scummvm/audio/audiostream.cpp
2023-12-24 13:19:25 +01:00

517 lines
16 KiB
C++

/* ScummVM - Graphic Adventure Engine
*
* ScummVM is the legal property of its developers, whose names
* are too numerous to list here. Please refer to the COPYRIGHT
* file distributed with this source distribution.
*
* 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 3 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, see <http://www.gnu.org/licenses/>.
*
*/
#include "common/debug.h"
#include "common/file.h"
#include "common/mutex.h"
#include "common/textconsole.h"
#include "common/queue.h"
#include "common/util.h"
#include "audio/audiostream.h"
#include "audio/decoders/flac.h"
#include "audio/decoders/mp3.h"
#include "audio/decoders/quicktime.h"
#include "audio/decoders/raw.h"
#include "audio/decoders/vorbis.h"
#include "audio/decoders/wave.h"
#include "audio/mixer.h"
namespace Audio {
struct StreamFileFormat {
/** Decodername */
const char *decoderName;
const char *fileExtension;
/**
* Pointer to a function which tries to open a file of type StreamFormat.
* Return NULL in case of an error (invalid/nonexisting file).
*/
SeekableAudioStream *(*openStreamFile)(Common::SeekableReadStream *stream, DisposeAfterUse::Flag disposeAfterUse);
};
static const StreamFileFormat STREAM_FILEFORMATS[] = {
/* decoderName, fileExt, openStreamFunction */
#ifdef USE_FLAC
{ "FLAC", ".flac", makeFLACStream },
{ "FLAC", ".fla", makeFLACStream },
#endif
#ifdef USE_VORBIS
{ "Ogg Vorbis", ".ogg", makeVorbisStream },
#endif
#ifdef USE_MAD
{ "MPEG Layer 3", ".mp3", makeMP3Stream },
#endif
{ "MPEG-4 Audio", ".m4a", makeQuickTimeStream },
{ "WAV", ".wav", makeWAVStream },
};
SeekableAudioStream *SeekableAudioStream::openStreamFile(const Common::Path &basename) {
SeekableAudioStream *stream = nullptr;
Common::File *fileHandle = new Common::File();
for (int i = 0; i < ARRAYSIZE(STREAM_FILEFORMATS); ++i) {
Common::Path filename = basename.append(STREAM_FILEFORMATS[i].fileExtension);
fileHandle->open(filename);
if (fileHandle->isOpen()) {
// Create the stream object
stream = STREAM_FILEFORMATS[i].openStreamFile(fileHandle, DisposeAfterUse::YES);
fileHandle = nullptr;
break;
}
}
delete fileHandle;
if (stream == nullptr)
debug(1, "SeekableAudioStream::openStreamFile: Could not open compressed AudioFile %s", basename.toString(Common::Path::kNativeSeparator).c_str());
return stream;
}
#pragma mark -
#pragma mark --- LoopingAudioStream ---
#pragma mark -
LoopingAudioStream::LoopingAudioStream(Common::DisposablePtr<RewindableAudioStream>&& stream, uint loops, bool rewind)
: _parent(Common::move(stream)), _loops(loops), _completeIterations(0) {
assert(_parent);
if (rewind && !_parent->rewind()) {
// TODO: Properly indicate error
_loops = _completeIterations = 1;
}
if (_parent->endOfStream()) {
// Apparently this is an empty stream
_loops = _completeIterations = 1;
}
}
LoopingAudioStream::LoopingAudioStream(RewindableAudioStream *stream, uint loops, DisposeAfterUse::Flag disposeAfterUse, bool rewind)
: LoopingAudioStream(Common::move(Common::DisposablePtr<RewindableAudioStream>(stream, disposeAfterUse)), loops, rewind) {}
int LoopingAudioStream::readBuffer(int16 *buffer, const int numSamples) {
if ((_loops && _completeIterations == _loops) || !numSamples)
return 0;
