shsprandn and shsprand don't guarantee that entries have unique values; ie I can have an entry [3,4] = .4 and [3,5] = -.2. This is fine if I'm only multiplying by the matrix (the entries are just added) but makes no sense for indexing or factorizations. This can be fixed by making shmem_randsample sample without replacement within each column.
shsprandn and shsprand don't guarantee that entries have unique values; ie I can have an entry [3,4] = .4 and [3,5] = -.2. This is fine if I'm only multiplying by the matrix (the entries are just added) but makes no sense for indexing or factorizations. This can be fixed by making shmem_randsample sample without replacement within each column.