RFR(S): 8235673: [C1, C2] Split inlining control flags

[Doerr, Martin]

Hi Tobias,

thanks a lot for your help.

> Okay, I'll run it through our startup benchmark suite. Will report back once it
> finished.
I'm using an aggressive setting of C1InlineStackLimit=5 with TieredCompilation. This saves more stack space, but should be evaluated carefully.
So I highly appreciate your startup benchmarks.

> I meant that the authors of these tests might have intended to tweak C1
> behavior when adding the
> flag whereas with your change, only C2 behavior is affected. That doesn't
> necessarily mean that the
> test will fail now but it could mean that the regression the test was written
> for is not triggered
> anymore. You might just want to add the C1 flag as well (in another @run),
> Vladimir K. also
> mentioned this in his review.
I got your point. I'll take a look and reply to Vladimir's email.

Best regards,
Martin


> -----Original Message-----
> From: Tobias Hartmann <tobias.hartmann at oracle.com>
> Sent: Freitag, 8. Mai 2020 08:40
> To: Doerr, Martin <martin.doerr at sap.com>; Nils Eliasson
> <nils.eliasson at oracle.com>; hotspot-compiler-dev at openjdk.java.net
> Subject: Re: RFR(S): 8235673: [C1, C2] Split inlining control flags
> 
> Hi Martin,
> 
> On 07.05.20 15:42, Doerr, Martin wrote:
> > Hi Tobias,
> >
> > thanks for looking at my change.
> > It is only expected to influence startup, not peak performance.
> >
> > Nevertheless, I've run benchmarks to check peak performance as well:
> SPEC jvm 2008, SPEC jbb 2015
> > No regressions observable, as expected.
> 
> Okay, great.
> 
> > For startup performance, I've ran SPEC jbb 2005 with throughput
> measurements every 1.5 seconds like I had shown in my fosdem talk
> (https://fosdem.org/2020/schedule/event/jit2020/).
> > I couldn't observe any regression, either.
> >
> > It would be very helpful if other people (e.g. from Oracle) could run
> additional benchmarks. I don't know what you use to check startup
> performance.
> 
> Okay, I'll run it through our startup benchmark suite. Will report back once it
> finished.
> 
> >> Did you verify that tests using these flags are still working as expected
> (i.e.,
> >> intend to only adjust C2's behavior)?
> > Using the existing flags still works for C2. So there's no issue with C2 tests.
> > I'm not aware of any test which requires one of these flags to modify C1
> behavior.
> >
> > I've run a substantial amount of tests and couldn't find any related issues:
> > jtreg, jck (normal, with -Xcomp, with -Xcomp -XX:-TieredCompilation),
> SAP's proprietary tests
> 
> I meant that the authors of these tests might have intended to tweak C1
> behavior when adding the
> flag whereas with your change, only C2 behavior is affected. That doesn't
> necessarily mean that the
> test will fail now but it could mean that the regression the test was written
> for is not triggered
> anymore. You might just want to add the C1 flag as well (in another @run),
> Vladimir K. also
> mentioned this in his review.
> 
> Best regards,
> Tobias