MatGetSubMatrix and some of the matrix-matrix kernels might benefit from this operation. I don't know if it's a bottleneck, but if it is, this shows how to make it fast.
https://highlyscalable.wordpress.com/2012/06/05/fast-intersection-sorted-lists-sse/ Not a deep insight or surprising, but still a nice write-up. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20120607/f14fbb2a/attachment.html>
