2 * Copyright (c) 2017 Paul B Mahol
4 * This file is part of FFmpeg.
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 * An arbitrary audio FIR filter
28 #include "libavutil/audio_fifo.h"
29 #include "libavutil/common.h"
30 #include "libavutil/float_dsp.h"
31 #include "libavutil/intreadwrite.h"
32 #include "libavutil/opt.h"
33 #include "libavutil/xga_font_data.h"
34 #include "libavcodec/avfft.h"
42 static void fcmul_add_c(float *sum, const float *t, const float *c, ptrdiff_t len)
46 for (n = 0; n < len; n++) {
47 const float cre = c[2 * n ];
48 const float cim = c[2 * n + 1];
49 const float tre = t[2 * n ];
50 const float tim = t[2 * n + 1];
52 sum[2 * n ] += tre * cre - tim * cim;
53 sum[2 * n + 1] += tre * cim + tim * cre;
56 sum[2 * n] += t[2 * n] * c[2 * n];
59 static int fir_channel(AVFilterContext *ctx, void *arg, int ch, int nb_jobs)
61 AudioFIRContext *s = ctx->priv;
62 const float *src = (const float *)s->in[0]->extended_data[ch];
63 int index1 = (s->index + 1) % 3;
64 int index2 = (s->index + 2) % 3;
65 float *sum = s->sum[ch];
71 memset(sum, 0, sizeof(*sum) * s->fft_length);
72 block = s->block[ch] + s->part_index * s->block_size;
73 memset(block, 0, sizeof(*block) * s->fft_length);
75 s->fdsp->vector_fmul_scalar(block + s->part_size, src, s->dry_gain, FFALIGN(s->nb_samples, 4));
78 av_rdft_calc(s->rdft[ch], block);
79 block[2 * s->part_size] = block[1];
84 for (i = 0; i < s->nb_partitions; i++) {
85 const int coffset = i * s->coeff_size;
86 const FFTComplex *coeff = s->coeff[ch * !s->one2many] + coffset;
88 block = s->block[ch] + j * s->block_size;
89 s->fcmul_add(sum, block, (const float *)coeff, s->part_size);
96 sum[1] = sum[2 * s->part_size];
97 av_rdft_calc(s->irdft[ch], sum);
99 dst = (float *)s->buffer->extended_data[ch] + index1 * s->part_size;
100 for (n = 0; n < s->part_size; n++) {
104 dst = (float *)s->buffer->extended_data[ch] + index2 * s->part_size;
106 memcpy(dst, sum + s->part_size, s->part_size * sizeof(*dst));
108 dst = (float *)s->buffer->extended_data[ch] + s->index * s->part_size;
111 float *ptr = (float *)out->extended_data[ch];
112 s->fdsp->vector_fmul_scalar(ptr, dst, s->wet_gain, FFALIGN(out->nb_samples, 4));
119 static int fir_frame(AudioFIRContext *s, AVFilterLink *outlink)
121 AVFilterContext *ctx = outlink->src;
125 s->nb_samples = FFMIN(s->part_size, av_audio_fifo_size(s->fifo[0]));
128 out = ff_get_audio_buffer(outlink, s->nb_samples);
130 return AVERROR(ENOMEM);
133 s->in[0] = ff_get_audio_buffer(ctx->inputs[0], s->nb_samples);
136 return AVERROR(ENOMEM);
139 av_audio_fifo_peek(s->fifo[0], (void **)s->in[0]->extended_data, s->nb_samples);
141 ctx->internal->execute(ctx, fir_channel, out, NULL, outlink->channels);
143 s->part_index = (s->part_index + 1) % s->nb_partitions;
145 av_audio_fifo_drain(s->fifo[0], s->nb_samples);
149 if (s->pts != AV_NOPTS_VALUE)
150 s->pts += av_rescale_q(out->nb_samples, (AVRational){1, outlink->sample_rate}, outlink->time_base);
157 av_frame_free(&s->in[0]);
159 if (s->want_skip == 1) {
163 ret = ff_filter_frame(outlink, out);
169 static void drawtext(AVFrame *pic, int x, int y, const char *txt, uint32_t color)
175 font = avpriv_cga_font, font_height = 8;
177 for (i = 0; txt[i]; i++) {
180 uint8_t *p = pic->data[0] + y * pic->linesize[0] + (x + i * 8) * 4;
181 for (char_y = 0; char_y < font_height; char_y++) {
182 for (mask = 0x80; mask; mask >>= 1) {
183 if (font[txt[i] * font_height + char_y] & mask)
187 p += pic->linesize[0] - 8 * 4;
192 static void draw_line(AVFrame *out, int x0, int y0, int x1, int y1, uint32_t color)
194 int dx = FFABS(x1-x0);
195 int dy = FFABS(y1-y0), sy = y0 < y1 ? 1 : -1;
196 int err = (dx>dy ? dx : -dy) / 2, e2;
199 AV_WL32(out->data[0] + y0 * out->linesize[0] + x0 * 4, color);
201 if (x0 == x1 && y0 == y1)
218 static void draw_response(AVFilterContext *ctx, AVFrame *out)
220 AudioFIRContext *s = ctx->priv;
221 float *mag, *phase, min = FLT_MAX, max = FLT_MIN;
222 int prev_ymag = -1, prev_yphase = -1;
226 memset(out->data[0], 0, s->h * out->linesize[0]);
228 phase = av_malloc_array(s->w, sizeof(*phase));
229 mag = av_malloc_array(s->w, sizeof(*mag));
233 channel = av_clip(s->ir_channel, 0, s->in[1]->channels - 1);
234 for (i = 0; i < s->w; i++) {
235 const float *src = (const float *)s->in[1]->extended_data[channel];
236 double w = i * M_PI / (s->w - 1);
240 for (x = 0; x < s->nb_taps; x++) {
241 real += cos(-x * w) * src[x];
242 imag += sin(-x * w) * src[x];
245 mag[i] = hypot(real, imag);
246 phase[i] = atan2(imag, real);
247 min = fminf(min, mag[i]);
248 max = fmaxf(max, mag[i]);
251 for (i = 0; i < s->w; i++) {
252 int ymag = mag[i] / max * (s->h - 1);
253 int yphase = (0.5 * (1. + phase[i] / M_PI)) * (s->h - 1);
255 ymag = s->h - 1 - av_clip(ymag, 0, s->h - 1);
256 yphase = s->h - 1 - av_clip(yphase, 0, s->h - 1);
261 prev_yphase = yphase;
263 draw_line(out, i, ymag, FFMAX(i - 1, 0), prev_ymag, 0xFFFF00FF);
264 draw_line(out, i, yphase, FFMAX(i - 1, 0), prev_yphase, 0xFF00FF00);
267 prev_yphase = yphase;
270 if (s->w > 400 && s->h > 100) {
271 drawtext(out, 2, 2, "Max Magnitude:", 0xDDDDDDDD);
272 snprintf(text, sizeof(text), "%.2f", max);
273 drawtext(out, 15 * 8 + 2, 2, text, 0xDDDDDDDD);
275 drawtext(out, 2, 12, "Min Magnitude:", 0xDDDDDDDD);
276 snprintf(text, sizeof(text), "%.2f", min);
277 drawtext(out, 15 * 8 + 2, 12, text, 0xDDDDDDDD);
285 static int convert_coeffs(AVFilterContext *ctx)
287 AudioFIRContext *s = ctx->priv;
290 s->nb_taps = av_audio_fifo_size(s->fifo[1]);
292 return AVERROR(EINVAL);
294 for (n = 4; (1 << n) < s->nb_taps; n++);
296 s->ir_length = 1 << n;
297 s->fft_length = (1 << (N + 1)) + 1;
298 s->part_size = 1 << (N - 1);
299 s->block_size = FFALIGN(s->fft_length, 32);
300 s->coeff_size = FFALIGN(s->part_size + 1, 32);
301 s->nb_partitions = (s->nb_taps + s->part_size - 1) / s->part_size;
302 s->nb_coeffs = s->ir_length + s->nb_partitions;
304 for (ch = 0; ch < ctx->inputs[0]->channels; ch++) {
305 s->sum[ch] = av_calloc(s->fft_length, sizeof(**s->sum));
307 return AVERROR(ENOMEM);
310 for (ch = 0; ch < ctx->inputs[1]->channels; ch++) {
311 s->coeff[ch] = av_calloc(s->nb_partitions * s->coeff_size, sizeof(**s->coeff));
313 return AVERROR(ENOMEM);
316 for (ch = 0; ch < ctx->inputs[0]->channels; ch++) {
317 s->block[ch] = av_calloc(s->nb_partitions * s->block_size, sizeof(**s->block));
319 return AVERROR(ENOMEM);
322 for (ch = 0; ch < ctx->inputs[0]->channels; ch++) {
323 s->rdft[ch] = av_rdft_init(N, DFT_R2C);
324 s->irdft[ch] = av_rdft_init(N, IDFT_C2R);
