Improve coalesce performance

[pitrou]

Before:

--------------------------------------------------------------------------------------
Benchmark                            Time             CPU   Iterations UserCounters...
--------------------------------------------------------------------------------------
CoalesceBench64/0              3366683 ns      3364669 ns          213 bytes_per_second=2.32192G/s items_per_second=311.643M/s length=1048.58k null%=1 num_args=2
CoalesceBench64/1              2851663 ns      2849846 ns          245 bytes_per_second=2.74138G/s items_per_second=367.941M/s length=1048.58k null%=1 num_args=4
CoalesceBench64/2              7499865 ns      7495813 ns           95 bytes_per_second=1067.26M/s items_per_second=139.888M/s length=1048.58k null%=25 num_args=2
CoalesceBench64/3             11773437 ns     11766272 ns           60 bytes_per_second=679.909M/s items_per_second=89.1171M/s length=1048.58k null%=25 num_args=4
CoalesceBench64/4              9636207 ns      9631169 ns           73 bytes_per_second=830.636M/s items_per_second=108.873M/s length=1048.58k null%=50 num_args=2
CoalesceBench64/5             19456855 ns     19445858 ns           36 bytes_per_second=411.399M/s items_per_second=53.9228M/s length=1048.58k null%=50 num_args=4
CoalesceBench64/6              3288217 ns      3286426 ns          214 bytes_per_second=2.3772G/s items_per_second=319.063M/s length=1048.58k null%=99 num_args=2
CoalesceBench64/7              7603232 ns      7599720 ns           92 bytes_per_second=1052.67M/s items_per_second=137.976M/s length=1048.58k null%=99 num_args=4
CoalesceScalarBench64/0         775260 ns       774797 ns          904 bytes_per_second=10.0833G/s items_per_second=1.35336G/s length=1048.58k null%=1 num_args=2
CoalesceScalarBench64/2        3500267 ns      3498388 ns          201 bytes_per_second=2.23317G/s items_per_second=299.731M/s length=1048.58k null%=25 num_args=2
CoalesceScalarBench64/4        4815186 ns      4812821 ns          146 bytes_per_second=1.62327G/s items_per_second=217.871M/s length=1048.58k null%=50 num_args=2
CoalesceScalarBench64/6         446897 ns       446783 ns         1541 bytes_per_second=17.4861G/s items_per_second=2.34695G/s length=1048.58k null%=99 num_args=2
CoalesceScalarStringBench/0   74138532 ns     74089097 ns           10 bytes_per_second=6.72872G/s items_per_second=14.1529M/s length=1048.58k null%=1 num_args=2
CoalesceScalarStringBench/2   58106933 ns     58064020 ns            9 bytes_per_second=6.52407G/s items_per_second=18.059M/s length=1048.58k null%=25 num_args=2
CoalesceScalarStringBench/4   52094990 ns     52064312 ns           10 bytes_per_second=4.88432G/s items_per_second=20.14M/s length=1048.58k null%=50 num_args=2
CoalesceScalarStringBench/6    5136540 ns      5133121 ns          138 bytes_per_second=1.7244G/s items_per_second=204.276M/s length=1048.58k null%=99 num_args=2

After:

--------------------------------------------------------------------------------------
Benchmark                            Time             CPU   Iterations UserCounters...
--------------------------------------------------------------------------------------
CoalesceBench64/0              1047061 ns      1046399 ns          661 bytes_per_second=7.46608G/s items_per_second=1002.08M/s length=1048.58k null%=1 num_args=2
CoalesceBench64/1              1377282 ns      1376405 ns          511 bytes_per_second=5.67602G/s items_per_second=761.822M/s length=1048.58k null%=1 num_args=4
CoalesceBench64/2              2804061 ns      2802178 ns          251 bytes_per_second=2.78801G/s items_per_second=374.2M/s length=1048.58k null%=25 num_args=2
CoalesceBench64/3              5234633 ns      5230898 ns          134 bytes_per_second=1.49353G/s items_per_second=200.458M/s length=1048.58k null%=25 num_args=4
CoalesceBench64/4              3700820 ns      3698116 ns          190 bytes_per_second=2.11256G/s items_per_second=283.543M/s length=1048.58k null%=50 num_args=2
CoalesceBench64/5              7731316 ns      7726379 ns           90 bytes_per_second=1035.41M/s items_per_second=135.714M/s length=1048.58k null%=50 num_args=4
CoalesceBench64/6              1004359 ns      1003745 ns          693 bytes_per_second=7.78335G/s items_per_second=1044.66M/s length=1048.58k null%=99 num_args=2
CoalesceBench64/7              4660379 ns      4658001 ns          151 bytes_per_second=1.67722G/s items_per_second=225.113M/s length=1048.58k null%=99 num_args=4
CoalesceScalarBench64/0         656265 ns       655870 ns         1067 bytes_per_second=11.9117G/s items_per_second=1.59876G/s length=1048.58k null%=1 num_args=2
CoalesceScalarBench64/2        2889294 ns      2887898 ns          242 bytes_per_second=2.70525G/s items_per_second=363.093M/s length=1048.58k null%=25 num_args=2
CoalesceScalarBench64/4        4015990 ns      4014054 ns          175 bytes_per_second=1.94629G/s items_per_second=261.226M/s length=1048.58k null%=50 num_args=2
CoalesceScalarBench64/6         390245 ns       390138 ns         1800 bytes_per_second=20.025G/s items_per_second=2.68771G/s length=1048.58k null%=99 num_args=2
CoalesceScalarStringBench/0   82277097 ns     82223643 ns            9 bytes_per_second=6.06303G/s items_per_second=12.7527M/s length=1048.58k null%=1 num_args=2
CoalesceScalarStringBench/2   70821126 ns     70771323 ns           10 bytes_per_second=5.35265G/s items_per_second=14.8164M/s length=1048.58k null%=25 num_args=2
CoalesceScalarStringBench/4   47119447 ns     47087724 ns           13 bytes_per_second=5.40053G/s items_per_second=22.2686M/s length=1048.58k null%=50 num_args=2
CoalesceScalarStringBench/6    4579486 ns      4576728 ns          150 bytes_per_second=1.93403G/s items_per_second=229.11M/s length=1048.58k null%=99 num_args=2

[pitrou]

@lidavidm You should probably enable Github Actions on this fork.

[lidavidm]

I re-enabled them (I keep them disabled usually because they generate so much email spam on every push 🙁)

[pitrou]

Ah, I disabled those notifications (not sure how I did that, but I don't get emails from Github Actions).

[lidavidm]

Thanks for taking a look at this, this is quite the impressive improvement.