This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [tree-ssa] live analysis on local static functions


> On Thu, Oct 16, 2003 at 09:36:44AM +0200, Jan Hubicka wrote:
> > This is no fun.  THere are number of functions in reload inlined because
> > they are called once and apparently one of them kills datastructures
> > (reload fails in the final sanity check on frame offsets to match again
> > only on compiling large function - it is able to get trought half of
> > stage3)
> 
> Didn't you say you have a patch that makes things work again?

I made patch to work around the unreferenced symbol problem (I am just
testing more proper fix but it is equivaelnt in the functionality on C
bootstrap), but still I get failures later in the process
when I enable unit-at-a-time.

With the snapshot I have at home I get the misscompilation in reload.
There are no differences in dump files of stage1 and stage2 compiler
(except for addresses) up to reload and reload manges to mess something
up, so it looks like debugging nightmare.  Only I know is that the
problem disappears when I disable inlining of functions called once

However with current snapshot I get failures much earlier and also bogus
warnings about unused variables.  So I will debug these first and
hopefully it will make the later to disappear.  In case it won't I will
try to debug it or we can for disabling of the functions called once
optimization instead of whole unit-at-a-time until the branch gets fixed
wrt inlining.

Thanks for all the help with the inlining issue!
Honza
> Diff dumps between that patch and the prospective.
> 
> 
> r~


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]