mirror of
https://git.ffmpeg.org/ffmpeg.git
synced 2024-09-19 21:06:42 +00:00
avcodec/aptx: split decoder and encoder into separate files
Signed-off-by: James Almer <jamrial@gmail.com>
This commit is contained in:
parent
a8a05340de
commit
2383021a7a
@ -194,10 +194,10 @@ OBJS-$(CONFIG_AMV_ENCODER) += mjpegenc.o mjpegenc_common.o \
|
||||
OBJS-$(CONFIG_ANM_DECODER) += anm.o
|
||||
OBJS-$(CONFIG_ANSI_DECODER) += ansi.o cga_data.o
|
||||
OBJS-$(CONFIG_APE_DECODER) += apedec.o
|
||||
OBJS-$(CONFIG_APTX_DECODER) += aptx.o
|
||||
OBJS-$(CONFIG_APTX_ENCODER) += aptx.o
|
||||
OBJS-$(CONFIG_APTX_HD_DECODER) += aptx.o
|
||||
OBJS-$(CONFIG_APTX_HD_ENCODER) += aptx.o
|
||||
OBJS-$(CONFIG_APTX_DECODER) += aptxdec.o aptx.o
|
||||
OBJS-$(CONFIG_APTX_ENCODER) += aptxenc.o aptx.o
|
||||
OBJS-$(CONFIG_APTX_HD_DECODER) += aptxdec.o aptx.o
|
||||
OBJS-$(CONFIG_APTX_HD_ENCODER) += aptxenc.o aptx.o
|
||||
OBJS-$(CONFIG_APNG_DECODER) += png.o pngdec.o pngdsp.o
|
||||
OBJS-$(CONFIG_APNG_ENCODER) += png.o pngenc.o
|
||||
OBJS-$(CONFIG_ARBC_DECODER) += arbc.o
|
||||
|
@ -20,81 +20,7 @@
|
||||
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
||||
*/
|
||||
|
||||
#include "libavutil/intreadwrite.h"
|
||||
#include "avcodec.h"
|
||||
#include "internal.h"
|
||||
#include "mathops.h"
|
||||
#include "audio_frame_queue.h"
|
||||
|
||||
|
||||
enum channels {
|
||||
LEFT,
|
||||
RIGHT,
|
||||
NB_CHANNELS
|
||||
};
|
||||
|
||||
enum subbands {
|
||||
LF, // Low Frequency (0-5.5 kHz)
|
||||
MLF, // Medium-Low Frequency (5.5-11kHz)
|
||||
MHF, // Medium-High Frequency (11-16.5kHz)
|
||||
HF, // High Frequency (16.5-22kHz)
|
||||
NB_SUBBANDS
|
||||
};
|
||||
|
||||
#define NB_FILTERS 2
|
||||
#define FILTER_TAPS 16
|
||||
|
||||
typedef struct {
|
||||
int pos;
|
||||
int32_t buffer[2*FILTER_TAPS];
|
||||
} FilterSignal;
|
||||
|
||||
typedef struct {
|
||||
FilterSignal outer_filter_signal[NB_FILTERS];
|
||||
FilterSignal inner_filter_signal[NB_FILTERS][NB_FILTERS];
|
||||
} QMFAnalysis;
|
||||
|
||||
typedef struct {
|
||||
int32_t quantized_sample;
|
||||
int32_t quantized_sample_parity_change;
|
||||
int32_t error;
|
||||
} Quantize;
|
||||
|
||||
typedef struct {
|
||||
int32_t quantization_factor;
|
||||
int32_t factor_select;
|
||||
int32_t reconstructed_difference;
|
||||
} InvertQuantize;
|
||||
|
||||
typedef struct {
|
||||
int32_t prev_sign[2];
|
||||
int32_t s_weight[2];
|
||||
int32_t d_weight[24];
|
||||
int32_t pos;
|
||||
int32_t reconstructed_differences[48];
|
||||
int32_t previous_reconstructed_sample;
|
||||
int32_t predicted_difference;
|
||||
int32_t predicted_sample;
|
||||
} Prediction;
|
||||
|
||||
typedef struct {
|
||||
int32_t codeword_history;
|
||||
int32_t dither_parity;
|
||||
int32_t dither[NB_SUBBANDS];
|
||||
|
||||
QMFAnalysis qmf;
|
||||
Quantize quantize[NB_SUBBANDS];
|
||||
InvertQuantize invert_quantize[NB_SUBBANDS];
|
||||
Prediction prediction[NB_SUBBANDS];
|
||||
} Channel;
|
||||
|
||||
typedef struct {
|
||||
int hd;
|
||||
int block_size;
|
||||
int32_t sync_idx;
|
||||
Channel channels[NB_CHANNELS];
|
||||
AudioFrameQueue afq;
|
||||
} AptXContext;
|
||||
#include "aptx.h"
|
||||
|
||||
|
||||
static const int32_t quantize_intervals_LF[65] = {
|
||||
@ -383,17 +309,7 @@ static const int16_t hd_quantize_factor_select_offset_HF[17] = {
|
||||
70, 90, 115, 147, 192, 264, 398, 521, 521,
|
||||
};
|
||||
|
||||
typedef const struct {
|
||||
const int32_t *quantize_intervals;
|
||||
const int32_t *invert_quantize_dither_factors;
|
||||
const int32_t *quantize_dither_factors;
|
||||
const int16_t *quantize_factor_select_offset;
|
||||
int tables_size;
|
||||
int32_t factor_max;
|
||||
int32_t prediction_order;
|
||||
} ConstTables;
|
||||
|
||||
static ConstTables tables[2][NB_SUBBANDS] = {
|
||||
ConstTables ff_aptx_quant_tables[2][NB_SUBBANDS] = {
|
||||
{
|
||||
[LF] = { quantize_intervals_LF,
|
||||
invert_quantize_dither_factors_LF,
|
||||
@ -456,24 +372,6 @@ static const int16_t quantization_factors[32] = {
|
||||
};
|
||||
|
||||
|
||||
/* Rounded right shift with optionnal clipping */
|
||||
#define RSHIFT_SIZE(size) \
|
||||
av_always_inline \
|
||||
static int##size##_t rshift##size(int##size##_t value, int shift) \
|
||||
{ \
|
||||
int##size##_t rounding = (int##size##_t)1 << (shift - 1); \
|
||||
int##size##_t mask = ((int##size##_t)1 << (shift + 1)) - 1; \
|
||||
return ((value + rounding) >> shift) - ((value & mask) == rounding); \
|
||||
} \
|
||||
av_always_inline \
|
||||
static int##size##_t rshift##size##_clip24(int##size##_t value, int shift) \
|
||||
{ \
|
||||
return av_clip_intp2(rshift##size(value, shift), 23); \
|
||||
}
|
||||
RSHIFT_SIZE(32)
|
||||
RSHIFT_SIZE(64)
|
||||
|
||||
|
||||
av_always_inline
|
||||
static void aptx_update_codeword_history(Channel *channel)
|
||||
{
|
||||
@ -483,7 +381,7 @@ static void aptx_update_codeword_history(Channel *channel)
|
||||
channel->codeword_history = (cw << 8) + ((unsigned)channel->codeword_history << 4);
|
||||
}
|
||||
|
||||
static void aptx_generate_dither(Channel *channel)
|
||||
void ff_aptx_generate_dither(Channel *channel)
|
||||
{
|
||||
int subband;
|
||||
int64_t m;
|
||||
@ -498,256 +396,6 @@ static void aptx_generate_dither(Channel *channel)
|
||||
channel->dither_parity = (d >> 25) & 1;
|
||||
}
|
||||
|
||||
/*
|
||||
* Convolution filter coefficients for the outer QMF of the QMF tree.
|
||||
* The 2 sets are a mirror of each other.
|
||||
*/
|
||||
static const int32_t aptx_qmf_outer_coeffs[NB_FILTERS][FILTER_TAPS] = {
|
||||
{
|
||||
730, -413, -9611, 43626, -121026, 269973, -585547, 2801966,
|
||||
697128, -160481, 27611, 8478, -10043, 3511, 688, -897,
|
||||
},
|
||||
{
|
||||
-897, 688, 3511, -10043, 8478, 27611, -160481, 697128,
|
||||
2801966, -585547, 269973, -121026, 43626, -9611, -413, 730,
|
||||
},
|
||||
};
|
||||
|
||||
/*
|
||||
* Convolution filter coefficients for the inner QMF of the QMF tree.
|
||||
* The 2 sets are a mirror of each other.
|
||||
*/
|
||||
static const int32_t aptx_qmf_inner_coeffs[NB_FILTERS][FILTER_TAPS] = {
|
||||
{
|
||||
1033, -584, -13592, 61697, -171156, 381799, -828088, 3962579,
|
||||
985888, -226954, 39048, 11990, -14203, 4966, 973, -1268,
|
||||
},
|
||||
{
|
||||
-1268, 973, 4966, -14203, 11990, 39048, -226954, 985888,
|
||||
3962579, -828088, 381799, -171156, 61697, -13592, -584, 1033,
|
||||
},
|
||||
};
|
||||
|
||||
/*
|
||||
* Push one sample into a circular signal buffer.
|
||||
*/
|
||||
av_always_inline
|
||||
static void aptx_qmf_filter_signal_push(FilterSignal *signal, int32_t sample)
|
||||
{
|
||||
signal->buffer[signal->pos ] = sample;
|
||||
signal->buffer[signal->pos+FILTER_TAPS] = sample;
|
||||
signal->pos = (signal->pos + 1) & (FILTER_TAPS - 1);
|
||||
}
|
||||
|
||||
/*
|
||||
* Compute the convolution of the signal with the coefficients, and reduce
|
||||
* to 24 bits by applying the specified right shifting.
|
||||
*/
|
||||
av_always_inline
|
||||
static int32_t aptx_qmf_convolution(FilterSignal *signal,
|
||||
const int32_t coeffs[FILTER_TAPS],
|
||||
int shift)
|
||||
{
|
||||
int32_t *sig = &signal->buffer[signal->pos];
|
||||
int64_t e = 0;
|
||||
int i;
|
||||
|
||||
for (i = 0; i < FILTER_TAPS; i++)
|
||||
e += MUL64(sig[i], coeffs[i]);
|
||||
|
||||
return rshift64_clip24(e, shift);
|
||||
}
|
||||
|
||||
/*
|
||||
* Half-band QMF analysis filter realized with a polyphase FIR filter.
|
||||
* Split into 2 subbands and downsample by 2.
|
||||
* So for each pair of samples that goes in, one sample goes out,
|
||||
* split into 2 separate subbands.
|
||||
*/
|
||||
av_always_inline
|
||||
static void aptx_qmf_polyphase_analysis(FilterSignal signal[NB_FILTERS],
|
||||
const int32_t coeffs[NB_FILTERS][FILTER_TAPS],
|
||||
int shift,
|
||||
int32_t samples[NB_FILTERS],
|
||||
int32_t *low_subband_output,
|
||||
int32_t *high_subband_output)
|
||||
{
|
||||
int32_t subbands[NB_FILTERS];
|
||||
int i;
|
||||
|
||||
for (i = 0; i < NB_FILTERS; i++) {
|
||||
aptx_qmf_filter_signal_push(&signal[i], samples[NB_FILTERS-1-i]);
|
||||
subbands[i] = aptx_qmf_convolution(&signal[i], coeffs[i], shift);
|
||||
}
|
||||
|
||||
*low_subband_output = av_clip_intp2(subbands[0] + subbands[1], 23);
|
||||
*high_subband_output = av_clip_intp2(subbands[0] - subbands[1], 23);
|
||||
}
|
||||
|
||||
/*
|
||||
* Two stage QMF analysis tree.
|
||||
* Split 4 input samples into 4 subbands and downsample by 4.
|
||||
* So for each group of 4 samples that goes in, one sample goes out,
|
||||
* split into 4 separate subbands.
|
||||
*/
|
||||
static void aptx_qmf_tree_analysis(QMFAnalysis *qmf,
|
||||
int32_t samples[4],
|
||||
int32_t subband_samples[4])
|
||||
{
|
||||
int32_t intermediate_samples[4];
|
||||
int i;
|
||||
|
||||
/* Split 4 input samples into 2 intermediate subbands downsampled to 2 samples */
|
||||
for (i = 0; i < 2; i++)
|
||||
aptx_qmf_polyphase_analysis(qmf->outer_filter_signal,
|
||||
aptx_qmf_outer_coeffs, 23,
|
||||
&samples[2*i],
|
||||
&intermediate_samples[0+i],
|
||||
&intermediate_samples[2+i]);
|
||||
|
||||
/* Split 2 intermediate subband samples into 4 final subbands downsampled to 1 sample */
|
||||
for (i = 0; i < 2; i++)
|
||||
aptx_qmf_polyphase_analysis(qmf->inner_filter_signal[i],
|
||||
aptx_qmf_inner_coeffs, 23,
|
||||
&intermediate_samples[2*i],
|
||||
&subband_samples[2*i+0],
|
||||
&subband_samples[2*i+1]);
|
||||
}
|
||||
|
||||
/*
|
||||
* Half-band QMF synthesis filter realized with a polyphase FIR filter.
|
||||
* Join 2 subbands and upsample by 2.
|
||||
* So for each 2 subbands sample that goes in, a pair of samples goes out.
|
||||
*/
|
||||
av_always_inline
|
||||
static void aptx_qmf_polyphase_synthesis(FilterSignal signal[NB_FILTERS],
|
||||
const int32_t coeffs[NB_FILTERS][FILTER_TAPS],
|
||||
int shift,
|
||||
int32_t low_subband_input,
|
||||
int32_t high_subband_input,
|
||||
int32_t samples[NB_FILTERS])
|
||||
{
|
||||
int32_t subbands[NB_FILTERS];
|
||||
int i;
|
||||
|
||||
subbands[0] = low_subband_input + high_subband_input;
|
||||
subbands[1] = low_subband_input - high_subband_input;
|
||||
|
||||
for (i = 0; i < NB_FILTERS; i++) {
|
||||
aptx_qmf_filter_signal_push(&signal[i], subbands[1-i]);
|
||||
samples[i] = aptx_qmf_convolution(&signal[i], coeffs[i], shift);
|
||||
}
|
||||
}
|
||||
|
||||
/*
|
||||
* Two stage QMF synthesis tree.
|
||||
* Join 4 subbands and upsample by 4.
|
||||
* So for each 4 subbands sample that goes in, a group of 4 samples goes out.
|
||||
*/
|
||||
static void aptx_qmf_tree_synthesis(QMFAnalysis *qmf,
|
||||
int32_t subband_samples[4],
|
||||
int32_t samples[4])
|
||||
{
|
||||
int32_t intermediate_samples[4];
|
||||
int i;
|
||||
|
||||
/* Join 4 subbands into 2 intermediate subbands upsampled to 2 samples. */
|
||||
for (i = 0; i < 2; i++)
|
||||
aptx_qmf_polyphase_synthesis(qmf->inner_filter_signal[i],
|
||||
aptx_qmf_inner_coeffs, 22,
|
||||
subband_samples[2*i+0],
|
||||
subband_samples[2*i+1],
|
||||
&intermediate_samples[2*i]);
|
||||
|
||||
/* Join 2 samples from intermediate subbands upsampled to 4 samples. */
|
||||
for (i = 0; i < 2; i++)
|
||||
aptx_qmf_polyphase_synthesis(qmf->outer_filter_signal,
|
||||
aptx_qmf_outer_coeffs, 21,
|
||||
intermediate_samples[0+i],
|
||||
intermediate_samples[2+i],
|
||||
&samples[2*i]);
|
||||
}
|
||||
|
||||
|
||||
av_always_inline
|
||||
static int32_t aptx_bin_search(int32_t value, int32_t factor,
|
||||
const int32_t *intervals, int32_t nb_intervals)
|
||||
{
|
||||
int32_t idx = 0;
|
||||
int i;
|
||||
|
||||
for (i = nb_intervals >> 1; i > 0; i >>= 1)
|
||||
if (MUL64(factor, intervals[idx + i]) <= ((int64_t)value << 24))
|
||||
idx += i;
|
||||
|
||||
return idx;
|
||||
}
|
||||
|
||||
static void aptx_quantize_difference(Quantize *quantize,
|
||||
int32_t sample_difference,
|
||||
int32_t dither,
|
||||
int32_t quantization_factor,
|
||||
ConstTables *tables)
|
||||
{
|
||||
const int32_t *intervals = tables->quantize_intervals;
|
||||
int32_t quantized_sample, dithered_sample, parity_change;
|
||||
int32_t d, mean, interval, inv, sample_difference_abs;
|
||||
int64_t error;
|
||||
|
||||
sample_difference_abs = FFABS(sample_difference);
|
||||
sample_difference_abs = FFMIN(sample_difference_abs, (1 << 23) - 1);
|
||||
|
||||
quantized_sample = aptx_bin_search(sample_difference_abs >> 4,
|
||||
quantization_factor,
|
||||
intervals, tables->tables_size);
|
||||
|
||||
d = rshift32_clip24(MULH(dither, dither), 7) - (1 << 23);
|
||||
d = rshift64(MUL64(d, tables->quantize_dither_factors[quantized_sample]), 23);
|
||||
|
||||
intervals += quantized_sample;
|
||||
mean = (intervals[1] + intervals[0]) / 2;
|
||||
interval = (intervals[1] - intervals[0]) * (-(sample_difference < 0) | 1);
|
||||
|
||||
