From: noer AT cygnus DOT com (Geoffrey Noer) Subject: Re: Problems with transcendentals in libm 22 Apr 1997 20:04:05 -0700 Approved: cygnus DOT gnu-win32 AT cygnus DOT com Distribution: cygnus Message-ID: <199704221825.LAA06655.cygnus.gnu-win32@cirdan.cygnus.com> Content-Type: text Original-To: mjo AT mrao DOT cam DOT ac DOT uk (Martin Oldfield) Original-Cc: gnu-win32 AT cygnus DOT com In-Reply-To: from "Martin Oldfield" at Jul 11, 96 00:23:00 am X-Mailer: ELM [version 2.4 PL23] Original-Sender: owner-gnu-win32 AT cygnus DOT com Greetings, I've confirmed that the following problem is fixed in the upcoming beta18 release. I now get the same behavior for gnu-win32 gcc as for Solaris gcc, with and without optimization on. Thanks for the test case! (I believe this was caused by errors in floating point initialization in the crt0). -- Geoffrey Noer noer AT cygnus DOT com Martin Oldfield wrote: > > The following program produces bogus output. This appears to imply a > bug in the cosh(x) and tanh(x) functions. Is this a known problem and > are there any workarounds ? [...] > Broken lines running from cygwin32 > > x: logx expx tanhx coshx sinhx > 0.368: -1 1.44 0.352 NaN 0.376 > 2.72: 1 15.2 NaN NaN 7.54 > 7.39: 2 1.62e+03 NaN NaN 809 > 20.1: 3 5.28e+08 NaN NaN 2.64e+08 > > -0.368: -Inf 0.692 -0.352 NaN -0.376 > -2.72: -Inf 0.066 NaN NaN -7.54 > -7.39: -Inf 0.000618 NaN NaN -809 > -20.1: -Inf 1.89e-09 NaN NaN -2.64e+08 > > ------------------------------------------------------------------------ > > Comparable lines running under linux > > x: logx expx tanhx coshx sinhx > 0.368: -1 1.44 0.352 1.07 0.376 > 2.72: 1 15.2 0.991 7.61 7.54 > 7.39: 2 1.62e+03 1 809 809 > 20.1: 3 5.28e+08 1 2.64e+08 2.64e+08 > > -0.368: NaN 0.692 -0.352 1.07 -0.376 > -2.72: NaN 0.066 -0.991 7.61 -7.54 > -7.39: NaN 0.000618 -1 809 -809 > -20.1: NaN 1.89e-09 -1 2.64e+08 -2.64e+08 - For help on using this list (especially unsubscribing), send a message to "gnu-win32-request AT cygnus DOT com" with one line of text: "help".