FLINT3.7.0-dev
/

FLINT 3.1

FLINT 3.1.2 (2024-03-18)#

Main contributors: Albin Ahlbäck (AA).

  • Maintenance

    • Remove the need for ldconfig completely in FLINT's Autotools build system (AA).

FLINT 3.1.1 (2024-03-07)#

Main contributors: Albin Ahlbäck (AA).

  • Bug fixes

    • Add padic_types.h to headers in Makefile.in (AA).

  • Maintenance

    • Add check for aligned_alloc and _aligned_malloc for systems that may not provide any of these functions (AA).
    • Add options for setting -march=ARCH (AA).

FLINT 3.1.0 (2024-02-25)#

Main contributors: Fredrik Johansson (FJ), Albin Ahlbäck (AA), Jean Kieffer (JK).

  • License

    • Changed license from "LGPL 2.1 or later" to "LGPL 3 or later".

  • Major interface changes

    • The methods in the fmpz_mod_mat module now use a context object (FJ).
    • Changed fq_mat_rref and others to allow separate input and output matrices (FJ).

  • Features

  • Bugs

    • Fixed threading problem in gr_method_tab_init: FLINT would occasionally crash when calling generics-based internal code when using a large number of threads (FJ, after debugging by Alexander Smirnov).
    • Fixed comparison of gr vectors with fmpq elements (FJ).
    • Fixed allocation bug in gr_mpoly_mul_monomial (FJ).
    • Fixed aliasing in fmpzi_divrem_approx (FJ).
    • Avoid division by zero in acb_poly_refine_roots_durand_kerner: in rare instances, computing roots of an integer polynomial could hang (FJ).
    • Allow large arguments in arb_atan_frac_bsplit (FJ).
    • Fixed printing large coefficients in nmod_mpoly (Alexander Smirnov, AA).
    • Fixed fq_*_poly_powmod (AA).
    • Fixed initialization of fq_default_ctx (Claus Fieker, Tommy Hofmann).
    • Fixed memory leak in gr_poly_write (FJ).
    • Fixed printing nmod32 elements on 32-bit systems (FJ).
    • Fixed ldconfig for BSD systems (AA).
    • Fixed FLINT_WANT_ASSERT for CMake (AA).
    • Fixed memory leak in arb_nint (FJ).

  • Performance

    • FLINT is now built with -O3 -march=native by default (AA).
    • FLINT is no longer built with -funroll-loops by default except for select modules. This reduces the library size by more than 20%. (AA).
    • Assembly routines are now used as intended on ARM64 when compiling with GCC (AA).
    • New basecase code for flint_mpn_mul, flint_mpn_mul_n and flint_mpn_sqr (generic C versions, assembly for x86-64). This can yield up to a 2x speedup over GMP for short integer multiplications when calling mpn functions directly, though few applications currently benefit significantly due to wrapper overheads (some Arb benchmarks run ~5% faster with this change) (AA, FJ).
    • Added hardcoded low-level routines for high multiplications of two operands of the same size for Broadwell-type CPUs – flint_mpn_mulhigh_* and flint_mpn_sqrhigh_* (AA).
    • Added hardcoded low-level routines for high normalised multiplications of two normalised operands of the same size for Broadwell-type CPUs – flint_mpn_mulhigh_normalised_* (AA).
    • Added flint_mpn_mulhigh_basecase and flint_mpn_sqrhigh_basecase for Broadwell-type CPUs (AA).
    • Use toom22 on top of custom basecase code for intermediate operands in flint_mpn_mul (FJ, based on GMP code).
    • Use mulx in umul_ppmm when available (AA).
    • Faster _fmpz_vec_dot (FJ).
    • Faster _fmpz_mod_vec_dot (FJ).
    • Faster fmpz_poly and fmpz_mat basecase algorithms based on dot products (FJ).
    • Optimized fmpz_mat_mul_classical (FJ).
    • Added fmpz_mat_mul_waksman, speeding up fmpz_mat multiplication for balanced matrices with huge entries (Éric Schost, Vincent Neiger, FJ).
    • Tweaks to improve GCC's code generation for fmpz_mpoly and nmod_mpoly multiplication (FJ).
    • Improved fmpz_mod_mat_set_fmpz_mat (FJ).
    • Improved tuning for fmpz_mat_sqr (FJ).
    • Improved fmpz_mat_sqr_bodrato with small coefficients (Marco Bodrato).
    • Squaring optimizations in fmpz_mat_mul_multi_mod (FJ).
    • Use Bodrato's sequence for Strassen multiplication (Marco Bodrato).
    • Strip trailing zeros in fmpz_poly_gcd: this gives a 50x speedup computing gcd(x^1000, x^1001) (FJ).
    • Faster CLD bound computation, speeding up fmpz_poly_factor for some polynomials (FJ).
    • Use new formulas from Jorge Zuniga and Jesús Guillera to compute log(2), Catalan's constant, zeta(3) and gamma(1/3) faster (FJ).
    • Compressed the database of Conway polynomials to 10% of the original size (AA).
    • Optimized context initializers for fq, fq_zech, fq_nmod and qadic (AA).
    • Avoid calls to the slow standard library function ldexp (FJ).
    • Improved fmpz_is_probabprime for word-size input (FJ).
    • Make gr_poly_resultant use the quasilinear hgcd algorithm by default over finite rings and remove obsolete fmpz_mod_poly implementations (FJ).
    • Optimized header files (AA).
    • Changed several internal helper functions to forced inlines (AA).
    • Merged some sources files to speed up compilation (AA).
    • Faster computation of Swinnerton-Dyer polynomials (FJ).
    • Added benchmark script (dev/bench.py) that should be used to check performance increases and regressions (FJ).
    • Add special cases for Clang and MSVC in longlong.h (AA).

