FORTRAN floating point accuracy problems

[HP67]

(04-03-2016 05:21 PM)Dieter Wrote:  

(04-03-2016 05:17 PM)HP67 Wrote:  Again, I am only carrying 6 decimal places of expected results although I am displaying 7.

This means that "differences" in the 7th place actually are not present. ;-)

Yes I know. Therefore I was not troubled by the diff values ;-)

(04-03-2016 05:21 PM)Dieter Wrote:  Maybe you can update the "expect" values to seven decimals.

I'll have to look at it. It could be with 7 decimal places I can't fit 4 values on one line of FORTRAN code...

(04-03-2016 05:17 PM)HP67 Wrote:  Thanks, I'll try your suggestions. But first I'm going to try to write an assembler subroutine for DINT. Then I can use your rounding routine which would really be great.

(04-03-2016 05:21 PM)Dieter Wrote:  Hmmm... I would love to know whether the suggested rounding fixes the issue. Maybe you can add these little changes and report the results... ?-)

The good news is your rounding routine works perfectly up to 15 decimal places (possibly more, I will check) with my brand new shiny double precision (8 byte) DINT routine in assembler. I didn't have time to go back to our old versions of the driver code to see if it works better. I'll get back to that as soon as I can.

Thanks for the help! I am really excited about having the rounding routine in my FORTRAN IV toolbag. I have wanted one for a long time and just couldn't figure out how to do it. I do have some rounding code I use for assembler control blocks but I just couldn't extrapolate it to real numbers.