This is the mail archive of the
libstdc++@gcc.gnu.org
mailing list for the libstdc++ project.
Re: libstdc++/10277
- From: Gabriel Dos Reis <gdr at integrable-solutions dot net>
- To: "Voykov" <voykov at casaccia dot enea dot it>
- Cc: gcc-bugs at gcc dot gnu dot org, libstdc++ at gcc dot gnu dot org
- Date: 20 May 2003 13:26:14 +0200
- Subject: Re: libstdc++/10277
- Organization: Integrable Solutions
- References: <003601c31ebe$640d0420$531710ac@dhcpnet.casaccia>
"Voykov" <voykov@casaccia.enea.it> writes:
| Mr. Dos Reis,
|
| ?
|
| '1.0 - exp(x)' jumps to 0.0 for 'double x' < 1.25e-16 and 'double x' >
| -1.00e-16, values near to the 'mach_eps<double>'. It is strange that for
| 'float x' the corresponding values are 5.68e-20 and -2.84e-20, far from the
| 'mach_eps<float>'.
Using GCC-3.3 and mainline, I did the test on a GNU/Linux running a
Pentium IV and for those values, the differences were consistently
zero, so the jumps were not there. I suspect this is an issue with
the mathematical functions from your C library.
-- Gaby