Complex Division Patch (reprise)

[craig@jcb-sc.com]

>> Here's a second version of the Complex Division patch
>
>This gets through the complex variable tests that I have available,
>but
>
>  1) It does poorly when -ffast-math is turned on.  Please do not
>  use fast-math for this switch.  Make up a new switch with a
>  different name.

What specifically do you mean "it does poorly"?  Also, what language are
you using?

>  2) The new and old methods do not always give identical results.
>  Please leave the old method as the default (for C language).
>  C9X will have some kind of pragma to control the behavior.

Unless you can establish that the old results were (more) correct than
the new, I don't think we will worry about not giving identical results.

(egcs/gcc is *nowhere* near to being the sort of compiler that ensures
identical FP results when the same code is compiled using different
versions of the compiler, especially on systems, like x86, with
extended-range FP registers.)

But, for the C language, I think it is reasonable to leave the old method
as the default for other reasons.  And I agree we need a distinct option
to control which method is used.

        tq vm, (burley)