int samplesRead = _parent->readBuffer(buffer, numSamples);
if (_parent->endOfStream()) {
++_completeIterations;
if (_completeIterations == _loops)
return samplesRead;
const int remainingSamples = numSamples - samplesRead;
if (!_parent->rewind()) {
// TODO: Properly indicate error
_loops = _completeIterations;
return samplesRead;
}
if (_parent->endOfStream()) {
// Apparently this is an empty stream
_loops = _completeIterations;
}
return samplesRead + readBuffer(buffer + samplesRead, remainingSamples);
}
return samplesRead;
}
bool LoopingAudioStream::endOfData() const {
return (_loops != 0 && _completeIterations == _loops) || _parent->endOfData();
}
bool LoopingAudioStream::endOfStream() const {
return _loops != 0 && _completeIterations == _loops;
}
AudioStream *makeLoopingAudioStream(RewindableAudioStream *stream, uint loops) {
if (loops != 1)
return new LoopingAudioStream(stream, loops);
else
return stream;
}
AudioStream *makeLoopingAudioStream(SeekableAudioStream *stream, Timestamp start, Timestamp end, uint loops) {
if (!start.totalNumberOfFrames() && (!end.totalNumberOfFrames() || end == stream->getLength())) {
return makeLoopingAudioStream(stream, loops);
} else {
if (!end.totalNumberOfFrames())
end = stream->getLength();
if (start >= end) {
warning("makeLoopingAudioStream: start (%d) >= end (%d)", start.msecs(), end.msecs());
delete stream;
return nullptr;
}
return makeLoopingAudioStream(new SubSeekableAudioStream(stream, start, end), loops);
}
}
#pragma mark -
#pragma mark --- SubLoopingAudioStream ---
#pragma mark -
SubLoopingAudioStream::SubLoopingAudioStream(SeekableAudioStream *stream,
uint loops,
const Timestamp loopStart,
const Timestamp loopEnd,
DisposeAfterUse::Flag disposeAfterUse)
: _parent(stream, disposeAfterUse), _loops(loops), _completeIterations(0),
_pos(0, getRate() * (isStereo() ? 2 : 1)),
_loopStart(convertTimeToStreamPos(loopStart, getRate(), isStereo())),
_loopEnd(convertTimeToStreamPos(loopEnd, getRate(), isStereo())) {
assert(loopStart < loopEnd);
assert(stream);
if (!_parent->rewind())
_loops = _completeIterations = 1;
}
int SubLoopingAudioStream::readBuffer(int16 *buffer, const int numSamples) {
if ((_loops && _completeIterations == _loops) || !numSamples)
return 0;
int framesLeft = MIN(_loopEnd.frameDiff(_pos), numSamples);
int framesRead = _parent->readBuffer(buffer, framesLeft);
_pos = _pos.addFrames(framesRead);
if (framesRead < framesLeft && _parent->endOfStream()) {
// TODO: Proper error indication.
if (!_completeIterations)
_completeIterations = 1;
_loops = _completeIterations;
return framesRead;
} else if (_pos == _loopEnd) {
++_completeIterations;
if (_completeIterations == _loops)
return framesRead;
if (!_parent->seek(_loopStart)) {
// TODO: Proper error indication.
_loops = _completeIterations;
return framesRead;
}
_pos = _loopStart;
framesLeft = numSamples - framesLeft;
return framesRead + readBuffer(buffer + framesRead, framesLeft);
} else {
return framesRead;
}
}
bool SubLoopingAudioStream::endOfData() const {
// We're out of data if this stream is finished or the parent
// has run out of data for now.
return (_loops != 0 && _completeIterations == _loops) || _parent->endOfData();
}
bool SubLoopingAudioStream::endOfStream() const {
// The end of the stream has been reached only when we've gone
// through all the iterations.
