• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 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     FFTComplex *temp_fft[2];
73     FFTComplex *temp_afft[2];
74 
75     FFTContext *fft[2], *ifft[2];
76     FFTComplex *data_hrtf[2];
77 
78     float (*scalarproduct_float)(const float *v1, const float *v2, int len);
79     struct hrir_inputs {
80         int          ir_len;
81         int          eof;
82     } hrir_in[64];
83     uint64_t mapping[64];
84 } HeadphoneContext;
85 
parse_channel_name(const char * arg,uint64_t * rchannel)86 static int parse_channel_name(const char *arg, uint64_t *rchannel)
87 {
88     uint64_t layout = av_get_channel_layout(arg);
89 
90     if (av_get_channel_layout_nb_channels(layout) != 1)
91         return AVERROR(EINVAL);
92     *rchannel = layout;
93     return 0;
94 }
95 
parse_map(AVFilterContext * ctx)96 static void parse_map(AVFilterContext *ctx)
97 {
98     HeadphoneContext *s = ctx->priv;
99     char *arg, *tokenizer, *p;
100     uint64_t used_channels = 0;
101 
102     p = s->map;
103     while ((arg = av_strtok(p, "|", &tokenizer))) {
104         uint64_t out_channel;
105 
106         p = NULL;
107         if (parse_channel_name(arg, &out_channel)) {
108             av_log(ctx, AV_LOG_WARNING, "Failed to parse \'%s\' as channel name.\n", arg);
109             continue;
110         }
111         if (used_channels & out_channel) {
112             av_log(ctx, AV_LOG_WARNING, "Ignoring duplicate channel '%s'.\n", arg);
113             continue;
114         }
115         used_channels        |= out_channel;
116         s->mapping[s->nb_irs] = out_channel;
117         s->nb_irs++;
118     }
119 
120     if (s->hrir_fmt == HRIR_MULTI)
121         s->nb_hrir_inputs = 1;
122     else
123         s->nb_hrir_inputs = s->nb_irs;
124 }
125 
126 typedef struct ThreadData {
127     AVFrame *in, *out;
128     int *write;
129     float **ir;
130     int *n_clippings;
131     float **ringbuffer;
132     float **temp_src;
133     FFTComplex **temp_fft;
134     FFTComplex **temp_afft;
135 } ThreadData;
136 
headphone_convolute(AVFilterContext * ctx,void * arg,int jobnr,int nb_jobs)137 static int headphone_convolute(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
138 {
139     HeadphoneContext *s = ctx->priv;
140     ThreadData *td = arg;
141     AVFrame *in = td->in, *out = td->out;
142     int offset = jobnr;
143     int *write = &td->write[jobnr];
144     const float *const ir = td->ir[jobnr];
145     int *n_clippings = &td->n_clippings[jobnr];
146     float *ringbuffer = td->ringbuffer[jobnr];
147     float *temp_src = td->temp_src[jobnr];
148     const int ir_len = s->ir_len;
149     const int air_len = s->air_len;
150     const float *src = (const float *)in->data[0];
151     float *dst = (float *)out->data[0];
152     const int in_channels = in->channels;
153     const int buffer_length = s->buffer_length;
154     const uint32_t modulo = (uint32_t)buffer_length - 1;
155     float *buffer[64];
156     int wr = *write;
157     int read;
158     int i, l;
159 
160     dst += offset;
161     for (l = 0; l < in_channels; l++) {
162         buffer[l] = ringbuffer + l * buffer_length;
163     }
164 
165     for (i = 0; i < in->nb_samples; i++) {
166         const float *cur_ir = ir;
167 
168         *dst = 0;
169         for (l = 0; l < in_channels; l++) {
170             *(buffer[l] + wr) = src[l];
171         }
172 
173         for (l = 0; l < in_channels; cur_ir += air_len, l++) {
174             const float *const bptr = buffer[l];
175 
176             if (l == s->lfe_channel) {
177                 *dst += *(buffer[s->lfe_channel] + wr) * s->gain_lfe;
178                 continue;
