I am not sure about the best way to raise this so e-mailing directly.
Some users had reported that recent versions of lfric_apps fail to build using the CCE compiler on ARCHER2 with errors about use of unallocated arrays in the routine ukca_volcanic_so2.
I have opened issue:81 on the GitHub repository and associated PR is in progress.
Since the error/ warnings are so far being encountered on ARCHER2 only I tried to test my branch, as well as re-test a vn3.0 trunk clone with the fix, but while the ukca_volcanic_so2 routine is compiled the CCE build still fails with an internal compiler error for a Socrates routine – see message (from job.out) below. It is not clear if this is a known issue, or not encountered previously due to the UKCA failure happening prior to this routine.
For information: ~mdalvi/cylc-run/vn3.0_git_trunk/run1 = Vanilla run ~mdalvi/cylc-run/vn3.0_git_trunk/run2 = With UKCA fix in cloned copy
to be clear, the issue is that there is a compiler bug with CCE15 which crashes when building a file? Does it build with gnu (there are equivalent spack environments for lfric_apps 2.2 in a central place (see rose stem) which should also build 3.0)?
At the risk of talking like an answerphone message: If you want to develop on Archer then is it ok to use gnu? If you’d like to escalate the CCE compiler bug then can you make a small reproducible example, as the Archer2 team can test that?
Yes, builds using GNU are successful (with a warning for the ukca_volcanic_so2 arrays being fixed by issue:#81). The error was reported by someone attempting to build lfric_apps using CCE compiler, and I am testing this with a lfric_atm task available for the ncas site (lfric_atm_nwp_gal9_noukca_1T-C48_MG_ex_cce_fast-debug-64bit) but not currently part of default groups.
The is being highlighted in case anyone else attempts to build with the current CCE, unless this has been a known issue for ARCHER2.