Browse code

avfilter: add SOFAlizer audio filter

Signed-off-by: Paul B Mahol <onemda@gmail.com>

Paul B Mahol authored on 2015/12/09 21:40:03
Showing 9 changed files
... ...
@@ -44,6 +44,7 @@ version <next>:
44 44
 - mips32r5 option has been removed
45 45
 - mips64r6 option has been removed
46 46
 - DXVA2-accelerated VP9 decoding
47
+- SOFAlizer: virtual binaural acoustics filter
47 48
 
48 49
 
49 50
 version 2.8:
... ...
@@ -279,6 +279,7 @@ External library support:
279 279
   --disable-lzma           disable lzma [autodetect]
280 280
   --enable-decklink        enable Blackmagic DeckLink I/O support [no]
281 281
   --enable-mmal            enable decoding via MMAL [no]
282
+  --enable-netcdf          enable NetCDF, needed for sofalizer filter [no]
282 283
   --enable-nvenc           enable NVIDIA NVENC support [no]
283 284
   --enable-openal          enable OpenAL 1.1 capture support [no]
284 285
   --enable-opencl          enable OpenCL code
... ...
@@ -1503,6 +1504,7 @@ EXTERNAL_LIBRARY_LIST="
1503 1503
     libzvbi
1504 1504
     lzma
1505 1505
     mmal
1506
+    netcdf
1506 1507
     nvenc
1507 1508
     openal
1508 1509
     opencl
... ...
@@ -2890,6 +2892,7 @@ showfreqs_filter_deps="avcodec"
2890 2890
 showfreqs_filter_select="fft"
2891 2891
 showspectrum_filter_deps="avcodec"
2892 2892
 showspectrum_filter_select="rdft"
2893
+sofalizer_filter_deps="netcdf"
2893 2894
 spp_filter_deps="gpl avcodec"
2894 2895
 spp_filter_select="fft idctdsp fdctdsp me_cmp pixblockdsp"
2895 2896
 stereo3d_filter_deps="gpl"
... ...
@@ -5494,6 +5497,7 @@ enabled mmal              && { check_lib interface/mmal/mmal.h mmal_port_connect
5494 5494
                                     check_lib interface/mmal/mmal.h mmal_port_connect ; }
5495 5495
                                 check_lib interface/mmal/mmal.h mmal_port_connect ; } ||
5496 5496
                                die "ERROR: mmal not found"; }
5497
+enabled netcdf            && require_pkg_config netcdf netcdf.h nc_inq_libvers
5497 5498
 enabled nvenc             && { check_header nvEncodeAPI.h || die "ERROR: nvEncodeAPI.h not found."; } &&
5498 5499
                              { check_cpp_condition nvEncodeAPI.h "NVENCAPI_MAJOR_VERSION >= 5" ||
5499 5500
                                die "ERROR: NVENC API version 4 or older is not supported"; } &&
... ...
@@ -2889,6 +2889,35 @@ silenceremove=1:5:0.02
2889 2889
 @end example
2890 2890
 @end itemize
2891 2891
 
2892
+@section sofalizer
2893
+
2894
+SOFAlizer uses head-related transfer functions (HRTFs) to create virtual
2895
+loudspeakers around the user for binaural listening via headphones (audio
2896
+formats up to 9 channels supported).
2897
+The HRTFs are stored in SOFA files (see www.sofacoustics.org for a database).
2898
+SOFAlizer is developed at the Acoustics Research Institute (ARI) of the
2899
+Austrian Academy of Sciences.
2900
+
2901
+The filter accepts the following options:
2902
+
2903
+@table @option
2904
+@item sofa
2905
+Set the SOFA file used for rendering.
2906
+
2907
+@item gain
2908
+Set gain applied to audio. Value is in dB. Default is 0.
2909
+
2910
+@item rotation
2911
+Set rotation of virtual loudspeakers in deg. Default is 0.
2912
+
2913
+@item elevation
2914
+Set elevation of virtual speakers in deg. Default is 0.
2915
+
2916
+@item radius
2917
+Set distance in meters between loudspeakers and the listener with near-field
2918
+HRTFs. Default is 1.
2919
+@end table
2920
+
2892 2921
 @section stereotools
2893 2922
 
2894 2923
 This filter has some handy utilities to manage stereo signals, for converting
... ...
@@ -87,6 +87,7 @@ OBJS-$(CONFIG_SIDECHAINCOMPRESS_FILTER)      += af_sidechaincompress.o
87 87
 OBJS-$(CONFIG_SIDECHAINGATE_FILTER)          += af_agate.o
88 88
 OBJS-$(CONFIG_SILENCEDETECT_FILTER)          += af_silencedetect.o
89 89
 OBJS-$(CONFIG_SILENCEREMOVE_FILTER)          += af_silenceremove.o
90
+OBJS-$(CONFIG_SOFALIZER_FILTER)              += af_sofalizer.o
90 91
 OBJS-$(CONFIG_STEREOTOOLS_FILTER)            += af_stereotools.o
91 92
 OBJS-$(CONFIG_STEREOWIDEN_FILTER)            += af_stereowiden.o
92 93
 OBJS-$(CONFIG_TREBLE_FILTER)                 += af_biquads.o
93 94
new file mode 100644
... ...
@@ -0,0 +1,1018 @@
0
+/*****************************************************************************
1
+ * sofalizer.c : SOFAlizer filter for virtual binaural acoustics
2
+ *****************************************************************************
3
+ * Copyright (C) 2013-2015 Andreas Fuchs, Wolfgang Hrauda,
4
+ *                         Acoustics Research Institute (ARI), Vienna, Austria
5
+ *
6
+ * Authors: Andreas Fuchs <andi.fuchs.mail@gmail.com>
7
+ *          Wolfgang Hrauda <wolfgang.hrauda@gmx.at>
8
+ *
9
+ * SOFAlizer project coordinator at ARI, main developer of SOFA:
10
+ *          Piotr Majdak <piotr@majdak.at>
11
+ *
12
+ * This program is free software; you can redistribute it and/or modify it
13
+ * under the terms of the GNU Lesser General Public License as published by
14
+ * the Free Software Foundation; either version 2.1 of the License, or
15
+ * (at your option) any later version.
16
+ *
17
+ * This program is distributed in the hope that it will be useful,
18
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
19
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
20
+ * GNU Lesser General Public License for more details.
21
+ *
22
+ * You should have received a copy of the GNU Lesser General Public License
23
+ * along with this program; if not, write to the Free Software Foundation,
24
+ * Inc., 51 Franklin Street, Fifth Floor, Boston MA 02110-1301, USA.
