Bug 215183 – Preload graphics drivers in Mac WebProcess

• Status: RESOLVED
• Resolution: FIXED
• Component: New Bugs

[Ben Nham]

Asynchronously preload graphics drivers in Mac WebProcess

[Ben Nham]

Created attachment 406023 [details]
Patch

[Radar WebKit Bug Importer]

<rdar://problem/66587606>

[Ben Nham]

Created attachment 406025 [details]
Patch

[Simon Fraser (smfr)]

Comment on attachment 406025 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=406025&action=review

> Source/WebKit/WebProcess/WebPage/mac/TiledCoreAnimationDrawingArea.mm:227
> +    if (settings.acceleratedDrawingEnabled()) {

I would not bother checking this. It's the default, and even if it isn't, canvas or WebGL can use Metal.

[Tim Horton]

Also, Darin will show up shortly to tell you that you don’t need the dispatch_once since this code is already not threadsafe; just use a static bool.

[Ben Nham]

If I don't need to check the setting, then I'll probably move this to earlier in the init process (e.g. to WebProcess::platformInitializeWebProcess or WebProcess::platformSetWebsiteDataStoreParameters).

[Tim Horton]

Definitely don't need to.

[Ben Nham]

Created attachment 406110 [details]
Patch

[Alexey Proskuryakov]

Comment on attachment 406110 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=406110&action=review

> Source/WebCore/page/ProcessWarming.cpp:47
> +#include <Metal/Metal.h>

Looks like this is an Objective-C header.

[Alexey Proskuryakov]

(so the patch breaks the build)

[Geoffrey Garen]

Comment on attachment 406110 [details]
Patch

Seems like you could call WebCore::GPUDevice::tryCreate() to make this behavior C++-friendly.

[Ben Nham]

Created attachment 406236 [details]
Patch

[Ben Nham]

Comment on attachment 406236 [details]
Patch

I profiled CompPLT with this patch applied and now see all MTLCopyAllDevices work happening in prewarmGlobally as expected.

[Darin Adler]

Comment on attachment 406236 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=406236&action=review

> Source/WebCore/page/ProcessWarming.cpp:89
> +    GPUDevice::tryCreate(gpuRequestAdapterOptions);

Why is the local variable needed? Isn’t this the same as passing WTF::nullopt?

[Ben Nham]

Created attachment 406261 [details]
Patch

[EWS]

Committed r265418: <https://trac.webkit.org/changeset/265418>

All reviewed patches have been landed. Closing bug and clearing flags on attachment 406261 [details].