https://hgpu.org/?p=9241
Reduce, Reuse, Recycle (R^3): a Design Methodology for Sparse Matrix Vector Multiplication on Reconfigurable Platforms