This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: Non-dominating loop bounds in tree-ssa-loop-niter 3/4
On Wed, 31 Oct 2012, Jan Hubicka wrote:
> > On Wed, Oct 31, 2012 at 01:30:02PM +0100, Richard Biener wrote:
> > > On Wed, 31 Oct 2012, Jan Hubicka wrote:
> > > > > unroll you mean. Because unrolling mutates the CFG too much.
> > > > > Well - it was just a starting point, populating -Og with as little
> > > > > as possible and 100% profitable transforms (in both debug and speed
> > > > > metric). In late opts we only do (early opt queue is shared):
> > > >
> > > > Well, and what about early cunrolli?
> > >
> > > It's not there (and I would disable it for -Og).
> >
> > Generally, most of the loop transforms are undesirable for -Og.
>
> Hmm, do we have some guidelines what is and is not desirable for -Og?
> Unrolling/header copying being code duplicating optimization do not really
> throw away any info by themself only require debugger to understand that some
> code may get duplicated, but it is not different from i.e. inlining. Of course
> the subsequent const prop will make it impossible to write into the iv
> variable, but do we want to support reliable writting into user vars (i.e.
> disable cprop on user vars) at -Og?
No, we don't support reliable writing.
We should first sort out issues with the current passes and debug info,
look at compile-time and runtime performance and then add new passes.
Richard.