Newsgroups: php.internals
Path: news.php.net
Xref: news.php.net php.internals:122252
MIME-Version: 1.0
References: <CAK7R4ayXCC9_EWveFAt-BS8cJV-h9JxoLP5c+j9Y3LzPWS9xnQ@mail.gmail.com>
 <CAO__Xv6SkuD3DyNV5VM-DJKPHZboX=Pg+=5F=3aGS1tZaZAeaQ@mail.gmail.com>
 <CAJmy8YHyq6D8aB+eP8dL4jL4_Up_VO6NZTVoeNMLNkS404p3Xw@mail.gmail.com>
 <CAJmy8YHUA2yQVs3VeBpVes_rbjmsrnsAvnqxOBMY1pET91uLxw@mail.gmail.com>
 <CAO__Xv4GEnTFZkyHMH2giVRto_ebsMfdpyp4wQ6JnynvZSV52A@mail.gmail.com>
 <CAK7R4awBXVa+sKsBh1xddLhO9h0HUfN1+FcPLJmmVE2av9Ue4Q@mail.gmail.com>
 <CAO__Xv5k3Jis-VdOL9oKhhQeHW=8Mf6-URpX=bknpp6GVBviYA@mail.gmail.com>
 <CADyq6sJz1OHW_=KNYYFvj9WpRQE_d4kkB1iJcqhcPKeZ8zAMuw@mail.gmail.com> <CAJmy8YE5Q81jv4rENfNmLifNVL5erCF2rknGKGSaa2Eo1ww2eg@mail.gmail.com>
In-Reply-To: <CAJmy8YE5Q81jv4rENfNmLifNVL5erCF2rknGKGSaa2Eo1ww2eg@mail.gmail.com>
Date: Thu, 25 Jan 2024 07:59:41 +0100
Message-ID: <CADyq6sLy4XARVE-gDgHEA9FM1oZ9k0Ay5mTrXdf8r1epALJdJg@mail.gmail.com>
To: Hans Henrik Bergan <divinity76@gmail.com>
Cc: tag Knife <fenniclog@gmail.com>, Hans Henrik Bergan <hans@loltek.net>, 
	PHP Internals List <internals@lists.php.net>
Content-Type: multipart/alternative; boundary="000000000000da7490060fbfb920"
Subject: Re: [PHP-DEV] BLAKE3 hash
From: ocramius@gmail.com (Marco Pivetta)

--000000000000da7490060fbfb920
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Thu, 25 Jan 2024, 06:22 Hans Henrik Bergan, <divinity76@gmail.com> wrote=
:

> On Wed, 24 Jan 2024 at 17:59, Marco Pivetta <ocramius@gmail.com> wrote:
> >
> > Depends on the actual numbers: is there any way to make a comparison th=
at
> > is relatively stable across architectures?
> >
> > Would it be feasible to start with the
> > cross-platform-let-the-compiler-do-its-job version (that somebody may
> > actually be capable of auditing), and then introduce other versions whe=
n
> > the jump is significant enough?
> >
>
> don't know about "relatively stable across architectures" but wrote
> some benchmarking code, keep reading.
>
>
>
> On Wed, 24 Jan 2024 at 17:55, tag Knife <fenniclog@gmail.com> wrote:
> > Should we even be considering the specific instruction implementations?
> > I've always been in the camp
> > of you are not smarter than the compiler. As even the best human writte=
n
> > ASM code can be slower
> > than the obscure instructions the compiler might choose to use in a wei=
rd
> > and wonderful way.
>
> The BLAKE3 team is smarter than GCC11.4, even with -march=3Dnative
> -mtune=3Dnative, which is *not* commonly used in PHP,
> the compiler didn't stand a chance against the hand-optimized assembly
> versions,
>
> wrote some benchmarks, but the TL;DR is:
> portable -O2 usually used by PHP managed 1126MB/s,
> portable -O2 -march=3Dnative managed 533MB/s (wtf? gcc obviously got
> something wrong here),
> hand-written -O2 SSE2  managed 3144MB/s,
> hand-written -O2 SSE41 managed 3332MB/s,
> hand-written -O2 avx2 managed 6554MB/s,
> hand-writen -O2 AVX512 managed 8913MB/s,
> on my AMD Ryzen 9 7950x,
> benchmarking code:
> https://gist.github.com/divinity76/5729472dd5d77e94cd0acb245aac2226
> raw output:
> array(6) {
>   ["O2-portable-march"]=3D>
>   array(2) {
>     ["microseconds_for_16_kib"]=3D>
>     int(29295)
>     ["mb_per_second"]=3D>
>     float(533.3674688513398)
>   }
>   ["O2-portable"]=3D>
>   array(2) {
>     ["microseconds_for_16_kib"]=3D>
>     int(13876)
>     ["mb_per_second"]=3D>
>     float(1126.0449697319111)
>   }
>   ["O2-sse2"]=3D>
>   array(2) {
>     ["microseconds_for_16_kib"]=3D>
>     int(4969)
>     ["mb_per_second"]=3D>
>     float(3144.4958744214127)
>   }
>   ["O2-sse41"]=3D>
>   array(2) {
>     ["microseconds_for_16_kib"]=3D>
>     int(4688)
>     ["mb_per_second"]=3D>
>     float(3332.977815699659)
>   }
>   ["O2-avx2"]=3D>
>   array(2) {
>     ["microseconds_for_16_kib"]=3D>
>     int(2384)
>     ["mb_per_second"]=3D>
>     float(6554.1107382550335)
>   }
>   ["O2-avx512"]=3D>
>   array(2) {
>     ["microseconds_for_16_kib"]=3D>
>     int(1753)
>     ["mb_per_second"]=3D>
>     float(8913.291500285226)
>   }
> }
>

Oh yes, the AVX jump is impressive =F0=9F=98=B5

--000000000000da7490060fbfb920--