179             }
180 
181             read = (wr - (ir_len - 1)) & modulo;
182 
183             if (read + ir_len < buffer_length) {
184                 memcpy(temp_src, bptr + read, ir_len * sizeof(*temp_src));
185             } else {
186                 int len = FFMIN(air_len - (read % ir_len), buffer_length - read);
187 
188                 memcpy(temp_src, bptr + read, len * sizeof(*temp_src));
189                 memcpy(temp_src + len, bptr, (air_len - len) * sizeof(*temp_src));
190             }
191 
192             dst[0] += s->scalarproduct_float(cur_ir, temp_src, FFALIGN(ir_len, 32));
193         }
194 
195         if (fabsf(dst[0]) > 1)
196             n_clippings[0]++;
197 
198         dst += 2;
199         src += in_channels;
200         wr   = (wr + 1) & modulo;
201     }
202 
203     *write = wr;
204 
205     return 0;
206 }
207 
headphone_fast_convolute(AVFilterContext * ctx,void * arg,int jobnr,int nb_jobs)208 static int headphone_fast_convolute(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
209 {
210     HeadphoneContext *s = ctx->priv;
211     ThreadData *td = arg;
212     AVFrame *in = td->in, *out = td->out;
213     int offset = jobnr;
214     int *write = &td->write[jobnr];
215     FFTComplex *hrtf = s->data_hrtf[jobnr];
216     int *n_clippings = &td->n_clippings[jobnr];
217     float *ringbuffer = td->ringbuffer[jobnr];
218     const int ir_len = s->ir_len;
219     const float *src = (const float *)in->data[0];
220     float *dst = (float *)out->data[0];
221     const int in_channels = in->channels;
222     const int buffer_length = s->buffer_length;
223     const uint32_t modulo = (uint32_t)buffer_length - 1;
224     FFTComplex *fft_in = s->temp_fft[jobnr];
225     FFTComplex *fft_acc = s->temp_afft[jobnr];
226     FFTContext *ifft = s->ifft[jobnr];
227     FFTContext *fft = s->fft[jobnr];
228     const int n_fft = s->n_fft;
229     const float fft_scale = 1.0f / s->n_fft;
230     FFTComplex *hrtf_offset;
231     int wr = *write;
232     int n_read;
233     int i, j;
234 
235     dst += offset;
236 
237     n_read = FFMIN(ir_len, in->nb_samples);
238     for (j = 0; j < n_read; j++) {
239         dst[2 * j]     = ringbuffer[wr];
240         ringbuffer[wr] = 0.0;
241         wr  = (wr + 1) & modulo;
242     }
243 
244     for (j = n_read; j < in->nb_samples; j++) {
245         dst[2 * j] = 0;
246     }
247 
248     memset(fft_acc, 0, sizeof(FFTComplex) * n_fft);
249 
250     for (i = 0; i < in_channels; i++) {
251         if (i == s->lfe_channel) {
252             for (j = 0; j < in->nb_samples; j++) {
253                 dst[2 * j] += src[i + j * in_channels] * s->gain_lfe;
254             }
255             continue;
256         }
257 
258         offset = i * n_fft;
259         hrtf_offset = hrtf + offset;
260 
261         memset(fft_in, 0, sizeof(FFTComplex) * n_fft);
262 
263         for (j = 0; j < in->nb_samples; j++) {
264             fft_in[j].re = src[j * in_channels + i];
265         }
266 
267         av_fft_permute(fft, fft_in);
268         av_fft_calc(fft, fft_in);
269         for (j = 0; j < n_fft; j++) {
270             const FFTComplex *hcomplex = hrtf_offset + j;
271             const float re = fft_in[j].re;
272             const float im = fft_in[j].im;
273 
274             fft_acc[j].re += re * hcomplex->re - im * hcomplex->im;
275             fft_acc[j].im += re * hcomplex->im + im * hcomplex->re;
276         }
277     }
278 
279     av_fft_permute(ifft, fft_acc);
280     av_fft_calc(ifft, fft_acc);
281 
282     for (j = 0; j < in->nb_samples; j++) {
283         dst[2 * j] += fft_acc[j].re * fft_scale;
284         if (fabsf(dst[2 * j]) > 1)
285             n_clippings[0]++;
