Minutes 2023 05 24

GPU Web 2023-05-24

Note that unless stated otherwise this is a GPU for the Web community group meeting and not a working group meeting.

Chair: KG

Scribe:

Location: Google Meet

Tentative agenda

• Administrivia
  • Fingerprinting Surface #3101
    • This conversation seems stalled. To help move forward we should organize to attend a PING meeting with interested WebGPU folks (meetings are 9 AM PT Thursdays). Collect a list of interested folks, find a date to discuss with the PING.
• CTS Update
• “Tacit Resolution” Queue
• Handling of omitted labels is unclear #4083
• More promise ordering guarantees #4117
• Timestamp query resets #4069
• Copying depth32float to depth32float #4125
• Agenda for next meeting

Attendance

• Apple
  • Mike Wyrzykowski
  • Myles C. Maxfield
• Google
  • Brandon Jones
  • Kai Ninomiya
• Microsoft
  • Rafael Cintron
• Mozilla
  • Jim Blandy
  • Kelsey Gilbert
• Nvidia
  • Anders Leino

Administrivia

• Fingerprinting Surface #3101
  • This conversation seems stalled. To help move forward we should organize to attend a PING meeting with interested WebGPU folks (meetings are 9 AM PT Thursdays). Collect a list of interested folks, find a date to discuss with the PING.

CTS Update

• Google team working on:
  • AbstractFloat behavior/precision testing
  • Texture filtering testing
  • Fixed an issue with WPT build
  • etc.
• KG: Mozilla getting test expectations stood up
  • KG:
  • KN: Happy to talk about chunking less finely
  • KG: Think it’s
  • KG: Find CTS and TypeScript hard to develop with
  • MM: Agree about TypeScript
  • MM: Finding opposite problem - chunked too coarsely
  • KG: CI import sees every single subtest
  • MM: Sounds like we’re picking opposite extremes
  • …
  • MM: Want bundles of tests that take a few seconds to run. Likely to want to run an offline script with infinite timeouts that measures how long each subtest takes. Then an offline tool can build bundles of tests which will add to a particular time target
  • KN: Happy to help get default bundles into the CTS

“Tacit Resolution” Queue

• None (editors have been meeting less frequently)

Handling of omitted labels is unclear #4083

• JB: Defaulting to empty string seems like a perfectly good solution
• (Agreement)
• Resolved

More promise ordering guarantees #4117

• JB: (confirmation of the motivation for these guarantees)
• (Agreement)
• Resolved: accept

Timestamp query resets #4069

• MM: Possible for rollover to happen but to still not see timestamp move backward.
• all: Think we can’t/shouldn’t do try to paper over anything here.
• BJ: Text currently in the spec …
• KN: … needs work.
• MM: To be developer friendly basically warn developers about this happening and how it can happen
• KG: We want to warn devs not to panic or e.g. throw when timestamps appear to go backwards.
• Resolved: do that

Copying depth32float to depth32float #4125

• <general agreement that copying depth32float to depth32float should work>
• <general un-sure-ness about copying depth32float to anything else>
• Resolved: Allow depth32float-depth32float. Don’t bother to investigate depth<->depth-stencil, and keep disallowing it for now (until someone asks for it later on)

Agenda for next meeting

• Won’t meet next week due to low agenda
• Please add any items (not just post-v1) to the agenda below!
  • Otherwise if they are not in the milestone we’re currently working on, we might not get to them.
• Agenda
  • (socialization) Space efficient wide color support #4108
  • (socialization) Cannot upload/download to UMA storage buffer without an unnecessary copy and unnecessary memory use #2388
    • Recap the design constraints for this problem