv_table, 2, s->b_code);

  1750.             }

  1751.             if(type == CANDIDATE_MB_TYPE_FORWARD_I || type == CANDIDATE_MB_TYPE_BIDIR_I){

  1752.                 s->me.skip=0;

  1753.                 s->me.current_mv_penalty= s->me.mv_penalty[s->f_code] + MAX_MV;

  1754.                 interlaced_search(s, 0,

  1755.                                         s->b_field_mv_table[0], s->b_field_select_table[0],

  1756.                                         s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1], 1);

  1757.             }

  1758.             if(type == CANDIDATE_MB_TYPE_BACKWARD_I || type == CANDIDATE_MB_TYPE_BIDIR_I){

  1759.                 s->me.skip=0;

  1760.                 s->me.current_mv_penalty= s->me.mv_penalty[s->b_code] + MAX_MV;

  1761.                 interlaced_search(s, 2,

  1762.                                         s->b_field_mv_table[1], s->b_field_select_table[1],

  1763.                                         s->b_back_mv_table[xy][0], s->b_back_mv_table[xy][1], 1);

  1764.             }

  1765.             return;

  1766.         }

  1767.     }

  1768. 

  1769.     if (s->codec_id == CODEC_ID_MPEG4)

  1770.         dmin= direct_search(s, mb_x, mb_y);

  1771.     else

  1772.         dmin= INT_MAX;

  1773. //FIXME penalty stuff for non mpeg4

  1774.     s->me.skip=0;

  1775.     fmin= ff_estimate_motion_b(s, mb_x, mb_y, s->b_forw_mv_table, 0, s->f_code) + 3*penalty_factor;

  1776.     

  1777.     s->me.skip=0;

  1778.     bmin= ff_estimate_motion_b(s, mb_x, mb_y, s->b_back_mv_table, 2, s->b_code) + 2*penalty_factor;

  1779. //printf(" %d %d ", s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1]);

  1780. 

  1781.     s->me.skip=0;

  1782.     fbmin= bidir_refine(s, mb_x, mb_y) + penalty_factor;

  1783. //printf("%d %d %d %d\n", dmin, fmin, bmin, fbmin);

  1784.     

  1785.     if(s->flags & CODEC_FLAG_INTERLACED_ME){

  1786. //FIXME mb type penalty

  1787.         s->me.skip=0;

  1788.         s->me.current_mv_penalty= s->me.mv_penalty[s->f_code] + MAX_MV;

  1789.         fimin= interlaced_search(s, 0,

  1790.                                  s->b_field_mv_table[0], s->b_field_select_table[0],

  1791.                                  s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1], 0);

  1792.         s->me.current_mv_penalty= s->me.mv_penalty[s->b_code] + MAX_MV;

  1793.         bimin= interlaced_search(s, 2,

  1794.                                  s->b_field_mv_table[1], s->b_field_select_table[1],

  1795.                                  s->b_back_mv_table[xy][0], s->b_back_mv_table[xy][1], 0);

  1796.     }else

  1797.         fimin= bimin= INT_MAX;

  1798. 

  1799.     {

  1800.         int score= fmin;

  1801.         type = CANDIDATE_MB_TYPE_FORWARD;

  1802.         

  1803.         if (dmin <= score){

  1804.             score = dmin;

  1805.             type = CANDIDATE_MB_TYPE_DIRECT;

  1806.         }

  1807.         if(bmin<score){

  1808.             score=bmin;

  1809.             type= CANDIDATE_MB_TYPE_BACKWARD; 

  1810.         }

  1811.         if(fbmin<score){

  1812.             score=fbmin;

  1813.             type= CANDIDATE_MB_TYPE_BIDIR;

  1814.         }

  1815.         if(fimin<score){

  1816.             score=fimin;

  1817.             type= CANDIDATE_MB_TYPE_FORWARD_I;

  1818.         }

  1819.         if(bimin<score){

  1820.             score=bimin;

  1821.             type= CANDIDATE_MB_TYPE_BACKWARD_I;

  1822.         }

  1823.         

  1824.         score= ((unsigned)(score*score + 128*256))>>16;

  1825.         s->mc_mb_var_sum_temp += score;

  1826.         s->current_picture.mc_mb_var[mb_y*s->mb_stride + mb_x] = score; //FIXME use SSE

  1827.     }

  1828. 

  1829.     if(s->avctx->mb_decision > FF_MB_DECISION_SIMPLE){

  1830.         type= CANDIDATE_MB_TYPE_FORWARD | CANDIDATE_MB_TYPE_BACKWARD | CANDIDATE_MB_TYPE_BIDIR | CANDIDATE_MB_TYPE_DIRECT;

  1831.         if(fimin < INT_MAX)

  1832.             type |= CANDIDATE_MB_TYPE_FORWARD_I;

  1833.         if(bimin < INT_MAX)

  1834.             type |= CANDIDATE_MB_TYPE_BACKWARD_I;

  1835.         if(fimin < INT_MAX && bimin < INT_MAX){

  1836.             type |= CANDIDATE_MB_TYPE_BIDIR_I;

  1837.         }

  1838.          //FIXME something smarter

  1839.         if(dmin>256*256*16) type&= ~CANDIDATE_MB_TYPE_DIRECT; //dont try direct mode if its invalid for this MB

  1840. #if 0        

  1841.         if(s->out_format == FMT_MPEG1)

  1842.             type |= CANDIDATE_MB_TYPE_INTRA;

  1843. #endif

  1844.     }

  1845. 

  1846.     s->mb_type[mb_y*s->mb_stride + mb_x]= type;

  1847. }

  1848. 

  1849. /* find best f_code for ME which do unlimited searches */

  1850. int ff_get_best_fcode(MpegEncContext * s, int16_t (*mv_table)[2], int type)

  1851. {

  1852.     if(s->me_method>=ME_EPZS){

  1853.         int score[8];

  1854.         int i, y;

  1855.         uint8_t * fcode_tab= s->fcode_tab;

  1856.         int best_fcode=-1;

  1857.         int best_score=-10000000;

  1858. 

  1859.         for(i=0; i<8; i++) score[i]= s->mb_num*(8-i);

  1860. 

  1861.         for(y=0; y<s->mb_height; y++){

  1862.             int x;

  1863.             int xy= y*s->mb_stride;

  1864.             for(x=0; x<s->mb_width; x++){

  1865.                 if(s->mb_type[xy] & type){

  1866.                     int fcode= FFMAX(fcode_tab[mv_table[xy][0] + MAX_MV],

  1867.                                      fcode_tab[mv_table[xy][1] + MAX_MV]);

  1868.                     int j;

  1869.                     

  1870.                     for(j=0; j<fcode && j<8; j++){

  1871.                         if(s->pict_type==B_TYPE || s->current_picture.mc_mb_var[xy] < s->current_picture.mb_var[xy])

  1872.                             score[j]-= 170;

  1873.                     }

  1874.                 }

  1875.                 xy++;

  1876.             }

  1877.         }

  1878.         

  1879.         for(i=1; i<8; i++){

  1880.             if(score[i] > best_score){

  1881.                 best_score= score[i];

  1882.                 best_fcode= i;

  1883.             }

  1884. //            printf("%d %d\n", i, score[i]);

  1885.         }

  1886. 

  1887. //    printf("fcode: %d type: %d\n", i, s->pict_type);

  1888.         return best_fcode;

  1889. /*        for(i=0; i<=MAX_FCODE; i++){

  1890.             printf("%d ", mv_num[i]);

  1891.         }

  1892.         printf("\n");*/

  1893.     }else{

  1894.         return 1;

  1895.     }

  1896. }

  1897. 

  1898. void ff_fix_long_p_mvs(MpegEncContext * s)

  1899. {

  1900.     const int f_code= s->f_code;

  1901.     int y, range;

  1902.     assert(s->pict_type==P_TYPE);

  1903. 

  1904.     range = (((s->out_format == FMT_MPEG1) ? 8 : 16) << f_code);

  1905.     

  1906.     if(s->msmpeg4_version) range= 16;

  1907.     

