xref: /netbsd-src/external/lgpl3/gmp/dist/NEWS (revision 2dd295436a0082eb4f8d294f4aa73c223413d0f2)
1Copyright 1996, 1999-2016, 2018-2020 Free Software Foundation, Inc.
2
3Verbatim copying and distribution of this entire article is permitted in any
4medium, provided this notice is preserved.
5
6Changes between GMP version 6.2.0 and 6.2.1
7
8  BUGS FIXED
9  * A possible overflow of type int is avoided for mpz_cmp on huge operands.
10
11  * Overflows are more carefully detected and reported for mpz_pow_ui.
12
13  * A bug in longlong.h for aarch64 sub_ddmmss, not affecting GMP, was healed.
14
15  * mini-gmp: mpz_out_str and mpq_out_str now correctly handle out of
16    range bases.
17
18  FEATURES
19  * C90 compliance.
20
21  * Initial support for Darwin on arm64, and improved portability.
22
23  * Support for more processors.
24
25  SPEEDUPS
26  * None, except indirectly through recognition of new CPUs.
27
28  MISC
29  * None.
30
31Changes between GMP version 6.1.* and 6.2.0
32
33  BUGS FIXED
34  * gmp_snprintf now correctly reports an error (returns -1) when snprintf
35    does.
36
37  * Conversion to double is now more robust even if the configuration process
38    does not recognize the float format.
39
40  * mpz_powm could return a not completely normalised value, when the
41    exponent was 1 and the base < 0.
42
43  * mpf_set_str could create invalid mpf_t variables for input strings with
44    many leading zeros.
45
46  FEATURES
47  * New C++ functions factorial, primorial and fibonacci for mpz_class.
48
49  * Functions to detect primality now substitute the first 24 Miller-Rabin
50    iterations with the BPSW test.
51
52  * Mini-GMP: new functions mpz_2fac_ui and mpz_mfac_uiui.
53
54  * Mini-GMP: mpz_sizeinbase, mpz_get_str, and mpz_set_str now support bases up
55    to 62.
56
57  * Mini-GMP: added support for the mpq_t layer.
58
59  * MIPS r6 cores are now supported.
60
61  SPEEDUPS
62  * Major speedup for AMD Ryzen and Epyc thanks to an extensive set of assembly
63    code.
64
65  * Major speedup for IBM POWER9 thanks to assembly code making use of new madd
66    instruction.
67
68  * Speedup for 64-bit ARM CPUs thanks to new/improved assembly code.
69
70  * The n-over-k function mpz_bin_ui has been reimplemented for great speedups
71    for large operands.
72
73  * Speedup for the worst case of mpz_perfect_power_p.
74
75  * Speedup for gcd for small and medium size operands.
76
77  * Speedup for really huge multiplies thanks to much larger FFT tables.
78
79  MISC
80  * Internal representation of the mpz_t variables now supports lazy
81    allocation; memory is allocated only when a value is stored.
82
83  * Small improvements and better coverage for the test suite.
84
85  * The tune/speed program can measure some more functions.
86
87  * The low-level function mpn_mul no longer diverts to mpn_sqr, users should
88    call mpn_sqr directly when applicable.
89
90  * New installed files gmp.pc and gmpxx.pc, for use with pkg-config.
91    Contributed by Hugh McMaster.
92
93Changes between GMP version 6.1.1 and 6.1.2
94
95  BUGS FIXED
96  * Mini-GMP: Fixed a division bug, which on a machine with 64-bit
97    unsigned long affects approximately 1 out of 2^32 divisors.
98
99  * Mini-GMP: Fix mpz_set_str crash on inputs with a large number of
100    leading zeros. Also stricter input validation, rejecting inputs
101    with no digits.
102
103  FEATURES
104  * Handle more systems which require PIC code in static libraries (e.g.,
105    "hardened" Gentoo and Debian 9).
106
107  * Configuration for arm (-32 and -64) has been rewritten, fixing poor
108    code selection for many CPUs.
109
110  * Mini-GMP: Updated to the latest development version, including
111    new functions mpn_com and mpn_neg.
112
113  SPEEDUPS
114  * None, except for arm CPUs affected by the configuration rewrite.
115
116  MISC
117  -
118
119Changes between GMP version 6.1.0 and 6.1.1
120
121  BUGS FIXED
122  * Make Intel Broadwell configurations work on Windows.
123
124  FEATURES
125  * Work around faulty cpuid on some recent Intel chips (this allows GMP to run
126    on Skylake Pentiums).
127
128  * Support thumb-less ARM chips.
129
130Changes between GMP version 6.0.* and 6.1.0
131
132  BUGS FIXED
133  * The public function mpn_com is now correctly declared in gmp.h.
134
135  * Healed possible failures of mpn_sec_sqr for non-cryptographic sizes for
136    some obsolete CPUs.
137
138  * The option --disable-assembly now disables all inlined asm.
139
140  * Fixed bug affecting mini-gmp's bitwise functions mpz_setbit, mpz_clrbit,
141    and mpz_combit.
142
143  * Various problems related to precision for mpf have been fixed.
144
145  * Fixed ABI incompatible stack alignment in calls from assembly code.
146
147  * Fixed PIC bug in popcount affecting Intel processors using the 32-bit ABI.
148
149  SPEEDUPS
150  * Speedup for Intel Broadwell and Skylake through assembly code making use of
151    new ADX instructions.
152
153  * Square root is now faster when the remainder is not needed. Also the speed
154    to compute the k-th root improved, for small sizes.
155
156  * Improved arm64 support.
157
158  FEATURES
159  * New C++ functions gcd and lcm for mpz_class.
160
161  * New public mpn functions mpn_divexact_1, mpn_zero_p, and mpn_cnd_swap.
162
163  * New public mpq_cmp_z function, to efficiently compare rationals with
164    integers.
165
166  * Support for Darwin in all x86 code, thereby enabling fat builds on Darwin.
167
168  * Support for more 32-bit arm processors.
169
170  * Support for compilation with clang/llvm on more platforms.  Caution: GMP
171    triggers mis-compilation bugs in clang for many platforms, such as arm, x86
172    (32-bit and 64-bit), powerpc, mips.
173
174  * Support for AVX-less modern x86 CPUs. (Such support might be missing either
175    because the CPU vendor chose to disable AVX, or because the running kernel
176    lacks AVX context switch support.)
177
178  * Stack usage trimmed; we believe 512 KiB is now sufficient for any GMP
179    call, irrespective of operand size.
180
181  * Support for NetBSD under Xen; we switch off AVX unconditionally under
182    NetBSD since a bug in NetBSD makes AVX fail under Xen.
183
184  MISC
185  * We now use manufacturers' code names for x86 CPUs, e.g., "haswell" instead
186    of names derived from the commercial brands.
187
188  * Small improvements and better coverage for the test suite.
189
190  * The various FreeBSD problems listed for 6.0.0 affect this release too.
191
192  * Tuned values for FFT multiplications are provided for larger number on
193    many platforms.
194
195Changes between GMP version 5.1.* and 6.0.0
196
197  BUGS FIXED
198  * The function mpz_invert now considers any number invertible in Z/1Z.
199
200  * The mpn multiply code now handles operands of more than 2^31 limbs
201    correctly.  (Note however that the mpz code is limited to 2^32 bits on
202    32-bit hosts and 2^37 bits on 64-bit hosts.)
203
204  * Contains all fixes from release 5.1.3.
205
206  SPEEDUPS
207  * Plain division of large operands is faster and more monotonous in operand
208    size.
209
210  * Major speedup for ARM, in particular ARM Cortex-A15, thanks to improved
211    assembly.
212
213  * Major speedup for SPARC T4/T5 and speedup also for T3, thanks to a lot of
214    new assembly.
215
216  * Speedup for Intel Sandy Bridge, Ivy Bridge, Haswell, thanks to rewritten
217    and vastly expanded assembly support.  Speedup also for the older Core 2
218    and Nehalem.
219
220  * Faster mixed arithmetic between mpq_class and double.
221
222  * With g++, optimise more operations when one argument is a simple constant.
223
224  FEATURES
225  * Support for new Intel and AMD CPUs.
226
227  * Support for ARM64 alias Aarch64 alias ARMv8.
228
229  * New public functions mpn_sec_mul and mpn_sec_sqr, implementing side-channel
230    silent multiplication and squaring.
231
232  * New public functions mpn_sec_div_qr and mpn_sec_div_r, implementing
233    side-channel silent division.
234
235  * New public functions mpn_cnd_add_n and mpn_cnd_sub_n.  Side-channel silent
236    conditional addition and subtraction.
237
238  * New public function mpn_sec_powm, implementing side-channel silent modexp.
239
240  * New public function mpn_sec_invert, implementing side-channel silent
241    modular inversion.
242
243  * Better support for applications which use the mpz_t type, but nevertheless
244    need to call some of the lower-level mpn functions.  See the documentation
245    for mpz_limbs_read and related functions.
246
247  MISC
248  * This release will not work on NetBSD 5.x, FreeBSD 7.x, 8.x or 9 series
249    before 9.3.  The reason is that the m4 command is not correctly
250    implemented.  (Workaround: Use an older GMP release, or install GNU m4 from
251    /usr/ports and tell GMP to use it.)
252
253  * This release will not build properly on FreeBSD/amd64 before version 10
254    using the 32-bit ABI (once a working m4 is installed).  The reason is
255    broken limits.h.  (Workaround: Use an older GMP release if using the 32-bit
256    ABI on these FreeBSD releases is important.)
257
258  * This release will not work reliably on FreeBSD 10.0 for i386 or amd64 using
259    the 32-bit ABI.  The reason is bugs in the compiler 'clang'.  Depending on
260    CPU-dependent compiler flags, GMP may or may not be miscompiled in a
261    particular build.  (Workaround: Compiling gcc from /usr/ports should work,
262    except that gcc circularly depends on GMP; we have not been able to test
263    that workaround due to FreeBSD 10.0 bugs affecting its ability to run under
264    KVM and Xen.)
265
266  * This release will not compile on FreeBSD before version 10 for i386,
267    targeting any modern AMD processor.  The reason is bugs in the old gcc
268    bundled with FreeBSD.  (Workaround: install a less obsolete gcc from
269    /usr/ports and tell GMP to use it, or override the -march=amdfam10
270    GMP configure command line argument.)
271
272
273Changes between GMP version 5.1.2 and 5.1.3
274
275  BUGS FIXED
276  * The internal functions mpn_sbpi1_div_qr_sec mpn_sbpi1_div_r_sec could
277    compute garbage with a low probability.  They are now rewritten, and the
278    test code has been improved.
279
280  * A bug in the ia64 implementation of mpn_divrem_2, clobbering some
281    callee-save registers, has been fixed. This is an internal
282    function, with the bug manifesting itself as miscomputation in,
283    e.g., mpn_sqrtrem.
284
285  * The documentation now correctly says 'const' for input arguments.
286
287  SPEEDUPS
288  * None.
289
290  FEATURES
291  * None.
