SIMD slower than NoParallelization

[Octogonapus]

Describe the bug

The SIMD implementation is slower than NoParallelization across the board. I don't want to jump to conclusions and blame LoopVectorization here; I probably have written my loop wrong.

Steps To Reproduce

This happens on all benchmarks (in the README).

Expected behavior

SIMD should be faster than NoParallelization. My "gold standard" (i.e. the fastest histogram implementation I can find so far) is AHTL. I would like to match AHTL's performance as much as possible.
As of writing, FastHistograms (without SIMD) needs 2.57596e8 ns to do what AHTL can do in 1.08249e8 ns.

Additional Information

Relevant code in AHTL: https://github.com/pcjung/AHTL/blob/master/src/fixed.cpp

Also, debatably I should remove SIMD from this package (for now) because it's slower.