GCC Code Coverage Report
Directory: ../../../ffmpeg/ Exec Total Coverage
File: src/libavcodec/rv34.c Lines: 932 1042 89.4 %
Date: 2021-01-20 23:14:43 Branches: 547 665 82.3 %

Line Branch Exec Source
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
24
 * RV30/40 decoder common data
25
 */
26
27
#include "libavutil/avassert.h"
28
#include "libavutil/imgutils.h"
29
#include "libavutil/internal.h"
30
#include "libavutil/mem_internal.h"
31
#include "libavutil/thread.h"
32
#include "libavutil/video_enc_params.h"
33
34
#include "avcodec.h"
35
#include "error_resilience.h"
36
#include "mpegutils.h"
37
#include "mpegvideo.h"
38
#include "golomb.h"
39
#include "internal.h"
40
#include "mathops.h"
41
#include "mpeg_er.h"
42
#include "qpeldsp.h"
43
#include "rectangle.h"
44
#include "thread.h"
45
46
#include "rv34vlc.h"
47
#include "rv34data.h"
48
#include "rv34.h"
49
50
454602
static inline void ZERO8x2(void* dst, int stride)
51
{
52
454602
    fill_rectangle(dst,                 1, 2, stride, 0, 4);
53
454602
    fill_rectangle(((uint8_t*)(dst))+4, 1, 2, stride, 0, 4);
54
454602
}
55
56
/** translation of RV30/40 macroblock types to lavc ones */
57
static const int rv34_mb_type_to_lavc[12] = {
58
    MB_TYPE_INTRA,
59
    MB_TYPE_INTRA16x16              | MB_TYPE_SEPARATE_DC,
60
    MB_TYPE_16x16   | MB_TYPE_L0,
61
    MB_TYPE_8x8     | MB_TYPE_L0,
62
    MB_TYPE_16x16   | MB_TYPE_L0,
63
    MB_TYPE_16x16   | MB_TYPE_L1,
64
    MB_TYPE_SKIP,
65
    MB_TYPE_DIRECT2 | MB_TYPE_16x16,
66
    MB_TYPE_16x8    | MB_TYPE_L0,
67
    MB_TYPE_8x16    | MB_TYPE_L0,
68
    MB_TYPE_16x16   | MB_TYPE_L0L1,
69
    MB_TYPE_16x16   | MB_TYPE_L0    | MB_TYPE_SEPARATE_DC
70
};
71
72
73
static RV34VLC intra_vlcs[NUM_INTRA_TABLES], inter_vlcs[NUM_INTER_TABLES];
74
75
static int rv34_decode_mv(RV34DecContext *r, int block_type);
76
77
/**
78
 * @name RV30/40 VLC generating functions
79
 * @{
80
 */
81
82
static VLC_TYPE table_data[117592][2];
83
84
/**
85
 * Generate VLC from codeword lengths.
86
 * @param bits   codeword lengths (zeroes are accepted)
87
 * @param size   length of input data
88
 * @param vlc    output VLC
89
 * @param insyms symbols for input codes (NULL for default ones)
90
 * @param num    VLC table number (for static initialization)
91
 */
92
895
static void rv34_gen_vlc(const uint8_t *bits, int size, VLC *vlc, const uint8_t *syms,
93
                         int *offset)
94
{
95
895
    int counts[17] = {0}, codes[17];
96
    uint16_t cw[MAX_VLC_SIZE];
97
    int maxbits;
98
99
291215
    for (int i = 0; i < size; i++)
100
290320
        counts[bits[i]]++;
101
102
    /* bits[0] is zero for some tables, i.e. syms actually starts at 1.
103
     * So we reset it here. The code assigned to this element is 0x00. */
104
895
    codes[0] = counts[0] = 0;
105
15215
    for (int i = 0; i < 16; i++) {
106
14320
        codes[i+1] = (codes[i] + counts[i]) << 1;
107
14320
        if (counts[i])
108
8095
            maxbits = i;
109
    }
110
291215
    for (int i = 0; i < size; i++)
111
290320
        cw[i] = codes[bits[i]]++;
112
113
895
    vlc->table           = &table_data[*offset];
114
895
    vlc->table_allocated = FF_ARRAY_ELEMS(table_data) - *offset;
115
895
    ff_init_vlc_sparse(vlc, FFMIN(maxbits, 9), size,
116
                       bits, 1, 1,
117
                       cw,    2, 2,
118
                       syms, !!syms, !!syms, INIT_VLC_STATIC_OVERLONG);
119
895
    *offset += vlc->table_size;
120
895
}
121
122
/**
123
 * Initialize all tables.
124
 */
125
5
static av_cold void rv34_init_tables(void)
126
{
127
5
    int i, j, k, offset = 0;
128
129
30
    for(i = 0; i < NUM_INTRA_TABLES; i++){
130
75
        for(j = 0; j < 2; j++){
131
50
            rv34_gen_vlc(rv34_table_intra_cbppat   [i][j], CBPPAT_VLC_SIZE,
132
                         &intra_vlcs[i].cbppattern[j],     NULL, &offset);
133
50
            rv34_gen_vlc(rv34_table_intra_secondpat[i][j], OTHERBLK_VLC_SIZE,
134
                         &intra_vlcs[i].second_pattern[j], NULL, &offset);
135
50
            rv34_gen_vlc(rv34_table_intra_thirdpat [i][j], OTHERBLK_VLC_SIZE,
136
                         &intra_vlcs[i].third_pattern[j],  NULL, &offset);
137
250
            for(k = 0; k < 4; k++){
138
200
                rv34_gen_vlc(rv34_table_intra_cbp[i][j+k*2],  CBP_VLC_SIZE,
139
                             &intra_vlcs[i].cbp[j][k], rv34_cbp_code, &offset);
140
            }
141
        }
142
125
        for(j = 0; j < 4; j++){
143
100
            rv34_gen_vlc(rv34_table_intra_firstpat[i][j], FIRSTBLK_VLC_SIZE,
144
                         &intra_vlcs[i].first_pattern[j], NULL, &offset);
145
        }
146
25
        rv34_gen_vlc(rv34_intra_coeff[i], COEFF_VLC_SIZE,
147
                     &intra_vlcs[i].coefficient, NULL, &offset);
148
    }
149
150
40
    for(i = 0; i < NUM_INTER_TABLES; i++){
151
35
        rv34_gen_vlc(rv34_inter_cbppat[i], CBPPAT_VLC_SIZE,
152
                     &inter_vlcs[i].cbppattern[0], NULL, &offset);
153
175
        for(j = 0; j < 4; j++){
154
140
            rv34_gen_vlc(rv34_inter_cbp[i][j], CBP_VLC_SIZE,
155
                         &inter_vlcs[i].cbp[0][j], rv34_cbp_code, &offset);
156
        }
157
105
        for(j = 0; j < 2; j++){
158
70
            rv34_gen_vlc(rv34_table_inter_firstpat [i][j], FIRSTBLK_VLC_SIZE,
159
                         &inter_vlcs[i].first_pattern[j],  NULL, &offset);
160
70
            rv34_gen_vlc(rv34_table_inter_secondpat[i][j], OTHERBLK_VLC_SIZE,
161
                         &inter_vlcs[i].second_pattern[j], NULL, &offset);
162
70
            rv34_gen_vlc(rv34_table_inter_thirdpat [i][j], OTHERBLK_VLC_SIZE,
163
                         &inter_vlcs[i].third_pattern[j],  NULL, &offset);
164
        }
165
35
        rv34_gen_vlc(rv34_inter_coeff[i], COEFF_VLC_SIZE,
166
                     &inter_vlcs[i].coefficient, NULL, &offset);
167
    }
168
5
}
169
170
/** @} */ // vlc group
171
172
/**
173
 * @name RV30/40 4x4 block decoding functions
174
 * @{
175
 */
176
177
/**
178
 * Decode coded block pattern.
179
 */
180
129189
static int rv34_decode_cbp(GetBitContext *gb, RV34VLC *vlc, int table)
181
{
182
129189
    int pattern, code, cbp=0;
183
    int ones;
184
    static const int cbp_masks[3] = {0x100000, 0x010000, 0x110000};
185
    static const int shifts[4] = { 0, 2, 8, 10 };
186
129189
    const int *curshift = shifts;
187
    int i, t, mask;
188
189
129189
    code = get_vlc2(gb, vlc->cbppattern[table].table, 9, 2);
190
129189
    pattern = code & 0xF;
191
129189
    code >>= 4;
192
193
129189
    ones = rv34_count_ones[pattern];
194
195
645945
    for(mask = 8; mask; mask >>= 1, curshift++){
196
516756
        if(pattern & mask)
197
181416
            cbp |= get_vlc2(gb, vlc->cbp[table][ones].table, vlc->cbp[table][ones].bits, 1) << curshift[0];
198
    }
199
200
645945
    for(i = 0; i < 4; i++){
201
516756
        t = (modulo_three_table[code] >> (6 - 2*i)) & 3;
202
516756
        if(t == 1)
203
121158
            cbp |= cbp_masks[get_bits1(gb)] << i;
204
516756
        if(t == 2)
205
59165
            cbp |= cbp_masks[2] << i;
206
    }
207
129189
    return cbp;
208
}
209
210
/**
211
 * Get one coefficient value from the bitstream and store it.
212
 */
213
2269770
static inline void decode_coeff(int16_t *dst, int coef, int esc, GetBitContext *gb, VLC* vlc, int q)
214
{
215
2269770
    if(coef){
216
1158695
        if(coef == esc){
217
158483
            coef = get_vlc2(gb, vlc->table, 9, 2);
218
158483
            if(coef > 23){
219
139
                coef -= 23;
220
139
                coef = 22 + ((1 << coef) | get_bits(gb, coef));
221
            }
222
158483
            coef += esc;
223
        }
224
1158695
        if(get_bits1(gb))
225
572454
            coef = -coef;
226
1158695
        *dst = (coef*q + 8) >> 4;
227
    }
228
2269770
}
229
230
/**
231
 * Decode 2x2 subblock of coefficients.
232
 */
233
134059
static inline void decode_subblock(int16_t *dst, int code, const int is_block2, GetBitContext *gb, VLC *vlc, int q)
234
{
235
134059
    int flags = modulo_three_table[code];
236
237
134059
    decode_coeff(    dst+0*4+0, (flags >> 6)    , 3, gb, vlc, q);
238
134059
    if(is_block2){
239
83261
        decode_coeff(dst+1*4+0, (flags >> 4) & 3, 2, gb, vlc, q);
240
83261
        decode_coeff(dst+0*4+1, (flags >> 2) & 3, 2, gb, vlc, q);
241
    }else{
242
50798
        decode_coeff(dst+0*4+1, (flags >> 4) & 3, 2, gb, vlc, q);
243
50798
        decode_coeff(dst+1*4+0, (flags >> 2) & 3, 2, gb, vlc, q);
244
    }
245
134059
    decode_coeff(    dst+1*4+1, (flags >> 0) & 3, 2, gb, vlc, q);
246
134059
}
247
248
/**
249
 * Decode a single coefficient.
250
 */
251
404286
static inline void decode_subblock1(int16_t *dst, int code, GetBitContext *gb, VLC *vlc, int q)
252
{
253
404286
    int coeff = modulo_three_table[code] >> 6;
254
404286
    decode_coeff(dst, coeff, 3, gb, vlc, q);
255
404286
}
256
257
332312
static inline void decode_subblock3(int16_t *dst, int code, GetBitContext *gb, VLC *vlc,
258
                                    int q_dc, int q_ac1, int q_ac2)
259
{
260
332312
    int flags = modulo_three_table[code];
261
262
332312
    decode_coeff(dst+0*4+0, (flags >> 6)    , 3, gb, vlc, q_dc);
263
332312
    decode_coeff(dst+0*4+1, (flags >> 4) & 3, 2, gb, vlc, q_ac1);
264
332312
    decode_coeff(dst+1*4+0, (flags >> 2) & 3, 2, gb, vlc, q_ac1);
265
332312
    decode_coeff(dst+1*4+1, (flags >> 0) & 3, 2, gb, vlc, q_ac2);
266
332312
}
267
268
/**
269
 * Decode coefficients for 4x4 block.
270
 *
271
 * This is done by filling 2x2 subblocks with decoded coefficients
272
 * in this order (the same for subblocks and subblock coefficients):
273
 *  o--o
274
 *    /
275
 *   /
276
 *  o--o
277
 */
278
279
736598
static int rv34_decode_block(int16_t *dst, GetBitContext *gb, RV34VLC *rvlc, int fc, int sc, int q_dc, int q_ac1, int q_ac2)
280
{
281
736598
    int code, pattern, has_ac = 1;
282
283
736598
    code = get_vlc2(gb, rvlc->first_pattern[fc].table, 9, 2);
284
285
736598
    pattern = code & 0x7;
286
287
736598
    code >>= 3;
288
289
736598
    if (modulo_three_table[code] & 0x3F) {
290
332312
        decode_subblock3(dst, code, gb, &rvlc->coefficient, q_dc, q_ac1, q_ac2);
291
    } else {
292
404286
        decode_subblock1(dst, code, gb, &rvlc->coefficient, q_dc);
293
404286
        if (!pattern)
294
363407
            return 0;
295
40879
        has_ac = 0;
296
    }
297
298
373191
    if(pattern & 4){
299
44632
        code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
300
44632
        decode_subblock(dst + 4*0+2, code, 0, gb, &rvlc->coefficient, q_ac2);
301
    }
302
373191
    if(pattern & 2){ // Looks like coefficients 1 and 2 are swapped for this block
303
83261
        code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
304
83261
        decode_subblock(dst + 4*2+0, code, 1, gb, &rvlc->coefficient, q_ac2);
305
    }
306
373191
    if(pattern & 1){
307
6166
        code = get_vlc2(gb, rvlc->third_pattern[sc].table, 9, 2);
308
6166
        decode_subblock(dst + 4*2+2, code, 0, gb, &rvlc->coefficient, q_ac2);
309
    }
310
373191
    return has_ac | pattern;
311
}
312
313
/**
314
 * @name RV30/40 bitstream parsing
315
 * @{
316
 */
317
318
/**
319
 * Decode starting slice position.
320
 * @todo Maybe replace with ff_h263_decode_mba() ?
321
 */
322
2770
int ff_rv34_get_start_offset(GetBitContext *gb, int mb_size)
323
{
324
    int i;
325
10368
    for(i = 0; i < 5; i++)
326
10368
        if(rv34_mb_max_sizes[i] >= mb_size - 1)
327
2770
            break;
328
2770
    return rv34_mb_bits_sizes[i];
329
}
330
331
/**
332
 * Select VLC set for decoding from current quantizer, modifier and frame type.
333
 */
334
130283
static inline RV34VLC* choose_vlc_set(int quant, int mod, int type)
335
{
336

130283
    if(mod == 2 && quant < 19) quant += 10;
337

70216
    else if(mod && quant < 26) quant += 5;
338
    av_assert2(quant >= 0 && quant < 32);
339
97819
    return type ? &inter_vlcs[rv34_quant_to_vlc_set[1][quant]]
340
130283
                : &intra_vlcs[rv34_quant_to_vlc_set[0][quant]];
341
}
342
343
/**
344
 * Decode intra macroblock header and return CBP in case of success, -1 otherwise.
345
 */
346
7640
static int rv34_decode_intra_mb_header(RV34DecContext *r, int8_t *intra_types)
347
{
348
7640
    MpegEncContext *s = &r->s;
349
7640
    GetBitContext *gb = &s->gb;
350
7640
    int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
351
    int t;
352
353
7640
    r->is16 = get_bits1(gb);
354
7640
    if(r->is16){
355
5830
        s->current_picture_ptr->mb_type[mb_pos] = MB_TYPE_INTRA16x16;
356
5830
        r->block_type = RV34_MB_TYPE_INTRA16x16;
357
5830
        t = get_bits(gb, 2);
358
5830
        fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0]));
359
5830
        r->luma_vlc   = 2;
360
    }else{
361
1810
        if(!r->rv30){
362
467
            if(!get_bits1(gb))
363
                av_log(s->avctx, AV_LOG_ERROR, "Need DQUANT\n");
364
        }
365
1810
        s->current_picture_ptr->mb_type[mb_pos] = MB_TYPE_INTRA;
366
1810
        r->block_type = RV34_MB_TYPE_INTRA;
367
1810
        if(r->decode_intra_types(r, gb, intra_types) < 0)
368
            return -1;
369
1810
        r->luma_vlc   = 1;
370
    }
371
372
7640
    r->chroma_vlc = 0;
373
7640
    r->cur_vlcs   = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
374
375
7640
    return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
376
}
377
378
/**
379
 * Decode inter macroblock header and return CBP in case of success, -1 otherwise.
380
 */
381
284640
static int rv34_decode_inter_mb_header(RV34DecContext *r, int8_t *intra_types)
382
{
383
284640
    MpegEncContext *s = &r->s;
384
284640
    GetBitContext *gb = &s->gb;
385
284640
    int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
386
    int i, t;
387
388
284640
    r->block_type = r->decode_mb_info(r);
389
284640
    if(r->block_type == -1)
390
        return -1;
391
284640
    s->current_picture_ptr->mb_type[mb_pos] = rv34_mb_type_to_lavc[r->block_type];
392
284640
    r->mb_type[mb_pos] = r->block_type;
393
284640
    if(r->block_type == RV34_MB_SKIP){
394
163091
        if(s->pict_type == AV_PICTURE_TYPE_P)
395
31305
            r->mb_type[mb_pos] = RV34_MB_P_16x16;
396
163091
        if(s->pict_type == AV_PICTURE_TYPE_B)
397
131786
            r->mb_type[mb_pos] = RV34_MB_B_DIRECT;
398
    }
399
284640
    r->is16 = !!IS_INTRA16x16(s->current_picture_ptr->mb_type[mb_pos]);
400
284640
    if (rv34_decode_mv(r, r->block_type) < 0)
401
        return -1;
402
284640
    if(r->block_type == RV34_MB_SKIP){
403
163091
        fill_rectangle(intra_types, 4, 4, r->intra_types_stride, 0, sizeof(intra_types[0]));
404
163091
        return 0;
405
    }
406
121549
    r->chroma_vlc = 1;
407
121549
    r->luma_vlc   = 0;
408
409
121549
    if(IS_INTRA(s->current_picture_ptr->mb_type[mb_pos])){
410
24277
        if(r->is16){
411
10976
            t = get_bits(gb, 2);
412
10976
            fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0]));
413
10976
            r->luma_vlc   = 2;
414
        }else{
415
13301
            if(r->decode_intra_types(r, gb, intra_types) < 0)
416
                return -1;
417
13301
            r->luma_vlc   = 1;
418
        }
419
24277
        r->chroma_vlc = 0;
420
24277
        r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
421
    }else{
422
1653624
        for(i = 0; i < 16; i++)
423
1556352
            intra_types[(i & 3) + (i>>2) * r->intra_types_stride] = 0;
424
97272
        r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
425
97272
        if(r->mb_type[mb_pos] == RV34_MB_P_MIX16x16){
426
547
            r->is16 = 1;
427
547
            r->chroma_vlc = 1;
428
547
            r->luma_vlc   = 2;
429
547
            r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
430
        }
431
    }
432
433
121549
    return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
434
}
435
436
/** @} */ //bitstream functions
437
438
/**
439
 * @name motion vector related code (prediction, reconstruction, motion compensation)
440
 * @{
441
 */
442
443
/** macroblock partition width in 8x8 blocks */
444
static const uint8_t part_sizes_w[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 2, 1, 2, 2 };
445
446
/** macroblock partition height in 8x8 blocks */
447
static const uint8_t part_sizes_h[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 1, 2, 2, 2 };
448
449
/** availability index for subblocks */
450
static const uint8_t avail_indexes[4] = { 6, 7, 10, 11 };
451
452
/**
453
 * motion vector prediction
454
 *
455
 * Motion prediction performed for the block by using median prediction of
456
 * motion vectors from the left, top and right top blocks but in corner cases
457
 * some other vectors may be used instead.
458
 */
459
42620
static void rv34_pred_mv(RV34DecContext *r, int block_type, int subblock_no, int dmv_no)
460
{
461
42620
    MpegEncContext *s = &r->s;
462
42620
    int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
463
42620
    int A[2] = {0}, B[2], C[2];
464
    int i, j;
465
    int mx, my;
466
42620
    int* avail = r->avail_cache + avail_indexes[subblock_no];
467
42620
    int c_off = part_sizes_w[block_type];
468
469
42620
    mv_pos += (subblock_no & 1) + (subblock_no >> 1)*s->b8_stride;
470
42620
    if(subblock_no == 3)
471
4526
        c_off = -1;
472
473
42620
    if(avail[-1]){
474
41140
        A[0] = s->current_picture_ptr->motion_val[0][mv_pos-1][0];
475
41140
        A[1] = s->current_picture_ptr->motion_val[0][mv_pos-1][1];
476
    }
477
42620
    if(avail[-4]){
478
30363
        B[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][0];
479
30363
        B[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][1];
480
    }else{
481
12257
        B[0] = A[0];
482
12257
        B[1] = A[1];
483
    }
484
42620
    if(!avail[c_off-4]){
485

13924
        if(avail[-4] && (avail[-1] || r->rv30)){
486
1813
            C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][0];
487
1813
            C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][1];
488
        }else{
489
12111
            C[0] = A[0];
490
12111
            C[1] = A[1];
491
        }
492
    }else{
493
28696
        C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][0];
494
28696
        C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][1];
495
    }
496
42620
    mx = mid_pred(A[0], B[0], C[0]);
497
42620
    my = mid_pred(A[1], B[1], C[1]);
498
42620
    mx += r->dmv[dmv_no][0];
499
42620
    my += r->dmv[dmv_no][1];
500
107064
    for(j = 0; j < part_sizes_h[block_type]; j++){
501
165336
        for(i = 0; i < part_sizes_w[block_type]; i++){
502
100892
            s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][0] = mx;
503
100892
            s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][1] = my;
504
        }
505
    }
506
42620
}
507
508
#define GET_PTS_DIFF(a, b) (((a) - (b) + 8192) & 0x1FFF)
509
510
/**
511
 * Calculate motion vector component that should be added for direct blocks.
512
 */
513
802736
static int calc_add_mv(RV34DecContext *r, int dir, int val)
514
{
515
802736
    int mul = dir ? -r->mv_weight2 : r->mv_weight1;
516
517
802736
    return (int)(val * (SUINT)mul + 0x2000) >> 14;
518
}
519
520
/**
521
 * Predict motion vector for B-frame macroblock.
522
 */
523
47002
static inline void rv34_pred_b_vector(int A[2], int B[2], int C[2],
524
                                      int A_avail, int B_avail, int C_avail,
525
                                      int *mx, int *my)
526
{
527
47002
    if(A_avail + B_avail + C_avail != 3){
528
34307
        *mx = A[0] + B[0] + C[0];
529
34307
        *my = A[1] + B[1] + C[1];
530
34307
        if(A_avail + B_avail + C_avail == 2){
531
11581
            *mx /= 2;
532
11581
            *my /= 2;
533
        }
534
    }else{
535
12695
        *mx = mid_pred(A[0], B[0], C[0]);
536
12695
        *my = mid_pred(A[1], B[1], C[1]);
537
    }
538
47002
}
539
540
/**
541
 * motion vector prediction for B-frames
542
 */
543
47002
static void rv34_pred_mv_b(RV34DecContext *r, int block_type, int dir)
544
{
545
47002
    MpegEncContext *s = &r->s;
546
47002
    int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
547
47002
    int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
548
47002
    int A[2] = { 0 }, B[2] = { 0 }, C[2] = { 0 };
549
47002
    int has_A = 0, has_B = 0, has_C = 0;
550
    int mx, my;
551
    int i, j;
552
47002
    Picture *cur_pic = s->current_picture_ptr;
553
47002
    const int mask = dir ? MB_TYPE_L1 : MB_TYPE_L0;
554
47002
    int type = cur_pic->mb_type[mb_pos];
555
556
47002
    if((r->avail_cache[6-1] & type) & mask){
557
29561
        A[0] = cur_pic->motion_val[dir][mv_pos - 1][0];
558
29561
        A[1] = cur_pic->motion_val[dir][mv_pos - 1][1];
559
29561
        has_A = 1;
560
    }
561
47002
    if((r->avail_cache[6-4] & type) & mask){
562
24041
        B[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][0];
563
24041
        B[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][1];
564
24041
        has_B = 1;
565
    }
566

47002
    if(r->avail_cache[6-4] && (r->avail_cache[6-2] & type) & mask){
567
21963
        C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][0];
568
21963
        C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][1];
569
21963
        has_C = 1;
570

25039
    }else if((s->mb_x+1) == s->mb_width && (r->avail_cache[6-5] & type) & mask){
571
386
        C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][0];
572
386
        C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][1];
573
386
        has_C = 1;
574
    }
575
576
47002
    rv34_pred_b_vector(A, B, C, has_A, has_B, has_C, &mx, &my);
577
578
47002
    mx += r->dmv[dir][0];
579
47002
    my += r->dmv[dir][1];
580
581
141006
    for(j = 0; j < 2; j++){
582
282012
        for(i = 0; i < 2; i++){
583
188008
            cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][0] = mx;
584
188008
            cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][1] = my;
585
        }
586
    }
587

47002
    if(block_type == RV34_MB_B_BACKWARD || block_type == RV34_MB_B_FORWARD){
588
28476
        ZERO8x2(cur_pic->motion_val[!dir][mv_pos], s->b8_stride);
589
    }
590
47002
}
591
592
/**
593
 * motion vector prediction - RV3 version
594
 */
595
9134
static void rv34_pred_mv_rv3(RV34DecContext *r, int block_type, int dir)
596
{
597
9134
    MpegEncContext *s = &r->s;
598
9134
    int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
599
9134
    int A[2] = {0}, B[2], C[2];
600
    int i, j, k;
601
    int mx, my;
602
9134
    int* avail = r->avail_cache + avail_indexes[0];
603
604
9134
    if(avail[-1]){
605
8666
        A[0] = s->current_picture_ptr->motion_val[0][mv_pos - 1][0];
606
8666
        A[1] = s->current_picture_ptr->motion_val[0][mv_pos - 1][1];
607
    }
608
9134
    if(avail[-4]){
609
8618
        B[0] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride][0];
610
8618
        B[1] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride][1];
611
    }else{
612
516
        B[0] = A[0];
613
516
        B[1] = A[1];
614
    }
615
9134
    if(!avail[-4 + 2]){
616

950
        if(avail[-4] && (avail[-1])){
617
434
            C[0] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride - 1][0];
618
434
            C[1] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride - 1][1];
619
        }else{
620
516
            C[0] = A[0];
621
516
            C[1] = A[1];
622
        }
623
    }else{
624
8184
        C[0] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride + 2][0];
625
8184
        C[1] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride + 2][1];
626
    }
627
9134
    mx = mid_pred(A[0], B[0], C[0]);
628
9134
    my = mid_pred(A[1], B[1], C[1]);
629
9134
    mx += r->dmv[0][0];
630
9134
    my += r->dmv[0][1];
631
27402
    for(j = 0; j < 2; j++){
632
54804
        for(i = 0; i < 2; i++){
633
109608
            for(k = 0; k < 2; k++){
634
73072
                s->current_picture_ptr->motion_val[k][mv_pos + i + j*s->b8_stride][0] = mx;
635
73072
                s->current_picture_ptr->motion_val[k][mv_pos + i + j*s->b8_stride][1] = my;
636
            }
637
        }
638
    }
639
9134
}
640
641
static const int chroma_coeffs[3] = { 0, 3, 5 };
642
643
/**
644
 * generic motion compensation function
645
 *
646
 * @param r decoder context
647
 * @param block_type type of the current block
648
 * @param xoff horizontal offset from the start of the current block
649
 * @param yoff vertical offset from the start of the current block
650
 * @param mv_off offset to the motion vector information
651
 * @param width width of the current partition in 8x8 blocks
652
 * @param height height of the current partition in 8x8 blocks
653
 * @param dir motion compensation direction (i.e. from the last or the next reference frame)
654
 * @param thirdpel motion vectors are specified in 1/3 of pixel
655
 * @param qpel_mc a set of functions used to perform luma motion compensation
656
 * @param chroma_mc a set of functions used to perform chroma motion compensation
657
 */
658
511743
static inline void rv34_mc(RV34DecContext *r, const int block_type,
659
                          const int xoff, const int yoff, int mv_off,
660
                          const int width, const int height, int dir,
661
                          const int thirdpel, int weighted,
662
                          qpel_mc_func (*qpel_mc)[16],
663
                          h264_chroma_mc_func (*chroma_mc))
664
{
665
511743
    MpegEncContext *s = &r->s;
666
    uint8_t *Y, *U, *V, *srcY, *srcU, *srcV;
667
    int dxy, mx, my, umx, umy, lx, ly, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
668
511743
    int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride + mv_off;
669
511743
    int is16x16 = 1;
670
511743
    int emu = 0;
671
672
511743
    if(thirdpel){
673
        int chroma_mx, chroma_my;
674
127708
        mx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) / 3 - (1 << 24);
675
127708
        my = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) / 3 - (1 << 24);
676
127708
        lx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) % 3;
677
127708
        ly = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) % 3;
678
127708
        chroma_mx = s->current_picture_ptr->motion_val[dir][mv_pos][0] / 2;
679
127708
        chroma_my = s->current_picture_ptr->motion_val[dir][mv_pos][1] / 2;
680
127708
        umx = (chroma_mx + (3 << 24)) / 3 - (1 << 24);
681
127708
        umy = (chroma_my + (3 << 24)) / 3 - (1 << 24);
682
127708
        uvmx = chroma_coeffs[(chroma_mx + (3 << 24)) % 3];
683
127708
        uvmy = chroma_coeffs[(chroma_my + (3 << 24)) % 3];
684
    }else{
685
        int cx, cy;
686
384035
        mx = s->current_picture_ptr->motion_val[dir][mv_pos][0] >> 2;
687
384035
        my = s->current_picture_ptr->motion_val[dir][mv_pos][1] >> 2;
688
384035
        lx = s->current_picture_ptr->motion_val[dir][mv_pos][0] & 3;
689
384035
        ly = s->current_picture_ptr->motion_val[dir][mv_pos][1] & 3;
690
384035
        cx = s->current_picture_ptr->motion_val[dir][mv_pos][0] / 2;
691
384035
        cy = s->current_picture_ptr->motion_val[dir][mv_pos][1] / 2;
692
384035
        umx = cx >> 2;
693
384035
        umy = cy >> 2;
694
384035
        uvmx = (cx & 3) << 1;
695
384035
        uvmy = (cy & 3) << 1;
696
        //due to some flaw RV40 uses the same MC compensation routine for H2V2 and H3V3
697

384035
        if(uvmx == 6 && uvmy == 6)
698
7377
            uvmx = uvmy = 4;
699
    }
700
701
511743
    if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
702
        /* wait for the referenced mb row to be finished */
703
        int mb_row = s->mb_y + ((yoff + my + 5 + 8 * height) >> 4);
704
        ThreadFrame *f = dir ? &s->next_picture_ptr->tf : &s->last_picture_ptr->tf;
705
        ff_thread_await_progress(f, mb_row, 0);
706
    }
707
708
511743
    dxy = ly*4 + lx;
709
511743
    srcY = dir ? s->next_picture_ptr->f->data[0] : s->last_picture_ptr->f->data[0];
710
511743
    srcU = dir ? s->next_picture_ptr->f->data[1] : s->last_picture_ptr->f->data[1];
711
511743
    srcV = dir ? s->next_picture_ptr->f->data[2] : s->last_picture_ptr->f->data[2];
712
511743
    src_x = s->mb_x * 16 + xoff + mx;
713
511743
    src_y = s->mb_y * 16 + yoff + my;
714
511743
    uvsrc_x = s->mb_x * 8 + (xoff >> 1) + umx;
715
511743
    uvsrc_y = s->mb_y * 8 + (yoff >> 1) + umy;
716
511743
    srcY += src_y * s->linesize + src_x;
717
511743
    srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
718
511743
    srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
719

