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.
1592 Commits
32 Branches
303MB
Tree: 70ac76c0eb
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '70ac76c0eb'
${ noResults }
FFmpeg/libavcodec/motion_est.c

1544 lines
50KB
Raw Blame History 

  1. /*

  2.  * Motion estimation 

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

  4.  * Copyright (c) 2002 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. #include <stdlib.h>

  24. #include <stdio.h>

  25. #include "avcodec.h"

  26. #include "dsputil.h"

  27. #include "mpegvideo.h"

  28. 

  29. //#undef NDEBUG

  30. //#include <assert.h>

  31. 

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

  33. 

  34. #define P_LEFT P[1]

  35. #define P_TOP P[2]

  36. #define P_TOPRIGHT P[3]

  37. #define P_MEDIAN P[4]

  38. #define P_MV1 P[9]

  39. 

  40. static inline int sad_hpel_motion_search(MpegEncContext * s,

  41. 				  int *mx_ptr, int *my_ptr, int dmin,

  42. 				  int xmin, int ymin, int xmax, int ymax,

  43.                                   int pred_x, int pred_y, Picture *picture,

  44.                                   int n, int size, uint16_t * const mv_penalty);

  45. 

  46. static inline int update_map_generation(MpegEncContext * s)

  47. {

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

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

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

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

  52.     }

  53.     return s->me.map_generation;

  54. }

  55. 

  56. /* shape adaptive search stuff */

  57. typedef struct Minima{

  58.     int height;

  59.     int x, y;

  60.     int checked;

  61. }Minima;

  62. 

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

  64.     Minima *da = (Minima *) a;

  65.     Minima *db = (Minima *) b;

  66.     

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

  68. }

  69.                                   

  70. /* SIMPLE */

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

  72. 

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

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

  75. 

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

  77. {\

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

  79.     hpel_put[0][dxy](s->me.scratchpad, (ref_y) + (x) + (y)*(stride), stride, (16>>size));\

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

  81. }

  82. 

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

  84. {\

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

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

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

  88. }

  89. 

  90. #include "motion_est_template.c"

  91. #undef RENAME

  92. #undef CMP

  93. #undef CMP_HPEL

  94. #undef CMP_QPEL

  95. #undef INIT

  96. 

  97. /* SIMPLE CHROMA */

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

  99. 

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

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

  102. if(chroma_cmp){\

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

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

  105. \

  106.     chroma_hpel_put[0][dxy](s->me.scratchpad, ref_u + c, uvstride, 8);\

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

  108.     chroma_hpel_put[0][dxy](s->me.scratchpad, ref_v + c, uvstride, 8);\

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

  110. }

  111. 

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

  113. {\

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

  115.     hpel_put[0][dxy](s->me.scratchpad, (ref_y) + (x) + (y)*(stride), stride, (16>>size));\

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

  117.     if(chroma_cmp_sub){\

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

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

  120.         chroma_hpel_put[0][cxy](s->me.scratchpad, ref_u + c, uvstride, 8);\

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

  122.         chroma_hpel_put[0][cxy](s->me.scratchpad, ref_v + c, uvstride, 8);\

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

  124.     }\

  125. }

  126. 

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

  128. {\

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

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

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

  132.     if(chroma_cmp_sub){\

  133.         int cxy, c;\

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

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

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

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

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

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

  140.         chroma_hpel_put[0][cxy](s->me.scratchpad, ref_u + c, uvstride, 8);\

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

  142.         chroma_hpel_put[0][cxy](s->me.scratchpad, ref_v + c, uvstride, 8);\

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

  144.     }\

  145. }

  146. 

  147. #include "motion_est_template.c"

  148. #undef RENAME

  149. #undef CMP

  150. #undef CMP_HPEL

  151. #undef CMP_QPEL

  152. #undef INIT

  153. 

  154. /* SIMPLE DIRECT HPEL */

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

  156. //FIXME precalc divisions stuff

  157. 

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

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

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

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

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

  163.         int i;\

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

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

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

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

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

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

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

  171. \

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

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

  174.             hpel_avg[1][bxy](dst, (ref2_y) + (bx>>1) + (by>>1)*(stride), stride, 8);\

  175.         }\

  176.     }else{\

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

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

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

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

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

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

  183. \

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

  185.         hpel_avg[0][bxy](s->me.scratchpad, (ref2_y) + (bx>>1) + (by>>1)*(stride), stride, 16);\

  186.     }\

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

  188. }else\

  189.     d= 256*256*256*32;

  190. 

  191. 

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

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

  194. 

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

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

  197.     

  198. #include "motion_est_template.c"

  199. #undef RENAME

  200. #undef CMP

  201. #undef CMP_HPEL

  202. #undef CMP_QPEL

  203. #undef INIT

  204. #undef CMP_DIRECT

  205. 

  206. /* SIMPLE DIRECT QPEL */

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

  208. 

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

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

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

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

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

  214.         int i;\

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

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

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

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

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

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

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

  222. \

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

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

  225.             qpel_avg[1][bxy](dst, (ref2_y) + (bx>>2) + (by>>2)*(stride), stride);\

  226.         }\

  227.     }else{\

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

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

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

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

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

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

  234. \

  235.         qpel_put[0][fxy](s->me.scratchpad, (ref_y ) + (fx>>2) + (fy>>2)*(stride), stride);\

  236.         qpel_avg[0][bxy](s->me.scratchpad, (ref2_y) + (bx>>2) + (by>>2)*(stride), stride);\

  237.     }\

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

  239. }else\

  240.     d= 256*256*256*32;

  241. 

  242. 

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

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

  245. 

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

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

  248. 

  249. #include "motion_est_template.c"

  250. #undef RENAME

  251. #undef CMP

  252. #undef CMP_HPEL

  253. #undef CMP_QPEL

  254. #undef INIT

  255. #undef CMP__DIRECT

  256. 

  257. 

  258. static int zero_cmp(void *s, uint8_t *a, uint8_t *b, int stride){

  259.     return 0;

  260. }

  261. 

  262. static void set_cmp(MpegEncContext *s, me_cmp_func *cmp, int type){

  263.     DSPContext* c= &s->dsp;

  264.     int i;

  265.     

  266.     memset(cmp, 0, sizeof(void*)*11);

  267. 

  268.     switch(type&0xFF){

  269.     case FF_CMP_SAD:

  270.         cmp[0]= c->sad[0];

  271.         cmp[1]= c->sad[1];

  272.         break;

  273.     case FF_CMP_SATD:

  274.         cmp[0]= c->hadamard8_diff[0];

  275.         cmp[1]= c->hadamard8_diff[1];

  276.         break;

  277.     case FF_CMP_SSE:

  278.         cmp[0]= c->sse[0];

  279.         cmp[1]= c->sse[1];

  280.         break;

  281.     case FF_CMP_DCT:

  282.         cmp[0]= c->dct_sad[0];

  283.         cmp[1]= c->dct_sad[1];

  284.         break;

  285.     case FF_CMP_PSNR:

  286.         cmp[0]= c->quant_psnr[0];

  287.         cmp[1]= c->quant_psnr[1];

  288.         break;

  289.     case FF_CMP_ZERO:

  290.         for(i=0; i<7; i++){

  291.             cmp[i]= zero_cmp;

  292.         }

  293.         break;

  294.     default:

  295.         fprintf(stderr,"internal error in cmp function selection\n");

  296.     }

  297. };

  298. 

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

  300. 

  301.     switch(type){

  302.     default:

  303.     case FF_CMP_SAD:

  304.         return s->qscale;

  305.     case FF_CMP_SSE:

  306. //        return s->qscale*8;

  307.     case FF_CMP_DCT:

  308.     case FF_CMP_SATD:

  309.         return s->qscale*8;

  310.     }

  311. }

  312. 

