MPFR

GNU MPFR (for GNU Multiple Precision Floating-Point Reliably^[1]) is a portable C library for arbitrary-precision binary floating-point computation with correct rounding, based on GNU Multi-Precision Library. The computation is both efficient and has a well-defined semantics. It copies the ideas from the ANSI/IEEE-754 standard for fixed-precision floating-point arithmetic. More precisely, its main features are:

• Support for special numbers: signed zeros (−0), infinities and not-a-number (a single NaN is currently supported).
• Each number has its own precision (in bits since MPFR uses radix 2). The floating-point results are correctly rounded to the precision of the target variable, in any of the four IEEE-754 rounding modes.
• Supported functions: MPFR implements all mathematical functions from C99: the logarithm and exponential in natural base, base 2 and base 10, the log(1+x) and exp(x)-1 functions (log1p and expm1), the six trigonometric and hyperbolic functions and their inverses, the gamma, zeta and error functions, the arithmetic geometric mean, the power (x^y) function. All those functions are correctly rounded over their complete range.
• Subnormals are not supported, but can be emulated with the mpfr_subnormalize function.

MPFR is not able to track the accuracy of numbers in a whole program or expression; this is not its goal. Interval arithmetic packages like MPFI, or Real RAM implementations like iRRAM, which may be based on MPFR, can do that for the user.

External links

• Official MPFR web site
• MPFR project page

References