https://hgpu.org/?p=16914
An N log N Parallel Fast Direct Solver for Kernel Matrices