On Thu, Feb 16, 2023 at 02:21:04PM +0000, Richard Biener wrote:
> > That said, while this patch fixes the regression by allowing DSE of
> > IFN_DEFERRED_INIT again, I think we probably have some latent bug in FRE
> > where without this patch it seems to be fre5 that sees one unconditional
> > c = 1; store, one conditional c = 0; store and in the last bb before return
> > another c = 1; store and decides that the last store is redundant, which is
> > not the case, the first two stores are redundant or if they can't be
> > removed, none of them is. Richard, could you please have a look?
>
> That's before this patch only? I'll have a look.
Yes.
Jakub
