LanguageExtensions.rst

``instancetype`` as its result type.  ``instancetype`` is a contextual keyword
that is only permitted in the result type of an Objective-C method, e.g.

.. code-block:: objc

  @interface A
  + (instancetype)constructAnA;
  @end

The related result type can also be inferred for some methods.  To determine
whether a method has an inferred related result type, the first word in the
camel-case selector (e.g., "``init``" in "``initWithObjects``") is considered,
and the method will have a related result type if its return type is compatible
with the type of its class and if:

* the first word is "``alloc``" or "``new``", and the method is a class method,
  or

* the first word is "``autorelease``", "``init``", "``retain``", or "``self``",
  and the method is an instance method.

If a method with a related result type is overridden by a subclass method, the
subclass method must also return a type that is compatible with the subclass
type.  For example:

.. code-block:: objc

  @interface NSString : NSObject
  - (NSUnrelated *)init; // incorrect usage: NSUnrelated is not NSString or a superclass of NSString
  @end

Related result types only affect the type of a message send or property access
via the given method.  In all other respects, a method with a related result
type is treated the same way as method that returns ``id``.

Use ``__has_feature(objc_instancetype)`` to determine whether the
``instancetype`` contextual keyword is available.

Automatic reference counting
----------------------------

Clang provides support for :doc:`automated reference counting
<AutomaticReferenceCounting>` in Objective-C, which eliminates the need
for manual ``retain``/``release``/``autorelease`` message sends.  There are two
feature macros associated with automatic reference counting:
``__has_feature(objc_arc)`` indicates the availability of automated reference
counting in general, while ``__has_feature(objc_arc_weak)`` indicates that
automated reference counting also includes support for ``__weak`` pointers to
Objective-C objects.

.. _objc-fixed-enum:

Enumerations with a fixed underlying type
-----------------------------------------

Clang provides support for C++11 enumerations with a fixed underlying type
within Objective-C.  For example, one can write an enumeration type as:

.. code-block:: c++

  typedef enum : unsigned char { Red, Green, Blue } Color;

This specifies that the underlying type, which is used to store the enumeration
value, is ``unsigned char``.

Use ``__has_feature(objc_fixed_enum)`` to determine whether support for fixed
underlying types is available in Objective-C.

Interoperability with C++11 lambdas
-----------------------------------

Clang provides interoperability between C++11 lambdas and blocks-based APIs, by
permitting a lambda to be implicitly converted to a block pointer with the
corresponding signature.  For example, consider an API such as ``NSArray``'s
array-sorting method:

.. code-block:: objc

  - (NSArray *)sortedArrayUsingComparator:(NSComparator)cmptr;

``NSComparator`` is simply a typedef for the block pointer ``NSComparisonResult
(^)(id, id)``, and parameters of this type are generally provided with block
literals as arguments.  However, one can also use a C++11 lambda so long as it
provides the same signature (in this case, accepting two parameters of type
``id`` and returning an ``NSComparisonResult``):

.. code-block:: objc

  NSArray *array = @[@"string 1", @"string 21", @"string 12", @"String 11",
                     @"String 02"];
  const NSStringCompareOptions comparisonOptions
    = NSCaseInsensitiveSearch | NSNumericSearch |
      NSWidthInsensitiveSearch | NSForcedOrderingSearch;
  NSLocale *currentLocale = [NSLocale currentLocale];
  NSArray *sorted
    = [array sortedArrayUsingComparator:[=](id s1, id s2) -> NSComparisonResult {
               NSRange string1Range = NSMakeRange(0, [s1 length]);
               return [s1 compare:s2 options:comparisonOptions
               range:string1Range locale:currentLocale];
       }];
  NSLog(@"sorted: %@", sorted);

This code relies on an implicit conversion from the type of the lambda
expression (an unnamed, local class type called the *closure type*) to the
corresponding block pointer type.  The conversion itself is expressed by a
conversion operator in that closure type that produces a block pointer with the
same signature as the lambda itself, e.g.,

.. code-block:: objc

  operator NSComparisonResult (^)(id, id)() const;

This conversion function returns a new block that simply forwards the two
parameters to the lambda object (which it captures by copy), then returns the
result.  The returned block is first copied (with ``Block_copy``) and then
autoreleased.  As an optimization, if a lambda expression is immediately
converted to a block pointer (as in the first example, above), then the block
is not copied and autoreleased: rather, it is given the same lifetime as a
block literal written at that point in the program, which avoids the overhead
of copying a block to the heap in the common case.

