[Cfp-interest] expression transformations for decimal FP

[Jim Thomas]

Here's a shot at words for Part 2 to address the optimization issue of preserving quantum exponents for decimal FP.

-Jim
=================

In F.9.2, insert at the beginning:

[1a] Valid expression transformations must preserve values in all cases.

[1b] The validity of the transformations noted below applies to expressions of generic floating types.

[1c] …

In F.9.2, append:

[2] For expressions of decimal floating types, transformations must preserve quantum exponents, as well as numerical, infinity, and NaN values (5.4.2.2).

[3] EXAMPLE:  1. × x -> x is valid for decimal floating-point expressions x, but 1.0 × x -> x is not:

1. × 12.34 = (1, 1, 0) x (1, 1234, -2) = (1, 1234, -4) = 12.34

1.0  × 12.34 = (1, 10, -1) x (1, 1234, -2) = (1, 12340, -5) = 12.340

The results are numerically equal, but have different quantum exponents, hence have different values.

 

 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.oakapple.net/pipermail/cfp-interest/attachments/20121214/71cff4d8/attachment-0001.html