• Sources: survey, discussion
  • Summary: The survey gives per-crate feature and instruction-set tables and concludes that support for runtime multiversioning is the practical dividing line between the libraries. It names two traps in particular: std::simd silently emitting scalar code when LLVM lacks a matching operation, and wide being incompatible with runtime feature dispatch. Published 2026-09-25.
  • Why it matters: The per-crate tables and the named traps let a team pick a SIMD crate against its actual dispatch requirements rather than by popularity.

send feedback on this story