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_fadd(AudioFIRContext * s,float * dst,const float * src,int nb_samples)67 static void fir_fadd(AudioFIRContext *s, float *dst, const float *src, int nb_samples)
68 {
69 if ((nb_samples & 15) == 0 && nb_samples >= 16) {
70 s->fdsp->vector_fmac_scalar(dst, src, 1.f, nb_samples);
71 } else {
72 for (int n = 0; n < nb_samples; n++)
73 dst[n] += src[n];
74 }
75 }
76
fir_quantum(AVFilterContext * ctx,AVFrame * out,int ch,int offset)77 static int fir_quantum(AVFilterContext *ctx, AVFrame *out, int ch, int offset)
78 {
79 AudioFIRContext *s = ctx->priv;
80 const float *in = (const float *)s->in->extended_data[ch] + offset;
81 float *block, *buf, *ptr = (float *)out->extended_data[ch] + offset;
82 const int nb_samples = FFMIN(s->min_part_size, out->nb_samples - offset);
83 int n, i, j;
84
85 for (int segment = 0; segment < s->nb_segments; segment++) {
86 AudioFIRSegment *seg = &s->seg[segment];
87 float *src = (float *)seg->input->extended_data[ch];
88 float *dst = (float *)seg->output->extended_data[ch];
89 float *sum = (float *)seg->sum->extended_data[ch];
90
91 if (s->min_part_size >= 8) {
92 s->fdsp->vector_fmul_scalar(src + seg->input_offset, in, s->dry_gain, FFALIGN(nb_samples, 4));
93 emms_c();
94 } else {
95 for (n = 0; n < nb_samples; n++)
96 src[seg->input_offset + n] = in[n] * s->dry_gain;
97 }
98
99 seg->output_offset[ch] += s->min_part_size;
100 if (seg->output_offset[ch] == seg->part_size) {
101 seg->output_offset[ch] = 0;
102 } else {
103 memmove(src, src + s->min_part_size, (seg->input_size - s->min_part_size) * sizeof(*src));
104
105 dst += seg->output_offset[ch];
106 fir_fadd(s, ptr, dst, nb_samples);
107 continue;
108 }
109
110 if (seg->part_size < 8) {
111 memset(dst, 0, sizeof(*dst) * seg->part_size * seg->nb_partitions);
112
113 j = seg->part_index[ch];
114
115 for (i = 0; i < seg->nb_partitions; i++) {
116 const int coffset = j * seg->coeff_size;
117 const FFTComplex *coeff = (const FFTComplex *)seg->coeff->extended_data[ch * !s->one2many] + coffset;
118
119 direct(src, coeff, nb_samples, dst);
120
121 if (j == 0)
122 j = seg->nb_partitions;
123 j--;
124 }
125
126 seg->part_index[ch] = (seg->part_index[ch] + 1) % seg->nb_partitions;
127
128 memmove(src, src + s->min_part_size, (seg->input_size - s->min_part_size) * sizeof(*src));
129
130 for (n = 0; n < nb_samples; n++) {
131 ptr[n] += dst[n];
132 }
133 continue;
134 }
135
136 memset(sum, 0, sizeof(*sum) * seg->fft_length);
137 block = (float *)seg->block->extended_data[ch] + seg->part_index[ch] * seg->block_size;
138 memset(block + seg->part_size, 0, sizeof(*block) * (seg->fft_length - seg->part_size));
139
140 memcpy(block, src, sizeof(*src) * seg->part_size);
141
142 av_rdft_calc(seg->rdft[ch], block);
143 block[2 * seg->part_size] = block[1];
144 block[1] = 0;
145
146 j = seg->part_index[ch];
147
148 for (i = 0; i < seg->nb_partitions; i++) {
149 const int coffset = j * seg->coeff_size;
150 const float *block = (const float *)seg->block->extended_data[ch] + i * seg->block_size;
151 const FFTComplex *coeff = (const FFTComplex *)seg->coeff->extended_data[ch * !s->one2many] + coffset;
152
153 s->afirdsp.fcmul_add(sum, block, (const float *)coeff, seg->part_size);
154
155 if (j == 0)
156 j = seg->nb_partitions;
157 j--;
158 }
159
160 sum[1] = sum[2 * seg->part_size];
161 av_rdft_calc(seg->irdft[ch], sum);
162
163 buf = (float *)seg->buffer->extended_data[ch];
164 fir_fadd(s, buf, sum, seg->part_size);
165
166 memcpy(dst, buf, seg->part_size * sizeof(*dst));
167
168 buf = (float *)seg->buffer->extended_data[ch];
169 memcpy(buf, sum + seg->part_size, seg->part_size * sizeof(*buf));
170
171 seg->part_index[ch] = (seg->part_index[ch] + 1) % seg->nb_partitions;
172
173 memmove(src, src + s->min_part_size, (seg->input_size - s->min_part_size) * sizeof(*src));
174
175 fir_fadd(s, ptr, dst, nb_samples);
176 }
177
178 if (s->min_part_size >= 8) {
179 s->fdsp->vector_fmul_scalar(ptr, ptr, s->wet_gain, FFALIGN(nb_samples, 4));
180 emms_c();
181 } else {
182 for (n = 0; n < nb_samples; n++)
183 ptr[n] *= s->wet_gain;
184 }
185
186 return 0;
187 }
188
fir_channel(AVFilterContext * ctx,AVFrame * out,int ch)189 static int fir_channel(AVFilterContext *ctx, AVFrame *out, int ch)
190 {
191 AudioFIRContext *s = ctx->priv;
192
193 for (int offset = 0; offset < out->nb_samples; offset += s->min_part_size) {
194 fir_quantum(ctx, out, ch, offset);
195 }
196
197 return 0;
198 }
199
fir_channels(AVFilterContext * ctx,void * arg,int jobnr,int nb_jobs)200 static int fir_channels(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
201 {
202 AVFrame *out = arg;
203 const int start = (out->channels * jobnr) / nb_jobs;
204 const int end = (out->channels * (jobnr+1)) / nb_jobs;
205
206 for (int ch = start; ch < end; ch++) {
207 fir_channel(ctx, out, ch);
208 }
209
210 return 0;
211 }
212
fir_frame(AudioFIRContext * s,AVFrame * in,AVFilterLink * outlink)213 static int fir_frame(AudioFIRContext *s, AVFrame *in, AVFilterLink *outlink)
214 {
215 AVFilterContext *ctx = outlink->src;
216 AVFrame *out = NULL;
217
218 out = ff_get_audio_buffer(outlink, in->nb_samples);
219 if (!out) {
220 av_frame_free(&in);
221 return AVERROR(ENOMEM);
222 }
223
224 if (s->pts == AV_NOPTS_VALUE)
225 s->pts = in->pts;
226 s->in = in;
227 ctx->internal->execute(ctx, fir_channels, out, NULL, FFMIN(outlink->channels,
228 ff_filter_get_nb_threads(ctx)));
229
230 out->pts = s->pts;
231 if (s->pts != AV_NOPTS_VALUE)
232 s->pts += av_rescale_q(out->nb_samples, (AVRational){1, outlink->sample_rate}, outlink->time_base);
233
234 av_frame_free(&in);
235 s->in = NULL;
236
237 return ff_filter_frame(outlink, out);
238 }
239
drawtext(AVFrame * pic,int x,int y,const char * txt,uint32_t color)240 static void drawtext(AVFrame *pic, int x, int y, const char *txt, uint32_t color)
241 {
242 const uint8_t *font;
243 int font_height;
244 int i;
245
246 font = avpriv_cga_font, font_height = 8;
247
248 for (i = 0; txt[i]; i++) {
249 int char_y, mask;
250
251 uint8_t *p = pic->data[0] + y * pic->linesize[0] + (x + i * 8) * 4;
252 for (char_y = 0; char_y < font_height; char_y++) {
253 for (mask = 0x80; mask; mask >>= 1) {
254 if (font[txt[i] * font_height + char_y] & mask)
255 AV_WL32(p, color);
256 p += 4;
257 }
258 p += pic->linesize[0] - 8 * 4;
259 }
260 }
261 }
262
draw_line(AVFrame * out,int x0,int y0,int x1,int y1,uint32_t color)263 static void draw_line(AVFrame *out, int x0, int y0, int x1, int y1, uint32_t color)
264 {
265 int dx = FFABS(x1-x0);
266 int dy = FFABS(y1-y0), sy = y0 < y1 ? 1 : -1;
267 int err = (dx>dy ? dx : -dy) / 2, e2;
268
269 for (;;) {
270 AV_WL32(out->data[0] + y0 * out->linesize[0] + x0 * 4, color);
271
272 if (x0 == x1 && y0 == y1)
273 break;
274
275 e2 = err;
276
277 if (e2 >-dx) {
278 err -= dy;
279 x0--;
280 }
281
282 if (e2 < dy) {
283 err += dx;
284 y0 += sy;
285 }
286 }
287 }
288
draw_response(AVFilterContext * ctx,AVFrame * out)289 static void draw_response(AVFilterContext *ctx, AVFrame *out)
290 {
291 AudioFIRContext *s = ctx->priv;
292 float *mag, *phase, *delay, min = FLT_MAX, max = FLT_MIN;
293 float min_delay = FLT_MAX, max_delay = FLT_MIN;
294 int prev_ymag = -1, prev_yphase = -1, prev_ydelay = -1;
295 char text[32];
296 int channel, i, x;
297
298 memset(out->data[0], 0, s->h * out->linesize[0]);
299
300 phase = av_malloc_array(s->w, sizeof(*phase));
301 mag = av_malloc_array(s->w, sizeof(*mag));
302 delay = av_malloc_array(s->w, sizeof(*delay));
303 if (!mag || !phase || !delay)
304 goto end;
305
306 channel = av_clip(s->ir_channel, 0, s->ir[s->selir]->channels - 1);
307 for (i = 0; i < s->w; i++) {
308 const float *src = (const float *)s->ir[s->selir]->extended_data[channel];
309 double w = i * M_PI / (s->w - 1);
310 double div, real_num = 0., imag_num = 0., real = 0., imag = 0.;
311
312 for (x = 0; x < s->nb_taps; x++) {
313 real += cos(-x * w) * src[x];
314 imag += sin(-x * w) * src[x];
315 real_num += cos(-x * w) * src[x] * x;
316 imag_num += sin(-x * w) * src[x] * x;
317 }
318
319 mag[i] = hypot(real, imag);
320 phase[i] = atan2(imag, real);
321 div = real * real + imag * imag;
322 delay[i] = (real_num * real + imag_num * imag) / div;
323 min = fminf(min, mag[i]);
324 max = fmaxf(max, mag[i]);
325 min_delay = fminf(min_delay, delay[i]);
326 max_delay = fmaxf(max_delay, delay[i]);
327 }
328
329 for (i = 0; i < s->w; i++) {
330 int ymag = mag[i] / max * (s->h - 1);
331 int ydelay = (delay[i] - min_delay) / (max_delay - min_delay) * (s->h - 1);
332 int yphase = (0.5 * (1. + phase[i] / M_PI)) * (s->h - 1);
333
334 ymag = s->h - 1 - av_clip(ymag, 0, s->h - 1);
335 yphase = s->h - 1 - av_clip(yphase, 0, s->h - 1);
336 ydelay = s->h - 1 - av_clip(ydelay, 0, s->h - 1);
337
338 if (prev_ymag < 0)
339 prev_ymag = ymag;
340 if (prev_yphase < 0)
341 prev_yphase = yphase;
342 if (prev_ydelay < 0)
343 prev_ydelay = ydelay;
344
345 draw_line(out, i, ymag, FFMAX(i - 1, 0), prev_ymag, 0xFFFF00FF);
346 draw_line(out, i, yphase, FFMAX(i - 1, 0), prev_yphase, 0xFF00FF00);
347 draw_line(out, i, ydelay, FFMAX(i - 1, 0), prev_ydelay, 0xFF00FFFF);
348
349 prev_ymag = ymag;
350 prev_yphase = yphase;
351 prev_ydelay = ydelay;
352 }
353
354 if (s->w > 400 && s->h > 100) {
355 drawtext(out, 2, 2, "Max Magnitude:", 0xDDDDDDDD);
356 snprintf(text, sizeof(text), "%.2f", max);
357 drawtext(out, 15 * 8 + 2, 2, text, 0xDDDDDDDD);
358
359 drawtext(out, 2, 12, "Min Magnitude:", 0xDDDDDDDD);
360 snprintf(text, sizeof(text), "%.2f", min);
361 drawtext(out, 15 * 8 + 2, 12, text, 0xDDDDDDDD);
362
363 drawtext(out, 2, 22, "Max Delay:", 0xDDDDDDDD);
364 snprintf(text, sizeof(text), "%.2f", max_delay);
365 drawtext(out, 11 * 8 + 2, 22, text, 0xDDDDDDDD);
366
367 drawtext(out, 2, 32, "Min Delay:", 0xDDDDDDDD);
368 snprintf(text, sizeof(text), "%.2f", min_delay);
369 drawtext(out, 11 * 8 + 2, 32, text, 0xDDDDDDDD);
370 }
371
372 end:
373 av_free(delay);
374 av_free(phase);
375 av_free(mag);
376 }
377
init_segment(AVFilterContext * ctx,AudioFIRSegment * seg,int offset,int nb_partitions,int part_size)378 static int init_segment(AVFilterContext *ctx, AudioFIRSegment *seg,
379 int offset, int nb_partitions, int part_size)
380 {
381 AudioFIRContext *s = ctx->priv;
382
383 seg->rdft = av_calloc(ctx->inputs[0]->channels, sizeof(*seg->rdft));
384 seg->irdft = av_calloc(ctx->inputs[0]->channels, sizeof(*seg->irdft));
385 if (!seg->rdft || !seg->irdft)
386 return AVERROR(ENOMEM);
387
388 seg->fft_length = part_size * 2 + 1;
389 seg->part_size = part_size;
