Performance craft is back in fashion

A cluster of threads is about squeezing machines. A write-up raced C against C# on VLC's Gaussian blur, and commenters were struck that AOT C# lands close to C. Richard Feldman's piece on how close pure functional programming can get to C drew debate about copy elision and memory bandwidth. Another article covered running a city builder on 16 MHz. And "Twelve years of distributed SQLite" prompted a user to ask what to move to from dqlite, with rqlite mentioned as a simpler alternative to heavy clustered databases.

The through-line is a taste for simple, fast and small. Commenters keep asking why the original VLC filter doesn't use SIMD, and why we spend money on modern CPUs whose features sit idle. SQLite-based clustering appeals because it avoids the complexity of big database setups.

Some pushback: one commenter noted libdeflate produces similar output size to zstd but may take far longer, a reminder that benchmarks depend on what you compare.