Re: IEEE 754 compliance
There are some resources for testing by Nelson Beebe at http://www.math.utah.edu/~beebe/software/ieee/timops.html From what I understand, the main issues are handling of edge effects (underflow, overflow, divide by 0, etc.) where compilers may do some things different from the standard's prescription. There is, of course, an interaction here with hardware that may not provide ways to get at the bits (literally). In my efforts to set up some Gnumeric test worksheets, I've tried to contact Beebe without success. He may have retired (I believe he is older than I, and I'm on the brink of retirement from teaching, but not from Gnumeric!) If there is interest, and in particular an example where IEEE754 may be important, I'll be happy to dig a bit. I was a corresponding (ie vote by mail) member of the IEEE 754 committee back in late 70s. Given the arcane detail, it will take a bit of review for me to get fully up to speed. JN ___ gnumeric-list mailing list gnumeric-list@gnome.org http://mail.gnome.org/mailman/listinfo/gnumeric-list
Re: IEEE 754 compliance
The 754 standard is being updated and the result is being voted on now, I think. It's called 754r and can be found on the web. 16-bit and 128-bit floating point formats are what led me to it. Dave Feustel On Tue, May 13, 2008 at 08:50:45AM -0400, Prof J C Nash wrote: There are some resources for testing by Nelson Beebe at http://www.math.utah.edu/~beebe/software/ieee/timops.html From what I understand, the main issues are handling of edge effects (underflow, overflow, divide by 0, etc.) where compilers may do some things different from the standard's prescription. There is, of course, an interaction here with hardware that may not provide ways to get at the bits (literally). In my efforts to set up some Gnumeric test worksheets, I've tried to contact Beebe without success. He may have retired (I believe he is older than I, and I'm on the brink of retirement from teaching, but not from Gnumeric!) If there is interest, and in particular an example where IEEE754 may be important, I'll be happy to dig a bit. I was a corresponding (ie vote by mail) member of the IEEE 754 committee back in late 70s. Given the arcane detail, it will take a bit of review for me to get fully up to speed. JN ___ gnumeric-list mailing list gnumeric-list@gnome.org http://mail.gnome.org/mailman/listinfo/gnumeric-list ___ gnumeric-list mailing list gnumeric-list@gnome.org http://mail.gnome.org/mailman/listinfo/gnumeric-list
Re: IEEE 754 compliance
Gnumeric does not let you access NaN etc. It would interfere with the desired semantics. Morten ___ gnumeric-list mailing list gnumeric-list@gnome.org http://mail.gnome.org/mailman/listinfo/gnumeric-list
Re: IEEE 754 compliance
The complexities of the edge effects are best kept out of the spreadsheet, as Morten indicates. However, there are some computations that might be influenced by how a particular internal calculation is performed. I was earlier looking at the ends of the Gaussian (normal) distribution where one gets some weirdness in Excel. This could be because very small numbers are handled poorly. It is in the special functions etc. that I would think Gnumeric and other spreadsheets are most likely to be interested in IEEE 754 and its revision. Must admit I was unaware of 754r activity. Thanks Dave. For info, there's a nice Wikipedia item at http://en.wikipedia.org/wiki/IEEE_754r with a link to a recent essay by Velvel Kahan (who was really the person who got all the floating point stuff going, and in the 1960s pretty well embarassed IBM into retrofitting the 360 with guard digits for floating point) that gives a nice and nasty example using Excel. I've tried this in Gnumeric and get somewhat different results, but which I'm sure would upset novice users. JN Morten Welinder wrote: Gnumeric does not let you access NaN etc. It would interfere with the desired semantics. Morten ___ gnumeric-list mailing list gnumeric-list@gnome.org http://mail.gnome.org/mailman/listinfo/gnumeric-list