1 <chapter xmlns="http://docbook.org/ns/docbook" version="5.0"
2 xml:id="manual.ext.concurrency" xreflabel="Concurrency Extensions">
3 <?dbhtml filename="ext_concurrency.html"?>
5 <info><title>Concurrency</title>
18 <section xml:id="manual.ext.concurrency.design" xreflabel="Design"><info><title>Design</title></info>
21 <section xml:id="manual.ext.concurrency.design.threads" xreflabel="Threads API"><info><title>Interface to Locks and Mutexes</title></info>
24 <para>The file <filename class="headerfile"><ext/concurrence.h></filename>
25 contains all the higher-level
26 constructs for playing with threads. In contrast to the atomics layer,
27 the concurrence layer consists largely of types. All types are defined within <code>namespace __gnu_cxx</code>.
31 These types can be used in a portable manner, regardless of the
32 specific environment. They are carefully designed to provide optimum
33 efficiency and speed, abstracting out underlying thread calls and
34 accesses when compiling for single-threaded situations (even on hosts
35 that support multiple threads.)
38 <para>The enumerated type <code>_Lock_policy</code> details the set of
40 policies: <code>_S_single</code>, <code>_S_mutex</code>,
41 and <code>_S_atomic</code>.
45 <listitem><para><code>_S_single</code></para>
46 <para>Indicates single-threaded code that does not need locking.
50 <listitem><para><code>_S_mutex</code></para>
51 <para>Indicates multi-threaded code using thread-layer abstractions.
54 <listitem><para><code>_S_atomic</code></para>
55 <para>Indicates multi-threaded code using atomic operations.
60 <para>The compile-time constant <code>__default_lock_policy</code> is set
61 to one of the three values above, depending on characteristics of the
62 host environment and the current compilation flags.
65 <para>Two more datatypes make up the rest of the
66 interface: <code>__mutex</code>, and <code>__scoped_lock</code>.
69 <para>The scoped lock idiom is well-discussed within the C++
70 community. This version takes a <code>__mutex</code> reference, and
71 locks it during construction of <code>__scoped_lock</code> and
72 unlocks it during destruction. This is an efficient way of locking
73 critical sections, while retaining exception-safety.
74 These types have been superseded in the ISO C++ 2011 standard by the
75 mutex and lock types defined in the header
76 <filename class="headerfile"><mutex></filename>.
80 <section xml:id="manual.ext.concurrency.design.atomics" xreflabel="Atomic API"><info><title>Interface to Atomic Functions</title></info>
85 Two functions and one type form the base of atomic support.
89 <para>The type <code>_Atomic_word</code> is a signed integral type
90 supporting atomic operations.
94 The two functions functions are:
99 __exchange_and_add_dispatch(volatile _Atomic_word*, int);
102 __atomic_add_dispatch(volatile _Atomic_word*, int);
105 <para>Both of these functions are declared in the header file
106 <ext/atomicity.h>, and are in <code>namespace __gnu_cxx</code>.
112 __exchange_and_add_dispatch
115 <para>Adds the second argument's value to the first argument. Returns the old value.
120 __atomic_add_dispatch
123 <para>Adds the second argument's value to the first argument. Has no return value.
129 These functions forward to one of several specialized helper
130 functions, depending on the circumstances. For instance,
135 __exchange_and_add_dispatch
140 Calls through to either of:
144 <listitem><para><code>__exchange_and_add</code>
146 <para>Multi-thread version. Inlined if compiler-generated builtin atomics
147 can be used, otherwise resolved at link time to a non-builtin code
152 <listitem><para><code>__exchange_and_add_single</code>
154 <para>Single threaded version. Inlined.</para>
158 <para>However, only <code>__exchange_and_add_dispatch</code>
159 and <code>__atomic_add_dispatch</code> should be used. These functions
160 can be used in a portable manner, regardless of the specific
161 environment. They are carefully designed to provide optimum efficiency
162 and speed, abstracting out atomic accesses when they are not required
163 (even on hosts that support compiler intrinsics for atomic
168 In addition, there are two macros
173 _GLIBCXX_READ_MEM_BARRIER
178 _GLIBCXX_WRITE_MEM_BARRIER
183 Which expand to the appropriate write and read barrier required by the
184 host hardware and operating system.