dithered_sample = rshift64_clip24(MUL64(dither, interval) + ((int64_t)av_clip_intp2(mean + d, 23) << 32), 32);
|
||||
error = ((int64_t)sample_difference_abs << 20) - MUL64(dithered_sample, quantization_factor);
|
||||
quantize->error = FFABS(rshift64(error, 23));
|
||||
|
||||
parity_change = quantized_sample;
|
||||
if (error < 0)
|
||||
quantized_sample--;
|
||||
else
|
||||
parity_change--;
|
||||
|
||||
inv = -(sample_difference < 0);
|
||||
quantize->quantized_sample = quantized_sample ^ inv;
|
||||
quantize->quantized_sample_parity_change = parity_change ^ inv;
|
||||
}
|
||||
|
||||
static void aptx_encode_channel(Channel *channel, int32_t samples[4], int hd)
|
||||
{
|
||||
int32_t subband_samples[4];
|
||||
int subband;
|
||||
aptx_qmf_tree_analysis(&channel->qmf, samples, subband_samples);
|
||||
aptx_generate_dither(channel);
|
||||
for (subband = 0; subband < NB_SUBBANDS; subband++) {
|
||||
int32_t diff = av_clip_intp2(subband_samples[subband] - channel->prediction[subband].predicted_sample, 23);
|
||||
aptx_quantize_difference(&channel->quantize[subband], diff,
|
||||
channel->dither[subband],
|
||||
channel->invert_quantize[subband].quantization_factor,
|
||||
&tables[hd][subband]);
|
||||
}
|
||||
}
|
||||
|
||||
static void aptx_decode_channel(Channel *channel, int32_t samples[4])
|
||||
{
|
||||
int32_t subband_samples[4];
|
||||
int subband;
|
||||
for (subband = 0; subband < NB_SUBBANDS; subband++)
|
||||
subband_samples[subband] = channel->prediction[subband].previous_reconstructed_sample;
|
||||
aptx_qmf_tree_synthesis(&channel->qmf, subband_samples, samples);
|
||||
}
|
||||
|
||||
|
||||
static void aptx_invert_quantization(InvertQuantize *invert_quantize,
|
||||
int32_t quantized_sample, int32_t dither,
|
||||
ConstTables *tables)
|
||||
@ -845,7 +493,7 @@ static void aptx_process_subband(InvertQuantize *invert_quantize,
|
||||
tables->prediction_order);
|
||||
}
|
||||
|
||||
static void aptx_invert_quantize_and_prediction(Channel *channel, int hd)
|
||||
void ff_aptx_invert_quantize_and_prediction(Channel *channel, int hd)
|
||||
{
|
||||
int subband;
|
||||
for (subband = 0; subband < NB_SUBBANDS; subband++)
|
||||
@ -853,138 +501,10 @@ static void aptx_invert_quantize_and_prediction(Channel *channel, int hd)
|
||||
&channel->prediction[subband],
|
||||
channel->quantize[subband].quantized_sample,
|
||||
channel->dither[subband],
|
||||
&tables[hd][subband]);
|
||||
&ff_aptx_quant_tables[hd][subband]);
|
||||
}
|
||||
|
||||
static int32_t aptx_quantized_parity(Channel *channel)
|
||||
{
|
||||
int32_t parity = channel->dither_parity;
|
||||
int subband;
|
||||
|
||||
for (subband = 0; subband < NB_SUBBANDS; subband++)
|
||||
parity ^= channel->quantize[subband].quantized_sample;
|
||||
|
||||
return parity & 1;
|
||||
}
|
||||
|
||||
/* For each sample, ensure that the parity of all subbands of all channels
|
||||
* is 0 except once every 8 samples where the parity is forced to 1. */
|
||||
static int aptx_check_parity(Channel channels[NB_CHANNELS], int32_t *idx)
|
||||
{
|
||||
int32_t parity = aptx_quantized_parity(&channels[LEFT])
|
||||
^ aptx_quantized_parity(&channels[RIGHT]);
|
||||
|
||||
int eighth = *idx == 7;
|
||||
*idx = (*idx + 1) & 7;
|
||||
|
||||
return parity ^ eighth;
|
||||
}
|
||||
|
||||
static void aptx_insert_sync(Channel channels[NB_CHANNELS], int32_t *idx)
|
||||
{
|
||||
if (aptx_check_parity(channels, idx)) {
|
||||
int i;
|
||||
Channel *c;
|
||||
static const int map[] = { 1, 2, 0, 3 };
|
||||
Quantize *min = &channels[NB_CHANNELS-1].quantize[map[0]];
|
||||
for (c = &channels[NB_CHANNELS-1]; c >= channels; c--)
|
||||
for (i = 0; i < NB_SUBBANDS; i++)
|
||||
if (c->quantize[map[i]].error < min->error)
|
||||
min = &c->quantize[map[i]];
|
||||
|
||||
/* Forcing the desired parity is done by offsetting by 1 the quantized
|
||||
* sample from the subband featuring the smallest quantization error. */
|
||||
min->quantized_sample = min->quantized_sample_parity_change;
|
||||
}
|
||||
}
|
||||
|
||||
static uint16_t aptx_pack_codeword(Channel *channel)
|
||||
{
|
||||
int32_t parity = aptx_quantized_parity(channel);
|
||||
return (((channel->quantize[3].quantized_sample & 0x06) | parity) << 13)
|
||||
| (((channel->quantize[2].quantized_sample & 0x03) ) << 11)
|
||||
| (((channel->quantize[1].quantized_sample & 0x0F) ) << 7)
|
||||
| (((channel->quantize[0].quantized_sample & 0x7F) ) << 0);
|
||||
}
|
||||
|
||||
static uint32_t aptxhd_pack_codeword(Channel *channel)
|
||||
{
|
||||
int32_t parity = aptx_quantized_parity(channel);
|
||||
return (((channel->quantize[3].quantized_sample & 0x01E) | parity) << 19)
|
||||
| (((channel->quantize[2].quantized_sample & 0x00F) ) << 15)
|
||||
| (((channel->quantize[1].quantized_sample & 0x03F) ) << 9)
|
||||
| (((channel->quantize[0].quantized_sample & 0x1FF) ) << 0);
|
||||
}
|
||||
|
||||
static void aptx_unpack_codeword(Channel *channel, uint16_t codeword)
|
||||
{
|
||||
channel->quantize[0].quantized_sample = sign_extend(codeword >> 0, 7);
|
||||
channel->quantize[1].quantized_sample = sign_extend(codeword >> 7, 4);
|
||||
channel->quantize[2].quantized_sample = sign_extend(codeword >> 11, 2);
|
||||
channel->quantize[3].quantized_sample = sign_extend(codeword >> 13, 3);
|
||||
channel->quantize[3].quantized_sample = (channel->quantize[3].quantized_sample & ~1)
|
||||
| aptx_quantized_parity(channel);
|
||||
}
|
||||
|
||||
static void aptxhd_unpack_codeword(Channel *channel, uint32_t codeword)
|
||||
{
|
||||
channel->quantize[0].quantized_sample = sign_extend(codeword >> 0, 9);
|
||||
channel->quantize[1].quantized_sample = sign_extend(codeword >> 9, 6);
|
||||
channel->quantize[2].quantized_sample = sign_extend(codeword >> 15, 4);
|
||||
channel->quantize[3].quantized_sample = sign_extend(codeword >> 19, 5);
|
||||
channel->quantize[3].quantized_sample = (channel->quantize[3].quantized_sample & ~1)
|
||||
| aptx_quantized_parity(channel);
|
||||
}
|
||||
|
||||
static void aptx_encode_samples(AptXContext *ctx,
|
||||
int32_t samples[NB_CHANNELS][4],
|
||||
uint8_t *output)
|
||||
{
|
||||
int channel;
|
||||
for (channel = 0; channel < NB_CHANNELS; channel++)
|
||||
aptx_encode_channel(&ctx->channels[channel], samples[channel], ctx->hd);
|
||||
|
||||
aptx_insert_sync(ctx->channels, &ctx->sync_idx);
|
||||
|
||||
for (channel = 0; channel < NB_CHANNELS; channel++) {
|
||||
aptx_invert_quantize_and_prediction(&ctx->channels[channel], ctx->hd);
|
||||
if (ctx->hd)
|
||||
AV_WB24(output + 3*channel,
|
||||
aptxhd_pack_codeword(&ctx->channels[channel]));
|
||||
else
|
||||
AV_WB16(output + 2*channel,
|
||||
aptx_pack_codeword(&ctx->channels[channel]));
|
||||
}
|
||||
}
|
||||
|
||||
static int aptx_decode_samples(AptXContext *ctx,
|
||||
const uint8_t *input,
|
||||
int32_t samples[NB_CHANNELS][4])
|
||||
{
|
||||
int channel, ret;
|
||||
|
||||
for (channel = 0; channel < NB_CHANNELS; channel++) {
|
||||
aptx_generate_dither(&ctx->channels[channel]);
|
||||
|
||||
if (ctx->hd)
|
||||
aptxhd_unpack_codeword(&ctx->channels[channel],
|
||||
AV_RB24(input + 3*channel));
|
||||
else
|
||||
aptx_unpack_codeword(&ctx->channels[channel],
|
||||
AV_RB16(input + 2*channel));
|
||||
aptx_invert_quantize_and_prediction(&ctx->channels[channel], ctx->hd);
|
||||
}
|
||||
|
||||
ret = aptx_check_parity(ctx->channels, &ctx->sync_idx);
|
||||
|
||||
for (channel = 0; channel < NB_CHANNELS; channel++)
|
||||
aptx_decode_channel(&ctx->channels[channel], samples[channel]);
|
||||
|
||||
return ret;
|
||||
}
|
||||
|
||||
|
||||
static av_cold int aptx_init(AVCodecContext *avctx)
|
||||
av_cold int ff_aptx_init(AVCodecContext *avctx)
|
||||
{
|
||||
AptXContext *s = avctx->priv_data;
|
||||
int chan, subband;
|
||||
@ -1016,150 +536,3 @@ static av_cold int aptx_init(AVCodecContext *avctx)
|
||||
ff_af_queue_init(avctx, &s->afq);
|
||||
return 0;
|
||||
}
|
||||
|
||||
static int aptx_decode_frame(AVCodecContext *avctx, void *data,
|
||||
int *got_frame_ptr, AVPacket *avpkt)
|
||||
{
|
||||
AptXContext *s = avctx->priv_data;
|
||||
AVFrame *frame = data;
|
||||
int pos, opos, channel, sample, ret;
|
||||
|
||||
if (avpkt->size < s->block_size) {
|
||||
av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
|
||||
return AVERROR_INVALIDDATA;
|
||||
}
|
||||
|
||||
/* get output buffer */
|
||||
frame->channels = NB_CHANNELS;
|
||||
frame->format = AV_SAMPLE_FMT_S32P;
|
||||
frame->nb_samples = 4 * avpkt->size / s->block_size;
|
||||
if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
|
||||
return ret;
|
||||
|
||||
for (pos = 0, opos = 0; opos < frame->nb_samples; pos += s->block_size, opos += 4) {
|
||||
int32_t samples[NB_CHANNELS][4];
|
||||
|
||||
if (aptx_decode_samples(s, &avpkt->data[pos], samples)) {
|
||||
av_log(avctx, AV_LOG_ERROR, "Synchronization error\n");
|
||||
return AVERROR_INVALIDDATA;
|
||||
}
|
||||
|
||||
for (channel = 0; channel < NB_CHANNELS; channel++)
|
||||
for (sample = 0; sample < 4; sample++)
|
||||
AV_WN32A(&frame->data[channel][4*(opos+sample)],
|
||||
samples[channel][sample] * 256);
|
||||
}
|
||||
|
||||
*got_frame_ptr = 1;
|
||||
return s->block_size * frame->nb_samples / 4;
|
||||
}
|
||||
|
||||
static int aptx_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
|
||||
const AVFrame *frame, int *got_packet_ptr)
|
||||
{
|
||||
AptXContext *s = avctx->priv_data;
|
||||
int pos, ipos, channel, sample, output_size, ret;
|
||||
|
||||
if ((ret = ff_af_queue_add(&s->afq, frame)) < 0)
|
||||
return ret;
|
||||
|
||||
output_size = s->block_size * frame->nb_samples/4;
|
||||
if ((ret = ff_alloc_packet2(avctx, avpkt, output_size, 0)) < 0)
|
||||
return ret;
|
||||
|
||||
for (pos = 0, ipos = 0; pos < output_size; pos += s->block_size, ipos += 4) {
|
||||
int32_t samples[NB_CHANNELS][4];
|
||||
|
||||
for (channel = 0; channel < NB_CHANNELS; channel++)
|
||||
for (sample = 0; sample < 4; sample++)
|
||||
samples[channel][sample] = (int32_t)AV_RN32A(&frame->data[channel][4*(ipos+sample)]) >> 8;
|
||||
|
||||
aptx_encode_samples(s, samples, avpkt->data + pos);
|
||||
}
|
||||
|
||||
ff_af_queue_remove(&s->afq, frame->nb_samples, &avpkt->pts, &avpkt->duration);
|
||||
*got_packet_ptr = 1;
|
||||
return 0;
|
||||
}
|
||||
|
||||
static av_cold int aptx_close(AVCodecContext *avctx)
|
||||
{
|
||||
AptXContext *s = avctx->priv_data;
|
||||
ff_af_queue_close(&s->afq);
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
||||
#if CONFIG_APTX_DECODER
|
||||
AVCodec ff_aptx_decoder = {
|
||||
.name = "aptx",
|
||||
.long_name = NULL_IF_CONFIG_SMALL("aptX (Audio Processing Technology for Bluetooth)"),
|
||||
.type = AVMEDIA_TYPE_AUDIO,
|
||||
.id = AV_CODEC_ID_APTX,
|
||||
.priv_data_size = sizeof(AptXContext),
|
||||
.init = aptx_init,
|
||||
.decode = aptx_decode_frame,
|
||||
.close = aptx_close,
|
||||
.capabilities = AV_CODEC_CAP_DR1,
|
||||
.caps_internal = FF_CODEC_CAP_INIT_THREADSAFE,
|
||||
.channel_layouts = (const uint64_t[]) { AV_CH_LAYOUT_STEREO, 0},
|
||||
.sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S32P,
|
||||
AV_SAMPLE_FMT_NONE },
|
||||
};
|
||||
#endif
|
||||
|
||||
#if CONFIG_APTX_HD_DECODER
|
||||
AVCodec ff_aptx_hd_decoder = {
|
||||
.name = "aptx_hd",
|
||||
.long_name = NULL_IF_CONFIG_SMALL("aptX HD (Audio Processing Technology for Bluetooth)"),
|
||||
.type = AVMEDIA_TYPE_AUDIO,
|
||||
.id = AV_CODEC_ID_APTX_HD,
|
||||
.priv_data_size = sizeof(AptXContext),
|
||||
.init = aptx_init,
|
||||
.decode = aptx_decode_frame,
|
||||
.close = aptx_close,
|
||||
.capabilities = AV_CODEC_CAP_DR1,
|
||||
.caps_internal = FF_CODEC_CAP_INIT_THREADSAFE,
|
||||
.channel_layouts = (const uint64_t[]) { AV_CH_LAYOUT_STEREO, 0},
|
||||
.sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S32P,
|
||||
AV_SAMPLE_FMT_NONE },
|
||||
};
|
||||
#endif
|
||||
|
||||
#if CONFIG_APTX_ENCODER
|
||||
AVCodec ff_aptx_encoder = {
|
||||
.name = "aptx",
|
||||
.long_name = NULL_IF_CONFIG_SMALL("aptX (Audio Processing Technology for Bluetooth)"),
|
||||
.type = AVMEDIA_TYPE_AUDIO,
|
||||
.id = AV_CODEC_ID_APTX,
|
||||
.priv_data_size = sizeof(AptXContext),
|
||||
.init = aptx_init,
|
||||
.encode2 = aptx_encode_frame,
|
||||
.close = aptx_close,
|
||||
.capabilities = AV_CODEC_CAP_SMALL_LAST_FRAME,
|
||||
.caps_internal = FF_CODEC_CAP_INIT_THREADSAFE,
|
||||
.channel_layouts = (const uint64_t[]) { AV_CH_LAYOUT_STEREO, 0},
|
||||
.sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S32P,
|
||||
AV_SAMPLE_FMT_NONE },
|
||||
.supported_samplerates = (const int[]) {8000, 16000, 24000, 32000, 44100, 48000, 0},
|
||||
};
|
||||
#endif
|
||||
|
||||
#if CONFIG_APTX_HD_ENCODER
|
||||
AVCodec ff_aptx_hd_encoder = {
|
||||
.name = "aptx_hd",
|
||||
.long_name = NULL_IF_CONFIG_SMALL("aptX HD (Audio Processing Technology for Bluetooth)"),
|
||||
.type = AVMEDIA_TYPE_AUDIO,
|
||||
.id = AV_CODEC_ID_APTX_HD,
|
||||
.priv_data_size = sizeof(AptXContext),
|
||||
.init = aptx_init,
|
||||
.encode2 = aptx_encode_frame,
|
||||
.close = aptx_close,
|
||||
.capabilities = AV_CODEC_CAP_SMALL_LAST_FRAME,
|
||||
.caps_internal = FF_CODEC_CAP_INIT_THREADSAFE,
|
||||
.channel_layouts = (const uint64_t[]) { AV_CH_LAYOUT_STEREO, 0},
|
||||
.sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S32P,
|
||||
AV_SAMPLE_FMT_NONE },
|
||||
.supported_samplerates = (const int[]) {8000, 16000, 24000, 32000, 44100, 48000, 0},
|
||||
};
|
||||
#endif
|
||||
|
220
libavcodec/aptx.h
Normal file
220
libavcodec/aptx.h
Normal file
@ -0,0 +1,220 @@
|
||||
/*
|
||||
* Audio Processing Technology codec for Bluetooth (aptX)
|
||||
*
|
||||
* Copyright (C) 2017 Aurelien Jacobs <aurel@gnuage.org>
|
||||
*
|
||||
* This file is part of FFmpeg.
