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