From a05dc7a903f42358cb6c7a9685a81465c7e14953 Mon Sep 17 00:00:00 2001 From: Dan Bailey Date: Fri, 28 Mar 2025 22:50:55 -0700 Subject: [PATCH] Meeting Notes Signed-off-by: Dan Bailey --- tsc/meetings/2025-03-26.md | 69 ++++++++++++++++++++++++++++++++++++++ 1 file changed, 69 insertions(+) create mode 100644 tsc/meetings/2025-03-26.md diff --git a/tsc/meetings/2025-03-26.md b/tsc/meetings/2025-03-26.md new file mode 100644 index 0000000000..f467c50300 --- /dev/null +++ b/tsc/meetings/2025-03-26.md @@ -0,0 +1,69 @@ +Minutes from OpenVDB TSC meeting, March 26th, 2025 + +Attendees: *Dan* B., *Andre* P, *Jeff* L., *Nick* A., *Greg* H. + +Additional Attendees: Dhruv Govil (Apple), Mathew Cong (NVIDIA), + +Regrets: *Ken* M. + +Agenda: + +1) Confirm quorum +2) Secretary +3) PR1935 +4) PR2016 +5) NanoVDB PointsToGrid Hashing +6) New VDB Release +7) PR2008 +8) PR2011 +9) Next Meeting + +------------ + +1) Confirm quorum + +Quorum is present. + +2) Secretary + +Secretary is Dan Bailey. + +3) PR1935 + +Levelset Tapered Tubes. Dan has approved, does not need another approval as Greg is TSC member, +Greg can merge. + +4) PR2016 + +Jeff fixed a NanoVDB bug with a missing assignment during transpose that prevented round-tripping. +Remains to be seen how many downstream consumers will need to adapt as a result. Nick suggested +adding nodiscard keywords to the transpose methods to generate compiler warnings for these cases. + +5) NanoVDB PointsToGrid Hashing + +Jeff has implemented a faster version of points to grid hashing that does not sort. As a result +this method is non-deterministic. For cases where this doesn't matter can be much faster. Andre +confirmed that NVidia has a similar Cuda implementation but has not been made open-sourced. + +6) New VDB Release + +Nick to do the release now CI is looking healthier. + +7) PR2008 + +This has huge impact on Windows, but ultimately should apply to Linux as well. Needs another +approval. + +8) PR2011 + +Andre working through supporting half in math functions for Vec3 types. Currently failing +Windows CI. + +Autodesk still interested in instantiating half grids when reading float grids from disk. Main +motivation is to reduce peak memory usage. Brief discussion of a more substantial change to be +able to convert any convertible grids on load. Simpler solution of only handling float->half +conversion is less likely to stall. + +9) Next Meeting + +April 9th, 2025, usual time