[llvm-bugs] [Bug 49588] Clang/OpenMP schedule(dynamic): severe scaling issue, up to 2.5x slower than GCC/OpenMP

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=49588

Kim Walisch <kim.walisch at gmail.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |WORKSFORME
             Status|NEW                         |RESOLVED

--- Comment #3 from Kim Walisch <kim.walisch at gmail.com> ---
I was able to fix my particular scaling issue by using
schedule(monotonic:dynamic, 1) instead of schedule(dynamic, 1) (which is the
same as schedule(nonmonotonic:dynamic, 1)).

My algorithms work best when the iterations are executed in "increasing logical
iteration order". This partially explains why the default schedule(dynamic, 1)
is not well suited for my particular use case as iterations may be executed in
any order. I will note however that GCC's schedule(nonmonotonic:dynamic, 1)
scales very nicely in my benchmarks, so I do think that there is room for
improvement in Clang/libomp's schedule(nonmonotonic:dynamic, 1) implementation.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20210922/8e5a04a3/attachment.html>