[CIG-SEISMO] Issue Configuring SPECFEM3D on Stampede2 Supercomputer

[Malladi, Rama Kishan V]

Hi Ian,
Not sure if your issue is related to an earlier post. Can you try using this compiler flag? : '-assume nobuffered_io'

Thanks
-Rama

From: CIG-SEISMO [mailto:cig-seismo-bounces at geodynamics.org] On Behalf Of Ian Stone
Sent: Thursday, May 17, 2018 11:19 PM
To: cig-seismo at geodynamics.org
Subject: [CIG-SEISMO] Issue Configuring SPECFEM3D on Stampede2 Supercomputer

Hello,

My name is Ian Stone, and I am a PhD student at the University of Washington. I am trying to install Specfem3D Cartesian on the TACC Stampede2 supercomputer so I can simulate earthquake scenarios.

I have run into an issue where MPI processes do not seem to execute correctly, where either xgenerate_databases is not producing proper database files, or xspecfem3D is not reading them correctly. This is the error message I receive when I attempt to run the MountStHelens example on 4 cores; it is created when xspecfem3D starts reading the database files:

-----------

forrtl: severe (24): end-of-file during read, unit 27, file /<path>/specfem3d/EXAMPLES/Mount_StHelens/OUTPUT_FILES/DATABASES_MPI/proc000001_external_mesh.bin

Image              PC                Routine            Line        Source

xspecfem3D         0000000001022F89  Unknown               Unknown  Unknown

xspecfem3D         00000000010489C4  Unknown               Unknown  Unknown

xspecfem3D         0000000000C7CE00  read_mesh_databas          68  read_mesh_databases.F90

xspecfem3D         0000000000DD6B43  MAIN__                    353  specfem3D.F90

xspecfem3D         000000000040423E  Unknown               Unknown  Unknown

libc.so.6          00002ABACC0FFC05  Unknown               Unknown  Unknown

xspecfem3D         0000000000404142  Unknown               Unknown  Unknown

forrtl: severe (24): end-of-file during read, unit 27, file /<path>/specfem3d/EXAMPLES/Mount_StHelens/OUTPUT_FILES/DATABASES_MPI/proc000002_external_mesh.bin

Image              PC                Routine            Line        Source

xspecfem3D         0000000001022F89  Unknown               Unknown  Unknown

xspecfem3D         00000000010489C4  Unknown               Unknown  Unknown

xspecfem3D         0000000000C7CE00  read_mesh_databas          68  read_mesh_databases.F90

xspecfem3D         0000000000DD6B43  MAIN__                    353  specfem3D.F90

xspecfem3D         000000000040423E  Unknown               Unknown  Unknown

libc.so.6          00002BA144606C05  Unknown               Unknown  Unknown

xspecfem3D         0000000000404142  Unknown               Unknown  Unknown

forrtl: severe (24): end-of-file during read, unit 27, file /<path>/specfem3d/EXAMPLES/Mount_StHelens/OUTPUT_FILES/DATABASES_MPI/proc000003_external_mesh.bin

Image              PC                Routine            Line        Source

xspecfem3D         0000000001022F89  Unknown               Unknown  Unknown

xspecfem3D         00000000010489C4  Unknown               Unknown  Unknown

xspecfem3D         0000000000C7CE00  read_mesh_databas          68  read_mesh_databases.F90

xspecfem3D         0000000000DD6B43  MAIN__                    353  specfem3D.F90

xspecfem3D         000000000040423E  Unknown               Unknown  Unknown

libc.so.6          00002B83ACB71C05  Unknown               Unknown  Unknown

xspecfem3D         0000000000404142  Unknown               Unknown  Unknown


-----------

I can run this simulation on a single core without invoking MPI, so I know specfem is working. It is just when I attempt to run using MPI that things go awry. The nodes & cores values match between my Par_file and sbatch script, so I figure that means something is wrong with the way I configured specfem initially.

To configure, I load intel version 16.0.3 and impi version 17.0.3, and use the command:


./configure CC=icc FC=ifort MPIFC=mpif90 MPI_INC="/opt/intel/compilers_and_libraries_2017.4.196/linux/mpi/intel64/include" --with-mpi

Does this look ok? Or do any of you have experience installing specfem either on Stampede or Stampede2? I should add that running after using the "--enable-debug" flag does not produce any outstanding messages.

Thank you,

Ian Stone
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.geodynamics.org/pipermail/cig-seismo/attachments/20180518/63219af8/attachment-0001.html>