It's a quote from David Goldberg at the cited URL. Not my idea.
= = =
Loren P Meissner
[log in to unmask]
> -----Original Message-----
> From: [log in to unmask] [mailto:[log in to unmask]]
> Sent: Friday, March 19, 1999 9:01 AM
> To: [log in to unmask]; [log in to unmask]; [log in to unmask]
> Cc: [log in to unmask]; [log in to unmask]
> Subject: RE: Pierre's Problem [-> Kahan summation formula]
>
>
> I disagree strongly with this. The whole purpose of allowing a
> processor to evaluate "mathematically equivalent" expressions
> is to optimize floating point expressions.
>
> It is the programmer who must be careful (using parentheses, for
> example) to prevent evaluation of a different expression.
>
> One can argue about whether this is the correct philosophy for
> Fortran (I think it is, and it has been always), but given the
> way the language is designed, optimizers should be very agressive
> (when asked to be).
>
> [Loren wrote]:
>
> >>>>
> An optimizer that believed floating-point arithmetic obeyed
> the laws of
> algebra would conclude that C = [T-S] - Y = [(S+Y)-S] - Y =
> 0, rendering
> the algorithm completely useless . . . optimizers should be extremely
> cautious when applying algebraic identities that hold for the
> mathematical real numbers to expressions involving floating-point
> variables.
> >>>>
>
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|