Comment Re:It's a question that WAS relevant (Score 1) 161
Back in the 1970s I worked at a computer manufacturer, writing code for their product's instruction set in assembler. The computers were designed and built around AMD2901 bit slices. The hardware guys implemented the instruction sets using microcode and, as the computers got bigger and more complicated some of the instructions got so elaborate that programmers found ways to do an operation faster using a few simpler instructions instead of one complicated one.
Nowadays, with the kind of speedups from using cache memory, branch prediction, and so on, I reckon it could be a whole different ballgame. I suspect though, that proving correctness might become the most important criteria, and simpler would make proving correctness easier.