Re: decimal floating point

[Dan Zuras <dan@xxxxxxxxx>]

> From: "Ivan Godard" <igodard@xxxxxxxxxxx>
> To: <stds-754@xxxxxxxx>
> Subject: decimal floating point
> Date: Mon, 24 Mar 2003 15:38:39 -0800
> 
> 
> . . .
> 
> Secondly, what is the interpretation of an operation that is applied to
> mixed formats? In particular, assuming there are interformat conversion
> operations in hardware, what are the promotion rules?
> 
> . . .
> 
> Or have you looked at mixed formats already and considered them hopeless?
> 
> Your help and advice appreciated.
> 
> Ivan Godard

	Wow.  In something like a year of discussing this extension,
	you are the first to raise this issue.  Amazing.

	I guess that, while the text of the standard states that an
	implementation may support either or both of decimal & binary,
	everyone involved has been considering one or the other but
	not both.

	This issue, of course, only comes up in a language that
	supports both.

	Blindly, one might calculate that the precision ordering is:

	decimal32 < binary32 < binary64 < decimal64 < decimal128 < binary128

	But this would get you into trouble as the middle relation
	(binary64 < decimal64) is an illusion in the sense that most
	binary64 numbers are more densely packed that most decimal64
	numbers due to the large "jaggy" introduced by decimal jumps.

	Also, the decimal types would come out on top WRT range so
	overflows are a problem.

	Not to mention calculations that are exact in one base & not
	in the other.

	I fear you have opened a rather large can of worms, my friend.

	We're gonna need a bigger can...


					Dan