Skip to content

Comment on CUDA Tile Open Sourcedparent

Comments

Yes, but why would you want to use this over the other MLIR dialects that are already cross platform?

That's not really the point. The point is that Nvidia is updating a lot of their higher level CUDA tooling to integrate with and compile to Tile IR. So this gives an escape hatch for tools built on top of CUDA to deploy outside the ecosystem.

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.