MPFR version 2.2.0 (released on 20 September 2005)

MPFR 2.2.0 requires GMP 4.1.0 or higher.

Download

The MPFR 2.2.0 source can be downloaded in the following archive formats:

• mpfr-2.2.0.tar.bz2 (tar/bzip2, 712 KB),
• mpfr-2.2.0.tar.gz (tar/gzip, 923 KB),
• mpfr-2.2.0.zip (zip, 1212 KB).
• Source RPM: mpfr-2.2.0-1.src.rpm (927 KB).

Note that the patches below have not been applied. You are strongly advised to apply the cumulative patch after extracting the MPFR archive.

MPFR is also available via third-party packages and ports. In particular:

• Official Debian packages libmpfr-dev, libmpfr1 (shared library) and source package. Patches up to 8 have been applied in version 2.2.0.dfsg.1-4. Patches up to 12 have been applied in version 2.2.0.dfsg.1-7.
• Gentoo MPFR packages. Patches up to 8 have been applied in version mpfr-2.2.0_p8. Patches up to 10 have been applied in version mpfr-2.2.0_p10. Patches up to 12 have been applied in version mpfr-2.2.0_p12. Patches up to 14 have been applied in version mpfr-2.2.0_p14.

Documentation

• The reference manual: [html] [dvi] [ps] [pdf]
• The reference manual translated into Japanese: [html] [pdf] [sources]
• The MPFR FAQ (common to all MPFR versions).

Bugs

The bugs listed below have been fixed (latest update: 2006-08-28). The following instructions are for Linux and may be similar for other operating systems. You can apply the patches in several ways:

• Apply the cumulative patch on the 2.2.0 release or a partially patched version with the command

  patch -N -Z -p1 < path_to_patches_file

  from the build directory. The -N option allows the patch command to ignore patches already applied. The -Z option sets the modification time of the patched files from time stamps given in the patch file, thus avoiding the need of some development utilities (such as autoconf). Under Microsoft Windows, in addition to the above options, you may need the --binary option (this depends whether the end-of-lines have been converted when you downloaded the patch file). Also note that the --dry-run option does not work with this cumulative patch.
• Or apply the individual patches below.
• Or checkout the 2.2 branch from the Subversion repository, e.g. with the command

  svn co svn://scm.gforge.inria.fr/svn/mpfr/branches/2.2 mpfr-2.2

  In this case, as said on the Gforge page, you should read the README.dev file for details on how to build MPFR. Also note that the branch may contain additional minor fixes.

Note: These patches are in unified diff format and some vendors' patch commands cannot deal with them; in this case, please use GNU patch (it may already be installed on your system, e.g. as /usr/local/bin/patch or gpatch).

Fixed bugs:

1. The computation of lngamma(2) yields an infinite loop or an assertion failure (this special case is not tested). This bug is fixed by the lngamma(2) patch (with testcase).
2. With gcc, the function mpfr_set_ui is also implemented by a macro, and when the second argument is the constant 0, the first argument is evaluated twice and the third argument is not evaluated at all; this may lead to a visible bug when the first and/or third arguments have side effects. The mpfr_set_si macro is also affected as it uses the mpfr_set_ui macro when the integer is the constant 0. This problem is fixed by the mpfr_set_ui macro patch (with testcase).
3. At the time of the MPFR 2.2.0 release, there was a bug in the libtool used to generate the MPFR configure script, preventing the creation of shared libraries on Darwin (Mac OS X). This script can be fixed by the configure patch (thanks to Dave Morrison for this very simple patch); alternatively, you may want to run autoreconf if you have the necessary tools.
4. One of the tests uses the fabs function, but -lm is not given when the executable is generated; this makes the linking fail under HP-UX. This problem is fixed by the tpow.c patch (the fabs function is no longer used).
5. Several problems fixed by patch 5:
   • Rounding in mpfr_hypot can be incorrect when the rounding mode is to nearest, x is much larger than y, and x is the middle of two consecutive numbers in the target precision (the patch provides a testcase for this bug).
   • Some cases where a computed value to be rounded to nearest is the middle of two consecutive numbers in the target precision are not optimized, with potential problems in case of intermediate underflow/overflow (the patch involved changes in internal macros, which were necessary to correct the mpfr_hypot bug).
   • Moved undocumented cache-related structure definition and function declarations out of the mpfr.h header file, to prevent the programmer from using them by mistake. Indeed it is planned to add these functions to the API in the future, but the current code is not working yet for every value (thus future programs using these functions should not be compiled with this version of MPFR).
   A bit more information about this patch is available in b86f4a9d (r3949).
6. The mpfr_div function where the divisor has a much smaller precision than the dividend is very slow and takes much more memory than necessary: the full-size division is performed, instead of a division of linear complexity (assuming the divisor has a bounded size). This is fixed by the small-divisor patch. Note that this efficiency bug was introduced after the rewrite of the division code for this version of MPFR, so that previous MPFR versions are not affected by this problem.
7. The mpfr_sin function fails to round correctly on some worst cases when rounding away from zero (on the tested worst cases: GMP_RNDU rounding mode for positive numbers between 0 and π). This bug, found on 2005-12-08, was introduced on 2005-01-29 in revision 3248, therefore MPFR versions previous to 2.2.0 are not affected. For MPFR 2.2.0, this bug is fixed by the sin.c patch.
8. The mpfr_get_f and mpfr_set_f functions are incorrect on the following cases: mpfr_get_f gives a wrong value when the result has a negative exponent and mpfr_set_f does not support overflows. These bugs are fixed by the mpfr_get_f/mpfr_set_f patch (which also potentially fixes other type problems and provides testcases).
9. The mpfr_random2 function does not change the sign of the destination number when the size is 0; therefore it can return either -0 or +0, instead of +0 all the time. This bug is fixed by the mpfr_random2 patch (with testcase).
10. MPFR cannot be compiled with some compilers due to empty macro arguments (allowed as of C99 only). This problem is fixed by the macro argument patch.
11. The division has a potential bug: the code is incorrect, but we do not know if there are values that can yield an incorrect result (this is not obvious and we have not tried to search for such values yet). This is fixed by the division patch. This patch also removes symbols that were stealing GMP's namespace and all code related to mpn_sub_nc (that led to build errors in some cases, when a new version of GMP was installed, but MPFR was not recompiled).
12. The mpfr_get_f function has another bug, not fixed by the mpfr_get_f/mpfr_set_f patch. This one is fixed by the mpfr_get_f patch (with testcase).

