Don't have Telegram yet? Try it now!
https://hgpu.org/?p=14440
Optimizing OpenCL Local Work Group Size With Machine Learning