ni...@lysator.liu.se (Niels Möller) writes: > I haven't paid much attention to x86_64 assembly for sha1 and md5 (there > is x86_64 assembly for sha1, which is a direct translation of an earlier > version of the x86 assembly, and there's no md5-assembly at all for > x86_64).
I just checked in a some basic md5 assembly for x86_64, pretty similar to the x86 code. Gave a speedup of 15% on intel and 35% on amd (testing on my closest machines). Still slower than openssl, but by a smaller margin, about 15%. And on the amd machine, nettle's sha1 seems to be 20% faster than openssl. I don't think I had really thought about that missing md5 assembly, which was a very low hanging fruit. Anyway, I'm not totally fanatic about performance in Nettle. When there's a tradeoff between performance and readability of the code, I don't always choose performance. I think it can be good enough even if it's not fastest for every single algorithm. Regards, /Niels -- Niels Möller. PGP-encrypted email is preferred. Keyid C0B98E26. Internet email is subject to wholesale government surveillance.