511743
    if(s->h_edge_pos - (width << 3) < 6 || s->v_edge_pos - (height << 3) < 6 ||
720

511743
       (unsigned)(src_x - !!lx*2) > s->h_edge_pos - !!lx*2 - (width <<3) - 4 ||
721

495678
       (unsigned)(src_y - !!ly*2) > s->v_edge_pos - !!ly*2 - (height<<3) - 4) {
722
43652
        srcY -= 2 + 2*s->linesize;
723
43652
        s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, srcY,
724
                                 s->linesize, s->linesize,
725
43652
                                 (width << 3) + 6, (height << 3) + 6,
726
                                 src_x - 2, src_y - 2,
727
                                 s->h_edge_pos, s->v_edge_pos);
728
43652
        srcY = s->sc.edge_emu_buffer + 2 + 2*s->linesize;
729
43652
        emu = 1;
730
    }
731
511743
    if(!weighted){
732
230597
        Y = s->dest[0] + xoff      + yoff     *s->linesize;
733
230597
        U = s->dest[1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
734
230597
        V = s->dest[2] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
735
    }else{
736
281146
        Y = r->tmp_b_block_y [dir]     +  xoff     +  yoff    *s->linesize;
737
281146
        U = r->tmp_b_block_uv[dir*2]   + (xoff>>1) + (yoff>>1)*s->uvlinesize;
738
281146
        V = r->tmp_b_block_uv[dir*2+1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
739
    }
740
741
511743
    if(block_type == RV34_MB_P_16x8){
742
2692
        qpel_mc[1][dxy](Y, srcY, s->linesize);
743
2692
        Y    += 8;
744
2692
        srcY += 8;
745
509051
    }else if(block_type == RV34_MB_P_8x16){
746
4946
        qpel_mc[1][dxy](Y, srcY, s->linesize);
747
4946
        Y    += 8 * s->linesize;
748
4946
        srcY += 8 * s->linesize;
749
    }
750

511743
    is16x16 = (block_type != RV34_MB_P_8x8) && (block_type != RV34_MB_P_16x8) && (block_type != RV34_MB_P_8x16);
751
511743
    qpel_mc[!is16x16][dxy](Y, srcY, s->linesize);
752
511743
    if (emu) {
753
43652
        uint8_t *uvbuf = s->sc.edge_emu_buffer;
754
755
43652
        s->vdsp.emulated_edge_mc(uvbuf, srcU,
756
                                 s->uvlinesize, s->uvlinesize,
757
43652
                                 (width << 2) + 1, (height << 2) + 1,
758
                                 uvsrc_x, uvsrc_y,
759
43652
                                 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
760
43652
        srcU = uvbuf;
761
43652
        uvbuf += 9*s->uvlinesize;
762
763
43652
        s->vdsp.emulated_edge_mc(uvbuf, srcV,
764
                                 s->uvlinesize, s->uvlinesize,
765
43652
                                 (width << 2) + 1, (height << 2) + 1,
766
                                 uvsrc_x, uvsrc_y,
767
43652
                                 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
768
43652
        srcV = uvbuf;
769
    }
770
511743
    chroma_mc[2-width]   (U, srcU, s->uvlinesize, height*4, uvmx, uvmy);
771
511743
    chroma_mc[2-width]   (V, srcV, s->uvlinesize, height*4, uvmx, uvmy);
772
511743
}
773
774
111535
static void rv34_mc_1mv(RV34DecContext *r, const int block_type,
775
                        const int xoff, const int yoff, int mv_off,
776
                        const int width, const int height, int dir)
777
{
778
111535
    rv34_mc(r, block_type, xoff, yoff, mv_off, width, height, dir, r->rv30, 0,
779
111535
            r->rdsp.put_pixels_tab,
780
111535
            r->rdsp.put_chroma_pixels_tab);
781
111535
}
782
783
115964
static void rv4_weight(RV34DecContext *r)
784
{
785
115964
    r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][0](r->s.dest[0],
786
                                                        r->tmp_b_block_y[0],
787
                                                        r->tmp_b_block_y[1],
788
                                                        r->weight1,
789
                                                        r->weight2,
790
                                                        r->s.linesize);
791
115964
    r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][1](r->s.dest[1],
792
                                                        r->tmp_b_block_uv[0],
793
                                                        r->tmp_b_block_uv[2],
794
                                                        r->weight1,
795
                                                        r->weight2,
796
                                                        r->s.uvlinesize);
797
115964
    r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][1](r->s.dest[2],
798
                                                        r->tmp_b_block_uv[1],
799
                                                        r->tmp_b_block_uv[3],
800
                                                        r->weight1,
801
                                                        r->weight2,
802
                                                        r->s.uvlinesize);
803
115964
}
804
805
154932
static void rv34_mc_2mv(RV34DecContext *r, const int block_type)
806
{
807

154932
    int weighted = !r->rv30 && block_type != RV34_MB_B_BIDIR && r->weight1 != 8192;
808
809
154932
    rv34_mc(r, block_type, 0, 0, 0, 2, 2, 0, r->rv30, weighted,
810
154932
            r->rdsp.put_pixels_tab,
811
154932
            r->rdsp.put_chroma_pixels_tab);
812
154932
    if(!weighted){
813
47171
        rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30, 0,
814
47171
                r->rdsp.avg_pixels_tab,
815
47171
                r->rdsp.avg_chroma_pixels_tab);
816
    }else{
817
107761
        rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30, 1,
818
107761
                r->rdsp.put_pixels_tab,
819
107761
                r->rdsp.put_chroma_pixels_tab);
820
107761
        rv4_weight(r);
821
    }
822
154932
}
823
824
11293
static void rv34_mc_2mv_skip(RV34DecContext *r)
825
{
826
    int i, j;
827

11293
    int weighted = !r->rv30 && r->weight1 != 8192;
828
829
33879
    for(j = 0; j < 2; j++)
830
67758
        for(i = 0; i < 2; i++){
831
45172
             rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 0, r->rv30,
832
                     weighted,
833
45172
                     r->rdsp.put_pixels_tab,
834
45172
                     r->rdsp.put_chroma_pixels_tab);
835

45172
             rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 1, r->rv30,
836
                     weighted,
837
                     weighted ? r->rdsp.put_pixels_tab : r->rdsp.avg_pixels_tab,
838
                     weighted ? r->rdsp.put_chroma_pixels_tab : r->rdsp.avg_chroma_pixels_tab);
839
        }
840
11293
    if(weighted)
841
8203
        rv4_weight(r);
842
11293
}
843
844
/** number of motion vectors in each macroblock type */
845
static const int num_mvs[RV34_MB_TYPES] = { 0, 0, 1, 4, 1, 1, 0, 0, 2, 2, 2, 1 };
846
847
/**
848
 * Decode motion vector differences
849
 * and perform motion vector reconstruction and motion compensation.
850
 */
851
284640
static int rv34_decode_mv(RV34DecContext *r, int block_type)
852
{
853
284640
    MpegEncContext *s = &r->s;
854
284640
    GetBitContext *gb = &s->gb;
855
    int i, j, k, l;
856
284640
    int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
857
    int next_bt;
858
859
284640
    memset(r->dmv, 0, sizeof(r->dmv));
860
383396
    for(i = 0; i < num_mvs[block_type]; i++){
861
98756
        r->dmv[i][0] = get_interleaved_se_golomb(gb);
862
98756
        r->dmv[i][1] = get_interleaved_se_golomb(gb);
863
98756
        if (r->dmv[i][0] == INVALID_VLC ||
864
98756
            r->dmv[i][1] == INVALID_VLC) {
865
            r->dmv[i][0] = r->dmv[i][1] = 0;
866
            return AVERROR_INVALIDDATA;
867
        }
868
    }
869


284640
    switch(block_type){
870
24277
    case RV34_MB_TYPE_INTRA:
871
    case RV34_MB_TYPE_INTRA16x16:
872
24277
        ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
873
24277
        return 0;
874
163091
    case RV34_MB_SKIP:
875
163091
        if(s->pict_type == AV_PICTURE_TYPE_P){
876
31305
            ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
877
31305
            rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
878
31305
            break;
879
        }
880
    case RV34_MB_B_DIRECT:
881
        //surprisingly, it uses motion scheme from next reference frame
882
        /* wait for the current mb row to be finished */
883
156962
        if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
884
            ff_thread_await_progress(&s->next_picture_ptr->tf, FFMAX(0, s->mb_y-1), 0);
885
886
156962
        next_bt = s->next_picture_ptr->mb_type[s->mb_x + s->mb_y * s->mb_stride];
887

156962
        if(IS_INTRA(next_bt) || IS_SKIP(next_bt)){
888
106791
            ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
889
106791
            ZERO8x2(s->current_picture_ptr->motion_val[1][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
890
        }else
891
150513
            for(j = 0; j < 2; j++)
892
301026
                for(i = 0; i < 2; i++)
893
602052
                    for(k = 0; k < 2; k++)
894
1204104
                        for(l = 0; l < 2; l++)
895
802736
                            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]);
896

