most other implementations do not use splicevm, which is a huge win for this specific problem.

Of course all the AVX and cache optimizations are also exceedingly clever.