Coverage Report

Created: 2025-07-11 06:40

/proc/self/cwd/libfaad/ps_dec.c
Line
Count
Source (jump to first uncovered line)
1
/*
2
** FAAD2 - Freeware Advanced Audio (AAC) Decoder including SBR decoding
3
** Copyright (C) 2003-2005 M. Bakker, Nero AG, http://www.nero.com
4
**
5
** This program is free software; you can redistribute it and/or modify
6
** it under the terms of the GNU General Public License as published by
7
** the Free Software Foundation; either version 2 of the License, or
8
** (at your option) any later version.
9
**
10
** This program is distributed in the hope that it will be useful,
11
** but WITHOUT ANY WARRANTY; without even the implied warranty of
12
** MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13
** GNU General Public License for more details.
14
**
15
** You should have received a copy of the GNU General Public License
16
** along with this program; if not, write to the Free Software
17
** Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
18
**
19
** Any non-GPL usage of this software or parts of this software is strictly
20
** forbidden.
21
**
22
** The "appropriate copyright message" mentioned in section 2c of the GPLv2
23
** must read: "Code from FAAD2 is copyright (c) Nero AG, www.nero.com"
24
**
25
** Commercial non-GPL licensing of this software is possible.
26
** For more info contact Nero AG through Mpeg4AAClicense@nero.com.
27
**
28
** $Id: ps_dec.c,v 1.16 2009/01/26 22:32:31 menno Exp $
29
**/
30
31
#include "common.h"
32
33
#ifdef PS_DEC
34
35
#include <stdlib.h>
36
#include <stdio.h>
37
#include "ps_dec.h"
38
#include "ps_tables.h"
39
40
/* constants */
41
14.1M
#define NEGATE_IPD_MASK            (0x1000)
42
107k
#define DECAY_SLOPE                FRAC_CONST(0.05)
43
#define COEF_SQRT2                 COEF_CONST(1.4142135623731)
44
45
/* tables */
46
/* filters are mirrored in coef 6, second half left out */
47
static const real_t p8_13_20[7] =
48
{
49
    FRAC_CONST(0.00746082949812),
50
    FRAC_CONST(0.02270420949825),
51
    FRAC_CONST(0.04546865930473),
52
    FRAC_CONST(0.07266113929591),
53
    FRAC_CONST(0.09885108575264),
54
    FRAC_CONST(0.11793710567217),
55
    FRAC_CONST(0.125)
56
};
57
58
static const real_t p2_13_20[7] =
59
{
60
    FRAC_CONST(0.0),
61
    FRAC_CONST(0.01899487526049),
62
    FRAC_CONST(0.0),
63
    FRAC_CONST(-0.07293139167538),
64
    FRAC_CONST(0.0),
65
    FRAC_CONST(0.30596630545168),
66
    FRAC_CONST(0.5)
67
};
68
69
static const real_t p12_13_34[7] =
70
{
71
    FRAC_CONST(0.04081179924692),
72
    FRAC_CONST(0.03812810994926),
73
    FRAC_CONST(0.05144908135699),
74
    FRAC_CONST(0.06399831151592),
75
    FRAC_CONST(0.07428313801106),
76
    FRAC_CONST(0.08100347892914),
77
    FRAC_CONST(0.08333333333333)
78
};
79
80
static const real_t p8_13_34[7] =
81
{
82
    FRAC_CONST(0.01565675600122),
83
    FRAC_CONST(0.03752716391991),
84
    FRAC_CONST(0.05417891378782),
85
    FRAC_CONST(0.08417044116767),
86
    FRAC_CONST(0.10307344158036),
87
    FRAC_CONST(0.12222452249753),
88
    FRAC_CONST(0.125)
89
};
90
91
static const real_t p4_13_34[7] =
92
{
93
    FRAC_CONST(-0.05908211155639),
94
    FRAC_CONST(-0.04871498374946),
95
    FRAC_CONST(0.0),
96
    FRAC_CONST(0.07778723915851),
97
    FRAC_CONST(0.16486303567403),
98
    FRAC_CONST(0.23279856662996),
99
    FRAC_CONST(0.25)
100
};
101
102
#ifdef PARAM_32KHZ
103
static const uint8_t delay_length_d[2][NO_ALLPASS_LINKS] = {
104
    { 1, 2, 3 } /* d_24kHz */,
105
    { 3, 4, 5 } /* d_48kHz */
106
};
107
#else
108
static const uint8_t delay_length_d[NO_ALLPASS_LINKS] = {
109
    3, 4, 5 /* d_48kHz */
110
};
111
#endif
112
static const real_t filter_a[NO_ALLPASS_LINKS] = { /* a(m) = exp(-d_48kHz(m)/7) */
113
    FRAC_CONST(0.65143905753106),
114
    FRAC_CONST(0.56471812200776),
115
    FRAC_CONST(0.48954165955695)
116
};
117
118
static const uint8_t group_border20[10+12 + 1] =
119
{
120
    6, 7, 0, 1, 2, 3, /* 6 subqmf subbands */
121
    9, 8,             /* 2 subqmf subbands */
122
    10, 11,           /* 2 subqmf subbands */
123
    3, 4, 5, 6, 7, 8, 9, 11, 14, 18, 23, 35, 64
124
};
125
126
static const uint8_t group_border34[32+18 + 1] =
127
{
128
     0,  1,  2,  3,  4,  5,  6,  7,  8,  9,  10, 11, /* 12 subqmf subbands */
129
     12, 13, 14, 15, 16, 17, 18, 19,                 /*  8 subqmf subbands */
130
     20, 21, 22, 23,                                 /*  4 subqmf subbands */
131
     24, 25, 26, 27,                                 /*  4 subqmf subbands */
132
     28, 29, 30, 31,                                 /*  4 subqmf subbands */
133
     32-27, 33-27, 34-27, 35-27, 36-27, 37-27, 38-27, 40-27, 42-27, 44-27, 46-27, 48-27, 51-27, 54-27, 57-27, 60-27, 64-27, 68-27, 91-27
134
};
135
136
static const uint16_t map_group2bk20[10+12] =
137
{
138
    (NEGATE_IPD_MASK | 1), (NEGATE_IPD_MASK | 0),
139
    0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19
140
};
141
142
static const uint16_t map_group2bk34[32+18] =
143
{
144
    0,  1,  2,  3,  4,  5,  6,  6,  7, (NEGATE_IPD_MASK | 2), (NEGATE_IPD_MASK | 1), (NEGATE_IPD_MASK | 0),
145
    10, 10, 4,  5,  6,  7,  8,  9,
146
    10, 11, 12, 9,
147
    14, 11, 12, 13,
148
    14, 15, 16, 13,
149
    16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33
150
};
151
152
/* type definitions */
153
typedef struct
154
{
155
    uint8_t frame_len;
156
    uint8_t resolution20[3];
157
    uint8_t resolution34[5];
158
159
    qmf_t *work;
160
    qmf_t **buffer;
161
    qmf_t **temp;
162
} hyb_info;
163
164
/* static function declarations */
165
static void ps_data_decode(ps_info *ps);
166
static hyb_info *hybrid_init(uint8_t numTimeSlotsRate);
167
static void channel_filter2(hyb_info *hyb, uint8_t frame_len, const real_t *filter,
168
                            qmf_t *buffer, qmf_t **X_hybrid);
169
static void INLINE DCT3_4_unscaled(real_t *y, real_t *x);
170
static void channel_filter8(hyb_info *hyb, uint8_t frame_len, const real_t *filter,
171
                            qmf_t *buffer, qmf_t **X_hybrid);
172
static void hybrid_analysis(hyb_info *hyb, qmf_t X[32][64], qmf_t X_hybrid[32][32],
173
                            uint8_t use34, uint8_t numTimeSlotsRate);
174
static void hybrid_synthesis(hyb_info *hyb, qmf_t X[32][64], qmf_t X_hybrid[32][32],
175
                             uint8_t use34, uint8_t numTimeSlotsRate);
176
static int8_t delta_clip(int8_t i, int8_t min, int8_t max);
177
static void delta_decode(uint8_t enable, int8_t *index, int8_t *index_prev,
178
                         uint8_t dt_flag, uint8_t nr_par, uint8_t stride,
179
                         int8_t min_index, int8_t max_index);
180
static void delta_modulo_decode(uint8_t enable, int8_t *index, int8_t *index_prev,
181
                                uint8_t dt_flag, uint8_t nr_par, uint8_t stride,
182
                                int8_t and_modulo);
183
static void map20indexto34(int8_t *index, uint8_t bins);
184
#ifdef PS_LOW_POWER
185
static void map34indexto20(int8_t *index, uint8_t bins);
186
#endif
187
static void ps_data_decode(ps_info *ps);
188
static void ps_decorrelate(ps_info *ps, qmf_t X_left[38][64], qmf_t X_right[38][64],
189
                           qmf_t X_hybrid_left[32][32], qmf_t X_hybrid_right[32][32]);
190
static void ps_mix_phase(ps_info *ps, qmf_t X_left[38][64], qmf_t X_right[38][64],
191
                         qmf_t X_hybrid_left[32][32], qmf_t X_hybrid_right[32][32]);
192
193
/*  */
194
195
196
static hyb_info *hybrid_init(uint8_t numTimeSlotsRate)
197
8.89k
{
198
8.89k
    uint8_t i;
199
200
8.89k
    hyb_info *hyb = (hyb_info*)faad_malloc(sizeof(hyb_info));
201
202
8.89k
    hyb->resolution34[0] = 12;
203
8.89k
    hyb->resolution34[1] = 8;
204
8.89k
    hyb->resolution34[2] = 4;
205
8.89k
    hyb->resolution34[3] = 4;
206
8.89k
    hyb->resolution34[4] = 4;
207
208
8.89k
    hyb->resolution20[0] = 8;
209
8.89k
    hyb->resolution20[1] = 2;
210
8.89k
    hyb->resolution20[2] = 2;
211
212
8.89k
    hyb->frame_len = numTimeSlotsRate;
213
214
8.89k
    hyb->work = (qmf_t*)faad_malloc((hyb->frame_len+12) * sizeof(qmf_t));
215
8.89k
    memset(hyb->work, 0, (hyb->frame_len+12) * sizeof(qmf_t));
216
217
8.89k
    hyb->buffer = (qmf_t**)faad_malloc(5 * sizeof(qmf_t*));
218
53.3k
    for (i = 0; i < 5; i++)
219
44.4k
    {
220
44.4k
        hyb->buffer[i] = (qmf_t*)faad_malloc(hyb->frame_len * sizeof(qmf_t));
221
44.4k
        memset(hyb->buffer[i], 0, hyb->frame_len * sizeof(qmf_t));
222
44.4k
    }
223
224
8.89k
    hyb->temp = (qmf_t**)faad_malloc(hyb->frame_len * sizeof(qmf_t*));
225
290k
    for (i = 0; i < hyb->frame_len; i++)
226
281k
    {
227
281k
        hyb->temp[i] = (qmf_t*)faad_malloc(12 /*max*/ * sizeof(qmf_t));
228
281k
    }
229
230
8.89k
    return hyb;
231
8.89k
}
232
233
static void hybrid_free(hyb_info *hyb)
234
8.89k
{
235
8.89k
    uint8_t i;
236
237
8.89k
  if (!hyb) return;
238
239
8.89k
    if (hyb->work)
240
8.89k
        faad_free(hyb->work);
241
242
53.3k
    for (i = 0; i < 5; i++)
243
44.4k
    {
244
44.4k
        if (hyb->buffer[i])
245
44.4k
            faad_free(hyb->buffer[i]);
246
44.4k
    }
247
8.89k
    if (hyb->buffer)
248
8.89k
        faad_free(hyb->buffer);
249
250
290k
    for (i = 0; i < hyb->frame_len; i++)
251
281k
    {
252
281k
        if (hyb->temp[i])
253
281k
            faad_free(hyb->temp[i]);
254
281k
    }
255
8.89k
    if (hyb->temp)
256
8.89k
        faad_free(hyb->temp);
257
258
8.89k
    faad_free(hyb);
259
8.89k
}
260
261
/* real filter, size 2 */
262
static void channel_filter2(hyb_info *hyb, uint8_t frame_len, const real_t *filter,
263
                            qmf_t *buffer, qmf_t **X_hybrid)
264
7.29k
{
265
7.29k
    uint8_t i;
266
7.29k
    (void)hyb;  /* TODO: remove parameter? */
267
268
236k
    for (i = 0; i < frame_len; i++)
269
228k
    {
270
228k
        real_t r0 = MUL_F(filter[0],(QMF_RE(buffer[0+i]) + QMF_RE(buffer[12+i])));
271
228k
        real_t r1 = MUL_F(filter[1],(QMF_RE(buffer[1+i]) + QMF_RE(buffer[11+i])));
272
228k
        real_t r2 = MUL_F(filter[2],(QMF_RE(buffer[2+i]) + QMF_RE(buffer[10+i])));
273
228k
        real_t r3 = MUL_F(filter[3],(QMF_RE(buffer[3+i]) + QMF_RE(buffer[9+i])));
274
228k
        real_t r4 = MUL_F(filter[4],(QMF_RE(buffer[4+i]) + QMF_RE(buffer[8+i])));
275
228k
        real_t r5 = MUL_F(filter[5],(QMF_RE(buffer[5+i]) + QMF_RE(buffer[7+i])));
276
228k
        real_t r6 = MUL_F(filter[6],QMF_RE(buffer[6+i]));
277
228k
        real_t i0 = MUL_F(filter[0],(QMF_IM(buffer[0+i]) + QMF_IM(buffer[12+i])));
278
228k
        real_t i1 = MUL_F(filter[1],(QMF_IM(buffer[1+i]) + QMF_IM(buffer[11+i])));
279
228k
        real_t i2 = MUL_F(filter[2],(QMF_IM(buffer[2+i]) + QMF_IM(buffer[10+i])));
280
228k
        real_t i3 = MUL_F(filter[3],(QMF_IM(buffer[3+i]) + QMF_IM(buffer[9+i])));
281
228k
        real_t i4 = MUL_F(filter[4],(QMF_IM(buffer[4+i]) + QMF_IM(buffer[8+i])));
282
228k
        real_t i5 = MUL_F(filter[5],(QMF_IM(buffer[5+i]) + QMF_IM(buffer[7+i])));
283
228k
        real_t i6 = MUL_F(filter[6],QMF_IM(buffer[6+i]));
284
285
        /* q = 0 */
286
228k
        QMF_RE(X_hybrid[i][0]) = r0 + r1 + r2 + r3 + r4 + r5 + r6;
287
228k
        QMF_IM(X_hybrid[i][0]) = i0 + i1 + i2 + i3 + i4 + i5 + i6;
288
289
        /* q = 1 */
290
228k
        QMF_RE(X_hybrid[i][1]) = r0 - r1 + r2 - r3 + r4 - r5 + r6;
291
228k
        QMF_IM(X_hybrid[i][1]) = i0 - i1 + i2 - i3 + i4 - i5 + i6;
292
228k
    }
293
7.29k
}
294
295
/* complex filter, size 4 */
296
static void channel_filter4(hyb_info *hyb, uint8_t frame_len, const real_t *filter,
297
                            qmf_t *buffer, qmf_t **X_hybrid)
298
6.02k
{
299
6.02k
    uint8_t i;
300
6.02k
    real_t input_re1[2], input_re2[2], input_im1[2], input_im2[2];
301
6.02k
    (void)hyb;  /* TODO: remove parameter? */
302
303
190k
    for (i = 0; i < frame_len; i++)
304
184k
    {
305
184k
        input_re1[0] = -MUL_F(filter[2], (QMF_RE(buffer[i+2]) + QMF_RE(buffer[i+10]))) +
306
184k
            MUL_F(filter[6], QMF_RE(buffer[i+6]));
307
184k
        input_re1[1] = MUL_F(FRAC_CONST(-0.70710678118655),
308
184k
            (MUL_F(filter[1], (QMF_RE(buffer[i+1]) + QMF_RE(buffer[i+11]))) +
309
184k
            MUL_F(filter[3], (QMF_RE(buffer[i+3]) + QMF_RE(buffer[i+9]))) -
310
184k
            MUL_F(filter[5], (QMF_RE(buffer[i+5]) + QMF_RE(buffer[i+7])))));
311
312
184k
        input_im1[0] = MUL_F(filter[0], (QMF_IM(buffer[i+0]) - QMF_IM(buffer[i+12]))) -
313
184k
            MUL_F(filter[4], (QMF_IM(buffer[i+4]) - QMF_IM(buffer[i+8])));
314
184k
        input_im1[1] = MUL_F(FRAC_CONST(0.70710678118655),
315
184k
            (MUL_F(filter[1], (QMF_IM(buffer[i+1]) - QMF_IM(buffer[i+11]))) -
316
184k
            MUL_F(filter[3], (QMF_IM(buffer[i+3]) - QMF_IM(buffer[i+9]))) -
317
184k
            MUL_F(filter[5], (QMF_IM(buffer[i+5]) - QMF_IM(buffer[i+7])))));
318
319
184k
        input_re2[0] = MUL_F(filter[0], (QMF_RE(buffer[i+0]) - QMF_RE(buffer[i+12]))) -
320
184k
            MUL_F(filter[4], (QMF_RE(buffer[i+4]) - QMF_RE(buffer[i+8])));
321
184k
        input_re2[1] = MUL_F(FRAC_CONST(0.70710678118655),
322
184k
            (MUL_F(filter[1], (QMF_RE(buffer[i+1]) - QMF_RE(buffer[i+11]))) -
323
184k
            MUL_F(filter[3], (QMF_RE(buffer[i+3]) - QMF_RE(buffer[i+9]))) -
324
184k
            MUL_F(filter[5], (QMF_RE(buffer[i+5]) - QMF_RE(buffer[i+7])))));
325
326
184k
        input_im2[0] = -MUL_F(filter[2], (QMF_IM(buffer[i+2]) + QMF_IM(buffer[i+10]))) +
327
184k
            MUL_F(filter[6], QMF_IM(buffer[i+6]));
328
184k
        input_im2[1] = MUL_F(FRAC_CONST(-0.70710678118655),
329
184k
            (MUL_F(filter[1], (QMF_IM(buffer[i+1]) + QMF_IM(buffer[i+11]))) +
330
184k
            MUL_F(filter[3], (QMF_IM(buffer[i+3]) + QMF_IM(buffer[i+9]))) -
331
184k
            MUL_F(filter[5], (QMF_IM(buffer[i+5]) + QMF_IM(buffer[i+7])))));
332
333
        /* q == 0 */
334
184k
        QMF_RE(X_hybrid[i][0]) =  input_re1[0] + input_re1[1] + input_im1[0] + input_im1[1];
335
184k
        QMF_IM(X_hybrid[i][0]) = -input_re2[0] - input_re2[1] + input_im2[0] + input_im2[1];
336
337
        /* q == 1 */
338
184k
        QMF_RE(X_hybrid[i][1]) =  input_re1[0] - input_re1[1] - input_im1[0] + input_im1[1];
339
184k
        QMF_IM(X_hybrid[i][1]) =  input_re2[0] - input_re2[1] + input_im2[0] - input_im2[1];
340
341
        /* q == 2 */
342
184k
        QMF_RE(X_hybrid[i][2]) =  input_re1[0] - input_re1[1] + input_im1[0] - input_im1[1];
343
184k
        QMF_IM(X_hybrid[i][2]) = -input_re2[0] + input_re2[1] + input_im2[0] - input_im2[1];
344
345
        /* q == 3 */
346
184k
        QMF_RE(X_hybrid[i][3]) =  input_re1[0] + input_re1[1] - input_im1[0] - input_im1[1];
347
184k
        QMF_IM(X_hybrid[i][3]) =  input_re2[0] + input_re2[1] + input_im2[0] + input_im2[1];
348
184k
    }
349
6.02k
}
350
351
static void INLINE DCT3_4_unscaled(real_t *y, real_t *x)
352
703k
{
353
703k
    real_t f0, f1, f2, f3, f4, f5, f6, f7, f8;
354
355
703k
    f0 = MUL_F(x[2], FRAC_CONST(0.7071067811865476));
356
703k
    f1 = x[0] - f0;
357
703k
    f2 = x[0] + f0;
358
703k
    f3 = x[1] + x[3];
359
703k
    f4 = MUL_C(x[1], COEF_CONST(1.3065629648763766));
360
703k
    f5 = MUL_F(f3, FRAC_CONST(-0.9238795325112866));
361
703k
    f6 = MUL_F(x[3], FRAC_CONST(-0.5411961001461967));
362
703k
    f7 = f4 + f5;
363
703k
    f8 = f6 - f5;
364
703k
    y[3] = f2 - f8;
365
703k
    y[0] = f2 + f8;
366
703k
    y[2] = f1 - f7;
367
703k
    y[1] = f1 + f7;
368
703k
}
369
370
/* complex filter, size 8 */
371
static void channel_filter8(hyb_info *hyb, uint8_t frame_len, const real_t *filter,
372
                            qmf_t *buffer, qmf_t **X_hybrid)
373
5.65k
{
374
5.65k
    uint8_t i, n;
375
5.65k
    real_t input_re1[4], input_re2[4], input_im1[4], input_im2[4];
376
5.65k
    real_t x[4];
377
5.65k
    (void)hyb;  /* TODO: remove parameter? */
378
379
181k
    for (i = 0; i < frame_len; i++)
380
175k
    {
381
175k
        input_re1[0] =  MUL_F(filter[6],QMF_RE(buffer[6+i]));
382
175k
        input_re1[1] =  MUL_F(filter[5],(QMF_RE(buffer[5+i]) + QMF_RE(buffer[7+i])));
383
175k
        input_re1[2] = -MUL_F(filter[0],(QMF_RE(buffer[0+i]) + QMF_RE(buffer[12+i]))) + MUL_F(filter[4],(QMF_RE(buffer[4+i]) + QMF_RE(buffer[8+i])));
384
175k
        input_re1[3] = -MUL_F(filter[1],(QMF_RE(buffer[1+i]) + QMF_RE(buffer[11+i]))) + MUL_F(filter[3],(QMF_RE(buffer[3+i]) + QMF_RE(buffer[9+i])));
385
386
175k
        input_im1[0] = MUL_F(filter[5],(QMF_IM(buffer[7+i]) - QMF_IM(buffer[5+i])));
387
175k
        input_im1[1] = MUL_F(filter[0],(QMF_IM(buffer[12+i]) - QMF_IM(buffer[0+i]))) + MUL_F(filter[4],(QMF_IM(buffer[8+i]) - QMF_IM(buffer[4+i])));
388
175k
        input_im1[2] = MUL_F(filter[1],(QMF_IM(buffer[11+i]) - QMF_IM(buffer[1+i]))) + MUL_F(filter[3],(QMF_IM(buffer[9+i]) - QMF_IM(buffer[3+i])));
389
175k
        input_im1[3] = MUL_F(filter[2],(QMF_IM(buffer[10+i]) - QMF_IM(buffer[2+i])));
390
391
879k
        for (n = 0; n < 4; n++)
392
703k
        {
393
703k
            x[n] = input_re1[n] - input_im1[3-n];
394
703k
        }
395
175k
        DCT3_4_unscaled(x, x);
396
175k
        QMF_RE(X_hybrid[i][7]) = x[0];
397
175k
        QMF_RE(X_hybrid[i][5]) = x[2];
398
175k
        QMF_RE(X_hybrid[i][3]) = x[3];
399
175k
        QMF_RE(X_hybrid[i][1]) = x[1];
400
401
879k
        for (n = 0; n < 4; n++)
402
703k
        {
403
703k
            x[n] = input_re1[n] + input_im1[3-n];
404
703k
        }
405
175k
        DCT3_4_unscaled(x, x);
406
175k
        QMF_RE(X_hybrid[i][6]) = x[1];
407
175k
        QMF_RE(X_hybrid[i][4]) = x[3];
408
175k
        QMF_RE(X_hybrid[i][2]) = x[2];
409
175k
        QMF_RE(X_hybrid[i][0]) = x[0];
410
411
175k
        input_im2[0] =  MUL_F(filter[6],QMF_IM(buffer[6+i]));
412
175k
        input_im2[1] =  MUL_F(filter[5],(QMF_IM(buffer[5+i]) + QMF_IM(buffer[7+i])));
413
175k
        input_im2[2] = -MUL_F(filter[0],(QMF_IM(buffer[0+i]) + QMF_IM(buffer[12+i]))) + MUL_F(filter[4],(QMF_IM(buffer[4+i]) + QMF_IM(buffer[8+i])));
414
175k
        input_im2[3] = -MUL_F(filter[1],(QMF_IM(buffer[1+i]) + QMF_IM(buffer[11+i]))) + MUL_F(filter[3],(QMF_IM(buffer[3+i]) + QMF_IM(buffer[9+i])));
415
416
175k
        input_re2[0] = MUL_F(filter[5],(QMF_RE(buffer[7+i]) - QMF_RE(buffer[5+i])));
417
175k
        input_re2[1] = MUL_F(filter[0],(QMF_RE(buffer[12+i]) - QMF_RE(buffer[0+i]))) + MUL_F(filter[4],(QMF_RE(buffer[8+i]) - QMF_RE(buffer[4+i])));
418
175k
        input_re2[2] = MUL_F(filter[1],(QMF_RE(buffer[11+i]) - QMF_RE(buffer[1+i]))) + MUL_F(filter[3],(QMF_RE(buffer[9+i]) - QMF_RE(buffer[3+i])));
419
175k
        input_re2[3] = MUL_F(filter[2],(QMF_RE(buffer[10+i]) - QMF_RE(buffer[2+i])));
420
421
879k
        for (n = 0; n < 4; n++)
422
703k
        {
423
703k
            x[n] = input_im2[n] + input_re2[3-n];
424
703k
        }
425
175k
        DCT3_4_unscaled(x, x);
426
175k
        QMF_IM(X_hybrid[i][7]) = x[0];
427
175k
        QMF_IM(X_hybrid[i][5]) = x[2];
428
175k
        QMF_IM(X_hybrid[i][3]) = x[3];
429
175k
        QMF_IM(X_hybrid[i][1]) = x[1];
430
431
879k
        for (n = 0; n < 4; n++)
432
703k
        {
433
703k
            x[n] = input_im2[n] - input_re2[3-n];
434
703k
        }
435
175k
        DCT3_4_unscaled(x, x);
436
175k
        QMF_IM(X_hybrid[i][6]) = x[1];
437
175k
        QMF_IM(X_hybrid[i][4]) = x[3];
438
175k
        QMF_IM(X_hybrid[i][2]) = x[2];
439
175k
        QMF_IM(X_hybrid[i][0]) = x[0];
440
175k
    }
441
5.65k
}
442
443
static void INLINE DCT3_6_unscaled(real_t *y, real_t *x)
444
246k
{
445
246k
    real_t f0, f1, f2, f3, f4, f5, f6, f7;
446
447
246k
    f0 = MUL_F(x[3], FRAC_CONST(0.70710678118655));
448
246k
    f1 = x[0] + f0;
449
246k
    f2 = x[0] - f0;
450
246k
    f3 = MUL_F((x[1] - x[5]), FRAC_CONST(0.70710678118655));
451
246k
    f4 = MUL_F(x[2], FRAC_CONST(0.86602540378444)) + MUL_F(x[4], FRAC_CONST(0.5));
452
246k
    f5 = f4 - x[4];
453
246k
    f6 = MUL_F(x[1], FRAC_CONST(0.96592582628907)) + MUL_F(x[5], FRAC_CONST(0.25881904510252));
454
246k
    f7 = f6 - f3;
455
246k
    y[0] = f1 + f6 + f4;
456
246k
    y[1] = f2 + f3 - x[4];
457
246k
    y[2] = f7 + f2 - f5;
458
246k
    y[3] = f1 - f7 - f5;
459
246k
    y[4] = f1 - f3 - x[4];
460
246k
    y[5] = f2 - f6 + f4;
461
246k
}
462
463
/* complex filter, size 12 */
464
static void channel_filter12(hyb_info *hyb, uint8_t frame_len, const real_t *filter,
465
                             qmf_t *buffer, qmf_t **X_hybrid)
466
2.00k
{
467
2.00k
    uint8_t i, n;
468
2.00k
    real_t input_re1[6], input_re2[6], input_im1[6], input_im2[6];
469
2.00k
    real_t out_re1[6], out_re2[6], out_im1[6], out_im2[6];
470
2.00k
    (void)hyb;  /* TODO: remove parameter? */
471
472
63.5k
    for (i = 0; i < frame_len; i++)
473
61.5k
    {
474
430k
        for (n = 0; n < 6; n++)
475
369k
        {
476
369k
            if (n == 0)
477
61.5k
            {
478
61.5k
                input_re1[0] = MUL_F(QMF_RE(buffer[6+i]), filter[6]);
479
61.5k
                input_re2[0] = MUL_F(QMF_IM(buffer[6+i]), filter[6]);
480
307k
            } else {
481
307k
                input_re1[6-n] = MUL_F((QMF_RE(buffer[n+i]) + QMF_RE(buffer[12-n+i])), filter[n]);
482
307k
                input_re2[6-n] = MUL_F((QMF_IM(buffer[n+i]) + QMF_IM(buffer[12-n+i])), filter[n]);
483
307k
            }
484
369k
            input_im2[n] = MUL_F((QMF_RE(buffer[n+i]) - QMF_RE(buffer[12-n+i])), filter[n]);
485
369k
            input_im1[n] = MUL_F((QMF_IM(buffer[n+i]) - QMF_IM(buffer[12-n+i])), filter[n]);
486
369k
        }
487
488
61.5k
        DCT3_6_unscaled(out_re1, input_re1);
489
61.5k
        DCT3_6_unscaled(out_re2, input_re2);
490
491
61.5k
        DCT3_6_unscaled(out_im1, input_im1);
492
61.5k
        DCT3_6_unscaled(out_im2, input_im2);
493
494
246k
        for (n = 0; n < 6; n += 2)
495
184k
        {
496
184k
            QMF_RE(X_hybrid[i][n]) = out_re1[n] - out_im1[n];
497
184k
            QMF_IM(X_hybrid[i][n]) = out_re2[n] + out_im2[n];
498
184k
            QMF_RE(X_hybrid[i][n+1]) = out_re1[n+1] + out_im1[n+1];
499
184k
            QMF_IM(X_hybrid[i][n+1]) = out_re2[n+1] - out_im2[n+1];
500
501
184k
            QMF_RE(X_hybrid[i][10-n]) = out_re1[n+1] - out_im1[n+1];
502
184k
            QMF_IM(X_hybrid[i][10-n]) = out_re2[n+1] + out_im2[n+1];
503
184k
            QMF_RE(X_hybrid[i][11-n]) = out_re1[n] + out_im1[n];
504
184k
            QMF_IM(X_hybrid[i][11-n]) = out_re2[n] - out_im2[n];
505
184k
        }
506
61.5k
    }
507
2.00k
}
508
509
/* Hybrid analysis: further split up QMF subbands
510
 * to improve frequency resolution
511
 */
512
static void hybrid_analysis(hyb_info *hyb, qmf_t X[32][64], qmf_t X_hybrid[32][32],
513
                            uint8_t use34, uint8_t numTimeSlotsRate)
514
5.65k
{
515
5.65k
    uint8_t k, n, band;
516
5.65k
    uint8_t offset = 0;
517
5.65k
    uint8_t qmf_bands = (use34) ? 5 : 3;
518
5.65k
    uint8_t *resolution = (use34) ? hyb->resolution34 : hyb->resolution20;
519
520
26.6k
    for (band = 0; band < qmf_bands; band++)
521
20.9k
    {
522
        /* build working buffer */
523
20.9k
        memcpy(hyb->work, hyb->buffer[band], 12 * sizeof(qmf_t));
524
525
        /* add new samples */
526
671k
        for (n = 0; n < hyb->frame_len; n++)
527
650k
        {
528
650k
            QMF_RE(hyb->work[12 + n]) = QMF_RE(X[n + 6 /*delay*/][band]);
529
650k
            QMF_IM(hyb->work[12 + n]) = QMF_IM(X[n + 6 /*delay*/][band]);
530
650k
        }
531
532
        /* store samples */
533
20.9k
        memcpy(hyb->buffer[band], hyb->work + hyb->frame_len, 12 * sizeof(qmf_t));
534
535
536
20.9k
        switch(resolution[band])
537
20.9k
        {
538
7.29k
        case 2:
539
            /* Type B real filter, Q[p] = 2 */
540
7.29k
            channel_filter2(hyb, hyb->frame_len, p2_13_20, hyb->work, hyb->temp);
541
7.29k
            break;
542
6.02k
        case 4:
543
            /* Type A complex filter, Q[p] = 4 */
544
6.02k
            channel_filter4(hyb, hyb->frame_len, p4_13_34, hyb->work, hyb->temp);
545
6.02k
            break;
546
5.65k
        case 8:
547
            /* Type A complex filter, Q[p] = 8 */
548
5.65k
            channel_filter8(hyb, hyb->frame_len, (use34) ? p8_13_34 : p8_13_20,
549
5.65k
                hyb->work, hyb->temp);
550
5.65k
            break;
551
2.00k
        case 12:
552
            /* Type A complex filter, Q[p] = 12 */
553
2.00k
            channel_filter12(hyb, hyb->frame_len, p12_13_34, hyb->work, hyb->temp);
554
2.00k
            break;
555
20.9k
        }
556
557
671k
        for (n = 0; n < hyb->frame_len; n++)
558
650k
        {
559
3.99M
            for (k = 0; k < resolution[band]; k++)
560
3.34M
            {
561
3.34M
                QMF_RE(X_hybrid[n][offset + k]) = QMF_RE(hyb->temp[n][k]);
562
3.34M
                QMF_IM(X_hybrid[n][offset + k]) = QMF_IM(hyb->temp[n][k]);
563
3.34M
            }
564
650k
        }
565
20.9k
        offset += resolution[band];
566
20.9k
    }
567
568
    /* group hybrid channels */
569
5.65k
    if (!use34)
570
3.64k
    {
571
118k
        for (n = 0; n < numTimeSlotsRate; n++)
572
114k
        {
573
114k
            QMF_RE(X_hybrid[n][3]) += QMF_RE(X_hybrid[n][4]);
574
114k
            QMF_IM(X_hybrid[n][3]) += QMF_IM(X_hybrid[n][4]);
575
114k
            QMF_RE(X_hybrid[n][4]) = 0;
576
114k
            QMF_IM(X_hybrid[n][4]) = 0;
577
578
114k
            QMF_RE(X_hybrid[n][2]) += QMF_RE(X_hybrid[n][5]);
579
114k
            QMF_IM(X_hybrid[n][2]) += QMF_IM(X_hybrid[n][5]);
580
114k
            QMF_RE(X_hybrid[n][5]) = 0;
581
114k
            QMF_IM(X_hybrid[n][5]) = 0;
582
114k
        }
583
3.64k
    }
584
5.65k
}
585
586
static void hybrid_synthesis(hyb_info *hyb, qmf_t X[32][64], qmf_t X_hybrid[32][32],
587
                             uint8_t use34, uint8_t numTimeSlotsRate)
588
11.3k
{
589
11.3k
    uint8_t k, n, band;
590
11.3k
    uint8_t offset = 0;
591
11.3k
    uint8_t qmf_bands = (use34) ? 5 : 3;
592
11.3k
    uint8_t *resolution = (use34) ? hyb->resolution34 : hyb->resolution20;
593
11.3k
    (void)numTimeSlotsRate;  /* TODO: remove parameter? */
594
595
53.2k
    for(band = 0; band < qmf_bands; band++)
596
41.9k
    {
597
1.34M
        for (n = 0; n < hyb->frame_len; n++)
598
1.30M
        {
599
1.30M
            QMF_RE(X[n][band]) = 0;
600
1.30M
            QMF_IM(X[n][band]) = 0;
601
602
7.98M
            for (k = 0; k < resolution[band]; k++)
603
6.68M
            {
604
6.68M
                QMF_RE(X[n][band]) += QMF_RE(X_hybrid[n][offset + k]);
605
6.68M
                QMF_IM(X[n][band]) += QMF_IM(X_hybrid[n][offset + k]);
606
6.68M
            }
607
1.30M
        }
608
41.9k
        offset += resolution[band];
609
41.9k
    }
610
11.3k
}
611
612
/* limits the value i to the range [min,max] */
613
static int8_t delta_clip(int8_t i, int8_t min, int8_t max)
614
81.2k
{
615
81.2k
    if (i < min)
616
9.00k
        return min;
617
72.2k
    else if (i > max)
618
1.84k
        return max;
619
70.4k
    else
620
70.4k
        return i;
621
81.2k
}
622
623
//int iid = 0;
624
625
/* delta decode array */
626
static void delta_decode(uint8_t enable, int8_t *index, int8_t *index_prev,
627
                         uint8_t dt_flag, uint8_t nr_par, uint8_t stride,
628
                         int8_t min_index, int8_t max_index)
629
18.5k
{
630
18.5k
    int8_t i;
631
632
18.5k
    if (enable == 1)
633
8.07k
    {
634
8.07k
        if (dt_flag == 0)
635
5.52k
        {
636
            /* delta coded in frequency direction */
637
5.52k
            index[0] = 0 + index[0];
638
5.52k
            index[0] = delta_clip(index[0], min_index, max_index);
639
640
59.3k
            for (i = 1; i < nr_par; i++)
641
53.7k
            {
642
53.7k
                index[i] = index[i-1] + index[i];
643
53.7k
                index[i] = delta_clip(index[i], min_index, max_index);
644
53.7k
            }
645
5.52k
        } else {
646
            /* delta coded in time direction */
647
24.4k
            for (i = 0; i < nr_par; i++)
648
21.9k
            {
649
                //int8_t tmp2;
650
                //int8_t tmp = index[i];
651
652
                //printf("%d %d\n", index_prev[i*stride], index[i]);
653
                //printf("%d\n", index[i]);
654
655
21.9k
                index[i] = index_prev[i*stride] + index[i];
656
                //tmp2 = index[i];
657
21.9k
                index[i] = delta_clip(index[i], min_index, max_index);
658
659
                //if (iid)
660
                //{
661
                //    if (index[i] == 7)
662
                //    {
663
                //        printf("%d %d %d\n", index_prev[i*stride], tmp, tmp2);
664
                //    }
665
                //}
666
21.9k
            }
667
2.55k
        }
668
10.4k
    } else {
669
        /* set indices to zero */
670
19.5k
        for (i = 0; i < nr_par; i++)
671
9.04k
        {
672
9.04k
            index[i] = 0;
673
9.04k
        }
674
10.4k
    }
675
676
    /* coarse */
677
18.5k
    if (stride == 2)
678
11.3k
    {
679
37.6k
        for (i = (nr_par<<1)-1; i > 0; i--)
680
26.3k
        {
681
26.3k
            index[i] = index[i>>1];
682
26.3k
        }
683
11.3k
    }
684
18.5k
}
685
686
/* delta modulo decode array */
687
/* in: log2 value of the modulo value to allow using AND instead of MOD */
688
static void delta_modulo_decode(uint8_t enable, int8_t *index, int8_t *index_prev,
689
                                uint8_t dt_flag, uint8_t nr_par, uint8_t stride,
690
                                int8_t and_modulo)
691
18.5k
{
692
18.5k
    int8_t i;
693
694
18.5k
    if (enable == 1)
695
5.57k
    {
696
5.57k
        if (dt_flag == 0)
697
2.30k
        {
698
            /* delta coded in frequency direction */
699
2.30k
            index[0] = 0 + index[0];
700
2.30k
            index[0] &= and_modulo;
701
702
8.04k
            for (i = 1; i < nr_par; i++)
703
5.74k
            {
704
5.74k
                index[i] = index[i-1] + index[i];
705
5.74k
                index[i] &= and_modulo;
706
5.74k
            }
707
3.26k
        } else {
708
            /* delta coded in time direction */
709
9.12k
            for (i = 0; i < nr_par; i++)
710
5.85k
            {
711
5.85k
                index[i] = index_prev[i*stride] + index[i];
712
5.85k
                index[i] &= and_modulo;
713
5.85k
            }
714
3.26k
        }
715
12.9k
    } else {
716
        /* set indices to zero */
717
46.9k
        for (i = 0; i < nr_par; i++)
718
33.9k
        {
719
33.9k
            index[i] = 0;
720
33.9k
        }
721
12.9k
    }
722
723
    /* coarse */
724
18.5k
    if (stride == 2)
725
0
    {
726
0
        index[0] = 0;
727
0
        for (i = (nr_par<<1)-1; i > 0; i--)
728
0
        {
729
0
            index[i] = index[i>>1];
730
0
        }
731
0
    }
732
18.5k
}
733
734
#ifdef PS_LOW_POWER
735
static void map34indexto20(int8_t *index, uint8_t bins)
736
{
737
    index[0] = (2*index[0]+index[1])/3;
738
    index[1] = (index[1]+2*index[2])/3;
739
    index[2] = (2*index[3]+index[4])/3;
740
    index[3] = (index[4]+2*index[5])/3;
741
    index[4] = (index[6]+index[7])/2;
742
    index[5] = (index[8]+index[9])/2;
743
    index[6] = index[10];
744
    index[7] = index[11];
745
    index[8] = (index[12]+index[13])/2;
746
    index[9] = (index[14]+index[15])/2;
747
    index[10] = index[16];
748
749
    if (bins == 34)
750
    {
751
        index[11] = index[17];
752
        index[12] = index[18];
753
        index[13] = index[19];
754
        index[14] = (index[20]+index[21])/2;
755
        index[15] = (index[22]+index[23])/2;
756
        index[16] = (index[24]+index[25])/2;
757
        index[17] = (index[26]+index[27])/2;
758
        index[18] = (index[28]+index[29]+index[30]+index[31])/4;
759
        index[19] = (index[32]+index[33])/2;
760
    }
761
}
762
#endif
763
764
static void map20indexto34(int8_t *index, uint8_t bins)
765
7.22k
{
766
7.22k
    index[0] = index[0];
767
7.22k
    index[1] = (index[0] + index[1])/2;
768
7.22k
    index[2] = index[1];
769
7.22k
    index[3] = index[2];
770
7.22k
    index[4] = (index[2] + index[3])/2;
771
7.22k
    index[5] = index[3];
772
7.22k
    index[6] = index[4];
773
7.22k
    index[7] = index[4];
774
7.22k
    index[8] = index[5];
775
7.22k
    index[9] = index[5];
776
7.22k
    index[10] = index[6];
777
7.22k
    index[11] = index[7];
778
7.22k
    index[12] = index[8];
779
7.22k
    index[13] = index[8];
780
7.22k
    index[14] = index[9];
781
7.22k
    index[15] = index[9];
782
7.22k
    index[16] = index[10];
783
784
7.22k
    if (bins == 34)
785
3.30k
    {
786
3.30k
        index[17] = index[11];
787
3.30k
        index[18] = index[12];
788
3.30k
        index[19] = index[13];
789
3.30k
        index[20] = index[14];
790
3.30k
        index[21] = index[14];
791
3.30k
        index[22] = index[15];
792
3.30k
        index[23] = index[15];
793
3.30k
        index[24] = index[16];
794
3.30k
        index[25] = index[16];
795
3.30k
        index[26] = index[17];
796
3.30k
        index[27] = index[17];
797
3.30k
        index[28] = index[18];
798
3.30k
        index[29] = index[18];
799
3.30k
        index[30] = index[18];
800
3.30k
        index[31] = index[18];
801
3.30k
        index[32] = index[19];
802
3.30k
        index[33] = index[19];
803
3.30k
    }
804
7.22k
}
805
806
/* parse the bitstream data decoded in ps_data() */
807
static void ps_data_decode(ps_info *ps)
808
5.65k
{
809
5.65k
    uint8_t env, bin;
810
811
    /* ps data not available, use data from previous frame */
812
5.65k
    if (ps->ps_data_available == 0)
813
1.56k
    {
814
1.56k
        ps->num_env = 0;
815
1.56k
    }
816
817
14.9k
    for (env = 0; env < ps->num_env; env++)
818
9.28k
    {
819
9.28k
        int8_t *iid_index_prev;
820
9.28k
        int8_t *icc_index_prev;
821
9.28k
        int8_t *ipd_index_prev;
822
9.28k
        int8_t *opd_index_prev;
823
824
9.28k
        int8_t num_iid_steps = (ps->iid_mode < 3) ? 7 : 15 /*fine quant*/;
825
826
9.28k
        if (env == 0)
827
2.71k
        {
828
            /* take last envelope from previous frame */
829
2.71k
            iid_index_prev = ps->iid_index_prev;
830
2.71k
            icc_index_prev = ps->icc_index_prev;
831
2.71k
            ipd_index_prev = ps->ipd_index_prev;
832
2.71k
            opd_index_prev = ps->opd_index_prev;
833
6.56k
        } else {
834
            /* take index values from previous envelope */
835
6.56k
            iid_index_prev = ps->iid_index[env - 1];
836
6.56k
            icc_index_prev = ps->icc_index[env - 1];
837
6.56k
            ipd_index_prev = ps->ipd_index[env - 1];
838
6.56k
            opd_index_prev = ps->opd_index[env - 1];
839
6.56k
        }
840
841
//        iid = 1;
842
        /* delta decode iid parameters */
843
9.28k
        delta_decode(ps->enable_iid, ps->iid_index[env], iid_index_prev,
844
9.28k
            ps->iid_dt[env], ps->nr_iid_par,
845
9.28k
            (ps->iid_mode == 0 || ps->iid_mode == 3) ? 2 : 1,
846
9.28k
            -num_iid_steps, num_iid_steps);
847
//        iid = 0;
848
849
        /* delta decode icc parameters */
850
9.28k
        delta_decode(ps->enable_icc, ps->icc_index[env], icc_index_prev,
851
9.28k
            ps->icc_dt[env], ps->nr_icc_par,
852
9.28k
            (ps->icc_mode == 0 || ps->icc_mode == 3) ? 2 : 1,
853
9.28k
            0, 7);
854
855
        /* delta modulo decode ipd parameters */
856
9.28k
        delta_modulo_decode(ps->enable_ipdopd, ps->ipd_index[env], ipd_index_prev,
857
9.28k
            ps->ipd_dt[env], ps->nr_ipdopd_par, 1, 7);
858
859
        /* delta modulo decode opd parameters */
860
9.28k
        delta_modulo_decode(ps->enable_ipdopd, ps->opd_index[env], opd_index_prev,
861
9.28k
            ps->opd_dt[env], ps->nr_ipdopd_par, 1, 7);
862
9.28k
    }
863
864
    /* handle error case */
865
5.65k
    if (ps->num_env == 0)
866
2.93k
    {
867
        /* force to 1 */
868
2.93k
        ps->num_env = 1;
869
870
2.93k
        if (ps->enable_iid)
871
2.04k
        {
872
71.5k
            for (bin = 0; bin < 34; bin++)
873
69.4k
                ps->iid_index[0][bin] = ps->iid_index_prev[bin];
874
2.04k
        } else {
875
31.3k
            for (bin = 0; bin < 34; bin++)
876
30.4k
                ps->iid_index[0][bin] = 0;
877
895
        }
878
879
2.93k
        if (ps->enable_icc)
880
1.64k
        {
881
57.5k
            for (bin = 0; bin < 34; bin++)
882
55.8k
                ps->icc_index[0][bin] = ps->icc_index_prev[bin];
883
1.64k
        } else {
884
45.3k
            for (bin = 0; bin < 34; bin++)
885
44.0k
                ps->icc_index[0][bin] = 0;
886
1.29k
        }
887
888
2.93k
        if (ps->enable_ipdopd)
889
463
        {
890
8.33k
            for (bin = 0; bin < 17; bin++)
891
7.87k
            {
892
7.87k
                ps->ipd_index[0][bin] = ps->ipd_index_prev[bin];
893
7.87k
                ps->opd_index[0][bin] = ps->opd_index_prev[bin];
894
7.87k
            }
895
2.47k
        } else {
896
44.5k
            for (bin = 0; bin < 17; bin++)
897
42.0k
            {
898
42.0k
                ps->ipd_index[0][bin] = 0;
899
42.0k
                ps->opd_index[0][bin] = 0;
900
42.0k
            }
901
2.47k
        }
902
2.93k
    }
903
904
    /* update previous indices */
905
197k
    for (bin = 0; bin < 34; bin++)
906
192k
        ps->iid_index_prev[bin] = ps->iid_index[ps->num_env-1][bin];
907
197k
    for (bin = 0; bin < 34; bin++)
908
192k
        ps->icc_index_prev[bin] = ps->icc_index[ps->num_env-1][bin];
909
101k
    for (bin = 0; bin < 17; bin++)
910
96.1k
    {
911
96.1k
        ps->ipd_index_prev[bin] = ps->ipd_index[ps->num_env-1][bin];
912
96.1k
        ps->opd_index_prev[bin] = ps->opd_index[ps->num_env-1][bin];
913
96.1k
    }
914
915
5.65k
    ps->ps_data_available = 0;
916
917
5.65k
    if (ps->frame_class == 0)
918
3.79k
    {
919
3.79k
        ps->border_position[0] = 0;
920
6.87k
        for (env = 1; env < ps->num_env; env++)
921
3.07k
        {
922
3.07k
            ps->border_position[env] = (env * ps->numTimeSlotsRate) / ps->num_env;
923
3.07k
        }
924
3.79k
        ps->border_position[ps->num_env] = ps->numTimeSlotsRate;
925
3.79k
    } else {
926
1.85k
        ps->border_position[0] = 0;
927
928
1.85k
        if (ps->border_position[ps->num_env] < ps->numTimeSlotsRate)
929
1.57k
        {
930
55.0k
            for (bin = 0; bin < 34; bin++)
931
53.5k
            {
932
53.5k
                ps->iid_index[ps->num_env][bin] = ps->iid_index[ps->num_env-1][bin];
933
53.5k
                ps->icc_index[ps->num_env][bin] = ps->icc_index[ps->num_env-1][bin];
934
53.5k
            }
935
28.3k
            for (bin = 0; bin < 17; bin++)
936
26.7k
            {
937
26.7k
                ps->ipd_index[ps->num_env][bin] = ps->ipd_index[ps->num_env-1][bin];
938
26.7k
                ps->opd_index[ps->num_env][bin] = ps->opd_index[ps->num_env-1][bin];
939
26.7k
            }
940
1.57k
            ps->num_env++;
941
1.57k
            ps->border_position[ps->num_env] = ps->numTimeSlotsRate;
942
1.57k
        }
943
944
6.92k
        for (env = 1; env < ps->num_env; env++)
945
5.06k
        {
946
5.06k
            int8_t thr = ps->numTimeSlotsRate - (ps->num_env - env);
947
948
5.06k
            if (ps->border_position[env] > thr)
949
944
            {
950
944
                ps->border_position[env] = thr;
951
4.12k
            } else {
952
4.12k
                thr = ps->border_position[env-1]+1;
953
4.12k
                if (ps->border_position[env] < thr)
954
2.00k
                {
955
2.00k
                    ps->border_position[env] = thr;
956
2.00k
                }
957
4.12k
            }
958
5.06k
        }
959
1.85k
    }
960
961
    /* make sure that the indices of all parameters can be mapped
962
     * to the same hybrid synthesis filterbank
963
     */
964
#ifdef PS_LOW_POWER
965
    for (env = 0; env < ps->num_env; env++)
966
    {
967
        if (ps->iid_mode == 2 || ps->iid_mode == 5)
968
            map34indexto20(ps->iid_index[env], 34);
969
        if (ps->icc_mode == 2 || ps->icc_mode == 5)
970
            map34indexto20(ps->icc_index[env], 34);
971
972
        /* disable ipd/opd */
973
        for (bin = 0; bin < 17; bin++)
974
        {
975
            ps->aaIpdIndex[env][bin] = 0;
976
            ps->aaOpdIndex[env][bin] = 0;
977
        }
978
    }
979
#else
980
5.65k
    if (ps->use34hybrid_bands)
981
2.00k
    {
982
5.43k
        for (env = 0; env < ps->num_env; env++)
983
3.42k
        {
984
3.42k
            if (ps->iid_mode != 2 && ps->iid_mode != 5)
985
1.96k
                map20indexto34(ps->iid_index[env], 34);
986
3.42k
            if (ps->icc_mode != 2 && ps->icc_mode != 5)
987
1.34k
                map20indexto34(ps->icc_index[env], 34);
988
3.42k
            if (ps->ipd_mode != 2 && ps->ipd_mode != 5)
989
1.96k
            {
990
1.96k
                map20indexto34(ps->ipd_index[env], 17);
991
1.96k
                map20indexto34(ps->opd_index[env], 17);
992
1.96k
            }
993
3.42k
        }
994
2.00k
    }
995
5.65k
#endif
996
997
#if 0
998
    for (env = 0; env < ps->num_env; env++)
999
    {
1000
        printf("iid[env:%d]:", env);
1001
        for (bin = 0; bin < 34; bin++)
1002
        {
1003
            printf(" %d", ps->iid_index[env][bin]);
1004
        }
1005
        printf("\n");
1006
    }
1007
    for (env = 0; env < ps->num_env; env++)
1008
    {
1009
        printf("icc[env:%d]:", env);
1010
        for (bin = 0; bin < 34; bin++)
1011
        {
1012
            printf(" %d", ps->icc_index[env][bin]);
1013
        }
1014
        printf("\n");
1015
    }
1016
    for (env = 0; env < ps->num_env; env++)
1017
    {
1018
        printf("ipd[env:%d]:", env);
1019
        for (bin = 0; bin < 17; bin++)
1020
        {
1021
            printf(" %d", ps->ipd_index[env][bin]);
1022
        }
1023
        printf("\n");
1024
    }
1025
    for (env = 0; env < ps->num_env; env++)
1026
    {
1027
        printf("opd[env:%d]:", env);
1028
        for (bin = 0; bin < 17; bin++)
1029
        {
1030
            printf(" %d", ps->opd_index[env][bin]);
1031
        }
1032
        printf("\n");
1033
    }
1034
    printf("\n");
1035
#endif
1036
5.65k
}
1037
1038
/* decorrelate the mono signal using an allpass filter */
1039
static void ps_decorrelate(ps_info *ps, qmf_t X_left[38][64], qmf_t X_right[38][64],
1040
                           qmf_t X_hybrid_left[32][32], qmf_t X_hybrid_right[32][32])
1041
5.65k
{
1042
5.65k
    uint8_t gr, n, bk;
1043
5.65k
    uint8_t temp_delay = 0;
1044
5.65k
    uint8_t sb, maxsb;
1045
5.65k
    const complex_t *Phi_Fract_SubQmf;
1046
5.65k
    uint8_t temp_delay_ser[NO_ALLPASS_LINKS];
1047
5.65k
    real_t P_SmoothPeakDecayDiffNrg, nrg;
1048
5.65k
    real_t P[32][34];
1049
5.65k
    real_t G_TransientRatio[32][34] = {{0}};
1050
5.65k
    complex_t inputLeft;
1051
1052
1053
    /* chose hybrid filterbank: 20 or 34 band case */
1054
5.65k
    if (ps->use34hybrid_bands)
1055
2.00k
    {
1056
2.00k
        Phi_Fract_SubQmf = Phi_Fract_SubQmf34;
1057
3.64k
    } else{
1058
3.64k
        Phi_Fract_SubQmf = Phi_Fract_SubQmf20;
1059
3.64k
    }
1060
1061
    /* clear the energy values */
1062
186k
    for (n = 0; n < 32; n++)
1063
180k
    {
1064
6.33M
        for (bk = 0; bk < 34; bk++)
1065
6.15M
        {
1066
6.15M
            P[n][bk] = 0;
1067
6.15M
        }
1068
180k
    }
1069
1070
    /* calculate the energy in each parameter band b(k) */
1071
186k
    for (gr = 0; gr < ps->num_groups; gr++)
1072
180k
    {
1073
        /* select the parameter index b(k) to which this group belongs */
1074
180k
        bk = (~NEGATE_IPD_MASK) & ps->map_group2bk[gr];
1075
1076
        /* select the upper subband border for this group */
1077
180k
        maxsb = (gr < ps->num_hybrid_groups) ? ps->group_border[gr]+1 : ps->group_border[gr+1];
1078
1079
622k
        for (sb = ps->group_border[gr]; sb < maxsb; sb++)
1080
441k
        {
1081
14.1M
            for (n = ps->border_position[0]; n < ps->border_position[ps->num_env]; n++)
1082
13.7M
            {
1083
#ifdef FIXED_POINT
1084
                uint32_t in_re, in_im;
1085
#endif
1086
1087
                /* input from hybrid subbands or QMF subbands */
1088
13.7M
                if (gr < ps->num_hybrid_groups)
1089
3.11M
                {
1090
3.11M
                    RE(inputLeft) = QMF_RE(X_hybrid_left[n][sb]);
1091
3.11M
                    IM(inputLeft) = QMF_IM(X_hybrid_left[n][sb]);
1092
10.6M
                } else {
1093
10.6M
                    RE(inputLeft) = QMF_RE(X_left[n][sb]);
1094
10.6M
                    IM(inputLeft) = QMF_IM(X_left[n][sb]);
1095
10.6M
                }
1096
1097
                /* accumulate energy */
1098
