1 /*
2 * Copyright (c) 2017 Paul B Mahol
3 *
4 * This file is part of FFmpeg.
5 *
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.
10 *
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.
15 *
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
19 */
20
21 /**
22 * @file
23 * An arbitrary audio FIR filter
24 */
25
26 #include <float.h>
27
28 #include "libavutil/avstring.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"
35
36 #include "audio.h"
37 #include "avfilter.h"
38 #include "filters.h"
39 #include "formats.h"
40 #include "internal.h"
41 #include "af_afir.h"
42
fcmul_add_c(float * sum,const float * t,const float * c,ptrdiff_t len)43 static void fcmul_add_c(float *sum, const float *t, const float *c, ptrdiff_t len)
44 {
45 int n;
46
47 for (n = 0; n < len; n++) {
48 const float cre = c[2 * n ];
49 const float cim = c[2 * n + 1];
50 const float tre = t[2 * n ];
51 const float tim = t[2 * n + 1];
52
53 sum[2 * n ] += tre * cre - tim * cim;
54 sum[2 * n + 1] += tre * cim + tim * cre;
55 }
56
57 sum[2 * n] += t[2 * n] * c[2 * n];
58 }
59
direct(const float * in,const FFTComplex * ir,int len,float * out)60 static void direct(const float *in, const FFTComplex *ir, int len, float *out)
61 {
62 for (int n = 0; n < len; n++)
63 for (int m = 0; m <= n; m++)
64 out[n] += ir[m].re * in[n - m];
65 }
66
fir_quantum(AVFilterContext * ctx,AVFrame * out,int ch,int offset)67 static int fir_quantum(AVFilterContext *ctx, AVFrame *out, int ch, int offset)
68 {
69 AudioFIRContext *s = ctx->priv;
70 const float *in = (const float *)s->in->extended_data[ch] + offset;
71 float *block, *buf, *ptr = (float *)out->extended_data[ch] + offset;
72 const int nb_samples = FFMIN(s->min_part_size, out->nb_samples - offset);
73 int n, i, j;
74
75 for (int segment = 0; segment < s->nb_segments; segment++) {
76 AudioFIRSegment *seg = &s->seg[segment];
77 float *src = (float *)seg->input->extended_data[ch];
78 float *dst = (float *)seg->output->extended_data[ch];
79 float *sum = (float *)seg->sum->extended_data[ch];
80
81 if (s->min_part_size >= 8) {
82 s->fdsp->vector_fmul_scalar(src + seg->input_offset, in, s->dry_gain, FFALIGN(nb_samples, 4));
83 emms_c();
84 } else {
85 for (n = 0; n < nb_samples; n++)
86 src[seg->input_offset + n] = in[n] * s->dry_gain;
87 }
88
89 seg->output_offset[ch] += s->min_part_size;
90 if (seg->output_offset[ch] == seg->part_size) {
91 seg->output_offset[ch] = 0;
92 } else {
93 memmove(src, src + s->min_part_size, (seg->input_size - s->min_part_size) * sizeof(*src));
94
95 dst += seg->output_offset[ch];
96 for (n = 0; n < nb_samples; n++) {
97 ptr[n] += dst[n];
98 }
99 continue;
100 }
101
102 if (seg->part_size < 8) {
103 memset(dst, 0, sizeof(*dst) * seg->part_size * seg->nb_partitions);
104
105 j = seg->part_index[ch];
106
107 for (i = 0; i < seg->nb_partitions; i++) {
108 const int coffset = j * seg->coeff_size;
109 const FFTComplex *coeff = (const FFTComplex *)seg->coeff->extended_data[ch * !s->one2many] + coffset;
110
111 direct(src, coeff, nb_samples, dst);
112
113 if (j == 0)
114 j = seg->nb_partitions;
115 j--;
116 }
117
118 seg->part_index[ch] = (seg->part_index[ch] + 1) % seg->nb_partitions;
119
120 memmove(src, src + s->min_part_size, (seg->input_size - s->min_part_size) * sizeof(*src));
121
122 for (n = 0; n < nb_samples; n++) {
123 ptr[n] += dst[n];
124 }
125 continue;
126 }
127
128 memset(sum, 0, sizeof(*sum) * seg->fft_length);
129 block = (float *)seg->block->extended_data[ch] + seg->part_index[ch] * seg->block_size;
130 memset(block + seg->part_size, 0, sizeof(*block) * (seg->fft_length - seg->part_size));
131
132 memcpy(block, src, sizeof(*src) * seg->part_size);
133
134 av_rdft_calc(seg->rdft[ch], block);
135 block[2 * seg->part_size] = block[1];
136 block[1] = 0;
137
138 j = seg->part_index[ch];
139
140 for (i = 0; i < seg->nb_partitions; i++) {
141 const int coffset = j * seg->coeff_size;
142 const float *block = (const float *)seg->block->extended_data[ch] + i * seg->block_size;
143 const FFTComplex *coeff = (const FFTComplex *)seg->coeff->extended_data[ch * !s->one2many] + coffset;
144
145 s->afirdsp.fcmul_add(sum, block, (const float *)coeff, seg->part_size);
146
147 if (j == 0)
148 j = seg->nb_partitions;
149 j--;
150 }
151
152 sum[1] = sum[2 * seg->part_size];
153 av_rdft_calc(seg->irdft[ch], sum);
154
155 buf = (float *)seg->buffer->extended_data[ch];
156 for (n = 0; n < seg->part_size; n++) {
157 buf[n] += sum[n];
158 }
159
160 memcpy(dst, buf, seg->part_size * sizeof(*dst));
161
162 buf = (float *)seg->buffer->extended_data[ch];
163 memcpy(buf, sum + seg->part_size, seg->part_size * sizeof(*buf));
164
165 seg->part_index[ch] = (seg->part_index[ch] + 1) % seg->nb_partitions;
166
167 memmove(src, src + s->min_part_size, (seg->input_size - s->min_part_size) * sizeof(*src));
168
169 for (n = 0; n < nb_samples; n++) {
170 ptr[n] += dst[n];
171 }
172 }
173
174 if (s->min_part_size >= 8) {
175 s->fdsp->vector_fmul_scalar(ptr, ptr, s->wet_gain, FFALIGN(nb_samples, 4));
176 emms_c();
177 } else {
178 for (n = 0; n < nb_samples; n++)
179 ptr[n] *= s->wet_gain;
180 }
181
182 return 0;
183 }
184
fir_channel(AVFilterContext * ctx,AVFrame * out,int ch)185 static int fir_channel(AVFilterContext *ctx, AVFrame *out, int ch)
186 {
187 AudioFIRContext *s = ctx->priv;
188
189 for (int offset = 0; offset < out->nb_samples; offset += s->min_part_size) {
190 fir_quantum(ctx, out, ch, offset);
191 }
192
193 return 0;
194 }
195
fir_channels(AVFilterContext * ctx,void * arg,int jobnr,int nb_jobs)196 static int fir_channels(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
197 {
198 AVFrame *out = arg;
199 const int start = (out->channels * jobnr) / nb_jobs;
200 const int end = (out->channels * (jobnr+1)) / nb_jobs;
201
202 for (int ch = start; ch < end; ch++) {
203 fir_channel(ctx, out, ch);
204 }
205
206 return 0;
207 }
208
fir_frame(AudioFIRContext * s,AVFrame * in,AVFilterLink * outlink)209 static int fir_frame(AudioFIRContext *s, AVFrame *in, AVFilterLink *outlink)
210 {
211 AVFilterContext *ctx = outlink->src;
212 AVFrame *out = NULL;
213
214 out = ff_get_audio_buffer(outlink, in->nb_samples);
215 if (!out) {
216 av_frame_free(&in);
217 return AVERROR(ENOMEM);
218 }
219
220 if (s->pts == AV_NOPTS_VALUE)
221 s->pts = in->pts;
222 s->in = in;
223 ctx->internal->execute(ctx, fir_channels, out, NULL, FFMIN(outlink->channels,
224 ff_filter_get_nb_threads(ctx)));
225
226 out->pts = s->pts;
227 if (s->pts != AV_NOPTS_VALUE)
228 s->pts += av_rescale_q(out->nb_samples, (AVRational){1, outlink->sample_rate}, outlink->time_base);
229
230 av_frame_free(&in);
231 s->in = NULL;
232
233 return ff_filter_frame(outlink, out);
234 }
235
drawtext(AVFrame * pic,int x,int y,const char * txt,uint32_t color)236 static void drawtext(AVFrame *pic, int x, int y, const char *txt, uint32_t color)
237 {
238 const uint8_t *font;
239 int font_height;
240 int i;
241
242 font = avpriv_cga_font, font_height = 8;
243
244 for (i = 0; txt[i]; i++) {
245 int char_y, mask;
246
247 uint8_t *p = pic->data[0] + y * pic->linesize[0] + (x + i * 8) * 4;
248 for (char_y = 0; char_y < font_height; char_y++) {
249 for (mask = 0x80; mask; mask >>= 1) {
250 if (font[txt[i] * font_height + char_y] & mask)
251 AV_WL32(p, color);
252 p += 4;
253 }
254 p += pic->linesize[0] - 8 * 4;
255 }
256 }
257 }
258
draw_line(AVFrame * out,int x0,int y0,int x1,int y1,uint32_t color)259 static void draw_line(AVFrame *out, int x0, int y0, int x1, int y1, uint32_t color)
260 {
261 int dx = FFABS(x1-x0);
262 int dy = FFABS(y1-y0), sy = y0 < y1 ? 1 : -1;
263 int err = (dx>dy ? dx : -dy) / 2, e2;
264
265 for (;;) {
266 AV_WL32(out->data[0] + y0 * out->linesize[0] + x0 * 4, color);
267
268 if (x0 == x1 && y0 == y1)
269 break;
270
271 e2 = err;
272
273 if (e2 >-dx) {
274 err -= dy;
275 x0--;
276 }
277
278 if (e2 < dy) {
279 err += dx;
280 y0 += sy;
281 }
282 }
283 }
284
draw_response(AVFilterContext * ctx,AVFrame * out)285 static void draw_response(AVFilterContext *ctx, AVFrame *out)
286 {
287 AudioFIRContext *s = ctx->priv;
288 float *mag, *phase, *delay, min = FLT_MAX, max = FLT_MIN;
289 float min_delay = FLT_MAX, max_delay = FLT_MIN;
290 int prev_ymag = -1, prev_yphase = -1, prev_ydelay = -1;
291 char text[32];
292 int channel, i, x;
293
294 memset(out->data[0], 0, s->h * out->linesize[0]);
295
296 phase = av_malloc_array(s->w, sizeof(*phase));
297 mag = av_malloc_array(s->w, sizeof(*mag));
298 delay = av_malloc_array(s->w, sizeof(*delay));
299 if (!mag || !phase || !delay)
300 goto end;
301
302 channel = av_clip(s->ir_channel, 0, s->ir[s->selir]->channels - 1);
303 for (i = 0; i < s->w; i++) {
304 const float *src = (const float *)s->ir[s->selir]->extended_data[channel];
305 double w = i * M_PI / (s->w - 1);
306 double div, real_num = 0., imag_num = 0., real = 0., imag = 0.;
307
308 for (x = 0; x < s->nb_taps; x++) {
309 real += cos(-x * w) * src[x];
310 imag += sin(-x * w) * src[x];
311 real_num += cos(-x * w) * src[x] * x;
312 imag_num += sin(-x * w) * src[x] * x;
313 }
314
315 mag[i] = hypot(real, imag);
316 phase[i] = atan2(imag, real);
317 div = real * real + imag * imag;
318 delay[i] = (real_num * real + imag_num * imag) / div;
319 min = fminf(min, mag[i]);
320 max = fmaxf(max, mag[i]);
321 min_delay = fminf(min_delay, delay[i]);
322 max_delay = fmaxf(max_delay, delay[i]);
323 }
324
325 for (i = 0; i < s->w; i++) {
326 int ymag = mag[i] / max * (s->h - 1);
327 int ydelay = (delay[i] - min_delay) / (max_delay - min_delay) * (s->h - 1);
328 int yphase = (0.5 * (1. + phase[i] / M_PI)) * (s->h - 1);
329
330 ymag = s->h - 1 - av_clip(ymag, 0, s->h - 1);
331 yphase = s->h - 1 - av_clip(yphase, 0, s->h - 1);
332 ydelay = s->h - 1 - av_clip(ydelay, 0, s->h - 1);
333
334 if (prev_ymag < 0)
335 prev_ymag = ymag;
336 if (prev_yphase < 0)
337 prev_yphase = yphase;
338 if (prev_ydelay < 0)
339 prev_ydelay = ydelay;
340
341 draw_line(out, i, ymag, FFMAX(i - 1, 0), prev_ymag, 0xFFFF00FF);
342 draw_line(out, i, yphase, FFMAX(i - 1, 0), prev_yphase, 0xFF00FF00);
343 draw_line(out, i, ydelay, FFMAX(i - 1, 0), prev_ydelay, 0xFF00FFFF);
344
345 prev_ymag = ymag;
346 prev_yphase = yphase;
347 prev_ydelay = ydelay;
348 }
349
350 if (s->w > 400 && s->h > 100) {
351 drawtext(out, 2, 2, "Max Magnitude:", 0xDDDDDDDD);
352 snprintf(text, sizeof(text), "%.2f", max);
353 drawtext(out, 15 * 8 + 2, 2, text, 0xDDDDDDDD);
354
355 drawtext(out, 2, 12, "Min Magnitude:", 0xDDDDDDDD);
356 snprintf(text, sizeof(text), "%.2f", min);
357 drawtext(out, 15 * 8 + 2, 12, text, 0xDDDDDDDD);
358
359 drawtext(out, 2, 22, "Max Delay:", 0xDDDDDDDD);
360 snprintf(text, sizeof(text), "%.2f", max_delay);
361 drawtext(out, 11 * 8 + 2, 22, text, 0xDDDDDDDD);
362
363 drawtext(out, 2, 32, "Min Delay:", 0xDDDDDDDD);
364 snprintf(text, sizeof(text), "%.2f", min_delay);
365 drawtext(out, 11 * 8 + 2, 32, text, 0xDDDDDDDD);
366 }
367
368 end:
369 av_free(delay);
370 av_free(phase);
371 av_free(mag);
372 }
373
init_segment(AVFilterContext * ctx,AudioFIRSegment * seg,int offset,int nb_partitions,int part_size)374 static int init_segment(AVFilterContext *ctx, AudioFIRSegment *seg,
375 int offset, int nb_partitions, int part_size)
376 {
377 AudioFIRContext *s = ctx->priv;
378
379 seg->rdft = av_calloc(ctx->inputs[0]->channels, sizeof(*seg->rdft));
380 seg->irdft = av_calloc(ctx->inputs[0]->channels, sizeof(*seg->irdft));
381 if (!seg->rdft || !seg->irdft)
382 return AVERROR(ENOMEM);
383
384 seg->fft_length = part_size * 2 + 1;
385 seg->part_size = part_size;
386 seg->block_size = FFALIGN(seg->fft_length, 32);
387 seg->coeff_size = FFALIGN(seg->part_size + 1, 32);
388 seg->nb_partitions = nb_partitions;
389 seg->input_size = offset + s->min_part_size;
390 seg->input_offset = offset;
391
392 seg->part_index = av_calloc(ctx->inputs[0]->channels, sizeof(*seg->part_index));
393 seg->output_offset = av_calloc(ctx->inputs[0]->channels, sizeof(*seg->output_offset));
394 if (!seg->part_index || !seg->output_offset)
395 return AVERROR(ENOMEM);
396
397 for (int ch = 0; ch < ctx->inputs[0]->channels && part_size >= 8; ch++) {
398 seg->rdft[ch] = av_rdft_init(av_log2(2 * part_size), DFT_R2C);
399 seg->irdft[ch] = av_rdft_init(av_log2(2 * part_size), IDFT_C2R);
400 if (!seg->rdft[ch] || !seg->irdft[ch])
401 return AVERROR(ENOMEM);
402 }
403
404 seg->sum = ff_get_audio_buffer(ctx->inputs[0], seg->fft_length);
405 seg->block = ff_get_audio_buffer(ctx->inputs[0], seg->nb_partitions * seg->block_size);
406 seg->buffer = ff_get_audio_buffer(ctx->inputs[0], seg->part_size);
407 seg->coeff = ff_get_audio_buffer(ctx->inputs[1 + s->selir], seg->nb_partitions * seg->coeff_size * 2);
408 seg->input = ff_get_audio_buffer(ctx->inputs[0], seg->input_size);
409 seg->output = ff_get_audio_buffer(ctx->inputs[0], seg->part_size);
410 if (!seg->buffer || !seg->sum || !seg->block || !seg->coeff || !seg->input || !seg->output)
411 return AVERROR(ENOMEM);
412
413 return 0;
414 }
415
uninit_segment(AVFilterContext * ctx,AudioFIRSegment * seg)416 static void uninit_segment(AVFilterContext *ctx, AudioFIRSegment *seg)
417 {
418 AudioFIRContext *s = ctx->priv;
419
420 if (seg->rdft) {
421 for (int ch = 0; ch < s->nb_channels; ch++) {
422 av_rdft_end(seg->rdft[ch]);
423 }
424 }
425 av_freep(&seg->rdft);
426
427 if (seg->irdft) {
428 for (int ch = 0; ch < s->nb_channels; ch++) {
429 av_rdft_end(seg->irdft[ch]);
430 }
431 }
432 av_freep(&seg->irdft);
433
434 av_freep(&seg->output_offset);
435 av_freep(&seg->part_index);
436
437 av_frame_free(&seg->block);
438 av_frame_free(&seg->sum);
439 av_frame_free(&seg->buffer);
440 av_frame_free(&seg->coeff);
441 av_frame_free(&seg->input);
442 av_frame_free(&seg->output);
443 seg->input_size = 0;
444 }
445
convert_coeffs(AVFilterContext * ctx)446 static int convert_coeffs(AVFilterContext *ctx)
447 {
448 AudioFIRContext *s = ctx->priv;
449 int ret, i, ch, n, cur_nb_taps;
450 float power = 0;
451
452 if (!s->nb_taps) {
453 int part_size, max_part_size;
454 int left, offset = 0;
455
456 s->nb_taps = ff_inlink_queued_samples(ctx->inputs[1 + s->selir]);
457 if (s->nb_taps <= 0)
458 return AVERROR(EINVAL);
459
460 if (s->minp > s->maxp) {
461 s->maxp = s->minp;
462 }
463
464 left = s->nb_taps;
465 part_size = 1 << av_log2(s->minp);
466 max_part_size = 1 << av_log2(s->maxp);
467
468 s->min_part_size = part_size;
469
470 for (i = 0; left > 0; i++) {
471 int step = part_size == max_part_size ? INT_MAX : 1 + (i == 0);
472 int nb_partitions = FFMIN(step, (left + part_size - 1) / part_size);
473
474 s->nb_segments = i + 1;
475 ret = init_segment(ctx, &s->seg[i], offset, nb_partitions, part_size);
476 if (ret < 0)
477 return ret;
478 offset += nb_partitions * part_size;
479 left -= nb_partitions * part_size;
480 part_size *= 2;
481 part_size = FFMIN(part_size, max_part_size);
482 }
483 }
484
485 if (!s->ir[s->selir]) {
486 ret = ff_inlink_consume_samples(ctx->inputs[1 + s->selir], s->nb_taps, s->nb_taps, &s->ir[s->selir]);
487 if (ret < 0)
488 return ret;
489 if (ret == 0)
490 return AVERROR_BUG;
491 }
492
493 if (s->response)
494 draw_response(ctx, s->video);
495
496 s->gain = 1;
497 cur_nb_taps = s->ir[s->selir]->nb_samples;
498
499 switch (s->gtype) {
500 case -1:
501 /* nothing to do */
502 break;
503 case 0:
504 for (ch = 0; ch < ctx->inputs[1 + s->selir]->channels; ch++) {
505 float *time = (float *)s->ir[s->selir]->extended_data[!s->one2many * ch];
506
507 for (i = 0; i < cur_nb_taps; i++)
508 power += FFABS(time[i]);
509 }
510 s->gain = ctx->inputs[1 + s->selir]->channels / power;
511 break;
512 case 1:
513 for (ch = 0; ch < ctx->inputs[1 + s->selir]->channels; ch++) {
514 float *time = (float *)s->ir[s->selir]->extended_data[!s->one2many * ch];
515
516 for (i = 0; i < cur_nb_taps; i++)
517 power += time[i];
518 }
519 s->gain = ctx->inputs[1 + s->selir]->channels / power;
520 break;
521 case 2:
522 for (ch = 0; ch < ctx->inputs[1 + s->selir]->channels; ch++) {
523 float *time = (float *)s->ir[s->selir]->extended_data[!s->one2many * ch];
524
525 for (i = 0; i < cur_nb_taps; i++)
526 power += time[i] * time[i];
527 }
528 s->gain = sqrtf(ch / power);
529 break;
530 default:
531 return AVERROR_BUG;
532 }
533
534 s->gain = FFMIN(s->gain * s->ir_gain, 1.f);
535 av_log(ctx, AV_LOG_DEBUG, "power %f, gain %f\n", power, s->gain);
536 for (ch = 0; ch < ctx->inputs[1 + s->selir]->channels; ch++) {
537 float *time = (float *)s->ir[s->selir]->extended_data[!s->one2many * ch];
538
539 s->fdsp->vector_fmul_scalar(time, time, s->gain, FFALIGN(cur_nb_taps, 4));
540 }
541
542 av_log(ctx, AV_LOG_DEBUG, "nb_taps: %d\n", cur_nb_taps);
543 av_log(ctx, AV_LOG_DEBUG, "nb_segments: %d\n", s->nb_segments);
544
545 for (ch = 0; ch < ctx->inputs[1 + s->selir]->channels; ch++) {
546 float *time = (float *)s->ir[s->selir]->extended_data[!s->one2many * ch];
547 int toffset = 0;
548
549 for (i = FFMAX(1, s->length * s->nb_taps); i < s->nb_taps; i++)
550 time[i] = 0;
551
552 av_log(ctx, AV_LOG_DEBUG, "channel: %d\n", ch);
553
554 for (int segment = 0; segment < s->nb_segments; segment++) {
555 AudioFIRSegment *seg = &s->seg[segment];
556 float *block = (float *)seg->block->extended_data[ch];
557 FFTComplex *coeff = (FFTComplex *)seg->coeff->extended_data[ch];
558
559 av_log(ctx, AV_LOG_DEBUG, "segment: %d\n", segment);
560
561 for (i = 0; i < seg->nb_partitions; i++) {
562 const float scale = 1.f / seg->part_size;
563 const int coffset = i * seg->coeff_size;
564 const int remaining = s->nb_taps - toffset;
565 const int size = remaining >= seg->part_size ? seg->part_size : remaining;
566
567 if (size < 8) {
568 for (n = 0; n < size; n++)
569 coeff[coffset + n].re = time[toffset + n];
570
571 toffset += size;
572 continue;
573 }
574
575 memset(block, 0, sizeof(*block) * seg->fft_length);
576 memcpy(block, time + toffset, size * sizeof(*block));
577
578 av_rdft_calc(seg->rdft[0], block);
579
580 coeff[coffset].re = block[0] * scale;
581 coeff[coffset].im = 0;
582 for (n = 1; n < seg->part_size; n++) {
583 coeff[coffset + n].re = block[2 * n] * scale;
584 coeff[coffset + n].im = block[2 * n + 1] * scale;
585 }
586 coeff[coffset + seg->part_size].re = block[1] * scale;
587 coeff[coffset + seg->part_size].im = 0;
588
589 toffset += size;
590 }
591
592 av_log(ctx, AV_LOG_DEBUG, "nb_partitions: %d\n", seg->nb_partitions);
593 av_log(ctx, AV_LOG_DEBUG, "partition size: %d\n", seg->part_size);
594 av_log(ctx, AV_LOG_DEBUG, "block size: %d\n", seg->block_size);
595 av_log(ctx, AV_LOG_DEBUG, "fft_length: %d\n", seg->fft_length);
596 av_log(ctx, AV_LOG_DEBUG, "coeff_size: %d\n", seg->coeff_size);
597 av_log(ctx, AV_LOG_DEBUG, "input_size: %d\n", seg->input_size);
598 av_log(ctx, AV_LOG_DEBUG, "input_offset: %d\n", seg->input_offset);
599 }
600 }
601
602 s->have_coeffs = 1;
603
604 return 0;
605 }
606
check_ir(AVFilterLink * link,AVFrame * frame)607 static int check_ir(AVFilterLink *link, AVFrame *frame)
608 {
609 AVFilterContext *ctx = link->dst;
610 AudioFIRContext *s = ctx->priv;
611 int nb_taps, max_nb_taps;
612
613 nb_taps = ff_inlink_queued_samples(link);
