ffmpeg/libavresample/resample.c
Peter Meerwald 12655c4804 libavresample: NEON optimized FIR audio resampling
modelled after aarch64 code

on Cortex-A8, s16 and s32 code is about 2x faster,
float code about 7x faster

Signed-off-by: Peter Meerwald <pmeerw@pmeerw.net>
Signed-off-by: Martin Storsjö <martin@martin.st>
2015-02-21 22:08:39 +02:00

504 lines
17 KiB
C

/*
* Copyright (c) 2004 Michael Niedermayer <michaelni@gmx.at>
* 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/libm.h"
#include "libavutil/log.h"
#include "internal.h"
#include "resample.h"
#include "audio_data.h"
/* double template */
#define CONFIG_RESAMPLE_DBL
#include "resample_template.c"
#undef CONFIG_RESAMPLE_DBL
/* float template */
#define CONFIG_RESAMPLE_FLT
#include "resample_template.c"
#undef CONFIG_RESAMPLE_FLT
/* s32 template */
#define CONFIG_RESAMPLE_S32
#include "resample_template.c"
#undef CONFIG_RESAMPLE_S32
/* s16 template */
#include "resample_template.c"
/* 0th order modified bessel function of the first kind. */
static double bessel(double x)
{
double v = 1;
double lastv = 0;
double t = 1;
int i;
x = x * x / 4;
for (i = 1; v != lastv; i++) {
lastv = v;
t *= x / (i * i);
v += t;
}
return v;
}
/* Build a polyphase filterbank. */
static int build_filter(ResampleContext *c, double factor)
{
int ph, i;
double x, y, w;
double *tab;
int tap_count = c->filter_length;
int phase_count = 1 << c->phase_shift;
const int center = (tap_count - 1) / 2;
tab = av_malloc(tap_count * sizeof(*tab));
if (!tab)
return AVERROR(ENOMEM);
for (ph = 0; ph < phase_count; ph++) {
double norm = 0;
for (i = 0; i < tap_count; i++) {
x = M_PI * ((double)(i - center) - (double)ph / phase_count) * factor;
if (x == 0) y = 1.0;
else y = sin(x) / x;
switch (c->filter_type) {
case AV_RESAMPLE_FILTER_TYPE_CUBIC: {
const float d = -0.5; //first order derivative = -0.5
x = fabs(((double)(i - center) - (double)ph / phase_count) * factor);
if (x < 1.0) y = 1 - 3 * x*x + 2 * x*x*x + d * ( -x*x + x*x*x);
else y = d * (-4 + 8 * x - 5 * x*x + x*x*x);
break;
}
case AV_RESAMPLE_FILTER_TYPE_BLACKMAN_NUTTALL:
w = 2.0 * x / (factor * tap_count) + M_PI;
y *= 0.3635819 - 0.4891775 * cos( w) +
0.1365995 * cos(2 * w) -
0.0106411 * cos(3 * w);
break;
case AV_RESAMPLE_FILTER_TYPE_KAISER:
w = 2.0 * x / (factor * tap_count * M_PI);
y *= bessel(c->kaiser_beta * sqrt(FFMAX(1 - w * w, 0)));
break;
}
tab[i] = y;
norm += y;
}
/* normalize so that an uniform color remains the same */
for (i = 0; i < tap_count; i++)
tab[i] = tab[i] / norm;
c->set_filter(c->filter_bank, tab, ph, tap_count);
}
av_free(tab);
return 0;
}
ResampleContext *ff_audio_resample_init(AVAudioResampleContext *avr)
{
ResampleContext *c;
int out_rate = avr->out_sample_rate;
int in_rate = avr->in_sample_rate;
double factor = FFMIN(out_rate * avr->cutoff / in_rate, 1.0);
int phase_count = 1 << avr->phase_shift;
int felem_size;
if (avr->internal_sample_fmt != AV_SAMPLE_FMT_S16P &&
avr->internal_sample_fmt != AV_SAMPLE_FMT_S32P &&
avr->internal_sample_fmt != AV_SAMPLE_FMT_FLTP &&
avr->internal_sample_fmt != AV_SAMPLE_FMT_DBLP) {
av_log(avr, AV_LOG_ERROR, "Unsupported internal format for "
"resampling: %s\n",
av_get_sample_fmt_name(avr->internal_sample_fmt));
return NULL;
}
c = av_mallocz(sizeof(*c));
if (!c)
return NULL;
c->avr = avr;
c->phase_shift = avr->phase_shift;
c->phase_mask = phase_count - 1;
c->linear = avr->linear_interp;
c->filter_length = FFMAX((int)ceil(avr->filter_size / factor), 1);
c->filter_type = avr->filter_type;
c->kaiser_beta = avr->kaiser_beta;
switch (avr->internal_sample_fmt) {
case AV_SAMPLE_FMT_DBLP:
c->resample_one = c->linear ? resample_linear_dbl : resample_one_dbl;
c->resample_nearest = resample_nearest_dbl;
c->set_filter = set_filter_dbl;
break;
case AV_SAMPLE_FMT_FLTP:
c->resample_one = c->linear ? resample_linear_flt : resample_one_flt;
c->resample_nearest = resample_nearest_flt;
c->set_filter = set_filter_flt;
break;
case AV_SAMPLE_FMT_S32P:
c->resample_one = c->linear ? resample_linear_s32 : resample_one_s32;
c->resample_nearest = resample_nearest_s32;
c->set_filter = set_filter_s32;
break;
case AV_SAMPLE_FMT_S16P:
c->resample_one = c->linear ? resample_linear_s16 : resample_one_s16;
c->resample_nearest = resample_nearest_s16;
c->set_filter = set_filter_s16;
break;
}
if (ARCH_AARCH64)
ff_audio_resample_init_aarch64(c, avr->internal_sample_fmt);
if (ARCH_ARM)
ff_audio_resample_init_arm(c, avr->internal_sample_fmt);
felem_size = av_get_bytes_per_sample(avr->internal_sample_fmt);
c->filter_bank = av_mallocz(c->filter_length * (phase_count + 1) * felem_size);
if (!c->filter_bank)
goto error;
if (build_filter(c, factor) < 0)
goto error;
memcpy(&c->filter_bank[(c->filter_length * phase_count + 1) * felem_size],
c->filter_bank, (c->filter_length - 1) * felem_size);
memcpy(&c->filter_bank[c->filter_length * phase_count * felem_size],
&c->filter_bank[(c->filter_length - 1) * felem_size], felem_size);
c->compensation_distance = 0;
if (!av_reduce(&c->src_incr, &c->dst_incr, out_rate,
in_rate * (int64_t)phase_count, INT32_MAX / 2))
goto error;
c->ideal_dst_incr = c->dst_incr;
c->padding_size = (c->filter_length - 1) / 2;
c->initial_padding_filled = 0;
c->index = 0;
c->frac = 0;
/* allocate internal buffer */
c->buffer = ff_audio_data_alloc(avr->resample_channels, c->padding_size,
avr->internal_sample_fmt,
"resample buffer");
if (!c->buffer)
goto error;
c->buffer->nb_samples = c->padding_size;
c->initial_padding_samples = c->padding_size;
av_log(avr, AV_LOG_DEBUG, "resample: %s from %d Hz to %d Hz\n",
av_get_sample_fmt_name(avr->internal_sample_fmt),
avr->in_sample_rate, avr->out_sample_rate);
return c;
error:
ff_audio_data_free(&c->buffer);
av_free(c->filter_bank);
av_free(c);
return NULL;
}
void ff_audio_resample_free(ResampleContext **c)
{
if (!*c)
return;
ff_audio_data_free(&(*c)->buffer);
av_free((*c)->filter_bank);
av_freep(c);
}
int avresample_set_compensation(AVAudioResampleContext *avr, int sample_delta,
int compensation_distance)
{
ResampleContext *c;
AudioData *fifo_buf = NULL;
int ret = 0;
if (compensation_distance < 0)
return AVERROR(EINVAL);
if (!compensation_distance && sample_delta)
return AVERROR(EINVAL);
if (!avr->resample_needed) {
#if FF_API_RESAMPLE_CLOSE_OPEN
/* if resampling was not enabled previously, re-initialize the
AVAudioResampleContext and force resampling */
int fifo_samples;
int restore_matrix = 0;
double matrix[AVRESAMPLE_MAX_CHANNELS * AVRESAMPLE_MAX_CHANNELS] = { 0 };
/* buffer any remaining samples in the output FIFO before closing */
