February 6th, 2023
Host: Carol Payne
Secretary: Carol Payne
Attendees:
- Rémi Achard (TSC) - DNEG
Mark Boorer (TSC) - Industrial Light & Magic
Mei Chu (TSC) - Sony Pictures Imageworks
Sean Cooper (TSC ACES TAC Rep) - ARRI
Michael Dolan (TSC) - Epic Games
Patrick Hodoul (TSC) - Autodesk
John Mertic - Academy Software Foundation / Linux Foundation
Carol Payne (TSC Chair) - Netflix
Mark Titchener (TSC) - Foundry
Carl Rand (TSC) - Weta Digital
Doug Walker (TSC Chief Architect) - Autodesk
Kevin Wheatley (TSC) - Framestore
- Thomas Mansencal - WetaFX
- Zach Lewis - Method Studios
- Eric Renaud-Houde - Epic Games
Apologies:
- Mei Chu
OCIO TSC Meeting Notes
- CI Build Matrix
- Remi: there are some breaking changes in the spreadsheet. Please take a look if you get a chance
- https://docs.google.com/spreadsheets/d/12fS8A3rcAHz5X05NVM11CeeD2N8EG6l1dRbuOqnHL7U/edit#gid=0
- Red: delete
- Green: add
- Blue: in-line changes
- Should work start on these changes before the 2023 docker is available?
- Not planning on changing the nightly, changes to CI only
- Michael: should we leave latest in the nightly though? To catch upcoming breakage quickly?
- Kevin: floating versions should be for notification purposes, not for code quality release support purposes.
- Remi: the nightly builds have been failing for months now, so it's hard to use/depend on
- Remi: optional jobs in the CI would be more visible
- Doug: hopefully we'll be able to get resources to fix at least the analysis nightly builds soon
- Michael: we could split up the analysis workflows and be more intentional about what we want to get out of them.
- We should update the yaml files with new workflows before we attempt to make the changes proposed by Remi. Michael: should be pretty simple, can take a look.
- Managing 3rd party security updates
- Doug: Cedric is working on a PR for that as discussed in prior meetings - adding min/max/recommended
- PR #1762
- 1D/2D Texture LUT - PR to let users disable the 1D LUT simplification as 1D is not supported in Unreal and other apps
- Default would stay the same, user/app would have to explicitly disallow to generate 2D only
- Remi: not sure there is a way to implement this without breaking ABI
- Mark: is there a huge performance hit to just remove the optimization to 1D lookup?
- Eric: probably not, but it would be breaking still as the client would need to be updated - as the height of the lookup would remain "1"
- API should have been based on the enum type and not the texture height
- Michael: can we overload the function to pass an enum on a patch release?
- Does this need fixed in 2.2 or can it wait for 2.3?
- Eric: Epic prefers waiting for an official release, will mitigate on their side until that is available
- Group: decision to do this "right" an add more of an enum to the methods which will break ABI & API so would target 2.3
- GradingPrimaryTransform fixes
- Remi: is anyone currently looking at a fix for this (clamping, saturation behavior differences)? If not, going to take a look
- Doug - on the list but not in progress, reach out for help as needed
- NaN/Inf handling:
- Doug: Cedric is looking at adding NEON support. Difficult to get the same behavior between the different cpu/gpu paths. Even can be different per graphic card. Right now, OCIO prioritizes efficiency and speed vs. 100% accuracy between paths. Wanted to check with the TSC - is the current approach satisfactory? Are the differences an issue?
- Michael: would prioritize parity over performance
- Remi: could we do both? CMAKE or runtime?
- Carol: what is the performance difference? Can we decide on a threshold that is okay?
- Doug: early days, no real benchmarking yet. But we're working on a PR and can discuss more as we go.
Overview
Content Tools