191 <section xml:id="manual.ext.concurrency.impl" xreflabel="Implementation"><info><title>Implementation</title></info>
193 <section xml:id="manual.ext.concurrency.impl.atomic_fallbacks" xreflabel="Atomic F"><info><title>Using Builtin Atomic Functions</title></info>
196 <para>The functions for atomic operations described above are either
197 implemented via compiler intrinsics (if the underlying host is
198 capable) or by library fallbacks.</para>
200 <para>Compiler intrinsics (builtins) are always preferred. However, as
201 the compiler builtins for atomics are not universally implemented,
202 using them directly is problematic, and can result in undefined
203 function calls. (An example of an undefined symbol from the use
204 of <code>__sync_fetch_and_add</code> on an unsupported host is a
205 missing reference to <code>__sync_fetch_and_add_4</code>.)
208 <para>In addition, on some hosts the compiler intrinsics are enabled
209 conditionally, via the <code>-march</code> command line flag. This makes
210 usage vary depending on the target hardware and the flags used during
218 Incomplete/inconsistent. This is only C++11.
223 If builtins are possible for bool-sized integral types,
224 <code>ATOMIC_BOOL_LOCK_FREE</code> will be defined.
225 If builtins are possible for int-sized integral types,
226 <code>ATOMIC_INT_LOCK_FREE</code> will be defined.
230 <para>For the following hosts, intrinsics are enabled by default.
234 <listitem><para>alpha</para></listitem>
235 <listitem><para>ia64</para></listitem>
236 <listitem><para>powerpc</para></listitem>
237 <listitem><para>s390</para></listitem>
240 <para>For others, some form of <code>-march</code> may work. On
241 non-ancient x86 hardware, <code>-march=native</code> usually does the
244 <para> For hosts without compiler intrinsics, but with capable
245 hardware, hand-crafted assembly is selected. This is the case for the following hosts:
249 <listitem><para>cris</para></listitem>
250 <listitem><para>hppa</para></listitem>
251 <listitem><para>i386</para></listitem>
252 <listitem><para>i486</para></listitem>
253 <listitem><para>m48k</para></listitem>
254 <listitem><para>mips</para></listitem>
255 <listitem><para>sparc</para></listitem>
258 <para>And for the rest, a simulated atomic lock via pthreads.
261 <para> Detailed information about compiler intrinsics for atomic operations can be found in the GCC <link xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="http://gcc.gnu.org/onlinedocs/gcc/Atomic-Builtins.html"> documentation</link>.
264 <para> More details on the library fallbacks from the porting <link linkend="internals.thread_safety">section</link>.
269 <section xml:id="manual.ext.concurrency.impl.thread" xreflabel="Pthread"><info><title>Thread Abstraction</title></info>
272 <para>A thin layer above IEEE 1003.1 (i.e. pthreads) is used to abstract
273 the thread interface for GCC. This layer is called "gthread," and is
274 comprised of one header file that wraps the host's default thread layer with
275 a POSIX-like interface.
278 <para> The file <gthr-default.h> points to the deduced wrapper for
279 the current host. In libstdc++ implementation files,
280 <bits/gthr.h> is used to select the proper gthreads file.
283 <para>Within libstdc++ sources, all calls to underlying thread functionality
284 use this layer. More detail as to the specific interface can be found in the source <link xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="http://gcc.gnu.org/onlinedocs/libstdc++/latest-doxygen/a00883_source.html">documentation</link>.
287 <para>By design, the gthread layer is interoperable with the types,
288 functions, and usage found in the usual <pthread.h> file,
289 including <code>pthread_t</code>, <code>pthread_once_t</code>, <code>pthread_create</code>,
296 <section xml:id="manual.ext.concurrency.use" xreflabel="Use"><info><title>Use</title></info>
300 <para>Typical usage of the last two constructs is demonstrated as follows:
304 #include <ext/concurrence.h>
308 __gnu_cxx::__mutex safe_base_mutex;
309 } // anonymous namespace
316 __gnu_cxx::__scoped_lock sentry(safe_base_mutex);
317 for (int i = 0; i < max; ++i)
319 _Safe_iterator_base* __old = __iter;
320 __iter = __iter-<_M_next;
321 __old-<_M_detach_single();
326 <para>In this sample code, an anonymous namespace is used to keep
327 the <code>__mutex</code> private to the compilation unit,
328 and <code>__scoped_lock</code> is used to guard access to the critical
329 section within the for loop, locking the mutex on creation and freeing
330 the mutex as control moves out of this block.
333 <para>Several exception classes are used to keep track of
334 concurrence-related errors. These classes
335 are: <code>__concurrence_lock_error</code>, <code>__concurrence_unlock_error</code>, <code>__concurrence_wait_error</code>,
336 and <code>__concurrence_broadcast_error</code>.