Statistics
| Branch: | Revision:

ffmpeg / libavcodec / cavs.c @ 6636b7e8

History | View | Annotate | Download (51.6 KB)

1
/*
2
 * Chinese AVS video (AVS1-P2, JiZhun profile) decoder.
3
 * Copyright (c) 2006  Stefan Gehrer <stefan.gehrer@gmx.de>
4
 *
5
 * This library is free software; you can redistribute it and/or
6
 * modify it under the terms of the GNU Lesser General Public
7
 * License as published by the Free Software Foundation; either
8
 * version 2 of the License, or (at your option) any later version.
9
 *
10
 * This library is distributed in the hope that it will be useful,
11
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13
 * Lesser General Public License for more details.
14
 *
15
 * You should have received a copy of the GNU Lesser General Public
16
 * License along with this library; if not, write to the Free Software
17
 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
18
 */
19

    
20
/**
21
 * @file cavs.c
22
 * Chinese AVS video (AVS1-P2, JiZhun profile) decoder
23
 * @author Stefan Gehrer <stefan.gehrer@gmx.de>
24
 */
25

    
26
#include "avcodec.h"
27
#include "bitstream.h"
28
#include "golomb.h"
29
#include "mpegvideo.h"
30
#include "cavsdata.h"
31

    
32
typedef struct {
33
    MpegEncContext s;
34
    Picture picture; ///< currently decoded frame
35
    Picture DPB[2];  ///< reference frames
36
    int dist[2];     ///< temporal distances from current frame to ref frames
37
    int profile, level;
38
    int aspect_ratio;
39
    int mb_width, mb_height;
40
    int pic_type;
41
    int progressive;
42
    int pic_structure;
43
    int skip_mode_flag; ///< select between skip_count or one skip_flag per MB
44
    int loop_filter_disable;
45
    int alpha_offset, beta_offset;
46
    int ref_flag;
47
    int mbx, mby;      ///< macroblock coordinates
48
    int flags;         ///< availability flags of neighbouring macroblocks
49
    int stc;           ///< last start code
50
    uint8_t *cy, *cu, *cv; ///< current MB sample pointers
51
    int left_qp;
52
    uint8_t *top_qp;
53

    
54
    /** mv motion vector cache
55
       0:    D3  B2  B3  C2
56
       4:    A1  X0  X1   -
57
       8:    A3  X2  X3   -
58

59
       X are the vectors in the current macroblock (5,6,9,10)
60
       A is the macroblock to the left (4,8)
61
       B is the macroblock to the top (1,2)
62
       C is the macroblock to the top-right (3)
63
       D is the macroblock to the top-left (0)
64

65
       the same is repeated for backward motion vectors */
66
    vector_t mv[2*4*3];
67
    vector_t *top_mv[2];
68
    vector_t *col_mv;
69

    
70
    /** luma pred mode cache
71
       0:    --  B2  B3
72
       3:    A1  X0  X1
73
       6:    A3  X2  X3   */
74
    int pred_mode_Y[3*3];
75
    int *top_pred_Y;
76
    int l_stride, c_stride;
77
    int luma_scan[4];
78
    int qp;
79
    int qp_fixed;
80
    int cbp;
81

    
82
    /** intra prediction is done with un-deblocked samples
83
     they are saved here before deblocking the MB  */
84
    uint8_t *top_border_y, *top_border_u, *top_border_v;
85
    uint8_t left_border_y[16], left_border_u[10], left_border_v[10];
86
    uint8_t topleft_border_y, topleft_border_u, topleft_border_v;
87

    
88
    void (*intra_pred_l[8])(uint8_t *d,uint8_t *top,uint8_t *left,int stride);
89
    void (*intra_pred_c[7])(uint8_t *d,uint8_t *top,uint8_t *left,int stride);
90
    uint8_t *col_type_base;
91
    uint8_t *col_type;
92

    
93
    /* scaling factors for MV prediction */
94
    int sym_factor;    ///< for scaling in symmetrical B block
95
    int direct_den[2]; ///< for scaling in direct B block
96
    int scale_den[2];  ///< for scaling neighbouring MVs
97

    
98
    int got_keyframe;
99
} AVSContext;
100

    
101
/*****************************************************************************
102
 *
103
 * in-loop deblocking filter
104
 *
105
 ****************************************************************************/
106

    
107
static inline int get_bs(vector_t *mvP, vector_t *mvQ, int b) {
108
    if((mvP->ref == REF_INTRA) || (mvQ->ref == REF_INTRA))
109
        return 2;
110
    if( (abs(mvP->x - mvQ->x) >= 4) ||  (abs(mvP->y - mvQ->y) >= 4) )
111
        return 1;
112
    if(b){
113
        mvP += MV_BWD_OFFS;
114
        mvQ += MV_BWD_OFFS;
115
        if( (abs(mvP->x - mvQ->x) >= 4) ||  (abs(mvP->y - mvQ->y) >= 4) )
116
            return 1;
117
    }else{
118
        if(mvP->ref != mvQ->ref)
119
            return 1;
120
    }
121
    return 0;
122
}
123

    
124
#define SET_PARAMS                                            \
125
    alpha = alpha_tab[clip(qp_avg + h->alpha_offset,0,63)];   \
126
    beta  =  beta_tab[clip(qp_avg + h->beta_offset, 0,63)];   \
127
    tc    =    tc_tab[clip(qp_avg + h->alpha_offset,0,63)];
128

    
129
/**
130
 * in-loop deblocking filter for a single macroblock
131
 *
132
 * boundary strength (bs) mapping:
133
 *
134
 * --4---5--
135
 * 0   2   |
136
 * | 6 | 7 |
137
 * 1   3   |
138
 * ---------
139
 *
140
 */
141
static void filter_mb(AVSContext *h, enum mb_t mb_type) {
142
    DECLARE_ALIGNED_8(uint8_t, bs[8]);
143
    int qp_avg, alpha, beta, tc;
144
    int i;
145

    
146
    /* save un-deblocked lines */
147
    h->topleft_border_y = h->top_border_y[h->mbx*16+15];
148
    h->topleft_border_u = h->top_border_u[h->mbx*10+8];
149
    h->topleft_border_v = h->top_border_v[h->mbx*10+8];
150
    memcpy(&h->top_border_y[h->mbx*16], h->cy + 15* h->l_stride,16);
151
    memcpy(&h->top_border_u[h->mbx*10+1], h->cu +  7* h->c_stride,8);
152
    memcpy(&h->top_border_v[h->mbx*10+1], h->cv +  7* h->c_stride,8);
153
    for(i=0;i<8;i++) {
154
        h->left_border_y[i*2+0] = *(h->cy + 15 + (i*2+0)*h->l_stride);
155
        h->left_border_y[i*2+1] = *(h->cy + 15 + (i*2+1)*h->l_stride);
156
        h->left_border_u[i+1] = *(h->cu + 7 + i*h->c_stride);
157
        h->left_border_v[i+1] = *(h->cv + 7 + i*h->c_stride);
158
    }
159
    if(!h->loop_filter_disable) {
160
        /* determine bs */
161
        if(mb_type == I_8X8)
162
            *((uint64_t *)bs) = 0x0202020202020202ULL;
163
        else{
164
            *((uint64_t *)bs) = 0;
165
            if(partition_flags[mb_type] & SPLITV){
166
                bs[2] = get_bs(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X1], mb_type > P_8X8);
167
                bs[3] = get_bs(&h->mv[MV_FWD_X2], &h->mv[MV_FWD_X3], mb_type > P_8X8);
168
            }
169
            if(partition_flags[mb_type] & SPLITH){
170
                bs[6] = get_bs(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X2], mb_type > P_8X8);
171
                bs[7] = get_bs(&h->mv[MV_FWD_X1], &h->mv[MV_FWD_X3], mb_type > P_8X8);
172
            }
173
            bs[0] = get_bs(&h->mv[MV_FWD_A1], &h->mv[MV_FWD_X0], mb_type > P_8X8);
174
            bs[1] = get_bs(&h->mv[MV_FWD_A3], &h->mv[MV_FWD_X2], mb_type > P_8X8);
175
            bs[4] = get_bs(&h->mv[MV_FWD_B2], &h->mv[MV_FWD_X0], mb_type > P_8X8);
176
            bs[5] = get_bs(&h->mv[MV_FWD_B3], &h->mv[MV_FWD_X1], mb_type > P_8X8);
177
        }
178
        if( *((uint64_t *)bs) ) {
179
            if(h->flags & A_AVAIL) {
180
                qp_avg = (h->qp + h->left_qp + 1) >> 1;
181
                SET_PARAMS;
182
                h->s.dsp.cavs_filter_lv(h->cy,h->l_stride,alpha,beta,tc,bs[0],bs[1]);
183
                h->s.dsp.cavs_filter_cv(h->cu,h->c_stride,alpha,beta,tc,bs[0],bs[1]);
184
                h->s.dsp.cavs_filter_cv(h->cv,h->c_stride,alpha,beta,tc,bs[0],bs[1]);
185
            }
186
            qp_avg = h->qp;
187
            SET_PARAMS;
188
            h->s.dsp.cavs_filter_lv(h->cy + 8,h->l_stride,alpha,beta,tc,bs[2],bs[3]);
189
            h->s.dsp.cavs_filter_lh(h->cy + 8*h->l_stride,h->l_stride,alpha,beta,tc,
190
                           bs[6],bs[7]);
191

    
192
            if(h->flags & B_AVAIL) {
193
                qp_avg = (h->qp + h->top_qp[h->mbx] + 1) >> 1;
194
                SET_PARAMS;
195
                h->s.dsp.cavs_filter_lh(h->cy,h->l_stride,alpha,beta,tc,bs[4],bs[5]);
196
                h->s.dsp.cavs_filter_ch(h->cu,h->c_stride,alpha,beta,tc,bs[4],bs[5]);
197
                h->s.dsp.cavs_filter_ch(h->cv,h->c_stride,alpha,beta,tc,bs[4],bs[5]);
198
            }
199
        }
200
    }
201
    h->left_qp = h->qp;
202
    h->top_qp[h->mbx] = h->qp;
203
}
204

    
205
#undef SET_PARAMS
206

    
207
/*****************************************************************************
208
 *
209
 * spatial intra prediction
210
 *
211
 ****************************************************************************/
212

    
213
static inline void load_intra_pred_luma(AVSContext *h, uint8_t *top,
214
                                        uint8_t *left, int block) {
215
    int i;
216

    
217
    switch(block) {
218
    case 0:
219
        memcpy(&left[1],h->left_border_y,16);
220
        left[0] = left[1];
221
        left[17] = left[16];
222
        memcpy(&top[1],&h->top_border_y[h->mbx*16],16);
223
        top[17] = top[16];
224
        top[0] = top[1];
225
        if((h->flags & A_AVAIL) && (h->flags & B_AVAIL))
226
            left[0] = top[0] = h->topleft_border_y;
227
        break;
228
    case 1:
229
        for(i=0;i<8;i++)
230
            left[i+1] = *(h->cy + 7 + i*h->l_stride);
231
        memset(&left[9],left[8],9);
232
        left[0] = left[1];
233
        memcpy(&top[1],&h->top_border_y[h->mbx*16+8],8);
234
        if(h->flags & C_AVAIL)
235
            memcpy(&top[9],&h->top_border_y[(h->mbx + 1)*16],8);
236
        else
237
            memset(&top[9],top[8],9);
238
        top[17] = top[16];
239
        top[0] = top[1];
240
        if(h->flags & B_AVAIL)
241
            left[0] = top[0] = h->top_border_y[h->mbx*16+7];
242
        break;
243
    case 2:
244
        memcpy(&left[1],&h->left_border_y[8],8);
245
        memset(&left[9],left[8],9);
246
        memcpy(&top[1],h->cy + 7*h->l_stride,16);
247
        top[17] = top[16];
248
        left[0] = h->left_border_y[7];
249
        top[0] = top[1];
250
        if(h->flags & A_AVAIL)
251
            top[0] = left[0];
252
        break;
253
    case 3:
254
        for(i=0;i<9;i++)
255
            left[i] = *(h->cy + 7 + (i+7)*h->l_stride);
256
        memset(&left[9],left[8],9);
257
        memcpy(&top[0],h->cy + 7 + 7*h->l_stride,9);
258
        memset(&top[9],top[8],9);
259
        break;
260
    }
261
}
262

    
263
static void intra_pred_vert(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
264
    int y;
265
    uint64_t a = unaligned64(&top[1]);
266
    for(y=0;y<8;y++) {
267
        *((uint64_t *)(d+y*stride)) = a;
268
    }
269
}
270

    
271
static void intra_pred_horiz(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
272
    int y;
273
    uint64_t a;
274
    for(y=0;y<8;y++) {
275
        a = left[y+1] * 0x0101010101010101ULL;
276
        *((uint64_t *)(d+y*stride)) = a;
277
    }
278
}
279

    
280
static void intra_pred_dc_128(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
281
    int y;
282
    uint64_t a = 0x8080808080808080ULL;
283
    for(y=0;y<8;y++)
284
        *((uint64_t *)(d+y*stride)) = a;
285
}
286

    
287
static void intra_pred_plane(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
288
    int x,y,ia;
289
    int ih = 0;
290
    int iv = 0;
291
    uint8_t *cm = cropTbl + MAX_NEG_CROP;
292

    
293
    for(x=0; x<4; x++) {
294
        ih += (x+1)*(top[5+x]-top[3-x]);
295
        iv += (x+1)*(left[5+x]-left[3-x]);
296
    }
297
    ia = (top[8]+left[8])<<4;
298
    ih = (17*ih+16)>>5;
299
    iv = (17*iv+16)>>5;
300
    for(y=0; y<8; y++)
301
        for(x=0; x<8; x++)
302
            d[y*stride+x] = cm[(ia+(x-3)*ih+(y-3)*iv+16)>>5];
303
}
304

    
305
#define LOWPASS(ARRAY,INDEX)                                            \
306
    (( ARRAY[(INDEX)-1] + 2*ARRAY[(INDEX)] + ARRAY[(INDEX)+1] + 2) >> 2)
307

    
308
static void intra_pred_lp(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
309
    int x,y;
310
    for(y=0; y<8; y++)
311
        for(x=0; x<8; x++)
312
            d[y*stride+x] = (LOWPASS(top,x+1) + LOWPASS(left,y+1)) >> 1;
313
}
314

    
315
static void intra_pred_down_left(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
316
    int x,y;
317
    for(y=0; y<8; y++)
318
        for(x=0; x<8; x++)
319
            d[y*stride+x] = (LOWPASS(top,x+y+2) + LOWPASS(left,x+y+2)) >> 1;
320
}
321

    
322
static void intra_pred_down_right(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
323
    int x,y;
324
    for(y=0; y<8; y++)
325
        for(x=0; x<8; x++)
326
            if(x==y)
327
                d[y*stride+x] = (left[1]+2*top[0]+top[1]+2)>>2;
328
            else if(x>y)
329
                d[y*stride+x] = LOWPASS(top,x-y);
330
            else
331
                d[y*stride+x] = LOWPASS(left,y-x);
332
}
333

    
334
static void intra_pred_lp_left(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
335
    int x,y;
336
    for(y=0; y<8; y++)
337
        for(x=0; x<8; x++)
338
            d[y*stride+x] = LOWPASS(left,y+1);
339
}
340

    
341
static void intra_pred_lp_top(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
342
    int x,y;
343
    for(y=0; y<8; y++)
344
        for(x=0; x<8; x++)
345
            d[y*stride+x] = LOWPASS(top,x+1);
346
}
347

    
348
#undef LOWPASS
349

    
350
static inline void modify_pred(const int_fast8_t *mod_table, int *mode) {
351
    *mode = mod_table[*mode];
352
    if(*mode < 0) {
353
        av_log(NULL, AV_LOG_ERROR, "Illegal intra prediction mode\n");
354
        *mode = 0;
355
    }
356
}
357

    
358
/*****************************************************************************
359
 *
360
 * motion compensation
361
 *
362
 ****************************************************************************/
363

    
364
static inline void mc_dir_part(AVSContext *h,Picture *pic,int square,
365
                        int chroma_height,int delta,int list,uint8_t *dest_y,
366
                        uint8_t *dest_cb,uint8_t *dest_cr,int src_x_offset,
367
                        int src_y_offset,qpel_mc_func *qpix_op,
368
                        h264_chroma_mc_func chroma_op,vector_t *mv){
369
    MpegEncContext * const s = &h->s;
370
    const int mx= mv->x + src_x_offset*8;
371
    const int my= mv->y + src_y_offset*8;
372
    const int luma_xy= (mx&3) + ((my&3)<<2);
373
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->l_stride;
374
    uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->c_stride;
375
    uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->c_stride;
376
    int extra_width= 0; //(s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
377
    int extra_height= extra_width;
378
    int emu=0;
379
    const int full_mx= mx>>2;
380
    const int full_my= my>>2;
381
    const int pic_width  = 16*h->mb_width;
382
    const int pic_height = 16*h->mb_height;
383

    
384
    if(!pic->data[0])
385
        return;
386
    if(mx&7) extra_width -= 3;
387
    if(my&7) extra_height -= 3;
388

    
389
    if(   full_mx < 0-extra_width
390
          || full_my < 0-extra_height
391
          || full_mx + 16/*FIXME*/ > pic_width + extra_width
392
          || full_my + 16/*FIXME*/ > pic_height + extra_height){
393
        ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*h->l_stride, h->l_stride,
394
                            16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
395
        src_y= s->edge_emu_buffer + 2 + 2*h->l_stride;
396
        emu=1;
397
    }
398

    
399
    qpix_op[luma_xy](dest_y, src_y, h->l_stride); //FIXME try variable height perhaps?
400
    if(!square){
401
        qpix_op[luma_xy](dest_y + delta, src_y + delta, h->l_stride);
402
    }
403

