ziotom78 / ducc0.jl Goto Github PK

Even with release 0.29.0+1 of ducc0_jll the CI builds still fail on Windows. It seems that Yggdrasil's PR#6306 was not enough to fix the problems with Julia, as now the tests are failing because the JLL package does not find a suitable DLL file.

Judging from the source code for ducc0_jll, it seems that if AVX is available then ducc0_jll wants to use an AVX DLL and is not able to fall back to the first available binary (in our case, x86_64). If I run augment_microarchitecture on my Windows 7 system, it sets platform["march"] to avx, and then this happens:

julia> using ducc0_jll

julia> ducc0_jll.is_available()
false

I fear this happens because we skipped the creation of the AVX DLL in the first place!

@giordano, do you have any idea about how to fix this and force Julia to revert to x86_64 if no avx DLL is available? I would avoid fixing platform_augmentation manually…

This issue is used to trigger TagBot; feel free to unsubscribe.

If you haven't already, you should update your TagBot.yml to include issue comment triggers.
Please see this post on Discourse for instructions and more details.

If you'd like for me to do this for you, comment TagBot fix on this issue.
I'll open a PR within a few hours, please be patient!

This is a bit of a shopping list of things to do before the package can be made public.

• When should this happen? Should it happen at all?
  Personally, I'd say we need it to be public if we expect other packages to depend on it, and I already have one third party package (https://github.com/JuliaMath/NFFT.jl) waiting for a release

• Choice of license
  I'm only mentioning this for completeness; I'm totally happy with using a liberal license (like the currently selected MIT) for the glue code in this package. Still, we need to add the appropriate copyright notices / licensing header to all source files.

Anything else? @ziotom78 @LeeoBianchi

It was suggested (JuliaMath/NFFT.jl#109 (comment)) that some of the functionality could be made available for a wider audience by providing implementations for already existing abstract Julia classes (https://github.com/JuliaMath/AbstractFFTs.jl and https://github.com/JuliaMath/NFFT.jl/blob/master/AbstractNFFTs/src/AbstractNFFTs.jl, to be specific). This looks like a very nice idea to me, but I admit I could never do this myself, at least with my current Julia knowledge.

@ziotom78, @LeeoBianchi how high do you estimate the effort necessary for that kind of integration? And should it be done (if at all) inside this package, or would separate small packages depending on Ducc0.jl be more appropriate?