It would be wonderful if someone else would create a super-concise explanation of unums! Thank you, thank you! I can't seem to do it. I must be getting long-winded in my old age.
I looked a https://github.com/tbreloff/Unums.jl and saw that it erroneously says the sign bit field can have flexible size. Nope, that's always a single bit. Otherwise, it looks like the web site is on the right track. Expand the variable size fields up to some maximum (like the example in the book for a 64-bit fixed size unum) but still follow the rules of correct unum math within that bound. This is likely how the first hardware versions will do things, also, and they will save power by knowing which bits are not in use. One thing any language implementation will want is to expand a packed set of unums (using the utag as a linked list pointer) into unpacked ones of fixed size, and the reverse operation. That way, block stores and loads can be optimized. I did not put that in the Mathematica reference implementation, but it needs to be there. On Wednesday, July 29, 2015 at 3:33:52 PM UTC-7, Job van der Zwan wrote: > > BTW, Tom, I was already working on a summary of the book (on an IJulia > notebook). I'm on mobile right now so don't have access to it, but I can > share it later. I think something like that might be useful to attract more > collaborators - we can't expect everyone to read it.
