Autres/audacity
1
0
Fork 0
mirror of https://github.com/cookiengineer/audacity synced 2025-10-31 14:13:50 +01:00
Code Issues Releases Wiki Activity
Files
b3539ed2b84e2c678f9e645c55150d6fee0c2ea8
audacity/lib-src/libmp3lame/quantize.c
James Crook 5e0efd1a25 Start on built-in LAME 
Using LAME 3.10
Windows project files substantially changed from original, and included into audacity solution.
2019-03-26 17:46:53 +00:00

2051 lines
64 KiB
C
Raw Blame History

/*
* MP3 quantization
*
* Copyright (c) 1999-2000 Mark Taylor
* Copyright (c) 1999-2003 Takehiro Tominaga
* Copyright (c) 2000-2011 Robert Hegemann
* Copyright (c) 2001-2005 Gabriel Bouvigne
*
* This library is free software; you can redistribute it and/or
* modify it under the terms of the GNU Library General Public
* License as published by the Free Software Foundation; either
* version 2 of the License, or (at your option) any later version.
*
* This library 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
* Library General Public License for more details.
*
* You should have received a copy of the GNU Library General Public
* License along with this library; if not, write to the
* Free Software Foundation, Inc., 59 Temple Place - Suite 330,
* Boston, MA 02111-1307, USA.
*/
/* $Id: quantize.c,v 1.219 2017/08/02 19:48:05 robert Exp $ */
#ifdef HAVE_CONFIG_H
# include <config.h>
#endif
#include "lame.h"
#include "machine.h"
#include "encoder.h"
#include "util.h"
#include "quantize_pvt.h"
#include "reservoir.h"
#include "bitstream.h"
#include "vbrquantize.h"
#include "quantize.h"
#ifdef HAVE_XMMINTRIN_H
#include "vector/lame_intrin.h"
#endif
/* convert from L/R <-> Mid/Side */
static void
ms_convert(III_side_info_t * l3_side, int gr)
{
int i;
for (i = 0; i < 576; ++i) {
FLOAT l, r;
l = l3_side->tt[gr][0].xr[i];
r = l3_side->tt[gr][1].xr[i];
l3_side->tt[gr][0].xr[i] = (l + r) * (FLOAT) (SQRT2 * 0.5);
l3_side->tt[gr][1].xr[i] = (l - r) * (FLOAT) (SQRT2 * 0.5);
}
}
/************************************************************************
*
* init_outer_loop()
* mt 6/99
*
* initializes cod_info, scalefac and xrpow
*
* returns 0 if all energies in xr are zero, else 1
*
************************************************************************/
static void
init_xrpow_core_c(gr_info * const cod_info, FLOAT xrpow[576], int upper, FLOAT * sum)
{
int i;
FLOAT tmp;
*sum = 0;
for (i = 0; i <= upper; ++i) {
tmp = fabs(cod_info->xr[i]);
*sum += tmp;
xrpow[i] = sqrt(tmp * sqrt(tmp));
if (xrpow[i] > cod_info->xrpow_max)
cod_info->xrpow_max = xrpow[i];
}
}
void
init_xrpow_core_init(lame_internal_flags * const gfc)
{
gfc->init_xrpow_core = init_xrpow_core_c;
#if defined(HAVE_XMMINTRIN_H)
if (gfc->CPU_features.SSE)
gfc->init_xrpow_core = init_xrpow_core_sse;
#endif
#ifndef HAVE_NASM
#ifdef MIN_ARCH_SSE
gfc->init_xrpow_core = init_xrpow_core_sse;
#endif
#endif
}
static int
init_xrpow(lame_internal_flags * gfc, gr_info * const cod_info, FLOAT xrpow[576])
{
FLOAT sum = 0;
int i;
int const upper = cod_info->max_nonzero_coeff;
assert(xrpow != NULL);
cod_info->xrpow_max = 0;
/* check if there is some energy we have to quantize
* and calculate xrpow matching our fresh scalefactors
*/
assert(0 <= upper && upper <= 575);
memset(&(xrpow[upper]), 0, (576 - upper) * sizeof(xrpow[0]));
gfc->init_xrpow_core(cod_info, xrpow, upper, &sum);
/* return 1 if we have something to quantize, else 0
*/
if (sum > (FLOAT) 1E-20) {
int j = 0;
if (gfc->sv_qnt.substep_shaping & 2)
j = 1;
for (i = 0; i < cod_info->psymax; i++)
gfc->sv_qnt.pseudohalf[i] = j;
return 1;
}
memset(&cod_info->l3_enc[0], 0, sizeof(int) * 576);
return 0;
}
/*
Gabriel Bouvigne feb/apr 2003
Analog silence detection in partitionned sfb21
or sfb12 for short blocks
From top to bottom of sfb, changes to 0
coeffs which are below ath. It stops on the first
coeff higher than ath.
*/
static void
psfb21_analogsilence(lame_internal_flags const *gfc, gr_info * const cod_info)
{
ATH_t const *const ATH = gfc->ATH;
FLOAT *const xr = cod_info->xr;
if (cod_info->block_type != SHORT_TYPE) { /* NORM, START or STOP type, but not SHORT blocks */
int gsfb;
int stop = 0;
for (gsfb = PSFB21 - 1; gsfb >= 0 && !stop; gsfb--) {
int const start = gfc->scalefac_band.psfb21[gsfb];
int const end = gfc->scalefac_band.psfb21[gsfb + 1];
int j;
FLOAT ath21;
ath21 = athAdjust(ATH->adjust_factor, ATH->psfb21[gsfb], ATH->floor, 0);
if (gfc->sv_qnt.longfact[21] > 1e-12f)
ath21 *= gfc->sv_qnt.longfact[21];
for (j = end - 1; j >= start; j--) {
if (fabs(xr[j]) < ath21)
xr[j] = 0;
else {
stop = 1;
break;
}
}
}
}
else {
/*note: short blocks coeffs are reordered */
int block;
for (block = 0; block < 3; block++) {
int gsfb;
int stop = 0;
for (gsfb = PSFB12 - 1; gsfb >= 0 && !stop; gsfb--) {
int const start = gfc->scalefac_band.s[12] * 3 +
(gfc->scalefac_band.s[13] - gfc->scalefac_band.s[12]) * block +
(gfc->scalefac_band.psfb12[gsfb] - gfc->scalefac_band.psfb12[0]);
int const end =
start + (gfc->scalefac_band.psfb12[gsfb + 1] - gfc->scalefac_band.psfb12[gsfb]);
int j;
FLOAT ath12;
ath12 = athAdjust(ATH->adjust_factor, ATH->psfb12[gsfb], ATH->floor, 0);
if (gfc->sv_qnt.shortfact[12] > 1e-12f)
ath12 *= gfc->sv_qnt.shortfact[12];
for (j = end - 1; j >= start; j--) {
if (fabs(xr[j]) < ath12)
xr[j] = 0;
else {
stop = 1;
break;
}
}
}
}
}
}
static void
init_outer_loop(lame_internal_flags const *gfc, gr_info * const cod_info)
{
SessionConfig_t const *const cfg = &gfc->cfg;
int sfb, j;
/* initialize fresh cod_info
*/
cod_info->part2_3_length = 0;
cod_info->big_values = 0;
cod_info->count1 = 0;
cod_info->global_gain = 210;
cod_info->scalefac_compress = 0;
/* mixed_block_flag, block_type was set in psymodel.c */
cod_info->table_select[0] = 0;
cod_info->table_select[1] = 0;
cod_info->table_select[2] = 0;
cod_info->subblock_gain[0] = 0;
cod_info->subblock_gain[1] = 0;
cod_info->subblock_gain[2] = 0;
cod_info->subblock_gain[3] = 0; /* this one is always 0 */
cod_info->region0_count = 0;
cod_info->region1_count = 0;
cod_info->preflag = 0;
cod_info->scalefac_scale = 0;
cod_info->count1table_select = 0;
cod_info->part2_length = 0;
if (cfg->samplerate_out <= 8000) {
cod_info->sfb_lmax = 17;
cod_info->sfb_smin = 9;
cod_info->psy_lmax = 17;
}
else {
cod_info->sfb_lmax = SBPSY_l;
cod_info->sfb_smin = SBPSY_s;
cod_info->psy_lmax = gfc->sv_qnt.sfb21_extra ? SBMAX_l : SBPSY_l;
}
cod_info->psymax = cod_info->psy_lmax;
cod_info->sfbmax = cod_info->sfb_lmax;
cod_info->sfbdivide = 11;
for (sfb = 0; sfb < SBMAX_l; sfb++) {
cod_info->width[sfb]
= gfc->scalefac_band.l[sfb + 1] - gfc->scalefac_band.l[sfb];
cod_info->window[sfb] = 3; /* which is always 0. */
}
if (cod_info->block_type == SHORT_TYPE) {
FLOAT ixwork[576];
FLOAT *ix;
cod_info->sfb_smin = 0;
cod_info->sfb_lmax = 0;
if (cod_info->mixed_block_flag) {
/*
* MPEG-1: sfbs 0-7 long block, 3-12 short blocks
* MPEG-2(.5): sfbs 0-5 long block, 3-12 short blocks
*/
cod_info->sfb_smin = 3;
cod_info->sfb_lmax = cfg->mode_gr * 2 + 4;
}
if (cfg->samplerate_out <= 8000) {
cod_info->psymax
= cod_info->sfb_lmax
+ 3 * (9 - cod_info->sfb_smin);
cod_info->sfbmax = cod_info->sfb_lmax + 3 * (9 - cod_info->sfb_smin);
}
else {
cod_info->psymax
= cod_info->sfb_lmax
+ 3 * ((gfc->sv_qnt.sfb21_extra ? SBMAX_s : SBPSY_s) - cod_info->sfb_smin);
cod_info->sfbmax = cod_info->sfb_lmax + 3 * (SBPSY_s - cod_info->sfb_smin);
}
cod_info->sfbdivide = cod_info->sfbmax - 18;
cod_info->psy_lmax = cod_info->sfb_lmax;
/* re-order the short blocks, for more efficient encoding below */
/* By Takehiro TOMINAGA */
/*
Within each scalefactor band, data is given for successive
time windows, beginning with window 0 and ending with window 2.
Within each window, the quantized values are then arranged in
order of increasing frequency...
*/
ix = &cod_info->xr[gfc->scalefac_band.l[cod_info->sfb_lmax]];
memcpy(ixwork, cod_info->xr, 576 * sizeof(FLOAT));
for (sfb = cod_info->sfb_smin; sfb < SBMAX_s; sfb++) {
int const start = gfc->scalefac_band.s[sfb];
int const end = gfc->scalefac_band.s[sfb + 1];
int window, l;
for (window = 0; window < 3; window++) {
for (l = start; l < end; l++) {
*ix++ = ixwork[3 * l + window];
}
}
}
j = cod_info->sfb_lmax;
for (sfb = cod_info->sfb_smin; sfb < SBMAX_s; sfb++) {
cod_info->width[j] = cod_info->width[j + 1] = cod_info->width[j + 2]
= gfc->scalefac_band.s[sfb + 1] - gfc->scalefac_band.s[sfb];
cod_info->window[j] = 0;
cod_info->window[j + 1] = 1;
cod_info->window[j + 2] = 2;
j += 3;
}
}
cod_info->count1bits = 0;
cod_info->sfb_partition_table = nr_of_sfb_block[0][0];
cod_info->slen[0] = 0;
cod_info->slen[1] = 0;
cod_info->slen[2] = 0;
cod_info->slen[3] = 0;
cod_info->max_nonzero_coeff = 575;
/* fresh scalefactors are all zero
*/
memset(cod_info->scalefac, 0, sizeof(cod_info->scalefac));
if (cfg->vbr != vbr_mt && cfg->vbr != vbr_mtrh && cfg->vbr != vbr_abr && cfg->vbr != vbr_off) {
psfb21_analogsilence(gfc, cod_info);
}
}
/************************************************************************
*
* bin_search_StepSize()
*
* author/date??
*
* binary step size search
* used by outer_loop to get a quantizer step size to start with
*
************************************************************************/
typedef enum {
BINSEARCH_NONE,
BINSEARCH_UP,
BINSEARCH_DOWN
} binsearchDirection_t;
static int
bin_search_StepSize(lame_internal_flags * const gfc, gr_info * const cod_info,
int desired_rate, const int ch, const FLOAT xrpow[576])
{
int nBits;
int CurrentStep = gfc->sv_qnt.CurrentStep[ch];
int flag_GoneOver = 0;
int const start = gfc->sv_qnt.OldValue[ch];
binsearchDirection_t Direction = BINSEARCH_NONE;
cod_info->global_gain = start;
desired_rate -= cod_info->part2_length;
assert(CurrentStep);
for (;;) {
int step;
nBits = count_bits(gfc, xrpow, cod_info, 0);
if (CurrentStep == 1 || nBits == desired_rate)
break; /* nothing to adjust anymore */
if (nBits > desired_rate) {
/* increase Quantize_StepSize */
if (Direction == BINSEARCH_DOWN)
flag_GoneOver = 1;
if (flag_GoneOver)
CurrentStep /= 2;
Direction = BINSEARCH_UP;
step = CurrentStep;
}
else {
/* decrease Quantize_StepSize */
if (Direction == BINSEARCH_UP)
flag_GoneOver = 1;
if (flag_GoneOver)
CurrentStep /= 2;
Direction = BINSEARCH_DOWN;
step = -CurrentStep;
}
cod_info->global_gain += step;
if (cod_info->global_gain < 0) {
cod_info->global_gain = 0;
flag_GoneOver = 1;
}
if (cod_info->global_gain > 255) {
cod_info->global_gain = 255;
flag_GoneOver = 1;
}
}
assert(cod_info->global_gain >= 0);
assert(cod_info->global_gain < 256);
while (nBits > desired_rate && cod_info->global_gain < 255) {
cod_info->global_gain++;
nBits = count_bits(gfc, xrpow, cod_info, 0);
}
gfc->sv_qnt.CurrentStep[ch] = (start - cod_info->global_gain >= 4) ? 4 : 2;
gfc->sv_qnt.OldValue[ch] = cod_info->global_gain;
cod_info->part2_3_length = nBits;
return nBits;
}
/************************************************************************
*
* trancate_smallspectrums()
*
* Takehiro TOMINAGA 2002-07-21
*
* trancate smaller nubmers into 0 as long as the noise threshold is allowed.
*
************************************************************************/
static int
floatcompare(const void *v1, const void *v2)
{
const FLOAT *const a = v1, *const b = v2;
if (*a > *b)
return 1;
if (*a < *b)
return -1;
return 0;
}
static void
trancate_smallspectrums(lame_internal_flags const *gfc,
gr_info * const gi, const FLOAT * const l3_xmin, FLOAT * const work)
{
int sfb, j, width;
FLOAT distort[SFBMAX];
calc_noise_result dummy;
if ((!(gfc->sv_qnt.substep_shaping & 4) && gi->block_type == SHORT_TYPE)
|| gfc->sv_qnt.substep_shaping & 0x80)
return;
(void) calc_noise(gi, l3_xmin, distort, &dummy, 0);
for (j = 0; j < 576; j++) {
FLOAT xr = 0.0;
if (gi->l3_enc[j] != 0)
xr = fabs(gi->xr[j]);
work[j] = xr;
}
j = 0;
sfb = 8;
if (gi->block_type == SHORT_TYPE)
sfb = 6;
do {
FLOAT allowedNoise, trancateThreshold;
int nsame, start;
width = gi->width[sfb];
j += width;
if (distort[sfb] >= 1.0)
continue;
qsort(&work[j - width], width, sizeof(FLOAT), floatcompare);
if (EQ(work[j - 1], 0.0))
continue; /* all zero sfb */
allowedNoise = (1.0 - distort[sfb]) * l3_xmin[sfb];
trancateThreshold = 0.0;
start = 0;
do {
FLOAT noise;
for (nsame = 1; start + nsame < width; nsame++)
if (NEQ(work[start + j - width], work[start + j + nsame - width]))
break;
noise = work[start + j - width] * work[start + j - width] * nsame;
if (allowedNoise < noise) {
if (start != 0)
trancateThreshold = work[start + j - width - 1];
break;
}
allowedNoise -= noise;
start += nsame;
} while (start < width);
if (EQ(trancateThreshold, 0.0))
continue;
/* printf("%e %e %e\n", */
/* trancateThreshold/l3_xmin[sfb], */
/* trancateThreshold/(l3_xmin[sfb]*start), */
/* trancateThreshold/(l3_xmin[sfb]*(start+width)) */
/* ); */
/* if (trancateThreshold > 1000*l3_xmin[sfb]*start) */
/* trancateThreshold = 1000*l3_xmin[sfb]*start; */
do {
if (fabs(gi->xr[j - width]) <= trancateThreshold)
gi->l3_enc[j - width] = 0;
} while (--width > 0);
} while (++sfb < gi->psymax);
gi->part2_3_length = noquant_count_bits(gfc, gi, 0);
}
/*************************************************************************
*
* loop_break()
*
* author/date??
