On Aug 7 14:10, Tim Prince wrote: > > On 8/7/2014 1:41 PM, Denis Excoffier wrote: > >Hello, > > > >Perhaps the values of some constants are not as accurate as they should be? > >See also M_LN2LO and M_LN2HI. > > > >diff -uNr cygwin-snapshot-20140807-1-original/newlib/libc/include/math.h > >cygwin-snapshot-20140807-1-patched/newlib/libc/include/math.h > >--- cygwin-snapshot-20140807-1-original/newlib/libc/include/math.h > >2014-08-07 18:26:21.000000000 +0200 > >+++ cygwin-snapshot-20140807-1-patched/newlib/libc/include/math.h > >2014-08-07 19:31:51.000000000 +0200 > >@@ -569,14 +569,14 @@ > > #ifndef __STRICT_ANSI__ > > #define M_TWOPI (M_PI * 2.0) > >-#define M_3PI_4 2.3561944901923448370E0 > >-#define M_SQRTPI 1.77245385090551602792981 > >+#define M_3PI_4 2.3561944901923449288E0 > >+#define M_SQRTPI 1.77245385090551602729817 > > #define M_LN2LO 1.9082149292705877000E-10 > > #define M_LN2HI 6.9314718036912381649E-1 > >-#define M_SQRT3 1.73205080756887719000 > >+#define M_SQRT3 1.73205080756887729353 > > #define M_IVLN10 0.43429448190325182765 /* 1 / log(10) */ > > #define M_LOG2_E _M_LN2 > >-#define M_INVLN2 1.4426950408889633870E0 /* 1 / log(2) */ > >+#define M_INVLN2 1.4426950408889634074E0 /* 1 / log(2) */ > > /* Global control over fdlibm error handling. */ > > > > > > > Without the L suffix, it doesn't look like it matters. > I just noticed this week the M_LN2LO and HI. They should add up to a long > double accurate constant. I've been using a similar pair in implementation > of exp() for 20 years.
Uhm, I'm not exactly a math expert so I won't try to contribute. However, given that this file is from newlib, it would be more helpful to discuss this on the newlib AT sourceware DOT org mailing list. Thanks, Corinna -- Corinna Vinschen Please, send mails regarding Cygwin to Cygwin Maintainer cygwin AT cygwin DOT com Red Hat
pgpivfvjBMity.pgp
Description: PGP signature
