RFR(S): 8244086: Following 8241492, strip mined loop may run extra iterations

[Doerr, Martin]

Hi Roland,

> Anyway, I disagree with the "huge amount of problems" comment. All non
> trivial C2 changes that I've been involved with have had a long bug
> tail. I don't think LSM is better or worse in that regard.
Well, we have seen a lot of crashes in production since JDK 10. But we don't have to discuss this here. Let's just fix it.

> Actually in the review thread for 8223051, John suggested some
> refactoring that would apply here as well. I'll update this webrev.
Ok. Thanks.

> > TestStripMinedLimitBelowInit.java:
> > I suggest to use -XX:-TieredCompilation
> 
> Why that?
Your test is constructed in a way such that it warms up and creates a highest tier nmethod. Then, this method is called once with other parameters. Why should we create a C1 nmethod, too?
-XX:-TieredCompilation makes it easier to follow when using the test case for debugging.

Best regards,
Martin


> -----Original Message-----
> From: Roland Westrelin <rwestrel at redhat.com>
> Sent: Dienstag, 5. Mai 2020 09:20
> To: Doerr, Martin <martin.doerr at sap.com>; Pengfei Li
> <Pengfei.Li at arm.com>; hotspot-compiler-dev at openjdk.java.net
> Cc: nd <nd at arm.com>
> Subject: RE: RFR(S): 8244086: Following 8241492, strip mined loop may run
> extra iterations
> 
> 
> Hi Martin,
> 
> > Given the huge amount of problems related to LSM I wonder if it's worth
> maintaining it. It comes with a high price.
> 
> And default back to a safepoint in every loop iteration for latency
> sensitive gcs? People involved with gcs would need to comment on whether
> they think it's acceptable.
> 
> Anyway, I disagree with the "huge amount of problems" comment. All non
> trivial C2 changes that I've been involved with have had a long bug
> tail. I don't think LSM is better or worse in that regard.
> 
> > Nevertheless, your version looks at least correct to me.
> 
> Thanks for the review.
> 
> > loopnode.cpp:
> > Please add your formula to the comment.
> 
> Actually in the review thread for 8223051, John suggested some
> refactoring that would apply here as well. I'll update this webrev.
> 
> > TestStripMinedLimitBelowInit.java:
> > I suggest to use -XX:-TieredCompilation
> 
> Why that?
> 
> Roland.