shmem failure when --shmem F specified

[drmackay]

The following occurs on Topaz: Here are the modules loaded:

module list
Currently Loaded Modulefiles:
1) java/1.8                3) mpi/sgimpt/2.13-11280
2) compiler/intel/16.0.0   4) pbs/13.1.1.162303

Next run tau init with following specifiers:

tau init --mpi T --shmem F

After several warnings about oshCC, oshcc and oshifort the following Critical error message is issued:

[TAU] 
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
[TAU]
[TAU] CRITICAL
[TAU]
[TAU] '/p/home/apps/sgi/mpt-2.13-11280_1/bin/oshfort' isn't actually a SOS SHMEM
[TAU]      Fortran compiler since it doesn't accept arguments ['-show'].
[TAU]
[TAU] Hint: Try `tau --help`
[TAU]
[TAU] TAU cannot proceed with the given inputs.
[TAU] Please check the configuration for errors or contact
[TAU]     <support@paratools.com> for assistance.
[TAU]
[TAU] XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX

There is a work around to this on topaz. By replacing the sgi mpi module with intel mpi module:

module swap mpi/sgimpt/2.13-11280 mpi/intelmpi/16.0.0

then initializing I can initialize tau commander:

drmackay@topaz03:~/testissue> tau init --mpi T --shmem F

This completes successfully after swapping mpi modules. If tau is incompatible with sgi mpt we should just list that upfront and provide more information to the user that sgi mpt is not supported. Otherwise if I explicitly turn off shmem with --shmem F then tau commander should not test for it.

[khsa1]

Is this still an issue?

I just tried tau init --mpi T --shmem F with the following modules loaded:

Currently Loaded Modulefiles:
  1) java/1.8                2) compiler/intel/16.0.0   3) mpi/sgimpt/2.15         4) pbs/13.1.1.162303

and it initialized without any errors.

[zbeekman]

Hi Sam,

When I try this I get warnings about osh{CC,cc,fort} being assumed to be SOS, but other than this, it seems to work.

[khsa1]

That is what I am seeing as well, which I think is the correct behavior. When you do tau init --shmem F, that turns off SHMEM in the measurements, but the SHMEM compiler is still part of the target. Is there anything that should actually be changed here to make it clearer to users?

[zbeekman]

Is there anything that should actually be changed here to make it clearer to users?

I'm not sure. If tau init is creating the target and we specify --shmem F then is there a good reason to include the SHMEM compiler in the target? @jlinford thoughts?

To me it doesn't strike me as very DWIM (do what I mean) to include the shmem compilers in the target if on tau init we've specifically asked for them to be disabled; however, there may be a more nuanced issue with underlying TAU that I am overlooking here.

At any rate, this particular issues seems more or less resolved and I think it can be closed.

[zbeekman]

@khsa1 This appears to be due to mpi/sgimpt/2.13-11280. With the default mpi/sgimpt module loaded this issue doesn't happen.

[wohlbier]

I see this problem on thunder. It seems related to this issue. Interestingly, as related to another issue, when I do tau init in $WORKDIR/path it seems fine. I'm on the master branch.

wohlbier@thunder03:~/devel/foam/foam-extend-3.2> tau init --mpi [TAU] System MPI C++ compiler '/p/home/apps/sgi/mpt-2.12-sgi712r26/bin/mpicxx' [TAU] wraps [TAU] '/p/home/apps/intel/compilers/16/compilers_and_libraries_2016.1.150/linux/bin/intel64/icpc' [TAU] System MPI C compiler '/p/home/apps/sgi/mpt-2.12-sgi712r26/bin/mpicc' [TAU] wraps [TAU] '/p/home/apps/intel/compilers/16/compilers_and_libraries_2016.1.150/linux/bin/intel64/icc' [TAU] System MPI Fortran compiler [TAU] '/p/home/apps/sgi/mpt-2.12-sgi712r26/bin/mpif90' wraps [TAU] '/p/home/apps/intel/compilers/16/compilers_and_libraries_2016.1.150/linux/bin/intel64/ifort' [TAU] Created a new project named 'foam-extend-3.2'. [TAU] Added application 'foam-extend-3.2' to project configuration [TAU] 'foam-extend-3.2'. [TAU] ** [TAU] [TAU] WARNING [TAU] [TAU] Assuming '/p/home/apps/sgi/mpt-2.12-sgi712r26/bin/oshCC' is a SOS compiler [TAU] but it could be to any of these: SOS, OpenSHMEM [TAU] If this assumption is incorrect then you should manually specify your [TAU] compilers [TAU] [TAU] ** [TAU] XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX [TAU] [TAU] CRITICAL [TAU] [TAU] '/p/home/apps/sgi/mpt-2.12-sgi712r26/bin/oshCC' isn't actually a SOS SHMEM [TAU] C++ compiler since it doesn't accept arguments ['-show']. [TAU] [TAU] Hint: Try tau --help [TAU] [TAU] TAU cannot proceed with the given inputs. [TAU] Please check the configuration for errors or contact [TAU] support@paratools.com for assistance. [TAU] [TAU] XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX