RFR(M): 8134579: [TESTBUG] Some bmi tests fail if can_access_local_variables is on.

Lindenmaier, Goetz goetz.lindenmaier at sap.com
Mon Sep 28 15:13:51 UTC 2015


Hi Filipp, 

unfortunately that problem did not reproduce well.  It only happened in our
nightly tests.  I added debug printout to the test, actually I enabled graph
printing, and there I see the uncommon traps. Analysis was very tedious, 
because I always had to await the nightly tests.

I didn't look at the succeeding tests, though.

The other problem reproduces easily if you pass and agentlib to the vm
called in the test.  (Our VM has can_access_local_vaiables on per default.)

Best regards,
  Goetz.



-----Original Message-----
From: Filipp Zhinkin [mailto:filipp.zhinkin at gmail.com] 
Sent: Montag, 28. September 2015 16:44
To: Lindenmaier, Goetz
Cc: hotspot compiler
Subject: Re: RFR(M): 8134579: [TESTBUG] Some bmi tests fail if can_access_local_variables is on.

Hi Goetz,

do you why some tests were not affected by the uncommon trap issue?

Regards,
Filipp.

On Mon, Sep 28, 2015 at 2:54 PM, Lindenmaier, Goetz
<goetz.lindenmaier at sap.com> wrote:
> Could someone please have a look at this?
>
>
>
> Thanks and best regards,
>
>   Goetz.
>
>
>
> From: Lindenmaier, Goetz
> Sent: Montag, 21. September 2015 12:45
> To: hotspot compiler
> Subject: RFR(M): 8134579: [TESTBUG] Some bmi tests fail if
> can_access_local_variables is on.
>
>
>
> Hi,
>
>
>
> I saw two problems in the intrinsics/bmi/verifycode tests in our nightly
> runs.
>
> Unfortunately they did not reproduce easily.
>
>
>
> With can_access_local_variables() switched on the pattern matched in
> AddnTestI/L
>
> is messed up with an edge to the Uncommon trap for the NPE.
>
> T/LzcntTestI/L were compiled with an uncommon trap before the call to the
>
> Interger method.
>
>
>
> This easily can be fixed by checking for NULL and calling the test function
> once
>
> Before compiling the method.
>
>
>
> I also fixed the name of AddnTestI/L.  They actually call TestAndnI/L
> internally
>
> and thus should be named AndnTestI/L.
>
>
>
> Please review this change.  I please need a sponsor.
>
> http://cr.openjdk.java.net/~goetz/webrevs/8134579-bmi/webrev.00/
>
>
>
> Best regards,
>
>   Goetz Lindenmaier
>
>


More information about the hotspot-compiler-dev mailing list