I Tomasz, > > > > Just thought about it once again: > > As now inside rte_eth_tx_prep() we do now: > > + > > + if (!dev->tx_pkt_prep) > > + return nb_pkts; > > > > Then there might be a better approach to set > > dev->tx_pkt_prep = NULL > > for simple and vector TX functions? > > > > After all, prep_simple() does nothing but returns an error if > > conditions are not met. > > And if simple TX was already selected, then that means that user > > deliberately disabled all HW TX offloads in favor of faster TX and > > there is no point to slow him down with extra checks here. > > Same for i40e and fm10k. > > What is your opinion? > > > > Konstantin > > > > Yes, if performance is a key, and, while the limitations of vector/simple > path are quite well documented, these additional checks are a > bit overzealous. We may assume that to made tx offloads working, we need to > configure driver in a right way, and this is a > configuration issue if something doesn't work. > > I will remove it.
Great, thanks. Konstantin
