From: Paul B Mahol <onemda@gmail.com>
To: FFmpeg development discussions and patches <ffmpeg-devel@ffmpeg.org>
Subject: [FFmpeg-devel] [PATCH] avfilter/af_firequalizer: switch to TX from lavu
Date: Sun, 13 Nov 2022 14:35:24 +0100
Message-ID: <CAPYw7P5HqsjWDbMmoRfUE8SVS149m2D1h6Gboe=dwtK7Oi2p7w@mail.gmail.com> (raw)
[-- Attachment #1: Type: text/plain, Size: 16 bytes --]
Patch attached.
[-- Attachment #2: 0001-avfilter-af_firequalizer-switch-to-TX-from-lavu.patch --]
[-- Type: text/x-patch, Size: 14419 bytes --]
From bf0c69dc29fa4a41de7e47308f6907adf1215df3 Mon Sep 17 00:00:00 2001
From: Paul B Mahol <onemda@gmail.com>
Date: Sat, 12 Nov 2022 12:02:08 +0100
Subject: [PATCH] avfilter/af_firequalizer: switch to TX from lavu
Signed-off-by: Paul B Mahol <onemda@gmail.com>
---
configure | 3 -
libavfilter/af_firequalizer.c | 119 +++++++++++++++++++---------------
2 files changed, 66 insertions(+), 56 deletions(-)
diff --git a/configure b/configure
index e6470dc03b..d2c5ca64b1 100755
--- a/configure
+++ b/configure
@@ -3671,8 +3671,6 @@ elbg_filter_deps="avcodec"
eq_filter_deps="gpl"
erosion_opencl_filter_deps="opencl"
find_rect_filter_deps="avcodec avformat gpl"
-firequalizer_filter_deps="avcodec"
-firequalizer_filter_select="rdft"
flip_vulkan_filter_deps="vulkan spirv_compiler"
flite_filter_deps="libflite"
framerate_filter_select="scene_sad"
@@ -7466,7 +7464,6 @@ enabled cover_rect_filter && prepend avfilter_deps "avformat avcodec"
enabled ebur128_filter && enabled swresample && prepend avfilter_deps "swresample"
enabled elbg_filter && prepend avfilter_deps "avcodec"
enabled find_rect_filter && prepend avfilter_deps "avformat avcodec"
-enabled firequalizer_filter && prepend avfilter_deps "avcodec"
enabled mcdeint_filter && prepend avfilter_deps "avcodec"
enabled movie_filter && prepend avfilter_deps "avformat avcodec"
enabled pan_filter && prepend avfilter_deps "swresample"
diff --git a/libavfilter/af_firequalizer.c b/libavfilter/af_firequalizer.c
index e0f35c139e..156e5e14d7 100644
--- a/libavfilter/af_firequalizer.c
+++ b/libavfilter/af_firequalizer.c
@@ -23,7 +23,7 @@
#include "libavutil/opt.h"
#include "libavutil/eval.h"
#include "libavutil/avassert.h"
-#include "libavcodec/avfft.h"
+#include "libavutil/tx.h"
#include "avfilter.h"
#include "internal.h"
#include "audio.h"
@@ -67,13 +67,20 @@ typedef struct OverlapIndex {
typedef struct FIREqualizerContext {
const AVClass *class;
- RDFTContext *analysis_rdft;
- RDFTContext *analysis_irdft;
- RDFTContext *rdft;
- RDFTContext *irdft;
- FFTContext *fft_ctx;
- RDFTContext *cepstrum_rdft;
- RDFTContext *cepstrum_irdft;
+ AVTXContext *analysis_rdft;
+ av_tx_fn analysis_rdft_fn;
+ AVTXContext *analysis_irdft;
+ av_tx_fn analysis_irdft_fn;
+ AVTXContext *rdft;
+ av_tx_fn rdft_fn;
+ AVTXContext *irdft;
+ av_tx_fn irdft_fn;
+ AVTXContext *fft_ctx;
+ av_tx_fn fft_fn;
+ AVTXContext *cepstrum_rdft;
