You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
AMD's HIP library offers at least a subset of CUDA operations, many of which have similar name and API. How feasible would it be to add this as a feature, to create a similar RustaCUDA API for high level programming with HIP? I'm thinking it would be a feature that creates an additional hip module that exposes a HIP version. This would make porting / targeting both platforms easier.
The text was updated successfully, but these errors were encountered:
Unfortunately, I don't have an AMD GPU. I wouldn't be able to develop something like this. Even if someone did contribute an implementation, I wouldn't be able to test it.
Ah ok. Yeah, currently in the same boat.
I'm asking specifically because I'm working adding support for AMD's ROCm to my autgraph library. That's pretty much done but I haven't been able to test it yet. It's just a quick and dirty port of the things I needed. Long term, I was hoping to have this be external, fully fleshed out, so that it could be used by others, in particular to extend functionality in autograph. And more broadly, use in other projects in the GPGPU space, potentially in a device agnostic way.
AMD's HIP library offers at least a subset of CUDA operations, many of which have similar name and API. How feasible would it be to add this as a feature, to create a similar RustaCUDA API for high level programming with HIP? I'm thinking it would be a feature that creates an additional hip module that exposes a HIP version. This would make porting / targeting both platforms easier.
The text was updated successfully, but these errors were encountered: