https://hgpu.org/?p=17887
Accelerated Sparse Matrix Operations in Nonlinear Least Squares Solvers