Re: [Jikesrvm-core] O1 for Bootimages

[Michael H. <hi...@us...>]

Hi Peter,

> I thought about forcing the method to be baseline compiled but that
> feels like we are just continuing to hide bugs and it will only jump
> up and bite us again. So I am proposing that for all the
> configurations that run in the core test run we force bootimage
> compilation to O1.

I'm sorry I can offer the time to help hunt down this bug, but I do want 
to mention something to look out for if you make the change above. Namely, 
inside the adaptive system there is some logic that says something like 
don't recompile any method that is at the max opt level already (unless 
you have ideas that inlining can be improved).  This logic means that in a 
current production image we don't spend any time trying to recompile the 
VM (except for the rare improved inlining scenario).   However, if you 
bootimage compile the VM at O1, the logic might result in some of the VM 
being recompiled at runtime, which on one hand is the right thing to do if 
the VM method is really important, but it could result in a performance 
degradation (particularly for short runs) in that you are paying extra 
compile time than you would in a config where the boot image is compiled 
at O2.

IMO, the *right* answer is to fix the bug or force the method to be 
compiled at O1, but since I'm not able to contribute any time on this, I 
can see how other alternatives are attractive.

Mike
-----------------------------------------------------------------------------------------------------------
Michael Hind, Senior Manager, Programming Technologies Department 
IBM T.J. Watson Research Center
http://www.research.ibm.com/people/h/hind
914 784-7589