Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264_direct.c @ e8d964fa

History | View | Annotate | Download (20.1 KB)

1
/*
2
 * H.26L/H.264/AVC/JVT/14496-10/... direct mb/block decoding
3
 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4
 *
5
 * This file is part of FFmpeg.
6
 *
7
 * FFmpeg 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 2.1 of the License, or (at your option) any later version.
11
 *
12
 * FFmpeg 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 FFmpeg; if not, write to the Free Software
19
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20
 */
21

    
22
/**
23
 * @file libavcodec/h264_direct.c
24
 * H.264 / AVC / MPEG4 part10 direct mb/block decoding.
25
 * @author Michael Niedermayer <michaelni@gmx.at>
26
 */
27

    
28
#include "internal.h"
29
#include "dsputil.h"
30
#include "avcodec.h"
31
#include "mpegvideo.h"
32
#include "h264.h"
33
#include "h264_mvpred.h"
34
#include "rectangle.h"
35

    
36
//#undef NDEBUG
37
#include <assert.h>
38

    
39

    
40
static int get_scale_factor(H264Context * const h, int poc, int poc1, int i){
41
    int poc0 = h->ref_list[0][i].poc;
42
    int td = av_clip(poc1 - poc0, -128, 127);
43
    if(td == 0 || h->ref_list[0][i].long_ref){
44
        return 256;
45
    }else{
46
        int tb = av_clip(poc - poc0, -128, 127);
47
        int tx = (16384 + (FFABS(td) >> 1)) / td;
48
        return av_clip((tb*tx + 32) >> 6, -1024, 1023);
49
    }
50
}
51

    
52
void ff_h264_direct_dist_scale_factor(H264Context * const h){
53
    MpegEncContext * const s = &h->s;
54
    const int poc = h->s.current_picture_ptr->field_poc[ s->picture_structure == PICT_BOTTOM_FIELD ];
55
    const int poc1 = h->ref_list[1][0].poc;
56
    int i, field;
57
    for(field=0; field<2; field++){
58
        const int poc  = h->s.current_picture_ptr->field_poc[field];
59
        const int poc1 = h->ref_list[1][0].field_poc[field];
60
        for(i=0; i < 2*h->ref_count[0]; i++)
61
            h->dist_scale_factor_field[field][i^field] = get_scale_factor(h, poc, poc1, i+16);
62
    }
63

    
64
    for(i=0; i<h->ref_count[0]; i++){
65
        h->dist_scale_factor[i] = get_scale_factor(h, poc, poc1, i);
66
    }
67
}
68

    
69
static void fill_colmap(H264Context *h, int map[2][16+32], int list, int field, int colfield, int mbafi){
70
    MpegEncContext * const s = &h->s;
71
    Picture * const ref1 = &h->ref_list[1][0];
72
    int j, old_ref, rfield;
73
    int start= mbafi ? 16                      : 0;
74
    int end  = mbafi ? 16+2*h->ref_count[0]    : h->ref_count[0];
75
    int interl= mbafi || s->picture_structure != PICT_FRAME;
76

    
77
    /* bogus; fills in for missing frames */
78
    memset(map[list], 0, sizeof(map[list]));
79

    
80
    for(rfield=0; rfield<2; rfield++){
81
        for(old_ref=0; old_ref<ref1->ref_count[colfield][list]; old_ref++){
82
            int poc = ref1->ref_poc[colfield][list][old_ref];
83

    
84
            if     (!interl)
85
                poc |= 3;
86
            else if( interl && (poc&3) == 3) //FIXME store all MBAFF references so this isnt needed
87
                poc= (poc&~3) + rfield + 1;
88

    
89
            for(j=start; j<end; j++){
90
                if(4*h->ref_list[0][j].frame_num + (h->ref_list[0][j].reference&3) == poc){
91
                    int cur_ref= mbafi ? (j-16)^field : j;
92
                    map[list][2*old_ref + (rfield^field) + 16] = cur_ref;
93
                    if(rfield == field || !interl)
94
                        map[list][old_ref] = cur_ref;
95
                    break;
96
                }
97
            }
98
        }
99
    }
100
}
101

    
102
void ff_h264_direct_ref_list_init(H264Context * const h){
103
    MpegEncContext * const s = &h->s;
104
    Picture * const ref1 = &h->ref_list[1][0];
105
    Picture * const cur = s->current_picture_ptr;
106
    int list, j, field;
107
    int sidx= (s->picture_structure&1)^1;
108
    int ref1sidx= (ref1->reference&1)^1;
109

    
110
    for(list=0; list<2; list++){
111
        cur->ref_count[sidx][list] = h->ref_count[list];
112
        for(j=0; j<h->ref_count[list]; j++)
113
            cur->ref_poc[sidx][list][j] = 4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3);
114
    }
115

