1 Copyright 1999-2023 Free Software Foundation, Inc. 2 Contributed by the AriC and Caramba projects, INRIA. 3 4 This file is part of the GNU MPFR Library. 5 6 The GNU MPFR Library is free software; you can redistribute it and/or modify 7 it under the terms of the GNU Lesser General Public License as published by 8 the Free Software Foundation; either version 3 of the License, or (at your 9 option) any later version. 10 11 The GNU MPFR Library is distributed in the hope that it will be useful, but 12 WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY 13 or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public 14 License for more details. 15 16 You should have received a copy of the GNU Lesser General Public License 17 along with the GNU MPFR Library; see the file COPYING.LESSER. If not, see 18 https://www.gnu.org/licenses/ or write to the Free Software Foundation, Inc., 19 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA. 20 21 22 23 Installing GNU MPFR 24 =================== 25 26 Note: In case of problem, please read this INSTALL file carefully before 27 reporting a bug, in particular Section "In case of problem" below. Some 28 problems are due to bad configuration on the user side (not specific to 29 MPFR). 30 31 0. You first need to install GMP. See <https://gmplib.org/>. 32 MPFR requires GMP version 5.0.0 or later. 33 You need a C compiler, preferably GCC, but any reasonable compiler should 34 work (C++ compilers should work too, under the condition that they do not 35 break type punning via union). 36 And you need the standard Unix "make" command, plus some other standard 37 Unix utility commands. 38 39 1. Extract the files from the archive. 40 41 2. It is strongly advised to apply the latest patches if this has 42 not been done yet and if patches are available. You can check 43 on the release page: 44 45 https://www.mpfr.org/mpfr-4.2.1/ 46 47 which may have additional information. The patches can be applied 48 with commands like: 49 50 wget --no-config https://www.mpfr.org/mpfr-4.2.1/allpatches 51 patch -N -Z -p1 < allpatches 52 53 or 54 55 curl https://www.mpfr.org/mpfr-4.2.1/allpatches | patch -N -Z -p1 56 57 (Those instructions are for the GNU patch command, for example 58 /usr/bin/gpatch on Solaris.) 59 60 3. In the MPFR directory, to detect your system, type: 61 ./configure 62 possibly with options (see below, in particular if this step or 63 one of the following fails). You should also check whether WARNING 64 lines have been output (such a problem may cause a failure in one 65 of the following steps). 66 Note: paths provided in configure options must always be absolute 67 (relative paths are not supported). 68 69 4. To build the library, type: 70 make 71 72 [optional] if you want to tune MPFR for your specific architecture, see 73 the section "Tuning MPFR" below. Note that for most common architectures, 74 MPFR includes some default tuning parameters which should be near from 75 optimal. 76 77 5. To check the built library (runs the test files), type: 78 make check 79 80 Note: If any test fails, information about this failure can be found in 81 the tests/test-suite.log file; you should provide this file in your bug 82 reports (in addition to other useful information, as mentioned later). 83 If you want the contents of this file to be automatically output in case 84 of failure, you can set the VERBOSE environment variable to 1 before 85 running "make check", for instance by typing: 86 VERBOSE=1 make check 87 88 6. [Optional / experimental] Binary distributions may also want to run: 89 make check-gmp-symbols 90 91 This will check that MPFR does not use GMP internal symbols, which 92 could yield failures in case of GMP upgrade without a MPFR rebuild. 93 But note that this is a heuristic and might give false positives or 94 false negatives. Please report any problem to the MPFR developers. 95 End users may also be interested in this check, unless they have 96 allowed GMP internals with configure options (see below). 