Different MOM6 solution in ACCESS-om3 using pure cmake build.sh vs spack workflow.

[chrisb13]

Re: this spack-package.

@ezhilsabareesh8 @minghangli-uni have observed different MOM6 solutions using the "pure cmake" build.sh build workflow as compared to the pre-built spack executables (e.g. size and timing of truncation errors differs). @micaeljtoliveira suggested that it could be related to the architecture dependent optimisations being turned off (e.g. “-march”; perhaps cmake adds this by default?). Unverified: could additionally mean that the spack built version is slower than the build.sh version.

The relevant compiler flags for the Fortran code base are (L98-101):

elseif(CMAKE_Fortran_COMPILER_ID MATCHES "Intel")
  set(CMAKE_Fortran_FLAGS "${CMAKE_Fortran_FLAGS} -qno-opt-dynamic-align  -convert big_endian -assume byterecl -ftz -traceback -assume realloc_lhs -fp-model source")
  set(CMAKE_Fortran_FLAGS_RELEASE "-O2 -debug minimal")
  set(CMAKE_Fortran_FLAGS_DEBUG "-O0 -g -check uninit -check bounds -check pointers -fpe0 -check noarg_temp_created")

source

@aidanheerdegen suggested posting this as an issue to answer.. Question: what flags are being used by Spack? Any other ideas why we get different solutions?

[chrisb13]

Question: what flags are being used by Spack?

@harshula has advised the following can be used to find the flags:

. spack-config/spack-enable.bash
spack find --paths | grep -i mom5
zcat /opt/release/linux-rocky8-x86_64/intel-2021.2.0/mom5-git.2024.08.23_2024.08.23-bi5ksivdcqfh7gj7a6tof3afp5dfoxo6/.spack/spack-build-out.txt.gz | less