54f5fd22 |
/*
* Simple free lossless/lossy audio codec
* Copyright (c) 2004 Alex Beregszaszi
* |
b78e7197 |
* This file is part of FFmpeg.
*
* FFmpeg is free software; you can redistribute it and/or |
54f5fd22 |
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either |
b78e7197 |
* version 2.1 of the License, or (at your option) any later version. |
54f5fd22 |
* |
b78e7197 |
* FFmpeg is distributed in the hope that it will be useful, |
54f5fd22 |
* 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 |
b78e7197 |
* License along with FFmpeg; if not, write to the Free Software |
5509bffa |
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
54f5fd22 |
*/
#include "avcodec.h" |
9106a698 |
#include "get_bits.h" |
54f5fd22 |
#include "golomb.h" |
a44cbc1c |
#include "internal.h" |
6026a5ad |
#include "rangecoder.h"
|
54f5fd22 |
/** |
ba87f080 |
* @file |
54f5fd22 |
* Simple free lossless/lossy audio codec
* Based on Paul Francis Harrison's Bonk (http://www.logarithmic.net/pfh/bonk)
* Written and designed by Alex Beregszaszi
*
* TODO:
* - CABAC put/get_symbol
* - independent quantizer for channels
* - >2 channels support
* - more decorrelation types
* - more tap_quant tests
* - selectable intlist writers/readers (bonk-style, golomb, cabac)
*/
#define MAX_CHANNELS 2
|
cc078b9e |
#define MID_SIDE 0
#define LEFT_SIDE 1
#define RIGHT_SIDE 2
|
54f5fd22 |
typedef struct SonicContext { |
4c32629b |
int version; |
c61daa68 |
int minor_version; |
cc078b9e |
int lossless, decorrelation; |
115329f1 |
|
54f5fd22 |
int num_taps, downsampling;
double quantization; |
115329f1 |
|
54f5fd22 |
int channels, samplerate, block_align, frame_size;
int *tap_quant;
int *int_samples;
int *coded_samples[MAX_CHANNELS];
// for encoding
int *tail;
int tail_size;
int *window;
int window_size;
// for decoding
int *predictor_k;
int *predictor_state[MAX_CHANNELS];
} SonicContext;
|
bb270c08 |
#define LATTICE_SHIFT 10
#define SAMPLE_SHIFT 4
#define LATTICE_FACTOR (1 << LATTICE_SHIFT)
#define SAMPLE_FACTOR (1 << SAMPLE_SHIFT) |
54f5fd22 |
|
bb270c08 |
#define BASE_QUANT 0.6
#define RATE_VARIATION 3.0 |
54f5fd22 |
static inline int shift(int a,int b)
{
return (a+(1<<(b-1))) >> b;
}
static inline int shift_down(int a,int b)
{ |
8689ee0e |
return (a>>b)+(a<0); |
54f5fd22 |
}
|
6026a5ad |
static av_always_inline av_flatten void put_symbol(RangeCoder *c, uint8_t *state, int v, int is_signed, uint64_t rc_stat[256][2], uint64_t rc_stat2[32][2]){
int i;
#define put_rac(C,S,B) \
do{\
if(rc_stat){\
rc_stat[*(S)][B]++;\
rc_stat2[(S)-state][B]++;\
}\
put_rac(C,S,B);\
}while(0)
if(v){
const int a= FFABS(v);
const int e= av_log2(a);
put_rac(c, state+0, 0);
if(e<=9){
for(i=0; i<e; i++){
put_rac(c, state+1+i, 1); //1..10
}
put_rac(c, state+1+i, 0);
for(i=e-1; i>=0; i--){
put_rac(c, state+22+i, (a>>i)&1); //22..31
}
if(is_signed)
put_rac(c, state+11 + e, v < 0); //11..21
}else{
for(i=0; i<e; i++){
put_rac(c, state+1+FFMIN(i,9), 1); //1..10
}
put_rac(c, state+1+9, 0);
for(i=e-1; i>=0; i--){
put_rac(c, state+22+FFMIN(i,9), (a>>i)&1); //22..31
}
if(is_signed)
put_rac(c, state+11 + 10, v < 0); //11..21
}
}else{
put_rac(c, state+0, 1);
}
#undef put_rac
}
static inline av_flatten int get_symbol(RangeCoder *c, uint8_t *state, int is_signed){
if(get_rac(c, state+0))
return 0;
else{
int i, e, a;
e= 0;
while(get_rac(c, state+1 + FFMIN(e,9))){ //1..10
e++;
}
a= 1;
for(i=e-1; i>=0; i--){
a += a + get_rac(c, state+22 + FFMIN(i,9)); //22..31
}
e= -(is_signed && get_rac(c, state+11 + FFMIN(e, 10))); //11..21
return (a^e)-e;
}
}
|
54f5fd22 |
#if 1 |
6026a5ad |
static inline int intlist_write(RangeCoder *c, uint8_t *state, int *buf, int entries, int base_2_part)
{
int i;
for (i = 0; i < entries; i++)
put_symbol(c, state, buf[i], 1, NULL, NULL);
return 1;
}
static inline int intlist_read(RangeCoder *c, uint8_t *state, int *buf, int entries, int base_2_part)
{
int i;
for (i = 0; i < entries; i++)
buf[i] = get_symbol(c, state, 1);
return 1;
}
#elif 1 |
54f5fd22 |
static inline int intlist_write(PutBitContext *pb, int *buf, int entries, int base_2_part)
{
int i;
for (i = 0; i < entries; i++) |
bb270c08 |
set_se_golomb(pb, buf[i]); |
54f5fd22 |
return 1;
}
static inline int intlist_read(GetBitContext *gb, int *buf, int entries, int base_2_part)
{
int i; |
115329f1 |
|
54f5fd22 |
for (i = 0; i < entries; i++) |
bb270c08 |
buf[i] = get_se_golomb(gb); |
54f5fd22 |
return 1;
}
#else
#define ADAPT_LEVEL 8
static int bits_to_store(uint64_t x)
{
int res = 0; |
115329f1 |
|
54f5fd22 |
while(x)
{ |
bb270c08 |
res++;
x >>= 1; |
54f5fd22 |
}
return res;
}
static void write_uint_max(PutBitContext *pb, unsigned int value, unsigned int max)
{
int i, bits;
if (!max) |
bb270c08 |
return; |
54f5fd22 |
bits = bits_to_store(max);
for (i = 0; i < bits-1; i++) |
bb270c08 |
put_bits(pb, 1, value & (1 << i)); |
54f5fd22 |
if ( (value | (1 << (bits-1))) <= max) |
bb270c08 |
put_bits(pb, 1, value & (1 << (bits-1))); |
54f5fd22 |
}
static unsigned int read_uint_max(GetBitContext *gb, int max)
{
int i, bits, value = 0; |
115329f1 |
|
54f5fd22 |
if (!max) |
bb270c08 |
return 0; |
54f5fd22 |
bits = bits_to_store(max);
for (i = 0; i < bits-1; i++) |
bb270c08 |
if (get_bits1(gb))
value += 1 << i; |
54f5fd22 |
if ( (value | (1<<(bits-1))) <= max) |
bb270c08 |
if (get_bits1(gb))
value += 1 << (bits-1); |
54f5fd22 |
return value;
}
static int intlist_write(PutBitContext *pb, int *buf, int entries, int base_2_part)
{
int i, j, x = 0, low_bits = 0, max = 0;
int step = 256, pos = 0, dominant = 0, any = 0;
int *copy, *bits;
|
bcb42fb6 |
copy = av_calloc(entries, sizeof(*copy)); |
54f5fd22 |
if (!copy) |
12de2933 |
return AVERROR(ENOMEM); |
115329f1 |
|
54f5fd22 |
if (base_2_part)
{ |
bb270c08 |
int energy = 0; |
115329f1 |
|
bb270c08 |
for (i = 0; i < entries; i++)
energy += abs(buf[i]); |
115329f1 |
|
bb270c08 |
low_bits = bits_to_store(energy / (entries * 2));
if (low_bits > 15)
low_bits = 15; |
115329f1 |
|
bb270c08 |
put_bits(pb, 4, low_bits); |
54f5fd22 |
} |
115329f1 |
|
54f5fd22 |
for (i = 0; i < entries; i++)
{ |
bb270c08 |
put_bits(pb, low_bits, abs(buf[i]));
copy[i] = abs(buf[i]) >> low_bits;
if (copy[i] > max)
max = abs(copy[i]); |
54f5fd22 |
}
|
bcb42fb6 |
bits = av_calloc(entries*max, sizeof(*bits)); |
54f5fd22 |
if (!bits)
{ |
ec4d761c |
av_free(copy); |
12de2933 |
return AVERROR(ENOMEM); |
54f5fd22 |
} |
115329f1 |
|
54f5fd22 |
for (i = 0; i <= max; i++)
{ |
bb270c08 |
for (j = 0; j < entries; j++)
if (copy[j] >= i)
bits[x++] = copy[j] > i; |
54f5fd22 |
}
// store bitstream
while (pos < x)
{ |
bb270c08 |
int steplet = step >> 8; |
115329f1 |
|
bb270c08 |
if (pos + steplet > x)
steplet = x - pos; |
115329f1 |
|
bb270c08 |
for (i = 0; i < steplet; i++)
if (bits[i+pos] != dominant)
any = 1; |
115329f1 |
|
bb270c08 |
put_bits(pb, 1, any); |
115329f1 |
|
bb270c08 |
if (!any)
{
pos += steplet;
step += step / ADAPT_LEVEL;
}
else
{
int interloper = 0; |
115329f1 |
|
bb270c08 |
while (((pos + interloper) < x) && (bits[pos + interloper] == dominant))
interloper++; |
54f5fd22 |
|
bb270c08 |
// note change
write_uint_max(pb, interloper, (step >> 8) - 1); |
115329f1 |
|
bb270c08 |
pos += interloper + 1;
step -= step / ADAPT_LEVEL;
} |
115329f1 |
|
bb270c08 |
if (step < 256)
{
step = 65536 / step;
dominant = !dominant;
} |
54f5fd22 |
} |
115329f1 |
|
54f5fd22 |
// store signs
