On 15 December 2014 at 19:37, Mike Holmes <[email protected]> wrote:
>
>
> On 15 December 2014 at 08:26, Ola Liljedahl <[email protected]>
> wrote:
>>
>> On 25 November 2014 at 21:07, Mike Holmes <[email protected]> wrote:
>> >
>> >
>> > On 25 November 2014 at 12:03, Ola Liljedahl <[email protected]>
>> > wrote:
>> >>
>> >> Signed-off-by: Ola Liljedahl <[email protected]>
>> >
>> >
>> >
>> > Works on x86 - reported the correct number of cores available
>> >
>> > http://www.gnu.org/software/libc/manual/html_node/Processor-Resources.html
>> >
>> > So I assume it fixes the big.LITTLE hang, however the API is called
>> > sysconf_core_count and the meager docs say it is "Core count". We
>> > might
>> > want to name this sysconf_avalable_core_count if the difference is going
>> > to
>> > matter in the future.
>> >
>> > I think this patch should improve the docs to clarify that it is the
>> > currently available core count even if we don't rename or add a new API.
>> Hasn't this API been better document in some other patch? I just
>> updated my repo and the description seems not to have changed:
>> * @return Core count
>>
>> I think this patch is still relevant. As ODP cannot make cores online
>> (e.g. hotplug cores), it only makes sense to report (the number of)
>> cores already online and available to ODP instead of some potential
>> number of cores.
>>
>> Do I need to resubmit the patch with a description of it reporting
>> "the number of online cores" and update the header file as well?
>
>
> As the commenter on the this patch, you and I need to agree on how to get it
> in.
I make the patch, you approve it, Maxim merges it.
> I would like to see this fix in and the docs updating to match because
> getting the doc updated is proving hard.
The name of the function (odp_core_count) is also wrong. We are
interested in what Linux calls CPU's. They map to HW threads if the
hardware uses such (and they are enabled). I think we should reuse the
term Linux uses since we seem not to be able to find something better.
>
>>
>> >
>> >>
>> >> ---
>> >> On ARM big.LITTLE systems with cluster switching, return the actual
>> >> number
>> >> of
>> >> currently usable CPU's (e.g. 4, not 8). Trying to bind threads to cores
>> >> 4..7
>> >> will just hang those threads.
>> >>
>> >> platform/linux-generic/odp_system_info.c | 2 +-
>> >> 1 file changed, 1 insertion(+), 1 deletion(-)
>> >>
>> >> diff --git a/platform/linux-generic/odp_system_info.c
>> >> b/platform/linux-generic/odp_system_info.c
>> >> index 10665bb..60804cb 100644
>> >> --- a/platform/linux-generic/odp_system_info.c
>> >> +++ b/platform/linux-generic/odp_system_info.c
>> >> @@ -51,7 +51,7 @@ static int sysconf_core_count(void)
>> >> {
>> >> long ret;
>> >>
>> >> - ret = sysconf(_SC_NPROCESSORS_CONF);
>> >> + ret = sysconf(_SC_NPROCESSORS_ONLN);
>> >> if (ret < 0)
>> >> return 0;
>> >>
>> >> --
>> >> 1.9.1
>> >>
>> >>
>> >> _______________________________________________
>> >> lng-odp mailing list
>> >> [email protected]
>> >> http://lists.linaro.org/mailman/listinfo/lng-odp
>> >
>> >
>> >
>> >
>> > --
>> > Mike Holmes
>> > Linaro Sr Technical Manager
>> > LNG - ODP
>
>
>
> --
> Mike Holmes
> Linaro Sr Technical Manager
> LNG - ODP
_______________________________________________
lng-odp mailing list
[email protected]
http://lists.linaro.org/mailman/listinfo/lng-odp
