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.
3301 Commits
32 Branches
303MB
Tree: df70de1f61
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'df70de1f61'
${ noResults }
FFmpeg/libavcodec/motion_est.c

1937 lines
68KB
Raw Blame History 

  1. /*

  2.  * Motion estimation 

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

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

  5.  * 

  6.  *

  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 inline void init_mc(MpegEncContext *s, int size, int flags){

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

  107. 

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

  109.     if(flags & FLAG_CHROMA){

  110.         if(s->no_rounding) c->chroma_hpel_put= &s->dsp.put_no_rnd_pixels_tab[size+1];

  111.         else               c->chroma_hpel_put= &s->dsp.put_pixels_tab[size+1];

  112.     }

  113. 

  114.     if(flags & FLAG_QPEL){

  115.         c->qpel_avg= &s->dsp.avg_qpel_pixels_tab[size];

  116.         if(s->no_rounding) c->qpel_put= &s->dsp.put_no_rnd_qpel_pixels_tab[size];

  117.         else               c->qpel_put= &s->dsp.put_qpel_pixels_tab[size];

  118.     }else{

  119.         c->hpel_avg= &s->dsp.avg_pixels_tab[size];

  120.         if(s->no_rounding) c->hpel_put= &s->dsp.put_no_rnd_pixels_tab[size];

  121.         else               c->hpel_put= &s->dsp.put_pixels_tab[size];

  122.     }

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

  124. }

  125. 

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

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

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

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

  130.     const int stride= c->stride;

  131.     const int uvstride= c->uvstride;

  132.     const int qpel= flags&FLAG_QPEL;

  133.     const int chroma= flags&FLAG_CHROMA;

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

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

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

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

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

  139.     int d;

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

  141.     if(flags&FLAG_DIRECT){

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

  143.             const int time_pp= s->pp_time;

  144.             const int time_pb= s->pb_time;

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

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

  147.                 int i;

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

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

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

  151.                     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));

  152.                     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));

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

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

  155.         

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

  157.                     if(qpel){

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

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

  160.                     }else{

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

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

  163.                     }

  164.                 }

  165.             }else{

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

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

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

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

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

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

  172.                 

  173.                 if(qpel){

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

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

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

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

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

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

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

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

  182.                 }else{            

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

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

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

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

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

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

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

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

  191. 

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

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

  194.                 }

  195.             }

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

  197.         }else

  198.             d= 256*256*256*32;

  199.     }else{

  200.         int uvdxy;

  201.         if(dxy){

  202.             if(qpel){

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

  204.                 if(chroma){

  205.                     int cx= hx/2;

  206.                     int cy= hy/2;

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

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

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

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

  211.                 }

  212.             }else{

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

  214.                 if(chroma)

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

  216.             }

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

  218.         }else{

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

  220.             if(chroma)

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

  222.         }

  223.         if(chroma){

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

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

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

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

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

  229.         }

  230.     }

  231. #if 0

  232.     if(full_pel){

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

  234.         score_map[index]= d;

  235.     }

  236. 

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

  238. #endif

  239.     return d;

  240. }

  241. 

  242. #include "motion_est_template.c"

  243. 

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

  245.     switch(type&0xFF){

  246.     default:

  247.     case FF_CMP_SAD:

  248.         return s->qscale*2;

  249.     case FF_CMP_DCT:

  250.         return s->qscale*3;

  251.     case FF_CMP_SATD:

  252.         return s->qscale*6;

  253.     case FF_CMP_SSE:

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

  255.     case FF_CMP_BIT:

  256.         return 1;

  257.     case FF_CMP_RD:

  258.     case FF_CMP_PSNR:

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

  260.     }

  261. }

  262. 

  263. void ff_init_me(MpegEncContext *s){

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

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

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

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

  268.     

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

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

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

  272. 

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

  274.         s->me.sub_motion_search= qpel_motion_search;

  275.     }else{

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

  277.             s->me.sub_motion_search= hpel_motion_search;

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

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

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

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

  282.         else

  283.             s->me.sub_motion_search= hpel_motion_search;

  284.     }

  285.     if(s->linesize){

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

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

  288.     }else{

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

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

  291.     }

  292. }

  293.       

  294. #if 0

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

  296. {

  297.     int s, i, j;

  298. 

  299.     s = 0;

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

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

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

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

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

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

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

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

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

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

  310. 	    pix += 8;

  311. 	}

  312. 	pix += line_size - 16;

  313.     }

  314.     return s;

  315. }

  316. #endif

  317. 

  318. static inline void no_motion_search(MpegEncContext * s,

  319. 				    int *mx_ptr, int *my_ptr)

  320. {

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

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

  323. }

  324. 

  325. static int full_motion_search(MpegEncContext * s,

  326.                               int *mx_ptr, int *my_ptr, int range,

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

  328. {

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

  330.     int mx, my, dmin, d;

  331.     uint8_t *pix;

  332. 

  333.     xx = 16 * s->mb_x;

  334.     yy = 16 * s->mb_y;

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

  336.     if (x1 < xmin)

  337. 	x1 = xmin;

  338.     x2 = xx + range - 1;

  339.     if (x2 > xmax)

  340. 	x2 = xmax;

  341.     y1 = yy - range + 1;

  342.     if (y1 < ymin)

  343. 	y1 = ymin;

  344.     y2 = yy + range - 1;

  345.     if (y2 > ymax)

  346. 	y2 = ymax;

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

  348.     dmin = 0x7fffffff;

  349.     mx = 0;

  350.     my = 0;

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

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

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

  354. 			     s->linesize, 16);

  355. 	    if (d < dmin ||

  356. 		(d == dmin &&

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

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

  359. 		dmin = d;

  360. 		mx = x;

  361. 		my = y;

  362. 	    }

  363. 	}

  364.     }

  365. 

  366.     *mx_ptr = mx;

  367.     *my_ptr = my;

  368. 

  369. #if 0

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

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

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

  373.     }

  374. #endif

  375.     return dmin;

  376. }

  377. 

  378. 

  379. static int log_motion_search(MpegEncContext * s,

  380.                              int *mx_ptr, int *my_ptr, int range,

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

  382. {

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

  384.     int mx, my, dmin, d;

  385.     uint8_t *pix;

  386. 

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

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

  389. 

  390.     /* Left limit */

  391.     x1 = xx - range;

  392.     if (x1 < xmin)

  393. 	x1 = xmin;

  394. 

  395.     /* Right limit */

  396.     x2 = xx + range;

  397.     if (x2 > xmax)

  398. 	x2 = xmax;

  399. 

  400.     /* Upper limit */

  401.     y1 = yy - range;

  402.     if (y1 < ymin)

  403. 	y1 = ymin;

  404. 

  405.     /* Lower limit */

  406.     y2 = yy + range;

  407.     if (y2 > ymax)

  408. 	y2 = ymax;

  409. 

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

  411.     dmin = 0x7fffffff;

  412.     mx = 0;

  413.     my = 0;

  414. 

  415.     do {

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

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

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

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

  420. 		    dmin = d;

  421. 		    mx = x;

  422. 		    my = y;

  423. 		}

  424. 	    }

  425. 	}

  426. 

  427. 	range = range >> 1;

  428. 

  429. 	x1 = mx - range;

  430. 	if (x1 < xmin)

  431. 	    x1 = xmin;

  432. 

  433. 	x2 = mx + range;

  434. 	if (x2 > xmax)

  435. 	    x2 = xmax;

  436. 

  437. 	y1 = my - range;

  438. 	if (y1 < ymin)

  439. 	    y1 = ymin;

  440. 

  441. 	y2 = my + range;

  442. 	if (y2 > ymax)

  443. 	    y2 = ymax;

  444. 

  445.     } while (range >= 1);

  446. 

  447. #ifdef DEBUG

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

  449. #endif

  450.     *mx_ptr = mx;

  451.     *my_ptr = my;

  452.     return dmin;

  453. }

  454. 

  455. static int phods_motion_search(MpegEncContext * s,

  456.                                int *mx_ptr, int *my_ptr, int range,

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

  458. {

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

  460.     int mx, my, dminx, dminy;

  461.     uint8_t *pix;

  462. 

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

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

  465. 

  466.     /* Left limit */

  467.     x1 = xx - range;

  468.     if (x1 < xmin)

  469. 	x1 = xmin;

  470. 

  471.     /* Right limit */

  472.     x2 = xx + range;

  473.     if (x2 > xmax)

  474. 	x2 = xmax;

  475. 

  476.     /* Upper limit */

  477.     y1 = yy - range;

  478.     if (y1 < ymin)

  479. 	y1 = ymin;

  480. 

  481.     /* Lower limit */

  482.     y2 = yy + range;

  483.     if (y2 > ymax)

  484. 	y2 = ymax;

  485. 

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

  487.     mx = 0;

  488.     my = 0;

  489. 

  490.     x = xx;

  491.     y = yy;

  492.     do {

  493.         dminx = 0x7fffffff;

  494.         dminy = 0x7fffffff;

  495. 

  496. 	lastx = x;

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

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

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

  500. 		dminx = d;

  501. 		mx = x;

  502. 	    }

  503. 	}

  504. 

  505. 	x = lastx;

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

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

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

  509. 		dminy = d;

  510. 		my = y;

  511. 	    }

  512. 	}

  513. 

  514. 	range = range >> 1;

  515. 

  516. 	x = mx;

  517. 	y = my;

  518. 	x1 = mx - range;

  519. 	if (x1 < xmin)

  520. 	    x1 = xmin;

  521. 

  522. 	x2 = mx + range;

  523. 	if (x2 > xmax)

  524. 	    x2 = xmax;

  525. 

  526. 	y1 = my - range;

  527. 	if (y1 < ymin)

  528. 	    y1 = ymin;

  529. 

  530. 	y2 = my + range;

  531. 	if (y2 > ymax)

  532. 	    y2 = ymax;

  533. 

  534.     } while (range >= 1);

  535. 

  536. #ifdef DEBUG

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

  538. #endif

  539. 

  540.     /* half pixel search */

  541.     *mx_ptr = mx;

  542.     *my_ptr = my;

  543.     return dminy;

  544. }

  545. 

  546. 

  547. #define Z_THRESHOLD 256

  548. 

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

  550. {\

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

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

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

  554. }

  555. 

  556. static inline int sad_hpel_motion_search(MpegEncContext * s,

  557. 				  int *mx_ptr, int *my_ptr, int dmin,

  558.                                   int src_index, int ref_index,

  559.                                   int size, int h)

  560. {

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

  562.     int mx, my, dminh;

  563.     uint8_t *pix, *ptr;

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

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

  566.     LOAD_COMMON

  567.     

  568.     assert(flags == 0);

  569. 

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

  571. //    printf("S");

  572.         *mx_ptr = 0;

  573.         *my_ptr = 0;

  574.         return dmin;

  575.     }

  576. //    printf("N");

  577.         

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

  579. 

  580.     mx = *mx_ptr;

  581.     my = *my_ptr;

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

  583.     

  584.     dminh = dmin;

  585. 

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

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

  588.         int dx=0, dy=0;

  589.         int d, pen_x, pen_y; 

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

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

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

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

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

  595.         mx<<=1;

  596.         my<<=1;

  597. 

  598.         

  599.         pen_x= pred_x + mx;

  600.         pen_y= pred_y + my;

  601. 

  602.         ptr-= stride;

  603.         if(t<=b){

  604.             CHECK_SAD_HALF_MV(y2 , 0, -1)

  605.             if(l<=r){

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

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

  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.             }else{

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

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

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

  619.                     ptr+= stride;

  620.                 }else{

  621.                     ptr+= stride;

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

  623.                 }

  624.                 CHECK_SAD_HALF_MV(x2 , +1,  0)

  625.             }

  626.         }else{

  627.             if(l<=r){

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

  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.             }else{

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

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

  640.                     ptr+= stride;

  641.                 }else{

  642.                     ptr+= stride;

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

  644.                 }

  645.                 CHECK_SAD_HALF_MV(x2 , +1,  0)

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

  647.             }

  648.             CHECK_SAD_HALF_MV(y2 ,  0, +1)

  649.         }

  650.         mx+=dx;

  651.         my+=dy;

  652. 

  653.     }else{

  654.         mx<<=1;

  655.         my<<=1;

  656.     }

  657. 

  658.     *mx_ptr = mx;

  659.     *my_ptr = my;

  660.     return dminh;

  661. }

  662. 

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

  664. {

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

  666.     

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

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

  669. 

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

  671.     if(mv4){

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

  673. 

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

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

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

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

  678. 

  679.         mot_xy += s->b8_stride;

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

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

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

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

  684.     }

  685. }

  686. 

  687. /**

  688.  * get fullpel ME search limits.

  689.  */

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

  691. {

  692. /*

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

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

  695. */

  696.     if (s->unrestricted_mv) {

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

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

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

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

  701.     } else {

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

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

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

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

  706.     }

  707. }

  708. 

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

  710. {

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

  712.     const int size= 1;

  713.     const int h=8;

  714.     int block;

  715.     int P[10][2];

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

  717.     int same=1;

  718.     const int stride= s->linesize;

  719.     const int uvstride= s->uvlinesize;

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

  721. 

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

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

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

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

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

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

  728.     init_mc(s, size, c->flags);

  729.     

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

  731.         int mx4, my4;

  732.         int pred_x4, pred_y4;

  733.         int dmin4;

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

  735.         const int mot_stride = s->b8_stride;

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

  737. 

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

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

  740. 

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

  742. 

  743.         /* special case for first line */

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

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

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

  747.         } else {

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

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

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

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

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

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

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

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

  756.     

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

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

  759. 

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

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

  762.         }

  763.         P_MV1[0]= mx;

  764.         P_MV1[1]= my;

  765. 

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

  767. 

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

  769.         

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

  771.             int dxy;

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

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

  774.             if(s->quarter_sample){

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

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

  777. 

  778.                 if(s->no_rounding)

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

  780.                 else

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

  782.             }else{

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

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

  785. 

  786.                 if(s->no_rounding)

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

  788.                 else

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

  790.             }

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

  792.         }else

  793.             dmin_sum+= dmin4;

  794. 

  795.         if(s->quarter_sample){

  796.             mx4_sum+= mx4/2;

  797.             my4_sum+= my4/2;

  798.         }else{

  799.             mx4_sum+= mx4;

  800.             my4_sum+= my4;

  801.         }

  802.             

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

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

  805. 

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

  807.     }

  808.     

  809.     if(same)

  810.         return INT_MAX;

  811.     

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

  813.         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);

  814.     }

  815.     

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

  817.         int dxy;

  818.         int mx, my;

  819.         int offset;

  820. 

  821.         mx= ff_h263_round_chroma(mx4_sum);

  822.         my= ff_h263_round_chroma(my4_sum);

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

  824.         

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

  826.        

  827.         if(s->no_rounding){

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

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

  830.         }else{

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

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

  833.         }

  834. 

  835.         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);

  836.         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);

  837.     }

  838.     

  839.     s->me.pred_x= mx;

  840.     s->me.pred_y= my;

  841. 

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

  843.     /*case FF_CMP_SSE:

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

  845.     case FF_CMP_RD:

  846.         return dmin_sum;

  847.     default:

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

  849.     }

  850. }

  851. 

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

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

  854. {

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

  856.     const int size=0;

  857.     const int h=8;

  858.     int block;

  859.     int P[10][2];

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

  861.     int same=1;

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

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

  864.     int dmin_sum= 0;

  865.     const int mot_stride= s->mb_stride;

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

  867.     

  868.     c->ymin>>=1;

  869.     c->ymax>>=1;

  870.     c->stride<<=1;

  871.     c->uvstride<<=1;

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

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

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

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

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

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

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

  879.     }

  880.     init_mc(s, size, c->flags);

  881.     

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

  883.         int field_select;

  884.         int best_dmin= INT_MAX;

  885.         int best_field= -1;

  886. 

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

  888.             int dmin, mx_i, my_i;

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

  890.             

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

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

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

  894.             

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

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

  897.             

  898.             if(!s->first_slice_line){

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

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

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

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

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

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

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

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

  907.     

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

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

  910.             }

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

  912.             P_MV1[1]= my / 2;

  913.             

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

  915. 

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

  917.             

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

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

  920.             

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

  922.                 int dxy;

  923. 

  924.                 //FIXME chroma ME

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

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

  927. 

  928.                 if(s->no_rounding){

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

  930.                 }else{

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

  932.                 }

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

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

  935.             }else

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

  937.                 

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

  939.             

  940.             if(dmin < best_dmin){

  941.                 best_dmin= dmin;

  942.                 best_field= field_select;

  943.             }

  944.         }

  945.         {

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

  947. 

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

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

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

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

  952.         }

  953. 

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

  955.         dmin_sum += best_dmin;

  956.     }

  957.     

  958.     c->ymin<<=1;

  959.     c->ymax<<=1;

  960.     c->stride>>=1;

  961.     c->uvstride>>=1;

  962. 

  963.     if(same)

  964.         return INT_MAX;

  965.     

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

  967.     /*case FF_CMP_SSE:

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

  969.     case FF_CMP_RD:

  970.         return dmin_sum;

  971.     default:

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

  973.     }

  974. }

  975. 

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

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

  978.     Picture *p= s->current_picture_ptr;

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

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

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

  982.     int flags= c->flags;

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

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

  985.     int x, y;

  986.     int d=0;

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

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

  989.     

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

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

  992.     

  993.     if(IS_INTERLACED(mb_type)){

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

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

  996.         c->stride<<=1;

  997.         c->uvstride<<=1;

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

  999.         c->ref[3][0] = c->ref[2][0] + s->linesize;

  1000.         c->src[1][0] = c->src[0][0] + s->linesize;

  1001.         if(c->flags & FLAG_CHROMA){

  1002.             c->ref[1][1] = c->ref[0][1] + s->uvlinesize;

  1003.             c->ref[1][2] = c->ref[0][2] + s->uvlinesize;

  1004.             c->ref[3][1] = c->ref[2][1] + s->uvlinesize;

  1005.             c->ref[3][2] = c->ref[2][2] + s->uvlinesize;

  1006.             c->src[1][1] = c->src[0][1] + s->uvlinesize;

  1007.             c->src[1][2] = c->src[0][2] + s->uvlinesize;

  1008.         }

  1009.         if(USES_LIST(mb_type, 0)){

  1010.             int field_select0= p->ref_index[0][xy ];

  1011.             int field_select1= p->ref_index[0][xy2];

  1012.             assert(field_select0==0 ||field_select0==1);

  1013.             assert(field_select1==0 ||field_select1==1);

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

  1041.             s->b_field_select_table[1][1][mb_xy]= field_select1;

  1042.             *(uint32_t*)s->b_field_mv_table[1][0][field_select0][mb_xy]= *(uint32_t*)p->motion_val[1][xy ];

  1043.             *(uint32_t*)s->b_field_mv_table[1][1][field_select1][mb_xy]= *(uint32_t*)p->motion_val[1][xy2];

  1044.             if(USES_LIST(mb_type, 0)){

  1045.                 s->mb_type[mb_xy]= CANDIDATE_MB_TYPE_BIDIR_I;

  1046.             }else{

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

  1048.             }

  1049. 

  1050.             x= p->motion_val[1][xy ][0]; 

  1051.             y= p->motion_val[1][xy ][1];

  1052.             d = cmp(s, x>>shift, y>>shift, x&mask, y&mask, 0, 8, field_select0+2, 0, cmpf, chroma_cmpf, flags);

  1053.             x= p->motion_val[1][xy2][0]; 

  1054.             y= p->motion_val[1][xy2][1];

  1055.             d+= cmp(s, x>>shift, y>>shift, x&mask, y&mask, 0, 8, field_select1+2, 1, cmpf, chroma_cmpf, flags);

  1056.             //FIXME bidir scores

  1057.         }

  1058.         c->stride>>=1;

  1059.         c->uvstride>>=1;

  1060.     }else{

  1061.         if(USES_LIST(mb_type, 0)){

  1062.             if(p_type){

  1063.                 *(uint32_t*)s->p_mv_table[mb_xy]= *(uint32_t*)p->motion_val[0][xy];

  1064.                 s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_INTER;

  1065.             }else if(USES_LIST(mb_type, 1)){

  1066.                 *(uint32_t*)s->b_bidir_forw_mv_table[mb_xy]= *(uint32_t*)p->motion_val[0][xy];

  1067.                 *(uint32_t*)s->b_bidir_back_mv_table[mb_xy]= *(uint32_t*)p->motion_val[1][xy];

  1068.                 s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_BIDIR;

  1069.             }else{

  1070.                 *(uint32_t*)s->b_forw_mv_table[mb_xy]= *(uint32_t*)p->motion_val[0][xy];

  1071.                 s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_FORWARD;

  1072.             }

  1073.             x= p->motion_val[0][xy][0]; 

  1074.             y= p->motion_val[0][xy][1];

  1075.             d = cmp(s, x>>shift, y>>shift, x&mask, y&mask, 0, 16, 0, 0, cmpf, chroma_cmpf, flags);

  1076.         }else if(USES_LIST(mb_type, 1)){

  1077.             *(uint32_t*)s->b_back_mv_table[mb_xy]= *(uint32_t*)p->motion_val[1][xy];

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

  1079.            

  1080.             x= p->motion_val[1][xy][0]; 

  1081.             y= p->motion_val[1][xy][1];

  1082.             d = cmp(s, x>>shift, y>>shift, x&mask, y&mask, 0, 16, 2, 0, cmpf, chroma_cmpf, flags);

  1083.         }else

  1084.             s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_INTRA;

  1085.     }

  1086.     return d;

  1087. }

  1088. 

  1089. void ff_estimate_p_frame_motion(MpegEncContext * s,

  1090.                                 int mb_x, int mb_y)

  1091. {

  1092.     MotionEstContext * const c= &s->me;

  1093.     uint8_t *pix, *ppix;

  1094.     int sum, varc, vard, mx, my, dmin, xx, yy;

  1095.     int P[10][2];

  1096.     const int shift= 1+s->quarter_sample;

  1097.     int mb_type=0;

  1098.     Picture * const pic= &s->current_picture;

  1099.     

  1100.     init_ref(s, s->new_picture.data, s->last_picture.data, NULL, 16*mb_x, 16*mb_y, 0);

  1101.     init_mc(s, 0, c->flags);

  1102. 

  1103.     assert(s->quarter_sample==0 || s->quarter_sample==1);

  1104.     assert(s->linesize == s->me.stride);

  1105.     assert(s->uvlinesize == s->me.uvstride);

  1106. 

  1107.     s->me.penalty_factor    = get_penalty_factor(s, s->avctx->me_cmp);

  1108.     s->me.sub_penalty_factor= get_penalty_factor(s, s->avctx->me_sub_cmp);

  1109.     s->me.mb_penalty_factor = get_penalty_factor(s, s->avctx->mb_cmp);

  1110.     s->me.current_mv_penalty= s->me.mv_penalty[s->f_code] + MAX_MV;

  1111. 

  1112.     get_limits(s, 16*mb_x, 16*mb_y);

  1113.     s->me.skip=0;

  1114. 

  1115.     if(s->avctx->me_threshold){

  1116.         vard= (check_input_motion(s, mb_x, mb_y, 1)+128)>>8;

  1117.         

  1118.         if(vard<s->avctx->me_threshold){

  1119.             pix = c->src[0][0];

  1120.             sum = s->dsp.pix_sum(pix, s->linesize);

  1121.             varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500 + 128)>>8;

  1122.         

  1123.             pic->mb_var   [s->mb_stride * mb_y + mb_x] = varc;

  1124.             pic->mc_mb_var[s->mb_stride * mb_y + mb_x] = vard;

  1125.             pic->mb_mean  [s->mb_stride * mb_y + mb_x] = (sum+128)>>8;

  1126.             s->mb_var_sum_temp    += varc;

  1127.             s->mc_mb_var_sum_temp += vard;

  1128.             if (vard <= 64 || vard < varc) { //FIXME

  1129.                 s->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);

  1130.             }else{

  1131.                 s->scene_change_score+= s->qscale;

  1132.             }

  1133.             return;

  1134.         }

  1135.     }

  1136. 

  1137.     switch(s->me_method) {

  1138.     case ME_ZERO:

  1139.     default:

  1140. 	no_motion_search(s, &mx, &my);

  1141.         mx-= mb_x*16;

  1142.         my-= mb_y*16;

  1143.         dmin = 0;

  1144.         break;

  1145. #if 0

  1146.     case ME_FULL:

  1147. 	dmin = full_motion_search(s, &mx, &my, range, ref_picture);

  1148.         mx-= mb_x*16;

  1149.         my-= mb_y*16;

  1150.         break;

  1151.     case ME_LOG:

  1152. 	dmin = log_motion_search(s, &mx, &my, range / 2, ref_picture);

  1153.         mx-= mb_x*16;

  1154.         my-= mb_y*16;

  1155.         break;

  1156.     case ME_PHODS:

  1157. 	dmin = phods_motion_search(s, &mx, &my, range / 2, ref_picture);

  1158.         mx-= mb_x*16;

  1159.         my-= mb_y*16;

  1160.         break;

  1161. #endif

  1162.     case ME_X1:

  1163.     case ME_EPZS:

  1164.        {

  1165.             const int mot_stride = s->b8_stride;

  1166.             const int mot_xy = s->block_index[0];

  1167. 

