libstdc++-v3/doc/html/manual/internals.html

   1 <html><head><meta http-equiv="Content-Type" content="text/html; charset=UTF-8"><title>Porting to New Hardware or Operating Systems</title><meta name="generator" content="DocBook XSL-NS Stylesheets V1.76.1"><meta name="keywords" content="
   2       ISO C++
   3     ,
   4       internals
   5     "><meta name="keywords" content="
   6       ISO C++
   7     ,
   8       library
   9     "><meta name="keywords" content="
  10       ISO C++
  11     ,
  12       runtime
  13     ,
  14       library
  15     "><link rel="home" href="../index.html" title="The GNU C++ Library"><link rel="up" href="appendix_porting.html" title="Appendix B.  Porting and Maintenance"><link rel="prev" href="documentation_hacking.html" title="Writing and Generating Documentation"><link rel="next" href="test.html" title="Test"></head><body bgcolor="white" text="black" link="#0000FF" vlink="#840084" alink="#0000FF"><div class="navheader"><table width="100%" summary="Navigation header"><tr><th colspan="3" align="center">Porting to New Hardware or Operating Systems</th></tr><tr><td width="20%" align="left"><a accesskey="p" href="documentation_hacking.html">Prev</a> </td><th width="60%" align="center">Appendix B.
  16   Porting and Maintenance
  17
  18 </th><td width="20%" align="right"> <a accesskey="n" href="test.html">Next</a></td></tr></table><hr></div><div class="section" title="Porting to New Hardware or Operating Systems"><div class="titlepage"><div><div><h2 class="title" style="clear: both"><a name="appendix.porting.internals"></a>Porting to New Hardware or Operating Systems</h2></div></div></div><p>
  19 </p><p>This document explains how to port libstdc++ (the GNU C++ library) to
  20 a new target.
  21 </p><p>In order to make the GNU C++ library (libstdc++) work with a new
  22 target, you must edit some configuration files and provide some new
  23 header files.  Unless this is done, libstdc++ will use generic
  24 settings which may not be correct for your target; even if they are
  25 correct, they will likely be inefficient.
  26    </p><p>Before you get started, make sure that you have a working C library on
  27 your target.  The C library need not precisely comply with any
  28 particular standard, but should generally conform to the requirements
  29 imposed by the ANSI/ISO standard.
  30    </p><p>In addition, you should try to verify that the C++ compiler generally
  31 works.  It is difficult to test the C++ compiler without a working
  32 library, but you should at least try some minimal test cases.
  33    </p><p>(Note that what we think of as a "target," the library refers to as
  34 a "host."  The comment at the top of <code class="code">configure.ac</code> explains why.)
  35    </p><div class="section" title="Operating System"><div class="titlepage"><div><div><h3 class="title"><a name="internals.os"></a>Operating System</h3></div></div></div><p>If you are porting to a new operating system (as opposed to a new chip
  36 using an existing operating system), you will need to create a new
  37 directory in the <code class="code">config/os</code> hierarchy.  For example, the IRIX
  38 configuration files are all in <code class="code">config/os/irix</code>.  There is no set
  39 way to organize the OS configuration directory.  For example,
  40 <code class="code">config/os/solaris/solaris-2.6</code> and
  41 <code class="code">config/os/solaris/solaris-2.7</code> are used as configuration
  42 directories for these two versions of Solaris.  On the other hand, both
  43 Solaris 2.7 and Solaris 2.8 use the <code class="code">config/os/solaris/solaris-2.7</code>
  44 directory.  The important information is that there needs to be a
  45 directory under <code class="code">config/os</code> to store the files for your operating
  46 system.
  47 </p><p>You might have to change the <code class="code">configure.host</code> file to ensure that
  48 your new directory is activated.  Look for the switch statement that sets
  49 <code class="code">os_include_dir</code>, and add a pattern to handle your operating system
  50 if the default will not suffice.  The switch statement switches on only
  51 the OS portion of the standard target triplet; e.g., the <code class="code">solaris2.8</code>
  52 in <code class="code">sparc-sun-solaris2.8</code>.  If the new directory is named after the
  53 OS portion of the triplet (the default), then nothing needs to be changed.
  54    </p><p>The first file to create in this directory, should be called
  55 <code class="code">os_defines.h</code>.  This file contains basic macro definitions
  56 that are required to allow the C++ library to work with your C library.
  57    </p><p>Several libstdc++ source files unconditionally define the macro
  58 <code class="code">_POSIX_SOURCE</code>.  On many systems, defining this macro causes
  59 large portions of the C library header files to be eliminated
  60 at preprocessing time.  Therefore, you may have to <code class="code">#undef</code> this
  61 macro, or define other macros (like <code class="code">_LARGEFILE_SOURCE</code> or
  62 <code class="code">__EXTENSIONS__</code>).  You won't know what macros to define or
  63 undefine at this point; you'll have to try compiling the library and
  64 seeing what goes wrong.  If you see errors about calling functions
  65 that have not been declared, look in your C library headers to see if
  66 the functions are declared there, and then figure out what macros you
  67 need to define.  You will need to add them to the
  68 <code class="code">CPLUSPLUS_CPP_SPEC</code> macro in the GCC configuration file for your
  69 target.  It will not work to simply define these macros in
  70 <code class="code">os_defines.h</code>.
  71    </p><p>At this time, there are a few libstdc++-specific macros which may be
  72 defined:
  73    </p><p><code class="code">_GLIBCXX_USE_C99_CHECK</code> may be defined to 1 to check C99
  74 function declarations (which are not covered by specialization below)
  75 found in system headers against versions found in the library headers
  76 derived from the standard.
  77    </p><p><code class="code">_GLIBCXX_USE_C99_DYNAMIC</code> may be defined to an expression that
  78 yields 0 if and only if the system headers are exposing proper support
  79 for C99 functions (which are not covered by specialization below).  If
  80 defined, it must be 0 while bootstrapping the compiler/rebuilding the
  81 library.
  82    </p><p><code class="code">_GLIBCXX_USE_C99_LONG_LONG_CHECK</code> may be defined to 1 to check
  83 the set of C99 long long function declarations found in system headers
  84 against versions found in the library headers derived from the
  85 standard.
  86
  87    </p><p><code class="code">_GLIBCXX_USE_C99_LONG_LONG_DYNAMIC</code> may be defined to an
  88 expression that yields 0 if and only if the system headers are
  89 exposing proper support for the set of C99 long long functions.  If
  90 defined, it must be 0 while bootstrapping the compiler/rebuilding the
  91 library.
  92    </p><p><code class="code">_GLIBCXX_USE_C99_FP_MACROS_DYNAMIC</code> may be defined to an
  93 expression that yields 0 if and only if the system headers
  94 are exposing proper support for the related set of macros.  If defined,
  95 it must be 0 while bootstrapping the compiler/rebuilding the library.
  96    </p><p><code class="code">_GLIBCXX_USE_C99_FLOAT_TRANSCENDENTALS_CHECK</code> may be defined
  97 to 1 to check the related set of function declarations found in system
  98 headers against versions found in the library headers derived from
  99 the standard.
 100    </p><p><code class="code">_GLIBCXX_USE_C99_FLOAT_TRANSCENDENTALS_DYNAMIC</code> may be defined
 101 to an expression that yields 0 if and only if the system headers
 102 are exposing proper support for the related set of functions.  If defined,
 103 it must be 0 while bootstrapping the compiler/rebuilding the library.
 104    </p><p>Finally, you should bracket the entire file in an include-guard, like
 105 this:
 106    </p><pre class="programlisting">
 107
 108 #ifndef _GLIBCXX_OS_DEFINES
 109 #define _GLIBCXX_OS_DEFINES
 110 ...
 111 #endif
 112 </pre><p>We recommend copying an existing <code class="code">os_defines.h</code> to use as a
 113 starting point.
