harmony-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Xiao-Feng Li" <xiaofeng...@gmail.com>
Subject Re: [drlvm][gcv5] is there an exclude list for "build test"?
Date Tue, 05 Dec 2006 00:15:33 GMT
On 12/4/06, Mikhail Fursov <mike.fursov@gmail.com> wrote:
> On 12/2/06, Xiao-Feng Li <xiaofeng.li@gmail.com> wrote:
> >
> >
> > At the same time, I wonder if that's better for JIT to check the
> > finalizable property for VM helper as well. I guess this can be easier
> > for VM helpers since they are GC independent with standard interface.
> > Only guess anyway, how do you think, Mikhail?
> >
> It's not a problem to check "isFinalizable" for JIT during the compilation.
> The question is how to use this knowledge.
> Today JIT has only 1 usage scenario: do not inline fast-path helper for
> objects with 'finalize' method.
> If JIT checks this property before generating a call to usual VM method what
> is the option? Should JIT call different VM helper for allocation in this
> case?

Mikhail, yes. I think you make the right point. My opinion is, JIT can
call different native alocation function (alloc_fast or common alloc)
as well as it does for helper inlining after it checks the
finalization property.

Actually in my experiments, I found the NEXT_TO_HIGH_BIT is no longer
marked by JIT before its invocation to native GC alloc functions. That
means, it is unused in current DRLVM. (That also means, the test case
should be removed). So if without JIT's help, we need check gc_vtable
for the finalization property even in the alloc_fast function (the
native one).


> --
> Mikhail Fursov

View raw message