Re: [PATCH v2] raid6: arm64: add SVE optimized implementation for syndrome generation

Mark Brown <[email protected]> Fri, 17 Apr 2026 16:36:44 +0100

Newsgroups	gmane.linux.kernel,gmane.linux.ports.arm.kernel,gmane.linux.raid
Message-ID	<[email protected]>

On Fri, Apr 17, 2026 at 04:43:06PM +0200, Ard Biesheuvel wrote:

> On arm64, kernel mode NEON is mostly used to gain access to AES and SHA
> instructions, and only to a lesser degree to speed up ordinary
> arithmetic, and so XOR is somewhat of an outlier here.

> Given that Neoverse V1 apparently already carves up ordinary arithmetic
> performed on 256-bit vectors and operates on 128 bits at a time, I am
> rather skeptical that we're likely to see any SVE implementations of the
> crypto extensions soon that are meaningfully faster, given that these
> are presumably much costlier to implement in terms of gate count, and
> therefore likely to be split up even on SVE implementations that can
> perform ordinary arithmetic on 256+ bit vectors in a single cycle. Note
> that even the arm64 SIMD accelerated CRC implementations rely heavily on
> 64x64->128 polynomial multiplication.

I'd not be surprised to see something that delivers useful benefits
using SVE at some point.

> IOW, before we consider kernel mode SVE, I'd like to see some benchmarks
> for other algorithms too.

Definitely, it needs a solid win to merge anything.  I do want to get
back to the situation where we've got out of tree infrastructure patches
so that people working on algorithms have something to base their work
on (and see the overheads using SVE incurs) but unless theres's a
practical user they should stay out of tree.

signature.asc (application/pgp-signature, 488 B)