avfilter/af_headphone: switch to TX FFT from avutil
This commit is contained in:
parent
f6aeb94d72
commit
88b6342c2b
2
configure
vendored
2
configure
vendored
@ -3608,8 +3608,6 @@ freezedetect_filter_select="scene_sad"
|
||||
frei0r_filter_deps="frei0r libdl"
|
||||
frei0r_src_filter_deps="frei0r libdl"
|
||||
fspp_filter_deps="gpl"
|
||||
headphone_filter_deps="avcodec"
|
||||
headphone_filter_select="fft"
|
||||
histeq_filter_deps="gpl"
|
||||
hqdn3d_filter_deps="gpl"
|
||||
interlace_filter_deps="gpl"
|
||||
|
@ -25,7 +25,7 @@
|
||||
#include "libavutil/float_dsp.h"
|
||||
#include "libavutil/intmath.h"
|
||||
#include "libavutil/opt.h"
|
||||
#include "libavcodec/avfft.h"
|
||||
#include "libavutil/tx.h"
|
||||
|
||||
#include "avfilter.h"
|
||||
#include "filters.h"
|
||||
@ -69,11 +69,13 @@ typedef struct HeadphoneContext {
|
||||
|
||||
float *data_ir[2];
|
||||
float *temp_src[2];
|
||||
FFTComplex *temp_fft[2];
|
||||
FFTComplex *temp_afft[2];
|
||||
AVComplexFloat *out_fft[2];
|
||||
AVComplexFloat *in_fft[2];
|
||||
AVComplexFloat *temp_afft[2];
|
||||
|
||||
FFTContext *fft[2], *ifft[2];
|
||||
FFTComplex *data_hrtf[2];
|
||||
AVTXContext *fft[2], *ifft[2];
|
||||
av_tx_fn tx_fn[2], itx_fn[2];
|
||||
AVComplexFloat *data_hrtf[2];
|
||||
|
||||
float (*scalarproduct_float)(const float *v1, const float *v2, int len);
|
||||
struct hrir_inputs {
|
||||
@ -130,8 +132,9 @@ typedef struct ThreadData {
|
||||
int *n_clippings;
|
||||
float **ringbuffer;
|
||||
float **temp_src;
|
||||
FFTComplex **temp_fft;
|
||||
FFTComplex **temp_afft;
|
||||
AVComplexFloat **out_fft;
|
||||
AVComplexFloat **in_fft;
|
||||
AVComplexFloat **temp_afft;
|
||||
} ThreadData;
|
||||
|
||||
static int headphone_convolute(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
|
||||
@ -212,7 +215,7 @@ static int headphone_fast_convolute(AVFilterContext *ctx, void *arg, int jobnr,
|
||||
AVFrame *in = td->in, *out = td->out;
|
||||
int offset = jobnr;
|
||||
int *write = &td->write[jobnr];
|
||||
FFTComplex *hrtf = s->data_hrtf[jobnr];
|
||||
AVComplexFloat *hrtf = s->data_hrtf[jobnr];
|
||||
int *n_clippings = &td->n_clippings[jobnr];
|
||||
float *ringbuffer = td->ringbuffer[jobnr];
|
||||
const int ir_len = s->ir_len;
|
||||
@ -221,13 +224,16 @@ static int headphone_fast_convolute(AVFilterContext *ctx, void *arg, int jobnr,
|
||||
const int in_channels = in->channels;
|
||||
const int buffer_length = s->buffer_length;
|
||||
const uint32_t modulo = (uint32_t)buffer_length - 1;
|
||||
FFTComplex *fft_in = s->temp_fft[jobnr];
|
||||
FFTComplex *fft_acc = s->temp_afft[jobnr];
|
||||
FFTContext *ifft = s->ifft[jobnr];
|
||||
FFTContext *fft = s->fft[jobnr];
|
||||
AVComplexFloat *fft_out = s->out_fft[jobnr];
|
||||
AVComplexFloat *fft_in = s->in_fft[jobnr];
|
||||
AVComplexFloat *fft_acc = s->temp_afft[jobnr];
|
||||
AVTXContext *ifft = s->ifft[jobnr];
|
||||
AVTXContext *fft = s->fft[jobnr];
|
||||
av_tx_fn tx_fn = s->tx_fn[jobnr];
|
||||
av_tx_fn itx_fn = s->itx_fn[jobnr];
|
||||
const int n_fft = s->n_fft;
|
||||
const float fft_scale = 1.0f / s->n_fft;
|
||||
FFTComplex *hrtf_offset;
|
||||
AVComplexFloat *hrtf_offset;
|
||||
int wr = *write;
|
||||
int n_read;
|
||||
int i, j;
|
||||
@ -245,7 +251,7 @@ static int headphone_fast_convolute(AVFilterContext *ctx, void *arg, int jobnr,
|
||||
dst[2 * j] = 0;
|
||||
}
|
||||
|
||||
memset(fft_acc, 0, sizeof(FFTComplex) * n_fft);
|
||||
memset(fft_acc, 0, sizeof(AVComplexFloat) * n_fft);
|
||||
|
||||
for (i = 0; i < in_channels; i++) {
|
||||
if (i == s->lfe_channel) {
|
||||
@ -258,29 +264,28 @@ static int headphone_fast_convolute(AVFilterContext *ctx, void *arg, int jobnr,
|
||||
offset = i * n_fft;
|
||||
hrtf_offset = hrtf + offset;
|
||||
|
||||
memset(fft_in, 0, sizeof(FFTComplex) * n_fft);
|
||||
memset(fft_in, 0, sizeof(AVComplexFloat) * n_fft);
|
||||
|
||||
for (j = 0; j < in->nb_samples; j++) {
|
||||
fft_in[j].re = src[j * in_channels + i];
|
||||
}
|
||||
|
||||
av_fft_permute(fft, fft_in);
|
||||
av_fft_calc(fft, fft_in);
|
||||
tx_fn(fft, fft_out, fft_in, sizeof(float));
|
||||
|
||||
for (j = 0; j < n_fft; j++) {
|
||||
const FFTComplex *hcomplex = hrtf_offset + j;
|
||||
const float re = fft_in[j].re;
|
||||
const float im = fft_in[j].im;
|
||||
const AVComplexFloat *hcomplex = hrtf_offset + j;
|
||||
const float re = fft_out[j].re;
|
||||
const float im = fft_out[j].im;
|
||||
|
||||
fft_acc[j].re += re * hcomplex->re - im * hcomplex->im;
|
||||
fft_acc[j].im += re * hcomplex->im + im * hcomplex->re;
|
||||
}
|
||||
}
|
||||
|
||||
av_fft_permute(ifft, fft_acc);
|
||||
av_fft_calc(ifft, fft_acc);
|
||||
itx_fn(ifft, fft_out, fft_acc, sizeof(float));
|
||||
|
||||
for (j = 0; j < in->nb_samples; j++) {
|
||||
dst[2 * j] += fft_acc[j].re * fft_scale;
|
||||
dst[2 * j] += fft_out[j].re * fft_scale;
|
||||
if (fabsf(dst[2 * j]) > 1)
|
||||
n_clippings[0]++;
|
||||
}
|
||||
@ -288,7 +293,7 @@ static int headphone_fast_convolute(AVFilterContext *ctx, void *arg, int jobnr,
|
||||
for (j = 0; j < ir_len - 1; j++) {
|
||||
int write_pos = (wr + j) & modulo;
|
||||
|
||||
*(ringbuffer + write_pos) += fft_acc[in->nb_samples + j].re * fft_scale;
|
||||
*(ringbuffer + write_pos) += fft_out[in->nb_samples + j].re * fft_scale;
|
||||
}
|
||||
|
||||
*write = wr;
|
||||
@ -331,7 +336,8 @@ static int headphone_frame(HeadphoneContext *s, AVFrame *in, AVFilterLink *outli
|
||||
td.in = in; td.out = out; td.write = s->write;
|
||||
td.ir = s->data_ir; td.n_clippings = n_clippings;
|
||||
td.ringbuffer = s->ringbuffer; td.temp_src = s->temp_src;
|
||||
td.temp_fft = s->temp_fft;
|
||||
td.out_fft = s->out_fft;
|
||||
td.in_fft = s->in_fft;
|
||||
td.temp_afft = s->temp_afft;
|
||||
|
||||
if (s->type == TIME_DOMAIN) {
|
||||
@ -369,10 +375,20 @@ static int convert_coeffs(AVFilterContext *ctx, AVFilterLink *inlink)
|
||||
s->n_fft = n_fft = 1 << (32 - ff_clz(ir_len + s->size));
|
||||
|
||||
if (s->type == FREQUENCY_DOMAIN) {
|
||||
s->fft[0] = av_fft_init(av_log2(s->n_fft), 0);
|
||||
s->fft[1] = av_fft_init(av_log2(s->n_fft), 0);
|
||||
s->ifft[0] = av_fft_init(av_log2(s->n_fft), 1);
|
||||
s->ifft[1] = av_fft_init(av_log2(s->n_fft), 1);
|
||||
float scale;
|
||||
|
||||
ret = av_tx_init(&s->fft[0], &s->tx_fn[0], AV_TX_FLOAT_FFT, 0, s->n_fft, &scale, 0);
|
||||
if (ret < 0)
|
||||
goto fail;
|
||||
ret = av_tx_init(&s->fft[1], &s->tx_fn[1], AV_TX_FLOAT_FFT, 0, s->n_fft, &scale, 0);
|
||||
if (ret < 0)
|
||||
goto fail;
|
||||
ret = av_tx_init(&s->ifft[0], &s->itx_fn[0], AV_TX_FLOAT_FFT, 1, s->n_fft, &scale, 0);
|
||||
if (ret < 0)
|
||||
goto fail;
|
||||
ret = av_tx_init(&s->ifft[1], &s->itx_fn[1], AV_TX_FLOAT_FFT, 1, s->n_fft, &scale, 0);
|
||||
if (ret < 0)
|
||||
goto fail;
|
||||
|
||||
if (!s->fft[0] || !s->fft[1] || !s->ifft[0] || !s->ifft[1]) {
|
||||
av_log(ctx, AV_LOG_ERROR, "Unable to create FFT contexts of size %d.\n", s->n_fft);
|
||||
@ -387,11 +403,14 @@ static int convert_coeffs(AVFilterContext *ctx, AVFilterLink *inlink)
|
||||
} else {
|
||||
s->ringbuffer[0] = av_calloc(s->buffer_length, sizeof(float));
|
||||
s->ringbuffer[1] = av_calloc(s->buffer_length, sizeof(float));
|
||||
s->temp_fft[0] = av_calloc(s->n_fft, sizeof(FFTComplex));
|
||||
s->temp_fft[1] = av_calloc(s->n_fft, sizeof(FFTComplex));
|
||||
s->temp_afft[0] = av_calloc(s->n_fft, sizeof(FFTComplex));
|
||||
s->temp_afft[1] = av_calloc(s->n_fft, sizeof(FFTComplex));
|
||||
if (!s->temp_fft[0] || !s->temp_fft[1] ||
|
||||
s->out_fft[0] = av_calloc(s->n_fft, sizeof(AVComplexFloat));
|
||||
s->out_fft[1] = av_calloc(s->n_fft, sizeof(AVComplexFloat));
|
||||
s->in_fft[0] = av_calloc(s->n_fft, sizeof(AVComplexFloat));
|
||||
s->in_fft[1] = av_calloc(s->n_fft, sizeof(AVComplexFloat));
|
||||
s->temp_afft[0] = av_calloc(s->n_fft, sizeof(AVComplexFloat));
|
||||
s->temp_afft[1] = av_calloc(s->n_fft, sizeof(AVComplexFloat));
|
||||
if (!s->in_fft[0] || !s->in_fft[1] ||
|
||||
!s->out_fft[0] || !s->out_fft[1] ||
|
||||
!s->temp_afft[0] || !s->temp_afft[1]) {
|
||||
ret = AVERROR(ENOMEM);
|
||||
goto fail;
|
||||
@ -445,18 +464,18 @@ static int convert_coeffs(AVFilterContext *ctx, AVFilterLink *inlink)
|
||||
data_ir_r[j] = ptr[len * 2 - j * 2 - 1] * gain_lin;
|
||||
}
|
||||
} else {
|
||||
FFTComplex *fft_in_l = s->data_hrtf[0] + idx * n_fft;
|
||||
FFTComplex *fft_in_r = s->data_hrtf[1] + idx * n_fft;
|
||||
AVComplexFloat *fft_out_l = s->data_hrtf[0] + idx * n_fft;
|
||||
AVComplexFloat *fft_out_r = s->data_hrtf[1] + idx * n_fft;
|
||||
AVComplexFloat *fft_in_l = s->in_fft[0];
|
||||
AVComplexFloat *fft_in_r = s->in_fft[1];
|
||||
|
||||
for (j = 0; j < len; j++) {
|
||||
fft_in_l[j].re = ptr[j * 2 ] * gain_lin;
|
||||
fft_in_r[j].re = ptr[j * 2 + 1] * gain_lin;
|
||||
}
|
||||
|
||||
av_fft_permute(s->fft[0], fft_in_l);
|
||||
av_fft_calc(s->fft[0], fft_in_l);
|
||||
av_fft_permute(s->fft[0], fft_in_r);
|
||||
av_fft_calc(s->fft[0], fft_in_r);
|
||||
s->tx_fn[0](s->fft[0], fft_out_l, fft_in_l, sizeof(float));
|
||||
s->tx_fn[0](s->fft[0], fft_out_r, fft_in_r, sizeof(float));
|
||||
}
|
||||
} else {
|
||||
int I, N = ctx->inputs[1]->channels;
|
||||
@ -477,18 +496,18 @@ static int convert_coeffs(AVFilterContext *ctx, AVFilterLink *inlink)
|
||||
data_ir_r[j] = ptr[len * N - j * N - N + I + 1] * gain_lin;
|
||||
}
|
||||
} else {
|
||||
FFTComplex *fft_in_l = s->data_hrtf[0] + idx * n_fft;
|
||||
FFTComplex *fft_in_r = s->data_hrtf[1] + idx * n_fft;
|
||||
AVComplexFloat *fft_out_l = s->data_hrtf[0] + idx * n_fft;
|
||||
AVComplexFloat *fft_out_r = s->data_hrtf[1] + idx * n_fft;
|
||||
AVComplexFloat *fft_in_l = s->in_fft[0];
|
||||
AVComplexFloat *fft_in_r = s->in_fft[1];
|
||||
|
||||
for (j = 0; j < len; j++) {
|
||||
fft_in_l[j].re = ptr[j * N + I ] * gain_lin;
|
||||
fft_in_r[j].re = ptr[j * N + I + 1] * gain_lin;
|
||||
}
|
||||
|
||||
av_fft_permute(s->fft[0], fft_in_l);
|
||||
av_fft_calc(s->fft[0], fft_in_l);
|
||||
av_fft_permute(s->fft[0], fft_in_r);
|
||||
av_fft_calc(s->fft[0], fft_in_r);
|
||||
s->tx_fn[0](s->fft[0], fft_out_l, fft_in_l, sizeof(float));
|
||||
s->tx_fn[0](s->fft[0], fft_out_r, fft_in_r, sizeof(float));
|
||||
}
|
||||
}
|
||||
}
|
||||
@ -695,18 +714,20 @@ static av_cold void uninit(AVFilterContext *ctx)
|
||||
{
|
||||
HeadphoneContext *s = ctx->priv;
|
||||
|
||||
av_fft_end(s->ifft[0]);
|
||||
av_fft_end(s->ifft[1]);
|
||||
av_fft_end(s->fft[0]);
|
||||
av_fft_end(s->fft[1]);
|
||||
av_tx_uninit(&s->ifft[0]);
|
||||
av_tx_uninit(&s->ifft[1]);
|
||||
av_tx_uninit(&s->fft[0]);
|
||||
av_tx_uninit(&s->fft[1]);
|
||||
av_freep(&s->data_ir[0]);
|
||||
av_freep(&s->data_ir[1]);
|
||||
av_freep(&s->ringbuffer[0]);
|
||||
av_freep(&s->ringbuffer[1]);
|
||||
av_freep(&s->temp_src[0]);
|
||||
av_freep(&s->temp_src[1]);
|
||||
av_freep(&s->temp_fft[0]);
|
||||
av_freep(&s->temp_fft[1]);
|
||||
av_freep(&s->out_fft[0]);
|
||||
av_freep(&s->out_fft[1]);
|
||||
av_freep(&s->in_fft[0]);
|
||||
av_freep(&s->in_fft[1]);
|
||||
av_freep(&s->temp_afft[0]);
|
||||
av_freep(&s->temp_afft[1]);
|
||||
av_freep(&s->data_hrtf[0]);
|
||||
|
Loading…
Reference in New Issue
Block a user