fifo_samples = av_audio_fifo_size(avr->out_fifo);
if (fifo_samples > 0) {
fifo_buf = ff_audio_data_alloc(avr->out_channels, fifo_samples,
avr->out_sample_fmt, NULL);
if (!fifo_buf)
return AVERROR(EINVAL);
ret = ff_audio_data_read_from_fifo(avr->out_fifo, fifo_buf,
fifo_samples);
if (ret < 0)
goto reinit_fail;
}
/* save the channel mixing matrix */
if (avr->am) {
ret = avresample_get_matrix(avr, matrix, AVRESAMPLE_MAX_CHANNELS);
if (ret < 0)
goto reinit_fail;
restore_matrix = 1;
}
/* close the AVAudioResampleContext */
avresample_close(avr);
avr->force_resampling = 1;
/* restore the channel mixing matrix */
if (restore_matrix) {
ret = avresample_set_matrix(avr, matrix, AVRESAMPLE_MAX_CHANNELS);
if (ret < 0)
goto reinit_fail;
}
/* re-open the AVAudioResampleContext */
ret = avresample_open(avr);
if (ret < 0)
goto reinit_fail;
/* restore buffered samples to the output FIFO */
if (fifo_samples > 0) {
ret = ff_audio_data_add_to_fifo(avr->out_fifo, fifo_buf, 0,
fifo_samples);
if (ret < 0)
goto reinit_fail;
ff_audio_data_free(&fifo_buf);
}
#else
av_log(avr, AV_LOG_ERROR, "Unable to set resampling compensation\n");
return AVERROR(EINVAL);
#endif
}
c = avr->resample;
c->compensation_distance = compensation_distance;
if (compensation_distance) {
c->dst_incr = c->ideal_dst_incr - c->ideal_dst_incr *
(int64_t)sample_delta / compensation_distance;
} else {
c->dst_incr = c->ideal_dst_incr;
}
return 0;
reinit_fail:
ff_audio_data_free(&fifo_buf);
return ret;
}
static int resample(ResampleContext *c, void *dst, const void *src,
int *consumed, int src_size, int dst_size, int update_ctx,
int nearest_neighbour)
{
int dst_index;
unsigned int index = c->index;
int frac = c->frac;
int dst_incr_frac = c->dst_incr % c->src_incr;
int dst_incr = c->dst_incr / c->src_incr;
int compensation_distance = c->compensation_distance;
if (!dst != !src)
return AVERROR(EINVAL);
if (nearest_neighbour) {
uint64_t index2 = ((uint64_t)index) << 32;
int64_t incr = (1LL << 32) * c->dst_incr / c->src_incr;
dst_size = FFMIN(dst_size,
(src_size-1-index) * (int64_t)c->src_incr /
c->dst_incr);
if (dst) {
for(dst_index = 0; dst_index < dst_size; dst_index++) {
c->resample_nearest(dst, dst_index, src, index2 >> 32);
index2 += incr;
}
} else {
dst_index = dst_size;
}
index += dst_index * dst_incr;
index += (frac + dst_index * (int64_t)dst_incr_frac) / c->src_incr;
frac = (frac + dst_index * (int64_t)dst_incr_frac) % c->src_incr;
} else {
for (dst_index = 0; dst_index < dst_size; dst_index++) {
int sample_index = index >> c->phase_shift;
if (sample_index + c->filter_length > src_size)
break;
if (dst)
c->resample_one(c, dst, dst_index, src, index, frac);
frac += dst_incr_frac;
index += dst_incr;
if (frac >= c->src_incr) {
frac -= c->src_incr;
index++;
}
if (dst_index + 1 == compensation_distance) {
compensation_distance = 0;
dst_incr_frac = c->ideal_dst_incr % c->src_incr;
dst_incr = c->ideal_dst_incr / c->src_incr;
}
}
}
if (consumed)
*consumed = index >> c->phase_shift;
if (update_ctx) {
index &= c->phase_mask;
if (compensation_distance) {
compensation_distance -= dst_index;
if (compensation_distance <= 0)
return AVERROR_BUG;
}
c->frac = frac;
c->index = index;
c->dst_incr = dst_incr_frac + c->src_incr*dst_incr;
