On Mon, Nov 20, 2023 at 07:54:54AM +0000, Richard Biener wrote:
> On Fri, 17 Nov 2023, Jakub Jelinek wrote:
> > Per the earlier discussions on this PR, the following patch folds
> > popcount (x) == 1 (and != 1) into (x ^ (x - 1)) > x - 1 (or <=)
> > if the corresponding popcount optab isn't implemented (I think any
> > double-word popcount or call will be necessarily slower than the
> > above cheap 3 op check and even for -Os larger or same size).
> >
> > I've noticed e.g. C++ aligned new starts with std::has_single_bit
> > which does popcount (x) == 1.
> >
> > As a follow-up, I'm considering changing in this routine the popcount
> > call to IFN_POPCOUNT with 2 arguments and during expansion test costs.
> >
> > Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
>
> Classically this would have been an RTL expansion alternative, given
> we want to do less of those the next place would have been the ISEL
> pass. Any particular reason you chose widening-mul for this (guess
> that pass just has a bad name and it's the effective "optimize" ISEL pass
> we have).
I think the ssa-math-opts pass does far more of this staff than the isel
pass which only deals with vector stuff right now and you've even mentioned
that pass for that in the PR90693 thread.
That said, I can move it into the isel pass as well if you prefer that.
Jakub
