On Tue, Sep 6, 2022 at 4:01 PM Robin Dapp <rd...@linux.ibm.com> wrote:
>
> > cost might also depend on the context in case flag setting
> > behavior differs for xor vs sub (on x86 sub looks strictly more
> > powerful here). The same is probably true when looking for
> > a combination with another bitwise operation.
> >
> > Btw, why not perform the optimization in expand_binop? That
> > for example already does
> >
> > if (binoptab == sub_optab && CONST_INT_P (op1))
> > {
> > op1 = negate_rtx (mode, op1);
> > binoptab = add_optab;
> > }
> >
> > alternatively a targets expander can do the selection as well.
>
> I was under the impression optabs/expand_binops is only supposed to
> "optimize" when it's clear that it is an optimization/canonicalization.
> I didn't see other functions there trying two alternatives and also none
> seems to use range information already.
>
> Regarding the proper costing (including the surroundings): is it even
> possible to encompass everything in such a localized decision? A
> target's expander decision would also not take this into account when
> deciding? If so, should we not perform this conversion generally and not
> only target specifc?
The question is really whether xor or sub is "better" statically. I can't
think of any reasons. On s390, why does xor end up "better"?
Richard.
>
> Regards
> Robin
