6.9 KiB
Minutes from 12th OpenVDB TSC meeting, Mar. 28, 2019
Attendees: Ken M., Nick A., Dan B., Peter C.
Additional Attendees: Andrew Pearce (DWA), Thanh Ha (LF)
Apologies: Jeff L.
Agenda:
-
Confirmation of quorum
-
Selection of secretary
-
TAC Update
-
Multi-Res Grid
-
Out-of-Core Read-only Grid
-
Instantiation / Compile-times
-
Tools vs Grid Methods
-
Release Schedule
-
CMake/CI Update
-
Improve / Update Cookbooks
-
CLA Check
-
Update Website
-
Documentation on how to file bug-reports or ask questions
-
Grid Iterator Performance
-
Next Meeting
-
A quorum was confirmed. Ken was phoning in, so Dan chaired.
-
Secretary - Dan Bailey
-
TAC Update
No-one available to represent OpenVDB in TAC Meeting the day before. Dan sent Daniel Heckenberg a brief summary of OpenVDB updates which were shared. Thanh gave a brief update from the TAC Meeting. The ASWF board are in support of using Circle CI but remain interested in being able to provide reproducible builds outside of Circle. Thanh also announced he was leaving the Linux Foundation for a startup. Andrew Grimberg is taking over his responsibilities and is already up-to-speed. It is envisioned that Andrew will attend the TAC and CI working group meetings, but likely not the project TSC meetings.
- Multi-Res Grid
Ken has received a comprehensive response from Autodesk answering the TSC questions regarding Autodesk's proposal for introducing the Multi-Res Grid to OpenVDB. He will be sending it around to the group shortly for discussion in the next meeting. First impressions look promising.
- Out-of-Core Read-only Grid
Ken has a working implementation of an out-of-core VDB grid. It is read-only, supports random access with a pre-defined fixed memory footprint and can be built from a standard out-of-core VDB grid. A fully out-of-core VDB grid requires half the memory footprint of a standard VDB grid in out-of-core mode. Savings are also possible through not supporting very old file formats. Ken plans to release at Weta and then to look to introduce to the OpenVDB codebase. Some concerns from Peter and Dan about introducing a new Grid structure, but keen on understanding more of the implementation details.
- Instantiation / Compile-times
Compilation of applications and plugins built against OpenVDB is slow and uses a lot of memory, which we wish to address. Peter wants to consider baking the configuration of the grid so as to improve compilation times and suggests asking users whether they ever change the grid configuration. Ken mentioned that he knows people change the size of the leaf node for GPU and other uses, but unaware of many people changing the configuration of the tree. He wants to find out if this is a consideration for Autodesk's Multi-Res Grid. Potential investigation around where the compilation time is going could help direct our efforts. Explicit template specialization also worth looking into.
- Tools vs Grid Methods
Over time, there has been a gradual migration towards using free functions in the tools directory instead of adding to the Grid and Tree classes. Question of whether it's a good time to introduce a policy around this and to look at removing some of the Grid and Tree methods. Ken proposed that the first step is to look at all the methods and confirm whether they can currently be removed or whether they need new access functions to be able to remove them. This change is also expected to improve compilation times. Removing virtual functions needs to be done in a major release, so they would need to be marked deprecated in a prior release.
- Release Schedule
Peter proposes we do a new release in the next few weeks. It was decided that the next release should be 6.1.0, as there is a lot of new functionality and some of the changes appear that they may affect the API. Unanimous vote in favour. Potentially hold off to ensure we can include CMake changes. Peter proposes adding static casts to fix the precision warnings as a temporary solution so that we can make this release compile with strict mode enabled without behavior changes. Ken proposes another minor/patch release shortly before Siggraph. Rough plan is to do a new 7.0 release in September / October which changes ABI, so we should start gathering ideas now. This new major release can also be when we mark methods as deprecated that we intend to remove in early 2020.
- CMake/CI Update
Nick gave an update on his progress with refactoring CMake, his stage branch contains further improvements. Main issues are around IlmBase/OpenEXR version suffixes and case-sensitivity of CMake modules. Peter is currently testing with DWA infrastructure and has run into a few issues, particularly around GLFW. Dan has added CircleCI support which now builds all third-party dependencies from scratch. More testing to come. Dan proposes changing the syntax, Nick suggested after this PR gets merged is a good time. Peter asked about improved ways to help users build against OpenVDB instead of using openvdb_print, Dan suggested a "Hello World" sample might be a nice way of getting people started. Nick says there is still some work needed on introducing a CMake module and improving the find package mechanism for building against OpenVDB.
- Improve / Update Cookbooks
Dan suggested building the cookbooks as part of the CI. Peter and Ken proposed instead turning the cookbook examples into samples modeled after the SideFX HDK samples with links from the documentation where relevant.
- CLA Check
Thanh says to follow up with John regarding the progress of this.
- Update Website
Website originally designed by Mihai Alden, needs some updating. All present should seek permission for images from their respective studios to add to the website with the aim for an update to be completed by Siggraph.
- Documentation on how to file bug-reports or ask questions
General desire to keep Google forum around to avoid having to ask 600 or so users to re-subscribe. Need a proposal on what to do with other communication mechanisms such as GitHub issues and mailing lists. Dan to do a first draft of changes to contributing.md.
- Grid Iterator Performance
Dan raised performance issues with the grid iterator. When used in tools::extrema(), it represented as much as 30x performance differential. Ken mentioned that it's a significant problem and often his first question to a user asking about performance is around which iterator they are using. The main issue is poor ability to split when threading. Peter highlighted that this iterator was designed for convenience, not performance and that performance varies based on the use. Ken believes people are using it by accident without realizing the performance consequences and better documentation wouldn't necessarily address this. Two potential solutions were discussed - removing the grid iterator entirely and re-writing it to offer a faster, underlying implementation.
- Next Meeting
Our next TSC meeting is scheduled for Thursday April 11, 2019 11am-12pm PST.