As briefly discussed in https://github.com/JuliaLang/julia/pull/22846 it would be quite useful for some algorithms to know the CPU cache size in order to branch based on the input size when looping over arrays. Right now I picked an arbitrary power of two that just happened to work on my laptop, and that really seems like a slippery sloppy slope.
On Linux, it looks like I can get the right information from /proc/cpuinfo, but I am somewhat clueless about and unable to test how to get this information on any other platform. So I'm a little wary of venturing into this area unguided ...
Does someone else have any experience with this? Or is this information already available somewhere undocumented? ... or documented? :slightly_frowning_face:
dalum
All 6 comments
KristofferC
on 21 Jul 2017
There is also https://github.com/m-j-w/CpuId.jl
simonbyrne
on 25 Jul 2017
We can have tighter integration of hwloc but don't use the cpuid package.
yuyichao
on 25 Jul 2017
Is there anything to be done in base?
vchuravy
on 29 Jul 2017
Depending on if there's any code in base that want to use this we might want to integrate it in base.
yuyichao
on 29 Jul 2017
Let's hold off on that. It's easier to move it in if that turns out to be good than move it out.
StefanKarpinski
on 29 Jul 2017
Related issues
musm
路
3Comments
omus
路
3Comments
TotalVerb
路
3Comments
wilburtownsend
路
3Comments
ararslan
路
3Comments
Most helpful comment
https://github.com/JuliaParallel/Hwloc.jl