highperfocused/ffmpeg
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

avfilter/af_afir: switch to lavu/tx

Browse Source
This commit is contained in:
Paul B Mahol 2022-01-29 11:35:40 +01:00
parent 8ca06a8148
commit d388dc20b9
2 changed files with 56 additions and 53 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

98
libavfilter/af_afir.c
View File

@ -25,6 +25,7 @@
#include <float.h> #include <float.h>
#include "libavutil/tx.h"
#include "libavutil/avstring.h" #include "libavutil/avstring.h"
#include "libavutil/channel_layout.h" #include "libavutil/channel_layout.h"
#include "libavutil/common.h" #include "libavutil/common.h"
@ -32,7 +33,6 @@
#include "libavutil/intreadwrite.h" #include "libavutil/intreadwrite.h"
#include "libavutil/opt.h" #include "libavutil/opt.h"
#include "libavutil/xga_font_data.h" #include "libavutil/xga_font_data.h"
#include "libavcodec/avfft.h"
#include "audio.h" #include "audio.h"
#include "avfilter.h" #include "avfilter.h"
@ -58,7 +58,7 @@ static void fcmul_add_c(float *sum, const float *t, const float *c, ptrdiff_t le
sum[2 * n] += t[2 * n] * c[2 * n]; sum[2 * n] += t[2 * n] * c[2 * n];
} }
static void direct(const float *in, const FFTComplex *ir, int len, float *out) static void direct(const float *in, const AVComplexFloat *ir, int len, float *out)
{ {
for (int n = 0; n < len; n++) for (int n = 0; n < len; n++)
for (int m = 0; m <= n; m++) for (int m = 0; m <= n; m++)
@ -79,7 +79,7 @@ static int fir_quantum(AVFilterContext *ctx, AVFrame *out, int ch, int offset)
{ {
AudioFIRContext *s = ctx->priv; AudioFIRContext *s = ctx->priv;
const float *in = (const float *)s->in->extended_data[ch] + offset; const float *in = (const float *)s->in->extended_data[ch] + offset;
float *block, *buf, *ptr = (float *)out->extended_data[ch] + offset; float *blockin, *blockout, *buf, *ptr = (float *)out->extended_data[ch] + offset;
const int nb_samples = FFMIN(s->min_part_size, out->nb_samples - offset); const int nb_samples = FFMIN(s->min_part_size, out->nb_samples - offset);
int n, i, j; int n, i, j;
@ -87,7 +87,8 @@ static int fir_quantum(AVFilterContext *ctx, AVFrame *out, int ch, int offset)
AudioFIRSegment *seg = &s->seg[segment]; AudioFIRSegment *seg = &s->seg[segment];
float *src = (float *)seg->input->extended_data[ch]; float *src = (float *)seg->input->extended_data[ch];
float *dst = (float *)seg->output->extended_data[ch]; float *dst = (float *)seg->output->extended_data[ch];
float *sum = (float *)seg->sum->extended_data[ch]; float *sumin = (float *)seg->sumin->extended_data[ch];
float *sumout = (float *)seg->sumout->extended_data[ch];
if (s->min_part_size >= 8) { if (s->min_part_size >= 8) {
s->fdsp->vector_fmul_scalar(src + seg->input_offset, in, s->dry_gain, FFALIGN(nb_samples, 4)); s->fdsp->vector_fmul_scalar(src + seg->input_offset, in, s->dry_gain, FFALIGN(nb_samples, 4));
@ -115,7 +116,7 @@ static int fir_quantum(AVFilterContext *ctx, AVFrame *out, int ch, int offset)
for (i = 0; i < seg->nb_partitions; i++) { for (i = 0; i < seg->nb_partitions; i++) {
const int coffset = j * seg->coeff_size; const int coffset = j * seg->coeff_size;
