Skip to main content
Best practices
2 votes
0 replies
0 views

How to signal a multithreaded worker pool at very high throughput?

Good question. I avoid atomics, so I might not be the best person to comment on this, but it seems that the amount of computational work you're doing in a single step is insufficient for performance ...
Simon Goater's user avatar
  • 2,067
Best practices
1 vote
0 replies
0 views

How to signal a multithreaded worker pool at very high throughput?

If you have jobs[0] and jobs[1] on the same cache line (64 bytes on x86-64 platforms), then false sharing may be an issue. You may want to ensure that each element of the jobs array is on its own ...
Andreas Wenzel's user avatar
Best practices
1 vote
0 replies
0 views

How to signal a multithreaded worker pool at very high throughput?

It may be helpful if you could provide a minimal reproducible example of the problem, which includes a function main and all #include directives, as well as the exact input required to reproduce the ...
Andreas Wenzel's user avatar
Best practices
1 vote
0 replies
0 views

How to signal a multithreaded worker pool at very high throughput?

Thanks for the answer. I'm not quite sure of the benefit of workers "pulling", since I'd still have to choose a chunk length per-job (I immediately know the total work length before ...
lzg's user avatar
  • 123

Only top scored, non community-wiki answers of a minimum length are eligible