bridge_softmix.c

Go to the documentation of this file.

/*
 * Asterisk -- An open source telephony toolkit.
 *
 * Copyright (C) 2011, Digium, Inc.
 *
 * Joshua Colp <jcolp@digium.com>
 * David Vossel <dvossel@digium.com>
 *
 * See http://www.asterisk.org for more information about
 * the Asterisk project. Please do not directly contact
 * any of the maintainers of this project for assistance;
 * the project provides a web site, mailing lists and IRC
 * channels for your use.
 *
 * This program is free software, distributed under the terms of
 * the GNU General Public License Version 2. See the LICENSE file
 * at the top of the source tree.
 */

/*! \file
 *
 * \brief Multi-party software based channel mixing
 *
 * \author Joshua Colp <jcolp@digium.com>
 * \author David Vossel <dvossel@digium.com>
 *
 * \ingroup bridges
 */

/*** MODULEINFO
   <support_level>core</support_level>
 ***/

#include "asterisk.h"

ASTERISK_FILE_VERSION(__FILE__, "$Revision: 356573 $")

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/time.h>
#include <signal.h>
#include <errno.h>
#include <unistd.h>

#include "asterisk/module.h"
#include "asterisk/channel.h"
#include "asterisk/bridging.h"
#include "asterisk/bridging_technology.h"
#include "asterisk/frame.h"
#include "asterisk/options.h"
#include "asterisk/logger.h"
#include "asterisk/slinfactory.h"
#include "asterisk/astobj2.h"
#include "asterisk/timing.h"
#include "asterisk/translate.h"

#define MAX_DATALEN 8096

/*! \brief Interval at which mixing will take place. Valid options are 10, 20, and 40. */
#define DEFAULT_SOFTMIX_INTERVAL 20

/*! \brief Size of the buffer used for sample manipulation */
#define SOFTMIX_DATALEN(rate, interval) ((rate/50) * (interval / 10))

/*! \brief Number of samples we are dealing with */
#define SOFTMIX_SAMPLES(rate, interval) (SOFTMIX_DATALEN(rate, interval) / 2)

/*! \brief Number of mixing iterations to perform between gathering statistics. */
#define SOFTMIX_STAT_INTERVAL 100

/* This is the threshold in ms at which a channel's own audio will stop getting
 * mixed out its own write audio stream because it is not talking. */
#define DEFAULT_SOFTMIX_SILENCE_THRESHOLD 2500
#define DEFAULT_SOFTMIX_TALKING_THRESHOLD 160

#define DEFAULT_ENERGY_HISTORY_LEN 150

struct video_follow_talker_data {
   /*! audio energy history */
   int energy_history[DEFAULT_ENERGY_HISTORY_LEN];
   /*! The current slot being used in the history buffer, this
    *  increments and wraps around */
   int energy_history_cur_slot;
   /*! The current energy sum used for averages. */
   int energy_accum;
   /*! The current energy average */
   int energy_average;
};

/*! \brief Structure which contains per-channel mixing information */
struct softmix_channel {
   /*! Lock to protect this structure */
   ast_mutex_t lock;
   /*! Factory which contains audio read in from the channel */
   struct ast_slinfactory factory;
   /*! Frame that contains mixed audio to be written out to the channel */
   struct ast_frame write_frame;
   /*! Frame that contains mixed audio read from the channel */
   struct ast_frame read_frame;
   /*! DSP for detecting silence */
   struct ast_dsp *dsp;
   /*! Bit used to indicate if a channel is talking or not. This affects how
    * the channel's audio is mixed back to it. */
   int talking:1;
   /*! Bit used to indicate that the channel provided audio for this mixing interval */
   int have_audio:1;
   /*! Bit used to indicate that a frame is available to be written out to the channel */
   int have_frame:1;
   /*! Buffer containing final mixed audio from all sources */
   short final_buf[MAX_DATALEN];
   /*! Buffer containing only the audio from the channel */
   short our_buf[MAX_DATALEN];
   /*! Data pertaining to talker mode for video conferencing */
   struct video_follow_talker_data video_talker;
};

struct softmix_bridge_data {
   struct ast_timer *timer;
   unsigned int internal_rate;
   unsigned int internal_mixing_interval;
};

struct softmix_stats {
      /*! Each index represents a sample rate used above the internal rate. */
      unsigned int sample_rates[16];
      /*! Each index represents the number of channels using the same index in the sample_rates array.  */
      unsigned int num_channels[16];
      /*! the number of channels above the internal sample rate */
      unsigned int num_above_internal_rate;
      /*! the number of channels at the internal sample rate */
      unsigned int num_at_internal_rate;
      /*! the absolute highest sample rate supported by any channel in the bridge */
      unsigned int highest_supported_rate;
      /*! Is the sample rate locked by the bridge, if so what is that rate.*/
      unsigned int locked_rate;
};

