Sparse computations, like multiplying sparse matrices with dense vectors, are difficult to optimize due to their irregularity and memory constraints. To enhance performance, solutions include hardware enhancements and specialized software formats and programs. These computations are typically done on unchanging sparse structures, with the data being variable but the sparsity structure remaining constant. Formats like CSR can be costly to modify. The main aim is to repeatedly compute on the same sparse structure without altering its sparsity.