97 98 7. To install it (default "/usr/local" | see "--prefix" option), type: 99 make install 100 101 If you installed MPFR (header and library) in directories that are 102 not searched by default by the compiler and/or linking tools, then, 103 like with other libraries, you may need to set up some environment 104 variables such as C_INCLUDE_PATH (to find the header mpfr.h), 105 LIBRARY_PATH (to find the library), and if the shared library has 106 been installed, LD_LIBRARY_PATH (before execution) or LD_RUN_PATH 107 (before linking); this list is not exhaustive and some environment 108 variables may be specific to your system. "make install" gives some 109 instructions; please read them. You can also find more information 110 in the manuals of your compiler and linker. The MPFR FAQ may also 111 give some information. 112 113 Remember that if you have several MPFR (or GMP) versions installed 114 (e.g., one with the system, and one, newer, by you), you will not 115 necessarily get a compilation/linking error if a wrong library is 116 used (e.g., because LD_LIBRARY_PATH has not been set correctly). 117 But unexpected results may occur. 118 119 Under Mac OS X, if the shared library was not installed and you use 120 Apple's linker (this is the default), you will also need to provide 121 the -search_paths_first linker flag ("-Wl,-search_paths_first" when 122 you link via gcc) to make sure that the right library is selected, 123 as by default, Apple's linker selects a shared library preferably, 124 even when it is farther in the library paths. We recall that if a 125 wrong library is selected due to this behavior, unexpected results 126 may occur. 127 128 129 Building the documentation 130 ========================== 131 132 To build the documentation in various formats, you may first need to 133 install recent versions of some utilities such as texinfo. 134 135 * Type "make info" to produce the documentation in the info format. 136 137 * Type "make pdf" to produce the documentation in the PDF format. 138 139 * Type "make dvi" to produce the documentation in the DVI format. 140 141 * Type "make ps" to produce the documentation in the PostScript format. 142 143 * Type "make html" to produce the documentation in the HTML format 144 (in several pages); if you want only one output HTML file, then 145 type "makeinfo --html --no-split mpfr.texi" from the doc directory 146 instead. 147 148 149 Building MPFR with internal GMP header files 150 ============================================ 151 152 MPFR built with internal GMP header files is a bit faster, so you may want 153 to build it with them. Just do this in step 1: 154 155 ./configure --with-gmp-build=GMPBUILD 156 157 where GMPBUILD is the GMP build directory. The needed header files are: 158 gmp-impl.h, longlong.h and all the necessary headers to use them, which 159 may be located either in the GMP source directory or in the GMP build 160 directory, in case they are different (MPFR takes care of that, as of 161 MPFR 3.1.0). 162 163 Warning: the library obtained in this way may use some internal GMP 164 symbols, and thus dynamically linking your software with a different 165 version of GMP might fail, even though it is declared as compatible 166 by Libtool's versioning system. 167 168 169 Tuning MPFR 170 =========== 171 172 [For the current GMP version (6.1.0), a Unix-like OS is required.] 173 174 For this, you need to build MPFR with a GMP build directory (see above). 175 In the GMP build directory, you also need to go into the "tune" subdirectory 176 and type "make speed". This will build the GMP speed library, which is used 177 by the MPFR tuning mechanism. 178 179 Then go back to the MPFR build directory, go into the "tune" subdirectory and 180 type "make tune". This will build an optimized file "mparam.h" for your 181 specific architecture. 182 183 184 ./configure options 185 =================== 186 187 --prefix=DIR installs MPFR headers and library in DIR/include and 188 DIR/lib respectively (the default is "/usr/local"). 