History log of /frameworks/av/services/audioflinger/FastMixer.cpp
Revision Date Author Comments (<<< Hide modified files) (Show modified files >>>)
80b3273cea8660fe8a5868d024d2788a1e083ffc 24-Sep-2012 Glenn Kasten <gkasten@google.com> Check clock_gettime(CLOCK_MONOTONIC) for failure

Bug: 7100774
Change-Id: I15a84a19bb6d6ef1d9dac4beaa03587638196404
fe3156ec6fd9fa57dde913fd8567530d095a6550 11-Sep-2012 Jean-Michel Trivi <jmtrivi@google.com> Communicate audio session ID to downmixer

The audio downmixer effect might need the audio session Id, pass it
from the track creation in AudioFlinger to the downmix effect
creation in AudioMixer.

Change-Id: I5e29540542ae89cf4a0cdb537b3e67f04442a20a
1ab212cf5cfa5b88c801840e11e3191cd10f48e4 07-Sep-2012 Glenn Kasten <gkasten@google.com> Add outlier statistics for fast mixer cycle times

Change-Id: I31c964caeb8b5d9ae0a426224f030cdcb01114a0
0d27c65ddb5c968baa6db0c26e80f5c451bc52bc 07-Aug-2012 Glenn Kasten <gkasten@google.com> Tune the overrun correction

The new overrun correction is: if the previous write() cycle time is <
50% of expected, then sleep enough extra to make the total cycle time =
95% of expected. This should help compensate for HAL implementations
that have uneven cycle times due to sample rate conversion.

Bug: 6881638
Change-Id: I5ab58afdfceee7ac523177f021fbf62d743d571a
2c3b2da3049627264b7c6b449a1622f002210f03 03-Aug-2012 John Grossman <johngro@google.com> AudioFlinger: fix timed audio

(cherry picked from commit e20ac92c564a2f4e8123885807abdf0a78de0dd7)

> AudioFlinger: fix timed audio
> Addresses Bug 6900517.
> Finish up support for timed audio in the new FastMixer world. Pay special
> attention to remaining lock-less and voluntary yield free on the FastMixer
> thread. This fixes audio playback for Q on JB.
> Change-Id: Iaf815e58a1b1d0a0190051794bec8dc5c9231785
> Signed-off-by: John Grossman <johngro@google.com>

Change-Id: I9bd687acc345a05867af48e71116690fdb0ce1b5
Signed-off-by: John Grossman <johngro@google.com>
eb15716b59020f342df62bce5b293f0603b94861 13-Jun-2012 Glenn Kasten <gkasten@google.com> Change definition of warmup period

Previously, warmup was considered done as soon as any write() took
more than 0.5 nominal cycle time. In practice, this was always the
first write() that turned on power to the output path, and it didn't
accurately account for filling the full kernel buffer queue, or for
buffering in the HAL sample rate conversion.

Now warmup is considered done when a write() _after_ the first write
takes more than 0.5 nominal cycle time.

This will throttle the initial pull rate after coming out of standby.
When combined with another change to throttle the pull rate for
devices with HAL sample rate conversion, it may help reduce some of the
notification glitches. The only downside is that it will increase the
warmup time a bit.

Bug: 6508317, 6650846, 6607056
Change-Id: I39f324c5195578170a55308e9601d3a1b41db3e0
972af221ae7253e406e0e1ea853e56a3010ed6b1 14-Jun-2012 Glenn Kasten <gkasten@google.com> FastMixer compensates for SRC jitter in audio HAL

Force a sleep if audio HAL write() returns too quickly.

Bug: 6661275
Change-Id: Ie810f5a087b89fb24dab229a0e500af1dee53571
0a14c4ce1a41bc09eb7855fa531a3af629a69139 13-Jun-2012 Glenn Kasten <gkasten@google.com> Make CPU frequency statistics optional

Certain CPUs with dynamic cluster swapping and hotplug
don't report CPU frequency accurately. The file descriptors
used to read the frequency become stale and report bogus data.
So make this feature a build time option for debugging only.
This will also improve performance of the fast mixer loop.

Change-Id: I602f81ec3281a37992769208be08084ed1469e8c
868c0abe9f26852c217bc201b337e72c68f829b8 13-Jun-2012 Glenn Kasten <gkasten@google.com> Shorten dumpsys media.audio_flinger output

Don't include in FastMixer if it's not present.
This removes confusing clutter especially for devices
with a separate deep buffer output thread, or for the
duplicating thread and A2DP output thread.

Change-Id: I84933f5555593256f11ba0895ec915c09cf16963
9a0d82df8fd681257bc5b29fb5e6c65fcc7cef5a 06-Jun-2012 Eric Laurent <elaurent@google.com> audioflinger: modify fast mixer overrun management

Do not force sleep when an overrun occurs. This was for non
blocking audio HAL and is unused.
Keep statistics on overruns.

Bug 6605215.

Change-Id: I295455d4562201aa3814c88bdd6e52c9226712af
1295bb4dcff7b29c75cd23746816df12a871d72c 31-May-2012 Glenn Kasten <gkasten@google.com> Fast track dumpsys

Bug: 6591648
Change-Id: I696f51c682e7233ba690d97da26012084989b412
04a4ca4217e9b30342d5865024c340013d7184e8 01-Jun-2012 Glenn Kasten <gkasten@google.com> Was calculating measured warmup time wrong

Bug: 6592923
Change-Id: I980c6fc69d78c569147ea44ce33a7f8a41739c91
fbae5dae5187aca9d974cbe15ec818e9c6f56705 21-May-2012 Glenn Kasten <gkasten@google.com> Keep a copy of most recent audio played

Change-Id: I6b2f97881c39998a2fae9ab79d669af6c0a37e94
99c99d00beb43b939dedc9ffb07adb89f6a85ba5 15-May-2012 Glenn Kasten <gkasten@google.com> systrace for audio

Trace fast track buffer fill status for underruns etc.

Move the definition of macro to Android.mk.

No overhead if disabled.

Change-Id: If0e83e21b61b059ca38f543f8a6ffb58e08c79ee
c059bd4246c1b3944965be921e5b334d51cd236c 15-May-2012 Glenn Kasten <gkasten@google.com> Skip bad kHz values, and display them accurately

One device reports zero CPU kHz occasionally, so skip those values.

Preserve accuracy of the low-order 4 bits of the the CPU frequency.

Change-Id: I7d934f697412277462464fca82d112d0306a214e
09474df67278c0cd621b57c4aef1deaec4d8447f 10-May-2012 Glenn Kasten <gkasten@google.com> Improve underrun handling for fast tracks

Maintain more accurate accounting of type of underrun.
Automatically remove track from active list after a series of "empty" underruns.

Change-Id: If042bf80e1790dcaaf195c99dc9c0ed9b55382c1
e58ccce45598bcf4b4874b0e87cd1eb8d05ba9a9 12-May-2012 Glenn Kasten <gkasten@google.com> Fix 2 bugs in fast mixer statistics

Was losing the upper half of the bounds, and MHz was off.

Change-Id: I61c98027b35ce7c3578ea6f3acf05aa5c48a5831
d8e6fd35ec2b59ee7d873daf1f1d9d348221c7bc 07-May-2012 Glenn Kasten <gkasten@google.com> Use audio tag for system tracing

Disabled by default; uncomment ATRACE_TAG to enable

Change-Id: I99af894022a859ee5644bd853cfd8a48e4735ff9
42d45cfd0c3d62357a6549c62f535e4d4fe08d91 02-May-2012 Glenn Kasten <gkasten@google.com> Update fast mixer statistics

Compute statistics on fast mixer elapsed time and CPU load per cycle using a
simple moving average rather than by fixed blocks. This has a couple advantages:
- remove burstiness of CPU usage due to former floating-point calculations in fast mixer
- gives us flexibility in how to report (e.g. could report over just the last 1 second)

Disadvantage is increased RAM, and since the samples are not updated
atomically, it is possible to have an error in the statistics. This
should not be much of an issue due to the relatively large number of samples.

Also add CPU raw ns and adjusted MHz statistics.

Change-Id: Iaa2cd13f18250c3162aff40409b3694b769d9505
288ed2103d96f3aabd7e6bea3c080ab6db164049 26-Apr-2012 Glenn Kasten <gkasten@google.com> Fix race condition for non-started fast tracks

This required re-implementing how fast tracks are considered active.
Now, they use the same logic as normal tracks, except underrun is ignored.

Other changes:
- add framesReady() to AudioBufferProvider interface
- rebased
- add track underrun counter state to fast mixer dump state
- move dumpsys header to Track::appendDumpHeader()
so it closer to where tracks are dumped
- display track state in dumpsys as a character code
- measure and display warmup time and cycles in dumpsys
- copy in the presentation complete code
- add ExtendedAudioBufferProvider for framesReady() which returns size_t
- simplify underrun tracking
- deferred reset track after stop()
- add comments

Change-Id: I7db8821bc565230ec76da1f9380fe3fb09735e5b
21e8c50bd13ebe44f3088e26c9c6df0e163c469c 12-Apr-2012 Glenn Kasten <gkasten@google.com> FastMixer update

- Add support for mono fast tracks
- Add support for optional sample rate conversion on fast tracks
- Log sample rate and frame count
- Enable statistics

Change-Id: Ife014edf4f452da361f3eaaae19209ef6ff6958b
9bd23229fdec1657398abc682ccccfce1c95f8aa 16-Apr-2012 Jean-Michel Trivi <jmtrivi@google.com> Fix multichannel downmix pause bug on video player

Pausing a video player will cause the track audio mixer to be
disabled, which causes the downmixer to be deleted. When reenabled,
the track channel mask hasn't changed but the downmixer is there
Fixed by:
- instanciating a downmixer when the AudioMixer track
gets initialized (in getTrackName(), now taking a channel mask
as input), and deleted when in deleteTrackName().
- when the channel changes on a track, check whether it
needs a downmixer or not. Preparing a track for downmix
automatically removes the old downmixer if there was one.

Also: initialize the track downmixerBufferProvider field
when AudioMixer is instanciated, so we can safely call
delete on it in AudioMixer's destructor, in case
deleteTrackName() wasn't called before the mixer was

Change-Id: I589b0781cda5b3c82f85b561c52b08546cac21f8
97b5d0d5b5ef766eb5dd680d05a5d199662d4ae0 24-Mar-2012 Glenn Kasten <gkasten@google.com> Fast mixer

Change-Id: I61552f83507e08e4c706076b9fb15362869e6265