charm AT lists.cs.illinois.edu

Subject: Charm++ parallel programming system

List archive

[charm] build charm++ ofi network layer on intel omni-path architecture

From: acyang <acyang0903 AT gmail.com>
To: charm AT lists.cs.illinois.edu
Subject: [charm] build charm++ ofi network layer on intel omni-path architecture
Date: Thu, 27 Dec 2018 10:40:44 +0800
Authentication-results: illinois.edu; spf=pass smtp.mailfrom=acyang0903 AT gmail.com; dkim=pass header.d=gmail.com header.s=20161025; dmarc=pass header.from=gmail.com

I want to run NAMD on our intel omni-path cluster.

According to the installation guide, ofi network layer should be chosen to get better performance.

I followed the recipet from intel developer zone :

https://software.intel.com/en-us/articles/recipe-building-namd-on-intel-xeon-and-intel-xeon-phi-processors-for-multi-node-runs

I built charm++ ofi-linux-x86_64 ofi-linux-x86_64, but when I tried to use charm++ to build app(simplearrayhello or megatest in charm++ built-in tests folder), I got the following error message:

../../../bin/../lib/libconv-core.a(convcore.o): In function `CmiTimerInit':
convcore.c:(.text+0x6435): undefined reference to `CmiBarrierZero'

Other users also mentioned this issue in the comment of intel recipet, but the author has not reply yet.

Any suggestions are welcome.

Thanks

A.C.

[charm] build charm++ ofi network layer on intel omni-path architecture, acyang, 12/26/2018