ref: 0b79b1c3c52c4ad1535bf8429ae5b7c701c4a907
parent: e82315d28bb5b1e866dad32fadc801c14c5cb1b4
author: menno <menno>
date: Wed Feb 20 08:05:57 EST 2002
- Nicer defines in common.h - 1% speedup in MDCT
--- a/libfaad/common.h
+++ b/libfaad/common.h
@@ -16,7 +16,7 @@
** along with this program; if not, write to the Free Software
** Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
**
-** $Id: common.h,v 1.1 2002/02/18 10:01:05 menno Exp $
+** $Id: common.h,v 1.2 2002/02/20 13:05:57 menno Exp $
**/
#ifndef __COMMON_H__
@@ -52,9 +52,18 @@
#endif
-//#define USE_DOUBLE_PRECISION
+/* COMPILE TIME DEFINITIONS */
+/* use double precision */
+/* #define USE_DOUBLE_PRECISION */
+/* use table lookup twiddle factors in MDCT [more memory, higher speed],
+ otherwise recurrence relations are used [no memory usage, lower speed] */
+#define USE_TWIDDLE_TABLE
+
+/* END COMPILE TIME DEFINITIONS */
+
+
#if defined(_WIN32)
@@ -66,16 +75,7 @@
typedef __int8 int8_t;
typedef float float32_t;
-#ifndef USE_DOUBLE_PRECISION
-typedef float real_t;
-#ifdef __ICL /* only Intel C compiler has fmath ??? */
-#define USE_FMATH
-#endif
-#else
-typedef double real_t;
-#endif
-
#elif defined(LINUX) || defined(DJGPP)
@@ -91,27 +91,57 @@
typedef float float32_t;
#endif
-#ifndef USE_DOUBLE_PRECISION
-typedef float real_t;
-#else
-typedef double real_t;
-#endif
-
#else /* Some other OS */
#include <inttypes.h>
-#ifndef USE_DOUBLE_PRECISION
-typedef float real_t;
-#else
-typedef double real_t;
#endif
+#ifndef USE_DOUBLE_PRECISION
+
+ typedef float real_t;
+
+ #ifdef __ICL /* only Intel C compiler has fmath ??? */
+
+ #include <mathf.h>
+
+ #define sin sinf
+ #define cos cosf
+ #define pow powf
+ #define floor floorf
+ #define sqrt sqrtf
+
+ #else
+
+ #include <math.h>
+
+#ifdef HAVE_SINF
+# define sin sinf
#endif
+#ifdef HAVE_COSF
+# define cos cosf
+#endif
+#ifdef HAVE_POWF
+# define pow powf
+#endif
+#ifdef HAVE_FLOORF
+# define floor floorf
+#endif
+#ifdef HAVE_SQRTF
+# define sqrt sqrtf
+#endif
+ #endif
+
+#else
+
+ typedef double real_t;
+ #include <math.h>
+
+#endif
#ifdef __cplusplus
}
--- a/libfaad/decoder.c
+++ b/libfaad/decoder.c
@@ -16,7 +16,7 @@
** along with this program; if not, write to the Free Software
** Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
**
-** $Id: decoder.c,v 1.8 2002/02/18 10:01:05 menno Exp $
+** $Id: decoder.c,v 1.9 2002/02/20 13:05:57 menno Exp $
**/
#include <stdlib.h>
@@ -250,7 +250,7 @@
#else
real_t *pow2_table = NULL;
#endif
- uint8_t *window_shape_prev = hDecoder->window_shape_prev;
+ uint8_t *window_shape_prev = hDecoder->window_shape_prev;
real_t **time_state = hDecoder->time_state;
real_t **time_out = hDecoder->time_out;
fb_info *fb = &hDecoder->fb;
--- a/libfaad/drc.c
+++ b/libfaad/drc.c
@@ -16,16 +16,11 @@
** along with this program; if not, write to the Free Software
** Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
**
-** $Id: drc.c,v 1.2 2002/02/18 10:01:05 menno Exp $
+** $Id: drc.c,v 1.3 2002/02/20 13:05:57 menno Exp $
**/
#include "common.h"
-#ifdef USE_FMATH
-#include <mathf.h>
-#else
-#include <math.h>
-#endif
#include <memory.h>
#include "syntax.h"
#include "drc.h"
@@ -58,17 +53,9 @@
/* Decode DRC gain factor */
if (drc->dyn_rng_sgn[bd]) /* compress */
-#ifdef USE_FMATH
- factor = powf(2.0f, (-drc->ctrl1 * drc->dyn_rng_ctl[bd]/24.0f));
-#else
factor = (real_t)pow(2.0, (-drc->ctrl1 * drc->dyn_rng_ctl[bd]/24.0));
-#endif
else /* boost */
-#ifdef USE_FMATH
- factor = powf(2.0f, (drc->ctrl2 * drc->dyn_rng_ctl[bd]/24.0f));
-#else
factor = (real_t)pow(2.0, (drc->ctrl2 * drc->dyn_rng_ctl[bd]/24.0));
-#endif
/* Level alignment between different programs (if desired) */
/* If program reference normalization is done in the digital domain,
@@ -78,11 +65,7 @@
modification avoids problems with reduced DAC SNR (if signal is
attenuated) or clipping (if signal is boosted)
*/
