Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Tue, 16 Apr 2013 11:06:15 +0200
From: magnum <john.magnum@...hmail.com>
To: "john-dev@...ts.openwall.com" <john-dev@...ts.openwall.com>
Subject: Re: [patch] sse/xop implementation of raw-sha256

On 16 Apr, 2013, at 4:26 , jfoug@....net wrote:
> ---- magnum <john.magnum@...hmail.com> wrote: 
>> On 11 Apr, 2013, at 15:19 , jfoug@....net wrote:
>>> Since you are still working on this, I would like to ask for these changes:
>>> 1.  change the wrapper #define.
>>> Currently  
>>> #ifdef __SSE2__
>>> Please change to:
>>> #include "arch.h"
>>> #ifdef MMX_COEF
>>> ...
>>> It does build just fine on VC, but not until the above changes.  VC does not define __SSE2__
>> 
>> If __SSE2__ is not defined, there are numerous of other uses of this macro (eg. in DES, XSHA512, Blake, john.c) that should be problematic (or result in suboptimal code). Maybe we should define it in arch.h when applicable and unless already defined? Then all a format like this needs is to include arch.h.
> 
> There are additional problems, so it may be a case by case.  In the new rawSHA512_ng.c, only the 64 bit VC can compile many of the intrinsic code.  

But the offenders are #ifdef __XOP__ or __SSSE3__, no? Or do you mean a 32-bit VC doesn't support all SSE2 intrinsics?

magnum

Powered by blists - more mailing lists

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.