[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Axiom-developer] Re: differences

From: gdr
Subject: [Axiom-developer] Re: differences
Date: Mon, 02 Apr 2007 10:40:15 -0500
User-agent: Internet Messaging Program (IMP) H3 (4.0.2)

Quoting address@hidden:

As far as I know so far Axiom just uses the underlying lisp values
from cos and acos.

I don't consider it a bug but I don't consider it "stable" either.
All other values of all other functions seem to give the same
result and that result matches the input in this case. It's a
curious result though. I'd expect that they are using Chebyshev
polynomials and I'd have expected a symmetric result. Somebody
clearly "optimized" the world by saving an instruction somewhere.

Note also that in GCC-4.3, the compiler will be constant-folding
"non-trivial" functions called with constant arguments, e.g. sin(9.3),
thanks to MPFR and all that.  Consequently, the test must be precise
about what it is testing and how to reproduce the results across
platforms, otherwise we could just remove it.

The problem arose trying to explain a failing case in the
regression tests. Now that I understand it better it's not an issue.


reply via email to

[Prev in Thread] Current Thread [Next in Thread]