Improving the Performance of Multithreaded Sparse Matrix-Vector Multiplication Using Index and Value Compression