Nyaaori/rockbox
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity
rockbox/lib/rbcodec/codecs/libopus/celt/mdct.c
Andree Buschmann 0ae8d16643 opus: replace copy loops with calls to memcpy 
speeds up decoding of a 64kbps test file by 0.5Hz on h300 (cf)
0.9MHz on c200 (pp) and 0.2MHz on fuzev1 (amsv1)

Change-Id: Ib537c2393fa6dca0b61e4e9f80eef5e688c2c2bd
Signed-off-by: Nils Wallménius <nils@rockbox.org>
2012-10-07 12:34:54 +02:00

345 lines
11 KiB
C
Raw Blame History

/* Copyright (c) 2007-2008 CSIRO
Copyright (c) 2007-2008 Xiph.Org Foundation
Written by Jean-Marc Valin */
/*
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions
are met:
- Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer.
- Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the following disclaimer in the
documentation and/or other materials provided with the distribution.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
/* This is a simple MDCT implementation that uses a N/4 complex FFT
to do most of the work. It should be relatively straightforward to
plug in pretty much and FFT here.
This replaces the Vorbis FFT (and uses the exact same API), which
was a bit too messy and that was ending up duplicating code
(might as well use the same FFT everywhere).
The algorithm is similar to (and inspired from) Fabrice Bellard's
MDCT implementation in FFMPEG, but has differences in signs, ordering
and scaling in many places.
*/
#ifndef SKIP_CONFIG_H
#ifdef HAVE_CONFIG_H
#include "opus_config.h"
#endif
#endif
#include "mdct.h"
#include "kiss_fft.h"
#include "_kiss_fft_guts.h"
#include <math.h>
#include "os_support.h"
#include "mathops.h"
#include "stack_alloc.h"
#ifdef CUSTOM_MODES
int clt_mdct_init(mdct_lookup *l,int N, int maxshift)
{
int i;
int N4;
kiss_twiddle_scalar *trig;
#if defined(FIXED_POINT)
int N2=N>>1;
#endif
l->n = N;
N4 = N>>2;
l->maxshift = maxshift;
for (i=0;i<=maxshift;i++)
{
if (i==0)
l->kfft[i] = opus_fft_alloc(N>>2>>i, 0, 0);
else
l->kfft[i] = opus_fft_alloc_twiddles(N>>2>>i, 0, 0, l->kfft[0]);
#ifndef ENABLE_TI_DSPLIB55
if (l->kfft[i]==NULL)
return 0;
#endif
}
l->trig = trig = (kiss_twiddle_scalar*)opus_alloc((N4+1)*sizeof(kiss_twiddle_scalar));
if (l->trig==NULL)
return 0;
/* We have enough points that sine isn't necessary */
#if defined(FIXED_POINT)
for (i=0;i<=N4;i++)
trig[i] = TRIG_UPSCALE*celt_cos_norm(DIV32(ADD32(SHL32(EXTEND32(i),17),N2),N));
#else
for (i=0;i<=N4;i++)
trig[i] = (kiss_twiddle_scalar)cos(2*PI*i/N);
#endif
return 1;
}
void clt_mdct_clear(mdct_lookup *l)
{
int i;
for (i=0;i<=l->maxshift;i++)
opus_fft_free(l->kfft[i]);
opus_free((kiss_twiddle_scalar*)l->trig);
}
#endif /* CUSTOM_MODES */
#if 0
/* Forward MDCT trashes the input array */
void clt_mdct_forward(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scalar * OPUS_RESTRICT out,
const opus_val16 *window, int overlap, int shift, int stride)
{
int i;
int N, N2, N4;
kiss_twiddle_scalar sine;
VARDECL(kiss_fft_scalar, f);
SAVE_STACK;
N = l->n;
N >>= shift;
N2 = N>>1;
