Clarification about cusparse layout in spmm

Accelerated Computing GPU-Accelerated Libraries
1
  1. According to cusparseOrder_t, “only column-major layout is supported [for dense matrices]”.
  2. Coo format doc says, “A sparse matrix in COO format is assumed to be stored in row-major format”.

In cusparseSpMM doc, “Row-major layout provides higher performance than column-major.”,
But seems like, sparse coo matrix is always in row major format and dense matrix can’t support row major format, I am not sure what the above statement is pointing to?
It would be great if each matrix layout reference mentions explicitly the matrix (name/type[sparse/dense]) in the algorithms description (cusparseSpMM supports the following algorithms:).