[Bug tree-optimization/32044] [4.3/4.4 Regression] final value replacement too aggressive for e.g. targets with no native div/mod insns

[rakdver at kam dot mff dot cuni dot cz]

------- Comment #57 from rakdver at kam dot mff dot cuni dot cz  2008-12-10 22:02 -------
Subject: Re:  [4.3/4.4 Regression] final value replacement too aggressive for
e.g. targets with no native div/mod insns

> I think Zdenek is right in comment #54: We should reincarnate
> expression_expensive_p in some form such that it takes into account (or makes a
> reasonable guess about) which instructions are particularly expensive. 
> Apparently, the old cost check was too conservative (it tested the expression
> cost against spill cost, ignoring further optimization opportunities).  So we
> need to test for something else than before.
> 
> Maybe we should not replace if the final value is not a constant?

That is way too conservative (replacing the final values like say "n + 1"
should usually be profitable).

> Or if it is an expression involving div/mod? 

That is what I currently plan to do.

> Can we test somehow how many insns the final
> value replacement computation would take

I think that has the same problems as the original solution (it is
not possible to find a bound that would both prevent expensive things to be
moved out and to allow the non-expensive ones, as the exact costs are
target-specific and we are not estimating them precisely enough).

> (maybe re-use the inliner heuristics)?

That would be difficult, as estimate_num_insns works on gimple
statements, not on expressions.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32044