390 seg->block_size = FFALIGN(seg->fft_length, 32);
391 seg->coeff_size = FFALIGN(seg->part_size + 1, 32);
392 seg->nb_partitions = nb_partitions;
393 seg->input_size = offset + s->min_part_size;
394 seg->input_offset = offset;
395
396 seg->part_index = av_calloc(ctx->inputs[0]->channels, sizeof(*seg->part_index));
397 seg->output_offset = av_calloc(ctx->inputs[0]->channels, sizeof(*seg->output_offset));
398 if (!seg->part_index || !seg->output_offset)
399 return AVERROR(ENOMEM);
400
401 for (int ch = 0; ch < ctx->inputs[0]->channels && part_size >= 8; ch++) {
402 seg->rdft[ch] = av_rdft_init(av_log2(2 * part_size), DFT_R2C);
403 seg->irdft[ch] = av_rdft_init(av_log2(2 * part_size), IDFT_C2R);
404 if (!seg->rdft[ch] || !seg->irdft[ch])
405 return AVERROR(ENOMEM);
406 }
407
408 seg->sum = ff_get_audio_buffer(ctx->inputs[0], seg->fft_length);
409 seg->block = ff_get_audio_buffer(ctx->inputs[0], seg->nb_partitions * seg->block_size);
410 seg->buffer = ff_get_audio_buffer(ctx->inputs[0], seg->part_size);
411 seg->coeff = ff_get_audio_buffer(ctx->inputs[1 + s->selir], seg->nb_partitions * seg->coeff_size * 2);
412 seg->input = ff_get_audio_buffer(ctx->inputs[0], seg->input_size);
413 seg->output = ff_get_audio_buffer(ctx->inputs[0], seg->part_size);
414 if (!seg->buffer || !seg->sum || !seg->block || !seg->coeff || !seg->input || !seg->output)
415 return AVERROR(ENOMEM);
416
417 return 0;
418 }
419
uninit_segment(AVFilterContext * ctx,AudioFIRSegment * seg)420 static void uninit_segment(AVFilterContext *ctx, AudioFIRSegment *seg)
421 {
422 AudioFIRContext *s = ctx->priv;
423
424 if (seg->rdft) {
425 for (int ch = 0; ch < s->nb_channels; ch++) {
426 av_rdft_end(seg->rdft[ch]);
427 }
428 }
429 av_freep(&seg->rdft);
430
431 if (seg->irdft) {
432 for (int ch = 0; ch < s->nb_channels; ch++) {
433 av_rdft_end(seg->irdft[ch]);
434 }
435 }
436 av_freep(&seg->irdft);
437
438 av_freep(&seg->output_offset);
439 av_freep(&seg->part_index);
440
441 av_frame_free(&seg->block);
442 av_frame_free(&seg->sum);
443 av_frame_free(&seg->buffer);
444 av_frame_free(&seg->coeff);
445 av_frame_free(&seg->input);
446 av_frame_free(&seg->output);
447 seg->input_size = 0;
448 }
449
convert_coeffs(AVFilterContext * ctx)450 static int convert_coeffs(AVFilterContext *ctx)
451 {
452 AudioFIRContext *s = ctx->priv;
453 int ret, i, ch, n, cur_nb_taps;
454 float power = 0;
455
456 if (!s->nb_taps) {
457 int part_size, max_part_size;
458 int left, offset = 0;
459
460 s->nb_taps = ff_inlink_queued_samples(ctx->inputs[1 + s->selir]);
461 if (s->nb_taps <= 0)
462 return AVERROR(EINVAL);
463
464 if (s->minp > s->maxp) {
465 s->maxp = s->minp;
466 }
467
468 left = s->nb_taps;
469 part_size = 1 << av_log2(s->minp);
470 max_part_size = 1 << av_log2(s->maxp);
471
472 s->min_part_size = part_size;
473
474 for (i = 0; left > 0; i++) {
475 int step = part_size == max_part_size ? INT_MAX : 1 + (i == 0);
476 int nb_partitions = FFMIN(step, (left + part_size - 1) / part_size);
477
478 s->nb_segments = i + 1;
479 ret = init_segment(ctx, &s->seg[i], offset, nb_partitions, part_size);
480 if (ret < 0)
481 return ret;
482 offset += nb_partitions * part_size;
483 left -= nb_partitions * part_size;
484 part_size *= 2;
485 part_size = FFMIN(part_size, max_part_size);
486 }
487 }
488
489 if (!s->ir[s->selir]) {
490 ret = ff_inlink_consume_samples(ctx->inputs[1 + s->selir], s->nb_taps, s->nb_taps, &s->ir[s->selir]);
491 if (ret < 0)
492 return ret;
493 if (ret == 0)
494 return AVERROR_BUG;
495 }
496
497 if (s->response)
498 draw_response(ctx, s->video);
499
500 s->gain = 1;
501 cur_nb_taps = s->ir[s->selir]->nb_samples;
502
503 switch (s->gtype) {
504 case -1:
505 /* nothing to do */
506 break;
507 case 0:
508 for (ch = 0; ch < ctx->inputs[1 + s->selir]->channels; ch++) {
509 float *time = (float *)s->ir[s->selir]->extended_data[!s->one2many * ch];
510
511 for (i = 0; i < cur_nb_taps; i++)
512 power += FFABS(time[i]);
513 }
514 s->gain = ctx->inputs[1 + s->selir]->channels / power;
515 break;
516 case 1:
517 for (ch = 0; ch < ctx->inputs[1 + s->selir]->channels; ch++) {
518 float *time = (float *)s->ir[s->selir]->extended_data[!s->one2many * ch];
519
520 for (i = 0; i < cur_nb_taps; i++)
521 power += time[i];
522 }
523 s->gain = ctx->inputs[1 + s->selir]->channels / power;
524 break;
525 case 2:
526 for (ch = 0; ch < ctx->inputs[1 + s->selir]->channels; ch++) {
527 float *time = (float *)s->ir[s->selir]->extended_data[!s->one2many * ch];
528
529 for (i = 0; i < cur_nb_taps; i++)
530 power += time[i] * time[i];
531 }
532 s->gain = sqrtf(ch / power);
533 break;
534 default:
535 return AVERROR_BUG;
536 }
537
538 s->gain = FFMIN(s->gain * s->ir_gain, 1.f);
539 av_log(ctx, AV_LOG_DEBUG, "power %f, gain %f\n", power, s->gain);
540 for (ch = 0; ch < ctx->inputs[1 + s->selir]->channels; ch++) {
541 float *time = (float *)s->ir[s->selir]->extended_data[!s->one2many * ch];
542
543 s->fdsp->vector_fmul_scalar(time, time, s->gain, FFALIGN(cur_nb_taps, 4));
544 }
545
546 av_log(ctx, AV_LOG_DEBUG, "nb_taps: %d\n", cur_nb_taps);
547 av_log(ctx, AV_LOG_DEBUG, "nb_segments: %d\n", s->nb_segments);
548
549 for (ch = 0; ch < ctx->inputs[1 + s->selir]->channels; ch++) {
550 float *time = (float *)s->ir[s->selir]->extended_data[!s->one2many * ch];
551 int toffset = 0;
552
553 for (i = FFMAX(1, s->length * s->nb_taps); i < s->nb_taps; i++)
554 time[i] = 0;
555
556 av_log(ctx, AV_LOG_DEBUG, "channel: %d\n", ch);
557
558 for (int segment = 0; segment < s->nb_segments; segment++) {
559 AudioFIRSegment *seg = &s->seg[segment];
560 float *block = (float *)seg->block->extended_data[ch];
561 FFTComplex *coeff = (FFTComplex *)seg->coeff->extended_data[ch];
562
563 av_log(ctx, AV_LOG_DEBUG, "segment: %d\n", segment);
564
565 for (i = 0; i < seg->nb_partitions; i++) {
566 const float scale = 1.f / seg->part_size;
567 const int coffset = i * seg->coeff_size;
568 const int remaining = s->nb_taps - toffset;
569 const int size = remaining >= seg->part_size ? seg->part_size : remaining;
570
571 if (size < 8) {
572 for (n = 0; n < size; n++)
573 coeff[coffset + n].re = time[toffset + n];
574
575 toffset += size;
576 continue;
577 }
578
579 memset(block, 0, sizeof(*block) * seg->fft_length);
580 memcpy(block, time + toffset, size * sizeof(*block));
581
582 av_rdft_calc(seg->rdft[0], block);
583
584 coeff[coffset].re = block[0] * scale;
585 coeff[coffset].im = 0;
586 for (n = 1; n < seg->part_size; n++) {
587 coeff[coffset + n].re = block[2 * n] * scale;
588 coeff[coffset + n].im = block[2 * n + 1] * scale;
589 }
590 coeff[coffset + seg->part_size].re = block[1] * scale;
591 coeff[coffset + seg->part_size].im = 0;
592
593 toffset += size;
594 }
595
596 av_log(ctx, AV_LOG_DEBUG, "nb_partitions: %d\n", seg->nb_partitions);
597 av_log(ctx, AV_LOG_DEBUG, "partition size: %d\n", seg->part_size);
598 av_log(ctx, AV_LOG_DEBUG, "block size: %d\n", seg->block_size);
599 av_log(ctx, AV_LOG_DEBUG, "fft_length: %d\n", seg->fft_length);
600 av_log(ctx, AV_LOG_DEBUG, "coeff_size: %d\n", seg->coeff_size);
601 av_log(ctx, AV_LOG_DEBUG, "input_size: %d\n", seg->input_size);
602 av_log(ctx, AV_LOG_DEBUG, "input_offset: %d\n", seg->input_offset);
603 }
604 }
605
606 s->have_coeffs = 1;
607
608 return 0;
609 }
610
check_ir(AVFilterLink * link)611 static int check_ir(AVFilterLink *link)
612 {
613 AVFilterContext *ctx = link->dst;
614 AudioFIRContext *s = ctx->priv;
615 int nb_taps, max_nb_taps;
616
617 nb_taps = ff_inlink_queued_samples(link);
618 max_nb_taps = s->max_ir_len * ctx->outputs[0]->sample_rate;
619 if (nb_taps > max_nb_taps) {
620 av_log(ctx, AV_LOG_ERROR, "Too big number of coefficients: %d > %d.\n", nb_taps, max_nb_taps);
621 return AVERROR(EINVAL);
622 }
623
624 return 0;
625 }
626
activate(AVFilterContext * ctx)627 static int activate(AVFilterContext *ctx)
628 {
629 AudioFIRContext *s = ctx->priv;
630 AVFilterLink *outlink = ctx->outputs[0];
631 int ret, status, available, wanted;
632 AVFrame *in = NULL;
633 int64_t pts;
634
635 FF_FILTER_FORWARD_STATUS_BACK_ALL(ctx->outputs[0], ctx);
636 if (s->response)
637 FF_FILTER_FORWARD_STATUS_BACK_ALL(ctx->outputs[1], ctx);
638 if (!s->eof_coeffs[s->selir]) {
639 ret = check_ir(ctx->inputs[1 + s->selir]);
640 if (ret < 0)
641 return ret;
642
643 if (ff_outlink_get_status(ctx->inputs[1 + s->selir]) == AVERROR_EOF)
644 s->eof_coeffs[s->selir] = 1;
645
646 if (!s->eof_coeffs[s->selir]) {
647 if (ff_outlink_frame_wanted(ctx->outputs[0]))
648 ff_inlink_request_frame(ctx->inputs[1 + s->selir]);
649 else if (s->response && ff_outlink_frame_wanted(ctx->outputs[1]))
650 ff_inlink_request_frame(ctx->inputs[1 + s->selir]);
651 return 0;
652 }
653 }
654
655 if (!s->have_coeffs && s->eof_coeffs[s->selir]) {
656 ret = convert_coeffs(ctx);
657 if (ret < 0)
658 return ret;
659 }
660
661 available = ff_inlink_queued_samples(ctx->inputs[0]);
662 wanted = FFMAX(s->min_part_size, (available / s->min_part_size) * s->min_part_size);
663 ret = ff_inlink_consume_samples(ctx->inputs[0], wanted, wanted, &in);
664 if (ret > 0)
665 ret = fir_frame(s, in, outlink);
666
667 if (ret < 0)
668 return ret;
669
670 if (s->response && s->have_coeffs) {
671 int64_t old_pts = s->video->pts;
672 int64_t new_pts = av_rescale_q(s->pts, ctx->inputs[0]->time_base, ctx->outputs[1]->time_base);
673
674 if (ff_outlink_frame_wanted(ctx->outputs[1]) && old_pts < new_pts) {
675 AVFrame *clone;
676 s->video->pts = new_pts;
677 clone = av_frame_clone(s->video);
678 if (!clone)
679 return AVERROR(ENOMEM);
680 return ff_filter_frame(ctx->outputs[1], clone);
681 }
682 }
683
684 if (ff_inlink_queued_samples(ctx->inputs[0]) >= s->min_part_size) {
685 ff_filter_set_ready(ctx, 10);
686 return 0;
687 }
688
689 if (ff_inlink_acknowledge_status(ctx->inputs[0], &status, &pts)) {
690 if (status == AVERROR_EOF) {
691 ff_outlink_set_status(ctx->outputs[0], status, pts);
692 if (s->response)
693 ff_outlink_set_status(ctx->outputs[1], status, pts);
694 return 0;
695 }
696 }
697
698 if (ff_outlink_frame_wanted(ctx->outputs[0]) &&
699 !ff_outlink_get_status(ctx->inputs[0])) {
700 ff_inlink_request_frame(ctx->inputs[0]);
701 return 0;
702 }
703
704 if (s->response &&
705 ff_outlink_frame_wanted(ctx->outputs[1]) &&
706 !ff_outlink_get_status(ctx->inputs[0])) {
707 ff_inlink_request_frame(ctx->inputs[0]);
708 return 0;
709 }
710
711 return FFERROR_NOT_READY;
712 }
713
query_formats(AVFilterContext * ctx)714 static int query_formats(AVFilterContext *ctx)
715 {
716 AudioFIRContext *s = ctx->priv;
717 AVFilterFormats *formats;
718 AVFilterChannelLayouts *layouts;
