[PATCH] D115416: [lld-macho] Prevent writing map files on the critical path

[Vincent Lee via Phabricator via llvm-commits]

thevinster added a comment.

Not meant as a commit for submitting, but wanted to poll the audience on what pattern we'd like to use with threadPool. We essentially have three options here:

- Option 1 - continue using `parallelForEach` methods
- Option 2 - Re-use our declared class level `threadPool`, but areas where we only want to wait on a subset of threads, we have to re-implement the fork-join model by declaring some vector and waiting on those threads that are pushed back to the vector
- Option 3 - Declare class-level and method-level `threadPools`. Simpifies the for loop from Option 2, but might be harder to manage since we now have to control N number of threadPools with a potential of N different concurrency settings.

I did run a benchmark across all options and there's no reasonable performance improvement with either options. So, it seems like this all comes down to syntax and what looks best.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D115416/new/

https://reviews.llvm.org/D115416