Date: Sat, 8 Sep 2018 12:44:14 +0200 (CEST)
From: Andreas Joseph Krogh <andreas@visena.com>
To: pgsql-hackers@lists.postgresql.org
Message-ID: <VisenaEmail.6.ca44a189d8b47f10.165b8a8b449@tc7-visena>
In-Reply-To: <VisenaEmail.2.c105fe39c156917b.165659f1429@tc7-visena>
Subject: Sv: Re: Query is over 2x slower with jit=on
MIME-Version: 1.0
Content-Type: multipart/mixed;
	boundary="----=_Part_25_1108310266.1536403454402"
Precedence: bulk

------=_Part_25_1108310266.1536403454402
Content-Type: multipart/related;
	boundary="----=_Part_26_467446946.1536403454402"

------=_Part_26_467446946.1536403454402
Content-Type: multipart/alternative;
	boundary="----=_Part_27_646758791.1536403454412"

------=_Part_27_646758791.1536403454412
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: quoted-printable

P=C3=A5 torsdag 23. august 2018 kl. 09:14:56, skrev Andreas Joseph Krogh <
andreas@visena.com <mailto:andreas@visena.com>>:
P=C3=A5 torsdag 23. august 2018 kl. 03:00:42, skrev Jonathan S. Katz <
jkatz@postgresql.org <mailto:jkatz@postgresql.org>>:

 > On Aug 22, 2018, at 7:13 PM, Andres Freund <andres@anarazel.de> wrote:
 [snip]
 > For the archives sake: This likely largely is the consequence of
 > building with LLVM's expensive assertions enabled, as confirmed by
 > Jonathan over IM.

 I recompiled with the release version of LLVM. jit=3Don was still slower,
 but the discrepancy was not as bad as the previously reported result:

 jit =3D off
 Planning Time: 0.938 ms
 Execution Time: 935.599 ms

 jit =3D on
 Planning Time: 0.951 ms
 JIT:
 =C2=A0 Functions: 184
 =C2=A0 Generation Time: 17.605 ms
 =C2=A0 Inlining: true
 =C2=A0 Inlining Time: 20.522 ms
 =C2=A0 Optimization: true
 =C2=A0 Optimization Time: 1001.034 ms
 =C2=A0 Emission Time: 665.319 ms
 Execution Time: 2491.560 ms

 However, it was still 2x+ slower, so still +1ing for open items.
=C2=A0
=C2=A0
I compiled with whatever switches LLVM=C2=A0that comes with=C2=A0Ubuntu 18.=
04=C2=A0is built=20
with, and without debugging or=C2=A0assertions.
=C2=A0
With 11b3 as of 825f10fbda7a5d8a48d187b8193160e5e44e4011 I'm repeatedly=20
getting these results with jit=3Don, after 10 runs:
=C2=A0
=C2=A0
=C2=A0
Planning Time: 0.266 ms=20
 JIT:
 =C2=A0Functions: 686
 =C2=A0Generation Time: 71.895 ms
 =C2=A0Inlining: false
 =C2=A0Inlining Time: 0.000 ms
 =C2=A0Optimization: false
 =C2=A0Optimization Time: 39.906 ms
 =C2=A0Emission Time: 589.944 ms
 Execution Time: 2198.928 ms
 =C2=A0
=C2=A0
=C2=A0
Turning jit=3Doff gives this:
=C2=A0
Planning Time: 0.180 ms
Execution Time: 938.451 ms
=C2=A0
I can provide dataset offlist if anyone wants to look into this.
=C2=A0
--
 Andreas Joseph Krogh


------=_Part_27_646758791.1536403454412
Content-Type: text/html;charset=UTF-8
Content-Transfer-Encoding: quoted-printable

<div>P=C3=A5 torsdag 23. august 2018 kl. 09:14:56, skrev Andreas Joseph Kro=
gh &lt;<a href=3D"mailto:andreas@visena.com">andreas@visena.com</a>&gt;:</d=
iv>

