[Webkit-unassigned] [Bug 235114] REGRESSION (iOS 15): Massive variation in microbenchmark results when tab returns to foreground

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=235114

--- Comment #9 from Simon Taylor <simontaylor1 at ntlworld.com> ---
(In reply to Chris Dumez from comment #7)
> I am starting to wonder if something is wrong with the benchmark itself?

Certainly possible.

In the WebGL one I added console.log() calls (which show up as syscalls if WebInspector is open) and verified that the section where the calls are counted was 1ms long in both fast and slow cases. The idea of accumulating to a global var was to stop the optimizer removing the calls entirely. Different JIT levels being used does feel most likely.

I'm not able to reproduce very readily on M1 Pro MBP with Safari Tech Preview. There's a vague hint of a 28k / 32k difference (perhaps some non-determinism but at the same JIT tier?), but I haven't seen the ~11k case there.

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.webkit.org/pipermail/webkit-unassigned/attachments/20220112/36043d4d/attachment.htm>