5th IEEE op

[Stu Anderson 5-3595]

> If the accumulation of floating-point numbers and of products of
> such numbers into a fixed-point register is introduced
> as a fifth floating-point operation, many unnecessary
> errors that appear in floating-point computations can be avoided.

There is a book based on extensions of the above idea.  The first article is 
of interest. 

    G. Bohlender [1990].  What Do We Need Beyond IEEE Arithmetic?  in Computer 
    Arithmetic and Self-Validating Numerical Methods (edited by Christian 
    Ullrich), Academic Press.

I personally have little need for such accumulation since I am more concerned 
about error in the terms than I am about their sum.  In addition, I am more 
concerned about poor implementation of the IEEE standard than I am about its 
extension.  <Pun accidental, but kept deliberately.> 

--Stu Anderson (Boeing Computer Services, slaaespresso.boeing.com)