UXL SIGs : New Contributions, SYCL in Clang and Enabling open standards for AI processors.

During February and March we had a flurry of Special Interest Group (SIG) meetings with the Math, AI and Hardware groups getting together to cover a wide range of fascinating topics.

Once you have subscribed to the mailing list for the relevant SIG, you can catch up with the meetings through the Linux Foundation https://openprofile.dev/ website.

The Language SIG was the first group to meet this year and there was a presentation from Intel’s Tom Honermann on the progress being made to upstream the DPC++ SYCL compiler project to the LLVM Clang project. This was an interesting insight into the technical decisions and work involved in integrating this SYCL compiler into the Clang compiler project.

The Hardware SIG welcomed a group of presenters from Embecosm and the University of Southampton where they showed how they have been using the oneAPI Construction Kit to enable PyTorch custom operators to run on new RISC-V based hardware.

The Math SIG had a presentation from Augustin Degomme from SiPearl who talked about the contributions they have made to the oneMath project, bringing support for Arm processors to the project through the Arm Performance Library. This work makes it possible to target processors from AMD, Arm, Intel and Nvidia using a single math library covering the likes of BLAS, LAPACK and RNG.

The AI SIG saw two presentations, the first from Leping Wang, Beijing Institute of Open Source (BOSC) who discussed the importance of extending the oneDNN project APIs for unified computing platforms. This contribution aims to ensure that oneDNN is the go-library backend for AI frameworks.

The second presentation was made by Ashok Bhat from Arm, discussing the hardware features of Arm that aid in Machine Learning (ML), including Bfloat16 support and matrix multiply extensions. He explained how the software team makes these features available to higher-level frameworks like PyTorch and TensorFlow through the oneDNN library. Ashok also presented some interesting performance figures showing oneDNN with PyTorch on Arm processors.

Sign up to the mailing lists to find out about upcoming UXL Foundation SIG meetings.

Find the meeting notes from all the UXL Foundation SIGs on our GitHub repository, and view the meeting recordings through https://openprofile.dev/ or on our YouTube channel.