614 max_nb_taps = s->max_ir_len * ctx->outputs[0]->sample_rate;
615 if (nb_taps > max_nb_taps) {
616 av_log(ctx, AV_LOG_ERROR, "Too big number of coefficients: %d > %d.\n", nb_taps, max_nb_taps);
617 return AVERROR(EINVAL);
618 }
619
620 return 0;
621 }
622
activate(AVFilterContext * ctx)623 static int activate(AVFilterContext *ctx)
624 {
625 AudioFIRContext *s = ctx->priv;
626 AVFilterLink *outlink = ctx->outputs[0];
627 int ret, status, available, wanted;
628 AVFrame *in = NULL;
629 int64_t pts;
630
631 FF_FILTER_FORWARD_STATUS_BACK_ALL(ctx->outputs[0], ctx);
632 if (s->response)
633 FF_FILTER_FORWARD_STATUS_BACK_ALL(ctx->outputs[1], ctx);
634 if (!s->eof_coeffs[s->selir]) {
635 AVFrame *ir = NULL;
636
637 ret = check_ir(ctx->inputs[1 + s->selir], ir);
638 if (ret < 0)
639 return ret;
640
641 if (ff_outlink_get_status(ctx->inputs[1 + s->selir]) == AVERROR_EOF)
642 s->eof_coeffs[s->selir] = 1;
643
644 if (!s->eof_coeffs[s->selir]) {
645 if (ff_outlink_frame_wanted(ctx->outputs[0]))
646 ff_inlink_request_frame(ctx->inputs[1 + s->selir]);
647 else if (s->response && ff_outlink_frame_wanted(ctx->outputs[1]))
648 ff_inlink_request_frame(ctx->inputs[1 + s->selir]);
649 return 0;
650 }
651 }
652
653 if (!s->have_coeffs && s->eof_coeffs[s->selir]) {
654 ret = convert_coeffs(ctx);
655 if (ret < 0)
656 return ret;
657 }
658
659 available = ff_inlink_queued_samples(ctx->inputs[0]);
660 wanted = FFMAX(s->min_part_size, (available / s->min_part_size) * s->min_part_size);
661 ret = ff_inlink_consume_samples(ctx->inputs[0], wanted, wanted, &in);
662 if (ret > 0)
663 ret = fir_frame(s, in, outlink);
664
665 if (ret < 0)
666 return ret;
667
668 if (s->response && s->have_coeffs) {
669 int64_t old_pts = s->video->pts;
670 int64_t new_pts = av_rescale_q(s->pts, ctx->inputs[0]->time_base, ctx->outputs[1]->time_base);
671
672 if (ff_outlink_frame_wanted(ctx->outputs[1]) && old_pts < new_pts) {
673 AVFrame *clone;
674 s->video->pts = new_pts;
675 clone = av_frame_clone(s->video);
676 if (!clone)
677 return AVERROR(ENOMEM);
678 return ff_filter_frame(ctx->outputs[1], clone);
679 }
680 }
681
682 if (ff_inlink_queued_samples(ctx->inputs[0]) >= s->min_part_size) {
683 ff_filter_set_ready(ctx, 10);
684 return 0;
685 }
686
687 if (ff_inlink_acknowledge_status(ctx->inputs[0], &status, &pts)) {
688 if (status == AVERROR_EOF) {
689 ff_outlink_set_status(ctx->outputs[0], status, pts);
690 if (s->response)
691 ff_outlink_set_status(ctx->outputs[1], status, pts);
692 return 0;
693 }
694 }
695
696 if (ff_outlink_frame_wanted(ctx->outputs[0]) &&
697 !ff_outlink_get_status(ctx->inputs[0])) {
698 ff_inlink_request_frame(ctx->inputs[0]);
699 return 0;
700 }
701
702 if (s->response &&
703 ff_outlink_frame_wanted(ctx->outputs[1]) &&
704 !ff_outlink_get_status(ctx->inputs[0])) {
705 ff_inlink_request_frame(ctx->inputs[0]);
706 return 0;
707 }
708
709 return FFERROR_NOT_READY;
710 }
711
query_formats(AVFilterContext * ctx)712 static int query_formats(AVFilterContext *ctx)
713 {
714 AudioFIRContext *s = ctx->priv;
715 AVFilterFormats *formats;
716 AVFilterChannelLayouts *layouts;
717 static const enum AVSampleFormat sample_fmts[] = {
718 AV_SAMPLE_FMT_FLTP,
719 AV_SAMPLE_FMT_NONE
720 };
721 static const enum AVPixelFormat pix_fmts[] = {
722 AV_PIX_FMT_RGB0,
723 AV_PIX_FMT_NONE
724 };
725 int ret;
726
727 if (s->response) {
728 AVFilterLink *videolink = ctx->outputs[1];
729 formats = ff_make_format_list(pix_fmts);
730 if ((ret = ff_formats_ref(formats, &videolink->in_formats)) < 0)
731 return ret;
732 }
733
734 layouts = ff_all_channel_counts();
735 if (!layouts)
736 return AVERROR(ENOMEM);
737
738 if (s->ir_format) {
739 ret = ff_set_common_channel_layouts(ctx, layouts);
740 if (ret < 0)
741 return ret;
742 } else {
743 AVFilterChannelLayouts *mono = NULL;
744
745 ret = ff_add_channel_layout(&mono, AV_CH_LAYOUT_MONO);
746 if (ret)
747 return ret;
748
749 if ((ret = ff_channel_layouts_ref(layouts, &ctx->inputs[0]->out_channel_layouts)) < 0)
750 return ret;
751 if ((ret = ff_channel_layouts_ref(layouts, &ctx->outputs[0]->in_channel_layouts)) < 0)
752 return ret;
753 for (int i = 1; i < ctx->nb_inputs; i++) {
754 if ((ret = ff_channel_layouts_ref(mono, &ctx->inputs[i]->out_channel_layouts)) < 0)
755 return ret;
756 }
757 }
758
759 formats = ff_make_format_list(sample_fmts);
760 if ((ret = ff_set_common_formats(ctx, formats)) < 0)
761 return ret;
762
763 formats = ff_all_samplerates();
764 return ff_set_common_samplerates(ctx, formats);
765 }
766
config_output(AVFilterLink * outlink)767 static int config_output(AVFilterLink *outlink)
768 {
769 AVFilterContext *ctx = outlink->src;
770 AudioFIRContext *s = ctx->priv;
771
772 s->one2many = ctx->inputs[1 + s->selir]->channels == 1;
773 outlink->sample_rate = ctx->inputs[0]->sample_rate;
774 outlink->time_base = ctx->inputs[0]->time_base;
775 outlink->channel_layout = ctx->inputs[0]->channel_layout;
776 outlink->channels = ctx->inputs[0]->channels;
777
778 s->nb_channels = outlink->channels;
779 s->nb_coef_channels = ctx->inputs[1 + s->selir]->channels;
780 s->pts = AV_NOPTS_VALUE;
781
782 return 0;
783 }
784
uninit(AVFilterContext * ctx)785 static av_cold void uninit(AVFilterContext *ctx)
786 {
787 AudioFIRContext *s = ctx->priv;
788
789 for (int i = 0; i < s->nb_segments; i++) {
790 uninit_segment(ctx, &s->seg[i]);
791 }
792
793 av_freep(&s->fdsp);
794
795 for (int i = 0; i < s->nb_irs; i++) {
796 av_frame_free(&s->ir[i]);
797 }
798
799 for (int i = 0; i < ctx->nb_inputs; i++)
800 av_freep(&ctx->input_pads[i].name);
801
802 for (int i = 0; i < ctx->nb_outputs; i++)
803 av_freep(&ctx->output_pads[i].name);
804 av_frame_free(&s->video);
805 }
806
config_video(AVFilterLink * outlink)807 static int config_video(AVFilterLink *outlink)
808 {
809 AVFilterContext *ctx = outlink->src;
810 AudioFIRContext *s = ctx->priv;
811
812 outlink->sample_aspect_ratio = (AVRational){1,1};
813 outlink->w = s->w;
814 outlink->h = s->h;
815 outlink->frame_rate = s->frame_rate;
816 outlink->time_base = av_inv_q(outlink->frame_rate);
817
818 av_frame_free(&s->video);
819 s->video = ff_get_video_buffer(outlink, outlink->w, outlink->h);
820 if (!s->video)
821 return AVERROR(ENOMEM);
822
823 return 0;
824 }
825
ff_afir_init(AudioFIRDSPContext * dsp)826 void ff_afir_init(AudioFIRDSPContext *dsp)
827 {
828 dsp->fcmul_add = fcmul_add_c;
829
830 if (ARCH_X86)
831 ff_afir_init_x86(dsp);
832 }
833
init(AVFilterContext * ctx)834 static av_cold int init(AVFilterContext *ctx)
835 {
836 AudioFIRContext *s = ctx->priv;
837 AVFilterPad pad, vpad;
838 int ret;
839
840 pad = (AVFilterPad) {
841 .name = av_strdup("main"),
842 .type = AVMEDIA_TYPE_AUDIO,
843 };
844
845 if (!pad.name)
846 return AVERROR(ENOMEM);
847
848 ret = ff_insert_inpad(ctx, 0, &pad);
849 if (ret < 0) {
850 av_freep(&pad.name);
851 return ret;
852 }
853
854 for (int n = 0; n < s->nb_irs; n++) {
855 pad = (AVFilterPad) {
856 .name = av_asprintf("ir%d", n),
857 .type = AVMEDIA_TYPE_AUDIO,
858 };
859
860 if (!pad.name)
861 return AVERROR(ENOMEM);
862
863 ret = ff_insert_inpad(ctx, n + 1, &pad);
864 if (ret < 0) {
865 av_freep(&pad.name);
866 return ret;
867 }
868 }
869
870 pad = (AVFilterPad) {
871 .name = av_strdup("default"),
872 .type = AVMEDIA_TYPE_AUDIO,
873 .config_props = config_output,
874 };
875
876 if (!pad.name)
877 return AVERROR(ENOMEM);
878
879 ret = ff_insert_outpad(ctx, 0, &pad);
880 if (ret < 0) {
881 av_freep(&pad.name);
882 return ret;
883 }
884
885 if (s->response) {
886 vpad = (AVFilterPad){
887 .name = av_strdup("filter_response"),
888 .type = AVMEDIA_TYPE_VIDEO,
889 .config_props = config_video,
