On Thu, Apr 19, 2018 at 10:21:32AM +0200, Richard Biener wrote:
>
> The following fixes the fact that the vectorizer doesn't bother to
> preserve restrict information on its generated loads/stores which
> in turn causes missed predictive commonings. The regression happened
> because there's now a IPA-CP clone which wrecks points-to info,
> but restrict is properly preserved. But with the vectorizer
> throwing it away we lost all of the info.
>
> Bootstrap and regtest running on x86_64-unknown-linux-gnu.
>
> Dependend on how close we are to a RC I consider to fix this
> only for GCC 8.2.
I'd put it in now, there is still work to do on the C++ P1s.
> 2018-04-19 Richard Biener <rguent...@suse.de>
>
> PR tree-optimization/84737
> * tree-vect-data-refs.c (vect_copy_ref_info): New function
> copying restrict info.
> (vect_setup_realignment): Use it.
> * tree-vectorizer.h (vect_copy_ref_info): Declare.
> * tree-vect-stmts.c (vectorizable_store): Copy ref info from
> the first DR to all generated stores.
> (vectorizable_load): Likewise for loads.
Jakub
