Remove unnecessary extra wrapping of valtypes

[KevinDCarlson]

There's no difference in functionality between f(::Type{Val{n}}) where n =... and f(::Val{n}) where n = ..., but the latter is both cleaner and the intended usage: you plug in the unique instance of the type Val{n} rather than plugging in the type. I'm not able to find any difference in performance between the two approaches but I wanted to clean this up anyway.

[epatters]

Have you done benchmarks? I'm not sure how it shakes out in this case, but my understanding is that there is sometimes a reason to prefer ::Type{Val{T}} to ::Val{T}, namely that Val{t}, where t is a literal, is a constant available at compile time and can therefore be constant propagated, whereas Val(t) is value that might only be constructed and dispatched at runtime.

Or, at least at one time in the past, this was a distinction that bit me. Perhaps, due to compiler improvements, it is no longer relevant. I would suggest getting advice from someone who actually understands how the Julia compiler works (not me).

[jpfairbanks]

This PR would make things much more idiomatic, but we do need the benchmarks, since this code is so critical to decapodes performance. A quick and dirty way to check, since you are touching such a large fraction of the codebase is to just run the tests before and after the change. If the tests take longer, then likely this is the culprit.

[KevinDCarlson]

This PR would make things much more idiomatic, but we do need the benchmarks, since this code is so critical to decapodes performance. A quick and dirty way to check, since you are touching such a large fraction of the codebase is to just run the tests before and after the change. If the tests take longer, then likely this is the culprit.

Oh I definitely did that, it gets like 1-2% faster IIRC. Also tried some synthetic tests but wasn’t able to show the difference ever affects performance.