  1908.     if(s->avctx->me_range && range > s->avctx->me_range) range= s->avctx->me_range;

  1909.     

  1910. //printf("%d no:%d %d//\n", clip, noclip, f_code);

  1911.     if(s->flags&CODEC_FLAG_4MV){

  1912.         const int wrap= s->b8_stride;

  1913. 

  1914.         /* clip / convert to intra 8x8 type MVs */

  1915.         for(y=0; y<s->mb_height; y++){

  1916.             int xy= y*2*wrap;

  1917.             int i= y*s->mb_stride;

  1918.             int x;

  1919. 

  1920.             for(x=0; x<s->mb_width; x++){

  1921.                 if(s->mb_type[i]&CANDIDATE_MB_TYPE_INTER4V){

  1922.                     int block;

  1923.                     for(block=0; block<4; block++){

  1924.                         int off= (block& 1) + (block>>1)*wrap;

  1925.                         int mx= s->current_picture.motion_val[0][ xy + off ][0];

  1926.                         int my= s->current_picture.motion_val[0][ xy + off ][1];

  1927. 

  1928.                         if(   mx >=range || mx <-range

  1929.                            || my >=range || my <-range){

  1930.                             s->mb_type[i] &= ~CANDIDATE_MB_TYPE_INTER4V;

  1931.                             s->mb_type[i] |= CANDIDATE_MB_TYPE_INTRA;

  1932.                             s->current_picture.mb_type[i]= CANDIDATE_MB_TYPE_INTRA;

  1933.                         }

  1934.                     }

  1935.                 }

  1936.                 xy+=2;

  1937.                 i++;

  1938.             }

  1939.         }

  1940.     }

  1941. }

  1942. 

  1943. /**

  1944.  *

  1945.  * @param truncate 1 for truncation, 0 for using intra

  1946.  */

  1947. void ff_fix_long_mvs(MpegEncContext * s, uint8_t *field_select_table, int field_select, 

  1948.                      int16_t (*mv_table)[2], int f_code, int type, int truncate)

  1949. {

  1950.     int y, h_range, v_range;

  1951. 

  1952.     // RAL: 8 in MPEG-1, 16 in MPEG-4

  1953.     int range = (((s->out_format == FMT_MPEG1) ? 8 : 16) << f_code);

  1954. 

  1955.     if(s->msmpeg4_version) range= 16;

  1956.     if(s->avctx->me_range && range > s->avctx->me_range) range= s->avctx->me_range;

  1957. 

  1958.     h_range= range;

  1959.     v_range= field_select_table ? range>>1 : range;

  1960. 

  1961.     /* clip / convert to intra 16x16 type MVs */

  1962.     for(y=0; y<s->mb_height; y++){

  1963.         int x;

  1964.         int xy= y*s->mb_stride;

  1965.         for(x=0; x<s->mb_width; x++){

  1966.             if (s->mb_type[xy] & type){    // RAL: "type" test added...

  1967.                 if(field_select_table==NULL || field_select_table[xy] == field_select){

  1968.                     if(   mv_table[xy][0] >=h_range || mv_table[xy][0] <-h_range

  1969.                        || mv_table[xy][1] >=v_range || mv_table[xy][1] <-v_range){

  1970. 

  1971.                         if(truncate){

  1972.                             if     (mv_table[xy][0] > h_range-1) mv_table[xy][0]=  h_range-1;

  1973.                             else if(mv_table[xy][0] < -h_range ) mv_table[xy][0]= -h_range;

  1974.                             if     (mv_table[xy][1] > v_range-1) mv_table[xy][1]=  v_range-1;

  1975.                             else if(mv_table[xy][1] < -v_range ) mv_table[xy][1]= -v_range;

  1976.                         }else{

  1977.                             s->mb_type[xy] &= ~type;

  1978.                             s->mb_type[xy] |= CANDIDATE_MB_TYPE_INTRA;

  1979.                             mv_table[xy][0]=

  1980.                             mv_table[xy][1]= 0;

  1981.                         }

  1982.                     }

  1983.                 }

  1984.             }

  1985.             xy++;

  1986.         }

  1987.     }

  1988. }