Files
UnrealEngine/Engine/Source/ThirdParty/OpenVDB/openvdb-12.0.0/tsc/meetings/2024-09-24.md
Brandyn / Techy fcc1b09210 init
2026-04-04 15:40:51 -05:00

3.0 KiB

Minutes from OpenVDB TSC meeting, September 9th, 2024

Attendees: Ken M., Jeff L., Andre P, Dan B., Greg H.

Additional Attendees: John Mertic (Linux Foundation), Dhruv Govil (Apple), Mathew Cong (NVIDIA), Jonathan S, JT Nelson (Blender)

Regrets: Nick A., Richard J.

Agenda:

  1. Confirm quorum
  2. Secretary
  3. Various license changes
  4. OpenVDB 12
  5. Broken build system
  6. fVDB GPU testing
  7. PIP install
  8. Boost
  9. Next Meeting

  1. Confirm quorum

Quorum is present.

  1. Secretary

Secretary is Greg Hurst.

  1. Various license changes

Need 2/3 majority approval for technical charter

Ken, Jeff, Dan, Greg, Andre all approve -- 5/7 TSC approval

SideFX and UT approve CCLA change, awaiting NVIDIA and other relevant players, Weta and ILM

Links for reference and review:

ICLA - https://cla-signature-files-prod.s3.amazonaws.com/contract-group/28a0d3d8-b869-44e0-a4f4-06b50c49822f/template/icla-2024-09-04T09-58-58Z.pdf CCLA - https://cla-signature-files-prod.s3.amazonaws.com/contract-group/28a0d3d8-b869-44e0-a4f4-06b50c49822f/template/ccla-2024-09-04T09-58-59Z.pdf

  1. OpenVDB 12

fVDB not ready for version 12

  • OpenVDB fillet filter PR 1793
  • OpenVDB Half Grid PR 1787
  • OpenVDB tube complex (TBD!)

Access method on the root node. Currently can only access node via coordinate not index -- tease these things out ABI could be a question mark here. Iterate root table + random access

nanoBind support for OpenVDB and NanoVDB (?)

Beyond 12: Volume advection is slow, which uses a slow grid iterator. Relatively straightforward to fix Dilation more expensive than advection -- 6 nn can be sped up.

  1. Broken build system

2 failures left on the weekly CI 1 is trivial and 1 is a little more complicated.

ASWF says we used all of their budget this mont for their CI

fVDB + license updates triggered runs

  1. fVDB GPU testing

fVDB needs ampere or above GPU's to test 32 bit architecture.

  1. Azure can give us a 1/16th of an A10 -- not to build but run tests. Is it worth the cost?

  2. Could possibly run within internally at NVIDIA and push once verified

  3. Could push for other hardware to be donated to ASWF

Can we connect the dots between ASWF and NVIDIA since NVIDIA can see this as a business need / priority

FWIW OpenColorIO has GPU runners

  1. PIP install

Helpful for distributing would be

pip install openvdb and pip install nanovdb

pip install nanovdb is working without openvdb dependency

pip install openvdb is almost working -- Windows is still a headache

fVDB currently uses pybind

Historically we haven't supported package type management ourselves and left it to the users

Just have ability to pip install and clone -- should be minimal support on our end. Still need to build on user's end.

Go all in on nanoBind for OpenVDB and nanoVDB and then pybind for fVDB -- then transition when time is right in the future.

  1. Boost

No longer require boost libraries -- optional requirement except for delayed loading

  1. Next Meeting

Next Meeting 10-08-2024