Infinite loop if core pinning fails #103
Comments
|
This file could also benefit from a complete rewrite. You can already send your PR I think. |
IagoAbal
added a commit
to IagoAbal/parmap
that referenced
this issue
Jul 5, 2021
This is known to happen, under some circumstances, on Apple M1. Plus code cleanup. Closes rdicosmo#103
IagoAbal
added a commit
to IagoAbal/parmap
that referenced
this issue
Jul 5, 2021
This is known to happen, under some circumstances, on Apple M1. Plus code cleanup. Closes rdicosmo#103
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Core pinning does not always work on Apple M1 and as a result Parmap enters an infinite loop. See semgrep/semgrep#2432 and facebook/infer#1410. We at @returntocorp have only observed this behavior with HomeBrew builds, presumably because in HomeBrew's build environment HAVE_MACH_THREAD_POLICY_H is set? Even then, this only occurs under certain values of
ncoresandchunksize. So running on 8 files using 8 cores and chunk size of 1 works, but if there are 9 files it hangs. Running on 9 files with 4 cores and a chunk size of 2 it does not hang.Our current work around is to simply disable core pinning.
Independently of why core pinning is failing in this case, could it make sense to change
while (finished==0) { ... }todo { ... } while (finished==0 && w>0)or bound the number of iterations of that loop in some other way ? I am happy to open a PR if we agree on this.The text was updated successfully, but these errors were encountered: