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 "libavcodec/avfft.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 mapping[64];
56
57 int nb_inputs;
58
59 int nb_irs;
60
61 float gain;
62 float lfe_gain, gain_lfe;
63
64 float *ringbuffer[2];
65 int write[2];
66
67 int buffer_length;
68 int n_fft;
69 int size;
70 int hrir_fmt;
71
72 int *delay[2];
73 float *data_ir[2];
74 float *temp_src[2];
75 FFTComplex *temp_fft[2];
76 FFTComplex *temp_afft[2];
77
78 FFTContext *fft[2], *ifft[2];
79 FFTComplex *data_hrtf[2];
80
81 AVFloatDSPContext *fdsp;
82 struct headphone_inputs {
83 AVFrame *frame;
84 int ir_len;
85 int delay_l;
86 int delay_r;
87 int eof;
88 } *in;
89 } HeadphoneContext;
90
parse_channel_name(HeadphoneContext * s,int x,char ** arg,int * rchannel,char * buf)91 static int parse_channel_name(HeadphoneContext *s, int x, char **arg, int *rchannel, char *buf)
92 {
93 int len, i, channel_id = 0;
94 int64_t layout, layout0;
95
96 if (sscanf(*arg, "%7[A-Z]%n", buf, &len)) {
97 layout0 = layout = av_get_channel_layout(buf);
98 if (layout == AV_CH_LOW_FREQUENCY)
99 s->lfe_channel = x;
100 for (i = 32; i > 0; i >>= 1) {
101 if (layout >= 1LL << i) {
102 channel_id += i;
103 layout >>= i;
104 }
105 }
106 if (channel_id >= 64 || layout0 != 1LL << channel_id)
107 return AVERROR(EINVAL);
108 *rchannel = channel_id;
109 *arg += len;
110 return 0;
111 }
112 return AVERROR(EINVAL);
113 }
114
parse_map(AVFilterContext * ctx)115 static void parse_map(AVFilterContext *ctx)
116 {
117 HeadphoneContext *s = ctx->priv;
118 char *arg, *tokenizer, *p, *args = av_strdup(s->map);
119 int i;
120
121 if (!args)
122 return;
123 p = args;
124
125 s->lfe_channel = -1;
126 s->nb_inputs = 1;
127
128 for (i = 0; i < 64; i++) {
129 s->mapping[i] = -1;
130 }
131
132 while ((arg = av_strtok(p, "|", &tokenizer))) {
133 int out_ch_id;
134 char buf[8];
135
136 p = NULL;
137 if (parse_channel_name(s, s->nb_irs, &arg, &out_ch_id, buf)) {
138 av_log(ctx, AV_LOG_WARNING, "Failed to parse \'%s\' as channel name.\n", buf);
139 continue;
140 }
141 s->mapping[s->nb_irs] = out_ch_id;
142 s->nb_irs++;
143 }
144
145 if (s->hrir_fmt == HRIR_MULTI)
146 s->nb_inputs = 2;
147 else
148 s->nb_inputs = s->nb_irs + 1;
149
150 av_free(args);
151 }
152
153 typedef struct ThreadData {
154 AVFrame *in, *out;
155 int *write;
156 int **delay;
157 float **ir;
158 int *n_clippings;
159 float **ringbuffer;
160 float **temp_src;
161 FFTComplex **temp_fft;
162 FFTComplex **temp_afft;
163 } ThreadData;
164
headphone_convolute(AVFilterContext * ctx,void * arg,int jobnr,int nb_jobs)165 static int headphone_convolute(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
166 {
167 HeadphoneContext *s = ctx->priv;
168 ThreadData *td = arg;
169 AVFrame *in = td->in, *out = td->out;
170 int offset = jobnr;
171 int *write = &td->write[jobnr];
172 const int *const delay = td->delay[jobnr];
173 const float *const ir = td->ir[jobnr];
174 int *n_clippings = &td->n_clippings[jobnr];
175 float *ringbuffer = td->ringbuffer[jobnr];
