da Silva, Joe wrote:
Emil Jerabek wrote:
Well, this one is a little longer than the previous ones, so should I use it?
[Joe da Silva]
Probably. It looks more reliable than the previous version.
The only concern I still have, is that the test program, emil9, does not test for accuracy with large "arctan" arguments (only up to 1e17). Since "tan" can produce huge results even with just real arguments, an "arctan" function must conversely provide accurate results with huge arguments.
Indeed. So for all checks containing A (1e17) I'm adding similar tests with B=MaxReal (except those whose correct results contain B/2 or Sqr(B) in which cases I'm using MaxReal/2 and SqRt(MaxReal), resp.). Alright? In fact, both the last and the previous version of Complex_ArcTan passes all these tests on IA32 (tested on Linux and DJGPP).
One platform where I'm getting overflows with the previous version (even without the new tests) is alpha-dec-osf4.0b. I'll try the last version there.
Frank