189 190 --with-gmp-include=DIR assumes that DIR contains gmp.h 191 --with-gmp-lib=DIR assumes that DIR contains the GMP library 192 --with-gmp=DIR assumes that DIR is where you have installed GMP. 193 same as --with-gmp-lib=DIR/lib 194 and --with-gmp-include=DIR/include 195 (use either --with-gmp alone or one or both of 196 --with-gmp-lib/--with-gmp-include) 197 Warning! Do not use these options if you have 198 CPPFLAGS and/or LDFLAGS containing a -I or -L 199 option with a directory that contains a GMP 200 header or library file, as these options just 201 add -I and -L options to CPPFLAGS and LDFLAGS 202 *after* the ones that are currently declared, 203 so that DIR will have a lower precedence. Also, 204 this may not work if DIR is a system directory 205 (typically /usr or /usr/local); see below. 206 207 --with-gmp-build=DIR assumes that DIR contains the GMP build directory, 208 and enables the use of GMP internals (see above). 209 Warning! This option and the group of options 210 --with-gmp are mutually exclusive. 211 212 --enable-assert build MPFR with assertions. 213 214 --enable-thread-safe build MPFR as thread safe, using compiler-level 215 Thread-Local Storage (TLS). Note: TLS support is 216 roughly tested by configure. If configure detects 217 that TLS does not work (because of the compiler, 218 linker or system libraries), it will output an 219 error message, telling you to build MPFR without 220 thread safe. For instance, though Mac OS X uses 221 GCC, it may not currently support GCC's __thread 222 storage class. 223 224 --disable-thread-safe build MPFR without TLS. By default, TLS support 225 is detected automatically, and MPFR is built as 226 thread safe if supported. However, this detection 227 is only a heuristic: TLS can be detected as 228 supported while its support is incomplete or 229 buggy (MPFR tests may fail). In such a case, 230 this option is useful. 231 232 --enable-decimal-float build conversion functions from/to decimal floats. 233 Note that detection by the configure script is 234 limited in case of cross-compilation. 235 Accepted arguments: 236 yes Decimal support is requested and the configure 237 script fails if it detects that decimals do not 238 work. 239 The encoding (BID or DPD) will automatically be 240 detected at configure time or at compile time if 241 possible (if not, generic code will be used). 242 no Decimal support is explicitly disabled. 243 auto Decimal support is enabled if the configure script 244 detects that it works. This is the default when 245 --{enable,disable}-decimal-float is not given. 246 bid Decimal support is requested and the encoding is 247 assumed to be BID (some check may be done). 248 dpd Decimal support is requested and the encoding is 249 assumed to be DPD (some check may be done). 250 generic Decimal support is requested and the generic code 251 is used (mainly for developers). 252 253 --enable-gmp-internals allows the MPFR build to use GMP's undocumented 254 functions (not from the public API). Note that 255 library versioning is not guaranteed to work if 256 this option is used. Thus it must not be used in 257 binary distributions. 258 259 --with-sysroot=DIR Search for dependent libraries within DIR (which 260 may be useful in cross-compilation). If you use 261 this option, you need to have Libtool 2.4+ on 262 the target system. See Libtool 2.4+'s NEWS file. 263 264 Note: By default, the configure script tries to set CC / CFLAGS to GMP's 265 ones from gmp.h (__GMP_CC / __GMP_CFLAGS) in order to ensure that MPFR is 266 built with the same ABI as GMP. The reason is that when GMP is built, it 267 may set CC / CFLAGS to select an ABI that is not the default one in order 268 to have a better performance. The -pedantic option in GMP's CFLAGS, when 269 present (which is the case by default), is removed, because the MPFR 270 build system uses some C extensions (when this script detects that they 271 are supported) and -pedantic yields too many useless warnings. However, 272 this setting from GMP is not guaranteed to work as the configure script 273 does some compiler tests earlier, and a conflict may arise. Also, the 274 values obtained from GMP may be incorrect for the MPFR build if GMP has 275 been built on a different machine; in such a case, the user may need to 276 specify CC / CFLAGS, as explained below. 