Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

Re: Empty interval representations (general)



> Date: Wed, 05 May 2010 10:12:56 -0500
> From: "R. Baker Kearfott" <rbk@xxxxxxxxxxxxx>
> To: John Pryce <j.d.pryce@xxxxxxxxxxxx>
> CC: P1788 <stds-1788@xxxxxxxxxxxxxxxxx>
> Subject: Re: Empty interval representations (general)
> 
> That's an interesting stance since, I think (correct me if
> I am wrong), that, except for "big-endian" and "small-endian",
> 754 does specify formats.  Of course, an argument for
> not specifying is:  "it does not matter what's inside
> if what is actually seen is as specified."
> 
> Baker

	Actually, John turns out to be correct.  And,
	you too, in a way.  But John has the intent
	correct.

	While 754 DOES specify formats, it is only the
	interchange formats.  How a system chooses to
	represent its numbers for itself is not of any
	concern to 754.  Well, 754-2008, anyway.

	What IS specified down to the last element is
	the set of numbers that are represented in each
	floating-point datatype, whether interchange or
	not.

	So if you say you are doing arithmetic in, say,
	Binary64, everyone knows exactly what numbers
	you are talking about.  And Binary64 also has
	an interchange format so there are bit patterns
	available to allow you to give your numbers to
	another system.

	But you could just as easily say you are doing
	arithmetic in a Binary65 & we would STILL know
	what your numbers are.  (Well, it needs to be
	specified a bit more than that but its all in
	the document.)  Its just that in this case there
	is no interchange format defined.

	As for big-endian versus little-endian, even back
	in 1985 that was solved by another data interchange
	standard.  (Was it 1596.5?  I forget.)  So 754
	basically punts that problem by appealing to the
	other standard by reference.

	I think John is correct in that 1788 can pretty
	much do the same thing.

	More to the point, I think we should.


				Dan


	P.S. - This is different from but related to the
	problem of mid-rad versus inf-sup.  If they both
	specified exactly the same set of intervals we
	could ignore the implementer's choice & move on.
	But they do not have the same set & we must deal
	with that.  The 6.1 text takes the first step in
	coercing mid-rad into inf-sup but it does not
	solve the problem.  I'm not sure what the solution
	is or even if we SHOULD coerce them to be the same.
	I guess I'm hoping something will come up in our
	discussions.

> 
> On 5/5/2010 4:34 AM, John Pryce wrote:
> > Nate, Dan, everyone
> >
> > On 4 May 2010, at 14:56, Nate Hayes wrote:
> >> I still partly think P1788 only needs to specify bit-encodings for
> >> interchange formats and not internal representations of various conforming
> >> implementations.
> >
> > This remains my default position and I don't feel Dan or anyone else has made a convincing case for doing otherwise.
> >
> > John
> >
>