Don't have Telegram yet? Try it now!
https://hgpu.org/?p=12795
Using hybrid GPU/CPU kernel splitting to accelerate spherical convolutions