Parallelized SparseBlockMatrix construction
AlePalu opened this issue · 0 comments
AlePalu commented
With multithreading support, it would be possible to speed-up the construction of a SparseBlockMatrix by assigning the computation of each block to a thread.
Ideally, each thread should fill a std::vector of triplets (or fill a unique shared thread-safe container), each corresponding to a block. The final assembly of the overall matrix is then performed sequentially (setFromTriplets is fast) from one working thread.
Some interesting ideas here: https://gitlab.com/libeigen/eigen/-/issues/2631