[Bug tree-optimization/40768] PRE backfires when given a sequence of "if" statements

[amylaar at gcc dot gnu dot org]

------- Comment #5 from amylaar at gcc dot gnu dot org  2009-07-16 00:11 -------
(In reply to comment #3)
> Hi Steven,
> 
> Maybe I'm missing something, but what do patches talking about
> SMALL_REGISTER_CLASSES have to do with this issue?  For ARM, the registers
> involved are general purpose and ! SMALL_REGISTER_CLASSES.  There's about 14 of
> them available, and these PHIs want a virtually unbounded number of registers. 
> And we are not optimizing for size (-O3).  So I can't see how your patch is
> going to help.
> 
> Don't get me wrong, I'm happy to close it again if they will :-)  But I don't
> think either of the checks in want_to_pre_p will apply.

Look at comment #14.  My patch attempts to do a cost-benefit analysis
of phi-node (with implied copies) insertion versus the saved arithmetic.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40768