The conversion from a lambda to a block pointer is only available in
Objective-C++, and not in C++ with blocks, due to its use of Objective-C memory
management (autorelease).

Object Literals and Subscripting
--------------------------------

Clang provides support for :doc:`Object Literals and Subscripting
<ObjectiveCLiterals>` in Objective-C, which simplifies common Objective-C
programming patterns, makes programs more concise, and improves the safety of
container creation.  There are several feature macros associated with object
literals and subscripting: ``__has_feature(objc_array_literals)`` tests the
availability of array literals; ``__has_feature(objc_dictionary_literals)``
tests the availability of dictionary literals;
``__has_feature(objc_subscripting)`` tests the availability of object
subscripting.

Objective-C Autosynthesis of Properties
---------------------------------------

Clang provides support for autosynthesis of declared properties.  Using this
feature, clang provides default synthesis of those properties not declared
@dynamic and not having user provided backing getter and setter methods.
``__has_feature(objc_default_synthesize_properties)`` checks for availability
of this feature in version of clang being used.

.. _langext-objc-retain-release:

Objective-C retaining behavior attributes
-----------------------------------------

In Objective-C, functions and methods are generally assumed to follow the
`Cocoa Memory Management 
<http://developer.apple.com/library/mac/#documentation/Cocoa/Conceptual/MemoryMgmt/Articles/mmRules.html>`_
conventions for ownership of object arguments and
return values. However, there are exceptions, and so Clang provides attributes
to allow these exceptions to be documented. This are used by ARC and the
`static analyzer <http://clang-analyzer.llvm.org>`_ Some exceptions may be
better described using the ``objc_method_family`` attribute instead.

**Usage**: The ``ns_returns_retained``, ``ns_returns_not_retained``,
``ns_returns_autoreleased``, ``cf_returns_retained``, and
``cf_returns_not_retained`` attributes can be placed on methods and functions
that return Objective-C or CoreFoundation objects. They are commonly placed at
the end of a function prototype or method declaration:

.. code-block:: objc

  id foo() __attribute__((ns_returns_retained));

  - (NSString *)bar:(int)x __attribute__((ns_returns_retained));

The ``*_returns_retained`` attributes specify that the returned object has a +1
retain count.  The ``*_returns_not_retained`` attributes specify that the return
object has a +0 retain count, even if the normal convention for its selector
would be +1.  ``ns_returns_autoreleased`` specifies that the returned object is
+0, but is guaranteed to live at least as long as the next flush of an
autorelease pool.

**Usage**: The ``ns_consumed`` and ``cf_consumed`` attributes can be placed on
an parameter declaration; they specify that the argument is expected to have a
+1 retain count, which will be balanced in some way by the function or method.
The ``ns_consumes_self`` attribute can only be placed on an Objective-C
method; it specifies that the method expects its ``self`` parameter to have a
+1 retain count, which it will balance in some way.

.. code-block:: objc

  void foo(__attribute__((ns_consumed)) NSString *string);

  - (void) bar __attribute__((ns_consumes_self));
  - (void) baz:(id) __attribute__((ns_consumed)) x;

Further examples of these attributes are available in the static analyzer's `list of annotations for analysis
<http://clang-analyzer.llvm.org/annotations.html#cocoa_mem>`_.

Query for these features with ``__has_attribute(ns_consumed)``,
``__has_attribute(ns_returns_retained)``, etc.


Objective-C++ ABI: protocol-qualifier mangling of parameters
------------------------------------------------------------

Starting with LLVM 3.4, Clang produces a new mangling for parameters whose
type is a qualified-``id`` (e.g., ``id<Foo>``).  This mangling allows such
parameters to be differentiated from those with the regular unqualified ``id``
type.

This was a non-backward compatible mangling change to the ABI.  This change
allows proper overloading, and also prevents mangling conflicts with template
parameters of protocol-qualified type.

Query the presence of this new mangling with
``__has_feature(objc_protocol_qualifier_mangling)``.

