Mercurial > hg > svapp
view audioio/AudioCallbackPlaySource.h @ 35:37e3c693af0c
* Make a single base Thread class for RT and non-RT threads
* Pull ResizeableBitmap out from the MatrixFile's ColumnBitmap
* Reorder SpectrogramLayer::paint somewhat so as to improve cache hit ratio
in the FFT file cache
author | Chris Cannam |
---|---|
date | Mon, 08 May 2006 16:44:47 +0000 |
parents | 4593a7ebad93 |
children | 4ed2448582cc |
line wrap: on
line source
/* -*- c-basic-offset: 4 indent-tabs-mode: nil -*- vi:set ts=8 sts=4 sw=4: */ /* Sonic Visualiser An audio file viewer and annotation editor. Centre for Digital Music, Queen Mary, University of London. This file copyright 2006 Chris Cannam. 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. See the file COPYING included with this distribution for more information. */ #ifndef _AUDIO_CALLBACK_PLAY_SOURCE_H_ #define _AUDIO_CALLBACK_PLAY_SOURCE_H_ #include "base/RingBuffer.h" #include "base/AudioPlaySource.h" #include "base/Scavenger.h" #include <QObject> #include <QMutex> #include <QWaitCondition> #include "base/Thread.h" #include <samplerate.h> #include <set> #include <map> class Model; class ViewManager; class AudioGenerator; class PlayParameters; class IntegerTimeStretcher; /** * AudioCallbackPlaySource manages audio data supply to callback-based * audio APIs such as JACK or CoreAudio. It maintains one ring buffer * per channel, filled during playback by a non-realtime thread, and * provides a method for a realtime thread to pick up the latest * available sample data from these buffers. */ class AudioCallbackPlaySource : public virtual QObject, public AudioPlaySource { Q_OBJECT public: AudioCallbackPlaySource(ViewManager *); virtual ~AudioCallbackPlaySource(); /** * Add a data model to be played from. The source can mix * playback from a number of sources including dense and sparse * models. The models must match in sample rate, but they don't * have to have identical numbers of channels. */ virtual void addModel(Model *model); /** * Remove a model. */ virtual void removeModel(Model *model); /** * Remove all models. (Silence will ensue.) */ virtual void clearModels(); /** * Start making data available in the ring buffers for playback, * from the given frame. If playback is already under way, reseek * to the given frame and continue. */ virtual void play(size_t startFrame); /** * Stop playback and ensure that no more data is returned. */ virtual void stop(); /** * Return whether playback is currently supposed to be happening. */ virtual bool isPlaying() const { return m_playing; } /** * Return the frame number that is currently expected to be coming * out of the speakers. (i.e. compensating for playback latency.) */ virtual size_t getCurrentPlayingFrame(); /** * Set the block size of the target audio device. This should * be called by the target class. */ void setTargetBlockSize(size_t); /** * Get the block size of the target audio device. */ size_t getTargetBlockSize() const; /** * Set the playback latency of the target audio device, in frames * at the target sample rate. This is the difference between the * frame currently "leaving the speakers" and the last frame (or * highest last frame across all channels) requested via * getSamples(). The default is zero. */ void setTargetPlayLatency(size_t); /** * Get the playback latency of the target audio device. */ size_t getTargetPlayLatency() const; /** * Specify that the target audio device has a fixed sample rate * (i.e. cannot accommodate arbitrary sample rates based on the * source). If the target sets this to something other than the * source sample rate, this class will resample automatically to * fit. */ void setTargetSampleRate(size_t); /** * Return the sample rate set by the target audio device (or the * source sample rate if the target hasn't set one). */ virtual size_t getTargetSampleRate() const; /** * Set the current output levels for metering (for call from the * target) */ void setOutputLevels(float left, float right); /** * Return the current (or thereabouts) output levels in the range * 0.0 -> 1.0, for metering purposes. */ virtual bool getOutputLevels(float &left, float &right); /** * Get the number of channels of audio that in the source models. * This may safely be called from a realtime thread. Returns 0 if * there is no source yet available. */ size_t getSourceChannelCount() const; /** * Get the number of channels of audio that will be provided * to the play target. This may be more than the source channel * count: for example, a mono source will provide 2 channels * after pan. * This may safely be called from a realtime thread. Returns 0 if * there is no source yet available. */ size_t getTargetChannelCount() const; /** * Get the actual sample rate of the source material. This may * safely be called from a realtime thread. Returns 0 if there is * no source yet available. */ size_t getSourceSampleRate() const; /** * Get "count" samples (at the target sample rate) of the mixed * audio data, in all channels. This may safely be called from a * realtime thread. */ size_t getSourceSamples(size_t count, float **buffer); void setSlowdownFactor(size_t factor); signals: void modelReplaced(); void playStatusChanged(bool isPlaying); void sampleRateMismatch(size_t requested, size_t available, bool willResample); protected slots: void selectionChanged(); void playLoopModeChanged(); void playSelectionModeChanged(); void playParametersChanged(PlayParameters *); protected: ViewManager *m_viewManager; AudioGenerator *m_audioGenerator; class RingBufferVector : public std::vector<RingBuffer<float> *> { public: virtual ~RingBufferVector() { while (!empty()) { delete *begin(); erase(begin()); } } }; std::set<Model *> m_models; RingBufferVector *m_readBuffers; RingBufferVector *m_writeBuffers; size_t m_readBufferFill; size_t m_writeBufferFill; Scavenger<RingBufferVector> m_bufferScavenger; size_t m_sourceChannelCount; size_t m_blockSize; size_t m_sourceSampleRate; size_t m_targetSampleRate; size_t m_playLatency; bool m_playing; bool m_exiting; size_t m_lastModelEndFrame; static const size_t m_ringBufferSize; float m_outputLeft; float m_outputRight; RingBuffer<float> *getWriteRingBuffer(size_t c) { if (m_writeBuffers && c < m_writeBuffers->size()) { return (*m_writeBuffers)[c]; } else { return 0; } } RingBuffer<float> *getReadRingBuffer(size_t c) { RingBufferVector *rb = m_readBuffers; if (rb && c < rb->size()) { return (*rb)[c]; } else { return 0; } } void clearRingBuffers(bool haveLock = false, size_t count = 0); void unifyRingBuffers(); class TimeStretcherData { public: TimeStretcherData(size_t channels, size_t factor, size_t blockSize); ~TimeStretcherData(); size_t getFactor() const { return m_factor; } IntegerTimeStretcher *getStretcher(size_t channel); double *getOutputBuffer(size_t channel); double *getInputBuffer(); void run(size_t channel); protected: TimeStretcherData(const TimeStretcherData &); // not provided TimeStretcherData &operator=(const TimeStretcherData &); // not provided typedef std::pair<IntegerTimeStretcher *, double *> StretcherBuffer; std::map<size_t, StretcherBuffer> m_stretcher; double *m_stretchInputBuffer; size_t m_factor; size_t m_blockSize; }; size_t m_slowdownCounter; TimeStretcherData *m_timeStretcher; Scavenger<TimeStretcherData> m_timeStretcherScavenger; // Called from fill thread, m_playing true, mutex held // Return true if work done bool fillBuffers(); // Called from fillBuffers. Return the number of frames written, // which will be count or fewer. Return in the frame argument the // new buffered frame position (which may be earlier than the // frame argument passed in, in the case of looping). size_t mixModels(size_t &frame, size_t count, float **buffers); class AudioCallbackPlaySourceFillThread : public Thread { public: AudioCallbackPlaySourceFillThread(AudioCallbackPlaySource &source) : Thread(Thread::NonRTThread), m_source(source) { } virtual void run(); protected: AudioCallbackPlaySource &m_source; }; QMutex m_mutex; QWaitCondition m_condition; AudioCallbackPlaySourceFillThread *m_fillThread; SRC_STATE *m_converter; }; #endif