286     }
287 
288     for (j = 0; j < ir_len - 1; j++) {
289         int write_pos = (wr + j) & modulo;
290 
291         *(ringbuffer + write_pos) += fft_acc[in->nb_samples + j].re * fft_scale;
292     }
293 
294     *write = wr;
295 
296     return 0;
297 }
298 
check_ir(AVFilterLink * inlink,int input_number)299 static int check_ir(AVFilterLink *inlink, int input_number)
300 {
301     AVFilterContext *ctx = inlink->dst;
302     HeadphoneContext *s = ctx->priv;
303     int ir_len, max_ir_len;
304 
305     ir_len = ff_inlink_queued_samples(inlink);
306     max_ir_len = 65536;
307     if (ir_len > max_ir_len) {
308         av_log(ctx, AV_LOG_ERROR, "Too big length of IRs: %d > %d.\n", ir_len, max_ir_len);
309         return AVERROR(EINVAL);
310     }
311     s->hrir_in[input_number].ir_len = ir_len;
312     s->ir_len = FFMAX(ir_len, s->ir_len);
313 
314     return 0;
315 }
316 
headphone_frame(HeadphoneContext * s,AVFrame * in,AVFilterLink * outlink)317 static int headphone_frame(HeadphoneContext *s, AVFrame *in, AVFilterLink *outlink)
318 {
319     AVFilterContext *ctx = outlink->src;
320     int n_clippings[2] = { 0 };
321     ThreadData td;
322     AVFrame *out;
323 
324     out = ff_get_audio_buffer(outlink, in->nb_samples);
325     if (!out) {
326         av_frame_free(&in);
327         return AVERROR(ENOMEM);
328     }
329     out->pts = in->pts;
330 
331     td.in = in; td.out = out; td.write = s->write;
332     td.ir = s->data_ir; td.n_clippings = n_clippings;
333     td.ringbuffer = s->ringbuffer; td.temp_src = s->temp_src;
334     td.temp_fft = s->temp_fft;
335     td.temp_afft = s->temp_afft;
336 
337     if (s->type == TIME_DOMAIN) {
338         ctx->internal->execute(ctx, headphone_convolute, &td, NULL, 2);
339     } else {
340         ctx->internal->execute(ctx, headphone_fast_convolute, &td, NULL, 2);
341     }
342     emms_c();
343 
344     if (n_clippings[0] + n_clippings[1] > 0) {
345         av_log(ctx, AV_LOG_WARNING, "%d of %d samples clipped. Please reduce gain.\n",
346                n_clippings[0] + n_clippings[1], out->nb_samples * 2);
347     }
348 
349     av_frame_free(&in);
350     return ff_filter_frame(outlink, out);
351 }
352 
convert_coeffs(AVFilterContext * ctx,AVFilterLink * inlink)353 static int convert_coeffs(AVFilterContext *ctx, AVFilterLink *inlink)
354 {
355     struct HeadphoneContext *s = ctx->priv;
356     const int ir_len = s->ir_len;
357     int nb_input_channels = ctx->inputs[0]->channels;
358     float gain_lin = expf((s->gain - 3 * nb_input_channels) / 20 * M_LN10);
359     AVFrame *frame;
360     int ret = 0;
361     int n_fft;
362     int i, j, k;
363 
364     s->air_len = 1 << (32 - ff_clz(ir_len));
365     if (s->type == TIME_DOMAIN) {
366         s->air_len = FFALIGN(s->air_len, 32);
367     }
368     s->buffer_length = 1 << (32 - ff_clz(s->air_len));
369     s->n_fft = n_fft = 1 << (32 - ff_clz(ir_len + s->size));
370 
371     if (s->type == FREQUENCY_DOMAIN) {
372         s->fft[0] = av_fft_init(av_log2(s->n_fft), 0);
373         s->fft[1] = av_fft_init(av_log2(s->n_fft), 0);
374         s->ifft[0] = av_fft_init(av_log2(s->n_fft), 1);
375         s->ifft[1] = av_fft_init(av_log2(s->n_fft), 1);
376 
377         if (!s->fft[0] || !s->fft[1] || !s->ifft[0] || !s->ifft[1]) {
378             av_log(ctx, AV_LOG_ERROR, "Unable to create FFT contexts of size %d.\n", s->n_fft);
379             ret = AVERROR(ENOMEM);
380             goto fail;
381         }
382     }
383 
384     if (s->type == TIME_DOMAIN) {
385         s->ringbuffer[0] = av_calloc(s->buffer_length, sizeof(float) * nb_input_channels);
386         s->ringbuffer[1] = av_calloc(s->buffer_length, sizeof(float) * nb_input_channels);
387     } else {
388         s->ringbuffer[0] = av_calloc(s->buffer_length, sizeof(float));
389         s->ringbuffer[1] = av_calloc(s->buffer_length, sizeof(float));
390         s->temp_fft[0] = av_calloc(s->n_fft, sizeof(FFTComplex));
391         s->temp_fft[1] = av_calloc(s->n_fft, sizeof(FFTComplex));
392         s->temp_afft[0] = av_calloc(s->n_fft, sizeof(FFTComplex));
393         s->temp_afft[1] = av_calloc(s->n_fft, sizeof(FFTComplex));
394         if (!s->temp_fft[0] || !s->temp_fft[1] ||
395             !s->temp_afft[0] || !s->temp_afft[1]) {
396             ret = AVERROR(ENOMEM);
397             goto fail;
398         }
399     }
400 
401     if (!s->ringbuffer[0] || !s->ringbuffer[1]) {
402         ret = AVERROR(ENOMEM);
403         goto fail;
404     }
405 
406     if (s->type == TIME_DOMAIN) {
407         s->temp_src[0] = av_calloc(s->air_len, sizeof(float));
408         s->temp_src[1] = av_calloc(s->air_len, sizeof(float));
409 
410         s->data_ir[0] = av_calloc(nb_input_channels * s->air_len, sizeof(*s->data_ir[0]));
411         s->data_ir[1] = av_calloc(nb_input_channels * s->air_len, sizeof(*s->data_ir[1]));
412         if (!s->data_ir[0] || !s->data_ir[1] || !s->temp_src[0] || !s->temp_src[1]) {
413             ret = AVERROR(ENOMEM);
414             goto fail;
415         }
416     } else {
417         s->data_hrtf[0] = av_calloc(n_fft, sizeof(*s->data_hrtf[0]) * nb_input_channels);
418         s->data_hrtf[1] = av_calloc(n_fft, sizeof(*s->data_hrtf[1]) * nb_input_channels);
419         if (!s->data_hrtf[0] || !s->data_hrtf[1]) {
420             ret = AVERROR(ENOMEM);
421             goto fail;
422         }
423     }
424 
425     for (i = 0; i < s->nb_hrir_inputs; av_frame_free(&frame), i++) {
426         int len = s->hrir_in[i].ir_len;
427         float *ptr;
428 
429         ret = ff_inlink_consume_samples(ctx->inputs[i + 1], len, len, &frame);
430         if (ret < 0)
431             goto fail;
432         ptr = (float *)frame->extended_data[0];
433 
434         if (s->hrir_fmt == HRIR_STEREO) {
435             int idx = av_get_channel_layout_channel_index(inlink->channel_layout,
436                                                           s->mapping[i]);
437             if (idx < 0)
438                 continue;
439             if (s->type == TIME_DOMAIN) {
440                 float *data_ir_l = s->data_ir[0] + idx * s->air_len;
441                 float *data_ir_r = s->data_ir[1] + idx * s->air_len;
442 
443                 for (j = 0; j < len; j++) {
444                     data_ir_l[j] = ptr[len * 2 - j * 2 - 2] * gain_lin;
445                     data_ir_r[j] = ptr[len * 2 - j * 2 - 1] * gain_lin;
446                 }
447             } else {
448                 FFTComplex *fft_in_l = s->data_hrtf[0] + idx * n_fft;
449                 FFTComplex *fft_in_r = s->data_hrtf[1] + idx * n_fft;
450 
451                 for (j = 0; j < len; j++) {
452                     fft_in_l[j].re = ptr[j * 2    ] * gain_lin;
453                     fft_in_r[j].re = ptr[j * 2 + 1] * gain_lin;
454                 }
455 
456                 av_fft_permute(s->fft[0], fft_in_l);
457                 av_fft_calc(s->fft[0], fft_in_l);
458                 av_fft_permute(s->fft[0], fft_in_r);
459                 av_fft_calc(s->fft[0], fft_in_r);
460             }
461         } else {
462             int I, N = ctx->inputs[1]->channels;
463 
464             for (k = 0; k < N / 2; k++) {
465                 int idx = av_get_channel_layout_channel_index(inlink->channel_layout,
466                                                               s->mapping[k]);
467                 if (idx < 0)
468                     continue;
469 
470                 I = k * 2;
471                 if (s->type == TIME_DOMAIN) {
472                     float *data_ir_l = s->data_ir[0] + idx * s->air_len;
473                     float *data_ir_r = s->data_ir[1] + idx * s->air_len;
474 
475                     for (j = 0; j < len; j++) {
476                         data_ir_l[j] = ptr[len * N - j * N - N + I    ] * gain_lin;
477                         data_ir_r[j] = ptr[len * N - j * N - N + I + 1] * gain_lin;
478                     }
479                 } else {
480                     FFTComplex *fft_in_l = s->data_hrtf[0] + idx * n_fft;
481                     FFTComplex *fft_in_r = s->data_hrtf[1] + idx * n_fft;
482 
483                     for (j = 0; j < len; j++) {
484                         fft_in_l[j].re = ptr[j * N + I    ] * gain_lin;
485                         fft_in_r[j].re = ptr[j * N + I + 1] * gain_lin;
486                     }
487 
488                     av_fft_permute(s->fft[0], fft_in_l);
489                     av_fft_calc(s->fft[0], fft_in_l);
490                     av_fft_permute(s->fft[0], fft_in_r);
491                     av_fft_calc(s->fft[0], fft_in_r);
492                 }
493             }
494         }
495     }
496 
497     s->have_hrirs = 1;
498 
499 fail:
500     return ret;
501 }
502 
activate(AVFilterContext * ctx)503 static int activate(AVFilterContext *ctx)
504 {
505     HeadphoneContext *s = ctx->priv;
506     AVFilterLink *inlink = ctx->inputs[0];
507     AVFilterLink *outlink = ctx->outputs[0];
508     AVFrame *in = NULL;
509     int i, ret;
510 
511     FF_FILTER_FORWARD_STATUS_BACK_ALL(ctx->outputs[0], ctx);
512     if (!s->eof_hrirs) {
513         int eof = 1;
514         for (i = 0; i < s->nb_hrir_inputs; i++) {
515             AVFilterLink *input = ctx->inputs[i + 1];
516 
517             if (s->hrir_in[i].eof)
518                 continue;
519 
520             if ((ret = check_ir(input, i)) < 0)
521                 return ret;
522 
523             if (ff_outlink_get_status(input) == AVERROR_EOF) {
524                 if (!ff_inlink_queued_samples(input)) {
525                     av_log(ctx, AV_LOG_ERROR, "No samples provided for "
526                            "HRIR stream %d.\n", i);
527                     return AVERROR_INVALIDDATA;
528                 }
529                 s->hrir_in[i].eof = 1;
530             } else {
531                 if (ff_outlink_frame_wanted(ctx->outputs[0]))
532                     ff_inlink_request_frame(input);
533                 eof = 0;
534             }
535         }
536         if (!eof)
537             return 0;
538         s->eof_hrirs = 1;
539 
540         ret = convert_coeffs(ctx, inlink);
541         if (ret < 0)
542             return ret;
543     } else if (!s->have_hrirs)
544         return AVERROR_EOF;
545 
546     if ((ret = ff_inlink_consume_samples(ctx->inputs[0], s->size, s->size, &in)) > 0) {
547         ret = headphone_frame(s, in, outlink);
548         if (ret < 0)
549             return ret;
550     }
551 
552     if (ret < 0)
553         return ret;
554 
555     FF_FILTER_FORWARD_STATUS(ctx->inputs[0], ctx->outputs[0]);
556     if (ff_outlink_frame_wanted(ctx->outputs[0]))
557         ff_inlink_request_frame(ctx->inputs[0]);
558 
559     return 0;
560 }
561 
query_formats(AVFilterContext * ctx)562 static int query_formats(AVFilterContext *ctx)
563 {
564     struct HeadphoneContext *s = ctx->priv;
565     AVFilterFormats *formats = NULL;
566     AVFilterChannelLayouts *layouts = NULL;
567     AVFilterChannelLayouts *stereo_layout = NULL;
568     AVFilterChannelLayouts *hrir_layouts = NULL;
569     int ret, i;
