GCC Code Coverage Report
Directory: ../../../ffmpeg/ Exec Total Coverage
File: src/libavcodec/atrac3.c Lines: 292 451 64.7 %
Date: 2020-11-28 20:53:16 Branches: 132 234 56.4 %

Line Branch Exec Source
1
/*
2
 * ATRAC3 compatible decoder
3
 * Copyright (c) 2006-2008 Maxim Poliakovski
4
 * Copyright (c) 2006-2008 Benjamin Larsson
5
 *
6
 * This file is part of FFmpeg.
7
 *
8
 * FFmpeg is free software; you can redistribute it and/or
9
 * modify it under the terms of the GNU Lesser General Public
10
 * License as published by the Free Software Foundation; either
11
 * version 2.1 of the License, or (at your option) any later version.
12
 *
13
 * FFmpeg is distributed in the hope that it will be useful,
14
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
16
 * Lesser General Public License for more details.
17
 *
18
 * You should have received a copy of the GNU Lesser General Public
19
 * License along with FFmpeg; if not, write to the Free Software
20
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21
 */
22
23
/**
24
 * @file
25
 * ATRAC3 compatible decoder.
26
 * This decoder handles Sony's ATRAC3 data.
27
 *
28
 * Container formats used to store ATRAC3 data:
29
 * RealMedia (.rm), RIFF WAV (.wav, .at3), Sony OpenMG (.oma, .aa3).
30
 *
31
 * To use this decoder, a calling application must supply the extradata
32
 * bytes provided in the containers above.
33
 */
34
35
#include <math.h>
36
#include <stddef.h>
37
#include <stdio.h>
38
39
#include "libavutil/attributes.h"
40
#include "libavutil/float_dsp.h"
41
#include "libavutil/libm.h"
42
#include "avcodec.h"
43
#include "bytestream.h"
44
#include "fft.h"
45
#include "get_bits.h"
46
#include "internal.h"
47
48
#include "atrac.h"
49
#include "atrac3data.h"
50
51
#define MIN_CHANNELS    1
52
#define MAX_CHANNELS    8
53
#define MAX_JS_PAIRS    8 / 2
54
55
#define JOINT_STEREO    0x12
56
#define SINGLE          0x2
57
58
#define SAMPLES_PER_FRAME 1024
59
#define MDCT_SIZE          512
60
61
#define ATRAC3_VLC_BITS 8
62
63
typedef struct GainBlock {
64
    AtracGainInfo g_block[4];
65
} GainBlock;
66
67
typedef struct TonalComponent {
68
    int pos;
69
    int num_coefs;
70
    float coef[8];
71
} TonalComponent;
72
73
typedef struct ChannelUnit {
74
    int            bands_coded;
75
    int            num_components;
76
    float          prev_frame[SAMPLES_PER_FRAME];
77
    int            gc_blk_switch;
78
    TonalComponent components[64];
79
    GainBlock      gain_block[2];
80
81
    DECLARE_ALIGNED(32, float, spectrum)[SAMPLES_PER_FRAME];
82
    DECLARE_ALIGNED(32, float, imdct_buf)[SAMPLES_PER_FRAME];
83
84
    float          delay_buf1[46]; ///<qmf delay buffers
85
    float          delay_buf2[46];
86
    float          delay_buf3[46];
87
} ChannelUnit;
88
89
typedef struct ATRAC3Context {
90
    GetBitContext gb;
91
    //@{
92
    /** stream data */
93
    int coding_mode;
94
95
    ChannelUnit *units;
96
    //@}
97
    //@{
98
    /** joint-stereo related variables */
99
    int matrix_coeff_index_prev[MAX_JS_PAIRS][4];
100
    int matrix_coeff_index_now[MAX_JS_PAIRS][4];
101
    int matrix_coeff_index_next[MAX_JS_PAIRS][4];
102
    int weighting_delay[MAX_JS_PAIRS][6];
103
    //@}
104
    //@{
105
    /** data buffers */
106
    uint8_t *decoded_bytes_buffer;
107
    float temp_buf[1070];
108
    //@}
109
    //@{
110
    /** extradata */
111
    int scrambled_stream;
112
    //@}
113
114
    AtracGCContext    gainc_ctx;
115
    FFTContext        mdct_ctx;
116
    void (*vector_fmul)(float *dst, const float *src0, const float *src1,
117
                        int len);
118
} ATRAC3Context;
119
120
static DECLARE_ALIGNED(32, float, mdct_window)[MDCT_SIZE];
121
static VLC_TYPE atrac3_vlc_table[7 * 1 << ATRAC3_VLC_BITS][2];
122
static VLC   spectral_coeff_tab[7];
123
124
/**
125
 * Regular 512 points IMDCT without overlapping, with the exception of the
126
 * swapping of odd bands caused by the reverse spectra of the QMF.
127
 *
128
 * @param odd_band  1 if the band is an odd band
129
 */
130
2800
static void imlt(ATRAC3Context *q, float *input, float *output, int odd_band)
131
{
132
    int i;
133
134
2800
    if (odd_band) {
135
        /**
136
         * Reverse the odd bands before IMDCT, this is an effect of the QMF
137
         * transform or it gives better compression to do it this way.
138
         * FIXME: It should be possible to handle this in imdct_calc
139
         * for that to happen a modification of the prerotation step of
140
         * all SIMD code and C code is needed.
141
         * Or fix the functions before so they generate a pre reversed spectrum.
142
         */
143
109392
        for (i = 0; i < 128; i++)
144
108544
            FFSWAP(float, input[i], input[255 - i]);
145
    }
146
147
2800
    q->mdct_ctx.imdct_calc(&q->mdct_ctx, output, input);
148
149
    /* Perform windowing on the output. */
150
2800
    q->vector_fmul(output, output, mdct_window, MDCT_SIZE);
151
2800
}
152
153
/*
154
 * indata descrambling, only used for data coming from the rm container
155
 */
156
static int decode_bytes(const uint8_t *input, uint8_t *out, int bytes)
157
{
158
    int i, off;
159
    uint32_t c;
160
    const uint32_t *buf;
161
    uint32_t *output = (uint32_t *)out;
162
163
    off = (intptr_t)input & 3;
164
    buf = (const uint32_t *)(input - off);
165
    if (off)
166
        c = av_be2ne32((0x537F6103U >> (off * 8)) | (0x537F6103U << (32 - (off * 8))));
167
    else
168
        c = av_be2ne32(0x537F6103U);
169
    bytes += 3 + off;
170
    for (i = 0; i < bytes / 4; i++)
171
        output[i] = c ^ buf[i];
172
173
    if (off)
174
        avpriv_request_sample(NULL, "Offset of %d", off);
175
176
    return off;
177
}
178
179
4
static av_cold void init_imdct_window(void)
180
{
181
    int i, j;
182
183
    /* generate the mdct window, for details see
184
     * http://wiki.multimedia.cx/index.php?title=RealAudio_atrc#Windows */
185
516
    for (i = 0, j = 255; i < 128; i++, j--) {
186
512
        float wi = sin(((i + 0.5) / 256.0 - 0.5) * M_PI) + 1.0;
187
512
        float wj = sin(((j + 0.5) / 256.0 - 0.5) * M_PI) + 1.0;
188
512
        float w  = 0.5 * (wi * wi + wj * wj);
189
512
        mdct_window[i] = mdct_window[511 - i] = wi / w;
190
512
        mdct_window[j] = mdct_window[511 - j] = wj / w;
191
    }
192
4
}
193
194
7
static av_cold int atrac3_decode_close(AVCodecContext *avctx)
195
{
196
7
    ATRAC3Context *q = avctx->priv_data;
197
198
7
    av_freep(&q->units);
199
7
    av_freep(&q->decoded_bytes_buffer);
200
201
7
    ff_mdct_end(&q->mdct_ctx);
202
203
7
    return 0;
204
}
205
206
/**
207
 * Mantissa decoding
208
 *
209
 * @param selector     which table the output values are coded with
210
 * @param coding_flag  constant length coding or variable length coding
211
 * @param mantissas    mantissa output table
212
 * @param num_codes    number of values to get
213
 */
214
28209
static void read_quant_spectral_coeffs(GetBitContext *gb, int selector,
215
                                       int coding_flag, int *mantissas,
216
                                       int num_codes)
217
{
218
    int i, code, huff_symb;
219
220
28209
    if (selector == 1)
221
15517
        num_codes /= 2;
222
223
28209
    if (coding_flag != 0) {
224
        /* constant length coding (CLC) */
225
        int num_bits = clc_length_tab[selector];
226
227
        if (selector > 1) {
228
            for (i = 0; i < num_codes; i++) {
229
                if (num_bits)
230
                    code = get_sbits(gb, num_bits);
231
                else
232
                    code = 0;
233
                mantissas[i] = code;
234
            }
235
        } else {
236
            for (i = 0; i < num_codes; i++) {
237
                if (num_bits)
238
                    code = get_bits(gb, num_bits); // num_bits is always 4 in this case
239
                else
240
                    code = 0;
241
                mantissas[i * 2    ] = mantissa_clc_tab[code >> 2];
242
                mantissas[i * 2 + 1] = mantissa_clc_tab[code &  3];
243
            }
244
        }
245
    } else {
246
        /* variable length coding (VLC) */
247
28209
        if (selector != 1) {
248
192516
            for (i = 0; i < num_codes; i++) {
249
179824
                huff_symb = get_vlc2(gb, spectral_coeff_tab[selector-1].table,
250
                                     ATRAC3_VLC_BITS, 1);
251
179824
                huff_symb += 1;
252
179824
                code = huff_symb >> 1;
253
179824
                if (huff_symb & 1)
254
139733
                    code = -code;
255
179824
                mantissas[i] = code;
256
            }
257
        } else {
258
250405
            for (i = 0; i < num_codes; i++) {
259
234888
                huff_symb = get_vlc2(gb, spectral_coeff_tab[selector - 1].table,
260
                                     ATRAC3_VLC_BITS, 1);
261
234888
                mantissas[i * 2    ] = mantissa_vlc_tab[huff_symb * 2    ];
262
234888
                mantissas[i * 2 + 1] = mantissa_vlc_tab[huff_symb * 2 + 1];
263
            }
264
        }
265
    }
266
28209
}
267
268
/**
269
 * Restore the quantized band spectrum coefficients
270
 *
271
 * @return subband count, fix for broken specification/files
272
 */
273
1108
static int decode_spectrum(GetBitContext *gb, float *output)
274
{
275
    int num_subbands, coding_mode, i, j, first, last, subband_size;
276
    int subband_vlc_index[32], sf_index[32];
277
    int mantissas[128];
278
    float scale_factor;
279
280
1108
    num_subbands = get_bits(gb, 5);  // number of coded subbands
281
1108
    coding_mode  = get_bits1(gb);    // coding Mode: 0 - VLC/ 1-CLC
282
283
    /* get the VLC selector table for the subbands, 0 means not coded */
284
29317
    for (i = 0; i <= num_subbands; i++)
285
28209
        subband_vlc_index[i] = get_bits(gb, 3);
286
287
    /* read the scale factor indexes from the stream */
288
29317
    for (i = 0; i <= num_subbands; i++) {
289
28209
        if (subband_vlc_index[i] != 0)
290
28209
            sf_index[i] = get_bits(gb, 6);
291
    }
292
293
29317
    for (i = 0; i <= num_subbands; i++) {
294
28209
        first = subband_tab[i    ];
295
28209
        last  = subband_tab[i + 1];
296
297
28209
        subband_size = last - first;
298
299
28209
        if (subband_vlc_index[i] != 0) {
300
            /* decode spectral coefficients for this subband */
301
            /* TODO: This can be done faster is several blocks share the
302
             * same VLC selector (subband_vlc_index) */
303
28209
            read_quant_spectral_coeffs(gb, subband_vlc_index[i], coding_mode,
304
                                       mantissas, subband_size);
305
306
            /* decode the scale factor for this subband */
307
28209
            scale_factor = ff_atrac_sf_table[sf_index[i]] *
308
28209
                           inv_max_quant[subband_vlc_index[i]];
309
310
            /* inverse quantize the coefficients */
311
677809
            for (j = 0; first < last; first++, j++)
312
649600
                output[first] = mantissas[j] * scale_factor;
313
        } else {
314
            /* this subband was not coded, so zero the entire subband */
315
            memset(output + first, 0, subband_size * sizeof(*output));
316
        }
317
    }
318
319
    /* clear the subbands that were not coded */
320
1108
    first = subband_tab[i];
321
1108
    memset(output + first, 0, (SAMPLES_PER_FRAME - first) * sizeof(*output));
322
1108
    return num_subbands;
323
}
324
325
/**
326
 * Restore the quantized tonal components
327
 *
328
 * @param components tonal components
329
 * @param num_bands  number of coded bands
330
 */
331
1108
static int decode_tonal_components(GetBitContext *gb,
332
                                   TonalComponent *components, int num_bands)
333
{
334
    int i, b, c, m;
335
    int nb_components, coding_mode_selector, coding_mode;
336
    int band_flags[4], mantissa[8];
337
1108
    int component_count = 0;
338
339
1108
    nb_components = get_bits(gb, 5);
340
341
    /* no tonal components */
342
1108
    if (nb_components == 0)
343
1108
        return 0;
344
345
    coding_mode_selector = get_bits(gb, 2);
346
    if (coding_mode_selector == 2)
347
        return AVERROR_INVALIDDATA;
348
349
    coding_mode = coding_mode_selector & 1;
350
351
    for (i = 0; i < nb_components; i++) {
352
        int coded_values_per_component, quant_step_index;
353
354
        for (b = 0; b <= num_bands; b++)
355
            band_flags[b] = get_bits1(gb);
356
357
        coded_values_per_component = get_bits(gb, 3);
358
359
        quant_step_index = get_bits(gb, 3);
360
        if (quant_step_index <= 1)
361
            return AVERROR_INVALIDDATA;
362
363
        if (coding_mode_selector == 3)
364
            coding_mode = get_bits1(gb);
365
366
        for (b = 0; b < (num_bands + 1) * 4; b++) {
367
            int coded_components;
368
369
            if (band_flags[b >> 2] == 0)
370
                continue;
371
372
            coded_components = get_bits(gb, 3);
373
374
            for (c = 0; c < coded_components; c++) {
375
                TonalComponent *cmp = &components[component_count];
376
                int sf_index, coded_values, max_coded_values;
377
                float scale_factor;
378
379
                sf_index = get_bits(gb, 6);
380
                if (component_count >= 64)
381
                    return AVERROR_INVALIDDATA;
382
383
                cmp->pos = b * 64 + get_bits(gb, 6);
384
385
                max_coded_values = SAMPLES_PER_FRAME - cmp->pos;
386
                coded_values     = coded_values_per_component + 1;
387
                coded_values     = FFMIN(max_coded_values, coded_values);
388
389
                scale_factor = ff_atrac_sf_table[sf_index] *
390
                               inv_max_quant[quant_step_index];
391
392
                read_quant_spectral_coeffs(gb, quant_step_index, coding_mode,
393
                                           mantissa, coded_values);
394
395
                cmp->num_coefs = coded_values;
396
397
                /* inverse quant */
398
                for (m = 0; m < coded_values; m++)
399
                    cmp->coef[m] = mantissa[m] * scale_factor;
400
401
                component_count++;
402
            }
403
        }
404
    }
405
406
    return component_count;
407
}
408
409
/**
410
 * Decode gain parameters for the coded bands
411
 *
412
 * @param block      the gainblock for the current band
413
 * @param num_bands  amount of coded bands
414
 */
415
1108
static int decode_gain_control(GetBitContext *gb, GainBlock *block,
416
                               int num_bands)
417
{
418
    int b, j;
419
    int *level, *loc;
420
421
1108
    AtracGainInfo *gain = block->g_block;
422
423
3912
    for (b = 0; b <= num_bands; b++) {
424
2804
        gain[b].num_points = get_bits(gb, 3);
