mirrors/librempeg
1
0
Fork 0
mirror of https://github.com/librempeg/librempeg synced 2024-11-22 00:51:37 +00:00
Code Issues Packages Projects Releases Wiki Activity

avcodec: add QOA encoder

Browse Source
This commit is contained in:
Paul B Mahol 2024-05-25 22:26:02 +02:00
parent 7d3a0051dc
commit 3c231cb19d
5 changed files with 234 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
Changelog
View File

@ -11,6 +11,7 @@ version <next>:
- qsv_params option added for QSV encoders
- dcblock audio filter
- amae audio filter
- QOA encoder and muxer
version 7.0:
- DXV DXT1 encoder

2
doc/general_contents.texi
View File

@ -607,6 +607,7 @@ library:
@item TechnoTrend PVA @tab @tab X
@tab Used by TechnoTrend DVB PCI boards.
@item QCP @tab @tab X
@item QOA @tab X @tab X
@item raw ADTS (AAC) @tab X @tab X
@item raw AC-3 @tab X @tab X
@item raw AMR-NB @tab @tab X
@ -1364,6 +1365,7 @@ following image formats are supported:
@item QDesign Music Codec 1 @tab @tab X
@item QDesign Music Codec 2 @tab @tab X
@tab There are still some distortions.
@item QOA (Quite OK Audio) @tab X @tab X
@item RealAudio 1.0 (14.4K) @tab X @tab X
@tab Real 14400 bit/s codec
@item RealAudio 2.0 (28.8K) @tab @tab X

1
libavcodec/Makefile
View File

@ -625,6 +625,7 @@ OBJS-$(CONFIG_QDM2_DECODER) += qdm2.o
OBJS-$(CONFIG_QDMC_DECODER) += qdmc.o
OBJS-$(CONFIG_QDRAW_DECODER) += qdrw.o
OBJS-$(CONFIG_QOA_DECODER) += qoadec.o
OBJS-$(CONFIG_QOA_ENCODER) += qoaenc.o
OBJS-$(CONFIG_QOI_DECODER) += qoidec.o
OBJS-$(CONFIG_QOI_ENCODER) += qoienc.o
OBJS-$(CONFIG_QPEG_DECODER) += qpeg.o

1
libavcodec/allcodecs.c
View File

@ -515,6 +515,7 @@ extern const FFCodec ff_paf_audio_decoder;
extern const FFCodec ff_qcelp_decoder;
extern const FFCodec ff_qdm2_decoder;
extern const FFCodec ff_qdmc_decoder;
extern const FFCodec ff_qoa_encoder;
extern const FFCodec ff_qoa_decoder;
extern const FFCodec ff_ra_144_encoder;
extern const FFCodec ff_ra_144_decoder;

229
libavcodec/qoaenc.c Normal file
View File

@ -0,0 +1,229 @@
/*
* 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 "avcodec.h"
#include "bytestream.h"
#include "codec_internal.h"
#include "encode.h"
#define QOA_SLICE_LEN 20
#define QOA_LMS_LEN 4
typedef struct QOAChannel {
int16_t history[QOA_LMS_LEN];
int16_t weights[QOA_LMS_LEN];
int error;
} QOAChannel;
typedef struct QOAContext {
QOAChannel ch[255];
} QOAContext;
static const uint8_t qoa_quant_tab[17] = {
7, 7, 7, 5, 5, 3, 3, 1, 0, 0, 2, 2, 4, 4, 6, 6, 6,
};
static const int16_t qoa_dequant_tab[16][8] = {
{ 1, -1, 3, -3, 5, -5, 7, -7},
{ 5, -5, 18, -18, 32, -32, 49, -49},
{ 16, -16, 53, -53, 95, -95, 147, -147},
{ 34, -34, 113, -113, 203, -203, 315, -315},
{ 63, -63, 210, -210, 378, -378, 588, -588},
{ 104, -104, 345, -345, 621, -621, 966, -966},
{ 158, -158, 528, -528, 950, -950, 1477, -1477},
{ 228, -228, 760, -760, 1368, -1368, 2128, -2128},
{ 316, -316, 1053, -1053, 1895, -1895, 2947, -2947},
{ 422, -422, 1405, -1405, 2529, -2529, 3934, -3934},
{ 548, -548, 1828, -1828, 3290, -3290, 5117, -5117},
{ 696, -696, 2320, -2320, 4176, -4176, 6496, -6496},
{ 868, -868, 2893, -2893, 5207, -5207, 8099, -8099},
{1064, -1064, 3548, -3548, 6386, -6386, 9933, -9933},
{1286, -1286, 4288, -4288, 7718, -7718, 12005, -12005},
{1536, -1536, 5120, -5120, 9216, -9216, 14336, -14336},
};
static const int32_t qoa_reciprocal_tab[16] = {
65536, 9363, 3121, 1457, 781, 475, 311, 216, 156, 117, 90, 71, 57, 47, 39, 32
};
static av_cold int qoa_encode_init(AVCodecContext *avctx)
{
const int nb_channels = avctx->ch_layout.nb_channels;
QOAContext *s = avctx->priv_data;
if (nb_channels > 255) {
av_log(avctx, AV_LOG_ERROR, "Invalid number of channels\n");
return AVERROR(EINVAL);
}
avctx->frame_size = FFMIN((((1<<16)-1)-8-QOA_LMS_LEN*4*nb_channels) / (QOA_SLICE_LEN*8), 256) * QOA_SLICE_LEN;
for (int ch = 0; ch < nb_channels; ch++) {
s->ch[ch].weights[0] = s->ch[ch].weights[1] = 0;
s->ch[ch].weights[2] = -(1<<13);
s->ch[ch].weights[3] = (1<<14);
}
return 0;
}
static ptrdiff_t qoa_encode_header(AVCodecContext *avctx, QOAContext *s,
int nb_samples, uint32_t fsize,
uint8_t *buf, int bufsize)
{
const int nb_channels = avctx->ch_layout.nb_channels;
uint8_t *orig = buf;
bytestream_put_byte(&buf, avctx->ch_layout.nb_channels);
bytestream_put_be24(&buf, avctx->sample_rate);
bytestream_put_be16(&buf, nb_samples);
bytestream_put_be16(&buf, fsize);
for (int ch = 0; ch < nb_channels; ch++) {
QOAChannel *qch = &s->ch[ch];
for (int n = 0; n < QOA_LMS_LEN; n++)
bytestream_put_be16(&buf, qch->history[n]);
for (int n = 0; n < QOA_LMS_LEN; n++)
bytestream_put_be16(&buf, qch->weights[n]);
}
return buf - orig;
}
static int qoa_lms_predict(QOAChannel *lms)
{
int prediction = 0;
for (int i = 0; i < QOA_LMS_LEN; i++)
prediction += (int32_t)lms->weights[i] * lms->history[i];
prediction >>= 13;
return prediction;
}
static void qoa_lms_update(QOAChannel *lms, int sample, int residual)
{
int delta = residual >> 4;
for (int i = 0; i < QOA_LMS_LEN; i++)
lms->weights[i] += lms->history[i] < 0 ? -delta : delta;
for (int i = 0; i < QOA_LMS_LEN-1; i++)
lms->history[i] = lms->history[i+1];
lms->history[QOA_LMS_LEN-1] = sample;
}
static inline int qoa_div(int v, int scalefactor)
{
int64_t reciprocal = qoa_reciprocal_tab[scalefactor];
int n = (v * reciprocal + (1 << 15)) / (1 << 16);
n = n + ((v > 0) - (v < 0)) - ((n > 0) - (n < 0)); /* round away from 0 */
return n;
}
static int qoa_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
const AVFrame *frame, int *got_packet_ptr)
{
QOAContext *s = avctx->priv_data;
const int nb_channels = avctx->ch_layout.nb_channels;
const int16_t *samples = (const int16_t *)frame->data[0];
int prev_scalefactor[255] = {0};
int64_t out_size;
uint8_t *dst;
int ret;
out_size = 8 + QOA_LMS_LEN * 4 * nb_channels + nb_channels *
8LL * ((frame->nb_samples + QOA_SLICE_LEN - 1) / QOA_SLICE_LEN);
if (out_size > UINT16_MAX)
return AVERROR(EINVAL);
if ((ret = ff_get_encode_buffer(avctx, avpkt, out_size, 0)) < 0)
return ret;
dst = avpkt->data;
dst += qoa_encode_header(avctx, s, frame->nb_samples,
out_size, dst, avpkt->size);
for (int sample_index = 0; sample_index < frame->nb_samples;
sample_index += QOA_SLICE_LEN) {
for (int ch = 0; ch < nb_channels; ch++) {
QOAChannel *lms = &s->ch[ch];
int slice_len = FFMIN(QOA_SLICE_LEN, frame->nb_samples - sample_index);
int slice_start = sample_index * nb_channels + ch;
int slice_end = (sample_index + slice_len) * nb_channels + ch;
uint64_t best_error = UINT64_MAX, best_slice = 0;
QOAChannel best_lms = {0};
int best_scalefactor = 0;
for (int sfi = 0; sfi < 16; sfi++) {
int scalefactor = (sfi + prev_scalefactor[ch]) % 16;
uint64_t slice = scalefactor;
uint64_t current_error = 0;
QOAChannel new_lms = *lms;
for (int si = slice_start; si < slice_end; si += nb_channels) {
int sample = samples[si] - new_lms.error;
int predicted = qoa_lms_predict(&new_lms);
int residual = sample - predicted;
int scaled = qoa_div(residual, scalefactor);
int clamped = av_clip(scaled, -8, 8);
int quantized = qoa_quant_tab[clamped + 8];
int dequantized = qoa_dequant_tab[scalefactor][quantized];
int reconstructed = av_clip_int16(predicted + dequantized);
int error = sample - reconstructed;
current_error += (int64_t)error * error;
if (current_error > best_error)
break;
qoa_lms_update(&new_lms, reconstructed, dequantized);
new_lms.error = error;
slice = (slice << 3) | quantized;
}
if (current_error < best_error) {
best_error = current_error;
best_slice = slice;
best_lms = new_lms;
best_scalefactor = scalefactor;
}
}
prev_scalefactor[ch] = best_scalefactor;
memcpy(lms, &best_lms, sizeof(*lms));
best_slice <<= (QOA_SLICE_LEN - slice_len) * 3;
bytestream_put_be64(&dst, best_slice);
}
}
*got_packet_ptr = 1;
return 0;
}
const FFCodec ff_qoa_encoder = {
.p.name = "qoa",
CODEC_LONG_NAME("QOA (Quite OK Audio)"),
.p.type = AVMEDIA_TYPE_AUDIO,
.p.id = AV_CODEC_ID_QOA,
.p.capabilities = AV_CODEC_CAP_DR1 |
AV_CODEC_CAP_SMALL_LAST_FRAME,
.priv_data_size = sizeof(QOAContext),
.init = qoa_encode_init,
FF_CODEC_ENCODE_CB(qoa_encode_frame),
.p.sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16,
AV_SAMPLE_FMT_NONE },
};