On Thu, May 13, 2010 at 20:00, Julian Foad <julian.f...@wandisco.com> wrote: >... > The fast_memcpy() optimization raised an eyebrow: one developer (Greg > Stein) mentioned to me that optimizing standard memcpy felt wrong in > principle. I don't know what was behind his feeling. I guess one kind > of concern that could be considered against it is that newer compilers > and hardware in a few years could possibly negate the optimization, yet
Not even in a few years. I figured it to be "now". But with that said, Stefan makes a very good point: the memcpy() that compilers will use are optimized for a different workload than we are using here. Thus, it seems fair to craft one tuned for *this* workload. >... The couple things that stood out in my quick look at the patch have already been responded to by Julian, so I'll just wait for the thing to be committed and do a review then. It seems that, with Julian's input and re-roll of the patch, it is ready for commit. Cheers, -g