for (i = 0; i < entries; i++) |
bb270c08 |
if (buf[i])
put_bits(pb, 1, buf[i] < 0); |
54f5fd22 |
|
ec4d761c |
av_free(bits);
av_free(copy); |
54f5fd22 |
return 0;
}
static int intlist_read(GetBitContext *gb, int *buf, int entries, int base_2_part)
{
int i, low_bits = 0, x = 0;
int n_zeros = 0, step = 256, dominant = 0;
int pos = 0, level = 0; |
bcb42fb6 |
int *bits = av_calloc(entries, sizeof(*bits)); |
54f5fd22 |
if (!bits) |
12de2933 |
return AVERROR(ENOMEM); |
115329f1 |
|
54f5fd22 |
if (base_2_part)
{ |
bb270c08 |
low_bits = get_bits(gb, 4); |
54f5fd22 |
|
bb270c08 |
if (low_bits)
for (i = 0; i < entries; i++)
buf[i] = get_bits(gb, low_bits); |
54f5fd22 |
}
// av_log(NULL, AV_LOG_INFO, "entries: %d, low bits: %d\n", entries, low_bits);
while (n_zeros < entries)
{ |
bb270c08 |
int steplet = step >> 8; |
115329f1 |
|
bb270c08 |
if (!get_bits1(gb))
{
for (i = 0; i < steplet; i++)
bits[x++] = dominant; |
115329f1 |
|
bb270c08 |
if (!dominant)
n_zeros += steplet; |
115329f1 |
|
bb270c08 |
step += step / ADAPT_LEVEL;
}
else
{
int actual_run = read_uint_max(gb, steplet-1); |
115329f1 |
|
bb270c08 |
// av_log(NULL, AV_LOG_INFO, "actual run: %d\n", actual_run); |
115329f1 |
|
bb270c08 |
for (i = 0; i < actual_run; i++)
bits[x++] = dominant; |
115329f1 |
|
bb270c08 |
bits[x++] = !dominant; |
115329f1 |
|
bb270c08 |
if (!dominant)
n_zeros += actual_run;
else
n_zeros++; |
115329f1 |
|
bb270c08 |
step -= step / ADAPT_LEVEL;
} |
115329f1 |
|
bb270c08 |
if (step < 256)
{
step = 65536 / step;
dominant = !dominant;
} |
54f5fd22 |
} |
115329f1 |
|
54f5fd22 |
// reconstruct unsigned values
n_zeros = 0;
for (i = 0; n_zeros < entries; i++)
{ |
bb270c08 |
while(1)
{
if (pos >= entries)
{
pos = 0;
level += 1 << low_bits;
} |
115329f1 |
|
bb270c08 |
if (buf[pos] >= level)
break; |
115329f1 |
|
bb270c08 |
pos++;
} |
115329f1 |
|
bb270c08 |
if (bits[i])
buf[pos] += 1 << low_bits;
else
n_zeros++; |
115329f1 |
|
bb270c08 |
pos++; |
54f5fd22 |
} |
ec4d761c |
av_free(bits); |
115329f1 |
|
54f5fd22 |
// read signs
for (i = 0; i < entries; i++) |
bb270c08 |
if (buf[i] && get_bits1(gb))
buf[i] = -buf[i]; |
54f5fd22 |
// av_log(NULL, AV_LOG_INFO, "zeros: %d pos: %d\n", n_zeros, pos);
return 0;
}
#endif
static void predictor_init_state(int *k, int *state, int order)
{
int i;
for (i = order-2; i >= 0; i--)
{ |
bb270c08 |
int j, p, x = state[i]; |
54f5fd22 |
|
bb270c08 |
for (j = 0, p = i+1; p < order; j++,p++)
{
int tmp = x + shift_down(k[j] * state[p], LATTICE_SHIFT);
state[p] += shift_down(k[j]*x, LATTICE_SHIFT);
x = tmp;
} |
54f5fd22 |
}
}
static int predictor_calc_error(int *k, int *state, int order, int error)
{
int i, x = error - shift_down(k[order-1] * state[order-1], LATTICE_SHIFT);
#if 1
int *k_ptr = &(k[order-2]), |
bb270c08 |
*state_ptr = &(state[order-2]); |
54f5fd22 |
for (i = order-2; i >= 0; i--, k_ptr--, state_ptr--)
{ |
bb270c08 |
int k_value = *k_ptr, state_value = *state_ptr;
x -= shift_down(k_value * state_value, LATTICE_SHIFT);
state_ptr[1] = state_value + shift_down(k_value * x, LATTICE_SHIFT); |
54f5fd22 |
}
#else
for (i = order-2; i >= 0; i--)
{ |
bb270c08 |
x -= shift_down(k[i] * state[i], LATTICE_SHIFT);
state[i+1] = state[i] + shift_down(k[i] * x, LATTICE_SHIFT); |
54f5fd22 |
}
#endif
|
115329f1 |
// don't drift too far, to avoid overflows |
54f5fd22 |
if (x > (SAMPLE_FACTOR<<16)) x = (SAMPLE_FACTOR<<16);
if (x < -(SAMPLE_FACTOR<<16)) x = -(SAMPLE_FACTOR<<16);
state[0] = x;
return x;
}
|
b250f9c6 |
#if CONFIG_SONIC_ENCODER || CONFIG_SONIC_LS_ENCODER |
54f5fd22 |
// Heavily modified Levinson-Durbin algorithm which
// copes better with quantization, and calculates the
// actual whitened result as it goes.
static void modified_levinson_durbin(int *window, int window_entries, |
bb270c08 |
int *out, int out_entries, int channels, int *tap_quant) |
54f5fd22 |
{
int i; |
bcb42fb6 |
int *state = av_calloc(window_entries, sizeof(*state)); |
115329f1 |
|
54f5fd22 |
memcpy(state, window, 4* window_entries); |
115329f1 |
|
54f5fd22 |
for (i = 0; i < out_entries; i++)
{ |
bb270c08 |
int step = (i+1)*channels, k, j;
double xx = 0.0, xy = 0.0; |
54f5fd22 |
#if 1 |
694c2d1a |
int *x_ptr = &(window[step]);
int *state_ptr = &(state[0]); |
bb270c08 |
j = window_entries - step; |
ddefb80c |
for (;j>0;j--,x_ptr++,state_ptr++) |
bb270c08 |
{ |
694c2d1a |
double x_value = *x_ptr;
double state_value = *state_ptr; |
bb270c08 |
xx += state_value*state_value;
xy += x_value*state_value;
} |
54f5fd22 |
#else |
bb270c08 |
for (j = 0; j <= (window_entries - step); j++);
{ |
694c2d1a |
double stepval = window[step+j];
double stateval = window[j]; |
bb270c08 |
// xx += (double)window[j]*(double)window[j];
// xy += (double)window[step+j]*(double)window[j];
xx += stateval*stateval;
xy += stepval*stateval;
} |
54f5fd22 |
#endif |
bb270c08 |
if (xx == 0.0)
k = 0;
else
k = (int)(floor(-xy/xx * (double)LATTICE_FACTOR / (double)(tap_quant[i]) + 0.5)); |
115329f1 |
|
bb270c08 |
if (k > (LATTICE_FACTOR/tap_quant[i]))
k = LATTICE_FACTOR/tap_quant[i];
if (-k > (LATTICE_FACTOR/tap_quant[i]))
k = -(LATTICE_FACTOR/tap_quant[i]); |
115329f1 |
|
bb270c08 |
out[i] = k;
k *= tap_quant[i]; |
54f5fd22 |
#if 1 |
bb270c08 |
x_ptr = &(window[step]);
state_ptr = &(state[0]);
j = window_entries - step; |
ddefb80c |
for (;j>0;j--,x_ptr++,state_ptr++) |
bb270c08 |
{ |
694c2d1a |
int x_value = *x_ptr;
int state_value = *state_ptr; |
bb270c08 |
*x_ptr = x_value + shift_down(k*state_value,LATTICE_SHIFT);
*state_ptr = state_value + shift_down(k*x_value, LATTICE_SHIFT);
} |
115329f1 |
#else |
bb270c08 |
for (j=0; j <= (window_entries - step); j++)
{ |
694c2d1a |
int stepval = window[step+j];
int stateval=state[j]; |
bb270c08 |
window[step+j] += shift_down(k * stateval, LATTICE_SHIFT);
state[j] += shift_down(k * stepval, LATTICE_SHIFT);
} |
54f5fd22 |
#endif
} |
115329f1 |
|
54f5fd22 |
av_free(state);
}
static inline int code_samplerate(int samplerate)
{
switch (samplerate)
{ |
bb270c08 |
case 44100: return 0;
case 22050: return 1;
case 11025: return 2;
case 96000: return 3;
case 48000: return 4;
case 32000: return 5;
case 24000: return 6;
case 16000: return 7;
case 8000: return 8; |
54f5fd22 |
} |
12de2933 |
return AVERROR(EINVAL); |
54f5fd22 |
}
|
98a6fff9 |
static av_cold int sonic_encode_init(AVCodecContext *avctx) |
54f5fd22 |
{
SonicContext *s = avctx->priv_data;
PutBitContext pb; |
4c32629b |
int i; |
54f5fd22 |
|
6026a5ad |
s->version = 2;
|
54f5fd22 |
if (avctx->channels > MAX_CHANNELS) |
ef859ca3 |
{ |
bb270c08 |
av_log(avctx, AV_LOG_ERROR, "Only mono and stereo streams are supported by now\n"); |
12de2933 |
return AVERROR(EINVAL); /* only stereo or mono for now */ |
ef859ca3 |
} |
54f5fd22 |
if (avctx->channels == 2) |
bb270c08 |
s->decorrelation = MID_SIDE; |
85d7f546 |
else
s->decorrelation = 3; |
ef859ca3 |
|
7a72695c |
if (avctx->codec->id == AV_CODEC_ID_SONIC_LS) |
54f5fd22 |
{ |
bb270c08 |
s->lossless = 1;
s->num_taps = 32;
s->downsampling = 1;
s->quantization = 0.0; |
54f5fd22 |
}
else
{ |
bb270c08 |
s->num_taps = 128;
s->downsampling = 2;
s->quantization = 1.0; |
54f5fd22 |
}
// max tap 2048 |
69d0a292 |
if (s->num_taps < 32 || s->num_taps > 1024 || s->num_taps % 32) { |
bb270c08 |
av_log(avctx, AV_LOG_ERROR, "Invalid number of taps\n"); |
12de2933 |
return AVERROR_INVALIDDATA; |
54f5fd22 |
}
// generate taps |
bcb42fb6 |
s->tap_quant = av_calloc(s->num_taps, sizeof(*s->tap_quant)); |
54f5fd22 |
for (i = 0; i < s->num_taps; i++) |
730e07f1 |
s->tap_quant[i] = ff_sqrt(i+1); |
54f5fd22 |
s->channels = avctx->channels;
s->samplerate = avctx->sample_rate;
|
081a7f3e |
s->block_align = 2048LL*s->samplerate/(44100*s->downsampling); |
54f5fd22 |
s->frame_size = s->channels*s->block_align*s->downsampling;
|
0b2ac679 |
s->tail_size = s->num_taps*s->channels; |
bcb42fb6 |
s->tail = av_calloc(s->tail_size, sizeof(*s->tail)); |
54f5fd22 |
if (!s->tail) |
12de2933 |
return AVERROR(ENOMEM); |
54f5fd22 |
|
bcb42fb6 |
s->predictor_k = av_calloc(s->num_taps, sizeof(*s->predictor_k) ); |
54f5fd22 |
if (!s->predictor_k) |
12de2933 |
return AVERROR(ENOMEM); |
54f5fd22 |
for (i = 0; i < s->channels; i++)
{ |
bcb42fb6 |
s->coded_samples[i] = av_calloc(s->block_align, sizeof(**s->coded_samples)); |
bb270c08 |
if (!s->coded_samples[i]) |
12de2933 |
return AVERROR(ENOMEM); |
54f5fd22 |
} |
115329f1 |
|
bcb42fb6 |
s->int_samples = av_calloc(s->frame_size, sizeof(*s->int_samples)); |
54f5fd22 |
s->window_size = ((2*s->tail_size)+s->frame_size); |
bcb42fb6 |
s->window = av_calloc(s->window_size, sizeof(*s->window)); |
54f5fd22 |
if (!s->window) |
12de2933 |
return AVERROR(ENOMEM); |
54f5fd22 |
avctx->extradata = av_mallocz(16);
if (!avctx->extradata) |
12de2933 |
return AVERROR(ENOMEM); |
54f5fd22 |
init_put_bits(&pb, avctx->extradata, 16*8);
|
4c32629b |
put_bits(&pb, 2, s->version); // version |
c61daa68 |
if (s->version >= 1) |
54f5fd22 |
{ |
c61daa68 |
if (s->version >= 2) {
put_bits(&pb, 8, s->version);
put_bits(&pb, 8, s->minor_version);
} |
bb270c08 |
put_bits(&pb, 2, s->channels);
put_bits(&pb, 4, code_samplerate(s->samplerate)); |
54f5fd22 |
}
put_bits(&pb, 1, s->lossless);
if (!s->lossless) |
bb270c08 |
put_bits(&pb, 3, SAMPLE_SHIFT); // XXX FIXME: sample precision |
cc078b9e |
put_bits(&pb, 2, s->decorrelation); |
54f5fd22 |
put_bits(&pb, 2, s->downsampling);
put_bits(&pb, 5, (s->num_taps >> 5)-1); // 32..1024
put_bits(&pb, 1, 0); // XXX FIXME: no custom tap quant table
flush_put_bits(&pb);
avctx->extradata_size = put_bits_count(&pb)/8;
|
c61daa68 |
av_log(avctx, AV_LOG_INFO, "Sonic: ver: %d.%d ls: %d dr: %d taps: %d block: %d frame: %d downsamp: %d\n",
s->version, s->minor_version, s->lossless, s->decorrelation, s->num_taps, s->block_align, s->frame_size, s->downsampling); |
54f5fd22 |
avctx->frame_size = s->block_align*s->downsampling;
return 0;
}
|
98a6fff9 |
static av_cold int sonic_encode_close(AVCodecContext *avctx) |
54f5fd22 |
{
SonicContext *s = avctx->priv_data;
int i;
for (i = 0; i < s->channels; i++) |
9375f500 |
av_freep(&s->coded_samples[i]); |
54f5fd22 |
|
9375f500 |
av_freep(&s->predictor_k);
av_freep(&s->tail);
av_freep(&s->tap_quant);
av_freep(&s->window);
av_freep(&s->int_samples); |
54f5fd22 |
return 0;
}
|
a44cbc1c |
static int sonic_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
const AVFrame *frame, int *got_packet_ptr) |
54f5fd22 |
{
SonicContext *s = avctx->priv_data; |
6026a5ad |
RangeCoder c; |
54f5fd22 |
int i, j, ch, quant = 0, x = 0; |
a44cbc1c |
int ret;
const short *samples = (const int16_t*)frame->data[0]; |
6026a5ad |
uint8_t state[32]; |
54f5fd22 |
|
bcaf64b6 |
if ((ret = ff_alloc_packet2(avctx, avpkt, s->frame_size * 5 + 1000)) < 0) |
a44cbc1c |
return ret;
|
6026a5ad |
ff_init_range_encoder(&c, avpkt->data, avpkt->size);
ff_build_rac_states(&c, 0.05*(1LL<<32), 256-8);
memset(state, 128, sizeof(state)); |
54f5fd22 |
// short -> internal
for (i = 0; i < s->frame_size; i++) |
bb270c08 |
s->int_samples[i] = samples[i]; |
54f5fd22 |
if (!s->lossless) |
bb270c08 |
for (i = 0; i < s->frame_size; i++)
s->int_samples[i] = s->int_samples[i] << SAMPLE_SHIFT; |
54f5fd22 |
|
cc078b9e |
switch(s->decorrelation)
{ |
bb270c08 |
case MID_SIDE:
for (i = 0; i < s->frame_size; i += s->channels)
{
s->int_samples[i] += s->int_samples[i+1];
s->int_samples[i+1] -= shift(s->int_samples[i], 1);
}
break;
case LEFT_SIDE:
for (i = 0; i < s->frame_size; i += s->channels)
s->int_samples[i+1] -= s->int_samples[i];
break;
case RIGHT_SIDE:
for (i = 0; i < s->frame_size; i += s->channels)
s->int_samples[i] -= s->int_samples[i+1];
break; |
cc078b9e |
} |
54f5fd22 |
memset(s->window, 0, 4* s->window_size); |
115329f1 |
|
54f5fd22 |
for (i = 0; i < s->tail_size; i++) |
bb270c08 |
s->window[x++] = s->tail[i]; |
54f5fd22 |
for (i = 0; i < s->frame_size; i++) |
bb270c08 |
s->window[x++] = s->int_samples[i]; |
115329f1 |
|
54f5fd22 |
for (i = 0; i < s->tail_size; i++) |
bb270c08 |
s->window[x++] = 0; |
54f5fd22 |
for (i = 0; i < s->tail_size; i++) |
bb270c08 |
s->tail[i] = s->int_samples[s->frame_size - s->tail_size + i]; |
54f5fd22 |
// generate taps
modified_levinson_durbin(s->window, s->window_size, |
bb270c08 |
s->predictor_k, s->num_taps, s->channels, s->tap_quant); |
6026a5ad |
if ((ret = intlist_write(&c, state, s->predictor_k, s->num_taps, 0)) < 0) |
12de2933 |
return ret; |
54f5fd22 |
for (ch = 0; ch < s->channels; ch++)
{ |
bb270c08 |
x = s->tail_size+ch;
for (i = 0; i < s->block_align; i++)
{
int sum = 0;
for (j = 0; j < s->downsampling; j++, x += s->channels)
sum += s->window[x];
s->coded_samples[ch][i] = sum;
} |
54f5fd22 |
} |
115329f1 |
// simple rate control code |
54f5fd22 |
if (!s->lossless)
{ |
bb270c08 |
double energy1 = 0.0, energy2 = 0.0;
for (ch = 0; ch < s->channels; ch++)
{
for (i = 0; i < s->block_align; i++)
{
double sample = s->coded_samples[ch][i];
energy2 += sample*sample;
energy1 += fabs(sample);
}
} |
115329f1 |
|
bb270c08 |
energy2 = sqrt(energy2/(s->channels*s->block_align)); |
8477e63d |
energy1 = M_SQRT2*energy1/(s->channels*s->block_align); |
115329f1 |
|
bb270c08 |
// increase bitrate when samples are like a gaussian distribution
// reduce bitrate when samples are like a two-tailed exponential distribution |
115329f1 |
|
bb270c08 |
if (energy2 > energy1)
energy2 += (energy2-energy1)*RATE_VARIATION; |
115329f1 |
|
bb270c08 |
quant = (int)(BASE_QUANT*s->quantization*energy2/SAMPLE_FACTOR);
// av_log(avctx, AV_LOG_DEBUG, "quant: %d energy: %f / %f\n", quant, energy1, energy2); |
54f5fd22 |
|
4ec7ef56 |
quant = av_clip(quant, 1, 65534); |
115329f1 |
|
6026a5ad |
put_symbol(&c, state, quant, 0, NULL, NULL); |
115329f1 |
|
bb270c08 |
quant *= SAMPLE_FACTOR; |
54f5fd22 |
}
// write out coded samples
for (ch = 0; ch < s->channels; ch++)
{ |
bb270c08 |
if (!s->lossless)
for (i = 0; i < s->block_align; i++) |
db27dadc |
s->coded_samples[ch][i] = ROUNDED_DIV(s->coded_samples[ch][i], quant); |
54f5fd22 |
|
6026a5ad |
if ((ret = intlist_write(&c, state, s->coded_samples[ch], s->block_align, 1)) < 0) |
12de2933 |
return ret; |
54f5fd22 |
}
// av_log(avctx, AV_LOG_DEBUG, "used bytes: %d\n", (put_bits_count(&pb)+7)/8);
|
6026a5ad |
avpkt->size = ff_rac_terminate(&c); |
a44cbc1c |
*got_packet_ptr = 1;
return 0; |
6026a5ad |
|
54f5fd22 |
} |
b250f9c6 |
#endif /* CONFIG_SONIC_ENCODER || CONFIG_SONIC_LS_ENCODER */ |
54f5fd22 |
|
b250f9c6 |
#if CONFIG_SONIC_DECODER |
359a9979 |
static const int samplerate_table[] =
{ 44100, 22050, 11025, 96000, 48000, 32000, 24000, 16000, 8000 };
|
98a6fff9 |
static av_cold int sonic_decode_init(AVCodecContext *avctx) |
54f5fd22 |
{
SonicContext *s = avctx->priv_data;
GetBitContext gb; |
4c32629b |
int i; |
115329f1 |
|
54f5fd22 |
s->channels = avctx->channels;
s->samplerate = avctx->sample_rate; |
115329f1 |
|
54f5fd22 |
if (!avctx->extradata)
{ |
bb270c08 |
av_log(avctx, AV_LOG_ERROR, "No mandatory headers present\n"); |
12de2933 |
return AVERROR_INVALIDDATA; |
54f5fd22 |
} |
115329f1 |
|
6d05039c |
init_get_bits8(&gb, avctx->extradata, avctx->extradata_size); |
115329f1 |
|
4c32629b |
s->version = get_bits(&gb, 2); |
c61daa68 |
if (s->version >= 2) {
s->version = get_bits(&gb, 8);
s->minor_version = get_bits(&gb, 8);
} |
6026a5ad |
if (s->version != 2) |
54f5fd22 |
{ |
