/*********************************************************************** * AUTHOR: Marcus Overhagen, Jérôme Duval * FILE: SoundPlayNode.cpp * DESCR: This is the BBufferProducer, used internally by BSoundPlayer * This belongs into a private namespace, but isn't for * compatibility reasons. ***********************************************************************/ #include #include #include #include #include "SoundPlayNode.h" #include "debug.h" #define DPRINTF 1 #if DPRINTF #undef DPRINTF #define DPRINTF printf #else #undef DPRINTF #define DPRINTF if (1) {} else printf #endif #define SEND_NEW_BUFFER_EVENT (BTimedEventQueue::B_USER_EVENT + 1) _SoundPlayNode::_SoundPlayNode(const char *name, const media_multi_audio_format *format, BSoundPlayer *player) : BMediaNode(name), BBufferProducer(B_MEDIA_RAW_AUDIO), BMediaEventLooper(), mPlayer(player), mInitCheckStatus(B_OK), mOutputEnabled(true), mBufferGroup(NULL), mFramesSent(0), mTooEarlyCount(0) { CALLED(); mFormat.type = B_MEDIA_RAW_AUDIO; mFormat.u.raw_audio = *format; DPRINTF("Format Info:\n"); DPRINTF(" frame_rate: %f\n",mFormat.u.raw_audio.frame_rate); DPRINTF(" channel_count: %ld\n",mFormat.u.raw_audio.channel_count); DPRINTF(" byte_order: %ld (",mFormat.u.raw_audio.byte_order); switch (mFormat.u.raw_audio.byte_order) { case B_MEDIA_BIG_ENDIAN: DPRINTF("B_MEDIA_BIG_ENDIAN)\n"); break; case B_MEDIA_LITTLE_ENDIAN: DPRINTF("B_MEDIA_LITTLE_ENDIAN)\n"); break; default: DPRINTF("unknown)\n"); break; } DPRINTF(" buffer_size: %ld\n",mFormat.u.raw_audio.buffer_size); DPRINTF(" format: %ld (",mFormat.u.raw_audio.format); switch (mFormat.u.raw_audio.format) { case media_raw_audio_format::B_AUDIO_FLOAT: DPRINTF("B_AUDIO_FLOAT)\n"); break; case media_raw_audio_format::B_AUDIO_SHORT: DPRINTF("B_AUDIO_SHORT)\n"); break; case media_raw_audio_format::B_AUDIO_INT: DPRINTF("B_AUDIO_INT)\n"); break; case media_raw_audio_format::B_AUDIO_CHAR: DPRINTF("B_AUDIO_CHAR)\n"); break; case media_raw_audio_format::B_AUDIO_UCHAR: DPRINTF("B_AUDIO_UCHAR)\n"); break; default: DPRINTF("unknown)\n"); break; } } _SoundPlayNode::~_SoundPlayNode() { CALLED(); Quit(); } bool _SoundPlayNode::IsPlaying() { return RunState() == B_STARTED; } bigtime_t _SoundPlayNode::Latency() { return EventLatency(); } media_multi_audio_format _SoundPlayNode::Format() const { return mFormat.u.raw_audio; } // -------------------------------------------------------- // // implementation of BMediaNode // -------------------------------------------------------- // BMediaAddOn * _SoundPlayNode::AddOn(int32 * internal_id) const { CALLED(); // BeBook says this only gets called if we were in an add-on. return NULL; } void _SoundPlayNode::Preroll(void) { CALLED(); // XXX:Performance opportunity BMediaNode::Preroll(); } status_t _SoundPlayNode::HandleMessage(int32 message, const void * data, size_t size) { CALLED(); return B_ERROR; } void _SoundPlayNode::NodeRegistered(void) { CALLED(); if (mInitCheckStatus != B_OK) { ReportError(B_NODE_IN_DISTRESS); return; } SetPriority(B_URGENT_PRIORITY); mOutput.format = mFormat; mOutput.destination = media_destination::null; mOutput.source.port = ControlPort(); mOutput.source.id = 1; mOutput.node = Node(); sprintf(mOutput.name, "output %ld", mOutput.source.id); Run(); } status_t _SoundPlayNode::RequestCompleted(const media_request_info &info) { CALLED(); return B_OK; } void _SoundPlayNode::SetTimeSource(BTimeSource *timeSource) { CALLED(); BMediaNode::SetTimeSource(timeSource); } void _SoundPlayNode::SetRunMode(run_mode mode) { TRACE("_SoundPlayNode::SetRunMode mode:%i\n", mode); BMediaNode::SetRunMode(mode); } // -------------------------------------------------------- // // implementation