GCC Code Coverage Report
Directory: ../../../ffmpeg/ Exec Total Coverage
File: src/libavcodec/motion_est.c Lines: 861 1000 86.1 %
Date: 2019-11-18 18:00:01 Branches: 500 619 80.8 %

Line Branch Exec Source
1
/*
2
 * Motion estimation
3
 * Copyright (c) 2000,2001 Fabrice Bellard
4
 * Copyright (c) 2002-2004 Michael Niedermayer
5
 *
6
 * new motion estimation (X1/EPZS) by Michael Niedermayer <michaelni@gmx.at>
7
 *
8
 * This file is part of FFmpeg.
9
 *
10
 * FFmpeg is free software; you can redistribute it and/or
11
 * modify it under the terms of the GNU Lesser General Public
12
 * License as published by the Free Software Foundation; either
13
 * version 2.1 of the License, or (at your option) any later version.
14
 *
15
 * FFmpeg is distributed in the hope that it will be useful,
16
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
18
 * Lesser General Public License for more details.
19
 *
20
 * You should have received a copy of the GNU Lesser General Public
21
 * License along with FFmpeg; if not, write to the Free Software
22
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23
 */
24
25
/**
26
 * @file
27
 * Motion estimation.
28
 */
29
30
#include <stdlib.h>
31
#include <stdio.h>
32
#include <limits.h>
33
34
#include "avcodec.h"
35
#include "internal.h"
36
#include "mathops.h"
37
#include "motion_est.h"
38
#include "mpegutils.h"
39
#include "mpegvideo.h"
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
#define ME_MAP_SHIFT 3
48
#define ME_MAP_MV_BITS 11
49
50
static int sad_hpel_motion_search(MpegEncContext * s,
51
                                  int *mx_ptr, int *my_ptr, int dmin,
52
                                  int src_index, int ref_index,
53
                                  int size, int h);
54
55
5662581
static inline unsigned update_map_generation(MotionEstContext *c)
56
{
57
5662581
    c->map_generation+= 1<<(ME_MAP_MV_BITS*2);
58
5662581
    if(c->map_generation==0){
59
5460
        c->map_generation= 1<<(ME_MAP_MV_BITS*2);
60
5460
        memset(c->map, 0, sizeof(uint32_t)*ME_MAP_SIZE);
61
    }
62
5662581
    return c->map_generation;
63
}
64
65
/* shape adaptive search stuff */
66
typedef struct Minima{
67
    int height;
68
    int x, y;
69
    int checked;
70
}Minima;
71
72
static int minima_cmp(const void *a, const void *b){
73
    const Minima *da = (const Minima *) a;
74
    const Minima *db = (const Minima *) b;
75
76
    return da->height - db->height;
77
}
78
79
#define FLAG_QPEL   1 //must be 1
80
#define FLAG_CHROMA 2
81
#define FLAG_DIRECT 4
82
83
2173014
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){
84
2173014
    const int offset[3]= {
85
2173014
          y*c->  stride + x,
86
2173014
        ((y*c->uvstride + x)>>1),
87
2173014
        ((y*c->uvstride + x)>>1),
88
    };
89
    int i;
90
8692056
    for(i=0; i<3; i++){
91
6519042
        c->src[0][i]= src [i] + offset[i];
92
6519042
        c->ref[0][i]= ref [i] + offset[i];
93
    }
94
2173014
    if(ref_index){
95
1633248
        for(i=0; i<3; i++){
96
1224936
            c->ref[ref_index][i]= ref2[i] + offset[i];
97
        }
98
    }
99
2173014
}
100
101
28383
static int get_flags(MotionEstContext *c, int direct, int chroma){
102
28383
    return   ((c->avctx->flags&AV_CODEC_FLAG_QPEL) ? FLAG_QPEL : 0)
103
28383
           + (direct ? FLAG_DIRECT : 0)
104
28383
           + (chroma ? FLAG_CHROMA : 0);
105
}
106
107
2666114
static av_always_inline int cmp_direct_inline(MpegEncContext *s, const int x, const int y, const int subx, const int suby,
108
                      const int size, const int h, int ref_index, int src_index,
109
                      me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, int qpel){
110
2666114
    MotionEstContext * const c= &s->me;
111
2666114
    const int stride= c->stride;
112
2666114
    const int hx= subx + (x<<(1+qpel));
113
2666114
    const int hy= suby + (y<<(1+qpel));
114
2666114
    uint8_t * const * const ref= c->ref[ref_index];
115
2666114
    uint8_t * const * const src= c->src[src_index];
116
    int d;
117
    //FIXME check chroma 4mv, (no crashes ...)
118
        av_assert2(x >= c->xmin && hx <= c->xmax<<(qpel+1) && y >= c->ymin && hy <= c->ymax<<(qpel+1));
119


5332228
        if(x >= c->xmin && hx <= c->xmax<<(qpel+1) && y >= c->ymin && hy <= c->ymax<<(qpel+1)){
120
2666114
            const int time_pp= s->pp_time;
121
2666114
            const int time_pb= s->pb_time;
122
2666114
            const int mask= 2*qpel+1;
123
2666114
            if(s->mv_type==MV_TYPE_8X8){
124
                int i;
125
3320665
                for(i=0; i<4; i++){
126
2656532
                    int fx = c->direct_basis_mv[i][0] + hx;
127
2656532
                    int fy = c->direct_basis_mv[i][1] + hy;
128
2656532
                    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));
129
2656532
                    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));
130
2656532
                    int fxy= (fx&mask) + ((fy&mask)<<(qpel+1));
131
2656532
                    int bxy= (bx&mask) + ((by&mask)<<(qpel+1));
132
133
2656532
                    uint8_t *dst= c->temp + 8*(i&1) + 8*stride*(i>>1);
134
2656532
                    if(qpel){
135
961440
                        c->qpel_put[1][fxy](dst, ref[0] + (fx>>2) + (fy>>2)*stride, stride);
136
961440
                        c->qpel_avg[1][bxy](dst, ref[8] + (bx>>2) + (by>>2)*stride, stride);
137
                    }else{
138
1695092
                        c->hpel_put[1][fxy](dst, ref[0] + (fx>>1) + (fy>>1)*stride, stride, 8);
139
1695092
                        c->hpel_avg[1][bxy](dst, ref[8] + (bx>>1) + (by>>1)*stride, stride, 8);
140
                    }
141
                }
142
            }else{
143
2001981
                int fx = c->direct_basis_mv[0][0] + hx;
144
2001981
                int fy = c->direct_basis_mv[0][1] + hy;
145
2001981
                int bx = hx ? fx - c->co_located_mv[0][0] : (c->co_located_mv[0][0]*(time_pb - time_pp)/time_pp);
146
2001981
                int by = hy ? fy - c->co_located_mv[0][1] : (c->co_located_mv[0][1]*(time_pb - time_pp)/time_pp);
147
2001981
                int fxy= (fx&mask) + ((fy&mask)<<(qpel+1));
148
2001981
                int bxy= (bx&mask) + ((by&mask)<<(qpel+1));
149
150
2001981
                if(qpel){
151
433681
                    c->qpel_put[1][fxy](c->temp               , ref[0] + (fx>>2) + (fy>>2)*stride               , stride);
152
433681
                    c->qpel_put[1][fxy](c->temp + 8           , ref[0] + (fx>>2) + (fy>>2)*stride + 8           , stride);
153
433681
                    c->qpel_put[1][fxy](c->temp     + 8*stride, ref[0] + (fx>>2) + (fy>>2)*stride     + 8*stride, stride);
154
433681
                    c->qpel_put[1][fxy](c->temp + 8 + 8*stride, ref[0] + (fx>>2) + (fy>>2)*stride + 8 + 8*stride, stride);
155
433681
                    c->qpel_avg[1][bxy](c->temp               , ref[8] + (bx>>2) + (by>>2)*stride               , stride);
156
433681
                    c->qpel_avg[1][bxy](c->temp + 8           , ref[8] + (bx>>2) + (by>>2)*stride + 8           , stride);
157
433681
                    c->qpel_avg[1][bxy](c->temp     + 8*stride, ref[8] + (bx>>2) + (by>>2)*stride     + 8*stride, stride);
158
433681
                    c->qpel_avg[1][bxy](c->temp + 8 + 8*stride, ref[8] + (bx>>2) + (by>>2)*stride + 8 + 8*stride, stride);
159
                }else{
160
                    av_assert2((fx>>1) + 16*s->mb_x >= -16);
161
                    av_assert2((fy>>1) + 16*s->mb_y >= -16);
162
                    av_assert2((fx>>1) + 16*s->mb_x <= s->width);
163
                    av_assert2((fy>>1) + 16*s->mb_y <= s->height);
164
                    av_assert2((bx>>1) + 16*s->mb_x >= -16);
165
                    av_assert2((by>>1) + 16*s->mb_y >= -16);
166
                    av_assert2((bx>>1) + 16*s->mb_x <= s->width);
167
                    av_assert2((by>>1) + 16*s->mb_y <= s->height);
168
169
1568300
                    c->hpel_put[0][fxy](c->temp, ref[0] + (fx>>1) + (fy>>1)*stride, stride, 16);
170
1568300
                    c->hpel_avg[0][bxy](c->temp, ref[8] + (bx>>1) + (by>>1)*stride, stride, 16);
171
                }
172
            }
173
2666114
            d = cmp_func(s, c->temp, src[0], stride, 16);
174
        }else
175
            d= 256*256*256*32;
176
2666114
    return d;
177
}
178
179
56753788
static av_always_inline int cmp_inline(MpegEncContext *s, const int x, const int y, const int subx, const int suby,
180
                      const int size, const int h, int ref_index, int src_index,
181
                      me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, int qpel, int chroma){
182
56753788
    MotionEstContext * const c= &s->me;
183
56753788
    const int stride= c->stride;
184
56753788
    const int uvstride= c->uvstride;
185
56753788
    const int dxy= subx + (suby<<(1+qpel)); //FIXME log2_subpel?
186
56753788
    const int hx= subx + x*(1<<(1+qpel));
187
56753788
    const int hy= suby + y*(1<<(1+qpel));
188
56753788
    uint8_t * const * const ref= c->ref[ref_index];
189
56753788
    uint8_t * const * const src= c->src[src_index];
190
    int d;
191
    //FIXME check chroma 4mv, (no crashes ...)
192
        int uvdxy;              /* no, it might not be used uninitialized */
193
56753788
        if(dxy){
194
7986326
            if(qpel){
195
6065816
                if (h << size == 16) {
196
6065816
                    c->qpel_put[size][dxy](c->temp, ref[0] + x + y*stride, stride); //FIXME prototype (add h)
197
                } else if (size == 0 && h == 8) {
198
                    c->qpel_put[1][dxy](c->temp    , ref[0] + x + y*stride    , stride);
199
                    c->qpel_put[1][dxy](c->temp + 8, ref[0] + x + y*stride + 8, stride);
200
                } else
201
                    av_assert2(0);
202
6065816
                if(chroma){
203
                    int cx= hx/2;
204
                    int cy= hy/2;
205
                    cx= (cx>>1)|(cx&1);
206
                    cy= (cy>>1)|(cy&1);
207
                    uvdxy= (cx&1) + 2*(cy&1);
208
                    // FIXME x/y wrong, but MPEG-4 qpel is sick anyway, we should drop as much of it as possible in favor for H.264
209
                }
210
            }else{
211
1920510
                c->hpel_put[size][dxy](c->temp, ref[0] + x + y*stride, stride, h);
212
1920510
                if(chroma)
213
                    uvdxy= dxy | (x&1) | (2*(y&1));
214
            }
215
7986326
            d = cmp_func(s, c->temp, src[0], stride, h);
216
        }else{
217
48767462
            d = cmp_func(s, src[0], ref[0] + x + y*stride, stride, h);
218
48767462
            if(chroma)
219
                uvdxy= (x&1) + 2*(y&1);
220
        }
221
56753788
        if(chroma){
222
            uint8_t * const uvtemp= c->temp + 16*stride;
223
            c->hpel_put[size+1][uvdxy](uvtemp  , ref[1] + (x>>1) + (y>>1)*uvstride, uvstride, h>>1);
224
            c->hpel_put[size+1][uvdxy](uvtemp+8, ref[2] + (x>>1) + (y>>1)*uvstride, uvstride, h>>1);
225
            d += chroma_cmp_func(s, uvtemp  , src[1], uvstride, h>>1);
226
            d += chroma_cmp_func(s, uvtemp+8, src[2], uvstride, h>>1);
227
        }
228
56753788
    return d;
229
}
230
231
static int cmp_simple(MpegEncContext *s, const int x, const int y,
232
                      int ref_index, int src_index,
233
                      me_cmp_func cmp_func, me_cmp_func chroma_cmp_func){
234
    return cmp_inline(s,x,y,0,0,0,16,ref_index,src_index, cmp_func, chroma_cmp_func, 0, 0);
235
}
236
237
static int cmp_fpel_internal(MpegEncContext *s, const int x, const int y,
238
                      const int size, const int h, int ref_index, int src_index,
239
                      me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){
240
    if(flags&FLAG_DIRECT){
241
        return cmp_direct_inline(s,x,y,0,0,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, flags&FLAG_QPEL);
242
    }else{
243
        return cmp_inline(s,x,y,0,0,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, 0, flags&FLAG_CHROMA);
244
    }
245
}
246
247
51210230
static int cmp_internal(MpegEncContext *s, const int x, const int y, const int subx, const int suby,
248
                      const int size, const int h, int ref_index, int src_index,
249
                      me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){
250
51210230
    if(flags&FLAG_DIRECT){
251
1794670
        return cmp_direct_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, flags&FLAG_QPEL);
252
    }else{
253
49415560
        return cmp_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, flags&FLAG_QPEL, flags&FLAG_CHROMA);
254
    }
255
}
256
257
/** @brief compares a block (either a full macroblock or a partition thereof)
258
    against a proposed motion-compensated prediction of that block
259
 */
260
51210230
static av_always_inline int cmp(MpegEncContext *s, const int x, const int y, const int subx, const int suby,
261
                      const int size, const int h, int ref_index, int src_index,
262
                      me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){
263
    if(av_builtin_constant_p(flags) && av_builtin_constant_p(h) && av_builtin_constant_p(size)
264
       && av_builtin_constant_p(subx) && av_builtin_constant_p(suby)
265
       && flags==0 && h==16 && size==0 && subx==0 && suby==0){
266
        return cmp_simple(s,x,y,ref_index,src_index, cmp_func, chroma_cmp_func);
267
    }else if(av_builtin_constant_p(subx) && av_builtin_constant_p(suby)
268
       && subx==0 && suby==0){
269
        return cmp_fpel_internal(s,x,y,size,h,ref_index,src_index, cmp_func, chroma_cmp_func,flags);
270
    }else{
271
51210230
        return cmp_internal(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, flags);
272
    }
273
}
274
275
2331640
static int cmp_hpel(MpegEncContext *s, const int x, const int y, const int subx, const int suby,
276
                      const int size, const int h, int ref_index, int src_index,
277
                      me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){
278
2331640
    if(flags&FLAG_DIRECT){
279
598036
        return cmp_direct_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, 0);
280
    }else{
281
1733604
        return cmp_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, 0, flags&FLAG_CHROMA);
282
    }
283
}
284
285
5878032
static int cmp_qpel(MpegEncContext *s, const int x, const int y, const int subx, const int suby,
286
                      const int size, const int h, int ref_index, int src_index,
287
                      me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){
288
5878032
    if(flags&FLAG_DIRECT){
289
273408
        return cmp_direct_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, 1);
290
    }else{
291
5604624
        return cmp_inline(s,x,y,subx,suby,size,h,ref_index,src_index, cmp_func, chroma_cmp_func, 1, flags&FLAG_CHROMA);
292
    }
293
}
294
295
#include "motion_est_template.c"
296
297
static int zero_cmp(MpegEncContext *s, uint8_t *a, uint8_t *b,
298
                    ptrdiff_t stride, int h)
299
{
300
    return 0;
301
}
302
303
static void zero_hpel(uint8_t *a, const uint8_t *b, ptrdiff_t stride, int h){
304
}
305
306
9461
int ff_init_me(MpegEncContext *s){
307
9461
    MotionEstContext * const c= &s->me;
308
9461
    int cache_size= FFMIN(ME_MAP_SIZE>>ME_MAP_SHIFT, 1<<ME_MAP_SHIFT);
309
9461
    int dia_size= FFMAX(FFABS(s->avctx->dia_size)&255, FFABS(s->avctx->pre_dia_size)&255);
310
311
9461
    if(FFMIN(s->avctx->dia_size, s->avctx->pre_dia_size) < -FFMIN(ME_MAP_SIZE, MAX_SAB_SIZE)){
312
        av_log(s->avctx, AV_LOG_ERROR, "ME_MAP size is too small for SAB diamond\n");
313
        return -1;
314
    }
315
316
9461
    c->avctx= s->avctx;
317
318
9461
    if(s->codec_id == AV_CODEC_ID_H261)
319
300
        c->avctx->me_sub_cmp = c->avctx->me_cmp;
320
321

