I was there at the time and could say a lot of things about the CRISP, but will restrict myself to three observations:
-
I was there at the time and could say a lot of things about the CRISP, but will restrict myself to three observations:
1. There was this thinking that the architecture meant the compiler could do a better job, but that was putting the cart before the horse. Better compilers were a better idea.
2. The stack architecture put too many gates on the performance-critical traces in the CPU.
3. I am the photographer of those photos at the bottom of the die.AT&T’s CRISP Hobbits
An unexpected journey for AT&T with it's own low power processor
(thechipletter.substack.com)
-
@robpike Re (2), a few years back I stumbled upon this 2002 paper: https://citeseerx.ist.psu.edu/document?repid=rep1&type=pdf&doi=0366820530b662bd1c3a912720ce23795862d1ba, which describes a stack machine microarchitecture with out-of-order execution, achieving ILP comparable to register machines ([*] in mathematical modeling, didn't make it to silicon afaict).
This was in the heyday of getting java to run on the metal, but the paper's conclusion points to how things actually panned out: you can have any microarchitecture you want, as long as the frontend's x86_64.