[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Bug-gnulib] [RFC] Module: long double math functions
From: |
Bruno Haible |
Subject: |
Re: [Bug-gnulib] [RFC] Module: long double math functions |
Date: |
Mon, 10 Feb 2003 20:28:10 +0100 (CET) |
Paolo Bonzini writes:
> I have adapted the long double math functions (sinl & co.) in glibc so that
> they don't make any reference to the format of the numbers. This caused a
> loss in speed and probably in accuracy as well, but it is fine for
> portability purposes.
Good. Can you put this into gnulib, together with autoconf tests that
verify the assumptions of the code? I mean: does it assume
sizeof (long double) > sizeof (double)
precisionof (long double) <= 2 * precisionof (double)
IEEE format
sizeof (long double) == 12 or == 16 ?
If you have precision problems, you can look at the algorithms of
Keith Briggs' doubledouble package (a.k.a. xdouble in LIDIA), see
http://members.lycos.co.uk/keithmbriggs/doubledouble.html
- his algorithms provide 106 bits of precision.
Bruno