NEWS

Copyright 1996, 1999-2016 Free Software Foundation, Inc.

Verbatim copying and distribution of this entire article is permitted in any
medium, provided this notice is preserved.


Changes between GMP version 6.1.1 and 6.1.2

  BUGS FIXED
  * Mini-GMP: Fixed a division bug, which on a machine with 64-bit
    unsigned long affects approximately 1 out of 2^32 divisors.

  * Mini-GMP: Fix mpz_set_str crash on inputs with a large number of
    leading zeros. Also stricter input validation, rejecting inputs
    with no digits.

  FEATURES
  * Handle more systems which require PIC code in static libraries (e.g.,
    "hardened" Gentoo and Debian 9).

  * Configuration for arm (-32 and -64) has been rewritten, fixing poor
    code selection for many CPUs.

  * Mini-GMP: Updated to the latest development version, including
    new functions mpn_com and mpn_neg.

  SPEEDUPS
  * None, except for arm CPUs affected by the configuration rewrite.

  MISC
  -

Changes between GMP version 6.1.0 and 6.1.1

  BUGS FIXED
  * Make Intel Broadwell configurations work on Windows.

  FEATURES
  * Work around faulty cpuid on some recent Intel chips (this allows GMP to run
    on Skylake Pentiums).

  * Support thumb-less ARM chips.

Changes between GMP version 6.0.* and 6.1.0

  BUGS FIXED
  * The public function mpn_com is now correctly declared in gmp.h.

  * Healed possible failures of mpn_sec_sqr for non-cryptographic sizes for
    some obsolete CPUs.

  * The option --disable-assembly now disables all inlined asm.

  * Fixed bug affecting mini-gmp's bitwise functions mpz_setbit, mpz_clrbit,
    and mpz_combit.

  * Various problems related to precision for mpf have been fixed.

  * Fixed ABI incompatible stack alignment in calls from assembly code.

  * Fixed PIC bug in popcount affecting Intel processors using the 32-bit ABI.

  SPEEDUPS
  * Speedup for Intel Broadwell and Skylake through assembly code making use of
    new ADX instructions.

  * Square root is now faster when the remainder is not needed. Also the speed
    to compute the k-th root improved, for small sizes.

  * Improved arm64 support.

  FEATURES
  * New C++ functions gcd and lcm for mpz_class.

  * New public mpn functions mpn_divexact_1, mpn_zero_p, and mpn_cnd_swap.

  * New public mpq_cmp_z function, to efficiently compare rationals with
    integers.

  * Support for Darwin in all x86 code, thereby enabling fat builds on Darwin.

  * Support for more 32-bit arm processors.

  * Support for compilation with clang/llvm on more platforms.  Caution: GMP
    triggers mis-compilation bugs in clang for many platforms, such as arm, x86
    (32-bit and 64-bit), powerpc, mips.

  * Support for AVX-less modern x86 CPUs. (Such support might be missing either
    because the CPU vendor chose to disable AVX, or because the running kernel
    lacks AVX context switch support.)

  * Stack usage trimmed; we believe 512 KiB is now sufficient for any GMP
    call, irrespective of operand size.

  * Support for NetBSD under Xen; we switch off AVX unconditionally under
    NetBSD since a bug in NetBSD makes AVX fail under Xen.

  MISC
  * We now use manufacturers' code names for x86 CPUs, e.g., "haswell" instead
    of names derived from the commercial brands.

  * Small improvements and better coverage for the test suite.

  * The various FreeBSD problems listed for 6.0.0 affect this release too.

  * Tuned values for FFT multiplications are provided for larger number on
    many platforms.

Changes between GMP version 5.1.* and 6.0.0

  BUGS FIXED
  * The function mpz_invert now considers any number invertible in Z/1Z.

  * The mpn multiply code now handles operands of more than 2^31 limbs
    correctly.  (Note however that the mpz code is limited to 2^32 bits on
    32-bit hosts and 2^37 bits on 64-bit hosts.)

  * Contains all fixes from release 5.1.3.

  SPEEDUPS
  * Plain division of large operands is faster and more monotonous in operand
    size.

  * Major speedup for ARM, in particular ARM Cortex-A15, thanks to improved
    assembly.

  * Major speedup for SPARC T4/T5 and speedup also for T3, thanks to a lot of
    new assembly.

  * Speedup for Intel Sandy Bridge, Ivy Bridge, Haswell, thanks to rewritten
    and vastly expanded assembly support.  Speedup also for the older Core 2
    and Nehalem.

  * Faster mixed arithmetic between mpq_class and double.

  * With g++, optimise more operations when one argument is a simple constant.

  FEATURES
  * Support for new Intel and AMD CPUs.

  * Support for ARM64 alias Aarch64 alias ARMv8.

