1. 19

  2. 4

    “ your architecture should allow to run HLL code much faster than a compiler emitting something like RISC instructions, without significant physical size penalties”

    My reason is more about running reliably and/or securely even if there is a penalty. I’m curious if it could speed things up, though. To support the possibility, Intel’s is already a higher-level architecture than the RISC-like instructions the micro-architecture uses. Intel’s is also one of the highest-performance implementations. Perhaps a full-custom design of a HLL-centric processor could similarly boost things. I don’t know, though. I will address two of these.

    “JWZ’s Lisp-can-be-efficient-on-stock-hardware claim isn’t much better than Smalltalk-can-be-efficient-on-custom-hardware, I find. Just how can it be?”

    Something like this that achieves lower, performance penalty than the 25% author allows for.

    “There are various other kinds of computers, such as convenient realizations of neural networks or cellular automata, but they’re nowhere as popular either, at least not yet””

    Deep, neural networks got super popular. Then, engineers figured out you could ignore a lot about how modern chips were designed, esp precision, when implementing them. Also, that analog implemented them really well. Quite a diverse array of custom architectures making them faster, use less energy, etc. Some are analog/digital hybrids. Most common deployment involves regular CPU’s with GPU’s since they’re commodities. Hardware implementation on cutting-edge nodes, especially analog, has high cost.

    1. 2

      My reason is more about running reliably and/or securely even if there is a penalty.

      I think this too. I’d rather work on an architecture that makes compilation simpler and, more importantly, memory management safer.

    2. 1
      1. 3

        ARM has Jazelle for this purpose too, although I don’t think it’s very popular.

        1. 1

          There’s also Java processors like aJile and JOP. JOP’s comparison page lists more. I’m not sure how often they’re used. I can see the benefits of some, though.

        2. 1

          Yeah, but I am pretty sure that it is dead because sitting to regular arm code was faster.