Non-blocking audio I/O interface, WIP
Yet another abstraction similar to AudioTrack::Buffer and AudioBufferProvider,
but with support for streaming, non-blocking, and eventually PTS.
This is intended to be used as follows:
- primary HAL output stream implements a Sink
- primary HAL input stream implements a Source
- Pipe implements a Sink
- PipeReader implements a Source or TimedSource (not shown yet),
which supports "read at PTS"
- fast AudioTrack on server side will implement a Source using cblk
- normal AudioTrack on server side will not be changed initially
- fast AudioRecord on server side will implement a Sink using cblk
- normal AudioRecord on server side will not be changed initially
- fast mixer thread will read from Sources and write to a Sink,
or (unlikely) implement a Source and multiple Sinks
- Visualization and PCM logger will read from Source or TimedSource
- A2DP normal mixer will be connected directly to its output stream
and there will be a kind of OutputTrack for duplication that will
read from a Sink with non-blocking write fed by the fast mixer.
Patch set 3 changes:
- Add more implementations of NBAIO interfaces:
added SourceAudioBufferProvider, MonoPipe, MonoPipeReader.
- Added Format_sampleRate and Format_channelCount.
- Extract out the roundUp() method.
- Respond to most comments from previous code review.
- The new classes are untested.
Patch set 4 changes:
- Fix bugs in MonoPipe::write() and MonoPipeReader::read()
- Fix bug initializing mFrameBitShift too early
- renamed roundUp() to roundup()
- Fix Android.mk
- Add LOG_TAG an LOG_NDEBUG, use ALOG_ASSERT and utils/Log.h instead of assert
- Fix build warnings
- Move constructor and destructor bodies from .h to .cpp
- Line length 100
- Following naming conventions for #include double-include protector macros
- Include what you use
- More NBAIO logging
- MonoPipe write can be blocking
Patch set 5 changes:
- Address code review comments
- Use a static library so unused implementations don't take memory
- Comment out libsndfile dependency
- Remove debugging LOGV and LOG_NDEBUG
Patch set 6 changes (would be 6 at old location, actually 2 at new location):
- Address code review comments on patchset 5
- For MonoPipe, allow the full pipe to be used, no need to omit one slot
- Don't do atomic releasing stores unless needed
Still to do:
- I'm not happy with the Pipe class names
- Update build/ for new static library?
Change-Id: Ie6c61f05ce06b676b033be448a8ef9025a2ffcfd
diff --git a/services/audioflinger/MonoPipe.h b/services/audioflinger/MonoPipe.h
new file mode 100644
index 0000000..45e6bb4
--- /dev/null
+++ b/services/audioflinger/MonoPipe.h
@@ -0,0 +1,75 @@
+/*
+ * Copyright (C) 2012 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef ANDROID_AUDIO_MONO_PIPE_H
+#define ANDROID_AUDIO_MONO_PIPE_H
+
+#include <time.h>
+#include "NBAIO.h"
+
+namespace android {
+
+// MonoPipe is similar to Pipe except:
+// - supports only a single reader, called MonoPipeReader
+// - write() cannot overrun; instead it will return a short actual count if insufficient space
+// - write() can optionally block if the pipe is full
+// Like Pipe, it is not multi-thread safe for either writer or reader
+// but writer and reader can be different threads.
+class MonoPipe : public NBAIO_Sink {
+
+ friend class MonoPipeReader;
+
+public:
+ // maxFrames will be rounded up to a power of 2, and all slots are available. Must be >= 2.
+ // Note: whatever shares this object with another thread needs to do so in an SMP-safe way (like
+ // creating it the object before creating the other thread, or storing the object with a
+ // release_store). Otherwise the other thread could see a partially-constructed object.
+ MonoPipe(size_t maxFrames, NBAIO_Format format, bool writeCanBlock = false);
+ virtual ~MonoPipe();
+
+ // NBAIO_Port interface
+
+ //virtual ssize_t negotiate(const NBAIO_Format offers[], size_t numOffers,
+ // NBAIO_Format counterOffers[], size_t& numCounterOffers);
+ //virtual NBAIO_Format format() const;
+
+ // NBAIO_Sink interface
+
+ //virtual size_t framesWritten() const;
+ //virtual size_t framesUnderrun() const;
+ //virtual size_t underruns() const;
+
+ virtual ssize_t availableToWrite() const;
+ virtual ssize_t write(const void *buffer, size_t count);
+ //virtual ssize_t writeVia(writeVia_t via, size_t total, void *user, size_t block);
+
+private:
+ const size_t mMaxFrames; // always a power of 2
+ void * const mBuffer;
+ // mFront and mRear will never be separated by more than mMaxFrames.
+ // 32-bit overflow is possible if the pipe is active for a long time, but if that happens it's
+ // safe because we "&" with (mMaxFrames-1) at end of computations to calculate a buffer index.
+ volatile int32_t mFront; // written by reader with android_atomic_release_store,
+ // read by writer with android_atomic_acquire_load
+ volatile int32_t mRear; // written by writer with android_atomic_release_store,
+ // read by reader with android_atomic_acquire_load
+ const bool mWriteCanBlock; // whether write() should block if the pipe is full
+ struct timespec mSleep; // time to sleep if blocking is enabled and the pipe is full
+};
+
+} // namespace android
+
+#endif // ANDROID_AUDIO_MONO_PIPE_H