RE: [PATCH v1] x86/csum: rewrite csum_partial()

From: David Laight <David.Laight@ACULAB.COM>
To: 'Alexander Duyck' <alexander.duyck@gmail.com>,
	Eric Dumazet <eric.dumazet@gmail.com>
Cc: "David S . Miller" <davem@davemloft.net>,
	Jakub Kicinski <kuba@kernel.org>, netdev <netdev@vger.kernel.org>,
	Eric Dumazet <edumazet@google.com>,
	the arch/x86 maintainers <x86@kernel.org>,
	"Peter Zijlstra" <peterz@infradead.org>
Subject: RE: [PATCH v1] x86/csum: rewrite csum_partial()
Date: Sun, 14 Nov 2021 14:21:20 +0000	[thread overview]
Message-ID: <3518c78fab894d01b391c764efffbb62@AcuMS.aculab.com> (raw)
In-Reply-To: <CAKgT0UdmECakQTinbTagiG4PWfaniP_GP6T3rLvWdP+mVrB4xw@mail.gmail.com>

From: Alexander Duyck
> Sent: 11 November 2021 21:56
...
> It might be worthwhile to beef up the odd check to account for
> anything 7 bytes or less. To address it you could do something along
> the lines of:
>     unaligned = 7 & (unsigned long) buff;
>     if (unaligned) {
>         shift = unaligned * 8;
>         temp64 = (*(unsigned long)buff >> shift) << shift;
>         buff += 8 - unaligned;
>         if (len < 8 - unaligned) {
>             shift = (8 - len - unaligned) * 8;
>             temp64 <<= shift;
>             temp64 >>= shift;
>             len = 0;
>         } else {
>             len -= 8 - unaligned;
>         }
>         result += temp64;
>         result += result < temp64;
>    }

I tried doing that.
Basically it is likely to take longer that just doing the memory reads.
The register dependency chain is just too long.

	David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)