Statistics
| Branch: | Revision:

ffmpeg / libavcodec / rv34.c @ 77f7156d

History | View | Annotate | Download (51.6 KB)

1
/*
2
 * RV30/40 decoder common data
3
 * Copyright (c) 2007 Mike Melanson, Konstantin Shishkov
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/rv34.c
24
 * RV30/40 decoder common data
25
 */
26

    
27
#include "avcodec.h"
28
#include "dsputil.h"
29
#include "mpegvideo.h"
30
#include "golomb.h"
31
#include "mathops.h"
32
#include "rectangle.h"
33

    
34
#include "rv34vlc.h"
35
#include "rv34data.h"
36
#include "rv34.h"
37

    
38
//#define DEBUG
39

    
40
/** translation of RV30/40 macroblock types to lavc ones */
41
static const int rv34_mb_type_to_lavc[12] = {
42
    MB_TYPE_INTRA,
43
    MB_TYPE_INTRA16x16              | MB_TYPE_SEPARATE_DC,
44
    MB_TYPE_16x16   | MB_TYPE_L0,
45
    MB_TYPE_8x8     | MB_TYPE_L0,
46
    MB_TYPE_16x16   | MB_TYPE_L0,
47
    MB_TYPE_16x16   | MB_TYPE_L1,
48
    MB_TYPE_SKIP,
49
    MB_TYPE_DIRECT2 | MB_TYPE_16x16,
50
    MB_TYPE_16x8    | MB_TYPE_L0,
51
    MB_TYPE_8x16    | MB_TYPE_L0,
52
    MB_TYPE_16x16   | MB_TYPE_L0L1,
53
    MB_TYPE_16x16   | MB_TYPE_L0    | MB_TYPE_SEPARATE_DC
54
};
55

    
56

    
57
static RV34VLC intra_vlcs[NUM_INTRA_TABLES], inter_vlcs[NUM_INTER_TABLES];
58

    
59
/**
60
 * @defgroup vlc RV30/40 VLC generating functions
61
 * @{
62
 */
63

    
64
/**
65
 * Generate VLC from codeword lengths.
66
 * @param bits   codeword lengths (zeroes are accepted)
67
 * @param size   length of input data
68
 * @param vlc    output VLC
69
 * @param insyms symbols for input codes (NULL for default ones)
70
 */
71
static void rv34_gen_vlc(const uint8_t *bits, int size, VLC *vlc, const uint8_t *insyms)
72
{
73
    int i;
74
    int counts[17] = {0}, codes[17];
75
    uint16_t cw[size], syms[size];
76
    uint8_t bits2[size];
77
    int maxbits = 0, realsize = 0;
78

    
79
    for(i = 0; i < size; i++){
80
        if(bits[i]){
81
            bits2[realsize] = bits[i];
82
            syms[realsize] = insyms ? insyms[i] : i;
83
            realsize++;
84
            maxbits = FFMAX(maxbits, bits[i]);
85
            counts[bits[i]]++;
86
        }
87
    }
88

    
89
    codes[0] = 0;
90
    for(i = 0; i < 16; i++)
91
        codes[i+1] = (codes[i] + counts[i]) << 1;
92
    for(i = 0; i < realsize; i++)
93
        cw[i] = codes[bits2[i]]++;
94

    
95
    init_vlc_sparse(vlc, FFMIN(maxbits, 9), realsize,
96
                    bits2, 1, 1,
97
                    cw,    2, 2,
98
                    syms,  2, 2, INIT_VLC_USE_STATIC);
99
}
100

    
101
/**
102
 * Initialize all tables.
103
 */
104
static av_cold void rv34_init_tables(void)
105
{
106
    int i, j, k;
107

    
108
    for(i = 0; i < NUM_INTRA_TABLES; i++){
109
        for(j = 0; j < 2; j++){
110
            rv34_gen_vlc(rv34_table_intra_cbppat   [i][j], CBPPAT_VLC_SIZE,   &intra_vlcs[i].cbppattern[j],     NULL);
111
            rv34_gen_vlc(rv34_table_intra_secondpat[i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].second_pattern[j], NULL);
112
            rv34_gen_vlc(rv34_table_intra_thirdpat [i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].third_pattern[j],  NULL);
113
            for(k = 0; k < 4; k++)
114
                rv34_gen_vlc(rv34_table_intra_cbp[i][j+k*2],  CBP_VLC_SIZE,   &intra_vlcs[i].cbp[j][k],         rv34_cbp_code);
115
        }
116
        for(j = 0; j < 4; j++)
117
            rv34_gen_vlc(rv34_table_intra_firstpat[i][j], FIRSTBLK_VLC_SIZE, &intra_vlcs[i].first_pattern[j], NULL);
118
        rv34_gen_vlc(rv34_intra_coeff[i], COEFF_VLC_SIZE, &intra_vlcs[i].coefficient, NULL);
119
    }
120

    
121
    for(i = 0; i < NUM_INTER_TABLES; i++){
122
        rv34_gen_vlc(rv34_inter_cbppat[i], CBPPAT_VLC_SIZE, &inter_vlcs[i].cbppattern[0], NULL);
123
        for(j = 0; j < 4; j++)
124
            rv34_gen_vlc(rv34_inter_cbp[i][j], CBP_VLC_SIZE, &inter_vlcs[i].cbp[0][j], rv34_cbp_code);
125
        for(j = 0; j < 2; j++){
126
            rv34_gen_vlc(rv34_table_inter_firstpat [i][j], FIRSTBLK_VLC_SIZE, &inter_vlcs[i].first_pattern[j],  NULL);
127
            rv34_gen_vlc(rv34_table_inter_secondpat[i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].second_pattern[j], NULL);
128
            rv34_gen_vlc(rv34_table_inter_thirdpat [i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].third_pattern[j],  NULL);
129
        }
130
        rv34_gen_vlc(rv34_inter_coeff[i], COEFF_VLC_SIZE, &inter_vlcs[i].coefficient, NULL);
131
    }
132
}
133

    
134
/** @} */ // vlc group
135

    
136

    
137
/**
138
 * @defgroup transform RV30/40 inverse transform functions
139
 * @{
140
 */
141

    
142
static av_always_inline void rv34_row_transform(int temp[16], DCTELEM *block)
143
{
144
    int i;
145

    
146
    for(i=0; i<4; i++){
147
        const int z0= 13*(block[i+8*0] +    block[i+8*2]);
148
        const int z1= 13*(block[i+8*0] -    block[i+8*2]);
149
        const int z2=  7* block[i+8*1] - 17*block[i+8*3];
150
        const int z3= 17* block[i+8*1] +  7*block[i+8*3];
151

    
152
        temp[4*i+0]= z0+z3;
153
        temp[4*i+1]= z1+z2;
154
        temp[4*i+2]= z1-z2;
155
        temp[4*i+3]= z0-z3;
156
    }
157
}
158

    
159
/**
160
 * Real Video 3.0/4.0 inverse transform
161
 * Code is almost the same as in SVQ3, only scaling is different.
162
 */
163
static void rv34_inv_transform(DCTELEM *block){
164
    int temp[16];
165
    int i;
166

    
167
    rv34_row_transform(temp, block);
168

    
169
    for(i=0; i<4; i++){
170
        const int z0= 13*(temp[4*0+i] +    temp[4*2+i]) + 0x200;
171
        const int z1= 13*(temp[4*0+i] -    temp[4*2+i]) + 0x200;
172
        const int z2=  7* temp[4*1+i] - 17*temp[4*3+i];
173
        const int z3= 17* temp[4*1+i] +  7*temp[4*3+i];
174

    
175
        block[i*8+0]= (z0 + z3)>>10;
176
        block[i*8+1]= (z1 + z2)>>10;
177
        block[i*8+2]= (z1 - z2)>>10;
178
        block[i*8+3]= (z0 - z3)>>10;
179
    }
180

    
181
}
182

    
183
/**
184
 * RealVideo 3.0/4.0 inverse transform for DC block
185
 *
186
 * Code is almost the same as rv34_inv_transform()
187
 * but final coefficients are multiplied by 1.5 and have no rounding.
188
 */
189
static void rv34_inv_transform_noround(DCTELEM *block){
190
    int temp[16];
191
    int i;
192

    
193
    rv34_row_transform(temp, block);
194

    
195
    for(i=0; i<4; i++){
196
        const int z0= 13*(temp[4*0+i] +    temp[4*2+i]);
197
        const int z1= 13*(temp[4*0+i] -    temp[4*2+i]);
198
        const int z2=  7* temp[4*1+i] - 17*temp[4*3+i];
199
        const int z3= 17* temp[4*1+i] +  7*temp[4*3+i];
200

    
201
        block[i*8+0]= ((z0 + z3)*3)>>11;
202
        block[i*8+1]= ((z1 + z2)*3)>>11;
203
        block[i*8+2]= ((z1 - z2)*3)>>11;
204
        block[i*8+3]= ((z0 - z3)*3)>>11;
205
    }
206

    
207
}
208

    
209
/** @} */ // transform
210

    
211

    
212
/**
213
 * @defgroup block RV30/40 4x4 block decoding functions
214
 * @{
215
 */
216

    
217
/**
218
 * Decode coded block pattern.
219
 */
220
static int rv34_decode_cbp(GetBitContext *gb, RV34VLC *vlc, int table)
221
{
222
    int pattern, code, cbp=0;
223
    int ones;
224
    static const int cbp_masks[3] = {0x100000, 0x010000, 0x110000};
225
    static const int shifts[4] = { 0, 2, 8, 10 };
226
    const int *curshift = shifts;
227
    int i, t, mask;
228

    
229
    code = get_vlc2(gb, vlc->cbppattern[table].table, 9, 2);
230
    pattern = code & 0xF;
231
    code >>= 4;
232

    
233
    ones = rv34_count_ones[pattern];
234

    
235
    for(mask = 8; mask; mask >>= 1, curshift++){
236
        if(pattern & mask)
237
            cbp |= get_vlc2(gb, vlc->cbp[table][ones].table, vlc->cbp[table][ones].bits, 1) << curshift[0];
238
    }
239

    
240
    for(i = 0; i < 4; i++){
241
        t = modulo_three_table[code][i];
242
        if(t == 1)
243
            cbp |= cbp_masks[get_bits1(gb)] << i;
244
        if(t == 2)
245
            cbp |= cbp_masks[2] << i;
246
    }
247
    return cbp;
248
}
249

    
250
/**
251
 * Get one coefficient value from the bistream and store it.
252
 */
253
static inline void decode_coeff(DCTELEM *dst, int coef, int esc, GetBitContext *gb, VLC* vlc)
254
{
255
    if(coef){
256
        if(coef == esc){
257
            coef = get_vlc2(gb, vlc->table, 9, 2);
258
            if(coef > 23){
259
                coef -= 23;
260
                coef = 22 + ((1 << coef) | get_bits(gb, coef));
261
            }
262
            coef += esc;
263
        }
264
        if(get_bits1(gb))
265
            coef = -coef;
266
        *dst = coef;
267
    }
268
}
269

    
270
/**
271
 * Decode 2x2 subblock of coefficients.
272
 */
273
static inline void decode_subblock(DCTELEM *dst, int code, const int is_block2, GetBitContext *gb, VLC *vlc)
274
{
275
    int coeffs[4];
276

    
277
    coeffs[0] = modulo_three_table[code][0];
278
    coeffs[1] = modulo_three_table[code][1];
279
    coeffs[2] = modulo_three_table[code][2];
280
    coeffs[3] = modulo_three_table[code][3];
281
    decode_coeff(dst  , coeffs[0], 3, gb, vlc);
282
    if(is_block2){
283
        decode_coeff(dst+8, coeffs[1], 2, gb, vlc);
284
        decode_coeff(dst+1, coeffs[2], 2, gb, vlc);
285
    }else{
286
        decode_coeff(dst+1, coeffs[1], 2, gb, vlc);
287
        decode_coeff(dst+8, coeffs[2], 2, gb, vlc);
288
    }
289
    decode_coeff(dst+9, coeffs[3], 2, gb, vlc);
290
}
291

    
292
/**
293
 * Decode coefficients for 4x4 block.
294
 *
295
 * This is done by filling 2x2 subblocks with decoded coefficients
296
 * in this order (the same for subblocks and subblock coefficients):
297
 *  o--o
298
 *    /
299
 *   /
300
 *  o--o
301
 */
302

    
303
static inline void rv34_decode_block(DCTELEM *dst, GetBitContext *gb, RV34VLC *rvlc, int fc, int sc)
304
{
305
    int code, pattern;
306

    
307
    code = get_vlc2(gb, rvlc->first_pattern[fc].table, 9, 2);
308

    
309
    pattern = code & 0x7;
310

    
311
    code >>= 3;
312
    decode_subblock(dst, code, 0, gb, &rvlc->coefficient);
313

    
314
    if(pattern & 4){
315
        code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
316
        decode_subblock(dst + 2, code, 0, gb, &rvlc->coefficient);
317
    }
318
    if(pattern & 2){ // Looks like coefficients 1 and 2 are swapped for this block
319
        code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
320
        decode_subblock(dst + 8*2, code, 1, gb, &rvlc->coefficient);
321
    }
322
    if(pattern & 1){
323
        code = get_vlc2(gb, rvlc->third_pattern[sc].table, 9, 2);
324
        decode_subblock(dst + 8*2+2, code, 0, gb, &rvlc->coefficient);
325
    }
326

    
327
}
328

    
329
/**
330
 * Dequantize ordinary 4x4 block.
331
 * @todo optimize
332
 */
333
static inline void rv34_dequant4x4(DCTELEM *block, int Qdc, int Q)
334
{
335
    int i, j;
336

    
337
    block[0] = (block[0] * Qdc + 8) >> 4;
338
    for(i = 0; i < 4; i++)
339
        for(j = !i; j < 4; j++)
340
            block[j + i*8] = (block[j + i*8] * Q + 8) >> 4;
341
}
342

    
343
/**
344
 * Dequantize 4x4 block of DC values for 16x16 macroblock.
345
 * @todo optimize
346
 */
347
static inline void rv34_dequant4x4_16x16(DCTELEM *block, int Qdc, int Q)
348
{
349
    int i;
350

    
351
    for(i = 0; i < 3; i++)
352
         block[rv34_dezigzag[i]] = (block[rv34_dezigzag[i]] * Qdc + 8) >> 4;
353
    for(; i < 16; i++)
354
         block[rv34_dezigzag[i]] = (block[rv34_dezigzag[i]] * Q + 8) >> 4;
355
}
356
/** @} */ //block functions
357

    
358

    
359
/**
360
 * @defgroup bitstream RV30/40 bitstream parsing
361
 * @{
362
 */
363

    
364
/**
365
 * Decode starting slice position.
366
 * @todo Maybe replace with ff_h263_decode_mba() ?
367
 */
368
int ff_rv34_get_start_offset(GetBitContext *gb, int mb_size)
369
{
370
    int i;
371
    for(i = 0; i < 5; i++)
372
        if(rv34_mb_max_sizes[i] >= mb_size - 1)
373
            break;
374
    return rv34_mb_bits_sizes[i];
375
}
376

    
377
/**
378
 * Select VLC set for decoding from current quantizer, modifier and frame type.
379
 */
380
static inline RV34VLC* choose_vlc_set(int quant, int mod, int type)
381
{
382
    if(mod == 2 && quant < 19) quant += 10;
383
    else if(mod && quant < 26) quant += 5;
384
    return type ? &inter_vlcs[rv34_quant_to_vlc_set[1][av_clip(quant, 0, 30)]]
385
                : &intra_vlcs[rv34_quant_to_vlc_set[0][av_clip(quant, 0, 30)]];
386
}
387

    
388
/**
389
 * Decode quantizer difference and return modified quantizer.
390
 */
391
static inline int rv34_decode_dquant(GetBitContext *gb, int quant)
392
{
393
    if(get_bits1(gb))
394
        return rv34_dquant_tab[get_bits1(gb)][quant];
395
    else
396
        return get_bits(gb, 5);
397
}
398

    
399
/** @} */ //bitstream functions
400

    
401
/**
402
 * @defgroup mv motion vector related code (prediction, reconstruction, motion compensation)
403
 * @{
404
 */
405

    
406
/** macroblock partition width in 8x8 blocks */
407
static const uint8_t part_sizes_w[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 2, 1, 2, 2 };
408

    
409
/** macroblock partition height in 8x8 blocks */
410
static const uint8_t part_sizes_h[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 1, 2, 2, 2 };
411

    
412
/** availability index for subblocks */
413
static const uint8_t avail_indexes[4] = { 5, 6, 9, 10 };
414

    
415
/**
416
 * motion vector prediction
417
 *
418
 * Motion prediction performed for the block by using median prediction of
419
 * motion vectors from the left, top and right top blocks but in corner cases
420
 * some other vectors may be used instead.
421
 */
422
static void rv34_pred_mv(RV34DecContext *r, int block_type, int subblock_no, int dmv_no)
423
{
424
    MpegEncContext *s = &r->s;
425
    int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
426
    int A[2] = {0}, B[2], C[2];
427
    int i, j;
428
    int mx, my;
429
    int avail_index = avail_indexes[subblock_no];
430
    int c_off = part_sizes_w[block_type];
431

