Author Topic: Floating point Arithmetic precision  (Read 317 times)


  • New Member
  • *
  • Posts: 22
Floating point Arithmetic precision
« on: March 26, 2021, 11:52:31 am »
Reply to @MarkMLI :
This is in fact a "real" question on the "mathematical functions" in Fpc-PASCAL. When you want to uses ArcTan2 for real calculation with a good precision for instance, we need to have more a 10**(-16) precision, to do it (numerical recipes in PASCAL) you have to tackkle many sources (as we did in Fortran) to be confident in the calculation method. The complex numbers formulations for Exp/Log need a "fine tuning" of the interval definitions for the trigonometric functions. Some functions like "de haar wavelets" or "Lanczos matrix calculations" can "explode" on the lake of accuracy, so we need the source with the unit ppu. Friendl. Yves .
« Last Edit: March 26, 2021, 05:03:18 pm by marcov »


  • Global Moderator
  • Hero Member
  • *****
  • Posts: 9437
  • FPC developer.
Re: Floating point Arithmetic precision
« Reply #1 on: March 26, 2021, 05:03:48 pm »
(I've split off the post into a separate thread, Mark you might want to revise or remove your post)


TinyPortal © 2005-2018