Statistics
| Branch: | Revision:

ffmpeg / libavcodec / rv34.c @ 199436b9

History | View | Annotate | Download (50.1 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 "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 insyms symbols for input codes (NULL for default ones)
69
 */
70
static void rv34_gen_vlc(const uint8_t *bits, int size, VLC *vlc, const uint8_t *insyms)
71
{
72
    int i;
73
    int counts[17] = {0}, codes[17];
74
    uint16_t cw[size], syms[size];
75
    uint8_t bits2[size];
76
    int maxbits = 0, realsize = 0;
77

    
78
    for(i = 0; i < size; i++){
79
        if(bits[i]){
80
            bits2[realsize] = bits[i];
81
            syms[realsize] = insyms ? insyms[i] : i;
82
            realsize++;
83
            maxbits = FFMAX(maxbits, bits[i]);
84
            counts[bits[i]]++;
85
        }
86
    }
87

    
88
    codes[0] = 0;
89
    for(i = 0; i < 16; i++)
90
        codes[i+1] = (codes[i] + counts[i]) << 1;
91
    for(i = 0; i < realsize; i++)
92
        cw[i] = codes[bits2[i]]++;
93

    
94
    init_vlc_sparse(vlc, FFMIN(maxbits, 9), realsize,
95
                    bits2, 1, 1,
96
                    cw,    2, 2,
97
                    syms,  2, 2, INIT_VLC_USE_STATIC);
98
}
99

    
100
/**
101
 * Initialize all tables.
102
 */
103
static av_cold void rv34_init_tables()
104
{
105
    int i, j, k;
106

    
107
    for(i = 0; i < NUM_INTRA_TABLES; i++){
108
        for(j = 0; j < 2; j++){
109
            rv34_gen_vlc(rv34_table_intra_cbppat   [i][j], CBPPAT_VLC_SIZE,   &intra_vlcs[i].cbppattern[j],     NULL);
110
            rv34_gen_vlc(rv34_table_intra_secondpat[i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].second_pattern[j], NULL);
111
            rv34_gen_vlc(rv34_table_intra_thirdpat [i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].third_pattern[j],  NULL);
112
            for(k = 0; k < 4; k++)
113
                rv34_gen_vlc(rv34_table_intra_cbp[i][j+k*2],  CBP_VLC_SIZE,   &intra_vlcs[i].cbp[j][k],         rv34_cbp_code);
114
        }
115
        for(j = 0; j < 4; j++)
116
            rv34_gen_vlc(rv34_table_intra_firstpat[i][j], FIRSTBLK_VLC_SIZE, &intra_vlcs[i].first_pattern[j], NULL);
117
        rv34_gen_vlc(rv34_intra_coeff[i], COEFF_VLC_SIZE, &intra_vlcs[i].coefficient, NULL);
118
    }
119

    
120
    for(i = 0; i < NUM_INTER_TABLES; i++){
121
        rv34_gen_vlc(rv34_inter_cbppat[i], CBPPAT_VLC_SIZE, &inter_vlcs[i].cbppattern[0], NULL);
122
        for(j = 0; j < 4; j++)
123
            rv34_gen_vlc(rv34_inter_cbp[i][j], CBP_VLC_SIZE, &inter_vlcs[i].cbp[0][j], rv34_cbp_code);
124
        for(j = 0; j < 2; j++){
125
            rv34_gen_vlc(rv34_table_inter_firstpat [i][j], FIRSTBLK_VLC_SIZE, &inter_vlcs[i].first_pattern[j],  NULL);
126
            rv34_gen_vlc(rv34_table_inter_secondpat[i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].second_pattern[j], NULL);
127
            rv34_gen_vlc(rv34_table_inter_thirdpat [i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].third_pattern[j],  NULL);
128
        }
129
        rv34_gen_vlc(rv34_inter_coeff[i], COEFF_VLC_SIZE, &inter_vlcs[i].coefficient, NULL);
130
    }
131
}
132

    
133
/** @} */ // vlc group
134

    
135

    
136
/**
137
 * @defgroup transform RV30/40 inverse transform functions
138
 * @{
139
 */
140

    
141
static av_always_inline void rv34_row_transform(int temp[16], DCTELEM *block)
142
{
143
    int i;
144

    
145
    for(i=0; i<4; i++){
146
        const int z0= 13*(block[i+8*0] +    block[i+8*2]);
147
        const int z1= 13*(block[i+8*0] -    block[i+8*2]);
148
        const int z2=  7* block[i+8*1] - 17*block[i+8*3];
149
        const int z3= 17* block[i+8*1] +  7*block[i+8*3];
150

    
151
        temp[4*i+0]= z0+z3;
152
        temp[4*i+1]= z1+z2;
153
        temp[4*i+2]= z1-z2;
154
        temp[4*i+3]= z0-z3;
155
    }
156
}
157

    
158
/**
159
 * Real Video 3.0/4.0 inverse transform
160
 * Code is almost the same as in SVQ3, only scaling is different.
161
 */
162
static void rv34_inv_transform(DCTELEM *block){
163
    int temp[16];
164
    int i;
165

    
166
    rv34_row_transform(temp, block);
167

    
168
    for(i=0; i<4; i++){
169
        const int z0= 13*(temp[4*0+i] +    temp[4*2+i]) + 0x200;
170
        const int z1= 13*(temp[4*0+i] -    temp[4*2+i]) + 0x200;
171
        const int z2=  7* temp[4*1+i] - 17*temp[4*3+i];
172
        const int z3= 17* temp[4*1+i] +  7*temp[4*3+i];
173

    
174
        block[i*8+0]= (z0 + z3)>>10;
175
        block[i*8+1]= (z1 + z2)>>10;
176
        block[i*8+2]= (z1 - z2)>>10;
177
        block[i*8+3]= (z0 - z3)>>10;
178
    }
179

    
180
}
181

    
182
/**
183
 * RealVideo 3.0/4.0 inverse transform for DC block
184
 *
185
 * Code is almost the same as rv34_inv_transform()
186
 * but final coefficients are multiplied by 1.5 and have no rounding.
187
 */
188
static void rv34_inv_transform_noround(DCTELEM *block){
189
    int temp[16];
190
    int i;
191

    
192
    rv34_row_transform(temp, block);
193

    
194
    for(i=0; i<4; i++){
195
        const int z0= 13*(temp[4*0+i] +    temp[4*2+i]);
196
        const int z1= 13*(temp[4*0+i] -    temp[4*2+i]);
197
        const int z2=  7* temp[4*1+i] - 17*temp[4*3+i];
198
        const int z3= 17* temp[4*1+i] +  7*temp[4*3+i];
199

    
200
        block[i*8+0]= ((z0 + z3)*3)>>11;
201
        block[i*8+1]= ((z1 + z2)*3)>>11;
202
        block[i*8+2]= ((z1 - z2)*3)>>11;
203
        block[i*8+3]= ((z0 - z3)*3)>>11;
204
    }
205

    
206
}
207

    
208
/** @} */ // transform
209

    
210

    
211
/**
212
 * @defgroup block RV30/40 4x4 block decoding functions
213
 * @{
214
 */
215

    
216
/**
217
 * Decode coded block pattern.
218
 */
219
static int rv34_decode_cbp(GetBitContext *gb, RV34VLC *vlc, int table)
220
{
221
    int pattern, code, cbp=0;
222
    int ones;
223
    static const int cbp_masks[3] = {0x100000, 0x010000, 0x110000};
224
    static const int shifts[4] = { 0, 2, 8, 10 };
225
    int *curshift = shifts;
226
    int i, t, mask;
227

    
228
    code = get_vlc2(gb, vlc->cbppattern[table].table, 9, 2);
229
    pattern = code & 0xF;
230
    code >>= 4;
231

    
232
    ones = rv34_count_ones[pattern];
233

    
234
    for(mask = 8; mask; mask >>= 1, curshift++){
235
        if(pattern & mask)
236
            cbp |= get_vlc2(gb, vlc->cbp[table][ones].table, vlc->cbp[table][ones].bits, 1) << curshift[0];
237
    }
238

    
239
    for(i = 0; i < 4; i++){
240
        t = modulo_three_table[code][i];
241
        if(t == 1)
242
            cbp |= cbp_masks[get_bits1(gb)] << i;
243
        if(t == 2)
244
            cbp |= cbp_masks[2] << i;
245
    }
246
    return cbp;
247
}
248

    
249
/**
250
 * Get one coefficient value from the bistream and store it.
251
 */
252
static inline void decode_coeff(DCTELEM *dst, int coef, int esc, GetBitContext *gb, VLC* vlc)
253
{
254
    if(coef){
255
        if(coef == esc){
256
            coef = get_vlc2(gb, vlc->table, 9, 2);
257
            if(coef > 23){
258
                coef -= 23;
259
                coef = 22 + ((1 << coef) | get_bits(gb, coef));
260
            }
261
            coef += esc;
262
        }
263
        if(get_bits1(gb))
264
            coef = -coef;
265
        *dst = coef;
266
    }
267
}
268

    
269
/**
270
 * Decode 2x2 subblock of coefficients.
271
 */
272
static inline void decode_subblock(DCTELEM *dst, int code, const int is_block2, GetBitContext *gb, VLC *vlc)
273
{
274
    int coeffs[4];
275

    
276
    coeffs[0] = modulo_three_table[code][0];
277
    coeffs[1] = modulo_three_table[code][1];
278
    coeffs[2] = modulo_three_table[code][2];
279
    coeffs[3] = modulo_three_table[code][3];
280
    decode_coeff(dst  , coeffs[0], 3, gb, vlc);
281
    if(is_block2){
282
        decode_coeff(dst+8, coeffs[1], 2, gb, vlc);
283
        decode_coeff(dst+1, coeffs[2], 2, gb, vlc);
284
    }else{
285
        decode_coeff(dst+1, coeffs[1], 2, gb, vlc);
286
        decode_coeff(dst+8, coeffs[2], 2, gb, vlc);
287
    }
288
    decode_coeff(dst+9, coeffs[3], 2, gb, vlc);
289
}
290

    
291
/**
292
 * Decode coefficients for 4x4 block.
293
 *
294
 * This is done by filling 2x2 subblocks with decoded coefficients
295
 * in this order (the same for subblocks and subblock coefficients):
296
 *  o--o
297
 *    /
298
 *   /
299
 *  o--o
300
 */
301

    
302
static inline void rv34_decode_block(DCTELEM *dst, GetBitContext *gb, RV34VLC *rvlc, int fc, int sc)
303
{
304
    int code, pattern;
305

    
306
    code = get_vlc2(gb, rvlc->first_pattern[fc].table, 9, 2);
307

    
308
    pattern = code & 0x7;
309

    
310
    code >>= 3;
311
    decode_subblock(dst, code, 0, gb, &rvlc->coefficient);
312

    
313
    if(pattern & 4){
314
        code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
315
        decode_subblock(dst + 2, code, 0, gb, &rvlc->coefficient);
316
    }
317
    if(pattern & 2){ // Looks like coefficients 1 and 2 are swapped for this block
318
        code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
319
        decode_subblock(dst + 8*2, code, 1, gb, &rvlc->coefficient);
320
    }
321
    if(pattern & 1){
322
        code = get_vlc2(gb, rvlc->third_pattern[sc].table, 9, 2);
323
        decode_subblock(dst + 8*2+2, code, 0, gb, &rvlc->coefficient);
324
    }
325

    
326
}
327

    
328
/**
329
 * Dequantize ordinary 4x4 block.
330
 * @todo optimize
331
 */
332
static inline void rv34_dequant4x4(DCTELEM *block, int Qdc, int Q)
333
{
334
    int i, j;
335

    
336
    block[0] = (block[0] * Qdc + 8) >> 4;
337
    for(i = 0; i < 4; i++)
338
        for(j = !i; j < 4; j++)
339
            block[j + i*8] = (block[j + i*8] * Q + 8) >> 4;
340
}
341

    
342
/**
343
 * Dequantize 4x4 block of DC values for 16x16 macroblock.
344
 * @todo optimize
345
 */
346
static inline void rv34_dequant4x4_16x16(DCTELEM *block, int Qdc, int Q)
347
{
348
    int i;
349

    
350
    for(i = 0; i < 3; i++)
351
         block[rv34_dezigzag[i]] = (block[rv34_dezigzag[i]] * Qdc + 8) >> 4;
352
    for(; i < 16; i++)
353
         block[rv34_dezigzag[i]] = (block[rv34_dezigzag[i]] * Q + 8) >> 4;
354
}
355
/** @} */ //block functions
356

    
357

    
358
/**
359
 * @defgroup bitstream RV30/40 bitstream parsing
360
 * @{
361
 */
362

    
363
/**
364
 * Decode starting slice position.
365
 * @todo Maybe replace with ff_h263_decode_mba() ?
366
 */
367
int ff_rv34_get_start_offset(GetBitContext *gb, int mb_size)
368
{
369
    int i;
370
    for(i = 0; i < 5; i++)
371
        if(rv34_mb_max_sizes[i] >= mb_size - 1)
372
            break;
373
    return rv34_mb_bits_sizes[i];
374
}
375

    
376
/**
377
 * Select VLC set for decoding from current quantizer, modifier and frame type.
378
 */
379
static inline RV34VLC* choose_vlc_set(int quant, int mod, int type)
380
{
381
    if(mod == 2 && quant < 19) quant += 10;
382
    else if(mod && quant < 26) quant += 5;
383
    return type ? &inter_vlcs[rv34_quant_to_vlc_set[1][av_clip(quant, 0, 30)]]
384
                : &intra_vlcs[rv34_quant_to_vlc_set[0][av_clip(quant, 0, 30)]];
385
}
386

    
387
/**
388
 * Decode quantizer difference and return modified quantizer.
389
 */
390
static inline int rv34_decode_dquant(GetBitContext *gb, int quant)
391
{
392
    if(get_bits1(gb))
393
        return rv34_dquant_tab[get_bits1(gb)][quant];
394
    else
395
        return get_bits(gb, 5);
396
}
397

    
398
/** @} */ //bitstream functions
399

    
400
/**
401
 * @defgroup mv motion vector related code (prediction, reconstruction, motion compensation)
402
 * @{
403
 */
404

    
405
/** macroblock partition width in 8x8 blocks */
406
static const uint8_t part_sizes_w[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 2, 1, 2, 2 };
407

    
408
/** macroblock partition height in 8x8 blocks */
409
static const uint8_t part_sizes_h[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 1, 2, 2, 2 };
410

    
411
/** availability index for subblocks */
412
static const uint8_t avail_indexes[4] = { 5, 6, 9, 10 };
413

    
414
/**
415
 * motion vector prediction
416
 *
417
 * Motion prediction performed for the block by using median prediction of
418
 * motion vectors from the left, top and right top blocks but in corner cases
419
 * some other vectors may be used instead.
420
 */
421
static void rv34_pred_mv(RV34DecContext *r, int block_type, int subblock_no, int dmv_no)
422
{
423
    MpegEncContext *s = &r->s;
424
    int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
425
    int A[2] = {0}, B[2], C[2];
426
    int i, j;
427
    int mx, my;
428
    int avail_index = avail_indexes[subblock_no];
429
    int c_off = part_sizes_w[block_type];
430

    
431
    mv_pos += (subblock_no & 1) + (subblock_no >> 1)*s->b8_stride;
432
    if(subblock_no == 3)
433
        c_off = -1;
434

    
435
    if(r->avail_cache[avail_index - 1]){
436
        A[0] = s->current_picture_ptr->motion_val[0][mv_pos-1][0];
437
        A[1] = s->current_picture_ptr->motion_val[0][mv_pos-1][1];
438
    }
439
    if(r->avail_cache[avail_index - 4]){
440
        B[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][0];
441
        B[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][1];
442
    }else{
443
        B[0] = A[0];
444
        B[1] = A[1];
445
    }
446
    if(!r->avail_cache[avail_index - 4 + c_off]){
447
        if(r->avail_cache[avail_index - 4] && (r->avail_cache[avail_index - 1] || r->rv30)){
448
            C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][0];
449
            C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][1];
450
        }else{
451
            C[0] = A[0];
452
            C[1] = A[1];
453
        }
454
    }else{
455
        C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][0];
456
        C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][1];
457
    }
458
    mx = mid_pred(A[0], B[0], C[0]);
459
    my = mid_pred(A[1], B[1], C[1]);
460
    mx += r->dmv[dmv_no][0];
461
    my += r->dmv[dmv_no][1];
462
    for(j = 0; j < part_sizes_h[block_type]; j++){
463
        for(i = 0; i < part_sizes_w[block_type]; i++){
464
            s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][0] = mx;
465
            s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][1] = my;
466
        }
467
    }
468
}
469

    
470
#define GET_PTS_DIFF(a, b) ((a - b + 8192) & 0x1FFF)
471

    
472
/**
473
 * Calculate motion vector component that should be added for direct blocks.
474
 */
475
static int calc_add_mv(RV34DecContext *r, int dir, int val)
476
{
477
    int refdist = GET_PTS_DIFF(r->next_pts, r->last_pts);
478
    int dist = dir ? -GET_PTS_DIFF(r->next_pts, r->cur_pts) : GET_PTS_DIFF(r->cur_pts, r->last_pts);
479
    int mul;
480

    
481
    if(!refdist) return 0;
482
    mul = (dist << 14) / refdist;
483
    return (val * mul + 0x2000) >> 14;
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
/**
561
 * motion vector prediction - RV3 version
562
 */
563
static void rv34_pred_mv_rv3(RV34DecContext *r, int block_type, int dir)
564
{
565
    MpegEncContext *s = &r->s;
566
    int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
567
    int A[2] = {0}, B[2], C[2];
568
    int i, j, k;
569
    int mx, my;
570
    int avail_index = avail_indexes[0];
571

    
572
    if(r->avail_cache[avail_index - 1]){
573
        A[0] = s->current_picture_ptr->motion_val[0][mv_pos-1][0];
574
        A[1] = s->current_picture_ptr->motion_val[0][mv_pos-1][1];
575
    }
576
    if(r->avail_cache[avail_index - 4]){
577
        B[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][0];
578
        B[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][1];
579
    }else{
580
        B[0] = A[0];
581
        B[1] = A[1];
582
    }
583
    if(!r->avail_cache[avail_index - 4 + 2]){
584
        if(r->avail_cache[avail_index - 4] && (r->avail_cache[avail_index - 1])){
585
            C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][0];
586
            C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][1];
587
        }else{
588
            C[0] = A[0];
589
            C[1] = A[1];
590
        }
591
    }else{
592
        C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+2][0];
593
        C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+2][1];
594
    }
595
    mx = mid_pred(A[0], B[0], C[0]);
596
    my = mid_pred(A[1], B[1], C[1]);
597
    mx += r->dmv[0][0];
598
    my += r->dmv[0][1];
599
    for(j = 0; j < 2; j++){
600
        for(i = 0; i < 2; i++){
601
            for(k = 0; k < 2; k++){
602
                s->current_picture_ptr->motion_val[k][mv_pos + i + j*s->b8_stride][0] = mx;
603
                s->current_picture_ptr->motion_val[k][mv_pos + i + j*s->b8_stride][1] = my;
604
            }
605
        }
606
    }
607
}
608

    
609
static const int chroma_coeffs[3] = { 0, 3, 5 };
610

    
611
/**
612
 * generic motion compensation function
613
 *
614
 * @param r decoder context
615
 * @param block_type type of the current block
616
 * @param xoff horizontal offset from the start of the current block
617
 * @param yoff vertical offset from the start of the current block
618
 * @param mv_off offset to the motion vector information
619
 * @param width width of the current partition in 8x8 blocks
620
 * @param height height of the current partition in 8x8 blocks
621
 */
622
static inline void rv34_mc(RV34DecContext *r, const int block_type,
623
                          const int xoff, const int yoff, int mv_off,
624
                          const int width, const int height, int dir,
625
                          const int thirdpel,
626
                          qpel_mc_func (*qpel_mc)[16],
627
                          h264_chroma_mc_func (*chroma_mc))
628
{
629
    MpegEncContext *s = &r->s;
630
    uint8_t *Y, *U, *V, *srcY, *srcU, *srcV;
631
    int dxy, mx, my, umx, umy, lx, ly, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
632
    int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride + mv_off;
633
    int is16x16 = 1;
634

    
635
    if(thirdpel){
636
        int chroma_mx, chroma_my;
637
        mx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) / 3 - (1 << 24);
638
        my = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) / 3 - (1 << 24);
639
        lx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) % 3;
640
        ly = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) % 3;
641
        chroma_mx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + 1) >> 1;
642
        chroma_my = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + 1) >> 1;
643
        umx = (chroma_mx + (3 << 24)) / 3 - (1 << 24);
644
        umy = (chroma_my + (3 << 24)) / 3 - (1 << 24);
645
        uvmx = chroma_coeffs[(chroma_mx + (3 << 24)) % 3];
646
        uvmy = chroma_coeffs[(chroma_my + (3 << 24)) % 3];
647
    }else{
648
        int cx, cy;
649
        mx = s->current_picture_ptr->motion_val[dir][mv_pos][0] >> 2;
650
        my = s->current_picture_ptr->motion_val[dir][mv_pos][1] >> 2;
651
        lx = s->current_picture_ptr->motion_val[dir][mv_pos][0] & 3;
652
        ly = s->current_picture_ptr->motion_val[dir][mv_pos][1] & 3;
653
        cx = s->current_picture_ptr->motion_val[dir][mv_pos][0] / 2;
654
        cy = s->current_picture_ptr->motion_val[dir][mv_pos][1] / 2;
655
        umx = cx >> 2;
656
        umy = cy >> 2;
657
        uvmx = (cx & 3) << 1;
658
        uvmy = (cy & 3) << 1;
659
        //due to some flaw RV40 uses the same MC compensation routine for H2V2 and H3V3
660
        if(uvmx == 6 && uvmy == 6)
661
            uvmx = uvmy = 4;
662
    }
663
    dxy = ly*4 + lx;
664
    srcY = dir ? s->next_picture_ptr->data[0] : s->last_picture_ptr->data[0];
665
    srcU = dir ? s->next_picture_ptr->data[1] : s->last_picture_ptr->data[1];
666
    srcV = dir ? s->next_picture_ptr->data[2] : s->last_picture_ptr->data[2];
667
    src_x = s->mb_x * 16 + xoff + mx;
668
    src_y = s->mb_y * 16 + yoff + my;
669
    uvsrc_x = s->mb_x * 8 + (xoff >> 1) + umx;
670
    uvsrc_y = s->mb_y * 8 + (yoff >> 1) + umy;
671
    srcY += src_y * s->linesize + src_x;
672
    srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
673
    srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
674
    if(   (unsigned)(src_x - !!lx*2) > s->h_edge_pos - !!lx*2 - (width <<3) - 4
675
       || (unsigned)(src_y - !!ly*2) > s->v_edge_pos - !!ly*2 - (height<<3) - 4){
676
        uint8_t *uvbuf= s->edge_emu_buffer + 22 * s->linesize;
677

    
678
        srcY -= 2 + 2*s->linesize;
679
        ff_emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, (width<<3)+6, (height<<3)+6,
680
                            src_x - 2, src_y - 2, s->h_edge_pos, s->v_edge_pos);
681
        srcY = s->edge_emu_buffer + 2 + 2*s->linesize;
682
        ff_emulated_edge_mc(uvbuf     , srcU, s->uvlinesize, (width<<2)+1, (height<<2)+1,
683
                            uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
684
        ff_emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, (width<<2)+1, (height<<2)+1,
685
                            uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
686
        srcU = uvbuf;
687
        srcV = uvbuf + 16;
688
    }
689
    Y = s->dest[0] + xoff      + yoff     *s->linesize;
690
    U = s->dest[1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
691
    V = s->dest[2] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
692

    
693
    if(block_type == RV34_MB_P_16x8){
694
        qpel_mc[1][dxy](Y, srcY, s->linesize);
695
        Y    += 8;
696
        srcY += 8;
697
    }else if(block_type == RV34_MB_P_8x16){
698
        qpel_mc[1][dxy](Y, srcY, s->linesize);
699
        Y    += 8 * s->linesize;
700
        srcY += 8 * s->linesize;
701
    }
702
    is16x16 = (block_type != RV34_MB_P_8x8) && (block_type != RV34_MB_P_16x8) && (block_type != RV34_MB_P_8x16);
703
    qpel_mc[!is16x16][dxy](Y, srcY, s->linesize);
704
    chroma_mc[2-width]   (U, srcU, s->uvlinesize, height*4, uvmx, uvmy);
705
    chroma_mc[2-width]   (V, srcV, s->uvlinesize, height*4, uvmx, uvmy);
706
}
707

    
708
static void rv34_mc_1mv(RV34DecContext *r, const int block_type,
709
                        const int xoff, const int yoff, int mv_off,
710
                        const int width, const int height, int dir)
711
{
712
    rv34_mc(r, block_type, xoff, yoff, mv_off, width, height, dir, r->rv30,
713
            r->rv30 ? r->s.dsp.put_rv30_tpel_pixels_tab
714
                    : r->s.dsp.put_rv40_qpel_pixels_tab,
715
            r->rv30 ? r->s.dsp.put_h264_chroma_pixels_tab
716
                    : r->s.dsp.put_rv40_chroma_pixels_tab);
717
}
718

    
719
static void rv34_mc_2mv(RV34DecContext *r, const int block_type)
720
{
721
    rv34_mc(r, block_type, 0, 0, 0, 2, 2, 0, r->rv30,
722
            r->rv30 ? r->s.dsp.put_rv30_tpel_pixels_tab
723
                    : r->s.dsp.put_rv40_qpel_pixels_tab,
724
            r->rv30 ? r->s.dsp.put_h264_chroma_pixels_tab
725
                    : r->s.dsp.put_rv40_chroma_pixels_tab);
726
    rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30,
727
            r->rv30 ? r->s.dsp.avg_rv30_tpel_pixels_tab
728
                    : r->s.dsp.avg_rv40_qpel_pixels_tab,
729
            r->rv30 ? r->s.dsp.avg_h264_chroma_pixels_tab
730
                    : r->s.dsp.avg_rv40_chroma_pixels_tab);
731
}
732

    
733
static void rv34_mc_2mv_skip(RV34DecContext *r)
734
{
735
    int i, j;
736
    for(j = 0; j < 2; j++)
737
        for(i = 0; i < 2; i++){
738
             rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 0, r->rv30,
739
                    r->rv30 ? r->s.dsp.put_rv30_tpel_pixels_tab
740
                            : r->s.dsp.put_rv40_qpel_pixels_tab,
741
                    r->rv30 ? r->s.dsp.put_h264_chroma_pixels_tab
742
                            : r->s.dsp.put_rv40_chroma_pixels_tab);
743
             rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 1, r->rv30,
744
                    r->rv30 ? r->s.dsp.avg_rv30_tpel_pixels_tab
745
                            : r->s.dsp.avg_rv40_qpel_pixels_tab,
746
                    r->rv30 ? r->s.dsp.avg_h264_chroma_pixels_tab
747
                            : r->s.dsp.avg_rv40_chroma_pixels_tab);
748
        }
749
}
750

    
751
/** number of motion vectors in each macroblock type */
752
static const int num_mvs[RV34_MB_TYPES] = { 0, 0, 1, 4, 1, 1, 0, 0, 2, 2, 2, 1 };
753

    
754
/**
755
 * Decode motion vector differences
756
 * and perform motion vector reconstruction and motion compensation.
757
 */
758
static int rv34_decode_mv(RV34DecContext *r, int block_type)
759
{
760
    MpegEncContext *s = &r->s;
761
    GetBitContext *gb = &s->gb;
762
    int i, j, k, l;
763
    int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
764
    int next_bt;
765

    
766
    memset(r->dmv, 0, sizeof(r->dmv));
767
    for(i = 0; i < num_mvs[block_type]; i++){
768
        r->dmv[i][0] = svq3_get_se_golomb(gb);
769
        r->dmv[i][1] = svq3_get_se_golomb(gb);
770
    }
771
    switch(block_type){
772
    case RV34_MB_TYPE_INTRA:
773
    case RV34_MB_TYPE_INTRA16x16:
774
        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);
775
        return 0;
776
    case RV34_MB_SKIP:
777
        if(s->pict_type == FF_P_TYPE){
778
            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);
779
            rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
780
            break;
781
        }
782
    case RV34_MB_B_DIRECT:
783
        //surprisingly, it uses motion scheme from next reference frame
784
        next_bt = s->next_picture_ptr->mb_type[s->mb_x + s->mb_y * s->mb_stride];
785
        for(j = 0; j < 2; j++)
786
            for(i = 0; i < 2; i++)
787
                for(k = 0; k < 2; k++)
788
                    for(l = 0; l < 2; l++)
789
                        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]);
790
        if(IS_16X16(next_bt)) //we can use whole macroblock MC
791
            rv34_mc_2mv(r, block_type);
792
        else
793
            rv34_mc_2mv_skip(r);
794
        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);
795
        break;
796
    case RV34_MB_P_16x16:
797
    case RV34_MB_P_MIX16x16:
798
        rv34_pred_mv(r, block_type, 0, 0);
799
        rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
800
        break;
801
    case RV34_MB_B_FORWARD:
802
    case RV34_MB_B_BACKWARD:
803
        r->dmv[1][0] = r->dmv[0][0];
804
        r->dmv[1][1] = r->dmv[0][1];
805
        if(r->rv30)
806
            rv34_pred_mv_rv3(r, block_type, block_type == RV34_MB_B_BACKWARD);
807
        else
808
            rv34_pred_mv_b  (r, block_type, block_type == RV34_MB_B_BACKWARD);
809
        rv34_mc_1mv     (r, block_type, 0, 0, 0, 2, 2, block_type == RV34_MB_B_BACKWARD);
810
        break;
811
    case RV34_MB_P_16x8:
812
    case RV34_MB_P_8x16:
813
        rv34_pred_mv(r, block_type, 0, 0);
814
        rv34_pred_mv(r, block_type, 1 + (block_type == RV34_MB_P_16x8), 1);
815
        if(block_type == RV34_MB_P_16x8){
816
            rv34_mc_1mv(r, block_type, 0, 0, 0,            2, 1, 0);
817
            rv34_mc_1mv(r, block_type, 0, 8, s->b8_stride, 2, 1, 0);
818
        }
819
        if(block_type == RV34_MB_P_8x16){
820
            rv34_mc_1mv(r, block_type, 0, 0, 0, 1, 2, 0);
821
            rv34_mc_1mv(r, block_type, 8, 0, 1, 1, 2, 0);
822
        }
823
        break;
824
    case RV34_MB_B_BIDIR:
825
        rv34_pred_mv_b  (r, block_type, 0);
826
        rv34_pred_mv_b  (r, block_type, 1);
827
        rv34_mc_2mv     (r, block_type);
828
        break;
829
    case RV34_MB_P_8x8:
830
        for(i=0;i< 4;i++){
831
            rv34_pred_mv(r, block_type, i, i);
832
            rv34_mc_1mv (r, block_type, (i&1)<<3, (i&2)<<2, (i&1)+(i>>1)*s->b8_stride, 1, 1, 0);
833
        }
834
        break;
835
    }
836

    
837
    return 0;
838
}
839
/** @} */ // mv group
840

    
841
/**
842
 * @defgroup recons Macroblock reconstruction functions
843
 * @{
844
 */
845
/** mapping of RV30/40 intra prediction types to standard H.264 types */
846
static const int ittrans[9] = {
847
 DC_PRED, VERT_PRED, HOR_PRED, DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_LEFT_PRED,
848
 VERT_RIGHT_PRED, VERT_LEFT_PRED, HOR_UP_PRED, HOR_DOWN_PRED,
849
};
850

    
851
/** mapping of RV30/40 intra 16x16 prediction types to standard H.264 types */
852
static const int ittrans16[4] = {
853
 DC_PRED8x8, VERT_PRED8x8, HOR_PRED8x8, PLANE_PRED8x8,
854
};
855

    
856
/**
857
 * Perform 4x4 intra prediction.
858
 */
859
static void rv34_pred_4x4_block(RV34DecContext *r, uint8_t *dst, int stride, int itype, int up, int left, int down, int right)
860
{
861
    uint8_t *prev = dst - stride + 4;
862
    uint32_t topleft;
863

    
864
    if(!up && !left)
865
        itype = DC_128_PRED;
866
    else if(!up){
867
        if(itype == VERT_PRED) itype = HOR_PRED;
868
        if(itype == DC_PRED)   itype = LEFT_DC_PRED;
869
    }else if(!left){
870
        if(itype == HOR_PRED)  itype = VERT_PRED;
871
        if(itype == DC_PRED)   itype = TOP_DC_PRED;
872
        if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
873
    }
874
    if(!down){
875
        if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
876
        if(itype == HOR_UP_PRED) itype = HOR_UP_PRED_RV40_NODOWN;
877
        if(itype == VERT_LEFT_PRED) itype = VERT_LEFT_PRED_RV40_NODOWN;
878
    }
879
    if(!right && up){
880
        topleft = dst[-stride + 3] * 0x01010101;
881
        prev = &topleft;
882
    }
883
    r->h.pred4x4[itype](dst, prev, stride);
884
}
885

    
886
/** add_pixels_clamped for 4x4 block */
887
static void rv34_add_4x4_block(uint8_t *dst, int stride, DCTELEM block[64], int off)
888
{
889
    int x, y;
890
    for(y = 0; y < 4; y++)
891
        for(x = 0; x < 4; x++)
892
            dst[x + y*stride] = av_clip_uint8(dst[x + y*stride] + block[off + x+y*8]);
893
}
894

    
895
static inline int adjust_pred16(int itype, int up, int left)
896
{
897
    if(!up && !left)
898
        itype = DC_128_PRED8x8;
899
    else if(!up){
900
        if(itype == PLANE_PRED8x8)itype = HOR_PRED8x8;
901
        if(itype == VERT_PRED8x8) itype = HOR_PRED8x8;
902
        if(itype == DC_PRED8x8)   itype = LEFT_DC_PRED8x8;
903
    }else if(!left){
904
        if(itype == PLANE_PRED8x8)itype = VERT_PRED8x8;
905
        if(itype == HOR_PRED8x8)  itype = VERT_PRED8x8;
906
        if(itype == DC_PRED8x8)   itype = TOP_DC_PRED8x8;
907
    }
908
    return itype;
909
}
910

    
911
static void rv34_output_macroblock(RV34DecContext *r, int8_t *intra_types, int cbp, int is16)
912
{
913
    MpegEncContext *s = &r->s;
914
    DSPContext *dsp = &s->dsp;
915
    int i, j;
916
    uint8_t *Y, *U, *V;
917
    int itype;
918
    int avail[6*8] = {0};
919
    int idx;
920

    
921
    // Set neighbour information.
922
    if(r->avail_cache[0])
923
        avail[0] = 1;
924
    if(r->avail_cache[1])
925
        avail[1] = avail[2] = 1;
926
    if(r->avail_cache[2])
927
        avail[3] = avail[4] = 1;
928
    if(r->avail_cache[3])
929
        avail[5] = 1;
930
    if(r->avail_cache[4])
931
        avail[8] = avail[16] = 1;
932
    if(r->avail_cache[8])
933
        avail[24] = avail[32] = 1;
934

    
935
    Y = s->dest[0];
936
    U = s->dest[1];
937
    V = s->dest[2];
938
    if(!is16){
939
        for(j = 0; j < 4; j++){
940
            idx = 9 + j*8;
941
            for(i = 0; i < 4; i++, cbp >>= 1, Y += 4, idx++){
942
                rv34_pred_4x4_block(r, Y, s->linesize, ittrans[intra_types[i]], avail[idx-8], avail[idx-1], avail[idx+7], avail[idx-7]);
943
                avail[idx] = 1;
944
                if(cbp & 1)
945
                    rv34_add_4x4_block(Y, s->linesize, s->block[(i>>1)+(j&2)], (i&1)*4+(j&1)*32);
946
            }
947
            Y += s->linesize * 4 - 4*4;
948
            intra_types += s->b4_stride;
949
        }
950
        intra_types -= s->b4_stride * 4;
951
        fill_rectangle(r->avail_cache + 5, 2, 2, 4, 0, 4);
952
        for(j = 0; j < 2; j++){
953
            idx = 5 + j*4;
954
            for(i = 0; i < 2; i++, cbp >>= 1, idx++){
955
                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]);
956
                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]);
957
                r->avail_cache[idx] = 1;
958
                if(cbp & 0x01)
959
                    rv34_add_4x4_block(U + i*4 + j*4*s->uvlinesize, s->uvlinesize, s->block[4], i*4+j*32);
960
                if(cbp & 0x10)
961
                    rv34_add_4x4_block(V + i*4 + j*4*s->uvlinesize, s->uvlinesize, s->block[5], i*4+j*32);
962
            }
963
        }
964
    }else{
965
        itype = ittrans16[intra_types[0]];
966
        itype = adjust_pred16(itype, r->avail_cache[5-4], r->avail_cache[5-1]);
967
        r->h.pred16x16[itype](Y, s->linesize);
968
        dsp->add_pixels_clamped(s->block[0], Y,     s->linesize);
969
        dsp->add_pixels_clamped(s->block[1], Y + 8, s->linesize);
970
        Y += s->linesize * 8;
971
        dsp->add_pixels_clamped(s->block[2], Y,     s->linesize);
972
        dsp->add_pixels_clamped(s->block[3], Y + 8, s->linesize);
973

