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 <link rel="stylesheet" href="llvm.css" type="text/css">
11 <div class="doc_title"> LLVM Language Reference Manual </div>
13 <li><a href="#abstract">Abstract</a></li>
14 <li><a href="#introduction">Introduction</a></li>
15 <li><a href="#identifiers">Identifiers</a></li>
16 <li><a href="#typesystem">Type System</a>
18 <li><a href="#t_primitive">Primitive Types</a>
20 <li><a href="#t_classifications">Type Classifications</a></li>
23 <li><a href="#t_derived">Derived Types</a>
25 <li><a href="#t_array">Array Type</a></li>
26 <li><a href="#t_function">Function Type</a></li>
27 <li><a href="#t_pointer">Pointer Type</a></li>
28 <li><a href="#t_struct">Structure Type</a></li>
29 <!-- <li><a href="#t_packed" >Packed Type</a> -->
34 <li><a href="#highlevel">High Level Structure</a>
36 <li><a href="#modulestructure">Module Structure</a></li>
37 <li><a href="#globalvars">Global Variables</a></li>
38 <li><a href="#functionstructure">Function Structure</a></li>
41 <li><a href="#instref">Instruction Reference</a>
43 <li><a href="#terminators">Terminator Instructions</a>
45 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
46 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
47 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
48 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
49 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
52 <li><a href="#binaryops">Binary Operations</a>
54 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
55 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
56 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
57 <li><a href="#i_div">'<tt>div</tt>' Instruction</a></li>
58 <li><a href="#i_rem">'<tt>rem</tt>' Instruction</a></li>
59 <li><a href="#i_setcc">'<tt>set<i>cc</i></tt>' Instructions</a></li>
62 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
64 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
65 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
66 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
67 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
68 <li><a href="#i_shr">'<tt>shr</tt>' Instruction</a></li>
71 <li><a href="#memoryops">Memory Access Operations</a>
73 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
74 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
75 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
76 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
77 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
78 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
81 <li><a href="#otherops">Other Operations</a>
83 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
84 <li><a href="#i_cast">'<tt>cast .. to</tt>' Instruction</a></li>
85 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
86 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
87 <li><a href="#i_vanext">'<tt>vanext</tt>' Instruction</a></li>
88 <li><a href="#i_vaarg">'<tt>vaarg</tt>' Instruction</a></li>
93 <li><a href="#intrinsics">Intrinsic Functions</a>
95 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
97 <li><a href="#i_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
98 <li><a href="#i_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
99 <li><a href="#i_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
102 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
104 <li><a href="#i_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
105 <li><a href="#i_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
106 <li><a href="#i_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
109 <li><a href="#int_codegen">Code Generator Intrinsics</a>
111 <li><a href="#i_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
112 <li><a href="#i_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
115 <li><a href="#int_os">Operating System Intrinsics</a>
117 <li><a href="#i_readport">'<tt>llvm.readport</tt>' Intrinsic</a></li>
118 <li><a href="#i_writeport">'<tt>llvm.writeport</tt>' Intrinsic</a></li>
119 <li><a href="#i_readio">'<tt>llvm.readio</tt>' Intrinsic</a></li>
120 <li><a href="#i_writeio">'<tt>llvm.writeio</tt>' Intrinsic</a></li>
122 <li><a href="#int_libc">Standard C Library Intrinsics</a>
124 <li><a href="#i_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a></li>
125 <li><a href="#i_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a></li>
126 <li><a href="#i_memset">'<tt>llvm.memset</tt>' Intrinsic</a></li>
129 <li><a href="#int_debugger">Debugger intrinsics</a></li>
134 <div class="doc_author">
135 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
136 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
139 <!-- *********************************************************************** -->
140 <div class="doc_section"> <a name="abstract">Abstract </a></div>
141 <!-- *********************************************************************** -->
143 <div class="doc_text">
144 <p>This document is a reference manual for the LLVM assembly language.
145 LLVM is an SSA based representation that provides type safety,
146 low-level operations, flexibility, and the capability of representing
147 'all' high-level languages cleanly. It is the common code
148 representation used throughout all phases of the LLVM compilation
152 <!-- *********************************************************************** -->
153 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
154 <!-- *********************************************************************** -->
156 <div class="doc_text">
158 <p>The LLVM code representation is designed to be used in three
159 different forms: as an in-memory compiler IR, as an on-disk bytecode
160 representation (suitable for fast loading by a Just-In-Time compiler),
161 and as a human readable assembly language representation. This allows
162 LLVM to provide a powerful intermediate representation for efficient
163 compiler transformations and analysis, while providing a natural means
164 to debug and visualize the transformations. The three different forms
165 of LLVM are all equivalent. This document describes the human readable
166 representation and notation.</p>
168 <p>The LLVM representation aims to be a light-weight and low-level
169 while being expressive, typed, and extensible at the same time. It
170 aims to be a "universal IR" of sorts, by being at a low enough level
171 that high-level ideas may be cleanly mapped to it (similar to how
172 microprocessors are "universal IR's", allowing many source languages to
173 be mapped to them). By providing type information, LLVM can be used as
174 the target of optimizations: for example, through pointer analysis, it
175 can be proven that a C automatic variable is never accessed outside of
176 the current function... allowing it to be promoted to a simple SSA
177 value instead of a memory location.</p>
181 <!-- _______________________________________________________________________ -->
182 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
184 <div class="doc_text">
186 <p>It is important to note that this document describes 'well formed'
187 LLVM assembly language. There is a difference between what the parser
188 accepts and what is considered 'well formed'. For example, the
189 following instruction is syntactically okay, but not well formed:</p>
192 %x = <a href="#i_add">add</a> int 1, %x
195 <p>...because the definition of <tt>%x</tt> does not dominate all of
196 its uses. The LLVM infrastructure provides a verification pass that may
197 be used to verify that an LLVM module is well formed. This pass is
198 automatically run by the parser after parsing input assembly, and by
199 the optimizer before it outputs bytecode. The violations pointed out
200 by the verifier pass indicate bugs in transformation passes or input to
203 <!-- Describe the typesetting conventions here. --> </div>
205 <!-- *********************************************************************** -->
206 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
207 <!-- *********************************************************************** -->
209 <div class="doc_text">
211 <p>LLVM uses three different forms of identifiers, for different
215 <li>Numeric constants are represented as you would expect: 12, -3
216 123.421, etc. Floating point constants have an optional hexadecimal
218 <li>Named values are represented as a string of characters with a '%'
219 prefix. For example, %foo, %DivisionByZero,
220 %a.really.long.identifier. The actual regular expression used is '<tt>%[a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
221 Identifiers which require other characters in their names can be
222 surrounded with quotes. In this way, anything except a <tt>"</tt>
223 character can be used in a name.</li>
224 <li>Unnamed values are represented as an unsigned numeric value with
225 a '%' prefix. For example, %12, %2, %44.</li>
227 <p>LLVM requires that values start with a '%' sign for two reasons:
228 Compilers don't need to worry about name clashes with reserved words,
229 and the set of reserved words may be expanded in the future without
230 penalty. Additionally, unnamed identifiers allow a compiler to quickly
231 come up with a temporary variable without having to avoid symbol table
233 <p>Reserved words in LLVM are very similar to reserved words in other
234 languages. There are keywords for different opcodes ('<tt><a
235 href="#i_add">add</a></tt>', '<tt><a href="#i_cast">cast</a></tt>', '<tt><a
236 href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
237 href="#t_void">void</a></tt>', '<tt><a href="#t_uint">uint</a></tt>',
238 etc...), and others. These reserved words cannot conflict with
239 variable names, because none of them start with a '%' character.</p>
240 <p>Here is an example of LLVM code to multiply the integer variable '<tt>%X</tt>'
243 <pre> %result = <a href="#i_mul">mul</a> uint %X, 8<br></pre>
244 <p>After strength reduction:</p>
245 <pre> %result = <a href="#i_shl">shl</a> uint %X, ubyte 3<br></pre>
246 <p>And the hard way:</p>
247 <pre> <a href="#i_add">add</a> uint %X, %X <i>; yields {uint}:%0</i>
249 href="#i_add">add</a> uint %0, %0 <i>; yields {uint}:%1</i>
251 href="#i_add">add</a> uint %1, %1<br></pre>
252 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
253 important lexical features of LLVM:</p>
255 <li>Comments are delimited with a '<tt>;</tt>' and go until the end
257 <li>Unnamed temporaries are created when the result of a computation
258 is not assigned to a named value.</li>
259 <li>Unnamed temporaries are numbered sequentially</li>
261 <p>...and it also show a convention that we follow in this document.
262 When demonstrating instructions, we will follow an instruction with a
263 comment that defines the type and name of value produced. Comments are
264 shown in italic text.</p>
265 <p>The one non-intuitive notation for constants is the optional
266 hexidecimal form of floating point constants. For example, the form '<tt>double
267 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
268 4.5e+15</tt>' which is also supported by the parser. The only time
269 hexadecimal floating point constants are useful (and the only time that
270 they are generated by the disassembler) is when an FP constant has to
271 be emitted that is not representable as a decimal floating point number
272 exactly. For example, NaN's, infinities, and other special cases are
273 represented in their IEEE hexadecimal format so that assembly and
274 disassembly do not cause any bits to change in the constants.</p>
276 <!-- *********************************************************************** -->
277 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
278 <!-- *********************************************************************** -->
279 <div class="doc_text">
280 <p>The LLVM type system is one of the most important features of the
281 intermediate representation. Being typed enables a number of
282 optimizations to be performed on the IR directly, without having to do
283 extra analyses on the side before the transformation. A strong type
284 system makes it easier to read the generated code and enables novel
285 analyses and transformations that are not feasible to perform on normal
286 three address code representations.</p>
287 <!-- The written form for the type system was heavily influenced by the
288 syntactic problems with types in the C language<sup><a
289 href="#rw_stroustrup">1</a></sup>.<p> --> </div>
290 <!-- ======================================================================= -->
291 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
292 <div class="doc_text">
293 <p>The primitive types are the fundamental building blocks of the LLVM
294 system. The current set of primitive types are as follows:</p>
296 <table border="0" style="align: center">
300 <table border="1" cellspacing="0" cellpadding="4" style="align: center">
303 <td><tt>void</tt></td>
307 <td><tt>ubyte</tt></td>
308 <td>Unsigned 8 bit value</td>
311 <td><tt>ushort</tt></td>
312 <td>Unsigned 16 bit value</td>
315 <td><tt>uint</tt></td>
316 <td>Unsigned 32 bit value</td>
319 <td><tt>ulong</tt></td>
320 <td>Unsigned 64 bit value</td>
323 <td><tt>float</tt></td>
324 <td>32 bit floating point value</td>
327 <td><tt>label</tt></td>
328 <td>Branch destination</td>
334 <table border="1" cellspacing="0" cellpadding="4">
337 <td><tt>bool</tt></td>
338 <td>True or False value</td>
341 <td><tt>sbyte</tt></td>
342 <td>Signed 8 bit value</td>
345 <td><tt>short</tt></td>
346 <td>Signed 16 bit value</td>
349 <td><tt>int</tt></td>
350 <td>Signed 32 bit value</td>
353 <td><tt>long</tt></td>
354 <td>Signed 64 bit value</td>
357 <td><tt>double</tt></td>
358 <td>64 bit floating point value</td>
368 <!-- _______________________________________________________________________ -->
369 <div class="doc_subsubsection"> <a name="t_classifications">Type
370 Classifications</a> </div>
371 <div class="doc_text">
372 <p>These different primitive types fall into a few useful
375 <table border="1" cellspacing="0" cellpadding="4">
378 <td><a name="t_signed">signed</a></td>
379 <td><tt>sbyte, short, int, long, float, double</tt></td>
382 <td><a name="t_unsigned">unsigned</a></td>
383 <td><tt>ubyte, ushort, uint, ulong</tt></td>
386 <td><a name="t_integer">integer</a></td>
387 <td><tt>ubyte, sbyte, ushort, short, uint, int, ulong, long</tt></td>
390 <td><a name="t_integral">integral</a></td>
391 <td><tt>bool, ubyte, sbyte, ushort, short, uint, int, ulong, long</tt></td>
394 <td><a name="t_floating">floating point</a></td>
395 <td><tt>float, double</tt></td>
398 <td><a name="t_firstclass">first class</a></td>
399 <td><tt>bool, ubyte, sbyte, ushort, short,<br>
400 uint, int, ulong, long, float, double, <a href="#t_pointer">pointer</a></tt></td>
405 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
406 most important. Values of these types are the only ones which can be
407 produced by instructions, passed as arguments, or used as operands to
408 instructions. This means that all structures and arrays must be
409 manipulated either by pointer or by component.</p>
411 <!-- ======================================================================= -->
412 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
413 <div class="doc_text">
414 <p>The real power in LLVM comes from the derived types in the system.
415 This is what allows a programmer to represent arrays, functions,
416 pointers, and other useful types. Note that these derived types may be
417 recursive: For example, it is possible to have a two dimensional array.</p>
419 <!-- _______________________________________________________________________ -->
420 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
421 <div class="doc_text">
423 <p>The array type is a very simple derived type that arranges elements
424 sequentially in memory. The array type requires a size (number of
425 elements) and an underlying data type.</p>
427 <pre> [<# elements> x <elementtype>]<br></pre>
428 <p>The number of elements is a constant integer value, elementtype may
429 be any type with a size.</p>
431 <p> <tt>[40 x int ]</tt>: Array of 40 integer values.<br>
432 <tt>[41 x int ]</tt>: Array of 41 integer values.<br>
433 <tt>[40 x uint]</tt>: Array of 40 unsigned integer values.</p>
435 <p>Here are some examples of multidimensional arrays:</p>
437 <table border="0" cellpadding="0" cellspacing="0">
440 <td><tt>[3 x [4 x int]]</tt></td>
441 <td>: 3x4 array integer values.</td>
444 <td><tt>[12 x [10 x float]]</tt></td>
445 <td>: 12x10 array of single precision floating point values.</td>
448 <td><tt>[2 x [3 x [4 x uint]]]</tt></td>
449 <td>: 2x3x4 array of unsigned integer values.</td>
455 <!-- _______________________________________________________________________ -->
456 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
457 <div class="doc_text">
459 <p>The function type can be thought of as a function signature. It
460 consists of a return type and a list of formal parameter types.
461 Function types are usually used to build virtual function tables
462 (which are structures of pointers to functions), for indirect function
463 calls, and when defining a function.</p>
465 The return type of a function type cannot be an aggregate type.
468 <pre> <returntype> (<parameter list>)<br></pre>
469 <p>Where '<tt><parameter list></tt>' is a comma-separated list of
470 type specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
471 which indicates that the function takes a variable number of arguments.
472 Variable argument functions can access their arguments with the <a
473 href="#int_varargs">variable argument handling intrinsic</a> functions.</p>
476 <table border="0" cellpadding="0" cellspacing="0">
479 <td><tt>int (int)</tt></td>
480 <td>: function taking an <tt>int</tt>, returning an <tt>int</tt></td>
483 <td><tt>float (int, int *) *</tt></td>
484 <td>: <a href="#t_pointer">Pointer</a> to a function that takes
485 an <tt>int</tt> and a <a href="#t_pointer">pointer</a> to <tt>int</tt>,
486 returning <tt>float</tt>.</td>
489 <td><tt>int (sbyte *, ...)</tt></td>
490 <td>: A vararg function that takes at least one <a
491 href="#t_pointer">pointer</a> to <tt>sbyte</tt> (signed char in C),
492 which returns an integer. This is the signature for <tt>printf</tt>
499 <!-- _______________________________________________________________________ -->
500 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
501 <div class="doc_text">
503 <p>The structure type is used to represent a collection of data members
504 together in memory. The packing of the field types is defined to match
505 the ABI of the underlying processor. The elements of a structure may
506 be any type that has a size.</p>
507 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
508 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
509 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
512 <pre> { <type list> }<br></pre>
515 <table border="0" cellpadding="0" cellspacing="0">
518 <td><tt>{ int, int, int }</tt></td>
519 <td>: a triple of three <tt>int</tt> values</td>
522 <td><tt>{ float, int (int) * }</tt></td>
523 <td>: A pair, where the first element is a <tt>float</tt> and the
524 second element is a <a href="#t_pointer">pointer</a> to a <a
525 href="#t_function">function</a> that takes an <tt>int</tt>, returning
526 an <tt>int</tt>.</td>
532 <!-- _______________________________________________________________________ -->
533 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
534 <div class="doc_text">
536 <p>As in many languages, the pointer type represents a pointer or
537 reference to another object, which must live in memory.</p>
539 <pre> <type> *<br></pre>
542 <table border="0" cellpadding="0" cellspacing="0">
545 <td><tt>[4x int]*</tt></td>
546 <td>: <a href="#t_pointer">pointer</a> to <a href="#t_array">array</a>
547 of four <tt>int</tt> values</td>
550 <td><tt>int (int *) *</tt></td>
551 <td>: A <a href="#t_pointer">pointer</a> to a <a
552 href="#t_function">function</a> that takes an <tt>int</tt>, returning
553 an <tt>int</tt>.</td>
559 <!-- _______________________________________________________________________ --><!--
560 <div class="doc_subsubsection">
561 <a name="t_packed">Packed Type</a>
564 <div class="doc_text">
566 Mention/decide that packed types work with saturation or not. Maybe have a packed+saturated type in addition to just a packed type.<p>
568 Packed types should be 'nonsaturated' because standard data types are not saturated. Maybe have a saturated packed type?<p>
572 --><!-- *********************************************************************** -->
573 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
574 <!-- *********************************************************************** --><!-- ======================================================================= -->
575 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a> </div>
576 <div class="doc_text">
577 <p>LLVM programs are composed of "Module"s, each of which is a
578 translation unit of the input programs. Each module consists of
579 functions, global variables, and symbol table entries. Modules may be
580 combined together with the LLVM linker, which merges function (and
581 global variable) definitions, resolves forward declarations, and merges
582 symbol table entries. Here is an example of the "hello world" module:</p>
583 <pre><i>; Declare the string constant as a global constant...</i>
584 <a href="#identifiers">%.LC0</a> = <a href="#linkage_internal">internal</a> <a
585 href="#globalvars">constant</a> <a href="#t_array">[13 x sbyte]</a> c"hello world\0A\00" <i>; [13 x sbyte]*</i>
587 <i>; External declaration of the puts function</i>
588 <a href="#functionstructure">declare</a> int %puts(sbyte*) <i>; int(sbyte*)* </i>
590 <i>; Definition of main function</i>
591 int %main() { <i>; int()* </i>
592 <i>; Convert [13x sbyte]* to sbyte *...</i>
594 href="#i_getelementptr">getelementptr</a> [13 x sbyte]* %.LC0, long 0, long 0 <i>; sbyte*</i>
596 <i>; Call puts function to write out the string to stdout...</i>
598 href="#i_call">call</a> int %puts(sbyte* %cast210) <i>; int</i>
600 href="#i_ret">ret</a> int 0<br>}<br></pre>
601 <p>This example is made up of a <a href="#globalvars">global variable</a>
602 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
603 function, and a <a href="#functionstructure">function definition</a>
604 for "<tt>main</tt>".</p>
605 <a name="linkage"> In general, a module is made up of a list of global
606 values, where both functions and global variables are global values.
607 Global values are represented by a pointer to a memory location (in
608 this case, a pointer to an array of char, and a pointer to a function),
609 and have one of the following linkage types:</a>
612 <dt><tt><b><a name="linkage_internal">internal</a></b></tt> </dt>
613 <dd>Global values with internal linkage are only directly accessible
614 by objects in the current module. In particular, linking code into a
615 module with an internal global value may cause the internal to be
616 renamed as necessary to avoid collisions. Because the symbol is
617 internal to the module, all references can be updated. This
618 corresponds to the notion of the '<tt>static</tt>' keyword in C, or the
619 idea of "anonymous namespaces" in C++.
622 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
623 <dd>"<tt>linkonce</tt>" linkage is similar to <tt>internal</tt>
624 linkage, with the twist that linking together two modules defining the
625 same <tt>linkonce</tt> globals will cause one of the globals to be
626 discarded. This is typically used to implement inline functions.
627 Unreferenced <tt>linkonce</tt> globals are allowed to be discarded.
630 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
631 <dd>"<tt>weak</tt>" linkage is exactly the same as <tt>linkonce</tt>
632 linkage, except that unreferenced <tt>weak</tt> globals may not be
633 discarded. This is used to implement constructs in C such as "<tt>int
634 X;</tt>" at global scope.
637 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
638 <dd>"<tt>appending</tt>" linkage may only be applied to global
639 variables of pointer to array type. When two global variables with
640 appending linkage are linked together, the two global arrays are
641 appended together. This is the LLVM, typesafe, equivalent of having
642 the system linker append together "sections" with identical names when
646 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
647 <dd>If none of the above identifiers are used, the global is
648 externally visible, meaning that it participates in linkage and can be
649 used to resolve external symbol references.
654 <p><a name="linkage_external">For example, since the "<tt>.LC0</tt>"
655 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
656 variable and was linked with this one, one of the two would be renamed,
657 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
658 external (i.e., lacking any linkage declarations), they are accessible
659 outside of the current module. It is illegal for a function <i>declaration</i>
660 to have any linkage type other than "externally visible".</a></p>
663 <!-- ======================================================================= -->
664 <div class="doc_subsection">
665 <a name="globalvars">Global Variables</a>
668 <div class="doc_text">
670 <p>Global variables define regions of memory allocated at compilation
671 time instead of run-time. Global variables may optionally be
672 initialized. A variable may be defined as a global "constant", which
673 indicates that the contents of the variable will never be modified
674 (opening options for optimization).</p>
676 <p>As SSA values, global variables define pointer values that are in
677 scope (i.e. they dominate) for all basic blocks in the program. Global
678 variables always define a pointer to their "content" type because they
679 describe a region of memory, and all memory objects in LLVM are
680 accessed through pointers.</p>
685 <!-- ======================================================================= -->
686 <div class="doc_subsection">
687 <a name="functionstructure">Functions</a>
690 <div class="doc_text">
692 <p>LLVM function definitions are composed of a (possibly empty) argument list,
693 an opening curly brace, a list of basic blocks, and a closing curly brace. LLVM
694 function declarations are defined with the "<tt>declare</tt>" keyword, a
695 function name, and a function signature.</p>
697 <p>A function definition contains a list of basic blocks, forming the CFG for
698 the function. Each basic block may optionally start with a label (giving the
699 basic block a symbol table entry), contains a list of instructions, and ends
700 with a <a href="#terminators">terminator</a> instruction (such as a branch or
701 function return).</p>
703 <p>The first basic block in program is special in two ways: it is immediately
704 executed on entrance to the function, and it is not allowed to have predecessor
705 basic blocks (i.e. there can not be any branches to the entry block of a
706 function). Because the block can have no predecessors, it also cannot have any
707 <a href="#i_phi">PHI nodes</a>.</p>
709 <p>LLVM functions are identified by their name and type signature. Hence, two
710 functions with the same name but different parameter lists or return values are
711 considered different functions, and LLVM will resolves references to each
717 <!-- *********************************************************************** -->
718 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
719 <!-- *********************************************************************** -->
720 <div class="doc_text">
721 <p>The LLVM instruction set consists of several different
722 classifications of instructions: <a href="#terminators">terminator
723 instructions</a>, <a href="#binaryops">binary instructions</a>, <a
724 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
725 instructions</a>.</p>
727 <!-- ======================================================================= -->
728 <div class="doc_subsection"> <a name="terminators">Terminator
729 Instructions</a> </div>
730 <div class="doc_text">
731 <p>As mentioned <a href="#functionstructure">previously</a>, every
732 basic block in a program ends with a "Terminator" instruction, which
733 indicates which block should be executed after the current block is
734 finished. These terminator instructions typically yield a '<tt>void</tt>'
735 value: they produce control flow, not values (the one exception being
736 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
737 <p>There are five different terminator instructions: the '<a
738 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
739 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
740 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, and the '<a
741 href="#i_unwind"><tt>unwind</tt></a>' instruction.</p>
743 <!-- _______________________________________________________________________ -->
744 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
745 Instruction</a> </div>
746 <div class="doc_text">
748 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
749 ret void <i>; Return from void function</i>
752 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
753 value) from a function, back to the caller.</p>
754 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
755 returns a value and then causes control flow, and one that just causes
756 control flow to occur.</p>
758 <p>The '<tt>ret</tt>' instruction may return any '<a
759 href="#t_firstclass">first class</a>' type. Notice that a function is
760 not <a href="#wellformed">well formed</a> if there exists a '<tt>ret</tt>'
761 instruction inside of the function that returns a value that does not
762 match the return type of the function.</p>
764 <p>When the '<tt>ret</tt>' instruction is executed, control flow
765 returns back to the calling function's context. If the caller is a "<a
766 href="#i_call"><tt>call</tt></a> instruction, execution continues at
767 the instruction after the call. If the caller was an "<a
768 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
769 at the beginning "normal" of the destination block. If the instruction
770 returns a value, that value shall set the call or invoke instruction's
773 <pre> ret int 5 <i>; Return an integer value of 5</i>
774 ret void <i>; Return from a void function</i>
777 <!-- _______________________________________________________________________ -->
778 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
779 <div class="doc_text">
781 <pre> br bool <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
784 <p>The '<tt>br</tt>' instruction is used to cause control flow to
785 transfer to a different basic block in the current function. There are
786 two forms of this instruction, corresponding to a conditional branch
787 and an unconditional branch.</p>
789 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
790 single '<tt>bool</tt>' value and two '<tt>label</tt>' values. The
791 unconditional form of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>'
792 value as a target.</p>
794 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>bool</tt>'
795 argument is evaluated. If the value is <tt>true</tt>, control flows
796 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
797 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
799 <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
800 href="#i_ret">ret</a> int 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> int 0<br></pre>
802 <!-- _______________________________________________________________________ -->
803 <div class="doc_subsubsection">
804 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
807 <div class="doc_text">
811 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
816 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
817 several different places. It is a generalization of the '<tt>br</tt>'
818 instruction, allowing a branch to occur to one of many possible
824 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
825 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
826 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
827 table is not allowed to contain duplicate constant entries.</p>
831 <p>The <tt>switch</tt> instruction specifies a table of values and
832 destinations. When the '<tt>switch</tt>' instruction is executed, this
833 table is searched for the given value. If the value is found, the
834 corresponding destination is branched to, otherwise the default value
835 it transfered to.</p>
837 <h5>Implementation:</h5>
839 <p>Depending on properties of the target machine and the particular
840 <tt>switch</tt> instruction, this instruction may be code generated in different
841 ways, for example as a series of chained conditional branches, or with a lookup
847 <i>; Emulate a conditional br instruction</i>
848 %Val = <a href="#i_cast">cast</a> bool %value to int
849 switch int %Val, label %truedest [int 0, label %falsedest ]
851 <i>; Emulate an unconditional br instruction</i>
852 switch uint 0, label %dest [ ]
854 <i>; Implement a jump table:</i>
855 switch uint %val, label %otherwise [ uint 0, label %onzero
857 uint 2, label %ontwo ]
860 <!-- _______________________________________________________________________ -->
861 <div class="doc_subsubsection"> <a name="i_invoke">'<tt>invoke</tt>'
862 Instruction</a> </div>
863 <div class="doc_text">
865 <pre> <result> = invoke <ptr to function ty> %<function ptr val>(<function args>)<br> to label <normal label> except label <exception label><br></pre>
867 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a
868 specified function, with the possibility of control flow transfer to
869 either the '<tt>normal</tt>' <tt>label</tt> label or the '<tt>exception</tt>'<tt>label</tt>.
870 If the callee function returns with the "<tt><a href="#i_ret">ret</a></tt>"
871 instruction, control flow will return to the "normal" label. If the
872 callee (or any indirect callees) returns with the "<a href="#i_unwind"><tt>unwind</tt></a>"
873 instruction, control is interrupted, and continued at the dynamically
874 nearest "except" label.</p>
876 <p>This instruction requires several arguments:</p>
878 <li>'<tt>ptr to function ty</tt>': shall be the signature of the
879 pointer to function value being invoked. In most cases, this is a
880 direct function invocation, but indirect <tt>invoke</tt>s are just as
881 possible, branching off an arbitrary pointer to function value. </li>
882 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer
883 to a function to be invoked. </li>
884 <li>'<tt>function args</tt>': argument list whose types match the
885 function signature argument types. If the function signature indicates
886 the function accepts a variable number of arguments, the extra
887 arguments can be specified. </li>
888 <li>'<tt>normal label</tt>': the label reached when the called
889 function executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
890 <li>'<tt>exception label</tt>': the label reached when a callee
891 returns with the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
894 <p>This instruction is designed to operate as a standard '<tt><a
895 href="#i_call">call</a></tt>' instruction in most regards. The
896 primary difference is that it establishes an association with a label,
897 which is used by the runtime library to unwind the stack.</p>
898 <p>This instruction is used in languages with destructors to ensure
899 that proper cleanup is performed in the case of either a <tt>longjmp</tt>
900 or a thrown exception. Additionally, this is important for
901 implementation of '<tt>catch</tt>' clauses in high-level languages that
904 <pre> %retval = invoke int %Test(int 15)<br> to label %Continue<br> except label %TestCleanup <i>; {int}:retval set</i>
907 <!-- _______________________________________________________________________ -->
908 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
909 Instruction</a> </div>
910 <div class="doc_text">
912 <pre> unwind<br></pre>
914 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing
915 control flow at the first callee in the dynamic call stack which used
916 an <a href="#i_invoke"><tt>invoke</tt></a> instruction to perform the
917 call. This is primarily used to implement exception handling.</p>
919 <p>The '<tt>unwind</tt>' intrinsic causes execution of the current
920 function to immediately halt. The dynamic call stack is then searched
921 for the first <a href="#i_invoke"><tt>invoke</tt></a> instruction on
922 the call stack. Once found, execution continues at the "exceptional"
923 destination block specified by the <tt>invoke</tt> instruction. If
924 there is no <tt>invoke</tt> instruction in the dynamic call chain,
925 undefined behavior results.</p>
927 <!-- ======================================================================= -->
928 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
929 <div class="doc_text">
930 <p>Binary operators are used to do most of the computation in a
931 program. They require two operands, execute an operation on them, and
932 produce a single value. The result value of a binary operator is not
933 necessarily the same type as its operands.</p>
934 <p>There are several different binary operators:</p>
936 <!-- _______________________________________________________________________ -->
937 <div class="doc_subsubsection"> <a name="i_add">'<tt>add</tt>'
938 Instruction</a> </div>
939 <div class="doc_text">
941 <pre> <result> = add <ty> <var1>, <var2> <i>; yields {ty}:result</i>
944 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
946 <p>The two arguments to the '<tt>add</tt>' instruction must be either <a
947 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
948 values. Both arguments must have identical types.</p>
950 <p>The value produced is the integer or floating point sum of the two
953 <pre> <result> = add int 4, %var <i>; yields {int}:result = 4 + %var</i>
956 <!-- _______________________________________________________________________ -->
957 <div class="doc_subsubsection"> <a name="i_sub">'<tt>sub</tt>'
958 Instruction</a> </div>
959 <div class="doc_text">
961 <pre> <result> = sub <ty> <var1>, <var2> <i>; yields {ty}:result</i>
964 <p>The '<tt>sub</tt>' instruction returns the difference of its two
966 <p>Note that the '<tt>sub</tt>' instruction is used to represent the '<tt>neg</tt>'
967 instruction present in most other intermediate representations.</p>
969 <p>The two arguments to the '<tt>sub</tt>' instruction must be either <a
970 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
971 values. Both arguments must have identical types.</p>
973 <p>The value produced is the integer or floating point difference of
974 the two operands.</p>
976 <pre> <result> = sub int 4, %var <i>; yields {int}:result = 4 - %var</i>
977 <result> = sub int 0, %val <i>; yields {int}:result = -%var</i>
980 <!-- _______________________________________________________________________ -->
981 <div class="doc_subsubsection"> <a name="i_mul">'<tt>mul</tt>'
982 Instruction</a> </div>
983 <div class="doc_text">
985 <pre> <result> = mul <ty> <var1>, <var2> <i>; yields {ty}:result</i>
988 <p>The '<tt>mul</tt>' instruction returns the product of its two
991 <p>The two arguments to the '<tt>mul</tt>' instruction must be either <a
992 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
993 values. Both arguments must have identical types.</p>
995 <p>The value produced is the integer or floating point product of the
997 <p>There is no signed vs unsigned multiplication. The appropriate
998 action is taken based on the type of the operand.</p>
1000 <pre> <result> = mul int 4, %var <i>; yields {int}:result = 4 * %var</i>
1003 <!-- _______________________________________________________________________ -->
1004 <div class="doc_subsubsection"> <a name="i_div">'<tt>div</tt>'
1005 Instruction</a> </div>
1006 <div class="doc_text">
1008 <pre> <result> = div <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1011 <p>The '<tt>div</tt>' instruction returns the quotient of its two
1014 <p>The two arguments to the '<tt>div</tt>' instruction must be either <a
1015 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1016 values. Both arguments must have identical types.</p>
1018 <p>The value produced is the integer or floating point quotient of the
1021 <pre> <result> = div int 4, %var <i>; yields {int}:result = 4 / %var</i>
1024 <!-- _______________________________________________________________________ -->
1025 <div class="doc_subsubsection"> <a name="i_rem">'<tt>rem</tt>'
1026 Instruction</a> </div>
1027 <div class="doc_text">
1029 <pre> <result> = rem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1032 <p>The '<tt>rem</tt>' instruction returns the remainder from the
1033 division of its two operands.</p>
1035 <p>The two arguments to the '<tt>rem</tt>' instruction must be either <a
1036 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1037 values. Both arguments must have identical types.</p>
1039 <p>This returns the <i>remainder</i> of a division (where the result
1040 has the same sign as the divisor), not the <i>modulus</i> (where the
1041 result has the same sign as the dividend) of a value. For more
1042 information about the difference, see: <a
1043 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
1046 <pre> <result> = rem int 4, %var <i>; yields {int}:result = 4 % %var</i>
1049 <!-- _______________________________________________________________________ -->
1050 <div class="doc_subsubsection"> <a name="i_setcc">'<tt>set<i>cc</i></tt>'
1051 Instructions</a> </div>
1052 <div class="doc_text">
1054 <pre> <result> = seteq <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1055 <result> = setne <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1056 <result> = setlt <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1057 <result> = setgt <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1058 <result> = setle <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1059 <result> = setge <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1062 <p>The '<tt>set<i>cc</i></tt>' family of instructions returns a boolean
1063 value based on a comparison of their two operands.</p>
1065 <p>The two arguments to the '<tt>set<i>cc</i></tt>' instructions must
1066 be of <a href="#t_firstclass">first class</a> type (it is not possible
1067 to compare '<tt>label</tt>'s, '<tt>array</tt>'s, '<tt>structure</tt>'
1068 or '<tt>void</tt>' values, etc...). Both arguments must have identical
1071 <p>The '<tt>seteq</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1072 value if both operands are equal.<br>
1073 The '<tt>setne</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1074 value if both operands are unequal.<br>
1075 The '<tt>setlt</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1076 value if the first operand is less than the second operand.<br>
1077 The '<tt>setgt</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1078 value if the first operand is greater than the second operand.<br>
1079 The '<tt>setle</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1080 value if the first operand is less than or equal to the second operand.<br>
1081 The '<tt>setge</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1082 value if the first operand is greater than or equal to the second
1085 <pre> <result> = seteq int 4, 5 <i>; yields {bool}:result = false</i>
1086 <result> = setne float 4, 5 <i>; yields {bool}:result = true</i>
1087 <result> = setlt uint 4, 5 <i>; yields {bool}:result = true</i>
1088 <result> = setgt sbyte 4, 5 <i>; yields {bool}:result = false</i>
1089 <result> = setle sbyte 4, 5 <i>; yields {bool}:result = true</i>
1090 <result> = setge sbyte 4, 5 <i>; yields {bool}:result = false</i>
1093 <!-- ======================================================================= -->
1094 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
1095 Operations</a> </div>
1096 <div class="doc_text">
1097 <p>Bitwise binary operators are used to do various forms of
1098 bit-twiddling in a program. They are generally very efficient
1099 instructions, and can commonly be strength reduced from other
1100 instructions. They require two operands, execute an operation on them,
1101 and produce a single value. The resulting value of the bitwise binary
1102 operators is always the same type as its first operand.</p>
1104 <!-- _______________________________________________________________________ -->
1105 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
1106 Instruction</a> </div>
1107 <div class="doc_text">
1109 <pre> <result> = and <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1112 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
1113 its two operands.</p>
1115 <p>The two arguments to the '<tt>and</tt>' instruction must be <a
1116 href="#t_integral">integral</a> values. Both arguments must have
1117 identical types.</p>
1119 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
1121 <div style="align: center">
1122 <table border="1" cellspacing="0" cellpadding="4">
1153 <pre> <result> = and int 4, %var <i>; yields {int}:result = 4 & %var</i>
1154 <result> = and int 15, 40 <i>; yields {int}:result = 8</i>
1155 <result> = and int 4, 8 <i>; yields {int}:result = 0</i>
1158 <!-- _______________________________________________________________________ -->
1159 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
1160 <div class="doc_text">
1162 <pre> <result> = or <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1165 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
1166 or of its two operands.</p>
1168 <p>The two arguments to the '<tt>or</tt>' instruction must be <a
1169 href="#t_integral">integral</a> values. Both arguments must have
1170 identical types.</p>
1172 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
1174 <div style="align: center">
1175 <table border="1" cellspacing="0" cellpadding="4">
1206 <pre> <result> = or int 4, %var <i>; yields {int}:result = 4 | %var</i>
1207 <result> = or int 15, 40 <i>; yields {int}:result = 47</i>
1208 <result> = or int 4, 8 <i>; yields {int}:result = 12</i>
1211 <!-- _______________________________________________________________________ -->
1212 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
1213 Instruction</a> </div>
1214 <div class="doc_text">
1216 <pre> <result> = xor <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1219 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
1220 or of its two operands. The <tt>xor</tt> is used to implement the
1221 "one's complement" operation, which is the "~" operator in C.</p>
1223 <p>The two arguments to the '<tt>xor</tt>' instruction must be <a
1224 href="#t_integral">integral</a> values. Both arguments must have
1225 identical types.</p>
1227 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
1229 <div style="align: center">
1230 <table border="1" cellspacing="0" cellpadding="4">
1262 <pre> <result> = xor int 4, %var <i>; yields {int}:result = 4 ^ %var</i>
1263 <result> = xor int 15, 40 <i>; yields {int}:result = 39</i>
1264 <result> = xor int 4, 8 <i>; yields {int}:result = 12</i>
1265 <result> = xor int %V, -1 <i>; yields {int}:result = ~%V</i>
1268 <!-- _______________________________________________________________________ -->
1269 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
1270 Instruction</a> </div>
1271 <div class="doc_text">
1273 <pre> <result> = shl <ty> <var1>, ubyte <var2> <i>; yields {ty}:result</i>
1276 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
1277 the left a specified number of bits.</p>
1279 <p>The first argument to the '<tt>shl</tt>' instruction must be an <a
1280 href="#t_integer">integer</a> type. The second argument must be an '<tt>ubyte</tt>'
1283 <p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup>.</p>
1285 <pre> <result> = shl int 4, ubyte %var <i>; yields {int}:result = 4 << %var</i>
1286 <result> = shl int 4, ubyte 2 <i>; yields {int}:result = 16</i>
1287 <result> = shl int 1, ubyte 10 <i>; yields {int}:result = 1024</i>
1290 <!-- _______________________________________________________________________ -->
1291 <div class="doc_subsubsection"> <a name="i_shr">'<tt>shr</tt>'
1292 Instruction</a> </div>
1293 <div class="doc_text">
1295 <pre> <result> = shr <ty> <var1>, ubyte <var2> <i>; yields {ty}:result</i>
1298 <p>The '<tt>shr</tt>' instruction returns the first operand shifted to
1299 the right a specified number of bits.</p>
1301 <p>The first argument to the '<tt>shr</tt>' instruction must be an <a
1302 href="#t_integer">integer</a> type. The second argument must be an '<tt>ubyte</tt>'
1305 <p>If the first argument is a <a href="#t_signed">signed</a> type, the
1306 most significant bit is duplicated in the newly free'd bit positions.
1307 If the first argument is unsigned, zero bits shall fill the empty
1310 <pre> <result> = shr int 4, ubyte %var <i>; yields {int}:result = 4 >> %var</i>
1311 <result> = shr uint 4, ubyte 1 <i>; yields {uint}:result = 2</i>
1312 <result> = shr int 4, ubyte 2 <i>; yields {int}:result = 1</i>
1313 <result> = shr sbyte 4, ubyte 3 <i>; yields {sbyte}:result = 0</i>
1314 <result> = shr sbyte -2, ubyte 1 <i>; yields {sbyte}:result = -1</i>
1317 <!-- ======================================================================= -->
1318 <div class="doc_subsection"> <a name="memoryops">Memory Access
1319 Operations</a></div>
1320 <div class="doc_text">
1321 <p>A key design point of an SSA-based representation is how it
1322 represents memory. In LLVM, no memory locations are in SSA form, which
1323 makes things very simple. This section describes how to read, write,
1324 allocate and free memory in LLVM.</p>
1326 <!-- _______________________________________________________________________ -->
1327 <div class="doc_subsubsection"> <a name="i_malloc">'<tt>malloc</tt>'
1328 Instruction</a> </div>
1329 <div class="doc_text">
1331 <pre> <result> = malloc <type>, uint <NumElements> <i>; yields {type*}:result</i>
1332 <result> = malloc <type> <i>; yields {type*}:result</i>
1335 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
1336 heap and returns a pointer to it.</p>
1338 <p>The '<tt>malloc</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
1339 bytes of memory from the operating system and returns a pointer of the
1340 appropriate type to the program. The second form of the instruction is
1341 a shorter version of the first instruction that defaults to allocating
1343 <p>'<tt>type</tt>' must be a sized type.</p>
1345 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
1346 a pointer is returned.</p>
1348 <pre> %array = malloc [4 x ubyte ] <i>; yields {[%4 x ubyte]*}:array</i>
1351 href="#i_add">add</a> uint 2, 2 <i>; yields {uint}:size = uint 4</i>
1352 %array1 = malloc ubyte, uint 4 <i>; yields {ubyte*}:array1</i>
1353 %array2 = malloc [12 x ubyte], uint %size <i>; yields {[12 x ubyte]*}:array2</i>
1356 <!-- _______________________________________________________________________ -->
1357 <div class="doc_subsubsection"> <a name="i_free">'<tt>free</tt>'
1358 Instruction</a> </div>
1359 <div class="doc_text">
1361 <pre> free <type> <value> <i>; yields {void}</i>
1364 <p>The '<tt>free</tt>' instruction returns memory back to the unused
1365 memory heap, to be reallocated in the future.</p>
1368 <p>'<tt>value</tt>' shall be a pointer value that points to a value
1369 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
1372 <p>Access to the memory pointed to by the pointer is not longer defined
1373 after this instruction executes.</p>
1375 <pre> %array = <a href="#i_malloc">malloc</a> [4 x ubyte] <i>; yields {[4 x ubyte]*}:array</i>
1376 free [4 x ubyte]* %array
1379 <!-- _______________________________________________________________________ -->
1380 <div class="doc_subsubsection"> <a name="i_alloca">'<tt>alloca</tt>'
1381 Instruction</a> </div>
1382 <div class="doc_text">
1384 <pre> <result> = alloca <type>, uint <NumElements> <i>; yields {type*}:result</i>
1385 <result> = alloca <type> <i>; yields {type*}:result</i>
1388 <p>The '<tt>alloca</tt>' instruction allocates memory on the current
1389 stack frame of the procedure that is live until the current function
1390 returns to its caller.</p>
1392 <p>The the '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
1393 bytes of memory on the runtime stack, returning a pointer of the
1394 appropriate type to the program. The second form of the instruction is
1395 a shorter version of the first that defaults to allocating one element.</p>
1396 <p>'<tt>type</tt>' may be any sized type.</p>
1398 <p>Memory is allocated, a pointer is returned. '<tt>alloca</tt>'d
1399 memory is automatically released when the function returns. The '<tt>alloca</tt>'
1400 instruction is commonly used to represent automatic variables that must
1401 have an address available. When the function returns (either with the <tt><a
1402 href="#i_ret">ret</a></tt> or <tt><a href="#i_invoke">invoke</a></tt>
1403 instructions), the memory is reclaimed.</p>
1405 <pre> %ptr = alloca int <i>; yields {int*}:ptr</i>
1406 %ptr = alloca int, uint 4 <i>; yields {int*}:ptr</i>
1409 <!-- _______________________________________________________________________ -->
1410 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
1411 Instruction</a> </div>
1412 <div class="doc_text">
1414 <pre> <result> = load <ty>* <pointer><br> <result> = volatile load <ty>* <pointer><br></pre>
1416 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
1418 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
1419 address to load from. The pointer must point to a <a
1420 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
1421 marked as <tt>volatile</tt> then the optimizer is not allowed to modify
1422 the number or order of execution of this <tt>load</tt> with other
1423 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
1426 <p>The location of memory pointed to is loaded.</p>
1428 <pre> %ptr = <a href="#i_alloca">alloca</a> int <i>; yields {int*}:ptr</i>
1430 href="#i_store">store</a> int 3, int* %ptr <i>; yields {void}</i>
1431 %val = load int* %ptr <i>; yields {int}:val = int 3</i>
1434 <!-- _______________________________________________________________________ -->
1435 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
1436 Instruction</a> </div>
1438 <pre> store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
1439 volatile store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
1442 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
1444 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
1445 to store and an address to store it into. The type of the '<tt><pointer></tt>'
1446 operand must be a pointer to the type of the '<tt><value></tt>'
1447 operand. If the <tt>store</tt> is marked as <tt>volatile</tt> then the
1448 optimizer is not allowed to modify the number or order of execution of
1449 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
1450 href="#i_store">store</a></tt> instructions.</p>
1452 <p>The contents of memory are updated to contain '<tt><value></tt>'
1453 at the location specified by the '<tt><pointer></tt>' operand.</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>
1460 <!-- _______________________________________________________________________ -->
1461 <div class="doc_subsubsection">
1462 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
1465 <div class="doc_text">
1468 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
1474 The '<tt>getelementptr</tt>' instruction is used to get the address of a
1475 subelement of an aggregate data structure.</p>
1479 <p>This instruction takes a list of integer constants that indicate what
1480 elements of the aggregate object to index to. The actual types of the arguments
1481 provided depend on the type of the first pointer argument. The
1482 '<tt>getelementptr</tt>' instruction is used to index down through the type
1483 levels of a structure. When indexing into a structure, only <tt>uint</tt>
1484 integer constants are allowed. When indexing into an array or pointer
1485 <tt>int</tt> and <tt>long</tt> indexes are allowed of any sign.</p>
1487 <p>For example, let's consider a C code fragment and how it gets
1488 compiled to LLVM:</p>
1502 int *foo(struct ST *s) {
1503 return &s[1].Z.B[5][13];
1507 <p>The LLVM code generated by the GCC frontend is:</p>
1510 %RT = type { sbyte, [10 x [20 x int]], sbyte }
1511 %ST = type { int, double, %RT }
1513 int* "foo"(%ST* %s) {
1514 %reg = getelementptr %ST* %s, int 1, uint 2, uint 1, int 5, int 13<br>
1521 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
1522 on the pointer type that is being index into. <a href="#t_pointer">Pointer</a>
1523 and <a href="#t_array">array</a> types require <tt>uint</tt>, <tt>int</tt>,
1524 <tt>ulong</tt>, or <tt>long</tt> values, and <a href="#t_struct">structure</a>
1525 types require <tt>uint</tt> <b>constants</b>.</p>
1527 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
1528 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ int, double, %RT
1529 }</tt>' type, a structure. The second index indexes into the third element of
1530 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ sbyte, [10 x [20 x int]],
1531 sbyte }</tt>' type, another structure. The third index indexes into the second
1532 element of the structure, yielding a '<tt>[10 x [20 x int]]</tt>' type, an
1533 array. The two dimensions of the array are subscripted into, yielding an
1534 '<tt>int</tt>' type. The '<tt>getelementptr</tt>' instruction return a pointer
1535 to this element, thus computing a value of '<tt>int*</tt>' type.</p>
1537 <p>Note that it is perfectly legal to index partially through a
1538 structure, returning a pointer to an inner element. Because of this,
1539 the LLVM code for the given testcase is equivalent to:</p>
1542 int* "foo"(%ST* %s) {
1543 %t1 = getelementptr %ST* %s, int 1 <i>; yields %ST*:%t1</i>
1544 %t2 = getelementptr %ST* %t1, int 0, uint 2 <i>; yields %RT*:%t2</i>
1545 %t3 = getelementptr %RT* %t2, int 0, uint 1 <i>; yields [10 x [20 x int]]*:%t3</i>
1546 %t4 = getelementptr [10 x [20 x int]]* %t3, int 0, int 5 <i>; yields [20 x int]*:%t4</i>
1547 %t5 = getelementptr [20 x int]* %t4, int 0, int 13 <i>; yields int*:%t5</i>
1553 <i>; yields [12 x ubyte]*:aptr</i>
1554 %aptr = getelementptr {int, [12 x ubyte]}* %sptr, long 0, uint 1
1558 <!-- ======================================================================= -->
1559 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
1560 <div class="doc_text">
1561 <p>The instructions in this category are the "miscellaneous"
1562 instructions, which defy better classification.</p>
1564 <!-- _______________________________________________________________________ -->
1565 <div class="doc_subsubsection"> <a name="i_phi">'<tt>phi</tt>'
1566 Instruction</a> </div>
1567 <div class="doc_text">
1569 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
1571 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
1572 the SSA graph representing the function.</p>
1574 <p>The type of the incoming values are specified with the first type
1575 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
1576 as arguments, with one pair for each predecessor basic block of the
1577 current block. Only values of <a href="#t_firstclass">first class</a>
1578 type may be used as the value arguments to the PHI node. Only labels
1579 may be used as the label arguments.</p>
1580 <p>There must be no non-phi instructions between the start of a basic
1581 block and the PHI instructions: i.e. PHI instructions must be first in
1584 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the
1585 value specified by the parameter, depending on which basic block we
1586 came from in the last <a href="#terminators">terminator</a> instruction.</p>
1588 <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>
1591 <!-- _______________________________________________________________________ -->
1592 <div class="doc_subsubsection">
1593 <a name="i_cast">'<tt>cast .. to</tt>' Instruction</a>
1596 <div class="doc_text">
1601 <result> = cast <ty> <value> to <ty2> <i>; yields ty2</i>
1607 The '<tt>cast</tt>' instruction is used as the primitive means to convert
1608 integers to floating point, change data type sizes, and break type safety (by
1616 The '<tt>cast</tt>' instruction takes a value to cast, which must be a first
1617 class value, and a type to cast it to, which must also be a <a
1618 href="#t_firstclass">first class</a> type.
1624 This instruction follows the C rules for explicit casts when determining how the
1625 data being cast must change to fit in its new container.
1629 When casting to bool, any value that would be considered true in the context of
1630 a C '<tt>if</tt>' condition is converted to the boolean '<tt>true</tt>' values,
1631 all else are '<tt>false</tt>'.
1635 When extending an integral value from a type of one signness to another (for
1636 example '<tt>sbyte</tt>' to '<tt>ulong</tt>'), the value is sign-extended if the
1637 <b>source</b> value is signed, and zero-extended if the source value is
1638 unsigned. <tt>bool</tt> values are always zero extended into either zero or
1645 %X = cast int 257 to ubyte <i>; yields ubyte:1</i>
1646 %Y = cast int 123 to bool <i>; yields bool:true</i>
1650 <!-- _______________________________________________________________________ -->
1651 <div class="doc_subsubsection">
1652 <a name="i_select">'<tt>select</tt>' Instruction</a>
1655 <div class="doc_text">
1660 <result> = select bool <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
1666 The '<tt>select</tt>' instruction is used to choose one value based on a
1667 condition, without branching.
1674 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.
1680 If the boolean condition evaluates to true, the instruction returns the first
1681 value argument, otherwise it returns the second value argument.
1687 %X = select bool true, ubyte 17, ubyte 42 <i>; yields ubyte:17</i>
1695 <!-- _______________________________________________________________________ -->
1696 <div class="doc_subsubsection"> <a name="i_call">'<tt>call</tt>'
1697 Instruction</a> </div>
1698 <div class="doc_text">
1700 <pre> <result> = call <ty>* <fnptrval>(<param list>)<br></pre>
1702 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
1704 <p>This instruction requires several arguments:</p>
1707 <p>'<tt>ty</tt>': shall be the signature of the pointer to function
1708 value being invoked. The argument types must match the types implied
1709 by this signature.</p>
1712 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a
1713 function to be invoked. In most cases, this is a direct function
1714 invocation, but indirect <tt>call</tt>s are just as possible,
1715 calling an arbitrary pointer to function values.</p>
1718 <p>'<tt>function args</tt>': argument list whose types match the
1719 function signature argument types. If the function signature
1720 indicates the function accepts a variable number of arguments, the
1721 extra arguments can be specified.</p>
1725 <p>The '<tt>call</tt>' instruction is used to cause control flow to
1726 transfer to a specified function, with its incoming arguments bound to
1727 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
1728 instruction in the called function, control flow continues with the
1729 instruction after the function call, and the return value of the
1730 function is bound to the result argument. This is a simpler case of
1731 the <a href="#i_invoke">invoke</a> instruction.</p>
1733 <pre> %retval = call int %test(int %argc)<br> call int(sbyte*, ...) *%printf(sbyte* %msg, int 12, sbyte 42);<br></pre>
1735 <!-- _______________________________________________________________________ -->
1736 <div class="doc_subsubsection"> <a name="i_vanext">'<tt>vanext</tt>'
1737 Instruction</a> </div>
1738 <div class="doc_text">
1740 <pre> <resultarglist> = vanext <va_list> <arglist>, <argty><br></pre>
1742 <p>The '<tt>vanext</tt>' instruction is used to access arguments passed
1743 through the "variable argument" area of a function call. It is used to
1744 implement the <tt>va_arg</tt> macro in C.</p>
1746 <p>This instruction takes a <tt>valist</tt> value and the type of the
1747 argument. It returns another <tt>valist</tt>.</p>
1749 <p>The '<tt>vanext</tt>' instruction advances the specified <tt>valist</tt>
1750 past an argument of the specified type. In conjunction with the <a
1751 href="#i_vaarg"><tt>vaarg</tt></a> instruction, it is used to implement
1752 the <tt>va_arg</tt> macro available in C. For more information, see
1753 the variable argument handling <a href="#int_varargs">Intrinsic
1755 <p>It is legal for this instruction to be called in a function which
1756 does not take a variable number of arguments, for example, the <tt>vfprintf</tt>
1758 <p><tt>vanext</tt> is an LLVM instruction instead of an <a
1759 href="#intrinsics">intrinsic function</a> because it takes an type as
1762 <p>See the <a href="#int_varargs">variable argument processing</a>
1765 <!-- _______________________________________________________________________ -->
1766 <div class="doc_subsubsection"> <a name="i_vaarg">'<tt>vaarg</tt>'
1767 Instruction</a> </div>
1768 <div class="doc_text">
1770 <pre> <resultval> = vaarg <va_list> <arglist>, <argty><br></pre>
1772 <p>The '<tt>vaarg</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 a value of the specified argument type.</p>
1779 <p>The '<tt>vaarg</tt>' instruction loads an argument of the specified
1780 type from the specified <tt>va_list</tt>. In conjunction with the <a
1781 href="#i_vanext"><tt>vanext</tt></a> instruction, it is used to
1782 implement the <tt>va_arg</tt> macro available in C. For more
1783 information, see 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>vaarg</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>
1796 <!-- *********************************************************************** -->
1797 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
1798 <!-- *********************************************************************** -->
1800 <div class="doc_text">
1802 <p>LLVM supports the notion of an "intrinsic function". These functions have
1803 well known names and semantics, and are required to follow certain
1804 restrictions. Overall, these instructions represent an extension mechanism for
1805 the LLVM language that does not require changing all of the transformations in
1806 LLVM to add to the language (or the bytecode reader/writer, the parser,
1809 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix, this
1810 prefix is reserved in LLVM for intrinsic names, thus functions may not be named
1811 this. Intrinsic functions must always be external functions: you cannot define
1812 the body of intrinsic functions. Intrinsic functions may only be used in call
1813 or invoke instructions: it is illegal to take the address of an intrinsic
1814 function. Additionally, because intrinsic functions are part of the LLVM
1815 language, it is required that they all be documented here if any are added.</p>
1819 Adding an intrinsic to LLVM is straight-forward if it is possible to express the
1820 concept in LLVM directly (ie, code generator support is not _required_). To do
1821 this, extend the default implementation of the IntrinsicLowering class to handle
1822 the intrinsic. Code generators use this class to lower intrinsics they do not
1823 understand to raw LLVM instructions that they do.
