Ray, A detailed and complete explanation as always, thanks
I think I've finally understood it ! IEEE standards say zero is 2.91E-11 , which as a fraction is roughly 1 / pwr(2,35) So, the binary representation of IEEE zero has an exponent of -35 If you bias this by 1023 you get 988 or 001111011100 This is the leftmost 12 bits of hex 0x3dc00000 (UV default wide zero) By IEEE rules if the exponent of the difference (expressed in binary and biased by 1023) is less than 0x3dc00000 then the difference is effectively zero. By varying "wide zero" in Universe you can change this, in my case to 0x3de00000 (1023 - 33) and in Bruce's case to 0x3eb00000 (1023 - 20) Each time you add 1 into the leftmost 12 bits of wide zero you are increasing the value of the exponent by 1 and multiplying the value of "effective zero" by 2 If my maths is correct (which would be a miracle) 0x3eb00000 or 1023-20 means 2.91E-11 multiplied by pwr(2,15) , or approx 0.000001 So, on a system with a wide-zero value of 0x3eb00000 any difference of more than 0.000001 is significant, any difference of less than 0.0000001 is not significant, and the boundary is somewhere in between. ... and I've just run a program to demonstrate it !! Precision = 8 Wide zero is 0x3eb00000 A difference of 1 times 10 to the power nn is : nn = -4:- Significant nn = -5:- Significant nn = -6:- Significant nn = -7:- Not significant nn = -8:- Not significant Time for a coffee I think, before I change my client's uvconfig file Thanks again Ray, Bruce and Adrian - what a wonderful list... -----Original Message----- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] Behalf Of Ray Wurlod Sent: 22 September 2004 11:33 To: [EMAIL PROTECTED] Subject: RE: [U2] [UV] - Wide Zero This is slightly adapted from an old UniVerse Technical Bulletin (about WIDE0) ------- u2-users mailing list [EMAIL PROTECTED] To unsubscribe please visit http://listserver.u2ug.org/
