MIME-Version: 1.0
References: 
 <CAOzEurSW8cNr6TPKsjrstnPfhf4QyQqB4tnPXGGe8N4e_v7Jig@mail.gmail.com>
 <CAN55FZ247JdiT8Sd1SRiyOJxk3Ei=pDCL4kpdP=HqLRjOhKf1Q@mail.gmail.com>
 <CAN55FZ2AxiwSah7TiQoMB==r=JKT0bOtooCB7ov4xRrGkVmJ1A@mail.gmail.com>
 <CAOzEurR5nFt=-SijfU7y0BHVcrT6RG9ovvdVfKt_uBZfEQew9w@mail.gmail.com>
 <CAOzEurSqgA69er9SzhPnXwmsVpO7-piUOuOy3dXcHOi__nSQcg@mail.gmail.com>
 <CA+K2RumC79NwWxBdofHOYo8SCSs0YCJic05Du=xOszRmoPf9FA@mail.gmail.com>
 <CAN55FZ0houfWHn8_MEEefhprZvc33jr07GrBYo+Bp2yw=TVnKA@mail.gmail.com>
 <CA+K2Ru=jHuz_Wpgar4Sobtxeb33qxx=o59ToOhZ=vpmkMqErnA@mail.gmail.com>
 <CAN55FZ1J+6eM=F5GreWEBMJcNV_gifYyYY1b6xpYzun=nWPhMQ@mail.gmail.com>
 <CAN55FZ109W90Ux_EBEqkkU2TyNqBNhdhN_1XPRGo3iiZ2L9b=A@mail.gmail.com>
 <8615c983-1662-43b4-b0c9-49d194ac33aa@dunslane.net>
 <CAN55FZ1KF7XNpm2XyG=M-sFUODai=6Z8a11xE3s4YRBeBKY3tA@mail.gmail.com>
In-Reply-To: 
 <CAN55FZ1KF7XNpm2XyG=M-sFUODai=6Z8a11xE3s4YRBeBKY3tA@mail.gmail.com>
From: KAZAR Ayoub <ma_kazar@esi.dz>
Date: Sat, 18 Oct 2025 20:46:29 +0200
Message-ID: 
 <CA+K2RunFNDMxCWMX3PFSBa_r6REVwfEekaKHwg1C8KYYGePsnA@mail.gmail.com>
Subject: Re: Speed up COPY FROM text/CSV parsing using SIMD
To: Nazir Bilal Yavuz <byavuz81@gmail.com>
Cc: Andrew Dunstan <andrew@dunslane.net>,
 Shinya Kato <shinya11.kato@gmail.com>,
	pgsql-hackers@postgresql.org
Content-Type: multipart/alternative; boundary="0000000000005e9b6e0641734596"
Archived-At: 
 <https://www.postgresql.org/message-id/CA%2BK2RunFNDMxCWMX3PFSBa_r6REVwfEekaKHwg1C8KYYGePsnA%40mail.gmail.com>
Precedence: bulk

--0000000000005e9b6e0641734596
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

Hello,

I=E2=80=99ve rebenchmarked the new heuristic patch, We still have the previ=
ous
improvements ranging from 15% to 30%. For regressions i see at maximum 3%
or 4% in the worst case, so this is solid.

I'm also trying the idea of doing SIMD inside quotes with prefix XOR using
carry less multiplication avoiding the slow path in all cases even with
weird looking input, but it needs to take into consideration the
availability of PCLMULQDQ instruction set with <wmmintrin.h> and here we
go, it quickly starts to become dirty OR we can wait for the decision to
start requiring x86-64-v2 or v3 which has SSE4.2 and AVX2.


Regards,
Ayoub Kazar

--0000000000005e9b6e0641734596
Content-Type: text/html; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

<div dir=3D"ltr"><div dir=3D"ltr"><div id=3D"gmail-:v0" class=3D"gmail-Am g=
mail-aiL gmail-Al editable gmail-LW-avf gmail-tS-tW gmail-tS-tY" aria-label=
=3D"Corps du message" role=3D"textbox" aria-multiline=3D"true" style=3D"dir=
ection:ltr;min-height:85px" tabindex=3D"1" aria-controls=3D":xm" aria-expan=
ded=3D"false"><div>Hello,</div><p dir=3D"auto" style=3D"white-space:pre-wra=
p">I=E2=80=99ve rebenchmarked the new heuristic patch, We still have the pr=
evious improvements ranging from 15% to 30%. For regressions i see at maxim=
um 3% or 4% in the worst case, so this is solid.<br></p><p style=3D"white-s=
pace:pre-wrap">I&#39;m also trying the idea of doing SIMD inside quotes wit=
h prefix XOR using carry less multiplication avoiding the slow path in all =
cases even with weird looking input, but it needs to take into consideratio=
n the availability of PCLMULQDQ instruction set with &lt;wmmintrin.h&gt; an=
d here we go, it quickly starts to become dirty OR we can wait for the deci=
sion to start requiring x86-64-v2 or v3 which has SSE4.2 and AVX2.</p><p st=
yle=3D"white-space:pre-wrap"><br></p><p style=3D"white-space:pre-wrap">Rega=
rds,<br>Ayoub Kazar</p></div></div></div>

--0000000000005e9b6e0641734596--