|
||||
*
|
||||
* FFmpeg is free software; you can redistribute it and/or
|
||||
* modify it under the terms of the GNU Lesser General Public
|
||||
* License as published by the Free Software Foundation; either
|
||||
* version 2.1 of the License, or (at your option) any later version.
|
||||
*
|
||||
* FFmpeg is distributed in the hope that it will be useful,
|
||||
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||||
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
||||
* Lesser General Public License for more details.
|
||||
*
|
||||
* You should have received a copy of the GNU Lesser General Public
|
||||
* License along with FFmpeg; if not, write to the Free Software
|
||||
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
||||
*/
|
||||
|
||||
#ifndef AVCODEC_APTX_H
|
||||
#define AVCODEC_APTX_H
|
||||
|
||||
#include "libavutil/intreadwrite.h"
|
||||
#include "avcodec.h"
|
||||
#include "internal.h"
|
||||
#include "mathops.h"
|
||||
#include "audio_frame_queue.h"
|
||||
|
||||
|
||||
enum channels {
|
||||
LEFT,
|
||||
RIGHT,
|
||||
NB_CHANNELS
|
||||
};
|
||||
|
||||
enum subbands {
|
||||
LF, // Low Frequency (0-5.5 kHz)
|
||||
MLF, // Medium-Low Frequency (5.5-11kHz)
|
||||
MHF, // Medium-High Frequency (11-16.5kHz)
|
||||
HF, // High Frequency (16.5-22kHz)
|
||||
NB_SUBBANDS
|
||||
};
|
||||
|
||||
#define NB_FILTERS 2
|
||||
#define FILTER_TAPS 16
|
||||
|
||||
typedef struct {
|
||||
int pos;
|
||||
int32_t buffer[2*FILTER_TAPS];
|
||||
} FilterSignal;
|
||||
|
||||
typedef struct {
|
||||
FilterSignal outer_filter_signal[NB_FILTERS];
|
||||
FilterSignal inner_filter_signal[NB_FILTERS][NB_FILTERS];
|
||||
} QMFAnalysis;
|
||||
|
||||
typedef struct {
|
||||
int32_t quantized_sample;
|
||||
int32_t quantized_sample_parity_change;
|
||||
int32_t error;
|
||||
} Quantize;
|
||||
|
||||
typedef struct {
|
||||
int32_t quantization_factor;
|
||||
int32_t factor_select;
|
||||
int32_t reconstructed_difference;
|
||||
} InvertQuantize;
|
||||
|
||||
typedef struct {
|
||||
int32_t prev_sign[2];
|
||||
int32_t s_weight[2];
|
||||
int32_t d_weight[24];
|
||||
int32_t pos;
|
||||
int32_t reconstructed_differences[48];
|
||||
int32_t previous_reconstructed_sample;
|
||||
int32_t predicted_difference;
|
||||
int32_t predicted_sample;
|
||||
} Prediction;
|
||||
|
||||
typedef struct {
|
||||
int32_t codeword_history;
|
||||
int32_t dither_parity;
|
||||
int32_t dither[NB_SUBBANDS];
|
||||
|
||||
QMFAnalysis qmf;
|
||||
Quantize quantize[NB_SUBBANDS];
|
||||
InvertQuantize invert_quantize[NB_SUBBANDS];
|
||||
Prediction prediction[NB_SUBBANDS];
|
||||
} Channel;
|
||||
|
||||
typedef struct {
|
||||
int hd;
|
||||
int block_size;
|
||||
int32_t sync_idx;
|
||||
Channel channels[NB_CHANNELS];
|
||||
AudioFrameQueue afq;
|
||||
} AptXContext;
|
||||
|
||||
typedef const struct {
|
||||
const int32_t *quantize_intervals;
|
||||
const int32_t *invert_quantize_dither_factors;
|
||||
const int32_t *quantize_dither_factors;
|
||||
const int16_t *quantize_factor_select_offset;
|
||||
int tables_size;
|
||||
int32_t factor_max;
|
||||
int32_t prediction_order;
|
||||
} ConstTables;
|
||||
|
||||
extern ConstTables ff_aptx_quant_tables[2][NB_SUBBANDS];
|
||||
|
||||
/* Rounded right shift with optionnal clipping */
|
||||
#define RSHIFT_SIZE(size) \
|
||||
av_always_inline \
|
||||
static int##size##_t rshift##size(int##size##_t value, int shift) \
|
||||
{ \
|
||||
int##size##_t rounding = (int##size##_t)1 << (shift - 1); \
|
||||
int##size##_t mask = ((int##size##_t)1 << (shift + 1)) - 1; \
|
||||
return ((value + rounding) >> shift) - ((value & mask) == rounding); \
|
||||
} \
|
||||
av_always_inline \
|
||||
static int##size##_t rshift##size##_clip24(int##size##_t value, int shift) \
|
||||
{ \
|
||||
return av_clip_intp2(rshift##size(value, shift), 23); \
|
||||
}
|
||||
RSHIFT_SIZE(32)
|
||||
RSHIFT_SIZE(64)
|
||||
|
||||
/*
|
||||
* Convolution filter coefficients for the outer QMF of the QMF tree.
|
||||
* The 2 sets are a mirror of each other.
|
||||
*/
|
||||
static const int32_t aptx_qmf_outer_coeffs[NB_FILTERS][FILTER_TAPS] = {
|
||||
{
|
||||
730, -413, -9611, 43626, -121026, 269973, -585547, 2801966,
|
||||
697128, -160481, 27611, 8478, -10043, 3511, 688, -897,
|
||||
},
|
||||
{
|
||||
-897, 688, 3511, -10043, 8478, 27611, -160481, 697128,
|
||||
2801966, -585547, 269973, -121026, 43626, -9611, -413, 730,
|
||||
},
|
||||
};
|
||||
|
||||
/*
|
||||
* Convolution filter coefficients for the inner QMF of the QMF tree.
|
||||
* The 2 sets are a mirror of each other.
|
||||
*/
|
||||
static const int32_t aptx_qmf_inner_coeffs[NB_FILTERS][FILTER_TAPS] = {
|
||||
{
|
||||
1033, -584, -13592, 61697, -171156, 381799, -828088, 3962579,
|
||||
985888, -226954, 39048, 11990, -14203, 4966, 973, -1268,
|
||||
},
|
||||
{
|
||||
-1268, 973, 4966, -14203, 11990, 39048, -226954, 985888,
|
||||
3962579, -828088, 381799, -171156, 61697, -13592, -584, 1033,
|
||||
},
|
||||
};
|
||||
|
||||
/*
|
||||
* Push one sample into a circular signal buffer.
|
||||
*/
|
||||
av_always_inline
|
||||
static void aptx_qmf_filter_signal_push(FilterSignal *signal, int32_t sample)
|
||||
{
|
||||
signal->buffer[signal->pos ] = sample;
|
||||
signal->buffer[signal->pos+FILTER_TAPS] = sample;
|
||||
signal->pos = (signal->pos + 1) & (FILTER_TAPS - 1);
|
||||
}
|
||||
|
||||
/*
|
||||
* Compute the convolution of the signal with the coefficients, and reduce
|
||||
* to 24 bits by applying the specified right shifting.
|
||||
*/
|
||||
av_always_inline
|
||||
static int32_t aptx_qmf_convolution(FilterSignal *signal,
|
||||
const int32_t coeffs[FILTER_TAPS],
|
||||
int shift)
|
||||
{
|
||||
int32_t *sig = &signal->buffer[signal->pos];
|
||||
int64_t e = 0;
|
||||
int i;
|
||||
|
||||
for (i = 0; i < FILTER_TAPS; i++)
|
||||
e += MUL64(sig[i], coeffs[i]);
|
||||
|
||||
return rshift64_clip24(e, shift);
|
||||
}
|
||||
|
||||
static inline int32_t aptx_quantized_parity(Channel *channel)
|
||||
{
|
||||
int32_t parity = channel->dither_parity;
|
||||
int subband;
|
||||
|
||||
for (subband = 0; subband < NB_SUBBANDS; subband++)
|
||||
parity ^= channel->quantize[subband].quantized_sample;
|
||||
|
||||
return parity & 1;
|
||||
}
|
||||
|
||||
/* For each sample, ensure that the parity of all subbands of all channels
|
||||
* is 0 except once every 8 samples where the parity is forced to 1. */
|
||||
static inline int aptx_check_parity(Channel channels[NB_CHANNELS], int32_t *idx)
|
||||
{
|
||||
int32_t parity = aptx_quantized_parity(&channels[LEFT])
|
||||
^ aptx_quantized_parity(&channels[RIGHT]);
|
||||
|
||||
int eighth = *idx == 7;
|
||||
*idx = (*idx + 1) & 7;
|
||||
|
||||
return parity ^ eighth;
|
||||
}
|
||||
|
||||
void ff_aptx_invert_quantize_and_prediction(Channel *channel, int hd);
|
||||
void ff_aptx_generate_dither(Channel *channel);
|
||||
|
||||
int ff_aptx_init(AVCodecContext *avctx);
|
||||
|
||||
#endif /* AVCODEC_APTX_H */
|
204
libavcodec/aptxdec.c
Normal file
204
libavcodec/aptxdec.c
Normal file
@ -0,0 +1,204 @@
|
||||
/*
|
||||
* Audio Processing Technology codec for Bluetooth (aptX)
|
||||
*
|
||||
* Copyright (C) 2017 Aurelien Jacobs <aurel@gnuage.org>
|
||||
*
|
||||
* This file is part of FFmpeg.
|
||||
*
|
||||
* FFmpeg is free software; you can redistribute it and/or
|
||||
* modify it under the terms of the GNU Lesser General Public
|
||||
* License as published by the Free Software Foundation; either
|
||||
* version 2.1 of the License, or (at your option) any later version.
|
||||
*
|
||||
* FFmpeg is distributed in the hope that it will be useful,
|
||||
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||||
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
||||
* Lesser General Public License for more details.
|
||||
*
|
||||
* You should have received a copy of the GNU Lesser General Public
|
||||
* License along with FFmpeg; if not, write to the Free Software
|
||||
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
||||
*/
|
||||
|
||||
#include "aptx.h"
|
||||
|
||||
/*
|
||||
* Half-band QMF synthesis filter realized with a polyphase FIR filter.
|
||||
* Join 2 subbands and upsample by 2.
|
||||
* So for each 2 subbands sample that goes in, a pair of samples goes out.
|
||||
*/
|
||||
av_always_inline
|
||||
static void aptx_qmf_polyphase_synthesis(FilterSignal signal[NB_FILTERS],
|
||||
const int32_t coeffs[NB_FILTERS][FILTER_TAPS],
|
||||
int shift,
|
||||
int32_t low_subband_input,
|
||||
int32_t high_subband_input,
|
||||
int32_t samples[NB_FILTERS])
|
||||
{
|
||||
int32_t subbands[NB_FILTERS];
|
||||
int i;
|
||||
|
||||
subbands[0] = low_subband_input + high_subband_input;
|
||||
subbands[1] = low_subband_input - high_subband_input;
|
||||
|
||||
for (i = 0; i < NB_FILTERS; i++) {
|
||||
aptx_qmf_filter_signal_push(&signal[i], subbands[1-i]);
|
||||
samples[i] = aptx_qmf_convolution(&signal[i], coeffs[i], shift);
|
||||
}
|
||||
}
|
||||
|
||||
/*
|
||||
* Two stage QMF synthesis tree.
|
||||
* Join 4 subbands and upsample by 4.
|
||||
* So for each 4 subbands sample that goes in, a group of 4 samples goes out.
|
||||
*/
|
||||
static void aptx_qmf_tree_synthesis(QMFAnalysis *qmf,
|
||||
int32_t subband_samples[4],
|
||||
int32_t samples[4])
|
||||
{
|
||||
int32_t intermediate_samples[4];
|
||||
int i;
|
||||
|
||||
/* Join 4 subbands into 2 intermediate subbands upsampled to 2 samples. */
|
||||
for (i = 0; i < 2; i++)
|
||||
aptx_qmf_polyphase_synthesis(qmf->inner_filter_signal[i],
|
||||
aptx_qmf_inner_coeffs, 22,
|
||||
subband_samples[2*i+0],
|
||||
subband_samples[2*i+1],
|
||||
&intermediate_samples[2*i]);
|
||||
|
||||
/* Join 2 samples from intermediate subbands upsampled to 4 samples. */
|
||||
for (i = 0; i < 2; i++)
|
||||
aptx_qmf_polyphase_synthesis(qmf->outer_filter_signal,
|
||||
aptx_qmf_outer_coeffs, 21,
|
||||
intermediate_samples[0+i],
|
||||
intermediate_samples[2+i],
|
||||
&samples[2*i]);
|
||||
}
|
||||
|
||||
|
||||
static void aptx_decode_channel(Channel *channel, int32_t samples[4])
|
||||
{
|
||||
int32_t subband_samples[4];
|
||||
int subband;
|
||||
for (subband = 0; subband < NB_SUBBANDS; subband++)
|
||||
subband_samples[subband] = channel->prediction[subband].previous_reconstructed_sample;
|
||||
aptx_qmf_tree_synthesis(&channel->qmf, subband_samples, samples);
|
||||
}
|
||||
|
||||
static void aptx_unpack_codeword(Channel *channel, uint16_t codeword)
|
||||
{
|
||||
channel->quantize[0].quantized_sample = sign_extend(codeword >> 0, 7);
|
||||
channel->quantize[1].quantized_sample = sign_extend(codeword >> 7, 4);
|
||||
channel->quantize[2].quantized_sample = sign_extend(codeword >> 11, 2);
|
||||
channel->quantize[3].quantized_sample = sign_extend(codeword >> 13, 3);
|
||||
channel->quantize[3].quantized_sample = (channel->quantize[3].quantized_sample & ~1)
|
||||
| aptx_quantized_parity(channel);
|
||||
}
|
||||
|
||||
static void aptxhd_unpack_codeword(Channel *channel, uint32_t codeword)
|
||||
{
|
||||
channel->quantize[0].quantized_sample = sign_extend(codeword >> 0, 9);
|
||||
channel->quantize[1].quantized_sample = sign_extend(codeword >> 9, 6);
|
||||
channel->quantize[2].quantized_sample = sign_extend(codeword >> 15, 4);
|
||||
channel->quantize[3].quantized_sample = sign_extend(codeword >> 19, 5);
|
||||
channel->quantize[3].quantized_sample = (channel->quantize[3].quantized_sample & ~1)
|
||||
| aptx_quantized_parity(channel);
|
||||
}
|
||||
|
||||
static int aptx_decode_samples(AptXContext *ctx,
|
||||
const uint8_t *input,
|
||||
int32_t samples[NB_CHANNELS][4])
|
||||
{
|
||||
int channel, ret;
|
||||
|
||||
for (channel = 0; channel < NB_CHANNELS; channel++) {
|
||||
ff_aptx_generate_dither(&ctx->channels[channel]);
|
||||
|
||||
if (ctx->hd)
|
||||
aptxhd_unpack_codeword(&ctx->channels[channel],
|
||||
AV_RB24(input + 3*channel));
|
||||
else
|
||||
aptx_unpack_codeword(&ctx->channels[channel],
|
||||
AV_RB16(input + 2*channel));
|
||||
ff_aptx_invert_quantize_and_prediction(&ctx->channels[channel], ctx->hd);
|
||||
}
|
||||
|
||||
ret = aptx_check_parity(ctx->channels, &ctx->sync_idx);
|
||||
|
||||
for (channel = 0; channel < NB_CHANNELS; channel++)
|
||||
aptx_decode_channel(&ctx->channels[channel], samples[channel]);
|
||||
|
||||
return ret;
|
||||
}
|
||||
|
||||