 114    </p></div><div class="section" title="CPU"><div class="titlepage"><div><div><h3 class="title"><a name="internals.cpu"></a>CPU</h3></div></div></div><p>If you are porting to a new chip (as opposed to a new operating system
 115 running on an existing chip), you will need to create a new directory in the
 116 <code class="code">config/cpu</code> hierarchy.  Much like the <a class="link" href="internals.html#internals.os" title="Operating System">Operating system</a> setup,
 117 there are no strict rules on how to organize the CPU configuration
 118 directory, but careful naming choices will allow the configury to find your
 119 setup files without explicit help.
 120 </p><p>We recommend that for a target triplet <code class="code">&lt;CPU&gt;-&lt;vendor&gt;-&lt;OS&gt;</code>, you
 121 name your configuration directory <code class="code">config/cpu/&lt;CPU&gt;</code>.  If you do this,
 122 the configury will find the directory by itself.  Otherwise you will need to
 123 edit the <code class="code">configure.host</code> file and, in the switch statement that sets
 124 <code class="code">cpu_include_dir</code>, add a pattern to handle your chip.
 125    </p><p>Note that some chip families share a single configuration directory, for
 126 example, <code class="code">alpha</code>, <code class="code">alphaev5</code>, and <code class="code">alphaev6</code> all use the
 127 <code class="code">config/cpu/alpha</code> directory, and there is an entry in the
 128 <code class="code">configure.host</code> switch statement to handle this.
 129    </p><p>The <code class="code">cpu_include_dir</code> sets default locations for the files controlling
 130 <a class="link" href="internals.html#internals.thread_safety" title="Thread Safety">Thread safety</a> and <a class="link" href="internals.html#internals.numeric_limits" title="Numeric Limits">Numeric limits</a>, if the defaults are not
 131 appropriate for your chip.
 132    </p></div><div class="section" title="Character Types"><div class="titlepage"><div><div><h3 class="title"><a name="internals.char_types"></a>Character Types</h3></div></div></div><p>The library requires that you provide three header files to implement
 133 character classification, analogous to that provided by the C libraries
 134 <code class="code">&lt;ctype.h&gt;</code> header.  You can model these on the files provided in
 135 <code class="code">config/os/generic</code>.  However, these files will almost
 136 certainly need some modification.
 137 </p><p>The first file to write is <code class="code">ctype_base.h</code>.  This file provides
 138 some very basic information about character classification.  The libstdc++
 139 library assumes that your C library implements <code class="code">&lt;ctype.h&gt;</code> by using
 140 a table (indexed by character code) containing integers, where each of
 141 these integers is a bit-mask indicating whether the character is
 142 upper-case, lower-case, alphabetic, etc.  The <code class="code">ctype_base.h</code>
 143 file gives the type of the integer, and the values of the various bit
 144 masks.  You will have to peer at your own <code class="code">&lt;ctype.h&gt;</code> to figure out
 145 how to define the values required by this file.
 146    </p><p>The <code class="code">ctype_base.h</code> header file does not need include guards.
 147 It should contain a single <code class="code">struct</code> definition called
 148 <code class="code">ctype_base</code>.  This <code class="code">struct</code> should contain two type
 149 declarations, and one enumeration declaration, like this example, taken
 150 from the IRIX configuration:
 151    </p><pre class="programlisting">
 152   struct ctype_base
 153      {
 154        typedef unsigned int     mask;
 155        typedef int*             __to_type;
 156
 157        enum
 158        {
 159          space = _ISspace,
 160          print = _ISprint,
 161          cntrl = _IScntrl,
 162          upper = _ISupper,
 163          lower = _ISlower,
 164          alpha = _ISalpha,
 165          digit = _ISdigit,
 166          punct = _ISpunct,
 167          xdigit = _ISxdigit,
 168          alnum = _ISalnum,
 169          graph = _ISgraph
 170        };
 171      };
 172 </pre><p>The <code class="code">mask</code> type is the type of the elements in the table.  If your
 173 C library uses a table to map lower-case numbers to upper-case numbers,
 174 and vice versa, you should define <code class="code">__to_type</code> to be the type of the
 175 elements in that table.  If you don't mind taking a minor performance
 176 penalty, or if your library doesn't implement <code class="code">toupper</code> and
 177 <code class="code">tolower</code> in this way, you can pick any pointer-to-integer type,
 178 but you must still define the type.
 179 </p><p>The enumeration should give definitions for all the values in the above
 180 example, using the values from your native <code class="code">&lt;ctype.h&gt;</code>.  They can
 181 be given symbolically (as above), or numerically, if you prefer.  You do
 182 not have to include <code class="code">&lt;ctype.h&gt;</code> in this header; it will always be
 183 included before <code class="code">ctype_base.h</code> is included.
 184    </p><p>The next file to write is <code class="code">ctype_configure_char.cc</code>.
 185 The first function that must be written is the <code class="code">ctype&lt;char&gt;::ctype</code> constructor.  Here is the IRIX example:
 186    </p><pre class="programlisting">
 187 ctype&lt;char&gt;::ctype(const mask* __table = 0, bool __del = false,
 188            size_t __refs = 0)
 189        : _Ctype_nois&lt;char&gt;(__refs), _M_del(__table != 0 &amp;&amp; __del),
 190          _M_toupper(NULL),
 191          _M_tolower(NULL),
 192          _M_ctable(NULL),
 193          _M_table(!__table
 194                   ? (const mask*) (__libc_attr._ctype_tbl-&gt;_class + 1)
 195                   : __table)
 196        { }
 197 </pre><p>There are two parts of this that you might choose to alter. The first,
 198 and most important, is the line involving <code class="code">__libc_attr</code>.  That is
 199 IRIX system-dependent code that gets the base of the table mapping
 200 character codes to attributes.  You need to substitute code that obtains
 201 the address of this table on your system.  If you want to use your
 202 operating system's tables to map upper-case letters to lower-case, and
 203 vice versa, you should initialize <code class="code">_M_toupper</code> and
 204 <code class="code">_M_tolower</code> with those tables, in similar fashion.
 205 </p><p>Now, you have to write two functions to convert from upper-case to
 206 lower-case, and vice versa.  Here are the IRIX versions:
 207    </p><pre class="programlisting">
 208      char
 209      ctype&lt;char&gt;::do_toupper(char __c) const
 210      { return _toupper(__c); }
 211
 212      char
 213      ctype&lt;char&gt;::do_tolower(char __c) const
 214      { return _tolower(__c); }
 215 </pre><p>Your C library provides equivalents to IRIX's <code class="code">_toupper</code> and
 216 <code class="code">_tolower</code>.  If you initialized <code class="code">_M_toupper</code> and
 217 <code class="code">_M_tolower</code> above, then you could use those tables instead.
 218 </p><p>Finally, you have to provide two utility functions that convert strings
 219 of characters.  The versions provided here will always work - but you
 220 could use specialized routines for greater performance if you have
 221 machinery to do that on your system:
 222    </p><pre class="programlisting">
 223      const char*
 224      ctype&lt;char&gt;::do_toupper(char* __low, const char* __high) const
 225      {
 226        while (__low &lt; __high)
 227          {
 228            *__low = do_toupper(*__low);
 229            ++__low;
 230          }
 231        return __high;
 232      }
 233
 234      const char*
 235      ctype&lt;char&gt;::do_tolower(char* __low, const char* __high) const
 236      {
 237        while (__low &lt; __high)
 238          {
 239            *__low = do_tolower(*__low);
 240            ++__low;
 241          }
 242        return __high;
 243      }
 244 </pre><p>You must also provide the <code class="code">ctype_inline.h</code> file, which
 245 contains a few more functions.  On most systems, you can just copy
 246 <code class="code">config/os/generic/ctype_inline.h</code> and use it on your system.
 247    </p><p>In detail, the functions provided test characters for particular
 248 properties; they are analogous to the functions like <code class="code">isalpha</code> and
 249 <code class="code">islower</code> provided by the C library.
 250    </p><p>The first function is implemented like this on IRIX:
 251    </p><pre class="programlisting">
 252      bool
 253      ctype&lt;char&gt;::
 254      is(mask __m, char __c) const throw()
 255      { return (_M_table)[(unsigned char)(__c)] &amp; __m; }
 256 </pre><p>The <code class="code">_M_table</code> is the table passed in above, in the constructor.
