mirror of
https://git.ffmpeg.org/ffmpeg.git
synced 2024-12-25 16:52:31 +00:00
avfilter/avf_showcqt: switch to TX FFT from avutil
This commit is contained in:
parent
925d41ebd4
commit
0068b3d0f0
5
configure
vendored
5
configure
vendored
@ -3652,9 +3652,8 @@ scale_qsv_filter_deps="libmfx"
|
||||
scdet_filter_select="scene_sad"
|
||||
select_filter_select="scene_sad"
|
||||
sharpness_vaapi_filter_deps="vaapi"
|
||||
showcqt_filter_deps="avcodec avformat swscale"
|
||||
showcqt_filter_deps="avformat swscale"
|
||||
showcqt_filter_suggest="libfontconfig libfreetype"
|
||||
showcqt_filter_select="fft"
|
||||
showfreqs_filter_deps="avcodec"
|
||||
showfreqs_filter_select="fft"
|
||||
showspatial_filter_deps="avcodec"
|
||||
@ -7281,7 +7280,7 @@ enabled removelogo_filter && prepend avfilter_deps "avformat avcodec swscale"
|
||||
enabled sab_filter && prepend avfilter_deps "swscale"
|
||||
enabled scale_filter && prepend avfilter_deps "swscale"
|
||||
enabled scale2ref_filter && prepend avfilter_deps "swscale"
|
||||
enabled showcqt_filter && prepend avfilter_deps "avformat avcodec swscale"
|
||||
enabled showcqt_filter && prepend avfilter_deps "avformat swscale"
|
||||
enabled showfreqs_filter && prepend avfilter_deps "avcodec"
|
||||
enabled signature_filter && prepend avfilter_deps "avcodec avformat"
|
||||
enabled smartblur_filter && prepend avfilter_deps "swscale"
|
||||
|
@ -19,7 +19,7 @@
|
||||
*/
|
||||
|
||||
#include "config.h"
|
||||
#include "libavcodec/avfft.h"
|
||||
#include "libavutil/tx.h"
|
||||
#include "libavutil/channel_layout.h"
|
||||
#include "libavutil/opt.h"
|
||||
#include "libavutil/xga_font_data.h"
|
||||
@ -144,13 +144,13 @@ static void common_uninit(ShowCQTContext *s)
|
||||
|
||||
av_frame_free(&s->axis_frame);
|
||||
av_frame_free(&s->sono_frame);
|
||||
av_fft_end(s->fft_ctx);
|
||||
s->fft_ctx = NULL;
|
||||
av_tx_uninit(&s->fft_ctx);
|
||||
if (s->coeffs)
|
||||
for (k = 0; k < s->cqt_len; k++)
|
||||
av_freep(&s->coeffs[k].val);
|
||||
av_freep(&s->coeffs);
|
||||
av_freep(&s->fft_data);
|
||||
av_freep(&s->fft_input);
|
||||
av_freep(&s->fft_result);
|
||||
av_freep(&s->cqt_result);
|
||||
av_freep(&s->attack_data);
|
||||
@ -267,15 +267,15 @@ error:
|
||||
return ret;
|
||||
}
|
||||
|
||||
static void cqt_calc(FFTComplex *dst, const FFTComplex *src, const Coeffs *coeffs,
|
||||
static void cqt_calc(AVComplexFloat *dst, const AVComplexFloat *src, const Coeffs *coeffs,
|
||||
int len, int fft_len)
|
||||
{
|
||||
int k, x, i, j;
|
||||
for (k = 0; k < len; k++) {
|
||||
FFTComplex l, r, a = {0,0}, b = {0,0};
|
||||
AVComplexFloat l, r, a = {0,0}, b = {0,0};
|
||||
|
||||
for (x = 0; x < coeffs[k].len; x++) {
|
||||
FFTSample u = coeffs[k].val[x];
|
||||
float u = coeffs[k].val[x];
|
||||
i = coeffs[k].start + x;
|
||||
j = fft_len - i;
|
||||
a.re += u * src[i].re;
|
||||
@ -730,7 +730,7 @@ static float calculate_gamma(float v, float g)
|
||||
return expf(logf(v) / g);
|
||||
}
|
||||
|
||||
static void rgb_from_cqt(ColorFloat *c, const FFTComplex *v, float g, int len, float cscheme[6])
|
||||
static void rgb_from_cqt(ColorFloat *c, const AVComplexFloat *v, float g, int len, float cscheme[6])
|
||||
{
|
||||
int x;
|
||||
for (x = 0; x < len; x++) {
|
||||
@ -740,7 +740,7 @@ static void rgb_from_cqt(ColorFloat *c, const FFTComplex *v, float g, int len, f
|
||||
}
|
||||
}
|
||||
|
||||
static void yuv_from_cqt(ColorFloat *c, const FFTComplex *v, float gamma, int len, float cm[3][3], float cscheme[6])
|
||||
static void yuv_from_cqt(ColorFloat *c, const AVComplexFloat *v, float gamma, int len, float cm[3][3], float cscheme[6])
|
||||
{
|
||||
int x;
|
||||
for (x = 0; x < len; x++) {
|
||||
@ -1110,7 +1110,7 @@ static void process_cqt(ShowCQTContext *s)
|
||||
if (s->fcount > 1) {
|
||||
float rcp_fcount = 1.0f / s->fcount;
|
||||
for (x = 0; x < s->width; x++) {
|
||||
FFTComplex result = {0.0f, 0.0f};
|
||||
AVComplexFloat result = {0.0f, 0.0f};
|
||||
for (i = 0; i < s->fcount; i++) {
