Vectorize lexicographical_compare!

[AlexGuteniev]

Resolves #3998

Using existing mismatch vectorization and existing classifiers.

Benchmark

Before:

bm<uint8_t, op::lexi>/8/3                  2.29 ns         1.08 ns   1000000000
bm<uint8_t, op::lexi>/24/22                2.77 ns         1.48 ns   1000000000
bm<uint8_t, op::lexi>/105/-1               4.29 ns         2.16 ns   1000000000
bm<uint8_t, op::lexi>/4021/3056            73.6 ns         35.7 ns     50779001
bm<int8_t, op::lexi>/8/3                   2.35 ns         1.20 ns   1000000000
bm<int8_t, op::lexi>/24/22                 12.2 ns         5.40 ns    248888889
bm<int8_t, op::lexi>/105/-1                61.6 ns         26.7 ns     52705882
bm<int8_t, op::lexi>/4021/3056             1485 ns          755 ns      2357895
bm<uint16_t, op::lexi>/8/3                 2.27 ns         1.16 ns   1000000000
bm<uint16_t, op::lexi>/24/22               11.9 ns         6.05 ns    263529412
bm<uint16_t, op::lexi>/105/-1              60.7 ns         30.7 ns     44800000
bm<uint16_t, op::lexi>/4021/3056           1486 ns          680 ns      1723077
bm<uint32_t, op::lexi>/8/3                 2.18 ns         1.03 ns   1000000000
bm<uint32_t, op::lexi>/24/22               11.9 ns         5.85 ns    229743590
bm<uint32_t, op::lexi>/105/-1              59.1 ns         35.6 ns     37333333
bm<uint32_t, op::lexi>/4021/3056           1488 ns          643 ns      1629091
bm<uint64_t, op::lexi>/8/3                 2.17 ns         1.05 ns   1000000000
bm<uint64_t, op::lexi>/24/22               12.0 ns         4.90 ns    242162162
bm<uint64_t, op::lexi>/105/-1              59.1 ns         27.5 ns     47157895
bm<uint64_t, op::lexi>/4021/3056           1474 ns          750 ns      1792000

After:

bm<uint8_t, op::lexi>/8/3                  2.69 ns         1.98 ns    995555556
bm<uint8_t, op::lexi>/24/22                2.65 ns         2.09 ns    560000000
bm<uint8_t, op::lexi>/105/-1               4.30 ns         2.59 ns    640000000
bm<uint8_t, op::lexi>/4021/3056            59.4 ns         38.8 ns     42666667
bm<int8_t, op::lexi>/8/3                   2.67 ns         1.61 ns    814545455
bm<int8_t, op::lexi>/24/22                 2.66 ns         1.52 ns    689230769
bm<int8_t, op::lexi>/105/-1                4.25 ns         2.78 ns    426666667
bm<int8_t, op::lexi>/4021/3056             58.9 ns         39.1 ns     27151515
bm<uint16_t, op::lexi>/8/3                 3.09 ns         1.59 ns    689230769
bm<uint16_t, op::lexi>/24/22               3.76 ns         2.73 ns    527058824
bm<uint16_t, op::lexi>/105/-1              6.14 ns         3.57 ns    280000000
bm<uint16_t, op::lexi>/4021/3056           99.3 ns         67.8 ns     18666667
bm<uint32_t, op::lexi>/8/3                 2.73 ns         1.99 ns    746666667
bm<uint32_t, op::lexi>/24/22               3.76 ns         2.17 ns    597333333
bm<uint32_t, op::lexi>/105/-1              8.80 ns         6.49 ns    373333333
bm<uint32_t, op::lexi>/4021/3056            193 ns          143 ns     12800000
bm<uint64_t, op::lexi>/8/3                 2.68 ns         1.66 ns    896000000
bm<uint64_t, op::lexi>/24/22               5.10 ns         3.85 ns    426666667
bm<uint64_t, op::lexi>/105/-1              15.1 ns         9.51 ns    154482759
bm<uint64_t, op::lexi>/4021/3056            379 ns          291 ns      5270588

[StephanTLavavej]

Thanks - this is so much easier to follow! 😻 🧠

[StephanTLavavej]

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

[StephanTLavavej]

Thanks for vectorizing one of the STL's most important algorithms! 😻 🚀 🥇