Thanks, Jeff.
I'll let you know what happens.
Best.....John
On 2/16/16 10:19 AM, Jeff Squyres (jsquyres) wrote:
-----Original Message-----
From: JR Cary <c...@txcorp.com>
Reply: Open MPI Users <us...@open-mpi.org>
Date: February 16, 2016 at 9:39:23 AM
To: us...@open-mpi.org <us...@open-mpi.org>
Subject: Re: [OMPI users] readv failed How to debug?
Thanks, Gilles,
Yes, this binary was built a few years ago.
You mention a user error, but do you mean developer error? I.e., it
would have to be in the code?
We can't say for sure, but it is *likely* that this error is in the code itself.
What does "--mca coll ^tuned" do?
It turns off the "tuned" collective component and instead uses the "basic" collective
component. The "basic" component generally has lower performance, but *if* the problem is that the
code is using equivalent-but-different datatypes, the problem should go away. If the problem is that one of
the MPI processes is simply providing an incorrect count, then it's going to fail like this no matter what
mpirun CLI options are used.
--
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to:
http://www.cisco.com/web/about/doing_business/legal/cri/
