It's a pity that this has the top slot because it provides no technical information about what optimizations he made and it doesn't have any benchmark information.
It's just "Hey, look at me, I'm Linus and I optimized some C code to make it fast". I would much rather see him give us some insight into optimization strategies for different architectures. He appears to know about that, but not share it.
Think of lightweight articles about SHA1 rewrites in order to pull the requirements of the OpenSSL libraries out of GIT as the HN version of idle gossip.
Not every article has to be serious, and of foundational importance. Some sites have pictures of funny animals. This one has Kernel Developer insight into their OCD desires to defeat the compiler from messing up their code...
He's not rolling out his own libcrypt. He's fixing a bug: i.e. getting rid of the unnecessary dependency on a huge library (OpenSSH) just because of a hashing function. This is systems programming, not a brainless enterprise data pumping: you get rid of unnecessary dependencies if you want to run on cell phones and blade servers at the same time.
If all he wanted was to dump a dependency, it would have been easier to cut-and-paste a slightly slower implementation from another library. I think he's saying he enjoyed it. The benchmarks are just after-the-fact justification.
This provides a valuable insight into Linux culture. There have been many cases where the micro-optimization obsession has displaced effort from the "big picture".
It's just "Hey, look at me, I'm Linus and I optimized some C code to make it fast". I would much rather see him give us some insight into optimization strategies for different architectures. He appears to know about that, but not share it.
Oddly, he was talking about how the Mozilla SHA1 code was faster than OpenSSL back in 2005: http://www.mail-archive.com/git@vger.kernel.org/msg01138.htm...