On 11/03/2025 20:28, Bernd Helmle wrote:
> Please find a new rebased version of this patch.
Hmm, if we implement sortsupport function for GiST, we can register it
for B-tree opfamily as well. The range comparison function has quite
high overhead thanks to detoasting, but I'm nevertheless seeing a tiny
speedup. I tested that with:
setup:
create table intranges (r int4range);
insert into intranges select int4range(g, g+10) from generate_series(1,
1000000) g;
vacuum freeze intranges;
test script 'rangesort.sql':
set work_mem='100 MB';
explain analyze select r from intranges order by r;
test:
pgbench -n -f rangesort.sql -P1 postgres -t100
On my laptop, that reports avg latency of 152 ms on master, and 149 ms
latency with the patch. Nothing to write home about, but we might as
well take it if we have the sortsupport function for gist anyway.
So I added it for the btree opfamily too, and moved the function to
rangetypes.c since it's not just for gist anymore. I Ccmmitted that
part, and will start looking more closely at the remaining btree_gist
parts now.
--
Heikki Linnakangas
Neon (https://neon.tech)