    
432
    mv_pos += (subblock_no & 1) + (subblock_no >> 1)*s->b8_stride;
433
    if(subblock_no == 3)
434
        c_off = -1;
435

    
436
    if(r->avail_cache[avail_index - 1]){
437
        A[0] = s->current_picture_ptr->motion_val[0][mv_pos-1][0];
438
        A[1] = s->current_picture_ptr->motion_val[0][mv_pos-1][1];
439
    }
440
    if(r->avail_cache[avail_index - 4]){
441
        B[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][0];
442
        B[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][1];
443
    }else{
444
        B[0] = A[0];
445
        B[1] = A[1];
446
    }
447
    if(!r->avail_cache[avail_index - 4 + c_off]){
448
        if(r->avail_cache[avail_index - 4] && (r->avail_cache[avail_index - 1] || r->rv30)){
449
            C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][0];
450
            C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][1];
451
        }else{
452
            C[0] = A[0];
453
            C[1] = A[1];
454
        }
455
    }else{
456
        C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][0];
457
        C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][1];
458
    }
459
    mx = mid_pred(A[0], B[0], C[0]);
460
    my = mid_pred(A[1], B[1], C[1]);
461
    mx += r->dmv[dmv_no][0];
462
    my += r->dmv[dmv_no][1];
463
    for(j = 0; j < part_sizes_h[block_type]; j++){
464
        for(i = 0; i < part_sizes_w[block_type]; i++){
465
            s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][0] = mx;
466
            s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][1] = my;
467
        }
468
    }
469
}
470

    
471
#define GET_PTS_DIFF(a, b) ((a - b + 8192) & 0x1FFF)
472

    
473
/**
474
 * Calculate motion vector component that should be added for direct blocks.
475
 */
476
static int calc_add_mv(RV34DecContext *r, int dir, int val)
477
{
478
    int refdist = GET_PTS_DIFF(r->next_pts, r->last_pts);
479
    int dist = dir ? -GET_PTS_DIFF(r->next_pts, r->cur_pts) : GET_PTS_DIFF(r->cur_pts, r->last_pts);
480
    int mul;
481

    
482
    if(!refdist) return 0;
483
    mul = (dist << 14) / refdist;
484
    return (val * mul + 0x2000) >> 14;
485
}
486

    
487
/**
488
 * Predict motion vector for B-frame macroblock.
489
 */
490
static inline void rv34_pred_b_vector(int A[2], int B[2], int C[2],
491
                                      int A_avail, int B_avail, int C_avail,
492
                                      int *mx, int *my)
493
{
494
    if(A_avail + B_avail + C_avail != 3){
495
        *mx = A[0] + B[0] + C[0];
496
        *my = A[1] + B[1] + C[1];
497
        if(A_avail + B_avail + C_avail == 2){
498
            *mx /= 2;
499
            *my /= 2;
500
        }
501
    }else{
502
        *mx = mid_pred(A[0], B[0], C[0]);
503
        *my = mid_pred(A[1], B[1], C[1]);
504
    }
505
}
506

    
507
/**
508
 * motion vector prediction for B-frames
509
 */
510
static void rv34_pred_mv_b(RV34DecContext *r, int block_type, int dir)
511
{
512
    MpegEncContext *s = &r->s;
513
    int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
514
    int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
515
    int A[2], B[2], C[2];
516
    int has_A = 0, has_B = 0, has_C = 0;
517
    int mx, my;
518
    int i, j;
519
    Picture *cur_pic = s->current_picture_ptr;
520
    const int mask = dir ? MB_TYPE_L1 : MB_TYPE_L0;
521
    int type = cur_pic->mb_type[mb_pos];
522

    
523
    memset(A, 0, sizeof(A));
524
    memset(B, 0, sizeof(B));
525
    memset(C, 0, sizeof(C));
526
    if((r->avail_cache[5-1] & type) & mask){
527
        A[0] = cur_pic->motion_val[dir][mv_pos - 1][0];
528
        A[1] = cur_pic->motion_val[dir][mv_pos - 1][1];
529
        has_A = 1;
530
    }
531
    if((r->avail_cache[5-4] & type) & mask){
532
        B[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][0];
533
        B[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][1];
534
        has_B = 1;
535
    }
536
    if((r->avail_cache[5-2] & type) & mask){
537
        C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][0];
538
        C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][1];
539
        has_C = 1;
540
    }else if((s->mb_x+1) == s->mb_width && (r->avail_cache[5-5] & type) & mask){
541
        C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][0];
542
        C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][1];
543
        has_C = 1;
544
    }
545

    
546
    rv34_pred_b_vector(A, B, C, has_A, has_B, has_C, &mx, &my);
547

    
548
    mx += r->dmv[dir][0];
549
    my += r->dmv[dir][1];
550

    
551
    for(j = 0; j < 2; j++){
552
        for(i = 0; i < 2; i++){
553
            cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][0] = mx;
554
            cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][1] = my;
555
        }
556
    }
557
    if(block_type == RV34_MB_B_BACKWARD || block_type == RV34_MB_B_FORWARD)
558
        fill_rectangle(cur_pic->motion_val[!dir][mv_pos], 2, 2, s->b8_stride, 0, 4);
559
}
560

    
561
/**
562
 * motion vector prediction - RV3 version
563
 */
564
static void rv34_pred_mv_rv3(RV34DecContext *r, int block_type, int dir)
565
{
566
    MpegEncContext *s = &r->s;
567
    int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
568
    int A[2] = {0}, B[2], C[2];
569
    int i, j, k;
570
    int mx, my;
571
    int avail_index = avail_indexes[0];
572

    
573
    if(r->avail_cache[avail_index - 1]){
574
        A[0] = s->current_picture_ptr->motion_val[0][mv_pos-1][0];
575
        A[1] = s->current_picture_ptr->motion_val[0][mv_pos-1][1];
576
    }
577
    if(r->avail_cache[avail_index - 4]){
578
        B[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][0];
579
        B[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][1];
580
    }else{
581
        B[0] = A[0];
582
        B[1] = A[1];
583
    }
584
    if(!r->avail_cache[avail_index - 4 + 2]){
585
        if(r->avail_cache[avail_index - 4] && (r->avail_cache[avail_index - 1])){
586
            C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][0];
587
            C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][1];
588
        }else{
589
            C[0] = A[0];
590
            C[1] = A[1];
591
        }
592
    }else{
593
        C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+2][0];
594
        C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+2][1];
595
    }
596
    mx = mid_pred(A[0], B[0], C[0]);
597
    my = mid_pred(A[1], B[1], C[1]);
598
    mx += r->dmv[0][0];
599
    my += r->dmv[0][1];
600
    for(j = 0; j < 2; j++){
601
        for(i = 0; i < 2; i++){
602
            for(k = 0; k < 2; k++){
603
                s->current_picture_ptr->motion_val[k][mv_pos + i + j*s->b8_stride][0] = mx;
604
                s->current_picture_ptr->motion_val[k][mv_pos + i + j*s->b8_stride][1] = my;
605
            }
606
        }
607
    }
608
}
609

    
610
static const int chroma_coeffs[3] = { 0, 3, 5 };
611

    
612
/**
613
 * generic motion compensation function
614
 *
615
 * @param r decoder context
616
 * @param block_type type of the current block
617
 * @param xoff horizontal offset from the start of the current block
618
 * @param yoff vertical offset from the start of the current block
619
 * @param mv_off offset to the motion vector information
620
 * @param width width of the current partition in 8x8 blocks
621
 * @param height height of the current partition in 8x8 blocks
622
 * @param dir motion compensation direction (i.e. from the last or the next reference frame)
623
 * @param thirdpel motion vectors are specified in 1/3 of pixel
624
 * @param qpel_mc a set of functions used to perform luma motion compensation
625
 * @param chroma_mc a set of functions used to perform chroma motion compensation
626
 */
627
static inline void rv34_mc(RV34DecContext *r, const int block_type,
628
                          const int xoff, const int yoff, int mv_off,
629
                          const int width, const int height, int dir,
630
                          const int thirdpel,
631
                          qpel_mc_func (*qpel_mc)[16],
632
                          h264_chroma_mc_func (*chroma_mc))
633
{
634
    MpegEncContext *s = &r->s;
635
    uint8_t *Y, *U, *V, *srcY, *srcU, *srcV;
636
    int dxy, mx, my, umx, umy, lx, ly, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
637
    int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride + mv_off;
638
    int is16x16 = 1;
639

