https://hgpu.org/?p=13378
Global finite element matrix construction based on a CPU-GPU implementation