325 if (!s->rdft[ch] || !s->irdft[ch])
326 return AVERROR(ENOMEM);
329 s->in[1] = ff_get_audio_buffer(ctx->inputs[1], s->nb_taps);
331 return AVERROR(ENOMEM);
333 s->buffer = ff_get_audio_buffer(ctx->inputs[0], s->part_size * 3);
335 return AVERROR(ENOMEM);
337 av_audio_fifo_read(s->fifo[1], (void **)s->in[1]->extended_data, s->nb_taps);
340 draw_response(ctx, s->video);
345 for (ch = 0; ch < ctx->inputs[1]->channels; ch++) {
346 float *time = (float *)s->in[1]->extended_data[!s->one2many * ch];
348 for (i = 0; i < s->nb_taps; i++)
349 power += FFABS(time[i]);
352 s->gain = sqrtf(1.f / (ctx->inputs[1]->channels * power)) / (sqrtf(ctx->inputs[1]->channels));
353 for (ch = 0; ch < ctx->inputs[1]->channels; ch++) {
354 float *time = (float *)s->in[1]->extended_data[!s->one2many * ch];
356 s->fdsp->vector_fmul_scalar(time, time, s->gain, FFALIGN(s->nb_taps, 4));
360 for (ch = 0; ch < ctx->inputs[1]->channels; ch++) {
361 float *time = (float *)s->in[1]->extended_data[!s->one2many * ch];
362 float *block = s->block[ch];
363 FFTComplex *coeff = s->coeff[ch];
365 for (i = FFMAX(1, s->length * s->nb_taps); i < s->nb_taps; i++)
368 for (i = 0; i < s->nb_partitions; i++) {
369 const float scale = 1.f / s->part_size;
370 const int toffset = i * s->part_size;
371 const int coffset = i * s->coeff_size;
372 const int boffset = s->part_size;
373 const int remaining = s->nb_taps - (i * s->part_size);
374 const int size = remaining >= s->part_size ? s->part_size : remaining;
376 memset(block, 0, sizeof(*block) * s->fft_length);
377 memcpy(block + boffset, time + toffset, size * sizeof(*block));
379 av_rdft_calc(s->rdft[0], block);
381 coeff[coffset].re = block[0] * scale;
382 coeff[coffset].im = 0;
383 for (n = 1; n < s->part_size; n++) {
384 coeff[coffset + n].re = block[2 * n] * scale;
385 coeff[coffset + n].im = block[2 * n + 1] * scale;
387 coeff[coffset + s->part_size].re = block[1] * scale;
388 coeff[coffset + s->part_size].im = 0;
392 av_frame_free(&s->in[1]);
393 av_log(ctx, AV_LOG_DEBUG, "nb_taps: %d\n", s->nb_taps);
394 av_log(ctx, AV_LOG_DEBUG, "nb_partitions: %d\n", s->nb_partitions);
395 av_log(ctx, AV_LOG_DEBUG, "partition size: %d\n", s->part_size);
396 av_log(ctx, AV_LOG_DEBUG, "ir_length: %d\n", s->ir_length);
403 static int read_ir(AVFilterLink *link, AVFrame *frame)
405 AVFilterContext *ctx = link->dst;
406 AudioFIRContext *s = ctx->priv;
407 int nb_taps, max_nb_taps, ret;
409 ret = av_audio_fifo_write(s->fifo[1], (void **)frame->extended_data,
411 av_frame_free(&frame);
415 nb_taps = av_audio_fifo_size(s->fifo[1]);
416 max_nb_taps = s->max_ir_len * ctx->outputs[0]->sample_rate;
417 if (nb_taps > max_nb_taps) {
418 av_log(ctx, AV_LOG_ERROR, "Too big number of coefficients: %d > %d.\n", nb_taps, max_nb_taps);
419 return AVERROR(EINVAL);
425 static int filter_frame(AVFilterLink *link, AVFrame *frame)
427 AVFilterContext *ctx = link->dst;
428 AudioFIRContext *s = ctx->priv;
429 AVFilterLink *outlink = ctx->outputs[0];
432 ret = av_audio_fifo_write(s->fifo[0], (void **)frame->extended_data,
434 if (ret > 0 && s->pts == AV_NOPTS_VALUE)
437 av_frame_free(&frame);
442 if (!s->have_coeffs && s->eof_coeffs) {
443 ret = convert_coeffs(ctx);
448 if (s->response && s->have_coeffs) {
449 s->video->pts = s->pts;
450 ret = ff_filter_frame(ctx->outputs[1], av_frame_clone(s->video));
455 if (s->have_coeffs) {
456 while (av_audio_fifo_size(s->fifo[0]) >= s->part_size) {
457 ret = fir_frame(s, outlink);
465 static int request_frame(AVFilterLink *outlink)
467 AVFilterContext *ctx = outlink->src;
468 AudioFIRContext *s = ctx->priv;
471 if (!s->eof_coeffs) {
472 ret = ff_request_frame(ctx->inputs[1]);
473 if (ret == AVERROR_EOF) {
479 ret = ff_request_frame(ctx->inputs[0]);
480 if (ret == AVERROR_EOF && s->have_coeffs) {
481 if (s->need_padding) {
482 AVFrame *silence = ff_get_audio_buffer(outlink, s->part_size);
485 return AVERROR(ENOMEM);
486 ret = av_audio_fifo_write(s->fifo[0], (void **)silence->extended_data,
487 silence->nb_samples);
488 av_frame_free(&silence);
494 while (av_audio_fifo_size(s->fifo[0]) > 0) {
495 ret = fir_frame(s, outlink);
504 static int query_formats(AVFilterContext *ctx)
506 AudioFIRContext *s = ctx->priv;
507 AVFilterFormats *formats;
508 AVFilterChannelLayouts *layouts;
509 static const enum AVSampleFormat sample_fmts[] = {
513 static const enum AVPixelFormat pix_fmts[] = {
520 AVFilterLink *videolink = ctx->outputs[1];
521 formats = ff_make_format_list(pix_fmts);
522 if ((ret = ff_formats_ref(formats, &videolink->in_formats)) < 0)
526 layouts = ff_all_channel_counts();
527 if ((ret = ff_channel_layouts_ref(layouts, &ctx->outputs[0]->in_channel_layouts)) < 0)
530 for (i = 0; i < 2; i++) {
531 layouts = ff_all_channel_counts();
532 if ((ret = ff_channel_layouts_ref(layouts, &ctx->inputs[i]->out_channel_layouts)) < 0)
536 formats = ff_make_format_list(sample_fmts);
537 if ((ret = ff_set_common_formats(ctx, formats)) < 0)
540 formats = ff_all_samplerates();
541 return ff_set_common_samplerates(ctx, formats);
544 static int config_output(AVFilterLink *outlink)
546 AVFilterContext *ctx = outlink->src;
547 AudioFIRContext *s = ctx->priv;
549 if (ctx->inputs[0]->channels != ctx->inputs[1]->channels &&
550 ctx->inputs[1]->channels != 1) {
551 av_log(ctx, AV_LOG_ERROR,
552 "Second input must have same number of channels as first input or "
553 "exactly 1 channel.\n");
554 return AVERROR(EINVAL);
557 s->one2many = ctx->inputs[1]->channels == 1;
558 outlink->sample_rate = ctx->inputs[0]->sample_rate;
559 outlink->time_base = ctx->inputs[0]->time_base;
560 outlink->channel_layout = ctx->inputs[0]->channel_layout;
561 outlink->channels = ctx->inputs[0]->channels;
563 s->fifo[0] = av_audio_fifo_alloc(ctx->inputs[0]->format, ctx->inputs[0]->channels, 1024);
564 s->fifo[1] = av_audio_fifo_alloc(ctx->inputs[1]->format, ctx->inputs[1]->channels, 1024);
565 if (!s->fifo[0] || !s->fifo[1])
566 return AVERROR(ENOMEM);
568 s->sum = av_calloc(outlink->channels, sizeof(*s->sum));
569 s->coeff = av_calloc(ctx->inputs[1]->channels, sizeof(*s->coeff));
570 s->block = av_calloc(ctx->inputs[0]->channels, sizeof(*s->block));
571 s->rdft = av_calloc(outlink->channels, sizeof(*s->rdft));
572 s->irdft = av_calloc(outlink->channels, sizeof(*s->irdft));
573 if (!s->sum || !s->coeff || !s->block || !s->rdft || !s->irdft)
574 return AVERROR(ENOMEM);
576 s->nb_channels = outlink->channels;
577 s->nb_coef_channels = ctx->inputs[1]->channels;
580 s->pts = AV_NOPTS_VALUE;