  • Test code

    • Unified test programs per module: compiling FLINT's test suite is now an order of magnitude faster (AA).
    • Added pretty-printing and timing output for unit tests (AA).
    • Improve use of test multiplier in some long-running unit tests (AA, FJ).
    • Improved test coverage (AA, FJ).
    • Allow gr_ctx_init_random to generate composite rings (FJ).
    • Fixed the test code of fq_*_poly_powmod (AA).

  • Maintenance

    • Require GMP >= 6.2.1 and MPFR >= 4.1.0 (AA).
    • Drop support for Itanium (AA).
    • Drop support for MPIR (AA).
    • Cleaned up longlong.h (AA).
    • Removed ARB_VERSION, ANTIC_VERSION, CALCIUM_VERSION and associated constants (AA).
    • Removed _long_vec_print and _perm_print (use flint_printf instead) (AA).
    • Removed unused functions in the d_mat and aprcl modules (AA).
    • Removed fmpq_get_mpz_frac (AA).
    • Removed arb_fmpz_poly_cos_minpoly (AA).
    • Removed fmpz_mat_mul_classical_inline (FJ).
    • Removed fmpz_mat_rref_mod (use fmpz_mod_mat_rref instead) (FJ).
    • Introduce FLINT_SWAP macro to replace several older macros (FJ).
    • Replaced invert_limb by n_preinvert_limb_prenorm (AA).
    • Renamed _perm_set_one to _perm_one (AA).
    • Only define some multithreaded "divides" function when the CPU is strongly ordered (AA).
    • Enable fft_small for MSVC builds (AA).
    • Use binary format instead of decimal format in qsieve, removing the need of conversions between said formats (AA).
    • Use C11 atomics in the fmpz memory manager (AA).
    • Merged some repeated code in the mpoly modules (FJ).
    • Refactored fq_default to use gr generics internally (FJ).
    • Replaced more functions by generics-based versions (FJ).
    • Do not include pthread.h when opted out (AA).
    • Test ARM NEON in CI via Github's Apple M1 runner (AA).
    • Test examples in CI (AA).
    • Detect MPFR and GMP internals in configure (AA).
    • Add -lflint to PKG-CONFIG (Josh Rickmar).
    • Silence GCC compiler warnings (AA).
    • Unified exception handling (AA).
    • Corrected some function signatures in the documentation (Vincent Delecroix, Joel-Dahne, Edgar Costa).
    • Rename the default Git-branch from trunk to main (FJ).
    • Document some macros defined in flint.h (AA).
    • Other code cleanup and modernisation (AA).
    • Major cleanup in configure.ac and acinclude.m4.
    • Use parts of GMP's configuration to configure assembly properly.
    • Change instances of #ifdef FLINT_WANT_ASSERT to #if FLINT_WANT_ASSERT (AA).
Cite this page

The FLINT team, “FLINT 3.1,” FLINT: Fast Library for Number Theory, documentation, version 3.7.0-dev, 2026. https://flintlib.org/doc/history/flint3/v3-1.html (generated 2026-10-20).