FLINT3.7.0-dev
/

FLINT 3.2

FLINT 3.2.2 (2025-03-31)#

Main contributors: Albin Ahlbäck (AA)

  • Continuous integration

    • Enable redoing upload release CI (AA).

  • Build system

    • Fix installation of headers for out-of-tree builds (AA).
    • Use __ARM_ACLE to guard inclusion of arm_acle.h (Lars Göttgens).

  • Maintenance

    • Remove assertion with uninitialized variable (AA).

FLINT 3.2.1 (2025-03-17)#

  • Bug fixes

    • Properly recognize AMD Zen during configuration (Albin Ahlbäck, reported by Doug Torrance).

FLINT 3.2.0 (2025-03-12)#

Main contributors: Albin Ahlbäck (AA), Bill Allombert (BA), Ricardo Buring (RB), Edgar Costa (EC), Tobias Diez (TD), Isuru Fernando (IF), Tommy Hofmann (TH), Max Horn (MH), Fredrik Johansson (FJ), Marc Mezzarobba (MM), Jake Moss (JM), Vincent Neiger (VN), Giacomo Pope (GP), Daniel Schultz (DS).

  • Features

    • Add new module mpn_mod for fixed-size few-word modulo arithmetic, with additional vector, matrix and polynomial arithmetic (FJ).
    • Add new module nfloat for packed fixed-size floating-point numbers with additional vector and matrix arithmetic (FJ).
    • Add fixed-point arithmetic under the nfloat module (FJ).

    • Major work on gr modules, including:

      • Toom-Cook multiplications for gr_poly,
      • Add Aberth and WDK root refinement for gr_poly,
      • Handle valuations and exact results in gr_series_div,
      • Addition of missing modular arithmetic routines,
      • Addition of comparison routines,
      • Matrix norms,
      • Improvements of Strassen matrix multiplication,
      • Addition of Waksman matrix multiplication,
      • Add all remaining scalar arithmetic methods for gr_mat,
      • Refactoring of gr_mpoly contexts,
      • Improve equality testing code for gr_mpoly.

      All was contributed by (FJ).

    • Implement computing qqbar roots of qqbar polynomials (FJ).
    • Implement generic flint_mpn_mulhigh and flint_mpn_sqrhigh for all ranges based on Mulders' algorithm (FJ).
    • Implement n_factor_evaluate (AA).
    • Wrap some more methods in flint_ctypes (FJ).
    • Let flint_mpn random functions take flint_randstate_t as input (FJ).
    • Add fmpz_randbits_unsigned (AA).
    • Add special case for _acb_dirichlet_zeta whenever s is a positive integer (FJ).
    • Typedef ulong and slong based on C primitives rather than relying on GMP's mp_limb_t and mp_limb_signed_t (AA).
    • Refactor FLINT's base randomisation, and rename the types and functions to be in proper FLINT-style.
    • Add some missing GR methods for modular arithmetic (FJ).
    • Add vec4n_(zero|mul|permute) in machine_vectors.h (VN).
    • Add cmpabs routines for acb and acf (FJ).
    • Add fq_default_ctx_init_randtest (GP).
    • Add some missing fmpz_mod_mpoly_compose functions (JM).
    • Improve enclosures in exponential integrals (FJ).
    • Improve transition regions and on wide intervals in acb_hypgeom_bessel_j (FJ).
    • Allow evaluation of the deflated Riemann zeta function on small intervals containing 1 (FJ).
    • Allow evaluation of a[rc]b_poly_sinc_series and a[rc]b_poly_sinc_pi_series on intervals containing 0 (FJ).

  • Examples

    • Add AKS primality example program (FJ).
    • Add example on double exponential integration (Hartmut Monien).
    • Update integral examples (FJ).

  • Performance

    • Implement flint_mpn_2add_n_inplace for x86_64 architectures supporting the ADX instruction set for adding two n limbed integers onto another n limbed integer inplace, returning the carry (AA).
    • Replace flint_mpn_divexact_1 with mpn_divexact_1 (AA).
    • Add flint_mpn_mul_Xn for X < 16 on Arm v8, outperforming GMP on Apple M1 (AA).
    • Add flint_mpn_mul_2 for Arm v8 (AA).
    • Push some parameters into flint-mparam.h, mainly thresholds for FFT multiplication, currently only for Skylake, Zen 3 and Apple M1 (AA).
    • Add inline assembly version of MPN_IORD_U, where the x86 version is taken from GMP, and an Arm v8 version was added as well (AA).
    • Use flint_mpn_mulhigh in mpn_*_preinvn methods (FJ).
    • Avoid 8x excessive memory allocation in fmpz preinvn functions (FJ).
    • Use flint_mpn_mulhigh for unbalanced preinvn divisions (FJ).
    • Minimize 32-bit instructions in x86_64 assembly.
    • Add flint_mpn_mullow_N for N \le 8 on x86_64 architectures supporting the ADX instruction set (AA).
    • Add flint_mpn_mullow_basecase for x86_64 architectures supporting the ADX instruction set (AA).
    • Micro-optimize flint_mpn_mulhigh (AA).
    • Add flint_mpn_sqr_N routines for N \le 9 for Arm v8 (AA).
    • Add flint_mpn_mulhigh_N routines for N \le 8 for Arm v8 (AA).
    • Add flint_mpn_sqrhigh_N routines for N \le 8 for Arm v8 (AA).
    • Add _flint_mpn_mulhigh_basecase routines Arm v8 optimized for Apple M1 (AA).
    • In gr matrix rings, call gr_mat_mul rather than gr_mat_mul_classical (FJ).
    • Change generic truncated power series to use gr_poly instead of gr_series as the data type (FJ).
    • Optimize nmod_poly_gcd with 64-bit moduli (FJ).
    • Do not calculate remainder in exact division of fq_nmod_mpoly unless assertion is needed.
    • Implement fast flint_mpn_mullow_n for large n, and use this in preinversion functions (FJ).
    • Simplify FLINT_MPN_MUL_2X2 (FJ).
    • Reduce inlining of flint_mpn_mulhigh_normalised (FJ).
    • Implement faster version of flint_mpn_addmod_2 (FJ).
    • Use flint_mpn_(add|sub)mod_n in the fmpz_factor module (FJ).
    • Use fmpz_poly_divexact instead of fmpz_poly_divides in one place (FJ).
    • Inline n_randint to avoid division operations when limit is compile-time constant (AA).
    • Simplify fmpz_randprime, and rely on n_randprime for bitsizes up to FLINT_BITS (AA).
    • Skip up to one call to fmpz_equal in fmpz_is_strong_probabprime (Mathieu Gouttenoire).
    • Optimize n_revbin for Aarch64 (AA).
    • Use builtin check arithmetic in ulong_extras.h and long_extras.h, where it was useful (AA).
    • Use _fmpz_new_mpz instead of mpz_init to allow the use of already allocated mpz (AA).
    • Improve formula selection for acb_hypgeom_beta_lower (FJ).
    • Write proper function for fmpz_mul2_uiui (AA).
    • Replace some n_[cf]log(X, 2) to FLINT_[CF]LOG2(X) (AA).
    • Replace malloc + memset calls by calloc in block_lanczos (AA).
    • Improve tuning for _mpz_tdiv_qr_preinvn (FJ).
    • Use preinversions in more cases in fmpz_mod (FJ).
    • Rewrite fmpz_lucas_chain to use various \mathbb{Z} / n \mathbb{Z} implementations (FJ).
    • Use Bini's algorithm to select initial values in acb_poly_find_roots (FJ).
    • Speed up nmod dot products, including vectorized routines (VN).
    • Use nmod dot products more throughout FLINT (VN).
    • Implement more modular multiplication routines with precomputated inverses (VN).
    • Add tuning for Intel Icelake (VN).
    • Avoid computing some entries twice in Strassen multiplication for (gr|fmpz)_mat (FJ).
    • Utilize symmetry when squaring in gr_mat_mul_strassen (FJ).
    • Improve algorithm selection in fmpz_mat_rref (FJ).
    • Use acb_dot in recurrence for Eisenstein series instead of multiple addmuls (FJ).

  • Bug fixes

    • Fix segfault in examples/fq_poly (FJ, reported by Andrea Lesavourey).
    • Fix uninitialized and uncleared variables (FJ).
    • Fix wrong stack usage in x86 and Arm assembly routines (AA).
    • Fix missing break statement in example program (FJ).
    • Fix assertion in flint_mpn_mulhigh (AA).
    • Fix bug in gr_series (FJ).
    • Fix primitive root prime (VN).
    • Fix gr_ctx_is_finite_characteristic for fmpz_mod (FJ).
    • Fix setting generator names for univariate gr rings (FJ).
    • Fix memory usage in x86 sqrhigh assembly where out-of-bound elements were accessed (AA).
    • Fix bugs in fq_default context initialisers (AA).
    • mpz_remove may reduce the number of limbs allocated, which may silently reduce the number of limbs contained in a big fmpz. Thus, ensure that after each call to mpz_remove, the result has a proper number of limbs allocated (AA).
    • Fix leading coefficient bug in fmpz_mpoly_factor (DS).
    • Fix bug in fq_default_poly_evaluate_fq_default where the check for fmpz_mod representation was incorrect (GP).
    • Fix division by zero in mpoly_monomials_deflate (JM).
    • Fix bug introduced in new function fq_default_ctx_init_randtest (AA).
    • Fix crash potential crashes in fmpz_poly_sub_fmpz (MH)
    • Fix modulus 1 for nmod_poly and fmpz_poly modules (TH).
    • Fix name clashes of variables in macros, as this becomes a problem whenever macros are layered (AA).
    • Fix NEED_CLZ_TAB (AA).
    • Patch a bug in Clang where it assumes __builtin_clz always returns something non-zero (AA).
    • Patch certain tests who caused segfaults during compilations in Clang (AA).
    • Avoid inline assembly for big aors_n in GCC on ARM since they lack enough registers to compile this (AA).
    • Fix wrong arithmetic function used in n_is_probabprime_lucas (FJ, reported by Mikhail Hogrefe).
    • Correct pointer casting for ARM in machine_vectors.h (AA).

  • Build system

    • Add GMP's config.guess and utilize it (AA).
    • Check for more CPUs in config.guess, including Intel Comet Lake and Github's Apple M1 virtual runner (AA).
    • Valdidate more CPUs in config.guess, including x86_64v3 (AA).
    • Set compiler architecture dependent flags depending on $host in Autotools (AA).
    • Add check for Aarch64/Arm v8 (AA).
    • Search for more GMP internal functions (AA).
    • Implement parameter files flint-mparam.h based on architecture (AA).
    • Create an include directory to build examples in Makefile (BA).
    • Fix missing header (VN).
    • Use CXX when testing NTL (George Huebner).
    • Check for GMP's mpn_invert_limb during build (AA).
    • Add -Wno-maybe-uninitialized to default C flags (AA).
    • Enable backup Cflags in the configuration, such as if -march=znver4 is prioritised over -march=znver3, then it will first try -march=znver4 and fall back to -march=znver3 if necessary.
    • Deprecate the --enable-tls option in the configuration (AA).
    • Add maintainer-level only option for fast build where optimizations are disabled (AA).
    • Recognize Comet Lake U in config.guess (AA).
    • Push non-executable stack when compiling assembly files (AA).
    • Remove unused compiler options for Clang, hence silencing warnings (AA).
    • Make Autotools build work out of tree (AA).
    • Don't assume AVX is available if user specified their own CFLAGS (AA).
    • Use cp -pRP instead of cp -a in Make as the latter is not supported by POSIX, causing installation failures on some systems (AA).
    • Add C11 atomics for MSVC builds (IF).
    • Recognize Apple M3 and Apple M4 (Pro) in config.guess (AA).
    • Fix static build for MSVC (IF, TD).
    • Only check assembly labels when it is actually being used (AA).
    • Update X86_64_PATTERN and X86_64_ADX_PATTERN and add some missing patterns (AA).
    • Recognize and map Zen 5 with correct CFLAGS and set some initial parameters for the architecture (AA).
    • Check presence of install-sh instead of config.guess during bootstrapping as config.guess is always present (AA).