292
293  MISC
294  * None.
295
296
297Changes between GMP version 5.1.1 and 5.1.2
298
299  BUGS FIXED
300  * A bug in mpz_powm_ui triggered by base arguments of at least 15000 decimal
301    digits or mod arguments of at least 7500 decimal digits has been fixed.
302
303  * An AMD Bulldozer specific bug affecting the 64-bit Windows ABI has been
304    fixed.  This bug was in a key function (mpn_mul_1) and made both Bulldozer
305    specific builds and fat builds run on Bulldozer completely non-functional.
306
307  SPEEDUPS
308  * None.
309
310  FEATURES
311  * None.
312
313  MISC
314  * Fixes and generalisations to the test suite.
315
316  * Minor portability enhancements.
317
318
319Changes between GMP version 5.1.0 and 5.1.1
320
321  BUGS FIXED
322  * On Windows 64-bit, an error causing link errors about
323    __gmp_binvert_limb_table has been fixed.
324
325  * Aarch64 alias ARM64 support now works.
326
327  * A possible buffer overrun in mpz_ior has been fixed.
328
329  * A rare sign flip in mpz_remove has been fixed.
330
331  * A bug causing problems with mpf numbers with absolute value >= 2^31 has
332    been fixed.
333
334  * Several bugs in mini-gmp have been fixed.
335
336  * A bug caused by automake, related to the 'distcheck' target, has been fixed
337    by upgrading the automake used for GMP release engineering.
338
339  SPEEDUPS
340  * None.
341
342  FEATURES
343  * Preliminary support for the x32 ABI under x86-64.
344
345  MISC
346  * The mini-gmp testsuite now tests the entire set of functions.
347
348  * Various improvements of the GMP testsuite.
349
350
351Changes between GMP version 5.0.* and 5.1.0
352
353  BUGS FIXED
354  * When reading a C++ number (like mpz_class) in an istream reaches the end
355    of the stream, the eofbit is now set.
356
357  * The result sign of mpz_rootrem's remainder is now always correct.
358
359  * The mpz_remove function now handles negative divisors.
360
361  * Contains all fixes from release 5.0.5.
362
363  SPEEDUPS
364  * The n-factorial and n-over-k functions have been reimplemented for great
365    speedups for small and large operands.
366
367  * New subquadratic algorithm for the Kronecker/Jacobi/Legendre symbol.
368
369  * Major speedup for ARM, in particular ARM Cortex-A9 and A15, thanks to broad
370    assembly support.
371
372  * Significant speedup for POWER6 and POWER7 thanks to improved assembly.
373
374  * The performance under M$ Windows' 64-bit ABI has been greatly improved
375    thanks to complete assembly support.
376
377  * Minor speed improvements of many functions and for many platforms.
378
379  FEATURES
380  * Many new CPUs recognised.
381
382  * New functions for multi-factorials, and primorial: mpz_2fac_ui,
383    mpz_mfac_uiui and mpz_primorial_ui.
384
385  * The mpz_powm_sec function now uses side-channel silent division for
386    converting into Montgomery residues.
387
388  * The fat binary mechanism is now more robust in its CPU recognition.
389
390  MISC
391  * Inclusion of assembly code is now controlled by the configure options
392    --enable-assembly and --disable-assembly.  The "none" CPU target is gone.
393
394  * In C++, the conversions mpq_class->mpz_class, mpf_class->mpz_class and
395    mpf_class->mpq_class are now explicit.
396
397  * Includes "mini-gmp", a small, portable, but less efficient, implementation
398    of a subset of GMP's mpn and mpz interfaces. Used in GMP bootstrap, but it
399    can also be bundled with applications as a fallback when the real GMP
400    library is unavailable.
401
402  * The ABIs under AIX are no longer called aix32 and aix64, but mode64 and 32.
403    This is more consistent with other powerpc systems.
404
405  * The coverage of the testsuite has been improved, using the lcov tool.  See
406    also https://gmplib.org/devel/lcov/.
407
408  * It is now possible to compile GMP using a C++ compiler.
409
410  * K&R C compilers are no longer supported.
411
412  * The BSD MP compatibility functions have been removed.
413
414
415Changes between GMP version 5.0.4 and 5.0.5
416
417  BUGS FIXED
418  * A bug causing AMD 11h processors to be treated like AMD 10h has been fixed.
419    The 11h processors do not correctly handle all 10h (aka K10) instructions,
420    and GMP's use of these instructions results in major miscomputations (not
421    as one would have hoped CPU traps of some 'illegal instruction' sort).
422
423  * A bug affecting recent Intel Sandy Bridge CPUs resulting in configuration
424    failures has been fixed.
425
426  SPEEDUPS
427  * None.
428
429  FEATURES
430  * A couple of tests added to the self-check suite.
431
432  MISC
433  * None.
434
435
436Changes between GMP version 5.0.3 and 5.0.4
437
438  BUGS FIXED
439  * Thresholds in mpn_powm_sec for both fat and non-fat builds are now used
440    safely, plugging a one-word buffer overrun introduced in the 5.0.3 release
441    (for non-fat) and a multi-word buffer overrun that existed since 5.0 (for
442    fat).  (We have not been able to provoke malign stack smashing in any of
443    the ~100 configurations explored by the GMP nightly builds, but the bug
444    should be assumed to be exploitable.)
445
446  * Two bugs in multiplication code causing incorrect computation with
447    extremely low probability have been fixed.
448
449  * A bug in the test suite causing buffer overruns during "make check",
450    sometimes leading to subsequent malloc crashes, has been fixed.
451
452  * Two bugs in the gcd code have been fixed.  They could lead to incorrect
453    results, but for uniformly distributed random operands, the likelihood for
454    that is infinitesimally small.  (There was also a third bug, but that was
455    an incorrect ASSERT, which furthermore was not enabled by default.)
456
457  * A bug affecting 32-bit PowerPC division has been fixed.  The bug caused
458    miscomputation for certain divisors in the range 2^32 ... 2^64-1 (about 1
459    in 2^30 of these).
460
461  SPEEDUPS
462  * None, except indirectly through recognition of new CPUs, and through better
463    tuning parameters.
464
465  FEATURES
466  * Some more tests added to the self-check suite.
467
468  * The AMD "Bulldozer" CPU is now recognised.
469
470  MISC
471  * None.
472
473
474Changes between GMP version 5.0.2 and 5.0.3
475
476  BUGS FIXED
477  * A few minor bugs related to portability fixed.
478
479  * A slight timing leak of the powm_sec functions have been sealed.  (This
480    leak could possibly be used to extract the most significant few bits of the
481    exponent.  "Few" here means at most 10.)
482
483  * The mpz_nextprime function now runs a safer number of pseudo-random prime
484    tests.
485
486  * A bug in division code possibly causing incorrect computation was fixed.
487
488  SPEEDUPS
489  * None, except indirectly through recognition of new CPUs, and through better
490    tuning parameters.
491
492  FEATURES
493  * New CPUs recognised.
494
495  * IBM S/390 are now supported in both 31/32-bit and 64-bit mode.  (We have
496    not been able to fully test this on any multilib machine, since IBM expired
497    our guest account a few days before our release.)
498
499  MISC
500  * None.
501
502
503Changes between GMP version 5.0.1 and 5.0.2
504
505  BUGS FIXED
506  * Many minor bugs related to portability fixed.
507
508  * The support for HPPA 2.0N now works, after an assembly bug fix.
509
510  * A test case type error has been fixed.  The symptom of this bug was
511    spurious 'make check' failures.
512
513  SPEEDUPS
514  * None, except indirectly through recognition of new CPUs.
515
516  FEATURES
517  * Fat builds are now supported for 64-bit x86 processors also under Darwin.
518
519  MISC
520  * None.
521
522
523Changes between GMP version 5.0.0 and 5.0.1
524
525  BUGS FIXED
526  * Fat builds fixed.
527
528  * Fixed crash for huge multiplies when old FFT_TABLE2 type of parameter
529    selection tables' sentinel was smaller than multiplied operands.
530
531  * The solib numbers now reflect the removal of the documented but preliminary
532    mpn_bdivmod function; we correctly flag incompatibility with GMP 4.3.  GMP
533    5.0.0 has this wrong, and should perhaps be uninstalled to avoid confusion.
534
535  SPEEDUPS
536  * Multiplication of large numbers has indirectly been sped up through better
537    FFT tuning and processor recognition.  Since many operations depend on
538    multiplication, there will be a general speedup.
539
540  FEATURES
541  * More Core i3, i5 an Core i7 processor models are recognised.
542
543  * Fixes and workarounds for Mac OS quirks should make this GMP version build
544    using many of the different versions of "Xcode".
545
546  MISC
547  * The amount of scratch memory needed for multiplication of huge numbers has
548    been reduced substantially (but is still larger than in GMP 4.3.)
549
550  * Likewise, the amount of scratch memory needed for division of large numbers
551    has been reduced substantially.
552
553  * The FFT tuning code of tune/tuneup.c has been completely rewritten, and
554    new, large FFT parameter selection tables are provided for many machines.
555
556  * Upgraded to the latest autoconf, automake, libtool.
557
558
559Changes between GMP version 4.3.X and 5.0.0
560
561  BUGS FIXED
562  * None (contains the same fixes as release 4.3.2).
563
564  SPEEDUPS
565  * Multiplication has been overhauled:
566    (1) Multiplication of larger same size operands has been improved with
567        the addition of two new Toom functions and a new internal function
568        mpn_mulmod_bnm1 (computing U * V mod (B^n-1), B being the word base.
569        This latter function is used for the largest products, waiting for a
570        better Schoenhage-Strassen U * V mod (B^n+1) implementation.
571    (2) Likewise for squaring.
572    (3) Multiplication of different size operands has been improved with the
573        addition of many new Toom function, and by selecting underlying
574        functions better from the main multiply functions.
575
576  * Division and mod have been overhauled:
577    (1) Plain "schoolbook" division is reimplemented using faster quotient
578        approximation.
579    (2) Division Q = N/D, R = N mod D where both the quotient and remainder
580        are needed now runs in time O(M(log(N))).  This is an improvement of
581        a factor log(log(N))
582    (3) Division where just the quotient is needed is now O(M(log(Q))) on
583        average.
584    (4) Modulo operations using Montgomery REDC form now take time O(M(n)).
585    (5) Exact division Q = N/D by means of mpz_divexact has been improved
586        for all sizes, and now runs in time O(M(log(N))).
587
588  * The function mpz_powm is now faster for all sizes.  Its complexity has
589    gone from O(M(n)log(n)m) to O(M(n)m) where n is the size of the modulo
590    argument and m is the size of the exponent.  It is also radically
591    faster for even modulus, since it now partially factors such modulus
592    and performs two smaller modexp operations, then uses CRT.
593
594  * The internal support for multiplication yielding just the lower n limbs
595    has been improved by using Mulders' algorithm.
596
597  * Computation of inverses, both plain 1/N and 1/N mod B^n have been
598    improved by using well-tuned Newton iterations, and wrap-around
599    multiplication using mpn_mulmod_bnm1.
600
601  * A new algorithm makes mpz_perfect_power_p asymptotically faster.
602
603  * The function mpz_remove uses a much faster algorithm, is better tuned,
604    and also benefits from the division improvements.
605
606  * Intel Atom and VIA Nano specific optimisations.
607
608  * Plus hundreds of smaller improvements and tweaks!
609
610  FEATURES
611  * New mpz function: mpz_powm_sec for side-channel quiet modexp
612    computations.
613
614  * New mpn functions: mpn_sqr, mpn_and_n, mpn_ior_n, mpn_xor_n, mpn_nand_n,
615    mpn_nior_n, mpn_xnor_n, mpn_andn_n, mpn_iorn_n, mpn_com, mpn_neg,
616    mpn_copyi, mpn_copyd, mpn_zero.
617
618  * The function mpn_tdiv_qr now allows certain argument overlap.
619
620  * Support for fat binaries for 64-bit x86 processors has been added.
621
622  * A new type, mp_bitcnt_t for bignum bit counts, has been introduced.
623
624  * Support for Windows64 through mingw64 has been added.
625
626  * The cofactors of mpz_gcdext and mpn_gcdext are now more strictly
627    normalised, returning to how GMP 4.2 worked.  (Note that also release
628    4.3.2 has this change.)
629
630  MISC
631  * The mpn_mul function should no longer be used for squaring,
632    instead use the new mpn_sqr.
633
634  * The algorithm selection has been improved, the number of thresholds have
635    more than doubled, and the tuning and use of existing thresholds have
636    been improved.
637
638  * The tune/speed program can measure many of new functions.
639
640  * The mpn_bdivmod function has been removed.  We do not consider this an
641    incompatible change, since the function was marked as preliminary.
642
643  * The testsuite has been enhanced in various ways.
644
645
646Changes between GMP version 4.3.1 and 4.3.2
647
648  Bugs:
649  * Fixed bug in mpf_eq.
650  * Fixed overflow issues in mpz_set_str, mpz_inp_str, mpf_set_str, and
651    mpf_get_str.
652  * Avoid unbounded stack allocation for unbalanced multiplication.
653  * Fixed bug in FFT multiplication.
654
655  Speedups:
656  * None, except that proper processor recognition helps affected processors.
657
658  Features:
659  * Recognise more "Core 2" processor variants.
660  * The cofactors of mpz_gcdext and mpn_gcdext are now more strictly
661    normalised, returning to how GMP 4.2 worked.
662
663
664Changes between GMP version 4.3.0 and 4.3.1
665
666  Bugs:
667  * Fixed bug in mpn_gcdext, affecting also mpz_gcdext and mpz_invert.
668    The bug could cause a cofactor to have a leading zero limb, which
669    could lead to crashes or miscomputation later on.
670  * Fixed some minor documentation issues.
671
672  Speedups:
673  * None.
674
675  Features:
676  * Workarounds for various issues with Mac OS X's build tools.
677  * Recognise more IBM "POWER" processor variants.
678
679
680Changes between GMP version 4.2.X and 4.3.0
681
682  Bugs:
683  * Fixed bug in mpz_perfect_power_p with recognition of negative perfect
684    powers that can be written both as an even and odd power.
685  * We might accidentally have added bugs since there is a large amount of
686    new code in this release.
687
688  Speedups:
689  * Vastly improved assembly code for x86-64 processors from AMD and Intel.
690  * Major improvements also for many other processor families, such as
691    Alpha, PowerPC, and Itanium.
692  * New sub-quadratic mpn_gcd and mpn_gcdext, as well as improved basecase
693    gcd code.
694  * The multiply FFT code has been slightly improved.
695  * Balanced multiplication now uses 4-way Toom in addition to schoolbook,
696    Karatsuba, 3-way Toom, and FFT.
697  * Unbalanced multiplication has been vastly improved.
698  * Improved schoolbook division by means of faster quotient approximation.
699  * Several new algorithms for division and mod by single limbs, giving
700    many-fold speedups.
701  * Improved nth root computations.
702  * The mpz_nextprime function uses sieving and is much faster.
703  * Countless minor tweaks.
704
705  Features:
706  * Updated support for fat binaries for x86_32 include current processors
707  * Lots of new mpn internal interfaces.  Some of them will become public
708    in a future GMP release.
709  * Support for the 32-bit ABI under x86-apple-darwin.
710  * x86 CPU recognition code should now default better for future
711    processors.
712  * The experimental nails feature does not work in this release, but
713    it might be re-enabled in the future.
714
715  Misc:
716  * The gmp_version variable now always contains three parts.  For this
717    release, it is "4.3.0".
718
719
720Changes between GMP version 4.2.3 and 4.2.4
721
722  Bugs:
723  * Fix bug with parsing exponent '+' sign in mpf.
724  * Fix an allocation bug in mpf_set_str, also affecting mpf_init_set_str, and
725    mpf_inp_str.
726
727  Speedups:
728  * None, except that proper processor recognition helps affected processors.
729
730  Features:
731  * Recognize new AMD processors.
732
733
734Changes between GMP version 4.2.2 and 4.2.3
735
736  Bugs:
737  * Fix x86 CPU recognition code to properly identify recent AMD and Intel
738    64-bit processors.
739  * The >> operator of the C++ wrapper gmpxx.h now does floor rounding, not
740    truncation.
741  * Inline semantics now follow the C99 standard, and works with recent GCC
742    releases.
743  * C++ bitwise logical operations work for more types.