  * New public functions mpn_sec_mul and mpn_sec_sqr, implementing side-channel
    silent multiplication and squaring.

  * New public functions mpn_sec_div_qr and mpn_sec_div_r, implementing
    side-channel silent division.

  * New public functions mpn_cnd_add_n and mpn_cnd_sub_n.  Side-channel silent
    conditional addition and subtraction.

  * New public function mpn_sec_powm, implementing side-channel silent modexp.

  * New public function mpn_sec_invert, implementing side-channel silent
    modular inversion.

  * Better support for applications which use the mpz_t type, but nevertheless
    need to call some of the lower-level mpn functions.  See the documentation
    for mpz_limbs_read and related functions.

  MISC
  * This release will not work on NetBSD 5.x, FreeBSD 7.x, 8.x or 9 series
    before 9.3.  The reason is that the m4 command is not correctly
    implemented.  (Workaround: Use an older GMP release, or install GNU m4 from
    /usr/ports and tell GMP to use it.)

  * This release will not build properly on FreeBSD/amd64 before version 10
    using the 32-bit ABI (once a working m4 is installed).  The reason is
    broken limits.h.  (Workaround: Use an older GMP release if using the 32-bit
    ABI on these FreeBSD releases is important.)

  * This release will not work reliably on FreeBSD 10.0 for i386 or amd64 using
    the 32-bit ABI.  The reason is bugs in the compiler 'clang'.  Depending on
    CPU-dependent compiler flags, GMP may or may not be miscompiled in a
    particular build.  (Workaround: Compiling gcc from /usr/ports should work,
    except that gcc circularly depends on GMP; we have not been able to test
    that workaround due to FreeBSD 10.0 bugs affecting its ability to run under
    KVM and Xen.)

  * This release will not compile on FreeBSD before version 10 for i386,
    targeting any modern AMD processor.  The reason is bugs in the old gcc
    bundled with FreeBSD.  (Workaround: install a less obsolete gcc from
    /usr/ports and tell GMP to use it, or override the -march=amdfam10
    GMP configure command line argument.)


Changes between GMP version 5.1.2 and 5.1.3

  BUGS FIXED
  * The internal functions mpn_sbpi1_div_qr_sec mpn_sbpi1_div_r_sec could
    compute garbage with a low probability.  They are now rewritten, and the
    test code has been improved.

  * A bug in the ia64 implementation of mpn_divrem_2, clobbering some
    callee-save registers, has been fixed. This is an internal
    function, with the bug manifesting itself as miscomputation in,
    e.g., mpn_sqrtrem.

  * The documentation now correctly says 'const' for input arguments.

  SPEEDUPS
  * None.

  FEATURES
  * None.

  MISC
  * None.


Changes between GMP version 5.1.1 and 5.1.2

  BUGS FIXED
  * A bug in mpz_powm_ui triggered by base arguments of at least 15000 decimal
    digits or mod arguments of at least 7500 decimal digits has been fixed.

  * An AMD Bulldozer specific bug affecting the 64-bit Windows ABI has been
    fixed.  This bug was in a key function (mpn_mul_1) and made both Bulldozer
    specific builds and fat builds run on Bulldozer completely non-functional.

  SPEEDUPS
  * None.

  FEATURES
  * None.

  MISC
  * Fixes and generalisations to the test suite.

  * Minor portability enhancements.


Changes between GMP version 5.1.0 and 5.1.1

  BUGS FIXED
  * On Windows 64-bit, an error causing link errors about
    __gmp_binvert_limb_table has been fixed.

  * Aarch64 alias ARM64 support now works.

  * A possible buffer overrun in mpz_ior has been fixed.

  * A rare sign flip in mpz_remove has been fixed.

  * A bug causing problems with mpf numbers with absolute value >= 2^31 has
    been fixed.

  * Several bugs in mini-gmp have been fixed.

  * A bug caused by automake, related to the 'distcheck' target, has been fixed
    by upgrading the automake used for GMP release engineering.

  SPEEDUPS
  * None.

  FEATURES
  * Preliminary support for the x32 ABI under x86-64.

  MISC
  * The mini-gmp testsuite now tests the entire set of functions.

  * Various improvements of the GMP testsuite.


Changes between GMP version 5.0.* and 5.1.0

  BUGS FIXED
  * When reading a C++ number (like mpz_class) in an istream reaches the end
    of the stream, the eofbit is now set.

  * The result sign of mpz_rootrem's remainder is now always correct.

  * The mpz_remove function now handles negative divisors.

  * Contains all fixes from release 5.0.5.

  SPEEDUPS
  * The n-factorial and n-over-k functions have been reimplemented for great
    speedups for small and large operands.

  * New subquadratic algorithm for the Kronecker/Jacobi/Legendre symbol.

  * Major speedup for ARM, in particular ARM Cortex-A9 and A15, thanks to broad
    assembly support.

  * Significant speedup for POWER6 and POWER7 thanks to improved assembly.

  * The performance under M$ Windows' 64-bit ABI has been greatly improved
    thanks to complete assembly support.

  * Minor speed improvements of many functions and for many platforms.

  FEATURES
  * Many new CPUs recognised.

  * New functions for multi-factorials, and primorial: mpz_2fac_ui,
    mpz_mfac_uiui and mpz_primorial_ui.

  * The mpz_powm_sec function now uses side-channel silent division for
    converting into Montgomery residues.

  * The fat binary mechanism is now more robust in its CPU recognition.

  MISC
  * Inclusion of assembly code is now controlled by the configure options
    --enable-assembly and --disable-assembly.  The "none" CPU target is gone.

  * In C++, the conversions mpq_class->mpz_class, mpf_class->mpz_class and
    mpf_class->mpq_class are now explicit.

  * Includes "mini-gmp", a small, portable, but less efficient, implementation
    of a subset of GMP's mpn and mpz interfaces. Used in GMP bootstrap, but it
    can also be bundled with applications as a fallback when the real GMP
    library is unavailable.

  * The ABIs under AIX are no longer called aix32 and aix64, but mode64 and 32.
    This is more consistent with other powerpc systems.

  * The coverage of the testsuite has been improved, using the lcov tool.  See
    also https://gmplib.org/devel/lcov/.

  * It is now possible to compile GMP using a C++ compiler.

  * K&R C compilers are no longer supported.

  * The BSD MP compatibility functions have been removed.


Changes between GMP version 5.0.4 and 5.0.5

  BUGS FIXED
  * A bug causing AMD 11h processors to be treated like AMD 10h has been fixed.
    The 11h processors do not correctly handle all 10h (aka K10) instructions,
    and GMP's use of these instructions results in major miscomputations (not
    as one would have hoped CPU traps of some 'illegal instruction' sort).

  * A bug affecting recent Intel Sandy Bridge CPUs resulting in configuration
    failures has been fixed.

  SPEEDUPS
  * None.

  FEATURES
  * A couple of tests added to the self-check suite.

  MISC
  * None.


Changes between GMP version 5.0.3 and 5.0.4

  BUGS FIXED
  * Thresholds in mpn_powm_sec for both fat and non-fat builds are now used
    safely, plugging a one-word buffer overrun introduced in the 5.0.3 release
    (for non-fat) and a multi-word buffer overrun that existed since 5.0 (for
    fat).  (We have not been able to provoke malign stack smashing in any of
    the ~100 configurations explored by the GMP nightly builds, but the bug
    should be assumed to be exploitable.)

  * Two bugs in multiplication code causing incorrect computation with
    extremely low probability have been fixed.

  * A bug in the test suite causing buffer overruns during "make check",
    sometimes leading to subsequent malloc crashes, has been fixed.

  * Two bugs in the gcd code have been fixed.  They could lead to incorrect
    results, but for uniformly distributed random operands, the likelihood for
    that is infinitesimally small.  (There was also a third bug, but that was
    an incorrect ASSERT, which furthermore was not enabled by default.)

  * A bug affecting 32-bit PowerPC division has been fixed.  The bug caused
    miscomputation for certain divisors in the range 2^32 ... 2^64-1 (about 1
    in 2^30 of these).

  SPEEDUPS
  * None, except indirectly through recognition of new CPUs, and through better
    tuning parameters.

  FEATURES
  * Some more tests added to the self-check suite.

  * The AMD "Bulldozer" CPU is now recognised.

  MISC
  * None.


Changes between GMP version 5.0.2 and 5.0.3

  BUGS FIXED
  * A few minor bugs related to portability fixed.

  * A slight timing leak of the powm_sec functions have been sealed.  (This
    leak could possibly be used to extract the most significant few bits of the
    exponent.  "Few" here means at most 10.)

  * The mpz_nextprime function now runs a safer number of pseudo-random prime
    tests.

  * A bug in division code possibly causing incorrect computation was fixed.

  SPEEDUPS
  * None, except indirectly through recognition of new CPUs, and through better
    tuning parameters.

  FEATURES
  * New CPUs recognised.

  * IBM S/390 are now supported in both 31/32-bit and 64-bit mode.  (We have
    not been able to fully test this on any multilib machine, since IBM expired
    our guest account a few days before our release.)

  MISC
  * None.


Changes between GMP version 5.0.1 and 5.0.2

  BUGS FIXED
  * Many minor bugs related to portability fixed.

  * The support for HPPA 2.0N now works, after an assembly bug fix.

  * A test case type error has been fixed.  The symptom of this bug was
    spurious 'make check' failures.

  SPEEDUPS
  * None, except indirectly through recognition of new CPUs.

  FEATURES
  * Fat builds are now supported for 64-bit x86 processors also under Darwin.

  MISC
  * None.


Changes between GMP version 5.0.0 and 5.0.1

  BUGS FIXED
  * Fat builds fixed.

  * Fixed crash for huge multiplies when old FFT_TABLE2 type of parameter
    selection tables' sentinel was smaller than multiplied operands.

  * The solib numbers now reflect the removal of the documented but preliminary
    mpn_bdivmod function; we correctly flag incompatibility with GMP 4.3.  GMP
    5.0.0 has this wrong, and should perhaps be uninstalled to avoid confusion.

  SPEEDUPS
  * Multiplication of large numbers has indirectly been sped up through better
    FFT tuning and processor recognition.  Since many operations depend on
    multiplication, there will be a general speedup.

  FEATURES
  * More Core i3, i5 an Core i7 processor models are recognised.

  * Fixes and workarounds for Mac OS quirks should make this GMP version build
    using many of the different versions of "Xcode".

  MISC
  * The amount of scratch memory needed for multiplication of huge numbers has
    been reduced substantially (but is still larger than in GMP 4.3.)

  * Likewise, the amount of scratch memory needed for division of large numbers
    has been reduced substantially.

  * The FFT tuning code of tune/tuneup.c has been completely rewritten, and
    new, large FFT parameter selection tables are provided for many machines.

  * Upgraded to the latest autoconf, automake, libtool.


Changes between GMP version 4.3.X and 5.0.0

  BUGS FIXED
  * None (contains the same fixes as release 4.3.2).

  SPEEDUPS
  * Multiplication has been overhauled:
    (1) Multiplication of larger same size operands has been improved with
        the addition of two new Toom functions and a new internal function
        mpn_mulmod_bnm1 (computing U * V mod (B^n-1), B being the word base.
        This latter function is used for the largest products, waiting for a
        better Schoenhage-Strassen U * V mod (B^n+1) implementation.
    (2) Likewise for squaring.
    (3) Multiplication of different size operands has been improved with the
        addition of many new Toom function, and by selecting underlying
        functions better from the main multiply functions.

  * Division and mod have been overhauled:
    (1) Plain "schoolbook" division is reimplemented using faster quotient
        approximation.
    (2) Division Q = N/D, R = N mod D where both the quotient and remainder
        are needed now runs in time O(M(log(N))).  This is an improvement of
        a factor log(log(N))
    (3) Division where just the quotient is needed is now O(M(log(Q))) on
        average.
    (4) Modulo operations using Montgomery REDC form now take time O(M(n)).
    (5) Exact division Q = N/D by means of mpz_divexact has been improved
        for all sizes, and now runs in time O(M(log(N))).

  * The function mpz_powm is now faster for all sizes.  Its complexity has
    gone from O(M(n)log(n)m) to O(M(n)m) where n is the size of the modulo
    argument and m is the size of the exponent.  It is also radically
    faster for even modulus, since it now partially factors such modulus
    and performs two smaller modexp operations, then uses CRT.

  * The internal support for multiplication yielding just the lower n limbs
    has been improved by using Mulders' algorithm.

  * Computation of inverses, both plain 1/N and 1/N mod B^n have been
    improved by using well-tuned Newton iterations, and wrap-around
    multiplication using mpn_mulmod_bnm1.

  * A new algorithm makes mpz_perfect_power_p asymptotically faster.

  * The function mpz_remove uses a much faster algorithm, is better tuned,
    and also benefits from the division improvements.

  * Intel Atom and VIA Nano specific optimisations.

  * Plus hundreds of smaller improvements and tweaks!

  FEATURES
  * New mpz function: mpz_powm_sec for side-channel quiet modexp
    computations.

  * New mpn functions: mpn_sqr, mpn_and_n, mpn_ior_n, mpn_xor_n, mpn_nand_n,
    mpn_nior_n, mpn_xnor_n, mpn_andn_n, mpn_iorn_n, mpn_com, mpn_neg,
    mpn_copyi, mpn_copyd, mpn_zero.

  * The function mpn_tdiv_qr now allows certain argument overlap.

  * Support for fat binaries for 64-bit x86 processors has been added.

  * A new type, mp_bitcnt_t for bignum bit counts, has been introduced.

  * Support for Windows64 through mingw64 has been added.

  * The cofactors of mpz_gcdext and mpn_gcdext are now more strictly
    normalised, returning to how GMP 4.2 worked.  (Note that also release
    4.3.2 has this change.)

  MISC
  * The mpn_mul function should no longer be used for squaring,
    instead use the new mpn_sqr.

  * The algorithm selection has been improved, the number of thresholds have
    more than doubled, and the tuning and use of existing thresholds have
    been improved.

  * The tune/speed program can measure many of new functions.

  * The mpn_bdivmod function has been removed.  We do not consider this an
    incompatible change, since the function was marked as preliminary.

  * The testsuite has been enhanced in various ways.


Changes between GMP version 4.3.1 and 4.3.2

  Bugs:
  * Fixed bug in mpf_eq.
  * Fixed overflow issues in mpz_set_str, mpz_inp_str, mpf_set_str, and
    mpf_get_str.
  * Avoid unbounded stack allocation for unbalanced multiplication.
  * Fixed bug in FFT multiplication.

  Speedups:
  * None, except that proper processor recognition helps affected processors.

  Features:
  * Recognise more "Core 2" processor variants.
  * The cofactors of mpz_gcdext and mpn_gcdext are now more strictly
    normalised, returning to how GMP 4.2 worked.


Changes between GMP version 4.3.0 and 4.3.1

  Bugs:
  * Fixed bug in mpn_gcdext, affecting also mpz_gcdext and mpz_invert.
    The bug could cause a cofactor to have a leading zero limb, which
    could lead to crashes or miscomputation later on.
  * Fixed some minor documentation issues.

  Speedups:
  * None.

  Features:
  * Workarounds for various issues with Mac OS X's build tools.
  * Recognise more IBM "POWER" processor variants.


Changes between GMP version 4.2.X and 4.3.0

  Bugs:
  * Fixed bug in mpz_perfect_power_p with recognition of negative perfect
    powers that can be written both as an even and odd power.
  * We might accidentally have added bugs since there is a large amount of
    new code in this release.

  Speedups:
  * Vastly improved assembly code for x86-64 processors from AMD and Intel.
  * Major improvements also for many other processor families, such as
    Alpha, PowerPC, and Itanium.
  * New sub-quadratic mpn_gcd and mpn_gcdext, as well as improved basecase
    gcd code.
  * The multiply FFT code has been slightly improved.
  * Balanced multiplication now uses 4-way Toom in addition to schoolbook,
    Karatsuba, 3-way Toom, and FFT.
  * Unbalanced multiplication has been vastly improved.
  * Improved schoolbook division by means of faster quotient approximation.
  * Several new algorithms for division and mod by single limbs, giving
    many-fold speedups.
  * Improved nth root computations.
  * The mpz_nextprime function uses sieving and is much faster.
  * Countless minor tweaks.

  Features:
  * Updated support for fat binaries for x86_32 include current processors
  * Lots of new mpn internal interfaces.  Some of them will become public
    in a future GMP release.
  * Support for the 32-bit ABI under x86-apple-darwin.
  * x86 CPU recognition code should now default better for future
    processors.
  * The experimental nails feature does not work in this release, but
    it might be re-enabled in the future.

  Misc:
  * The gmp_version variable now always contains three parts.  For this
    release, it is "4.3.0".


Changes between GMP version 4.2.3 and 4.2.4

  Bugs:
  * Fix bug with parsing exponent '+' sign in mpf.
  * Fix an allocation bug in mpf_set_str, also affecting mpf_init_set_str, and
    mpf_inp_str.

  Speedups:
  * None, except that proper processor recognition helps affected processors.

  Features:
  * Recognize new AMD processors.


Changes between GMP version 4.2.2 and 4.2.3

  Bugs:
  * Fix x86 CPU recognition code to properly identify recent AMD and Intel
    64-bit processors.
  * The >> operator of the C++ wrapper gmpxx.h now does floor rounding, not
    truncation.
  * Inline semantics now follow the C99 standard, and works with recent GCC
    releases.
  * C++ bitwise logical operations work for more types.
  * For C++, gmp.h now includes cstdio, improving compiler compatibility.
  * Bases > 36 now work properly in mpf_set_str.

  Speedups:
  * None, except that proper processor recognition helps affected processors.

  Features:
  * The allocation functions now detect overflow of the mpz_t type.  This means
    that overflow will now cause an abort, except when the allocation
    computation itself overflows.  (Such overflow can probably only happen in
    powering functions; we will detect powering overflow in the future.)


Changes between GMP version 4.2.1 and 4.2.2

  * License is now LGPL version 3.

  Bugs:
  * Shared library numbers corrected for libcxx.
  * Fixed serious bug in gmpxx.h where a=a+b*c would generate garbage.
    Note that this only affects C++ programs.
  * Fix crash in mpz_set_d for arguments with large negative exponent.
  * Fix 32-bit ABI bug with Itanium assembly for popcount and hamdist.
  * Fix assembly syntax problem for powerpc-ibm-aix with AIX native assembler.
  * Fix problems with x86 --enable-fat, where the compiler where told to
    generate code for the build machine, not plain i386 code as it should.
  * Improved recognition of powerpc systems wrt Altivec/VMX capability.
  * Misc minor fixes, mainly workarounds for compiler/assembler bugs.

  Speedups:
  * "Core 2" and Pentium 4 processors, running in 64-bit mode will get a
     slight boost as they are now specifically recognized.

  Features:
  * New support for x86_64-solaris
  * New, rudimentary support for x86-apple-darwin and x86_64-apple-darwin.
    (Please see https://gmplib.org/macos.html for more information.)


Changes between GMP version 4.2 and 4.2.1

  Bugs:
  * Shared library numbers corrected.
  * Broken support for 32-bit AIX fixed.
  * Misc minor fixes.

  Speedups:
  * Exact division (mpz_divexact) now falls back to plain division for large
    operands.

  Features:
  * Support for some new systems.


Changes between GMP version 4.1.4 and 4.2

  Bugs:
  * Minor bug fixes and code generalizations.
  * Expanded and improved test suite.

  Speedups:
  * Many minor optimizations, too many to mention here.
  * Division now always subquadratic.
  * Computation of n-factorial much faster.
  * Added basic x86-64 assembly code.
  * Floating-point output is now subquadratic for all bases.
  * FFT multiply code now about 25% faster.
  * Toom3 multiply code faster.

  Features:
  * Much improved configure.
  * Workarounds for many more compiler bugs.
  * Temporary allocations are now made on the stack only if small.
  * New systems supported: HPPA-2.0 gcc, IA-64 HP-UX, PowerPC-64 Darwin,
    Sparc64 GNU/Linux.
  * New i386 fat binaries, selecting optimised code at runtime (--enable-fat).
  * New build option: --enable-profiling=instrument.
  * New memory function: mp_get_memory_functions.
  * New Mersenne Twister random numbers: gmp_randinit_mt, also now used for
    gmp_randinit_default.
  * New random functions: gmp_randinit_set, gmp_urandomb_ui, gmp_urandomm_ui.
  * New integer functions: mpz_combit, mpz_rootrem.
  * gmp_printf etc new type "M" for mp_limb_t.
  * gmp_scanf and friends now accept C99 hex floats.
  * Numeric input and output can now be in bases up to 62.
  * Comparisons mpz_cmp_d, mpz_cmpabs_d, mpf_cmp_d recognise infinities.
  * Conversions mpz_get_d, mpq_get_d, mpf_get_d truncate towards zero,
    previously their behaviour was unspecified.
  * Fixes for overflow issues with operands >= 2^31 bits.

  Caveats:
  * mpfr is gone, and will from now on be released only separately.  Please see
    www.mpfr.org.


Changes between GMP version 4.1.3 and 4.1.4

* Bug fix to FFT multiplication code (crash for huge operands).
* Bug fix to mpf_sub (miscomputation).
* Support for powerpc64-gnu-linux.
* Better support for AMD64 in 32-bit mode.
* Upwardly binary compatible with 4.1.3, 4.1.2, 4.1.1, 4.1, 4.0.1, 4.0,
  and 3.x versions.


Changes between GMP version 4.1.2 and 4.1.3

* Bug fix for FFT multiplication code (miscomputation).
* Bug fix to K6 assembly code for gcd.
* Bug fix to IA-64 assembly code for population count.
* Portability improvements, most notably functional AMD64 support.
* mpz_export allows NULL for countp parameter.
* Many minor bug fixes.
* mpz_export allows NULL for countp parameter.
* Upwardly binary compatible with 4.1.2, 4.1.1, 4.1, 4.0.1, 4.0, and 3.x
  versions.


Changes between GMP version 4.1.1 and 4.1.2

* Bug fixes.


Changes between GMP version 4.1 and 4.1.1

* Bug fixes.
* New systems supported: NetBSD and OpenBSD sparc64.


Changes between GMP version 4.0.1 and 4.1

* Bug fixes.
* Speed improvements.
* Upwardly binary compatible with 4.0, 4.0.1, and 3.x versions.
* Asymptotically fast conversion to/from strings (mpz, mpq, mpn levels), but
  also major speed improvements for tiny operands.
* mpn_get_str parameter restrictions relaxed.
* Major speed improvements for HPPA 2.0 systems.
* Major speed improvements for UltraSPARC systems.
* Major speed improvements for IA-64 systems (but still sub-optimal code).
* Extended test suite.
* mpfr is back, with many bug fixes and portability improvements.
* New function: mpz_ui_sub.
* New functions: mpz_export, mpz_import.
* Optimization for nth root functions (mpz_root, mpz_perfect_power_p).
* Optimization for extended gcd (mpz_gcdext, mpz_invert, mpn_gcdext).
* Generalized low-level number format, reserving a `nails' part of each
  limb.  (Please note that this is really experimental; some functions
  are likely to compute garbage when nails are enabled.)
* Nails-enabled Alpha 21264 assembly code, allowing up to 75% better
  performance.  (Use --enable-nails=4 to enable it.)


Changes between GMP version 4.0 and 4.0.1

* Bug fixes.


Changes between GMP version 3.1.1 and 4.0

* Bug fixes.
* Speed improvements.
* Upwardly binary compatible with 3.x versions.
* New CPU support: IA-64, Pentium 4.
* Improved CPU support: 21264, Cray vector systems.
* Support for all MIPS ABIs: o32, n32, 64.
* New systems supported: Darwin, SCO, Windows DLLs.
* New divide-and-conquer square root algorithm.
* New algorithms chapter in the manual.
* New malloc reentrant temporary memory method.
* New C++ class interface by Gerardo Ballabio (beta).
* Revamped configure, featuring ABI selection.
* Speed improvements for mpz_powm and mpz_powm_ui (mainly affecting small
  operands).
* mpz_perfect_power_p now properly recognizes 0, 1, and negative perfect
  powers.
* mpz_hamdist now supports negative operands.
* mpz_jacobi now accepts non-positive denominators.
* mpz_powm now supports negative exponents.
* mpn_mul_1 operand overlap requirements relaxed.
* Float input and output uses locale specific decimal point where available.
* New gmp_printf, gmp_scanf and related functions.
* New division functions: mpz_cdiv_q_2exp, mpz_cdiv_r_2exp, mpz_divexact_ui.
* New divisibility tests: mpz_divisible_p, mpz_divisible_ui_p,
  mpz_divisible_2exp_p, mpz_congruent_p, mpz_congruent_ui_p,
  mpz_congruent_2exp_p.
* New Fibonacci function: mpz_fib2_ui.
* New Lucas number functions: mpz_lucnum_ui, mpz_lucnum2_ui.
* Other new integer functions: mpz_cmp_d, mpz_cmpabs_d, mpz_get_d_2exp,
  mpz_init2, mpz_kronecker, mpz_lcm_ui, mpz_realloc2.
* New rational I/O: mpq_get_str, mpq_inp_str, mpq_out_str, mpq_set_str.
* Other new rational functions: mpq_abs, mpq_cmp_si, mpq_div_2exp,
  mpq_mul_2exp, mpq_set_f.
* New float tests: mpf_integer_p, mpf_fits_sint_p, mpf_fits_slong_p,
  mpf_fits_sshort_p, mpf_fits_uint_p, mpf_fits_ulong_p, mpf_fits_ushort_p.
* Other new float functions: mpf_cmp_d, mpf_get_default_prec, mpf_get_si,
  mpf_get_ui, mpf_get_d_2exp.
* New random functions: gmp_randinit_default, gmp_randinit_lc_2exp_size.
* New demo expression string parser (see demos/expr).
* New preliminary perl interface (see demos/perl).
* Tuned algorithm thresholds for many more CPUs.


Changes between GMP version 3.1 and 3.1.1

* Bug fixes for division (rare), mpf_get_str, FFT, and miscellaneous minor
  things.


Changes between GMP version 3.0 and 3.1

* Bug fixes.
* Improved `make check' running more tests.
* Tuned algorithm cutoff points for many machines.  This will improve speed for
  a lot of operations, in some cases by a large amount.
* Major speed improvements: Alpha 21264.
* Some speed improvements: Cray vector computers, AMD K6 and Athlon, Intel P5
  and Pentium Pro/II/III.
* The mpf_get_prec function now works as it did in GMP 2.
* New utilities for auto-tuning and speed measuring.
* Multiplication now optionally uses FFT for very large operands.  (To enable
  it, pass --enable-fft to configure.)
* Support for new systems: Solaris running on x86, FreeBSD 5, HP-UX 11, Cray
  vector computers, Rhapsody, Nextstep/Openstep, MacOS.
* Support for shared libraries on 32-bit HPPA.
* New integer functions: mpz_mul_si, mpz_odd_p, mpz_even_p.
* New Kronecker symbol functions: mpz_kronecker_si, mpz_kronecker_ui,
  mpz_si_kronecker, mpz_ui_kronecker.
* New rational functions: mpq_out_str, mpq_swap.
* New float functions: mpf_swap.
* New mpn functions: mpn_divexact_by3c, mpn_tdiv_qr.
* New EXPERIMENTAL function layer for accurate floating-point arithmetic, mpfr.
  To try it, pass --enable-mpfr to configure.  See the mpfr subdirectory for
  more information; it is not documented in the main GMP manual.


Changes between GMP version 3.0 and 3.0.1

* Memory leaks in gmp_randinit and mpz_probab_prime_p fixed.
* Documentation for gmp_randinit fixed.  Misc documentation errors fixed.


Changes between GMP version 2.0 and 3.0

* Source level compatibility with past releases (except mpn_gcd).
* Bug fixes.
* Much improved speed thanks to both host independent and host dependent
  optimizations.
* Switch to autoconf/automake/libtool.
* Support for building libgmp as a shared library.
* Multiplication and squaring using 3-way Toom-Cook.
* Division using the Burnikel-Ziegler method.
* New functions computing binomial coefficients: mpz_bin_ui, mpz_bin_uiui.
* New function computing Fibonacci numbers: mpz_fib_ui.
* New random number generators: mpf_urandomb, mpz_rrandomb, mpz_urandomb,
  mpz_urandomm, gmp_randclear, gmp_randinit, gmp_randinit_lc_2exp,
  gmp_randseed, gmp_randseed_ui.
* New function for quickly extracting limbs: mpz_getlimbn.
* New functions performing integer size tests: mpz_fits_sint_p,
  mpz_fits_slong_p, mpz_fits_sshort_p, mpz_fits_uint_p, mpz_fits_ulong_p,
  mpz_fits_ushort_p.
* New mpf functions: mpf_ceil, mpf_floor, mpf_pow_ui, mpf_trunc.
* New mpq function: mpq_set_d.
* New mpz functions: mpz_addmul_ui, mpz_cmpabs, mpz_cmpabs_ui, mpz_lcm,
  mpz_nextprime, mpz_perfect_power_p, mpz_remove, mpz_root, mpz_swap,
  mpz_tdiv_ui, mpz_tstbit, mpz_xor.
* New mpn function: mpn_divexact_by3.
* New CPU support: DEC Alpha 21264, AMD K6 and Athlon, HPPA 2.0 and 64,
  Intel Pentium Pro and Pentium-II/III, Sparc 64, PowerPC 64.
* Almost 10 times faster mpz_invert and mpn_gcdext.
* The interface of mpn_gcd has changed.
* Better support for MIPS R4x000 and R5000 under Irix 6.
* Improved support for SPARCv8 and SPARCv9 processors.


Changes between GMP version 2.0 and 2.0.2

* Many bug fixes.


Changes between GMP version 1.3.2 and 2.0

* Division routines in the mpz class have changed.  There are three classes of
  functions, that rounds the quotient to -infinity, 0, and +infinity,
  respectively.  The first class of functions have names that begin with
  mpz_fdiv (f is short for floor), the second class' names begin with mpz_tdiv
  (t is short for trunc), and the third class' names begin with mpz_cdiv (c is
  short for ceil).

  The old division routines beginning with mpz_m are similar to the new
  mpz_fdiv, with the exception that some of the new functions return useful
  values.

  The old function names can still be used.  All the old functions names will
  now do floor division, not trunc division as some of them used to.  This was
  changed to make the functions more compatible with common mathematical
  practice.

  The mpz_mod and mpz_mod_ui functions now compute the mathematical mod
  function.  I.e., the sign of the 2nd argument is ignored.

* The mpq assignment functions do not canonicalize their results.  A new
  function, mpq_canonicalize must be called by the user if the result is not
  known to be canonical.
* The mpn functions are now documented.  These functions are intended for
  very time critical applications, or applications that need full control over
  memory allocation.  Note that the mpn interface is irregular and hard to
  use.
* New functions for arbitrary precision floating point arithmetic.  Names
  begin with `mpf_'.  Associated type mpf_t.
* New and improved mpz functions, including much faster GCD, fast exact
  division (mpz_divexact), bit scan (mpz_scan0 and mpz_scan1), and number
  theoretical functions like Jacobi (mpz_jacobi) and multiplicative inverse
  (mpz_invert).
* New variable types (mpz_t and mpq_t) are available that makes syntax of
  mpz and mpq calls nicer (no need for & before variables).  The MP_INT and
  MP_RAT types are still available for compatibility.
* Uses GNU configure.  This makes it possible to choose target architecture
  and CPU variant, and to compile into a separate object directory.
* Carefully optimized assembly for important inner loops.  Support for DEC
  Alpha, Amd 29000, HPPA 1.0 and 1.1, Intel Pentium and generic x86, Intel
  i960, Motorola MC68000, MC68020, MC88100, and MC88110, Motorola/IBM
  PowerPC, National NS32000, IBM POWER, MIPS R3000, R4000, SPARCv7,
  SuperSPARC, generic SPARCv8, and DEC VAX.  Some support also for ARM,
  Clipper, IBM ROMP (RT), and Pyramid AP/XP.
* Faster.  Thanks to the assembler code, new algorithms, and general tuning.
  In particular, the speed on machines without GCC is improved.
* Support for machines without alloca.
* Now under the LGPL.

INCOMPATIBILITIES BETWEEN GMP 1 AND GMP 2

* mpq assignment functions do not canonicalize their results.
* mpz division functions round differently.
* mpz mod functions now really compute mod.
* mpz_powm and mpz_powm_ui now really use mod for reduction.