N4 = N>>2;
ALLOC(f, N2, kiss_fft_scalar);
/* sin(x) ~= x here */
#ifdef FIXED_POINT
sine = TRIG_UPSCALE*(QCONST16(0.7853981f, 15)+N2)/N;
#else
sine = (kiss_twiddle_scalar)2*PI*(.125f)/N;
#endif
/* Consider the input to be composed of four blocks: [a, b, c, d] */
/* Window, shuffle, fold */
{
/* Temp pointers to make it really clear to the compiler what we're doing */
const kiss_fft_scalar * OPUS_RESTRICT xp1 = in+(overlap>>1);
const kiss_fft_scalar * OPUS_RESTRICT xp2 = in+N2-1+(overlap>>1);
kiss_fft_scalar * OPUS_RESTRICT yp = f;
const opus_val16 * OPUS_RESTRICT wp1 = window+(overlap>>1);
const opus_val16 * OPUS_RESTRICT wp2 = window+(overlap>>1)-1;
for(i=0;i<(overlap>>2);i++)
{
/* Real part arranged as -d-cR, Imag part arranged as -b+aR*/
*yp++ = MULT16_32_Q15(*wp2, xp1[N2]) + MULT16_32_Q15(*wp1,*xp2);
*yp++ = MULT16_32_Q15(*wp1, *xp1) - MULT16_32_Q15(*wp2, xp2[-N2]);
xp1+=2;
xp2-=2;
wp1+=2;
wp2-=2;
}
wp1 = window;
wp2 = window+overlap-1;
for(;i<N4-(overlap>>2);i++)
{
/* Real part arranged as a-bR, Imag part arranged as -c-dR */
*yp++ = *xp2;
*yp++ = *xp1;
xp1+=2;
xp2-=2;
}
for(;i<N4;i++)
{
/* Real part arranged as a-bR, Imag part arranged as -c-dR */
*yp++ = -MULT16_32_Q15(*wp1, xp1[-N2]) + MULT16_32_Q15(*wp2, *xp2);
*yp++ = MULT16_32_Q15(*wp2, *xp1) + MULT16_32_Q15(*wp1, xp2[N2]);
xp1+=2;
xp2-=2;
wp1+=2;
wp2-=2;
}
}
/* Pre-rotation */
{
kiss_fft_scalar * OPUS_RESTRICT yp = f;
const kiss_twiddle_scalar *t = &l->trig[0];
for(i=0;i<N4;i++)
{
kiss_fft_scalar re, im, yr, yi;
re = yp[0];
im = yp[1];
yr = -S_MUL(re,t[i<<shift]) - S_MUL(im,t[(N4-i)<<shift]);
yi = -S_MUL(im,t[i<<shift]) + S_MUL(re,t[(N4-i)<<shift]);
/* works because the cos is nearly one */
*yp++ = yr + S_MUL(yi,sine);
*yp++ = yi - S_MUL(yr,sine);
}
}
/* N/4 complex FFT, down-scales by 4/N */
opus_fft(l->kfft[shift], (kiss_fft_cpx *)f, (kiss_fft_cpx *)in);
/* Post-rotate */
{
/* Temp pointers to make it really clear to the compiler what we're doing */
const kiss_fft_scalar * OPUS_RESTRICT fp = in;
kiss_fft_scalar * OPUS_RESTRICT yp1 = out;
kiss_fft_scalar * OPUS_RESTRICT yp2 = out+stride*(N2-1);
const kiss_twiddle_scalar *t = &l->trig[0];
/* Temp pointers to make it really clear to the compiler what we're doing */
for(i=0;i<N4;i++)
{
kiss_fft_scalar yr, yi;
yr = S_MUL(fp[1],t[(N4-i)<<shift]) + S_MUL(fp[0],t[i<<shift]);
yi = S_MUL(fp[0],t[(N4-i)<<shift]) - S_MUL(fp[1],t[i<<shift]);
/* works because the cos is nearly one */
*yp1 = yr - S_MUL(yi,sine);
*yp2 = yi + S_MUL(yr,sine);;
fp += 2;
yp1 += 2*stride;
yp2 -= 2*stride;
}
}
RESTORE_STACK;
}
#endif
#define S_F_BUF_SIZE (1920>>1) /* N = 1920 for static modes */
static kiss_fft_scalar s_f2[S_F_BUF_SIZE] IBSS_ATTR MEM_ALIGN_ATTR;
void clt_mdct_backward(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scalar * OPUS_RESTRICT out,
const opus_val16 * OPUS_RESTRICT window, int overlap, int shift, int stride)
{
int i;
int N, N2, N4;
kiss_twiddle_scalar sine;
VARDECL(kiss_fft_scalar, f);
VARDECL(kiss_fft_scalar, f2);
SAVE_STACK;
