> Microbenchmark performance on NovaLake/DiamondRapids shows no benefit from 
> SETcc.ZU encoding on these cores.

Please try *macro* benchmarks.

(Across a collection of a few dozen of them
I see an average perf drop of ~1%, with up
to ~4% in the worst cases. No a drop you'd
want for at scale deployments.)

--
C.

Reply via email to