Add Tests/OpenMP/atomicAdd

[WeiqunZhang]

No description provided.

[WeiqunZhang]

Oh, I did it the original way because the code does not work with GPU unless managed memory is used, but it does work even when OMP is off.

[WeiqunZhang]

If we let it run for GPU builds, we have to fix the source code. Use RunOn::Device and add a streamSychronize, and possible disable tiling when GPU is on. Also note that when both GPU and OMP are on, atomicAdd and lockAdd will do it on GPU. That effectively disable OMP.

[ax3l]

Oh, ok. I checked the logs and I could not find the test being built in it o.0

[WeiqunZhang]

Oh, I had an extra NOT in the logic of testing the backend

[WeiqunZhang]

It should read AMReX_GPU_BACKEND STREQUAL NONE, but somehow I had an extra NOT.

[ax3l]

Looking through the build logs: should this show up as Test_OpenMP...?

[ax3l]

Now I see the Test_OpenMP_atomicAdd_*d targets being built.

[ax3l]

@WeiqunZhang should this test be included in AMReX_TEST_TYPE=Small so we can run it in CI of package managers (e.g., conda-forge #3795)?

[WeiqunZhang]

@WeiqunZhang should this test be included in AMReX_TEST_TYPE=Small so we can run it in CI of package managers (e.g., conda-forge #3795)?

Sure