/* Copyright (C) 2000-2006 Paul Davis This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version. This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with this program; if not, write to the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "i18n.h" #include using namespace std; using namespace ARDOUR; using namespace PBD; size_t AudioDiskstream::_working_buffers_size = 0; Sample* AudioDiskstream::_mixdown_buffer = 0; gain_t* AudioDiskstream::_gain_buffer = 0; AudioDiskstream::AudioDiskstream (Session &sess, const string &name, Diskstream::Flag flag) : Diskstream(sess, name, flag) , deprecated_io_node(NULL) , channels (new ChannelList) { /* prevent any write sources from being created */ in_set_state = true; init(flag); use_new_playlist (); in_set_state = false; } AudioDiskstream::AudioDiskstream (Session& sess, const XMLNode& node) : Diskstream(sess, node) , deprecated_io_node(NULL) , channels (new ChannelList) { in_set_state = true; init (Recordable); if (set_state (node)) { in_set_state = false; throw failed_constructor(); } in_set_state = false; if (destructive()) { use_destructive_playlist (); } } void AudioDiskstream::init (Diskstream::Flag f) { Diskstream::init(f); /* there are no channels at this point, so these two calls just get speed_buffer_size and wrap_buffer size setup without duplicating their code. */ set_block_size (_session.get_block_size()); allocate_temporary_buffers (); add_channel (1); assert(_n_channels == 1); } AudioDiskstream::~AudioDiskstream () { notify_callbacks (); { RCUWriter writer (channels); boost::shared_ptr c = writer.get_copy(); for (ChannelList::iterator chan = c->begin(); chan != c->end(); ++chan) { delete *chan; } c->clear(); } channels.flush (); } void AudioDiskstream::allocate_working_buffers() { assert(disk_io_frames() > 0); _working_buffers_size = disk_io_frames(); _mixdown_buffer = new Sample[_working_buffers_size]; _gain_buffer = new gain_t[_working_buffers_size]; } void AudioDiskstream::free_working_buffers() { delete [] _mixdown_buffer; delete [] _gain_buffer; _working_buffers_size = 0; _mixdown_buffer = 0; _gain_buffer = 0; } void AudioDiskstream::non_realtime_input_change () { { Glib::Mutex::Lock lm (state_lock); if (input_change_pending == NoChange) { return; } { RCUWriter writer (channels); boost::shared_ptr c = writer.get_copy(); _n_channels = c->size(); if (_io->n_inputs() > _n_channels) { add_channel_to (c, _io->n_inputs() - _n_channels); } else if (_io->n_inputs() < _n_channels) { remove_channel_from (c, _n_channels - _io->n_inputs()); } } get_input_sources (); set_capture_offset (); if (first_input_change) { set_align_style (_persistent_alignment_style); first_input_change = false; } else { set_align_style_from_io (); } input_change_pending = NoChange; /* implicit unlock */ } /* reset capture files */ reset_write_sources (false); /* now refill channel buffers */ if (speed() != 1.0f || speed() != -1.0f) { seek ((nframes_t) (_session.transport_frame() * (double) speed())); } else { seek (_session.transport_frame()); } } void AudioDiskstream::get_input_sources () { boost::shared_ptr c = channels.reader(); uint32_t n; ChannelList::iterator chan; uint32_t ni = _io->n_inputs(); for (n = 0, chan = c->begin(); chan != c->end() && n < ni; ++chan, ++n) { const char **connections = _io->input(n)->get_connections (); if (connections == 0 || connections[0] == 0) { if ((*chan)->source) { // _source->disable_metering (); } (*chan)->source = 0; } else { (*chan)->source = _session.engine().get_port_by_name (connections[0]); } if (connections) { free (connections); } } } int AudioDiskstream::find_and_use_playlist (const string& name) { boost::shared_ptr playlist; if ((playlist = boost::dynamic_pointer_cast (_session.playlist_by_name (name))) == 0) { playlist = boost::dynamic_pointer_cast (PlaylistFactory::create (_session, name)); } if (!playlist) { error << string_compose(_("AudioDiskstream: Playlist \"%1\" isn't an audio playlist"), name) << endmsg; return -1; } return use_playlist (playlist); } int AudioDiskstream::use_playlist (boost::shared_ptr playlist) { assert(boost::dynamic_pointer_cast(playlist)); Diskstream::use_playlist(playlist); return 0; } int AudioDiskstream::use_new_playlist () { string newname; boost::shared_ptr playlist; if (!in_set_state && destructive()) { return 0; } if (_playlist) { newname = Playlist::bump_name (_playlist->name(), _session); } else { newname = Playlist::bump_name (_name, _session); } if ((playlist = boost::dynamic_pointer_cast (PlaylistFactory::create (_session, newname, hidden()))) != 0) { playlist->set_orig_diskstream_id (id()); return use_playlist (playlist); } else { return -1; } } int AudioDiskstream::use_copy_playlist () { assert(audio_playlist()); if (destructive()) { return 0; } if (_playlist == 0) { error << string_compose(_("AudioDiskstream %1: there is no existing playlist to make a copy of!"), _name) << endmsg; return -1; } string newname; boost::shared_ptr playlist; newname = Playlist::bump_name (_playlist->name(), _session); if ((playlist = boost::dynamic_pointer_cast(PlaylistFactory::create (audio_playlist(), newname))) != 0) { playlist->set_orig_diskstream_id (id()); return use_playlist (playlist); } else { return -1; } } void AudioDiskstream::setup_destructive_playlist () { SourceList srcs; boost::shared_ptr c = channels.reader(); for (ChannelList::iterator chan = c->begin(); chan != c->end(); ++chan) { srcs.push_back ((*chan)->write_source); } /* a single full-sized region */ boost::shared_ptr region (RegionFactory::create (srcs, 0, max_frames - srcs.front()->natural_position(), _name)); _playlist->add_region (region, srcs.front()->natural_position()); } void AudioDiskstream::use_destructive_playlist () { /* this is called from the XML-based constructor or ::set_destructive. when called, we already have a playlist and a region, but we need to set up our sources for write. we use the sources associated with the (presumed single, full-extent) region. */ boost::shared_ptr rp = _playlist->find_next_region (_session.current_start_frame(), Start, 1); if (!rp) { reset_write_sources (false, true); return; } boost::shared_ptr region = boost::dynamic_pointer_cast (rp); if (region == 0) { throw failed_constructor(); } /* be sure to stretch the region out to the maximum length */ region->set_length (max_frames - region->position(), this); uint32_t n; ChannelList::iterator chan; boost::shared_ptr c = channels.reader(); for (n = 0, chan = c->begin(); chan != c->end(); ++chan, ++n) { (*chan)->write_source = boost::dynamic_pointer_cast(region->source (n)); assert((*chan)->write_source); (*chan)->write_source->set_allow_remove_if_empty (false); /* this might be false if we switched modes, so force it */ (*chan)->write_source->set_destructive (true); } /* the source list will never be reset for a destructive track */ } void AudioDiskstream::check_record_status (nframes_t transport_frame, nframes_t nframes, bool can_record) { int possibly_recording; int rolling; int change; const int transport_rolling = 0x4; const int track_rec_enabled = 0x2; const int global_rec_enabled = 0x1; /* merge together the 3 factors that affect record status, and compute what has changed. */ rolling = _session.transport_speed() != 0.0f; possibly_recording = (rolling << 2) | (record_enabled() << 1) | can_record; change = possibly_recording ^ last_possibly_recording; if (possibly_recording == last_possibly_recording) { return; } /* change state */ /* if per-track or global rec-enable turned on while the other was already on, we've started recording */ if ((change & track_rec_enabled) && record_enabled() && (!(change & global_rec_enabled) && can_record) || ((change & global_rec_enabled) && can_record && (!(change & track_rec_enabled) && record_enabled()))) { /* starting to record: compute first+last frames */ first_recordable_frame = transport_frame + _capture_offset; last_recordable_frame = max_frames; capture_start_frame = transport_frame; if (!(last_possibly_recording & transport_rolling) && (possibly_recording & transport_rolling)) { /* was stopped, now rolling (and recording) */ if (_alignment_style == ExistingMaterial) { first_recordable_frame += _session.worst_output_latency(); } else { first_recordable_frame += _roll_delay; } } else { /* was rolling, but record state changed */ if (_alignment_style == ExistingMaterial) { if (!Config->get_punch_in()) { /* manual punch in happens at the correct transport frame because the user hit a button. but to get alignment correct we have to back up the position of the new region to the appropriate spot given the roll delay. */ capture_start_frame -= _roll_delay; /* XXX paul notes (august 2005): i don't know why this is needed. */ first_recordable_frame += _capture_offset; } else { /* autopunch toggles recording at the precise transport frame, and then the DS waits to start recording for a time that depends on the output latency. */ first_recordable_frame += _session.worst_output_latency(); } } else { if (Config->get_punch_in()) { first_recordable_frame += _roll_delay; } else { capture_start_frame -= _roll_delay; } } } if (_flags & Recordable) { boost::shared_ptr c = channels.reader(); for (ChannelList::iterator chan = c->begin(); chan != c->end(); ++chan) { RingBufferNPT::rw_vector transvec; (*chan)->capture_transition_buf->get_write_vector(&transvec); if (transvec.len[0] > 0) { transvec.buf[0]->type = CaptureStart; transvec.buf[0]->capture_val = capture_start_frame; (*chan)->capture_transition_buf->increment_write_ptr(1); } else { // bad! fatal << X_("programming error: capture_transition_buf is full on rec start! inconceivable!") << endmsg; } } } } else if (!record_enabled() || !can_record) { /* stop recording */ last_recordable_frame = transport_frame + _capture_offset; if (_alignment_style == ExistingMaterial) { last_recordable_frame += _session.worst_output_latency(); } else { last_recordable_frame += _roll_delay; } } last_possibly_recording = possibly_recording; } int AudioDiskstream::process (nframes_t transport_frame, nframes_t nframes, nframes_t offset, bool can_record, bool rec_monitors_input) { uint32_t n; boost::shared_ptr c = channels.reader(); ChannelList::iterator chan; int ret = -1; nframes_t rec_offset = 0; nframes_t rec_nframes = 0; bool nominally_recording; bool re = record_enabled (); bool collect_playback = false; /* if we've already processed the frames corresponding to this call, just return. this allows multiple routes that are taking input from this diskstream to call our ::process() method, but have this stuff only happen once. more commonly, it allows both the AudioTrack that is using this AudioDiskstream *and* the Session to call process() without problems. */ if (_processed) { return 0; } commit_should_unlock = false; check_record_status (transport_frame, nframes, can_record); nominally_recording = (can_record && re); if (nframes == 0) { _processed = true; return 0; } /* This lock is held until the end of AudioDiskstream::commit, so these two functions must always be called as a pair. The only exception is if this function returns a non-zero value, in which case, ::commit should not be called. */ // If we can't take the state lock return. if (!state_lock.trylock()) { return 1; } commit_should_unlock = true; adjust_capture_position = 0; for (chan = c->begin(); chan != c->end(); ++chan) { (*chan)->current_capture_buffer = 0; (*chan)->current_playback_buffer = 0; } if (nominally_recording || (_session.get_record_enabled() && Config->get_punch_in())) { OverlapType ot; ot = coverage (first_recordable_frame, last_recordable_frame, transport_frame, transport_frame + nframes); switch (ot) { case OverlapNone: rec_nframes = 0; break; case OverlapInternal: /* ---------- recrange |---| transrange */ rec_nframes = nframes; rec_offset = 0; break; case OverlapStart: /* |--------| recrange -----| transrange */ rec_nframes = transport_frame + nframes - first_recordable_frame; if (rec_nframes) { rec_offset = first_recordable_frame - transport_frame; } break; case OverlapEnd: /* |--------| recrange |-------- transrange */ rec_nframes = last_recordable_frame - transport_frame; rec_offset = 0; break; case OverlapExternal: /* |--------| recrange -------------- transrange */ rec_nframes = last_recordable_frame - last_recordable_frame; rec_offset = first_recordable_frame - transport_frame; break; } if (rec_nframes && !was_recording) { capture_captured = 0; was_recording = true; } } if (can_record && !_last_capture_regions.empty()) { _last_capture_regions.clear (); } if (nominally_recording || rec_nframes) { uint32_t limit = _io->n_inputs (); /* one or more ports could already have been removed from _io, but our channel setup hasn't yet been updated. prevent us from trying to use channels that correspond to missing ports. note that the process callback (from which this is called) is always atomic with respect to port removal/addition. */ for (n = 0, chan = c->begin(); chan != c->end() && n < limit; ++chan, ++n) { ChannelInfo* chaninfo (*chan); chaninfo->capture_buf->get_write_vector (&chaninfo->capture_vector); if (rec_nframes <= chaninfo->capture_vector.len[0]) { chaninfo->current_capture_buffer = chaninfo->capture_vector.buf[0]; /* note: grab the entire port buffer, but only copy what we were supposed to for recording, and use rec_offset */ memcpy (chaninfo->current_capture_buffer, _io->input(n)->get_buffer (rec_nframes) + offset + rec_offset, sizeof (Sample) * rec_nframes); } else { nframes_t total = chaninfo->capture_vector.len[0] + chaninfo->capture_vector.len[1]; if (rec_nframes > total) { DiskOverrun (); goto out; } Sample* buf = _io->input (n)->get_buffer (nframes) + offset; nframes_t first = chaninfo->capture_vector.len[0]; memcpy (chaninfo->capture_wrap_buffer, buf, sizeof (Sample) * first); memcpy (chaninfo->capture_vector.buf[0], buf, sizeof (Sample) * first); memcpy (chaninfo->capture_wrap_buffer+first, buf + first, sizeof (Sample) * (rec_nframes - first)); memcpy (chaninfo->capture_vector.buf[1], buf + first, sizeof (Sample) * (rec_nframes - first)); chaninfo->current_capture_buffer = chaninfo->capture_wrap_buffer; } } } else { if (was_recording) { finish_capture (rec_monitors_input, c); } } if (rec_nframes) { /* data will be written to disk */ if (rec_nframes == nframes && rec_offset == 0) { for (chan = c->begin(); chan != c->end(); ++chan) { (*chan)->current_playback_buffer = (*chan)->current_capture_buffer; } playback_distance = nframes; } else { /* we can't use the capture buffer as the playback buffer, because we recorded only a part of the current process' cycle data for capture. */ collect_playback = true; } adjust_capture_position = rec_nframes; } else if (nominally_recording) { /* can't do actual capture yet - waiting for latency effects to finish before we start*/ for (chan = c->begin(); chan != c->end(); ++chan) { (*chan)->current_playback_buffer = (*chan)->current_capture_buffer; } playback_distance = nframes; } else { collect_playback = true; } if (collect_playback) { /* we're doing playback */ nframes_t necessary_samples; /* no varispeed playback if we're recording, because the output .... TBD */ if (rec_nframes == 0 && _actual_speed != 1.0f) { necessary_samples = (nframes_t) floor ((nframes * fabs (_actual_speed))) + 1; } else { necessary_samples = nframes; } for (chan = c->begin(); chan != c->end(); ++chan) { (*chan)->playback_buf->get_read_vector (&(*chan)->playback_vector); } n = 0; for (chan = c->begin(); chan != c->end(); ++chan, ++n) { ChannelInfo* chaninfo (*chan); if (necessary_samples <= chaninfo->playback_vector.len[0]) { chaninfo->current_playback_buffer = chaninfo->playback_vector.buf[0]; } else { nframes_t total = chaninfo->playback_vector.len[0] + chaninfo->playback_vector.len[1]; if (necessary_samples > total) { DiskUnderrun (); goto out; } else { memcpy ((char *) chaninfo->playback_wrap_buffer, chaninfo->playback_vector.buf[0], chaninfo->playback_vector.len[0] * sizeof (Sample)); memcpy (chaninfo->playback_wrap_buffer + chaninfo->playback_vector.len[0], chaninfo->playback_vector.buf[1], (necessary_samples - chaninfo->playback_vector.len[0]) * sizeof (Sample)); chaninfo->current_playback_buffer = chaninfo->playback_wrap_buffer; } } } if (rec_nframes == 0 && _actual_speed != 1.0f && _actual_speed != -1.0f) { uint64_t phase = last_phase; nframes_t i = 0; // Linearly interpolate into the alt buffer // using 40.24 fixp maths (swh) for (chan = c->begin(); chan != c->end(); ++chan) { float fr; ChannelInfo* chaninfo (*chan); i = 0; phase = last_phase; for (nframes_t outsample = 0; outsample < nframes; ++outsample) { i = phase >> 24; fr = (phase & 0xFFFFFF) / 16777216.0f; chaninfo->speed_buffer[outsample] = chaninfo->current_playback_buffer[i] * (1.0f - fr) + chaninfo->current_playback_buffer[i+1] * fr; phase += phi; } chaninfo->current_playback_buffer = chaninfo->speed_buffer; } playback_distance = i + 1; last_phase = (phase & 0xFFFFFF); } else { playback_distance = nframes; } } ret = 0; out: _processed = true; if (ret) { /* we're exiting with failure, so ::commit will not be called. unlock the state lock. */ commit_should_unlock = false; state_lock.unlock(); } return ret; } bool AudioDiskstream::commit (nframes_t nframes) { bool need_butler = false; if (_actual_speed < 0.0) { playback_sample -= playback_distance; } else { playback_sample += playback_distance; } boost::shared_ptr c = channels.reader(); for (ChannelList::iterator chan = c->begin(); chan != c->end(); ++chan) { (*chan)->playback_buf->increment_read_ptr (playback_distance); if (adjust_capture_position) { (*chan)->capture_buf->increment_write_ptr (adjust_capture_position); } } if (adjust_capture_position != 0) { capture_captured += adjust_capture_position; adjust_capture_position = 0; } if (_slaved) { need_butler = c->front()->playback_buf->write_space() >= c->front()->playback_buf->bufsize() / 2; } else { need_butler = c->front()->playback_buf->write_space() >= disk_io_chunk_frames || c->front()->capture_buf->read_space() >= disk_io_chunk_frames; } if (commit_should_unlock) { state_lock.unlock(); } _processed = false; return need_butler; } void AudioDiskstream::set_pending_overwrite (bool yn) { /* called from audio thread, so we can use the read ptr and playback sample as we wish */ pending_overwrite = yn; overwrite_frame = playback_sample; overwrite_offset = channels.reader()->front()->playback_buf->get_read_ptr(); } int AudioDiskstream::overwrite_existing_buffers () { boost::shared_ptr c = channels.reader(); Sample* mixdown_buffer; float* gain_buffer; int ret = -1; bool reversed = (_visible_speed * _session.transport_speed()) < 0.0f; overwrite_queued = false; /* assume all are the same size */ nframes_t size = c->front()->playback_buf->bufsize(); mixdown_buffer = new Sample[size]; gain_buffer = new float[size]; /* reduce size so that we can fill the buffer correctly. */ size--; uint32_t n=0; nframes_t start; for (ChannelList::iterator chan = c->begin(); chan != c->end(); ++chan, ++n) { start = overwrite_frame; nframes_t cnt = size; /* to fill the buffer without resetting the playback sample, we need to do it one or two chunks (normally two). |----------------------------------------------------------------------| ^ overwrite_offset |<- second chunk->||<----------------- first chunk ------------------>| */ nframes_t to_read = size - overwrite_offset; if (read ((*chan)->playback_buf->buffer() + overwrite_offset, mixdown_buffer, gain_buffer, start, to_read, *chan, n, reversed)) { error << string_compose(_("AudioDiskstream %1: when refilling, cannot read %2 from playlist at frame %3"), _id, size, playback_sample) << endmsg; goto out; } if (cnt > to_read) { cnt -= to_read; if (read ((*chan)->playback_buf->buffer(), mixdown_buffer, gain_buffer, start, cnt, *chan, n, reversed)) { error << string_compose(_("AudioDiskstream %1: when refilling, cannot read %2 from playlist at frame %3"), _id, size, playback_sample) << endmsg; goto out; } } } ret = 0; out: pending_overwrite = false; delete [] gain_buffer; delete [] mixdown_buffer; return ret; } int AudioDiskstream::seek (nframes_t frame, bool complete_refill) { uint32_t n; int ret; ChannelList::iterator chan; boost::shared_ptr c = channels.reader(); Glib::Mutex::Lock lm (state_lock); for (n = 0, chan = c->begin(); chan != c->end(); ++chan, ++n) { (*chan)->playback_buf->reset (); (*chan)->capture_buf->reset (); } /* can't rec-enable in destructive mode if transport is before start */ if (destructive() && record_enabled() && frame < _session.current_start_frame()) { disengage_record_enable (); } playback_sample = frame; file_frame = frame; if (complete_refill) { while ((ret = do_refill_with_alloc ()) > 0) ; } else { ret = do_refill_with_alloc (); } return ret; } int AudioDiskstream::can_internal_playback_seek (nframes_t distance) { ChannelList::iterator chan; boost::shared_ptr c = channels.reader(); for (chan = c->begin(); chan != c->end(); ++chan) { if ((*chan)->playback_buf->read_space() < distance) { return false; } } return true; } int AudioDiskstream::internal_playback_seek (nframes_t distance) { ChannelList::iterator chan; boost::shared_ptr c = channels.reader(); for (chan = c->begin(); chan != c->end(); ++chan) { (*chan)->playback_buf->increment_read_ptr (distance); } first_recordable_frame += distance; playback_sample += distance; return 0; } int AudioDiskstream::read (Sample* buf, Sample* mixdown_buffer, float* gain_buffer, nframes_t& start, nframes_t cnt, ChannelInfo* channel_info, int channel, bool reversed) { nframes_t this_read = 0; bool reloop = false; nframes_t loop_end = 0; nframes_t loop_start = 0; nframes_t loop_length = 0; nframes_t offset = 0; Location *loc = 0; /* XXX we don't currently play loops in reverse. not sure why */ if (!reversed) { /* Make the use of a Location atomic for this read operation. Note: Locations don't get deleted, so all we care about when I say "atomic" is that we are always pointing to the same one and using a start/length values obtained just once. */ if ((loc = loop_location) != 0) { loop_start = loc->start(); loop_end = loc->end(); loop_length = loop_end - loop_start; } /* if we are looping, ensure that the first frame we read is at the correct position within the loop. */ if (loc && start >= loop_end) { //cerr << "start adjusted from " << start; start = loop_start + ((start - loop_start) % loop_length); //cerr << "to " << start << endl; } //cerr << "start is " << start << " loopstart: " << loop_start << " loopend: " << loop_end << endl; } while (cnt) { if (reversed) { start -= cnt; } /* take any loop into account. we can't read past the end of the loop. */ if (loc && (loop_end - start < cnt)) { this_read = loop_end - start; //cerr << "reloop true: thisread: " << this_read << " cnt: " << cnt << endl; reloop = true; } else { reloop = false; this_read = cnt; } if (this_read == 0) { break; } this_read = min(cnt,this_read); if (audio_playlist()->read (buf+offset, mixdown_buffer, gain_buffer, start, this_read, channel) != this_read) { error << string_compose(_("AudioDiskstream %1: cannot read %2 from playlist at frame %3"), _id, this_read, start) << endmsg; return -1; } _read_data_count = _playlist->read_data_count(); if (reversed) { swap_by_ptr (buf, buf + this_read - 1); } else { /* if we read to the end of the loop, go back to the beginning */ if (reloop) { start = loop_start; } else { start += this_read; } } cnt -= this_read; offset += this_read; } return 0; } int AudioDiskstream::do_refill_with_alloc () { Sample* mix_buf = new Sample[disk_io_chunk_frames]; float* gain_buf = new float[disk_io_chunk_frames]; int ret = _do_refill(mix_buf, gain_buf); delete [] mix_buf; delete [] gain_buf; return ret; } int AudioDiskstream::_do_refill (Sample* mixdown_buffer, float* gain_buffer) { int32_t ret = 0; nframes_t to_read; RingBufferNPT::rw_vector vector; bool reversed = (_visible_speed * _session.transport_speed()) < 0.0f; nframes_t total_space; nframes_t zero_fill; uint32_t chan_n; ChannelList::iterator i; boost::shared_ptr c = channels.reader(); nframes_t ts; if (c->empty()) { return 0; } assert(mixdown_buffer); assert(gain_buffer); vector.buf[0] = 0; vector.len[0] = 0; vector.buf[1] = 0; vector.len[1] = 0; c->front()->playback_buf->get_write_vector (&vector); if ((total_space = vector.len[0] + vector.len[1]) == 0) { return 0; } /* if there are 2+ chunks of disk i/o possible for this track, let the caller know so that it can arrange for us to be called again, ASAP. */ if (total_space >= (_slaved?3:2) * disk_io_chunk_frames) { ret = 1; } /* if we're running close to normal speed and there isn't enough space to do disk_io_chunk_frames of I/O, then don't bother. at higher speeds, just do it because the sync between butler and audio thread may not be good enough. */ if ((total_space < disk_io_chunk_frames) && fabs (_actual_speed) < 2.0f) { return 0; } /* when slaved, don't try to get too close to the read pointer. this leaves space for the buffer reversal to have something useful to work with. */ if (_slaved && total_space < (c->front()->playback_buf->bufsize() / 2)) { return 0; } /* never do more than disk_io_chunk_frames worth of disk input per call (limit doesn't apply for memset) */ total_space = min (disk_io_chunk_frames, total_space); if (reversed) { if (file_frame == 0) { /* at start: nothing to do but fill with silence */ for (chan_n = 0, i = c->begin(); i != c->end(); ++i, ++chan_n) { ChannelInfo* chan (*i); chan->playback_buf->get_write_vector (&vector); memset (vector.buf[0], 0, sizeof(Sample) * vector.len[0]); if (vector.len[1]) { memset (vector.buf[1], 0, sizeof(Sample) * vector.len[1]); } chan->playback_buf->increment_write_ptr (vector.len[0] + vector.len[1]); } return 0; } if (file_frame < total_space) { /* too close to the start: read what we can, and then zero fill the rest */ zero_fill = total_space - file_frame; total_space = file_frame; file_frame = 0; } else { zero_fill = 0; } } else { if (file_frame == max_frames) { /* at end: nothing to do but fill with silence */ for (chan_n = 0, i = c->begin(); i != c->end(); ++i, ++chan_n) { ChannelInfo* chan (*i); chan->playback_buf->get_write_vector (&vector); memset (vector.buf[0], 0, sizeof(Sample) * vector.len[0]); if (vector.len[1]) { memset (vector.buf[1], 0, sizeof(Sample) * vector.len[1]); } chan->playback_buf->increment_write_ptr (vector.len[0] + vector.len[1]); } return 0; } if (file_frame > max_frames - total_space) { /* to close to the end: read what we can, and zero fill the rest */ zero_fill = total_space - (max_frames - file_frame); total_space = max_frames - file_frame; } else { zero_fill = 0; } } nframes_t file_frame_tmp = 0; for (chan_n = 0, i = c->begin(); i != c->end(); ++i, ++chan_n) { ChannelInfo* chan (*i); Sample* buf1; Sample* buf2; nframes_t len1, len2; chan->playback_buf->get_write_vector (&vector); if (vector.len[0] > disk_io_chunk_frames) { /* we're not going to fill the first chunk, so certainly do not bother with the other part. it won't be connected with the part we do fill, as in: .... => writable space ++++ => readable space ^^^^ => 1 x disk_io_chunk_frames that would be filled |......