156962
        if(!(IS_16X8(next_bt) || IS_8X16(next_bt) || IS_8X8(next_bt))) //we can use whole macroblock MC
897
145669
            rv34_mc_2mv(r, block_type);
898
        else
899
11293
            rv34_mc_2mv_skip(r);
900
156962
        ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
901
156962
        break;
902
16878
    case RV34_MB_P_16x16:
903
    case RV34_MB_P_MIX16x16:
904
16878
        rv34_pred_mv(r, block_type, 0, 0);
905
16878
        rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
906
16878
        break;
907
37610
    case RV34_MB_B_FORWARD:
908
    case RV34_MB_B_BACKWARD:
909
37610
        r->dmv[1][0] = r->dmv[0][0];
910
37610
        r->dmv[1][1] = r->dmv[0][1];
911
37610
        if(r->rv30)
912
9134
            rv34_pred_mv_rv3(r, block_type, block_type == RV34_MB_B_BACKWARD);
913
        else
914
28476
            rv34_pred_mv_b  (r, block_type, block_type == RV34_MB_B_BACKWARD);
915
37610
        rv34_mc_1mv     (r, block_type, 0, 0, 0, 2, 2, block_type == RV34_MB_B_BACKWARD);
916
37610
        break;
917
3819
    case RV34_MB_P_16x8:
918
    case RV34_MB_P_8x16:
919
3819
        rv34_pred_mv(r, block_type, 0, 0);
920
3819
        rv34_pred_mv(r, block_type, 1 + (block_type == RV34_MB_P_16x8), 1);
921
3819
        if(block_type == RV34_MB_P_16x8){
922
1346
            rv34_mc_1mv(r, block_type, 0, 0, 0,            2, 1, 0);
923
1346
            rv34_mc_1mv(r, block_type, 0, 8, s->b8_stride, 2, 1, 0);
924
        }
925
3819
        if(block_type == RV34_MB_P_8x16){
926
2473
            rv34_mc_1mv(r, block_type, 0, 0, 0, 1, 2, 0);
927
2473
            rv34_mc_1mv(r, block_type, 8, 0, 1, 1, 2, 0);
928
        }
929
3819
        break;
930
9263
    case RV34_MB_B_BIDIR:
931
9263
        rv34_pred_mv_b  (r, block_type, 0);
932
9263
        rv34_pred_mv_b  (r, block_type, 1);
933
9263
        rv34_mc_2mv     (r, block_type);
934
9263
        break;
935
4526
    case RV34_MB_P_8x8:
936
22630
        for(i=0;i< 4;i++){
937
18104
            rv34_pred_mv(r, block_type, i, i);
938
18104
            rv34_mc_1mv (r, block_type, (i&1)<<3, (i&2)<<2, (i&1)+(i>>1)*s->b8_stride, 1, 1, 0);
939
        }
940
4526
        break;
941
    }
942
943
260363
    return 0;
944
}
945
/** @} */ // mv group
946
947
/**
948
 * @name Macroblock reconstruction functions
949
 * @{
950
 */
951
/** mapping of RV30/40 intra prediction types to standard H.264 types */
952
static const int ittrans[9] = {
953
 DC_PRED, VERT_PRED, HOR_PRED, DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_LEFT_PRED,
954
 VERT_RIGHT_PRED, VERT_LEFT_PRED, HOR_UP_PRED, HOR_DOWN_PRED,
955
};
956
957
/** mapping of RV30/40 intra 16x16 prediction types to standard H.264 types */
958
static const int ittrans16[4] = {
959
 DC_PRED8x8, VERT_PRED8x8, HOR_PRED8x8, PLANE_PRED8x8,
960
};
961
962
/**
963
 * Perform 4x4 intra prediction.
964
 */
965
362664
static void rv34_pred_4x4_block(RV34DecContext *r, uint8_t *dst, int stride, int itype, int up, int left, int down, int right)
966
{
967
362664
    uint8_t *prev = dst - stride + 4;
968
    uint32_t topleft;
969
970

362664
    if(!up && !left)
971
906
        itype = DC_128_PRED;
972
361758
    else if(!up){
973
43702
        if(itype == VERT_PRED) itype = HOR_PRED;
974
43702
        if(itype == DC_PRED)   itype = LEFT_DC_PRED;
975
318056
    }else if(!left){
976
5318
        if(itype == HOR_PRED)  itype = VERT_PRED;
977
5318
        if(itype == DC_PRED)   itype = TOP_DC_PRED;
978
5318
        if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
979
    }
980
362664
    if(!down){
981
289443
        if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
982
289443
        if(itype == HOR_UP_PRED) itype = HOR_UP_PRED_RV40_NODOWN;
983
289443
        if(itype == VERT_LEFT_PRED) itype = VERT_LEFT_PRED_RV40_NODOWN;
984
    }
985

362664
    if(!right && up){
986
76047
        topleft = dst[-stride + 3] * 0x01010101u;
987
76047
        prev = (uint8_t*)&topleft;
988
    }
989
362664
    r->h.pred4x4[itype](dst, prev, stride);
990
362664
}
991
992
33612
static inline int adjust_pred16(int itype, int up, int left)
993
{
994

33612
    if(!up && !left)
995
54
        itype = DC_128_PRED8x8;
996
33558
    else if(!up){
997
5754
        if(itype == PLANE_PRED8x8)itype = HOR_PRED8x8;
998
5754
        if(itype == VERT_PRED8x8) itype = HOR_PRED8x8;
999
5754
        if(itype == DC_PRED8x8)   itype = LEFT_DC_PRED8x8;
1000
27804
    }else if(!left){
1001
836
        if(itype == PLANE_PRED8x8)itype = VERT_PRED8x8;
1002
836
        if(itype == HOR_PRED8x8)  itype = VERT_PRED8x8;
1003
836
        if(itype == DC_PRED8x8)   itype = TOP_DC_PRED8x8;
1004
    }
1005
33612
    return itype;
1006
}
1007
1008
704977
static inline void rv34_process_block(RV34DecContext *r,
1009
                                      uint8_t *pdst, int stride,
1010
                                      int fc, int sc, int q_dc, int q_ac)
1011
{
1012
704977
    MpegEncContext *s = &r->s;
1013
704977
    int16_t *ptr = s->block[0];
1014
704977
    int has_ac = rv34_decode_block(ptr, &s->gb, r->cur_vlcs,
1015
                                   fc, sc, q_dc, q_ac, q_ac);
1016
704977
    if(has_ac){
1017
350745
        r->rdsp.rv34_idct_add(pdst, stride, ptr);
1018
    }else{
1019
354232
        r->rdsp.rv34_idct_dc_add(pdst, stride, ptr[0]);
1020
354232
        ptr[0] = 0;
1021
    }
1022
704977
}
1023
1024
16806
static void rv34_output_i16x16(RV34DecContext *r, int8_t *intra_types, int cbp)
1025
{
1026
16806
    LOCAL_ALIGNED_16(int16_t, block16, [16]);
1027
16806
    MpegEncContext *s    = &r->s;
1028
16806
    GetBitContext  *gb   = &s->gb;
1029
16806
    int             q_dc = rv34_qscale_tab[ r->luma_dc_quant_i[s->qscale] ],
1030
16806
                    q_ac = rv34_qscale_tab[s->qscale];
1031
16806
    uint8_t        *dst  = s->dest[0];
1032
16806
    int16_t        *ptr  = s->block[0];
1033
    int i, j, itype, has_ac;
1034
1035
16806
    memset(block16, 0, 16 * sizeof(*block16));
1036
1037
16806
    has_ac = rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac);
1038
16806
    if(has_ac)
1039
7912
        r->rdsp.rv34_inv_transform(block16);
1040
    else
1041
8894
        r->rdsp.rv34_inv_transform_dc(block16);
1042
1043
16806
    itype = ittrans16[intra_types[0]];
1044
16806
    itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
1045
16806
    r->h.pred16x16[itype](dst, s->linesize);
1046
1047
84030
    for(j = 0; j < 4; j++){
1048
336120
        for(i = 0; i < 4; i++, cbp >>= 1){
1049
268896
            int dc = block16[i + j*4];
1050
1051
268896
            if(cbp & 1){
1052
14262
                has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);
1053
            }else
1054
254634
                has_ac = 0;
1055
1056
268896
            if(has_ac){
1057
14262
                ptr[0] = dc;
1058
14262
                r->rdsp.rv34_idct_add(dst+4*i, s->linesize, ptr);
1059
            }else
1060
254634
                r->rdsp.rv34_idct_dc_add(dst+4*i, s->linesize, dc);
1061
        }
1062
1063
67224
        dst += 4*s->linesize;
1064
    }
1065
1066
16806
    itype = ittrans16[intra_types[0]];
1067
16806
    if(itype == PLANE_PRED8x8) itype = DC_PRED8x8;
1068
16806
    itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
1069
1070
16806
    q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
1071
16806
    q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
1072
1073
50418
    for(j = 1; j < 3; j++){
1074
33612
        dst = s->dest[j];
1075
33612
        r->h.pred8x8[itype](dst, s->uvlinesize);
1076
168060
        for(i = 0; i < 4; i++, cbp >>= 1){
1077
            uint8_t *pdst;
1078
134448
            if(!(cbp & 1)) continue;
1079
24665
            pdst   = dst + (i&1)*4 + (i&2)*2*s->uvlinesize;
1080
1081
24665
            rv34_process_block(r, pdst, s->uvlinesize,
1082
                               r->chroma_vlc, 1, q_dc, q_ac);
1083
        }
1084
    }
1085
16806
}
1086
1087
15111
static void rv34_output_intra(RV34DecContext *r, int8_t *intra_types, int cbp)
1088
{
1089
15111
    MpegEncContext *s   = &r->s;
1090
15111
    uint8_t        *dst = s->dest[0];
1091
15111
    int      avail[6*8] = {0};
1092
    int i, j, k;
1093
    int idx, q_ac, q_dc;
1094
1095
    // Set neighbour information.
1096
15111
    if(r->avail_cache[1])
1097
8914
        avail[0] = 1;
1098
15111
    if(r->avail_cache[2])
1099
9535
        avail[1] = avail[2] = 1;
1100
15111
    if(r->avail_cache[3])
1101
9535
        avail[3] = avail[4] = 1;
1102
15111
    if(r->avail_cache[4])
1103
9499
        avail[5] = 1;
1104
15111
    if(r->avail_cache[5])
1105
14333
        avail[8] = avail[16] = 1;
1106
15111
    if(r->avail_cache[9])
1107
14333
        avail[24] = avail[32] = 1;
1108
1109
15111
    q_ac = rv34_qscale_tab[s->qscale];
1110
75555
    for(j = 0; j < 4; j++){
1111
60444
        idx = 9 + j*8;
1112
302220
        for(i = 0; i < 4; i++, cbp >>= 1, dst += 4, idx++){
1113
241776
            rv34_pred_4x4_block(r, dst, s->linesize, ittrans[intra_types[i]], avail[idx-8], avail[idx-1], avail[idx+7], avail[idx-7]);
1114
241776
            avail[idx] = 1;
1115
241776
            if(!(cbp & 1)) continue;
1116
1117
148411
            rv34_process_block(r, dst, s->linesize,
1118
                               r->luma_vlc, 0, q_ac, q_ac);
1119
        }
1120
60444
        dst += s->linesize * 4 - 4*4;
1121
60444
        intra_types += r->intra_types_stride;
1122
    }
1123
1124
15111
    intra_types -= r->intra_types_stride * 4;
1125
1126
15111
    q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
1127
15111
    q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
1128
1129
45333
    for(k = 0; k < 2; k++){
1130
30222
        dst = s->dest[1+k];
1131
30222
        fill_rectangle(r->avail_cache + 6, 2, 2, 4, 0, 4);
1132
1133
90666
        for(j = 0; j < 2; j++){
1134
60444
            int* acache = r->avail_cache + 6 + j*4;
1135
181332
            for(i = 0; i < 2; i++, cbp >>= 1, acache++){
1136
120888
                int itype = ittrans[intra_types[i*2+j*2*r->intra_types_stride]];
1137

120888
                rv34_pred_4x4_block(r, dst+4*i, s->uvlinesize, itype, acache[-4], acache[-1], !i && !j, acache[-3]);
1138
120888
                acache[0] = 1;
1139
1140
120888
                if(!(cbp&1)) continue;
1141
1142
71067
                rv34_process_block(r, dst + 4*i, s->uvlinesize,
1143
                                   r->chroma_vlc, 1, q_dc, q_ac);
1144
            }
1145
1146
60444
            dst += 4*s->uvlinesize;
1147
        }
1148
    }
1149
15111
}
1150
1151
2246368
static int is_mv_diff_gt_3(int16_t (*motion_val)[2], int step)
1152
{
1153
    int d;
1154
2246368
    d = motion_val[0][0] - motion_val[-step][0];
1155

2246368
    if(d < -3 || d > 3)
1156
188252
        return 1;
1157
2058116
    d = motion_val[0][1] - motion_val[-step][1];
1158

2058116
    if(d < -3 || d > 3)
1159
31852
        return 1;
1160
2026264
    return 0;
1161
}
1162
1163
284640
static int rv34_set_deblock_coef(RV34DecContext *r)
1164
{
1165
284640
    MpegEncContext *s = &r->s;
1166
284640
    int hmvmask = 0, vmvmask = 0, i, j;
1167
284640
    int midx = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
1168
284640
    int16_t (*motion_val)[2] = &s->current_picture_ptr->motion_val[0][midx];
1169
853920
    for(j = 0; j < 16; j += 8){
1170
1707840
        for(i = 0; i < 2; i++){
1171
1138560
            if(is_mv_diff_gt_3(motion_val + i, 1))
1172
106472
                vmvmask |= 0x11 << (j + i*2);
1173

1138560
            if((j || s->mb_y) && is_mv_diff_gt_3(motion_val + i, s->b8_stride))
1174
113632
                hmvmask |= 0x03 << (j + i*2);
1175
        }
1176
569280
        motion_val += s->b8_stride;
1177
    }
1178
284640
    if(s->first_slice_line)
1179
42347
        hmvmask &= ~0x000F;
1180
284640
    if(!s->mb_x)
1181
9120
        vmvmask &= ~0x1111;
1182
284640
    if(r->rv30){ //RV30 marks both subblocks on the edge for filtering
1183
68640
        vmvmask |= (vmvmask & 0x4444) >> 1;
1184
68640
        hmvmask |= (hmvmask & 0x0F00) >> 4;
1185
68640
        if(s->mb_x)
1186
65520
            r->deblock_coefs[s->mb_x - 1 + s->mb_y*s->mb_stride] |= (vmvmask & 0x1111) << 3;
1187
68640
        if(!s->first_slice_line)
1188
61878
            r->deblock_coefs[s->mb_x + (s->mb_y - 1)*s->mb_stride] |= (hmvmask & 0xF) << 12;
1189
    }
1190
284640
    return hmvmask | vmvmask;
1191
}
1192
1193
284640
static int rv34_decode_inter_macroblock(RV34DecContext *r, int8_t *intra_types)
1194
{
1195
284640
    MpegEncContext *s   = &r->s;
1196
284640
    GetBitContext  *gb  = &s->gb;
1197
284640
    uint8_t        *dst = s->dest[0];
1198
284640
    int16_t        *ptr = s->block[0];
1199
284640
    int          mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1200
    int cbp, cbp2;
1201
    int q_dc, q_ac, has_ac;
1202
    int i, j;
1203
    int dist;
1204
1205
    // Calculate which neighbours are available. Maybe it's worth optimizing too.
1206
284640
    memset(r->avail_cache, 0, sizeof(r->avail_cache));
1207
284640
    fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);
1208
284640
    dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
1209

284640
    if(s->mb_x && dist)
1210
275017
        r->avail_cache[5] =
1211
275017
        r->avail_cache[9] = s->current_picture_ptr->mb_type[mb_pos - 1];
1212
284640
    if(dist >= s->mb_width)
1213
242293
        r->avail_cache[2] =
1214
242293
        r->avail_cache[3] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride];
1215

284640
    if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
1216
234929
        r->avail_cache[4] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride + 1];
1217

284640
    if(s->mb_x && dist > s->mb_width)
1218
234004
        r->avail_cache[1] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride - 1];
1219
1220
284640
    s->qscale = r->si.quant;
1221
284640
    cbp = cbp2 = rv34_decode_inter_mb_header(r, intra_types);
1222
284640
    r->cbp_luma  [mb_pos] = cbp;
1223
284640
    r->cbp_chroma[mb_pos] = cbp >> 16;
1224
284640
    r->deblock_coefs[mb_pos] = rv34_set_deblock_coef(r) | r->cbp_luma[mb_pos];
1225
284640
    s->current_picture_ptr->qscale_table[mb_pos] = s->qscale;
1226
1227
284640
    if(cbp == -1)
1228
        return -1;
1229
1230
284640
    if (IS_INTRA(s->current_picture_ptr->mb_type[mb_pos])){
1231
24277
        if(r->is16) rv34_output_i16x16(r, intra_types, cbp);
1232
13301
        else        rv34_output_intra(r, intra_types, cbp);
1233
24277
        return 0;
1234
    }
1235
1236
260363
    if(r->is16){
1237
        // Only for RV34_MB_P_MIX16x16
1238
547
        LOCAL_ALIGNED_16(int16_t, block16, [16]);
1239
547
        memset(block16, 0, 16 * sizeof(*block16));
1240
547
        q_dc = rv34_qscale_tab[ r->luma_dc_quant_p[s->qscale] ];
1241
547
        q_ac = rv34_qscale_tab[s->qscale];
1242
547
        if (rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac))
1243
266
            r->rdsp.rv34_inv_transform(block16);
1244
        else
1245
281
            r->rdsp.rv34_inv_transform_dc(block16);
1246
1247
547
        q_ac = rv34_qscale_tab[s->qscale];
1248
1249
2735
        for(j = 0; j < 4; j++){
1250
10940
            for(i = 0; i < 4; i++, cbp >>= 1){
1251
8752
                int      dc   = block16[i + j*4];
1252
1253
8752
                if(cbp & 1){
1254
6
                    has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);
1255
                }else
1256
8746
                    has_ac = 0;
1257
1258
8752
                if(has_ac){
1259
6
                    ptr[0] = dc;
1260
6
                    r->rdsp.rv34_idct_add(dst+4*i, s->linesize, ptr);
1261
                }else
1262
8746
                    r->rdsp.rv34_idct_dc_add(dst+4*i, s->linesize, dc);
1263
            }
1264
1265
2188
            dst += 4*s->linesize;
1266
        }
1267
1268
547
        r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
1269
    }else{
1270
259816
        q_ac = rv34_qscale_tab[s->qscale];
1271
1272
1299080
        for(j = 0; j < 4; j++){
1273
5196320
            for(i = 0; i < 4; i++, cbp >>= 1){
1274
4157056
                if(!(cbp & 1)) continue;
1275
1276
317078
                rv34_process_block(r, dst + 4*i, s->linesize,
1277
                                   r->luma_vlc, 0, q_ac, q_ac);
1278
            }
1279
1039264
            dst += 4*s->linesize;
1280
        }
1281
    }
1282
1283
260363
    q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
1284
260363
    q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
1285
1286
781089
    for(j = 1; j < 3; j++){
1287
520726
        dst = s->dest[j];
1288
2603630
        for(i = 0; i < 4; i++, cbp >>= 1){
1289
            uint8_t *pdst;
1290
2082904
            if(!(cbp & 1)) continue;
1291
143756
            pdst = dst + (i&1)*4 + (i&2)*2*s->uvlinesize;
1292
1293
143756
            rv34_process_block(r, pdst, s->uvlinesize,
1294
                               r->chroma_vlc, 1, q_dc, q_ac);
1295
        }
1296
    }
1297
1298
260363
    return 0;
1299
}
1300
1301
7640
static int rv34_decode_intra_macroblock(RV34DecContext *r, int8_t *intra_types)
1302
{
1303
7640
    MpegEncContext *s = &r->s;
1304
    int cbp, dist;
1305
7640
    int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1306
1307
    // Calculate which neighbours are available. Maybe it's worth optimizing too.
1308
7640
    memset(r->avail_cache, 0, sizeof(r->avail_cache));
1309
7640
    fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);
1310
7640
    dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
1311

7640
    if(s->mb_x && dist)
1312
7325
        r->avail_cache[5] =
1313
7325
        r->avail_cache[9] = s->current_picture_ptr->mb_type[mb_pos - 1];
1314
7640
    if(dist >= s->mb_width)
1315
6133
        r->avail_cache[2] =
1316
6133
        r->avail_cache[3] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride];
1317

7640
    if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
1318
5928
        r->avail_cache[4] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride + 1];
1319

7640
    if(s->mb_x && dist > s->mb_width)
1320
5878
        r->avail_cache[1] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride - 1];
1321
1322
7640
    s->qscale = r->si.quant;
1323
7640
    cbp = rv34_decode_intra_mb_header(r, intra_types);
1324
7640
    r->cbp_luma  [mb_pos] = cbp;
1325
7640
    r->cbp_chroma[mb_pos] = cbp >> 16;
1326
7640
    r->deblock_coefs[mb_pos] = 0xFFFF;
1327
7640
    s->current_picture_ptr->qscale_table[mb_pos] = s->qscale;
1328
1329
7640
    if(cbp == -1)
1330
        return -1;
1331
1332
7640
    if(r->is16){
1333
5830
        rv34_output_i16x16(r, intra_types, cbp);
1334
5830
        return 0;
1335
    }
1336
1337
1810
    rv34_output_intra(r, intra_types, cbp);
1338
1810
    return 0;
1339
}
1340
1341
293665
static int check_slice_end(RV34DecContext *r, MpegEncContext *s)
1342
{
1343
    int bits;
1344
293665
    if(s->mb_y >= s->mb_height)
1345
526
        return 1;
1346
293139
    if(!s->mb_num_left)
1347
859
        return 1;
1348
292280
    if(r->s.mb_skip_run > 1)
1349
128319
        return 0;
1350
163961
    bits = get_bits_left(&s->gb);
1351

