multiple compiler support broken in newcase and/or setup scripts #1344
Comments
|
Yes this happens on Anvil with current master comparing default (intel) and --compiler gnu. |
|
Note that this is critical because members of both ocn and atm teams are using the intel compiler for their runs on titan, while much of the development is still using pgi. Without this, Titan support is still missing. |
|
OK, I will look at it today. It's probably an easy fix. |
|
@worleyph I fixed (b) in my fixes PR. I am not clear on (a) since intel build seemed to work fine. Are you saying that changing the compiler with case.setup is what's broken? |
|
If you look closely, the --intel is quietly building with the pgi options. Pgi works so everything seems to work. Compare the Macros.make file between the same case with 2 different compilers. They should be very different. |
Restore Titan support after CIME5.2 merge. A number of issues prevented successful create, set-up, build and run phases on Titan after the merge of CIME5.2. Even with these resolved, some functionality was lost, e.g. support for using a different number of MPI tasks per node for nonoverlapping components and support for more than one compiler. Finally, support for netcdf4 files was missing when using PGI.
Remove CESM_REPO.
Add better support for netcdf and pnetcdf in config*.xml .
Use findNetcdf.cmake from pio2 in pio1.
Use newer cmake, required by pio2.
Fix aprun command construction.
Fix Macro maker, to enable multiple compiler support.
Update modules and paths to enable netcdf4 support and multiple compiler support.
Fixes #1327
Fixes #1330
Fixes #1344
Also, (already fixed, but verified here)
Fixes #1239
Testing has focused on production-like cases, including low resolution water cycle and high resolution GMPAS, both with and without the v3 MPAS meshes and with a number of different PE layouts. acme_developer is not yet working, for reasons distinct from the pull request, and without these changes, progress will not be possible on getting the acme_developer to work again.
A unit test was added for the updated aprun command construction capability, and this was passed.
[BFB]
simplify netcdf lib args on hobart Using nf-config is overly complicated and in this case unnecessary, remove it. Test suite: scripts_regression_tests + SMS.f19_g16.X for all supported compilers on hobart Test baseline: Test namelist changes: Test status: bit for bit Fixes #1344 User interface changes?: Code review: Bill Sacks
a) ./create_new case followed by ./case.setup with -compiler intel and with -compiler pgi on Titan generates identical Macros.make files (the pgi version, I think).
The same thing occurs on Edison, comparing -compiler gnu with -compiler intel (don't know which one it generates, probably intel?)
@rljacob indicates that a similar issue occurs on Anvil.
(The compiler selection is correction in env_build.xml.)
b) the env_mach_specific.xml script generated (from config_machines.xml) on Titan has hardcoded paths for NETCDFROOT and PNETCDFROOT that point at compiler specific locations. This needs to be moved into the compiler-specific logic. Don't know if this is true for other systems as well or not.
The text was updated successfully, but these errors were encountered: