ARM Strlen from Godbolt

[jamierpond]

-------------------------------------------------------------------------------------------------------
Benchmark                                                             Time             CPU   Iterations
-------------------------------------------------------------------------------------------------------
runBenchmark<Corpus8DecimalDigits, InvokeLemire>                   1963 ns         1917 ns       365642
runBenchmark<Corpus8DecimalDigits, InvokeZoo>                      1953 ns         1916 ns       365751
runBenchmark<Corpus8DecimalDigits, InvokeLIBC>                     9521 ns         9344 ns        74932
runBenchmark<CorpusStringLength, InvokeLIBC_STRLEN>                4283 ns         4203 ns       164312
runBenchmark<CorpusStringLength, InvokeZOO_STRLEN>                 4472 ns         4388 ns       159545
runBenchmark<CorpusStringLength, InvokeZOO_NATURAL_STRLEN>         5009 ns         4911 ns       135583
runBenchmark<CorpusStringLength, InvokeGENERIC_GLIBC_STRLEN>       6802 ns         6681 ns       101432
runBenchmark<CorpusStringLength, InvokeZOO_NEON>                   3540 ns         3473 ns       206814
runBenchmark<CorpusStringLength, RepeatZooStrlen>                  4479 ns         4394 ns       158624

[thecppzoo]

Gosh, another top-benchmarking!!

Jamie: are you up for replicating GLIBC by unrolling manually in the C++ code 1 iteration? so that we have an idea of what happens with the unrolling. I'm not understanding why there is such a difference (not complaining since the code we are working on is consistently coming on top, regardless of how we "slice and dice it")