Statistics
| Branch: | Revision:

ffmpeg / libavcodec / motion_est.c @ aacb6acc

History | View | Annotate | Download (73.9 KB)

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(MotionEstContext *c)
53
{
54
    c->map_generation+= 1<<(ME_MAP_MV_BITS*2);
55
    if(c->map_generation==0){
56
        c->map_generation= 1<<(ME_MAP_MV_BITS*2);
57
        memset(c->map, 0, sizeof(uint32_t)*ME_MAP_SIZE);
58
    }
59
    return c->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(MotionEstContext *c, uint8_t *src[3], uint8_t *ref[3], uint8_t *ref2[3], int x, int y, int ref_index){
81
    const int offset[3]= {
82
          y*c->  stride + x,
83
        ((y*c->uvstride + x)>>1),
84
        ((y*c->uvstride + x)>>1),
85
    };
86
    int i;
87
    for(i=0; i<3; i++){
88
        c->src[0][i]= src [i] + offset[i];
89
        c->ref[0][i]= ref [i] + offset[i];
90
    }
91
    if(ref_index){
92
        for(i=0; i<3; i++){
93
            c->ref[ref_index][i]= ref2[i] + offset[i];
94
        }
95
    }
96
}
97

    
98
static int get_flags(MotionEstContext *c, int direct, int chroma){
99
    return   ((c->avctx->flags&CODEC_FLAG_QPEL) ? FLAG_QPEL : 0)
100
           + (direct ? FLAG_DIRECT : 0)
101
           + (chroma ? FLAG_CHROMA : 0);
102
}
103

    
104
static always_inline int cmp(MpegEncContext *s, const int x, const int y, const int subx, const int suby,
105
                      const int size, const int h, int ref_index, int src_index,
106
                      me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){
107
    MotionEstContext * const c= &s->me;
108
    const int stride= c->stride;
109
    const int uvstride= c->uvstride;
110
    const int qpel= flags&FLAG_QPEL;
111
    const int chroma= flags&FLAG_CHROMA;
112
    const int dxy= subx + (suby<<(1+qpel)); //FIXME log2_subpel?
113
    const int hx= subx + (x<<(1+qpel));
114
    const int hy= suby + (y<<(1+qpel));
115
    uint8_t * const * const ref= c->ref[ref_index];
116
    uint8_t * const * const src= c->src[src_index];
117
    int d;
118
    //FIXME check chroma 4mv, (no crashes ...)
119
    if(flags&FLAG_DIRECT){
120
        if(x >= c->xmin && hx <= c->xmax<<(qpel+1) && y >= c->ymin && hy <= c->ymax<<(qpel+1)){
121
            const int time_pp= s->pp_time;
122
            const int time_pb= s->pb_time;
123
            const int mask= 2*qpel+1;
124
            if(s->mv_type==MV_TYPE_8X8){
125
                int i;
126
                for(i=0; i<4; i++){
127
                    int fx = c->direct_basis_mv[i][0] + hx;
128
                    int fy = c->direct_basis_mv[i][1] + hy;
129
                    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));
130
                    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));
131
                    int fxy= (fx&mask) + ((fy&mask)<<(qpel+1));
132
                    int bxy= (bx&mask) + ((by&mask)<<(qpel+1));
133

    
134
                    uint8_t *dst= c->temp + 8*(i&1) + 8*stride*(i>>1);
135
                    if(qpel){
136
                        c->qpel_put[1][fxy](dst, ref[0] + (fx>>2) + (fy>>2)*stride, stride);
137
                        c->qpel_avg[1][bxy](dst, ref[8] + (bx>>2) + (by>>2)*stride, stride);
138
                    }else{
139
                        c->hpel_put[1][fxy](dst, ref[0] + (fx>>1) + (fy>>1)*stride, stride, 8);
140
                        c->hpel_avg[1][bxy](dst, ref[8] + (bx>>1) + (by>>1)*stride, stride, 8);
141
                    }
142
                }
143
            }else{
144
                int fx = c->direct_basis_mv[0][0] + hx;
145
                int fy = c->direct_basis_mv[0][1] + hy;
146
                int bx = hx ? fx - c->co_located_mv[0][0] : (c->co_located_mv[0][0]*(time_pb - time_pp)/time_pp);
147
                int by = hy ? fy - c->co_located_mv[0][1] : (c->co_located_mv[0][1]*(time_pb - time_pp)/time_pp);
148
                int fxy= (fx&mask) + ((fy&mask)<<(qpel+1));
149
                int bxy= (bx&mask) + ((by&mask)<<(qpel+1));
150

    
151
                if(qpel){
152
                    c->qpel_put[1][fxy](c->temp               , ref[0] + (fx>>2) + (fy>>2)*stride               , stride);
153
                    c->qpel_put[1][fxy](c->temp + 8           , ref[0] + (fx>>2) + (fy>>2)*stride + 8           , stride);
154
                    c->qpel_put[1][fxy](c->temp     + 8*stride, ref[0] + (fx>>2) + (fy>>2)*stride     + 8*stride, stride);
155
                    c->qpel_put[1][fxy](c->temp + 8 + 8*stride, ref[0] + (fx>>2) + (fy>>2)*stride + 8 + 8*stride, stride);
156
                    c->qpel_avg[1][bxy](c->temp               , ref[8] + (bx>>2) + (by>>2)*stride               , stride);
157
                    c->qpel_avg[1][bxy](c->temp + 8           , ref[8] + (bx>>2) + (by>>2)*stride + 8           , stride);
158
                    c->qpel_avg[1][bxy](c->temp     + 8*stride, ref[8] + (bx>>2) + (by>>2)*stride     + 8*stride, stride);
159
                    c->qpel_avg[1][bxy](c->temp + 8 + 8*stride, ref[8] + (bx>>2) + (by>>2)*stride + 8 + 8*stride, stride);
160
                }else{
161
                    assert((fx>>1) + 16*s->mb_x >= -16);
162
                    assert((fy>>1) + 16*s->mb_y >= -16);
163
                    assert((fx>>1) + 16*s->mb_x <= s->width);
164
                    assert((fy>>1) + 16*s->mb_y <= s->height);
165
                    assert((bx>>1) + 16*s->mb_x >= -16);
166
                    assert((by>>1) + 16*s->mb_y >= -16);
167
                    assert((bx>>1) + 16*s->mb_x <= s->width);
168
                    assert((by>>1) + 16*s->mb_y <= s->height);
169

    
170
                    c->hpel_put[0][fxy](c->temp, ref[0] + (fx>>1) + (fy>>1)*stride, stride, 16);
171
                    c->hpel_avg[0][bxy](c->temp, ref[8] + (bx>>1) + (by>>1)*stride, stride, 16);
172
                }
173
            }
174
            d = cmp_func(s, c->temp, src[0], stride, 16);
175
        }else
176
            d= 256*256*256*32;
177
    }else{
178
        int uvdxy;              /* no, it might not be used uninitialized */
179
        if(dxy){
180
            if(qpel){
181
                c->qpel_put[size][dxy](c->temp, ref[0] + x + y*stride, stride); //FIXME prototype (add h)
182
                if(chroma){
183
                    int cx= hx/2;
184
                    int cy= hy/2;
185
                    cx= (cx>>1)|(cx&1);
186
                    cy= (cy>>1)|(cy&1);
187
                    uvdxy= (cx&1) + 2*(cy&1);
188
                    //FIXME x/y wrong, but mpeg4 qpel is sick anyway, we should drop as much of it as possible in favor for h264
189
                }
190
            }else{
191
                c->hpel_put[size][dxy](c->temp, ref[0] + x + y*stride, stride, h);
192
                if(chroma)
193
                    uvdxy= dxy | (x&1) | (2*(y&1));
194
            }
195
            d = cmp_func(s, c->temp, src[0], stride, h);
196
        }else{
197
            d = cmp_func(s, src[0], ref[0] + x + y*stride, stride, h);
198
            if(chroma)
199
                uvdxy= (x&1) + 2*(y&1);
200
        }
201
        if(chroma){
202
            uint8_t * const uvtemp= c->temp + 16*stride;
203
            c->hpel_put[size+1][uvdxy](uvtemp  , ref[1] + (x>>1) + (y>>1)*uvstride, uvstride, h>>1);
204
            c->hpel_put[size+1][uvdxy](uvtemp+8, ref[2] + (x>>1) + (y>>1)*uvstride, uvstride, h>>1);
205
            d += chroma_cmp_func(s, uvtemp  , src[1], uvstride, h>>1);
206
            d += chroma_cmp_func(s, uvtemp+8, src[2], uvstride, h>>1);
207
        }
208
    }
209
#if 0
210
    if(full_pel){
211
        const int index= (((y)<<ME_MAP_SHIFT) + (x))&(ME_MAP_SIZE-1);
212
        score_map[index]= d;
213
    }
214

215
    d += (c->mv_penalty[hx - c->pred_x] + c->mv_penalty[hy - c->pred_y])*c->penalty_factor;
216
#endif
217
    return d;
218
}
219

    
220
#include "motion_est_template.c"
221

    
222
static int zero_cmp(void *s, uint8_t *a, uint8_t *b, int stride, int h){
223
    return 0;
224
}
225

    
226
static void zero_hpel(uint8_t *a, const uint8_t *b, int stride, int h){
227
}
228

    
229
void ff_init_me(MpegEncContext *s){
230
    MotionEstContext * const c= &s->me;
231
    c->avctx= s->avctx;
232

    
233
    ff_set_cmp(&s->dsp, s->dsp.me_pre_cmp, c->avctx->me_pre_cmp);
234
    ff_set_cmp(&s->dsp, s->dsp.me_cmp, c->avctx->me_cmp);
235
    ff_set_cmp(&s->dsp, s->dsp.me_sub_cmp, c->avctx->me_sub_cmp);
236
    ff_set_cmp(&s->dsp, s->dsp.mb_cmp, c->avctx->mb_cmp);
237

    
238
    c->flags    = get_flags(c, 0, c->avctx->me_cmp    &FF_CMP_CHROMA);
239
    c->sub_flags= get_flags(c, 0, c->avctx->me_sub_cmp&FF_CMP_CHROMA);
240
    c->mb_flags = get_flags(c, 0, c->avctx->mb_cmp    &FF_CMP_CHROMA);
241

    
242
/*FIXME s->no_rounding b_type*/
243
    if(s->flags&CODEC_FLAG_QPEL){
244
        c->sub_motion_search= qpel_motion_search;
245
        c->qpel_avg= s->dsp.avg_qpel_pixels_tab;
246
        if(s->no_rounding) c->qpel_put= s->dsp.put_no_rnd_qpel_pixels_tab;
247
        else               c->qpel_put= s->dsp.put_qpel_pixels_tab;
248
    }else{
249
        if(c->avctx->me_sub_cmp&FF_CMP_CHROMA)
250
            c->sub_motion_search= hpel_motion_search;
251
        else if(   c->avctx->me_sub_cmp == FF_CMP_SAD
252
                && c->avctx->    me_cmp == FF_CMP_SAD
253
                && c->avctx->    mb_cmp == FF_CMP_SAD)
254
            c->sub_motion_search= sad_hpel_motion_search; // 2050 vs. 2450 cycles
255
        else
256
            c->sub_motion_search= hpel_motion_search;
257
    }
258
    c->hpel_avg= s->dsp.avg_pixels_tab;
259
    if(s->no_rounding) c->hpel_put= s->dsp.put_no_rnd_pixels_tab;
260
    else               c->hpel_put= s->dsp.put_pixels_tab;
261

    
262
    if(s->linesize){
263
        c->stride  = s->linesize;
264
        c->uvstride= s->uvlinesize;
265
    }else{
266
        c->stride  = 16*s->mb_width + 32;
267
        c->uvstride=  8*s->mb_width + 16;
268
    }
269

    
270
    // 8x8 fullpel search would need a 4x4 chroma compare, which we dont have yet, and even if we had the motion estimation code doesnt expect it
271
    if(s->codec_id != CODEC_ID_SNOW){
272
        if((c->avctx->me_cmp&FF_CMP_CHROMA)/* && !s->dsp.me_cmp[2]*/){
273
            s->dsp.me_cmp[2]= zero_cmp;
274
        }
275
        if((c->avctx->me_sub_cmp&FF_CMP_CHROMA) && !s->dsp.me_sub_cmp[2]){
276
            s->dsp.me_sub_cmp[2]= zero_cmp;
277
        }
278
        c->hpel_put[2][0]= c->hpel_put[2][1]=
279
        c->hpel_put[2][2]= c->hpel_put[2][3]= zero_hpel;
280
    }
281

    
282
    if(s->codec_id == CODEC_ID_H261){
283
        c->sub_motion_search= no_sub_motion_search;
284
    }
285

    
286
    c->temp= c->scratchpad;
287
}
288

    
289
#if 0
290
static int pix_dev(uint8_t * pix, int line_size, int mean)
291
{
292
    int s, i, j;
293

294
    s = 0;
295
    for (i = 0; i < 16; i++) {
296
        for (j = 0; j < 16; j += 8) {
297
            s += ABS(pix[0]-mean);
298
            s += ABS(pix[1]-mean);
299
            s += ABS(pix[2]-mean);
300
            s += ABS(pix[3]-mean);
301
            s += ABS(pix[4]-mean);
302
            s += ABS(pix[5]-mean);
303
            s += ABS(pix[6]-mean);
304
            s += ABS(pix[7]-mean);
305
            pix += 8;
306
        }
307
        pix += line_size - 16;
308
    }
309
    return s;
310
}
311
#endif
312

    
313
static inline void no_motion_search(MpegEncContext * s,
314
                                    int *mx_ptr, int *my_ptr)
315
{
316
    *mx_ptr = 16 * s->mb_x;
317
    *my_ptr = 16 * s->mb_y;
318
}
319

    
320
#if 0  /* the use of these functions is inside #if 0 */
321
static int full_motion_search(MpegEncContext * s,
322
                              int *mx_ptr, int *my_ptr, int range,
323
                              int xmin, int ymin, int xmax, int ymax, uint8_t *ref_picture)
324
{
325
    int x1, y1, x2, y2, xx, yy, x, y;
326
    int mx, my, dmin, d;
327
    uint8_t *pix;
328

329
    xx = 16 * s->mb_x;
330
    yy = 16 * s->mb_y;
331
    x1 = xx - range + 1;        /* we loose one pixel to avoid boundary pb with half pixel pred */
332
    if (x1 < xmin)
333
        x1 = xmin;
334
    x2 = xx + range - 1;
335
    if (x2 > xmax)
336
        x2 = xmax;
337
    y1 = yy - range + 1;
338
    if (y1 < ymin)
339
        y1 = ymin;
340
    y2 = yy + range - 1;
341
    if (y2 > ymax)
342
        y2 = ymax;
343
    pix = s->new_picture.data[0] + (yy * s->linesize) + xx;
344
    dmin = 0x7fffffff;
345
    mx = 0;
346
    my = 0;
347
    for (y = y1; y <= y2; y++) {
348
        for (x = x1; x <= x2; x++) {
349
            d = s->dsp.pix_abs[0][0](NULL, pix, ref_picture + (y * s->linesize) + x,
350
                             s->linesize, 16);
351
            if (d < dmin ||
352
                (d == dmin &&
353
                 (abs(x - xx) + abs(y - yy)) <
354
                 (abs(mx - xx) + abs(my - yy)))) {
355
                dmin = d;
356
                mx = x;
357
                my = y;
358
            }
359
        }
360
    }
361

362
    *mx_ptr = mx;
363
    *my_ptr = my;
364

365
#if 0
366
    if (*mx_ptr < -(2 * range) || *mx_ptr >= (2 * range) ||
367
        *my_ptr < -(2 * range) || *my_ptr >= (2 * range)) {
368
        fprintf(stderr, "error %d %d\n", *mx_ptr, *my_ptr);
369
    }
370
#endif
371
    return dmin;
372
}
373

    
374

    
375
static int log_motion_search(MpegEncContext * s,
376
                             int *mx_ptr, int *my_ptr, int range,
377
                             int xmin, int ymin, int xmax, int ymax, uint8_t *ref_picture)
378
{
379
    int x1, y1, x2, y2, xx, yy, x, y;
380
    int mx, my, dmin, d;
381
    uint8_t *pix;
382

    
383
    xx = s->mb_x << 4;
384
    yy = s->mb_y << 4;
385

    
386
    /* Left limit */
387
    x1 = xx - range;
388
    if (x1 < xmin)
389
        x1 = xmin;
390

    
391
    /* Right limit */
392
    x2 = xx + range;
393
    if (x2 > xmax)
394
        x2 = xmax;
395

    
396
    /* Upper limit */
397
    y1 = yy - range;
398
    if (y1 < ymin)
399
        y1 = ymin;
400

    
401
    /* Lower limit */
402
    y2 = yy + range;
403
    if (y2 > ymax)
404
        y2 = ymax;
405

    
406
    pix = s->new_picture.data[0] + (yy * s->linesize) + xx;
407
    dmin = 0x7fffffff;
408
    mx = 0;
409
    my = 0;
410

    
411
    do {
412
        for (y = y1; y <= y2; y += range) {
413
            for (x = x1; x <= x2; x += range) {
414
                d = s->dsp.pix_abs[0][0](NULL, pix, ref_picture + (y * s->linesize) + x, s->linesize, 16);
415
                if (d < dmin || (d == dmin && (abs(x - xx) + abs(y - yy)) < (abs(mx - xx) + abs(my - yy)))) {
416
                    dmin = d;
417
                    mx = x;
418
                    my = y;
419
                }
420
            }
421
        }
422

    
423
        range = range >> 1;
424

    
425
        x1 = mx - range;
426
        if (x1 < xmin)
427
            x1 = xmin;
428

    
429
        x2 = mx + range;
430
        if (x2 > xmax)
431
            x2 = xmax;
432

    
433
        y1 = my - range;
434
        if (y1 < ymin)
435
            y1 = ymin;
436

    
437
        y2 = my + range;
438
        if (y2 > ymax)
439
            y2 = ymax;
440

    
441
    } while (range >= 1);
442

    
443
#ifdef DEBUG
444
    av_log(s->avctx, AV_LOG_DEBUG, "log       - MX: %d\tMY: %d\n", mx, my);
445
#endif
446
    *mx_ptr = mx;
447
    *my_ptr = my;
448
    return dmin;
449
}
450

    
451
static int phods_motion_search(MpegEncContext * s,
452
                               int *mx_ptr, int *my_ptr, int range,
453
                               int xmin, int ymin, int xmax, int ymax, uint8_t *ref_picture)
454
{
455
    int x1, y1, x2, y2, xx, yy, x, y, lastx, d;
456
    int mx, my, dminx, dminy;
457
    uint8_t *pix;
458

    
459
    xx = s->mb_x << 4;
460
    yy = s->mb_y << 4;
461

    
462
    /* Left limit */
463
    x1 = xx - range;
464
    if (x1 < xmin)
465
        x1 = xmin;
466

    
467
    /* Right limit */
468
    x2 = xx + range;
469
    if (x2 > xmax)
470
        x2 = xmax;
471

    
472
    /* Upper limit */
473
    y1 = yy - range;
474
    if (y1 < ymin)
475
        y1 = ymin;
476

    
477
    /* Lower limit */
478
    y2 = yy + range;
479
    if (y2 > ymax)
480
        y2 = ymax;
481

    
482
    pix = s->new_picture.data[0] + (yy * s->linesize) + xx;
483
    mx = 0;
484
    my = 0;
485

    
486
    x = xx;
487
    y = yy;
488
    do {
489
        dminx = 0x7fffffff;
490
        dminy = 0x7fffffff;
491

    
492
        lastx = x;
493
        for (x = x1; x <= x2; x += range) {
494
            d = s->dsp.pix_abs[0][0](NULL, pix, ref_picture + (y * s->linesize) + x, s->linesize, 16);
495
            if (d < dminx || (d == dminx && (abs(x - xx) + abs(y - yy)) < (abs(mx - xx) + abs(my - yy)))) {
496
                dminx = d;
497
                mx = x;
498
            }
499
        }
500

    
501
        x = lastx;
502
        for (y = y1; y <= y2; y += range) {
503
            d = s->dsp.pix_abs[0][0](NULL, pix, ref_picture + (y * s->linesize) + x, s->linesize, 16);
504
            if (d < dminy || (d == dminy && (abs(x - xx) + abs(y - yy)) < (abs(mx - xx) + abs(my - yy)))) {
505
                dminy = d;
506
                my = y;
507
            }
508
        }
509

    
510
        range = range >> 1;
511

    
512
        x = mx;
513
        y = my;
514
        x1 = mx - range;
515
        if (x1 < xmin)
516
            x1 = xmin;
517

    
518
        x2 = mx + range;
519
        if (x2 > xmax)
520
            x2 = xmax;
521

    
522
        y1 = my - range;
523
        if (y1 < ymin)
524
            y1 = ymin;
525

    
526
        y2 = my + range;
527
        if (y2 > ymax)
528
            y2 = ymax;
529

    
530
    } while (range >= 1);
531

    
532
#ifdef DEBUG
533
    av_log(s->avctx, AV_LOG_DEBUG, "phods     - MX: %d\tMY: %d\n", mx, my);
534
#endif
535

    
536
    /* half pixel search */
537
    *mx_ptr = mx;
538
    *my_ptr = my;
539
    return dminy;
540
}
541
#endif /* 0 */
542

    
543
#define Z_THRESHOLD 256
544

    
545
#define CHECK_SAD_HALF_MV(suffix, x, y) \
546
{\
547
    d= s->dsp.pix_abs[size][(x?1:0)+(y?2:0)](NULL, pix, ptr+((x)>>1), stride, h);\
548
    d += (mv_penalty[pen_x + x] + mv_penalty[pen_y + y])*penalty_factor;\
549
    COPY3_IF_LT(dminh, d, dx, x, dy, y)\
550
}
551

    
552
static inline int sad_hpel_motion_search(MpegEncContext * s,
553
                                  int *mx_ptr, int *my_ptr, int dmin,
554
                                  int src_index, int ref_index,
555
                                  int size, int h)
556
{
557
    MotionEstContext * const c= &s->me;
558
    const int penalty_factor= c->sub_penalty_factor;
559
    int mx, my, dminh;
560
    uint8_t *pix, *ptr;
561
    int stride= c->stride;
562
    const int flags= c->sub_flags;
563
    LOAD_COMMON
564

    
565
    assert(flags == 0);
566

    
567
    if(c->skip){
568
//    printf("S");
569
        *mx_ptr = 0;
570
        *my_ptr = 0;
571
        return dmin;
572
    }
573
//    printf("N");
574

    
575
    pix = c->src[src_index][0];
576

    
577
    mx = *mx_ptr;
578
    my = *my_ptr;
579
    ptr = c->ref[ref_index][0] + (my * stride) + mx;
580

    
581
    dminh = dmin;
582

    
583
    if (mx > xmin && mx < xmax &&
584
        my > ymin && my < ymax) {
585
        int dx=0, dy=0;
586
        int d, pen_x, pen_y;
587
        const int index= (my<<ME_MAP_SHIFT) + mx;
588
        const int t= score_map[(index-(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)];
589
        const int l= score_map[(index- 1               )&(ME_MAP_SIZE-1)];
590
        const int r= score_map[(index+ 1               )&(ME_MAP_SIZE-1)];
591
        const int b= score_map[(index+(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)];
592
        mx<<=1;
593
        my<<=1;
594

    
595

    
596
        pen_x= pred_x + mx;
597
        pen_y= pred_y + my;
598

    
599
        ptr-= stride;
600
        if(t<=b){
601
            CHECK_SAD_HALF_MV(y2 , 0, -1)
602
            if(l<=r){
603
                CHECK_SAD_HALF_MV(xy2, -1, -1)
604
                if(t+r<=b+l){
605
                    CHECK_SAD_HALF_MV(xy2, +1, -1)
606
                    ptr+= stride;
607
                }else{
608
                    ptr+= stride;
609
                    CHECK_SAD_HALF_MV(xy2, -1, +1)
610
                }
611
                CHECK_SAD_HALF_MV(x2 , -1,  0)
612
            }else{
613
                CHECK_SAD_HALF_MV(xy2, +1, -1)
614
                if(t+l<=b+r){
615
                    CHECK_SAD_HALF_MV(xy2, -1, -1)
616
                    ptr+= stride;
617
                }else{
618
                    ptr+= stride;
619
                    CHECK_SAD_HALF_MV(xy2, +1, +1)
620
                }
621
                CHECK_SAD_HALF_MV(x2 , +1,  0)
622
            }
623
        }else{
624
            if(l<=r){
625
                if(t+l<=b+r){
626
                    CHECK_SAD_HALF_MV(xy2, -1, -1)
627
                    ptr+= stride;
628
                }else{
629
                    ptr+= stride;
630
                    CHECK_SAD_HALF_MV(xy2, +1, +1)
631
                }
632
                CHECK_SAD_HALF_MV(x2 , -1,  0)
633
                CHECK_SAD_HALF_MV(xy2, -1, +1)
634
            }else{
635
                if(t+r<=b+l){
636
                    CHECK_SAD_HALF_MV(xy2, +1, -1)
637
                    ptr+= stride;
638
                }else{
639
                    ptr+= stride;
640
                    CHECK_SAD_HALF_MV(xy2, -1, +1)
641
                }
642
                CHECK_SAD_HALF_MV(x2 , +1,  0)
643
                CHECK_SAD_HALF_MV(xy2, +1, +1)
644
            }
645
            CHECK_SAD_HALF_MV(y2 ,  0, +1)
646
        }
647
        mx+=dx;
648
        my+=dy;
649

    
650
    }else{
651
        mx<<=1;
652
        my<<=1;
653
    }
654

    
655
    *mx_ptr = mx;
656
    *my_ptr = my;
657
    return dminh;
658
}
659

    
660
static inline void set_p_mv_tables(MpegEncContext * s, int mx, int my, int mv4)
661
{
662
    const int xy= s->mb_x + s->mb_y*s->mb_stride;
663

    
664
    s->p_mv_table[xy][0] = mx;
665
    s->p_mv_table[xy][1] = my;
666

    
667
    /* has already been set to the 4 MV if 4MV is done */
668
    if(mv4){
669
        int mot_xy= s->block_index[0];
670

    
671
        s->current_picture.motion_val[0][mot_xy  ][0]= mx;
672
        s->current_picture.motion_val[0][mot_xy  ][1]= my;
673
        s->current_picture.motion_val[0][mot_xy+1][0]= mx;
674
        s->current_picture.motion_val[0][mot_xy+1][1]= my;
675

    
676
        mot_xy += s->b8_stride;
677
        s->current_picture.motion_val[0][mot_xy  ][0]= mx;
678
        s->current_picture.motion_val[0][mot_xy  ][1]= my;
679
        s->current_picture.motion_val[0][mot_xy+1][0]= mx;
680
        s->current_picture.motion_val[0][mot_xy+1][1]= my;
681
    }
682
}
683

    
684
/**
685
 * get fullpel ME search limits.
686
 */
687
static inline void get_limits(MpegEncContext *s, int x, int y)
688
{
689
    MotionEstContext * const c= &s->me;
690
/*
691
    if(c->avctx->me_range) c->range= c->avctx->me_range >> 1;
692
    else                   c->range= 16;
693
*/
694
    if (s->unrestricted_mv) {
695
        c->xmin = - x - 16;
696
        c->ymin = - y - 16;
697
        c->xmax = - x + s->mb_width *16;
698
        c->ymax = - y + s->mb_height*16;
699
    } else if (s->out_format == FMT_H261){
700
        // Search range of H261 is different from other codec standards
701
        c->xmin = (x > 15) ? - 15 : 0;
702
        c->ymin = (y > 15) ? - 15 : 0;
703
        c->xmax = (x < s->mb_width * 16 - 16) ? 15 : 0;
704
        c->ymax = (y < s->mb_height * 16 - 16) ? 15 : 0;
705
    } else {
706
        c->xmin = - x;
707
        c->ymin = - y;
708
        c->xmax = - x + s->mb_width *16 - 16;
709
        c->ymax = - y + s->mb_height*16 - 16;
710
    }
711
}
712

    
713
static inline void init_mv4_ref(MotionEstContext *c){
714
    const int stride= c->stride;
715

    
716
    c->ref[1][0] = c->ref[0][0] + 8;
717
    c->ref[2][0] = c->ref[0][0] + 8*stride;
718
    c->ref[3][0] = c->ref[2][0] + 8;
719
    c->src[1][0] = c->src[0][0] + 8;
720
    c->src[2][0] = c->src[0][0] + 8*stride;
721
    c->src[3][0] = c->src[2][0] + 8;
722
}
723

    
724
static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift)
725
{
726
    MotionEstContext * const c= &s->me;
727
    const int size= 1;
728
    const int h=8;
729
    int block;
730
    int P[10][2];
731
    int dmin_sum=0, mx4_sum=0, my4_sum=0;
732
    int same=1;
733
    const int stride= c->stride;
734
    uint8_t *mv_penalty= c->current_mv_penalty;
735

    
736
    init_mv4_ref(c);
737

    
738
    for(block=0; block<4; block++){
739
        int mx4, my4;
740
        int pred_x4, pred_y4;
741
        int dmin4;
742
        static const int off[4]= {2, 1, 1, -1};
743
        const int mot_stride = s->b8_stride;
744
        const int mot_xy = s->block_index[block];
745

    
746
        P_LEFT[0] = s->current_picture.motion_val[0][mot_xy - 1][0];
747
        P_LEFT[1] = s->current_picture.motion_val[0][mot_xy - 1][1];
748

    
749
        if(P_LEFT[0]       > (c->xmax<<shift)) P_LEFT[0]       = (c->xmax<<shift);
750

    
751
        /* special case for first line */
752
        if (s->first_slice_line && block<2) {
753
            c->pred_x= pred_x4= P_LEFT[0];
754
            c->pred_y= pred_y4= P_LEFT[1];
755
        } else {
756
            P_TOP[0]      = s->current_picture.motion_val[0][mot_xy - mot_stride             ][0];
757
            P_TOP[1]      = s->current_picture.motion_val[0][mot_xy - mot_stride             ][1];
758
            P_TOPRIGHT[0] = s->current_picture.motion_val[0][mot_xy - mot_stride + off[block]][0];
759
            P_TOPRIGHT[1] = s->current_picture.motion_val[0][mot_xy - mot_stride + off[block]][1];
760
            if(P_TOP[1]      > (c->ymax<<shift)) P_TOP[1]     = (c->ymax<<shift);
761
            if(P_TOPRIGHT[0] < (c->xmin<<shift)) P_TOPRIGHT[0]= (c->xmin<<shift);
762
            if(P_TOPRIGHT[0] > (c->xmax<<shift)) P_TOPRIGHT[0]= (c->xmax<<shift);
763
            if(P_TOPRIGHT[1] > (c->ymax<<shift)) P_TOPRIGHT[1]= (c->ymax<<shift);
764

    
765
            P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
766
            P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
767

    
768
            c->pred_x= pred_x4 = P_MEDIAN[0];
769
            c->pred_y= pred_y4 = P_MEDIAN[1];
770
        }
771
        P_MV1[0]= mx;
772
        P_MV1[1]= my;
773

    
774
        dmin4 = epzs_motion_search4(s, &mx4, &my4, P, block, block, s->p_mv_table, (1<<16)>>shift);
775

    
776
        dmin4= c->sub_motion_search(s, &mx4, &my4, dmin4, block, block, size, h);
777

    
778
        if(s->dsp.me_sub_cmp[0] != s->dsp.mb_cmp[0]){
779
            int dxy;
780
            const int offset= ((block&1) + (block>>1)*stride)*8;
781
            uint8_t *dest_y = c->scratchpad + offset;
782
            if(s->quarter_sample){
783
                uint8_t *ref= c->ref[block][0] + (mx4>>2) + (my4>>2)*stride;
784
                dxy = ((my4 & 3) << 2) | (mx4 & 3);
785

    
786
                if(s->no_rounding)
787
                    s->dsp.put_no_rnd_qpel_pixels_tab[1][dxy](dest_y   , ref    , stride);
788
                else
789
                    s->dsp.put_qpel_pixels_tab       [1][dxy](dest_y   , ref    , stride);
790
            }else{
791
                uint8_t *ref= c->ref[block][0] + (mx4>>1) + (my4>>1)*stride;
792
                dxy = ((my4 & 1) << 1) | (mx4 & 1);
793

    
794
                if(s->no_rounding)
795
                    s->dsp.put_no_rnd_pixels_tab[1][dxy](dest_y    , ref    , stride, h);
796
                else
797
                    s->dsp.put_pixels_tab       [1][dxy](dest_y    , ref    , stride, h);
798
            }
799
            dmin_sum+= (mv_penalty[mx4-pred_x4] + mv_penalty[my4-pred_y4])*c->mb_penalty_factor;
800
        }else
801
            dmin_sum+= dmin4;
802

    
803
        if(s->quarter_sample){
804
            mx4_sum+= mx4/2;
805
            my4_sum+= my4/2;
806
        }else{
807
            mx4_sum+= mx4;
808
            my4_sum+= my4;
809
        }
810

    
811
        s->current_picture.motion_val[0][ s->block_index[block] ][0]= mx4;
812
        s->current_picture.motion_val[0][ s->block_index[block] ][1]= my4;
813

    
814
        if(mx4 != mx || my4 != my) same=0;
815
    }
816

    
817
    if(same)
818
        return INT_MAX;
819

    
820
    if(s->dsp.me_sub_cmp[0] != s->dsp.mb_cmp[0]){
821
        dmin_sum += s->dsp.mb_cmp[0](s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*16*stride, c->scratchpad, stride, 16);
822
    }
823

    
824
    if(c->avctx->mb_cmp&FF_CMP_CHROMA){
825
        int dxy;
826
        int mx, my;
827
        int offset;
828

    
829
        mx= ff_h263_round_chroma(mx4_sum);
830
        my= ff_h263_round_chroma(my4_sum);
831
        dxy = ((my & 1) << 1) | (mx & 1);
832

    
833
        offset= (s->mb_x*8 + (mx>>1)) + (s->mb_y*8 + (my>>1))*s->uvlinesize;
834

    
835
        if(s->no_rounding){
836
            s->dsp.put_no_rnd_pixels_tab[1][dxy](c->scratchpad    , s->last_picture.data[1] + offset, s->uvlinesize, 8);
837
            s->dsp.put_no_rnd_pixels_tab[1][dxy](c->scratchpad+8  , s->last_picture.data[2] + offset, s->uvlinesize, 8);
838
        }else{
839
            s->dsp.put_pixels_tab       [1][dxy](c->scratchpad    , s->last_picture.data[1] + offset, s->uvlinesize, 8);
840
            s->dsp.put_pixels_tab       [1][dxy](c->scratchpad+8  , s->last_picture.data[2] + offset, s->uvlinesize, 8);
841
        }
842

    
843
        dmin_sum += s->dsp.mb_cmp[1](s, s->new_picture.data[1] + s->mb_x*8 + s->mb_y*8*s->uvlinesize, c->scratchpad  , s->uvlinesize, 8);
844
        dmin_sum += s->dsp.mb_cmp[1](s, s->new_picture.data[2] + s->mb_x*8 + s->mb_y*8*s->uvlinesize, c->scratchpad+8, s->uvlinesize, 8);
845
    }
846

    
847
    c->pred_x= mx;
848
    c->pred_y= my;
849

    
850
    switch(c->avctx->mb_cmp&0xFF){
851
    /*case FF_CMP_SSE:
852
        return dmin_sum+ 32*s->qscale*s->qscale;*/
853
    case FF_CMP_RD:
854
        return dmin_sum;
855
    default:
856
        return dmin_sum+ 11*c->mb_penalty_factor;
857
    }
858
}
859

    
860
static inline void init_interlaced_ref(MpegEncContext *s, int ref_index){
861
    MotionEstContext * const c= &s->me;
862

    
863
    c->ref[1+ref_index][0] = c->ref[0+ref_index][0] + s->linesize;
864
    c->src[1][0] = c->src[0][0] + s->linesize;
865
    if(c->flags & FLAG_CHROMA){
866
        c->ref[1+ref_index][1] = c->ref[0+ref_index][1] + s->uvlinesize;
867
        c->ref[1+ref_index][2] = c->ref[0+ref_index][2] + s->uvlinesize;
868
        c->src[1][1] = c->src[0][1] + s->uvlinesize;
869
        c->src[1][2] = c->src[0][2] + s->uvlinesize;
870
    }
871
}
872

    
873
static int interlaced_search(MpegEncContext *s, int ref_index,
874
                             int16_t (*mv_tables[2][2])[2], uint8_t *field_select_tables[2], int mx, int my, int user_field_select)
875
{
876
    MotionEstContext * const c= &s->me;
877
    const int size=0;
878
    const int h=8;
879
    int block;
880
    int P[10][2];
881
    uint8_t * const mv_penalty= c->current_mv_penalty;
882
    int same=1;
883
    const int stride= 2*s->linesize;
884
    int dmin_sum= 0;
885
    const int mot_stride= s->mb_stride;
886
    const int xy= s->mb_x + s->mb_y*mot_stride;
887

    
888
    c->ymin>>=1;
889
    c->ymax>>=1;
890
    c->stride<<=1;
891
    c->uvstride<<=1;
892
    init_interlaced_ref(s, ref_index);
893

    
894
    for(block=0; block<2; block++){
895
        int field_select;
896
        int best_dmin= INT_MAX;
897
        int best_field= -1;
898

    
899
        for(field_select=0; field_select<2; field_select++){
900
            int dmin, mx_i, my_i;
901
            int16_t (*mv_table)[2]= mv_tables[block][field_select];
902

    
903
            if(user_field_select){
904
                if(field_select_tables[block][xy] != field_select)
905
                    continue;
906
            }
907

    
908
            P_LEFT[0] = mv_table[xy - 1][0];
909
            P_LEFT[1] = mv_table[xy - 1][1];
910
            if(P_LEFT[0]       > (c->xmax<<1)) P_LEFT[0]       = (c->xmax<<1);
911

    
912
            c->pred_x= P_LEFT[0];
913
            c->pred_y= P_LEFT[1];
914

    
915
            if(!s->first_slice_line){
916
                P_TOP[0]      = mv_table[xy - mot_stride][0];
917
                P_TOP[1]      = mv_table[xy - mot_stride][1];
918
                P_TOPRIGHT[0] = mv_table[xy - mot_stride + 1][0];
919
                P_TOPRIGHT[1] = mv_table[xy - mot_stride + 1][1];
920
                if(P_TOP[1]      > (c->ymax<<1)) P_TOP[1]     = (c->ymax<<1);
921
                if(P_TOPRIGHT[0] < (c->xmin<<1)) P_TOPRIGHT[0]= (c->xmin<<1);
922
                if(P_TOPRIGHT[0] > (c->xmax<<1)) P_TOPRIGHT[0]= (c->xmax<<1);
923
                if(P_TOPRIGHT[1] > (c->ymax<<1)) P_TOPRIGHT[1]= (c->ymax<<1);
924

    
925
                P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
926
                P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
927
            }
928
            P_MV1[0]= mx; //FIXME not correct if block != field_select
929
            P_MV1[1]= my / 2;
930

    
931
            dmin = epzs_motion_search2(s, &mx_i, &my_i, P, block, field_select+ref_index, mv_table, (1<<16)>>1);
932

    
933
            dmin= c->sub_motion_search(s, &mx_i, &my_i, dmin, block, field_select+ref_index, size, h);
934

    
935
            mv_table[xy][0]= mx_i;
936
            mv_table[xy][1]= my_i;
937

    
938
            if(s->dsp.me_sub_cmp[0] != s->dsp.mb_cmp[0]){
939
                int dxy;
940

    
941
                //FIXME chroma ME
942
                uint8_t *ref= c->ref[field_select+ref_index][0] + (mx_i>>1) + (my_i>>1)*stride;
943
                dxy = ((my_i & 1) << 1) | (mx_i & 1);
944

    
945
                if(s->no_rounding){
946
                    s->dsp.put_no_rnd_pixels_tab[size][dxy](c->scratchpad, ref    , stride, h);
947
                }else{
948
                    s->dsp.put_pixels_tab       [size][dxy](c->scratchpad, ref    , stride, h);
949
                }
950
                dmin= s->dsp.mb_cmp[size](s, c->src[block][0], c->scratchpad, stride, h);
951
                dmin+= (mv_penalty[mx_i-c->pred_x] + mv_penalty[my_i-c->pred_y] + 1)*c->mb_penalty_factor;
952
            }else
953
                dmin+= c->mb_penalty_factor; //field_select bits
954

    
955
            dmin += field_select != block; //slightly prefer same field
956

    
957
            if(dmin < best_dmin){
958
                best_dmin= dmin;
959
                best_field= field_select;
960
            }
961
        }
962
        {
963
            int16_t (*mv_table)[2]= mv_tables[block][best_field];
964

    
965
            if(mv_table[xy][0] != mx) same=0; //FIXME check if these checks work and are any good at all
966
            if(mv_table[xy][1]&1) same=0;
967
            if(mv_table[xy][1]*2 != my) same=0;
968
            if(best_field != block) same=0;
969
        }
970

    
971
        field_select_tables[block][xy]= best_field;
972
        dmin_sum += best_dmin;
973
    }
974

    
975
    c->ymin<<=1;
976
    c->ymax<<=1;
977
    c->stride>>=1;
978
    c->uvstride>>=1;
979

    
980
    if(same)
981
        return INT_MAX;
982

    
983
    switch(c->avctx->mb_cmp&0xFF){
984
    /*case FF_CMP_SSE:
985
        return dmin_sum+ 32*s->qscale*s->qscale;*/
986
    case FF_CMP_RD:
987
        return dmin_sum;
988
    default:
989
        return dmin_sum+ 11*c->mb_penalty_factor;
990
    }
991
}
992

    
993
static void clip_input_mv(MpegEncContext * s, int16_t *mv, int interlaced){
994
    int ymax= s->me.ymax>>interlaced;
995
    int ymin= s->me.ymin>>interlaced;
996

    
997
    if(mv[0] < s->me.xmin) mv[0] = s->me.xmin;
998
    if(mv[0] > s->me.xmax) mv[0] = s->me.xmax;
999
    if(mv[1] <       ymin) mv[1] =       ymin;
1000
    if(mv[1] >       ymax) mv[1] =       ymax;
1001
}
1002

    
1003
static inline int check_input_motion(MpegEncContext * s, int mb_x, int mb_y, int p_type){
1004
    MotionEstContext * const c= &s->me;
1005
    Picture *p= s->current_picture_ptr;
1006
    int mb_xy= mb_x + mb_y*s->mb_stride;
1007
    int xy= 2*mb_x + 2*mb_y*s->b8_stride;
1008
    int mb_type= s->current_picture.mb_type[mb_xy];
1009
    int flags= c->flags;
1010
    int shift= (flags&FLAG_QPEL) + 1;
1011
    int mask= (1<<shift)-1;
1012
    int x, y, i;
1013
    int d=0;
1014
    me_cmp_func cmpf= s->dsp.sse[0];
1015
    me_cmp_func chroma_cmpf= s->dsp.sse[1];
1016

    
1017
    if(p_type && USES_LIST(mb_type, 1)){
1018
        av_log(c->avctx, AV_LOG_ERROR, "backward motion vector in P frame\n");
1019
        return INT_MAX/2;
1020
    }
1021
    assert(IS_INTRA(mb_type) || USES_LIST(mb_type,0) || USES_LIST(mb_type,1));
1022

    
1023
    for(i=0; i<4; i++){
1024
        int xy= s->block_index[i];
1025
        clip_input_mv(s, p->motion_val[0][xy], !!IS_INTERLACED(mb_type));
1026
        clip_input_mv(s, p->motion_val[1][xy], !!IS_INTERLACED(mb_type));
1027
    }
1028

    
1029
    if(IS_INTERLACED(mb_type)){
1030
        int xy2= xy  + s->b8_stride;
1031
        s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_INTRA;
1032
        c->stride<<=1;
1033
        c->uvstride<<=1;
1034

    
1035
        if(!(s->flags & CODEC_FLAG_INTERLACED_ME)){
1036
            av_log(c->avctx, AV_LOG_ERROR, "Interlaced macroblock selected but interlaced motion estimation disabled\n");
1037
            return INT_MAX/2;
1038
        }
1039

    
1040
        if(USES_LIST(mb_type, 0)){
1041
            int field_select0= p->ref_index[0][xy ];
1042
            int field_select1= p->ref_index[0][xy2];
1043
            assert(field_select0==0 ||field_select0==1);
1044
            assert(field_select1==0 ||field_select1==1);
1045
            init_interlaced_ref(s, 0);
1046

    
1047
            if(p_type){
1048
                s->p_field_select_table[0][mb_xy]= field_select0;
1049
                s->p_field_select_table[1][mb_xy]= field_select1;
1050
                *(uint32_t*)s->p_field_mv_table[0][field_select0][mb_xy]= *(uint32_t*)p->motion_val[0][xy ];
1051
                *(uint32_t*)s->p_field_mv_table[1][field_select1][mb_xy]= *(uint32_t*)p->motion_val[0][xy2];
1052
                s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_INTER_I;
1053
            }else{
1054
                s->b_field_select_table[0][0][mb_xy]= field_select0;
1055
                s->b_field_select_table[0][1][mb_xy]= field_select1;
1056
                *(uint32_t*)s->b_field_mv_table[0][0][field_select0][mb_xy]= *(uint32_t*)p->motion_val[0][xy ];
1057
                *(uint32_t*)s->b_field_mv_table[0][1][field_select1][mb_xy]= *(uint32_t*)p->motion_val[0][xy2];
1058
                s->mb_type[mb_xy]= CANDIDATE_MB_TYPE_FORWARD_I;
1059
            }
1060

    
1061
            x= p->motion_val[0][xy ][0];
1062
            y= p->motion_val[0][xy ][1];
1063
            d = cmp(s, x>>shift, y>>shift, x&mask, y&mask, 0, 8, field_select0, 0, cmpf, chroma_cmpf, flags);
1064
            x= p->motion_val[0][xy2][0];
1065
            y= p->motion_val[0][xy2][1];
1066
            d+= cmp(s, x>>shift, y>>shift, x&mask, y&mask, 0, 8, field_select1, 1, cmpf, chroma_cmpf, flags);
1067
        }
1068
        if(USES_LIST(mb_type, 1)){
1069
            int field_select0= p->ref_index[1][xy ];
1070
            int field_select1= p->ref_index[1][xy2];
1071
            assert(field_select0==0 ||field_select0==1);
1072
            assert(field_select1==0 ||field_select1==1);
1073
            init_interlaced_ref(s, 2);
1074

    
1075
            s->b_field_select_table[1][0][mb_xy]= field_select0;
1076
            s->b_field_select_table[1][1][mb_xy]= field_select1;
1077
            *(uint32_t*)s->b_field_mv_table[1][0][field_select0][mb_xy]= *(uint32_t*)p->motion_val[1][xy ];
1078
            *(uint32_t*)s->b_field_mv_table[1][1][field_select1][mb_xy]= *(uint32_t*)p->motion_val[1][xy2];
1079
            if(USES_LIST(mb_type, 0)){
1080
                s->mb_type[mb_xy]= CANDIDATE_MB_TYPE_BIDIR_I;
1081
            }else{
1082
                s->mb_type[mb_xy]= CANDIDATE_MB_TYPE_BACKWARD_I;
1083
            }
1084

    
1085
            x= p->motion_val[1][xy ][0];
1086
            y= p->motion_val[1][xy ][1];
1087
            d = cmp(s, x>>shift, y>>shift, x&mask, y&mask, 0, 8, field_select0+2, 0, cmpf, chroma_cmpf, flags);
1088
            x= p->motion_val[1][xy2][0];
1089
            y= p->motion_val[1][xy2][1];
1090
            d+= cmp(s, x>>shift, y>>shift, x&mask, y&mask, 0, 8, field_select1+2, 1, cmpf, chroma_cmpf, flags);
1091
            //FIXME bidir scores
1092
        }
1093
        c->stride>>=1;
1094
        c->uvstride>>=1;
1095
    }else if(IS_8X8(mb_type)){
1096
        if(!(s->flags & CODEC_FLAG_4MV)){
1097
            av_log(c->avctx, AV_LOG_ERROR, "4MV macroblock selected but 4MV encoding disabled\n");
1098
            return INT_MAX/2;
1099
        }
1100
        cmpf= s->dsp.sse[1];
1101
        chroma_cmpf= s->dsp.sse[1];
1102
        init_mv4_ref(c);
1103
        for(i=0; i<4; i++){
1104
            xy= s->block_index[i];
1105
            x= p->motion_val[0][xy][0];
1106
            y= p->motion_val[0][xy][1];
1107
            d+= cmp(s, x>>shift, y>>shift, x&mask, y&mask, 1, 8, i, i, cmpf, chroma_cmpf, flags);
1108
        }
1109
        s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_INTER4V;
1110
    }else{
1111
        if(USES_LIST(mb_type, 0)){
1112
            if(p_type){
1113
                *(uint32_t*)s->p_mv_table[mb_xy]= *(uint32_t*)p->motion_val[0][xy];
1114
                s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_INTER;
1115
            }else if(USES_LIST(mb_type, 1)){
1116
                *(uint32_t*)s->b_bidir_forw_mv_table[mb_xy]= *(uint32_t*)p->motion_val[0][xy];
1117
                *(uint32_t*)s->b_bidir_back_mv_table[mb_xy]= *(uint32_t*)p->motion_val[1][xy];
1118
                s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_BIDIR;
1119
            }else{
1120
                *(uint32_t*)s->b_forw_mv_table[mb_xy]= *(uint32_t*)p->motion_val[0][xy];
1121
                s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_FORWARD;
1122
            }
1123
            x= p->motion_val[0][xy][0];
1124
            y= p->motion_val[0][xy][1];
1125
            d = cmp(s, x>>shift, y>>shift, x&mask, y&mask, 0, 16, 0, 0, cmpf, chroma_cmpf, flags);
1126
        }else if(USES_LIST(mb_type, 1)){
1127
            *(uint32_t*)s->b_back_mv_table[mb_xy]= *(uint32_t*)p->motion_val[1][xy];
1128
            s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_BACKWARD;
1129

