Open MPI logo

Hardware Locality Development Mailing List Archives

  |   Home   |   Support   |   FAQ   |  

This web mail archive is frozen.

This page is part of a frozen web archive of this mailing list.

You can still navigate around this archive, but know that no new mails have been added to it since July of 2016.

Click here to be taken to the new web archives of this list; it includes all the mails that are in this frozen archive plus all new mails that have been sent to the list since it was migrated to the new archives.

Subject: Re: [hwloc-devel] hwloc-1.3.1 and 1.4 failures on MIPS64/qemu
From: Paul H. Hargrove (PHHargrove_at_[hidden])
Date: 2012-02-01 00:49:31


The failure is only w/ {C,CXX}FLAGS=-mabi=64.
With the "32" or "n32" ABI's there is no problem.
That is why it was not seen on the 32-bit system.

I have no other gcc for that system at the moment.
So, I can't determine if it is a compiler bug.
It may also be a matter of which libs exist in 32-vs-64 bit versions on
this system.

-Paul

On 1/31/2012 8:57 PM, Paul H. Hargrove wrote:
> I have 2 QEMU-emulated Linux/MIPS systems.
> On one hwloc-1.3.1 and 1.4 are working fine.
> On the other I see strange "make check" failures.
> Unfortunately they differ in both distro version and 32- vs 64-bit,
> making it hard to pin down the cause.
>
> The newer distro w/ the 32-bt kernel is OK:
>> $ cat /etc/debian_version ; uname -a
>> 6.0.3
>> Linux debian-mips 2.6.32-5-4kc-malta #1 Thu Nov 3 05:21:00 UTC 2011
>> mips GNU/Linux
>
> The older distro w/ the 64-bit kernel is failing:
>> $ cat /etc/debian_version ; uname -a
>> 5.0.9
>> Linux qemu-hargrove-mips 2.6.24-etchnhalf.1-5kc-malta #1 Thu Oct 16
>> 01:21:56 BST 2008 mips64 GNU/Linux
>
> W/ hwloc-1.3.1 the failure is
>> FAIL: 256ia64-64n2s2c.output
> with numerous differences in the diff output.
>
> W/ hwloc-1.4 the failure is
>> FAIL: test-hwloc-distances.sh
> with just one difference:
>> -Latency matrix between 16 PUs (depth 4) by logical indexes (below
>> Machine L#0):
>> +Latency matrix between 16 PUs (depth 3) by logical indexes (below
>> Machine L#0):
> BUT, there is also a bus error that is being reported as a PASS:
>> Machine (247MB) + PU L#0 (P#0)
>> ./test-hwloc-ls.sh: line 50: 7061 Bus error $ls
>> $tmp/test.txt
>> PASS: test-hwloc-ls.sh
>
> At the moment my suspicion falls on the compiler, as I can't see how a
> failure of 256ia64-64n2s2c.output could be h/w dependent.
>
> -Paul
>
>

-- 
Paul H. Hargrove                          PHHargrove_at_[hidden]
Future Technologies Group
HPC Research Department                   Tel: +1-510-495-2352
Lawrence Berkeley National Laboratory     Fax: +1-510-486-6900