2
** FAAD2 - Freeware Advanced Audio (AAC) Decoder including SBR decoding
3
** Copyright (C) 2003-2004 M. Bakker, Ahead Software AG, http://www.nero.com
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.
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.
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.
19
** Any non-GPL usage of this software or parts of this software is strictly
22
** Commercial non-GPL licensing of this software is possible.
23
** For more info contact Ahead Software through Mpeg4AAClicense@nero.com.
25
** $Id: sbr_qmf.c,v 1.19 2004/01/05 14:05:12 menno Exp $
38
#include "sbr_qmf_c.h"
39
#include "sbr_syntax.h"
42
qmfa_info *qmfa_init(uint8_t channels)
44
qmfa_info *qmfa = (qmfa_info*)faad_malloc(sizeof(qmfa_info));
45
qmfa->x = (real_t*)faad_malloc(channels * 10 * sizeof(real_t));
46
memset(qmfa->x, 0, channels * 10 * sizeof(real_t));
48
qmfa->channels = channels;
53
void qmfa_end(qmfa_info *qmfa)
57
if (qmfa->x) faad_free(qmfa->x);
62
void sbr_qmf_analysis_32(sbr_info *sbr, qmfa_info *qmfa, const real_t *input,
63
qmf_t X[MAX_NTSRHFG][32], uint8_t offset, uint8_t kx)
67
ALIGN real_t x[64], y[64];
75
for (l = 0; l < sbr->numTimeSlotsRate; l++)
79
/* shift input buffer x */
80
memmove(qmfa->x + 32, qmfa->x, (320-32)*sizeof(real_t));
82
/* add new samples to input buffer x */
83
for (n = 32 - 1; n >= 0; n--)
86
qmfa->x[n] = (input[in++]) >> 5;
88
qmfa->x[n] = input[in++];
92
/* window and summation to create array u */
93
for (n = 0; n < 64; n++)
95
u[n] = MUL_F(qmfa->x[n], qmf_c[2*n]) +
96
MUL_F(qmfa->x[n + 64], qmf_c[2*(n + 64)]) +
97
MUL_F(qmfa->x[n + 128], qmf_c[2*(n + 128)]) +
98
MUL_F(qmfa->x[n + 192], qmf_c[2*(n + 192)]) +
99
MUL_F(qmfa->x[n + 256], qmf_c[2*(n + 256)]);
102
/* calculate 32 subband samples by introducing X */
105
for (n = 1; n < 16; n++)
106
y[n] = u[n+48] + u[48-n];
107
for (n = 16; n < 32; n++)
108
y[n] = -u[n-16] + u[48-n];
110
DCT3_32_unscaled(u, y);
112
for (n = 0; n < 32; n++)
117
QMF_RE(X[l + offset][n]) = u[n] << 1;
119
QMF_RE(X[l + offset][n]) = 2. * u[n];
122
QMF_RE(X[l + offset][n]) = 0;
127
for (n = 0; n < 31; n++)
129
x[2*n+1] = u[n+1] + u[63-n];
130
x[2*n+2] = u[n+1] - u[63-n];
134
DCT4_64_kernel(y, x);
136
for (n = 0; n < 32; n++)
141
QMF_RE(X[l + offset][n]) = y[n] << 1;
142
QMF_IM(X[l + offset][n]) = -y[63-n] << 1;
144
QMF_RE(X[l + offset][n]) = 2. * y[n];
145
QMF_IM(X[l + offset][n]) = -2. * y[63-n];
148
QMF_RE(X[l + offset][n]) = 0;
149
QMF_IM(X[l + offset][n]) = 0;
156
qmfs_info *qmfs_init(uint8_t channels)
158
qmfs_info *qmfs = (qmfs_info*)faad_malloc(sizeof(qmfs_info));
160
#ifndef SBR_LOW_POWER
161
qmfs->v[0] = (real_t*)faad_malloc(channels * 10 * sizeof(real_t));
162
memset(qmfs->v[0], 0, channels * 10 * sizeof(real_t));
163
qmfs->v[1] = (real_t*)faad_malloc(channels * 10 * sizeof(real_t));
164
memset(qmfs->v[1], 0, channels * 10 * sizeof(real_t));
166
qmfs->v[0] = (real_t*)faad_malloc(channels * 20 * sizeof(real_t));
167
memset(qmfs->v[0], 0, channels * 20 * sizeof(real_t));
173
qmfs->channels = channels;
178
qmfs->qmf_func = sbr_qmf_synthesis_64_sse;
180
qmfs->qmf_func = sbr_qmf_synthesis_64;
187
void qmfs_end(qmfs_info *qmfs)
191
if (qmfs->v[0]) faad_free(qmfs->v[0]);
192
#ifndef SBR_LOW_POWER
193
if (qmfs->v[1]) faad_free(qmfs->v[1]);
200
void sbr_qmf_synthesis_64(sbr_info *sbr, qmfs_info *qmfs, qmf_t X[MAX_NTSRHFG][64],
205
int16_t n, k, out = 0;
209
/* qmf subsample l */
210
for (l = 0; l < sbr->numTimeSlotsRate; l++)
215
//memmove(qmfs->v[0] + 64, qmfs->v[0], (640-64)*sizeof(real_t));
216
//memmove(qmfs->v[1] + 64, qmfs->v[1], (640-64)*sizeof(real_t));
217
memmove(qmfs->v[0] + 128, qmfs->v[0], (1280-128)*sizeof(real_t));
219
//v0 = qmfs->v[qmfs->v_index];
220
//v1 = qmfs->v[(qmfs->v_index + 1) & 0x1];
221
//qmfs->v_index = (qmfs->v_index + 1) & 0x1;
223
/* calculate 128 samples */
224
for (k = 0; k < 64; k++)
227
x[k] = QMF_RE(X[l][k]);
229
x[k] = QMF_RE(X[l][k]) / 32.;
233
for (n = 0; n < 32; n++)
239
DCT2_64_unscaled(x, x);
241
for (n = 0; n < 64; n++)
243
qmfs->v[0][n+32] = x[n];
245
for (n = 0; n < 32; n++)
247
qmfs->v[0][31 - n] = x[n + 1];
249
DST2_64_unscaled(x, y);
251
for (n = 1; n < 32; n++)
253
qmfs->v[0][n + 96] = x[n-1];
256
/* calculate 64 output samples and window */
257
for (k = 0; k < 64; k++)
260
output[out++] = MUL_F(qmfs->v[0][k], qmf_c[k]) +
261
MUL_F(qmfs->v[0][192 + k], qmf_c[64 + k]) +
262
MUL_F(qmfs->v[0][256 + k], qmf_c[128 + k]) +
263
MUL_F(qmfs->v[0][256 + 192 + k], qmf_c[128 + 64 + k]) +
264
MUL_F(qmfs->v[0][512 + k], qmf_c[256 + k]) +
265
MUL_F(qmfs->v[0][512 + 192 + k], qmf_c[256 + 64 + k]) +
266
MUL_F(qmfs->v[0][768 + k], qmf_c[384 + k]) +
267
MUL_F(qmfs->v[0][768 + 192 + k], qmf_c[384 + 64 + k]) +
268
MUL_F(qmfs->v[0][1024 + k], qmf_c[512 + k]) +