    
1130
            x= p->motion_val[1][xy][0];
1131
            y= p->motion_val[1][xy][1];
1132
            d = cmp(s, x>>shift, y>>shift, x&mask, y&mask, 0, 16, 2, 0, cmpf, chroma_cmpf, flags);
1133
        }else
1134
            s->mb_type[mb_xy]=CANDIDATE_MB_TYPE_INTRA;
1135
    }
1136
    return d;
1137
}
1138

    
1139
void ff_estimate_p_frame_motion(MpegEncContext * s,
1140
                                int mb_x, int mb_y)
1141
{
1142
    MotionEstContext * const c= &s->me;
1143
    uint8_t *pix, *ppix;
1144
    int sum, varc, vard, mx, my, dmin;
1145
    int P[10][2];
1146
    const int shift= 1+s->quarter_sample;
1147
    int mb_type=0;
1148
    Picture * const pic= &s->current_picture;
1149

    
1150
    init_ref(c, s->new_picture.data, s->last_picture.data, NULL, 16*mb_x, 16*mb_y, 0);
1151

    
1152
    assert(s->quarter_sample==0 || s->quarter_sample==1);
1153
    assert(s->linesize == c->stride);
1154
    assert(s->uvlinesize == c->uvstride);
1155

    
1156
    c->penalty_factor    = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_cmp);
1157
    c->sub_penalty_factor= get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_sub_cmp);
1158
    c->mb_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->mb_cmp);
1159
    c->current_mv_penalty= c->mv_penalty[s->f_code] + MAX_MV;
1160

    
1161
    get_limits(s, 16*mb_x, 16*mb_y);
1162
    c->skip=0;
1163

    
1164
    /* intra / predictive decision */
1165
    pix = c->src[0][0];
1166
    sum = s->dsp.pix_sum(pix, s->linesize);
1167
    varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500 + 128)>>8;
1168

    
1169
    pic->mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
1170
    pic->mb_var [s->mb_stride * mb_y + mb_x] = varc;
1171
    c->mb_var_sum_temp += varc;
1172

    
1173
    if(c->avctx->me_threshold){
1174
        vard= (check_input_motion(s, mb_x, mb_y, 1)+128)>>8;
1175

    
1176
        if(vard<c->avctx->me_threshold){
1177
            pic->mc_mb_var[s->mb_stride * mb_y + mb_x] = vard;
1178
            c->mc_mb_var_sum_temp += vard;
1179
            if (vard <= 64 || vard < varc) { //FIXME
1180
                c->scene_change_score+= ff_sqrt(vard<<8) - ff_sqrt(varc<<8);
1181
            }else{
1182
                c->scene_change_score+= s->qscale;
1183
            }
1184
            return;
1185
        }
1186
        if(vard<c->avctx->mb_threshold)
1187
            mb_type= s->mb_type[mb_x + mb_y*s->mb_stride];
1188
    }
1189

    
1190
    switch(s->me_method) {
1191
    case ME_ZERO:
1192
    default:
1193
        no_motion_search(s, &mx, &my);
1194
        mx-= mb_x*16;
1195
        my-= mb_y*16;
1196
        dmin = 0;
1197
        break;
1198
#if 0
1199
    case ME_FULL:
1200
        dmin = full_motion_search(s, &mx, &my, range, ref_picture);
1201
        mx-= mb_x*16;
1202
        my-= mb_y*16;
1203
        break;
1204
    case ME_LOG:
1205
        dmin = log_motion_search(s, &mx, &my, range / 2, ref_picture);
1206
        mx-= mb_x*16;
1207
        my-= mb_y*16;
1208
        break;
1209
    case ME_PHODS:
1210
        dmin = phods_motion_search(s, &mx, &my, range / 2, ref_picture);
1211
        mx-= mb_x*16;
1212
        my-= mb_y*16;
1213
        break;
1214
#endif
1215
    case ME_X1:
1216
    case ME_EPZS:
1217
       {
1218
            const int mot_stride = s->b8_stride;
1219
            const int mot_xy = s->block_index[0];
1220

    
1221
            P_LEFT[0]       = s->current_picture.motion_val[0][mot_xy - 1][0];
1222
            P_LEFT[1]       = s->current_picture.motion_val[0][mot_xy - 1][1];
1223

    
1224
            if(P_LEFT[0]       > (c->xmax<<shift)) P_LEFT[0]       = (c->xmax<<shift);
1225

    
1226
            if(!s->first_slice_line) {
1227
                P_TOP[0]      = s->current_picture.motion_val[0][mot_xy - mot_stride    ][0];
1228
                P_TOP[1]      = s->current_picture.motion_val[0][mot_xy - mot_stride    ][1];
1229
                P_TOPRIGHT[0] = s->current_picture.motion_val[0][mot_xy - mot_stride + 2][0];
1230
                P_TOPRIGHT[1] = s->current_picture.motion_val[0][mot_xy - mot_stride + 2][1];
1231
                if(P_TOP[1]      > (c->ymax<<shift)) P_TOP[1]     = (c->ymax<<shift);
1232
                if(P_TOPRIGHT[0] < (c->xmin<<shift)) P_TOPRIGHT[0]= (c->xmin<<shift);
1233
                if(P_TOPRIGHT[1] > (c->ymax<<shift)) P_TOPRIGHT[1]= (c->ymax<<shift);
1234

    
1235
                P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
1236
                P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
1237

    
1238
                if(s->out_format == FMT_H263){
1239
                    c->pred_x = P_MEDIAN[0];
1240
                    c->pred_y = P_MEDIAN[1];
1241
                }else { /* mpeg1 at least */
1242
                    c->pred_x= P_LEFT[0];
1243
                    c->pred_y= P_LEFT[1];
1244
                }
1245
            }else{
1246
                c->pred_x= P_LEFT[0];
1247
                c->pred_y= P_LEFT[1];
1248
            }
1249

    
1250
        }
1251
        dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, 0, s->p_mv_table, (1<<16)>>shift, 0, 16);
1252

    
1253
        break;
1254
    }
1255

    
1256
    /* At this point (mx,my) are full-pell and the relative displacement */
1257
    ppix = c->ref[0][0] + (my * s->linesize) + mx;
1258

    
1259
    vard = (s->dsp.sse[0](NULL, pix, ppix, s->linesize, 16)+128)>>8;
1260

    
1261
    pic->mc_mb_var[s->mb_stride * mb_y + mb_x] = vard;
1262
//    pic->mb_cmp_score[s->mb_stride * mb_y + mb_x] = dmin;
1263
    c->mc_mb_var_sum_temp += vard;
1264

    
1265
#if 0
1266
    printf("varc=%4d avg_var=%4d (sum=%4d) vard=%4d mx=%2d my=%2d\n",
1267
           varc, s->avg_mb_var, sum, vard, mx - xx, my - yy);
1268
#endif
1269
    if(mb_type){
1270
        if (vard <= 64 || vard < varc)
1271
            c->scene_change_score+= ff_sqrt(vard<<8) - ff_sqrt(varc<<8);
1272
        else
1273
            c->scene_change_score+= s->qscale;
1274

    
1275
        if(mb_type == CANDIDATE_MB_TYPE_INTER){
1276
            c->sub_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
1277
            set_p_mv_tables(s, mx, my, 1);
1278
        }else{
1279
            mx <<=shift;
1280
            my <<=shift;
1281
        }
1282
        if(mb_type == CANDIDATE_MB_TYPE_INTER4V){
1283
            h263_mv4_search(s, mx, my, shift);
1284

    
1285
            set_p_mv_tables(s, mx, my, 0);
1286
        }
1287
        if(mb_type == CANDIDATE_MB_TYPE_INTER_I){
1288
            interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my, 1);
1289
        }
1290
    }else if(c->avctx->mb_decision > FF_MB_DECISION_SIMPLE){
1291
        if (vard <= 64 || vard < varc)
1292
            c->scene_change_score+= ff_sqrt(vard<<8) - ff_sqrt(varc<<8);
1293
        else
1294
            c->scene_change_score+= s->qscale;
1295

    
1296
        if (vard*2 + 200 > varc)
1297
            mb_type|= CANDIDATE_MB_TYPE_INTRA;
1298
        if (varc*2 + 200 > vard){
1299
            mb_type|= CANDIDATE_MB_TYPE_INTER;
1300
            c->sub_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
1301
            if(s->flags&CODEC_FLAG_MV0)
1302
                if(mx || my)
1303
                    mb_type |= CANDIDATE_MB_TYPE_SKIPPED; //FIXME check difference
1304
        }else{
1305
            mx <<=shift;
1306
            my <<=shift;
1307
        }
1308
        if((s->flags&CODEC_FLAG_4MV)
1309
           && !c->skip && varc>50 && vard>10){
1310
            if(h263_mv4_search(s, mx, my, shift) < INT_MAX)
1311
                mb_type|=CANDIDATE_MB_TYPE_INTER4V;
1312

    
1313
            set_p_mv_tables(s, mx, my, 0);
1314
        }else
1315
            set_p_mv_tables(s, mx, my, 1);
1316
        if((s->flags&CODEC_FLAG_INTERLACED_ME)
1317
           && !c->skip){ //FIXME varc/d checks
1318
            if(interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my, 0) < INT_MAX)
1319
                mb_type |= CANDIDATE_MB_TYPE_INTER_I;
1320
        }
1321
    }else{
1322
        int intra_score, i;
1323
        mb_type= CANDIDATE_MB_TYPE_INTER;
1324

    
1325
        dmin= c->sub_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
1326
        if(c->avctx->me_sub_cmp != c->avctx->mb_cmp && !c->skip)
1327
            dmin= ff_get_mb_score(s, mx, my, 0, 0, 0, 16, 1);
1328

    
1329
        if((s->flags&CODEC_FLAG_4MV)
1330
           && !c->skip && varc>50 && vard>10){
1331
            int dmin4= h263_mv4_search(s, mx, my, shift);
1332
            if(dmin4 < dmin){
1333
                mb_type= CANDIDATE_MB_TYPE_INTER4V;
1334
                dmin=dmin4;
1335
            }
1336
        }
1337
        if((s->flags&CODEC_FLAG_INTERLACED_ME)
1338
           && !c->skip){ //FIXME varc/d checks
1339
            int dmin_i= interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my, 0);
1340
            if(dmin_i < dmin){
1341
                mb_type = CANDIDATE_MB_TYPE_INTER_I;
1342
                dmin= dmin_i;
1343
            }
1344
        }
1345

    
1346
//        pic->mb_cmp_score[s->mb_stride * mb_y + mb_x] = dmin;
1347
        set_p_mv_tables(s, mx, my, mb_type!=CANDIDATE_MB_TYPE_INTER4V);
1348

    
1349
        /* get intra luma score */
1350
        if((c->avctx->mb_cmp&0xFF)==FF_CMP_SSE){
1351
            intra_score= (varc<<8) - 500; //FIXME dont scale it down so we dont have to fix it
1352
        }else{
1353
            int mean= (sum+128)>>8;
1354
            mean*= 0x01010101;
1355

    
1356
            for(i=0; i<16; i++){
1357
                *(uint32_t*)(&c->scratchpad[i*s->linesize+ 0]) = mean;
1358
                *(uint32_t*)(&c->scratchpad[i*s->linesize+ 4]) = mean;
1359
                *(uint32_t*)(&c->scratchpad[i*s->linesize+ 8]) = mean;
1360
                *(uint32_t*)(&c->scratchpad[i*s->linesize+12]) = mean;
1361
            }
1362

    
1363
            intra_score= s->dsp.mb_cmp[0](s, c->scratchpad, pix, s->linesize, 16);
1364
        }
1365
#if 0 //FIXME
1366
        /* get chroma score */
1367
        if(c->avctx->mb_cmp&FF_CMP_CHROMA){
1368
            for(i=1; i<3; i++){
1369
                uint8_t *dest_c;
1370
                int mean;
1371

1372
                if(s->out_format == FMT_H263){
1373
                    mean= (s->dc_val[i][mb_x + mb_y*s->b8_stride] + 4)>>3; //FIXME not exact but simple ;)
1374
                }else{
1375
                    mean= (s->last_dc[i] + 4)>>3;
1376
                }
1377
                dest_c = s->new_picture.data[i] + (mb_y * 8  * (s->uvlinesize)) + mb_x * 8;
1378

