Why is the performance of my custom functions so different from that of numerical operations?

[wodejiong]

I implemented a date_trunc operator that handles one of timestamps.

It takes about 3 to 4 seconds to execute the following SQL statement:
SELECT date_trunc('hour', time) AS time, count(*)AS PV FROM log GROUP BY time ORDER BY time

However, when I do numerical calculations, similar queries, with the same amount of data, are within 1 second.
SELECT (__time/(60601000)(60601000)) AS time, count()AS PV FROM log GROUP BY time ORDER BY time

What's the difference between the two, and how do I optimize my operator.

[hashhar]

The logic is obviously much more involved in the date_trunc case compared to just numerical operations. You should consider recording a JFR profile to see where the time is spent in both cases.

Also for some operations IIRC the engine can apply them in parallel to an entire Page instead of per-row (I'm not 100% sure about this to be honest).