1<?xml version="1.0" encoding="iso-8859-1"?> 2<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Strict//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd"> 3 4<!-- 5Copyright 2000-2023 Free Software Foundation, Inc. 6Contributed by the AriC and Caramba projects, INRIA. 7 8This file is part of the GNU MPFR Library. 9 10The GNU MPFR Library is free software; you can redistribute it and/or modify 11it under the terms of the GNU Lesser General Public License as published by 12the Free Software Foundation; either version 3 of the License, or (at your 13option) any later version. 14 15The GNU MPFR Library is distributed in the hope that it will be useful, but 16WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY 17or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public 18License for more details. 19 20You should have received a copy of the GNU Lesser General Public License 21along with the GNU MPFR Library; see the file COPYING.LESSER. If not, see 22https://www.gnu.org/licenses/ or write to the Free Software Foundation, Inc., 2351 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA. 24--> 25 26<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"> 27 28<head><meta http-equiv="Content-Type" content="text/html; charset=iso-8859-1" /> 29<title>Frequently Asked Questions about GNU MPFR</title> 30<style type="text/css">/*<![CDATA[*/ 31/* Global stylesheet for visual media */ 32 33html, body 34{ 35 background: white; 36 color: black; 37} 38 39div.logo { float: right } 40div.logo img { border: 0 } 41 42div.footer img { border: 0 } 43 44dt 45{ 46 margin-top: 2ex; 47 margin-bottom: 1ex; 48 font-weight: bolder; 49} 50 51/* For testing: dd { background: #ddddff } */ 52 53table { margin: 0.5ex auto } 54 55li { margin-top: 0.5ex; margin-bottom: 0.5ex } 56 57dd + dd 58{ 59 margin-top: 0; 60 margin-bottom: 0; 61 padding-top: 0.5ex; 62 padding-bottom: 0; 63} 64 65li > p, dd > p 66{ 67 margin-top: 0; 68 margin-bottom: 0; 69 padding-top: 0.5ex; 70 padding-bottom: 0.5ex; 71} 72 73.block-code code, code.block-code, 74.block-code samp, samp.block-code 75{ 76 display: block; 77 padding: 0.5ex 0; 78 margin-left: 2em; 79} 80 81.highlight { background: #dddddd } 82 83.nowrap { white-space: nowrap } 84 85/* 86dl.faq { counter-reset: faq } 87 88dl.faq dt:before 89{ 90 counter-increment: faq; 91 content: counter(faq) ". "; 92} 93*/ 94 95dl.faq dt { background: #dddddd } 96 97dl.faq dd 98{ 99 border-left: 4px solid; 100 border-color: transparent; 101 margin-left: 0em; 102 padding-left: 1.5em; 103} 104 105dl.faq dt:target + dd 106{ 107 border-left-color: #aaaaaa; 108} 109 110var.env { font-style: normal } 111/*]]>*/</style> 112</head> 113 114<body> 115 116<h1>Frequently Asked Questions about <cite><acronym>GNU</acronym> <acronym>MPFR</acronym></cite></h1> 117 118<p><strong>Important notice: Problems with a particular version of 119<cite><acronym>MPFR</acronym></cite> are discussed in the corresponding 120bugs page.</strong></p> 121 122<p>The latest version of this <acronym>FAQ</acronym> is available at 123<a href="https://www.mpfr.org/faq.html">https://www.mpfr.org/faq.html</a>. 124Please look at this version if possible.</p> 125 126<ol> 127<li><a href="#mpfr_vs_mpf">What are the differences between 128<cite><acronym>MPF</acronym></cite> from <cite><acronym>GMP</acronym></cite> 129and <cite><acronym>MPFR</acronym></cite>?</a></li> 130<li><a href="#mpf2mpfr">How to convert my program written using 131<cite><acronym>MPF</acronym></cite> to 132<cite><acronym>MPFR</acronym></cite>?</a></li> 133<li><a href="#no_libgmp">At configure time, I get the error: <q>libgmp not found or uses a different ABI.</q></a></li> 134<li><a href="#undef_ref1">I get undefined reference to <code>__gmp_get_memory_functions</code>.</a></li> 135<li><a href="#undef_ref2">When I link my program with 136<cite><acronym>MPFR</acronym></cite>, I get undefined reference 137to <code>__gmpXXXX</code>.