425
2804
        level              = gain[b].lev_code;
426
2804
        loc                = gain[b].loc_code;
427
428
3678
        for (j = 0; j < gain[b].num_points; j++) {
429
874
            level[j] = get_bits(gb, 4);
430
874
            loc[j]   = get_bits(gb, 5);
431

874
            if (j && loc[j] <= loc[j - 1])
432
                return AVERROR_INVALIDDATA;
433
        }
434
    }
435
436
    /* Clear the unused blocks. */
437
2736
    for (; b < 4 ; b++)
438
1628
        gain[b].num_points = 0;
439
440
1108
    return 0;
441
}
442
443
/**
444
 * Combine the tonal band spectrum and regular band spectrum
445
 *
446
 * @param spectrum        output spectrum buffer
447
 * @param num_components  number of tonal components
448
 * @param components      tonal components for this band
449
 * @return                position of the last tonal coefficient
450
 */
451
1108
static int add_tonal_components(float *spectrum, int num_components,
452
                                TonalComponent *components)
453
{
454
1108
    int i, j, last_pos = -1;
455
    float *input, *output;
456
457
1108
    for (i = 0; i < num_components; i++) {
458
        last_pos = FFMAX(components[i].pos + components[i].num_coefs, last_pos);
459
        input    = components[i].coef;
460
        output   = &spectrum[components[i].pos];
461
462
        for (j = 0; j < components[i].num_coefs; j++)
463
            output[j] += input[j];
464
    }
465
466
1108
    return last_pos;
467
}
468
469
#define INTERPOLATE(old, new, nsample) \
470
    ((old) + (nsample) * 0.125 * ((new) - (old)))
471
472
260
static void reverse_matrixing(float *su1, float *su2, int *prev_code,
473
                              int *curr_code)
474
{
475
    int i, nsample, band;
476
    float mc1_l, mc1_r, mc2_l, mc2_r;
477
478
1300
    for (i = 0, band = 0; band < 4 * 256; band += 256, i++) {
479
1040
        int s1 = prev_code[i];
480
1040
        int s2 = curr_code[i];
481
1040
        nsample = band;
482
483
1040
        if (s1 != s2) {
484
            /* Selector value changed, interpolation needed. */
485
38
            mc1_l = matrix_coeffs[s1 * 2    ];
486
38
            mc1_r = matrix_coeffs[s1 * 2 + 1];
487
38
            mc2_l = matrix_coeffs[s2 * 2    ];
488
38
            mc2_r = matrix_coeffs[s2 * 2 + 1];
489
490
            /* Interpolation is done over the first eight samples. */
491
342
            for (; nsample < band + 8; nsample++) {
492
304
                float c1 = su1[nsample];
493
304
                float c2 = su2[nsample];
494
304
                c2 = c1 * INTERPOLATE(mc1_l, mc2_l, nsample - band) +
495
304
                     c2 * INTERPOLATE(mc1_r, mc2_r, nsample - band);
496
304
                su1[nsample] = c2;
497
304
                su2[nsample] = c1 * 2.0 - c2;
498
            }
499
        }
500
501
        /* Apply the matrix without interpolation. */
502

1040
        switch (s2) {
503
45
        case 0:     /* M/S decoding */
504
11413
            for (; nsample < band + 256; nsample++) {
505
11368
                float c1 = su1[nsample];
506
11368
                float c2 = su2[nsample];
507
11368
                su1[nsample] =  c2       * 2.0;
508
11368
                su2[nsample] = (c1 - c2) * 2.0;
509
            }
510
45
            break;
511
        case 1:
512
            for (; nsample < band + 256; nsample++) {
513
                float c1 = su1[nsample];
514
                float c2 = su2[nsample];
515
                su1[nsample] = (c1 + c2) *  2.0;
516
                su2[nsample] =  c2       * -2.0;
517
            }
518
            break;
519
995
        case 2:
520
        case 3:
521
255563
            for (; nsample < band + 256; nsample++) {
522
254568
                float c1 = su1[nsample];
523
254568
                float c2 = su2[nsample];
524
254568
                su1[nsample] = c1 + c2;
525
254568
                su2[nsample] = c1 - c2;
526
            }
527
995
            break;
528
1040
        default:
529
            av_assert1(0);
530
        }
531
    }
532
260
}
533
534
492
static void get_channel_weights(int index, int flag, float ch[2])
535
{
536
492
    if (index == 7) {
537
53
        ch[0] = 1.0;
538
53
        ch[1] = 1.0;
539
    } else {
540
439
        ch[0] = (index & 7) / 7.0;
541
439
        ch[1] = sqrt(2 - ch[0] * ch[0]);
542
439
        if (flag)
543
115
            FFSWAP(float, ch[0], ch[1]);
544
    }
545
492
}
546
547
260
static void channel_weighting(float *su1, float *su2, int *p3)
548
{
549
    int band, nsample;
550
    /* w[x][y] y=0 is left y=1 is right */
551
    float w[2][2];
552
553

260
    if (p3[1] != 7 || p3[3] != 7) {
554
246
        get_channel_weights(p3[1], p3[0], w[0]);
555
246
        get_channel_weights(p3[3], p3[2], w[1]);
556
557
984
        for (band = 256; band < 4 * 256; band += 256) {
558
6642
            for (nsample = band; nsample < band + 8; nsample++) {
559
5904
                su1[nsample] *= INTERPOLATE(w[0][0], w[0][1], nsample - band);
560
5904
                su2[nsample] *= INTERPOLATE(w[1][0], w[1][1], nsample - band);
561
            }
562
183762
            for(; nsample < band + 256; nsample++) {
563
183024
                su1[nsample] *= w[1][0];
564
183024
                su2[nsample] *= w[1][1];
565
            }
566
        }
567
    }
568
260
}
569
570
/**
571
 * Decode a Sound Unit
572
 *
573
 * @param snd           the channel unit to be used
574
 * @param output        the decoded samples before IQMF in float representation
575
 * @param channel_num   channel number
576
 * @param coding_mode   the coding mode (JOINT_STEREO or single channels)
577
 */
578
1108
static int decode_channel_sound_unit(ATRAC3Context *q, GetBitContext *gb,
579
                                     ChannelUnit *snd, float *output,
580
                                     int channel_num, int coding_mode)
581
{
582
    int band, ret, num_subbands, last_tonal, num_bands;
583
1108
    GainBlock *gain1 = &snd->gain_block[    snd->gc_blk_switch];
584
1108
    GainBlock *gain2 = &snd->gain_block[1 - snd->gc_blk_switch];
585
586

1108
    if (coding_mode == JOINT_STEREO && (channel_num % 2) == 1) {
587
260
        if (get_bits(gb, 2) != 3) {
588
            av_log(NULL,AV_LOG_ERROR,"JS mono Sound Unit id != 3.\n");
589
            return AVERROR_INVALIDDATA;
590
        }
591
    } else {
592
848
        if (get_bits(gb, 6) != 0x28) {
593
            av_log(NULL,AV_LOG_ERROR,"Sound Unit id != 0x28.\n");
594
            return AVERROR_INVALIDDATA;
595
        }
596
    }
597
598
    /* number of coded QMF bands */
599
1108
    snd->bands_coded = get_bits(gb, 2);
600
601
1108
    ret = decode_gain_control(gb, gain2, snd->bands_coded);
602
1108
    if (ret)
603
        return ret;
604
605
1108
    snd->num_components = decode_tonal_components(gb, snd->components,
606
                                                  snd->bands_coded);
607
1108
    if (snd->num_components < 0)
608
        return snd->num_components;
609
610
1108
    num_subbands = decode_spectrum(gb, snd->spectrum);
611
612
    /* Merge the decoded spectrum and tonal components. */
613
1108
    last_tonal = add_tonal_components(snd->spectrum, snd->num_components,
614
1108
                                      snd->components);
615
616
617
    /* calculate number of used MLT/QMF bands according to the amount of coded
618
       spectral lines */
619
1108
    num_bands = (subband_tab[num_subbands] - 1) >> 8;
620
1108
    if (last_tonal >= 0)
621
        num_bands = FFMAX((last_tonal + 256) >> 8, num_bands);
622
623
624
    /* Reconstruct time domain samples. */
625
5540
    for (band = 0; band < 4; band++) {
626
        /* Perform the IMDCT step without overlapping. */
627
4432
        if (band <= num_bands)
628
2800
            imlt(q, &snd->spectrum[band * 256], snd->imdct_buf, band & 1);
629
        else
630
1632
            memset(snd->imdct_buf, 0, 512 * sizeof(*snd->imdct_buf));
631
632
        /* gain compensation and overlapping */
633
4432
        ff_atrac_gain_compensation(&q->gainc_ctx, snd->imdct_buf,
634
4432
                                   &snd->prev_frame[band * 256],
635
                                   &gain1->g_block[band], &gain2->g_block[band],
636
4432
                                   256, &output[band * 256]);
637
    }
638
639
    /* Swap the gain control buffers for the next frame. */
640
1108
    snd->gc_blk_switch ^= 1;
641
642
1108
    return 0;
643
}
644
645
554
static int decode_frame(AVCodecContext *avctx, const uint8_t *databuf,
646
                        float **out_samples)
647
{
648
554
    ATRAC3Context *q = avctx->priv_data;
649
    int ret, i, ch;
650
    uint8_t *ptr1;
651
652
554
    if (q->coding_mode == JOINT_STEREO) {
653
        /* channel coupling mode */
654
655
        /* Decode sound unit pairs (channels are expected to be even).
656
         * Multichannel joint stereo interleaves pairs (6ch: 2ch + 2ch + 2ch) */
657
        const uint8_t *js_databuf;
658
        int js_pair, js_block_align;
659
660
260
        js_block_align = (avctx->block_align / avctx->channels) * 2; /* block pair */
661
662
520
        for (ch = 0; ch < avctx->channels; ch = ch + 2) {
663
260
            js_pair = ch/2;
664
260
            js_databuf = databuf + js_pair * js_block_align; /* align to current pair */
665
666
            /* Set the bitstream reader at the start of first channel sound unit. */
667
260
            init_get_bits(&q->gb,
668
                          js_databuf, js_block_align * 8);
669
670
            /* decode Sound Unit 1 */
671
260
            ret = decode_channel_sound_unit(q, &q->gb, &q->units[ch],
672
260
                                            out_samples[ch], ch, JOINT_STEREO);
673
260
            if (ret != 0)
674
                return ret;
675
676
            /* Framedata of the su2 in the joint-stereo mode is encoded in
677
             * reverse byte order so we need to swap it first. */
678
260
            if (js_databuf == q->decoded_bytes_buffer) {
679
                uint8_t *ptr2 = q->decoded_bytes_buffer + js_block_align - 1;
680
                ptr1          = q->decoded_bytes_buffer;
681
                for (i = 0; i < js_block_align / 2; i++, ptr1++, ptr2--)
682
                    FFSWAP(uint8_t, *ptr1, *ptr2);