<blockquote style=3D"border-left: 1px solid rgb(204, 204, 204); margin: 0pt=
 0pt 0pt 0.8ex; padding-left: 1ex;">
<div>P=C3=A5 torsdag 23. august 2018 kl. 03:00:42, skrev Jonathan S. Katz &=
lt;<a href=3D"mailto:jkatz@postgresql.org">jkatz@postgresql.org</a>&gt;:</d=
iv>

<blockquote style=3D"border-left: 1px solid rgb(204, 204, 204); margin: 0pt=
 0pt 0pt 0.8ex; padding-left: 1ex;">
<div style=3D"display:inline; font-family: monospace; font-size: 12px;"><br=
>
&gt; On Aug 22, 2018, at 7:13 PM, Andres Freund &lt;andres@anarazel.de&gt; =
wrote:<br>
[snip]<br>
&gt; For the archives sake: This likely largely is the consequence of<br>
&gt; building with LLVM's expensive assertions enabled, as confirmed by<br>
&gt; Jonathan over IM.<br>
<br>
I recompiled with the release version of LLVM. jit=3Don was still slower,<b=
r>
but the discrepancy was not as bad as the previously reported result:<br>
<br>
jit =3D off<br>
Planning Time: 0.938 ms<br>
Execution Time: 935.599 ms<br>
<br>
jit =3D on<br>
Planning Time: 0.951 ms<br>
JIT:<br>
=C2=A0 Functions: 184<br>
=C2=A0 Generation Time: 17.605 ms<br>
=C2=A0 Inlining: true<br>
=C2=A0 Inlining Time: 20.522 ms<br>
=C2=A0 Optimization: true<br>
=C2=A0 Optimization Time: 1001.034 ms<br>
=C2=A0 Emission Time: 665.319 ms<br>
Execution Time: 2491.560 ms<br>
<br>
However, it was still 2x+ slower, so still +1ing for open items.</div>
</blockquote>

<div>=C2=A0</div>

<div>=C2=A0</div>

<div>I compiled with whatever switches LLVM=C2=A0that comes with=C2=A0Ubunt=
u 18.04=C2=A0is built with, and without debugging or=C2=A0assertions.</div>
</blockquote>

<div>=C2=A0</div>

<div>With 11b3 as of 825f10fbda7a5d8a48d187b8193160e5e44e4011 I'm repeatedl=
y getting these results with jit=3Don, after 10 runs:</div>

<div>=C2=A0</div>

<div>=C2=A0</div>

<div>=C2=A0</div>

<div><span style=3D"font-family:monospace"><span style=3D"color: rgb(0, 0, =
0);">Planning Time: 0.266 ms </span><br>
JIT:<br>
=C2=A0Functions: 686<br>
=C2=A0Generation Time: 71.895 ms<br>
=C2=A0Inlining: false<br>
=C2=A0Inlining Time: 0.000 ms<br>
=C2=A0Optimization: false<br>
=C2=A0Optimization Time: 39.906 ms<br>
=C2=A0Emission Time: 589.944 ms<br>
Execution Time: 2198.928 ms</span><br>
=C2=A0</div>

<div>=C2=A0</div>

<div>=C2=A0</div>

<div>Turning jit=3Doff gives this:</div>

<div>=C2=A0</div>

<div><span style=3D"font-family:monospace"><span style=3D"color: rgb(0, 0, =
0);">Planning Time: 0.180 ms</span></span></div>

<div><span style=3D"font-family:monospace">Execution Time: 938.451 ms</span=
></div>

<div>=C2=A0</div>

<div>I can provide dataset offlist if anyone wants to look into this.</div>

<div>=C2=A0</div>

<div class=3D"visena-email-signature" data-visena-message-id=3D"&lt;VisenaE=
mail.6.ca44a189d8b47f10.165b8a8b449@tc7-visena>">
<div>--<br>
Andreas Joseph Krogh</div>
</div>
------=_Part_27_646758791.1536403454412--

------=_Part_26_467446946.1536403454402--

------=_Part_25_1108310266.1536403454402--