This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: Patch for PR16967
On Tuesday 10 August 2004 22:19, Zack Weinberg wrote:
> Steven Bosscher <stevenb@suse.de> writes:
> > In fact LABEL_REFs too. Fun. I think I'll see what the effect
> > is of the following patch :-/
>
> That seems a sounder way to go about this than an ever-out-of-sync
> list of RTL constants.
>
> I do share the concerns of "f" (on IRC) about its being desirable to
"f" is Andi Kleen in real life ;-)
> GCSE expensive constant loads. Perhaps this is another place we ought
> to be looking at RTX_COSTS?
Yup.
> If we did this we would want to have
> antisymmetric checks in GCSE and CPROP so that they both agreed which
> one was responsible for any given expression.
Right. The basic problem as it is now is that CPROP propagates
the constant, then PRE un-propagates it, and then CPROP propagates
it again. CPROP always wins because it runs last.
In the mean time I'm just going to try and see what the patch does
to the code we generate with/without PIC. Perhaps it's a non-issue.
Gr.
Steven