Re: [patches] Re: ARM libstdc++ failure fix

[Jan Hubicka <jh at suse dot cz>]

> On Sun, Jul 29, 2001 at 10:02:51PM +0200, Jan Hubicka wrote:
> > This way we keep the original behaviour.  The comment is out of date,
> > can you suggest better wording?
> 
> Hmm.  I suggest just stopping, rather than aborting.
Thats what I do, if the boundary == code_label
(as it should be the case of gcse - otherwise we hit label inside basic block).
> 
> > Whats about the Graham's idea of checking CALL_INSN too?
> 
> Probably a good idea.  Would be helpful on targets whose argument
> registers overlap call-return registers.  For ppc, something like
> 
> 	foo(bar())
> 
> you could wind up with a noop move that gets removed.  So there
> _is_ no independant set of r3.
OH, I see, so I will add code to stop on CALL_INSN too.
> 
> > Choose ppc, mips, sparc (or less prefebraly arm or sh) :)
> 
> Any of them.  The fastest.  ;-)
Hard question - I will need to benchmark first.
> 
> > OK after checking?
> 
> Yes.
I will send updated version in momentarity and except it approved - you
have 20 minutes to cancel :)

Honza