9461
    if(cache_size < 2*dia_size && !c->stride){
322
        av_log(s->avctx, AV_LOG_INFO, "ME_MAP size may be a little small for the selected diamond size\n");
323
    }
324
325
9461
    ff_set_cmp(&s->mecc, s->mecc.me_pre_cmp, c->avctx->me_pre_cmp);
326
9461
    ff_set_cmp(&s->mecc, s->mecc.me_cmp,     c->avctx->me_cmp);
327
9461
    ff_set_cmp(&s->mecc, s->mecc.me_sub_cmp, c->avctx->me_sub_cmp);
328
9461
    ff_set_cmp(&s->mecc, s->mecc.mb_cmp,     c->avctx->mb_cmp);
329
330
9461
    c->flags    = get_flags(c, 0, c->avctx->me_cmp    &FF_CMP_CHROMA);
331
9461
    c->sub_flags= get_flags(c, 0, c->avctx->me_sub_cmp&FF_CMP_CHROMA);
332
9461
    c->mb_flags = get_flags(c, 0, c->avctx->mb_cmp    &FF_CMP_CHROMA);
333
334
/*FIXME s->no_rounding b_type*/
335
9461
    if (s->avctx->flags & AV_CODEC_FLAG_QPEL) {
336
528
        c->sub_motion_search= qpel_motion_search;
337
528
        c->qpel_avg = s->qdsp.avg_qpel_pixels_tab;
338
528
        if (s->no_rounding)
339
99
            c->qpel_put = s->qdsp.put_no_rnd_qpel_pixels_tab;
340
        else
341
429
            c->qpel_put = s->qdsp.put_qpel_pixels_tab;
342
    }else{
343
8933
        if(c->avctx->me_sub_cmp&FF_CMP_CHROMA)
344
            c->sub_motion_search= hpel_motion_search;
345
8933
        else if(   c->avctx->me_sub_cmp == FF_CMP_SAD
346
7998
                && c->avctx->    me_cmp == FF_CMP_SAD
347
7998
                && c->avctx->    mb_cmp == FF_CMP_SAD)
348
7998
            c->sub_motion_search= sad_hpel_motion_search; // 2050 vs. 2450 cycles
349
        else
350
935
            c->sub_motion_search= hpel_motion_search;
351
    }
352
9461
    c->hpel_avg = s->hdsp.avg_pixels_tab;
353
9461
    if (s->no_rounding)
354
1623
        c->hpel_put = s->hdsp.put_no_rnd_pixels_tab;
355
    else
356
7838
        c->hpel_put = s->hdsp.put_pixels_tab;
357
358
9461
    if(s->linesize){
359
9461
        c->stride  = s->linesize;
360
9461
        c->uvstride= s->uvlinesize;
361
    }else{
362
        c->stride  = 16*s->mb_width + 32;
363
        c->uvstride=  8*s->mb_width + 16;
364
    }
365
366
    /* 8x8 fullpel search would need a 4x4 chroma compare, which we do
367
     * not have yet, and even if we had, the motion estimation code
368
     * does not expect it. */
369
9461
    if (s->codec_id != AV_CODEC_ID_SNOW) {
370
9461
        if ((c->avctx->me_cmp & FF_CMP_CHROMA) /* && !s->mecc.me_cmp[2] */)
371
            s->mecc.me_cmp[2] = zero_cmp;
372

9461
        if ((c->avctx->me_sub_cmp & FF_CMP_CHROMA) && !s->mecc.me_sub_cmp[2])
373
            s->mecc.me_sub_cmp[2] = zero_cmp;
374
9461
        c->hpel_put[2][0]= c->hpel_put[2][1]=
375
9461
        c->hpel_put[2][2]= c->hpel_put[2][3]= zero_hpel;
376
    }
377
378
9461
    if(s->codec_id == AV_CODEC_ID_H261){
379
300
        c->sub_motion_search= no_sub_motion_search;
380
    }
381
382
9461
    return 0;
383
}
384
385
#define CHECK_SAD_HALF_MV(suffix, x, y) \
386
{\
387
    d  = s->mecc.pix_abs[size][(x ? 1 : 0) + (y ? 2 : 0)](NULL, pix, ptr + ((x) >> 1), stride, h); \
388
    d += (mv_penalty[pen_x + x] + mv_penalty[pen_y + y])*penalty_factor;\
389
    COPY3_IF_LT(dminh, d, dx, x, dy, y)\
390
}
391
392
4215198
static int sad_hpel_motion_search(MpegEncContext * s,
393
                                  int *mx_ptr, int *my_ptr, int dmin,
394
                                  int src_index, int ref_index,
395
                                  int size, int h)
396
{
397
4215198
    MotionEstContext * const c= &s->me;
398
4215198
    const int penalty_factor= c->sub_penalty_factor;
399
    int mx, my, dminh;
400
    uint8_t *pix, *ptr;
401
4215198
    int stride= c->stride;
402
4215198
    LOAD_COMMON
403
404
    av_assert2(c->sub_flags == 0);
405
406
4215198
    if(c->skip){
407
21434
        *mx_ptr = 0;
408
21434
        *my_ptr = 0;
409
21434
        return dmin;
410
    }
411
412
4193764
    pix = c->src[src_index][0];
413
414
4193764
    mx = *mx_ptr;
415
4193764
    my = *my_ptr;
416
4193764
    ptr = c->ref[ref_index][0] + (my * stride) + mx;
417
418
4193764
    dminh = dmin;
419
420

4193764
    if (mx > xmin && mx < xmax &&
421
3977192
        my > ymin && my < ymax) {
422
3899321
        int dx=0, dy=0;
423
        int d, pen_x, pen_y;
424
3899321
        const int index= my*(1<<ME_MAP_SHIFT) + mx;
425
3899321
        const int t= score_map[(index-(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)];
426
3899321
        const int l= score_map[(index- 1               )&(ME_MAP_SIZE-1)];
427
3899321
        const int r= score_map[(index+ 1               )&(ME_MAP_SIZE-1)];
428
3899321
        const int b= score_map[(index+(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)];
429
3899321
        mx += mx;
430
3899321
        my += my;
431
432
433
3899321
        pen_x= pred_x + mx;
434
3899321
        pen_y= pred_y + my;
435
436
3899321
        ptr-= stride;
437
3899321
        if(t<=b){
438
1974834
            CHECK_SAD_HALF_MV(y2 , 0, -1)
439
1974834
            if(l<=r){
440
1232648
                CHECK_SAD_HALF_MV(xy2, -1, -1)
441
1232648
                if(t+r<=b+l){
442
719905
                    CHECK_SAD_HALF_MV(xy2, +1, -1)
443
719905
                    ptr+= stride;
444
                }else{
445
512743
                    ptr+= stride;
446
512743
                    CHECK_SAD_HALF_MV(xy2, -1, +1)
447
                }
448
1232648
                CHECK_SAD_HALF_MV(x2 , -1,  0)
449
            }else{
450
742186
                CHECK_SAD_HALF_MV(xy2, +1, -1)
451
742186
                if(t+l<=b+r){
452
409012
                    CHECK_SAD_HALF_MV(xy2, -1, -1)
453
409012
                    ptr+= stride;
454
                }else{
455
333174
                    ptr+= stride;
456
333174
                    CHECK_SAD_HALF_MV(xy2, +1, +1)
457
                }
458
742186
                CHECK_SAD_HALF_MV(x2 , +1,  0)
459
            }
460
        }else{
461
1924487
            if(l<=r){
462
726198
                if(t+l<=b+r){
463
305615
                    CHECK_SAD_HALF_MV(xy2, -1, -1)
464
305615
                    ptr+= stride;
465
                }else{
466
420583
                    ptr+= stride;
467
420583
                    CHECK_SAD_HALF_MV(xy2, +1, +1)
468
                }
469
726198
                CHECK_SAD_HALF_MV(x2 , -1,  0)
470
726198
                CHECK_SAD_HALF_MV(xy2, -1, +1)
471
            }else{
472
1198289
                if(t+r<=b+l){
473
530205
                    CHECK_SAD_HALF_MV(xy2, +1, -1)
474
530205
                    ptr+= stride;
475
                }else{
476
668084
                    ptr+= stride;
477
668084
                    CHECK_SAD_HALF_MV(xy2, -1, +1)
478
                }
479
1198289
                CHECK_SAD_HALF_MV(x2 , +1,  0)
480
1198289
                CHECK_SAD_HALF_MV(xy2, +1, +1)
481
            }
482
1924487
            CHECK_SAD_HALF_MV(y2 ,  0, +1)
483
        }
484
3899321
        mx+=dx;
485
3899321
        my+=dy;
486
487
    }else{
488
294443
        mx += mx;
489
294443
        my += my;
490
    }
491
492
4193764
    *mx_ptr = mx;
493
4193764
    *my_ptr = my;
494
4193764
    return dminh;
495
}
496
497
1764702
static inline void set_p_mv_tables(MpegEncContext * s, int mx, int my, int mv4)
498
{
499
1764702
    const int xy= s->mb_x + s->mb_y*s->mb_stride;
500
501
1764702
    s->p_mv_table[xy][0] = mx;
502
1764702
    s->p_mv_table[xy][1] = my;
503
504
    /* has already been set to the 4 MV if 4MV is done */
505
1764702
    if(mv4){
506
1512871
        int mot_xy= s->block_index[0];
507
508
1512871
        s->current_picture.motion_val[0][mot_xy    ][0] = mx;
509
1512871
        s->current_picture.motion_val[0][mot_xy    ][1] = my;
510
1512871
        s->current_picture.motion_val[0][mot_xy + 1][0] = mx;
511
1512871
        s->current_picture.motion_val[0][mot_xy + 1][1] = my;
512
513
1512871
        mot_xy += s->b8_stride;
514
1512871
        s->current_picture.motion_val[0][mot_xy    ][0] = mx;
515
1512871
        s->current_picture.motion_val[0][mot_xy    ][1] = my;
516
1512871
        s->current_picture.motion_val[0][mot_xy + 1][0] = mx;
517
1512871
        s->current_picture.motion_val[0][mot_xy + 1][1] = my;
518
    }
519
1764702
}
520
521
/**
522
 * get fullpel ME search limits.
523
 */
524
3180817
static inline void get_limits(MpegEncContext *s, int x, int y)
525
{
526
3180817
    MotionEstContext * const c= &s->me;
527
3180817
    int range= c->avctx->me_range >> (1 + !!(c->flags&FLAG_QPEL));
528
3180817
    int max_range = MAX_MV >> (1 + !!(c->flags&FLAG_QPEL));
529
/*
530
    if(c->avctx->me_range) c->range= c->avctx->me_range >> 1;
531
    else                   c->range= 16;
532
*/
533
3180817
    if (s->unrestricted_mv) {
534
1688914
        c->xmin = - x - 16;
535
1688914
        c->ymin = - y - 16;
536
1688914
        c->xmax = - x + s->width;
537
1688914
        c->ymax = - y + s->height;
538
1491903
    } else if (s->out_format == FMT_H261){
539
        // Search range of H.261 is different from other codec standards
540
106920
        c->xmin = (x > 15) ? - 15 : 0;
541
106920
        c->ymin = (y > 15) ? - 15 : 0;
542
106920
        c->xmax = (x < s->mb_width * 16 - 16) ? 15 : 0;
543
106920
        c->ymax = (y < s->mb_height * 16 - 16) ? 15 : 0;
544
    } else {
545
1384983
        c->xmin = - x;
546
1384983
        c->ymin = - y;
547
1384983
        c->xmax = - x + s->mb_width *16 - 16;
548
1384983
        c->ymax = - y + s->mb_height*16 - 16;
549
    }
550

3180817
    if(!range || range > max_range)
551
3180817
        range = max_range;
552
3180817
    if(range){
553
3180817
        c->xmin = FFMAX(c->xmin,-range);
554
3180817
        c->xmax = FFMIN(c->xmax, range);
555
3180817
        c->ymin = FFMAX(c->ymin,-range);
556
3180817
        c->ymax = FFMIN(c->ymax, range);
557
    }
558
3180817
}
559
560
251831
static inline void init_mv4_ref(MotionEstContext *c){
561
251831
    const int stride= c->stride;
562
563
251831
    c->ref[1][0] = c->ref[0][0] + 8;
564
251831
    c->ref[2][0] = c->ref[0][0] + 8*stride;
565
251831
    c->ref[3][0] = c->ref[2][0] + 8;
566
251831
    c->src[1][0] = c->src[0][0] + 8;
567
251831
    c->src[2][0] = c->src[0][0] + 8*stride;
568
251831
    c->src[3][0] = c->src[2][0] + 8;
569
251831
}
570
571
251831
static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift)
572
{
573
251831
    MotionEstContext * const c= &s->me;
574
251831
    const int size= 1;
575
251831
    const int h=8;
576
    int block;
577
    int P[10][2];
578
251831
    int dmin_sum=0, mx4_sum=0, my4_sum=0, i;
579
251831
    int same=1;
580
251831
    const int stride= c->stride;
581
251831
    uint8_t *mv_penalty= c->current_mv_penalty;
582

251831
    int safety_clipping= s->unrestricted_mv && (s->width&15) && (s->height&15);
583
584
251831
    init_mv4_ref(c);
585
586
1259155
    for(block=0; block<4; block++){
587
        int mx4, my4;
588
        int pred_x4, pred_y4;
589
        int dmin4;
590
        static const int off[4]= {2, 1, 1, -1};
591
1007324
        const int mot_stride = s->b8_stride;
592
1007324
        const int mot_xy = s->block_index[block];
593
594
1007324
        if(safety_clipping){
595
8240
            c->xmax = - 16*s->mb_x + s->width  - 8*(block &1);
596
8240
            c->ymax = - 16*s->mb_y + s->height - 8*(block>>1);
597
        }
598
599
1007324
        P_LEFT[0] = s->current_picture.motion_val[0][mot_xy - 1][0];
600
1007324
        P_LEFT[1] = s->current_picture.motion_val[0][mot_xy - 1][1];
601
602
1007324
        if(P_LEFT[0]       > (c->xmax<<shift)) P_LEFT[0]       = (c->xmax<<shift);
603
604
        /* special case for first line */
605

1007324
        if (s->first_slice_line && block<2) {
606
31178
            c->pred_x= pred_x4= P_LEFT[0];
607
31178
            c->pred_y= pred_y4= P_LEFT[1];
608
        } else {
609
976146
            P_TOP[0]      = s->current_picture.motion_val[0][mot_xy - mot_stride             ][0];
610
976146
            P_TOP[1]      = s->current_picture.motion_val[0][mot_xy - mot_stride             ][1];
611
976146
            P_TOPRIGHT[0] = s->current_picture.motion_val[0][mot_xy - mot_stride + off[block]][0];
612
976146
            P_TOPRIGHT[1] = s->current_picture.motion_val[0][mot_xy - mot_stride + off[block]][1];
613
976146
            if(P_TOP[1]      > (c->ymax<<shift)) P_TOP[1]     = (c->ymax<<shift);
614
976146
            if(P_TOPRIGHT[0] < (c->xmin<<shift)) P_TOPRIGHT[0]= (c->xmin<<shift);
615
976146
            if(P_TOPRIGHT[0] > (c->xmax<<shift)) P_TOPRIGHT[0]= (c->xmax<<shift);
616
976146
            if(P_TOPRIGHT[1] > (c->ymax<<shift)) P_TOPRIGHT[1]= (c->ymax<<shift);
617
618
976146
            P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
619
976146
            P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
620
621
976146
            c->pred_x= pred_x4 = P_MEDIAN[0];
622
976146
            c->pred_y= pred_y4 = P_MEDIAN[1];
623
        }
624
1007324
        P_MV1[0]= mx;
625
1007324
        P_MV1[1]= my;
626
1007324
        if(safety_clipping)
627
82400
            for(i=1; i<10; i++){
628


74160
                if (s->first_slice_line && block<2 && i>1 && i<9)
629
10234
                    continue;
630

63926
                if (i>4 && i<9)
631
27112
                    continue;
632
36814
                if(P[i][0] > (c->xmax<<shift)) P[i][0]= (c->xmax<<shift);
633
36814
                if(P[i][1] > (c->ymax<<shift)) P[i][1]= (c->ymax<<shift);
634
            }
635
636
1007324
        dmin4 = epzs_motion_search2(s, &mx4, &my4, P, block, block, s->p_mv_table, (1<<16)>>shift, 1);
637
638
1007324
        dmin4= c->sub_motion_search(s, &mx4, &my4, dmin4, block, block, size, h);
639
640
1007324
        if (s->mecc.me_sub_cmp[0] != s->mecc.mb_cmp[0]) {
641
            int dxy;
642
167784
            const int offset= ((block&1) + (block>>1)*stride)*8;
643
167784
            uint8_t *dest_y = c->scratchpad + offset;
644
167784
            if(s->quarter_sample){
645
56744
                uint8_t *ref= c->ref[block][0] + (mx4>>2) + (my4>>2)*stride;
646
56744
                dxy = ((my4 & 3) << 2) | (mx4 & 3);
647
648
56744
                if(s->no_rounding)
649
39304
                    s->qdsp.put_no_rnd_qpel_pixels_tab[1][dxy](dest_y, ref, stride);
650
                else
651
17440
                    s->qdsp.put_qpel_pixels_tab[1][dxy](dest_y, ref, stride);
652
            }else{
653
111040
                uint8_t *ref= c->ref[block][0] + (mx4>>1) + (my4>>1)*stride;
654
111040
                dxy = ((my4 & 1) << 1) | (mx4 & 1);
655
656
111040
                if(s->no_rounding)
657
77264
                    s->hdsp.put_no_rnd_pixels_tab[1][dxy](dest_y    , ref    , stride, h);
658
                else
659
33776
                    s->hdsp.put_pixels_tab       [1][dxy](dest_y    , ref    , stride, h);
660
            }
661
167784
            dmin_sum+= (mv_penalty[mx4-pred_x4] + mv_penalty[my4-pred_y4])*c->mb_penalty_factor;
662
        }else
663
839540
            dmin_sum+= dmin4;
664
665
1007324
        if(s->quarter_sample){
666
56744
            mx4_sum+= mx4/2;
667
56744
            my4_sum+= my4/2;
668
        }else{
669
950580
            mx4_sum+= mx4;
670
950580
            my4_sum+= my4;
671
        }
672
673
1007324
        s->current_picture.motion_val[0][s->block_index[block]][0] = mx4;
674
1007324
        s->current_picture.motion_val[0][s->block_index[block]][1] = my4;
675
676

1007324
        if(mx4 != mx || my4 != my) same=0;
677
    }
678
679
251831
    if(same)
680
41713
        return INT_MAX;
681
682
210118
    if (s->mecc.me_sub_cmp[0] != s->mecc.mb_cmp[0]) {
683
41517
        dmin_sum += s->mecc.mb_cmp[0](s,
684
41517
                                      s->new_picture.f->data[0] +
685
41517
                                      s->mb_x * 16 + s->mb_y * 16 * stride,
686
                                      c->scratchpad, stride, 16);
687
    }
688
689
210118
    if(c->avctx->mb_cmp&FF_CMP_CHROMA){
690
        int dxy;
691
        int mx, my;
692
        int offset;
693
694
        mx= ff_h263_round_chroma(mx4_sum);
695
        my= ff_h263_round_chroma(my4_sum);
696
        dxy = ((my & 1) << 1) | (mx & 1);
697
698
        offset= (s->mb_x*8 + (mx>>1)) + (s->mb_y*8 + (my>>1))*s->uvlinesize;
699
700
        if(s->no_rounding){
701
            s->hdsp.put_no_rnd_pixels_tab[1][dxy](c->scratchpad    , s->last_picture.f->data[1] + offset, s->uvlinesize, 8);
702
            s->hdsp.put_no_rnd_pixels_tab[1][dxy](c->scratchpad + 8, s->last_picture.f->data[2] + offset, s->uvlinesize, 8);
703
        }else{
704
            s->hdsp.put_pixels_tab       [1][dxy](c->scratchpad    , s->last_picture.f->data[1] + offset, s->uvlinesize, 8);
705
            s->hdsp.put_pixels_tab       [1][dxy](c->scratchpad + 8, s->last_picture.f->data[2] + offset, s->uvlinesize, 8);
706
        }
707
708
        dmin_sum += s->mecc.mb_cmp[1](s, s->new_picture.f->data[1] + s->mb_x * 8 + s->mb_y * 8 * s->uvlinesize, c->scratchpad,     s->uvlinesize, 8);
709
        dmin_sum += s->mecc.mb_cmp[1](s, s->new_picture.f->data[2] + s->mb_x * 8 + s->mb_y * 8 * s->uvlinesize, c->scratchpad + 8, s->uvlinesize, 8);
710
    }
711
712
210118
    c->pred_x= mx;
713
210118
    c->pred_y= my;
714
715
210118
    switch(c->avctx->mb_cmp&0xFF){
716
    /*case FF_CMP_SSE:
717
        return dmin_sum+ 32*s->qscale*s->qscale;*/
718
    case FF_CMP_RD:
719
        return dmin_sum;
720
210118
    default:
721
210118
        return dmin_sum+ 11*c->mb_penalty_factor;
722
    }
723
}
724
725
331233
static inline void init_interlaced_ref(MpegEncContext *s, int ref_index){
726
331233
    MotionEstContext * const c= &s->me;
727
728
331233
    c->ref[1+ref_index][0] = c->ref[0+ref_index][0] + s->linesize;
729
331233
    c->src[1][0] = c->src[0][0] + s->linesize;
730
331233
    if(c->flags & FLAG_CHROMA){
731
        c->ref[1+ref_index][1] = c->ref[0+ref_index][1] + s->uvlinesize;
732
        c->ref[1+ref_index][2] = c->ref[0+ref_index][2] + s->uvlinesize;
733
        c->src[1][1] = c->src[0][1] + s->uvlinesize;
734
        c->src[1][2] = c->src[0][2] + s->uvlinesize;
735
    }
736
331233
}
737
738
331233
static int interlaced_search(MpegEncContext *s, int ref_index,
739
                             int16_t (*mv_tables[2][2])[2], uint8_t *field_select_tables[2], int mx, int my, int user_field_select)
740
{
741
331233
    MotionEstContext * const c= &s->me;
742
331233
    const int size=0;
743
331233
    const int h=8;
744
    int block;
745
    int P[10][2];
746
331233
    uint8_t * const mv_penalty= c->current_mv_penalty;
747
331233
    int same=1;
748
331233
    const int stride= 2*s->linesize;
749
331233
    int dmin_sum= 0;
750
331233
    const int mot_stride= s->mb_stride;
751
331233
    const int xy= s->mb_x + s->mb_y*mot_stride;
752
753
331233
    c->ymin>>=1;
754
331233
    c->ymax>>=1;
755
331233
    c->stride<<=1;
756
331233
    c->uvstride<<=1;
757
331233
    init_interlaced_ref(s, ref_index);
758
759
993699
    for(block=0; block<2; block++){
760
        int field_select;
761
662466
        int best_dmin= INT_MAX;
762
662466
        int best_field= -1;
763
764
1987398
        for(field_select=0; field_select<2; field_select++){
765
            int dmin, mx_i, my_i;
766
1324932
            int16_t (*mv_table)[2]= mv_tables[block][field_select];
767
768
1324932
            if(user_field_select){
769
                av_assert1(field_select==0 || field_select==1);
770
                av_assert1(field_select_tables[block][xy]==0 || field_select_tables[block][xy]==1);
771
                if(field_select_tables[block][xy] != field_select)
772
                    continue;
773
            }
774
775
1324932
            P_LEFT[0] = mv_table[xy - 1][0];
776
1324932
            P_LEFT[1] = mv_table[xy - 1][1];
777
1324932
            if(P_LEFT[0]       > (c->xmax<<1)) P_LEFT[0]       = (c->xmax<<1);
778
779
1324932
            c->pred_x= P_LEFT[0];
780
1324932
            c->pred_y= P_LEFT[1];
781
782
1324932
            if(!s->first_slice_line){
783
1206192
                P_TOP[0]      = mv_table[xy - mot_stride][0];
784
1206192
                P_TOP[1]      = mv_table[xy - mot_stride][1];
785
1206192
                P_TOPRIGHT[0] = mv_table[xy - mot_stride + 1][0];
786
1206192
                P_TOPRIGHT[1] = mv_table[xy - mot_stride + 1][1];
787
1206192
                if(P_TOP[1]      > (c->ymax<<1)) P_TOP[1]     = (c->ymax<<1);
788
1206192
                if(P_TOPRIGHT[0] < (c->xmin<<1)) P_TOPRIGHT[0]= (c->xmin<<1);
789
1206192
                if(P_TOPRIGHT[0] > (c->xmax<<1)) P_TOPRIGHT[0]= (c->xmax<<1);
790
1206192
                if(P_TOPRIGHT[1] > (c->ymax<<1)) P_TOPRIGHT[1]= (c->ymax<<1);
791
792
1206192
                P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
793
1206192
                P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
794
            }
795
1324932
            P_MV1[0]= mx; //FIXME not correct if block != field_select
796
1324932
            P_MV1[1]= my / 2;
797
798
1324932
            dmin = epzs_motion_search2(s, &mx_i, &my_i, P, block, field_select+ref_index, mv_table, (1<<16)>>1, 0);
799
800
1324932
            dmin= c->sub_motion_search(s, &mx_i, &my_i, dmin, block, field_select+ref_index, size, h);
801
802
1324932
            mv_table[xy][0]= mx_i;
803
1324932
            mv_table[xy][1]= my_i;
804
805
1324932
            if (s->mecc.me_sub_cmp[0] != s->mecc.mb_cmp[0]) {
806
                int dxy;
807
808
                //FIXME chroma ME
809
                uint8_t *ref= c->ref[field_select+ref_index][0] + (mx_i>>1) + (my_i>>1)*stride;
810
                dxy = ((my_i & 1) << 1) | (mx_i & 1);
811
812
                if(s->no_rounding){
813
                    s->hdsp.put_no_rnd_pixels_tab[size][dxy](c->scratchpad, ref    , stride, h);
814
                }else{
815
                    s->hdsp.put_pixels_tab       [size][dxy](c->scratchpad, ref    , stride, h);
816
                }
817
                dmin = s->mecc.mb_cmp[size](s, c->src[block][0], c->scratchpad, stride, h);
818
                dmin+= (mv_penalty[mx_i-c->pred_x] + mv_penalty[my_i-c->pred_y] + 1)*c->mb_penalty_factor;
819
            }else
820
1324932
                dmin+= c->mb_penalty_factor; //field_select bits
821
822
1324932
            dmin += field_select != block; //slightly prefer same field
823
824
1324932
            if(dmin < best_dmin){
825
994986
                best_dmin= dmin;
826
994986
                best_field= field_select;
827
            }
828
        }
829
        {
830
662466
            int16_t (*mv_table)[2]= mv_tables[block][best_field];
831
832
662466
            if(mv_table[xy][0] != mx) same=0; //FIXME check if these checks work and are any good at all
833
662466
            if(mv_table[xy][1]&1) same=0;
834
662466
            if(mv_table[xy][1]*2 != my) same=0;
835
662466
            if(best_field != block) same=0;
836
        }
837
838
662466
        field_select_tables[block][xy]= best_field;
839
662466
        dmin_sum += best_dmin;
840
    }
841
842
331233
    c->ymin<<=1;
843
331233
    c->ymax<<=1;
844
331233
    c->stride>>=1;
845
331233
    c->uvstride>>=1;
846
847
331233
    if(same)
848
53962
        return INT_MAX;
849
850
277271
    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
277271
    default:
856
277271
        return dmin_sum+ 11*c->mb_penalty_factor;
857
    }
858
}
859
860
7743270
static inline int get_penalty_factor(int lambda, int lambda2, int type){
861

7743270
    switch(type&0xFF){
862
6844435
    default:
863
    case FF_CMP_SAD:
864
6844435
        return lambda>>FF_LAMBDA_SHIFT;
865
    case FF_CMP_DCT:
866
        return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
867
    case FF_CMP_W53:
868
        return (4*lambda)>>(FF_LAMBDA_SHIFT);
869
    case FF_CMP_W97:
870
        return (2*lambda)>>(FF_LAMBDA_SHIFT);
871
552905
    case FF_CMP_SATD:
872
    case FF_CMP_DCT264:
873
552905
        return (2*lambda)>>FF_LAMBDA_SHIFT;
874
345930
    case FF_CMP_RD:
875
    case FF_CMP_PSNR:
876
    case FF_CMP_SSE:
877
    case FF_CMP_NSSE:
878
345930
        return lambda2>>FF_LAMBDA_SHIFT;
879
    case FF_CMP_BIT:
880
    case FF_CMP_MEDIAN_SAD:
881
        return 1;
882
    }
883
}
884
885
1764702
void ff_estimate_p_frame_motion(MpegEncContext * s,
886
                                int mb_x, int mb_y)
887
{
888
1764702
    MotionEstContext * const c= &s->me;
889
    uint8_t *pix, *ppix;
890
1764702
    int sum, mx = 0, my = 0, dmin = 0;
891
    int varc;            ///< the variance of the block (sum of squared (p[y][x]-average))
892
    int vard;            ///< sum of squared differences with the estimated motion vector
893
    int P[10][2];
894
1764702
    const int shift= 1+s->quarter_sample;
895
1764702
    int mb_type=0;
896
1764702
    Picture * const pic= &s->current_picture;
897
898
1764702
    init_ref(c, s->new_picture.f->data, s->last_picture.f->data, NULL, 16*mb_x, 16*mb_y, 0);
899
900

1764702
    av_assert0(s->quarter_sample==0 || s->quarter_sample==1);
901
1764702
    av_assert0(s->linesize == c->stride);
902
1764702
    av_assert0(s->uvlinesize == c->uvstride);
903
904
1764702
    c->penalty_factor    = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_cmp);
905
1764702
    c->sub_penalty_factor= get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_sub_cmp);
906
1764702
    c->mb_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->mb_cmp);
907
1764702
    c->current_mv_penalty= c->mv_penalty[s->f_code] + MAX_DMV;
908
909
1764702
    get_limits(s, 16*mb_x, 16*mb_y);
910
1764702
    c->skip=0;
911
912
    /* intra / predictive decision */
913
1764702
    pix = c->src[0][0];
914
1764702
    sum  = s->mpvencdsp.pix_sum(pix, s->linesize);
915
1764702
    varc = s->mpvencdsp.pix_norm1(pix, s->linesize) -
916
1764702
           (((unsigned) sum * sum) >> 8) + 500;
917
918
1764702
    pic->mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
919
1764702
    pic->mb_var [s->mb_stride * mb_y + mb_x] = (varc+128)>>8;
920
1764702
    c->mb_var_sum_temp += (varc+128)>>8;
921
922
1764702
    if (s->motion_est != FF_ME_ZERO) {
923
1764702
        const int mot_stride = s->b8_stride;
924
1764702
        const int mot_xy = s->block_index[0];
925
926
1764702
        P_LEFT[0] = s->current_picture.motion_val[0][mot_xy - 1][0];
927
1764702
        P_LEFT[1] = s->current_picture.motion_val[0][mot_xy - 1][1];
928
929
1764702
        if (P_LEFT[0] > (c->xmax << shift))
930
19163
            P_LEFT[0] =  c->xmax << shift;
931
932
1764702
        if (!s->first_slice_line) {
933
1662583
            P_TOP[0]      = s->current_picture.motion_val[0][mot_xy - mot_stride    ][0];
934
1662583
            P_TOP[1]      = s->current_picture.motion_val[0][mot_xy - mot_stride    ][1];
935
1662583
            P_TOPRIGHT[0] = s->current_picture.motion_val[0][mot_xy - mot_stride + 2][0];
936
1662583
            P_TOPRIGHT[1] = s->current_picture.motion_val[0][mot_xy - mot_stride + 2][1];
937
1662583
            if (P_TOP[1] > (c->ymax << shift))
938
21562
                P_TOP[1] =  c->ymax << shift;
939
1662583
            if (P_TOPRIGHT[0] < (c->xmin * (1 << shift)))
940
20959
                P_TOPRIGHT[0] =  c->xmin * (1 << shift);
941
1662583
            if (P_TOPRIGHT[1] > (c->ymax * (1 << shift)))
942
19794
                P_TOPRIGHT[1] =  c->ymax * (1 << shift);
943
944
1662583
            P_MEDIAN[0] = mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
945
1662583
            P_MEDIAN[1] = mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
946
947
1662583
            if (s->out_format == FMT_H263) {
948
1078981
                c->pred_x = P_MEDIAN[0];
949
1078981
                c->pred_y = P_MEDIAN[1];
950
            } else { /* MPEG-1 at least */
951
583602
                c->pred_x = P_LEFT[0];
952
583602
                c->pred_y = P_LEFT[1];
953
            }
954
        } else {
955
102119
            c->pred_x = P_LEFT[0];
956
102119
            c->pred_y = P_LEFT[1];
957
        }
958
1764702
        dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, 0, s->p_mv_table, (1<<16)>>shift, 0, 16);
959
    }
960
961
    /* At this point (mx,my) are full-pell and the relative displacement */
962
1764702
    ppix = c->ref[0][0] + (my * s->linesize) + mx;
963
964
1764702
    vard = s->mecc.sse[0](NULL, pix, ppix, s->linesize, 16);
965
966
1764702
    pic->mc_mb_var[s->mb_stride * mb_y + mb_x] = (vard+128)>>8;
967
1764702
    c->mc_mb_var_sum_temp += (vard+128)>>8;
968
969
1764702
    if (c->avctx->mb_decision > FF_MB_DECISION_SIMPLE) {
970
362361
        int p_score= FFMIN(vard, varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*100);
971
362361
        int i_score= varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*20;
972
362361
        c->scene_change_score+= ff_sqrt(p_score) - ff_sqrt(i_score);
973
974

362361
        if (vard*2 + 200*256 > varc && !s->intra_penalty)
975
97665
            mb_type|= CANDIDATE_MB_TYPE_INTRA;
976

362361
        if (varc*2 + 200*256 > vard || s->qscale > 24){
977
//        if (varc*2 + 200*256 + 50*(s->lambda2>>FF_LAMBDA_SHIFT) > vard){
978
360249
            mb_type|= CANDIDATE_MB_TYPE_INTER;
979
360249
            c->sub_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
980
360249
            if (s->mpv_flags & FF_MPV_FLAG_MV0)
981

61151
                if(mx || my)
982
60625
                    mb_type |= CANDIDATE_MB_TYPE_SKIPPED; //FIXME check difference
983
        }else{
984
2112
            mx <<=shift;
985
2112
            my <<=shift;
986
        }
987
362361
        if ((s->avctx->flags & AV_CODEC_FLAG_4MV)
988

277722
           && !c->skip && varc>50<<8 && vard>10<<8){
989
251831
            if(h263_mv4_search(s, mx, my, shift) < INT_MAX)
990
210118
                mb_type|=CANDIDATE_MB_TYPE_INTER4V;
991
992
251831
            set_p_mv_tables(s, mx, my, 0);
993
        }else
994
110530
            set_p_mv_tables(s, mx, my, 1);
995
362361
        if ((s->avctx->flags & AV_CODEC_FLAG_INTERLACED_ME)
996
15600
           && !c->skip){ //FIXME varc/d checks
997
15600
            if(interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my, 0) < INT_MAX)
998
13629
                mb_type |= CANDIDATE_MB_TYPE_INTER_I;
999
        }
1000
    }else{
1001
        int intra_score, i;
1002
1402341
        mb_type= CANDIDATE_MB_TYPE_INTER;
1003
1004
1402341
        dmin= c->sub_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
1005

1402341
        if(c->avctx->me_sub_cmp != c->avctx->mb_cmp && !c->skip)
1006
            dmin= get_mb_score(s, mx, my, 0, 0, 0, 16, 1);
1007
1008
1402341
        if ((s->avctx->flags & AV_CODEC_FLAG_4MV)
1009
           && !c->skip && varc>50<<8 && vard>10<<8){
1010
            int dmin4= h263_mv4_search(s, mx, my, shift);
1011
            if(dmin4 < dmin){
1012
                mb_type= CANDIDATE_MB_TYPE_INTER4V;
1013
                dmin=dmin4;
1014
            }
1015
        }
1016
1402341
        if ((s->avctx->flags & AV_CODEC_FLAG_INTERLACED_ME)
1017
85236
           && !c->skip){ //FIXME varc/d checks
1018
85233
            int dmin_i= interlaced_search(s, 0, s->p_field_mv_table, s->p_field_select_table, mx, my, 0);
1019
85233
            if(dmin_i < dmin){
1020
11208
                mb_type = CANDIDATE_MB_TYPE_INTER_I;
1021
11208
                dmin= dmin_i;
1022
            }
1023
        }
1024
1025
1402341
        set_p_mv_tables(s, mx, my, mb_type!=CANDIDATE_MB_TYPE_INTER4V);
1026
1027
        /* get intra luma score */
1028
1402341
        if((c->avctx->mb_cmp&0xFF)==FF_CMP_SSE){
1029
            intra_score= varc - 500;
1030
        }else{
1031
1402341
            unsigned mean = (sum+128)>>8;
1032
1402341
            mean*= 0x01010101;
1033
1034
23839797
            for(i=0; i<16; i++){
1035
22437456
                *(uint32_t*)(&c->scratchpad[i*s->linesize+ 0]) = mean;
1036
22437456
                *(uint32_t*)(&c->scratchpad[i*s->linesize+ 4]) = mean;
1037
22437456
                *(uint32_t*)(&c->scratchpad[i*s->linesize+ 8]) = mean;
1038
22437456
                *(uint32_t*)(&c->scratchpad[i*s->linesize+12]) = mean;
1039
            }
1040
1041
1402341
            intra_score= s->mecc.mb_cmp[0](s, c->scratchpad, pix, s->linesize, 16);
1042
        }
1043
1402341
        intra_score += c->mb_penalty_factor*16 + s->intra_penalty;
1044
1045
1402341
        if(intra_score < dmin){
1046
69978
            mb_type= CANDIDATE_MB_TYPE_INTRA;
1047
69978
            s->current_picture.mb_type[mb_y*s->mb_stride + mb_x] = CANDIDATE_MB_TYPE_INTRA; //FIXME cleanup
1048
        }else
1049
1332363
            s->current_picture.mb_type[mb_y*s->mb_stride + mb_x] = 0;
1050
1051
        {
1052
1402341
            int p_score= FFMIN(vard, varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*100);
1053
1402341
            int i_score= varc-500+(s->lambda2>>FF_LAMBDA_SHIFT)*20;
1054
1402341
            c->scene_change_score+= ff_sqrt(p_score) - ff_sqrt(i_score);
1055
        }
1056
    }
1057
1058
1764702
    s->mb_type[mb_y*s->mb_stride + mb_x]= mb_type;
1059
1764702
}
1060
1061
int ff_pre_estimate_p_frame_motion(MpegEncContext * s,
1062
                                    int mb_x, int mb_y)
1063
{
1064
    MotionEstContext * const c= &s->me;
1065
    int mx, my, dmin;
1066
    int P[10][2];
1067
    const int shift= 1+s->quarter_sample;
1068
    const int xy= mb_x + mb_y*s->mb_stride;
1069
    init_ref(c, s->new_picture.f->data, s->last_picture.f->data, NULL, 16*mb_x, 16*mb_y, 0);
1070
1071
    av_assert0(s->quarter_sample==0 || s->quarter_sample==1);
1072
1073
    c->pre_penalty_factor    = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_pre_cmp);
1074
    c->current_mv_penalty= c->mv_penalty[s->f_code] + MAX_DMV;
1075
1076
    get_limits(s, 16*mb_x, 16*mb_y);
1077
    c->skip=0;
1078
1079
    P_LEFT[0]       = s->p_mv_table[xy + 1][0];
1080
    P_LEFT[1]       = s->p_mv_table[xy + 1][1];
1081
1082
    if(P_LEFT[0]       < (c->xmin<<shift)) P_LEFT[0]       = (c->xmin<<shift);
1083
1084
    /* special case for first line */
1085
    if (s->first_slice_line) {
1086
        c->pred_x= P_LEFT[0];
1087
        c->pred_y= P_LEFT[1];
1088
        P_TOP[0]= P_TOPRIGHT[0]= P_MEDIAN[0]=
1089
        P_TOP[1]= P_TOPRIGHT[1]= P_MEDIAN[1]= 0; //FIXME
1090
    } else {
1091
        P_TOP[0]      = s->p_mv_table[xy + s->mb_stride    ][0];
1092
        P_TOP[1]      = s->p_mv_table[xy + s->mb_stride    ][1];
1093
        P_TOPRIGHT[0] = s->p_mv_table[xy + s->mb_stride - 1][0];
1094
        P_TOPRIGHT[1] = s->p_mv_table[xy + s->mb_stride - 1][1];
1095
        if(P_TOP[1]      < (c->ymin<<shift)) P_TOP[1]     = (c->ymin<<shift);
1096
        if(P_TOPRIGHT[0] > (c->xmax<<shift)) P_TOPRIGHT[0]= (c->xmax<<shift);
1097
        if(P_TOPRIGHT[1] < (c->ymin<<shift)) P_TOPRIGHT[1]= (c->ymin<<shift);
1098
1099
        P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
1100
        P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
1101
1102
        c->pred_x = P_MEDIAN[0];
1103
        c->pred_y = P_MEDIAN[1];
1104
    }
1105
1106
    dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, 0, s->p_mv_table, (1<<16)>>shift, 0, 16);
1107
1108
    s->p_mv_table[xy][0] = mx<<shift;
1109
    s->p_mv_table[xy][1] = my<<shift;
1110
1111
    return dmin;
1112
}
1113
1114
816388
static int estimate_motion_b(MpegEncContext *s, int mb_x, int mb_y,
1115
                             int16_t (*mv_table)[2], int ref_index, int f_code)
1116
{
1117
816388
    MotionEstContext * const c= &s->me;
1118
816388
    int mx = 0, my = 0, dmin = 0;
1119
    int P[10][2];
1120
816388
    const int shift= 1+s->quarter_sample;
1121
816388
    const int mot_stride = s->mb_stride;
1122
816388
    const int mot_xy = mb_y*mot_stride + mb_x;
1123
816388
    uint8_t * const mv_penalty= c->mv_penalty[f_code] + MAX_DMV;
1124
    int mv_scale;
1125
1126
816388
    c->penalty_factor    = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_cmp);
1127
816388
    c->sub_penalty_factor= get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_sub_cmp);
1128
816388
    c->mb_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->mb_cmp);
1129
816388
    c->current_mv_penalty= mv_penalty;
1130
1131
816388
    get_limits(s, 16*mb_x, 16*mb_y);
1132
1133
816388
    if (s->motion_est != FF_ME_ZERO) {
1134
816388
        P_LEFT[0] = mv_table[mot_xy - 1][0];
1135
816388
        P_LEFT[1] = mv_table[mot_xy - 1][1];
1136
1137
816388
        if (P_LEFT[0] > (c->xmax << shift)) P_LEFT[0] = (c->xmax << shift);
1138
1139
        /* special case for first line */
1140
816388
        if (!s->first_slice_line) {
1141
756196
            P_TOP[0]      = mv_table[mot_xy - mot_stride    ][0];
1142
756196
            P_TOP[1]      = mv_table[mot_xy - mot_stride    ][1];
1143
756196
            P_TOPRIGHT[0] = mv_table[mot_xy - mot_stride + 1][0];
1144
756196
            P_TOPRIGHT[1] = mv_table[mot_xy - mot_stride + 1][1];
1145
756196
            if (P_TOP[1] > (c->ymax << shift)) P_TOP[1] = (c->ymax << shift);
1146
756196
            if (P_TOPRIGHT[0] < (c->xmin << shift)) P_TOPRIGHT[0] = (c->xmin << shift);
1147
756196
            if (P_TOPRIGHT[1] > (c->ymax << shift)) P_TOPRIGHT[1] = (c->ymax << shift);
1148
1149
756196
            P_MEDIAN[0] = mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
1150
756196
            P_MEDIAN[1] = mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
1151
        }
1152
816388
        c->pred_x = P_LEFT[0];
1153
816388
        c->pred_y = P_LEFT[1];
1154
1155
816388
        if(mv_table == s->b_forw_mv_table){
1156
408194
            mv_scale= (s->pb_time<<16) / (s->pp_time<<shift);
1157
        }else{
1158
408194
            mv_scale= ((s->pb_time - s->pp_time)<<16) / (s->pp_time<<shift);
1159
        }
1160
1161
816388
        dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, ref_index, s->p_mv_table, mv_scale, 0, 16);
1162
    }
1163
1164
816388
    dmin= c->sub_motion_search(s, &mx, &my, dmin, 0, ref_index, 0, 16);
1165
1166

