RFC: fix think-o in tree-ssa-loop-ivopts.c

[Eric Botcazou]

> The "cost" is never referred to again in the fallback code, so it seems
> like deleting the return statement was an unintended mistake.

Indeed, how embarassing. :-(  This affects 4.5.x as well.

> So, I'm wondering:  Eric, were the runtime speed improvements you claimed
> you saw here
>
> http://gcc.gnu.org/ml/gcc-patches/2009-05/msg01858.html
>
> actually present in the checked-in version of the patch, where the result
> of the fancy cost computation is simply discarded, or in some other version
> of the patch that implemented what I presume was the intended behavior?

Probably in both versions, at least partially.  The main effect (saving 1 
register for loops with non-constant iteration origin) is always present.

> Do we know for sure that the fancy cost computation actually even an
> improvement over the fallback case?

The "fancy" computation was an improvement, yes.

> 2010-06-09  Sandra Loosemore  <sandra@codesourcery.com>
>
> 	gcc/
> 	* tree-ssa-loop-ivopts.c (get_computation_cost_at): Return the
> 	computed cost.

I'd add "again".  Thanks for spotting this.

-- 
Eric Botcazou