falkTX
/
FFmpeg
mirror of https://github.com/falkTX/FFmpeg.git
1
0
Fork 0
Code Issues 0 Releases 338 Wiki Activity
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
79113 Commits
32 Branches
292MB
Tree: 124526ba1a
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '124526ba1a'
${ noResults }
FFmpeg/libavfilter/af_sofalizer.c

1152 lines
44KB
Raw Blame History 

  1. /*****************************************************************************

  2.  * sofalizer.c : SOFAlizer filter for virtual binaural acoustics

  3.  *****************************************************************************

  4.  * Copyright (C) 2013-2015 Andreas Fuchs, Wolfgang Hrauda,

  5.  *                         Acoustics Research Institute (ARI), Vienna, Austria

  6.  *

  7.  * Authors: Andreas Fuchs <andi.fuchs.mail@gmail.com>

  8.  *          Wolfgang Hrauda <wolfgang.hrauda@gmx.at>

  9.  *

  10.  * SOFAlizer project coordinator at ARI, main developer of SOFA:

  11.  *          Piotr Majdak <piotr@majdak.at>

  12.  *

  13.  * This program is free software; you can redistribute it and/or modify it

  14.  * under the terms of the GNU Lesser General Public License as published by

  15.  * the Free Software Foundation; either version 2.1 of the License, or

  16.  * (at your option) any later version.

  17.  *

  18.  * This program is distributed in the hope that it will be useful,

  19.  * but WITHOUT ANY WARRANTY; without even the implied warranty of

  20.  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the

  21.  * GNU Lesser General Public License for more details.

  22.  *

  23.  * You should have received a copy of the GNU Lesser General Public License

  24.  * along with this program; if not, write to the Free Software Foundation,

  25.  * Inc., 51 Franklin Street, Fifth Floor, Boston MA 02110-1301, USA.

  26.  *****************************************************************************/

  27. 

  28. #include <math.h>

  29. #include <netcdf.h>

  30. 

  31. #include "libavcodec/avfft.h"

  32. #include "libavutil/float_dsp.h"

  33. #include "libavutil/intmath.h"

  34. #include "libavutil/opt.h"

  35. #include "avfilter.h"

  36. #include "internal.h"

  37. #include "audio.h"

  38. 

  39. #define TIME_DOMAIN      0

  40. #define FREQUENCY_DOMAIN 1

  41. 

  42. typedef struct NCSofa {  /* contains data of one SOFA file */

  43.     int ncid;            /* netCDF ID of the opened SOFA file */

  44.     int n_samples;       /* length of one impulse response (IR) */

  45.     int m_dim;           /* number of measurement positions */

  46.     int *data_delay;     /* broadband delay of each IR */

  47.                          /* all measurement positions for each receiver (i.e. ear): */

  48.     float *sp_a;         /* azimuth angles */

  49.     float *sp_e;         /* elevation angles */

  50.     float *sp_r;         /* radii */

  51.                          /* data at each measurement position for each receiver: */

  52.     float *data_ir;      /* IRs (time-domain) */

  53. } NCSofa;

  54. 

  55. typedef struct SOFAlizerContext {

  56.     const AVClass *class;

  57. 

  58.     char *filename;             /* name of SOFA file */

  59.     NCSofa sofa;                /* contains data of the SOFA file */

  60. 

  61.     int sample_rate;            /* sample rate from SOFA file */

  62.     float *speaker_azim;        /* azimuth of the virtual loudspeakers */

  63.     float *speaker_elev;        /* elevation of the virtual loudspeakers */

  64.     float gain_lfe;             /* gain applied to LFE channel */

  65.     int lfe_channel;            /* LFE channel position in channel layout */

  66. 

  67.     int n_conv;                 /* number of channels to convolute */

  68. 

  69.                                 /* buffer variables (for convolution) */

  70.     float *ringbuffer[2];       /* buffers input samples, length of one buffer: */

  71.                                 /* no. input ch. (incl. LFE) x buffer_length */

  72.     int write[2];               /* current write position to ringbuffer */

  73.     int buffer_length;          /* is: longest IR plus max. delay in all SOFA files */

  74.                                 /* then choose next power of 2 */

  75.     int n_fft;                  /* number of samples in one FFT block */

  76. 

  77.                                 /* netCDF variables */

  78.     int *delay[2];              /* broadband delay for each channel/IR to be convolved */

  79. 

  80.     float *data_ir[2];          /* IRs for all channels to be convolved */

  81.                                 /* (this excludes the LFE) */

  82.     float *temp_src[2];

  83.     FFTComplex *temp_fft[2];

  84. 

  85.                          /* control variables */

  86.     float gain;          /* filter gain (in dB) */

  87.     float rotation;      /* rotation of virtual loudspeakers (in degrees)  */

  88.     float elevation;     /* elevation of virtual loudspeakers (in deg.) */

  89.     float radius;        /* distance virtual loudspeakers to listener (in metres) */

  90.     int type;            /* processing type */

  91. 

  92.     FFTContext *fft[2], *ifft[2];

  93.     FFTComplex *data_hrtf[2];

  94. 

  95.     AVFloatDSPContext *fdsp;

  96. } SOFAlizerContext;

  97. 

  98. static int close_sofa(struct NCSofa *sofa)

  99. {

  100.     av_freep(&sofa->data_delay);

  101.     av_freep(&sofa->sp_a);

  102.     av_freep(&sofa->sp_e);

  103.     av_freep(&sofa->sp_r);

  104.     av_freep(&sofa->data_ir);

  105.     nc_close(sofa->ncid);

  106.     sofa->ncid = 0;

  107. 

  108.     return 0;

  109. }

  110. 

  111. static int load_sofa(AVFilterContext *ctx, char *filename, int *samplingrate)

  112. {

  113.     struct SOFAlizerContext *s = ctx->priv;

  114.     /* variables associated with content of SOFA file: */

  115.     int ncid, n_dims, n_vars, n_gatts, n_unlim_dim_id, status;

  116.     char data_delay_dim_name[NC_MAX_NAME];

  117.     float *sp_a, *sp_e, *sp_r, *data_ir;

  118.     char *sofa_conventions;

  119.     char dim_name[NC_MAX_NAME];   /* names of netCDF dimensions */

  120.     size_t *dim_length;           /* lengths of netCDF dimensions */

  121.     char *text;

  122.     unsigned int sample_rate;

  123.     int data_delay_dim_id[2];

  124.     int samplingrate_id;

  125.     int data_delay_id;

  126.     int n_samples;

  127.     int m_dim_id = -1;

  128.     int n_dim_id = -1;

  129.     int data_ir_id;

  130.     size_t att_len;

  131.     int m_dim;

  132.     int *data_delay;

  133.     int sp_id;

  134.     int i, ret;

  135. 

  136.     s->sofa.ncid = 0;

  137.     status = nc_open(filename, NC_NOWRITE, &ncid); /* open SOFA file read-only */

  138.     if (status != NC_NOERR) {

  139.         av_log(ctx, AV_LOG_ERROR, "Can't find SOFA-file '%s'\n", filename);

  140.         return AVERROR(EINVAL);

  141.     }

  142. 

  143.     /* get number of dimensions, vars, global attributes and Id of unlimited dimensions: */

  144.     nc_inq(ncid, &n_dims, &n_vars, &n_gatts, &n_unlim_dim_id);

  145. 

  146.     /* -- get number of measurements ("M") and length of one IR ("N") -- */

  147.     dim_length = av_malloc_array(n_dims, sizeof(*dim_length));

  148.     if (!dim_length) {

  149.         nc_close(ncid);

  150.         return AVERROR(ENOMEM);

  151.     }

  152. 