176 float *temp_src = td->temp_src[jobnr];
177 const int ir_len = s->ir_len;
178 const int air_len = s->air_len;
179 const float *src = (const float *)in->data[0];
180 float *dst = (float *)out->data[0];
181 const int in_channels = in->channels;
182 const int buffer_length = s->buffer_length;
183 const uint32_t modulo = (uint32_t)buffer_length - 1;
184 float *buffer[16];
185 int wr = *write;
186 int read;
187 int i, l;
188
189 dst += offset;
190 for (l = 0; l < in_channels; l++) {
191 buffer[l] = ringbuffer + l * buffer_length;
192 }
193
194 for (i = 0; i < in->nb_samples; i++) {
195 const float *temp_ir = ir;
196
197 *dst = 0;
198 for (l = 0; l < in_channels; l++) {
199 *(buffer[l] + wr) = src[l];
200 }
201
202 for (l = 0; l < in_channels; l++) {
203 const float *const bptr = buffer[l];
204
205 if (l == s->lfe_channel) {
206 *dst += *(buffer[s->lfe_channel] + wr) * s->gain_lfe;
207 temp_ir += air_len;
208 continue;
209 }
210
211 read = (wr - *(delay + l) - (ir_len - 1) + buffer_length) & modulo;
212
213 if (read + ir_len < buffer_length) {
214 memcpy(temp_src, bptr + read, ir_len * sizeof(*temp_src));
215 } else {
216 int len = FFMIN(air_len - (read % ir_len), buffer_length - read);
217
218 memcpy(temp_src, bptr + read, len * sizeof(*temp_src));
219 memcpy(temp_src + len, bptr, (air_len - len) * sizeof(*temp_src));
220 }
221
222 dst[0] += s->fdsp->scalarproduct_float(temp_ir, temp_src, FFALIGN(ir_len, 32));
223 temp_ir += air_len;
224 }
225
226 if (fabsf(dst[0]) > 1)
227 n_clippings[0]++;
228
229 dst += 2;
230 src += in_channels;
231 wr = (wr + 1) & modulo;
232 }
233
234 *write = wr;
235
236 return 0;
237 }
238
headphone_fast_convolute(AVFilterContext * ctx,void * arg,int jobnr,int nb_jobs)239 static int headphone_fast_convolute(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
240 {
241 HeadphoneContext *s = ctx->priv;
242 ThreadData *td = arg;
243 AVFrame *in = td->in, *out = td->out;
244 int offset = jobnr;
245 int *write = &td->write[jobnr];
246 FFTComplex *hrtf = s->data_hrtf[jobnr];
247 int *n_clippings = &td->n_clippings[jobnr];
248 float *ringbuffer = td->ringbuffer[jobnr];
249 const int ir_len = s->ir_len;
250 const float *src = (const float *)in->data[0];
251 float *dst = (float *)out->data[0];
252 const int in_channels = in->channels;
253 const int buffer_length = s->buffer_length;
254 const uint32_t modulo = (uint32_t)buffer_length - 1;
255 FFTComplex *fft_in = s->temp_fft[jobnr];
256 FFTComplex *fft_acc = s->temp_afft[jobnr];
257 FFTContext *ifft = s->ifft[jobnr];
258 FFTContext *fft = s->fft[jobnr];
259 const int n_fft = s->n_fft;
260 const float fft_scale = 1.0f / s->n_fft;
261 FFTComplex *hrtf_offset;
262 int wr = *write;
263 int n_read;
264 int i, j;
265
266 dst += offset;
267
268 n_read = FFMIN(ir_len, in->nb_samples);
269 for (j = 0; j < n_read; j++) {
270 dst[2 * j] = ringbuffer[wr];
271 ringbuffer[wr] = 0.0;
272 wr = (wr + 1) & modulo;
273 }
274
275 for (j = n_read; j < in->nb_samples; j++) {
276 dst[2 * j] = 0;
277 }
278
279 memset(fft_acc, 0, sizeof(FFTComplex) * n_fft);
280
281 for (i = 0; i < in_channels; i++) {
282 if (i == s->lfe_channel) {
283 for (j = 0; j < in->nb_samples; j++) {
284 dst[2 * j] += src[i + j * in_channels] * s->gain_lfe;
285 }
286 continue;
287 }
288
289 offset = i * n_fft;
290 hrtf_offset = hrtf + offset;
291
292 memset(fft_in, 0, sizeof(FFTComplex) * n_fft);
293
294 for (j = 0; j < in->nb_samples; j++) {
295 fft_in[j].re = src[j * in_channels + i];
296 }
297
298 av_fft_permute(fft, fft_in);
299 av_fft_calc(fft, fft_in);
300 for (j = 0; j < n_fft; j++) {
301 const FFTComplex *hcomplex = hrtf_offset + j;
302 const float re = fft_in[j].re;
303 const float im = fft_in[j].im;
304
305 fft_acc[j].re += re * hcomplex->re - im * hcomplex->im;
306 fft_acc[j].im += re * hcomplex->im + im * hcomplex->re;
307 }
308 }
309
310 av_fft_permute(ifft, fft_acc);
311 av_fft_calc(ifft, fft_acc);
312
313 for (j = 0; j < in->nb_samples; j++) {
314 dst[2 * j] += fft_acc[j].re * fft_scale;
315 }
316
317 for (j = 0; j < ir_len - 1; j++) {
318 int write_pos = (wr + j) & modulo;
319
320 *(ringbuffer + write_pos) += fft_acc[in->nb_samples + j].re * fft_scale;
321 }
322
323 for (i = 0; i < out->nb_samples; i++) {
324 if (fabsf(dst[0]) > 1) {
325 n_clippings[0]++;
326 }
327
328 dst += 2;
329 }
330
331 *write = wr;
332
333 return 0;
334 }
335
check_ir(AVFilterLink * inlink,int input_number)336 static int check_ir(AVFilterLink *inlink, int input_number)
337 {
338 AVFilterContext *ctx = inlink->dst;
339 HeadphoneContext *s = ctx->priv;
340 int ir_len, max_ir_len;
341
342 ir_len = ff_inlink_queued_samples(inlink);
343 max_ir_len = 65536;
344 if (ir_len > max_ir_len) {
345 av_log(ctx, AV_LOG_ERROR, "Too big length of IRs: %d > %d.\n", ir_len, max_ir_len);
346 return AVERROR(EINVAL);
347 }
348 s->in[input_number].ir_len = ir_len;
349 s->ir_len = FFMAX(ir_len, s->ir_len);
350
351 return 0;
352 }
353
headphone_frame(HeadphoneContext * s,AVFrame * in,AVFilterLink * outlink)354 static int headphone_frame(HeadphoneContext *s, AVFrame *in, AVFilterLink *outlink)
355 {
356 AVFilterContext *ctx = outlink->src;
357 int n_clippings[2] = { 0 };
358 ThreadData td;
359 AVFrame *out;
360
361 out = ff_get_audio_buffer(outlink, in->nb_samples);
362 if (!out) {
363 av_frame_free(&in);
364 return AVERROR(ENOMEM);
365 }
366 out->pts = in->pts;
367
368 td.in = in; td.out = out; td.write = s->write;
369 td.delay = s->delay; td.ir = s->data_ir; td.n_clippings = n_clippings;
370 td.ringbuffer = s->ringbuffer; td.temp_src = s->temp_src;
371 td.temp_fft = s->temp_fft;
372 td.temp_afft = s->temp_afft;
373
374 if (s->type == TIME_DOMAIN) {
375 ctx->internal->execute(ctx, headphone_convolute, &td, NULL, 2);
376 } else {
377 ctx->internal->execute(ctx, headphone_fast_convolute, &td, NULL, 2);
378 }
379 emms_c();
380
381 if (n_clippings[0] + n_clippings[1] > 0) {
382 av_log(ctx, AV_LOG_WARNING, "%d of %d samples clipped. Please reduce gain.\n",
383 n_clippings[0] + n_clippings[1], out->nb_samples * 2);
384 }
385
386 av_frame_free(&in);
387 return ff_filter_frame(outlink, out);
388 }
389
convert_coeffs(AVFilterContext * ctx,AVFilterLink * inlink)390 static int convert_coeffs(AVFilterContext *ctx, AVFilterLink *inlink)
391 {
392 struct HeadphoneContext *s = ctx->priv;
393 const int ir_len = s->ir_len;
394 int nb_irs = s->nb_irs;
395 int nb_input_channels = ctx->inputs[0]->channels;
396 float gain_lin = expf((s->gain - 3 * nb_input_channels) / 20 * M_LN10);
397 FFTComplex *data_hrtf_l = NULL;
398 FFTComplex *data_hrtf_r = NULL;
399 FFTComplex *fft_in_l = NULL;
400 FFTComplex *fft_in_r = NULL;
401 float *data_ir_l = NULL;
402 float *data_ir_r = NULL;
403 int offset = 0, ret = 0;
404 int n_fft;
405 int i, j, k;
406
407 s->air_len = 1 << (32 - ff_clz(ir_len));
408 s->buffer_length = 1 << (32 - ff_clz(s->air_len));
409 s->n_fft = n_fft = 1 << (32 - ff_clz(ir_len + s->size));
410
411 if (s->type == FREQUENCY_DOMAIN) {
412 fft_in_l = av_calloc(n_fft, sizeof(*fft_in_l));
413 fft_in_r = av_calloc(n_fft, sizeof(*fft_in_r));
414 if (!fft_in_l || !fft_in_r) {
415 ret = AVERROR(ENOMEM);
416 goto fail;
417 }
418
419 av_fft_end(s->fft[0]);
420 av_fft_end(s->fft[1]);
421 s->fft[0] = av_fft_init(av_log2(s->n_fft), 0);
422 s->fft[1] = av_fft_init(av_log2(s->n_fft), 0);
423 av_fft_end(s->ifft[0]);
424 av_fft_end(s->ifft[1]);
425 s->ifft[0] = av_fft_init(av_log2(s->n_fft), 1);
426 s->ifft[1] = av_fft_init(av_log2(s->n_fft), 1);
427
428 if (!s->fft[0] || !s->fft[1] || !s->ifft[0] || !s->ifft[1]) {
429 av_log(ctx, AV_LOG_ERROR, "Unable to create FFT contexts of size %d.\n", s->n_fft);
430 ret = AVERROR(ENOMEM);
431 goto fail;
432 }
433 }
434
435 s->data_ir[0] = av_calloc(s->air_len, sizeof(float) * s->nb_irs);
436 s->data_ir[1] = av_calloc(s->air_len, sizeof(float) * s->nb_irs);
437 s->delay[0] = av_calloc(s->nb_irs, sizeof(float));
438 s->delay[1] = av_calloc(s->nb_irs, sizeof(float));
439
440 if (s->type == TIME_DOMAIN) {
441 s->ringbuffer[0] = av_calloc(s->buffer_length, sizeof(float) * nb_input_channels);
442 s->ringbuffer[1] = av_calloc(s->buffer_length, sizeof(float) * nb_input_channels);
443 } else {
444 s->ringbuffer[0] = av_calloc(s->buffer_length, sizeof(float));
445 s->ringbuffer[1] = av_calloc(s->buffer_length, sizeof(float));
446 s->temp_fft[0] = av_calloc(s->n_fft, sizeof(FFTComplex));
447 s->temp_fft[1] = av_calloc(s->n_fft, sizeof(FFTComplex));
448 s->temp_afft[0] = av_calloc(s->n_fft, sizeof(FFTComplex));
449 s->temp_afft[1] = av_calloc(s->n_fft, sizeof(FFTComplex));
450 if (!s->temp_fft[0] || !s->temp_fft[1] ||
451 !s->temp_afft[0] || !s->temp_afft[1]) {
452 ret = AVERROR(ENOMEM);
453 goto fail;
454 }
455 }
456
457 if (!s->data_ir[0] || !s->data_ir[1] ||
458 !s->ringbuffer[0] || !s->ringbuffer[1]) {
459 ret = AVERROR(ENOMEM);
460 goto fail;
461 }
462
463 if (s->type == TIME_DOMAIN) {
464 s->temp_src[0] = av_calloc(s->air_len, sizeof(float));
465 s->temp_src[1] = av_calloc(s->air_len, sizeof(float));
466
467 data_ir_l = av_calloc(nb_irs * s->air_len, sizeof(*data_ir_l));