#ifdef FIXED_POINT
1099
                /* NOTE: all input is scaled by 2^(-5) because of fixed point QMF
1100
                 * meaning that P will be scaled by 2^(-10) compared to floating point version
1101
                 */
1102
                in_re = ((abs(RE(inputLeft))+(1<<(REAL_BITS-1)))>>REAL_BITS);
1103
                in_im = ((abs(IM(inputLeft))+(1<<(REAL_BITS-1)))>>REAL_BITS);
1104
                P[n][bk] += in_re*in_re + in_im*in_im;
1105
#else
1106
13.7M
                P[n][bk] += MUL_R(RE(inputLeft),RE(inputLeft)) + MUL_R(IM(inputLeft),IM(inputLeft));
1107
13.7M
#endif
1108
13.7M
            }
1109
441k
        }
1110
180k
    }
1111
1112
#if 0
1113
    for (n = 0; n < 32; n++)
1114
    {
1115
        for (bk = 0; bk < 34; bk++)
1116
        {
1117
#ifdef FIXED_POINT
1118
            printf("%d %d: %d\n", n, bk, P[n][bk] /*/(float)REAL_PRECISION*/);
1119
#else
1120
            printf("%d %d: %f\n", n, bk, P[n][bk]/1024.0);
1121
#endif
1122
        }
1123
    }
1124
#endif
1125
1126
    /* calculate transient reduction ratio for each parameter band b(k) */
1127
146k
    for (bk = 0; bk < ps->nr_par_bands; bk++)
1128
141k
    {
1129
4.52M
        for (n = ps->border_position[0]; n < ps->border_position[ps->num_env]; n++)
1130
4.38M
        {
1131
4.38M
            const real_t gamma = COEF_CONST(1.5);
1132
1133
4.38M
            ps->P_PeakDecayNrg[bk] = MUL_F(ps->P_PeakDecayNrg[bk], ps->alpha_decay);
1134
4.38M
            if (ps->P_PeakDecayNrg[bk] < P[n][bk])
1135
51.6k
                ps->P_PeakDecayNrg[bk] = P[n][bk];
1136
1137
            /* apply smoothing filter to peak decay energy */
1138
4.38M
            P_SmoothPeakDecayDiffNrg = ps->P_SmoothPeakDecayDiffNrg_prev[bk];
1139
4.38M
            P_SmoothPeakDecayDiffNrg += MUL_F((ps->P_PeakDecayNrg[bk] - P[n][bk] - ps->P_SmoothPeakDecayDiffNrg_prev[bk]), ps->alpha_smooth);
1140
4.38M
            ps->P_SmoothPeakDecayDiffNrg_prev[bk] = P_SmoothPeakDecayDiffNrg;
1141
1142
            /* apply smoothing filter to energy */
1143
4.38M
            nrg = ps->P_prev[bk];
1144
4.38M
            nrg += MUL_F((P[n][bk] - ps->P_prev[bk]), ps->alpha_smooth);
1145
4.38M
            ps->P_prev[bk] = nrg;
1146
1147
            /* calculate transient ratio */
1148
4.38M
            if (MUL_C(P_SmoothPeakDecayDiffNrg, gamma) <= nrg)
1149
4.34M
            {
1150
4.34M
                G_TransientRatio[n][bk] = REAL_CONST(1.0);
1151
4.34M
            } else {
1152
38.5k
                G_TransientRatio[n][bk] = DIV_R(nrg, (MUL_C(P_SmoothPeakDecayDiffNrg, gamma)));
1153
38.5k
            }
1154
4.38M
        }
1155
141k
    }
1156
1157
#if 0
1158
    for (n = 0; n < 32; n++)
1159
    {
1160
        for (bk = 0; bk < 34; bk++)
1161
        {
1162
#ifdef FIXED_POINT
1163
            printf("%d %d: %f\n", n, bk, G_TransientRatio[n][bk]/(float)REAL_PRECISION);
1164
#else
1165
            printf("%d %d: %f\n", n, bk, G_TransientRatio[n][bk]);
1166
#endif
1167
        }
1168
    }
1169
#endif
1170
1171
    /* apply stereo decorrelation filter to the signal */
1172
186k
    for (gr = 0; gr < ps->num_groups; gr++)
1173
180k
    {
1174
180k
        if (gr < ps->num_hybrid_groups)
1175
100k
            maxsb = ps->group_border[gr] + 1;
1176
79.9k
        else
1177
79.9k
            maxsb = ps->group_border[gr + 1];
1178
1179
        /* QMF channel */
1180
622k
        for (sb = ps->group_border[gr]; sb < maxsb; sb++)
1181
441k
        {
1182
441k
            real_t g_DecaySlope;
1183
441k
            real_t g_DecaySlope_filt[NO_ALLPASS_LINKS];
1184
1185
            /* g_DecaySlope: [0..1] */
1186
441k
            if (gr < ps->num_hybrid_groups || sb <= ps->decay_cutoff)
1187
106k
            {
1188
106k
                g_DecaySlope = FRAC_CONST(1.0);
1189
335k
            } else {
1190
335k
                int8_t decay = ps->decay_cutoff - sb;
1191
335k
                if (decay <= -20 /* -1/DECAY_SLOPE */)
1192
227k
                {
1193
227k
                    g_DecaySlope = 0;
1194
227k
                } else {
1195
                    /* decay(int)*decay_slope(frac) = g_DecaySlope(frac) */
1196
107k
                    g_DecaySlope = FRAC_CONST(1.0) + DECAY_SLOPE * decay;
1197
107k
                }
1198
335k
            }
1199
1200
            /* calculate g_DecaySlope_filt for every n multiplied by filter_a[n] */
1201
1.76M
            for (n = 0; n < NO_ALLPASS_LINKS; n++)
1202
1.32M
            {
1203
1.32M
                g_DecaySlope_filt[n] = MUL_F(g_DecaySlope, filter_a[n]);
1204
1.32M
            }
1205
1206
1207
            /* set delay indices */
1208
441k
            temp_delay = ps->saved_delay;
1209
1.76M
            for (n = 0; n < NO_ALLPASS_LINKS; n++)
1210
1.32M
                temp_delay_ser[n] = ps->delay_buf_index_ser[n];
1211
1212
14.1M
            for (n = ps->border_position[0]; n < ps->border_position[ps->num_env]; n++)
1213
13.7M
            {
1214
13.7M
                complex_t tmp, tmp0, R0;
1215
13.7M
                uint8_t m;
1216
1217
13.7M
                if (gr < ps->num_hybrid_groups)
1218
3.11M
                {
1219
                    /* hybrid filterbank input */
1220
3.11M
                    RE(inputLeft) = QMF_RE(X_hybrid_left[n][sb]);
1221
3.11M
                    IM(inputLeft) = QMF_IM(X_hybrid_left[n][sb]);
1222
10.6M
                } else {
1223
                    /* QMF filterbank input */
1224
10.6M
                    RE(inputLeft) = QMF_RE(X_left[n][sb]);
1225
10.6M
                    IM(inputLeft) = QMF_IM(X_left[n][sb]);
1226
10.6M
                }
1227
1228
13.7M
                if (sb > ps->nr_allpass_bands && gr >= ps->num_hybrid_groups)
1229
7.22M
                {
1230
                    /* delay */
1231
1232
                    /* never hybrid subbands here, always QMF subbands */
1233
7.22M
                    RE(tmp) = RE(ps->delay_Qmf[ps->delay_buf_index_delay[sb]][sb]);
1234
7.22M
                    IM(tmp) = IM(ps->delay_Qmf[ps->delay_buf_index_delay[sb]][sb]);
1235
7.22M
                    RE(R0) = RE(tmp);
1236
7.22M
                    IM(R0) = IM(tmp);
1237
7.22M
                    RE(ps->delay_Qmf[ps->delay_buf_index_delay[sb]][sb]) = RE(inputLeft);
1238
7.22M
                    IM(ps->delay_Qmf[ps->delay_buf_index_delay[sb]][sb]) = IM(inputLeft);
1239
7.22M
                } else {
1240
                    /* allpass filter */
1241
6.52M
                    complex_t Phi_Fract;
1242
1243
                    /* fetch parameters */
1244
6.52M
                    if (gr < ps->num_hybrid_groups)
1245
3.11M
                    {
1246
                        /* select data from the hybrid subbands */
1247
3.11M
                        RE(tmp0) = RE(ps->delay_SubQmf[temp_delay][sb]);
1248
3.11M
                        IM(tmp0) = IM(ps->delay_SubQmf[temp_delay][sb]);
1249
1250
3.11M
                        RE(ps->delay_SubQmf[temp_delay][sb]) = RE(inputLeft);
1251
3.11M
                        IM(ps->delay_SubQmf[temp_delay][sb]) = IM(inputLeft);
1252
1253
3.11M
                        RE(Phi_Fract) = RE(Phi_Fract_SubQmf[sb]);
1254
3.11M
                        IM(Phi_Fract) = IM(Phi_Fract_SubQmf[sb]);
1255
3.40M
                    } else {
1256
                        /* select data from the QMF subbands */
1257
3.40M
                        RE(tmp0) = RE(ps->delay_Qmf[temp_delay][sb]);
1258
3.40M
                        IM(tmp0) = IM(ps->delay_Qmf[temp_delay][sb]);
1259
1260
3.40M
                        RE(ps->delay_Qmf[temp_delay][sb]) = RE(inputLeft);
1261
3.40M
                        IM(ps->delay_Qmf[temp_delay][sb]) = IM(inputLeft);
1262
1263
3.40M
                        RE(Phi_Fract) = RE(Phi_Fract_Qmf[sb]);
1264
3.40M
                        IM(Phi_Fract) = IM(Phi_Fract_Qmf[sb]);
1265
3.40M
                    }
1266
1267
                    /* z^(-2) * Phi_Fract[k] */
1268
6.52M
                    ComplexMult(&RE(tmp), &IM(tmp), RE(tmp0), IM(tmp0), RE(Phi_Fract), IM(Phi_Fract));
1269
1270
6.52M
                    RE(R0) = RE(tmp);
1271
6.52M
                    IM(R0) = IM(tmp);
1272
26.0M
                    for (m = 0; m < NO_ALLPASS_LINKS; m++)
1273
19.5M
                    {
1274
19.5M
                        complex_t Q_Fract_allpass, tmp2;
1275
1276
                        /* fetch parameters */
1277
19.5M
                        if (gr < ps->num_hybrid_groups)
1278
9.35M
                        {
1279
                            /* select data from the hybrid subbands */
1280
9.35M
                            RE(tmp0) = RE(ps->delay_SubQmf_ser[m][temp_delay_ser[m]][sb]);
1281
9.35M
                            IM(tmp0) = IM(ps->delay_SubQmf_ser[m][temp_delay_ser[m]][sb]);
1282
1283
9.35M
                            if (ps->use34hybrid_bands)
1284
5.90M
                            {
1285
5.90M
                                RE(Q_Fract_allpass) = RE(Q_Fract_allpass_SubQmf34[sb][m]);
1286
5.90M
                                IM(Q_Fract_allpass) = IM(Q_Fract_allpass_SubQmf34[sb][m]);
1287
5.90M
                            } else {
1288
3.44M
                                RE(Q_Fract_allpass) = RE(Q_Fract_allpass_SubQmf20[sb][m]);
1289
3.44M
                                IM(Q_Fract_allpass) = IM(Q_Fract_allpass_SubQmf20[sb][m]);
1290
3.44M
                            }
1291
10.2M
                        } else {
1292
                            /* select data from the QMF subbands */
1293
10.2M
                            RE(tmp0) = RE(ps->delay_Qmf_ser[m][temp_delay_ser[m]][sb]);
1294
10.2M
                            IM(tmp0) = IM(ps->delay_Qmf_ser[m][temp_delay_ser[m]][sb]);
1295
1296
10.2M
                            RE(Q_Fract_allpass) = RE(Q_Fract_allpass_Qmf[sb][m]);
1297
10.2M
                            IM(Q_Fract_allpass) = IM(Q_Fract_allpass_Qmf[sb][m]);
1298
10.2M
                        }
1299
1300
                        /* delay by a fraction */
1301
                        /* z^(-d(m)) * Q_Fract_allpass[k,m] */
1302
19.5M
                        ComplexMult(&RE(tmp), &IM(tmp), RE(tmp0), IM(tmp0), RE(Q_Fract_allpass), IM(Q_Fract_allpass));
1303
1304
                        /* -a(m) * g_DecaySlope[k] */
1305
19.5M
                        RE(tmp) += -MUL_F(g_DecaySlope_filt[m], RE(R0));
1306
19.5M
                        IM(tmp) += -MUL_F(g_DecaySlope_filt[m], IM(R0));
1307
1308
                        /* -a(m) * g_DecaySlope[k] * Q_Fract_allpass[k,m] * z^(-d(m)) */
1309
19.5M
                        RE(tmp2) = RE(R0) + MUL_F(g_DecaySlope_filt[m], RE(tmp));
1310
19.5M
                        IM(tmp2) = IM(R0) + MUL_F(g_DecaySlope_filt[m], IM(tmp));
1311
1312
                        /* store sample */
1313
19.5M
                        if (gr < ps->num_hybrid_groups)
1314
9.35M
                        {
1315
9.35M
                            RE(ps->delay_SubQmf_ser[m][temp_delay_ser[m]][sb]) = RE(tmp2);
1316
9.35M
                            IM(ps->delay_SubQmf_ser[m][temp_delay_ser[m]][sb]) = IM(tmp2);
1317
10.2M
                        } else {
1318
10.2M
                            RE(ps->delay_Qmf_ser[m][temp_delay_ser[m]][sb]) = RE(tmp2);
1319
10.2M