.. _langext-overloading:

Initializer lists for complex numbers in C
==========================================

clang supports an extension which allows the following in C:

.. code-block:: c++

  #include <math.h>
  #include <complex.h>
  complex float x = { 1.0f, INFINITY }; // Init to (1, Inf)

This construct is useful because there is no way to separately initialize the
real and imaginary parts of a complex variable in standard C, given that clang
does not support ``_Imaginary``.  (Clang also supports the ``__real__`` and
``__imag__`` extensions from gcc, which help in some cases, but are not usable
in static initializers.)

Note that this extension does not allow eliding the braces; the meaning of the
following two lines is different:

.. code-block:: c++

  complex float x[] = { { 1.0f, 1.0f } }; // [0] = (1, 1)
  complex float x[] = { 1.0f, 1.0f }; // [0] = (1, 0), [1] = (1, 0)

This extension also works in C++ mode, as far as that goes, but does not apply
to the C++ ``std::complex``.  (In C++11, list initialization allows the same
syntax to be used with ``std::complex`` with the same meaning.)

Builtin Functions
=================

Clang supports a number of builtin library functions with the same syntax as
GCC, including things like ``__builtin_nan``, ``__builtin_constant_p``,
``__builtin_choose_expr``, ``__builtin_types_compatible_p``,
``__builtin_assume_aligned``, ``__sync_fetch_and_add``, etc.  In addition to
the GCC builtins, Clang supports a number of builtins that GCC does not, which
are listed here.

Please note that Clang does not and will not support all of the GCC builtins
for vector operations.  Instead of using builtins, you should use the functions
defined in target-specific header files like ``<xmmintrin.h>``, which define
portable wrappers for these.  Many of the Clang versions of these functions are
implemented directly in terms of :ref:`extended vector support
<langext-vectors>` instead of builtins, in order to reduce the number of
builtins that we need to implement.

``__builtin_assume``
------------------------------

``__builtin_assume`` is used to provide the optimizer with a boolean
invariant that is defined to be true.

**Syntax**:

.. code-block:: c++

  __builtin_assume(bool)

**Example of Use**:

.. code-block:: c++

  int foo(int x) {
    __builtin_assume(x != 0);

    // The optimizer may short-circuit this check using the invariant.
    if (x == 0)
      return do_something();

    return do_something_else();
  }

**Description**:

The boolean argument to this function is defined to be true. The optimizer may
analyze the form of the expression provided as the argument and deduce from
that information used to optimize the program. If the condition is violated
during execution, the behavior is undefined. The argument itself is never
evaluated, so any side effects of the expression will be discarded.

Query for this feature with ``__has_builtin(__builtin_assume)``.

``__builtin_readcyclecounter``
------------------------------

``__builtin_readcyclecounter`` is used to access the cycle counter register (or
a similar low-latency, high-accuracy clock) on those targets that support it.

**Syntax**:

.. code-block:: c++

  __builtin_readcyclecounter()

**Example of Use**:

.. code-block:: c++

  unsigned long long t0 = __builtin_readcyclecounter();
  do_something();
  unsigned long long t1 = __builtin_readcyclecounter();
  unsigned long long cycles_to_do_something = t1 - t0; // assuming no overflow

**Description**:

The ``__builtin_readcyclecounter()`` builtin returns the cycle counter value,
which may be either global or process/thread-specific depending on the target.
As the backing counters often overflow quickly (on the order of seconds) this
should only be used for timing small intervals.  When not supported by the
target, the return value is always zero.  This builtin takes no arguments and
produces an unsigned long long result.

Query for this feature with ``__has_builtin(__builtin_readcyclecounter)``. Note
that even if present, its use may depend on run-time privilege or other OS
controlled state.

.. _langext-__builtin_shufflevector:

``__builtin_shufflevector``
---------------------------

``__builtin_shufflevector`` is used to express generic vector
permutation/shuffle/swizzle operations.  This builtin is also very important
for the implementation of various target-specific header files like
``<xmmintrin.h>``.