const FFTComplex *coeff = (const FFTComplex *)seg->coeff->extended_data[ch * !s->one2many] + coffset; const AVComplexFloat *coeff = (const AVComplexFloat *)seg->coeff->extended_data[ch * !s->one2many] + coffset;
direct(src, coeff, nb_samples, dst); direct(src, coeff, nb_samples, dst);
@ -134,40 +135,38 @@ static int fir_quantum(AVFilterContext *ctx, AVFrame *out, int ch, int offset)
continue; continue;
} }
memset(sum, 0, sizeof(*sum) * seg->fft_length); memset(sumin, 0, sizeof(*sumin) * seg->fft_length);
block = (float *)seg->block->extended_data[ch] + seg->part_index[ch] * seg->block_size; blockin = (float *)seg->blockin->extended_data[ch] + seg->part_index[ch] * seg->block_size;
memset(block + seg->part_size, 0, sizeof(*block) * (seg->fft_length - seg->part_size)); blockout = (float *)seg->blockout->extended_data[ch] + seg->part_index[ch] * seg->block_size;
memset(blockin + seg->part_size, 0, sizeof(*blockin) * (seg->fft_length - seg->part_size));
memcpy(block, src, sizeof(*src) * seg->part_size); memcpy(blockin, src, sizeof(*src) * seg->part_size);
av_rdft_calc(seg->rdft[ch], block); seg->tx_fn(seg->tx[ch], blockout, blockin, sizeof(float));
block[2 * seg->part_size] = block[1];
block[1] = 0;
j = seg->part_index[ch]; j = seg->part_index[ch];
for (i = 0; i < seg->nb_partitions; i++) { for (i = 0; i < seg->nb_partitions; i++) {
const int coffset = j * seg->coeff_size; const int coffset = j * seg->coeff_size;
const float *block = (const float *)seg->block->extended_data[ch] + i * seg->block_size; const float *blockout = (const float *)seg->blockout->extended_data[ch] + i * seg->block_size;
const FFTComplex *coeff = (const FFTComplex *)seg->coeff->extended_data[ch * !s->one2many] + coffset; const AVComplexFloat *coeff = (const AVComplexFloat *)seg->coeff->extended_data[ch * !s->one2many] + coffset;
s->afirdsp.fcmul_add(sum, block, (const float *)coeff, seg->part_size); s->afirdsp.fcmul_add(sumin, blockout, (const float *)coeff, seg->part_size);
if (j == 0) if (j == 0)
j = seg->nb_partitions; j = seg->nb_partitions;
j--; j--;
} }
sum[1] = sum[2 * seg->part_size]; seg->itx_fn(seg->itx[ch], sumout, sumin, sizeof(float));
av_rdft_calc(seg->irdft[ch], sum);
buf = (float *)seg->buffer->extended_data[ch]; buf = (float *)seg->buffer->extended_data[ch];
fir_fadd(s, buf, sum, seg->part_size); fir_fadd(s, buf, sumout, seg->part_size);
memcpy(dst, buf, seg->part_size * sizeof(*dst)); memcpy(dst, buf, seg->part_size * sizeof(*dst));
buf = (float *)seg->buffer->extended_data[ch]; buf = (float *)seg->buffer->extended_data[ch];
memcpy(buf, sum + seg->part_size, seg->part_size * sizeof(*buf)); memcpy(buf, sumout + seg->part_size, seg->part_size * sizeof(*buf));
seg->part_index[ch] = (seg->part_index[ch] + 1) % seg->nb_partitions; seg->part_index[ch] = (seg->part_index[ch] + 1) % seg->nb_partitions;
@ -381,9 +380,9 @@ static int init_segment(AVFilterContext *ctx, AudioFIRSegment *seg,
{ {
AudioFIRContext *s = ctx->priv; AudioFIRContext *s = ctx->priv;
seg->rdft = av_calloc(ctx->inputs[0]->channels, sizeof(*seg->rdft)); seg->tx = av_calloc(ctx->inputs[0]->channels, sizeof(*seg->tx));
seg->irdft = av_calloc(ctx->inputs[0]->channels, sizeof(*seg->irdft)); seg->itx = av_calloc(ctx->inputs[0]->channels, sizeof(*seg->itx));