                            IM(ps->delay_Qmf_ser[m][temp_delay_ser[m]][sb]) = IM(tmp2);
1320
10.2M
                        }
1321
1322
                        /* store for next iteration (or as output value if last iteration) */
1323
19.5M
                        RE(R0) = RE(tmp);
1324
19.5M
                        IM(R0) = IM(tmp);
1325
19.5M
                    }
1326
6.52M
                }
1327
1328
                /* select b(k) for reading the transient ratio */
1329
13.7M
                bk = (~NEGATE_IPD_MASK) & ps->map_group2bk[gr];
1330
1331
                /* duck if a past transient is found */
1332
13.7M
                RE(R0) = MUL_R(G_TransientRatio[n][bk], RE(R0));
1333
13.7M
                IM(R0) = MUL_R(G_TransientRatio[n][bk], IM(R0));
1334
1335
13.7M
                if (gr < ps->num_hybrid_groups)
1336
3.11M
                {
1337
                    /* hybrid */
1338
3.11M
                    QMF_RE(X_hybrid_right[n][sb]) = RE(R0);
1339
3.11M
                    QMF_IM(X_hybrid_right[n][sb]) = IM(R0);
1340
10.6M
                } else {
1341
                    /* QMF */
1342
10.6M
                    QMF_RE(X_right[n][sb]) = RE(R0);
1343
10.6M
                    QMF_IM(X_right[n][sb]) = IM(R0);
1344
10.6M
                }
1345
1346
                /* Update delay buffer index */
1347
13.7M
                if (++temp_delay >= 2)
1348
6.87M
                {
1349
6.87M
                    temp_delay = 0;
1350
6.87M
                }
1351
1352
                /* update delay indices */
1353
13.7M
                if (sb > ps->nr_allpass_bands && gr >= ps->num_hybrid_groups)
1354
7.22M
                {
1355
                    /* delay_D depends on the samplerate, it can hold the values 14 and 1 */
1356
7.22M
                    if (++ps->delay_buf_index_delay[sb] >= ps->delay_D[sb])
1357
5.24M
                    {
1358
5.24M
                        ps->delay_buf_index_delay[sb] = 0;
1359
5.24M
                    }
1360
7.22M
                }
1361
1362
55.0M
                for (m = 0; m < NO_ALLPASS_LINKS; m++)
1363
41.2M
                {
1364
41.2M
                    if (++temp_delay_ser[m] >= ps->num_sample_delay_ser[m])
1365
10.5M
                    {
1366
10.5M
                        temp_delay_ser[m] = 0;
1367
10.5M
                    }
1368
41.2M
                }
1369
13.7M
            }
1370
441k
        }
1371
180k
    }
1372
1373
    /* update delay indices */
1374
5.65k
    ps->saved_delay = temp_delay;
1375
22.6k
    for (n = 0; n < NO_ALLPASS_LINKS; n++)
1376
16.9k
        ps->delay_buf_index_ser[n] = temp_delay_ser[n];
1377
5.65k
}
1378
1379
#if 0
1380
#ifdef FIXED_POINT
1381
#define step(shift) \
1382
    if ((0x40000000l >> shift) + root <= value)       \
1383
    {                                                 \
1384
        value -= (0x40000000l >> shift) + root;       \
1385
        root = (root >> 1) | (0x40000000l >> shift);  \
1386
    } else {                                          \
1387
        root = root >> 1;                             \
1388
    }
1389
1390
/* fixed point square root approximation */
1391
static real_t ps_sqrt(real_t value)
1392
{
1393
    real_t root = 0;
1394
1395
    step( 0); step( 2); step( 4); step( 6);
1396
    step( 8); step(10); step(12); step(14);
1397
    step(16); step(18); step(20); step(22);
1398
    step(24); step(26); step(28); step(30);
1399
1400
    if (root < value)
1401
        ++root;
1402
1403
    root <<= (REAL_BITS/2);
1404
1405
    return root;
1406
}
1407
#else
1408
#define ps_sqrt(A) sqrt(A)
1409
#endif
1410
#endif
1411
1412
static const real_t ipdopd_cos_tab[] = {
1413
    FRAC_CONST(1.000000000000000),
1414
    FRAC_CONST(0.707106781186548),
1415
    FRAC_CONST(0.000000000000000),
1416
    FRAC_CONST(-0.707106781186547),
1417
    FRAC_CONST(-1.000000000000000),
1418
    FRAC_CONST(-0.707106781186548),
1419
    FRAC_CONST(-0.000000000000000),
1420
    FRAC_CONST(0.707106781186547),
1421
    FRAC_CONST(1.000000000000000)
1422
};
1423
1424
static const real_t ipdopd_sin_tab[] = {
1425
    FRAC_CONST(0.000000000000000),
1426
    FRAC_CONST(0.707106781186547),
1427
    FRAC_CONST(1.000000000000000),
1428
    FRAC_CONST(0.707106781186548),
1429
    FRAC_CONST(0.000000000000000),
1430
    FRAC_CONST(-0.707106781186547),
1431
    FRAC_CONST(-1.000000000000000),
1432
    FRAC_CONST(-0.707106781186548),
1433
    FRAC_CONST(-0.000000000000000)
1434
};
1435
1436
static real_t magnitude_c(complex_t c)
1437
105k
{
1438
#ifdef FIXED_POINT
1439
#define ps_abs(A) (((A) > 0) ? (A) : (-(A)))
1440
#define ALPHA FRAC_CONST(0.948059448969)
1441
#define BETA  FRAC_CONST(0.392699081699)
1442
1443
    real_t abs_inphase = ps_abs(RE(c));
1444
    real_t abs_quadrature = ps_abs(IM(c));
1445
1446
    if (abs_inphase > abs_quadrature) {
1447
        return MUL_F(abs_inphase, ALPHA) + MUL_F(abs_quadrature, BETA);
1448
    } else {
1449
        return MUL_F(abs_quadrature, ALPHA) + MUL_F(abs_inphase, BETA);
1450
    }
1451
#else
1452
105k
    return sqrt(RE(c)*RE(c) + IM(c)*IM(c));
1453
105k
#endif
1454
105k
}
1455
1456
static void ps_mix_phase(ps_info *ps, qmf_t X_left[38][64], qmf_t X_right[38][64],
1457
                         qmf_t X_hybrid_left[32][32], qmf_t X_hybrid_right[32][32])
1458
5.65k
{
1459
5.65k
    uint8_t n;
1460
5.65k
    uint8_t gr;
1461
5.65k
    uint8_t bk = 0;
1462
5.65k
    uint8_t sb, maxsb;
1463
5.65k
    uint8_t env;
1464
5.65k
    uint8_t nr_ipdopd_par;
1465
5.65k
    complex_t h11, h12, h21, h22;  // COEF
1466
5.65k
    complex_t H11, H12, H21, H22;  // COEF
1467
5.65k
    complex_t deltaH11, deltaH12, deltaH21, deltaH22;  // COEF
1468
5.65k
    complex_t tempLeft, tempRight; // FRAC
1469
5.65k
    complex_t phaseLeft, phaseRight; // FRAC
1470
5.65k
    real_t L;
1471
5.65k
    const real_t *sf_iid;
1472
5.65k
    uint8_t no_iid_steps;
1473
1474
5.65k
    if (ps->iid_mode >= 3)
1475
2.34k
    {
1476
2.34k
        no_iid_steps = 15;
1477
2.34k
        sf_iid = sf_iid_fine;
1478
3.31k
    } else {
1479
3.31k
        no_iid_steps = 7;
1480
3.31k
        sf_iid = sf_iid_normal;
1481
3.31k
    }
1482
1483
5.65k
    if (ps->ipd_mode == 0 || ps->ipd_mode == 3)
1484
3.12k
    {
1485
3.12k
        nr_ipdopd_par = 11; /* resolution */
1486
3.12k
    } else {
1487
2.53k
        nr_ipdopd_par = ps->nr_ipdopd_par;
1488
2.53k
    }
1489
1490
186k
    for (gr = 0; gr < ps->num_groups; gr++)
1491
180k
    {
1492
180k
        bk = (~NEGATE_IPD_MASK) & ps->map_group2bk[gr];
1493
1494
        /* use one channel per group in the subqmf domain */
1495
180k
        maxsb = (gr < ps->num_hybrid_groups) ? ps->group_border[gr] + 1 : ps->group_border[gr + 1];
1496
1497
580k
        for (env = 0; env < ps->num_env; env++)
1498
399k
        {
1499
399k
            uint8_t abs_iid = (uint8_t)abs(ps->iid_index[env][bk]);
1500
            /* index range is supposed to be -7...7 or -15...15 depending on iid_mode
1501
                (Table 8.24, ISO/IEC 14496-3:2005).
1502
                if it is outside these boundaries, this is most likely an error. sanitize
1503
                it and try to process further. */
1504
399k
            if (ps->iid_index[env][bk] < -no_iid_steps) {
1505
129
                fprintf(stderr, "Warning: invalid iid_index: %d < %d\n", ps->iid_index[env][bk],
1506
129
                    -no_iid_steps);
1507
129
                ps->iid_index[env][bk] = -no_iid_steps;
1508
129
                abs_iid = no_iid_steps;
1509
399k
            } else if (ps->iid_index[env][bk] > no_iid_steps) {
1510
83
                fprintf(stderr, "Warning: invalid iid_index: %d > %d\n", ps->iid_index[env][bk],
1511
83
                    no_iid_steps);
1512
83
                ps->iid_index[env][bk] = no_iid_steps;
1513
83
                abs_iid = no_iid_steps;
1514
83
            }
1515
399k
            if (ps->icc_index[env][bk] < 0) {
1516
306
                fprintf(stderr, "Warning: invalid icc_index: %d < 0\n", ps->icc_index[env][bk]);
1517
306
                ps->icc_index[env][bk] = 0;
1518
399k
            } else if (ps->icc_index[env][bk] > 7) {
1519
0
                fprintf(stderr, "Warning: invalid icc_index: %d > 7\n", ps->icc_index[env][bk]);
1520
0
                ps->icc_index[env][bk] = 7;
1521
0
            }
1522
1523
399k
            if (ps->icc_mode < 3)
1524
244k
            {
1525
                /* type 'A' mixing as described in 8.6.4.6.2.1 */
1526
244k
                real_t c_1, c_2;  // COEF
1527
244k
                real_t cosa, sina;  // COEF
1528
244k
                real_t cosb, sinb;  // COEF
1529
244k
                real_t ab1, ab2;  // COEF
1530
244k
                real_t ab3, ab4;  // COEF
1531
1532
                /*
1533
                c_1 = sqrt(2.0 / (1.0 + pow(10.0, quant_iid[no_iid_steps + iid_index] / 10.0)));
1534
                c_2 = sqrt(2.0 / (1.0 + pow(10.0, quant_iid[no_iid_steps - iid_index] / 10.0)));
1535
                alpha = 0.5 * acos(quant_rho[icc_index]);
1536
                beta = alpha * ( c_1 - c_2 ) / sqrt(2.0);
1537
                */
1538
1539
                //printf("%d\n", ps->iid_index[env][bk]);
1540
1541
                /* calculate the scalefactors c_1 and c_2 from the intensity differences */
1542
244k
                c_1 = sf_iid[no_iid_steps + ps->iid_index[env][bk]];
1543
244k
                c_2 = sf_iid[no_iid_steps - ps->iid_index[env][bk]];
1544
1545
                /* calculate alpha and beta using the ICC parameters */
1546
244k
                cosa = cos_alphas[ps->icc_index[env][bk]];
1547
244k
                sina = sin_alphas[ps->icc_index[env][bk]];
1548
1549
244k
                if (ps->iid_mode >= 3)
1550
89.3k
                {
1551
89.3k
                    cosb = cos_betas_fine[abs_iid][ps->icc_index[env][bk]];
1552
89.3k
                    sinb = sin_betas_fine[abs_iid][ps->icc_index[env][bk]];
1553
155k
                } else {
1554
155k
                    cosb = cos_betas_normal[abs_iid][ps->icc_index[env][bk]];
1555
155k
                    sinb = sin_betas_normal[abs_iid][ps->icc_index[env][bk]];
1556
155k
                }
1557
1558
244k
                ab1 = MUL_C(cosb, cosa);
1559
244k
                ab2 = MUL_C(sinb, sina);
1560
244k
                ab3 = MUL_C(sinb, cosa);
1561
244k
                ab4 = MUL_C(cosb, sina);
1562
1563
                /* h_xy: COEF */
1564
244k
                RE(h11) = MUL_C(c_2, (ab1 - ab2));
1565
244k
                RE(h12) = MUL_C(c_1, (ab1 + ab2));
1566
244k
                RE(h21) = MUL_C(c_2, (ab3 + ab4));
1567
244k
                RE(h22) = MUL_C(c_1, (ab3 - ab4));
1568
244k
            } else {
1569
                /* type 'B' mixing as described in 8.6.4.6.2.2 */
1570
155k
                real_t sina, cosa;  // COEF
1571
155k
                real_t cosg, sing;  // COEF
1572
1573
                /*
1574
                real_t c, rho, mu, alpha, gamma;
1575
                uint8_t i;
1576
1577
                i = ps->iid_index[env][bk];
1578
                c = (real_t)pow(10.0, ((i)?(((i>0)?1:-1)*quant_iid[((i>0)?i:-i)-1]):0.)/20.0);
1579
                rho = quant_rho[ps->icc_index[env][bk]];
1580
1581
                if (rho == 0.0f && c == 1.)