*
* Function: Returns zero if there is a scalefac which has not been
* amplified. Otherwise it returns one.
*
*************************************************************************/
inline static int
loop_break(const gr_info * const cod_info)
{
int sfb;
for (sfb = 0; sfb < cod_info->sfbmax; sfb++)
if (cod_info->scalefac[sfb]
+ cod_info->subblock_gain[cod_info->window[sfb]] == 0)
return 0;
return 1;
}
/* mt 5/99: Function: Improved calc_noise for a single channel */
/*************************************************************************
*
* quant_compare()
*
* author/date??
*
* several different codes to decide which quantization is better
*
*************************************************************************/
static double
penalties(double noise)
{
return FAST_LOG10(0.368 + 0.632 * noise * noise * noise);
}
static double
get_klemm_noise(const FLOAT * distort, const gr_info * const gi)
{
int sfb;
double klemm_noise = 1E-37;
for (sfb = 0; sfb < gi->psymax; sfb++)
klemm_noise += penalties(distort[sfb]);
return Max(1e-20, klemm_noise);
}
inline static int
quant_compare(const int quant_comp,
const calc_noise_result * const best,
calc_noise_result * const calc, const gr_info * const gi, const FLOAT * distort)
{
/*
noise is given in decibels (dB) relative to masking thesholds.
over_noise: ??? (the previous comment is fully wrong)
tot_noise: ??? (the previous comment is fully wrong)
max_noise: max quantization noise
*/
int better;
switch (quant_comp) {
default:
case 9:{
if (best->over_count > 0) {
/* there are distorted sfb */
better = calc->over_SSD <= best->over_SSD;
if (calc->over_SSD == best->over_SSD)
better = calc->bits < best->bits;
}
else {
/* no distorted sfb */
better = ((calc->max_noise < 0) &&
((calc->max_noise * 10 + calc->bits) <=
(best->max_noise * 10 + best->bits)));
}
break;
}
case 0:
better = calc->over_count < best->over_count
|| (calc->over_count == best->over_count && calc->over_noise < best->over_noise)
|| (calc->over_count == best->over_count &&
EQ(calc->over_noise, best->over_noise) && calc->tot_noise < best->tot_noise);
break;
case 8:
calc->max_noise = get_klemm_noise(distort, gi);
/*lint --fallthrough */
case 1:
better = calc->max_noise < best->max_noise;
break;
case 2:
better = calc->tot_noise < best->tot_noise;
break;
case 3:
better = (calc->tot_noise < best->tot_noise)
&& (calc->max_noise < best->max_noise);
break;
case 4:
better = (calc->max_noise <= 0.0 && best->max_noise > 0.2)
|| (calc->max_noise <= 0.0 &&
best->max_noise < 0.0 &&
best->max_noise > calc->max_noise - 0.2 && calc->tot_noise < best->tot_noise)
|| (calc->max_noise <= 0.0 &&
best->max_noise > 0.0 &&
best->max_noise > calc->max_noise - 0.2 &&
calc->tot_noise < best->tot_noise + best->over_noise)
|| (calc->max_noise > 0.0 &&
best->max_noise > -0.05 &&
best->max_noise > calc->max_noise - 0.1 &&
calc->tot_noise + calc->over_noise < best->tot_noise + best->over_noise)
|| (calc->max_noise > 0.0 &&
best->max_noise > -0.1 &&
best->max_noise > calc->max_noise - 0.15 &&
calc->tot_noise + calc->over_noise + calc->over_noise <
best->tot_noise + best->over_noise + best->over_noise);
break;
case 5:
better = calc->over_noise < best->over_noise
|| (EQ(calc->over_noise, best->over_noise) && calc->tot_noise < best->tot_noise);
break;
case 6:
better = calc->over_noise < best->over_noise
|| (EQ(calc->over_noise, best->over_noise) &&
(calc->max_noise < best->max_noise
|| (EQ(calc->max_noise, best->max_noise) && calc->tot_noise <= best->tot_noise)
));
break;
case 7:
better = calc->over_count < best->over_count || calc->over_noise < best->over_noise;
break;
}
if (best->over_count == 0) {
/*
If no distorted bands, only use this quantization
if it is better, and if it uses less bits.
Unfortunately, part2_3_length is sometimes a poor
estimator of the final size at low bitrates.
*/
better = better && calc->bits < best->bits;
}
return better;
}
/*************************************************************************
*
* amp_scalefac_bands()
*
* author/date??
*
* Amplify the scalefactor bands that violate the masking threshold.
* See ISO 11172-3 Section C.1.5.4.3.5
*
* distort[] = noise/masking
* distort[] > 1 ==> noise is not masked
* distort[] < 1 ==> noise is masked
* max_dist = maximum value of distort[]
*
* Three algorithms:
* noise_shaping_amp
* 0 Amplify all bands with distort[]>1.
*
* 1 Amplify all bands with distort[] >= max_dist^(.5);
* ( 50% in the db scale)
*
* 2 Amplify first band with distort[] >= max_dist;
*
*
* For algorithms 0 and 1, if max_dist < 1, then amplify all bands
* with distort[] >= .95*max_dist. This is to make sure we always
* amplify at least one band.
*
*
*************************************************************************/
static void
amp_scalefac_bands(lame_internal_flags * gfc,
gr_info * const cod_info, FLOAT const *distort, FLOAT xrpow[576], int bRefine)
{
SessionConfig_t const *const cfg = &gfc->cfg;
int j, sfb;
FLOAT ifqstep34, trigger;
int noise_shaping_amp;
if (cod_info->scalefac_scale == 0) {
ifqstep34 = 1.29683955465100964055; /* 2**(.75*.5) */
}
else {
ifqstep34 = 1.68179283050742922612; /* 2**(.75*1) */
}
/* compute maximum value of distort[] */
trigger = 0;
for (sfb = 0; sfb < cod_info->sfbmax; sfb++) {
if (trigger < distort[sfb])
trigger = distort[sfb];
}
noise_shaping_amp = cfg->noise_shaping_amp;
if (noise_shaping_amp == 3) {
if (bRefine == 1)
noise_shaping_amp = 2;
else
noise_shaping_amp = 1;
}
switch (noise_shaping_amp) {
case 2:
/* amplify exactly 1 band */
break;
case 1:
/* amplify bands within 50% of max (on db scale) */
if (trigger > 1.0)
trigger = pow(trigger, .5);
else
trigger *= .95;
break;
case 0:
default:
/* ISO algorithm. amplify all bands with distort>1 */
if (trigger > 1.0)
trigger = 1.0;
else
trigger *= .95;
break;
}
j = 0;
for (sfb = 0; sfb < cod_info->sfbmax; sfb++) {
int const width = cod_info->width[sfb];
int l;
j += width;
if (distort[sfb] < trigger)
continue;
if (gfc->sv_qnt.substep_shaping & 2) {
gfc->sv_qnt.pseudohalf[sfb] = !gfc->sv_qnt.pseudohalf[sfb];
if (!gfc->sv_qnt.pseudohalf[sfb] && cfg->noise_shaping_amp == 2)
return;
}
cod_info->scalefac[sfb]++;
for (l = -width; l < 0; l++) {
xrpow[j + l] *= ifqstep34;
if (xrpow[j + l] > cod_info->xrpow_max)
cod_info->xrpow_max = xrpow[j + l];
}
if (cfg->noise_shaping_amp == 2)
return;
}
}
/*************************************************************************
*
* inc_scalefac_scale()
*
* Takehiro Tominaga 2000-xx-xx
*
* turns on scalefac scale and adjusts scalefactors
*
*************************************************************************/
static void
inc_scalefac_scale(gr_info * const cod_info, FLOAT xrpow[576])
{
int l, j, sfb;
const FLOAT ifqstep34 = 1.29683955465100964055;
j = 0;
for (sfb = 0; sfb < cod_info->sfbmax; sfb++) {
int const width = cod_info->width[sfb];
int s = cod_info->scalefac[sfb];
if (cod_info->preflag)
s += pretab[sfb];
j += width;
if (s & 1) {
s++;
for (l = -width; l < 0; l++) {
xrpow[j + l] *= ifqstep34;
if (xrpow[j + l] > cod_info->xrpow_max)
cod_info->xrpow_max = xrpow[j + l];
}
}
cod_info->scalefac[sfb] = s >> 1;
}
cod_info->preflag = 0;
cod_info->scalefac_scale = 1;
}
/*************************************************************************
*
* inc_subblock_gain()
*
* Takehiro Tominaga 2000-xx-xx
*
* increases the subblock gain and adjusts scalefactors
*
*************************************************************************/
static int
inc_subblock_gain(const lame_internal_flags * const gfc, gr_info * const cod_info, FLOAT xrpow[576])
{
int sfb, window;
int *const scalefac = cod_info->scalefac;
/* subbloc_gain can't do anything in the long block region */
for (sfb = 0; sfb < cod_info->sfb_lmax; sfb++) {
if (scalefac[sfb] >= 16)
return 1;
}
for (window = 0; window < 3; window++) {
int s1, s2, l, j;
s1 = s2 = 0;
for (sfb = cod_info->sfb_lmax + window; sfb < cod_info->sfbdivide; sfb += 3) {
if (s1 < scalefac[sfb])
s1 = scalefac[sfb];
}
for (; sfb < cod_info->sfbmax; sfb += 3) {
if (s2 < scalefac[sfb])
s2 = scalefac[sfb];
}
if (s1 < 16 && s2 < 8)
continue;
if (cod_info->subblock_gain[window] >= 7)
return 1;
/* even though there is no scalefactor for sfb12
* subblock gain affects upper frequencies too, that's why
* we have to go up to SBMAX_s
*/
cod_info->subblock_gain[window]++;
j = gfc->scalefac_band.l[cod_info->sfb_lmax];
for (sfb = cod_info->sfb_lmax + window; sfb < cod_info->sfbmax; sfb += 3) {
FLOAT amp;
int const width = cod_info->width[sfb];
int s = scalefac[sfb];
assert(s >= 0);
s = s - (4 >> cod_info->scalefac_scale);
if (s >= 0) {
scalefac[sfb] = s;
j += width * 3;
continue;
}
scalefac[sfb] = 0;
{
int const gain = 210 + (s << (cod_info->scalefac_scale + 1));
amp = IPOW20(gain);
}
j += width * (window + 1);
for (l = -width; l < 0; l++) {
xrpow[j + l] *= amp;
if (xrpow[j + l] > cod_info->xrpow_max)
cod_info->xrpow_max = xrpow[j + l];
}
j += width * (3 - window - 1);
}
{
FLOAT const amp = IPOW20(202);
j += cod_info->width[sfb] * (window + 1);
for (l = -cod_info->width[sfb]; l < 0; l++) {
xrpow[j + l] *= amp;
if (xrpow[j + l] > cod_info->xrpow_max)
cod_info->xrpow_max = xrpow[j + l];
}
}
}
return 0;
}
/********************************************************************
*
* balance_noise()
*
* Takehiro Tominaga /date??
* Robert Hegemann 2000-09-06: made a function of it
*
* amplifies scalefactor bands,
* - if all are already amplified returns 0
* - if some bands are amplified too much:
* * try to increase scalefac_scale
* * if already scalefac_scale was set
* try on short blocks to increase subblock gain
*
********************************************************************/
inline static int
balance_noise(lame_internal_flags * gfc,
gr_info * const cod_info, FLOAT const *distort, FLOAT xrpow[576], int bRefine)
{
SessionConfig_t const *const cfg = &gfc->cfg;
int status;
amp_scalefac_bands(gfc, cod_info, distort, xrpow, bRefine);
/* check to make sure we have not amplified too much
* loop_break returns 0 if there is an unamplified scalefac
* scale_bitcount returns 0 if no scalefactors are too large
*/
status = loop_break(cod_info);
if (status)
return 0; /* all bands amplified */
/* not all scalefactors have been amplified. so these
* scalefacs are possibly valid. encode them:
*/
status = scale_bitcount(gfc, cod_info);
if (!status)
return 1; /* amplified some bands not exceeding limits */
/* some scalefactors are too large.
* lets try setting scalefac_scale=1
*/
if (cfg->noise_shaping > 1) {
memset(&gfc->sv_qnt.pseudohalf[0], 0, sizeof(gfc->sv_qnt.pseudohalf));
if (!cod_info->scalefac_scale) {
inc_scalefac_scale(cod_info, xrpow);
status = 0;
}
else {
if (cod_info->block_type == SHORT_TYPE && cfg->subblock_gain > 0) {
status = inc_subblock_gain(gfc, cod_info, xrpow)
|| loop_break(cod_info);
}
}
}
if (!status) {
status = scale_bitcount(gfc, cod_info);
}
return !status;
}
/************************************************************************
*
* outer_loop ()
*
* Function: The outer iteration loop controls the masking conditions
* of all scalefactorbands. It computes the best scalefac and
* global gain. This module calls the inner iteration loop
*
* mt 5/99 completely rewritten to allow for bit reservoir control,
* mid/side channels with L/R or mid/side masking thresholds,
* and chooses best quantization instead of last quantization when
* no distortion free quantization can be found.
*
* added VBR support mt 5/99
*
* some code shuffle rh 9/00
************************************************************************/
static int
outer_loop(lame_internal_flags * gfc, gr_info * const cod_info, const FLOAT * const l3_xmin, /* allowed distortion */
FLOAT xrpow[576], /* coloured magnitudes of spectral */
const int ch, const int targ_bits)
{ /* maximum allowed bits */
SessionConfig_t const *const cfg = &gfc->cfg;
gr_info cod_info_w;
FLOAT save_xrpow[576];
FLOAT distort[SFBMAX];
calc_noise_result best_noise_info;
int huff_bits;
int better;
int age;
calc_noise_data prev_noise;
int best_part2_3_length = 9999999;
int bEndOfSearch = 0;
int bRefine = 0;
int best_ggain_pass1 = 0;
(void) bin_search_StepSize(gfc, cod_info, targ_bits, ch, xrpow);
if (!cfg->noise_shaping)
/* fast mode, no noise shaping, we are ready */
return 100; /* default noise_info.over_count */
memset(&prev_noise, 0, sizeof(calc_noise_data));
/* compute the distortion in this quantization */
/* coefficients and thresholds both l/r (or both mid/side) */
(void) calc_noise(cod_info, l3_xmin, distort, &best_noise_info, &prev_noise);
best_noise_info.bits = cod_info->part2_3_length;
cod_info_w = *cod_info;
age = 0;
/* if (cfg->vbr == vbr_rh || cfg->vbr == vbr_mtrh) */
memcpy(save_xrpow, xrpow, sizeof(FLOAT) * 576);
while (!bEndOfSearch) {
/* BEGIN MAIN LOOP */
do {
calc_noise_result noise_info;
int search_limit;
int maxggain = 255;
/* When quantization with no distorted bands is found,
* allow up to X new unsuccesful tries in serial. This
* gives us more possibilities for different quant_compare modes.
* Much more than 3 makes not a big difference, it is only slower.
*/
if (gfc->sv_qnt.substep_shaping & 2) {
search_limit = 20;
}
else {
search_limit = 3;
}
/* Check if the last scalefactor band is distorted.
* in VBR mode we can't get rid of the distortion, so quit now
* and VBR mode will try again with more bits.
* (makes a 10% speed increase, the files I tested were
* binary identical, 2000/05/20 Robert Hegemann)
* distort[] > 1 means noise > allowed noise
*/
if (gfc->sv_qnt.sfb21_extra) {
if (distort[cod_info_w.sfbmax] > 1.0)
break;
if (cod_info_w.block_type == SHORT_TYPE
&& (distort[cod_info_w.sfbmax + 1] > 1.0
|| distort[cod_info_w.sfbmax + 2] > 1.0))
break;
}
/* try a new scalefactor conbination on cod_info_w */
if (balance_noise(gfc, &cod_info_w, distort, xrpow, bRefine) == 0)
break;
if (cod_info_w.scalefac_scale)
maxggain = 254;
/* inner_loop starts with the initial quantization step computed above
* and slowly increases until the bits < huff_bits.
* Thus it is important not to start with too large of an inital
* quantization step. Too small is ok, but inner_loop will take longer
*/
huff_bits = targ_bits - cod_info_w.part2_length;
if (huff_bits <= 0)
break;
/* increase quantizer stepsize until needed bits are below maximum
*/
while ((cod_info_w.part2_3_length
= count_bits(gfc, xrpow, &cod_info_w, &prev_noise)) > huff_bits
&& cod_info_w.global_gain <= maxggain)
cod_info_w.global_gain++;
if (cod_info_w.global_gain > maxggain)
break;
if (best_noise_info.over_count == 0) {
while ((cod_info_w.part2_3_length
= count_bits(gfc, xrpow, &cod_info_w, &prev_noise)) > best_part2_3_length
&& cod_info_w.global_gain <= maxggain)
cod_info_w.global_gain++;
if (cod_info_w.global_gain > maxggain)
break;
}
/* compute the distortion in this quantization */
(void) calc_noise(&cod_info_w, l3_xmin, distort, &noise_info, &prev_noise);
noise_info.bits = cod_info_w.part2_3_length;
/* check if this quantization is better
* than our saved quantization */
if (cod_info->block_type != SHORT_TYPE) /* NORM, START or STOP type */
better = cfg->quant_comp;
else
better = cfg->quant_comp_short;
better = quant_compare(better, &best_noise_info, &noise_info, &cod_info_w, distort);
/* save data so we can restore this quantization later */
if (better) {
best_part2_3_length = cod_info->part2_3_length;
best_noise_info = noise_info;
*cod_info = cod_info_w;
age = 0;
/* save data so we can restore this quantization later */
/*if (cfg->vbr == vbr_rh || cfg->vbr == vbr_mtrh) */ {
/* store for later reuse */
memcpy(save_xrpow, xrpow, sizeof(FLOAT) * 576);
}
}
else {
/* early stop? */
if (cfg->full_outer_loop == 0) {
if (++age > search_limit && best_noise_info.over_count == 0)
break;
if ((cfg->noise_shaping_amp == 3) && bRefine && age > 30)
break;
if ((cfg->noise_shaping_amp == 3) && bRefine &&
(cod_info_w.global_gain - best_ggain_pass1) > 15)
break;
}
}
}
while ((cod_info_w.global_gain + cod_info_w.scalefac_scale) < 255);
if (cfg->noise_shaping_amp == 3) {
if (!bRefine) {
/* refine search */
cod_info_w = *cod_info;
memcpy(xrpow, save_xrpow, sizeof(FLOAT) * 576);
age = 0;
best_ggain_pass1 = cod_info_w.global_gain;
bRefine = 1;
}
else {
/* search already refined, stop */
bEndOfSearch = 1;
}
}
else {
bEndOfSearch = 1;
}
}
assert((cod_info->global_gain + cod_info->scalefac_scale) <= 255);
/* finish up
*/
if (cfg->vbr == vbr_rh || cfg->vbr == vbr_mtrh || cfg->vbr == vbr_mt)
/* restore for reuse on next try */
memcpy(xrpow, save_xrpow, sizeof(FLOAT) * 576);
/* do the 'substep shaping'
*/
else if (gfc->sv_qnt.substep_shaping & 1)
trancate_smallspectrums(gfc, cod_info, l3_xmin, xrpow);
return best_noise_info.over_count;
}
/************************************************************************
*
* iteration_finish_one()
*
* Robert Hegemann 2000-09-06
*
* update reservoir status after FINAL quantization/bitrate
*
************************************************************************/
static void
iteration_finish_one(lame_internal_flags * gfc, int gr, int ch)
{
SessionConfig_t const *const cfg = &gfc->cfg;
III_side_info_t *const l3_side = &gfc->l3_side;
gr_info *const cod_info = &l3_side->tt[gr][ch];
/* try some better scalefac storage
*/
best_scalefac_store(gfc, gr, ch, l3_side);
/* best huffman_divide may save some bits too
*/
if (cfg->use_best_huffman == 1)
best_huffman_divide(gfc, cod_info);
/* update reservoir status after FINAL quantization/bitrate
*/
ResvAdjust(gfc, cod_info);
}
/*********************************************************************
*
* VBR_encode_granule()
*
* 2000-09-04 Robert Hegemann
*
*********************************************************************/
static void
VBR_encode_granule(lame_internal_flags * gfc, gr_info * const cod_info, const FLOAT * const l3_xmin, /* allowed distortion of the scalefactor */
FLOAT xrpow[576], /* coloured magnitudes of spectral values */
const int ch, int min_bits, int max_bits)
{
gr_info bst_cod_info;
FLOAT bst_xrpow[576];
int const Max_bits = max_bits;
int real_bits = max_bits + 1;
int this_bits = (max_bits + min_bits) / 2;
int dbits, over, found = 0;
int const sfb21_extra = gfc->sv_qnt.sfb21_extra;
assert(Max_bits <= MAX_BITS_PER_CHANNEL);
memset(bst_cod_info.l3_enc, 0, sizeof(bst_cod_info.l3_enc));
/* search within round about 40 bits of optimal
*/
do {
assert(this_bits >= min_bits);
assert(this_bits <= max_bits);
assert(min_bits <= max_bits);
if (this_bits > Max_bits - 42)
gfc->sv_qnt.sfb21_extra = 0;
else
gfc->sv_qnt.sfb21_extra = sfb21_extra;
over = outer_loop(gfc, cod_info, l3_xmin, xrpow, ch, this_bits);
/* is quantization as good as we are looking for ?
* in this case: is no scalefactor band distorted?
*/
if (over <= 0) {
found = 1;
/* now we know it can be done with "real_bits"
* and maybe we can skip some iterations
*/
real_bits = cod_info->part2_3_length;
/* store best quantization so far
*/
bst_cod_info = *cod_info;
memcpy(bst_xrpow, xrpow, sizeof(FLOAT) * 576);
/* try with fewer bits
*/
max_bits = real_bits - 32;
dbits = max_bits - min_bits;
this_bits = (max_bits + min_bits) / 2;
}
else {
/* try with more bits
*/
min_bits = this_bits + 32;
dbits = max_bits - min_bits;
this_bits = (max_bits + min_bits) / 2;
if (found) {
found = 2;
/* start again with best quantization so far
*/
*cod_info = bst_cod_info;
memcpy(xrpow, bst_xrpow, sizeof(FLOAT) * 576);
}
}
} while (dbits > 12);
gfc->sv_qnt.sfb21_extra = sfb21_extra;
/* found=0 => nothing found, use last one
* found=1 => we just found the best and left the loop
* found=2 => we restored a good one and have now l3_enc to restore too
*/
if (found == 2) {
memcpy(cod_info->l3_enc, bst_cod_info.l3_enc, sizeof(int) * 576);
}
assert(cod_info->part2_3_length <= Max_bits);
}
/************************************************************************
*
* get_framebits()
*
* Robert Hegemann 2000-09-05
*
* calculates
* * how many bits are available for analog silent granules
* * how many bits to use for the lowest allowed bitrate
* * how many bits each bitrate would provide
*
************************************************************************/
static void
get_framebits(lame_internal_flags * gfc, int frameBits[15])
{
SessionConfig_t const *const cfg = &gfc->cfg;
EncResult_t *const eov = &gfc->ov_enc;
int bitsPerFrame, i;
/* always use at least this many bits per granule per channel
* unless we detect analog silence, see below
*/
eov->bitrate_index = cfg->vbr_min_bitrate_index;
bitsPerFrame = getframebits(gfc);
/* bits for analog silence
*/
eov->bitrate_index = 1;
bitsPerFrame = getframebits(gfc);
for (i = 1; i <= cfg->vbr_max_bitrate_index; i++) {
eov->bitrate_index = i;
frameBits[i] = ResvFrameBegin(gfc, &bitsPerFrame);
}
}
/*********************************************************************
*
* VBR_prepare()
*
* 2000-09-04 Robert Hegemann
*
* * converts LR to MS coding when necessary
* * calculates allowed/adjusted quantization noise amounts
* * detects analog silent frames
*
* some remarks:
* - lower masking depending on Quality setting
* - quality control together with adjusted ATH MDCT scaling
* on lower quality setting allocate more noise from
* ATH masking, and on higher quality setting allocate
* less noise from ATH masking.
* - experiments show that going more than 2dB over GPSYCHO's
* limits ends up in very annoying artefacts
*
*********************************************************************/
/* RH: this one needs to be overhauled sometime */
static int
VBR_old_prepare(lame_internal_flags * gfc,
const FLOAT pe[2][2], FLOAT const ms_ener_ratio[2],
const III_psy_ratio ratio[2][2],
FLOAT l3_xmin[2][2][SFBMAX],
int frameBits[16], int min_bits[2][2], int max_bits[2][2], int bands[2][2])
{
SessionConfig_t const *const cfg = &gfc->cfg;
EncResult_t *const eov = &gfc->ov_enc;
FLOAT masking_lower_db, adjust = 0.0;
int gr, ch;
int analog_silence = 1;
int avg, mxb, bits = 0;
eov->bitrate_index = cfg->vbr_max_bitrate_index;
avg = ResvFrameBegin(gfc, &avg) / cfg->mode_gr;
get_framebits(gfc, frameBits);
for (gr = 0; gr < cfg->mode_gr; gr++) {
mxb = on_pe(gfc, pe, max_bits[gr], avg, gr, 0);
if (gfc->ov_enc.mode_ext == MPG_MD_MS_LR) {
ms_convert(&gfc->l3_side, gr);
reduce_side(max_bits[gr], ms_ener_ratio[gr], avg, mxb);
}
for (ch = 0; ch < cfg->channels_out; ++ch) {
gr_info *const cod_info = &gfc->l3_side.tt[gr][ch];
if (cod_info->block_type != SHORT_TYPE) { /* NORM, START or STOP type */
adjust = 1.28 / (1 + exp(3.5 - pe[gr][ch] / 300.)) - 0.05;
masking_lower_db = gfc->sv_qnt.mask_adjust - adjust;
}
else {
adjust = 2.56 / (1 + exp(3.5 - pe[gr][ch] / 300.)) - 0.14;
masking_lower_db = gfc->sv_qnt.mask_adjust_short - adjust;
}
gfc->sv_qnt.masking_lower = pow(10.0, masking_lower_db * 0.1);
init_outer_loop(gfc, cod_info);
bands[gr][ch] = calc_xmin(gfc, &ratio[gr][ch], cod_info, l3_xmin[gr][ch]);
if (bands[gr][ch])
analog_silence = 0;
min_bits[gr][ch] = 126;
bits += max_bits[gr][ch];
}
}
for (gr = 0; gr < cfg->mode_gr; gr++) {
for (ch = 0; ch < cfg->channels_out; ch++) {
if (bits > frameBits[cfg->vbr_max_bitrate_index] && bits > 0) {
max_bits[gr][ch] *= frameBits[cfg->vbr_max_bitrate_index];
max_bits[gr][ch] /= bits;
}
if (min_bits[gr][ch] > max_bits[gr][ch])
min_bits[gr][ch] = max_bits[gr][ch];
} /* for ch */
} /* for gr */
return analog_silence;
}
static void
bitpressure_strategy(lame_internal_flags const *gfc,
FLOAT l3_xmin[2][2][SFBMAX], const int min_bits[2][2], int max_bits[2][2])
{
SessionConfig_t const *const cfg = &gfc->cfg;
int gr, ch, sfb;
for (gr = 0; gr < cfg->mode_gr; gr++) {
for (ch = 0; ch < cfg->channels_out; ch++) {
gr_info const *const gi = &gfc->l3_side.tt[gr][ch];
FLOAT *pxmin = l3_xmin[gr][ch];
for (sfb = 0; sfb < gi->psy_lmax; sfb++)
*pxmin++ *= 1. + .029 * sfb * sfb / SBMAX_l / SBMAX_l;
if (gi->block_type == SHORT_TYPE) {
for (sfb = gi->sfb_smin; sfb < SBMAX_s; sfb++) {
*pxmin++ *= 1. + .029 * sfb * sfb / SBMAX_s / SBMAX_s;
*pxmin++ *= 1. + .029 * sfb * sfb / SBMAX_s / SBMAX_s;
*pxmin++ *= 1. + .029 * sfb * sfb / SBMAX_s / SBMAX_s;
}
}
max_bits[gr][ch] = Max(min_bits[gr][ch], 0.9 * max_bits[gr][ch]);
}
}
}
/************************************************************************
*
* VBR_iteration_loop()
*
* tries to find out how many bits are needed for each granule and channel
* to get an acceptable quantization. An appropriate bitrate will then be
* choosed for quantization. rh 8/99
*
* Robert Hegemann 2000-09-06 rewrite
*
************************************************************************/
void
VBR_old_iteration_loop(lame_internal_flags * gfc, const FLOAT pe[2][2],
const FLOAT ms_ener_ratio[2], const III_psy_ratio ratio[2][2])
{
SessionConfig_t const *const cfg = &gfc->cfg;
EncResult_t *const eov = &gfc->ov_enc;
FLOAT l3_xmin[2][2][SFBMAX];
FLOAT xrpow[576];
int bands[2][2];
int frameBits[15];
int used_bits;
int bits;
int min_bits[2][2], max_bits[2][2];
int mean_bits;
int ch, gr, analog_silence;
III_side_info_t *const l3_side = &gfc->l3_side;
analog_silence = VBR_old_prepare(gfc, pe, ms_ener_ratio, ratio,
l3_xmin, frameBits, min_bits, max_bits, bands);
/*---------------------------------*/
for (;;) {
/* quantize granules with lowest possible number of bits
*/
used_bits = 0;
for (gr = 0; gr < cfg->mode_gr; gr++) {
for (ch = 0; ch < cfg->channels_out; ch++) {
int ret;
gr_info *const cod_info = &l3_side->tt[gr][ch];
/* init_outer_loop sets up cod_info, scalefac and xrpow
*/
ret = init_xrpow(gfc, cod_info, xrpow);
if (ret == 0 || max_bits[gr][ch] == 0) {
/* xr contains no energy
* l3_enc, our encoding data, will be quantized to zero
*/
continue; /* with next channel */
}
VBR_encode_granule(gfc, cod_info, l3_xmin[gr][ch], xrpow,
ch, min_bits[gr][ch], max_bits[gr][ch]);
/* do the 'substep shaping'
*/
if (gfc->sv_qnt.substep_shaping & 1) {
trancate_smallspectrums(gfc, &l3_side->tt[gr][ch], l3_xmin[gr][ch], xrpow);
}
ret = cod_info->part2_3_length + cod_info->part2_length;
used_bits += ret;
} /* for ch */
} /* for gr */
/* find lowest bitrate able to hold used bits
*/
if (analog_silence && !cfg->enforce_min_bitrate)
/* we detected analog silence and the user did not specify
* any hard framesize limit, so start with smallest possible frame
*/
eov->bitrate_index = 1;
else
eov->bitrate_index = cfg->vbr_min_bitrate_index;
for (; eov->bitrate_index < cfg->vbr_max_bitrate_index; eov->bitrate_index++) {
if (used_bits <= frameBits[eov->bitrate_index])
break;
}
bits = ResvFrameBegin(gfc, &mean_bits);
if (used_bits <= bits)
break;
bitpressure_strategy(gfc, l3_xmin, (const int (*)[2])min_bits, max_bits);
} /* breaks adjusted */
/*--------------------------------------*/
for (gr = 0; gr < cfg->mode_gr; gr++) {
for (ch = 0; ch < cfg->channels_out; ch++) {
iteration_finish_one(gfc, gr, ch);
} /* for ch */
} /* for gr */
ResvFrameEnd(gfc, mean_bits);
}
static int
VBR_new_prepare(lame_internal_flags * gfc,
const FLOAT pe[2][2], const III_psy_ratio ratio[2][2],
FLOAT l3_xmin[2][2][SFBMAX], int frameBits[16], int max_bits[2][2],
int* max_resv)
{
SessionConfig_t const *const cfg = &gfc->cfg;
