[PATCH] match: Add simplification of `(a*zero_one_valued_p) & b` if `a & b` simplifies [PR119402]

[Andrew Pinski]

On Tue, Jan 13, 2026 at 12:20 PM Jeffrey Law
<jeffrey.law@oss.qualcomm.com> wrote:
>
>
>
> On 1/13/2026 10:39 AM, Andrew Pinski wrote:
> > This is a small reassociation for `a*bool & b` into `(a & b) * bool` checking if
> > `a & b` simplifies. Since it could be the case `b` is `~a` or `a` or something
> > else that might simplify when anding with `a`.
> >
> > Note this fixes a regression for aarch64 where the cost of a multiply vs `&-` changed
> > in GCC 14 and can no longer optimize some cases at the RTL level.
> >
> > Bootstrapped and tested on x86_64-linux-gnu.
> >
> >       PR tree-optimization/119402
> > gcc/ChangeLog:
> >
> >       * match.pd (`(a*zero_one_valued_p) & b`): New pattern.
> >
> > gcc/testsuite/ChangeLog:
> >
> >       * gcc.dg/tree-ssa/bitops-14.c: New test.
> >       * gcc.dg/tree-ssa/bitops-15.c: New test.
> Last I looked we didn't have scalar support for a & ~b as a named
> pattern,  but it seems like that's another good reason (at some point)
> to trying and bring the logical terms together in general since multiple
> ISAs have a full complement (pun intended) of logical-with-complement
> style operations.

GCC does have an internal function for `a&~b` though and it works with
the scalar modes with the optabs.
That is one thing I made sure worked as I had to rename the optab to
andn from andc (due to csi being the complex si mode and being
matched).
https://inbox.sourceware.org/gcc-patches/20240724024953.3118043-1-quic_apinski@quicinc.com/

At the time I was looking into use the internal function but this was
before fold_before_rtl_expansion_p/PROP_last_full_fold was around.
This is something I will be looking into finishing up during stage 1 though.

Thanks,
Andrew

>
> And more generally, there are guidelines (from an ancient HP paper) on
> reassociation and this would seem to generally fit.  An AND operation
> would largely be similar to PLUS and would have a different priority
> than MULT.  So when you rewrite in the way you're doing you'll generally
> tend to improve CSE, LICM and IVopts.  It's been eons since I studied
> that work, so don't ask for more details :-)
>
> Anyway, OK for the trunk.  Thanks.
>
> jeff
>
>