On Sat, Mar 16, 2024 at 2:40 AM Nathan Bossart <nathandbossart@gmail.com> wrote:
>
> On Fri, Mar 15, 2024 at 12:41:49PM -0500, Nathan Bossart wrote:
> > I've also attached the results of running this benchmark on my machine at
> > HEAD, after applying 0001, and after applying both 0001 and 0002. 0001
> > appears to work pretty well. When there is a small "tail," it regresses a
> > small amount, but overall, it seems to improve more cases than it harms.
> > 0002 does regress searches on smaller arrays quite a bit, since it
> > postpones the SIMD optimizations until the arrays are longer. It might be
> > possible to mitigate by using 2 registers when the "tail" is long enough,
> > but I have yet to try that.
>
> The attached 0003 is a sketch of what such mitigation might look like. It
> appears to help with the regressions nicely. I omitted the benchmarking
> patch in v3 to appease cfbot.
I haven't looked at the patches, but the graphs look good.