Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264_direct.c @ 2b3649f6

History | View | Annotate | Download (19.9 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[list] : h->ref_count[list];
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[list][j].frame_num + (h->ref_list[list][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)
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
    for(list=0; list<2; list++){
127
        fill_colmap(h, h->map_col_to_list0, list, sidx, ref1sidx, 0);
128
        for(field=0; field<2; field++)
129
            fill_colmap(h, h->map_col_to_list0_field[field], list, field, field, 1);
130
    }
131
}
132

    
133
void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type){
134
    MpegEncContext * const s = &h->s;
135
    int b8_stride = h->b8_stride;
136
    int b4_stride = h->b_stride;
137
    int mb_xy = h->mb_xy;
138
    int mb_type_col[2];
139
    const int16_t (*l1mv0)[2], (*l1mv1)[2];
140
    const int8_t *l1ref0, *l1ref1;
141
    const int is_b8x8 = IS_8X8(*mb_type);
142
    unsigned int sub_mb_type;
143
    int i8, i4;
144

    
145
    assert(h->ref_list[1][0].reference&3);
146

    
147
#define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
148

    
149
    if(IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])){ // AFL/AFR/FR/FL -> AFL/FL
150
        if(!IS_INTERLACED(*mb_type)){                    //     AFR/FR    -> AFL/FL
151
            int cur_poc = s->current_picture_ptr->poc;
152
            int *col_poc = h->ref_list[1]->field_poc;
153
            int col_parity = FFABS(col_poc[0] - cur_poc) >= FFABS(col_poc[1] - cur_poc);
154
            mb_xy= s->mb_x + ((s->mb_y&~1) + col_parity)*s->mb_stride;
155
            b8_stride = 0;
156
        }else if(!(s->picture_structure & h->ref_list[1][0].reference) && !h->ref_list[1][0].mbaff){// FL -> FL & differ parity
157
            int fieldoff= 2*(h->ref_list[1][0].reference)-3;
158
            mb_xy += s->mb_stride*fieldoff;
159
        }
160
        goto single_col;
161
    }else{                                               // AFL/AFR/FR/FL -> AFR/FR
162
        if(IS_INTERLACED(*mb_type)){                     // AFL       /FL -> AFR/FR
163
            mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride;
164
            mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];
165
            mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + s->mb_stride];
166
            b8_stride *= 3;
167
            b4_stride *= 6;
168
            //FIXME IS_8X8(mb_type_col[0]) && !h->sps.direct_8x8_inference_flag
169
            if(    (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)
170
                && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)
171
                && !is_b8x8){
172
                sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
173
                *mb_type   |= MB_TYPE_16x8 |MB_TYPE_L0L1|MB_TYPE_DIRECT2; /* B_16x8 */
174
            }else{
175
                sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
176
                *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
177
            }
178
        }else{                                           //     AFR/FR    -> AFR/FR
179
single_col:
180
            mb_type_col[0] =
181
            mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy];
182
            if(IS_8X8(mb_type_col[0]) && !h->sps.direct_8x8_inference_flag){
183
                /* FIXME save sub mb types from previous frames (or derive from MVs)
184
                * so we know exactly what block size to use */
185
                sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
186
                *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
187
            }else if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
188
                sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
189
                *mb_type   |= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
190
            }else{
191
                sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
192
                *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
193
            }
194
        }
195
    }
196

    
197
    l1mv0  = &h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]];
198
    l1mv1  = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]];
199
    l1ref0 = &h->ref_list[1][0].ref_index [0][h->mb2b8_xy[mb_xy]];
200
    l1ref1 = &h->ref_list[1][0].ref_index [1][h->mb2b8_xy[mb_xy]];
201
    if(!b8_stride){
202
        if(s->mb_y&1){
203
            l1ref0 += h->b8_stride;
204
            l1ref1 += h->b8_stride;
205
            l1mv0  +=  2*b4_stride;
206
            l1mv1  +=  2*b4_stride;
207
        }
208
    }
209

    
210
    if(h->direct_spatial_mv_pred){
211
        int ref[2];
212
        int mv[2][2];
213
        int list;
214

    
215
        /* FIXME interlacing + spatial direct uses wrong colocated block positions */
216

    
217
        /* ref = min(neighbors) */
218
        for(list=0; list<2; list++){
219
            int refa = h->ref_cache[list][scan8[0] - 1];
220
            int refb = h->ref_cache[list][scan8[0] - 8];
221
            int refc = h->ref_cache[list][scan8[0] - 8 + 4];
222
            if(refc == PART_NOT_AVAILABLE)
223
                refc = h->ref_cache[list][scan8[0] - 8 - 1];
224
            ref[list] = FFMIN3((unsigned)refa, (unsigned)refb, (unsigned)refc);
225
            if(ref[list] < 0)
226
                ref[list] = -1;
227
        }
228

    
229
        if(ref[0] < 0 && ref[1] < 0){
230
            ref[0] = ref[1] = 0;
231
            mv[0][0] = mv[0][1] =
232
            mv[1][0] = mv[1][1] = 0;
233
        }else{
234
            for(list=0; list<2; list++){
235
                if(ref[list] >= 0)
236
                    pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
237
                else
238
                    mv[list][0] = mv[list][1] = 0;
239
            }
240
        }
241

    
242
        if(ref[1] < 0){
243
            if(!is_b8x8)
244
                *mb_type &= ~MB_TYPE_L1;
245
            sub_mb_type &= ~MB_TYPE_L1;
246
        }else if(ref[0] < 0){
247
            if(!is_b8x8)
248
                *mb_type &= ~MB_TYPE_L0;
249
            sub_mb_type &= ~MB_TYPE_L0;
250
        }
251

    
252
        if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
253
            for(i8=0; i8<4; i8++){
254
                int x8 = i8&1;
255
                int y8 = i8>>1;
256
                int xy8 = x8+y8*b8_stride;
257
                int xy4 = 3*x8+y8*b4_stride;
258
                int a=0, b=0;
259

    
260
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
261
                    continue;
262
                h->sub_mb_type[i8] = sub_mb_type;
263

    
264
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
265
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
266
                if(!IS_INTRA(mb_type_col[y8])
267
                   && (   (l1ref0[xy8] == 0 && FFABS(l1mv0[xy4][0]) <= 1 && FFABS(l1mv0[xy4][1]) <= 1)
268
                       || (l1ref0[xy8]  < 0 && l1ref1[xy8] == 0 && FFABS(l1mv1[xy4][0]) <= 1 && FFABS(l1mv1[xy4][1]) <= 1))){
269
                    if(ref[0] > 0)
270
                        a= pack16to32(mv[0][0],mv[0][1]);
271
                    if(ref[1] > 0)
272
                        b= pack16to32(mv[1][0],mv[1][1]);
273
                }else{
274
                    a= pack16to32(mv[0][0],mv[0][1]);
275
                    b= pack16to32(mv[1][0],mv[1][1]);
276
                }
277
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, a, 4);
278
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, b, 4);
279
            }
280
        }else if(IS_16X16(*mb_type)){
281
            int a=0, b=0;
282

    
283
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
284
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
285
            if(!IS_INTRA(mb_type_col[0])
286
               && (   (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
287
                   || (l1ref0[0]  < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
288
                       && (h->x264_build>33 || !h->x264_build)))){
289
                if(ref[0] > 0)
290
                    a= pack16to32(mv[0][0],mv[0][1]);
291
                if(ref[1] > 0)
292
                    b= pack16to32(mv[1][0],mv[1][1]);
293
            }else{
294
                a= pack16to32(mv[0][0],mv[0][1]);
295
                b= pack16to32(mv[1][0],mv[1][1]);
296
            }
297
            fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
298
            fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
299
        }else{
300
            for(i8=0; i8<4; i8++){
301
                const int x8 = i8&1;
302
                const int y8 = i8>>1;
303

    
304
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
305
                    continue;
306
                h->sub_mb_type[i8] = sub_mb_type;
307

    
308
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
309
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
310
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
311
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
312

    
313
                /* col_zero_flag */
314
                if(!IS_INTRA(mb_type_col[0]) && (   l1ref0[x8 + y8*b8_stride] == 0
315
                                              || (l1ref0[x8 + y8*b8_stride] < 0 && l1ref1[x8 + y8*b8_stride] == 0
316
                                                  && (h->x264_build>33 || !h->x264_build)))){
317
                    const int16_t (*l1mv)[2]= l1ref0[x8 + y8*b8_stride] == 0 ? l1mv0 : l1mv1;
318
                    if(IS_SUB_8X8(sub_mb_type)){
319
                        const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
320
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
321
                            if(ref[0] == 0)
322
                                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
323
                            if(ref[1] == 0)
324
                                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
325
                        }
326
                    }else
327
                    for(i4=0; i4<4; i4++){
328
                        const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
329
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
330
                            if(ref[0] == 0)
331
                                *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
332
                            if(ref[1] == 0)
333
                                *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
334
                        }
335
                    }
336
                }
337
            }
338
        }
339
    }else{ /* direct temporal mv pred */
340
        const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
341
        const int *dist_scale_factor = h->dist_scale_factor;
342
        int ref_offset= 0;
343

