Merge branch 'AcademySoftwareFoundation:master' into master

tsc/meetings/2024-03-12.md

@@ -0,0 +1,76 @@
+Minutes from OpenVDB TSC meeting, March 12th, 2024
+
+Attendees: *Ken* M., *Jeff* L., *Andre* P, *Dan* B., *Greg* H., *Richard* J.
+
+Additional Attendees: Ivo (Autodesk), John Mertic (Linux Foundation), Dhruv Govil (Apple)
+
+Regrets: *Nick* A.
+
+Agenda:
+
+1) Confirm quorum
+2) Secretary
+3) Moving to Apache 2.0 License
+4) PR-1760
+5) Half-Grid Support Presentation by Greg
+6) Next meeting
+
+------------
+
+1) Confirm quorum
+
+Quorum is present.
+
+2) Secretary
+
+Secretary is Andre Pradhana.
+
+3) Moving to Apache 2.0 License
+
+We decided to try to move from Mozilla Public License to Apache 2.
+
+The key part of the process is that you cannot change the entire license until
+every contributor is on board.
+
+The reason we are moving to Apache 2 are: (1) many other ASWF projects are using
+it, (2) it protects the project from patents disputes. The disadvantages of MPL
+are: (1) any contributor can pull out their contributor and (2) the difficulty to
+reliable proof that what is in the system comes from a particular source code.
+
+Ken will share a google doc with a summary on why we are moving to Apache 2.0.
+We will use this draft to communicate to contributors to sign their approval for
+moving to Apache 2.0.
+
+John Mertic will contact DreamWorks to inform them of this decision and to
+follow through with the proces.
+
+It is noted that what we need are contributors after the year 2018, since the
+previous contributors are covered by DreamWorks.
+
+4) PR-1760
+Dan talks about Initial Adaptive Grid Prototype (PR-1760).
+
+Dan tried to use the Point Advection class and go through the interpolation
+through a different type of Grid. To make it work, Dan needs to modify the use of
+Value Accessor, which was designed for Sparse Grid. He changed it to use an
+Accessor from the Tree class itself. Essentially, this is pushing the creation of
+an Accessor down into the Tree.
+
+All the unit tests pass.
+
+Dan asks for people to look at the PR and to provide feedback. He will try to
+fix the Windows build problem. He suggests that we notify AutoDesk of this PR.
+
+5) Half-Grid Support Presentation by Greg
+
+Greg extended the support of HalfGrid to other tools, such as
+CreateLevelSetSphere, FastSweeping, LevelSetFilter, LevelSetMeasure,
+LevelSetMorph, LevelSetSphere, LevelSetTracker, LevelSetUtil, MeshToVolume,
+RayIntersector, RayTracer, SignedFloodFill. He experiments with LevelSetFilter.
+
+We discuss ValueType and ComputeType, which is similar to StoreType and
+ReturnType.
+
+6) Next meeting
+
+Next meeting is on March 19th, 2024. 2pm-3pm EST (GMT-5).

tsc/meetings/2024-03-19.md

@@ -0,0 +1,80 @@
+Minutes from OpenVDB TSC meeting, March 19th, 2024
+
+Attendees: *Jeff* L., *Andre* P, *Dan* B., *Greg* H., *Rich* J.
+
+Additional Attendees: 
+Dhruv Govil, Ivo Kondapaneni
+
+Regrets: *Ken* M., *Nick* A.
+
+Agenda:
+
+1) Confirm quorum
+2) Secretary
+3) OpenVDB Half Grids
+4) Load as Half Patch
+5) Next meeting
+
+------------
+
+1) Confirm quorum
+
+Quorum is present.
+
+2) Secretary
+
+Secretary is Jeff Lait.
+
+3) OpenVDB Half Grids
+
+Support for half without a compute type is much slower and has
+significant errors.  Proposal is to add a Tree::ComputeType that is
+identity with ValueType, but float for Half.
+
+For many tools, like stencils or volumes, should they return Compute
+Type?  Does this break ABI?  It shouldn't as the typedef should decay
+to the base type for mangling?  We could use an enable if to work
+around if it somehow does mess up ABI.
+
+What should be compute types?  
+
+Stencils have a value type that is there value type, different from
+the tree's value type, so they can logically set this to the tree's
+ComputeType.
+
+How to get accessors or setters to use ComputeType?  We could have
+getComputeValue or getValue for the two types.   But widening is free?
+
+Should I be able to make stencils any compute type?  With the tree's
+compute type just the default?  Eventually do we want float grids with
+a compute type of double?  Probably yes, but beyond this PR.
+
+Import and export doesn't conflict with save float as half - they are
+independent and this flag is ignored with half.
+
+What half type do we use? All of our half types are the same
+underlying bit pattern so it doesn't matter.
+
+4) Arnold support for Half
+
+How to load a float grid as a half grid.  Add a scalarConversion type
+for what the new grid's type should be.  
+
+We set that to the stream's metadata, and the grid descriptor uses
+that.  This creates a CreatingReaderFactory to read the source and
+write out as the desired type.  
+
+Raw is.read() with reinterpret are replaced by the converting reader
+that might do some casting internally.
+
+seek should be seekElement() and take the count and position to make
+it clear what it is doing.
+
+Delayed loading will work as the stream has the converter attached to
+it.
+
+Will be made as a PR for further discussion.
+
+5) Next meeting
+
+Next meeting is on March 26th, 2024. 2pm-3pm EDT (GMT-4)