You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 

665 lines
25 KiB

/*
* Copyright (c) 2012 Justin Ruggles <justin.ruggles@gmail.com>
*
* This file is part of Libav.
*
* Libav is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* Libav is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with Libav; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#include "libavutil/common.h"
#include "libavutil/dict.h"
#include "libavutil/error.h"
#include "libavutil/log.h"
#include "libavutil/mem.h"
#include "libavutil/opt.h"
#include "avresample.h"
#include "internal.h"
#include "audio_data.h"
#include "audio_convert.h"
#include "audio_mix.h"
#include "resample.h"
int avresample_open(AVAudioResampleContext *avr)
{
int ret;
if (avresample_is_open(avr)) {
av_log(avr, AV_LOG_ERROR, "The resampling context is already open.\n");
return AVERROR(EINVAL);
}
/* set channel mixing parameters */
avr->in_channels = av_get_channel_layout_nb_channels(avr->in_channel_layout);
if (avr->in_channels <= 0 || avr->in_channels > AVRESAMPLE_MAX_CHANNELS) {
av_log(avr, AV_LOG_ERROR, "Invalid input channel layout: %"PRIu64"\n",
avr->in_channel_layout);
return AVERROR(EINVAL);
}
avr->out_channels = av_get_channel_layout_nb_channels(avr->out_channel_layout);
if (avr->out_channels <= 0 || avr->out_channels > AVRESAMPLE_MAX_CHANNELS) {
av_log(avr, AV_LOG_ERROR, "Invalid output channel layout: %"PRIu64"\n",
avr->out_channel_layout);
return AVERROR(EINVAL);
}
avr->resample_channels = FFMIN(avr->in_channels, avr->out_channels);
avr->downmix_needed = avr->in_channels > avr->out_channels;
avr->upmix_needed = avr->out_channels > avr->in_channels ||
(!avr->downmix_needed && (avr->mix_matrix ||
avr->in_channel_layout != avr->out_channel_layout));
avr->mixing_needed = avr->downmix_needed || avr->upmix_needed;
/* set resampling parameters */
avr->resample_needed = avr->in_sample_rate != avr->out_sample_rate ||
avr->force_resampling;
/* select internal sample format if not specified by the user */
if (avr->internal_sample_fmt == AV_SAMPLE_FMT_NONE &&
(avr->mixing_needed || avr->resample_needed)) {
enum AVSampleFormat in_fmt = av_get_planar_sample_fmt(avr->in_sample_fmt);
enum AVSampleFormat out_fmt = av_get_planar_sample_fmt(avr->out_sample_fmt);
int max_bps = FFMAX(av_get_bytes_per_sample(in_fmt),
av_get_bytes_per_sample(out_fmt));
if (max_bps <= 2) {
avr->internal_sample_fmt = AV_SAMPLE_FMT_S16P;
} else if (avr->mixing_needed) {
avr->internal_sample_fmt = AV_SAMPLE_FMT_FLTP;
} else {
if (max_bps <= 4) {
if (in_fmt == AV_SAMPLE_FMT_S32P ||
out_fmt == AV_SAMPLE_FMT_S32P) {
if (in_fmt == AV_SAMPLE_FMT_FLTP ||
out_fmt == AV_SAMPLE_FMT_FLTP) {
/* if one is s32 and the other is flt, use dbl */
avr->internal_sample_fmt = AV_SAMPLE_FMT_DBLP;
} else {
/* if one is s32 and the other is s32, s16, or u8, use s32 */
avr->internal_sample_fmt = AV_SAMPLE_FMT_S32P;
}
} else {
/* if one is flt and the other is flt, s16 or u8, use flt */
avr->internal_sample_fmt = AV_SAMPLE_FMT_FLTP;
}
} else {
/* if either is dbl, use dbl */
avr->internal_sample_fmt = AV_SAMPLE_FMT_DBLP;
}
}
av_log(avr, AV_LOG_DEBUG, "Using %s as internal sample format\n",
av_get_sample_fmt_name(avr->internal_sample_fmt));
}
/* treat all mono as planar for easier comparison */
if (avr->in_channels == 1)
avr->in_sample_fmt = av_get_planar_sample_fmt(avr->in_sample_fmt);
if (avr->out_channels == 1)
avr->out_sample_fmt = av_get_planar_sample_fmt(avr->out_sample_fmt);
/* we may need to add an extra conversion in order to remap channels if
the output format is not planar */
if (avr->use_channel_map && !avr->mixing_needed && !avr->resample_needed &&
!av_sample_fmt_is_planar(avr->out_sample_fmt)) {
avr->internal_sample_fmt = av_get_planar_sample_fmt(avr->out_sample_fmt);
}
/* set sample format conversion parameters */
if (avr->resample_needed || avr->mixing_needed)
avr->in_convert_needed = avr->in_sample_fmt != avr->internal_sample_fmt;
else
avr->in_convert_needed = avr->use_channel_map &&
!av_sample_fmt_is_planar(avr->out_sample_fmt);
if (avr->resample_needed || avr->mixing_needed || avr->in_convert_needed)
avr->out_convert_needed = avr->internal_sample_fmt != avr->out_sample_fmt;
else
avr->out_convert_needed = avr->in_sample_fmt != avr->out_sample_fmt;
avr->in_copy_needed = !avr->in_convert_needed && (avr->mixing_needed ||
(avr->use_channel_map && avr->resample_needed));
if (avr->use_channel_map) {
if (avr->in_copy_needed) {
avr->remap_point = REMAP_IN_COPY;
av_dlog(avr, "remap channels during in_copy\n");
} else if (avr->in_convert_needed) {
avr->remap_point = REMAP_IN_CONVERT;
av_dlog(avr, "remap channels during in_convert\n");
} else if (avr->out_convert_needed) {
avr->remap_point = REMAP_OUT_CONVERT;
av_dlog(avr, "remap channels during out_convert\n");
} else {
avr->remap_point = REMAP_OUT_COPY;
av_dlog(avr, "remap channels during out_copy\n");
}
#ifdef DEBUG
{
int ch;
av_dlog(avr, "output map: ");
if (avr->ch_map_info.do_remap)
for (ch = 0; ch < avr->in_channels; ch++)
av_dlog(avr, " % 2d", avr->ch_map_info.channel_map[ch]);
else
av_dlog(avr, "n/a");
av_dlog(avr, "\n");
av_dlog(avr, "copy map: ");
if (avr->ch_map_info.do_copy)
for (ch = 0; ch < avr->in_channels; ch++)
av_dlog(avr, " % 2d", avr->ch_map_info.channel_copy[ch]);
else
av_dlog(avr, "n/a");
av_dlog(avr, "\n");
av_dlog(avr, "zero map: ");
if (avr->ch_map_info.do_zero)
for (ch = 0; ch < avr->in_channels; ch++)
av_dlog(avr, " % 2d", avr->ch_map_info.channel_zero[ch]);
else
av_dlog(avr, "n/a");
av_dlog(avr, "\n");
av_dlog(avr, "input map: ");
for (ch = 0; ch < avr->in_channels; ch++)
av_dlog(avr, " % 2d", avr->ch_map_info.input_map[ch]);
av_dlog(avr, "\n");
}
#endif
} else
avr->remap_point = REMAP_NONE;
/* allocate buffers */
if (avr->in_copy_needed || avr->in_convert_needed) {
avr->in_buffer = ff_audio_data_alloc(FFMAX(avr->in_channels, avr->out_channels),
0, avr->internal_sample_fmt,
"in_buffer");
if (!avr->in_buffer) {
ret = AVERROR(EINVAL);
goto error;
}
}
if (avr->resample_needed) {
avr->resample_out_buffer = ff_audio_data_alloc(avr->out_channels,
1024, avr->internal_sample_fmt,
"resample_out_buffer");
if (!avr->resample_out_buffer) {
ret = AVERROR(EINVAL);
goto error;
}
}
if (avr->out_convert_needed) {
avr->out_buffer = ff_audio_data_alloc(avr->out_channels, 0,
avr->out_sample_fmt, "out_buffer");
if (!avr->out_buffer) {
ret = AVERROR(EINVAL);
goto error;
}
}
avr->out_fifo = av_audio_fifo_alloc(avr->out_sample_fmt, avr->out_channels,
1024);
if (!avr->out_fifo) {
ret = AVERROR(ENOMEM);
goto error;
}
/* setup contexts */
if (avr->in_convert_needed) {
avr->ac_in = ff_audio_convert_alloc(avr, avr->internal_sample_fmt,
avr->in_sample_fmt, avr->in_channels,
avr->in_sample_rate,
avr->remap_point == REMAP_IN_CONVERT);
if (!avr->ac_in) {
ret = AVERROR(ENOMEM);
goto error;
}
}
if (avr->out_convert_needed) {
enum AVSampleFormat src_fmt;
if (avr->in_convert_needed)
src_fmt = avr->internal_sample_fmt;
else
src_fmt = avr->in_sample_fmt;
avr->ac_out = ff_audio_convert_alloc(avr, avr->out_sample_fmt, src_fmt,
avr->out_channels,
avr->out_sample_rate,
avr->remap_point == REMAP_OUT_CONVERT);
if (!avr->ac_out) {
ret = AVERROR(ENOMEM);
goto error;
}
}
if (avr->resample_needed) {
avr->resample = ff_audio_resample_init(avr);
if (!avr->resample) {
ret = AVERROR(ENOMEM);
goto error;
}
}
if (avr->mixing_needed) {
avr->am = ff_audio_mix_alloc(avr);
if (!avr->am) {
ret = AVERROR(ENOMEM);
goto error;
}
}
return 0;
error:
avresample_close(avr);
return ret;
}
int avresample_is_open(AVAudioResampleContext *avr)
{
return !!avr->out_fifo;
}
void avresample_close(AVAudioResampleContext *avr)
{
ff_audio_data_free(&avr->in_buffer);
ff_audio_data_free(&avr->resample_out_buffer);
ff_audio_data_free(&avr->out_buffer);
av_audio_fifo_free(avr->out_fifo);
avr->out_fifo = NULL;
ff_audio_convert_free(&avr->ac_in);
ff_audio_convert_free(&avr->ac_out);
ff_audio_resample_free(&avr->resample);
ff_audio_mix_free(&avr->am);
av_freep(&avr->mix_matrix);
avr->use_channel_map = 0;
}
void avresample_free(AVAudioResampleContext **avr)
{
if (!*avr)
return;
avresample_close(*avr);
av_opt_free(*avr);
av_freep(avr);
}
static int handle_buffered_output(AVAudioResampleContext *avr,
AudioData *output, AudioData *converted)
{
int ret;
if (!output || av_audio_fifo_size(avr->out_fifo) > 0 ||
(converted && output->allocated_samples < converted->nb_samples)) {
if (converted) {
/* if there are any samples in the output FIFO or if the
user-supplied output buffer is not large enough for all samples,
we add to the output FIFO */
av_dlog(avr, "[FIFO] add %s to out_fifo\n", converted->name);
ret = ff_audio_data_add_to_fifo(avr->out_fifo, converted, 0,
converted->nb_samples);
if (ret < 0)
return ret;
}
/* if the user specified an output buffer, read samples from the output
FIFO to the user output */
if (output && output->allocated_samples > 0) {
av_dlog(avr, "[FIFO] read from out_fifo to output\n");
av_dlog(avr, "[end conversion]\n");
return ff_audio_data_read_from_fifo(avr->out_fifo, output,
output->allocated_samples);
}
} else if (converted) {
/* copy directly to output if it is large enough or there is not any
data in the output FIFO */
av_dlog(avr, "[copy] %s to output\n", converted->name);
output->nb_samples = 0;
ret = ff_audio_data_copy(output, converted,
avr->remap_point == REMAP_OUT_COPY ?
