The main change is that FSoftSkinVertex, used by FSkeletalMeshLODModel, in now stores weights as 16-bit after conversion from the import data. This increases the size of each FSoftSkinVertex from 144 bytes to 160 bytes (about 10% increase). By default render meshes still use 8-bit skin weights, with weights downshifted from the 16-bit modeling data, so no change in GPU memory consumption there. However, the vertex buffer will automatically return a 16-bit skin weights when requested from the GPU side (e.g. for CPU skinning and viewing tangents).
This change in the model data and vertex buffer CPU-side query, resulted in many changes throughout the codebase and will have an effect on licensees who are actively reading from and writing to these two storage locations.
The GPU skin cache shader has had one more permutation added when not using unlimited skin weights. The vertex factory is not affected.
#jira UE-164386
#rb alexis.matte, josie.yang
#preflight 632c0c5ab4515b7e22b4804d
[CL 22215219 by halfdan ingvarsson in ue5-main branch]
