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/PipeReader.cpp b/services/audioflinger/PipeReader.cpp
new file mode 100644
index 0000000..43bcb42
--- /dev/null
+++ b/services/audioflinger/PipeReader.cpp
@@ -0,0 +1,95 @@
+/*
+ * 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.
+ */
+
+#define LOG_TAG "PipeReader"
+//#define LOG_NDEBUG 0
+
+#include <cutils/compiler.h>
+#include <utils/Log.h>
+#include "PipeReader.h"
+
+namespace android {
+
+PipeReader::PipeReader(Pipe& pipe) :
+ NBAIO_Source(pipe.mFormat),
+ mPipe(pipe),
+ // any data already in the pipe is not visible to this PipeReader
+ mFront(android_atomic_acquire_load(&pipe.mRear)),
+ mFramesOverrun(0),
+ mOverruns(0)
+{
+ android_atomic_inc(&pipe.mReaders);
+}
+
+PipeReader::~PipeReader()
+{
+ int32_t readers = android_atomic_dec(&mPipe.mReaders);
+ ALOG_ASSERT(readers > 0);
+}
+
+ssize_t PipeReader::availableToRead()
+{
+ if (CC_UNLIKELY(!mNegotiated)) {
+ return NEGOTIATE;
+ }
+ int32_t rear = android_atomic_acquire_load(&mPipe.mRear);
+ // read() is not multi-thread safe w.r.t. itself, so no mutex or atomic op needed to read mFront
+ size_t avail = rear - mFront;
+ if (CC_UNLIKELY(avail > mPipe.mMaxFrames)) {
+ // Discard all but 3/4 of the most recent data in pipe to avoid another overrun immediately
+ int32_t oldFront = mFront;
+ mFront = rear - mPipe.mMaxFrames + (mPipe.mMaxFrames >> 2);
+ mFramesOverrun += (size_t) (mFront - oldFront);
+ ++mOverruns;
+ return OVERRUN;
+ }
+ return avail;
+}
+
+ssize_t PipeReader::read(void *buffer, size_t count)
+{
+ ssize_t avail = availableToRead();
+ if (CC_UNLIKELY(avail <= 0)) {
+ return avail;
+ }
+ // An overrun can occur from here on and be silently ignored,
+ // but it will be caught at next read()
+ if (CC_LIKELY(count > (size_t) avail)) {
+ count = avail;
+ }
+ size_t front = mFront & (mPipe.mMaxFrames - 1);
+ size_t red = mPipe.mMaxFrames - front;
+ if (CC_LIKELY(red > count)) {
+ red = count;
+ }
+ // In particular, an overrun during the memcpy will result in reading corrupt data
+ memcpy(buffer, (char *) mPipe.mBuffer + (front << mBitShift), red << mBitShift);
+ // We could re-read the rear pointer here to detect the corruption, but why bother?
+ if (CC_UNLIKELY(front + red == mPipe.mMaxFrames)) {
+ if (CC_UNLIKELY((count -= red) > front)) {
+ count = front;
+ }
+ if (CC_LIKELY(count > 0)) {
+ memcpy((char *) buffer + (red << mBitShift), mPipe.mBuffer, count << mBitShift);
+ red += count;
+ }
+ }
+ mFront += red;
+ mFramesRead += red;
+ return red;
+}
+
+} // namespace android