Subject: Re: [PATCH] audo-detect and use -moutline-atomics compilation flag
 for
 aarch64
Thread-Topic: [PATCH] audo-detect and use -moutline-atomics compilation flag
 for aarch64
From: "Pop, Sebastian" <spop@amazon.com>
To: Christoph Berg <myon@debian.org>
CC: PostgreSQL in Debian <pgsql-pkg-debian@postgresql.org>, "Saidi, Ali"
	<alisaidi@amazon.com>, "Csoma, Csaba" <csabac@amazon.com>
Thread-Index: AQHX1xz/aMnwG7BEBkCMiUUHc3Ygaav+jI+UgABHwWU=
Date: Thu, 11 Nov 2021 21:21:30 +0000
Message-ID: <1636665689666.18737@amazon.com>
References: <1636497120606.54267@amazon.com>
 <1635266397152.45198@amazon.com>,<YY1LHr9YYQb0Ds8h@msg.df7cb.de>,<1636650661046.35954@amazon.com>
In-Reply-To: <1636650661046.35954@amazon.com>
Accept-Language: en-US
Content-Language: en-US
Content-Type: text/plain; charset="iso-8859-1"
Content-Transfer-Encoding: quoted-printable
MIME-Version: 1.0
Archived-At: 
 <https://www.postgresql.org/message-id/1636665689666.18737%40amazon.com>
Precedence: bulk

>> The benchmarks in=0A=
>> https://www.postgresql.org/message-id/1C8D0E58-FB33-4105-AC00-8FA07621F5=
DD%40amazon.com=0A=
>> don't look like the option is a game-changer.=0A=
>=0A=
> Since then we have seen benchmarks that are several orders of magnitude f=
aster with LSE instructions.=0A=
> I will get the new data and share.=0A=
=0A=
On an m6gd.metal with PostgreSQL v11=0A=
test: pgbench - select-only=0A=
Measure tps (transactions/s), higher is better.=0A=
=0A=
build without LSE: 132143 tps =0A=
with LSE atomics: 844552 tps=0A=
=0A=
That is a 6.39x speedup.=0A=
=0A=
Sebastian=