I just did a minor bug-fix release of PLPA v1.3.1 (an off-by-one error
in the plpa-taskset executable). Hopefully, it'll be the last-ever
PLPA release. I have updated the PLPA web site to have big red
notices that it will soon be deprecated in favor of hwloc.
I also filed https://svn.open
Hi,
this is my first post on this mailing list -- I just found hwloc and it
seems like a useful tool for my research.
As an initial test I ran hwloc-info on our 8 socket Opteron 8356 system
(32 cores), the full results are attached below. The node information is
correct, as is the level 2 cache in
Hello,
Peter Thoman, le Tue 22 Sep 2009 16:36:39 +0200, a écrit :
> However, the L1 cache size is wrong
512K is quite big indeed :)
hwloc gets it from /sys/devices/system/cpu/cpu0/cache/index*/size
> and the L3 cache is in fact shared between each set of 4 cores
> (its size is correct though).
Samuel Thibault, le Tue 22 Sep 2009 16:44:11 +0200, a écrit :
> I'm afraid the bug is most probably in the kernel,
BTW, I know that on x86 at least glibc and libgomp use the cpuid
instruction themselves to discover the cpu topology. Such backend could
be written to compensate for kernel bugs, but
Samuel Thibault wrote:
> Samuel Thibault, le Tue 22 Sep 2009 16:44:11 +0200, a écrit :
>
>> I'm afraid the bug is most probably in the kernel,
>>
>
> BTW, I know that on x86 at least glibc and libgomp use the cpuid
> instruction themselves to discover the cpu topology. Such backend could
>
Brice Goglin, le Tue 22 Sep 2009 19:29:33 +0200, a écrit :
> Samuel Thibault wrote:
> > Such backend could be written to compensate for kernel bugs,
>
> Well, the kernel also uses cpuid to discover the cpu topology.
Yes, but from a user point of view it's easier to upgrade to a newer
hwloc than t