Statistics
| Branch: | Revision:

ffmpeg / libavcodec / wma.c @ fe2ff6d2

History | View | Annotate | Download (16.5 KB)

1
/*
2
 * WMA compatible codec
3
 * Copyright (c) 2002-2007 The FFmpeg Project
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
#include "avcodec.h"
23
#include "wma.h"
24
#include "wmadata.h"
25

    
26
#undef NDEBUG
27
#include <assert.h>
28

    
29
/* XXX: use same run/length optimization as mpeg decoders */
30
//FIXME maybe split decode / encode or pass flag
31
static void init_coef_vlc(VLC *vlc, uint16_t **prun_table,
32
                          float **plevel_table, uint16_t **pint_table,
33
                          const CoefVLCTable *vlc_table)
34
{
35
    int n = vlc_table->n;
36
    const uint8_t  *table_bits   = vlc_table->huffbits;
37
    const uint32_t *table_codes  = vlc_table->huffcodes;
38
    const uint16_t *levels_table = vlc_table->levels;
39
    uint16_t *run_table, *level_table, *int_table;
40
    float *flevel_table;
41
    int i, l, j, k, level;
42

    
43
    init_vlc(vlc, VLCBITS, n, table_bits, 1, 1, table_codes, 4, 4, 0);
44

    
45
    run_table   = av_malloc(n * sizeof(uint16_t));
46
    level_table = av_malloc(n * sizeof(uint16_t));
47
    flevel_table= av_malloc(n * sizeof(*flevel_table));
48
    int_table   = av_malloc(n * sizeof(uint16_t));
49
    i = 2;
50
    level = 1;
51
    k = 0;
52
    while (i < n) {
53
        int_table[k] = i;
54
        l = levels_table[k++];
55
        for (j = 0; j < l; j++) {
56
            run_table[i]   = j;
57
            level_table[i] = level;
58
            flevel_table[i]= level;
59
            i++;
60
        }
61
        level++;
62
    }
63
    *prun_table   = run_table;
64
    *plevel_table = flevel_table;
65
    *pint_table   = int_table;
66
    av_free(level_table);
67
}
68

    
69
/**
70
 *@brief Get the samples per frame for this stream.
71
 *@param sample_rate output sample_rate
72
 *@param version wma version
73
 *@param decode_flags codec compression features
74
 *@return log2 of the number of output samples per frame
75
 */
76
int av_cold ff_wma_get_frame_len_bits(int sample_rate, int version,
77
                                      unsigned int decode_flags)
78
{
79

    
80
    int frame_len_bits;
81

    
82
    if (sample_rate <= 16000) {
83
        frame_len_bits = 9;
84
    } else if (sample_rate <= 22050 ||
85
             (sample_rate <= 32000 && version == 1)) {
86
        frame_len_bits = 10;
87
    } else if (sample_rate <= 48000) {
88
        frame_len_bits = 11;
89
    } else if (sample_rate <= 96000) {
90
        frame_len_bits = 12;
91
    } else {
92
        frame_len_bits = 13;
93
    }
94

    
95
    if (version == 3) {
96
        int tmp = decode_flags & 0x6;
97
        if (tmp == 0x2) {
98
            ++frame_len_bits;
99
        } else if (tmp == 0x4) {
100
            --frame_len_bits;
101
        } else if (tmp == 0x6) {
102
            frame_len_bits -= 2;
103
        }
104
    }
105

    
106
    return frame_len_bits;
107
}
108

    
109
int ff_wma_init(AVCodecContext *avctx, int flags2)
110
{
111
    WMACodecContext *s = avctx->priv_data;
112
    int i;
113
    float bps1, high_freq;
114
    volatile float bps;
115
    int sample_rate1;
116
    int coef_vlc_table;
117

    
118
    if (   avctx->sample_rate <= 0 || avctx->sample_rate > 50000
119
        || avctx->channels    <= 0 || avctx->channels    > 8
120
        || avctx->bit_rate    <= 0)
121
        return -1;
122

