[Openmp-commits] [PATCH] D107121: [OpenMP] Fix performance regression reported in bug #51235

Sat Jul 31 18:58:15 PDT 2021

tianshilei1992 marked an inline comment as done.
tianshilei1992 added a comment.

In D107121#2918601 <https://reviews.llvm.org/D107121#2918601>, @protze.joachim wrote:

> I execute in BUILD/runtime/runtime-bin and use LD_PRELOAD to make sure, that I use the right libomp.
> The system I use for testing has two sockets with 24 cores each. So, using core 24,25 I make sure to get the extra latency from communication between the sockets.
>
> I added a print statement before the assertion and found that v=-1.
> I added an atomic counter to count the executed tasks for each iteration of the for loop. If the assertion triggers, the task count is 2 instead of 3.

Okay, thanks. Sadly, I still cannot reproduce it, even with the same configuration as yours:
`for i in $(seq 1000); do numactl -C 0,1,29,30 ./gtid & done; wait`
It ran 1000 times and nothing happens…

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D107121/new/

https://reviews.llvm.org/D107121