- Skip full-mesh normals computation in OffsetMeshRegion if using vertex normals.
- Add operator[] to MeshConnectedComponents.
- Add FSparseDynamicOctree3::ContainmentQuery() to allow for single-point range query.
- Add FMeshFaceSelection::IsSelectionBoundaryEdge() and optimize ExpandToOneRingNeighbours() and ContractBorderByOneRingNeighbours(), now significantly faster for very large selections. Convert VtxTrianglesItr calls to EnumerateVtxTriangles
- Use TestAnyHitTriangle() instead of FindNearestHitTriangle() in TRemoveOccludedTriangles. Reduces cost from 6.4s to 1.2s on a 64-core threadripper for a large-scale (1.4m triangles) visibility-based removal problem
- Optimize FMarchingCubes implementation for large thread counts. Replace single FCriticalSection with indexed sets of FCriticalSection to reduce thread contention. Instead of directly writing to mesh, accumulate mesh vertices and triangles in separate blocks (hashed on grid coordinates) and then build output mesh after generation completes. Add bEnableValueCaching option to skip caching of corner values, which causes lots of thread contention if Implicit function is very cheap (eg a grid-based SDF evaluation). Enable this option in Morphology meshing. Reduces marching cubes cost from ~3.5s to <500ms on a 64-core threadripper for a large-scale (1.4m triangles) Morphology problem.
#rb jimmy.andrews
#rnx
#jira none
#ROBOMERGE-SOURCE: CL 16370702 in //UE5/Private-Frosty/...
#ROBOMERGE-BOT: STARSHIP (Private-Frosty -> Main) (v804-16311228)
[CL 16372561 by ryan schmidt in ue5-main branch]