 257 This is the table that contains the bitmasks for each character.  The
 258 implementation here should work on all systems.
 259 </p><p>The next function is:
 260    </p><pre class="programlisting">
 261      const char*
 262      ctype&lt;char&gt;::
 263      is(const char* __low, const char* __high, mask* __vec) const throw()
 264      {
 265        while (__low &lt; __high)
 266          *__vec++ = (_M_table)[(unsigned char)(*__low++)];
 267        return __high;
 268      }
 269 </pre><p>This function is similar; it copies the masks for all the characters
 270 from <code class="code">__low</code> up until <code class="code">__high</code> into the vector given by
 271 <code class="code">__vec</code>.
 272 </p><p>The last two functions again are entirely generic:
 273    </p><pre class="programlisting">
 274      const char*
 275      ctype&lt;char&gt;::
 276      scan_is(mask __m, const char* __low, const char* __high) const throw()
 277      {
 278        while (__low &lt; __high &amp;&amp; !this-&gt;is(__m, *__low))
 279          ++__low;
 280        return __low;
 281      }
 282
 283      const char*
 284      ctype&lt;char&gt;::
 285      scan_not(mask __m, const char* __low, const char* __high) const throw()
 286      {
 287        while (__low &lt; __high &amp;&amp; this-&gt;is(__m, *__low))
 288          ++__low;
 289        return __low;
 290      }
 291 </pre></div><div class="section" title="Thread Safety"><div class="titlepage"><div><div><h3 class="title"><a name="internals.thread_safety"></a>Thread Safety</h3></div></div></div><p>The C++ library string functionality requires a couple of atomic
 292 operations to provide thread-safety.  If you don't take any special
 293 action, the library will use stub versions of these functions that are
 294 not thread-safe.  They will work fine, unless your applications are
 295 multi-threaded.
 296 </p><p>If you want to provide custom, safe, versions of these functions, there
 297 are two distinct approaches.  One is to provide a version for your CPU,
 298 using assembly language constructs.  The other is to use the
 299 thread-safety primitives in your operating system.  In either case, you
 300 make a file called <code class="code">atomicity.h</code>, and the variable
 301 <code class="code">ATOMICITYH</code> must point to this file.
 302    </p><p>If you are using the assembly-language approach, put this code in
 303 <code class="code">config/cpu/&lt;chip&gt;/atomicity.h</code>, where chip is the name of
 304 your processor (see <a class="link" href="internals.html#internals.cpu" title="CPU">CPU</a>).  No additional changes are necessary to
 305 locate the file in this case; <code class="code">ATOMICITYH</code> will be set by default.
 306    </p><p>If you are using the operating system thread-safety primitives approach,
 307 you can also put this code in the same CPU directory, in which case no more
 308 work is needed to locate the file.  For examples of this approach,
 309 see the <code class="code">atomicity.h</code> file for IRIX or IA64.
 310    </p><p>Alternatively, if the primitives are more closely related to the OS
 311 than they are to the CPU, you can put the <code class="code">atomicity.h</code> file in
 312 the <a class="link" href="internals.html#internals.os" title="Operating System">Operating system</a> directory instead.  In this case, you must
 313 edit <code class="code">configure.host</code>, and in the switch statement that handles
 314 operating systems, override the <code class="code">ATOMICITYH</code> variable to point to
 315 the appropriate <code class="code">os_include_dir</code>.  For examples of this approach,
 316 see the <code class="code">atomicity.h</code> file for AIX.
 317    </p><p>With those bits out of the way, you have to actually write
 318 <code class="code">atomicity.h</code> itself.  This file should be wrapped in an
 319 include guard named <code class="code">_GLIBCXX_ATOMICITY_H</code>.  It should define one
 320 type, and two functions.
 321    </p><p>The type is <code class="code">_Atomic_word</code>.  Here is the version used on IRIX:
 322    </p><pre class="programlisting">
 323 typedef long _Atomic_word;
 324 </pre><p>This type must be a signed integral type supporting atomic operations.