bb270c08 |
av_log(avctx, AV_LOG_ERROR, "Unsupported Sonic version, please report\n"); |
12de2933 |
return AVERROR_INVALIDDATA; |
54f5fd22 |
}
|
c61daa68 |
if (s->version >= 1) |
54f5fd22 |
{ |
bb270c08 |
s->channels = get_bits(&gb, 2);
s->samplerate = samplerate_table[get_bits(&gb, 4)];
av_log(avctx, AV_LOG_INFO, "Sonicv2 chans: %d samprate: %d\n",
s->channels, s->samplerate); |
54f5fd22 |
}
if (s->channels > MAX_CHANNELS)
{ |
bb270c08 |
av_log(avctx, AV_LOG_ERROR, "Only mono and stereo streams are supported by now\n"); |
12de2933 |
return AVERROR_INVALIDDATA; |
54f5fd22 |
}
s->lossless = get_bits1(&gb);
if (!s->lossless) |
bb270c08 |
skip_bits(&gb, 3); // XXX FIXME |
cc078b9e |
s->decorrelation = get_bits(&gb, 2); |
1426291e |
if (s->decorrelation != 3 && s->channels != 2) {
av_log(avctx, AV_LOG_ERROR, "invalid decorrelation %d\n", s->decorrelation);
return AVERROR_INVALIDDATA;
} |
54f5fd22 |
s->downsampling = get_bits(&gb, 2); |
8a0cd587 |
if (!s->downsampling) {
av_log(avctx, AV_LOG_ERROR, "invalid downsampling value\n");
return AVERROR_INVALIDDATA;
}
|
54f5fd22 |
s->num_taps = (get_bits(&gb, 5)+1)<<5;
if (get_bits1(&gb)) // XXX FIXME |
bb270c08 |
av_log(avctx, AV_LOG_INFO, "Custom quant table\n"); |
115329f1 |
|
081a7f3e |
s->block_align = 2048LL*s->samplerate/(44100*s->downsampling); |
54f5fd22 |
s->frame_size = s->channels*s->block_align*s->downsampling;
// avctx->frame_size = s->block_align;
|
c61daa68 |
av_log(avctx, AV_LOG_INFO, "Sonic: ver: %d.%d ls: %d dr: %d taps: %d block: %d frame: %d downsamp: %d\n",
s->version, s->minor_version, s->lossless, s->decorrelation, s->num_taps, s->block_align, s->frame_size, s->downsampling); |
54f5fd22 |
// generate taps |
bcb42fb6 |
s->tap_quant = av_calloc(s->num_taps, sizeof(*s->tap_quant)); |
54f5fd22 |
for (i = 0; i < s->num_taps; i++) |
730e07f1 |
s->tap_quant[i] = ff_sqrt(i+1); |
115329f1 |
|
bcb42fb6 |
s->predictor_k = av_calloc(s->num_taps, sizeof(*s->predictor_k)); |
115329f1 |
|
54f5fd22 |
for (i = 0; i < s->channels; i++)
{ |
bcb42fb6 |
s->predictor_state[i] = av_calloc(s->num_taps, sizeof(**s->predictor_state)); |
bb270c08 |
if (!s->predictor_state[i]) |
12de2933 |
return AVERROR(ENOMEM); |
54f5fd22 |
}
for (i = 0; i < s->channels; i++)
{ |
bcb42fb6 |
s->coded_samples[i] = av_calloc(s->block_align, sizeof(**s->coded_samples)); |
bb270c08 |
if (!s->coded_samples[i]) |
12de2933 |
return AVERROR(ENOMEM); |
54f5fd22 |
} |
bcb42fb6 |
s->int_samples = av_calloc(s->frame_size, sizeof(*s->int_samples)); |
54f5fd22 |
|
5d6e4c16 |
avctx->sample_fmt = AV_SAMPLE_FMT_S16; |
54f5fd22 |
return 0;
}
|
98a6fff9 |
static av_cold int sonic_decode_close(AVCodecContext *avctx) |
54f5fd22 |
{
SonicContext *s = avctx->priv_data;
int i; |
115329f1 |
|
9375f500 |
av_freep(&s->int_samples);
av_freep(&s->tap_quant);
av_freep(&s->predictor_k); |
115329f1 |
|
54f5fd22 |
for (i = 0; i < s->channels; i++)
{ |
9375f500 |
av_freep(&s->predictor_state[i]);
av_freep(&s->coded_samples[i]); |
54f5fd22 |
} |
115329f1 |
|
54f5fd22 |
return 0;
}
static int sonic_decode_frame(AVCodecContext *avctx, |
6f9803e5 |
void *data, int *got_frame_ptr, |
7a00bbad |
AVPacket *avpkt) |
54f5fd22 |
{ |
7a00bbad |
const uint8_t *buf = avpkt->data;
int buf_size = avpkt->size; |
54f5fd22 |
SonicContext *s = avctx->priv_data; |
6026a5ad |
RangeCoder c;
uint8_t state[32]; |
6f9803e5 |
int i, quant, ch, j, ret; |
89cd95b1 |
int16_t *samples; |
4aa85033 |
AVFrame *frame = data; |
54f5fd22 |
if (buf_size == 0) return 0;
|
4aa85033 |
frame->nb_samples = s->frame_size / avctx->channels;
if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) |
6f9803e5 |
return ret; |
4aa85033 |
samples = (int16_t *)frame->data[0]; |
6f9803e5 |
|
54f5fd22 |
// av_log(NULL, AV_LOG_INFO, "buf_size: %d\n", buf_size); |
115329f1 |
|
6026a5ad |
memset(state, 128, sizeof(state));
ff_init_range_decoder(&c, buf, buf_size);
ff_build_rac_states(&c, 0.05*(1LL<<32), 256-8); |
115329f1 |
|
6026a5ad |
intlist_read(&c, state, s->predictor_k, s->num_taps, 0); |
54f5fd22 |
// dequantize
for (i = 0; i < s->num_taps; i++) |
bb270c08 |
s->predictor_k[i] *= s->tap_quant[i]; |
54f5fd22 |
if (s->lossless) |
bb270c08 |
quant = 1; |
54f5fd22 |
else |
6026a5ad |
quant = get_symbol(&c, state, 0) * SAMPLE_FACTOR; |
54f5fd22 |
// av_log(NULL, AV_LOG_INFO, "quant: %d\n", quant);
for (ch = 0; ch < s->channels; ch++)
{ |
bb270c08 |
int x = ch; |
54f5fd22 |
|
bb270c08 |
predictor_init_state(s->predictor_k, s->predictor_state[ch], s->num_taps); |
115329f1 |
|
6026a5ad |
intlist_read(&c, state, s->coded_samples[ch], s->block_align, 1); |
54f5fd22 |
|
bb270c08 |
for (i = 0; i < s->block_align; i++)
{
for (j = 0; j < s->downsampling - 1; j++)
{
s->int_samples[x] = predictor_calc_error(s->predictor_k, s->predictor_state[ch], s->num_taps, 0);
x += s->channels;
} |
115329f1 |
|
bb270c08 |
s->int_samples[x] = predictor_calc_error(s->predictor_k, s->predictor_state[ch], s->num_taps, s->coded_samples[ch][i] * quant);
x += s->channels;
} |
54f5fd22 |
|
bb270c08 |
for (i = 0; i < s->num_taps; i++)
s->predictor_state[ch][i] = s->int_samples[s->frame_size - s->channels + ch - i*s->channels]; |
54f5fd22 |
} |
115329f1 |
|
cc078b9e |
switch(s->decorrelation)
{ |
bb270c08 |
case MID_SIDE:
for (i = 0; i < s->frame_size; i += s->channels)
{
s->int_samples[i+1] += shift(s->int_samples[i], 1);
s->int_samples[i] -= s->int_samples[i+1];
}
break;
case LEFT_SIDE:
for (i = 0; i < s->frame_size; i += s->channels)
s->int_samples[i+1] += s->int_samples[i];
break;
case RIGHT_SIDE:
for (i = 0; i < s->frame_size; i += s->channels)
s->int_samples[i] += s->int_samples[i+1];
break; |
cc078b9e |
} |
54f5fd22 |
if (!s->lossless) |
bb270c08 |
for (i = 0; i < s->frame_size; i++)
s->int_samples[i] = shift(s->int_samples[i], SAMPLE_SHIFT); |
54f5fd22 |
// internal -> short
for (i = 0; i < s->frame_size; i++) |
aee481ce |
samples[i] = av_clip_int16(s->int_samples[i]); |
54f5fd22 |
|
6f9803e5 |
*got_frame_ptr = 1; |
54f5fd22 |
|
6026a5ad |
return buf_size; |
54f5fd22 |
} |
359a9979 |
|
e7e2df27 |
AVCodec ff_sonic_decoder = { |
ba10207b |
.name = "sonic", |
b46f1910 |
.long_name = NULL_IF_CONFIG_SMALL("Sonic"), |
ba10207b |
.type = AVMEDIA_TYPE_AUDIO, |
7a72695c |
.id = AV_CODEC_ID_SONIC, |
ba10207b |
.priv_data_size = sizeof(SonicContext),
.init = sonic_decode_init,
.close = sonic_decode_close,
.decode = sonic_decode_frame, |
7ed9abf7 |
.capabilities = CODEC_CAP_DR1 | CODEC_CAP_EXPERIMENTAL, |
359a9979 |
}; |
2a43a093 |
#endif /* CONFIG_SONIC_DECODER */ |
54f5fd22 |
|
b250f9c6 |
#if CONFIG_SONIC_ENCODER |
e7e2df27 |
AVCodec ff_sonic_encoder = { |
ba10207b |
.name = "sonic", |
b46f1910 |
.long_name = NULL_IF_CONFIG_SMALL("Sonic"), |
ba10207b |
.type = AVMEDIA_TYPE_AUDIO, |
7a72695c |
.id = AV_CODEC_ID_SONIC, |
ba10207b |
.priv_data_size = sizeof(SonicContext),
.init = sonic_encode_init, |
a44cbc1c |
.encode2 = sonic_encode_frame, |
6df61c3a |
.sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE }, |
7ed9abf7 |
.capabilities = CODEC_CAP_EXPERIMENTAL, |
ba10207b |
.close = sonic_encode_close, |
54f5fd22 |
}; |
f544a5fc |
#endif |
54f5fd22 |
|
b250f9c6 |
#if CONFIG_SONIC_LS_ENCODER |
e7e2df27 |
AVCodec ff_sonic_ls_encoder = { |
ba10207b |
.name = "sonicls", |
b46f1910 |
.long_name = NULL_IF_CONFIG_SMALL("Sonic lossless"), |
ba10207b |
.type = AVMEDIA_TYPE_AUDIO, |
7a72695c |
.id = AV_CODEC_ID_SONIC_LS, |
ba10207b |
.priv_data_size = sizeof(SonicContext),
.init = sonic_encode_init, |
a44cbc1c |
.encode2 = sonic_encode_frame, |
6df61c3a |
.sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE }, |
7ed9abf7 |
.capabilities = CODEC_CAP_EXPERIMENTAL, |
ba10207b |
.close = sonic_encode_close, |
54f5fd22 |
};
#endif |