Re: [linus:master] [x86] adfcf4231b: blogbench.read_score -10.9% regression

[Linus Torvalds]

On Tue, May 9, 2023 at 5:30 AM Feng Tang <feng.tang@xxxxxxxxx> wrote:
>
> > At a *minimum* that benchmark should have different threads using
> > different read() buffers.
>
> We tried a debug patch which allocates a dedicated buffer for
> each reader thread, run it on the same Cacade Lake platform, and
> the regression is _gone_, after the noise of cache false sharing is
> reduced.

Thanks for verifying.

I was pretty sure that was the issue, but it's always good to
double-check that there wasn't something else going on,

                 Linus