890 };
891 if (!vpad.name)
892 return AVERROR(ENOMEM);
893
894 ret = ff_insert_outpad(ctx, 1, &vpad);
895 if (ret < 0) {
896 av_freep(&vpad.name);
897 return ret;
898 }
899 }
900
901 s->fdsp = avpriv_float_dsp_alloc(0);
902 if (!s->fdsp)
903 return AVERROR(ENOMEM);
904
905 ff_afir_init(&s->afirdsp);
906
907 return 0;
908 }
909
process_command(AVFilterContext * ctx,const char * cmd,const char * arg,char * res,int res_len,int flags)910 static int process_command(AVFilterContext *ctx,
911 const char *cmd,
912 const char *arg,
913 char *res,
914 int res_len,
915 int flags)
916 {
917 AudioFIRContext *s = ctx->priv;
918 int prev_ir = s->selir;
919 int ret = ff_filter_process_command(ctx, cmd, arg, res, res_len, flags);
920
921 if (ret < 0)
922 return ret;
923
924 s->selir = FFMIN(s->nb_irs - 1, s->selir);
925
926 if (prev_ir != s->selir) {
927 s->have_coeffs = 0;
928 }
929
930 return 0;
931 }
932
933 #define AF AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
934 #define AFR AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_RUNTIME_PARAM
935 #define VF AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
936 #define OFFSET(x) offsetof(AudioFIRContext, x)
937
938 static const AVOption afir_options[] = {
939 { "dry", "set dry gain", OFFSET(dry_gain), AV_OPT_TYPE_FLOAT, {.dbl=1}, 0, 10, AF },
940 { "wet", "set wet gain", OFFSET(wet_gain), AV_OPT_TYPE_FLOAT, {.dbl=1}, 0, 10, AF },
941 { "length", "set IR length", OFFSET(length), AV_OPT_TYPE_FLOAT, {.dbl=1}, 0, 1, AF },
942 { "gtype", "set IR auto gain type",OFFSET(gtype), AV_OPT_TYPE_INT, {.i64=0}, -1, 2, AF, "gtype" },
943 { "none", "without auto gain", 0, AV_OPT_TYPE_CONST, {.i64=-1}, 0, 0, AF, "gtype" },
944 { "peak", "peak gain", 0, AV_OPT_TYPE_CONST, {.i64=0}, 0, 0, AF, "gtype" },
945 { "dc", "DC gain", 0, AV_OPT_TYPE_CONST, {.i64=1}, 0, 0, AF, "gtype" },
946 { "gn", "gain to noise", 0, AV_OPT_TYPE_CONST, {.i64=2}, 0, 0, AF, "gtype" },
947 { "irgain", "set IR gain", OFFSET(ir_gain), AV_OPT_TYPE_FLOAT, {.dbl=1}, 0, 1, AF },
948 { "irfmt", "set IR format", OFFSET(ir_format), AV_OPT_TYPE_INT, {.i64=1}, 0, 1, AF, "irfmt" },
949 { "mono", "single channel", 0, AV_OPT_TYPE_CONST, {.i64=0}, 0, 0, AF, "irfmt" },
950 { "input", "same as input", 0, AV_OPT_TYPE_CONST, {.i64=1}, 0, 0, AF, "irfmt" },
951 { "maxir", "set max IR length", OFFSET(max_ir_len), AV_OPT_TYPE_FLOAT, {.dbl=30}, 0.1, 60, AF },
952 { "response", "show IR frequency response", OFFSET(response), AV_OPT_TYPE_BOOL, {.i64=0}, 0, 1, VF },
953 { "channel", "set IR channel to display frequency response", OFFSET(ir_channel), AV_OPT_TYPE_INT, {.i64=0}, 0, 1024, VF },
954 { "size", "set video size", OFFSET(w), AV_OPT_TYPE_IMAGE_SIZE, {.str = "hd720"}, 0, 0, VF },
955 { "rate", "set video rate", OFFSET(frame_rate), AV_OPT_TYPE_VIDEO_RATE, {.str = "25"}, 0, INT32_MAX, VF },
956 { "minp", "set min partition size", OFFSET(minp), AV_OPT_TYPE_INT, {.i64=8192}, 1, 32768, AF },
957 { "maxp", "set max partition size", OFFSET(maxp), AV_OPT_TYPE_INT, {.i64=8192}, 8, 32768, AF },
958 { "nbirs", "set number of input IRs",OFFSET(nb_irs),AV_OPT_TYPE_INT, {.i64=1}, 1, 32, AF },
959 { "ir", "select IR", OFFSET(selir), AV_OPT_TYPE_INT, {.i64=0}, 0, 31, AFR },
960 { NULL }
961 };
962
963 AVFILTER_DEFINE_CLASS(afir);
964
965 AVFilter ff_af_afir = {
966 .name = "afir",
967 .description = NULL_IF_CONFIG_SMALL("Apply Finite Impulse Response filter with supplied coefficients in additional stream(s)."),
968 .priv_size = sizeof(AudioFIRContext),
969 .priv_class = &afir_class,
970 .query_formats = query_formats,
971 .init = init,
972 .activate = activate,
973 .uninit = uninit,
974 .process_command = process_command,
975 .flags = AVFILTER_FLAG_DYNAMIC_INPUTS |
976 AVFILTER_FLAG_DYNAMIC_OUTPUTS |
977 AVFILTER_FLAG_SLICE_THREADS,
978 };
979