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