How UXL Foundation redefines the landscape of Accelerated Computing

### Overview At the Open Source Summit in Bilbao, Spain, host Swapnil Bhartiya sits down with Rod Burns, VP of Ecosystem at Codeplay and Chair of the Steering Committee for the newly announced Unified Acceleration Foundation (UXL Foundation). Hosted under the Joint Development Foundation within the Linux Foundation, the UXL Foundation was established to address the fragmented programming landscape across heterogeneous and accelerated computing architectures. By transitioning specifications and open-source projects originally developed under Intel's oneAPI initiative into open governance, the foundation aims to build an open, multivendor, multi-architecture software ecosystem for GPUs, FPGAs, and dedicated AI processors.

### Key Topics - **Evolution of Accelerated Computing:** Compute workloads have rapidly transitioned from pure CPU architectures to heterogeneous environments. Modern high-performance computing (HPC), top supercomputers, and modern AI workloads (such as ChatGPT) rely heavily on accelerators like GPUs. However, software development lacks a unified, cross-vendor programming model. - **Origin and Governance:** While Codeplay trace its roots back to graphics programming tools for video games in the late 1980s and 1990s, its mission expanded to open standards and ecosystems. Intel previously led the oneAPI initiative; establishing the UXL Foundation under the Linux Foundation's Joint Development Foundation creates neutral, open governance to democratize specifications and implementations across industry competitors. - **Founding Member Ecosystem:** The foundation brings together key industry players across hardware IP, cloud, and systems, including Arm, Fujitsu, Google Cloud, Imagination Technologies, Intel, Qualcomm, and Samsung, alongside contributions from Codeplay. - **Expanding Beyond HPC:** Although oneAPI originally concentrated on supercomputers, data centers, and clusters (illustrated by scientific projects like GROMACS using oneAPI and the SYCL standard), the UXL Foundation aims to expand into broader domains, particularly AI and automotive systems requiring safety-critical standards.

### Technical Details The foundation governs core open-source libraries and standard API definitions that serve as foundational building blocks for accelerated software: - **oneMKL (Math Kernel Library):** Provides math operations such as linear algebra (BLAS) and random number generation, enabling developers to target different architectures (including AMD, Nvidia, and Intel GPUs) using a common API. - **oneDNN (Deep Neural Network Library):** Supplies accelerated operations for AI applications and frameworks, with an emphasis on key primitives like convolutions. - **oneDPL (Device Parallel Library):** Implements ISO C++ parallel algorithms (e.g., parallel sort) to dispatch data-parallel tasks to GPUs and other accelerators rather than running sequentially on CPUs.

### Takeaways The primary near-term focus for the UXL Foundation is the smooth migration of oneAPI codebases and specifications into the Linux Foundation governance model. Looking forward, the steering committee is prioritizing conformance and compatibility testing to ensure identical behavioral guarantees across distinct hardware vendors. Success relies on dual engagement: hardware vendors bringing processor targets to the projects, and the developer community actively shaping emerging API standards—especially in fast-evolving AI workflows where standardized interfaces have yet to be established.