EncResult_t *const eov = &gfc->ov_enc;
int gr, ch;
int analog_silence = 1;
int avg, bits = 0;
int maximum_framebits;
if (!cfg->free_format) {
eov->bitrate_index = cfg->vbr_max_bitrate_index;
(void) ResvFrameBegin(gfc, &avg);
*max_resv = gfc->sv_enc.ResvMax;
get_framebits(gfc, frameBits);
maximum_framebits = frameBits[cfg->vbr_max_bitrate_index];
}
else {
eov->bitrate_index = 0;
maximum_framebits = ResvFrameBegin(gfc, &avg);
frameBits[0] = maximum_framebits;
*max_resv = gfc->sv_enc.ResvMax;
}
for (gr = 0; gr < cfg->mode_gr; gr++) {
(void) on_pe(gfc, pe, max_bits[gr], avg, gr, 0);
if (gfc->ov_enc.mode_ext == MPG_MD_MS_LR) {
ms_convert(&gfc->l3_side, gr);
}
for (ch = 0; ch < cfg->channels_out; ++ch) {
gr_info *const cod_info = &gfc->l3_side.tt[gr][ch];
gfc->sv_qnt.masking_lower = pow(10.0, gfc->sv_qnt.mask_adjust * 0.1);
init_outer_loop(gfc, cod_info);
if (0 != calc_xmin(gfc, &ratio[gr][ch], cod_info, l3_xmin[gr][ch]))
analog_silence = 0;
bits += max_bits[gr][ch];
}
}
for (gr = 0; gr < cfg->mode_gr; gr++) {
for (ch = 0; ch < cfg->channels_out; ch++) {
if (bits > maximum_framebits && bits > 0) {
max_bits[gr][ch] *= maximum_framebits;
max_bits[gr][ch] /= bits;
}
} /* for ch */
} /* for gr */
if (analog_silence) {
*max_resv = 0;
}
return analog_silence;
}
void
VBR_new_iteration_loop(lame_internal_flags * gfc, const FLOAT pe[2][2],
const FLOAT ms_ener_ratio[2], const III_psy_ratio ratio[2][2])
{
SessionConfig_t const *const cfg = &gfc->cfg;
EncResult_t *const eov = &gfc->ov_enc;
FLOAT l3_xmin[2][2][SFBMAX];
FLOAT xrpow[2][2][576];
int frameBits[15];
int used_bits;
int max_bits[2][2];
int ch, gr, analog_silence, pad;
III_side_info_t *const l3_side = &gfc->l3_side;
const FLOAT (*const_l3_xmin)[2][SFBMAX] = (const FLOAT (*)[2][SFBMAX])l3_xmin;
const FLOAT (*const_xrpow)[2][576] = (const FLOAT (*)[2][576])xrpow;
const int (*const_max_bits)[2] = (const int (*)[2])max_bits;
(void) ms_ener_ratio; /* not used */
memset(xrpow, 0, sizeof(xrpow));
analog_silence = VBR_new_prepare(gfc, pe, ratio, l3_xmin, frameBits, max_bits, &pad);
for (gr = 0; gr < cfg->mode_gr; gr++) {
for (ch = 0; ch < cfg->channels_out; ch++) {
gr_info *const cod_info = &l3_side->tt[gr][ch];
/* init_outer_loop sets up cod_info, scalefac and xrpow
*/
if (0 == init_xrpow(gfc, cod_info, xrpow[gr][ch])) {
max_bits[gr][ch] = 0; /* silent granule needs no bits */
}
} /* for ch */
} /* for gr */
/* quantize granules with lowest possible number of bits
*/
used_bits = VBR_encode_frame(gfc, const_xrpow, const_l3_xmin, const_max_bits);
if (!cfg->free_format) {
int i, j;
/* find lowest bitrate able to hold used bits
*/
if (analog_silence && !cfg->enforce_min_bitrate) {
/* we detected analog silence and the user did not specify
* any hard framesize limit, so start with smallest possible frame
*/
i = 1;
}
else {
i = cfg->vbr_min_bitrate_index;
}
for (; i < cfg->vbr_max_bitrate_index; i++) {
if (used_bits <= frameBits[i])
break;
}
if (i > cfg->vbr_max_bitrate_index) {
i = cfg->vbr_max_bitrate_index;
}
if (pad > 0) {
for (j = cfg->vbr_max_bitrate_index; j > i; --j) {
int const unused = frameBits[j] - used_bits;
if (unused <= pad)
break;
}
eov->bitrate_index = j;
}
else {
eov->bitrate_index = i;
}
}
else {
#if 0
static int mmm = 0;
int fff = getFramesize_kbps(gfc, used_bits);
int hhh = getFramesize_kbps(gfc, MAX_BITS_PER_GRANULE * cfg->mode_gr);
if (mmm < fff)
mmm = fff;
printf("demand=%3d kbps max=%3d kbps limit=%3d kbps\n", fff, mmm, hhh);
#endif
eov->bitrate_index = 0;
}
if (used_bits <= frameBits[eov->bitrate_index]) {
/* update Reservoire status */
int mean_bits, fullframebits;
fullframebits = ResvFrameBegin(gfc, &mean_bits);
assert(used_bits <= fullframebits);
for (gr = 0; gr < cfg->mode_gr; gr++) {
for (ch = 0; ch < cfg->channels_out; ch++) {
gr_info const *const cod_info = &l3_side->tt[gr][ch];
ResvAdjust(gfc, cod_info);
}
}
ResvFrameEnd(gfc, mean_bits);
}
else {
/* SHOULD NOT HAPPEN INTERNAL ERROR
*/
ERRORF(gfc, "INTERNAL ERROR IN VBR NEW CODE, please send bug report\n");
exit(-1);
}
}
/********************************************************************
*
* calc_target_bits()
*
* calculates target bits for ABR encoding
*
* mt 2000/05/31
*
********************************************************************/
static void
calc_target_bits(lame_internal_flags * gfc,
const FLOAT pe[2][2],
FLOAT const ms_ener_ratio[2],
int targ_bits[2][2], int *analog_silence_bits, int *max_frame_bits)
{
SessionConfig_t const *const cfg = &gfc->cfg;
EncResult_t *const eov = &gfc->ov_enc;
III_side_info_t const *const l3_side = &gfc->l3_side;
FLOAT res_factor;
int gr, ch, totbits, mean_bits;
int framesize = 576 * cfg->mode_gr;
eov->bitrate_index = cfg->vbr_max_bitrate_index;
*max_frame_bits = ResvFrameBegin(gfc, &mean_bits);
eov->bitrate_index = 1;
mean_bits = getframebits(gfc) - cfg->sideinfo_len * 8;
*analog_silence_bits = mean_bits / (cfg->mode_gr * cfg->channels_out);
mean_bits = cfg->vbr_avg_bitrate_kbps * framesize * 1000;
if (gfc->sv_qnt.substep_shaping & 1)
mean_bits *= 1.09;
mean_bits /= cfg->samplerate_out;
mean_bits -= cfg->sideinfo_len * 8;
mean_bits /= (cfg->mode_gr * cfg->channels_out);
/*
res_factor is the percentage of the target bitrate that should
be used on average. the remaining bits are added to the
bitreservoir and used for difficult to encode frames.
Since we are tracking the average bitrate, we should adjust
res_factor "on the fly", increasing it if the average bitrate
is greater than the requested bitrate, and decreasing it
otherwise. Reasonable ranges are from .9 to 1.0
Until we get the above suggestion working, we use the following
tuning:
compression ratio res_factor
5.5 (256kbps) 1.0 no need for bitreservoir
11 (128kbps) .93 7% held for reservoir
with linear interpolation for other values.