468 data_ir_r = av_calloc(nb_irs * s->air_len, sizeof(*data_ir_r));
469 if (!data_ir_r || !data_ir_l || !s->temp_src[0] || !s->temp_src[1]) {
470 ret = AVERROR(ENOMEM);
471 goto fail;
472 }
473 } else {
474 data_hrtf_l = av_calloc(n_fft, sizeof(*data_hrtf_l) * nb_irs);
475 data_hrtf_r = av_calloc(n_fft, sizeof(*data_hrtf_r) * nb_irs);
476 if (!data_hrtf_r || !data_hrtf_l) {
477 ret = AVERROR(ENOMEM);
478 goto fail;
479 }
480 }
481
482 for (i = 0; i < s->nb_inputs - 1; i++) {
483 int len = s->in[i + 1].ir_len;
484 int delay_l = s->in[i + 1].delay_l;
485 int delay_r = s->in[i + 1].delay_r;
486 float *ptr;
487
488 ret = ff_inlink_consume_samples(ctx->inputs[i + 1], len, len, &s->in[i + 1].frame);
489 if (ret < 0)
490 goto fail;
491 ptr = (float *)s->in[i + 1].frame->extended_data[0];
492
493 if (s->hrir_fmt == HRIR_STEREO) {
494 int idx = -1;
495
496 for (j = 0; j < inlink->channels; j++) {
497 if (s->mapping[i] < 0) {
498 continue;
499 }
500
501 if ((av_channel_layout_extract_channel(inlink->channel_layout, j)) == (1LL << s->mapping[i])) {
502 idx = i;
503 break;
504 }
505 }
506
507 if (idx == -1)
508 continue;
509 if (s->type == TIME_DOMAIN) {
510 offset = idx * s->air_len;
511 for (j = 0; j < len; j++) {
512 data_ir_l[offset + j] = ptr[len * 2 - j * 2 - 2] * gain_lin;
513 data_ir_r[offset + j] = ptr[len * 2 - j * 2 - 1] * gain_lin;
514 }
515 } else {
516 memset(fft_in_l, 0, n_fft * sizeof(*fft_in_l));
517 memset(fft_in_r, 0, n_fft * sizeof(*fft_in_r));
518
519 offset = idx * n_fft;
520 for (j = 0; j < len; j++) {
521 fft_in_l[delay_l + j].re = ptr[j * 2 ] * gain_lin;
522 fft_in_r[delay_r + j].re = ptr[j * 2 + 1] * gain_lin;
523 }
524
525 av_fft_permute(s->fft[0], fft_in_l);
526 av_fft_calc(s->fft[0], fft_in_l);
527 memcpy(data_hrtf_l + offset, fft_in_l, n_fft * sizeof(*fft_in_l));
528 av_fft_permute(s->fft[0], fft_in_r);
529 av_fft_calc(s->fft[0], fft_in_r);
530 memcpy(data_hrtf_r + offset, fft_in_r, n_fft * sizeof(*fft_in_r));
531 }
532 } else {
533 int I, N = ctx->inputs[1]->channels;
534
535 for (k = 0; k < N / 2; k++) {
536 int idx = -1;
537
538 for (j = 0; j < inlink->channels; j++) {
539 if (s->mapping[k] < 0) {
540 continue;
541 }
542
543 if ((av_channel_layout_extract_channel(inlink->channel_layout, j)) == (1LL << s->mapping[k])) {
544 idx = k;
545 break;
546 }
547 }
548 if (idx == -1)
549 continue;
550
551 I = idx * 2;
552 if (s->type == TIME_DOMAIN) {
553 offset = idx * s->air_len;
554 for (j = 0; j < len; j++) {
555 data_ir_l[offset + j] = ptr[len * N - j * N - N + I ] * gain_lin;
556 data_ir_r[offset + j] = ptr[len * N - j * N - N + I + 1] * gain_lin;
557 }
558 } else {
559 memset(fft_in_l, 0, n_fft * sizeof(*fft_in_l));
560 memset(fft_in_r, 0, n_fft * sizeof(*fft_in_r));
561
562 offset = idx * n_fft;
563 for (j = 0; j < len; j++) {
564 fft_in_l[delay_l + j].re = ptr[j * N + I ] * gain_lin;
565 fft_in_r[delay_r + j].re = ptr[j * N + I + 1] * gain_lin;
566 }
567
568 av_fft_permute(s->fft[0], fft_in_l);
569 av_fft_calc(s->fft[0], fft_in_l);
