Intel installation issue

170 views
Skip to first unread message

Jordi Sanglas Molist

unread,
Jun 30, 2021, 6:46:52 AM6/30/21
to pflotran-dev
Hi all,

I was trying to install PETSc and PFLOTRAN using Intel compilers on Linux, but I have some issues.

After cloning PETSc, I run the following configuration command:

./configure --CFLAGS=-O3 --CXXFLAGS=-O3 --FFLAGS=-O3 --with-debugging=no --download-hdf5=yes --download-fblaslapack=yes --download-ptscotch=yes --with-cc=mpicc --with-cxx=mpicxx --with-fc=mpif90

I verified that mpicc, mpicxx and mpif90 have the correct paths (by using which mpif90, which mpicc...). Then I use the command make all and PETSc is compiled successfully. However, after running the PETSc tests using make check, one of them fails:

Running check examples to verify correct installation
Using PETSC_DIR=/home/altopt/intel_pflotran/petsc and PETSC_ARCH=arch-linux2-c-opt
C/C++ example src/snes/tutorials/ex19 run successfully with 1 MPI process
Possible error running C/C++ src/snes/tutorials/ex19 with 2 MPI processes
[1625044409.741787] [ord-19-18:87818:0]         select.c:406  UCX  ERROR no remote registered memory access transport to <no debug data>: self/self - Destination is unreachable, mm/sysv - Destination is unreachable, mm/posix - Destination is unreachable, cma/cma - no put short
[1625044409.741824] [ord-19-18:87817:0]         select.c:406  UCX  ERROR no remote registered memory access transport to <no debug data>: self/self - Destination is unreachable, mm/sysv - Destination is unreachable, mm/posix - Destination is unreachable, cma/cma - no put short
Abort(1091215) on node 1 (rank 1 in comm 0): Fatal error in PMPI_Init_thread: Other MPI error, error stack:
MPIR_Init_thread(138)........: 
MPID_Init(1139)..............: 
MPIDI_OFI_mpi_init_hook(1728): OFI get address vector map failed
Abort(1091215) on node 0 (rank 0 in comm 0): Fatal error in PMPI_Init_thread: Other MPI error, error stack:
MPIR_Init_thread(138)........: 
MPID_Init(1139)..............: 
MPIDI_OFI_mpi_init_hook(1728): OFI get address vector map failed
C/C++ example src/vec/vec/tests/ex47 run successfully with hdf5
Fortran example src/snes/tutorials/ex5f run successfully with 1 MPI process
Completed test examples

Note that the test that fails is the one that uses more than 1 CPU. If I then clone and compile PFLOTRAN, I can only run models using 1 process. If I try to run them using more than one process, PFLOTRAN crashes. I am not sure what the problem is. I tried searching the error on the Internet, but I have not found anything useful.

Thanks,
Jordi

Hammond, Glenn E

unread,
Jun 30, 2021, 10:59:54 AM6/30/21
to pflotr...@googlegroups.com

Jordi,

 

I recommend sending error message below to petsc...@mcs.anl.gov.

 

Glenn

 

From: pflotr...@googlegroups.com <pflotr...@googlegroups.com> On Behalf Of Jordi Sanglas Molist
Sent: Wednesday, June 30, 2021 3:47 AM
To: pflotran-dev <pflotr...@googlegroups.com>
Subject: [pflotran-dev: 6072] Intel installation issue

 

Check twice before you click! This email originated from outside PNNL.

 

--
You received this message because you are subscribed to the Google Groups "pflotran-dev" group.
To unsubscribe from this group and stop receiving emails from it, send an email to pflotran-dev...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/pflotran-dev/d437f253-2415-4e0e-b0aa-dcc19bc0eb87n%40googlegroups.com.

Reply all
Reply to author
Forward
0 new messages