FFmpeg/libavcodec/ac3enc_fixed.c
Justin Ruggles ac05f9030e Convert floating-point MDCT coefficients to 24-bit fixed-point all at once
instead of doing it separately in 2 different functions.
This makes float AC-3 encoding approx. 3-7% faster overall.
Also, the coefficient conversion can now be easily SIMD-optimized.

Originally committed as revision 26232 to svn://svn.ffmpeg.org/ffmpeg/trunk
2011-01-05 20:35:36 +00:00

441 lines
11 KiB
C

/*
* The simplest AC-3 encoder
* Copyright (c) 2000 Fabrice Bellard
* Copyright (c) 2006-2010 Justin Ruggles <justin.ruggles@gmail.com>
* Copyright (c) 2006-2010 Prakash Punnoor <prakash@punnoor.de>
*
* This file is part of FFmpeg.
*
* FFmpeg is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* FFmpeg 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
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with FFmpeg; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
/**
* @file
* fixed-point AC-3 encoder.
*/
#undef CONFIG_AC3ENC_FLOAT
#include "ac3enc.c"
/** Scale a float value by 2^15, convert to an integer, and clip to range -32767..32767. */
#define FIX15(a) av_clip(SCALE_FLOAT(a, 15), -32767, 32767)
/**
* Finalize MDCT and free allocated memory.
*/
static av_cold void mdct_end(AC3MDCTContext *mdct)
{
mdct->nbits = 0;
av_freep(&mdct->costab);
av_freep(&mdct->sintab);
av_freep(&mdct->xcos1);
av_freep(&mdct->xsin1);
av_freep(&mdct->rot_tmp);
av_freep(&mdct->cplx_tmp);
}
/**
* Initialize FFT tables.
* @param ln log2(FFT size)
*/
static av_cold int fft_init(AVCodecContext *avctx, AC3MDCTContext *mdct, int ln)
{
int i, n, n2;
float alpha;
n = 1 << ln;
n2 = n >> 1;
FF_ALLOC_OR_GOTO(avctx, mdct->costab, n2 * sizeof(*mdct->costab), fft_alloc_fail);
FF_ALLOC_OR_GOTO(avctx, mdct->sintab, n2 * sizeof(*mdct->sintab), fft_alloc_fail);
for (i = 0; i < n2; i++) {
alpha = 2.0 * M_PI * i / n;
mdct->costab[i] = FIX15(cos(alpha));
mdct->sintab[i] = FIX15(sin(alpha));
}
return 0;
fft_alloc_fail:
mdct_end(mdct);
return AVERROR(ENOMEM);
}
/**
* Initialize MDCT tables.
* @param nbits log2(MDCT size)
*/
static av_cold int mdct_init(AVCodecContext *avctx, AC3MDCTContext *mdct,
int nbits)
{
int i, n, n4, ret;
n = 1 << nbits;
n4 = n >> 2;
mdct->nbits = nbits;
ret = fft_init(avctx, mdct, nbits - 2);
if (ret)
return ret;
mdct->window = ff_ac3_window;
FF_ALLOC_OR_GOTO(avctx, mdct->xcos1, n4 * sizeof(*mdct->xcos1), mdct_alloc_fail);
FF_ALLOC_OR_GOTO(avctx, mdct->xsin1, n4 * sizeof(*mdct->xsin1), mdct_alloc_fail);
FF_ALLOC_OR_GOTO(avctx, mdct->rot_tmp, n * sizeof(*mdct->rot_tmp), mdct_alloc_fail);
FF_ALLOC_OR_GOTO(avctx, mdct->cplx_tmp, n4 * sizeof(*mdct->cplx_tmp), mdct_alloc_fail);
for (i = 0; i < n4; i++) {
float alpha = 2.0 * M_PI * (i + 1.0 / 8.0) / n;
mdct->xcos1[i] = FIX15(-cos(alpha));
mdct->xsin1[i] = FIX15(-sin(alpha));
}
return 0;
mdct_alloc_fail:
mdct_end(mdct);
return AVERROR(ENOMEM);
}
/** Butterfly op */
#define BF(pre, pim, qre, qim, pre1, pim1, qre1, qim1) \
{ \
int ax, ay, bx, by; \
bx = pre1; \
by = pim1; \
ax = qre1; \
ay = qim1; \
pre = (bx + ax) >> 1; \
pim = (by + ay) >> 1; \
qre = (bx - ax) >> 1; \
qim = (by - ay) >> 1; \
}
/** Complex multiply */
#define CMUL(pre, pim, are, aim, bre, bim) \
{ \
pre = (MUL16(are, bre) - MUL16(aim, bim)) >> 15; \
pim = (MUL16(are, bim) + MUL16(bre, aim)) >> 15; \
}
/**
* Calculate a 2^n point complex FFT on 2^ln points.
* @param z complex input/output samples
* @param ln log2(FFT size)
*/
static void fft(AC3MDCTContext *mdct, IComplex *z, int ln)
{
int j, l, np, np2;
int nblocks, nloops;
register IComplex *p,*q;
int tmp_re, tmp_im;
np = 1 << ln;
/* reverse */
for (j = 0; j < np; j++) {
int k = av_reverse[j] >> (8 - ln);
if (k < j)
FFSWAP(IComplex, z[k], z[j]);
}
/* pass 0 */
p = &z[0];
j = np >> 1;
do {
BF(p[0].re, p[0].im, p[1].re, p[1].im,
p[0].re, p[0].im, p[1].re, p[1].im);
p += 2;
} while (--j);
/* pass 1 */
p = &z[0];
j = np >> 2;
do {
BF(p[0].re, p[0].im, p[2].re, p[2].im,
p[0].re, p[0].im, p[2].re, p[2].im);
BF(p[1].re, p[1].im, p[3].re, p[3].im,
p[1].re, p[1].im, p[3].im, -p[3].re);
p+=4;
} while (--j);
/* pass 2 .. ln-1 */
nblocks = np >> 3;
nloops = 1 << 2;
np2 = np >> 1;
do {
p = z;
q = z + nloops;
for (j = 0; j < nblocks; j++) {
BF(p->re, p->im, q->re, q->im,
p->re, p->im, q->re, q->im);
p++;
q++;
for(l = nblocks; l < np2; l += nblocks) {
CMUL(tmp_re, tmp_im, mdct->costab[l], -mdct->sintab[l], q->re, q->im);
BF(p->re, p->im, q->re, q->im,
p->re, p->im, tmp_re, tmp_im);
p++;
q++;
}
p += nloops;
q += nloops;
}
nblocks = nblocks >> 1;
nloops = nloops << 1;
} while (nblocks);
}
/**
* Calculate a 512-point MDCT
* @param out 256 output frequency coefficients
* @param in 512 windowed input audio samples
*/
static void mdct512(AC3MDCTContext *mdct, int32_t *out, int16_t *in)
{
int i, re, im, n, n2, n4;
int16_t *rot = mdct->rot_tmp;
IComplex *x = mdct->cplx_tmp;
n = 1 << mdct->nbits;
n2 = n >> 1;
n4 = n >> 2;
/* shift to simplify computations */
for (i = 0; i <n4; i++)
rot[i] = -in[i + 3*n4];
memcpy(&rot[n4], &in[0], 3*n4*sizeof(*in));
/* pre rotation */
for (i = 0; i < n4; i++) {
re = ((int)rot[ 2*i] - (int)rot[ n-1-2*i]) >> 1;
im = -((int)rot[n2+2*i] - (int)rot[n2-1-2*i]) >> 1;
CMUL(x[i].re, x[i].im, re, im, -mdct->xcos1[i], mdct->xsin1[i]);
}
fft(mdct, x, mdct->nbits - 2);
/* post rotation */
for (i = 0; i < n4; i++) {
re = x[i].re;
im = x[i].im;
CMUL(out[n2-1-2*i], out[2*i], re, im, mdct->xsin1[i], mdct->xcos1[i]);
}
}
/**
* Apply KBD window to input samples prior to MDCT.
*/
static void apply_window(int16_t *output, const int16_t *input,
const int16_t *window, int n)
{
int i;
int n2 = n >> 1;
for (i = 0; i < n2; i++) {
output[i] = MUL16(input[i], window[i]) >> 15;
output[n-i-1] = MUL16(input[n-i-1], window[i]) >> 15;
}
}
/**
* Calculate the log2() of the maximum absolute value in an array.
* @param tab input array
* @param n number of values in the array
* @return log2(max(abs(tab[])))
*/
static int log2_tab(int16_t *tab, int n)
{
int i, v;
v = 0;
for (i = 0; i < n; i++)
v |= abs(tab[i]);
return av_log2(v);
}
/**
* Left-shift each value in an array by a specified amount.
* @param tab input array
* @param n number of values in the array
* @param lshift left shift amount. a negative value means right shift.