816388
    if(c->avctx->me_sub_cmp != c->avctx->mb_cmp && !c->skip)
1167
306340
        dmin= get_mb_score(s, mx, my, 0, ref_index, 0, 16, 1);
1168
1169
//    s->mb_type[mb_y*s->mb_width + mb_x]= mb_type;
1170
816388
    mv_table[mot_xy][0]= mx;
1171
816388
    mv_table[mot_xy][1]= my;
1172
1173
816388
    return dmin;
1174
}
1175
1176
5444630
static inline int check_bidir_mv(MpegEncContext * s,
1177
                   int motion_fx, int motion_fy,
1178
                   int motion_bx, int motion_by,
1179
                   int pred_fx, int pred_fy,
1180
                   int pred_bx, int pred_by,
1181
                   int size, int h)
1182
{
1183
    //FIXME optimize?
1184
    //FIXME better f_code prediction (max mv & distance)
1185
    //FIXME pointers
1186
5444630
    MotionEstContext * const c= &s->me;
1187
5444630
    uint8_t * const mv_penalty_f= c->mv_penalty[s->f_code] + MAX_DMV; // f_code of the prev frame
1188
5444630
    uint8_t * const mv_penalty_b= c->mv_penalty[s->b_code] + MAX_DMV; // f_code of the prev frame
1189
5444630
    int stride= c->stride;
1190
5444630
    uint8_t *dest_y = c->scratchpad;
1191
    uint8_t *ptr;
1192
    int dxy;
1193
    int src_x, src_y;
1194
    int fbmin;
1195
5444630
    uint8_t **src_data= c->src[0];
1196
5444630
    uint8_t **ref_data= c->ref[0];
1197
5444630
    uint8_t **ref2_data= c->ref[2];
1198
1199
5444630
    if(s->quarter_sample){
1200
625749
        dxy = ((motion_fy & 3) << 2) | (motion_fx & 3);
1201
625749
        src_x = motion_fx >> 2;
1202
625749
        src_y = motion_fy >> 2;
1203
1204
625749
        ptr = ref_data[0] + (src_y * stride) + src_x;
1205
625749
        s->qdsp.put_qpel_pixels_tab[0][dxy](dest_y, ptr, stride);
1206
1207
625749
        dxy = ((motion_by & 3) << 2) | (motion_bx & 3);
1208
625749
        src_x = motion_bx >> 2;
1209
625749
        src_y = motion_by >> 2;
1210
1211
625749
        ptr = ref2_data[0] + (src_y * stride) + src_x;
1212
625749
        s->qdsp.avg_qpel_pixels_tab[size][dxy](dest_y, ptr, stride);
1213
    }else{
1214
4818881
        dxy = ((motion_fy & 1) << 1) | (motion_fx & 1);
1215
4818881
        src_x = motion_fx >> 1;
1216
4818881
        src_y = motion_fy >> 1;
1217
1218
4818881
        ptr = ref_data[0] + (src_y * stride) + src_x;
1219
4818881
        s->hdsp.put_pixels_tab[size][dxy](dest_y    , ptr    , stride, h);
1220
1221
4818881
        dxy = ((motion_by & 1) << 1) | (motion_bx & 1);
1222
4818881
        src_x = motion_bx >> 1;
1223
4818881
        src_y = motion_by >> 1;
1224
1225
4818881
        ptr = ref2_data[0] + (src_y * stride) + src_x;
1226
4818881
        s->hdsp.avg_pixels_tab[size][dxy](dest_y    , ptr    , stride, h);
1227
    }
1228
1229
10889260
    fbmin = (mv_penalty_f[motion_fx-pred_fx] + mv_penalty_f[motion_fy-pred_fy])*c->mb_penalty_factor
1230
5444630
           +(mv_penalty_b[motion_bx-pred_bx] + mv_penalty_b[motion_by-pred_by])*c->mb_penalty_factor
1231
5444630
           + s->mecc.mb_cmp[size](s, src_data[0], dest_y, stride, h); // FIXME new_pic
1232
1233
5444630
    if(c->avctx->mb_cmp&FF_CMP_CHROMA){
1234
    }
1235
    //FIXME CHROMA !!!
1236
1237
5444630
    return fbmin;
1238
}
1239
1240
/* refine the bidir vectors in hq mode and return the score in both lq & hq mode*/
1241
408194
static inline int bidir_refine(MpegEncContext * s, int mb_x, int mb_y)
1242
{
1243
408194
    MotionEstContext * const c= &s->me;
1244
408194
    const int mot_stride = s->mb_stride;
1245
408194
    const int xy = mb_y *mot_stride + mb_x;
1246
    int fbmin;
1247
408194
    int pred_fx= s->b_bidir_forw_mv_table[xy-1][0];
1248
408194
    int pred_fy= s->b_bidir_forw_mv_table[xy-1][1];
1249
408194
    int pred_bx= s->b_bidir_back_mv_table[xy-1][0];
1250
408194
    int pred_by= s->b_bidir_back_mv_table[xy-1][1];
1251
408194
    int motion_fx= s->b_bidir_forw_mv_table[xy][0]= s->b_forw_mv_table[xy][0];
1252
408194
    int motion_fy= s->b_bidir_forw_mv_table[xy][1]= s->b_forw_mv_table[xy][1];
1253
408194
    int motion_bx= s->b_bidir_back_mv_table[xy][0]= s->b_back_mv_table[xy][0];
1254
408194
    int motion_by= s->b_bidir_back_mv_table[xy][1]= s->b_back_mv_table[xy][1];
1255
408194
    const int flags= c->sub_flags;
1256
408194
    const int qpel= flags&FLAG_QPEL;
1257
408194
    const int shift= 1+qpel;
1258
408194
    const int xmin= c->xmin<<shift;
1259
408194
    const int ymin= c->ymin<<shift;
1260
408194
    const int xmax= c->xmax<<shift;
1261
408194
    const int ymax= c->ymax<<shift;
1262
#define HASH(fx,fy,bx,by) ((fx)+17*(fy)+63*(bx)+117*(by))
1263
#define HASH8(fx,fy,bx,by) ((uint8_t)HASH(fx,fy,bx,by))
1264
408194
    int hashidx= HASH(motion_fx,motion_fy, motion_bx, motion_by);
1265
408194
    uint8_t map[256] = { 0 };
1266
1267
408194
    map[hashidx&255] = 1;
1268
1269
408194
    fbmin= check_bidir_mv(s, motion_fx, motion_fy,
1270
                          motion_bx, motion_by,
1271
                          pred_fx, pred_fy,
1272
                          pred_bx, pred_by,
1273
                          0, 16);
1274
1275
408194
    if(s->avctx->bidir_refine){
1276
        int end;
1277
        static const uint8_t limittab[5]={0,8,32,64,80};
1278
408194
        const int limit= limittab[s->avctx->bidir_refine];
1279
        static const int8_t vect[][4]={
1280
{ 0, 0, 0, 1}, { 0, 0, 0,-1}, { 0, 0, 1, 0}, { 0, 0,-1, 0}, { 0, 1, 0, 0}, { 0,-1, 0, 0}, { 1, 0, 0, 0}, {-1, 0, 0, 0},
1281
1282
{ 0, 0, 1, 1}, { 0, 0,-1,-1}, { 0, 1, 1, 0}, { 0,-1,-1, 0}, { 1, 1, 0, 0}, {-1,-1, 0, 0}, { 1, 0, 0, 1}, {-1, 0, 0,-1},
1283
{ 0, 1, 0, 1}, { 0,-1, 0,-1}, { 1, 0, 1, 0}, {-1, 0,-1, 0},
1284
{ 0, 0,-1, 1}, { 0, 0, 1,-1}, { 0,-1, 1, 0}, { 0, 1,-1, 0}, {-1, 1, 0, 0}, { 1,-1, 0, 0}, { 1, 0, 0,-1}, {-1, 0, 0, 1},
1285
{ 0,-1, 0, 1}, { 0, 1, 0,-1}, {-1, 0, 1, 0}, { 1, 0,-1, 0},
1286
1287
{ 0, 1, 1, 1}, { 0,-1,-1,-1}, { 1, 1, 1, 0}, {-1,-1,-1, 0}, { 1, 1, 0, 1}, {-1,-1, 0,-1}, { 1, 0, 1, 1}, {-1, 0,-1,-1},
1288
{ 0,-1, 1, 1}, { 0, 1,-1,-1}, {-1, 1, 1, 0}, { 1,-1,-1, 0}, { 1, 1, 0,-1}, {-1,-1, 0, 1}, { 1, 0,-1, 1}, {-1, 0, 1,-1},
1289
{ 0, 1,-1, 1}, { 0,-1, 1,-1}, { 1,-1, 1, 0}, {-1, 1,-1, 0}, {-1, 1, 0, 1}, { 1,-1, 0,-1}, { 1, 0, 1,-1}, {-1, 0,-1, 1},
1290
{ 0, 1, 1,-1}, { 0,-1,-1, 1}, { 1, 1,-1, 0}, {-1,-1, 1, 0}, { 1,-1, 0, 1}, {-1, 1, 0,-1}, {-1, 0, 1, 1}, { 1, 0,-1,-1},
1291
1292
{ 1, 1, 1, 1}, {-1,-1,-1,-1},
1293
{ 1, 1, 1,-1}, {-1,-1,-1, 1}, { 1, 1,-1, 1}, {-1,-1, 1,-1}, { 1,-1, 1, 1}, {-1, 1,-1,-1}, {-1, 1, 1, 1}, { 1,-1,-1,-1},
1294
{ 1, 1,-1,-1}, {-1,-1, 1, 1}, { 1,-1,-1, 1}, {-1, 1, 1,-1}, { 1,-1, 1,-1}, {-1, 1,-1, 1},
1295
        };
1296
        static const uint8_t hash[]={
1297
HASH8( 0, 0, 0, 1), HASH8( 0, 0, 0,-1), HASH8( 0, 0, 1, 0), HASH8( 0, 0,-1, 0), HASH8( 0, 1, 0, 0), HASH8( 0,-1, 0, 0), HASH8( 1, 0, 0, 0), HASH8(-1, 0, 0, 0),
1298
1299
HASH8( 0, 0, 1, 1), HASH8( 0, 0,-1,-1), HASH8( 0, 1, 1, 0), HASH8( 0,-1,-1, 0), HASH8( 1, 1, 0, 0), HASH8(-1,-1, 0, 0), HASH8( 1, 0, 0, 1), HASH8(-1, 0, 0,-1),
1300
HASH8( 0, 1, 0, 1), HASH8( 0,-1, 0,-1), HASH8( 1, 0, 1, 0), HASH8(-1, 0,-1, 0),
1301
HASH8( 0, 0,-1, 1), HASH8( 0, 0, 1,-1), HASH8( 0,-1, 1, 0), HASH8( 0, 1,-1, 0), HASH8(-1, 1, 0, 0), HASH8( 1,-1, 0, 0), HASH8( 1, 0, 0,-1), HASH8(-1, 0, 0, 1),
1302
HASH8( 0,-1, 0, 1), HASH8( 0, 1, 0,-1), HASH8(-1, 0, 1, 0), HASH8( 1, 0,-1, 0),
1303
1304
HASH8( 0, 1, 1, 1), HASH8( 0,-1,-1,-1), HASH8( 1, 1, 1, 0), HASH8(-1,-1,-1, 0), HASH8( 1, 1, 0, 1), HASH8(-1,-1, 0,-1), HASH8( 1, 0, 1, 1), HASH8(-1, 0,-1,-1),
1305
HASH8( 0,-1, 1, 1), HASH8( 0, 1,-1,-1), HASH8(-1, 1, 1, 0), HASH8( 1,-1,-1, 0), HASH8( 1, 1, 0,-1), HASH8(-1,-1, 0, 1), HASH8( 1, 0,-1, 1), HASH8(-1, 0, 1,-1),
1306
HASH8( 0, 1,-1, 1), HASH8( 0,-1, 1,-1), HASH8( 1,-1, 1, 0), HASH8(-1, 1,-1, 0), HASH8(-1, 1, 0, 1), HASH8( 1,-1, 0,-1), HASH8( 1, 0, 1,-1), HASH8(-1, 0,-1, 1),
1307
HASH8( 0, 1, 1,-1), HASH8( 0,-1,-1, 1), HASH8( 1, 1,-1, 0), HASH8(-1,-1, 1, 0), HASH8( 1,-1, 0, 1), HASH8(-1, 1, 0,-1), HASH8(-1, 0, 1, 1), HASH8( 1, 0,-1,-1),
1308
1309
HASH8( 1, 1, 1, 1), HASH8(-1,-1,-1,-1),
1310
HASH8( 1, 1, 1,-1), HASH8(-1,-1,-1, 1), HASH8( 1, 1,-1, 1), HASH8(-1,-1, 1,-1), HASH8( 1,-1, 1, 1), HASH8(-1, 1,-1,-1), HASH8(-1, 1, 1, 1), HASH8( 1,-1,-1,-1),
1311
HASH8( 1, 1,-1,-1), HASH8(-1,-1, 1, 1), HASH8( 1,-1,-1, 1), HASH8(-1, 1, 1,-1), HASH8( 1,-1, 1,-1), HASH8(-1, 1,-1, 1),
1312
};
1313
1314
#define CHECK_BIDIR(fx,fy,bx,by)\
1315
    if( !map[(hashidx+HASH(fx,fy,bx,by))&255]\
1316
       &&(fx<=0 || motion_fx+fx<=xmax) && (fy<=0 || motion_fy+fy<=ymax) && (bx<=0 || motion_bx+bx<=xmax) && (by<=0 || motion_by+by<=ymax)\
1317
       &&(fx>=0 || motion_fx+fx>=xmin) && (fy>=0 || motion_fy+fy>=ymin) && (bx>=0 || motion_bx+bx>=xmin) && (by>=0 || motion_by+by>=ymin)){\
1318
        int score;\
1319
        map[(hashidx+HASH(fx,fy,bx,by))&255] = 1;\
1320
        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);\
1321
        if(score < fbmin){\
1322
            hashidx += HASH(fx,fy,bx,by);\
1323
            fbmin= score;\
1324
            motion_fx+=fx;\
1325
            motion_fy+=fy;\
1326
            motion_bx+=bx;\
1327
            motion_by+=by;\
1328
            end=0;\
1329
        }\
1330
    }
1331
#define CHECK_BIDIR2(a,b,c,d)\
1332
CHECK_BIDIR(a,b,c,d)\
1333
CHECK_BIDIR(-(a),-(b),-(c),-(d))
1334
1335
        do{
1336
            int i;
1337
829046
            int borderdist=0;
1338
829046
            end=1;
1339
1340



