The paper reports results that separate the actual SpGEMM runs from the csr-to-tile conversion. In the code, csr2tile_row_major(matrixA) is timed, but csr2tile_col_major(matrixB) is not timed (I assume csr2tile_col_major() is considered part of the csr-to-tile conversion phase?). Is there any particular reason why csr2tile_col_major() is not timed?