**Syntax**:

.. code-block:: c++

  __builtin_shufflevector(vec1, vec2, index1, index2, ...)

**Examples**:

.. code-block:: c++

  // identity operation - return 4-element vector v1.
  __builtin_shufflevector(v1, v1, 0, 1, 2, 3)

  // "Splat" element 0 of V1 into a 4-element result.
  __builtin_shufflevector(V1, V1, 0, 0, 0, 0)

  // Reverse 4-element vector V1.
  __builtin_shufflevector(V1, V1, 3, 2, 1, 0)

  // Concatenate every other element of 4-element vectors V1 and V2.
  __builtin_shufflevector(V1, V2, 0, 2, 4, 6)

  // Concatenate every other element of 8-element vectors V1 and V2.
  __builtin_shufflevector(V1, V2, 0, 2, 4, 6, 8, 10, 12, 14)

  // Shuffle v1 with some elements being undefined
  __builtin_shufflevector(v1, v1, 3, -1, 1, -1)

**Description**:

The first two arguments to ``__builtin_shufflevector`` are vectors that have
the same element type.  The remaining arguments are a list of integers that
specify the elements indices of the first two vectors that should be extracted
and returned in a new vector.  These element indices are numbered sequentially
starting with the first vector, continuing into the second vector.  Thus, if
``vec1`` is a 4-element vector, index 5 would refer to the second element of
``vec2``. An index of -1 can be used to indicate that the corresponding element
in the returned vector is a don't care and can be optimized by the backend.

The result of ``__builtin_shufflevector`` is a vector with the same element
type as ``vec1``/``vec2`` but that has an element count equal to the number of
indices specified.

Query for this feature with ``__has_builtin(__builtin_shufflevector)``.

.. _langext-__builtin_convertvector:

``__builtin_convertvector``
---------------------------

``__builtin_convertvector`` is used to express generic vector
type-conversion operations. The input vector and the output vector
type must have the same number of elements.

**Syntax**:

.. code-block:: c++

  __builtin_convertvector(src_vec, dst_vec_type)

**Examples**:

.. code-block:: c++

  typedef double vector4double __attribute__((__vector_size__(32)));
  typedef float  vector4float  __attribute__((__vector_size__(16)));
  typedef short  vector4short  __attribute__((__vector_size__(8)));
  vector4float vf; vector4short vs;

  // convert from a vector of 4 floats to a vector of 4 doubles.
  __builtin_convertvector(vf, vector4double)
  // equivalent to:
  (vector4double) { (double) vf[0], (double) vf[1], (double) vf[2], (double) vf[3] }

  // convert from a vector of 4 shorts to a vector of 4 floats.
  __builtin_convertvector(vs, vector4float)
  // equivalent to:
  (vector4float) { (float) vs[0], (float) vs[1], (float) vs[2], (float) vs[3] }

**Description**:

The first argument to ``__builtin_convertvector`` is a vector, and the second
argument is a vector type with the same number of elements as the first
argument.

The result of ``__builtin_convertvector`` is a vector with the same element
type as the second argument, with a value defined in terms of the action of a
C-style cast applied to each element of the first argument.

Query for this feature with ``__has_builtin(__builtin_convertvector)``.

``__builtin_unreachable``
-------------------------

``__builtin_unreachable`` is used to indicate that a specific point in the
program cannot be reached, even if the compiler might otherwise think it can.
This is useful to improve optimization and eliminates certain warnings.  For
example, without the ``__builtin_unreachable`` in the example below, the
compiler assumes that the inline asm can fall through and prints a "function
declared '``noreturn``' should not return" warning.

**Syntax**:

.. code-block:: c++

    __builtin_unreachable()

**Example of use**:

.. code-block:: c++

  void myabort(void) __attribute__((noreturn));
  void myabort(void) {
    asm("int3");
    __builtin_unreachable();
  }

**Description**:

The ``__builtin_unreachable()`` builtin has completely undefined behavior.
Since it has undefined behavior, it is a statement that it is never reached and
the optimizer can take advantage of this to produce better code.  This builtin
takes no arguments and produces a void result.

Query for this feature with ``__has_builtin(__builtin_unreachable)``.

``__sync_swap``
---------------

``__sync_swap`` is used to atomically swap integers or pointers in memory.

**Syntax**:

.. code-block:: c++

  type __sync_swap(type *ptr, type value, ...)

**Example of Use**:

.. code-block:: c++

  int old_value = __sync_swap(&value, new_value);

**Description**:

The ``__sync_swap()`` builtin extends the existing ``__sync_*()`` family of
atomic intrinsics to allow code to atomically swap the current value with the
new value.  More importantly, it helps developers write more efficient and
correct code by avoiding expensive loops around
``__sync_bool_compare_and_swap()`` or relying on the platform specific
implementation details of ``__sync_lock_test_and_set()``.  The
``__sync_swap()`` builtin is a full barrier.