c->compensation_distance = compensation_distance;
}
return dst_index;
}
int ff_audio_resample(ResampleContext *c, AudioData *dst, AudioData *src)
{
int ch, in_samples, in_leftover, consumed = 0, out_samples = 0;
int ret = AVERROR(EINVAL);
int nearest_neighbour = (c->compensation_distance == 0 &&
c->filter_length == 1 &&
c->phase_shift == 0);
in_samples = src ? src->nb_samples : 0;
in_leftover = c->buffer->nb_samples;
/* add input samples to the internal buffer */
if (src) {
ret = ff_audio_data_combine(c->buffer, in_leftover, src, 0, in_samples);
if (ret < 0)
return ret;
} else if (in_leftover <= c->final_padding_samples) {
/* no remaining samples to flush */
return 0;
}
if (!c->initial_padding_filled) {
int bps = av_get_bytes_per_sample(c->avr->internal_sample_fmt);
int i;
if (src && c->buffer->nb_samples < 2 * c->padding_size)
return 0;
for (i = 0; i < c->padding_size; i++)
for (ch = 0; ch < c->buffer->channels; ch++) {
if (c->buffer->nb_samples > 2 * c->padding_size - i) {
memcpy(c->buffer->data[ch] + bps * i,
c->buffer->data[ch] + bps * (2 * c->padding_size - i), bps);
} else {
memset(c->buffer->data[ch] + bps * i, 0, bps);
}
}
c->initial_padding_filled = 1;
}
if (!src && !c->final_padding_filled) {
int bps = av_get_bytes_per_sample(c->avr->internal_sample_fmt);
int i;
ret = ff_audio_data_realloc(c->buffer, in_samples + c->padding_size);
if (ret < 0) {
av_log(c->avr, AV_LOG_ERROR, "Error reallocating resampling buffer\n");
return AVERROR(ENOMEM);
}
for (i = 0; i < c->padding_size; i++)
for (ch = 0; ch < c->buffer->channels; ch++) {
if (in_leftover > i) {
memcpy(c->buffer->data[ch] + bps * (in_leftover + i),
c->buffer->data[ch] + bps * (in_leftover - i - 1),
bps);
} else {
memset(c->buffer->data[ch] + bps * (in_leftover + i),
0, bps);
}
}
c->buffer->nb_samples += c->padding_size;
c->final_padding_samples = c->padding_size;
c->final_padding_filled = 1;
}
/* calculate output size and reallocate output buffer if needed */
/* TODO: try to calculate this without the dummy resample() run */
if (!dst->read_only && dst->allow_realloc) {
out_samples = resample(c, NULL, NULL, NULL, c->buffer->nb_samples,
INT_MAX, 0, nearest_neighbour);
ret = ff_audio_data_realloc(dst, out_samples);
if (ret < 0) {
av_log(c->avr, AV_LOG_ERROR, "error reallocating output\n");
return ret;
}
}
/* resample each channel plane */
for (ch = 0; ch < c->buffer->channels; ch++) {
out_samples = resample(c, (void *)dst->data[ch],
(const void *)c->buffer->data[ch], &consumed,
c->buffer->nb_samples, dst->allocated_samples,
ch + 1 == c->buffer->channels, nearest_neighbour);
}
if (out_samples < 0) {
av_log(c->avr, AV_LOG_ERROR, "error during resampling\n");
return out_samples;
}
/* drain consumed samples from the internal buffer */
ff_audio_data_drain(c->buffer, consumed);
c->initial_padding_samples = FFMAX(c->initial_padding_samples - consumed, 0);
av_dlog(c->avr, "resampled %d in + %d leftover to %d out + %d leftover\n",
in_samples, in_leftover, out_samples, c->buffer->nb_samples);
dst->nb_samples = out_samples;
return 0;
}
int avresample_get_delay(AVAudioResampleContext *avr)
{
ResampleContext *c = avr->resample;
if (!avr->resample_needed || !avr->resample)
return 0;
return FFMAX(c->buffer->nb_samples - c->padding_size, 0);
}