    
640
    if(thirdpel){
641
        int chroma_mx, chroma_my;
642
        mx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) / 3 - (1 << 24);
643
        my = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) / 3 - (1 << 24);
644
        lx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) % 3;
645
        ly = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) % 3;
646
        chroma_mx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + 1) >> 1;
647
        chroma_my = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + 1) >> 1;
648
        umx = (chroma_mx + (3 << 24)) / 3 - (1 << 24);
649
        umy = (chroma_my + (3 << 24)) / 3 - (1 << 24);
650
        uvmx = chroma_coeffs[(chroma_mx + (3 << 24)) % 3];
651
        uvmy = chroma_coeffs[(chroma_my + (3 << 24)) % 3];
652
    }else{
653
        int cx, cy;
654
        mx = s->current_picture_ptr->motion_val[dir][mv_pos][0] >> 2;
655
        my = s->current_picture_ptr->motion_val[dir][mv_pos][1] >> 2;
656
        lx = s->current_picture_ptr->motion_val[dir][mv_pos][0] & 3;
657
        ly = s->current_picture_ptr->motion_val[dir][mv_pos][1] & 3;
658
        cx = s->current_picture_ptr->motion_val[dir][mv_pos][0] / 2;
659
        cy = s->current_picture_ptr->motion_val[dir][mv_pos][1] / 2;
660
        umx = cx >> 2;
661
        umy = cy >> 2;
662
        uvmx = (cx & 3) << 1;
663
        uvmy = (cy & 3) << 1;
664
        //due to some flaw RV40 uses the same MC compensation routine for H2V2 and H3V3
665
        if(uvmx == 6 && uvmy == 6)
666
            uvmx = uvmy = 4;
667
    }
668
    dxy = ly*4 + lx;
669
    srcY = dir ? s->next_picture_ptr->data[0] : s->last_picture_ptr->data[0];
670
    srcU = dir ? s->next_picture_ptr->data[1] : s->last_picture_ptr->data[1];
671
    srcV = dir ? s->next_picture_ptr->data[2] : s->last_picture_ptr->data[2];
672
    src_x = s->mb_x * 16 + xoff + mx;
673
    src_y = s->mb_y * 16 + yoff + my;
674
    uvsrc_x = s->mb_x * 8 + (xoff >> 1) + umx;
675
    uvsrc_y = s->mb_y * 8 + (yoff >> 1) + umy;
676
    srcY += src_y * s->linesize + src_x;
677
    srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
678
    srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
679
    if(   (unsigned)(src_x - !!lx*2) > s->h_edge_pos - !!lx*2 - (width <<3) - 4
680
       || (unsigned)(src_y - !!ly*2) > s->v_edge_pos - !!ly*2 - (height<<3) - 4){
681
        uint8_t *uvbuf= s->edge_emu_buffer + 22 * s->linesize;
682

    
683
        srcY -= 2 + 2*s->linesize;
684
        ff_emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, (width<<3)+6, (height<<3)+6,
685
                            src_x - 2, src_y - 2, s->h_edge_pos, s->v_edge_pos);
686
        srcY = s->edge_emu_buffer + 2 + 2*s->linesize;
687
        ff_emulated_edge_mc(uvbuf     , srcU, s->uvlinesize, (width<<2)+1, (height<<2)+1,
688
                            uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
689
        ff_emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, (width<<2)+1, (height<<2)+1,
690
                            uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
691
        srcU = uvbuf;
692
        srcV = uvbuf + 16;
693
    }
694
    Y = s->dest[0] + xoff      + yoff     *s->linesize;
695
    U = s->dest[1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
696
    V = s->dest[2] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
697

    
698
    if(block_type == RV34_MB_P_16x8){
699
        qpel_mc[1][dxy](Y, srcY, s->linesize);
700
        Y    += 8;
701
        srcY += 8;
702
    }else if(block_type == RV34_MB_P_8x16){
703
        qpel_mc[1][dxy](Y, srcY, s->linesize);
704
        Y    += 8 * s->linesize;
705
        srcY += 8 * s->linesize;
706
    }
707
    is16x16 = (block_type != RV34_MB_P_8x8) && (block_type != RV34_MB_P_16x8) && (block_type != RV34_MB_P_8x16);
708
    qpel_mc[!is16x16][dxy](Y, srcY, s->linesize);
709
    chroma_mc[2-width]   (U, srcU, s->uvlinesize, height*4, uvmx, uvmy);
710
    chroma_mc[2-width]   (V, srcV, s->uvlinesize, height*4, uvmx, uvmy);
711
}
712

    
713
static void rv34_mc_1mv(RV34DecContext *r, const int block_type,
714
                        const int xoff, const int yoff, int mv_off,
715
                        const int width, const int height, int dir)
716
{
717
    rv34_mc(r, block_type, xoff, yoff, mv_off, width, height, dir, r->rv30,
718
            r->rv30 ? r->s.dsp.put_rv30_tpel_pixels_tab
719
                    : r->s.dsp.put_rv40_qpel_pixels_tab,
720
            r->rv30 ? r->s.dsp.put_h264_chroma_pixels_tab
721
                    : r->s.dsp.put_rv40_chroma_pixels_tab);
722
}
723

    
724
static void rv34_mc_2mv(RV34DecContext *r, const int block_type)
725
{
726
    rv34_mc(r, block_type, 0, 0, 0, 2, 2, 0, r->rv30,
727
            r->rv30 ? r->s.dsp.put_rv30_tpel_pixels_tab
728
                    : r->s.dsp.put_rv40_qpel_pixels_tab,
729
            r->rv30 ? r->s.dsp.put_h264_chroma_pixels_tab
730
                    : r->s.dsp.put_rv40_chroma_pixels_tab);
731
    rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30,
732
            r->rv30 ? r->s.dsp.avg_rv30_tpel_pixels_tab
733
                    : r->s.dsp.avg_rv40_qpel_pixels_tab,
734
            r->rv30 ? r->s.dsp.avg_h264_chroma_pixels_tab
735
                    : r->s.dsp.avg_rv40_chroma_pixels_tab);
736
}
737

    
738
static void rv34_mc_2mv_skip(RV34DecContext *r)
739
{
740
    int i, j;
741
    for(j = 0; j < 2; j++)
742
        for(i = 0; i < 2; i++){
743
             rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 0, r->rv30,
744
                    r->rv30 ? r->s.dsp.put_rv30_tpel_pixels_tab
745
                            : r->s.dsp.put_rv40_qpel_pixels_tab,
746
                    r->rv30 ? r->s.dsp.put_h264_chroma_pixels_tab
747
                            : r->s.dsp.put_rv40_chroma_pixels_tab);
748
             rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 1, r->rv30,
749
                    r->rv30 ? r->s.dsp.avg_rv30_tpel_pixels_tab
750
                            : r->s.dsp.avg_rv40_qpel_pixels_tab,
751
                    r->rv30 ? r->s.dsp.avg_h264_chroma_pixels_tab
752
                            : r->s.dsp.avg_rv40_chroma_pixels_tab);
753
        }
754
}
755

    
756
/** number of motion vectors in each macroblock type */
757
static const int num_mvs[RV34_MB_TYPES] = { 0, 0, 1, 4, 1, 1, 0, 0, 2, 2, 2, 1 };
758

    
759
/**
760
 * Decode motion vector differences
761
 * and perform motion vector reconstruction and motion compensation.
762
 */
763
static int rv34_decode_mv(RV34DecContext *r, int block_type)
764
{
765
    MpegEncContext *s = &r->s;
766
    GetBitContext *gb = &s->gb;
767
    int i, j, k, l;
768
    int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
769
    int next_bt;
770