1379
                mean*= 0x01010101;
1380
                for(i=0; i<8; i++){
1381
                    *(uint32_t*)(&c->scratchpad[i*s->uvlinesize+ 0]) = mean;
1382
                    *(uint32_t*)(&c->scratchpad[i*s->uvlinesize+ 4]) = mean;
1383
                }
1384

1385
                intra_score+= s->dsp.mb_cmp[1](s, c->scratchpad, dest_c, s->uvlinesize);
1386
            }
1387
        }
1388
#endif
1389
        intra_score += c->mb_penalty_factor*16;
1390

    
1391
        if(intra_score < dmin){
1392
            mb_type= CANDIDATE_MB_TYPE_INTRA;
1393
            s->current_picture.mb_type[mb_y*s->mb_stride + mb_x]= CANDIDATE_MB_TYPE_INTRA; //FIXME cleanup
1394
        }else
1395
            s->current_picture.mb_type[mb_y*s->mb_stride + mb_x]= 0;
1396

    
1397
        if (vard <= 64 || vard < varc) { //FIXME
1398
            c->scene_change_score+= ff_sqrt(vard<<8) - ff_sqrt(varc<<8);
1399
        }else{
1400
            c->scene_change_score+= s->qscale;
1401
        }
1402
    }
1403

    
1404
    s->mb_type[mb_y*s->mb_stride + mb_x]= mb_type;
1405
}
1406

    
1407
int ff_pre_estimate_p_frame_motion(MpegEncContext * s,
1408
                                    int mb_x, int mb_y)
1409
{
1410
    MotionEstContext * const c= &s->me;
1411
    int mx, my, dmin;
1412
    int P[10][2];
1413
    const int shift= 1+s->quarter_sample;
1414
    const int xy= mb_x + mb_y*s->mb_stride;
1415
    init_ref(c, s->new_picture.data, s->last_picture.data, NULL, 16*mb_x, 16*mb_y, 0);
1416

    
1417
    assert(s->quarter_sample==0 || s->quarter_sample==1);
1418

    
1419
    c->pre_penalty_factor    = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_pre_cmp);
1420
    c->current_mv_penalty= c->mv_penalty[s->f_code] + MAX_MV;
1421

    
1422
    get_limits(s, 16*mb_x, 16*mb_y);
1423
    c->skip=0;
1424

    
1425
    P_LEFT[0]       = s->p_mv_table[xy + 1][0];
1426
    P_LEFT[1]       = s->p_mv_table[xy + 1][1];
1427

    
1428
    if(P_LEFT[0]       < (c->xmin<<shift)) P_LEFT[0]       = (c->xmin<<shift);
1429

    
1430
    /* special case for first line */
1431
    if (s->first_slice_line) {
1432
        c->pred_x= P_LEFT[0];
1433
        c->pred_y= P_LEFT[1];
1434
        P_TOP[0]= P_TOPRIGHT[0]= P_MEDIAN[0]=
1435
        P_TOP[1]= P_TOPRIGHT[1]= P_MEDIAN[1]= 0; //FIXME
1436
    } else {
1437
        P_TOP[0]      = s->p_mv_table[xy + s->mb_stride    ][0];
1438
        P_TOP[1]      = s->p_mv_table[xy + s->mb_stride    ][1];
1439
        P_TOPRIGHT[0] = s->p_mv_table[xy + s->mb_stride - 1][0];
1440
        P_TOPRIGHT[1] = s->p_mv_table[xy + s->mb_stride - 1][1];
1441
        if(P_TOP[1]      < (c->ymin<<shift)) P_TOP[1]     = (c->ymin<<shift);
1442
        if(P_TOPRIGHT[0] > (c->xmax<<shift)) P_TOPRIGHT[0]= (c->xmax<<shift);
1443
        if(P_TOPRIGHT[1] < (c->ymin<<shift)) P_TOPRIGHT[1]= (c->ymin<<shift);
1444

    
1445
        P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
1446
        P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
1447

    
1448
        c->pred_x = P_MEDIAN[0];
1449
        c->pred_y = P_MEDIAN[1];
1450
    }
1451

    
1452
    dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, 0, s->p_mv_table, (1<<16)>>shift, 0, 16);
1453

    
1454
    s->p_mv_table[xy][0] = mx<<shift;
1455
    s->p_mv_table[xy][1] = my<<shift;
1456

    
1457
    return dmin;
1458
}
1459

    
1460
static int ff_estimate_motion_b(MpegEncContext * s,
1461
                       int mb_x, int mb_y, int16_t (*mv_table)[2], int ref_index, int f_code)
1462
{
1463
    MotionEstContext * const c= &s->me;
1464
    int mx, my, dmin;
1465
    int P[10][2];
1466
    const int shift= 1+s->quarter_sample;
1467
    const int mot_stride = s->mb_stride;
1468
    const int mot_xy = mb_y*mot_stride + mb_x;
1469
    uint8_t * const mv_penalty= c->mv_penalty[f_code] + MAX_MV;
1470
    int mv_scale;
1471

    
1472
    c->penalty_factor    = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_cmp);
1473
    c->sub_penalty_factor= get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_sub_cmp);
1474
    c->mb_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->mb_cmp);
1475
    c->current_mv_penalty= mv_penalty;
1476

    
1477
    get_limits(s, 16*mb_x, 16*mb_y);
1478

    
1479
    switch(s->me_method) {
1480
    case ME_ZERO:
1481
    default:
1482
        no_motion_search(s, &mx, &my);
1483
        dmin = 0;
1484
        mx-= mb_x*16;
1485
        my-= mb_y*16;
1486
        break;
1487
#if 0
1488
    case ME_FULL:
1489
        dmin = full_motion_search(s, &mx, &my, range, ref_picture);
1490
        mx-= mb_x*16;
1491
        my-= mb_y*16;
1492
        break;
1493
    case ME_LOG:
1494
        dmin = log_motion_search(s, &mx, &my, range / 2, ref_picture);
1495
        mx-= mb_x*16;
1496
        my-= mb_y*16;
1497
        break;
1498
    case ME_PHODS:
1499
        dmin = phods_motion_search(s, &mx, &my, range / 2, ref_picture);
1500
        mx-= mb_x*16;
1501
        my-= mb_y*16;
1502
        break;
1503
#endif
1504
    case ME_X1:
1505
    case ME_EPZS:
1506
       {
1507
            P_LEFT[0]        = mv_table[mot_xy - 1][0];
1508
            P_LEFT[1]        = mv_table[mot_xy - 1][1];
1509

    
1510
            if(P_LEFT[0]       > (c->xmax<<shift)) P_LEFT[0]       = (c->xmax<<shift);
1511

    
1512
            /* special case for first line */
1513
            if (!s->first_slice_line) {
1514
                P_TOP[0] = mv_table[mot_xy - mot_stride             ][0];
1515
                P_TOP[1] = mv_table[mot_xy - mot_stride             ][1];
1516
                P_TOPRIGHT[0] = mv_table[mot_xy - mot_stride + 1         ][0];
1517
                P_TOPRIGHT[1] = mv_table[mot_xy - mot_stride + 1         ][1];
1518
                if(P_TOP[1] > (c->ymax<<shift)) P_TOP[1]= (c->ymax<<shift);
1519
                if(P_TOPRIGHT[0] < (c->xmin<<shift)) P_TOPRIGHT[0]= (c->xmin<<shift);
1520
                if(P_TOPRIGHT[1] > (c->ymax<<shift)) P_TOPRIGHT[1]= (c->ymax<<shift);
1521

    
1522
                P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
1523
                P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
1524
            }
1525
            c->pred_x= P_LEFT[0];
1526
            c->pred_y= P_LEFT[1];
1527
        }
1528

    
1529
        if(mv_table == s->b_forw_mv_table){
1530
            mv_scale= (s->pb_time<<16) / (s->pp_time<<shift);
1531
        }else{
1532
            mv_scale= ((s->pb_time - s->pp_time)<<16) / (s->pp_time<<shift);
1533
        }
1534

    
1535
        dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, ref_index, s->p_mv_table, mv_scale, 0, 16);
1536

    
1537
        break;
1538
    }
1539

    
1540
    dmin= c->sub_motion_search(s, &mx, &my, dmin, 0, ref_index, 0, 16);
1541

    
1542
    if(c->avctx->me_sub_cmp != c->avctx->mb_cmp && !c->skip)
1543
        dmin= ff_get_mb_score(s, mx, my, 0, ref_index, 0, 16, 1);
1544

    
1545
//printf("%d %d %d %d//", s->mb_x, s->mb_y, mx, my);
1546
//    s->mb_type[mb_y*s->mb_width + mb_x]= mb_type;
1547
    mv_table[mot_xy][0]= mx;
1548
    mv_table[mot_xy][1]= my;
1549

    
1550
    return dmin;
1551
}
1552

    
1553
static inline int check_bidir_mv(MpegEncContext * s,
1554
                   int motion_fx, int motion_fy,
1555
                   int motion_bx, int motion_by,
1556
                   int pred_fx, int pred_fy,
1557
                   int pred_bx, int pred_by,
1558
                   int size, int h)
1559
{
1560
    //FIXME optimize?
1561
    //FIXME better f_code prediction (max mv & distance)
1562
    //FIXME pointers
1563
    MotionEstContext * const c= &s->me;
1564
    uint8_t * const mv_penalty_f= c->mv_penalty[s->f_code] + MAX_MV; // f_code of the prev frame
1565
    uint8_t * const mv_penalty_b= c->mv_penalty[s->b_code] + MAX_MV; // f_code of the prev frame
1566
    int stride= c->stride;
1567
    uint8_t *dest_y = c->scratchpad;
1568
    uint8_t *ptr;
1569
    int dxy;
1570
    int src_x, src_y;
1571
    int fbmin;
1572
    uint8_t **src_data= c->src[0];
1573
    uint8_t **ref_data= c->ref[0];
1574
    uint8_t **ref2_data= c->ref[2];
1575

    
1576
    if(s->quarter_sample){
1577
        dxy = ((motion_fy & 3) << 2) | (motion_fx & 3);
1578
        src_x = motion_fx >> 2;
1579
        src_y = motion_fy >> 2;
1580

    
1581
        ptr = ref_data[0] + (src_y * stride) + src_x;
1582
        s->dsp.put_qpel_pixels_tab[0][dxy](dest_y    , ptr    , stride);
1583

    
1584
        dxy = ((motion_by & 3) << 2) | (motion_bx & 3);
1585
        src_x = motion_bx >> 2;
1586
        src_y = motion_by >> 2;
1587

    
1588
        ptr = ref2_data[0] + (src_y * stride) + src_x;
1589
        s->dsp.avg_qpel_pixels_tab[size][dxy](dest_y    , ptr    , stride);
1590
    }else{
1591
        dxy = ((motion_fy & 1) << 1) | (motion_fx & 1);
1592
        src_x = motion_fx >> 1;
1593
        src_y = motion_fy >> 1;
1594

    
1595
        ptr = ref_data[0] + (src_y * stride) + src_x;
1596
        s->dsp.put_pixels_tab[size][dxy](dest_y    , ptr    , stride, h);
1597

    
1598
        dxy = ((motion_by & 1) << 1) | (motion_bx & 1);
1599
        src_x = motion_bx >> 1;
1600
        src_y = motion_by >> 1;
1601

    
1602
        ptr = ref2_data[0] + (src_y * stride) + src_x;
1603
        s->dsp.avg_pixels_tab[size][dxy](dest_y    , ptr    , stride, h);
1604
    }
1605

    
1606
    fbmin = (mv_penalty_f[motion_fx-pred_fx] + mv_penalty_f[motion_fy-pred_fy])*c->mb_penalty_factor
1607
           +(mv_penalty_b[motion_bx-pred_bx] + mv_penalty_b[motion_by-pred_by])*c->mb_penalty_factor
1608
           + s->dsp.mb_cmp[size](s, src_data[0], dest_y, stride, h); //FIXME new_pic
1609

    
1610
    if(c->avctx->mb_cmp&FF_CMP_CHROMA){
1611
    }
1612
    //FIXME CHROMA !!!
1613

