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

1 files changed, 95 insertions, 0 deletions

diff --git a/services/audioflinger/PipeReader.cpp b/services/audioflinger/PipeReader.cpp
new file mode 100644
index 00000000..43bcb42f
--- /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