Valentin Churavy
Valentin Churavy
Yeah essentially matching: ``` julia> zeros(Base.OneTo(10)) 10-element Vector{Float64}: 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 ```
I assume that the MWE I provided in the issue, is not minimal enough? ``` OPENBLAS_NUM_THREADS=1 ENABLE_JITPROFILING=1 perf record --call-graph dwarf -k 1 julia -g2 -e "fib(x) = x
Probably better than invalidating `Base.rand` we should have `KA.rand` like `KA.ones`
I think it does make sense to query the size of maximum number of workgroup size, and how many SM/Core are available. We just need to be defensive and allow...
Oof, I have been avoiding adding a notion of device (besides the device switching API https://github.com/JuliaGPU/KernelAbstractions.jl/blob/1ac546fc59cc611d749fa7a50e4a1efa3393851b/src/KernelAbstractions.jl#L601-L627)
I would say yes. But maybe @maleadt has some thoughts.
I tagged a KernelAbstraction release, but importantly you can't use `@Const`. If you annotate a function with `@Const` it is still possible to hit this bug
It's sadly still broken if the user uses `@Const`, so let's leave it open.
> I understood Julia emits traps/unreachables only in debug mode so this should be non issue for performance runs? No we emit trap/unreachable generally; essentially, we are modeling exceptions here....
>If the programmer doesn't enable bounds checking It's the other way around. Bounds-checking is enabled by default, and it is up to the programmer to locally opt-out. > nor uses...