829046
            CHECK_BIDIR2(0,0,0,1)
1341



829046
            CHECK_BIDIR2(0,0,1,0)
1342



829046
            CHECK_BIDIR2(0,1,0,0)
1343



829046
            CHECK_BIDIR2(1,0,0,0)
1344
1345
829046
            for(i=8; i<limit; i++){
1346
                int fx= motion_fx+vect[i][0];
1347
                int fy= motion_fy+vect[i][1];
1348
                int bx= motion_bx+vect[i][2];
1349
                int by= motion_by+vect[i][3];
1350
                if(borderdist<=0){
1351
                    int a= (xmax - FFMAX(fx,bx))|(FFMIN(fx,bx) - xmin);
1352
                    int b= (ymax - FFMAX(fy,by))|(FFMIN(fy,by) - ymin);
1353
                    if((a|b) < 0)
1354
                        map[(hashidx+hash[i])&255] = 1;
1355
                }
1356
                if(!map[(hashidx+hash[i])&255]){
1357
                    int score;
1358
                    map[(hashidx+hash[i])&255] = 1;
1359
                    score= check_bidir_mv(s, fx, fy, bx, by, pred_fx, pred_fy, pred_bx, pred_by, 0, 16);
1360
                    if(score < fbmin){
1361
                        hashidx += hash[i];
1362
                        fbmin= score;
1363
                        motion_fx=fx;
1364
                        motion_fy=fy;
1365
                        motion_bx=bx;
1366
                        motion_by=by;
1367
                        end=0;
1368
                        borderdist--;
1369
                        if(borderdist<=0){
1370
                            int a= FFMIN(xmax - FFMAX(fx,bx), FFMIN(fx,bx) - xmin);
1371
                            int b= FFMIN(ymax - FFMAX(fy,by), FFMIN(fy,by) - ymin);
1372
                            borderdist= FFMIN(a,b);
1373
                        }
1374
                    }
1375
                }
1376
            }
1377
829046
        }while(!end);
1378
    }
1379
1380
408194
    s->b_bidir_forw_mv_table[xy][0]= motion_fx;
1381
408194
    s->b_bidir_forw_mv_table[xy][1]= motion_fy;
1382
408194
    s->b_bidir_back_mv_table[xy][0]= motion_bx;
1383
408194
    s->b_bidir_back_mv_table[xy][1]= motion_by;
1384
1385
408194
    return fbmin;
1386
}
1387
1388
191520
static inline int direct_search(MpegEncContext * s, int mb_x, int mb_y)
1389
{
1390
191520
    MotionEstContext * const c= &s->me;
1391
    int P[10][2];
1392
191520
    const int mot_stride = s->mb_stride;
1393
191520
    const int mot_xy = mb_y*mot_stride + mb_x;
1394
191520
    const int shift= 1+s->quarter_sample;
1395
    int dmin, i;
1396
191520
    const int time_pp= s->pp_time;
1397
191520
    const int time_pb= s->pb_time;
1398
    int mx, my, xmin, xmax, ymin, ymax;
1399
191520
    int16_t (*mv_table)[2]= s->b_direct_mv_table;
1400
1401
191520
    c->current_mv_penalty= c->mv_penalty[1] + MAX_DMV;
1402
191520
    ymin= xmin=(-32)>>shift;
1403
191520
    ymax= xmax=   31>>shift;
1404
1405
191520
    if (IS_8X8(s->next_picture.mb_type[mot_xy])) {
1406
54358
        s->mv_type= MV_TYPE_8X8;
1407
    }else{
1408
137162
        s->mv_type= MV_TYPE_16X16;
1409
    }
1410
1411
408952
    for(i=0; i<4; i++){
1412
354594
        int index= s->block_index[i];
1413
        int min, max;
1414
1415
354594
        c->co_located_mv[i][0] = s->next_picture.motion_val[0][index][0];
1416
354594
        c->co_located_mv[i][1] = s->next_picture.motion_val[0][index][1];
1417
354594
        c->direct_basis_mv[i][0]= c->co_located_mv[i][0]*time_pb/time_pp + ((i& 1)<<(shift+3));
1418
354594
        c->direct_basis_mv[i][1]= c->co_located_mv[i][1]*time_pb/time_pp + ((i>>1)<<(shift+3));
1419
//        c->direct_basis_mv[1][i][0]= c->co_located_mv[i][0]*(time_pb - time_pp)/time_pp + ((i &1)<<(shift+3);
1420
//        c->direct_basis_mv[1][i][1]= c->co_located_mv[i][1]*(time_pb - time_pp)/time_pp + ((i>>1)<<(shift+3);
1421
1422
354594
        max= FFMAX(c->direct_basis_mv[i][0], c->direct_basis_mv[i][0] - c->co_located_mv[i][0])>>shift;
1423
354594
        min= FFMIN(c->direct_basis_mv[i][0], c->direct_basis_mv[i][0] - c->co_located_mv[i][0])>>shift;
1424
354594
        max+= 16*mb_x + 1; // +-1 is for the simpler rounding
1425
354594
        min+= 16*mb_x - 1;
1426
354594
        xmax= FFMIN(xmax, s->width - max);
1427
354594
        xmin= FFMAX(xmin, - 16     - min);
1428
1429
354594
        max= FFMAX(c->direct_basis_mv[i][1], c->direct_basis_mv[i][1] - c->co_located_mv[i][1])>>shift;
1430
354594
        min= FFMIN(c->direct_basis_mv[i][1], c->direct_basis_mv[i][1] - c->co_located_mv[i][1])>>shift;
1431
354594
        max+= 16*mb_y + 1; // +-1 is for the simpler rounding
1432
354594
        min+= 16*mb_y - 1;
1433
354594
        ymax= FFMIN(ymax, s->height - max);
1434
354594
        ymin= FFMAX(ymin, - 16      - min);
1435
1436
354594
        if(s->mv_type == MV_TYPE_16X16) break;
1437
    }
1438
1439
    av_assert2(xmax <= 15 && ymax <= 15 && xmin >= -16 && ymin >= -16);
1440
1441


191520
    if(xmax < 0 || xmin >0 || ymax < 0 || ymin > 0){
1442
105
        s->b_direct_mv_table[mot_xy][0]= 0;
1443
105
        s->b_direct_mv_table[mot_xy][1]= 0;
1444
1445
105
        return 256*256*256*64;
1446
    }
1447
1448
191415
    c->xmin= xmin;
1449
191415
    c->ymin= ymin;
1450
191415
    c->xmax= xmax;
1451
191415
    c->ymax= ymax;
1452
191415
    c->flags     |= FLAG_DIRECT;
1453
191415
    c->sub_flags |= FLAG_DIRECT;
1454
191415
    c->pred_x=0;
1455
191415
    c->pred_y=0;
1456
1457
191415
    P_LEFT[0]        = av_clip(mv_table[mot_xy - 1][0], xmin<<shift, xmax<<shift);
1458
191415
    P_LEFT[1]        = av_clip(mv_table[mot_xy - 1][1], ymin<<shift, ymax<<shift);
1459
1460
    /* special case for first line */
1461
191415
    if (!s->first_slice_line) { //FIXME maybe allow this over thread boundary as it is clipped
1462
178176
        P_TOP[0]      = av_clip(mv_table[mot_xy - mot_stride             ][0], xmin<<shift, xmax<<shift);
1463
178176
        P_TOP[1]      = av_clip(mv_table[mot_xy - mot_stride             ][1], ymin<<shift, ymax<<shift);
1464
178176
        P_TOPRIGHT[0] = av_clip(mv_table[mot_xy - mot_stride + 1         ][0], xmin<<shift, xmax<<shift);
1465
178176
        P_TOPRIGHT[1] = av_clip(mv_table[mot_xy - mot_stride + 1         ][1], ymin<<shift, ymax<<shift);
1466
1467
178176
        P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
1468
178176
        P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
1469
    }
1470
1471
191415
    dmin = ff_epzs_motion_search(s, &mx, &my, P, 0, 0, mv_table, 1<<(16-shift), 0, 16);
1472
191415
    if(c->sub_flags&FLAG_QPEL)
1473
38282
        dmin = qpel_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
1474
    else
1475
153133
        dmin = hpel_motion_search(s, &mx, &my, dmin, 0, 0, 0, 16);
1476
1477

191415
    if(c->avctx->me_sub_cmp != c->avctx->mb_cmp && !c->skip)
1478
114819
        dmin= get_mb_score(s, mx, my, 0, 0, 0, 16, 1);
1479
1480
191415
    get_limits(s, 16*mb_x, 16*mb_y); //restore c->?min/max, maybe not needed
1481
1482
191415
    mv_table[mot_xy][0]= mx;
1483
191415
    mv_table[mot_xy][1]= my;
1484
191415
    c->flags     &= ~FLAG_DIRECT;
1485
191415
    c->sub_flags &= ~FLAG_DIRECT;
1486
1487
191415
    return dmin;
1488
}
1489
1490
408312
void ff_estimate_b_frame_motion(MpegEncContext * s,
1491
                             int mb_x, int mb_y)
1492
{
1493
408312
    MotionEstContext * const c= &s->me;
1494
408312
    const int penalty_factor= c->mb_penalty_factor;
1495
    int fmin, bmin, dmin, fbmin, bimin, fimin;
1496
408312
    int type=0;
1497
408312
    const int xy = mb_y*s->mb_stride + mb_x;
1498
408312
    init_ref(c, s->new_picture.f->data, s->last_picture.f->data,
1499
408312
             s->next_picture.f->data, 16 * mb_x, 16 * mb_y, 2);
1500
1501
408312
    get_limits(s, 16*mb_x, 16*mb_y);
1502
1503
408312
    c->skip=0;
1504
1505

