Date
Monday, October 26, 2026
Name
LLVM/Offload
Session Type
Pre-Event Session
Description

Presentations and discussions with the audience about the status of the LLVM/Offload project, offload tools and binary formats. We discuss CUDA, HIP, OpenMP, OpenACC, OpenCL and SYCL plans/requirements/use-cases, as well as features and API design, the interaction of Offload with tools and driver artifacts.

Half-day workshop
Abstract

1:00 – 1:05 PM | Introduction / Greeting
Dinos, Yury, Kevin

1:05 – 1:30 PM | Building a Lock-Free GPU Allocator in LLVM
Joseph Huber

1:30 – 1:50 PM | Native Offloading Interfaces Unified via LLVM Offload
Kevin Sala Perades

1:50 – 2:05 PM | SYCL Offloading Update
Yury Plyakhin

2:05 – 2:30 PM | Portable Just-in-Time Compilation of C/C++ Using Proteus
Zane Fink

2:30 – 3:00 PM | Bringing OpenACC to the Offload Stack
Ivan R. Ivanov

3:00 – 3:30 PM | Break

3:30 – 3:55 PM | LLVM/Offload as a GPU Runtime Abstraction Layer for SYCL
Piotr Balcer

3:55 – 4:10 PM | OpenMP SPIR-V Offload
Konstantinos Parasyris

4:20 – 4:45 PM | CUDA and HIP in ClangIR: Merging Host and Device Compilation for Cross-Boundary Optimization
David Rivera

4:45 – 5:00 PM | Using Host Compilation Analysis to Inform Device-Side Optimization in SYCL
Victor Mustya

5:00 PM | Closing Remarks
Yury, Kevin, Dinos

Location Name
Grand Ballroom ABC