search

pnnl / lamellar-runtime

Lamellar is an asynchronous tasking runtime for HPC systems developed in RUST
Other
43 stars 5 forks source link

CUDA as backend? #30

Open Wandalen opened 1 year ago

Wandalen commented 1 year ago

Hi! Can it use CUDA as backend?

rdfriese commented 1 year ago

We don't directly support CUDA as a backend, but we can integrate with other crates that target accelerators.

For example, we also develop the mcl-rs crate, a Rust frontend around the mcl library which enables asynchronous execution of OpenCL kernels on heterogeneous systems with multiple types of accelerators.

How we have targeted GPUs in the past was to use Lamellar as the distributed data manager, and then wrap kernel tasks within Lamellar active messages

Wandalen commented 1 year ago

I see. Thnaks!