struct softmix_mixing_array {
   int max_num_entries;
   int used_entries;
   int16_t **buffers;
};

struct softmix_translate_helper_entry {
   int num_times_requested; /*!< Once this entry is no longer requested, free the trans_pvt
                                 and re-init if it was usable. */
   struct ast_format dst_format; /*!< The destination format for this helper */
   struct ast_trans_pvt *trans_pvt; /*!< the translator for this slot. */
   struct ast_frame *out_frame; /*!< The output frame from the last translation */
   AST_LIST_ENTRY(softmix_translate_helper_entry) entry;
};

struct softmix_translate_helper {
   struct ast_format slin_src; /*!< the source format expected for all the translators */
   AST_LIST_HEAD_NOLOCK(, softmix_translate_helper_entry) entries;
};

static struct softmix_translate_helper_entry *softmix_translate_helper_entry_alloc(struct ast_format *dst)
{
   struct softmix_translate_helper_entry *entry;
   if (!(entry = ast_calloc(1, sizeof(*entry)))) {
      return NULL;
   }
   ast_format_copy(&entry->dst_format, dst);
   return entry;
}

static void *softmix_translate_helper_free_entry(struct softmix_translate_helper_entry *entry)
{
   if (entry->trans_pvt) {
      ast_translator_free_path(entry->trans_pvt);
   }
   if (entry->out_frame) {
      ast_frfree(entry->out_frame);
   }
   ast_free(entry);
   return NULL;
}

static void softmix_translate_helper_init(struct softmix_translate_helper *trans_helper, unsigned int sample_rate)
{
   memset(trans_helper, 0, sizeof(*trans_helper));
   ast_format_set(&trans_helper->slin_src, ast_format_slin_by_rate(sample_rate), 0);
}

static void softmix_translate_helper_destroy(struct softmix_translate_helper *trans_helper)
{
   struct softmix_translate_helper_entry *entry;

   while ((entry = AST_LIST_REMOVE_HEAD(&trans_helper->entries, entry))) {
      softmix_translate_helper_free_entry(entry);
   }
}

static void softmix_translate_helper_change_rate(struct softmix_translate_helper *trans_helper, unsigned int sample_rate)
{
   struct softmix_translate_helper_entry *entry;

   ast_format_set(&trans_helper->slin_src, ast_format_slin_by_rate(sample_rate), 0);
   AST_LIST_TRAVERSE_SAFE_BEGIN(&trans_helper->entries, entry, entry) {
      if (entry->trans_pvt) {
         ast_translator_free_path(entry->trans_pvt);
         if (!(entry->trans_pvt = ast_translator_build_path(&entry->dst_format, &trans_helper->slin_src))) {
            AST_LIST_REMOVE_CURRENT(entry);
            entry = softmix_translate_helper_free_entry(entry);
         }
      }
   }
   AST_LIST_TRAVERSE_SAFE_END;
}

/*!
 * \internal
 * \brief Get the next available audio on the softmix channel's read stream
 * and determine if it should be mixed out or not on the write stream. 
 *
 * \retval pointer to buffer containing the exact number of samples requested on success.
 * \retval NULL if no samples are present
 */
static int16_t *softmix_process_read_audio(struct softmix_channel *sc, unsigned int num_samples)
{
   if ((ast_slinfactory_available(&sc->factory) >= num_samples) &&
      ast_slinfactory_read(&sc->factory, sc->our_buf, num_samples)) {
      sc->have_audio = 1;
      return sc->our_buf;
   }
   sc->have_audio = 0;
   return NULL;
}

/*!
 * \internal
 * \brief Process a softmix channel's write audio
 *
 * \details This function will remove the channel's talking from its own audio if present and
 * possibly even do the channel's write translation for it depending on how many other
 * channels use the same write format.
 */
static void softmix_process_write_audio(struct softmix_translate_helper *trans_helper,
   struct ast_format *raw_write_fmt,
   struct softmix_channel *sc)
{
   struct softmix_translate_helper_entry *entry = NULL;
   int i;

   /* If we provided audio that was not determined to be silence,
    * then take it out while in slinear format. */
   if (sc->have_audio && sc->talking) {
      for (i = 0; i < sc->write_frame.samples; i++) {
         ast_slinear_saturated_subtract(&sc->final_buf[i], &sc->our_buf[i]);
      }
      /* do not do any special write translate optimization if we had to make
       * a special mix for them to remove their own audio. */
      return;
   }

   AST_LIST_TRAVERSE(&trans_helper->entries, entry, entry) {
      if (ast_format_cmp(&entry->dst_format, raw_write_fmt) == AST_FORMAT_CMP_EQUAL) {
         entry->num_times_requested++;
      } else {
         continue;
      }
      if (!entry->trans_pvt && (entry->num_times_requested > 1)) {
         entry->trans_pvt = ast_translator_build_path(&entry->dst_format, &trans_helper->slin_src);
      }
      if (entry->trans_pvt && !entry->out_frame) {
         entry->out_frame = ast_translate(entry->trans_pvt, &sc->write_frame, 0);
      }
      if (entry->out_frame && (entry->out_frame->datalen < MAX_DATALEN)) {
         ast_format_copy(&sc->write_frame.subclass.format, &entry->out_frame->subclass.format);
         memcpy(sc->final_buf, entry->out_frame->data.ptr, entry->out_frame->datalen);
         sc->write_frame.datalen = entry->out_frame->datalen;
         sc->write_frame.samples = entry->out_frame->samples;
      }
      break;
   }

   /* add new entry into list if this format destination was not matched. */
   if (!entry && (entry = softmix_translate_helper_entry_alloc(raw_write_fmt))) {
      AST_LIST_INSERT_HEAD(&trans_helper->entries, entry, entry);
   }
}

static void softmix_translate_helper_cleanup(struct softmix_translate_helper *trans_helper)
{
   struct softmix_translate_helper_entry *entry = NULL;
   AST_LIST_TRAVERSE(&trans_helper->entries, entry, entry) {
      if (entry->out_frame) {
         ast_frfree(entry->out_frame);
         entry->out_frame = NULL;
      }
      entry->num_times_requested = 0;
   }
}

static void softmix_bridge_data_destroy(void *obj)
{
   struct softmix_bridge_data *softmix_data = obj;
   ast_timer_close(softmix_data->timer);
}

/*! \brief Function called when a bridge is created */
static int softmix_bridge_create(struct ast_bridge *bridge)
{
   struct softmix_bridge_data *softmix_data;

   if (!(softmix_data = ao2_alloc(sizeof(*softmix_data), softmix_bridge_data_destroy))) {
      return -1;
   }
   if (!(softmix_data->timer = ast_timer_open())) {
      ao2_ref(softmix_data, -1);
      return -1;
   }

   /* start at 8khz, let it grow from there */
   softmix_data->internal_rate = 8000;
   softmix_data->internal_mixing_interval = DEFAULT_SOFTMIX_INTERVAL;

   bridge->bridge_pvt = softmix_data;
   return 0;
}

/*! \brief Function called when a bridge is destroyed */
static int softmix_bridge_destroy(struct ast_bridge *bridge)
{
   struct softmix_bridge_data *softmix_data = bridge->bridge_pvt;
   if (!bridge->bridge_pvt) {
      return -1;
   }
   ao2_ref(softmix_data, -1);
   bridge->bridge_pvt = NULL;
   return 0;
}

static void set_softmix_bridge_data(int rate, int interval, struct ast_bridge_channel *bridge_channel, int reset)
{
   struct softmix_channel *sc = bridge_channel->bridge_pvt;
   unsigned int channel_read_rate = ast_format_rate(ast_channel_rawreadformat(bridge_channel->chan));

   ast_mutex_lock(&sc->lock);
   if (reset) {
      ast_slinfactory_destroy(&sc->factory);
      ast_dsp_free(sc->dsp);
   }
   /* Setup read/write frame parameters */
   sc->write_frame.frametype = AST_FRAME_VOICE;
   ast_format_set(&sc->write_frame.subclass.format, ast_format_slin_by_rate(rate), 0);
   sc->write_frame.data.ptr = sc->final_buf;
   sc->write_frame.datalen = SOFTMIX_DATALEN(rate, interval);
   sc->write_frame.samples = SOFTMIX_SAMPLES(rate, interval);

   sc->read_frame.frametype = AST_FRAME_VOICE;
   ast_format_set(&sc->read_frame.subclass.format, ast_format_slin_by_rate(channel_read_rate), 0);
   sc->read_frame.data.ptr = sc->our_buf;
   sc->read_frame.datalen = SOFTMIX_DATALEN(channel_read_rate, interval);
   sc->read_frame.samples = SOFTMIX_SAMPLES(channel_read_rate, interval);

   /* Setup smoother */
   ast_slinfactory_init_with_format(&sc->factory, &sc->write_frame.subclass.format);

   /* set new read and write formats on channel. */
   ast_set_read_format(bridge_channel->chan, &sc->read_frame.subclass.format);
   ast_set_write_format(bridge_channel->chan, &sc->write_frame.subclass.format);

   /* set up new DSP.  This is on the read side only right before the read frame enters the smoother.  */
   sc->dsp = ast_dsp_new_with_rate(channel_read_rate);
   /* we want to aggressively detect silence to avoid feedback */
   if (bridge_channel->tech_args.talking_threshold) {
      ast_dsp_set_threshold(sc->dsp, bridge_channel->tech_args.talking_threshold);
   } else {
      ast_dsp_set_threshold(sc->dsp, DEFAULT_SOFTMIX_TALKING_THRESHOLD);
   }

   ast_mutex_unlock(&sc->lock);
}

/*! \brief Function called when a channel is joined into the bridge */
static int softmix_bridge_join(struct ast_bridge *bridge, struct ast_bridge_channel *bridge_channel)
{
   struct softmix_channel *sc = NULL;
   struct softmix_bridge_data *softmix_data = bridge->bridge_pvt;

   /* Create a new softmix_channel structure and allocate various things on it */
   if (!(sc = ast_calloc(1, sizeof(*sc)))) {
      return -1;
   }

   /* Can't forget the lock */
   ast_mutex_init(&sc->lock);

   /* Can't forget to record our pvt structure within the bridged channel structure */
   bridge_channel->bridge_pvt = sc;

   set_softmix_bridge_data(softmix_data->internal_rate,
      softmix_data->internal_mixing_interval ? softmix_data->internal_mixing_interval : DEFAULT_SOFTMIX_INTERVAL,
      bridge_channel, 0);

   return 0;
}

/*! \brief Function called when a channel leaves the bridge */
static int softmix_bridge_leave(struct ast_bridge *bridge, struct ast_bridge_channel *bridge_channel)
{
   struct softmix_channel *sc = bridge_channel->bridge_pvt;

   if (!(bridge_channel->bridge_pvt)) {
      return 0;
   }
   bridge_channel->bridge_pvt = NULL;

   /* Drop mutex lock */
   ast_mutex_destroy(&sc->lock);

   /* Drop the factory */
   ast_slinfactory_destroy(&sc->factory);

   /* Drop the DSP */
   ast_dsp_free(sc->dsp);

   /* Eep! drop ourselves */
   ast_free(sc);

   return 0;
}

/*!
 * \internal
 * \brief If the bridging core passes DTMF to us, then they want it to be distributed out to all memebers. Do that here.
 */
static void softmix_pass_dtmf(struct ast_bridge *bridge, struct ast_bridge_channel *bridge_channel, struct ast_frame *frame)
{
   struct ast_bridge_channel *tmp;
   AST_LIST_TRAVERSE(&bridge->channels, tmp, entry) {
      if (tmp == bridge_channel) {
         continue;
      }
      ast_write(tmp->chan, frame);
   }
}

static void softmix_pass_video_top_priority(struct ast_bridge *bridge, struct ast_frame *frame)
{
   struct ast_bridge_channel *tmp;
   AST_LIST_TRAVERSE(&bridge->channels, tmp, entry) {
      if (tmp->suspended) {
         continue;
      }
      if (ast_bridge_is_video_src(bridge, tmp->chan) == 1) {
         ast_write(tmp->chan, frame);
         break;
      }
   }
}

static void softmix_pass_video_all(struct ast_bridge *bridge, struct ast_bridge_channel *bridge_channel, struct ast_frame *frame, int echo)
{
   struct ast_bridge_channel *tmp;
   AST_LIST_TRAVERSE(&bridge->channels, tmp, entry) {
      if (tmp->suspended) {
         continue;
      }
      if ((tmp->chan == bridge_channel->chan) && !echo) {
         continue;
      }
      ast_write(tmp->chan, frame);
   }
}

/*! \brief Function called when a channel writes a frame into the bridge */
static enum ast_bridge_write_result softmix_bridge_write(struct ast_bridge *bridge, struct ast_bridge_channel *bridge_channel, struct ast_frame *frame)
{
   struct softmix_channel *sc = bridge_channel->bridge_pvt;
   struct softmix_bridge_data *softmix_data = bridge->bridge_pvt;
   int totalsilence = 0;
   int cur_energy = 0;
   int silence_threshold = bridge_channel->tech_args.silence_threshold ?
      bridge_channel->tech_args.silence_threshold :
      DEFAULT_SOFTMIX_SILENCE_THRESHOLD;
   char update_talking = -1;  /* if this is set to 0 or 1, tell the bridge that the channel has started or stopped talking. */
   int res = AST_BRIDGE_WRITE_SUCCESS;

   /* Only accept audio frames, all others are unsupported */
   if (frame->frametype == AST_FRAME_DTMF_END || frame->frametype == AST_FRAME_DTMF_BEGIN) {
      softmix_pass_dtmf(bridge, bridge_channel, frame);
      goto bridge_write_cleanup;
   } else if (frame->frametype != AST_FRAME_VOICE && frame->frametype != AST_FRAME_VIDEO) {
      res = AST_BRIDGE_WRITE_UNSUPPORTED;
      goto bridge_write_cleanup;
   } else if (frame->datalen == 0) {
      goto bridge_write_cleanup;
   }

   /* Determine if this video frame should be distributed or not */
   if (frame->frametype == AST_FRAME_VIDEO) {
      int num_src = ast_bridge_number_video_src(bridge);
      int video_src_priority = ast_bridge_is_video_src(bridge, bridge_channel->chan);

      switch (bridge->video_mode.mode) {
      case AST_BRIDGE_VIDEO_MODE_NONE:
         break;
      case AST_BRIDGE_VIDEO_MODE_SINGLE_SRC:
         if (video_src_priority == 1) {
            softmix_pass_video_all(bridge, bridge_channel, frame, 1);
         }
         break;
      case AST_BRIDGE_VIDEO_MODE_TALKER_SRC:
         ast_mutex_lock(&sc->lock);
         ast_bridge_update_talker_src_video_mode(bridge, bridge_channel->chan, sc->video_talker.energy_average, ast_format_get_video_mark(&frame->subclass.format));
         ast_mutex_unlock(&sc->lock);
         if (video_src_priority == 1) {
            int echo = num_src > 1 ? 0 : 1;
            softmix_pass_video_all(bridge, bridge_channel, frame, echo);
         } else if (video_src_priority == 2) {
            softmix_pass_video_top_priority(bridge, frame);
         }
         break;
      }
      goto bridge_write_cleanup;
   }

   /* If we made it here, we are going to write the frame into the conference */
   ast_mutex_lock(&sc->lock);
   ast_dsp_silence_with_energy(sc->dsp, frame, &totalsilence, &cur_energy);

   if (bridge->video_mode.mode == AST_BRIDGE_VIDEO_MODE_TALKER_SRC) {
      int cur_slot = sc->video_talker.energy_history_cur_slot;
      sc->video_talker.energy_accum -= sc->video_talker.energy_history[cur_slot];
      sc->video_talker.energy_accum += cur_energy;
      sc->video_talker.energy_history[cur_slot] = cur_energy;
      sc->video_talker.energy_average = sc->video_talker.energy_accum / DEFAULT_ENERGY_HISTORY_LEN;
      sc->video_talker.energy_history_cur_slot++;
      if (sc->video_talker.energy_history_cur_slot == DEFAULT_ENERGY_HISTORY_LEN) {
         sc->video_talker.energy_history_cur_slot = 0; /* wrap around */
      }
   }

   if (totalsilence < silence_threshold) {
      if (!sc->talking) {
         update_talking = 1;
      }
      sc->talking = 1; /* tell the write process we have audio to be mixed out */
   } else {
      if (sc->talking) {
         update_talking = 0;
      }
      sc->talking = 0;
   }

   /* Before adding audio in, make sure we haven't fallen behind. If audio has fallen
    * behind 4 times the amount of samples mixed on every iteration of the mixer, Re-sync
    * the audio by flushing the buffer before adding new audio in. */
   if (ast_slinfactory_available(&sc->factory) > (4 * SOFTMIX_SAMPLES(softmix_data->internal_rate, softmix_data->internal_mixing_interval))) {
      ast_slinfactory_flush(&sc->factory);
   }

   /* If a frame was provided add it to the smoother, unless drop silence is enabled and this frame
    * is not determined to be talking. */
   if (!(bridge_channel->tech_args.drop_silence && !sc->talking) &&
      (frame->frametype == AST_FRAME_VOICE && ast_format_is_slinear(&frame->subclass.format))) {
      ast_slinfactory_feed(&sc->factory, frame);
   }

   /* If a frame is ready to be written out, do so */
   if (sc->have_frame) {
      ast_write(bridge_channel->chan, &sc->write_frame);
      sc->have_frame = 0;
   }

   /* Alllll done */
   ast_mutex_unlock(&sc->lock);

   if (update_talking != -1) {
      ast_bridge_notify_talking(bridge, bridge_channel, update_talking);
   }

   return res;

bridge_write_cleanup:
   /* Even though the frame is not being written into the conference because it is not audio,
    * we should use this opportunity to check to see if a frame is ready to be written out from
    * the conference to the channel. */
   ast_mutex_lock(&sc->lock);
   if (sc->have_frame) {
      ast_write(bridge_channel->chan, &sc->write_frame);
      sc->have_frame = 0;
   }
   ast_mutex_unlock(&sc->lock);

   return res;
}

/*! \brief Function called when the channel's thread is poked */
static int softmix_bridge_poke(struct ast_bridge *bridge, struct ast_bridge_channel *bridge_channel)
{
   struct softmix_channel *sc = bridge_channel->bridge_pvt;