*/
res_factor = .93 + .07 * (11.0 - cfg->compression_ratio) / (11.0 - 5.5);
if (res_factor < .90)
res_factor = .90;
if (res_factor > 1.00)
res_factor = 1.00;
for (gr = 0; gr < cfg->mode_gr; gr++) {
int sum = 0;
for (ch = 0; ch < cfg->channels_out; ch++) {
targ_bits[gr][ch] = res_factor * mean_bits;
if (pe[gr][ch] > 700) {
int add_bits = (pe[gr][ch] - 700) / 1.4;
gr_info const *const cod_info = &l3_side->tt[gr][ch];
targ_bits[gr][ch] = res_factor * mean_bits;
/* short blocks use a little extra, no matter what the pe */
if (cod_info->block_type == SHORT_TYPE) {
if (add_bits < mean_bits / 2)
add_bits = mean_bits / 2;
}
/* at most increase bits by 1.5*average */
if (add_bits > mean_bits * 3 / 2)
add_bits = mean_bits * 3 / 2;
else if (add_bits < 0)
add_bits = 0;
targ_bits[gr][ch] += add_bits;
}
if (targ_bits[gr][ch] > MAX_BITS_PER_CHANNEL) {
targ_bits[gr][ch] = MAX_BITS_PER_CHANNEL;
}
sum += targ_bits[gr][ch];
} /* for ch */
if (sum > MAX_BITS_PER_GRANULE) {
for (ch = 0; ch < cfg->channels_out; ++ch) {
targ_bits[gr][ch] *= MAX_BITS_PER_GRANULE;
targ_bits[gr][ch] /= sum;
}
}
} /* for gr */
if (gfc->ov_enc.mode_ext == MPG_MD_MS_LR)
for (gr = 0; gr < cfg->mode_gr; gr++) {
reduce_side(targ_bits[gr], ms_ener_ratio[gr], mean_bits * cfg->channels_out,
MAX_BITS_PER_GRANULE);
}
/* sum target bits
*/
totbits = 0;
for (gr = 0; gr < cfg->mode_gr; gr++) {
for (ch = 0; ch < cfg->channels_out; ch++) {
if (targ_bits[gr][ch] > MAX_BITS_PER_CHANNEL)
targ_bits[gr][ch] = MAX_BITS_PER_CHANNEL;
totbits += targ_bits[gr][ch];
}
}
/* repartion target bits if needed
*/
if (totbits > *max_frame_bits && totbits > 0) {
for (gr = 0; gr < cfg->mode_gr; gr++) {
for (ch = 0; ch < cfg->channels_out; ch++) {
targ_bits[gr][ch] *= *max_frame_bits;
targ_bits[gr][ch] /= totbits;
}
}
}
}
/********************************************************************
*
* ABR_iteration_loop()
*
* encode a frame with a disired average bitrate
*
* mt 2000/05/31
*
********************************************************************/
void
ABR_iteration_loop(lame_internal_flags * gfc, const FLOAT pe[2][2],
const FLOAT ms_ener_ratio[2], const III_psy_ratio ratio[2][2])
{
SessionConfig_t const *const cfg = &gfc->cfg;
EncResult_t *const eov = &gfc->ov_enc;
FLOAT l3_xmin[SFBMAX];
FLOAT xrpow[576];
int targ_bits[2][2];
int mean_bits, max_frame_bits;
int ch, gr, ath_over;
int analog_silence_bits;
gr_info *cod_info;
III_side_info_t *const l3_side = &gfc->l3_side;
mean_bits = 0;
calc_target_bits(gfc, pe, ms_ener_ratio, targ_bits, &analog_silence_bits, &max_frame_bits);
/* encode granules
*/
for (gr = 0; gr < cfg->mode_gr; gr++) {
if (gfc->ov_enc.mode_ext == MPG_MD_MS_LR) {
ms_convert(&gfc->l3_side, gr);
}
for (ch = 0; ch < cfg->channels_out; ch++) {
FLOAT adjust, masking_lower_db;
cod_info = &l3_side->tt[gr][ch];
if (cod_info->block_type != SHORT_TYPE) { /* NORM, START or STOP type */
/* adjust = 1.28/(1+exp(3.5-pe[gr][ch]/300.))-0.05; */
adjust = 0;
masking_lower_db = gfc->sv_qnt.mask_adjust - adjust;
}
else {
/* adjust = 2.56/(1+exp(3.5-pe[gr][ch]/300.))-0.14; */
adjust = 0;
masking_lower_db = gfc->sv_qnt.mask_adjust_short - adjust;
}
gfc->sv_qnt.masking_lower = pow(10.0, masking_lower_db * 0.1);
/* cod_info, scalefac and xrpow get initialized in init_outer_loop
*/
init_outer_loop(gfc, cod_info);
if (init_xrpow(gfc, cod_info, xrpow)) {
/* xr contains energy we will have to encode
* calculate the masking abilities
* find some good quantization in outer_loop
*/
ath_over = calc_xmin(gfc, &ratio[gr][ch], cod_info, l3_xmin);
if (0 == ath_over) /* analog silence */
targ_bits[gr][ch] = analog_silence_bits;
(void) outer_loop(gfc, cod_info, l3_xmin, xrpow, ch, targ_bits[gr][ch]);
}
iteration_finish_one(gfc, gr, ch);
} /* ch */
} /* gr */
/* find a bitrate which can refill the resevoir to positive size.
*/
for (eov->bitrate_index = cfg->vbr_min_bitrate_index;
eov->bitrate_index <= cfg->vbr_max_bitrate_index; eov->bitrate_index++) {
if (ResvFrameBegin(gfc, &mean_bits) >= 0)
break;
}
assert(eov->bitrate_index <= cfg->vbr_max_bitrate_index);
ResvFrameEnd(gfc, mean_bits);
}
/************************************************************************
*
* CBR_iteration_loop()
*
* author/date??
*
* encodes one frame of MP3 data with constant bitrate
*
************************************************************************/
void
CBR_iteration_loop(lame_internal_flags * gfc, const FLOAT pe[2][2],
const FLOAT ms_ener_ratio[2], const III_psy_ratio ratio[2][2])
{
SessionConfig_t const *const cfg = &gfc->cfg;
FLOAT l3_xmin[SFBMAX];
FLOAT xrpow[576];
int targ_bits[2];
int mean_bits, max_bits;
int gr, ch;
III_side_info_t *const l3_side = &gfc->l3_side;
gr_info *cod_info;
(void) ResvFrameBegin(gfc, &mean_bits);
/* quantize! */
for (gr = 0; gr < cfg->mode_gr; gr++) {
/* calculate needed bits
*/
max_bits = on_pe(gfc, pe, targ_bits, mean_bits, gr, gr);
if (gfc->ov_enc.mode_ext == MPG_MD_MS_LR) {
ms_convert(&gfc->l3_side, gr);
reduce_side(targ_bits, ms_ener_ratio[gr], mean_bits, max_bits);
}
for (ch = 0; ch < cfg->channels_out; ch++) {
FLOAT adjust, masking_lower_db;
cod_info = &l3_side->tt[gr][ch];
if (cod_info->block_type != SHORT_TYPE) { /* NORM, START or STOP type */
/* adjust = 1.28/(1+exp(3.5-pe[gr][ch]/300.))-0.05; */
adjust = 0;
masking_lower_db = gfc->sv_qnt.mask_adjust - adjust;
}
else {
/* adjust = 2.56/(1+exp(3.5-pe[gr][ch]/300.))-0.14; */
adjust = 0;
masking_lower_db = gfc->sv_qnt.mask_adjust_short - adjust;
}
gfc->sv_qnt.masking_lower = pow(10.0, masking_lower_db * 0.1);
/* init_outer_loop sets up cod_info, scalefac and xrpow
*/
init_outer_loop(gfc, cod_info);
if (init_xrpow(gfc, cod_info, xrpow)) {
/* xr contains energy we will have to encode
* calculate the masking abilities
* find some good quantization in outer_loop
*/
(void) calc_xmin(gfc, &ratio[gr][ch], cod_info, l3_xmin);
(void) outer_loop(gfc, cod_info, l3_xmin, xrpow, ch, targ_bits[ch]);
}
iteration_finish_one(gfc, gr, ch);
assert(cod_info->part2_3_length <= MAX_BITS_PER_CHANNEL);
assert(cod_info->part2_3_length <= targ_bits[ch]);
} /* for ch */
} /* for gr */
ResvFrameEnd(gfc, mean_bits);
}
Reference in New Issue View Git Blame Copy Permalink