Valentin Churavy comments

Results 1413 comments of


                                            Valentin Churavy

Auto-tuning workgroupsize when localmem consumption depends on it

This is #11 KA doesn't support dynamic shared memory.

Auto-tuning workgroupsize when localmem consumption depends on it

No #11 was started before we added auto-tuning, and stalled since no-one had a clear need for it.

CUDADevice not defined

You need to import CUDAKernels which contains the CUDA implementation of KernelAbstractions

`nested task error: unsupported or misplaced expression aliasscope` when using `julia --compile=min`

Can you check that https://github.com/JuliaLang/julia/pull/40275 fixes the issue?

eps(::AbstractFloat) inside a kernel doesn't compile on Julia 1.5.1

@jakebolewski I must have been to tired the last time I read your message, and didn't read the part about `@noinline`. `@noinline` is something we explicitly honor (see https://github.com/JuliaGPU/KernelAbstractions.jl/blob/8c7052d23c32d6950ca9abb326af144cce8dae1c/src/compiler/pass.jl#L15-L19) mostly...

eps(::AbstractFloat) inside a kernel doesn't compile on Julia 1.5.1

Well it depends, even for the GPU it is probably good to outline these error calls. But yes we can make this context dependent.

Broken link to `@uniform` in docs

It should point to https://github.com/JuliaGPU/KernelAbstractions.jl/blob/17469f21826970e4ff14a938e16159c2fb1df524/src/KernelAbstractions.jl#L163

Docs for AMD support.

cc: @jpsamaroo you can use `AMDGPU` and `ROCKernels` as the backend

Degraded performance using kwargs in kernels

Your measuring is okayish :) I need to add some better infrastructure. @lcw has it basically correct. keyword arguments require sorting and make the code more sensitive and more prone...