MAM4xx sycl errors

[oksanaguba]

Tagging @odiazib @overfelt @tcclevenger @bartgol .

There is an error i do not know how to resolve

In file included from /home/onguba/scream-b2/components/../externals/mam4xx/src/mam4xx/calcsize.hpp:14:
/home/onguba/scream-b2/components/../externals/mam4xx/src/mam4xx/conversions.hpp:253:19: error: SYCL kernel cannot call through a function pointer
  253 |   const auto ws = wsat(T, p);
      |                   ^

pointing to code

KOKKOS_INLINE_FUNCTION Real relative_humidity_from_vapor_mixing_ratio(
    Real w, Real T, Real p,
    Real (*wsat)(Real, Real) = saturation_mixing_ratio_hardy) {
  const auto ws = wsat(T, p);
  return w / ws;
}

Other errros so far are only about printf, where i use the same workaround as in homme,

#ifdef KOKKOS_ENABLE_SYCL
    Kokkos::printf ("k = %d\n",k);
#else
    printf ("k = %d\n",k);
#endif

though i am not sure why it is not just kokkos::printf for all builds (maybe i tried that and it did not work).

[tcclevenger]

Re printf: I believe this is something that is fixed in later Kokkos versions (I think 4.2 on, and E3SM is using 3.7). Once we can update kokkos I think these workarounds won't be necessary.

[oksanaguba]

@tcclevenger thanks -- my kokkos is pretty updated (branch develop from Apr 26, 2024)

[bartgol]

I guess passing a function pointer is problematic. Perhaps you could template on a function type, and pass a lambda.

[tcclevenger]

@tcclevenger thanks -- my kokkos is pretty updated (branch develop from Apr 26, 2024)

Oh interesting. I can find some time to look into the specifics of that change and the SYCL backend.. But if your using april develop those changes would definitely be included.

[odiazib]

Hi @oksanaguba, do you require MAM4XX for your simulation? If not, you can disable it by using -DSCREAM_ENABLE_MAM=OFF as suggested by @jeff-cohere in the Emaxx change discussion on Slack. I will proceed to open an issue in the MAM4XX repository to address this.