[Beowulf] Detecting binaries that are limited to an architecture?

Paul McIntosh paul.mcintosh at monash.edu
Sun May 8 18:18:54 PDT 2016



Thanks! I’ll use a combination of proactive tests based on this and reactive based on dmesg/message log. I’ll also investigate Chris’ cupid.


Here’s a quick test – our broken python (due to libraries which only run on new nodes) and the fixed python…


for i in `cat files_broken`; do echo $i; ./analyze-x86 $i | grep AVX | grep -v " 0"; done


        AVX:        361794

        AVX2:           17







        AVX:          5115

        AVX2:           29


        AVX:          6120

        AVX2:           29



for i in `cat files_fixed`; do echo $i; ./analyze-x86 $i | grep AVX | grep -v " 0"; done













From: Carlos Fenoy [mailto:minibit at gmail.com] 
Sent: Friday, 6 May 2016 7:32 PM
To: Carsten Aulbert <Carsten.Aulbert at aei.mpg.de>
Cc: Paul McIntosh <paul.mcintosh at monash.edu>; beowulf at beowulf.org
Subject: Re: [Beowulf] Detecting binaries that are limited to an architecture?




Here you have a more updated version of the script. This one supports up to AVX2, although it will not tell you exactly the CPU version it is compatible with.








On Fri, May 6, 2016 at 7:06 AM, Carsten Aulbert <Carsten.Aulbert at aei.mpg.de <mailto:Carsten.Aulbert at aei.mpg.de> > wrote:


On 05.05.2016 22:51, Paul McIntosh wrote:

> We have /usr/local mounted across nodes with some Intel Xeon X5650
> (Westmere) and some E5-2670 (SandyBridge). Some code spits out "Illegal
> Instruction" when run on the old nodes and it appears to be due to hitting
> shared libraries compiled on the newer nodes. We are going to have a similar
> situation on the newer clusters also.

some years ago I accidentally found this here


which "simply" wades through all instructions and spits out for which
target platform the binary is ok.

Given that this is 4yrs old and I had to apply a small fix about
variable naming (I think/attached) it will most certainly not find all
edge cases, but may already help to get started.



Beowulf mailing list, Beowulf at beowulf.org <mailto:Beowulf at beowulf.org>  sponsored by Penguin Computing
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf



Carles Fenoy

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.beowulf.org/pipermail/beowulf/attachments/20160509/434a1921/attachment.html>

More information about the Beowulf mailing list