static int aptx_decode_frame(AVCodecContext *avctx, void *data,
|
||||
int *got_frame_ptr, AVPacket *avpkt)
|
||||
{
|
||||
AptXContext *s = avctx->priv_data;
|
||||
AVFrame *frame = data;
|
||||
int pos, opos, channel, sample, ret;
|
||||
|
||||
if (avpkt->size < s->block_size) {
|
||||
av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
|
||||
return AVERROR_INVALIDDATA;
|
||||
}
|
||||
|
||||
/* get output buffer */
|
||||
frame->channels = NB_CHANNELS;
|
||||
frame->format = AV_SAMPLE_FMT_S32P;
|
||||
frame->nb_samples = 4 * avpkt->size / s->block_size;
|
||||
if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
|
||||
return ret;
|
||||
|
||||
for (pos = 0, opos = 0; opos < frame->nb_samples; pos += s->block_size, opos += 4) {
|
||||
int32_t samples[NB_CHANNELS][4];
|
||||
|
||||
if (aptx_decode_samples(s, &avpkt->data[pos], samples)) {
|
||||
av_log(avctx, AV_LOG_ERROR, "Synchronization error\n");
|
||||
return AVERROR_INVALIDDATA;
|
||||
}
|
||||
|
||||
for (channel = 0; channel < NB_CHANNELS; channel++)
|
||||
for (sample = 0; sample < 4; sample++)
|
||||
AV_WN32A(&frame->data[channel][4*(opos+sample)],
|
||||
samples[channel][sample] * 256);
|
||||
}
|
||||
|
||||
*got_frame_ptr = 1;
|
||||
return s->block_size * frame->nb_samples / 4;
|
||||
}
|
||||
|
||||
#if CONFIG_APTX_DECODER
|
||||
AVCodec ff_aptx_decoder = {
|
||||
.name = "aptx",
|
||||
.long_name = NULL_IF_CONFIG_SMALL("aptX (Audio Processing Technology for Bluetooth)"),
|
||||
.type = AVMEDIA_TYPE_AUDIO,
|
||||
.id = AV_CODEC_ID_APTX,
|
||||
.priv_data_size = sizeof(AptXContext),
|
||||
.init = ff_aptx_init,
|
||||
.decode = aptx_decode_frame,
|
||||
.capabilities = AV_CODEC_CAP_DR1,
|
||||
.caps_internal = FF_CODEC_CAP_INIT_THREADSAFE,
|
||||
.channel_layouts = (const uint64_t[]) { AV_CH_LAYOUT_STEREO, 0},
|
||||
.sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S32P,
|
||||
AV_SAMPLE_FMT_NONE },
|
||||
};
|
||||
#endif
|
||||
|
||||
#if CONFIG_APTX_HD_DECODER
|
||||
AVCodec ff_aptx_hd_decoder = {
|
||||
.name = "aptx_hd",
|
||||
.long_name = NULL_IF_CONFIG_SMALL("aptX HD (Audio Processing Technology for Bluetooth)"),
|
||||
.type = AVMEDIA_TYPE_AUDIO,
|
||||
.id = AV_CODEC_ID_APTX_HD,
|
||||
.priv_data_size = sizeof(AptXContext),
|
||||
.init = ff_aptx_init,
|
||||
.decode = aptx_decode_frame,
|
||||
.capabilities = AV_CODEC_CAP_DR1,
|
||||
.caps_internal = FF_CODEC_CAP_INIT_THREADSAFE,
|
||||
.channel_layouts = (const uint64_t[]) { AV_CH_LAYOUT_STEREO, 0},
|
||||
.sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S32P,
|
||||
AV_SAMPLE_FMT_NONE },
|
||||
};
|
||||
#endif
|
278
libavcodec/aptxenc.c
Normal file
278
libavcodec/aptxenc.c
Normal file
@ -0,0 +1,278 @@
|
||||
/*
|
||||
* Audio Processing Technology codec for Bluetooth (aptX)
|
||||
*
|
||||
* Copyright (C) 2017 Aurelien Jacobs <aurel@gnuage.org>
|
||||
*
|
||||
* This file is part of FFmpeg.
|
||||
*
|
||||
* FFmpeg is free software; you can redistribute it and/or
|
||||
* modify it under the terms of the GNU Lesser General Public
|
||||
* License as published by the Free Software Foundation; either
|
||||
* version 2.1 of the License, or (at your option) any later version.
|
||||
*
|
||||
* FFmpeg is distributed in the hope that it will be useful,
|
||||
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||||
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
||||
* Lesser General Public License for more details.
|
||||
*
|
||||
* You should have received a copy of the GNU Lesser General Public
|
||||
* License along with FFmpeg; if not, write to the Free Software
|
||||
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
||||
*/
|
||||
|
||||
#include "aptx.h"
|
||||
|
||||
/*
|
||||
* Half-band QMF analysis filter realized with a polyphase FIR filter.
|
||||
* Split into 2 subbands and downsample by 2.
|
||||
* So for each pair of samples that goes in, one sample goes out,
|
||||
* split into 2 separate subbands.
|
||||
*/
|
||||
av_always_inline
|
||||
static void aptx_qmf_polyphase_analysis(FilterSignal signal[NB_FILTERS],
|
||||
const int32_t coeffs[NB_FILTERS][FILTER_TAPS],
|
||||
int shift,
|
||||
int32_t samples[NB_FILTERS],
|
||||
int32_t *low_subband_output,
|
||||
int32_t *high_subband_output)
|
||||
{
|
||||
int32_t subbands[NB_FILTERS];
|
||||
int i;
|
||||
|
||||
for (i = 0; i < NB_FILTERS; i++) {
|
||||
aptx_qmf_filter_signal_push(&signal[i], samples[NB_FILTERS-1-i]);
|
||||
subbands[i] = aptx_qmf_convolution(&signal[i], coeffs[i], shift);
|
||||
}
|
||||
|
||||
*low_subband_output = av_clip_intp2(subbands[0] + subbands[1], 23);
|
||||
*high_subband_output = av_clip_intp2(subbands[0] - subbands[1], 23);
|
||||
}
|
||||
|
||||
/*
|
||||
* Two stage QMF analysis tree.
|
||||
* Split 4 input samples into 4 subbands and downsample by 4.
|
||||
* So for each group of 4 samples that goes in, one sample goes out,
|
||||
* split into 4 separate subbands.
|
||||
*/
|
||||
static void aptx_qmf_tree_analysis(QMFAnalysis *qmf,
|
||||
int32_t samples[4],
|
||||
int32_t subband_samples[4])
|
||||
{
|
||||
int32_t intermediate_samples[4];
|
||||
int i;
|
||||
|
||||
/* Split 4 input samples into 2 intermediate subbands downsampled to 2 samples */
|
||||
for (i = 0; i < 2; i++)
|
||||
aptx_qmf_polyphase_analysis(qmf->outer_filter_signal,
|
||||
aptx_qmf_outer_coeffs, 23,
|
||||
&samples[2*i],
|
||||
&intermediate_samples[0+i],
|
||||
&intermediate_samples[2+i]);
|
||||
|
||||
/* Split 2 intermediate subband samples into 4 final subbands downsampled to 1 sample */
|
||||
for (i = 0; i < 2; i++)
|
||||
aptx_qmf_polyphase_analysis(qmf->inner_filter_signal[i],
|
||||
aptx_qmf_inner_coeffs, 23,
|
||||
&intermediate_samples[2*i],
|
||||
&subband_samples[2*i+0],
|
||||
&subband_samples[2*i+1]);
|
||||
}
|
||||
|
||||
av_always_inline
|
||||
static int32_t aptx_bin_search(int32_t value, int32_t factor,
|
||||
const int32_t *intervals, int32_t nb_intervals)
|
||||
{
|
||||
int32_t idx = 0;
|
||||
int i;
|
||||
|
||||
for (i = nb_intervals >> 1; i > 0; i >>= 1)
|
||||
if (MUL64(factor, intervals[idx + i]) <= ((int64_t)value << 24))
|
||||
idx += i;
|
||||
|
||||
return idx;
|
||||
}
|
||||
|
||||
static void aptx_quantize_difference(Quantize *quantize,
|
||||
int32_t sample_difference,
|
||||
int32_t dither,
|
||||
int32_t quantization_factor,
|
||||
ConstTables *tables)
|
||||
{
|
||||
const int32_t *intervals = tables->quantize_intervals;
|
||||
int32_t quantized_sample, dithered_sample, parity_change;
|
||||
int32_t d, mean, interval, inv, sample_difference_abs;
|
||||
int64_t error;
|
||||
|
||||
sample_difference_abs = FFABS(sample_difference);
|
||||
sample_difference_abs = FFMIN(sample_difference_abs, (1 << 23) - 1);
|
||||
|
||||
quantized_sample = aptx_bin_search(sample_difference_abs >> 4,
|
||||
quantization_factor,
|
||||