  313. void ff_init_me(MpegEncContext *s){

  314.     set_cmp(s, s->dsp.me_cmp, s->avctx->me_cmp);

  315.     set_cmp(s, s->dsp.me_sub_cmp, s->avctx->me_sub_cmp);

  316.     set_cmp(s, s->dsp.mb_cmp, s->avctx->mb_cmp);

  317. 

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

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

  320.             s->me.sub_motion_search= simple_chroma_qpel_motion_search;

  321.         else

  322.             s->me.sub_motion_search= simple_qpel_motion_search;

  323.     }else{

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

  325.             s->me.sub_motion_search= simple_chroma_hpel_motion_search;

  326.         else if(s->avctx->me_sub_cmp == FF_CMP_SAD && s->avctx->me_cmp == FF_CMP_SAD)

  327.             s->me.sub_motion_search= sad_hpel_motion_search;

  328.         else

  329.             s->me.sub_motion_search= simple_hpel_motion_search;

  330.     }

  331. 

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

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

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

  335.     }else{

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

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

  338.     }

  339. }

  340.       

  341. static int pix_dev(UINT8 * pix, int line_size, int mean)

  342. {

  343.     int s, i, j;

  344. 

  345.     s = 0;

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

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

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

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

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

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

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

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

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

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

  356. 	    pix += 8;

  357. 	}

  358. 	pix += line_size - 16;

  359.     }

  360.     return s;

  361. }

  362. 

  363. static inline void no_motion_search(MpegEncContext * s,

  364. 				    int *mx_ptr, int *my_ptr)

  365. {

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

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

  368. }

  369. 

  370. static int full_motion_search(MpegEncContext * s,

  371.                               int *mx_ptr, int *my_ptr, int range,

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

  373. {

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

  375.     int mx, my, dmin, d;

  376.     UINT8 *pix;

  377. 

  378.     xx = 16 * s->mb_x;

  379.     yy = 16 * s->mb_y;

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

  381.     if (x1 < xmin)

  382. 	x1 = xmin;

  383.     x2 = xx + range - 1;

  384.     if (x2 > xmax)

  385. 	x2 = xmax;

  386.     y1 = yy - range + 1;

  387.     if (y1 < ymin)

  388. 	y1 = ymin;

  389.     y2 = yy + range - 1;

  390.     if (y2 > ymax)

  391. 	y2 = ymax;

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

  393.     dmin = 0x7fffffff;

  394.     mx = 0;

  395.     my = 0;

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

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

  398. 	    d = s->dsp.pix_abs16x16(pix, ref_picture + (y * s->linesize) + x,

  399. 			     s->linesize);

  400. 	    if (d < dmin ||

  401. 		(d == dmin &&

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

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

  404. 		dmin = d;

  405. 		mx = x;

  406. 		my = y;

  407. 	    }

  408. 	}

  409.     }

  410. 

  411.     *mx_ptr = mx;

  412.     *my_ptr = my;

  413. 

  414. #if 0

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

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

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

  418.     }

  419. #endif

  420.     return dmin;

  421. }

  422. 

  423. 

  424. static int log_motion_search(MpegEncContext * s,

  425.                              int *mx_ptr, int *my_ptr, int range,

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

  427. {

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

  429.     int mx, my, dmin, d;

  430.     UINT8 *pix;

  431. 

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

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

  434. 

  435.     /* Left limit */

  436.     x1 = xx - range;

  437.     if (x1 < xmin)

  438. 	x1 = xmin;

  439. 

  440.     /* Right limit */

  441.     x2 = xx + range;

  442.     if (x2 > xmax)

  443. 	x2 = xmax;

  444. 

  445.     /* Upper limit */

  446.     y1 = yy - range;

  447.     if (y1 < ymin)

  448. 	y1 = ymin;

  449. 

  450.     /* Lower limit */

  451.     y2 = yy + range;

  452.     if (y2 > ymax)

  453. 	y2 = ymax;

  454. 

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

  456.     dmin = 0x7fffffff;

  457.     mx = 0;

  458.     my = 0;

  459. 

  460.     do {

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

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

  463. 		d = s->dsp.pix_abs16x16(pix, ref_picture + (y * s->linesize) + x, s->linesize);

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

  465. 		    dmin = d;

  466. 		    mx = x;

  467. 		    my = y;

  468. 		}

  469. 	    }

  470. 	}

  471. 

  472. 	range = range >> 1;

  473. 

  474. 	x1 = mx - range;

  475. 	if (x1 < xmin)

  476. 	    x1 = xmin;

  477. 

  478. 	x2 = mx + range;

  479. 	if (x2 > xmax)

  480. 	    x2 = xmax;

  481. 

  482. 	y1 = my - range;

  483. 	if (y1 < ymin)

  484. 	    y1 = ymin;

  485. 

  486. 	y2 = my + range;

  487. 	if (y2 > ymax)

  488. 	    y2 = ymax;

  489. 

  490.     } while (range >= 1);

  491. 

  492. #ifdef DEBUG

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

  494. #endif

  495.     *mx_ptr = mx;

  496.     *my_ptr = my;

  497.     return dmin;

  498. }

  499. 

  500. static int phods_motion_search(MpegEncContext * s,

  501.                                int *mx_ptr, int *my_ptr, int range,

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

  503. {

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

  505.     int mx, my, dminx, dminy;

  506.     UINT8 *pix;

  507. 

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

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

  510. 

  511.     /* Left limit */

  512.     x1 = xx - range;

  513.     if (x1 < xmin)

  514. 	x1 = xmin;

  515. 

  516.     /* Right limit */

  517.     x2 = xx + range;

  518.     if (x2 > xmax)

  519. 	x2 = xmax;

  520. 

  521.     /* Upper limit */

  522.     y1 = yy - range;

  523.     if (y1 < ymin)

  524. 	y1 = ymin;

  525. 

  526.     /* Lower limit */

  527.     y2 = yy + range;

  528.     if (y2 > ymax)

  529. 	y2 = ymax;

  530. 

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

  532.     mx = 0;

  533.     my = 0;

  534. 

  535.     x = xx;

  536.     y = yy;

  537.     do {

  538.         dminx = 0x7fffffff;

  539.         dminy = 0x7fffffff;

  540. 

  541. 	lastx = x;

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

  543. 	    d = s->dsp.pix_abs16x16(pix, ref_picture + (y * s->linesize) + x, s->linesize);

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

  545. 		dminx = d;

  546. 		mx = x;

  547. 	    }

  548. 	}

  549. 

  550. 	x = lastx;

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

  552. 	    d = s->dsp.pix_abs16x16(pix, ref_picture + (y * s->linesize) + x, s->linesize);

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

  554. 		dminy = d;

  555. 		my = y;

  556. 	    }

  557. 	}

  558. 

  559. 	range = range >> 1;

  560. 

  561. 	x = mx;

  562. 	y = my;

  563. 	x1 = mx - range;

  564. 	if (x1 < xmin)

  565. 	    x1 = xmin;

  566. 

  567. 	x2 = mx + range;

  568. 	if (x2 > xmax)

  569. 	    x2 = xmax;

  570. 

  571. 	y1 = my - range;

  572. 	if (y1 < ymin)

  573. 	    y1 = ymin;

  574. 

  575. 	y2 = my + range;

  576. 	if (y2 > ymax)

  577. 	    y2 = ymax;

  578. 

  579.     } while (range >= 1);

  580. 

  581. #ifdef DEBUG

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

  583. #endif

  584. 

  585.     /* half pixel search */

  586.     *mx_ptr = mx;

  587.     *my_ptr = my;

  588.     return dminy;

  589. }

  590. 

  591. 

  592. #define Z_THRESHOLD 256

  593. 

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

  595. {\

  596.     d= pix_abs_ ## suffix(pix, ptr+((x)>>1), s->linesize);\

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

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

  599. }

  600. 

  601. static inline int sad_hpel_motion_search(MpegEncContext * s,

  602. 				  int *mx_ptr, int *my_ptr, int dmin,

  603. 				  int xmin, int ymin, int xmax, int ymax,

  604.                                   int pred_x, int pred_y, Picture *picture,

  605.                                   int n, int size, uint16_t * const mv_penalty)

  606. {

  607.     uint8_t *ref_picture= picture->data[0];

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

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

  610.     int mx, my, xx, yy, dminh;

  611.     UINT8 *pix, *ptr;

  612.     op_pixels_abs_func pix_abs_x2;

  613.     op_pixels_abs_func pix_abs_y2;

  614.     op_pixels_abs_func pix_abs_xy2;

  615.     

  616.     if(size==0){

  617.         pix_abs_x2 = s->dsp.pix_abs16x16_x2;

  618.         pix_abs_y2 = s->dsp.pix_abs16x16_y2;

  619.         pix_abs_xy2= s->dsp.pix_abs16x16_xy2;

  620.     }else{

  621.         pix_abs_x2 = s->dsp.pix_abs8x8_x2;

  622.         pix_abs_y2 = s->dsp.pix_abs8x8_y2;

  623.         pix_abs_xy2= s->dsp.pix_abs8x8_xy2;

  624.     }

  625. 

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

  627. //    printf("S");

  628.         *mx_ptr = 0;

  629.         *my_ptr = 0;

  630.         return dmin;

  631.     }

  632. //    printf("N");

  633.         

  634.     xx = 16 * s->mb_x + 8*(n&1);

  635.     yy = 16 * s->mb_y + 8*(n>>1);

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

  637. 

  638.     mx = *mx_ptr;

  639.     my = *my_ptr;

  640.     ptr = ref_picture + ((yy + my) * s->linesize) + (xx + mx);

  641.     

  642.     dminh = dmin;

  643. 

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

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

  646.         int dx=0, dy=0;

  647.         int d, pen_x, pen_y; 

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

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

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

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

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

  653.         mx<<=1;

  654.         my<<=1;

  655. 

  656.         

  657.         pen_x= pred_x + mx;

  658.         pen_y= pred_y + my;

  659. 

  660.         ptr-= s->linesize;

  661.         if(t<=b){

  662.             CHECK_SAD_HALF_MV(y2 , 0, -1)

  663.             if(l<=r){

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

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

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

  667.                     ptr+= s->linesize;

  668.                 }else{

  669.                     ptr+= s->linesize;

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

  671.                 }

  672.                 CHECK_SAD_HALF_MV(x2 , -1,  0)

  673.             }else{

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

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

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

  677.                     ptr+= s->linesize;

  678.                 }else{

  679.                     ptr+= s->linesize;

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

  681.                 }

  682.                 CHECK_SAD_HALF_MV(x2 , +1,  0)

  683.             }

  684.         }else{

  685.             if(l<=r){

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

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

  688.                     ptr+= s->linesize;

  689.                 }else{

  690.                     ptr+= s->linesize;

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

  692.                 }

  693.                 CHECK_SAD_HALF_MV(x2 , -1,  0)

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

  695.             }else{

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

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

  698.                     ptr+= s->linesize;

  699.                 }else{

  700.                     ptr+= s->linesize;

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

  702.                 }

  703.                 CHECK_SAD_HALF_MV(x2 , +1,  0)

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

  705.             }

  706.             CHECK_SAD_HALF_MV(y2 ,  0, +1)

  707.         }

  708.         mx+=dx;

  709.         my+=dy;

  710. 

  711.     }else{

  712.         mx<<=1;

  713.         my<<=1;

  714.     }

  715. 

  716.     *mx_ptr = mx;

  717.     *my_ptr = my;

  718.     return dminh;

  719. }

  720. 

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

  722. {

  723.     const int xy= s->mb_x + 1 + (s->mb_y + 1)*(s->mb_width + 2);

  724.     

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

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

  727. 

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

  729.     if(mv4){

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

  731. 

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

  733.         s->motion_val[mot_xy  ][1]= my;

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

  735.         s->motion_val[mot_xy+1][1]= my;

  736. 

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

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

  739.         s->motion_val[mot_xy  ][1]= my;

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

  741.         s->motion_val[mot_xy+1][1]= my;

  742.     }

  743. }

  744. 

  745. static inline void get_limits(MpegEncContext *s, int *range, int *xmin, int *ymin, int *xmax, int *ymax, int f_code)

  746. {

  747.     *range = 8 * (1 << (f_code - 1));

  748.     /* XXX: temporary kludge to avoid overflow for msmpeg4 */

  749.     if (s->out_format == FMT_H263 && !s->h263_msmpeg4)

  750. 	*range *= 2;

  751. 

  752.     if (s->unrestricted_mv) {

  753.         *xmin = -16;

  754.         *ymin = -16;

  755.         if (s->h263_plus)

  756.             *range *= 2;

  757.         if(s->avctx->codec->id!=CODEC_ID_MPEG4){

  758.             *xmax = s->mb_width*16;

  759.             *ymax = s->mb_height*16;

  760.         }else {

  761.             *xmax = s->width;

  762.             *ymax = s->height;

  763.         }

  764.     } else {

  765.         *xmin = 0;

  766.         *ymin = 0;

  767.         *xmax = s->mb_width*16 - 16;

  768.         *ymax = s->mb_height*16 - 16;

  769.     }

  770. }

  771. 

  772. static inline int mv4_search(MpegEncContext *s, int xmin, int ymin, int xmax, int ymax, int mx, int my, int shift)

  773. {

  774.     int block;

  775.     int P[10][2];

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

  777.     int dmin_sum=0;

  778.     uint16_t * const mv_penalty= s->me.mv_penalty[s->f_code] + MAX_MV;

  779. 

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

  781.         int mx4, my4;

  782.         int pred_x4, pred_y4;

  783.         int dmin4;

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

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

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

  787. //        const int block_x= (block&1);

  788. //        const int block_y= (block>>1);

  789. #if 1 // this saves us a bit of cliping work and shouldnt affect compression in a negative way

  790.         const int rel_xmin4= xmin;

  791.         const int rel_xmax4= xmax;

  792.         const int rel_ymin4= ymin;

  793.         const int rel_ymax4= ymax;

  794. #else

  795.         const int rel_xmin4= xmin - block_x*8;

  796.         const int rel_xmax4= xmax - block_x*8 + 8;

  797.         const int rel_ymin4= ymin - block_y*8;

  798.         const int rel_ymax4= ymax - block_y*8 + 8;

  799. #endif

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

  801.         P_LEFT[1] = s->motion_val[mot_xy - 1][1];

  802. 

  803.         if(P_LEFT[0]       > (rel_xmax4<<shift)) P_LEFT[0]       = (rel_xmax4<<shift);

  804. 

  805.         /* special case for first line */

  806.         if (s->mb_y == 0 && block<2) {

  807.             pred_x4= P_LEFT[0];

  808.             pred_y4= P_LEFT[1];

  809.         } else {

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

  811.             P_TOP[1]      = s->motion_val[mot_xy - mot_stride             ][1];

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

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

  814.             if(P_TOP[1]      > (rel_ymax4<<shift)) P_TOP[1]     = (rel_ymax4<<shift);

  815.             if(P_TOPRIGHT[0] < (rel_xmin4<<shift)) P_TOPRIGHT[0]= (rel_xmin4<<shift);

  816.             if(P_TOPRIGHT[0] > (rel_xmax4<<shift)) P_TOPRIGHT[0]= (rel_xmax4<<shift);

  817.             if(P_TOPRIGHT[1] > (rel_ymax4<<shift)) P_TOPRIGHT[1]= (rel_ymax4<<shift);

  818.     

