paireks
commented
6 months ago Is it something we want to add to v2 some day or maybe design v3 in a way that already such feature will be possible? Cause it looks like a pretty big change to me. Or maybe I overexaggerate :)
xeolabs
Problem
Currently we are unable to move individual objects long distances within xeokit's double-precision 3D world coordinate system. This is because they need to stay more or less within the bounds of their RTC (relative-to-center) tiles, where their geometries are able to have the requisite single-precision coordinates as offsets from the RTC tile center. This is neccessary to support xeokit's ability to render double-precision coordinates on today's GPUs, which are single-precision, without rendering jitter.
Idea
For each mesh, store an integer ID in the VBO or data texture that identifies which RTC tile it is currently in. When we update the transform for that mesh such that it moves the mesh into another RTC tile, then update that ID.
The VBO or data texture layer would then be rendered once for each distinct ID value that occurs within it, each time updating the RTC view matrix according. At each render, only the meshes whose IDs match the RTC for the current view matrix. This is effectively culling those meshes that don't belong to the current RTC render.
The downside is that layers would get rendered multiple times. However, if there is not a huge number of RTC tiles occupied at any moment, then this could be viable. The VBO/data texture layers would need to keep counts of how many meshes are in each RTC tile, so we can avoid needlessly rendering layers.
The modeling transform matrices for the meshes would also need to be dynamically modified for their meshes' RTC tile origins, by post-concatenating a subtraction-translation, so that they are effectively then relative to the RTC tile origin.