XFAIL const-elim-1.c on 32-bit PA

[Mark Mitchell]

Zack Weinberg wrote:

> const-elim-1.c has the same problem on 32-bit PA as it does on ARM and
> Xtensa: the 'S' temporary is initialized with a block move which the
> RTL optimizers cannot eliminate.  Making them smarter is just not in
> the cards.  Accordingly, I XFAIL this.  The pattern is intended to match
> "hppa2.0w-hp-hpux" but not "hppa64-hp-hpux" and does seem to do the job.
> 
> I've successfully bootstrapped and tested this on both these targets,
> 3.4 branch.  Unfortunately I cannot bootstrap any PA target on
> mainline right now.  I suspect that by the time I get round to fixing
> that, tree-ssa will have landed, and it is likely to squelch the
> underlying problem.  Therefore I am proposing this patch for 3.4
> branch ONLY at this time.  Mark, OK?

The regexp seems a little subtle, but the comment in the code makes it 
relatively understandable.

I think you should put this on the mainline too; when tree-ssa fixes 
this magically, it will start XPASSing, but in case, for some reason, it 
does not, it should stay XFAILed.

In any case, it's OK for 3.4.

Thanks,

-- 
Mark Mitchell
CodeSourcery, LLC
mark@codesourcery.com