408312
    if (s->codec_id == AV_CODEC_ID_MPEG4 && s->next_picture.mbskip_table[xy]) {
1506
118
        int score= direct_search(s, mb_x, mb_y); //FIXME just check 0,0
1507
1508
118
        score= ((unsigned)(score*score + 128*256))>>16;
1509
118
        c->mc_mb_var_sum_temp += score;
1510
118
        s->current_picture.mc_mb_var[mb_y*s->mb_stride + mb_x] = score; //FIXME use SSE
1511
118
        s->mb_type[mb_y*s->mb_stride + mb_x]= CANDIDATE_MB_TYPE_DIRECT0;
1512
1513
118
        return;
1514
    }
1515
1516
408194
    if (s->codec_id == AV_CODEC_ID_MPEG4)
1517
191402
        dmin= direct_search(s, mb_x, mb_y);
1518
    else
1519
216792
        dmin= INT_MAX;
1520
// FIXME penalty stuff for non-MPEG-4
1521
408194
    c->skip=0;
1522
408194
    fmin = estimate_motion_b(s, mb_x, mb_y, s->b_forw_mv_table, 0, s->f_code) +
1523
408194
           3 * penalty_factor;
1524
1525
408194
    c->skip=0;
1526
408194
    bmin = estimate_motion_b(s, mb_x, mb_y, s->b_back_mv_table, 2, s->b_code) +
1527
408194
           2 * penalty_factor;
1528
    ff_dlog(s, " %d %d ", s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1]);
1529
1530
408194
    c->skip=0;
1531
408194
    fbmin= bidir_refine(s, mb_x, mb_y) + penalty_factor;
1532
    ff_dlog(s, "%d %d %d %d\n", dmin, fmin, bmin, fbmin);
1533
1534
408194
    if (s->avctx->flags & AV_CODEC_FLAG_INTERLACED_ME) {
1535
//FIXME mb type penalty
1536
115200
        c->skip=0;
1537
115200
        c->current_mv_penalty= c->mv_penalty[s->f_code] + MAX_DMV;
1538
115200
        fimin= interlaced_search(s, 0,
1539
115200
                                 s->b_field_mv_table[0], s->b_field_select_table[0],
1540
115200
                                 s->b_forw_mv_table[xy][0], s->b_forw_mv_table[xy][1], 0);
1541
115200
        c->current_mv_penalty= c->mv_penalty[s->b_code] + MAX_DMV;
1542
115200
        bimin= interlaced_search(s, 2,
1543
115200
                                 s->b_field_mv_table[1], s->b_field_select_table[1],
1544
115200
                                 s->b_back_mv_table[xy][0], s->b_back_mv_table[xy][1], 0);
1545
    }else
1546
292994
        fimin= bimin= INT_MAX;
1547
1548
    {
1549
408194
        int score= fmin;
1550
408194
        type = CANDIDATE_MB_TYPE_FORWARD;
1551
1552
408194
        if (dmin <= score){
1553
55255
            score = dmin;
1554
55255
            type = CANDIDATE_MB_TYPE_DIRECT;
1555
        }
1556
408194
        if(bmin<score){
1557
202836
            score=bmin;
1558
202836
            type= CANDIDATE_MB_TYPE_BACKWARD;
1559
        }
1560
408194
        if(fbmin<score){
1561
237884
            score=fbmin;
1562
237884
            type= CANDIDATE_MB_TYPE_BIDIR;
1563
        }
1564
408194
        if(fimin<score){
1565
3925
            score=fimin;
1566
3925
            type= CANDIDATE_MB_TYPE_FORWARD_I;
1567
        }
1568
408194
        if(bimin<score){
1569
4080
            score=bimin;
1570
4080
            type= CANDIDATE_MB_TYPE_BACKWARD_I;
1571
        }
1572
1573
408194
        score= ((unsigned)(score*score + 128*256))>>16;
1574
408194
        c->mc_mb_var_sum_temp += score;
1575
408194
        s->current_picture.mc_mb_var[mb_y*s->mb_stride + mb_x] = score; //FIXME use SSE
1576
    }
1577
1578
408194
    if(c->avctx->mb_decision > FF_MB_DECISION_SIMPLE){
1579
229850
        type= CANDIDATE_MB_TYPE_FORWARD | CANDIDATE_MB_TYPE_BACKWARD | CANDIDATE_MB_TYPE_BIDIR | CANDIDATE_MB_TYPE_DIRECT;
1580
229850
        if(fimin < INT_MAX)
1581
31700
            type |= CANDIDATE_MB_TYPE_FORWARD_I;
1582
229850
        if(bimin < INT_MAX)
1583
31688
            type |= CANDIDATE_MB_TYPE_BACKWARD_I;
1584

229850
        if(fimin < INT_MAX && bimin < INT_MAX){
1585
26999
            type |= CANDIDATE_MB_TYPE_BIDIR_I;
1586
        }
1587
         //FIXME something smarter
1588
229850
        if(dmin>256*256*16) type&= ~CANDIDATE_MB_TYPE_DIRECT; //do not try direct mode if it is invalid for this MB
1589

229850
        if (s->codec_id == AV_CODEC_ID_MPEG4 && type&CANDIDATE_MB_TYPE_DIRECT &&
1590

153046
            s->mpv_flags & FF_MPV_FLAG_MV0 && *(uint32_t*)s->b_direct_mv_table[xy])
1591
47465
            type |= CANDIDATE_MB_TYPE_DIRECT0;
1592
    }
1593
1594
408194
    s->mb_type[mb_y*s->mb_stride + mb_x]= type;
1595
}
1596
1597
/* find best f_code for ME which do unlimited searches */
1598
10718
int ff_get_best_fcode(MpegEncContext * s, int16_t (*mv_table)[2], int type)
1599
{
1600
10718
    if (s->motion_est != FF_ME_ZERO) {
1601
        int score[8];
1602
10718
        int i, y, range= s->avctx->me_range ? s->avctx->me_range : (INT_MAX/2);
1603
10718
        uint8_t * fcode_tab= s->fcode_tab;
1604
10718
        int best_fcode=-1;
1605
10718
        int best_score=-10000000;
1606
1607
10718
        if(s->msmpeg4_version)
1608
742
            range= FFMIN(range, 16);
1609

9976
        else if(s->codec_id == AV_CODEC_ID_MPEG2VIDEO && s->avctx->strict_std_compliance >= FF_COMPLIANCE_NORMAL)
1610
3786
            range= FFMIN(range, 256);
1611
1612
96462
        for(i=0; i<8; i++) score[i]= s->mb_num*(8-i);
1613
1614
171914
        for(y=0; y<s->mb_height; y++){
1615
            int x;
1616
161196
            int xy= y*s->mb_stride;
1617
3644826
            for(x=0; x<s->mb_width; x++){
1618
3483630
                if(s->mb_type[xy] & type){
1619
2798548
                    int mx= mv_table[xy][0];
1620
2798548
                    int my= mv_table[xy][1];
1621
2798548
                    int fcode= FFMAX(fcode_tab[mx + MAX_MV],
1622
                                     fcode_tab[my + MAX_MV]);
1623
                    int j;
1624
1625

2798548
                        if(mx >= range || mx < -range ||
1626
2786705
                           my >= range || my < -range)
1627
14408
                            continue;
1628
1629

5758770
                    for(j=0; j<fcode && j<8; j++){
1630

2974630
                        if(s->pict_type==AV_PICTURE_TYPE_B || s->current_picture.mc_mb_var[xy] < s->current_picture.mb_var[xy])
1631
2877948
                            score[j]-= 170;
1632
                    }
1633
                }
1634
3469222
                xy++;
1635
            }
1636
        }
1637
1638
85744
        for(i=1; i<8; i++){
1639
75026
            if(score[i] > best_score){
1640
14480
                best_score= score[i];
1641
14480
                best_fcode= i;
1642
            }
1643
        }
1644
1645
10718
        return best_fcode;
1646
    }else{
1647
        return 1;
1648
    }
1649
}
1650
1651
5206
void ff_fix_long_p_mvs(MpegEncContext * s, int type)
1652
{
1653
5206
    MotionEstContext * const c= &s->me;
1654
5206
    const int f_code= s->f_code;
1655
    int y, range;
1656
5206
    av_assert0(s->pict_type==AV_PICTURE_TYPE_P);
1657
1658

5206
    range = (((s->out_format == FMT_MPEG1 || s->msmpeg4_version) ? 8 : 16) << f_code);
1659
1660

5206
    av_assert0(range <= 16 || !s->msmpeg4_version);
1661

5206
    av_assert0(range <=256 || !(s->codec_id == AV_CODEC_ID_MPEG2VIDEO && s->avctx->strict_std_compliance >= FF_COMPLIANCE_NORMAL));
1662
1663

5206
    if(c->avctx->me_range && range > c->avctx->me_range) range= c->avctx->me_range;
1664
1665
5206
    if (s->avctx->flags & AV_CODEC_FLAG_4MV) {
1666
927
        const int wrap= s->b8_stride;
1667
1668
        /* clip / convert to intra 8x8 type MVs */
1669
14148
        for(y=0; y<s->mb_height; y++){
1670
13221
            int xy= y*2*wrap;
1671
13221
            int i= y*s->mb_stride;
1672
            int x;
1673
1674
290916
            for(x=0; x<s->mb_width; x++){
1675
277695
                if(s->mb_type[i]&CANDIDATE_MB_TYPE_INTER4V){
1676
                    int block;
1677
1050460
                    for(block=0; block<4; block++){
1678
840368
                        int off= (block& 1) + (block>>1)*wrap;
1679
840368
                        int mx = s->current_picture.motion_val[0][ xy + off ][0];
1680
840368
                        int my = s->current_picture.motion_val[0][ xy + off ][1];
1681
1682

840368
                        if(   mx >=range || mx <-range
1683

839337
                           || my >=range || my <-range){
1684
1409
                            s->mb_type[i] &= ~CANDIDATE_MB_TYPE_INTER4V;
1685
1409
                            s->mb_type[i] |= type;
1686
1409
                            s->current_picture.mb_type[i] = type;
1687
                        }
1688
                    }
1689
                }
1690
277695
                xy+=2;
1691
277695
                i++;
1692
            }
1693
        }
1694
    }
1695
5206
}
1696
1697
/**
1698
 * @param truncate 1 for truncation, 0 for using intra
1699
 */
1700
14994
void ff_fix_long_mvs(MpegEncContext * s, uint8_t *field_select_table, int field_select,
1701
                     int16_t (*mv_table)[2], int f_code, int type, int truncate)
1702
{
1703
14994
    MotionEstContext * const c= &s->me;
1704
    int y, h_range, v_range;
1705
1706
    // RAL: 8 in MPEG-1, 16 in MPEG-4
1707

14994
    int range = (((s->out_format == FMT_MPEG1 || s->msmpeg4_version) ? 8 : 16) << f_code);
1708
1709

14994
    if(c->avctx->me_range && range > c->avctx->me_range) range= c->avctx->me_range;
1710
1711
14994
    h_range= range;
1712
14994
    v_range= field_select_table ? range>>1 : range;
1713
1714
    /* clip / convert to intra 16x16 type MVs */
1715
234451
    for(y=0; y<s->mb_height; y++){
1716
        int x;
1717
219457
        int xy= y*s->mb_stride;
1718
4888246
        for(x=0; x<s->mb_width; x++){
1719
4668789
            if (s->mb_type[xy] & type){    // RAL: "type" test added...
1720

3174945
                if (!field_select_table || field_select_table[xy] == field_select) {
1721

2992651
                    if(   mv_table[xy][0] >=h_range || mv_table[xy][0] <-h_range
1722

2988508
                       || mv_table[xy][1] >=v_range || mv_table[xy][1] <-v_range){
1723
1724
6659
                        if(truncate){
1725
1528
                            if     (mv_table[xy][0] > h_range-1) mv_table[xy][0]=  h_range-1;
1726
1075
                            else if(mv_table[xy][0] < -h_range ) mv_table[xy][0]= -h_range;
1727
1528
                            if     (mv_table[xy][1] > v_range-1) mv_table[xy][1]=  v_range-1;
1728
1185
                            else if(mv_table[xy][1] < -v_range ) mv_table[xy][1]= -v_range;
1729
                        }else{
1730
5131
                            s->mb_type[xy] &= ~type;
1731
5131
                            s->mb_type[xy] |= CANDIDATE_MB_TYPE_INTRA;
1732
5131
                            mv_table[xy][0]=
1733
5131
                            mv_table[xy][1]= 0;
1734
                        }
1735
                    }
1736
                }
1737
            }
1738
4668789
            xy++;
1739
        }
1740
    }
1741
14994
}