Parallel matrix creation

Question

Are there algorithms that allow you to efficiently create (padding) sparse (for example, CSR or coordinate) matrix?

Answer 1

There are no efficient algorithms for creating sparse matrices in parallel access. Credible is the type of coordinate matrix, which requires sorting after filling the contents, but this type is slow for matrix products, etc.

Solution: you do not build a sparse matrix - you do not store it in memory; you perform implicit operations when you calculate the elements of a sparse matrix.

Answer 2

, , , .

Java ConcurrentHashMap, .NET 4 ConcurrentDictionary, (afaik) .