Improve unrolled size estimates
Jan Hubicka
hubicka@ucw.cz
Tue May 12 09:11:00 GMT 2009
> On Mon, 11 May 2009, Jan Hubicka wrote:
>
> > Hi,
> > it turned out that there is some extra testsuite compensation neccesary
> > since in vectorizer testsuite we now tend fully unroll internal loops
> > preventing vectorization. In some cases unrolling makes sense, in some
> > cases we make code size growth because we account loads and stores to
> > have size of 0 I will fix by followup patch. Still it seems to make
>
> Stores have size 0? They never should ;)
Yes, both load and stores are estimated to size of 0 :) It was so
because metric was originally intended to only drive inliner and I
conlcuded that most of load/stores will be optimized away later anyway
so it makes sense to only count the real stuff done in the function.
Honza
>
> Richard.
More information about the Gcc-patches
mailing list