25
+ *****************************************************************************/
26
+
27
+#include <math.h>
28
+#include <netcdf.h>
29
+
30
+#include "libavutil/float_dsp.h"
31
+#include "libavutil/opt.h"
32
+#include "avfilter.h"
33
+#include "internal.h"
34
+#include "audio.h"
35
+
36
+typedef struct NCSofa {  /* contains data of one SOFA file */
37
+    int ncid;            /* netCDF ID of the opened SOFA file */
38
+    int n_samples;       /* length of one impulse response (IR) */
39
+    int m_dim;           /* number of measurement positions */
40
+    int *data_delay;     /* broadband delay of each IR */
41
+                         /* all measurement positions for each receiver (i.e. ear): */
42
+    float *sp_a;         /* azimuth angles */
43
+    float *sp_e;         /* elevation angles */
44
+    float *sp_r;         /* radii */
45
+                         /* data at each measurement position for each receiver: */
46
+    float *data_ir;      /* IRs (time-domain) */
47
+} NCSofa;
48
+
49
+typedef struct SOFAlizerContext {
50
+    const AVClass *class;
51
+
52
+    char *filename;             /* name of SOFA file */
53
+    NCSofa sofa;                /* contains data of the SOFA file */
54
+
55
+    const int8_t *reorder;      /* reorder in SOFA channel order */
56
+    int sample_rate;            /* sample rate from SOFA file */
57
+    float *speaker_pos;         /* positions of the virtual loudspekaers */
58
+    float gain_lfe;             /* gain applied to LFE channel */
59
+
60
+    int n_conv;                 /* number of channels to convolute */
61
+
62
+                                /* buffer variables (for convolution) */
63
+    float *ringbuffer[2];       /* buffers input samples, length of one buffer: */
64
+                                /* no. input ch. (incl. LFE) x buffer_length */
65
+    int write[2];               /* current write position to ringbuffer */
66
+    int buffer_length;          /* is: longest IR plus max. delay in all SOFA files */
67
+                                /* then choose next power of 2 */
68
+
69
+                                /* netCDF variables */
70
+    int *delay[2];              /* broadband delay for each channel/IR to be convolved */
71
+
72
+    float *data_ir[2];          /* IRs for all channels to be convolved */
73
+                                /* (this excludes the LFE) */
74
+    float *temp_src[2];
75
+
76
+                         /* control variables */
77
+    float gain;          /* filter gain (in dB) */
78
+    float rotation;      /* rotation of virtual loudspeakers (in degrees)  */
79
+    float elevation;     /* elevation of virtual loudspeakers (in deg.) */
80
+    float radius;        /* distance virtual loudspeakers to listener (in metres) */
81
+
82
+    int lfe;             /* whether or not the LFE channel is used */
83
+
84
+    AVFloatDSPContext *fdsp;
85
+} SOFAlizerContext;
86
+
87
+static int close_sofa(struct NCSofa *sofa)
88
+{
89
+    av_freep(&sofa->data_delay);
90
+    av_freep(&sofa->sp_a);
91
+    av_freep(&sofa->sp_e);
92
+    av_freep(&sofa->sp_r);
93
+    av_freep(&sofa->data_ir);
94
+    nc_close(sofa->ncid);
95
+    sofa->ncid = 0;
96
+
97
+    return 0;
98
+}
99
+
100
+static int load_sofa(AVFilterContext *ctx, char *filename, int *samplingrate)
101
+{
102
+    struct SOFAlizerContext *s = ctx->priv;
103
+    /* variables associated with content of SOFA file: */
104
+    int ncid, n_dims, n_vars, n_gatts, n_unlim_dim_id, status;
105
+    char data_delay_dim_name[NC_MAX_NAME];
106
+    float *sp_a, *sp_e, *sp_r, *data_ir;
107
+    char *sofa_conventions;
108
+    char dim_name[NC_MAX_NAME];   /* names of netCDF dimensions */
109
+    size_t *dim_length;           /* lengths of netCDF dimensions */
110
+    char *psz_conventions;
111
+    unsigned int sample_rate;
112
+    int data_delay_dim_id[2];
113
+    int samplingrate_id;
114
+    int data_delay_id;
115
+    int n_samples;
116
+    int m_dim_id = -1;
117
+    int n_dim_id = -1;
118
+    int data_ir_id;
119
+    size_t att_len;
120
+    int m_dim;
121
+    int *data_delay;
122
+    int sp_id;
123
+    int i, ret;
124
+
125
+    s->sofa.ncid = 0;
126
+    status = nc_open(filename, NC_NOWRITE, &ncid); /* open SOFA file read-only */
127
+    if (status != NC_NOERR) {
128
+        av_log(ctx, AV_LOG_ERROR, "Can't find SOFA-file '%s'\n", filename);
129
+        return AVERROR(EINVAL);
130
+    }
131
+
132
+    /* get number of dimensions, vars, global attributes and Id of unlimited dimensions: */
133
+    nc_inq(ncid, &n_dims, &n_vars, &n_gatts, &n_unlim_dim_id);
134
+
135
+    /* -- get number of measurements ("M") and length of one IR ("N") -- */
136
+    dim_length = av_malloc_array(n_dims, sizeof(*dim_length));
137
+    if (!dim_length) {
138
+        nc_close(ncid);
139
+        return AVERROR(ENOMEM);
140
+    }
141
+
142
+    for (i = 0; i < n_dims; i++) { /* go through all dimensions of file */
143
+        nc_inq_dim(ncid, i, (char *)&dim_name, &dim_length[i]); /* get dimensions */
144
+        if (!strncmp("M", (const char *)&dim_name, 1)) /* get ID of dimension "M" */
145
+            m_dim_id = i;
146
+        if (!strncmp("N", (const char *)&dim_name, 1)) /* get ID of dimension "N" */
147
+            n_dim_id = i;
148
+    }
149
+
150
+    if ((m_dim_id == -1) || (n_dim_id == -1)) { /* dimension "M" or "N" couldn't be found */
151
+        av_log(ctx, AV_LOG_ERROR, "Can't find required dimensions in SOFA file.\n");
152
+        av_freep(&dim_length);
153
+        nc_close(ncid);
154
+        return AVERROR(EINVAL);
155
+    }
156
+
157
+    n_samples = dim_length[n_dim_id]; /* get number of measurements */
158
+    m_dim     = dim_length[m_dim_id]; /* get length of one IR */
159
+
160
+    av_freep(&dim_length);
161
+
162
+    /* -- check file type -- */
163
+    /* get length of attritube "Conventions" */
164
+    status = nc_inq_attlen(ncid, NC_GLOBAL, "Conventions", &att_len);
165
+    if (status != NC_NOERR) {
166
+        av_log(ctx, AV_LOG_ERROR, "Can't get length of attribute \"Conventions\".\n");
167
+        nc_close(ncid);
168
+        return AVERROR_INVALIDDATA;
169
+    }
170
+
171
+    /* check whether file is SOFA file */
172
+    psz_conventions = av_malloc(att_len + 1);
173
+    if (!psz_conventions) {
174
+        nc_close(ncid);
175
+        return AVERROR(ENOMEM);
176
+    }
177
+
178
+    nc_get_att_text(ncid, NC_GLOBAL, "Conventions", psz_conventions);
179
+    *(psz_conventions + att_len) = 0;
180
+    if (strncmp("SOFA", psz_conventions, 4)) {
181
+        av_log(ctx, AV_LOG_ERROR, "Not a SOFA file!\n");
182
+        av_freep(&psz_conventions);
183
+        nc_close(ncid);
184
+        return AVERROR(EINVAL);
185
+    }
186
+    av_freep(&psz_conventions);
187
+
188
+    status = nc_inq_attlen(ncid, NC_GLOBAL, "SOFAConventions", &att_len);
189
+    if (status != NC_NOERR) {
190
+        av_log(ctx, AV_LOG_ERROR, "Can't get length of attribute \"SOFAConventions\".\n");
191
+        nc_close(ncid);
192
+        return AVERROR_INVALIDDATA;
193
+    }
194
+
195
+    sofa_conventions = av_malloc(att_len + 1);
196
+    if (!sofa_conventions) {
197
+        nc_close(ncid);
198
+        return AVERROR(ENOMEM);
199
+    }
200
+
201
+    nc_get_att_text(ncid, NC_GLOBAL, "SOFAConventions", sofa_conventions);
202
+    *(sofa_conventions + att_len) = 0;
203
+    if (strncmp("SimpleFreeFieldHRIR", sofa_conventions, att_len)) {
204
+        av_log(ctx, AV_LOG_ERROR, "Not a SimpleFreeFieldHRIR file!\n");
205
+        av_freep(&sofa_conventions);
206
+        nc_close(ncid);
207
+        return AVERROR(EINVAL);
208
+    }
209
+    av_freep(&sofa_conventions);
210
+
211
+    /* -- get sampling rate of HRTFs -- */
212
+    /* read ID, then value */
213
+    status  = nc_inq_varid(ncid, "Data.SamplingRate", &samplingrate_id);
214
+    status += nc_get_var_uint(ncid, samplingrate_id, &sample_rate);
215
+    if (status != NC_NOERR) {
216
+        av_log(ctx, AV_LOG_ERROR, "Couldn't read Data.SamplingRate.\n");
217
+        nc_close(ncid);
218
+        return AVERROR(EINVAL);
219
+    }
220
+    *samplingrate = sample_rate; /* remember sampling rate */
221
+
222
+    /* -- allocate memory for one value for each measurement position: -- */
223
+    sp_a = s->sofa.sp_a = av_malloc_array(m_dim, sizeof(float));
224
+    sp_e = s->sofa.sp_e = av_malloc_array(m_dim, sizeof(float));
225
+    sp_r = s->sofa.sp_r = av_malloc_array(m_dim, sizeof(float));
226
+    /* delay and IR values required for each ear and measurement position: */
227
+    data_delay = s->sofa.data_delay = av_calloc(m_dim, 2 * sizeof(int));
228
+    data_ir = s->sofa.data_ir = av_malloc_array(m_dim * n_samples, sizeof(float) * 2);
229
+    s->temp_src[0] = av_calloc(FFALIGN(n_samples, 16), sizeof(float));
230
+    s->temp_src[1] = av_calloc(FFALIGN(n_samples, 16), sizeof(float));
231
+
232
+    if (!data_delay || !sp_a || !sp_e || !sp_r || !data_ir ||
233
+        !s->temp_src[0] || !s->temp_src[1]) {
234
+        /* if memory could not be allocated */
235
+        close_sofa(&s->sofa);
236
+        return AVERROR(ENOMEM);
237
+    }
238
+
239
+    /* get impulse responses (HRTFs): */
240
+    /* get corresponding ID */
241
+    status = nc_inq_varid(ncid, "Data.IR", &data_ir_id);
242
+    status += nc_get_var_float(ncid, data_ir_id, data_ir); /* read and store IRs */
243
+    if (status != NC_NOERR) {
244
+        av_log(ctx, AV_LOG_ERROR, "Couldn't read Data.IR!\n");
245
+        ret = AVERROR(EINVAL);
246
+        goto error;
247
+    }
248
+
249
+    /* get source positions of the HRTFs in the SOFA file: */
250
+    status  = nc_inq_varid(ncid, "SourcePosition", &sp_id); /* get corresponding ID */
251
+    status += nc_get_vara_float(ncid, sp_id, (size_t[2]){ 0, 0 } ,
252
+                (size_t[2]){ m_dim, 1}, sp_a); /* read & store azimuth angles */
253
+    status += nc_get_vara_float(ncid, sp_id, (size_t[2]){ 0, 1 } ,
254
+                (size_t[2]){ m_dim, 1}, sp_e); /* read & store elevation angles */
255
+    status += nc_get_vara_float(ncid, sp_id, (size_t[2]){ 0, 2 } ,
256
+                (size_t[2]){ m_dim, 1}, sp_r); /* read & store radii */
257
+    if (status != NC_NOERR) { /* if any source position variable coudn't be read */
258
+        av_log(ctx, AV_LOG_ERROR, "Couldn't read SourcePosition.\n");
259
+        ret = AVERROR(EINVAL);
260
+        goto error;
261
+    }
262
+
263
+    /* read Data.Delay, check for errors and fit it to data_delay */
264
+    status  = nc_inq_varid(ncid, "Data.Delay", &data_delay_id);
265
+    status += nc_inq_vardimid(ncid, data_delay_id, &data_delay_dim_id[0]);
266
+    status += nc_inq_dimname(ncid, data_delay_dim_id[0], data_delay_dim_name);
267
+    if (status != NC_NOERR) {
268
+        av_log(ctx, AV_LOG_ERROR, "Couldn't read Data.Delay.\n");
269
+        ret = AVERROR(EINVAL);
270
+        goto error;
271
+    }
272
+
273
+    /* Data.Delay dimension check */
274
+    /* dimension of Data.Delay is [I R]: */
275
+    if (!strncmp(data_delay_dim_name, "I", 2)) {
276
+        /* check 2 characters to assure string is 0-terminated after "I" */
277
+        int delay[2]; /* delays get from SOFA file: */
278
+
279
+        av_log(ctx, AV_LOG_DEBUG, "Data.Delay has dimension [I R]\n");
280
+        status = nc_get_var_int(ncid, data_delay_id, &delay[0]);
281
+        if (status != NC_NOERR) {
282
+            av_log(ctx, AV_LOG_ERROR, "Couldn't read Data.Delay\n");
283
+            ret = AVERROR(EINVAL);
284
+            goto error;
285
+        }
286
+        int *data_delay_r = data_delay + m_dim;
287
+        for (i = 0; i < m_dim; i++) { /* extend given dimension [I R] to [M R] */
288
+            /* assign constant delay value for all measurements to data_delay fields */
289
+            data_delay[i]   = delay[0];
290
+            data_delay_r[i] = delay[1];
291
+        }
292
+        /* dimension of Data.Delay is [M R] */
293
+    } else if (!strncmp(data_delay_dim_name, "M", 2)) {
294
+        av_log(ctx, AV_LOG_ERROR, "Data.Delay in dimension [M R]\n");
295
+        /* get delays from SOFA file: */
296
+        status = nc_get_var_int(ncid, data_delay_id, data_delay);
297
+        if (status != NC_NOERR) {
298
+            av_log(ctx, AV_LOG_ERROR, "Couldn't read Data.Delay\n");
299
+            ret = AVERROR(EINVAL);
300
+            goto error;
301
+        }
302
+    } else { /* dimension of Data.Delay is neither [I R] nor [M R] */
303
+        av_log(ctx, AV_LOG_ERROR, "Data.Delay does not have the required dimensions [I R] or [M R].\n");
304
+        ret = AVERROR(EINVAL);
305
+        goto error;
306
+    }
307
+
308
+    /* save information in SOFA struct: */
309
+    s->sofa.m_dim = m_dim; /* no. measurement positions */
