1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <title>LLVM Assembly Language Reference Manual</title>
6 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
7 <meta name="author" content="Chris Lattner">
8 <meta name="description"
9 content="LLVM Assembly Language Reference Manual.">
10 <link rel="stylesheet" href="llvm.css" type="text/css">
15 <div class="doc_title"> LLVM Language Reference Manual </div>
17 <li><a href="#abstract">Abstract</a></li>
18 <li><a href="#introduction">Introduction</a></li>
19 <li><a href="#identifiers">Identifiers</a></li>
20 <li><a href="#typesystem">Type System</a>
22 <li><a href="#t_primitive">Primitive Types</a>
24 <li><a href="#t_classifications">Type Classifications</a></li>
27 <li><a href="#t_derived">Derived Types</a>
29 <li><a href="#t_array">Array Type</a></li>
30 <li><a href="#t_function">Function Type</a></li>
31 <li><a href="#t_pointer">Pointer Type</a></li>
32 <li><a href="#t_struct">Structure Type</a></li>
33 <li><a href="#t_packed">Packed Type</a></li>
38 <li><a href="#highlevel">High Level Structure</a>
40 <li><a href="#modulestructure">Module Structure</a></li>
41 <li><a href="#globalvars">Global Variables</a></li>
42 <li><a href="#functionstructure">Function Structure</a></li>
45 <li><a href="#instref">Instruction Reference</a>
47 <li><a href="#terminators">Terminator Instructions</a>
49 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
50 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
51 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
52 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
53 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
56 <li><a href="#binaryops">Binary Operations</a>
58 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
59 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
60 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
61 <li><a href="#i_div">'<tt>div</tt>' Instruction</a></li>
62 <li><a href="#i_rem">'<tt>rem</tt>' Instruction</a></li>
63 <li><a href="#i_setcc">'<tt>set<i>cc</i></tt>' Instructions</a></li>
66 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
68 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
69 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
70 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
71 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
72 <li><a href="#i_shr">'<tt>shr</tt>' Instruction</a></li>
75 <li><a href="#memoryops">Memory Access Operations</a>
77 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
78 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
79 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
80 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
81 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
82 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
85 <li><a href="#otherops">Other Operations</a>
87 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
88 <li><a href="#i_cast">'<tt>cast .. to</tt>' Instruction</a></li>
89 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
90 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
91 <li><a href="#i_vanext">'<tt>vanext</tt>' Instruction</a></li>
92 <li><a href="#i_vaarg">'<tt>vaarg</tt>' Instruction</a></li>
97 <li><a href="#intrinsics">Intrinsic Functions</a>
99 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
101 <li><a href="#i_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
102 <li><a href="#i_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
103 <li><a href="#i_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
106 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
108 <li><a href="#i_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
109 <li><a href="#i_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
110 <li><a href="#i_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
113 <li><a href="#int_codegen">Code Generator Intrinsics</a>
115 <li><a href="#i_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
116 <li><a href="#i_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
119 <li><a href="#int_os">Operating System Intrinsics</a>
121 <li><a href="#i_readport">'<tt>llvm.readport</tt>' Intrinsic</a></li>
122 <li><a href="#i_writeport">'<tt>llvm.writeport</tt>' Intrinsic</a></li>
123 <li><a href="#i_readio">'<tt>llvm.readio</tt>' Intrinsic</a></li>
124 <li><a href="#i_writeio">'<tt>llvm.writeio</tt>' Intrinsic</a></li>
126 <li><a href="#int_libc">Standard C Library Intrinsics</a>
128 <li><a href="#i_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a></li>
129 <li><a href="#i_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a></li>
130 <li><a href="#i_memset">'<tt>llvm.memset</tt>' Intrinsic</a></li>
131 <li><a href="#i_isunordered">'<tt>llvm.isunordered</tt>' Intrinsic</a></li>
134 <li><a href="#int_debugger">Debugger intrinsics</a></li>
139 <div class="doc_author">
140 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
141 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
144 <!-- *********************************************************************** -->
145 <div class="doc_section"> <a name="abstract">Abstract </a></div>
146 <!-- *********************************************************************** -->
148 <div class="doc_text">
149 <p>This document is a reference manual for the LLVM assembly language.
150 LLVM is an SSA based representation that provides type safety,
151 low-level operations, flexibility, and the capability of representing
152 'all' high-level languages cleanly. It is the common code
153 representation used throughout all phases of the LLVM compilation
157 <!-- *********************************************************************** -->
158 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
159 <!-- *********************************************************************** -->
161 <div class="doc_text">
163 <p>The LLVM code representation is designed to be used in three
164 different forms: as an in-memory compiler IR, as an on-disk bytecode
165 representation (suitable for fast loading by a Just-In-Time compiler),
166 and as a human readable assembly language representation. This allows
167 LLVM to provide a powerful intermediate representation for efficient
168 compiler transformations and analysis, while providing a natural means
169 to debug and visualize the transformations. The three different forms
170 of LLVM are all equivalent. This document describes the human readable
171 representation and notation.</p>
173 <p>The LLVM representation aims to be a light-weight and low-level
174 while being expressive, typed, and extensible at the same time. It
175 aims to be a "universal IR" of sorts, by being at a low enough level
176 that high-level ideas may be cleanly mapped to it (similar to how
177 microprocessors are "universal IR's", allowing many source languages to
178 be mapped to them). By providing type information, LLVM can be used as
179 the target of optimizations: for example, through pointer analysis, it
180 can be proven that a C automatic variable is never accessed outside of
181 the current function... allowing it to be promoted to a simple SSA
182 value instead of a memory location.</p>
186 <!-- _______________________________________________________________________ -->
187 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
189 <div class="doc_text">
191 <p>It is important to note that this document describes 'well formed'
192 LLVM assembly language. There is a difference between what the parser
193 accepts and what is considered 'well formed'. For example, the
194 following instruction is syntactically okay, but not well formed:</p>
197 %x = <a href="#i_add">add</a> int 1, %x
200 <p>...because the definition of <tt>%x</tt> does not dominate all of
201 its uses. The LLVM infrastructure provides a verification pass that may
202 be used to verify that an LLVM module is well formed. This pass is
203 automatically run by the parser after parsing input assembly, and by
204 the optimizer before it outputs bytecode. The violations pointed out
205 by the verifier pass indicate bugs in transformation passes or input to
208 <!-- Describe the typesetting conventions here. --> </div>
210 <!-- *********************************************************************** -->
211 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
212 <!-- *********************************************************************** -->
214 <div class="doc_text">
216 <p>LLVM uses three different forms of identifiers, for different
220 <li>Numeric constants are represented as you would expect: 12, -3
221 123.421, etc. Floating point constants have an optional hexadecimal
223 <li>Named values are represented as a string of characters with a '%'
224 prefix. For example, %foo, %DivisionByZero,
225 %a.really.long.identifier. The actual regular expression used is '<tt>%[a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
226 Identifiers which require other characters in their names can be
227 surrounded with quotes. In this way, anything except a <tt>"</tt>
228 character can be used in a name.</li>
229 <li>Unnamed values are represented as an unsigned numeric value with
230 a '%' prefix. For example, %12, %2, %44.</li>
232 <p>LLVM requires that values start with a '%' sign for two reasons:
233 Compilers don't need to worry about name clashes with reserved words,
234 and the set of reserved words may be expanded in the future without
235 penalty. Additionally, unnamed identifiers allow a compiler to quickly
236 come up with a temporary variable without having to avoid symbol table
238 <p>Reserved words in LLVM are very similar to reserved words in other
239 languages. There are keywords for different opcodes ('<tt><a
240 href="#i_add">add</a></tt>', '<tt><a href="#i_cast">cast</a></tt>', '<tt><a
241 href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
242 href="#t_void">void</a></tt>', '<tt><a href="#t_uint">uint</a></tt>',
243 etc...), and others. These reserved words cannot conflict with
244 variable names, because none of them start with a '%' character.</p>
245 <p>Here is an example of LLVM code to multiply the integer variable '<tt>%X</tt>'
248 <pre> %result = <a href="#i_mul">mul</a> uint %X, 8<br></pre>
249 <p>After strength reduction:</p>
250 <pre> %result = <a href="#i_shl">shl</a> uint %X, ubyte 3<br></pre>
251 <p>And the hard way:</p>
252 <pre> <a href="#i_add">add</a> uint %X, %X <i>; yields {uint}:%0</i>
254 href="#i_add">add</a> uint %0, %0 <i>; yields {uint}:%1</i>
256 href="#i_add">add</a> uint %1, %1<br></pre>
257 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
258 important lexical features of LLVM:</p>
260 <li>Comments are delimited with a '<tt>;</tt>' and go until the end
262 <li>Unnamed temporaries are created when the result of a computation
263 is not assigned to a named value.</li>
264 <li>Unnamed temporaries are numbered sequentially</li>
266 <p>...and it also show a convention that we follow in this document.
267 When demonstrating instructions, we will follow an instruction with a
268 comment that defines the type and name of value produced. Comments are
269 shown in italic text.</p>
270 <p>The one non-intuitive notation for constants is the optional
271 hexidecimal form of floating point constants. For example, the form '<tt>double
272 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
273 4.5e+15</tt>' which is also supported by the parser. The only time
274 hexadecimal floating point constants are useful (and the only time that
275 they are generated by the disassembler) is when an FP constant has to
276 be emitted that is not representable as a decimal floating point number
277 exactly. For example, NaN's, infinities, and other special cases are
278 represented in their IEEE hexadecimal format so that assembly and
279 disassembly do not cause any bits to change in the constants.</p>
281 <!-- *********************************************************************** -->
282 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
283 <!-- *********************************************************************** -->
284 <div class="doc_text">
285 <p>The LLVM type system is one of the most important features of the
286 intermediate representation. Being typed enables a number of
287 optimizations to be performed on the IR directly, without having to do
288 extra analyses on the side before the transformation. A strong type
289 system makes it easier to read the generated code and enables novel
290 analyses and transformations that are not feasible to perform on normal
291 three address code representations.</p>
292 <!-- The written form for the type system was heavily influenced by the
293 syntactic problems with types in the C language<sup><a
294 href="#rw_stroustrup">1</a></sup>.<p> --> </div>
295 <!-- ======================================================================= -->
296 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
297 <div class="doc_text">
298 <p>The primitive types are the fundamental building blocks of the LLVM
299 system. The current set of primitive types are as follows:</p>
301 <table border="0" style="align: center">
305 <table border="1" cellspacing="0" cellpadding="4" style="align: center">
308 <td><tt>void</tt></td>
312 <td><tt>ubyte</tt></td>
313 <td>Unsigned 8 bit value</td>
316 <td><tt>ushort</tt></td>
317 <td>Unsigned 16 bit value</td>
320 <td><tt>uint</tt></td>
321 <td>Unsigned 32 bit value</td>
324 <td><tt>ulong</tt></td>
325 <td>Unsigned 64 bit value</td>
328 <td><tt>float</tt></td>
329 <td>32 bit floating point value</td>
332 <td><tt>label</tt></td>
333 <td>Branch destination</td>
339 <table border="1" cellspacing="0" cellpadding="4">
342 <td><tt>bool</tt></td>
343 <td>True or False value</td>
346 <td><tt>sbyte</tt></td>
347 <td>Signed 8 bit value</td>
350 <td><tt>short</tt></td>
351 <td>Signed 16 bit value</td>
354 <td><tt>int</tt></td>
355 <td>Signed 32 bit value</td>
358 <td><tt>long</tt></td>
359 <td>Signed 64 bit value</td>
362 <td><tt>double</tt></td>
363 <td>64 bit floating point value</td>
373 <!-- _______________________________________________________________________ -->
374 <div class="doc_subsubsection"> <a name="t_classifications">Type
375 Classifications</a> </div>
376 <div class="doc_text">
377 <p>These different primitive types fall into a few useful
380 <table border="1" cellspacing="0" cellpadding="4">
383 <td><a name="t_signed">signed</a></td>
384 <td><tt>sbyte, short, int, long, float, double</tt></td>
387 <td><a name="t_unsigned">unsigned</a></td>
388 <td><tt>ubyte, ushort, uint, ulong</tt></td>
391 <td><a name="t_integer">integer</a></td>
392 <td><tt>ubyte, sbyte, ushort, short, uint, int, ulong, long</tt></td>
395 <td><a name="t_integral">integral</a></td>
396 <td><tt>bool, ubyte, sbyte, ushort, short, uint, int, ulong, long</tt>
400 <td><a name="t_floating">floating point</a></td>
401 <td><tt>float, double</tt></td>
404 <td><a name="t_firstclass">first class</a></td>
405 <td><tt>bool, ubyte, sbyte, ushort, short, uint, int, ulong, long,<br>
406 float, double, <a href="#t_pointer">pointer</a>,
407 <a href="#t_packed">packed</a></tt></td>
412 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
413 most important. Values of these types are the only ones which can be
414 produced by instructions, passed as arguments, or used as operands to
415 instructions. This means that all structures and arrays must be
416 manipulated either by pointer or by component.</p>
418 <!-- ======================================================================= -->
419 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
420 <div class="doc_text">
421 <p>The real power in LLVM comes from the derived types in the system.
422 This is what allows a programmer to represent arrays, functions,
423 pointers, and other useful types. Note that these derived types may be
424 recursive: For example, it is possible to have a two dimensional array.</p>
426 <!-- _______________________________________________________________________ -->
427 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
428 <div class="doc_text">
430 <p>The array type is a very simple derived type that arranges elements
431 sequentially in memory. The array type requires a size (number of
432 elements) and an underlying data type.</p>
434 <pre> [<# elements> x <elementtype>]<br></pre>
435 <p>The number of elements is a constant integer value, elementtype may
436 be any type with a size.</p>
438 <p> <tt>[40 x int ]</tt>: Array of 40 integer values.<br>
439 <tt>[41 x int ]</tt>: Array of 41 integer values.<br>
440 <tt>[40 x uint]</tt>: Array of 40 unsigned integer values.</p>
442 <p>Here are some examples of multidimensional arrays:</p>
444 <table border="0" cellpadding="0" cellspacing="0">
447 <td><tt>[3 x [4 x int]]</tt></td>
448 <td>: 3x4 array integer values.</td>
451 <td><tt>[12 x [10 x float]]</tt></td>
452 <td>: 12x10 array of single precision floating point values.</td>
455 <td><tt>[2 x [3 x [4 x uint]]]</tt></td>
456 <td>: 2x3x4 array of unsigned integer values.</td>
462 <!-- _______________________________________________________________________ -->
463 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
464 <div class="doc_text">
466 <p>The function type can be thought of as a function signature. It
467 consists of a return type and a list of formal parameter types.
468 Function types are usually used to build virtual function tables
469 (which are structures of pointers to functions), for indirect function
470 calls, and when defining a function.</p>
472 The return type of a function type cannot be an aggregate type.
475 <pre> <returntype> (<parameter list>)<br></pre>
476 <p>Where '<tt><parameter list></tt>' is a comma-separated list of type
477 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
478 which indicates that the function takes a variable number of arguments.
479 Variable argument functions can access their arguments with the <a
480 href="#int_varargs">variable argument handling intrinsic</a> functions.</p>
483 <table border="0" cellpadding="0" cellspacing="0">
486 <td><tt>int (int)</tt></td>
487 <td>: function taking an <tt>int</tt>, returning an <tt>int</tt></td>
490 <td><tt>float (int, int *) *</tt></td>
491 <td>: <a href="#t_pointer">Pointer</a> to a function that takes an
492 <tt>int</tt> and a <a href="#t_pointer">pointer</a> to <tt>int</tt>,
493 returning <tt>float</tt>.</td>
496 <td><tt>int (sbyte *, ...)</tt></td>
497 <td>: A vararg function that takes at least one <a
498 href="#t_pointer">pointer</a> to <tt>sbyte</tt> (signed char in C), which
499 returns an integer. This is the signature for <tt>printf</tt> in
506 <!-- _______________________________________________________________________ -->
507 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
508 <div class="doc_text">
510 <p>The structure type is used to represent a collection of data members
511 together in memory. The packing of the field types is defined to match
512 the ABI of the underlying processor. The elements of a structure may
513 be any type that has a size.</p>
514 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
515 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
516 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
519 <pre> { <type list> }<br></pre>
522 <table border="0" cellpadding="0" cellspacing="0">
525 <td><tt>{ int, int, int }</tt></td>
526 <td>: a triple of three <tt>int</tt> values</td>
529 <td><tt>{ float, int (int) * }</tt></td>
530 <td>: A pair, where the first element is a <tt>float</tt> and the second
531 element is a <a href="#t_pointer">pointer</a> to a <a
532 href="#t_function">function</a> that takes an <tt>int</tt>, returning an
539 <!-- _______________________________________________________________________ -->
540 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
541 <div class="doc_text">
543 <p>As in many languages, the pointer type represents a pointer or
544 reference to another object, which must live in memory.</p>
546 <pre> <type> *<br></pre>
549 <table border="0" cellpadding="0" cellspacing="0">
552 <td><tt>[4x int]*</tt></td>
553 <td>: <a href="#t_pointer">pointer</a> to <a href="#t_array">array</a> of
554 four <tt>int</tt> values</td>
557 <td><tt>int (int *) *</tt></td>
558 <td>: A <a href="#t_pointer">pointer</a> to a <a
559 href="#t_function">function</a> that takes an <tt>int</tt>, returning an
565 <!-- _______________________________________________________________________ -->
566 <div class="doc_subsubsection"> <a name="t_packed">Packed Type</a> </div>
567 <div class="doc_text">
569 <p>A packed type is a simple derived type that represents a vector
570 of elements. Packed types are used when multiple primitive data
571 are operated in parallel using a single instruction (SIMD).
572 A packed type requires a size (number of
573 elements) and an underlying primitive data type. Packed types are
574 considered <a href="#t_firstclass">first class</a>.</p>
576 <pre> < <# elements> x <elementtype> ><br></pre>
577 <p>The number of elements is a constant integer value, elementtype may
578 be any integral or floating point type.</p>
580 <p> <tt><4 x int></tt>: Packed vector of 4 integer values.<br>
581 <tt><8 x float></tt>: Packed vector of 8 floating-point values.<br>
582 <tt><2 x uint></tt>: Packed vector of 2 unsigned integer values.</p>
586 <!-- *********************************************************************** -->
587 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
588 <!-- *********************************************************************** -->
589 <!-- ======================================================================= -->
590 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
592 <div class="doc_text">
593 <p>LLVM programs are composed of "Module"s, each of which is a
594 translation unit of the input programs. Each module consists of
595 functions, global variables, and symbol table entries. Modules may be
596 combined together with the LLVM linker, which merges function (and
597 global variable) definitions, resolves forward declarations, and merges
598 symbol table entries. Here is an example of the "hello world" module:</p>
599 <pre><i>; Declare the string constant as a global constant...</i>
600 <a href="#identifiers">%.LC0</a> = <a href="#linkage_internal">internal</a> <a
601 href="#globalvars">constant</a> <a href="#t_array">[13 x sbyte]</a> c"hello world\0A\00" <i>; [13 x sbyte]*</i>
603 <i>; External declaration of the puts function</i>
604 <a href="#functionstructure">declare</a> int %puts(sbyte*) <i>; int(sbyte*)* </i>
606 <i>; Definition of main function</i>
607 int %main() { <i>; int()* </i>
608 <i>; Convert [13x sbyte]* to sbyte *...</i>
610 href="#i_getelementptr">getelementptr</a> [13 x sbyte]* %.LC0, long 0, long 0 <i>; sbyte*</i>
612 <i>; Call puts function to write out the string to stdout...</i>
614 href="#i_call">call</a> int %puts(sbyte* %cast210) <i>; int</i>
616 href="#i_ret">ret</a> int 0<br>}<br></pre>
617 <p>This example is made up of a <a href="#globalvars">global variable</a>
618 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
619 function, and a <a href="#functionstructure">function definition</a>
620 for "<tt>main</tt>".</p>
621 <a name="linkage"> In general, a module is made up of a list of global
622 values, where both functions and global variables are global values.