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

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

  821. 

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

  823.                 pred_x4 = P_MEDIAN[0];

  824.                 pred_y4 = P_MEDIAN[1];

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

  826.                 pred_x4= P_LEFT[0];

  827.                 pred_y4= P_LEFT[1];

  828.             }

  829.         }

  830.         P_MV1[0]= mx;

  831.         P_MV1[1]= my;

  832. 

  833.         dmin4 = s->me.motion_search[1](s, block, &mx4, &my4, P, pred_x4, pred_y4, rel_xmin4, rel_ymin4, rel_xmax4, rel_ymax4, 

  834.                                        &s->last_picture, s->p_mv_table, (1<<16)>>shift, mv_penalty);

  835. 

  836.         dmin4= s->me.sub_motion_search(s, &mx4, &my4, dmin4, rel_xmin4, rel_ymin4, rel_xmax4, rel_ymax4, 

  837. 					  pred_x4, pred_y4, &s->last_picture, block, 1, mv_penalty);

  838.  

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

  840.         s->motion_val[ s->block_index[block] ][1]= my4;

  841.         dmin_sum+= dmin4;

  842.     }

  843.     return dmin_sum;

  844. }

  845. 

  846. void ff_estimate_p_frame_motion(MpegEncContext * s,

  847.                                 int mb_x, int mb_y)

  848. {

  849.     UINT8 *pix, *ppix;

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

  851.     int xmin, ymin, xmax, ymax;

  852.     int rel_xmin, rel_ymin, rel_xmax, rel_ymax;

  853.     int pred_x=0, pred_y=0;

  854.     int P[10][2];

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

  856.     int mb_type=0;

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

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

  859.     uint16_t * const mv_penalty= s->me.mv_penalty[s->f_code] + MAX_MV;

  860.     

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

  862. 

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

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

  865. 

  866.     get_limits(s, &range, &xmin, &ymin, &xmax, &ymax, s->f_code);

  867.     rel_xmin= xmin - mb_x*16;

  868.     rel_xmax= xmax - mb_x*16;

  869.     rel_ymin= ymin - mb_y*16;

  870.     rel_ymax= ymax - mb_y*16;

  871.     s->me.skip=0;

  872. 

  873.     switch(s->me_method) {

  874.     case ME_ZERO:

  875.     default:

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

  877.         mx-= mb_x*16;

  878.         my-= mb_y*16;

  879.         dmin = 0;

  880.         break;

  881.     case ME_FULL:

  882. 	dmin = full_motion_search(s, &mx, &my, range, xmin, ymin, xmax, ymax, ref_picture);

  883.         mx-= mb_x*16;

  884.         my-= mb_y*16;

  885.         break;

  886.     case ME_LOG:

  887. 	dmin = log_motion_search(s, &mx, &my, range / 2, xmin, ymin, xmax, ymax, ref_picture);

  888.         mx-= mb_x*16;

  889.         my-= mb_y*16;

  890.         break;

  891.     case ME_PHODS:

  892. 	dmin = phods_motion_search(s, &mx, &my, range / 2, xmin, ymin, xmax, ymax, ref_picture);

  893.         mx-= mb_x*16;

  894.         my-= mb_y*16;

  895.         break;

  896.     case ME_X1:

  897.     case ME_EPZS:

  898.        {

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

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

  901. 

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

  903.             P_LEFT[1]       = s->motion_val[mot_xy - 1][1];

  904. 

  905.             if(P_LEFT[0]       > (rel_xmax<<shift)) P_LEFT[0]       = (rel_xmax<<shift);

  906. 

  907.             if(mb_y) {

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

  909.                 P_TOP[1]      = s->motion_val[mot_xy - mot_stride    ][1];

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

  911.                 P_TOPRIGHT[1] = s->motion_val[mot_xy - mot_stride + 2][1];

  912.                 if(P_TOP[1]      > (rel_ymax<<shift)) P_TOP[1]     = (rel_ymax<<shift);

  913.                 if(P_TOPRIGHT[0] < (rel_xmin<<shift)) P_TOPRIGHT[0]= (rel_xmin<<shift);

  914.                 if(P_TOPRIGHT[1] > (rel_ymax<<shift)) P_TOPRIGHT[1]= (rel_ymax<<shift);

  915.         

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

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

  918. 

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

  920.                     pred_x = P_MEDIAN[0];

  921.                     pred_y = P_MEDIAN[1];

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

  923.                     pred_x= P_LEFT[0];

  924.                     pred_y= P_LEFT[1];

  925.                 }

  926.             }else{

  927.                 pred_x= P_LEFT[0];

  928.                 pred_y= P_LEFT[1];

  929.             }

  930. 

  931.         }

  932.         dmin = s->me.motion_search[0](s, 0, &mx, &my, P, pred_x, pred_y, rel_xmin, rel_ymin, rel_xmax, rel_ymax, 

  933.                                       &s->last_picture, s->p_mv_table, (1<<16)>>shift, mv_penalty);

  934.  

  935.         break;

  936.     }

  937. 

  938.     /* intra / predictive decision */

  939.     xx = mb_x * 16;

  940.     yy = mb_y * 16;

  941. 

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

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

  944.     ppix = ref_picture + ((yy+my) * s->linesize) + (xx+mx);

  945.     

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

  947.     

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

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

  950. 

  951. //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);

  952.     pic->mb_var   [s->mb_width * mb_y + mb_x] = varc;

  953.     pic->mc_mb_var[s->mb_width * mb_y + mb_x] = vard;

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

  955.     pic->mb_var_sum    += varc;

  956.     pic->mc_mb_var_sum += vard;

  957. //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);

  958.     

  959. #if 0

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

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

  962. #endif

  963.     if(s->flags&CODEC_FLAG_HQ){

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

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

  966.         else

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

  968. 

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

  970.             mb_type|= MB_TYPE_INTRA;

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

  972.             mb_type|= MB_TYPE_INTER;

  973.             s->me.sub_motion_search(s, &mx, &my, dmin, rel_xmin, rel_ymin, rel_xmax, rel_ymax,

  974. 				   pred_x, pred_y, &s->last_picture, 0, 0, mv_penalty);

  975.         }else{

  976.             mx <<=shift;

  977.             my <<=shift;

  978.         }

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

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

  981.             mv4_search(s, rel_xmin, rel_ymin, rel_xmax, rel_ymax, mx, my, shift);

  982.             mb_type|=MB_TYPE_INTER4V;

  983. 

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

  985.         }else

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

  987.     }else{

  988.         if (vard <= 64 || vard < varc) {

  989. //        if (sadP <= 32 || sadP < sadI + 500) {

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

  991.             mb_type|= MB_TYPE_INTER;

  992.             if (s->me_method != ME_ZERO) {

  993.                 dmin= s->me.sub_motion_search(s, &mx, &my, dmin, rel_xmin, rel_ymin, rel_xmax, rel_ymax,

  994.                                             pred_x, pred_y, &s->last_picture, 0, 0, mv_penalty);

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

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

  997.                     int dmin4= mv4_search(s, rel_xmin, rel_ymin, rel_xmax, rel_ymax, mx, my, shift);

  998.                     if(dmin4 + 128 <dmin)

  999.                         mb_type= MB_TYPE_INTER4V;

  1000.                 }

  1001.                 set_p_mv_tables(s, mx, my, mb_type!=MB_TYPE_INTER4V);

  1002. 

