GCC Code Coverage Report
Directory: ../../../ffmpeg/ Exec Total Coverage
File: src/libavcodec/aacsbr_fixed.c Lines: 272 292 93.2 %
Date: 2019-11-20 04:07:19 Branches: 163 174 93.7 %

Line Branch Exec Source
1
/*
2
 * Copyright (c) 2013
3
 *      MIPS Technologies, Inc., California.
4
 *
5
 * Redistribution and use in source and binary forms, with or without
6
 * modification, are permitted provided that the following conditions
7
 * are met:
8
 * 1. Redistributions of source code must retain the above copyright
9
 *    notice, this list of conditions and the following disclaimer.
10
 * 2. Redistributions in binary form must reproduce the above copyright
11
 *    notice, this list of conditions and the following disclaimer in the
12
 *    documentation and/or other materials provided with the distribution.
13
 * 3. Neither the name of the MIPS Technologies, Inc., nor the names of its
14
 *    contributors may be used to endorse or promote products derived from
15
 *    this software without specific prior written permission.
16
 *
17
 * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND
18
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20
 * ARE DISCLAIMED.  IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE
21
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27
 * SUCH DAMAGE.
28
 *
29
 * AAC Spectral Band Replication decoding functions (fixed-point)
30
 * Copyright (c) 2008-2009 Robert Swain ( rob opendot cl )
31
 * Copyright (c) 2009-2010 Alex Converse <alex.converse@gmail.com>
32
 *
33
 * This file is part of FFmpeg.
34
 *
35
 * FFmpeg is free software; you can redistribute it and/or
36
 * modify it under the terms of the GNU Lesser General Public
37
 * License as published by the Free Software Foundation; either
38
 * version 2.1 of the License, or (at your option) any later version.
39
 *
40
 * FFmpeg is distributed in the hope that it will be useful,
41
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
42
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
43
 * Lesser General Public License for more details.
44
 *
45
 * You should have received a copy of the GNU Lesser General Public
46
 * License along with FFmpeg; if not, write to the Free Software
47
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
48
 */
49
50
/**
51
 * @file
52
 * AAC Spectral Band Replication decoding functions (fixed-point)
53
 * Note: Rounding-to-nearest used unless otherwise stated
54
 * @author Robert Swain ( rob opendot cl )
55
 * @author Stanislav Ocovaj ( stanislav.ocovaj imgtec com )
56
 */
57
#define USE_FIXED 1
58
59
#include "aac.h"
60
#include "sbr.h"
61
#include "aacsbr.h"
62
#include "aacsbrdata.h"
63
#include "aacsbr_fixed_tablegen.h"
64
#include "fft.h"
65
#include "aacps.h"
66
#include "sbrdsp.h"
67
#include "libavutil/internal.h"
68
#include "libavutil/libm.h"
69
#include "libavutil/avassert.h"
70
71
#include <stdint.h>
72
#include <float.h>
73
#include <math.h>
74
75
static VLC vlc_sbr[10];
76
static void aacsbr_func_ptr_init(AACSBRContext *c);
77
static const int CONST_LN2       = Q31(0.6931471806/256);  // ln(2)/256
78
static const int CONST_RECIP_LN2 = Q31(0.7213475204);      // 0.5/ln(2)
79
static const int CONST_076923    = Q31(0.76923076923076923077f);
80
81
static const int fixed_log_table[10] =
82
{
83
    Q31(1.0/2), Q31(1.0/3), Q31(1.0/4), Q31(1.0/5), Q31(1.0/6),
84
    Q31(1.0/7), Q31(1.0/8), Q31(1.0/9), Q31(1.0/10), Q31(1.0/11)
85
};
86
87
24
static int fixed_log(int x)
88
{
89
    int i, ret, xpow, tmp;
90
91
24
    ret = x;
92
24
    xpow = x;
93
144
    for (i=0; i<10; i+=2){
94
120
        xpow = (int)(((int64_t)xpow * x + 0x40000000) >> 31);
95
120
        tmp = (int)(((int64_t)xpow * fixed_log_table[i] + 0x40000000) >> 31);
96
120
        ret -= tmp;
97
98
120
        xpow = (int)(((int64_t)xpow * x + 0x40000000) >> 31);
99
120
        tmp = (int)(((int64_t)xpow * fixed_log_table[i+1] + 0x40000000) >> 31);
100
120
        ret += tmp;
101
    }
102
103
24
    return ret;
104
}
105
106
static const int fixed_exp_table[7] =
107
{
108
    Q31(1.0/2), Q31(1.0/6), Q31(1.0/24), Q31(1.0/120),
109
    Q31(1.0/720), Q31(1.0/5040), Q31(1.0/40320)
110
};
111
112
12
static int fixed_exp(int x)
113
{
114
    int i, ret, xpow, tmp;
115
116
12
    ret = 0x800000 + x;
117
12
    xpow = x;
118
96
    for (i=0; i<7; i++){
119
84
        xpow = (int)(((int64_t)xpow * x + 0x400000) >> 23);
120
84
        tmp = (int)(((int64_t)xpow * fixed_exp_table[i] + 0x40000000) >> 31);
121
84
        ret += tmp;
122
    }
123
124
12
    return ret;
125
}
126
127
12
static void make_bands(int16_t* bands, int start, int stop, int num_bands)
128
{
129
    int k, previous, present;
130
12
    int base, prod, nz = 0;
131
132
12
    base = (stop << 23) / start;
133
85
    while (base < 0x40000000){
134
73
        base <<= 1;
135
73
        nz++;
136
    }
137
12
    base = fixed_log(base - 0x80000000);
138
12
    base = (((base + 0x80) >> 8) + (8-nz)*CONST_LN2) / num_bands;
139
12
    base = fixed_exp(base);
140
141
12
    previous = start;
142
12
    prod = start << 23;
143
144
123
    for (k = 0; k < num_bands-1; k++) {
145
111
        prod = (int)(((int64_t)prod * base + 0x400000) >> 23);
146
111
        present = (prod + 0x400000) >> 23;
147
111
        bands[k] = present - previous;
148
111
        previous = present;
149
    }
150
12
    bands[num_bands-1] = stop - previous;
151
12
}
152
153
/// Dequantization and stereo decoding (14496-3 sp04 p203)
154
2082
static void sbr_dequant(SpectralBandReplication *sbr, int id_aac)
155
{
156
    int k, e;
157
    int ch;
158
159

3362
    if (id_aac == TYPE_CPE && sbr->bs_coupling) {
160
1280
        int alpha      = sbr->data[0].bs_amp_res ?  2 :  1;
161
1280
        int pan_offset = sbr->data[0].bs_amp_res ? 12 : 24;
162
2922
        for (e = 1; e <= sbr->data[0].bs_num_env; e++) {
163
19065
            for (k = 0; k < sbr->n[sbr->data[0].bs_freq_res[e]]; k++) {
164
                SoftFloat temp1, temp2, fac;
165
166
17423
                temp1.exp = sbr->data[0].env_facs_q[e][k] * alpha + 14;
167
17423
                if (temp1.exp & 1)
168
4733
                  temp1.mant = 759250125;
169
                else
170
12690
                  temp1.mant = 0x20000000;
171
17423
                temp1.exp = (temp1.exp >> 1) + 1;
172
17423
                if (temp1.exp > 66) { // temp1 > 1E20
173
                    av_log(NULL, AV_LOG_ERROR, "envelope scalefactor overflow in dequant\n");
174
                    temp1 = FLOAT_1;
175
                }
176
177
17423
                temp2.exp = (pan_offset - sbr->data[1].env_facs_q[e][k]) * alpha;
178
17423
                if (temp2.exp & 1)
179
                  temp2.mant = 759250125;
180
                else
181
17423
                  temp2.mant = 0x20000000;
182
17423
                temp2.exp = (temp2.exp >> 1) + 1;
183
17423
                fac   = av_div_sf(temp1, av_add_sf(FLOAT_1, temp2));
184
17423
                sbr->data[0].env_facs[e][k] = fac;
185
17423
                sbr->data[1].env_facs[e][k] = av_mul_sf(fac, temp2);
186
            }
187
        }
188
2837
        for (e = 1; e <= sbr->data[0].bs_num_noise; e++) {
189
5699
            for (k = 0; k < sbr->n_q; k++) {
190
                SoftFloat temp1, temp2, fac;
191
192
4142
                temp1.exp = NOISE_FLOOR_OFFSET - \
193
4142
                    sbr->data[0].noise_facs_q[e][k] + 2;
194
4142
                temp1.mant = 0x20000000;
195
4142
                av_assert0(temp1.exp <= 66);
196
4142
                temp2.exp = 12 - sbr->data[1].noise_facs_q[e][k] + 1;
197
4142
                temp2.mant = 0x20000000;
198
4142
                fac   = av_div_sf(temp1, av_add_sf(FLOAT_1, temp2));
199
4142
                sbr->data[0].noise_facs[e][k] = fac;
200
4142
                sbr->data[1].noise_facs[e][k] = av_mul_sf(fac, temp2);
201
            }
202
        }
203
    } else { // SCE or one non-coupled CPE
204

2029
        for (ch = 0; ch < (id_aac == TYPE_CPE) + 1; ch++) {
205
1227
            int alpha = sbr->data[ch].bs_amp_res ? 2 : 1;
206
3531
            for (e = 1; e <= sbr->data[ch].bs_num_env; e++)
207
26435
                for (k = 0; k < sbr->n[sbr->data[ch].bs_freq_res[e]]; k++){
208
                    SoftFloat temp1;
209
210
24131
                    temp1.exp = alpha * sbr->data[ch].env_facs_q[e][k] + 12;
211
24131
                    if (temp1.exp & 1)
212
2958
                        temp1.mant = 759250125;
213
                    else
214
21173
                        temp1.mant = 0x20000000;
215
24131
                    temp1.exp = (temp1.exp >> 1) + 1;
216
24131
                    if (temp1.exp > 66) { // temp1 > 1E20
217
                        av_log(NULL, AV_LOG_ERROR, "envelope scalefactor overflow in dequant\n");
218
                        temp1 = FLOAT_1;
219
                    }
220
24131
                    sbr->data[ch].env_facs[e][k] = temp1;
221
                }
222
3181
            for (e = 1; e <= sbr->data[ch].bs_num_noise; e++)
223
7656
                for (k = 0; k < sbr->n_q; k++){
224
5702
                    sbr->data[ch].noise_facs[e][k].exp = NOISE_FLOOR_OFFSET - \
225
5702
                        sbr->data[ch].noise_facs_q[e][k] + 1;
226
5702
                    sbr->data[ch].noise_facs[e][k].mant = 0x20000000;
227
                }
228
        }
229
    }
230
2082
}
231
232
/** High Frequency Generation (14496-3 sp04 p214+) and Inverse Filtering
233
 * (14496-3 sp04 p214)
234
 * Warning: This routine does not seem numerically stable.
235
 */
236
3787
static void sbr_hf_inverse_filter(SBRDSPContext *dsp,
237
                                  int (*alpha0)[2], int (*alpha1)[2],
238
                                  const int X_low[32][40][2], int k0)
239
{
240
    int k;
241
    int shift, round;
242
243
71885
    for (k = 0; k < k0; k++) {
244
        SoftFloat phi[3][2][2];
245
        SoftFloat a00, a01, a10, a11;
246
        SoftFloat dk;
247
248
68098
        dsp->autocorrelate(X_low[k], phi);
249
250
68098
        dk = av_sub_sf(av_mul_sf(phi[2][1][0], phi[1][0][0]),
251
             av_mul_sf(av_add_sf(av_mul_sf(phi[1][1][0], phi[1][1][0]),
252
             av_mul_sf(phi[1][1][1], phi[1][1][1])), FLOAT_0999999));
253
254
68098
        if (!dk.mant) {
255
1708
            a10 = FLOAT_0;
256
1708
            a11 = FLOAT_0;
257
        } else {
258
            SoftFloat temp_real, temp_im;
259
66390
            temp_real = av_sub_sf(av_sub_sf(av_mul_sf(phi[0][0][0], phi[1][1][0]),
260
                                            av_mul_sf(phi[0][0][1], phi[1][1][1])),
261
                                  av_mul_sf(phi[0][1][0], phi[1][0][0]));
262
66390
            temp_im   = av_sub_sf(av_add_sf(av_mul_sf(phi[0][0][0], phi[1][1][1]),
263
                                            av_mul_sf(phi[0][0][1], phi[1][1][0])),
264
                                  av_mul_sf(phi[0][1][1], phi[1][0][0]));
265
266
66390
            a10 = av_div_sf(temp_real, dk);
267
66390
            a11 = av_div_sf(temp_im,   dk);
268
        }
269
270
68098
        if (!phi[1][0][0].mant) {
271
1708
            a00 = FLOAT_0;
272
1708
            a01 = FLOAT_0;
273
        } else {
274
            SoftFloat temp_real, temp_im;
275
66390
            temp_real = av_add_sf(phi[0][0][0],
276
                                  av_add_sf(av_mul_sf(a10, phi[1][1][0]),
277
                                            av_mul_sf(a11, phi[1][1][1])));
278
66390
            temp_im   = av_add_sf(phi[0][0][1],
279
                                  av_sub_sf(av_mul_sf(a11, phi[1][1][0]),
280
                                            av_mul_sf(a10, phi[1][1][1])));
281
282
66390
            temp_real.mant = -temp_real.mant;
283
66390
            temp_im.mant   = -temp_im.mant;
284
66390
            a00 = av_div_sf(temp_real, phi[1][0][0]);
285
66390
            a01 = av_div_sf(temp_im,   phi[1][0][0]);
286
        }
287
288
68098
        shift = a00.exp;
289
68098
        if (shift >= 3)
290
2
            alpha0[k][0] = 0x7fffffff;
291
68096
        else if (shift <= -30)
292
2123
            alpha0[k][0] = 0;
293
        else {
294
65973
            shift = 1-shift;
295
65973
            if (shift <= 0)
296
1838
                alpha0[k][0] = a00.mant * (1<<-shift);
297
            else {
298
64135
                round = 1 << (shift-1);
299
64135
                alpha0[k][0] = (a00.mant + round) >> shift;
300
            }
301
        }
302
303
68098
        shift = a01.exp;
304
68098
        if (shift >= 3)
305
3
            alpha0[k][1] = 0x7fffffff;
306
68095
        else if (shift <= -30)
307
2078
            alpha0[k][1] = 0;
308
        else {
309
66017
            shift = 1-shift;
310
66017
            if (shift <= 0)
311
14382
                alpha0[k][1] = a01.mant * (1<<-shift);
312
            else {
313
51635
                round = 1 << (shift-1);
314
51635
                alpha0[k][1] = (a01.mant + round) >> shift;
315
            }
316
        }
317
68098
        shift = a10.exp;
318
68098
        if (shift >= 3)
319
5
            alpha1[k][0] = 0x7fffffff;
320
68093
        else if (shift <= -30)
321
2024
            alpha1[k][0] = 0;
322
        else {
323
66069
            shift = 1-shift;
324
66069
            if (shift <= 0)
325
76
                alpha1[k][0] = a10.mant * (1<<-shift);
326
            else {
327
65993
                round = 1 << (shift-1);
328
65993
                alpha1[k][0] = (a10.mant + round) >> shift;
329
            }
330
        }
331
332
68098
        shift = a11.exp;
333
68098
        if (shift >= 3)
334
            alpha1[k][1] = 0x7fffffff;
335
68098
        else if (shift <= -30)
336
2114
            alpha1[k][1] = 0;
337
        else {
338
65984
            shift = 1-shift;
339
65984
            if (shift <= 0)
340
84
                alpha1[k][1] = a11.mant * (1<<-shift);
341
            else {
342
65900
                round = 1 << (shift-1);
343
65900
                alpha1[k][1] = (a11.mant + round) >> shift;
344
            }
345
        }
346
347
68098
        shift = (int)(((int64_t)(alpha1[k][0]>>1) * (alpha1[k][0]>>1) + \
348
68098
                       (int64_t)(alpha1[k][1]>>1) * (alpha1[k][1]>>1) + \
349
68098
                       0x40000000) >> 31);
350
68098
        if (shift >= 0x20000000){
351
            alpha1[k][0] = 0;
352
            alpha1[k][1] = 0;
353
            alpha0[k][0] = 0;
354
            alpha0[k][1] = 0;
355
        }
356
357
68098
        shift = (int)(((int64_t)(alpha0[k][0]>>1) * (alpha0[k][0]>>1) + \
358
68098
                       (int64_t)(alpha0[k][1]>>1) * (alpha0[k][1]>>1) + \
359
68098
                       0x40000000) >> 31);
360
68098
        if (shift >= 0x20000000){
361
            alpha1[k][0] = 0;
362
            alpha1[k][1] = 0;
363
            alpha0[k][0] = 0;
364
            alpha0[k][1] = 0;
365
        }
366
    }
367
3787
}
368
369
/// Chirp Factors (14496-3 sp04 p214)
370
3787
static void sbr_chirp(SpectralBandReplication *sbr, SBRData *ch_data)
371
{
372
    int i;
373
    int new_bw;
374
    static const int bw_tab[] = { 0, 1610612736, 1932735283, 2104533975 };
375
    int64_t accu;
376
377
14000
    for (i = 0; i < sbr->n_q; i++) {
378
10213
        if (ch_data->bs_invf_mode[0][i] + ch_data->bs_invf_mode[1][i] == 1)
379
799
            new_bw = 1288490189;
380
        else
381
9414
            new_bw = bw_tab[ch_data->bs_invf_mode[0][i]];
382
383
10213
        if (new_bw < ch_data->bw_array[i]){
384
2080
            accu  = (int64_t)new_bw * 1610612736;
385
2080
            accu += (int64_t)ch_data->bw_array[i] * 0x20000000;
386
2080
            new_bw = (int)((accu + 0x40000000) >> 31);
387
        } else {
388
8133
            accu  = (int64_t)new_bw * 1946157056;
389
8133
            accu += (int64_t)ch_data->bw_array[i] * 201326592;
390
8133
            new_bw = (int)((accu + 0x40000000) >> 31);
391
        }
392
10213
        ch_data->bw_array[i] = new_bw < 0x2000000 ? 0 : new_bw;
393
    }
394
3787
}
395
396
/**
397
 * Calculation of levels of additional HF signal components (14496-3 sp04 p219)
398
 * and Calculation of gain (14496-3 sp04 p219)
399
 */
400
3787
static void sbr_gain_calc(AACContext *ac, SpectralBandReplication *sbr,
401
                          SBRData *ch_data, const int e_a[2])
402
{
403
    int e, k, m;
404
    // max gain limits : -3dB, 0dB, 3dB, inf dB (limiter off)
405
    static const SoftFloat limgain[4] = { { 760155524,  0 }, { 0x20000000,  1 },
406
                                            { 758351638,  1 }, { 625000000, 34 } };
407
408
9375
    for (e = 0; e < ch_data->bs_num_env; e++) {
409

5588
        int delta = !((e == e_a[1]) || (e == e_a[0]));
410
23292