623 Global values are represented by a pointer to a memory location (in
624 this case, a pointer to an array of char, and a pointer to a function),
625 and have one of the following linkage types:</a>
628 <dt><tt><b><a name="linkage_internal">internal</a></b></tt> </dt>
629 <dd>Global values with internal linkage are only directly accessible
630 by objects in the current module. In particular, linking code into a
631 module with an internal global value may cause the internal to be
632 renamed as necessary to avoid collisions. Because the symbol is
633 internal to the module, all references can be updated. This
634 corresponds to the notion of the '<tt>static</tt>' keyword in C, or the
635 idea of "anonymous namespaces" in C++.
638 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
639 <dd>"<tt>linkonce</tt>" linkage is similar to <tt>internal</tt>
640 linkage, with the twist that linking together two modules defining the
641 same <tt>linkonce</tt> globals will cause one of the globals to be
642 discarded. This is typically used to implement inline functions.
643 Unreferenced <tt>linkonce</tt> globals are allowed to be discarded.
646 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
647 <dd>"<tt>weak</tt>" linkage is exactly the same as <tt>linkonce</tt>
648 linkage, except that unreferenced <tt>weak</tt> globals may not be
649 discarded. This is used to implement constructs in C such as "<tt>int
650 X;</tt>" at global scope.
653 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
654 <dd>"<tt>appending</tt>" linkage may only be applied to global
655 variables of pointer to array type. When two global variables with
656 appending linkage are linked together, the two global arrays are
657 appended together. This is the LLVM, typesafe, equivalent of having
658 the system linker append together "sections" with identical names when
662 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
663 <dd>If none of the above identifiers are used, the global is
664 externally visible, meaning that it participates in linkage and can be
665 used to resolve external symbol references.
670 <p><a name="linkage_external">For example, since the "<tt>.LC0</tt>"
671 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
672 variable and was linked with this one, one of the two would be renamed,
673 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
674 external (i.e., lacking any linkage declarations), they are accessible
675 outside of the current module. It is illegal for a function <i>declaration</i>
676 to have any linkage type other than "externally visible".</a></p>
679 <!-- ======================================================================= -->
680 <div class="doc_subsection">
681 <a name="globalvars">Global Variables</a>
684 <div class="doc_text">
686 <p>Global variables define regions of memory allocated at compilation
687 time instead of run-time. Global variables may optionally be
688 initialized. A variable may be defined as a global "constant", which
689 indicates that the contents of the variable will never be modified
690 (opening options for optimization).</p>
692 <p>As SSA values, global variables define pointer values that are in
693 scope (i.e. they dominate) for all basic blocks in the program. Global
694 variables always define a pointer to their "content" type because they
695 describe a region of memory, and all memory objects in LLVM are
696 accessed through pointers.</p>
701 <!-- ======================================================================= -->
702 <div class="doc_subsection">
703 <a name="functionstructure">Functions</a>
706 <div class="doc_text">
708 <p>LLVM function definitions are composed of a (possibly empty) argument list,
709 an opening curly brace, a list of basic blocks, and a closing curly brace. LLVM
710 function declarations are defined with the "<tt>declare</tt>" keyword, a
711 function name, and a function signature.</p>
713 <p>A function definition contains a list of basic blocks, forming the CFG for
714 the function. Each basic block may optionally start with a label (giving the
715 basic block a symbol table entry), contains a list of instructions, and ends
716 with a <a href="#terminators">terminator</a> instruction (such as a branch or
717 function return).</p>
719 <p>The first basic block in program is special in two ways: it is immediately
720 executed on entrance to the function, and it is not allowed to have predecessor
721 basic blocks (i.e. there can not be any branches to the entry block of a
722 function). Because the block can have no predecessors, it also cannot have any
723 <a href="#i_phi">PHI nodes</a>.</p>
725 <p>LLVM functions are identified by their name and type signature. Hence, two
726 functions with the same name but different parameter lists or return values are
727 considered different functions, and LLVM will resolves references to each
733 <!-- *********************************************************************** -->
734 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
735 <!-- *********************************************************************** -->
736 <div class="doc_text">
737 <p>The LLVM instruction set consists of several different
738 classifications of instructions: <a href="#terminators">terminator
739 instructions</a>, <a href="#binaryops">binary instructions</a>, <a
740 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
741 instructions</a>.</p>
743 <!-- ======================================================================= -->
744 <div class="doc_subsection"> <a name="terminators">Terminator
745 Instructions</a> </div>
746 <div class="doc_text">
747 <p>As mentioned <a href="#functionstructure">previously</a>, every
748 basic block in a program ends with a "Terminator" instruction, which
749 indicates which block should be executed after the current block is
750 finished. These terminator instructions typically yield a '<tt>void</tt>'
751 value: they produce control flow, not values (the one exception being
752 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
753 <p>There are five different terminator instructions: the '<a
754 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
755 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
756 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, and the '<a
757 href="#i_unwind"><tt>unwind</tt></a>' instruction.</p>
759 <!-- _______________________________________________________________________ -->
760 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
761 Instruction</a> </div>
762 <div class="doc_text">
764 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
765 ret void <i>; Return from void function</i>
768 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
769 value) from a function, back to the caller.</p>
770 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
771 returns a value and then causes control flow, and one that just causes
772 control flow to occur.</p>
774 <p>The '<tt>ret</tt>' instruction may return any '<a
775 href="#t_firstclass">first class</a>' type. Notice that a function is
776 not <a href="#wellformed">well formed</a> if there exists a '<tt>ret</tt>'
777 instruction inside of the function that returns a value that does not
778 match the return type of the function.</p>
780 <p>When the '<tt>ret</tt>' instruction is executed, control flow
781 returns back to the calling function's context. If the caller is a "<a
782 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
783 the instruction after the call. If the caller was an "<a
784 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
785 at the beginning "normal" of the destination block. If the instruction
786 returns a value, that value shall set the call or invoke instruction's
789 <pre> ret int 5 <i>; Return an integer value of 5</i>
790 ret void <i>; Return from a void function</i>
793 <!-- _______________________________________________________________________ -->
794 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
795 <div class="doc_text">
797 <pre> br bool <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
800 <p>The '<tt>br</tt>' instruction is used to cause control flow to
801 transfer to a different basic block in the current function. There are
802 two forms of this instruction, corresponding to a conditional branch
803 and an unconditional branch.</p>
805 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
806 single '<tt>bool</tt>' value and two '<tt>label</tt>' values. The
807 unconditional form of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>'
808 value as a target.</p>
810 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>bool</tt>'
811 argument is evaluated. If the value is <tt>true</tt>, control flows
812 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
813 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
815 <pre>Test:<br> %cond = <a href="#i_setcc">seteq</a> int %a, %b<br> br bool %cond, label %IfEqual, label %IfUnequal<br>IfEqual:<br> <a
816 href="#i_ret">ret</a> int 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> int 0<br></pre>
818 <!-- _______________________________________________________________________ -->
819 <div class="doc_subsubsection">
820 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
823 <div class="doc_text">
827 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
832 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
833 several different places. It is a generalization of the '<tt>br</tt>'
834 instruction, allowing a branch to occur to one of many possible
840 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
841 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
842 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
843 table is not allowed to contain duplicate constant entries.</p>
847 <p>The <tt>switch</tt> instruction specifies a table of values and
848 destinations. When the '<tt>switch</tt>' instruction is executed, this
849 table is searched for the given value. If the value is found, control flow is
850 transfered to the corresponding destination; otherwise, control flow is
851 transfered to the default destination.</p>
853 <h5>Implementation:</h5>
855 <p>Depending on properties of the target machine and the particular
856 <tt>switch</tt> instruction, this instruction may be code generated in different
857 ways. For example, it could be generated as a series of chained conditional
858 branches or with a lookup table.</p>
863 <i>; Emulate a conditional br instruction</i>
864 %Val = <a href="#i_cast">cast</a> bool %value to int
865 switch int %Val, label %truedest [int 0, label %falsedest ]
867 <i>; Emulate an unconditional br instruction</i>
868 switch uint 0, label %dest [ ]
870 <i>; Implement a jump table:</i>
871 switch uint %val, label %otherwise [ uint 0, label %onzero
873 uint 2, label %ontwo ]
876 <!-- _______________________________________________________________________ -->
877 <div class="doc_subsubsection"> <a name="i_invoke">'<tt>invoke</tt>'
878 Instruction</a> </div>
879 <div class="doc_text">
881 <pre> <result> = invoke <ptr to function ty> %<function ptr val>(<function args>)<br> to label <normal label> except label <exception label><br></pre>
883 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a
884 specified function, with the possibility of control flow transfer to
885 either the '<tt>normal</tt>' <tt>label</tt> label or the '<tt>exception</tt>'<tt>label</tt>.
