FFmpeg coverage


Directory: ../../../ffmpeg/
File: src/libavfilter/af_headphone.c
Date: 2024-03-28 14:59:00
Exec Total Coverage
Lines: 0 421 0.0%
Functions: 0 13 0.0%
Branches: 0 210 0.0%

Line Branch Exec Source
1 /*
2 * Copyright (C) 2017 Paul B Mahol
3 * Copyright (C) 2013-2015 Andreas Fuchs, Wolfgang Hrauda
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 #include <math.h>
22
23 #include "libavutil/avstring.h"
24 #include "libavutil/channel_layout.h"
25 #include "libavutil/float_dsp.h"
26 #include "libavutil/intmath.h"
27 #include "libavutil/opt.h"
28 #include "libavutil/tx.h"
29
30 #include "avfilter.h"
31 #include "filters.h"
32 #include "formats.h"
33 #include "internal.h"
34 #include "audio.h"
35
36 #define TIME_DOMAIN 0
37 #define FREQUENCY_DOMAIN 1
38
39 #define HRIR_STEREO 0
40 #define HRIR_MULTI 1
41
42 typedef struct HeadphoneContext {
43 const AVClass *class;
44
45 char *map;
46 int type;
47
48 int lfe_channel;
49
50 int have_hrirs;
51 int eof_hrirs;
52
53 int ir_len;
54 int air_len;
55
56 int nb_hrir_inputs;
57
58 int nb_irs;
59
60 float gain;
61 float lfe_gain, gain_lfe;
62
63 float *ringbuffer[2];
64 int write[2];
65
66 int buffer_length;
67 int n_fft;
68 int size;
69 int hrir_fmt;
70
71 float *data_ir[2];
72 float *temp_src[2];
73 AVComplexFloat *out_fft[2];
74 AVComplexFloat *in_fft[2];
75 AVComplexFloat *temp_afft[2];
76
77 AVTXContext *fft[2], *ifft[2];
78 av_tx_fn tx_fn[2], itx_fn[2];
79 AVComplexFloat *data_hrtf[2];
80
81 float (*scalarproduct_float)(const float *v1, const float *v2, int len);
82 struct hrir_inputs {
83 int ir_len;
84 int eof;
85 } hrir_in[64];
86 AVChannelLayout map_channel_layout;
87 enum AVChannel mapping[64];
88 uint8_t hrir_map[64];
89 } HeadphoneContext;
90
91 static int parse_channel_name(const char *arg, enum AVChannel *rchannel)
92 {
93 int channel = av_channel_from_string(arg);
94
95 if (channel < 0 || channel >= 64)
96 return AVERROR(EINVAL);
97 *rchannel = channel;
98 return 0;
99 }
100
101 static void parse_map(AVFilterContext *ctx)
102 {
103 HeadphoneContext *s = ctx->priv;
104 char *arg, *tokenizer, *p;
105 uint64_t used_channels = 0;
106
107 p = s->map;
108 while ((arg = av_strtok(p, "|", &tokenizer))) {
109 enum AVChannel out_channel;
110
111 p = NULL;
112 if (parse_channel_name(arg, &out_channel)) {
113 av_log(ctx, AV_LOG_WARNING, "Failed to parse \'%s\' as channel name.\n", arg);
114 continue;
115 }
116 if (used_channels & (1ULL << out_channel)) {
117 av_log(ctx, AV_LOG_WARNING, "Ignoring duplicate channel '%s'.\n", arg);
118 continue;
119 }
120 used_channels |= (1ULL << out_channel);
121 s->mapping[s->nb_irs] = out_channel;
122 s->nb_irs++;
123 }
124 av_channel_layout_from_mask(&s->map_channel_layout, used_channels);
125
126 if (s->hrir_fmt == HRIR_MULTI)
127 s->nb_hrir_inputs = 1;
128 else
129 s->nb_hrir_inputs = s->nb_irs;
130 }
131
132 typedef struct ThreadData {
133 AVFrame *in, *out;
134 int *write;
135 float **ir;
136 int *n_clippings;
137 float **ringbuffer;
138 float **temp_src;
139 AVComplexFloat **out_fft;
140 AVComplexFloat **in_fft;
141 AVComplexFloat **temp_afft;
142 } ThreadData;
143
144 static int headphone_convolute(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
145 {
146 HeadphoneContext *s = ctx->priv;
147 ThreadData *td = arg;
148 AVFrame *in = td->in, *out = td->out;
149 int offset = jobnr;
150 int *write = &td->write[jobnr];
151 const float *const ir = td->ir[jobnr];
152 int *n_clippings = &td->n_clippings[jobnr];
153 float *ringbuffer = td->ringbuffer[jobnr];
154 float *temp_src = td->temp_src[jobnr];
155 const int ir_len = s->ir_len;
156 const int air_len = s->air_len;
157 const float *src = (const float *)in->data[0];
158 float *dst = (float *)out->data[0];
159 const int in_channels = in->ch_layout.nb_channels;
160 const int buffer_length = s->buffer_length;
161 const uint32_t modulo = (uint32_t)buffer_length - 1;
162 float *buffer[64];
163 int wr = *write;
164 int read;
165 int i, l;
166
167 dst += offset;
168 for (l = 0; l < in_channels; l++) {
169 buffer[l] = ringbuffer + l * buffer_length;
170 }
171
172 for (i = 0; i < in->nb_samples; i++) {
173 const float *cur_ir = ir;
174
175 *dst = 0;
176 for (l = 0; l < in_channels; l++) {
177 *(buffer[l] + wr) = src[l];
178 }
179
180 for (l = 0; l < in_channels; cur_ir += air_len, l++) {
181 const float *const bptr = buffer[l];
182
183 if (l == s->lfe_channel) {
184 *dst += *(buffer[s->lfe_channel] + wr) * s->gain_lfe;
185 continue;
186 }
187
188 read = (wr - (ir_len - 1)) & modulo;
189
190 if (read + ir_len < buffer_length) {
191 memcpy(temp_src, bptr + read, ir_len * sizeof(*temp_src));
192 } else {
193 int len = FFMIN(air_len - (read % ir_len), buffer_length - read);
194
195 memcpy(temp_src, bptr + read, len * sizeof(*temp_src));
196 memcpy(temp_src + len, bptr, (air_len - len) * sizeof(*temp_src));
197 }
198
199 dst[0] += s->scalarproduct_float(cur_ir, temp_src, FFALIGN(ir_len, 32));
200 }
201
202 if (fabsf(dst[0]) > 1)
203 n_clippings[0]++;
204
205 dst += 2;
206 src += in_channels;
207 wr = (wr + 1) & modulo;
208 }
209
210 *write = wr;
211
212 return 0;
213 }
214
215 static int headphone_fast_convolute(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
216 {
217 HeadphoneContext *s = ctx->priv;
218 ThreadData *td = arg;
219 AVFrame *in = td->in, *out = td->out;
220 int offset = jobnr;
221 int *write = &td->write[jobnr];
222 AVComplexFloat *hrtf = s->data_hrtf[jobnr];
223 int *n_clippings = &td->n_clippings[jobnr];
224 float *ringbuffer = td->ringbuffer[jobnr];
225 const int ir_len = s->ir_len;
226 const float *src = (const float *)in->data[0];
227 float *dst = (float *)out->data[0];
228 const int in_channels = in->ch_layout.nb_channels;
229 const int buffer_length = s->buffer_length;
230 const uint32_t modulo = (uint32_t)buffer_length - 1;
231 AVComplexFloat *fft_out = s->out_fft[jobnr];
232 AVComplexFloat *fft_in = s->in_fft[jobnr];
233 AVComplexFloat *fft_acc = s->temp_afft[jobnr];
234 AVTXContext *ifft = s->ifft[jobnr];
235 AVTXContext *fft = s->fft[jobnr];
236 av_tx_fn tx_fn = s->tx_fn[jobnr];
237 av_tx_fn itx_fn = s->itx_fn[jobnr];
238 const int n_fft = s->n_fft;
239 const float fft_scale = 1.0f / s->n_fft;
240 AVComplexFloat *hrtf_offset;
241 int wr = *write;
242 int n_read;
243 int i, j;
244
245 dst += offset;
246
247 n_read = FFMIN(ir_len, in->nb_samples);
248 for (j = 0; j < n_read; j++) {
249 dst[2 * j] = ringbuffer[wr];
250 ringbuffer[wr] = 0.0;
251 wr = (wr + 1) & modulo;
252 }
253
254 for (j = n_read; j < in->nb_samples; j++) {
255 dst[2 * j] = 0;
256 }
257
258 memset(fft_acc, 0, sizeof(AVComplexFloat) * n_fft);
259
260 for (i = 0; i < in_channels; i++) {
261 if (i == s->lfe_channel) {
262 for (j = 0; j < in->nb_samples; j++) {
263 dst[2 * j] += src[i + j * in_channels] * s->gain_lfe;
264 }
265 continue;
266 }
267
268 offset = i * n_fft;
269 hrtf_offset = hrtf + s->hrir_map[i] * n_fft;
270
271 memset(fft_in, 0, sizeof(AVComplexFloat) * n_fft);
272
273 for (j = 0; j < in->nb_samples; j++) {
274 fft_in[j].re = src[j * in_channels + i];
275 }
276
277 tx_fn(fft, fft_out, fft_in, sizeof(*fft_in));
278
279 for (j = 0; j < n_fft; j++) {
280 const AVComplexFloat *hcomplex = hrtf_offset + j;
281 const float re = fft_out[j].re;
282 const float im = fft_out[j].im;
283
284 fft_acc[j].re += re * hcomplex->re - im * hcomplex->im;
285 fft_acc[j].im += re * hcomplex->im + im * hcomplex->re;
286 }
287 }
288
289 itx_fn(ifft, fft_out, fft_acc, sizeof(*fft_acc));
290
291 for (j = 0; j < in->nb_samples; j++) {
292 dst[2 * j] += fft_out[j].re * fft_scale;
293 if (fabsf(dst[2 * j]) > 1)
294 n_clippings[0]++;
295 }
296
297 for (j = 0; j < ir_len - 1; j++) {
298 int write_pos = (wr + j) & modulo;
299
300 *(ringbuffer + write_pos) += fft_out[in->nb_samples + j].re * fft_scale;
301 }
302
303 *write = wr;
304
305 return 0;
306 }
307
308 static int check_ir(AVFilterLink *inlink, int input_number)
309 {
310 AVFilterContext *ctx = inlink->dst;
311 HeadphoneContext *s = ctx->priv;
312 int ir_len, max_ir_len;
313
314 ir_len = ff_inlink_queued_samples(inlink);
315 max_ir_len = 65536;
316 if (ir_len > max_ir_len) {
317 av_log(ctx, AV_LOG_ERROR, "Too big length of IRs: %d > %d.\n", ir_len, max_ir_len);
318 return AVERROR(EINVAL);
319 }
320 s->hrir_in[input_number].ir_len = ir_len;
321 s->ir_len = FFMAX(ir_len, s->ir_len);
322
323 if (ff_inlink_check_available_samples(inlink, ir_len + 1) == 1) {
324 s->hrir_in[input_number].eof = 1;
325 return 1;
326 }
327
328 if (!s->hrir_in[input_number].eof) {
329 ff_inlink_request_frame(inlink);
330 return 0;
331 }
332
333 return 0;
334 }
335
336 static int headphone_frame(HeadphoneContext *s, AVFrame *in, AVFilterLink *outlink)