  • Tests

    • Fix test for flint_mpn_mul where testing for big inputs was not done correctly (AA).
    • Cleanup tests related to assembly routines in mpn_extras (AA).
    • Add more test code for gr_series (FJ).
    • Fix warning in test (AA).
    • Add functions gr_mat_test_mul, gr_mat_test_lu and gr_poly_test_mullow for testing generics overrides (FJ).
    • Avoid mpn_random2 in tests (FJ).
    • Check correct type initialisation in fq_default (AA).
    • Add stricter bound to test of flint_mpn_mulhigh (AA).
    • Revise tests for ulong_extras (AA).
    • Reduce test multiplier where it was excessive (AA).
    • Speedup the test for flint_mpn_mul_toom22 (AA).
    • Merge tests for nmod_mat_(add|sub)mul (AA).
    • Simplify test for fmpz_randprime, and remove the risk this test failing (AA).
    • Revise test for fmpz_init_set_readonly (AA).
    • Add missing abort in test for arb_mat_solve_preapprox (AA).
    • Add more tests for gr module (FJ).
    • Add test for dlog_rho_t (FJ).
    • Add test for fq_default_poly_evaluate (GP).
    • Correct initialisation-utilisation-clearing order in test for fq_default_init (GP).
    • Add test for division by zero in fmpz_mpoly_(inflate|deflate) (JM).
    • Uncomment some previously commented code in fmpz_mod_mpoly_compose test, use smaller primes for this test (JM).
    • Use TEST_FUNCTION_FAIL macro in more multivariate polynomial tests (JM).
    • Explicitly check zero polynomial compositions for most multivariate polynomial modules (JM).
    • Randomize modulus in multivariate polynomial tests (JM).

  • Profiling

    • Add profile program for flint_mpn_divrem_preinvn (FJ).
    • Add profile program for flint_mpn_mullow (AA).
    • Fix p-mulmod_preinvn (FJ).
    • Add profile program for dot products in the nmod_vec module (VN).
    • Add profile program for nmod_vec_scalar_addmul* (VN).
    • General profile program fixes and touchups on the nmod modules (VN).

  • Maintenance

    • Add notes that CMake is only recommended for Windows users (AA).
    • Convert TODO from txt-format to Markdown (AA).
    • Generate and install CMake configuration files (Mehdi Chinoune).
    • Define GMP internal functions, if they are available, in mpn_extras.h (AA).
    • Change #ifdef FLINT_HAVE_FFT_SMALL to #if FLINT_HAVE_FFT_SMALL (AA).
    • Enable running specific tests in modules via make check MOD=XXX ARGS=YYY (AA).
    • Add Codecov key to CI (AA, FJ).
    • Force Unix-type newlines through git on Cygwin CI (AA).
    • Remove debug code from qqbar_roots_poly_squarefree (FJ).
    • Remove FLINT_HAVE_AVX* definitions (AA).
    • Update flint-config.h.in (AA).
    • Fix a macro (?) (FJ).
    • Simplify definition of mp_limb_pair_t (FJ).
    • Use #include <flint/xxx.h> in examples (BA).
    • Change __mpz_struct * to mpz_ptr (EC).
    • Move mpn macros from flint.h to mpn_extras.h (AA)
    • Enable compiling with -Wextra -Werror for a big part of the library (AA).
    • Disable static build by default (AA).
    • Do not remove intermediate assembly files for making debugging easier (AA).
    • Add make debug MOD=XXX ARGS=YYY shortcut for debugging with GDB (AA).
    • Add gdb_history and vgcore.* to .gitignore (AA).
    • Replace "FLINT2" by "FLINT" (Julian Rüth).
    • Remove obviously outdated entries in TODO.md (FJ).
    • Remove unnecessary inclusions of mpn_extras.h (FJ).
    • Remove some undocumented and/or unused functions, including most of mpfr_* modules and n_cbrt_newton_iteration (AA, FJ).
    • Change __mpz_struct * to mpz_ptr (AA).
    • Add keywords FLINT_MALLOC and FLINT_RETURNS_NONNULL. Use these along with FLINT_WARN_UNUSED whenever suitable in the memory managing functions in flint.h (AA).
    • Remove inclusion of gmp.h and limits.h in flint.h and other headers (AA).
    • Use FLINT types instead of GMP types in source code (AA).
    • Remove signed_mpn_sub_n in favor of flint_mpn_signed_sub_n (AA).
    • Cleanup copyright claims in longlong.h (AA).
    • Remove byte_swap from longlong.h (AA).
    • Remove flint_mul_sizes and use z_mul_checked instead (AA).
    • Remove r_shift and l_shift from flint.h, and only define where needed (AA).
    • Remove inclusion of longlong.h from flint.h (AA).
    • Remove old deprecations in fmpz.h (AA).
    • General header cleanups and removal of unnecessary inclusions in them (AA).
    • Replace calls to mpz_realloc2 to mpz_realloc, and only do these calls if it is truly necessary (AA).
    • Add constant MPZ_MIN_ALLOC that decides the minimum number of allocated limbs that the fmpz-collector may receive such that _fmpz_new_mpz always returns an mpz with at least MPZ_MIN_ALLOC limbs allocated (AA).
    • Merge several source files (AA).
    • Simplify MAG_FIXMUL for 32-bit systems (AA).
    • Delete acb_vec_printd_index, and only define where needed in test files (AA).
    • Add flags to gr_acb_ctx, currently only analytic flag implemented (FJ).
    • Further maintenance of fft_small (DS).
    • Increase code coverage in tests (AA, FJ).
    • Use _mm256_srli_epi64 instead of _mm256_srl_epi64 in machine_vectors.h (VN).
    • Clarify that libtool is needed in INSTALL.md and README.md (AA).
    • Do not use I as a variable name as it is reserved in C99 (Dima Pasechnik).
    • Deprecate aliases in the arith module (AA).
    • Move add_s{n}a{2n} and friends from crt_helpers.h to longlong.h (FJ).
    • Rename FLINT_TEST_CLEANUP to FLINT_TEST_CLEAR (VN).
    • Add multiple missing flint_rand_clear (VN).
    • Fix typos in source code (Frédéric Chapoton).
    • Add flint-mparam.h for Zen 4 (VN).
    • Remove prototypes of non-existing functions arb_poly_mullow_ztrunc (MM).
    • Only include headers before extern "C" (AA).
    • Undefine small macro in qsieve.h as it may cause clashes on Windows systems (AA).
    • Fix noreturn attribute for C23, as it uses [[noreturn]] for this (Michael Orlitzky).
    • Add newlines to end of files that where missing them (AA).
    • Add mpn_extras/inlines.c (AA).
    • Replace compound literal with a struct to accommodate MSVC (TD).
    • Cast pointers in machine_vectors.h to suppress warnings about incompatible pointers (AA).
    • Make GCC pragmas dispatch only for GCC (AA).
    • Comment out unused static functions (AA).
    • Check availability of rbit instruction for ARM in n_revbin (AA).

  • Continuous integration

    • Add nightly build to Github (EC).
    • Add SHAsum (EC).
    • Remove Ubuntu CMake runner as we no longer recommend CMake for building FLINT on non-Windows systems (AA).
    • Add runner that checks against regression when compiling with -Wextra -Werror (AA).
    • Publish prepreleases only as drafts (Mahrud Sayrafi).
    • Multiple updates of the runners according to Github's runners (AA, EC).
    • Exclude profiler source code from Codecov (AA).
    • Assume lcov version 2.1 is available (AA).
    • Touchups and fixes for the release CI (EC).
    • Add Linux ARM runner (AA).
    • Fix release CI to automatically make releases when new tags are created, and push related files and generate website accordingly (AA, EC).
    • Rebuild citation block on website after each release (AA).

  • Documentation

    • Fix typos (BA, Frédéric Chapoton, EC, MM, VN).
    • Clarify usage of inline assembly addition and subtraction macros such as add_ssaaaa (AA).
    • Add some explanations for mpn_ctx_mpn_mul (FJ).
    • Add human-readable text to documentation of ordering_t (RB).
    • Document Generic Ring setters for infinities and extended values (RB).
    • Fix documentation of gr_cmp_other (MM).
    • Fix reference to Granlund-Montgomery division (AA).
    • Correct names of the SFB-TRR 195 grant by the DFG (MH).
    • Correct fq_default_default to fq_default (AA).
    • Clarify documentation for fq_default context initialisers (AA).
    • Revise some fmpz docstrings (AA).
    • Document MPZ_MIN_ALLOC and low-level fmpz memory management functions (AA).
    • Correct docstring for udiv_qrnnd (VN).
    • Rework docstring for n_mulmod_precomp_shoup, and clarify its constraints (VN).
    • Document nmod_mat_set_mod (VN).
    • Document fmpz_mod_poly-ulong scalar functions (AA).
    • Clarify that releases do not require Autotools (AA).
    • Fix signatures for acb_mat_get_(real|imag) (Marek Kaluba).
Cite this page

The FLINT team, “FLINT 3.2,” FLINT: Fast Library for Number Theory, documentation, version 3.7.0-dev, 2026. https://flintlib.org/doc/history/flint3/v3-2.html (generated 2026-10-20).