(09-19-2022, 01:48 AM)Jack Wrote: Pete, why did you give up on numeric computations?
it took me a long time, but dogged determination brought me success, writing math routines from scratch can be very hard to debug and to get working because of many dependent routines.
just for fun I am continuing to develop this 32-byte decfloat package but I decided to change from 8 digits per element to 9, which took me a while to get working right, this way you get 63 digits but some operations tend to loose accuracy so the the print routine is restricted to 55 digits with round up, I got the basic 4 arithmetic and some helper functions done
@Jack
Keep in mind I started migrating to string math before some of the better underscore math types were built for QB64. When I took a second look, and tried the larger data types, I got some stuff working with numeric computation but sure enough, when I ramped up the size of the the computations enough, fail.
With string math very large computations involving addition, subtraction, multiplication, non-decimal powers, square roots, and percentages are done remarkably well and now, fast.
What I'm finding a problem with is general roots (includes decimal roots) and decimal powers. With solutions involving approximation routines, especially those which rely on decimal factors in the equation like .333, .666, etc. well, errors just keep getting compounded. There is way too much irony involved with having a math system that can print out 10,000 digits in a decent amount of time but, in some instances, every one of those digits could be off!
So what is needed with any of the mat routines is fraction manipulations with long division when needed. That seems quite difficult to develop for general roots and decimal powers.
Pete