1342384 - Regression on misc-basic-array-forof due to TraceLogger changes for multithreaded runtimes

Status: RESOLVED WONTFIX

(Core :: JavaScript Engine: JIT, defect, P1)

Description

[Hannes Verschore [:h4writer]]

See bug 1342369. This is about the third regression. Locally it changed score from 45 to 47 (ms):

https://arewefastyet.com/#machine=29&view=single&suite=misc&subtest=basic-array-forof&start=1486531747&end=1486587463

The first regression visible on that graph is (click on jsc in the legend to not see that line):
Bug 1334194 - TraceLogger changes for multithreaded runtimes, r=h4writer.

Comment 1

[Hannes Verschore [:h4writer]]

The second regression visible on that graph is:
Brian Hackett <[e-mail]> - Wed, 08 Feb 2017 05:23:32 -0700 - rev 341355
Bug 1334212 - Handle multiple contexts per runtime in compiled JIT code, r=h4writer.

Comment 2

[Brian Hackett [Laid off!]]

Bug 1334212 shouldn't affect behavior when JS_TRACE_LOGGING is not defined.

Comment 3

[Jan de Mooij [:jandem]]

(In reply to Brian Hackett (:bhackett) from comment #2)
> Bug 1334212 shouldn't affect behavior when JS_TRACE_LOGGING is not defined.

But I think JS_TRACE_LOGGING is #defined by default nowadays...

Comment 4

[Hannes Verschore [:h4writer]]

(In reply to Hannes Verschore [:h4writer] from comment #0)
> See bug 1342369. This is about the third regression. Locally it changed
> score from 45 to 47 (ms):
> 
> https://arewefastyet.com/#machine=29&view=single&suite=misc&subtest=basic-
> array-forof&start=1486531747&end=1486587463
> 
> The first regression visible on that graph is (click on jsc in the legend to
> not see that line):
> Bug 1334194 - TraceLogger changes for multithreaded runtimes, r=h4writer.

This will be solved by bug 1345148. The locking was taking some overhead. Bug 1345148 makes sure we need no locking when tracelogger is not active.

(In reply to Hannes Verschore [:h4writer] from comment #1)
> The second regression visible on that graph is:
> Brian Hackett <[e-mail]> - Wed, 08 Feb 2017 05:23:32 -0700 -
> rev 341355
> Bug 1334212 - Handle multiple contexts per runtime in compiled JIT code,
> r=h4writer.

https://hg.mozilla.org/integration/mozilla-inbound/rev/85849dc84129

Seems like this is caused by the need to load values instead of comparing values out of memory immediately.
That part added a lot of these. I don't think we can actually do something about it though.
I'll mark it as WONTFIX