[Darshan-users] Darshan trace file creation

[Shane Snyder]

Thank you for the info, Wadud. I was not aware, but it is good to hear 
that Intel resolved this issue in the new 2017 version.

I'll go ahead and close your ticket, but I'll create a new one that 
discusses this issue more clearly. I thought we had an open ticket on 
GitLab describing the Intel MPI issue, but we don't, so it will be good 
to track in case we decide to revisit and intercept Fortran symbols 
directly in the future.

--Shane

On 11/30/2016 04:31 AM, Wadud Miah wrote:
>
> Hi Shane,
>
> Many thanks for letting me know about Intel MPI’s issue with Fortran. 
> Yes, this is for a Fortran code. It seems like Intel MPI version 
> 2017.0.098 has fixed this issue as it is now creating trace files. You 
> may now close this ticket.
>
> Best regards,
>
> Wadud.
>
> *From:*Shane Snyder [mailto:ssnyder at mcs.anl.gov]
> *Sent:* 29 November 2016 20:50
> *To:* Wadud Miah <wadud.miah at nag.co.uk>; darshan-users at lists.mcs.anl.gov
> *Subject:* Re: [Darshan-users] Darshan trace file creation
>
> Hi Wadud,
>
> I'm assuming this is a Fortran app, right?
>
> If so, I believe this is tied back to a known issue with Intel MPI's 
> Fortran bindings. Long story short, these bindings are calling the 
> PMPI variants of the MPI function calls directly which bypasses any 
> profiling tool that wraps the MPI interface at the C layer (e.g., 
> Darshan). See this post for more details: 
> https://software.intel.com/en-us/forums/intel-clusters-and-hpc-technology/topic/279445
>
> I haven't personally confirmed this on Intel MPI 5.x versions (the 
> post I reference above is regarding a 4.x version of Intel MPI), but I 
> believe other users have reported the problem still exists.
>
> It should be possible to work around the issue by modifying Darshan to 
> provide an additional interface for intercepting Fortran MPI calls 
> natively at the Fortran layer, but we have not undertaken this 
> development effort as of yet. Lacking that, I'm not sure what, if any, 
> other solutions there are, unfortunately.
>
> Thanks,
> --Shane
>
> On 11/28/2016 02:58 PM, Wadud Miah wrote:
>
>     Hello,
>
>     I built Darshan with Intel MPI 5.1.3 and gcc 4.3.4 with the
>     following configure flags:
>
>     |./configure --prefix=$HOME/darshan-3.1.2-impi --with-mem-align=8
>     --with-jobid-env=NONE --with-log-path-by-env=DARSHAN_LOGPATH|
>
>     which built fine. I set |DARSHAN_LOGPATH|to |darshan-trace-dir|and
>     I created the directory |darshan-trace-dir|. I then executed my
>     MPI code which does I/O using:
>
>     |mpirun -n 9 -env LD_PRELOAD
>     /home/pr1ebt00/pr1ebt05/darshan-3.1.2-impi/lib/libdarshan.so
>     ./bt.A.9.mpi_io_full|
>
>     in the same directory that has the |darshan-trace-dir|directory.
>     However, I do not see any tracefile in the directory
>     |darshan-trace-dir|and no error messages. Any help will be greatly
>     appreciated.
>
>     Thanks in advance,
>
>     Wadud.
>
>     ---------------------------------------
>
>     Dr. Wadud Miah
>
>     Computational Scientist
>
>     Numerical Algorithms Group
>
>     01865 518035
>
>     ------------------------------------------------------------------------
>
>     The Numerical Algorithms Group Ltd is a company registered in
>     England and Wales with company number 1249803. The registered
>     office is:
>
>     Wilkinson House, Jordan Hill Road, Oxford OX2 8DR, United Kingdom.
>
>     This e-mail has been scanned for all viruses by Microsoft Office 365.
>
>     ------------------------------------------------------------------------
>
>
>
>
>     _______________________________________________
>
>     Darshan-users mailing list
>
>     Darshan-users at lists.mcs.anl.gov
>     <mailto:Darshan-users at lists.mcs.anl.gov>
>
>     https://lists.mcs.anl.gov/mailman/listinfo/darshan-users
>
>
>
> ________________________________________________________________________
> This e-mail has been scanned for all viruses by Star.
> ________________________________________________________________________
>
> ------------------------------------------------------------------------
>
> The Numerical Algorithms Group Ltd is a company registered in England 
> and Wales with company number 1249803. The registered office is:
>
> Wilkinson House, Jordan Hill Road, Oxford OX2 8DR, United Kingdom.
>
> This e-mail has been scanned for all viruses by Microsoft Office 365.
>
> ------------------------------------------------------------------------

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/darshan-users/attachments/20161130/62f1ea3a/attachment-0001.html>