    
771
    memset(r->dmv, 0, sizeof(r->dmv));
772
    for(i = 0; i < num_mvs[block_type]; i++){
773
        r->dmv[i][0] = svq3_get_se_golomb(gb);
774
        r->dmv[i][1] = svq3_get_se_golomb(gb);
775
    }
776
    switch(block_type){
777
    case RV34_MB_TYPE_INTRA:
778
    case RV34_MB_TYPE_INTRA16x16:
779
        fill_rectangle(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], 2, 2, s->b8_stride, 0, 4);
780
        return 0;
781
    case RV34_MB_SKIP:
782
        if(s->pict_type == FF_P_TYPE){
783
            fill_rectangle(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], 2, 2, s->b8_stride, 0, 4);
784
            rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
785
            break;
786
        }
787
    case RV34_MB_B_DIRECT:
788
        //surprisingly, it uses motion scheme from next reference frame
789
        next_bt = s->next_picture_ptr->mb_type[s->mb_x + s->mb_y * s->mb_stride];
790
        if(IS_INTRA(next_bt))
791
            fill_rectangle(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], 2, 2, s->b8_stride, 0, 4);
792
        else
793
        for(j = 0; j < 2; j++)
794
            for(i = 0; i < 2; i++)
795
                for(k = 0; k < 2; k++)
796
                    for(l = 0; l < 2; l++)
797
                        s->current_picture_ptr->motion_val[l][mv_pos + i + j*s->b8_stride][k] = calc_add_mv(r, l, s->next_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][k]);
798
        if(IS_16X16(next_bt)) //we can use whole macroblock MC
799
            rv34_mc_2mv(r, block_type);
800
        else
801
            rv34_mc_2mv_skip(r);
802
        fill_rectangle(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], 2, 2, s->b8_stride, 0, 4);
803
        break;
804
    case RV34_MB_P_16x16:
805
    case RV34_MB_P_MIX16x16:
806
        rv34_pred_mv(r, block_type, 0, 0);
807
        rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
808
        break;
809
    case RV34_MB_B_FORWARD:
810
    case RV34_MB_B_BACKWARD:
811
        r->dmv[1][0] = r->dmv[0][0];
812
        r->dmv[1][1] = r->dmv[0][1];
813
        if(r->rv30)
814
            rv34_pred_mv_rv3(r, block_type, block_type == RV34_MB_B_BACKWARD);
815
        else
816
            rv34_pred_mv_b  (r, block_type, block_type == RV34_MB_B_BACKWARD);
817
        rv34_mc_1mv     (r, block_type, 0, 0, 0, 2, 2, block_type == RV34_MB_B_BACKWARD);
818
        break;
819
    case RV34_MB_P_16x8:
820
    case RV34_MB_P_8x16:
821
        rv34_pred_mv(r, block_type, 0, 0);
822
        rv34_pred_mv(r, block_type, 1 + (block_type == RV34_MB_P_16x8), 1);
823
        if(block_type == RV34_MB_P_16x8){
824
            rv34_mc_1mv(r, block_type, 0, 0, 0,            2, 1, 0);
825
            rv34_mc_1mv(r, block_type, 0, 8, s->b8_stride, 2, 1, 0);
826
        }
827
        if(block_type == RV34_MB_P_8x16){
828
            rv34_mc_1mv(r, block_type, 0, 0, 0, 1, 2, 0);
829
            rv34_mc_1mv(r, block_type, 8, 0, 1, 1, 2, 0);
830
        }
831
        break;
832
    case RV34_MB_B_BIDIR:
833
        rv34_pred_mv_b  (r, block_type, 0);
834
        rv34_pred_mv_b  (r, block_type, 1);
835
        rv34_mc_2mv     (r, block_type);
836
        break;
837
    case RV34_MB_P_8x8:
838
        for(i=0;i< 4;i++){
839
            rv34_pred_mv(r, block_type, i, i);
840
            rv34_mc_1mv (r, block_type, (i&1)<<3, (i&2)<<2, (i&1)+(i>>1)*s->b8_stride, 1, 1, 0);
841
        }
842
        break;
843
    }
844

    
845
    return 0;
846
}
847
/** @} */ // mv group
848

    
849
/**
850
 * @defgroup recons Macroblock reconstruction functions
851
 * @{
852
 */
853
/** mapping of RV30/40 intra prediction types to standard H.264 types */
854
static const int ittrans[9] = {
855
 DC_PRED, VERT_PRED, HOR_PRED, DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_LEFT_PRED,
856
 VERT_RIGHT_PRED, VERT_LEFT_PRED, HOR_UP_PRED, HOR_DOWN_PRED,
857
};
858

    
859
/** mapping of RV30/40 intra 16x16 prediction types to standard H.264 types */
860
static const int ittrans16[4] = {
861
 DC_PRED8x8, VERT_PRED8x8, HOR_PRED8x8, PLANE_PRED8x8,
862
};
863

    
864
/**
865
 * Perform 4x4 intra prediction.
866
 */
867
static void rv34_pred_4x4_block(RV34DecContext *r, uint8_t *dst, int stride, int itype, int up, int left, int down, int right)
868
{
869
    uint8_t *prev = dst - stride + 4;
870
    uint32_t topleft;
871

    
872
    if(!up && !left)
873
        itype = DC_128_PRED;
874
    else if(!up){
875
        if(itype == VERT_PRED) itype = HOR_PRED;
876
        if(itype == DC_PRED)   itype = LEFT_DC_PRED;
877
    }else if(!left){
878
        if(itype == HOR_PRED)  itype = VERT_PRED;
879
        if(itype == DC_PRED)   itype = TOP_DC_PRED;
880
        if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
881
    }
882
    if(!down){
883
        if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
884
        if(itype == HOR_UP_PRED) itype = HOR_UP_PRED_RV40_NODOWN;
885
        if(itype == VERT_LEFT_PRED) itype = VERT_LEFT_PRED_RV40_NODOWN;
886
    }
887
    if(!right && up){
888
        topleft = dst[-stride + 3] * 0x01010101;
889
        prev = (uint8_t*)&topleft;
890
    }
891
    r->h.pred4x4[itype](dst, prev, stride);
892
}
893

    
894
/** add_pixels_clamped for 4x4 block */
895
static void rv34_add_4x4_block(uint8_t *dst, int stride, DCTELEM block[64], int off)
896
{
897
    int x, y;
898
    for(y = 0; y < 4; y++)
899
        for(x = 0; x < 4; x++)
900
            dst[x + y*stride] = av_clip_uint8(dst[x + y*stride] + block[off + x+y*8]);
901
}
902

    
903
static inline int adjust_pred16(int itype, int up, int left)
904
{
905
    if(!up && !left)
906
        itype = DC_128_PRED8x8;
907
    else if(!up){
908
        if(itype == PLANE_PRED8x8)itype = HOR_PRED8x8;
909
        if(itype == VERT_PRED8x8) itype = HOR_PRED8x8;
910
        if(itype == DC_PRED8x8)   itype = LEFT_DC_PRED8x8;
911
    }else if(!left){
912
        if(itype == PLANE_PRED8x8)itype = VERT_PRED8x8;
913
        if(itype == HOR_PRED8x8)  itype = VERT_PRED8x8;
914
        if(itype == DC_PRED8x8)   itype = TOP_DC_PRED8x8;
915
    }
916
    return itype;
917
}
918

    
919
static void rv34_output_macroblock(RV34DecContext *r, int8_t *intra_types, int cbp, int is16)
920
{
921
    MpegEncContext *s = &r->s;
922
    DSPContext *dsp = &s->dsp;
923
    int i, j;
924
    uint8_t *Y, *U, *V;
925
    int itype;
926
    int avail[6*8] = {0};
927
    int idx;
928

    
929
    // Set neighbour information.
930
    if(r->avail_cache[0])
931
        avail[0] = 1;
932
    if(r->avail_cache[1])
933
        avail[1] = avail[2] = 1;
934
    if(r->avail_cache[2])
935
        avail[3] = avail[4] = 1;
936
    if(r->avail_cache[3])
937
        avail[5] = 1;
938
    if(r->avail_cache[4])
939
        avail[8] = avail[16] = 1;
940
    if(r->avail_cache[8])
941
        avail[24] = avail[32] = 1;
942

    
943
    Y = s->dest[0];
944
    U = s->dest[1];
945
    V = s->dest[2];
946
    if(!is16){
947
        for(j = 0; j < 4; j++){
948
            idx = 9 + j*8;
949
            for(i = 0; i < 4; i++, cbp >>= 1, Y += 4, idx++){
950
                rv34_pred_4x4_block(r, Y, s->linesize, ittrans[intra_types[i]], avail[idx-8], avail[idx-1], avail[idx+7], avail[idx-7]);
951
                avail[idx] = 1;
952
                if(cbp & 1)
953
                    rv34_add_4x4_block(Y, s->linesize, s->block[(i>>1)+(j&2)], (i&1)*4+(j&1)*32);
954
            }
955
            Y += s->linesize * 4 - 4*4;
956
            intra_types += s->b4_stride;
957
        }
958
        intra_types -= s->b4_stride * 4;
959
        fill_rectangle(r->avail_cache + 5, 2, 2, 4, 0, 4);
960
        for(j = 0; j < 2; j++){
961
            idx = 5 + j*4;
962
            for(i = 0; i < 2; i++, cbp >>= 1, idx++){
963
                rv34_pred_4x4_block(r, U + i*4 + j*4*s->uvlinesize, s->uvlinesize, ittrans[intra_types[i*2+j*2*s->b4_stride]], r->avail_cache[idx-4], r->avail_cache[idx-1], !i && !j, r->avail_cache[idx-3]);
964
                rv34_pred_4x4_block(r, V + i*4 + j*4*s->uvlinesize, s->uvlinesize, ittrans[intra_types[i*2+j*2*s->b4_stride]], r->avail_cache[idx-4], r->avail_cache[idx-1], !i && !j, r->avail_cache[idx-3]);
965
                r->avail_cache[idx] = 1;
966
                if(cbp & 0x01)
967
                    rv34_add_4x4_block(U + i*4 + j*4*s->uvlinesize, s->uvlinesize, s->block[4], i*4+j*32);
968
                if(cbp & 0x10)
969
                    rv34_add_4x4_block(V + i*4 + j*4*s->uvlinesize, s->uvlinesize, s->block[5], i*4+j*32);
970
            }
971
        }
972
    }else{
973
        itype = ittrans16[intra_types[0]];
974
        itype = adjust_pred16(itype, r->avail_cache[5-4], r->avail_cache[5-1]);
975
        r->h.pred16x16[itype](Y, s->linesize);
976
        dsp->add_pixels_clamped(s->block[0], Y,     s->linesize);
977
        dsp->add_pixels_clamped(s->block[1], Y + 8, s->linesize);
978
        Y += s->linesize * 8;
979
        dsp->add_pixels_clamped(s->block[2], Y,     s->linesize);
980
        dsp->add_pixels_clamped(s->block[3], Y + 8, s->linesize);
981

    
982
        itype = ittrans16[intra_types[0]];
983
        if(itype == PLANE_PRED8x8) itype = DC_PRED8x8;
984
        itype = adjust_pred16(itype, r->avail_cache[5-4], r->avail_cache[5-1]);
985
        r->h.pred8x8[itype](U, s->uvlinesize);
986
        dsp->add_pixels_clamped(s->block[4], U, s->uvlinesize);
987
        r->h.pred8x8[itype](V, s->uvlinesize);
988
        dsp->add_pixels_clamped(s->block[5], V, s->uvlinesize);
989
    }
990
}
991

    
992
/** @} */ // recons group
993

    
994
/**
995
 * @addtogroup bitstream
996
 * Decode macroblock header and return CBP in case of success, -1 otherwise.
997
 */
998
static int rv34_decode_mb_header(RV34DecContext *r, int8_t *intra_types)
999
{
1000
    MpegEncContext *s = &r->s;
1001
    GetBitContext *gb = &s->gb;
1002
    int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1003
    int i, t;
1004

    
1005
    if(!r->si.type){
1006
        r->is16 = get_bits1(gb);
1007
        if(!r->is16 && !r->rv30){
1008
            if(!get_bits1(gb))
1009
                av_log(s->avctx, AV_LOG_ERROR, "Need DQUANT\n");
1010
        }
1011
        s->current_picture_ptr->mb_type[mb_pos] = r->is16 ? MB_TYPE_INTRA16x16 : MB_TYPE_INTRA;
1012
        r->block_type = r->is16 ? RV34_MB_TYPE_INTRA16x16 : RV34_MB_TYPE_INTRA;
1013
    }else{
1014
        r->block_type = r->decode_mb_info(r);
1015
        if(r->block_type == -1)
1016
            return -1;
1017
        s->current_picture_ptr->mb_type[mb_pos] = rv34_mb_type_to_lavc[r->block_type];
1018
        r->mb_type[mb_pos] = r->block_type;
1019
        if(r->block_type == RV34_MB_SKIP){
1020
            if(s->pict_type == FF_P_TYPE)
1021
                r->mb_type[mb_pos] = RV34_MB_P_16x16;
1022
            if(s->pict_type == FF_B_TYPE)
1023
                r->mb_type[mb_pos] = RV34_MB_B_DIRECT;
1024
        }
1025
        r->is16 = !!IS_INTRA16x16(s->current_picture_ptr->mb_type[mb_pos]);
1026
        rv34_decode_mv(r, r->block_type);
1027
        if(r->block_type == RV34_MB_SKIP){
1028
            fill_rectangle(intra_types, 4, 4, s->b4_stride, 0, sizeof(intra_types[0]));
1029
            return 0;
1030
        }
1031
        r->chroma_vlc = 1;
1032
        r->luma_vlc   = 0;
1033
    }
1034
    if(IS_INTRA(s->current_picture_ptr->mb_type[mb_pos])){
1035
        if(r->is16){
1036
            t = get_bits(gb, 2);
1037
            fill_rectangle(intra_types, 4, 4, s->b4_stride, t, sizeof(intra_types[0]));
1038
            r->luma_vlc   = 2;
1039
        }else{
1040
            if(r->decode_intra_types(r, gb, intra_types) < 0)
1041
                return -1;
1042
            r->luma_vlc   = 1;
1043
        }
1044
        r->chroma_vlc = 0;
1045
        r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
1046
    }else{
1047
        for(i = 0; i < 16; i++)
1048
            intra_types[(i & 3) + (i>>2) * s->b4_stride] = 0;
1049
        r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
1050
        if(r->mb_type[mb_pos] == RV34_MB_P_MIX16x16){
1051
            r->is16 = 1;
1052
            r->chroma_vlc = 1;
1053
            r->luma_vlc   = 2;
1054
            r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
1055
        }
1056
    }
1057

    
1058
    return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
1059
}
1060

    
1061
/**
1062
 * @addtogroup recons
1063
 * @{
1064
 */
1065
/**
1066
 * mask for retrieving all bits in coded block pattern
1067
 * corresponding to one 8x8 block
1068
 */
1069
#define LUMA_CBP_BLOCK_MASK 0x33
1070

    
1071
#define U_CBP_MASK 0x0F0000
1072
#define V_CBP_MASK 0xF00000
1073

    
1074

    
1075
static void rv34_apply_differences(RV34DecContext *r, int cbp)
1076
{
1077
    static const int shifts[4] = { 0, 2, 8, 10 };
1078
    MpegEncContext *s = &r->s;
1079
    int i;
1080

    
1081
    for(i = 0; i < 4; i++)
1082
        if((cbp & (LUMA_CBP_BLOCK_MASK << shifts[i])) || r->block_type == RV34_MB_P_MIX16x16)
1083
            s->dsp.add_pixels_clamped(s->block[i], s->dest[0] + (i & 1)*8 + (i&2)*4*s->linesize, s->linesize);
1084
    if(cbp & U_CBP_MASK)
1085
        s->dsp.add_pixels_clamped(s->block[4], s->dest[1], s->uvlinesize);
1086
    if(cbp & V_CBP_MASK)
1087
        s->dsp.add_pixels_clamped(s->block[5], s->dest[2], s->uvlinesize);
1088
}
1089

    
1090
static int is_mv_diff_gt_3(int16_t (*motion_val)[2], int step)
1091
{
1092
    int d;
1093
    d = motion_val[0][0] - motion_val[-step][0];
1094
    if(d < -3 || d > 3)
1095
        return 1;
1096
    d = motion_val[0][1] - motion_val[-step][1];
1097
    if(d < -3 || d > 3)
1098
        return 1;
1099
    return 0;
1100
}
1101

    
1102
static int rv34_set_deblock_coef(RV34DecContext *r)
1103
{
1104
    MpegEncContext *s = &r->s;
1105
    int hmvmask = 0, vmvmask = 0, i, j;
1106
    int midx = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
1107
    int16_t (*motion_val)[2] = s->current_picture_ptr->motion_val[0][midx];
1108
    for(j = 0; j < 16; j += 8){
1109
        for(i = 0; i < 2; i++){
1110
            if(is_mv_diff_gt_3(motion_val + i, 1))
1111
                vmvmask |= 0x11 << (j + i*2);
1112
            if((j || s->mb_y) && is_mv_diff_gt_3(motion_val + i, s->b8_stride))
1113
                hmvmask |= 0x03 << (j + i*2);
1114
        }
1115
        motion_val += s->b8_stride;
1116
    }
1117
    if(s->first_slice_line)
1118
        hmvmask &= ~0x000F;
1119
    if(!s->mb_x)
1120
        vmvmask &= ~0x1111;
1121
    if(r->rv30){ //RV30 marks both subblocks on the edge for filtering
1122
        vmvmask |= (vmvmask & 0x4444) >> 1;
1123
        hmvmask |= (hmvmask & 0x0F00) >> 4;
1124
        if(s->mb_x)
1125
            r->deblock_coefs[s->mb_x - 1 + s->mb_y*s->mb_stride] |= (vmvmask & 0x1111) << 3;
1126
        if(!s->first_slice_line)
1127
            r->deblock_coefs[s->mb_x + (s->mb_y - 1)*s->mb_stride] |= (hmvmask & 0xF) << 12;
1128
    }
1129
    return hmvmask | vmvmask;
1130
}
1131

