Skip to content

LDWG Meeting minutes: March 26, 2018

Nate Foster edited this page Apr 2, 2018 · 2 revisions

Attendees

  • Antonin Bas (Barefoot)
  • Mihai Budiu (VMware)
  • Gordon Brebner (Xilinx)
  • Calin Cascaval (Barefoot)
  • James Coole (Cisco)
  • Chris Dodd (Barefoot)
  • Andy Fingerhut (Cisco)
  • Nate Foster (Cornell)
  • Yitzhak Gottlieb
  • Vladimir Gurevich (Barefoot)
  • Andy Keep (Cisco)
  • Chris Sommers (Keysight)
  • SK Sodhi (?)
  • Yuki Hisata (One Net Systems)
  • Milad Sharif (Barefoot)
  • Han Wang (Barefoot)

Status Update and Announcements (Nate / Gordon)

  • PSA has 7 target-specific types---e.g., Portid_t
  • But P4 Runtime wants to use a uniform type to manipulate these---e.g., bit<16>
  • Goals:
    • Generate target-independent P4Info from myp4prog.p4
    • Generate target-specific compiled versions from myp4prog.p4 and psa.p4
  • Proposals:
    • Add annotations to typedefs
    • Add newtype declaration
  • Two issues:
    • Want capability to link different modules together
    • Want to expose/manipulate different types in data plane and control plane
  • Action Item: Andy F to convene sub working group
  • Status update:

    • Size added as a constructor parameter
    • PR modifies the p4c IR to reflect this change
  • Question: should we write “tuple” for the elements of the value set? Or “struct” so the control-plane API becomes simpler?

  • Action Item: Han to incorporate improvements from offline discussion into PR; we’ll finalize next time.

  • See also notes from Andy F.
  • Proposal: add size as a standard, but optional attribute of tables
  • Challenge: size is difficult to talk about on hardware targets due to complex encodings of patterns.
  • Proposal: meaning is that compiler must reserve space for at least N entries, but doesn’t necessarily mean that adding N entries will succeed.
  • Action Item: Andy F to create PR with link to notes.
  • API WG has proposed adding top-level metadata to the program through annotations.
  • Desire to standardize the structure of these annocations, which would appear at the start of the document.
  • Question: should we provide special annotation syntax for key-value pairs? Action Item: Chris to refine proposal and discuss with Mihai on syntactic improvements
  • Andy K has produced a document summarizing the design Link to Andy K’s PDF.
  • Questions:
    • How to ensure type safety with enums and non-contiguous values? Action Item: Andy K to lead sub-working group
  • Tradeoffs between adding a new type and a new operation.
  • Action Item: Calin to address lingering issues through offline discussion, and create PR.

Announcements

  • Next meeting: April 16th @ Barefoot
Clone this wiki locally