summaryrefslogtreecommitdiff
path: root/libavcodec/twinvq.c
blob: 08a7a9f4e881a087378d4b91196ae292eb0c0e68 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
/*
 * TwinVQ decoder
 * Copyright (c) 2009 Vitor Sessak
 *
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * FFmpeg is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with FFmpeg; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

#include <math.h>
#include <stdint.h>

#include "libavutil/channel_layout.h"
#include "libavutil/float_dsp.h"
#include "avcodec.h"
#include "fft.h"
#include "internal.h"
#include "lsp.h"
#include "sinewin.h"
#include "twinvq.h"

/**
 * Evaluate a single LPC amplitude spectrum envelope coefficient from the line
 * spectrum pairs.
 *
 * @param lsp a vector of the cosine of the LSP values
 * @param cos_val cos(PI*i/N) where i is the index of the LPC amplitude
 * @param order the order of the LSP (and the size of the *lsp buffer). Must
 *        be a multiple of four.
 * @return the LPC value
 *
 * @todo reuse code from Vorbis decoder: vorbis_floor0_decode
 */
static float eval_lpc_spectrum(const float *lsp, float cos_val, int order)
{
    int j;
    float p         = 0.5f;
    float q         = 0.5f;
    float two_cos_w = 2.0f * cos_val;

    for (j = 0; j + 1 < order; j += 2 * 2) {
        // Unroll the loop once since order is a multiple of four
        q *= lsp[j]     - two_cos_w;
        p *= lsp[j + 1] - two_cos_w;

        q *= lsp[j + 2] - two_cos_w;
        p *= lsp[j + 3] - two_cos_w;
    }

    p *= p * (2.0f - two_cos_w);
    q *= q * (2.0f + two_cos_w);

    return 0.5 / (p + q);
}

/**
 * Evaluate the LPC amplitude spectrum envelope from the line spectrum pairs.
 */
static void eval_lpcenv(TwinVQContext *tctx, const float *cos_vals, float *lpc)
{
    int i;
    const TwinVQModeTab *mtab = tctx->mtab;
    int size_s = mtab->size / mtab->fmode[TWINVQ_FT_SHORT].sub;

    for (i = 0; i < size_s / 2; i++) {
        float cos_i = tctx->cos_tabs[0][i];
        lpc[i]              = eval_lpc_spectrum(cos_vals,  cos_i, mtab->n_lsp);
        lpc[size_s - i - 1] = eval_lpc_spectrum(cos_vals, -cos_i, mtab->n_lsp);
    }
}

static void interpolate(float *out, float v1, float v2, int size)
{
    int i;
    float step = (v1 - v2) / (size + 1);

    for (i = 0; i < size; i++) {
        v2    += step;
        out[i] = v2;
    }
}

static inline float get_cos(int idx, int part, const float *cos_tab, int size)
{
    return part ? -cos_tab[size - idx - 1]
                :  cos_tab[idx];
}

/**
 * Evaluate the LPC amplitude spectrum envelope from the line spectrum pairs.
 * Probably for speed reasons, the coefficients are evaluated as
 * siiiibiiiisiiiibiiiisiiiibiiiisiiiibiiiis ...
 * where s is an evaluated value, i is a value interpolated from the others
 * and b might be either calculated or interpolated, depending on an
 * unexplained condition.
 *
 * @param step the size of a block "siiiibiiii"
 * @param in the cosine of the LSP data
 * @param part is 0 for 0...PI (positive cosine values) and 1 for PI...2PI
 *        (negative cosine values)
 * @param size the size of the whole output
 */
static inline void eval_lpcenv_or_interp(TwinVQContext *tctx,
                                         enum TwinVQFrameType ftype,
                                         float *out, const float *in,
                                         int size, int step, int part)
{
    int i;
    const TwinVQModeTab *mtab = tctx->mtab;
    const float *cos_tab      = tctx->cos_tabs[ftype];

    // Fill the 's'
    for (i = 0; i < size; i += step)
        out[i] =
            eval_lpc_spectrum(in,
                              get_cos(i, part, cos_tab, size),
                              mtab->n_lsp);

    // Fill the 'iiiibiiii'
    for (i = step; i <= size - 2 * step; i += step) {
        if (out[i + step] + out[i - step] > 1.95 * out[i] ||
            out[i + step]                 >= out[i - step]) {
            interpolate(out + i - step + 1, out[i], out[i - step], step - 1);
        } else {
            out[i - step / 2] =
                eval_lpc_spectrum(in,
                                  get_cos(i - step / 2, part, cos_tab, size),
                                  mtab->n_lsp);
            interpolate(out + i - step + 1, out[i - step / 2],
                        out[i - step], step / 2 - 1);
            interpolate(out + i - step / 2 + 1, out[i],
                        out[i - step / 2], step / 2 - 1);
        }
    }

    interpolate(out + size - 2 * step + 1, out[size - step],
                out[size - 2 * step], step - 1);
}

static void eval_lpcenv_2parts(TwinVQContext *tctx, enum TwinVQFrameType ftype,
                               const float *buf, float *lpc,
                               int size, int step)
{
    eval_lpcenv_or_interp(tctx, ftype, lpc, buf, size / 2, step, 0);
    eval_lpcenv_or_interp(tctx, ftype, lpc + size / 2, buf, size / 2,
                          2 * step, 1);

    interpolate(lpc + size / 2 - step + 1, lpc[size / 2],
                lpc[size / 2 - step], step);

    twinvq_memset_float(lpc + size - 2 * step + 1, lpc[size - 2 * step],
                        2 * step - 1);
}

/**
 * Inverse quantization. Read CB coefficients for cb1 and cb2 from the
 * bitstream, sum the corresponding vectors and write the result to *out
 * after permutation.
 */
static void dequant(TwinVQContext *tctx, const uint8_t *cb_bits, float *out,
                    enum TwinVQFrameType ftype,
                    const int16_t *cb0, const int16_t *cb1, int cb_len)
{
    int pos = 0;
    int i, j;

    for (i = 0; i < tctx->n_div[ftype]; i++) {
        int tmp0, tmp1;
        int sign0 = 1;
        int sign1 = 1;
        const int16_t *tab0, *tab1;
        int length = tctx->length[ftype][i >= tctx->length_change[ftype]];
        int bitstream_second_part = (i >= tctx->bits_main_spec_change[ftype]);

        int bits = tctx->bits_main_spec[0][ftype][bitstream_second_part];
        tmp0 = *cb_bits++;
        if (bits == 7) {
            if (tmp0 & 0x40)
                sign0 = -1;
            tmp0 &= 0x3F;
        }

        bits = tctx->bits_main_spec[1][ftype][bitstream_second_part];
        tmp1 = *cb_bits++;
        if (bits == 7) {
            if (tmp1 & 0x40)
                sign1 = -1;
            tmp1 &= 0x3F;
        }

        tab0 = cb0 + tmp0 * cb_len;
        tab1 = cb1 + tmp1 * cb_len;

        for (j = 0; j < length; j++)
            out[tctx->permut[ftype][pos + j]] = sign0 * tab0[j] +
                                                sign1 * tab1[j];

        pos += length;
    }
}

static void dec_gain(TwinVQContext *tctx,
                     enum TwinVQFrameType ftype, float *out)
{
    const TwinVQModeTab   *mtab =  tctx->mtab;
    const TwinVQFrameData *bits = &tctx->bits[tctx->cur_frame];
    int i, j;
    int sub        = mtab->fmode[ftype].sub;
    float step     = TWINVQ_AMP_MAX     / ((1 << TWINVQ_GAIN_BITS)     - 1);
    float sub_step = TWINVQ_SUB_AMP_MAX / ((1 << TWINVQ_SUB_GAIN_BITS) - 1);

    if (ftype == TWINVQ_FT_LONG) {
        for (i = 0; i < tctx->avctx->channels; i++)
            out[i] = (1.0 / (1 << 13)) *
                     twinvq_mulawinv(step * 0.5 + step * bits->gain_bits[i],
                                     TWINVQ_AMP_MAX, TWINVQ_MULAW_MU);
    } else {
        for (i = 0; i < tctx->avctx->channels; i++) {
            float val = (1.0 / (1 << 23)) *
                        twinvq_mulawinv(step * 0.5 + step * bits->gain_bits[i],
                                        TWINVQ_AMP_MAX, TWINVQ_MULAW_MU);

            for (j = 0; j < sub; j++)
                out[i * sub + j] =
                    val * twinvq_mulawinv(sub_step * 0.5 +
                                          sub_step * bits->sub_gain_bits[i * sub + j],
                                          TWINVQ_SUB_AMP_MAX, TWINVQ_MULAW_MU);
        }
    }
}

/**
 * Rearrange the LSP coefficients so that they have a minimum distance of
 * min_dist. This function does it exactly as described in section of 3.2.4
 * of the G.729 specification (but interestingly is different from what the
 * reference decoder actually does).
 */
static void rearrange_lsp(int order, float *lsp, float min_dist)
{
    int i;
    float min_dist2 = min_dist * 0.5;
    for (i = 1; i < order; i++)
        if (lsp[i] - lsp[i - 1] < min_dist) {
            float avg = (lsp[i] + lsp[i - 1]) * 0.5;

            lsp[i - 1] = avg - min_dist2;
            lsp[i]     = avg + min_dist2;
        }
}

static void decode_lsp(TwinVQContext *tctx, int lpc_idx1, uint8_t *lpc_idx2,
                       int lpc_hist_idx, float *lsp, float *hist)
{
    const TwinVQModeTab *mtab = tctx->mtab;
    int i, j;

    const float *cb  = mtab->lspcodebook;
    const float *cb2 = cb  + (1 << mtab->lsp_bit1) * mtab->n_lsp;
    const float *cb3 = cb2 + (1 << mtab->lsp_bit2) * mtab->n_lsp;

    const int8_t funny_rounding[4] = {
        -2,
        mtab->lsp_split == 4 ? -2 : 1,
        mtab->lsp_split == 4 ? -2 : 1,
        0
    };

    j = 0;
    for (i = 0; i < mtab->lsp_split; i++) {
        int chunk_end = ((i + 1) * mtab->n_lsp + funny_rounding[i]) /
                        mtab->lsp_split;
        for (; j < chunk_end; j++)
            lsp[j] = cb[lpc_idx1     * mtab->n_lsp + j] +
                     cb2[lpc_idx2[i] * mtab->n_lsp + j];
    }

    rearrange_lsp(mtab->n_lsp, lsp, 0.0001);

    for (i = 0; i < mtab->n_lsp; i++) {
        float tmp1 = 1.0     - cb3[lpc_hist_idx * mtab->n_lsp + i];
        float tmp2 = hist[i] * cb3[lpc_hist_idx * mtab->n_lsp + i];
        hist[i] = lsp[i];
        lsp[i]  = lsp[i] * tmp1 + tmp2;
    }

    rearrange_lsp(mtab->n_lsp, lsp, 0.0001);
    rearrange_lsp(mtab->n_lsp, lsp, 0.000095);
    ff_sort_nearly_sorted_floats(lsp, mtab->n_lsp);
}

static void dec_lpc_spectrum_inv(TwinVQContext *tctx, float *lsp,
                                 enum TwinVQFrameType ftype, float *lpc)
{
    int i;
    int size = tctx->mtab->size / tctx->mtab->fmode[ftype].sub;

    for (i = 0; i < tctx->mtab->n_lsp; i++)
        lsp[i] = 2 * cos(lsp[i]);

    switch (ftype) {
    case TWINVQ_FT_LONG:
        eval_lpcenv_2parts(tctx, ftype, lsp, lpc, size, 8);
        break;
    case TWINVQ_FT_MEDIUM:
        eval_lpcenv_2parts(tctx, ftype, lsp, lpc, size, 2);
        break;
    case TWINVQ_FT_SHORT:
        eval_lpcenv(tctx, lsp, lpc);
        break;
    }
}

static const uint8_t wtype_to_wsize[] = { 0, 0, 2, 2, 2, 1, 0, 1, 1 };

static void imdct_and_window(TwinVQContext *tctx, enum TwinVQFrameType ftype,
                             int wtype, float *in, float *prev, int ch)
{
    FFTContext *mdct = &tctx->mdct_ctx[ftype];
    const TwinVQModeTab *mtab = tctx->mtab;
    int bsize = mtab->size / mtab->fmode[ftype].sub;
    int size  = mtab->size;
    float *buf1 = tctx->tmp_buf;
    int j, first_wsize, wsize; // Window size
    float *out  = tctx->curr_frame + 2 * ch * mtab->size;
    float *out2 = out;
    float *prev_buf;
    int types_sizes[] = {
        mtab->size /  mtab->fmode[TWINVQ_FT_LONG].sub,
        mtab->size /  mtab->fmode[TWINVQ_FT_MEDIUM].sub,
        mtab->size / (mtab->fmode[TWINVQ_FT_SHORT].sub * 2),
    };

    wsize       = types_sizes[wtype_to_wsize[wtype]];
    first_wsize = wsize;
    prev_buf    = prev + (size - bsize) / 2;

    for (j = 0; j < mtab->fmode[ftype].sub; j++) {
        int sub_wtype = ftype == TWINVQ_FT_MEDIUM ? 8 : wtype;

        if (!j && wtype == 4)
            sub_wtype = 4;
        else if (j == mtab->fmode[ftype].sub - 1 && wtype == 7)
            sub_wtype = 7;

        wsize = types_sizes[wtype_to_wsize[sub_wtype]];

        mdct->imdct_half(mdct, buf1 + bsize * j, in + bsize * j);

        tctx->fdsp.vector_fmul_window(out2, prev_buf + (bsize - wsize) / 2,
                                      buf1 + bsize * j,
                                      ff_sine_windows[av_log2(wsize)],
                                      wsize / 2);
        out2 += wsize;

        memcpy(out2, buf1 + bsize * j + wsize / 2,
               (bsize - wsize / 2) * sizeof(float));

        out2 += ftype == TWINVQ_FT_MEDIUM ? (bsize - wsize) / 2 : bsize - wsize;

        prev_buf = buf1 + bsize * j + bsize / 2;
    }

    tctx->last_block_pos[ch] = (size + first_wsize) / 2;
}

static void imdct_output(TwinVQContext *tctx, enum TwinVQFrameType ftype,
                         int wtype, float **out, int offset)
{
    const TwinVQModeTab *mtab = tctx->mtab;
    float *prev_buf           = tctx->prev_frame + tctx->last_block_pos[0];
    int size1, size2, i;
    float *out1, *out2;

    for (i = 0; i < tctx->avctx->channels; i++)
        imdct_and_window(tctx, ftype, wtype,
                         tctx->spectrum + i * mtab->size,
                         prev_buf + 2 * i * mtab->size,
                         i);

    if (!out)
        return;

    size2 = tctx->last_block_pos[0];
    size1 = mtab->size - size2;

    out1 = &out[0][0] + offset;
    memcpy(out1,         prev_buf,         size1 * sizeof(*out1));
    memcpy(out1 + size1, tctx->curr_frame, size2 * sizeof(*out1));

    if (tctx->avctx->channels == 2) {
        out2 = &out[1][0] + offset;
        memcpy(out2, &prev_buf[2 * mtab->size],
               size1 * sizeof(*out2));
        memcpy(out2 + size1, &tctx->curr_frame[2 * mtab->size],
               size2 * sizeof(*out2));
        tctx->fdsp.butterflies_float(out1, out2, mtab->size);
    }
}

static void read_and_decode_spectrum(TwinVQContext *tctx, float *out,
                                     enum TwinVQFrameType ftype)
{
    const TwinVQModeTab *mtab = tctx->mtab;
    TwinVQFrameData *bits     = &tctx->bits[tctx->cur_frame];
    int channels              = tctx->avctx->channels;
    int sub        = mtab->fmode[ftype].sub;
    int block_size = mtab->size / sub;
    float gain[TWINVQ_CHANNELS_MAX * TWINVQ_SUBBLOCKS_MAX];
    float ppc_shape[TWINVQ_PPC_SHAPE_LEN_MAX * TWINVQ_CHANNELS_MAX * 4];

    int i, j;

    dequant(tctx, bits->main_coeffs, out, ftype,
            mtab->fmode[ftype].cb0, mtab->fmode[ftype].cb1,
            mtab->fmode[ftype].cb_len_read);

    dec_gain(tctx, ftype, gain);

    if (ftype == TWINVQ_FT_LONG) {
        int cb_len_p = (tctx->n_div[3] + mtab->ppc_shape_len * channels - 1) /
                       tctx->n_div[3];
        dequant(tctx, bits->ppc_coeffs, ppc_shape,
                TWINVQ_FT_PPC, mtab->ppc_shape_cb,
                mtab->ppc_shape_cb + cb_len_p * TWINVQ_PPC_SHAPE_CB_SIZE,
                cb_len_p);
    }

    for (i = 0; i < channels; i++) {
        float *chunk = out + mtab->size * i;
        float lsp[TWINVQ_LSP_COEFS_MAX];

        for (j = 0; j < sub; j++) {
            tctx->dec_bark_env(tctx, bits->bark1[i][j],
                               bits->bark_use_hist[i][j], i,
                               tctx->tmp_buf, gain[sub * i + j], ftype);

            tctx->fdsp.vector_fmul(chunk + block_size * j,
                                   chunk + block_size * j,
                                   tctx->tmp_buf, block_size);
        }

        if (ftype == TWINVQ_FT_LONG)
            tctx->decode_ppc(tctx, bits->p_coef[i], bits->g_coef[i],
                             ppc_shape + i * mtab->ppc_shape_len, chunk);

        decode_lsp(tctx, bits->lpc_idx1[i], bits->lpc_idx2[i],
                   bits->lpc_hist_idx[i], lsp, tctx->lsp_hist[i]);

        dec_lpc_spectrum_inv(tctx, lsp, ftype, tctx->tmp_buf);

        for (j = 0; j < mtab->fmode[ftype].sub; j++) {
            tctx->fdsp.vector_fmul(chunk, chunk, tctx->tmp_buf, block_size);
            chunk += block_size;
        }
    }
}

const enum TwinVQFrameType ff_twinvq_wtype_to_ftype_table[] = {
    TWINVQ_FT_LONG,   TWINVQ_FT_LONG, TWINVQ_FT_SHORT, TWINVQ_FT_LONG,
    TWINVQ_FT_MEDIUM, TWINVQ_FT_LONG, TWINVQ_FT_LONG,  TWINVQ_FT_MEDIUM,
    TWINVQ_FT_MEDIUM
};

int ff_twinvq_decode_frame(AVCodecContext *avctx, void *data,
                           int *got_frame_ptr, AVPacket *avpkt)
{
    AVFrame *frame     = data;
    const uint8_t *buf = avpkt->data;
    int buf_size       = avpkt->size;
    TwinVQContext *tctx = avctx->priv_data;
    const TwinVQModeTab *mtab = tctx->mtab;
    float **out = NULL;
    int ret;

    /* get output buffer */
    if (tctx->discarded_packets >= 2) {
        frame->nb_samples = mtab->size * tctx->frames_per_packet;
        if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
            return ret;
        out = (float **)frame->extended_data;
    }

    if (buf_size < avctx->block_align) {
        av_log(avctx, AV_LOG_ERROR,
               "Frame too small (%d bytes). Truncated file?\n", buf_size);
        return AVERROR(EINVAL);
    }

    if ((ret = tctx->read_bitstream(avctx, tctx, buf, buf_size)) < 0)
        return ret;

    for (tctx->cur_frame = 0; tctx->cur_frame < tctx->frames_per_packet;
         tctx->cur_frame++) {
        read_and_decode_spectrum(tctx, tctx->spectrum,
                                 tctx->bits[tctx->cur_frame].ftype);

        imdct_output(tctx, tctx->bits[tctx->cur_frame].ftype,
                     tctx->bits[tctx->cur_frame].window_type, out,
                     tctx->cur_frame * mtab->size);

        FFSWAP(float *, tctx->curr_frame, tctx->prev_frame);
    }

    if (tctx->discarded_packets < 2) {
        tctx->discarded_packets++;
        *got_frame_ptr = 0;
        return buf_size;
    }

    *got_frame_ptr = 1;

    // VQF can deliver packets 1 byte greater than block align
    if (buf_size == avctx->block_align + 1)
        return buf_size;
    return avctx->block_align;
}

/**
 * Init IMDCT and windowing tables
 */
static av_cold int init_mdct_win(TwinVQContext *tctx)
{
    int i, j, ret;
    const TwinVQModeTab *mtab = tctx->mtab;
    int size_s = mtab->size / mtab->fmode[TWINVQ_FT_SHORT].sub;
    int size_m = mtab->size / mtab->fmode[TWINVQ_FT_MEDIUM].sub;
    int channels = tctx->avctx->channels;
    float norm = channels == 1 ? 2.0 : 1.0;

    for (i = 0; i < 3; i++) {
        int bsize = tctx->mtab->size / tctx->mtab->fmode[i].sub;
        if ((ret = ff_mdct_init(&tctx->mdct_ctx[i], av_log2(bsize) + 1, 1,
                                -sqrt(norm / bsize) / (1 << 15))))
            return ret;
    }

    FF_ALLOC_OR_GOTO(tctx->avctx, tctx->tmp_buf,
                     mtab->size * sizeof(*tctx->tmp_buf), alloc_fail);

    FF_ALLOC_OR_GOTO(tctx->avctx, tctx->spectrum,
                     2 * mtab->size * channels * sizeof(*tctx->spectrum),
                     alloc_fail);
    FF_ALLOC_OR_GOTO(tctx->avctx, tctx->curr_frame,
                     2 * mtab->size * channels * sizeof(*tctx->curr_frame),
                     alloc_fail);
    FF_ALLOC_OR_GOTO(tctx->avctx, tctx->prev_frame,
                     2 * mtab->size * channels * sizeof(*tctx->prev_frame),
                     alloc_fail);

    for (i = 0; i < 3; i++) {
        int m       = 4 * mtab->size / mtab->fmode[i].sub;
        double freq = 2 * M_PI / m;
        FF_ALLOC_OR_GOTO(tctx->avctx, tctx->cos_tabs[i],
                         (m / 4) * sizeof(*tctx->cos_tabs[i]), alloc_fail);

        for (j = 0; j <= m / 8; j++)
            tctx->cos_tabs[i][j] = cos((2 * j + 1) * freq);
        for (j = 1; j < m / 8; j++)
            tctx->cos_tabs[i][m / 4 - j] = tctx->cos_tabs[i][j];
    }

    ff_init_ff_sine_windows(av_log2(size_m));
    ff_init_ff_sine_windows(av_log2(size_s / 2));
    ff_init_ff_sine_windows(av_log2(mtab->size));

    return 0;

alloc_fail:
    return AVERROR(ENOMEM);
}

/**
 * Interpret the data as if it were a num_blocks x line_len[0] matrix and for
 * each line do a cyclic permutation, i.e.
 * abcdefghijklm -> defghijklmabc
 * where the amount to be shifted is evaluated depending on the column.
 */
static void permutate_in_line(int16_t *tab, int num_vect, int num_blocks,
                              int block_size,
                              const uint8_t line_len[2], int length_div,
                              enum TwinVQFrameType ftype)
{
    int i, j;

    for (i = 0; i < line_len[0]; i++) {
        int shift;

        if (num_blocks == 1                                    ||
            (ftype == TWINVQ_FT_LONG && num_vect % num_blocks) ||
            (ftype != TWINVQ_FT_LONG && num_vect & 1)          ||
            i == line_len[1]) {
            shift = 0;
        } else if (ftype == TWINVQ_FT_LONG) {
            shift = i;
        } else
            shift = i * i;

        for (j = 0; j < num_vect && (j + num_vect * i < block_size * num_blocks); j++)
            tab[i * num_vect + j] = i * num_vect + (j + shift) % num_vect;
    }
}