  1003.             } else {

  1004.                 mx <<=shift;

  1005.                 my <<=shift;

  1006.             }

  1007. #if 0

  1008.             if (vard < 10) {

  1009.                 skip++;

  1010.                 fprintf(stderr,"\nEarly skip: %d vard: %2d varc: %5d dmin: %d", 

  1011.                                 skip, vard, varc, dmin);

  1012.             }

  1013. #endif

  1014.         }else{

  1015.             s->scene_change_score+= 20;

  1016.             mb_type|= MB_TYPE_INTRA;

  1017.             mx = 0;

  1018.             my = 0;

  1019.         }

  1020.     }

  1021. 

  1022.     s->mb_type[mb_y*s->mb_width + mb_x]= mb_type;

  1023. }

  1024. 

  1025. int ff_pre_estimate_p_frame_motion(MpegEncContext * s,

  1026.                                     int mb_x, int mb_y)

  1027. {

  1028.     int mx, my, range, dmin;

  1029.     int xmin, ymin, xmax, ymax;

  1030.     int rel_xmin, rel_ymin, rel_xmax, rel_ymax;

  1031.     int pred_x=0, pred_y=0;

  1032.     int P[10][2];

  1033.     const int shift= 1+s->quarter_sample;

  1034.     uint16_t * const mv_penalty= s->me.mv_penalty[s->f_code] + MAX_MV;

  1035.     const int mv_stride= s->mb_width + 2;

  1036.     const int xy= mb_x + 1 + (mb_y + 1)*mv_stride;

  1037.     

  1038.     assert(s->quarter_sample==0 || s->quarter_sample==1);

  1039. 

  1040.     s->me.penalty_factor    = get_penalty_factor(s, s->avctx->me_cmp);

  1041. 

  1042.     get_limits(s, &range, &xmin, &ymin, &xmax, &ymax, s->f_code);

  1043.     rel_xmin= xmin - mb_x*16;

  1044.     rel_xmax= xmax - mb_x*16;

  1045.     rel_ymin= ymin - mb_y*16;

  1046.     rel_ymax= ymax - mb_y*16;

  1047.     s->me.skip=0;

  1048. 

  1049.     P_LEFT[0]       = s->p_mv_table[xy + 1][0];

  1050.     P_LEFT[1]       = s->p_mv_table[xy + 1][1];

  1051. 

  1052.     if(P_LEFT[0]       < (rel_xmin<<shift)) P_LEFT[0]       = (rel_xmin<<shift);

  1053. 

  1054.     /* special case for first line */

  1055.     if (mb_y == s->mb_height-1) {

  1056.         pred_x= P_LEFT[0];

  1057.         pred_y= P_LEFT[1];

  1058.         P_TOP[0]= P_TOPRIGHT[0]= P_MEDIAN[0]=

  1059.         P_TOP[1]= P_TOPRIGHT[1]= P_MEDIAN[1]= 0; //FIXME 

  1060.     } else {

  1061.         P_TOP[0]      = s->p_mv_table[xy + mv_stride    ][0];

  1062.         P_TOP[1]      = s->p_mv_table[xy + mv_stride    ][1];

  1063.         P_TOPRIGHT[0] = s->p_mv_table[xy + mv_stride - 1][0];

  1064.         P_TOPRIGHT[1] = s->p_mv_table[xy + mv_stride - 1][1];

  1065.         if(P_TOP[1]      < (rel_ymin<<shift)) P_TOP[1]     = (rel_ymin<<shift);

  1066.         if(P_TOPRIGHT[0] > (rel_xmax<<shift)) P_TOPRIGHT[0]= (rel_xmax<<shift);

  1067.         if(P_TOPRIGHT[1] < (rel_ymin<<shift)) P_TOPRIGHT[1]= (rel_ymin<<shift);

  1068.     

  1069.         P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);

  1070.         P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);

  1071. 

  1072.         pred_x = P_MEDIAN[0];

  1073.         pred_y = P_MEDIAN[1];

  1074.     }

  1075.     dmin = s->me.motion_search[0](s, 0, &mx, &my, P, pred_x, pred_y, rel_xmin, rel_ymin, rel_xmax, rel_ymax, 

  1076.                                   &s->last_picture, s->p_mv_table, (1<<16)>>shift, mv_penalty);

  1077. 

  1078.     s->p_mv_table[xy][0] = mx<<shift;

  1079.     s->p_mv_table[xy][1] = my<<shift;

  1080.     

  1081.     return dmin;

  1082. }

  1083. 

  1084. int ff_estimate_motion_b(MpegEncContext * s,

  1085.                        int mb_x, int mb_y, int16_t (*mv_table)[2], Picture *picture, int f_code)

  1086. {

  1087.     int mx, my, range, dmin;

  1088.     int xmin, ymin, xmax, ymax;

  1089.     int rel_xmin, rel_ymin, rel_xmax, rel_ymax;

  1090.     int pred_x=0, pred_y=0;

  1091.     int P[10][2];

  1092.     const int shift= 1+s->quarter_sample;

  1093.     const int mot_stride = s->mb_width + 2;

  1094.     const int mot_xy = (mb_y + 1)*mot_stride + mb_x + 1;

  1095.     uint8_t * const ref_picture= picture->data[0];

  1096.     uint16_t * const mv_penalty= s->me.mv_penalty[f_code] + MAX_MV;

  1097.     int mv_scale;

  1098.         

  1099.     s->me.penalty_factor    = get_penalty_factor(s, s->avctx->me_cmp);

  1100.     s->me.sub_penalty_factor= get_penalty_factor(s, s->avctx->me_sub_cmp);

  1101. 

  1102.     get_limits(s, &range, &xmin, &ymin, &xmax, &ymax, f_code);

  1103.     rel_xmin= xmin - mb_x*16;

  1104.     rel_xmax= xmax - mb_x*16;

  1105.     rel_ymin= ymin - mb_y*16;

  1106.     rel_ymax= ymax - mb_y*16;

  1107. 

  1108.     switch(s->me_method) {

  1109.     case ME_ZERO:

  1110.     default:

  1111. 	no_motion_search(s, &mx, &my);

  1112.         dmin = 0;

  1113.         mx-= mb_x*16;

  1114.         my-= mb_y*16;

  1115.         break;

  1116.     case ME_FULL:

  1117. 	dmin = full_motion_search(s, &mx, &my, range, xmin, ymin, xmax, ymax, ref_picture);

  1118.         mx-= mb_x*16;

  1119.         my-= mb_y*16;

  1120.         break;

  1121.     case ME_LOG:

  1122. 	dmin = log_motion_search(s, &mx, &my, range / 2, xmin, ymin, xmax, ymax, ref_picture);

  1123.         mx-= mb_x*16;

  1124.         my-= mb_y*16;

  1125.         break;

  1126.     case ME_PHODS:

  1127. 	dmin = phods_motion_search(s, &mx, &my, range / 2, xmin, ymin, xmax, ymax, ref_picture);

  1128.         mx-= mb_x*16;

  1129.         my-= mb_y*16;

  1130.         break;

  1131.     case ME_X1:

  1132.     case ME_EPZS:

  1133.        {

  1134.             P_LEFT[0]        = mv_table[mot_xy - 1][0];

  1135.             P_LEFT[1]        = mv_table[mot_xy - 1][1];

  1136. 

  1137.             if(P_LEFT[0]       > (rel_xmax<<shift)) P_LEFT[0]       = (rel_xmax<<shift);

  1138. 