886 If the callee function returns with the "<tt><a href="#i_ret">ret</a></tt>"
887 instruction, control flow will return to the "normal" label. If the
888 callee (or any indirect callees) returns with the "<a href="#i_unwind"><tt>unwind</tt></a>"
889 instruction, control is interrupted, and continued at the dynamically
890 nearest "except" label.</p>
892 <p>This instruction requires several arguments:</p>
894 <li>'<tt>ptr to function ty</tt>': shall be the signature of the
895 pointer to function value being invoked. In most cases, this is a
896 direct function invocation, but indirect <tt>invoke</tt>s are just as
897 possible, branching off an arbitrary pointer to function value. </li>
898 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer
899 to a function to be invoked. </li>
900 <li>'<tt>function args</tt>': argument list whose types match the
901 function signature argument types. If the function signature indicates
902 the function accepts a variable number of arguments, the extra
903 arguments can be specified. </li>
904 <li>'<tt>normal label</tt>': the label reached when the called
905 function executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
906 <li>'<tt>exception label</tt>': the label reached when a callee
907 returns with the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
910 <p>This instruction is designed to operate as a standard '<tt><a
911 href="#i_call">call</a></tt>' instruction in most regards. The
912 primary difference is that it establishes an association with a label,
913 which is used by the runtime library to unwind the stack.</p>
914 <p>This instruction is used in languages with destructors to ensure
915 that proper cleanup is performed in the case of either a <tt>longjmp</tt>
916 or a thrown exception. Additionally, this is important for
917 implementation of '<tt>catch</tt>' clauses in high-level languages that
920 <pre> %retval = invoke int %Test(int 15)<br> to label %Continue<br> except label %TestCleanup <i>; {int}:retval set</i>
923 <!-- _______________________________________________________________________ -->
924 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
925 Instruction</a> </div>
926 <div class="doc_text">
928 <pre> unwind<br></pre>
930 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing
931 control flow at the first callee in the dynamic call stack which used
932 an <a href="#i_invoke"><tt>invoke</tt></a> instruction to perform the
933 call. This is primarily used to implement exception handling.</p>
935 <p>The '<tt>unwind</tt>' intrinsic causes execution of the current
936 function to immediately halt. The dynamic call stack is then searched
937 for the first <a href="#i_invoke"><tt>invoke</tt></a> instruction on
938 the call stack. Once found, execution continues at the "exceptional"
939 destination block specified by the <tt>invoke</tt> instruction. If
940 there is no <tt>invoke</tt> instruction in the dynamic call chain,
941 undefined behavior results.</p>
943 <!-- ======================================================================= -->
944 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
945 <div class="doc_text">
946 <p>Binary operators are used to do most of the computation in a
947 program. They require two operands, execute an operation on them, and
948 produce a single value. Although, that single value might represent
949 multiple data, as is the case with the <a href="#t_packed">packed</a> data type.
950 The result value of a binary operator is not
951 necessarily the same type as its operands.</p>
952 <p>There are several different binary operators:</p>
954 <!-- _______________________________________________________________________ -->
955 <div class="doc_subsubsection"> <a name="i_add">'<tt>add</tt>'
956 Instruction</a> </div>
957 <div class="doc_text">
959 <pre> <result> = add <ty> <var1>, <var2> <i>; yields {ty}:result</i>
962 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
964 <p>The two arguments to the '<tt>add</tt>' instruction must be either <a
965 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a> values.
966 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
967 Both arguments must have identical types.</p>
969 <p>The value produced is the integer or floating point sum of the two
972 <pre> <result> = add int 4, %var <i>; yields {int}:result = 4 + %var</i>
975 <!-- _______________________________________________________________________ -->
976 <div class="doc_subsubsection"> <a name="i_sub">'<tt>sub</tt>'
977 Instruction</a> </div>
978 <div class="doc_text">
980 <pre> <result> = sub <ty> <var1>, <var2> <i>; yields {ty}:result</i>
983 <p>The '<tt>sub</tt>' instruction returns the difference of its two
985 <p>Note that the '<tt>sub</tt>' instruction is used to represent the '<tt>neg</tt>'
986 instruction present in most other intermediate representations.</p>
988 <p>The two arguments to the '<tt>sub</tt>' instruction must be either <a
989 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
991 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
992 Both arguments must have identical types.</p>
994 <p>The value produced is the integer or floating point difference of
995 the two operands.</p>
997 <pre> <result> = sub int 4, %var <i>; yields {int}:result = 4 - %var</i>
998 <result> = sub int 0, %val <i>; yields {int}:result = -%var</i>
1001 <!-- _______________________________________________________________________ -->
1002 <div class="doc_subsubsection"> <a name="i_mul">'<tt>mul</tt>'
1003 Instruction</a> </div>
1004 <div class="doc_text">
1006 <pre> <result> = mul <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1009 <p>The '<tt>mul</tt>' instruction returns the product of its two
1012 <p>The two arguments to the '<tt>mul</tt>' instruction must be either <a
1013 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1015 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1016 Both arguments must have identical types.</p>
1018 <p>The value produced is the integer or floating point product of the
1020 <p>There is no signed vs unsigned multiplication. The appropriate
1021 action is taken based on the type of the operand.</p>
1023 <pre> <result> = mul int 4, %var <i>; yields {int}:result = 4 * %var</i>
1026 <!-- _______________________________________________________________________ -->
1027 <div class="doc_subsubsection"> <a name="i_div">'<tt>div</tt>'
1028 Instruction</a> </div>
1029 <div class="doc_text">
1031 <pre> <result> = div <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1034 <p>The '<tt>div</tt>' instruction returns the quotient of its two
1037 <p>The two arguments to the '<tt>div</tt>' instruction must be either <a
1038 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1040 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1041 Both arguments must have identical types.</p>
1043 <p>The value produced is the integer or floating point quotient of the
1046 <pre> <result> = div int 4, %var <i>; yields {int}:result = 4 / %var</i>
1049 <!-- _______________________________________________________________________ -->
1050 <div class="doc_subsubsection"> <a name="i_rem">'<tt>rem</tt>'
1051 Instruction</a> </div>
1052 <div class="doc_text">
1054 <pre> <result> = rem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1057 <p>The '<tt>rem</tt>' instruction returns the remainder from the
1058 division of its two operands.</p>
1060 <p>The two arguments to the '<tt>rem</tt>' instruction must be either <a
1061 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1063 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1064 Both arguments must have identical types.</p>
1066 <p>This returns the <i>remainder</i> of a division (where the result
1067 has the same sign as the divisor), not the <i>modulus</i> (where the
1068 result has the same sign as the dividend) of a value. For more
1069 information about the difference, see: <a
1070 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
1073 <pre> <result> = rem int 4, %var <i>; yields {int}:result = 4 % %var</i>
1076 <!-- _______________________________________________________________________ -->
1077 <div class="doc_subsubsection"> <a name="i_setcc">'<tt>set<i>cc</i></tt>'
1078 Instructions</a> </div>
1079 <div class="doc_text">
1081 <pre> <result> = seteq <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1082 <result> = setne <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1083 <result> = setlt <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1084 <result> = setgt <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1085 <result> = setle <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1086 <result> = setge <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1089 <p>The '<tt>set<i>cc</i></tt>' family of instructions returns a boolean
1090 value based on a comparison of their two operands.</p>
1092 <p>The two arguments to the '<tt>set<i>cc</i></tt>' instructions must
1093 be of <a href="#t_firstclass">first class</a> type (it is not possible
1094 to compare '<tt>label</tt>'s, '<tt>array</tt>'s, '<tt>structure</tt>'
1095 or '<tt>void</tt>' values, etc...). Both arguments must have identical
1098 <p>The '<tt>seteq</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1099 value if both operands are equal.<br>
1100 The '<tt>setne</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1101 value if both operands are unequal.<br>
1102 The '<tt>setlt</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1103 value if the first operand is less than the second operand.<br>
1104 The '<tt>setgt</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1105 value if the first operand is greater than the second operand.<br>
1106 The '<tt>setle</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1107 value if the first operand is less than or equal to the second operand.<br>
1108 The '<tt>setge</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1109 value if the first operand is greater than or equal to the second
1112 <pre> <result> = seteq int 4, 5 <i>; yields {bool}:result = false</i>
1113 <result> = setne float 4, 5 <i>; yields {bool}:result = true</i>
1114 <result> = setlt uint 4, 5 <i>; yields {bool}:result = true</i>
1115 <result> = setgt sbyte 4, 5 <i>; yields {bool}:result = false</i>
1116 <result> = setle sbyte 4, 5 <i>; yields {bool}:result = true</i>
1117 <result> = setge sbyte 4, 5 <i>; yields {bool}:result = false</i>
1120 <!-- ======================================================================= -->
1121 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
1122 Operations</a> </div>
1123 <div class="doc_text">
1124 <p>Bitwise binary operators are used to do various forms of
1125 bit-twiddling in a program. They are generally very efficient
1126 instructions, and can commonly be strength reduced from other
1127 instructions. They require two operands, execute an operation on them,
1128 and produce a single value. The resulting value of the bitwise binary
1129 operators is always the same type as its first operand.</p>
1131 <!-- _______________________________________________________________________ -->
1132 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
1133 Instruction</a> </div>
1134 <div class="doc_text">
1136 <pre> <result> = and <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1139 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
1140 its two operands.</p>
1142 <p>The two arguments to the '<tt>and</tt>' instruction must be <a
1143 href="#t_integral">integral</a> values. Both arguments must have
1144 identical types.</p>
1146 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
1148 <div style="align: center">
1149 <table border="1" cellspacing="0" cellpadding="4">
1180 <pre> <result> = and int 4, %var <i>; yields {int}:result = 4 & %var</i>
1181 <result> = and int 15, 40 <i>; yields {int}:result = 8</i>
1182 <result> = and int 4, 8 <i>; yields {int}:result = 0</i>
1185 <!-- _______________________________________________________________________ -->
1186 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
1187 <div class="doc_text">
1189 <pre> <result> = or <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1192 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
1193 or of its two operands.</p>
1195 <p>The two arguments to the '<tt>or</tt>' instruction must be <a
1196 href="#t_integral">integral</a> values. Both arguments must have
1197 identical types.</p>
1199 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
1201 <div style="align: center">
1202 <table border="1" cellspacing="0" cellpadding="4">
1233 <pre> <result> = or int 4, %var <i>; yields {int}:result = 4 | %var</i>
1234 <result> = or int 15, 40 <i>; yields {int}:result = 47</i>
1235 <result> = or int 4, 8 <i>; yields {int}:result = 12</i>
1238 <!-- _______________________________________________________________________ -->
1239 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
1240 Instruction</a> </div>
1241 <div class="doc_text">
1243 <pre> <result> = xor <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1246 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
1247 or of its two operands. The <tt>xor</tt> is used to implement the
1248 "one's complement" operation, which is the "~" operator in C.</p>
1250 <p>The two arguments to the '<tt>xor</tt>' instruction must be <a
1251 href="#t_integral">integral</a> values. Both arguments must have
1252 identical types.</p>
1254 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
1256 <div style="align: center">
1257 <table border="1" cellspacing="0" cellpadding="4">
1289 <pre> <result> = xor int 4, %var <i>; yields {int}:result = 4 ^ %var</i>
1290 <result> = xor int 15, 40 <i>; yields {int}:result = 39</i>
1291 <result> = xor int 4, 8 <i>; yields {int}:result = 12</i>
1292 <result> = xor int %V, -1 <i>; yields {int}:result = ~%V</i>
1295 <!-- _______________________________________________________________________ -->
1296 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
1297 Instruction</a> </div>
1298 <div class="doc_text">
1300 <pre> <result> = shl <ty> <var1>, ubyte <var2> <i>; yields {ty}:result</i>
1303 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
1304 the left a specified number of bits.</p>
1306 <p>The first argument to the '<tt>shl</tt>' instruction must be an <a
1307 href="#t_integer">integer</a> type. The second argument must be an '<tt>ubyte</tt>'
1310 <p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup>.</p>
1312 <pre> <result> = shl int 4, ubyte %var <i>; yields {int}:result = 4 << %var</i>
1313 <result> = shl int 4, ubyte 2 <i>; yields {int}:result = 16</i>
1314 <result> = shl int 1, ubyte 10 <i>; yields {int}:result = 1024</i>
1317 <!-- _______________________________________________________________________ -->
1318 <div class="doc_subsubsection"> <a name="i_shr">'<tt>shr</tt>'
1319 Instruction</a> </div>
1320 <div class="doc_text">
1322 <pre> <result> = shr <ty> <var1>, ubyte <var2> <i>; yields {ty}:result</i>
1325 <p>The '<tt>shr</tt>' instruction returns the first operand shifted to
1326 the right a specified number of bits.</p>
1328 <p>The first argument to the '<tt>shr</tt>' instruction must be an <a
1329 href="#t_integer">integer</a> type. The second argument must be an '<tt>ubyte</tt>'
1332 <p>If the first argument is a <a href="#t_signed">signed</a> type, the
1333 most significant bit is duplicated in the newly free'd bit positions.
1334 If the first argument is unsigned, zero bits shall fill the empty
1337 <pre> <result> = shr int 4, ubyte %var <i>; yields {int}:result = 4 >> %var</i>
1338 <result> = shr uint 4, ubyte 1 <i>; yields {uint}:result = 2</i>
1339 <result> = shr int 4, ubyte 2 <i>; yields {int}:result = 1</i>
1340 <result> = shr sbyte 4, ubyte 3 <i>; yields {sbyte}:result = 0</i>
1341 <result> = shr sbyte -2, ubyte 1 <i>; yields {sbyte}:result = -1</i>
1344 <!-- ======================================================================= -->
1345 <div class="doc_subsection"> <a name="memoryops">Memory Access
1346 Operations</a></div>
1347 <div class="doc_text">
1348 <p>A key design point of an SSA-based representation is how it
1349 represents memory. In LLVM, no memory locations are in SSA form, which
1350 makes things very simple. This section describes how to read, write,
1351 allocate and free memory in LLVM.</p>
1353 <!-- _______________________________________________________________________ -->
1354 <div class="doc_subsubsection"> <a name="i_malloc">'<tt>malloc</tt>'
1355 Instruction</a> </div>
1356 <div class="doc_text">
1358 <pre> <result> = malloc <type>, uint <NumElements> <i>; yields {type*}:result</i>
1359 <result> = malloc <type> <i>; yields {type*}:result</i>
1362 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
1363 heap and returns a pointer to it.</p>
1365 <p>The '<tt>malloc</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
1366 bytes of memory from the operating system and returns a pointer of the
1367 appropriate type to the program. The second form of the instruction is
1368 a shorter version of the first instruction that defaults to allocating
1370 <p>'<tt>type</tt>' must be a sized type.</p>
1372 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
1373 a pointer is returned.</p>
1375 <pre> %array = malloc [4 x ubyte ] <i>; yields {[%4 x ubyte]*}:array</i>
1378 href="#i_add">add</a> uint 2, 2 <i>; yields {uint}:size = uint 4</i>
1379 %array1 = malloc ubyte, uint 4 <i>; yields {ubyte*}:array1</i>
1380 %array2 = malloc [12 x ubyte], uint %size <i>; yields {[12 x ubyte]*}:array2</i>
1383 <!-- _______________________________________________________________________ -->
1384 <div class="doc_subsubsection"> <a name="i_free">'<tt>free</tt>'
1385 Instruction</a> </div>
1386 <div class="doc_text">
1388 <pre> free <type> <value> <i>; yields {void}</i>
1391 <p>The '<tt>free</tt>' instruction returns memory back to the unused
1392 memory heap, to be reallocated in the future.</p>
1395 <p>'<tt>value</tt>' shall be a pointer value that points to a value
1396 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
1399 <p>Access to the memory pointed to by the pointer is not longer defined
1400 after this instruction executes.</p>
1402 <pre> %array = <a href="#i_malloc">malloc</a> [4 x ubyte] <i>; yields {[4 x ubyte]*}:array</i>
1403 free [4 x ubyte]* %array
1406 <!-- _______________________________________________________________________ -->
1407 <div class="doc_subsubsection"> <a name="i_alloca">'<tt>alloca</tt>'
1408 Instruction</a> </div>
1409 <div class="doc_text">
1411 <pre> <result> = alloca <type>, uint <NumElements> <i>; yields {type*}:result</i>
1412 <result> = alloca <type> <i>; yields {type*}:result</i>
1415 <p>The '<tt>alloca</tt>' instruction allocates memory on the current
1416 stack frame of the procedure that is live until the current function
1417 returns to its caller.</p>
1419 <p>The the '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
1420 bytes of memory on the runtime stack, returning a pointer of the
1421 appropriate type to the program. The second form of the instruction is
1422 a shorter version of the first that defaults to allocating one element.</p>
1423 <p>'<tt>type</tt>' may be any sized type.</p>
1425 <p>Memory is allocated, a pointer is returned. '<tt>alloca</tt>'d
1426 memory is automatically released when the function returns. The '<tt>alloca</tt>'
1427 instruction is commonly used to represent automatic variables that must
1428 have an address available. When the function returns (either with the <tt><a
1429 href="#i_ret">ret</a></tt> or <tt><a href="#i_invoke">invoke</a></tt>
1430 instructions), the memory is reclaimed.</p>
1432 <pre> %ptr = alloca int <i>; yields {int*}:ptr</i>
1433 %ptr = alloca int, uint 4 <i>; yields {int*}:ptr</i>
1436 <!-- _______________________________________________________________________ -->
1437 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
1438 Instruction</a> </div>
1439 <div class="doc_text">
1441 <pre> <result> = load <ty>* <pointer><br> <result> = volatile load <ty>* <pointer><br></pre>
1443 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
1445 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
1446 address to load from. The pointer must point to a <a
1447 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
1448 marked as <tt>volatile</tt> then the optimizer is not allowed to modify
1449 the number or order of execution of this <tt>load</tt> with other
1450 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
1453 <p>The location of memory pointed to is loaded.</p>
1455 <pre> %ptr = <a href="#i_alloca">alloca</a> int <i>; yields {int*}:ptr</i>
1457 href="#i_store">store</a> int 3, int* %ptr <i>; yields {void}</i>
1458 %val = load int* %ptr <i>; yields {int}:val = int 3</i>
1461 <!-- _______________________________________________________________________ -->
1462 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
1463 Instruction</a> </div>
1465 <pre> store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
1466 volatile store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
1469 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
1471 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
1472 to store and an address to store it into. The type of the '<tt><pointer></tt>'
1473 operand must be a pointer to the type of the '<tt><value></tt>'
1474 operand. If the <tt>store</tt> is marked as <tt>volatile</tt> then the
1475 optimizer is not allowed to modify the number or order of execution of
1476 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
1477 href="#i_store">store</a></tt> instructions.</p>
1479 <p>The contents of memory are updated to contain '<tt><value></tt>'
1480 at the location specified by the '<tt><pointer></tt>' operand.</p>
1482 <pre> %ptr = <a href="#i_alloca">alloca</a> int <i>; yields {int*}:ptr</i>
1484 href="#i_store">store</a> int 3, int* %ptr <i>; yields {void}</i>
1485 %val = load int* %ptr <i>; yields {int}:val = int 3</i>
1487 <!-- _______________________________________________________________________ -->
1488 <div class="doc_subsubsection">
1489 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
1492 <div class="doc_text">
1495 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
1501 The '<tt>getelementptr</tt>' instruction is used to get the address of a
1502 subelement of an aggregate data structure.</p>
1506 <p>This instruction takes a list of integer constants that indicate what
1507 elements of the aggregate object to index to. The actual types of the arguments
1508 provided depend on the type of the first pointer argument. The
1509 '<tt>getelementptr</tt>' instruction is used to index down through the type
1510 levels of a structure. When indexing into a structure, only <tt>uint</tt>
1511 integer constants are allowed. When indexing into an array or pointer
1512 <tt>int</tt> and <tt>long</tt> indexes are allowed of any sign.</p>
1514 <p>For example, let's consider a C code fragment and how it gets
1515 compiled to LLVM:</p>
1529 int *foo(struct ST *s) {
1530 return &s[1].Z.B[5][13];
1534 <p>The LLVM code generated by the GCC frontend is:</p>
1537 %RT = type { sbyte, [10 x [20 x int]], sbyte }
1538 %ST = type { int, double, %RT }
1542 int* %foo(%ST* %s) {
1544 %reg = getelementptr %ST* %s, int 1, uint 2, uint 1, int 5, int 13
1551 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
1552 on the pointer type that is being index into. <a href="#t_pointer">Pointer</a>
1553 and <a href="#t_array">array</a> types require <tt>uint</tt>, <tt>int</tt>,
1554 <tt>ulong</tt>, or <tt>long</tt> values, and <a href="#t_struct">structure</a>
1555 types require <tt>uint</tt> <b>constants</b>.</p>
1557 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
1558 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ int, double, %RT
1559 }</tt>' type, a structure. The second index indexes into the third element of
1560 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ sbyte, [10 x [20 x int]],
1561 sbyte }</tt>' type, another structure. The third index indexes into the second
1562 element of the structure, yielding a '<tt>[10 x [20 x int]]</tt>' type, an
1563 array. The two dimensions of the array are subscripted into, yielding an
1564 '<tt>int</tt>' type. The '<tt>getelementptr</tt>' instruction return a pointer
1565 to this element, thus computing a value of '<tt>int*</tt>' type.</p>
1567 <p>Note that it is perfectly legal to index partially through a
1568 structure, returning a pointer to an inner element. Because of this,
1569 the LLVM code for the given testcase is equivalent to:</p>
1572 int* "foo"(%ST* %s) {
1573 %t1 = getelementptr %ST* %s, int 1 <i>; yields %ST*:%t1</i>
1574 %t2 = getelementptr %ST* %t1, int 0, uint 2 <i>; yields %RT*:%t2</i>
1575 %t3 = getelementptr %RT* %t2, int 0, uint 1 <i>; yields [10 x [20 x int]]*:%t3</i>
1576 %t4 = getelementptr [10 x [20 x int]]* %t3, int 0, int 5 <i>; yields [20 x int]*:%t4</i>
1577 %t5 = getelementptr [20 x int]* %t4, int 0, int 13 <i>; yields int*:%t5</i>
1583 <i>; yields [12 x ubyte]*:aptr</i>
1584 %aptr = getelementptr {int, [12 x ubyte]}* %sptr, long 0, uint 1
1588 <!-- ======================================================================= -->
1589 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
1590 <div class="doc_text">
1591 <p>The instructions in this category are the "miscellaneous"
1592 instructions, which defy better classification.</p>
1594 <!-- _______________________________________________________________________ -->
1595 <div class="doc_subsubsection"> <a name="i_phi">'<tt>phi</tt>'
1596 Instruction</a> </div>
1597 <div class="doc_text">
1599 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
1601 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
1602 the SSA graph representing the function.</p>
1604 <p>The type of the incoming values are specified with the first type
1605 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
1606 as arguments, with one pair for each predecessor basic block of the
1607 current block. Only values of <a href="#t_firstclass">first class</a>
1608 type may be used as the value arguments to the PHI node. Only labels
1609 may be used as the label arguments.</p>
1610 <p>There must be no non-phi instructions between the start of a basic
1611 block and the PHI instructions: i.e. PHI instructions must be first in
1614 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the
1615 value specified by the parameter, depending on which basic block we
1616 came from in the last <a href="#terminators">terminator</a> instruction.</p>
1618 <pre>Loop: ; Infinite loop that counts from 0 on up...<br> %indvar = phi uint [ 0, %LoopHeader ], [ %nextindvar, %Loop ]<br> %nextindvar = add uint %indvar, 1<br> br label %Loop<br></pre>
1621 <!-- _______________________________________________________________________ -->
1622 <div class="doc_subsubsection">
1623 <a name="i_cast">'<tt>cast .. to</tt>' Instruction</a>
1626 <div class="doc_text">
1631 <result> = cast <ty> <value> to <ty2> <i>; yields ty2</i>
1637 The '<tt>cast</tt>' instruction is used as the primitive means to convert
1638 integers to floating point, change data type sizes, and break type safety (by
1646 The '<tt>cast</tt>' instruction takes a value to cast, which must be a first
1647 class value, and a type to cast it to, which must also be a <a
1648 href="#t_firstclass">first class</a> type.
1654 This instruction follows the C rules for explicit casts when determining how the
1655 data being cast must change to fit in its new container.
1659 When casting to bool, any value that would be considered true in the context of
1660 a C '<tt>if</tt>' condition is converted to the boolean '<tt>true</tt>' values,
1661 all else are '<tt>false</tt>'.
1665 When extending an integral value from a type of one signness to another (for
1666 example '<tt>sbyte</tt>' to '<tt>ulong</tt>'), the value is sign-extended if the
1667 <b>source</b> value is signed, and zero-extended if the source value is
1668 unsigned. <tt>bool</tt> values are always zero extended into either zero or
1675 %X = cast int 257 to ubyte <i>; yields ubyte:1</i>
1676 %Y = cast int 123 to bool <i>; yields bool:true</i>
1680 <!-- _______________________________________________________________________ -->
1681 <div class="doc_subsubsection">
1682 <a name="i_select">'<tt>select</tt>' Instruction</a>
1685 <div class="doc_text">
1690 <result> = select bool <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
1696 The '<tt>select</tt>' instruction is used to choose one value based on a
1697 condition, without branching.
1704 The '<tt>select</tt>' instruction requires a boolean value indicating the condition, and two values of the same <a href="#t_firstclass">first class</a> type.
1710 If the boolean condition evaluates to true, the instruction returns the first
1711 value argument, otherwise it returns the second value argument.
1717 %X = select bool true, ubyte 17, ubyte 42 <i>; yields ubyte:17</i>
1725 <!-- _______________________________________________________________________ -->
1726 <div class="doc_subsubsection"> <a name="i_call">'<tt>call</tt>'
1727 Instruction</a> </div>
1728 <div class="doc_text">
1730 <pre> <result> = call <ty>* <fnptrval>(<param list>)<br></pre>
1732 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
1734 <p>This instruction requires several arguments:</p>
1737 <p>'<tt>ty</tt>': shall be the signature of the pointer to function
1738 value being invoked. The argument types must match the types implied
1739 by this signature.</p>
1742 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a
1743 function to be invoked. In most cases, this is a direct function
1744 invocation, but indirect <tt>call</tt>s are just as possible,
1745 calling an arbitrary pointer to function values.</p>
1748 <p>'<tt>function args</tt>': argument list whose types match the
1749 function signature argument types. If the function signature
1750 indicates the function accepts a variable number of arguments, the
1751 extra arguments can be specified.</p>
1755 <p>The '<tt>call</tt>' instruction is used to cause control flow to
1756 transfer to a specified function, with its incoming arguments bound to
1757 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
1758 instruction in the called function, control flow continues with the
1759 instruction after the function call, and the return value of the
1760 function is bound to the result argument. This is a simpler case of
1761 the <a href="#i_invoke">invoke</a> instruction.</p>
1763 <pre> %retval = call int %test(int %argc)<br> call int(sbyte*, ...) *%printf(sbyte* %msg, int 12, sbyte 42);<br></pre>
1765 <!-- _______________________________________________________________________ -->
1766 <div class="doc_subsubsection"> <a name="i_vanext">'<tt>vanext</tt>'
1767 Instruction</a> </div>
1768 <div class="doc_text">
1770 <pre> <resultarglist> = vanext <va_list> <arglist>, <argty><br></pre>
1772 <p>The '<tt>vanext</tt>' instruction is used to access arguments passed
1773 through the "variable argument" area of a function call. It is used to
1774 implement the <tt>va_arg</tt> macro in C.</p>
1776 <p>This instruction takes a <tt>valist</tt> value and the type of the
1777 argument. It returns another <tt>valist</tt>.</p>
1779 <p>The '<tt>vanext</tt>' instruction advances the specified <tt>valist</tt>
1780 past an argument of the specified type. In conjunction with the <a
1781 href="#i_vaarg"><tt>vaarg</tt></a> instruction, it is used to implement
1782 the <tt>va_arg</tt> macro available in C. For more information, see
1783 the variable argument handling <a href="#int_varargs">Intrinsic
1785 <p>It is legal for this instruction to be called in a function which
1786 does not take a variable number of arguments, for example, the <tt>vfprintf</tt>
1788 <p><tt>vanext</tt> is an LLVM instruction instead of an <a
1789 href="#intrinsics">intrinsic function</a> because it takes an type as
1792 <p>See the <a href="#int_varargs">variable argument processing</a>
1795 <!-- _______________________________________________________________________ -->
1796 <div class="doc_subsubsection"> <a name="i_vaarg">'<tt>vaarg</tt>'
1797 Instruction</a> </div>
1798 <div class="doc_text">
1800 <pre> <resultval> = vaarg <va_list> <arglist>, <argty><br></pre>
1802 <p>The '<tt>vaarg</tt>' instruction is used to access arguments passed
1803 through the "variable argument" area of a function call. It is used to
1804 implement the <tt>va_arg</tt> macro in C.</p>
1806 <p>This instruction takes a <tt>valist</tt> value and the type of the
1807 argument. It returns a value of the specified argument type.</p>
1809 <p>The '<tt>vaarg</tt>' instruction loads an argument of the specified
1810 type from the specified <tt>va_list</tt>. In conjunction with the <a
1811 href="#i_vanext"><tt>vanext</tt></a> instruction, it is used to
1812 implement the <tt>va_arg</tt> macro available in C. For more
1813 information, see the variable argument handling <a href="#int_varargs">Intrinsic
1815 <p>It is legal for this instruction to be called in a function which
1816 does not take a variable number of arguments, for example, the <tt>vfprintf</tt>
1818 <p><tt>vaarg</tt> is an LLVM instruction instead of an <a
1819 href="#intrinsics">intrinsic function</a> because it takes an type as
1822 <p>See the <a href="#int_varargs">variable argument processing</a>
1826 <!-- *********************************************************************** -->
1827 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
1828 <!-- *********************************************************************** -->
1830 <div class="doc_text">
1832 <p>LLVM supports the notion of an "intrinsic function". These functions have
1833 well known names and semantics, and are required to follow certain
1834 restrictions. Overall, these instructions represent an extension mechanism for
1835 the LLVM language that does not require changing all of the transformations in
1836 LLVM to add to the language (or the bytecode reader/writer, the parser,
1839 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix, this
1840 prefix is reserved in LLVM for intrinsic names, thus functions may not be named
1841 this. Intrinsic functions must always be external functions: you cannot define
1842 the body of intrinsic functions. Intrinsic functions may only be used in call
1843 or invoke instructions: it is illegal to take the address of an intrinsic
1844 function. Additionally, because intrinsic functions are part of the LLVM
1845 language, it is required that they all be documented here if any are added.</p>
1849 Adding an intrinsic to LLVM is straight-forward if it is possible to express the
1850 concept in LLVM directly (ie, code generator support is not _required_). To do
1851 this, extend the default implementation of the IntrinsicLowering class to handle
1852 the intrinsic. Code generators use this class to lower intrinsics they do not
1853 understand to raw LLVM instructions that they do.