    
974
        itype = ittrans16[intra_types[0]];
975
        if(itype == PLANE_PRED8x8) itype = DC_PRED8x8;
976
        itype = adjust_pred16(itype, r->avail_cache[5-4], r->avail_cache[5-1]);
977
        r->h.pred8x8[itype](U, s->uvlinesize);
978
        dsp->add_pixels_clamped(s->block[4], U, s->uvlinesize);
979
        r->h.pred8x8[itype](V, s->uvlinesize);
980
        dsp->add_pixels_clamped(s->block[5], V, s->uvlinesize);
981
    }
982
}
983

    
984
/** @} */ // recons group
985

    
986
/**
987
 * @addtogroup bitstream
988
 * Decode macroblock header and return CBP in case of success, -1 otherwise.
989
 */
990
static int rv34_decode_mb_header(RV34DecContext *r, int8_t *intra_types)
991
{
992
    MpegEncContext *s = &r->s;
993
    GetBitContext *gb = &s->gb;
994
    int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
995
    int i, t;
996

    
997
    if(!r->si.type){
998
        r->is16 = get_bits1(gb);
999
        if(!r->is16 && !r->rv30){
1000
            if(!get_bits1(gb))
1001
                av_log(s->avctx, AV_LOG_ERROR, "Need DQUANT\n");
1002
        }
1003
        s->current_picture_ptr->mb_type[mb_pos] = r->is16 ? MB_TYPE_INTRA16x16 : MB_TYPE_INTRA;
1004
        r->block_type = r->is16 ? RV34_MB_TYPE_INTRA16x16 : RV34_MB_TYPE_INTRA;
1005
    }else{
1006
        r->block_type = r->decode_mb_info(r);
1007
        if(r->block_type == -1)
1008
            return -1;
1009
        s->current_picture_ptr->mb_type[mb_pos] = rv34_mb_type_to_lavc[r->block_type];
1010
        r->mb_type[mb_pos] = r->block_type;
1011
        if(r->block_type == RV34_MB_SKIP){
1012
            if(s->pict_type == FF_P_TYPE)
1013
                r->mb_type[mb_pos] = RV34_MB_P_16x16;
1014
            if(s->pict_type == FF_B_TYPE)
1015
                r->mb_type[mb_pos] = RV34_MB_B_DIRECT;
1016
        }
1017
        r->is16 = !!IS_INTRA16x16(s->current_picture_ptr->mb_type[mb_pos]);
1018
        rv34_decode_mv(r, r->block_type);
1019
        if(r->block_type == RV34_MB_SKIP){
1020
            fill_rectangle(intra_types, 4, 4, s->b4_stride, 0, sizeof(intra_types[0]));
1021
            return 0;
1022
        }
1023
        r->chroma_vlc = 1;
1024
        r->luma_vlc   = 0;
1025
    }
1026
    if(IS_INTRA(s->current_picture_ptr->mb_type[mb_pos])){
1027
        if(r->is16){
1028
            t = get_bits(gb, 2);
1029
            fill_rectangle(intra_types, 4, 4, s->b4_stride, t, sizeof(intra_types[0]));
1030
            r->luma_vlc   = 2;
1031
        }else{
1032
            if(r->decode_intra_types(r, gb, intra_types) < 0)
1033
                return -1;
1034
            r->luma_vlc   = 1;
1035
        }
1036
        r->chroma_vlc = 0;
1037
        r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
1038
    }else{
1039
        for(i = 0; i < 16; i++)
1040
            intra_types[(i & 3) + (i>>2) * s->b4_stride] = 0;
1041
        r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
1042
        if(r->mb_type[mb_pos] == RV34_MB_P_MIX16x16){
1043
            r->is16 = 1;
1044
            r->chroma_vlc = 1;
1045
            r->luma_vlc   = 2;
1046
            r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
1047
        }
1048
    }
1049

    
1050
    return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
1051
}
1052

    
1053
/**
1054
 * @addtogroup recons
1055
 * @{
1056
 */
1057
/**
1058
 * mask for retrieving all bits in coded block pattern
1059
 * corresponding to one 8x8 block
1060
 */
1061
#define LUMA_CBP_BLOCK_MASK 0x33
1062

    
1063
#define U_CBP_MASK 0x0F0000
1064
#define V_CBP_MASK 0xF00000
1065

    
1066

    
1067
static void rv34_apply_differences(RV34DecContext *r, int cbp)
1068
{
1069
    static const int shifts[4] = { 0, 2, 8, 10 };
1070
    MpegEncContext *s = &r->s;
1071
    int i;
1072

    
1073
    for(i = 0; i < 4; i++)
1074
        if((cbp & (LUMA_CBP_BLOCK_MASK << shifts[i])) || r->block_type == RV34_MB_P_MIX16x16)
1075
            s->dsp.add_pixels_clamped(s->block[i], s->dest[0] + (i & 1)*8 + (i&2)*4*s->linesize, s->linesize);
1076
    if(cbp & U_CBP_MASK)
1077
        s->dsp.add_pixels_clamped(s->block[4], s->dest[1], s->uvlinesize);
1078
    if(cbp & V_CBP_MASK)
1079
        s->dsp.add_pixels_clamped(s->block[5], s->dest[2], s->uvlinesize);
1080
}
1081

    
1082
static int is_mv_diff_gt_3(int16_t (*motion_val)[2], int step)
1083
{
1084
    int d;
1085
    d = motion_val[0][0] - motion_val[-step][0];
1086
    if(d < -3 || d > 3)
1087
        return 1;
1088
    d = motion_val[0][1] - motion_val[-step][1];
1089
    if(d < -3 || d > 3)
1090
        return 1;
1091
    return 0;
1092
}
1093

    
1094
static int rv34_set_deblock_coef(RV34DecContext *r)
1095
{
1096
    MpegEncContext *s = &r->s;
1097
    int hmvmask = 0, vmvmask = 0, i, j;
1098
    int midx = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
1099
    int16_t (*motion_val)[2] = s->current_picture_ptr->motion_val[0][midx];
1100
    for(j = 0; j < 16; j += 8){
1101
        for(i = 0; i < 2; i++){
1102
            if(is_mv_diff_gt_3(motion_val + i, 1))
1103
                vmvmask |= 0x11 << (j + i*2);
1104
            if((j || s->mb_y) && is_mv_diff_gt_3(motion_val + i, s->b8_stride))
1105
                hmvmask |= 0x03 << (j + i*2);
1106
        }
1107
        motion_val += s->b8_stride;
1108
    }
1109
    if(s->first_slice_line)
1110
        hmvmask &= ~0x000F;
1111
    if(!s->mb_x)
1112
        vmvmask &= ~0x1111;
1113
    if(r->rv30){ //RV30 marks both subblocks on the edge for filtering
1114
        vmvmask |= (vmvmask & 0x4444) >> 1;
1115
        hmvmask |= (hmvmask & 0x0F00) >> 4;
1116
        if(s->mb_x)
1117
            r->deblock_coefs[s->mb_x - 1 + s->mb_y*s->mb_stride] |= (vmvmask & 0x1111) << 3;
1118
        if(!s->first_slice_line)
1119
            r->deblock_coefs[s->mb_x + (s->mb_y - 1)*s->mb_stride] |= (hmvmask & 0xF) << 12;
1120
    }
1121
    return hmvmask | vmvmask;
1122
}
1123

    
1124
static int rv34_decode_macroblock(RV34DecContext *r, int8_t *intra_types)
1125
{
1126
    MpegEncContext *s = &r->s;
1127
    GetBitContext *gb = &s->gb;
1128
    int cbp, cbp2;
1129
    int i, blknum, blkoff;
1130
    DCTELEM block16[64];
1131
    int luma_dc_quant;
1132
    int dist;
1133
    int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1134

    
1135
    // Calculate which neighbours are available. Maybe it's worth optimizing too.
1136
    memset(r->avail_cache, 0, sizeof(r->avail_cache));
1137
    fill_rectangle(r->avail_cache + 5, 2, 2, 4, 1, 4);
1138
    dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
1139
    if(s->mb_x && dist)
1140
        r->avail_cache[4] =
1141
        r->avail_cache[8] = s->current_picture_ptr->mb_type[mb_pos - 1];
1142
    if(dist >= s->mb_width)
1143
        r->avail_cache[1] =
1144
        r->avail_cache[2] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride];
1145
    if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
1146
        r->avail_cache[3] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride + 1];
1147
    if(s->mb_x && dist > s->mb_width)
1148
        r->avail_cache[0] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride - 1];
1149

    
1150
    s->qscale = r->si.quant;
1151
    cbp = cbp2 = rv34_decode_mb_header(r, intra_types);
1152
    r->cbp_luma  [mb_pos] = cbp;
1153
    r->cbp_chroma[mb_pos] = cbp >> 16;
1154
    if(s->pict_type == FF_I_TYPE)
1155
        r->deblock_coefs[mb_pos] = 0xFFFF;
1156
    else
1157
        r->deblock_coefs[mb_pos] = rv34_set_deblock_coef(r) | r->cbp_luma[mb_pos];
1158
    s->current_picture_ptr->qscale_table[mb_pos] = s->qscale;
1159

    
1160
    if(cbp == -1)
1161
        return -1;
1162

    
1163
    luma_dc_quant = r->block_type == RV34_MB_P_MIX16x16 ? r->luma_dc_quant_p[s->qscale] : r->luma_dc_quant_i[s->qscale];
1164
    if(r->is16){
1165
        memset(block16, 0, sizeof(block16));
1166
        rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0);
1167
        rv34_dequant4x4_16x16(block16, rv34_qscale_tab[luma_dc_quant],rv34_qscale_tab[s->qscale]);
1168
        rv34_inv_transform_noround(block16);
1169
    }
1170

    
1171
    for(i = 0; i < 16; i++, cbp >>= 1){
1172
        if(!r->is16 && !(cbp & 1)) continue;
1173
        blknum = ((i & 2) >> 1) + ((i & 8) >> 2);
1174
        blkoff = ((i & 1) << 2) + ((i & 4) << 3);
1175
        if(cbp & 1)
1176
            rv34_decode_block(s->block[blknum] + blkoff, gb, r->cur_vlcs, r->luma_vlc, 0);
1177
        rv34_dequant4x4(s->block[blknum] + blkoff, rv34_qscale_tab[s->qscale],rv34_qscale_tab[s->qscale]);
1178
        if(r->is16) //FIXME: optimize
1179
            s->block[blknum][blkoff] = block16[(i & 3) | ((i & 0xC) << 1)];
1180
        rv34_inv_transform(s->block[blknum] + blkoff);
1181
    }
1182
    if(r->block_type == RV34_MB_P_MIX16x16)
1183
        r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
1184
    for(; i < 24; i++, cbp >>= 1){
1185
        if(!(cbp & 1)) continue;
1186
        blknum = ((i & 4) >> 2) + 4;
1187
        blkoff = ((i & 1) << 2) + ((i & 2) << 4);
1188
        rv34_decode_block(s->block[blknum] + blkoff, gb, r->cur_vlcs, r->chroma_vlc, 1);
1189
        rv34_dequant4x4(s->block[blknum] + blkoff, rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]],rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]]);
1190
        rv34_inv_transform(s->block[blknum] + blkoff);
1191
    }
1192
    if(IS_INTRA(s->current_picture_ptr->mb_type[mb_pos]))
1193
        rv34_output_macroblock(r, intra_types, cbp2, r->is16);
1194
    else
1195
        rv34_apply_differences(r, cbp2);
1196

    
1197
    return 0;
1198
}
1199

    
1200
static int check_slice_end(RV34DecContext *r, MpegEncContext *s)
1201
{
1202
    int bits;
1203
    if(s->mb_y >= s->mb_height)
1204
        return 1;
1205
    if(!s->mb_num_left)
1206
        return 1;
1207
    if(r->s.mb_skip_run > 1)
1208
        return 0;
1209
    bits = r->bits - get_bits_count(&s->gb);
1210
    if(bits < 0 || (bits < 8 && !show_bits(&s->gb, bits)))
1211
        return 1;
1212
    return 0;
1213
}
1214

    
1215
static inline int slice_compare(SliceInfo *si1, SliceInfo *si2)
1216
{
1217
    return si1->type   != si2->type  ||
1218
           si1->start  >= si2->start ||
1219
           si1->width  != si2->width ||
1220
           si1->height != si2->height||
1221
           si1->pts    != si2->pts;
1222
}
1223

    
1224
static int rv34_decode_slice(RV34DecContext *r, int end, const uint8_t* buf, int buf_size)
1225
{
1226
    MpegEncContext *s = &r->s;
1227
    GetBitContext *gb = &s->gb;
1228
    int mb_pos;
1229
    int res;
1230

    
1231
    init_get_bits(&r->s.gb, buf, buf_size*8);
1232
    res = r->parse_slice_header(r, gb, &r->si);
1233
    if(res < 0){
1234
        av_log(s->avctx, AV_LOG_ERROR, "Incorrect or unknown slice header\n");
1235
        return -1;
1236
    }
1237

    
1238
    if ((s->mb_x == 0 && s->mb_y == 0) || s->current_picture_ptr==NULL) {
1239
        if(s->width != r->si.width || s->height != r->si.height){
1240
            av_log(s->avctx, AV_LOG_DEBUG, "Changing dimensions to %dx%d\n", r->si.width,r->si.height);
1241
            MPV_common_end(s);
1242
            s->width  = r->si.width;
1243
            s->height = r->si.height;
1244
            if(MPV_common_init(s) < 0)
1245
                return -1;
1246
            r->intra_types_hist = av_realloc(r->intra_types_hist, s->b4_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1247
            r->intra_types = r->intra_types_hist + s->b4_stride * 4;
1248
            r->mb_type = av_realloc(r->mb_type, r->s.mb_stride * r->s.mb_height * sizeof(*r->mb_type));
1249
            r->cbp_luma   = av_realloc(r->cbp_luma,   r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_luma));
1250
            r->cbp_chroma = av_realloc(r->cbp_chroma, r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_chroma));
1251
            r->deblock_coefs = av_realloc(r->deblock_coefs, r->s.mb_stride * r->s.mb_height * sizeof(*r->deblock_coefs));
1252
        }
1253
        s->pict_type = r->si.type ? r->si.type : FF_I_TYPE;
1254
        if(MPV_frame_start(s, s->avctx) < 0)
1255
            return -1;
1256
        ff_er_frame_start(s);
1257
        r->cur_pts = r->si.pts;
1258
        if(s->pict_type != FF_B_TYPE){
1259
            r->last_pts = r->next_pts;
1260
            r->next_pts = r->cur_pts;
1261
        }
1262
        s->mb_x = s->mb_y = 0;
1263
    }
1264

    
1265
    r->si.end = end;
1266
    s->qscale = r->si.quant;
1267
    r->bits = buf_size*8;
1268
    s->mb_num_left = r->si.end - r->si.start;
1269
    r->s.mb_skip_run = 0;
1270

    
1271
    mb_pos = s->mb_x + s->mb_y * s->mb_width;
1272
    if(r->si.start != mb_pos){
1273
        av_log(s->avctx, AV_LOG_ERROR, "Slice indicates MB offset %d, got %d\n", r->si.start, mb_pos);
1274
        s->mb_x = r->si.start % s->mb_width;
1275
        s->mb_y = r->si.start / s->mb_width;
1276
    }
1277
    memset(r->intra_types_hist, -1, s->b4_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1278
    s->first_slice_line = 1;
1279
    s->resync_mb_x= s->mb_x;
1280
    s->resync_mb_y= s->mb_y;
1281

    
1282
    ff_init_block_index(s);
1283
    while(!check_slice_end(r, s)) {
1284
        ff_update_block_index(s);
1285
        s->dsp.clear_blocks(s->block[0]);
1286

    
1287
        if(rv34_decode_macroblock(r, r->intra_types + s->mb_x * 4 + 1) < 0){
1288
            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);
1289
            return -1;
1290
        }
1291
        if (++s->mb_x == s->mb_width) {
1292
            s->mb_x = 0;
1293
            s->mb_y++;
1294
            ff_init_block_index(s);
1295

    
1296
            memmove(r->intra_types_hist, r->intra_types, s->b4_stride * 4 * sizeof(*r->intra_types_hist));
1297
            memset(r->intra_types, -1, s->b4_stride * 4 * sizeof(*r->intra_types_hist));
1298

    
1299
            if(r->loop_filter && s->mb_y >= 2)
1300
                r->loop_filter(r, s->mb_y - 2);
1301
        }
1302
        if(s->mb_x == s->resync_mb_x)
1303
            s->first_slice_line=0;
1304
        s->mb_num_left--;
1305
    }
1306
    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);
1307

    
1308
    return s->mb_y == s->mb_height;
1309
}
1310

    
1311
/** @} */ // recons group end
1312

    
1313
/**
1314
 * Initialize decoder.
1315
 */
1316
av_cold int ff_rv34_decode_init(AVCodecContext *avctx)
1317
{
1318
    RV34DecContext *r = avctx->priv_data;
1319
    MpegEncContext *s = &r->s;
1320

    
1321
    MPV_decode_defaults(s);
1322
    s->avctx= avctx;
1323
    s->out_format = FMT_H263;
1324
    s->codec_id= avctx->codec_id;
1325

    
1326
    s->width = avctx->width;
1327
    s->height = avctx->height;
1328

    
1329
    r->s.avctx = avctx;
1330
    avctx->flags |= CODEC_FLAG_EMU_EDGE;
1331
    r->s.flags |= CODEC_FLAG_EMU_EDGE;
1332
    avctx->pix_fmt = PIX_FMT_YUV420P;
1333
    avctx->has_b_frames = 1;
1334
    s->low_delay = 0;
1335

    
1336
    if (MPV_common_init(s) < 0)
1337
        return -1;
1338

    
1339
    ff_h264_pred_init(&r->h, CODEC_ID_RV40);
1340

    
1341
    r->intra_types_hist = av_malloc(s->b4_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1342
    r->intra_types = r->intra_types_hist + s->b4_stride * 4;
1343

    
1344
    r->mb_type = av_mallocz(r->s.mb_stride * r->s.mb_height * sizeof(*r->mb_type));
1345

    
1346
    r->cbp_luma   = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_luma));
1347
    r->cbp_chroma = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_chroma));
1348
    r->deblock_coefs = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->deblock_coefs));
1349

    
1350
    if(!intra_vlcs[0].cbppattern[0].bits)
1351
        rv34_init_tables();
1352

    
1353
    return 0;
1354
}
1355

    
1356
static int get_slice_offset(AVCodecContext *avctx, const uint8_t *buf, int n)
1357
{
1358
    if(avctx->slice_count) return avctx->slice_offset[n];
1359
    else                   return AV_RL32(buf + n*8 - 4) == 1 ? AV_RL32(buf + n*8) :  AV_RB32(buf + n*8);
1360
}
1361

    
1362
int ff_rv34_decode_frame(AVCodecContext *avctx,
1363
                            void *data, int *data_size,
1364
                            const uint8_t *buf, int buf_size)
1365
{
1366
    RV34DecContext *r = avctx->priv_data;
1367
    MpegEncContext *s = &r->s;
1368
    AVFrame *pict = data;
1369
    SliceInfo si;
1370
    int i;
1371
    int slice_count;
1372
    const uint8_t *slices_hdr = NULL;
1373
    int last = 0;
1374

    
1375
    /* no supplementary picture */
1376
    if (buf_size == 0) {
1377
        /* special case for last picture */
1378
        if (s->low_delay==0 && s->next_picture_ptr) {
1379
            *pict= *(AVFrame*)s->next_picture_ptr;
1380
            s->next_picture_ptr= NULL;
1381

    
1382
            *data_size = sizeof(AVFrame);
1383
        }
1384
        return 0;
1385
    }
1386

    
1387
    if(!avctx->slice_count){
1388
        slice_count = (*buf++) + 1;
1389
        slices_hdr = buf + 4;
1390
        buf += 8 * slice_count;
1391
    }else
1392
        slice_count = avctx->slice_count;
1393

    
1394
    for(i=0; i<slice_count; i++){
1395
        int offset= get_slice_offset(avctx, slices_hdr, i);
1396
        int size;
1397
        if(i+1 == slice_count)
1398
            size= buf_size - offset;
1399
        else
1400
            size= get_slice_offset(avctx, slices_hdr, i+1) - offset;
1401

    
1402
        if(offset > buf_size){
1403
            av_log(avctx, AV_LOG_ERROR, "Slice offset is greater than frame size\n");
1404
            break;
1405
        }
1406

    
1407
        r->si.end = s->mb_width * s->mb_height;
1408
        if(i+1 < slice_count){
1409
            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);
1410
            if(r->parse_slice_header(r, &r->s.gb, &si) < 0){
1411
                if(i+2 < slice_count)
1412
                    size = get_slice_offset(avctx, slices_hdr, i+2) - offset;
1413
                else
1414
                    size = buf_size - offset;
1415
            }else
1416
                r->si.end = si.start;
1417
        }
1418
        if(!i && si.type == FF_B_TYPE && (!s->last_picture_ptr || !s->last_picture_ptr->data[0]))
1419
            return -1;
1420
        last = rv34_decode_slice(r, r->si.end, buf + offset, size);
1421
        s->mb_num_left = r->s.mb_x + r->s.mb_y*r->s.mb_width - r->si.start;
1422
        if(last)
1423
            break;
1424
    }
1425

    
1426
    if(last){
1427
        if(r->loop_filter)
1428
            r->loop_filter(r, s->mb_height - 1);
1429
        ff_er_frame_end(s);
1430
        MPV_frame_end(s);
1431
        if (s->pict_type == FF_B_TYPE || s->low_delay) {
1432
            *pict= *(AVFrame*)s->current_picture_ptr;
1433
        } else if (s->last_picture_ptr != NULL) {
1434
            *pict= *(AVFrame*)s->last_picture_ptr;
1435
        }
1436

    
1437
        if(s->last_picture_ptr || s->low_delay){
1438
            *data_size = sizeof(AVFrame);
1439
            ff_print_debug_info(s, pict);
1440
        }
1441
        s->current_picture_ptr= NULL; //so we can detect if frame_end wasnt called (find some nicer solution...)
1442
    }
1443
    return buf_size;
1444
}
1445

    
1446
av_cold int ff_rv34_decode_end(AVCodecContext *avctx)
1447
{
1448
    RV34DecContext *r = avctx->priv_data;
1449

    
1450
    MPV_common_end(&r->s);
1451

    
1452
    av_freep(&r->intra_types_hist);
1453
    r->intra_types = NULL;
1454
    av_freep(&r->mb_type);
1455
    av_freep(&r->cbp_luma);
1456
    av_freep(&r->cbp_chroma);
1457
    av_freep(&r->deblock_coefs);
1458

    
1459
    return 0;
1460
}