The
BlockSparseArrays.jl
package provides block-sparse vectors and matrices.
What this package needs next is a generic sparse LU decomposition implemented in Julia. "Generic" here means that it must work with arbitrary types, not just the standard floating-point types. Without such an implementation, it will not be possible to be more efficient than standard Julia sparse arrays when solving linear systems.
This package is similar to BlockSparseMatrices.jl. Different from that packge, this package here encodes the block size in the type, which can create more efficient code.