intervals, tables->tables_size);
|
||||
|
||||
d = rshift32_clip24(MULH(dither, dither), 7) - (1 << 23);
|
||||
d = rshift64(MUL64(d, tables->quantize_dither_factors[quantized_sample]), 23);
|
||||
|
||||
intervals += quantized_sample;
|
||||
mean = (intervals[1] + intervals[0]) / 2;
|
||||
interval = (intervals[1] - intervals[0]) * (-(sample_difference < 0) | 1);
|
||||
|
||||
dithered_sample = rshift64_clip24(MUL64(dither, interval) + ((int64_t)av_clip_intp2(mean + d, 23) << 32), 32);
|
||||
error = ((int64_t)sample_difference_abs << 20) - MUL64(dithered_sample, quantization_factor);
|
||||
quantize->error = FFABS(rshift64(error, 23));
|
||||
|
||||
parity_change = quantized_sample;
|
||||
if (error < 0)
|
||||
quantized_sample--;
|
||||
else
|
||||
parity_change--;
|
||||
|
||||
inv = -(sample_difference < 0);
|
||||
quantize->quantized_sample = quantized_sample ^ inv;
|
||||
quantize->quantized_sample_parity_change = parity_change ^ inv;
|
||||
}
|
||||
|
||||
static void aptx_encode_channel(Channel *channel, int32_t samples[4], int hd)
|
||||
{
|
||||
int32_t subband_samples[4];
|
||||
int subband;
|
||||
aptx_qmf_tree_analysis(&channel->qmf, samples, subband_samples);
|
||||
ff_aptx_generate_dither(channel);
|
||||
for (subband = 0; subband < NB_SUBBANDS; subband++) {
|
||||
int32_t diff = av_clip_intp2(subband_samples[subband] - channel->prediction[subband].predicted_sample, 23);
|
||||
aptx_quantize_difference(&channel->quantize[subband], diff,
|
||||
channel->dither[subband],
|
||||
channel->invert_quantize[subband].quantization_factor,
|
||||
&ff_aptx_quant_tables[hd][subband]);
|
||||
}
|
||||
}
|
||||
|
||||
static void aptx_insert_sync(Channel channels[NB_CHANNELS], int32_t *idx)
|
||||
{
|
||||
if (aptx_check_parity(channels, idx)) {
|
||||
int i;
|
||||
Channel *c;
|
||||
static const int map[] = { 1, 2, 0, 3 };
|
||||
Quantize *min = &channels[NB_CHANNELS-1].quantize[map[0]];
|
||||
for (c = &channels[NB_CHANNELS-1]; c >= channels; c--)
|
||||
for (i = 0; i < NB_SUBBANDS; i++)
|
||||
if (c->quantize[map[i]].error < min->error)
|
||||
min = &c->quantize[map[i]];
|
||||
|
||||
/* Forcing the desired parity is done by offsetting by 1 the quantized
|
||||
* sample from the subband featuring the smallest quantization error. */
|
||||
min->quantized_sample = min->quantized_sample_parity_change;
|
||||
}
|
||||
}
|
||||
|
||||
static uint16_t aptx_pack_codeword(Channel *channel)
|
||||
{
|
||||
int32_t parity = aptx_quantized_parity(channel);
|
||||
return (((channel->quantize[3].quantized_sample & 0x06) | parity) << 13)
|
||||
| (((channel->quantize[2].quantized_sample & 0x03) ) << 11)
|
||||
| (((channel->quantize[1].quantized_sample & 0x0F) ) << 7)
|
||||
| (((channel->quantize[0].quantized_sample & 0x7F) ) << 0);
|
||||
}
|
||||
|
||||
static uint32_t aptxhd_pack_codeword(Channel *channel)
|
||||
{
|
||||
int32_t parity = aptx_quantized_parity(channel);
|
||||
return (((channel->quantize[3].quantized_sample & 0x01E) | parity) << 19)
|
||||
| (((channel->quantize[2].quantized_sample & 0x00F) ) << 15)
|
||||
| (((channel->quantize[1].quantized_sample & 0x03F) ) << 9)
|
||||
| (((channel->quantize[0].quantized_sample & 0x1FF) ) << 0);
|
||||
}
|
||||
|
||||
static void aptx_encode_samples(AptXContext *ctx,
|
||||
int32_t samples[NB_CHANNELS][4],
|
||||
uint8_t *output)
|
||||
{
|
||||
int channel;
|
||||
for (channel = 0; channel < NB_CHANNELS; channel++)
|
||||
aptx_encode_channel(&ctx->channels[channel], samples[channel], ctx->hd);
|
||||
|
||||
aptx_insert_sync(ctx->channels, &ctx->sync_idx);
|
||||
|
||||
for (channel = 0; channel < NB_CHANNELS; channel++) {
|
||||
ff_aptx_invert_quantize_and_prediction(&ctx->channels[channel], ctx->hd);
|
||||
if (ctx->hd)
|
||||
AV_WB24(output + 3*channel,
|
||||
aptxhd_pack_codeword(&ctx->channels[channel]));
|
||||
else
|
||||
AV_WB16(output + 2*channel,
|
||||
aptx_pack_codeword(&ctx->channels[channel]));
|
||||
}
|
||||
}
|
||||
|
||||
static int aptx_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
|
||||
const AVFrame *frame, int *got_packet_ptr)
|
||||
{
|
||||
AptXContext *s = avctx->priv_data;
|
||||
int pos, ipos, channel, sample, output_size, ret;
|
||||
|
||||
if ((ret = ff_af_queue_add(&s->afq, frame)) < 0)
|
||||
return ret;
|
||||
|
||||
output_size = s->block_size * frame->nb_samples/4;
|
||||
if ((ret = ff_alloc_packet2(avctx, avpkt, output_size, 0)) < 0)
|
||||
return ret;
|
||||
|
||||
for (pos = 0, ipos = 0; pos < output_size; pos += s->block_size, ipos += 4) {
|
||||
int32_t samples[NB_CHANNELS][4];
|
||||
|
||||
for (channel = 0; channel < NB_CHANNELS; channel++)
|
||||
for (sample = 0; sample < 4; sample++)
|
||||
samples[channel][sample] = (int32_t)AV_RN32A(&frame->data[channel][4*(ipos+sample)]) >> 8;
|
||||
|
||||
aptx_encode_samples(s, samples, avpkt->data + pos);
|
||||
}
|
||||
|
||||
ff_af_queue_remove(&s->afq, frame->nb_samples, &avpkt->pts, &avpkt->duration);
|
||||
*got_packet_ptr = 1;
|
||||
return 0;
|
||||
}
|
||||
|
||||
static av_cold int aptx_close(AVCodecContext *avctx)
|
||||
{
|
||||
AptXContext *s = avctx->priv_data;
|
||||
ff_af_queue_close(&s->afq);
|
||||
return 0;
|
||||
}
|
||||
|
||||
#if CONFIG_APTX_ENCODER
|
||||
AVCodec ff_aptx_encoder = {
|
||||
.name = "aptx",
|
||||
.long_name = NULL_IF_CONFIG_SMALL("aptX (Audio Processing Technology for Bluetooth)"),
|
||||
.type = AVMEDIA_TYPE_AUDIO,
|
||||
.id = AV_CODEC_ID_APTX,
|
||||
.priv_data_size = sizeof(AptXContext),
|
||||
.init = ff_aptx_init,
|
||||
.encode2 = aptx_encode_frame,
|
||||
.close = aptx_close,
|
||||
.capabilities = AV_CODEC_CAP_SMALL_LAST_FRAME,
|
||||
.caps_internal = FF_CODEC_CAP_INIT_THREADSAFE,
|
||||
.channel_layouts = (const uint64_t[]) { AV_CH_LAYOUT_STEREO, 0},
|
||||
.sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S32P,
|
||||
AV_SAMPLE_FMT_NONE },
|
||||
.supported_samplerates = (const int[]) {8000, 16000, 24000, 32000, 44100, 48000, 0},
|
||||
};
|
||||
#endif
|
||||
|
||||
#if CONFIG_APTX_HD_ENCODER
|
||||
AVCodec ff_aptx_hd_encoder = {
|
||||
.name = "aptx_hd",
|
||||
.long_name = NULL_IF_CONFIG_SMALL("aptX HD (Audio Processing Technology for Bluetooth)"),
|
||||
.type = AVMEDIA_TYPE_AUDIO,
|
||||
.id = AV_CODEC_ID_APTX_HD,
|
||||
.priv_data_size = sizeof(AptXContext),
|
||||
.init = ff_aptx_init,
|
||||
.encode2 = aptx_encode_frame,
|
||||
.close = aptx_close,
|
||||
.capabilities = AV_CODEC_CAP_SMALL_LAST_FRAME,
|
||||
.caps_internal = FF_CODEC_CAP_INIT_THREADSAFE,
|
||||
.channel_layouts = (const uint64_t[]) { AV_CH_LAYOUT_STEREO, 0},
|
||||
.sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S32P,
|
||||
AV_SAMPLE_FMT_NONE },
|
||||
.supported_samplerates = (const int[]) {8000, 16000, 24000, 32000, 44100, 48000, 0},
|
||||
};
|
||||
#endif
|
Loading…
Reference in New Issue
Block a user