269
MUL_F(qmfs->v[0][1024 + 192 + k], qmf_c[512 + 64 + k]);
271
output[out++] = MUL_F(v0[k], qmf_c[k]) +
272
MUL_F(v0[64 + k], qmf_c[64 + k]) +
273
MUL_F(v0[128 + k], qmf_c[128 + k]) +
274
MUL_F(v0[192 + k], qmf_c[192 + k]) +
275
MUL_F(v0[256 + k], qmf_c[256 + k]) +
276
MUL_F(v0[320 + k], qmf_c[320 + k]) +
277
MUL_F(v0[384 + k], qmf_c[384 + k]) +
278
MUL_F(v0[448 + k], qmf_c[448 + k]) +
279
MUL_F(v0[512 + k], qmf_c[512 + k]) +
280
MUL_F(v0[576 + k], qmf_c[576 + k]);
286
void sbr_qmf_synthesis_64_sse(sbr_info *sbr, qmfs_info *qmfs, qmf_t X[MAX_NTSRHFG][64],
292
int16_t n, k, out = 0;
295
/* qmf subsample l */
296
for (l = 0; l < sbr->numTimeSlotsRate; l++)
301
//memmove(qmfs->v[0] + 64, qmfs->v[0], (640-64)*sizeof(real_t));
302
//memmove(qmfs->v[1] + 64, qmfs->v[1], (640-64)*sizeof(real_t));
303
memmove(qmfs->v[0] + 128, qmfs->v[0], (1280-128)*sizeof(real_t));
305
//v0 = qmfs->v[qmfs->v_index];
306
//v1 = qmfs->v[(qmfs->v_index + 1) & 0x1];
307
//qmfs->v_index = (qmfs->v_index + 1) & 0x1;
309
/* calculate 128 samples */
310
for (k = 0; k < 64; k++)
313
x[k] = QMF_RE(X[l][k]);
315
x[k] = QMF_RE(X[l][k]) / 32.;
319
for (n = 0; n < 32; n++)
325
DCT2_64_unscaled(x, x);
327
for (n = 0; n < 64; n++)
329
qmfs->v[0][n+32] = x[n];
331
for (n = 0; n < 32; n++)
333
qmfs->v[0][31 - n] = x[n + 1];
336
DST2_64_unscaled(x, y);
338
for (n = 1; n < 32; n++)
340
qmfs->v[0][n + 96] = x[n-1];
343
/* calculate 64 output samples and window */
344
for (k = 0; k < 64; k++)
347
output[out++] = MUL_F(qmfs->v[0][k], qmf_c[k]) +
348
MUL_F(qmfs->v[0][192 + k], qmf_c[64 + k]) +
349
MUL_F(qmfs->v[0][256 + k], qmf_c[128 + k]) +
350
MUL_F(qmfs->v[0][256 + 192 + k], qmf_c[128 + 64 + k]) +
351
MUL_F(qmfs->v[0][512 + k], qmf_c[256 + k]) +
352
MUL_F(qmfs->v[0][512 + 192 + k], qmf_c[256 + 64 + k]) +
353
MUL_F(qmfs->v[0][768 + k], qmf_c[384 + k]) +
354
MUL_F(qmfs->v[0][768 + 192 + k], qmf_c[384 + 64 + k]) +
355
MUL_F(qmfs->v[0][1024 + k], qmf_c[512 + k]) +
356
MUL_F(qmfs->v[0][1024 + 192 + k], qmf_c[512 + 64 + k]);
358
output[out++] = MUL_F(v0[k], qmf_c[k]) +
359
MUL_F(v0[64 + k], qmf_c[64 + k]) +
360
MUL_F(v0[128 + k], qmf_c[128 + k]) +
361
MUL_F(v0[192 + k], qmf_c[192 + k]) +
362
MUL_F(v0[256 + k], qmf_c[256 + k]) +
363
MUL_F(v0[320 + k], qmf_c[320 + k]) +
364
MUL_F(v0[384 + k], qmf_c[384 + k]) +
365
MUL_F(v0[448 + k], qmf_c[448 + k]) +
366