  1168.             P_LEFT[0]       = s->current_picture.motion_val[0][mot_xy - 1][0];

  1169.             P_LEFT[1]       = s->current_picture.motion_val[0][mot_xy - 1][1];

  1170. 

  1171.             if(P_LEFT[0]       > (s->me.xmax<<shift)) P_LEFT[0]       = (s->me.xmax<<shift);

  1172. 

  1173.             if(!s->first_slice_line) {

  1174.                 P_TOP[0]      = s->current_picture.motion_val[0][mot_xy - mot_stride    ][0];

  1175.                 P_TOP[1]      = s->current_picture.motion_val[0][mot_xy - mot_stride    ][1];

  1176.                 P_TOPRIGHT[0] = s->current_picture.motion_val[0][mot_xy - mot_stride + 2][0];

  1177.                 P_TOPRIGHT[1] = s->current_picture.motion_val[0][mot_xy - mot_stride + 2][1];

  1178.                 if(P_TOP[1]      > (s->me.ymax<<shift)) P_TOP[1]     = (s->me.ymax<<shift);

  1179.                 if(P_TOPRIGHT[0] < (s->me.xmin<<shift)) P_TOPRIGHT[0]= (s->me.xmin<<shift);

  1180.                 if(P_TOPRIGHT[1] > (s->me.ymax<<shift)) P_TOPRIGHT[1]= (s->me.ymax<<shift);

  1181.         

  1182.                 P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);

  1183.                 P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);

  1184. 

  1185.                 if(s->out_format == FMT_H263){

  1186.                     c->pred_x = P_MEDIAN[0];

  1187.                     c->pred_y = P_MEDIAN[1];

  1188.                 }else { /* mpeg1 at least */

  1189.                     c->pred_x= P_LEFT[0];

  1190.                     c->pred_y= P_LEFT[1];

  1191.                 }

  1192.             }else{

  1193.                 c->pred_x= P_LEFT[0];

  1194.                 c->pred_y= P_LEFT[1];

  1195.             }

  1196. 

  1197.         }

  1198.         dmin = epzs_motion_search(s, &mx, &my, P, 0, 0, s->p_mv_table, (1<<16)>>shift);       

  1199. 

  1200.         break;

  1201.     }

  1202. 

  1203.     /* intra / predictive decision */

  1204.     xx = mb_x * 16;

  1205.     yy = mb_y * 16;

  1206. 

  1207.     pix = c->src[0][0];

  1208.     /* At this point (mx,my) are full-pell and the relative displacement */

  1209.     ppix = c->ref[0][0] + (my * s->linesize) + mx;

  1210.     

  1211.     sum = s->dsp.pix_sum(pix, s->linesize);

  1212.     

  1213.     varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500 + 128)>>8;

  1214.     vard = (s->dsp.sse[0](NULL, pix, ppix, s->linesize, 16)+128)>>8;

  1215. 

  1216. //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);

  1217.     pic->mb_var   [s->mb_stride * mb_y + mb_x] = varc;

  1218.     pic->mc_mb_var[s->mb_stride * mb_y + mb_x] = vard;

  1219.     pic->mb_mean  [s->mb_stride * mb_y + mb_x] = (sum+128)>>8;

  1220. //    pic->mb_cmp_score[s->mb_stride * mb_y + mb_x] = dmin; 

  1221.     s->mb_var_sum_temp    += varc;

  1222.     s->mc_mb_var_sum_temp += vard;

  1223. //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);

  1224.     

  1225. #if 0

  1226.     printf("varc=%4d avg_var=%4d (sum=%4d) vard=%4d mx=%2d my=%2d\n",

  1227. 	   varc, s->avg_mb_var, sum, vard, mx - xx, my - yy);

  1228. #endif

  1229.     if(s->avctx->mb_decision > FF_MB_DECISION_SIMPLE){

  1230.         if (vard <= 64 || vard < varc)

  1231.             s->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);

  1232.         else

  1233.             s->scene_change_score+= s->qscale;

  1234. 

  1235.         if (vard*2 + 200 > varc)

  1236.             mb_type|= CANDIDATE_MB_TYPE_INTRA;

  1237.         if (varc*2 + 200 > vard){

  1238.             mb_type|= CANDIDATE_MB_TYPE_INTER;

  1239.             s->me.sub_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);

  1240.             if(s->flags&CODEC_FLAG_MV0)

  1241.                 if(mx || my)

  1242.                     mb_type |= CANDIDATE_MB_TYPE_SKIPED; //FIXME check difference

  1243.         }else{

  1244.             mx <<=shift;

  1245.             my <<=shift;

  1246.         }

  1247.         if((s->flags&CODEC_FLAG_4MV)

  1248.            && !s->me.skip && varc>50 && vard>10){

  1249.             if(h263_mv4_search(s, mx, my, shift) < INT_MAX)

  1250.                 mb_type|=CANDIDATE_MB_TYPE_INTER4V;

  1251. 

  1252.             set_p_mv_tables(s, mx, my, 0);

  1253.         }else

  1254.             set_p_mv_tables(s, mx, my, 1);

  1255.         if((s->flags&CODEC_FLAG_INTERLACED_ME)

  1256.            && !s->me.skip){ //FIXME varc/d checks

  1257.             if(interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my) < INT_MAX)

  1258.                 mb_type |= CANDIDATE_MB_TYPE_INTER_I;

  1259.         }

  1260.     }else{

  1261.         int intra_score, i;

  1262.         mb_type= CANDIDATE_MB_TYPE_INTER;

  1263. 

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

  1265.         if(s->avctx->me_sub_cmp != s->avctx->mb_cmp && !s->me.skip)

  1266.             dmin= get_mb_score(s, mx, my, 0, 0);

  1267. 

  1268.         if((s->flags&CODEC_FLAG_4MV)

  1269.            && !s->me.skip && varc>50 && vard>10){

  1270.             int dmin4= h263_mv4_search(s, mx, my, shift);

  1271.             if(dmin4 < dmin){

  1272.                 mb_type= CANDIDATE_MB_TYPE_INTER4V;

  1273.                 dmin=dmin4;

  1274.             }

  1275.         }

  1276.         if((s->flags&CODEC_FLAG_INTERLACED_ME)

  1277.            && !s->me.skip){ //FIXME varc/d checks

  1278.             int dmin_i= interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my);

  1279.             if(dmin_i < dmin){

  1280.                 mb_type = CANDIDATE_MB_TYPE_INTER_I;

  1281.                 dmin= dmin_i;

  1282.             }

  1283.         }

  1284.                 

  1285. //        pic->mb_cmp_score[s->mb_stride * mb_y + mb_x] = dmin; 

  1286.         set_p_mv_tables(s, mx, my, mb_type!=CANDIDATE_MB_TYPE_INTER4V);

  1287. 

  1288.         /* get intra luma score */

  1289.         if((s->avctx->mb_cmp&0xFF)==FF_CMP_SSE){

  1290.             intra_score= (varc<<8) - 500; //FIXME dont scale it down so we dont have to fix it

  1291.         }else{

  1292.             int mean= (sum+128)>>8;

  1293.             mean*= 0x01010101;

  1294.             

  1295.             for(i=0; i<16; i++){

  1296.                 *(uint32_t*)(&s->me.scratchpad[i*s->linesize+ 0]) = mean;

  1297.                 *(uint32_t*)(&s->me.scratchpad[i*s->linesize+ 4]) = mean;

  1298.                 *(uint32_t*)(&s->me.scratchpad[i*s->linesize+ 8]) = mean;

  1299.                 *(uint32_t*)(&s->me.scratchpad[i*s->linesize+12]) = mean;

  1300.             }

  1301. 

  1302.             intra_score= s->dsp.mb_cmp[0](s, s->me.scratchpad, pix, s->linesize, 16);

  1303.         }

  1304. #if 0 //FIXME

  1305.         /* get chroma score */

  1306.         if(s->avctx->mb_cmp&FF_CMP_CHROMA){

  1307.             for(i=1; i<3; i++){

  1308.                 uint8_t *dest_c;

  1309.                 int mean;

  1310.                 

