Bug 226541 - All CodeBlocks should be asynchronously sweepable.
Summary: All CodeBlocks should be asynchronously sweepable.
Status: NEW
Alias: None
Product: WebKit
Classification: Unclassified
Component: New Bugs (show other bugs)
Version: WebKit Nightly Build
Hardware: Unspecified Unspecified
: P2 Normal
Assignee: Keith Miller
URL:
Keywords: InRadar
Depends on:
Blocks:
 
Reported: 2021-06-02 09:54 PDT by Keith Miller
Modified: 2021-06-05 18:39 PDT (History)
9 users (show)

See Also:

Attachments
WIP (26.11 KB, patch)
2021-06-02 09:55 PDT, Keith Miller 		no flags Details | Formatted Diff | Diff
LastPatch (35.23 KB, patch)
2021-06-05 18:38 PDT, Keith Miller 		no flags Details | Formatted Diff | Diff
Show Obsolete (1) View All Add an attachment (proposed patch, testcase, etc.)

Note You need to log in before you can comment on or make changes to this bug.
Description Keith Miller 2021-06-02 09:54:43 PDT 
All CodeBlocks should be asynchronously sweepable.
Comment 1 Keith Miller 2021-06-02 09:55:02 PDT 
Created attachment 430370 [details]
WIP
Comment 2 Radar WebKit Bug Importer 2021-06-02 14:06:35 PDT 
<rdar://problem/78782125>
Comment 3 Keith Miller 2021-06-05 18:38:40 PDT 
Created attachment 430668 [details]
LastPatch
Comment 4 Keith Miller 2021-06-05 18:39:44 PDT 
Seems to be a 1-2% Speedo regression. I couldn't figure out where the regression came from. It doesn't seem to come from iterating the codeblock set with Subspace::forEachLiveObject instead of the CodeBlockSet. We also don't spend more than .3% of our time in CodeBlock destructors so it's not that the code blocks themselves got slower.

My assumption is that the regression came from lost cache locality when destroying the code blocks.

I'm gonna give up on this for now since I can't identify the reasoning for the slowdown.