|+++++++++++++|...............................| buf1 buf0 ^^^^^^^^^^^^^^^ So, just pretend that the buf1 part isn't there. */ vector.buf[1] = 0; vector.len[1] = 0; } ts = total_space; file_frame_tmp = file_frame; buf1 = vector.buf[0]; len1 = vector.len[0]; buf2 = vector.buf[1]; len2 = vector.len[1]; to_read = min (ts, len1); to_read = min (to_read, disk_io_chunk_frames); if (to_read) { if (read (buf1, mixdown_buffer, gain_buffer, file_frame_tmp, to_read, chan, chan_n, reversed)) { ret = -1; goto out; } chan->playback_buf->increment_write_ptr (to_read); ts -= to_read; } to_read = min (ts, len2); if (to_read) { /* we read all of vector.len[0], but it wasn't an entire disk_io_chunk_frames of data, so read some or all of vector.len[1] as well. */ if (read (buf2, mixdown_buffer, gain_buffer, file_frame_tmp, to_read, chan, chan_n, reversed)) { ret = -1; goto out; } chan->playback_buf->increment_write_ptr (to_read); } if (zero_fill) { /* do something */ } } file_frame = file_frame_tmp; out: return ret; } /** Flush pending data to disk. * * Important note: this function will write *AT MOST* disk_io_chunk_frames * of data to disk. it will never write more than that. If it writes that * much and there is more than that waiting to be written, it will return 1, * otherwise 0 on success or -1 on failure. * * If there is less than disk_io_chunk_frames to be written, no data will be * written at all unless @a force_flush is true. */ int AudioDiskstream::do_flush (Session::RunContext context, bool force_flush) { uint32_t to_write; int32_t ret = 0; RingBufferNPT::rw_vector vector; RingBufferNPT::rw_vector transvec; nframes_t total; _write_data_count = 0; transvec.buf[0] = 0; transvec.buf[1] = 0; vector.buf[0] = 0; vector.buf[1] = 0; boost::shared_ptr c = channels.reader(); for (ChannelList::iterator chan = c->begin(); chan != c->end(); ++chan) { (*chan)->capture_buf->get_read_vector (&vector); total = vector.len[0] + vector.len[1]; if (total == 0 || (total < disk_io_chunk_frames && !force_flush && was_recording)) { goto out; } /* if there are 2+ chunks of disk i/o possible for this track, let the caller know so that it can arrange for us to be called again, ASAP. if we are forcing a flush, then if there is* any* extra work, let the caller know. if we are no longer recording and there is any extra work, let the caller know too. */ if (total >= 2 * disk_io_chunk_frames || ((force_flush || !was_recording) && total > disk_io_chunk_frames)) { ret = 1; } to_write = min (disk_io_chunk_frames, (nframes_t) vector.len[0]); // check the transition buffer when recording destructive // important that we get this after the capture buf if (destructive()) { (*chan)->capture_transition_buf->get_read_vector(&transvec); size_t transcount = transvec.len[0] + transvec.len[1]; bool have_start = false; size_t ti; for (ti=0; ti < transcount; ++ti) { CaptureTransition & captrans = (ti < transvec.len[0]) ? transvec.buf[0][ti] : transvec.buf[1][ti-transvec.len[0]]; if (captrans.type == CaptureStart) { // by definition, the first data we got above represents the given capture pos (*chan)->write_source->mark_capture_start (captrans.capture_val); (*chan)->curr_capture_cnt = 0; have_start = true; } else if (captrans.type == CaptureEnd) { // capture end, the capture_val represents total frames in capture if (captrans.capture_val <= (*chan)->curr_capture_cnt + to_write) { // shorten to make the write a perfect fit uint32_t nto_write = (captrans.capture_val - (*chan)->curr_capture_cnt); if (nto_write < to_write) { ret = 1; // should we? } to_write = nto_write; (*chan)->write_source->mark_capture_end (); // increment past this transition, but go no further ++ti; break; } else { // actually ends just beyond this chunk, so force more work ret = 1; break; } } } if (ti > 0) { (*chan)->capture_transition_buf->increment_read_ptr(ti); } } if ((!(*chan)->write_source) || (*chan)->write_source->write (vector.buf[0], to_write) != to_write) { error << string_compose(_("AudioDiskstream %1: cannot write to disk"), _id) << endmsg; return -1; } (*chan)->capture_buf->increment_read_ptr (to_write); (*chan)->curr_capture_cnt += to_write; if ((to_write == vector.len[0]) && (total > to_write) && (to_write < disk_io_chunk_frames) && !destructive()) { /* we wrote all of vector.len[0] but it wasn't an entire disk_io_chunk_frames of data, so arrange for some part of vector.len[1] to be flushed to disk as well. */ to_write = min ((nframes_t)(disk_io_chunk_frames - to_write), (nframes_t) vector.len[1]); if ((*chan)->write_source->write (vector.buf[1], to_write) != to_write) { error << string_compose(_("AudioDiskstream %1: cannot write to disk"), _id) << endmsg; return -1; } _write_data_count += (*chan)->write_source->write_data_count(); (*chan)->capture_buf->increment_read_ptr (to_write); (*chan)->curr_capture_cnt += to_write; } } out: return ret; } void AudioDiskstream::transport_stopped (struct tm& when, time_t twhen, bool abort_capture) { uint32_t buffer_position; bool more_work = true; int err = 0; boost::shared_ptr region; nframes_t total_capture; SourceList srcs; SourceList::iterator src; ChannelList::iterator chan; vector::iterator ci; boost::shared_ptr c = channels.reader(); uint32_t n = 0; bool mark_write_completed = false; finish_capture (true, c); /* butler is already stopped, but there may be work to do to flush remaining data to disk. */ while (more_work && !err) { switch (do_flush (Session::TransportContext, true)) { case 0: more_work = false; break; case 1: break; case -1: error << string_compose(_("AudioDiskstream \"%1\": cannot flush captured data to disk!"), _name) << endmsg; err++; } } /* XXX is there anything we can do if err != 0 ? */ Glib::Mutex::Lock lm (capture_info_lock); if (capture_info.empty()) { return; } if (abort_capture) { if (destructive()) { goto outout; } for (ChannelList::iterator chan = c->begin(); chan != c->end(); ++chan) { if ((*chan)->write_source) { (*chan)->write_source->mark_for_remove (); (*chan)->write_source->drop_references (); (*chan)->write_source.reset (); } /* new source set up in "out" below */ } goto out; } for (total_capture = 0, ci = capture_info.begin(); ci != capture_info.end(); ++ci) { total_capture += (*ci)->frames; } /* figure out the name for this take */ for (n = 0, chan = c->begin(); chan != c->end(); ++chan, ++n) { boost::shared_ptr s = (*chan)->write_source; if (s) { srcs.push_back (s); s->update_header (capture_info.front()->start, when, twhen); s->set_captured_for (_name); s->mark_immutable (); } } /* destructive tracks have a single, never changing region */ if (destructive()) { /* send a signal that any UI can pick up to do the right thing. there is a small problem here in that a UI may need the peak data to be ready for the data that was recorded and this isn't interlocked with that process. this problem is deferred to the UI. */ _playlist->Modified(); } else { string whole_file_region_name; whole_file_region_name = region_name_from_path (c->front()->write_source->name(), true); /* Register a new region with the Session that describes the entire source. Do this first so that any sub-regions will obviously be children of this one (later!) */ try { boost::shared_ptr rx (RegionFactory::create (srcs, c->front()->write_source->last_capture_start_frame(), total_capture, whole_file_region_name, 0, AudioRegion::Flag (AudioRegion::DefaultFlags|AudioRegion::Automatic|AudioRegion::WholeFile))); region = boost::dynamic_pointer_cast (rx); region->special_set_position (capture_info.front()->start); } catch (failed_constructor& err) { error << string_compose(_("%1: could not create region for complete audio file"), _name) << endmsg; /* XXX what now? */ } _last_capture_regions.push_back (region); // cerr << _name << ": there are " << capture_info.size() << " capture_info records\n"; XMLNode &before = _playlist->get_state(); _playlist->freeze (); for (buffer_position = c->front()->write_source->last_capture_start_frame(), ci = capture_info.begin(); ci != capture_info.end(); ++ci) { string region_name; _session.region_name (region_name, whole_file_region_name, false); // cerr << _name << ": based on ci of " << (*ci)->start << " for " << (*ci)->frames << " add region " << region_name << endl; try { boost::shared_ptr rx (RegionFactory::create (srcs, buffer_position, (*ci)->frames, region_name)); region = boost::dynamic_pointer_cast (rx); } catch (failed_constructor& err) { error << _("AudioDiskstream: could not create region for captured audio!") << endmsg; continue; /* XXX is this OK? */ } region->GoingAway.connect (bind (mem_fun (*this, &Diskstream::remove_region_from_last_capture), boost::weak_ptr(region))); _last_capture_regions.push_back (region); i_am_the_modifier++; _playlist->add_region (region, (*ci)->start); i_am_the_modifier--; buffer_position += (*ci)->frames; } _playlist->thaw (); XMLNode &after = _playlist->get_state(); _session.add_command (new MementoCommand(*_playlist, &before, &after)); } mark_write_completed = true; out: reset_write_sources (mark_write_completed); outout: for (ci = capture_info.begin(); ci != capture_info.end(); ++ci) { delete *ci; } capture_info.clear (); capture_start_frame = 0; } void AudioDiskstream::finish_capture (bool rec_monitors_input, boost::shared_ptr c) { was_recording = false; if (capture_captured == 0) { return; } if (recordable() && destructive()) { for (ChannelList::iterator chan = c->begin(); chan != c->end(); ++chan) { RingBufferNPT::rw_vector transvec; (*chan)->capture_transition_buf->get_write_vector(&transvec); if (transvec.len[0] > 0) { transvec.buf[0]->type = CaptureEnd; transvec.buf[0]->capture_val = capture_captured; (*chan)->capture_transition_buf->increment_write_ptr(1); } else { // bad! fatal << string_compose (_("programmer error: %1"), X_("capture_transition_buf is full when stopping record! inconceivable!")) << endmsg; } } } CaptureInfo* ci = new CaptureInfo; ci->start = capture_start_frame; ci->frames = capture_captured; /* XXX theoretical race condition here. Need atomic exchange ? However, the circumstances when this is called right now (either on record-disable or transport_stopped) mean that no actual race exists. I think ... We now have a capture_info_lock, but it is only to be used to synchronize in the transport_stop and the capture info accessors, so that invalidation will not occur (both non-realtime). */ // cerr << "Finish capture, add new CI, " << ci->start << '+' << ci->frames << endl; capture_info.push_back (ci); capture_captured = 0; } void AudioDiskstream::set_record_enabled (bool yn) { if (!recordable() || !_session.record_enabling_legal() || _io->n_inputs() == 0) { return; } /* can't rec-enable in destructive mode if transport is before start */ if (destructive() && yn && _session.transport_frame() < _session.current_start_frame()) { return; } if (yn && channels.reader()->front()->source == 0) { /* pick up connections not initiated *from* the IO object we're associated with. */ get_input_sources (); } /* yes, i know that this not proof against race conditions, but its good enough. i think. */ if (record_enabled() != yn) { if (yn) { engage_record_enable (); } else { disengage_record_enable (); } } } void AudioDiskstream::engage_record_enable () { bool rolling = _session.transport_speed() != 0.0f; boost::shared_ptr c = channels.reader(); g_atomic_int_set (&_record_enabled, 1); capturing_sources.clear (); if (Config->get_monitoring_model() == HardwareMonitoring) { for (ChannelList::iterator chan = c->begin(); chan != c->end(); ++chan) { if ((*chan)->source) { (*chan)->source->ensure_monitor_input (!(Config->get_auto_input() && rolling)); } capturing_sources.push_back ((*chan)->write_source); } } else { for (ChannelList::iterator chan = c->begin(); chan != c->end(); ++chan) { capturing_sources.push_back ((*chan)->write_source); } } RecordEnableChanged (); /* EMIT SIGNAL */ } void AudioDiskstream::disengage_record_enable () { g_atomic_int_set (&_record_enabled, 0); boost::shared_ptr c = channels.reader(); for (ChannelList::iterator chan = c->begin(); chan != c->end(); ++chan) { if (Config->get_monitoring_model() == HardwareMonitoring) { if ((*chan)->source) { (*chan)->source->ensure_monitor_input (false); } } } capturing_sources.clear (); RecordEnableChanged (); /* EMIT SIGNAL */ } XMLNode& AudioDiskstream::get_state () { XMLNode* node = new XMLNode ("AudioDiskstream"); char buf[64] = ""; LocaleGuard lg (X_("POSIX")); boost::shared_ptr c = channels.reader(); node->add_property ("flags", enum_2_string (_flags)); snprintf (buf, sizeof(buf), "%zd", c->size()); node->add_property ("channels", buf); node->add_property ("playlist", _playlist->name()); snprintf (buf, sizeof(buf), "%.12g", _visible_speed); node->add_property ("speed", buf); node->add_property("name", _name); id().print (buf, sizeof (buf)); node->add_property("id", buf); if (!capturing_sources.empty() && _session.get_record_enabled()) { XMLNode* cs_child = new XMLNode (X_("CapturingSources")); XMLNode* cs_grandchild; for (vector >::iterator i = capturing_sources.begin(); i != capturing_sources.end(); ++i) { cs_grandchild = new XMLNode (X_("file")); cs_grandchild->add_property (X_("path"), (*i)->path()); cs_child->add_child_nocopy (*cs_grandchild); } /* store the location where capture will start */ Location* pi; if (Config->get_punch_in() && ((pi = _session.locations()->auto_punch_location()) != 0)) { snprintf (buf, sizeof (buf), "%" PRIu32, pi->start()); } else { snprintf (buf, sizeof (buf), "%" PRIu32, _session.transport_frame()); } cs_child->add_property (X_("at"), buf); node->add_child_nocopy (*cs_child); } if (_extra_xml) { node->add_child_copy (*_extra_xml); } return* node; } int AudioDiskstream::set_state (const XMLNode& node) { const XMLProperty* prop; XMLNodeList nlist = node.children(); XMLNodeIterator niter; uint32_t nchans = 1; XMLNode* capture_pending_node = 0; LocaleGuard lg (X_("POSIX")); in_set_state = true; for (niter = nlist.begin(); niter != nlist.end(); ++niter) { if ((*niter)->name() == IO::state_node_name) { deprecated_io_node = new XMLNode (**niter); } if ((*niter)->name() == X_("CapturingSources")) { capture_pending_node = *niter; } } /* prevent write sources from being created */ in_set_state = true; if ((prop = node.property ("name")) != 0) { _name = prop->value(); } if (deprecated_io_node) { if ((prop = deprecated_io_node->property ("id")) != 0) { _id = prop->value (); } } else { if ((prop = node.property ("id")) != 0) { _id = prop->value (); } } if ((prop = node.property ("flags")) != 0) { _flags = Flag (string_2_enum (prop->value(), _flags)); } if ((prop = node.property ("channels")) != 0) { nchans = atoi (prop->value().c_str()); } // create necessary extra channels // we are always constructed with one and we always need one _n_channels = channels.reader()->size(); if (nchans > _n_channels) { add_channel (nchans - _n_channels); } else if (nchans < _n_channels) { remove_channel (_n_channels - nchans); } if ((prop = node.property ("playlist")) == 0) { return -1; } { bool had_playlist = (_playlist != 0); if (find_and_use_playlist (prop->value())) { return -1; } if (!had_playlist) { _playlist->set_orig_diskstream_id (_id); } if (!destructive() && capture_pending_node) { /* destructive streams have one and only one source per channel, and so they never end up in pending capture in any useful sense. */ use_pending_capture_data (*capture_pending_node); } } if ((prop = node.property ("speed")) != 0) { double sp = atof (prop->value().c_str()); if (realtime_set_speed (sp, false)) { non_realtime_set_speed (); } } in_set_state = false; /* make sure this is clear before we do anything else */ capturing_sources.clear (); /* write sources are handled when we handle the input set up of the IO that owns this DS (::non_realtime_input_change()) */ in_set_state = false; return 0; } int AudioDiskstream::use_new_write_source (uint32_t n) { boost::shared_ptr c = channels.reader(); if (!recordable()) { return 1; } if (n >= c->size()) { error << string_compose (_("AudioDiskstream: channel %1 out of range"), n) << endmsg; return -1; } ChannelInfo* chan = (*c)[n]; if (chan->write_source) { chan->write_source->done_with_peakfile_writes (); chan->write_source->set_allow_remove_if_empty (true); chan->write_source.reset (); } try { if ((chan->write_source = _session.create_audio_source_for_session (*this, n, destructive())) == 0) { throw failed_constructor(); } } catch (failed_constructor &err) { error << string_compose (_("%1:%2 new capture file not initialized correctly"), _name, n) << endmsg; chan->write_source.reset (); return -1; } /* do not remove destructive files even if they are empty */ chan->write_source->set_allow_remove_if_empty (!destructive()); return 0; } void AudioDiskstream::reset_write_sources (bool mark_write_complete, bool force) { ChannelList::iterator chan; boost::shared_ptr c = channels.reader(); uint32_t n; if (!recordable()) { return; } capturing_sources.clear (); for (chan = c->begin(), n = 0; chan != c->end(); ++chan, ++n) { if (!destructive()) { if ((*chan)->write_source && mark_write_complete) { (*chan)->write_source->mark_streaming_write_completed (); } use_new_write_source (n); if (record_enabled()) { capturing_sources.push_back ((*chan)->write_source); } } else { if ((*chan)->write_source == 0) { use_new_write_source (n); } } } if (destructive()) { /* we now have all our write sources set up, so create the playlist's single region. */ if (_playlist->empty()) { setup_destructive_playlist (); } } } int AudioDiskstream::rename_write_sources () { ChannelList::iterator chan; boost::shared_ptr c = channels.reader(); uint32_t n; for (chan = c->begin(), n = 0; chan != c->end(); ++chan, ++n) { if ((*chan)->write_source != 0) { (*chan)->write_source->set_name (_name, destructive()); /* XXX what to do if one of them fails ? */ } } return 0; } void AudioDiskstream::set_block_size (nframes_t nframes) { if (_session.get_block_size() > speed_buffer_size) { speed_buffer_size = _session.get_block_size(); boost::shared_ptr c = channels.reader(); for (ChannelList::iterator chan = c->begin(); chan != c->end(); ++chan) { if ((*chan)->speed_buffer) delete [] (*chan)->speed_buffer; (*chan)->speed_buffer = new Sample[speed_buffer_size]; } } allocate_temporary_buffers (); } void AudioDiskstream::allocate_temporary_buffers () { /* make sure the wrap buffer is at least large enough to deal with the speeds up to 1.2, to allow for micro-variation when slaving to MTC, SMPTE etc. */ double sp = max (fabsf (_actual_speed), 1.2f); nframes_t required_wrap_size = (nframes_t) floor (_session.get_block_size() * sp) + 1; if (required_wrap_size > wrap_buffer_size) { boost::shared_ptr c = channels.reader(); for (ChannelList::iterator chan = c->begin(); chan != c->end(); ++chan) { if ((*chan)->playback_wrap_buffer) delete [] (*chan)->playback_wrap_buffer; (*chan)->playback_wrap_buffer = new Sample[required_wrap_size]; if ((*chan)->capture_wrap_buffer) delete [] (*chan)->capture_wrap_buffer; (*chan)->capture_wrap_buffer = new Sample[required_wrap_size]; } wrap_buffer_size = required_wrap_size; } } void