1858 <!-- ======================================================================= -->
1859 <div class="doc_subsection">
1860 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
1863 <div class="doc_text">
1865 <p>Variable argument support is defined in LLVM with the <a
1866 href="#i_vanext"><tt>vanext</tt></a> instruction and these three
1867 intrinsic functions. These functions are related to the similarly
1868 named macros defined in the <tt><stdarg.h></tt> header file.</p>
1870 <p>All of these functions operate on arguments that use a
1871 target-specific value type "<tt>va_list</tt>". The LLVM assembly
1872 language reference manual does not define what this type is, so all
1873 transformations should be prepared to handle intrinsics with any type
1876 <p>This example shows how the <a href="#i_vanext"><tt>vanext</tt></a>
1877 instruction and the variable argument handling intrinsic functions are
1881 int %test(int %X, ...) {
1882 ; Initialize variable argument processing
1883 %ap = call sbyte* %<a href="#i_va_start">llvm.va_start</a>()
1885 ; Read a single integer argument
1886 %tmp = vaarg sbyte* %ap, int
1888 ; Advance to the next argument
1889 %ap2 = vanext sbyte* %ap, int
1891 ; Demonstrate usage of llvm.va_copy and llvm.va_end
1892 %aq = call sbyte* %<a href="#i_va_copy">llvm.va_copy</a>(sbyte* %ap2)
1893 call void %<a href="#i_va_end">llvm.va_end</a>(sbyte* %aq)
1895 ; Stop processing of arguments.
1896 call void %<a href="#i_va_end">llvm.va_end</a>(sbyte* %ap2)
1902 <!-- _______________________________________________________________________ -->
1903 <div class="doc_subsubsection">
1904 <a name="i_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
1908 <div class="doc_text">
1910 <pre> call va_list ()* %llvm.va_start()<br></pre>
1912 <p>The '<tt>llvm.va_start</tt>' intrinsic returns a new <tt><arglist></tt>
1913 for subsequent use by the variable argument intrinsics.</p>
1915 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
1916 macro available in C. In a target-dependent way, it initializes and
1917 returns a <tt>va_list</tt> element, so that the next <tt>vaarg</tt>
1918 will produce the first variable argument passed to the function. Unlike
1919 the C <tt>va_start</tt> macro, this intrinsic does not need to know the
1920 last argument of the function, the compiler can figure that out.</p>
1921 <p>Note that this intrinsic function is only legal to be called from
1922 within the body of a variable argument function.</p>
1925 <!-- _______________________________________________________________________ -->
1926 <div class="doc_subsubsection">
1927 <a name="i_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
1930 <div class="doc_text">
1932 <pre> call void (va_list)* %llvm.va_end(va_list <arglist>)<br></pre>
1934 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt><arglist></tt>
1935 which has been initialized previously with <tt><a href="#i_va_start">llvm.va_start</a></tt>
1936 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
1938 <p>The argument is a <tt>va_list</tt> to destroy.</p>
1940 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
1941 macro available in C. In a target-dependent way, it destroys the <tt>va_list</tt>.
1942 Calls to <a href="#i_va_start"><tt>llvm.va_start</tt></a> and <a
1943 href="#i_va_copy"><tt>llvm.va_copy</tt></a> must be matched exactly
1944 with calls to <tt>llvm.va_end</tt>.</p>
1947 <!-- _______________________________________________________________________ -->
1948 <div class="doc_subsubsection">
1949 <a name="i_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
1952 <div class="doc_text">
1957 call va_list (va_list)* %llvm.va_copy(va_list <destarglist>)
1962 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
1963 from the source argument list to the destination argument list.</p>
1967 <p>The argument is the <tt>va_list</tt> to copy.</p>
1971 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
1972 macro available in C. In a target-dependent way, it copies the source
1973 <tt>va_list</tt> element into the returned list. This intrinsic is necessary
1974 because the <tt><a href="#i_va_start">llvm.va_start</a></tt> intrinsic may be
1975 arbitrarily complex and require memory allocation, for example.</p>
1979 <!-- ======================================================================= -->
1980 <div class="doc_subsection">
1981 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
1984 <div class="doc_text">
1987 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
1988 Collection</a> requires the implementation and generation of these intrinsics.
1989 These intrinsics allow identification of <a href="#i_gcroot">GC roots on the
1990 stack</a>, as well as garbage collector implementations that require <a
1991 href="#i_gcread">read</a> and <a href="#i_gcwrite">write</a> barriers.
1992 Front-ends for type-safe garbage collected languages should generate these
1993 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
1994 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
1998 <!-- _______________________________________________________________________ -->
1999 <div class="doc_subsubsection">
2000 <a name="i_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
2003 <div class="doc_text">
2008 call void (<ty>**, <ty2>*)* %llvm.gcroot(<ty>** %ptrloc, <ty2>* %metadata)
2013 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existance of a GC root to
2014 the code generator, and allows some metadata to be associated with it.</p>
2018 <p>The first argument specifies the address of a stack object that contains the
2019 root pointer. The second pointer (which must be either a constant or a global
2020 value address) contains the meta-data to be associated with the root.</p>
2024 <p>At runtime, a call to this intrinsics stores a null pointer into the "ptrloc"
2025 location. At compile-time, the code generator generates information to allow
2026 the runtime to find the pointer at GC safe points.
2032 <!-- _______________________________________________________________________ -->
2033 <div class="doc_subsubsection">
2034 <a name="i_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
2037 <div class="doc_text">
2042 call sbyte* (sbyte**)* %llvm.gcread(sbyte** %Ptr)
2047 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
2048 locations, allowing garbage collector implementations that require read
2053 <p>The argument is the address to read from, which should be an address
2054 allocated from the garbage collector.</p>
2058 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
2059 instruction, but may be replaced with substantially more complex code by the
2060 garbage collector runtime, as needed.</p>
2065 <!-- _______________________________________________________________________ -->
2066 <div class="doc_subsubsection">
2067 <a name="i_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
2070 <div class="doc_text">
2075 call void (sbyte*, sbyte**)* %llvm.gcwrite(sbyte* %P1, sbyte** %P2)
2080 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
2081 locations, allowing garbage collector implementations that require write
2082 barriers (such as generational or reference counting collectors).</p>
2086 <p>The first argument is the reference to store, and the second is the heap
2087 location to store to.</p>
2091 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
2092 instruction, but may be replaced with substantially more complex code by the
2093 garbage collector runtime, as needed.</p>
2099 <!-- ======================================================================= -->
2100 <div class="doc_subsection">
2101 <a name="int_codegen">Code Generator Intrinsics</a>
2104 <div class="doc_text">
2106 These intrinsics are provided by LLVM to expose special features that may only
2107 be implemented with code generator support.
2112 <!-- _______________________________________________________________________ -->
2113 <div class="doc_subsubsection">
2114 <a name="i_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
2117 <div class="doc_text">
2121 call void* ()* %llvm.returnaddress(uint <level>)
2127 The '<tt>llvm.returnaddress</tt>' intrinsic returns a target-specific value
2128 indicating the return address of the current function or one of its callers.
2134 The argument to this intrinsic indicates which function to return the address
2135 for. Zero indicates the calling function, one indicates its caller, etc. The
2136 argument is <b>required</b> to be a constant integer value.
2142 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
2143 the return address of the specified call frame, or zero if it cannot be
2144 identified. The value returned by this intrinsic is likely to be incorrect or 0
2145 for arguments other than zero, so it should only be used for debugging purposes.
2149 Note that calling this intrinsic does not prevent function inlining or other
2150 aggressive transformations, so the value returned may not that of the obvious
2151 source-language caller.
2156 <!-- _______________________________________________________________________ -->
2157 <div class="doc_subsubsection">
2158 <a name="i_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
2161 <div class="doc_text">
2165 call void* ()* %llvm.frameaddress(uint <level>)
2171 The '<tt>llvm.frameaddress</tt>' intrinsic returns the target-specific frame
2172 pointer value for the specified stack frame.
2178 The argument to this intrinsic indicates which function to return the frame
2179 pointer for. Zero indicates the calling function, one indicates its caller,
2180 etc. The argument is <b>required</b> to be a constant integer value.
2186 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
2187 the frame address of the specified call frame, or zero if it cannot be
2188 identified. The value returned by this intrinsic is likely to be incorrect or 0
2189 for arguments other than zero, so it should only be used for debugging purposes.
2193 Note that calling this intrinsic does not prevent function inlining or other
2194 aggressive transformations, so the value returned may not that of the obvious
2195 source-language caller.
2199 <!-- ======================================================================= -->
2200 <div class="doc_subsection">
2201 <a name="int_os">Operating System Intrinsics</a>
2204 <div class="doc_text">
2206 These intrinsics are provided by LLVM to support the implementation of
2207 operating system level code.