    
1132
static int rv34_decode_macroblock(RV34DecContext *r, int8_t *intra_types)
1133
{
1134
    MpegEncContext *s = &r->s;
1135
    GetBitContext *gb = &s->gb;
1136
    int cbp, cbp2;
1137
    int i, blknum, blkoff;
1138
    DCTELEM block16[64];
1139
    int luma_dc_quant;
1140
    int dist;
1141
    int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1142

    
1143
    // Calculate which neighbours are available. Maybe it's worth optimizing too.
1144
    memset(r->avail_cache, 0, sizeof(r->avail_cache));
1145
    fill_rectangle(r->avail_cache + 5, 2, 2, 4, 1, 4);
1146
    dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
1147
    if(s->mb_x && dist)
1148
        r->avail_cache[4] =
1149
        r->avail_cache[8] = s->current_picture_ptr->mb_type[mb_pos - 1];
1150
    if(dist >= s->mb_width)
1151
        r->avail_cache[1] =
1152
        r->avail_cache[2] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride];
1153
    if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
1154
        r->avail_cache[3] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride + 1];
1155
    if(s->mb_x && dist > s->mb_width)
1156
        r->avail_cache[0] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride - 1];
1157

    
1158
    s->qscale = r->si.quant;
1159
    cbp = cbp2 = rv34_decode_mb_header(r, intra_types);
1160
    r->cbp_luma  [mb_pos] = cbp;
1161
    r->cbp_chroma[mb_pos] = cbp >> 16;
1162
    if(s->pict_type == FF_I_TYPE)
1163
        r->deblock_coefs[mb_pos] = 0xFFFF;
1164
    else
1165
        r->deblock_coefs[mb_pos] = rv34_set_deblock_coef(r) | r->cbp_luma[mb_pos];
1166
    s->current_picture_ptr->qscale_table[mb_pos] = s->qscale;
1167

    
1168
    if(cbp == -1)
1169
        return -1;
1170

    
1171
    luma_dc_quant = r->block_type == RV34_MB_P_MIX16x16 ? r->luma_dc_quant_p[s->qscale] : r->luma_dc_quant_i[s->qscale];
1172
    if(r->is16){
1173
        memset(block16, 0, sizeof(block16));
1174
        rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0);
1175
        rv34_dequant4x4_16x16(block16, rv34_qscale_tab[luma_dc_quant],rv34_qscale_tab[s->qscale]);
1176
        rv34_inv_transform_noround(block16);
1177
    }
1178

    
1179
    for(i = 0; i < 16; i++, cbp >>= 1){
1180
        if(!r->is16 && !(cbp & 1)) continue;
1181
        blknum = ((i & 2) >> 1) + ((i & 8) >> 2);
1182
        blkoff = ((i & 1) << 2) + ((i & 4) << 3);
1183
        if(cbp & 1)
1184
            rv34_decode_block(s->block[blknum] + blkoff, gb, r->cur_vlcs, r->luma_vlc, 0);
1185
        rv34_dequant4x4(s->block[blknum] + blkoff, rv34_qscale_tab[s->qscale],rv34_qscale_tab[s->qscale]);
1186
        if(r->is16) //FIXME: optimize
1187
            s->block[blknum][blkoff] = block16[(i & 3) | ((i & 0xC) << 1)];
1188
        rv34_inv_transform(s->block[blknum] + blkoff);
1189
    }
1190
    if(r->block_type == RV34_MB_P_MIX16x16)
1191
        r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
1192
    for(; i < 24; i++, cbp >>= 1){
1193
        if(!(cbp & 1)) continue;
1194
        blknum = ((i & 4) >> 2) + 4;
1195
        blkoff = ((i & 1) << 2) + ((i & 2) << 4);
1196
        rv34_decode_block(s->block[blknum] + blkoff, gb, r->cur_vlcs, r->chroma_vlc, 1);
1197
        rv34_dequant4x4(s->block[blknum] + blkoff, rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]],rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]]);
1198
        rv34_inv_transform(s->block[blknum] + blkoff);
1199
    }
1200
    if(IS_INTRA(s->current_picture_ptr->mb_type[mb_pos]))
1201
        rv34_output_macroblock(r, intra_types, cbp2, r->is16);
1202
    else
1203
        rv34_apply_differences(r, cbp2);
1204

    
1205
    return 0;
1206
}
1207

    
1208
static int check_slice_end(RV34DecContext *r, MpegEncContext *s)
1209
{
1210
    int bits;
1211
    if(s->mb_y >= s->mb_height)
1212
        return 1;
1213
    if(!s->mb_num_left)
1214
        return 1;
1215
    if(r->s.mb_skip_run > 1)
1216
        return 0;
1217
    bits = r->bits - get_bits_count(&s->gb);
1218
    if(bits < 0 || (bits < 8 && !show_bits(&s->gb, bits)))
1219
        return 1;
1220
    return 0;
1221
}
1222

    
1223
static inline int slice_compare(SliceInfo *si1, SliceInfo *si2)
1224
{
1225
    return si1->type   != si2->type  ||
1226
           si1->start  >= si2->start ||
1227
           si1->width  != si2->width ||
1228
           si1->height != si2->height||
1229
           si1->pts    != si2->pts;
1230
}
1231

    
1232
static int rv34_decode_slice(RV34DecContext *r, int end, const uint8_t* buf, int buf_size)
1233
{
1234
    MpegEncContext *s = &r->s;
1235
    GetBitContext *gb = &s->gb;
1236
    int mb_pos;
1237
    int res;
1238

    
1239
    init_get_bits(&r->s.gb, buf, buf_size*8);
1240
    res = r->parse_slice_header(r, gb, &r->si);
1241
    if(res < 0){
1242
        av_log(s->avctx, AV_LOG_ERROR, "Incorrect or unknown slice header\n");
1243
        return -1;
1244
    }
1245

    
1246
    if ((s->mb_x == 0 && s->mb_y == 0) || s->current_picture_ptr==NULL) {
1247
        if(s->width != r->si.width || s->height != r->si.height){
1248
            av_log(s->avctx, AV_LOG_DEBUG, "Changing dimensions to %dx%d\n", r->si.width,r->si.height);
1249
            MPV_common_end(s);
1250
            s->width  = r->si.width;
1251
            s->height = r->si.height;
1252
            if(MPV_common_init(s) < 0)
1253
                return -1;
1254
            r->intra_types_hist = av_realloc(r->intra_types_hist, s->b4_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1255
            r->intra_types = r->intra_types_hist + s->b4_stride * 4;
1256
            r->mb_type = av_realloc(r->mb_type, r->s.mb_stride * r->s.mb_height * sizeof(*r->mb_type));
1257
            r->cbp_luma   = av_realloc(r->cbp_luma,   r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_luma));
1258
            r->cbp_chroma = av_realloc(r->cbp_chroma, r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_chroma));
1259
            r->deblock_coefs = av_realloc(r->deblock_coefs, r->s.mb_stride * r->s.mb_height * sizeof(*r->deblock_coefs));
1260
        }
1261
        s->pict_type = r->si.type ? r->si.type : FF_I_TYPE;
1262
        if(MPV_frame_start(s, s->avctx) < 0)
1263
            return -1;
1264
        ff_er_frame_start(s);
1265
        r->cur_pts = r->si.pts;
1266
        if(s->pict_type != FF_B_TYPE){
1267
            r->last_pts = r->next_pts;
1268
            r->next_pts = r->cur_pts;
1269
        }
1270
        s->mb_x = s->mb_y = 0;
1271
    }
1272

    
1273
    r->si.end = end;
1274
    s->qscale = r->si.quant;
1275
    r->bits = buf_size*8;
1276
    s->mb_num_left = r->si.end - r->si.start;
1277
    r->s.mb_skip_run = 0;
1278

    
1279
    mb_pos = s->mb_x + s->mb_y * s->mb_width;
1280
    if(r->si.start != mb_pos){
1281
        av_log(s->avctx, AV_LOG_ERROR, "Slice indicates MB offset %d, got %d\n", r->si.start, mb_pos);
1282
        s->mb_x = r->si.start % s->mb_width;
1283
        s->mb_y = r->si.start / s->mb_width;
1284
    }
1285
    memset(r->intra_types_hist, -1, s->b4_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1286
    s->first_slice_line = 1;
1287
    s->resync_mb_x= s->mb_x;
1288
    s->resync_mb_y= s->mb_y;
1289

    
1290
    ff_init_block_index(s);
1291
    while(!check_slice_end(r, s)) {
1292
        ff_update_block_index(s);
1293
        s->dsp.clear_blocks(s->block[0]);
1294