``__builtin_addressof``
-----------------------

``__builtin_addressof`` performs the functionality of the built-in ``&``
operator, ignoring any ``operator&`` overload.  This is useful in constant
expressions in C++11, where there is no other way to take the address of an
object that overloads ``operator&``.

**Example of use**:

.. code-block:: c++

  template<typename T> constexpr T *addressof(T &value) {
    return __builtin_addressof(value);
  }

``__builtin_operator_new`` and ``__builtin_operator_delete``
------------------------------------------------------------

``__builtin_operator_new`` allocates memory just like a non-placement non-class
*new-expression*. This is exactly like directly calling the normal
non-placement ``::operator new``, except that it allows certain optimizations
that the C++ standard does not permit for a direct function call to
``::operator new`` (in particular, removing ``new`` / ``delete`` pairs and
merging allocations).

Likewise, ``__builtin_operator_delete`` deallocates memory just like a
non-class *delete-expression*, and is exactly like directly calling the normal
``::operator delete``, except that it permits optimizations. Only the unsized
form of ``__builtin_operator_delete`` is currently available.

These builtins are intended for use in the implementation of ``std::allocator``
and other similar allocation libraries, and are only available in C++.

Multiprecision Arithmetic Builtins
----------------------------------

Clang provides a set of builtins which expose multiprecision arithmetic in a
manner amenable to C. They all have the following form:

.. code-block:: c

  unsigned x = ..., y = ..., carryin = ..., carryout;
  unsigned sum = __builtin_addc(x, y, carryin, &carryout);

Thus one can form a multiprecision addition chain in the following manner:

.. code-block:: c

  unsigned *x, *y, *z, carryin=0, carryout;
  z[0] = __builtin_addc(x[0], y[0], carryin, &carryout);
  carryin = carryout;
  z[1] = __builtin_addc(x[1], y[1], carryin, &carryout);
  carryin = carryout;
  z[2] = __builtin_addc(x[2], y[2], carryin, &carryout);
  carryin = carryout;
  z[3] = __builtin_addc(x[3], y[3], carryin, &carryout);

The complete list of builtins are:

.. code-block:: c

  unsigned char      __builtin_addcb (unsigned char x, unsigned char y, unsigned char carryin, unsigned char *carryout);
  unsigned short     __builtin_addcs (unsigned short x, unsigned short y, unsigned short carryin, unsigned short *carryout);
  unsigned           __builtin_addc  (unsigned x, unsigned y, unsigned carryin, unsigned *carryout);
  unsigned long      __builtin_addcl (unsigned long x, unsigned long y, unsigned long carryin, unsigned long *carryout);
  unsigned long long __builtin_addcll(unsigned long long x, unsigned long long y, unsigned long long carryin, unsigned long long *carryout);
  unsigned char      __builtin_subcb (unsigned char x, unsigned char y, unsigned char carryin, unsigned char *carryout);
  unsigned short     __builtin_subcs (unsigned short x, unsigned short y, unsigned short carryin, unsigned short *carryout);
  unsigned           __builtin_subc  (unsigned x, unsigned y, unsigned carryin, unsigned *carryout);
  unsigned long      __builtin_subcl (unsigned long x, unsigned long y, unsigned long carryin, unsigned long *carryout);
  unsigned long long __builtin_subcll(unsigned long long x, unsigned long long y, unsigned long long carryin, unsigned long long *carryout);

Checked Arithmetic Builtins
---------------------------

Clang provides a set of builtins that implement checked arithmetic for security
critical applications in a manner that is fast and easily expressable in C. As
an example of their usage:

.. code-block:: c

  errorcode_t security_critical_application(...) {
    unsigned x, y, result;
    ...
    if (__builtin_umul_overflow(x, y, &result))
      return kErrorCodeHackers;
    ...
    use_multiply(result);
    ...
  }

A complete enumeration of the builtins are:

.. code-block:: c

  bool __builtin_uadd_overflow  (unsigned x, unsigned y, unsigned *sum);
  bool __builtin_uaddl_overflow (unsigned long x, unsigned long y, unsigned long *sum);
  bool __builtin_uaddll_overflow(unsigned long long x, unsigned long long y, unsigned long long *sum);
  bool __builtin_usub_overflow  (unsigned x, unsigned y, unsigned *diff);
  bool __builtin_usubl_overflow (unsigned long x, unsigned long y, unsigned long *diff);
  bool __builtin_usubll_overflow(unsigned long long x, unsigned long long y, unsigned long long *diff);
  bool __builtin_umul_overflow  (unsigned x, unsigned y, unsigned *prod);
  bool __builtin_umull_overflow (unsigned long x, unsigned long y, unsigned long *prod);
  bool __builtin_umulll_overflow(unsigned long long x, unsigned long long y, unsigned long long *prod);
  bool __builtin_sadd_overflow  (int x, int y, int *sum);
  bool __builtin_saddl_overflow (long x, long y, long *sum);
  bool __builtin_saddll_overflow(long long x, long long y, long long *sum);
  bool __builtin_ssub_overflow  (int x, int y, int *diff);
  bool __builtin_ssubl_overflow (long x, long y, long *diff);
  bool __builtin_ssubll_overflow(long long x, long long y, long long *diff);
  bool __builtin_smul_overflow  (int x, int y, int *prod);
  bool __builtin_smull_overflow (long x, long y, long *prod);
  bool __builtin_smulll_overflow(long long x, long long y, long long *prod);


.. _langext-__c11_atomic:

__c11_atomic builtins
---------------------

Clang provides a set of builtins which are intended to be used to implement
C11's ``<stdatomic.h>`` header.  These builtins provide the semantics of the
``_explicit`` form of the corresponding C11 operation, and are named with a
``__c11_`` prefix.  The supported operations, and the differences from
the corresponding C11 operations, are:

* ``__c11_atomic_init``
* ``__c11_atomic_thread_fence``
* ``__c11_atomic_signal_fence``
* ``__c11_atomic_is_lock_free`` (The argument is the size of the
  ``_Atomic(...)`` object, instead of its address)
* ``__c11_atomic_store``
* ``__c11_atomic_load``
* ``__c11_atomic_exchange``
* ``__c11_atomic_compare_exchange_strong``
* ``__c11_atomic_compare_exchange_weak``
* ``__c11_atomic_fetch_add``
* ``__c11_atomic_fetch_sub``
* ``__c11_atomic_fetch_and``
* ``__c11_atomic_fetch_or``
* ``__c11_atomic_fetch_xor``

The macros ``__ATOMIC_RELAXED``, ``__ATOMIC_CONSUME``, ``__ATOMIC_ACQUIRE``,
``__ATOMIC_RELEASE``, ``__ATOMIC_ACQ_REL``, and ``__ATOMIC_SEQ_CST`` are
provided, with values corresponding to the enumerators of C11's
``memory_order`` enumeration.

Low-level ARM exclusive memory builtins
---------------------------------------

Clang provides overloaded builtins giving direct access to the three key ARM
instructions for implementing atomic operations.

.. code-block:: c

  T __builtin_arm_ldrex(const volatile T *addr);
  T __builtin_arm_ldaex(const volatile T *addr);
  int __builtin_arm_strex(T val, volatile T *addr);
  int __builtin_arm_stlex(T val, volatile T *addr);
  void __builtin_arm_clrex(void);

The types ``T`` currently supported are:
* Integer types with width at most 64 bits (or 128 bits on AArch64).
* Floating-point types
* Pointer types.

Note that the compiler does not guarantee it will not insert stores which clear
the exclusive monitor in between an ``ldrex`` type operation and its paired
``strex``. In practice this is only usually a risk when the extra store is on
the same cache line as the variable being modified and Clang will only insert
stack stores on its own, so it is best not to use these operations on variables
with automatic storage duration.

Also, loads and stores may be implicit in code written between the ``ldrex`` and
``strex``. Clang will not necessarily mitigate the effects of these either, so
care should be exercised.

For these reasons the higher level atomic primitives should be preferred where
possible.

Non-standard C++11 Attributes
=============================

Clang's non-standard C++11 attributes live in the ``clang`` attribute
namespace.

Clang supports GCC's ``gnu`` attribute namespace. All GCC attributes which
are accepted with the ``__attribute__((foo))`` syntax are also accepted as
``[[gnu::foo]]``. This only extends to attributes which are specified by GCC
(see the list of `GCC function attributes
<http://gcc.gnu.org/onlinedocs/gcc/Function-Attributes.html>`_, `GCC variable
attributes <http://gcc.gnu.org/onlinedocs/gcc/Variable-Attributes.html>`_, and
`GCC type attributes
<http://gcc.gnu.org/onlinedocs/gcc/Type-Attributes.html>`_). As with the GCC
implementation, these attributes must appertain to the *declarator-id* in a
declaration, which means they must go either at the start of the declaration or
immediately after the name being declared.

For example, this applies the GNU ``unused`` attribute to ``a`` and ``f``, and
also applies the GNU ``noreturn`` attribute to ``f``.

.. code-block:: c++

  [[gnu::unused]] int a, f [[gnu::noreturn]] ();

Target-Specific Extensions
==========================

Clang supports some language features conditionally on some targets.

ARM/AArch64 Language Extensions
-------------------------------

Memory Barrier Intrinsics
^^^^^^^^^^^^^^^^^^^^^^^^^
Clang implements the ``__dmb``, ``__dsb`` and ``__isb`` intrinsics as defined
in the `ARM C Language Extensions Release 2.0
<http://infocenter.arm.com/help/topic/com.arm.doc.ihi0053c/IHI0053C_acle_2_0.pdf>`_.
Note that these intrinsics are implemented as motion barriers that block
reordering of memory accesses and side effect instructions. Other instructions
like simple arithmatic may be reordered around the intrinsic. If you expect to
have no reordering at all, use inline assembly instead.

X86/X86-64 Language Extensions
------------------------------

The X86 backend has these language extensions:

Memory references off the GS segment
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

Annotating a pointer with address space #256 causes it to be code generated
relative to the X86 GS segment register, and address space #257 causes it to be
relative to the X86 FS segment.  Note that this is a very very low-level
feature that should only be used if you know what you're doing (for example in
an OS kernel).

Here is an example:

.. code-block:: c++

  #define GS_RELATIVE __attribute__((address_space(256)))
  int foo(int GS_RELATIVE *P) {
    return *P;
  }

Which compiles to (on X86-32):

.. code-block:: gas

  _foo:
          movl    4(%esp), %eax
          movl    %gs:(%eax), %eax
          ret

Extensions for Static Analysis
==============================

Clang supports additional attributes that are useful for documenting program
invariants and rules for static analysis tools, such as the `Clang Static
Analyzer <http://clang-analyzer.llvm.org/>`_. These attributes are documented
in the analyzer's `list of source-level annotations
<http://clang-analyzer.llvm.org/annotations.html>`_.


Extensions for Dynamic Analysis
===============================

Use ``__has_feature(address_sanitizer)`` to check if the code is being built
with :doc:`AddressSanitizer`.

Use ``__has_feature(thread_sanitizer)`` to check if the code is being built
with :doc:`ThreadSanitizer`.

Use ``__has_feature(memory_sanitizer)`` to check if the code is being built
with :doc:`MemorySanitizer`.


Extensions for selectively disabling optimization
=================================================

Clang provides a mechanism for selectively disabling optimizations in functions
and methods.

To disable optimizations in a single function definition, the GNU-style or C++11
non-standard attribute ``optnone`` can be used.

.. code-block:: c++

  // The following functions will not be optimized.
  // GNU-style attribute
  __attribute__((optnone)) int foo() {
    // ... code
  }
  // C++11 attribute
  [[clang::optnone]] int bar() {
    // ... code
  }

To facilitate disabling optimization for a range of function definitions, a
range-based pragma is provided. Its syntax is ``#pragma clang optimize``
followed by ``off`` or ``on``.

All function definitions in the region between an ``off`` and the following
``on`` will be decorated with the ``optnone`` attribute unless doing so would
conflict with explicit attributes already present on the function (e.g. the
ones that control inlining).

.. code-block:: c++

  #pragma clang optimize off
  // This function will be decorated with optnone.
  int foo() {
    // ... code
  }

  // optnone conflicts with always_inline, so bar() will not be decorated.
  __attribute__((always_inline)) int bar() {
    // ... code
  }
  #pragma clang optimize on

If no ``on`` is found to close an ``off`` region, the end of the region is the
end of the compilation unit.

Note that a stray ``#pragma clang optimize on`` does not selectively enable
additional optimizations when compiling at low optimization levels. This feature
can only be used to selectively disable optimizations.

The pragma has an effect on functions only at the point of their definition; for
function templates, this means that the state of the pragma at the point of an
instantiation is not necessarily relevant. Consider the following example:

.. code-block:: c++

  template<typename T> T twice(T t) {
    return 2 * t;
  }

  #pragma clang optimize off
  template<typename T> T thrice(T t) {
    return 3 * t;
  }

  int container(int a, int b) {
    return twice(a) + thrice(b);
  }
  #pragma clang optimize on

In this example, the definition of the template function ``twice`` is outside
the pragma region, whereas the definition of ``thrice`` is inside the region.
The ``container`` function is also in the region and will not be optimized, but
it causes the instantiation of ``twice`` and ``thrice`` with an ``int`` type; of
these two instantiations, ``twice`` will be optimized (because its definition
was outside the region) and ``thrice`` will not be optimized.

Extensions for loop hint optimizations
======================================

The ``#pragma clang loop`` directive is used to specify hints for optimizing the
subsequent for, while, do-while, or c++11 range-based for loop. The directive
provides options for vectorization, interleaving, and unrolling. Loop hints can
be specified before any loop and will be ignored if the optimization is not safe
to apply.

Vectorization and Interleaving
------------------------------

A vectorized loop performs multiple iterations of the original loop
in parallel using vector instructions. The instruction set of the target
processor determines which vector instructions are available and their vector
widths. This restricts the types of loops that can be vectorized. The vectorizer
automatically determines if the loop is safe and profitable to vectorize. A
vector instruction cost model is used to select the vector width.

Interleaving multiple loop iterations allows modern processors to further
improve instruction-level parallelism (ILP) using advanced hardware features,
such as multiple execution units and out-of-order execution. The vectorizer uses
a cost model that depends on the register pressure and generated code size to
select the interleaving count.

Vectorization is enabled by ``vectorize(enable)`` and interleaving is enabled
by ``interleave(enable)``. This is useful when compiling with ``-Os`` to
manually enable vectorization or interleaving.

.. code-block:: c++

  #pragma clang loop vectorize(enable)
  #pragma clang loop interleave(enable)
  for(...) {
    ...
  }

The vector width is specified by ``vectorize_width(_value_)`` and the interleave
count is specified by ``interleave_count(_value_)``, where
_value_ is a positive integer. This is useful for specifying the optimal
width/count of the set of target architectures supported by your application.

.. code-block:: c++

  #pragma clang loop vectorize_width(2)
  #pragma clang loop interleave_count(2)
  for(...) {
    ...
  }

Specifying a width/count of 1 disables the optimization, and is equivalent to
``vectorize(disable)`` or ``interleave(disable)``.

Loop Unrolling
--------------

Unrolling a loop reduces the loop control overhead and exposes more
opportunities for ILP. Loops can be fully or partially unrolled. Full unrolling
eliminates the loop and replaces it with an enumerated sequence of loop
iterations. Full unrolling is only possible if the loop trip count is known at
compile time. Partial unrolling replicates the loop body within the loop and
reduces the trip count.

If ``unroll(full)`` is specified the unroller will attempt to fully unroll the
loop if the trip count is known at compile time. If the loop count is not known
or the fully unrolled code size is greater than the limit specified by the
`-pragma-unroll-threshold` command line option the loop will be partially
unrolled subject to the same limit.

.. code-block:: c++

  #pragma clang loop unroll(full)
  for(...) {
    ...
  }

The unroll count can be specified explicitly with ``unroll_count(_value_)`` where
_value_ is a positive integer. If this value is greater than the trip count the
loop will be fully unrolled. Otherwise the loop is partially unrolled subject
to the `-pragma-unroll-threshold` limit.

.. code-block:: c++

  #pragma clang loop unroll_count(8)
  for(...) {
    ...
  }

Unrolling of a loop can be prevented by specifying ``unroll(disable)``.

Additional Information
----------------------

For convenience multiple loop hints can be specified on a single line.

.. code-block:: c++

  #pragma clang loop vectorize_width(4) interleave_count(8)
  for(...) {
    ...
  }

If an optimization cannot be applied any hints that apply to it will be ignored.
For example, the hint ``vectorize_width(4)`` is ignored if the loop is not
proven safe to vectorize. To identify and diagnose optimization issues use
`-Rpass`, `-Rpass-missed`, and `-Rpass-analysis` command line options. See the
user guide for details.