[PATCH]: Handle lonely CC0 setter in sched-deps.c

Maxim Kuvyrkov maxim@codesourcery.com
Thu Apr 17 18:41:00 GMT 2008


Hans-Peter Nilsson wrote:
> On Mon, 7 Apr 2008, Maxim Kuvyrkov wrote:
>> Paolo Bonzini wrote:
>> Here's a proposed patch for seqabstr pass.  The patch implements conservative
>> fix to not collect cc0 instructions into sequences for abstraction.  This is
>> done similarly to handling live stack regs.  Any comments?
> 
> This looks much too conservative.  I had a loook and ISTM it'd
> be better to fix rtl-factoring.c:matching_length to make sure
> cc0 setter and users (must be the next_nonnote_insn) are *both*
> (or neither) added to such a sequence.

I agree, but.  I don't think it's worth adding any extensive support for 
CC0, which is being deprecated.

> 
> Can you please enter a bugzilla report for your original bug?
> I'll have a look as I'm interested in making scheduling work for
> CRIS (currently a cc0 target).

The bug occurs only with the patch for scheduling for ColdFire, which is 
not yet committed to mainline.  So submitting the bug report would be 
rather meaningless.


Thanks,

Maxim



More information about the Gcc-patches mailing list