    
1295
        if(rv34_decode_macroblock(r, r->intra_types + s->mb_x * 4 + 1) < 0){
1296
            ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, AC_ERROR|DC_ERROR|MV_ERROR);
1297
            return -1;
1298
        }
1299
        if (++s->mb_x == s->mb_width) {
1300
            s->mb_x = 0;
1301
            s->mb_y++;
1302
            ff_init_block_index(s);
1303

    
1304
            memmove(r->intra_types_hist, r->intra_types, s->b4_stride * 4 * sizeof(*r->intra_types_hist));
1305
            memset(r->intra_types, -1, s->b4_stride * 4 * sizeof(*r->intra_types_hist));
1306

    
1307
            if(r->loop_filter && s->mb_y >= 2)
1308
                r->loop_filter(r, s->mb_y - 2);
1309
        }
1310
        if(s->mb_x == s->resync_mb_x)
1311
            s->first_slice_line=0;
1312
        s->mb_num_left--;
1313
    }
1314
    ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, AC_END|DC_END|MV_END);
1315

    
1316
    return s->mb_y == s->mb_height;
1317
}
1318

    
1319
/** @} */ // recons group end
1320

    
1321
/**
1322
 * Initialize decoder.
1323
 */
1324
av_cold int ff_rv34_decode_init(AVCodecContext *avctx)
1325
{
1326
    RV34DecContext *r = avctx->priv_data;
1327
    MpegEncContext *s = &r->s;
1328

    
1329
    MPV_decode_defaults(s);
1330
    s->avctx= avctx;
1331
    s->out_format = FMT_H263;
1332
    s->codec_id= avctx->codec_id;
1333

    
1334
    s->width = avctx->width;
1335
    s->height = avctx->height;
1336

    
1337
    r->s.avctx = avctx;
1338
    avctx->flags |= CODEC_FLAG_EMU_EDGE;
1339
    r->s.flags |= CODEC_FLAG_EMU_EDGE;
1340
    avctx->pix_fmt = PIX_FMT_YUV420P;
1341
    avctx->has_b_frames = 1;
1342
    s->low_delay = 0;
1343

    
1344
    if (MPV_common_init(s) < 0)
1345
        return -1;
1346

    
1347
    ff_h264_pred_init(&r->h, CODEC_ID_RV40);
1348

    
1349
    r->intra_types_hist = av_malloc(s->b4_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1350
    r->intra_types = r->intra_types_hist + s->b4_stride * 4;
1351

    
1352
    r->mb_type = av_mallocz(r->s.mb_stride * r->s.mb_height * sizeof(*r->mb_type));
1353

    
1354
    r->cbp_luma   = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_luma));
1355
    r->cbp_chroma = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_chroma));
1356
    r->deblock_coefs = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->deblock_coefs));
1357

    
1358
    if(!intra_vlcs[0].cbppattern[0].bits)
1359
        rv34_init_tables();
1360

    
1361
    return 0;
1362
}
1363

    
1364
static int get_slice_offset(AVCodecContext *avctx, const uint8_t *buf, int n)
1365
{
1366
    if(avctx->slice_count) return avctx->slice_offset[n];
1367
    else                   return AV_RL32(buf + n*8 - 4) == 1 ? AV_RL32(buf + n*8) :  AV_RB32(buf + n*8);
1368
}
1369

    
1370
int ff_rv34_decode_frame(AVCodecContext *avctx,
1371
                            void *data, int *data_size,
1372
                            const uint8_t *buf, int buf_size)
1373
{
1374
    RV34DecContext *r = avctx->priv_data;
1375
    MpegEncContext *s = &r->s;
1376
    AVFrame *pict = data;
1377
    SliceInfo si;
1378
    int i;
1379
    int slice_count;
1380
    const uint8_t *slices_hdr = NULL;
1381
    int last = 0;
1382

    
1383
    /* no supplementary picture */
1384
    if (buf_size == 0) {
1385
        /* special case for last picture */
1386
        if (s->low_delay==0 && s->next_picture_ptr) {
1387
            *pict= *(AVFrame*)s->next_picture_ptr;
1388
            s->next_picture_ptr= NULL;
1389

    
1390
            *data_size = sizeof(AVFrame);
1391
        }
1392
        return 0;
1393
    }
1394

    
1395
    if(!avctx->slice_count){
1396
        slice_count = (*buf++) + 1;
1397
        slices_hdr = buf + 4;
1398
        buf += 8 * slice_count;
1399
    }else
1400
        slice_count = avctx->slice_count;
1401

    
1402
    //parse first slice header to check whether this frame can be decoded
1403
    if(get_slice_offset(avctx, slices_hdr, 0) > buf_size){
1404
        av_log(avctx, AV_LOG_ERROR, "Slice offset is greater than frame size\n");
1405
        return -1;
1406
    }
1407
    init_get_bits(&s->gb, buf+get_slice_offset(avctx, slices_hdr, 0), buf_size-get_slice_offset(avctx, slices_hdr, 0));
1408
    if(r->parse_slice_header(r, &r->s.gb, &si) < 0 || si.start){
1409
        av_log(avctx, AV_LOG_ERROR, "First slice header is incorrect\n");
1410
        return -1;
1411
    }
1412
    if((!s->last_picture_ptr || !s->last_picture_ptr->data[0]) && si.type == FF_B_TYPE)
1413
        return -1;
1414
    /* skip b frames if we are in a hurry */
1415
    if(avctx->hurry_up && si.type==FF_B_TYPE) return buf_size;
1416
    if(   (avctx->skip_frame >= AVDISCARD_NONREF && si.type==FF_B_TYPE)
1417
       || (avctx->skip_frame >= AVDISCARD_NONKEY && si.type!=FF_I_TYPE)
1418
       ||  avctx->skip_frame >= AVDISCARD_ALL)
1419
        return buf_size;
1420
    /* skip everything if we are in a hurry>=5 */
1421
    if(avctx->hurry_up>=5)
1422
        return buf_size;
1423

    
1424
    for(i=0; i<slice_count; i++){
1425
        int offset= get_slice_offset(avctx, slices_hdr, i);
1426
        int size;
1427
        if(i+1 == slice_count)
1428
            size= buf_size - offset;
1429
        else
1430
            size= get_slice_offset(avctx, slices_hdr, i+1) - offset;
1431

    
1432
        if(offset > buf_size){
1433
            av_log(avctx, AV_LOG_ERROR, "Slice offset is greater than frame size\n");
1434
            break;
1435
        }
1436

    
1437
        r->si.end = s->mb_width * s->mb_height;
1438
        if(i+1 < slice_count){
1439
            init_get_bits(&s->gb, buf+get_slice_offset(avctx, slices_hdr, i+1), (buf_size-get_slice_offset(avctx, slices_hdr, i+1))*8);
1440
            if(r->parse_slice_header(r, &r->s.gb, &si) < 0){
1441
                if(i+2 < slice_count)
1442
                    size = get_slice_offset(avctx, slices_hdr, i+2) - offset;
1443
                else
1444
                    size = buf_size - offset;
1445
            }else
1446
                r->si.end = si.start;
1447
        }
1448
        last = rv34_decode_slice(r, r->si.end, buf + offset, size);
1449
        s->mb_num_left = r->s.mb_x + r->s.mb_y*r->s.mb_width - r->si.start;
1450
        if(last)
1451
            break;
1452
    }
1453

    
1454
    if(last){
1455
        if(r->loop_filter)
1456
            r->loop_filter(r, s->mb_height - 1);
1457
        ff_er_frame_end(s);
1458
        MPV_frame_end(s);
1459
        if (s->pict_type == FF_B_TYPE || s->low_delay) {
1460
            *pict= *(AVFrame*)s->current_picture_ptr;
1461
        } else if (s->last_picture_ptr != NULL) {
1462
            *pict= *(AVFrame*)s->last_picture_ptr;
1463
        }
1464

    
1465
        if(s->last_picture_ptr || s->low_delay){
1466
            *data_size = sizeof(AVFrame);
1467
            ff_print_debug_info(s, pict);
1468
        }
1469
        s->current_picture_ptr= NULL; //so we can detect if frame_end wasnt called (find some nicer solution...)
1470
    }
1471
    return buf_size;
1472
}
1473

    
1474
av_cold int ff_rv34_decode_end(AVCodecContext *avctx)
1475
{
1476
    RV34DecContext *r = avctx->priv_data;
1477

    
1478
    MPV_common_end(&r->s);
1479

    
1480
    av_freep(&r->intra_types_hist);
1481
    r->intra_types = NULL;
1482
    av_freep(&r->mb_type);
1483
    av_freep(&r->cbp_luma);
1484
    av_freep(&r->cbp_chroma);
1485
    av_freep(&r->deblock_coefs);
1486

    
1487
    return 0;
1488
}