Date: Sat, 7 May 2011 11:33:25 -0500 From: "JimF" <jfoug@....net> To: <john-dev@...ts.openwall.com> Subject: Re: patches to integrate into jumbo I fully agree with both of these ----- Original Message ----- From: "magnum" <rawsmooth@...dband.net> >I really want the UTF-8 stuff into a Jumbo because it includes stuff that >others can use for supporting UTF-8 in the future. Right now, md5_gen has a 'trival' unicode support. It would be nice to be able to keep the trival, but also add utf8->unicode support. I currently only have 7-bit ASCII to unicode. Simply converting bytes into words. > The intrinsics patches are so important for performance that they should > IMO trigger a new Jumbo themselves, whenever they are ready! I agree with this one also. I can not use intrisics (well, I 'can' build with them, but 32 bit gcc sux), but for people that do have 64 bit systems, this is a HUGE speedup, almost as much as going from x86 to original SSE2 in asm is on the 32 bit builds. I am not sure whether these are prime time 'ready' for a J-2. However, isn't the even builds, builds that add new stuff, and the odds usually are bug fix/performance fix releases? I do not believe that intrisics as writting (other than in md5_gen) are 'ready'. I am pretty sure that intrisics replace most of the formats they use, vs working with them IIRC. As for utf8 to unicode, it can probably be done without ifdef's by making changes in the init() functions of the formats, to change the proper function pointers (and possibly testing vectors, when the command line lists utf8). I am pretty sure that is what Magnum did in the end, and if not, it can readly be done. In that way, a single build (without having to do a 'special' uft8) would properly give you all the usablity you need, based upon command line driven switches. Jim.
Powered by blists - more mailing lists
Powered by Openwall GNU/*/Linux - Powered by OpenVZ