    
404
    if(emu){
405
        ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, h->c_stride,
406
                            9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
407
        src_cb= s->edge_emu_buffer;
408
    }
409
    chroma_op(dest_cb, src_cb, h->c_stride, chroma_height, mx&7, my&7);
410

    
411
    if(emu){
412
        ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, h->c_stride,
413
                            9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
414
        src_cr= s->edge_emu_buffer;
415
    }
416
    chroma_op(dest_cr, src_cr, h->c_stride, chroma_height, mx&7, my&7);
417
}
418

    
419
static inline void mc_part_std(AVSContext *h,int square,int chroma_height,int delta,
420
                        uint8_t *dest_y,uint8_t *dest_cb,uint8_t *dest_cr,
421
                        int x_offset, int y_offset,qpel_mc_func *qpix_put,
422
                        h264_chroma_mc_func chroma_put,qpel_mc_func *qpix_avg,
423
                        h264_chroma_mc_func chroma_avg, vector_t *mv){
424
    qpel_mc_func *qpix_op=  qpix_put;
425
    h264_chroma_mc_func chroma_op= chroma_put;
426

    
427
    dest_y  += 2*x_offset + 2*y_offset*h->l_stride;
428
    dest_cb +=   x_offset +   y_offset*h->c_stride;
429
    dest_cr +=   x_offset +   y_offset*h->c_stride;
430
    x_offset += 8*h->mbx;
431
    y_offset += 8*h->mby;
432

    
433
    if(mv->ref >= 0){
434
        Picture *ref= &h->DPB[mv->ref];
435
        mc_dir_part(h, ref, square, chroma_height, delta, 0,
436
                    dest_y, dest_cb, dest_cr, x_offset, y_offset,
437
                    qpix_op, chroma_op, mv);
438

    
439
        qpix_op=  qpix_avg;
440
        chroma_op= chroma_avg;
441
    }
442

    
443
    if((mv+MV_BWD_OFFS)->ref >= 0){
444
        Picture *ref= &h->DPB[0];
445
        mc_dir_part(h, ref, square, chroma_height, delta, 1,
446
                    dest_y, dest_cb, dest_cr, x_offset, y_offset,
447
                    qpix_op, chroma_op, mv+MV_BWD_OFFS);
448
    }
449
}
450

    
451
static void inter_pred(AVSContext *h, enum mb_t mb_type) {
452
    if(partition_flags[mb_type] == 0){ // 16x16
453
        mc_part_std(h, 1, 8, 0, h->cy, h->cu, h->cv, 0, 0,
454
                h->s.dsp.put_cavs_qpel_pixels_tab[0],
455
                h->s.dsp.put_h264_chroma_pixels_tab[0],
456
                h->s.dsp.avg_cavs_qpel_pixels_tab[0],
457
                h->s.dsp.avg_h264_chroma_pixels_tab[0],&h->mv[MV_FWD_X0]);
458
    }else{
459
        mc_part_std(h, 1, 4, 0, h->cy, h->cu, h->cv, 0, 0,
460
                h->s.dsp.put_cavs_qpel_pixels_tab[1],
461
                h->s.dsp.put_h264_chroma_pixels_tab[1],
462
                h->s.dsp.avg_cavs_qpel_pixels_tab[1],
463
                h->s.dsp.avg_h264_chroma_pixels_tab[1],&h->mv[MV_FWD_X0]);
464
        mc_part_std(h, 1, 4, 0, h->cy, h->cu, h->cv, 4, 0,
465
                h->s.dsp.put_cavs_qpel_pixels_tab[1],
466
                h->s.dsp.put_h264_chroma_pixels_tab[1],
467
                h->s.dsp.avg_cavs_qpel_pixels_tab[1],
468
                h->s.dsp.avg_h264_chroma_pixels_tab[1],&h->mv[MV_FWD_X1]);
469
        mc_part_std(h, 1, 4, 0, h->cy, h->cu, h->cv, 0, 4,
470
                h->s.dsp.put_cavs_qpel_pixels_tab[1],
471
                h->s.dsp.put_h264_chroma_pixels_tab[1],
472
                h->s.dsp.avg_cavs_qpel_pixels_tab[1],
473
                h->s.dsp.avg_h264_chroma_pixels_tab[1],&h->mv[MV_FWD_X2]);
474
        mc_part_std(h, 1, 4, 0, h->cy, h->cu, h->cv, 4, 4,
475
                h->s.dsp.put_cavs_qpel_pixels_tab[1],
476
                h->s.dsp.put_h264_chroma_pixels_tab[1],
477
                h->s.dsp.avg_cavs_qpel_pixels_tab[1],
478
                h->s.dsp.avg_h264_chroma_pixels_tab[1],&h->mv[MV_FWD_X3]);
479
    }
480
    /* set intra prediction modes to default values */
481
    h->pred_mode_Y[3] =  h->pred_mode_Y[6] = INTRA_L_LP;
482
    h->top_pred_Y[h->mbx*2+0] = h->top_pred_Y[h->mbx*2+1] = INTRA_L_LP;
483
}
484

    
485
/*****************************************************************************
486
 *
487
 * motion vector prediction
488
 *
489
 ****************************************************************************/
490

    
491
static inline void set_mvs(vector_t *mv, enum block_t size) {
492
    switch(size) {
493
    case BLK_16X16:
494
        mv[MV_STRIDE  ] = mv[0];
495
        mv[MV_STRIDE+1] = mv[0];
496
    case BLK_16X8:
497
        mv[1] = mv[0];
498
        break;
499
    case BLK_8X16:
500
        mv[MV_STRIDE] = mv[0];
501
        break;
502
    }
503
}
504

    
505
static inline void store_mvs(AVSContext *h) {
506
    h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + 0] = h->mv[MV_FWD_X0];
507
    h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + 1] = h->mv[MV_FWD_X1];
508
    h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + 2] = h->mv[MV_FWD_X2];
509
    h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + 3] = h->mv[MV_FWD_X3];
510
}
511

    
512
static inline void scale_mv(AVSContext *h, int *d_x, int *d_y, vector_t *src, int distp) {
513
    int den = h->scale_den[src->ref];
514

    
515
    *d_x = (src->x*distp*den + 256 + (src->x>>31)) >> 9;
516
    *d_y = (src->y*distp*den + 256 + (src->y>>31)) >> 9;
517
}
518

    
519
static inline void mv_pred_median(AVSContext *h, vector_t *mvP, vector_t *mvA, vector_t *mvB, vector_t *mvC) {
520
    int ax, ay, bx, by, cx, cy;
521
    int len_ab, len_bc, len_ca, len_mid;
522

    
523
    /* scale candidates according to their temporal span */
524
    scale_mv(h, &ax, &ay, mvA, mvP->dist);
525
    scale_mv(h, &bx, &by, mvB, mvP->dist);
526
    scale_mv(h, &cx, &cy, mvC, mvP->dist);
527
    /* find the geometrical median of the three candidates */
528
    len_ab = abs(ax - bx) + abs(ay - by);
529
    len_bc = abs(bx - cx) + abs(by - cy);
530
    len_ca = abs(cx - ax) + abs(cy - ay);
531
    len_mid = mid_pred(len_ab, len_bc, len_ca);
532
    if(len_mid == len_ab) {
533
        mvP->x = cx;
534
        mvP->y = cy;
535
    } else if(len_mid == len_bc) {
536
        mvP->x = ax;
537
        mvP->y = ay;
538
    } else {
539
        mvP->x = bx;
540
        mvP->y = by;
541
    }
542
}
543

    
544
static inline void mv_pred_direct(AVSContext *h, vector_t *pmv_fw,
545
                                  vector_t *col_mv) {
546
    vector_t *pmv_bw = pmv_fw + MV_BWD_OFFS;
547
    int den = h->direct_den[col_mv->ref];
548
    int m = col_mv->x >> 31;
549

    
550
    pmv_fw->dist = h->dist[1];
551
    pmv_bw->dist = h->dist[0];
552
    pmv_fw->ref = 1;
553
    pmv_bw->ref = 0;
554
    /* scale the co-located motion vector according to its temporal span */
555
    pmv_fw->x = (((den+(den*col_mv->x*pmv_fw->dist^m)-m-1)>>14)^m)-m;
556
    pmv_bw->x = m-(((den+(den*col_mv->x*pmv_bw->dist^m)-m-1)>>14)^m);
557
    m = col_mv->y >> 31;
558
    pmv_fw->y = (((den+(den*col_mv->y*pmv_fw->dist^m)-m-1)>>14)^m)-m;
559
    pmv_bw->y = m-(((den+(den*col_mv->y*pmv_bw->dist^m)-m-1)>>14)^m);
560
}
561

    
562
static inline void mv_pred_sym(AVSContext *h, vector_t *src, enum block_t size) {
563
    vector_t *dst = src + MV_BWD_OFFS;
564

    
565
    /* backward mv is the scaled and negated forward mv */
566
    dst->x = -((src->x * h->sym_factor + 256) >> 9);
567
    dst->y = -((src->y * h->sym_factor + 256) >> 9);
568
    dst->ref = 0;
569
    dst->dist = h->dist[0];
570
    set_mvs(dst, size);
571
}
572