570 memcpy(data_hrtf_l + offset, fft_in_l, n_fft * sizeof(*fft_in_l));
571 av_fft_permute(s->fft[0], fft_in_r);
572 av_fft_calc(s->fft[0], fft_in_r);
573 memcpy(data_hrtf_r + offset, fft_in_r, n_fft * sizeof(*fft_in_r));
574 }
575 }
576 }
577
578 av_frame_free(&s->in[i + 1].frame);
579 }
580
581 if (s->type == TIME_DOMAIN) {
582 memcpy(s->data_ir[0], data_ir_l, sizeof(float) * nb_irs * s->air_len);
583 memcpy(s->data_ir[1], data_ir_r, sizeof(float) * nb_irs * s->air_len);
584 } else {
585 s->data_hrtf[0] = av_calloc(n_fft * s->nb_irs, sizeof(FFTComplex));
586 s->data_hrtf[1] = av_calloc(n_fft * s->nb_irs, sizeof(FFTComplex));
587 if (!s->data_hrtf[0] || !s->data_hrtf[1]) {
588 ret = AVERROR(ENOMEM);
589 goto fail;
590 }
591
592 memcpy(s->data_hrtf[0], data_hrtf_l,
593 sizeof(FFTComplex) * nb_irs * n_fft);
594 memcpy(s->data_hrtf[1], data_hrtf_r,
595 sizeof(FFTComplex) * nb_irs * n_fft);
596 }
597
598 s->have_hrirs = 1;
599
600 fail:
601
602 for (i = 0; i < s->nb_inputs - 1; i++)
603 av_frame_free(&s->in[i + 1].frame);
604
605 av_freep(&data_ir_l);
606 av_freep(&data_ir_r);
607
608 av_freep(&data_hrtf_l);
609 av_freep(&data_hrtf_r);
610
611 av_freep(&fft_in_l);
612 av_freep(&fft_in_r);
613
614 return ret;
615 }
616
activate(AVFilterContext * ctx)617 static int activate(AVFilterContext *ctx)
618 {
619 HeadphoneContext *s = ctx->priv;
620 AVFilterLink *inlink = ctx->inputs[0];
621 AVFilterLink *outlink = ctx->outputs[0];
622 AVFrame *in = NULL;
623 int i, ret;
624
625 FF_FILTER_FORWARD_STATUS_BACK_ALL(ctx->outputs[0], ctx);
626 if (!s->eof_hrirs) {
627 for (i = 1; i < s->nb_inputs; i++) {
628 if (s->in[i].eof)
629 continue;
630
631 if ((ret = check_ir(ctx->inputs[i], i)) < 0)
632 return ret;
633
634 if (!s->in[i].eof) {
635 if (ff_outlink_get_status(ctx->inputs[i]) == AVERROR_EOF)
636 s->in[i].eof = 1;
637 }
638 }
639
640 for (i = 1; i < s->nb_inputs; i++) {
641 if (!s->in[i].eof)
642 break;
643 }
644
645 if (i != s->nb_inputs) {
646 if (ff_outlink_frame_wanted(ctx->outputs[0])) {
647 for (i = 1; i < s->nb_inputs; i++) {
648 if (!s->in[i].eof)
649 ff_inlink_request_frame(ctx->inputs[i]);
650 }
651 }
652
653 return 0;
654 } else {
655 s->eof_hrirs = 1;
656 }
657 }
658
659 if (!s->have_hrirs && s->eof_hrirs) {
660 ret = convert_coeffs(ctx, inlink);
661 if (ret < 0)
662 return ret;
663 }
664
665 if ((ret = ff_inlink_consume_samples(ctx->inputs[0], s->size, s->size, &in)) > 0) {
666 ret = headphone_frame(s, in, outlink);
667 if (ret < 0)
668 return ret;
669 }
670
671 if (ret < 0)
672 return ret;
673
674 FF_FILTER_FORWARD_STATUS(ctx->inputs[0], ctx->outputs[0]);
675 if (ff_outlink_frame_wanted(ctx->outputs[0]))
676 ff_inlink_request_frame(ctx->inputs[0]);
677
678 return 0;
679 }
680
query_formats(AVFilterContext * ctx)681 static int query_formats(AVFilterContext *ctx)
682 {
683 struct HeadphoneContext *s = ctx->priv;
684 AVFilterFormats *formats = NULL;
685 AVFilterChannelLayouts *layouts = NULL;
686 AVFilterChannelLayouts *stereo_layout = NULL;
687 AVFilterChannelLayouts *hrir_layouts = NULL;
688 int ret, i;
689
690 ret = ff_add_format(&formats, AV_SAMPLE_FMT_FLT);
691 if (ret)
692 return ret;
693 ret = ff_set_common_formats(ctx, formats);
694 if (ret)
695 return ret;
696
697 layouts = ff_all_channel_layouts();
698 if (!layouts)
699 return AVERROR(ENOMEM);
700
701 ret = ff_channel_layouts_ref(layouts, &ctx->inputs[0]->out_channel_layouts);
702 if (ret)
703 return ret;
704
705 ret = ff_add_channel_layout(&stereo_layout, AV_CH_LAYOUT_STEREO);
706 if (ret)
707 return ret;
708
709 if (s->hrir_fmt == HRIR_MULTI) {
710 hrir_layouts = ff_all_channel_counts();
711 if (!hrir_layouts)
712 return AVERROR(ENOMEM);
713 ret = ff_channel_layouts_ref(hrir_layouts, &ctx->inputs[1]->out_channel_layouts);
714 if (ret)
715 return ret;
716 } else {
717 for (i = 1; i < s->nb_inputs; i++) {
718 ret = ff_channel_layouts_ref(stereo_layout, &ctx->inputs[i]->out_channel_layouts);
719 if (ret)
720 return ret;
721 }
722 }
723
724 ret = ff_channel_layouts_ref(stereo_layout, &ctx->outputs[0]->in_channel_layouts);
725 if (ret)
726 return ret;
727
728 formats = ff_all_samplerates();
729 if (!formats)
730 return AVERROR(ENOMEM);
731 return ff_set_common_samplerates(ctx, formats);
732 }
733
config_input(AVFilterLink * inlink)734 static int config_input(AVFilterLink *inlink)
735 {
736 AVFilterContext *ctx = inlink->dst;
737 HeadphoneContext *s = ctx->priv;
738
739 if (s->nb_irs < inlink->channels) {
740 av_log(ctx, AV_LOG_ERROR, "Number of HRIRs must be >= %d.\n", inlink->channels);
741 return AVERROR(EINVAL);
742 }
743
744 return 0;
745 }
746
init(AVFilterContext * ctx)747 static av_cold int init(AVFilterContext *ctx)
748 {
749 HeadphoneContext *s = ctx->priv;
750 int i, ret;
751
752 AVFilterPad pad = {
753 .name = "in0",
754 .type = AVMEDIA_TYPE_AUDIO,
755 .config_props = config_input,
756 };
757 if ((ret = ff_insert_inpad(ctx, 0, &pad)) < 0)
758 return ret;
759
760 if (!s->map) {
761 av_log(ctx, AV_LOG_ERROR, "Valid mapping must be set.\n");
762 return AVERROR(EINVAL);
763 }
764
765 parse_map(ctx);
766
767 s->in = av_calloc(s->nb_inputs, sizeof(*s->in));
768 if (!s->in)
769 return AVERROR(ENOMEM);
770
771 for (i = 1; i < s->nb_inputs; i++) {
772 char *name = av_asprintf("hrir%d", i - 1);
773 AVFilterPad pad = {
774 .name = name,
775 .type = AVMEDIA_TYPE_AUDIO,
776 };
777 if (!name)
778 return AVERROR(ENOMEM);
779 if ((ret = ff_insert_inpad(ctx, i, &pad)) < 0) {
780 av_freep(&pad.name);
781 return ret;
782 }
783 }
784
785 s->fdsp = avpriv_float_dsp_alloc(0);
786 if (!s->fdsp)
787 return AVERROR(ENOMEM);
788
789 return 0;
790 }
791
config_output(AVFilterLink * outlink)792 static int config_output(AVFilterLink *outlink)
793 {
794 AVFilterContext *ctx = outlink->src;
795 HeadphoneContext *s = ctx->priv;
796 AVFilterLink *inlink = ctx->inputs[0];
797
798 if (s->hrir_fmt == HRIR_MULTI) {
799 AVFilterLink *hrir_link = ctx->inputs[1];
800
801 if (hrir_link->channels < inlink->channels * 2) {
802 av_log(ctx, AV_LOG_ERROR, "Number of channels in HRIR stream must be >= %d.\n", inlink->channels * 2);