for BBufferProducer // -------------------------------------------------------- // status_t _SoundPlayNode::FormatSuggestionRequested(media_type type, int32 /*quality*/, media_format* format) { // FormatSuggestionRequested() is not necessarily part of the format negotiation // process; it's simply an interrogation -- the caller wants to see what the node's // preferred data format is, given a suggestion by the caller. CALLED(); // a wildcard type is okay; but we only support raw audio if (type != B_MEDIA_RAW_AUDIO && type != B_MEDIA_UNKNOWN_TYPE) return B_MEDIA_BAD_FORMAT; // this is the format we'll be returning (our preferred format) *format = mFormat; return B_OK; } status_t _SoundPlayNode::FormatProposal(const media_source& output, media_format* format) { // FormatProposal() is the first stage in the BMediaRoster::Connect() process. We hand // out a suggested format, with wildcards for any variations we support. CALLED(); // is this a proposal for our one output? if (output != mOutput.source) { fprintf(stderr, "_SoundPlayNode::FormatProposal returning B_MEDIA_BAD_SOURCE\n"); return B_MEDIA_BAD_SOURCE; } // we only support floating-point raw audio, so we always return that, but we // supply an error code depending on whether we found the proposal acceptable. media_type requestedType = format->type; *format = mFormat; if ((requestedType != B_MEDIA_UNKNOWN_TYPE) && (requestedType != B_MEDIA_RAW_AUDIO)) { fprintf(stderr, "_SoundPlayNode::FormatProposal returning B_MEDIA_BAD_FORMAT\n"); return B_MEDIA_BAD_FORMAT; } else return B_OK; // raw audio or wildcard type, either is okay by us } status_t _SoundPlayNode::FormatChangeRequested(const media_source& source, const media_destination& destination, media_format* io_format, int32* _deprecated_) { CALLED(); // we don't support any other formats, so we just reject any format changes. return B_ERROR; } status_t _SoundPlayNode::GetNextOutput(int32* cookie, media_output* out_output) { CALLED(); if ((*cookie < 1) && (*cookie >= 0)) { *out_output = mOutput; *cookie += 1; return B_OK; } else return B_BAD_INDEX; } status_t _SoundPlayNode::DisposeOutputCookie(int32 cookie) { CALLED(); // do nothing because we don't use the cookie for anything special return B_OK; } status_t _SoundPlayNode::SetBufferGroup(const media_source& for_source, BBufferGroup* newGroup) { CALLED(); // is this our output? if (for_source != mOutput.source) { fprintf(stderr, "_SoundPlayNode::SetBufferGroup returning B_MEDIA_BAD_SOURCE\n"); return B_MEDIA_BAD_SOURCE; } // Are we being passed the buffer group we're already using? if (newGroup == mBufferGroup) return B_OK; // Ahh, someone wants us to use a different buffer group. At this point we delete // the one we are using and use the specified one instead. If the specified group is // NULL, we need to recreate one ourselves, and use *that*. Note that if we're // caching a BBuffer that we requested earlier, we have to Recycle() that buffer // *before* deleting the buffer group, otherwise we'll deadlock waiting for that // buffer to be recycled! delete mBufferGroup; // waits for all buffers to recycle if (newGroup != NULL) { // we were given a valid group; just use that one from now on mBufferGroup = newGroup; } else { // we were passed a NULL group pointer; that means we construct // our own buffer group to use from now on size_t size = mOutput.format.u.raw_audio.buffer_size; int32 count = int32(mLatency / BufferDuration() + 1 + 1); if (count < 3) count = 3; mBufferGroup = new BBufferGroup(size, count); } return B_OK; } status_t _SoundPlayNode::GetLatency(bigtime_t* out_latency) { CALLED(); // report our *total* latency: internal plus downstream plus scheduling *out_latency = EventLatency() + SchedulingLatency(); return B_OK; } status_t _SoundPlayNode::PrepareToConnect(const media_source& what, const media_destination& where, media_format* format, media_source* out_source, char* out_name) { // PrepareToConnect() is the second stage of format negotiations that happens // inside BMediaRoster::Connect(). At this point, the consumer's AcceptFormat() // method has been called, and that node has potentially changed the proposed // format. It may also have left wildcards in the format. PrepareToConnect() // *must* fully specialize the format before returning! CALLED(); // is this our output? if (what != mOutput.source) { fprintf(stderr, "_SoundPlayNode::PrepareToConnect returning B_MEDIA_BAD_SOURCE\n"); return B_MEDIA_BAD_SOURCE; } // are we already connected? if (mOutput.destination != media_destination::null) return B_MEDIA_ALREADY_CONNECTED; // the format may not yet be fully specialized (the consumer might have // passed back some wildcards). Finish specializing it now, and return an // error if we don't support the requested format. if (format->type != B_MEDIA_RAW_AUDIO) { fprintf(stderr, "\tnon-raw-audio format?!\n"); return B_MEDIA_BAD_FORMAT; } // !!! validate all other fields except for buffer_size here, because the consumer might have // supplied different values from AcceptFormat()? // check the buffer size, which may still be wildcarded if (format->u.raw_audio.buffer_size == media_raw_audio_format::wildcard.buffer_size) { format->u.raw_audio.buffer_size = 2048; // pick something comfortable to suggest fprintf(stderr, "\tno buffer size provided, suggesting %lu\n", format->u.raw_audio.buffer_size); } else { fprintf(stderr, "\tconsumer suggested buffer_size %lu\n", format->u.raw_audio.buffer_size); } // Now reserve the connection, and return information about it mOutput.destination = where; mOutput.format = *format; *out_source = mOutput.source; strncpy(out_name, mOutput.name, B_MEDIA_NAME_LENGTH); return B_OK; } void _SoundPlayNode::Connect(status_t error, const media_source& source, const media_destination& destination, const media_format& format, char* io_name) { CALLED(); // is this our output? if (source != mOutput.source) { fprintf(stderr, "_SoundPlayNode::Connect returning\n"); return; } // If something earlier failed, Connect() might still be called, but with a non-zero // error code. When that happens we simply unreserve the connection and do // nothing else. if (error) { mOutput.destination = media_destination::null; mOutput.format = mFormat; return; } // Okay, the connection has been confirmed. Record the destination and format // that we agreed on, and report our connection name again. mOutput.destination = destination; mOutput.format = format; strncpy(io_name, mOutput.name, B_MEDIA_NAME_LENGTH); // Now that we're connected, we can determine our downstream latency. // Do so, then make sure we get our events early enough. media_node_id id; FindLatencyFor(mOutput.destination, &mLatency, &id); fprintf(stderr, "\tdownstream latency = %Ld\n", mLatency); mInternalLatency = 10000LL; fprintf(stderr, "\tbuffer-filling took %Ld usec on this machine\n", mInternalLatency); SetEventLatency(mLatency + mInternalLatency); // reset our buffer duration, etc. to avoid later calculations bigtime_t duration = mOutput.format.u.raw_audio.buffer_size * 10000 / ( (mOutput.format.u.raw_audio.format & media_raw_audio_format::B_AUDIO_SIZE_MASK) * mOutput.format.u.raw_audio.channel_count) / ((int32)(mOutput.format.u.raw_audio.frame_rate / 100)); //bigtime_t duration = bigtime_t(1000000) * samplesPerBuffer / bigtime_t(mOutput.format.u.raw_audio.frame_rate); SetBufferDuration(duration); // Set up the buffer group for our connection, as long as nobody handed us a // buffer group (via SetBufferGroup()) prior to this. That can happen, for example, // if the consumer calls SetOutputBuffersFor() on us from within its Connected() // method. if (!mBufferGroup) AllocateBuffers(); } void _SoundPlayNode::Disconnect(const media_source& what, const media_destination& where) { CALLED(); // is this our output? if (what != mOutput.source) { fprintf(stderr, "_SoundPlayNode::Disconnect returning\n"); return; } // Make sure that our connection is the one being disconnected if ((where == mOutput.destination) && (what == mOutput.source)) { mOutput.destination = media_destination::null; mOutput.format = mFormat; delete mBufferGroup; mBufferGroup = NULL; } else { fprintf(stderr, "\tDisconnect() called with wrong source/destination (%ld/%ld), ours is (%ld/%ld)\n", what.id, where.id, mOutput.source.id, mOutput.destination.id); } } void _SoundPlayNode::LateNoticeReceived(const media_source& what, bigtime_t how_much, bigtime_t performance_time) { CALLED(); printf("_SoundPlayNode::LateNoticeReceived, %Ld too late at %Ld\n", how_much, performance_time); // is this our output? if (what != mOutput.source) { fprintf(stderr, "_SoundPlayNode::LateNoticeReceived returning\n"); return; } if (RunMode() != B_DROP_DATA) { // We're late, and our run mode dictates that we try to produce buffers // earlier in order to catch up. This argues that the downstream nodes are // not properly reporting their latency, but there's not much we can do about // that at the moment, so we try to start producing buffers earlier to // compensate. mInternalLatency += how_much; if (mInternalLatency > 30000) // avoid getting a too high latency mInternalLatency = 30000; SetEventLatency(mLatency + mInternalLatency); fprintf(stderr, "\tincreasing latency to %Ld\n", mLatency + mInternalLatency); } else { // The other run modes dictate various strategies for sacrificing data quality // in the interests of timely data delivery. The way *we* do this is to skip // a buffer, which catches us up in time by one buffer duration. size_t nFrames = mOutput.format.u.raw_audio.buffer_size / ((mOutput.format.u.raw_audio.format & media_raw_audio_format::B_AUDIO_SIZE_MASK) * mOutput.format.u.raw_audio.channel_count); mFramesSent += nFrames; fprintf(stderr, "\tskipping a buffer to try to catch up\n"); } } void _SoundPlayNode::EnableOutput(const media_source& what, bool enabled, int32* _deprecated_) { CALLED(); // If I had more than one output, I'd have to walk my list of output records to see // which one matched the given source, and then enable/disable that one. But this // node only has one output, so I just make sure the given source matches, then set // the enable state accordingly. // is this our output? if (what != mOutput.source) { fprintf(stderr, "_SoundPlayNode::EnableOutput returning\n"); return; } mOutputEnabled = enabled; } void _SoundPlayNode::AdditionalBufferRequested(const media_source& source, media_buffer_id prev_buffer, bigtime_t prev_time, const media_seek_tag* prev_tag) { CALLED(); // we don't support offline mode return; } void _SoundPlayNode::LatencyChanged(const media_source& source, const media_destination& destination, bigtime_t new_latency, uint32 flags) { CALLED(); printf("_SoundPlayNode::LatencyChanged: new_latency %Ld\n", new_latency); // something downstream changed latency, so we need to start producing // buffers earlier (or later) than we were previously. Make sure that the // connection that changed is ours, and adjust to the new downstream // latency if so. if ((source == mOutput.source) && (destination == mOutput.destination)) { mLatency = new_latency; SetEventLatency(mLatency + mInternalLatency); } else { printf("_SoundPlayNode::LatencyChanged: ignored\n"); } } // -------------------------------------------------------- // // implementation for BMediaEventLooper // -------------------------------------------------------- // void _SoundPlayNode::HandleEvent( const media_timed_event *event, bigtime_t lateness, bool realTimeEvent = false) { CALLED(); switch (event->type) { case BTimedEventQueue::B_START: HandleStart(event,lateness,realTimeEvent); break; case BTimedEventQueue::B_SEEK: HandleSeek(event,lateness,realTimeEvent); break; case BTimedEventQueue::B_WARP: HandleWarp(event,lateness,realTimeEvent); break; case BTimedEventQueue::B_STOP: HandleStop(event,lateness,realTimeEvent); break; case BTimedEventQueue::B_HANDLE_BUFFER: // we don't get any buffers break; case SEND_NEW_BUFFER_EVENT: if (RunState() == BMediaEventLooper::B_STARTED) { SendNewBuffer(event, lateness, realTimeEvent); } break; case BTimedEventQueue::B_DATA_STATUS: HandleDataStatus(event,lateness,realTimeEvent); break; case BTimedEventQueue::B_PARAMETER: HandleParameter(event,lateness,realTimeEvent); break; default: fprintf(stderr," unknown event type: %li\n",event->type); break; } } // protected: // how should we handle late buffers? drop them? // notify the producer? status_t _SoundPlayNode::SendNewBuffer(const media_timed_event *event, bigtime_t lateness, bool realTimeEvent) { CALLED(); // printf("latency = %12Ld, event = %12Ld, sched = %5Ld, arrive at %12Ld, now %12Ld, current lateness %12Ld\n", EventLatency() + SchedulingLatency(), EventLatency(), SchedulingLatency(), event->event_time, TimeSource()->Now(), lateness); // make sure we're both started *and* connected before delivering a buffer if ((RunState() != BMediaEventLooper::B_STARTED) || (mOutput.destination == media_destination::null)) return B_OK; // The event->event_time is the time at which the buffer we are preparing here should // arrive at it's destination. The MediaEventLooper should have scheduled us early enough // (based on EventLatency() and the SchedulingLatency()) to make this possible. // lateness is independent of EventLatency()! if (lateness > 1000) { printf("_SoundPlayNode::SendNewBuffer, event scheduled much too late, lateness is %Ld\n", lateness); } // skip buffer creation if output not enabled if (mOutputEnabled) { // Get the next buffer of data BBuffer* buffer = FillNextBuffer(event->event_time); if (buffer) { // If we are ready way too early, decrase internal latency /* bigtime_t how_early = event->event_time - TimeSource()->Now() - mLatency - mInternalLatency; if (how_early > 5000) { printf("_SoundPlayNode::SendNewBuffer, event scheduled too early, how_early is %Ld\n", how_early); if (mTooEarlyCount++ == 5) { mInternalLatency -= how_early; if (mInternalLatency < 500) mInternalLatency = 500; printf("_SoundPlayNode::SendNewBuffer setting internal latency to %Ld\n", mInternalLatency); SetEventLatency(mLatency + mInternalLatency); mTooEarlyCount = 0; } } */ // send the buffer downstream if and only if output is enabled if (B_OK != SendBuffer(buffer, mOutput.destination)) { // we need to recycle the buffer // if the call to SendBuffer() fails printf("Buffer sending failed\n"); buffer->Recycle(); } } } // track how much media we've delivered so far size_t nFrames = mOutput.format.u.raw_audio.buffer_size / ((mOutput.format.u.raw_audio.format & media_raw_audio_format::B_AUDIO_SIZE_MASK) * mOutput.format.u.raw_audio.channel_count); mFramesSent += nFrames; // The buffer is on its way; now schedule the next one to go // nextEvent is the time at which the buffer should arrive at it's destination bigtime_t nextEvent = mStartTime + bigtime_t((1000000LL * mFramesSent) / mOutput.format.u.raw_audio.frame_rate); media_timed_event nextBufferEvent(nextEvent, SEND_NEW_BUFFER_EVENT); EventQueue()->AddEvent(nextBufferEvent); return B_OK; } status_t _SoundPlayNode::HandleDataStatus( const media_timed_event *event, bigtime_t lateness, bool realTimeEvent = false) { TRACE("_SoundPlayNode::HandleDataStatus status:%li, lateness:%li\n", event->data, lateness); switch(event->data) { case B_DATA_NOT_AVAILABLE: break; case B_DATA_AVAILABLE: break; case B_PRODUCER_STOPPED: break; default: break; } return B_OK; } status_t _SoundPlayNode::HandleStart( const media_timed_event *event, bigtime_t lateness, bool realTimeEvent = false) { CALLED(); // don't do anything if we're already running if (RunState() != B_STARTED) { // We want to start sending buffers now, so we set up the buffer-sending bookkeeping // and fire off the first "produce a buffer" event. mFramesSent = 0; mStartTime = event->event_time; media_timed_event firstBufferEvent(event->event_time, SEND_NEW_BUFFER_EVENT); // Alternatively, we could call HandleEvent() directly with this event, to avoid a trip through // the event queue, like this: // // this->HandleEvent(&firstBufferEvent, 0, false); // EventQueue()->AddEvent(firstBufferEvent); } return B_OK; } status_t _SoundPlayNode::HandleSeek( const media_timed_event *event, bigtime_t lateness, bool realTimeEvent = false) { CALLED(); DPRINTF("_SoundPlayNode::HandleSeek(t=%lld,d=%li,bd=%lld)\n",event->event_time,event->data,event->bigdata); return B_OK; } status_t _SoundPlayNode::HandleWarp( const media_timed_event *event, bigtime_t lateness, bool realTimeEvent = false) { CALLED(); return B_OK; } status_t _SoundPlayNode::HandleStop( const media_timed_event *event, bigtime_t lateness, bool realTimeEvent = false) { CALLED(); // flush the queue so downstreamers don't get any more EventQueue()->FlushEvents(0, BTimedEventQueue::B_ALWAYS, true, SEND_NEW_BUFFER_EVENT); return B_OK; } status_t _SoundPlayNode::HandleParameter( const media_timed_event *event, bigtime_t lateness, bool realTimeEvent = false) { CALLED(); return B_OK; } void _SoundPlayNode::AllocateBuffers() { CALLED(); // allocate enough buffers to span our downstream latency, plus one size_t size = mOutput.format.u.raw_audio.buffer_size; int32 count = int32(mLatency / BufferDuration() + 1 + 1); DPRINTF("\tlatency = %Ld, buffer duration = %Ld\n", mLatency, BufferDuration()); DPRINTF("\tcreating group of %ld buffers, size = %lu\n", count, size); if (count < 3) count = 3; mBufferGroup = new BBufferGroup(size, count); } BBuffer* _SoundPlayNode::FillNextBuffer(bigtime_t event_time) { CALLED(); // get a buffer from our buffer group BBuffer* buf = mBufferGroup->RequestBuffer(mOutput.format.u.raw_audio.buffer_size, BufferDuration() / 2); // if we fail to get a buffer (for example, if the request times out), we skip this // buffer and go on to the next, to avoid locking up the control thread if (!buf) { printf("RequestBuffer failed\n"); return NULL; } memset(buf->Data(), 0, mOutput.format.u.raw_audio.buffer_size); if(mPlayer->HasData()) { mPlayer->PlayBuffer(buf->Data(), mOutput.format.u.raw_audio.buffer_size, mOutput.format.u.raw_audio); } // fill in the buffer header media_header* hdr = buf->Header(); hdr->type = B_MEDIA_RAW_AUDIO; hdr->size_used = mOutput.format.u.raw_audio.buffer_size; hdr->time_source = TimeSource()->ID(); hdr->start_time = event_time; return buf; }