   ast_mutex_lock(&sc->lock);

   if (sc->have_frame) {
      ast_write(bridge_channel->chan, &sc->write_frame);
      sc->have_frame = 0;
   }

   ast_mutex_unlock(&sc->lock);

   return 0;
}

static void gather_softmix_stats(struct softmix_stats *stats,
   const struct softmix_bridge_data *softmix_data,
   struct ast_bridge_channel *bridge_channel)
{
   int channel_native_rate;
   int i;
   /* Gather stats about channel sample rates. */
   channel_native_rate = MAX(ast_format_rate(ast_channel_rawwriteformat(bridge_channel->chan)),
      ast_format_rate(ast_channel_rawreadformat(bridge_channel->chan)));

   if (channel_native_rate > stats->highest_supported_rate) {
      stats->highest_supported_rate = channel_native_rate;
   }
   if (channel_native_rate > softmix_data->internal_rate) {
      for (i = 0; i < ARRAY_LEN(stats->sample_rates); i++) {
         if (stats->sample_rates[i] == channel_native_rate) {
            stats->num_channels[i]++;
            break;
         } else if (!stats->sample_rates[i]) {
            stats->sample_rates[i] = channel_native_rate;
            stats->num_channels[i]++;
            break;
         }
      }
      stats->num_above_internal_rate++;
   } else if (channel_native_rate == softmix_data->internal_rate) {
      stats->num_at_internal_rate++;
   }
}
/*!
 * \internal
 * \brief Analyse mixing statistics and change bridges internal rate
 * if necessary.
 *
 * \retval 0, no changes to internal rate 
 * \ratval 1, internal rate was changed, update all the channels on the next mixing iteration.
 */
static unsigned int analyse_softmix_stats(struct softmix_stats *stats, struct softmix_bridge_data *softmix_data)
{
   int i;
   /* Re-adjust the internal bridge sample rate if
    * 1. The bridge's internal sample rate is locked in at a sample
    *    rate other than the current sample rate being used.
    * 2. two or more channels support a higher sample rate
    * 3. no channels support the current sample rate or a higher rate
    */
   if (stats->locked_rate) {
      /* if the rate is locked by the bridge, only update it if it differs
       * from the current rate we are using. */
      if (softmix_data->internal_rate != stats->locked_rate) {
         softmix_data->internal_rate = stats->locked_rate;
         ast_debug(1, " Bridge is locked in at sample rate %d\n", softmix_data->internal_rate);
         return 1;
      }
   } else if (stats->num_above_internal_rate >= 2) {
      /* the highest rate is just used as a starting point */
      unsigned int best_rate = stats->highest_supported_rate;
      int best_index = -1;

      for (i = 0; i < ARRAY_LEN(stats->num_channels); i++) {
         if (stats->num_channels[i]) {
            break;
         }
         /* best_rate starts out being the first sample rate
          * greater than the internal sample rate that 2 or
          * more channels support. */
         if (stats->num_channels[i] >= 2 && (best_index == -1)) {
            best_rate = stats->sample_rates[i];
            best_index = i;
         /* If it has been detected that multiple rates above
          * the internal rate are present, compare those rates
          * to each other and pick the highest one two or more
          * channels support. */
         } else if (((best_index != -1) &&
            (stats->num_channels[i] >= 2) &&
            (stats->sample_rates[best_index] < stats->sample_rates[i]))) {
            best_rate = stats->sample_rates[i];
            best_index = i;
         /* It is possible that multiple channels exist with native sample
          * rates above the internal sample rate, but none of those channels
          * have the same rate in common.  In this case, the lowest sample
          * rate among those channels is picked. Over time as additional
          * statistic runs are made the internal sample rate number will
          * adjust to the most optimal sample rate, but it may take multiple
          * iterations. */
         } else if (best_index == -1) {
            best_rate = MIN(best_rate, stats->sample_rates[i]);
         }
      }

      ast_debug(1, " Bridge changed from %d To %d\n", softmix_data->internal_rate, best_rate);
      softmix_data->internal_rate = best_rate;
      return 1;
   } else if (!stats->num_at_internal_rate && !stats->num_above_internal_rate) {
      /* In this case, the highest supported rate is actually lower than the internal rate */
      softmix_data->internal_rate = stats->highest_supported_rate;
      ast_debug(1, " Bridge changed from %d to %d\n", softmix_data->internal_rate, stats->highest_supported_rate);
      return 1;
   }
   return 0;
}

static int softmix_mixing_array_init(struct softmix_mixing_array *mixing_array, unsigned int starting_num_entries)
{
   memset(mixing_array, 0, sizeof(*mixing_array));
   mixing_array->max_num_entries = starting_num_entries;
   if (!(mixing_array->buffers = ast_calloc(mixing_array->max_num_entries, sizeof(int16_t *)))) {
      ast_log(LOG_NOTICE, "Failed to allocate softmix mixing structure. \n");
      return -1;
   }
   return 0;
}

