diff --git a/meetings/notes/2025-04-22.rst b/meetings/notes/2025-04-22.rst index 1b60a1b..71aff89 100644 --- a/meetings/notes/2025-04-22.rst +++ b/meetings/notes/2025-04-22.rst @@ -116,4 +116,4 @@ unsure about the problem statement and the benefits of this proposal. John agreed to bring in Vasanth to explain the proposal in more detail. -.. _`slides`: ../presentations/2025-03-25-UXLCIPoC.pdf +.. _`slides`: ../presentations/2025-04-22-UXLCIPoC.pdf diff --git a/meetings/notes/2025-05-27.rst b/meetings/notes/2025-05-27.rst new file mode 100644 index 0000000..5fbdddf --- /dev/null +++ b/meetings/notes/2025-05-27.rst @@ -0,0 +1,114 @@ +========================== +Open Source WG: 05/27/2025 +========================== + +Recording: A recording of the meeting is available in the Linux Foundation https://openprofile.dev/ profile. If you are +a member of the Working Group you can access this through your account. + +Attendees +========= + +* Megan Knight - Arm +* Nick Dingle - Arm + +* Aaron Dron - Codeplay +* Rod Burns - Codeplay + +* Ragesh Hajela - Fujitsu + +* Kevan Ahmadi - Imagination Technologies + +* Michael Voss - Intel +* John Melonakos - Intel +* Timmie Smith - Intel +* Maria Petrova - Intel +* Alexey Kukanov - Intel +* Mourad Gouicem - Intel +* Nikolay Petrov - Intel +* Augustin Degomme - Intel +* Vasanth Tovinkere - Intel +* Maria Kraynyuk - Intel +* Alison Richards - Intel + +* Roman Zhukov - Red Hat + +* Biagio Cosenza - University of Salerno + + +Next Steps +========== + +* John to send out Vasanth's presentation slides to the group. (see below) +* John to set up a GitHub discussion thread for further conversation on CPU inclusivity and the C API specification. +* John to reach out to Qualcomm to rekindle the discussion on CPU inclusivity and compiler leveraging. +* John to organize a focused discussion with the oneAPI Math (oneMKL) team on specific domains (BLAS, LAPACK, FFT, RNG) + regarding CPU inclusivity. +* John to reach out to software ISVs to participate in the GitHub discussion on CPU inclusivity. +* John to share the sanctions discussion topic with Larry at Intel and bring it up in the next meeting. +* Mike to send the Linux Foundation discussion link about sanctions to John and Megan via email. + +Summary + +Security Work Package Progress Review (`UXL community infrastructure slides`_) +===================================== + +Rod presented on security work package progress, noting that most issues had been resolved or moved to the "in progress" +column. He highlighted Coverity and OSS Fuzz as potential pain points, particularly regarding Google account +requirements for OSS Fuzz, and mentioned that a guide was being developed to assist users. + +Security Testing Flexibility for Projects +========================================= + +The team discussed security and quality testing approaches for UXL Foundation projects, agreeing that projects can +determine their own priorities on a case-by-case basis rather than having a unified set of rules. They decided to share +best practices and positive experiences from different projects rather than establishing one-size-fits-all requirements +for static analysis. The group also addressed questions about security email responses, maintainer selection processes, +and external testing requirements, with Megan suggesting they review OpenSSF materials for guidance on maintainer +selection. Rod noted that while current OpenSSF scorecard scores were decent, security representatives should review +their current threats and next steps, potentially through Slack discussions. + +Hardware Runners Status and Planning +==================================== + +Rod discussed the status of various hardware runners, noting that the Codeplay host runner would go offline in a few +days but that the free ARM runners were now available. He mentioned three in-progress runners from Codeplay and Intel, +including the BattleMage B580 and Nvidia H100, with Intel planning to provide more B580s. Rod also highlighted the +completion of the Intel GPU Max 1550 migration and thanked projects for providing infrastructure requirements. He +identified gaps in testing, such as PowerPC and RISC-V requirements, and suggested potential solutions like emulation or +hardware options. Rod also mentioned exploring Common Forge for binary releases and ongoing discussions with Intel about +packaging DPC++. He encouraged further conversations about release planning and naming strategies. + +API CPU Inclusivity Customer Concerns (`CPU-inclusive C API slides`_) +===================================== + +Vasanth presented on customer pain points regarding API specifications, particularly around CPU inclusivity and library +support. Customers expressed concerns about the need to accommodate both CPU and accelerator paths, with questions about +default targets and fallback mechanisms. The discussion highlighted the desire for a stable abstraction layer that would +allow software innovation while providing hardware vendors with a consistent API to implement. Customers requested CPU +inclusivity as a first-order discussion, with a preference for reference implementations and the ability to set +different policy settings for different libraries, while also emphasizing the importance of conformance tests. + +Math and Imaging Library Proposals +================================== + +Vasanth discussed the primary requests regarding math and imaging libraries, noting that 80-90% of the requests centered +around math functions, with the remaining 20% focusing on imaging library specifications. He explained that discussions +had culminated in proposals to form work groups at the UXL level to address these issues. Maria asked about the +programming model for C API, particularly in relation to offloading tasks to GPUs and accelerators. Vasanth clarified +that the initial pain point was resistance to changing existing C APIs to a new specification that required queuing and +target selection, which software vendors found burdensome. He emphasized the need for a unified abstraction that would +reduce workload and allow incremental implementation by hardware vendors, with a fallback option to CPU for seamless +integration and testing. + +Compiler Support for CPU Inclusivity +==================================== + +The meeting focused on discussing compiler support for CPU inclusivity, particularly for oneMath projects. Vasanth +shared feedback from various stakeholders, including Qualcomm, who expressed interest in leveraging the compiler for +specific work. The group agreed to create a GitHub discussion thread to continue the conversation and gather input from +those who couldn't attend. + + + +.. _`UXL community infrastructure slides`: ../presentations/2025-05-27-UXLCIPoC.pdf +.. _`CPU-inclusive C API slides`: ../presentations/2025-05-27-UXL-Library-brainstorm.pptx diff --git a/meetings/notes/README.rst b/meetings/notes/README.rst index 3359be1..ff78ce6 100644 --- a/meetings/notes/README.rst +++ b/meetings/notes/README.rst @@ -2,6 +2,10 @@ Meeting Notes =============== +Past Meetings +============= + +* `2025-05-27 <2025-05-27.rst>`__ * `2025-04-22 <2025-04-22.rst>`__ * `2025-03-25 <2025-03-25.rst>`__ * `2025-02-25 <2025-02-25.rst>`__ diff --git a/meetings/presentations/2025-05-27-UXL-Library-brainstorm.pptx b/meetings/presentations/2025-05-27-UXL-Library-brainstorm.pptx new file mode 100644 index 0000000..e77d85a Binary files /dev/null and b/meetings/presentations/2025-05-27-UXL-Library-brainstorm.pptx differ diff --git a/meetings/presentations/2025-05-27-UXLCIPoC.pdf b/meetings/presentations/2025-05-27-UXLCIPoC.pdf new file mode 100644 index 0000000..6a572ff Binary files /dev/null and b/meetings/presentations/2025-05-27-UXLCIPoC.pdf differ