Donate to e Foundation | Murena handsets with /e/OS | Own a part of Murena! Learn more

Commit be02f32c authored by John Grossman's avatar John Grossman
Browse files

TimedAudioTrack: Optimize the queue trim operation. 



DO NOT MERGE
this change must be hand-merged into master due to directory
restructuring.

Don't perform the end PTS calculation for each buffer during trimming.
Instead, only calculate the ending PTS of a buffer if there is no next
buffer in the queue.  This optimization assumes that the buffers being
queued are in monotonic media time order (a fair assumption for now)
and that the timestamps in the audio are contiguous (not a requirement
for this API, but a reality of how it is being used right now).

In the case where the audio is discontinuous on purpose, it is
that this optimization will cause the system hold one extra buffer
which it could have safely trimmed.  It should not be much of an issue
since in real life the audio is almost always contiguous, and as long
as the media clock is running and the mixer is mixing, the buffer will
be used up and discard as part of the normal flow anyway.

Change-Id: I00061e85ee7d5651fcf80751646c7d7415894a14
Signed-off-by: default avatarJohn Grossman <johngro@google.com>
parent f609b661

services/audioflinger/AudioFlinger.cpp

+29 −11
Original line number Diff line number Diff line
@@ -3936,20 +3936,38 @@ void AudioFlinger::PlaybackThread::TimedTrack::trimTimedBufferQueue_l() { 


    size_t trimEnd;

    for (trimEnd = 0; trimEnd < mTimedBufferQueue.size(); trimEnd++) {

        int64_t bufEnd;



        if ((trimEnd + 1) < mTimedBufferQueue.size()) {

            // We have a next buffer.  Just use its PTS as the PTS of the frame

            // following the last frame in this buffer.  If the stream is sparse

            // (ie, there are deliberate gaps left in the stream which should be

            // filled with silence by the TimedAudioTrack), then this can result

            // in one extra buffer being left un-trimmed when it could have

            // been.  In general, this is not typical, and we would rather

            // optimized away the TS calculation below for the more common case

            // where PTSes are contiguous.

            bufEnd = mTimedBufferQueue[trimEnd + 1].pts();

        } else {

            // We have no next buffer.  Compute the PTS of the frame following

            // the last frame in this buffer by computing the duration of of

            // this frame in media time units and adding it to the PTS of the

            // buffer.

            int64_t frameCount = mTimedBufferQueue[trimEnd].buffer()->size()

                               / mCblk->frameSize;

        int64_t bufEnd;



            if (!mMediaTimeToSampleTransform.doReverseTransform(frameCount,

                                                                &bufEnd)) {

            LOGE("Failed to convert frame count of %lld to media time duration"

                 " (scale factor %d/%u) in %s", frameCount,

                LOGE("Failed to convert frame count of %lld to media time"

                     " duration" " (scale factor %d/%u) in %s",

                     frameCount,

                     mMediaTimeToSampleTransform.a_to_b_numer,

                     mMediaTimeToSampleTransform.a_to_b_denom,

                     __PRETTY_FUNCTION__);

                break;

            }

            bufEnd += mTimedBufferQueue[trimEnd].pts();

        }



        if (bufEnd > mediaTimeNow)

            break;