Re: [RFB] IRA and regclass both ignore '!' attribute

[Vladimir Makarov <vmakarov at redhat dot com>]

Andrew Pinski wrote:
> I noticed while looking into a register allocation issue in 4.1.1,
> that regclass was ignoring '!' attribute so it would sometimes give
> the wrong result so we get performance issue (selecting GPR instead of
> FPR).  So I saw that IRA also ignore it too which seems wrong and
> seems like the code was copied from regclass.  preprocess_constraints
> and find_reloads both take '!' into account.
>
> Can someone benchmark this on x86 and PPC? I can only really benchmark
> with a 4.3 or 4.1.1 based compiler with a subset of SPEC.  The 4.1.1
> based compiler gets a nice but small 0.6% performance increase and a
> small code decrease.
>
> Thanks,
> Andrew Pinski
>
> ChangeLog:
> * regclass.c (record_reg_classes): Take into account '!'.
> (record_reg_classes): Likewise.
>   
Andrew, thanks for looking into RA problems.  I think your patch should 
be thoroughly tested not only for performance.  Such changes  might 
break some ports.  I tried this change when I worked on YARA project and 
that time the result was negative.   Therefore it is not used in IRA.  I 
think that there were some reasons for ignoring this in regclass too.  
But may be I am wrong.

The problem you mentioned (selecting GPR instead of FPR) could be solved 
by putting one or more ? in the alternative to get the same effect you 
are trying to achieve with this patch.  Imho, the patch you are 
proposing probably decrease some gcc functionality. Probably ! was 
supposed to be used only for reload.

But if you gets an approval you should change the documentation (for ! 
constraint) too.