    
1614
    return fbmin;
1615
}
1616

    
1617
/* refine the bidir vectors in hq mode and return the score in both lq & hq mode*/
1618
static inline int bidir_refine(MpegEncContext * s, int mb_x, int mb_y)
1619
{
1620
    MotionEstContext * const c= &s->me;
1621
    const int mot_stride = s->mb_stride;
1622
    const int xy = mb_y *mot_stride + mb_x;
1623
    int fbmin;
1624
    int pred_fx= s->b_bidir_forw_mv_table[xy-1][0];
1625
    int pred_fy= s->b_bidir_forw_mv_table[xy-1][1];
1626
    int pred_bx= s->b_bidir_back_mv_table[xy-1][0];
1627
    int pred_by= s->b_bidir_back_mv_table[xy-1][1];
1628
    int motion_fx= s->b_bidir_forw_mv_table[xy][0]= s->b_forw_mv_table[xy][0];
1629
    int motion_fy= s->b_bidir_forw_mv_table[xy][1]= s->b_forw_mv_table[xy][1];
1630
    int motion_bx= s->b_bidir_back_mv_table[xy][0]= s->b_back_mv_table[xy][0];
1631
    int motion_by= s->b_bidir_back_mv_table[xy][1]= s->b_back_mv_table[xy][1];
1632
    const int flags= c->sub_flags;
1633
    const int qpel= flags&FLAG_QPEL;
1634
    const int shift= 1+qpel;
1635
    const int xmin= c->xmin<<shift;
1636
    const int ymin= c->ymin<<shift;
1637
    const int xmax= c->xmax<<shift;
1638
    const int ymax= c->ymax<<shift;
1639

    
1640
    fbmin= check_bidir_mv(s, motion_fx, motion_fy,
1641
                          motion_bx, motion_by,
1642
                          pred_fx, pred_fy,
1643
                          pred_bx, pred_by,
1644
                          0, 16);
1645

    
1646
    if(s->avctx->bidir_refine){
1647
        int score, end;
1648
#define CHECK_BIDIR(fx,fy,bx,by)\
1649
    if(  (fx<=0 || motion_fx+fx<=xmax) && (fy<=0 || motion_fy+fy<=ymax) && (bx<=0 || motion_bx+bx<=xmax) && (by<=0 || motion_by+by<=ymax)\
1650
       &&(fx>=0 || motion_fx+fx>=xmin) && (fy>=0 || motion_fy+fy>=ymin) && (bx>=0 || motion_bx+bx>=xmin) && (by>=0 || motion_by+by>=ymin)){\
1651
        score= check_bidir_mv(s, motion_fx+fx, motion_fy+fy, motion_bx+bx, motion_by+by, pred_fx, pred_fy, pred_bx, pred_by, 0, 16);\
1652
        if(score < fbmin){\
1653
            fbmin= score;\
1654
            motion_fx+=fx;\
1655
            motion_fy+=fy;\
1656
            motion_bx+=bx;\
1657
            motion_by+=by;\
1658
            end=0;\
1659
        }\
1660
    }
1661
#define CHECK_BIDIR2(a,b,c,d)\
1662
CHECK_BIDIR(a,b,c,d)\
1663
CHECK_BIDIR(-a,-b,-c,-d)
1664

    
1665
#define CHECK_BIDIRR(a,b,c,d)\
1666
CHECK_BIDIR2(a,b,c,d)\
1667
CHECK_BIDIR2(b,c,d,a)\
1668
CHECK_BIDIR2(c,d,a,b)\
1669
CHECK_BIDIR2(d,a,b,c)
1670

    
1671
        do{
1672
            end=1;
1673

    
1674
            CHECK_BIDIRR( 0, 0, 0, 1)
1675
            if(s->avctx->bidir_refine > 1){
1676
                CHECK_BIDIRR( 0, 0, 1, 1)
1677
                CHECK_BIDIR2( 0, 1, 0, 1)
1678
                CHECK_BIDIR2( 1, 0, 1, 0)
1679
                CHECK_BIDIRR( 0, 0,-1, 1)
1680
                CHECK_BIDIR2( 0,-1, 0, 1)
1681
                CHECK_BIDIR2(-1, 0, 1, 0)
1682
                if(s->avctx->bidir_refine > 2){
1683
                    CHECK_BIDIRR( 0, 1, 1, 1)
1684
                    CHECK_BIDIRR( 0,-1, 1, 1)
1685
                    CHECK_BIDIRR( 0, 1,-1, 1)
1686
                    CHECK_BIDIRR( 0, 1, 1,-1)
1687
                    if(s->avctx->bidir_refine > 3){
1688
                        CHECK_BIDIR2( 1, 1, 1, 1)
1689
                        CHECK_BIDIRR( 1, 1, 1,-1)
1690
                        CHECK_BIDIR2( 1, 1,-1,-1)
1691
                        CHECK_BIDIR2( 1,-1,-1, 1)
1692
                        CHECK_BIDIR2( 1,-1, 1,-1)
1693
                    }
1694
                }
1695
            }
1696
        }while(!end);
1697
    }
1698

    
1699
    return fbmin;
1700
}
1701

    
1702
static inline int direct_search(MpegEncContext * s, int mb_x, int mb_y)
1703
{
1704
    MotionEstContext * const c= &s->me;
1705
    int P[10][2];
1706
    const int mot_stride = s->mb_stride;
1707
    const int mot_xy = mb_y*mot_stride + mb_x;
1708
    const int shift= 1+s->quarter_sample;
1709
    int dmin, i;
1710
    const int time_pp= s->pp_time;
1711
    const int time_pb= s->pb_time;
1712
    int mx, my, xmin, xmax, ymin, ymax;
1713
    int16_t (*mv_table)[2]= s->b_direct_mv_table;
1714

    
1715
    c->current_mv_penalty= c->mv_penalty[1] + MAX_MV;
1716
    ymin= xmin=(-32)>>shift;
1717
    ymax= xmax=   31>>shift;
1718

    
1719
    if(IS_8X8(s->next_picture.mb_type[mot_xy])){
1720
        s->mv_type= MV_TYPE_8X8;
1721
    }else{
1722
        s->mv_type= MV_TYPE_16X16;
1723
    }
1724

    
1725
    for(i=0; i<4; i++){
1726
        int index= s->block_index[i];
1727
        int min, max;
1728

    
1729
        c->co_located_mv[i][0]= s->next_picture.motion_val[0][index][0];
1730
        c->co_located_mv[i][1]= s->next_picture.motion_val[0][index][1];
1731
        c->direct_basis_mv[i][0]= c->co_located_mv[i][0]*time_pb/time_pp + ((i& 1)<<(shift+3));
1732
        c->direct_basis_mv[i][1]= c->co_located_mv[i][1]*time_pb/time_pp + ((i>>1)<<(shift+3));
1733
//        c->direct_basis_mv[1][i][0]= c->co_located_mv[i][0]*(time_pb - time_pp)/time_pp + ((i &1)<<(shift+3);
1734
//        c->direct_basis_mv[1][i][1]= c->co_located_mv[i][1]*(time_pb - time_pp)/time_pp + ((i>>1)<<(shift+3);
1735

    
1736
        max= FFMAX(c->direct_basis_mv[i][0], c->direct_basis_mv[i][0] - c->co_located_mv[i][0])>>shift;
1737
        min= FFMIN(c->direct_basis_mv[i][0], c->direct_basis_mv[i][0] - c->co_located_mv[i][0])>>shift;
1738
        max+= 16*mb_x + 1; // +-1 is for the simpler rounding
1739
        min+= 16*mb_x - 1;
1740
        xmax= FFMIN(xmax, s->width - max);
1741
        xmin= FFMAX(xmin, - 16     - min);
1742

    
1743
        max= FFMAX(c->direct_basis_mv[i][1], c->direct_basis_mv[i][1] - c->co_located_mv[i][1])>>shift;
1744
        min= FFMIN(c->direct_basis_mv[i][1], c->direct_basis_mv[i][1] - c->co_located_mv[i][1])>>shift;
1745
        max+= 16*mb_y + 1; // +-1 is for the simpler rounding
1746
        min+= 16*mb_y - 1;
1747
        ymax= FFMIN(ymax, s->height - max);
1748
        ymin= FFMAX(ymin, - 16      - min);
1749

    
1750
        if(s->mv_type == MV_TYPE_16X16) break;
1751
    }
1752

    
1753
    assert(xmax <= 15 && ymax <= 15 && xmin >= -16 && ymin >= -16);
1754

    
1755
    if(xmax < 0 || xmin >0 || ymax < 0 || ymin > 0){
1756
        s->b_direct_mv_table[mot_xy][0]= 0;
1757
        s->b_direct_mv_table[mot_xy][1]= 0;
1758

    
1759
        return 256*256*256*64;
1760
    }
1761

    
1762
    c->xmin= xmin;
1763
    c->ymin= ymin;
1764
    c->xmax= xmax;
1765
    c->ymax= ymax;
1766
    c->flags     |= FLAG_DIRECT;
1767
    c->sub_flags |= FLAG_DIRECT;
1768
    c->pred_x=0;
1769
    c->pred_y=0;
1770

    
1771
    P_LEFT[0]        = clip(mv_table[mot_xy - 1][0], xmin<<shift, xmax<<shift);
1772
    P_LEFT[1]        = clip(mv_table[mot_xy - 1][1], ymin<<shift, ymax<<shift);
1773

    
1774
    /* special case for first line */
1775
    if (!s->first_slice_line) { //FIXME maybe allow this over thread boundary as its cliped
1776
        P_TOP[0]      = clip(mv_table[mot_xy - mot_stride             ][0], xmin<<shift, xmax<<shift);
1777
        P_TOP[1]      = clip(mv_table[mot_xy - mot_stride             ][1], ymin<<shift, ymax<<shift);
1778
        P_TOPRIGHT[0] = clip(mv_table[mot_xy - mot_stride + 1         ][0], xmin<<shift, xmax<<shift);
1779
        P_TOPRIGHT[1] = clip(mv_table[mot_xy - mot_stride + 1         ][1], ymin<<shift, ymax<<shift);
1780

    
1781
        P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
1782
        P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
1783
    }
1784

    
1785
    dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, 0, mv_table, 1<<(16-shift), 0, 16);
1786
    if(c->sub_flags&FLAG_QPEL)
1787
        dmin = qpel_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
1788
    else
1789
        dmin = hpel_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
1790

    
1791
    if(c->avctx->me_sub_cmp != c->avctx->mb_cmp && !c->skip)
1792
        dmin= ff_get_mb_score(s, mx, my, 0, 0, 0, 16, 1);
1793

    
1794
    get_limits(s, 16*mb_x, 16*mb_y); //restore c->?min/max, maybe not needed
1795

    
1796
    s->b_direct_mv_table[mot_xy][0]= mx;
1797
    s->b_direct_mv_table[mot_xy][1]= my;
1798
    c->flags     &= ~FLAG_DIRECT;
1799
    c->sub_flags &= ~FLAG_DIRECT;
1800

    
1801
    return dmin;
1802
}
1803

    
1804
void ff_estimate_b_frame_motion(MpegEncContext * s,
1805
                             int mb_x, int mb_y)
1806
{
1807
    MotionEstContext * const c= &s->me;
1808
    const int penalty_factor= c->mb_penalty_factor;
1809
    int fmin, bmin, dmin, fbmin, bimin, fimin;
1810
    int type=0;
1811
    const int xy = mb_y*s->mb_stride + mb_x;
1812
    init_ref(c, s->new_picture.data, s->last_picture.data, s->next_picture.data, 16*mb_x, 16*mb_y, 2);
1813

    
1814
    get_limits(s, 16*mb_x, 16*mb_y);
1815

    
1816
    c->skip=0;
1817
    if(c->avctx->me_threshold){
1818
        int vard= (check_input_motion(s, mb_x, mb_y, 0)+128)>>8;
1819

    
1820
        if(vard<c->avctx->me_threshold){
1821
//            pix = c->src[0][0];
1822
//            sum = s->dsp.pix_sum(pix, s->linesize);
1823
//            varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500 + 128)>>8;
1824