683
            } else {
684
260
                const uint8_t *ptr2 = js_databuf + js_block_align - 1;
685
50180
                for (i = 0; i < js_block_align; i++)
686
49920
                    q->decoded_bytes_buffer[i] = *ptr2--;
687
            }
688
689
            /* Skip the sync codes (0xF8). */
690
260
            ptr1 = q->decoded_bytes_buffer;
691
260
            for (i = 4; *ptr1 == 0xF8; i++, ptr1++) {
692
                if (i >= js_block_align)
693
                    return AVERROR_INVALIDDATA;
694
            }
695
696
697
            /* set the bitstream reader at the start of the second Sound Unit */
698
260
            ret = init_get_bits8(&q->gb,
699
260
                           ptr1, q->decoded_bytes_buffer + js_block_align - ptr1);
700
260
            if (ret < 0)
701
                return ret;
702
703
            /* Fill the Weighting coeffs delay buffer */
704
260
            memmove(q->weighting_delay[js_pair], &q->weighting_delay[js_pair][2],
705
                    4 * sizeof(*q->weighting_delay[js_pair]));
706
260
            q->weighting_delay[js_pair][4] = get_bits1(&q->gb);
707
260
            q->weighting_delay[js_pair][5] = get_bits(&q->gb, 3);
708
709
1300
            for (i = 0; i < 4; i++) {
710
1040
                q->matrix_coeff_index_prev[js_pair][i] = q->matrix_coeff_index_now[js_pair][i];
711
1040
                q->matrix_coeff_index_now[js_pair][i]  = q->matrix_coeff_index_next[js_pair][i];
712
1040
                q->matrix_coeff_index_next[js_pair][i] = get_bits(&q->gb, 2);
713
            }
714
715
            /* Decode Sound Unit 2. */
716
260
            ret = decode_channel_sound_unit(q, &q->gb, &q->units[ch+1],
717
260
                                            out_samples[ch+1], ch+1, JOINT_STEREO);
718
260
            if (ret != 0)
719
                return ret;
720
721
            /* Reconstruct the channel coefficients. */
722
260
            reverse_matrixing(out_samples[ch], out_samples[ch+1],
723
260
                              q->matrix_coeff_index_prev[js_pair],
724
260
                              q->matrix_coeff_index_now[js_pair]);
725
726
260
            channel_weighting(out_samples[ch], out_samples[ch+1], q->weighting_delay[js_pair]);
727
        }
728
    } else {
729
        /* single channels */
730
        /* Decode the channel sound units. */
731
882
        for (i = 0; i < avctx->channels; i++) {
732
            /* Set the bitstream reader at the start of a channel sound unit. */
733
588
            init_get_bits(&q->gb,
734
588
                          databuf + i * avctx->block_align / avctx->channels,
735
588
                          avctx->block_align * 8 / avctx->channels);
736
737
588
            ret = decode_channel_sound_unit(q, &q->gb, &q->units[i],
738
588
                                            out_samples[i], i, q->coding_mode);
739
588
            if (ret != 0)
740
                return ret;
741
        }
742
    }
743
744
    /* Apply the iQMF synthesis filter. */
745
1662
    for (i = 0; i < avctx->channels; i++) {
746
1108
        float *p1 = out_samples[i];
747
1108
        float *p2 = p1 + 256;
748
1108
        float *p3 = p2 + 256;
749
1108
        float *p4 = p3 + 256;
750
1108
        ff_atrac_iqmf(p1, p2, 256, p1, q->units[i].delay_buf1, q->temp_buf);
751
1108
        ff_atrac_iqmf(p4, p3, 256, p3, q->units[i].delay_buf2, q->temp_buf);
752
1108
        ff_atrac_iqmf(p1, p3, 512, p1, q->units[i].delay_buf3, q->temp_buf);
753
    }
754
755
554
    return 0;
756
}
757
758
static int al_decode_frame(AVCodecContext *avctx, const uint8_t *databuf,
759
                           int size, float **out_samples)
760
{
761
    ATRAC3Context *q = avctx->priv_data;
762
    int ret, i;
763
764
    /* Set the bitstream reader at the start of a channel sound unit. */
765
    init_get_bits(&q->gb, databuf, size * 8);
766
    /* single channels */
767
    /* Decode the channel sound units. */
768
    for (i = 0; i < avctx->channels; i++) {
769
        ret = decode_channel_sound_unit(q, &q->gb, &q->units[i],
770
                                        out_samples[i], i, q->coding_mode);
771
        if (ret != 0)
772
            return ret;
773
        while (i < avctx->channels && get_bits_left(&q->gb) > 6 && show_bits(&q->gb, 6) != 0x28) {
774
            skip_bits(&q->gb, 1);
775
        }
776
    }
777
778
    /* Apply the iQMF synthesis filter. */
779
    for (i = 0; i < avctx->channels; i++) {
780
        float *p1 = out_samples[i];
781
        float *p2 = p1 + 256;
782
        float *p3 = p2 + 256;
783
        float *p4 = p3 + 256;
784
        ff_atrac_iqmf(p1, p2, 256, p1, q->units[i].delay_buf1, q->temp_buf);
785
        ff_atrac_iqmf(p4, p3, 256, p3, q->units[i].delay_buf2, q->temp_buf);
786
        ff_atrac_iqmf(p1, p3, 512, p1, q->units[i].delay_buf3, q->temp_buf);
787
    }
788
789
    return 0;
790
}
791
792
557
static int atrac3_decode_frame(AVCodecContext *avctx, void *data,
793
                               int *got_frame_ptr, AVPacket *avpkt)
794
{
795
557
    AVFrame *frame     = data;
796
557
    const uint8_t *buf = avpkt->data;