    
573
static void mv_pred(AVSContext *h, enum mv_loc_t nP, enum mv_loc_t nC,
574
                    enum mv_pred_t mode, enum block_t size, int ref) {
575
    vector_t *mvP = &h->mv[nP];
576
    vector_t *mvA = &h->mv[nP-1];
577
    vector_t *mvB = &h->mv[nP-4];
578
    vector_t *mvC = &h->mv[nC];
579
    vector_t *mvP2 = NULL;
580

    
581
    mvP->ref = ref;
582
    mvP->dist = h->dist[mvP->ref];
583
    if(mvC->ref == NOT_AVAIL)
584
        mvC = &h->mv[nP-5]; // set to top-left (mvD)
585
    if(mode == MV_PRED_PSKIP) {
586
        if((mvA->ref == NOT_AVAIL) || (mvB->ref == NOT_AVAIL) ||
587
           ((mvA->x | mvA->y | mvA->ref) == 0)  ||
588
           ((mvB->x | mvB->y | mvB->ref) == 0) ) {
589
            mvP->x = mvP->y = 0;
590
            set_mvs(mvP,size);
591
            return;
592
        }
593
    }
594
    /* if there is only one suitable candidate, take it */
595
    if((mvA->ref >= 0) && (mvB->ref < 0) && (mvC->ref < 0)) {
596
        mvP2= mvA;
597
    } else if((mvA->ref < 0) && (mvB->ref >= 0) && (mvC->ref < 0)) {
598
        mvP2= mvB;
599
    } else if((mvA->ref < 0) && (mvB->ref < 0) && (mvC->ref >= 0)) {
600
        mvP2= mvC;
601
    } else if(mode == MV_PRED_LEFT     && mvA->ref == ref){
602
        mvP2= mvA;
603
    } else if(mode == MV_PRED_TOP      && mvB->ref == ref){
604
        mvP2= mvB;
605
    } else if(mode == MV_PRED_TOPRIGHT && mvC->ref == ref){
606
        mvP2= mvC;
607
    }
608
    if(mvP2){
609
        mvP->x = mvP2->x;
610
        mvP->y = mvP2->y;
611
    }else
612
        mv_pred_median(h, mvP, mvA, mvB, mvC);
613

    
614
    if(mode < MV_PRED_PSKIP) {
615
        mvP->x += get_se_golomb(&h->s.gb);
616
        mvP->y += get_se_golomb(&h->s.gb);
617
    }
618
    set_mvs(mvP,size);
619
}
620

    
621
/*****************************************************************************
622
 *
623
 * residual data decoding
624
 *
625
 ****************************************************************************/
626

    
627
/** kth-order exponential golomb code */
628
static inline int get_ue_code(GetBitContext *gb, int order) {
629
    if(order) {
630
        int ret = get_ue_golomb(gb) << order;
631
        return ret + get_bits(gb,order);
632
    }
633
    return get_ue_golomb(gb);
634
}
635

    
636
/**
637
 * decode coefficients from one 8x8 block, dequantize, inverse transform
638
 *  and add them to sample block
639
 * @param r pointer to 2D VLC table
640
 * @param esc_golomb_order escape codes are k-golomb with this order k
641
 * @param qp quantizer
642
 * @param dst location of sample block
643
 * @param stride line stride in frame buffer
644
 */
645
static int decode_residual_block(AVSContext *h, GetBitContext *gb,
646
                                 const residual_vlc_t *r, int esc_golomb_order,
647
                                 int qp, uint8_t *dst, int stride) {
648
    int i,pos = -1;
649
    int level_code, esc_code, level, run, mask;
650
    int level_buf[64];
651
    int run_buf[64];
652
    int dqm = dequant_mul[qp];
653
    int dqs = dequant_shift[qp];
654
    int dqa = 1 << (dqs - 1);
655
    const uint8_t *scantab = ff_zigzag_direct;
656
    DCTELEM block[64];
657

    
658
    memset(block,0,64*sizeof(DCTELEM));
659
    for(i=0;i<65;i++) {
660
        level_code = get_ue_code(gb,r->golomb_order);
661
        if(level_code >= ESCAPE_CODE) {
662
            run = (level_code - ESCAPE_CODE) >> 1;
663
            esc_code = get_ue_code(gb,esc_golomb_order);
664
            level = esc_code + (run > r->max_run ? 1 : r->level_add[run]);
665
            while(level > r->inc_limit)
666
                r++;
667
            mask = -(level_code & 1);
668
            level = (level^mask) - mask;
669
        } else {
670
            if(level_code < 0)
671
                return -1;
672
            level = r->rltab[level_code][0];
673
            if(!level) //end of block signal
674
                break;
675
            run   = r->rltab[level_code][1];
676
            r += r->rltab[level_code][2];
677
        }
678
        level_buf[i] = level;
679
        run_buf[i] = run;
680
    }
681
    /* inverse scan and dequantization */
682
    while(--i >= 0){
683
        pos += 1 + run_buf[i];
684
        if(pos > 63) {
685
            av_log(h->s.avctx, AV_LOG_ERROR,
686
                   "position out of block bounds at pic %d MB(%d,%d)\n",
687
                   h->picture.poc, h->mbx, h->mby);
688
            return -1;
689
        }
690
        block[scantab[pos]] = (level_buf[i]*dqm + dqa) >> dqs;
691
    }
692
    h->s.dsp.cavs_idct8_add(dst,block,stride);
693
    return 0;
694
}
695

    
696

    
697
static inline void decode_residual_chroma(AVSContext *h) {
698
    if(h->cbp & (1<<4))
699
        decode_residual_block(h,&h->s.gb,chroma_2dvlc,0, chroma_qp[h->qp],
700
                              h->cu,h->c_stride);
701
    if(h->cbp & (1<<5))
702
        decode_residual_block(h,&h->s.gb,chroma_2dvlc,0, chroma_qp[h->qp],
703
                              h->cv,h->c_stride);
704
}
705

    
706
static inline int decode_residual_inter(AVSContext *h) {
707
    int block;
708

    
709
    /* get coded block pattern */
710
    int cbp= get_ue_golomb(&h->s.gb);
711
    if(cbp > 63){
712
        av_log(h->s.avctx, AV_LOG_ERROR, "illegal inter cbp\n");
713
        return -1;
714
    }
715
    h->cbp = cbp_tab[cbp][1];
716

    
717
    /* get quantizer */
718
    if(h->cbp && !h->qp_fixed)
719
        h->qp += get_se_golomb(&h->s.gb);
720
    for(block=0;block<4;block++)
721
        if(h->cbp & (1<<block))
722
            decode_residual_block(h,&h->s.gb,inter_2dvlc,0,h->qp,
723
                                  h->cy + h->luma_scan[block], h->l_stride);
724
    decode_residual_chroma(h);
725

    
726
    return 0;
727
}
728

    
729
/*****************************************************************************
730
 *
731
 * macroblock level
732
 *
733
 ****************************************************************************/
734

    
735
/**
736
 * initialise predictors for motion vectors and intra prediction
737
 */
738
static inline void init_mb(AVSContext *h) {
739
    int i;
740

    
741
    /* copy predictors from top line (MB B and C) into cache */
742
    for(i=0;i<3;i++) {
743
        h->mv[MV_FWD_B2+i] = h->top_mv[0][h->mbx*2+i];
744
        h->mv[MV_BWD_B2+i] = h->top_mv[1][h->mbx*2+i];
745
    }
746
    h->pred_mode_Y[1] = h->top_pred_Y[h->mbx*2+0];
747
    h->pred_mode_Y[2] = h->top_pred_Y[h->mbx*2+1];
748
    /* clear top predictors if MB B is not available */
749
    if(!(h->flags & B_AVAIL)) {
750
        h->mv[MV_FWD_B2] = un_mv;
751
        h->mv[MV_FWD_B3] = un_mv;
752
        h->mv[MV_BWD_B2] = un_mv;
753
        h->mv[MV_BWD_B3] = un_mv;
754
        h->pred_mode_Y[1] = h->pred_mode_Y[2] = NOT_AVAIL;
755
        h->flags &= ~(C_AVAIL|D_AVAIL);
756
    } else if(h->mbx) {
757
        h->flags |= D_AVAIL;
758
    }
759
    if(h->mbx == h->mb_width-1) //MB C not available
760
        h->flags &= ~C_AVAIL;
761
    /* clear top-right predictors if MB C is not available */
762
    if(!(h->flags & C_AVAIL)) {
763
        h->mv[MV_FWD_C2] = un_mv;
764
        h->mv[MV_BWD_C2] = un_mv;
765
    }
766
    /* clear top-left predictors if MB D is not available */
767
    if(!(h->flags & D_AVAIL)) {
768
        h->mv[MV_FWD_D3] = un_mv;
769
        h->mv[MV_BWD_D3] = un_mv;
770
    }
771
    /* set pointer for co-located macroblock type */
772
    h->col_type = &h->col_type_base[h->mby*h->mb_width + h->mbx];
773
}
774

    
775
static inline void check_for_slice(AVSContext *h);
776

    
777
/**
778
 * save predictors for later macroblocks and increase
779
 * macroblock address
780
 * @returns 0 if end of frame is reached, 1 otherwise
781
 */
782
static inline int next_mb(AVSContext *h) {
783
    int i;
784

    
785
    h->flags |= A_AVAIL;
786
    h->cy += 16;
787
    h->cu += 8;
788
    h->cv += 8;
789
    /* copy mvs as predictors to the left */
790
    for(i=0;i<=20;i+=4)
791
        h->mv[i] = h->mv[i+2];
792
    /* copy bottom mvs from cache to top line */
793
    h->top_mv[0][h->mbx*2+0] = h->mv[MV_FWD_X2];
794
    h->top_mv[0][h->mbx*2+1] = h->mv[MV_FWD_X3];
795
    h->top_mv[1][h->mbx*2+0] = h->mv[MV_BWD_X2];
796
    h->top_mv[1][h->mbx*2+1] = h->mv[MV_BWD_X3];
797
    /* next MB address */
798
    h->mbx++;
799
    if(h->mbx == h->mb_width) { //new mb line
800
        h->flags = B_AVAIL|C_AVAIL;
801
        /* clear left pred_modes */
802
        h->pred_mode_Y[3] = h->pred_mode_Y[6] = NOT_AVAIL;
803
        /* clear left mv predictors */
804
        for(i=0;i<=20;i+=4)
805
            h->mv[i] = un_mv;
806
        h->mbx = 0;
807
        h->mby++;
808
        /* re-calculate sample pointers */
809
        h->cy = h->picture.data[0] + h->mby*16*h->l_stride;
810
        h->cu = h->picture.data[1] + h->mby*8*h->c_stride;
811
        h->cv = h->picture.data[2] + h->mby*8*h->c_stride;
812
        if(h->mby == h->mb_height) { //frame end
813
            return 0;
814
        } else {
815
            //check_for_slice(h);
816
        }
817
    }
818
    return 1;
819
}
820

    
821
static int decode_mb_i(AVSContext *h, int cbp_code) {
822
    GetBitContext *gb = &h->s.gb;
823
    int block, pred_mode_uv;
824
    uint8_t top[18];
825
    uint8_t left[18];
826
    uint8_t *d;
827

    
828
    init_mb(h);
829

    
830
    /* get intra prediction modes from stream */
831
    for(block=0;block<4;block++) {
832
        int nA,nB,predpred;
833
        int pos = scan3x3[block];
834

    
835
        nA = h->pred_mode_Y[pos-1];
836
        nB = h->pred_mode_Y[pos-3];
837
        predpred = FFMIN(nA,nB);
838
        if(predpred == NOT_AVAIL) // if either is not available
839
            predpred = INTRA_L_LP;
840
        if(!get_bits1(gb)){
841
            int rem_mode= get_bits(gb, 2);
842
            predpred = rem_mode + (rem_mode >= predpred);
843
        }
844
        h->pred_mode_Y[pos] = predpred;
845
    }
846
    pred_mode_uv = get_ue_golomb(gb);
847
    if(pred_mode_uv > 6) {
848
        av_log(h->s.avctx, AV_LOG_ERROR, "illegal intra chroma pred mode\n");
849
        return -1;
850
    }
851

    
852
    /* save pred modes before they get modified */
853
    h->pred_mode_Y[3] =  h->pred_mode_Y[5];
854
    h->pred_mode_Y[6] =  h->pred_mode_Y[8];
855
    h->top_pred_Y[h->mbx*2+0] = h->pred_mode_Y[7];
856
    h->top_pred_Y[h->mbx*2+1] = h->pred_mode_Y[8];
857

    
858
    /* modify pred modes according to availability of neighbour samples */
859
    if(!(h->flags & A_AVAIL)) {
860
        modify_pred(left_modifier_l, &h->pred_mode_Y[4] );
861
        modify_pred(left_modifier_l, &h->pred_mode_Y[7] );
862
        modify_pred(left_modifier_c, &pred_mode_uv );
863
    }
864
    if(!(h->flags & B_AVAIL)) {
865
        modify_pred(top_modifier_l, &h->pred_mode_Y[4] );
866
        modify_pred(top_modifier_l, &h->pred_mode_Y[5] );
867
        modify_pred(top_modifier_c, &pred_mode_uv );
868
    }
869

    
870
    /* get coded block pattern */
871
    if(h->pic_type == FF_I_TYPE)
872
        cbp_code = get_ue_golomb(gb);
873
    if(cbp_code > 63){
874
        av_log(h->s.avctx, AV_LOG_ERROR, "illegal intra cbp\n");
875
        return -1;
876
    }
877
    h->cbp = cbp_tab[cbp_code][0];
878
    if(h->cbp && !h->qp_fixed)
879
        h->qp += get_se_golomb(gb); //qp_delta
880

    
881
    /* luma intra prediction interleaved with residual decode/transform/add */
882
    for(block=0;block<4;block++) {
883
        d = h->cy + h->luma_scan[block];
884
        load_intra_pred_luma(h, top, left, block);
885
        h->intra_pred_l[h->pred_mode_Y[scan3x3[block]]]
886
            (d, top, left, h->l_stride);
887
        if(h->cbp & (1<<block))
888
            decode_residual_block(h,gb,intra_2dvlc,1,h->qp,d,h->l_stride);
889
    }
890

    
891
    /* chroma intra prediction */
892
    /* extend borders by one pixel */
893
    h->left_border_u[9] = h->left_border_u[8];
894
    h->left_border_v[9] = h->left_border_v[8];
895
    h->top_border_u[h->mbx*10+9] = h->top_border_u[h->mbx*10+8];
896
    h->top_border_v[h->mbx*10+9] = h->top_border_v[h->mbx*10+8];
897
    if(h->mbx && h->mby) {
898
        h->top_border_u[h->mbx*10] = h->left_border_u[0] = h->topleft_border_u;
899
        h->top_border_v[h->mbx*10] = h->left_border_v[0] = h->topleft_border_v;
900
    } else {
901
        h->left_border_u[0] = h->left_border_u[1];
902
        h->left_border_v[0] = h->left_border_v[1];
903
        h->top_border_u[h->mbx*10] = h->top_border_u[h->mbx*10+1];
904
        h->top_border_v[h->mbx*10] = h->top_border_v[h->mbx*10+1];
905
    }
906
    h->intra_pred_c[pred_mode_uv](h->cu, &h->top_border_u[h->mbx*10],
907
                                  h->left_border_u, h->c_stride);
908
    h->intra_pred_c[pred_mode_uv](h->cv, &h->top_border_v[h->mbx*10],
909
                                  h->left_border_v, h->c_stride);
910

    
911
    decode_residual_chroma(h);
912
    filter_mb(h,I_8X8);
913

    
914
    /* mark motion vectors as intra */
915
    h->mv[MV_FWD_X0] = intra_mv;
916
    set_mvs(&h->mv[MV_FWD_X0], BLK_16X16);
917
    h->mv[MV_BWD_X0] = intra_mv;
918
    set_mvs(&h->mv[MV_BWD_X0], BLK_16X16);
919
    if(h->pic_type != FF_B_TYPE)
920
        *h->col_type = I_8X8;
921

    
922
    return 0;
923
}
924

    
925
static void decode_mb_p(AVSContext *h, enum mb_t mb_type) {
926
    GetBitContext *gb = &h->s.gb;
927
    int ref[4];
928

    
929
    init_mb(h);
930
    switch(mb_type) {
931
    case P_SKIP:
932
        mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_PSKIP, BLK_16X16, 0);
933
        break;
934
    case P_16X16:
935
        ref[0] = h->ref_flag ? 0 : get_bits1(gb);
936
        mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_MEDIAN,   BLK_16X16,ref[0]);
937
        break;
938
    case P_16X8:
939
        ref[0] = h->ref_flag ? 0 : get_bits1(gb);
940
        ref[2] = h->ref_flag ? 0 : get_bits1(gb);
941
        mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_TOP,      BLK_16X8, ref[0]);
942
        mv_pred(h, MV_FWD_X2, MV_FWD_A1, MV_PRED_LEFT,     BLK_16X8, ref[2]);
943
        break;
944
    case P_8X16:
945
        ref[0] = h->ref_flag ? 0 : get_bits1(gb);
946
        ref[1] = h->ref_flag ? 0 : get_bits1(gb);
947
        mv_pred(h, MV_FWD_X0, MV_FWD_B3, MV_PRED_LEFT,     BLK_8X16, ref[0]);
948
        mv_pred(h, MV_FWD_X1, MV_FWD_C2, MV_PRED_TOPRIGHT, BLK_8X16, ref[1]);
949
        break;
950
    case P_8X8:
951
        ref[0] = h->ref_flag ? 0 : get_bits1(gb);
952
        ref[1] = h->ref_flag ? 0 : get_bits1(gb);
953
        ref[2] = h->ref_flag ? 0 : get_bits1(gb);
954
        ref[3] = h->ref_flag ? 0 : get_bits1(gb);
955
        mv_pred(h, MV_FWD_X0, MV_FWD_B3, MV_PRED_MEDIAN,   BLK_8X8, ref[0]);
956
        mv_pred(h, MV_FWD_X1, MV_FWD_C2, MV_PRED_MEDIAN,   BLK_8X8, ref[1]);
957
        mv_pred(h, MV_FWD_X2, MV_FWD_X1, MV_PRED_MEDIAN,   BLK_8X8, ref[2]);
958
        mv_pred(h, MV_FWD_X3, MV_FWD_X0, MV_PRED_MEDIAN,   BLK_8X8, ref[3]);
959
    }
960
    inter_pred(h, mb_type);
961
    store_mvs(h);
962
    if(mb_type != P_SKIP)
963
        decode_residual_inter(h);
964
    filter_mb(h,mb_type);
965
    *h->col_type = mb_type;
966
}
967

