Optimize find_first_of for one element needle

[AlexGuteniev]

Look for /325/1 lines in the result:

Before:

---------------------------------------------------------------
Benchmark                     Time             CPU   Iterations
---------------------------------------------------------------
bm<uint8_t>/2/3            3.58 ns         2.47 ns    640000000
bm<uint8_t>/7/4            13.8 ns         11.8 ns     99555556
bm<uint8_t>/9/3            9.76 ns         7.25 ns    213333333
bm<uint8_t>/22/5           10.1 ns         8.79 ns    186666667
bm<uint8_t>/58/2           11.4 ns         10.2 ns    100000000
bm<uint8_t>/102/4          13.4 ns         11.9 ns    100000000
bm<uint8_t>/325/1          33.0 ns         27.6 ns     49777778
bm<uint8_t>/1011/11        94.6 ns         79.4 ns     19478261
bm<uint8_t>/3056/7          276 ns          222 ns      5973333
bm<uint16_t>/2/3           2.72 ns         2.25 ns    896000000
bm<uint16_t>/7/4           16.5 ns         13.3 ns     89600000
bm<uint16_t>/9/3           10.5 ns         8.80 ns    154482759
bm<uint16_t>/22/5          11.1 ns         10.8 ns    100000000
bm<uint16_t>/58/2          15.1 ns         12.6 ns    128000000
bm<uint16_t>/102/4         22.3 ns         20.1 ns     89600000
bm<uint16_t>/325/1         59.8 ns         52.7 ns     24888889
bm<uint16_t>/1011/11       3534 ns         3214 ns       471579
bm<uint16_t>/3056/7         541 ns          455 ns      2986667

After:

---------------------------------------------------------------
Benchmark                     Time             CPU   Iterations
---------------------------------------------------------------
bm<uint8_t>/2/3            2.88 ns         2.30 ns    597333333
bm<uint8_t>/7/4            13.9 ns         11.7 ns    112000000
bm<uint8_t>/9/3            9.94 ns         7.85 ns    179200000
bm<uint8_t>/22/5           10.4 ns         6.98 ns    179200000
bm<uint8_t>/58/2           11.5 ns         8.32 ns    169056604
bm<uint8_t>/102/4          13.4 ns         11.2 ns    154482759
bm<uint8_t>/325/1          5.72 ns         3.74 ns    308965517
bm<uint8_t>/1011/11        93.7 ns         85.0 ns     16000000
bm<uint8_t>/3056/7          275 ns          243 ns      5973333
bm<uint16_t>/2/3           3.51 ns         3.11 ns    471578947
bm<uint16_t>/7/4           13.9 ns         11.7 ns    100000000
bm<uint16_t>/9/3           10.8 ns         8.91 ns    128000000
bm<uint16_t>/22/5          11.3 ns         10.5 ns    100000000
bm<uint16_t>/58/2          15.3 ns         13.4 ns    112000000
bm<uint16_t>/102/4         22.2 ns         20.0 ns     64000000
bm<uint16_t>/325/1         8.40 ns         7.54 ns    194782609
bm<uint16_t>/1011/11       3369 ns         3024 ns       527059
bm<uint16_t>/3056/7         516 ns          466 ns      2986667

[StephanTLavavej]

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

[StephanTLavavej]

Thanks for another awesome optimization! 😹 🚀 🪄