</a></li> 138<li><a href="#crash_high_prec">My program crashes with high precisions.</a></li> 139<li><a href="#accuracy">Though I have increased the precision, the results 140are not more accurate.</a></li> 141<li><a href="#detect_mpfr">How can I detect <cite><acronym>MPFR</acronym></cite> 142installation using <cite>autoconf</cite> or <cite>pkg-config</cite>?</a></li> 143<li><a href="#cite">How to cite <cite><acronym>MPFR</acronym></cite> in a 144scientific publication?</a></li> 145<li><a href="#fpic">When I build <cite><acronym>MPFR</acronym></cite>, I get 146an error asking me to recompile with <samp>-fPIC</samp>.</a></li> 147</ol> 148 149<dl class="faq"> 150 151<dt id="mpfr_vs_mpf">1. What are the differences between 152<cite><acronym>MPF</acronym></cite> from <cite><acronym>GMP</acronym></cite> 153and <cite><acronym>MPFR</acronym></cite>?</dt> 154 155<dd><p>The main differences are:</p> 156<ul> 157<li><p>The precision of a <cite><acronym>MPFR</acronym></cite> variable 158is the <em>exact</em> number of bits used for its mantissa, whereas in 159<cite><acronym>MPF</acronym></cite>, the precision requested by the user 160is a minimum value (<cite><acronym>MPF</acronym></cite> generally uses a 161higher precision). With the additional difference below, this implies that 162the <cite><acronym>MPFR</acronym></cite> results do not depend on the 163number of bits (16, 32, 64 or more) of the underlying architecture.</p></li> 164<li><p>As a consequence, <cite><acronym>MPFR</acronym></cite> uses a 165base-2 exponent, whereas in <cite><acronym>MPF</acronym></cite>, this 166is a base-2<sup>32</sup> or base-2<sup>64</sup> exponent, depending on 167the limb size. For this reason (and other internal ones), the maximum 168exponent range in <cite><acronym>MPFR</acronym></cite> is different 169(and smaller, if the exponent is represented by the same type as in 170<cite><acronym>MPF</acronym></cite>).</p></li> 171<li><p><cite><acronym>MPFR</acronym></cite> provides an additional rounding 172mode argument to its functions; furthermore, it is guaranteed that the 173result of any operation is the nearest possible floating-point value from 174the exact result (considering the input variables as exact values), taking 175into account the precision of the destination variable and the rounding 176mode. <cite><acronym>MPFR</acronym></cite> also says whether the rounded 177result is above or below the exact result.</p></li> 178<li><p><cite><acronym>MPFR</acronym></cite> supports much more functions 179(in particular transcendental functions such as exponentials, logarithms, 180trigonometric functions and so on) and special values: signed zeros, 181infinities, not-a-number (NaN).</p></li> 182</ul></dd> 183 184<dt id="mpf2mpfr">2. How to convert my program written using 185<cite><acronym>MPF</acronym></cite> to 186<cite><acronym>MPFR</acronym></cite>?</dt> 187 188<dd><p>You need to add <q><code>r</code></q> to the function names, and to 189specify the rounding mode (<code>MPFR_RNDN</code> for rounding to nearest, 190<code>MPFR_RNDZ</code> for rounding toward zero, <code>MPFR_RNDU</code> 191for rounding toward positive infinity, <code>MPFR_RNDD</code> for rounding 192toward negative infinity). You can also define macros as follows: 193<code class="block-code">#define mpf_add(a, b, c) mpfr_add(a, b, c, MPFR_RNDN)</code></p> 194<p>The header file <samp>mpf2mpfr.h</samp> from the 195<cite><acronym>MPFR</acronym></cite> distribution automatically 196redefines all <cite><acronym>MPF</acronym></cite> functions in this 197way, using the default <cite><acronym>MPFR</acronym></cite> rounding 198mode. Thus you simply need to add the following line in all your files 199using <cite><acronym>MPF</acronym></cite> functions: 200<code class="block-code">#include <mpf2mpfr.h></code> 201just after the <samp>gmp.h</samp> and <samp>mpfr.h</samp> 202header files. If the program uses <cite><acronym>MPF</acronym></cite> 203internals (such as direct access to <code>__mpf_struct</code> members), 204additional changes will be needed.