return _loops != 0 && _completeIterations == _loops;
}
#pragma mark -
#pragma mark --- SubSeekableAudioStream ---
#pragma mark -
SubSeekableAudioStream::SubSeekableAudioStream(SeekableAudioStream *parent, const Timestamp start, const Timestamp end, DisposeAfterUse::Flag disposeAfterUse)
: _parent(parent, disposeAfterUse),
_start(convertTimeToStreamPos(start, getRate(), isStereo())),
_pos(0, getRate() * (isStereo() ? 2 : 1)),
_length(convertTimeToStreamPos(end, getRate(), isStereo()) - _start) {
assert(_length.totalNumberOfFrames() % (isStereo() ? 2 : 1) == 0);
_parent->seek(_start);
}
int SubSeekableAudioStream::readBuffer(int16 *buffer, const int numSamples) {
int framesLeft = MIN(_length.frameDiff(_pos), numSamples);
int framesRead = _parent->readBuffer(buffer, framesLeft);
_pos = _pos.addFrames(framesRead);
return framesRead;
}
bool SubSeekableAudioStream::seek(const Timestamp &where) {
_pos = convertTimeToStreamPos(where, getRate(), isStereo());
if (_pos > _length) {
_pos = _length;
return false;
}
if (_parent->seek(_pos + _start)) {
return true;
} else {
_pos = _length;
return false;
}
}
#pragma mark -
#pragma mark --- Queueing audio stream ---
#pragma mark -
void QueuingAudioStream::queueBuffer(byte *data, uint32 size, DisposeAfterUse::Flag disposeAfterUse, byte flags) {
AudioStream *stream = makeRawStream(data, size, getRate(), flags, disposeAfterUse);
queueAudioStream(stream, DisposeAfterUse::YES);
}
class QueuingAudioStreamImpl : public QueuingAudioStream {
private:
/**
* We queue a number of (pointers to) audio stream objects.
* In addition, we need to remember for each stream whether
* to dispose it after all data has been read from it.
* Hence, we don't store pointers to stream objects directly,
* but rather StreamHolder structs.
*/
struct StreamHolder {
AudioStream *_stream;
DisposeAfterUse::Flag _disposeAfterUse;
StreamHolder(AudioStream *stream, DisposeAfterUse::Flag disposeAfterUse)
: _stream(stream),
_disposeAfterUse(disposeAfterUse) {}
};
/**
* The sampling rate of this audio stream.
*/
const int _rate;
/**
* Whether this audio stream is mono (=false) or stereo (=true).
*/
const int _stereo;
/**
* This flag is set by the finish() method only. See there for more details.
*/
bool _finished;
/**
* A mutex to avoid access problems (causing e.g. corruption of
* the linked list) in thread aware environments.
*/
Common::Mutex _mutex;
/**
* The queue of audio streams.
*/
Common::Queue<StreamHolder> _queue;
public:
QueuingAudioStreamImpl(int rate, bool stereo)
: _rate(rate), _stereo(stereo), _finished(false) {}
~QueuingAudioStreamImpl();
// Implement the AudioStream API
int readBuffer(int16 *buffer, const int numSamples) override;
bool isStereo() const override { return _stereo; }
int getRate() const override { return _rate; }
bool endOfData() const override {
Common::StackLock lock(_mutex);
return _queue.empty() || _queue.front()._stream->endOfData();
}
bool endOfStream() const override {
Common::StackLock lock(_mutex);
return _finished && _queue.empty();
}
// Implement the QueuingAudioStream API
void queueAudioStream(AudioStream *stream, DisposeAfterUse::Flag disposeAfterUse) override;
void finish() override {
Common::StackLock lock(_mutex);
_finished = true;
}
uint32 numQueuedStreams() const override {
Common::StackLock lock(_mutex);
return _queue.size();
}
};
QueuingAudioStreamImpl::~QueuingAudioStreamImpl() {
while (!_queue.empty()) {
StreamHolder tmp = _queue.pop();
if (tmp._disposeAfterUse == DisposeAfterUse::YES)
delete tmp._stream;
}
}
void QueuingAudioStreamImpl::queueAudioStream(AudioStream *stream, DisposeAfterUse::Flag disposeAfterUse) {
assert(!_finished);
if ((stream->getRate() != getRate()) || (stream->isStereo() != isStereo()))
error("QueuingAudioStreamImpl::queueAudioStream: stream has mismatched parameters");
Common::StackLock lock(_mutex);
_queue.push(StreamHolder(stream, disposeAfterUse));
}
int QueuingAudioStreamImpl::readBuffer(int16 *buffer, const int numSamples) {
Common::StackLock lock(_mutex);
int samplesDecoded = 0;
while (samplesDecoded < numSamples && !_queue.empty()) {
AudioStream *stream = _queue.front()._stream;
samplesDecoded += stream->readBuffer(buffer + samplesDecoded, numSamples - samplesDecoded);
// Done with the stream completely
if (stream->endOfStream()) {
StreamHolder tmp = _queue.pop();
if (tmp._disposeAfterUse == DisposeAfterUse::YES)
delete stream;
continue;
}
// Done with data but not the stream, bail out
if (stream->endOfData())
break;
}
return samplesDecoded;
}
QueuingAudioStream *makeQueuingAudioStream(int rate, bool stereo) {
return new QueuingAudioStreamImpl(rate, stereo);
}
Timestamp convertTimeToStreamPos(const Timestamp &where, int rate, bool isStereo) {
Timestamp result(where.convertToFramerate(rate * (isStereo ? 2 : 1)));
// When the Stream is a stereo stream, we have to assure
// that the sample position is an even number.
if (isStereo && (result.totalNumberOfFrames() & 1))
result = result.addFrames(-1); // We cut off one sample here.