  1311.                 if(s->out_format == FMT_H263){

  1312.                     mean= (s->dc_val[i][mb_x + mb_y*s->b8_stride] + 4)>>3; //FIXME not exact but simple ;)

  1313.                 }else{

  1314.                     mean= (s->last_dc[i] + 4)>>3;

  1315.                 }

  1316.                 dest_c = s->new_picture.data[i] + (mb_y * 8  * (s->uvlinesize)) + mb_x * 8;

  1317.                 

  1318.                 mean*= 0x01010101;

  1319.                 for(i=0; i<8; i++){

  1320.                     *(uint32_t*)(&s->me.scratchpad[i*s->uvlinesize+ 0]) = mean;

  1321.                     *(uint32_t*)(&s->me.scratchpad[i*s->uvlinesize+ 4]) = mean;

  1322.                 }

  1323.                 

  1324.                 intra_score+= s->dsp.mb_cmp[1](s, s->me.scratchpad, dest_c, s->uvlinesize);

  1325.             }                

  1326.         }

  1327. #endif

  1328.         intra_score += s->me.mb_penalty_factor*16;

  1329.         

  1330.         if(intra_score < dmin){

  1331.             mb_type= CANDIDATE_MB_TYPE_INTRA;

  1332.             s->current_picture.mb_type[mb_y*s->mb_stride + mb_x]= CANDIDATE_MB_TYPE_INTRA; //FIXME cleanup

  1333.         }else

  1334.             s->current_picture.mb_type[mb_y*s->mb_stride + mb_x]= 0;

  1335.         

  1336.         if (vard <= 64 || vard < varc) { //FIXME

  1337.             s->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);

  1338.         }else{

  1339.             s->scene_change_score+= s->qscale;

  1340.         }

  1341.     }

  1342. 

  1343.     s->mb_type[mb_y*s->mb_stride + mb_x]= mb_type;

  1344. }

  1345. 

  1346. int ff_pre_estimate_p_frame_motion(MpegEncContext * s,

  1347.                                     int mb_x, int mb_y)

  1348. {

  1349.     MotionEstContext * const c= &s->me;

  1350.     int mx, my, dmin;

  1351.     int P[10][2];

  1352.     const int shift= 1+s->quarter_sample;

  1353.     const int xy= mb_x + mb_y*s->mb_stride;

  1354.     init_ref(s, s->new_picture.data, s->last_picture.data, NULL, 16*mb_x, 16*mb_y, 0);

  1355.     init_mc(s, 0, c->flags);

  1356.     

  1357.     assert(s->quarter_sample==0 || s->quarter_sample==1);

  1358. 

  1359.     s->me.pre_penalty_factor    = get_penalty_factor(s, s->avctx->me_pre_cmp);

  1360.     s->me.current_mv_penalty= s->me.mv_penalty[s->f_code] + MAX_MV;

  1361. 

  1362.     get_limits(s, 16*mb_x, 16*mb_y);

  1363.     s->me.skip=0;

  1364. 

  1365.     P_LEFT[0]       = s->p_mv_table[xy + 1][0];

  1366.     P_LEFT[1]       = s->p_mv_table[xy + 1][1];

  1367. 

  1368.     if(P_LEFT[0]       < (s->me.xmin<<shift)) P_LEFT[0]       = (s->me.xmin<<shift);

  1369. 

  1370.     /* special case for first line */

  1371.     if (s->first_slice_line) {

  1372.         c->pred_x= P_LEFT[0];

  1373.         c->pred_y= P_LEFT[1];

  1374.         P_TOP[0]= P_TOPRIGHT[0]= P_MEDIAN[0]=

  1375.         P_TOP[1]= P_TOPRIGHT[1]= P_MEDIAN[1]= 0; //FIXME 

  1376.     } else {

  1377.         P_TOP[0]      = s->p_mv_table[xy + s->mb_stride    ][0];

  1378.         P_TOP[1]      = s->p_mv_table[xy + s->mb_stride    ][1];

  1379.         P_TOPRIGHT[0] = s->p_mv_table[xy + s->mb_stride - 1][0];

  1380.         P_TOPRIGHT[1] = s->p_mv_table[xy + s->mb_stride - 1][1];

  1381.         if(P_TOP[1]      < (s->me.ymin<<shift)) P_TOP[1]     = (s->me.ymin<<shift);

  1382.         if(P_TOPRIGHT[0] > (s->me.xmax<<shift)) P_TOPRIGHT[0]= (s->me.xmax<<shift);

  1383.         if(P_TOPRIGHT[1] < (s->me.ymin<<shift)) P_TOPRIGHT[1]= (s->me.ymin<<shift);

  1384.     

  1385.         P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);

  1386.         P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);

  1387. 

  1388.         c->pred_x = P_MEDIAN[0];

  1389.         c->pred_y = P_MEDIAN[1];

  1390.     }

  1391. 

  1392.     dmin = epzs_motion_search(s, &mx, &my, P, 0, 0, s->p_mv_table, (1<<16)>>shift);       

  1393. 

  1394.     s->p_mv_table[xy][0] = mx<<shift;

  1395.     s->p_mv_table[xy][1] = my<<shift;

  1396.     

  1397.     return dmin;

  1398. }

  1399. 

  1400. static int ff_estimate_motion_b(MpegEncContext * s,

  1401.                        int mb_x, int mb_y, int16_t (*mv_table)[2], int ref_index, int f_code)

  1402. {

  1403.     int mx, my, dmin;

  1404.     int P[10][2];

  1405.     const int shift= 1+s->quarter_sample;

  1406.     const int mot_stride = s->mb_stride;

  1407.     const int mot_xy = mb_y*mot_stride + mb_x;

  1408.     uint8_t * const mv_penalty= s->me.mv_penalty[f_code] + MAX_MV;

  1409.     int mv_scale;

  1410.         

  1411.     s->me.penalty_factor    = get_penalty_factor(s, s->avctx->me_cmp);

  1412.     s->me.sub_penalty_factor= get_penalty_factor(s, s->avctx->me_sub_cmp);

  1413.     s->me.mb_penalty_factor = get_penalty_factor(s, s->avctx->mb_cmp);

  1414.     s->me.current_mv_penalty= mv_penalty;

  1415. 

  1416.     get_limits(s, 16*mb_x, 16*mb_y);

  1417. 

  1418.     switch(s->me_method) {

  1419.     case ME_ZERO:

  1420.     default:

  1421. 	no_motion_search(s, &mx, &my);

  1422.         dmin = 0;

  1423.         mx-= mb_x*16;

  1424.         my-= mb_y*16;

  1425.         break;

  1426. #if 0

  1427.     case ME_FULL:

  1428. 	dmin = full_motion_search(s, &mx, &my, range, ref_picture);

  1429.         mx-= mb_x*16;

  1430.         my-= mb_y*16;

  1431.         break;

  1432.     case ME_LOG:

  1433. 	dmin = log_motion_search(s, &mx, &my, range / 2, ref_picture);

  1434.         mx-= mb_x*16;

  1435.         my-= mb_y*16;

  1436.         break;

  1437.     case ME_PHODS:

  1438. 	dmin = phods_motion_search(s, &mx, &my, range / 2, ref_picture);

  1439.         mx-= mb_x*16;

  1440.         my-= mb_y*16;

  1441.         break;

  1442. #endif

  1443.     case ME_X1:

  1444.     case ME_EPZS:

  1445.        {

  1446.             P_LEFT[0]        = mv_table[mot_xy - 1][0];

  1447.             P_LEFT[1]        = mv_table[mot_xy - 1][1];

  1448. 

  1449.             if(P_LEFT[0]       > (s->me.xmax<<shift)) P_LEFT[0]       = (s->me.xmax<<shift);

  1450. 

  1451.             /* special case for first line */

  1452.             if (!s->first_slice_line) {

  1453.                 P_TOP[0] = mv_table[mot_xy - mot_stride             ][0];

  1454.                 P_TOP[1] = mv_table[mot_xy - mot_stride             ][1];

  1455.                 P_TOPRIGHT[0] = mv_table[mot_xy - mot_stride + 1         ][0];

  1456.                 P_TOPRIGHT[1] = mv_table[mot_xy - mot_stride + 1         ][1];

  1457.                 if(P_TOP[1] > (s->me.ymax<<shift)) P_TOP[1]= (s->me.ymax<<shift);

  1458.                 if(P_TOPRIGHT[0] < (s->me.xmin<<shift)) P_TOPRIGHT[0]= (s->me.xmin<<shift);

  1459.                 if(P_TOPRIGHT[1] > (s->me.ymax<<shift)) P_TOPRIGHT[1]= (s->me.ymax<<shift);

  1460.         

  1461.                 P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);

  1462.                 P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);

  1463.             }

  1464.             s->me.pred_x= P_LEFT[0];

  1465.             s->me.pred_y= P_LEFT[1];

  1466.         }

  1467.         

  1468.         if(mv_table == s->b_forw_mv_table){

  1469.             mv_scale= (s->pb_time<<16) / (s->pp_time<<shift);

  1470.         }else{

  1471.             mv_scale= ((s->pb_time - s->pp_time)<<16) / (s->pp_time<<shift);

  1472.         }

  1473.         

  1474.         dmin = epzs_motion_search(s, &mx, &my, P, 0, ref_index, s->p_mv_table, mv_scale);

  1475.  

  1476.         break;

  1477.     }

  1478.     

  1479.     dmin= s->me.sub_motion_search(s, &mx, &my, dmin, 0, ref_index, 0, 16);

  1480.                                    

  1481.     if(s->avctx->me_sub_cmp != s->avctx->mb_cmp && !s->me.skip)

  1482.         dmin= get_mb_score(s, mx, my, 0, ref_index);

  1483. 

  1484. //printf("%d %d %d %d//", s->mb_x, s->mb_y, mx, my);

  1485. //    s->mb_type[mb_y*s->mb_width + mb_x]= mb_type;

  1486.     mv_table[mot_xy][0]= mx;

  1487.     mv_table[mot_xy][1]= my;

  1488. 

  1489.     return dmin;

  1490. }

  1491. 

  1492. static inline int check_bidir_mv(MpegEncContext * s,

  1493.                    int motion_fx, int motion_fy,

  1494.                    int motion_bx, int motion_by,

  1495.                    int pred_fx, int pred_fy,

  1496.                    int pred_bx, int pred_by,

  1497.                    int size, int h)

  1498. {

  1499.     //FIXME optimize?

  1500.     //FIXME better f_code prediction (max mv & distance)

  1501.     //FIXME pointers

  1502.     MotionEstContext * const c= &s->me;

  1503.     uint8_t * const mv_penalty= s->me.mv_penalty[s->f_code] + MAX_MV; // f_code of the prev frame

  1504.     int stride= s->me.stride;

  1505.     int uvstride= s->me.uvstride;

  1506.     uint8_t *dest_y = s->me.scratchpad;

  1507.     uint8_t *ptr;

  1508.     int dxy;

  1509.     int src_x, src_y;

  1510.     int fbmin;

  1511.     uint8_t **src_data= c->src[0];

  1512.     uint8_t **ref_data= c->ref[0];

  1513.     uint8_t **ref2_data= c->ref[2];

  1514. 

  1515.     if(s->quarter_sample){

  1516.         dxy = ((motion_fy & 3) << 2) | (motion_fx & 3);

  1517.         src_x = motion_fx >> 2;

  1518.         src_y = motion_fy >> 2;

  1519. 

  1520.         ptr = ref_data[0] + (src_y * stride) + src_x;

  1521.         s->dsp.put_qpel_pixels_tab[0][dxy](dest_y    , ptr    , stride);

  1522. 

  1523.         dxy = ((motion_by & 3) << 2) | (motion_bx & 3);

  1524.         src_x = motion_bx >> 2;

  1525.         src_y = motion_by >> 2;

  1526.     

  1527.         ptr = ref2_data[0] + (src_y * stride) + src_x;

  1528.         s->dsp.avg_qpel_pixels_tab[size][dxy](dest_y    , ptr    , stride);

  1529.     }else{

  1530.         dxy = ((motion_fy & 1) << 1) | (motion_fx & 1);

  1531.         src_x = motion_fx >> 1;

  1532.         src_y = motion_fy >> 1;

  1533. 

  1534.         ptr = ref_data[0] + (src_y * stride) + src_x;

  1535.         s->dsp.put_pixels_tab[size][dxy](dest_y    , ptr    , stride, h);

  1536. 

  1537.         dxy = ((motion_by & 1) << 1) | (motion_bx & 1);

  1538.         src_x = motion_bx >> 1;

  1539.         src_y = motion_by >> 1;

  1540.     

  1541.         ptr = ref2_data[0] + (src_y * stride) + src_x;

  1542.         s->dsp.avg_pixels_tab[size][dxy](dest_y    , ptr    , stride, h);

  1543.     }

  1544. 

  1545.     fbmin = (mv_penalty[motion_fx-pred_fx] + mv_penalty[motion_fy-pred_fy])*s->me.mb_penalty_factor

  1546.            +(mv_penalty[motion_bx-pred_bx] + mv_penalty[motion_by-pred_by])*s->me.mb_penalty_factor

  1547.            + s->dsp.mb_cmp[size](s, src_data[0], dest_y, stride, h); //FIXME new_pic

  1548.            

  1549.     if(s->avctx->mb_cmp&FF_CMP_CHROMA){

  1550.     }

  1551.     //FIXME CHROMA !!!

  1552.            

  1553.     return fbmin;

  1554. }

  1555. 

  1556. /* refine the bidir vectors in hq mode and return the score in both lq & hq mode*/

  1557. static inline int bidir_refine(MpegEncContext * s, int mb_x, int mb_y)

  1558. {

  1559.     const int mot_stride = s->mb_stride;

  1560.     const int xy = mb_y *mot_stride + mb_x;

  1561.     int fbmin;

  1562.     int pred_fx= s->b_bidir_forw_mv_table[xy-1][0];

  1563.     int pred_fy= s->b_bidir_forw_mv_table[xy-1][1];

  1564.     int pred_bx= s->b_bidir_back_mv_table[xy-1][0];

  1565.     int pred_by= s->b_bidir_back_mv_table[xy-1][1];

  1566.     int motion_fx= s->b_bidir_forw_mv_table[xy][0]= s->b_forw_mv_table[xy][0];

  1567.     int motion_fy= s->b_bidir_forw_mv_table[xy][1]= s->b_forw_mv_table[xy][1];

  1568.     int motion_bx= s->b_bidir_back_mv_table[xy][0]= s->b_back_mv_table[xy][0];

  1569.     int motion_by= s->b_bidir_back_mv_table[xy][1]= s->b_back_mv_table[xy][1];

  1570. 

  1571.     //FIXME do refinement and add flag

  1572.     

  1573.     fbmin= check_bidir_mv(s, motion_fx, motion_fy,

  1574.                           motion_bx, motion_by,

  1575.                           pred_fx, pred_fy,

  1576.                           pred_bx, pred_by,

  1577.                           0, 16);

  1578. 

  1579.    return fbmin;

  1580. }

  1581. 

  1582. static inline int direct_search(MpegEncContext * s, int mb_x, int mb_y)

  1583. {

  1584.     int P[10][2];

  1585.     const int mot_stride = s->mb_stride;

  1586.     const int mot_xy = mb_y*mot_stride + mb_x;

  1587.     const int shift= 1+s->quarter_sample;

  1588.     int dmin, i;

  1589.     const int time_pp= s->pp_time;

  1590.     const int time_pb= s->pb_time;

  1591.     int mx, my, xmin, xmax, ymin, ymax;

  1592.     int16_t (*mv_table)[2]= s->b_direct_mv_table;

  1593.     

  1594.     s->me.current_mv_penalty= s->me.mv_penalty[1] + MAX_MV;

  1595.     ymin= xmin=(-32)>>shift;

  1596.     ymax= xmax=   31>>shift;

  1597. 

  1598.     if(IS_8X8(s->next_picture.mb_type[mot_xy])){

  1599.         s->mv_type= MV_TYPE_8X8;

  1600.     }else{

  1601.         s->mv_type= MV_TYPE_16X16;

  1602.     }

  1603. 

  1604.     for(i=0; i<4; i++){

  1605.         int index= s->block_index[i];

  1606.         int min, max;

  1607.     

  1608.         s->me.co_located_mv[i][0]= s->next_picture.motion_val[0][index][0];

  1609.         s->me.co_located_mv[i][1]= s->next_picture.motion_val[0][index][1];

  1610.         s->me.direct_basis_mv[i][0]= s->me.co_located_mv[i][0]*time_pb/time_pp + ((i& 1)<<(shift+3));

  1611.         s->me.direct_basis_mv[i][1]= s->me.co_located_mv[i][1]*time_pb/time_pp + ((i>>1)<<(shift+3));

  1612. //        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);

  1613. //        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);

  1614. 

  1615.         max= FFMAX(s->me.direct_basis_mv[i][0], s->me.direct_basis_mv[i][0] - s->me.co_located_mv[i][0])>>shift;

  1616.         min= FFMIN(s->me.direct_basis_mv[i][0], s->me.direct_basis_mv[i][0] - s->me.co_located_mv[i][0])>>shift;

  1617.         max+= 16*mb_x + 1; // +-1 is for the simpler rounding

  1618.         min+= 16*mb_x - 1;

  1619.         xmax= FFMIN(xmax, s->width - max);

  1620.         xmin= FFMAX(xmin, - 16     - min);

  1621. 

  1622.         max= FFMAX(s->me.direct_basis_mv[i][1], s->me.direct_basis_mv[i][1] - s->me.co_located_mv[i][1])>>shift;

  1623.         min= FFMIN(s->me.direct_basis_mv[i][1], s->me.direct_basis_mv[i][1] - s->me.co_located_mv[i][1])>>shift;

  1624.         max+= 16*mb_y + 1; // +-1 is for the simpler rounding

  1625.         min+= 16*mb_y - 1;

  1626.         ymax= FFMIN(ymax, s->height - max);

  1627.         ymin= FFMAX(ymin, - 16      - min);

  1628.         

  1629.         if(s->mv_type == MV_TYPE_16X16) break;

  1630.     }

  1631.     

  1632.     assert(xmax <= 15 && ymax <= 15 && xmin >= -16 && ymin >= -16);

  1633.     

  1634.     if(xmax < 0 || xmin >0 || ymax < 0 || ymin > 0){

  1635.         s->b_direct_mv_table[mot_xy][0]= 0;

  1636.         s->b_direct_mv_table[mot_xy][1]= 0;

  1637. 

  1638.         return 256*256*256*64;

  1639.     }

  1640.     

  1641.     s->me.xmin= xmin;

  1642.     s->me.ymin= ymin;

  1643.     s->me.xmax= xmax;

  1644.     s->me.ymax= ymax;

  1645.     s->me.flags     |= FLAG_DIRECT;

  1646.     s->me.sub_flags |= FLAG_DIRECT;

  1647.     s->me.pred_x=0;

  1648.     s->me.pred_y=0;

  1649. 

  1650.     P_LEFT[0]        = clip(mv_table[mot_xy - 1][0], xmin<<shift, xmax<<shift);

  1651.     P_LEFT[1]        = clip(mv_table[mot_xy - 1][1], ymin<<shift, ymax<<shift);

  1652. 

  1653.     /* special case for first line */

  1654.     if (!s->first_slice_line) { //FIXME maybe allow this over thread boundary as its cliped

  1655.         P_TOP[0]      = clip(mv_table[mot_xy - mot_stride             ][0], xmin<<shift, xmax<<shift);

  1656.         P_TOP[1]      = clip(mv_table[mot_xy - mot_stride             ][1], ymin<<shift, ymax<<shift);

  1657.         P_TOPRIGHT[0] = clip(mv_table[mot_xy - mot_stride + 1         ][0], xmin<<shift, xmax<<shift);

  1658.         P_TOPRIGHT[1] = clip(mv_table[mot_xy - mot_stride + 1         ][1], ymin<<shift, ymax<<shift);

  1659.     

  1660.         P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);

  1661.         P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);

  1662.     }

  1663.  

  1664.     dmin = epzs_motion_search(s, &mx, &my, P, 0, 0, mv_table, 1<<(16-shift));

  1665.     if(s->me.sub_flags&FLAG_QPEL) 

  1666.         dmin = qpel_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);

  1667.     else

  1668.         dmin = hpel_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);

  1669.     

  1670.     if(s->avctx->me_sub_cmp != s->avctx->mb_cmp && !s->me.skip)

  1671.         dmin= get_mb_score(s, mx, my, 0, 0);

  1672.     

  1673.     get_limits(s, 16*mb_x, 16*mb_y); //restore s->me.?min/max, maybe not needed

  1674. 

  1675.     s->b_direct_mv_table[mot_xy][0]= mx;

  1676.     s->b_direct_mv_table[mot_xy][1]= my;

  1677.     s->me.flags     &= ~FLAG_DIRECT;

  1678.     s->me.sub_flags &= ~FLAG_DIRECT;

  1679. 

  1680.     return dmin;

  1681. }

  1682. 

  1683. void ff_estimate_b_frame_motion(MpegEncContext * s,

  1684.                              int mb_x, int mb_y)

  1685. {

  1686.     const int penalty_factor= s->me.mb_penalty_factor;

  1687.     int fmin, bmin, dmin, fbmin, bimin, fimin;

  1688.     int type=0;

  1689.     init_ref(s, s->new_picture.data, s->last_picture.data, s->next_picture.data, 16*mb_x, 16*mb_y, 2);

  1690.     init_mc(s, 0, s->me.flags);

  1691.     

  1692.     s->me.skip=0;

  1693.     if(s->avctx->me_threshold){

  1694.         int vard= (check_input_motion(s, mb_x, mb_y, 0)+128)>>8;

  1695.         

  1696.         if(vard<s->avctx->me_threshold){

  1697. //            pix = c->src[0][0];

  1698. //            sum = s->dsp.pix_sum(pix, s->linesize);

  1699. //            varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500 + 128)>>8;

  1700.         

  1701. //            pic->mb_var   [s->mb_stride * mb_y + mb_x] = varc;

  1702.              s->current_picture.mc_mb_var[s->mb_stride * mb_y + mb_x] = vard;

  1703. /*            pic->mb_mean  [s->mb_stride * mb_y + mb_x] = (sum+128)>>8;

  1704.             s->mb_var_sum_temp    += varc;*/

  1705.             s->mc_mb_var_sum_temp += vard;

  1706. /*            if (vard <= 64 || vard < varc) {

  1707.                 s->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);

  1708.             }else{

  1709.                 s->scene_change_score+= s->qscale;

  1710.             }*/

  1711.             return;

  1712.         }

  1713.     }

  1714. 

  1715.     if (s->codec_id == CODEC_ID_MPEG4)

  1716.         dmin= direct_search(s, mb_x, mb_y);

  1717.     else

  1718.         dmin= INT_MAX;

  1719. //FIXME penalty stuff for non mpeg4

  1720.     s->me.skip=0;

  1721.     fmin= ff_estimate_motion_b(s, mb_x, mb_y, s->b_forw_mv_table, 0, s->f_code) + 3*penalty_factor;

  1722.     

  1723.     s->me.skip=0;

  1724.     bmin= ff_estimate_motion_b(s, mb_x, mb_y, s->b_back_mv_table, 2, s->b_code) + 2*penalty_factor;

  1725. //printf(" %d %d ", s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1]);

  1726. 

  1727.     s->me.skip=0;

  1728.     fbmin= bidir_refine(s, mb_x, mb_y) + penalty_factor;

  1729. //printf("%d %d %d %d\n", dmin, fmin, bmin, fbmin);

  1730.     

  1731.     if(s->flags & CODEC_FLAG_INTERLACED_ME){

  1732.         const int xy = mb_y*s->mb_stride + mb_x;

  1733. 

  1734. //FIXME mb type penalty

  1735.         s->me.skip=0;

  1736.         s->me.current_mv_penalty= s->me.mv_penalty[s->f_code] + MAX_MV;

  1737.         fimin= interlaced_search(s, 0,

  1738.                                  s->b_field_mv_table[0], s->b_field_select_table[0],

  1739.                                  s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1]);

  1740.         s->me.current_mv_penalty= s->me.mv_penalty[s->b_code] + MAX_MV;

  1741.         bimin= interlaced_search(s, 2,

  1742.                                  s->b_field_mv_table[1], s->b_field_select_table[1],

  1743.                                  s->b_back_mv_table[xy][0], s->b_back_mv_table[xy][1]);

  1744.     }else

  1745.         fimin= bimin= INT_MAX;

  1746. 

  1747.     {

  1748.         int score= fmin;

  1749.         type = CANDIDATE_MB_TYPE_FORWARD;

  1750.         

  1751.         if (dmin <= score){

  1752.             score = dmin;

  1753.             type = CANDIDATE_MB_TYPE_DIRECT;

  1754.         }

  1755.         if(bmin<score){

  1756.             score=bmin;

  1757.             type= CANDIDATE_MB_TYPE_BACKWARD; 

  1758.         }

  1759.         if(fbmin<score){

  1760.             score=fbmin;

  1761.             type= CANDIDATE_MB_TYPE_BIDIR;

  1762.         }

  1763.         if(fimin<score){

  1764.             score=fimin;

  1765.             type= CANDIDATE_MB_TYPE_FORWARD_I;

  1766.         }

  1767.         if(bimin<score){

  1768.             score=bimin;

  1769.             type= CANDIDATE_MB_TYPE_BACKWARD_I;

  1770.         }

  1771.         

  1772.         score= ((unsigned)(score*score + 128*256))>>16;

  1773.         s->mc_mb_var_sum_temp += score;

  1774.         s->current_picture.mc_mb_var[mb_y*s->mb_stride + mb_x] = score; //FIXME use SSE

  1775.     }

  1776. 

  1777.     if(s->avctx->mb_decision > FF_MB_DECISION_SIMPLE){

  1778.         type= CANDIDATE_MB_TYPE_FORWARD | CANDIDATE_MB_TYPE_BACKWARD | CANDIDATE_MB_TYPE_BIDIR | CANDIDATE_MB_TYPE_DIRECT;

  1779.         if(fimin < INT_MAX)

  1780.             type |= CANDIDATE_MB_TYPE_FORWARD_I;

  1781.         if(bimin < INT_MAX)

  1782.             type |= CANDIDATE_MB_TYPE_BACKWARD_I;

  1783.         if(fimin < INT_MAX && bimin < INT_MAX){

  1784.             type |= CANDIDATE_MB_TYPE_BIDIR_I;

  1785.         }

  1786.          //FIXME something smarter

  1787.         if(dmin>256*256*16) type&= ~CANDIDATE_MB_TYPE_DIRECT; //dont try direct mode if its invalid for this MB

  1788. #if 0        

  1789.         if(s->out_format == FMT_MPEG1)

  1790.             type |= CANDIDATE_MB_TYPE_INTRA;

  1791. #endif

  1792.     }

  1793. 

  1794.     s->mb_type[mb_y*s->mb_stride + mb_x]= type;

  1795. }

  1796. 

  1797. /* find best f_code for ME which do unlimited searches */

  1798. int ff_get_best_fcode(MpegEncContext * s, int16_t (*mv_table)[2], int type)

  1799. {

  1800.     if(s->me_method>=ME_EPZS){

  1801.         int score[8];

  1802.         int i, y;

  1803.         uint8_t * fcode_tab= s->fcode_tab;

  1804.         int best_fcode=-1;

  1805.         int best_score=-10000000;

  1806. 

  1807.         for(i=0; i<8; i++) score[i]= s->mb_num*(8-i);

  1808. 

  1809.         for(y=0; y<s->mb_height; y++){

  1810.             int x;

  1811.             int xy= y*s->mb_stride;

  1812.             for(x=0; x<s->mb_width; x++){

  1813.                 if(s->mb_type[xy] & type){

  1814.                     int fcode= FFMAX(fcode_tab[mv_table[xy][0] + MAX_MV],

  1815.                                      fcode_tab[mv_table[xy][1] + MAX_MV]);

  1816.                     int j;

  1817.                     

  1818.                     for(j=0; j<fcode && j<8; j++){

  1819.                         if(s->pict_type==B_TYPE || s->current_picture.mc_mb_var[xy] < s->current_picture.mb_var[xy])

  1820.                             score[j]-= 170;

  1821.                     }

  1822.                 }

  1823.                 xy++;

  1824.             }

  1825.         }

  1826.         

  1827.         for(i=1; i<8; i++){

  1828.             if(score[i] > best_score){

  1829.                 best_score= score[i];

  1830.                 best_fcode= i;

  1831.             }

  1832. //            printf("%d %d\n", i, score[i]);

  1833.         }

  1834. 

  1835. //    printf("fcode: %d type: %d\n", i, s->pict_type);

  1836.         return best_fcode;

  1837. /*        for(i=0; i<=MAX_FCODE; i++){

  1838.             printf("%d ", mv_num[i]);

  1839.         }

  1840.         printf("\n");*/

  1841.     }else{

  1842.         return 1;

  1843.     }

  1844. }

  1845. 

  1846. void ff_fix_long_p_mvs(MpegEncContext * s)

  1847. {

  1848.     const int f_code= s->f_code;

  1849.     int y, range;

  1850.     assert(s->pict_type==P_TYPE);

  1851. 

  1852.     range = (((s->out_format == FMT_MPEG1) ? 8 : 16) << f_code);

  1853.     

  1854.     if(s->msmpeg4_version) range= 16;

  1855.     

  1856.     if(s->avctx->me_range && range > s->avctx->me_range) range= s->avctx->me_range;

  1857.     

  1858. //printf("%d no:%d %d//\n", clip, noclip, f_code);

  1859.     if(s->flags&CODEC_FLAG_4MV){

  1860.         const int wrap= s->b8_stride;

  1861. 

  1862.         /* clip / convert to intra 8x8 type MVs */

  1863.         for(y=0; y<s->mb_height; y++){

  1864.             int xy= y*2*wrap;

  1865.             int i= y*s->mb_stride;

  1866.             int x;

  1867. 

  1868.             for(x=0; x<s->mb_width; x++){

  1869.                 if(s->mb_type[i]&CANDIDATE_MB_TYPE_INTER4V){

  1870.                     int block;

  1871.                     for(block=0; block<4; block++){

  1872.                         int off= (block& 1) + (block>>1)*wrap;

  1873.                         int mx= s->current_picture.motion_val[0][ xy + off ][0];

  1874.                         int my= s->current_picture.motion_val[0][ xy + off ][1];

  1875. 

  1876.                         if(   mx >=range || mx <-range

  1877.                            || my >=range || my <-range){

  1878.                             s->mb_type[i] &= ~CANDIDATE_MB_TYPE_INTER4V;

  1879.                             s->mb_type[i] |= CANDIDATE_MB_TYPE_INTRA;

  1880.                             s->current_picture.mb_type[i]= CANDIDATE_MB_TYPE_INTRA;

  1881.                         }

  1882.                     }

  1883.                 }

  1884.                 xy+=2;

  1885.                 i++;

  1886.             }

  1887.         }

  1888.     }

  1889. }

  1890. 

  1891. /**

  1892.  *

  1893.  * @param truncate 1 for truncation, 0 for using intra

  1894.  */

  1895. void ff_fix_long_mvs(MpegEncContext * s, uint8_t *field_select_table, int field_select, 

  1896.                      int16_t (*mv_table)[2], int f_code, int type, int truncate)

  1897. {

  1898.     int y, h_range, v_range;

  1899. 

  1900.     // RAL: 8 in MPEG-1, 16 in MPEG-4

  1901.     int range = (((s->out_format == FMT_MPEG1) ? 8 : 16) << f_code);

  1902. 

  1903.     if(s->msmpeg4_version) range= 16;

  1904.     if(s->avctx->me_range && range > s->avctx->me_range) range= s->avctx->me_range;

  1905. 

  1906.     h_range= range;

  1907.     v_range= field_select_table ? range>>1 : range;

  1908. 

  1909.     /* clip / convert to intra 16x16 type MVs */

  1910.     for(y=0; y<s->mb_height; y++){

  1911.         int x;

  1912.         int xy= y*s->mb_stride;

  1913.         for(x=0; x<s->mb_width; x++){

  1914.             if (s->mb_type[xy] & type){    // RAL: "type" test added...

  1915.                 if(field_select_table==NULL || field_select_table[xy] == field_select){

  1916.                     if(   mv_table[xy][0] >=h_range || mv_table[xy][0] <-h_range

  1917.                        || mv_table[xy][1] >=v_range || mv_table[xy][1] <-v_range){

  1918. 

  1919.                         if(truncate){

  1920.                             if     (mv_table[xy][0] > h_range-1) mv_table[xy][0]=  h_range-1;

  1921.                             else if(mv_table[xy][0] < -h_range ) mv_table[xy][0]= -h_range;

  1922.                             if     (mv_table[xy][1] > v_range-1) mv_table[xy][1]=  v_range-1;

  1923.                             else if(mv_table[xy][1] < -v_range ) mv_table[xy][1]= -v_range;

  1924.                         }else{

  1925.                             s->mb_type[xy] &= ~type;

  1926.                             s->mb_type[xy] |= CANDIDATE_MB_TYPE_INTRA;

  1927.                             mv_table[xy][0]=

  1928.                             mv_table[xy][1]= 0;

  1929.                         }

  1930.                     }

  1931.                 }

  1932.             }

  1933.             xy++;

  1934.         }

  1935.     }

  1936. }