2212 <!-- _______________________________________________________________________ -->
2213 <div class="doc_subsubsection">
2214 <a name="i_readport">'<tt>llvm.readport</tt>' Intrinsic</a>
2217 <div class="doc_text">
2221 call <integer type> (<integer type>)* %llvm.readport (<integer type> <address>)
2227 The '<tt>llvm.readport</tt>' intrinsic reads data from the specified hardware
2234 The argument to this intrinsic indicates the hardware I/O address from which
2235 to read the data. The address is in the hardware I/O address namespace (as
2236 opposed to being a memory location for memory mapped I/O).
2242 The '<tt>llvm.readport</tt>' intrinsic reads data from the hardware I/O port
2243 specified by <i>address</i> and returns the value. The address and return
2244 value must be integers, but the size is dependent upon the platform upon which
2245 the program is code generated. For example, on x86, the address must be an
2246 unsigned 16 bit value, and the return value must be 8, 16, or 32 bits.
2251 <!-- _______________________________________________________________________ -->
2252 <div class="doc_subsubsection">
2253 <a name="i_writeport">'<tt>llvm.writeport</tt>' Intrinsic</a>
2256 <div class="doc_text">
2260 call void (<integer type>, <integer type>)* %llvm.writeport (<integer type> <value>, <integer type> <address>)
2266 The '<tt>llvm.writeport</tt>' intrinsic writes data to the specified hardware
2273 The first argument is the value to write to the I/O port.
2277 The second argument indicates the hardware I/O address to which data should be
2278 written. The address is in the hardware I/O address namespace (as opposed to
2279 being a memory location for memory mapped I/O).
2285 The '<tt>llvm.writeport</tt>' intrinsic writes <i>value</i> to the I/O port
2286 specified by <i>address</i>. The address and value must be integers, but the
2287 size is dependent upon the platform upon which the program is code generated.
2288 For example, on x86, the address must be an unsigned 16 bit value, and the
2289 value written must be 8, 16, or 32 bits in length.
2294 <!-- _______________________________________________________________________ -->
2295 <div class="doc_subsubsection">
2296 <a name="i_readio">'<tt>llvm.readio</tt>' Intrinsic</a>
2299 <div class="doc_text">
2303 call <result> (<ty>*)* %llvm.readio (<ty> * <pointer>)
2309 The '<tt>llvm.readio</tt>' intrinsic reads data from a memory mapped I/O
2316 The argument to this intrinsic is a pointer indicating the memory address from
2317 which to read the data. The data must be a
2318 <a href="#t_firstclass">first class</a> type.
2324 The '<tt>llvm.readio</tt>' intrinsic reads data from a memory mapped I/O
2325 location specified by <i>pointer</i> and returns the value. The argument must
2326 be a pointer, and the return value must be a
2327 <a href="#t_firstclass">first class</a> type. However, certain architectures
2328 may not support I/O on all first class types. For example, 32 bit processors
2329 may only support I/O on data types that are 32 bits or less.
2333 This intrinsic enforces an in-order memory model for llvm.readio and
2334 llvm.writeio calls on machines that use dynamic scheduling. Dynamically
2335 scheduled processors may execute loads and stores out of order, re-ordering at
2336 run time accesses to memory mapped I/O registers. Using these intrinsics
2337 ensures that accesses to memory mapped I/O registers occur in program order.
2342 <!-- _______________________________________________________________________ -->
2343 <div class="doc_subsubsection">
2344 <a name="i_writeio">'<tt>llvm.writeio</tt>' Intrinsic</a>
2347 <div class="doc_text">
2351 call void (<ty1>, <ty2>*)* %llvm.writeio (<ty1> <value>, <ty2> * <pointer>)
2357 The '<tt>llvm.writeio</tt>' intrinsic writes data to the specified memory
2364 The first argument is the value to write to the memory mapped I/O location.
2365 The second argument is a pointer indicating the memory address to which the
2366 data should be written.
2372 The '<tt>llvm.writeio</tt>' intrinsic writes <i>value</i> to the memory mapped
2373 I/O address specified by <i>pointer</i>. The value must be a
2374 <a href="#t_firstclass">first class</a> type. However, certain architectures
2375 may not support I/O on all first class types. For example, 32 bit processors
2376 may only support I/O on data types that are 32 bits or less.
2380 This intrinsic enforces an in-order memory model for llvm.readio and
2381 llvm.writeio calls on machines that use dynamic scheduling. Dynamically
2382 scheduled processors may execute loads and stores out of order, re-ordering at
2383 run time accesses to memory mapped I/O registers. Using these intrinsics
2384 ensures that accesses to memory mapped I/O registers occur in program order.
2389 <!-- ======================================================================= -->
2390 <div class="doc_subsection">
2391 <a name="int_libc">Standard C Library Intrinsics</a>
2394 <div class="doc_text">
2396 LLVM provides intrinsics for a few important standard C library functions.
2397 These intrinsics allow source-language front-ends to pass information about the
2398 alignment of the pointer arguments to the code generator, providing opportunity
2399 for more efficient code generation.
2404 <!-- _______________________________________________________________________ -->
2405 <div class="doc_subsubsection">
2406 <a name="i_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
2409 <div class="doc_text">
2413 call void (sbyte*, sbyte*, uint, uint)* %llvm.memcpy(sbyte* <dest>, sbyte* <src>,
2414 uint <len>, uint <align>)
2420 The '<tt>llvm.memcpy</tt>' intrinsic copies a block of memory from the source
2421 location to the destination location.
2425 Note that, unlike the standard libc function, the <tt>llvm.memcpy</tt> intrinsic
2426 does not return a value, and takes an extra alignment argument.
2432 The first argument is a pointer to the destination, the second is a pointer to
2433 the source. The third argument is an (arbitrarily sized) integer argument
2434 specifying the number of bytes to copy, and the fourth argument is the alignment
2435 of the source and destination locations.
2439 If the call to this intrinisic has an alignment value that is not 0 or 1, then
2440 the caller guarantees that the size of the copy is a multiple of the alignment
2441 and that both the source and destination pointers are aligned to that boundary.
2447 The '<tt>llvm.memcpy</tt>' intrinsic copies a block of memory from the source
2448 location to the destination location, which are not allowed to overlap. It
2449 copies "len" bytes of memory over. If the argument is known to be aligned to
2450 some boundary, this can be specified as the fourth argument, otherwise it should
2456 <!-- _______________________________________________________________________ -->
2457 <div class="doc_subsubsection">
2458 <a name="i_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
2461 <div class="doc_text">
2465 call void (sbyte*, sbyte*, uint, uint)* %llvm.memmove(sbyte* <dest>, sbyte* <src>,
2466 uint <len>, uint <align>)
2472 The '<tt>llvm.memmove</tt>' intrinsic moves a block of memory from the source
2473 location to the destination location. It is similar to the '<tt>llvm.memcpy</tt>'
2474 intrinsic but allows the two memory locations to overlap.
2478 Note that, unlike the standard libc function, the <tt>llvm.memmove</tt> intrinsic
2479 does not return a value, and takes an extra alignment argument.
2485 The first argument is a pointer to the destination, the second is a pointer to
2486 the source. The third argument is an (arbitrarily sized) integer argument
2487 specifying the number of bytes to copy, and the fourth argument is the alignment
2488 of the source and destination locations.
2492 If the call to this intrinisic has an alignment value that is not 0 or 1, then
2493 the caller guarantees that the size of the copy is a multiple of the alignment
2494 and that both the source and destination pointers are aligned to that boundary.
2500 The '<tt>llvm.memmove</tt>' intrinsic copies a block of memory from the source
2501 location to the destination location, which may overlap. It
2502 copies "len" bytes of memory over. If the argument is known to be aligned to
2503 some boundary, this can be specified as the fourth argument, otherwise it should
2509 <!-- _______________________________________________________________________ -->
2510 <div class="doc_subsubsection">
2511 <a name="i_memset">'<tt>llvm.memset</tt>' Intrinsic</a>
2514 <div class="doc_text">
2518 call void (sbyte*, ubyte, uint, uint)* %llvm.memset(sbyte* <dest>, ubyte <val>,
2519 uint <len>, uint <align>)
2525 The '<tt>llvm.memset</tt>' intrinsic fills a block of memory with a particular
2530 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
2531 does not return a value, and takes an extra alignment argument.
2537 The first argument is a pointer to the destination to fill, the second is the
2538 byte value to fill it with, the third argument is an (arbitrarily sized) integer
2539 argument specifying the number of bytes to fill, and the fourth argument is the
2540 known alignment of destination location.
2544 If the call to this intrinisic has an alignment value that is not 0 or 1, then
2545 the caller guarantees that the size of the copy is a multiple of the alignment
2546 and that the destination pointer is aligned to that boundary.
2552 The '<tt>llvm.memset</tt>' intrinsic fills "len" bytes of memory starting at the
2553 destination location. If the argument is known to be aligned to some boundary,
2554 this can be specified as the fourth argument, otherwise it should be set to 0 or
2560 <!-- _______________________________________________________________________ -->
2561 <div class="doc_subsubsection">
2562 <a name="i_isunordered">'<tt>llvm.isunordered</tt>' Intrinsic</a>
2565 <div class="doc_text">
2569 call bool (<float or double>, <float or double>)* %llvm.isunordered(<float or double> Val1,
2570 <float or double> Val2)
2576 The '<tt>llvm.isunordered</tt>' intrinsic returns true if either or both of the
2577 specified floating point values is a NAN.
2583 The arguments are floating point numbers of the same type.
2589 If either or both of the arguments is a SNAN or QNAN, it returns true, otherwise
2597 <!-- ======================================================================= -->
2598 <div class="doc_subsection">
2599 <a name="int_debugger">Debugger Intrinsics</a>
2602 <div class="doc_text">
2604 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
2605 are described in the <a
2606 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
2607 Debugging</a> document.
2612 <!-- *********************************************************************** -->
2615 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
2616 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
2617 <a href="http://validator.w3.org/check/referer"><img
2618 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
2620 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
2621 <a href="http://llvm.cs.uiuc.edu">The LLVM Compiler Infrastructure</a><br>
2622 Last modified: $Date$