2023-11-01 22:58:48

by Yury Norov

Subject: [PATCH 0/4] lib/group_cpus: rework grp_spread_init_one() and make it O(1)

grp_spread_init_one() implementation is sub-optimal now because it
traverses bitmaps from the beginning, instead of picking from the
previous iteration.

Fix it and use find_bit API where appropriate.

Yury Norov (4):
cpumask: introduce for_each_cpu_and_from()
lib/group_cpus: relax atomicity requirement in grp_spread_init_one()
lib/group_cpus: optimize inner loop in grp_spread_init_one()
lib/group_cpus: optimize outer loop in grp_spread_init_one()

include/linux/cpumask.h | 11 +++++++++++
include/linux/find.h | 3 +++
lib/group_cpus.c | 29 ++++++++++++-----------------
3 files changed, 26 insertions(+), 17 deletions(-)

--
2.39.2

2023-11-01 22:58:54

by Yury Norov

Subject: [PATCH 4/4] lib/group_cpus: optimize outer loop in grp_spread_init_one()

Similarly to the inner loop, in the outer loop we can use for_each_cpu()
macro, and skip cpus that have been copied.

With this patch, the function becomes O(1), despite that it's a
double-loop.

Signed-off-by: Yury Norov <[email protected]>
---
lib/group_cpus.c | 12 ++++--------
1 file changed, 4 insertions(+), 8 deletions(-)

diff --git a/lib/group_cpus.c b/lib/group_cpus.c
index 8afe0c71d204..2442b38cbf37 100644
--- a/lib/group_cpus.c
+++ b/lib/group_cpus.c
@@ -17,16 +17,11 @@ static void grp_spread_init_one(struct cpumask *irqmsk, struct cpumask *nmsk,
const struct cpumask *siblmsk;
int cpu, sibl;

- for ( ; cpus_per_grp > 0; ) {
- cpu = cpumask_first(nmsk);
-
- /* Should not happen, but I'm too lazy to think about it */
- if (cpu >= nr_cpu_ids)
- return;
-
+ for_each_cpu(cpu, nmsk) {
__cpumask_clear_cpu(cpu, nmsk);
__cpumask_set_cpu(cpu, irqmsk);
- cpus_per_grp--;
+ if (cpus_per_grp-- == 0)
+ return;

/* If the cpu has siblings, use them first */
siblmsk = topology_sibling_cpumask(cpu);
@@ -37,6 +32,7 @@ static void grp_spread_init_one(struct cpumask *irqmsk, struct cpumask *nmsk,
__cpumask_set_cpu(sibl, irqmsk);
if (cpus_per_grp-- == 0)
return;
+ cpu = sibl + 1;
}
}
}
--
2.39.2

2023-12-04 22:01:36

by Yury Norov

Subject: Re: [PATCH 0/4] lib/group_cpus: rework grp_spread_init_one() and make it O(1)

Ping?

On Wed, Nov 01, 2023 at 03:58:16PM -0700, Yury Norov wrote:
> grp_spread_init_one() implementation is sub-optimal now because it
> traverses bitmaps from the beginning, instead of picking from the
> previous iteration.
>
> Fix it and use find_bit API where appropriate.
>
> Yury Norov (4):
> cpumask: introduce for_each_cpu_and_from()
> lib/group_cpus: relax atomicity requirement in grp_spread_init_one()
> lib/group_cpus: optimize inner loop in grp_spread_init_one()
> lib/group_cpus: optimize outer loop in grp_spread_init_one()
>
> include/linux/cpumask.h | 11 +++++++++++
> include/linux/find.h | 3 +++
> lib/group_cpus.c | 29 ++++++++++++-----------------
> 3 files changed, 26 insertions(+), 17 deletions(-)
>
> --
> 2.39.2