[PATCH] D147493: [ELF] Cap parallel::strategy to 8 cores when --threads= is unspecified

[Mateusz Guzik via Phabricator via llvm-commits]

mjguzik added a comment.

In D147493#4243823 <https://reviews.llvm.org/D147493#4243823>, @andrewng wrote:

>> Thanks for the results. Does not look like the win is necessarily there, and even if so, it is unclear if it is justified by the computing power thrown at it.
>
> I guess that somewhat depends on what is important to you, i.e. performance per cycle or wall time duration to complete task.

I'm trying to say the reported difference is very easily within measurement error.

What most likely happens apart from it is that the cycle count spent on it skyrockets, making the entire thing into a net *loss*.

Unfortunately I have negative Windows clue, so if you can't take a closer look maybe someone else will volunteer.

>> That is to say, can you redo these while collecting cycle counts?
>
> I'm not setup to do that and IIRC (like most things) it's not so trivial to do so on Windows. But as I mentioned above, it partly comes down to what is considered "good" performance.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D147493/new/

https://reviews.llvm.org/D147493