`HlsExtractorFactory` instances are mutable, so storing one in a static
field is not safe, and can lead to state accidentally/surprisingly being
shared between different player instances.
PiperOrigin-RevId: 685701062
This ensures these values are still updated even if the delegate
`DataSource.open()` throws an exception (e.g. an HTTP 404).
PiperOrigin-RevId: 685687810
The `DefaultLoadControl` implementation of onTracksSelected only utilizes the `Renderer[]` parameter for use in stream type, of which it can collect from the `ExoTrackSelection[]` parameter.
PiperOrigin-RevId: 685677726
The `DefaultPreloadManager.Builder` is able to build the `DefaultPreloadManager` and `ExoPlayer` instances with the consistently shared configurations. Apps can:
* Simply setup the `DefaultPreloadManager` and `ExoPlayer` with all default configurations via `build()` and `buildExoPlayer()`;
* Or customize the shared configurations by the setters on `DefaultPreloadManager.Builder` and setup via `build()` and `buildExoPlayer()`;
* Or customize the player-only configurations for `ExoPlayer` via `buildExoPlayer(ExoPlayer.Builder)`.
PiperOrigin-RevId: 684852808
This CL is prework for implementing
`RandomParameterizedSpeedChangingAudioProcessorTest`, which will build
on logic present in `RandomParameterizedSonicTest`.
This is a non-functional change.
PiperOrigin-RevId: 684838017
Inconsistent rounding modes between `currentTimeUs` and
`bytesUntilNextSpeedChange` would cause `SpeedChangingAudioProcessor`
to miss calling `queueEndOfStream()` on `SonicAudioProcessor` on a speed
change, and thus the final output samples of that `SonicAudioProcessor`
"configuration" would be missed.
This change is also a partial revert of 971486f5f9, which fixed a hang
of `SpeedChangingAudioProcessor`, but introduced the dropped output
frames issue fixed in this CL (see b/372203420). To avoid reintroducing
the hang, we are now ignoring any mid-sample speed changes and will only
apply speed changes that are effective at a whole sample position.
PiperOrigin-RevId: 684824218
Removed `Android.mk` and `Application.mk`, allowing `CMake` to run directly from the build.gradle file. Users no longer need to check out `NDK` or depend on it, simplifying the usage of the Flac extension.
Also fixed a copy-pasted comment in `CMakeLists.txt` of Opus and IAMF.
PiperOrigin-RevId: 684769561
Removed `Android.mk` and `Application.mk`, allowing `CMake` to run directly from the build.gradle file. Users no longer need to check out `NDK` or depend on it, simplifying the usage of the Opus extension.
PiperOrigin-RevId: 684489927
We need to rebuild any native components of the app to prevent crashes on devices with 16 KB page support.
Tested on a device that supports 16 KB pages and runs Android 15, as well as on older Android devices.
Issue: androidx/media#1685
PiperOrigin-RevId: 684488244
Removed `Android.mk` and `Application.mk`, allowing `CMake` to run directly from the `build.gradle` file. Users no longer need to check out `NDK` or depend on it, simplifying the usage of the IAMF extension.
PiperOrigin-RevId: 684471874
The version and flags are stored in a single integer,
with the version in the higher 8 bits and the flags in
the lower 24 bits. The version should be 1 and the
flags should be 0.
Surprisingly the incorrect value was ignored by many
players and hence the bug was never caught.
With the bug, the video does not play on
`Samsung Galaxy S22 Ultra` and works well
after fixing the bug.
PiperOrigin-RevId: 684433371
Minor improvement to allow an Media3 browser to pass extras
when connecting the initial browser in `MediaControllerImplLegacy`.
Before this change an empty bundle was sent. After this change
the connection hints of the `Media3 browser is used as root hints
of the initial browser that connects when the Media3 browser is
built in `MediaBrowser.buildAsync`.
#cherrypick
PiperOrigin-RevId: 684372552
This helps avoid overflows in intermediate calculations.
Verified the value in the test using `BigInteger`:
```
jshell> BigInteger.valueOf(1L << 52).multiply(BigInteger.valueOf(90000)).divide(BigInteger.valueOf(1000000))
$3 ==> 405323966463344
```
Issue: androidx/media#1763
#cherrypick
PiperOrigin-RevId: 684028178
The implementation of these methods was updated from direct java integer
arithmetic in 885ddb167e.
