BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:DRM Fabric: do we want a common topology API\, or three more sysfs
  layouts?
DTSTART;VALUE=DATE-TIME:20260929T211500Z
DTEND;VALUE=DATE-TIME:20260929T212000Z
DTSTAMP;VALUE=DATE-TIME:20260924T200810Z
UID:indico-contribution-630@indico.freedesktop.org
DESCRIPTION:Speakers: Rodrigo Vivi ()\nScale-up accelerator fabrics are al
 ready here: amdgpu has exposed xGMI topology through its own sysfs for yea
 rs\, AMD recently posted patches of UALink pod infrastructure\, and Intel'
 s XeLink/IAF support was posted and never became shared infrastructure. Ev
 ery one of these describes the same thing: which accelerators are directly
  connected\, through which ports\, in what state. And everyone describes i
 t differently.\n \nThe drm/fabric RFC proposes one vendor-neutral model fo
 r that: fabric → endpoint → port → peer\, over Generic Netlink\, wit
 h the core recording direct adjacency only and leaving routing\, transport
  and hardware programming to the drivers. It ships with a synthetic provid
 er and no real hardware backend on purpose\, so the model can be argued ab
 out before it's tied to anyone's silicon.\n \nI'll spend five minutes on t
 he model and then ask the room the question the RFC actually needs answere
 d: is this the right minimum common representation\, and will vendors impl
 ement it\, or do we accept per-driver topology forever?\n\nhttps://indico.
 freedesktop.org/event/12/contributions/630/
LOCATION:
URL:https://indico.freedesktop.org/event/12/contributions/630/
END:VEVENT
END:VCALENDAR