Optimizing Sparse Matrix-Vector Multiplication on Emerging Many-Core Architectures