[v3] Possible fix for PR 5730

[Gabriel Dos Reis]

Phil Edwards <phil@jaj.com> writes:

| On Fri, Mar 21, 2003 at 12:24:17AM +0100, Gabriel Dos Reis wrote:
| > Phil Edwards <phil@jaj.com> writes:
| > | 
| > | That's debatable, but I don't feel as strongly about it as you do.
| > 
| > Installation and compiler specific flags are put in c++config.  I don't
| > see why this particular one  has to be put somewhere else.
| 
| As I said, I don't feel strongly about it.
| 
| Do you have any thoughts on Jerry's question; why/how the distinction is
| causing such a slowdown?

Jerry asked:

  Can someone that actually understands floating point explain the accuracy
  improvement with the current implementation?  Mathematically (ignoring fp
  limitations), they are equivalent.  So how does the distinction show up with
  actual fp math?

The distinction happens on border cases.  I recall I came across
values where the slower version gave slightly better gradual overflow
(radix was 16 I believe).  But here, I don't  have the numbers rigt
with me.  I seem to recall the archive should have some record of
discussions on this topic.

-- Gaby