&avr->ch_map_info : NULL);
if (ret < 0)
return ret;
av_dlog(avr, "[end conversion]\n");
return output->nb_samples;
}
av_dlog(avr, "[end conversion]\n");
return 0;
}
int attribute_align_arg avresample_convert(AVAudioResampleContext *avr,
uint8_t **output, int out_plane_size,
int out_samples, uint8_t **input,
int in_plane_size, int in_samples)
{
AudioData input_buffer;
AudioData output_buffer;
AudioData *current_buffer;
int ret, direct_output;
/* reset internal buffers */
if (avr->in_buffer) {
avr->in_buffer->nb_samples = 0;
ff_audio_data_set_channels(avr->in_buffer,
avr->in_buffer->allocated_channels);
}
if (avr->resample_out_buffer) {
avr->resample_out_buffer->nb_samples = 0;
ff_audio_data_set_channels(avr->resample_out_buffer,
avr->resample_out_buffer->allocated_channels);
}
if (avr->out_buffer) {
avr->out_buffer->nb_samples = 0;
ff_audio_data_set_channels(avr->out_buffer,
avr->out_buffer->allocated_channels);
}
av_dlog(avr, "[start conversion]\n");
/* initialize output_buffer with output data */
direct_output = output && av_audio_fifo_size(avr->out_fifo) == 0;
if (output) {
ret = ff_audio_data_init(&output_buffer, output, out_plane_size,
avr->out_channels, out_samples,
avr->out_sample_fmt, 0, "output");
if (ret < 0)
return ret;
output_buffer.nb_samples = 0;
}
if (input) {
/* initialize input_buffer with input data */
ret = ff_audio_data_init(&input_buffer, input, in_plane_size,
avr->in_channels, in_samples,
avr->in_sample_fmt, 1, "input");
if (ret < 0)
return ret;
current_buffer = &input_buffer;
if (avr->upmix_needed && !avr->in_convert_needed && !avr->resample_needed &&
!avr->out_convert_needed && direct_output && out_samples >= in_samples) {
/* in some rare cases we can copy input to output and upmix
directly in the output buffer */
av_dlog(avr, "[copy] %s to output\n", current_buffer->name);
ret = ff_audio_data_copy(&output_buffer, current_buffer,
avr->remap_point == REMAP_OUT_COPY ?
&avr->ch_map_info : NULL);
if (ret < 0)
return ret;
current_buffer = &output_buffer;
} else if (avr->remap_point == REMAP_OUT_COPY &&
(!direct_output || out_samples < in_samples)) {
/* if remapping channels during output copy, we may need to
* use an intermediate buffer in order to remap before adding
* samples to the output fifo */
av_dlog(avr, "[copy] %s to out_buffer\n", current_buffer->name);
ret = ff_audio_data_copy(avr->out_buffer, current_buffer,
&avr->ch_map_info);
if (ret < 0)
return ret;
current_buffer = avr->out_buffer;
} else if (avr->in_copy_needed || avr->in_convert_needed) {
/* if needed, copy or convert input to in_buffer, and downmix if
applicable */
if (avr->in_convert_needed) {
ret = ff_audio_data_realloc(avr->in_buffer,
current_buffer->nb_samples);
if (ret < 0)
return ret;
av_dlog(avr, "[convert] %s to in_buffer\n", current_buffer->name);
ret = ff_audio_convert(avr->ac_in, avr->in_buffer,
current_buffer);
if (ret < 0)
return ret;
} else {
av_dlog(avr, "[copy] %s to in_buffer\n", current_buffer->name);
ret = ff_audio_data_copy(avr->in_buffer, current_buffer,
avr->remap_point == REMAP_IN_COPY ?
&avr->ch_map_info : NULL);
if (ret < 0)
return ret;
}
ff_audio_data_set_channels(avr->in_buffer, avr->in_channels);
if (avr->downmix_needed) {
av_dlog(avr, "[downmix] in_buffer\n");
ret = ff_audio_mix(avr->am, avr->in_buffer);
if (ret < 0)
return ret;
}
current_buffer = avr->in_buffer;
}
} else {
/* flush resampling buffer and/or output FIFO if input is NULL */
if (!avr->resample_needed)
return handle_buffered_output(avr, output ? &output_buffer : NULL,
NULL);
current_buffer = NULL;
}
if (avr->resample_needed) {
AudioData *resample_out;
if (!avr->out_convert_needed && direct_output && out_samples > 0)
resample_out = &output_buffer;
else
resample_out = avr->resample_out_buffer;
av_dlog(avr, "[resample] %s to %s\n",
current_buffer ? current_buffer->name : "null",
resample_out->name);
ret = ff_audio_resample(avr->resample, resample_out,
current_buffer);
if (ret < 0)
return ret;
/* if resampling did not produce any samples, just return 0 */
if (resample_out->nb_samples == 0) {
av_dlog(avr, "[end conversion]\n");
return 0;
}
current_buffer = resample_out;
}
if (avr->upmix_needed) {
av_dlog(avr, "[upmix] %s\n", current_buffer->name);
ret = ff_audio_mix(avr->am, current_buffer);
if (ret < 0)
return ret;
}
/* if we resampled or upmixed directly to output, return here */
if (current_buffer == &output_buffer) {
av_dlog(avr, "[end conversion]\n");
return current_buffer->nb_samples;
}
if (avr->out_convert_needed) {
if (direct_output && out_samples >= current_buffer->nb_samples) {
/* convert directly to output */
av_dlog(avr, "[convert] %s to output\n", current_buffer->name);
ret = ff_audio_convert(avr->ac_out, &output_buffer, current_buffer);
if (ret < 0)
return ret;
av_dlog(avr, "[end conversion]\n");
return output_buffer.nb_samples;
} else {
ret = ff_audio_data_realloc(avr->out_buffer,
current_buffer->nb_samples);
if (ret < 0)
return ret;
av_dlog(avr, "[convert] %s to out_buffer\n", current_buffer->name);
ret = ff_audio_convert(avr->ac_out, avr->out_buffer,
current_buffer);
if (ret < 0)
return ret;
current_buffer = avr->out_buffer;
}
}
return handle_buffered_output(avr, output ? &output_buffer : NULL,
current_buffer);
}
int avresample_get_matrix(AVAudioResampleContext *avr, double *matrix,
int stride)
{
int in_channels, out_channels, i, o;
if (avr->am)
return ff_audio_mix_get_matrix(avr->am, matrix, stride);
in_channels = av_get_channel_layout_nb_channels(avr->in_channel_layout);
out_channels = av_get_channel_layout_nb_channels(avr->out_channel_layout);
if ( in_channels <= 0 || in_channels > AVRESAMPLE_MAX_CHANNELS ||
out_channels <= 0 || out_channels > AVRESAMPLE_MAX_CHANNELS) {
av_log(avr, AV_LOG_ERROR, "Invalid channel layouts\n");
return AVERROR(EINVAL);
}
if (!avr->mix_matrix) {
av_log(avr, AV_LOG_ERROR, "matrix is not set\n");
return AVERROR(EINVAL);
}
for (o = 0; o < out_channels; o++)
for (i = 0; i < in_channels; i++)
matrix[o * stride + i] = avr->mix_matrix[o * in_channels + i];
return 0;
}
int avresample_set_matrix(AVAudioResampleContext *avr, const double *matrix,
int stride)
{
int in_channels, out_channels, i, o;
if (avr->am)
return ff_audio_mix_set_matrix(avr->am, matrix, stride);