/**
 * Interpret the input data as in the following table:
 *
 * @verbatim
 *
 * abcdefgh
 * ijklmnop
 * qrstuvw
 * x123456
 *
 * @endverbatim
 *
 * and transpose it, giving the output
 * aiqxbjr1cks2dlt3emu4fvn5gow6hp
 */
static void transpose_perm(int16_t *out, int16_t *in, int num_vect,
                           const uint8_t line_len[2], int length_div)
{
    int i, j;
    int cont = 0;

    for (i = 0; i < num_vect; i++)
        for (j = 0; j < line_len[i >= length_div]; j++)
            out[cont++] = in[j * num_vect + i];
}

static void linear_perm(int16_t *out, int16_t *in, int n_blocks, int size)
{
    int block_size = size / n_blocks;
    int i;

    for (i = 0; i < size; i++)
        out[i] = block_size * (in[i] % n_blocks) + in[i] / n_blocks;
}

static av_cold void construct_perm_table(TwinVQContext *tctx,
                                         enum TwinVQFrameType ftype)
{
    int block_size, size;
    const TwinVQModeTab *mtab = tctx->mtab;
    int16_t *tmp_perm = (int16_t *)tctx->tmp_buf;

    if (ftype == TWINVQ_FT_PPC) {
        size       = tctx->avctx->channels;
        block_size = mtab->ppc_shape_len;
    } else {
        size       = tctx->avctx->channels * mtab->fmode[ftype].sub;
        block_size = mtab->size / mtab->fmode[ftype].sub;
    }

    permutate_in_line(tmp_perm, tctx->n_div[ftype], size,
                      block_size, tctx->length[ftype],
                      tctx->length_change[ftype], ftype);

    transpose_perm(tctx->permut[ftype], tmp_perm, tctx->n_div[ftype],
                   tctx->length[ftype], tctx->length_change[ftype]);

    linear_perm(tctx->permut[ftype], tctx->permut[ftype], size,
                size * block_size);
}

static av_cold void init_bitstream_params(TwinVQContext *tctx)
{
    const TwinVQModeTab *mtab = tctx->mtab;
    int n_ch                  = tctx->avctx->channels;
    int total_fr_bits         = tctx->avctx->bit_rate * mtab->size /
                                tctx->avctx->sample_rate;

    int lsp_bits_per_block = n_ch * (mtab->lsp_bit0 + mtab->lsp_bit1 +
                                     mtab->lsp_split * mtab->lsp_bit2);

    int ppc_bits = n_ch * (mtab->pgain_bit + mtab->ppc_shape_bit +
                           mtab->ppc_period_bit);

    int bsize_no_main_cb[3], bse_bits[3], i;
    enum TwinVQFrameType frametype;

    for (i = 0; i < 3; i++)
        // +1 for history usage switch
        bse_bits[i] = n_ch *
                      (mtab->fmode[i].bark_n_coef *
                       mtab->fmode[i].bark_n_bit + 1);

    bsize_no_main_cb[2] = bse_bits[2] + lsp_bits_per_block + ppc_bits +
                          TWINVQ_WINDOW_TYPE_BITS + n_ch * TWINVQ_GAIN_BITS;

    for (i = 0; i < 2; i++)
        bsize_no_main_cb[i] =
            lsp_bits_per_block + n_ch * TWINVQ_GAIN_BITS +
            TWINVQ_WINDOW_TYPE_BITS +
            mtab->fmode[i].sub * (bse_bits[i] + n_ch * TWINVQ_SUB_GAIN_BITS);

    if (tctx->codec == TWINVQ_CODEC_METASOUND && !tctx->is_6kbps) {
        bsize_no_main_cb[1] += 2;
        bsize_no_main_cb[2] += 2;
    }

    // The remaining bits are all used for the main spectrum coefficients
    for (i = 0; i < 4; i++) {
        int bit_size, vect_size;
        int rounded_up, rounded_down, num_rounded_down, num_rounded_up;
        if (i == 3) {
            bit_size  = n_ch * mtab->ppc_shape_bit;
            vect_size = n_ch * mtab->ppc_shape_len;
        } else {
            bit_size  = total_fr_bits - bsize_no_main_cb[i];
            vect_size = n_ch * mtab->size;
        }

        tctx->n_div[i] = (bit_size + 13) / 14;

        rounded_up                     = (bit_size + tctx->n_div[i] - 1) /
                                         tctx->n_div[i];
        rounded_down                   = (bit_size) / tctx->n_div[i];
        num_rounded_down               = rounded_up * tctx->n_div[i] - bit_size;
        num_rounded_up                 = tctx->n_div[i] - num_rounded_down;
        tctx->bits_main_spec[0][i][0]  = (rounded_up + 1)   / 2;
        tctx->bits_main_spec[1][i][0]  =  rounded_up        / 2;
        tctx->bits_main_spec[0][i][1]  = (rounded_down + 1) / 2;
        tctx->bits_main_spec[1][i][1]  =  rounded_down      / 2;
        tctx->bits_main_spec_change[i] = num_rounded_up;

        rounded_up             = (vect_size + tctx->n_div[i] - 1) /
                                 tctx->n_div[i];
        rounded_down           = (vect_size) / tctx->n_div[i];
        num_rounded_down       = rounded_up * tctx->n_div[i] - vect_size;
        num_rounded_up         = tctx->n_div[i] - num_rounded_down;
        tctx->length[i][0]     = rounded_up;
        tctx->length[i][1]     = rounded_down;
        tctx->length_change[i] = num_rounded_up;
    }

    for (frametype = TWINVQ_FT_SHORT; frametype <= TWINVQ_FT_PPC; frametype++)
        construct_perm_table(tctx, frametype);
}

av_cold int ff_twinvq_decode_close(AVCodecContext *avctx)
{
    TwinVQContext *tctx = avctx->priv_data;
    int i;

    for (i = 0; i < 3; i++) {
        ff_mdct_end(&tctx->mdct_ctx[i]);
        av_free(tctx->cos_tabs[i]);
    }

    av_free(tctx->curr_frame);
    av_free(tctx->spectrum);
    av_free(tctx->prev_frame);
    av_free(tctx->tmp_buf);

    return 0;
}

av_cold int ff_twinvq_decode_init(AVCodecContext *avctx)
{
    int ret;
    TwinVQContext *tctx = avctx->priv_data;

    tctx->avctx       = avctx;
    avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;

    if (!avctx->block_align) {
        avctx->block_align = tctx->frame_size + 7 >> 3;
    } else if (avctx->block_align * 8 < tctx->frame_size) {
        av_log(avctx, AV_LOG_ERROR, "Block align is %d bits, expected %d\n",
               avctx->block_align * 8, tctx->frame_size);
        return AVERROR_INVALIDDATA;
    }
    tctx->frames_per_packet = avctx->block_align * 8 / tctx->frame_size;
    if (tctx->frames_per_packet > TWINVQ_MAX_FRAMES_PER_PACKET) {
        av_log(avctx, AV_LOG_ERROR, "Too many frames per packet (%d)\n",
               tctx->frames_per_packet);
        return AVERROR_INVALIDDATA;
    }

    avpriv_float_dsp_init(&tctx->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
    if ((ret = init_mdct_win(tctx))) {
        av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n");
        ff_twinvq_decode_close(avctx);
        return ret;
    }
    init_bitstream_params(tctx);

    twinvq_memset_float(tctx->bark_hist[0][0], 0.1,
                        FF_ARRAY_ELEMS(tctx->bark_hist));

    return 0;
}