        for (k = 0; k < sbr->n_lim; k++) {
411
            SoftFloat gain_boost, gain_max;
412
            SoftFloat sum[2];
413
17704
            sum[0] = sum[1] = FLOAT_0;
414
145541
            for (m = sbr->f_tablelim[k] - sbr->kx[1]; m < sbr->f_tablelim[k + 1] - sbr->kx[1]; m++) {
415
127837
                const SoftFloat temp = av_div_sf(sbr->e_origmapped[e][m],
416
                                            av_add_sf(FLOAT_1, sbr->q_mapped[e][m]));
417
127837
                sbr->q_m[e][m] = av_sqrt_sf(av_mul_sf(temp, sbr->q_mapped[e][m]));
418
127837
                sbr->s_m[e][m] = av_sqrt_sf(av_mul_sf(temp, av_int2sf(ch_data->s_indexmapped[e + 1][m], 0)));
419
127837
                if (!sbr->s_mapped[e][m]) {
420
123205
                    if (delta) {
421
115357
                      sbr->gain[e][m] = av_sqrt_sf(av_div_sf(sbr->e_origmapped[e][m],
422
                                            av_mul_sf(av_add_sf(FLOAT_1, sbr->e_curr[e][m]),
423
                                            av_add_sf(FLOAT_1, sbr->q_mapped[e][m]))));
424
                    } else {
425
7848
                      sbr->gain[e][m] = av_sqrt_sf(av_div_sf(sbr->e_origmapped[e][m],
426
                                            av_add_sf(FLOAT_1, sbr->e_curr[e][m])));
427
                    }
428
                } else {
429
4632
                    sbr->gain[e][m] = av_sqrt_sf(
430
                                        av_div_sf(
431
                                            av_mul_sf(sbr->e_origmapped[e][m], sbr->q_mapped[e][m]),
432
                                            av_mul_sf(
433
                                                av_add_sf(FLOAT_1, sbr->e_curr[e][m]),
434
                                                av_add_sf(FLOAT_1, sbr->q_mapped[e][m]))));
435
                }
436
127837
                sbr->gain[e][m] = av_add_sf(sbr->gain[e][m], FLOAT_MIN);
437
            }
438
145541
            for (m = sbr->f_tablelim[k] - sbr->kx[1]; m < sbr->f_tablelim[k + 1] - sbr->kx[1]; m++) {
439
127837
                sum[0] = av_add_sf(sum[0], sbr->e_origmapped[e][m]);
440
127837
                sum[1] = av_add_sf(sum[1], sbr->e_curr[e][m]);
441
            }
442
17704
            gain_max = av_mul_sf(limgain[sbr->bs_limiter_gains],
443
                            av_sqrt_sf(
444
                                av_div_sf(
445
                                    av_add_sf(FLOAT_EPSILON, sum[0]),
446
                                    av_add_sf(FLOAT_EPSILON, sum[1]))));
447
17704
            if (av_gt_sf(gain_max, FLOAT_100000))
448
3741
              gain_max = FLOAT_100000;
449
145541
            for (m = sbr->f_tablelim[k] - sbr->kx[1]; m < sbr->f_tablelim[k + 1] - sbr->kx[1]; m++) {
450
127837
                SoftFloat q_m_max = av_div_sf(
451
                                        av_mul_sf(sbr->q_m[e][m], gain_max),
452
                                        sbr->gain[e][m]);
453
127837
                if (av_gt_sf(sbr->q_m[e][m], q_m_max))
454
33577
                  sbr->q_m[e][m] = q_m_max;
455
127837
                if (av_gt_sf(sbr->gain[e][m], gain_max))
456
33577
                  sbr->gain[e][m] = gain_max;
457
            }
458
17704
            sum[0] = sum[1] = FLOAT_0;
459
145541
            for (m = sbr->f_tablelim[k] - sbr->kx[1]; m < sbr->f_tablelim[k + 1] - sbr->kx[1]; m++) {
460
127837
                sum[0] = av_add_sf(sum[0], sbr->e_origmapped[e][m]);
461
127837
                sum[1] = av_add_sf(sum[1],
462
                            av_mul_sf(
463
                                av_mul_sf(sbr->e_curr[e][m],
