Newsgroups: php.internals
Path: news.php.net
Xref: news.php.net php.internals:112154
MIME-Version: 1.0
References: <CAJmy8YHhx+Q+9mAY64oyHAN+KVErREZ+sH2oXdFrMxp9n4+ShQ@mail.gmail.com>
In-Reply-To: <CAJmy8YHhx+Q+9mAY64oyHAN+KVErREZ+sH2oXdFrMxp9n4+ShQ@mail.gmail.com>
Date: Sat, 31 Oct 2020 01:07:49 +0100
Message-ID: <CAJmy8YFV5y4U8N+8F4vfyS=xFoy-+bQWKmMvUg_JoOSSCamTQQ@mail.gmail.com>
To: PHP internals <internals@lists.php.net>
Content-Type: text/plain; charset="UTF-8"
Subject: Re: hash BLAKE3
From: divinity76@gmail.com (Hans Henrik Bergan)

hmm i'll try again,
made another PR: https://github.com/php/php-src/pull/6393

this PR is an alternative to https://github.com/php/php-src/pull/6358
with some notable differences,

- this PR is roughly 1700 SLOC, compared to #6358 at roughly 31,000 SLOC

- this PR only bundles the "portable C" version of BLAKE3, it does not
contain any of the SSE/AVX/NEON cpu-optimized non-portable
implementations (hence the huge SLOC difference)

the idea here is that just having BLAKE3 supported in PHP, even if
it's not the most performant implementation, is better than no support
at all,

unsurprisingly, the portable version of BLAKE3 isn't nearly as fast as
in #6358 , being beaten by MD4 (still beats MD5 though), but it still
does a pretty good job in /ext/hash/bech.php, being roughly 1.9 times
faster than sha3-224 on my laptop (compared to 4.2 times faster in
#6358 ), and being the fastest secure hash on the list, for exact
numbers check the PR.

thoughts?

On Wed, 21 Oct 2020 at 21:34, Hans Henrik Bergan <divinity76@gmail.com> wrote:
>
> i want BLAKE3 hash support in php,
>
> some justifications can be found here https://bugs.php.net/bug.php?id=79492
> primary reasons being that it's a "cryptographically secure hash",
> predecessor BLAKE was nearly crowned "SHA3" (SHA3 finalist),
> and it's very fast in software,
> when i run it against php-src/ext/hash/bench.php , on a cpu without
> the most beneficial instruction-set available, it was still over 4
> times faster than sha3-224, outperforming even several
> non-crypto-hashes, exact numbers here:
> https://github.com/php/php-src/pull/6358#issuecomment-713506971 )
>
> i have a PR here: https://github.com/php/php-src/pull/6358
> initially had several issues, but i think most of them are resolved now,
>
> - SSE/AVX optimizations are added for "gcc+(unix|windows)+x86_64" targets,
> and i have asked the php-windows mailing list for help in adding
> SSE/AVX optimizations to the MSVC builds,
> but for the time being, MSVC builds are just using a portable generic
> C implementation.
> it would also be possible to add an optimized implementation for
> ARM+Neon targets, i haven't looked into that at all.
>
> - currently the code is rolling it's own cpu feature detection,
> instead of asking Zend/zend_cpuinfo.c  (specifically
> ext/hash/blake3/blake3_dispatch.c ),
> is that something that "must be fixed" before upstream merging can be
> considered? or is it ok-ish?
>
> - frankly, there's just *a lot* of code in there, is it possible that
> it will be rejected just because "it's too much code" ? (there's a
> reason there's so much code though, it contains cpu-specific
> implementations for a lot of different environments, compilers, OS's,
> and CPU's, from SSE2 to AVX512 to ARM+Neon, also a much smaller
> implementation using just the "portable generic C implementation"
> would be possible, at the cost of performance)