Profiling on S390

[Ulrich Weigand]

Richard Henderson wrote:

>It could, but if you don't have add/load/store insns that don't
>clobber CC, then you'll get in trouble with reload too.

We can load/store without clobbering CC; while we don't have
a general addition operation that does not clobber CC, we do
have a load-address operation, which is all reload actually
needs (if things are set up correctly).  However, LA is a 31-bit
addition and cannot be used to implement 32-bit or even 64-bit
integer arithmetic.

However, it is certainly true that the use of gen_add2_insn in
reload is really broken as it doesn't care at all whether CC
gets clobbered.  It is only by setting things up so that reload
actually never tries to use gen_add2_insn that things work out ...

>This is all about a movstrsi or cmpstrsi pattern?  Perhaps it
>would be better to keep this as one unit until after reload?

It's only about cmpstrsi.  The idea behind exposing the loop
early was to potentially benefit from loop unrolling ...


Mit freundlichen Gruessen / Best Regards

Ulrich Weigand

--
  Dr. Ulrich Weigand
  Linux for S/390 Design & Development
  IBM Deutschland Entwicklung GmbH, Schoenaicher Str. 220, 71032 Boeblingen
  Phone: +49-7031/16-3727   ---   Email: Ulrich.Weigand@de.ibm.com