N = l->n;
N >>= shift;
N2 = N>>1;
N4 = N>>2;
kiss_fft_scalar s_f[S_F_BUF_SIZE];
if (S_F_BUF_SIZE >= N2)
{
f = s_f;
f2 = s_f2;
}
else
{
ALLOC(f , N2, kiss_fft_scalar);
ALLOC(f2, N2, kiss_fft_scalar);
}
/* sin(x) ~= x here */
#ifdef FIXED_POINT
sine = TRIG_UPSCALE*(QCONST16(0.7853981f, 15)+N2)/N;
#else
sine = (kiss_twiddle_scalar)2*PI*(.125f)/N;
#endif
/* Pre-rotate */
{
/* Temp pointers to make it really clear to the compiler what we're doing */
const kiss_fft_scalar * OPUS_RESTRICT xp1 = in;
const kiss_fft_scalar * OPUS_RESTRICT xp2 = in+stride*(N2-1);
kiss_fft_scalar * OPUS_RESTRICT yp = f2;
const kiss_twiddle_scalar *t = &l->trig[0];
for(i=0;i<N4;i++)
{
kiss_fft_scalar yr, yi;
yr = -S_MUL(*xp2, t[i<<shift]) + S_MUL(*xp1,t[(N4-i)<<shift]);
yi = -S_MUL(*xp2, t[(N4-i)<<shift]) - S_MUL(*xp1,t[i<<shift]);
/* works because the cos is nearly one */
*yp++ = yr - S_MUL(yi,sine);
*yp++ = yi + S_MUL(yr,sine);
xp1+=2*stride;
xp2-=2*stride;
}
}
/* Inverse N/4 complex FFT. This one should *not* downscale even in fixed-point */
opus_ifft(l->kfft[shift], (kiss_fft_cpx *)f2, (kiss_fft_cpx *)f);
/* Post-rotate */
{
kiss_fft_scalar * OPUS_RESTRICT fp = f;
const kiss_twiddle_scalar *t = &l->trig[0];
for(i=0;i<N4;i++)
{
kiss_fft_scalar re, im, yr, yi;
re = fp[0];
im = fp[1];
/* We'd scale up by 2 here, but instead it's done when mixing the windows */
yr = S_MUL(re,t[i<<shift]) - S_MUL(im,t[(N4-i)<<shift]);
yi = S_MUL(im,t[i<<shift]) + S_MUL(re,t[(N4-i)<<shift]);
/* works because the cos is nearly one */
*fp++ = yr - S_MUL(yi,sine);
*fp++ = yi + S_MUL(yr,sine);
}
}
/* De-shuffle the components for the middle of the window only */
{
const kiss_fft_scalar * OPUS_RESTRICT fp1 = f;
const kiss_fft_scalar * OPUS_RESTRICT fp2 = f+N2-1;
kiss_fft_scalar * OPUS_RESTRICT yp = f2;
for(i = 0; i < N4; i++)
{
*yp++ =-*fp1;
*yp++ = *fp2;
fp1 += 2;
fp2 -= 2;
}
}
out -= (N2-overlap)>>1;
/* Mirror on both sides for TDAC */
{
kiss_fft_scalar * OPUS_RESTRICT fp1 = f2+N4-1;
kiss_fft_scalar * OPUS_RESTRICT xp1 = out+N2-1;
kiss_fft_scalar * OPUS_RESTRICT yp1 = out+N4-overlap/2;
const opus_val16 * OPUS_RESTRICT wp1 = window;
const opus_val16 * OPUS_RESTRICT wp2 = window+overlap-1;
i = N4-overlap/2;
xp1 -= N4-overlap/2;
fp1 -= N4-overlap/2;
OPUS_COPY(xp1+1, fp1+1, N4-overlap/2);
for(; i < N4; i++)
{
kiss_fft_scalar x1;
x1 = *fp1--;
*yp1++ +=-MULT16_32_Q15(*wp1, x1);
*xp1-- += MULT16_32_Q15(*wp2, x1);
wp1++;
wp2--;
}
}
{
kiss_fft_scalar * OPUS_RESTRICT fp2 = f2+N4;
kiss_fft_scalar * OPUS_RESTRICT xp2 = out+N2;
kiss_fft_scalar * OPUS_RESTRICT yp2 = out+N-1-(N4-overlap/2);
const opus_val16 * OPUS_RESTRICT wp1 = window;
const opus_val16 * OPUS_RESTRICT wp2 = window+overlap-1;
i = N4-overlap/2;
OPUS_COPY(xp2, fp2, N4-overlap/2);
xp2 += N4-overlap/2;
fp2 += N4-overlap/2;
for(; i < N4; i++)
{
kiss_fft_scalar x2;
x2 = *fp2++;
*yp2-- = MULT16_32_Q15(*wp1, x2);
*xp2++ = MULT16_32_Q15(*wp2, x2);
wp1++;
wp2--;
}
}
RESTORE_STACK;
}
Reference in a new issue View git blame Copy permalink