+ av_tx_fn cepstrum_rdft_fn;
+ AVTXContext *cepstrum_irdft;
+ av_tx_fn cepstrum_irdft_fn;
int analysis_rdft_len;
int rdft_len;
int cepstrum_len;
@@ -151,13 +158,13 @@ AVFILTER_DEFINE_CLASS(firequalizer);
static void common_uninit(FIREqualizerContext *s)
{
- av_rdft_end(s->analysis_rdft);
- av_rdft_end(s->analysis_irdft);
- av_rdft_end(s->rdft);
- av_rdft_end(s->irdft);
- av_fft_end(s->fft_ctx);
- av_rdft_end(s->cepstrum_rdft);
- av_rdft_end(s->cepstrum_irdft);
+ av_tx_uninit(&s->analysis_rdft);
+ av_tx_uninit(&s->analysis_irdft);
+ av_tx_uninit(&s->rdft);
+ av_tx_uninit(&s->irdft);
+ av_tx_uninit(&s->fft_ctx);
+ av_tx_uninit(&s->cepstrum_rdft);
+ av_tx_uninit(&s->cepstrum_irdft);
s->analysis_rdft = s->analysis_irdft = s->rdft = s->irdft = NULL;
s->fft_ctx = NULL;
s->cepstrum_rdft = NULL;
@@ -193,16 +200,14 @@ static void fast_convolute(FIREqualizerContext *av_restrict s, const float *av_r
memset(buf, 0, center * sizeof(*data));
memcpy(buf + center, data, nsamples * sizeof(*data));
memset(buf + center + nsamples, 0, (s->rdft_len - nsamples - center) * sizeof(*data));
- av_rdft_calc(s->rdft, buf);
+ s->rdft_fn(s->rdft, buf, buf, sizeof(float));
- buf[0] *= kernel_buf[0];
- buf[1] *= kernel_buf[s->rdft_len/2];
- for (k = 1; k < s->rdft_len/2; k++) {
+ for (k = 0; k <= s->rdft_len/2; k++) {
buf[2*k] *= kernel_buf[k];
buf[2*k+1] *= kernel_buf[k];
}
- av_rdft_calc(s->irdft, buf);
+ s->irdft_fn(s->irdft, buf, buf, sizeof(float));
for (k = 0; k < s->rdft_len - idx->overlap_idx; k++)
buf[k] += obuf[k];
memcpy(data, buf, nsamples * sizeof(*data));
@@ -230,11 +235,9 @@ static void fast_convolute_nonlinear(FIREqualizerContext *av_restrict s, const f
memcpy(buf, data, nsamples * sizeof(*data));
memset(buf + nsamples, 0, (s->rdft_len - nsamples) * sizeof(*data));
- av_rdft_calc(s->rdft, buf);
+ s->rdft_fn(s->rdft, buf, buf, sizeof(float));
- buf[0] *= kernel_buf[0];
- buf[1] *= kernel_buf[1];
- for (k = 2; k < s->rdft_len; k += 2) {
+ for (k = 0; k <= s->rdft_len; k += 2) {
float re, im;
re = buf[k] * kernel_buf[k] - buf[k+1] * kernel_buf[k+1];
im = buf[k] * kernel_buf[k+1] + buf[k+1] * kernel_buf[k];
@@ -242,7 +245,7 @@ static void fast_convolute_nonlinear(FIREqualizerContext *av_restrict s, const f
buf[k+1] = im;
}
- av_rdft_calc(s->irdft, buf);
+ s->irdft_fn(s->irdft, buf, buf, sizeof(float));
for (k = 0; k < s->rdft_len - idx->overlap_idx; k++)
buf[k] += obuf[k];
memcpy(data, buf, nsamples * sizeof(*data));
@@ -259,12 +262,12 @@ static void fast_convolute_nonlinear(FIREqualizerContext *av_restrict s, const f
}
}
-static void fast_convolute2(FIREqualizerContext *av_restrict s, const float *av_restrict kernel_buf, FFTComplex *av_restrict conv_buf,
+static void fast_convolute2(FIREqualizerContext *av_restrict s, const float *av_restrict kernel_buf, AVComplexFloat *av_restrict conv_buf,
OverlapIndex *av_restrict idx, float *av_restrict data0, float *av_restrict data1, int nsamples)
{
if (nsamples <= s->nsamples_max) {
- FFTComplex *buf = conv_buf + idx->buf_idx * s->rdft_len;
- FFTComplex *obuf = conv_buf + !idx->buf_idx * s->rdft_len + idx->overlap_idx;
+ AVComplexFloat *buf = conv_buf + idx->buf_idx * s->rdft_len;
+ AVComplexFloat *obuf = conv_buf + !idx->buf_idx * s->rdft_len + idx->overlap_idx;
int center = s->fir_len/2;
int k;
float tmp;
@@ -275,8 +278,7 @@ static void fast_convolute2(FIREqualizerContext *av_restrict s, const float *av_
buf[center+k].im = data1[k];
}
memset(buf + center + nsamples, 0, (s->rdft_len - nsamples - center) * sizeof(*buf));
- av_fft_permute(s->fft_ctx, buf);
- av_fft_calc(s->fft_ctx, buf);
+ s->fft_fn(s->fft_ctx, buf, buf, sizeof(float));
/* swap re <-> im, do backward fft using forward fft_ctx */
/* normalize with 0.5f */
@@ -296,8 +298,7 @@ static void fast_convolute2(FIREqualizerContext *av_restrict s, const float *av_
buf[k].re = 0.5f * kernel_buf[k] * buf[k].im;
buf[k].im = 0.5f * kernel_buf[k] * tmp;
- av_fft_permute(s->fft_ctx, buf);
- av_fft_calc(s->fft_ctx, buf);
+ s->fft_fn(s->fft_ctx, buf, buf, sizeof(float));
for (k = 0; k < s->rdft_len - idx->overlap_idx; k++) {
buf[k].re += obuf[k].re;
@@ -361,7 +362,7 @@ static void dump_fir(AVFilterContext *ctx, FILE *fp, int ch)
fprintf(fp, "%15.10f %15.10f\n", (double)x / rate, (double) s->analysis_buf[x]);
}
- av_rdft_calc(s->analysis_rdft, s->analysis_buf);
+ s->analysis_rdft_fn(s->analysis_rdft, s->analysis_buf, s->analysis_buf, sizeof(float));
fprintf(fp, "\n\n# freq[%d] (frequency desired_gain actual_gain)\n", ch);
@@ -533,7 +534,7 @@ static void generate_min_phase_kernel(FIREqualizerContext *s, float *rdft_buf)
memcpy(s->cepstrum_buf, rdft_buf, rdft_len/2 * sizeof(*rdft_buf));
memcpy(s->cepstrum_buf + cepstrum_len - rdft_len/2, rdft_buf + rdft_len/2, rdft_len/2 * sizeof(*rdft_buf));
- av_rdft_calc(s->cepstrum_rdft, s->cepstrum_buf);
+ s->cepstrum_rdft_fn(s->cepstrum_rdft, s->cepstrum_buf, s->cepstrum_buf, sizeof(float));
s->cepstrum_buf[0] = log(FFMAX(s->cepstrum_buf[0], minval));
s->cepstrum_buf[1] = log(FFMAX(s->cepstrum_buf[1], minval));
@@ -543,13 +544,13 @@ static void generate_min_phase_kernel(FIREqualizerContext *s, float *rdft_buf)
s->cepstrum_buf[k+1] = 0;
}
- av_rdft_calc(s->cepstrum_irdft, s->cepstrum_buf);
+ s->cepstrum_irdft_fn(s->cepstrum_irdft, s->cepstrum_buf, s->cepstrum_buf, sizeof(float));
memset(s->cepstrum_buf + cepstrum_len/2 + 1, 0, (cepstrum_len/2 - 1) * sizeof(*s->cepstrum_buf));
for (k = 1; k < cepstrum_len/2; k++)
s->cepstrum_buf[k] *= 2;
- av_rdft_calc(s->cepstrum_rdft, s->cepstrum_buf);
+ s->cepstrum_rdft_fn(s->cepstrum_rdft, s->cepstrum_buf, s->cepstrum_buf, sizeof(float));
s->cepstrum_buf[0] = exp(s->cepstrum_buf[0] * norm) * norm;
s->cepstrum_buf[1] = exp(s->cepstrum_buf[1] * norm) * norm;
@@ -560,7 +561,7 @@ static void generate_min_phase_kernel(FIREqualizerContext *s, float *rdft_buf)
s->cepstrum_buf[k+1] = mag * sin(ph);
}
- av_rdft_calc(s->cepstrum_irdft, s->cepstrum_buf);
+ s->cepstrum_irdft_fn(s->cepstrum_irdft, s->cepstrum_buf, s->cepstrum_buf, sizeof(float));
memset(rdft_buf, 0, s->rdft_len * sizeof(*rdft_buf));
memcpy(rdft_buf, s->cepstrum_buf, s->fir_len * sizeof(*rdft_buf));
@@ -568,7 +569,6 @@ static void generate_min_phase_kernel(FIREqualizerContext *s, float *rdft_buf)
memset(s->analysis_buf, 0, s->analysis_rdft_len * sizeof(*s->analysis_buf));
memcpy(s->analysis_buf, s->cepstrum_buf, s->fir_len * sizeof(*s->analysis_buf));
}
-
}
static int generate_kernel(AVFilterContext *ctx, const char *gain, const char *gain_entry)
@@ -641,7 +641,7 @@ static int generate_kernel(AVFilterContext *ctx, const char *gain, const char *g
if (s->dump_buf)
memcpy(s->dump_buf, s->analysis_buf, s->analysis_rdft_len * sizeof(*s->analysis_buf));
- av_rdft_calc(s->analysis_irdft, s->analysis_buf);
+ s->analysis_irdft_fn(s->analysis_irdft, s->analysis_buf, s->analysis_buf, sizeof(float));
center = s->fir_len / 2;
for (k = 0; k <= center; k++) {
@@ -691,7 +691,7 @@ static int generate_kernel(AVFilterContext *ctx, const char *gain, const char *g
memcpy(rdft_buf + s->rdft_len/2, s->analysis_buf + s->analysis_rdft_len - s->rdft_len/2, s->rdft_len/2 * sizeof(*s->analysis_buf));
if (s->min_phase)
generate_min_phase_kernel(s, rdft_buf);
- av_rdft_calc(s->rdft, rdft_buf);
+ s->rdft_fn(s->rdft, rdft_buf, rdft_buf, sizeof(float));
for (k = 0; k < s->rdft_len; k++) {
if (isnan(rdft_buf[k]) || isinf(rdft_buf[k])) {
@@ -731,7 +731,8 @@ static int config_input(AVFilterLink *inlink)
{
AVFilterContext *ctx = inlink->dst;
FIREqualizerContext *s = ctx->priv;
- int rdft_bits;
+ float iscale, scale = 1.f;
+ int rdft_bits, ret;
common_uninit(s);
@@ -753,11 +754,15 @@ static int config_input(AVFilterLink *inlink)
return AVERROR(EINVAL);
}
- if (!(s->rdft = av_rdft_init(rdft_bits, DFT_R2C)) || !(s->irdft = av_rdft_init(rdft_bits, IDFT_C2R)))
- return AVERROR(ENOMEM);
+ iscale = 0.5f;
+ if (((ret = av_tx_init(&s->rdft, &s->rdft_fn, AV_TX_FLOAT_RDFT, 0, 1 << rdft_bits, &scale, AV_TX_INPLACE)) < 0) ||
+ ((ret = av_tx_init(&s->irdft, &s->irdft_fn, AV_TX_FLOAT_RDFT, 1, 1 << rdft_bits, &iscale, AV_TX_INPLACE)) < 0))
+ return ret;
- if (s->fft2 && !s->multi && inlink->ch_layout.nb_channels > 1 && !(s->fft_ctx = av_fft_init(rdft_bits, 0)))
- return AVERROR(ENOMEM);
+ scale = 1.f;
+ if (s->fft2 && !s->multi && inlink->ch_layout.nb_channels > 1 &&
+ ((ret = av_tx_init(&s->fft_ctx, &s->fft_fn, AV_TX_FLOAT_FFT, 0, 1 << rdft_bits, &scale, AV_TX_INPLACE)) < 0))
+ return ret;
if (s->min_phase) {
int cepstrum_bits = rdft_bits + 2;
@@ -767,10 +772,15 @@ static int config_input(AVFilterLink *inlink)
}
cepstrum_bits = FFMIN(RDFT_BITS_MAX, cepstrum_bits + 1);
