History of changes to inline-math:

No. Date    Comments

1.0 20APR95 Initial version.

1.1 28APR95 Fixed bug in sgn(); replaced macros with inline functions.

1.2 28APR95 Introduced more efficient alogrithm for log1p().

1.3 04MAY95 Fixed bug in tan(); allow compilation under gcc -ansi.

1.4 16MAY95 Allow compilation under g++ -pedantic.

1.5 12JUN95 Removed copysign; sgn1() and sgn() simplified and return double.

1.7 17NOV95 Rewrote exp(), expm1(), and atan2() to avoid FP register stack
            overflow. Also added pow(), ceil(), floor(), extensions log2()
            and pow2(). 

1.8 29NOV95 Changed copyright to LGPL; included test based on Moshier's
            netlib.att.com/netlib/cephes/cmath.shar.Z consistency check.

1.9 04DEC95 Fixed asinh(0.0) floating exception.

2.0 27JUL96 Made minor cosmetic changes to header file; algorithms unchanged.

2.1 12AUG96 Prepended underscores to local variables.

2.2 31JAN97 Added miscellaneous missing prototypes; #define logb __logb

2.4 19AUG97 Merged with GLIBC version; added float and long double versions
            of all routines. 

2.5 31AUG97 Fixed compatibility with gcc-2.8.

2.6 18SEP97 Fixed C++ comment.

2.7 06MAY99 Workaround for gcc-2.8.1 inequality bug on i386; inlined
            versions of miscellaneous routines (eg. __sincos and __expm1)
			available even under -ansi, regardless of optimization level. 