  153.     for (i = 0; i < n_dims; i++) { /* go through all dimensions of file */

  154.         nc_inq_dim(ncid, i, (char *)&dim_name, &dim_length[i]); /* get dimensions */

  155.         if (!strncmp("M", (const char *)&dim_name, 1)) /* get ID of dimension "M" */

  156.             m_dim_id = i;

  157.         if (!strncmp("N", (const char *)&dim_name, 1)) /* get ID of dimension "N" */

  158.             n_dim_id = i;

  159.     }

  160. 

  161.     if ((m_dim_id == -1) || (n_dim_id == -1)) { /* dimension "M" or "N" couldn't be found */

  162.         av_log(ctx, AV_LOG_ERROR, "Can't find required dimensions in SOFA file.\n");

  163.         av_freep(&dim_length);

  164.         nc_close(ncid);

  165.         return AVERROR(EINVAL);

  166.     }

  167. 

  168.     n_samples = dim_length[n_dim_id]; /* get length of one IR */

  169.     m_dim     = dim_length[m_dim_id]; /* get number of measurements */

  170. 

  171.     av_freep(&dim_length);

  172. 

  173.     /* -- check file type -- */

  174.     /* get length of attritube "Conventions" */

  175.     status = nc_inq_attlen(ncid, NC_GLOBAL, "Conventions", &att_len);

  176.     if (status != NC_NOERR) {

  177.         av_log(ctx, AV_LOG_ERROR, "Can't get length of attribute \"Conventions\".\n");

  178.         nc_close(ncid);

  179.         return AVERROR_INVALIDDATA;

  180.     }

  181. 

  182.     /* check whether file is SOFA file */

  183.     text = av_malloc(att_len + 1);

  184.     if (!text) {

  185.         nc_close(ncid);

  186.         return AVERROR(ENOMEM);

  187.     }

  188. 

  189.     nc_get_att_text(ncid, NC_GLOBAL, "Conventions", text);

  190.     *(text + att_len) = 0;

  191.     if (strncmp("SOFA", text, 4)) {

  192.         av_log(ctx, AV_LOG_ERROR, "Not a SOFA file!\n");

  193.         av_freep(&text);

  194.         nc_close(ncid);

  195.         return AVERROR(EINVAL);

  196.     }

  197.     av_freep(&text);

  198. 

  199.     status = nc_inq_attlen(ncid, NC_GLOBAL, "License", &att_len);

  200.     if (status == NC_NOERR) {

  201.         text = av_malloc(att_len + 1);

  202.         if (text) {

  203.             nc_get_att_text(ncid, NC_GLOBAL, "License", text);

  204.             *(text + att_len) = 0;

  205.             av_log(ctx, AV_LOG_INFO, "SOFA file License: %s\n", text);

  206.             av_freep(&text);

  207.         }

  208.     }

  209. 

  210.     status = nc_inq_attlen(ncid, NC_GLOBAL, "SourceDescription", &att_len);

  211.     if (status == NC_NOERR) {

  212.         text = av_malloc(att_len + 1);

  213.         if (text) {

  214.             nc_get_att_text(ncid, NC_GLOBAL, "SourceDescription", text);

  215.             *(text + att_len) = 0;

  216.             av_log(ctx, AV_LOG_INFO, "SOFA file SourceDescription: %s\n", text);

  217.             av_freep(&text);

  218.         }

  219.     }

  220. 

  221.     status = nc_inq_attlen(ncid, NC_GLOBAL, "Comment", &att_len);

  222.     if (status == NC_NOERR) {

  223.         text = av_malloc(att_len + 1);

  224.         if (text) {

  225.             nc_get_att_text(ncid, NC_GLOBAL, "Comment", text);

  226.             *(text + att_len) = 0;

  227.             av_log(ctx, AV_LOG_INFO, "SOFA file Comment: %s\n", text);

  228.             av_freep(&text);

  229.         }

  230.     }

  231. 

  232.     status = nc_inq_attlen(ncid, NC_GLOBAL, "SOFAConventions", &att_len);

  233.     if (status != NC_NOERR) {

  234.         av_log(ctx, AV_LOG_ERROR, "Can't get length of attribute \"SOFAConventions\".\n");

  235.         nc_close(ncid);

  236.         return AVERROR_INVALIDDATA;

  237.     }

  238. 

  239.     sofa_conventions = av_malloc(att_len + 1);

  240.     if (!sofa_conventions) {

  241.         nc_close(ncid);

  242.         return AVERROR(ENOMEM);

  243.     }

  244. 

  245.     nc_get_att_text(ncid, NC_GLOBAL, "SOFAConventions", sofa_conventions);

  246.     *(sofa_conventions + att_len) = 0;

  247.     if (strncmp("SimpleFreeFieldHRIR", sofa_conventions, att_len)) {

  248.         av_log(ctx, AV_LOG_ERROR, "Not a SimpleFreeFieldHRIR file!\n");

  249.         av_freep(&sofa_conventions);

  250.         nc_close(ncid);

  251.         return AVERROR(EINVAL);

  252.     }

  253.     av_freep(&sofa_conventions);

  254. 

  255.     /* -- get sampling rate of HRTFs -- */

  256.     /* read ID, then value */

  257.     status  = nc_inq_varid(ncid, "Data.SamplingRate", &samplingrate_id);

  258.     status += nc_get_var_uint(ncid, samplingrate_id, &sample_rate);

  259.     if (status != NC_NOERR) {

  260.         av_log(ctx, AV_LOG_ERROR, "Couldn't read Data.SamplingRate.\n");

  261.         nc_close(ncid);

  262.         return AVERROR(EINVAL);

  263.     }

  264.     *samplingrate = sample_rate; /* remember sampling rate */

  265. 

  266.     /* -- allocate memory for one value for each measurement position: -- */

  267.     sp_a = s->sofa.sp_a = av_malloc_array(m_dim, sizeof(float));

  268.     sp_e = s->sofa.sp_e = av_malloc_array(m_dim, sizeof(float));

  269.     sp_r = s->sofa.sp_r = av_malloc_array(m_dim, sizeof(float));

  270.     /* delay and IR values required for each ear and measurement position: */

  271.     data_delay = s->sofa.data_delay = av_calloc(m_dim, 2 * sizeof(int));

  272.     data_ir = s->sofa.data_ir = av_calloc(m_dim * FFALIGN(n_samples, 16), sizeof(float) * 2);

  273. 

  274.     if (!data_delay || !sp_a || !sp_e || !sp_r || !data_ir) {

  275.         /* if memory could not be allocated */

  276.         close_sofa(&s->sofa);

  277.         return AVERROR(ENOMEM);

  278.     }

  279. 

  280.     /* get impulse responses (HRTFs): */

  281.     /* get corresponding ID */

  282.     status = nc_inq_varid(ncid, "Data.IR", &data_ir_id);

  283.     status += nc_get_var_float(ncid, data_ir_id, data_ir); /* read and store IRs */

  284.     if (status != NC_NOERR) {

  285.         av_log(ctx, AV_LOG_ERROR, "Couldn't read Data.IR!\n");

  286.         ret = AVERROR(EINVAL);

  287.         goto error;

  288.     }

  289. 

