Re: [PATCH v2 2/2] random: use BLAKE2s instead of SHA1 in extraction

From: "Jason A. Donenfeld" <Jason@zx2c4.com>
To: David Laight <David.Laight@aculab.com>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
	Theodore Tso <tytso@mit.edu>,
	Greg KH <gregkh@linuxfoundation.org>,
	Jean-Philippe Aumasson <jeanphilippe.aumasson@gmail.com>
Subject: Re: [PATCH v2 2/2] random: use BLAKE2s instead of SHA1 in extraction
Date: Tue, 11 Jan 2022 19:26:03 +0100	[thread overview]
Message-ID: <CAHmME9rvRZVgy+1wr94dC8pCV-qgLSm9mk7FkqbtjM0xz=HT1w@mail.gmail.com> (raw)
In-Reply-To: <caed82818cdb466aade033501f57d183@AcuMS.aculab.com>

On Tue, Jan 11, 2022 at 4:47 PM David Laight <David.Laight@aculab.com> wrote:
> > -     int i;
> > +     int i, j;
>
> Use unsigned int i, j;
> Ensures the '% 4' are done as '& 3' and the divides as shifts.
> Unless the compiler manages to track the valid values that will
> even generate better code on x86-64.
> (Saves a sign extension prior to the array indexes.)

Ack.

> I think I'd look at doing [0..3] then [4..7] to save execution time.

I actually wound up making the same change to sha1 instead of blake2s
for v2 of this, and achieved pretty similar results, but I think
that's more satisfactory of a conclusion. v2 is here:
https://lore.kernel.org/linux-crypto/20220111181037.632969-1-Jason@zx2c4.com/T/#u

Jason