1582
                {
1583
                    alpha = (real_t)M_PI/4.0f;
1584
                    rho = 0.05f;
1585
                } else {
1586
                    if (rho <= 0.05f)
1587
                    {
1588
                        rho = 0.05f;
1589
                    }
1590
                    alpha = 0.5f*(real_t)atan( (2.0f*c*rho) / (c*c-1.0f) );
1591
1592
                    if (alpha < 0.)
1593
                    {
1594
                        alpha += (real_t)M_PI/2.0f;
1595
                    }
1596
                    if (rho < 0.)
1597
                    {
1598
                        alpha += (real_t)M_PI;
1599
                    }
1600
                }
1601
                mu = c+1.0f/c;
1602
                mu = 1+(4.0f*rho*rho-4.0f)/(mu*mu);
1603
                gamma = (real_t)atan(sqrt((1.0f-sqrt(mu))/(1.0f+sqrt(mu))));
1604
                */
1605
1606
155k
                if (ps->iid_mode >= 3)
1607
70.5k
                {
1608
70.5k
                    cosa = sincos_alphas_B_fine[no_iid_steps + ps->iid_index[env][bk]][ps->icc_index[env][bk]];
1609
70.5k
                    sina = sincos_alphas_B_fine[30 - (no_iid_steps + ps->iid_index[env][bk])][ps->icc_index[env][bk]];
1610
70.5k
                    cosg = cos_gammas_fine[abs_iid][ps->icc_index[env][bk]];
1611
70.5k
                    sing = sin_gammas_fine[abs_iid][ps->icc_index[env][bk]];
1612
84.4k
                } else {
1613
84.4k
                    cosa = sincos_alphas_B_normal[no_iid_steps + ps->iid_index[env][bk]][ps->icc_index[env][bk]];
1614
84.4k
                    sina = sincos_alphas_B_normal[14 - (no_iid_steps + ps->iid_index[env][bk])][ps->icc_index[env][bk]];
1615
84.4k
                    cosg = cos_gammas_normal[abs_iid][ps->icc_index[env][bk]];
1616
84.4k
                    sing = sin_gammas_normal[abs_iid][ps->icc_index[env][bk]];
1617
84.4k
                }
1618
1619
155k
                RE(h11) = MUL_C(COEF_SQRT2, MUL_C(cosa, cosg));
1620
155k
                RE(h12) = MUL_C(COEF_SQRT2, MUL_C(sina, cosg));
1621
155k
                RE(h21) = MUL_C(COEF_SQRT2, MUL_C(-cosa, sing));
1622
155k
                RE(h22) = MUL_C(COEF_SQRT2, MUL_C(sina, sing));
1623
155k
            }
1624
399k
            IM(h11) = IM(h12) = IM(h21) = IM(h22) = 0;
1625
1626
            /* calculate phase rotation parameters H_xy */
1627
            /* note that the imaginary part of these parameters are only calculated when
1628
               IPD and OPD are enabled
1629
             */
1630
399k
            if ((ps->enable_ipdopd) && (bk < nr_ipdopd_par))
1631
52.6k
            {
1632
52.6k
                int8_t i;
1633
52.6k
                real_t xy, pq, xypq;  // FRAC
1634
1635
                /* ringbuffer index */
1636
52.6k
                i = ps->phase_hist;
1637
1638
                /* previous value */
1639
#ifdef FIXED_POINT
1640
                /* divide by 4*2, shift right 3 bits;
1641
                   extra halving to avoid overflows; it is ok, because result is normalized */
1642
                RE(tempLeft)  = RE(ps->ipd_prev[bk][i]) >> 3;
1643
                IM(tempLeft)  = IM(ps->ipd_prev[bk][i]) >> 3;
1644
                RE(tempRight) = RE(ps->opd_prev[bk][i]) >> 3;
1645
                IM(tempRight) = IM(ps->opd_prev[bk][i]) >> 3;
1646
#else
1647
52.6k
                RE(tempLeft)  = MUL_F(RE(ps->ipd_prev[bk][i]), FRAC_CONST(0.25));
1648
52.6k
                IM(tempLeft)  = MUL_F(IM(ps->ipd_prev[bk][i]), FRAC_CONST(0.25));
1649
52.6k
                RE(tempRight) = MUL_F(RE(ps->opd_prev[bk][i]), FRAC_CONST(0.25));
1650
52.6k
                IM(tempRight) = MUL_F(IM(ps->opd_prev[bk][i]), FRAC_CONST(0.25));
1651
52.6k
#endif
1652
1653
                /* save current value */
1654
52.6k
                RE(ps->ipd_prev[bk][i]) = ipdopd_cos_tab[abs(ps->ipd_index[env][bk])];
1655
52.6k
                IM(ps->ipd_prev[bk][i]) = ipdopd_sin_tab[abs(ps->ipd_index[env][bk])];
1656
52.6k
                RE(ps->opd_prev[bk][i]) = ipdopd_cos_tab[abs(ps->opd_index[env][bk])];
1657
52.6k
                IM(ps->opd_prev[bk][i]) = ipdopd_sin_tab[abs(ps->opd_index[env][bk])];
1658
1659
                /* add current value */
1660
#ifdef FIXED_POINT
1661
                /* extra halving to avoid overflows */
1662
                RE(tempLeft)  += RE(ps->ipd_prev[bk][i]) >> 1;
1663
                IM(tempLeft)  += IM(ps->ipd_prev[bk][i]) >> 1;
1664
                RE(tempRight) += RE(ps->opd_prev[bk][i]) >> 1;
1665
                IM(tempRight) += IM(ps->opd_prev[bk][i]) >> 1;
1666
#else
1667
52.6k
                RE(tempLeft)  += RE(ps->ipd_prev[bk][i]);
1668
52.6k
                IM(tempLeft)  += IM(ps->ipd_prev[bk][i]);
1669
52.6k
                RE(tempRight) += RE(ps->opd_prev[bk][i]);
1670
52.6k
                IM(tempRight) += IM(ps->opd_prev[bk][i]);
1671
52.6k
#endif
1672
1673
                /* ringbuffer index */
1674
52.6k
                if (i == 0)
1675
26.6k
                {
1676
26.6k
                    i = 2;
1677
26.6k
                }
1678
52.6k
                i--;
1679
1680
                /* get value before previous */
1681
#ifdef FIXED_POINT
1682
                /* dividing by 2*2, shift right 2 bits; extra halving to avoid overflows */
1683
                RE(tempLeft)  += (RE(ps->ipd_prev[bk][i]) >> 2);
1684
                IM(tempLeft)  += (IM(ps->ipd_prev[bk][i]) >> 2);
1685
                RE(tempRight) += (RE(ps->opd_prev[bk][i]) >> 2);
1686
                IM(tempRight) += (IM(ps->opd_prev[bk][i]) >> 2);
1687
#else
1688
52.6k
                RE(tempLeft)  += MUL_F(RE(ps->ipd_prev[bk][i]), FRAC_CONST(0.5));
1689
52.6k
                IM(tempLeft)  += MUL_F(IM(ps->ipd_prev[bk][i]), FRAC_CONST(0.5));
1690
52.6k
                RE(tempRight) += MUL_F(RE(ps->opd_prev[bk][i]), FRAC_CONST(0.5));
1691
52.6k
                IM(tempRight) += MUL_F(IM(ps->opd_prev[bk][i]), FRAC_CONST(0.5));
1692
52.6k
#endif
1693
1694
#if 0 /* original code */
1695
                ipd = (float)atan2(IM(tempLeft), RE(tempLeft));
1696
                opd = (float)atan2(IM(tempRight), RE(tempRight));
1697
1698
                /* phase rotation */
1699
                RE(phaseLeft) = (float)cos(opd);
1700
                IM(phaseLeft) = (float)sin(opd);
1701
                opd -= ipd;
1702
                RE(phaseRight) = (float)cos(opd);
1703
                IM(phaseRight) = (float)sin(opd);
1704
#else
1705
1706
                // x = IM(tempLeft)
1707
                // y = RE(tempLeft)
1708
                // p = IM(tempRight)
1709
                // q = RE(tempRight)
1710
                // cos(atan2(x,y)) = y/sqrt((x*x) + (y*y))
1711
                // sin(atan2(x,y)) = x/sqrt((x*x) + (y*y))
1712
                // cos(atan2(x,y)-atan2(p,q)) = (y*q + x*p) / ( sqrt((x*x) + (y*y)) * sqrt((p*p) + (q*q)) );
1713
                // sin(atan2(x,y)-atan2(p,q)) = (x*q - y*p) / ( sqrt((x*x) + (y*y)) * sqrt((p*p) + (q*q)) );
1714
1715
52.6k
                xy = magnitude_c(tempRight);
1716
52.6k
                pq = magnitude_c(tempLeft);
1717
1718
52.6k
                if (xy != 0)
1719
52.6k
                {
1720
52.6k
                    RE(phaseLeft) = DIV_F(RE(tempRight), xy);
1721
52.6k
                    IM(phaseLeft) = DIV_F(IM(tempRight), xy);
1722
52.6k
                } else {
1723
0
                    RE(phaseLeft) = 0;
1724
0
                    IM(phaseLeft) = 0;
1725
0
                }
1726
1727
52.6k
                xypq = MUL_F(xy, pq);
1728
1729
52.6k
                if (xypq != 0)
1730
52.6k
                {
1731
52.6k
                    real_t tmp1 = MUL_F(RE(tempRight), RE(tempLeft)) + MUL_F(IM(tempRight), IM(tempLeft));
1732
52.6k
                    real_t tmp2 = MUL_F(IM(tempRight), RE(tempLeft)) - MUL_F(RE(tempRight), IM(tempLeft));
1733
1734
52.6k
                    RE(phaseRight) = DIV_F(tmp1, xypq);
1735
52.6k
                    IM(phaseRight) = DIV_F(tmp2, xypq);
1736
52.6k
                } else {
1737
0
                    RE(phaseRight) = 0;
1738
0
                    IM(phaseRight) = 0;
1739
0
                }
1740
1741
52.6k
#endif
1742
1743
                /* MUL_F(COEF, REAL) = COEF */
1744
52.6k
                IM(h11) = MUL_F(RE(h11), IM(phaseLeft));
1745
52.6k
                IM(h12) = MUL_F(RE(h12), IM(phaseRight));
1746
52.6k
                IM(h21) = MUL_F(RE(h21), IM(phaseLeft));
1747
52.6k
                IM(h22) = MUL_F(RE(h22), IM(phaseRight));
1748
1749
52.6k
                RE(h11) = MUL_F(RE(h11), RE(phaseLeft));
1750
52.6k
                RE(h12) = MUL_F(RE(h12), RE(phaseRight));
1751
52.6k
                RE(h21) = MUL_F(RE(h21), RE(phaseLeft));
1752
52.6k
                RE(h22) = MUL_F(RE(h22), RE(phaseRight));
1753
52.6k
            }
1754
1755
            /* length of the envelope n_e+1 - n_e (in time samples) */
1756
            /* 0 < L <= 32: integer */
1757
399k
            L = (real_t)(ps->border_position[env + 1] - ps->border_position[env]);
1758
1759
            /* obtain final H_xy by means of linear interpolation */
1760
399k
            RE(deltaH11) = (RE(h11) - RE(ps->h11_prev[gr])) / L;
1761
399k
            RE(deltaH12) = (RE(h12) - RE(ps->h12_prev[gr])) / L;
1762
399k
            RE(deltaH21) = (RE(h21) - RE(ps->h21_prev[gr])) / L;
1763
399k
            RE(deltaH22) = (RE(h22) - RE(ps->h22_prev[gr])) / L;
1764
1765
399k
            RE(H11) = RE(ps->h11_prev[gr]);
1766
399k
            RE(H12) = RE(ps->h12_prev[gr]);
1767
399k
            RE(H21) = RE(ps->h21_prev[gr]);
1768
399k
            RE(H22) = RE(ps->h22_prev[gr]);
1769
399k
            IM(H11) = IM(H12) = IM(H21) = IM(H22) = 0;
1770
1771
399k
            RE(ps->h11_prev[gr]) = RE(h11);
1772
399k
            RE(ps->h12_prev[gr]) = RE(h12);
1773
399k
            RE(ps->h21_prev[gr]) = RE(h21);
1774
399k
            RE(ps->h22_prev[gr]) = RE(h22);
1775
1776
            /* only calculate imaginary part when needed */
1777
399k
            if ((ps->enable_ipdopd) && (bk < nr_ipdopd_par))
1778
52.6k
            {
1779
                /* obtain final H_xy by means of linear interpolation */
1780
52.6k
                IM(deltaH11) = (IM(h11) - IM(ps->h11_prev[gr])) / L;
1781
52.6k
                IM(deltaH12) = (IM(h12) - IM(ps->h12_prev[gr])) / L;
1782
52.6k
                IM(deltaH21) = (IM(h21) - IM(ps->h21_prev[gr])) / L;
1783
52.6k
                IM(deltaH22) = (IM(h22) - IM(ps->h22_prev[gr])) / L;
1784
1785
52.6k
                IM(H11) = IM(ps->h11_prev[gr]);
1786
52.6k
                IM(H12) = IM(ps->h12_prev[gr]);
1787
52.6k
                IM(H21) = IM(ps->h21_prev[gr]);
1788
52.6k
                IM(H22) = IM(ps->h22_prev[gr]);
1789
1790
52.6k
                if ((NEGATE_IPD_MASK & ps->map_group2bk[gr]) != 0)
1791
7.28k
                {
1792
7.28k
                    IM(deltaH11) = -IM(deltaH11);
1793
7.28k
                    IM(deltaH12) = -IM(deltaH12);
1794
7.28k
                    IM(deltaH21) = -IM(deltaH21);
1795
7.28k
                    IM(deltaH22) = -IM(deltaH22);
1796
1797
7.28k
                    IM(H11) = -IM(H11);
1798
7.28k
                    IM(H12) = -IM(H12);
1799
7.28k
                    IM(H21) = -IM(H21);
1800
7.28k
                    IM(H22) = -IM(H22);
1801
7.28k
                }
1802
1803
52.6k
                IM(ps->h11_prev[gr]) = IM(h11);
1804
52.6k
                IM(ps->h12_prev[gr]) = IM(h12);
1805
52.6k
                IM(ps->h21_prev[gr]) = IM(h21);
1806
52.6k
                IM(ps->h22_prev[gr]) = IM(h22);
1807
52.6k
            }
1808
1809
            /* apply H_xy to the current envelope band of the decorrelated subband */
1810
6.00M
            for (n = ps->border_position[env]; n < ps->border_position[env + 1]; n++)
1811
5.60M
            {
1812
                /* addition finalises the interpolation over every n */
1813
5.60M
                RE(H11) += RE(deltaH11);
1814
5.60M
                RE(H12) += RE(deltaH12);
1815
5.60M
                RE(H21) += RE(deltaH21);
1816
5.60M
                RE(H22) += RE(deltaH22);
1817
5.60M
                if ((ps->enable_ipdopd) && (bk < nr_ipdopd_par))
1818
639k
                {
1819
639k
                    IM(H11) += IM(deltaH11);
1820
639k
                    IM(H12) += IM(deltaH12);
1821
639k
                    IM(H21) += IM(deltaH21);
1822
639k
                    IM(H22) += IM(deltaH22);
1823
639k
                }
1824
1825
                /* channel is an alias to the subband */
1826
19.3M
                for (sb = ps->group_border[gr]; sb < maxsb; sb++)
1827
13.7M
                {
1828
13.7M
                    complex_t inLeft, inRight;  // precision_of in(Left|Right) == precision_of X_(left|right)
1829
1830
                    /* load decorrelated samples */
1831
13.7M
                    if (gr < ps->num_hybrid_groups)
1832
3.11M
                    {
1833
3.11M
                        RE(inLeft) =  RE(X_hybrid_left[n][sb]);
1834
3.11M
                        IM(inLeft) =  IM(X_hybrid_left[n][sb]);
1835
3.11M
                        RE(inRight) = RE(X_hybrid_right[n][sb]);
1836
3.11M
                        IM(inRight) = IM(X_hybrid_right[n][sb]);
1837
10.6M
                    } else {
1838
10.6M
                        RE(inLeft) =  RE(X_left[n][sb]);
1839
10.6M
                        IM(inLeft) =  IM(X_left[n][sb]);
1840
10.6M
                        RE(inRight) = RE(X_right[n][sb]);
1841
10.6M
                        IM(inRight) = IM(X_right[n][sb]);
1842
10.6M
                    }
1843
1844
                    /* precision_of temp(Left|Right) == precision_of X_(left|right) */
1845
1846
                    /* apply mixing */
1847
13.7M
                    RE(tempLeft) =  MUL_C(RE(H11), RE(inLeft)) + MUL_C(RE(H21), RE(inRight));
1848
13.7M
                    IM(tempLeft) =  MUL_C(RE(H11), IM(inLeft)) + MUL_C(RE(H21), IM(inRight));
1849
13.7M
                    RE(tempRight) = MUL_C(RE(H12), RE(inLeft)) + MUL_C(RE(H22), RE(inRight));
1850
13.7M
                    IM(tempRight) = MUL_C(RE(H12), IM(inLeft)) + MUL_C(RE(H22), IM(inRight));
1851
1852
                    /* only perform imaginary operations when needed */
1853
13.7M
                    if ((ps->enable_ipdopd) && (bk < nr_ipdopd_par))
1854
641k
                    {
1855
                        /* apply rotation */
1856
641k
                        RE(tempLeft)  -= MUL_C(IM(H11), IM(inLeft)) + MUL_C(IM(H21), IM(inRight));
1857
641k
                        IM(tempLeft)  += MUL_C(IM(H11), RE(inLeft)) + MUL_C(IM(H21), RE(inRight));
1858
641k
                        RE(tempRight) -= MUL_C(IM(H12), IM(inLeft)) + MUL_C(IM(H22), IM(inRight));
1859
641k
                        IM(tempRight) += MUL_C(IM(H12), RE(inLeft)) + MUL_C(IM(H22), RE(inRight));
1860
641k
                    }
1861
1862
                    /* store final samples */
1863
13.7M
                    if (gr < ps->num_hybrid_groups)
1864
3.11M
                    {
1865
3.11M
                        RE(X_hybrid_left[n][sb])  = RE(tempLeft);
1866
3.11M
                        IM(X_hybrid_left[n][sb])  = IM(tempLeft);
1867
3.11M
                        RE(X_hybrid_right[n][sb]) = RE(tempRight);
1868
3.11M
                        IM(X_hybrid_right[n][sb]) = IM(tempRight);
1869
10.6M
                    } else {
1870
10.6M
                        RE(X_left[n][sb])  = RE(tempLeft);
1871
10.6M
                        IM(X_left[n][sb])  = IM(tempLeft);
1872
10.6M
                        RE(X_right[n][sb]) = RE(tempRight);
1873
10.6M
                        IM(X_right[n][sb]) = IM(tempRight);
1874
10.6M
                    }
1875
13.7M
                }
1876
5.60M
            }
1877
1878
            /* shift phase smoother's circular buffer index */
1879
399k
            ps->phase_hist++;
1880
399k
            if (ps->phase_hist == 2)
1881
199k
            {
1882
199k
                ps->phase_hist = 0;
1883
199k
            }
1884
399k
        }
1885
180k
    }
1886
5.65k
}
1887
1888
void ps_free(ps_info *ps)
1889
8.89k
{
1890
    /* free hybrid filterbank structures */
1891
8.89k
    hybrid_free(ps->hyb);
1892
1893
8.89k
    faad_free(ps);
1894
8.89k
}
1895
1896
ps_info *ps_init(uint8_t sr_index, uint8_t numTimeSlotsRate)
1897
8.89k
{
1898
8.89k
    uint8_t i;
1899
8.89k
    uint8_t short_delay_band;
1900
1901
8.89k
    ps_info *ps = (ps_info*)faad_malloc(sizeof(ps_info));
1902
8.89k
    memset(ps, 0, sizeof(ps_info));
1903
1904
8.89k
    ps->hyb = hybrid_init(numTimeSlotsRate);
1905
8.89k
    ps->numTimeSlotsRate = numTimeSlotsRate;
1906
1907
8.89k
    ps->ps_data_available = 0;
1908
1909
    /* delay stuff*/
1910
8.89k
    ps->saved_delay = 0;
1911
1912
578k
    for (i = 0; i < 64; i++)
1913
569k
    {
1914
569k
        ps->delay_buf_index_delay[i] = 0;
1915
569k
    }
1916
1917
35.5k
    for (i = 0; i < NO_ALLPASS_LINKS; i++)
1918
26.6k
    {
1919
26.6k
        ps->delay_buf_index_ser[i] = 0;
1920
#ifdef PARAM_32KHZ
1921
        if (sr_index <= 5) /* >= 32 kHz*/
1922
        {
1923
            ps->num_sample_delay_ser[i] = delay_length_d[1][i];
1924
        } else {
1925
            ps->num_sample_delay_ser[i] = delay_length_d[0][i];
1926
        }
1927
#else
1928
26.6k
        (void)sr_index;
1929
        /* THESE ARE CONSTANTS NOW */
1930
26.6k
        ps->num_sample_delay_ser[i] = delay_length_d[i];
1931
26.6k
#endif
1932
26.6k
    }
1933
1934
#ifdef PARAM_32KHZ
1935
    if (sr_index <= 5) /* >= 32 kHz*/
1936
    {
1937
        short_delay_band = 35;
1938
        ps->nr_allpass_bands = 22;
1939
        ps->alpha_decay = FRAC_CONST(0.76592833836465);
1940
        ps->alpha_smooth = FRAC_CONST(0.25);
1941
    } else {
1942
        short_delay_band = 64;
1943
        ps->nr_allpass_bands = 45;
1944
        ps->alpha_decay = FRAC_CONST(0.58664621951003);
1945
        ps->alpha_smooth = FRAC_CONST(0.6);
1946
    }
1947
#else
1948
    /* THESE ARE CONSTANTS NOW */
1949
8.89k
    short_delay_band = 35;
1950
8.89k
    ps->nr_allpass_bands = 22;
1951
8.89k
    ps->alpha_decay = FRAC_CONST(0.76592833836465);
1952
8.89k
    ps->alpha_smooth = FRAC_CONST(0.25);
1953
8.89k
#endif
1954
1955
    /* THESE ARE CONSTANT NOW IF PS IS INDEPENDANT OF SAMPLERATE */
1956
320k
    for (i = 0; i < short_delay_band; i++)
1957
311k
    {
1958
311k
        ps->delay_D[i] = 14;
1959
311k
    }
1960
266k
    for (i = short_delay_band; i < 64; i++)
1961
257k
    {
1962
257k
        ps->delay_D[i] = 1;
1963
257k
    }
1964
1965
    /* mixing and phase */
1966
453k
    for (i = 0; i < 50; i++)
1967
444k
    {
1968
444k
        RE(ps->h11_prev[i]) = 1;
1969
444k
        IM(ps->h11_prev[i]) = 1;
1970
444k
        RE(ps->h12_prev[i]) = 1;
1971
444k
        IM(ps->h12_prev[i]) = 1;
1972
444k
    }
1973
1974
8.89k
    ps->phase_hist = 0;
1975
1976
186k
    for (i = 0; i < 20; i++)
1977
177k
    {
1978
177k
        RE(ps->ipd_prev[i][0]) = 0;
1979
177k
        IM(ps->ipd_prev[i][0]) = 0;
1980
177k
        RE(ps->ipd_prev[i][1]) = 0;
1981
177k
        IM(ps->ipd_prev[i][1]) = 0;
1982
177k
        RE(ps->opd_prev[i][0]) = 0;
1983
177k
        IM(ps->opd_prev[i][0]) = 0;
1984
177k
        RE(ps->opd_prev[i][1]) = 0;
1985
177k
        IM(ps->opd_prev[i][1]) = 0;
1986
177k
    }
1987
1988
8.89k
    return ps;
1989
8.89k
}
1990
1991
/* main Parametric Stereo decoding function */
1992
uint8_t ps_decode(ps_info *ps, qmf_t X_left[38][64], qmf_t X_right[38][64])
1993
5.65k
{
1994
5.65k
    qmf_t X_hybrid_left[32][32] = {{{0}}};
1995
5.65k
    qmf_t X_hybrid_right[32][32] = {{{0}}};
1996
1997
    /* delta decoding of the bitstream data */
1998
5.65k
    ps_data_decode(ps);
1999
2000
    /* set up some parameters depending on filterbank type */
2001
5.65k
    if (ps->use34hybrid_bands)
2002
2.00k
    {
2003
2.00k
        ps->group_border = (uint8_t*)group_border34;
2004
2.00k
        ps->map_group2bk = (uint16_t*)map_group2bk34;
2005
2.00k
        ps->num_groups = 32+18;
2006
2.00k
        ps->num_hybrid_groups = 32;
2007
2.00k
        ps->nr_par_bands = 34;
2008
2.00k
        ps->decay_cutoff = 5;
2009
3.64k
    } else {
2010
3.64k
        ps->group_border = (uint8_t*)group_border20;
2011
3.64k
        ps->map_group2bk = (uint16_t*)map_group2bk20;
2012
3.64k
        ps->num_groups = 10+12;
2013
3.64k
        ps->num_hybrid_groups = 10;
2014
3.64k
        ps->nr_par_bands = 20;
2015
3.64k
        ps->decay_cutoff = 3;
2016
3.64k
    }
2017
2018
    /* Perform further analysis on the lowest subbands to get a higher
2019
     * frequency resolution
2020
     */
2021
5.65k
    hybrid_analysis((hyb_info*)ps->hyb, X_left, X_hybrid_left,
2022
5.65k
        ps->use34hybrid_bands, ps->numTimeSlotsRate);
2023
2024
    /* decorrelate mono signal */
2025
5.65k
    ps_decorrelate(ps, X_left, X_right, X_hybrid_left, X_hybrid_right);
2026
2027
    /* apply mixing and phase parameters */
2028
5.65k
    ps_mix_phase(ps, X_left, X_right, X_hybrid_left, X_hybrid_right);
2029
2030
    /* hybrid synthesis, to rebuild the SBR QMF matrices */
2031
5.65k
    hybrid_synthesis((hyb_info*)ps->hyb, X_left, X_hybrid_left,
2032
5.65k
        ps->use34hybrid_bands, ps->numTimeSlotsRate);
2033
2034
5.65k
    hybrid_synthesis((hyb_info*)ps->hyb, X_right, X_hybrid_right,
2035
5.65k
        ps->use34hybrid_bands, ps->numTimeSlotsRate);
2036
2037
5.65k
    return 0;
2038
5.65k
}
2039
2040
#endif