464
                                          sbr->gain[e][m]),
465
                                sbr->gain[e][m]));
466
127837
                sum[1] = av_add_sf(sum[1],
467
                            av_mul_sf(sbr->s_m[e][m], sbr->s_m[e][m]));
468

127837
                if (delta && !sbr->s_m[e][m].mant)
469
117689
                  sum[1] = av_add_sf(sum[1],
470
                                av_mul_sf(sbr->q_m[e][m], sbr->q_m[e][m]));
471
            }
472
17704
            gain_boost = av_sqrt_sf(
473
                            av_div_sf(
474
                                av_add_sf(FLOAT_EPSILON, sum[0]),
475
                                av_add_sf(FLOAT_EPSILON, sum[1])));
476
17704
            if (av_gt_sf(gain_boost, FLOAT_1584893192))
477
1483
              gain_boost = FLOAT_1584893192;
478
479
145541
            for (m = sbr->f_tablelim[k] - sbr->kx[1]; m < sbr->f_tablelim[k + 1] - sbr->kx[1]; m++) {
480
127837
                sbr->gain[e][m] = av_mul_sf(sbr->gain[e][m], gain_boost);
481
127837
                sbr->q_m[e][m]  = av_mul_sf(sbr->q_m[e][m], gain_boost);
482
127837
                sbr->s_m[e][m]  = av_mul_sf(sbr->s_m[e][m], gain_boost);
483
            }
484
        }
485
    }
486
3787
}
487
488
/// Assembling HF Signals (14496-3 sp04 p220)
489
3787
static void sbr_hf_assemble(int Y1[38][64][2],
490
                            const int X_high[64][40][2],
491
                            SpectralBandReplication *sbr, SBRData *ch_data,
492
                            const int e_a[2])
493
{
494
    int e, i, j, m;
495
3787
    const int h_SL = 4 * !sbr->bs_smoothing_mode;
496
3787
    const int kx = sbr->kx[1];
497
3787
    const int m_max = sbr->m[1];
498
    static const SoftFloat h_smooth[5] = {
499
      { 715827883, -1 },
500
      { 647472402, -1 },
501
      { 937030863, -2 },
502
      { 989249804, -3 },
503
      { 546843842, -4 },
504
    };
505
3787
    SoftFloat (*g_temp)[48] = ch_data->g_temp, (*q_temp)[48] = ch_data->q_temp;
506
3787
    int indexnoise = ch_data->f_indexnoise;
507
3787
    int indexsine  = ch_data->f_indexsine;
508
509
3787
    if (sbr->reset) {
510
17
        for (i = 0; i < h_SL; i++) {
511
8
            memcpy(g_temp[i + 2*ch_data->t_env[0]], sbr->gain[0], m_max * sizeof(sbr->gain[0][0]));
512
8
            memcpy(q_temp[i + 2*ch_data->t_env[0]], sbr->q_m[0],  m_max * sizeof(sbr->q_m[0][0]));
513
        }
514
3778
    } else if (h_SL) {
515
5730
        for (i = 0; i < 4; i++) {
516
4584
            memcpy(g_temp[i + 2 * ch_data->t_env[0]],
517
4584
                   g_temp[i + 2 * ch_data->t_env_num_env_old],
518
                   sizeof(g_temp[0]));
519
4584
            memcpy(q_temp[i + 2 * ch_data->t_env[0]],
520
4584
                   q_temp[i + 2 * ch_data->t_env_num_env_old],
521
                   sizeof(q_temp[0]));
522
        }
523
    }
524
525
9375
    for (e = 0; e < ch_data->bs_num_env; e++) {
526
126772
        for (i = 2 * ch_data->t_env[e]; i < 2 * ch_data->t_env[e + 1]; i++) {
527
121184
            memcpy(g_temp[h_SL + i], sbr->gain[e], m_max * sizeof(sbr->gain[0][0]));
528
121184
            memcpy(q_temp[h_SL + i], sbr->q_m[e],  m_max * sizeof(sbr->q_m[0][0]));
529
        }
530
    }
531
532
9375
    for (e = 0; e < ch_data->bs_num_env; e++) {
533
126772
        for (i = 2 * ch_data->t_env[e]; i < 2 * ch_data->t_env[e + 1]; i++) {
534
            SoftFloat g_filt_tab[48];
535
            SoftFloat q_filt_tab[48];
536
            SoftFloat *g_filt, *q_filt;
537
538

121184
            if (h_SL && e != e_a[0] && e != e_a[1]) {
539
