>Patch looks ok to me, and AI review finds no issues. I've flagged a couple of >minor things below. > >Main concern is performance. What's the performance difference - if any - >measured on this Rx path with this change compared to without?
Thanks for the review. On E810, 64B single-core rxonly (--record-core-cycles, Rx core saturated): with RTE_ETH_RX_OFFLOAD_TIMESTAMP enabled, the scalar fallback was ~159 cyc/pkt; this patch keeps it on the vector path at ~78 (AVX512) / ~75 (AVX2), roughly a 2x reduction. The non-timestamp path is unchanged. >Patch needs a rebase, there is already a section in release notes for ice >driver >updates, and the ice.ini file has conflicting updates too (though both these >are easy fixes) Done in v2 - rebased, release note folded into the existing ice section, ice.ini conflict resolved. >Nit: I'd try to keep this on two lines rather than 3. Fixed in v2. Thanks, Sandeep
