Has anyone implemented John Gustafson's Unum numeric format in Delphi? I have C implementation of it from a research paper but was hoping someone may have done this before I take off on it.
This debate between Gustafson (of unums) and Kahan (of IEEE-754) is interesting. Posted on Gustafson's website, and I'm not sure what to make of the promotional tone of much of his writing. Read with a pinch of salt. PDF. johngustafson.net - johngustafson.net/pdfs/DebateTranscription.pdf
Going back decades, Kahan is well known for being a critic of things he didnt do himself. He most often doesnt want to talk about the errors that the Intel trigonometric functions will produce for arguments that are scaled modulo 2pi, which is his doings as well.
There is a basic implementation of Unum in Julia and it is slow but accurate.
Jeroen Wiert Pluimers I've changed my view since I posted that in 2016 - I actually feel much more convinced about unums/posits now.
ReplyDeleteThis debate between Gustafson (of unums) and Kahan (of IEEE-754) is interesting. Posted on Gustafson's website, and I'm not sure what to make of the promotional tone of much of his writing. Read with a pinch of salt. PDF. johngustafson.net - johngustafson.net/pdfs/DebateTranscription.pdf
ReplyDeleteHere is more in this topic posithub.org - Unum & Posit- Next Generation Arithmetic
ReplyDeleteGoing back decades, Kahan is well known for being a critic of things he didnt do himself. He most often doesnt want to talk about the errors that the Intel trigonometric functions will produce for arguments that are scaled modulo 2pi, which is his doings as well.
There is a basic implementation of Unum in Julia and it is slow but accurate.