modules/webaudio/AudioBufferSourceNode.cpp

/*
 * Copyright (C) 2010, Google Inc. All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1.  Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2.  Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 *
 * THIS SOFTWARE IS PROVIDED BY APPLE INC. AND ITS CONTRIBUTORS ``AS IS'' AND ANY
 * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
 * DISCLAIMED. IN NO EVENT SHALL APPLE INC. OR ITS CONTRIBUTORS BE LIABLE FOR ANY
 * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
 * ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */

#include "config.h"

#if ENABLE(WEB_AUDIO)

#include "modules/webaudio/AudioBufferSourceNode.h"

#include "bindings/v8/ExceptionState.h"
#include "core/dom/ExceptionCode.h"
#include "core/page/PageConsole.h"
#include "platform/audio/AudioUtilities.h"
#include "modules/webaudio/AudioContext.h"
#include "modules/webaudio/AudioNodeOutput.h"
#include "platform/FloatConversion.h"
#include "wtf/MainThread.h"
#include "wtf/MathExtras.h"
#include <algorithm>

using namespace std;

namespace WebCore {

const double DefaultGrainDuration = 0.020; // 20ms

// Arbitrary upper limit on playback rate.
// Higher than expected rates can be useful when playing back oversampled buffers
// to minimize linear interpolation aliasing.
const double MaxRate = 1024;

PassRefPtr<AudioBufferSourceNode> AudioBufferSourceNode::create(AudioContext* context, float sampleRate)
{
    return adoptRef(new AudioBufferSourceNode(context, sampleRate));
}

AudioBufferSourceNode::AudioBufferSourceNode(AudioContext* context, float sampleRate)
    : AudioScheduledSourceNode(context, sampleRate)
    , m_buffer(0)
    , m_isLooping(false)
    , m_loopStart(0)
    , m_loopEnd(0)
    , m_virtualReadIndex(0)
    , m_isGrain(false)
    , m_grainOffset(0.0)
    , m_grainDuration(DefaultGrainDuration)
    , m_lastGain(1.0)
    , m_pannerNode(0)
{
    ScriptWrappable::init(this);
    setNodeType(NodeTypeAudioBufferSource);

    m_gain = AudioParam::create(context, "gain", 1.0, 0.0, 1.0);
    m_playbackRate = AudioParam::create(context, "playbackRate", 1.0, 0.0, MaxRate);

    // Default to mono.  A call to setBuffer() will set the number of output channels to that of the buffer.
    addOutput(adoptPtr(new AudioNodeOutput(this, 1)));

    initialize();
}

AudioBufferSourceNode::~AudioBufferSourceNode()
{
    clearPannerNode();
    uninitialize();
}

void AudioBufferSourceNode::process(size_t framesToProcess)
{
    AudioBus* outputBus = output(0)->bus();

    if (!isInitialized()) {
        outputBus->zero();
        return;
    }

    // The audio thread can't block on this lock, so we call tryLock() instead.
    MutexTryLocker tryLocker(m_processLock);
    if (tryLocker.locked()) {
        if (!buffer()) {
            outputBus->zero();
            return;
        }

        // After calling setBuffer() with a buffer having a different number of channels, there can in rare cases be a slight delay
        // before the output bus is updated to the new number of channels because of use of tryLocks() in the context's updating system.
        // In this case, if the the buffer has just been changed and we're not quite ready yet, then just output silence.
        if (numberOfChannels() != buffer()->numberOfChannels()) {
            outputBus->zero();
            return;
        }

        size_t quantumFrameOffset;
        size_t bufferFramesToProcess;

        updateSchedulingInfo(framesToProcess,
                             outputBus,
                             quantumFrameOffset,
                             bufferFramesToProcess);

        if (!bufferFramesToProcess) {
            outputBus->zero();
            return;
        }

        for (unsigned i = 0; i < outputBus->numberOfChannels(); ++i)
            m_destinationChannels[i] = outputBus->channel(i)->mutableData();

        // Render by reading directly from the buffer.
        if (!renderFromBuffer(outputBus, quantumFrameOffset, bufferFramesToProcess)) {
            outputBus->zero();
            return;
        }

        // Apply the gain (in-place) to the output bus.
        float totalGain = gain()->value() * m_buffer->gain();
        outputBus->copyWithGainFrom(*outputBus, &m_lastGain, totalGain);
        outputBus->clearSilentFlag();
    } else {
        // Too bad - the tryLock() failed.  We must be in the middle of changing buffers and were already outputting silence anyway.
        outputBus->zero();
    }
}

// Returns true if we're finished.
bool AudioBufferSourceNode::renderSilenceAndFinishIfNotLooping(AudioBus*, unsigned index, size_t framesToProcess)
{
    if (!loop()) {
        // If we're not looping, then stop playing when we get to the end.

        if (framesToProcess > 0) {
            // We're not looping and we've reached the end of the sample data, but we still need to provide more output,
            // so generate silence for the remaining.
            for (unsigned i = 0; i < numberOfChannels(); ++i)
                memset(m_destinationChannels[i] + index, 0, sizeof(float) * framesToProcess);
        }

        finish();
        return true;
    }
    return false;
}

bool AudioBufferSourceNode::renderFromBuffer(AudioBus* bus, unsigned destinationFrameOffset, size_t numberOfFrames)
{
    ASSERT(context()->isAudioThread());

    // Basic sanity checking
    ASSERT(bus);
    ASSERT(buffer());
    if (!bus || !buffer())
        return false;

    unsigned numberOfChannels = this->numberOfChannels();
    unsigned busNumberOfChannels = bus->numberOfChannels();

    bool channelCountGood = numberOfChannels && numberOfChannels == busNumberOfChannels;
    ASSERT(channelCountGood);
    if (!channelCountGood)
        return false;

    // Sanity check destinationFrameOffset, numberOfFrames.
    size_t destinationLength = bus->length();

    bool isLengthGood = destinationLength <= 4096 && numberOfFrames <= 4096;
    ASSERT(isLengthGood);
    if (!isLengthGood)
        return false;

    bool isOffsetGood = destinationFrameOffset <= destinationLength && destinationFrameOffset + numberOfFrames <= destinationLength;
    ASSERT(isOffsetGood);
    if (!isOffsetGood)
        return false;

    // Potentially zero out initial frames leading up to the offset.
    if (destinationFrameOffset) {
        for (unsigned i = 0; i < numberOfChannels; ++i)
            memset(m_destinationChannels[i], 0, sizeof(float) * destinationFrameOffset);
    }

    // Offset the pointers to the correct offset frame.
    unsigned writeIndex = destinationFrameOffset;

    size_t bufferLength = buffer()->length();
    double bufferSampleRate = buffer()->sampleRate();

    // Avoid converting from time to sample-frames twice by computing
    // the grain end time first before computing the sample frame.
    unsigned endFrame = m_isGrain ? AudioUtilities::timeToSampleFrame(m_grainOffset + m_grainDuration, bufferSampleRate) : bufferLength;

    // This is a HACK to allow for HRTF tail-time - avoids glitch at end.
    // FIXME: implement tailTime for each AudioNode for a more general solution to this problem.
    // https://bugs.webkit.org/show_bug.cgi?id=77224
    if (m_isGrain)
        endFrame += 512;

    // Do some sanity checking.
    if (endFrame > bufferLength)
        endFrame = bufferLength;
    if (m_virtualReadIndex >= endFrame)
        m_virtualReadIndex = 0; // reset to start

    // If the .loop attribute is true, then values of m_loopStart == 0 && m_loopEnd == 0 implies
    // that we should use the entire buffer as the loop, otherwise use the loop values in m_loopStart and m_loopEnd.
    double virtualEndFrame = endFrame;
    double virtualDeltaFrames = endFrame;

    if (loop() && (m_loopStart || m_loopEnd) && m_loopStart >= 0 && m_loopEnd > 0 && m_loopStart < m_loopEnd) {
        // Convert from seconds to sample-frames.
        double loopStartFrame = m_loopStart * buffer()->sampleRate();
        double loopEndFrame = m_loopEnd * buffer()->sampleRate();

        virtualEndFrame = min(loopEndFrame, virtualEndFrame);
        virtualDeltaFrames = virtualEndFrame - loopStartFrame;
    }


    double pitchRate = totalPitchRate();

    // Sanity check that our playback rate isn't larger than the loop size.
    if (pitchRate >= virtualDeltaFrames)
        return false;

    // Get local copy.
    double virtualReadIndex = m_virtualReadIndex;

    // Render loop - reading from the source buffer to the destination using linear interpolation.
    int framesToProcess = numberOfFrames;

    const float** sourceChannels = m_sourceChannels.get();
    float** destinationChannels = m_destinationChannels.get();

    // Optimize for the very common case of playing back with pitchRate == 1.
    // We can avoid the linear interpolation.
    if (pitchRate == 1 && virtualReadIndex == floor(virtualReadIndex)
        && virtualDeltaFrames == floor(virtualDeltaFrames)
        && virtualEndFrame == floor(virtualEndFrame)) {
        unsigned readIndex = static_cast<unsigned>(virtualReadIndex);
        unsigned deltaFrames = static_cast<unsigned>(virtualDeltaFrames);
        endFrame = static_cast<unsigned>(virtualEndFrame);
        while (framesToProcess > 0) {
            int framesToEnd = endFrame - readIndex;
            int framesThisTime = min(framesToProcess, framesToEnd);
            framesThisTime = max(0, framesThisTime);

            for (unsigned i = 0; i < numberOfChannels; ++i)
                memcpy(destinationChannels[i] + writeIndex, sourceChannels[i] + readIndex, sizeof(float) * framesThisTime);

            writeIndex += framesThisTime;
            readIndex += framesThisTime;
            framesToProcess -= framesThisTime;

            // Wrap-around.
            if (readIndex >= endFrame) {
                readIndex -= deltaFrames;
                if (renderSilenceAndFinishIfNotLooping(bus, writeIndex, framesToProcess))
                    break;
            }
        }
        virtualReadIndex = readIndex;
    } else {
        while (framesToProcess--) {
            unsigned readIndex = static_cast<unsigned>(virtualReadIndex);
            double interpolationFactor = virtualReadIndex - readIndex;

            // For linear interpolation we need the next sample-frame too.
            unsigned readIndex2 = readIndex + 1;
            if (readIndex2 >= bufferLength) {
                if (loop()) {
                    // Make sure to wrap around at the end of the buffer.
                    readIndex2 = static_cast<unsigned>(virtualReadIndex + 1 - virtualDeltaFrames);
                } else
                    readIndex2 = readIndex;
            }

            // Final sanity check on buffer access.
            // FIXME: as an optimization, try to get rid of this inner-loop check and put assertions and guards before the loop.
            if (readIndex >= bufferLength || readIndex2 >= bufferLength)
                break;

            // Linear interpolation.
            for (unsigned i = 0; i < numberOfChannels; ++i) {
                float* destination = destinationChannels[i];
                const float* source = sourceChannels[i];

                double sample1 = source[readIndex];
                double sample2 = source[readIndex2];
                double sample = (1.0 - interpolationFactor) * sample1 + interpolationFactor * sample2;

                destination[writeIndex] = narrowPrecisionToFloat(sample);
            }
            writeIndex++;

            virtualReadIndex += pitchRate;

            // Wrap-around, retaining sub-sample position since virtualReadIndex is floating-point.
            if (virtualReadIndex >= virtualEndFrame) {
                virtualReadIndex -= virtualDeltaFrames;
                if (renderSilenceAndFinishIfNotLooping(bus, writeIndex, framesToProcess))
                    break;
            }
        }
    }

    bus->clearSilentFlag();

    m_virtualReadIndex = virtualReadIndex;

    return true;
}


void AudioBufferSourceNode::reset()
{
    m_virtualReadIndex = 0;
    m_lastGain = gain()->value();
}

void AudioBufferSourceNode::setBuffer(AudioBuffer* buffer, ExceptionState& exceptionState)
{
    ASSERT(isMainThread());
    // FIXME: It does not look like we should throw if the buffer is null as
    // the attribute is nullable in the specification.
    if (!buffer) {
        exceptionState.throwTypeError("buffer cannot be null");
        return;
    }

    // The context must be locked since changing the buffer can re-configure the number of channels that are output.
    AudioContext::AutoLocker contextLocker(context());

    // This synchronizes with process().
    MutexLocker processLocker(m_processLock);

    if (buffer) {
        // Do any necesssary re-configuration to the buffer's number of channels.
        unsigned numberOfChannels = buffer->numberOfChannels();

        if (numberOfChannels > AudioContext::maxNumberOfChannels()) {
            exceptionState.throwTypeError("number of input channels (" + String::number(numberOfChannels)
                + ") exceeds maximum ("
                + String::number(AudioContext::maxNumberOfChannels()) + ").");
            return;
        }

        output(0)->setNumberOfChannels(numberOfChannels);

        m_sourceChannels = adoptArrayPtr(new const float* [numberOfChannels]);
        m_destinationChannels = adoptArrayPtr(new float* [numberOfChannels]);

        for (unsigned i = 0; i < numberOfChannels; ++i)
            m_sourceChannels[i] = buffer->getChannelData(i)->data();
    }

    m_virtualReadIndex = 0;
    m_buffer = buffer;
}

unsigned AudioBufferSourceNode::numberOfChannels()
{
    return output(0)->numberOfChannels();
}

void AudioBufferSourceNode::start(ExceptionState& exceptionState)
{
    startPlaying(false, 0, 0, buffer() ? buffer()->duration() : 0, exceptionState);
}

void AudioBufferSourceNode::start(double when, ExceptionState& exceptionState)
{
    startPlaying(false, when, 0, buffer() ? buffer()->duration() : 0, exceptionState);
}

void AudioBufferSourceNode::start(double when, double grainOffset, ExceptionState& exceptionState)
{
    startPlaying(true, when, grainOffset, buffer() ? buffer()->duration() : 0, exceptionState);
}

void AudioBufferSourceNode::start(double when, double grainOffset, double grainDuration, ExceptionState& exceptionState)
{
    startPlaying(true, when, grainOffset, grainDuration, exceptionState);
}

void AudioBufferSourceNode::startPlaying(bool isGrain, double when, double grainOffset, double grainDuration, ExceptionState& exceptionState)
{
    ASSERT(isMainThread());

    if (m_playbackState != UNSCHEDULED_STATE) {
        exceptionState.throwDOMException(
            InvalidStateError,
            "cannot call start more than once.");
        return;
    }

    if (!buffer())
        return;

    if (isGrain) {
        // Do sanity checking of grain parameters versus buffer size.
        double bufferDuration = buffer()->duration();

        grainOffset = max(0.0, grainOffset);
        grainOffset = min(bufferDuration, grainOffset);
        m_grainOffset = grainOffset;

        double maxDuration = bufferDuration - grainOffset;

        grainDuration = max(0.0, grainDuration);
        grainDuration = min(maxDuration, grainDuration);
        m_grainDuration = grainDuration;
    }

    m_isGrain = isGrain;
    m_startTime = when;

    // We call timeToSampleFrame here since at playbackRate == 1 we don't want to go through linear interpolation
    // at a sub-sample position since it will degrade the quality.
    // When aligned to the sample-frame the playback will be identical to the PCM data stored in the buffer.
    // Since playbackRate == 1 is very common, it's worth considering quality.
    m_virtualReadIndex = AudioUtilities::timeToSampleFrame(m_grainOffset, buffer()->sampleRate());

    m_playbackState = SCHEDULED_STATE;
}

void AudioBufferSourceNode::noteGrainOn(double when, double grainOffset, double grainDuration, ExceptionState& exceptionState)
{
    // Handle unspecified duration where 0 means the rest of the buffer.
    if (!grainDuration && buffer())
        grainDuration = buffer()->duration();
    startPlaying(true, when, grainOffset, grainDuration, exceptionState);
}

double AudioBufferSourceNode::totalPitchRate()
{
    double dopplerRate = 1.0;
    if (m_pannerNode)
        dopplerRate = m_pannerNode->dopplerRate();

    // Incorporate buffer's sample-rate versus AudioContext's sample-rate.
    // Normally it's not an issue because buffers are loaded at the AudioContext's sample-rate, but we can handle it in any case.
    double sampleRateFactor = 1.0;
    if (buffer())
        sampleRateFactor = buffer()->sampleRate() / sampleRate();

    double basePitchRate = playbackRate()->value();

    double totalRate = dopplerRate * sampleRateFactor * basePitchRate;

    // Sanity check the total rate.  It's very important that the resampler not get any bad rate values.
    totalRate = max(0.0, totalRate);
    if (!totalRate)
        totalRate = 1; // zero rate is considered illegal
    totalRate = min(MaxRate, totalRate);

    bool isTotalRateValid = !std::isnan(totalRate) && !std::isinf(totalRate);
    ASSERT(isTotalRateValid);
    if (!isTotalRateValid)
        totalRate = 1.0;

    return totalRate;
}

bool AudioBufferSourceNode::propagatesSilence() const
{
    return !isPlayingOrScheduled() || hasFinished() || !m_buffer;
}

void AudioBufferSourceNode::setPannerNode(PannerNode* pannerNode)
{
    if (m_pannerNode != pannerNode && !hasFinished()) {
        if (pannerNode)
            pannerNode->ref(AudioNode::RefTypeConnection);
        if (m_pannerNode)
            m_pannerNode->deref(AudioNode::RefTypeConnection);

        m_pannerNode = pannerNode;
    }
}

void AudioBufferSourceNode::clearPannerNode()
{
    if (m_pannerNode) {
        m_pannerNode->deref(AudioNode::RefTypeConnection);
        m_pannerNode = 0;
    }
}

void AudioBufferSourceNode::finish()
{
    clearPannerNode();
    ASSERT(!m_pannerNode);
    AudioScheduledSourceNode::finish();
}

} // namespace WebCore

#endif // ENABLE(WEB_AUDIO)