337 {
338 AVFilterContext *ctx = outlink->src;
339 int n_clippings[2] = { 0 };
340 ThreadData td;
341 AVFrame *out;
342
343 out = ff_get_audio_buffer(outlink, in->nb_samples);
344 if (!out) {
345 av_frame_free(&in);
346 return AVERROR(ENOMEM);
347 }
348 out->pts = in->pts;
349
350 td.in = in; td.out = out; td.write = s->write;
351 td.ir = s->data_ir; td.n_clippings = n_clippings;
352 td.ringbuffer = s->ringbuffer; td.temp_src = s->temp_src;
353 td.out_fft = s->out_fft;
354 td.in_fft = s->in_fft;
355 td.temp_afft = s->temp_afft;
356
357 if (s->type == TIME_DOMAIN) {
358 ff_filter_execute(ctx, headphone_convolute, &td, NULL, 2);
359 } else {
360 ff_filter_execute(ctx, headphone_fast_convolute, &td, NULL, 2);
361 }
362
363 if (n_clippings[0] + n_clippings[1] > 0) {
364 av_log(ctx, AV_LOG_WARNING, "%d of %d samples clipped. Please reduce gain.\n",
365 n_clippings[0] + n_clippings[1], out->nb_samples * 2);
366 }
367
368 av_frame_free(&in);
369 return ff_filter_frame(outlink, out);
370 }
371
372 static int convert_coeffs(AVFilterContext *ctx, AVFilterLink *inlink)
373 {
374 struct HeadphoneContext *s = ctx->priv;
375 const int ir_len = s->ir_len;
376 int nb_input_channels = ctx->inputs[0]->ch_layout.nb_channels;
377 const int nb_hrir_channels = s->nb_hrir_inputs == 1 ? ctx->inputs[1]->ch_layout.nb_channels : s->nb_hrir_inputs * 2;
378 float gain_lin = expf((s->gain - 3 * nb_input_channels) / 20 * M_LN10);
379 AVFrame *frame;
380 int ret = 0;
381 int n_fft;
382 int i, j, k;
383
384 s->air_len = 1 << (32 - ff_clz(ir_len));
385 if (s->type == TIME_DOMAIN) {
386 s->air_len = FFALIGN(s->air_len, 32);
387 }
388 s->buffer_length = 1 << (32 - ff_clz(s->air_len));
389 s->n_fft = n_fft = 1 << (32 - ff_clz(ir_len + s->size));
390
391 if (s->type == FREQUENCY_DOMAIN) {
392 float scale = 1.f;
393
394 ret = av_tx_init(&s->fft[0], &s->tx_fn[0], AV_TX_FLOAT_FFT, 0, s->n_fft, &scale, 0);
395 if (ret < 0)
396 goto fail;
397 ret = av_tx_init(&s->fft[1], &s->tx_fn[1], AV_TX_FLOAT_FFT, 0, s->n_fft, &scale, 0);
398 if (ret < 0)
399 goto fail;
400 ret = av_tx_init(&s->ifft[0], &s->itx_fn[0], AV_TX_FLOAT_FFT, 1, s->n_fft, &scale, 0);
401 if (ret < 0)
402 goto fail;
403 ret = av_tx_init(&s->ifft[1], &s->itx_fn[1], AV_TX_FLOAT_FFT, 1, s->n_fft, &scale, 0);
404 if (ret < 0)
405 goto fail;
406
407 if (!s->fft[0] || !s->fft[1] || !s->ifft[0] || !s->ifft[1]) {
408 av_log(ctx, AV_LOG_ERROR, "Unable to create FFT contexts of size %d.\n", s->n_fft);
409 ret = AVERROR(ENOMEM);
410 goto fail;
411 }
412 }
413
414 if (s->type == TIME_DOMAIN) {
415 s->ringbuffer[0] = av_calloc(s->buffer_length, sizeof(float) * nb_input_channels);
416 s->ringbuffer[1] = av_calloc(s->buffer_length, sizeof(float) * nb_input_channels);
417 } else {
418 s->ringbuffer[0] = av_calloc(s->buffer_length, sizeof(float));