MUL_F(v0[512 + k], qmf_c[512 + k]) +
367
MUL_F(v0[576 + k], qmf_c[576 + k]);
373
void sbr_qmf_synthesis_64(sbr_info *sbr, qmfs_info *qmfs, qmf_t X[MAX_NTSRHFG][64],
376
ALIGN real_t x1[64], x2[64];
377
real_t scale = 1.f/64.f;
378
int16_t n, k, out = 0;
382
/* qmf subsample l */
383
for (l = 0; l < sbr->numTimeSlotsRate; l++)
388
memmove(qmfs->v[0] + 64, qmfs->v[0], (640-64)*sizeof(real_t));
389
memmove(qmfs->v[1] + 64, qmfs->v[1], (640-64)*sizeof(real_t));
391
v0 = qmfs->v[qmfs->v_index];
392
v1 = qmfs->v[(qmfs->v_index + 1) & 0x1];
393
qmfs->v_index = (qmfs->v_index + 1) & 0x1;
395
/* calculate 128 samples */
396
x1[0] = scale*QMF_RE(X[l][0]);
397
x2[63] = scale*QMF_IM(X[l][0]);
398
for (k = 0; k < 31; k++)
400
x1[2*k+1] = scale*(QMF_RE(X[l][2*k+1]) - QMF_RE(X[l][2*k+2]));
401
x1[2*k+2] = scale*(QMF_RE(X[l][2*k+1]) + QMF_RE(X[l][2*k+2]));
403
x2[61 - 2*k] = scale*(QMF_IM(X[l][2*k+2]) - QMF_IM(X[l][2*k+1]));
404
x2[62 - 2*k] = scale*(QMF_IM(X[l][2*k+2]) + QMF_IM(X[l][2*k+1]));
406
x1[63] = scale*QMF_RE(X[l][63]);
407
x2[0] = scale*QMF_IM(X[l][63]);
409
DCT4_64_kernel(x1, x1);
410
DCT4_64_kernel(x2, x2);
412
for (n = 0; n < 32; n++)
414
v0[ 2*n] = x2[2*n] - x1[2*n];
415
v1[63-2*n] = x2[2*n] + x1[2*n];
416
v0[ 2*n+1] = -x2[2*n+1] - x1[2*n+1];
417
v1[62-2*n] = -x2[2*n+1] + x1[2*n+1];
420
/* calculate 64 output samples and window */
421
for (k = 0; k < 64; k++)
423
output[out++] = MUL_F(v0[k], qmf_c[k]) +
424
MUL_F(v0[64 + k], qmf_c[64 + k]) +
425
MUL_F(v0[128 + k], qmf_c[128 + k]) +
426
MUL_F(v0[192 + k], qmf_c[192 + k]) +
427
MUL_F(v0[256 + k], qmf_c[256 + k]) +
428
MUL_F(v0[320 + k], qmf_c[320 + k]) +
429
MUL_F(v0[384 + k], qmf_c[384 + k]) +
430
MUL_F(v0[448 + k], qmf_c[448 + k]) +
431
MUL_F(v0[512 + k], qmf_c[512 + k]) +
432
MUL_F(v0[576 + k], qmf_c[576 + k]);
438
void memmove_sse_576(real_t *out, const real_t *in)
443
for (i = 0; i < 144; i++)
445
m[i] = _mm_load_ps(&in[i*4]);
447
for (i = 0; i < 144; i++)
449
_mm_store_ps(&out[i*4], m[i]);
453
void sbr_qmf_synthesis_64_sse(sbr_info *sbr, qmfs_info *qmfs, qmf_t X[MAX_NTSRHFG][64],
456
ALIGN real_t x1[64], x2[64];
457
real_t scale = 1.f/64.f;
458
int16_t n, k, out = 0;
462
/* qmf subsample l */
463
for (l = 0; l < sbr->numTimeSlotsRate; l++)
468
memmove_sse_576(qmfs->v[0] + 64, qmfs->v[0]);
469
memmove_sse_576(qmfs->v[1] + 64, qmfs->v[1]);
471
v0 = qmfs->v[qmfs->v_index];
472
v1 = qmfs->v[(qmfs->v_index + 1) & 0x1];
473
qmfs->v_index = (qmfs->v_index + 1) & 0x1;
475
/* calculate 128 samples */
476
x1[0] = scale*QMF_RE(X[l][0]);
477
x2[63] = scale*QMF_IM(X[l][0]);
478
for (k = 0; k < 31; k++)
480
x1[2*k+1] = scale*(QMF_RE(X[l][2*k+1]) - QMF_RE(X[l][2*k+2]));
481
x1[2*k+2] = scale*(QMF_RE(X[l][2*k+1]) + QMF_RE(X[l][2*k+2]));
483
x2[61 - 2*k] = scale*(QMF_IM(X[l][2*k+2]) - QMF_IM(X[l][2*k+1]));
484
x2[62 - 2*k] = scale*(QMF_IM(X[l][2*k+2]) + QMF_IM(X[l][2*k+1]));
486
x1[63] = scale*QMF_RE(X[l][63]);
487
x2[0] = scale*QMF_IM(X[l][63]);
489
DCT4_64_kernel(x1, x1);
490
DCT4_64_kernel(x2, x2);
492
for (n = 0; n < 32; n++)
494
v0[ 2*n ] = x2[2*n] - x1[2*n];
495
v1[63- 2*n ] = x2[2*n] + x1[2*n];
496
v0[ 2*n+1 ] = -x2[2*n+1] - x1[2*n+1];
497
v1[63-(2*n+1)] = -x2[2*n+1] + x1[2*n+1];
500
/* calculate 64 output samples and window */
501
for (k = 0; k < 64; k+=4)
503
__m128 m0, m1, m2, m3, m4, m5, m6, m7, m8, m9;
504
__m128 c0, c1, c2, c3, c4, c5, c6, c7, c8, c9;
505
__m128 s1, s2, s3, s4, s5, s6, s7, s8, s9;
507
m0 = _mm_load_ps(&v0[k]);
508
m1 = _mm_load_ps(&v0[k + 64]);
509
m2 = _mm_load_ps(&v0[k + 128]);
510
m3 = _mm_load_ps(&v0[k + 192]);
511
m4 = _mm_load_ps(&v0[k + 256]);
512
c0 = _mm_load_ps(&qmf_c[k]);
513
c1 = _mm_load_ps(&qmf_c[k + 64]);
514
c2 = _mm_load_ps(&qmf_c[k + 128]);
515
c3 = _mm_load_ps(&qmf_c[k + 192]);
516
c4 = _mm_load_ps(&qmf_c[k + 256]);
518
m0 = _mm_mul_ps(m0, c0);
519
m1 = _mm_mul_ps(m1, c1);
520
m2 = _mm_mul_ps(m2, c2);
521
m3 = _mm_mul_ps(m3, c3);
522
m4 = _mm_mul_ps(m4, c4);
524
s1 = _mm_add_ps(m0, m1);
525
s2 = _mm_add_ps(m2, m3);
526
s6 = _mm_add_ps(s1, s2);
528
m5 = _mm_load_ps(&v0[k + 320]);
529
m6 = _mm_load_ps(&v0[k + 384]);
530
m7 = _mm_load_ps(&v0[k + 448]);
531
m8 = _mm_load_ps(&v0[k + 512]);
532
m9 = _mm_load_ps(&v0[k + 576]);
533
c5 = _mm_load_ps(&qmf_c[k + 320]);
534
c6 = _mm_load_ps(&qmf_c[k + 384]);
535
c7 = _mm_load_ps(&qmf_c[k + 448]);
536
c8 = _mm_load_ps(&qmf_c[k + 512]);
537
c9 = _mm_load_ps(&qmf_c[k + 576]);
539
m5 = _mm_mul_ps(m5, c5);
540
m6 = _mm_mul_ps(m6, c6);
541
m7 = _mm_mul_ps(m7, c7);
542
m8 = _mm_mul_ps(m8, c8);
543
m9 = _mm_mul_ps(m9, c9);
545
s3 = _mm_add_ps(m4, m5);
546
s4 = _mm_add_ps(m6, m7);
547
s5 = _mm_add_ps(m8, m9);
548
s7 = _mm_add_ps(s3, s4);
549
s8 = _mm_add_ps(s5, s6);
550
s9 = _mm_add_ps(s7, s8);
552
_mm_store_ps(&output[out], s9);