570 
571     ret = ff_add_format(&formats, AV_SAMPLE_FMT_FLT);
572     if (ret)
573         return ret;
574     ret = ff_set_common_formats(ctx, formats);
575     if (ret)
576         return ret;
577 
578     layouts = ff_all_channel_layouts();
579     if (!layouts)
580         return AVERROR(ENOMEM);
581 
582     ret = ff_channel_layouts_ref(layouts, &ctx->inputs[0]->outcfg.channel_layouts);
583     if (ret)
584         return ret;
585 
586     ret = ff_add_channel_layout(&stereo_layout, AV_CH_LAYOUT_STEREO);
587     if (ret)
588         return ret;
589     ret = ff_channel_layouts_ref(stereo_layout, &ctx->outputs[0]->incfg.channel_layouts);
590     if (ret)
591         return ret;
592 
593     if (s->hrir_fmt == HRIR_MULTI) {
594         hrir_layouts = ff_all_channel_counts();
595         if (!hrir_layouts)
596             return AVERROR(ENOMEM);
597         ret = ff_channel_layouts_ref(hrir_layouts, &ctx->inputs[1]->outcfg.channel_layouts);
598         if (ret)
599             return ret;
600     } else {
601         for (i = 1; i <= s->nb_hrir_inputs; i++) {
602             ret = ff_channel_layouts_ref(stereo_layout, &ctx->inputs[i]->outcfg.channel_layouts);
603             if (ret)
604                 return ret;
605         }
606     }
607 
608     formats = ff_all_samplerates();
609     if (!formats)
610         return AVERROR(ENOMEM);
611     return ff_set_common_samplerates(ctx, formats);
612 }
613 
config_input(AVFilterLink * inlink)614 static int config_input(AVFilterLink *inlink)
615 {
616     AVFilterContext *ctx = inlink->dst;
617     HeadphoneContext *s = ctx->priv;
618 
619     if (s->nb_irs < inlink->channels) {
620         av_log(ctx, AV_LOG_ERROR, "Number of HRIRs must be >= %d.\n", inlink->channels);
621         return AVERROR(EINVAL);
622     }
623 
624     s->lfe_channel = av_get_channel_layout_channel_index(inlink->channel_layout,
625                                                          AV_CH_LOW_FREQUENCY);
626     return 0;
627 }
628 
init(AVFilterContext * ctx)629 static av_cold int init(AVFilterContext *ctx)
630 {
631     HeadphoneContext *s = ctx->priv;
632     int i, ret;
633 
634     AVFilterPad pad = {
635         .name         = "in0",
636         .type         = AVMEDIA_TYPE_AUDIO,
637         .config_props = config_input,
638     };
639     if ((ret = ff_insert_inpad(ctx, 0, &pad)) < 0)
640         return ret;
641 
642     if (!s->map) {
643         av_log(ctx, AV_LOG_ERROR, "Valid mapping must be set.\n");
644         return AVERROR(EINVAL);
645     }
646 
647     parse_map(ctx);
648 
649     for (i = 0; i < s->nb_hrir_inputs; i++) {
650         char *name = av_asprintf("hrir%d", i);
651         AVFilterPad pad = {
652             .name         = name,
653             .type         = AVMEDIA_TYPE_AUDIO,
654         };
655         if (!name)
656             return AVERROR(ENOMEM);
657         if ((ret = ff_insert_inpad(ctx, i + 1, &pad)) < 0) {
658             av_freep(&pad.name);
659             return ret;
660         }
661     }
662 
663     if (s->type == TIME_DOMAIN) {
664         AVFloatDSPContext *fdsp = avpriv_float_dsp_alloc(0);
665         if (!fdsp)
666             return AVERROR(ENOMEM);
667         s->scalarproduct_float = fdsp->scalarproduct_float;
668         av_free(fdsp);
669     }
670 
671     return 0;
672 }
673 
config_output(AVFilterLink * outlink)674 static int config_output(AVFilterLink *outlink)
675 {
676     AVFilterContext *ctx = outlink->src;
677     HeadphoneContext *s = ctx->priv;
678     AVFilterLink *inlink = ctx->inputs[0];
679 
680     if (s->hrir_fmt == HRIR_MULTI) {
681         AVFilterLink *hrir_link = ctx->inputs[1];
682 
683         if (hrir_link->channels < inlink->channels * 2) {
684             av_log(ctx, AV_LOG_ERROR, "Number of channels in HRIR stream must be >= %d.\n", inlink->channels * 2);
685             return AVERROR(EINVAL);
686         }
687     }
688 
689     s->gain_lfe = expf((s->gain - 3 * inlink->channels + s->lfe_gain) / 20 * M_LN10);
690 
691     return 0;
692 }
693 
uninit(AVFilterContext * ctx)694 static av_cold void uninit(AVFilterContext *ctx)
695 {
696     HeadphoneContext *s = ctx->priv;
697 
698     av_fft_end(s->ifft[0]);
699     av_fft_end(s->ifft[1]);
700     av_fft_end(s->fft[0]);
701     av_fft_end(s->fft[1]);
702     av_freep(&s->data_ir[0]);
703     av_freep(&s->data_ir[1]);
704     av_freep(&s->ringbuffer[0]);
705     av_freep(&s->ringbuffer[1]);
706     av_freep(&s->temp_src[0]);
707     av_freep(&s->temp_src[1]);
708     av_freep(&s->temp_fft[0]);
709     av_freep(&s->temp_fft[1]);
710     av_freep(&s->temp_afft[0]);
711     av_freep(&s->temp_afft[1]);
712     av_freep(&s->data_hrtf[0]);
713     av_freep(&s->data_hrtf[1]);
714 
715     for (unsigned i = 1; i < ctx->nb_inputs; i++)
716         av_freep(&ctx->input_pads[i].name);
717 }
718 
719 #define OFFSET(x) offsetof(HeadphoneContext, x)
720 #define FLAGS AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
721 
722 static const AVOption headphone_options[] = {
723     { "map",       "set channels convolution mappings",  OFFSET(map),      AV_OPT_TYPE_STRING, {.str=NULL},            .flags = FLAGS },
724     { "gain",      "set gain in dB",                     OFFSET(gain),     AV_OPT_TYPE_FLOAT,  {.dbl=0},     -20,  40, .flags = FLAGS },
725     { "lfe",       "set lfe gain in dB",                 OFFSET(lfe_gain), AV_OPT_TYPE_FLOAT,  {.dbl=0},     -20,  40, .flags = FLAGS },
726     { "type",      "set processing",                     OFFSET(type),     AV_OPT_TYPE_INT,    {.i64=1},       0,   1, .flags = FLAGS, "type" },
727     { "time",      "time domain",                        0,                AV_OPT_TYPE_CONST,  {.i64=0},       0,   0, .flags = FLAGS, "type" },
728     { "freq",      "frequency domain",                   0,                AV_OPT_TYPE_CONST,  {.i64=1},       0,   0, .flags = FLAGS, "type" },
729     { "size",      "set frame size",                     OFFSET(size),     AV_OPT_TYPE_INT,    {.i64=1024},1024,96000, .flags = FLAGS },
730     { "hrir",      "set hrir format",                    OFFSET(hrir_fmt), AV_OPT_TYPE_INT,    {.i64=HRIR_STEREO}, 0, 1, .flags = FLAGS, "hrir" },
731     { "stereo",    "hrir files have exactly 2 channels", 0,                AV_OPT_TYPE_CONST,  {.i64=HRIR_STEREO}, 0, 0, .flags = FLAGS, "hrir" },
732     { "multich",   "single multichannel hrir file",      0,                AV_OPT_TYPE_CONST,  {.i64=HRIR_MULTI},  0, 0, .flags = FLAGS, "hrir" },
733     { NULL }
734 };
735 
736 AVFILTER_DEFINE_CLASS(headphone);
737 
738 static const AVFilterPad outputs[] = {
739     {
740         .name          = "default",
741         .type          = AVMEDIA_TYPE_AUDIO,
742         .config_props  = config_output,
743     },
744     { NULL }
745 };
746 
747 AVFilter ff_af_headphone = {
748     .name          = "headphone",
749     .description   = NULL_IF_CONFIG_SMALL("Apply headphone binaural spatialization with HRTFs in additional streams."),
750     .priv_size     = sizeof(HeadphoneContext),
751     .priv_class    = &headphone_class,
752     .init          = init,
753     .uninit        = uninit,
754     .query_formats = query_formats,
755     .activate      = activate,
756     .inputs        = NULL,
757     .outputs       = outputs,
758     .flags         = AVFILTER_FLAG_SLICE_THREADS | AVFILTER_FLAG_DYNAMIC_INPUTS,
759 };
760