    
968
static void decode_mb_b(AVSContext *h, enum mb_t mb_type) {
969
    int block;
970
    enum sub_mb_t sub_type[4];
971
    int flags;
972

    
973
    init_mb(h);
974

    
975
    /* reset all MVs */
976
    h->mv[MV_FWD_X0] = dir_mv;
977
    set_mvs(&h->mv[MV_FWD_X0], BLK_16X16);
978
    h->mv[MV_BWD_X0] = dir_mv;
979
    set_mvs(&h->mv[MV_BWD_X0], BLK_16X16);
980
    switch(mb_type) {
981
    case B_SKIP:
982
    case B_DIRECT:
983
        if(!(*h->col_type)) {
984
            /* intra MB at co-location, do in-plane prediction */
985
            mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_BSKIP, BLK_16X16, 1);
986
            mv_pred(h, MV_BWD_X0, MV_BWD_C2, MV_PRED_BSKIP, BLK_16X16, 0);
987
        } else
988
            /* direct prediction from co-located P MB, block-wise */
989
            for(block=0;block<4;block++)
990
                mv_pred_direct(h,&h->mv[mv_scan[block]],
991
                            &h->col_mv[(h->mby*h->mb_width+h->mbx)*4 + block]);
992
        break;
993
    case B_FWD_16X16:
994
        mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_MEDIAN, BLK_16X16, 1);
995
        break;
996
    case B_SYM_16X16:
997
        mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_MEDIAN, BLK_16X16, 1);
998
        mv_pred_sym(h, &h->mv[MV_FWD_X0], BLK_16X16);
999
        break;
1000
    case B_BWD_16X16:
1001
        mv_pred(h, MV_BWD_X0, MV_BWD_C2, MV_PRED_MEDIAN, BLK_16X16, 0);
1002
        break;
1003
    case B_8X8:
1004
        for(block=0;block<4;block++)
1005
            sub_type[block] = get_bits(&h->s.gb,2);
1006
        for(block=0;block<4;block++) {
1007
            switch(sub_type[block]) {
1008
            case B_SUB_DIRECT:
1009
                if(!(*h->col_type)) {
1010
                    /* intra MB at co-location, do in-plane prediction */
1011
                    mv_pred(h, mv_scan[block], mv_scan[block]-3,
1012
                            MV_PRED_BSKIP, BLK_8X8, 1);
1013
                    mv_pred(h, mv_scan[block]+MV_BWD_OFFS,
1014
                            mv_scan[block]-3+MV_BWD_OFFS,
1015
                            MV_PRED_BSKIP, BLK_8X8, 0);
1016
                } else
1017
                    mv_pred_direct(h,&h->mv[mv_scan[block]],
1018
                                   &h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + block]);
1019
                break;
1020
            case B_SUB_FWD:
1021
                mv_pred(h, mv_scan[block], mv_scan[block]-3,
1022
                        MV_PRED_MEDIAN, BLK_8X8, 1);
1023
                break;
1024
            case B_SUB_SYM:
1025
                mv_pred(h, mv_scan[block], mv_scan[block]-3,
1026
                        MV_PRED_MEDIAN, BLK_8X8, 1);
1027
                mv_pred_sym(h, &h->mv[mv_scan[block]], BLK_8X8);
1028
                break;
1029
            }
1030
        }
1031
        for(block=0;block<4;block++) {
1032
            if(sub_type[block] == B_SUB_BWD)
1033
                mv_pred(h, mv_scan[block]+MV_BWD_OFFS,
1034
                        mv_scan[block]+MV_BWD_OFFS-3,
1035
                        MV_PRED_MEDIAN, BLK_8X8, 0);
1036
        }
1037
        break;
1038
    default:
1039
        assert((mb_type > B_SYM_16X16) && (mb_type < B_8X8));
1040
        flags = partition_flags[mb_type];
1041
        if(mb_type & 1) { /* 16x8 macroblock types */
1042
            if(flags & FWD0)
1043
                mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_TOP,  BLK_16X8, 1);
1044
            if(flags & SYM0) {
1045
                mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_TOP,  BLK_16X8, 1);
1046
                mv_pred_sym(h, &h->mv[MV_FWD_X0], BLK_16X8);
1047
            }
1048
            if(flags & FWD1)
1049
                mv_pred(h, MV_FWD_X2, MV_FWD_A1, MV_PRED_LEFT, BLK_16X8, 1);
1050
            if(flags & SYM1) {
1051
                mv_pred(h, MV_FWD_X2, MV_FWD_A1, MV_PRED_LEFT, BLK_16X8, 1);
1052
                mv_pred_sym(h, &h->mv[9], BLK_16X8);
1053
            }
1054
            if(flags & BWD0)
1055
                mv_pred(h, MV_BWD_X0, MV_BWD_C2, MV_PRED_TOP,  BLK_16X8, 0);
1056
            if(flags & BWD1)
1057
                mv_pred(h, MV_BWD_X2, MV_BWD_A1, MV_PRED_LEFT, BLK_16X8, 0);
1058
        } else {          /* 8x16 macroblock types */
1059
            if(flags & FWD0)
1060
                mv_pred(h, MV_FWD_X0, MV_FWD_B3, MV_PRED_LEFT, BLK_8X16, 1);
1061
            if(flags & SYM0) {
1062
                mv_pred(h, MV_FWD_X0, MV_FWD_B3, MV_PRED_LEFT, BLK_8X16, 1);
1063
                mv_pred_sym(h, &h->mv[MV_FWD_X0], BLK_8X16);
1064
            }
1065
            if(flags & FWD1)
1066
                mv_pred(h, MV_FWD_X1, MV_FWD_C2, MV_PRED_TOPRIGHT,BLK_8X16, 1);
1067
            if(flags & SYM1) {
1068
                mv_pred(h, MV_FWD_X1, MV_FWD_C2, MV_PRED_TOPRIGHT,BLK_8X16, 1);
1069
                mv_pred_sym(h, &h->mv[6], BLK_8X16);
1070
            }
1071
            if(flags & BWD0)
1072
                mv_pred(h, MV_BWD_X0, MV_BWD_B3, MV_PRED_LEFT, BLK_8X16, 0);
1073
            if(flags & BWD1)
1074
                mv_pred(h, MV_BWD_X1, MV_BWD_C2, MV_PRED_TOPRIGHT,BLK_8X16, 0);
1075
        }
1076
    }
1077
    inter_pred(h, mb_type);
1078
    if(mb_type != B_SKIP)
1079
        decode_residual_inter(h);
1080
    filter_mb(h,mb_type);
1081
}
1082

    
1083
/*****************************************************************************
1084
 *
1085
 * slice level
1086
 *
1087
 ****************************************************************************/
1088

    
1089
static inline int decode_slice_header(AVSContext *h, GetBitContext *gb) {
1090
    if(h->stc > 0xAF)
1091
        av_log(h->s.avctx, AV_LOG_ERROR, "unexpected start code 0x%02x\n", h->stc);
1092
    h->mby = h->stc;
1093
    if((h->mby == 0) && (!h->qp_fixed)){
1094
        h->qp_fixed = get_bits1(gb);
1095
        h->qp = get_bits(gb,6);
1096
    }
1097
    /* inter frame or second slice can have weighting params */
1098
    if((h->pic_type != FF_I_TYPE) || (!h->pic_structure && h->mby >= h->mb_width/2))
1099
        if(get_bits1(gb)) { //slice_weighting_flag
1100
            av_log(h->s.avctx, AV_LOG_ERROR,
1101
                   "weighted prediction not yet supported\n");
1102
        }
1103
    return 0;
1104
}
1105

    
1106
static inline void check_for_slice(AVSContext *h) {
1107
    GetBitContext *gb = &h->s.gb;
1108
    int align;
1109
    align = (-get_bits_count(gb)) & 7;
1110
    if((show_bits_long(gb,24+align) & 0xFFFFFF) == 0x000001) {
1111
        get_bits_long(gb,24+align);
1112
        h->stc = get_bits(gb,8);
1113
        decode_slice_header(h,gb);
1114
    }
1115
}
1116

    
1117
/*****************************************************************************
1118
 *
1119
 * frame level
1120
 *
1121
 ****************************************************************************/
1122