  290.     /* get source positions of the HRTFs in the SOFA file: */

  291.     status  = nc_inq_varid(ncid, "SourcePosition", &sp_id); /* get corresponding ID */

  292.     status += nc_get_vara_float(ncid, sp_id, (size_t[2]){ 0, 0 } ,

  293.                 (size_t[2]){ m_dim, 1}, sp_a); /* read & store azimuth angles */

  294.     status += nc_get_vara_float(ncid, sp_id, (size_t[2]){ 0, 1 } ,

  295.                 (size_t[2]){ m_dim, 1}, sp_e); /* read & store elevation angles */

  296.     status += nc_get_vara_float(ncid, sp_id, (size_t[2]){ 0, 2 } ,

  297.                 (size_t[2]){ m_dim, 1}, sp_r); /* read & store radii */

  298.     if (status != NC_NOERR) { /* if any source position variable coudn't be read */

  299.         av_log(ctx, AV_LOG_ERROR, "Couldn't read SourcePosition.\n");

  300.         ret = AVERROR(EINVAL);

  301.         goto error;

  302.     }

  303. 

  304.     /* read Data.Delay, check for errors and fit it to data_delay */

  305.     status  = nc_inq_varid(ncid, "Data.Delay", &data_delay_id);

  306.     status += nc_inq_vardimid(ncid, data_delay_id, &data_delay_dim_id[0]);

  307.     status += nc_inq_dimname(ncid, data_delay_dim_id[0], data_delay_dim_name);

  308.     if (status != NC_NOERR) {

  309.         av_log(ctx, AV_LOG_ERROR, "Couldn't read Data.Delay.\n");

  310.         ret = AVERROR(EINVAL);

  311.         goto error;

  312.     }

  313. 

  314.     /* Data.Delay dimension check */

  315.     /* dimension of Data.Delay is [I R]: */

  316.     if (!strncmp(data_delay_dim_name, "I", 2)) {

  317.         /* check 2 characters to assure string is 0-terminated after "I" */

  318.         int delay[2]; /* delays get from SOFA file: */

  319.         int *data_delay_r;

  320. 

  321.         av_log(ctx, AV_LOG_DEBUG, "Data.Delay has dimension [I R]\n");

  322.         status = nc_get_var_int(ncid, data_delay_id, &delay[0]);

  323.         if (status != NC_NOERR) {

  324.             av_log(ctx, AV_LOG_ERROR, "Couldn't read Data.Delay\n");

  325.             ret = AVERROR(EINVAL);

  326.             goto error;

  327.         }

  328.         data_delay_r = data_delay + m_dim;

  329.         for (i = 0; i < m_dim; i++) { /* extend given dimension [I R] to [M R] */

  330.             /* assign constant delay value for all measurements to data_delay fields */

  331.             data_delay[i]   = delay[0];

  332.             data_delay_r[i] = delay[1];

  333.         }

  334.         /* dimension of Data.Delay is [M R] */

  335.     } else if (!strncmp(data_delay_dim_name, "M", 2)) {

  336.         av_log(ctx, AV_LOG_ERROR, "Data.Delay in dimension [M R]\n");

  337.         /* get delays from SOFA file: */

  338.         status = nc_get_var_int(ncid, data_delay_id, data_delay);

  339.         if (status != NC_NOERR) {

  340.             av_log(ctx, AV_LOG_ERROR, "Couldn't read Data.Delay\n");

  341.             ret = AVERROR(EINVAL);

  342.             goto error;

  343.         }

  344.     } else { /* dimension of Data.Delay is neither [I R] nor [M R] */

  345.         av_log(ctx, AV_LOG_ERROR, "Data.Delay does not have the required dimensions [I R] or [M R].\n");

  346.         ret = AVERROR(EINVAL);

  347.         goto error;

  348.     }

  349. 

  350.     /* save information in SOFA struct: */

  351.     s->sofa.m_dim = m_dim; /* no. measurement positions */

  352.     s->sofa.n_samples = n_samples; /* length on one IR */

  353.     s->sofa.ncid = ncid; /* netCDF ID of SOFA file */

  354.     nc_close(ncid); /* close SOFA file */

  355. 

  356.     av_log(ctx, AV_LOG_DEBUG, "m_dim: %d n_samples %d\n", m_dim, n_samples);

  357. 

  358.     return 0;

  359. 

  360. error:

  361.     close_sofa(&s->sofa);

  362.     return ret;

  363. }

  364. 

  365. static int get_speaker_pos(AVFilterContext *ctx,

  366.                            float *speaker_azim, float *speaker_elev)

  367. {

  368.     struct SOFAlizerContext *s = ctx->priv;

  369.     uint64_t channels_layout = ctx->inputs[0]->channel_layout;

  370.     float azim[16] = { 0 };

  371.     float elev[16] = { 0 };

  372.     int m, ch, n_conv = ctx->inputs[0]->channels; /* get no. input channels */

  373. 

  374.     if (n_conv > 16)

  375.         return AVERROR(EINVAL);

  376. 

  377.     s->lfe_channel = -1;

  378. 

  379.     /* set speaker positions according to input channel configuration: */

  380.     for (m = 0, ch = 0; ch < n_conv && m < 64; m++) {

  381.         uint64_t mask = channels_layout & (1 << m);

  382. 

  383.         switch (mask) {

  384.         case AV_CH_FRONT_LEFT:            azim[ch] =  30;      break;

  385.         case AV_CH_FRONT_RIGHT:           azim[ch] = 330;      break;

  386.         case AV_CH_FRONT_CENTER:          azim[ch] =   0;      break;

  387.         case AV_CH_LOW_FREQUENCY:

  388.         case AV_CH_LOW_FREQUENCY_2:       s->lfe_channel = ch; break;

  389.         case AV_CH_BACK_LEFT:             azim[ch] = 150;      break;

  390.         case AV_CH_BACK_RIGHT:            azim[ch] = 210;      break;

  391.         case AV_CH_BACK_CENTER:           azim[ch] = 180;      break;

  392.         case AV_CH_SIDE_LEFT:             azim[ch] =  90;      break;

  393.         case AV_CH_SIDE_RIGHT:            azim[ch] = 270;      break;

  394.         case AV_CH_FRONT_LEFT_OF_CENTER:  azim[ch] =  15;      break;

  395.         case AV_CH_FRONT_RIGHT_OF_CENTER: azim[ch] = 345;      break;

  396.         case AV_CH_TOP_CENTER:            azim[ch] =   0;

  397.                                           elev[ch] =  90;      break;

  398.         case AV_CH_TOP_FRONT_LEFT:        azim[ch] =  30;

  399.                                           elev[ch] =  45;      break;

  400.         case AV_CH_TOP_FRONT_CENTER:      azim[ch] =   0;

  401.                                           elev[ch] =  45;      break;

  402.         case AV_CH_TOP_FRONT_RIGHT:       azim[ch] = 330;

  403.                                           elev[ch] =  45;      break;

  404.         case AV_CH_TOP_BACK_LEFT:         azim[ch] = 150;

  405.                                           elev[ch] =  45;      break;

  406.         case AV_CH_TOP_BACK_RIGHT:        azim[ch] = 210;

  407.                                           elev[ch] =  45;      break;

  408.         case AV_CH_TOP_BACK_CENTER:       azim[ch] = 180;

  409.                                           elev[ch] =  45;      break;

  410.         case AV_CH_WIDE_LEFT:             azim[ch] =  90;      break;

  411.         case AV_CH_WIDE_RIGHT:            azim[ch] = 270;      break;

  412.         case AV_CH_SURROUND_DIRECT_LEFT:  azim[ch] =  90;      break;

  413.         case AV_CH_SURROUND_DIRECT_RIGHT: azim[ch] = 270;      break;

  414.         case AV_CH_STEREO_LEFT:           azim[ch] =  90;      break;

  415.         case AV_CH_STEREO_RIGHT:          azim[ch] = 270;      break;

  416.         case 0:                                                break;

  417.         default:

  418.             return AVERROR(EINVAL);

  419.         }

  420.         if (mask)

  421.             ch++;

  422.     }

  423. 