    
1825
//            pic->mb_var   [s->mb_stride * mb_y + mb_x] = varc;
1826
             s->current_picture.mc_mb_var[s->mb_stride * mb_y + mb_x] = vard;
1827
/*            pic->mb_mean  [s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
1828
            c->mb_var_sum_temp    += varc;*/
1829
            c->mc_mb_var_sum_temp += vard;
1830
/*            if (vard <= 64 || vard < varc) {
1831
                c->scene_change_score+= ff_sqrt(vard<<8) - ff_sqrt(varc<<8);
1832
            }else{
1833
                c->scene_change_score+= s->qscale;
1834
            }*/
1835
            return;
1836
        }
1837
        if(vard<c->avctx->mb_threshold){
1838
            type= s->mb_type[mb_y*s->mb_stride + mb_x];
1839
            if(type == CANDIDATE_MB_TYPE_DIRECT){
1840
                direct_search(s, mb_x, mb_y);
1841
            }
1842
            if(type == CANDIDATE_MB_TYPE_FORWARD || type == CANDIDATE_MB_TYPE_BIDIR){
1843
                c->skip=0;
1844
                ff_estimate_motion_b(s, mb_x, mb_y, s->b_forw_mv_table, 0, s->f_code);
1845
            }
1846
            if(type == CANDIDATE_MB_TYPE_BACKWARD || type == CANDIDATE_MB_TYPE_BIDIR){
1847
                c->skip=0;
1848
                ff_estimate_motion_b(s, mb_x, mb_y, s->b_back_mv_table, 2, s->b_code);
1849
            }
1850
            if(type == CANDIDATE_MB_TYPE_FORWARD_I || type == CANDIDATE_MB_TYPE_BIDIR_I){
1851
                c->skip=0;
1852
                c->current_mv_penalty= c->mv_penalty[s->f_code] + MAX_MV;
1853
                interlaced_search(s, 0,
1854
                                        s->b_field_mv_table[0], s->b_field_select_table[0],
1855
                                        s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1], 1);
1856
            }
1857
            if(type == CANDIDATE_MB_TYPE_BACKWARD_I || type == CANDIDATE_MB_TYPE_BIDIR_I){
1858
                c->skip=0;
1859
                c->current_mv_penalty= c->mv_penalty[s->b_code] + MAX_MV;
1860
                interlaced_search(s, 2,
1861
                                        s->b_field_mv_table[1], s->b_field_select_table[1],
1862
                                        s->b_back_mv_table[xy][0], s->b_back_mv_table[xy][1], 1);
1863
            }
1864
            return;
1865
        }
1866
    }
1867

    
1868
    if (s->codec_id == CODEC_ID_MPEG4)
1869
        dmin= direct_search(s, mb_x, mb_y);
1870
    else
1871
        dmin= INT_MAX;
1872
//FIXME penalty stuff for non mpeg4
1873
    c->skip=0;
1874
    fmin= ff_estimate_motion_b(s, mb_x, mb_y, s->b_forw_mv_table, 0, s->f_code) + 3*penalty_factor;
1875

    
1876
    c->skip=0;
1877
    bmin= ff_estimate_motion_b(s, mb_x, mb_y, s->b_back_mv_table, 2, s->b_code) + 2*penalty_factor;
1878
//printf(" %d %d ", s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1]);
1879

    
1880
    c->skip=0;
1881
    fbmin= bidir_refine(s, mb_x, mb_y) + penalty_factor;
1882
//printf("%d %d %d %d\n", dmin, fmin, bmin, fbmin);
1883

    
1884
    if(s->flags & CODEC_FLAG_INTERLACED_ME){
1885
//FIXME mb type penalty
1886
        c->skip=0;
1887
        c->current_mv_penalty= c->mv_penalty[s->f_code] + MAX_MV;
1888
        fimin= interlaced_search(s, 0,
1889
                                 s->b_field_mv_table[0], s->b_field_select_table[0],
1890
                                 s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1], 0);
1891
        c->current_mv_penalty= c->mv_penalty[s->b_code] + MAX_MV;
1892
        bimin= interlaced_search(s, 2,
1893
                                 s->b_field_mv_table[1], s->b_field_select_table[1],
1894
                                 s->b_back_mv_table[xy][0], s->b_back_mv_table[xy][1], 0);
1895
    }else
1896
        fimin= bimin= INT_MAX;
1897

    
1898
    {
1899
        int score= fmin;
1900
        type = CANDIDATE_MB_TYPE_FORWARD;
1901

    
1902
        if (dmin <= score){
1903
            score = dmin;
1904
            type = CANDIDATE_MB_TYPE_DIRECT;
1905
        }
1906
        if(bmin<score){
1907
            score=bmin;
1908
            type= CANDIDATE_MB_TYPE_BACKWARD;
1909
        }
1910
        if(fbmin<score){
1911
            score=fbmin;
1912
            type= CANDIDATE_MB_TYPE_BIDIR;
1913
        }
1914
        if(fimin<score){
1915
            score=fimin;
1916
            type= CANDIDATE_MB_TYPE_FORWARD_I;
1917
        }
1918
        if(bimin<score){
1919
            score=bimin;
1920
            type= CANDIDATE_MB_TYPE_BACKWARD_I;
1921
        }
1922

    
1923
        score= ((unsigned)(score*score + 128*256))>>16;
1924
        c->mc_mb_var_sum_temp += score;
1925
        s->current_picture.mc_mb_var[mb_y*s->mb_stride + mb_x] = score; //FIXME use SSE
1926
    }
1927

    
1928
    if(c->avctx->mb_decision > FF_MB_DECISION_SIMPLE){
1929
        type= CANDIDATE_MB_TYPE_FORWARD | CANDIDATE_MB_TYPE_BACKWARD | CANDIDATE_MB_TYPE_BIDIR | CANDIDATE_MB_TYPE_DIRECT;
1930
        if(fimin < INT_MAX)
1931
            type |= CANDIDATE_MB_TYPE_FORWARD_I;
1932
        if(bimin < INT_MAX)
1933
            type |= CANDIDATE_MB_TYPE_BACKWARD_I;
1934
        if(fimin < INT_MAX && bimin < INT_MAX){
1935
            type |= CANDIDATE_MB_TYPE_BIDIR_I;
1936
        }
1937
         //FIXME something smarter
1938
        if(dmin>256*256*16) type&= ~CANDIDATE_MB_TYPE_DIRECT; //dont try direct mode if its invalid for this MB
1939
#if 0
1940
        if(s->out_format == FMT_MPEG1)
1941
            type |= CANDIDATE_MB_TYPE_INTRA;
1942
#endif
1943
    }
1944

    
1945
    s->mb_type[mb_y*s->mb_stride + mb_x]= type;
1946
}
1947

    
1948
/* find best f_code for ME which do unlimited searches */
1949
int ff_get_best_fcode(MpegEncContext * s, int16_t (*mv_table)[2], int type)
1950
{
1951
    if(s->me_method>=ME_EPZS){
1952
        int score[8];
1953
        int i, y, range= s->avctx->me_range ? s->avctx->me_range : (INT_MAX/2);
1954
        uint8_t * fcode_tab= s->fcode_tab;
1955
        int best_fcode=-1;
1956
        int best_score=-10000000;
1957

    
1958
        if(s->msmpeg4_version)
1959
            range= FFMIN(range, 16);
1960
        else if(s->codec_id == CODEC_ID_MPEG2VIDEO && s->avctx->strict_std_compliance >= FF_COMPLIANCE_NORMAL)
1961
            range= FFMIN(range, 256);
1962

    
1963
        for(i=0; i<8; i++) score[i]= s->mb_num*(8-i);
1964

    
1965
        for(y=0; y<s->mb_height; y++){
1966
            int x;
1967
            int xy= y*s->mb_stride;
1968
            for(x=0; x<s->mb_width; x++){
1969
                if(s->mb_type[xy] & type){
1970
                    int mx= mv_table[xy][0];
1971
                    int my= mv_table[xy][1];
1972
                    int fcode= FFMAX(fcode_tab[mx + MAX_MV],
1973
                                     fcode_tab[my + MAX_MV]);
1974
                    int j;
1975

    
1976
                        if(mx >= range || mx < -range ||
1977
                           my >= range || my < -range)
1978
                            continue;
1979

    
1980
                    for(j=0; j<fcode && j<8; j++){
1981
                        if(s->pict_type==B_TYPE || s->current_picture.mc_mb_var[xy] < s->current_picture.mb_var[xy])
1982
                            score[j]-= 170;
1983
                    }
1984
                }
1985
                xy++;
1986
            }
1987
        }
1988

    
1989
        for(i=1; i<8; i++){
1990
            if(score[i] > best_score){
1991
                best_score= score[i];
1992
                best_fcode= i;
1993
            }
1994
//            printf("%d %d\n", i, score[i]);
1995
        }
1996

    
1997
//    printf("fcode: %d type: %d\n", i, s->pict_type);
1998
        return best_fcode;
1999
/*        for(i=0; i<=MAX_FCODE; i++){
2000
            printf("%d ", mv_num[i]);
2001
        }
2002
        printf("\n");*/
2003
    }else{
2004
        return 1;
2005
    }
2006
}
2007

    
2008
void ff_fix_long_p_mvs(MpegEncContext * s)
2009
{
2010
    MotionEstContext * const c= &s->me;
2011
    const int f_code= s->f_code;
2012
    int y, range;
2013
    assert(s->pict_type==P_TYPE);
2014

    
2015
    range = (((s->out_format == FMT_MPEG1 || s->msmpeg4_version) ? 8 : 16) << f_code);
2016

    
2017
    assert(range <= 16 || !s->msmpeg4_version);
2018
    assert(range <=256 || !(s->codec_id == CODEC_ID_MPEG2VIDEO && s->avctx->strict_std_compliance >= FF_COMPLIANCE_NORMAL));
2019

    
2020
    if(c->avctx->me_range && range > c->avctx->me_range) range= c->avctx->me_range;
2021

    
2022
//printf("%d no:%d %d//\n", clip, noclip, f_code);
2023
    if(s->flags&CODEC_FLAG_4MV){
2024
        const int wrap= s->b8_stride;
2025

    
2026
        /* clip / convert to intra 8x8 type MVs */
2027
        for(y=0; y<s->mb_height; y++){
2028
            int xy= y*2*wrap;
2029
            int i= y*s->mb_stride;
2030
            int x;
2031

    
2032
            for(x=0; x<s->mb_width; x++){
2033
                if(s->mb_type[i]&CANDIDATE_MB_TYPE_INTER4V){
2034
                    int block;
2035
                    for(block=0; block<4; block++){
2036
                        int off= (block& 1) + (block>>1)*wrap;
2037
                        int mx= s->current_picture.motion_val[0][ xy + off ][0];
2038
                        int my= s->current_picture.motion_val[0][ xy + off ][1];
2039

    
2040
                        if(   mx >=range || mx <-range
2041
                           || my >=range || my <-range){
2042
                            s->mb_type[i] &= ~CANDIDATE_MB_TYPE_INTER4V;
2043
                            s->mb_type[i] |= CANDIDATE_MB_TYPE_INTRA;
2044
                            s->current_picture.mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
2045
                        }
2046
                    }
2047
                }
2048
                xy+=2;
2049
                i++;
2050
            }
2051
        }
2052
    }
2053
}
2054

    
2055
/**
2056
 *
2057
 * @param truncate 1 for truncation, 0 for using intra
2058
 */
2059
void ff_fix_long_mvs(MpegEncContext * s, uint8_t *field_select_table, int field_select,
2060
                     int16_t (*mv_table)[2], int f_code, int type, int truncate)
2061
{
2062
    MotionEstContext * const c= &s->me;
2063
    int y, h_range, v_range;
2064

    
2065
    // RAL: 8 in MPEG-1, 16 in MPEG-4
2066
    int range = (((s->out_format == FMT_MPEG1 || s->msmpeg4_version) ? 8 : 16) << f_code);
2067

    
2068
    if(c->avctx->me_range && range > c->avctx->me_range) range= c->avctx->me_range;
2069

    
2070
    h_range= range;
2071
    v_range= field_select_table ? range>>1 : range;
2072

    
2073
    /* clip / convert to intra 16x16 type MVs */
2074
    for(y=0; y<s->mb_height; y++){
2075
        int x;
2076
        int xy= y*s->mb_stride;
2077
        for(x=0; x<s->mb_width; x++){
2078
            if (s->mb_type[xy] & type){    // RAL: "type" test added...
2079
                if(field_select_table==NULL || field_select_table[xy] == field_select){
2080
                    if(   mv_table[xy][0] >=h_range || mv_table[xy][0] <-h_range
2081
                       || mv_table[xy][1] >=v_range || mv_table[xy][1] <-v_range){
2082

    
2083
                        if(truncate){
2084
                            if     (mv_table[xy][0] > h_range-1) mv_table[xy][0]=  h_range-1;
2085
                            else if(mv_table[xy][0] < -h_range ) mv_table[xy][0]= -h_range;
2086
                            if     (mv_table[xy][1] > v_range-1) mv_table[xy][1]=  v_range-1;
2087
                            else if(mv_table[xy][1] < -v_range ) mv_table[xy][1]= -v_range;
2088
                        }else{
2089
                            s->mb_type[xy] &= ~type;
2090
                            s->mb_type[xy] |= CANDIDATE_MB_TYPE_INTRA;
2091
                            mv_table[xy][0]=
2092
                            mv_table[xy][1]= 0;
2093
                        }
2094
                    }
2095
                }
2096
            }
2097
            xy++;
2098
        }
2099
    }
2100
}