Re: [PATCH v8 6/6] elf: Optimize _dl_new_hash in dl-new-hash.h

[Noah Goldstein <goldstein.w.n@gmail.com>]

On Mon, May 16, 2022 at 11:44 AM Siddhesh Poyarekar <siddhesh@gotplt.org> wrote:
>
> On 16/05/2022 22:08, Noah Goldstein wrote:
> >> Thanks for the context, this should go into comments.  A wishlist bug
> >> would be nice but I suspect it'll just gather dust.  Maybe it's still
> >> useful for someone coming in after 10-15 years looking for more context
> >> on it.
> >
> > I'll add a comment in the next version.
>
> Thanks.
>
> >> I would prefer the earlier variant in generic code, with (if necessary)
> >> the scheduling hack being a sysdep for x86.  Other architectures that
> >> want to use the latter should #include it and also post microbenchmark
> >> results so that we keep track of how we arrived at that decision.
> >
> > I'm happy to switch it back but I don't think the scheduling hack is x86
> > oriented. I don't think re-ordering could ever de-optimize things.
> > The only real architectural assumption is a reasonably fast
> > 32-bit multiply which is true for both the more generic earlier version
> > and the current one.
>
> I don't entirely disagree, but I think the conservative stance here is
> to keep the scheduling hack in the sysdep that has actually been tested
> and then bring it out into generic if it has been experimentally
> verified to be a universal win for all architectures we support.  That
> is, if we find that every architecture is including the sysdep version,
> then it's time to bring it out to replace the generic version.
>
> FWIW, I'm lowering the bar for acceptance because you only need to
> verify that the scheduling hack is better for the architecture you're
> interested in, not all architectures we support :)

Made the asm statements arch specific in V9.

>
> Thanks,
> Siddhesh