719 static const enum AVSampleFormat sample_fmts[] = {
720 AV_SAMPLE_FMT_FLTP,
721 AV_SAMPLE_FMT_NONE
722 };
723 static const enum AVPixelFormat pix_fmts[] = {
724 AV_PIX_FMT_RGB0,
725 AV_PIX_FMT_NONE
726 };
727 int ret;
728
729 if (s->response) {
730 AVFilterLink *videolink = ctx->outputs[1];
731 formats = ff_make_format_list(pix_fmts);
732 if ((ret = ff_formats_ref(formats, &videolink->incfg.formats)) < 0)
733 return ret;
734 }
735
736 layouts = ff_all_channel_counts();
737 if (!layouts)
738 return AVERROR(ENOMEM);
739
740 if (s->ir_format) {
741 ret = ff_set_common_channel_layouts(ctx, layouts);
742 if (ret < 0)
743 return ret;
744 } else {
745 AVFilterChannelLayouts *mono = NULL;
746
747 if ((ret = ff_channel_layouts_ref(layouts, &ctx->inputs[0]->outcfg.channel_layouts)) < 0)
748 return ret;
749 if ((ret = ff_channel_layouts_ref(layouts, &ctx->outputs[0]->incfg.channel_layouts)) < 0)
750 return ret;
751
752 ret = ff_add_channel_layout(&mono, AV_CH_LAYOUT_MONO);
753 if (ret)
754 return ret;
755 for (int i = 1; i < ctx->nb_inputs; i++) {
756 if ((ret = ff_channel_layouts_ref(mono, &ctx->inputs[i]->outcfg.channel_layouts)) < 0)
757 return ret;
758 }
759 }
760
761 formats = ff_make_format_list(sample_fmts);
762 if ((ret = ff_set_common_formats(ctx, formats)) < 0)
763 return ret;
764
765 formats = ff_all_samplerates();
766 return ff_set_common_samplerates(ctx, formats);
767 }
768
config_output(AVFilterLink * outlink)769 static int config_output(AVFilterLink *outlink)
770 {
771 AVFilterContext *ctx = outlink->src;
772 AudioFIRContext *s = ctx->priv;
773
774 s->one2many = ctx->inputs[1 + s->selir]->channels == 1;
775 outlink->sample_rate = ctx->inputs[0]->sample_rate;
776 outlink->time_base = ctx->inputs[0]->time_base;
777 outlink->channel_layout = ctx->inputs[0]->channel_layout;
778 outlink->channels = ctx->inputs[0]->channels;
779
780 s->nb_channels = outlink->channels;
781 s->nb_coef_channels = ctx->inputs[1 + s->selir]->channels;
782 s->pts = AV_NOPTS_VALUE;
783
784 return 0;
785 }
786
uninit(AVFilterContext * ctx)787 static av_cold void uninit(AVFilterContext *ctx)
788 {
789 AudioFIRContext *s = ctx->priv;
790
791 for (int i = 0; i < s->nb_segments; i++) {
792 uninit_segment(ctx, &s->seg[i]);
793 }
794
795 av_freep(&s->fdsp);
796
797 for (int i = 0; i < s->nb_irs; i++) {
798 av_frame_free(&s->ir[i]);
799 }
800
801 for (unsigned i = 1; i < ctx->nb_inputs; i++)
802 av_freep(&ctx->input_pads[i].name);
803
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 = "main",
842 .type = AVMEDIA_TYPE_AUDIO,
843 };
844
845 ret = ff_insert_inpad(ctx, 0, &pad);
846 if (ret < 0)
847 return ret;
848
849 for (int n = 0; n < s->nb_irs; n++) {
850 pad = (AVFilterPad) {
851 .name = av_asprintf("ir%d", n),
852 .type = AVMEDIA_TYPE_AUDIO,
853 };
854
855 if (!pad.name)
856 return AVERROR(ENOMEM);
857
858 ret = ff_insert_inpad(ctx, n + 1, &pad);
859 if (ret < 0) {
860 av_freep(&pad.name);
861 return ret;
862 }
863 }
864
865 pad = (AVFilterPad) {
866 .name = "default",
867 .type = AVMEDIA_TYPE_AUDIO,
868 .config_props = config_output,
869 };
870
871 ret = ff_insert_outpad(ctx, 0, &pad);
872 if (ret < 0)
873 return ret;
874
875 if (s->response) {
876 vpad = (AVFilterPad){
877 .name = "filter_response",
878 .type = AVMEDIA_TYPE_VIDEO,
879 .config_props = config_video,
880 };
881
882 ret = ff_insert_outpad(ctx, 1, &vpad);
883 if (ret < 0)
884 return ret;
885 }
886
887 s->fdsp = avpriv_float_dsp_alloc(0);
888 if (!s->fdsp)
889 return AVERROR(ENOMEM);
890
891 ff_afir_init(&s->afirdsp);
892
893 return 0;
894 }
895
process_command(AVFilterContext * ctx,const char * cmd,const char * arg,char * res,int res_len,int flags)896 static int process_command(AVFilterContext *ctx,
897 const char *cmd,
898 const char *arg,
899 char *res,
900 int res_len,
901 int flags)
902 {
903 AudioFIRContext *s = ctx->priv;
904 int prev_ir = s->selir;
905 int ret = ff_filter_process_command(ctx, cmd, arg, res, res_len, flags);
906
907 if (ret < 0)
908 return ret;
909
910 s->selir = FFMIN(s->nb_irs - 1, s->selir);
911
912 if (prev_ir != s->selir) {
913 s->have_coeffs = 0;
914 }
915
916 return 0;
917 }
918
919 #define AF AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
920 #define AFR AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_RUNTIME_PARAM
921 #define VF AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
922 #define OFFSET(x) offsetof(AudioFIRContext, x)
923
924 static const AVOption afir_options[] = {
925 { "dry", "set dry gain", OFFSET(dry_gain), AV_OPT_TYPE_FLOAT, {.dbl=1}, 0, 10, AF },
926 { "wet", "set wet gain", OFFSET(wet_gain), AV_OPT_TYPE_FLOAT, {.dbl=1}, 0, 10, AF },
927 { "length", "set IR length", OFFSET(length), AV_OPT_TYPE_FLOAT, {.dbl=1}, 0, 1, AF },
928 { "gtype", "set IR auto gain type",OFFSET(gtype), AV_OPT_TYPE_INT, {.i64=0}, -1, 2, AF, "gtype" },
929 { "none", "without auto gain", 0, AV_OPT_TYPE_CONST, {.i64=-1}, 0, 0, AF, "gtype" },
930 { "peak", "peak gain", 0, AV_OPT_TYPE_CONST, {.i64=0}, 0, 0, AF, "gtype" },
931 { "dc", "DC gain", 0, AV_OPT_TYPE_CONST, {.i64=1}, 0, 0, AF, "gtype" },
932 { "gn", "gain to noise", 0, AV_OPT_TYPE_CONST, {.i64=2}, 0, 0, AF, "gtype" },
933 { "irgain", "set IR gain", OFFSET(ir_gain), AV_OPT_TYPE_FLOAT, {.dbl=1}, 0, 1, AF },
934 { "irfmt", "set IR format", OFFSET(ir_format), AV_OPT_TYPE_INT, {.i64=1}, 0, 1, AF, "irfmt" },
935 { "mono", "single channel", 0, AV_OPT_TYPE_CONST, {.i64=0}, 0, 0, AF, "irfmt" },
936 { "input", "same as input", 0, AV_OPT_TYPE_CONST, {.i64=1}, 0, 0, AF, "irfmt" },
937 { "maxir", "set max IR length", OFFSET(max_ir_len), AV_OPT_TYPE_FLOAT, {.dbl=30}, 0.1, 60, AF },
938 { "response", "show IR frequency response", OFFSET(response), AV_OPT_TYPE_BOOL, {.i64=0}, 0, 1, VF },
939 { "channel", "set IR channel to display frequency response", OFFSET(ir_channel), AV_OPT_TYPE_INT, {.i64=0}, 0, 1024, VF },
940 { "size", "set video size", OFFSET(w), AV_OPT_TYPE_IMAGE_SIZE, {.str = "hd720"}, 0, 0, VF },
941 { "rate", "set video rate", OFFSET(frame_rate), AV_OPT_TYPE_VIDEO_RATE, {.str = "25"}, 0, INT32_MAX, VF },
942 { "minp", "set min partition size", OFFSET(minp), AV_OPT_TYPE_INT, {.i64=8192}, 1, 32768, AF },
943 { "maxp", "set max partition size", OFFSET(maxp), AV_OPT_TYPE_INT, {.i64=8192}, 8, 32768, AF },
944 { "nbirs", "set number of input IRs",OFFSET(nb_irs),AV_OPT_TYPE_INT, {.i64=1}, 1, 32, AF },
945 { "ir", "select IR", OFFSET(selir), AV_OPT_TYPE_INT, {.i64=0}, 0, 31, AFR },
946 { NULL }
947 };
948
949 AVFILTER_DEFINE_CLASS(afir);
950
951 AVFilter ff_af_afir = {
952 .name = "afir",
953 .description = NULL_IF_CONFIG_SMALL("Apply Finite Impulse Response filter with supplied coefficients in additional stream(s)."),
954 .priv_size = sizeof(AudioFIRContext),
955 .priv_class = &afir_class,
956 .query_formats = query_formats,
957 .init = init,
958 .activate = activate,
959 .uninit = uninit,
960 .process_command = process_command,
961 .flags = AVFILTER_FLAG_DYNAMIC_INPUTS |
962 AVFILTER_FLAG_DYNAMIC_OUTPUTS |
963 AVFILTER_FLAG_SLICE_THREADS,
964 };
965