DRAFT - avoid optimization for setuprad.f90 with Intel

[climbfuji]

Description

This PR is a suggestion/workaround to address the GSI failures with Intel when compiler optimization is enabled. Please ignore the changes to the Orion LUA module, I needed that to (unsuccessfully) attempt running the regression tests myself.

I got far enough to run into a permissions issue for the global_3dvar test, because I am not in group rstprod. But I couldn't figure out how to run global_T62, at least this test doesn't exist when I do

cmake -DCONTROLPATH=$PWD/../install/bin -DBUILD_REG_TESTING=ON -DCMAKE_INSTALL_PREFIX=$PWD/../install ../GSI-lower-opt-intel 2>&1 | tee log.cmake

This PR may fix #447, but someone needs to test this. @DavidHuber-NOAA?

Type of change

Please delete options that are not relevant.

• Bug fix/workaround (non-breaking change which fixes an issue)

How Has This Been Tested?

See above

Checklist

• My code follows the style guidelines of this project
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• New and existing tests pass with my changes
• Any dependent changes have been merged and published

[aerorahul]

Thanks @climbfuji
Just one TODO placeholder, so this is not forgotten.

[DavidHuber-NOAA]

Thanks @climbfuji I will give this a try today.

[DavidHuber-NOAA]

@climbfuji I gave this a try and the global_3dvar test failed with a crash on gsi/readfiles.f90:724 when compiled with Intel 2022.3.0 on Hera. Here is the stack trace:

0 0x000000000004d455 ucs_debug_print_backtrace() ???:0
1 0x0000000001e7a09c do_deallocate_all() for_alloc_copy.c:0
2 0x00000000012d17c5 read_files_() /scratch1/NESDIS/nesdis-rdo2/David.Huber/gsi_test/src/gsi/read_files.f90:724
3 0x0000000000f9d0ae gesinfo_() /scratch1/NESDIS/nesdis-rdo2/David.Huber/gsi_test/src/gsi/gesinfo.F90:606
4 0x000000000063a02f gsisub_() /scratch1/NESDIS/nesdis-rdo2/David.Huber/gsi_test/src/gsi/gsisub.F90:131
5 0x00000000004164fd gsimod_mp_gsimain_run_() /scratch1/NESDIS/nesdis-rdo2/David.Huber/gsi_test/src/gsi/gsimod.F90:2266
6 0x000000000041643f MAIN__() /scratch1/NESDIS/nesdis-rdo2/David.Huber/gsi_test/src/gsi/gsimain.f90:631
7 0x00000000004163dd main() ???:0
8 0x0000000000022555 __libc_start_main() ???:0
9 0x00000000004162f6 _start() ???:0

I will keep adding the OPTIMIZE:0 directive to the broken modules until we have something working or it fails to fix an issue.

[climbfuji]

Hmm. Something isn't right with this code. Thanks for testing!

[DavidHuber-NOAA]

@climbfuji Unfortunately, adding the directive to read_files.F90 did not fix that crash. On inspection, I noticed that this crash occurs on the last line of the module, according to the backtrace, so I went up another level and tried deoptimizing gesinfo.F90, which also didn't work; same crash. This work is in port_2_i2021.

Reading up on the directive itself, setting optimization flags is a 'suggestion' to the compiler and may be ignored without any diagnostic output. So instead, I'm going to modify the CMake file to set -O0 for these three modules.

[climbfuji]

@climbfuji Unfortunately, adding the directive to read_files.F90 did not fix that crash. On inspection, I noticed that this crash occurs on the last line of the module, according to the backtrace, so I went up another level and tried deoptimizing gesinfo.F90, which also didn't work; same crash. This work is in port_2_i2021.

Reading up on the directive itself, setting optimization flags is a 'suggestion' to the compiler and may be ignored without any diagnostic output. So instead, I'm going to modify the CMake file to set -O0 for these three modules.

Thanks for all the tests, @DavidHuber-NOAA. This is very annoying, but if instead of using directives setting cmake compiler flags works, then that's fine!

[DavidHuber-NOAA]

@climbfuji Unfortunately, deoptimizing the 3 modules still resulted in a segmentation fault in read_files.f90. I took it another step and deoptimized every module that was called by read_files.f90 and gesinfo.F90, but I am still getting the same seg fault.

[climbfuji]

Closing this, since we have a better solution in #447