As the SHM and Padded Alltoall modifications are orthogonal, they can be combined in a straightforward way. Each simply copies its data into the shared buffer as before, but instead of packing contiguously as before, instead the data is packed so that every block is spaced equally, according to the size of the largest individual message size.
However, as seen in section 5 this method does not give as good performance as the SHM Alltoallv approach on its own, so it has not been included in the final code.