Don't have Telegram yet? Try it now!
https://hgpu.org/?p=9079
Improving Performance Portability in OpenCL Programs