Portability — FLINT

FLINT runs on 64-bit and 32-bit platforms, on Unix-like systems and on Windows. Code using FLINT stays portable if it follows a few rules about word sizes.

Word types and constants#

The C types long and unsigned long have 64 bits on 64-bit Unix systems but only 32 bits on 64-bit Windows. FLINT therefore uses its own types ulong and slong, which always have the size of a GMP limb, FLINT_BITS bits: 64 on all 64-bit platforms, including Windows, and 32 on 32-bit platforms. Use these types for all values passed to FLINT functions taking ulong or slong, and do not assume that they are unsigned long and long.

Integer constants of word size are written with the macros UWORD and WORD, as in UWORD(1) << 40 or WORD(-123456789012), instead of the suffixes UL and L, which give 32-bit constants on 64-bit Windows. The extreme values are UWORD_MAX, WORD_MIN and WORD_MAX. Code that depends on the word size can test FLINT_BITS (or FLINT64) with the preprocessor, and the same holds for constants that only fit in 64 bits.

Printing and reading words#

The format specifiers %lu and %ld of the C library are wrong for ulong and slong on 64-bit Windows. FLINT's I/O functions flint_printf, flint_fprintf, flint_sprintf, flint_snprintf, flint_scanf, flint_fscanf and flint_sscanf accept the length modifier w instead: %wu and %wd print a ulong and an slong on every platform (and %wx in hexadecimal). With the C library functions, the macro WORD_FMT gives the right length modifier, as in printf(WORD_FMT "d", x). The flint_printf functions also print FLINT objects directly; see Formatted output.

Platform-dependent behaviour#

The results of FLINT functions do not depend on the platform, with a few documented exceptions:

Functions that take or return words naturally have a word-size range: fmpz_get_si returns a 64-bit or 32-bit value, nmod moduli are limited to FLINT_BITS bits, and the random functions of ulong_extras produce word-size values, so random test inputs differ between 32-bit and 64-bit platforms.
Tuning cutoffs, and therefore the choice of algorithms, depend on the processor. Results are the same, but timings, and the exact error bounds of some ball arithmetic functions, may differ.
The fft_small.h module is only available on x86-64 processors with AVX2 and on ARM64; elsewhere other algorithms are used.
flint_sort uses qsort_r with the GNU C library and a merge sort elsewhere, so the order of equal elements may differ.

Building on Windows#

On Windows, FLINT can be built with MSYS2 or Cygwin using configure, or natively with CMake and MSVC; see CMake on Windows. With MinGW-w64, FLINT is built with 64-bit long long words (FLINT_LONG_LONG is defined in flint-config.h). Programs compiled with MSVC can use the FLINT headers as long as they are compiled with the same configuration as the library.