Spack package to add build options for CICE5 in esm1.6

[anton-seaice]

@harshula

When I try to use this package, it builds the wrong executable

I run

spack install [email protected]=access-esm1.6%[email protected]

and it builds the executable for access-cm2 ?

From the spack-build-out.txt file, I can see that spack is sending the wrong arguments to spack-build.sh :

==> cice5: Executing phase: 'build'
==> [2024-11-29-11:34:55.729889] '/g/data/tm70/as2285/spack0.22/spack/var/spack/environments/esm1_5/cice5/bld/spack-build.sh' 'access-cm2' '360x300' '12x1' '12'

Any ideas ?

[anton-seaice]

I think the issue is the use of strings as versions ...

Reading the spack docs, on version comparison, I think 'access-esm1.6' is > 'access-cm2' because e>c ... Which is pretty confusing.

Reording the with statements could fix the problem, but is pretty confusing ... I am inclined to use variants instead ?

[anton-seaice]

@harshula - I made the suggested changes, but it didn't generate the __targets correctly

I moved the __targets to the build() step and now it works

For access-om2, see the pre-release for this version of spack-packages:
ACCESS-NRI/ACCESS-OM2#88

The pre-release folder contains the 4 expected cice5 executables for access-om2. The repro test for this is running.

For the future access-esm1.6, in my local instance this works.

It solves correctly (i.e. doesn't try to include libaccessom2 which is only needed for the master version ) :

Input spec
--------------------------------
 -   [email protected]

Concretized
--------------------------------
 -   [email protected]=access-esm1.6%[email protected]~deterministic~optimisation_report build_system=makefile dev_path=/g/data/tm70/as2285/spack0.22/environments/esm1_5/cice5 arch=linux-rocky8-x86_64_v4
[+]      ^[email protected]%[email protected] build_system=autotools patches=80b9577 arch=linux-rocky8-x86_64_v4
[e]      ^[email protected]%[email protected] build_system=autotools arch=linux-rocky8-x86_64_v4
[+]      ^[email protected]%[email protected]~guile build_system=generic arch=linux-rocky8-x86_64_v4
[+]      ^[email protected]%[email protected]~blosc~byterange~dap~fsync~hdf4~jna+mpi~nczarr_zip+optimize~parallel-netcdf+pic+shared~szip~zstd build_system=autotools arch=linux-rocky8-x86_64_v4
[+]          ^[email protected]%[email protected]~cxx~fortran+hl~ipo~java+mpi+shared~szip~threadsafe+tools api=default build_system=cmake build_type=Release generator=make arch=linux-rocky8-x86_64_v4
[e]              ^[email protected]%[email protected]~doc+ncurses+ownlibs build_system=generic build_type=Release arch=linux-rocky8-x86_64_v4
[+]              ^[email protected]%[email protected] build_system=autotools arch=linux-rocky8-x86_64_v4
[+]              ^[email protected]%[email protected]+compat+new_strategies+opt+pic+shared build_system=autotools arch=linux-rocky8-x86_64_v4
[+]      ^[email protected]%[email protected]~doc+pic+shared build_system=autotools patches=b050dbd arch=linux-rocky8-x86_64_v4
[+]      ^[email protected]_2024.05.24=access-esm1.5%[email protected]~deterministic~optimisation_report build_system=makefile arch=linux-rocky8-x86_64_v4
[e]      ^[email protected]%[email protected]~atomics~cuda~cxx~cxx_exceptions~gpfs~internal-hwloc~internal-libevent~internal-pmix~java~legacylaunchers~lustre~memchecker~openshmem~orterunprefix~romio+rsh~singularity~static+vt+wrapper-rpath build_system=autotools fabrics=none patches=073477a,60ce20b romio-filesystem=none schedulers=none arch=linux-rocky8-x86_64_v4

and generates the one expected executable:

cice_access-esm1.6_360x300_12x1_12p.exe

[anton-seaice]

New build is bitwise reproducible with the old, see ACCESS-NRI/access-om2-configs#151 (comment)

[anton-seaice]

Closes #156

[harshula]

@harshula - I made the suggested changes, but it didn't generate the __targets correctly
I moved the __targets to the build() step and now it works

Did it generate the targets correctly before you made the suggested changes?

[anton-seaice]

@harshula - I made the suggested changes, but it didn't generate the __targets correctly
I moved the __targets to the build() step and now it works

Did it generate the targets correctly before you made the suggested changes?

No - see first comment in the PR.

[harshula]

Thanks for the clarification. It is best to avoid declaring member variables (self.__targets = __targets) from a member function. Ideally, the targets data structure should be declared after the class declaration. I'll look into whether with when() allows variable declaration/definition. This might be a with when() limitation.

[anton-seaice]

Ok thanks. The targets are only used in the build function, so it could just me a local
Variable instead of a member variable

[harshula]

No, it is also used in the install member function.

[anton-seaice]

Hi @ofa001 - just confirming a couple of science build settings for CICE5 in access-esm1.6

I have set these two lines:

Zero layer ice:
setenv NICELYR 1 #1 for ktherm=0, zero-layer thermodynamics

No BGC tracers:
setenv TRBGCS 0

Is 5 thickness categories and 1 snow layer correct ? Do these two lines need changing?:

setenv NSNWLYR    1       # number of vertical layers in the snow
setenv NICECAT    5       # number of ice thickness categories

This is set to build with 12 tasks (to match esm1.5) but also will build a version with 16 tasks if needed. (Obviously these can be refined in the future).

[anton-seaice]

@harshula this is ready for review.

Produced two binaries expected:

$ ls /g/data/tm70/as2285/spack0.22/restricted/ukmo/release/linux-rocky8-x86_64_v4/intel-19.0.3.199/cice5-git.access-esm1.6_access-esm1.6-nipftaviflymvjbwuxhzqycxpfucyexn/bin/
cice_access-esm1.6_360x300_12x1_12p.exe  cice_access-esm1.6_360x300_8x2_16p.exe

[ofa001]

HI @anton-seaice @harshula, @DaveBi is holding off finishing the coupling the work as I understand to the New Year whilst he gets the cice4 version with the icebergs linked up with Pearce's Generic tracer for the test run's in Rachel's table. So whilst your settings for cice5 ice thickness categoriex and snow ice layers are thin Anton, it wont run, not sure it can pick tha tit doesnt have the correct links in to the UM in this spack set up as I am not sure how sophisticated it is. We can talk about it at the mini-hackathon tomorrow if you are around.

[anton-seaice]

Thanks @ofa001 - this is just build infrastructure, I am just making sure its ready / available so we can build the model so the cice code can be updated.

So whilst your settings for cice5 ice thickness categoriex and snow ice layers are thin Anton

What did you mean here ? We can change these compile time options later ofcourse, but its good if they are close now.

[ofa001]

Sorry @anton-seaice, I was distracted, I had someone here fixing my aircon, I meant "fine"