[Darshan-users] compiling statically with intel toolchain

Harms, Kevin harms at alcf.anl.gov
Wed May 18 11:02:01 CDT 2016


Rene,

  ok, two items below.



>If I look at the binary generated I see this.
>
>nm $binfile | grep -i "mpi_init" 
>                 U MPI_Init
>                 U MPI_Initialized
>

  I think this means you are using dynamic linking. You could try darshan using the dynamic linking method that uses LD_PRELOAD.

http://www.mcs.anl.gov/research/projects/darshan/docs/darshan-runtime.html#_instrumenting_dynamically_linked_fortran_applications

>
>Maybe this is known problem.  I seemed to have missed this section of the documentation 
>
>http://www.mcs.anl.gov/research/projects/darshan/docs/darshan-runtime.html#_linux_clusters_using_intel_mpi
>
>Caveat
>Darshan is only known to work with C and C++ executables generated by the Intel MPI suite. Darshan will not produce instrumentation for Fortran executables. For more details please check this Intel forum discussion:
>
>
>Unfortunately a very big chunk of our codes are in fortran and we use intel MPI and ifort for that.  It looks like I may be out of luck on this? 

  Yes, possibly. Intel may have changed their setup but unlikely. I believe the issue here is that that the Intel Fortran MPI functions call the PMPI_* routines directly and thus can not be intercepted unless the specific Fortran interface is intercepted. Do you do I/O from FORTAN directly or are you using an I/O library?

kevin
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 4090 bytes
Desc: not available
URL: <http://lists.mcs.anl.gov/pipermail/darshan-users/attachments/20160518/c12ad70a/attachment.bin>


More information about the Darshan-users mailing list