    
344
        if(FRAME_MBAFF && IS_INTERLACED(*mb_type)){
345
            map_col_to_list0[0] = h->map_col_to_list0_field[s->mb_y&1][0];
346
            map_col_to_list0[1] = h->map_col_to_list0_field[s->mb_y&1][1];
347
            dist_scale_factor   =h->dist_scale_factor_field[s->mb_y&1];
348
        }
349
        if(h->ref_list[1][0].mbaff && IS_INTERLACED(mb_type_col[0]))
350
            ref_offset += 16;
351

    
352
        if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
353
            /* FIXME assumes direct_8x8_inference == 1 */
354
            int y_shift  = 2*!IS_INTERLACED(*mb_type);
355

    
356
            for(i8=0; i8<4; i8++){
357
                const int x8 = i8&1;
358
                const int y8 = i8>>1;
359
                int ref0, scale;
360
                const int16_t (*l1mv)[2]= l1mv0;
361

    
362
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
363
                    continue;
364
                h->sub_mb_type[i8] = sub_mb_type;
365

    
366
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
367
                if(IS_INTRA(mb_type_col[y8])){
368
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
369
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
370
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
371
                    continue;
372
                }
373

    
374
                ref0 = l1ref0[x8 + y8*b8_stride];
375
                if(ref0 >= 0)
376
                    ref0 = map_col_to_list0[0][ref0 + ref_offset];
377
                else{
378
                    ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
379
                    l1mv= l1mv1;
380
                }
381
                scale = dist_scale_factor[ref0];
382
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
383

    
384
                {
385
                    const int16_t *mv_col = l1mv[x8*3 + y8*b4_stride];
386
                    int my_col = (mv_col[1]<<y_shift)/2;
387
                    int mx = (scale * mv_col[0] + 128) >> 8;
388
                    int my = (scale * my_col + 128) >> 8;
389
                    fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
390
                    fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
391
                }
392
            }
393
            return;
394
        }
395

    
396
        /* one-to-one mv scaling */
397

    
398
        if(IS_16X16(*mb_type)){
399
            int ref, mv0, mv1;
400

    
401
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
402
            if(IS_INTRA(mb_type_col[0])){
403
                ref=mv0=mv1=0;
404
            }else{
405
                const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0] + ref_offset]
406
                                                : map_col_to_list0[1][l1ref1[0] + ref_offset];
407
                const int scale = dist_scale_factor[ref0];
408
                const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
409
                int mv_l0[2];
410
                mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
411
                mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
412
                ref= ref0;
413
                mv0= pack16to32(mv_l0[0],mv_l0[1]);
414
                mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
415
            }
416
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
417
            fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);
418
            fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4);
419
        }else{
420
            for(i8=0; i8<4; i8++){
421
                const int x8 = i8&1;
422
                const int y8 = i8>>1;
423
                int ref0, scale;
424
                const int16_t (*l1mv)[2]= l1mv0;
425

    
426
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
427
                    continue;
428
                h->sub_mb_type[i8] = sub_mb_type;
429
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
430
                if(IS_INTRA(mb_type_col[0])){
431
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
432
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
433
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
434
                    continue;
435
                }
436

    
437
                ref0 = l1ref0[x8 + y8*b8_stride] + ref_offset;
438
                if(ref0 >= 0)
439
                    ref0 = map_col_to_list0[0][ref0];
440
                else{
441
                    ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
442
                    l1mv= l1mv1;
443
                }
444
                scale = dist_scale_factor[ref0];
445

    
446
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
447
                if(IS_SUB_8X8(sub_mb_type)){
448
                    const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
449
                    int mx = (scale * mv_col[0] + 128) >> 8;
450
                    int my = (scale * mv_col[1] + 128) >> 8;
451
                    fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
452
                    fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
453
                }else
454
                for(i4=0; i4<4; i4++){
455
                    const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
456
                    int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
457
                    mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
458
                    mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
459
                    *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
460
                        pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
461
                }
462
            }
463
        }
464
    }
465
}