  1139.             /* special case for first line */

  1140.             if (mb_y) {

  1141.                 P_TOP[0] = mv_table[mot_xy - mot_stride             ][0];

  1142.                 P_TOP[1] = mv_table[mot_xy - mot_stride             ][1];

  1143.                 P_TOPRIGHT[0] = mv_table[mot_xy - mot_stride + 1         ][0];

  1144.                 P_TOPRIGHT[1] = mv_table[mot_xy - mot_stride + 1         ][1];

  1145.                 if(P_TOP[1] > (rel_ymax<<shift)) P_TOP[1]= (rel_ymax<<shift);

  1146.                 if(P_TOPRIGHT[0] < (rel_xmin<<shift)) P_TOPRIGHT[0]= (rel_xmin<<shift);

  1147.                 if(P_TOPRIGHT[1] > (rel_ymax<<shift)) P_TOPRIGHT[1]= (rel_ymax<<shift);

  1148.         

  1149.                 P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);

  1150.                 P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);

  1151.             }

  1152.             pred_x= P_LEFT[0];

  1153.             pred_y= P_LEFT[1];

  1154.         }

  1155.         

  1156.         if(mv_table == s->b_forw_mv_table){

  1157.             mv_scale= (s->pb_time<<16) / (s->pp_time<<shift);

  1158.         }else{

  1159.             mv_scale= ((s->pb_time - s->pp_time)<<16) / (s->pp_time<<shift);

  1160.         }

  1161.         

  1162.         dmin = s->me.motion_search[0](s, 0, &mx, &my, P, pred_x, pred_y, rel_xmin, rel_ymin, rel_xmax, rel_ymax, 

  1163.                                       picture, s->p_mv_table, mv_scale, mv_penalty);

  1164.  

  1165.         break;

  1166.     }

  1167.     

  1168.     dmin= s->me.sub_motion_search(s, &mx, &my, dmin, rel_xmin, rel_ymin, rel_xmax, rel_ymax,

  1169. 				   pred_x, pred_y, picture, 0, 0, mv_penalty);

  1170. //printf("%d %d %d %d//", s->mb_x, s->mb_y, mx, my);

  1171. //    s->mb_type[mb_y*s->mb_width + mb_x]= mb_type;

  1172.     mv_table[mot_xy][0]= mx;

  1173.     mv_table[mot_xy][1]= my;

  1174. 

  1175.     return dmin;

  1176. }

  1177. 

  1178. static inline int check_bidir_mv(MpegEncContext * s,

  1179.                    int mb_x, int mb_y,

  1180.                    int motion_fx, int motion_fy,

  1181.                    int motion_bx, int motion_by,

  1182.                    int pred_fx, int pred_fy,

  1183.                    int pred_bx, int pred_by)

  1184. {

  1185.     //FIXME optimize?

  1186.     //FIXME move into template?

  1187.     //FIXME better f_code prediction (max mv & distance)

  1188.     UINT16 *mv_penalty= s->me.mv_penalty[s->f_code] + MAX_MV; // f_code of the prev frame

  1189.     uint8_t *dest_y = s->me.scratchpad;

  1190.     uint8_t *ptr;

  1191.     int dxy;

  1192.     int src_x, src_y;

  1193.     int fbmin;

  1194. 

  1195.     if(s->quarter_sample){

  1196.         dxy = ((motion_fy & 3) << 2) | (motion_fx & 3);

  1197.         src_x = mb_x * 16 + (motion_fx >> 2);

  1198.         src_y = mb_y * 16 + (motion_fy >> 2);

  1199.         assert(src_x >=-16 && src_x<=s->width);

  1200.         assert(src_y >=-16 && src_y<=s->height);

  1201. 

  1202.         ptr = s->last_picture.data[0] + (src_y * s->linesize) + src_x;

  1203.         s->dsp.put_qpel_pixels_tab[0][dxy](dest_y    , ptr    , s->linesize);

  1204. 

  1205.         dxy = ((motion_by & 3) << 2) | (motion_bx & 3);

  1206.         src_x = mb_x * 16 + (motion_bx >> 2);

  1207.         src_y = mb_y * 16 + (motion_by >> 2);

  1208.         assert(src_x >=-16 && src_x<=s->width);

  1209.         assert(src_y >=-16 && src_y<=s->height);

  1210.     

  1211.         ptr = s->next_picture.data[0] + (src_y * s->linesize) + src_x;

  1212.         s->dsp.avg_qpel_pixels_tab[0][dxy](dest_y    , ptr    , s->linesize);

  1213.     }else{

  1214.         dxy = ((motion_fy & 1) << 1) | (motion_fx & 1);

  1215.         src_x = mb_x * 16 + (motion_fx >> 1);

  1216.         src_y = mb_y * 16 + (motion_fy >> 1);

  1217.         assert(src_x >=-16 && src_x<=s->width);

  1218.         assert(src_y >=-16 && src_y<=s->height);

  1219. 

  1220.         ptr = s->last_picture.data[0] + (src_y * s->linesize) + src_x;

  1221.         s->dsp.put_pixels_tab[0][dxy](dest_y    , ptr    , s->linesize, 16);

  1222. 

  1223.         dxy = ((motion_by & 1) << 1) | (motion_bx & 1);

  1224.         src_x = mb_x * 16 + (motion_bx >> 1);

  1225.         src_y = mb_y * 16 + (motion_by >> 1);

  1226.         assert(src_x >=-16 && src_x<=s->width);

  1227.         assert(src_y >=-16 && src_y<=s->height);

  1228.     

  1229.         ptr = s->next_picture.data[0] + (src_y * s->linesize) + src_x;

  1230.         s->dsp.avg_pixels_tab[0][dxy](dest_y    , ptr    , s->linesize, 16);

  1231.     }

  1232. 

  1233.     fbmin = (mv_penalty[motion_fx-pred_fx] + mv_penalty[motion_fy-pred_fy])*s->me.sub_penalty_factor

  1234.            +(mv_penalty[motion_bx-pred_bx] + mv_penalty[motion_by-pred_by])*s->me.sub_penalty_factor;

  1235.            + s->dsp.me_sub_cmp[0](s, s->new_picture.data[0] + mb_x*16 + mb_y*16*s->linesize, dest_y, s->linesize);

  1236. 

  1237.     return fbmin;

  1238. }

  1239. 

  1240. /* refine the bidir vectors in hq mode and return the score in both lq & hq mode*/

  1241. static inline int bidir_refine(MpegEncContext * s,

  1242.                                   int mb_x, int mb_y)

  1243. {

  1244.     const int mot_stride = s->mb_width + 2;

  1245.     const int xy = (mb_y + 1)*mot_stride + mb_x + 1;

  1246.     int fbmin;

  1247.     int pred_fx= s->b_bidir_forw_mv_table[xy-1][0];

  1248.     int pred_fy= s->b_bidir_forw_mv_table[xy-1][1];

  1249.     int pred_bx= s->b_bidir_back_mv_table[xy-1][0];

  1250.     int pred_by= s->b_bidir_back_mv_table[xy-1][1];

  1251.     int motion_fx= s->b_bidir_forw_mv_table[xy][0]= s->b_forw_mv_table[xy][0];

  1252.     int motion_fy= s->b_bidir_forw_mv_table[xy][1]= s->b_forw_mv_table[xy][1];

  1253.     int motion_bx= s->b_bidir_back_mv_table[xy][0]= s->b_back_mv_table[xy][0];

  1254.     int motion_by= s->b_bidir_back_mv_table[xy][1]= s->b_back_mv_table[xy][1];

  1255. 

  1256.     //FIXME do refinement and add flag

  1257.     