AudioDiskstream::monitor_input (bool yn) { boost::shared_ptr c = channels.reader(); for (ChannelList::iterator chan = c->begin(); chan != c->end(); ++chan) { if ((*chan)->source) { (*chan)->source->ensure_monitor_input (yn); } } } void AudioDiskstream::set_align_style_from_io () { bool have_physical = false; if (_io == 0) { return; } get_input_sources (); boost::shared_ptr c = channels.reader(); for (ChannelList::iterator chan = c->begin(); chan != c->end(); ++chan) { if ((*chan)->source && (*chan)->source->flags() & JackPortIsPhysical) { have_physical = true; break; } } if (have_physical) { set_align_style (ExistingMaterial); } else { set_align_style (CaptureTime); } } int AudioDiskstream::add_channel_to (boost::shared_ptr c, uint32_t how_many) { while (how_many--) { c->push_back (new ChannelInfo(_session.diskstream_buffer_size(), speed_buffer_size, wrap_buffer_size)); } _n_channels = c->size(); return 0; } int AudioDiskstream::add_channel (uint32_t how_many) { RCUWriter writer (channels); boost::shared_ptr c = writer.get_copy(); return add_channel_to (c, how_many); } int AudioDiskstream::remove_channel_from (boost::shared_ptr c, uint32_t how_many) { while (--how_many && !c->empty()) { delete c->back(); c->pop_back(); } _n_channels = c->size(); return 0; } int AudioDiskstream::remove_channel (uint32_t how_many) { RCUWriter writer (channels); boost::shared_ptr c = writer.get_copy(); return remove_channel_from (c, how_many); } float AudioDiskstream::playback_buffer_load () const { boost::shared_ptr c = channels.reader(); return (float) ((double) c->front()->playback_buf->read_space()/ (double) c->front()->playback_buf->bufsize()); } float AudioDiskstream::capture_buffer_load () const { boost::shared_ptr c = channels.reader(); return (float) ((double) c->front()->capture_buf->write_space()/ (double) c->front()->capture_buf->bufsize()); } int AudioDiskstream::use_pending_capture_data (XMLNode& node) { const XMLProperty* prop; XMLNodeList nlist = node.children(); XMLNodeIterator niter; boost::shared_ptr fs; boost::shared_ptr first_fs; SourceList pending_sources; nframes_t position; if ((prop = node.property (X_("at"))) == 0) { return -1; } if (sscanf (prop->value().c_str(), "%" PRIu32, &position) != 1) { return -1; } for (niter = nlist.begin(); niter != nlist.end(); ++niter) { if ((*niter)->name() == X_("file")) { if ((prop = (*niter)->property (X_("path"))) == 0) { continue; } try { fs = boost::dynamic_pointer_cast (SourceFactory::createWritable (_session, prop->value(), false, _session.frame_rate())); } catch (failed_constructor& err) { error << string_compose (_("%1: cannot restore pending capture source file %2"), _name, prop->value()) << endmsg; return -1; } pending_sources.push_back (fs); if (first_fs == 0) { first_fs = fs; } fs->set_captured_for (_name); } } if (pending_sources.size() == 0) { /* nothing can be done */ return 1; } if (pending_sources.size() != _n_channels) { error << string_compose (_("%1: incorrect number of pending sources listed - ignoring them all"), _name) << endmsg; return -1; } boost::shared_ptr region; try { region = boost::dynamic_pointer_cast (RegionFactory::create (pending_sources, 0, first_fs->length(), region_name_from_path (first_fs->name(), true), 0, AudioRegion::Flag (AudioRegion::DefaultFlags|AudioRegion::Automatic|AudioRegion::WholeFile))); region->special_set_position (0); } catch (failed_constructor& err) { error << string_compose (_("%1: cannot create whole-file region from pending capture sources"), _name) << endmsg; return -1; } try { region = boost::dynamic_pointer_cast (RegionFactory::create (pending_sources, 0, first_fs->length(), region_name_from_path (first_fs->name(), true))); } catch (failed_constructor& err) { error << string_compose (_("%1: cannot create region from pending capture sources"), _name) << endmsg; return -1; } _playlist->add_region (region, position); return 0; } int AudioDiskstream::set_destructive (bool yn) { bool bounce_ignored; if (yn != destructive()) { if (yn) { /* requestor should already have checked this and bounced if necessary and desired */ if (!can_become_destructive (bounce_ignored)) { return -1; } _flags = Flag (_flags | Destructive); use_destructive_playlist (); } else { _flags = Flag (_flags & ~Destructive); reset_write_sources (true, true); } } return 0; } bool AudioDiskstream::can_become_destructive (bool& requires_bounce) const { if (!_playlist) { requires_bounce = false; return false; } /* is there only one region ? */ if (_playlist->n_regions() != 1) { requires_bounce = true; return false; } boost::shared_ptr first = _playlist->find_next_region (_session.current_start_frame(), Start, 1); assert (first); /* do the source(s) for the region cover the session start position ? */ if (first->position() != _session.current_start_frame()) { if (first->start() > _session.current_start_frame()) { requires_bounce = true; return false; } } /* is the source used by only 1 playlist ? */ boost::shared_ptr afirst = boost::dynamic_pointer_cast (first); assert (afirst); if (afirst->source()->used() > 1) { requires_bounce = true; return false; } requires_bounce = false; return true; } AudioDiskstream::ChannelInfo::ChannelInfo (nframes_t bufsize, nframes_t speed_size, nframes_t wrap_size) { peak_power = 0.0f; source = 0; current_capture_buffer = 0; current_playback_buffer = 0; curr_capture_cnt = 0; speed_buffer = new Sample[speed_size]; playback_wrap_buffer = new Sample[wrap_size]; capture_wrap_buffer = new Sample[wrap_size]; playback_buf = new RingBufferNPT (bufsize); capture_buf = new RingBufferNPT (bufsize); capture_transition_buf = new RingBufferNPT (128); /* touch the ringbuffer buffers, which will cause them to be mapped into locked physical RAM if we're running with mlockall(). this doesn't do much if we're not. */ memset (playback_buf->buffer(), 0, sizeof (Sample) * playback_buf->bufsize()); memset (capture_buf->buffer(), 0, sizeof (Sample) * capture_buf->bufsize()); memset (capture_transition_buf->buffer(), 0, sizeof (CaptureTransition) * capture_transition_buf->bufsize()); } AudioDiskstream::ChannelInfo::~ChannelInfo () { if (write_source) { write_source.reset (); } if (speed_buffer) { delete [] speed_buffer; speed_buffer = 0; } if (playback_wrap_buffer) { delete [] playback_wrap_buffer; playback_wrap_buffer = 0; } if (capture_wrap_buffer) { delete [] capture_wrap_buffer; capture_wrap_buffer = 0; } if (playback_buf) { delete playback_buf; playback_buf = 0; } if (capture_buf) { delete capture_buf; capture_buf = 0; } if (capture_transition_buf) { delete capture_transition_buf; capture_transition_buf = 0; } }