In this change, `RoundingMode.FLOOR` was used to try and maintain
compatibility with java integer division. This was incorrect, because
java integer division uses `DOWN` (i.e. towards zero), rather than
`FLOOR` (i.e. towards negative infinity) semantics.
This change fixes the compatibility.
The dump file changes in this CL relate to tests that exercise edit
list behaviour. This involves manipulating negative timestamps, which
explains why they are impacted by this change.
PiperOrigin-RevId: 684013175
- Updated `DefaultTrackSelector.SpatializerWrapperV32.canBeSpatialized` to handle IAMF format.
- Modified `isAudioFormatWithinAudioChannelCountConstraints` to check for `NO_VALUE` of `channelCount` to improve readability.
Note: `DefaultTrackSelector.SpatializerWrapperV32.canBeSpatialized` is not triggered for the IAMF format due to the unset channel count (`Format.NO_VALUE`). The update ensures completeness.
#cherrypick
PiperOrigin-RevId: 684003980
Object-based audio is more efficient and flexible than channel-based audio, supporting a broader range of devices. This update makes `DefaultTrackSelector` prefer object-based audio when other factors are equal, ensuring its use whenever possible.
#cherrypick
PiperOrigin-RevId: 683990051
When calling `MediaController.getCommandButtonForMediaItem(MediaItem)`
command buttons with custom commands that are not available
shouldn't be advertised to the controller when connected to
a Media3 session.
In contrast, when connected to a legacy session, available commands
are not enforced when advertising commands. Similarly, when sending
a custom commands that is referenced by a command button for media
items, sending is permitted without the command being available.
This is required because available commands match to custom actions
in `PlaybackStateCompat` of the legacy session. Adding commands for
media items to custom action of the `PlaybackStateCompat` would
interfere with other use cases.
Issue: androidx/media#1474
#cherrypick
PiperOrigin-RevId: 683717723
This was missed in da724c8cc4
I tried to write a test for this, but got stuck crafting valid test
data. I was able to create a new fragmented MP4 file containing only a
TTML track:
```shell
$ MP4Box -add simple.ttml -frag 2000 sample_fragmented_ttml.mp4
```
Then I tried naively removing the `ftyp` and `moov` boxes with a hex
editor, but using this in `FragmentedMp4ExtractorNoSniffingTest` gave
me an `EOFException` that I didn't get to the root cause of.
Issue: androidx/media#1779
#cherrypick
PiperOrigin-RevId: 683667850
This refactors the `supportsFormat` method on the MediaCodecVideoRenderer to be a static method that can be called from external components to determine if a video format is supported by the device renderers. Since `context` is the only component that is part of the MCVR, it can easily be obtained externally and passed to the method.
This can help optimize format support decisions ahead of time of a player having been instantiated, such as removing unsupported representations from a dash manifest.
This could also be done for the MCAR but would also require passing in the AudioSink and isn't required at this time.
This change is a no-op, because
`SubtitleTranscodingExtractorOutput.seekMap` forwards directly to the
delegate implementation, but it seems clearer to always use the
wrapper.
Also remove a no-op assignment in `MatroskaExtractor`.
This is a follow-up to Issue: androidx/media#1779 where I manually checked every
implementation of `Extractor.init` for a similar mistake.
#cherrypick
PiperOrigin-RevId: 683607090
These Uris are not widely supported yet and were only meant to be
used with content Uris. Restricting this more tightly allows
controllers to use these Uris more easily as they have a stricter
guarentee on what it's needed to load these Uris. Media session
apps with different types of Uris can convert them by setting up
a ContentProvider if needed.
Issue: androidx/media#1783
PiperOrigin-RevId: 683539747
PlaybackVideoGraphWrapper will soon contain an input and an output video
sink, as the rendering of the VideoGraph output frames will be handled
by a DefaultVideoSink instance.
PiperOrigin-RevId: 683167795
This CL also fixes EOS handling to account for not-yet-copied samples in
`remainingInputToCopyFrameCount`, which would throw off the final output
sample count calculation.
For testing, we allow a tolerance of 0.000017% drift between expected
and actual number of output samples. The value was obtained from running
100 iterations of `timeStretching_returnsExpectedNumberOfSamples()` and
calculating the average delta percentage between expected and actual
number of output samples. Roughly, this means a tolerance of 40 samples
on a 90 min mono stream @48KHz.
PiperOrigin-RevId: 683133461
Just clearing the overrides only helps if a text override was
previously set. If the text is shown because of app defined track
selection parameters for language or role flags, the "none" button is
currently not working and we need to clear these flags explicitly.
PiperOrigin-RevId: 682373821
IMA always starts midrolls at index 1. So if there is no preroll ad,
the ad group index in AdPlaybackState is off by 1 all the time, and
may also lead to ArrayIndexOutOfBoundsExceptions when trying to access
the last midroll ad
Issue: androidx/media#1741
PiperOrigin-RevId: 682324368
The workaround in ExternalTextureManager.forceSignalEndOfStream
was being applied even when the decoder did output all frames, and only
the GL pipeline was slow.
This change ensures that workaround is not applied when the decoder
has already produced all output frames.
PiperOrigin-RevId: 680471587
If a codec received the EOS signal and already returned the last
output buffer, we should expect the output EOS very quickly. If
it doesn't arrive within 100ms, we can proceed to end the stream
manually without waiting any further to prevent cases where the
codec is completely stuck otherwise.
PiperOrigin-RevId: 679633116
Some devices can encode portrait 720x1080 but not landscape
1080x720. But VideoSampleExporter always prefers encoding
landscape. Have `assumeFormatsSupported` mirror sample exporter
logic more closely
PiperOrigin-RevId: 679495210
The buffered position was last updated before the beginning of
the renderer loop in doSomeWork. As the loading happens on a
background thread, it may have progressed further already
depending on how long it took to run the renderer loop.
It's slightly more correct to pass in an updated value to
shouldStartPlayback so that playback can start quicker if the
buffering is particularly fast.
PiperOrigin-RevId: 679203465
We began to use a different preload thread than the main thread in the `DefaultPreloadManagerTest`, then in the test execution, we should also ensure that the events queued on the preload looper have been executed.
Also, there is an edge case also causing flakiness. Assume that `DefaultPreloadManager` is preloading source B, then the app invalidates and triggers another sequence of preloading, say they are A, B and C. There is possibility that source B finishes preloading (started before `invalidate`) when A starting to preload, then `onPreloadSkipped` will be triggered for source B instead of `onPreloadCompleted`. However, the functional block inside of `onPreloadSkipped` is dispatched asynchronously, and by then it's possible that B starts to preload again at the consequence of A being completed, then the functional block just mentioned may think that the current source matches at B, and will advance the current preloading source to C, without informing the `Listener.onPreloadCompleted` for it. As the result, the `Listener.onPreloadCompleted` can never be triggered for B for the second sequence. To fix this, we should prevent the functional block in `onPreloadCompleted`, `onPreloadError`, `onPreloadSkipped` to be even dispatched, when the source doesn't match the current preloading one.
PiperOrigin-RevId: 679145353
The MetadataRenderer by default supports icy and vnd.dvb.ait content. Those tracks should therefore be set with the `C.TrackType` `TRACK_TYPE_METADATA` rather than `TRACK_TYPE_UNKNOWN`.
PiperOrigin-RevId: 679132680
When a Media3 controller calls play on a Media3 session, the call
is currently not routed through the platform session at all.
This means the usual exemption to start a FGS for media controller
interactions is not triggered.
We can manually ensure this exemption is given by sending a custom
platform command to the session. The Media3 session will never
receive this command as it's not a known Media3 custom command.
Sessions will see a single onConnect call with a platform controller
from the sender app though. We can prevent this on newer versions of
the code by dropping the onCommand call early.
PiperOrigin-RevId: 679115247
The max number of commands for media items of a browser or controller
can be configured with `setMaxCommandsForMediaItems(int)` on
`MediaController.Builder` and `MediaBrowser.Builder`. An app that has
only limited space for displaying commands can hint this limit to the
session.
A session can receive the value of a connected browser or controller
through `getMaxCommandsForMediaItems()` of a `ControllerInfo` that
is passed into every callback method. The session can then pick the most
sensible commands instead of making the browser app truncating the commands
rather randomly.
When a `MediaBrowser` is connected against a legacy `MediaBrowserServiceCompat`,
the max number of commands is automatically added to the root hints. Conversely,
the value passed in with the root hints to `MediaLibraryService` by a legacy
`MediaBrowserCompat`, is read into `ControllerInfo` like for a Media3 browser.