744  * For C++, gmp.h now includes cstdio, improving compiler compatibility.
745  * Bases > 36 now work properly in mpf_set_str.
746
747  Speedups:
748  * None, except that proper processor recognition helps affected processors.
749
750  Features:
751  * The allocation functions now detect overflow of the mpz_t type.  This means
752    that overflow will now cause an abort, except when the allocation
753    computation itself overflows.  (Such overflow can probably only happen in
754    powering functions; we will detect powering overflow in the future.)
755
756
757Changes between GMP version 4.2.1 and 4.2.2
758
759  * License is now LGPL version 3.
760
761  Bugs:
762  * Shared library numbers corrected for libcxx.
763  * Fixed serious bug in gmpxx.h where a=a+b*c would generate garbage.
764    Note that this only affects C++ programs.
765  * Fix crash in mpz_set_d for arguments with large negative exponent.
766  * Fix 32-bit ABI bug with Itanium assembly for popcount and hamdist.
767  * Fix assembly syntax problem for powerpc-ibm-aix with AIX native assembler.
768  * Fix problems with x86 --enable-fat, where the compiler where told to
769    generate code for the build machine, not plain i386 code as it should.
770  * Improved recognition of powerpc systems wrt Altivec/VMX capability.
771  * Misc minor fixes, mainly workarounds for compiler/assembler bugs.
772
773  Speedups:
774  * "Core 2" and Pentium 4 processors, running in 64-bit mode will get a
775     slight boost as they are now specifically recognized.
776
777  Features:
778  * New support for x86_64-solaris
779  * New, rudimentary support for x86-apple-darwin and x86_64-apple-darwin.
780    (Please see https://gmplib.org/macos.html for more information.)
781
782
783Changes between GMP version 4.2 and 4.2.1
784
785  Bugs:
786  * Shared library numbers corrected.
787  * Broken support for 32-bit AIX fixed.
788  * Misc minor fixes.
789
790  Speedups:
791  * Exact division (mpz_divexact) now falls back to plain division for large
792    operands.
793
794  Features:
795  * Support for some new systems.
796
797
798Changes between GMP version 4.1.4 and 4.2
799
800  Bugs:
801  * Minor bug fixes and code generalizations.
802  * Expanded and improved test suite.
803
804  Speedups:
805  * Many minor optimizations, too many to mention here.
806  * Division now always subquadratic.
807  * Computation of n-factorial much faster.
808  * Added basic x86-64 assembly code.
809  * Floating-point output is now subquadratic for all bases.
810  * FFT multiply code now about 25% faster.
811  * Toom3 multiply code faster.
812
813  Features:
814  * Much improved configure.
815  * Workarounds for many more compiler bugs.
816  * Temporary allocations are now made on the stack only if small.
817  * New systems supported: HPPA-2.0 gcc, IA-64 HP-UX, PowerPC-64 Darwin,
818    Sparc64 GNU/Linux.
819  * New i386 fat binaries, selecting optimised code at runtime (--enable-fat).
820  * New build option: --enable-profiling=instrument.
821  * New memory function: mp_get_memory_functions.
822  * New Mersenne Twister random numbers: gmp_randinit_mt, also now used for
823    gmp_randinit_default.
824  * New random functions: gmp_randinit_set, gmp_urandomb_ui, gmp_urandomm_ui.
825  * New integer functions: mpz_combit, mpz_rootrem.
826  * gmp_printf etc new type "M" for mp_limb_t.
827  * gmp_scanf and friends now accept C99 hex floats.
828  * Numeric input and output can now be in bases up to 62.
829  * Comparisons mpz_cmp_d, mpz_cmpabs_d, mpf_cmp_d recognise infinities.
830  * Conversions mpz_get_d, mpq_get_d, mpf_get_d truncate towards zero,
831    previously their behaviour was unspecified.
832  * Fixes for overflow issues with operands >= 2^31 bits.
833
834  Caveats:
835  * mpfr is gone, and will from now on be released only separately.  Please see
836    www.mpfr.org.
837
838
839Changes between GMP version 4.1.3 and 4.1.4
840
841* Bug fix to FFT multiplication code (crash for huge operands).
842* Bug fix to mpf_sub (miscomputation).
843* Support for powerpc64-gnu-linux.
844* Better support for AMD64 in 32-bit mode.
845* Upwardly binary compatible with 4.1.3, 4.1.2, 4.1.1, 4.1, 4.0.1, 4.0,
846  and 3.x versions.
847
848
849Changes between GMP version 4.1.2 and 4.1.3
850
851* Bug fix for FFT multiplication code (miscomputation).
852* Bug fix to K6 assembly code for gcd.
853* Bug fix to IA-64 assembly code for population count.
854* Portability improvements, most notably functional AMD64 support.
855* mpz_export allows NULL for countp parameter.
856* Many minor bug fixes.
857* mpz_export allows NULL for countp parameter.
858* Upwardly binary compatible with 4.1.2, 4.1.1, 4.1, 4.0.1, 4.0, and 3.x
859  versions.
860
861
862Changes between GMP version 4.1.1 and 4.1.2
863
864* Bug fixes.
865
866
867Changes between GMP version 4.1 and 4.1.1
868
869* Bug fixes.
870* New systems supported: NetBSD and OpenBSD sparc64.
871
872
873Changes between GMP version 4.0.1 and 4.1
874
875* Bug fixes.
876* Speed improvements.
877* Upwardly binary compatible with 4.0, 4.0.1, and 3.x versions.
878* Asymptotically fast conversion to/from strings (mpz, mpq, mpn levels), but
879  also major speed improvements for tiny operands.