    
1123
static void init_pic(AVSContext *h) {
1124
    int i;
1125

    
1126
    /* clear some predictors */
1127
    for(i=0;i<=20;i+=4)
1128
        h->mv[i] = un_mv;
1129
    h->mv[MV_BWD_X0] = dir_mv;
1130
    set_mvs(&h->mv[MV_BWD_X0], BLK_16X16);
1131
    h->mv[MV_FWD_X0] = dir_mv;
1132
    set_mvs(&h->mv[MV_FWD_X0], BLK_16X16);
1133
    h->pred_mode_Y[3] = h->pred_mode_Y[6] = NOT_AVAIL;
1134
    h->cy = h->picture.data[0];
1135
    h->cu = h->picture.data[1];
1136
    h->cv = h->picture.data[2];
1137
    h->l_stride = h->picture.linesize[0];
1138
    h->c_stride = h->picture.linesize[1];
1139
    h->luma_scan[2] = 8*h->l_stride;
1140
    h->luma_scan[3] = 8*h->l_stride+8;
1141
    h->mbx = h->mby = 0;
1142
    h->flags = 0;
1143
}
1144

    
1145
static int decode_pic(AVSContext *h) {
1146
    MpegEncContext *s = &h->s;
1147
    int skip_count;
1148
    enum mb_t mb_type;
1149

    
1150
    if (!s->context_initialized) {
1151
        if (MPV_common_init(s) < 0)
1152
            return -1;
1153
    }
1154
    get_bits(&s->gb,16);//bbv_dwlay
1155
    if(h->stc == PIC_PB_START_CODE) {
1156
        h->pic_type = get_bits(&s->gb,2) + FF_I_TYPE;
1157
        /* make sure we have the reference frames we need */
1158
        if(!h->DPB[0].data[0] ||
1159
          (!h->DPB[1].data[0] && h->pic_type == FF_B_TYPE))
1160
            return -1;
1161
    } else {
1162
        h->pic_type = FF_I_TYPE;
1163
        if(get_bits1(&s->gb))
1164
            get_bits(&s->gb,16);//time_code
1165
    }
1166
    /* release last B frame */
1167
    if(h->picture.data[0])
1168
        s->avctx->release_buffer(s->avctx, (AVFrame *)&h->picture);
1169

    
1170
    s->avctx->get_buffer(s->avctx, (AVFrame *)&h->picture);
1171
    init_pic(h);
1172
    h->picture.poc = get_bits(&s->gb,8)*2;
1173

    
1174
    /* get temporal distances and MV scaling factors */
1175
    if(h->pic_type != FF_B_TYPE) {
1176
        h->dist[0] = (h->picture.poc - h->DPB[0].poc  + 512) % 512;
1177
    } else {
1178
        h->dist[0] = (h->DPB[0].poc  - h->picture.poc + 512) % 512;
1179
    }
1180
    h->dist[1] = (h->picture.poc - h->DPB[1].poc  + 512) % 512;
1181
    h->scale_den[0] = h->dist[0] ? 512/h->dist[0] : 0;
1182
    h->scale_den[1] = h->dist[1] ? 512/h->dist[1] : 0;
1183
    if(h->pic_type == FF_B_TYPE) {
1184
        h->sym_factor = h->dist[0]*h->scale_den[1];
1185
    } else {
1186
        h->direct_den[0] = h->dist[0] ? 16384/h->dist[0] : 0;
1187
        h->direct_den[1] = h->dist[1] ? 16384/h->dist[1] : 0;
1188
    }
1189

    
1190
    if(s->low_delay)
1191
        get_ue_golomb(&s->gb); //bbv_check_times
1192
    h->progressive             = get_bits1(&s->gb);
1193
    if(h->progressive)
1194
        h->pic_structure = 1;
1195
    else if(!(h->pic_structure = get_bits1(&s->gb) && (h->stc == PIC_PB_START_CODE)) )
1196
        get_bits1(&s->gb);     //advanced_pred_mode_disable
1197
    skip_bits1(&s->gb);        //top_field_first
1198
    skip_bits1(&s->gb);        //repeat_first_field
1199
    h->qp_fixed                = get_bits1(&s->gb);
1200
    h->qp                      = get_bits(&s->gb,6);
1201
    if(h->pic_type == FF_I_TYPE) {
1202
        if(!h->progressive && !h->pic_structure)
1203
            skip_bits1(&s->gb);//what is this?
1204
        skip_bits(&s->gb,4);   //reserved bits
1205
    } else {
1206
        if(!(h->pic_type == FF_B_TYPE && h->pic_structure == 1))
1207
            h->ref_flag        = get_bits1(&s->gb);
1208
        skip_bits(&s->gb,4);   //reserved bits
1209
        h->skip_mode_flag      = get_bits1(&s->gb);
1210
    }
1211
    h->loop_filter_disable     = get_bits1(&s->gb);
1212
    if(!h->loop_filter_disable && get_bits1(&s->gb)) {
1213
        h->alpha_offset        = get_se_golomb(&s->gb);
1214
        h->beta_offset         = get_se_golomb(&s->gb);
1215
    } else {
1216
        h->alpha_offset = h->beta_offset  = 0;
1217
    }
1218
    check_for_slice(h);
1219
    if(h->pic_type == FF_I_TYPE) {
1220
        do {
1221
            decode_mb_i(h, 0);
1222
        } while(next_mb(h));
1223
    } else if(h->pic_type == FF_P_TYPE) {
1224
        do {
1225
            if(h->skip_mode_flag) {
1226
                skip_count = get_ue_golomb(&s->gb);
1227
                while(skip_count--) {
1228
                    decode_mb_p(h,P_SKIP);
1229
                    if(!next_mb(h))
1230
                        goto done;
1231
                }
1232
                mb_type = get_ue_golomb(&s->gb) + P_16X16;
1233
            } else
1234
                mb_type = get_ue_golomb(&s->gb) + P_SKIP;
1235
            if(mb_type > P_8X8) {
1236
                decode_mb_i(h, mb_type - P_8X8 - 1);
1237
            } else
1238
                decode_mb_p(h,mb_type);
1239
        } while(next_mb(h));
1240
    } else { /* FF_B_TYPE */
1241
        do {
1242
            if(h->skip_mode_flag) {
1243
                skip_count = get_ue_golomb(&s->gb);
1244
                while(skip_count--) {
1245
                    decode_mb_b(h,B_SKIP);
1246
                    if(!next_mb(h))
1247
                        goto done;
1248
                }
1249
                mb_type = get_ue_golomb(&s->gb) + B_DIRECT;
1250
            } else
1251
                mb_type = get_ue_golomb(&s->gb) + B_SKIP;
1252
            if(mb_type > B_8X8) {
1253
                decode_mb_i(h, mb_type - B_8X8 - 1);
1254
            } else
1255
                decode_mb_b(h,mb_type);
1256
        } while(next_mb(h));
1257
    }
1258
 done:
1259
    if(h->pic_type != FF_B_TYPE) {
1260
        if(h->DPB[1].data[0])
1261
            s->avctx->release_buffer(s->avctx, (AVFrame *)&h->DPB[1]);
1262
        memcpy(&h->DPB[1], &h->DPB[0], sizeof(Picture));
1263
        memcpy(&h->DPB[0], &h->picture, sizeof(Picture));
1264
        memset(&h->picture,0,sizeof(Picture));
1265
    }
1266
    return 0;
1267
}
1268

    
1269
/*****************************************************************************
1270
 *
1271
 * headers and interface
1272
 *
1273
 ****************************************************************************/
1274

    
1275
/**
1276
 * some predictions require data from the top-neighbouring macroblock.
1277
 * this data has to be stored for one complete row of macroblocks
1278
 * and this storage space is allocated here
1279
 */
1280
static void init_top_lines(AVSContext *h) {
1281
    /* alloc top line of predictors */
1282
    h->top_qp       = av_malloc( h->mb_width);
1283
    h->top_mv[0]    = av_malloc((h->mb_width*2+1)*sizeof(vector_t));
1284
    h->top_mv[1]    = av_malloc((h->mb_width*2+1)*sizeof(vector_t));
1285
    h->top_pred_Y   = av_malloc( h->mb_width*2*sizeof(*h->top_pred_Y));
1286
    h->top_border_y = av_malloc((h->mb_width+1)*16);
1287
    h->top_border_u = av_malloc((h->mb_width)*10);
1288
    h->top_border_v = av_malloc((h->mb_width)*10);
1289

    
1290
    /* alloc space for co-located MVs and types */
1291
    h->col_mv       = av_malloc( h->mb_width*h->mb_height*4*sizeof(vector_t));
1292
    h->col_type_base = av_malloc(h->mb_width*h->mb_height);
1293
}
1294

    
1295
static int decode_seq_header(AVSContext *h) {
1296
    MpegEncContext *s = &h->s;
1297
    extern const AVRational ff_frame_rate_tab[];
1298
    int frame_rate_code;
1299

    
1300
    h->profile =         get_bits(&s->gb,8);
1301
    h->level =           get_bits(&s->gb,8);
1302
    skip_bits1(&s->gb); //progressive sequence
1303
    s->width =           get_bits(&s->gb,14);
1304
    s->height =          get_bits(&s->gb,14);
1305
    skip_bits(&s->gb,2); //chroma format
1306
    skip_bits(&s->gb,3); //sample_precision
1307
    h->aspect_ratio =    get_bits(&s->gb,4);
1308
    frame_rate_code =    get_bits(&s->gb,4);
1309
    skip_bits(&s->gb,18);//bit_rate_lower
1310
    skip_bits1(&s->gb);  //marker_bit
1311
    skip_bits(&s->gb,12);//bit_rate_upper
1312
    s->low_delay =       get_bits1(&s->gb);
1313
    h->mb_width  = (s->width  + 15) >> 4;
1314
    h->mb_height = (s->height + 15) >> 4;
1315
    h->s.avctx->time_base.den = ff_frame_rate_tab[frame_rate_code].num;
1316
    h->s.avctx->time_base.num = ff_frame_rate_tab[frame_rate_code].den;
1317
    h->s.avctx->width  = s->width;
1318
    h->s.avctx->height = s->height;
1319
    if(!h->top_qp)
1320
        init_top_lines(h);
1321
    return 0;
1322
}
1323

    
1324
/**
1325
 * finds the end of the current frame in the bitstream.
1326
 * @return the position of the first byte of the next frame, or -1
1327
 */
1328
int ff_cavs_find_frame_end(ParseContext *pc, const uint8_t *buf, int buf_size) {
1329
    int pic_found, i;
1330
    uint32_t state;
1331

    
1332
    pic_found= pc->frame_start_found;
1333
    state= pc->state;
1334

    
1335
    i=0;
1336
    if(!pic_found){
1337
        for(i=0; i<buf_size; i++){
1338
            state= (state<<8) | buf[i];
1339
            if(state == PIC_I_START_CODE || state == PIC_PB_START_CODE){
1340
                i++;
1341
                pic_found=1;
1342
                break;
1343
            }
1344
        }
1345
    }
1346

    
1347
    if(pic_found){
1348
        /* EOF considered as end of frame */
1349
        if (buf_size == 0)
1350
            return 0;
1351
        for(; i<buf_size; i++){
1352
            state= (state<<8) | buf[i];
1353
            if((state&0xFFFFFF00) == 0x100){
1354
                if(state < SLICE_MIN_START_CODE || state > SLICE_MAX_START_CODE){
1355
                    pc->frame_start_found=0;
1356
                    pc->state=-1;
1357
                    return i-3;
1358
                }
1359
            }
1360
        }
1361
    }
1362
    pc->frame_start_found= pic_found;
1363
    pc->state= state;
1364
    return END_NOT_FOUND;
1365
}
1366

    
1367
void ff_cavs_flush(AVCodecContext * avctx) {
1368
    AVSContext *h = avctx->priv_data;
1369
    h->got_keyframe = 0;
1370
}
1371

    
1372
static int cavs_decode_frame(AVCodecContext * avctx,void *data, int *data_size,
1373
                             uint8_t * buf, int buf_size) {
1374
    AVSContext *h = avctx->priv_data;
1375
    MpegEncContext *s = &h->s;
1376
    int input_size;
1377
    const uint8_t *buf_end;
1378
    const uint8_t *buf_ptr;
1379
    AVFrame *picture = data;
1380
    uint32_t stc;
1381

    
1382
    s->avctx = avctx;
1383

    
1384
    if (buf_size == 0) {
1385
        if(!s->low_delay && h->DPB[0].data[0]) {
1386
            *data_size = sizeof(AVPicture);
1387
            *picture = *(AVFrame *) &h->DPB[0];
1388
        }
1389
        return 0;
1390
    }
1391

    
1392
    buf_ptr = buf;
1393
    buf_end = buf + buf_size;
1394
    for(;;) {
1395
        buf_ptr = ff_find_start_code(buf_ptr,buf_end, &stc);
1396
        if(stc & 0xFFFFFE00)
1397
            return FFMAX(0, buf_ptr - buf - s->parse_context.last_index);
1398
        input_size = (buf_end - buf_ptr)*8;
1399
        switch(stc) {
1400
        case SEQ_START_CODE:
1401
            init_get_bits(&s->gb, buf_ptr, input_size);
1402
            decode_seq_header(h);
1403
            break;
1404
        case PIC_I_START_CODE:
1405
            if(!h->got_keyframe) {
1406
                if(h->DPB[0].data[0])
1407
                    avctx->release_buffer(avctx, (AVFrame *)&h->DPB[0]);
1408
                if(h->DPB[1].data[0])
1409
                    avctx->release_buffer(avctx, (AVFrame *)&h->DPB[1]);
1410
                h->got_keyframe = 1;
1411
            }
1412
        case PIC_PB_START_CODE:
1413
            *data_size = 0;
1414
            if(!h->got_keyframe)
1415
                break;
1416
            init_get_bits(&s->gb, buf_ptr, input_size);
1417
            h->stc = stc;
1418
            if(decode_pic(h))
1419
                break;
1420
            *data_size = sizeof(AVPicture);
1421
            if(h->pic_type != FF_B_TYPE) {
1422
                if(h->DPB[1].data[0]) {
1423
                    *picture = *(AVFrame *) &h->DPB[1];
1424
                } else {
1425
                    *data_size = 0;
1426
                }
1427
            } else
1428
                *picture = *(AVFrame *) &h->picture;
1429
            break;
1430
        case EXT_START_CODE:
1431
            //mpeg_decode_extension(avctx,buf_ptr, input_size);
1432
            break;
1433
        case USER_START_CODE:
1434
            //mpeg_decode_user_data(avctx,buf_ptr, input_size);
1435
            break;
1436
        default:
1437
            if (stc >= SLICE_MIN_START_CODE &&
1438
                stc <= SLICE_MAX_START_CODE) {
1439
                init_get_bits(&s->gb, buf_ptr, input_size);
1440
                decode_slice_header(h, &s->gb);
1441
            }
1442
            break;
1443
        }
1444
    }
1445
}
1446

    
1447
static int cavs_decode_init(AVCodecContext * avctx) {
1448
    AVSContext *h = avctx->priv_data;
1449
    MpegEncContext * const s = &h->s;
1450

    
1451
    MPV_decode_defaults(s);
1452
    s->avctx = avctx;
1453

    
1454
    avctx->pix_fmt= PIX_FMT_YUV420P;
1455

    
1456
    h->luma_scan[0] = 0;
1457
    h->luma_scan[1] = 8;
1458
    h->intra_pred_l[      INTRA_L_VERT] = intra_pred_vert;
1459
    h->intra_pred_l[     INTRA_L_HORIZ] = intra_pred_horiz;
1460
    h->intra_pred_l[        INTRA_L_LP] = intra_pred_lp;
1461
    h->intra_pred_l[ INTRA_L_DOWN_LEFT] = intra_pred_down_left;
1462
    h->intra_pred_l[INTRA_L_DOWN_RIGHT] = intra_pred_down_right;
1463
    h->intra_pred_l[   INTRA_L_LP_LEFT] = intra_pred_lp_left;
1464
    h->intra_pred_l[    INTRA_L_LP_TOP] = intra_pred_lp_top;
1465
    h->intra_pred_l[    INTRA_L_DC_128] = intra_pred_dc_128;
1466
    h->intra_pred_c[        INTRA_C_LP] = intra_pred_lp;
1467
    h->intra_pred_c[     INTRA_C_HORIZ] = intra_pred_horiz;
1468
    h->intra_pred_c[      INTRA_C_VERT] = intra_pred_vert;
1469
    h->intra_pred_c[     INTRA_C_PLANE] = intra_pred_plane;
1470
    h->intra_pred_c[   INTRA_C_LP_LEFT] = intra_pred_lp_left;
1471
    h->intra_pred_c[    INTRA_C_LP_TOP] = intra_pred_lp_top;
1472
    h->intra_pred_c[    INTRA_C_DC_128] = intra_pred_dc_128;
1473
    h->mv[ 7] = un_mv;
1474
    h->mv[19] = un_mv;
1475
    return 0;
1476
}
1477

    
1478
static int cavs_decode_end(AVCodecContext * avctx) {
1479
    AVSContext *h = avctx->priv_data;
1480

    
1481
    av_free(h->top_qp);
1482
    av_free(h->top_mv[0]);
1483
    av_free(h->top_mv[1]);
1484
    av_free(h->top_pred_Y);
1485
    av_free(h->top_border_y);
1486
    av_free(h->top_border_u);
1487
    av_free(h->top_border_v);
1488
    av_free(h->col_mv);
1489
    av_free(h->col_type_base);
1490
    return 0;
1491
}
1492

    
1493
AVCodec cavs_decoder = {
1494
    "cavs",
1495
    CODEC_TYPE_VIDEO,
1496
    CODEC_ID_CAVS,
1497
    sizeof(AVSContext),
1498
    cavs_decode_init,
1499
    NULL,
1500
    cavs_decode_end,
1501
    cavs_decode_frame,
1502
    CODEC_CAP_DR1 | CODEC_CAP_DELAY,
1503
    .flush= ff_cavs_flush,
1504
};