MIME-Version: 1.0
References: 
 <CAEudQApbWon+3Eb9x4WW_D-JkSt2mvfx99dXu9VZ4AeCuTh=fw@mail.gmail.com>
 <CAEudQApEfvhNT1fEPURzVcQH7G0A1ukh_ugoCGaErV6_dbndCQ@mail.gmail.com>
In-Reply-To: 
 <CAEudQApEfvhNT1fEPURzVcQH7G0A1ukh_ugoCGaErV6_dbndCQ@mail.gmail.com>
From: Bryan Green <dbryan.green@gmail.com>
Date: Mon, 9 Mar 2026 08:45:56 -0600
Message-ID: 
 <CAF+pBj_RS2KErTqQ6ORXjhVzmukG7Ve0wHU1Kq56xjJfFKwVqA@mail.gmail.com>
Subject: Re: Avoid multiple calls to memcpy (src/backend/access/index/genam.c)
To: Ranier Vilela <ranier.vf@gmail.com>
Cc: Pg Hackers <pgsql-hackers@postgresql.org>
Content-Type: multipart/alternative; boundary="0000000000006619ca064c987ad1"
Archived-At: 
 <https://www.postgresql.org/message-id/CAF%2BpBj_RS2KErTqQ6ORXjhVzmukG7Ve0wHU1Kq56xjJfFKwVqA%40mail.gmail.com>
Precedence: bulk

--0000000000006619ca064c987ad1
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

I created an example that is a little bit closer to the actual code and
changed the compiler from C++ to C.

It is interesting the optimization that the compiler has chosen for version
1 versus version 2.  One calls
memcpy and one doesn't.  There is a good chance the inlining of memcpy as
SSE+scalar per iteration
will be faster for syscache scans-- which I believe are usually small (1-4
keys?).

Probably the only reason to do this patch would be if N is normally large
or if this is considered an
improvement in code clarity without a detrimental impact on small N
syscache scans.
I realize you only said "possible small optimization".  It might be
worthwhile to benchmark the code for
different values of n to determine if there is a tipping point either way?

https://godbolt.org/z/dM18cGfE6

-- bg

On Mon, Mar 9, 2026 at 8:05=E2=80=AFAM Ranier Vilela <ranier.vf@gmail.com> =
wrote:

>
> Em seg., 9 de mar. de 2026 =C3=A0s 10:16, Ranier Vilela <ranier.vf@gmail.=
com>
> escreveu:
>
>> Hi.
>>
>> In the functions *systable_beginscan* and *systable_beginscan_ordered*,
>> is possible a small optimization.
>> The array *idxkey* can be constructed in one go with a single call to
>> mempcy.
>> The excess might not make much of a difference, but I think it's worth
>> the effort.
>>
>> patch attached.
>>
> Someone asked me if O2 does not do the work.
> Apparently not.
>
> https://godbolt.org/z/h5dndz33x
>
> best regards,
> Ranier Vilela
>

--0000000000006619ca064c987ad1
Content-Type: text/html; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

<div dir=3D"ltr">I created an example that is a little bit closer to the ac=
tual code and changed the compiler from C++ to C.=C2=A0<div><br></div><div>=
It is interesting the optimization that the compiler has chosen for version=
 1 versus version 2.=C2=A0 One calls</div><div>memcpy and one doesn&#39;t.=
=C2=A0 There is a good chance the inlining of memcpy as SSE+scalar per iter=
ation</div><div>will be faster for syscache scans-- which I believe are usu=
ally small (1-4 keys?).=C2=A0=C2=A0</div><div><br></div><div>Probably the o=
nly reason to do this patch would be if N is normally large or if this is c=
onsidered an</div><div>improvement in code clarity without a detrimental=C2=
=A0impact on small N syscache scans.=C2=A0=C2=A0</div><div>I realize you on=
ly said &quot;possible small optimization&quot;.=C2=A0 It might be worthwhi=
le to benchmark the code for=C2=A0</div><div>different values of n to deter=
mine if there is a tipping point either way?</div><div><br></div><div><a hr=
ef=3D"https://godbolt.org/z/dM18cGfE6">https://godbolt.org/z/dM18cGfE6</a><=
/div><div><br></div><div>-- bg</div></div><br><div class=3D"gmail_quote gma=
il_quote_container"><div dir=3D"ltr" class=3D"gmail_attr">On Mon, Mar 9, 20=
26 at 8:05=E2=80=AFAM Ranier Vilela &lt;<a href=3D"mailto:ranier.vf@gmail.c=
om">ranier.vf@gmail.com</a>&gt; wrote:<br></div><blockquote class=3D"gmail_=
quote" style=3D"margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,=
204);padding-left:1ex"><div dir=3D"ltr"><br><div class=3D"gmail_quote"><div=
 dir=3D"ltr" class=3D"gmail_attr">Em seg., 9 de mar. de 2026 =C3=A0s 10:16,=
 Ranier Vilela &lt;<a href=3D"mailto:ranier.vf@gmail.com" target=3D"_blank"=
>ranier.vf@gmail.com</a>&gt; escreveu:<br></div><blockquote class=3D"gmail_=
quote" style=3D"margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,=
204);padding-left:1ex"><div dir=3D"ltr"><div>Hi.</div><div><br></div><div>I=
n the functions *systable_beginscan* and *systable_beginscan_ordered*,</div=
><div>is possible a small optimization.</div><div>The array *idxkey* can be=
 constructed in one go with a single call to mempcy.</div><div>The excess m=
ight not make much of a difference, but I think it&#39;s worth the effort.<=
/div><div><br></div><div>patch attached.</div></div></blockquote><div>Someo=
ne asked me if O2 does not do the work.</div><div>Apparently not.</div><div=
><br></div><div><a href=3D"https://godbolt.org/z/h5dndz33x" target=3D"_blan=
k">https://godbolt.org/z/h5dndz33x</a></div><div><br></div><div>best regard=
s,</div><div>Ranier Vilela</div></div></div>
</blockquote></div>

--0000000000006619ca064c987ad1--