803 return AVERROR(EINVAL);
804 }
805 }
806
807 s->gain_lfe = expf((s->gain - 3 * inlink->channels + s->lfe_gain) / 20 * M_LN10);
808
809 return 0;
810 }
811
uninit(AVFilterContext * ctx)812 static av_cold void uninit(AVFilterContext *ctx)
813 {
814 HeadphoneContext *s = ctx->priv;
815 int i;
816
817 av_fft_end(s->ifft[0]);
818 av_fft_end(s->ifft[1]);
819 av_fft_end(s->fft[0]);
820 av_fft_end(s->fft[1]);
821 av_freep(&s->delay[0]);
822 av_freep(&s->delay[1]);
823 av_freep(&s->data_ir[0]);
824 av_freep(&s->data_ir[1]);
825 av_freep(&s->ringbuffer[0]);
826 av_freep(&s->ringbuffer[1]);
827 av_freep(&s->temp_src[0]);
828 av_freep(&s->temp_src[1]);
829 av_freep(&s->temp_fft[0]);
830 av_freep(&s->temp_fft[1]);
831 av_freep(&s->temp_afft[0]);
832 av_freep(&s->temp_afft[1]);
833 av_freep(&s->data_hrtf[0]);
834 av_freep(&s->data_hrtf[1]);
835 av_freep(&s->fdsp);
836
837 for (i = 0; i < s->nb_inputs; i++) {
838 if (ctx->input_pads && i)
839 av_freep(&ctx->input_pads[i].name);
840 }
841 av_freep(&s->in);
842 }
843
844 #define OFFSET(x) offsetof(HeadphoneContext, x)
845 #define FLAGS AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
846
847 static const AVOption headphone_options[] = {
848 { "map", "set channels convolution mappings", OFFSET(map), AV_OPT_TYPE_STRING, {.str=NULL}, .flags = FLAGS },
849 { "gain", "set gain in dB", OFFSET(gain), AV_OPT_TYPE_FLOAT, {.dbl=0}, -20, 40, .flags = FLAGS },
850 { "lfe", "set lfe gain in dB", OFFSET(lfe_gain), AV_OPT_TYPE_FLOAT, {.dbl=0}, -20, 40, .flags = FLAGS },
851 { "type", "set processing", OFFSET(type), AV_OPT_TYPE_INT, {.i64=1}, 0, 1, .flags = FLAGS, "type" },
852 { "time", "time domain", 0, AV_OPT_TYPE_CONST, {.i64=0}, 0, 0, .flags = FLAGS, "type" },
853 { "freq", "frequency domain", 0, AV_OPT_TYPE_CONST, {.i64=1}, 0, 0, .flags = FLAGS, "type" },
854 { "size", "set frame size", OFFSET(size), AV_OPT_TYPE_INT, {.i64=1024},1024,96000, .flags = FLAGS },
855 { "hrir", "set hrir format", OFFSET(hrir_fmt), AV_OPT_TYPE_INT, {.i64=HRIR_STEREO}, 0, 1, .flags = FLAGS, "hrir" },
856 { "stereo", "hrir files have exactly 2 channels", 0, AV_OPT_TYPE_CONST, {.i64=HRIR_STEREO}, 0, 0, .flags = FLAGS, "hrir" },
857 { "multich", "single multichannel hrir file", 0, AV_OPT_TYPE_CONST, {.i64=HRIR_MULTI}, 0, 0, .flags = FLAGS, "hrir" },
858 { NULL }
859 };
860
861 AVFILTER_DEFINE_CLASS(headphone);
862
863 static const AVFilterPad outputs[] = {
864 {
865 .name = "default",
866 .type = AVMEDIA_TYPE_AUDIO,
867 .config_props = config_output,
868 },
869 { NULL }
870 };
871
872 AVFilter ff_af_headphone = {
873 .name = "headphone",
874 .description = NULL_IF_CONFIG_SMALL("Apply headphone binaural spatialization with HRTFs in additional streams."),
875 .priv_size = sizeof(HeadphoneContext),
876 .priv_class = &headphone_class,
877 .init = init,
878 .uninit = uninit,
879 .query_formats = query_formats,
880 .activate = activate,
881 .inputs = NULL,
882 .outputs = outputs,
883 .flags = AVFILTER_FLAG_SLICE_THREADS | AVFILTER_FLAG_DYNAMIC_INPUTS,
884 };
885