There are Office Hours for the GNU Toolchain every 3rd Thursday!

There was broad consensus that we should send out the Office Hours minutes both 
to draw attention
to the Office Hours and make it easy to see what was discussed without having 
to go to the wiki.

I'll start posting them monthly the list.

Please see https://gcc.gnu.org/wiki/OfficeHours for more details.

The following was the agenda and discussions for both office hours:

Meeting: 2026-07-16 @ 1100h EST5EDT

Attendees:

Carlos, David, Malcolm, Jason, Adhemerval, Andrew, Claudio, Michael, Ricahrd, 
Saurov, Thomas, Tobias.

Agenda:

[Carlos] To answer last question about SIP for the room. We do not have SIP 
connection according to the hosting provider.
[Sudi] What are we doing with CRA (Cyber Resilience Act)?
Long discussion current CRA requirements for stewards and manufacturers.
[Carlos] We don't have additional obligations but could decide to self-report 
issues if we have a security team willing or interested in doing that work.
[Tobias] Greg KH talk, if I understood it correctly, implies that there is 
nothing to do for us (as community) except for providing a way to submit 
security bugs (even if it goes to /dev/null) → 
https://kernel-recipes.org/en/2025/schedule/the-cra-and-what-it-means-for-us/
[David Malcolm] What is the status of the GCC SC review of the AI policy WG 
draft?
[Jason] It is under discussion.
[David Malcolm] I'm in the happy position that Red Hat is paying for my tokens. 
I would want the GNU Toolchain projects to retain a certain level of human 
scale and accessibility to new contributors that they don't need LLMs.
[Carlos] The same question applies to hardware too, but we do want a low-level 
entry to the process.
[David Edelsohn] The leadership for the projects are reaching out to the IHVs 
to support the projects.
[Sudi] If you want to be considered a tier 1 target, please provide your 
machines so we can test this?
[David Edelsohn] We want to avoid pay to play and take it off the table.
[Carlos] For glibc I'm saying that if pre-commit CI is present for your 
hardware, that we will monitor those issues as patches arrive and weekly review 
occurs.
[David Edelsohn] We don't want to have barriers, we also don't want to limit 
what developers can do.
[Claudio] https://sashiko.dev/

Meeting: 2026-07-16 @ 0900h "Asia/Kolkata"

Attendees: Ramana Radhakrishnan, Maxim Kuvyrkov, Avinash Jayakar, Jeevitha, 
Kishan Parmar, Saurov Shyam, Surya Kumari J, Vijay Shankar, David Edelsohn, 
Soumya AR

Agenda:

* Avinash - Matrix extensions and intrinsics and the status of Graphite in GCC

Powerpc and other ISAs have matrix multiply extensions and most usage is via 
intrinsics and no optimization is using these intrinsics. Graphite (and 
polyhedral opts) in GCC can identify the matrix multiply operations. Fortran is 
possibly the closest in terms of frontends that support a builtin matmul 
operation . Discussed that Graphite was in maintenance mode and the work could 
be considered in 2 parts - one is improving loop optimizations driven via 
fortran using polyhedral methods and the second was mapping the matmul to 
actual C / C++ / Fortran codes. Both appear difficult but would require quite 
substantial thinking around what benchmarks / workloads / applications would 
benefit from this and consider it from there.

* Maxim - Status AutoFDO?

AArch64 is functional for GCC 16. Two parts: ingestion of perf.data profiles 
and use of profile information in the various passes. Seen some issues with the 
upstream Google AutoFDO tools and working on a rewrite (RFC now upstream). 
Occasionally see crashes with use of profiles in the compiler though the 
frequency has reduced. Plan for GCC17 is to continue to work on performance 
gaps and functional

--
Cheers,
Carlos.

Reply via email to