GEOS-ESM / GEOSctm

Fixture for chemical transport scenarios
Apache License 2.0
0 stars 2 forks source link

Update ctm_setup for MPT and MAPL 2.0 #26

Closed mathomp4 closed 4 years ago

mathomp4 commented 4 years ago

MPT plus MAPL 2.0 requires some new options.

mmanyin commented 4 years ago

I'm puzzled. Shouldn't this be part of a larger PR, that includes

mathomp4 commented 4 years ago

I'm puzzled. Shouldn't this be part of a larger PR, that includes

  • revised g5_modules
  • revised RUN_CMD
  • revised env vars ?

@mmanyin I am not going to be the one to tell you to run MPT. All this PR does is (I think) let you if you want. If ctm_setup detects you have built with MPT, it will set the flags it thinks are best. If you build it Open MPI or Intel MPI, it will try to do the same as well.

My guess is any for MVAPICH2 or MPICH might be out-of-date, but if you decided to use those MPI stacks I can try to figure some out.

mmanyin commented 4 years ago

Is the "Build Error" any cause for concern?

mathomp4 commented 4 years ago

Is the "Build Error" any cause for concern?

@mmanyin I think we will get those until PR #23 is merged into master. CircleCI is looking for things to do on GEOSctm but not finding any configuration files :)

JulesKouatchou commented 4 years ago

@mathomp4 I have a question. If I check out the CTM code, will the g5_modules file automatically include MPT?

mmanyin commented 4 years ago

@JulesKouatchou Looking at g5_modules, it is set to use Intel MPI on NCCS, and MPT at NAS. What #26 did was to change the setup script, not the build environment.

mathomp4 commented 4 years ago

@JulesKouatchou Looking at g5_modules, it is set to use Intel MPI on NCCS, and MPT at NAS. What #26 did was to change the setup script, not the build environment.

Correct. I wanted the CTM to have the ability to use MPT if needed. It wouldn't be hard to make a version of ESMA_env that defaults to MPT, but then ctm_setup should have a check that if someone tries to run on the Skylakes at NCCS, it stops.