    
116
    if(s->picture_structure == PICT_FRAME){
117
        memcpy(cur->ref_count[1], cur->ref_count[0], sizeof(cur->ref_count[0]));
118
        memcpy(cur->ref_poc  [1], cur->ref_poc  [0], sizeof(cur->ref_poc  [0]));
119
    }
120

    
121
    cur->mbaff= FRAME_MBAFF;
122

    
123
    if(cur->pict_type != FF_B_TYPE || h->direct_spatial_mv_pred)
124
        return;
125

    
126
    if(s->picture_structure == PICT_FRAME){
127
        int cur_poc = s->current_picture_ptr->poc;
128
        int *col_poc = h->ref_list[1]->field_poc;
129
        ref1sidx=sidx= (FFABS(col_poc[0] - cur_poc) >= FFABS(col_poc[1] - cur_poc));
130
    }
131

    
132
    for(list=0; list<2; list++){
133
        fill_colmap(h, h->map_col_to_list0, list, sidx, ref1sidx, 0);
134
        if(FRAME_MBAFF)
135
        for(field=0; field<2; field++)
136
            fill_colmap(h, h->map_col_to_list0_field[field], list, field, field, 1);
137
    }
138
}
139

    
140
void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type){
141
    MpegEncContext * const s = &h->s;
142
    int b8_stride = h->b8_stride;
143
    int b4_stride = h->b_stride;
144
    int mb_xy = h->mb_xy;
145
    int mb_type_col[2];
146
    const int16_t (*l1mv0)[2], (*l1mv1)[2];
147
    const int8_t *l1ref0, *l1ref1;
148
    const int is_b8x8 = IS_8X8(*mb_type);
149
    unsigned int sub_mb_type;
150
    int i8, i4;
151

    
152
    assert(h->ref_list[1][0].reference&3);
153

    
154
#define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
155

    
156
    if(IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])){ // AFL/AFR/FR/FL -> AFL/FL
157
        if(!IS_INTERLACED(*mb_type)){                    //     AFR/FR    -> AFL/FL
158
            int cur_poc = s->current_picture_ptr->poc;
159
            int *col_poc = h->ref_list[1]->field_poc;
160
            int col_parity = FFABS(col_poc[0] - cur_poc) >= FFABS(col_poc[1] - cur_poc);
161
            mb_xy= s->mb_x + ((s->mb_y&~1) + col_parity)*s->mb_stride;
162
            b8_stride = 0;
163
        }else if(!(s->picture_structure & h->ref_list[1][0].reference) && !h->ref_list[1][0].mbaff){// FL -> FL & differ parity
164
            int fieldoff= 2*(h->ref_list[1][0].reference)-3;
165
            mb_xy += s->mb_stride*fieldoff;
166
        }
167
        goto single_col;
168
    }else{                                               // AFL/AFR/FR/FL -> AFR/FR
169
        if(IS_INTERLACED(*mb_type)){                     // AFL       /FL -> AFR/FR
170
            mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride;
171
            mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];
172
            mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + s->mb_stride];
173
            b8_stride *= 3;
174
            b4_stride *= 6;
175

    
176
            sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
177
            if(    (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)
178
                && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)
179
                && !is_b8x8){
180
                *mb_type   |= MB_TYPE_16x8 |MB_TYPE_L0L1|MB_TYPE_DIRECT2; /* B_16x8 */
181
            }else{
182
                *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
183
            }
184
        }else{                                           //     AFR/FR    -> AFR/FR
185
single_col:
186
            mb_type_col[0] =
187
            mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy];
188
            if(IS_8X8(mb_type_col[0]) && !h->sps.direct_8x8_inference_flag){
189
                /* FIXME save sub mb types from previous frames (or derive from MVs)
190
                * so we know exactly what block size to use */
191
                sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
192
                *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
193
            }else if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
194
                sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
195
                *mb_type   |= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
196
            }else{
197
                sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
198
                *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
199
            }
200
        }
201
    }
202

    
203
    l1mv0  = &h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]];
204
    l1mv1  = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]];
205
    l1ref0 = &h->ref_list[1][0].ref_index [0][h->mb2b8_xy[mb_xy]];
206
    l1ref1 = &h->ref_list[1][0].ref_index [1][h->mb2b8_xy[mb_xy]];
207
    if(!b8_stride){
208
        if(s->mb_y&1){
209
            l1ref0 += h->b8_stride;
210
            l1ref1 += h->b8_stride;
211
            l1mv0  +=  2*b4_stride;
212
            l1mv1  +=  2*b4_stride;
213
        }
214
    }
215

    
216
    if(h->direct_spatial_mv_pred){
217
        int ref[2];
218
        int mv[2][2];
219
        int list;
220

    
221
        /* FIXME interlacing + spatial direct uses wrong colocated block positions */
222

    
223
        /* ref = min(neighbors) */
224
        for(list=0; list<2; list++){
225
            int refa = h->ref_cache[list][scan8[0] - 1];
226
            int refb = h->ref_cache[list][scan8[0] - 8];
227
            int refc = h->ref_cache[list][scan8[0] - 8 + 4];
228
            if(refc == PART_NOT_AVAILABLE)
229
                refc = h->ref_cache[list][scan8[0] - 8 - 1];
230
            ref[list] = FFMIN3((unsigned)refa, (unsigned)refb, (unsigned)refc);
231
            if(ref[list] < 0)
232
                ref[list] = -1;
233
        }
234

    
235
        if(ref[0] < 0 && ref[1] < 0){
236
            ref[0] = ref[1] = 0;
237
            mv[0][0] = mv[0][1] =
238
            mv[1][0] = mv[1][1] = 0;
239
        }else{
240
            for(list=0; list<2; list++){
241
                if(ref[list] >= 0)
242
                    pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
243
                else
244
                    mv[list][0] = mv[list][1] = 0;
245
            }
246
        }
247

    
248
        if(ref[1] < 0){
249
            if(!is_b8x8)
250
                *mb_type &= ~MB_TYPE_L1;
251
            sub_mb_type &= ~MB_TYPE_L1;
252
        }else if(ref[0] < 0){
253
            if(!is_b8x8)
254
                *mb_type &= ~MB_TYPE_L0;
255
            sub_mb_type &= ~MB_TYPE_L0;
256
        }
257

    
258
        if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
259
            for(i8=0; i8<4; i8++){
260
                int x8 = i8&1;
261
                int y8 = i8>>1;
262
                int xy8 = x8+y8*b8_stride;
263
                int xy4 = 3*x8+y8*b4_stride;
264
                int a=0, b=0;
265

    
266
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
267
                    continue;
268
                h->sub_mb_type[i8] = sub_mb_type;
269

    
270
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
271
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
272
                if(!IS_INTRA(mb_type_col[y8]) && !h->ref_list[1][0].long_ref
273
                   && (   (l1ref0[xy8] == 0 && FFABS(l1mv0[xy4][0]) <= 1 && FFABS(l1mv0[xy4][1]) <= 1)
274
                       || (l1ref0[xy8]  < 0 && l1ref1[xy8] == 0 && FFABS(l1mv1[xy4][0]) <= 1 && FFABS(l1mv1[xy4][1]) <= 1))){
275
                    if(ref[0] > 0)
276
                        a= pack16to32(mv[0][0],mv[0][1]);
277
                    if(ref[1] > 0)
278
                        b= pack16to32(mv[1][0],mv[1][1]);
279
                }else{
280
                    a= pack16to32(mv[0][0],mv[0][1]);
281
                    b= pack16to32(mv[1][0],mv[1][1]);
282
                }
283
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, a, 4);
284
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, b, 4);
285
            }
286
        }else if(IS_16X16(*mb_type)){
287
            int a=0, b=0;
288

    
289
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
290
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
291
            if(!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref
292
               && (   (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
293
                   || (l1ref0[0]  < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
294
                       && (h->x264_build>33 || !h->x264_build)))){
295
                if(ref[0] > 0)
296
                    a= pack16to32(mv[0][0],mv[0][1]);
297
                if(ref[1] > 0)
298
                    b= pack16to32(mv[1][0],mv[1][1]);
299
            }else{
300
                a= pack16to32(mv[0][0],mv[0][1]);
301
                b= pack16to32(mv[1][0],mv[1][1]);
302
            }
303
            fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
304
            fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
305
        }else{
306
            for(i8=0; i8<4; i8++){
307
                const int x8 = i8&1;
308
                const int y8 = i8>>1;
309

    
310
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
311
                    continue;
312
                h->sub_mb_type[i8] = sub_mb_type;
313

    
314
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
315
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
316
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
317
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
318

    
319
                /* col_zero_flag */
320
                if(!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref && (   l1ref0[x8 + y8*b8_stride] == 0
321
                                              || (l1ref0[x8 + y8*b8_stride] < 0 && l1ref1[x8 + y8*b8_stride] == 0
322
                                                  && (h->x264_build>33 || !h->x264_build)))){
323
                    const int16_t (*l1mv)[2]= l1ref0[x8 + y8*b8_stride] == 0 ? l1mv0 : l1mv1;
324
                    if(IS_SUB_8X8(sub_mb_type)){
325
                        const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
326
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
327
                            if(ref[0] == 0)
328
                                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
329
                            if(ref[1] == 0)
330
                                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
331
                        }
332
                    }else
333
                    for(i4=0; i4<4; i4++){
334
                        const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
335
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
336
                            if(ref[0] == 0)
337
                                *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
338
                            if(ref[1] == 0)
339
                                *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
340
                        }
341
                    }
342
                }
343
            }
344
        }
345
    }else{ /* direct temporal mv pred */
346
        const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
347
        const int *dist_scale_factor = h->dist_scale_factor;
348
        int ref_offset= 0;
349

    
350
        if(FRAME_MBAFF && IS_INTERLACED(*mb_type)){
351
            map_col_to_list0[0] = h->map_col_to_list0_field[s->mb_y&1][0];
352
            map_col_to_list0[1] = h->map_col_to_list0_field[s->mb_y&1][1];
353
            dist_scale_factor   =h->dist_scale_factor_field[s->mb_y&1];
354
        }
355
        if(h->ref_list[1][0].mbaff && IS_INTERLACED(mb_type_col[0]))
356
            ref_offset += 16;
357

    
358
        if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
359
            int y_shift  = 2*!IS_INTERLACED(*mb_type);
360
            assert(h->sps.direct_8x8_inference_flag);
361

    
362
            for(i8=0; i8<4; i8++){
363
                const int x8 = i8&1;
364
                const int y8 = i8>>1;
365
                int ref0, scale;
366
                const int16_t (*l1mv)[2]= l1mv0;
367

    
368
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
369
                    continue;
370
                h->sub_mb_type[i8] = sub_mb_type;
371

    
372
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
373
                if(IS_INTRA(mb_type_col[y8])){
374
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
375
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
376
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
377
                    continue;
378
                }
379

    
380
                ref0 = l1ref0[x8 + y8*b8_stride];
381
                if(ref0 >= 0)
382
                    ref0 = map_col_to_list0[0][ref0 + ref_offset];
383
                else{
384
                    ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
385
                    l1mv= l1mv1;
386
                }
387
                scale = dist_scale_factor[ref0];
388
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
389

    
390
                {
391
                    const int16_t *mv_col = l1mv[x8*3 + y8*b4_stride];
392
                    int my_col = (mv_col[1]<<y_shift)/2;
393
                    int mx = (scale * mv_col[0] + 128) >> 8;
394
                    int my = (scale * my_col + 128) >> 8;
395
                    fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
396
                    fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
397
                }
398
            }
399
            return;
400
        }
401

    
402
        /* one-to-one mv scaling */
403

    
404
        if(IS_16X16(*mb_type)){
405
            int ref, mv0, mv1;
406

    
407
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
408
            if(IS_INTRA(mb_type_col[0])){
409
                ref=mv0=mv1=0;
410
            }else{
411
                const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0] + ref_offset]
412
                                                : map_col_to_list0[1][l1ref1[0] + ref_offset];
413
                const int scale = dist_scale_factor[ref0];
414
                const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
415
                int mv_l0[2];
416
                mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
417
                mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
418
                ref= ref0;
419
                mv0= pack16to32(mv_l0[0],mv_l0[1]);
420
                mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
421
            }
422
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
423
            fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);
424
            fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4);
425
        }else{
426
            for(i8=0; i8<4; i8++){
427
                const int x8 = i8&1;
428
                const int y8 = i8>>1;
429
                int ref0, scale;
430
                const int16_t (*l1mv)[2]= l1mv0;
431

    
432
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
433
                    continue;
434
                h->sub_mb_type[i8] = sub_mb_type;
435
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
436
                if(IS_INTRA(mb_type_col[0])){
437
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
438
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
439
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
440
                    continue;
441
                }
442

    
443
                ref0 = l1ref0[x8 + y8*b8_stride];
444
                if(ref0 >= 0)
445
                    ref0 = map_col_to_list0[0][ref0 + ref_offset];
446
                else{
447
                    ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
448
                    l1mv= l1mv1;
449
                }
450
                scale = dist_scale_factor[ref0];
451

    
452
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
453
                if(IS_SUB_8X8(sub_mb_type)){
454
                    const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
455
                    int mx = (scale * mv_col[0] + 128) >> 8;
456
                    int my = (scale * mv_col[1] + 128) >> 8;
457
                    fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
458
                    fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
459
                }else
460
                for(i4=0; i4<4; i4++){
461
                    const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
462
                    int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
463
                    mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
464
                    mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
465
                    *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
466
                        pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
467
                }
468
            }
469
        }
470
    }
471
}