TSC Meeting Notes 2021-04-22

Owned by Cary Phillips
Nov 04, 2022
2 min read

Attending:

  • Cary Phillips

  • Christina Tempelaar-Lietz

  • Joseph Goldstone

  • Kimball Thurston

  • Larry Gritz

  • Nick Porcino

  • Peter Hillman

  • Stela Espindola

Discussion:

Cmake targets provide wrong include dirs and libraries #136

  • CMake:

    • Kimball: Can we simplify further? I tried to replicate behavior of autoconf.

    • It's already been simplfied, but if anyone can point to any features that can be removed, please do.

    • Should we backport the CMake setup from 2.5 to 2.4? Nick: It should be possible.

    • Required version of cmake? Imath PR #138 suggests setting the minimum at 3.10 instead of 3.12.

    • Kimball: python bindings require 3.12, so technically we could require 3.12 only if compiling the python bindings.

    • Larry: It’s too complicated to sort out all the scenarios, easier to just pick a more modern version and require that.

    • Nick: CMP077 improves handling of non-cached variabes and is necessary to make Vstudio and Xcode liveable. 3.12 is the last bad version. 3.13 is the first version that enables CMP077.

    • Larry: 3.13 is the first with -B -S

    • Nick: submoduling is becoming more common.

    • Larry: we shouldn’t downgrade and then find a problem and change our minds.

    • Larry: it’s a nothing-burger to install cmake, it shouldn't be a problem to require anyone a more modern version.

    • Kimball: How to suggest to people to use the standard FindPackage? Nvidia has a crazy FindOpenEXR for Optix.

  • VFX reference platform:

    • USD is using a new feature of TBB that requires VP 2019.

    • Everybody’s got to compile USD these days.

    • Can we forget about 2017 and 2018? Seems so.

  • Iex:

    • Kimball: can we combine all of OpenEXR into a single library?

    • Cary: I would consider retiring Iex completely, just have it throw std::exceptions.

    • Larry: Those kind of changes require a lot of work by downstream projects. Better to be cautious. Changes like that should be at a 4.0 break.

  • Imath PR #136

    • Nick: The variables that he refers to refer to interface libraries, but we don’t have transitive inclusions.

    • Peter: He’s looking for the compile flags, but looking for them in the wrong place.

  • New WIP C Core:

    • KDT: I have most of the C library side of things ready to go. It’s getting close. Merge to master?

    • Cary: It’s purely additional, doesn't modify existing code, so it should be fine. But best to make an initial release as a beta.

  • C version of half

    • Kimball: I couldn't help myself. A guy came up with much smaller version of the half to float conversion. It produces bitwise identical results to current version, but with a much smaller lookup table, combined with some shift operations.

    • Nick: have you tried it on godbolt?

    • Peter: You can always construct a test that is slower, if your test access the lookup table sequentially to avoid cache thrashing. But on average it may be faster.

    • Go ahead and submit, we can play with it.

  • Kimball: time to start doing the fp16 extensions.

  • Metadata

    • Joseph: The best way to determine which of the camera metadata the in PR #880 is to talk to users of it and find out what they want. Usually not an option because the users aren't available to us, or won’t talk. Proposal is in, a metadata spreadsheet comparing SMPTE standard and the current reference standard. ASC definitely wants to pick this up. We were trying to do everything the ACES container was trying to do, but we should drop that. We can drop the problematic ones, only bringin the good ones.