3.3 KiB
Minutes from 129-th OpenVDB TSC meeting, April 12th, 2022, (EDT)
Attendees: Jeff L., Nick A, Dan B., Rich J., Andre P.
Additional Attendees: Sebastian Gaida, Serjio Rojas, Greg Hurst (United Therapeutics), Kyle Wardlow (United Therapeutics).
Regrets: Ken M.
Agenda:
- Confirm quorum
- Secretary
- Forum
- Typelists (PR #1336)
- Binary re-structure (PR #1319)
- Wolfram/Mathematica Integration (PR #1351)
- Next meeting
- Confirm quorum
Quorum is present.
- Secretary
Secretary is Dan Bailey.
- Forum
a) #1350
Questioning the choice of std::deque on Windows. No benchmarking data shared, in many cases using a deque was a conscious choice as the data is not moved when extended so works better when multi-threading. Jeff to reply. Nick removed bug label, default should be to close issues such as this where there's no issue that needs to be addressed.
b) Two dimensional grid Qs
This has been on the roadmap for years. A single slice of a 3D grid can work in many cases, although care must be taken with level sets. Main issue with this approach is a significantly larger memory footprint.
- Typelists (PR #1336)
Nick has extended slightly to improve unique implementation. Now uniqueness is enforced on append instead of at the end of the template expansion which is more efficient. Uses std::conditional which sometimes instantiates false branch and sometimes doesn't. Const Expr If with C++17 should handle this better.
Where to add custom grid support? Makes most sense to put it in openvdb::initialize(), however this may mean that methods that use the dynamic dispatch will not be able to support this custom grid.
Most important is to align the categories - with this change, they will exist in three places as they are also in CMake for explicit instantiation and in Houdini plugins. One inconsistency is scalar vs numeric grid types for example.
- Binary re-structure (PR #1319)
Might have to leave OPENVDB_BUILD_BINARIES for now. Not trivial to remove it as not clear how custom allocators or building binaries against an external libopenvdb will work. Dan to comment on PR for further discussion.
- Wolfram/Mathematica Integration (PR #1351)
Greg to do a half hour presentation next time on some of the integration questions.
Work still to do be done on improving the docs. Need to be some minor changes to the whitespace CI rules to exclude certain subdirs, particularly around autogen code.
This PR adds a lot of code. Question about whether or not it might be feasible to exclude the 400k+ lines of autogen code and to do this on the fly?
Main issue to investigate is the CI and licensing. Wolfram Engine is available for this kind of purpose, but need to follow up with TAC / John Mertic to find out more. Three options to pursue:
- Full CI integration to build and evaluate unit tests using Wolfram/Mathematica licenses.
- Houdini-style build-only integration that ensures compilation is successful but does no evaluation.
- Minimal build only components that do not include Wolfram headers/libs.
After discussions with ASWF, should be clearer which one is preferred.
Question about openvdb_wolfram instead of openvdb_mathematica? Although Mathematica is the main product, there is also Wolfram Engine. Greg's call on which makes most sense.
- Next meeting
Next meeting is on April 19th, 2022. 1pm-2pm EDT (GMT-4).