Statistics
| Branch: | Revision:

ffmpeg / libavcodec / rv34.c @ 39902a8c

History | View | Annotate | Download (45.2 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 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 "rectangle.h"
32

    
33
#include "rv34vlc.h"
34
#include "rv34data.h"
35
#include "rv34.h"
36

    
37
//#define DEBUG
38

    
39
/** translation of RV30/40 macroblock types to lavc ones */
40
static const int rv34_mb_type_to_lavc[12] = {
41
    MB_TYPE_INTRA,
42
    MB_TYPE_INTRA16x16,
43
    MB_TYPE_16x16   | MB_TYPE_L0,
44
    MB_TYPE_8x8     | MB_TYPE_L0,
45
    MB_TYPE_16x16   | MB_TYPE_L0,
46
    MB_TYPE_16x16   | MB_TYPE_L1,
47
    MB_TYPE_SKIP,
48
    MB_TYPE_DIRECT2 | MB_TYPE_16x16,
49
    MB_TYPE_16x8    | MB_TYPE_L0,
50
    MB_TYPE_8x16    | MB_TYPE_L0,
51
    MB_TYPE_16x16   | MB_TYPE_L0L1,
52
    MB_TYPE_16x16   | MB_TYPE_L0
53
};
54

    
55

    
56
static RV34VLC intra_vlcs[NUM_INTRA_TABLES], inter_vlcs[NUM_INTER_TABLES];
57

    
58
/**
59
 * @defgroup vlc RV30/40 VLC generating functions
60
 * @{
61
 */
62

    
63
/**
64
 * Generate VLC from codeword lengths.
65
 * @param bits   codeword lengths (zeroes are accepted)
66
 * @param size   length of input data
67
 * @param insyms symbols for input codes (NULL for default ones)
68
 */
69
static void rv34_gen_vlc(const uint8_t *bits, int size, VLC *vlc, const uint8_t *insyms)
70
{
71
    int i;
72
    int counts[17] = {0}, codes[17];
73
    uint16_t cw[size], syms[size];
74
    uint8_t bits2[size];
75
    int maxbits = 0, realsize = 0;
76

    
77
    for(i = 0; i < size; i++){
78
        if(bits[i]){
79
            bits2[realsize] = bits[i];
80
            syms[realsize] = insyms ? insyms[i] : i;
81
            realsize++;
82
            maxbits = FFMAX(maxbits, bits[i]);
83
            counts[bits[i]]++;
84
        }
85
    }
86

    
87
    codes[0] = 0;
88
    for(i = 0; i < 16; i++)
89
        codes[i+1] = (codes[i] + counts[i]) << 1;
90
    for(i = 0; i < realsize; i++)
91
        cw[i] = codes[bits2[i]]++;
92

    
93
    init_vlc_sparse(vlc, FFMIN(maxbits, 9), realsize,
94
                    bits2, 1, 1,
95
                    cw,    2, 2,
96
                    syms,  2, 2, INIT_VLC_USE_STATIC);
97
}
98

    
99
/**
100
 * Initialize all tables.
101
 */
102
static av_cold void rv34_init_tables()
103
{
104
    int i, j, k;
105

    
106
    for(i = 0; i < NUM_INTRA_TABLES; i++){
107
        for(j = 0; j < 2; j++){
108
            rv34_gen_vlc(rv34_table_intra_cbppat   [i][j], CBPPAT_VLC_SIZE,   &intra_vlcs[i].cbppattern[j],     NULL);
109
            rv34_gen_vlc(rv34_table_intra_secondpat[i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].second_pattern[j], NULL);
110
            rv34_gen_vlc(rv34_table_intra_thirdpat [i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].third_pattern[j],  NULL);
111
            for(k = 0; k < 4; k++)
112
                rv34_gen_vlc(rv34_table_intra_cbp[i][j+k*2],  CBP_VLC_SIZE,   &intra_vlcs[i].cbp[j][k],         rv34_cbp_code);
113
        }
114
        for(j = 0; j < 4; j++)
115
            rv34_gen_vlc(rv34_table_intra_firstpat[i][j], FIRSTBLK_VLC_SIZE, &intra_vlcs[i].first_pattern[j], NULL);
116
        rv34_gen_vlc(rv34_intra_coeff[i], COEFF_VLC_SIZE, &intra_vlcs[i].coefficient, NULL);
117
    }
118

    
119
    for(i = 0; i < NUM_INTER_TABLES; i++){
120
        rv34_gen_vlc(rv34_inter_cbppat[i], CBPPAT_VLC_SIZE, &inter_vlcs[i].cbppattern[0], NULL);
121
        for(j = 0; j < 4; j++)
122
            rv34_gen_vlc(rv34_inter_cbp[i][j], CBP_VLC_SIZE, &inter_vlcs[i].cbp[0][j], rv34_cbp_code);
123
        for(j = 0; j < 2; j++){
124
            rv34_gen_vlc(rv34_table_inter_firstpat [i][j], FIRSTBLK_VLC_SIZE, &inter_vlcs[i].first_pattern[j],  NULL);
125
            rv34_gen_vlc(rv34_table_inter_secondpat[i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].second_pattern[j], NULL);
126
            rv34_gen_vlc(rv34_table_inter_thirdpat [i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].third_pattern[j],  NULL);
127
        }
128
        rv34_gen_vlc(rv34_inter_coeff[i], COEFF_VLC_SIZE, &inter_vlcs[i].coefficient, NULL);
129
    }
130
}
131

    
132
/** @} */ // vlc group
133

    
134

    
135
/**
136
 * @defgroup transform RV30/40 inverse transform functions
137
 * @{
138
 */
139

    
140
static av_always_inline void rv34_row_transform(int temp[16], DCTELEM *block)
141
{
142
    int i;
143

    
144
    for(i=0; i<4; i++){
145
        const int z0= 13*(block[i+8*0] +    block[i+8*2]);
146
        const int z1= 13*(block[i+8*0] -    block[i+8*2]);
147
        const int z2=  7* block[i+8*1] - 17*block[i+8*3];
148
        const int z3= 17* block[i+8*1] +  7*block[i+8*3];
149

    
150
        temp[4*i+0]= z0+z3;
151
        temp[4*i+1]= z1+z2;
152
        temp[4*i+2]= z1-z2;
153
        temp[4*i+3]= z0-z3;
154
    }
155
}
156

    
157
/**
158
 * Real Video 3.0/4.0 inverse transform
159
 * Code is almost the same as in SVQ3, only scaling is different.
160
 */
161
static void rv34_inv_transform(DCTELEM *block){
162
    int temp[16];
163
    int i;
164

    
165
    rv34_row_transform(temp, block);
166

    
167
    for(i=0; i<4; i++){
168
        const int z0= 13*(temp[4*0+i] +    temp[4*2+i]) + 0x200;
169
        const int z1= 13*(temp[4*0+i] -    temp[4*2+i]) + 0x200;
170
        const int z2=  7* temp[4*1+i] - 17*temp[4*3+i];
171
        const int z3= 17* temp[4*1+i] +  7*temp[4*3+i];
172

    
173
        block[i*8+0]= (z0 + z3)>>10;
174
        block[i*8+1]= (z1 + z2)>>10;
175
        block[i*8+2]= (z1 - z2)>>10;
176
        block[i*8+3]= (z0 - z3)>>10;
177
    }
178

    
179
}
180

    
181
/**
182
 * RealVideo 3.0/4.0 inverse transform for DC block
183
 *
184
 * Code is almost the same as rv34_inv_transform()
185
 * but final coefficients are multiplied by 1.5 and have no rounding.
186
 */
187
static void rv34_inv_transform_noround(DCTELEM *block){
188
    int temp[16];
189
    int i;
190

    
191
    rv34_row_transform(temp, block);
192

    
193
    for(i=0; i<4; i++){
194
        const int z0= 13*(temp[4*0+i] +    temp[4*2+i]);
195
        const int z1= 13*(temp[4*0+i] -    temp[4*2+i]);
196
        const int z2=  7* temp[4*1+i] - 17*temp[4*3+i];
197
        const int z3= 17* temp[4*1+i] +  7*temp[4*3+i];
198

    
199
        block[i*8+0]= ((z0 + z3)*3)>>11;
200
        block[i*8+1]= ((z1 + z2)*3)>>11;
201
        block[i*8+2]= ((z1 - z2)*3)>>11;
202
        block[i*8+3]= ((z0 - z3)*3)>>11;
203
    }
204

    
205
}
206

    
207
/** @} */ // transform
208

    
209

    
210
/**
211
 * @defgroup block RV30/40 4x4 block decoding functions
212
 * @{
213
 */
214

    
215
/**
216
 * Decode coded block pattern.
217
 */
218
static int rv34_decode_cbp(GetBitContext *gb, RV34VLC *vlc, int table)
219
{
220
    int pattern, code, cbp=0;
221
    int ones;
222
    static const int cbp_masks[3] = {0x100000, 0x010000, 0x110000};
223
    static const int shifts[4] = { 0, 2, 8, 10 };
224
    int *curshift = shifts;
225
    int i, t, mask;
226

    
227
    code = get_vlc2(gb, vlc->cbppattern[table].table, 9, 2);
228
    pattern = code & 0xF;
229
    code >>= 4;
230

    
231
    ones = rv34_count_ones[pattern];
232

    
233
    for(mask = 8; mask; mask >>= 1, curshift++){
234
        if(pattern & mask)
235
            cbp |= get_vlc2(gb, vlc->cbp[table][ones].table, vlc->cbp[table][ones].bits, 1) << curshift[0];
236
    }
237

    
238
    for(i = 0; i < 4; i++){
239
        t = modulo_three_table[code][i];
240
        if(t == 1)
241
            cbp |= cbp_masks[get_bits1(gb)] << i;
242
        if(t == 2)
243
            cbp |= cbp_masks[2] << i;
244
    }
245
    return cbp;
246
}
247

    
248
/**
249
 * Get one coefficient value from the bistream and store it.
250
 */
251
static inline void decode_coeff(DCTELEM *dst, int coef, int esc, GetBitContext *gb, VLC* vlc)
252
{
253
    if(coef){
254
        if(coef == esc){
255
            coef = get_vlc2(gb, vlc->table, 9, 2);
256
            if(coef > 23){
257
                coef -= 23;
258
                coef = 22 + ((1 << coef) | get_bits(gb, coef));
259
            }
260
            coef += esc;
261
        }
262
        if(get_bits1(gb))
263
            coef = -coef;
264
        *dst = coef;
265
    }
266
}
267

    
268
/**
269
 * Decode 2x2 subblock of coefficients.
270
 */
271
static inline void decode_subblock(DCTELEM *dst, int code, const int is_block2, GetBitContext *gb, VLC *vlc)
272
{
273
    int coeffs[4];
274

    
275
    coeffs[0] = modulo_three_table[code][0];
276
    coeffs[1] = modulo_three_table[code][1];
277
    coeffs[2] = modulo_three_table[code][2];
278
    coeffs[3] = modulo_three_table[code][3];
279
    decode_coeff(dst  , coeffs[0], 3, gb, vlc);
280
    if(is_block2){
281
        decode_coeff(dst+8, coeffs[1], 2, gb, vlc);
282
        decode_coeff(dst+1, coeffs[2], 2, gb, vlc);
283
    }else{
284
        decode_coeff(dst+1, coeffs[1], 2, gb, vlc);
285
        decode_coeff(dst+8, coeffs[2], 2, gb, vlc);
286
    }
287
    decode_coeff(dst+9, coeffs[3], 2, gb, vlc);
288
}
289

    
290
/**
291
 * Decode coefficients for 4x4 block.
292
 *
293
 * This is done by filling 2x2 subblocks with decoded coefficients
294
 * in this order (the same for subblocks and subblock coefficients):
295
 *  o--o
296
 *    /
297
 *   /
298
 *  o--o
299
 */
300

    
301
static inline void rv34_decode_block(DCTELEM *dst, GetBitContext *gb, RV34VLC *rvlc, int fc, int sc)
302
{
303
    int code, pattern;
304

    
305
    code = get_vlc2(gb, rvlc->first_pattern[fc].table, 9, 2);
306

    
307
    pattern = code & 0x7;
308

    
309
    code >>= 3;
310
    decode_subblock(dst, code, 0, gb, &rvlc->coefficient);
311

    
312
    if(pattern & 4){
313
        code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
314
        decode_subblock(dst + 2, code, 0, gb, &rvlc->coefficient);
315
    }
316
    if(pattern & 2){ // Looks like coefficients 1 and 2 are swapped for this block
317
        code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
318
        decode_subblock(dst + 8*2, code, 1, gb, &rvlc->coefficient);
319
    }
320
    if(pattern & 1){
321
        code = get_vlc2(gb, rvlc->third_pattern[sc].table, 9, 2);
322
        decode_subblock(dst + 8*2+2, code, 0, gb, &rvlc->coefficient);
323
    }
324

    
325
}
326

    
327
/**
328
 * Dequantize ordinary 4x4 block.
329
 * @todo optimize
330
 */
331
static inline void rv34_dequant4x4(DCTELEM *block, int Qdc, int Q)
332
{
333
    int i, j;
334

    
335
    block[0] = (block[0] * Qdc + 8) >> 4;
336
    for(i = 0; i < 4; i++)
337
        for(j = !i; j < 4; j++)
338
            block[j + i*8] = (block[j + i*8] * Q + 8) >> 4;
339
}
340

    
341
/**
342
 * Dequantize 4x4 block of DC values for 16x16 macroblock.
343
 * @todo optimize
344
 */
345
static inline void rv34_dequant4x4_16x16(DCTELEM *block, int Qdc, int Q)
346
{
347
    int i;
348

    
349
    for(i = 0; i < 3; i++)
350
         block[rv34_dezigzag[i]] = (block[rv34_dezigzag[i]] * Qdc + 8) >> 4;
351
    for(; i < 16; i++)
352
         block[rv34_dezigzag[i]] = (block[rv34_dezigzag[i]] * Q + 8) >> 4;
353
}
354
/** @} */ //block functions
355

    
356

    
357
/**
358
 * @defgroup bitstream RV30/40 bitstream parsing
359
 * @{
360
 */
361

    
362
/**
363
 * Decode starting slice position.
364
 * @todo Maybe replace with ff_h263_decode_mba() ?
365
 */
366
int ff_rv34_get_start_offset(GetBitContext *gb, int mb_size)
367
{
368
    int i;
369
    for(i = 0; i < 5; i++)
370
        if(rv34_mb_max_sizes[i] > mb_size)
371
            break;
372
    return rv34_mb_bits_sizes[i];
373
}
374

    
375
/**
376
 * Select VLC set for decoding from current quantizer, modifier and frame type.
377
 */
378
static inline RV34VLC* choose_vlc_set(int quant, int mod, int type)
379
{
380
    if(mod == 2 && quant < 19) quant += 10;
381
    else if(mod && quant < 26) quant += 5;
382
    return type ? &inter_vlcs[rv34_quant_to_vlc_set[1][av_clip(quant, 0, 30)]]
383
                : &intra_vlcs[rv34_quant_to_vlc_set[0][av_clip(quant, 0, 30)]];
384
}
385

    
386
/**
387
 * Decode quantizer difference and return modified quantizer.
388
 */
389
static inline int rv34_decode_dquant(GetBitContext *gb, int quant)
390
{
391
    if(get_bits1(gb))
392
        return rv34_dquant_tab[get_bits1(gb)][quant];
393
    else
394
        return get_bits(gb, 5);
395
}
396

    
397
/** @} */ //bitstream functions
398

    
399
/**
400
 * @defgroup mv motion vector related code (prediction, reconstruction, motion compensation)
401
 * @{
402
 */
403

    
404
/** macroblock partition width in 8x8 blocks */
405
static const uint8_t part_sizes_w[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 2, 1, 2, 2 };
406

    
407
/** macroblock partition height in 8x8 blocks */
408
static const uint8_t part_sizes_h[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 1, 2, 2, 2 };
409

    
410
/** availability index for subblocks */
411
static const uint8_t avail_indexes[4] = { 5, 6, 9, 10 };
412

    
413
/**
414
 * motion vector prediction
415
 *
416
 * Motion prediction performed for the block by using median prediction of
417
 * motion vectors from the left, top and right top blocks but in corner cases
418
 * some other vectors may be used instead.
419
 */
420
static void rv34_pred_mv(RV34DecContext *r, int block_type, int subblock_no, int dmv_no)
421
{
422
    MpegEncContext *s = &r->s;
423
    int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
424
    int A[2] = {0}, B[2], C[2];
425
    int i, j;
426
    int mx, my;
427
    int avail_index = avail_indexes[subblock_no];
428
    int c_off = part_sizes_w[block_type];
429

    
430
    mv_pos += (subblock_no & 1) + (subblock_no >> 1)*s->b8_stride;
431
    if(subblock_no == 3)
432
        c_off = -1;
433

    
434
    if(r->avail_cache[avail_index - 1]){
435
        A[0] = s->current_picture_ptr->motion_val[0][mv_pos-1][0];
436
        A[1] = s->current_picture_ptr->motion_val[0][mv_pos-1][1];
437
    }
438
    if(r->avail_cache[avail_index - 4]){
439
        B[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][0];
440
        B[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][1];
441
    }else{
442
        B[0] = A[0];
443
        B[1] = A[1];
444
    }
445
    if(!r->avail_cache[avail_index - 4 + c_off]){
446
        if(r->avail_cache[avail_index - 4] && (r->avail_cache[avail_index - 1] || r->rv30)){
447
            C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][0];
448
            C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][1];
449
        }else{
450
            C[0] = A[0];
451
            C[1] = A[1];
452
        }
453
    }else{
454
        C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][0];
455
        C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][1];
456
    }
457
    mx = mid_pred(A[0], B[0], C[0]);
458
    my = mid_pred(A[1], B[1], C[1]);
459
    mx += r->dmv[dmv_no][0];
460
    my += r->dmv[dmv_no][1];
461
    for(j = 0; j < part_sizes_h[block_type]; j++){
462
        for(i = 0; i < part_sizes_w[block_type]; i++){
463
            s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][0] = mx;
464
            s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][1] = my;
465
        }
466
    }
467
}
468

    
469
#define GET_PTS_DIFF(a, b) ((a - b + 8192) & 0x1FFF)
470

    
471
/**
472
 * Calculate motion vector component that should be added for direct blocks.
473
 */
474
static int calc_add_mv(RV34DecContext *r, int dir, int val)
475
{
476
    int refdist = GET_PTS_DIFF(r->next_pts, r->last_pts);
477
    int dist = dir ? GET_PTS_DIFF(r->next_pts, r->cur_pts) : GET_PTS_DIFF(r->cur_pts, r->last_pts);
478

    
479
    if(!refdist) return 0;
480
    if(!dir)
481
        return (val * dist + refdist - 1) / refdist;
482
    else
483
        return -(val * dist / refdist);
484
}
485

    
486
/**
487
 * Predict motion vector for B-frame macroblock.
488
 */
489
static inline void rv34_pred_b_vector(int A[2], int B[2], int C[2],
490
                                      int A_avail, int B_avail, int C_avail,
491
                                      int *mx, int *my)
492
{
493
    if(A_avail + B_avail + C_avail != 3){
494
        *mx = A[0] + B[0] + C[0];
495
        *my = A[1] + B[1] + C[1];
496
        if(A_avail + B_avail + C_avail == 2){
497
            *mx /= 2;
498
            *my /= 2;
499
        }
500
    }else{
501
        *mx = mid_pred(A[0], B[0], C[0]);
502
        *my = mid_pred(A[1], B[1], C[1]);
503
    }
504
}
505

    
506
/**
507
 * motion vector prediction for B-frames
508
 */
509
static void rv34_pred_mv_b(RV34DecContext *r, int block_type, int dir)
510
{
511
    MpegEncContext *s = &r->s;
512
    int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
513
    int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
514
    int A[2], B[2], C[2];
515
    int has_A = 0, has_B = 0, has_C = 0;
516
    int mx, my;
517
    int i, j;
518
    Picture *cur_pic = s->current_picture_ptr;
519
    const int mask = dir ? MB_TYPE_L1 : MB_TYPE_L0;
520
    int type = cur_pic->mb_type[mb_pos];
521

    
522
    memset(A, 0, sizeof(A));
523
    memset(B, 0, sizeof(B));
524
    memset(C, 0, sizeof(C));
525
    if((r->avail_cache[5-1] & type) & mask){
526
        A[0] = cur_pic->motion_val[dir][mv_pos - 1][0];
527
        A[1] = cur_pic->motion_val[dir][mv_pos - 1][1];
528
        has_A = 1;
529
    }
530
    if((r->avail_cache[5-4] & type) & mask){
531
        B[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][0];
532
        B[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][1];
533
        has_B = 1;
534
    }
535
    if((r->avail_cache[5-2] & type) & mask){
536
        C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][0];
537
        C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][1];
538
        has_C = 1;
539
    }else if((s->mb_x+1) == s->mb_width && (r->avail_cache[5-5] & type) & mask){
540
        C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][0];
541
        C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][1];
542
        has_C = 1;
543
    }
544

    
545
    rv34_pred_b_vector(A, B, C, has_A, has_B, has_C, &mx, &my);
546

    
547
    mx += r->dmv[dir][0];
548
    my += r->dmv[dir][1];
549

    
550
    for(j = 0; j < 2; j++){
551
        for(i = 0; i < 2; i++){
552
            cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][0] = mx;
553
            cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][1] = my;
554
        }
555
    }
556
    if(block_type == RV34_MB_B_BACKWARD || block_type == RV34_MB_B_FORWARD)
557
        fill_rectangle(cur_pic->motion_val[!dir][mv_pos], 2, 2, s->b8_stride, 0, 4);
558
}
559

    
560
static const int chroma_coeffs[3] = { 8, 5, 3 };
561

    
562
/**
563
 * generic motion compensation function
564
 *
565
 * @param r decoder context
566
 * @param block_type type of the current block
567
 * @param xoff horizontal offset from the start of the current block
568
 * @param yoff vertical offset from the start of the current block
569
 * @param mv_off offset to the motion vector information
570
 * @param width width of the current partition in 8x8 blocks
571
 * @param height height of the current partition in 8x8 blocks
572
 */
573
static inline void rv34_mc(RV34DecContext *r, const int block_type,
574
                          const int xoff, const int yoff, int mv_off,
575
                          const int width, const int height, int dir,
576
                          const int thirdpel,
577
                          qpel_mc_func (*qpel_mc)[16],
578
                          h264_chroma_mc_func (*chroma_mc))
579
{
580
    MpegEncContext *s = &r->s;
581
    uint8_t *Y, *U, *V, *srcY, *srcU, *srcV;
582
    int dxy, mx, my, lx, ly, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
583
    int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride + mv_off;
584
    int is16x16 = 1;
585

    
586
    if(thirdpel){
587
        mx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) / 3 - (1 << 24);
588
        my = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) / 3 - (1 << 24);
589
        lx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) % 3;
590
        ly = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) % 3;
591
        uvmx = chroma_coeffs[(3*(mx&1) + lx) >> 1];
592
        uvmy = chroma_coeffs[(3*(my&1) + ly) >> 1];
593
    }else{
594
        mx = s->current_picture_ptr->motion_val[dir][mv_pos][0] >> 2;
595
        my = s->current_picture_ptr->motion_val[dir][mv_pos][1] >> 2;
596
        lx = s->current_picture_ptr->motion_val[dir][mv_pos][0] & 3;
597
        ly = s->current_picture_ptr->motion_val[dir][mv_pos][1] & 3;
598
        uvmx = mx & 6;
599
        uvmy = my & 6;
600
    }
601
    dxy = ly*4 + lx;
602
    srcY = dir ? s->next_picture_ptr->data[0] : s->last_picture_ptr->data[0];
603
    srcU = dir ? s->next_picture_ptr->data[1] : s->last_picture_ptr->data[1];
604
    srcV = dir ? s->next_picture_ptr->data[2] : s->last_picture_ptr->data[2];
605
    src_x = s->mb_x * 16 + xoff + mx;
606
    src_y = s->mb_y * 16 + yoff + my;
607
    uvsrc_x = s->mb_x * 8 + (xoff >> 1) + (mx >> 1);
608
    uvsrc_y = s->mb_y * 8 + (yoff >> 1) + (my >> 1);
609
    srcY += src_y * s->linesize + src_x;
610
    srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
611
    srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
612
    if(   (unsigned)(src_x - !!lx*2) > s->h_edge_pos - !!lx*2 - (width <<3) - 3
613
       || (unsigned)(src_y - !!ly*2) > s->v_edge_pos - !!ly*2 - (height<<3) - 3){
614
        uint8_t *uvbuf= s->edge_emu_buffer + 20 * s->linesize;
615

    
616
        srcY -= 2 + 2*s->linesize;
617
        ff_emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, (width<<3)+4, (height<<3)+4,
618
                            src_x - 2, src_y - 2, s->h_edge_pos, s->v_edge_pos);
619
        srcY = s->edge_emu_buffer + 2 + 2*s->linesize;
620
        ff_emulated_edge_mc(uvbuf     , srcU, s->uvlinesize, (width<<2)+1, (height<<2)+1,
621
                            uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
622
        ff_emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, (width<<2)+1, (height<<2)+1,
623
                            uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
624
        srcU = uvbuf;
625
        srcV = uvbuf + 16;
626
    }
627
    Y = s->dest[0] + xoff      + yoff     *s->linesize;
628
    U = s->dest[1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
629
    V = s->dest[2] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
630

    
631
    if(block_type == RV34_MB_P_16x8){
632
        qpel_mc[1][dxy](Y, srcY, s->linesize);
633
        Y    += 8;
634
        srcY += 8;
635
    }else if(block_type == RV34_MB_P_8x16){
636
        qpel_mc[1][dxy](Y, srcY, s->linesize);
637
        Y    += 8 * s->linesize;
638
        srcY += 8 * s->linesize;
639
    }
640
    is16x16 = (block_type != RV34_MB_P_8x8) && (block_type != RV34_MB_P_16x8) && (block_type != RV34_MB_P_8x16);
641
    qpel_mc[!is16x16][dxy](Y, srcY, s->linesize);
642
    chroma_mc[2-width]   (U, srcU, s->uvlinesize, height*4, uvmx, uvmy);
643
    chroma_mc[2-width]   (V, srcV, s->uvlinesize, height*4, uvmx, uvmy);
644
}
645

    
646
static void rv34_mc_1mv(RV34DecContext *r, const int block_type,
647
                        const int xoff, const int yoff, int mv_off,
648
                        const int width, const int height, int dir)
649
{
650
    rv34_mc(r, block_type, xoff, yoff, mv_off, width, height, dir, r->rv30,
651
            r->rv30 ? r->s.dsp.put_rv30_tpel_pixels_tab
652
                    : r->s.dsp.put_h264_qpel_pixels_tab,
653
            r->s.dsp.put_h264_chroma_pixels_tab);
654
}
655

    
656
static void rv34_mc_2mv(RV34DecContext *r, const int block_type)
657
{
658
    rv34_mc(r, block_type, 0, 0, 0, 2, 2, 0, r->rv30,
659
            r->rv30 ? r->s.dsp.put_rv30_tpel_pixels_tab
660
                    : r->s.dsp.put_h264_qpel_pixels_tab,
661
            r->s.dsp.put_h264_chroma_pixels_tab);
662
    rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30,
663
            r->rv30 ? r->s.dsp.avg_rv30_tpel_pixels_tab
664
                    : r->s.dsp.avg_h264_qpel_pixels_tab,
665
            r->s.dsp.avg_h264_chroma_pixels_tab);
666
}
667

    
668
static void rv34_mc_2mv_skip(RV34DecContext *r)
669
{
670
    int i, j, k;
671
    for(j = 0; j < 2; j++)
672
        for(i = 0; i < 2; i++){
673
             rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 0, r->rv30,
674
                    r->rv30 ? r->s.dsp.put_rv30_tpel_pixels_tab
675
                            : r->s.dsp.put_h264_qpel_pixels_tab,
676
                    r->s.dsp.put_h264_chroma_pixels_tab);
677
             rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 1, r->rv30,
678
                    r->rv30 ? r->s.dsp.avg_rv30_tpel_pixels_tab
679
                            : r->s.dsp.avg_h264_qpel_pixels_tab,
680
                    r->s.dsp.avg_h264_chroma_pixels_tab);
681
        }
682
}
683

    
684
/** number of motion vectors in each macroblock type */
685
static const int num_mvs[RV34_MB_TYPES] = { 0, 0, 1, 4, 1, 1, 0, 0, 2, 2, 2, 1 };
686

    
687
/**
688
 * Decode motion vector differences
689
 * and perform motion vector reconstruction and motion compensation.
690
 */
691
static int rv34_decode_mv(RV34DecContext *r, int block_type)
692
{
693
    MpegEncContext *s = &r->s;
694
    GetBitContext *gb = &s->gb;
695
    int i, j, k, l;
696
    int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
697
    int next_bt;
698

    
699
    memset(r->dmv, 0, sizeof(r->dmv));
700
    for(i = 0; i < num_mvs[block_type]; i++){
701
        r->dmv[i][0] = svq3_get_se_golomb(gb);
702
        r->dmv[i][1] = svq3_get_se_golomb(gb);
703
    }
704
    switch(block_type){
705
    case RV34_MB_TYPE_INTRA:
706
    case RV34_MB_TYPE_INTRA16x16:
707
        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);
708
        return 0;
709
    case RV34_MB_SKIP:
710
        if(s->pict_type == FF_P_TYPE){
711
            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);
712
            rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
713
            break;
714
        }
715
    case RV34_MB_B_DIRECT:
716
        //surprisingly, it uses motion scheme from next reference frame
717
        next_bt = s->next_picture_ptr->mb_type[s->mb_x + s->mb_y * s->mb_stride];
718
        for(j = 0; j < 2; j++)
719
            for(i = 0; i < 2; i++)
720
                for(k = 0; k < 2; k++)
721
                    for(l = 0; l < 2; l++)
722
                        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]);
723
        if(IS_16X16(next_bt)) //we can use whole macroblock MC
724
            rv34_mc_2mv(r, block_type);
725
        else
726
            rv34_mc_2mv_skip(r);
727
        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);
728
        break;
729
    case RV34_MB_P_16x16:
730
    case RV34_MB_P_MIX16x16:
731
        rv34_pred_mv(r, block_type, 0, 0);
732
        rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
733
        break;
734
    case RV34_MB_B_FORWARD:
735
    case RV34_MB_B_BACKWARD:
736
        r->dmv[1][0] = r->dmv[0][0];
737
        r->dmv[1][1] = r->dmv[0][1];
738
        rv34_pred_mv_b  (r, block_type, block_type == RV34_MB_B_BACKWARD);
739
        rv34_mc_1mv     (r, block_type, 0, 0, 0, 2, 2, block_type == RV34_MB_B_BACKWARD);
740
        break;
741
    case RV34_MB_P_16x8:
742
    case RV34_MB_P_8x16:
743
        rv34_pred_mv(r, block_type, 0, 0);
744
        rv34_pred_mv(r, block_type, 1 + (block_type == RV34_MB_P_16x8), 1);
745
        if(block_type == RV34_MB_P_16x8){
746
            rv34_mc_1mv(r, block_type, 0, 0, 0,            2, 1, 0);
747
            rv34_mc_1mv(r, block_type, 0, 8, s->b8_stride, 2, 1, 0);
748
        }
749
        if(block_type == RV34_MB_P_8x16){
750
            rv34_mc_1mv(r, block_type, 0, 0, 0, 1, 2, 0);
751
            rv34_mc_1mv(r, block_type, 8, 0, 1, 1, 2, 0);
752
        }
753
        break;
754
    case RV34_MB_B_BIDIR:
755
        rv34_pred_mv_b  (r, block_type, 0);
756
        rv34_pred_mv_b  (r, block_type, 1);
757
        rv34_mc_2mv     (r, block_type);
758
        break;
759
    case RV34_MB_P_8x8:
760
        for(i=0;i< 4;i++){
761
            rv34_pred_mv(r, block_type, i, i);
762
            rv34_mc_1mv (r, block_type, (i&1)<<3, (i&2)<<2, (i&1)+(i>>1)*s->b8_stride, 1, 1, 0);
763
        }
764
        break;
765
    }
766

    
767
    return 0;
768
}
769
/** @} */ // mv group
770

    
771
/**
772
 * @defgroup recons Macroblock reconstruction functions
773
 * @{
774
 */
775
/** mapping of RV30/40 intra prediction types to standard H.264 types */
776
static const int ittrans[9] = {
777
 DC_PRED, VERT_PRED, HOR_PRED, DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_LEFT_PRED,
778
 VERT_RIGHT_PRED, VERT_LEFT_PRED, HOR_UP_PRED, HOR_DOWN_PRED,
779
};
780

    
781
/** mapping of RV30/40 intra 16x16 prediction types to standard H.264 types */
782
static const int ittrans16[4] = {
783
 DC_PRED8x8, VERT_PRED8x8, HOR_PRED8x8, PLANE_PRED8x8,
784
};
785

    
786
/**
787
 * Perform 4x4 intra prediction.
788
 */
789
static void rv34_pred_4x4_block(RV34DecContext *r, uint8_t *dst, int stride, int itype, int up, int left, int down, int right)
790
{
791
    uint8_t *prev = dst - stride + 4;
792
    uint32_t topleft;
793

    
794
    if(!up && !left)
795
        itype = DC_128_PRED;
796
    else if(!up){
797
        if(itype == VERT_PRED) itype = HOR_PRED;
798
        if(itype == DC_PRED)   itype = LEFT_DC_PRED;
799
    }else if(!left){
800
        if(itype == HOR_PRED)  itype = VERT_PRED;
801
        if(itype == DC_PRED)   itype = TOP_DC_PRED;
802
        if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
803
    }
804
    if(!down){
805
        if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
806
        if(itype == HOR_UP_PRED) itype = HOR_UP_PRED_RV40_NODOWN;
807
        if(itype == VERT_LEFT_PRED) itype = VERT_LEFT_PRED_RV40_NODOWN;
808
    }
809
    if(!right && up){
810
        topleft = dst[-stride + 3] * 0x01010101;
811
        prev = &topleft;
812
    }
813
    r->h.pred4x4[itype](dst, prev, stride);
814
}
815

    
816
/** add_pixels_clamped for 4x4 block */
817
static void rv34_add_4x4_block(uint8_t *dst, int stride, DCTELEM block[64], int off)
818
{
819
    int x, y;
820
    for(y = 0; y < 4; y++)
821
        for(x = 0; x < 4; x++)
822
            dst[x + y*stride] = av_clip_uint8(dst[x + y*stride] + block[off + x+y*8]);
823
}
824

    
825
static inline int adjust_pred16(int itype, int up, int left)
826
{
827
    if(!up && !left)
828
        itype = DC_128_PRED8x8;
829
    else if(!up){
830
        if(itype == PLANE_PRED8x8)itype = HOR_PRED8x8;
831
        if(itype == VERT_PRED8x8) itype = HOR_PRED8x8;
832
        if(itype == DC_PRED8x8)   itype = LEFT_DC_PRED8x8;
833
    }else if(!left){
834
        if(itype == PLANE_PRED8x8)itype = VERT_PRED8x8;
835
        if(itype == HOR_PRED8x8)  itype = VERT_PRED8x8;
836
        if(itype == DC_PRED8x8)   itype = TOP_DC_PRED8x8;
837
    }
838
    return itype;
839
}
840

    
841
static void rv34_output_macroblock(RV34DecContext *r, int8_t *intra_types, int cbp, int is16)
842
{
843
    MpegEncContext *s = &r->s;
844
    DSPContext *dsp = &s->dsp;
845
    int i, j;
846
    uint8_t *Y, *U, *V;
847
    int itype;
848
    int avail[6*8] = {0};
849
    int idx;
850

    
851
    // Set neighbour information.
852
    if(r->avail_cache[0])
853
        avail[0] = 1;
854
    if(r->avail_cache[1])
855
        avail[1] = avail[2] = 1;
856
    if(r->avail_cache[2])
857
        avail[3] = avail[4] = 1;
858
    if(r->avail_cache[3])
859
        avail[5] = 1;
860
    if(r->avail_cache[4])
861
        avail[8] = avail[16] = 1;
862
    if(r->avail_cache[8])
863
        avail[24] = avail[32] = 1;
864

    
865
    Y = s->dest[0];
866
    U = s->dest[1];
867
    V = s->dest[2];
868
    if(!is16){
869
        for(j = 0; j < 4; j++){
870
            idx = 9 + j*8;
871
            for(i = 0; i < 4; i++, cbp >>= 1, Y += 4, idx++){
872
                rv34_pred_4x4_block(r, Y, s->linesize, ittrans[intra_types[i]], avail[idx-8], avail[idx-1], avail[idx+7], avail[idx-7]);
873
                avail[idx] = 1;
874
                if(cbp & 1)
875
                    rv34_add_4x4_block(Y, s->linesize, s->block[(i>>1)+(j&2)], (i&1)*4+(j&1)*32);
876
            }
877
            Y += s->linesize * 4 - 4*4;
878
            intra_types += s->b4_stride;
879
        }
880
        intra_types -= s->b4_stride * 4;
881
        fill_rectangle(r->avail_cache + 5, 2, 2, 4, 0, 4);
882
        for(j = 0; j < 2; j++){
883
            idx = 5 + j*4;
884
            for(i = 0; i < 2; i++, cbp >>= 1, idx++){
885
                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]);
886
                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]);
887
                r->avail_cache[idx] = 1;
888
                if(cbp & 0x01)
889
                    rv34_add_4x4_block(U + i*4 + j*4*s->uvlinesize, s->uvlinesize, s->block[4], i*4+j*32);
890
                if(cbp & 0x10)
891
                    rv34_add_4x4_block(V + i*4 + j*4*s->uvlinesize, s->uvlinesize, s->block[5], i*4+j*32);
892
            }
893
        }
894
    }else{
895
        itype = ittrans16[intra_types[0]];
896
        itype = adjust_pred16(itype, r->avail_cache[5-4], r->avail_cache[5-1]);
897
        r->h.pred16x16[itype](Y, s->linesize);
898
        dsp->add_pixels_clamped(s->block[0], Y,     s->current_picture.linesize[0]);
899
        dsp->add_pixels_clamped(s->block[1], Y + 8, s->current_picture.linesize[0]);
900
        Y += s->current_picture.linesize[0] * 8;
901
        dsp->add_pixels_clamped(s->block[2], Y,     s->current_picture.linesize[0]);
902
        dsp->add_pixels_clamped(s->block[3], Y + 8, s->current_picture.linesize[0]);
903

    
904
        itype = ittrans16[intra_types[0]];
905
        if(itype == PLANE_PRED8x8) itype = DC_PRED8x8;
906
        itype = adjust_pred16(itype, r->avail_cache[5-4], r->avail_cache[5-1]);
907
        r->h.pred8x8[itype](U, s->uvlinesize);
908
        dsp->add_pixels_clamped(s->block[4], U, s->uvlinesize);
909
        r->h.pred8x8[itype](V, s->uvlinesize);
910
        dsp->add_pixels_clamped(s->block[5], V, s->uvlinesize);
911
    }
912
}
913

    
914
/** @} */ // recons group
915

    
916
/**
917
 * @addtogroup bitstream
918
 * Decode macroblock header and return CBP in case of success, -1 otherwise.
919
 */
920
static int rv34_decode_mb_header(RV34DecContext *r, int8_t *intra_types)
921
{
922
    MpegEncContext *s = &r->s;
923
    GetBitContext *gb = &s->gb;
924
    int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
925
    int i, t;
926

    
927
    if(!r->si.type){
928
        r->is16 = get_bits1(gb);
929
        if(!r->is16 && !r->rv30){
930
            if(!get_bits1(gb))
931
                av_log(s->avctx, AV_LOG_ERROR, "Need DQUANT\n");
932
        }
933
        s->current_picture_ptr->mb_type[mb_pos] = r->is16 ? MB_TYPE_INTRA16x16 : MB_TYPE_INTRA;
934
        r->block_type = r->is16 ? RV34_MB_TYPE_INTRA16x16 : RV34_MB_TYPE_INTRA;
935
    }else{
936
        r->block_type = r->decode_mb_info(r);
937
        if(r->block_type == -1)
938
            return -1;
939
        s->current_picture_ptr->mb_type[mb_pos] = rv34_mb_type_to_lavc[r->block_type];
940
        r->mb_type[mb_pos] = r->block_type;
941
        if(r->block_type == RV34_MB_SKIP){
942
            if(s->pict_type == FF_P_TYPE)
943
                r->mb_type[mb_pos] = RV34_MB_P_16x16;
944
            if(s->pict_type == FF_B_TYPE)
945
                r->mb_type[mb_pos] = RV34_MB_B_DIRECT;
946
        }
947
        r->is16 = !!IS_INTRA16x16(s->current_picture_ptr->mb_type[mb_pos]);
948
        rv34_decode_mv(r, r->block_type);
949
        if(r->block_type == RV34_MB_SKIP){
950
            fill_rectangle(intra_types, 4, 4, s->b4_stride, 0, sizeof(intra_types[0]));
951
            return 0;
952
        }
953
        r->chroma_vlc = 1;
954
        r->luma_vlc   = 0;
955
    }
956
    if(IS_INTRA(s->current_picture_ptr->mb_type[mb_pos])){
957
        if(r->is16){
958
            t = get_bits(gb, 2);
959
            fill_rectangle(intra_types, 4, 4, s->b4_stride, t, sizeof(intra_types[0]));
960
            r->luma_vlc   = 2;
961
        }else{
962
            if(r->decode_intra_types(r, gb, intra_types) < 0)
963
                return -1;
964
            r->luma_vlc   = 1;
965
        }
966
        r->chroma_vlc = 0;
967
        r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
968
    }else{
969
        for(i = 0; i < 16; i++)
970
            intra_types[(i & 3) + (i>>2) * s->b4_stride] = 0;
971
        r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
972
        if(r->mb_type[mb_pos] == RV34_MB_P_MIX16x16){
973
            r->is16 = 1;
974
            r->chroma_vlc = 1;
975
            r->luma_vlc   = 2;
976
            r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
977
        }
978
    }
979

    
980
    return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
981
}
982

    
983
/**
984
 * @addtogroup recons
985
 * @{
986
 */
987
/**
988
 * mask for retrieving all bits in coded block pattern
989
 * corresponding to one 8x8 block
990
 */
991
#define LUMA_CBP_BLOCK_MASK 0x303
992

    
993
#define U_CBP_MASK 0x0F0000
994
#define V_CBP_MASK 0xF00000
995

    
996

    
997
static void rv34_apply_differences(RV34DecContext *r, int cbp)
998
{
999
    static const int shifts[4] = { 0, 2, 8, 10 };
1000
    MpegEncContext *s = &r->s;
1001
    int i;
1002

    
1003
    for(i = 0; i < 4; i++)
1004
        if(cbp & (LUMA_CBP_BLOCK_MASK << shifts[i]))
1005
            s->dsp.add_pixels_clamped(s->block[i], s->dest[0] + (i & 1)*8 + (i&2)*4*s->linesize, s->linesize);
1006
    if(cbp & U_CBP_MASK)
1007
        s->dsp.add_pixels_clamped(s->block[4], s->dest[1], s->uvlinesize);
1008
    if(cbp & V_CBP_MASK)
1009
        s->dsp.add_pixels_clamped(s->block[5], s->dest[2], s->uvlinesize);
1010
}
1011

    
1012
static int rv34_decode_macroblock(RV34DecContext *r, int8_t *intra_types)
1013
{
1014
    MpegEncContext *s = &r->s;
1015
    GetBitContext *gb = &s->gb;
1016
    int cbp, cbp2;
1017
    int i, blknum, blkoff;
1018
    DCTELEM block16[64];
1019
    int luma_dc_quant;
1020
    int dist;
1021
    int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1022

    
1023
    // Calculate which neighbours are available. Maybe it's worth optimizing too.
1024
    memset(r->avail_cache, 0, sizeof(r->avail_cache));
1025
    fill_rectangle(r->avail_cache + 5, 2, 2, 4, 1, 4);
1026
    dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
1027
    if(s->mb_x && dist)
1028
        r->avail_cache[4] =
1029
        r->avail_cache[8] = s->current_picture_ptr->mb_type[mb_pos - 1];
1030
    if(dist >= s->mb_width)
1031
        r->avail_cache[1] =
1032
        r->avail_cache[2] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride];
1033
    if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
1034
        r->avail_cache[3] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride + 1];
1035
    if(s->mb_x && dist > s->mb_width)
1036
        r->avail_cache[0] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride - 1];
1037

    
1038
    s->qscale = r->si.quant;
1039
    cbp = cbp2 = rv34_decode_mb_header(r, intra_types);
1040
    r->cbp_luma  [s->mb_x + s->mb_y * s->mb_stride] = cbp;
1041
    r->cbp_chroma[s->mb_x + s->mb_y * s->mb_stride] = cbp >> 16;
1042
    s->current_picture.qscale_table[s->mb_x + s->mb_y * s->mb_stride] = s->qscale;
1043

    
1044
    if(cbp == -1)
1045
        return -1;
1046

    
1047
    luma_dc_quant = r->si.type ? r->luma_dc_quant_p[s->qscale] : r->luma_dc_quant_i[s->qscale];
1048
    if(r->is16){
1049
        memset(block16, 0, sizeof(block16));
1050
        rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0);
1051
        rv34_dequant4x4_16x16(block16, rv34_qscale_tab[luma_dc_quant],rv34_qscale_tab[s->qscale]);
1052
        rv34_inv_transform_noround(block16);
1053
    }
1054

    
1055
    for(i = 0; i < 16; i++, cbp >>= 1){
1056
        if(!r->is16 && !(cbp & 1)) continue;
1057
        blknum = ((i & 2) >> 1) + ((i & 8) >> 2);
1058
        blkoff = ((i & 1) << 2) + ((i & 4) << 3);
1059
        if(cbp & 1)
1060
            rv34_decode_block(s->block[blknum] + blkoff, gb, r->cur_vlcs, r->luma_vlc, 0);
1061
        rv34_dequant4x4(s->block[blknum] + blkoff, rv34_qscale_tab[luma_dc_quant],rv34_qscale_tab[s->qscale]);
1062
        if(r->is16) //FIXME: optimize
1063
            s->block[blknum][blkoff] = block16[(i & 3) | ((i & 0xC) << 1)];
1064
        rv34_inv_transform(s->block[blknum] + blkoff);
1065
    }
1066
    if(r->block_type == RV34_MB_P_MIX16x16)
1067
        r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
1068
    for(; i < 24; i++, cbp >>= 1){
1069
        if(!(cbp & 1)) continue;
1070
        blknum = ((i & 4) >> 2) + 4;
1071
        blkoff = ((i & 1) << 2) + ((i & 2) << 4);
1072
        rv34_decode_block(s->block[blknum] + blkoff, gb, r->cur_vlcs, r->chroma_vlc, 1);
1073
        rv34_dequant4x4(s->block[blknum] + blkoff, rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]],rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]]);
1074
        rv34_inv_transform(s->block[blknum] + blkoff);
1075
    }
1076
    if(IS_INTRA(s->current_picture_ptr->mb_type[s->mb_x + s->mb_y*s->mb_stride]))
1077
        rv34_output_macroblock(r, intra_types, cbp2, r->is16);
1078
    else
1079
        rv34_apply_differences(r, cbp2);
1080

    
1081
    return 0;
1082
}
1083

    
1084
static int check_slice_end(RV34DecContext *r, MpegEncContext *s)
1085
{
1086
    int bits;
1087
    if(s->mb_y >= s->mb_height)
1088
        return 1;
1089
    if(!s->mb_num_left)
1090
        return 1;
1091
    if(r->s.mb_skip_run > 1)
1092
        return 0;
1093
    bits = r->bits - get_bits_count(&s->gb);
1094
    if(bits < 0 || (bits < 8 && !show_bits(&s->gb, bits)))
1095
        return 1;
1096
    return 0;
1097
}
1098

    
1099
static inline int slice_compare(SliceInfo *si1, SliceInfo *si2)
1100
{
1101
    return si1->type   != si2->type  ||
1102
           si1->start  >= si2->start ||
1103
           si1->width  != si2->width ||
1104
           si1->height != si2->height||
1105
           si1->pts    != si2->pts;
1106
}
1107

    
1108
static int rv34_decode_slice(RV34DecContext *r, int end, uint8_t* buf, int buf_size)
1109
{
1110
    MpegEncContext *s = &r->s;
1111
    GetBitContext *gb = &s->gb;
1112
    int mb_pos;
1113
    int res;
1114

    
1115
    init_get_bits(&r->s.gb, buf, buf_size*8);
1116
    res = r->parse_slice_header(r, gb, &r->si);
1117
    if(res < 0){
1118
        av_log(s->avctx, AV_LOG_ERROR, "Incorrect or unknown slice header\n");
1119
        return -1;
1120
    }
1121

    
1122
    if ((s->mb_x == 0 && s->mb_y == 0) || s->current_picture_ptr==NULL) {
1123
        if(s->width != r->si.width || s->height != r->si.height){
1124
            av_log(s->avctx, AV_LOG_DEBUG, "Changing dimensions to %dx%d\n", r->si.width,r->si.height);
1125
            MPV_common_end(s);
1126
            s->width  = r->si.width;
1127
            s->height = r->si.height;
1128
            if(MPV_common_init(s) < 0)
1129
                return -1;
1130
            r->intra_types_hist = av_realloc(r->intra_types_hist, s->b4_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1131
            r->intra_types = r->intra_types_hist + s->b4_stride * 4;
1132
            r->mb_type = av_realloc(r->mb_type, r->s.mb_stride * r->s.mb_height * sizeof(*r->mb_type));
1133
            r->cbp_luma   = av_realloc(r->cbp_luma,   r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_luma));
1134
            r->cbp_chroma = av_realloc(r->cbp_chroma, r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_chroma));
1135
        }
1136
        s->pict_type = r->si.type ? r->si.type : FF_I_TYPE;
1137
        if(MPV_frame_start(s, s->avctx) < 0)
1138
            return -1;
1139
        ff_er_frame_start(s);
1140
        s->current_picture_ptr = &s->current_picture;
1141
        r->cur_pts = r->si.pts;
1142
        if(s->pict_type != FF_B_TYPE){
1143
            r->last_pts = r->next_pts;
1144
            r->next_pts = r->cur_pts;
1145
        }
1146
        s->mb_x = s->mb_y = 0;
1147
    }
1148

    
1149
    r->si.end = end;
1150
    s->qscale = r->si.quant;
1151
    r->bits = buf_size*8;
1152
    s->mb_num_left = r->si.end - r->si.start;
1153
    r->s.mb_skip_run = 0;
1154

    
1155
    mb_pos = s->mb_x + s->mb_y * s->mb_width;
1156
    if(r->si.start != mb_pos){
1157
        av_log(s->avctx, AV_LOG_ERROR, "Slice indicates MB offset %d, got %d\n", r->si.start, mb_pos);
1158
        s->mb_x = r->si.start % s->mb_width;
1159
        s->mb_y = r->si.start / s->mb_width;
1160
    }
1161
    memset(r->intra_types_hist, -1, s->b4_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1162
    s->first_slice_line = 1;
1163
    s->resync_mb_x= s->mb_x;
1164
    s->resync_mb_y= s->mb_y;
1165

    
1166
    ff_init_block_index(s);
1167
    while(!check_slice_end(r, s)) {
1168
        ff_update_block_index(s);
1169
        s->dsp.clear_blocks(s->block[0]);
1170

    
1171
        if(rv34_decode_macroblock(r, r->intra_types + s->mb_x * 4 + 1) < 0){
1172
            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);
1173
            return -1;
1174
        }
1175
        if (++s->mb_x == s->mb_width) {
1176
            s->mb_x = 0;
1177
            s->mb_y++;
1178
            ff_init_block_index(s);
1179

    
1180
            memmove(r->intra_types_hist, r->intra_types, s->b4_stride * 4 * sizeof(*r->intra_types_hist));
1181
            memset(r->intra_types, -1, s->b4_stride * 4 * sizeof(*r->intra_types_hist));
1182
        }
1183
        if(s->mb_x == s->resync_mb_x)
1184
            s->first_slice_line=0;
1185
        s->mb_num_left--;
1186
    }
1187
    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);
1188

    
1189
    return (s->mb_y == s->mb_height);
1190
}
1191

    
1192
/** @} */ // recons group end
1193

    
1194
/**
1195
 * Initialize decoder.
1196
 */
1197
av_cold int ff_rv34_decode_init(AVCodecContext *avctx)
1198
{
1199
    RV34DecContext *r = avctx->priv_data;
1200
    MpegEncContext *s = &r->s;
1201

    
1202
    MPV_decode_defaults(s);
1203
    s->avctx= avctx;
1204
    s->out_format = FMT_H263;
1205
    s->codec_id= avctx->codec_id;
1206

    
1207
    s->width = avctx->width;
1208
    s->height = avctx->height;
1209

    
1210
    r->s.avctx = avctx;
1211
    avctx->flags |= CODEC_FLAG_EMU_EDGE;
1212
    r->s.flags |= CODEC_FLAG_EMU_EDGE;
1213
    avctx->pix_fmt = PIX_FMT_YUV420P;
1214
    avctx->has_b_frames = 1;
1215
    s->low_delay = 0;
1216

    
1217
    if (MPV_common_init(s) < 0)
1218
        return -1;
1219

    
1220
    ff_h264_pred_init(&r->h, CODEC_ID_RV40);
1221

    
1222
    r->intra_types_hist = av_malloc(s->b4_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1223
    r->intra_types = r->intra_types_hist + s->b4_stride * 4;
1224

    
1225
    r->mb_type = av_mallocz(r->s.mb_stride * r->s.mb_height * sizeof(*r->mb_type));
1226

    
1227
    r->cbp_luma   = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_luma));
1228
    r->cbp_chroma = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_chroma));
1229

    
1230
    if(!intra_vlcs[0].cbppattern[0].bits)
1231
        rv34_init_tables();
1232

    
1233
    return 0;
1234
}
1235

    
1236
static int get_slice_offset(AVCodecContext *avctx, uint8_t *buf, int n)
1237
{
1238
    if(avctx->slice_count) return avctx->slice_offset[n];
1239
    else                   return AV_RL32(buf + n*8 - 4) == 1 ? AV_RL32(buf + n*8) :  AV_RB32(buf + n*8);
1240
}
1241

    
1242
int ff_rv34_decode_frame(AVCodecContext *avctx,
1243
                            void *data, int *data_size,
1244
                            uint8_t *buf, int buf_size)
1245
{
1246
    RV34DecContext *r = avctx->priv_data;
1247
    MpegEncContext *s = &r->s;
1248
    AVFrame *pict = data;
1249
    SliceInfo si;
1250
    int i;
1251
    int slice_count;
1252
    uint8_t *slices_hdr = NULL;
1253
    int last = 0;
1254

    
1255
    /* no supplementary picture */
1256
    if (buf_size == 0) {
1257
        /* special case for last picture */
1258
        if (s->low_delay==0 && s->next_picture_ptr) {
1259
            *pict= *(AVFrame*)s->next_picture_ptr;
1260
            s->next_picture_ptr= NULL;
1261

    
1262
            *data_size = sizeof(AVFrame);
1263
        }
1264
        return 0;
1265
    }
1266

    
1267
    if(!avctx->slice_count){
1268
        slice_count = (*buf++) + 1;
1269
        slices_hdr = buf + 4;
1270
        buf += 8 * slice_count;
1271
    }else
1272
        slice_count = avctx->slice_count;
1273

    
1274
    for(i=0; i<slice_count; i++){
1275
        int offset= get_slice_offset(avctx, slices_hdr, i);
1276
        int size;
1277
        if(i+1 == slice_count)
1278
            size= buf_size - offset;
1279
        else
1280
            size= get_slice_offset(avctx, slices_hdr, i+1) - offset;
1281

    
1282
        r->si.end = s->mb_width * s->mb_height;
1283
        if(i+1 < slice_count){
1284
            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);
1285
            if(r->parse_slice_header(r, &r->s.gb, &si) < 0){
1286
                if(i+2 < slice_count)
1287
                    size = get_slice_offset(avctx, slices_hdr, i+2) - offset;
1288
                else
1289
                    size = buf_size - offset;
1290
            }else
1291
                r->si.end = si.start;
1292
        }
1293
        last = rv34_decode_slice(r, r->si.end, buf + offset, size);
1294
        s->mb_num_left = r->s.mb_x + r->s.mb_y*r->s.mb_width - r->si.start;
1295
        if(last)
1296
            break;
1297
    }
1298

    
1299
    if(last){
1300
        if(r->loop_filter)
1301
            r->loop_filter(r);
1302
        ff_er_frame_end(s);
1303
        MPV_frame_end(s);
1304
        if (s->pict_type == FF_B_TYPE || s->low_delay) {
1305
            *pict= *(AVFrame*)s->current_picture_ptr;
1306
        } else if (s->last_picture_ptr != NULL) {
1307
            *pict= *(AVFrame*)s->last_picture_ptr;
1308
        }
1309

    
1310
        if(s->last_picture_ptr || s->low_delay){
1311
            *data_size = sizeof(AVFrame);
1312
            ff_print_debug_info(s, pict);
1313
        }
1314
        s->current_picture_ptr= NULL; //so we can detect if frame_end wasnt called (find some nicer solution...)
1315
    }
1316
    return buf_size;
1317
}
1318

    
1319
av_cold int ff_rv34_decode_end(AVCodecContext *avctx)
1320
{
1321
    RV34DecContext *r = avctx->priv_data;
1322

    
1323
    MPV_common_end(&r->s);
1324

    
1325
    av_freep(&r->intra_types_hist);
1326
    r->intra_types = NULL;
1327
    av_freep(&r->mb_type);
1328

    
1329
    return 0;
1330
}