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] dead const/pure/alloca call removal

> On Sat, Nov 08, 2003 at 09:51:08PM +0100, Jan Hubicka wrote:
> > This has a problem.  We first gimplify functions, then analyze them and
> > mark as CONST/PURE and then optimize.  So we lose the information.
> > 
> > Wehre do you think this flag should be updated?
> 
> I don't unserstand why we'd lose this during gimplification.
> Please explain.  It definitely sounds like if we lose the bit
> then we've got a trivial bug somewhere.

We don't loose it, but we first gimplify, then realize that calee is
const (during it's RTL expansion, but soon from the trees in global
cgraph optimization pass) and then optimize the caller with bit still
set to side effects.  Thus we need to re-consider it.

Honza
> 
> 
> r~
Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]