ref: 064740662b4b78775c62f610a84dc17969a1aabf
dir: /libfaad/mdct.c/
/*
** FAAD2 - Freeware Advanced Audio (AAC) Decoder including SBR decoding
** Copyright (C) 2003 M. Bakker, Ahead Software AG, http://www.nero.com
**
** This program is free software; you can redistribute it and/or modify
** it under the terms of the GNU General Public License as published by
** the Free Software Foundation; either version 2 of the License, or
** (at your option) any later version.
**
** This program is distributed in the hope that it will be useful,
** but WITHOUT ANY WARRANTY; without even the implied warranty of
** MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
** GNU General Public License for more details.
**
** You should have received a copy of the GNU General Public License
** along with this program; if not, write to the Free Software
** Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
**
** Any non-GPL usage of this software or parts of this software is strictly
** forbidden.
**
** Commercial non-GPL licensing of this software is possible.
** For more info contact Ahead Software through [email protected].
**
** $Id: mdct.c,v 1.32 2003/11/04 21:43:30 menno Exp $
**/
/*
* Fast (I)MDCT Implementation using (I)FFT ((Inverse) Fast Fourier Transform)
* and consists of three steps: pre-(I)FFT complex multiplication, complex
* (I)FFT, post-(I)FFT complex multiplication,
*
* As described in:
* P. Duhamel, Y. Mahieux, and J.P. Petit, "A Fast Algorithm for the
* Implementation of Filter Banks Based on 'Time Domain Aliasing
* Cancellation�," IEEE Proc. on ICASSP�91, 1991, pp. 2209-2212.
*
*
* As of April 6th 2002 completely rewritten.
* This (I)MDCT can now be used for any data size n, where n is divisible by 8.
*
*/
#include "common.h"
#include "structs.h"
#include <stdlib.h>
#ifdef _WIN32_WCE
#define assert(x)
#else
#include <assert.h>
#endif
#include "cfft.h"
#include "mdct.h"
/* const_tab[]:
0: sqrt(2 / N)
1: cos(2 * PI / N)
2: sin(2 * PI / N)
3: cos(2 * PI * (1/8) / N)
4: sin(2 * PI * (1/8) / N)
*/
#ifndef FIXED_POINT
#ifdef _MSC_VER
#pragma warning(disable:4305)
#pragma warning(disable:4244)
#endif
real_t const_tab[][5] =
{
{ COEF_CONST(0.0312500000), COEF_CONST(0.9999952938), COEF_CONST(0.0030679568),
COEF_CONST(0.9999999265), COEF_CONST(0.0003834952) }, /* 2048 */
{ COEF_CONST(0.0322748612), COEF_CONST(0.9999946356), COEF_CONST(0.0032724866),
COEF_CONST(0.9999999404), COEF_CONST(0.0004090615) }, /* 1920 */
{ COEF_CONST(0.0441941738), COEF_CONST(0.9999811649), COEF_CONST(0.0061358847),
COEF_CONST(0.9999997020), COEF_CONST(0.0007669903) }, /* 1024 */
{ COEF_CONST(0.0456435465), COEF_CONST(0.9999786019), COEF_CONST(0.0065449383),
COEF_CONST(0.9999996424), COEF_CONST(0.0008181230) }, /* 960 */
{ COEF_CONST(0.0883883476), COEF_CONST(0.9996988177), COEF_CONST(0.0245412290),
COEF_CONST(0.9999952912), COEF_CONST(0.0030679568) }, /* 256 */
{ COEF_CONST(0.0912870929), COEF_CONST(0.9996573329), COEF_CONST(0.0261769500),
COEF_CONST(0.9999946356), COEF_CONST(0.0032724866) } /* 240 */
#ifdef SSR_DEC
,{ COEF_CONST(0.062500000), COEF_CONST(0.999924702), COEF_CONST(0.012271538),
COEF_CONST(0.999998823), COEF_CONST(0.00153398) }, /* 512 */
{ COEF_CONST(0.176776695), COEF_CONST(0.995184727), COEF_CONST(0.09801714),
COEF_CONST(0.999924702), COEF_CONST(0.012271538) } /* 64 */
#endif
};
#else
real_t const_tab[][5] =
{
{ COEF_CONST(1), COEF_CONST(0.9999952938), COEF_CONST(0.0030679568),
COEF_CONST(0.9999999265), COEF_CONST(0.0003834952) }, /* 2048 */
{ COEF_CONST(/* sqrt(1024/960) */ 1.03279556), COEF_CONST(0.9999946356), COEF_CONST(0.0032724866),
COEF_CONST(0), COEF_CONST(0.0004090615) }, /* 1920 */
{ COEF_CONST(1), COEF_CONST(0.9999811649), COEF_CONST(0.0061358847),
COEF_CONST(0.9999997020), COEF_CONST(0.0007669903) }, /* 1024 */
{ COEF_CONST(/* sqrt(512/480) */ 1.03279556), COEF_CONST(0.9999786019), COEF_CONST(0.0065449383),
COEF_CONST(0.9999996424), COEF_CONST(0.0008181230) }, /* 960 */
{ COEF_CONST(1), COEF_CONST(0.9996988177), COEF_CONST(0.0245412290),
COEF_CONST(0.9999952912), COEF_CONST(0.0030679568) }, /* 256 */
{ COEF_CONST(/* sqrt(256/240) */ 1.03279556), COEF_CONST(0.9996573329), COEF_CONST(0.0261769500),
COEF_CONST(0.9999946356), COEF_CONST(0.0032724866) } /* 240 */
#ifdef SSR_DEC
,{ COEF_CONST(0), COEF_CONST(0.999924702), COEF_CONST(0.012271538),
COEF_CONST(0.999998823), COEF_CONST(0.00153398) }, /* 512 */
{ COEF_CONST(0), COEF_CONST(0.995184727), COEF_CONST(0.09801714),
COEF_CONST(0.999924702), COEF_CONST(0.012271538) } /* 64 */
#endif
};
#endif
uint8_t map_N_to_idx(uint16_t N)
{
/* gives an index into const_tab above */
/* for normal AAC deocding (eg. no scalable profile) only */
/* index 0 and 4 will be used */
switch(N)
{
case 2048: return 0;
case 1920: return 1;
case 1024: return 2;
case 960: return 3;
case 256: return 4;
case 240: return 5;
#ifdef SSR_DEC
case 512: return 6;
case 64: return 7;
#endif
}
return 0;
}
mdct_info *faad_mdct_init(uint16_t N)
{
uint16_t k, N_idx;
real_t cangle, sangle, c, s, cold;
real_t scale;
mdct_info *mdct = (mdct_info*)malloc(sizeof(mdct_info));
assert(N % 8 == 0);
mdct->N = N;
mdct->sincos = (complex_t*)malloc(N/4*sizeof(complex_t));
N_idx = map_N_to_idx(N);
scale = const_tab[N_idx][0];
cangle = const_tab[N_idx][1];
sangle = const_tab[N_idx][2];
c = const_tab[N_idx][3];
s = const_tab[N_idx][4];
/* (co)sine table build using recurrence relations */
/* this can also be done using static table lookup or */
/* some form of interpolation */
for (k = 0; k < N/4; k++)
{
#if 1
RE(mdct->sincos[k]) = -1*MUL_C_C(c,scale);
IM(mdct->sincos[k]) = -1*MUL_C_C(s,scale);
cold = c;
c = MUL_C_C(c,cangle) - MUL_C_C(s,sangle);
s = MUL_C_C(s,cangle) + MUL_C_C(cold,sangle);
#else
/* no recurrence, just sines */
RE(mdct->sincos[k]) = -scale*cos(2.0*M_PI*(k+1./8.) / (float)N);
IM(mdct->sincos[k]) = -scale*sin(2.0*M_PI*(k+1./8.) / (float)N);
#endif
}
/* initialise fft */
mdct->cfft = cffti(N/4);
return mdct;
}
void faad_mdct_end(mdct_info *mdct)
{
if (mdct != NULL)
{
cfftu(mdct->cfft);
if (mdct->sincos) free(mdct->sincos);
free(mdct);
}
}
void faad_imdct(mdct_info *mdct, real_t *X_in, real_t *X_out)
{
uint16_t k;
complex_t x;
complex_t Z1[512];
complex_t *sincos = mdct->sincos;
uint16_t N = mdct->N;
uint16_t N2 = N >> 1;
uint16_t N4 = N >> 2;
uint16_t N8 = N >> 3;
/* pre-IFFT complex multiplication */
for (k = 0; k < N4; k++)
{
RE(Z1[k]) = MUL_R_C(X_in[N2 - 1 - 2*k], RE(sincos[k])) - MUL_R_C(X_in[2*k], IM(sincos[k]));
IM(Z1[k]) = MUL_R_C(X_in[2*k], RE(sincos[k])) + MUL_R_C(X_in[N2 - 1 - 2*k], IM(sincos[k]));
}
/* complex IFFT, any non-scaling FFT can be used here */
cfftb(mdct->cfft, Z1);
/* post-IFFT complex multiplication */
for (k = 0; k < N4; k++)
{
RE(x) = RE(Z1[k]);
IM(x) = IM(Z1[k]);
RE(Z1[k]) = MUL_R_C(RE(x), RE(sincos[k])) - MUL_R_C(IM(x), IM(sincos[k]));
IM(Z1[k]) = MUL_R_C(IM(x), RE(sincos[k])) + MUL_R_C(RE(x), IM(sincos[k]));
#if (REAL_BITS == 16)
if (abs(RE(Z1[k])) > REAL_CONST(16383.5))
{
if (RE(Z1[k]) > 0) RE(Z1[k]) = REAL_CONST(32767.0);
else RE(Z1[k]) = REAL_CONST(-32767.0);
} else {
RE(Z1[k]) *= 2;
}
if (abs(IM(Z1[k])) > REAL_CONST(16383.5))
{
if (IM(Z1[k]) > 0) IM(Z1[k]) = REAL_CONST(32767.0);
else IM(Z1[k]) = REAL_CONST(-32767.0);
} else {
IM(Z1[k]) *= 2;
}
#endif
}
/* reordering */
for (k = 0; k < N8; k++)
{
X_out[ 2*k] = IM(Z1[N8 + k]);
X_out[ 1 + 2*k] = -RE(Z1[N8 - 1 - k]);
X_out[N4 + 2*k] = RE(Z1[ k]);
X_out[N4 + 1 + 2*k] = -IM(Z1[N4 - 1 - k]);
X_out[N2 + 2*k] = RE(Z1[N8 + k]);
X_out[N2 + 1 + 2*k] = -IM(Z1[N8 - 1 - k]);
X_out[N2 + N4 + 2*k] = -IM(Z1[ k]);
X_out[N2 + N4 + 1 + 2*k] = RE(Z1[N4 - 1 - k]);
}
#if 0
{
float max_fp = 0;
for (k = 0; k < N; k++)
{
if (fabs(X_out[k]/(float)(REAL_PRECISION)) > max_fp)
max_fp = fabs(X_out[k]/(float)(REAL_PRECISION));
}
if (max_fp > 32767>>1)
printf("m: %f\n", max_fp);
}
#endif
}
#ifdef LTP_DEC
void faad_mdct(mdct_info *mdct, real_t *X_in, real_t *X_out)
{
uint16_t k;
complex_t x;
complex_t Z1[512];
complex_t *sincos = mdct->sincos;
uint16_t N = mdct->N;
uint16_t N2 = N >> 1;
uint16_t N4 = N >> 2;
uint16_t N8 = N >> 3;
#ifndef FIXED_POINT
real_t scale = REAL_CONST(N);
#else
real_t scale = REAL_CONST(4.0/N);
#endif
/* pre-FFT complex multiplication */
for (k = 0; k < N8; k++)
{
uint16_t n = k << 1;
RE(x) = X_in[N - N4 - 1 - n] + X_in[N - N4 + n];
IM(x) = X_in[ N4 + n] - X_in[ N4 - 1 - n];
RE(Z1[k]) = -MUL_R_C(RE(x), RE(sincos[k])) - MUL_R_C(IM(x), IM(sincos[k]));
IM(Z1[k]) = -MUL_R_C(IM(x), RE(sincos[k])) + MUL_R_C(RE(x), IM(sincos[k]));
RE(Z1[k]) = MUL(RE(Z1[k]), scale);
IM(Z1[k]) = MUL(IM(Z1[k]), scale);
RE(x) = X_in[N2 - 1 - n] - X_in[ n];
IM(x) = X_in[N2 + n] + X_in[N - 1 - n];
RE(Z1[k + N8]) = -MUL_R_C(RE(x), RE(sincos[k + N8])) - MUL_R_C(IM(x), IM(sincos[k + N8]));
IM(Z1[k + N8]) = -MUL_R_C(IM(x), RE(sincos[k + N8])) + MUL_R_C(RE(x), IM(sincos[k + N8]));
RE(Z1[k + N8]) = MUL(RE(Z1[k + N8]), scale);
IM(Z1[k + N8]) = MUL(IM(Z1[k + N8]), scale);
}
/* complex FFT, any non-scaling FFT can be used here */
cfftf(mdct->cfft, Z1);
/* post-FFT complex multiplication */
for (k = 0; k < N4; k++)
{
uint16_t n = k << 1;
RE(x) = MUL_R_C(RE(Z1[k]), RE(sincos[k])) + MUL_R_C(IM(Z1[k]), IM(sincos[k]));
IM(x) = MUL_R_C(IM(Z1[k]), RE(sincos[k])) - MUL_R_C(RE(Z1[k]), IM(sincos[k]));
X_out[ n] = RE(x);
X_out[N2 - 1 - n] = -IM(x);
X_out[N2 + n] = IM(x);
X_out[N - 1 - n] = -RE(x);
}
}
#endif