> I'd like to tell valgrind not to worry about certain uninitialized memory > (and/or to stop Crypto++ from using uninitialized memory entirely): > > http://markmail.org/message/u44non7a3hgqefjt
Thanks, I'll take a look at this. > In a related topic, I'd like to have a deterministic RNG which doesn't > read from the time or any other non-reproducible inputs. You can use any keystream-based stream cipher as a deterministic RNG. For example Salsa20, or AES-CTR. They all implement the RandomNumberGenerator interface. > Oh, per our earlier discussion, I still want Poly1305, even though I like > VMAC a lot. It is okay with me if it is a non-optimized implementation > of Poly1305 for starters. I'm willing to contribute a patch for that, > too. Well, here's my response from the earlier discussion, which you didn't respond to: > I'd be interested if there is a way to add Poly1305 in a way that is > portable and maintainable for me without a huge amount of work, but it's > not clear how to do that. If you take a look at > http://cr.yp.to/mac/poly1305_athlon.s, it can't be compiled with MSVC, and > I can't even begin to understand it without learning a new language, > namely DJB's qhasm. (The last time I tried putting code into Crypto++ that > I didn't understand, i.e. the deflate compression code, it went pretty > badly in the long run and I ended up having to rewrite it.) > > From what I can tell from your graphs, on Pentium M Crypto++'s > VMAC-AES-128 has a higher per-message overhead than Poly1305-AES, but > their per-additional-byte cost is pretty close. I'm curious how many > messages you are planning to MAC per second, and why? If the per-message > overhead is really a serious problem for you, I can look into if anything > can be done to reduce it. My position is that adding an unoptimized Poly1305 doesn't seem useful, since it's not a standard, but adding and maintaining an optimized version doesn't seem worth the payoff (i.e. marginal performance improvement over VMAC on some old platforms). Is there some reason why you're so interested in this algorithm? BTW, looking at your old posts, I noticed one about Brainpool ECC curves. I'd be happy to take a patch for that if you want to submit one. --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the "Crypto++ Users" Google Group. To unsubscribe, send an email to [EMAIL PROTECTED] More information about Crypto++ and this group is available at http://www.cryptopp.com. -~----------~----~----~----~------~----~------~--~---