- s->cepstrum_rdft = av_rdft_init(cepstrum_bits, DFT_R2C);
- s->cepstrum_irdft = av_rdft_init(cepstrum_bits, IDFT_C2R);
- if (!s->cepstrum_rdft || !s->cepstrum_irdft)
- return AVERROR(ENOMEM);
+ scale = 1.f;
+ ret = av_tx_init(&s->cepstrum_rdft, &s->cepstrum_rdft_fn, AV_TX_FLOAT_RDFT, 0, 1 << cepstrum_bits, &scale, AV_TX_INPLACE);
+ if (ret < 0)
+ return ret;
+
+ iscale = 8.f / (1 << cepstrum_bits);
+ ret = av_tx_init(&s->cepstrum_irdft, &s->cepstrum_irdft_fn, AV_TX_FLOAT_RDFT, 1, 1 << cepstrum_bits, &iscale, AV_TX_INPLACE);
+ if (ret < 0)
+ return ret;
s->cepstrum_len = 1 << cepstrum_bits;
s->cepstrum_buf = av_malloc_array(s->cepstrum_len, sizeof(*s->cepstrum_buf));
@@ -789,11 +799,14 @@ static int config_input(AVFilterLink *inlink)
return AVERROR(EINVAL);
}
- if (!(s->analysis_irdft = av_rdft_init(rdft_bits, IDFT_C2R)))
- return AVERROR(ENOMEM);
+ iscale = 0.5f;
+ if ((ret = av_tx_init(&s->analysis_irdft, &s->analysis_irdft_fn, AV_TX_FLOAT_RDFT, 1, 1 << rdft_bits, &iscale, AV_TX_INPLACE)) < 0)
+ return ret;
if (s->dumpfile) {
- s->analysis_rdft = av_rdft_init(rdft_bits, DFT_R2C);
+ scale = 1.f;
+ if ((ret = av_tx_init(&s->analysis_rdft, &s->analysis_rdft_fn, AV_TX_FLOAT_RDFT, 0, 1 << rdft_bits, &scale, AV_TX_INPLACE)) < 0)
+ return ret;
s->dump_buf = av_malloc_array(s->analysis_rdft_len, sizeof(*s->dump_buf));
}
@@ -822,7 +835,7 @@ static int filter_frame(AVFilterLink *inlink, AVFrame *frame)
if (!s->min_phase) {
for (ch = 0; ch + 1 < inlink->ch_layout.nb_channels && s->fft_ctx; ch += 2) {
- fast_convolute2(s, s->kernel_buf, (FFTComplex *)(s->conv_buf + 2 * ch * s->rdft_len),
+ fast_convolute2(s, s->kernel_buf, (AVComplexFloat *)(s->conv_buf + 2 * ch * s->rdft_len),
s->conv_idx + ch, (float *) frame->extended_data[ch],
(float *) frame->extended_data[ch+1], frame->nb_samples);
}
--
2.37.2
[-- Attachment #3: Type: text/plain, Size: 251 bytes --]
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
reply other threads:[~2022-11-13 13:35 UTC|newest]
Thread overview: [no followups] expand[flat|nested] mbox.gz Atom feed
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to='CAPYw7P5HqsjWDbMmoRfUE8SVS149m2D1h6Gboe=dwtK7Oi2p7w@mail.gmail.com' \
--to=onemda@gmail.com \
--cc=ffmpeg-devel@ffmpeg.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Git Inbox Mirror of the ffmpeg-devel mailing list - see https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
This inbox may be cloned and mirrored by anyone:
git clone --mirror https://master.gitmailbox.com/ffmpegdev/0 ffmpegdev/git/0.git
# If you have public-inbox 1.1+ installed, you may
# initialize and index your mirror using the following commands:
public-inbox-init -V2 ffmpegdev ffmpegdev/ https://master.gitmailbox.com/ffmpegdev \
ffmpegdev@gitmailbox.com
public-inbox-index ffmpegdev
Example config snippet for mirrors.
AGPL code for this site: git clone https://public-inbox.org/public-inbox.git