419 s->ringbuffer[1] = av_calloc(s->buffer_length, sizeof(float));
420 s->out_fft[0] = av_calloc(s->n_fft, sizeof(AVComplexFloat));
421 s->out_fft[1] = av_calloc(s->n_fft, sizeof(AVComplexFloat));
422 s->in_fft[0] = av_calloc(s->n_fft, sizeof(AVComplexFloat));
423 s->in_fft[1] = av_calloc(s->n_fft, sizeof(AVComplexFloat));
424 s->temp_afft[0] = av_calloc(s->n_fft, sizeof(AVComplexFloat));
425 s->temp_afft[1] = av_calloc(s->n_fft, sizeof(AVComplexFloat));
426 if (!s->in_fft[0] || !s->in_fft[1] ||
427 !s->out_fft[0] || !s->out_fft[1] ||
428 !s->temp_afft[0] || !s->temp_afft[1]) {
429 ret = AVERROR(ENOMEM);
430 goto fail;
431 }
432 }
433
434 if (!s->ringbuffer[0] || !s->ringbuffer[1]) {
435 ret = AVERROR(ENOMEM);
436 goto fail;
437 }
438
439 if (s->type == TIME_DOMAIN) {
440 s->temp_src[0] = av_calloc(s->air_len, sizeof(float));
441 s->temp_src[1] = av_calloc(s->air_len, sizeof(float));
442
443 s->data_ir[0] = av_calloc(nb_hrir_channels * s->air_len, sizeof(*s->data_ir[0]));
444 s->data_ir[1] = av_calloc(nb_hrir_channels * s->air_len, sizeof(*s->data_ir[1]));
445 if (!s->data_ir[0] || !s->data_ir[1] || !s->temp_src[0] || !s->temp_src[1]) {
446 ret = AVERROR(ENOMEM);
447 goto fail;
448 }
449 } else {
450 s->data_hrtf[0] = av_calloc(n_fft, sizeof(*s->data_hrtf[0]) * nb_hrir_channels);
451 s->data_hrtf[1] = av_calloc(n_fft, sizeof(*s->data_hrtf[1]) * nb_hrir_channels);
452 if (!s->data_hrtf[0] || !s->data_hrtf[1]) {
453 ret = AVERROR(ENOMEM);
454 goto fail;
455 }
456 }
457
458 for (i = 0; i < s->nb_hrir_inputs; av_frame_free(&frame), i++) {
459 int len = s->hrir_in[i].ir_len;
460 float *ptr;
461
462 ret = ff_inlink_consume_samples(ctx->inputs[i + 1], len, len, &frame);
463 if (ret < 0)
464 goto fail;
465 ptr = (float *)frame->extended_data[0];
466
467 if (s->hrir_fmt == HRIR_STEREO) {
468 int idx = av_channel_layout_index_from_channel(&s->map_channel_layout,
469 s->mapping[i]);
470 if (idx < 0)
471 continue;
472
473 s->hrir_map[i] = idx;
474 if (s->type == TIME_DOMAIN) {
475 float *data_ir_l = s->data_ir[0] + idx * s->air_len;
476 float *data_ir_r = s->data_ir[1] + idx * s->air_len;
477
478 for (j = 0; j < len; j++) {
479 data_ir_l[j] = ptr[len * 2 - j * 2 - 2] * gain_lin;
480 data_ir_r[j] = ptr[len * 2 - j * 2 - 1] * gain_lin;
481 }
482 } else {
483 AVComplexFloat *fft_out_l = s->data_hrtf[0] + idx * n_fft;
484 AVComplexFloat *fft_out_r = s->data_hrtf[1] + idx * n_fft;
485 AVComplexFloat *fft_in_l = s->in_fft[0];
486 AVComplexFloat *fft_in_r = s->in_fft[1];
487
488 for (j = 0; j < len; j++) {
489 fft_in_l[j].re = ptr[j * 2 ] * gain_lin;
490 fft_in_r[j].re = ptr[j * 2 + 1] * gain_lin;
491 }
492
493 s->tx_fn[0](s->fft[0], fft_out_l, fft_in_l, sizeof(*fft_in_l));
494 s->tx_fn[0](s->fft[0], fft_out_r, fft_in_r, sizeof(*fft_in_r));
495 }
496 } else {
497 int I, N = ctx->inputs[1]->ch_layout.nb_channels;
498
499 for (k = 0; k < N / 2; k++) {
500 int idx = av_channel_layout_index_from_channel(&inlink->ch_layout,
501 s->mapping[k]);
502 if (idx < 0)
503 continue;
504
505 s->hrir_map[k] = idx;
506 I = k * 2;
507 if (s->type == TIME_DOMAIN) {
508 float *data_ir_l = s->data_ir[0] + idx * s->air_len;
509 float *data_ir_r = s->data_ir[1] + idx * s->air_len;
510
511 for (j = 0; j < len; j++) {
512 data_ir_l[j] = ptr[len * N - j * N - N + I ] * gain_lin;
513 data_ir_r[j] = ptr[len * N - j * N - N + I + 1] * gain_lin;
514 }
515 } else {
516 AVComplexFloat *fft_out_l = s->data_hrtf[0] + idx * n_fft;
517 AVComplexFloat *fft_out_r = s->data_hrtf[1] + idx * n_fft;
518 AVComplexFloat *fft_in_l = s->in_fft[0];
519 AVComplexFloat *fft_in_r = s->in_fft[1];
520
521 for (j = 0; j < len; j++) {
522 fft_in_l[j].re = ptr[j * N + I ] * gain_lin;
523 fft_in_r[j].re = ptr[j * N + I + 1] * gain_lin;
524 }
525
526 s->tx_fn[0](s->fft[0], fft_out_l, fft_in_l, sizeof(*fft_in_l));
527 s->tx_fn[0](s->fft[0], fft_out_r, fft_in_r, sizeof(*fft_in_r));
528 }
529 }
530 }
531 }
532
533 s->have_hrirs = 1;
534
535 fail:
536 return ret;
537 }
538
539 static int activate(AVFilterContext *ctx)
540 {
541 HeadphoneContext *s = ctx->priv;
542 AVFilterLink *inlink = ctx->inputs[0];
543 AVFilterLink *outlink = ctx->outputs[0];
544 AVFrame *in = NULL;
545 int i, ret;
546
547 FF_FILTER_FORWARD_STATUS_BACK_ALL(outlink, ctx);
548 if (!s->eof_hrirs) {
549 int eof = 1;
550 for (i = 0; i < s->nb_hrir_inputs; i++) {
551 AVFilterLink *input = ctx->inputs[i + 1];
552
553 if (s->hrir_in[i].eof)
554 continue;
555
556 if ((ret = check_ir(input, i)) <= 0)
557 return ret;
558
559 if (s->hrir_in[i].eof) {
560 if (!ff_inlink_queued_samples(input)) {
561 av_log(ctx, AV_LOG_ERROR, "No samples provided for "
562 "HRIR stream %d.\n", i);
563 return AVERROR_INVALIDDATA;
564 }
565 } else {
566 eof = 0;
567 }
568 }
569 if (!eof) {
570 ff_filter_set_ready(ctx, 100);
571 return 0;
572 }
573 s->eof_hrirs = 1;
574
575 ret = convert_coeffs(ctx, inlink);
576 if (ret < 0)
577 return ret;
578 } else if (!s->have_hrirs)
579 return AVERROR_EOF;
580
581 if ((ret = ff_inlink_consume_samples(inlink, s->size, s->size, &in)) > 0) {
582 ret = headphone_frame(s, in, outlink);
583 if (ret < 0)
584 return ret;
585 }
586
587 if (ret < 0)
588 return ret;
589
590 FF_FILTER_FORWARD_STATUS(inlink, outlink);
591 if (ff_outlink_frame_wanted(outlink))
592 ff_inlink_request_frame(inlink);
593
594 return 0;
595 }
596
597 static int query_formats(AVFilterContext *ctx)
598 {
599 struct HeadphoneContext *s = ctx->priv;
600 AVFilterFormats *formats = NULL;
601 AVFilterChannelLayouts *layouts = NULL;
602 AVFilterChannelLayouts *stereo_layout = NULL;
603 AVFilterChannelLayouts *hrir_layouts = NULL;
604 int ret, i;
605
606 ret = ff_add_format(&formats, AV_SAMPLE_FMT_FLT);
607 if (ret)
608 return ret;
609 ret = ff_set_common_formats(ctx, formats);
610 if (ret)
611 return ret;
612
613 layouts = ff_all_channel_layouts();
614 if (!layouts)
615 return AVERROR(ENOMEM);
616
617 ret = ff_channel_layouts_ref(layouts, &ctx->inputs[0]->outcfg.channel_layouts);
618 if (ret)
619 return ret;
620
621 ret = ff_add_channel_layout(&stereo_layout, &(AVChannelLayout)AV_CHANNEL_LAYOUT_STEREO);
622 if (ret)
623 return ret;
624 ret = ff_channel_layouts_ref(stereo_layout, &ctx->outputs[0]->incfg.channel_layouts);
625 if (ret)
626 return ret;
627
628 if (s->hrir_fmt == HRIR_MULTI) {
629 hrir_layouts = ff_all_channel_counts();
630 if (!hrir_layouts)
631 return AVERROR(ENOMEM);
632 ret = ff_channel_layouts_ref(hrir_layouts, &ctx->inputs[1]->outcfg.channel_layouts);
633 if (ret)
634 return ret;
635 } else {
636 for (i = 1; i <= s->nb_hrir_inputs; i++) {
637 ret = ff_channel_layouts_ref(stereo_layout, &ctx->inputs[i]->outcfg.channel_layouts);
638 if (ret)
639 return ret;
640 }
641 }
642
643 return ff_set_common_all_samplerates(ctx);
644 }
645
646 static int config_input(AVFilterLink *inlink)
647 {
648 AVFilterContext *ctx = inlink->dst;
649 HeadphoneContext *s = ctx->priv;
650
651 if (s->nb_irs < inlink->ch_layout.nb_channels) {
652 av_log(ctx, AV_LOG_ERROR, "Number of HRIRs must be >= %d.\n", inlink->ch_layout.nb_channels);
653 return AVERROR(EINVAL);
654 }
655
656 s->lfe_channel = av_channel_layout_index_from_channel(&inlink->ch_layout,
657 AV_CHAN_LOW_FREQUENCY);
658 return 0;
659 }
660
661 static av_cold int init(AVFilterContext *ctx)
662 {
663 HeadphoneContext *s = ctx->priv;
664 int i, ret;
665
666 AVFilterPad pad = {
667 .name = "in0",
668 .type = AVMEDIA_TYPE_AUDIO,
669 .config_props = config_input,
670 };
671 if ((ret = ff_append_inpad(ctx, &pad)) < 0)
672 return ret;
673
674 if (!s->map) {
675 av_log(ctx, AV_LOG_ERROR, "Valid mapping must be set.\n");
676 return AVERROR(EINVAL);
677 }
678
679 parse_map(ctx);
680
681 for (i = 0; i < s->nb_hrir_inputs; i++) {
682 char *name = av_asprintf("hrir%d", i);
683 AVFilterPad pad = {
684 .name = name,
685 .type = AVMEDIA_TYPE_AUDIO,
686 };
687 if (!name)
688 return AVERROR(ENOMEM);
689 if ((ret = ff_append_inpad_free_name(ctx, &pad)) < 0)
690 return ret;
691 }
692
693 if (s->type == TIME_DOMAIN) {
694 AVFloatDSPContext *fdsp = avpriv_float_dsp_alloc(0);
695 if (!fdsp)
696 return AVERROR(ENOMEM);
697 s->scalarproduct_float = fdsp->scalarproduct_float;
698 av_free(fdsp);
699 }
700
701 return 0;
702 }
703
704 static int config_output(AVFilterLink *outlink)
705 {
706 AVFilterContext *ctx = outlink->src;
707 HeadphoneContext *s = ctx->priv;
708 AVFilterLink *inlink = ctx->inputs[0];
709
710 if (s->hrir_fmt == HRIR_MULTI) {
711 AVFilterLink *hrir_link = ctx->inputs[1];
712
713 if (hrir_link->ch_layout.nb_channels < inlink->ch_layout.nb_channels * 2) {
714 av_log(ctx, AV_LOG_ERROR, "Number of channels in HRIR stream must be >= %d.\n", inlink->ch_layout.nb_channels * 2);
715 return AVERROR(EINVAL);
716 }
717 }
718
719 s->gain_lfe = expf((s->gain - 3 * inlink->ch_layout.nb_channels + s->lfe_gain) / 20 * M_LN10);
720
721 return 0;
722 }
723
724 static av_cold void uninit(AVFilterContext *ctx)
725 {
726 HeadphoneContext *s = ctx->priv;
727
728 av_tx_uninit(&s->ifft[0]);
729 av_tx_uninit(&s->ifft[1]);
730 av_tx_uninit(&s->fft[0]);
731 av_tx_uninit(&s->fft[1]);
732 av_freep(&s->data_ir[0]);
733 av_freep(&s->data_ir[1]);
734 av_freep(&s->ringbuffer[0]);
735 av_freep(&s->ringbuffer[1]);
736 av_freep(&s->temp_src[0]);
737 av_freep(&s->temp_src[1]);
738 av_freep(&s->out_fft[0]);
739 av_freep(&s->out_fft[1]);
740 av_freep(&s->in_fft[0]);
741 av_freep(&s->in_fft[1]);
742 av_freep(&s->temp_afft[0]);
743 av_freep(&s->temp_afft[1]);
744 av_freep(&s->data_hrtf[0]);
745 av_freep(&s->data_hrtf[1]);
746 }
747
748 #define OFFSET(x) offsetof(HeadphoneContext, x)
749 #define FLAGS AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
750
751 static const AVOption headphone_options[] = {
752 { "map", "set channels convolution mappings", OFFSET(map), AV_OPT_TYPE_STRING, {.str=NULL}, .flags = FLAGS },
753 { "gain", "set gain in dB", OFFSET(gain), AV_OPT_TYPE_FLOAT, {.dbl=0}, -20, 40, .flags = FLAGS },
754 { "lfe", "set lfe gain in dB", OFFSET(lfe_gain), AV_OPT_TYPE_FLOAT, {.dbl=0}, -20, 40, .flags = FLAGS },
755 { "type", "set processing", OFFSET(type), AV_OPT_TYPE_INT, {.i64=1}, 0, 1, .flags = FLAGS, .unit = "type" },
756 { "time", "time domain", 0, AV_OPT_TYPE_CONST, {.i64=0}, 0, 0, .flags = FLAGS, .unit = "type" },
757 { "freq", "frequency domain", 0, AV_OPT_TYPE_CONST, {.i64=1}, 0, 0, .flags = FLAGS, .unit = "type" },
758 { "size", "set frame size", OFFSET(size), AV_OPT_TYPE_INT, {.i64=1024},1024,96000, .flags = FLAGS },
759 { "hrir", "set hrir format", OFFSET(hrir_fmt), AV_OPT_TYPE_INT, {.i64=HRIR_STEREO}, 0, 1, .flags = FLAGS, .unit = "hrir" },
760 { "stereo", "hrir files have exactly 2 channels", 0, AV_OPT_TYPE_CONST, {.i64=HRIR_STEREO}, 0, 0, .flags = FLAGS, .unit = "hrir" },
761 { "multich", "single multichannel hrir file", 0, AV_OPT_TYPE_CONST, {.i64=HRIR_MULTI}, 0, 0, .flags = FLAGS, .unit = "hrir" },
762 { NULL }
763 };
764
765 AVFILTER_DEFINE_CLASS(headphone);
766
767 static const AVFilterPad outputs[] = {
768 {
769 .name = "default",
770 .type = AVMEDIA_TYPE_AUDIO,
771 .config_props = config_output,
772 },
773 };
774
775 const AVFilter ff_af_headphone = {
776 .name = "headphone",
777 .description = NULL_IF_CONFIG_SMALL("Apply headphone binaural spatialization with HRTFs in additional streams."),
778 .priv_size = sizeof(HeadphoneContext),
779 .priv_class = &headphone_class,
780 .init = init,
781 .uninit = uninit,
782 .activate = activate,
783 .inputs = NULL,
784 FILTER_OUTPUTS(outputs),
785 FILTER_QUERY_FUNC(query_formats),
786 .flags = AVFILTER_FLAG_SLICE_THREADS | AVFILTER_FLAG_DYNAMIC_INPUTS,
787 };
788