| From: | "Devulapalli, Raghuveer" <raghuveer(dot)devulapalli(at)intel(dot)com> |
|---|---|
| To: | John Naylor <johncnaylorls(at)gmail(dot)com>, "R, Rakshit" <rakshit(dot)r(at)intel(dot)com> |
| Cc: | Andrei Lepikhov <lepihov(at)gmail(dot)com>, "Giacchino, Luca" <luca(dot)giacchino(at)intel(dot)com>, "pgsql-hackers(at)lists(dot)postgresql(dot)org" <pgsql-hackers(at)lists(dot)postgresql(dot)org>, "Shankaran, Akash" <akash(dot)shankaran(at)intel(dot)com>, "Paul, Sourav Kumar" <sourav(dot)kumar(dot)paul(at)intel(dot)com> |
| Subject: | RE: SIMD optimization for list_sort |
| Date: | 2025-02-20 17:25:49 |
| Message-ID: | PH8PR11MB82861E07C866BA864E258F2BFBC42@PH8PR11MB8286.namprd11.prod.outlook.com |
| Views: | Raw Message | Whole Thread | Download mbox | Resend email |
| Thread: | |
| Lists: | pgsql-hackers |
> Note that our current implemention is highly optimized for low-cardinality inputs.
> This is needed for aggregate queries. I found this write-up of a couple scalar and
> vectorized sorts, and they show this library doing very poorly on very-low
> cardinality inputs. I would look into that before trying to get something in shape to
> share.
>
> https://github.com/Voultapher/sort-research-
> rs/blob/main/writeup/intel_avx512/text.md
That write up is fairly old and those perf problems has subsequently been fixed. See https://github.com/intel/x86-simd-sort/pull/127 and https://github.com/intel/x86-simd-sort/pull/168. I still suggest measuring perf here for thoroughness.
>
> There is also the question of hardware support. It seems AVX-512 is not
> supported well on client side, where most developers work. And availability of
> any flavor is not guaranteed on server either.
> Something to keep in mind.
simd-sort also works on avx2 which is widely available. I would suggest benchmarking on one of the client laptops to measure the perf.
Raghuveer
| From | Date | Subject | |
|---|---|---|---|
| Next Message | Jacob Champion | 2025-02-20 17:28:36 | Re: [PoC] Federated Authn/z with OAUTHBEARER |
| Previous Message | Pavel Stehule | 2025-02-20 17:15:36 | Re: Proposal: pg_is_volatile function |