  424.     memcpy(speaker_azim, azim, n_conv * sizeof(float));

  425.     memcpy(speaker_elev, elev, n_conv * sizeof(float));

  426. 

  427.     return 0;

  428. 

  429. }

  430. 

  431. static int max_delay(struct NCSofa *sofa)

  432. {

  433.     int i, max = 0;

  434. 

  435.     for (i = 0; i < sofa->m_dim * 2; i++) {

  436.         /* search maximum delay in given SOFA file */

  437.         max = FFMAX(max, sofa->data_delay[i]);

  438.     }

  439. 

  440.     return max;

  441. }

  442. 

  443. static int find_m(SOFAlizerContext *s, int azim, int elev, float radius)

  444. {

  445.     /* get source positions and M of currently selected SOFA file */

  446.     float *sp_a = s->sofa.sp_a; /* azimuth angle */

  447.     float *sp_e = s->sofa.sp_e; /* elevation angle */

  448.     float *sp_r = s->sofa.sp_r; /* radius */

  449.     int m_dim = s->sofa.m_dim; /* no. measurements */

  450.     int best_id = 0; /* index m currently closest to desired source pos. */

  451.     float delta = 1000; /* offset between desired and currently best pos. */

  452.     float current;

  453.     int i;

  454. 

  455.     for (i = 0; i < m_dim; i++) {

  456.         /* search through all measurements in currently selected SOFA file */

  457.         /* distance of current to desired source position: */

  458.         current = fabs(sp_a[i] - azim) +

  459.                   fabs(sp_e[i] - elev) +

  460.                   fabs(sp_r[i] - radius);

  461.         if (current <= delta) {

  462.             /* if current distance is smaller than smallest distance so far */

  463.             delta = current;

  464.             best_id = i; /* remember index */

  465.         }

  466.     }

  467. 

  468.     return best_id;

  469. }

  470. 

  471. static int compensate_volume(AVFilterContext *ctx)

  472. {

  473.     struct SOFAlizerContext *s = ctx->priv;

  474.     float compensate;

  475.     float energy = 0;

  476.     float *ir;

  477.     int m;

  478. 

  479.     if (s->sofa.ncid) {

  480.         /* find IR at front center position in the SOFA file (IR closest to 0°,0°,1m) */

  481.         struct NCSofa *sofa = &s->sofa;

  482.         m = find_m(s, 0, 0, 1);

  483.         /* get energy of that IR and compensate volume */

  484.         ir = sofa->data_ir + 2 * m * sofa->n_samples;

  485.         if (sofa->n_samples & 31) {

  486.             energy = avpriv_scalarproduct_float_c(ir, ir, sofa->n_samples);

  487.         } else {

  488.             energy = s->fdsp->scalarproduct_float(ir, ir, sofa->n_samples);

  489.         }

  490.         compensate = 256 / (sofa->n_samples * sqrt(energy));

  491.         av_log(ctx, AV_LOG_DEBUG, "Compensate-factor: %f\n", compensate);

  492.         ir = sofa->data_ir;

  493.         /* apply volume compensation to IRs */

  494.         if (sofa->n_samples & 31) {

  495.             int i;

  496.             for (i = 0; i < sofa->n_samples * sofa->m_dim * 2; i++) {

  497.                 ir[i] = ir[i] * compensate;

  498.             }

  499.         } else {

  500.             s->fdsp->vector_fmul_scalar(ir, ir, compensate, sofa->n_samples * sofa->m_dim * 2);

  501.             emms_c();

  502.         }

  503.     }

  504. 

  505.     return 0;

  506. }

  507. 

  508. typedef struct ThreadData {

  509.     AVFrame *in, *out;

  510.     int *write;

  511.     int **delay;

  512.     float **ir;

  513.     int *n_clippings;

  514.     float **ringbuffer;

  515.     float **temp_src;

  516.     FFTComplex **temp_fft;

  517. } ThreadData;

  518. 

  519. static int sofalizer_convolute(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)

  520. {

  521.     SOFAlizerContext *s = ctx->priv;

  522.     ThreadData *td = arg;

  523.     AVFrame *in = td->in, *out = td->out;

  524.     int offset = jobnr;

  525.     int *write = &td->write[jobnr];

  526.     const int *const delay = td->delay[jobnr];

  527.     const float *const ir = td->ir[jobnr];

  528.     int *n_clippings = &td->n_clippings[jobnr];

  529.     float *ringbuffer = td->ringbuffer[jobnr];

  530.     float *temp_src = td->temp_src[jobnr];

  531.     const int n_samples = s->sofa.n_samples; /* length of one IR */

  532.     const float *src = (const float *)in->data[0]; /* get pointer to audio input buffer */

  533.     float *dst = (float *)out->data[0]; /* get pointer to audio output buffer */

  534.     const int in_channels = s->n_conv; /* number of input channels */

  535.     /* ring buffer length is: longest IR plus max. delay -> next power of 2 */

  536.     const int buffer_length = s->buffer_length;

  537.     /* -1 for AND instead of MODULO (applied to powers of 2): */

  538.     const uint32_t modulo = (uint32_t)buffer_length - 1;

  539.     float *buffer[16]; /* holds ringbuffer for each input channel */

  540.     int wr = *write;

  541.     int read;

  542.     int i, l;

  543. 

  544.     dst += offset;

  545.     for (l = 0; l < in_channels; l++) {

  546.         /* get starting address of ringbuffer for each input channel */

  547.         buffer[l] = ringbuffer + l * buffer_length;

  548.     }

  549. 

  550.     for (i = 0; i < in->nb_samples; i++) {

  551.         const float *temp_ir = ir; /* using same set of IRs for each sample */

  552. 

  553.         *dst = 0;

  554.         for (l = 0; l < in_channels; l++) {

  555.             /* write current input sample to ringbuffer (for each channel) */

  556.             *(buffer[l] + wr) = src[l];

  557.         }

  558. 

  559.         /* loop goes through all channels to be convolved */

  560.         for (l = 0; l < in_channels; l++) {

  561.             const float *const bptr = buffer[l];

  562. 

  563.             if (l == s->lfe_channel) {

  564.                 /* LFE is an input channel but requires no convolution */

  565.                 /* apply gain to LFE signal and add to output buffer */

  566.                 *dst += *(buffer[s->lfe_channel] + wr) * s->gain_lfe;

  567.                 temp_ir += FFALIGN(n_samples, 16);

  568.                 continue;

  569.             }

  570. 

  571.             /* current read position in ringbuffer: input sample write position

  572.              * - delay for l-th ch. + diff. betw. IR length and buffer length

  573.              * (mod buffer length) */

  574.             read = (wr - *(delay + l) - (n_samples - 1) + buffer_length) & modulo;

  575. 

  576.             if (read + n_samples < buffer_length) {

  577.                 memcpy(temp_src, bptr + read, n_samples * sizeof(*temp_src));

  578.             } else {

  579.                 int len = FFMIN(n_samples - (read % n_samples), buffer_length - read);

  580. 

  581.                 memcpy(temp_src, bptr + read, len * sizeof(*temp_src));

  582.                 memcpy(temp_src + len, bptr, (n_samples - len) * sizeof(*temp_src));

  583.             }

  584. 

  585.             /* multiply signal and IR, and add up the results */

  586.             dst[0] += s->fdsp->scalarproduct_float(temp_ir, temp_src, n_samples);

  587.             temp_ir += FFALIGN(n_samples, 16);

  588.         }

  589. 

  590.         /* clippings counter */

  591.         if (fabs(*dst) > 1)

  592.             *n_clippings += 1;

  593. 

  594.         /* move output buffer pointer by +2 to get to next sample of processed channel: */

  595.         dst += 2;

  596.         src += in_channels;

  597.         wr   = (wr + 1) & modulo; /* update ringbuffer write position */

  598.     }

  599. 

  600.     *write = wr; /* remember write position in ringbuffer for next call */

  601. 

  602.     return 0;

  603. }

  604. 

  605. static int sofalizer_fast_convolute(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)

  606. {

  607.     SOFAlizerContext *s = ctx->priv;

  608.     ThreadData *td = arg;

  609.     AVFrame *in = td->in, *out = td->out;

  610.     int offset = jobnr;

  611.     int *write = &td->write[jobnr];

  612.     FFTComplex *hrtf = s->data_hrtf[jobnr]; /* get pointers to current HRTF data */

  613.     int *n_clippings = &td->n_clippings[jobnr];

  614.     float *ringbuffer = td->ringbuffer[jobnr];

  615.     const int n_samples = s->sofa.n_samples; /* length of one IR */

  616.     const float *src = (const float *)in->data[0]; /* get pointer to audio input buffer */

  617.     float *dst = (float *)out->data[0]; /* get pointer to audio output buffer */

  618.     const int in_channels = s->n_conv; /* number of input channels */

  619.     /* ring buffer length is: longest IR plus max. delay -> next power of 2 */

  620.     const int buffer_length = s->buffer_length;

  621.     /* -1 for AND instead of MODULO (applied to powers of 2): */

  622.     const uint32_t modulo = (uint32_t)buffer_length - 1;

  623.     FFTComplex *fft_in = s->temp_fft[jobnr]; /* temporary array for FFT input/output data */

  624.     FFTContext *ifft = s->ifft[jobnr];

  625.     FFTContext *fft = s->fft[jobnr];

  626.     const int n_conv = s->n_conv;

  627.     const int n_fft = s->n_fft;

  628.     int wr = *write;

  629.     int n_read;

  630.     int i, j;

  631. 

  632.     dst += offset;

  633. 

  634.     /* find minimum between number of samples and output buffer length:

  635.      * (important, if one IR is longer than the output buffer) */

  636.     n_read = FFMIN(s->sofa.n_samples, in->nb_samples);

  637.     for (j = 0; j < n_read; j++) {

  638.         /* initialize output buf with saved signal from overflow buf */

  639.         dst[2 * j]     = ringbuffer[wr];

  640.         ringbuffer[wr] = 0.0; /* re-set read samples to zero */

  641.         /* update ringbuffer read/write position */

  642.         wr  = (wr + 1) & modulo;

  643.     }

  644. 

  645.     /* initialize rest of output buffer with 0 */

  646.     for (j = n_read; j < in->nb_samples; j++) {

  647.         dst[2 * j] = 0;

  648.     }

  649. 

  650.     for (i = 0; i < n_conv; i++) {

  651.         if (i == s->lfe_channel) { /* LFE */

  652.             for (j = 0; j < in->nb_samples; j++) {

  653.                 /* apply gain to LFE signal and add to output buffer */

  654.                 dst[2 * j] += src[i + j * in_channels] * s->gain_lfe;

  655.             }

  656.             continue;

  657.         }

  658. 

  659.         /* outer loop: go through all input channels to be convolved */

  660.         offset = i * n_fft; /* no. samples already processed */

  661. 

  662.         /* fill FFT input with 0 (we want to zero-pad) */

  663.         memset(fft_in, 0, sizeof(FFTComplex) * n_fft);

  664. 

  665.         for (j = 0; j < in->nb_samples; j++) {

  666.             /* prepare input for FFT */

  667.             /* write all samples of current input channel to FFT input array */

  668.             fft_in[j].re = src[j * in_channels + i];

  669.         }

  670. 

  671.         /* transform input signal of current channel to frequency domain */

  672.         av_fft_permute(fft, fft_in);

  673.         av_fft_calc(fft, fft_in);

  674.         for (j = 0; j < n_fft; j++) {

  675.             const float re = fft_in[j].re;

  676.             const float im = fft_in[j].im;

  677. 

  678.             /* complex multiplication of input signal and HRTFs */

  679.             /* output channel (real): */

  680.             fft_in[j].re = re * (hrtf + offset + j)->re - im * (hrtf + offset + j)->im;

  681.             /* output channel (imag): */

  682.             fft_in[j].im = re * (hrtf + offset + j)->im + im * (hrtf + offset + j)->re;

  683.         }

  684. 

  685.         /* transform output signal of current channel back to time domain */

  686.         av_fft_permute(ifft, fft_in);

  687.         av_fft_calc(ifft, fft_in);

  688. 

  689.         for (j = 0; j < in->nb_samples; j++) {

  690.             /* write output signal of current channel to output buffer */

  691.             dst[2 * j] += fft_in[j].re / (float)n_fft;

  692.         }

  693. 

  694.         for (j = 0; j < n_samples - 1; j++) { /* overflow length is IR length - 1 */

  695.             /* write the rest of output signal to overflow buffer */

  696.             int write_pos = (wr + j) & modulo;

  697. 

  698.             *(ringbuffer + write_pos) += fft_in[in->nb_samples + j].re / (float)n_fft;

  699.         }

  700.     }

  701. 

  702.     /* go through all samples of current output buffer: count clippings */

  703.     for (i = 0; i < out->nb_samples; i++) {

  704.         /* clippings counter */

  705.         if (fabs(*dst) > 1) { /* if current output sample > 1 */

  706.             *n_clippings = *n_clippings + 1;

  707.         }

  708. 

  709.         /* move output buffer pointer by +2 to get to next sample of processed channel: */

  710.         dst += 2;

  711.     }

  712. 

  713.     /* remember read/write position in ringbuffer for next call */

  714.     *write = wr;

  715. 

  716.     return 0;

  717. }

  718. 

  719. static int filter_frame(AVFilterLink *inlink, AVFrame *in)

  720. {

  721.     AVFilterContext *ctx = inlink->dst;

  722.     SOFAlizerContext *s = ctx->priv;

  723.     AVFilterLink *outlink = ctx->outputs[0];

  724.     int n_clippings[2] = { 0 };

  725.     ThreadData td;

  726.     AVFrame *out;

  727. 

  728.     out = ff_get_audio_buffer(outlink, in->nb_samples);

  729.     if (!out) {

  730.         av_frame_free(&in);

  731.         return AVERROR(ENOMEM);

  732.     }

  733.     av_frame_copy_props(out, in);

  734. 

  735.     td.in = in; td.out = out; td.write = s->write;

  736.     td.delay = s->delay; td.ir = s->data_ir; td.n_clippings = n_clippings;

  737.     td.ringbuffer = s->ringbuffer; td.temp_src = s->temp_src;

  738.     td.temp_fft = s->temp_fft;

  739. 

  740.     if (s->type == TIME_DOMAIN) {

  741.         ctx->internal->execute(ctx, sofalizer_convolute, &td, NULL, 2);

  742.     } else {

  743.         ctx->internal->execute(ctx, sofalizer_fast_convolute, &td, NULL, 2);

  744.     }

  745.     emms_c();

  746. 

