Bug 180341

Summary: GC should have a clock
Product: WebKit Reporter: Filip Pizlo <fpizlo>
Component: JavaScriptCoreAssignee: Filip Pizlo <fpizlo>
Status: NEW    
Severity: Normal CC: keith_miller, rniwa
Priority: P2    
Version: WebKit Nightly Build   
Hardware: All   
OS: All   
Attachments:
Description Flags
it's a start none

Filip Pizlo
Reported 2017-12-03 18:52:22 PST
The JSC GC should have a time source that advances when the GC is keeping the main thread from running. If we expose this (via internal APIs in the shell or whatever) to benchmarks, then we could switch a benchmark from measuring actual time (performance.now, preciseTime, whatever) to measuring gcTime. Then, we would be able to see what aspect of a performance change is due to changes in amount of time spent GCing. For changes that are meant to optimize the GC itself, this kind of clock-based analysis might be the best way of understanding the exact effect a change is having.
Attachments
it's a start (25.14 KB, patch)
2017-12-03 18:54 PST, Filip Pizlo 		no flags
DetailsFormatted DiffDiff
View All Add attachment proposed patch, testcase, etc.
Filip Pizlo
Comment 1 2017-12-03 18:52:54 PST
@Keith: we could use this to confirm why adding those two fields to JSObject is a regression. Presumably, it'll be a gcTime regression.
Filip Pizlo
Comment 2 2017-12-03 18:54:18 PST
Created attachment 328321 [details] it's a start
Keith Miller
Comment 3 2017-12-04 10:20:47 PST
Yeah, this would probably be useful. It might also be useful to time a particular gc phase. I suppose I could do that by commenting things out though.
Note You need to log in before you can comment on or make changes to this bug.