  1258.     fbmin= check_bidir_mv(s, mb_x, mb_y, 

  1259.                           motion_fx, motion_fy,

  1260.                           motion_bx, motion_by,

  1261.                           pred_fx, pred_fy,

  1262.                           pred_bx, pred_by);

  1263. 

  1264.    return fbmin;

  1265. }

  1266. 

  1267. static inline int direct_search(MpegEncContext * s,

  1268.                                 int mb_x, int mb_y)

  1269. {

  1270.     int P[10][2];

  1271.     const int mot_stride = s->mb_width + 2;

  1272.     const int mot_xy = (mb_y + 1)*mot_stride + mb_x + 1;

  1273.     const int shift= 1+s->quarter_sample;

  1274.     int dmin, i;

  1275.     const int time_pp= s->pp_time;

  1276.     const int time_pb= s->pb_time;

  1277.     int mx, my, xmin, xmax, ymin, ymax;

  1278.     int16_t (*mv_table)[2]= s->b_direct_mv_table;

  1279.     uint16_t * const mv_penalty= s->me.mv_penalty[1] + MAX_MV;

  1280.     

  1281.     ymin= xmin=(-32)>>shift;

  1282.     ymax= xmax=   31>>shift;

  1283. 

  1284.     if(s->co_located_type_table[mb_x + mb_y*s->mb_width]==CO_LOCATED_TYPE_4MV){

  1285.         s->mv_type= MV_TYPE_8X8;

  1286.     }else{

  1287.         s->mv_type= MV_TYPE_16X16;

  1288.     }

  1289. 

  1290.     for(i=0; i<4; i++){

  1291.         int index= s->block_index[i];

  1292.         int min, max;

  1293.     

  1294.         s->me.co_located_mv[i][0]= s->motion_val[index][0];

  1295.         s->me.co_located_mv[i][1]= s->motion_val[index][1];

  1296.         s->me.direct_basis_mv[i][0]= s->me.co_located_mv[i][0]*time_pb/time_pp + ((i& 1)<<(shift+3));

  1297.         s->me.direct_basis_mv[i][1]= s->me.co_located_mv[i][1]*time_pb/time_pp + ((i>>1)<<(shift+3));

  1298. //        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);

  1299. //        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);

  1300. 

  1301.         max= FFMAX(s->me.direct_basis_mv[i][0], s->me.direct_basis_mv[i][0] - s->me.co_located_mv[i][0])>>shift;

  1302.         min= FFMIN(s->me.direct_basis_mv[i][0], s->me.direct_basis_mv[i][0] - s->me.co_located_mv[i][0])>>shift;

  1303.         max+= (2*mb_x + (i& 1))*8 - 1; // +-1 is for the simpler rounding

  1304.         min+= (2*mb_x + (i& 1))*8 + 1;

  1305.         if(max >= s->width) xmax= s->width - max - 1;

  1306.         if(min < -16      ) xmin= - 32 - min;

  1307. 

  1308.         max= FFMAX(s->me.direct_basis_mv[i][1], s->me.direct_basis_mv[i][1] - s->me.co_located_mv[i][1])>>shift;

  1309.         min= FFMIN(s->me.direct_basis_mv[i][1], s->me.direct_basis_mv[i][1] - s->me.co_located_mv[i][1])>>shift;

  1310.         max+= (2*mb_y + (i>>1))*8 - 1; // +-1 is for the simpler rounding

  1311.         min+= (2*mb_y + (i>>1))*8 + 1;

  1312.         if(max >= s->height) ymax= s->height - max - 1;

  1313.         if(min < -16       ) ymin= - 32 - min;

  1314.         

  1315.         if(s->mv_type == MV_TYPE_16X16) break;

  1316.     }

  1317.     

  1318.     assert(xmax <= 15 && ymax <= 15 && xmin >= -16 && ymin >= -16);

  1319.     

  1320.     if(xmax < 0 || xmin >0 || ymax < 0 || ymin > 0){

  1321.         s->b_direct_mv_table[mot_xy][0]= 0;

  1322.         s->b_direct_mv_table[mot_xy][1]= 0;

  1323. 

  1324.         return 256*256*256*64;

  1325.     }

  1326. 

  1327.     P_LEFT[0]        = clip(mv_table[mot_xy - 1][0], xmin<<shift, xmax<<shift);

  1328.     P_LEFT[1]        = clip(mv_table[mot_xy - 1][1], ymin<<shift, ymax<<shift);

  1329. 

  1330.     /* special case for first line */

  1331.     if (mb_y) {

  1332.         P_TOP[0]      = clip(mv_table[mot_xy - mot_stride             ][0], xmin<<shift, xmax<<shift);

  1333.         P_TOP[1]      = clip(mv_table[mot_xy - mot_stride             ][1], ymin<<shift, ymax<<shift);

  1334.         P_TOPRIGHT[0] = clip(mv_table[mot_xy - mot_stride + 1         ][0], xmin<<shift, xmax<<shift);

  1335.         P_TOPRIGHT[1] = clip(mv_table[mot_xy - mot_stride + 1         ][1], ymin<<shift, ymax<<shift);

  1336.     

  1337.         P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);

  1338.         P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);

  1339.     }

  1340.     

  1341.     if(s->flags&CODEC_FLAG_QPEL){

  1342.         dmin = simple_direct_qpel_epzs_motion_search(s, 0, &mx, &my, P, 0, 0, xmin, ymin, xmax, ymax, 

  1343.                                                      &s->last_picture, mv_table, 1<<14, mv_penalty);

  1344.         dmin = simple_direct_qpel_qpel_motion_search(s, &mx, &my, dmin, xmin, ymin, xmax, ymax,

  1345.                                                 0, 0, &s->last_picture, 0, 0, mv_penalty);

  1346.     }else{

  1347.         dmin = simple_direct_hpel_epzs_motion_search(s, 0, &mx, &my, P, 0, 0, xmin, ymin, xmax, ymax, 

  1348.                                                      &s->last_picture, mv_table, 1<<15, mv_penalty);

  1349.         dmin = simple_direct_hpel_hpel_motion_search(s, &mx, &my, dmin, xmin, ymin, xmax, ymax,

  1350.                                                 0, 0, &s->last_picture, 0, 0, mv_penalty);

  1351.     }

  1352. 

  1353.     s->b_direct_mv_table[mot_xy][0]= mx;

  1354.     s->b_direct_mv_table[mot_xy][1]= my;

  1355.     return dmin;

  1356. }

  1357. 

  1358. void ff_estimate_b_frame_motion(MpegEncContext * s,

  1359.                              int mb_x, int mb_y)

  1360. {

  1361.     const int penalty_factor= s->me.penalty_factor;

  1362.     int fmin, bmin, dmin, fbmin;

  1363.     int type=0;

  1364.     

  1365.     dmin= direct_search(s, mb_x, mb_y);

  1366. 

  1367.     fmin= ff_estimate_motion_b(s, mb_x, mb_y, s->b_forw_mv_table, &s->last_picture, s->f_code);

  1368.     bmin= ff_estimate_motion_b(s, mb_x, mb_y, s->b_back_mv_table, &s->next_picture, s->b_code) - penalty_factor;

  1369. //printf(" %d %d ", s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1]);

  1370. 

  1371.     fbmin= bidir_refine(s, mb_x, mb_y);

  1372. 

  1373.     {

  1374.         int score= dmin;

  1375.         type=MB_TYPE_DIRECT;

  1376.         