    
123
    s->sample_rate = avctx->sample_rate;
124
    s->nb_channels = avctx->channels;
125
    s->bit_rate    = avctx->bit_rate;
126
    s->block_align = avctx->block_align;
127

    
128
    dsputil_init(&s->dsp, avctx);
129
    ff_fmt_convert_init(&s->fmt_conv, avctx);
130

    
131
    if (avctx->codec->id == CODEC_ID_WMAV1) {
132
        s->version = 1;
133
    } else {
134
        s->version = 2;
135
    }
136

    
137
    /* compute MDCT block size */
138
    s->frame_len_bits = ff_wma_get_frame_len_bits(s->sample_rate, s->version, 0);
139

    
140
    s->frame_len = 1 << s->frame_len_bits;
141
    if (s->use_variable_block_len) {
142
        int nb_max, nb;
143
        nb = ((flags2 >> 3) & 3) + 1;
144
        if ((s->bit_rate / s->nb_channels) >= 32000)
145
            nb += 2;
146
        nb_max = s->frame_len_bits - BLOCK_MIN_BITS;
147
        if (nb > nb_max)
148
            nb = nb_max;
149
        s->nb_block_sizes = nb + 1;
150
    } else {
151
        s->nb_block_sizes = 1;
152
    }
153

    
154
    /* init rate dependent parameters */
155
    s->use_noise_coding = 1;
156
    high_freq = s->sample_rate * 0.5;
157

    
158
    /* if version 2, then the rates are normalized */
159
    sample_rate1 = s->sample_rate;
160
    if (s->version == 2) {
161
        if (sample_rate1 >= 44100) {
162
            sample_rate1 = 44100;
163
        } else if (sample_rate1 >= 22050) {
164
            sample_rate1 = 22050;
165
        } else if (sample_rate1 >= 16000) {
166
            sample_rate1 = 16000;
167
        } else if (sample_rate1 >= 11025) {
168
            sample_rate1 = 11025;
169
        } else if (sample_rate1 >= 8000) {
170
            sample_rate1 = 8000;
171
        }
172
    }
173

    
174
    bps = (float)s->bit_rate / (float)(s->nb_channels * s->sample_rate);
175
    s->byte_offset_bits = av_log2((int)(bps * s->frame_len / 8.0 + 0.5)) + 2;
176

    
177
    /* compute high frequency value and choose if noise coding should
178
       be activated */
179
    bps1 = bps;
180
    if (s->nb_channels == 2)
181
        bps1 = bps * 1.6;
182
    if (sample_rate1 == 44100) {
183
        if (bps1 >= 0.61) {
184
            s->use_noise_coding = 0;
185
        } else {
186
            high_freq = high_freq * 0.4;
187
        }
188
    } else if (sample_rate1 == 22050) {
189
        if (bps1 >= 1.16) {
190
            s->use_noise_coding = 0;
191
        } else if (bps1 >= 0.72) {
192
            high_freq = high_freq * 0.7;
193
        } else {
194
            high_freq = high_freq * 0.6;
195
        }
196
    } else if (sample_rate1 == 16000) {
197
        if (bps > 0.5) {
198
            high_freq = high_freq * 0.5;
199
        } else {
200
            high_freq = high_freq * 0.3;
201
        }
202
    } else if (sample_rate1 == 11025) {
203
        high_freq = high_freq * 0.7;
204
    } else if (sample_rate1 == 8000) {
205
        if (bps <= 0.625) {
206
            high_freq = high_freq * 0.5;
207
        } else if (bps > 0.75) {
208
            s->use_noise_coding = 0;
209
        } else {
210
            high_freq = high_freq * 0.65;
211
        }
212
    } else {
213
        if (bps >= 0.8) {
214
            high_freq = high_freq * 0.75;
215
        } else if (bps >= 0.6) {
216
            high_freq = high_freq * 0.6;
217
        } else {
218
            high_freq = high_freq * 0.5;
219
        }
220
    }
221
    av_dlog(s->avctx, "flags2=0x%x\n", flags2);
222
    av_dlog(s->avctx, "version=%d channels=%d sample_rate=%d bitrate=%d block_align=%d\n",
223
            s->version, s->nb_channels, s->sample_rate, s->bit_rate,
224
            s->block_align);
225
    av_dlog(s->avctx, "bps=%f bps1=%f high_freq=%f bitoffset=%d\n",
226
            bps, bps1, high_freq, s->byte_offset_bits);
227
    av_dlog(s->avctx, "use_noise_coding=%d use_exp_vlc=%d nb_block_sizes=%d\n",
228
            s->use_noise_coding, s->use_exp_vlc, s->nb_block_sizes);
229

    
230
    /* compute the scale factor band sizes for each MDCT block size */
231
    {
232
        int a, b, pos, lpos, k, block_len, i, j, n;
233
        const uint8_t *table;
234

    
235
        if (s->version == 1) {
236
            s->coefs_start = 3;
237
        } else {
238
            s->coefs_start = 0;
239
        }
240
        for (k = 0; k < s->nb_block_sizes; k++) {
241
            block_len = s->frame_len >> k;
242

    
243
            if (s->version == 1) {
244
                lpos = 0;
245
                for (i = 0; i < 25; i++) {
246
                    a = ff_wma_critical_freqs[i];
247
                    b = s->sample_rate;
248
                    pos = ((block_len * 2 * a) + (b >> 1)) / b;
249
                    if (pos > block_len)
250
                        pos = block_len;
251
                    s->exponent_bands[0][i] = pos - lpos;
252
                    if (pos >= block_len) {
253
                        i++;
254
                        break;
255
                    }
256
                    lpos = pos;
257
                }
258
                s->exponent_sizes[0] = i;
259
            } else {
260
                /* hardcoded tables */
261
                table = NULL;
262
                a = s->frame_len_bits - BLOCK_MIN_BITS - k;
263
                if (a < 3) {
264
                    if (s->sample_rate >= 44100) {
265
                        table = exponent_band_44100[a];
266
                    } else if (s->sample_rate >= 32000) {
267
                        table = exponent_band_32000[a];
268
                    } else if (s->sample_rate >= 22050) {
269
                        table = exponent_band_22050[a];
270
                    }
271
                }
272
                if (table) {
273
                    n = *table++;
274
                    for (i = 0; i < n; i++)
275
                        s->exponent_bands[k][i] = table[i];
276
                    s->exponent_sizes[k] = n;
277
                } else {
278
                    j = 0;
279
                    lpos = 0;
280
                    for (i = 0; i < 25; i++) {
281
                        a = ff_wma_critical_freqs[i];
282
                        b = s->sample_rate;
283
                        pos = ((block_len * 2 * a) + (b << 1)) / (4 * b);
284
                        pos <<= 2;
285
                        if (pos > block_len)
286
                            pos = block_len;
287
                        if (pos > lpos)
288
                            s->exponent_bands[k][j++] = pos - lpos;
289
                        if (pos >= block_len)
290
                            break;
291
                        lpos = pos;
292
                    }
293
                    s->exponent_sizes[k] = j;
294
                }
295
            }
296

    
297
            /* max number of coefs */
298
            s->coefs_end[k] = (s->frame_len - ((s->frame_len * 9) / 100)) >> k;
299
            /* high freq computation */
300
            s->high_band_start[k] = (int)((block_len * 2 * high_freq) /
301
                                          s->sample_rate + 0.5);
302
            n = s->exponent_sizes[k];
303
            j = 0;
304
            pos = 0;
305
            for (i = 0; i < n; i++) {
306
                int start, end;
307
                start = pos;
308
                pos += s->exponent_bands[k][i];
309
                end = pos;
310
                if (start < s->high_band_start[k])
311
                    start = s->high_band_start[k];
312
                if (end > s->coefs_end[k])
313
                    end = s->coefs_end[k];
314
                if (end > start)
315
                    s->exponent_high_bands[k][j++] = end - start;
316
            }
317
            s->exponent_high_sizes[k] = j;
318
#if 0
319
            tprintf(s->avctx, "%5d: coefs_end=%d high_band_start=%d nb_high_bands=%d: ",
320
                    s->frame_len >> k,
321
                    s->coefs_end[k],
322
                    s->high_band_start[k],
323
                    s->exponent_high_sizes[k]);
324
            for (j = 0; j < s->exponent_high_sizes[k]; j++)
325
                tprintf(s->avctx, " %d", s->exponent_high_bands[k][j]);
326
            tprintf(s->avctx, "\n");
327
#endif
328
        }
329
    }
330

    
331
#ifdef TRACE
332
    {
333
        int i, j;
334
        for (i = 0; i < s->nb_block_sizes; i++) {
335
            tprintf(s->avctx, "%5d: n=%2d:",
336
                    s->frame_len >> i,
337
                    s->exponent_sizes[i]);
338
            for (j = 0; j < s->exponent_sizes[i]; j++)
339
                tprintf(s->avctx, " %d", s->exponent_bands[i][j]);
340
            tprintf(s->avctx, "\n");
341
        }
342
    }
343
#endif
344

    
345
    /* init MDCT windows : simple sinus window */
346
    for (i = 0; i < s->nb_block_sizes; i++) {
347
        ff_init_ff_sine_windows(s->frame_len_bits - i);
348
        s->windows[i] = ff_sine_windows[s->frame_len_bits - i];
349
    }
350

    
351
    s->reset_block_lengths = 1;
352

    
353
    if (s->use_noise_coding) {
354

    
355
        /* init the noise generator */
356
        if (s->use_exp_vlc) {
357
            s->noise_mult = 0.02;
358
        } else {
359
            s->noise_mult = 0.04;
360
        }
361

    
362
#ifdef TRACE
363
        for (i = 0; i < NOISE_TAB_SIZE; i++)
364
            s->noise_table[i] = 1.0 * s->noise_mult;
365
#else
366
        {
367
            unsigned int seed;
368
            float norm;
369
            seed = 1;
370
            norm = (1.0 / (float)(1LL << 31)) * sqrt(3) * s->noise_mult;
371
            for (i = 0; i < NOISE_TAB_SIZE; i++) {
372
                seed = seed * 314159 + 1;
373
                s->noise_table[i] = (float)((int)seed) * norm;
374
            }
375
        }
376
#endif
377
    }
378

    
379
    /* choose the VLC tables for the coefficients */
380
    coef_vlc_table = 2;
381
    if (s->sample_rate >= 32000) {
382
        if (bps1 < 0.72) {
383
            coef_vlc_table = 0;
384
        } else if (bps1 < 1.16) {
385
            coef_vlc_table = 1;
386
        }
387
    }
388
    s->coef_vlcs[0]= &coef_vlcs[coef_vlc_table * 2    ];
389
    s->coef_vlcs[1]= &coef_vlcs[coef_vlc_table * 2 + 1];
390
    init_coef_vlc(&s->coef_vlc[0], &s->run_table[0], &s->level_table[0], &s->int_table[0],
391
                  s->coef_vlcs[0]);
392
    init_coef_vlc(&s->coef_vlc[1], &s->run_table[1], &s->level_table[1], &s->int_table[1],
393
                  s->coef_vlcs[1]);
394

    
395
    return 0;
396
}
397

    
398
int ff_wma_total_gain_to_bits(int total_gain)
399
{
400
         if (total_gain < 15) return 13;
401
    else if (total_gain < 32) return 12;
402
    else if (total_gain < 40) return 11;
403
    else if (total_gain < 45) return 10;
404
    else                      return  9;
405
}
406

    
407
int ff_wma_end(AVCodecContext *avctx)
408
{
409
    WMACodecContext *s = avctx->priv_data;
410
    int i;
411

    
412
    for (i = 0; i < s->nb_block_sizes; i++)
413
        ff_mdct_end(&s->mdct_ctx[i]);
414

    
415
    if (s->use_exp_vlc) {
416
        free_vlc(&s->exp_vlc);
417
    }
418
    if (s->use_noise_coding) {
419
        free_vlc(&s->hgain_vlc);
420
    }
421
    for (i = 0; i < 2; i++) {
422
        free_vlc(&s->coef_vlc[i]);
423
        av_free(s->run_table[i]);
424
        av_free(s->level_table[i]);
425
        av_free(s->int_table[i]);
426
    }
427

    
428
    return 0;
429
}
430

    
431
/**
432
 * Decode an uncompressed coefficient.
433
 * @param gb GetBitContext
434
 * @return the decoded coefficient
435
 */
436
unsigned int ff_wma_get_large_val(GetBitContext* gb)
437
{
438
    /** consumes up to 34 bits */
439
    int n_bits = 8;
440
    /** decode length */
441
    if (get_bits1(gb)) {
442
        n_bits += 8;
443
        if (get_bits1(gb)) {
444
            n_bits += 8;
445
            if (get_bits1(gb)) {
446
                n_bits += 7;
447
            }
448
        }
449
    }
450
    return get_bits_long(gb, n_bits);
451
}
452

    
453
/**
454
 * Decode run level compressed coefficients.
455
 * @param avctx codec context
456
 * @param gb bitstream reader context
457
 * @param vlc vlc table for get_vlc2
458
 * @param level_table level codes
459
 * @param run_table run codes
460
 * @param version 0 for wma1,2 1 for wmapro
461
 * @param ptr output buffer
462
 * @param offset offset in the output buffer
463
 * @param num_coefs number of input coefficents
464
 * @param block_len input buffer length (2^n)
465
 * @param frame_len_bits number of bits for escaped run codes
466
 * @param coef_nb_bits number of bits for escaped level codes
467
 * @return 0 on success, -1 otherwise
468
 */
469
int ff_wma_run_level_decode(AVCodecContext* avctx, GetBitContext* gb,
470
                            VLC *vlc,
471
                            const float *level_table, const uint16_t *run_table,
472
                            int version, WMACoef *ptr, int offset,
473
                            int num_coefs, int block_len, int frame_len_bits,
474
                            int coef_nb_bits)
475
{
476
    int code, level, sign;
477
    const uint32_t *ilvl = (const uint32_t*)level_table;
478
    uint32_t *iptr = (uint32_t*)ptr;
479
    const unsigned int coef_mask = block_len - 1;
480
    for (; offset < num_coefs; offset++) {
481
        code = get_vlc2(gb, vlc->table, VLCBITS, VLCMAX);
482
        if (code > 1) {
483
            /** normal code */
484
            offset += run_table[code];
485
            sign = get_bits1(gb) - 1;
486
            iptr[offset & coef_mask] = ilvl[code] ^ sign<<31;
487
        } else if (code == 1) {
488
            /** EOB */
489
            break;
490
        } else {
491
            /** escape */
492
            if (!version) {
493
                level = get_bits(gb, coef_nb_bits);
494
                /** NOTE: this is rather suboptimal. reading
495
                    block_len_bits would be better */
496
                offset += get_bits(gb, frame_len_bits);
497
            } else {
498
                level = ff_wma_get_large_val(gb);
499
                /** escape decode */
500
                if (get_bits1(gb)) {
501
                    if (get_bits1(gb)) {
502
                        if (get_bits1(gb)) {
503
                            av_log(avctx,AV_LOG_ERROR,
504
                                "broken escape sequence\n");
505
                            return -1;
506
                        } else
507
                            offset += get_bits(gb, frame_len_bits) + 4;
508
                    } else
509
                        offset += get_bits(gb, 2) + 1;
510
                }
511
            }
512
            sign = get_bits1(gb) - 1;
513
            ptr[offset & coef_mask] = (level^sign) - sign;
514
        }
515
    }
516
    /** NOTE: EOB can be omitted */
517
    if (offset > num_coefs) {
518
        av_log(avctx, AV_LOG_ERROR, "overflow in spectral RLE, ignoring\n");
519
        return -1;
520
    }
521

    
522
    return 0;
523
}
524