3.0 KiB
Minutes from 107th OpenVDB TSC meeting, October 5th, 2021, (EDT)
Attendees: Ken M., Jeff L., Andre P, Nick A., Dan B.
Additional Attendees: JT Nelson (Blender), Richard Jones (DNeg), Johannes Meng (Intel), Greg Hurst (United Therapeutics)
Regrets:
Agenda:
-
Confirm Quorum
-
Secretary
-
Google Forum
-
TSC Membership
-
NanoVDB
-
VDB Render + OpenEXR
-
Interrupters
-
Next meeting
-
Confirm Quorum
Quorum is present.
- Secretary
Secretary is Dan Bailey.
- Google Forum
Briefly discussed issue 1134. It is a known problem that the same stream cannot be written to multiple times. It is desirable to fix this, but it's not clear whether it is possible due to the nature of needing to write a header at the front of the file.
- TSC Membership
Unanimous vote that Richard Jones should join the TSC. Rich will now be added to the mailing lists and given repo write access.
- NanoVDB
Still working towards bringing NanoVDB into OpenVDB. The FindOpenVDB.cmake is broken, but not deemed necessary to fix it. A CMake config would be better.
Nick points out that NanoVDB currently has three nested levels and that only two would technically be required, Ken to look at addressing this in the PR.
Decision to create a v8.2 release branch and then make the master branch the v9.0 candidate. v9.0 is the priority due to the hard release date.
- VDB Render + OpenEXR
Nick has a PR incoming to make OpenEXR optional for the vdb_render binary. The only remaining migration work for OpenEXR 3 is updating the internal version of Half, however this is not required for a v9.0 release.
Brief discussion about switching on half runtime grids. Greg mentioned he has tried this and it currently fails to compile. Likely won't make it into v9. Greg also asked about whether there was any interest in adding support for minifloat, an 8-bit floating point type. Not known among the TSC so something to potentially be investigated.
- Interrupters
Dan presented PR1193 where the interrupter methods are made into weak symbols and overrided in the Houdini plugin. Motivation here is to reduce overall instantiation cost and to allow the Houdini plugin to benefit from explicit template instantiation. Jeff felt that it was potentially hard to debug if something went wrong, particularly on different platforms and that using polymorphism would be a better approach. Dan to revisit this implementation and attempt this.
General discussion about the benefit of being able to compile out the NullInterrupter in the core library. Dan's testing shows that there is negigible performance to be gained in doing so unless the interrupter is polled too frequently and in this case client code such as Houdini would suffer worse performance. As a result, all agreed that for the benefit of improved compilation times, this feature should be removed. Will need to continue to be careful with how frequently interrupters are polled.
- Next meeting
Next meeting is October 12th, 2021. 12pm-1pm EST (GMT-5). Greg Hurst to present.