310
+    s->sofa.n_samples = n_samples; /* length on one IR */
311
+    s->sofa.ncid = ncid; /* netCDF ID of SOFA file */
312
+    nc_close(ncid); /* close SOFA file */
313
+
314
+    return 0;
315
+
316
+error:
317
+    close_sofa(&s->sofa);
318
+    return ret;
319
+}
320
+
321
+static const int8_t reorder[18][9] = {
322
+    { 0, -1, -1, -1, -1, -1, -1, -1, -1 },
323
+    { 0,  1, -1, -1, -1, -1, -1, -1, -1 },
324
+    { 0,  1,  2, -1, -1, -1, -1, -1, -1 },
325
+    { 0,  1,  2, -1, -1, -1, -1, -1, -1 },
326
+    { 0,  1,  2,  3, -1, -1, -1, -1, -1 },
327
+    { 0,  1,  2,  3, -1, -1, -1, -1, -1 },
328
+    { 0,  1,  2,  3, -1, -1, -1, -1, -1 },
329
+    { 0,  1,  3,  4,  2, -1, -1, -1, -1 },
330
+    { 0,  1,  3,  4,  2, -1, -1, -1, -1 },
331
+    { 0,  1,  4,  5,  2,  3, -1, -1, -1 },
332
+    { 0,  1,  4,  5,  2,  3, -1, -1, -1 },
333
+    { 0,  1,  5,  6,  4,  2,  3, -1, -1 },
334
+    { 0,  1,  5,  6,  3,  4,  2, -1, -1 },
335
+    { 0,  1,  6,  7,  4,  5,  2,  3, -1 },
336
+    { 0,  1,  2,  3,  4,  5,  6,  7,  8 },
337
+    { 0,  1,  2,  3,  4,  5,  6,  7, -1 },
338
+    { 0,  1,  3,  4,  2,  5, -1, -1, -1 },
339
+    { 0,  1,  4,  5,  2,  6,  3, -1, -1 },
340
+};
341
+
342
+static int get_speaker_pos(AVFilterContext *ctx, float *speaker_pos)
343
+{
344
+    struct SOFAlizerContext *s = ctx->priv;
345
+    uint64_t channels_layout = ctx->inputs[0]->channel_layout;
346
+    float pos_temp[9];
347
+    int nb_input_channels = ctx->inputs[0]->channels; /* get no. input channels */
348
+    int n_conv = nb_input_channels;
349
+
350
+    if (channels_layout & AV_CH_LOW_FREQUENCY) { /* if LFE is used */
351
+        /* decrease number of channels to be convolved: */
352
+        n_conv = nb_input_channels - 1;
353
+    }
354
+
355
+    /* set speaker positions according to input channel configuration: */
356
+    switch (channels_layout) {
357
+    case AV_CH_LAYOUT_MONO:
358
+                            pos_temp[0] = 0;
359
+                            break;
360
+    case AV_CH_LAYOUT_STEREO:
361
+    case AV_CH_LAYOUT_2POINT1:
362
+                            pos_temp[0] = 30;
363
+                            pos_temp[1] = 330;
364
+                            break;
365
+    case AV_CH_LAYOUT_SURROUND:
366
+    case AV_CH_LAYOUT_3POINT1:
367
+                            pos_temp[0] = 30;
368
+                            pos_temp[1] = 330;
369
+                            pos_temp[2] = 0;
370
+                            break;
371
+    case AV_CH_LAYOUT_2_1:
372
+                            pos_temp[0] = 30;
373
+                            pos_temp[1] = 330;
374
+                            pos_temp[2] = 180;
375
+                            break;
376
+    case AV_CH_LAYOUT_2_2:
377
+                            pos_temp[0] = 30;
378
+                            pos_temp[1] = 330;
379
+                            pos_temp[2] = 90;
380
+                            pos_temp[3] = 270;
381
+                            break;
382
+    case AV_CH_LAYOUT_QUAD:
383
+                            pos_temp[0] = 30;
384
+                            pos_temp[1] = 330;
385
+                            pos_temp[2] = 120;
386
+                            pos_temp[3] = 240;
387
+                            break;
388
+    case AV_CH_LAYOUT_4POINT0:
389
+    case AV_CH_LAYOUT_4POINT1:
390
+                            pos_temp[0] = 30;
391
+                            pos_temp[1] = 330;
392
+                            pos_temp[2] = 0;
393
+                            pos_temp[3] = 180;
394
+                            break;
395
+    case AV_CH_LAYOUT_5POINT0:
396
+    case AV_CH_LAYOUT_5POINT1:
397
+                            pos_temp[0] = 30;
398
+                            pos_temp[1] = 330;
399
+                            pos_temp[2] = 90;
400
+                            pos_temp[3] = 270;
401
+                            pos_temp[4] = 0;
402
+                            break;
403
+    case AV_CH_LAYOUT_5POINT0_BACK:
404
+    case AV_CH_LAYOUT_5POINT1_BACK:
405
+                            pos_temp[0] = 30;
406
+                            pos_temp[1] = 330;
407
+                            pos_temp[2] = 120;
408
+                            pos_temp[3] = 240;
409
+                            pos_temp[4] = 0;
410
+                            break;
411
+    case AV_CH_LAYOUT_6POINT0:
412
+    case AV_CH_LAYOUT_6POINT1:
413
+                            pos_temp[0] = 30;
414
+                            pos_temp[1] = 330;
415
+                            pos_temp[2] = 90;
416
+                            pos_temp[3] = 270;
417
+                            pos_temp[4] = 0;
418
+                            pos_temp[5] = 180;
419
+                            break;
420
+    case AV_CH_LAYOUT_6POINT1_BACK:
421
+                            pos_temp[0] = 30;
422
+                            pos_temp[1] = 330;
423
+                            pos_temp[2] = 120;
424
+                            pos_temp[3] = 240;
425
+                            pos_temp[4] = 0;
426
+                            pos_temp[4] = 180;
427
+                            break;
428
+    case AV_CH_LAYOUT_HEXAGONAL:
429
+                            pos_temp[0] = 30;
430
+                            pos_temp[1] = 330;
431
+                            pos_temp[2] = 120;
432
+                            pos_temp[3] = 240;
433
+                            pos_temp[4] = 0;
434
+                            pos_temp[5] = 180;
435
+                            break;
436
+    case AV_CH_LAYOUT_7POINT0:
437
+    case AV_CH_LAYOUT_7POINT1:
438
+                            pos_temp[0] = 30;
439
+                            pos_temp[1] = 330;
440
+                            pos_temp[2] = 90;
441
+                            pos_temp[3] = 270;
442
+                            pos_temp[4] = 150;
443
+                            pos_temp[5] = 210;
444
+                            pos_temp[6] = 0;
445
+                            break;
446
+    case AV_CH_LAYOUT_OCTAGONAL:
447
+                            pos_temp[0] = 30;
448
+                            pos_temp[1] = 330;
449
+                            pos_temp[2] = 0;
450
+                            pos_temp[3] = 150;
451
+                            pos_temp[4] = 210;
452
+                            pos_temp[5] = 180;
453
+                            pos_temp[6] = 90;
454
+                            pos_temp[7] = 270;
455
+                            break;
456
+    default:
457
+                            return -1;
458
+    }
459
+
460
+    switch (channels_layout) {
461
+    case AV_CH_LAYOUT_MONO:
462
+                            s->reorder = reorder[0];
463
+                            break;
464
+    case AV_CH_LAYOUT_STEREO:
465
+                            s->reorder = reorder[1];
466
+                            break;
467
+    case AV_CH_LAYOUT_2_1:
468
+    case AV_CH_LAYOUT_2POINT1:
469
+                            s->reorder = reorder[2];
470
+                            break;
471
+    case AV_CH_LAYOUT_SURROUND:
472
+                            s->reorder = reorder[3];
473
+                            break;
474
+    case AV_CH_LAYOUT_3POINT1:
475
+    case AV_CH_LAYOUT_2_2:
476
+                            s->reorder = reorder[4];
477
+                            break;
478
+    case AV_CH_LAYOUT_QUAD:
479
+                            s->reorder = reorder[5];
480
+                            break;
481
+    case AV_CH_LAYOUT_4POINT0:
482
+                            s->reorder = reorder[6];
483
+                            break;
484
+    case AV_CH_LAYOUT_4POINT1:
485
+                            s->reorder = reorder[7];
486
+                            break;
487
+    case AV_CH_LAYOUT_5POINT0:
488
+    case AV_CH_LAYOUT_5POINT0_BACK:
489
+                            s->reorder = reorder[8];
490
+                            break;
491
+    case AV_CH_LAYOUT_5POINT1:
492
+    case AV_CH_LAYOUT_5POINT1_BACK:
493
+                            s->reorder = reorder[9];
494
+                            break;
495
+    case AV_CH_LAYOUT_6POINT0:
496
+                            s->reorder = reorder[10];
497
+                            break;
498
+    case AV_CH_LAYOUT_HEXAGONAL:
499
+                            s->reorder = reorder[16];
500
+                            break;
501
+    case AV_CH_LAYOUT_6POINT1:
502
+                            s->reorder = reorder[11];
503
+                            break;
504
+    case AV_CH_LAYOUT_6POINT1_BACK:
505
+                            s->reorder = reorder[17];
506
+                            break;
507
+    case AV_CH_LAYOUT_7POINT0:
508
+                            s->reorder = reorder[12];
509
+                            break;
510
+    case AV_CH_LAYOUT_7POINT1:
511
+                            s->reorder = reorder[13];
512
+                            break;
513
+    case AV_CH_LAYOUT_OCTAGONAL:
514
+                            s->reorder = reorder[15];
515
+                            break;
516
+    default:
517
+                            return -1;
518
+    }
519
+
520
+    memcpy(speaker_pos, pos_temp, n_conv * sizeof(float));
521
+
522
+    return 0;
523
+
524
+}
525
+
526
+static int max_delay(struct NCSofa *sofa)
527
+{
528
+    int i, max = 0;
529
+
530
+    for (i = 0; i < sofa->m_dim * 2; i++) {
531
+        /* search maximum delay in given SOFA file */
532
+        max = FFMAX(max, sofa->data_delay[i]);
533
+    }
534
+
535
+    return max;
536
+}
537
+
538
+static int find_m(SOFAlizerContext *s, int azim, int elev, float radius)
539
+{
540
+    /* get source positions and M of currently selected SOFA file */
541
+    float *sp_a = s->sofa.sp_a; /* azimuth angle */
542
+    float *sp_e = s->sofa.sp_e; /* elevation angle */
543
+    float *sp_r = s->sofa.sp_r; /* radius */
544
+    int m_dim = s->sofa.m_dim; /* no. measurements */
545
+    int best_id = 0; /* index m currently closest to desired source pos. */
546
+    float delta = 1000; /* offset between desired and currently best pos. */
547
+    float current;
548
+    int i;
549
+
550
+    for (i = 0; i < m_dim; i++) {
551
+        /* search through all measurements in currently selected SOFA file */
552
+        /* distance of current to desired source position: */
553
+        current = fabs(sp_a[i] - azim) +
554
+                  fabs(sp_e[i] - elev) +
555
+                  fabs(sp_r[i] - radius);
556
+        if (current <= delta) {
557
+            /* if current distance is smaller than smallest distance so far */
558
+            delta = current;
559
+            best_id = i; /* remember index */
560
+        }
561
+    }
562
+
563
+    return best_id;
564
+}
565
+
566
+static int compensate_volume(AVFilterContext *ctx)
567
+{
568
+    struct SOFAlizerContext *s = ctx->priv;
569
+    float compensate;
570
+    float energy = 0;
571
+    float *ir;
572
+    int m, j;
573
+
574
+    if (s->sofa.ncid) {
575
+        /* find IR at front center position in the SOFA file (IR closest to 0°,0°,1m) */
576
+        struct NCSofa *sofa = &s->sofa;
577
+        m = find_m(s, 0, 0, 1);
578
+        /* get energy of that IR and compensate volume */
579
+        ir = sofa->data_ir + 2 * m * sofa->n_samples;
580
+        for (j = 0; j < sofa->n_samples; j++) {
581
+            energy += *(ir + j) * *(ir + j);
582
+        }
583
+        compensate = 256 / (sofa->n_samples * sqrt(energy));
584
+        av_log(ctx, AV_LOG_DEBUG, "Compensate-factor: %f\n", compensate);
585
+        ir = sofa->data_ir;
586
+        for (j = 0; j < sofa->n_samples * sofa->m_dim * 2; j++) {
587
+            ir[j] *= compensate; /* apply volume compensation to IRs */
588
+        }
589
+    }
590
+
591
+    return 0;
592
+}
593
+
594
+typedef struct ThreadData {
595
+    AVFrame *in, *out;
596
+    int *write;
597
+    int **delay;
598
+    float **ir;
599
+    int *n_clippings;
600
+    float **ringbuffer;
601
+    float **temp_src;
602
+} ThreadData;
603
+
604
+static int sofalizer_convolute(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
605
+{
606
+    SOFAlizerContext *s = ctx->priv;
607
+    ThreadData *td = arg;
608
+    AVFrame *in = td->in, *out = td->out;
609
+    int offset = jobnr;
610
+    int *write = &td->write[jobnr];
611
+    const int *const delay = td->delay[jobnr];
612
+    const float *const ir = td->ir[jobnr];
613
+    int *n_clippings = &td->n_clippings[jobnr];
614
+    float *ringbuffer = td->ringbuffer[jobnr];
615
+    float *temp_src = td->temp_src[jobnr];
616
+    const int n_samples = s->sofa.n_samples; /* length of one IR */
617
+    const float *src = (const float *)in->data[0]; /* get pointer to audio input buffer */
618