880* mpn_get_str parameter restrictions relaxed.
881* Major speed improvements for HPPA 2.0 systems.
882* Major speed improvements for UltraSPARC systems.
883* Major speed improvements for IA-64 systems (but still sub-optimal code).
884* Extended test suite.
885* mpfr is back, with many bug fixes and portability improvements.
886* New function: mpz_ui_sub.
887* New functions: mpz_export, mpz_import.
888* Optimization for nth root functions (mpz_root, mpz_perfect_power_p).
889* Optimization for extended gcd (mpz_gcdext, mpz_invert, mpn_gcdext).
890* Generalized low-level number format, reserving a `nails' part of each
891  limb.  (Please note that this is really experimental; some functions
892  are likely to compute garbage when nails are enabled.)
893* Nails-enabled Alpha 21264 assembly code, allowing up to 75% better
894  performance.  (Use --enable-nails=4 to enable it.)
895
896
897Changes between GMP version 4.0 and 4.0.1
898
899* Bug fixes.
900
901
902Changes between GMP version 3.1.1 and 4.0
903
904* Bug fixes.
905* Speed improvements.
906* Upwardly binary compatible with 3.x versions.
907* New CPU support: IA-64, Pentium 4.
908* Improved CPU support: 21264, Cray vector systems.
909* Support for all MIPS ABIs: o32, n32, 64.
910* New systems supported: Darwin, SCO, Windows DLLs.
911* New divide-and-conquer square root algorithm.
912* New algorithms chapter in the manual.
913* New malloc reentrant temporary memory method.
914* New C++ class interface by Gerardo Ballabio (beta).
915* Revamped configure, featuring ABI selection.
916* Speed improvements for mpz_powm and mpz_powm_ui (mainly affecting small
917  operands).
918* mpz_perfect_power_p now properly recognizes 0, 1, and negative perfect
919  powers.
920* mpz_hamdist now supports negative operands.
921* mpz_jacobi now accepts non-positive denominators.
922* mpz_powm now supports negative exponents.
923* mpn_mul_1 operand overlap requirements relaxed.
924* Float input and output uses locale specific decimal point where available.
925* New gmp_printf, gmp_scanf and related functions.
926* New division functions: mpz_cdiv_q_2exp, mpz_cdiv_r_2exp, mpz_divexact_ui.
927* New divisibility tests: mpz_divisible_p, mpz_divisible_ui_p,
928  mpz_divisible_2exp_p, mpz_congruent_p, mpz_congruent_ui_p,
929  mpz_congruent_2exp_p.
930* New Fibonacci function: mpz_fib2_ui.
931* New Lucas number functions: mpz_lucnum_ui, mpz_lucnum2_ui.
932* Other new integer functions: mpz_cmp_d, mpz_cmpabs_d, mpz_get_d_2exp,
933  mpz_init2, mpz_kronecker, mpz_lcm_ui, mpz_realloc2.
934* New rational I/O: mpq_get_str, mpq_inp_str, mpq_out_str, mpq_set_str.
935* Other new rational functions: mpq_abs, mpq_cmp_si, mpq_div_2exp,
936  mpq_mul_2exp, mpq_set_f.
937* New float tests: mpf_integer_p, mpf_fits_sint_p, mpf_fits_slong_p,
938  mpf_fits_sshort_p, mpf_fits_uint_p, mpf_fits_ulong_p, mpf_fits_ushort_p.
939* Other new float functions: mpf_cmp_d, mpf_get_default_prec, mpf_get_si,
940  mpf_get_ui, mpf_get_d_2exp.
941* New random functions: gmp_randinit_default, gmp_randinit_lc_2exp_size.
942* New demo expression string parser (see demos/expr).
943* New preliminary perl interface (see demos/perl).
944* Tuned algorithm thresholds for many more CPUs.
945
946
947Changes between GMP version 3.1 and 3.1.1
948
949* Bug fixes for division (rare), mpf_get_str, FFT, and miscellaneous minor
950  things.
951
952
953Changes between GMP version 3.0 and 3.1
954
955* Bug fixes.
956* Improved `make check' running more tests.
957* Tuned algorithm cutoff points for many machines.  This will improve speed for
958  a lot of operations, in some cases by a large amount.
959* Major speed improvements: Alpha 21264.
960* Some speed improvements: Cray vector computers, AMD K6 and Athlon, Intel P5
961  and Pentium Pro/II/III.
962* The mpf_get_prec function now works as it did in GMP 2.
963* New utilities for auto-tuning and speed measuring.
964* Multiplication now optionally uses FFT for very large operands.  (To enable
965  it, pass --enable-fft to configure.)
966* Support for new systems: Solaris running on x86, FreeBSD 5, HP-UX 11, Cray
967  vector computers, Rhapsody, Nextstep/Openstep, MacOS.
968* Support for shared libraries on 32-bit HPPA.
969* New integer functions: mpz_mul_si, mpz_odd_p, mpz_even_p.
970* New Kronecker symbol functions: mpz_kronecker_si, mpz_kronecker_ui,
971  mpz_si_kronecker, mpz_ui_kronecker.
972* New rational functions: mpq_out_str, mpq_swap.
973* New float functions: mpf_swap.
974* New mpn functions: mpn_divexact_by3c, mpn_tdiv_qr.
975* New EXPERIMENTAL function layer for accurate floating-point arithmetic, mpfr.
976  To try it, pass --enable-mpfr to configure.  See the mpfr subdirectory for
977  more information; it is not documented in the main GMP manual.
978
979
980Changes between GMP version 3.0 and 3.0.1
981
982* Memory leaks in gmp_randinit and mpz_probab_prime_p fixed.
983* Documentation for gmp_randinit fixed.  Misc documentation errors fixed.
984
985
986Changes between GMP version 2.0 and 3.0
987
988* Source level compatibility with past releases (except mpn_gcd).
989* Bug fixes.
990* Much improved speed thanks to both host independent and host dependent
991  optimizations.
992* Switch to autoconf/automake/libtool.
993* Support for building libgmp as a shared library.
994* Multiplication and squaring using 3-way Toom-Cook.
995* Division using the Burnikel-Ziegler method.
996* New functions computing binomial coefficients: mpz_bin_ui, mpz_bin_uiui.
997* New function computing Fibonacci numbers: mpz_fib_ui.
998* New random number generators: mpf_urandomb, mpz_rrandomb, mpz_urandomb,
999  mpz_urandomm, gmp_randclear, gmp_randinit, gmp_randinit_lc_2exp,
1000  gmp_randseed, gmp_randseed_ui.
1001* New function for quickly extracting limbs: mpz_getlimbn.
1002* New functions performing integer size tests: mpz_fits_sint_p,
1003  mpz_fits_slong_p, mpz_fits_sshort_p, mpz_fits_uint_p, mpz_fits_ulong_p,
1004  mpz_fits_ushort_p.
1005* New mpf functions: mpf_ceil, mpf_floor, mpf_pow_ui, mpf_trunc.
1006* New mpq function: mpq_set_d.
1007* New mpz functions: mpz_addmul_ui, mpz_cmpabs, mpz_cmpabs_ui, mpz_lcm,
1008  mpz_nextprime, mpz_perfect_power_p, mpz_remove, mpz_root, mpz_swap,
1009  mpz_tdiv_ui, mpz_tstbit, mpz_xor.
1010* New mpn function: mpn_divexact_by3.
1011* New CPU support: DEC Alpha 21264, AMD K6 and Athlon, HPPA 2.0 and 64,
1012  Intel Pentium Pro and Pentium-II/III, Sparc 64, PowerPC 64.
1013* Almost 10 times faster mpz_invert and mpn_gcdext.
1014* The interface of mpn_gcd has changed.
1015* Better support for MIPS R4x000 and R5000 under Irix 6.
1016* Improved support for SPARCv8 and SPARCv9 processors.
1017
1018
1019Changes between GMP version 2.0 and 2.0.2
1020
1021* Many bug fixes.
1022
1023
1024Changes between GMP version 1.3.2 and 2.0
1025
1026* Division routines in the mpz class have changed.  There are three classes of
1027  functions, that rounds the quotient to -infinity, 0, and +infinity,
1028  respectively.  The first class of functions have names that begin with
1029  mpz_fdiv (f is short for floor), the second class' names begin with mpz_tdiv
1030  (t is short for trunc), and the third class' names begin with mpz_cdiv (c is
1031  short for ceil).
1032
1033  The old division routines beginning with mpz_m are similar to the new
1034  mpz_fdiv, with the exception that some of the new functions return useful
1035  values.
1036
1037  The old function names can still be used.  All the old functions names will
1038  now do floor division, not trunc division as some of them used to.  This was
1039  changed to make the functions more compatible with common mathematical
1040  practice.
1041
1042  The mpz_mod and mpz_mod_ui functions now compute the mathematical mod
1043  function.  I.e., the sign of the 2nd argument is ignored.
1044
1045* The mpq assignment functions do not canonicalize their results.  A new
1046  function, mpq_canonicalize must be called by the user if the result is not
1047  known to be canonical.
1048* The mpn functions are now documented.  These functions are intended for
1049  very time critical applications, or applications that need full control over
1050  memory allocation.  Note that the mpn interface is irregular and hard to
1051  use.
1052* New functions for arbitrary precision floating point arithmetic.  Names
1053  begin with `mpf_'.  Associated type mpf_t.
1054* New and improved mpz functions, including much faster GCD, fast exact
1055  division (mpz_divexact), bit scan (mpz_scan0 and mpz_scan1), and number
1056  theoretical functions like Jacobi (mpz_jacobi) and multiplicative inverse
1057  (mpz_invert).
1058* New variable types (mpz_t and mpq_t) are available that makes syntax of
1059  mpz and mpq calls nicer (no need for & before variables).  The MP_INT and
1060  MP_RAT types are still available for compatibility.
1061* Uses GNU configure.  This makes it possible to choose target architecture
1062  and CPU variant, and to compile into a separate object directory.
1063* Carefully optimized assembly for important inner loops.  Support for DEC
1064  Alpha, Amd 29000, HPPA 1.0 and 1.1, Intel Pentium and generic x86, Intel
1065  i960, Motorola MC68000, MC68020, MC88100, and MC88110, Motorola/IBM
1066  PowerPC, National NS32000, IBM POWER, MIPS R3000, R4000, SPARCv7,
1067  SuperSPARC, generic SPARCv8, and DEC VAX.  Some support also for ARM,
1068  Clipper, IBM ROMP (RT), and Pyramid AP/XP.
1069* Faster.  Thanks to the assembler code, new algorithms, and general tuning.
1070  In particular, the speed on machines without GCC is improved.
1071* Support for machines without alloca.
1072* Now under the LGPL.
1073
1074INCOMPATIBILITIES BETWEEN GMP 1 AND GMP 2
1075
1076* mpq assignment functions do not canonicalize their results.
1077* mpz division functions round differently.
1078* mpz mod functions now really compute mod.
1079* mpz_powm and mpz_powm_ui now really use mod for reduction.
1080