Mill Computing, Inc. Forums The Mill Architecture 40/80-bit computing

  • Author
    Posts
  • jstopard
    Participant
    Post count: 3
    #4025 |

    What are some difficulties in implementing a fully 40/80-bit computer (i.e., not supporting 32-bit, 64-bit word sizes)?

    I realize that this could be trivial to do if one could be permitted to use word addressing instead of byte-level addressing in the computer; but, given that much of the code on github assumes that the architecture will use byte addressing, do you think it’s possible that a move could be made to this new type of architecture?

    The difficulty, I think, lies in the fact that there needs to be a correspondence between the cache size and word size (such that one is divisible by the other); the page size needs to be divisible by both, as well. Page sizes have to be power-of-2 based so that simple bit-wise indexing can be used to find the nearest page when doing TLB stuff.

    I have a vested interest in this, as I am working with audio/computer music/synthesis.

  • Art
    Participant
    Post count: 14

    This would have a severe impact on address computation for indexed loads.

    There are likely other problems with non-power of 2 sized words, but just the difficulty with indexed loads is sufficient to rule out 40/80 bit word sizes. Bit aligning and masking an index is trivial compared to the integer multiplication required to be part of the indexed load instruction with those word sizes.

You must be logged in to reply to this topic.