perf: optimize spark_lpad (up to 2x faster)

[andygrove]

Which issue does this PR close?

N/A

Rationale for this change

Optimize existing expression.

What changes are included in this PR?

Pad rows with one memcpy from a precomputed repeating pad buffer and write pieces directly into the string builder instead of a scratch String, plus an ASCII fast path and a fix for a rows*rows capacity hint in the column-length path.

How are these changes tested?

Existing tests + new unit tests.

Benchmark (criterion):

• spark_rpad_ default padding: 40.449% faster (base 130815ns -> cand 77902ns)
• spark_rpad_ length from column, multi-char padding: 41.531% faster (base 128111ns -> cand 74905ns)
• spark_lpad_ default padding: 37.778% faster (base 126685ns -> cand 78826ns)
• spark_lpad_ length from column: 31.923% faster (base 110963ns -> cand 75540ns)
• spark_lpad_ custom padding: 35.618% faster (base 123106ns -> cand 79258ns)
• spark_rpad_ multi-char padding: 52.516% faster (base 165109ns -> cand 78401ns)
• spark_lpad_ with truncation: 50.694% faster (base 110062ns -> cand 54268ns)
• spark_lpad_ multi-char padding: 57.246% faster (base 181356ns -> cand 77536ns)
• spark_rpad_ length from column: 32.23% faster (base 110303ns -> cand 74752ns)
• spark_rpad_ custom padding: 40.385% faster (base 130616ns -> cand 77866ns)
• spark_rpad_ with truncation: 50.556% faster (base 109976ns -> cand 54377ns)

Full criterion output:

spark_lpad: default padding
                        time:   [78.908 µs 79.083 µs 79.261 µs]
                        change: [−37.968% −37.778% −37.621%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 6 outliers among 100 measurements (6.00%)
  6 (6.00%) high mild

spark_lpad: custom padding
                        time:   [78.999 µs 79.093 µs 79.196 µs]
                        change: [−35.699% −35.618% −35.554%] (p = 0.00 < 0.05)
                        Performance has improved.

spark_rpad: default padding
                        time:   [78.086 µs 78.410 µs 78.731 µs]
                        change: [−40.596% −40.449% −40.294%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 20 outliers among 100 measurements (20.00%)
  3 (3.00%) high mild
  17 (17.00%) high severe

spark_rpad: custom padding
                        time:   [77.806 µs 77.855 µs 77.931 µs]
                        change: [−40.474% −40.385% −40.290%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 6 outliers among 100 measurements (6.00%)
  3 (3.00%) high mild
  3 (3.00%) high severe

spark_lpad: multi-char padding
                        time:   [77.424 µs 77.539 µs 77.677 µs]
                        change: [−57.312% −57.246% −57.187%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 3 outliers among 100 measurements (3.00%)
  3 (3.00%) high mild

spark_rpad: multi-char padding
                        time:   [78.306 µs 78.409 µs 78.530 µs]
                        change: [−52.583% −52.516% −52.442%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 7 outliers among 100 measurements (7.00%)
  6 (6.00%) high mild
  1 (1.00%) high severe

spark_lpad: with truncation
                        time:   [54.249 µs 54.317 µs 54.393 µs]
                        change: [−50.755% −50.694% −50.637%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 13 outliers among 100 measurements (13.00%)
  1 (1.00%) low mild
  11 (11.00%) high mild
  1 (1.00%) high severe

spark_rpad: with truncation
                        time:   [54.146 µs 54.203 µs 54.268 µs]
                        change: [−50.698% −50.556% −50.370%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 3 outliers among 100 measurements (3.00%)
  3 (3.00%) high severe

spark_lpad: length from column
                        time:   [75.318 µs 75.499 µs 75.686 µs]
                        change: [−32.039% −31.923% −31.795%] (p = 0.00 < 0.05)
                        Performance has improved.

spark_rpad: length from column
                        time:   [74.721 µs 74.759 µs 74.818 µs]
                        change: [−32.298% −32.230% −32.168%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 4 outliers among 100 measurements (4.00%)
  2 (2.00%) high mild
  2 (2.00%) high severe

spark_rpad: length from column, multi-char padding
                        time:   [74.876 µs 74.937 µs 75.016 µs]
                        change: [−41.976% −41.531% −41.269%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 5 outliers among 100 measurements (5.00%)
  3 (3.00%) high mild
  2 (2.00%) high severe

[comphead]

Hopefully this every row cast is not expensive on modern CPUs

[comphead]

Thanks @andygrove