KernelAbstractions.jl icon indicating copy to clipboard operation
KernelAbstractions.jl copied to clipboard

Pull `@unroll` out into its own tiny package

Open DilumAluthge opened this issue 4 years ago • 7 comments

The @unroll macro used to live in GPUifyLoops.jl. Now it lives in the extras folder of KernelAbstractions.jl.

It doesn't seem like the @unroll functionality depends on anything else in KernelAbstractions. Would it be okay if I pulled it out into its own small package?

E.g. GEMMKernels.jl uses @unroll but nothing else - no need to make GEMMKernels load all of KernelAbstractions if it only needs @unroll.

DilumAluthge avatar Jan 11 '21 13:01 DilumAluthge

I am not a fan of tiny packages ;)

But you have my encouragement to upstream it https://github.com/JuliaLang/julia/pull/31376

vchuravy avatar Jan 11 '21 15:01 vchuravy

Do you mind if I rebase the branch in https://github.com/JuliaLang/julia/pull/31376? And push some more commits to that branch?

DilumAluthge avatar Jan 12 '21 20:01 DilumAluthge

Please go ahead.

On Tue, Jan 12, 2021 at 3:37 PM Dilum Aluthge [email protected] wrote:

Do you mind if I rebase the branch in JuliaLang/julia#31376 https://github.com/JuliaLang/julia/pull/31376? And push some more commits to that branch?

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/JuliaGPU/KernelAbstractions.jl/issues/199#issuecomment-758944290, or unsubscribe https://github.com/notifications/unsubscribe-auth/AABDO2XYMJORLE3H4H3HIG3SZSXIZANCNFSM4V5RXDSA .

vchuravy avatar Jan 12 '21 20:01 vchuravy

Who do you think the right people are to review that PR?

DilumAluthge avatar Jan 12 '21 20:01 DilumAluthge

Jeff/Keno/Jameson/Yichao/Tim B./Me but the reason I never finished it was because I never came up wit a composable user interface, that should also incorporate parts of @simd

vchuravy avatar Jan 12 '21 21:01 vchuravy

Jeff/Keno/Jameson/Yichao/Tim B./Me but the reason I never finished it was because I never came up wit a composable user interface, that should also incorporate parts of @simd

Maybe we shouldn't let the perfect get it the way of the good :) I think it would be very useful to provide the functionality of the PR to users - be it in a package or in form of a revised PR.

carstenbauer avatar Jul 04 '23 09:07 carstenbauer

Wanna create a small package at JuliaPerf with that? I do think it's worth finding out a good user interface, but the burden of contribution is higher in base.

vchuravy avatar Jul 04 '23 16:07 vchuravy