harmony-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Mikhail Fursov" <mike.fur...@gmail.com>
Subject Re: [drlvm] [jitrino] inliner status
Date Thu, 07 Jun 2007 12:45:13 GMT
It's ok. Having 1M instructions in a method will cause Jitrino to consume
100Mb-1Gb memory during compilation.
This is too high and risky value to be used in real workloads.

On 6/7/07, Yves Ineichen <yves.ineichen@dplanet.ch> wrote:
>
> Mikhail Fursov wrote:
> > On 6/7/07, Yves Ineichen <yves.ineichen@dplanet.ch> wrote:
> >>
> >> hi mikhail,
> >>
> >> thanks for your hint. the current svn version (svn = r545097) seems to
> >> work with the optimzier rules but now the code generator asserts:
> >>
> >> java:
> >>
> >>
> /data/home/iyves/src/harmony/working_vm/vm/jitrino/src/shared/CGSupport.cpp:84:
> >>
> >> void Jitrino::setBCMappingEntry(void*, uint32, uint16): Assertion
> >> `key<1000*1000' failed.
> >> SIGABRT in VM code.
> >> Aborted
> >>
> >> did i miss anything in my path.codegen?:
> >
> >
> > I think that this problem is not about optimizations. The 'key' here is
> an
> > instruction ID.
> > I put this assertion into code because I do not believe we can have
> methods
> > with 1M instructions (both java method size and inliner limits will not
> > allow us to have methods of this size) and if you have ones this could
> be a
> > bug.
> > Could you check the HIR of your method and tell us how many instructions
> it
> > has?
> >
> > You can use "-verbose:em" option to know the name of the method that
> fails
> > and "-XX:jit.f.filter=class::method -XX:jit.f.arg.log=ct,irdump" to get
> a
> > HIR log.
> >
> >
>
> ah sorry my mistake... i played around with the inliner parameters and
> forgot to reset them to the default values.
>
> --
> Yves Ineichen
> ::p = "This statement cannot be proven"::
>



-- 
Mikhail Fursov

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message