585 static av_cold void uninit(AVFilterContext *ctx)
587 AudioFIRContext *s = ctx->priv;
591 for (ch = 0; ch < s->nb_channels; ch++) {
592 av_freep(&s->sum[ch]);
598 for (ch = 0; ch < s->nb_coef_channels; ch++) {
599 av_freep(&s->coeff[ch]);
605 for (ch = 0; ch < s->nb_channels; ch++) {
606 av_freep(&s->block[ch]);
612 for (ch = 0; ch < s->nb_channels; ch++) {
613 av_rdft_end(s->rdft[ch]);
619 for (ch = 0; ch < s->nb_channels; ch++) {
620 av_rdft_end(s->irdft[ch]);
625 av_frame_free(&s->in[0]);
626 av_frame_free(&s->in[1]);
627 av_frame_free(&s->buffer);
629 av_audio_fifo_free(s->fifo[0]);
630 av_audio_fifo_free(s->fifo[1]);
634 av_freep(&ctx->output_pads[0].name);
636 av_freep(&ctx->output_pads[1].name);
637 av_frame_free(&s->video);
640 static int config_video(AVFilterLink *outlink)
642 AVFilterContext *ctx = outlink->src;
643 AudioFIRContext *s = ctx->priv;
645 outlink->sample_aspect_ratio = (AVRational){1,1};
649 av_frame_free(&s->video);
650 s->video = ff_get_video_buffer(outlink, outlink->w, outlink->h);
652 return AVERROR(ENOMEM);
657 static av_cold int init(AVFilterContext *ctx)
659 AudioFIRContext *s = ctx->priv;
660 AVFilterPad pad, vpad;
663 .name = av_strdup("default"),
664 .type = AVMEDIA_TYPE_AUDIO,
665 .config_props = config_output,
666 .request_frame = request_frame,
670 return AVERROR(ENOMEM);
673 vpad = (AVFilterPad){
674 .name = av_strdup("filter_response"),
675 .type = AVMEDIA_TYPE_VIDEO,
676 .config_props = config_video,
679 return AVERROR(ENOMEM);
682 ff_insert_outpad(ctx, 0, &pad);
685 ff_insert_outpad(ctx, 1, &vpad);
687 s->fcmul_add = fcmul_add_c;
689 s->fdsp = avpriv_float_dsp_alloc(0);
691 return AVERROR(ENOMEM);
699 static const AVFilterPad afir_inputs[] = {
702 .type = AVMEDIA_TYPE_AUDIO,
703 .filter_frame = filter_frame,
706 .type = AVMEDIA_TYPE_AUDIO,
707 .filter_frame = read_ir,
712 #define AF AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
713 #define VF AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
714 #define OFFSET(x) offsetof(AudioFIRContext, x)
716 static const AVOption afir_options[] = {
717 { "dry", "set dry gain", OFFSET(dry_gain), AV_OPT_TYPE_FLOAT, {.dbl=1}, 0, 10, AF },
718 { "wet", "set wet gain", OFFSET(wet_gain), AV_OPT_TYPE_FLOAT, {.dbl=1}, 0, 10, AF },
719 { "length", "set IR length", OFFSET(length), AV_OPT_TYPE_FLOAT, {.dbl=1}, 0, 1, AF },
720 { "again", "enable auto gain", OFFSET(again), AV_OPT_TYPE_BOOL, {.i64=1}, 0, 1, AF },
721 { "maxir", "set max IR length", OFFSET(max_ir_len), AV_OPT_TYPE_FLOAT, {.dbl=30}, 0.1, 60, AF },
722 { "response", "show IR frequency response", OFFSET(response), AV_OPT_TYPE_BOOL, {.i64=0}, 0, 1, VF },
723 { "channel", "set IR channel to display frequency response", OFFSET(ir_channel), AV_OPT_TYPE_INT, {.i64=0}, 0, 1024, VF },
724 { "size", "set video size", OFFSET(w), AV_OPT_TYPE_IMAGE_SIZE, {.str = "hd720"}, 0, 0, VF },
728 AVFILTER_DEFINE_CLASS(afir);
730 AVFilter ff_af_afir = {
732 .description = NULL_IF_CONFIG_SMALL("Apply Finite Impulse Response filter with supplied coefficients in 2nd stream."),
733 .priv_size = sizeof(AudioFIRContext),
734 .priv_class = &afir_class,
735 .query_formats = query_formats,
738 .inputs = afir_inputs,
739 .flags = AVFILTER_FLAG_DYNAMIC_OUTPUTS |
740 AVFILTER_FLAG_SLICE_THREADS,