Add single precision builds and tests in the CI

[valassi]

This is the second part of #167, which initially had this name but actually had two parts

• Modify Makefiles and headers to allow external choice of float/double: this is now only in #167 and is done by #213
• Based on the above, modify the github CI infrestaruture

[valassi]

@hageboeck you want to do this? I would start with eemumu only (wherethe above is done).

About ggtgg that requires much more work (not only porting #213, but also porting various fixes to the tests from other PRs, and possibly additional fixes in the c++/cuda), before the CI stage.

Thanks Andrea

[hageboeck]

I'll have a look on how to activate a build & test for eemumu. I understand that for gg, it's better to wait?

[valassi]

I'll have a look on how to activate a build & test for eemumu. I understand that for gg, it's better to wait?

Thanks! Yes just wait for gg*...

[hageboeck]

• [x] eemumu is addressed in #217.
• [ ] gg* to be done.

[hageboeck]

Reopening (github closed it), since gg* is missing.

[valassi]

Reviewing old issues... @hageboeck I am tempted to close this. I think the real need was to have AT LEAST ONE float build for both c++ and cuda in the CI. I am not sure we need to have every single process there (and even, we are testing epochX cudacpp there, it would instead be interestung to add at least one sycl/kokkos/alpaka eventually).

This is to say: I do not think we also need ggtt and ggttgg float from epochX/cudacpp in the CI. But then, as you prefer.

Thanks! Andrea

[hageboeck]

Less CI jobs are of course better as long as we cover all code paths. 👍

[valassi]

Hi @hageboeck thanks again for this. I am closing this. In the meantime we have also added mixed precision to the CI (see #580, #581, MR #582).

This can beclosed.