On Sun, 31 May 2026 02:46:07 +0800 [email protected] wrote: > +static __rte_always_inline void > +sxe2_tx_desc_fill_one_avx2(volatile union sxe2_tx_data_desc *desc, struct > rte_mbuf *pkt, > + uint64_t desc_cmd, bool with_offloads) > +{
Why do you need to always inline here? After investigating lots of performance cases, my conclusion has been that no new code should use always_inline unless it has special __rte_constant() optimizations and/or can show significant benefit on a benchmark. Just inline should do fine, the compiler cost optimizer will figure out if inline makes sense.
