Introduction

gmpy is a Python module that provides access to the GNU Multiple Precision (GMP) libary. GMP provides very fast and highly optimised routines for working with arbitrary precision integers (MPZ), rationals (MPQ), and floating point numbers (MPF). The GMP types are exposed to Python as gmpy.mpz, gmpy.mpq, and gmpy.mpf.

gmpy supports both Python 2 and 3. Due to differences between Python 2 and 3, a small number of behavioral changes were introduced in version 1.10 and later relative to previous releases (version 1.04 and earlier). To allow more extensive changes without silently breaking applications, the development version has been renamed to gmpy2. gmpy2 is in beta status and the API should be stable (i.e. no more backwards incompatible changes). To import gmpy2, you must use import gmpy2.

Enhancements in gmpy2

• Floating point support is provided by the MPFR library instead of using the MPF type from GMP. MPFR provides correctly rounded floating point arithmetic and a wide variety of transcendental functions. MPFR is exposed to Python as gmpy2.mpfr. The old mpf type no longer exists in gmpy2. Partial support was introduced in Alpha1. Improved support (context manager, subnormals, etc.) will be included in Alpha2.
• Arbitrary precision complex numbers are provided by the MPC library.
• A mutable integer type is available as gmpy2.xmpz. In addition to slightly faster performance for in-place operations, the xmpz allows slices to be used for setting and clearing individual bits.
• A large number of minor changes were made:
  • Wrapped rootrem, fib2, lucas, and lucas2.
  • More primality tests.
  • Wrapped is_even and is_odd.
  • Cached hash values to improve performance.
  • Improved random number support.
  • Added "slice" access to the bits of an mpz and xmpz. Bit manipulations work in-place on xmpz.
  • Renamed the bit manipulation methods; i.e. gmpy.scan1 is now gmpy2.bit_scan1.

Which version to use?

gmpy2 is now the recommended version, especially if you use the pre-compiled versions for Windows.

If you are compiling from source, gmpy2 is still the recommended version. However, gmpy2 requires recent versions of GMP, MPFR, and MPC and therefore gmpy2 may be more complicated to build. If you just need fast integer or rational arithmetic, you can use gmpy. The performance difference between gmpy and gmpy2 for integer and rational arithmetic is negligible.