Attendees
- Rod Burns (Codeplay) Chairperson
- Priyanka Sharma (Fujitsu)
- Vinesh Sukumar (Qualcomm)
- Andrew Wafaa (Arm)
- Robert Cohn (Intel)
- Masahiro Doteguchi (Fujitsu)
- Penporn Koanakatool (Google)
- Jory Burson (Linux Foundation)
- Alison Richards (Intel)
Agenda
- Press Activity
- Webinar Planning
- UXL Strategy and Architecture
- An increasing number of organizations are applying for membership
- New marketing materials available for signage at events
- Agreed to make sure group leaders are able to record their meetings and get the recording and attendee information easily
The group discussed the recent articles that reference the UXL Foundation from Reuters and others. The journalist talked to members back in October and did not share the article ahead of time. The articles don't fully reflect the goal of the foundation, that being to build an inclusive and open ecosystem for all processors. It also contained some inaccuracies, that the projects can already target Nvidia GPUs for example.
The outcome of discussions were to use the webinar to educate people about the objectives of the foundation, and also to request a review of articles in the future.
An action was taken to find out who are authorised to be spokespeople for our Steering Committee organisations.
The group discussed the best format for the webinar on 30th April and how they will contribute. The schedule will be shared. We have one more Steering Committee meeting before the webinar.
Following Andrew Richards' presentation to the group the group discussed what should be the output and next steps.
It was agreed that we didn't get a full chance to discuss all the details of the presentation, but that the organisation does need to form a strategy and high level roadmap for the UXL Foundation.
Some members are keen to figure out how the UXL Foundation fits in with the AI software frameworks. The frameworks currently tend to use different backends for different processors. How can oneDNN and oneMKL in particular be used to enable frameworks to target multiple different processors through a unified API?
It seems like oneDNN can be used to enable new targets, like the Arm CPU target that has been added. It was suggested that a Work Package could be formed to investigate this. The topic can be brought to the Open Source Working Group.
Action taken to organise a series of strategy sessions to define this for the UXL Foundation for publication. Action for Rod to bring together an initial proposal for this based on Andrew's content and other feedback and input so far.