On Wed, Jun 17, 2015 at 03:02:03PM +0200, Stefan Agner wrote:

On 2015-06-09 22:07, Sebastian Andrzej Siewior wrote:

quoted

yeah, do you have any numbers by chance?

Just reevaluated the "performance optimizations". On a VF610 SoC (with
L2 cache) the improvements increase performance by 2.5% and below. On a
VF500 SoC (without L2 cache) it seems to have slightly more impact, up
to 4%. Overall, it seems to influence write more than read.

it could have something to do with the CPU clock and how long it is blocked
due to the sync operation.

Back then, when I implemented the improvements it certainly had a bigger
impact. I don't have strong opinion on that...

Thanks for doing that. If you could take those numbers, put them in a table
like with/without relaxed and so one, write how you got them (modprobe bla)
and make it part of the commit message then everybody could see about how much
we talk here. Thanks again.

--
Stefan

Sebastian