This is the mail archive of the gcc@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] New regressions as of 2003-11-04


> On Tue, 2003-11-04 at 16:55, law@redhat.com wrote:
> > In message <1067981299.7039.71.camel@p4>, Andrew MacLeod writes:
> >  >Has remove_useless_stmts_and_vars_cond() been modified since/during
> >  >lowering?
> > The code to handle this kind of stuff certainly hasn't been changed
> > in any significant way.
> > 
> >  > it pretty much has to be since there is never anything
> >  >actually *IN* the arms now, so it oughta be looking in the target block,
> >  >and it wouild pick this up wouldn't it?
> > Right.    The only difficulty is you don't have easy access to the
> > target block in that code since your CFG is potentially horked. 
> > 
> > [ Then again, maybe the lowering changes mean the CFG isn't horked
> >   anymore..  I haven't though much about that. ]
> > 
> Well, SSA->normal doesn't hork anything... So it should be fine coming
> out of there.

Modulo the edge splitting bugs, yes, I tested that :)
I can verify the CFG after out-of-SSA form and use it for the expansion
and RTL cleanup_cfg afterwards.

I think there is not much reason for remove_useless_stmts_and_vars_cond
once gimple is in low form.  At least not the parts that affects
instruction stream and kill CFG.

Honza

> 
> Although its about to not be pure GIMPLE any more coming out of SSA :-),
> but then I dont think you were depending on anything like that...
> 
> Andrew
> 


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