-#ifdef USE_FMATH
- factor *= powf(0.5f, ((DRC_REF_LEVEL - drc->prog_ref_level)/24.0f));
-#else
factor *= (real_t)pow(0.5, ((DRC_REF_LEVEL - drc->prog_ref_level)/24.0));
-#endif
/* Apply gain factor */
for (i = bottom; i < top; i++)
--- a/libfaad/filtbank.c
+++ b/libfaad/filtbank.c
@@ -16,17 +16,12 @@
** along with this program; if not, write to the Free Software
** Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
**
-** $Id: filtbank.c,v 1.2 2002/02/18 10:01:05 menno Exp $
+** $Id: filtbank.c,v 1.3 2002/02/20 13:05:57 menno Exp $
**/
#include "common.h"
#include <stdlib.h>
-#ifdef USE_FMATH
-#include <mathf.h>
-#else
-#include <math.h>
-#endif
#include <assert.h>
#include "filtbank.h"
#include "syntax.h"
@@ -41,7 +36,8 @@
{uint16_t i;
- make_fft_order(fb->unscrambled64, fb->unscrambled512);
+ mdct_init(&(fb->mdct256), 256);
+ mdct_init(&(fb->mdct2048), 2048);
fb->sin_long = malloc(BLOCK_LEN_LONG*sizeof(real_t));
fb->sin_short = malloc(BLOCK_LEN_SHORT*sizeof(real_t));
@@ -53,21 +49,16 @@
/* calculate the sine windows */
for (i = 0; i < BLOCK_LEN_LONG; i++)
-#ifdef USE_FMATH
- fb->sin_long[i] = sinf(M_PI / (2.0f * BLOCK_LEN_LONG) * (i + 0.5));
-#else
fb->sin_long[i] = (real_t)sin(M_PI / (2.0 * BLOCK_LEN_LONG) * (i + 0.5));
-#endif
for (i = 0; i < BLOCK_LEN_SHORT; i++)
-#ifdef USE_FMATH
- fb->sin_short[i] = sinf(M_PI / (2.0f * BLOCK_LEN_SHORT) * (i + 0.5));
-#else
fb->sin_short[i] = (real_t)sin(M_PI / (2.0 * BLOCK_LEN_SHORT) * (i + 0.5));
-#endif
}
void filter_bank_end(fb_info *fb)
{+ mdct_end(&(fb->mdct256));
+ mdct_end(&(fb->mdct2048));
+
if (fb->sin_long) free(fb->sin_long);
if (fb->sin_short) free(fb->sin_short);
}
@@ -164,10 +155,10 @@
switch (len)
{case 2048:
- IMDCT_long(in_data, out_data, fb->unscrambled512);
+ IMDCT_long(&(fb->mdct2048), in_data, out_data);
return;
case 256:
- IMDCT_short(in_data, out_data, fb->unscrambled64);
+ IMDCT_short(&(fb->mdct256), in_data, out_data);
return;
}
}
@@ -177,10 +168,10 @@
switch (len)
{case 2048:
- MDCT_long(in_data, out_data, fb->unscrambled512);
+ MDCT_long(&(fb->mdct2048), in_data, out_data);
return;
case 256:
- MDCT_short(in_data, out_data, fb->unscrambled64);
+ MDCT_short(&(fb->mdct256), in_data, out_data);
return;
}
}
--- a/libfaad/filtbank.h
+++ b/libfaad/filtbank.h
@@ -16,7 +16,7 @@
** along with this program; if not, write to the Free Software
** Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
**
-** $Id: filtbank.h,v 1.2 2002/02/18 10:01:05 menno Exp $
+** $Id: filtbank.h,v 1.3 2002/02/20 13:05:57 menno Exp $
**/
#ifndef __FILTBANK_H__
@@ -26,6 +26,7 @@
extern "C" {#endif
+#include "mdct.h"
#define BLOCK_LEN_LONG 1024
#define BLOCK_LEN_SHORT 128
@@ -33,11 +34,11 @@
typedef struct
{- uint16_t unscrambled64[64];
- uint16_t unscrambled512[512];
-
real_t *sin_long;
real_t *sin_short;
+
+ mdct_info mdct256;
+ mdct_info mdct2048;
} fb_info;
void filter_bank_init(fb_info *fb);
--- a/libfaad/is.c
+++ b/libfaad/is.c
@@ -16,16 +16,11 @@
** along with this program; if not, write to the Free Software
** Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
**
-** $Id: is.c,v 1.2 2002/02/18 10:01:05 menno Exp $
+** $Id: is.c,v 1.3 2002/02/20 13:05:57 menno Exp $
**/
#include "common.h"
-#ifdef USE_FMATH
-#include <mathf.h>
-#else
-#include <math.h>
-#endif
#include "syntax.h"
#include "is.h"
@@ -55,11 +50,7 @@
scale = is_intensity(icsr, g, sfb) *
invert_intensity(ics, g, sfb) *
-#ifdef USE_FMATH
- powf(0.5f, (0.25f*icsr->scale_factors[g][sfb]));
-#else
(real_t)pow(0.5, (0.25*icsr->scale_factors[g][sfb]));
-#endif
/* Scale from left to right channel,
do not touch left channel */
--- a/libfaad/mdct.c
+++ b/libfaad/mdct.c
@@ -16,36 +16,60 @@
** along with this program; if not, write to the Free Software
** Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
**
-** $Id: mdct.c,v 1.2 2002/02/18 10:01:05 menno Exp $
+** $Id: mdct.c,v 1.3 2002/02/20 13:05:57 menno Exp $
