[CFP][LLVM-DEV'25] LLVM/Offload Workshop

Announcements
, , , ,
1

As announced in the conference agenda, there will be a GPU / offloading half-day workshop at the 2025 LLVM developer’s meeting (On October 27th 8:30 AM - 12:30 PM).

LLVM/Offload - Where are we, where are we going?
Presentations and discussions on the status of the new LLVM/Offload project. We discuss CUDA, HIP, OpenMP, and SYCL plans, as well as features and API design. With the speakers and audience we will engage in discussions about the next steps and requirements for various use cases.

For this workshop, I would like to structure it as series of mega-roundtables, centered around an introductory talk.That is, open discussion interleaved with presentations. Please DM @arsenm by October 6th topic ideas, and if you would like to give a 5-10 minute introductory talk.

Open discussion potential topics:

  • Runtime library support
  • Offload API
  • Subregister handling in regalloc/scheduling
  • SPIR-V
  • Language support work
  • ???
5 Likes
2

I’d like to talk a lot about where we came from and where we are going with GPU offloading in LLVM. Ideally we’ll be able to have some longer discussions on the future of things like SPIR-V, SYCL, and MLIR runtime lowerings. I’ll likely modify my conference slides.

1 Like