  1377.         if(fmin<score){

  1378.             score=fmin;

  1379.             type= MB_TYPE_FORWARD; 

  1380.         }

  1381.         if(bmin<score){

  1382.             score=bmin;

  1383.             type= MB_TYPE_BACKWARD; 

  1384.         }

  1385.         if(fbmin<score){

  1386.             score=fbmin;

  1387.             type= MB_TYPE_BIDIR;

  1388.         }

  1389.         score= ((unsigned)(score*score + 128*256))>>16;

  1390.         s->current_picture.mc_mb_var_sum += score;

  1391.         s->current_picture.mc_mb_var[mb_y*s->mb_width + mb_x] = score; //FIXME use SSD

  1392.     }

  1393. 

  1394.     if(s->flags&CODEC_FLAG_HQ){

  1395.         type= MB_TYPE_FORWARD | MB_TYPE_BACKWARD | MB_TYPE_BIDIR | MB_TYPE_DIRECT; //FIXME something smarter

  1396.         if(dmin>256*256*16) type&= ~MB_TYPE_DIRECT; //dont try direct mode if its invalid for this MB

  1397.     }

  1398. 

  1399.     s->mb_type[mb_y*s->mb_width + mb_x]= type;

  1400. }

  1401. 

  1402. /* find best f_code for ME which do unlimited searches */

  1403. int ff_get_best_fcode(MpegEncContext * s, int16_t (*mv_table)[2], int type)

  1404. {

  1405.     if(s->me_method>=ME_EPZS){

  1406.         int score[8];

  1407.         int i, y;

  1408.         UINT8 * fcode_tab= s->fcode_tab;

  1409.         int best_fcode=-1;

  1410.         int best_score=-10000000;

  1411. 

  1412.         for(i=0; i<8; i++) score[i]= s->mb_num*(8-i);

  1413. 

  1414.         for(y=0; y<s->mb_height; y++){

  1415.             int x;

  1416.             int xy= (y+1)* (s->mb_width+2) + 1;

  1417.             i= y*s->mb_width;

  1418.             for(x=0; x<s->mb_width; x++){

  1419.                 if(s->mb_type[i] & type){

  1420.                     int fcode= FFMAX(fcode_tab[mv_table[xy][0] + MAX_MV],

  1421.                                      fcode_tab[mv_table[xy][1] + MAX_MV]);

  1422.                     int j;

  1423.                     

  1424.                     for(j=0; j<fcode && j<8; j++){

  1425.                         if(s->pict_type==B_TYPE || s->current_picture.mc_mb_var[i] < s->current_picture.mb_var[i])

  1426.                             score[j]-= 170;

  1427.                     }

  1428.                 }

  1429.                 i++;

  1430.                 xy++;

  1431.             }

  1432.         }

  1433.         

  1434.         for(i=1; i<8; i++){

  1435.             if(score[i] > best_score){

  1436.                 best_score= score[i];

  1437.                 best_fcode= i;

  1438.             }

  1439. //            printf("%d %d\n", i, score[i]);

  1440.         }

  1441. 

  1442. //    printf("fcode: %d type: %d\n", i, s->pict_type);

  1443.         return best_fcode;

  1444. /*        for(i=0; i<=MAX_FCODE; i++){

  1445.             printf("%d ", mv_num[i]);

  1446.         }

  1447.         printf("\n");*/

  1448.     }else{

  1449.         return 1;

  1450.     }

  1451. }

  1452. 

  1453. void ff_fix_long_p_mvs(MpegEncContext * s)

  1454. {

  1455.     const int f_code= s->f_code;

  1456.     int y;

  1457.     UINT8 * fcode_tab= s->fcode_tab;

  1458. //int clip=0;

  1459. //int noclip=0;

  1460.     /* clip / convert to intra 16x16 type MVs */

  1461.     for(y=0; y<s->mb_height; y++){

  1462.         int x;

  1463.         int xy= (y+1)* (s->mb_width+2)+1;

  1464.         int i= y*s->mb_width;

  1465.         for(x=0; x<s->mb_width; x++){

  1466.             if(s->mb_type[i]&MB_TYPE_INTER){

  1467.                 if(   fcode_tab[s->p_mv_table[xy][0] + MAX_MV] > f_code

  1468.                    || fcode_tab[s->p_mv_table[xy][0] + MAX_MV] == 0

  1469.                    || fcode_tab[s->p_mv_table[xy][1] + MAX_MV] > f_code

  1470.                    || fcode_tab[s->p_mv_table[xy][1] + MAX_MV] == 0 ){

  1471.                     s->mb_type[i] &= ~MB_TYPE_INTER;

  1472.                     s->mb_type[i] |= MB_TYPE_INTRA;

  1473.                     s->p_mv_table[xy][0] = 0;

  1474.                     s->p_mv_table[xy][1] = 0;

  1475. //clip++;

  1476.                 }

  1477. //else

  1478. //  noclip++;

  1479.             }

  1480.             xy++;

  1481.             i++;

  1482.         }

  1483.     }

  1484. //printf("%d no:%d %d//\n", clip, noclip, f_code);

  1485.     if(s->flags&CODEC_FLAG_4MV){

  1486.         const int wrap= 2+ s->mb_width*2;

  1487. 

  1488.         /* clip / convert to intra 8x8 type MVs */

  1489.         for(y=0; y<s->mb_height; y++){

  1490.             int xy= (y*2 + 1)*wrap + 1;

  1491.             int i= y*s->mb_width;

  1492.             int x;

  1493. 

  1494.             for(x=0; x<s->mb_width; x++){

  1495.                 if(s->mb_type[i]&MB_TYPE_INTER4V){

  1496.                     int block;

  1497.                     for(block=0; block<4; block++){

  1498.                         int off= (block& 1) + (block>>1)*wrap;

  1499.                         int mx= s->motion_val[ xy + off ][0];

  1500.                         int my= s->motion_val[ xy + off ][1];

  1501. 

  1502.                         if(   fcode_tab[mx + MAX_MV] > f_code

  1503.                            || fcode_tab[mx + MAX_MV] == 0

  1504.                            || fcode_tab[my + MAX_MV] > f_code

  1505.                            || fcode_tab[my + MAX_MV] == 0 ){

  1506.                             s->mb_type[i] &= ~MB_TYPE_INTER4V;

  1507.                             s->mb_type[i] |= MB_TYPE_INTRA;

  1508.                         }

  1509.                     }

  1510.                 }

  1511.                 xy+=2;

  1512.                 i++;

  1513.             }

  1514.         }

  1515.     }

  1516. }

  1517. 

  1518. void ff_fix_long_b_mvs(MpegEncContext * s, int16_t (*mv_table)[2], int f_code, int type)

  1519. {

  1520.     int y;

  1521.     UINT8 * fcode_tab= s->fcode_tab;

  1522. 

  1523.     /* clip / convert to intra 16x16 type MVs */

  1524.     for(y=0; y<s->mb_height; y++){

  1525.         int x;

  1526.         int xy= (y+1)* (s->mb_width+2)+1;

  1527.         int i= y*s->mb_width;

  1528.         for(x=0; x<s->mb_width; x++){

  1529.             if(   fcode_tab[mv_table[xy][0] + MAX_MV] > f_code

  1530.                || fcode_tab[mv_table[xy][0] + MAX_MV] == 0){

  1531.                 if(mv_table[xy][0]>0) mv_table[xy][0]=  (16<<f_code)-1;

  1532.                 else                  mv_table[xy][0]= -(16<<f_code);

  1533.             }

  1534.             if(   fcode_tab[mv_table[xy][1] + MAX_MV] > f_code

  1535.                || fcode_tab[mv_table[xy][1] + MAX_MV] == 0){

  1536.                 if(mv_table[xy][1]>0) mv_table[xy][1]=  (16<<f_code)-1;

  1537.                 else                  mv_table[xy][1]= -(16<<f_code);

  1538.             }

  1539.             xy++;

  1540.             i++;

  1541.         }

  1542.     }

  1543. }