(j3.2006) IEEE comparisons

[Cohen Malcolm]

Hi Robert,

I wrote:
> I am not adamantly opposed here; just that the costs seem to outweigh the
> likely benefits.

I will think further, but perhaps we should at least straw vote this (next 
meeting I mean, not via email!).

<<<
On a slightly different topic, ISO/IEC/IEEE 60559:2011
forbids replacing the expression

       .NOT. (X .LT. Y)

with

       X .GE. Y

(see the last paragraph of Subclause 5.11).
Subclause 7.1.5.4.2 allows a logical expression to
be evaluated as any other expression that is
logically equivalent.  I assume that that means
the optimization is allowed unless IEEE_SUPPORT_NAN
is true for X and Y.  Do we need to say something
about that?
>>>

Well, if IEEE_SUPPORT_NAN is true for X and Y, and IEEE_ARITHMETIC (or some 
IEEE_FEATURE thingy) is accessible, then .GE. is not the logical inverse of 
.LT. (in pseudo-operator form, in this case the opposite of "<" is "?>="). 
I don't think we need say anything in normative text, as this is clear from 
reading c14 and 60559, but a note would probably not hurt since reading all 
that is a bit of a chore.  OTOH the optimiser writers ought to know this 
stuff... and the ones who don't probably won't read the note anyway...

Cheers,
-- 
........................Malcolm Cohen, Nihon NAG, Tokyo.