From: Denys Vlasenko Subject: Re: [PATCH 1/4] camellia: 64-bit optimization Date: Thu, 22 Nov 2007 14:29:36 -0800 Message-ID: <200711221429.36309.vda.linux@googlemail.com> References: <200711210022.28510.vda.linux@googlemail.com> <200711210023.49410.vda.linux@googlemail.com> <20071122122528.GA9556@gondor.apana.org.au> Mime-Version: 1.0 Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: 7bit Cc: Noriaki TAKAMIYA , davem@davemloft.net, linux-crypto@vger.kernel.org To: Herbert Xu Return-path: Received: from rv-out-0910.google.com ([209.85.198.185]:25866 "EHLO rv-out-0910.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754938AbXKVW3o (ORCPT ); Thu, 22 Nov 2007 17:29:44 -0500 Received: by rv-out-0910.google.com with SMTP id k20so2478480rvb for ; Thu, 22 Nov 2007 14:29:43 -0800 (PST) In-Reply-To: <20071122122528.GA9556@gondor.apana.org.au> Content-Disposition: inline Sender: linux-crypto-owner@vger.kernel.org List-Id: linux-crypto.vger.kernel.org On Thursday 22 November 2007 04:25, Herbert Xu wrote: > On Wed, Nov 21, 2007 at 12:23:49AM -0800, Denys Vlasenko wrote: > > On Wednesday 21 November 2007 00:22, Denys Vlasenko wrote: > > > Patches start from 5 because pathches 1..4 are already > > > in cryptodev-2.6. > > > > > > camellia5: > > > adds 64-bit key setup, it is used if BITS_PER_LONG is 64. > > > > Signed-off-by: Denys Vlasenko > > Denys, I really appreciate your work in cleaning this algorithm up. > > However, I just can't accept 600 lines of new code just to do setkey > faster/smaller on 64-bit platforms. You know, some people have to > study these files line-by-line in order have them pass government > certifications and such. 480 lines after all patches are applied. > OK I went back and looked at the figures, so we're talking about > either 1K extra on 64-bit if we use the 32-bit version or 2K extra > on 32-bit if we went for the 64-bit version. > > Either way it doesn't seem to be a big deal. So how about just > picking one and sticking with it? "No big deal" approach resulted in thish module having 30k of code. I cut it down to 20-18k. In aic7xxx driver, similarly, I rediced size from 300k to 200k. Still not in mainline. This is depressing. I am an -Os guy and for me 1k out of 20k module doesn't sound too small. I will rework and repost patches so that 64-bit code is added by last patch, feel free to not apply it. -- vda