
Boost : 
From: Paul A Bristow (pbristow_at_[hidden])
Date: 20051202 13:11:39
 Original Message
 From: boostbounces_at_[hidden]
 [mailto:boostbounces_at_[hidden]] On Behalf Of John Maddock
 Sent: 02 December 2005 17:25
 To: boost_at_[hidden]
 Subject: Re: [boost] [math] floating point classification 
 testinghelpwanted

 Actually if that's correct then things are a lot better than
 I feared: the
 text implies that the type behaves as a real type with
 106bit's in the
 significand. It also implies that numeric_limits<>::epsilon
 must be buggy
 for that platform. So I'm more confused than ever :(
If you want to be confused even further, attached is my inexpert stab at
numeric_limits for NTL quad_float which I think may be the same.
But I have asked the package author for his advice and help on this, but so
far no reply (though there are regular package updates).
I'm not sure that epsilon has a exactly defined value but I concluded
static quad_float epsilon() throw() { return
2.4651903288156618919116517665087069677288E32;}; // 2^(1106) ==
2^105
was my best guess.
But without confirmation of this, in 'testing' NTL math functions, I cheated
using the number of decimal digits
quad_float tolpc = to_quad_float("1e30"); // 31 guaranteed decimal
digits, so percent tolerance is about 1e29
BOOST_CHECK_CLOSE(x, one, tolpc);
But I'm ready (anxious even) to have someone agree, or correct me.
Paul
 Paul A Bristow Prizet Farmhouse, Kendal, Cumbria UK LA8 8AB Phone and SMS text +44 1539 561830, Mobile and SMS text +44 7714 330204 mailto: pbristow_at_[hidden] http://www.hetp.unet.com/index.html http://www.hetp.unet.com/Paul%20A%20Bristow%20info.html
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk