The new LLVM/Offload serves as a unification point for all offloading languages (CUDA, HIP, SYCL, OpenMP, OpenACC, ...) and the various backends (CPU, NVIDIA GPU, AMD GPU, ...). In this workshop we look at the new subproject, discuss goals, and current development efforts. Our presenters will discuss the newest capabilities and ongoing efforts such that attendees can get a head start in joining the effort and reuse the system for their needs.

Agenda:
8:00 AM - 8:10 AM: Welcome, Johannes Doerfert
8:10 AM - 10:00 AM:

1. OMPT Device Support in LLVM, Dhruva Chakrabarti (AMD)
2. Xbc An Extensible Compiler For Heterogeneous Computing, Fabian Mora (University of Delaware)
3. Towards Rust (GPU) Offload, Manuel Drehwald (University of Toronto)
4. Thoughts and Results for an Offload-Specific Sanitizer, Johannes Doerfert (LLNL)
5. Offloading with Flang via OpenMP codistribute, Johannes Doerfert (LLNL), Ivan Ivanov (Tokyo Institute of Technology)
6. Building GPU Runtimes With The LLVM Multi-Lib Infrastructure, Joseph Huber (AMD)

10:00 AM - 10:30 AM: Break
10:30 AM - 11:50 AM: Open Discussion
11:50 AM - 12:00 PM: Close, Johannes Doerfert