This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: frv: asm reloads for pseudos assigned to memory fail
- From: Joern Rennecke <joern dot rennecke at superh dot com>
- To: aoliva at redhat dot com (Alexandre Oliva)
- Cc: joern dot rennecke at superh dot com (Joern Rennecke), rth at redhat dot com (Richard Henderson), gcc-patches at gcc dot gnu dot org
- Date: Fri, 30 Jul 2004 12:24:50 +0100 (BST)
- Subject: Re: frv: asm reloads for pseudos assigned to memory fail
> > and you can avoid inadvertantly using a load in the non-PIC case by making
> > the load alternatives expensive enough.
>
> Bzzzt! Wrong. :-)
>
> Making the alternatives expensive enough will just make it more
> difficult for them to be chosen. It won't avoid it entirely, unless
> you add so many `?'s that it becomes equivalent to `!' as far as
> reload is concerned.
As I said before, frv_cannot_force_const_mem already takes care of the
PIC problem, so the remaining issues are just performance.
That being said, there are some sub-optimal paths that reload could
theoretically take in the SH port, but I haven't ever seen them taken
once the right cost cues and logic were in place.
At any rate, having separate load patterns will probably cost you more
in performance than it can possibly save you by avoiding stray
force_const_mem, since reload inheritance can only work well when
you have your move patterns written properly.