Blog
UXL Foundation: Advancing Portable Acceleration Across Architectures in 2025 — and What’s Next for 2026
March 10, 2026
On January 15, 2026, the UXL Foundation community gathered for its annual member meeting to reflect on progress made in 2025 and outline priorities for the year ahead. The discussion highlighted steady momentum across the ecosystem—from expanding architecture support and advancing open-source libraries to strengthening governance and community engagement.
The UXL Foundation was established to guide the development of SYCL-based, cross-architecture software and promote an open, vendor-neutral acceleration ecosystem. Through collaborative development and open standards, the foundation aims to enable developers to build high-performance applications that run across a diverse range of hardware platforms.
In 2025, the community made significant progress toward that goal.
Demonstrating Portable Performance
One of the central themes of the past year was portable performance across architectures, particularly for HPC and AI workloads.
At UXL DevSummit 2025, several sessions showcased practical advances in enabling high-performance computing across hardware platforms. These included demonstrations of oneDNN running on ARM processors and optimizations to math operators integrated into the oneMath/oneMKL ecosystem.
Developers also shared techniques for leveraging Arm machine learning features, including BF16 and matrix extensions, enabling improved performance for popular frameworks such as PyTorch and TensorFlow through oneDNN.
Another milestone was the oneDAL on ARM developer story, which highlighted portability progress on Fujitsu’s MONAKA architecture. Working with partners including Arm and Fujitsu, the community continued advancing the conversation around energy-efficient AI, emphasizing both performance and efficiency across platforms.
The AI Special Interest Group (AI-SIG) also continued to expand, with nearly 100 members and more than 40 discussion topics. The group meets quarterly and is co-led by Penporn Koanantakool (Google) and Jianhui Li (Intel), helping drive collaboration around AI workloads within the UXL ecosystem.
Expanding Architecture Support
Support for new architectures continued to grow throughout the year.
The release of oneAPI Construction Kit 4.0 introduced RISC-V host CPU support, opening the door for portable acceleration stacks on emerging open hardware platforms.
Meanwhile, conversations within the one Memory Centric Compute (oneMCC) working group explored new architectural approaches that place memory closer to compute. These discussions reflect growing interest in memory-centric computing models as workloads continue to scale in both data size and complexity.
Advancing the Open Source Library Ecosystem
UXL’s open-source libraries continued evolving throughout 2025, with updates across projects such as oneDNN, oneMath, and oneCCL.
A key focus area was multi-vendor portability. Technical webinars and community presentations emphasized enabling these libraries to work effectively across multiple hardware platforms.
oneMath Progress
The oneMath library introduced several important updates across releases v0.7 through v0.9, including:
- New ARM performance backends for BLAS, LAPACK, RNG, and DFT
- Support for the rocSPARSE backend
- Continuous integration testing for the Arm Performance Libraries backend
- Support for AdaptiveCpp with cuFFT and rocFFT backends
- Introduction of backend maintainer roles to improve project sustainability
One notable milestone was the addition of an ARM backend for aarch64, significantly expanding portability for ARM-based platforms.
At the same time, the community continues addressing gaps such as public CI infrastructure and expanding the pool of community maintainers.
Strengthening Open Source Processes
The Open Source Working Group also developed an active security tracker workflow, with a process document currently under review. The goal is to establish consistent security practices across all UXL organization repositories, helping improve transparency and coordination across projects.
From Demonstrations to Developer Guidance
Looking ahead to 2026, the foundation is shifting its focus from isolated demos to practical guidance that helps developers build portable software more easily.
One major initiative is a documentation-first developer experience, including prescriptive interoperability and conformance guidance. These resources will help developers understand how to integrate UXL technologies into their workflows and build applications that run across architectures.
To demonstrate real-world portability, the community plans to deliver three public proof-of-concept implementations, targeting areas such as:
- Memory-centric compute (oneMCC)
- RISC-V portability
- Framework interoperability for HPC and AI
Launching the oneMCC Special Interest Group
A major technical initiative for 2026 is the launch of the one Memory Centric Compute (oneMCC) Special Interest Group.
The group will explore architectures designed around processing-in-memory (PIM) and processing-near-memory (PNM) models. Early work will include drafting extensions and publishing sample kernels for workloads such as reductions and embeddings.
Industry interest in the effort is already growing, with companies including Samsung, SK Hynix, Micron, and Qualcomm evaluating participation.
Strengthening Governance and Community Transparency
Another priority for 2026 is reinforcing the foundation’s open and vendor-neutral governance model.
Planned improvements include:
- Public CI dashboards
- Publishing Steering Committee minutes
- Sharing monthly health metrics through GitHub
- Member-driven roadmaps
These steps are intended to strengthen trust and reinforce the foundation’s position as an open, cross-vendor acceleration framework.
Growing the Ecosystem
Community awareness and education will also be key themes for the coming year.
New initiatives include:
Awareness and Technical Outreach
The foundation will launch “UXL in 5,” a quarterly series of short videos and blogs highlighting portability stories across architectures such as ARM, RISC-V, and memory-centric systems.
UXL contributors also plan to increase technical talks and tutorials at industry events including Open Source Summit North America, RISC-V ecosystem events, and major HPC conferences such as ISC and SC.
Developer Education
To support new contributors and developers, the community will introduce several educational programs:
- UXL 101, a short micro-course with four introductory modules
- University starter kits built around oneDNN and oneMath
- Quarterly mentor office hours for community questions and guidance
Measuring Progress in 2026
The foundation has defined several key metrics to track ecosystem growth and stability:
Technology
- Three public proof-of-concept demonstrations
- Updated domain profiles and specifications
- Public CI dashboards
Community
- 20% growth in GitHub stars
- Increased working group engagement
Ecosystem
- Two new steering members
- Ten new contributing member organizations
Education
- 500 new newsletter subscribers
- 25% growth in documentation traffic
- Three university pilot programs
Building the Future of Portable Acceleration
As heterogeneous computing continues to expand, developers increasingly need tools that enable applications to run across architectures without sacrificing performance.
The UXL Foundation’s work—spanning SYCL development, open-source libraries, and cross-vendor collaboration—aims to make that vision possible.
By focusing on portability, transparency, and practical interoperability, the community is building an ecosystem where accelerated computing works across vendors, domains, and emerging architectures.The progress of 2025 and the roadmap for 2026 reflect a shared goal: making open, portable acceleration a reality for the next generation of HPC and AI applications.