36634
                g_filt = g_filt_tab;
540
36634
                q_filt = q_filt_tab;
541
659412
                for (m = 0; m < m_max; m++) {
542
622778
                    const int idx1 = i + h_SL;
543
622778
                    g_filt[m].mant = g_filt[m].exp = 0;
544
622778
                    q_filt[m].mant = q_filt[m].exp = 0;
545
3736668
                    for (j = 0; j <= h_SL; j++) {
546
3113890
                        g_filt[m] = av_add_sf(g_filt[m],
547
3113890
                                        av_mul_sf(g_temp[idx1 - j][m],
548
                                            h_smooth[j]));
549
3113890
                        q_filt[m] = av_add_sf(q_filt[m],
550
3113890
                                        av_mul_sf(q_temp[idx1 - j][m],
551
                                            h_smooth[j]));
552
                    }
553
                }
554
            } else {
555
84550
                g_filt = g_temp[i + h_SL];
556
84550
                q_filt = q_temp[i];
557
            }
558
559
121184
            sbr->dsp.hf_g_filt(Y1[i] + kx, X_high + kx, g_filt, m_max,
560
121184
                               i + ENVELOPE_ADJUSTMENT_OFFSET);
561
562

121184
            if (e != e_a[0] && e != e_a[1]) {
563
119686
                sbr->dsp.hf_apply_noise[indexsine](Y1[i] + kx, sbr->s_m[e],
564
                                                   q_filt, indexnoise,
565
                                                   kx, m_max);
566
            } else {
567
1498
                int idx = indexsine&1;
568
1498
                int A = (1-((indexsine+(kx & 1))&2));
569
1498
                int B = (A^(-idx)) + idx;
570
1498
                unsigned *out = &Y1[i][kx][idx];
571
                int shift;
572
                unsigned round;
573
574
1498
                SoftFloat *in  = sbr->s_m[e];
575
18898
                for (m = 0; m+1 < m_max; m+=2) {
576
                    int shift2;
577
17400
                    shift = 22 - in[m  ].exp;
578
17400
                    shift2= 22 - in[m+1].exp;
579

17400
                    if (shift < 1 || shift2 < 1) {
580
                        av_log(NULL, AV_LOG_ERROR, "Overflow in sbr_hf_assemble, shift=%d,%d\n", shift, shift2);
581
                        return;
582
                    }
583
17400
                    if (shift < 32) {
584
600
                        round = 1 << (shift-1);
585
600
                        out[2*m  ] += (int)(in[m  ].mant * A + round) >> shift;
586
                    }
587
588
17400
                    if (shift2 < 32) {
589
152
                        round = 1 << (shift2-1);
590
152
                        out[2*m+2] += (int)(in[m+1].mant * B + round) >> shift2;
591
                    }
592
                }
593
1498
                if(m_max&1)
594
                {
595
1004
                    shift = 22 - in[m  ].exp;
596
1004
                    if (shift < 1) {
597
                        av_log(NULL, AV_LOG_ERROR, "Overflow in sbr_hf_assemble, shift=%d\n", shift);
598
                        return;
599
1004
                    } else if (shift < 32) {
600
                        round = 1 << (shift-1);
601
                        out[2*m  ] += (int)(in[m  ].mant * A + round) >> shift;
602
                    }
603
                }
604
            }
605
121184
            indexnoise = (indexnoise + m_max) & 0x1ff;
606
121184
            indexsine = (indexsine + 1) & 3;
607
        }
608
    }
609
3787
    ch_data->f_indexnoise = indexnoise;
610
3787
    ch_data->f_indexsine  = indexsine;
611
}
612
613
#include "aacsbr_template.c"