From mboxrd@z Thu Jan 1 00:00:00 1970 From: "H. Peter Anvin" Subject: Re: raid6's using not the best bandwidth method && raid6 algo is significantly slower in x86_64. Date: Fri, 21 Nov 2008 12:15:10 -0800 Message-ID: <492716CE.8040900@zytor.com> References: <200811172235.mAHMZAQt032187@terminus.zytor.com> <43d009740811180403v6308dbc0y8ba093524700bb72@mail.gmail.com> <4922E381.20501@zytor.com> <43d009740811211122m36809523mff9521ca8af95a1e@mail.gmail.com> <49270CA5.2090508@zytor.com> <43d009740811211133p7321bc39q1a1eccd9c53d6ca0@mail.gmail.com> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <43d009740811211133p7321bc39q1a1eccd9c53d6ca0@mail.gmail.com> Sender: linux-raid-owner@vger.kernel.org To: for.poige+linux@gmail.com Cc: linux-raid@vger.kernel.org List-Id: linux-raid.ids Igor Podlesny wrote: > 2008/11/22 H. Peter Anvin : > [...] >> It sort of makes sense that -Os would break this stuff. For newer gccs, >> it would be better to use SSE2 intrinsics rather than inline assembly. >> The problem is that it breaks older gcc. >> > Well, #ifdef could be helpful then, couldn't it? :-) > Yes, and that's probably the way to go. I just tested a version using gcc intrinsics with gcc 4.3, and it is almost 20% faster than the inline assembly version. That, plus the fact that the code is actually readable, makes me really want to figure out how best to deploy this. -hpa