Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S268321AbUJDCPr (ORCPT ); Sun, 3 Oct 2004 22:15:47 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S268326AbUJDCPr (ORCPT ); Sun, 3 Oct 2004 22:15:47 -0400 Received: from twinlark.arctic.org ([168.75.98.6]:35814 "EHLO twinlark.arctic.org") by vger.kernel.org with ESMTP id S268321AbUJDCPp (ORCPT ); Sun, 3 Oct 2004 22:15:45 -0400 Date: Sun, 3 Oct 2004 19:15:45 -0700 (PDT) From: dean gaudet To: Andi Kleen cc: Florian Bohrer , linux-kernel@vger.kernel.org, discuss@x86-64.org Subject: Re: [PATCH] AES x86-64-asm impl. In-Reply-To: Message-ID: References: <2KWl4-wq-25@gated-at.bofh.it> MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII; format=flowed Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 1235 Lines: 30 On Sat, 2 Oct 2004, Andi Kleen wrote: > Unfortunately it's still fundamentally 32bit. Anybody interested > in doing a true 64bit AES? i doubt it helps any -- except for benchmark-only purposes. there's a description of the 32-bit T-table approach in section 7.3 of basically the tables are 8-bit -> 32-bit maps, and there are 4 of them (2 for each direction). to go to 64-bit you'd need 16-bit -> 64-bit maps... 512KiB per table. there are some other variations on the tables which are smaller, but nothing as small as the 1024 bytes per table of the 32-bit implementation. there's a completely different approach using bit-slicing (basically consider each register as 64 1-bit registers), which yields great throughput but cruddy latency -- you basically need lots of non-dependant streams to make this pay off (i.e. it might work for disk crypto processing multiple blocks simultaneously). -dean - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/