 325 If you're using the OS approach, use the same type used by your system's
 326 primitives.  Otherwise, use the type for which your CPU provides atomic
 327 primitives.
 328 </p><p>Then, you must provide two functions.  The bodies of these functions
 329 must be equivalent to those provided here, but using atomic operations:
 330    </p><pre class="programlisting">
 331      static inline _Atomic_word
 332      __attribute__ ((__unused__))
 333      __exchange_and_add (_Atomic_word* __mem, int __val)
 334      {
 335        _Atomic_word __result = *__mem;
 336        *__mem += __val;
 337        return __result;
 338      }
 339
 340      static inline void
 341      __attribute__ ((__unused__))
 342      __atomic_add (_Atomic_word* __mem, int __val)
 343      {
 344        *__mem += __val;
 345      }
 346 </pre></div><div class="section" title="Numeric Limits"><div class="titlepage"><div><div><h3 class="title"><a name="internals.numeric_limits"></a>Numeric Limits</h3></div></div></div><p>The C++ library requires information about the fundamental data types,
 347 such as the minimum and maximum representable values of each type.
 348 You can define each of these values individually, but it is usually
 349 easiest just to indicate how many bits are used in each of the data
 350 types and let the library do the rest.  For information about the
 351 macros to define, see the top of <code class="code">include/bits/std_limits.h</code>.
 352 </p><p>If you need to define any macros, you can do so in <code class="code">os_defines.h</code>.
 353 However, if all operating systems for your CPU are likely to use the
 354 same values, you can provide a CPU-specific file instead so that you
 355 do not have to provide the same definitions for each operating system.
 356 To take that approach, create a new file called <code class="code">cpu_limits.h</code> in
 357 your CPU configuration directory (see <a class="link" href="internals.html#internals.cpu" title="CPU">CPU</a>).
 358    </p></div><div class="section" title="Libtool"><div class="titlepage"><div><div><h3 class="title"><a name="internals.libtool"></a>Libtool</h3></div></div></div><p>The C++ library is compiled, archived and linked with libtool.
 359 Explaining the full workings of libtool is beyond the scope of this
 360 document, but there are a few, particular bits that are necessary for
 361 porting.
 362 </p><p>Some parts of the libstdc++ library are compiled with the libtool
 363 <code class="code">--tags CXX</code> option (the C++ definitions for libtool).  Therefore,
 364 <code class="code">ltcf-cxx.sh</code> in the top-level directory needs to have the correct
 365 logic to compile and archive objects equivalent to the C version of libtool,
 366 <code class="code">ltcf-c.sh</code>.  Some libtool targets have definitions for C but not
 367 for C++, or C++ definitions which have not been kept up to date.
 368    </p><p>The C++ run-time library contains initialization code that needs to be
 369 run as the library is loaded.  Often, that requires linking in special
 370 object files when the C++ library is built as a shared library, or
 371 taking other system-specific actions.
 372    </p><p>The libstdc++ library is linked with the C version of libtool, even
 373 though it is a C++ library.  Therefore, the C version of libtool needs to
 374 ensure that the run-time library initializers are run.  The usual way to
 375 do this is to build the library using <code class="code">gcc -shared</code>.
 376    </p><p>If you need to change how the library is linked, look at
 377 <code class="code">ltcf-c.sh</code> in the top-level directory.  Find the switch statement
 378 that sets <code class="code">archive_cmds</code>.  Here, adjust the setting for your
 379 operating system.
 380    </p></div></div><div class="navfooter"><hr><table width="100%" summary="Navigation footer"><tr><td width="40%" align="left"><a accesskey="p" href="documentation_hacking.html">Prev</a> </td><td width="20%" align="center"><a accesskey="u" href="appendix_porting.html">Up</a></td><td width="40%" align="right"> <a accesskey="n" href="test.html">Next</a></td></tr><tr><td width="40%" align="left" valign="top">Writing and Generating Documentation </td><td width="20%" align="center"><a accesskey="h" href="../index.html">Home</a></td><td width="40%" align="right" valign="top"> Test</td></tr></table></div></body></html>