I've been reading about a new floating-ponit number format unum that seems to be gaining traction in the HPC world that could eventually superceed IEEE. Fortran is usually very slow about adopting new ideas but this does look to be a bread-and-buter issue for it so I was wondering if there are any plans to support (through either Intel or the Fortran Standard). Thanks.
An interesting article with a provocative title (The End of Error). I skimmed through the article, and it occurred to me that implementing the idea could need/benefit-from machine instructions to allow bit addressing and some architecture changes to avoid the misaligned operand penalties that are common now. Without such machine level support, the unum idea may only reach a status similar to that of multiple-precision arithmetic packages.
First I have heard of this. Fortran is all about performance and portability - unless IEEE adopts this format and multiple processor vendors support it, it's unlikely in the extreme that Fortran vendors would support it. The standard doesn't need to do anything here - it would just be another REAL kind in an implementation. The bigger issue is data interchange and interoperability.