Bug 219873 – [GPUProcess] Bump QoS to UserInteractive for real-time audio rendering thread

• Status: RESOLVED
• Resolution: FIXED
• Component: Web Audio

[Chris Dumez]

Bump QoS to UserInteractive for real-time audio rendering thread. We create our own thread on WebContent side when the GPU Process is enabled since the CoreAudio rendering thread lives in the GPU Process. To ensure the same quality of service, we need to make sure the thread we create gets very high QoS.

[Chris Dumez]

Created attachment 416194 [details]
Patch

[Chris Dumez]

Created attachment 416196 [details]
Patch

[Peng Liu]

Comment on attachment 416196 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=416196&action=review

> Source/WTF/ChangeLog:8
> +        Add QOS parameter to the Thread construct (similarly to WorkQueue) so that the caller can

Maybe priority is a better name for the parameter?

[Chris Dumez]

(In reply to Peng Liu from comment #3)
> Comment on attachment 416196 [details]
> Patch
> 
> View in context:
> https://bugs.webkit.org/attachment.cgi?id=416196&action=review
> 
> > Source/WTF/ChangeLog:8
> > +        Add QOS parameter to the Thread construct (similarly to WorkQueue) so that the caller can
> 
> Maybe priority is a better name for the parameter?

QoS (quality of service) is the naming on our platforms. The concept of process/thread priority also exists on our platforms but has a different meaning. As a result, I wouldn't want to conflate the 2. Also note that I am merely following the pattern in WorkQueue, not introducing a new concept in this patch.

[Geoffrey Garen]

Comment on attachment 416196 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=416196&action=review

> Source/WebCore/Modules/webaudio/AudioWorkletThread.cpp:69
> +    }, ThreadType::Audio, m_parameters.isAudioContextRealTime ? Thread::QOS::UserInteractive : Thread::QOS::Default);

What's the non-realtime case for audio?

> Source/WebCore/worklets/WorkletParameters.h:40
> +    bool isAudioContextRealTime;

Where does this get initialized?

[Chris Dumez]

(In reply to Geoffrey Garen from comment #5)
> Comment on attachment 416196 [details]
> Patch
> 
> View in context:
> https://bugs.webkit.org/attachment.cgi?id=416196&action=review
> 
> > Source/WebCore/Modules/webaudio/AudioWorkletThread.cpp:69
> > +    }, ThreadType::Audio, m_parameters.isAudioContextRealTime ? Thread::QOS::UserInteractive : Thread::QOS::Default);
> 
> What's the non-realtime case for audio?

When you construct an OfflineAudioContext (instead of an AudioContext), audio is rendered offline into a buffer instead of in real time on the speakers.

> 
> > Source/WebCore/worklets/WorkletParameters.h:40
> > +    bool isAudioContextRealTime;
> 
> Where does this get initialized?

Yes, it is, based on whether or not the AudioContext is offline or not, which answers your question above.

[Chris Dumez]

Comment on attachment 416196 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=416196&action=review

> Source/WebCore/Modules/webaudio/AudioWorkletMessagingProxy.cpp:57
> +        worklet.audioContext() ? !worklet.audioContext()->isOfflineContext() : false

For Geoff, this is where the isAudioContextRealTime flag gets initialized.

[Geoffrey Garen]

👍🏻

[EWS]

Committed r270806: <https://trac.webkit.org/changeset/270806>

All reviewed patches have been landed. Closing bug and clearing flags on attachment 416196 [details].

[Radar WebKit Bug Importer]

<rdar://problem/72317707>