*/
static void lshift_tab(int16_t *tab, int n, int lshift)
{
int i;
if (lshift > 0) {
for (i = 0; i < n; i++)
tab[i] <<= lshift;
} else if (lshift < 0) {
lshift = -lshift;
for (i = 0; i < n; i++)
tab[i] >>= lshift;
}
}
/**
* Normalize the input samples to use the maximum available precision.
* This assumes signed 16-bit input samples. Exponents are reduced by 9 to
* match the 24-bit internal precision for MDCT coefficients.
*
* @return exponent shift
*/
static int normalize_samples(AC3EncodeContext *s)
{
int v = 14 - log2_tab(s->windowed_samples, AC3_WINDOW_SIZE);
v = FFMAX(0, v);
lshift_tab(s->windowed_samples, AC3_WINDOW_SIZE, v);
return v - 9;
}
/**
* Scale MDCT coefficients from float to fixed-point.
*/
static void scale_coefficients(AC3EncodeContext *s)
{
/* scaling/conversion is obviously not needed for the fixed-point encoder
since the coefficients are already fixed-point. */
return;
}
#ifdef TEST
/*************************************************************************/
/* TEST */
#include "libavutil/lfg.h"
#define MDCT_NBITS 9
#define MDCT_SAMPLES (1 << MDCT_NBITS)
#define FN (MDCT_SAMPLES/4)
static void fft_test(AC3MDCTContext *mdct, AVLFG *lfg)
{
IComplex in[FN], in1[FN];
int k, n, i;
float sum_re, sum_im, a;
for (i = 0; i < FN; i++) {
in[i].re = av_lfg_get(lfg) % 65535 - 32767;
in[i].im = av_lfg_get(lfg) % 65535 - 32767;
in1[i] = in[i];
}
fft(mdct, in, 7);
/* do it by hand */
for (k = 0; k < FN; k++) {
sum_re = 0;
sum_im = 0;
for (n = 0; n < FN; n++) {
a = -2 * M_PI * (n * k) / FN;
sum_re += in1[n].re * cos(a) - in1[n].im * sin(a);
sum_im += in1[n].re * sin(a) + in1[n].im * cos(a);
}
av_log(NULL, AV_LOG_DEBUG, "%3d: %6d,%6d %6.0f,%6.0f\n",
k, in[k].re, in[k].im, sum_re / FN, sum_im / FN);
}
}
static void mdct_test(AC3MDCTContext *mdct, AVLFG *lfg)
{
int16_t input[MDCT_SAMPLES];
int32_t output[AC3_MAX_COEFS];
float input1[MDCT_SAMPLES];
float output1[AC3_MAX_COEFS];
float s, a, err, e, emax;
int i, k, n;
for (i = 0; i < MDCT_SAMPLES; i++) {
input[i] = (av_lfg_get(lfg) % 65535 - 32767) * 9 / 10;
input1[i] = input[i];
}
mdct512(mdct, output, input);
/* do it by hand */
for (k = 0; k < AC3_MAX_COEFS; k++) {
s = 0;
for (n = 0; n < MDCT_SAMPLES; n++) {
a = (2*M_PI*(2*n+1+MDCT_SAMPLES/2)*(2*k+1) / (4 * MDCT_SAMPLES));
s += input1[n] * cos(a);
}
output1[k] = -2 * s / MDCT_SAMPLES;
}
err = 0;
emax = 0;
for (i = 0; i < AC3_MAX_COEFS; i++) {
av_log(NULL, AV_LOG_DEBUG, "%3d: %7d %7.0f\n", i, output[i], output1[i]);
e = output[i] - output1[i];
if (e > emax)
emax = e;
err += e * e;
}
av_log(NULL, AV_LOG_DEBUG, "err2=%f emax=%f\n", err / AC3_MAX_COEFS, emax);
}
int main(void)
{
AVLFG lfg;
AC3MDCTContext mdct;
mdct.avctx = NULL;
av_log_set_level(AV_LOG_DEBUG);
mdct_init(&mdct, 9);
fft_test(&mdct, &lfg);
mdct_test(&mdct, &lfg);
return 0;
}
#endif /* TEST */
AVCodec ac3_fixed_encoder = {
"ac3_fixed",
AVMEDIA_TYPE_AUDIO,
CODEC_ID_AC3,
sizeof(AC3EncodeContext),
ac3_encode_init,
ac3_encode_frame,
ac3_encode_close,
NULL,
.sample_fmts = (const enum AVSampleFormat[]){AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE},
.long_name = NULL_IF_CONFIG_SMALL("ATSC A/52A (AC-3)"),
.channel_layouts = ac3_channel_layouts,
};