The issue is the generated CUDA code InformationFilterUpdate.cu calls a device function mwGetGlobalThreadIndex in another CUDA file, MWCudaDimUtility.cu. In this case, you need to pass "-rdc=true" to nvcc, and link the device code using nvcc.
I don't think mexcuda supports either of them. You can not pass flags to nvcc and you can not use nvcc to link the object files with mexcuda.
The solution I suggest is to modify InformationFilterUpdate.cu to copy the device functions it calls to the file. We have fixed this issue in the coming 20b release by moving the device functions to the header. For now, you can try replacing MWCudaDimUtility.hpp with the file I attached here, which is copied from 20b release.