On 24 June 2010 15:03, Sergey Bochkanov <sergey.bochka...@alglib.net> wrote: > Hello, Bill. > > You wrote 24 июня 2010 г., 17:20:17: >> How would you deal with say a 6x6 limb multiply (where a limb is by >> the old definition). How would thid maintain performance for such >> sizes? > > You are right. Now I see that it will work only for operations where > both operands are large. Tricks which work in linear algebra won't > work in bignum. Well, anyone can make a mistake. > > BTW, did you tried to contact LLVM team with a question regarding > better carry support? May be it is simpler than we think...
Actually, I did. It's a fair bit of work, and not something they currently have a developer that could work on it. But I'm pretty certain they'd let us implement it ourselves and contribute it. But it requires making patches at quite a few levels of their project. I think we figured out we wouldn't have to change the parser, but we'd have to change all the "back ends" (its kinda funny that LLVM as a back end has itself a parser and a back end). It would be a major project for which we'd probably need to get grant funding. It's certainly not something simple. But I haven't ruled out applying for a grant to research some ideas which would cover this stuff. I would need to do that jointly with someone in Computer Science here in the UK. So I'm still educating myself about the CS literature, because ultimately you need to convince referees in CS not maths, that it is a good idea to have such a project. For example I'm very interested in Jason Moxham's superoptimiser work, but the CS literature actually already has a fair bit about superoptimisers. Quite what a superoptimiser for a backend is and whether it would be useful for us, I am not yet certain. Still more to learn.... Bill. > > -- > With best regards, > Sergey mailto:sergey.bochka...@alglib.net > > -- > You received this message because you are subscribed to the Google Groups > "mpir-devel" group. > To post to this group, send email to mpir-de...@googlegroups.com. > To unsubscribe from this group, send email to > mpir-devel+unsubscr...@googlegroups.com. > For more options, visit this group at > http://groups.google.com/group/mpir-devel?hl=en. > > -- You received this message because you are subscribed to the Google Groups "mpir-devel" group. To post to this group, send email to mpir-de...@googlegroups.com. To unsubscribe from this group, send email to mpir-devel+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/mpir-devel?hl=en.
