Skip to content

Comment on Intel proposes XeGPU dialect for LLVM MLIRparent

Comments

It looks like offload is mostly about generic abstractions for accelerator devices. XeGPU on the other hand seems to be a way of encoding Xe-specific hardware features/instructions, so pretty different? I don't see how these are anything but orthogonal.

AboutSource Built by g1lg1l

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