163961
    if(bits <= 0 || (bits < 8 && !show_bits(&s->gb, bits)))
1352
        return 1;
1353
163961
    return 0;
1354
}
1355
1356
1357
9
static void rv34_decoder_free(RV34DecContext *r)
1358
{
1359
9
    av_freep(&r->intra_types_hist);
1360
9
    r->intra_types = NULL;
1361
9
    av_freep(&r->tmp_b_block_base);
1362
9
    av_freep(&r->mb_type);
1363
9
    av_freep(&r->cbp_luma);
1364
9
    av_freep(&r->cbp_chroma);
1365
9
    av_freep(&r->deblock_coefs);
1366
9
}
1367
1368
1369
9
static int rv34_decoder_alloc(RV34DecContext *r)
1370
{
1371
9
    r->intra_types_stride = r->s.mb_width * 4 + 4;
1372
1373
9
    r->cbp_chroma       = av_mallocz(r->s.mb_stride * r->s.mb_height *
1374
                                    sizeof(*r->cbp_chroma));
1375
9
    r->cbp_luma         = av_mallocz(r->s.mb_stride * r->s.mb_height *
1376
                                    sizeof(*r->cbp_luma));
1377
9
    r->deblock_coefs    = av_mallocz(r->s.mb_stride * r->s.mb_height *
1378
                                    sizeof(*r->deblock_coefs));
1379
9
    r->intra_types_hist = av_malloc(r->intra_types_stride * 4 * 2 *
1380
                                    sizeof(*r->intra_types_hist));
1381
9
    r->mb_type          = av_mallocz(r->s.mb_stride * r->s.mb_height *
1382
                                     sizeof(*r->mb_type));
1383
1384

9
    if (!(r->cbp_chroma       && r->cbp_luma && r->deblock_coefs &&
1385

9
          r->intra_types_hist && r->mb_type)) {
1386
        rv34_decoder_free(r);
1387
        return AVERROR(ENOMEM);
1388
    }
1389
1390
9
    r->intra_types = r->intra_types_hist + r->intra_types_stride * 4;
1391
1392
9
    return 0;
1393
}
1394
1395
1396
static int rv34_decoder_realloc(RV34DecContext *r)
1397
{
1398
    rv34_decoder_free(r);
1399
    return rv34_decoder_alloc(r);
1400
}
1401
1402
1403
1385
static int rv34_decode_slice(RV34DecContext *r, int end, const uint8_t* buf, int buf_size)
1404
{
1405
1385
    MpegEncContext *s = &r->s;
1406
1385
    GetBitContext *gb = &s->gb;
1407
    int mb_pos, slice_type;
1408
    int res;
1409
1410
1385
    init_get_bits(&r->s.gb, buf, buf_size*8);
1411
1385
    res = r->parse_slice_header(r, gb, &r->si);
1412
1385
    if(res < 0){
1413
        av_log(s->avctx, AV_LOG_ERROR, "Incorrect or unknown slice header\n");
1414
        return -1;
1415
    }
1416
1417
1385
    slice_type = r->si.type ? r->si.type : AV_PICTURE_TYPE_I;
1418
1385
    if (slice_type != s->pict_type) {
1419
        av_log(s->avctx, AV_LOG_ERROR, "Slice type mismatch\n");
1420
        return AVERROR_INVALIDDATA;
1421
    }
1422

1385
    if (s->width != r->si.width || s->height != r->si.height) {
1423
        av_log(s->avctx, AV_LOG_ERROR, "Size mismatch\n");
1424
        return AVERROR_INVALIDDATA;
1425
    }
1426
1427
1385
    r->si.end = end;
1428
1385
    s->qscale = r->si.quant;
1429
1385
    s->mb_num_left = r->si.end - r->si.start;
1430
1385
    r->s.mb_skip_run = 0;
1431
1432
1385
    mb_pos = s->mb_x + s->mb_y * s->mb_width;
1433
1385
    if(r->si.start != mb_pos){
1434
        av_log(s->avctx, AV_LOG_ERROR, "Slice indicates MB offset %d, got %d\n", r->si.start, mb_pos);
1435
        s->mb_x = r->si.start % s->mb_width;
1436
        s->mb_y = r->si.start / s->mb_width;
1437
    }
1438
1385
    memset(r->intra_types_hist, -1, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1439
1385
    s->first_slice_line = 1;
1440
1385
    s->resync_mb_x = s->mb_x;
1441
1385
    s->resync_mb_y = s->mb_y;
1442
1443
1385
    ff_init_block_index(s);
1444
293665
    while(!check_slice_end(r, s)) {
1445
292280
        ff_update_block_index(s);
1446
1447
292280
        if(r->si.type)
1448
284640
            res = rv34_decode_inter_macroblock(r, r->intra_types + s->mb_x * 4 + 4);
1449
        else
1450
7640
            res = rv34_decode_intra_macroblock(r, r->intra_types + s->mb_x * 4 + 4);
1451
292280
        if(res < 0){
1452
            ff_er_add_slice(&s->er, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_ERROR);
1453
            return -1;
1454
        }
1455
292280
        if (++s->mb_x == s->mb_width) {
1456
9408
            s->mb_x = 0;
1457
9408
            s->mb_y++;
1458
9408
            ff_init_block_index(s);
1459
1460
9408
            memmove(r->intra_types_hist, r->intra_types, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
1461
9408
            memset(r->intra_types, -1, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
1462
1463

9408
            if(r->loop_filter && s->mb_y >= 2)
1464
8882
                r->loop_filter(r, s->mb_y - 2);
1465
1466
9408
            if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
1467
                ff_thread_report_progress(&s->current_picture_ptr->tf,
1468
                                          s->mb_y - 2, 0);
1469
1470
        }
1471
292280
        if(s->mb_x == s->resync_mb_x)
1472
9031
            s->first_slice_line=0;
1473
292280
        s->mb_num_left--;
1474
    }
1475
1385
    ff_er_add_slice(&s->er, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_END);
1476
1477
1385
    return s->mb_y == s->mb_height;
1478
}
1479
1480
/** @} */ // reconstruction group end
1481
1482
/**
1483
 * Initialize decoder.
1484
 */
1485
9
av_cold int ff_rv34_decode_init(AVCodecContext *avctx)
1486
{
1487
    static AVOnce init_static_once = AV_ONCE_INIT;
1488
9
    RV34DecContext *r = avctx->priv_data;
1489
9
    MpegEncContext *s = &r->s;
1490
    int ret;
1491
1492
9
    ff_mpv_decode_init(s, avctx);
1493
9
    s->out_format = FMT_H263;
1494
1495
9
    avctx->pix_fmt = AV_PIX_FMT_YUV420P;
1496
9
    avctx->has_b_frames = 1;
1497
9
    s->low_delay = 0;
1498
1499
9
    ff_mpv_idct_init(s);
1500
9
    if ((ret = ff_mpv_common_init(s)) < 0)
1501
        return ret;
1502
1503
9
    ff_h264_pred_init(&r->h, AV_CODEC_ID_RV40, 8, 1);
1504
1505
#if CONFIG_RV30_DECODER
1506
9
    if (avctx->codec_id == AV_CODEC_ID_RV30)
1507
5
        ff_rv30dsp_init(&r->rdsp);
1508
#endif
1509
#if CONFIG_RV40_DECODER
1510
9
    if (avctx->codec_id == AV_CODEC_ID_RV40)
1511
4
        ff_rv40dsp_init(&r->rdsp);
1512
#endif
1513
1514
9
    if ((ret = rv34_decoder_alloc(r)) < 0) {
1515
        ff_mpv_common_end(&r->s);
1516
        return ret;
1517
    }
1518
1519
9
    ff_thread_once(&init_static_once, rv34_init_tables);
1520
1521
9
    return 0;
1522
}
1523
1524
int ff_rv34_decode_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)
1525
{
1526
    RV34DecContext *r = dst->priv_data, *r1 = src->priv_data;
1527
    MpegEncContext * const s = &r->s, * const s1 = &r1->s;
1528
    int err;
1529
1530
    if (dst == src || !s1->context_initialized)
1531
        return 0;
1532
1533
    if (s->height != s1->height || s->width != s1->width) {
1534
        s->height = s1->height;
1535
        s->width  = s1->width;
1536
        if ((err = ff_mpv_common_frame_size_change(s)) < 0)
1537
            return err;
1538
        if ((err = rv34_decoder_realloc(r)) < 0)
1539
            return err;
1540
    }
1541
1542
    r->cur_pts  = r1->cur_pts;
1543
    r->last_pts = r1->last_pts;
1544
    r->next_pts = r1->next_pts;
1545
1546
    memset(&r->si, 0, sizeof(r->si));
1547
1548
    // Do no call ff_mpeg_update_thread_context on a partially initialized
1549
    // decoder context.
1550
    if (!s1->context_initialized)
1551
        return 0;
1552
1553
    return ff_mpeg_update_thread_context(dst, src);
1554
}
1555
1556
4156
static int get_slice_offset(AVCodecContext *avctx, const uint8_t *buf, int n, int slice_count, int buf_size)
1557
{
1558
4156
    if (n < slice_count) {
1559
3273
        if(avctx->slice_count) return avctx->slice_offset[n];
1560
3273
        else                   return AV_RL32(buf + n*8 - 4) == 1 ? AV_RL32(buf + n*8) :  AV_RB32(buf + n*8);
1561
    } else
1562
883
        return buf_size;
1563
}
1564
1565
526
static int finish_frame(AVCodecContext *avctx, AVFrame *pict)
1566
{
1567
526
    RV34DecContext *r = avctx->priv_data;
1568
526
    MpegEncContext *s = &r->s;
1569
526
    int got_picture = 0, ret;
1570
1571
526
    ff_er_frame_end(&s->er);
1572
526
    ff_mpv_frame_end(s);
1573
526
    s->mb_num_left = 0;
1574
1575
526
    if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
1576
        ff_thread_report_progress(&s->current_picture_ptr->tf, INT_MAX, 0);
1577
1578

526
    if (s->pict_type == AV_PICTURE_TYPE_B || s->low_delay) {
1579
387
        if ((ret = av_frame_ref(pict, s->current_picture_ptr->f)) < 0)
1580
            return ret;
1581
387
        ff_print_debug_info(s, s->current_picture_ptr, pict);
1582
387
        ff_mpv_export_qp_table(s, pict, s->current_picture_ptr, FF_QSCALE_TYPE_MPEG1);
1583
387
        got_picture = 1;
1584
139
    } else if (s->last_picture_ptr) {
1585
130
        if ((ret = av_frame_ref(pict, s->last_picture_ptr->f)) < 0)
1586
            return ret;
1587
130
        ff_print_debug_info(s, s->last_picture_ptr, pict);
1588
130
        ff_mpv_export_qp_table(s, pict, s->last_picture_ptr, FF_QSCALE_TYPE_MPEG1);
1589
130
        got_picture = 1;
1590
    }
1591
1592
526
    return got_picture;
1593
}
1594
1595
static AVRational update_sar(int old_w, int old_h, AVRational sar, int new_w, int new_h)
1596
{
1597
    // attempt to keep aspect during typical resolution switches
1598
    if (!sar.num)
1599
        sar = (AVRational){1, 1};
1600
1601
    sar = av_mul_q(sar, av_mul_q((AVRational){new_h, new_w}, (AVRational){old_w, old_h}));
1602
    return sar;
1603
}
1604
1605
532
int ff_rv34_decode_frame(AVCodecContext *avctx,
1606
                            void *data, int *got_picture_ptr,
1607
                            AVPacket *avpkt)
1608
{
1609
532
    const uint8_t *buf = avpkt->data;
1610
532
    int buf_size = avpkt->size;
1611
532
    RV34DecContext *r = avctx->priv_data;
1612
532
    MpegEncContext *s = &r->s;
1613
532
    AVFrame *pict = data;
1614
    SliceInfo si;
1615
    int i, ret;
1616
    int slice_count;
1617
532
    const uint8_t *slices_hdr = NULL;
1618
532
    int last = 0;
1619
532
    int faulty_b = 0;
1620
    int offset;
1621
1622
    /* no supplementary picture */
1623
532
    if (buf_size == 0) {
1624
        /* special case for last picture */
1625

6
        if (s->low_delay==0 && s->next_picture_ptr) {
1626
4
            if ((ret = av_frame_ref(pict, s->next_picture_ptr->f)) < 0)
1627
                return ret;
1628
4
            s->next_picture_ptr = NULL;
1629
1630
4
            *got_picture_ptr = 1;
1631
        }
1632
6
        return 0;
1633
    }
1634
1635
526
    if(!avctx->slice_count){
1636
526
        slice_count = (*buf++) + 1;
1637
526
        slices_hdr = buf + 4;
1638
526
        buf += 8 * slice_count;
1639
526
        buf_size -= 1 + 8 * slice_count;
1640
    }else
1641
        slice_count = avctx->slice_count;
1642
1643
526
    offset = get_slice_offset(avctx, slices_hdr, 0, slice_count, buf_size);
1644
    //parse first slice header to check whether this frame can be decoded
1645

526
    if(offset < 0 || offset > buf_size){
1646
        av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
1647
        return AVERROR_INVALIDDATA;
1648
    }
1649
526
    init_get_bits(&s->gb, buf+offset, (buf_size-offset)*8);
1650

526
    if(r->parse_slice_header(r, &r->s.gb, &si) < 0 || si.start){
1651
        av_log(avctx, AV_LOG_ERROR, "First slice header is incorrect\n");
1652
        return AVERROR_INVALIDDATA;
1653
    }
1654

526
    if ((!s->last_picture_ptr || !s->last_picture_ptr->f->data[0]) &&
1655
13
        si.type == AV_PICTURE_TYPE_B) {
1656
        av_log(avctx, AV_LOG_ERROR, "Invalid decoder state: B-frame without "
1657
               "reference data.\n");
1658
        faulty_b = 1;
1659
    }
1660

526
    if(   (avctx->skip_frame >= AVDISCARD_NONREF && si.type==AV_PICTURE_TYPE_B)
1661

526
       || (avctx->skip_frame >= AVDISCARD_NONKEY && si.type!=AV_PICTURE_TYPE_I)
1662
526
       ||  avctx->skip_frame >= AVDISCARD_ALL)
1663
        return avpkt->size;
1664
1665
    /* first slice */
1666
526
    if (si.start == 0) {
1667

526
        if (s->mb_num_left > 0 && s->current_picture_ptr) {
1668
            av_log(avctx, AV_LOG_ERROR, "New frame but still %d MB left.\n",
1669
                   s->mb_num_left);
1670
            ff_er_frame_end(&s->er);
1671
            ff_mpv_frame_end(s);
1672
        }
1673
1674

526
        if (s->width != si.width || s->height != si.height) {
1675
            int err;
1676
1677
            av_log(s->avctx, AV_LOG_WARNING, "Changing dimensions to %dx%d\n",
1678
                   si.width, si.height);
1679
1680
            if (av_image_check_size(si.width, si.height, 0, s->avctx))
1681
                return AVERROR_INVALIDDATA;
1682
1683
            s->avctx->sample_aspect_ratio = update_sar(
1684
                s->width, s->height, s->avctx->sample_aspect_ratio,
1685
                si.width, si.height);
1686
            s->width  = si.width;
1687
            s->height = si.height;
1688
1689
            err = ff_set_dimensions(s->avctx, s->width, s->height);
1690
            if (err < 0)
1691
                return err;
1692
1693
            if ((err = ff_mpv_common_frame_size_change(s)) < 0)
1694
                return err;
1695
            if ((err = rv34_decoder_realloc(r)) < 0)
1696
                return err;
1697
        }
1698
526
        if (faulty_b)
1699
            return AVERROR_INVALIDDATA;
1700
526
        s->pict_type = si.type ? si.type : AV_PICTURE_TYPE_I;
1701
526
        if (ff_mpv_frame_start(s, s->avctx) < 0)
1702
            return -1;
1703
526
        ff_mpeg_er_frame_start(s);
1704
526
        if (!r->tmp_b_block_base) {
1705
            int i;
1706
1707
9
            r->tmp_b_block_base = av_malloc(s->linesize * 48);
1708
27
            for (i = 0; i < 2; i++)
1709
18
                r->tmp_b_block_y[i] = r->tmp_b_block_base
1710
18
                                      + i * 16 * s->linesize;
1711
45
            for (i = 0; i < 4; i++)
1712
36
                r->tmp_b_block_uv[i] = r->tmp_b_block_base + 32 * s->linesize
1713
36
                                       + (i >> 1) * 8 * s->uvlinesize
1714
36
                                       + (i &  1) * 16;
1715
        }
1716
526
        r->cur_pts = si.pts;
1717
526
        if (s->pict_type != AV_PICTURE_TYPE_B) {
1718
139
            r->last_pts = r->next_pts;
1719
139
            r->next_pts = r->cur_pts;
1720
        } else {
1721
387
            int refdist = GET_PTS_DIFF(r->next_pts, r->last_pts);
1722
387
            int dist0   = GET_PTS_DIFF(r->cur_pts,  r->last_pts);
1723
387
            int dist1   = GET_PTS_DIFF(r->next_pts, r->cur_pts);
1724
1725
387
            if(!refdist){
1726
                r->mv_weight1 = r->mv_weight2 = r->weight1 = r->weight2 = 8192;
1727
                r->scaled_weight = 0;
1728
            }else{
1729
387
                if (FFMAX(dist0, dist1) > refdist)
1730
                    av_log(avctx, AV_LOG_TRACE, "distance overflow\n");
1731
1732
387
                r->mv_weight1 = (dist0 << 14) / refdist;
1733
387
                r->mv_weight2 = (dist1 << 14) / refdist;
1734
387
                if((r->mv_weight1|r->mv_weight2) & 511){
1735
349
                    r->weight1 = r->mv_weight1;
1736
349
                    r->weight2 = r->mv_weight2;
1737
349
                    r->scaled_weight = 0;
1738
                }else{
1739
38
                    r->weight1 = r->mv_weight1 >> 9;
1740
38
                    r->weight2 = r->mv_weight2 >> 9;
1741
38
                    r->scaled_weight = 1;
1742
                }
1743
            }
1744
        }
1745
526
        s->mb_x = s->mb_y = 0;
1746
526
        ff_thread_finish_setup(s->avctx);
1747
    } else if (HAVE_THREADS &&
1748
               (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
1749
        av_log(s->avctx, AV_LOG_ERROR, "Decoder needs full frames in frame "
1750
               "multithreading mode (start MB is %d).\n", si.start);
1751
        return AVERROR_INVALIDDATA;
1752
    }
1753
1754
1385
    for(i = 0; i < slice_count; i++){
1755
1385
        int offset  = get_slice_offset(avctx, slices_hdr, i  , slice_count, buf_size);
1756
1385
        int offset1 = get_slice_offset(avctx, slices_hdr, i+1, slice_count, buf_size);
1757
        int size;
1758
1759

1385
        if(offset < 0 || offset > offset1 || offset1 > buf_size){
1760
            av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
1761
            break;
1762
        }
1763
1385
        size = offset1 - offset;
1764
1765
1385
        r->si.end = s->mb_width * s->mb_height;
1766
1385
        s->mb_num_left = r->s.mb_x + r->s.mb_y*r->s.mb_width - r->si.start;
1767
1768
1385
        if(i+1 < slice_count){
1769
860
            int offset2 = get_slice_offset(avctx, slices_hdr, i+2, slice_count, buf_size);
1770

860
            if (offset2 < offset1 || offset2 > buf_size) {
1771
                av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
1772
                break;
1773
            }
1774
860
            init_get_bits(&s->gb, buf+offset1, (buf_size-offset1)*8);
1775
860
            if(r->parse_slice_header(r, &r->s.gb, &si) < 0){
1776
1
                size = offset2 - offset;
1777
            }else
1778
859
                r->si.end = si.start;
1779
        }
1780

1385
        av_assert0 (size >= 0 && size <= buf_size - offset);
1781
1385
        last = rv34_decode_slice(r, r->si.end, buf + offset, size);
1782
1385
        if(last)
1783
526
            break;
1784
    }
1785
1786
526
    if (s->current_picture_ptr) {
1787
526
        if (last) {
1788
526
            if(r->loop_filter)
1789
526
                r->loop_filter(r, s->mb_height - 1);
1790
1791
526
            ret = finish_frame(avctx, pict);
1792
526
            if (ret < 0)
1793
                return ret;
1794
526
            *got_picture_ptr = ret;
1795
        } else if (HAVE_THREADS &&
1796
                   (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
1797
            av_log(avctx, AV_LOG_INFO, "marking unfished frame as finished\n");
1798
            /* always mark the current frame as finished, frame-mt supports
1799
             * only complete frames */
1800
            ff_er_frame_end(&s->er);
1801
            ff_mpv_frame_end(s);
1802
            s->mb_num_left = 0;
1803
            ff_thread_report_progress(&s->current_picture_ptr->tf, INT_MAX, 0);
1804
            return AVERROR_INVALIDDATA;
1805
        }
1806
    }
1807
1808
526
    return avpkt->size;
1809
}
1810
1811
9
av_cold int ff_rv34_decode_end(AVCodecContext *avctx)
1812
{
1813
9
    RV34DecContext *r = avctx->priv_data;
1814
1815
9
    ff_mpv_common_end(&r->s);
1816
9
    rv34_decoder_free(r);
1817
1818
9
    return 0;
1819
}