in_channels = av_get_channel_layout_nb_channels(avr->in_channel_layout);
out_channels = av_get_channel_layout_nb_channels(avr->out_channel_layout);
if ( in_channels <= 0 || in_channels > AVRESAMPLE_MAX_CHANNELS ||
out_channels <= 0 || out_channels > AVRESAMPLE_MAX_CHANNELS) {
av_log(avr, AV_LOG_ERROR, "Invalid channel layouts\n");
return AVERROR(EINVAL);
}
if (avr->mix_matrix)
av_freep(&avr->mix_matrix);
avr->mix_matrix = av_malloc(in_channels * out_channels *
sizeof(*avr->mix_matrix));
if (!avr->mix_matrix)
return AVERROR(ENOMEM);
for (o = 0; o < out_channels; o++)
for (i = 0; i < in_channels; i++)
avr->mix_matrix[o * in_channels + i] = matrix[o * stride + i];
return 0;
}
int avresample_set_channel_mapping(AVAudioResampleContext *avr,
const int *channel_map)
{
ChannelMapInfo *info = &avr->ch_map_info;
int in_channels, ch, i;
in_channels = av_get_channel_layout_nb_channels(avr->in_channel_layout);
if (in_channels <= 0 || in_channels > AVRESAMPLE_MAX_CHANNELS) {
av_log(avr, AV_LOG_ERROR, "Invalid input channel layout\n");
return AVERROR(EINVAL);
}
memset(info, 0, sizeof(*info));
memset(info->input_map, -1, sizeof(info->input_map));
for (ch = 0; ch < in_channels; ch++) {
if (channel_map[ch] >= in_channels) {
av_log(avr, AV_LOG_ERROR, "Invalid channel map\n");
return AVERROR(EINVAL);
}
if (channel_map[ch] < 0) {
info->channel_zero[ch] = 1;
info->channel_map[ch] = -1;
info->do_zero = 1;
} else if (info->input_map[channel_map[ch]] >= 0) {
info->channel_copy[ch] = info->input_map[channel_map[ch]];
info->channel_map[ch] = -1;
info->do_copy = 1;
} else {
info->channel_map[ch] = channel_map[ch];
info->input_map[channel_map[ch]] = ch;
info->do_remap = 1;
}
}
/* Fill-in unmapped input channels with unmapped output channels.
This is used when remapping during conversion from interleaved to
planar format. */
for (ch = 0, i = 0; ch < in_channels && i < in_channels; ch++, i++) {
while (ch < in_channels && info->input_map[ch] >= 0)
ch++;
while (i < in_channels && info->channel_map[i] >= 0)
i++;
if (ch >= in_channels || i >= in_channels)
break;
info->input_map[ch] = i;
}
avr->use_channel_map = 1;
return 0;
}
int avresample_available(AVAudioResampleContext *avr)
{
return av_audio_fifo_size(avr->out_fifo);
}
int avresample_get_out_samples(AVAudioResampleContext *avr, int in_nb_samples)
{
int64_t samples = avresample_get_delay(avr) + (int64_t)in_nb_samples;
if (avr->resample_needed) {
samples = av_rescale_rnd(samples,
avr->out_sample_rate,
avr->in_sample_rate,
AV_ROUND_UP);
}
samples += avresample_available(avr);
if (samples > INT_MAX)
return AVERROR(EINVAL);
return samples;
}
int avresample_read(AVAudioResampleContext *avr, uint8_t **output, int nb_samples)
{
if (!output)
return av_audio_fifo_drain(avr->out_fifo, nb_samples);
return av_audio_fifo_read(avr->out_fifo, (void**)output, nb_samples);
}
unsigned avresample_version(void)
{
return LIBAVRESAMPLE_VERSION_INT;
}
const char *avresample_license(void)
{
#define LICENSE_PREFIX "libavresample license: "
return LICENSE_PREFIX LIBAV_LICENSE + sizeof(LICENSE_PREFIX) - 1;
}
const char *avresample_configuration(void)
{
return LIBAV_CONFIGURATION;
}