1828 <!-- ======================================================================= -->
1829 <div class="doc_subsection">
1830 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
1833 <div class="doc_text">
1835 <p>Variable argument support is defined in LLVM with the <a
1836 href="#i_vanext"><tt>vanext</tt></a> instruction and these three
1837 intrinsic functions. These functions are related to the similarly
1838 named macros defined in the <tt><stdarg.h></tt> header file.</p>
1840 <p>All of these functions operate on arguments that use a
1841 target-specific value type "<tt>va_list</tt>". The LLVM assembly
1842 language reference manual does not define what this type is, so all
1843 transformations should be prepared to handle intrinsics with any type
1846 <p>This example shows how the <a href="#i_vanext"><tt>vanext</tt></a>
1847 instruction and the variable argument handling intrinsic functions are
1851 int %test(int %X, ...) {
1852 ; Initialize variable argument processing
1853 %ap = call sbyte* %<a href="#i_va_start">llvm.va_start</a>()
1855 ; Read a single integer argument
1856 %tmp = vaarg sbyte* %ap, int
1858 ; Advance to the next argument
1859 %ap2 = vanext sbyte* %ap, int
1861 ; Demonstrate usage of llvm.va_copy and llvm.va_end
1862 %aq = call sbyte* %<a href="#i_va_copy">llvm.va_copy</a>(sbyte* %ap2)
1863 call void %<a href="#i_va_end">llvm.va_end</a>(sbyte* %aq)
1865 ; Stop processing of arguments.
1866 call void %<a href="#i_va_end">llvm.va_end</a>(sbyte* %ap2)
1872 <!-- _______________________________________________________________________ -->
1873 <div class="doc_subsubsection">
1874 <a name="i_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
1878 <div class="doc_text">
1880 <pre> call va_list ()* %llvm.va_start()<br></pre>
1882 <p>The '<tt>llvm.va_start</tt>' intrinsic returns a new <tt><arglist></tt>
1883 for subsequent use by the variable argument intrinsics.</p>
1885 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
1886 macro available in C. In a target-dependent way, it initializes and
1887 returns a <tt>va_list</tt> element, so that the next <tt>vaarg</tt>
1888 will produce the first variable argument passed to the function. Unlike
1889 the C <tt>va_start</tt> macro, this intrinsic does not need to know the
1890 last argument of the function, the compiler can figure that out.</p>
1891 <p>Note that this intrinsic function is only legal to be called from
1892 within the body of a variable argument function.</p>
1895 <!-- _______________________________________________________________________ -->
1896 <div class="doc_subsubsection">
1897 <a name="i_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
1900 <div class="doc_text">
1902 <pre> call void (va_list)* %llvm.va_end(va_list <arglist>)<br></pre>
1904 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt><arglist></tt>
1905 which has been initialized previously with <tt><a href="#i_va_start">llvm.va_start</a></tt>
1906 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
1908 <p>The argument is a <tt>va_list</tt> to destroy.</p>
1910 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
1911 macro available in C. In a target-dependent way, it destroys the <tt>va_list</tt>.
1912 Calls to <a href="#i_va_start"><tt>llvm.va_start</tt></a> and <a
1913 href="#i_va_copy"><tt>llvm.va_copy</tt></a> must be matched exactly
1914 with calls to <tt>llvm.va_end</tt>.</p>
1917 <!-- _______________________________________________________________________ -->
1918 <div class="doc_subsubsection">
1919 <a name="i_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
1922 <div class="doc_text">
1927 call va_list (va_list)* %llvm.va_copy(va_list <destarglist>)
1932 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
1933 from the source argument list to the destination argument list.</p>
1937 <p>The argument is the <tt>va_list</tt> to copy.</p>
1941 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
1942 macro available in C. In a target-dependent way, it copies the source
1943 <tt>va_list</tt> element into the returned list. This intrinsic is necessary
1944 because the <tt><a href="i_va_start">llvm.va_start</a></tt> intrinsic may be
1945 arbitrarily complex and require memory allocation, for example.</p>
1949 <!-- ======================================================================= -->
1950 <div class="doc_subsection">
1951 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
1954 <div class="doc_text">
1957 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
1958 Collection</a> requires the implementation and generation of these intrinsics.
1959 These intrinsics allow identification of <a href="#i_gcroot">GC roots on the
1960 stack</a>, as well as garbage collector implementations that require <a
1961 href="#i_gcread">read</a> and <a href="#i_gcwrite">write</a> barriers.
1962 Front-ends for type-safe garbage collected languages should generate these
1963 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
1964 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
1968 <!-- _______________________________________________________________________ -->
1969 <div class="doc_subsubsection">
1970 <a name="i_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
1973 <div class="doc_text">
1978 call void (<ty>**, <ty2>*)* %llvm.gcroot(<ty>** %ptrloc, <ty2>* %metadata)
1983 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existance of a GC root to
1984 the code generator, and allows some metadata to be associated with it.</p>
1988 <p>The first argument specifies the address of a stack object that contains the
1989 root pointer. The second pointer (which must be either a constant or a global
1990 value address) contains the meta-data to be associated with the root.</p>
1994 <p>At runtime, a call to this intrinsics stores a null pointer into the "ptrloc"
1995 location. At compile-time, the code generator generates information to allow
1996 the runtime to find the pointer at GC safe points.
2002 <!-- _______________________________________________________________________ -->
2003 <div class="doc_subsubsection">
2004 <a name="i_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
2007 <div class="doc_text">
2012 call sbyte* (sbyte**)* %llvm.gcread(sbyte** %Ptr)
2017 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
2018 locations, allowing garbage collector implementations that require read
2023 <p>The argument is the address to read from, which should be an address
2024 allocated from the garbage collector.</p>
2028 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
2029 instruction, but may be replaced with substantially more complex code by the
2030 garbage collector runtime, as needed.</p>
2035 <!-- _______________________________________________________________________ -->
2036 <div class="doc_subsubsection">
2037 <a name="i_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
2040 <div class="doc_text">
2045 call void (sbyte*, sbyte**)* %llvm.gcwrite(sbyte* %P1, sbyte** %P2)
2050 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
2051 locations, allowing garbage collector implementations that require write
2052 barriers (such as generational or reference counting collectors).</p>
2056 <p>The first argument is the reference to store, and the second is the heap
2057 location to store to.</p>
2061 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
2062 instruction, but may be replaced with substantially more complex code by the
2063 garbage collector runtime, as needed.</p>
2069 <!-- ======================================================================= -->
2070 <div class="doc_subsection">
2071 <a name="int_codegen">Code Generator Intrinsics</a>
2074 <div class="doc_text">
2076 These intrinsics are provided by LLVM to expose special features that may only
2077 be implemented with code generator support.
2082 <!-- _______________________________________________________________________ -->
2083 <div class="doc_subsubsection">
2084 <a name="i_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
2087 <div class="doc_text">
2091 call void* ()* %llvm.returnaddress(uint <level>)
2097 The '<tt>llvm.returnaddress</tt>' intrinsic returns a target-specific value
2098 indicating the return address of the current function or one of its callers.
2104 The argument to this intrinsic indicates which function to return the address
2105 for. Zero indicates the calling function, one indicates its caller, etc. The
2106 argument is <b>required</b> to be a constant integer value.
2112 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
2113 the return address of the specified call frame, or zero if it cannot be
2114 identified. The value returned by this intrinsic is likely to be incorrect or 0
2115 for arguments other than zero, so it should only be used for debugging purposes.
2119 Note that calling this intrinsic does not prevent function inlining or other
2120 aggressive transformations, so the value returned may not that of the obvious
2121 source-language caller.
2126 <!-- _______________________________________________________________________ -->
2127 <div class="doc_subsubsection">
2128 <a name="i_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
2131 <div class="doc_text">
2135 call void* ()* %llvm.frameaddress(uint <level>)
2141 The '<tt>llvm.frameaddress</tt>' intrinsic returns the target-specific frame
2142 pointer value for the specified stack frame.
2148 The argument to this intrinsic indicates which function to return the frame
2149 pointer for. Zero indicates the calling function, one indicates its caller,
2150 etc. The argument is <b>required</b> to be a constant integer value.
2156 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
2157 the frame address of the specified call frame, or zero if it cannot be
2158 identified. The value returned by this intrinsic is likely to be incorrect or 0
2159 for arguments other than zero, so it should only be used for debugging purposes.
2163 Note that calling this intrinsic does not prevent function inlining or other
2164 aggressive transformations, so the value returned may not that of the obvious
2165 source-language caller.
2169 <!-- ======================================================================= -->
2170 <div class="doc_subsection">
2171 <a name="int_os">Operating System Intrinsics</a>
2174 <div class="doc_text">
2176 These intrinsics are provided by LLVM to support the implementation of
2177 operating system level code.
2182 <!-- _______________________________________________________________________ -->
2183 <div class="doc_subsubsection">
2184 <a name="i_readport">'<tt>llvm.readport</tt>' Intrinsic</a>
2187 <div class="doc_text">
2191 call <integer type> (<integer type>)* %llvm.readport (<integer type> <address>)
2197 The '<tt>llvm.readport</tt>' intrinsic reads data from the specified hardware
2204 The argument to this intrinsic indicates the hardware I/O address from which
2205 to read the data. The address is in the hardware I/O address namespace (as
2206 opposed to being a memory location for memory mapped I/O).
2212 The '<tt>llvm.readport</tt>' intrinsic reads data from the hardware I/O port
2213 specified by <i>address</i> and returns the value. The address and return
2214 value must be integers, but the size is dependent upon the platform upon which
2215 the program is code generated. For example, on x86, the address must be an
2216 unsigned 16 bit value, and the return value must be 8, 16, or 32 bits.
2221 <!-- _______________________________________________________________________ -->
2222 <div class="doc_subsubsection">
2223 <a name="i_writeport">'<tt>llvm.writeport</tt>' Intrinsic</a>
2226 <div class="doc_text">
2230 call void (<integer type>, <integer type>)* %llvm.writeport (<integer type> <value>, <integer type> <address>)
2236 The '<tt>llvm.writeport</tt>' intrinsic writes data to the specified hardware
2243 The first argument is the value to write to the I/O port.
2247 The second argument indicates the hardware I/O address to which data should be
2248 written. The address is in the hardware I/O address namespace (as opposed to
2249 being a memory location for memory mapped I/O).
2255 The '<tt>llvm.writeport</tt>' intrinsic writes <i>value</i> to the I/O port
2256 specified by <i>address</i>. The address and value must be integers, but the
2257 size is dependent upon the platform upon which the program is code generated.
2258 For example, on x86, the address must be an unsigned 16 bit value, and the
2259 value written must be 8, 16, or 32 bits in length.
2264 <!-- _______________________________________________________________________ -->
2265 <div class="doc_subsubsection">
2266 <a name="i_readio">'<tt>llvm.readio</tt>' Intrinsic</a>
2269 <div class="doc_text">
2273 call <result> (<ty>*)* %llvm.readio (<ty> * <pointer>)
2279 The '<tt>llvm.readio</tt>' intrinsic reads data from a memory mapped I/O
2286 The argument to this intrinsic is a pointer indicating the memory address from
2287 which to read the data. The data must be a
2288 <a href="#t_firstclass">first class</a> type.
2294 The '<tt>llvm.readio</tt>' intrinsic reads data from a memory mapped I/O
2295 location specified by <i>pointer</i> and returns the value. The argument must
2296 be a pointer, and the return value must be a
2297 <a href="#t_firstclass">first class</a> type. However, certain architectures
2298 may not support I/O on all first class types. For example, 32 bit processors
2299 may only support I/O on data types that are 32 bits or less.
2303 This intrinsic enforces an in-order memory model for llvm.readio and
2304 llvm.writeio calls on machines that use dynamic scheduling. Dynamically
2305 scheduled processors may execute loads and stores out of order, re-ordering at
2306 run time accesses to memory mapped I/O registers. Using these intrinsics
2307 ensures that accesses to memory mapped I/O registers occur in program order.
2312 <!-- _______________________________________________________________________ -->
2313 <div class="doc_subsubsection">
2314 <a name="i_writeio">'<tt>llvm.writeio</tt>' Intrinsic</a>
2317 <div class="doc_text">
2321 call void (<ty1>, <ty2>*)* %llvm.writeio (<ty1> <value>, <ty2> * <pointer>)
2327 The '<tt>llvm.writeio</tt>' intrinsic writes data to the specified memory
2334 The first argument is the value to write to the memory mapped I/O location.
2335 The second argument is a pointer indicating the memory address to which the
2336 data should be written.
2342 The '<tt>llvm.writeio</tt>' intrinsic writes <i>value</i> to the memory mapped
2343 I/O address specified by <i>pointer</i>. The value must be a
2344 <a href="#t_firstclass">first class</a> type. However, certain architectures
2345 may not support I/O on all first class types. For example, 32 bit processors
2346 may only support I/O on data types that are 32 bits or less.
2350 This intrinsic enforces an in-order memory model for llvm.readio and
2351 llvm.writeio calls on machines that use dynamic scheduling. Dynamically
2352 scheduled processors may execute loads and stores out of order, re-ordering at
2353 run time accesses to memory mapped I/O registers. Using these intrinsics
2354 ensures that accesses to memory mapped I/O registers occur in program order.
2360 <!-- ======================================================================= -->
2361 <div class="doc_subsection">
2362 <a name="int_libc">Standard C Library Intrinsics</a>
2365 <div class="doc_text">
2367 LLVM provides intrinsics for a few important standard C library functions.
2368 These intrinsics allow source-language front-ends to pass information about the
2369 alignment of the pointer arguments to the code generator, providing opportunity
2370 for more efficient code generation.
2375 <!-- _______________________________________________________________________ -->
2376 <div class="doc_subsubsection">
2377 <a name="i_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
2380 <div class="doc_text">
2384 call void (sbyte*, sbyte*, uint, uint)* %llvm.memcpy(sbyte* <dest>, sbyte* <src>,
2385 uint <len>, uint <align>)
2391 The '<tt>llvm.memcpy</tt>' intrinsic copies a block of memory from the source
2392 location to the destination location.
2396 Note that, unlike the standard libc function, the <tt>llvm.memcpy</tt> intrinsic
2397 does not return a value, and takes an extra alignment argument.
2403 The first argument is a pointer to the destination, the second is a pointer to
2404 the source. The third argument is an (arbitrarily sized) integer argument
2405 specifying the number of bytes to copy, and the fourth argument is the alignment
2406 of the source and destination locations.
2410 If the call to this intrinisic has an alignment value that is not 0 or 1, then
2411 the caller guarantees that the size of the copy is a multiple of the alignment
2412 and that both the source and destination pointers are aligned to that boundary.
2418 The '<tt>llvm.memcpy</tt>' intrinsic copies a block of memory from the source
2419 location to the destination location, which are not allowed to overlap. It
2420 copies "len" bytes of memory over. If the argument is known to be aligned to
2421 some boundary, this can be specified as the fourth argument, otherwise it should
2427 <!-- _______________________________________________________________________ -->
2428 <div class="doc_subsubsection">
2429 <a name="i_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
2432 <div class="doc_text">
2436 call void (sbyte*, sbyte*, uint, uint)* %llvm.memmove(sbyte* <dest>, sbyte* <src>,
2437 uint <len>, uint <align>)
2443 The '<tt>llvm.memmove</tt>' intrinsic moves a block of memory from the source
2444 location to the destination location. It is similar to the '<tt>llvm.memcpy</tt>'
2445 intrinsic but allows the two memory locations to overlap.
2449 Note that, unlike the standard libc function, the <tt>llvm.memmove</tt> intrinsic
2450 does not return a value, and takes an extra alignment argument.
2456 The first argument is a pointer to the destination, the second is a pointer to
2457 the source. The third argument is an (arbitrarily sized) integer argument
2458 specifying the number of bytes to copy, and the fourth argument is the alignment
2459 of the source and destination locations.
2463 If the call to this intrinisic has an alignment value that is not 0 or 1, then
2464 the caller guarantees that the size of the copy is a multiple of the alignment
2465 and that both the source and destination pointers are aligned to that boundary.
2471 The '<tt>llvm.memmove</tt>' intrinsic copies a block of memory from the source
2472 location to the destination location, which may overlap. It
2473 copies "len" bytes of memory over. If the argument is known to be aligned to
2474 some boundary, this can be specified as the fourth argument, otherwise it should
2480 <!-- _______________________________________________________________________ -->
2481 <div class="doc_subsubsection">
2482 <a name="i_memset">'<tt>llvm.memset</tt>' Intrinsic</a>
2485 <div class="doc_text">
2489 call void (sbyte*, ubyte, uint, uint)* %llvm.memset(sbyte* <dest>, ubyte <val>,
2490 uint <len>, uint <align>)
2496 The '<tt>llvm.memset</tt>' intrinsic fills a block of memory with a particular
2501 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
2502 does not return a value, and takes an extra alignment argument.
2508 The first argument is a pointer to the destination to fill, the second is the
2509 byte value to fill it with, the third argument is an (arbitrarily sized) integer
2510 argument specifying the number of bytes to fill, and the fourth argument is the
2511 known alignment of destination location.
2515 If the call to this intrinisic has an alignment value that is not 0 or 1, then
2516 the caller guarantees that the size of the copy is a multiple of the alignment
2517 and that the destination pointer is aligned to that boundary.
2523 The '<tt>llvm.memset</tt>' intrinsic fills "len" bytes of memory starting at the
2524 destination location. If the argument is known to be aligned to some boundary,
2525 this can be specified as the fourth argument, otherwise it should be set to 0 or
2531 <!-- ======================================================================= -->
2532 <div class="doc_subsection">
2533 <a name="int_debugger">Debugger Intrinsics</a>
2536 <div class="doc_text">
2538 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
2539 are described in the <a
2540 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
2541 Debugging</a> document.
2546 <!-- *********************************************************************** -->
2549 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
2550 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
2551 <a href="http://validator.w3.org/check/referer"><img
2552 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
2554 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
2555 <a href="http://llvm.cs.uiuc.edu">The LLVM Compiler Infrastructure</a><br>
2556 Last modified: $Date$