Open alex-w opened 1 week ago
Is this sorting parallelization so crucial for performance that we want to enable all the experimental features of libc++?
I see a good increase in performance with a large solarSystem. However, experimental lib support always sounds a bit scary for "safe and trusted" programs. It has been now built to be not an error if a compiler/library does not support std::execution::par, so I have no clear opinion on that. I understand only CLang/FreeBSD is affected? Maybe enable behind yet another CMake flag for this platform, and instruct users with problems to self-compile? (Any estimate of affected users on this platform?)
Oops, I forget convert this PR to draft, sorry.
This is an experimental PR and in theory these changes affected FreeBSD and macOS. Of course we have very limited number of FreeBSD users and it looks as not critical changes for this platform. The number of macOS users are more than FreeBSD users, but AppleClang has some specifics and the latest release of AppleClang has version 16.0 (macOS 15.0 Sequoia), so, this patch is not applicable here. Probably I should add a separate solution for macOS via using oneDPL library.
Enabling support of parallel algorithms for Clang 17+ and libc++ (like FreeBSD) via adding
-fexperimental-library
option for compiler. Of course other way to use parallel algorithms is adding Intel's oneDPL (Intel's oneAPI) library.See:
Our CI instance FreeBSD/Clang 18 is passed the test
Type of change
Checklist: