[Darshan-users] darshan3.1.5 issue on Cray XC40 cle6.up05

Bilel Hadri bilel.hadri at kaust.edu.sa
Tue Feb 27 10:36:46 CST 2018


Hi Phil,

Awesome, your workaround fixed the issue.

Regarding the linker, we are under CLE6Up5, and SUSE12 SP3 on the front end node, we are using xalt ( the wrapper to track library usage at linking and execution does not conflict with darshan), so the ld used is xalt wrapper and it works perfectly with both version.
which ld
/sw/xc40cle6/xalt/1.1.2/sles12.3_cce8.6.5/bin/ld

ld --version
GNU ld (GNU Binutils) 2.26.0.20160224
Copyright (C) 2015 Free Software Foundation, Inc.
This program is free software; you may redistribute it under the terms of
the GNU General Public License version 3 or (at your option) a later version.
This program has absolutely no warranty.

I really appreciate your kind and prompt help.

Regards,

Bilel

-------

Bilel Hadri, PhD
Computational Scientist
KAUST Supercomputing Lab
Al Khawarizmi Bldg. (1)  Office 126
4700 King Abdullah University of Science and Technology
Thuwal 23955-6900
Kingdom of Saudi Arabia
Office Phone: +966 12 808 0654
Cell Phone: + 966 544 700 893


From: Phil Carns <carns at mcs.anl.gov>
Date: Tuesday, February 27, 2018 at 6:18 PM
To: "bilel.hadri at kaust.edu.sa" <bilel.hadri at kaust.edu.sa>, "darshan-users at lists.mcs.anl.gov" <darshan-users at lists.mcs.anl.gov>
Subject: Re: [Darshan-users] darshan3.1.5 issue on Cray XC40 cle6.up05

On 02/27/2018 12:34 AM, Bilel Hadri wrote:
Hi Phil,
Sure, test_scalapack.f90 call ScaLAPACK routines and uses inside MPI library. The output of the failing ftn command with the craype-verbose flag



Thanks Bilel.  The petsc example helps too.  This clarifies that the trigger on Cray up05 is to compile a program that doesn't directly  use MPI symbols on its own but instead links in another library  that indirectly uses MPI symbols on its behalf.

This isn't an unusual use case (scalapack and petsc have done this forever), and the linker options in Darshan are set up to handle it, but something has obviously changed in the linker in this respect in this update.

We are still trying to work out how to reproduce this but it may be a little bit.  In the mean time would you be willing to try something in your 3.1.5 install to help debug?

Can you manually edit /sw/xc40cle6/darshan/3.1.5/sles12.3_gcc7.2.0/share/ld-opts/darshan-base-ld-opts and add this line to the top of the file, then try again to compile your examples with 3.1.5 loaded?

--undefined=__wrap_MPI_Init

By my (admittedly slightly murky) interpretation of the link order and the flags that we are setting, that shouldn't be necessary, but it would be helpful to see what it does for your example.

thanks,

-Phil

________________________________
This message and its contents including attachments are intended solely for the original recipient. If you are not the intended recipient or have received this message in error, please notify me immediately and delete this message from your computer system. Any unauthorized use or distribution is prohibited. Please consider the environment before printing this email.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/darshan-users/attachments/20180227/dc83d692/attachment-0001.html>


More information about the Darshan-users mailing list