[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Fri, 04 Apr 2014 21:46:41 +0100
From: Samuel Neves <sneves@....uc.pt>
To: discussions@...sword-hashing.net
Subject: Re: [PHC] Gambit review
On 04-04-2014 21:37, Bill Cox wrote:
> Now that I've seen a couple entries that use the new AESENC
> instruction, I'm more of a fan. It has a 2 cycle latency, and the CPU
> can queue a sick number of them per core, like 8.
Just a bit of pedantry: the best latency for AESENC I know of is 7 cycles on Haswell. You can dispatch 1 every cycle
(per core) in every chip, though. What this roughly means is that you can perform ~7-8 parallel AES encryptions for the
cost of one.
Powered by blists - more mailing lists