if (!seg->rdft || !seg->irdft) if (!seg->tx || !seg->itx)
return AVERROR(ENOMEM); return AVERROR(ENOMEM);
seg->fft_length = part_size * 2 + 1; seg->fft_length = part_size * 2 + 1;
@ -400,19 +399,22 @@ static int init_segment(AVFilterContext *ctx, AudioFIRSegment *seg,
return AVERROR(ENOMEM); return AVERROR(ENOMEM);
for (int ch = 0; ch < ctx->inputs[0]->channels && part_size >= 8; ch++) { for (int ch = 0; ch < ctx->inputs[0]->channels && part_size >= 8; ch++) {
seg->rdft[ch] = av_rdft_init(av_log2(2 * part_size), DFT_R2C); float scale = 1.f, iscale = 1.f / part_size;
seg->irdft[ch] = av_rdft_init(av_log2(2 * part_size), IDFT_C2R); av_tx_init(&seg->tx[ch], &seg->tx_fn, AV_TX_FLOAT_RDFT, 0, 2 * part_size, &scale, 0);
if (!seg->rdft[ch] || !seg->irdft[ch]) av_tx_init(&seg->itx[ch], &seg->itx_fn, AV_TX_FLOAT_RDFT, 1, 2 * part_size, &iscale, 0);
if (!seg->tx[ch] || !seg->itx[ch])
return AVERROR(ENOMEM); return AVERROR(ENOMEM);
} }
seg->sum = ff_get_audio_buffer(ctx->inputs[0], seg->fft_length); seg->sumin = ff_get_audio_buffer(ctx->inputs[0], seg->fft_length);
seg->block = ff_get_audio_buffer(ctx->inputs[0], seg->nb_partitions * seg->block_size); seg->sumout = ff_get_audio_buffer(ctx->inputs[0], seg->fft_length);
seg->blockin = ff_get_audio_buffer(ctx->inputs[0], seg->nb_partitions * seg->block_size);
seg->blockout = ff_get_audio_buffer(ctx->inputs[0], seg->nb_partitions * seg->block_size);
seg->buffer = ff_get_audio_buffer(ctx->inputs[0], seg->part_size); seg->buffer = ff_get_audio_buffer(ctx->inputs[0], seg->part_size);
seg->coeff = ff_get_audio_buffer(ctx->inputs[1 + s->selir], seg->nb_partitions * seg->coeff_size * 2); seg->coeff = ff_get_audio_buffer(ctx->inputs[1 + s->selir], seg->nb_partitions * seg->coeff_size * 2);
seg->input = ff_get_audio_buffer(ctx->inputs[0], seg->input_size); seg->input = ff_get_audio_buffer(ctx->inputs[0], seg->input_size);
seg->output = ff_get_audio_buffer(ctx->inputs[0], seg->part_size); seg->output = ff_get_audio_buffer(ctx->inputs[0], seg->part_size);
if (!seg->buffer || !seg->sum || !seg->block || !seg->coeff || !seg->input || !seg->output) if (!seg->buffer || !seg->sumin || !seg->sumout || !seg->blockin || !seg->blockout || !seg->coeff || !seg->input || !seg->output)
return AVERROR(ENOMEM); return AVERROR(ENOMEM);
return 0; return 0;
@ -422,25 +424,27 @@ static void uninit_segment(AVFilterContext *ctx, AudioFIRSegment *seg)
{ {
AudioFIRContext *s = ctx->priv; AudioFIRContext *s = ctx->priv;
if (seg->rdft) { if (seg->tx) {
for (int ch = 0; ch < s->nb_channels; ch++) { for (int ch = 0; ch < s->nb_channels; ch++) {
av_rdft_end(seg->rdft[ch]); av_tx_uninit(&seg->tx[ch]);
} }
} }
av_freep(&seg->rdft); av_freep(&seg->tx);
if (seg->irdft) { if (seg->itx) {
for (int ch = 0; ch < s->nb_channels; ch++) { for (int ch = 0; ch < s->nb_channels; ch++) {
av_rdft_end(seg->irdft[ch]); av_tx_uninit(&seg->itx[ch]);
} }
} }
av_freep(&seg->irdft); av_freep(&seg->itx);
av_freep(&seg->output_offset); av_freep(&seg->output_offset);
av_freep(&seg->part_index); av_freep(&seg->part_index);
av_frame_free(&seg->block); av_frame_free(&seg->blockin);
av_frame_free(&seg->sum); av_frame_free(&seg->blockout);
av_frame_free(&seg->sumin);
av_frame_free(&seg->sumout);
av_frame_free(&seg->buffer); av_frame_free(&seg->buffer);
av_frame_free(&seg->coeff); av_frame_free(&seg->coeff);
av_frame_free(&seg->input); av_frame_free(&seg->input);
@ -558,13 +562,13 @@ static int convert_coeffs(AVFilterContext *ctx)
for (int segment = 0; segment < s->nb_segments; segment++) { for (int segment = 0; segment < s->nb_segments; segment++) {
AudioFIRSegment *seg = &s->seg[segment]; AudioFIRSegment *seg = &s->seg[segment];
float *block = (float *)seg->block->extended_data[ch]; float *blockin = (float *)seg->blockin->extended_data[ch];
FFTComplex *coeff = (FFTComplex *)seg->coeff->extended_data[ch]; float *blockout = (float *)seg->blockout->extended_data[ch];
AVComplexFloat *coeff = (AVComplexFloat *)seg->coeff->extended_data[ch];
av_log(ctx, AV_LOG_DEBUG, "segment: %d\n", segment); av_log(ctx, AV_LOG_DEBUG, "segment: %d\n", segment);
for (i = 0; i < seg->nb_partitions; i++) { for (i = 0; i < seg->nb_partitions; i++) {
const float scale = 1.f / seg->part_size;
const int coffset = i * seg->coeff_size; const int coffset = i * seg->coeff_size;
const int remaining = s->nb_taps - toffset; const int remaining = s->nb_taps - toffset;
const int size = remaining >= seg->part_size ? seg->part_size : remaining; const int size = remaining >= seg->part_size ? seg->part_size : remaining;
@ -577,19 +581,15 @@ static int convert_coeffs(AVFilterContext *ctx)
continue; continue;
} }
memset(block, 0, sizeof(*block) * seg->fft_length); memset(blockin, 0, sizeof(*blockin) * seg->fft_length);
memcpy(block, time + toffset, size * sizeof(*block)); memcpy(blockin, time + toffset, size * sizeof(*blockin));
av_rdft_calc(seg->rdft[0], block); seg->tx_fn(seg->tx[0], blockout, blockin, sizeof(float));
coeff[coffset].re = block[0] * scale; for (n = 0; n < seg->part_size + 1; n++) {
coeff[coffset].im = 0; coeff[coffset + n].re = blockout[2 * n];
for (n = 1; n < seg->part_size; n++) { coeff[coffset + n].im = blockout[2 * n + 1];
coeff[coffset + n].re = block[2 * n] * scale;
coeff[coffset + n].im = block[2 * n + 1] * scale;
} }
coeff[coffset + seg->part_size].re = block[1] * scale;
coeff[coffset + seg->part_size].im = 0;
toffset += size; toffset += size;
} }

11
libavfilter/af_afir.h
View File

@ -21,10 +21,10 @@
#ifndef AVFILTER_AFIR_H #ifndef AVFILTER_AFIR_H
#define AVFILTER_AFIR_H #define AVFILTER_AFIR_H
#include "libavutil/tx.h"
#include "libavutil/common.h" #include "libavutil/common.h"
#include "libavutil/float_dsp.h" #include "libavutil/float_dsp.h"
#include "libavutil/opt.h" #include "libavutil/opt.h"
#include "libavcodec/avfft.h"
#include "audio.h" #include "audio.h"
#include "avfilter.h" #include "avfilter.h"
@ -43,14 +43,17 @@ typedef struct AudioFIRSegment {
int *output_offset; int *output_offset;
int *part_index; int *part_index;
AVFrame *sum; AVFrame *sumin;
AVFrame *block; AVFrame *sumout;
AVFrame *blockin;
AVFrame *blockout;
AVFrame *buffer; AVFrame *buffer;
AVFrame *coeff; AVFrame *coeff;
AVFrame *input; AVFrame *input;
AVFrame *output; AVFrame *output;
RDFTContext **rdft, **irdft; AVTXContext **tx, **itx;
av_tx_fn tx_fn, itx_fn;
} AudioFIRSegment; } AudioFIRSegment;
typedef struct AudioFIRDSPContext { typedef struct AudioFIRDSPContext {