-
Notifications
You must be signed in to change notification settings - Fork 18
Open
Description
Using HIP (AMD's answer to CUDA)
https://rocm.docs.amd.com/projects/HIP/en/latest/what_is_hip.html
we should be able to write a backend for AMD GPUs
This is because the existing CUDA code-base should be easily convertible to HIP (at least according to advertising for HIP) using their hipify tool https://rocm.docs.amd.com/projects/HIPIFY/en/amd-staging/index.html
(It also seemingly supports cusparse)
There also is a re-implementation of thrust in HIP
https://github.com/ROCm/rocm-libraries/tree/develop
which should help with the device vector and memory management
I would expect this to be a straightforward task, the risks are of course that some details may take time to figure out or find support for.
Metadata
Metadata
Assignees
Labels
No labels