vectorize find_first_of for long needle 🪡

[Alex Guteniev (AlexGuteniev)]

In benchmark results only the last line changed:

before

-----------------------------------------------------------------
Benchmark                       Time             CPU   Iterations
-----------------------------------------------------------------
bm            2.70 ns         2.35 ns    597333333
bm           2.71 ns         2.13 ns    689230769
bm            14.1 ns         10.9 ns    112000000
bm           12.4 ns         10.6 ns    100000000
bm            9.77 ns         7.70 ns    190638298
bm           10.4 ns         8.06 ns    162909091
bm           10.1 ns         8.02 ns    208372093
bm          11.0 ns         8.64 ns    151864407
bm          273 ns          233 ns      5973333
bm         561 ns          425 ns      3089655
bm        95.5 ns         79.2 ns     17568627
bm       2872 ns         2448 ns       497778

after

-----------------------------------------------------------------
Benchmark                       Time             CPU   Iterations
-----------------------------------------------------------------
bm            2.71 ns         1.80 ns    746666667
bm           2.71 ns         1.55 ns   1000000000
bm            14.1 ns         7.47 ns    175686275
bm           12.3 ns         5.82 ns    190638298
bm            9.78 ns         6.75 ns    208372093
bm           10.7 ns         7.54 ns    190638298
bm           10.3 ns         7.21 ns    169056604
bm          11.1 ns         8.63 ns    175686275
bm          276 ns          205 ns     11200000
bm         552 ns          432 ns      2890323
bm        94.4 ns         76.2 ns     19478261
bm        481 ns          365 ns      3895652

[Alex Guteniev (AlexGuteniev)]

There are real test failures here.

There were multiple errors.

After fixing all of them, I thought it would be better to preserve the <= 16 version, as the generic becomes noticeable slower.

[Alex Guteniev (AlexGuteniev)]

Review with "hide whitespace difference" to see the <= 16 branch preserved.

[Stephan T. Lavavej (StephanTLavavej)]

I'm mirroring this to the MSVC-internal repo - please notify me if any further changes are pushed.

[Stephan T. Lavavej (StephanTLavavej)]

Thanks for improving your vectorized implementation even further! 🚀 🚀 🚀