mirror of
https://github.com/FFmpeg/FFmpeg.git
synced 2024-12-23 12:43:46 +02:00
ac3enc: use generic fixed-point mdct
This makes the AC3 encoder use the shared fixed-point MDCT rather than its own implementation. The checksum changes are due to different rounding in the MDCT. Signed-off-by: Mans Rullgard <mans@mansr.com>
This commit is contained in:
Mans Rullgard
parent
aadfc9ee74
commit
79997def65
@ -29,6 +29,8 @@
|
||||
//#define DEBUG
|
||||
//#define ASSERT_LEVEL 2
|
||||
|
||||
#include <stdint.h>
|
||||
|
||||
#include "libavutil/audioconvert.h"
|
||||
#include "libavutil/avassert.h"
|
||||
#include "libavutil/crc.h"
|
||||
@ -39,6 +41,7 @@
|
||||
#include "ac3dsp.h"
|
||||
#include "ac3.h"
|
||||
#include "audioconvert.h"
|
||||
#include "fft.h"
|
||||
|
||||
|
||||
#ifndef CONFIG_AC3ENC_FLOAT
|
||||
@ -55,16 +58,22 @@
|
||||
#define AC3_REMATRIXING_NONE 1
|
||||
#define AC3_REMATRIXING_ALWAYS 3
|
||||
|
||||
/** Scale a float value by 2^bits and convert to an integer. */
|
||||
#define SCALE_FLOAT(a, bits) lrintf((a) * (float)(1 << (bits)))
|
||||
|
||||
|
||||
#if CONFIG_AC3ENC_FLOAT
|
||||
#include "ac3enc_float.h"
|
||||
#define MAC_COEF(d,a,b) ((d)+=(a)*(b))
|
||||
typedef float SampleType;
|
||||
typedef float CoefType;
|
||||
typedef float CoefSumType;
|
||||
#else
|
||||
#include "ac3enc_fixed.h"
|
||||
#define MAC_COEF(d,a,b) MAC64(d,a,b)
|
||||
typedef int16_t SampleType;
|
||||
typedef int32_t CoefType;
|
||||
typedef int64_t CoefSumType;
|
||||
#endif
|
||||
|
||||
typedef struct AC3MDCTContext {
|
||||
const SampleType *window; ///< MDCT window function
|
||||
FFTContext fft; ///< FFT context for MDCT calculation
|
||||
} AC3MDCTContext;
|
||||
|
||||
/**
|
||||
* Encoding Options used by AVOption.
|
||||
@ -279,8 +288,6 @@ static av_cold void mdct_end(AC3MDCTContext *mdct);
|
||||
static av_cold int mdct_init(AVCodecContext *avctx, AC3MDCTContext *mdct,
|
||||
int nbits);
|
||||
|
||||
static void mdct512(AC3MDCTContext *mdct, CoefType *out, SampleType *in);
|
||||
|
||||
static void apply_window(DSPContext *dsp, SampleType *output, const SampleType *input,
|
||||
const SampleType *window, unsigned int len);
|
||||
|
||||
@ -386,7 +393,8 @@ static void apply_mdct(AC3EncodeContext *s)
|
||||
|
||||
block->coeff_shift[ch] = normalize_samples(s);
|
||||
|
||||
mdct512(&s->mdct, block->mdct_coef[ch], s->windowed_samples);
|
||||
s->mdct.fft.mdct_calcw(&s->mdct.fft, block->mdct_coef[ch],
|
||||
s->windowed_samples);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
@ -26,54 +26,17 @@
|
||||
* fixed-point AC-3 encoder.
|
||||
*/
|
||||
|
||||
#define CONFIG_FFT_FLOAT 0
|
||||
#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);
|
||||
ff_fft_end(&mdct->fft);
|
||||
}
|
||||
|
||||
|
||||
@ -84,167 +47,9 @@ fft_alloc_fail:
|
||||
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;
|
||||
|
||||
int ret = ff_mdct_init(&mdct->fft, nbits, 0, 1.0);
|
||||
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, rshift) \
|
||||
{ \
|
||||
pre = (MUL16(are, bre) - MUL16(aim, bim)) >> rshift; \
|
||||
pim = (MUL16(are, bim) + MUL16(bre, aim)) >> rshift; \
|
||||
}
|
||||
|
||||
|
||||
/**
|
||||
* 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, 15);
|
||||
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], 15);
|
||||
}
|
||||
|
||||
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], 0);
|
||||
}
|
||||
return ret;
|
||||
}
|
||||
|
||||
|
||||
@ -304,101 +109,6 @@ static void scale_coefficients(AC3EncodeContext *s)
|
||||
}
|
||||
|
||||
|
||||
#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 ff_ac3_fixed_encoder = {
|
||||
"ac3_fixed",
|
||||
AVMEDIA_TYPE_AUDIO,
|
||||
|
||||
@ -1,61 +0,0 @@
|
||||
/*
|
||||
* 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 Libav.
|
||||
*
|
||||
* Libav 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.
|
||||
*
|
||||
* Libav 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 Libav; 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 header.
|
||||
*/
|
||||
|
||||
#ifndef AVCODEC_AC3ENC_FIXED_H
|
||||
#define AVCODEC_AC3ENC_FIXED_H
|
||||
|
||||
#include <stdint.h>
|
||||
|
||||
|
||||
typedef int16_t SampleType;
|
||||
typedef int32_t CoefType;
|
||||
typedef int64_t CoefSumType;
|
||||
|
||||
#define MAC_COEF(d,a,b) MAC64(d,a,b)
|
||||
|
||||
|
||||
/**
|
||||
* Compex number.
|
||||
* Used in fixed-point MDCT calculation.
|
||||
*/
|
||||
typedef struct IComplex {
|
||||
int16_t re,im;
|
||||
} IComplex;
|
||||
|
||||
typedef struct AC3MDCTContext {
|
||||
const int16_t *window; ///< MDCT window function
|
||||
int nbits; ///< log2(transform size)
|
||||
int16_t *costab; ///< FFT cos table
|
||||
int16_t *sintab; ///< FFT sin table
|
||||
int16_t *xcos1; ///< MDCT cos table
|
||||
int16_t *xsin1; ///< MDCT sin table
|
||||
int16_t *rot_tmp; ///< temp buffer for pre-rotated samples
|
||||
IComplex *cplx_tmp; ///< temp buffer for complex pre-rotated samples
|
||||
} AC3MDCTContext;
|
||||
|
||||
#endif /* AVCODEC_AC3ENC_FIXED_H */
|
||||
@ -68,17 +68,6 @@ static av_cold int mdct_init(AVCodecContext *avctx, AC3MDCTContext *mdct,
|
||||
}
|
||||
|
||||
|
||||
/**
|
||||
* Calculate a 512-point MDCT
|
||||
* @param out 256 output frequency coefficients
|
||||
* @param in 512 windowed input audio samples
|
||||
*/
|
||||
static void mdct512(AC3MDCTContext *mdct, float *out, float *in)
|
||||
{
|
||||
mdct->fft.mdct_calc(&mdct->fft, out, in);
|
||||
}
|
||||
|
||||
|
||||
/**
|
||||
* Apply KBD window to input samples prior to MDCT.
|
||||
*/
|
||||
|
||||
@ -1,47 +0,0 @@
|
||||
/*
|
||||
* 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 Libav.
|
||||
*
|
||||
* Libav 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.
|
||||
*
|
||||
* Libav 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 Libav; if not, write to the Free Software
|
||||
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
||||
*/
|
||||
|
||||
/**
|
||||
* @file
|
||||
* floating-point AC-3 encoder header.
|
||||
*/
|
||||
|
||||
#ifndef AVCODEC_AC3ENC_FLOAT_H
|
||||
#define AVCODEC_AC3ENC_FLOAT_H
|
||||
|
||||
#include "fft.h"
|
||||
|
||||
|
||||
typedef float SampleType;
|
||||
typedef float CoefType;
|
||||
typedef float CoefSumType;
|
||||
|
||||
#define MAC_COEF(d,a,b) ((d)+=(a)*(b))
|
||||
|
||||
|
||||
typedef struct AC3MDCTContext {
|
||||
const float *window; ///< MDCT window function
|
||||
FFTContext fft; ///< FFT context for MDCT calculation
|
||||
} AC3MDCTContext;
|
||||
|
||||
#endif /* AVCODEC_AC3ENC_FLOAT_H */
|
||||
@ -1,2 +1,2 @@
|
||||
346073c97eada69330f61e103a170ca1 *./tests/data/acodec/ac3.rm
|
||||
408f47ee5a60866fc751f7bc2314cbd6 *./tests/data/acodec/ac3.rm
|
||||
98751 ./tests/data/acodec/ac3.rm
|
||||
|
||||
@ -1,2 +1,2 @@
|
||||
7b7ede9548a09346675edad36acfbf19 *./tests/data/lavf/lavf.rm
|
||||
da74fe749c2eb21bbead7de81d2f3078 *./tests/data/lavf/lavf.rm
|
||||
346706 ./tests/data/lavf/lavf.rm
|
||||
|
||||
Loading…
Reference in New Issue
Block a user