Reduce use of ᶜspecific

[charleskawczynski]

No description provided.

[charleskawczynski]

@tapios, I tried lumping everything into the one function, but I'm getting an error:

ERROR: LoadError: The function body AST defined by this @generated function is not pure. This likely means it contains a closure, a comprehension or a generator.
  | Stacktrace:
  | [1] macro expansion
  | @ /central/scratch/esm/slurm-buildkite/climaatmos-ci/24693/climaatmos-ci/src/prognostic_equations/hyperdiffusion.jl:245 [inlined]
  |  

Is it alright if I break this back down into smaller function calls (which worked) and only do the generated function where it's needed? Or maybe @dennisYatunin has a better idea?

[tapios]

@tapios, I tried lumping everything into the one function, but I'm getting an error:

ERROR: LoadError: The function body AST defined by this @generated function is not pure. This likely means it contains a closure, a comprehension or a generator.
  | Stacktrace:
  | [1] macro expansion
  | @ /central/scratch/esm/slurm-buildkite/climaatmos-ci/24693/climaatmos-ci/src/prognostic_equations/hyperdiffusion.jl:245 [inlined]
  |  

Is it alright if I break this back down into smaller function calls (which worked) and only do the generated function where it's needed? Or maybe @dennisYatunin has a better idea?

I'd like to understand what fails here. The problem is in hyperdiffusion, l. 245. Nothing here needs to change if what we use is the old specific_gs. It looks like the problem is with the changes from specific_gs to all_specific_gs (which weren't needed here from what I know).

I'm fine with breaking this down into smaller functions, but I'm afraid the problem is something else. @dennisYatunin can weigh in.

[tapios]

You need to remove the duplicative specific_gs above. It should be the same thing as all_specific_gs.

[charleskawczynski]

I recommend we leave the duplicate definition until we are able to fully remove ᶜspecific from the cache, at which point we can remove specific_gs trivially (as nothing will depend on it). If I were to try to remove it right now, it would take me some time to figure out how to make all_specific_gs be able to return a field (all_specific_gs returns a NamedTuple of Broadcasted objects) like specific_gs does.

[tapios]

@tapios, I tried lumping everything into the one function, but I'm getting an error:

ERROR: LoadError: The function body AST defined by this @generated function is not pure. This likely means it contains a closure, a comprehension or a generator.
  | Stacktrace:
  | [1] macro expansion
  | @ /central/scratch/esm/slurm-buildkite/climaatmos-ci/24693/climaatmos-ci/src/prognostic_equations/hyperdiffusion.jl:245 [inlined]
  |  

Is it alright if I break this back down into smaller function calls (which worked) and only do the generated function where it's needed? Or maybe @dennisYatunin has a better idea?

Isn't the right thing to do here is using the same logic of looping over scalar names that you used several times in #3858?

[charleskawczynski]

Is it alright if I break this back down into smaller function calls (which worked) and only do the generated function where it's needed? Or maybe @dennisYatunin has a better idea?

@dennisYatunin ?

[tapios]

Is it alright if I break this back down into smaller function calls (which worked) and only do the generated function where it's needed? Or maybe @dennisYatunin has a better idea?

@dennisYatunin ?

As I mentioned, you'd want to use a (corrected form) of foreach_tracer here, restricted to grid-scale variables. Then this issue will disappear.

[charleskawczynski]

As I mentioned, you'd want to use a (corrected form) of foreach_tracer here, restricted to grid-scale variables. Then this issue will disappear.

It's not clear to me how to use foreach_tracer in this function. Are you suggesting to scrap the function altogether and just use foreach_tracer? If so, we'll need foreach_tracer to emit both ρχ_name and χ_symbol, as both are needed in the hyperdiffusion. Specifically, how do we use foreach_tracer for:

    for χ_name in propertynames(ᶜ∇²specific_tracers)
        @. ᶜ∇²specific_tracers.:($$χ_name) = wdivₕ(gradₕ(ᶜspecific.:($$χ_name)))
    end

?