Performance improvement in cudf::strings::all_characters_of_type

[davidwendt]

Description

Improves performance for cudf::strings::all_characters_of_type() API which covers many cudf is_X functions. The solution improves performance for all string lengths as measured by the new benchmark included in this PR.
Additionally, the code was cleaned up to help with maintenance and clarity.

Reference #13048

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[davidwendt]

Here is the diff from the nvbench_compare tool:

|  row_width  |  num_rows  |   Ref Time |   Cmp Time |          Diff |   %Diff |
|-------------|------------|------------|------------|---------------|---------|
|     32      |    4096    |  34.701 us |  32.349 us |     -2.352 us |  -6.78% |
|     64      |    4096    |  34.414 us |  32.179 us |     -2.235 us |  -6.50% |
|     128     |    4096    |  41.044 us |  32.793 us |     -8.251 us | -20.10% |
|     256     |    4096    |  60.614 us |  32.353 us |    -28.261 us | -46.63% |
|     512     |    4096    | 100.592 us |  33.944 us |    -66.648 us | -66.26% |
|    1024     |    4096    | 178.231 us |  34.122 us |   -144.109 us | -80.86% |
|    2048     |    4096    | 324.128 us |  33.723 us |   -290.405 us | -89.60% |
|    4096     |    4096    | 616.879 us |  32.786 us |   -584.093 us | -94.69% |
|     32      |   32768    |  34.873 us |  32.656 us |     -2.216 us |  -6.35% |
|     64      |   32768    |  36.433 us |  34.036 us |     -2.397 us |  -6.58% |
|     128     |   32768    |  45.096 us |  36.951 us |     -8.145 us | -18.06% |
|     256     |   32768    |  65.402 us |  35.736 us |    -29.666 us | -45.36% |
|     512     |   32768    | 113.118 us |  36.259 us |    -76.859 us | -67.95% |
|    1024     |   32768    | 199.634 us |  37.704 us |   -161.931 us | -81.11% |
|    2048     |   32768    | 365.835 us |  37.985 us |   -327.850 us | -89.62% |
|    4096     |   32768    | 688.672 us |  38.201 us |   -650.472 us | -94.45% |
|     32      |   262144   |  47.768 us |  42.404 us |     -5.365 us | -11.23% |
|     64      |   262144   |  61.138 us |  52.997 us |     -8.141 us | -13.32% |
|     128     |   262144   | 102.089 us |  63.134 us |    -38.955 us | -38.16% |
|     256     |   262144   | 305.667 us |  67.241 us |   -238.426 us | -78.00% |
|     512     |   262144   | 651.223 us |  67.515 us |   -583.708 us | -89.63% |
|    1024     |   262144   |   1.648 ms |  69.991 us |  -1578.019 us | -95.75% |
|    2048     |   262144   |   5.159 ms |  80.804 us |  -5077.801 us | -98.43% |
|    4096     |   262144   |  12.695 ms |  85.679 us | -12608.903 us | -99.33% |
|     32      |  2097152   | 360.893 us | 261.914 us |    -98.979 us | -27.43% |
|     64      |  2097152   | 417.313 us | 357.012 us |    -60.300 us | -14.45% |
|     128     |  2097152   | 788.544 us | 439.015 us |   -349.529 us | -44.33% |
|     256     |  2097152   |   2.518 ms | 488.347 us |  -2029.795 us | -80.61% |
|     512     |  2097152   |   5.052 ms | 551.573 us |  -4500.046 us | -89.08% |
|     32      |  16777216  |   2.042 ms |   1.893 ms |   -149.908 us |  -7.34% |
|     64      |  16777216  |   2.630 ms |   2.342 ms |   -288.120 us | -10.96% |

[PointKernel]

Looks good.

Just one small question: does the performance improvement come from bypassing the continuation char?

[karthikeyann]

Nice speedups.

[karthikeyann]

LGTM 👍

[davidwendt]

Looks good.

Just one small question: does the performance improvement come from bypassing the continuation char?

That's basically the case here. Overall this changes from a character-based iteration to a byte-based iteration which is a bit faster here by limiting character counting. Iterating by characters is not always necessary but also not always slower.

[davidwendt]

/merge