// Since Timestamp allows sub-frame-precision it might lead to odd behaviors
// when we would just return result.
//
// An example is when converting the timestamp 500ms to a 11025 Hz based
// stream. It would have an internal frame counter of 5512.5. Now when
// doing calculations at frame precision, this might lead to unexpected
// results: The frame difference between a timestamp 1000ms and the above
// mentioned timestamp (both with 11025 as framerate) would be 5512,
// instead of 5513, which is what a frame-precision based code would expect.
//
// By creating a new Timestamp with the given parameters, we create a
// Timestamp with frame-precision, which just drops a sub-frame-precision
// information (i.e. rounds down).
return Timestamp(result.secs(), result.numberOfFrames(), result.framerate());
}
/**
* An AudioStream wrapper that cuts off the amount of samples read after a
* given time length is reached.
*/
class LimitingAudioStream : public AudioStream {
public:
LimitingAudioStream(AudioStream *parentStream, const Audio::Timestamp &length, DisposeAfterUse::Flag disposeAfterUse) :
_parentStream(parentStream), _samplesRead(0), _disposeAfterUse(disposeAfterUse),
_totalSamples(length.convertToFramerate(getRate()).totalNumberOfFrames() * getChannels()) {}
~LimitingAudioStream() {
if (_disposeAfterUse == DisposeAfterUse::YES)
delete _parentStream;
}
int readBuffer(int16 *buffer, const int numSamples) override {
// Cap us off so we don't read past _totalSamples
int samplesRead = _parentStream->readBuffer(buffer, MIN<int>(numSamples, _totalSamples - _samplesRead));
_samplesRead += samplesRead;
return samplesRead;
}
bool endOfData() const override { return _parentStream->endOfData() || reachedLimit(); }
bool endOfStream() const override { return _parentStream->endOfStream() || reachedLimit(); }
bool isStereo() const override { return _parentStream->isStereo(); }
int getRate() const override { return _parentStream->getRate(); }
private:
int getChannels() const { return isStereo() ? 2 : 1; }
bool reachedLimit() const { return _samplesRead >= _totalSamples; }
AudioStream *_parentStream;
DisposeAfterUse::Flag _disposeAfterUse;
uint32 _totalSamples, _samplesRead;
};
AudioStream *makeLimitingAudioStream(AudioStream *parentStream, const Timestamp &length, DisposeAfterUse::Flag disposeAfterUse) {
return new LimitingAudioStream(parentStream, length, disposeAfterUse);
}
/**
* An AudioStream that plays nothing and immediately returns that
* the endOfStream() has been reached
*/
class NullAudioStream : public AudioStream {
public:
bool isStereo() const override { return false; }
int getRate() const override;
int readBuffer(int16 *data, const int numSamples) override { return 0; }
bool endOfData() const override { return true; }
};
int NullAudioStream::getRate() const {
return g_system->getMixer()->getOutputRate();
}
AudioStream *makeNullAudioStream() {
return new NullAudioStream();
}
/**
* An AudioStream that just returns silent samples and runs infinitely.
*/
class SilentAudioStream : public AudioStream {
public:
SilentAudioStream(int rate, bool stereo) : _rate(rate), _isStereo(stereo) {}
int readBuffer(int16 *buffer, const int numSamples) override {
memset(buffer, 0, numSamples * 2);
return numSamples;
}
bool endOfData() const override { return false; } // it never ends!
bool isStereo() const override { return _isStereo; }
int getRate() const override { return _rate; }
private:
int _rate;
bool _isStereo;
};
AudioStream *makeSilentAudioStream(int rate, bool stereo) {
return new SilentAudioStream(rate, stereo);
}
} // End of namespace Audio