Hacker News Reader

Even without vector intrinsics, you can get a lot of mileage out of spending almost all your time in loop bodies that map primitive operations over large arrays. Interpretive overhead tends to grow a lot slower than the user program's input size.

IIRC J didn't use intrinsics until maybe two years ago. They used mostly C and everything was built with for-loops. Some special verbs or phrases were however implemented in x86 assembly.

Go assembler can get to vector instructions, for instance as used by Cloudflare to implement crypto speedups: https://blog.cloudflare.com/go-crypto-bridging-the-performan...

Indeed but without intrinsics, functions using those vector instructions cannot be inlined, and the go calling convention's overhead is still big.

I think part of the speed of APL is the way that parts of it are small enough to fit in L2 cache (at least 1 of the APL descended languages, Q / K, bragged about this). Whether the Go implementation seen here will fit in typical caches of today's CPUs I don't know.