|
||||
result.re += s->cqt_result[s->fcount * x + i].re;
|
||||
result.im += s->cqt_result[s->fcount * x + i].im;
|
||||
@ -1139,17 +1139,16 @@ static int plot_cqt(AVFilterContext *ctx, AVFrame **frameout)
|
||||
|
||||
last_time = av_gettime_relative();
|
||||
|
||||
memcpy(s->fft_result, s->fft_data, s->fft_len * sizeof(*s->fft_data));
|
||||
memcpy(s->fft_input, s->fft_data, s->fft_len * sizeof(*s->fft_data));
|
||||
if (s->attack_data) {
|
||||
int k;
|
||||
for (k = 0; k < s->remaining_fill_max; k++) {
|
||||
s->fft_result[s->fft_len/2+k].re *= s->attack_data[k];
|
||||
s->fft_result[s->fft_len/2+k].im *= s->attack_data[k];
|
||||
s->fft_input[s->fft_len/2+k].re *= s->attack_data[k];
|
||||
s->fft_input[s->fft_len/2+k].im *= s->attack_data[k];
|
||||
}
|
||||
}
|
||||
|
||||
av_fft_permute(s->fft_ctx, s->fft_result);
|
||||
av_fft_calc(s->fft_ctx, s->fft_result);
|
||||
s->tx_fn(s->fft_ctx, s->fft_result, s->fft_input, sizeof(float));
|
||||
s->fft_result[s->fft_len] = s->fft_result[0];
|
||||
UPDATE_TIME(s->fft_time);
|
||||
|
||||
@ -1355,6 +1354,7 @@ static int config_output(AVFilterLink *outlink)
|
||||
AVFilterContext *ctx = outlink->src;
|
||||
AVFilterLink *inlink = ctx->inputs[0];
|
||||
ShowCQTContext *s = ctx->priv;
|
||||
float scale;
|
||||
int ret;
|
||||
|
||||
common_uninit(s);
|
||||
@ -1380,9 +1380,10 @@ static int config_output(AVFilterLink *outlink)
|
||||
s->fft_len = 1 << s->fft_bits;
|
||||
av_log(ctx, AV_LOG_INFO, "fft_len = %d, cqt_len = %d.\n", s->fft_len, s->cqt_len);
|
||||
|
||||
s->fft_ctx = av_fft_init(s->fft_bits, 0);
|
||||
ret = av_tx_init(&s->fft_ctx, &s->tx_fn, AV_TX_FLOAT_FFT, 0, s->fft_len, &scale, 0);
|
||||
s->fft_data = av_calloc(s->fft_len, sizeof(*s->fft_data));
|
||||
s->fft_result = av_calloc(s->fft_len + 64, sizeof(*s->fft_result));
|
||||
s->fft_input = av_calloc(FFALIGN(s->fft_len + 64, 256), sizeof(*s->fft_input));
|
||||
s->fft_result = av_calloc(FFALIGN(s->fft_len + 64, 256), sizeof(*s->fft_result));
|
||||
s->cqt_result = av_malloc_array(s->cqt_len, sizeof(*s->cqt_result));
|
||||
if (!s->fft_ctx || !s->fft_data || !s->fft_result || !s->cqt_result)
|
||||
return AVERROR(ENOMEM);
|
||||
|
@ -21,12 +21,12 @@
|
||||
#ifndef AVFILTER_SHOWCQT_H
|
||||
#define AVFILTER_SHOWCQT_H
|
||||
|
||||
#include "libavcodec/avfft.h"
|
||||
#include "libavutil/tx.h"
|
||||
#include "avfilter.h"
|
||||
#include "internal.h"
|
||||
|
||||
typedef struct Coeffs {
|
||||
FFTSample *val;
|
||||
float *val;
|
||||
int start, len;
|
||||
} Coeffs;
|
||||
|
||||
@ -58,11 +58,13 @@ typedef struct ShowCQTContext {
|
||||
int remaining_fill_max;
|
||||
int64_t next_pts;
|
||||
double *freq;
|
||||
FFTContext *fft_ctx;
|
||||
AVTXContext *fft_ctx;
|
||||
av_tx_fn tx_fn;
|
||||
Coeffs *coeffs;
|
||||
FFTComplex *fft_data;
|
||||
FFTComplex *fft_result;
|
||||
FFTComplex *cqt_result;
|
||||
AVComplexFloat *fft_data;
|
||||
AVComplexFloat *fft_input;
|
||||
AVComplexFloat *fft_result;
|
||||
AVComplexFloat *cqt_result;
|
||||
float *attack_data;
|
||||
int fft_bits;
|
||||
int fft_len;
|
||||
@ -76,7 +78,7 @@ typedef struct ShowCQTContext {
|
||||
float cmatrix[3][3];
|
||||
float cscheme_v[6];
|
||||
/* callback */
|
||||
void (*cqt_calc)(FFTComplex *dst, const FFTComplex *src, const Coeffs *coeffs,
|
||||
void (*cqt_calc)(AVComplexFloat *dst, const AVComplexFloat *src, const Coeffs *coeffs,
|
||||
int len, int fft_len);
|
||||
void (*permute_coeffs)(float *v, int len);
|
||||
void (*draw_bar)(AVFrame *out, const float *h, const float *rcp_h,
|
||||
|
@ -24,7 +24,7 @@
|
||||
#include "libavfilter/avf_showcqt.h"
|
||||
|
||||
#define DECLARE_CQT_CALC(type) \
|
||||
void ff_showcqt_cqt_calc_##type(FFTComplex *dst, const FFTComplex *src, \
|
||||
void ff_showcqt_cqt_calc_##type(AVComplexFloat *dst, const AVComplexFloat *src, \
|
||||
const Coeffs *coeffs, int len, int fft_len)
|
||||
|
||||
DECLARE_CQT_CALC(sse);
|
||||
|
Loading…
Reference in New Issue
Block a user