**/
#include "common.h"
-#ifdef USE_FMATH
-#include <mathf.h>
-#else
-#include <math.h>
-#endif
+#include <stdlib.h>
#include "mdct.h"
+void mdct_init(mdct_info *mdct, uint16_t len)
+{+ uint16_t i;
-void MDCT_long(fftw_real *in_data, fftw_real *out_data, uint16_t *unscrambled)
+ mdct->len = len;
+
+ mdct->unscrambled = malloc(len/4*sizeof(uint16_t));
+ make_fft_order(mdct->unscrambled, len/4);
+
+#ifdef USE_TWIDDLE_TABLE
+ mdct->twiddlers = malloc(len/2*sizeof(real_t));
+ for (i = 0; i < len/4; i++)
+ {+ real_t angle = 2.0f * M_PI * (i + 1.0f/8.0f) / (real_t)len;
+ mdct->twiddlers[2*i] = cos(angle);
+ mdct->twiddlers[2*i + 1] = sin(angle);
+ }
+#endif
+}
+
+void mdct_end(mdct_info *mdct)
{+ if (mdct->unscrambled) free(mdct->unscrambled);
+#ifdef USE_TWIDDLE_TABLE
+ if (mdct->twiddlers) free(mdct->twiddlers);
+#endif
+}
+
+void MDCT_long(mdct_info *mdct, fftw_real *in_data, fftw_real *out_data)
+{fftw_complex FFTarray[512];
- fftw_real tempr, tempi, c, s, cold, cfreq, sfreq;
- fftw_real fac,cosfreq8,sinfreq8;
+ fftw_real tempr, tempi, fac;
+
+#ifdef USE_TWIDDLE_TABLE
+ /* use twiddle factor tables */
+ real_t *twiddlers = mdct->twiddlers;
+#else
+ /* temps for pre and post twiddle */
+ real_t cosfreq8, sinfreq8, c, s, cold, cfreq, sfreq;
+#endif
uint16_t i;
- uint16_t b = 2048 >> 1;
- uint16_t N4 = 2048 >> 2;
- uint16_t N2 = 2048 >> 1;
- uint16_t a = 2048 - b;
- uint16_t a2 = a >> 1;
- uint16_t a4 = a >> 2;
- uint16_t b4 = b >> 2;
fac = 2.; /* 2 from MDCT inverse to forward */
+#ifndef USE_TWIDDLE_TABLE
/* prepare for recurrence relation in pre-twiddle */
cfreq = 0.99999529123306274f;
sfreq = 0.0030679567717015743f;
@@ -54,22 +78,27 @@
c = cosfreq8;
s = sinfreq8;
+#endif
- for (i = 0; i < N4; i++)
+ for (i = 0; i < 512; i++)
{- uint16_t n = 2048 / 2 - 1 - 2 * i;
- if (i < b4)
- tempr = in_data[a2 + n] + in_data[2048 + a2 - 1 - n];
+ uint16_t n = 1023 - (i << 1);
+ if (i < 256)
+ tempr = in_data[512 + n] + in_data[2559 - n];
else
- tempr = in_data[a2 + n] - in_data[a2 - 1 - n];
+ tempr = in_data[512 + n] - in_data[511 - n];
- n = 2 * i;
- if (i < a4)
- tempi = in_data[a2 + n] - in_data[a2 - 1 - n];
+ n = (i << 1);
+ if (i < 256)
+ tempi = in_data[512 + n] - in_data[511 - n];
else
- tempi = in_data[a2 + n] + in_data[2048 + a2 - 1 - n];
+ tempi = in_data[512 + n] + in_data[2559 - n];
/* calculate pre-twiddled FFT input */
+#ifdef USE_TWIDDLE_TABLE
+ FFTarray[i].re = tempr * twiddlers[n] + tempi * twiddlers[n + 1];
+ FFTarray[i].im = tempi * twiddlers[n] - tempr * twiddlers[n + 1];
+#else
FFTarray[i].re = tempr * c + tempi * s;
FFTarray[i].im = tempi * c - tempr * s;
@@ -77,6 +106,7 @@
cold = c;
c = c * cfreq - s * sfreq;
s = s * cfreq + cold * sfreq;
+#endif
}
/* Perform in-place complex FFT of length N/4 */
@@ -83,50 +113,61 @@
pfftw_512(FFTarray);
+#ifndef USE_TWIDDLE_TABLE
/* prepare for recurrence relations in post-twiddle */
c = cosfreq8;
s = sinfreq8;
+#endif
/* post-twiddle FFT output and then get output data */
- for (i = 0; i < N4; i++)
+ for (i = 0; i < 512; i++)
{- /* get post-twiddled FFT output */
- uint16_t unscr = unscrambled[i];
+ uint16_t n = i << 1;
+ uint16_t unscr = mdct->unscrambled[i];
+ /* get post-twiddled FFT output */
+#ifdef USE_TWIDDLE_TABLE
+ tempr = fac * (FFTarray[unscr].re * twiddlers[n] + FFTarray[unscr].im * twiddlers[n + 1]);
+ tempi = fac * (FFTarray[unscr].im * twiddlers[n] - FFTarray[unscr].re * twiddlers[n + 1]);
+#else
tempr = fac * (FFTarray[unscr].re * c + FFTarray[unscr].im * s);
tempi = fac * (FFTarray[unscr].im * c - FFTarray[unscr].re * s);
+#endif
/* fill in output values */
- out_data[2 * i] = -tempr; /* first half even */
- out_data[N2 - 1 - 2 * i] = tempi; /* first half odd */
- out_data[N2 + 2 * i] = -tempi; /* second half even */
- out_data[2048 - 1 - 2 * i] = tempr; /* second half odd */
+ out_data[n] = -tempr; /* first half even */
+ out_data[1023 - n] = tempi; /* first half odd */
+ out_data[1024 + n] = -tempi; /* second half even */
+ out_data[2047 - n] = tempr; /* second half odd */
+#ifndef USE_TWIDDLE_TABLE
/* use recurrence to prepare cosine and sine for next value of i */
cold = c;
c = c * cfreq - s * sfreq;
s = s * cfreq + cold * sfreq;
+#endif
}
}
-void MDCT_short(fftw_real *in_data, fftw_real *out_data, uint16_t *unscrambled)
+void MDCT_short(mdct_info *mdct, fftw_real *in_data, fftw_real *out_data)
{fftw_complex FFTarray[64]; /* the array for in-place FFT */
- fftw_real tempr, tempi, c, s, cold, cfreq, sfreq; /* temps for pre and post twiddle */
- fftw_real fac,cosfreq8,sinfreq8;
+ fftw_real tempr, tempi, fac;
+
+#ifdef USE_TWIDDLE_TABLE
+ /* use twiddle factor tables */
+ real_t *twiddlers = mdct->twiddlers;
+#else
+ /* temps for pre and post twiddle */
+ real_t cosfreq8, sinfreq8, c, s, cold, cfreq, sfreq;
+#endif
uint16_t i;
- uint16_t b = 256 >> 1;
- uint16_t N4 = 256 >> 2;
- uint16_t N2 = 256 >> 1;
- uint16_t a = 256 - b;
- uint16_t a2 = a >> 1;
- uint16_t a4 = a >> 2;
- uint16_t b4 = b >> 2;
/* Choosing to allocate 2/N factor to Inverse Xform! */
fac = 2.; /* 2 from MDCT inverse to forward */
+#ifndef USE_TWIDDLE_TABLE
/* prepare for recurrence relation in pre-twiddle */
cfreq = 0.99969881772994995f;
sfreq = 0.024541229009628296f;
@@ -135,22 +176,27 @@
c = cosfreq8;
s = sinfreq8;
+#endif
- for (i = 0; i < N4; i++)
+ for (i = 0; i < 64; i++)
{- uint16_t n = 256 / 2 - 1 - 2 * i;
- if (i < b4)
- tempr = in_data[a2 + n] + in_data[256 + a2 - 1 - n];
+ uint16_t n = 127 - (i << 1);
+ if (i < 32)
+ tempr = in_data[64 + n] + in_data[319 - n];
else
- tempr = in_data[a2 + n] - in_data[a2 - 1 - n];
+ tempr = in_data[64 + n] - in_data[63 - n];
- n = 2 * i;
- if (i < a4)
- tempi = in_data[a2 + n] - in_data[a2 - 1 - n];
+ n = i << 1;
+ if (i < 32)
+ tempi = in_data[64 + n] - in_data[63 - n];
else
- tempi = in_data[a2 + n] + in_data[256 + a2 - 1 - n];
+ tempi = in_data[64 + n] + in_data[319 - n];
/* calculate pre-twiddled FFT input */
+#ifdef USE_TWIDDLE_TABLE
+ FFTarray[i].re = tempr * twiddlers[n] + tempi * twiddlers[n + 1];
+ FFTarray[i].im = tempi * twiddlers[n] - tempr * twiddlers[n + 1];
+#else
FFTarray[i].re = tempr * c + tempi * s;
FFTarray[i].im = tempi * c - tempr * s;
@@ -158,50 +204,65 @@
cold = c;
c = c * cfreq - s * sfreq;
s = s * cfreq + cold * sfreq;
+#endif
}
/* Perform in-place complex FFT of length N/4 */
pfftw_64(FFTarray);
+#ifndef USE_TWIDDLE_TABLE
/* prepare for recurrence relations in post-twiddle */
c = cosfreq8;
s = sinfreq8;
+#endif
/* post-twiddle FFT output and then get output data */
- for (i = 0; i < N4; i++)
+ for (i = 0; i < 64; i++)
{- uint16_t unscr = unscrambled[i];
+ uint16_t n = i << 1;
+ uint16_t unscr = mdct->unscrambled[i];
+#ifdef USE_TWIDDLE_TABLE
+ tempr = fac * (FFTarray[unscr].re * twiddlers[n] + FFTarray[unscr].im * twiddlers[n + 1]);
+ tempi = fac * (FFTarray[unscr].im * twiddlers[n] - FFTarray[unscr].re * twiddlers[n + 1]);
+#else
tempr = fac * (FFTarray[unscr].re * c + FFTarray[unscr].im * s);
tempi = fac * (FFTarray[unscr].im * c - FFTarray[unscr].re * s);
+#endif
/* fill in output values */
- out_data[2 * i] = -tempr; /* first half even */
- out_data[N2 - 1 - 2 * i] = tempi; /* first half odd */
- out_data[N2 + 2 * i] = -tempi; /* second half even */
- out_data[256 - 1 - 2 * i] = tempr; /* second half odd */
+ out_data[n] = -tempr; /* first half even */
+ out_data[127 - n] = tempi; /* first half odd */
+ out_data[128 + n] = -tempi; /* second half even */
+ out_data[255 - n] = tempr; /* second half odd */
+#ifndef USE_TWIDDLE_TABLE
/* use recurrence to prepare cosine and sine for next value of i */
cold = c;
c = c * cfreq - s * sfreq;
s = s * cfreq + cold * sfreq;
+#endif
}
}
-void IMDCT_long(fftw_real *in_data, fftw_real *out_data, uint16_t *unscrambled)
