<html> <head> <title>Clang Language Extensions</title> <link type="text/css" rel="stylesheet" href="../menu.css" /> <link type="text/css" rel="stylesheet" href="../content.css" /> <style type="text/css"> td { vertical-align: top; } </style> </head> <body> <!--#include virtual="../menu.html.incl"--> <div id="content"> <h1>Clang Language Extensions</h1> <ul> <li><a href="#intro">Introduction</a></li> <li><a href="#feature_check">Feature Checking Macros</a></li> <li><a href="#builtinmacros">Builtin Macros</a></li> <li><a href="#vectors">Vectors and Extended Vectors</a></li> <li><a href="#blocks">Blocks</a></li> <li><a href="#overloading-in-c">Function Overloading in C</a></li> <li><a href="#builtins">Builtin Functions</a> <ul> <li><a href="#__builtin_shufflevector">__builtin_shufflevector</a></li> </ul> </li> <li><a href="#targetspecific">Target-Specific Extensions</a> <ul> <li><a href="#x86-specific">X86/X86-64 Language Extensions</a></li> </ul> </li> <li><a href="#analyzerspecific">Static Analysis-Specific Extensions</a> <ul> <li><a href="#analyzerattributes">Analyzer Attributes</a></li> </ul> </li> </ul> <!-- ======================================================================= --> <h2 id="intro">Introduction</h2> <!-- ======================================================================= --> <p>This document describes the language extensions provided by Clang. In addition to the language extensions listed here, Clang aims to support a broad range of GCC extensions. Please see the <a href="http://gcc.gnu.org/onlinedocs/gcc/C-Extensions.html">GCC manual</a> for more information on these extensions.</p> <!-- ======================================================================= --> <h2 id="feature_check">Feature Checking Macros</h2> <!-- ======================================================================= --> <p>Language extensions can be very useful, but only if you know you can depend on them. In order to allow fine-grain features checks, we support two builtin function-like macros. This allows you to directly test for a feature in your code without having to resort to something like autoconf or fragile "compiler version checks".</p> <!-- ======================================================================= --> <h3 id="__has_builtin">__has_builtin</h3> <!-- ======================================================================= --> <p>This function-like macro takes a single identifier argument that is the name of a builtin function. It evaluates to 1 if the builtin is supported or 0 if not. It can be used like this:</p> <blockquote> <pre> #ifndef __has_builtin // Optional of course. #define __has_builtin(x) 0 // Compatibility with non-clang compilers. #endif ... #if __has_builtin(__builtin_trap) __builtin_trap(); #else abort(); #endif ... </pre> </blockquote> <!-- ======================================================================= --> <h3 id="__has_feature">__has_feature</h3> <!-- ======================================================================= --> <p>This function-like macro takes a single identifier argument that is the name of a feature. It evaluates to 1 if the feature is supported or 0 if not. It can be used like this:</p> <blockquote> <pre> #ifndef __has_feature // Optional of course. #define __has_feature(x) 0 // Compatibility with non-clang compilers. #endif ... #if __has_feature(attribute_overloadable) || \ __has_feature(blocks) ... #endif ... </pre> </blockquote> <p>The feature tag is described along with the language feature below.</p> <!-- ======================================================================= --> <h2 id="builtinmacros">Builtin Macros</h2> <!-- ======================================================================= --> <p>__BASE_FILE__, __INCLUDE_LEVEL__, __TIMESTAMP__, __COUNTER__</p> <!-- ======================================================================= --> <h2 id="vectors">Vectors and Extended Vectors</h2> <!-- ======================================================================= --> <p>Supports the GCC vector extensions, plus some stuff like V[1]. ext_vector with V.xyzw syntax and other tidbits. See also <a href="#__builtin_shufflevector">__builtin_shufflevector</a>.</p> <p>Query for this feature with __has_feature(attribute_ext_vector_type).</p> <!-- ======================================================================= --> <h2 id="blocks">Blocks</h2> <!-- ======================================================================= --> <p>The syntax and high level language feature description is in <a href="BlockLanguageSpec.txt">BlockLanguageSpec.txt</a>. Implementation and ABI details for the clang implementation are in <a href="BlockImplementation.txt">BlockImplementation.txt</a>.</p> <p>Query for this feature with __has_feature(blocks).</p> <!-- ======================================================================= --> <h2 id="overloading-in-c">Function Overloading in C</h2> <!-- ======================================================================= --> <p>Clang provides support for C++ function overloading in C. Function overloading in C is introduced using the <tt>overloadable</tt> attribute. For example, one might provide several overloaded versions of a <tt>tgsin</tt> function that invokes the appropriate standard function computing the sine of a value with <tt>float</tt>, <tt>double</tt>, or <tt>long double</tt> precision:</p> <blockquote> <pre> #include <math.h> float <b>__attribute__((overloadable))</b> tgsin(float x) { return sinf(x); } double <b>__attribute__((overloadable))</b> tgsin(double x) { return sin(x); } long double <b>__attribute__((overloadable))</b> tgsin(long double x) { return sinl(x); } </pre> </blockquote> <p>Given these declarations, one can call <tt>tgsin</tt> with a <tt>float</tt> value to receive a <tt>float</tt> result, with a <tt>double</tt> to receive a <tt>double</tt> result, etc. Function overloading in C follows the rules of C++ function overloading to pick the best overload given the call arguments, with a few C-specific semantics:</p> <ul> <li>Conversion from <tt>float</tt> or <tt>double</tt> to <tt>long double</tt> is ranked as a floating-point promotion (per C99) rather than as a floating-point conversion (as in C++).</li> <li>A conversion from a pointer of type <tt>T*</tt> to a pointer of type <tt>U*</tt> is considered a pointer conversion (with conversion rank) if <tt>T</tt> and <tt>U</tt> are compatible types.</li> <li>A conversion from type <tt>T</tt> to a value of type <tt>U</tt> is permitted if <tt>T</tt> and <tt>U</tt> are compatible types. This conversion is given "conversion" rank.</li> </ul> <p>The declaration of <tt>overloadable</tt> functions is restricted to function declarations and definitions. Most importantly, if any function with a given name is given the <tt>overloadable</tt> attribute, then all function declarations and definitions with that name (and in that scope) must have the <tt>overloadable</tt> attribute. This rule even applies to redeclarations of functions whose original declaration had the <tt>overloadable</tt> attribute, e.g.,</p> <blockquote> <pre> int f(int) __attribute__((overloadable)); float f(float); <i>// error: declaration of "f" must have the "overloadable" attribute</i> int g(int) __attribute__((overloadable)); int g(int) { } <i>// error: redeclaration of "g" must also have the "overloadable" attribute</i> </pre> </blockquote> <p>Functions marked <tt>overloadable</tt> must have prototypes. Therefore, the following code is ill-formed:</p> <blockquote> <pre> int h() __attribute__((overloadable)); <i>// error: h does not have a prototype</i> </pre> </blockquote> <p>However, <tt>overloadable</tt> functions are allowed to use a ellipsis even if there are no named parameters (as is permitted in C++). This feature is particularly useful when combined with the <tt>unavailable</tt> attribute:</p> <blockquote> <pre> void honeypot(...) __attribute__((overloadable, unavailable)); <i>// calling me is an error</i> </pre> </blockquote> <p>Functions declared with the <tt>overloadable</tt> attribute have their names mangled according to the same rules as C++ function names. For example, the three <tt>tgsin</tt> functions in our motivating example get the mangled names <tt>_Z5tgsinf</tt>, <tt>_Z5tgsind</tt>, and <tt>Z5tgsine</tt>, respectively. There are two caveats to this use of name mangling:</p> <ul> <li>Future versions of Clang may change the name mangling of functions overloaded in C, so you should not depend on an specific mangling. To be completely safe, we strongly urge the use of <tt>static inline</tt> with <tt>overloadable</tt> functions.</li> <li>The <tt>overloadable</tt> attribute has almost no meaning when used in C++, because names will already be mangled and functions are already overloadable. However, when an <tt>overloadable</tt> function occurs within an <tt>extern "C"</tt> linkage specification, it's name <i>will</i> be mangled in the same way as it would in C.</li> </ul> <p>Query for this feature with __has_feature(attribute_overloadable).</p> <!-- ======================================================================= --> <h2 id="builtins">Builtin Functions</h2> <!-- ======================================================================= --> <p>Clang supports a number of builtin library functions with the same syntax as GCC, including things like <tt>__builtin_nan</tt>, <tt>__builtin_constant_p</tt>, <tt>__builtin_choose_expr</tt>, <tt>__builtin_types_compatible_p</tt>, <tt>__sync_fetch_and_add</tt>, etc. In addition to the GCC builtins, Clang supports a number of builtins that GCC does not, which are listed here.</p> <p>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 <tt><xmmintrin.h></tt>, which define portable wrappers for these. Many of the Clang versions of these functions are implemented directly in terms of <a href="#vectors">extended vector support</a> instead of builtins, in order to reduce the number of builtins that we need to implement.</p> <!-- ======================================================================= --> <h3 id="__builtin_shufflevector">__builtin_shufflevector</h3> <!-- ======================================================================= --> <p><tt>__builtin_shufflevector</tt> is used to expression generic vector permutation/shuffle/swizzle operations. This builtin is also very important for the implementation of various target-specific header files like <tt><xmmintrin.h></tt>. </p> <p><b>Syntax:</b></p> <pre> __builtin_shufflevector(vec1, vec2, index1, index2, ...) </pre> <p><b>Examples:</b></p> <pre> // 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) </pre> <p><b>Description:</b></p> <p>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. </p> <p>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. </p> <!-- ======================================================================= --> <h2 id="targetspecific">Target-Specific Extensions</h2> <!-- ======================================================================= --> <p>Clang supports some language features conditionally on some targets.</p> <!-- ======================================================================= --> <h3 id="x86-specific">X86/X86-64 Language Extensions</h3> <!-- ======================================================================= --> <p>The X86 backend has these language extensions:</p> <!-- ======================================================================= --> <h4 id="x86-gs-segment">Memory references off the GS segment</h4> <!-- ======================================================================= --> <p>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).</p> <p>Here is an example:</p> <pre> #define GS_RELATIVE __attribute__((address_space(256))) int foo(int GS_RELATIVE *P) { return *P; } </pre> <p>Which compiles to (on X86-32):</p> <pre> _foo: movl 4(%esp), %eax movl %gs:(%eax), %eax ret </pre> <!-- ======================================================================= --> <h2 id="analyzerspecific">Static Analysis-Specific Extensions</h2> <!-- ======================================================================= --> <p>Clang supports additional attributes that are useful for documenting program invariants and rules for static analysis tools. The extensions documented here are used by the <a href="http://clang.llvm.org/StaticAnalysis.html">path-sensitive static analyzer engine</a> that is part of Clang's Analysis library.</p> <!-- ======================================================================= --> <h3 id="analyzerattributes">Analyzer Attributes</h3> <!-- ======================================================================= --> <h4 id="attr_analyzer_noreturn"><tt>analyzer_noreturn</tt></h4> <p>Clang's static analysis engine understands the standard <tt>noreturn</tt> attribute. This attribute, which is typically affixed to a function prototype, indicates that a call to a given function never returns. Function prototypes for common functions like <tt>exit</tt> are typically annotated with this attribute, as well as a variety of common assertion handlers. Users can educate the static analyzer about their own custom assertion handles (thus cutting down on false positives due to false paths) by marking their own "panic" functions with this attribute.</p> <p>While useful, <tt>noreturn</tt> is not applicable in all cases. Sometimes there are special functions that for all intents and purposes should be considered panic functions (i.e., they are only called when an internal program error occurs) but may actually return so that the program can fail gracefully. The <tt>analyzer_noreturn</tt> attribute allows one to annotate such functions as being interpreted as "no return" functions by the analyzer (thus pruning bogus paths) but will not affect compilation (as in the case of <tt>noreturn</tt>).</p> <p><b>Usage</b>: The <tt>analyzer_noreturn</tt> attribute can be placed in the same places where the <tt>noreturn</tt> attribute can be placed. It is commonly placed at the end of function prototypes:</p> <pre> void foo() <b>__attribute__((analyzer_noreturn))</b>; </pre> <p>Query for this feature with __has_feature(attribute_analyzer_noreturn).</p> </div> </body> </html>