dgh@validgh.com (David G. Hough on validgh) (06/13/91)
Several people have commented that IBM is now experimenting with logarithmic representations for floating-point numbers. This is not new; people have been doing that for a long time. The proceedings of any IEEE Symposium on Computer Arithmetic will have several papers on this subject. The next symposium is later this month in France. Nobody has found any worthwhile advantage for logarithmic numbers for general- purpose computation, but I have heard that they have been used in some special-purpose embedded systems. X3J11 adopted a conventional floating-point number representation as part of its standard for ANSI-C. I pointed out that there was no way that <float.h> could be meaningfully defined on a system with logarithmic numbers, and the committee's response was basically that was OK, those systems just wouldn't support ANSI-C. -- David Hough dgh@validgh.com uunet!validgh!dgh na.hough@na-net.ornl.gov