ThreadPinningCore

This package will be a backend of ThreadPinning.jl. You may use it directly as a more lightweight alternative. Note, however, that you will need to specify the CPU threads - to which the Julia threads should be pinned - by OS indices ("physical" indices).

Features

Pin Julia threads (get and set their processor affinity), including GC threads (on Julia >= 1.11).
Pin OpenBLAS threads (get and set their processor affinity)
Fake mode (pin threads without actually pinning them, to be used in conjuction with SysInfo.jl's TestSystems)
...

Supported operating systems

Only Linux is fully and officially supported. However, you can install and load the package (using ThreadPinningCore) without any issues on all operating systems. It's only when you call (most) functions that you will see error messages.

Usage

julia> using ThreadPinningCore

julia> getcpuid() # where the calling thread is currently running
232

julia> ispinned() # single CPU thread affinity?
false

julia> printaffinity()
1111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111

julia> pinthread(4) # pin to the CPU thread with ID 4 (OS ordering, not logical order)

julia> getcpuid()
4

julia> printaffinity()
0000100000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000

julia> ispinned()
true

API

See api.jl and the @public/export markers in ThreadPinningCore.jl.

Internal (developer) notes

Fake mode

Noteworthy limitations:

setaffinity(mask), for a mask that highlights more than one CPU-thread, will randomly choose one of the valid CPU-threads and then fake a single-1 affinity mask. Hence, setaffinity(mask); getaffinity() != mask.