People have shown that direction optimizing BFS fits very neatly inside graphblas actually [1]. Perhaps in a few years people will have figured out how to make delta stepping and storing predecessors fit too even if it's not clear at the moment.
For delta stepping, it seems all you would need is a priority queue that works in batches as opposed to individual elements. Then to make it performant and match delta stepping code written from scratch, you might need something that can fuse multiple graphblas operations together so that you don't have too many extra memory ops from the priority queue operations.
For delta stepping, it seems all you would need is a priority queue that works in batches as opposed to individual elements. Then to make it performant and match delta stepping code written from scratch, you might need something that can fuse multiple graphblas operations together so that you don't have too many extra memory ops from the priority queue operations.
[1] https://dl.acm.org/doi/pdf/10.1145/3225058.3225122