13. If -I/-L options are given to CPPFLAGS/LDFLAGS and the --with-gmp configure option (or a similar one) is used, then directories may be added in an inconsistent order to CPPFLAGS and LDFLAGS at configure time. If several GMP versions are installed, then this bug can lead to compilation or linking errors (such as undefined reference to __gmp_get_memory_functions), or perhaps runtime errors (such as incorrect results). This bug is fixed by the --with-gmp patch.

    Note that even with this patch, you may still get a warning saying that gmp.h and libgmp seem to have different versions or you cannot run a program linked with GMP. The test is currently not reliable enough in practice and the warning may be confusing. The MPFR library should still be built correctly, though. Also, you may want to read this new version of the INSTALL file, as some information about such paths has been added.

14. Darwin (Mac OS X) users should apply the -search_paths_first patch. It fixes a problem with Apple's linker when several GMP versions are installed and for one of them, only the static library is installed.
15. Patch 15 fixes the following bugs and adds some testcases.
    • Incorrect mpfr_log1p and mpfr_expm1 values for small negative input values (mpfr_expm1 may also be incorrect on some small positive numbers).
    • The mpfr_expm1 function is very slow (several hours) and may take several dozens of megabytes (gigabytes?) on large negative values, as the exact result is very close to -1; this particular case is taken into account by the patch.
    • Infinite loop in the mpfr_exp2 function on some large non-integer values (due to undetected overflow).
    • When the exponent range is such that the minimum exponent is larger than 1 (i.e., the value 1 is not representable), the mpfr_exp2 function gives an incorrect result on integers; in practice, this should never happen, but anyway, this is a bug.
    • The global inexact flag is not set in several functions.
    • In lngamma.c: The inexact flag is sometimes set when the result is NaN (though NaN is always regarded as an exact result). When the exponent range is very small, the result may be incorrect.
    • Correction in source code style to be able to use the ansi2knr utility, for pre-ANSI compilers (not tested, perhaps no longer used).
16. With the previous patch, the texp2 test fails on 64-bit machines. This problem is fixed by the texp2 patch (apply the previous patch first if you haven't done this yet). Note: this is a bug in the tests, not in the library itself (and some old test code has been fixed here, not just code introduced by the previous patch).

You can also download the new mpfr-longlong.h file from the 2.2 branch. In particular, it is needed with GMP 4.2.1 under HP-UX to avoid unsatisfied symbols errors about __gmpn_umul_ppmm.

The following bugs are fixed in MPFR 2.2.1:

• Major efficiency problems (that could lead to infinite loops and even crashes) in some subdomains for mpfr_csch, mpfr_sech, mpfr_coth, mpfr_eint, mpfr_erfc and mpfr_zeta.
• In mpfr_pow_z (thus mpfr_pow too, when mpfr_pow_z is called) and mpfr_pow_si, the overflow and underflow flags are not always set. Example: x = 0.11 in base 2, y = 1.0e38, rounding-to-nearest mode.
• Bugs in mpfr_pow and mpfr_exp.
• A bug in mpfr_set_d concerning signed/unsigned types; it should have no effects with most/all current compilers.
• Possible problems in mpfr_strtofr with some specific locales or compilers.
• The test of logging support is suboptimal.
• The configure.in file contains code that had no longer worked for a long time, e.g. the --with-irix64 option code. As a consequence, this configure option no longer exists in future MPFR versions.
• The test on the GMP header and library versions in configure is not reliable enough. It is greatly improved in future MPFR versions.

In addition to the bugs mentioned in the BUGS file:

• There are possible bugs with huge precisions (greater than 2³⁰) and with exponent ranges that do not allow to represent some numbers, such as 1 or 1/2.
• Wrong code may be generated on some platforms (e.g., s390). A patch that should improve the current behavior has been applied to the trunk and to the 2.2 branch. However, for technical reasons, we cannot provide a patch here. If you are concerned by this bug, then please use the Subversion repository on InriaForge if you can (you will need the autotools as explained in the README.dev file); alternatively, if your compiler is gcc, you can compile with -DHAVE_ATTRIBUTE_MODE by adding CPPFLAGS="-DHAVE_ATTRIBUTE_MODE" to the configure line (as explained by ./configure --help).
• If you get the following error:

  Error in pow_si(x,x,-2) for x=2

  this may be a bug in your compiler, thus not an MPFR bug. Such a bug has been found in GCC preversions and in Sun's compiler for Solaris/x86 (when optimizations are enabled, which is the default when building MPFR).

Changes from versions 2.1.* to version 2.2.0

• Bug fixes.
• New functions mpfr_set_overflow, mpfr_set_underflow, mpfr_set_inexflag, mpfr_set_erangeflag, mpfr_set_nanflag, mpfr_erfc, mpfr_atan2, mpfr_pow_z, mpfr_subnormalize, mpfr_const_catalan, mpfr_sec, mpfr_csc, mpfr_cot, mpfr_root, mpfr_eint, mpfr_get_f, mpfr_sech, mpfr_csch, mpfr_coth, mpfr_lngamma.
• New macro: MPFR_VERSION_STRING.
• The exported MPFR variables have been removed from mpfr.h (they were undocumented, so programs which respect the API still work).
• Grep CC and CFLAGS from GMP's Makefile if possible.
• Math functions are faster (both average and worst cases).
• Better support for long double.
• Shared library of MPFR.
• Binary compatible with previous versions if you do not use undocumented features.
• Thread safe (if built with --enable-thread-safe).
• Logging facility.
• Change in the semantics of mpfr_out_str / mpfr_get_str when n_digits = 0.
• Better locale support.

Platforms Known to Support MPFR

MPFR 2.2.0 has been successfully compiled and checked (with --enable-assert=full, in general) on the following platforms:

• alphaev6-unknown-linux-gnu (Alpha, gcc 3.3.5) with patches 1 to 4,
• alphaev67-unknown-freebsd5.4 (Alpha, gcc 3.4.2) with patches 1 to 4,
• alphaev68-dec-osf5.1 (Alpha, gcc 3.2.1),
• armv5tel-unknown-linux-gnu (XScale-PXA255, gcc 2.95.2),
• hppa2.0w-hp-hpux11.23 (PA-RISC 8900, HP cc) with patches 1 to 4 (patch 4 is necessary),
• i386-unknown-freebsd5.4 (Intel Pentium III, gcc 3.4.2) with patches 1 to 4,
• i386-unknown-netbsdelf2.0 (Intel Pentium III, gcc 3.3.3) with patches 1 to 4,
• i686-pc-linux-gnu (AMD Athlon, gcc 4.0.1),
• i686-pc-linux-gnu (Intel Pentium 4, gcc 4.0.2),
• i686-pc-linux-gnu (Intel Pentium M, gcc 4.0.1),
• ia64-unknown-linux-gnu (Itanium, gcc 3.2.3),
• MingGW32_NT_5.1 (Intel Pentium M, gcc 3.4.2),
• mips-sgi-irix6.5 (MIPSpro 7.3.1.3m),
• powerpc-apple-darwin8.2.0 (G5, gcc 4.0.0),
• powerpc-unknown-linux-gnu (G4, gcc 4.0.1),
• powerpc64-ibm-aix5.2.0.0,
• sparc-sun-solaris2.7 (UltraSPARC-II, Sun WorkShop 6 2000/04/07 C 5.1),
• x86_64-unknown-linux-gnu (AMD Opteron, gcc 3.3.5).

Back to the MPFR page.