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]

Re: bounds-check intentionally undocumented?


"John Hughes" <john@Calva.COM> writes:

> > > >   make gcc treat auto BPs as register-allocable triples, rather
> > > >   than its current behavior of forcing them (as it does most structs)
> > > >   into stack slots.
> >
> > > Will this optimisation be specific to fat pointers, or available to
> > > other structs?
> > 
> > Doing it for all structs is the only sensible way to proceed, IMO.
> 
> Magic.  This should help some of my code *LOTS*.
> 
> Is it hard to do? 

Probably.  I haven't looked very deeply into it yet, but since people
have been talking about it for years, and it hasn't been done yet, I
can only assume that it's difficult.  If someone who has thought about
the problem could give us a design proposal, that would be very useful
to me and to the community.

Greg

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