cause of all the mainline Sparc regressions

[David S. Miller]

   From: Jan Hubicka <jh@suse.cz>
   Date: Thu, 23 May 2002 19:02:35 +0200

   > I think it makes the most sense to have rest_of_compilation set
   > a global variable that has CLEANUP_PRE_SIBCALL, CLEANUP_PRE_LOOP,
   > CLEANUP_POST_REGSTACK, and whatever new one Dave needs.  Then
   > cleanup_cfg or's that into the mode that it is given.
   
   Hmm, I original had global variable to control these, perhaps re-adding
   it makes sense.

It looks like we're still figuring out the best way to implement this,
I'll hold off until this is resolved.

I personally think just calling delete_unreachable_blocks directly
from the EH code instead of doing cleanup_cfg invocations is fine.
But Richard may feel otherwise.