  747.     /* display error message if clipping occurred */

  748.     if (n_clippings[0] + n_clippings[1] > 0) {

  749.         av_log(ctx, AV_LOG_WARNING, "%d of %d samples clipped. Please reduce gain.\n",

  750.                n_clippings[0] + n_clippings[1], out->nb_samples * 2);

  751.     }

  752. 

  753.     av_frame_free(&in);

  754.     return ff_filter_frame(outlink, out);

  755. }

  756. 

  757. static int query_formats(AVFilterContext *ctx)

  758. {

  759.     struct SOFAlizerContext *s = ctx->priv;

  760.     AVFilterFormats *formats = NULL;

  761.     AVFilterChannelLayouts *layouts = NULL;

  762.     int ret, sample_rates[] = { 48000, -1 };

  763. 

  764.     ret = ff_add_format(&formats, AV_SAMPLE_FMT_FLT);

  765.     if (ret)

  766.         return ret;

  767.     ret = ff_set_common_formats(ctx, formats);

  768.     if (ret)

  769.         return ret;

  770. 

  771.     layouts = ff_all_channel_layouts();

  772.     if (!layouts)

  773.         return AVERROR(ENOMEM);

  774. 

  775.     ret = ff_channel_layouts_ref(layouts, &ctx->inputs[0]->out_channel_layouts);

  776.     if (ret)

  777.         return ret;

  778. 

  779.     layouts = NULL;

  780.     ret = ff_add_channel_layout(&layouts, AV_CH_LAYOUT_STEREO);

  781.     if (ret)

  782.         return ret;

  783. 

  784.     ret = ff_channel_layouts_ref(layouts, &ctx->outputs[0]->in_channel_layouts);

  785.     if (ret)

  786.         return ret;

  787. 

  788.     sample_rates[0] = s->sample_rate;

  789.     formats = ff_make_format_list(sample_rates);

  790.     if (!formats)

  791.         return AVERROR(ENOMEM);

  792.     return ff_set_common_samplerates(ctx, formats);

  793. }

  794. 

  795. static int load_data(AVFilterContext *ctx, int azim, int elev, float radius)

  796. {

  797.     struct SOFAlizerContext *s = ctx->priv;

  798.     const int n_samples = s->sofa.n_samples;

  799.     int n_conv = s->n_conv; /* no. channels to convolve */

  800.     int n_fft = s->n_fft;

  801.     int delay_l[16]; /* broadband delay for each IR */

  802.     int delay_r[16];

  803.     int nb_input_channels = ctx->inputs[0]->channels; /* no. input channels */

  804.     float gain_lin = expf((s->gain - 3 * nb_input_channels) / 20 * M_LN10); /* gain - 3dB/channel */

  805.     FFTComplex *data_hrtf_l = NULL;

  806.     FFTComplex *data_hrtf_r = NULL;

  807.     FFTComplex *fft_in_l = NULL;

  808.     FFTComplex *fft_in_r = NULL;

  809.     float *data_ir_l = NULL;

  810.     float *data_ir_r = NULL;

  811.     int offset = 0; /* used for faster pointer arithmetics in for-loop */

  812.     int m[16]; /* measurement index m of IR closest to required source positions */

  813.     int i, j, azim_orig = azim, elev_orig = elev;

  814. 

  815.     if (!s->sofa.ncid) { /* if an invalid SOFA file has been selected */

  816.         av_log(ctx, AV_LOG_ERROR, "Selected SOFA file is invalid. Please select valid SOFA file.\n");

  817.         return AVERROR_INVALIDDATA;

  818.     }

  819. 

  820.     if (s->type == TIME_DOMAIN) {

  821.         s->temp_src[0] = av_calloc(FFALIGN(n_samples, 16), sizeof(float));

  822.         s->temp_src[1] = av_calloc(FFALIGN(n_samples, 16), sizeof(float));

  823. 

  824.         /* get temporary IR for L and R channel */

  825.         data_ir_l = av_calloc(n_conv * FFALIGN(n_samples, 16), sizeof(*data_ir_l));

  826.         data_ir_r = av_calloc(n_conv * FFALIGN(n_samples, 16), sizeof(*data_ir_r));

  827.         if (!data_ir_r || !data_ir_l || !s->temp_src[0] || !s->temp_src[1]) {

  828.             av_free(data_ir_l);

  829.             av_free(data_ir_r);

  830.             return AVERROR(ENOMEM);

  831.         }

  832.     } else {

  833.         /* get temporary HRTF memory for L and R channel */

  834.         data_hrtf_l = av_malloc_array(n_fft, sizeof(*data_hrtf_l) * n_conv);

  835.         data_hrtf_r = av_malloc_array(n_fft, sizeof(*data_hrtf_r) * n_conv);

  836.         if (!data_hrtf_r || !data_hrtf_l) {

  837.             av_free(data_hrtf_l);

  838.             av_free(data_hrtf_r);

  839.             return AVERROR(ENOMEM);

  840.         }

  841.     }

  842. 

  843.     for (i = 0; i < s->n_conv; i++) {

  844.         /* load and store IRs and corresponding delays */

  845.         azim = (int)(s->speaker_azim[i] + azim_orig) % 360;

  846.         elev = (int)(s->speaker_elev[i] + elev_orig) % 90;

  847.         /* get id of IR closest to desired position */

  848.         m[i] = find_m(s, azim, elev, radius);

  849. 

  850.         /* load the delays associated with the current IRs */

  851.         delay_l[i] = *(s->sofa.data_delay + 2 * m[i]);

  852.         delay_r[i] = *(s->sofa.data_delay + 2 * m[i] + 1);

  853. 

  854.         if (s->type == TIME_DOMAIN) {

  855.             offset = i * FFALIGN(n_samples, 16); /* no. samples already written */

  856.             for (j = 0; j < n_samples; j++) {

  857.                 /* load reversed IRs of the specified source position

  858.                  * sample-by-sample for left and right ear; and apply gain */

  859.                 *(data_ir_l + offset + j) = /* left channel */

  860.                 *(s->sofa.data_ir + 2 * m[i] * n_samples + n_samples - 1 - j) * gain_lin;

  861.                 *(data_ir_r + offset + j) = /* right channel */

  862.                 *(s->sofa.data_ir + 2 * m[i] * n_samples + n_samples - 1 - j  + n_samples) * gain_lin;

  863.             }

  864.         } else {

  865.             fft_in_l = av_calloc(n_fft, sizeof(*fft_in_l));

  866.             fft_in_r = av_calloc(n_fft, sizeof(*fft_in_r));

  867.             if (!fft_in_l || !fft_in_r) {

  868.                 av_free(data_hrtf_l);

  869.                 av_free(data_hrtf_r);

  870.                 av_free(fft_in_l);

  871.                 av_free(fft_in_r);

  872.                 return AVERROR(ENOMEM);

  873.             }

  874. 

  875.             offset = i * n_fft; /* no. samples already written */

  876.             for (j = 0; j < n_samples; j++) {

  877.                 /* load non-reversed IRs of the specified source position

  878.                  * sample-by-sample and apply gain,

  879.                  * L channel is loaded to real part, R channel to imag part,

  880.                  * IRs ared shifted by L and R delay */

  881.                 fft_in_l[delay_l[i] + j].re = /* left channel */

  882.                 *(s->sofa.data_ir + 2 * m[i] * n_samples + j) * gain_lin;

  883.                 fft_in_r[delay_r[i] + j].re = /* right channel */

  884.                 *(s->sofa.data_ir + (2 * m[i] + 1) * n_samples + j) * gain_lin;

  885.             }

  886. 

