Hi all,
I still have these patches sitting in my queue and figured I'd repost them.
Last time Linus proposed a "+m" alternative approach, but that generates far
far worse code (I've lost the patch and not re-ran those numbers, but I suppose
I can redo if found important).
These patches have been through 0day for a while.
> On Jun 13, 2019, at 6:54 AM, Peter Zijlstra <[email protected]> wrote:
>
> Hi all,
>
> I still have these patches sitting in my queue and figured I'd repost them.
>
> Last time Linus proposed a "+m" alternative approach, but that generates far
> far worse code (I've lost the patch and not re-ran those numbers, but I suppose
> I can redo if found important).
I remember I tried it as well and got the same results.
> These patches have been through 0day for a while.
Finally, I would not need to cache smp_processor_id() on the stack when it
is used multiple times…