</p></dd> 205 206<dt id="no_libgmp">3. At configure time, I get the error: <q>libgmp not found or uses a different ABI.</q></dt> 207 208<dd><p>This test (<samp>checking for __gmpz_init in -lgmp</samp>) comes 209after the <samp>gmp.h</samp> detection. The failure occurs either because 210the <cite><acronym>GMP</acronym></cite> library could not be found 211(as it is not in the provided library search paths) or because the 212<cite><acronym>GMP</acronym></cite> library that was found does not have 213the expected <acronym title="Application Binary Interface">ABI</acronym> 214(<abbr>e.g.</abbr> 32-bit <abbr>vs</abbr> 64-bit). The former problem can be 215due to the fact that a static build of <cite><acronym>MPFR</acronym></cite> 216was requested while only a shared <cite><acronym>GMP</acronym></cite> library 217is installed (or the opposite, but another error can also show up in this 218case, see the <a href="#fpic">question about <samp>-fPIC</samp></a>). The 219latter problem can have several causes:</p> 220<ul> 221<li>A wrong libgmp library has been picked up. This can occur if you have 222several <cite><acronym>GMP</acronym></cite> versions installed on the 223machine and something is wrong with the provided library search paths.</li> 224<li>Wrong compiler options (<samp>CFLAGS</samp>) were given. In general, the 225presence or absence of the <samp>-m64</samp> compiler option must match the 226library <acronym title="Application Binary Interface">ABI</acronym>.</li> 227<li>A wrong <samp>gmp.h</samp> file has been picked up (if you have several 228<cite><acronym>GMP</acronym></cite> versions installed). Indeed, by default, 229<cite><acronym>MPFR</acronym></cite> gets the compiler options from the 230<samp>gmp.h</samp> file (with <cite><acronym>GMP</acronym></cite> 4.2.3 231or later); this is needed because <cite><acronym>GMP</acronym></cite> does 232not necessarily use the default <acronym>ABI</acronym>. The consequence is 233that if the <samp>gmp.h</samp> file is associated with a library using a 234different <acronym>ABI</acronym>, the <acronym>ABI</acronym>-related options 235will be incorrect. Hence the failure.</li> 236</ul> 237<p>Note: The <samp>config.log</samp> output gives more information 238than the error message. In particular, see the output of the test: 239<samp>checking for CC and CFLAGS in gmp.h</samp>; it should give you 240the default compiler options (from <samp>gmp.h</samp>).</p> 241 242<p>See also the answer to the <a href="#undef_ref1">next question</a>.</p></dd> 243 244<dt id="undef_ref1">4. I get undefined reference to <code>__gmp_get_memory_functions</code>.</dt> 245 246<dd><p>Note: this was mainly a problem when upgrading from 247<cite><acronym>GMP</acronym></cite> 4.1.4 to a later version, 248but information given below may still be useful in other cases, 249when several <cite><acronym>GMP</acronym></cite> libraries are 250installed on the same machine.</p> 251 252<p>If you get such an error, in particular when running 253<samp>make check</samp>, then this probably means that you are using 254the header file from <cite><acronym>GMP</acronym></cite> 4.2.x but the 255<cite><acronym>GMP</acronym></cite> 4.1.4 library. This can happen if 256several <cite><acronym>GMP</acronym></cite> versions are installed on 257your machine (<abbr>e.g.