https://hgpu.org/?p=10252
Improving the GPU space of computation under triangular domain problems