Issue: androidx/media#1474
#cherrypick
PiperOrigin-RevId: 679076506
To avoid rounding errors, set the `Rounding mode` of the `uvFromVu` and `vuFromUs` results to `HALF_UP`. This `Rounding mode` rounds numbers towards the "nearest neighbor" unless both neighbors are equidistant, in which case round up.
PiperOrigin-RevId: 679003943
See https://developer.android.com/training/cars/media#custom_browse_actions
- `MediaMetadata.supportedCommands` is converted to an array list of
strings into the extras of `MediaDescriptionCompat` with `DESCRIPTION_EXTRAS_KEY_CUSTOM_BROWSER_ACTION_ID_LIST` and vice versa.
- The set of media item command buttons of a session is passed in the
root hints to a legacy browser that connects. A Media3 browser connected
to a legacy service, gets the set of all commands after calling
`getLibraryRoot()`.
#cherrypick
PiperOrigin-RevId: 678807473
Note that unlike the legacy implementation, custom media items
commands can be used for any media items with Media3 API. This
includes `MediaItem` instances that are received from sources
different to `MediaLibraryService` methods.
Hence when connected against a Media3 session these custom commands
can be used with a `MediaController` as well as with a `MediaBrowser`.
Interoperability with `MediaBrowserServiceCompat` will
be added in a follow up CL.
Issue: androidx/media#1474
#cherrypick
PiperOrigin-RevId: 678782860
This check is also done when setting a surface on ExoPlayerImpl,
but by the time we configure the codec the surface may have become
invalid (e.g. when it is destroyed). Even though we immediately remove
a destroyed surface, we could still accidentally use it before the
removal is processed. To avoid these edge cases, we can simply not
configure the codec with an invalid surface.
PiperOrigin-RevId: 678741425
Access is package-private and it will allow the media controller logic
to interact with the underlying platform session directly if needed.
Interop: When a MediaController connects to an older session (before this
change), it won't get the platform token from the session directly.
Many controllers will be set up with a platform or compat token though
and we can simply keep the already known token and use it. The only
cases where we still don't have a platform token in the MediaController
are the cases where the controller is created with a SessionToken based
on a ComponentName.
PiperOrigin-RevId: 678230977
Some devices seem to throw an `IllegalArgumentException` when
attempting to set a valid media button broadcast receiver
for playback resumption. This change handles this exception
as a no-op to avoid crashing the app. As a result, playback
resumption with media keys isn't going to work on these
devices.
This change needs to be reverted once the root cause on these
devices has been fixed (see internal bug ref in source).
Issue: androidx/media#1730
PiperOrigin-RevId: 677904243
Allow apps to preload the first period of the next window in
the playlist of `ExoPlayer`. By default playlist preloading is
disabled. To enable preloading,
`ExoPlayer.setPreloadConfiguration(PreloadConfiguration)` can be
called.
`LoadControl` determines when to preload with its implemenation of `shouldContinuePreloading(timeline, mediaPeriodId, bufferedDurationUs)`.
The implementation in `DefaultLoadControl` allows preloading only when
the player isn't currently loading for playback. Apps can override this
behaviour.
Issue: androidx/media#468
PiperOrigin-RevId: 677786017
This CL adds `SonicTest` and `RandomParameterizedSonicTest` as
initial basic unit testing for `Sonic.java`. The tested scenarios
do not necessarily verify a correct implementation of Sonic, but rather
hope to catch any behaviour change from the current implementation.
The change includes a small fix for a lossy simplification and also
checks whether the output sample count matches the expected drift from
the truncation accumulation error present in Sonic's resampler. This is
important as pre-work for fixing issues with unexpected durations within
`SonicAudioProcessor` and `SpeedChangingAudioProcessor` that cause AV
sync issues for speed changing effects.
This is a partial roll forward of e88d6fe459, which was rolled back in
873d485056.
PiperOrigin-RevId: 677756854
AudioTrack doesn't automatically ramp up the volume after a flush
(only when resuming with play after a pause), which causes audible
pop sounds in most cases. The issue can be avoided by manually
applying a short 20ms volume ramp, the same duration used by the
platform for the automatic volume ramping where available.
Together with the already submitted 6147050b90, this fixes the
unwanted pop sounds for most cases in the desired way. It only
leaves two cases that are not handled perfectly:
- If the media file itself contains a volume ramp at the beginning,
we wouldn't need this additional ramping. Given the extremely
short duration, this seems ignorable and we can treat it as a
future feature request to mark the beginning of media in a special
way that can then disable the volume ramping.
- For seamless period transitions where we keep using the same
AudioTrack, we may still get a pop sound at the transition. To
solve this, we'd need a dedicated audio processor to either ramp
the end of media down and the beginning of the next item up, or
apply a very short cross-fade. Either way, we need new signalling
to identify cases where the media originates from the same source
and this effect should not be applied (e.g. when re-concatenating
clipped audio snippets from the same file).
PiperOrigin-RevId: 676860234
When sending a custom command with `browser.sendCustomCommand` when
connected to a legacy browser service, the custom command was delivered to `MediaSessionCompat.Callback.onCustomAction` instead of the service method
`onCustomAction`. The difference is that the service version can return an
async response with a bundle, while the session callback version doesn't
have a return value.
Hence, the service method was never called and it wasn't possible to send
a reponse or signal an error back to the browser. The resulting
`ListanableFuture` simply always immediately resolved to a success.
This change overrides `ListenableFuture<SessionResult> sendCustomCommand(SessionCommand command, Bundle args)` in
`MediaBrowserImplLegacy` to use the `MediaBrowserCompat` method to send
instead of the `MediaControlleCompat` method that was used by the subclass
`MediaControllerImplLegacy`. This involves the service callback instead of the
session callback and enables `MediaBrowser` to get the actual return value
from the legacy service.
Issue: androidx/media#1474
#cherrypick
PiperOrigin-RevId: 676519314
Last buffer was not flipped, so was writing the garbage data between
limit and capacity, rather than the actual data between position and
limit.
As a result, all PCM audio dump files need updating.
PiperOrigin-RevId: 676452990
- Added logic to parse media duration from the `mdhd` box for accurate frame rate calculation.
- Fallbacks to track duration from `tkhd` when `mdhd` contains invalid or missing data.
- Avoids incorrect frame rate calculations in MP4 files with an edit list (`elst`) box.
- Adds frame rate calculations for partially fragmented MP4 files.
- Verified accuracy with tools like `mediainfo` and `ffprobe`.
Issue: androidx/media#1531
**Note**: The slight difference in frame rate values in dump files that aren’t MP4s with an edit list or fragmented MP4s isn’t due to differences in `tkhd` and `mdhd` duration values (which should be identical for non-edited or non-fragmented files). Rather, it’s because they are calculated using different timescales. The `mvhd` box defines a global movie timescale, which is used for the track's `tkhd` duration. Meanwhile, each track’s `mdhd` box defines its own timescale specific to its content type, which we now use for more accurate frame rate calculation.
PiperOrigin-RevId: 676046744
This method is documented that it may only be called in `STATE_OPENED`
or `STATE_OPENED_WITH_KEYS`. It's possible for it to be called in other
states (like `STATE_ERROR`) without this guard.
Previously this didn't cause issues, but since 9d62845c45
we assume that the `sessionId` is non-null in this method, which results
in an `IllegalStateException` when the documented state restriction is
ignored.
PiperOrigin-RevId: 675969256
This is no longer needed now our `compileSdk` implies a new-enough AGP
which does this out-lining automatically via R8. See also
https://issuetracker.google.com/345472586#comment7
There's no plan to remove the `ApiXXX` classes, but no new ones need
to be added.
PiperOrigin-RevId: 675940634
This is to allow not setting the MediaFormat OPERATING_RATE and PRIORITY
altogether. The current behvaiour, if left the value `UNSET`, it'll apply the
our optimizations, but apps might want to disable this optimization.
PiperOrigin-RevId: 675923909
The Galaxy Tab S7 FE has a device issue that causes 60fps secure H264 streams to be marked as unsupported. This CL adds a workaround for this issue by checking the CDD required support for secure H264 in addition to the current check on standard H264. If the provided performance points do not cover the CDD requirement of support 720p H264 at 60fps, then it falls back to using legacy methods for checking frame rate and resolution support.
Issue: androidx/media#1619
PiperOrigin-RevId: 675920968
After the change in a879bc2154, the Sequence won't have repeated
EditedMediaItems. Thus if the sequence is looping, the last EditedMediaItems
in the Sequence object might not corresponds to the last item in the "logical"
sequence.
PiperOrigin-RevId: 675912197
NotificationChannel and NotificationChannelGroup are public APIs that were
added in Android O, so at this point it is okay to have them appear in API
signatures.
PiperOrigin-RevId: 675756005
The muxer might not have accepted the first sample, if it
is waiting for audio track.
This bug causes issue when
1. VideoSampleExporter gives first sample (timestamp = 0) to the muxer.
2. Muxer does not write it because its waiting for audio track.
3. The video pipleline has processed all the sample and they are ready
to be consumed.
4. VideoSampleExporter fetches the next available sample from encoder (which is still with timestamp = 0) but it changes its timestamp to last timestamp because VideoSampleExporter thinks it has muxed the sample at timestamp zero, but in reality it hasn't. This is because the flag `hasMuxedTimestampZero` is set when queueing the input, rather than actually muxing the input.
This scenario can happen when video is processed much faster than
the audio.
PiperOrigin-RevId: 675565603
Previously, track IDs were added to `trackIndicesPerSampleInQueuedOrder`
even when the sample was not committed. This caused issues where attempts
to read samples from the `SampleQueue` returned `C.RESULT_READ_NOTHING`,
which led to an exception being thrown due to the assumption that samples
were available to read.
This fix updates the logic to track sample commits by comparing the write index before and after calling `SampleQueue.sampleMetadata`. Track indices are only added if the sample was committed, ensuring accurate sample handling and avoiding exceptions.
PiperOrigin-RevId: 675526115
to
LAST_SAMPLE_DURATION_BEHAVIOR_SET_FROM_END_OF_STREAM_BUFFER_OR_DUPLICATE_PREVIOUS
This CL also combines LAST_SAMPLE_DURATION_BEHAVIOR_SET_FROM_END_OF_STREAM_BUFFER
and LAST_SAMPLE_DURATION_BEHAVIOR_DUPLICATE_PREVIOUS.
The reason for combining the two enums is that, when the option
to use END_OF_STREAM_BUFFER is selected and if the EOS buffer is
not provided then the muxer anyways fallbacks to duplicate
duration behavior.
The last sample with 0 durations seems less useful so
change the default behavior to non-zero duration.
This will also match the behavior with MediaMuxer.
PiperOrigin-RevId: 675189932
To support for 3gpp h263 codec in Mp4Muxer currently profile and level is hardcoded and provided to h263 box. Parse profile and level from MediaFormat and use those value to write h263 box.
PiperOrigin-RevId: 675004590
The tests were flaky because CompositionPlayer registers audio sequences'
formats to the audio pipeline in first-come-first-serve order. With the change
in bc8d82355f, the audio format is deterministic.
The test is turned off in 060356ea00
PiperOrigin-RevId: 674261801
`PreloadMediaSource` allows to have a `startPositionUs` passed when `preload` is called, then in `PreloadControl.onContinueLoadingRequested`, it can be more intuitive to see the buffered duration rather than the absolute buffered position as the preload progress. Similar in `DefaultPreloadManager`, we haven't allowed the apps to set a custom start position for individual sources though, once we add this support, using the "duration from the start position" than the absolute position will be less error-prone, otherwise, it can run into a case that the position that the apps set is smaller than the start position.
PiperOrigin-RevId: 674251362
Sonic would accumulate truncation errors on float to int conversions
that caused the final output sample count to drift noticeably, by
hundreds of samples on streams of a few minutes of length. The fix now
keeps track of the truncation error and compensates for it.
Other small fixes include eliminating lossy operations (e.g. int
division) and using doubles instead of floats for resampling where
helpful.
This CL also introduces `SonicParameterizedTest`, which helps test
resampling on an arbitrary number of randomly generated parameters,
with random sample data. `SonicParameterizedTest` uses `BigDecimal`s
for calculating sample count values, as to avoid precision issues with
large sample counts.
PiperOrigin-RevId: 673852768
`CPU_FEATURES_COMPILED_ANY_ARM_NEON` is always defined when `CPU_FEATURES_ARCH_ARM` is but it can be defined as `0` or `1` ([cpu_features source](8e60d3f9be/include/cpu_features_macros.h (L237-L245))). This patch makes sure to use the value of `CPU_FEATURES_COMPILED_ANY_ARM_NEON` instead of whether it is defined or not.
PiperOrigin-RevId: 673837522