+    float *dst = (float *)out->data[0]; /* get pointer to audio output buffer */
619
+    int in_channels = in->channels; /* number of input channels */
620
+    /* ring buffer length is: longest IR plus max. delay -> next power of 2 */
621
+    int buffer_length = s->buffer_length;
622
+    /* -1 for AND instead of MODULO (applied to powers of 2): */
623
+    uint32_t modulo = (uint32_t)buffer_length - 1;
624
+    float *buffer[10]; /* holds ringbuffer for each input channel */
625
+    int wr = *write;
626
+    int read;
627
+    int i, j, l;
628
+
629
+    dst += offset;
630
+    for (l = 0; l < in_channels; l++) {
631
+        /* get starting address of ringbuffer for each input channel */
632
+        buffer[l] = ringbuffer + l * buffer_length;
633
+    }
634
+
635
+    for (i = 0; i < in->nb_samples; i++) {
636
+        const float *temp_ir = ir; /* using same set of IRs for each sample */
637
+
638
+        *dst = 0;
639
+        for (l = 0; l < in_channels; l++) {
640
+            /* write current input sample to ringbuffer (for each channel) */
641
+            *(buffer[l] + wr) = src[s->reorder[l]];
642
+        }
643
+
644
+        /* loop goes through all channels to be convolved (excl. LFE): */
645
+        for (l = 0; l < s->n_conv; l++) {
646
+            const float *const bptr = buffer[l];
647
+
648
+            /* current read position in ringbuffer: input sample write position
649
+             * - delay for l-th ch. + diff. betw. IR length and buffer length
650
+             * (mod buffer length) */
651
+            read = (wr - *(delay + l) - (n_samples - 1) + buffer_length) & modulo;
652
+
653
+            for (j = 0; j < n_samples; j++)
654
+                temp_src[j] = bptr[(read + j) & modulo];
655
+
656
+            /* multiply signal and IR, and add up the results */
657
+            dst[0] += s->fdsp->scalarproduct_float(temp_ir, temp_src, n_samples);
658
+            temp_ir += n_samples;
659
+        }
660
+
661
+        if (s->lfe) { /* LFE */
662
+            /* apply gain to LFE signal and add to output buffer */
663
+            *dst += *(buffer[s->n_conv] + wr) * s->gain_lfe;
664
+        }
665
+
666
+        /* clippings counter */
667
+        if (fabs(*dst) > 1)
668
+            *n_clippings += 1;
669
+
670
+        /* move output buffer pointer by +2 to get to next sample of processed channel: */
671
+        dst += 2;
672
+        src += in_channels;
673
+        wr   = (wr + 1) & modulo; /* update ringbuffer write position */
674
+    }
675
+
676
+    *write = wr; /* remember write position in ringbuffer for next call */
677
+
678
+    return 0;
679
+}
680
+
681
+static int filter_frame(AVFilterLink *inlink, AVFrame *in)
682
+{
683
+    AVFilterContext *ctx = inlink->dst;
684
+    SOFAlizerContext *s = ctx->priv;
685
+    AVFilterLink *outlink = ctx->outputs[0];
686
+    int n_clippings[2] = { 0 };
687
+    ThreadData td;
688
+    AVFrame *out;
689
+
690
+    out = ff_get_audio_buffer(outlink, in->nb_samples);
691
+    if (!out) {
692
+        av_frame_free(&in);
693
+        return AVERROR(ENOMEM);
694
+    }
695
+    av_frame_copy_props(out, in);
696
+
697
+    td.in = in; td.out = out; td.write = s->write;
698
+    td.delay = s->delay; td.ir = s->data_ir; td.n_clippings = n_clippings;
699
+    td.ringbuffer = s->ringbuffer; td.temp_src = s->temp_src;
700
+
701
+    ctx->internal->execute(ctx, sofalizer_convolute, &td, NULL, 2);
702
+    emms_c();
703
+
704
+    /* display error message if clipping occured */
705
+    if (n_clippings[0] + n_clippings[1] > 0) {
706
+        av_log(ctx, AV_LOG_WARNING, "%d of %d samples clipped. Please reduce gain.\n",
707
+               n_clippings[0] + n_clippings[1], out->nb_samples * 2);
708
+    }
709
+
710
+    av_frame_free(&in);
711
+    return ff_filter_frame(outlink, out);
712
+}
713
+
714
+static int query_formats(AVFilterContext *ctx)
715
+{
716
+    struct SOFAlizerContext *s = ctx->priv;
717
+    AVFilterFormats *formats = NULL;
718
+    AVFilterChannelLayouts *layouts = NULL;
719
+    int ret, sample_rates[] = { 48000, -1 };
720
+    static const uint64_t channel_layouts[] = { AV_CH_LAYOUT_MONO,
721
+                                                AV_CH_LAYOUT_STEREO,
722
+                                                AV_CH_LAYOUT_2POINT1,
723
+                                                AV_CH_LAYOUT_SURROUND,
724
+                                                AV_CH_LAYOUT_2_1,
725
+                                                AV_CH_LAYOUT_4POINT0,
726
+                                                AV_CH_LAYOUT_QUAD,
727
+                                                AV_CH_LAYOUT_2_2,
728
+                                                AV_CH_LAYOUT_3POINT1,
729
+                                                AV_CH_LAYOUT_5POINT0_BACK,
730
+                                                AV_CH_LAYOUT_5POINT0,
731
+                                                AV_CH_LAYOUT_4POINT1,
732
+                                                AV_CH_LAYOUT_5POINT1_BACK,
733
+                                                AV_CH_LAYOUT_5POINT1,
734
+                                                AV_CH_LAYOUT_6POINT0,
735
+                                                AV_CH_LAYOUT_HEXAGONAL,
736
+                                                AV_CH_LAYOUT_6POINT1,
737
+                                                AV_CH_LAYOUT_6POINT1_BACK,
738
+                                                AV_CH_LAYOUT_7POINT0,
739
+                                                AV_CH_LAYOUT_7POINT1,
740
+                                                AV_CH_LAYOUT_OCTAGONAL,
741
+                                                0, };
742
+
743
+    ret = ff_add_format(&formats, AV_SAMPLE_FMT_FLT);
744
+    if (ret)
745
+        return ret;
746
+    ret = ff_set_common_formats(ctx, formats);
747
+    if (ret)
748
+        return ret;
749
+
750
+    layouts = ff_make_formatu64_list(channel_layouts);
751
+    if (!layouts)
752
+        return AVERROR(ENOMEM);
753
+
754
+    ret = ff_channel_layouts_ref(layouts, &ctx->inputs[0]->out_channel_layouts);
755
+    if (ret)
756
+        return ret;
757
+
758
+    layouts = NULL;
759
+    ret = ff_add_channel_layout(&layouts, AV_CH_LAYOUT_STEREO);
760
+    if (ret)
761
+        return ret;
762
+
763
+    ret = ff_channel_layouts_ref(layouts, &ctx->outputs[0]->in_channel_layouts);
764
+    if (ret)
765
+        return ret;
766
+
767
+    sample_rates[0] = s->sample_rate;
768
+    formats = ff_make_format_list(sample_rates);
769
+    if (!formats)
770
+        return AVERROR(ENOMEM);
771
+    return ff_set_common_samplerates(ctx, formats);
772
+}
773
+
774
+static int load_data(AVFilterContext *ctx, int azim, int elev, float radius)
775
+{
776
+    struct SOFAlizerContext *s = ctx->priv;
777
+    const int n_samples = s->sofa.n_samples;
778
+    int n_conv = s->n_conv; /* no. channels to convolve (excl. LFE) */
779
+    int delay_l[10]; /* broadband delay for each IR */
780
+    int delay_r[10];
781
+    int nb_input_channels = ctx->inputs[0]->channels; /* no. input channels */
782
+    float gain_lin = expf((s->gain - 3 * nb_input_channels) / 20 * M_LN10); /* gain - 3dB/channel */
783
+    float *data_ir_l = NULL;
784
+    float *data_ir_r = NULL;
785
+    int offset = 0; /* used for faster pointer arithmetics in for-loop */
786
+    int m[s->n_conv]; /* measurement index m of IR closest to required source positions */
787
+    int i, j, azim_orig = azim;
788
+
789
+    if (!s->sofa.ncid) { /* if an invalid SOFA file has been selected */
790
+        av_log(ctx, AV_LOG_ERROR, "Selected SOFA file is invalid. Please select valid SOFA file.\n");
791
+        return AVERROR_INVALIDDATA;
792
+    }
793
+
794
+    /* get temporary IR for L and R channel */
795
+    data_ir_l = av_malloc_array(n_conv * n_samples, sizeof(*data_ir_l));
796
+    data_ir_r = av_malloc_array(n_conv * n_samples, sizeof(*data_ir_r));
797
+    if (!data_ir_r || !data_ir_l) {
798
+        av_free(data_ir_l);
799
+        av_free(data_ir_r);
800
+        return AVERROR(ENOMEM);
801
+    }
802
+
803
+    for (i = 0; i < s->n_conv; i++) {
804
+        /* load and store IRs and corresponding delays */
805
+        azim = (int)(s->speaker_pos[i] + azim_orig) % 360;
806
+        /* get id of IR closest to desired position */
807
+        m[i] = find_m(s, azim, elev, radius);
808
+
809
+        /* load the delays associated with the current IRs */
810
+        delay_l[i] = *(s->sofa.data_delay + 2 * m[i]);
811
+        delay_r[i] = *(s->sofa.data_delay + 2 * m[i] + 1);
812
+
813
+        offset = i * n_samples; /* no. samples already written */
814
+        for (j = 0; j < n_samples; j++) {
815
+            /* load reversed IRs of the specified source position
816
+             * sample-by-sample for left and right ear; and apply gain */
817
+            *(data_ir_l + offset + j) = /* left channel */
818
+            *(s->sofa.data_ir + 2 * m[i] * n_samples + n_samples - 1 - j) * gain_lin;
819
+            *(data_ir_r + offset + j) = /* right channel */
820
+            *(s->sofa.data_ir + 2 * m[i] * n_samples + n_samples - 1 - j  + n_samples) * gain_lin;
821
+        }
822
+
823
+        av_log(ctx, AV_LOG_DEBUG, "Index: %d, Azimuth: %f, Elevation: %f, Radius: %f of SOFA file.\n",
824
+               m[i], *(s->sofa.sp_a + m[i]), *(s->sofa.sp_e + m[i]), *(s->sofa.sp_r + m[i]));
825
+    }
826
+
827
+    /* copy IRs and delays to allocated memory in the SOFAlizerContext struct: */
828
+    memcpy(s->data_ir[0], data_ir_l, sizeof(float) * n_conv * n_samples);
829
+    memcpy(s->data_ir[1], data_ir_r, sizeof(float) * n_conv * n_samples);
830
+
831
+    av_free(data_ir_l); /* free temporary IR memory */
832
+    av_free(data_ir_r);
833
+
834
+    memcpy(s->delay[0], &delay_l[0], sizeof(int) * s->n_conv);
835
+    memcpy(s->delay[1], &delay_r[0], sizeof(int) * s->n_conv);
836
+
837
+    return 0;
838
+}
839
+
840
+static av_cold int init(AVFilterContext *ctx)
841
+{
842
+    SOFAlizerContext *s = ctx->priv;
843
+    int ret;
844
+
845
+    /* load SOFA file, */
846
+    /* initialize file IDs to 0 before attempting to load SOFA files,
847
+     * this assures that in case of error, only the memory of already
848
+     * loaded files is free'd */
849
+    s->sofa.ncid = 0;
850
+    ret = load_sofa(ctx, s->filename, &s->sample_rate);
851
+    if (ret) {
852
+        /* file loading error */
853
+        av_log(ctx, AV_LOG_ERROR, "Error while loading SOFA file: '%s'\n", s->filename);
854
+    } else { /* no file loading error, resampling not required */
855
+        av_log(ctx, AV_LOG_DEBUG, "File '%s' loaded.\n", s->filename);
856
+    }
857
+
858
+    if (ret) {
859
+        av_log(ctx, AV_LOG_ERROR, "No valid SOFA file could be loaded. Please specify valid SOFA file.\n");
860
+        return ret;
861
+    }
862
+
863
+    s->fdsp = avpriv_float_dsp_alloc(0);
864
+    if (!s->fdsp)
865
+        return AVERROR(ENOMEM);
866
+
867
+    return 0;
868
+}
869
+
870
+static inline unsigned clz(unsigned x)
871
+{
872
+    unsigned i = sizeof(x) * 8;
873
+
874
+    while (x) {
875
+        x >>= 1;
876
+        i--;
877
+    }
878
+
879
+    return i;
880
+}
881
+
882
+static int config_input(AVFilterLink *inlink)
883
+{
884
+    AVFilterContext *ctx = inlink->dst;
885
+    SOFAlizerContext *s = ctx->priv;
886
+    int nb_input_channels = inlink->channels; /* no. input channels */
887
+    int n_max_ir = 0;
888
+    int n_current;
889
+    int n_max = 0;
890
+    int ret;
891
+
892
+    /* gain -3 dB per channel, -6 dB to get LFE on a similar level */
893
+    s->gain_lfe = expf((s->gain - 3 * inlink->channels - 6) / 20 * M_LN10);
894
+
895
+    s->lfe = !!(inlink->channel_layout & AV_CH_LOW_FREQUENCY);
896
+    /* LFE is an input channel but requires no convolution */
897
+    s->n_conv = nb_input_channels - s->lfe;
898
+
899
+    /* get size of ringbuffer (longest IR plus max. delay) */
900
+    /* then choose next power of 2 for performance optimization */
901
+    n_current = s->sofa.n_samples + max_delay(&s->sofa);
902
+    if (n_current > n_max) {
903
+        /* length of longest IR plus max. delay (in all SOFA files) */
904
+        n_max = n_current;
905
+        /* length of longest IR (without delay, in all SOFA files) */
906
+        n_max_ir = s->sofa.n_samples;
907
+    }
908
+    /* buffer length is longest IR plus max. delay -> next power of 2
909
+       (32 - count leading zeros gives required exponent)  */
910
+    s->buffer_length = exp2(32 - clz((uint32_t)n_max));
911
+
912
+    /* Allocate memory for the impulse responses, delays and the ringbuffers */
913
+    /* size: (longest IR) * (number of channels to convolute), without LFE */
914
+    s->data_ir[0] = av_malloc_array(n_max_ir, sizeof(float) * s->n_conv);
915
+    s->data_ir[1] = av_malloc_array(n_max_ir, sizeof(float) * s->n_conv);
916
+    /* length:  number of channels to convolute */
917
+    s->delay[0] = av_malloc_array(s->n_conv, sizeof(float));
918
+    s->delay[1] = av_malloc_array(s->n_conv, sizeof(float));
919
+    /* length: (buffer length) * (number of input channels),
920
+     * OR: buffer length (if frequency domain processing)
921
+     * calloc zero-initializes the buffer */
922
+    s->ringbuffer[0] = av_calloc(s->buffer_length, sizeof(float) * nb_input_channels);
923
+    s->ringbuffer[1] = av_calloc(s->buffer_length, sizeof(float) * nb_input_channels);
924
+    /* length: number of channels to convolute */
925
+    s->speaker_pos = av_malloc_array(s->n_conv, sizeof(*s->speaker_pos));
926
+
927
+    /* memory allocation failed: */
928
+    if (!s->data_ir[0] || !s->data_ir[1] || !s->delay[1] ||
929
+        !s->delay[0] || !s->ringbuffer[0] || !s->ringbuffer[1] ||
930
+        !s->speaker_pos)
931
+        return AVERROR(ENOMEM);
932
+
933
+    compensate_volume(ctx);
934
+
935
+    /* get speaker positions */
936
+    if ((ret = get_speaker_pos(ctx, s->speaker_pos)) < 0) {
937
+        av_log(ctx, AV_LOG_ERROR, "Couldn't get speaker positions. Input channel configuration not supported.\n");
938
+        return ret;
939
+    }
940
+    /* load IRs to data_ir[0] and data_ir[1] for required directions */
941
+    /* only load IRs if time-domain convolution is used. */
942
+    if ((ret = load_data(ctx, s->rotation, s->elevation, s->radius)) < 0)
943
+        return ret;
944
+
945
+    av_log(ctx, AV_LOG_DEBUG, "Samplerate: %d Channels to convolute: %d, Length of ringbuffer: %d x %d\n",
946
+        inlink->sample_rate, s->n_conv, nb_input_channels, s->buffer_length);
947
+
948
+    return 0;
949
+}
950
+
951
+static av_cold void uninit(AVFilterContext *ctx)
952
+{
953
+    SOFAlizerContext *s = ctx->priv;
954
+
955
+    if (s->sofa.ncid) {
956
+        av_freep(&s->sofa.sp_a);
957
+        av_freep(&s->sofa.sp_e);
958
+        av_freep(&s->sofa.sp_r);
959
+        av_freep(&s->sofa.data_delay);
960
+        av_freep(&s->sofa.data_ir);
961
+    }
962
+    av_freep(&s->delay[0]);
963
+    av_freep(&s->delay[1]);
964
+    av_freep(&s->data_ir[0]);
965
+    av_freep(&s->data_ir[1]);
966
+    av_freep(&s->ringbuffer[0]);
967
+    av_freep(&s->ringbuffer[1]);
968
+    av_freep(&s->speaker_pos);
969
+    av_freep(&s->temp_src[0]);
970
+    av_freep(&s->temp_src[1]);
971
+    av_freep(&s->fdsp);
972
+}
973
+
974
+#define OFFSET(x) offsetof(SOFAlizerContext, x)
975
+#define FLAGS AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
976
+
977
+static const AVOption sofalizer_options[] = {
978
+    { "sofa",      "sofa filename",  OFFSET(filename),  AV_OPT_TYPE_STRING, {.str=NULL},            .flags = FLAGS },
979
+    { "gain",      "set gain in dB", OFFSET(gain),      AV_OPT_TYPE_FLOAT,  {.dbl=0},     -20,  40, .flags = FLAGS },
980
+    { "rotation",  "set rotation"  , OFFSET(rotation),  AV_OPT_TYPE_FLOAT,  {.dbl=0},    -360, 360, .flags = FLAGS },
981
+    { "elevation", "set elevation",  OFFSET(elevation), AV_OPT_TYPE_FLOAT,  {.dbl=0},     -90,  90, .flags = FLAGS },
982
+    { "radius",    "set radius",     OFFSET(radius),    AV_OPT_TYPE_FLOAT,  {.dbl=1},       0,   3, .flags = FLAGS },
983
+    { NULL }
984
+};
985
+
986
+AVFILTER_DEFINE_CLASS(sofalizer);
987
+
988
+static const AVFilterPad inputs[] = {
989
+    {
990
+        .name         = "default",
991
+        .type         = AVMEDIA_TYPE_AUDIO,
992
+        .config_props = config_input,
993
+        .filter_frame = filter_frame,
994
+    },
995
+    { NULL }
996
+};
997
+
998
+static const AVFilterPad outputs[] = {
999
+    {
1000
+        .name = "default",
1001
+        .type = AVMEDIA_TYPE_AUDIO,
1002
+    },
1003
+    { NULL }
1004
+};
1005
+
1006
+AVFilter ff_af_sofalizer = {
1007
+    .name          = "sofalizer",
1008
+    .description   = NULL_IF_CONFIG_SMALL("SOFAlizer (Spatially Oriented Format for Acoustics)."),
1009
+    .priv_size     = sizeof(SOFAlizerContext),
1010
+    .priv_class    = &sofalizer_class,
1011
+    .init          = init,
1012
+    .uninit        = uninit,
1013
+    .query_formats = query_formats,
1014
+    .inputs        = inputs,
1015
+    .outputs       = outputs,
1016
+    .flags         = AVFILTER_FLAG_SLICE_THREADS,
1017
+};
... ...
@@ -109,6 +109,7 @@ void avfilter_register_all(void)
109 109
     REGISTER_FILTER(SIDECHAINGATE,  sidechaingate,  af);
110 110
     REGISTER_FILTER(SILENCEDETECT,  silencedetect,  af);
111 111
     REGISTER_FILTER(SILENCEREMOVE,  silenceremove,  af);
112
+    REGISTER_FILTER(SOFALIZER,      sofalizer,      af);
112 113
     REGISTER_FILTER(STEREOTOOLS,    stereotools,    af);
113 114
     REGISTER_FILTER(STEREOWIDEN,    stereowiden,    af);
114 115
     REGISTER_FILTER(TREBLE,         treble,         af);
... ...
@@ -289,6 +289,17 @@ AVFilterFormats *ff_make_format_list(const int *fmts)
289 289
     return formats;
290 290
 }
291 291
 
292
+AVFilterChannelLayouts *ff_make_formatu64_list(const uint64_t *fmts)
293
+{
294
+    MAKE_FORMAT_LIST(AVFilterChannelLayouts,
295
+                     channel_layouts, nb_channel_layouts);
296
+    if (count)
297
+        memcpy(formats->channel_layouts, fmts,
298
+               sizeof(*formats->channel_layouts) * count);
299
+
300
+    return formats;
301
+}
302
+
292 303
 AVFilterChannelLayouts *avfilter_make_format64_list(const int64_t *fmts)
293 304
 {
294 305
     MAKE_FORMAT_LIST(AVFilterChannelLayouts,
... ...
@@ -141,6 +141,9 @@ AVFilterChannelLayouts *ff_all_channel_counts(void);
141 141
 av_warn_unused_result
142 142
 AVFilterChannelLayouts *avfilter_make_format64_list(const int64_t *fmts);
143 143
 
144
+av_warn_unused_result
145
+AVFilterChannelLayouts *ff_make_formatu64_list(const uint64_t *fmts);
146
+
144 147
 
145 148
 /**
146 149
  * A helper for query_formats() which sets all links to the same list of channel
... ...
@@ -30,7 +30,7 @@
30 30
 #include "libavutil/version.h"
31 31
 
32 32
 #define LIBAVFILTER_VERSION_MAJOR   6
33
-#define LIBAVFILTER_VERSION_MINOR  20
33
+#define LIBAVFILTER_VERSION_MINOR  21
34 34
 #define LIBAVFILTER_VERSION_MICRO 100
35 35
 
36 36
 #define LIBAVFILTER_VERSION_INT AV_VERSION_INT(LIBAVFILTER_VERSION_MAJOR, \