According to the document linked the following approach is even faster
than the one I used due to CPU parallelization:
page[1].compound_nr = ( shift & ~shift) | (-order & shift);
for(int x =0; x< 11;x++){
unsigned int order = x;
unsigned long shift = 1U << order;
printf("order %d output : %lu\n", order, ( shift & ~shift) |
(-order & shift));
}
order 0 output : 0
order 1 output : 2
order 2 output : 4
order 3 output : 8
order 4 output : 16
order 5 output : 32
order 6 output : 64
order 7 output : 128
order 8 output : 256
-- Nico
On Tue, Dec 13, 2022 at 4:53 PM Nico Pache <npache@xxxxxxxxxx> wrote:
>
> Hi Mike,
>
> Thanks for the pointer! Would the branchless conditional be an
> improvement over the current approach? I'm not sure how hot this path
> is, but it may be worth the optimization.
>
> -- Nico
>
> On Tue, Dec 13, 2022 at 4:48 PM Mike Kravetz <mike.kravetz@xxxxxxxxxx> wrote:
> >
> > On 12/13/22 16:45, Nico Pache wrote:
> > > Since commit 1378a5ee451a ("mm: store compound_nr as well as
> > > compound_order") the page[1].compound_nr must be explicitly set to 0 if
> > > calling set_compound_order(page, 0).
> > >
> > > This can lead to bugs if the caller of set_compound_order(page, 0) forgets
> > > to explicitly set compound_nr=0. An example of this is commit ba9c1201beaa
> > > ("mm/hugetlb: clear compound_nr before freeing gigantic pages")
> >
> > There has been some recent work in this area. The latest patch being,
> > https://lore.kernel.org/linux-mm/20221213212053.106058-1-sidhartha.kumar@xxxxxxxxxx/
> >
> > --
> > Mike Kravetz
> >
