Update adastra toolchain's prepare.sh

[etiennemlb]

• Fixes an issue with ginkgo not being built using the appropriate architecture.
• Prevent sourcing.
• Update to ginkgo 1.9
• Avoid hardcoding ${SPACK_USER_PREFIX}/config_user_spack

[tpadioleau]

Ok for me. I think we should wait for the results with @AbdelhadiKara before merging.

Was your branch up to date with main ? I don't understand why the static analysis failed.

[EmilyBourne]

Does this PR close #151?

[github-actions]

This PR is failing tests so it has been put back into draft. Please remove the draft status when the tests pass.

[AbdelhadiKara]

A first Installation/compilation attempt led to fix the matrix_tools tests on device.
OpenMp must nevertheless be deactivated.

[AbdelhadiKara]

It will be necessary to revert my commit.
And replace generic by mi250 [ in environment.sh file ] (

gyselalibxx/toolchains/mi250.hipcc.adastra.spack/environment.sh

Line 23 in 14bc781

gcc/13.2.1.generic/zen3/libyaml \

)

[tpadioleau]

Ok for me, @AbdelhadiKara can you confirm this is the file you used to do your tests ?

[AbdelhadiKara]

After compiling and running the ginkgo-related tests on the amd gpu, it was working for me yesterday. There are still two other points to be clarified, not in the scope of this MR.
The first concerns the fact that having device and Openmp activated at the same time breaks certain tests in matrix_utils.
The second concerns the extension of the polar poisson test timeout.

[tpadioleau]

After compiling and running the ginkgo-related tests on the amd gpu, it was working for me yesterday. There are still two other points to be clarified, not in the scope of this MR. The first concerns the fact that having device and Openmp activated at the same time breaks certain tests in matrix_utils. The second concerns the extension of the polar poisson test timeout.

Ok thanks, can you open issues about this if they don't already exist ?

[EmilyBourne]

After compiling and running the ginkgo-related tests on the amd gpu, it was working for me yesterday. There are still two other points to be clarified, not in the scope of this MR. The first concerns the fact that having device and Openmp activated at the same time breaks certain tests in matrix_utils. The second concerns the extension of the polar poisson test timeout.

Ok thanks, can you open issues about this if they don't already exist ?

@AbdelhadiKara Can you take care of this. I will merge once this is done

[AbdelhadiKara]

Issues created :
Extend timeout
OpenMp issue

[etiennemlb]

the config given by cines has an "off by one" issue in the yaml syntax (the Inès config is slightly old). not an issue but spack does not like that when we want to change the config via the cli.

…

On 5 March 2025 14:11:12 UTC, Thomas Padioleau ***@***.***> wrote: @tpadioleau commented on this pull request. Ok for me, @AbdelhadiKara can you confirm this is the file you used to do your tests ? > " # ***@***.******@***.***~bignuma~consistent_fpcsr+dynamic_dispatch+fortran~ilp64+locking+pic+shared build_system=makefile symbol_suffix=none threads=none arch=linux-rhel8-zen3 which spack +spack config --scope user update --yes-to-all packages What does this line do exactly ? What config does it update, with respect to what ?