  887.             /* actually transform to frequency domain (IRs -> HRTFs) */

  888.             av_fft_permute(s->fft[0], fft_in_l);

  889.             av_fft_calc(s->fft[0], fft_in_l);

  890.             memcpy(data_hrtf_l + offset, fft_in_l, n_fft * sizeof(*fft_in_l));

  891.             av_fft_permute(s->fft[0], fft_in_r);

  892.             av_fft_calc(s->fft[0], fft_in_r);

  893.             memcpy(data_hrtf_r + offset, fft_in_r, n_fft * sizeof(*fft_in_r));

  894.         }

  895. 

  896.         av_log(ctx, AV_LOG_DEBUG, "Index: %d, Azimuth: %f, Elevation: %f, Radius: %f of SOFA file.\n",

  897.                m[i], *(s->sofa.sp_a + m[i]), *(s->sofa.sp_e + m[i]), *(s->sofa.sp_r + m[i]));

  898.     }

  899. 

  900.     if (s->type == TIME_DOMAIN) {

  901.         /* copy IRs and delays to allocated memory in the SOFAlizerContext struct: */

  902.         memcpy(s->data_ir[0], data_ir_l, sizeof(float) * n_conv * FFALIGN(n_samples, 16));

  903.         memcpy(s->data_ir[1], data_ir_r, sizeof(float) * n_conv * FFALIGN(n_samples, 16));

  904. 

  905.         av_freep(&data_ir_l); /* free temporary IR memory */

  906.         av_freep(&data_ir_r);

  907.     } else {

  908.         s->data_hrtf[0] = av_malloc_array(n_fft * s->n_conv, sizeof(FFTComplex));

  909.         s->data_hrtf[1] = av_malloc_array(n_fft * s->n_conv, sizeof(FFTComplex));

  910.         if (!s->data_hrtf[0] || !s->data_hrtf[1]) {

  911.             av_freep(&data_hrtf_l);

  912.             av_freep(&data_hrtf_r);

  913.             av_freep(&fft_in_l);

  914.             av_freep(&fft_in_r);

  915.             return AVERROR(ENOMEM); /* memory allocation failed */

  916.         }

  917. 

  918.         memcpy(s->data_hrtf[0], data_hrtf_l, /* copy HRTF data to */

  919.             sizeof(FFTComplex) * n_conv * n_fft); /* filter struct */

  920.         memcpy(s->data_hrtf[1], data_hrtf_r,

  921.             sizeof(FFTComplex) * n_conv * n_fft);

  922. 

  923.         av_freep(&data_hrtf_l); /* free temporary HRTF memory */

  924.         av_freep(&data_hrtf_r);

  925. 

  926.         av_freep(&fft_in_l); /* free temporary FFT memory */

  927.         av_freep(&fft_in_r);

  928.     }

  929. 

  930.     memcpy(s->delay[0], &delay_l[0], sizeof(int) * s->n_conv);

  931.     memcpy(s->delay[1], &delay_r[0], sizeof(int) * s->n_conv);

  932. 

  933.     return 0;

  934. }

  935. 

  936. static av_cold int init(AVFilterContext *ctx)

  937. {

  938.     SOFAlizerContext *s = ctx->priv;

  939.     int ret;

  940. 

  941.     if (!s->filename) {

  942.         av_log(ctx, AV_LOG_ERROR, "Valid SOFA filename must be set.\n");

  943.         return AVERROR(EINVAL);

  944.     }

  945. 

  946.     /* load SOFA file, */

  947.     /* initialize file IDs to 0 before attempting to load SOFA files,

  948.      * this assures that in case of error, only the memory of already

  949.      * loaded files is free'd */

  950.     s->sofa.ncid = 0;

  951.     ret = load_sofa(ctx, s->filename, &s->sample_rate);

  952.     if (ret) {

  953.         /* file loading error */

  954.         av_log(ctx, AV_LOG_ERROR, "Error while loading SOFA file: '%s'\n", s->filename);

  955.     } else { /* no file loading error, resampling not required */

  956.         av_log(ctx, AV_LOG_DEBUG, "File '%s' loaded.\n", s->filename);

  957.     }

  958. 

  959.     if (ret) {

  960.         av_log(ctx, AV_LOG_ERROR, "No valid SOFA file could be loaded. Please specify valid SOFA file.\n");

  961.         return ret;

  962.     }

  963. 

  964.     s->fdsp = avpriv_float_dsp_alloc(0);

  965.     if (!s->fdsp)

  966.         return AVERROR(ENOMEM);

  967. 

  968.     return 0;

  969. }

  970. 

  971. static int config_input(AVFilterLink *inlink)

  972. {

  973.     AVFilterContext *ctx = inlink->dst;

  974.     SOFAlizerContext *s = ctx->priv;

  975.     int nb_input_channels = inlink->channels; /* no. input channels */

  976.     int n_max_ir = 0;

  977.     int n_current;

  978.     int n_max = 0;

  979.     int ret;

  980. 

  981.     if (s->type == FREQUENCY_DOMAIN) {

  982.         inlink->partial_buf_size =

  983.         inlink->min_samples =

  984.         inlink->max_samples = inlink->sample_rate;

  985.     }

  986. 

  987.     /* gain -3 dB per channel, -6 dB to get LFE on a similar level */

  988.     s->gain_lfe = expf((s->gain - 3 * inlink->channels - 6) / 20 * M_LN10);

  989. 

  990.     s->n_conv = nb_input_channels;

  991. 

  992.     /* get size of ringbuffer (longest IR plus max. delay) */

  993.     /* then choose next power of 2 for performance optimization */

  994.     n_current = s->sofa.n_samples + max_delay(&s->sofa);

  995.     if (n_current > n_max) {

  996.         /* length of longest IR plus max. delay (in all SOFA files) */

  997.         n_max = n_current;

  998.         /* length of longest IR (without delay, in all SOFA files) */

  999.         n_max_ir = s->sofa.n_samples;

  1000.     }

  1001.     /* buffer length is longest IR plus max. delay -> next power of 2

  1002.        (32 - count leading zeros gives required exponent)  */

  1003.     s->buffer_length = 1 << (32 - ff_clz(n_max));

  1004.     s->n_fft         = 1 << (32 - ff_clz(n_max + inlink->sample_rate));

  1005. 

  1006.     if (s->type == FREQUENCY_DOMAIN) {

  1007.         av_fft_end(s->fft[0]);

  1008.         av_fft_end(s->fft[1]);

  1009.         s->fft[0] = av_fft_init(log2(s->n_fft), 0);

  1010.         s->fft[1] = av_fft_init(log2(s->n_fft), 0);

  1011.         av_fft_end(s->ifft[0]);

  1012.         av_fft_end(s->ifft[1]);

  1013.         s->ifft[0] = av_fft_init(log2(s->n_fft), 1);

  1014.         s->ifft[1] = av_fft_init(log2(s->n_fft), 1);

  1015. 

  1016.         if (!s->fft[0] || !s->fft[1] || !s->ifft[0] || !s->ifft[1]) {

  1017.             av_log(ctx, AV_LOG_ERROR, "Unable to create FFT contexts of size %d.\n", s->n_fft);

  1018.             return AVERROR(ENOMEM);

  1019.         }

  1020.     }

  1021. 

  1022.     /* Allocate memory for the impulse responses, delays and the ringbuffers */

  1023.     /* size: (longest IR) * (number of channels to convolute) */

  1024.     s->data_ir[0] = av_calloc(FFALIGN(n_max_ir, 16), sizeof(float) * s->n_conv);

  1025.     s->data_ir[1] = av_calloc(FFALIGN(n_max_ir, 16), sizeof(float) * s->n_conv);

  1026.     /* length:  number of channels to convolute */

  1027.     s->delay[0] = av_malloc_array(s->n_conv, sizeof(float));

  1028.     s->delay[1] = av_malloc_array(s->n_conv, sizeof(float));

  1029.     /* length: (buffer length) * (number of input channels),

  1030.      * OR: buffer length (if frequency domain processing)

  1031.      * calloc zero-initializes the buffer */

  1032. 

  1033.     if (s->type == TIME_DOMAIN) {

  1034.         s->ringbuffer[0] = av_calloc(s->buffer_length, sizeof(float) * nb_input_channels);

  1035.         s->ringbuffer[1] = av_calloc(s->buffer_length, sizeof(float) * nb_input_channels);

  1036.     } else {

  1037.         s->ringbuffer[0] = av_calloc(s->buffer_length, sizeof(float));

  1038.         s->ringbuffer[1] = av_calloc(s->buffer_length, sizeof(float));

  1039.         s->temp_fft[0] = av_malloc_array(s->n_fft, sizeof(FFTComplex));

  1040.         s->temp_fft[1] = av_malloc_array(s->n_fft, sizeof(FFTComplex));

  1041.         if (!s->temp_fft[0] || !s->temp_fft[1])

  1042.             return AVERROR(ENOMEM);

  1043.     }

  1044. 

  1045.     /* length: number of channels to convolute */

  1046.     s->speaker_azim = av_calloc(s->n_conv, sizeof(*s->speaker_azim));

  1047.     s->speaker_elev = av_calloc(s->n_conv, sizeof(*s->speaker_elev));

  1048. 

  1049.     /* memory allocation failed: */

  1050.     if (!s->data_ir[0] || !s->data_ir[1] || !s->delay[1] ||

  1051.         !s->delay[0] || !s->ringbuffer[0] || !s->ringbuffer[1] ||

  1052.         !s->speaker_azim || !s->speaker_elev)

  1053.         return AVERROR(ENOMEM);

  1054. 

  1055.     compensate_volume(ctx);

  1056. 

  1057.     /* get speaker positions */

  1058.     if ((ret = get_speaker_pos(ctx, s->speaker_azim, s->speaker_elev)) < 0) {

  1059.         av_log(ctx, AV_LOG_ERROR, "Couldn't get speaker positions. Input channel configuration not supported.\n");

  1060.         return ret;

  1061.     }

  1062. 

  1063.     /* load IRs to data_ir[0] and data_ir[1] for required directions */

  1064.     if ((ret = load_data(ctx, s->rotation, s->elevation, s->radius)) < 0)

  1065.         return ret;

  1066. 

  1067.     av_log(ctx, AV_LOG_DEBUG, "Samplerate: %d Channels to convolute: %d, Length of ringbuffer: %d x %d\n",

  1068.         inlink->sample_rate, s->n_conv, nb_input_channels, s->buffer_length);

  1069. 

  1070.     return 0;

  1071. }

  1072. 

  1073. static av_cold void uninit(AVFilterContext *ctx)

  1074. {

  1075.     SOFAlizerContext *s = ctx->priv;

  1076. 

  1077.     if (s->sofa.ncid) {

  1078.         av_freep(&s->sofa.sp_a);

  1079.         av_freep(&s->sofa.sp_e);

  1080.         av_freep(&s->sofa.sp_r);

  1081.         av_freep(&s->sofa.data_delay);

  1082.         av_freep(&s->sofa.data_ir);

  1083.     }

  1084.     av_fft_end(s->ifft[0]);

  1085.     av_fft_end(s->ifft[1]);

  1086.     av_fft_end(s->fft[0]);

  1087.     av_fft_end(s->fft[1]);

  1088.     av_freep(&s->delay[0]);

  1089.     av_freep(&s->delay[1]);

  1090.     av_freep(&s->data_ir[0]);

  1091.     av_freep(&s->data_ir[1]);

  1092.     av_freep(&s->ringbuffer[0]);

  1093.     av_freep(&s->ringbuffer[1]);

  1094.     av_freep(&s->speaker_azim);

  1095.     av_freep(&s->speaker_elev);

  1096.     av_freep(&s->temp_src[0]);

  1097.     av_freep(&s->temp_src[1]);

  1098.     av_freep(&s->temp_fft[0]);

  1099.     av_freep(&s->temp_fft[1]);

  1100.     av_freep(&s->data_hrtf[0]);

  1101.     av_freep(&s->data_hrtf[1]);

  1102.     av_freep(&s->fdsp);

  1103. }

  1104. 

  1105. #define OFFSET(x) offsetof(SOFAlizerContext, x)

  1106. #define FLAGS AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM

  1107. 

  1108. static const AVOption sofalizer_options[] = {

  1109.     { "sofa",      "sofa filename",  OFFSET(filename),  AV_OPT_TYPE_STRING, {.str=NULL},            .flags = FLAGS },

  1110.     { "gain",      "set gain in dB", OFFSET(gain),      AV_OPT_TYPE_FLOAT,  {.dbl=0},     -20,  40, .flags = FLAGS },

  1111.     { "rotation",  "set rotation"  , OFFSET(rotation),  AV_OPT_TYPE_FLOAT,  {.dbl=0},    -360, 360, .flags = FLAGS },

  1112.     { "elevation", "set elevation",  OFFSET(elevation), AV_OPT_TYPE_FLOAT,  {.dbl=0},     -90,  90, .flags = FLAGS },

  1113.     { "radius",    "set radius",     OFFSET(radius),    AV_OPT_TYPE_FLOAT,  {.dbl=1},       0,   3, .flags = FLAGS },

  1114.     { "type",      "set processing", OFFSET(type),      AV_OPT_TYPE_INT,    {.i64=1},       0,   1, .flags = FLAGS, "type" },

  1115.     { "time",      "time domain",      0,               AV_OPT_TYPE_CONST,  {.i64=0},       0,   0, .flags = FLAGS, "type" },

  1116.     { "freq",      "frequency domain", 0,               AV_OPT_TYPE_CONST,  {.i64=1},       0,   0, .flags = FLAGS, "type" },

  1117.     { NULL }

  1118. };

  1119. 

  1120. AVFILTER_DEFINE_CLASS(sofalizer);

  1121. 

  1122. static const AVFilterPad inputs[] = {

  1123.     {

  1124.         .name         = "default",

  1125.         .type         = AVMEDIA_TYPE_AUDIO,

  1126.         .config_props = config_input,

  1127.         .filter_frame = filter_frame,

  1128.     },

  1129.     { NULL }

  1130. };

  1131. 

  1132. static const AVFilterPad outputs[] = {

  1133.     {

  1134.         .name = "default",

  1135.         .type = AVMEDIA_TYPE_AUDIO,

  1136.     },

  1137.     { NULL }

  1138. };

  1139. 

  1140. AVFilter ff_af_sofalizer = {

  1141.     .name          = "sofalizer",

  1142.     .description   = NULL_IF_CONFIG_SMALL("SOFAlizer (Spatially Oriented Format for Acoustics)."),

  1143.     .priv_size     = sizeof(SOFAlizerContext),

  1144.     .priv_class    = &sofalizer_class,

  1145.     .init          = init,

  1146.     .uninit        = uninit,

  1147.     .query_formats = query_formats,

  1148.     .inputs        = inputs,

  1149.     .outputs       = outputs,

  1150.     .flags         = AVFILTER_FLAG_SLICE_THREADS,

  1151. };