From: Andi Kleen <ak@suse.de>
Subject: {twofish,aes}-{x86_64,i586} versus C implementations
Date: Mon, 20 Aug 2007 02:34:25 +0200
Message-ID: <200708200234.25620.ak@suse.de>
Mime-Version: 1.0
Content-Type: text/plain;
  charset="us-ascii"
Content-Transfer-Encoding: 7bit
To: linux-crypto@vger.kernel.org
Content-Disposition: inline
Sender: linux-crypto-owner@vger.kernel.org

Hallo,

Currently there are two twofish and two aes implementions on x86.
Worse when both are enabled as modules a modprobe aes
will get the C version which seems to be slower on K8 
and about the same speed on Core2 on my tests.
 
Is there a specific reason why anybody would prefer the C functions
over the assembler functions? 

Possible reasons I could think of:
- If the assembler functions are optimized for a specific CPU the compiler
might be able to do a better job on other CPUs.
Is there evidence for this? I suspect it's not true.

- They are not trusted and might be buggy. I assume they have
been validated against the C versions with a wide range of input
data, correct? 

If none of these reasons are valid it might make sense to disable
the C versions for x86 and only offer the assembler versions.
Then modprobe aes|twofish would DTRT automatically.

Comments?

-Andi