Profiling Nwchem 6.8 (with score-p)


Just Got Here
Hi all,

I'd like to profile nwchem on our cluster. For that purpose I intend to use score-p (https://www.vi-hps.org/projects/score-p/), but I would be fine with any other profiler as well.

According to the score-p documentation one only has to precede CC with 'scorep' (likewise for FC).

So my build process now looks as follows:

1) EXPORT ALL NWCHEM-related ENVIRONMENT VARIABLES
2) export CC='scorep gcc'
3) export FC='scorep gfortran'
4) make nwchem_config
5) make -j8

Notice that everything runs find if I omit steps 2 and 3.

The problem now is that the Nwchem (6.8) makefile doesn't seem to cope well with the fact that CC is not just 'gcc' but now 'scorep gcc' (e.g., the generated makefile.h contains statements like 'ifeq ($(FC)), gfrotran)))

Has anyone build nwchem with score-p before and managed to run it? Any help is greatly appriciated.

(If note score-p, which other profilers can you recommend?)

Thanks

Forum Vet
This options worked for me using scorep 4.0 on a Fedora 28 box.
The trick is to supply _FC=gfortran and _CC=gcc
make FC="scorep-mpif90" _FC=gfortran CC=scorep-mpicc _CC=gcc  MAYBE_HOST=--host=x86_64-pc-linux-gnu

Forum Vet
Files available
The NWChem 6.8.1-scorep release is now available. This is a maintenance release that contains fixes to allow the NWChem 6.8.1 release to work with the Score-P measurement infrastructure

The command to compile on GNU compilers is

make FC="scorep-gfortran" _FC=gfortran CC=scorep-gcc _CC=gcc

https://github.com/nwchemgit/nwchem/releases/tag/6.8.1-scorep


Forum >> NWChem's corner >> Compiling NWChem