277 278 Moreover, even without --with-gmp-build and --enable-gmp-internals, 279 MPFR might use some GMP internals by mistake. This would be a bug, 280 which should be reported to the MPFR developers. 281 282 Run "./configure --help" to see the other options (default options 283 from Autoconf and Automake). 284 285 286 If 'gmp.h' and 'libgmp' do not match 287 ==================================== 288 289 Under some circumstances, the configure script may output a message 290 saying: 291 292 'gmp.h' and 'libgmp' seem to have different versions or 293 we cannot run a program linked with GMP (if you cannot 294 see the version numbers above). [...] 295 296 Even though the configure script does not fail in such a case, this 297 message most often indicates a real error, which needs to be avoided. 298 Possible causes are: 299 300 * The wanted GMP library does not have the same ABI as the one chosen 301 to build MPFR. The consequences may be: 302 303 - A different GMP library (with the correct ABI) has been found, 304 but does not have the same version as 'gmp.h'. 305 306 - No other GMP libraries have been found (in this case, no version 307 numbers should be printed above the warning message). 308 309 This is incorrect and one of the following steps (make, make check) 310 will probably fail. GMP (actually gmp.h) now provides CC and CFLAGS 311 information to select the correct ABI, so that this problem should 312 no longer occur; but if GMP was provided by a binary package, such 313 information may not be valid. See the 314 315 checking for CC and CFLAGS in gmp.h... 316 317 line in the configure output (about the 11th line) and the following 318 few ones for more information. You may need to reinstall GMP or to 319 provide your own CC and/or CFLAGS. See also the remaining of this 320 INSTALL file. 321 322 * A configure option like --with-gmp or --with-gmp-include was used 323 with a system include directory, e.g. one of the following: 324 325 --with-gmp=/usr 326 --with-gmp=/usr/local 327 --with-gmp-include=/usr/include 328 --with-gmp-include=/usr/local/include 329 330 GCC (and possibly other compilers) will ignore such a directive for 331 include directories (but this rule is not applied for the library 332 itself!). This means that the library search paths will be reordered 333 as declared, but the specified include directory will still be near 334 the end of the include search paths (thus with a low precedence). 335 This is not a problem if only one GMP version is installed, but 336 otherwise, a wrong gmp.h may be chosen, so that the versions of 337 gmp.h and libgmp may not match. The suggestions are the following: 338 339 - If you want to use the GMP version under /usr, then you should 340 uninstall all the other GMP versions (header and library files) 341 that may be seen in the search paths, in particular those under 342 /usr/local. 343 344 - If you want to use the GMP version under /usr/local, then you 345 should uninstall all the other GMP versions (header and library 346 files) that may be seen in the search paths, but *NOT* the one 347 under /usr (the version under /usr is provided by the OS vendor, 348 and changing/removing anything related to it may break your 349 system, and /usr should have a lower precedence than /usr/local 350 anyway). 351 352 To find where GMP versions have been installed: 353 354 $ locate libgmp (if you have a locate database) 355 356 and if the compiler is GCC: 357 358 $ gcc -print-file-name=libgmp.so (under most systems) 359 $ gcc -print-file-name=libgmp.dylib (under Mac OS X) 360 361 and if this does not work, you may try: 362 363 $ gcc -print-search-dirs 364 365 * An official GCC version was used under Debian GNU/Linux. Problems 366 may come from the fact that Debian chose a different convention 367 for library search paths concerning 32-bit and 64-bit libraries. 368 A possible problem can be, for instance: 369 370 [Debian's GCC, correct library path] 371 $ gcc -print-file-name=libgmp.so 372 /home/vlefevre/gmp/athlon64/lib/../lib/libgmp.so 373 374 [Official GCC, incorrect library path] 375 $ gcc-4.3.1 -print-file-name=libgmp.so 376 /usr/lib/../lib64/libgmp.so 377 378 The solution: use a GCC provided by Debian or add symbolic links 379 such as lib64 -> lib (on 64-bit machines) for your library paths. 380 381 * The problem may also be temporary and only due to the fact that 382 libtool could not be used at this time. This is unlikely, though. 383 384 385 In case of problem 386 ================== 387 388 First, look for any warning message in the configure output. 389 390 Several documents may help you to solve the problem: 391 * this INSTALL file, in particular information given below; 392 * the FAQ (either the FAQ.html file distributed with MPFR, or the 393 on-line version <https://www.mpfr.org/faq.html>, which may be more 394 up-to-date); 395 * the MPFR web page for this version <https://www.mpfr.org/mpfr-4.2.1/>, 396 which lists bugs found in this version and provides some patches. 397 398 If the "configure" fails, please check that the C compiler and its 399 options are the same as those for the GMP build (specially the ABI). 400 You can see the latter with the following command: 401 402 grep "^CC\|^CFLAGS" GMPBUILD/Makefile 403 404 if the GMP build directory is available. Then type: 405 406 ./configure <configure options> CC=<C compiler> CFLAGS="<compiler options>" 407 408 (quotes are needed when there are spaces or other special characters 409 in the CC/CFLAGS value) and continue the install. On some platforms, 410 you should provide further options to match those used by GMP, or set 411 some environment variables. For instance, see the "Notes on AIX/PowerPC" 412 section below. 413 414 Warning! Do NOT use optimization options that can change the semantics 415 of math operations, such as GCC's -ffast-math or Sun CC's -fast. 416 Otherwise conversions from/to double's may be incorrect on infinities, 417 NaN's and signed zeros. Since native FP arithmetic is used in a few 418 places only, such options would not make MPFR faster anyway. 419 420 If some "make check" tests fail, you can try the --disable-thread-safe 421 configure option (see the configure options above): it has been reported 422 that some platforms have buggy TLS support. Before trying this option, 423 you may want to check in the configure output whether MPFR was built 424 with TLS support; if yes, you will have a line: 425 checking for TLS support... yes 426 Alternatively "grep MPFR_USE_THREAD_SAFE config.log" will show that 427 MPFR_USE_THREAD_SAFE is defined to 1. If it is "no" (or the variable 428 is not defined), the --disable-thread-safe option would be useless. 429 430 Some tests failure may be due to other compiler bugs, in particular 431 in optimization code. You can try to build MPFR without compiler 432 optimizations by giving -O0 (letter O, digit 0) in CFLAGS. If the 433 MPFR tests no longer fail, this was probably due to a compiler bug, 434 though we cannot exclude a bug in MPFR. You may want to contact us 435 (see below), possibly after looking at: 436 https://members.loria.fr/PZimmermann/software/compilerbugs.html 437 438 On some platforms, try with "gmake" (GNU make) instead of "make". 439 Problems have been reported with the Tru64 make. 440 441 If the configure script reports that gmp.h version and libgmp version 442 are different, or if the build was OK, but the tests failed to link 443 with GMP or gave an error like 444 445 undefined reference to `__gmp_get_memory_functions' 446 447 meaning that the GMP library was not found or a wrong GMP library was 448 selected by the linker, then your library search paths are probably 449 not correctly set (some paths are missing or they are specified in an 450 incorrect order). 451 452 Such problems commonly occur under some GNU/Linux machines, where the 453 default header and library search paths may be inconsistent: GCC is 454 configured to search /usr/local/include and /usr/local/lib by default, 455 while the dynamic linker ignores /usr/local/lib. If you have a GMP 456 version installed in /usr (provided by the OS vendor) and a new one 457 installed in /usr/local, then the header of the new GMP version and 458 the library of the old GMP version will be used! The best solution 459 is to make sure that the dynamic linker configuration is consistent 460 with GCC's behavior, for instance by having /usr/local/lib in 461 /etc/ld.so.conf or in some file from /etc/ld.so.conf.d (as Debian 462 did: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=395177). See 463 also https://gcc.gnu.org/legacy-ml/gcc-help/2010-01/msg00171.html for 464 more information. Alternatively you can use: 465 * environment variables. This may sometimes be necessary. If DIR 466 is the installation directory of GMP, add DIR/include to your 467 CPATH or C_INCLUDE_PATH (for compilers other than GCC, please 468 check the manual of your compiler), and add DIR/lib to your 469 LIBRARY_PATH and LD_LIBRARY_PATH (and/or LD_RUN_PATH); 470 * --with-gmp* configure options (described above), e.g. 471 --with-gmp=/opt/local (to use /opt/local/include for headers and 472 /opt/local/lib for libraries), but other software that uses GMP 473 and/or MPFR will need correct paths too, and environment variables 474 allow one to set such search paths in a global way. 475 Note about "--with-gmp=/usr/local". This option may appear to 476 solve the above inconsistency problem, but does not work as you 477 expect. Indeed it affects the library search path, in particular, 478 the one used by the dynamic linker (thus adding the missing 479 /usr/local/lib directory as wanted), but since /usr/local/include 480 is a "standard system include directory" for GCC, the include 481 search patch is not changed; this is often not a problem in this 482 particular case because usually, /usr/local/include is already 483 last in the include search patch, but this may fail under some 484 occasions and may trigger obscure errors. 485 486 For instance, under Unix, where paths are separated by a colon: 487 488 * With POSIX sh-compatible shells (e.g. sh, ksh, bash, zsh): 489 export C_INCLUDE_PATH="/usr/local/include:/other/path/include" 490 export LIBRARY_PATH="/usr/local/lib:/other/path/lib" 491 export LD_LIBRARY_PATH="$LIBRARY_PATH" 492 493 * With csh or tcsh: 494 setenv C_INCLUDE_PATH "/usr/local/include:/other/path/include" 495 setenv LIBRARY_PATH "/usr/local/lib:/other/path/lib" 496 setenv LD_LIBRARY_PATH "$LIBRARY_PATH" 497 498 If almost all the tests fail and the messages in the test-suite.log file 499 (or in the output, when running individual tests from the command line) 500 start with a line of the form: 501 502 Incorrect MPFR version! (xxx header vs yyy library) 503 504 then this means that an installed MPFR version is tested instead of the 505 one that has just been built. This is probably not a bug in MPFR, but a 506 problem caused by the user or system configuration (particular options, 507 environment variables, etc.) or a bug in the toolchain. In particular, 508 if LD_LIBRARY_PATH overrides the run path (set up by libtool) and an 509 installed ABI-compatible version of MPFR is in a directory listed in 510 the LD_LIBRARY_PATH search path, then this will break. An example with 511 GNU ld: 512 513 https://sourceware.org/bugzilla/show_bug.cgi?id=21476 514 515 If you can't solve your problem, you should contact us via the MPFR 516 mailing-list <mpfr (a] inria.fr>, indicating the machine and operating system 517 used (uname -a), the compiler and version used (gcc -v if you use gcc), 518 the configure options used if any (including variables such as CC and 519 CFLAGS), the version of GMP and MPFR used, and a description of the 520 problem encountered, in particular the tests/test-suite.log file if 521 "make check" failed. Please send us also the log of the "configure" 522 (config.log). 523 524 525 Notes about ABI 526 =============== 527 528 On 64-bit computers, it may happen that the default ABI (Application Binary 529 Interface) chosen by MPFR does not correspond to the default one chosen by 530 the compiler. 531 532 In particular, this kind of message may indicate the problem: 533 /usr/bin/ld: skipping incompatible mpfr/src/.libs/libmpfr.a when searching for -lmpfr 534 535 In fact, since MPFR relies on GMP, it uses the same ABI as GMP. 536 To be sure that your program compiles and links correctly, use the same 537 compiler flags as MPFR does (look for CFLAGS in config.log). 538 539 You might also recompile GMP with a different ABI, with for example 540 ./configure ABI=32. 541 542 543 Notes on Mac OS X 544 ================= 545 546 If you get an error of the form 547 548 ld: pointer in read-only segment not allowed in slidable image... 549 550 this can mean that the link is done against a static (GMP) library. 551 In such a case, you should configure MPFR with --disable-shared to 552 disable the build of the shared library. 553 554 555 Notes on FreeBSD 4.3 556 ==================== 557 558 FreeBSD 4.3 is provided with an incorrect <float.h> header file, and 559 MPFR tests related to long double's may fail. If you cannot upgrade 560 the system, you can still use MPFR with FreeBSD 4.3, but you should 561 not use conversions with the long double type. 562 563 564 Notes on AIX/PowerPC 565 ==================== 566 567 The following has been tested on AIX 7.1.3 (gcc111.fsffrance.org) 568 with gcc 4.8.1 and GMP 6.1.0. 569 570 If GMP was built with the 64-bit ABI, before building and testing MPFR, 571 you should set the OBJECT_MODE environment variable to 64, e.g., with: 572 573 export OBJECT_MODE=64 574 575 (in a sh-compatible shell). Alternatively add the following to the configure 576 line: AR="ar -X64" NM="nm -B -X64". 577 578 579 Notes on Solaris 580 ================ 581 582 Do not put a -R<dir> option in the LD_OPTIONS environment variable, at least 583 if the directory <dir> can contain an MPFR library. Otherwise this MPFR 584 library may be chosen for the tests (make check) instead of the one that has 585 just been built, in which case, either you will get errors due to unmatched 586 versions or this problem may remain undetected. The reason is that this 587 option will appear before the -R options added by libtool, such as the one 588 to the src/.libs directory containing the MPFR library that has just been 589 built, and will have the precedence. 590 591 592 MPFR for use with Windows Applications 593 ====================================== 594 595 There are several ways of building MPFR on Windows, the most appropriate 596 approach depending on how you intend to use the resulting libraries. 597 598 a. Using MinGW 599 ============== 600 601 1 - We advise to use Mingw-w64 (https://www.mingw-w64.org/), which is 602 simpler and less demanding than Cygwin. Contrary to Cygwin, it also 603 provides native Windows code. 604 605 If you also use MSYS, you should use "make" for MSYS instead of 606 the "make" utility from Mingw-w64 or from GCC, which causes the 607 following error: 608 609 libtool: warning: libobj name 'extract.Tpo -c -o extract.lo extract.lo' 610 may not contain shell special characters. 611 rm: unknown option -- c 612 613 References about this issue and solution: 614 https://sourceforge.net/p/msys2/tickets/223/ 615 https://sympa.inria.fr/sympa/arc/mpfr/2016-07/msg00009.html 616 617 2 - If you just want to make a binary with gcc, there is nothing to do: 618 GMP, MPFR and the program compile exactly as under Linux. (It is 619 recommended to pass --build=xxx-yyy-mingw64 to the GMP configure command, 620 or --build=xxx with xxx containing mingw.) 621 622 3 - To avoid using the Microsoft runtime (which might not be conform to ISO C), 623 you can use the MinGW runtime package (which is an integral part of MinGW). 624 625 With MinGW version v8.0.0 and later, the formatted output functions 626 (printf, etc.) are ISO/POSIX-conforming by default; however, this is 627 no longer true if -std=c89 is used at build time. Conversely, with 628 earlier MinGW versions, it is possible to get conforming functions 629 with either '-ansi', '-posix' or '-D__USE_MINGW_ANSI_STDIO'. Note 630 that if there is a conformity mismatch between the options used for 631 the GMP build (from which the MPFR build gets the output) and those 632 used for the MPFR tests, the tsprintf test may fail with one of the 633 errors below. Be careful that a non-conforming output may yield a 634 buffer overflow. 635 636 The following applies to old MinGW versions, and may be discouraged 637 with recent MinGW versions. 638 639 In order to have the MPFR formatted output functions based on an 640 ISO-conforming printf(), you need to compile GMP (not MPFR) with 641 CC="gcc -D__USE_MINGW_ANSI_STDIO" (since the standard printf modifiers 642 %e, %Ld and %td are passed to GMP). 643 644 To avoid failures of some of the printf-related tests, MPFR needs to 645 be compiled with the same __USE_MINGW_ANSI_STDIO as with GMP, i.e. 646 this macro should be defined for both or undefined for both; this 647 should be the case by default, unless CC or CFLAGS has been redefined. 648 For instance, if __USE_MINGW_ANSI_STDIO is defined in GMP but not in 649 MPFR, the following error may occur: 650 651 Error in mpfr_vsprintf (s, "%e", ...); 652 expected: "-1.250000e+000" 653 got: "-1.250000e+00" 654 FAIL tsprintf.exe (exit status: 1) 655 656 and in the opposite case: 657 658 Error in mpfr_vsprintf (s, "%e", ...); 659 expected: "-1.250000e+00" 660 got: "-1.250000e+000" 661 FAIL tsprintf.exe (exit status: 1) 662 663 Note with old GMP versions: Other issues could arise due to the fact 664 that the C functions vsnprintf and vsprintf both used internally in 665 old GMP versions do not produce the same output: 666 667 https://sympa.inria.fr/sympa/arc/mpfr/2016-03/msg00045.html 668 https://sympa.inria.fr/sympa/arc/mpfr/2016-03/msg00051.html 669 https://sympa.inria.fr/sympa/arc/mpfr/2016-03/msg00053.html 670 671 If __USE_MINGW_ANSI_STDIO has not been defined when building GMP, 672 then the length modifiers j, L and t are not supported with the GMP 673 formatted output functions, and as a consequence, also with MPFR. 674 This is automatically detected by the configure script, except when 675 cross-compiling (e.g. under Linux), in which case some macros need 676 to be defined explicitly, e.g. with 677 678 "CPPFLAGS=-DNPRINTF_J -DNPRINTF_L -DNPRINTF_T" 679 680 in order to avoid potential issues with the MPFR library and failures 681 in the test suite (the corresponding tests are disabled explicitly). 682 683 b. Using Cygwin 684 =============== 685 686 This build should be similar to that for MinGW except that the resulting 687 library depends on the Cygwin DLL and cannot therefore be used with 688 Visual Studio as with MinGW. Indeed, the binaries compiled with Cygwin 689 require a dynamic library (cygwin.dll) to work; there is a Cygwin option 690 -mno-cygwin to build native code, but it may require some non-portable tricks. 691 692 In case of failure, you may need to pass LDFLAGS='-shared-libgcc' at the 693 end of the configure line due to a bug in GCC. Otherwise, if threading 694 support is not needed, you can configure MPFR with --disable-thread-safe. 695 696 c. Using Microsoft Visual C++ and Intel C++ Compilers 697 ===================================================== 698 699 Static and dynamic MPFR libraries for the 32- and 64-bit versions of 700 Windows can be built with Microsoft Visual Studio 2015 using the 701 Microsoft Visual C++ compiler, see: 702 703 https://visualstudio.microsoft.com/ 704 705 The Intel C++ compiler provided as a part of Intel Parallel Studio XE 706 can also be used: 707 708 https://software.intel.com/en-us/parallel-studio-xe 709 710 The relevant build projects are available here: 711 712 https://github.com/BrianGladman 713 714 d. Using the CompCert compiler 715 ============================== 716 717 [Tested with CompCert 3.10 and MPFR master-11992-f75b0c388 on x86_64-linux] 718 719 CompCert (https://compcert.org/) is a formally verified compiler. 720 To compile MPFR with CompCert: 721 722 $ ./configure --disable-shared CC=ccomp CFLAGS="-flongdouble -fstruct-passing" 723 724 You also need to unset LD_LIBRARY_PATH, and/or you might need to change 725 wl="" into wl="-Wl," in the libtool file (after running configure). 726 727 All tests (make check) should pass (tget_set_d64, tget_set_d128 and 728 tset_float128 are skipped, since CompCert does not support _Decimal64, 729 _Decimal128 nor _Float128). 730 731 e. Using the Intel OneApi compiler 732 ================================== 733 734 When using the Intel OneApi compiler (icx), one should add -fp-model=strict 735 to CFLAGS so that the conversion routines from/to native floating-point 736 types (float, double, ...) work properly. Otherwise some tests will fail. 737
Indexes created Wed Apr 29 00:23:26 UTC 2026