static void softmix_mixing_array_destroy(struct softmix_mixing_array *mixing_array)
{
   ast_free(mixing_array->buffers);
}

static int softmix_mixing_array_grow(struct softmix_mixing_array *mixing_array, unsigned int num_entries)
{
   int16_t **tmp;
   /* give it some room to grow since memory is cheap but allocations can be expensive */
   mixing_array->max_num_entries = num_entries;
   if (!(tmp = ast_realloc(mixing_array->buffers, (mixing_array->max_num_entries * sizeof(int16_t *))))) {
      ast_log(LOG_NOTICE, "Failed to re-allocate softmix mixing structure. \n");
      return -1;
   }
   mixing_array->buffers = tmp;
   return 0;
}

/*! \brief Function which acts as the mixing thread */
static int softmix_bridge_thread(struct ast_bridge *bridge)
{
   struct softmix_stats stats = { { 0 }, };
   struct softmix_mixing_array mixing_array;
   struct softmix_bridge_data *softmix_data = bridge->bridge_pvt;
   struct ast_timer *timer;
   struct softmix_translate_helper trans_helper;
   int16_t buf[MAX_DATALEN] = { 0, };
   unsigned int stat_iteration_counter = 0; /* counts down, gather stats at zero and reset. */
   int timingfd;
   int update_all_rates = 0; /* set this when the internal sample rate has changed */
   int i, x;
   int res = -1;

   if (!(softmix_data = bridge->bridge_pvt)) {
      goto softmix_cleanup;
   }

   ao2_ref(softmix_data, 1);
   timer = softmix_data->timer;
   timingfd = ast_timer_fd(timer);
   softmix_translate_helper_init(&trans_helper, softmix_data->internal_rate);
   ast_timer_set_rate(timer, (1000 / softmix_data->internal_mixing_interval));

   /* Give the mixing array room to grow, memory is cheap but allocations are expensive. */
   if (softmix_mixing_array_init(&mixing_array, bridge->num + 10)) {
      ast_log(LOG_NOTICE, "Failed to allocate softmix mixing structure. \n");
      goto softmix_cleanup;
   }

   while (!bridge->stop && !bridge->refresh && bridge->array_num) {
      struct ast_bridge_channel *bridge_channel = NULL;
      int timeout = -1;
      enum ast_format_id cur_slin_id = ast_format_slin_by_rate(softmix_data->internal_rate);
      unsigned int softmix_samples = SOFTMIX_SAMPLES(softmix_data->internal_rate, softmix_data->internal_mixing_interval);
      unsigned int softmix_datalen = SOFTMIX_DATALEN(softmix_data->internal_rate, softmix_data->internal_mixing_interval);

      if (softmix_datalen > MAX_DATALEN) {
         /* This should NEVER happen, but if it does we need to know about it. Almost
          * all the memcpys used during this process depend on this assumption.  Rather
          * than checking this over and over again through out the code, this single
          * verification is done on each iteration. */
         ast_log(LOG_WARNING, "Conference mixing error, requested mixing length greater than mixing buffer.\n");
         goto softmix_cleanup;
      }

      /* Grow the mixing array buffer as participants are added. */
      if (mixing_array.max_num_entries < bridge->num && softmix_mixing_array_grow(&mixing_array, bridge->num + 5)) {
         goto softmix_cleanup;
      }

      /* init the number of buffers stored in the mixing array to 0.
       * As buffers are added for mixing, this number is incremented. */
      mixing_array.used_entries = 0;

      /* These variables help determine if a rate change is required */
      if (!stat_iteration_counter) {
         memset(&stats, 0, sizeof(stats));
         stats.locked_rate = bridge->internal_sample_rate;
      }

      /* If the sample rate has changed, update the translator helper */
      if (update_all_rates) {
         softmix_translate_helper_change_rate(&trans_helper, softmix_data->internal_rate);
      }

      /* Go through pulling audio from each factory that has it available */
      AST_LIST_TRAVERSE(&bridge->channels, bridge_channel, entry) {
         struct softmix_channel *sc = bridge_channel->bridge_pvt;

         /* Update the sample rate to match the bridge's native sample rate if necessary. */
         if (update_all_rates) {
            set_softmix_bridge_data(softmix_data->internal_rate, softmix_data->internal_mixing_interval, bridge_channel, 1);
         }

         /* If stat_iteration_counter is 0, then collect statistics during this mixing interation */
         if (!stat_iteration_counter) {
            gather_softmix_stats(&stats, softmix_data, bridge_channel);
         }

         /* if the channel is suspended, don't check for audio, but still gather stats */
         if (bridge_channel->suspended) {
            continue;
         }

         /* Try to get audio from the factory if available */
         ast_mutex_lock(&sc->lock);
         if ((mixing_array.buffers[mixing_array.used_entries] = softmix_process_read_audio(sc, softmix_samples))) {
            mixing_array.used_entries++;
         }
         ast_mutex_unlock(&sc->lock);
      }

      /* mix it like crazy */
      memset(buf, 0, softmix_datalen);
      for (i = 0; i < mixing_array.used_entries; i++) {
         for (x = 0; x < softmix_samples; x++) {
            ast_slinear_saturated_add(buf + x, mixing_array.buffers[i] + x);
         }
      }

      /* Next step go through removing the channel's own audio and creating a good frame... */
      AST_LIST_TRAVERSE(&bridge->channels, bridge_channel, entry) {
         struct softmix_channel *sc = bridge_channel->bridge_pvt;

         if (bridge_channel->suspended) {
            continue;
         }

         ast_mutex_lock(&sc->lock);

         /* Make SLINEAR write frame from local buffer */
         if (sc->write_frame.subclass.format.id != cur_slin_id) {
            ast_format_set(&sc->write_frame.subclass.format, cur_slin_id, 0);
         }
         sc->write_frame.datalen = softmix_datalen;
         sc->write_frame.samples = softmix_samples;
         memcpy(sc->final_buf, buf, softmix_datalen);

         /* process the softmix channel's new write audio */
         softmix_process_write_audio(&trans_helper, ast_channel_rawwriteformat(bridge_channel->chan), sc);

         /* The frame is now ready for use... */
         sc->have_frame = 1;

         ast_mutex_unlock(&sc->lock);

         /* Poke bridged channel thread just in case */
         pthread_kill(bridge_channel->thread, SIGURG);
      }

      update_all_rates = 0;
      if (!stat_iteration_counter) {
         update_all_rates = analyse_softmix_stats(&stats, softmix_data);
         stat_iteration_counter = SOFTMIX_STAT_INTERVAL;
      }
      stat_iteration_counter--;

      ao2_unlock(bridge);
      /* cleanup any translation frame data from the previous mixing iteration. */
      softmix_translate_helper_cleanup(&trans_helper);
      /* Wait for the timing source to tell us to wake up and get things done */
      ast_waitfor_n_fd(&timingfd, 1, &timeout, NULL);
      ast_timer_ack(timer, 1);
      ao2_lock(bridge);

      /* make sure to detect mixing interval changes if they occur. */
      if (bridge->internal_mixing_interval && (bridge->internal_mixing_interval != softmix_data->internal_mixing_interval)) {
         softmix_data->internal_mixing_interval = bridge->internal_mixing_interval;
         ast_timer_set_rate(timer, (1000 / softmix_data->internal_mixing_interval));
         update_all_rates = 1; /* if the interval changes, the rates must be adjusted as well just to be notified new interval.*/
      }
   }

   res = 0;

softmix_cleanup:
   softmix_translate_helper_destroy(&trans_helper);
   softmix_mixing_array_destroy(&mixing_array);
   if (softmix_data) {
      ao2_ref(softmix_data, -1);
   }
   return res;
}

static struct ast_bridge_technology softmix_bridge = {
   .name = "softmix",
   .capabilities = AST_BRIDGE_CAPABILITY_MULTIMIX | AST_BRIDGE_CAPABILITY_THREAD | AST_BRIDGE_CAPABILITY_MULTITHREADED | AST_BRIDGE_CAPABILITY_OPTIMIZE | AST_BRIDGE_CAPABILITY_VIDEO,
   .preference = AST_BRIDGE_PREFERENCE_LOW,
   .create = softmix_bridge_create,
   .destroy = softmix_bridge_destroy,
   .join = softmix_bridge_join,
   .leave = softmix_bridge_leave,
   .write = softmix_bridge_write,
   .thread = softmix_bridge_thread,
   .poke = softmix_bridge_poke,
};

static int unload_module(void)
{
   ast_format_cap_destroy(softmix_bridge.format_capabilities);
   return ast_bridge_technology_unregister(&softmix_bridge);
}

static int load_module(void)
{
   struct ast_format tmp;
   if (!(softmix_bridge.format_capabilities = ast_format_cap_alloc())) {
      return AST_MODULE_LOAD_DECLINE;
   }
   ast_format_cap_add(softmix_bridge.format_capabilities, ast_format_set(&tmp, AST_FORMAT_SLINEAR, 0));
   return ast_bridge_technology_register(&softmix_bridge);
}

AST_MODULE_INFO_STANDARD(ASTERISK_GPL_KEY, "Multi-party software based channel mixing");