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: Loop optimizer issues


Hello,

> > > >I think there's been a bit of a disconnect in this discussion.
> > > >
> > > >Some people are arguing against merging the rtlopt changes into 
> > > >tree-ssa.
> > > >Some people are arguing for merging Zdenek's loop optimizer work into
> > > >tree-ssa.
> > > >
> > > >Nobody, AFAICT, is arguing for merging all of rtlopt into tree-ssa.
> > > >
> > > >It seems to me that Zdenek is interested in making his loop optimizer 
> > > >work
> > > >at the tree level, and therefore it makes sense to bring it into 
> > > >tree-ssa.
> > > >
> > > As long as it's going to work on the tree-ssa SSA form, and not just 
> > > GIMPLE trees (without SSA), i have no problem with it.
> > > If it's not going to work on the SSA form, i'm against it because it 
> > > would mean that we have three options:
> > > 1. Run all loop opts based on the loop optimizer before going into SSA. 
> > > (Bad)
> > > 2. Run all loop opts based on the loop optimizer after going out of SSA 
> > > (Bad, since we want to lower at some point to allow the optimizers work 
> > > remove redundant address arithmetic and whatnot)
> > > 3. Go into and out of SSA multiple times (expensive).
> > 
> > I don't see a problem with keeping the ssa form (although not
> > neccesarily the minimal one).
> > 
> I think Daniel is referring to not just keeping the program in SSA form,
> but *using* the SSA form to do the analyses.

currently my code uses ud chains; obviously ssa can be used instead of
them.

Zdenek


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