Re: REAL

[Georg Bauhaus <rm.dash-bauhaus@futureapps.de>]

On 14.03.12 13:59, Dmitry A. Kazakov wrote:
> On Wed, 14 Mar 2012 12:14:05 +0100, stefan-lucks@see-the.signature wrote:
> 
>> Hardly anyone cares enough to analyze how the error grows from operation 
>> to operation. Hardly anyone even bothers to define how close the result 
>> should to be -- which depends on the application at hand, of course. 
> 
> It is largely the hardware to blame. When the machine rounds, it is more
> difficult to keep track on precision, comparing to a truncating machine (in
> either direction).
> 
> Of course, interval arithmetic should have replaced IEEE long ago. But
> nobody cares.

Programmers do care about floating point effects---when they
strike: but being able to explain possible causes of FPT "effects"
is an asset to human capital. So it is better to keep the current
types and effects and acquire the asset. Turns hazards into
sources of achievement. "So, you handled that effect? Good job!"

If interval arithmetic is a superior alternative, then
interval arithmetic popular in some way, but also make sure
that programmers can substitute assets! Seems to be one way
of getting technical changes adopted. Maybe write some nifty shader.