</abbr>, one provided by the system in 258<samp>/usr/{include,lib}</samp> and a new one installed by the owner or 259administrator of the machine in <samp>/usr/local/{include,lib}</samp>) 260and your include and library search paths are inconsistent. On various 261<acronym>GNU</acronym>/Linux machines, this is unfortunately the case 262by default (<samp>/usr/local/include</samp> is in the default include 263search path, but <samp>/usr/local/lib</samp> is <em>not</em> in the 264default library search path). Typical errors are: 265<samp class="block-code">undefined reference to `__gmp_get_memory_functions'</samp> 266in <samp>make check</samp>. The best solution is to add 267<samp>/usr/local/include</samp> to your <var class="env">C_INCLUDE_PATH</var> 268environment variable and to add <samp>/usr/local/lib</samp> to your 269<var class="env">LIBRARY_PATH</var> and <var class="env">LD_LIBRARY_PATH</var> 270environment variables (and/or <var class="env">LD_RUN_PATH</var>). 271Alternatively, you can use <samp>--with-gmp*</samp> configure options, 272<abbr>e.g.</abbr> <samp>--with-gmp=/usr/local</samp>, but <strong>this is 273not guaranteed to work</strong> (in particular with <samp>gcc</samp> and 274system directories such as <samp>/usr</samp> or <samp>/usr/local</samp>), 275and other software that uses <cite><acronym>GMP</acronym></cite> and/or 276<cite><acronym>MPFR</acronym></cite> will need correct paths too; 277environment variables allow you to set them in a global way.</p> 278<p>Other information can be given in the <samp>INSTALL</samp> file and 279<samp>ld</samp> manual. Please look at them for more details. See also 280the <a href="#undef_ref2">next question</a>.</p></dd> 281 282<dt id="undef_ref2">5. When I link my program with 283<cite><acronym>MPFR</acronym></cite>, I get undefined reference 284to <code>__gmpXXXX</code>.</dt> 285 286<dd><p>Link your program with <cite><acronym>GMP</acronym></cite>. Assuming 287that your program is <samp>foo.c</samp>, you should link it using: 288<samp class="block-code">cc link.c -lmpfr -lgmp</samp> 289<cite><acronym>MPFR</acronym></cite> library reference (<samp>-lmpfr</samp>) 290should be before <cite><acronym>GMP</acronym></cite>'s one 291(<samp>-lgmp</samp>). Another solution is, with <acronym>GNU</acronym> 292<samp>ld</samp>, to give all the libraries in any order inside a group: 293<samp class="block-code">gcc link.c -Wl,--start-group libgmp.a libmpfr.a -Wl,--end-group</samp> 294with the full path to the libraries, or 295<samp class="block-code">gcc link.c -Wl,--start-group -l:libgmp.a -l:libmpfr.a -Wl,--end-group</samp> 296to use the library path. 297See <samp>INSTALL</samp> file and <samp>ld</samp> manual for more 298details.</p> 299<p>If you used correct link options, but still get an error, this may mean 300that your include and library search paths are inconsistent. Please see the 301<a href="#undef_ref1">previous question</a>.</p></dd> 302 303<dt id="crash_high_prec">6. My program crashes with high precisions.</dt> 304 305<dd><p>Your stack size limit may be too small; indeed, by default, 306<cite><acronym>GMP</acronym></cite> 4.1.4 and below allocates all 307temporary results on the stack, and in very high precisions, this 308limit may be reached. You can solve this problem in different ways:</p> 309<ul> 310<li><p>You can upgrade to <cite><acronym>GMP</acronym></cite> 4.2 (or above), 311which now makes temporary allocations on the stack only when they are 312small.</p></li> 313<li><p>You can increase the stack size limit with the <samp>limit</samp>, 314<samp>unlimit</samp> or <samp>ulimit</samp> command, depending on your 315shell. This may fail on some systems, where the maximum stack size cannot 316be increased above some value.</p></li> 317<li><p>You can rebuild both <cite><acronym>GMP</acronym></cite> and 318<cite><acronym>MPFR</acronym></cite> to use another allocation method.</p></li> 319</ul></dd> 320 321<dt id="accuracy">7. Though I have increased the precision, the results 322are not more accurate.</dt> 323 324<dd><p>The reason may be the use of C floating-point numbers. If you want 325to store a floating-point constant to a <code>mpfr_t</code>, you should use 326<code>mpfr_set_str</code> (or one of the <cite><acronym>MPFR</acronym></cite> 327constant functions, such as <code>mpfr_const_pi</code> for π) instead 328of <code>mpfr_set_d</code> or <code>mpfr_set_ld</code>. Otherwise the 329floating-point constant will be first converted into a reduced-precision 330(<abbr>e.g.</abbr>, 53-bit) binary number before 331<cite><acronym>MPFR</acronym></cite> can work with it. This is the case 332in particular for most exact decimal numbers, such as 0.17, which are 333not exactly representable in binary.</p> 334<p>Also remember that <cite><acronym>MPFR</acronym></cite> does not track 335the accuracy of the results: copying a value <var>x</var> to <var>y</var> 336with <code>mpfr_set (y, x, MPFR_RNDN)</code> where the variable <var>y</var> 337is more precise than the variable <var>x</var> will not make it more 338accurate; the (binary) value will remain unchanged.</p></dd> 339 340<dt id="detect_mpfr">8. How can I detect <cite><acronym>MPFR</acronym></cite> 341installation using <cite>autoconf</cite> or <cite>pkg-config</cite>?</dt> 342 343<dd><p>The <cite><acronym>MPFR</acronym></cite> team does not currently 344recommend any <cite>autoconf</cite> code, but a section will later 345be added to the <cite><acronym>MPFR</acronym></cite> manual. 346Limited <cite>pkg-config</cite> support has been added for 347<cite><acronym>MPFR</acronym></cite>�4.0.0; example:</p> 348<pre style="margin-left: 2em">cc myprogram.c $(pkg-config --cflags --libs mpfr)</pre></dd> 349 350<dt id="cite">9. How to cite <cite><acronym>MPFR</acronym></cite> in a 351scientific publication?</dt> 352 353<dd><p>To properly cite <cite><acronym>MPFR</acronym></cite> in a scientific 354publication, please cite the 355<a href="https://doi.org/10.1145/1236463.1236468"><acronym title="Association for Computing Machinery">ACM</acronym> 356<acronym title="Transactions on Mathematical Software">TOMS</acronym> 357paper</a> 358and/or the library web page 359<a href="https://www.mpfr.org/">https://www.mpfr.org</a>. If your publication 360is related to a particular release of <cite><acronym>MPFR</acronym></cite>, 361for example if you report timings, please also indicate the release number 362for future reference.</p></dd> 363 364<dt id="fpic">10. When I build <cite><acronym>MPFR</acronym></cite>, I get 365an error asking me to recompile with <samp>-fPIC</samp>.</dt> 366 367<dd><p>A typical error looks like:</p> 368<p><tt>/usr/bin/ld: <em>/path/to/</em>libgmp.a(realloc.o): relocation 369R_X86_64_32 against `.rodata.str1.1' can not be used when making a 370shared object; recompile with -fPIC<br /> 371<em>/path/to/</em>libgmp.a: could not read symbols: Bad value<br /> 372collect2: ld returned 1 exit status</tt></p> 373<p>The probable reason is that you tried to build 374<cite><acronym>MPFR</acronym></cite> with the shared library enabled (this 375is the default), while only a static <cite><acronym>GMP</acronym></cite> 376library could be found. To solve this problem, either rebuild and reinstall 377<cite><acronym>GMP</acronym></cite> without the <samp>--disable-shared</samp> 378configure option, or configure <cite><acronym>MPFR</acronym></cite> with 379<samp>--disable-shared</samp>. If you did this and still get the above 380error, the cause may be conflicting <cite><acronym>GMP</acronym></cite> 381versions installed on your system; please check that your search path 382settings are correct.</p> 383<p>Additional note about the last sentence: Under <acronym>GNU</acronym>/Linux 384(for instance), the linker takes the first library found in the library search 385path, whether it is dynamic or static. The default behavior under darwin is 386different, but <cite><acronym>MPFR</acronym></cite> will change it.</p></dd> 387<!-- Reference concerning darwin: see MPFR_LD_SEARCH_PATHS_FIRST 388 in MPFR's configure.{ac,in} and acinclude.m4 --> 389 390</dl> 391 392</body> 393 394</html> 395