797
557
    int buf_size = avpkt->size;
798
557
    ATRAC3Context *q = avctx->priv_data;
799
    int ret;
800
    const uint8_t *databuf;
801
802
557
    if (buf_size < avctx->block_align) {
803
3
        av_log(avctx, AV_LOG_ERROR,
804
               "Frame too small (%d bytes). Truncated file?\n", buf_size);
805
3
        return AVERROR_INVALIDDATA;
806
    }
807
808
    /* get output buffer */
809
554
    frame->nb_samples = SAMPLES_PER_FRAME;
810
554
    if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
811
        return ret;
812
813
    /* Check if we need to descramble and what buffer to pass on. */
814
554
    if (q->scrambled_stream) {
815
        decode_bytes(buf, q->decoded_bytes_buffer, avctx->block_align);
816
        databuf = q->decoded_bytes_buffer;
817
    } else {
818
554
        databuf = buf;
819
    }
820
821
554
    ret = decode_frame(avctx, databuf, (float **)frame->extended_data);
822
554
    if (ret) {
823
        av_log(avctx, AV_LOG_ERROR, "Frame decoding error!\n");
824
        return ret;
825
    }
826
827
554
    *got_frame_ptr = 1;
828
829
554
    return avctx->block_align;
830
}
831
832
static int atrac3al_decode_frame(AVCodecContext *avctx, void *data,
833
                                 int *got_frame_ptr, AVPacket *avpkt)
834
{
835
    AVFrame *frame = data;
836
    int ret;
837
838
    frame->nb_samples = SAMPLES_PER_FRAME;
839
    if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
840
        return ret;
841
842
    ret = al_decode_frame(avctx, avpkt->data, avpkt->size,
843
                          (float **)frame->extended_data);
844
    if (ret) {
845
        av_log(avctx, AV_LOG_ERROR, "Frame decoding error!\n");
846
        return ret;
847
    }
848
849
    *got_frame_ptr = 1;
850
851
    return avpkt->size;
852
}
853
854
4
static av_cold void atrac3_init_static_data(void)
855
{
856
4
    VLC_TYPE (*table)[2] = atrac3_vlc_table;
857
    int i;
858
859
4
    init_imdct_window();
860
4
    ff_atrac_generate_tables();
861
862
    /* Initialize the VLC tables. */
863
32
    for (i = 0; i < 7; i++) {
864
28
        spectral_coeff_tab[i].table           = table;
865
28
        spectral_coeff_tab[i].table_allocated = 256;
866
28
        init_vlc(&spectral_coeff_tab[i], ATRAC3_VLC_BITS, huff_tab_sizes[i],
867
                 huff_bits[i],  1, 1,
868
                 huff_codes[i], 1, 1, INIT_VLC_USE_NEW_STATIC);
869
28
        table += 256;
870
    }
871
4
}
872
873
7
static av_cold int atrac3_decode_init(AVCodecContext *avctx)
874
{
875
    static int static_init_done;
876
    int i, js_pair, ret;
877
    int version, delay, samples_per_frame, frame_factor;
878
7
    const uint8_t *edata_ptr = avctx->extradata;
879
7
    ATRAC3Context *q = avctx->priv_data;
880
    AVFloatDSPContext *fdsp;
881
882

7
    if (avctx->channels < MIN_CHANNELS || avctx->channels > MAX_CHANNELS) {
883
        av_log(avctx, AV_LOG_ERROR, "Channel configuration error!\n");
884
        return AVERROR(EINVAL);
885
    }
886
887
7
    if (!static_init_done)
888
4
        atrac3_init_static_data();
889
7
    static_init_done = 1;
890
891
    /* Take care of the codec-specific extradata. */
892
7
    if (avctx->codec_id == AV_CODEC_ID_ATRAC3AL) {
893
        version           = 4;
894
        samples_per_frame = SAMPLES_PER_FRAME * avctx->channels;
895
        delay             = 0x88E;
896
        q->coding_mode    = SINGLE;
897
7
    } else if (avctx->extradata_size == 14) {
898
        /* Parse the extradata, WAV format */
899
7
        av_log(avctx, AV_LOG_DEBUG, "[0-1] %d\n",
900
               bytestream_get_le16(&edata_ptr));  // Unknown value always 1
901
7
        edata_ptr += 4;                             // samples per channel
902
7
        q->coding_mode = bytestream_get_le16(&edata_ptr);
903
7
        av_log(avctx, AV_LOG_DEBUG,"[8-9] %d\n",
904
               bytestream_get_le16(&edata_ptr));  //Dupe of coding mode
905
7
        frame_factor = bytestream_get_le16(&edata_ptr);  // Unknown always 1
906
7
        av_log(avctx, AV_LOG_DEBUG,"[12-13] %d\n",
907
               bytestream_get_le16(&edata_ptr));  // Unknown always 0
908
909
        /* setup */
910
7
        samples_per_frame    = SAMPLES_PER_FRAME * avctx->channels;
911
7
        version              = 4;
912
7
        delay                = 0x88E;
913
7
        q->coding_mode       = q->coding_mode ? JOINT_STEREO : SINGLE;
914
7
        q->scrambled_stream  = 0;
915
916
7
        if (avctx->block_align !=  96 * avctx->channels * frame_factor &&
917
5
            avctx->block_align != 152 * avctx->channels * frame_factor &&
918
3
            avctx->block_align != 192 * avctx->channels * frame_factor) {
919
            av_log(avctx, AV_LOG_ERROR, "Unknown frame/channel/frame_factor "
920
                   "configuration %d/%d/%d\n", avctx->block_align,
921
                   avctx->channels, frame_factor);
922
            return AVERROR_INVALIDDATA;
923
        }
924
    } else if (avctx->extradata_size == 12 || avctx->extradata_size == 10) {
925
        /* Parse the extradata, RM format. */
926
        version                = bytestream_get_be32(&edata_ptr);
927
        samples_per_frame      = bytestream_get_be16(&edata_ptr);
928
        delay                  = bytestream_get_be16(&edata_ptr);
929
        q->coding_mode         = bytestream_get_be16(&edata_ptr);
930
        q->scrambled_stream    = 1;
931
932
    } else {
933
        av_log(avctx, AV_LOG_ERROR, "Unknown extradata size %d.\n",
934
               avctx->extradata_size);
935
        return AVERROR(EINVAL);
936
    }
937
938
    /* Check the extradata */
939
940
7
    if (version != 4) {
941
        av_log(avctx, AV_LOG_ERROR, "Version %d != 4.\n", version);
942
        return AVERROR_INVALIDDATA;
943
    }
944
945
7
    if (samples_per_frame != SAMPLES_PER_FRAME * avctx->channels) {
946
        av_log(avctx, AV_LOG_ERROR, "Unknown amount of samples per frame %d.\n",
947
               samples_per_frame);
948
        return AVERROR_INVALIDDATA;
949
    }
950
951
7
    if (delay != 0x88E) {
952
        av_log(avctx, AV_LOG_ERROR, "Unknown amount of delay %x != 0x88E.\n",
953
               delay);
954
        return AVERROR_INVALIDDATA;
955
    }
956
957
7
    if (q->coding_mode == SINGLE)
958
5
        av_log(avctx, AV_LOG_DEBUG, "Single channels detected.\n");
959
2
    else if (q->coding_mode == JOINT_STEREO) {
960
2
        if (avctx->channels % 2 == 1) { /* Joint stereo channels must be even */
961
            av_log(avctx, AV_LOG_ERROR, "Invalid joint stereo channel configuration.\n");
962
            return AVERROR_INVALIDDATA;
963
        }
964
2
        av_log(avctx, AV_LOG_DEBUG, "Joint stereo detected.\n");
965
    } else {
966
        av_log(avctx, AV_LOG_ERROR, "Unknown channel coding mode %x!\n",
967
               q->coding_mode);
968
        return AVERROR_INVALIDDATA;
969
    }
970
971

7
    if (avctx->block_align > 4096 || avctx->block_align <= 0)
972
        return AVERROR(EINVAL);
973
974
7
    q->decoded_bytes_buffer = av_mallocz(FFALIGN(avctx->block_align, 4) +
975
                                         AV_INPUT_BUFFER_PADDING_SIZE);
976
7
    if (!q->decoded_bytes_buffer)
977
        return AVERROR(ENOMEM);
978
979
7
    avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
980
981
    /* initialize the MDCT transform */
982
7
    if ((ret = ff_mdct_init(&q->mdct_ctx, 9, 1, 1.0 / 32768)) < 0) {
983
        av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n");
984
        return ret;
985
    }
986
987
    /* init the joint-stereo decoding data */
988
35
    for (js_pair = 0; js_pair < MAX_JS_PAIRS; js_pair++) {
989
28
        q->weighting_delay[js_pair][0] = 0;
990
28
        q->weighting_delay[js_pair][1] = 7;
991
28
        q->weighting_delay[js_pair][2] = 0;
992
28
        q->weighting_delay[js_pair][3] = 7;
993
28
        q->weighting_delay[js_pair][4] = 0;
994
28
        q->weighting_delay[js_pair][5] = 7;
995
996
140
        for (i = 0; i < 4; i++) {
997
112
            q->matrix_coeff_index_prev[js_pair][i] = 3;
998
112
            q->matrix_coeff_index_now[js_pair][i]  = 3;
999
112
            q->matrix_coeff_index_next[js_pair][i] = 3;
1000
        }
1001
    }
1002
1003
7
    ff_atrac_init_gain_compensation(&q->gainc_ctx, 4, 3);
1004
7
    fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
1005
7
    if (!fdsp)
1006
        return AVERROR(ENOMEM);
1007
7
    q->vector_fmul = fdsp->vector_fmul;
1008
7
    av_free(fdsp);
1009
1010
7
    q->units = av_mallocz_array(avctx->channels, sizeof(*q->units));
1011
7
    if (!q->units)
1012
        return AVERROR(ENOMEM);
1013
1014
7
    return 0;
1015
}
1016
1017
AVCodec ff_atrac3_decoder = {
1018
    .name             = "atrac3",
1019
    .long_name        = NULL_IF_CONFIG_SMALL("ATRAC3 (Adaptive TRansform Acoustic Coding 3)"),
1020
    .type             = AVMEDIA_TYPE_AUDIO,
1021
    .id               = AV_CODEC_ID_ATRAC3,
1022
    .priv_data_size   = sizeof(ATRAC3Context),
1023
    .init             = atrac3_decode_init,
1024
    .close            = atrac3_decode_close,
1025
    .decode           = atrac3_decode_frame,
1026
    .capabilities     = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1,
1027
    .sample_fmts      = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1028
                                                        AV_SAMPLE_FMT_NONE },
1029
    .caps_internal    = FF_CODEC_CAP_INIT_CLEANUP,
1030
};
1031
1032
AVCodec ff_atrac3al_decoder = {
1033
    .name             = "atrac3al",
1034
    .long_name        = NULL_IF_CONFIG_SMALL("ATRAC3 AL (Adaptive TRansform Acoustic Coding 3 Advanced Lossless)"),
1035
    .type             = AVMEDIA_TYPE_AUDIO,
1036
    .id               = AV_CODEC_ID_ATRAC3AL,
1037
    .priv_data_size   = sizeof(ATRAC3Context),
1038
    .init             = atrac3_decode_init,
1039
    .close            = atrac3_decode_close,
1040
    .decode           = atrac3al_decode_frame,
1041
    .capabilities     = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1,
1042
    .sample_fmts      = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1043
                                                        AV_SAMPLE_FMT_NONE },
1044
    .caps_internal    = FF_CODEC_CAP_INIT_CLEANUP,
1045
};