Don't have Telegram yet? Try it now!
https://hgpu.org/?p=5223
Extending abstract GPU APIs to shared memory