+void IMDCT_long(mdct_info *mdct, fftw_real *in_data, fftw_real *out_data)
{fftw_complex FFTarray[512]; /* the array for in-place FFT */
- fftw_real tempr, tempi, c, s, cold, cfreq, sfreq; /* temps for pre and post twiddle */
+ fftw_real tempr, tempi, fac;
- fftw_real fac, cosfreq8, sinfreq8;
+#ifdef USE_TWIDDLE_TABLE
+ /* use twiddle factor tables */
+ real_t *twiddlers = mdct->twiddlers;
+#else
+ /* temps for pre and post twiddle */
+ real_t cosfreq8, sinfreq8, c, s, cold, cfreq, sfreq;
+#endif
uint16_t i;
- uint16_t Nd2 = 2048 >> 1;
- uint16_t Nd4 = 2048 >> 2;
- uint16_t Nd8 = 2048 >> 3;
/* Choosing to allocate 2/N factor to Inverse Xform! */
fac = 0.0009765625f;
+#ifndef USE_TWIDDLE_TABLE
/* prepare for recurrence relation in pre-twiddle */
cfreq = 0.99999529123306274f;
sfreq = 0.0030679567717015743f;
@@ -210,15 +271,21 @@
c = cosfreq8;
s = sinfreq8;
+#endif
- for (i = 0; i < Nd4; i++)
+ for (i = 0; i < 512; i++)
{- uint16_t unscr = unscrambled[i];
+ uint16_t n = i << 1;
+ uint16_t unscr = mdct->unscrambled[i];
- tempr = -in_data[2 * i];
- tempi = in_data[Nd2 - 1 - 2 * i];
+ tempr = -in_data[n];
+ tempi = in_data[1023 - n];
/* calculate pre-twiddled FFT input */
+#ifdef USE_TWIDDLE_TABLE
+ FFTarray[unscr].re = tempr * twiddlers[n] - tempi * twiddlers[n + 1];
+ FFTarray[unscr].im = tempi * twiddlers[n] + tempr * twiddlers[n + 1];
+#else
FFTarray[unscr].re = tempr * c - tempi * s;
FFTarray[unscr].im = tempi * c + tempr * s;
@@ -226,55 +293,72 @@
cold = c;
c = c * cfreq - s * sfreq;
s = s * cfreq + cold * sfreq;
+#endif
}
/* Perform in-place complex IFFT of length N/4 */
pfftwi_512(FFTarray);
+#ifndef USE_TWIDDLE_TABLE
/* prepare for recurrence relations in post-twiddle */
c = cosfreq8;
s = sinfreq8;
+#endif
/* post-twiddle FFT output and then get output data */
- for (i = 0; i < Nd4; i++)
+ for (i = 0; i < 512; i++)
{+ uint16_t n = i << 1;
/* get post-twiddled FFT output */
+#ifdef USE_TWIDDLE_TABLE
+ tempr = fac * (FFTarray[i].re * twiddlers[n] - FFTarray[i].im * twiddlers[n + 1]);
+ tempi = fac * (FFTarray[i].im * twiddlers[n] + FFTarray[i].re * twiddlers[n + 1]);
+#else
tempr = fac * (FFTarray[i].re * c - FFTarray[i].im * s);
tempi = fac * (FFTarray[i].im * c + FFTarray[i].re * s);
+#endif
/* fill in output values */
- out_data [Nd2 + Nd4 - 1 - 2 * i] = tempr;
- if (i < Nd8)
- out_data[Nd2 + Nd4 + 2 * i] = tempr;
+ out_data [1535 - n] = tempr;
+ if (i < 256)
+ out_data[1536 + n] = tempr;
else
- out_data[2 * i - Nd4] = -tempr;
+ out_data[n - 512] = -tempr;
- out_data [Nd4 + 2 * i] = tempi;
- if (i < Nd8)
- out_data[Nd4 - 1 - 2 * i] = -tempi;
+ out_data [512 + n] = tempi;
+ if (i < 256)
+ out_data[511 - n] = -tempi;
else
- out_data[Nd4 + 2048 - 1 - 2*i] = tempi;
+ out_data[2559 - n] = tempi;
+#ifndef USE_TWIDDLE_TABLE
/* use recurrence to prepare cosine and sine for next value of i */
cold = c;
c = c * cfreq - s * sfreq;
s = s * cfreq + cold * sfreq;
+#endif
}
}
-void IMDCT_short(fftw_real *in_data, fftw_real *out_data, uint16_t *unscrambled)
+void IMDCT_short(mdct_info *mdct, fftw_real *in_data, fftw_real *out_data)
{fftw_complex FFTarray[64]; /* the array for in-place FFT */
- fftw_real tempr, tempi, c, s, cold, cfreq, sfreq; /* temps for pre and post twiddle */
- fftw_real fac, cosfreq8, sinfreq8;
+ fftw_real tempr, tempi;
+ fftw_real fac;
+
+#ifdef USE_TWIDDLE_TABLE
+ /* use twiddle factor tables */
+ real_t *twiddlers = mdct->twiddlers;
+#else
+ /* temps for pre and post twiddle */
+ real_t cosfreq8, sinfreq8, c, s, cold, cfreq, sfreq;
+#endif
uint16_t i;
- uint16_t Nd2 = 256 >> 1;
- uint16_t Nd4 = 256 >> 2;
- uint16_t Nd8 = 256 >> 3;
/* Choosing to allocate 2/N factor to Inverse Xform! */
fac = 0.0078125f; /* remaining 2/N from 4/N IFFT factor */
+#ifndef USE_TWIDDLE_TABLE
/* prepare for recurrence relation in pre-twiddle */
cfreq = 0.99969881772994995f;
sfreq = 0.024541229009628296f;
@@ -283,15 +367,21 @@
c = cosfreq8;
s = sinfreq8;
+#endif
- for (i = 0; i < Nd4; i++)
+ for (i = 0; i < 64; i++)
{- uint16_t unscr = unscrambled[i];
+ uint16_t n = i << 1;
+ uint16_t unscr = mdct->unscrambled[i];
- tempr = -in_data[2 * i];
- tempi = in_data[Nd2 - 1 - 2 * i];
+ tempr = -in_data[n];
+ tempi = in_data[127 - n];
/* calculate pre-twiddled FFT input */
+#ifdef USE_TWIDDLE_TABLE
+ FFTarray[unscr].re = tempr * twiddlers[n] - tempi * twiddlers[n + 1];
+ FFTarray[unscr].im = tempi * twiddlers[n] + tempr * twiddlers[n + 1];
+#else
FFTarray[unscr].re = tempr * c - tempi * s;
FFTarray[unscr].im = tempi * c + tempr * s;
@@ -299,39 +389,52 @@
cold = c;
c = c * cfreq - s * sfreq;
s = s * cfreq + cold * sfreq;
+#endif
}
/* Perform in-place complex IFFT of length N/4 */
pfftwi_64(FFTarray);
+#ifndef USE_TWIDDLE_TABLE
/* prepare for recurrence relations in post-twiddle */
c = cosfreq8;
s = sinfreq8;
+#endif
/* post-twiddle FFT output and then get output data */
- for (i = 0; i < Nd4; i++)
+ for (i = 0; i < 64; i++)
{+ uint16_t n = i << 1;
+
+#ifdef USE_TWIDDLE_TABLE
/* get post-twiddled FFT output */
+ tempr = fac * (FFTarray[i].re * twiddlers[n] - FFTarray[i].im * twiddlers[n + 1]);
+ tempi = fac * (FFTarray[i].im * twiddlers[n] + FFTarray[i].re * twiddlers[n + 1]);
+#else
+ /* get post-twiddled FFT output */
tempr = fac * (FFTarray[i].re * c - FFTarray[i].im * s);
tempi = fac * (FFTarray[i].im * c + FFTarray[i].re * s);
+#endif
/* fill in output values */
- out_data [Nd2 + Nd4 - 1 - 2 * i] = tempr;
- if (i < Nd8)
- out_data[Nd2 + Nd4 + 2 * i] = tempr;
+ out_data [191 - n] = tempr;
+ if (i < 32)
+ out_data[192 + n] = tempr;
else
- out_data[2 * i - Nd4] = -tempr;
+ out_data[n - 64] = -tempr;
- out_data [Nd4 + 2 * i] = tempi;
- if (i < Nd8)
- out_data[Nd4 - 1 - 2 * i] = -tempi;
+ out_data [64 + n] = tempi;
+ if (i < 32)
+ out_data[63 - n] = -tempi;
else
- out_data[Nd4 + 256 - 1 - 2*i] = tempi;
+ out_data[319 - n] = tempi;
+#ifndef USE_TWIDDLE_TABLE
/* use recurrence to prepare cosine and sine for next value of i */
cold = c;
c = c * cfreq - s * sfreq;
s = s * cfreq + cold * sfreq;
+#endif
}
}
@@ -711,7 +814,7 @@
{ -0.999698818696204, 0.0245412285229123 },};
-void PFFTW(16) (fftw_complex * input) {+static void PFFTW(16) (fftw_complex * input) {fftw_real tmp332;
fftw_real tmp331;
fftw_real tmp330;
@@ -1018,7 +1121,7 @@
c_im(input[14]) = st8;
}
-void PFFTW(32) (fftw_complex * input) {+static void PFFTW(32) (fftw_complex * input) {fftw_real tmp714;
fftw_real tmp713;
fftw_real tmp712;
@@ -1801,7 +1904,7 @@
c_im(input[1]) = st1;
}
-void PFFTW(64)(fftw_complex *input)
+static void PFFTW(64)(fftw_complex *input)
{PFFTW(twiddle_4)(input, PFFTW(W_64), 16);
PFFTW(16)(input );
@@ -1810,7 +1913,7 @@
PFFTW(16)(input + 48);
}
-void PFFTW(128)(fftw_complex *input)
+static void PFFTW(128)(fftw_complex *input)
{PFFTW(twiddle_4)(input, PFFTW(W_128), 32);
PFFTW(32)(input );
@@ -1819,7 +1922,7 @@
PFFTW(32)(input + 96);
}
-void PFFTW(512)(fftw_complex *input)
+static void PFFTW(512)(fftw_complex *input)
{PFFTW(twiddle_4)(input, PFFTW(W_512), 128);
PFFTW(128)(input );
@@ -1828,7 +1931,7 @@
PFFTW(128)(input + 384);
}
-void PFFTWI(16) (fftw_complex * input) {+static void PFFTWI(16) (fftw_complex * input) {fftw_real tmp333;
fftw_real tmp332;
fftw_real tmp331;
@@ -2137,7 +2240,7 @@
c_im(input[2]) = st1;
}
-void PFFTWI(32) (fftw_complex * input) {+static void PFFTWI(32) (fftw_complex * input) {fftw_real tmp714;
fftw_real tmp713;
fftw_real tmp712;
@@ -2920,7 +3023,7 @@
c_re(input[3]) = st1;
}
-void PFFTWI(64)(fftw_complex *input)
+static void PFFTWI(64)(fftw_complex *input)
{PFFTWI(16)(input );
PFFTWI(16)(input + 16);
@@ -2929,7 +3032,7 @@
PFFTWI(twiddle_4)(input, PFFTW(W_64), 16);
}
-void PFFTWI(128)(fftw_complex *input)
+static void PFFTWI(128)(fftw_complex *input)
{PFFTWI(32)(input );
PFFTWI(32)(input + 32);
@@ -2938,7 +3041,7 @@
PFFTWI(twiddle_4)(input, PFFTW(W_128), 32);
}
-void PFFTWI(512)(fftw_complex *input)
+static void PFFTWI(512)(fftw_complex *input)
{PFFTWI(128)(input );
PFFTWI(128)(input + 128);
@@ -2947,7 +3050,7 @@
PFFTWI(twiddle_4)(input, PFFTW(W_512), 128);
}
-void PFFTW(twiddle_4) (fftw_complex * A, const fftw_complex * W, uint16_t iostride) {+static void PFFTW(twiddle_4) (fftw_complex * A, const fftw_complex * W, uint16_t iostride) {uint16_t i;
fftw_complex *inout;
inout = A;
@@ -3060,7 +3163,7 @@
} while (i > 0);
}
-void PFFTWI(twiddle_4) (fftw_complex * A, const fftw_complex * W, uint16_t iostride) {+static void PFFTWI(twiddle_4) (fftw_complex * A, const fftw_complex * W, uint16_t iostride) {uint16_t i;
fftw_complex *inout;
inout = A;
@@ -3177,7 +3280,7 @@
} while (i > 0);
}
-uint16_t PFFTW(permutation_64)(uint16_t i)
+static uint16_t PFFTW(permutation_64)(uint16_t i)
{uint16_t i1 = i % 4;
uint16_t i2 = i / 4;
@@ -3187,7 +3290,7 @@
return (i1 * 16 + ((i2 + 1) % 16));
}
-uint16_t PFFTW(permutation_128)(uint16_t i)
+static uint16_t PFFTW(permutation_128)(uint16_t i)
{uint16_t i1 = i % 4;
uint16_t i2 = i / 4;
@@ -3197,7 +3300,7 @@
return (i1 * 32 + ((i2 + 1) % 32));
}
-uint16_t PFFTW(permutation_512)(uint16_t i)
+static uint16_t PFFTW(permutation_512)(uint16_t i)
{uint16_t i1 = i % 4;
uint16_t i2 = i / 4;
@@ -3207,13 +3310,19 @@
return (i1 * 128 + PFFTW(permutation_128)((i2 + 1) % 128));
}
-void make_fft_order(uint16_t *unscrambled64, uint16_t *unscrambled512)
+static void make_fft_order(uint16_t *unscrambled, uint16_t len)
{uint16_t i;
- for (i = 0; i < 64; i++)
- unscrambled64[i] = PFFTW(permutation_64)(i);
-
- for (i = 0; i < 512; i++)
- unscrambled512[i] = PFFTW(permutation_512)(i);
+ switch (len)
+ {+ case 64:
+ for (i = 0; i < len; i++)
+ unscrambled[i] = PFFTW(permutation_64)(i);
+ break;
+ case 512:
+ for (i = 0; i < len; i++)
+ unscrambled[i] = PFFTW(permutation_512)(i);
+ break;
+ }
}
--- a/libfaad/mdct.h
+++ b/libfaad/mdct.h
@@ -16,7 +16,7 @@
** along with this program; if not, write to the Free Software
** Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
**
-** $Id: mdct.h,v 1.2 2002/02/18 10:01:05 menno Exp $
+** $Id: mdct.h,v 1.3 2002/02/20 13:05:57 menno Exp $
**/
#ifndef __MDCT_H__
@@ -26,6 +26,12 @@
extern "C" {#endif
+typedef struct {+ uint16_t len;
+ real_t *twiddlers;
+ uint16_t *unscrambled;
+} mdct_info;
+
typedef real_t fftw_real;
typedef struct {@@ -46,13 +52,17 @@
DEFINE_PFFTW(128)
DEFINE_PFFTW(512)
-void make_fft_order(uint16_t *unscrambled64, uint16_t *unscrambled512);
-void IMDCT_long(fftw_real *in_data, fftw_real *out_data, uint16_t *unscrambled);
-void IMDCT_short(fftw_real *in_data, fftw_real *out_data, uint16_t *unscrambled);
+void mdct_init(mdct_info *mdct, uint16_t len);
+void mdct_end(mdct_info *mdct);
-void MDCT_long(fftw_real *in_data, fftw_real *out_data, uint16_t *unscrambled);
-void MDCT_short(fftw_real *in_data, fftw_real *out_data, uint16_t *unscrambled);
+void IMDCT_long(mdct_info *mdct, fftw_real *in_data, fftw_real *out_data);
+void IMDCT_short(mdct_info *mdct, fftw_real *in_data, fftw_real *out_data);
+void MDCT_long(mdct_info *mdct, fftw_real *in_data, fftw_real *out_data);
+void MDCT_short(mdct_info *mdct, fftw_real *in_data, fftw_real *out_data);
+
+static void make_fft_order(uint16_t *unscrambled, uint16_t len);
+
#define PFFTW(name) CONCAT(pfftw_, name)
#define PFFTWI(name) CONCAT(pfftwi_, name)
#define CONCAT_AUX(a, b) a ## b
@@ -59,8 +69,8 @@
#define CONCAT(a, b) CONCAT_AUX(a,b)
#define FFTW_KONST(x) ((fftw_real) x)
-void PFFTW(twiddle_4)(fftw_complex *A, const fftw_complex *W, uint16_t iostride);
-void PFFTWI(twiddle_4)(fftw_complex *A, const fftw_complex *W, uint16_t iostride);
+static void PFFTW(twiddle_4)(fftw_complex *A, const fftw_complex *W, uint16_t iostride);
+static void PFFTWI(twiddle_4)(fftw_complex *A, const fftw_complex *W, uint16_t iostride);
#ifdef __cplusplus
}
--- a/libfaad/output.c
+++ b/libfaad/output.c
@@ -16,16 +16,11 @@
** along with this program; if not, write to the Free Software
** Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
**
-** $Id: output.c,v 1.3 2002/02/18 10:01:05 menno Exp $
+** $Id: output.c,v 1.4 2002/02/20 13:05:57 menno Exp $
**/
#include "common.h"
-#ifdef USE_FMATH
-#include <mathf.h>
-#else
-#include <math.h>
-#endif
#include "output.h"
#include "decoder.h"
@@ -32,11 +27,7 @@
#define ftol(A,B) {tmp = *(int32_t*) & A - 0x4B7F8000; \B = (int16_t)((tmp==(int16_t)tmp) ? tmp : (tmp>>31)^0x7FFF);}
-#ifdef USE_FMATH
-#define ROUND(x) ((int32_t)floorf((x) + 0.5f))
-#else
#define ROUND(x) ((int32_t)floor((x) + 0.5))
-#endif
#define HAVE_IEEE754_FLOAT
#ifdef HAVE_IEEE754_FLOAT
--- a/libfaad/pns.c
+++ b/libfaad/pns.c
@@ -16,16 +16,11 @@
** along with this program; if not, write to the Free Software
** Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
**
-** $Id: pns.c,v 1.3 2002/02/18 10:01:05 menno Exp $
+** $Id: pns.c,v 1.4 2002/02/20 13:05:57 menno Exp $
**/
#include "common.h"
-#ifdef USE_FMATH
-#include <mathf.h>
-#else
-#include <math.h>
-#endif
#include "pns.h"
@@ -73,13 +68,8 @@
/* 14496-3 says:
scale = 1.0f/(size * (real_t)sqrt(MEAN_NRG));
*/
-#ifdef USE_FMATH
- scale = 1.0f/sqrtf(size * MEAN_NRG);
- scale *= powf(2.0f, 0.25f*scale_factor);
-#else
scale = 1.0f/(real_t)sqrt(size * MEAN_NRG);
scale *= (real_t)pow(2.0, 0.25*scale_factor);
-#endif
/* Scale random vector to desired target energy */
for (i = 0; i < size; i++)
--- a/libfaad/specrec.c
+++ b/libfaad/specrec.c
@@ -16,7 +16,7 @@
** along with this program; if not, write to the Free Software
** Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
**
-** $Id: specrec.c,v 1.3 2002/02/18 10:01:05 menno Exp $
+** $Id: specrec.c,v 1.4 2002/02/20 13:05:57 menno Exp $
**/
/*
@@ -28,11 +28,6 @@
#include "common.h"
-#ifdef USE_FMATH
-#include <mathf.h>
-#else
-#include <math.h>
-#endif
#include "specrec.h"
#include "syntax.h"
#include "data.h"
@@ -220,21 +215,13 @@
/* build pow() table for inverse quantization */
for(i = 0; i < IQ_TABLE_SIZE; i++)
{-#ifdef USE_FMATH
- iq_table[i] = powf(i, 4.0f/3.0f);
-#else
iq_table[i] = (real_t)pow(i, 4.0/3.0);
-#endif
}
/* build pow(2, 0.25) table for scalefactors */
for(i = 0; i < POW_TABLE_SIZE; i++)
{-#ifdef USE_FMATH
- pow2_table[i] = powf(2.0f, 0.25f * (i-100));
-#else
pow2_table[i] = (real_t)pow(2.0, 0.25 * (i-100));
-#endif
}
}
@@ -245,21 +232,13 @@
if (q < IQ_TABLE_SIZE)
return iq_table[q];
else
-#ifdef USE_FMATH
- return powf(q, 4.0f/3.0f);
-#else
return (real_t)pow(q, 4.0/3.0);
-#endif
} else if (q < 0) {q = -q;
if (q < IQ_TABLE_SIZE)
return -iq_table[q];
else
-#ifdef USE_FMATH
- return -powf(q, 4.0f/3.0f);
-#else
return -(real_t)pow(q, 4.0/3.0);
-#endif
} else {return 0.0f;
}
@@ -289,11 +268,7 @@
if (scale_factor < POW_TABLE_SIZE)
return pow2_table[scale_factor];
else
-#ifdef USE_FMATH
- return powf(2.0f, 0.25f * (scale_factor - 100));
-#else
return (real_t)pow(2.0, 0.25 * (scale_factor - 100));
-#endif
}
void apply_scalefactors(ic_stream *ics, real_t *x_invquant, real_t *pow2_table)
--- a/libfaad/tns.c
+++ b/libfaad/tns.c
@@ -16,16 +16,11 @@
** along with this program; if not, write to the Free Software
** Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
**
-** $Id: tns.c,v 1.2 2002/02/18 10:01:05 menno Exp $
+** $Id: tns.c,v 1.3 2002/02/20 13:05:57 menno Exp $
**/
#include "common.h"
-#ifdef USE_FMATH
-#include <mathf.h>
-#else
-#include <math.h>
-#endif
#include "syntax.h"
#include "tns.h"
@@ -154,11 +149,7 @@
iqfac_m = ((1 << (coef_res_bits-1)) + 0.5f) / (M_PI/2.0f);
for (i = 0; i < order; i++)
-#ifdef USE_FMATH
- tmp2[i] = sinf(tmp[i] / ((tmp[i] >= 0) ? iqfac : iqfac_m));
-#else
tmp2[i] = (real_t)sin(tmp[i] / ((tmp[i] >= 0) ? iqfac : iqfac_m));
-#endif
/* Conversion to LPC coefficients */
a[0] = 1;