The mesh refit's pipelines (the triangle step and the B1a refit), compiled
once and shared by any number of GpuMeshRefits on one device. A
caller with several deforming meshes needs one GpuMeshRefit per mesh
(each has its own parameters and topology, written with
queue.writeBuffer, so two meshes can't share one in a frame), and
sharing the kernels means a mesh added later compiles nothing.
The mesh refit's pipelines (the triangle step and the B1a refit), compiled once and shared by any number of GpuMeshRefits on one device. A caller with several deforming meshes needs one
GpuMeshRefitper mesh (each has its own parameters and topology, written withqueue.writeBuffer, so two meshes can't share one in a frame), and sharing the kernels means a mesh added later compiles nothing.