[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <dzohdvfrh3jivw5mt7aepxezzrvq422xptx5zyz4mouxypnt5v@hj75gj622i6x>
Date: Thu, 4 Jul 2024 18:42:09 +0200
From: Uwe Kleine-König <u.kleine-koenig@...libre.com>
To: Andrew Morton <akpm@...ux-foundation.org>
Cc: Nicolas Pitre <nico@...xnic.net>, linux-kernel@...r.kernel.org,
Biju Das <biju.das.jz@...renesas.com>
Subject: Re: [PATCH v2 2/2] mul_u64_u64_div_u64: basic sanity test
Hello,
On Wed, Jul 03, 2024 at 11:36:31PM +0200, Uwe Kleine-König wrote:
> I intend to do a performance test of Nico's code. I hope I get to that
> tomorrow.
I'm really surprised, I expected a penalty for being more correct, but
there doesn't seem to be one.
I tested on an stm32mp135f SoC (i.e. armv7) with:
time pwmtestperf -p 3 -P 1000000000 -S 1000
which configures the PWM 1000001 times, each configuration calls
mul_u64_u64_div_u64() three times. And I got
real 0m 37.17s
user 0m 0.69s
sys 0m 36.40s
on 6.10.0-rc1 (+ various pwm related patches) and
real 0m 36.93s
user 0m 0.58s
sys 0m 36.26s
with Nico's patch applied on top.
Looking at the patch in detail, I always hit ilog2(a) + ilog2(b) <= 62,
so on 2nd thought it's not surprising that the performance is similar.
Tested-by: Uwe Kleine-König <u.kleine-koenig@...libre.com>
Thanks!
Uwe
Download attachment "signature.asc" of type "application/pgp-signature" (489 bytes)
Powered by blists - more mailing lists